Visit https://ebookgate.

com to download the full version and

explore more ebooks

Handbook of Computational Gr Theory 1st Edition

Derek F. Holt

_____ Click the link below to download _____

https://ebookgate.com/product/handbook-of-
computational-gr-theory-1st-edition-derek-f-holt/

Explore and download more ebooks at ebookgate.com

Here are some recommended products that might interest you.
You can download now and explore!

Handbook of Equine Wound Management 1st Edition Derek

Knottenbelt

https://ebookgate.com/product/handbook-of-equine-wound-management-1st-
edition-derek-knottenbelt/

ebookgate.com

Traditions Warriner s Handbook 1 First Course Teachers

Edition Grade 7 Holt [Holt]

https://ebookgate.com/product/traditions-warriner-s-handbook-1-first-
course-teachers-edition-grade-7-holt-holt/

ebookgate.com

Handbook of Self Regulation Research Theory and

Applications 1st Edition Roy F. Baumeister

https://ebookgate.com/product/handbook-of-self-regulation-research-
theory-and-applications-1st-edition-roy-f-baumeister/

ebookgate.com

Handbook of Self Regulation Second Edition Research Theory

and Applications Roy F. Baumeister

https://ebookgate.com/product/handbook-of-self-regulation-second-
edition-research-theory-and-applications-roy-f-baumeister/

ebookgate.com
Handbook of Computational Molecular Biology 1st Edition
Srinivas Aluru

https://ebookgate.com/product/handbook-of-computational-molecular-
biology-1st-edition-srinivas-aluru/

ebookgate.com

Compact handbook of computational biology 1st Edition

Andrzej K. Konopka

https://ebookgate.com/product/compact-handbook-of-computational-
biology-1st-edition-andrzej-k-konopka/

ebookgate.com

Handbook of Discrete and Computational Geometry 2nd

Edition Rajeev Bansal

https://ebookgate.com/product/handbook-of-discrete-and-computational-
geometry-2nd-edition-rajeev-bansal/

ebookgate.com

A Course in Computational Number Theory 1st Edition David

Bressoud

https://ebookgate.com/product/a-course-in-computational-number-
theory-1st-edition-david-bressoud/

ebookgate.com

The Evolution of Blast Beats by Derek Roddy Derek Roddy

https://ebookgate.com/product/the-evolution-of-blast-beats-by-derek-
roddy-derek-roddy/

ebookgate.com
 DISCRETE MATHEMATICS AND ITS APPLICATIONS
Continued Titles Series Editor KENNETH H. ROSEN
Richard A. Mollin, RSA and Public-Key Cryptography
Kenneth H. Rosen, Handbook of Discrete andCombinatorial Mathematics
Douglas R. Shier and K. t Wal/enius, Applied Mathematical Modeling: A Multidisciplinary
Approach
Douglas R. Stinson, Cryptography: Theory andPractice, Second Edition
Roberto Togneri andChristopher J. deSilva, Fundamentals of Information Theory and
Coding Design
HANDBOOK OF
. Lawrence C. Washington, Elliptic Curves: Number Theory and Cryptography
Kun-Mao Chao andBang Ye Wu, Spanning Trees and Optimization Problems
Juergen Bierbrauer, Introduction to Coding Theory
William Kocay andDonald L. Kreher, Graphs, Algorithms, and Optimization
COMPUTATIONAL
Derek F. Holt with Bettina Eick and Eamonn A. O'Brien, Handbook ofComputational Group Theory
GROUP THEORY

DEREKF: HOLT
BcJ;TrINA EICK
EAMONN A. O'BRIEN

t34
CHAPMAN &HALUCRC
A CRC Press Company
Boca Raton London New York Washington, D.C.
 Preface
This book is about computational group theory, which we shall frequently
abbreviate to CGT. The origins of this lively and active branch of mathematics
can be traced back to the late nineteenth and early twentieth centuries, but
it has been flourishing particularly during the past 30 to 40 years. The' aim
of this book is to provide as complete a treatment as possible of all of the
fundamental methods and algorithms in CGT, without straying above l\ level
suitable for a beginning postgraduate student. .!

There are currently three specialized books devoted to specific areas of

CGT, namely those of Greg Butler [But91] and Akos Seress [8er03] on al-
gorithms for permutation groups, and the book by Charles 8ims [~i~94] on
computation with finitely presented groups. These texts cover their respec-
tive topics in greater detail than we shall do here, although we have relied
heavily on some of the pseudocode presented in [8im94) in our treatment of
Library of Congress Cataloging-in-Publication Data coset enumeration in Chapter 5,
The most basic algorithms in CGT tend to be representation specific; that
Catalog record is available from the Library of Congress
is, there are separate methods for groups given as permutation or matrix
groups, groups defined by means of polycyclic presentations, and groups that
This book contains information obtained from authentic and highly regarded sources. Reprinted material are defined using a general finite presentation. We have devoted separate
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
chapters to algorithms that apply to groups in these different types of repre-
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use. sentations, but there are other chapters that cover important methods involv-
ing more than one type. For example, Chapter 6 is about finding presentations
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microfilming, and recording, or by any information storage or
of permutation groups and the connections between coset enumeration and
retrieval system, without prior permission in writing from the publisher. methods for finding the order of a finite permutation group.
We have also included a chapter (Chapter 11) on the increasing number of
The consent of CRC Press does not extend to copying for general distribution, for promotion, for creating
new works, or for resale. Specific permission must be obtained in writing from CRC Press for such precomputed stored libraries and databases of groups, character tables, etc.
copying. that are now publicly available. They have been playing a major r6le in CGT
in recent years, both as an invaluable resource for the general mathematical
Direct all inquiries to CRC Press, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
public, and as components for use in some advanced algorithms in CGT. The
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are library of all finite groups of order up to 2000 (except for order 1024) has
used only for identification and explanation, without intent to infringe.
proved to be particularly popular with the wider community.
Visit the CRC Press Web site at www.crcpress.com It is inevitable that our choice of topics and treatment of the individual
topics will reflect the authors' personal expertise and preferences to some
© 2005 by Chapman & Hall/CRC Press extent. On the positive side, the final two chapters of the book cover appli-
cations of string-rewriting techniques to CGT (which is, however, treated in
No claim to original U.S. Government works
International Standard Book Number 1-58488-372-3
much greater detail in [8im94)), and the application of finite state automata
Printed in the United States of America I 2 3 4 5 6 7 8 9 0 to the computation of automatic structures of finitely presented groups. On
Printed on acid-free paper the other hand, there may be some topics for which our treatment is more
superficial than it would ideally be.
One such area is the complexity analysis of the algorithms of CGT. During
the 1980s and 1990s some, for the most part friendly and respectful, rivalry
developed between those whose research in CGT was principally directed to-
wards producing better performance of their code, and those who were more
 vi vii

interested in proving theoretical results concerning the complexity of the al- At the end of each chapter, or sometimes section, we have attempted to
gorithms. .This study of complexity began with the work of Eugene Luks, direct the reader's attention to some applications of the techniques developed
who established a connection in his 1982 article [Luk82] between permutation iIIthat chapter either to other areas of mathematics or to other sciences, It is
group algorithms and the problem of testing two finite graphs for isomorphism. generally difficult to do this effectively. Although there are many important
Our emphasis in this book will be more geared towards algorithms that per- and interesting applications of CGT around, the most significant of them will
form well in practice, rather than those with the best theoretical complexity. typically use methods of CGT as only one of many components, and so it not
Fortunately, Seress' book [Ser03] includes a very thorough treatment of com- possible to do them full justice without venturing a long way outside of the
plexity issues, and so we can safely refer the interested reader there. In any main topic of the book.
c~se, as machines become faster, computer memories larger, and bigger and We shall assume that the reader is familiar with group theory up to an
bigger groups come within the range of practical computation, it is becom- advanced undergraduate level, and has a basic knowledge of other topics in
ing more and more the case that those algorithms with the more favourable algebra; such as ring and field theory. Chapter 2 includes a more or less
complexity will also run faster when implemented. complete survey of the required background material in group theory, but we
The important topic of computational group representation theory and shall assume that at least most of the topics reviewed will be already familiar
computations with group characters is perhaps not treated as thoroughly as to readers. Chapter 7 assumes some basic knowledge of group representation
it might be in this book. We have covered some of the basic material in theory, such as the equivalence between matrix representations of a group G
Cha~ter 7, but there is unfortunately no specialized book on this topic. For over a field K and KG-modules, but it is interesting to note that many of
a bnef survey of the area, we can refer the reader to the article by Gerhard the most fundamental algorithms in the area, such as the 'Meataxe', use only
Hiss [His03]. rather basic linear algebra.
One of the most active areas of research in CGT at the present time both A number of people have made helpful and detailed comments on draft
from the view~oint of complexity and of practical performance, is the develop- versions of the book, and I would particularly like to thank John Cannon,
ment of effective methods for computing with large finite groups of matrices. Bettina Eick, Joachim Neubiiser, and Colva Roney-Dougal in this regard.
Much o! this material is beyond the scope of this book. It is, in any case, Most of all, I am grateful to my two coauthors, Bettina Eick and Eamonn
?ev~lopmg and changing too rapidly to make it sensible to attempt to cover O'Brien for helping me to write parts of the book. The whole of Chapter 8,
It properly here. Some pointers to the literature will of course be provided on computing in polycyclic groups, was written by Bettina Eick, as were
mainly in Section 7.8. ' Sections 11.1 and 11.4 on the libraries of primitive groups and of small groups,
Yet another topic that is beyond the scope of this book but which is of respectively. Section 7.8 on computing in matrix groups and Section 9.4 on
. .. '
increasing Importance in CGT, is computational Lie theory. This includes computing p-quotients of finitely presented groups were written by Eamonn
comput~tions with C.oxeter groups, reflection groups, and groups of Lie type O'Brien.
an~ th:Ir representat~ons. It also connects with computations in Lie algebras,
which IS an area of independent importance. The article by Cohen, Mur- August 2004 Derek Holt
ray, and Taylor [CMT04] provides a possible starting point for the interested
reader.
The author firmly believes that the correct way to present a mathematical
algorith~ ~s by means.of pseudocode, since a textual description will generally
lac~ pre~lSl?n, and wil~, usually involve rather vague instructions like "carry
on m a similar manner . So we have included pseudocode for all of the most
basic algorithms, and it is only for the more advanced procedures that we have
occasionally lapsed into sketchy summaries. We are very grateful to Thomas
Cor~en wh~ has mad: his ]g,.'IEXpackage 'clrscode' for displaying algorithms
publicly available. ThIS was used by him and his coauthors in the well-known
textbook on algorithms [CLRS02].
Although working through all but the most trivial examples with proce-
dures that are intended to be run on a computer can be very tedious we
have attempted to include illustrative examples for as many algorithms as is
practical.
Contents

