Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge A Series of Modern Surveys in Mathematics 66
Luis Ribes Profi nite Graphs and Groups Ergebnisse der Mathematik und Volume 66 ihrer Grenzgebiete
3. Folge
A Series of Modern Surveys in Mathematics
Editorial Board L. Ambrosio, Pisa V. Baladi, Paris G.-M. Greuel, Kaiserslautern M. Gromov, Bures-sur-Yvette G. Huisken, Tübingen J. Jost, Leipzig J. Kollár, Princeton G. Laumon, Orsay U. Tillmann, Oxford J. Tits, Paris D.B. Zagier, Bonn
For further volumes: www.springer.com/series/728 Luis Ribes
Profinite Graphs and Groups Luis Ribes School of Mathematics and Statistics Carleton University Ottawa, ON, Canada
ISSN 0071-1136 ISSN 2197-5655 (electronic) Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics ISBN 978-3-319-61041-2 ISBN 978-3-319-61199-0 (eBook) DOI 10.1007/978-3-319-61199-0
Library of Congress Control Number: 2017952053
Mathematics Subject Classification: 20E18, 20E06, 20E08, 20F65, 20J05, 05C05, 20M35, 22C05
© Springer International Publishing AG 2017 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature The registered company is Springer International Publishing AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland Para Fabián
Para Luisito y Tomasín Preface
Profinite groups are Galois groups, which we view as topological groups. In this book the theory of profinite graphs is developed as a natural tool in the study of some aspects of profinite and abstract groups. Our approach is modelled on the by now classical BassÐSerre theory of abstract groups acting on abstract trees as it appears in J.-P. Serre’s monograph ‘Trees’. We think of a graph Γ as the union of its sets of vertices V and edges E. A graph Γ is profinite if it is endowed with a profinite topology (i.e., a compact, Hausdorff and totally disconnected topology), in such a way that the functions defining the origin and terminal points are continuous. A natural example of a profinite graph is the Cayley graph Γ(G,X)of a profinite group G with respect to a closed subset X, say finite, of G: the vertices of Γ are the elements of G, and its directed edges have the form (g, x) (g ∈ G, x ∈ X) with origin d0(g, x) = g and terminal d1(g, x) = gx. Then the topology of G naturally induces a profinite topology on Γ(G,X). Part I of this book contains an exposition of the theory of profinite graphs and how it relates to and is motivated by the theory of profinite groups. Part II deals with applications to profinite groups, while Part III is dedicated to the study of certain properties of abstract groups with the help of tools developed in Parts I and II. Our aim in Parts I and II has been to make the exposition self-contained, and familiarity with the theory of abstract graphs and groups is not strictly necessary. However, knowledge of the BassÐSerre theory certainly helps, and throughout these two parts we often indicate the interconnections. These connections are in fact the main tools for some of the applications to abstract groups in Part III, where results and ideas ranging from topology and abstract group theory to automata theory are used freely. One fundamental difference with the abstract case is that a profinite group acting freely on a profinite tree need not be a free profinite group (it is just projective). This leads to a study of Galois coverings of profinite graphs and fundamental groups of profinite graphs. Throughout the book we have tried to be as general as reasonably possible, and so we consider pro-C groups, where C is a class of finite groups, rather than profinite groups in general. Consequently the book includes studies of Galois C-coverings, C-trees, fundamental groups of graphs of pro-C groups, etc.
vii viii Preface
Part I (Chaps. 2Ð6) includes the development of free products of pro-C groups continuously indexed by a topological profinite space, and a full treatment of the fundamental pro-C group of a graph of pro-C groups. Part II (Chaps. 7Ð10) contains applications to the structure of profinite groups. In Chap. 7 we describe subgroups of fundamental groups of graphs of profinite groups; in particular, an analogue of the Kurosh subgroup theorem for open subgroups of free products of pro-C groups is established. Chapter 8 describes the properties of minimal G-invariant subtrees of a tree on which the group G acts; this is done for profinite as well as abstract groups and graphs. The study of such minimal trees was initiated by Tits when G is cyclic and acts without fixed points on an abstract tree. It turns out that the connections between these types of minimal subtrees in the abstract and profinite cases provides a powerful tool to study certain properties in abstract groups. Chapters 9 and 10 of Part II deal mainly with homology. Chapter 9 includes a theorem of Neukirch and a generalization of Mel’nikov characterizing homologically when a profinite group is the free product of a collection of sub- groups continuously indexed by a topological (profinite) space; this plays the role of the usual combinatorial description of free products in the case of abstract groups. This chapter also contains a Kurosh-like theorem for countably generated closed subgroups of free products of pro-p groups due to D. Haran and O. Mel’nikov in- dependently. Chapter 10 includes the well-known theorem of J.-P. Serre that asserts that a torsion-free pro-p group G with an open free pro-p subgroup must be free pro-p. There is also a generalization of this result due to C. Scheiderer, where one allows torsion in G. Using this, the chapter also contains a study of the subgroup of fixed points of an automorphism of a free pro-p group. Part III (Chaps. 11Ð15) contains applications to abstract groups. These include generalizations of a theorem of Marshall Hall that asserts that a finitely generated subgroup H of an abstract free group Φ is the intersection of the subgroups of fi- nite index in Φ that contain H ; an algorithm to compute the closure of a finitely generated subgroup H of an abstract free group Φ in the pro-p topology of Φ; and applications to the theory of formal languages and finite monoids. Also included is the study of certain properties that hold for an abstract group if and only if they hold for the finite quotients of that group, e.g., conjugacy separability for an ab- stract group R :forx,y ∈ R, these elements are conjugate in R if their images are conjugate in every finite quotient group of R. The book ranges over a large number of areas and results, but we have not in- tended to make this into an encyclopedia of the subject. Part I gives a fairly complete account of profinite graphs and their connection with profinite groups. However in Part II and, even more, in Part III, I have made a choice of topics to illustrate some results and methods. At the end of each of the three parts of the book there is a section with historical comments on the development of the fundamental ideas and theorems, statements of additional results, references to related topics, and open questions. In an effort to make the book self-contained, the first chapter includes a review of basic notions and results about profinite spaces, profinite groups and homology that are used frequently throughout the monograph. Appendix A deals with aspects Preface ix of abstract graphs that are of interest in the book. The main purpose has been to de- velop a terminology common to abstract and profinite graphs. Appendix B contains a proof of a theorem of M. Benois about rational languages in free abstract groups. I have been indebted to many colleagues during the writing of this book. Throughout the years I have had many mathematical discussions with my longtime collaborator Pavel Zalesskii that have helped to clarify some topics developed here; it is a pleasure to acknowledge with thanks my debt to him. I thank John Dixon, Wolfgang Herfort, Dan Segal and Benjaming Steinberg, who have read parts of the manuscript and have made very useful comments, corrections and suggestions. Jean-Eric Pin has provided helpful references, and I am very grateful to him for this. This book was written mainly in Ottawa and Madrid. In Ottawa my thanks go to Carleton University for continuous help throughout the years, and for sabbat- ical periods that have allowed me to concentrate on the writing of his book. In Madrid I have often used the facilities of the Universidad Complutense, the Uni- versidad Autónoma and ICMAT, and I thank all of them for their generosity, and my colleagues at these institutions for their welcome whenever I have spent time with them. Finally, I acknowledge with thanks the continued research support from NSERC. MadridÐOttawa Luis Ribes March, 2017 Contents
Preface ...... vii 1 Preliminaries ...... 1 1.1 InverseLimits...... 2 1.2 Profinite Spaces ...... 3 1.3 Profinite Groups ...... 4 Pseudovarieties C ...... 5 Generators ...... 6 G-Spaces and Continuous Sections ...... 6 Order of a Profinite Group and Sylow Subgroups ...... 7 1.4 Pro-C Topologies in Abstract Groups ...... 8 1.5 Free Groups ...... 9 1.6 Free and Amalgamated Products of Groups ...... 10 1.7 Profinite Rings and Modules ...... 11 Exact Sequences ...... 13 The Functors Hom(−, −) ...... 13 Projective and Injective Modules ...... 14 1.8 The Complete Group Algebra ...... 15 G-Modules ...... 15 Complete Tensor Products ...... 16 n − − Λ − − 1.9 The Functors ExtΛ( , ) and Torn ( , ) ...... 17 n − − The Functors ExtΛ( , ) ...... 17 Λ − − The Functors Torn ( , ) ...... 19 1.10 Homology and Cohomology of Profinite Groups ...... 20 Cohomology of Profinite Groups ...... 20 Special Maps in Cohomology ...... 23 Homology of Profinite Groups ...... 24 Duality Homology-Cohomology ...... 24 (Co)induced Modules and Shapiro’s Lemma ...... 25 1.11(Co)homologicalDimension...... 25
xi xii Contents
Part I Basic Theory 2 Profinite Graphs ...... 29 2.1 FirstNotionsandExamples...... 29 2.2 Groups Acting on Profinite Graphs ...... 41 2.3 TheChainComplexofaGraph...... 45 2.4 π-Trees and C-Trees...... 48 2.5 Cayley Graphs and C-Trees...... 57 3 The Fundamental Group of a Profinite Graph ...... 63 3.1 GaloisCoverings...... 63 3.2 G(Γ |) as a Subgroup of Aut(Γ ) ...... 72 3.3 Universal Galois Coverings and Fundamental Groups ...... 74 3.4 0-Transversalsand0-Sections...... 77 3.5 Existence of Universal Coverings ...... 82 3.6 Subgroups of Fundamental Groups of Graphs ...... 89 3.7 Universal Coverings and Simple Connectivity ...... 91 3.8 Fundamental Groups and Projective Groups ...... 95 3.9 Fundamental Groups of Quotient Graphs ...... 96 3.10 π-Trees and Simple Connectivity ...... 100 3.11 Free Pro-C Groups and Cayley Graphs ...... 105 3.12 Change of Pseudovariety ...... 107 4 Profinite Groups Acting on C-Trees ...... 111 4.1 FixedPoints...... 111 4.2 Faithful and Irreducible Actions ...... 119 5 Free Products of Pro-C Groups ...... 137 5.1 Free Pro-C Products: The External Viewpoint ...... 137 5.2 Subgroups Continuously Indexed by a Space ...... 145 5.3 Free Pro-C Products: The Internal Viewpoint ...... 148 5.4 Profinite G-Spaces vs the Weight w(G) of G ...... 153 5.5 Basic Properties of Free Pro-C Products ...... 157 5.6 Free Products and Change of Pseudovariety ...... 164 5.7 Constant and Pseudoconstant Sheaves ...... 167 6 Graphs of Pro-C Groups ...... 177 6.1 Graphs of Pro-C Groups and Specializations ...... 177 6.2 The Fundamental Group of a Graph of Pro-C Groups ...... 180 Uniqueness of the Fundamental Group ...... 185 6.3 The Standard Graph of a Graph of Pro-C Groups ...... 193 6.4 Injective Graphs of Pro-C Groups ...... 205 6.5 Abstract vs Profinite Graphs of Groups ...... 207 6.6 Action of a Pro-C Group on a Profinite Graph with Finite Quotient...... 213 6.7 Notes, Comments and Further Reading: Part I ...... 216 Abstract Graph of Finite Groups (G,Γ)over an Infinite Graph Γ 218 Contents xiii
Part II Applications to Profinite Groups
7 Subgroups of Fundamental Groups of Graphs of Groups ...... 223 7.1 Subgroups ...... 223 7.2 Normal Subgroups ...... 227 7.3 The Kurosh Theorem for Free Pro-C Products ...... 232
8 Minimal Subtrees ...... 237 8.1 Minimal Subtrees: The Abstract Case ...... 238 8.2 Minimal Subtrees: Abstract vs Profinite Trees ...... 241 Trees Associated with Virtually Free Groups ...... 242 8.3 Graphs of Residually Finite Groups and the Tits Line ...... 245 8.4 Graph of a Free Product of Groups and the Tits Line ...... 250
9 Homology and Graphs of Pro-C Groups ...... 257 9.1 Direct Sums of Modules and Homology ...... 257 9.2 Corestriction and Continuously Indexed Families of Subgroups 259 9.3 The Homology Sequence of the Action on a Tree ...... 265 9.4 MayerÐVietoris Sequences ...... 267 9.5 Homological Characterization of Free Pro-p Products ..... 270 9.6 Pro-p Groups Acting on C-Trees and the Kurosh Theorem . . . 272
10 The Virtual Cohomological Dimension of Profinite Groups ..... 279 10.1 Tensor Product of Complexes ...... 279 10.2 Tensor Product Induction for a Complex ...... 281 10.3TheTorsion-FreeCase...... 290 10.4 Groups Virtually of Finite Cohomological Dimension: Periodicity...... 291 10.5TheTorsionCase...... 295 10.6 Pro-p Groups with a Free Subgroup of Index p ...... 309 10.7 Counter Kurosh ...... 312 10.8 Fixed Points of Automorphisms of Free Pro-p Groups ..... 317 10.9 Notes, Comments and Further Reading: Part II ...... 322 M. Hall Pro-p Groups ...... 323
Part III Applications to Abstract Groups
11 Separability Conditions in Free and Polycyclic Groups ...... 329 11.1 Separability Conditions in Abstract Groups ...... 329 11.2 Subgroup Separability in Free-by-Finite Groups ...... 333 11.3 Products of Subgroups in Free Abstract Groups ...... 337 11.4 Separability Properties of Polycyclic Groups ...... 342 xiv Contents
12 Algorithms in Abstract Free Groups and Monoids ...... 349 12.1 Algorithms for Subgroups of Finite Index ...... 349 12.2 Closure of Finitely Generated Subgroups in Abstract Free Groups ...... 353 12.3 Algorithms for Monoids ...... 359 The Kernel of a Finite Monoid ...... 364 The Mal’cev Product of Pseudovarieties of Monoids ...... 366 13 Abstract Groups vs Their Profinite Completions ...... 369 13.1 Free-by-Finite Groups vs Their Profinite Completions ..... 369 13.2 Polycyclic-by-Finite Groups vs Their Profinite Completions . . 379 14 Conjugacy in Free Products and in Free-by-Finite Groups ..... 383 14.1 Conjugacy Separability in Free-by-Finite Groups ...... 383 14.2 Conjugacy Subgroup Separability in Free-by-Finite Groups . . 386 14.3 Conjugacy Distinguishedness in Free-by-Finite Groups ..... 389 15 Conjugacy Separability in Amalgamated Products ...... 391 15.1 Abstract Free Products with Cyclic Amalgamation ...... 392 15.2 Normalizers in Amalgamated Products of Groups ...... 396 15.3 Conjugacy Separability of Amalgamated Products ...... 399 15.4 Amalgamated Products, Quasi-potency and Subgroup Separability ...... 405 15.5 Amalgamated Products and Products of Cyclic Subgroups . . . 407 15.6 Amalgamated Products and Normalizers of Cyclic Subgroups . 411 15.7 Amalgamated Products and Intersections of Cyclic Subgroups . 412 15.8 Amalgamated Products and Conjugacy Distinguishedness . . . 415 15.9 Conjugacy Separability of Certain Iterated Amalgamated Products ...... 418 15.10 Examples of Conjugacy Separable Groups ...... 418 15.11 Notes, Comments and Further Reading: Part III ...... 422 Subgroup Separability and Free Products ...... 423 Conjugacy Separability, Subgroups and Extensions ...... 427 Conjugacy Distinguished Subgroups ...... 427 Appendix A Abstract Graphs ...... 429 A.1 The Fundamental Group of an Abstract Graph ...... 429 TheStarofaVertex...... 430 Paths...... 430 A.2 Coverings of Abstract Graphs ...... 435 A.3 Foldings...... 441 A.4 Algorithms...... 442 Intersection of Finitely Generated Subgroups ...... 443 A.5 Notes, Comments and Further Reading ...... 445 Contents xv
Appendix B Rational Sets in Free Groups and Automata ...... 447 B.1 FiniteStateAutomata:ReviewandNotation...... 447 B.2 The Classical Function ρ ...... 448 B.3 Rational Subsets in Free Groups ...... 449 B.4 Notes, Comments and Further Reading ...... 451 References ...... 453 Index of Symbols ...... 461 Index of Authors ...... 463 Index of Terms ...... 465 Chapter 1 Preliminaries
The purpose of this chapter is to review some notation, terminology, basic concepts and results that are frequently used throughout the book. For more details and proofs one can consult one of the standard references on general profinite groups, e.g., Serre (1994), Wilson (1998), Fried and Jarden (2008), Ribes and Zalesskii (2010). In general we follow the notation in Ribes and Zalesskii (2010), cited as RZ throughout this book.
Basic Notation
Z denotes the group of integers under addition (also the ring of integers). N denotes the set of natural numbers.
H ≤ G, H ≤f G, H ≤o G and H ≤c G indicate that H is a subgroup of the group G, respectively, of finite index, open, closed.
H G, H f G, H o G, H c G indicate that H is a normal subgroup of the group G, respectively, of finite index, open, closed.
S ⊆o T and S ⊆c T indicate that S is an open, respectively closed, subset of the topological space T . If H and K are subgroups of a group R, then −1 NK (H ) = x ∈ K x Hx = H denotes the normalizer of H in K, and −1 CK (H ) = x ∈ K x hx = h, ∀h ∈ H denotes the centralizer of H in K. Let x and y be elements of a group R. Then xy = y−1xy and yx = yxy−1.
© Springer International Publishing AG 2017 1 L. Ribes, Profinite Graphs and Groups, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics 66, DOI 10.1007/978-3-319-61199-0_1 2 1 Preliminaries
Cn denotes the cyclic group of order n, written multiplicatively.
Fp denotes the prime field with p elements (occasionally also the additive group of that field). For other pieces of notation the reader may consult the Index.
1.1 Inverse Limits
A poset (or partially ordered set) (I, ) is said to be directed if whenever i, j ∈ I , there exists some k ∈ I such that i, j k.Aninverse system of topological spaces (respectively, topological groups) over such a directed poset (I, ) consists of a collection
{Xi | i ∈ I} of topological spaces (respectively, topological groups) indexed by I , and a col- lection of continuous mappings (respectively, continuous group homomorphisms) ϕij : Xi −→ Xj , defined whenever i j, such that the diagrams of the form
ϕik Xi Xk
ϕij ϕjk
Xj commute whenever they are defined, i.e., whenever i, j, k ∈ I and i j k.LetX be a topological space (respectively, topological group) and let
ϕi : X −→ Xi (i ∈ I) be a collection of continuous mappings (respectively, continuous group homomor- phisms) that are compatible (i.e., ϕij ϕi = ϕj whenever j i). One says that X, together with these mappings, is an inverse limit of the inverse system {Xi,ϕij ,I} if the following universal property is satisfied:
ψ Y X
ϕi ψi Xi whenever Y is a topological space (respectively, topological group) and ψi : Y −→ Xi (i ∈ I) is a set of compatible continuous mappings (respectively, con- tinuous homomorphisms), then there is a unique continuous mapping (respectively, continuous homomorphism) ψ : Y −→ X such that ϕiψ = ψi , for all i ∈ I .The maps ϕi : X −→ Xi are called projections. It follows easily from the definition that inverse limits are unique, if they exist. An inverse limit X of an inverse system {Xi,ϕij ,I} of topological spaces (respectively, topological groups) over a directed poset I can be constructed as follows: define X 1.2 Profinite Spaces 3 as the subspace (respectively, subgroup) of the direct product i∈I Xi of topological spaces (respectively, topological groups) consisting of those tuples (xi) that satisfy the condition ϕij (xi) = xj ,ifi j.Let
ϕi : X −→ Xi −→ denote the restriction of the canonical projection i∈I Xi Xi . Then one easily checks that each ϕi is continuous (respectively, a continuous homomorphism), and that (X, ϕi) is an inverse limit. These definitions of inverse systems and inverse limits translate in an obvious manner to other categories: one can replace topological spaces and continuous func- tions with sets and maps of sets, rings and homomorphisms of rings, etc. Next we state some useful facts that we use often in this book. The first one says that the inverse limit of an inverse system of nonempty finite sets is nonempty, or in more generality:
1.1.1 Let {Xi,ϕij ,I} be an inverse system of compact Hausdorff nonempty topo- logical spaces Xi over the directed poset (I, ). Then
lim←− Xi i∈I is nonempty.
Let (I, ) be a directed poset. Assume that I is a subset of I in such a way that (I , ) becomes a directed poset. We say that I is cofinal in I if for every i ∈ I there is some i ∈ I such that i i .If{Xi,ϕij ,I} is an inverse system and I is cofinal in I , then {Xi,ϕij ,I } becomes an inverse system in an obvious way, and we say that {Xi,ϕij ,I } is a cofinal subsystem of {Xi,ϕij ,I}.
1.1.2 Let {Xi,ϕij ,I} be an inverse system of compact topological spaces (respec- tively, compact topological groups) over a directed poset I and assume that I is a cofinal subset of I . Then
=∼ lim←− Xi lim←− Xi . i∈I i ∈I
1.2 Profinite Spaces
A profinite space (sometimes called a boolean space) X is an inverse limit = X lim←− Xi (1.1) i∈I of an inverse system of finite spaces (endowed with the discrete topology) {Xi,ϕij ,I}. Recall that a topological space is totally disconnected if every point in the space is its own connected component. One can then describe a profinite space in terms of internal topological properties as follows: 4 1 Preliminaries
1.2.1 A topological space is profinite if and only if it is compact, Hausdorff and totally disconnected.
In one direction, this is a consequence of the construction of the inverse limit as a subspace of the direct product of finite spaces and Tychonoff’s theorem, which asserts that the direct product of compact spaces is compact. Another important property of profinite spaces that is very useful is that its topology is determined by its closed and open (or clopen) subsets:
1.2.2 A topological space is profinite if and only if it is compact, Hausdorff and its topology admits a base of clopen sets.
In (1.1) denote by ϕi : X → Xi the projection into Xi . Then one has
1.2.3 If Y ⊆ X, then the topological closure Y¯ of Y in the profinite space X is the inverse limit of the images ϕi(Y ). Consequently, two subspaces of X with the same images under every ϕi have the same closure.
Let X be a topological space. Define the weight w(X) of X to be the smallest cardinal of a base of open sets of X. We denote by ρ(X) the cardinal of the set of all clopen subsets of X. A topological space X is said to be second countable,orto satisfy the second axiom of countability, if it has a countable base of open sets.
1.2.4 If X is an infinite profinite space, then w(X) = ρ(X). In particular, the car- dinality of any base of open sets of X consisting of clopen sets is ρ(X).
1.2.5 A profinite space X is second countable if and only if = X lim←− Xi, i∈I where (I, ) is a countable totally ordered set and each Xi is a finite discrete space.
1.3 Profinite Groups
A profinite group G is an inverse limit of finite groups Gi = G lim←− Gi, (1.2) i∈I and so one checks that the class of profinite groups is exactly the class of those topological groups that are compact, Hausdorff and totally disconnected. Equivalently, profinite groups are precisely Galois groups of Galois extensions of fields. Other concrete examples of profinite groups arise (and this is of importance in Part III of this book) when taking completions of abstract groups endowed with certain natural topologies (see Sect. 1.4 below). 1.3 Profinite Groups 5
Pseudovarieties C
Sometimes it is convenient to be more precise and consider finite groups Gi in the above inverse limit with specific properties. For this purpose one introduces the notion of a pseudovariety of finite groups: a nonempty class of finite groups C is a pseudovariety1 if it is closed under taking subgroups, homomorphic images and finite direct products. Important examples of pseudovarieties of finite groups: the class of all finite groups; the class of all finite solvable groups; the class of all finite nilpotent groups; the class of all finite p-groups (where p is a fixed prime number); the class of all finite abelian groups, etc. Let C be a fixed pseudovariety of finite groups. A pro-C group G is an in- verse limit (1.2) of finite groups Gi in C. Thus one speaks of a prosolvable group, a pronilpotent group, a pro-p group or a proabelian group, depending on which pseudovariety C, among those mentioned above, has been selected. A pseudovariety of finite groups C is said to be extension-closed if whenever 1 −→ K −→ G −→ H −→ 1 is an exact sequence of finite groups with K,H ∈ C, then G ∈ C. The pseudovariety C is said to be closed under extensions with abelian kernel if whenever 1 −→ K −→ G −→ H −→ 1 is an exact sequence of finite groups with K,H ∈ C and K is abelian, then G ∈ C. For a pseudovariety of finite groups C, π(C) denotes the set of all prime numbers p that divide the order of some group G in C. We also say that π(C) is the set of primes involved in C. Of course, every pro-C group is automatically profinite. The topology of a profi- nite group is determined by its open normal subgroups, and one has the following useful way of describing pro-C groups:
1.3.1 A group G is pro-C if and only if G is a topological group with a fundamental system U of open neighbourhoods U of the identity element 1 such that each U is a normal subgroup of UG with G/ ∈ C, and = G lim←− G/U. U∈U
The following property is often useful.
1.3.2 Let G be a profinite group and let (1.2) be a decomposition of G as an inverse limit of finite groups Gi . Then every continuous homomorphism ϕ : G → H from G to a finite group H factors through one of the groups Gi ; i.e., there is some i ∈ I and some homomorphism ρ : Gi → H such that ϕ = ρϕi , where ϕi : G → Gi is the canonical projection.
A similar result holds for profinite spaces.
1In some publications, including RZ, this is called a ‘variety’ of finite groups. 6 1 Preliminaries
Generators
Let G be a profinite group and let X ⊆ G. One says that X generates G (as a topological group) if the abstract subgroup X of G generated by X is dense in G. We write G = X. We say that a subset X of a profinite group G converges to 1 if every open subgroup U of G contains all but a finite number of the elements in X.IfX generates G and converges to 1, then we say that X is a set of generators of G converging to 1. This definition is motivated by the following fact.
1.3.3 Every profinite group G admits a set of generators converging to 1.
If G is a profinite group, d(G) denotes the smallest cardinality of a set of gen- erators of G converging to 1. It turns out that the cardinal d(G) (or more precisely, whether or not this cardinal is at most ℵ0) plays an important role concerning the structure of certain groups acting on profinite trees, as we shall see later on in this book; and one reason for this is that some properties of such groups can be proved using induction, using the following characterization.
1.3.4 Let C be a pseudovariety of finite groups and let G be a pro-C group. Then d(G) ≤ℵ0 if and only if G has a countable descending chain of open normal sub- groups
G = G0 ≥ G1 ≥···≥Gi ≥··· ∞ = such that i=0 Gi 1.
Finitely generated profinite groups are in addition Hopfian, that is, they satisfy the following useful property.
1.3.5 Let G be a finitely generated profinite group. Then every continuous epimor- phism ϕ : G −→ G is an isomorphism.
G-Spaces and Continuous Sections
Let G be a profinite group. We say that a profinite space X is a left G-space,or simply a G-space,ifG acts continuously on X, that is, if there exists a continuous map G × X → X, denoted (g, x) → gx, such that (gh)x = g(hx) and 1x = x, for all g,h ∈ G, x ∈ X. A pointed topological space (X, ∗) is a G-space if X is a G-space in the above 1.3 Profinite Groups 7 sense, and in addition g∗=∗for all g ∈ G. The action of G on X (respectively (X, ∗))issaidtobefree if for every x ∈ X (respectively, for every x ∈ X −{∗}) and every 1 = g ∈ G one has gx = x.
1.3.6 Let G be a profinite group acting on a profinite space X. Then
(a) X admits a decomposition as an inverse limit of finite G-spaces Xi : = X lim←− Xi. i∈I (b) Suppose that G is finite and acts freely on X. Then X admits a decomposition as an inverse limit of finite free G-spaces Xi : = X lim←− Xi. i∈I
Similar results hold for pointed profinite G-spaces (X.∗). Associated with a G-space X there a quotient space G\X, whose elements are the G-orbits Gx (x ∈ X), which in turn is a profinite space if one endows G\X with the quotient topology. A section of the natural continuous epimorphism ϕ : X −→ G\X (1.3) is a map σ : G\X → X such that ϕσ = idG\X. In general, quotient maps of the form (1.3) do not admit continuous sections. In fact
1.3.7 For any nontrivial profinite group G, there is a profinite G-space X such that the natural continuous epimorphism ϕ : X −→ G\X does not admit a continuous section.
But the existence of continuous sections is guaranteed in the following important cases.
1.3.8 If H is a closed subgroup of a profinite group G, then the quotient map ϕ : G −→ H \G of G onto the space of right cosets of H admits a continuous section.
1.3.9 If a profinite group G acts freely on a profinite space X, then the quotient map ϕ : X −→ G\X admits a continuous section.
1.3.10 If a profinite group G acts on a second countable profinite space X, then the quotient map ϕ : X −→ G\X admits a continuous section.
Order of a Profinite Group and Sylow Subgroups
If G is an infinite profinite group, knowing its cardinality provides little informa- tion. There is, nevertheless, a very useful notion of order of a profinite group G that 8 1 Preliminaries reflects, in a global manner, the arithmetic properties of the finite quotient groups G/U , where U is an open normal subgroup of G. To each prime number p one associates an exponent n(p) (a natural number or the symbol ∞) that is the supre- mum (over all open normal subgroups of G) of all the natural numbers n such that pn divides the order of the finite group G/U . Then the order #G of the profinite group G is the formal product #G = pn(p) p (‘a supernatural number’). If p is a fixed prime number, a pro-p group has order pn, where 0 ≤ n ≤∞.Ap-Sylow subgroup P of a profinite group G is a maximal pro-p subgroup of G. Using Zorn’s Lemma one sees that G contains p-Sylow subgroups. They have properties analogous to those of the p-Sylow subgroups of finite groups: any two p-Sylow subgroups of G are conjugate and any pro-p subgroup of G is contained in one of its p-Sylow subgroups.
1.4 Pro-C Topologies in Abstract Groups
Let C be a pseudovariety of finite groups and let R be an abstract group. Define
N = NC(R) ={N R | R/N ∈ C}. One can make R into a topological group by taking N as a base of open neighbour- hoods of the identity element 1 of R; this topology is called the pro-C topology of R or, if emphasis is needed, the full pro-C topology of R. Note that the pro-C topology of R is Hausdorff if and only if N = 1. N∈N If that is the case, R is said to be residually C. When C is the pseudovariety of all finite groups (respectively, all finite p-groups, all finite solvable groups, etc.), one uses instead the expression residually finite (respectively, residually p, residually solvable,etc.). The pro-C completion of R is defined to be the pro-C group = RCˆ lim←− R/N. N∈N
The canonical homomorphism of R into RCˆ
ι : R −→ RCˆ is given by r → (rN)N∈N (r ∈ R). Note that ι is injective if and only if R is residu- ally C; ι(R) is a dense subgroup of RCˆ.IfC is the pseudovariety of all finite groups, ˆ one usually writes the completion as R, instead of RCˆ, and calls it the profinite com- pletion of R.IfC is the pseudovariety of all finite p-groups, where p is a fixed prime number, then one usually writes the completion as Rpˆ . 1.5 Free Groups 9
For example, the profinite completion of the group Z is denoted by Zˆ : ˆ = Z lim←− Z/nZ. n∈N
Its pro-p completion is denoted by Zp (rather than Zpˆ ), following a long tradition in number theory: = n Zp lim←− Z/p Z. n∈N ˆ Note that Z is naturally embedded in Z and in Zp. Observe also that one can think ˆ ˆ of Zp as being embedded in Z; in fact Zp is the unique p-Sylow subgroup of Z. Moreover, Zˆ is the direct product of its Sylow subgroups. An additional useful ob- ˆ servation is that Z and Zp naturally have a ring structure (they are examples of profinite rings). One usually refers to Zp as the ring of p-adic integers.
1.5 Free Groups
Let C be a pseudovariety of finite groups and let X be a finite set. Denote by Φ the free abstract groups with basis X. Then the pro-C completion
F = F(X)= ΦCˆ of Φ is called the free pro-C group on the finite set X (a basis of the pro-C group). It satisfies the expected universal property of a free object in the category of pro-C groups: if we denote by ι : X → F the natural map (it is an injection in general), then for every function ϕ : X → H into a pro-C group, there exists a unique continuous homomorphism ϕ¯ : F → H such that ϕι¯ = ϕ. One can define a free pro-C group with more general bases, namely when X is a profinite space (or a pointed profinite space). The universal property that defines such a free pro-C group F = F(X)is analogous to the universal property described above, one simply requires that ϕ be continuous (and a map of pointed spaces, if (X, ∗) is a pointed space). Again, one may assume that X is a subspace of F .One important peculiarity is that a free pro-C group on a pointed profinite space (X, ∗) is also a free pro-C group on a pointed space which is the one-point compactification of a set (a discrete space) S; then one can think of S as a set of generators of F converging to 1. The cardinality of such S is an invariant of the group F ; this allows us to define the rank of the free pro-C group F : rank(F ) =|S|. ˆ For example, Z is the free profinite group of rank 1, while Zp is the free pro-p group of rank 1.
1.5.1 If G is a pro-C group, there exists a free pro-C group F and a continuous epimorphism F → G. 10 1 Preliminaries
Closed subgroups of free pro-C groups need not be free pro-C (e.g., a p-Sylow subgroup of a free profinite group).
1.5.2 Open subgroups of free pro-C groups are free pro-C groups (if C is extension- closed).
1.5.3 Closed subgroups of free pro-p groups are free pro-p groups.
A profinite group P is called projective if it satisfies the following universal property: P ϕ¯ ϕ
H α G whenever H and G are profinite groups, ϕ : P → G is a continuous homomorphism and α : H → G is a continuous epimorphism, there exists a continuous homomor- phism ϕ¯ : P → H such that ϕ = αϕ¯.
1.5.4 Projective profinite groups are precisely the closed subgroups of free profinite groups.
1.6 Free and Amalgamated Products of Groups
A free pro-C product C G = G1 G2 = G1 G2 of two pro-C groups G1 and G2 is a pro-C group G together with continuous ho- momorphisms ϕ1 : G1 −→ G and ϕ2 : G2 −→ G satisfying the following universal property: G ψ ϕi
Gi L ψi for any pro-C group L and any continuous homomorphisms ψi : Gi −→ L (i = 1, 2), there is a unique continuous homomorphism ψ : G −→ L such that ψi = ψϕi (i = 1, 2). From this definition one easily deduces that the homomorphisms ϕ1 and ϕ2 are injective and that the free pro-C product G is unique up to isomorphism. One con- structs G as a completion of the free product G1 ∗ G2 of G1 and G2 as abstract groups, with respect to a certain pro-C topology on G1 ∗ G2 that ensures that ϕ1 and ϕ2 are continuous. For example, the free profinite group Fn of finite rank n is the free profinite product of n copies of Zˆ . 1.7 Profinite Rings and Modules 11
Next we remind the reader about free products with amalgamation (also called ‘amalgamated products’ or ‘amalgamated free products’) in the category of profinite groups (i.e., when C is the pseudovariety of all profinite groups). Let G1, G2 and K be profinite groups. Let fi : K −→ Gi (i = 1, 2) be continuous monomorphisms (here we think of them as inclusion maps). An amalgamated free profinite product of G1 and G2 with amalgamated subgroup K is defined to be a profinite group, denoted
G = G1 K G2, together with continuous homomorphisms
ϕi : Gi −→ G, (i = 1, 2), with ϕ1f1 = ϕ2f2, satisfying the following universal property: for any pair of con- tinuous homomorphisms ψ1 : G1 −→ L, ψ2 : G2 −→ L into a profinite group L with ψ1f1 = ψ2f2, there exists a unique continuous homomorphism ψ : G −→ L such that the following diagram is commutative
f1 K G1
f2 ϕ1
ϕ2 ψ1 G2 G ψ
ψ2 L Again the uniqueness of G is easily deduced from the definition. One constructs G as the completion of the free product G1 ∗K G2 with amalgamation of G1 and G2 amalgamating K as abstract groups, with respect to a certain profinite topology on G1 ∗K G2 to ensure that ϕ1 and ϕ2 are continuous. In this case it is not automatic that the homomorphisms ϕ1 and ϕ2 are injective; the main reason for this is that, in general, the abstract group G1 ∗K G2 is not residually finite. When ϕ1 and ϕ2 are injective, one says that the amalgamated free profinite prod- uct G1 K G2 is proper. Some proper amalgamated free profinite products are studied in detail in this book; this is an important prerequisite for using methods developed in this book to study certain properties of abstract groups of the form R1 ∗H R2.