Notation and displayed procedures xvi

1 A Historical Review of Computational Group Theory 1

2 Background Material 9
2.1 Fundamentals.. 9
2.1.1 Definitions. 9
2.1.2 Subgroups. 11
2.1.3 Cyclic and dihedral groups 12
2.1.4 Generators 13
2.1.5 Examples - permutation groups and matrix groups 13
2.1.6 Normal subgroups and quotient groups. . . . . . 14
2.1.7 Homomorphisms and the isomorphism theorems 15
2.2 Group actions . . . . . . . . . . 17
2.2.1 Definition and examples . . . . . . . . . 17
2.2.2 Orbits and stabilizers 19
2.2.3 Conjugacy, normalizers, and centralizers 20
2.2.4 Sylow's theorems . . . . . . 21
2.2.5 Transitivity and primitivity . . . . . . . 22
2.3 S e r i e s . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.1 Simple and characteristically simple groups 26
2.3.2 Series 27
2.3.3 The derived series and solvable groups 27
2.3.4 Central series and nilpotent groups 29
2.3.5 The socle of a finite group . . . . 31
2.3.6 The Frattini subgroup of a group 32
2.4 Presentations of groups . . . 33
2.4.1 Free groups . . . . . . . . . . . . 33
2.4.2 Group presentations . . . . . . . 36
2.4.3 Presentations of group extensions . 38
2.4.4 Tietze transformations . . . . . . . 40
2.5 Presentations of subgroups . . . . . . . . . 41
2.5.1 Subgroup presentations on Schreier generators 41
2.5.2 Subgroup presentations on a general generating set 44
2.6 Abelian group presentations . . . . . . . . . . . . . . . . . 46
x xi

2.7 Representation theory, modules, extensions, derivations, and 4.4.5 The random Schreier-Sims method 97
complements . . . . . . . . . . . . . . . . . . . . . . . . . . " 48 4.4.6 The solvable BSGS algorithm . . . 98
2.7.1 The terminology ofrepresentation theory 49 4.4.7 Change of base . . . . . . . . . . . 102
2.7.2 Semidirect products, complements, derivations, and 4.5 Homomorphisms from permutation groups 105
first cohomology groups . . . . . . . . . . . . . . . .. 50 4.5.1 The induced action on a union of orbits 105
2.7.3 Extensions of modules and the second cohomology 4.5.2 The induced action on a block system . 106
group 52 4.5.3 Homomorphisms between permutation groups . 107
2.7.4 The actions of automorphisms on cohomology groups. 54 4.6 Backtrack searches 108
2.8 Field theory . . . . . . . . . . . . . . . . . . 56 4.6.1 Searching through the elements of a group. . . 110
2.8.1 Field extensions and splitting fields . 56 4.6.2 Pruning the tree . . . . . . . . . . . . . . . . . 113
2.8.2 Finite fields . . . . . 58 4.6.3 Searching for subgroups and coset representatives. 114
2.8.3 Conway polynomials . . . . . 59 4.6.4 Automorphism groups of combinatorial structures and
partitions . . . . . . . . . . . . 118
3 Representing Groups on a Computer 61 4.6.5 Normalizers and centralizers. . . . . 121
3.1 Representing groups on computers. . 61 4.6.6 Intersections of subgroups . . . . . 124
3.1.1 The fundamental representation types 61 4.6.7 Transversals and actions on cosets . 126
3.1.2 Computational situations 62 4.6.8 Finding double coset representatives 131
3.1.3 Straight-line programs . . . . 64 4.7 Sylow subgroups, p-cores, and the solvable radical 132
3.1.4 Black-box groups . . . . . . . 65 4.7.1 Reductions involving intransitivity and imprimitivity . 133
3.2 The use of random methods in CGT . 67 4.7.2 Computing Sylow subgroups. . . . . . . . . . . . . 134
3.2.1 Randomized algorithms '" 67 4.7.3 A result on quotient groups of permutation groups . 137
3.2.2 Finding random elements of groups . 69 4.7.4 Computing the p-core .. . . . . . . . . 138
3.3 Some structural calculations . . . . . . 72 4.7.5 Computing the solvable radical . . . . . 140
3.3.1 Powers and orders of elements. . . . 72 4.7.6 Nonabelian regular normal subgroups 141
3.3.2 Normal closure . . . . . . . . . . . . 73 4.8 Applications................... 143
3.3.3 The commutator subgroup, derived series, and lower 4.8.1 Card shuffling. . . . . . . . . . . . . . 144
central series 73 4.8.2 Graphs, block designs, and error-correcting codes 145
3.4 Computing with homomorphisms . . . . . . . . . . . 74 4.8.3 Diameters of Cayley graphs . . . . . 147
3.4.1 Defining and verifying group homomorphisms 74 4.8.4 Processor interconnection networks . . . . . . .. . 148
3.4.2 Desirable facilities 75
5 Coset Enumeration 149
4 Computation in Finite Permutation Groups 77 5.1 The basic procedure . 150
4.1 The calculation of orbits and stabilizers. 77 5.1.1 Coset tables and their properties 151
4.1.1 Schreier vectors. . . . . 79 5.1.2 Defining and scanning . 152
4.2 Testing for Alt(O) and Sym(O) . 81 5.1.3 Coincidences . . . . . .. 156
4.3 Finding block systems . . . . . . 82 5.2 Strategies for coset enumeration . 162
4.3.1 Introduction....... 82 5.2.1 The relator-based method 162
4.3.2 The Atkinson algorithm 83 5.2.2 The coset table-based method . · 165
4.3.3 Implementation of the class merging process . 85 5.2.3 Compression and standardization . · 167
4.4 Bases and strong generating sets . . . 87 5.2.4 Recent developments and examples . 168
4.4.1 Definitions................ 87 5.2.5 Implementation issues . . . . . . . . · 170
4.4.2 The Schreier-Sims algorithm. . . . . . 90 5.2.6 The use of coset enumeration in practice . · 171
4.4.3 Complexity and implementation issues 93 5.3 Presentations of subgroups . . . . . . . . . . . . . 173
4.4.4 Modifying the strong generating set - shallow Schreier 5.3.1 Computing a presentation on Schreier generators 173
trees. . . . . . . . . . . . . . . . . . . . . . . . . . .. 95 5.3.2 Computing a presentation on the user generators 178
xii xiii

5.3.3 Simplifying presentations . 184 Structural investigation of matrix groups . . . . . . . . . . . . 264

7.8
5.4 Finding all subgroups up to a given index. . . 188 7.8.1 Methods based on bases and strong generating sets .. 264
It"D
5.4.1 Coset tables for a group presentation. 189 7.8.2 Computing in large-degree matrix groups 268
5.4.2 Details of the procedure . . . 190
5.4.3 Variations and improvements 196 8" Computation with Polycyclic Groups 273
5.5 Applications.......... 198 ;::.:8.1 Polycyclic presentations . · 274
~'A 8.1.1 Polycyclic sequences . · 274
6 Presentations of Given Groups 199 8.1.2 Polycyclic presentations and consistency · 278
6.1 Finding a presentation of a given group. . . . . . . . . 199 8.1.3 The collection algorithm . . . .280
6.2 Finding a presentation on a set of strong generators . 205 8.1.4 Changing the presentation . · 284
6.2.1 The known BSGS case. . . . . . . . . . . . . 205
Examples of polycyclic groups . . . . . . . . . . . . · 286
.6.2.2 The 'Iodd-Coxeter-Schreier-Sims algorithm . 207
8.2.1 Abelian, nilpotentcand supersolvable groups · 286
6.3 The Sims 'Verify' algorithm . . . . 208
8.2.2 Infinite polycyclic groups and number fields · 288
6.3.1 The single-generator case . 209
8.2.3 Application - crystallographic groups · 289
6.3.2 The general case . 213
Subgroups and membership testing . . · 290
6.3.3 Examples......... . 217
8.3.1 Induced polycyclic sequences . · 291
7 Representation Theory, Cohomology, and Characters 219 8.3.2 Canonical polycyclic sequences .296
7.1 Computation in finite fields. . . . . . . . . 220 Factor groups and homomorphisms .298
7.2 Elementary computational linear algebra . 221 8.4.1 Factor groups .. .298
7.3 Factorizing polynomials over finite fields . 226 8.4.2 Homomorphisms .299
7.3.1 Reduction to the squarefree case . 228 Subgroup series . . . . . .300
7.3.2 Reduction to constant-degree irreducibles . 229 Orbit-stabilizer methods · 302
7.3.3 The constant-degree case . 229 Complements and extensions . .304
7.4 Testing KG-modules for irreducibility - the Meataxe . . 230 8.7.1 Complements and the first cohomology group .304
7.4.1 The Meataxe algorithm .. . 230 8.7.2 Extensions and the second cohomology group · 307
7.4.2 Proof of correctness . 234 Intersections, centralizers, and normalizers . .311
7.4.3 The Ivanyos-Lux extension . 235 8.8.1 Intersections .. .311
7.4.4 Actions on submodules and quotient modules . 235 8.8.2 Centralizers.............. · 313
7.4.5 Applications.................. . 236 8.8.3 Normalizers......... . . . . . · 314
7.5 Related computations. . . . . . . . . . . . . . . . . . 237 8.8.4 Conjugacy problems and conjugacy classes. · 316
7.5.1 Testing modules for absolute irreducibility. . 237 Automorphism groups. . . . . . . . . . · 317
7.5.2 Finding module homomorphisms . . . . . . . 241 The structure of finite solvable groups . .320
7.5.3 Testing irreducible modules for isomorphism. 244 8.10.1 Sylow and Hall subgroups .320
7.5.4 Application - invariant bilinear forms. . . . . 245 8.10.2 Maximal subgroups. . . . . .. .322
7.5.5 Finding all irreducible representations over a finite
field . . . . . . . . . . . . . . . . . . . . 246 f:Qomputing Quotients of Finitely Presented Groups 325
7.6 Cohomology.................. . . 248 Finite quotients and automorphism groups of finite groups .326
7.6.1 Computing first cohomology groups .. 249 9.1.1 Description of the algorithm . .326
7.6.2 Deciding whether an extension splits .. 253 9.1.2 Performance issues . . . . . . . . . .. · 332
7.6.3 Computing second cohomology groups .. 254 9.1.3 Automorphism groups of finite groups .333
7.7 Computing character tables . . 255 Abelian quotients . . . . . . . . . . . . . . . . .335
7.7.1 The basic method . 256 9.2.1 The linear algebra of a free abelian group · 335
7.7.2 Working modulo a prime. . 257 9.2.2 Elementary row operations · 336
7.7.3 Further improvements . . . 260 9.2.3 The Hermite normal form . . . . . . . . . .337
xiv
xv

9.2.4 Elementary column matrices and the Smith normal Jwriting Systems and the Knuth-Bendix Completion
form . · 341 411
/rbcess
9.3 Practical computation of the HNF and SNF . .347 2..1 Monoid presentations ' . · . 412
9.3.1 Modular techniques . .347 " 12.1.1 Monoids and semigroups . · . 412
9.3.2 The use of norms and row reduction techniques .349 12.1.2 Free monoids and monoid presentations · . 415
9.3.3 Applications........... .352 Rewriting systems . . . . . . . . . . . . . . · .417
9.4 p-quotients of finitely presented groups · 353 Rewriting systems in monoids and groups . · .423
9.4.1 Power-conjugate presentations. · 353 Rewriting systems for polycyclic groups . · .426
9.4.2 The p-quotient algorithm . · 355 Verifying nilpotency . . . . . . . . . . . . . · .429
9.4.3 Other quotient algorithms . .364 Applications . . . . . . . . . . . . . . . . . . . . . . . · .431
9.4.4 Generating descriptions of p-groups . : 364
9.4.5 Testing finite p-groups for isomorphism Finite State Automata and Automatic Groups 433
· 371 ian Finite state automata . . . . . . . . . . . . . . . . 434
9.4.6 Automorphism groups of finite p-groups · 371 ; 13.1.1 Definitions and examples . 434
9.4.7 Applications . · 372 13.1.2 Enumerating and counting the language of a dfa .. 437
13.1.3 The use of fsa in rewriting systems .. 439
10 Advanced Computations in Finite Groups 375 13.1.4 Word-acceptors. . . . . . . . . . . . . . 441
10.1 Some useful subgroups . . . . . . . .376 13.1.5 2-variable fsa . . . . . . . . . . . . . . . 442
10.1.1 Definition of the subgroups . .376 13.1.6 Operations on finite state automata .. 442
10.1.2 Computing the subgroups - initial reductions .377 13.1.6.1 Making an fsa deterministic. . . 443
10.1.3 The O'Nan-Scott theorem . · 378 13.1.6.2 Minimizing an fsa .. . . . . . . 444
10.1.4 Finding the socle factors - the primitive case .379 13.1.6.3 Testing for language equality .. 446
10.2 Computing composition and chief series . . · 381 13.1.6.4 Negation, union, and intersection. . . 447
10.2.1 Refining abelian sections . · 381 13.1.6.5 Concatenation and star .. 447
10.2.2 Identifying the composition factors . · 382 13.1.7 Existential quantification . . 448
10.3 Applications of the solvable radical method. · 383 Automatic groups . . . . . . . . . . . . . . . . . . 451
10.4 Computing the subgroups of a finite group . · 385 13.2.1 Definitions, examples, and background .. 451
10.4.1 Identifying the TF-factor . · 386 13.2.2 Word-differences and word-difference automata . . 453
10.4.2 Lifting subgroups to the next layer . · 387 The algorithm to compute the shortlex automatic structures . 456
10.5 Application - enumerating finite unlabelled structures. · 390 13.3.1 Step 1 . . . . . . . . . . . . . . 457
13.3.2 Step 2 and word reduction . . . 459
11 Libraries and Databases 393 13.3.3 Step 3 . 460
11.1 Primitive permutation groups . .394 13.3.4 Step 4 . . . . . . . . . . . . . 462
11.1.1 Affine primitive permutation groups .. 13.3.5 Step 5 . . . . . . . . . . . . . 464
· 395
11.1.2 Nonaffine primitive permutation groups .396 13.3.6 Comments on the implementation and examples . 466
11.2 Transitive permutation groups . Related algorithms . . 468
.397
11.2.1 Summary of the method Applications .. 469
.397
11.2.2 Applications . .399 rences 471
11.3 Perfect groups . .400 dex of Displayed Procedures · .497
11.4 The small groups library . 402 thor Index . · .499
11.4.1' The Frattini extension method . 404 bject Index . .. 503
11.4.2 A random isomorphism test · 405
11.5 Crystallographic groups . . . . . .407
11.6 The "ATLAS of Finite Groups" . . .409
xvi

Notation and displayed procedures

As is usual, {a,b,c} will denote the (unordered) set with elementsa,b,c. We
apter 1
shall generally use square brackets, as in [a, b, c, c, a], to denote an ordered set
or list, which may of course contain repeated elements. There is an unfortu-
nate and occasionally serious conflict here with the notation for commutators istorical Review of
of subgroups and elements of a group. If there is a danger of ambiguity in
this respect, then we shall write comm(g, h) rather than [g, h] for the commu-
'putational Group Theory
tator g-lh-1gh. Here is some notation used in the book, which might not be
completely standard.
N the set of natural numbers 1,2,3, excluding 0
No the set of natural numbers 0,1,2,3, including 0
'·;jthe book with a brief review of the history of CGT. Some of the
JFq the finite field of order q
[3.. 8] the ordered list [3,4, 5, 6, 7,8]
'til, ..this chapter has been taken from Section 2 of the article by
'@·[Neu95].
[11 .. -1 by -3] the ordered list [11,8,5,2,-1]
{3 .. 8} the unordered set {3,4,5,6, 7,8} eory has always been a computational subject. For example, there
[1,3,2] cat [5,1] or [1,3,2] U [5,1] ted attempts during the nineteenth century, with varying degrees
the concatenation [1,3,2,5,1] of the two ordered lists y, to compile complete lists of the primitive and transitive permuta-
's. of low degree. This work began with Ruffini in 1799and continued
t 1912. It was of course all done painstakingly using hand calcula-
The 1l\1EXpackage 'clrscode' has been used for typesetting the displayed her than provide citations here, we refer the reader to the extensive
procedures in this book. This is the package used in the book [CLRS02], and Ii;' compiled by Short in [Sho92, Appendix A] for details. For what-
was downloaded from Thomas Cormen's Web site [Cor]. 11, very little further work was done on this particular problem until
We shall assume that the meaning of the constructs used in the procedures, 0, at which time computers could be used both to check the results
like while, do, if, then, elseif is sufficiently well-known to require no expla- er calculations and, over the following years, to extend the lists to
nation. Unlike in [CLRS02], however, we use a := b (rather than a *- b) to much higher degree. We shall be discussing recent progress in this
denote an assignment. By default, a command break or continue will cause apter 11 of this book.
the procedure to break out of or proceed to the next iteration of the current ,~genuine algorithms proposed for solving problems in group theory
innermost loop. We shall, however, occasionally use "continue a" to mean the days of mechanical computers. As early as 1911, in [Deh11],
proceed to the next iteration of the loop "for a E ... " . Jrnulated the word, conjugacy, and isomorphism problems for finitely
Procedures may return one or more values. So, if a procedure MULTI- ,qgroups, and he devised an algorithm, the Dehn algorithm,for solving
PROC(x) ends with a statement return a, b;, then we access the two returned d problem in certain types of small cancellation groups. Chapters 12
values using a statement like y, z := MULTIPROC(X). If an argument of a pro- will explore recent computational work on some of these problems.
cedure is preceded by a '" symbol, as in ApPEND('" l, x) then, on calling the ., problems of this type are usually undecidable in general: it was
procedure, the value of the argument may be changed within the procedure. y. Novikov [Nov55] and then by Boone in [Boo59] that the word
We shall not display code for a number of basic procedures, of which the ugacy problems are undecidable, and by Adian in [Adi58] and Rabin
meaning is assumed to be obvious. For example, when called with a list l and !~llat the isomorphism problem is undecidable.
an element x, the procedure call ApPEND('--':l,x) simply appends x to the end
1936, Todd and Coxeter described the 'coset enumeration' proce-
of the list l.
ch attempts to calculate the index of a finitely generated subgroup of
presented group, on the assumption that this index is finite. While
ance of Dehn's algorithm was more theoretical than practical, the
eter method was intended to be used for real calculations, which in-
. Coset enumeration remains one of the most fundamental and im-
rocedures in CGT, and we shall be studying it in detail in Chapter 5.
2 Handbook of Computational Group Theory A Historical Review of Computational Group Theory 3