1.7 Profinite Rings and Modules
A profinite ring Λ is an inverse limit of an inverse system {Λi,ϕij ,I} of finite rings. We always assume that rings have an identity element, usually denoted by 1, and that homomorphisms of rings send identity elements to identity elements. ˆ As pointed out earlier, Z and Zp, and more generally ZCˆ, are examples of profi- nite rings. Their quotient rings will also play an important role in parts of this book. 12 1 Preliminaries
We are interested in two types of Λ-modules M (‘Λ-module’ is meant in the usual abstract sense, but in addition we assume that M is a topological abelian group and that the action of Λ on M is continuous), namely those with the discrete topol- ogy and those with a profinite topology. In general, we assume that the action of Λ on M is on the left; for emphasis sometimes we write ‘left Λ-module’. If the action of Λ on M is on the right, we write ‘right Λ-module’. We will need both left and right Λ-modules in the sequel. The category of discrete Λ-modules and their Λ-homomorphisms is denoted by DMod(Λ). The category of profinite Λ-modules and their continuous Λ- homomorphisms is denoted by PMod(Λ). A profinite ring Λ has a fundamental system for the neighbourhoods of 0 consist- ing of open (two-sided) ideals; from this one deduces the following fact for discrete modules.
1.7.1 If M is a discrete Λ-module, then M is the union of its finite Λ-submodules; in particular, M is torsion as an abelian group.
Profinite Λ-modules have properties similar to those of profinite groups. For ex- ample,
1.7.2 If M is a profinite Λ-module, then M is the inverse limit of its finite quotient Λ-modules.
In analogy with the definition of a free profinite group, one defines a free profinite Λ-module on a pointed profinite space (X, ∗) to consist of a profinite Λ-module, denoted [[ Λ(X, ∗)]] , together with a map of pointed spaces
ι : (X, ∗) −→ Λ(X, ∗) (i.e., ι(∗) = 0) satisfying the following universal property: whenever ϕ : (X, ∗) −→ N is a continuous mapping of pointed spaces into a profinite Λ-module N, there exists a unique continuous Λ-homomorphism ϕ¯ :[[Λ(X, ∗)]] −→ N such that ϕι¯ = ϕ. One has a similar definition for a free profinite Λ-module, denoted [[ ΛX]] ,on a profinite space X.IfX is a set and R is any ring, we denote the free abstract R-module with basis X by [RX] (it is the direct sum of |X| copies of R, considered as an abstract R-module in a natural way). When X is finite, [[ ΛX]] = [ ΛX] (in this case we understand the topology on [ΛX] to be the product topology). As in the case of profinite groups, one develops the concepts of ‘set of generators converging to 0’ in a Λ-module and of ‘free profinite Λ-module on a set converging to 0’. There are results similar to 1.3.3 and 1.5.1: 1.7 Profinite Rings and Modules 13
1.7.3 Every profinite Λ-module admits a set of generators converging to 0 and every profinite Λ-module is the image of a continuous Λ-homomorphism from a free profinite Λ-module.
Exact Sequences
Let ϕi−1 ϕi ···−→Mi−1 −→ Mi −→ Mi+1 −→··· be a sequence of Λ-modules and Λ-homomorphisms. This sequence is said to be exact at Mi if Im(ϕi−1) = Ker(ϕi). If this is the case at each module of the sequence, we say that the sequence is exact. Sequences of the form 0 −→ A −→ B −→ C −→ 0, where 0 denotes the module consisting only of the element zero, are used frequently. When they are exact, one refers to them as ‘short exact sequences’.
The Functors Hom(−, −)
If M and N are Λ-modules, we denote by HomΛ(M, N) the abelian group of continuous homomorphisms from M to N.IfwefixM, then we denote by HomΛ(M, −) the function that assigns to the Λ-module N the abelian group HomΛ(M, N). This function behaves ‘functorially’ (in fact, we refer to HomΛ(M, −) as a functor, or more specifically as a ‘covariant’ functor) in the sense that a continuous Λ-homomorphism ϕ1 : N1 → N2 of Λ-modules determines a homomorphism of abelian groups
HomΛ(M, ϕ1) : HomΛ(M, N1) −→ HomΛ(M, N2) defined by
HomΛ(M, ϕ1)(ψ) = ϕ1ψ ψ ∈ HomΛ(M, N1) ; and furthermore, (1) if ϕ2 : N2 → N3 is also a continuous Λ-homomorphism of Λ-modules, then HomΛ(M, ϕ2ϕ1) = HomΛ(M, ϕ2)HomΛ(M, ϕ1), and (2) if idN : N → N is the identity homomorphism, then HomΛ(M, idN ) : HomΛ(M, N) → HomΛ(M, N) is the identity homomorphism. A functor that sends exact sequences to exact sequences is called an exact functor (to check whether this is the case, it is enough to consider short exact sequences). The functor HomΛ(M, −) is not exact in general, but it is always left exact, meaning that if 0 −→ A −→ B −→ C −→ 0 14 1 Preliminaries is a short exact sequence of Λ-modules and continuous Λ-homomorphisms, then the sequence of abelian groups
0 −→ HomΛ(M, A) −→ HomΛ(M, B) −→ HomΛ(M, C) is also exact. Similarly, Hom(−,N) is also a functor (it is called ‘contravariant’ because it reverses the arrows). In general it is not exact, but it is always left exact.
Projective and Injective Modules
A profinite Λ-module P is called projective (in the category of profinite Λ-modules PMod(Λ)) if is a direct summand of a free profinite Λ-module of the form [[ ΛX]] , i.e., [[ ΛX]] = P ⊕ P , where P is some profinite Λ-submodule of [[ ΛX]] .Inview of 1.7.3, this is equivalent to saying that P satisfies the following universal property: P ϕ¯ ϕ
B α A whenever ϕ : P → A is a continuous Λ-homomorphism and α : B → A is a continuous Λ-epimorphism of profinite Λ-modules, there exists a continuous Λ- homomorphism ϕ¯ : P → B such that ϕ = αϕ¯. This second definition is ‘better’ in the sense that it can be formulated in any category (not just for profinite modules). Still another equivalent way of indicating that P is projective is to say that the func- tor HomΛ(P, −) is exact. In particular, it follows from the first definition that every free profinite Λ-module is projective and so (see 1.7.3) every profinite Λ-module is the homomorphic image of a projective module in the category PMod(Λ). One sometimes expresses this by saying that the category PMod(Λ) ‘has enough projectives’. This is important when developing a homology theory for profinite modules, as we will see in Sect. 1.9. Another important observation is that to check whether or not a profinite Λ-module P is projective, one may assume that in the diagram above the modules A and B are finite. The dual concept of ‘projective object’ is ‘injective object’. We recall the def- inition explicitly for discrete Λ-modules (i.e., in the category DMod(Λ)). One says that a discrete Λ-module Q is injective if whenever α : A → B is a Λ- monomorphism of discrete Λ-modules and ϕ : A → Q is a Λ-homomorphism,
A α B
ϕ ϕ¯ Q 1.8 The Complete Group Algebra 15 then there exists a Λ-homomorphism ϕ¯ : B −→ Q making the diagram com- mutative, i.e., ϕα¯ = ϕ. Equivalently, Q is injective in DMod(Λ) if the functor Hom(−,Q)is exact, i.e., whenever 0 −→ A −→ B −→ C −→ 0 is an exact sequence in DMod(Λ), so is the corresponding sequence 0 −→ Hom(C, Q) −→ Hom(B, Q) −→ Hom(A, Q) −→ 0 of abelian groups.
1.7.4 Every discrete Λ-module can be embedded as a submodule of an injective discrete Λ-module, i.e., DMod(Λ) ‘has enough injectives’.
1.8 The Complete Group Algebra
Consider a commutative profinite ring R and a profinite group H . We denote the usual abstract group algebra (or group ring) by [RH]. Recall that it consists of all ∈ formal sums h∈H rhh (rh R, where rh is zero for all but a finite number of indices h ∈ H ), with natural addition and multiplication. As an abstract R-module, [RH] isfreeonthesetH . ∼ Assume that H is a finite group. Then [RH] is (as a set) a direct product [RH] = | | [ ] [ ] H R of H copies of R.Ifweimposeon RH the product topology, then RH becomes a topological ring, in fact a profinite ring (since this topology is compact, Hausdorff and totally disconnected). Suppose now that G is a profinite group. Define the complete group algebra (or complete group ring) [[ RG]] to be the inverse limit
[[ ]] = RG lim←− R(G/U) U∈U of the ordinary group algebras [R(G/U)], where U is the collection of all open normal subgroups of G. One can express [[ RG]] as an inverse limit of finite rings [(R/I)(G/U)], where I and U range over the open ideals of R and the open normal subgroups of G, respectively.
G-Modules
If G is a profinite group, a left G-module consists of a topological abelian group M together with a continuous map G × M → M, denoted by (g, a) → ga, satisfying the following conditions: (i) (gh)m = g(hm); (ii) g(m + m ) = gm + gm ; (iii) 1m = m, for all g,h ∈ G, m, m ∈ M, where 1 is the identity of G. 16 1 Preliminaries
For example, if N/K is a Galois extension of fields and G is its Galois group, then N + (the additive group of N) and the group of roots of unity in N (under multiplication) are examples of (discrete) G-modules. Every [[ RG]] -module is naturally a G-module. It is easy to see that a profinite abelian group naturally has the structure of a Zˆ -module, and so one has that a profi- nite G-module is the same as a profinite [[ Zˆ G]] -module.
1.8.1 Denote by DMod(G) the category of all discrete G-modules and G- homomorphisms; then DMod([[ZG]] ) coincides with the subcategory of DMod(G) consisting of the discrete torsion G-modules.
1.8.2 DMod(G) has enough injectives.
Complete Tensor Products
ˆ Denote by R a commutative profinite ring, for example, a quotient of Z: Zp (for some prime number p), or ZCˆ (for some pseudovariety of finite groups C), or Fp (the finite prime field with p elements). Let G be a profinite group and let Λ =[[RG]] be the corresponding complete group algebra. Let A be a profinite right Λ-module, B a profinite left Λ-module, and let M be an R-module. A continuous map ϕ : A × B −→ M is called middle linear if ϕ(a + a ,b)= ϕ(a,b)+ ϕ(a ,b), ϕ(a,b+ b ) = ϕ(a,b)+ ϕ(a,b ) and ϕ(ar,b) = ϕ(a,rb), for all a,a ∈ A, b,b ∈ B, r ∈ R. We say that a profinite R-module T together with a continuous middle linear map A × B −→ T , denoted (a, b) → a ⊗ b,isacomplete tensor product of A and B over Λ if the following universal property is satisfied: If M is a profinite R-module and ϕ : A × B −→ M a continuous middle lin- ear map, then there exists a unique map of R-modules ϕ¯ : T −→ M such that ϕ(a¯ ⊗ b) = ϕ(a,b). The complete tensor product of A and B is unique, up to iso- morphism, and it is denoted by T = A ⊗Λ B. It can be expressed in terms of tensor products of finite abstract modules:
1.8.3 If one writes A as an inverse limit of finite right Λ-modules Ai , and B as an inverse limit of finite left Λ-modules Bi , then A ⊗ B = lim(A ⊗ B ), Λ ←− i Λ j where Ai ⊗Λ Bj is the usual tensor product as abstract Λ-modules. n − − Λ − − 1.9 The Functors ExtΛ( , ) and Torn ( , ) 17
It is useful to recall some basic properties of complete tensor products; they allow us to work with ease without always having to appeal to the definition. 1.8.4 A ⊗Λ − is an additive functor, i.e., it sends finite direct sums to finite direct sums: A ⊗Λ (B1 ⊕ B2) = A ⊗Λ B1 ⊕ A ⊗Λ B2. 1.8.5 A ⊗Λ − is a right exact functor: if
0 → B1 → B2 → B3 → 0 is a short exact sequence of left Λ-modules, then the corresponding sequence of R-modules A ⊗Λ B1 −→ A ⊗Λ B2 −→ A ⊗Λ B3 −→ 0 is exact. 1.8.6 If P is a projective right Λ-module, then the functor P ⊗Λ − is exact. There are similar statements for the functor − ⊗ΛB.
n − − Λ − − 1.9 The Functors ExtΛ( , ) and Torn ( , ) We restrict ourselves to profinite rings of the form Λ =[[RG]] (complete group rings), where R is a commutative profinite ring and G is a profinite group, although most concepts can be described for more general profinite R-algebras Λ in a similar manner.
n − − The Functors ExtΛ( , )
Let A be a profinite Λ-module and B a discrete Λ-module. For each natural number = n n 0, 1,..., one defines a discrete R-module ExtΛ(A, B) as follows. Consider an exact sequence (called a projective resolution of A)
···−→Pn+1 −→ Pn −→···−→P0 −→ A −→ 0 (1.4) where each Pi is a projective Λ-module; such a projective resolution exists because PMod(Λ) has enough projectives. Applying the functor Hom(−,B) we obtain a sequence
fn 0 −→ HomΛ(P0,B)−→···−→HomΛ(Pn,B)−→ HomΛ(Pn+1,B)−→··· n which is not exact in general. Then ExtΛ(A, B) is defined to be the n-th cohomology group of this sequence: n = ExtΛ(A, B) Ker(fn)/Im(fn−1). 18 1 Preliminaries
n − It turns out that ExtΛ(A, B) can also be computed using the functor HomΛ(A, ): consider an injective resolution of B + 0 −→ B −→ Q0 −→···−→Qn −→ Qn 1 −→···
(i.e., this is an exact sequence where each Qi is a discrete injective Λ-module; it exists because DMod(Λ) has enough injectives). Apply HomΛ(A, −) to obtain n 0 n g n+1 0 −→ HomΛ A,Q −→···−→HomΛ A,Q −→ HomΛ A,Q −→··· This sequence is not exact in general and one has
n = n n−1 ExtΛ(A, B) Ker g /Im g . It is rather cumbersome to work directly with these computational definitions. The following properties provide enough information for most purposes. We just write n − n − them for ExtΛ(A, ) (there are similar properties for ExtΛ( ,B)).
0 = 1.9.1 ExtΛ(A, B) HomΛ(A, B). n = ≥ 1.9.2 If B is an injective discrete Λ-module, then ExtΛ(A, B) 0, for n 1. n − 1.9.3 For each n,ExtΛ(A, ) is an additive functor.
1.9.4 If 0 → B1 → B2 → B3 → 0 is a short exact sequence of discrete Λ-modules, then there exist ‘connecting homomorphisms’ = n : n −→ n+1 δ δ ExtΛ(A, B3) ExtΛ (A, B1) (n = 0, 1,...)satisfying the following conditions: (a) For every commutative diagram
0 B1 B2 B3 0
0 B1 B2 B3 0 of discrete Λ-modules with exact rows, the following diagram commutes for every n
n δ n+1 ExtΛ(A, B3) ExtΛ (A, B1)
n δ n+1 ExtΛ(A, B3) ExtΛ (A, B1) (b) The long sequence ···→ n−1 →δ n → n → n →···δ ExtΛ (A, B3) ExtΛ(A, B1) ExtΛ(A, B2) ExtΛ(A, B3) is exact. n − − Λ − − 1.9 The Functors ExtΛ( , ) and Torn ( , ) 19
{ n − } = The above properties of the sequence of functors ExtΛ(A, ) (n 0, 1,...) are sufficient to characterize them, in a sense that one can make precise; it says, for example, that any other sequence of functors of the same type satisfying these prop- { n − } = erties must coincide with ExtΛ(A, ) (n 0, 1,...). One refers to such sequences as being universal. This is often very useful because to understand the functors n − 0 − ExtΛ(A, ) in every dimension n, it suffices to study the behaviour of ExtΛ(A, ).
Λ − − The Functors Torn ( , )
Let A be a profinite right Λ-module and B a profinite left Λ-module. For n = Λ 0, 1,..., one defines a profinite R-module Torn (A, B) as follows. Assume that (1.4) Λ is a projective resolution of the profinite right Λ-module A. Then Torn (A, B) is the n-th homology group of the sequence
fn ···−→Pn+1 ⊗ΛB −→ Pn ⊗ΛB −→···−→P0 ⊗ΛB −→ 0, i.e., Λ = Torn (A, B) Ker(fn−1)/Im(fn). Λ ⊗ − It turns out that one can also calculate Torn (A, B) using the functor A Λ :itis the n-th homology group of the sequence obtained by applying A ⊗Λ− to a projec- n − − tive resolution of B. As in the case of ExtΛ( , ), this computational definition of Λ Torn (A, B) is hard to use; instead one relies on some basic properties that we list next.
Λ = ⊗ 1.9.5 Tor0 (A, B) A ΛB.
Λ = ≥ 1.9.6 If B is a projective profinite left Λ-module, then Torn (A, B) 0, for n 1.
Λ − 1.9.7 For each n,Torn (A, ) is an additive functor.
1.9.8 If 0 → B1 → B2 → B3 → 0 is a short exact sequence of profinite left Λ- modules, then there exist ‘connecting homomorphisms’ = : Λ −→ Λ δ δn Torn+1(A, B3) Torn (A, B1) (n = 0, 1,...)satisfying the following conditions: (a) For every commutative diagram
0 B1 B2 B3 0
0 B1 B2 B3 0 20 1 Preliminaries
of profinite left Λ-modules with exact rows, the following diagram commutes for every n
Λ δ Λ Torn+1(A, B3) Torn (A, B1)
Λ δ Λ Torn+1(A, B3) Torn (A, B1) (b) The long sequence ···→ Λ →δ Λ → Λ → Λ →···δ Torn+1(A, B3) Torn (A, B1) Torn (A, B2) Torn (A, B3) is exact.
{ Λ − } = The sequence Torn+1(A, ) (n 0, 1,...) is universal. Analogous properties Λ − hold for Torn ( ,B).
1.10 Homology and Cohomology of Profinite Groups
ˆ Let R be a commutative profinite ring; usually one is interested in R being Z or one = = of its quotient rings: Zp, ZCˆ, Zπˆ p∈π Zp, Z/nZ, Z/pZ Fp,..., where p is a prime number, π is a set of prime numbers, C is a pseudovariety of finite groups and n is a natural number.
Cohomology of Profinite Groups
Let G be a profinite group. Consider R as a profinite G-module with trivial action: gr = r, for all g ∈ G, r ∈ R. Then R becomes a profinite [[ RG]] -module. Given a discrete [[ RG]] -module A and a natural number n, define the n-th cohomology group H n(G, A) of G with coefficients in A as the discrete R-module n = n H (G, A) Ext[[ RG]] (R, A). Define AG ={a | a ∈ A,ga = a, ∀g ∈ G}. Then AG is an [[ RG]] -submodule of A. We call AG the submodule of fixed points of A. This submodule coincides with Hom[[ RG]] (R, A). It is convenient to restate explicitly the properties mentioned in Sect. 1.9 for { n − } = { n − } = Ext[[ RG]] (R, ) (n 0, 1,...), using the notation H (G, ) (n 0, 1,...). It is a universal sequence of functors and has the following properties.
0 G 1.10.1 H (G, A) = Hom[[ RG]] (R, A) = A . 1.10 Homology and Cohomology of Profinite Groups 21
1.10.2 Each H n(G, −) is an additive functor.
1.10.3 H n(G, Q) = 0 for every injective discrete [[ RG]] -module Q and n ≥ 1.
1.10.4 For each short exact sequence 0 −→ A1 −→ A2 −→ A3 −→ 0 in DMod([[ RG]] ), there exist connecting homomorphisms n n+1 δ : H (G, A3) −→ H (G, A1) for all n ≥ 0, such that the sequence
0 0 0 δ 1 1 0 → H (G, A1) → H (G, A2) → H (G, A3) → H (G, A1) → H (G, A2) →··· is exact; and For every commutative diagram
0 A1 A2 A3 0
α β γ
0 A1 A2 A3 0 in DMod([[ RG]] ) with exact rows, the following diagram commutes for every n ≥ 0
n δ n+1 H (G, A3) H (G, A1)
H n(G,γ ) H n+1(G,α)
n δ n+1 H (G, A3) H (G, A1).
n = n As mentioned above, to compute H (G, A) Ext[[ RG]] (R, A) explicitly one can start, for example, by finding a projective resolution for R, applying to it the functor Hom[[ RG]] (−,A) and then computing the n-th cohomology group of the resulting sequence. In this case, there are standard ways of doing all of this and they are not hard. Rather than specifying all the steps, we just describe the end result; this will permit us to recall easily the definitions of certain special maps that are needed later in the book. For a natural number n,letCn(G, A) consist of all continuous maps
n+1 f : Gn+1 = G × ··· ×G −→ A such that
f(xx0,xx1,...,xxn) = xf (x0,x1,...,xn) ∀x,xi ∈ G. Then Cn(G, A) is a discrete R-module. Consider the sequence of R-modules and R-homomorphisms
+ ∂n 1 + 0 −→ C0(G, A) −→···−→Cn(G, A) −→ Cn 1(G, A) −→··· , 22 1 Preliminaries where n+1 n+1 i ∂ f (x0,x1,...,xn+1) = (−1) f(x0,...,xˆi,...xn+1) i=0
(the symbol xˆi indicates that xi is to be omitted). Then one has the following explicit description:
n = n = n+1 n H (G, A) Ext[[ RG]] (R, A) Ker ∂ /Im ∂ . (1.5) The elements in Ker(∂n+1) are called n-cocycles, and the elements of Im(∂n), n- coboundaries. Using (1.5) one obtains a useful description of, for example, the first cohomology group H 1(G, A). To explain this, recall that a derivation d : G −→ A from a profinite group G to G-module A is a continuous function such that d(xy) = xd(y) + d(x), for all x,y ∈ G. We denote the abelian group of derivations from G to A by Der(G, A).Ifa ∈ A,the map
da : G −→ A given by the formula da(x) = xa−a (x ∈ G) is a derivation, called the inner deriva- tion determined by a. The inner derivations form a subgroup of Der(G, A) which is denoted by Ider(G, A). Then one has
1.10.5 With the notation above, H 1(G, A) = Der(G, A)/Ider(G, A).
Derivations are useful in several other contexts in this book. To explain this recall first that the augmentation ideal (( I G )) of the complete group ring [[ RG]] is the kernel of the continuous ring homomorphism (the augmentation map) ε :[[RG]] −→ R that sends R identically to R, and ε(g) = 1, for every g ∈ G.
1.10.6 (a) (( I G )) is a profinite free R-module on the pointed topological space G − 1 ={x − 1 | x ∈ G}, where 0 is the distinguished point of G − 1. (b) If T is a profinite subspace generating G such that 1 ∈ T , then (( I G )) is gener- ated by the pointed space T − 1 ={t − 1 | t ∈ T }, as an [[ RG]] -module.
Next we see how derivations give rise to homomorphisms. 1.10 Homology and Cohomology of Profinite Groups 23
1.10.7 There is a natural isomorphism
ϕ : Der(G, A) −→ Hom[[ RG]] (( I G )) , A defined by (ϕ(d))(x − 1) = d(x) (d ∈ Der(G, A), x ∈ G).
Special Maps in Cohomology
Let H be a closed subgroup of a profinite group G.Adiscrete[[ RG]] -module A is also a discrete [[ RH]] -module. For each natural number n, one has a homomorphism = G : n −→ n Res ResH H (G, A) H (H, A) called restriction. One can define these mappings using the universality of the se- quence of functors H n(H, −) by specifying Res in dimension zero: it is just the inclusion AG → AH . Or, more explicitly, one can describe Res for each dimen- sion if we use the formula (1.5): Let σ : Gn+1 −→ A be an n-cocycle represent- inganelementσ¯ ∈ H n(G, A); then a representative n-cocycle ρ : H n+1 −→ A of Res(σ)¯ ∈ H n(H, A) is given by
ρ(x0,...,xn) = σ(x0,...,xn), (x0,...,xn ∈ H).
Assume now that K is a closed normal subgroup of a profinite group G, and let A be a discrete [[ RG]] -module. Then AK becomes a discrete [[ R(G/K)]] -module in a natural way:
(xK)(a) = xa, x ∈ G, a ∈ AK .
Define a homomorphism
= G/K : n K −→ n Inf InfG H G/K, A H (G, A), called inflation, as follows. In dimension n = 0, define
Inf : H 0 G/K, AK = AK G/K −→ H 0(G, A) = AG to be the identity mapping. Assume n>0, and let σ ∈ Cn(G/K, AK ) represent an element σ¯ of H n(G/K, AK ), i.e., σ : (G/K)n+1 −→ AK is an n-cocycle. Then Inf(σ)¯ is represented by the n-cocycle
ρ : Gn+1 −→ A given by
ρ(x0,...,xn) = σ(x0K,...,xnK). 24 1 Preliminaries
Homology of Profinite Groups
Let G be a profinite group and let B be a profinite right [[ RG]] -module. Define the n-th homology group Hn(G, B) of G with coefficients in B by the formula = [[ RG]] Hn(G, B) Torn (B, R). In dimension zero one has
1.10.8 ∼ def H0(G, B) = B/B((IG)) = B/ bg − b | b ∈ B,g ∈ G = BG.
Duality Homology-Cohomology
Given a Λ-module M (discrete or profinite), consider the abelian group (the Pon- tryagin dual of M) ∗ M = Hom(M, Q/Z) of all continuous homomorphism from M to Q/Z (as abelian groups) with the compact-open topology. M∗ is profinite if M is discrete, and it is discrete if M is profinite. Define a right action of Λ on M∗ by
f λ (m) = f(λm) f ∈ M∗,λ∈ Λ,m ∈ M .
Then M∗ becomes a right Λ-module. Also, if M is a projective profinite Λ-module, then M∗ becomes an injective discrete Λ-module, and vice versa.
1.10.9 Let G be a profinite group and let B be a right [[ Zˆ G]] -module. Then
n ∗ Hn(G, B) and H G, B (n ∈ N) are Pontryagin dual, where B∗ denotes the Pontryagin dual of B.
This duality permits the automatic translation of information in cohomology to homology, or vice versa. For example, one can easily state the dual results of 1.10.1Ð 1.10.4 for homology groups. The dual of the homomorphism Res described above is called corestriction: = H : −→ Cor CorG Hn(H, B) Hn(G, B). Similarly, the dual of Inf is called coinflation, and denoted Coinf. In dimension 1, we have the following useful descriptions when considering con- venient coefficient modules: 1.11 (Co)homological Dimension 25
1.10.10 (a) Let G be a profinite group and let G act on Zˆ trivially. Then there is a natural isomorphism ˆ ∼ H1(G, Z) = G/[G, G].
(b) Let G be a pro-p group and let G act on Fp trivially. Then there is a natural isomorphism ∼ H1(G, Fp) = G/Φ(G), where Φ(G) = Gp[G, G] is the Frattini subgroup of G.
(Co)induced Modules and Shapiro’s Lemma
Let H ≤ G be profinite groups and R a commutative profinite ring. Let B be a profinite right [[ RH]] -module. Then G = ⊗ [[ ]] IndH (B) B [[ RH]] RG has a natural structure as a right profinite [[ RG]] -module and it is called an induced module.Dually,ifA is a left discrete [[ RH]] -module, one defines the coinduced module
G = [[ ]] CoindH (A) Hom[[ RH]] RG ,A . It has a natural structure as a left discrete [[ RG]] -module. The following result is traditionally known as ‘Shapiro’s Lemma’.
1.10.11
(a)
G =∼ ≥ Hn G, IndH (B) Hn(H, B) (n 0). (b) Dually,
n G =∼ n ≥ H G, CoindH (A) H (H, A) (n 0).
1.11 (Co)homological Dimension
Let G be a profinite group and let p be a prime number. If A is an abelian group, then Ap denotes its p-primary component (the subgroup of those elements of A of n order p ,forsomen). If A = Ap, we say that A is p-primary. The cohomological p-dimension
cdp(G) 26 1 Preliminaries
k of G is the smallest non-negative integer n such that H (G, A)p = 0 for all k>n ˆ and A ∈ DMod([[ ZG]] ), if such an n exists; otherwise we say that cdp(G) =∞. Dually, one can define the homological dimension of G, denoted hdp(G).Wehave cdp(G) = hdp(G).
1.11.1 If Gp is a p-Sylow subgroup of a profinite group G, then cdp(G) = cdp(Gp).
1.11.2 Let G be a profinite group and let n be a fixed natural number. Then cdp(G) ≤ n if and only if there exists a projective resolution
0 → Pn → Pn−1 →···→P0 → Fp → 0 of Fp in PMod([[ FpG]] ) of length n.
It is often important to compare cohomological dimensions of groups and sub- groups. The following result is particularly useful.
1.11.3 Let H ≤c G be profinite groups. Then cdp(H ) ≤ cdp(G). If cdp(G) is finite and H is open in G, then cdp(H ) = cdp(G).
In the case of pro-p groups, freeness can be characterized in terms of cohomo- logical dimension (as is the case for abstract groups):
1.11.4 Let G beapro-p group. Then G is free pro-p if and only if cdp(G) ≤ 1.
1.11.5 Let G be a profinite group. Then G is projective if and only if cdp(G) ≤ 1 for every prime number p. Part I Basic Theory
In this first part of the book (the next five chapters) we develop the fundamentals of profinite graphs and we relate them to profinite groups. We view a graph as a collection Γ = V ∪ E of vertices V and edges E together with functions d0,d1 defined on Γ that assign to each m ∈ Γ vertices, its ‘origin’ d0(m) and ‘terminus’ d1(m) (a vertex is assumed to be its own origin and terminus). We call such a graph ‘profinite’ if Γ has in addition the structure of a profinite space (i.e., a compact, Hausdorff and totally-disconnected topological space) in such a way that V is a closed subspace of Γ and d0 and d1 are continuous functions. Profinite graphs arise naturally in connection with profinite groups and origi- nally they were introduced precisely as tools in the study of those groups. Here we develop an analogue to the BassÐSerre theory of (abstract) groups and (abstract) graphs. Our development is self-contained and in principle it does not require any knowledge of Bass and Serre’s elegant theory of groups acting on trees (Serre 1980). However, we will frequently mention connections with abstract groups and graphs and the BassÐSerre results; and, in fact, as we will see later in Part III of this book, the interconnections with the BassÐSerre approach will be crucial for applications to abstract groups. In Chap, 2 the concept of a π-tree is introduced, where π is a set of prime num- bers; a paradigm of a π-tree is the Cayley graph of a free pro-π group with respect to a basis. Profinite graphs are just inverse limits of finite graphs, but a π-tree need not be the inverse limit of finite trees. In Chaps. 2Ð4 we study profinite groups acting on π-trees, the fundamental group of a profinite graph and simple connectivity of profi- nite graphs. Unlike the situation for abstract graphs, the class of simply connected profinite graphs does not coincide with the class of π-trees. Free products of profinite groups are studied in Chap. 5. The need to describe the structure of closed subgroups of a free product of even just two profinite groups leads to the concept of a collection of closed subgroups of a profinite group ‘contin- uously indexed by a profinite space’, and to free products of such collections. More general ‘free constructions’ of profinite groups (e.g., amalgamated prod- ucts, HNN extensions or, more generally, fundamental groups of graphs of profinite 28 groups) and standard profinite graphs associated with these constructions are con- sidered in Chap. 6. A free profinite product of profinite groups always contains the free factors. But, unlike the abstract case, a profinite fundamental group need not contain the vertex groups of the corresponding graph of profinite groups; when they do, we call the fundamental groups ‘proper’. This difficulty may already be encountered in the case of an amalgamated product of two groups amalgamating a common subgroup. As wewillseeinPartIII of this book, in certain applications to abstract groups, one of the key points is to prove that appropriate profinite fundamental groups are proper. Chapter 2 Profinite Graphs
Unless otherwise specified, in this chapter C is a pseudovariety of finite groups, i.e., a nonempty class of finite groups closed under subgroups, quotients and finite direct products.
2.1 First Notions and Examples
A profinite graph is a profinite space Γ with a distinguished nonempty subset V(Γ), the vertex set of the graph Γ , and two continuous maps
d0,d1 : Γ → V(Γ) whose restrictions to V(Γ) are the identity map idV(Γ) (to simplify the notation, we sometimes write dim, rather than di(m) (m ∈ Γ,i = 0, 1)). This implies that the distinguished subset V(Γ)is necessarily closed. The elements of V(Γ)are called the vertices of Γ , the elements of E(Γ ) = Γ − V(Γ)are the edges of Γ , and d0(e) and d1(e) are the initial and terminal vertices of an edge e, respectively (also called the origin and terminus of e). An edge e with d0(e) = d1(e) = v is called a loop or a loop based at v. We refer to d0 and d1 as the incidence maps of the graph Γ . Observe that a profinite graph is also a graph in the usual sense, or, more pre- cisely, an oriented graph (see Appendix A), if we dispense with the topology. The set of edges E(Γ ) of a profinite graph Γ need not be a closed subset of Γ .IfE(Γ ) is closed (and therefore compact), it is enough to check the continuity of d0 and d1 on V(Γ)and E(Γ ) separately, since then V(Γ)and E(Γ ) are disjoint and clopen. Associated with each edge e of Γ we introduce symbols e1 and e−1. We identify 1 −1 e with e. Define incidence maps for these symbols as follows: d0(e ) = d1(e) −1 and d1(e ) = d0(e). Given vertices v and w of Γ ,apath pvw from v to w is a ε1 εm ≥ ∈ =± = finite sequence e1 ,...,em , where m 0, ei E(Γ ), εi 1 (i 1,...,m)such ε1 = εm = εi = εi+1 = − that d0(e1 ) v, d1(em ) w and d1(ei ) d0(ei+1 ) for i 1,...,m 1. Such a path is said to have length m. Observe that a path is always meant to be finite. The underlying graph of the path pvw consists of the edges e1,...,em and their
© Springer International Publishing AG 2017 29 L. Ribes, Profinite Graphs and Groups, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics 66, DOI 10.1007/978-3-319-61199-0_2 30 2 Profinite Graphs vertices di(ej ) (i = 0, 1; j = 1,...,m). The path pvw is called reduced if whenever ei = ei+1, then εi = εi+1, for all i = 1,...,m− 1.
Example 2.1.1 (a) A finite abstract graph Γ (see Appendix A) with the discrete topology is a profinite graph. (b) Let N ={0, 1, 2,...} and N˜ ={˜n | n ∈ N} be copies of the set of natural numbers (with the discrete topology). Define I = N ∪. N˜ ∪. {∞} to be the one-point compactification of the space N ∪. N˜ . Recall that then in the topology of I each set {n} and {˜n} is open (n ∈ N), and the basic open neighbour- hoods of ∞ are the complements of finite subsets of N ∪. N˜ . Clearly I is a profinite space. We make I into a profinite graph by setting V(I)= N ∪. {∞}, E(I) = N˜ , d0(n)˜ = n, d1(n)˜ = n + 1, for n˜ ∈ E(I), and di(n) = n,forn ∈ V(I)(i= 1, 2). 0 123 ∞ ••••··· • 0˜ 1˜ 2˜ Observe that in this case the subset of edges E(I) is open, but not closed in I . (c) Let p be a prime number and let Zp be the additive group of the ring of p-adic integers. Define a graph
Γ = Γ Zp, {1} with set of vertices V = V(Γ)= Zp and whose set of edges is E = E(Γ ) ={(α, 1) | α ∈ Zp}. Then V(Γ)and E(Γ ) are profinite spaces. We define the topology of Γ = V(Γ)∪. E(Γ ) to be the disjoint topology: a subset A of Γ is open if and only if A ∩ V is open in V and A ∩ E is open in E. One easily sees that Γ is a profinite space. Observe that the subset of edges E = E(Γ ) of Γ is both open and closed (clopen) in the topology of Γ . The incidence maps are the continuous maps
di : Γ −→ V(i= 0, 1) defined as d0(α) = α, d0(α, 1) = α and d1(α) = α, d1(α, 1) = α + 1 (α ∈ Zp). With these definitions Γ becomes a profinite graph. [This is an instance of profinite graphs obtained from profinite groups in a standard manner, the so-called Cayley graphs: see Example 2.1.12.] The subgroup of integers Z = 1 is dense in Zp and the topology of Z induced by the topology of Zp is the discrete topology. Let Γ Z, {1} = α ∈ V(Γ) α ∈ Z ∪. (α, 1) α ∈ Z . Then Γ(Z, {1}) is an abstract discrete graph −2 −1 0 12 ··· •••••··· (−2, 1)(−1, 1)(0, 1)(1, 1) which is dense in the profinite graph Γ = Γ(Zp, {1}). 2.1 First Notions and Examples 31
More generally, let β be a fixed element of Zp, and define Γ Z + β,{1} = α ∈ V(Γ) α ∈ Z + β ∪. (α, 1) ∈ E(Γ ) α ∈ Z + β . Then Γ(Z + β,{1}) is an abstract discrete graph β − 2 β − 1 β β + 1 β + 2 ··· •••••··· (β − 2, 1)(β− 1, 1)(β,1)(β+ 1, 1) which is also dense in the profinite graph Γ = Γ(Zp, {1}). Note that Γ(Zp, {1}) is a disjoint union of uncountably many abstract discrete graphs of the form Γ(Z + β,{1}): . Γ Zp, {1} = Γ Z + βλ, {1} , λ∈Λ where {βλ | λ ∈ Λ} is a complete set of representatives of the cosets of the subgroup Z in the group Zp.
Let Γ and be profinite graphs. A qmorphism or a quasi-morphism of profinite graphs or a map of graphs α : Γ → is a continuous map such that dj (α(m)) = α(dj (m)), for all m ∈ Γ and j = 0, 1. If in addition α(e) ∈ E() for every e ∈ E(Γ ), we say that α is a morphism. The composition of qmorphisms of profinite graphs is again a qmorphism, so that profinite graphs and their qmorphisms form a category. Similarly profinite graphs and their morphisms form a category. If α is a surjective (respectively, injective, bi- jective) qmorphism, we say that α is an epimorphism (respectively, monomorphism, isomorphism). An isomorphism α : Γ → Γ of the graph Γ to itself is called an au- tomorphism. Note that a monomorphism of graphs sends edges to edges, and hence it is always a morphism. A nonempty closed subset Γ of a profinite graph is called a profinite subgraph of if whenever m ∈ Γ , then dj (m) ∈ Γ(j= 0, 1). The equality dj (α(m)) = α(dj (m)) (j = 0, 1; m ∈ Γ)implies that a qmorphism of profinite graphs maps vertices to vertices. However, the next example shows that a qmorphism can map an edge to a vertex.
Example 2.1.2 (Subgraph collapsing) Let be a profinite subgraph of a profi- nite graph Γ . Consider the natural continuous map α : Γ → Γ/ to the quo- tient space Γ/ with the quotient topology [the points of Γ/ are the equiva- lence classes of the relation ∼ on Γ defined as follows: if m, m ∈ Γ , then m ∼ m if and only if either m = m or m, m ∈ ;ifm ∈ Γ , then α(m) is the equiva- lence class of m; a subset U of Γ/ is open if α−1(U) is open in Γ ]. Define a structure of profinite graph on the space Γ/ as follows: V(Γ/)= α(V (Γ )), d0(α(m)) = α(d0(m)), d1(α(m)) = α(d1(m)), for all m ∈ Γ . Then clearly α is a qmorphism of graphs and Γ/ becomes a quotient graph of Γ . We shall say that Γ/is obtained from Γ by collapsing to a point. Observe that α maps any edge of Γ which is in toavertexofΓ/. 32 2 Profinite Graphs
We note that if α : Γ → is an epimorphism of profinite graphs, then has the quotient topology (i.e., for A ⊆ , one has that A is open in if and only if α−1(A) is open in Γ ), since Γ and are compact Hausdorff spaces. We then say that is a quotient graph of Γ and α is a quotient qmorphism of graphs. If Γ is a profinite graph and ϕ : Γ → Y is a continuous surjection onto a profinite space Y , there is no assurance that there exists a profinite graph structure on Y so that ϕ is a qmorphism of graphs. The following construction provides necessary and sufficient conditions for this to happen.
Construction 2.1.3 Let Γ be a profinite graph and let ϕ : Γ → Y be a continuous surjection onto a profinite space Y . Then we construct a quotient qmorphism of graphs
ϕ˜ : Γ → Γϕ with the following properties.
(a) There is a continuous surjection of topological spaces ψϕ : Γϕ → Y such that the diagram ϕ Γ Y
ϕ˜ ψϕ Γϕ commutes. (b) If Y admits a profinite graph structure so that ϕ is a qmorphism, then ψϕ is an isomorphism of profinite graphs. (c) Consequently, there exists a profinite graph structure on Y such that ϕ is a qmorphism of graphs if and only if whenever m, m ∈ Γ with ϕ(m) = ϕ(m ),
then ϕd0(m) = ϕd0(m ) and ϕd1(m) = ϕd1(m ). If this is the case, then that structure is unique (isomorphic to Γϕ) and the incidence maps of Y are defined by diϕ(m) = ϕdi(m) (m ∈ Γ,i = 0, 1). (d) If E(Γ ) is a closed subset of Γ and ϕ(E(Γ )) ∩ ϕ(V (Γ )) =∅, then ϕ˜ is a morphism of profinite graphs and ψϕ(E(Γϕ)) ∩ ψϕ(V (Γϕ)) =∅.
To construct Γϕ, define a map ϕ˜ : Γ −→ Y × Y × Y by
ϕ(m)˜ = ϕ(m),ϕd0(m), ϕd1(m) (m ∈ Γ).
Let Γϕ =˜ϕ(Γ ). Then Γϕ admits a unique graph structure such that ϕ˜ : Γ → Γϕ is a ˜ ˜ qmorphism of graphs, namely one is forced to define the incidence maps d0 and d1 of Γϕ by
˜ d0 ϕ(m),ϕd0(m), ϕd1(m) = ϕd0(m), ϕd0(m), ϕd0(m) (m ∈ Γ) 2.1 First Notions and Examples 33 and
˜ d1 ϕ(m),ϕd0(m), ϕd1(m) = ϕd1(m), ϕd1(m), ϕd1(m) (m ∈ Γ) (one easily checks that these are well defined, and that ϕ˜ is indeed a qmorphism of profinite graphs). Next note that there exists a unique map ψϕ : Γϕ → Y such that ψϕϕ˜ = ϕ, namely, ψϕ(ϕ(m), ϕd0(m), ϕd1(m)) = ϕ(m). If Y is a profinite graph and ϕ is a qmorphism of profinite graphs, then ψϕ is an isomorphism of graphs because in this case the map ρ : Y → Γϕ given by ρϕ(m) = (ϕ(m), ϕd0(m), ϕd1(m)) is a well-defined qmorphism of graphs and it is inverse to ψϕ. This proves properties (a) and (b). Property (c) is clear. Property (d) is easily verified.
Before stating the following proposition we recall briefly the concept of an in- verse limit in the category of graphs (see Sect. 1.1). Let (I, ) be a directed partially ordered set (a directed poset). An inverse system of profinite graphs {Γi,ϕij ,I} over the directed poset I consists of a collection of profinite graphs Γi indexed by I and qmorphisms of profinite graphs ϕij : Γi → Γj , whenever i j, in such a way that ϕii = Idi , for all i ∈ I , and ϕjkϕij = ϕik, whenever i j k. The inverse limit (or projective limit) of such a system = Γ lim←− Γi i∈I = is the subset of i∈I Γi consisting of those tuples (mi) with ϕij (mi) mj , when- ever i j. Such an inverse limit is in a natural way a profinite graph whose space of vertices is = V(Γ) lim←− V(Γi). i∈I
Observe that the natural projections ϕi : Γ → Γi are qmorphisms of profinite graphs. Note that if each ϕij is a morphism, then so are the canonical projections ϕi . Let Γ be a profinite graph and consider the set R of all open equivalence relations R on the set Γ (i.e., the equivalence classes xR are open for all x ∈ Γ ). For R ∈ R, denote by ϕR : Γ → Γ/R the corresponding quotient map as topological spaces. One defines a partial ordering on R as follows: for R1,R2 ∈ R, we say that : → R1 R2 if there exists a map ϕR1,R2 Γ/R1 Γ/R2 such that the diagram
Γ/R1 ϕR1
Γ ϕR1,R2
ϕR2 Γ/R2 commutes. Then (cf. RZ, Theorem 1.1.2) (R, ) is in fact a directed poset, { } R Γ/R,ϕR1,R2 is an inverse system over , and, as topological spaces, the collection 34 2 Profinite Graphs of quotient maps {ϕ | R ∈ R} induces a homeomorphism from Γ to lim ∈RΓ/R; R ←− R in fact we identify these two spaces by means of this homeomorphism and write = Γ lim←− Γ/R. (2.1) R∈R Consider now the subset R of R consisting of those R ∈ R such that Γ/R admits a graph structure (which is unique according to part (c) of Construction 2.1.3)so that ϕR : Γ → Γ/R is a qmorphism of profinite graphs. We check next that the poset (R , ) is directed. Indeed, let R1,R2 ∈ R . Since R is directed, there exists an R ∈ R such that R R1,R2.LetϕR : Γ → Γ/R be the corresponding quotient : → = ˜ map. Let ΓϕR and ϕR Γ ΓϕR be as in Construction 2.1.3. Then ΓϕR Γ/R, ˜ −1 where R is the equivalence relation on Γ whose equivalence classes are {ϕR (x) | ∈ } ˜ ∈ R ˜ ˜ x ΓϕR . Clearly R and R R; hence R R1,R2, as needed. ∈ R : → Observe that if R1,R2 and R1 R2, then the map ϕR1,R2 Γ/R1 Γ/R2 { } is in fact a qmorphism of finite graphs. Therefore the collection Γ/R,ϕR1,R2 of all finite quotient graphs of Γ is an inverse system of finite graphs and qmorphisms over the directed poset R .
Proposition 2.1.4 Let Γ be a profinite graph. (a) Γ is the inverse limit of all its finite quotient graphs: = Γ lim←− Γ/R. R∈R Consequently = V(Γ) lim←− V(Γ/R). R∈R (b) If the subset E(Γ ) of edges of Γ is closed, then a directed subposet R of R ∈ R : can be chosen so that whenever R1,R2 with R1 R2, then ϕR1,R2 Γ/R1 → Γ/R2 is a morphism of graphs and = Γ lim←− Γ/R. R∈R Consequently, = = V(Γ) lim←− V(Γ/R) and E(Γ ) lim←− E(Γ/R). R∈R R∈R
Proof (a) In view of (2.1) one simply has to show that R is cofinal in R, i.e., one has to show that whenever R ∈ R, there exists an R ∈ R with R R. But this is clear from property (a) of Construction 2.1.3. (b) Suppose that E(Γ ) is closed. Then Γ = V(Γ)∪. E(Γ ) and V(Γ)and E(Γ ) are clopen subsets of Γ .LetR˜ be the subset of R consisting of those equiva- lence relations R ∈ R whose equivalence classes xR are contained in either E(Γ ) 2.1 First Notions and Examples 35 or V(Γ); this implies that if ϕR : Γ → Γ/R is the canonical projection, then ˜ ϕR(V (Γ )) ∩ ϕR(E(Γ )) =∅. Then one shows that R is cofinal in R, so that = Γ lim←− Γ/R. R∈R˜ One can argue now as in part (a); we just indicate the main points: let R be the subset of R˜ consisting of those equivalence relations R such that Γ/R has the structure of a graph in such a way that ϕR : Γ → Γ/R is a morphism of profinite graphs; note that R is also a subset of R ; using property (d) of Construction 2.1.3 one shows that R is cofinal in R˜ , and hence the result easily follows as above.
Lemma 2.1.5 Let {Γi,ϕij ,I} be an inverse system of profinite graphs and qmor- phisms over a directed poset I , and set = Γ lim←− Γi. (2.2) i∈I Let ρ : Γ → be a qmorphism into a finite graph . Then there exists a k ∈ I such that ρ factors through Γk, i.e., there exists a qmorphism ρ : Γk → such that
ρ = ρ ϕk, where ϕk : Γ → Γk is the projection.
Proof For i ∈ I denote by Ri the set of all equivalence relations R of Γi such that the quotient Γi/R is a finite discrete graph and the natural projection Γi → Γi/R is a qmorphism. Define an ordering on the set of pairs A = (i, R) i ∈ I,R ∈ Ri by setting (i, Ri) (j, Rj ),ifi j and (ϕij × ϕij )(Ri) ⊆ Rj . Let us prove that (A, ) is a directed poset. Fix i, j ∈ I and Ri ∈ Ri , Rj ∈ Rj . Since I is a di- rected poset, there exists some k ∈ I with k i, j. By Proposition 2.1.4, Γk is the inverse limit of all its finite quotient graphs; therefore there exists an Rk ∈ Rk with (ϕki × ϕki)(Rk) ⊆ Ri and (ϕkj × ϕkj )(Rk) ⊆ Rj , so that (k, Rk) (i, Ri), (j, Rj ), as needed. Now it is easy to see that = Γ lim←− Γi/R. (i,R)∈A
Thus from now on we may assume that each Γi in the decomposition (2.2)isfinite. Assume first that each projection ϕi : Γ → Γi is surjective. Let S be the equiva- lence relation on Γ whose equivalence classes are the clopen sets ρ−1(m), m ∈ ; then Γ/S = and ρ is the natural projection Γ → Γ/S. Similarly, for i ∈ I , let Si be the equivalence relation on Γ whose equivalence classes are the clopen −1 ∈ → = sets ϕi (m), m Γi , so that ϕi is the natural projection Γ Γ/Si . Since Γ lim ∈ Γ , we have that S is the trivial equivalence relation, i.e., S = D, ←− i I i i∈I i i∈I i where D is the diagonal subset of Γ × Γ . Note that S and Si (i ∈ I)are clopen sub- sets of Γ × Γ . Hence, it follows from the compactness of Γ × Γ that there exists ⊆ a finite subset F of I such that j∈F Sj S. Since the poset I is directed, there 36 2 Profinite Graphs ∈ ⊆ ⊆ exists a k I with Sk j∈F Sj S. This means that there exists a qmorphism of graphs ρk : Γk = Γ/Sk → = Γ/S such that ρ = ρkϕk.
Consider now a general ϕi . By the above, there exists some k ∈ I andaqmor- phism of graphs ρk : ϕk (Γ ) → such that ρ = ρk ϕk . Since Γk is finite, there exists a k k such that ϕkk (Γk) ⊆ ϕk (Γ ). Then ρ = ρk ϕkk is the required qmor- phism.
An alternative proof of Lemma 2.1.5 above can be obtained along the lines of the proof of Lemma 1.1.16 in RZ. A profinite graph Γ is said to be connected if whenever ϕ : Γ → A is a qmor- phism of profinite graphs onto a finite graph, then A is connected as an abstract graph (see Sect. A.1 in Appendix A).
Proposition 2.1.6 (a) Every quotient graph of a connected profinite graph is connected. (b) If = Γ lim←− Γi i∈I
and each Γi is a connected profinite graph, then Γ is a connected profinite graph. (c) Let Γ be a connected profinite graph. If |Γ | > 1, then Γ has at least one edge. Furthermore, if the set of edges E(Γ ) of Γ is closed in Γ , then for any vertex v ∈ V(Γ), there exists an edge e ∈ E(Γ ) such that either v = d0(e) or v = d1(e). (d) Let Γ be a profinite graph, and let be a connected profinite subgraph of Γ . Consider the quotient graph Γ/ obtained by collapsing to a point and let α : Γ → Γ/ be the natural projection. Then the inverse image Λ˜ = α−1(Λ) in Γ of a connected profinite subgraph Λ of Γ/is a connected profinite sub- graph.
Proof Part (a) is obvious. Let A be a finite quotient graph of Γ . Then (see Lemma 2.1.5) there exists an i ∈ I such that A is also a quotient graph of Γi .It follows that A is connected, proving (b). To check the first assertion in (c) observe that by Proposition 2.1.4, Γ has a finite quotient graph with at least two elements; since such a finite quotient graph is connected, it has at least one edge, and hence so does Γ . To check the second assertion in (c), write Γ as an inverse limit Γ = lim ∈ Γ of finite quotient graphs ←− i I i Γi in such a way that = E(Γ ) lim←− E(Γi) i∈I
(see Proposition 2.1.4(b)). For i ∈ I ,letϕi : Γ → Γi denote the canonical pro- jection, and if i, j ∈ I with i j,letϕij : Γi → Γj denote the canonical mor- phism. Put vi = ϕi(v) (i ∈ I). Since Γi is a connected finite graph, the set Si = 2.1 First Notions and Examples 37
−1 ∪ −1 d0 (vi) d1 (vi) of edges of Γi starting or ending at vi is nonempty; moreover, ϕij (Si) ⊆ Sj . Hence the collection {Si}i∈I is an inverse system of nonempty finite sets. Thus = ∅ lim←− Si i∈I (see Sect. 1.1). Let e ∈ lim ∈ S . Then e is an edge of Γ with either d (e) = v or ←− i I i 0 d1(e) = v. (d) This is clear if Γ is finite. Write = Γ lim←− Γi, i∈I where each Γi is a connected finite quotient graph of Γ (see Proposition 2.1.4(a)). Let i be the image of in Γi under the canonical projection. Then = = lim←− i and Γ/ lim←− Γi/i. i∈I i∈I ˜ Let Λi be the image of Λ in Γi/i , and denote by Λi its inverse image in Γi . Since ˜ ˜ ˜ Λ = lim ∈ Λ , Λ is connected according to part (b). ←− i I i
Lemma 2.1.7 (a) Let D be an abstract subgraph of a profinite graph Γ . Then the topological closure D¯ of D in Γ is a profinite graph. If D is connected as an abstract graph (see Sect. A.1 in Appendix A), then D¯ is a connected profinite graph. { | ∈ } (b) Let j j J be a collection of connected profinite subgraphs of a profinite = ∅ = graph Γ . If j∈J j , then j∈J j is connected. ¯ Proof To prove (a), let m ∈ D. By the continuity of di , di(m) ∈ V(D)(i= 1, 2), so that D¯ is a (profinite) graph with V(D)¯ = V(D).Ifϕ : D¯ → A is a qmorphism of profinite graphs onto a finite graph, then ϕ(D)¯ = ϕ(D) = A by continuity. Since D is a connected abstract graph, one easily checks that ϕ(D) is a finite connected graph; hence D¯ is a connected profinite graph. This proves (a). For part (b) note that if α : → A is a qmorphism onto a finite graph A, then ∈ = α(j ) is a connected finite subgraph of A(j J). Since A j∈J α(j ), and = ∅ j∈J α(j ) , it follows that A is a connected abstract graph.
Example 2.1.8 (A connected profinite graph which is not connected as an abstract graph and with a vertex with no edge beginning or ending at it) Let I be the graph considered in Example 2.1.1(b): I = N ∪. N˜ ∪. {∞} is the one-point compactification of a disjoint union of two copies N and N˜ ={˜n | n ∈ N} of the natural numbers; ˜ V(I)= N ∪. {∞}, E(I) = N, d0(n)˜ = n, d1(n)˜ = n + 1forn˜ ∈ E(I), and di(n) = n for n ∈ V(I)(i= 1, 2). 0 123 ∞ ••••··· • 0˜ 1˜ 2˜ 38 2 Profinite Graphs
Then I is a connected profinite graph; to see this consider the connected finite graphs In 0 1 2 n − 1 n • • • ··· • • 0˜ 1˜ n− 1 ˜ ˜ with vertices V(In) ={0, 1, 2,...,n} and edges E(In) ={0, 1,...,n − 1} such that ˜ ˜ d0(i) = i, d1(i) = i + 1(i = 0,...,n− 1) and dj (i) = i (i = 0,...,n; j = 0, 1). If n ≤ m, define ϕm,n : Im → In to be the map of graphs that sends the segment [0,n] identically to [0,n], and the segment [n, m] to the vertex n. Then (In,ϕm,n) is an inverse system of graphs, and = I lim←− In, n∈N where ∞=(n)n∈N. Hence I is a connected profinite graph. We observe that there is no edge e of I which has ∞ as one of its vertices; and so I is not connected as an abstract graph.
Lemma 2.1.9 Let Γ = Γ1 ∪. Γ2 be a profinite graph which is the disjoint union of two open profinite subgraphs Γ1 and Γ2; then Γ is not connected. In particular, a profinite graph that contains two different vertices and no edges is not connected.
Proof Collapse Γ1 to a point v1 and Γ2 to a different point v2 (see Example 2.1.2), ˜ to get a disconnected finite quotient graph Γ ={v1}∪. {v2} consisting of two vertices and no edges.
A maximal connected profinite subgraph of a profinite graph Γ is called a con- nected profinite component of Γ .
Proposition 2.1.10 Let Γ be a profinite graph. (a) Let m ∈ Γ . Then there exists a unique connected profinite component of Γ con- taining m, which we shall denote by Γ ∗(m). (b) Any two connected profinite components of Γ are either equal or disjoint. (c) Γ is the union of its connected profinite components.
Proof Part (c) follows from (a). Part (b) follows from (a) and Lemma 2.1.7(b). To prove (a) observe first that the result is obvious if Γ is finite. By Proposi- tion 2.1.4, Γ can be represented as an inverse limit lim ∈ Γ of finite quotient ←− i I i graphs. For i ∈ I ,letϕi : Γ → Γi denote the projection. Since the image of a ∗ connected profinite graph is connected, the graphs Γ (ϕi(m)) form an inverse sys- i ∗ tem. It suffices to show that the profinite subgraph lim ∈ Γ (ϕ (m)) of Γ is the ←− i I i i connected profinite component of Γ containing m. This profinite subgraph is con- nected by Proposition 2.1.6(b). If Γ is a connected profinite subgraph of Γ contain- ∗ ing m, then Γ = lim ∈ ϕ (Γ ). Therefore ϕ (Γ ) ⊆ Γ (ϕ (m)) for all i ∈ I . Hence ←− i I i i i i 2.1 First Notions and Examples 39
∗ ∗ Γ ⊆ lim ∈ Γ (ϕ (m)); therefore lim ∈ Γ (ϕ (m)) is maximal connected contain- ←− i I i i ←− i I i i ing m, as desired. The uniqueness of connected profinite components containing m follows from Lemma 2.1.7(b).
Exercise 2.1.11 (a) Let be a profinite graph. Define the space of connected profinite components of as a quotient space /∼, where ∼ is the equivalence relation defined ∗ ∗ as follows: m1 ∼ m2 if and only if (m1) = (m2). Prove that /∼ is a profinite space. [Hint: write as an inverse limit of finite quotient graphs.] (b) Let be a profinite subgraph of a profinite graph Γ . Define the operation of collapsing the connected profinite components of to points as a natural map- ping to the quotient space Γ/∼, where ∼ is the equivalence relation defined as ∗ ∗ follows: m1 ∼ m2 if m1 = m2,form1,m2 ∈ Γ − ,or (m1) = (m2) for m1,m2 ∈ . Prove that Γ/∼ is a profinite quotient graph of Γ .
Example 2.1.12 (The Cayley graph) Let G be a profinite group (whose operation is denoted as multiplication and whose identity element is denoted by 1) and let X be a closed subset of G. Put X˜ = X ∪{1}. Define the Cayley graph Γ(G,X)of G with respect to the subset X as follows: Γ(G,X)= G × X,˜ where G × X˜ has the product topology. Define the space of vertices of Γ(G,X) to be V(Γ(G,X))={(g, 1) | g ∈ G}. We identify this space of vertices with G by means of the homeomorphism (g, 1) → g(g∈ G). Finally, the incidence maps
˜ d0,d1 : Γ(G,X)= G × X −→ V Γ(G,X) = G are defined by
d0(g, x) = g and d1(g, x) = gx, g ∈ G, x ∈ X ∪{1} .
Clearly d0 and d1 are continuous and they are the identity map when restricted to V(Γ(G,X))={(g, 1) | g ∈ G}=G. Therefore the Cayley graph Γ(G,X)is a profinite graph. Note that the space of edges is E(Γ (G,X)) = Γ(G,X)− V(Γ(G,X))= G × (X −{1}):
(g,x) g gx, where x ∈ X −{1}. It is a closed (and hence clopen) subset of Γ(G,X) if and only if 1 is an isolated point of X˜ . Observe that if 1 ∈/ X, then V(Γ(G,X))= G and E(Γ (G,X)) = G × X, and in this case E(Γ (G,X)) is clopen. If 1 ∈ X, then X˜ = X. If 1 is in X and it is an isolated point of X (for example, if X is finite), then X −{1} is also a closed subspace and we have Γ(G,X)= Γ(G,X−{1}). Note that the Cayley graph Γ(G,X)does not contain loops since the elements of the form (g, 1) are vertices by definition. 40 2 Profinite Graphs
Let ϕ : G → H be a continuous homomorphism of profinite groups and let X be a closed subset of G. Put Y = ϕ(X). Then ϕ induces a qmorphism of the corre- sponding Cayley graphs
ϕ˜ : Γ(G,X)−→ Γ(H,Y).
In particular, if U is an open normal subgroup of G and XU = ϕU (X), where ϕU : G → G/U is the canonical epimorphism, then ϕU induces a corresponding epimorphism of Cayley graphs ϕU : Γ(G,X)→ Γ(G/U,XU ). One easily checks that
= Γ(G,X) lim←− Γ(G/U,XU ) UoG is a decomposition of Γ(G,X)as an inverse limit of finite Cayley graphs.
Example 2.1.13 (An infinite connected profinite graph all of whose proper con- nected profinite subgraphs are finite) Let Γ = Γ(Z, {1}) be the Cayley graph of the free profinite group Z of rank one with respect the subset {1}. Then
= { } Γ lim←− Γ Z/nZ, 1 , n≥2 with canonical maps
ϕmn : Γ Z/mZ, {1} −→ Γ Z/nZ, {1} (n|m).
Let
ϕn : Γ −→ Γ Z/nZ, {1} denote the projection (n ∈ N). Assume that is a connected proper profinite sub- graph of Γ . Put n = ϕn(Γ ). Then n is a connected subgraph of the finite graph Γ(Z/nZ, {1}). = ∈ = { } Since Γ , there exists some n0 N such that n0 Γ(Z/n0Z, 1 ).Ob- serve that for every m ∈ N with n |m, the connected components of ϕ−1 ( ) are 0 mn0 n0 | |=| | isomorphic to n0 . Therefore, m n0 . Thus is finite. It is easy to check that if is a proper connected subgraph of Γ with t + 1 vertices, then there exists a γ ∈ Z such that the vertices of are γ,γ + 1,...,γ + t and with edges (γ, 1), (γ + 1, 1),...,(γ + t − 1, 1):
γ γ + 1 γ + 2 γ + t − 1 γ + t • • • ··· • • (γ, 1)(γ+ 1, 1)(γ+ t − 1, 1) 2.2 Groups Acting on Profinite Graphs 41
2.1.14 Circuits. Let ε = (ε1,...,εn), where εi =±1(i = 1,...,n) and n ≥ 1isa natural number. Define Circn(ε) to be a graph with n vertices (that we take to be the elements of Z/nZ) and n edges e1,...,en
12e2
e1 e3
Circn(ε) : 0 3 . en . . n − 1 such that d0(ei) = i − 1 and d1(ei) = i,ifεi = 1, and d0(ei) = i and d1(ei) = i − 1, if εi =−1. We refer to a graph of the form Circn(ε) as a circuit of length n or as a n-circuit. A circuit of length 1 is a loop. Note that if n ≥ 2 and ε = (1,...,1), then Circn(ε) = Γ(Z/nZ, {1}).
2.2 Groups Acting on Profinite Graphs
Let G be a profinite group and let Γ be a profinite graph. We say that the profinite group G acts on the profinite graph Γ on the left, or that Γ is a G-graph,if (i) G acts continuously on the topological space Γ on the left, i.e., there is a con- tinuous map G × Γ → Γ , denoted (g, m) −→ gm, g ∈ G, m ∈ Γ , such that (gh)m = g(hm) and 1m = m, for all g,h ∈ G, m ∈ Γ , where 1 is the identity element of G; and (ii) dj (gm) = gdj (m), for all g ∈ G, m ∈ Γ , j = 0, 1.
Observe that if G acts on Γ , then for a fixed g ∈ G,themapρg : Γ → Γ given by m → gm (m ∈ Γ ) is an automorphism of the graph Γ . Hence (cf. RZ, Remark 5.6.1), G acts on a profinite graph Γ if and only if there exists a continuous homomorphism ρ : G −→ Aut(Γ ), where Aut(Γ ) is the group of automorphisms of Γ as a profinite graph, and where the topology on Aut(Γ ) is induced by the compact-open topology. The kernel of the action of G on Γ is the kernel of ρ, i.e., the closed normal subgroup of G consisting of all the elements g ∈ G such that gm = m, for all m ∈ Γ . One defines actions on the right in a similar manner. We shall consider only left actions in this chapter. Let G be a profinite group that acts continuously on two profinite graphs Γ and Γ . A qmorphism of graphs ϕ : Γ −→ Γ 42 2 Profinite Graphs is called a G-map of graphs if ϕ(gm) = gϕ(m), for all m ∈ Γ,g ∈ G. Assume that a profinite group G acts on a profinite graph Γ and let m ∈ Γ . Define
Gm ={g ∈ G | gm = m} to be the stabilizer (or G-stabilizer, if one needs to specify the group G)ofthe element m. It follows from the continuity of the action and the compactness of G that Gm is a closed subgroup of G. Clearly, ≤ ∈ = Gm Gdj (m), for every m Γ,j 0, 1.
If the stabilizer Gm of every element m ∈ Γ is trivial, i.e., Gm = 1, we say that G acts freely on Γ .Ifm ∈ Γ ,theG-orbit of m is the closed subset Gm ={gm | g ∈ G}. If a profinite group G acts on a profinite graph Γ , then G acts on the profinite space V(Γ)of vertices and G acts on E(Γ ). The space G\Γ ={Gm | m ∈ Γ } of G-orbits with the quotient topology is a profinite space which admits a natural profinite graph structure as follows:
V(G\Γ)= G\V(Γ), dj (Gm) = Gdj (m), j = 0, 1. We say that G\Γ is the quotient graph of Γ under the action of G. The correspond- ing quotient map Γ −→ G\Γ is an epimorphism of profinite graphs given by m → Gm (m ∈ Γ , g ∈ G). We observe that it sends edges to edges (it is a morphism). If N c G, there is an induced action of G/N on N\Γ defined by (gN)(Nm) = N(gm), g ∈ G, m ∈ Γ. The following result is straightforward.
Lemma 2.2.1 Let a profinite group G act on a profinite graph Γ . (a) Let N be a collection of closed normal subgroups of G filtered from below (i.e., the intersection of any two groups in N contains a group in N ) and assume that = G lim←− G/N. N∈N Then the collection of graphs {N\Γ | N ∈ N } is an inverse system in a natural way and = \ Γ lim←− N Γ. N∈N 2.2 Groups Acting on Profinite Graphs 43
(b) Let N c G. For m ∈ Γ , denote by m the image of m in N\Γ . Consider the natural action of G/N on N\Γ defined above. Then (G/N)m is the image of Gm under the natural epimorphism G → G/N. In particular, if Gm ≤ N, for all m ∈ Γ , then G/N acts freely on N\Γ .
Let G be a profinite group. If {Γi,ϕij ,I} is an inverse system of profinite G- graphs and G-maps over the directed poset I , then = Γ lim←− Γi i∈I is in a natural way a profinite G-graph. Next we show that every profinite G-graph admits a decomposition as an inverse limit of finite G-graphs.
Proposition 2.2.2 Let a profinite group G act on a profinite graph Γ . (a) Then there exists a decomposition = Γ lim←− Γi i∈I
of Γ as the inverse limit of a system of finite quotient G-graphs Γi and G-maps ϕij : Γi → Γj (i j) over a directed poset (I, ). (b) If G is finite and acts freely on Γ , then the decomposition of part (a) can be chosen so that G acts freely on each Γi .
Proof The proof follows the same pattern as the proof of Proposition 2.1.4; we only indicate the main steps and changes. We prove (a) and (b) at the same time. Let R be an open equivalence relation on Γ . Assume that G acts continu- ously on the finite discrete space Γ/R in such a way that the canonical projection ϕR : Γ → Γ/R is a G-map of G-spaces: this is equivalent to saying that whenever m, m ∈ Γ and mR = m R, then (gm)R = (gm )R, for all g ∈ G (we term such R a G-invariant equivalence relation). Then (see Sect. 1.3) there exists a set R of G-invariant open equivalence relations on Γ such that (R, ) is a directed poset, {Γ/R,ϕRR } is an inverse system of finite G-spaces and G-maps over R and = Γ lim←− Γ/R (2.3) R∈R as topological G-spaces. Moreover, if G is finite and acts freely on Γ , one can modify the set R so that the action of G on each Γ/R is free and the decomposition (2.3) still holds. Let R be the subset of R consisting of those R ∈ R such that in addition Γ/R has the structure of a G-graph and ϕR : Γ → Γ/R is a G-map of G-graphs. ∈ R : → Let R and apply Construction 2.1.3 to get the maps ϕR Γ ΓϕR and ψϕ : Γϕ → Γ/R.Forg ∈ G and m ∈ Γ , define R R g ϕ(m),ϕd0(m), ϕd1(m) = gϕ(m), gϕd0(m), gϕd1(m) . 44 2 Profinite Graphs
This makes ΓϕR into a G-graph and one checks that ϕR is a G-map of G-graphs and ˜ ψϕR is a G-map of G-spaces. Let R be the open equivalence relation on Γ whose {−1 | ∈ } = ˜ ˜ equivalence classes are ϕR (x) x ΓϕR , so that ΓϕR Γ/R. Therefore R R. From this one sees, as in the proof of Proposition 2.1.4, that R is a directed poset R that is cofinal in . Observe that if G acts freely on Γ/R, then it acts freely on ΓϕR . Hence both (a) and (b) follow from the decomposition (2.3) (see Sect. 1.1).