Originally it was used for hand calculations, and there were successful enu- the O'Nan-Scott theorem, which describes the detailed structure of a primitive
merations by hand of the cosets of subgroups of index several hundred. The permutation group, and of the classification of finite simple groups.
current most versatile computer implementation, ACE3 [HR99], by George Other noteworthy developments in the 1970s include the development of (a
Havas and Colin Ramsay, can routinely handle subgroups of index up to at number of variations of) a procedure for computing finite quotients of prime
least 108 . power order of finitely presented groups. The resulting algorithm is now
Yet another example from the premachine era was an algorithm for the known as the p-quotient algorithm and will form the main topic of Section 9.4
classification of space groups proposed in [Zas48] by Zassenhaus. Some time of Chapter 9. In the mid-1970s, versions were developed and implemented by
was to elapse before this could be put to practical use; we shall be discussing Macdonald [Mac74], Wamsley [BKW74], and Havas and Newman [New76].
this application in Section 11.5 of Chapter II. Several new results were discovered using these implementations, including the
Given that group theory had been recognized as an intrinsically compu- orders of the restricted Burnside groups IR(2, 5) I = 534 and IR(4,4) I = 2422
tational branch of mathematics for over a century, it is not surprising that (where R(d, n) denotes the largest finite d-generator group of exponent n).
solving problems in group theory was one of the very earliest nonnumeri- More recently, in 2002, using a special version of the algorithm that uses
cal applications of programmable computers to be proposed. Indeed, some techniques involving Lie algebras, O'Brien and Vaughan-Lee [OVL02] proved
of these early suggestions were surprisingly ambitious, and demonstrated a that IR(2, 7)I = 720416. This computation involved about a year of CPU-time!
striking perceptivity for the eventual capabilities of CGT. In 1951, M.H.A. The p-quotient algorithm outputs finite quotients in the form of a power-
Newman [New51] proposed using computers to investigate groups of order 256 conjugate presentation or polycyclic presentation. Efficient algorithms for
- the complete classification of the 56092 isomorphism types of groups of this computing in finite p-groups, and later in finite solvable and general poly-
order was eventually completed by O'Brien in 1988 [O'B91]. As another ex- cyclic groups, using this type of presentation as the basic data structure for
ample, there is a quote from A. Turing from 1945, which suggests the idea of group elements, were first developed around the same period. These methods
writing a computer program to enumerate the groups of order 720. However, introduced a number of techniques, such as:
the first reported enumeration of the 840 isomorphism types of groups of this • solving a problem from the top down, by solving it successively in larger
order was in 1999, as part of the systematic construction of all groups of order and larger quotients G/ N of the group G;
up to 1000 undertaken by Besche and Eick [BE99b].
The first implementation of a group-theoretical algorithm was probably by • a divide-and conquer approach that uses group homomorphisms to split
B. Hazelgrove, who partially implemented Todd-Coxeter coset enumeration a problem up into smaller instances of the same problem on the kernel
on the EDSAC II in Cambridge in 1953. For details of this and other early and image of the homomorphism;
implementations of coset enumeration, see Leech's article [Lee63]. Possibly
which were later generalized to permutation and matrix groups. For example,
the first genuinely useful programs were those of Felsch and Neubiiser [Neu60,
Neubiiser and Felsch [FN79] described and implemented a top-down approach
Neu61] for computing the subgroup lattices of finite permutation groups, and
to computing conjugacy classes and centralizers in p-groups, which was used
of finite 2-groups given by polycyclic presentations.
to find a counterexample to an open conjecture, the 'class-breadth' conjecture.
A major breakthrough in the art of computing in finite permutation groups
Algorithms for computing in polycyclic groups using polycyclic presentations
came around 1970 with the base and strong generating set methods and
will form the topic of Chapter 8.
the Schreier-Sims algorithm introduced by Charles Sims in [Sim70, Sim71a].
Also worthy of mention from this period are:
These methods have formed the backbone of this area of CGT ever since, and
they now enable detailed structural computations to be carried out routinely • Dixon's algorithm for computing character tables of finite groups [Dix67];
in permutation groups of degree up to about 107 . Indeed, computational
group theory made the mathematical headlines for the first time in 1972 • the application of the Zassenhaus algorithm and the subgroup lattice
when Sims used his methods as the basis for his proof by computer of the algorithms mentioned above to the determination of the 4-dimensional
existence of the Lyons sporadic simple group [Sim73]. He constructed.jhe space groups [BBN+78];
group as a permutation group of degree 8835156. Later [Sim80], in 1977, he • the implementation by Havas of the Reidemeister-Schreier algorithm for
was able to prove the existence of the Baby Monster in a similar fashion· in computing and simplifying a presentation of a subgroup of finite index
this case the degree is 13571955000. The groups J3, He, and ON were ~so in a finitely presented group;
constructed by computer as permutation groups.
Computing in finite permutation groups will be covered in Chapter 4 and • a program of Felsch and Neubiiser [FN68, FN70] for computing auto-
also in several later chapters. More recent work in this area has made use of morphism groups of finite groups.
4 Handbook of Computational Group Theory A Historical Review of Computational Group Theory 5

Computation with characters and representations of finite groups has played automatic groups also spring to mind; the algorithms involved will form the
a significant role in CGT and continues to do so. The Dixon algorithm for subject of Chapters 12 and 13 of this book.
computing character tables was based on a method described by Burnside However, the prevailing philosophy today is that it is nearly always prefer-
in [Burll], which was used by John McKay in [McK70] in its original form to able to implement algorithms in CGT as a component of an already existing
compute the character tables of the first two Janko sporadic simple groups J 1 package or system. The programmer need not then be concerned with much of
and J 2 • It was later refined by Schneider in [Sch90b]. These methods will be the underlying functionality required, such as multiplication of permutations,
described in Section 7.7. and basic matrix operations, because implementations of 'these will already
As mentioned already, five of the larger sporadic simple groups were proved be provided by the system. Indeed, the user can make use of or build on top
to exist by means of computer constructions as permutation groups. Four oth- of anything that is already there. So the principal objectives of systems of
ers, Ru, Th, HN, and h were originally constructed computationally as matrix this kind are:
groups either over an algebraic number field or over a finite field; for example,
J4 was constructed by Norton and others as a subgroup of GL(1l2, 2) [Nor80]. (i) to provide efficient implementations of a number of basic mathematical
See Section 2 of [Gor82] or [CCN+85] for details; but be aware that the names procedures and data structures; .
used in [Gor82] for several of these groups are different from the ones we are
(ii) to include the most up-to-date and versatile implementations of the
using here, which are as in [CCN+85]. It should be mentioned that several,
fundamental algorithms of CGT;
but not all, of the sporadic simple groups that were originally proved to exist
by means of a computer calculation now have alternative existence proofs as (iii) to provide a high-level programming or scripting language to allow the
subgroups of the largest such group, the Monster, which was constructed by user either to use one of these algorithms to perform a single compu-
R.L. Griess, without computer assistance, as a group of automorphisms of a tation, or to write programs that call one or more of the algorithms,
certain 196854-dimensional algebra [Gri82]. possibly repeatedly.
The principal computational tool that has been used for working with large
groups of matrices over finite fields is a collection of basic linear algebra pro- The programs referred to in (iii) could either be for the purpose of a one-off
cedures designed principally by Richard Parker [Par84] which, among other computation or, at the other extreme, they could be designed to extend the
things, can test large dimensional representations of groups over finite fields functionality of the system, in which case they might eventually be incorpo-
for irreducibility. This has become known as the 'Meataxe' algorithm on rated into it.
account of the efficiency with which it can chop modules up into their irre- The first such system for CGT, known as 'Group', and written and admin-
ducible constituents. These methods will be treated in detail in Chapter 7 istered by John Cannon in Sydney and Joachim Neubiiser in Aachen, got off
of this book. Together with other techniques, they were used extensively in the ground in the mid-1970s, and was written in Fortran. By about 1980,
the construction of Brauer trees by Hiss and Lux in [HL89] and of Brauer this had developed into the Sydney-based CAYLEY system. Around 1993,
character tables by Jansen, Lux, Parker, and Wilson in [JLPW95]. CAYLEY was superceded by MAGMA, which is written in 'C' (see [CP93] or
These constructions involved an important additional technique, unfortu- the Web site and online manual [Mag04]).
nately beyond the scope of this book, known as condensation (see, for ex- A number of more specialized packages, such as CAS [NPP84] for calculat-
ample [Ryb90, LMR94, RybOl]), which can be used to reduce computations ing and manipulating characters of finite groups, were developed and main-
in large dimensional modules to equivalent computations over smaller dimen- tained in Aachen. They were superceded in 1986 by GAP ('Groups, Algo-
sional modules over a different but Morita equivalent algebra. rithms and Programming'); see [GAP04] for detailed information and a list of
The very earliest computer programs for working with groups, such as authors. It has been based in St Andrews, Scotland, since 1997, and is now
Hazelgrove's 1953 implementation of coset enumeration, were written in ma- developed and maintained by an international development team.
chine code. For later programs, Fortran was for many years the most popular GAP and MAGMA are the two systems for CGT that currently enjoy the
choice of language, and more recently 'C' and (less often) 'C++' have been most widespread usage, although the large computer algebra systems Maple
used extensively. There remain a small number of stand-alone programs and and Mathematica both include some facilities for CGT, including coset enu-
packages that are in widespread use, such as ACE3 [HR99] for coset enu- 'meration and the Schreier-Sims algorithm for permutation groups. Both GAP
meration, which we have mentioned already, and the most recent stand-alone and MAGMA aim to cover the whole of CGT and to keep up with all of the
version of the 'Meataxe' programs, written mainly by Michael Ringe [Rin92]. latest developments in the area. MAGMA aims for maximal efficiencyby imple-
Sims' Rutgers Knuth-Bendix Package (rkbp), and the author's KBMAG pack- menting many fundamental algorithms at the 'C' level, whereas GAP aims for
age for the Knuth-Bendix algorithm in groups and monoids and computing in maximal versatility by having more code written in the GAP language, which
6 Handbook of Computational Group Theory A Historical Review of Computational Group Theory 7