We remark that part (b) of the above proposition can be sharpened in the fol- lowing sense. When G is infinite, it obviously cannot act freely on a finite graph; hence, if G acts freely on Γ , it is not possible to obtain a G-decomposition of Γ as in part (a) if in addition one requires that G acts freely on each Γi . However, one can obtain a decomposition as in part (a) so that, for each i, a finite quotient Gi of G acts freely on Γi , and G is the inverse limit of the Gi . We make this precise in Proposition 3.1.3. The following example shows how to do this in the case of Cayley graphs.
Example 2.2.3 (The Cayley graph as a G-graph) Let G be a profinite group and let X be a closed subset of G.LetΓ(G,X)be the Cayley graph of G with respect to X (see Example 2.1.12). Define a left action of G on Γ(G,X)by setting
g · (g, x) = g g,x ∀x ∈ X˜ = X ∪{1},g,g∈ G.
Clearly gdi(m) = di(gm), for all g ∈ G, m ∈ Γ(G,X), i = 0, 1. Thus, G acts (con- tinuously and freely) on the Cayley graph Γ(G,X). Now, if N is the collection of all open normal subgroups of G,wehave = Γ(G,X) lim←− Γ(G/N,XN ), N∈N where XN is the image of X in G/N. Note that G/N acts freely on Γ(G/N,XN ).
The next lemma sometimes provides a useful way of checking whether certain G-graphs are connected.
Lemma 2.2.4 (a) Let G = X be an abstract group generated by a subset X. Assume that G acts on an abstract graph Γ . Let be a connected subgraph of Γ such that ∩ x = ∅, for all x ∈ X. Then G = g g∈G is a connected subgraph of Γ . (b) Let X be a closed subset of a profinite group G that generates the group topo- logically, i.e., G = X. Assume that G acts on a profinite graph Γ . Let be a connected profinite subgraph of Γ such that ∩ x = ∅, for all x ∈ X. Then G = g g∈G is a connected profinite subgraph of Γ . 2.3 The Chain Complex of a Graph 45
(c) Let G be a profinite group and let X be a closed subset of G. The Cayley graph Γ(G,X)is connected if and only if G = X.
Proof (a) Put Y = xε ε =±1,x∈ X , and let Yn be the set of elements of G that can be written as a product of not more = = ∞ ⊆ ⊆··· than n elements of Y (n 0, 1, 2,...). Since G n=0 Yn, and Y0 Y1 , it suffices to prove that Yn is a connected graph. We show this by induction on n. If n = 0, then Y0 = . Assume that Yn is connected. From our assumption that x ∩ = ∅, we deduce that x−1 ∩ = ∅, for all x ∈ X. Observe that if w is ε awordinY of length n + 1, then w = w x ,forsomew ∈ Yn and some x ∈ X; hence w ∩ w = ∅; and so, w ∪ Yn is connected. It follows that Yn+1 is connected. (b) By Proposition 2.2.2 there exists a decomposition Γ = lim Γ , where all Γ ←− i i are finite quotient G-graphs of Γ . Hence it suffices to prove the result for Γ finite. In that case the kernel K of the action of G on Γ is an open normal subgroup of G. Therefore, replacing G by its quotient G/K if necessary, we may assume that G is finite; and then the result follows from part (a). (c) Let U be the collection of all open normal subgroups of G. Then = Γ(G,X) lim←− Γ(G/U,XU ), U∈U where XU is the image of X on G/U under the canonical map G → G/U . Therefore we may assume that G is finite, in which case the result follows from part (a): consider the connected subgraph of Γ(G,X) consisting of the ver- tices 1 and {x | x ∈ X} and the collection of edges {(1,x) | x ∈ X −{1}}; then Γ(G,X)= G.
2.3 The Chain Complex of a Graph
We shall use the following notation and terminology. Given a pseudovariety of finite groups C, we say that R is a pro-C ring if it is an inverse limit of finite rings which are in C as abelian groups; if C is the class of all finite rings, we write profinite rather than pro-C.LetX be a profinite space and let R beapro-C ring. We denote by [[ RX]] the free profinite R-module on the space X. Similarly, [[ R(X,∗)]] denotes the free profinite R-module on a pointed space (X, ∗). The complete group algebra [[ RG]] is the inverse limit of the finite group algebras
[[ RG]] = lim (R/I)(G/U) , ←− where I and U range over the open ideals of R and the open normal subgroups of G, respectively. Let G be a profinite group, and let X be a profinite G-space. Then [[ RX]] natu- rally becomes a profinite [[ RG]] -module. Similarly, if (X, ∗) is a pointed profinite 46 2 Profinite Graphs
G-space, then the free profinite R-module [[ R(X,∗)]] is naturally a profinite [[ RG]] - module. Let Γ be a profinite graph. Define E∗(Γ ) = Γ/V(Γ) to be the quotient space of the space Γ modulo the subspace of vertices V(Γ).We think of E∗(Γ ) as a pointed space with the image of V(Γ) as the distinguished point. Let R be a profinite ring and consider the free profinite R-modules [[ R(E∗(Γ ), ∗)]] and [[ RV (Γ )]] on the pointed profinite space (E∗(Γ ), ∗) and on the profinite space V(Γ), respectively. Denote by C(Γ,R) the chain complex
∗ d ε 0 −→ R E (Γ ), ∗ −→ RV (Γ ) −→ R −→ 0 (2.4) of free profinite R-modules and continuous R-homomorphisms d and ε determined by ε(v) = 1, for every v ∈ V(Γ), d(e)¯ = d1(e) − d0(e), where e¯ is the image of an edge e ∈ E(Γ ) in the quotient space E∗(Γ ), and d(∗) = 0. Obviously, Im(d) ⊆ Ker(ε). If we need to emphasize the role of the ring R we sometimes write dR for the map d. Note that if E(Γ ) is closed in Γ , then ∗ is an isolated point of E∗(Γ ), and so [[ R(E∗(Γ ), ∗)]] = [[ RE(Γ )]] ; this is the case in many important examples. The homology groups of Γ are defined as the homology groups of the chain complex C(Γ,R) in the usual way:
H0(Γ, R) = Ker(ε)/Im(d), H1(Γ, R) = Ker(d). A qmorphism α : Γ −→ of profinite graphs naturally induces continuous maps
∗ ∗ αV : V(Γ)−→ V() and αE∗ : E (Γ ), ∗ −→ E (), ∗ , which in turn extend to continuous R-homomorphisms
α˜ : RV (Γ ) −→ RV () and V ∗ ∗ α˜ E∗ : R E (Γ ), ∗ −→ R E (), ∗ . Then the following diagram
0 [[ R(E∗(Γ ), ∗)]] d [[ RV (Γ )]] ε R 0
α˜E∗ α˜V idR 0 [[ R(E∗(), ∗)]] d [[ RV ()]] ε R 0 commutes. In other words, the triple α˜ = (α˜ E∗ , α˜V , idR) is a morphism α˜ : C(Γ,R) −→ C(,R) 2.3 The Chain Complex of a Graph 47 of complexes. Therefore, if = Γ lim←− Γi i∈I is an inverse limit of an inverse system of profinite graphs Γi , the corresponding chain complexes C(Γi,R)form an inverse system and = C(Γ,R) lim←− C(Γi,R). i∈I Furthermore, the homomorphism α˜ induces continuous homomorphisms of ho- mology groups ∗ : −→ ∗ : −→ α0 H0(Γ, R) H0(, R) and α1 H1(Γ, R) H1(, R). ∗ ˜ ∗ Of course, α1 is just the restriction of αE to Ker(d). The statements in the following lemma are easily verified and we leave them to the reader.
Lemma 2.3.1 Let R be a profinite ring. (a) Let α : Γ −→ be a qmorphism of profinite graphs. If α is surjective, then ∗ : −→ α0 H0(Γ, R) H0(, R) is surjective. If α is injective, so is ∗ : −→ α1 H1(Γ, R) H1(, R). (b) If Γ = lim Γ is the inverse limit of an inverse system of profinite graphs Γ , ←− i i then H (Γ, R) = lim H (Γ ,R) and H (Γ, R) = lim H (Γ ,R). 0 ←− 0 i 1 ←− 1 i
In the next proposition we prove that the connectivity of a profinite graph is equivalent to the triviality of its 0-homology group.
Proposition 2.3.2 A profinite graph Γ is connected if and only if H0(Γ, R) = 0, independently of the choice of the profinite ring R.
Proof Write Γ as an inverse limit Γ = lim ∈ Γ of finite quotient graphs Γ .By ←− i I i i Proposition 2.1.6, Γ is a connected profinite graph if and only if each Γi is con- nected as an abstract graph. On the other hand, by Lemma 2.3.1, H0(Γ, R) = 0if and only if H0(Γi,R)= 0, for each i. Hence it suffices to prove the theorem for finite Γ . In this case the sequence (2.4) becomes
d ε 0 −→ RE(Γ ) −→ RV (Γ ) −→ R −→ 0, 48 2 Profinite Graphs where if X is a set, [RX] denotes the free R-module on the set X. Observe that εd = 0, so that Im(d) ≤ Ker(ε). Assume first that Γ is connected. Let t t ε nivi = ni = 0 v1,...,vt ∈ V(Γ); n1,...,nt ∈ R . i=1 i=1
∈ t = t − Fix v0 V(Γ). Then i=1 nivi i=1 ni(vi v0); hence it suffices to check that for every pair of distinct vertices v,w of Γ , there exists some c ∈[RE(Γ )] = − ε1 εm with d(c) w v. To verify this let e1 ,...,em be a path from v to w. Define = s = − c i=1 εiei , where we think of εi as an element of R. Then d(c) w v. Hence the sequence is exact at [RV (Γ )], i.e., H0(Γ, R) = 0. Assume now that the sequence is exact at [RV (Γ )].Letv ∈ V(Γ) and let Γ be the connected component of v in Γ . Suppose that Γ = Γ , and let Γ be the complement of Γ in Γ ; then Γ is a subgraph of Γ . Choose v ∈ V(Γ ). Clearly v − v ∈ Ker(ε). Then there exists s niei ∈ RE(Γ ) ei ∈ E(Γ ),ni ∈ R,i = 1,...,s i=1
s = − such that d( i=1 niei) v v . We may assume that v is a vertex of e1 and e1,...,et ∈ Γ , while et+1,...,es ∈ Γ and v is a vertex of es . Clearly
d RE Γ ≤ RV Γ ,
d RE Γ ≤ RV Γ and
RV (Γ ) = RV Γ ⊕ RV Γ .
t = ∈ = Therefore d( i=1 niei) v . However, v / Ker(ε), a contradiction. Thus Γ Γ , and Γ is connected.
2.4 π-Trees and C-Trees
Let C be a pseudovariety of finite groups and consider the set of primes π = π(C) involved in C (see Sect. 1.3). Let ZCˆ denote the pro-C completion of the group of integers Z. This is the free pro-C group of rank 1; it also has, in a natural way, a ring structure. One has np ZCˆ = Zp/p Zp, p∈π where n np = np(C) = sup n n ∈ N,p ||C|,C ∈ C . 2.4 π-Trees and C-Trees 49
∞ If np =∞, then, by convention, we agree that p Zp = 0. Note that every abelian pro-C group is in a unique way a profinite ZCˆ-module. A profinite graph Γ is said to be a C-tree if Γ is connected and H1(Γ, ZCˆ) = 0. Thus Γ is a C-tree if and only if the sequence C(Γ,ZCˆ) (see Sect. 2.3)
∗ d ε 0 −→ ZCˆ E (Γ ), ∗ −→ ZCˆV(Γ) −→ ZCˆ −→ 0 (2.5) is exact. Note that if the set of edges E(Γ ) of Γ is closed, then the sequence (2.5) becomes
−→ d ε 0 ZCˆE(Γ ) −→ ZCˆV(Γ) −→ ZCˆ −→ 0.
Lemma 2.4.1 Let C be a pseudovariety of finite groups. A profinite graph Γ is a C-tree if and only if the sequence C(Γ,Fp)
∗ d ε 0 −→ Fp E (Γ ), ∗ −→ FpV(Γ) −→ Fp −→ 0 is exact for every p ∈ π(C), where Fp is the field with p-elements.
Proof First observe that a proabelian group is the direct product of its p-Sylow subgroups. So, for any profinite space X, np [[ ZCˆX]] = Zp/p Zp X . p∈π(C) Therefore, np C(Γ,ZCˆ) = C Γ,Zp/p Zp , p∈π(C) where np = np(C). Hence the sequence C(Γ,ZCˆ) is exact if and only if the se- n quence C(Γ,Zp/p p Zp) is exact for each p ∈ π(C). Therefore it suffices to prove n that C(Γ,Zp/p p Zp) is exact if and only if C(Γ,Fp) is exact. n We observe that C(Γ,Zp/p p Zp) and C(Γ,Fp) are sequences of free abelian pro-p groups of exponent pnp and free abelian pro-p groups of exponent p, respec- tively. Moreover, if X is a profinite space, [[ FpX]] is the Frattini quotient
np np Zp/p Zp X /Φ Zp/p Zp X
n of [[ (Zp/p p Zp)X]] : this is obvious if X is finite, and in general this can be deduced by a standard inverse limit argument. n n Exactness of C(Γ,Zp/p p Zp) at [[ (Zp/p p Zp)(V (Γ ))]] is equivalent to exact- ness of C(Γ,Fp) at [[ Fp(V (Γ ))]] , because any of these statements is equivalent to Γ being connected, according to Proposition 2.3.2. Hence from now on we assume that Γ is connected as a profinite graph, and we must show that injectivity of the n map d of C(Γ,Zp/p p Zp) is equivalent to injectivity of the map d of C(Γ,Fp). 50 2 Profinite Graphs
To prove this we will also work with the chain complex C(Γ,Zp). Consider the commutative diagram
∗ dZp ∗ [[ Zp(E (Γ ), ∗)]] d([[ Zp(E (Γ ), ∗)]] )
n ∗ d n ∗ [[ (Zp/p p Zp)(E (Γ ), ∗)]] d([[ (Zp/p p Zp)(E (Γ ), ∗)]] )
∗ dFp ∗ [[ Fp(E (Γ ), ∗)]] d([[ Fp(E (Γ ), ∗)]] ) where the vertical maps are the natural quotient maps, and the maps dZp , d and dFp n denote the maps induced by the homomorphisms d of C(Γ,Zp), C(Γ,Zp/p p Zp) and C(Γ,Fp), respectively. Since the sequence C(Γ,Zp) is exact at [[ ZpV(Γ)]] and since Zp is the free Zp-module of rank 1, the map ε splits, and we have
∗ ZpV(Γ) = d Zp E (Γ ), ∗ ⊕ Zp. Similarly, we have
np np ∗ np Zp/p Zp V(Γ) = d Zp/p Zp E (Γ ), ∗ ⊕ Zp/p Zp and
∗ FpV(Γ) = d Fp E (Γ ), ∗ ⊕ Fp. From this it follows that the last line of the diagram is obtained from the first or second line by taking quotients modulo the subgroups of p-th powers (the Frattini subgroups); and the second line is obtained from the first by taking quotients mod- ulo the subgroups of pnp -th powers. It follows that if dZp (respectively, d )isan isomorphism, then so is dFp . Conversely, assume that dFp is an isomorphism. Since ∗ d([[ Zp(E (Γ ), ∗)]] ) is a subgroup of [[ ZpV(Γ)]] , it is a torsion-free pro-p group, and so a free abelian pro-p group (cf. RZ, Theorem 4.3.3 and Example 3.3.8(c)). Therefore there exists a continuous homomorphism
∗ ∗ α : d Zp E (Γ ), ∗ −→ Zp E (Γ ), ∗ Z ∗ such that d p α is the identity map on d([[ Zp(E (Γ ), ∗)]] ); therefore α is injective. On the other hand,
Zp ∗ Ker d ≤ Φ Zp E (Γ ), ∗ and
Zp ∗ Ker d + Im(α) = Zp E (Γ ), ∗ , ∗ ∗ where Φ([[ Zp(E (Γ ), ∗)]] ) is the subgroup of p-th powers of [[ Zp(E (Γ ), ∗)]] , ∗ i.e., its Frattini subgroup. So Im(α) =[[Zp(E (Γ ), ∗)]] (cf. RZ, Corollary 2.8.5).
Therefore α is an isomorphism, and hence dZp is an isomorphism. Thus, d is also an isomorphism. 2.4 π-Trees and C-Trees 51
The above lemma shows that in fact the concept of a C-tree depends only on the primes involved in the pseudovariety C. This suggests the following definition. Let π be a nonempty set of prime numbers, and denote by Zπˆ the profinite group (ring) Zπˆ = Zp. p∈π We say that a profinite graph Γ is a π-tree if it is connected as a profinite graph and one has H1(Γ, Zπˆ ) = 0. In other words, Γ is a π-tree if and only if the sequence C(Γ,Zπˆ )
∗ d ε 0 −→ Zπˆ E (Γ ), ∗ −→ Zπˆ V(Γ) −→ Zπˆ −→ 0 (2.6) is exact. If π ={p} consists of only one prime, we write p-tree rather than {p}- tree. When π is the set of all prime numbers, we normally use the term profinite tree rather than π-tree. The following proposition is an immediate consequence of Lemma 2.4.1.
Proposition 2.4.2 Let C be a pseudovariety of finite groups and let Γ be a profinite graph. Let π = π(C). The following conditions are equivalent: (a) Γ is a C-tree; (b) Γ is a π-tree; (c) let R be a quotient ring of Z such that the order #R of R as a profinite group involves precisely the primes in the set π. Then the sequence
∗ d ε 0 −→ R E (Γ ), ∗ −→ RV (Γ ) −→ R −→ 0 is exact; (d) for a given prime p, let Rp denote one of the following rings: Zp, Fp or n Zp/p Zp, for some positive integer n. Then, for every p ∈ π, the sequence
∗ d ε 0 −→ Rp E (Γ ), ∗ −→ RpV(Γ) −→ Rp −→ 0 is exact.
Proposition 2.4.3 Let π be a nonempty set of prime numbers. Then the following statements hold. (a) Every finite tree is a π-tree. (b) Every connected profinite subgraph of a π-tree is a π-tree. (c) If 1 and 2 are π-subtrees of a π-tree such that 1 ∩ 2 = ∅, then 1 ∪ 2 is a π-subtree. (d) An inverse limit of π-trees is a π-tree. In particular, an inverse limit of finite trees is a π-tree. (e) If ∅=π ⊆ π, then every π-tree is a π -tree.
Proof Part (b) follows from Lemma 2.3.1(a). Part (c) follows from (b) and Lemma 2.1.7. The first statement in part (d) is a consequence of Lemma 2.3.1(b); 52 2 Profinite Graphs and the second then follows from (a). Part (e) is a consequence of the definition of a π-tree. To prove (a), let Γ be a finite tree. In this case the sequence (2.6) becomes
d ε 0 −→ ZpE(Γ ) −→ ZpV(Γ) −→ Zp −→ 0.
Since Γ is connected, this sequence is exact at [ZpV(Γ)] by Proposition 2.3.2.It remains to see that d is an injection. For this define a map
ρ : V(Γ)−→ ZpE(Γ ) as follows: fix a vertex v0 ∈ V(Γ); since Γ is an abstract tree, for each vertex v ∈ ε1 εt V(Γ)there is a unique path e ,...,et from v0 to v of minimal length; define 1 ρ(v)= ε1e1 +···+εt et e1,...,et ∈ E(Γ ); εi =±1,i = 1,...,t .
Since [ZpV(Γ)] is a free Zp-module, this map extends to a Zp-homomorphism (also denoted ρ) ρ :[ZpV(Γ)]→[ZpE(Γ )]. Then ρd is the identity map on [ZpE(Γ )]; thus d is an injection.
Exercise 2.4.4 Let T be a π-tree. (a) T does not contain circuits. (b) If v,w ∈ V(T)and there exists a path pvw from v to w, then there is a unique reduced path from v to w.
Example 2.4.5 (A π-tree which is not an inverse limit of finite trees) It is not al- ways possible to decompose a π-tree as an inverse limit of finite trees. For example, let p be a prime number. The Cayley graph Γ = Γ(Zp, {1}) is a p-tree (see Theo- rem 2.5.3 below). Let Γ˜ be a finite quotient graph of Γ . Then Γ˜ is also a quotient graph of a graph of the form Γ(Z/pnZ, {1}) (see Lemma 2.1.5), which is a circuit. Hence, if |Γ˜ |≥2, then Γ˜ is not a tree.
Lemma 2.4.6 Let be a profinite subgraph of a profinite graph Γ , and let R be a profinite ring. Then (a) V()is a closed subspace of V(Γ), and (E∗(), ∗) is naturally embedded in (E∗(Γ ), ∗); (b) V(Γ/) is naturally homeomorphic with V(Γ)/V(), and E∗(Γ /, ∗) is naturally homeomorphic with (E∗(Γ )/E∗(), ∗), where, in this last case, the distinguished point ∗ is the image of E∗() in E∗(Γ )/E∗(); (c) R E∗(Γ /), ∗ =∼ R E∗(Γ ), ∗ R E∗(), ∗ .
Proof Parts (a) and (b) are straightforward. To prove (c) consider the natural con- tinuous map ι : E∗(Γ /), ∗ −→ R E∗(Γ ), ∗ R E∗(), ∗ . 2.4 π-Trees and C-Trees 53
We must show that [[ R(E∗(Γ ), ∗)]] /[[ R(E∗(), ∗)]] is the free profinite R-module on the space (E∗(Γ /), ∗) with respect to the map ι (see Sect. 1.7). Let ϕ : (E∗(Γ /), ∗) → A be a continuous map of pointed spaces into a profinite R- module A. Then ϕ induces a continuous map
∗ ϕ1 : E (Γ ), ∗ −→ A, and this in turn induces a continuous R-homomorphism
∗ ϕ1 : R E (Γ ), ∗ −→ A ∗ such that ϕ1([[ R(E (), ∗)]] ) = 0. Therefore ϕ1 induces a continuous R-homomor- phism ϕ¯ : R E∗(Γ ), ∗ R E∗(), ∗ −→ A such that ϕι¯ = ϕ. The uniqueness of ϕ¯ is clear since ι(E∗(Γ /), ∗) generates [[ R(E∗(Γ ), ∗)]] /[[ R(E∗(), ∗)]] .
Lemma 2.4.7 Let be a π-subtree of a connected profinite graph Γ and let α : Γ −→ Γ/ be the corresponding canonical epimorphism of graphs. Then the induced homo- morphism ∗ : −→ α1 H1(Γ, Zπˆ ) H1(Γ /, Zπˆ ) is an isomorphism. In particular, if Γ is a π-tree, then so is Γ/.
Proof We may assume that π consists of just one prime p.Let β : −→ Γ ˜ be the natural embedding. Then β and α induce a monomorphism β : C(,Zp) → C(Γ,Zp) and an epimorphism α˜ : C(Γ,Zp) → C(Γ/,Zp) of chain complexes, respectively, and the following diagram 0
∗ d ε 0 [[ Zp(E (), ∗)]] [[ ZpV()]] Zp 0 ˜ ˜ βE∗ βV id ∗ dΓ εΓ [[ Zp(E (Γ ), ∗)]] [[ ZpV(Γ)]] Zp 0
α˜E∗ α˜V id ∗ dΓ/ εΓ/ [[ Zp(E (Γ /), ∗)]] [[ ZpV(Γ/)]] Zp 0
0 54 2 Profinite Graphs commutes. Note that the first row is exact because is a p-tree, the second row is exact because Γ is connected. ˜ ∗ By Lemma 2.4.6,Ker(α˜E∗ ) = βE∗ ([[ Zp(E (), ∗)]] ),inotherwords,thefirst column of the diagram is an exact sequence. From this it easily follows that ∗ ∈ ∗ = α1 is an injection. Indeed, let a H1(Γ, Zp) be such that α1 (a) 0; i.e., ∗ Γ a ∈[[Zp(E (Γ ), ∗)]] with d (a) = 0 and α˜ E∗ (a) = 0. Then there exists a b ∈ ∗ ˜ ¯ [[ Zp(E (), ∗)]] such that βE∗ (b) = a.Now,sinced and βV are injections, we deduce from the commutativity of the diagram that b = 0. Thus a = 0. ˜ Next we observe that Ker(α˜ V ) = βV (Ker(ε )); indeed, first we notice that this is straightforward if V(Γ)is finite; in general we use an inverse limit argument. ∗ ∈[[ ∗ ∗ ]] Now we can easily deduce that α1 is a surjection: if c Zp(E (Γ /), ) Γ/ ∗ Γ and d (c) = 0, choose c˜ ∈[[Zp(E (Γ ), ∗)]] such that α˜ E∗ (c)˜ = c; then d (c)˜ ∈ ˜ Γ Ker(α˜ V ), and so there exists a y ∈ Ker(ε ) with βV (y) = d (c)˜ ; hence there exists ∗ ˜ Γ a y ∈[[Zp(E (), ∗)]] with d (y ) = y; then c =˜c − βE∗ (y ) ∈ Ker(d ) and
α˜ E∗ (c ) = c, as needed.
Lemma 2.4.8 Let R be a profinite ring. Then the following statements hold. { | ∈ } (a) Let Xi i I be a collection of closed subspaces of a profinite space Y . Set = X i∈I Xi . Then [[ RX]] = [[ RXi]] . i { ∗ | ∈ } (b) Let (Xi, ) i I be a collection of closed pointed subspaces of a profinite ∗ ∗ = ∗ pointed space (Y, ). Set (X, ) i∈I (Xi, ). Then R(X,∗) = R(Xi, ∗) . i (c) Let Y and Z be closed subspaces of the profinite pointed space (X, ∗) such that ∗∈Y and ∗ ∈/ Z. Then there are natural isomorphisms
R(X,∗) /[[ RZ]] =∼ R(X/Z,∗) and R(X,∗) R(Y,∗) =∼ R(X/Y,∗) .
Proof The proofs of (a) and (b) are similar. We only prove (a). Assume first that I ={1, 2}, i.e., X = X1 ∩ X2. Write Y as the inverse limit = Y lim←− Yj j∈J of its finite quotient spaces. Denote by ϕj : Y → Yj the projection (j ∈ J), and define X1j = ϕj (X1) and X2j = ϕj (X2). Since X1j and X2j are finite, we have
R(X1j ∩ X2j ) =[[RX1j ]] ∩ [[ RX2j ]] . 2.4 π-Trees and C-Trees 55
It is easy to verify that ∩ = ∩ = ∩ X1 X2 lim←− X1j lim←− X2j lim←− (X1j X2j ). j∈J j∈J j∈J Hence ∩ = ∩ = ∩ R(X1 X2) R lim←− (X1j X2j ) lim←− R(X1j X2j ) j∈J j∈J = [[ ]] ∩ [[ ]] = [[ ]] ∩ [[ ]] lim←− RX1j RX2j lim←− RX1j lim←− RX2j j∈J j∈J j∈J =[[RX1]] ∩ [[ RX2]] (for the second and fourth equalities see RZ, Proposition 5.2.2). Assume now that I is any indexing set. By the case considered above we may as- sume that the collection {Xi | i ∈ I} is filtered from below, i.e., that the intersection of any two sets in the collection contains a set in the collection. So we may think of this collection as an inverse system of sets and = = X Xi lim←− Xi. i∈I i∈I
Also, using again the case above, the collection of profinite R-submodules {[[RXi]] | i ∈ I} of [[ RY]] is filtered from below. Therefore, [[ ]] = = [[ ]] = [[ ]] RX R lim←− Xi lim←− RXi RXi . i∈I i∈I i∈I (c) We prove the second statement, the first being similar. The quotient map (X, ∗) → (X/Y, ∗) induces a continuous epimorphism of free profinite modules f :[[R(X,∗)]] → [[ R(X/Y,∗)]] . Since f([[ R(Y,∗)]] ) = 0, f induces an epimor- phism ρ : R(X,∗) R(Y,∗) −→ R(X/Y,∗) . On the other hand, the natural map (X/Y, ∗) →[[R(X,∗)]] /[[ R(Y,∗)]] induces a continuous homomorphism ψ : R(X/Y,∗) −→ R(X,∗) R(Y,∗) . Finally, observe that the composition ψρ is the identity map on [[ R(X,∗)]] / [[ R(Y,∗)]] . Thus ρ is an isomorphism.
Proposition 2.4.9 Let π be a nonempty set of prime numbers . Suppose that {i | ∈ } = i I is a family of π-subtrees of a π-tree T , and let i∈I i . Then is either empty or a π-tree.
Proof Assume that = ∅. By Lemma 2.4.8 one has Zπˆ V() = Zπˆ V(i) i∈I 56 2 Profinite Graphs and ∗ ∗ Zπˆ E (), ∗ = Zπˆ E (i), ∗ . i∈I Consider the exact sequence
∗ d ε 0 −→ Zπˆ E (T ), ∗ −→ Zπˆ V(T) −→ Zπˆ −→ 0 associated with T . Denote by ε,εi ,d,di the restrictions of ε and d to and i , respectively. Then i Ker ε = Zπˆ V() ∩ Ker(ε) = Zπˆ V(i) ∩ Ker(ε) = Ker ε ; i∈I i∈I moreover, Im d = Im di i∈I because d is injective. Since each i is connected, we have Ker(ε i ) = Im(d i ), for every i, by Proposition 2.3.2. It follows that Im(d) = Ker(ε). So, by Proposition 2.3.2, is connected, and therefore a π-tree according to Proposi- tion 2.4.3(b).
It follows from Proposition 2.4.9 that given a nonempty subset W of a π-tree T , there exists a smallest π-subtree [W] containing W , namely the intersection of all π-subtrees containing W .IfW consists of two vertices v and w, we use the notation [v,w] rather than [{v,w}] and call it the chain connecting v and w. Observe that if [v,w] is finite, then it is just the underlying graph of the unique reduced path from v to w.
Lemma 2.4.10 A profinite subgraph of a π-tree T is a π-tree if and only if [v,w]⊆, for all v,w ∈ V().
Proof If is a π-tree, then by definition [v,w]⊆, for all v,w ∈ V(). Con- versely, suppose is a profinite subgraph of T and that [v,w]⊆, for all v,w ∈ V(). To prove that is a π-tree, it suffices to show that is connected (see Proposition 2.4.3(b)). Write T as an inverse limit of finite quotient graphs, = T lim←− Ti, i∈I and let ϕi : T → Ti denote the projection (i ∈ I ). It suffices to prove that ϕi() is a connected graph for each i ∈ I . Given vertices v¯ and w¯ of ϕi(),letv,w ∈ V() with ϕi(v) =¯v and ϕi(w) =¯w. Since [v,w]⊆ and [v,w] is a π-tree, we have that ϕi([v,w]) is a connected subgraph of the finite graph ϕi() containing v¯ and w¯ . Therefore, ϕi() is connected. 2.5 Cayley Graphs and C-Trees 57
Example 2.4.11 (A π-tree that coincides with its infinite chains) Let Γ = Γ(Z, 1) be the Cayley graph of the free profinite group Z of rank 1 with respect to its subset {1}.Thisisaπ-tree for any nonempty set of prime numbers π (see Theorem 2.5.3 below and Proposition 2.4.3(e)). The proper π-subtrees of Γ are precisely the proper connected profinite subgraphs of Γ , and these are precisely the finite π-subtrees (see Example 2.1.13). Therefore, if v,w are vertices of Γ , then [v,w]=Γ , unless [v,w] is finite, in which case [v,w] has vertices γ,γ + 1,...,γ + t, where γ = v or γ = w and t is a natural number.
Let G be a profinite group that acts on a π-tree T .Aπ-subtree T of T is G- invariant if whenever g ∈ G and m ∈ T , one has gm ∈ T ; and such T is minimal if it does not contain any proper G-invariant π-subtrees. Minimal G-invariant π- subtrees are especially useful when they are unique. In the next proposition we begin the study of minimal G-invariant π-subtrees T of T . A more systematic study is carried out in Chap. 8.
Proposition 2.4.12 Let G be a profinite group acting on a π-tree T . Then the following assertions hold. (a) There exists a minimal G-invariant π-subtree D of T . (b) If |D| > 1, then D is unique. In particular, if |G| > 1 and G acts freely on T or if G is infinite and the stabilizer of some m ∈ D is finite, then D is the unique minimal G-invariant π-subtree of T . (c) Assume that D is unique. Let N G be such that there exists a unique minimal N-invariant π-subtree L of T . Then L = D.
Proof (a) Consider the collection T of all G-invariant π-subtrees of T ordered by ∈ T T = ∅ { } T reverse inclusion. Since T , .Let Ti i∈I be a linearly ordered chain in . By the compactness of T ,theset Ti is nonempty. Then, by Proposition 2.4.9, Ti is a G-invariant π-subtree. So {Ti}i∈I possesses an upper bound in T . Therefore we can apply Zorn’s lemma to conclude that there exists a minimal G-invariant π- subtree. (b) This will be proved after Corollary 4.1.9. (c) Let g ∈ G; then N acts on gL and so gL is minimal N-invariant; hence gL = L. This means that G acts on L. Therefore D ⊆ L; but obviously L ⊆ D, since N acts on D; thus L = D.
2.5 Cayley Graphs and C-Trees
A pseudovariety of finite groups C0 is said to be closed under extensions with abelian kernel if whenever 1 −→ A −→ G −→ H −→ 1 58 2 Profinite Graphs is an exact sequence of finite groups with A,H ∈ C0 and A is abelian, then G ∈ C0. By the KaloujnineÐKrassner theorem (cf. Kargapolov and Merzljakov 1979, The- orem 6.2.8) such an extension group G can be embedded in the wreath product A by H ; it follows that to check that a pseudovariety of finite groups C is closed under extensions with abelian kernel, it suffices to verify that any semidirect product of an abelian group in C by a group in C is in C. Next we give an example showing that a pseudovariety which is closed under extensions with abelian kernel is not necessarily extension-closed.
Example 2.5.1 (A pseudovariety closed under extensions with abelian kernel that is not extension-closed) Let = A5 be the alternating group of degree 5. This is the finite simple nonabelian group with smallest order. Let C() be the collection of all the finite direct products of copies of . Observe that C() is closed under homomorphic images (cf. RZ, Lemma 8.2.4). For a finite group G, denote by S(G) its maximal solvable normal subgroup. Define V to be the set of all finite groups G such that G/S(G) ∈ C(). We shall show that V is a pseudovariety of finite groups that is closed under extensions with abelian kernel, but not extension-closed. We claim first that V is a pseudovariety. Clearly V is closed under finite direct products; moreover, since C() is closed under homomorphic images, so is V.It remains to prove that V is closed under taking subgroups. Let G ∈ V and let H be a proper subgroup of G. We use induction on the order of G to show that H ∈ V. If G is solvable or G =∼ , then H is solvable and the result is clear. Observe that H/S(H) is a quotient of H/H ∩ S(G).IfS(G) = 1, the result follows from the induction hypothesis since H/H ∩ S(G) =∼ H S(G)/S(G) ≤ G/S(G) and G/S(G) < |G|.
Thus from now on we may assume that G ∈ C() , i.e., G = 1 ×···×n (n ≥ 2), where each i is isomorphic to . Since H is a proper subgroup of G, there is some i such that Hi = H ∩ i = i ,1≤ i ≤ n. Then Hi is solvable. So Hi ≤ S(H) and S(H/Hi) = S(H)/Hi . Now, since H/Hi ≤ G/i ∈ V, we conclude from the induction hypothesis that
H/S(H)= (H/Hi)/S(H/Hi) ∈ C(). This proves the claim. It follows easily from the definition that V is closed under extensions with abelian kernel. Let us show now that it is not extension-closed. For this consider the wreath product R = C of with a group C of order 2; this is a semidirect product of B = × by C; both of these groups are in V; and the action of C on B permutes the two factors .LetK R and assume that K is solvable. We claim that K = 1. Note that K ∩ B = 1, for otherwise K must contain one of the copies of , contra- dicting the solvability of K.IfK = 1, we have R = B × K = B × C, contradicting the definition of R. This proves the claim. Therefore S(R) = 1. Finally, observe that R/∈ C(). Thus R/∈ V. 2.5 Cayley Graphs and C-Trees 59
If (X, ∗) is a pointed profinite space, we denote by F = FC(X, ∗) the free pro-C group on the pointed space (X, ∗). The next two results establish conditions under which the Cayley graph of a free pro-C group with respect to one of its bases is a C-tree. We begin with a study of the augmentation ideal (see Sect. 1.10)ofafree pro-C group.
Lemma 2.5.2 Let C be a pseudovariety of finite groups. Then C is closed under extensions with abelian kernel if and only if for every pointed profinite space (Y, ∗), the augmentation ideal (( I F )) of the complete group algebra [[ ZCˆF ]] of the free pro-C group F = FC(Y, ∗) is a free [[ ZCˆF ]] -module on the pointed space (Y, ∗) with respect to the map ι : (Y, ∗) → (( I F )) defined by ι(y) = y − 1 (y ∈ Y).
Proof The augmentation ideal (( I F )) is topologically generated by the space Y − 1 ={y − 1 | y ∈ Y } as an [[ ZCˆF ]] -module (see Sect. 1.10). Assume first that C is closed under extensions with abelian kernel. We shall prove that (( I F )) satisfies the required universal property of a free [[ ZCˆF ]] -module with respect to the map ι. We must prove that given a map of pointed spaces ψ : Y → M to a profinite [[ ZCˆF ]] -module M, there exists a unique continuous [[ ZCˆF ]] -module homomorphism ψ˜ : (( I F )) → M such that ψι˜ = ψ.
ψ˜ y − 1 (( I F )) M
ι ψ y Y Observe that if such a ψ˜ exists, then it is unique since ι(Y) generates (( I F )) as a [[ ZCˆF ]] -module. We may assume that M is finite since M is an inverse limit of finite [[ ZCˆF ]] - modules (see Sect. 1.7). Note that M ∈ C since M is automatically a ZCˆ-module and so an abelian pro-C group. Since M is in particular an F -module, we may construct the corresponding semidirect product M F . We remark that M F is a pro-C group since C is closed under extensions with abelian kernel. Since F is a free pro-C group on (Y, ∗), there exists a unique continuous homomorphism ρ : F −→ M F such that ρ(y)= (ψ(y), y) (y ∈ Y ). Define now a map δ : F −→ M by the equation (δ(f ), f ) = ρ(f), for all f ∈ F . Then δ is continuous and it is a derivation, that is,
δ(f1f2) = δ(f1) + f1δ(f2), ∀f1,f2 ∈ F (see Sect. 1.10). Now, (see 1.10.7 in Sect. 1.10), there exists an isomorphism
∼ = [[ ]] Der(F, M) Hom ZCˆF (( I F )) , M , 60 2 Profinite Graphs and under this isomorphism δ corresponds to a [[ ZCˆF ]] -homomorphism ψ˜ : (( I F )) −→ M such that ψ(f˜ − 1) = δ(f), for all f ∈ F . Then ψι(y)˜ = ψ(y˜ − 1) = δ(y) = ψ(y), ∀y ∈ Y, and thus ψι˜ = ψ. Conversely, assume that (( I F )) is a free [[ ZCˆF ]] -module on the pointed space (Y, ∗) with respect to the map ι, for every profinite pointed space (Y, ∗), where F = F(Y,∗) denotes the free pro-C group on the pointed profinite space (Y, ∗).Let A,H ∈ C, with A abelian. Assume that A is an H -module, and let G = A H be the corresponding semidirect product. To prove that C is closed under extensions with abelian kernel it suffices to show that G ∈ C, as pointed out above. Let {(ay,hy) | y ∈ Y } be a generating set of G = A H , with ay ∈ A,hy ∈ H , for all y ∈ Y , where (Y, ∗) is a certain finite pointed indexing set and a∗ = 1, h∗ = 1. Then H = hy | y ∈ Y .LetF = FC(Y, ∗) be the free pro-C group on the pointed space (Y, ∗) and let ϕ : F −→ H be the continuous epimorphism determined by ϕ(y) = hy (y ∈ Y ). Then the action of H on A induces an action of F on A via ϕ: f · a = ϕ(f)a, (a ∈ A,f ∈ F). Let G˜ = A F be the corresponding semidirect product, and let ϕ˜ : G˜ = A F −→ G = A H be the epimorphism induced by ϕ. Since, by assumption, (( I F )) is a free [[ ZCˆF ]] -module on (Y, ∗) and A is an [[ ZCˆF ]] -module, there exists a continuous [[ ZCˆF ]] -homomorphism ψ˜ : (( I F )) → A ˜ such that ψ(y − 1) = ay (y ∈ Y). Define d : F −→ A by d(f) = ψ(f˜ − 1)(f∈ F). Then d is a continuous derivation (see 1.10.7 in Sect. 1.10). Hence the map ρ : F −→ G˜ = A F given by ρ(f) = (d(f ), f ) (f ∈ F), is a continuous homomorphism (cf. RZ, Lemma 9.3.6). Define α : F → G to be the composite α =˜ϕρ. Observe that
α(y) = (ay,hy)(y∈ Y); therefore α is an epimorphism, and thus G ∈ C, as needed. 2.5 Cayley Graphs and C-Trees 61
Theorem 2.5.3 Let C be a pseudovariety of finite groups. Then C is closed under extensions with abelian kernel if and only if for every profinite pointed space (Y, ∗), the Cayley graph Γ = Γ(F,Y)ofthefreepro-C group F = F(Y,∗) with respect to Y is a C-tree.
Proof We think of (Y, ∗) as being embedded in F ; in particular ∗ is identified with 1. Since 1 ∈ Y , Γ = Γ(F,Y)= F × Y and V(Γ)= F ×{1}. Consider the sequence associated with the graph Γ and ZCˆ as in Eq. (2.4) of Sect. 2.3: d ε 0 −→ ZCˆ (F × Y)/ F ×{1} , ∗ −→ [[ ZCˆF ]] −→ ZCˆ −→ 0, where d(f,y) = fy − f (y ∈ Y ) and ε(f ) = 1(f ∈ F ). We have to prove that this sequence is exact for every (Y, ∗) if and only if C is closed under extensions with abelian kernel. By Lemma 2.2.4, Γ is a connected profinite graph since F is topologically gener- ated by Y . Therefore, by Proposition 2.3.2, the above sequence is exact at [[ ZCˆF ]] . It remains to prove that d is a monomorphism. Now, Ker(ε) is the augmentation ideal (( I F )) of [[ ZCˆF ]] , which is generated as a topological [[ ZCˆF ]] -module by the subspace {y − 1 | y ∈ Y } (see Sect. 1.10). On the other hand, [[ ZCˆ((F × Y)/(F ×{1}), ∗)]] is a free [[ ZCˆF ]] -module on the quotient space F \((F × Y)/(F ×{1}), ∗) (cf. RZ, Proposition 5.7.1). The space F \((F × Y)/(F ×{1}), ∗) can be identified with the pointed space ({(1,y)| y ∈ Y }, ∗). Since d(1,y)= 1 − y(y∈ Y), to show that d is a monomorphism is equivalent to showing that the augmentation ideal (( I F )) is free on the subspace ({1 − y | y ∈ Y }, ∗), as a profinite [[ ZCˆF ]] -module. But, according to Lemma 2.5.2, this is the case if and only if C is closed under extensions with abelian kernel. Chapter 3 The Fundamental Group of a Profinite Graph
3.1 Galois Coverings
If a profinite group G acts freely on a profinite graph Γ , then the natural epimor- phism of profinite graphs ζ : Γ → = G\Γ of Γ onto the quotient graph = G\Γ is called a Galois covering of the profinite graph . Observe that a Galois covering ζ always sends edges to edges. We shall refer to the group G as the group associated with the Galois covering ζ and denote it by G = G(Γ |) or by G = G(ζ ).IfΓ is finite, one says that the Galois covering ζ is finite (observe that if this is the case, then the associated group G(Γ |) is finite). The Galois covering ζ is said to be connected if Γ is connected.
Example 3.1.1 (a) Let ζ : Γ → be a Galois covering of the profinite graph with associ- ated group G = G(Γ |).LetK c G. Then G/K acts freely on K\Γ and K\Γ → is also a Galois covering of , with associated group G/K. (b) Let (X, ∗) be a pointed profinite space. Define a profinite graph B = B(X,∗) by B = X, V(B)={∗}, di(x) =∗(x ∈ X),fori = 0, 1. We shall refer to B(X,∗) as the bouquet of loops associated to (X, ∗). For example, if X has 5 points, B(X,∗) is the graph
•
Let G be a profinite group, X a closed subset of G and let Γ(G,X) be the Cayley graph of G with respect to X. Then the natural action of G on Γ(G,X) described in Example 2.2.3 is free. Hence the natural epimorphism
ζ : Γ(G,X)→ G\Γ(G,X)
© Springer International Publishing AG 2017 63 L. Ribes, Profinite Graphs and Groups, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics 66, DOI 10.1007/978-3-319-61199-0_3 64 3 The Fundamental Group of a Profinite Graph
is a Galois covering. Note that G\Γ(G,X) is just the bouquet of loops B(X ∪{1}, 1).
Let ζ1 : Γ1 → 1, ζ2 : Γ2 → 2 be Galois coverings and put G1 = G(Γ1|1) and G2 = G(Γ2|2).Amorphism of Galois coverings
ν : ζ1 → ζ2 consists of a pair ν = (γ, f ), where γ : Γ1 → Γ2 is a qmorphism of profinite graphs and f : G1 → G2 is a continuous homomorphism of groups such that γ(gm)= f(g)γ(m), for all g ∈ G1, m ∈ Γ1 (we also refer to this situation by saying that the qmorphism of graphs γ and the group homomorphism f are compatible or that the map γ is equivariant). The composition of morphisms of Galois coverings is again a morphism, and Galois coverings and their morphisms form a category. Note that such a morphism ν of Galois coverings induces a unique qmorphism of profinite graphs δ : 1 → 2 such that the diagram
γ Γ1 Γ2
ζ1 ζ2 δ 1 2 commutes. The morphism ν = (γ, f ) is called surjective or an epimorphism if γ and f (and hence δ) are epimorphisms. Assume that {ζi : Γi → i,νij = (γij ,fij ), I} is an inverse system of Galois coverings indexed by a directed poset I . Put Gi = G(ζi)(i∈ I). Then we have corresponding inverse systems {Γi,γij ,I} and {Gi,fij ,I} of profinite graphs and profinite groups, respectively, such that the profinite group
G = lim G ←− i acts continuously and freely on the profinite graph
Γ = lim Γ ; ←− i moreover, the quotient profinite graph G\Γ is isomorphic with the profinite graph = lim . Hence we have established the following proposition. ←− i
Proposition 3.1.2 The inverse limit lim ζ of Galois coverings ←− i
ζi : Γi → i with associated group Gi = G(ζi) is a Galois covering with associated group lim G . ←− i 3.1 Galois Coverings 65
Proposition 3.1.3 Any Galois covering ζ : Γ → of profinite graphs can be de- composed as an inverse limit of finite Galois coverings, with surjective projections.
Proof Put G = G(Γ |). Assume first that G is finite. By Proposition 2.2.2(b), there exists a G-decomposition = Γ lim←− Γi i∈I of Γ as an inverse limit of finite quotient graphs Γi of Γ on which G acts freely. Denote by
ζi : Γi −→ G\Γi the natural morphism of graphs. Then each ζi is a Galois covering with associated group G. Moreover, the ζi constitute an inverse system of Galois coverings and = ζ lim←− ζi i∈I is the desired decomposition, with surjective projections ζ → ζi . Now assume that G is infinite. Let N denote the set of all open normal subgroups of G. For every N ∈ N , the finite quotient group G/N acts freely on ΓN = N\Γ (see Lemma 2.2.1). By the finite case above, there exists a G/N-decomposition = ΓN lim←− ΓN,i, i∈IN where each ΓN,i is a finite quotient graph of ΓN with free action of G/N on each ΓN,i. Denote by
ϕN,i : ΓN −→ ΓN,i the canonical projection; note that each ϕN,i is surjective. Next we show that the finite graphs ΓN,i form in a natural way an inverse system whose maps (to be defined presently)
γ(M,j),(N,i) : ΓM,j −→ ΓN,i are compatible with the canonical homomorphisms G/M → G/N, whenever M,N ∈ N and M ≤ N; i.e., we shall show that we have an inverse system of fi- nite Galois coverings. First we specify the poset I over which this is an inverse system. We relabel the elements of the indexing set IN : an element i ∈ IN will be denoted from now on by (N, i). Define the indexing set of this inverse system to be I = IN . N∈N If (N, i), (M, j) ∈ I , we say that
(N, i) (M, j) 66 3 The Fundamental Group of a Profinite Graph if M ≤ N and there exists a qmorphism of graphs α : ΓM,j → ΓN,i, compatible with the canonical homomorphism G/M → G/N, such that the diagram
ϕN,i N\Γ = ΓN ΓN,i
Γ α
M\Γ = ΓM ΓM,j ϕM,j commutes. Observe that α is unique, if it exists, because ϕM,j is surjective. Hence, (I, ) is a partially ordered set. We observe that the restriction of to IN coincides with the partial ordering in IN (N ∈ N ). We claim that this ordering makes (I, ) into a directed poset. To see this, consider (N, i), (M, j) ∈ I . Put L = N ∩ M. Then (see Lemma 2.1.5), there exists some (L, k) ∈ IL and morphisms of graphs αi : ΓL,k → ΓN,i and αj : ΓL,k → ΓM,j such that the diagram
ΓN ΓN,i
αi
ΓL ΓL,k αj
ΓM ΓM,j commutes. All maps in this diagram are epimorphisms; it follows that αi and αj are unique. All maps except possibly αi and αj are compatible with the natural homomorphisms of the relevant groups acting on these graphs. One deduces that the qmorphisms αi and αj are compatible with the homomorphisms G/L → G/N and G/L → G/M, respectively. This shows that (I, ) is directed. For (N, i) ∈ I , consider the Galois covering
ζN,i : ΓN,i → (G/N)\ΓN,i with associated group G/N. Then one checks that = ζ lim←− ζN,i, N,i∈I with epimorphic projections ζ → ζN,i. 3.1 Galois Coverings 67
The following proposition gives an equivalent way of viewing morphisms of Ga- lois coverings.
Proposition 3.1.4 Let ζ1 : Γ1 → 1, ζ2 : Γ2 → 2 be connected Galois coverings with associated groups G1 = G(Γ1|1) and G2 = G(Γ2|2), respectively. Let γ : Γ1 → Γ2 and δ : 1 → 2 be morphisms of graphs such that the diagram
γ Γ1 Γ2
ζ1 ζ2 δ 1 2 commutes. Then there exists a unique continuous homomorphism
f : G1 −→ G2 which is compatible with γ . Explicitly, f is defined as follows: given g ∈ G1, choose any m ∈ Γ1; then f(g)is the unique element of G2 such that γ(gm)= f(g)γ(m).
Proof Fix g ∈ G1. The commutativity of the diagram implies that γ(gm)and γ(m) are in the same G2-orbit for any given m ∈ Γ1. Since G2 acts freely on Γ2, there exists a unique element h(m, g) ∈ G2 such that γ(gm)= h(m, g)γ (m). We claim that the element h(m, g) is independent of the choice of m and that the map f : G1 → G2 given by g → h(m, g) is a continuous homomorphism. To prove the claim we consider two cases.
Case 1. Assume that ζ2 : Γ2 → 2 is finite.
For k ∈ G2,set Γ1(k) = m ∈ Γ1 h(m, g) = k .
Observe that Γ1(k) is a closed subset of Γ1; indeed, if m ∈ Γ1 is such that γ(gm)= kγ(m), then, by the continuity of γ and of the actions, there exists some open neigh- bourhood U of m such that γ(gU)∩ kγ(U) =∅; hence U ∩ Γ1(k) =∅, showing that the complement of Γ1(k) is open; therefore Γ1(k) is closed. Since G2 is finite, the graph Γ1 is a finite union of subsets Γ1(k) with pairwise empty intersection. Let us check that Γ1(k) is a subgraph for any k ∈ G2. Assume that m ∈ Γ1(k).This means that γ(gm)= kγ(m); hence
γ gdj (m) = γ dj (gm) = dj γ(gm) = dj kγ(m) = kγ dj (m) , so that dj (m) ∈ Γ1(k) (j = 0, 1). Then by the connectedness of Γ1, Γ1 = Γ1(k0), for some k0 ∈ G2 (see Lemma 2.1.9), i.e., h(m, g) = k0 for every m ∈ Γ1. We are now in a position to define the homomorphism f .Forg ∈ G1,set
f(g)= h(m, g) = k0. 68 3 The Fundamental Group of a Profinite Graph
The above considerations show that f is well-defined. The equality
f(g1g2)γ (m) = γ(g1g2m) = f(g1)γ (g2m) = f(g1)f (g2)γ (m) for g1,g2 ∈ G1 implies that f is a homomorphism. It remains to show that f is continuous. To see this we use Proposition 3.1.3 to decompose the Galois covering ζ1 : G1 → 1 as an inverse limit
= ζ1 lim←− ζ1i i∈I of finite Galois coverings ζ1i : Γ1i → 1i with epimorphic projections (γ1i,f1i) : ζ1 → ζ1i . Denote by δ1i : 1 → 1i the induced qmorphism of profinite graphs.
γ Γ1 Γ2 γ1i
Γ1i ζ1 ζ2
δ ζ1i 1 2 δ1i
1i
By Lemma 2.1.5, γ factors through Γ1k,forsomek ∈ I ,sayγk : Γ1k → Γ2 is a qmorphism of profinite graphs such that γ = γkγ1k. Similarly, there exists a morphism δj : 1j → 2 such that δ = δj δ1j . Choose t ∈ I such that t j,k. Since γ1t is surjective, we deduce that ζ2γt = δt ζ1t . It follows from the construction above that there exists a unique map of finite groups
ft : G1t = G(Γ1t |1t ) −→ G2 such that γi(g1t m1t ) = ft (g1t )γt (m1t ), for any g1t ∈ G1t and any m1t ∈ Γ1t .Nowit follows from the uniqueness of f and ft that f = ft f1t . Since both ft and f1t are continuous, so is f , as desired.
Case 2. Assume that Γ2 is infinite.
By Proposition 3.1.3, one can represent ζ2 as an inverse limit
= ζ2 lim←− ζ2i i∈I 3.1 Galois Coverings 69 of finite quotient Galois coverings ζ2i : Γ2i → 2i with finite associated groups G2i = G(Γ2i|2i), where = G2 lim←− G2i. i∈I
Denote by ϕ2i = (γ2i,f2i) : ζ2 → ζ2i the projection. Then we have commutative diagrams
γ γ2i Γ1 Γ2 Γ2i
ζ1 ζ2 ζ2i
δ δ2i 1 2 2i where δ2i is induced by ϕ2i . By the case above, for each i ∈ I there exists a unique continuous group homomorphism fi : G → G2i , compatible with γ2iγ , correspond- ing to the commutative diagram
γ2i γ Γ1 Γ2i
ζ1 ζ2i
δ2i δ 1 2i
The uniqueness of the homomorphisms fi implies that fi = fjifj , whenever i j, where fji : G2j → G2i is the canonical homomorphism associated with the inverse system {G2i}i∈I . Therefore, the fi induce a unique continuous homomorphism : −→ = f G1 G2 lim←− G2i i∈I compatible with γ .
It follows from this proposition that a morphism ν = (γ, f ) : ζ1 → ζ2 of con- nected Galois coverings ζ1 : Γ1 → 1, ζ2 : Γ2 → 2 can be viewed equivalently as a pair of morphisms of profinite graphs γ : Γ1 → Γ2 and δ : 1 → 2 such that ζ2γ = δζ1.
Proposition 3.1.5 Let ζ : Γ → , ξ : Σ → be connected Galois coverings and let γ : Γ → Σ be a qmorphism such that ζ = ξγ.
γ Γ Σ
ζ ξ 70 3 The Fundamental Group of a Profinite Graph
Let f : G(Γ |) → G(Σ|) be the homomorphism constructed in Proposi- tion 3.1.4. Then (a) the maps γ and f are surjective; (b) γ is a Galois covering with G(Γ |Σ)= Ker(f ), and consequently G(Σ|) =∼ G(Γ |)/G(Γ |Σ).
Proof (a) First we show that the surjectivity of γ implies the surjectivity of f .Let k ∈ G(Σ|). Choose m ∈ Γ . Since γ is assumed to be onto, there exists some n ∈ Γ such that kγ(m) = γ(n). Clearly ζ(m)= ζ(n). Hence there exists some g ∈ G(Γ |) with gm = n. It follows from the definition of f that γ(gm)= f(g)γ(m); therefore, f(g)γ(m)= kγ(m). Thus f(g)= k, because G(Σ|) acts freely on Σ. Next we prove that γ is surjective. For any open normal subgroup N of G(Σ|), the finite quotient group G(Σ|)/N acts freely on the quotient graph N\Σ.So the natural qmorphism ξN : N\Σ → is a Galois covering with finite associated group G(Σ|)/N. Clearly ξ = lim ξ and the projections of this decomposition ←− N are surjective. Let μN : Σ → N\Σ be the natural morphism and set γN = μN γ . Then γ = lim γ , and so it is enough to prove that γ is surjective for every N. ←− N N Thus from now on we may assume that the Galois covering ξ has finite associated group. Let H = f(G(Γ|)) and let t1,...,tn be representatives of the left cosets of H in G(Σ|). Observe that Hγ(Γ) = γ(Γ), since f(g)γ(m) = γ(gm) (g ∈ G(Γ |), m ∈ Γ);so n Σ = tiγ(Γ). i=1
We claim that the subgraphs tiγ(Γ) (i = 1,...,n) of Σ are clopen. They are ob- viously closed. Hence it suffices to show that they are disjoint. To see this, let s ∈ tiγ(Γ) ∩ tj γ(Γ),forsomei, j ∈{1,...,n}. Then tiγ(m1) = tj γ(m2),for some m1,m2 ∈ Γ . Hence ζ(m1) = ζ(m2), and therefore m1 = gm2,forsome g ∈ G(Γ |). Then tif(g)γ(m2) = tj γ(m2). Since G(Σ|) acts freely on Σ, one concludes that tif(g)= tj ; hence, i = j. This proves the claim. Since Σ is con- nected, it follows (see Lemma 2.1.9) that n = 1, i.e., γ(Γ)= Σ. (b) We just need to prove that γ coincides with the natural epimorphism Γ −→ Ker(f )\Γ.
If gm1 = m2,form1,m2 ∈ Γ , g ∈ Ker(f ), then
γ(m2) = γ(gm1) = f(g)γ(m1) = γ(m1).
Conversely, assume that γ(m1) = γ(m2),forsomem1,m2 ∈ Γ . Then there exists a g ∈ G(Γ |) such that m1 = gm2. Therefore,
γ(m2) = γ(m1) = f(g)γ(m2); so f(g)= 1, since G(Σ|) acts freely on Σ, i.e., g ∈ Ker(f ). 3.1 Galois Coverings 71
Corollary 3.1.6 Let
γ Γ1 Γ2
ζ1 ζ2 δ 1 2 be a commutative diagram of profinite graphs and qmorphisms such that ζ1 and ζ2 are connected Galois coverings. Then γ is surjective if and only if δ is surjective.
Proof If γ is a surjection, it is clear that so is δ. Conversely, assume that δ is a surjection. By Proposition 3.1.4, there exists a continuous homomorphism
f : G(Γ1|1) −→ G(Γ2|2) which is compatible with γ . It follows that the restriction
ζ2|Im(γ ) : Im(γ ) −→ 2 of ζ2 to Im(γ ) is a Galois covering with associated group G(Im(γ )|2) = Im(f ). Then, by Proposition 3.1.5, the inclusion Im(γ ) → Γ2 is surjective; thus Im(γ ) = Γ2, i.e., γ is surjective.
Lemma 3.1.7 Let ζ : Γ → be a connected Galois covering and let Σ be a connected profinite graph. If β1,β2 : Σ → Γ are morphisms of profinite graphs with ζβ1 = ζβ2, and β1(m) = β2(m), for some m ∈ Σ, then β1 = β2.
Proof Assume first that Γ is finite. Let Σ be the set of elements of Σ where β1 and
β2 coincide. Then Σ is a closed nonempty subgraph of Σ. Since Γ is finite and = −1 ∩ −1 Σ β1 (x) β2 (x) , x∈Γ it follows that Σ is clopen in Σ. Suppose Σ = Σ . Then there exists an edge e of
Σ with e/∈ E(Σ ) such that dj (e) ∈ Σ for j = 0 or 1, because otherwise Σ − Σ is an open subgraph of Σ and Σ = Σ ∪ (Σ − Σ ) would be a disconnected graph
(see Lemma 2.1.9), contradicting our hypothesis. Say d0(e) ∈ Σ . Hence
d0 β1(e) = β1 d0(e) = β2 d0(e) = d0 β2(e) .
Since ζ(β1(e)) = ζ(β2(e)), there exists a nontrivial element g ∈ G(Γ |) such that gβ1(e) = β2(e). Therefore
gd0 β1(e) = d0 β2(e) = d0 β1(e) , contradicting the freeness of the action of G(Γ |) on Γ . 72 3 The Fundamental Group of a Profinite Graph
Suppose now that Γ is infinite. Then, by Proposition 3.1.3, there exists a decom- position ζ = lim ζ of ζ as an inverse limit of finite Galois coverings ζ : Γ → . ←− i i i i Let ϕi be the projection of Γ onto Γi . By the case above, ϕiβ1 = ϕiβ2. Since this is valid for every i, we deduce that β1 = β2, as required.
3.2 G(Γ |) as a Subgroup of Aut(Γ )
Let ζ : Γ → be a Galois covering of a profinite graph , and let G = G(Γ |) be the associated profinite group. Each element g of G determines a continuous automorphism
νg : Γ −→ Γ given by νg(m) = gm (g ∈ G, m ∈ Γ). Moreover, the map
ν : G −→ Aut(Γ ) which sends g to νg is a homomorphism, and it is injective because G acts freely on Γ . The first result in this section says that we may think of the topological group G as a closed subgroup of Aut(Γ ) if we identify G with ν(G) and assume that Aut(Γ ) is endowed with the compact-open topology. Recall that the compact-open topology on Aut(Γ ) is generated by a sub-base of open subsets of the form B(K,U) = f ∈ Aut(Γ ) f(K)⊆ U , where K ranges over the compact subsets of Γ and U ranges over the open subsets of Γ (cf. Bourbaki 1989,X,3,4).
Proposition 3.2.1 Let ζ : Γ → be a Galois covering of a profinite graph , and let G = G(Γ |) be the associated profinite group. Consider the group Aut(Γ ) of automorphisms of the profinite graph endowed with the compact-open topology. Then G is naturally embedded in Aut(Γ ) as a topological subgroup, i.e., there exists a topological isomorphism of G with a closed subgroup of Aut(Γ ).
Proof We claim first that the monomorphism ν : G → Aut(Γ ) defined above is continuous. To see this, let K be a compact subset of Γ and let U be an open subset of Γ .Let B(K,U) = f ∈ Aut(Γ ) f(K)⊆ U be the corresponding sub-basic open subset of Aut(Γ ). It suffices to show that ν−1(B(K, U)) is open in G. To see this it is enough to prove that if g ∈ G and νg ∈ B(K,U), then there exists an open neighbourhood W of g in G contained in ν−1(B(K, U)). We shall check that this is indeed the case. Since gK ⊆ U, for each 3.2 G(Γ |) as a Subgroup of Aut(Γ ) 73 x ∈ K, there exist open neighbourhoods Wx of g in G and Vx of x in Γ such that ⊆ WxVx U. Since K is compact, there exist finitely many points x1,...,xn in K so n = that i=1 Vxi K. Then n = W Wxi i=1 is the desired neighbourhood of g. This proves the claim. Finally, since G is compact, the above implies that ν maps G homeomorphically onto Im(ν), if we can prove the following assertion: the compact-open topology on Aut(Γ ) is Hausdorff. To show this, let f1,f2 ∈ Aut(Γ ) be distinct. Then there exists an m ∈ Γ with f1(m) = f2(m).LetUi be a neighbourhood of fi(m) (i = 1, 2) such that U1 ∩ U2 =∅. Then clearly fi ∈ B({m},Ui)(i= 1, 2) and B({m},U1) ∩ B({m},U2) =∅, proving the assertion.
Let ζ : Γ → be a connected Galois covering of a profinite graph , and put G = G(Γ |). By the above proposition we may think of G as a subgroup of Aut(Γ ).Let
H = NAut(Γ )(G), the normalizer of G in Aut(Γ ). Then H is a closed subgroup of Aut(Γ ). Observe that if f ∈ H, then f induces a map
Φ(f) : = G\Γ −→ = G\Γ defined by Φ(f )(Gm) = Gf (m) (m ∈ Γ).
Proposition 3.2.2 (a) If f ∈ H, then Φ(f) is a continuous automorphism of the profinite graph , i.e., Φ(f) ∈ Aut(). (b) The map Φ : H −→ Aut() is a continuous homomorphism. (c) Ker(Φ) = G. (d) If is finite, then the homomorphism Φ : H → Aut() is open and the group H = NAut(Γ )(G) is profinite.
Proof (a) It is easy to verify that Φ(f) is a map of graphs and a bijection. We shall prove that it is continuous. Consider an open subset U of . Then
Φ(f)−1(U) = ζf −1 ζ −1(U) , and this is an open set since f is continuous and ζ is continuous and open. 74 3 The Fundamental Group of a Profinite Graph
(b) Consider a sub-basic open subset B(K,U) of Aut(), where K is a compact subset of and U is an open subset of . Then putting K˜ = ζ −1(K) and U˜ = ζ −1(U),wehave
Φ(f)−1 B(K,U) = B(K,˜ U),˜ which is open in Aut(Γ ). (c) Clearly G ≤ Ker(Φ). Conversely, let f ∈ H be such that Φ(f) = Id. Choose m ∈ Γ . Then f(m)= gm,forsomeg ∈ G. Define
g˜ : Γ −→ Γ to be the graph morphism which is multiplication by g. Then ζf = ζ g˜, since f ∈ Ker(Φ). Therefore f =˜g by Lemma 3.1.7, since f(m)=˜g(m). (d) The map Φ is open since Aut() is discrete. If follows that H/G =∼ Im(Φ), as topological groups. Since H/G is finite and Ker(Φ) = G is profinite, the group H is profinite.
3.3 Universal Galois Coverings and Fundamental Groups
Let C be a pseudovariety of finite groups. A Galois covering ζ : Γ → is said to be a Galois C-covering if its associated group G(Γ |) is a pro-C group. Note that since a Galois C-covering is a Galois covering, all the results of the previous sections are valid for Galois C-coverings. A connected Galois C-covering
ζ : Γ˜ −→ Γ of a connected profinite graph Γ is said to be universal if the following universal property holds: given any qmorphism β : Γ → to a connected profinite graph , any connected Galois C-covering ξ : Σ → , and any points m ∈ Γ˜ , s ∈ Σ such that βζ(m) = ξ(s), there exists a qmorphism of profinite graphs α : Γ˜ → Σ, such that βζ = ξα and α(m) = s
α Γ˜ Σ
ζ ξ β Γ
We say that α lifts β, or that α is a lifting (qmorphism) of β. Once m ∈ Γ˜ and s ∈ Σ with βζ(m) = ξ(s) are given, the lifting qmorphism α is unique (see Lemma 3.1.7). Note also that if the map β is surjective, so is α by Corollary 3.1.6. Furthermore, it follows from Proposition 3.1.3 that it is sufficient to check the universal property above for finite Galois C-coverings ξ : Σ → . 3.3 Universal Galois Coverings and Fundamental Groups 75
We shall prove the existence of universal Galois coverings in Proposition 3.5.3 and Proposition 3.5.6 below. We prove uniqueness in the following proposition.
Proposition 3.3.1 Let ζ : Γ˜ → Γ be a universal Galois C-covering of a profinite connected graph Γ . (a) Assume that α : Γ˜ → Γ˜ is a morphism of profinite graphs such that ζα = ζ . Then α is an automorphism. (b) A universal Galois C-covering is unique, if it exists. More precisely, if ζ : Γ˜ → Γ is another universal Galois C-covering of Γ , then there exists an iso- morphism α : Γ˜ → Γ˜ of profinite graphs such that ζ α = ζ .
Proof (a) Let m ∈ Γ˜ and set m = α(m). By the universality of ζ , there exists a morphism β : Γ˜ → Γ˜ such that β(m ) = m and ζβ = ζ . Since βα(m) = m and = = = αβ(m ) m ,wehavebyLemma3.1.7 that βα αβ idΓ˜ . Thus α is an automor- phism. (b) Choose m ∈ Γ˜ , m ∈ Γ˜ such that ζ(m) = ζ (m ). Then there exist mor- phisms α : Γ˜ → Γ˜ , α : Γ˜ → Γ˜ with α(m) = m , α (m ) = m, which are lift- = ing morphisms of the identity morphism of Γ . By the argument in (a), α α idΓ˜ , = αα idΓ˜ . Thus α is an isomorphism.