allows nonexpert users more flexibility in adapting or extending it. But for There are or have been a number of other more specialized systems and
both systems, a large proportion of the new code, particularly for algorithms packages that deserve a brief mention.
under development, is written in the language of the system.
• Derek Holt and Sarah Rees wrote and maintain an interactive graphics
Until about 1990, CGT was able to exist more or less in isolation from
package QUOTPIC [HR93], which uses the methods to be described in
the rest of computer algebra. In more recent years, that situation has been
Chapter 9 together with other basic algorithms from CGT to compute
slowly but steadily changing, and many of the newer algorithms in CGT have
quotients of finitely presented groups.
required facilities from mainstream computer algebra. For example:
• The New York Group Theory Cooperative directed by Gilbert Baum-
• The best performing Smith and Hermite normal form procedures, which slag and others maintain and develop an interactive system MAGNUS
will arise in Section 9.2 in connection with computing abelian quotients (see the Web site [Mag]) for exploring properties of finitely presented
of finitely presented groups, need the LLL and MLLL (modified) lattice groups. It has the interesting feature that several approaches to solv-
reduction algorithms. ing a problem can be tried and run in parallel; for example, if the user
• The flexible version of the Meataxe algorithm described by Holt and asks whether a group is finite, it will simultaneously run one or more
Rees in [HR94] uses factorization of polynomials over finite fields. programs that attempt to establish finiteness and infiniteness.
• The CHEVIE package of Geck and others for symbolic calculations with
• The polycyclic quotient algorithm, proposed initially by Baumslag, Can-
generic character tables of groups of Lie type, Coxeter groups, Iwahori-
nonito, and Miller [BCM81b, BCM81a], is discussed in Chapters 10 and
Heeke algebras, and other related structures; see the Web site [Che].
11 of [Sim94] for the metabelian case, and has been implemented in the
general case by Lo [L098]. It uses Grabner bases over group rings 7l..G • SYMMETRICA, DISCRETA, and MOLGEN will be mentioned at the
of polycyclic groups. end of Section 10.5.
• The methods under development by Eick for computing with infinite Other recent historical reviews of CGT have been written by Seress [Ser97]
polycyclic groups, which we shall summarize in Subsection 8.2.2 of this and Sims [Sim03]. For a much more detailed account of the early history of
book, require computational facilities from algebraic number theory. CGT and the state of the art around 1970,see the article by Neubiiser [Neu70].
Indeed, there are several papers on CGT in the proceedings of the conference
We can expect interactions of this kind to become more and more common- on computational problems in abstract algebra, which took place in Oxford
place in the future. in 1967 [Lee70]. The first conference exclusively on CGT was in Durham,
Both GAP and MAGMA have responded to this situation by expanding the England, in 1982 [Atk84]. Since then, there have been four conferences on
basic facilities that they provide. GAP is able to do this by means of its CGT in Oberwolfach, Germany, but without proceedings. There have also
package mechanism, which allows external code written either in GAP or in been four organized by DIMACS in the United States, the first three of which
another language, like 'C', to be incorporated into the system by its users. have published proceedings [FK93, FK97, KS97]. In addition, a number of
For example, there is a package of this kind by L. Soicher for computing special issues of the Journal of Symbolic Computation have been devoted to
with combinatorial objects, which includes the 'Nauty' package of Brendan CGT: parts of Volumes 9 (1990), 12 (1991), and 24 (1997).
D. McKay for testing finite graphs for isomorphism. A package providing an
interface to the KANT system for algebraic number theory is also available.
Since its inception, MAGMA has been steadily increasing its basic remit.
Although it started life as a successor to the CGT system CAYLEY, it is
no longer accurate to describe MAGMA as a CGT system. It now offers a
complete and up-to-date range of facilities in computational algebraic number
theory, geometry and combinatorics, as well as for computation in groups,
rings, modules, etc.; indeed the most exciting developments at the present
time are probably "in algebraic geometry. The interaction between CGT and
other branches of computer algebra has been a two-way affair; for example,
some computations in algebraic number theory and algebraic geometry involve
computing with cohomology groups.
Chapter 2
Background Material

In this chapter we shall summarize some algebraic background material, gen-

erally without proofs or with only outline proofs. We are assuming that the
reader is already familiar with the majority of the contents of this chapter,
which is included only as a quick reference guide, and to introduce notation.
The bulk of this material is from group theory and group representation the-
ory, but we have also included a little field theory, including the basic theory
of finite fields, which plays an important role in a number of areas of CGT.

2.1 Fundamentals
Westart by stating some definitions and results from elementary group theory;
see [Rot94], for example, for discussion and proofs. Many readers will want
to skip this section entirely, but one of its functions is to introduce notation.

2.1.1 Definitions

DEFINITION 2.1 A group is a set G together with a binary operation

o : G x G --+ G that satisfies the following properties:
(i) (Closure) For all a, h E G, go h E G;
(ii) (Associativity) For all g, h, kEG, (g 0 h) 0 k = go (h 0 k);
(iii) There exists a unique element e E G satsifying:
(a) (Identity) for all 9 E G, go e = eo 9 = gj and
(b) (Inverse) for all 9 E G there exists a unique h E G such that
hog = go h = e.

This definition assumes more than is necessary; for example, the uniqueness
of the identity and inverse elements can be deduced. See Exercise 1 below.
10 Handbook of Computational Group Theory Background Material 11

The number of elements in a is called the order of G and is denoted by product of n copies of a cyclic group Cp of order p has order pn and is called
IGI. The element e E G satisfying (iii) of the definition is called the identity an elementary abelian p-group.
element of G and, for 9 E G, the element h that satisfies (iii)(b) of the In general, if A, B are subsets of a group G, then we define
definition (h 0 9 = e) is called the inverse of g.
AB := {ab I a E A, se B} and A- 1 := {a- 1 I a E A}.
DEFINITION 2.2 A group is called abelian or commutative if it satisfies
the additional property: (Commutativity) for all g, h E G, go h = hog. 2.1.2 Subgroups

We shall now change notation! The groups in this book will either be:
DEFINITION 2.8 A subset H of a group G is called a subgroup of G if
• multiplicative groups, where we omit the 0 sign (g 0 h becomes just gh), it forms a group under the same operation as that of G.
we denote the identity element by 1 rather than bye, and we denote
the inverse of 9 E G by g-l; or For any group a, G and {I} are subgroups. A subgroup other than G is
• additive groups, where we replace 0 by +, we denote the identity element called a proper subgroup, and a subgroup other than {I} is called a nontrivial
by 0, and we denote the inverse of 9 by -g. subgroup. We often abuse notation and denote the trivial subgroup {I} by 1.
We shall use H ~ G (respectively, H c G) to mean that H is a subset
If there is more than one group around, and we need to distinguish between (respectively proper subset) of G, and H ::::: G (respectively, H < G) to mean
the identity elements of G and H say, then we will write la and IH (or Oa that H is a subgroup (respectively, proper subgroup) of G.
and OH)' A subgroup H of G is called maximal, if H < G and there are no subgroups
Additive groups will always be commutative, but multiplicative groups may K with H < K < G. Note that all nontrivial finite groups have maximal
or may not be commutative. Multiplicative notation is the default. subgroups, but an infinite group need not have any.
Note that in a multiplicative group G, we have (gh)-l = h- 1g-1 for all If H ::::: G, then it is easy to show that IH = la. The following result gives a
g, hE G. The cancellation laws gh = gk =} h = k and hg = kg *
h = k are commonly used criterion for deciding whether a subset H of G is a subgroup.
also easily deduced.
In a multiplicative group G, if n E Nand 9 E a, then gn is defined in-
ductively by gl = 9 and gn+1 = ggn for n :::: 1. We also define gO to be PROPOSITION 2.4 Let H be a nonempty subset of a group G. Then
the identity element 1, and g-n to be the inverse of gn. Then gX+Y = gXgY H is a subgroup of G if and only if
for all x, y E Z. In an additive group, gn becomes ng, where Og = 9 and
(-n)g = -eng). h1,h2EH=}h1h2EH; and hEH=}h- 1EH.
Let 9 E G. Then the order of g, denoted by Igl, is the least n > 0 such that
gn = 1, if such an n exists. If there is no such n, then 9 has infinite order, Let 9 E G. Then the right coset Hg is the subset {hg I h E H} of G.
and we write Igl = 00. Note that if 9 has infinite order, then the elements gX Similarly, the left coset gH is the subset {gh I h E H} of G. (Warning: Of
are distinct for distinct values of z. Similarly, if 9 has finite order n, then the course, by the law of Universal Cussedness, some authors define these the
n elements gO = 1, gl = g, .. . , gn-1 are all distinct, and for any z E Z, gX is other way round!) For additive groups, we write H + 9 rather than Hg.
equal to exactly one of these n elements. Also Igl = 1 {:} 9 = 1, and if Igi ±= n The coset HI is of course H itself. Two cosets H g, and H k are equal if
then, for x E Z, gX = 1 {:} nix. and only if 9 E Hk, which is the case if and only if gk- 1 E H. Otherwise they
If G and H are two (multiplicative) groups, then we define the direct product are disjoint. So the distinct right cosets partition G. Hence we have what is
G x H of G and H to be the set {(g, h) I 9 E a, u e H} of ordered pairs of probably the best-known result in finite group theory:
elements from G and H, with the obvious component-wise multiplication of
elements (gl,h 1)(g2,h2) = (glg2,h 1h2) for gl,g2 E a and lii.ha E H. THEQREM 2.5 (Lagrange's theorem) Let G be a finite group and H a
It is straightforward to check that G x H is a group under this operation. subgroup of G. Then the order of H divides the order of G.
If the groups are additive, then it is usually called the direct sum rather than
the direct product, and written G E& H. The index IG : HI of H in G is defined to be the number of distinct right
More generally, we can define the direct product G" of n copies of G for cosets of H in G, which may be finite or infinite in general, and is equal to
n ~ 1, by a 1 = G and an = Gn-1 x G for n > 1. In particular, the direct a
IGI/IHI when G is finite. A system of representatives in of the right cosets
12 Handbook of Computational Group Theory Background Material 13

of H in G is called a right transversal of H in G, and a left transversal is 2.1.4 Generators