C = ˜ | C The profinite group π1 (Γ ) G(Γ Γ)is called the fundamental pro- group of the connected profinite graph Γ . By Proposition 3.3.1(b) and Proposition 3.1.5(b), the fundamental pro-C group of Γ is well-defined up to isomorphism. We say that C C = a connected profinite graph Γ is -simply connected if π1 (Γ ) 1. In Proposi- tion 3.3.3 and in Sect. 3.10 we shall study the relationship between simple connec- tivity and being a C-tree. Next we show that forming universal Galois C-coverings commutes with inverse limits.
Proposition 3.3.2 Let {Γi,ϕij ,I} be an inverse system of profinite connected graphs Γi over a poset (I, ), and let = Γ lim←− Γi. i∈I ˜ For each i ∈ I , let ζi : Γi → Γi be a universal Galois C-covering of Γi . Then
(a) The Galois C-coverings ζi form an inverse system over I and = ζ lim←− ζi i∈I
is a universal Galois C-covering ζ : Γ˜ → Γ of Γ . C (b) The fundamental groups π1 (Γi) form an inverse system over I and C = C π1 (Γ ) lim←− π1 (Γi). i∈I 76 3 The Fundamental Group of a Profinite Graph
Proof Denote by ϕi : Γ → Γi (i ∈ I)the projection. Fix some m ∈ Γ and put mi = ˜ ϕi(m) (i ∈ I). For each i ∈ I , choose m˜ i ∈ Γi such that ζi(m˜ i) = mi For i j in I , ˜ ˜ let ϕ˜ij : Γi → Γj be the (unique) lifting of ϕij : Γi → Γj such that ϕ˜ij (m˜ i) =˜mj . ˜ Then {Γi, ϕ˜ij ,I} is an inverse system over I , and therefore {ζi,(ϕ˜ij ,ϕij ), I} is an inverse system over I of Galois coverings. Now, (ϕ˜ij ,ϕij ) induces a canonically : C → C ˜ defined continuous homomorphism fij π1 (Γi) π1 (Γj ) compatible with ϕij { C } (see Proposition 3.1.4). Therefore, π1 (Γi), fij ,I is an inverse system over I . Put = C = ˜ G lim←− π1 (Γi) and Γ lim←− Γi. i∈I i∈I
Clearly G acts freely on Γ and G\Γ = Γ . So, the quotient map ζ : Γ → Γ is a Galois C-covering. Moreover, ζ can be identified with the map induced by the maps ζ , i.e., ζ = lim ∈ ζ . i ←− i I i We claim that ζ is a universal Galois C-covering of Γ (this will show that Γ = Γ˜ and ζ = ζ , i.e., that part (a) holds). Let ξ : Σ → be a connected finite Galois C-covering and let μ : Γ → be a morphism of profinite graphs. Choose points m ∈ Γ and s ∈ Σ with μζ (m ) = ξ(s). Since is finite, there exists some ∈ : → = i0 I and a morphism μ0 Γi0 with μ0ϕi0 μ (see Lemma 2.1.5).
˜ Γ Γi Σ α 0 i0 μ˜ 0
ζ ζi ξ
ϕi0 μ0 Γ Γi0
μ
˜ : ˜ → ˜ = : Let μ0 Γi0 Σ be the lifting morphism of μ0 with μ0αi0 (m ) s, where αi0 → ˜ ˜ Γ Γi0 is the projection. Then obviously, μ0αi0 is the required lifting morphism of μ. This proves the claim and part (a). Part (b) follows from Proposition 3.1.2.
Proposition 3.3.3 (a) Let T be a finite tree. Then T is C-simply connected for every pseudovariety of finite groups C. (b) Let Γ be a profinite graph which is an inverse limit of finite trees. Then, for every pseudovariety of finite groups C, Γ is both a C-tree and a C-simply connected profinite graph.
Proof To prove part (a) we need to show that the identity morphism T → T is a universal Galois C-covering. Let β : T → be a qmorphism of T to a finite graph and let ζ : Σ → be a finite Galois C-covering. Fix m ∈ T and s ∈ Σ with β(m) = ζ(s).LetT be a subtree of T containing m and maximal with respect to the property that there exists a qmorphism α : T → Σ such that ζα = β|T and 3.4 0-Transversals and 0-Sections 77
α(m) = s. We need to prove that T = T . Suppose not. Then there exists an edge e ∈ E(T ) − E(T ) with one of its vertices in T ,sayd1(e). Since ζ is a Galois −1 covering, d (α(d1(e))) contains an edge e with ζ(e ) = β(e). Note that d0(e) = 1 d1(e) since T is a tree. Extend α to a morphism of graphs on T ∪{e}∪{d0(e)} by putting α(e) = e and α(d0(e)) = d0(e ). This contradicts the maximality of T , and so T = T . (Note that this argument is independent of the class C.) (b) By Proposition 2.4.3(d), Γ is a C-tree. That it is also C-simply connected follows from Proposition 3.3.2(b).
For a converse of Proposition 3.3.3(a), see Corollary 3.5.5.
Exercise 3.3.4 Let C be a pseudovariety of finite groups. (a) Let L(0) be the loop 0 • 0 Prove that the universal Galois C-covering of L(0) is the Cayley graph Γ(ZCˆ, {1}) of the free pro-C group ZCˆ of rank 1 (written additively) with respect { } C =∼ to its generating subset 1 . Moreover, π1 (L(0)) ZCˆ. (b) Let n be a natural number. Consider the finite graph
0 12n − 1 n L(n) : •••··· •• − 0 1 n 1 n Prove that the universal Galois C-covering graph of L(n) consists of the Cayley graph Γ(ZCˆ, {1}) of the free pro-C group ZCˆ of rank 1 with respect to the subset {1} to which one attaches a copy of the finite tree 0 12n − 1 n I(n): •••··· •• 0 1 n − 1
{ } C =∼ C =∼ to each point of Γ(ZCˆ, 1 ). Moreover, π1 (L(n)) π1 (L(0)) ZCˆ. (Hint: see Proposition 3.5.3 below.)
3.4 0-Transversals and 0-Sections
Let Γ be a connected profinite graph. A spanning profinite subgraph of Γ is a profinite subgraph T of Γ with V(T)= V(Γ).IfΓ is finite, then it has a spanning subtree; more generally, it is well-known that any connected abstract graph admits a spanning subtree (cf. Serre 1980, Proposition I.11). However, in general it need not be the case that a connected profinite graph has a spanning profinite subgraph which is a C-tree or a C-simply connected profinite subgraph, as the following example shows. 78 3 The Fundamental Group of a Profinite Graph
Example 3.4.1 (Connected graph with no spanning C-simply connected profinite subgraph or C-tree) Let N ={0, 1, 2,...} be the set of natural numbers with the discrete topology and let N¯ = N ∪. {∞} be the one-point compactification of N.De- fine a profinite graph L = N¯ ×{0, 1} with space of vertices and edges V(L)= i = (i, 0) i ∈ N¯ ,E(L)= i = (i, 1) i ∈ N¯ , and incidence maps d0,d1 : E(L) → V(L)defined by d0(i) = i and d1(i) = i + 1 (i ∈ N¯ ), where ∞+1 =∞:
0 123 ∞ L : ••••··· • 0 1 2 ∞
Observe that V(L) and E(L) are disjoint and they are both profinite spaces. Note that L is the inverse limit of the following finite connected graphs L(n) (n ≥ 0)
0 12n − 1 ∞ L(n) : •••··· •• − 0 1 n 1 ∞ where the canonical map L(n + 1) → L(n) sends i to i identically, if i ≤ n − 1, and it sends n and ∞ to ∞. One deduces that L is a connected profinite graph. We claim that any connected profinite subgraph L of L which contains all the vertices of L coincides with L; indeed, since L is connected and contains all the vertices of L, it must contain all the edges of the form i (i = 0, 1,...); therefore, since L is compact, it also contains ∞; this proves the claim. On the other hand, if C is a C =∼ C =∼ pseudovariety of finite groups, one easily sees that π1 (L) π1 (L(0)) ZCˆ (see Exercise 3.3.4 or Proposition 3.5.3 below). Hence L does not contain any spanning C-simply connected profinite subgraph. Similarly, L does not contain any spanning C-subtree since obviously L is not a C-tree (see Exercise 2.4.4).
Let G be a profinite group that acts on a profinite graph , and let ϕ : → Γ = G\ be the canonical quotient map. Let Λ be a profinite subgraph of Γ ;if there is a profinite subgraph Λ of such that ϕ|Λ is a monomorphism of graphs with ϕ|Λ (Λ ) = Λ, we say that Λ is a lifting of Λ. In contrast with the situation for abstract groups that act on abstract graphs (cf. Serre 1980, Proposition I.14), a general π-subtree or C-simply connected profinite subgraph of Γ need not have a lifting to an isomorphic profinite subgraph of , as the following example shows.
Example 3.4.2 (Quotient G-graph with no lifting of trees or simply connected sub- graphs) Let X be a profinite space on which a pro-π group G acts continuously in such a way that the canonical epimorphism ϕ : X → G\X does not admit a contin- uous section (see Sect. 1.3). Construct a profinite graph C = C(X,P), where P is a point not in X, as follows: C = V(C)∪. E(C), where V(C)= X ∪. {P },E(C)= (x, P ) x ∈ X ,d0(x, P ) = P, d1(x, P ) = x. 3.4 0-Transversals and 0-Sections 79
We term C the cone of X. Extend the action of G on X to an action of G on V(C) by letting every g ∈ G fix P . Define an action of G on E(C) as follows:
g(x,P) = (gx, P ), (g ∈ G, x ∈ X).
One checks that this defines a continuous action of G on the profinite graph C. We claim that C is both a C-tree and a C-simply connected graph, for any pseu- dovariety of finite groups C; indeed, write X as an inverse limit of finite quotient spaces Xi ; then C = lim C(X ,P); ←− i therefore C is the inverse limit of finite trees, and so the claim follows from Propo- sition 3.3.3(b). The quotient graph of C under the action of G is the cone of G\X:
G\C = C(G\X, P ).
In particular, G\C is a C-tree and a C-simply connected graph which does not have a lifting to a profinite subgraph of C(X,P), i.e., there is no morphism of profinite graphs ψ : C(G\X, P ) → C(X,P) such that ϕψ = idC(G\X,P ).
Let G be a profinite group that acts on a connected profinite graph , and let ϕ : → Γ = G\ be the canonical quotient map. A G-transversal or a transversal of ϕ is a closed subset J of such that ϕ|J : J → Γ is a homeomorphism. Associated with such a transversal there is a continuous G-section or section of ϕ
j : Γ −→ , i.e., a continuous mapping such that ϕj = idΓ and j(Γ)= J . Note that, in general, J is not a graph. We say that a transversal J is a 0-transversal if d0(m) ∈ J , for each m ∈ J ; in that case we refer to j as a 0-section. Note that if j is a 0-section, then
jd0 = d0j.
If it happens that the quotient graph Γ = G\ admits a spanning C-simply con- nected profinite subgraph, then we say that a 0-transversal J is a fundamental 0- transversal and the corresponding 0-section j : Γ → is a fundamental 0-section if for some spanning C-simply connected profinite subgraph T of Γ , T = j(T) is a lifting of T , i.e., the restriction of j to T is a morphism of graphs. Example 3.4.2 above shows that G-transversals do not exist in general. However, the following re- sult proves that their existence is guaranteed when the action of G is free. See also Proposition 3.4.5, Corollary 3.7.2 and Theorem 3.7.4.
Lemma 3.4.3 Let ζ : → Γ be a Galois C-covering of a profinite graph Γ with associated group G = G(|Γ). Assume that Λ is a lifting of a profinite subgraph Λ of Γ . Then there exists a 0-transversal J ⊆ of ζ such that Λ ⊆ J . 80 3 The Fundamental Group of a Profinite Graph
Proof Note that G\V()= V(Γ). Since V()is a profinite space and G acts on it freely, there is a continuous section
jV : V(Γ)−→ V() of ζ|V() such that {d0(m) | m ∈ Λ }⊆jV (V (Γ )) (see Sect. 1.3). Define
= −1 J d0 jV V(Γ) .
Note that Λ ⊆ J . Clearly d0(J ) ⊆ J . It remains to prove that the restriction ζ|J : J → Γ of ζ to J is a homeomorphism of topological spaces. Since J is compact, it suffices to prove that ζ|J is a bijection. Let m ∈ Γ , and let m˜ ∈ be such that ζ(m)˜ = m. Put v = jV d0(m). Since ζd0(m)˜ = ζ(v)= d0(m), there exists some g ∈ G with gd0(m)˜ = v. Put m = gm˜ . Then ζ(m ) = m, and d0(m ) = v, i.e., m ∈ J .Soζ|J is onto. Now, if m1,m2 ∈ J and ζ(m1) = ζ(m2), then there exists some g ∈ G with gm1 = m2, and so gd0(m1) = d0(m2). Since d0(m1), d0(m2) ∈ jV (V (Γ )), we deduce that d0(m1) = d0(m2); and since the action of G is free, g = 1. Therefore m1 = m2, proving that ζ|J is also an injection.
Let ζ : Γ˜ → Γ be the universal Galois C-covering of a profinite graph Γ .By Lemma 3.4.3 there exists a continuous 0-section j : Γ → Γ˜ of ζ .LetJ = j(Γ) be the corresponding 0-transversal. Associated with this transversal we are going to define two continuous maps = : ˜ −→ C = : −→ C κ κj Γ π1 (Γ ) and χ χj Γ π1 (Γ ) (3.1) which will play an important role later. If m˜ ∈ Γ˜ , define κ(m)˜ to be the unique C element of π1 (Γ ) such that
κ(m)˜ jζ(m)˜ =˜m.
For an arbitrary m ∈ Γ one has
ζd1j(m)= d1ζj(m) = d1(m) = ζjd1(m). C We define χ(m) to be the unique element of π1 (Γ ) such that
χ(m) jd1(m) = d1j(m).
Lemma 3.4.4 The following properties hold for the functions κ = κj and χ = χj defined above.
(a) κd1j(m)= χ(m), for all m ∈ Γ . ˜ = ˜ ˜ ∈ ˜ ∈ C (b) κ(hm) hκ(m), for all m Γ,h π1 (Γ ). ˜ (c) κ(m)˜ = κd0(m)˜ , for all m˜ ∈ Γ . (d) κj (m) = 1, for all m ∈ Γ ; χ(v)= 1, for all v ∈ V(Γ). ˜ (e) κ(m)(χζ(˜ m))˜ = κd1(m)˜ , for all m˜ ∈ Γ . (f) The maps κ and χ are continuous. 3.4 0-Transversals and 0-Sections 81
Proof Parts (a), (b), (c) and (d) follow easily from the definitions of κ and χ.To show (e) note that
χζ(m)˜ jζd1(m)˜ = χζ(m)˜ jd1ζ(m)˜ = d1jζ(m)˜
−1 −1 = d1 κ(m)˜ m˜ = κ(m)˜ d1(m),˜ so that κ(m)(χζ(˜ m))(jζ˜ d1(m))˜ = d1(m)˜ , proving (e). In view of (a), to prove part (f) it suffices to show that κ is continuous since d1 and j are continuous. Showing the continuity of κ is equivalent to proving that the graph = ˜ ˜ ˜ ∈ ˜ ⊆ ˜ × C Gr(κ) m, κ(m) m Γ Γ π1 (Γ ) of the function κ is closed: indeed, just observe that then Gr(κ) would be compact and so the natural projection Gr(κ) → Γ˜ , which is bijective, would be a homeo- morphism; since κ is the inverse map of this homeomorphism composed with the → C projection Gr(κ) π1 (Γ ), the result would follow. To see that Gr(κ) is closed, define a map : ˜ × C −→ ˜ γ Γ π1 (Γ ) Γ by
˜ = −1 ˜ ∈ C ˜ ∈ ˜ γ(m, h) h m h π1 (Γ ), m Γ . Clearly γ is continuous, so it will suffice to show that Gr(κ) = γ −1(J ). Note that if m˜ ∈ Γ˜ , then jζ(m)˜ = κ(m)˜ −1m˜ ; therefore
m,˜ κ(m)˜ ∈ γ −1 jζ(m)˜ ∈ γ −1(J ).
Conversely, if (m,˜ h) ∈ γ −1(J ), then h−1m˜ = j(m),forsomem ∈ Γ ; so, using (d) and (b), κ(m)˜ = κ(hj(m)) = h(κj (m)) = h, i.e., (m,˜ h) = (m,˜ κ(m))˜ ∈ Gr(κ),as needed.
One can sharpen Lemma 3.4.3 when the quotient graph Γ = G\ is finite.
Proposition 3.4.5 Let a profinite group G act on a connected profinite graph and let ϕ : → Γ = G\ be the corresponding projection. (a) Let T be a finite subtree of the graph Γ and let T be a finite subtree of that ϕ sends injectively into T . Then T lifts to a subtree of containing T . (b) Assume further that the action of G on is free and that the quotient graph Γ = G\ is finite. Let m0 ∈ . Then there exists a fundamental 0-transversal J in containing m0.
Proof (a) Let L be the set of finite subtrees of containing T which are sent L injectively into T by means of ϕ.LetT0 be a maximal element of with respect to inclusion, and let T0 be its image in T . Suppose that T0 = T . Since T is finite 82 3 The Fundamental Group of a Profinite Graph and connected, there exists an edge e of T not belonging to T0 such that one of the vertices of e is in T0,sayd0(e) ∈ V(T0); then d1(e)∈ / V(T0).Letv be a vertex of
T whose image in T is d0(e).Lete ∈ with ϕ(e ) = e. Since v and d0(e ) are in 0 the same G-orbit, there exists some g ∈ G with gd0(e ) = v . Define e = ge . Then = = ∪{ }∈L d0(e ) v and ϕ(e ) e. Since T0 e ,d1(e ) , this contradict the maximality = = of T0. Thus ϕ(T0) T0 T , as desired. (b) Since Γ is finite and connected, it has a subtree T with V(T)= V(Γ)(i.e., T is a spanning simply connected profinite subgraph of Γ ). By part (a) there exists −1 a lifting T of T such that d0(m0) ∈ V(T ). Define J = d (V (T )). Note that 0 m0 ∈ J and T ⊆ J . Then arguing as in the proof of Lemma 3.4.3, we see that J is a 0-transversal, and since T is a lifting of a maximal tree of Γ , J is a fundamental 0-transversal. Equivalently, one can describe J more explicitly: for each edge e ∈
Γ − T , choose e ∈ such that d0(e ) ∈ T and ϕ(e ) = e (this can be done since d0(e) ∈ T , and every vertex of is in the G-orbit of a vertex of T ; furthermore, such e is unique because G acts freely on ); then J consists of T together with all the chosen edges e (e ∈ Γ − T ).
3.5 Existence of Universal Coverings
In this section we prove the existence of universal Galois C-coverings of a connected profinite graph Γ . We begin with a general construction. When a profinite graph ad- mits a C-simply connected profinite subgraph, this provides an explicit construction of a universal Galois C-covering. This is the case, in particular, for all finite graphs.
3.5.1 A Special Construction of Galois Coverings Let Γ be a connected profinite graph and let T be a connected profinite subgraph of Γ with V(T)= V(Γ)(T need not equal Γ ). Denote by X = Γ/T the corresponding quotient space with canonical map ω : Γ −→ X = Γ/T.
Consider the element ∗=ω(T) as a distinguished point of X.LetF = FC(X, ∗) be the free pro-C group on the pointed profinite space (X, ∗) and think of (X, ∗) as being a subspace of FC(X, ∗) in the natural way. Define a profinite graph ΥC(Γ, T ) as follows:
ΥC(Γ, T ) = FC(X, ∗) × Γ, V ΥC(Γ, T ) = FC(X, ∗) × V(Γ), d0(r, m) = (r, d0(m)) and d1(r, m) = (rω(m), d1(m)) (r ∈ F,m∈ Γ).
Next define an action of F on the graph ΥC(Γ, T ) by
r (r, m) = r r, m r, r ∈ F,m∈ Γ . 3.5 Existence of Universal Coverings 83
Clearly this is a free action and F \ΥC(Γ, T ) = Γ . Therefore the natural epimor- phism
υ : ΥC(Γ, T ) −→ Γ that sends (r, m) to m (r ∈ F,m∈ Γ ) is a Galois C-covering.
Lemma 3.5.2 The Galois covering υ : ΥC(Γ, T ) → Γ is connected.
Proof We need to show that the graph Υ = ΥC(Γ, T ) is connected. Write Γ as an inverse limit = Γ lim←− Γi i∈I of finite quotient graphs (see Proposition 2.1.4), and let ρi : Γ → Γi be the canonical epimorphism (i ∈ I). Put Ti = ρi(T ). Then V(Ti) = V(Γi) and Ti is connected. We view Xi = Γi/Ti as a pointed space whose distinguished point ∗ is the image of Ti in Xi . One has ∗ = = ∗ (X, ) Γ/T lim←− (Xi, ), i∈I so that (RZ, Proposition 3.3.9) ∗ = ∗ F(X, ) lim←− F(Xi, ). i∈I Clearly = Υ(Γ,T) lim←− Υ(Γi,Ti). i∈I Hence we may assume that Γ is finite. If U is an open normal subgroup of F(X,∗), consider the continuous map
w wU : Γ −→ (X, ∗)→ F(X,∗) −→ F(X,∗)/U. Define the structure of a graph on
ΥU = F(X,∗)/U × Γ by imitating the construction above: its vertices are
V(ΥU ) = F(X,∗)/U × V(Γ), while d0(τ, m) = τ,d0(m) ,d1(τ, m) = τwU (m), d1(m) τ ∈ F(X,∗)/U, m ∈ Γ .
Then ΥU is a finite graph on which F(X,∗)/U acts, and = Υ(Γ,T) lim←− ΥU . U 84 3 The Fundamental Group of a Profinite Graph
Therefore it suffices to show that each ΥU is connected. To verify this consider the connected subgraph of ΥU defined as
= {1U}×Γ ∪ d1 {1U}×Γ , and observe that ΥU = (F (X, ∗)/U). Next note that wU (Γ ) generates the group F(X,∗)/U; moreover, the graphs and wU (m) have the vertex d1({1U},m)= (wU (m), m) in common (m ∈ Γ). It follows from Lemma 2.2.4(a) that ΥU is con- nected.
We aim to show that
υ : ΥC(Γ, T ) → Γ is universal if T is a spanning C-simply connected profinite subgraph of Γ .We prove this for a finite graph Γ in the next proposition and in Theorem 3.7.4 in full generality.
Proposition 3.5.3 Let Γ be a finite connected graph and let T be a maximal sub- tree of Γ (T is a spanning C-simply connected profinite subgraph of Γ according to Proposition 3.3.3(a)). Then one has the following properties.
(a) The Galois C-covering υ : ΥC(Γ, T ) → Γ constructed in 3.5.1 is universal. (b) Let (X, ∗) = (Γ /T , ∗); then C = ∗ π1 (Γ ) FC(X, ) is a free pro-C group of finite rank |Γ |−|T |. (c) The universal Galois C-covering υ : ΥC(Γ, T ) → Γ is independent of the max- imal subtree T chosen.
Proof Part (b) follows from part (a) and the construction of ΥC(Γ, T ). Part (c) is a consequence of (a) and the uniqueness of universal coverings. To prove (a) we need to check the appropriate universal property for υ : ΥC(Γ, T ) → Γ .Letβ : Γ → be a qmorphism into a connected finite graph and let ξ : Σ → be a finite connected Galois C-covering. Consider the pullback of β and ξ
α Γ Σ
ζ ξ
Γ β
Then Γ is the profinite graph Γ = (m, s) ∈ Γ × Σ β(m) = ξ(s) 3.5 Existence of Universal Coverings 85 whose space of vertices is
V Γ = Γ ∩ V(Γ)× V(Σ) , and with incidence maps defined by
di(m, s) = di(m), di(s) i = 0, 1; (m, s) ∈ Γ .
The morphisms α and ζ are the obvious projections. Define an action of G(Σ|) on Γ by
g(m,s) = (m, gs) g ∈ G(Σ|), (m, s) ∈ Γ . This action is free and the quotient graph G(Σ|)\Γ = (m, k) ∈ Γ × β(m) = k can be identified with Γ , so that the projection ζ : Γ → Γ coincides with the natural epimorphism Γ → G(Σ|)\Γ . Let u0 = (r, z) ∈ ΥC(Γ, T ) = FC(X, ∗) × Γ and s0 ∈ Σ be such that βυ(u0) = ξ(s0). We need to prove that there exists a qmorphism
α : ΥC(Γ, T ) −→ Σ of profinite graphs such that ξα = βυ and α(u0) = s0. Since the map ΥC(Γ, T ) → − ΥC(Γ, T ) given by u → r 1u(u∈ ΥC(Γ, T )) is an isomorphism of profinite graphs that commutes with υ, we may assume that u0 = (1,z). Let ∗ ∗ Γ = Γ (z, s0) be the connected component of (z, s0) in Γ .
ΥC(Γ, T )
α θ
Γ ∗
α Γ Σ υ ∗ j ζ
ζ ξ
Γ β 86 3 The Fundamental Group of a Profinite Graph
We claim that ∗ = : ∗ −→ ζ ζ|Γ ∗ Γ Γ is a Galois C-covering of Γ . Indeed, let G∗ be the maximal subgroup of G(Σ|) ∗ ∗ ∗ ∗ that leaves Γ invariant. Suppose m1,m2 ∈ Γ . Then ζ (m1) = ζ (m2) if and only ∗ ∗ if gm1 = m2,forsomeg ∈ G(Σ|); if this is the case, m2 ∈ Γ ∩ gΓ , hence gΓ ∗ ∪ Γ ∗ is connected (see Lemma 2.1.7), and therefore
gΓ ∗ ∪ Γ ∗ = Γ ∗.
It follows that g ∈ G∗. Thus ζ ∗ coincides with the natural epimorphism Γ ∗ → G∗\Γ ∗. To prove the claim it remains to show that ζ ∗(Γ ∗) = Γ . Suppose not. Then, since Γ is a finite and connected graph, there exists an edge e ∈ E(Γ ) − ζ ∗(Γ ∗) ∈ ∗ ∗ = = = such that di0 (e) V(ζ (Γ )),fori0 0ori0 1. Put w di0 (e) and choose w∗ ∈ Γ ∗ with ζ ∗(w∗) = w.Lete ∈ Γ with ζ (e ) = e. Then w∗ is a vertex of he , ∗ for some h ∈ G(Σ|). Hence the graph Γ ∪{he ,d0(he ), d1(he )} is connected and therefore it is contained in Γ ∗. It follows that
e = ζ he ∈ ζ ∗ Γ ∗ , a contradiction. Thus ζ ∗(Γ ∗) = Γ , i.e., ζ ∗ is a Galois C-covering of Γ , proving the claim. Since Γ is finite, there exists a fundamental G∗-transversal J in Γ ∗ lifting T such that (z, s0) ∈ J (see Proposition 3.4.5(b)). In particular, there is a bijection
j : Γ −→ J
∗ such that ζ j = idΓ and j|T is a graph morphism. Note that j(z) = (z, s0).Let x ∈ X −{∗}=Γ − T ; then d1(j (x)) and j(d1(x)) have the same image d1(x) in Γ . ∗ So, there exists some gx ∈ G such that
d1 j(x) = gxj d1(x) . ∗ Set g∗ = 1. Let ψ : FC(X, ∗) → G be the continuous homomorphism determined by ψ(x)= gx (x ∈ X). Define a map ∗ θ : ΥC(Γ, T ) = FC(X, ∗) × Γ −→ Γ by
θ r ,m = ψ r j(m) r ∈ FC(X, ∗), m ∈ Γ .
Then θ is a graph morphism and θ(u0) = (z, s0) (the continuity of θ follows from the continuity of ψ and the continuity of the action of G∗ on Γ ∗). Put α = α θ; then α is a lifting morphism of β with α(u0) = s0.
Corollary 3.5.4 Let ζ : Γ˜ → Γ be a universal C-covering of a finite connected graph Γ , and let T be a maximal subtree of Γ . Choose a fundamental 0-section 3.5 Existence of Universal Coverings 87
: → ˜ : → C j Γ Γ of ζ lifting T , and let χj Γ π1 (Γ ) be the corresponding map (see Eq.(3.1)). Then the pointed space (χj (Γ ), 1) with distinguished point 1 is a basis C C for the free pro- group π1 (Γ ).
Proof Since the universal covering graph Γ˜ of Γ is unique up to isomorphism, we may replace ζ : Γ˜ → Γ with the explicit construction 3.5.1
υ : ΥC(Γ, T ) → Γ. ¯ ¯ First observe that the function j : Γ → ΥC(Γ, T ) given by j(m)= (1,m) (m ∈ Γ ) : → C is a fundamental 0-section, and its corresponding map χj¯ Γ π1 (Γ ) coincides with the map : → ∗ = ∗ ⊆ ∗ = C ω Γ (Γ /T , ) (X, ) FC(X, ) π1 (Γ ), so that for this choice of fundamental 0-section the result holds. Now, let e be = ∈ ∗ = C = an edge of T . Say j(e) (α, e), where α FC(X, ) π1 (Γ ). Then d0j(e) (α, d0(e)) and d1j(e) = (α, d1(e)). Since the tree T is finite, one easily deduces = ¯ = −1 = −1 ∈ that j αj. Hence χj (m) αχj¯(m)α αω(m)α , for all m Γ . Therefore = ∗ −1 ∗ = C (χj (Γ ), 1) α(X, )α , which is a basis for FC(X, ) π1 (Γ ), as asserted.
Corollary 3.5.5 Let Γ be a finite graph and let C be a pseudovariety of finite groups. Then Γ is a tree if and only if it is C-simply connected as a profinite graph.
Proof In one direction this was proved in Proposition 3.3.3(a). Suppose now that C = Γ is simply connected and let T be a maximal subtree of Γ . Since π1 (Γ ) 1, it follows from Proposition 3.5.3(b) that Γ = T .
Recall that a pro-C group G is C-projective if it is a projective object in the category of pro-C groups, i.e., if for every pair of epimorphisms of pro-C groups α : A → B and ϕ : G → B, there exists a continuous homomorphism ϕ¯ : G → A such that αϕ¯ = ϕ:
G ϕ¯ ϕ
A B α
A profinite group is called projective if it is C-projective for the pseudovariety of all finite groups C.
Proposition 3.5.6 (See also Theorem 3.7.4) Let Γ be a connected profinite graph. Then (a) there exists a universal Galois C-covering ζ : Γ˜ → Γ of Γ ; C C C (b) the fundamental pro- group π1 (Γ ) of Γ is a -projective profinite group. 88 3 The Fundamental Group of a Profinite Graph
Proof Express Γ as an inverse limit
lim←− Γi i∈I of finite connected graphs Γi . By Proposition 3.5.3, for each i ∈ I there exists a ˜ universal Galois C-covering ζi : Γi → Γi of Γi . Then the results follow from Propo- sition 3.3.2, since the inverse limit of free pro-C groups is C-projective.
Proposition 3.5.7 Let Γ be a connected finite graph and let be a connected C C subgraph of Γ . Then π1 () is a free factor of π1 (Γ ).
Proof Choose a maximal subtree D of and extend it to a maximal subtree T C = − C = − of Γ . Then by Proposition 3.5.3 π1 () F( D) and π1 (Γ ) F(Γ T),so the result follows.
The construction described above of the graph ΥC(Γ, T ) parallels the construc- tion of the abstract standard tree Υ abs(Γ, T ) of an abstract connected graph Γ (see Appendix A.2 or Theorem I.12 in Serre 1980). Let Γ be a finite connected ab- stract graph and let T be a maximal subtree of Γ ; consider the abstract free group Φ = Φ(Γ − T)on the set Γ − T , and construct an abstract graph
Υ abs = Υ abs(Γ, T ) = Φ(Γ − T)× Γ with set of vertices V(Υabs) = Φ(Γ − T)× V(Γ) and set of edges E(Γ ) = Φ(Γ − T)× E(Γ ); define the incidence maps d0 and d1 by
d0(f, e) = f,d0(e) and d1(f, e) = fρ(e), d1(e) ,(e∈ Γ,f ∈ Φ) where the map ρ : Γ → Φ is given by 1, if e ∈ T ; ρ(e)= e, if e ∈ E(Γ − T).
abs abs = − Then the abstract fundamental group π1 (Γ ) of Γ is π1 (Γ ) Φ(Γ T). It is useful to keep in mind the relationship between the abstract and pro-C con- structions of fundamental groups and universal coverings. We make this explicit in the following proposition.
Proposition 3.5.8 Let Γ be a finite connected graph and let T be a maximal sub- tree of Γ . Then C C C (a) the fundamental pro- group π1 (Γ ) of Γ is the pro- completion
C = abs = − = − π1 (Γ ) π1 (Γ ) Cˆ Φ(Γ T) Cˆ FC(Γ T) abs − of the abstract fundamental group π1 (Γ ) of Γ [here FC(Γ T)denotes the free pro-C group with basis Γ − T ]; 3.6 Subgroups of Fundamental Groups of Graphs 89
(b)
= \ = ∩ abs \ abs ; ΥC(Γ, T ) lim←− N ΥC(Γ, T ) lim←− N π1 (Γ ) Υ (Γ, T ) C C N oπ1 (Γ ) N oπ1 (Γ )
(c) Υ abs(Γ, T ) is canonically embedded as a dense subgraph of ΥC(Γ, T ); abs C abs (d) π1 (Γ ) consists of the elements of π1 (Γ ) that leave Υ (Γ, T ) invariant; (e) Υ abs(Γ, T ) is an abstract connected component of ΥC(Γ, T ), where ΥC(Γ, T ) is viewed as an abstract graph; (f) there exists a fundamental 0-transversal J of ν : ΥC(Γ, T ) → Γ such that J ⊆ Υ abs(Γ, T ), and the corresponding function χ (see Eq.(3.1)) has its values in abs ∈ abs ∈ π1 (Γ ), i.e., χ(m) π1 (Γ ) for all m Γ .
Proof Parts (a)Ð(d) are clear. To prove part (e), suppose on the contrary that e is an edge of ΥC(Γ, T ) − Υ abs(Γ, T ) with one of its vertices v ∈ Υ abs(Γ, T ),say = = abs v d0(e ) (if v d1(e ), the argument is similar). Let J be a π1 (Γ )-transversal in abs = C ∈ C = Υ (Γ, T ). Since ΥC(Γ, T ) π1 (Γ )J , there exists a g π1 (Γ ) such that ge e , where e is an edge of J .Letw = d0(e); then gw = v. Since v and w have the same ∈ abs = −1 image in Γ , there exists a g π1 (Γ ) with g w v. Hence g g stabilizes w and −1 = C = therefore g g 1, because π1 (Γ ) acts freely on ΥC(Γ, T ). Hence g g , and so e ∈ Υ abs(Γ, T ), a contradiction. Finally, part (f) follows from Lemma 3.4.3 (using the notation of that lemma, just let Λ consist of a vertex of Υ abs(Γ, T ) that lifts a vertex of T ), and the definition of χ.
3.6 Subgroups of Fundamental Groups of Graphs
Throughout this section C denotes an extension-closed pseudovariety of finite groups.
Proposition 3.6.1 Let ζ : Γ˜ → Γ be a universal Galois C-covering of a connected C profinite graph Γ and let H be a closed subgroup of π1 (Γ ). (a) The canonical epimorphism
ξ : Γ˜ −→ H\Γ˜
is a universal Galois C-covering. C \ ˜ = (b) π1 (H Γ) H .
Proof Part (b) follows from (a). We prove (a) in three steps.
C Step 1. Assume that Γ is finite and H is an open subgroup of π1 (Γ ). 90 3 The Fundamental Group of a Profinite Graph
Put = H \Γ˜ and let
= C : r π1 (Γ ) H C ˜ (the index of H in π1 (Γ )). It follows from the construction of Γ that is a finite graph and ||=r|Γ | and V() = rV(Γ). Let T and D be maximal subtrees of Γ and , respectively. Since the number of vertices in a finite tree equals the number of its edges plus one, we have ||−|D|=r|Γ |−2rV(Γ) + 1 = r |Γ |−2V(Γ) + 1
= r |Γ |− |T |+1 + 1 = r |Γ |−|T |−1 + 1.
Hence, by Proposition 3.5.3,
C =| |−| |= C − + rank π1 () D r rank π1 (Γ ) 1 1. = C − + On the other hand, rank(H ) r(rank(π1 (Γ )) 1) 1 (cf. RZ, Theorem 3.6.2(b)). Therefore,
= C rank(H ) rank π1 () . Let ˜ → be a universal Galois C-covering of . Since the natural morphism of ˜ ˜ graphs Γ → H \Γ = is a Galois C-covering, the identity morphism id : → lifts to a qmorphism α : ˜ → Γ˜ . By Proposition 3.1.4, there exists a continuous : C → homomorphism f π1 () H compatible with α. By Proposition 3.1.5, f is C an epimorphism. Since π1 () and H are free profinite groups with the same finite rank, we deduce that f is an isomorphism (see Sect. 1.3). From Proposition 3.1.5(b) it follows that α is an isomorphism of graphs.
C Step 2. Assume that Γ is finite and H is a closed subgroup of π1 (Γ ). U C Let be the collection of all open subgroups of π1 (Γ ) containing H .Bythe ˜ ˜ case above, the canonical epimorphism ξU : Γ → U\Γ is a universal Galois C- covering (U ∈ U). Since = ξ lim←− ξU , U∈U it follows from Proposition 3.3.2 that ξ is a universal Galois C-covering.
Step 3. General case. Express Γ as an inverse limit
= Γ lim←− Γi i∈I 3.7 Universal Coverings and Simple Connectivity 91 of finite graphs Γi (see Proposition 2.1.4). By Proposition 3.3.2, the universal Galois C-coverings ˜ {ζi : Γi −→ Γi | i ∈ I} form an inverse system such that = ζ lim←− ζi i∈I and C = C π1 (Γ ) lim←− π1 (Γi). i∈I C C → C Denote by Hi the image of H on π1 (Γi) under the projection π1 (Γ ) π1 (Γi). ˜ ˜ Clearly the canonical epimorphisms ξi : Γi → Hi\Γi form an inverse system and = ξ lim←− ξi. i∈I
By Step 2, ξi is a universal Galois C-covering. Thus, by Proposition 3.3.2, ξ is a universal Galois C-covering.
3.7 Universal Coverings and Simple Connectivity
Throughout this section C denotes an extension-closed pseudovariety of finite groups.
Theorem 3.7.1 (a) Let ζ : Γ → be a Galois C-covering. Then ζ is universal if and only if Γ is C-simply connected. (b) Let Γ be a profinite graph. Then the identity morphism id : Γ → Γ is a univer- sal Galois C-covering if and only if Γ is C-simply connected. (c) Let Γ be a C-simply connected profinite graph. Then any Galois C-covering ζ : Σ → Γ is trivial, i.e., G(Σ|Γ)= 1.
: → = C N Proof (a) Assume first that ζ Γ is universal and let G π1 ().Let be the collection of all closed normal subgroups N of G. According to Proposi- : → \ C \ = tion 3.6.1, the natural qmorphism ζN Γ N Γ is universal with π1 (N Γ) N. Since = ζ lim←− ζN , N∈N we have (see Proposition 3.3.2(b)) C = C \ = = π1 (Γ ) lim←− π1 (N Γ) N 1. N∈N N∈N 92 3 The Fundamental Group of a Profinite Graph
C C = : Conversely, suppose that Γ is -simply connected, i.e., that π1 (Γ ) 1. Let ξ ˜ → be a universal Galois C-covering. By the universal property of ξ, there exists a qmorphism of profinite graphs α : ˜ → Γ such that ζα= ξ. By Proposition 3.1.5, α is a Galois covering and G(˜ |Γ) is a closed subgroup of G(˜ |). It follows from Proposition 3.6.1 that α is a universal Galois C-covering. Since Γ is C-simply ˜ | = C = ˜ = connected, G( Γ) π1 (Γ ) 1; therefore, Γ , as required. (b) Follows from (a). (c) By part (b) the identity morphism id : Γ → Γ is a universal Galois C- covering. Let α : Γ → Σ be such that ζα= id. Then by Proposition 3.1.5, G(Σ|Γ) is a quotient of G(Γ |Γ)= 1, and so G(Σ|Γ)= 1.
Corollary 3.7.2 Let ζ : → Γ be a connected Galois C-covering and let T be a C-simply connected profinite subgraph of Γ . (a) Let m ∈ such that ζ(m ) = m ∈ T . Then there exists a unique lifting T of T in such that m ∈ T . (b) If T and T are liftings of T in , there exists a unique g ∈ G(ζ ) such that T = gT . (c) Assume that T is in fact a spanning C-simply connected profinite subgraph of Γ . Then there exists a fundamental 0-transversal J in lifting T . If J is an- other such fundamental 0-transversal, then there exists a unique g ∈ G(ζ ) such J = gJ .
Proof (a) By Theorem 3.7.1(b), id : T → T is universal. Therefore there exists a morphism τ : T → such that ζτ = idT and τ(ζ(m))= m. Then T = τ(T) is the desired lifting. Note that τ is unique by Lemma 3.1.7. (b) Let m ∈ T and m ∈ T with ζ(m ) = ζ(m ) = m. Then there exists a unique g ∈ G(ζ ) such that m = gm . Since m ∈ gT and gT is also a lifting of T , it follows from (a) that T = gT . (c) This follows from parts (a) and (b) and Lemma 3.4.3.
Proposition 3.7.3 Let be a connected profinite subgraph of a connected profi- nite graph Γ and assume that ζ : Γ˜ → Γ and ξ : ˜ → are universal Galois C-coverings. Then we have the following properties. (a) There exists an injective lifting morphism ψ˜ : ˜ → Γ˜ of the natural embedding ψ : → Γ . C C (b) π1 () is isomorphic to a closed subgroup of π1 (Γ ). ˜ −1 C (c) can be identified with a connected component of ζ (), and π1 () with C the maximal subgroup of π1 (Γ ) that stabilizes that component. (d) A connected profinite subgraph of a C-simply connected profinite graph is C- simply connected.
Proof Part (d) is an immediate consequence of (b). To prove the other parts, as- sume first that Γ is finite. Choose maximal trees D and T of and Γ , respectively, so that T ∩ = D. By Propositions 3.5.3 and 3.3.1(b) we may identify ˜ and Γ˜ 3.7 Universal Coverings and Simple Connectivity 93 with (the canonically constructed graphs) ΥC(, D) and ΥC(Γ, T ), respectively, and we may assume that ξ and ζ are the natural projections. By Proposition 3.5.3, C = − C = − π1 () FC( D) is a subgroup of π1 (Γ ) FC(Γ T). From the construc- tions of ΥC(Γ, T ) and ΥC(, D) (see 3.5.1) it follows that ΥC(, D) is a profinite subgraph of ΥC(Γ, T ). − ∗ We claim that ΥC(, D) is a connected component of ζ 1(). To see this let G C be the maximal subgroup of π1 (Γ ) which leaves invariant the connected component − ∗ − (ζ 1()) of ζ 1() containing ΥC(, D). The restriction
: −1 ∗ −→ ζ|(ζ −1())∗ ζ () is a Galois C-covering with associated group G∗: indeed, since (ζ −1())∗ is a −1 maximal connected profinite subgraph of ζ (),wehavethatgm1 = m2 implies ∈ ∗ ∈ −1 ∗ ∈ C g G , whenever m1,m2 (ζ ()) , g π1 (Γ ) (see Lemma 2.1.7). Then, by Proposition 3.1.5, the natural embedding
−1 ∗ ΥC(, D) −→ ζ () is surjective and so it is an isomorphism. This proves the claim, and also parts (a), (b) and (c) when Γ is finite. Now suppose Γ is infinite. Write Γ as an inverse limit = Γ lim←− Γi i∈I of an inverse system {Γi,ϕij ,I} of finite graphs. Then = lim←− i, i∈I where i is the image of in Γi under the projection ϕi : Γ → Γi . Choose a point m ∈ and denote by mi its image in i (i ∈ I ). Denote by ˜ ˜ ζi : Γi −→ Γi and ξi : i −→ i the universal Galois C-coverings of Γi and i , respectively. By the finite case con- ˜ sidered above, i can be identified canonically with a connected component of −1 ∈ ˜ ∈ −1 ∈ ζi (i) (i I ). Choose mi ξi (mi) for every i I . By Proposition 3.3.2, there ˜ ˜ ˜ exists an inverse system {Γi, ϕ˜ij ,I}, where ϕ˜ij : Γi → Γj is the (unique) lifting of ϕij : Γi → Γj such that ϕ˜ij (m˜ i) =˜mj (i, j ∈ I,i j); moreover,
˜ = ˜ Γ lim←− Γi. i∈I ˜ ˜ Then clearly ϕij (i) ⊆ j ; furthermore ϕ˜ij (i) ⊆ j , because ϕ˜ij maps the ˜ −1 ˜ connected component of mi in ζi (i) into the connected component of mj in −1 : → ˜ : ˜ → ˜ ζi (j ). Now, denote by ρij i j and ρij i j the restrictions of ϕij 94 3 The Fundamental Group of a Profinite Graph
˜ ˜ and ϕ˜ij to i and i , respectively (i, j ∈ I,i j). Therefore, {i, ρ˜ij ,I} is an inverse system and, by Proposition 3.3.2, ˜ = ˜ lim←− i. i∈I ˜ −1 ˜ ˜ Since i is the connected component of ζi (i) containing mi , we have that is the connected component of ζ −1(Γ ) containing m˜ . This proves parts (a) and (c). Part (b) follows from the finite case above and Proposition 3.3.2.
In contrast with the situation for abstract graphs, the fundamental pro-C group C C π1 (Γ ) of a connected profinite graph Γ is not always free pro- . In fact, we show below (Theorem 3.8.3) that every projective pro-C group can be realized as the fun- C C damental pro- group π1 (Γ ) of some connected profinite graph Γ . In the next theorem we give a sufficient condition for a connected graph Γ to have a free fun- damental pro-C group.
Theorem 3.7.4 Let Γ be a connected profinite graph having a spanning C-simply connected profinite subgraph T . Then the following properties hold.
(a) The Galois C-covering υ : ΥC(Γ, T ) → Γ constructed in 3.5.1 is universal. C C ∗ = ∗ (b) π1 (Γ ) is a free pro- group on the pointed profinite space (X, ) (Γ /T , ).
Proof Part (b) follows from (a) and the construction 3.5.1. To prove (a) let ζ : Γ˜ → Γ be a universal C-covering of Γ . Choose a fundamental 0-transversal J in Γ˜ lifting T (see Corollary 3.7.2). Denote by j : Γ → J the corresponding fun- damental 0-section. Recall that j|T : T → j(T)= T is an isomorphism of profinite graphs. We shall construct an isomorphism α : ΥC(Γ, T ) → Γ˜ such that ζα= υ.
α ΥC(Γ, T ) Γ˜
υ ζ Γ
: → C Consider the continuous function χ Γ π1 (Γ ) (see Eq. (3.1) in Sect. 3.4)given by the formula
d1 j(m) = χ(m)j d1(m) (m ∈ Γ). (3.2)
We use the notation of 3.5.1, in particular recall that ΥC(Γ, T ) = FC(X, ∗) × Γ , where the pointed space X is (X, ∗) = (Γ /T , ∗); and
ω : Γ → (X, ∗) = (Γ /T , ∗) is the canonical projection. Since ω sends every t ∈ T to ∗ and χ sends t ∈ T to 1, χ induces a continuous map of pointed spaces (that we still denote by χ) 3.8 Fundamental Groups and Projective Groups 95
: ∗ → C χ (X, ) π1 (Γ ).Let : ∗ −→ C θ FC(X, ) π1 (Γ ) be the continuous homomorphism extending χ. Define a map
α : ΥC(Γ, T ) = FC(X, ∗) × Γ −→ Γ,˜ by
α(f,m) = θ(f)j(m) f ∈ FC(X, ∗), m ∈ Γ . Using equality (3.2) one easily checks that α is a qmorphism of profinite graphs. Clearly ζα = υ. By Proposition 3.1.5, α is surjective. It then follows from the C- simple connectivity of Γ˜ that α and θ are isomorphisms (see Theorem 3.7.1 and Proposition 3.1.5).
3.8 Fundamental Groups and Projective Groups
Throughout this section C denotes an extension-closed pseudovariety of finite groups.
Proposition 3.8.1 The Cayley graph Γ(FC(X, ∗), X) of a free pro-C group on a pointed profinite space (X, ∗) with respect to X is C-simply connected. In fact, Γ(FC(X, ∗), X) is the universal Galois C-covering space of the bouquet of loops = ∗ C = ∗ B B(X, ) and π1 (B) FC(X, ).
Proof Consider the bouquet of loops B = B(X,∗) associated with (X, ∗) (see Ex- ample 3.1.1). Plainly the subgraph T ={∗}is a spanning C-simply connected profi- nite subgraph of B and ΥC(B, T ) coincides with the Cayley graph Γ(FC(X, ∗), X). By Theorem 3.7.4, Γ(FC(X, ∗), X) is the universal Galois C-covering graph of B, C = ∗ ∗ C with π1 (B) FC(X, ). So, by Theorem 3.7.1, Γ(FC(X, ), X) is -simply con- nected.
Corollary 3.8.2 Let C ⊆ C be extension-closed pseudovarieties of finite groups. Let (X, ∗) be a pointed topological space and consider the natural morphism of graphs
γ : Γ FC(X, ∗), X −→ Γ FC (X, ∗), X , where F = FC(X, ∗) (respectively, F = FC (X, ∗)) is the free pro-C (respectively, pro-C ) group on the pointed space (X, ∗). Then γ is a universal Galois C-covering map with fundamental group
C ∗ = ∗ −→ ∗ π1 Γ FC (X, ), X Ker FC(X, ) FC (X, ) . 96 3 The Fundamental Group of a Profinite Graph
Proof Let B = B(X,∗) be the bouquet of loops associated with (X, ∗) (see Exam- ple 3.1.1). We have a commutative diagram
γ Γ(FC(X, ∗), X) Γ(FC (X, ∗), X)
ζ ζ B where ζ and ζ are quotient maps modulo F and F , respectively. As pointed out in Proposition 3.8.1, ζ and ζ are the universal Galois C-covering and C -covering of B, respectively. By Theorem 3.7.1(a), Γ(FC(X, ∗), X) is C-simply connected. By Proposition 3.1.5, γ is a Galois covering whose associated group is the kernel of the epimorphism
FC(X, ∗) −→ FC (X, ∗). The result then follows from Theorem 3.7.1(a) again.
Recall that (under our standing assumption that C is an extension-closed pseu- dovariety of finite groups) a pro-C group G is C-projective if and only if it is a closed subgroup of a free pro-C group, or equivalently, if and only if it is projective, i.e., if and only if it is a closed subgroup of a free profinite group (see Sect. 1.5). The next theorem gives a profinite graph-theoretic characterization of projective pro-C groups.
Theorem 3.8.3 A profinite group G is projective if and only if there exists a con- C =∼ nected profinite graph Γ such that π1 (Γ ) G, for some extension-closed pseu- dovariety of finite groups C.
C =∼ Proof If π1 (Γ ) G, then G is projective by Proposition 3.5.6 and the above com- ment. Conversely, assume that G is projective. Then G is a closed subgroup of a free profinite group F(X,∗) on some pointed profinite space (X, ∗). Then G is the fundamental profinite group of the quotient graph G\Γ(F(X,∗), X) of the Cayley graph Γ(F(X,∗), X) (see Propositions 3.6.1 and 3.8.1).
3.9 Fundamental Groups of Quotient Graphs
Throughout this section C denotes an extension-closed pseudovariety of finite groups.
In the next two propositions we show that certain types of epimorphisms of connected profinite graphs preserve C-simple connectivity. And, more generally, in some cases they preserve fundamental groups. 3.9 Fundamental Groups of Quotient Graphs 97
Proposition 3.9.1 Let be a profinite subgraph of a connected profinite graph Γ such that every connected component of is C-simply connected. Let Γ be the profinite quotient graph of Γ obtained by collapsing each connected component of to a point (see Exercise 2.1.11). Then C = C π1 (Γ ) π1 (Γ).
In particular, if Γ is C-simply connected, so is Γ.
Proof Assume first that Γ is C-simply connected. In light of Theorem 3.7.1,we need to show that : −→ idΓ Γ Γ is a universal Galois C-covering. To see this it is necessary to show that for an arbi- trary qmorphism of connected profinite graphs β : Γ → Ω, an arbitrary connected Galois C-covering ξ : Σ → Ω and any m ∈ Γ,s ∈ Σ with β(m) = ξ(s), there ex- ists a qmorphism of profinite graphs α : Γ → Σ such that β = ξα and α(m) = s.
ω Γ Σ α τ ξ
Γ Ω β
−1 Let τ : Γ → Γ be the natural epimorphism and let m0 ∈ τ (m). Since Γ is C-simply connected, there exists a qmorphism of profinite graphs ω : Γ → Σ such ∗ that ξω = βτ and ω(m0) = s. We claim that the image ω( ) of any connected component ∗ of consists of a single point, i.e., one vertex. Since ω(∗) is connected, this is equivalent to showing that E(ω(∗)) =∅. Suppose not, and let e ∈ E(∗) be such that ω(e) ∈ E(ω(∗)). Then
ξω(e)= βτ(e) = β(v) ∈ V(Ω),
∗ where v is the vertex of Γ obtained from collapsing . However, since ξ is a Ga- lois C-covering, ξ(E(Σ)) ⊆ E(Ω). This is a contradiction, and therefore the claim is established. Thus the map
α : Γ −→ Σ −1 given by α(m) = ω(τ (m)) (m ∈ Γ) is a well-defined qmorphism of profinite graphs. Obviously α satisfies the desired properties. Now let Γ be an arbitrary connected profinite graph and let ζ : Γ˜ → Γ be a universal Galois C-covering. Let Θ denote the graph obtained by collapsing the connected components of ζ −1(). By the case above, the graph Θ is C-simply con- C ˜ C nected. The free action of π1 (Γ ) on Γ induces an action of π1 (Γ ) on Θ (since 98 3 The Fundamental Group of a Profinite Graph
−1 C C ζ () is π1 (Γ )-invariant and the action of an element of π1 (Γ ) sends a con- nected component to a connected component). The stabilizers of the connected com- ponents of ζ −1() are isomorphic to the fundamental groups of the connected com- ponents of (see Proposition 3.7.3). Therefore these stabilizers are trivial since, by assumption, the connected components of are C-simply connected. Thus the C C π1 (Γ )-stabilizers of all vertices of Θ are trivial, i.e., π1 (Γ ) acts freely on Θ. Since Θ is C-simply connected, −→ C \ Θ π1 (Γ ) Θ is a Galois C-covering which is universal by Theorem 3.7.1. Clearly Γ = C \ π1 (Γ ) Θ. So, C = C π1 (Γ ) π1 (Γ), as desired.
Proposition 3.9.2 Let G be a pro-C group acting on a C-simply connected profinite graph Γ . Suppose that G is generated by the stabilizers of the elements of Γ :
G = Gm | m ∈ Γ .
Then the quotient graph G\Γ is C-simply connected.
Proof By Theorem 3.7.1, it is sufficient to check that
idG\Γ : G\Γ −→ G\Γ is a universal Galois C-covering. Let β : G\Γ → be a qmorphism of profinite graphs and let ξ : Σ → be a Galois C-covering. Choose x ∈ G\Γ,s ∈ Σ with β(x) = ξ(s). To check the universal property of idG\Γ we need to construct a mor- phism α : G\Γ −→ Σ such that ξα = β and α(x) = s.Letτ : Γ → G\Γ be the natural epimorphism −1 and let x0 ∈ τ (x). Since Γ is C-simply connected, we have that idΓ is universal; hence, there exists a morphism ω : Γ → Σ such that ξω= βτ and ω(x0) = s.
ω Γ Σ α τ ξ
G\Γ β
We claim that the image ω(Gm0) of the G-orbit of an element m0 ∈ Γ consists of exactly one element, namely ω(m0). By continuity it suffices to show that if m ∈ Γ 3.9 Fundamental Groups of Quotient Graphs 99 and gm ∈ Gm , then
ω(gm m0) = ω(m0). Note that
ξ ω(gm m0) = βτ(gm m0) = βτ(m0) = ξ ω(m0) .
Therefore, ω(gm m0) = gω(m0),forsomeg ∈ G(Σ|). We shall show that g = 1. Denote by δ and γ the automorphisms of Γ and Σ defined by multiplication by gm and g, respectively.
ω Γ Σ
δ γ ω Γ Σ α τ ξ
G\Γ β
Consider the morphisms ωδ,γ ω : Γ → Σ. One easily checks that ξωδ = ξγω; moreover, by the definition of g, ωδ(m0) = γω(m0). Hence, by Lemma 3.1.7, ωδ = γω.
It follows that
ω m = ω gm m = ωδ m = γω m = gω m . By the freeness of the action of G(Σ|) on Σ,wehaveg = 1, as desired. This proves the claim. Thus, one can define α : G\Γ −→ Σ by α(m) = ω(τ−1(m)). Clearly α satisfies the desired properties.
Corollary 3.9.3 Let G be a pro-C group acting on a C-simply connected profinite graph Γ . Let
N = Gm | m ∈ Γ be the closed subgroup of G generated by the stabilizers Gm of the elements m ∈ Γ . Then N is normal in G and C \ =∼ π1 (G Γ) G/N. In particular, the group G/N is projective. Consequently, if G acts freely on Γ , then G is projective. 100 3 The Fundamental Group of a Profinite Graph
−1 Proof The normality of N follows from the equality gGmg = Ggm. Plainly G/N acts freely on N\Γ ; so the natural epimorphism ζ : N\Γ −→ G\Γ = (G/N)\(N\Γ) is a Galois C-covering with associated group G((N\Γ)| (G\Γ))= G/N.By Proposition 3.9.2, N\Γ is C-simply connected. Therefore, by Theorem 3.7.1, ζ is C \ = universal. This means that π1 (G Γ) G/N. Finally, G/N is projective by Theo- rem 3.8.3.
3.10 π-Trees and Simple Connectivity
In this section we study the relationship between the notions of π-tree and simple connectivity. For the concept of complete tensor product of profinite modules that we use in the next theorem, see Sect. 1.8.
Theorem 3.10.1 Let C be a pseudovariety of finite groups which is closed under extensions with abelian kernel. Assume that Γ is a connected profinite graph and let ζ : Γ˜ → Γ be its universal Galois C-covering graph. Then Γ˜ is a C-tree.
Proof Case 1. Γ is a finite graph. Let T be a maximal subtree of Γ . Put X = Γ − T , the set of edges of Γ not in T , and let F = FC(X) be the free pro-C group on the set X. In this case Γ˜ (see 3.5.1 and Proposition 3.5.3) can be described as having vertices V(Γ)˜ = F × V(Γ)= F × V(T), edges
E(Γ)˜ = F × E(Γ ) = (F × X)∪. F × E(T ) ˜ ˜ and incidence maps d0,d1 : E(Γ) → V(Γ) defined as follows: d0(f, x) = (f, d0(x)), d1(f, x) = (f x, d1(x)), for all f ∈ F , x ∈ X, and d0(f, e) = (f, d0(e)), d1(f, e) = (f, d1(e)), for all f ∈ F , e ∈ E(T ). One easily sees that, in this case, to show that Γ˜ is a C-tree we have to prove that the sequence
˜ d ˜ ε 0 −→ ZCˆE(Γ) −→ ZCˆV(Γ) −→ ZCˆ −→ 0 (3.3) is exact, where d is the continuous homomorphism of free profinite ZCˆ-modules ˜ ˜ that restricts to d1 − d0 on E(Γ), and ε is the map that sends each v ∈ V(Γ) to 1 ∈ ZCˆ. Since T is a finite tree we know that
−→ d ε 0 ZCˆE(T ) −→ ZCˆV(T) −→ ZCˆ −→ 0 (3.4) is exact. 3.10 π-Trees and Simple Connectivity 101
For a profinite space Y , let us denote the kernel of the augmentation homomor- phism ε :[[ZCˆY ]] → ZCˆ by (( I Y )) . Clearly
[[ ZCˆY ]] = (( I Y )) ⊕ ZCˆ.
One check that if Y1 and Y2 are profinite spaces, then
ZCˆ(Y1 × Y2) =[[ZCˆY1]] ⊗[[ZCˆY2]] . It follows that
˜ ZCˆE(Γ) = ZCˆ(F × X) ⊕ ZCˆ F × E(T ) and
˜ ZCˆV(Γ) =[[ZCˆF ]] ⊗ ZCˆV(T) = (( I F )) ⊕ ZCˆ ⊗ IV(T) ⊕ ZCˆ
= (( I F )) ⊗ ZCˆ ⊕ (ZCˆ ⊗ ZCˆ) ⊕ [[ ZCˆF ]] ⊗ IV(T)
= (( I F )) ⊕ ZCˆ ⊕ [[ ZCˆF ]] ⊗ IV(T) . (3.5)
Next observe that the restriction of d to [[ ZCˆ(F × X)]] is the map
ZCˆ(F × X) → (( I F )) that sends (f, x) to fx− f , which is an isomorphism as one easily deduces from Lemma 2.5.2. On the other hand, the restriction of d to [[ ZCˆ(F × E(T ))]] may be identified with the homomorphism
id⊗ˆ d [[ ZCˆF ]] ⊗ ZCˆE(T ) −→ [[ ZCˆF ]] ⊗ ZCˆV(T) , ˆ where d(e) = d1(e) − d0(e),fore ∈ E(T ). Now note that id⊗d is injective because it is obtained by tensoring (3.4) with [[ ZCˆF ]] , and this is a free ZCˆ-module (see ˜ Sect. 1.8). This proves that the sequence (3.3) is exact at [[ ZCˆE(Γ)]] . ˜ To prove exactness at [[ ZCˆV(Γ)]] first note that from the above considerations and the exactness of (3.4)wehave
˜ d ZCˆE(Γ) = (( I F )) ⊕ [[ ZCˆF ]] ⊗ IV(T) ; ˜ and, from the description of [[ ZCˆV(Γ)]] in (3.5), one has that this coincides with Ker(ε), as needed.
Case 2. General connected profinite graph Γ .
Write Γ as an inverse limit of its finite quotient graphs Γi (see Proposition 2.1.4). ˜ ˜ Then Γ is the inverse limit of the universal C-covering graphs Γi (see Proposi- ˜ ˜ tion 3.3.2). By Case 1, each Γi is a C-tree, and therefore so is Γ by Proposi- tion 2.4.3(d). 102 3 The Fundamental Group of a Profinite Graph
Corollary 3.10.2 Let C be an extension-closed pseudovariety of finite groups. Then every C-simply connected profinite graph Γ is a C-tree.
Proof By Theorem 3.7.1(b), Γ is its own universal Galois C-covering graph. Hence the result is a consequence of Theorem 3.10.1.
In contrast to the corollary above, see Example 3.10.6. Let R be a profinite ring and let G be a profinite group. Recall that a profinite G-space X is a profinite space on which G acts continuously. For a G-space X,the free R-module [[ RX]] with basis X naturally becomes a profinite [[ RG]] -module, where [[ RG]] is the complete group algebra. Similarly, if (X, ∗) is a pointed profinite G-space, then the free profinite R-module [[ R(X,∗)]] is naturally a profinite [[ RG]] - module.
Lemma 3.10.3 Let G be a profinite group and let R be a commutative profinite ring. Assume that B is an R-module, which we view as an [[ RG]] -module with trivial G action. (a) If X is a profinite G-space, there exists an isomorphism of complete tensor products
∼ B ⊗[[ RG]] [[ RX]] = B ⊗R R(G\X) . (b) If (X, ∗) is a pointed profinite G-space, there exists an isomorphism of complete tensor products
∼ B ⊗[[ RG]] R(X,∗) = B ⊗R R(G\X, ∗) .
Proof The proofs of parts (a) and (b) are similar; we only prove (a). Write X as an inverse limit X = lim X ←− i of finite G-spaces X (see Sect. 1.3). Then G\X = lim(G\X ). Correspondingly i ←− i one has decompositions ∼ B ⊗[[ ]] [[ RX]] = lim B ⊗[[ ]] [[ RX ]] RG ←− RG i and
B ⊗ R(G\X) =∼ lim B ⊗ R(G\X ) . R ←− R i Hence it suffices to prove the lemma when X is finite. In that case the complete tensor product coincides with the usual tensor product (cf. RZ, Proposition 5.5.3(d)) and the result becomes
∼ B ⊗[[ RG]] [RX] = B ⊗R R(G\X) . 3.10 π-Trees and Simple Connectivity 103
To verify this, consider the map ϕ : B × X → B ⊗R [R(G\X)] given by ϕ(b,x) = b ⊗ Gx (b ∈ B,x ∈ X). This map extends naturally to a middle G-linear map ϕ : B ×[RX]→B ⊗R [R(G\X)], and therefore it induces a homomorphism
ϕ : B ⊗[[ RG]] [RX]−→B ⊗R R(G\X) .
This homomorphism is an isomorphism because, as one easily checks, it has an inverse induced by the map
B × G\X −→ B ⊗[[ RG]] [RX] given by (b, Gx) → b ⊗ x(b∈ B,x ∈ X).
Proposition 3.10.4 Let Γ be a profinite connected graph and let C be a pseudova- riety of finite groups which is closed under extensions with abelian kernel. Then C (a) H1(Γ, ZCˆ) is the abelianized group of π1 (Γ ):
=∼ C =∼ C C C H1(Γ, ZCˆ) H1 π1 (Γ ), ZCˆ π1 (Γ )/ π1 (Γ ), π1 (Γ ) . (b) If C consists of solvable groups, then every C-tree is C-simply connected.
Proof Set π = π(C). Since C is closed under extensions with abelian kernel, one = = has ZCˆ Zπˆ p∈π Zp. (a) Let ζ : Γ˜ → Γ be a universal Galois C-covering. By Theorem 3.10.1, Γ˜ is a π-tree. Hence one has the following exact sequence of Zπˆ -modules
∗ ˜ d ˜ ε 0 [[ Zπˆ (E (Γ),∗)]] [[ Zπˆ (V (Γ))]] Zπˆ 0. (3.6)
= C ˜ ∗ ˜ ∗ Put G π1 (Γ ). The action of G on Γ , induces actions on the spaces (E (Γ), ) ˜ ∗ ˜ ˜ and V(Γ); this implies that [[ Zπˆ (E (Γ),∗)]] and [[ Zπˆ (V (Γ))]] are also [[ Zπˆ G]] - modules (cf. RZ, Proposition 5.3.6). We think of Zπˆ as a [[ Zπˆ G]] -module with trivial G-action. Furthermore, because of the definition of these actions and the definition of d and ε, we deduce that these maps are [[ Zπˆ G]] -linear. Therefore the sequence (3.6) is in fact a short exact sequence of [[ Zπˆ G]] -modules. Using the left derived [[ ]] { Zπˆ G − } ⊗ − functors Torn (Zπˆ , ) of the functor Zπˆ [[ Zπˆ G]] , we get a corresponding long exact sequence of Zπˆ -modules (see Sect. 1.9) [[ ]] ···→ Zπˆ G ˜ → Tor1 Zπˆ , Zπˆ V(Γ) H1(G, Zπˆ )
→ ⊗ ∗ ˜ ∗ Zπˆ [[ Zπˆ G]] Zπˆ E (Γ),
−→d ⊗ ˜ −→ε ⊗ → Zπˆ [[ Zπˆ G]] Zπˆ V(Γ) Zπˆ [[ Zπˆ G]] Zπˆ 0, ⊗ − where d and ε denote the maps obtained after applying the functor Zπˆ [[ Zπˆ G]] to the corresponding maps d and ε in the sequence (3.6). Now, since the action of G 104 3 The Fundamental Group of a Profinite Graph
˜ ˜ on V(Γ)is free, [[ Zπˆ (V (Γ))]] is a free [[ Zπˆ G]] -module (cf. RZ, Proposition 5.7.1). Therefore [[ ]] Zπˆ G ˜ = Tor1 Zπˆ , Zπˆ V(Γ) 0. On the other hand, ⊗ = = Zπˆ [[ Zπˆ G]] Zπˆ H0(G, Zπˆ ) Zπˆ , since the action of G on Zπˆ is trivial (see Sect. 1.10). Thus, using Lemma 3.10.3 and the fact that G\Γ˜ = Γ , the last terms of the above exact sequence become
∗ d ε 0 −→ H1(G, Zπˆ ) −→ Zπˆ E (Γ ), ∗ −→ Zπˆ V(Γ) −→ Zπˆ −→ 0.
This implies (see the definition of H1(Γ, Zπˆ ) in Sect. 2.3) that
H1(Γ, Zπˆ ) = H1(G, Zπˆ ).
Finally, it is known (see Sect. 1.10) that
H1(G, Zπˆ ) = G/[G, G].
(b) If T is a C-tree, then H1(T , Zπˆ ) = 0. So, by part (a),
C = C C π1 (T ) π1 (T ), π1 (T ) . C C C = If consists of solvable groups, π1 (T ) is prosolvable, and thus π1 (T ) 1, i.e., T is C-simply connected.
Corollary 3.10.5 Let C be a pseudovariety of finite groups closed under exten- sions with abelian kernel. Let Γ be a connected profinite graph. Then the following conditions are equivalent: (a) Γ is a C-tree; C C (b) the fundamental pro- group π1 (Γ ) of Γ is perfect, i.e., it coincides with its commutator subgroup.
Proof By definition, Γ is a C-tree if and only if H1(Γ, ZCˆ) is trivial. Therefore the result follows from Proposition 3.10.4(a).
The next example shows that a π-tree is not always simply connected.
Example 3.10.6 (A C-tree which is not C-simply connected) Let (X, ∗) be a pointed profinite space with |X| > 2. Let Cs be the pseudovariety of all finite solvable groups and let C be the pseudovariety of all finite groups. Then π(Cs) = π(C); put π = C = C = ∗ = ∗ π( s) π( ).LetFs FCs (X, ) and F FC(X, ) be the free prosolvable and the free profinite group on (X, ∗), respectively. Consider the Cayley graphs Γs = Γ(Fs,X)and Γ = Γ(F,X)of Fs and F with respect to X, respectively. According 3.11 Free Pro-C Groups and Cayley Graphs 105 to Theorem 2.5.3, Γs is a Cs -tree, i.e., a π-tree, or equivalently, C-tree. We claim that C C Γs is not -simply connected. Indeed, according to Corollary 3.8.2, π1 (Γs) equals the kernel of the natural continuous epimorphism = ∗ −→ = ∗ F FC(X, ) Fs FCs (X, ), which obviously is not trivial.