defined correspondingly. Note that T is a right transversal if and only if T-l Let A be any subset of a group G. The subgroup of G generated by A, which
is a left transversal. is denoted by (A), or by (Xl, Xz, ... , x r) if A = {Xl> Xz, ... , Xr}, is defined to
An easy consequence of Lagrange's thereom is that, in a finite group G, the be the intersection of all subgroups of G that contain A. In particular, we say
order Jglof each 9 E G. divides IGI· that A generates G (or is a generating set for G) if (A) = G.
An equivalent definition is that (A) is equal to the set of all elements of G
2.1.3 Cyclic and dihedral groups that can be written as products XIXZ ... Xr for some rENo, where each Xi is
in A or in A-1. The empty product (r = 0) is defined to be la·
A group G is called cyclic if it consists of the integral powers of a single
In practice, and particularly in computational applications, a generating set
element. In other words, G is cyclic if there exists an element 9 in G with the
{Xl> Xz, ... ,xr } often has an ordering that is defined implicitly by the sub-
property that, for all h E G, there exists x E IE with gX = h. The element 9
scripts of the Xi· It is also convenient to allow for the possibility that Xi = Xj
is called a generator of G. when i =I: j, which can result in conflicts with the standard mathematical
The most familiar examples of cyclic groups are additive groups rather than convention that sets may not contain repeated elements. We shall therefore
multiplicative. The group (IE, +) of integers under addition is cyclic, because
frequently refer to generating sequences [Xl,"" Xk], rather than generating
every integer z is equal to xl, and so 1 is a generator. If we fix some integer
sets, where the subgroup generated by the sequence is defined to be the sub-
n > 0 and let lEn = {O, 1,2, ... , n - I} with the operation of addition modulo
group generated by the set consisting of the elements in the sequence.
n, then we get a cyclic group of order n, where 1 is once again a generator,
as is any number in lEn that is coprime to n. Note that in an infinite cyclic
group, any generator has infinite order, and in a finite cyclic group of order 2.1.5 Examples - permutation groups and matrix groups
n, any generator has order n.
Let 0 be any set, and let Sym(O) denote the set of permutations of 0; that
Let n E N with n :?: 2 and let P be a regular n-sided polygon in the plane.
is, the bijections from 0 to itself. Then Sym(O) is a group under composition
The dihedral group of order 2n consists of the isometries of P; that is:
ofmaps. It is known as the symmetric group on O. A subgroup of Sym(O) is
(i) n rotations through the angles 21fkln (0 :::; k < n) about the centre of called a permutation group on O.
P; and For X E 0 and 9 E Sym(O), we shall denote the result of applying 9 to x
by xg rather than by the more usual g(x). The composite gh will mean "first
(ii) n reflections about lines that pass through the centre of P, and either apply g, then apply h", and so xgh = (xg)h.
\ pass through a vertex of P or bisect an edge of P (or both). Let us recall the cyclic notation for permutations. If a1, ... , a r are distinct
""
Unfortunately, some authors denote this group by D n and others by D z n ,
elements of 0, then the cycle (a1'az, ... , ar) denotes the permutation 9 of 0
which can be confusing! We shall use D zn . with af = ai+1 for 1 ::; i < r, a~ = a1, and bg = b for b E 0 \ {a1' az, ... , ar}.
When 0 is finite, any permutation of 0 can be written as a product (or
To study these groups, it is convenient to number the vertices 1,2, ... , n,
composite) of disjoint cycles. Note that a cycle (a1) of length 1 means that
and to regard the group elements as permutations of the vertices. Then the
n rot.ations are the powers ak for 0 :::; k < n, where a = (1,2,3, ... ,n) is a af = a1, and so this cycle can (and normally is) omitted.
rotation through the angle 21f In. Let b be the reflection through the bisector As an example of composition, if 0 = {I, 2, 3}, gl = (1,2,3) and gz = (1,2),
of P that passes through the vertex 1. Then b = (2, n)(3, n-l) (4, n-2)···. For then glgZ = (2,3), whereas 92g1 = (1,3). The inverse of a permutation
example, when n = 5, b = (2,5)(3,4) and when n = 6, b = (2,6)(3,5). Notice can be calculated by reversing all of the cycles. For example, the inverse of
that there is a difference between the odd and even cases. When n is odd (1,5,3,6)(2,8,7) is (6,3,5,1)(7,8,2) = (1,6,3,5)(2,7,8). (The cyclic repre-
b fixes no vertex other than 1, but when n is even, b fixes one other vertex: sentation is not unique: (a1' az, . . . ,ar) = (az, a3, . . . , ar, a1), etc.)
namely (~+ 2)/2. Now we can see, either geometrically or by multiplying A cycle of length 2 is called a transposition. Since any cycle can be written
permutations, that the n reflections of P are the elements akb for 0 < k < n. as a product of transpositions ( (1,2,3, ... , n) = (1,2)(1,3)(1,4) ... (1, n), for
Thus we have - example), it follows that any permutation on 0 can be written as a product
of transpositions. A permutation is called even if it is a product of an even
G = {ak I 0 ::; k < n} U {ak b I 0 ::; k < n}. number of transpositions, and odd if it is a product of an odd number of
transpositions. It can be proved that no permutation is both even and odd,
In all cases, we have ba = an - 1b (= a- 1b). so the definition makes sense.
14 Handbook of Computational Group Theory Background Material 15

The set H of even permutations forms a subgroup of Sym(O). This is known taining A. An equivalent definition of (AG) is the subgroup of G generated
as the alternating group on 0, and is denoted by Alt(O). If (x, y) is some fixed by the set A G := {g-l ag I 9 E G,a E A}.
transposition on 0 then, for any 9 E Sym(O), one of the two permutations 9
and gx is even and the other is odd. Hence G = H U H(x, y), so IG: HI = 2.
Let K be a field. Then, for any fixed d > 0, the set of d x d invertible 2.1.7 Homomorphisms and the isomorphism theorems
matrices with entries in K forms a group under multiplication. This group is
denoted by GL(d, K) or GLd(K).
DEFINITION 2.8 Let G and H be groups. A homomorphism cp from G
The d x d matrices over K with determinant 1 also form a group denoted by
to H is a map ip : G -+ H such that cp(glg2) = cp(gl)CP(g2) for all gl,g2 E G.
SL(d,K) or SLd(K). This is a subgroup of GL(d,K). WhenK is a finite field
JFq of order q, we use the alternative notation GL(d, q), etc. There are a num-
. It is straightforward to prove that any homomorphism sp : G -+ H maps IG
ber of important subgroups of GL( d, q) known as the classical groups. These
to IH, and that cp(g-l) = cp(g)-l for all 9 E G. . .
include the symplectic groups Sp(d, q), the unitary groups GU(d, q), SU(d, q),
A homomorphism cp is called, respectively, a monomorphism; an epunor-
and two types of orthogonal groups GO(d, q). We refer the reader to [CCN+85]
phism or an isomorphism, if it is, respectively, an injection, a surje~tion or a
for definitions, essential properties, and further references.
bijection. A homomorphism from G to G is called an endomorphism of G,
fuld an isomorphism from G to G is called an automorphism of G.
2.1.6 Normal subgroups and quotient groups The automorphisms of G form a group, denoted by Aut(G), under composi-
tion. This is of course a subgroup of Sym( G), and so we follow our convention
for permutation groups, and represent the result of applying a E Aut(G) to
DEFINITION 2.6 A subgroup H of a group G is called normal in G if
9 E G by ga rather than by a(g). So, for a,(3 E Aut(G), a(3 means first a
the left and right cosets gH and Hg are equal for all 9 E G.
then (3. For any kEG, the map ak : G --+ G defined by gak = k- 1 gk is an
automorphism of G. This type of automorphism is known as an inner auto-
For example, {I} and G are normal subgroups of any group G, and any
morphism, and the set of all inner automorphisms forms a normal subgroup
subgroup of index 2 is normal, but the subgroup {I, (1, 2)} of Sym( {I, 2, 3})
is not normal. ~~cl~~. .
An automorphism that is not inner is called an outer automorphism and
An equivalent condition, which is often used as the definition, is that H is the outer automorphism group Out( G) is defined to be the quotient group
a normal subgroup of G if and only if g-lhg E H for all hE H, 9 E G.
Aut(G)/ Inn(G).
The standard notation for "H is a normal subgroup of G" is H <l G or Two groups G and H are called isomorphic, and we write G ~ H, if there is
H <l G. We shall use H <l G to mean that H is a proper normal subgroup of an isomorphism cp : G -+ H. The inverse of an isomorphism is an isomorphism,
G .::::-i.e., H:f: G, but this usage is not universal. so G ~ H implies H ~ G, and G ~ Hand H ~ K imply G ~ K.
A critical property of normal subgroups N of groups G, which is not shared In many contexts isomorphic groups are regarded as being the same group.
by subgroups in general, is that the product of any element n1g of the coset It is easy to prove that any two cyclic groups of the same order (finite or
N 9 with any element n2h of the coset Nh lies in the coset N gh. Hence infinite) are isomorphic, so we talk about the infinite cyclic group, and the
(Ng)(Nh) = Ngh, and the set of right cosets Ng of N in G forms a group cyclic group of order n, which we shall denote by en in this book. Similarly,
of order IG: NI under multiplication. This group is denoted by G/ N and is if n,~ are sets with 101 = I~I, then Sym(n) ~ Sym(~) and Alt(O) ~ Alt(~),
called the quotient group (or sometimes factor group) of G by N. so, for any n > 0, we can talk about the symmetric and alternating groups
A group is called simple if IGI > 1 and the only normal subgroups of G are Sym(n) and Alt(n) acting on a set of size n. By default, we shall regard these
{I} and G. For example, cyclic groups of prime order are simple. groups as acting on the set 0 = {I .. n}.
'. Let cp : G -+ H be any group homomorphism. Then the kernel ker( cp) of cp
DEFINITION 2.7 Let A be a subset of a group G. The normal closu~e is defined to be the set of elements of G that map onto IHi that is,
of A in G, which is denoted by (AG), is defined to be the intersection orall
normal subgroups of G that contain A. ker(cp) = {g I 9 E G, cp(g) = IH}.
It can be shown that ker(cp) is always a normal subgroup of G, whereas the
Since the intersection of any collection of normal subgroups of G is easily image im(cp) of ip is a subgroup, but not necessarily a normal subgroup, of H.
shown to be normal in G, (AG) is the smallest normal subgroup of G con- Furthermore, if N is any normal subgroup of G, then there is an epimorphism,
16 Handbook of Computational Group Theory Background Material 17
known as the natural epimorphism 'P : G -+ G / N defined by 'P(g) = N g. So 4. Prove that the two definitions of the subgroup generated by a subset of a
the set of normal subgroups of G is just the set of kernels of homomorphisms group G given in Subsection 2.1.3 are equivalent, and that the two defini-
with domain G. tions of the normal closure of a subset of a group are equivalent.
We now state the three isomorphism theorems for groups. The first of these
is the most important. For the second, note that if H is any subgroup and 5. Prove that a cyclic group Cn of order n has exactly il'(n) generators, where
K is a normal subgroup of G, then HK = KH is a subgroup of G. It is a <I>(n) = I{m E [l..n] I gcd(m, n) = 1}1 is the Euler il'-function. Deduce
normal subgroup if His. that IAut(Cn )I = il'(n).

6. Prove that the cyclic groups of prime order are the only (finite or infinite)
THEOREM 2.9 (first isomorphism theorem) Lei ip : G -+ H be a homo- abelian simple groups.
morphism with kernel K. Then G/K ~ im('P)' More precisely, there is an
isomorphism 7j5 : G / K -+ im( 'P) defined by 7j5(Kg) = 'P(g) for all 9 E G.

THEOREM 2.10 (second isomorphism theorem) Let H be any subgroup

and let K be a normal subgroup of a group G. Then H n K is a normal 2.2 Group actions
subgroup of Hand H/(H n K) ~ HK/K.
2.2.1 Definition and examples

THEOREM 2.11 (third isomorphism theorem) Let K ~ H ~ G, where

Hand K are both normal subgroups oiG. Then (G/K)/(H/K) ~ G/H. DEFINITION 2.12 Let G be a group and n a set. An action of G on n
is defined to be a homomorphism 'P : G -+ Sym(n).
Note that if ak is the inner automorphism of a group G defined as described
above by kEG, then the map G -+ Aut(G) defined by k f--+ ak is a homomor- So, for each 9 E G, 'P(g) is a permutation of n. When we are talking about
phism with image Inn(G). Its kernel is {k E G I k-1gk = 9 Vg E G}, which a fixed action 'P, we shall usually just write a g rather than a<p(g) for the action
is defined to be the centre Z(G) of G. So, by the first isomorphism theorem, of 'P(g) on a E n. The fact that 'P is a homomorphism then translates into
we have G/Z(G) ~ Inn(G). the property a 9h = (a 9 ) h for all g,h E G, a En.
Notice that it is inherent in our notation a 9 that G is acting on n from
Exercises the right. Actions from the left are more common in the literature, largely
because this is more consistent with the convention of writing functions on the
1. Let G be a set endowed with an associative binary operation 0 : G x G -+ G igft of their arguments. However, both GAP and MAGMA use actions from the
that contains an element e E G such that: right, and we shall consistently adopt this convention in this book. Similarly,
elements of vector spaces over a field K will be represented by row vectors
(a) for all 9 E G, eo 9 = g; and v E s», and a linear map Kd -+ K d will be represented by a d x d matrix a
over K, which acts on the vectors from the right; that is, v f--+ v·a.
(b) for all 9 E G there exists h E G such that hog = e. The property 'P(1G) = 1sy m(n) translates into a 1 = a for all a E n. In fact,
Prove that G is a group under the operation o. an equivalent way of defining a group action is as a map n x G -+ n with
(a,g) f--+ a 9 , which satisfies a 1 = a and a 9 h = (a 9 ) h for all g,hE G, a E n.
2. Prove that the number IG: HI of distinct right cosets of a subgroup H of a Some authors define a group action like this, and call a homomorphism from
group G is equal to the number of distinct left cosets. (This is easy if G is G to Sym(n) a permutation representation of G on n, but we shall regard the
finite, since both numbers are IGI/IHI, but less easy when G is infinite.) 'expressions "action of G on n" and "permutation representation of G on D"
as being synonymous. The image im('P) of the action will be denoted by Gn.
3. Let a and b be the generators of the dihedral group D 2n , as in Subsec- Two actions !.pI and !.p2 of G on sets D1 and D2 are said to be equivalent,
tion 2.1.3. Let H be a subgroup of D2n that is contained in the cyclic Ii{ there is a bijection 7 : D1 -+ D2 such that 7(a 9) = 7(a)9, for all a E D1
subgroup (a). Show that H is normal in D2n and that either H = (a), or and 9 E G. This just says that the actions are the same up to a 'relabelling'
G/ H is isomorphic to a dihedral group. of the elements of the set on which G is acting.
18 Handbook of Computational Group Theory Background Material 19

The kernel K = ker( <p) of an action ip is equal to the normal subgroup The edges of the Cayley graph define the action of the generators X in the
right regular action, of G.
{g E G I a g = a for all a EO}.
The action is said to be faithful if K = {I}. In this case, Theorem 2.9 says 2.2.2 Orbits and stabilizers
that G ~ GiK ~ im(<p).

.DEFINITION 2.14 Let G act on the set O. We define a relation rv on

Example 2.1
If G is a subgroup of Sym(O), then we can define an action of G on 0 simply
o by a rv {3 if and only if there exists agE G with {3 = a • Then rv is an
g

equivalence relation. The equivalence classes of rv are called the orbits of G

by putting <p(g) = 9 for 9 E G. This action is faithful. 0 on O. In particular, the orbit of a specific element a E 0, which is denoted
by a G , is {a 9 I g E G}.
Example 2.2
Let G = {I, a, a2, as, a4,a5, b,ab,a2b, aSb, a4b, a5b} = D 12 be the group of DEFINITION 2.15 Let G act on 0 and let a E O. Then the stabilizer
isometries of a regular hexagon P. In Subsection 2.1.3, we defined a and b to of a in G, which is denoted by Ga or by StabG(a), is {g E G I 0. 9 = a}.
be the permutations (1,2,3,4,5,6) and (2,6)(3,5) of the set {I, 2,3,4,5, 6} of
vertices of P, and so we have an action of G on this vertex set. It is straightforward to check that G a is a subgroup of G and that the kernel
There are some other related actions, however. We could instead take 0
of the action is naEo Ga.
to be the set E = {e1' e2, es, e4,e5, e6} of edges of P, where e1 is the edge
joining 1 and 2, e2 joins 2 and 3, etc. The action <p of G on E is then given
by <p(a) = (el,e2,e3,e4,e5,e6)' <p(b) = (e1,e6)(e2,e5)(eS,e4). This action is Example 2.4
still faithful, but is not equivalent to the action on vertices. Let G act by right multiplication on the set of right cosets of the subgroup
As a third possibility, let D = {d1' d2, ds} be the set of diagonals of P, H, as in Example 2.3 above. Since H9 = Hg for any 9 E G, HG = 0 and
where d1 joins vertices 1 and 4, d2 joins 2 and 5, and ds joins 3 and 6. Then there is only one orbit. The stabilizer of the coset H9 is the subgroup g-l H9
the action of G on D is defined by <p(a) = (d1,d2,d s), and <p(b) = (d2,d3). 6f G, and so the kernel of the action is equal to n 9EG g-l Hg. This (normal)
This action is not faithful, and its kernel is the normal subgroup {I, as} of G. subgroup of G is called the core of H in G. It is denoted by HG, and is the
The image is isomorphic to D 6 • 0 largest normal subgroup of G that is contained in H. 0

Example 2.3 THEOREM 2.16 (the orbit-stabilizer theorem) Let the finite group G
There is a faithful action called the right regular action, which we can define I,J,ct on 0, and let a E O. Then IGI = laGIIGal.
for any group G. Here we put 0 to be the underlying set of G and simply
define a g to be ag for all 9 E G, a E n.
More generally, let H be any subgroup of G, and let 0 be the set of right PROOF Let {3 E (XG. Then there exists agE G with (X9 = {3. If g' E G
cosets of H in G. Again we can define an action of G on n by a g := ago This with (Xl = {3, then g'g-l EGa, so g' E Gag, and we see that the elements g'
is known as a coset action. The right regular action corresponds to the case with (Xl = {3 are precisely the elements of the coset Hg. But IHgj = IHI, so
H= 1. 0 for each {3 E (XG, there are precisely IHI elements g' of G with 0. 9 = {3, and
the total number of such {3 E a G must be IGI/IHI, which proves the result .
The right regular action of a group G is closely related to its Cayley graph
rx(G), which is defined with respect to a generating set X of G. • Dually to the stabilizer G a, for 9 E G, we define

DEFINITION 2.13 If G = (X), then the Cayley graph rx(G) is a 0 9 := { a E 0 I 0. 9 = a }

directed labelled graph with vertex set V := {v g I 9 E G} in one-one corre-
spondence with the set of elements of G and, for each 9 E G and x EX, an to be the fixed point set of g. The following result, which is often known as
edge labelled x from v g to Vg",. It is customary to identify each such edge with Burnside's lemma, but was actually known earlier to Cauchy and Frobenius,
an edge labelled x- 1 from Vg", to v9 . says that the number of orbits of an action is equal to the average number
20 Handbook of Computational Group Theory Background Material 21

of fixed points of the group elements. This result, together with refinements Example 2.5
and generalizations, is the basis of countless applications of group theory to Let G = Sym(!1) and let t, 9 E G. Write 9 in cyclic notation, and suppose that
the enumeration of combinatorial structures. We shall say some more about one of the cycles of 9 is (a1,aZ, ... ,ar). Then ai = az, and so ai! = a~ and
this later in the book (Section 10.5). hence (a{)r"g! = a~, and we find that f-1 gf has a cycle (ar, a~, ... , at).
In other words, given a permutation 9 in cyclic notation, we obtain the
LEMMA 2.17 (the Cauchy-Frobenius lemma) Let the finite group G act conjugate f-1 gf of 9 by replacing each element a E 0 in the cycles of 9
on O. Then the number of orbits of G on 0 is equal to 2:gEc 10gl· rbr by at. For example, if 0 = {I, 2, 3, 4, 5, 6, 7}, 9 = (1,5)(2,4,7,6) and f =
(1,3,5,7,2,4,6), then r 1
gf = (3,7)(4,6,2,1).
It follows that two permutations of 0 are conjugate in Sym(O) if and only
PROOF if they have the same cycle-type; that is, the same number of cycles of each
length. For example, Sym(4) has five conju~acy classes, with representatives
L 10gi = L L 1= L L 1= L IGal = IGI L la
cl- 1,
1, (1,2), (1,2)(3,4), (1,2,3), (1,2,3,4). U
gEG gEC aEflg aEfl gEG" aEn aEn