We conclude this section with an example that shows that the notion of a π-tree depends on the set of primes π.
Example 3.10.7 (The notion of a π-tree depends on π) Here we exhibit examples to make explicit that the notion of a π-tree depends on the choice of the set of primes π. Let p and q be different prime numbers. Let Cp be the pseudovariety of all finite p-groups, and let Cp,q be the pseudovariety of all finite groups whose order is of the form piqj , where i, j are natural numbers. Let (X, ∗) be a profinite pointed space with |X| > 1, and let
= ∗ Γ Γ FCp (X, ), X ∗ ∗ be the Cayley graph of the free pro-p group FCp (X, ) on (X, ) with respect to X. By Theorem 2.5.3, Γ is a p-tree. Cp,q By Corollary 3.8.2, π1 (Γ ) is the kernel of the natural continuous epimorphism ∗ −→ ∗ FCp,q (X, ) FCp (X, ).
C p,q = Hence π1 (Γ ) 1. On the other hand, by a well-known theorem of Burnside C C p,q (cf. Hall 1959, Theorem 9.3.2), all groups in p,q are solvable, so that π1 (Γ ) is prosolvable, and therefore nonperfect. So, according to Corollary 3.10.5, Γ is not a {p,q}-tree. Since Γ is a p-tree, it follows that it is not a q-tree (see Proposi- tion 2.4.3(e)).
3.11 Free Pro-C Groups and Cayley Graphs
Throughout this section C denotes an extension-closed pseudovariety of finite groups.
Let X be a closed subset of a pro-C group G.NextwegiveacriterionforG to be a free pro-C group on X in terms of Cayley graphs.
Theorem 3.11.1 Let G beapro-C group and let X be a closed subset of G such that 1 ∈ X. Then G is the free pro-C group on the pointed space (X, 1) if and only if the Cayley graph Γ(G,X)of G with respect to X is C-simply connected. 106 3 The Fundamental Group of a Profinite Graph
Proof Denote by μ : X → G the inclusion map. View X as a pointed space with distinguished point ∗=1. Let
f : FC(X, ∗) −→ G be the continuous homomorphism from the free pro-C group FC(X, ∗) on (X, ∗) into G induced by μ. Consider the bouquet of loops B = B(X,∗) (see Example 3.1.1). Its univer- sal Galois C-covering is ζ : Γ(FC(X, ∗), X) → B; therefore, by Proposition 3.8.1, Γ(FC(X, ∗), X) is C-simply connected. C C = Conversely, assume that Γ(G,X)is -simply connected, i.e., π1 (Γ (G, X)) 1. Then, by definition, Γ(G,X)is a connected profinite graph. [Hence G = X (see Lemma 2.2.4(c)), and so f is surjective.] Therefore ξ : Γ(G,X)→ B is a connected Galois C-covering of B with associated group G (see Example 3.1.1). Consider the commutative diagram
α Γ(FC(X, ∗), X) Γ(G,X)
ζ ξ B where α is the epimorphism of profinite graphs given by α(g,x) = (f (g), μ(x)). By Proposition 3.1.5, α is a Galois C-covering whose associated group is G(α) = Ker(f ). Since Γ(FC(X, ∗), X) is C-simply connected,
α : Γ FC(X, ∗), X → Γ(G,X) is the universal Galois C-covering of Γ(G,X).So
= C = = 1 π1 Γ(G,X) G(α) Ker(f ).
Thus f is an isomorphism and G =∼ F(X,∗), as desired.
Corollary 3.11.2 Assume that all the groups in the class C are solvable. Let G be apro-C group and let X be a closed subset of G such that 1 ∈ X. Then G is a free pro-C group on the pointed space (X, 1) if and only if the Cayley graph Γ(G,X)is a C-tree.
Proof Since the groups in C are solvable, a profinite graph is a C-tree if and only if it is C-simply connected (see Corollary 3.10.2 and Proposition 3.10.4(b)). Hence the result follows from Theorem 3.11.1. 3.12 Change of Pseudovariety 107
3.12 Change of Pseudovariety
In this section we study how universal Galois C-coverings, fundamental pro-C groups, etc. change when the pseudovariety C changes. Let C ⊆ C be pseudovarieties of finite groups. Let Γ be a given connected profi- nite graph. Denote by
ζ : Γ˜ → Γ and ζ : Γ˜ → Γ the universal Galois C-covering and universal Galois C -covering of Γ , respectively. Then there exists an epimorphism of profinite graphs γ : Γ˜ → Γ˜ making the fol- lowing diagram
γ Γ˜ Γ˜
ζ ζ Γ commutative, since ζ is a Galois C -covering. In addition, there exists a continuous : C → C epimorphism f π1 (Γ ) π1 (Γ ) such that γ(gm)= f(g)γ(m), ∈ C ∈ ˜ for all g π1 (Γ ), m Γ (see Proposition 3.1.4). By Proposition 3.1.5(b) and C ˜ ˜ | ˜ = C ˜ Theorem 3.7.1(a), γ is a universal Galois -covering of Γ and G(Γ Γ) π1 (Γ) can be identified with Ker(f ). Choose a 0-transversal j of ζ and a 0-transversal jγ of γ (see Lemma 3.4.3). Define j = jγ j. Then one easily verifies the following result.
Lemma 3.12.1 (a) j is a 0-transversal of ζ : Γ˜ → Γ .
(b) Let χ : Γ → πC(Γ ) and χ : Γ → πC (Γ ) be the functions defined 1 1 by χ(m)(jd1(m)) = d1j(m) and χ (m)(j d1(m)) = d1j (m) (m ∈ Γ) (see Eq.(3.1) in Sect. 3.4). Then
f χ (m) = χ(m) (m∈ Γ).
For a profinite group G, denote by RC(G) the smallest closed normal subgroup of G such that G/RC(G) is a pro-C group. Clearly, RC(G) is the intersection of all closed normal subgroups N of G such that G/N is a pro-C group. (See RZ, Sect. 3.4, for properties of RC(G).)
Proposition 3.12.2 Let C and C be extension-closed pseudovarieties of finite groups, with C ⊆ C , and let Γ be a connected profinite graph. Assume that
ζ : Γ˜ −→ Γ and ζ : Γ˜ −→ Γ 108 3 The Fundamental Group of a Profinite Graph are a universal Galois C-covering and a universal Galois C -covering of Γ , respec- tively. Then
˜ = C \ ˜ C = C C Γ RC π1 (Γ ) Γ and π1 (Γ ) π1 (Γ )/RC π1 (Γ ) .
Proof We continue with the notation in the diagram above. By Proposition 3.1.5
C C ˜ =∼ C π1 (Γ )/π1 (Γ) π1 (Γ ), and hence
C ˜ ≥ C π1 (Γ) RC π1 (Γ ) . C C ˜ C = It follows that the pro- group π1 (Γ)/RC(π1 (Γ )) acts freely on C \ ˜ ˜ RC(π1 (Γ )) Γ , and the quotient graph of modulo this action coincides with Γ . Since Γ˜ is C-simply connected, the group
| ˜ = C ˜ C G( Γ) π1 (Γ)/RC π1 (Γ ) associated with the Galois C-covering → Γ˜ is trivial, and therefore
C ˜ = C π1 (Γ) RC π1 (Γ ) .
Thus the two assertions in the proposition follow.
Proposition 3.12.3 Assume that C is a pseudovariety of finite groups that is closed under extensions with abelian kernel, and let Ce be the smallest extension-closed pseudovariety of finite groups containing C. Let Γ be a connected profinite graph and let ζ : Γ˜ −→ Γ be its universal Galois C-covering. Then Γ˜ is a Ce-tree.
Proof By Corollary 3.10.5, this is equivalent to showing that the profinite group Ce ˜ π1 (Γ) is perfect, i.e., it coincides with the closure of its commutator subgroup. Let ζ e : Γ˜ e → Γ be a universal Galois Ce-covering of Γ . By Proposition 3.1.5 there exists a Galois covering ψ : Γ˜ e → Γ˜ such that ζψ = ζ e. By Theorem 3.7.1 ψ Ce ˜ e| ˜ = Ce ˜ is a universal Galois -covering and G(Γ Γ) π1 (Γ). Consider the closure
= Ce ˜ Ce ˜ C π1 (Γ),π1 (Γ) Ce ˜ of the derived subgroup of π1 (Γ). Put
= C\Γ˜ e. 3.12 Change of Pseudovariety 109
The natural epimorphisms
β : Γ˜ e −→ and ϕ : −→ Γ˜ are Galois Ce-coverings with associated groups Ce ˜ C and π1 (Γ)/C, Ce ˜ respectively (see Proposition 3.6.1). According to Proposition 3.1.5, π1 (Γ) Ce Ce Ce π1 (Γ );soC π1 (Γ ). Hence the group π1 (Γ )/C acts freely on and
Ce \ = π1 (Γ )/C Γ.
Let ζ¯ e : → Γ be the corresponding Galois Ce-covering. Note that ζϕ= ζ¯ e.
ϕ β
α Γ˜ e Γ˜ ψ ζ¯ e ζ e ζ Γ
By Proposition 3.1.5 G(Γ˜ |Γ)= G(|Γ)/G(|Γ),˜ | ˜ | = C so that G( Γ)is an extension of G(Γ Γ) π1 (Γ ) with abelian kernel | ˜ = Ce ˜ G( Γ) π1 (Γ)/C. We observe that a finite abelian group belongs to a pseudovariety of finite groups that is closed under extensions with abelian kernel if and only if for every prime p that divides the order of this group, the cyclic group Cp of order p belongs to the e pseudovariety. Therefore, since Cp ∈ C if and only if Cp ∈ C , the group
| = Ce ˜ G( Γ) π1 (Γ)/C is a pro-C group. Therefore ζ¯ e is a Galois C-covering. Now, since ζ : Γ˜ → Γ is a universal Galois C-covering, there exists an epimorphism α : Γ˜ → such that ζ¯ eα = ζ . Then by Proposition 3.3.1 ϕα is an isomorphism, and therefore so is ϕ.It Ce ˜ follows that π1 (Γ)/Cis trivial, as required. Chapter 4 Profinite Groups Acting on C-Trees
4.1 Fixed Points
Throughout this section π denotes a nonempty set of prime numbers.
Here we begin to study the structure of pro-π groups acting on π-trees (a pro- π group is an inverse limit of finite π-groups, in other words, finite groups whose orders are divisible only by primes in π). This section is concerned with fixed points. In particular, we shall prove that if a pro-π group acts on a π-tree, the subset of fixed points is a π-subtree (if it is not empty) and that a finite π-group acting on a π-tree fixes a vertex. If F = F(x,y) is a free pro-π group of rank 2, then the Cayley graph Γ = Γ(F,{x,y}) is a π-tree on which F acts (see Theorem 2.5.3); note that the quotient graph F \Γ is not a π-tree, but a bouquet of two loops (see Example 3.1.1). In contrast we have the following result.
Proposition 4.1.1 Let a pro-π group G act on a π-tree T . Let
N = Gm | m ∈ T , where Gm denotes the G-stabilizer of m. Then N\T is a π-tree.
Proof Let Cs(π) be the class of all finite solvable π-groups. By Proposition 3.10.4(b) T is Cs(π)-simply connected. Then, by Proposition 3.9.2,soisN\T . Therefore, by Corollary 3.10.2, N\T is a π-tree.
Theorem 4.1.2 Let a pro-π group G act freely on a π-tree T . Then G is a projec- tive profinite group.
Proof By Proposition 2.4.3(e), T is a p-tree for every p ∈ π, i.e.,
∗ d ε 0 −→ Zp E (T ), ∗ −→ Zp V(T) −→ Zp −→ 0
© Springer International Publishing AG 2017 111 L. Ribes, Profinite Graphs and Groups, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics 66, DOI 10.1007/978-3-319-61199-0_4 112 4 Profinite Groups Acting on C-Trees is a short exact sequence of free Zp-modules. Since the action of G on T is ∗ free, [[ Zp(E (T ), ∗)]] and [[ Zp(V (T ))]] are free [[ ZpG]] -modules (cf. RZ, Proposi- tion 5.7.1). Therefore the above sequence is a projective [[ ZpG]] -resolution of Zp. Hence cdp(G) ≤ 1, for each p ∈ π. Thus G is projective (see Sect. 1.11).
See Corollary 9.3.2(b) for an alternative proof of the above theorem using ho- mology.
Corollary 4.1.3 Let a pro-π group G act on a π-tree T and let N = Gm | m ∈ T be the subgroup generated by the stabilizers of all m ∈ T . Then N is normal in G and G/N is projective.
Proof By Proposition 4.1.1, N\T is a π-tree. Clearly N is normal in G and G/N acts freely on N\T (see Lemma 2.2.1(b)). Then, by Theorem 4.1.2, G/N is projec- tive.
Lemma 4.1.4 Let G be a finite group of prime order p generated by an element g. Let M be a free profinite [FpG]-module on a pointed profinite space (X, ∗) and let ν and τ be the endomorphisms of M defined by
− ν(m) = 1 + g +···+gp 1 m and τ(m)= (g − 1)m (m ∈ M), respectively. Then Im(ν) = Ker(τ).
Proof We first assume that the space (X, ∗) is finite. Since τ and ν commute with direct sums, it suffices to check in this case the result for a free module of rank 1, i.e., when M =[FpG]. Clearly τν = 0, and so Ker(τ) ≥ Im(ν). To prove equality, let m ∈[FpG] with τ(m)= 0. Say p−1 i m = aig (ai ∈ Fp). i=0 Then p−1 i τ(m)= (ai−1 − ai)g + ap−1 − a0 = 0. i=1 Therefore, a0 = a1 = ··· = ap−1, and so m = ν(a0). Thus, Ker(τ) ≤ Im(ν),as needed. Now, for a general profinite pointed space (X, ∗), write (X, ∗) = lim(X , ∗), ←− i where each (Xi, ∗) is a finite pointed space. Then M = lim M , ←− i where Mi is a free [FpG]-module with basis (Xi, ∗). Since ν and τ commute with inverse limits, the result now follows from the case above. 4.1 Fixed Points 113
Theorem 4.1.5 Suppose that a pro-π group G acts on a π-tree T . Then the fol- lowing results hold. (a) The subset T G ={m ∈ T | gm = m, for all g ∈ G} of fixed points of T under the action of G is either empty or a π-subtree of T . (b) If N is a closed normal subgroup of G and T N = ∅, then T N is a G-invariant π-subtree of T .
Proof Part (b) follows from (a). Here we prove part (a). If T G = T (i.e., the action is trivial), then the result is obvious. So we may suppose that the action of G on T is not trivial. From the continuity of the action we have that T G is a closed subset of T . Assume that T G = ∅. It is clear that T G is a profinite subgraph of T with vertex set V(TG) = (V (T ))G, and that E∗(T G) = (E∗(T ))G. By Proposition 2.4.3(b), it is enough to check that T G is connected. We proceed in steps.
Step 1. Assume that G is cyclic of prime order p ∈ π. G By Proposition 2.3.2, it suffices to check that H0(T , Fp) = 0. One may con- ∗ G G sider the modules [[ Fp((E (T )) , ∗)]] and [[ Fp(V (T )) ]] as submodules of ∗ [[ Fp(E (T ), ∗)]] and [[ FpV(T)]] , respectively. Then one has ∗ ∗ G ∗ ∗ G Fp E (T ), ∗ Fp E (T ) , ∗ = Fp E (T )/ E (T ) , ∗ and G G FpV(T) Fp V(T) = Fp V(T)/ V(T) , ∗ (see Lemma 2.4.6(c)). Since the only proper subgroup of G is the trivial group, it follows that the actions of G on the spaces
∗ ∗ E (T )/ E (T ) G, ∗ and V(T)/ V(T) G, ∗ are free, where ∗ is the image of (E∗(T ))G and of (V (T ))G, respectively. Hence the ∗ ∗ G G modules [[ Fp(E (T )/(E (T )) , ∗)]] and [[ Fp(V (T )/(V (T )) , ∗)]] are free, and therefore projective, [[ FpG]] -modules (cf. RZ, Propositions 5.7.1 and 5.4.2). Hence the natural [[ FpG]] -epimorphisms
∗ ∗ ∗ G Fp E (T ), ∗ −→ Fp E (T )/ E (T ) , ∗ and
G FpV(T) −→ Fp V(T)/ V(T) , ∗ split. So we can write the complex C(T,Fp) associated with T and Fp (see Sect. 2.3) in the following way:
∗ G d G ε 0 −→ Fp E (T ) , ∗ ⊕ M −→ Fp V(T) ⊕ L −→ Fp −→ 0, ∼ ∗ ∗ G ∼ G where M = [[ Fp(E (T )/(E (T )) , ∗)]] and L = [[ Fp(V (T )/(V (T )) , ∗)]] . 114 4 Profinite Groups Acting on C-Trees
G We view C(T , Fp) as a subcomplex of the complex C(T,Fp) described above. G Since T is a p-tree, the sequence C(T,Fp) is exact. We note that T is connected if and only if
∗ G ∗ = G d Fp E (T ) , Ker(ε|[[Fp(V (T )) ]] ). To prove this equality observe first that clearly
∗ G ∗ ≤ G d Fp E (T ) , Ker(ε|[[Fp(V (T )) ]] ).
∈ G ∈ Next take b Ker(ε|[[Fp(V (T )) ]] ). Since C(T,Fp) is exact, there exists an a ∗ ∗ G [[ Fp(E (T ), ∗)]] with d(a) = b.Leta = a0 + a1 for some a0 ∈[[Fp((E (T )) , ∗)]] and a1 ∈ M. Then d(a1) = b − d(a0) is an element of d(M) whichisfixedbyG, since d(a0) and b arefixedbyG.Letg be a generator of G and define endomor- phisms ν and τ of [[ Zp(V (T ))]] by
p−1 ν(m) = 1 + g +···+g m and τ(m)= (g − 1)m m ∈ Zp V(T) . Then, according to Lemma 4.1.4, the image of ν coincides with the kernel of τ on any free [ZpG]-submodule of [[ Fp(V (T ))]] . Since d(M) is a free [[ FpG]] -module (because d in an injection), and since d(a1) ∈ Ker(τ) (because d(a1) is fixed by G), one has that d(a1) = ν(d(a2)),forsomea2 ∈ M. Write
G d(a2) = b0 + b1, for some b0 ∈ Fp V(T) and b1 ∈ L.
Then d(a1) = ν(b0) + ν(b1). Now, the element
ν(b1) = d(a1) − ν(b0) = b − d(a0) − ν(b0) G G lies in L ∩[[Fp(V (T ) )]] , since ν(b1) ∈ L and b,d(a0), ν(b0) ∈[[Fp(V (T ) )]] . Therefore, ν(b1) = 0. Hence
d(a1) = ν(b0) = pb0 = 0. ∗ G It follows that b = d(a0), and thus, b ∈ d([[ Fp((E (T )) , ∗)]] ).
Step 2. Let G =∼ Z/nZ, where n is a natural number which is divisible only by primes in π. Assume that T G is not connected. Consider a minimal such n with T G not con- nected. By Step 1, n is not prime. Pick a proper nontrivial subgroup H of G. Since |H| Step 3. Assume now that G is procyclic. For a natural number m, put Gm ={gm | g ∈ G}; then G/Gm is a finite cyclic π- group. Since T = T G, there exists some vertex of T with a nontrivial stabilizer, say n n n G 0 , where n0 is a natural number. For n0|n,wehaveG ≤ G 0 . Then = n G lim←− G/G . n0|n 4.1 Fixed Points 115 n By Proposition 4.1.1, the quotient graph Tn = G \T is a π-tree for any n which is a multiple of n0. Clearly = T lim←− Tn. n0|n n n G/G Furthermore, G/G acts on Tn in a natural way. By Step 2, Tn is a π-tree. Therefore, by Proposition 2.4.3(d), G = G/Gn T lim←− Tn n0|n is also a π-tree. Step 4. Let G be an arbitrary pro-π group. Obviously, T G = T g. g∈G By Step 3, T g is a π-subtree of T for every g ∈ G. Hence, by Proposition 2.4.9, T G is a π-tree. Corollary 4.1.6 Suppose that a pro-π group G acts on a π-tree T , and let v and w be two different vertices of T . Then the set of edges E([v,w]) of the chain [v,w] is nonempty, and Gv ∩ Gw ≤ Ge for every e ∈ E([v,w]). Proof Clearly E([v,w]) = ∅, because otherwise [v,w]=V([v,w]), which is not K connected (see Lemma 2.1.9). Let K = Gv ∩ Gw. Since v,w ∈ T , we have that T K is not empty, and hence, by Theorem 4.1.5, T K is a π-tree. So [v,w]⊆T K (see Lemma 2.4.10). Therefore K ≤ Ge, for every edge e ∈[v,w]. Corollary 4.1.7 Suppose that a pro-π group G acts on a π-tree T . If T contains two G-invariant π-subtrees T1 and T2 which are disjoint, then G (a) there exists an edge e ∈ T such that e/∈ T1 and e/∈ T2; G = ∅ = G (b) T1 T2 ; (c) if G acts freely on a subtree T1 of T , then it acts freely on T . Proof (a) Consider the profinite graph T obtained from T by collapsing T1 to a point v1 and T2 to a different point v2. By Lemma 2.4.7, T is a π-tree. Moreover, G G acts on T fixing v1 and v2. By Theorem 4.1.5, T is a π-tree; and by Corol- G G lary 4.1.6, the chain [v1,v2] in T contains an edge e. Therefore e ∈ T with e/∈ T1 and e/∈ T2 (by abuse of notation, we denote the edge e and its unique lifting in T by the same letter). In particular, T G is not empty, and so it is a π-tree. G (b) It follows that T1 ∩ T = ∅ (for otherwise, according to (a), there would exist some fixed edge of T not in T G, which is obviously not the case), and in particular G = ∅ G = ∅ T1 . Similarly T2 . 116 4 Profinite Groups Acting on C-Trees (c) Say G acts freely on T1.Letv be a vertex of T such that its stabilizer Gv is ={ } Gv = ∅ not trivial. Put T2 v . Then T1 and T2 are disjoint. So, according to (b), T1 , a contradiction. Hence G acts freely on T , as asserted. Next we prove that a finite π-group G (i.e., if a prime number p divides the order of G, then p ∈ π) acting on a π-tree must fix a vertex. Theorem 4.1.8 Suppose that a finite π-group G acts on a π-tree T . Then G = Gv, for some vertex v ∈ V(T). Proof We must prove that T G ={t ∈ T | gt = t,∀g ∈ G}=∅. We use induction on the order of G. Assume that |G| > 1 and that the result holds whenever a finite π-group L acts on a π-tree and |L| < |G|. Observe first that if G has prime order and we had T G =∅, then the action of G on T would be free; hence Theorem 4.1.2 would imply that G is projective, contradicting the fact that G is finite and nontrivial; so if G has prime order, the result follows. If G is not simple and N is a proper nontrivial normal subgroup of G, the induc- tion hypothesis implies that T N = ∅. Hence by Theorem 4.1.5, T N is a π-subtree of T . Furthermore, G/N acts naturally on T N ; hence by induction again, one has that T G = (T N )G/N = ∅. Therefore if G is not simple, the result follows. Let us consider the remaining case: suppose that G is a nonabelian simple group. G Assume that T =∅. Choose a minimal G-invariant π-subtree T0 of T (see Propo- sition 2.4.12(a)). Fix a maximal proper subgroup H of G. Then the set − J = x ∈ G − H xHx 1 ∩ H = 1 is not empty, since otherwise G would be a Frobenius group and hence not simple H = ∅ (cf. Huppert 1967, Chap. V, Theorem 7.6). By the induction hypothesis, T0 ;let ∈ H = = v V(T0 ). We note that H Gv,theG-stabilizer of v, that is, xv v, for every x ∈ G − H , because if xv = v for some x ∈ G − H , we would have that G = H,x fixes v, contradicting the assumption that T G =∅. Choose g ∈ J such that the chain P =[v,gv] is minimal in the set of chains {[v,xv]|x ∈ J }, ordered by inclusion. Observe that since v,gv ∈ T0, we have that P =[v,gv]⊆T0. Claim A For every h ∈ H , Wh ={w ∈ V(P)| ghw = w}=∅and v,gv∈ / Wh. The second statement is clear. To prove the first one consider the following sub- graph of T0 n−1 D = (gh)iP, i=0 where n is the order of gh. Since h fixes v, one has that + (gh)igv ∈ (gh)iP ∩ (gh)i 1P, for i = 0,...,n− 1; 4.1 Fixed Points 117 hence D is connected (see Lemma 2.1.7(b)), and therefore a π-subtree of T0 (see Proposition 2.4.3(b)). Since gh is a proper subgroup of G and acts on D, there exists a vertex w of D fixed by gh. Note that w ∈ P ; indeed, say w ∈ (gh)iP ; then w = (gh)−iw ∈ P .ThisprovesClaimA. Claim B For h ∈ H and every w ∈ Wh one has [v,w]=P = [w,gv]. We observe first that the inequalities [v,w]=P = [w,gv] are equivalent if w ∈ Wh. Indeed, let us suppose that [v,w]=P ; then ghP = gh[v,w]=[gv,w]⊆P ; therefore, P = (gh)nP ⊆ (gh)n−1P ⊆···⊆ghP =[gv,w]⊆P ; thus [gv,w]=P . Similarly, if [w,gv]=P , one sees that [v,w]=P . ˜ ∈ ∈ Next we prove that there exists at least one h H and some wh˜ Wh˜ such that [ ]= ∈ [ ]= v,wh˜ P . Assume on the contrary that for every h H , one has v,w P (and hence [w,gv]=P ), for all w ∈ Wh. Then ghP = P , for all h ∈ H . Since G is gen- erated by the set {gh | h ∈ H}, it follows that P is G-invariant. Set K = gHg−1 ∩H ; then, by the choice of g and v, K = 1 and v,gv ∈ T K . By Theorem 4.1.5, T K is a π-tree; therefore P ⊆ T K , i.e., K acts trivially on P . Since G is simple nonabelian, there is a maximal proper subgroup M of G with K ≤ M. Then G = K,M,so that P G = P M . By the induction hypothesis, P M = ∅, and this contradicts the as- sumption that T G =∅. ˜ ∈ ∈ [ ]= Therefore, there exist some h H and a vertex wh˜ Wh˜ such that v,wh˜ P [ ]= ∈ (and hence wh˜ ,gv P ). To complete the proof of Claim B,leth H , and let ∈ ∈[ ] ∈[ ] ∈[ ] [ ]⊆ w Wh. Clearly either w v,wh˜ or w wh˜ ,gv ;sayw v,wh˜ . Then v,w [ ] [ ]= v,wh˜ . Thus v,w P .ThisprovesClaimB. For each h ∈ H , choose a vertex v ∈ W . Set h h D1 = [v,vh],D2 = [vh,gv] and D0 = D1 ∩ D2. h∈H h∈H Note that D1,D2,D0 ⊆ P ; furthermore, v/∈ D2 and gv∈ / D1 by Claim B.The profinite subgraphs D1 and D2 are nonempty and connected (see Lemma 2.1.7); therefore, by Proposition 2.4.3(b), they are π-subtrees of T0. Since D0 contains all the vertices vh (h ∈ H), it is not empty. It follows from Proposition 2.4.9 that it is a ⊆ π-subtree of T0. Since T0 is G-invariant, one has that x∈G xD0 T0. = Claim C T0 x∈G xD0. Since T0 is a minimal G-invariant subtree of T , we just need to show that x∈G xD0 is a G-invariant π-subtree of T0. Since G is finite, x∈G xD0 is closed in T0, and obviously it is G-invariant. Since G = gh| h ∈ H and since gh fixes ∈ avertexinD0 for every h H , we deduce that x∈G xD0 is connected (see Lemma 2.2.4). Then, by Proposition 2.4.3(b), x∈G xD0 is a π-subtree. This proves Claim C. 118 4 Profinite Groups Acting on C-Trees Since v ∈ T0, it follows from Claim C that there exists a g0 ∈ G such that g0v ∈ D0 = D1 ∩ D2. Next we show that g0 ∈ J . Since v/∈ D2, we deduce that g0v = v, −1 hence g0 ∈/ H . As remarked in the proof of Claim B, the subgroup K = gHg ∩ H −1 acts trivially on P . Therefore, K is contained in the stabilizer g0Hg0 of the vertex g0v ∈ D0 ⊂ P . Since = ⊆ −1 ∩ 1 K g0Hg0 H, we have that g0 ∈ J . To finish the proof we first note that the chain [v,g0v] is contained in D1 because v,g0v ∈ D1. Since gv∈ / D1, we have that [v,g0v] is a proper subset of P =[v,gv]. This contradicts the minimality of [v,gv]. Therefore T G = ∅, as desired. Corollary 4.1.9 Let G be a pro-π group acting on a π-tree T and let D be a G- invariant π-subtree of T . Then for every vertex v in T , there exists a vertex w ∈ D such that Gv ≤ Gw. In particular, if the action of G on D is free, then so is the action of G on T (see also Corollary 4.1.7(c)). Proof Write = Gv lim←− Gv/N, N∈N where N is the collection of all open normal subgroups of Gv. For each N ∈ N , set TN = N\T . Consider the natural actions of Gv and Gv/N on TN . Since N is generated by the N-stabilizers of the vertices of T , it follows from Proposition 4.1.1 that each TN is a π-tree. Let DN be the image of D in TN under the projection T −→ TN . Since D is connected, so is DN ; hence DN is a π-tree on which the finite π-group Gv/N acts. G G /N G Since (DN ) v = (DN ) v , we have that (DN ) v = ∅, by Theorem 4.1.8. Observe that the collection {DN | N ∈ N } is naturally an inverse system, and = D lim←− DN . N∈N G We easily deduce that the corresponding collection {(DN ) v | N ∈ N } of sets of fixed points is an inverse system, and Gv = Gv D lim←− (DN ) . N∈N G G Therefore D v = ∅ (see Sect. 1.1). For every vertex w ∈ D v we have Gv ≤ Gw. Using this corollary we can now complete the proof of Proposition 2.4.12 on the uniqueness of minimal G-invariant subtrees of a G-tree, in certain cases. Proof of Proposition 2.4.12(b) Assume that |D| > 1, and let D0 be another minimal G-invariant π-subtree of T . Denote by T the profinite graph obtained from T by collapsing D0 to a point (see Example 2.1.2). Since D0 is G-invariant, the action of G carries over naturally to T . By Lemma 2.4.7, T is a π-tree. Since D and D0 4.2 Faithful and Irreducible Actions 119 are minimal G-invariant, D ∩ D0 =∅. This means that the π-tree D is isomorphic to its image D in T and, in particular, |D | > 1. Observe that D is a minimal G-invariant π-subtree of T .Letv be the vertex of T to which D0 collapses. Then G = Gv. By Corollary 4.1.9, there is a vertex w in V(D) such that Gv ≤ Gw; hence Gw = G, and so G fixes the vertex w. But since D is minimal G-invariant, we have D ={w}, a contradiction. 4.2 Faithful and Irreducible Actions We say that an action of a pro-π group G on a π-tree T is irreducible (or that G acts irreducibly on T )ifT is a minimal G-invariant π-subtree of T . We say that the action is faithful (or that G acts faithfully on T ) if the kernel CG(T ) of the action G −→ Aut(T ) is trivial, i.e., if whenever g ∈ G and gm = m for all m ∈ T , then g = 1. Remark 4.2.1 (a) If G acts faithfully and irreducibly on T , then either G is trivial (in which case |T |=1) or |V(T)| > 1. Consequently (see Theorem 4.1.8), if in addition G is finite, it must be trivial. (b) Let H be a closed subgroup of a pro-π group G, and assume that G acts on a π-tree T . Observe that if the action of G on T is faithful, so is the induced action of H on T ; and if the induced action of H on T is irreducible, so is the action of G on T . (c) If a pro-π group G acts freely on a π-tree T , then this action is faithful; more- over it is obvious that such a group acts faithfully and irreducibly on any mini- mal G-invariant π-subtree of T . An often essential part of the information carried by groups acting on profinite trees can be obtained through faithful and irreducible actions. Indeed, if a profinite group G acts on a profinite tree T , one can consider the action of G on a minimal G-invariant subtree D and factor out the kernel of this action CG(D). The resulting action of G/CG(D) on D is faithful and irreducible. In this section we study faithful and irreducible actions; this will lead us eventu- ally to a classification of pro-π groups acting on π-trees (Theorem 4.2.11). Proposition 4.2.2 Let G beapro-π group acting irreducibly on a π-tree T , and let N be a closed normal subgroup of G. (a) If N is contained in the stabilizer of some vertex, then N acts trivially on T . (b) If N is a finite normal subgroup of G, then N acts trivially on T . Proof Part (b) follows from part (a) and Theorem 4.1.8. To prove part (a) note first that since T N is not empty, it is a G-invariant π-subtree of T (see Theo- rem 4.1.5(b)). The irreducibility of the action of G implies that T N = T . 120 4 Profinite Groups Acting on C-Trees Recall that a closed subgroup N of a profinite group G is called subnormal if there exists a finite chain of closed subgroups of G, N = N0 N1 ···Nk = G, each normal in the next one. We refer to this series as a normal series joining N to G. Proposition 4.2.3 Let G be a pro-π group acting faithfully and irreducibly on a π-tree T . Then the following statements hold. (a) Every closed subgroup of G that contains a nontrivial closed subnormal sub- group of G acts irreducibly on T . (b) If G is abelian, then it acts freely on T . Consequently, either G is trivial or ∼ G = Zρˆ , for some ρ ⊆ π. Proof If |T |=1, then |G|=1 and the results are obvious. Assume |T | > 1. (a) It suffices to prove that every nontrivial closed subnormal subgroup of G acts irreducibly on T (see Remark 4.2.1(b)). For a closed subnormal subgroup N of G, let k = kN denote the smallest natural number such that N = N0 N1 ···Nk = G is a normal series joining N to G. We use induction on k. Let N be a nontrivial closed subnormal subgroup of G and assume that kN = 1, i.e., N is normal in G. Choose a minimal N-invariant π-subtree D of T (see Propo- sition 2.4.12(a)). We claim that |D| > 1. Indeed, if D ={v} for some v ∈ V(T), then N ≤ Gv. By Proposition 4.2.2, N must act trivially on the whole of T , contradict- ing the faithfulness of the action of G on T . Thus |D| > 1. Since N is normal in G, the shift gD of D by any element g of G is also a minimal N-invariant π-subtree of T . By Proposition 2.4.12(b), such a subtree is unique, so that gD = D for all g ∈ G. Since G acts irreducibly on T , one concludes that D = T , proving that N acts irreducibly on T . Let k = kN > 1 and assume that the result holds whenever M is a closed subnor- mal subgroup of G with kM