where the last equality follows from the orbit-stabilizer theorem. But since We can also take 0 to be the set of subgroups H of G, and use the corre-
L:aEn laG 1-1 is the total number of orbits of G on 0, the result follows. I sponding action Hg = g-l Hg of G on O. Again, subgroups in the same orbit
of this action are called conjugate subgroups. The stabilizer of a subgroup H
For future use, we introduce some further notation. is equal to {g E G I g-l Hg = H}, which is called the normalizer of H in G
and denoted by NG(H). Then clearly H ::::1 NG(H). By the orbit-stabilizer
theorem, the number of conjugates of H in G is equal to IG:NG(H)I.
DEFINITION 2.18 Let G act on 0 and let ~ ~ O. Then
The centralizer CG(H) of H in G, on the other hand, is defined to be the
(i) The setwise stabilizer {g E G I a g E ~ for all a E ~} of ~ in G is subgroup {g E G I gh = hg for all hE H} of G, and so Cc(G) = Z(G).
denoted by G t::..
(ii) The pointwise stabilizer {g E G I a g = a for all a E ~} of ~ in G is 2.2.4 Sylow's theorems
denoted by G(t::.).
We state Sylow's theorems as a single theorem. There are many different
proofs. See, for example, Chapter 4 of [Rot94] and Section 1.5 of [Cam99] for
2.2.3 Conjugacy, normalizers, and centralizers two different proofs. If G is a finite group, p is a prime, and IGI = pmq with
PAq, then a subgroup of G of order pm is called a Sylow p-subgroup of G, and
In Example 2.3, a group G was made to act on the set of its own elements
the set of all such subgroups is denoted by Sylp(G). In general, a group of
by multiplication on the right. There is another important action of G on
order a power of a prime p, including the trivial group of order po, is known
0= G, which is defined by
as a p-group.
a g = g-l a g for g, a E G.
This is called conjugation. The orbits of the action are called the conjugacy THEOREM 2.19 (Sylow's theorems.) Let G be a finite group, let p be a
classes of G, and elements in the same conjugacy class are said to be conjugate prime number, and let IGI = pmq where PAq. Then:
in G. So g, h E G are conjugate if and only if there exists f E G with h =
r 1gf · We shall write Clc(g) for the orbit of g; that is, the conjugacy class (i) For any 1 with 0 ::; 1::; m, the number of subgroups of G of order pi is
congruent to 1 modulo p. In particular, such subgroups exist for all such
containing g. Since f-1 gf is the image of 9 under the inner automorphism of
l, and ISylp(G)I == 1 mod p.
G defined by t, conjugate elements must have the same order.
The stabilizer Gg in this action consists of the elements f E G for which (ii) Any two subgroups in Sylp(G) are conjugate in G, and ISylp(G) I =
g! = g; that is, fg = st- In other words, it consists of those f that IG:NG(P)I for any P E Sylp(G).
commute with g. It is called the centralizer of 9 in G and is written as
Cc(g). By applying the orbit-stabilizer theorem, we get the well-known for- (iii) For 0 ::; 1 < m, any subgroup of G of order pi is contained in some
mula ICIG(g)! = IGI/ICG(g)! for all 9 E G. The kernel K of the action is the subgroup of order pl+!. Hence any such subgroup is contained in a Sylow
centre Z(G) of G, which we encountered at the end of Subsection 2.1.7. p-subgroup.
22 Handbook of Computational Group Theory Background Material 23

We state without proof a couple of easy consequences of Sylow's theorems. H:= Go:. For gl,g2 E G, we have a g1 = a g 2 if and only if Hg 1 := Hg 2, so we
can define a bijection r : ~ -t 0 by r(H g) = a g , and r is easily seen to be an
equivalence of actions. For any {3 E 0, we have G{3 = g-lG exg, where a g = {3.
PROPOSITION 2.20 Let G be a finite group, p a prime, P E Sylp(G),
It follows from (ii) that 'PHl and 'PH2 are equivalent if and only if H2 is the
and H a subgroup of G with NG(P) ::; H. Then NG(H) = H.
stabilizer of a point in the action <PH1 , which is the case if and only if H1 and
H 2 are conjugate in G. I
PROPOSITION 2.21 (the F'rattini lemma.) Let G be a finite group, p
a prime, N a normal subgroup of G and P E Sylp(N). Then G = NNG(P).
DEFINITION 2.24 If G acts on 0, then Gfl is regular if it is transitive
and Gex := 1 for some (and hence all) a E 0.
2.2.5 Transitivity and primitivity
The right regular action of G on its own underlying set defined in Exam-
ple 2.3 is indeed regular. By the orbit-stabilizer theorem, if G is finite and
DEFINITION 2.22 An action of G on a set 0 is called transitive if there Gfl is regular, then IGI = 101.
is a single orbit under the action. That is, for any a, {3 E 0, there exists 9 E G If G acts on 0, then a nonempty subset Ll. of 0 is called a block under the
with a g = {3. Otherwise, it is called intransitive. action if, for all 9 E G, we have Ll.g = Ll. or Ll.g n Ll. := <p. The block is called
Let n E N with n > O. The action is n-fold transitive (or just n-transitive for nontrivial if ILl. I > 1 and Ll. =!= 0. The orbits of any action are clearly blocks,
short) if 101 :?': n, and for any two ordered lists [al' a2,· . . ,an], [{31, {32, ... , {3n] but we normally restrict the consideration of blocks to transitive actions.
of distinct points in 0, there exists 9 E G with af = Pi for 1 ::; i ::; n.

Note that I-transitive is the same as transitive. Also, the condition 101 :?': n DEFINITION 2.25 A transitive action of G on a set 0 is called primitive
for n-fold transitivity ensures that n-transitive actions are (n-I)-transitive if there are no nontrivial blocks under the action. Otherwise, it is called
for n > 1. By "Gfl is transitive" we mean "the action of G on 0 is transitive" . imprimitive.

Example 2.6 If Ll. is a block under an action, then the distinct translates Ll.g of Ll. under G
partition 0. The set of translates is known as a block system. So a transitive
If 101 = n, then Sym(O) is l-transitive for alll with 1 ::; 1 ::; n, and Alt(O) is
(l-2)-transitive for 3::; 1::; n. 0 action is primitive if and only if it does not preserve a nontrivial partition of
n. Note also that 1Ll.1 = l6.gl, so all blocks in such a partition have the same
The following result says that all transitive actions are equivalent to the size. Hence if Gfl is transitive and 101 is prime, then en is primitive.
action described in Example 2.4 for some subgroup H of G.
Example 2.7
PROPOSITION 2.23 Let G = D 12 act on the vertices of a regular hexagon, as in Example 2.2
above. Then G preserves two nontrivial block systems, { {I,3,5}, {2,4,6}}
(i) Let H be any subgroup of G, and let <PH be the action of G by right and { {I, 4}, {2,5}, {3, 6} }. 0
multiplication on the right cosets of H in G (Example 2.4). Then 'PH
is transitive. If G ::; Sym(O) and we say that G is (n-fold) transitive or primitive, then
(ii) If <P : G -t Sym(O) is any transitive action of G and a E 0, then 'P is we are referring to the obvious action of G on 0 defined in Example 2.1.
equivalent to 'PH, where H = Gex• There is an alternative method of describing block systems, which will often
be convenient for us to employ.
(iii) If H1,H2 ::; G, then <PHl and <PH2 are equivalent if and only if H 1 ana
H 2 are conjugate in G.
DEFINITION 2.26 If G acts on 0, then an equivalence relation rv on 0
is called a G-congruence if a rv {3 implies a g rv {3g for all a, {3 E 0 and 9 E G.
PROOF (i) is clear since, for any cosets Hg 1 and Hg2 of H, we have
(Hg1)(g11g2) = Hg 2. For (ii), let ~ be the set of right cosets of H in G, with Assume that Gfl is transitive. It is clear that if GO has block system
24 Handbook of Computational Group Theory Background Material 25

{Ll9 I 9 E G}, then the equivalence relation on n defined by a '" /3 if and only PROOF Let Gn be 2-transitive, and let Ll be a block with I~I > 1. Let
if a and /3 lie in the same block is a G-congruence. Conversely, the equivalence a,} E ~ and a =1= , E n. Then by 2-transitivity, there exists 9 E G with
classes defined by any G-congruence on n form the blocks of a block system. a9 == a and /39 =="so a E Ll n ~9, and hence Ll == Ll9, and, E Ll. Thus
So, for transitive actions, G-congruences and block systems are essentially the Ll== n, which proves (i).
same thing. ' Let ~ = aN be an orbit of Nn, let 9 E G, and suppose that a E ~ n ~g.
We remind the reader that, formally speaking, an equivalence relation", on So a == /39 for some /3 E Ll. Then for any a h E Ll with hEN, normality of
a set n is a subset R: of n x n, where a '" /3 means the same as (a,/3) En. N gives gh == h' 9 for some h' E N, and a h = /39 h = /3h' 9 E Ll9. So Ll = Ll9,
proving (ii). (iii) follows easily from (ii). I
DEFINITION 2.27 If S ~ n x n, then the G-congruence generated by
,PROPOSITION 2.30 Let Gn be a transitive action and let a E n. Then
S is defined to be the intersection of all G-congruences that contain S.
(i) If K is a subgroup with GOt ::::: K ::::: G, then a K is a block and, conversely,
PROPOSITION 2.28 Let S ~ n x n, and let n be the G-congruence any block containing a is equal to a K for some K with GOt::::: K ::::: G.
generated by S. Call, a, /3 E n directly equivalent if, for some 9 E G, we have
'Yii) If Inl > 1, then GO is primitive if and only if GOt is a maximal subgroup
a =,9 and /3= 69, where either (,,6) E S or (6,,) E S. Then (a,/3) E R if
afG.
and only if there exist a = ao, aI, ... ,ar = /3 in n with r ?: 0, such that each
ai is directly equivalent to aiH'
PROOF Let GOt ::::: K ::::: G and Ll = a K . Suppose that /3 E Ll n Ll9 for
some 9 E G. So /3 = a h = a k9 for some h, k E K. But then kgh- l E GOt ::::: K,
PROOF Define the relation rv on n by a rv /3 if and only if there exists a = so 9 E K and Ll9 = Ll. Conversely, if ~ is a block containing a, then let
ao, aI, ... ,ar = /3 as in the statement of the proposition. Then clearly a rv /3
implies that (a, /3) is in any G-congruence containing S, and hence (a, /3) E R.
K = {g E G I a 9 ELl}. Certainly GOt::::: K. For any k E K, we have
ali, E ~n~k, so Ll = Llk, and hence K = {g E G I ~9 = ~}, which is a
Conversely, it is straightforward to check that '" is a G-congruence, and the subgroup of G. This proves (i), and (ii) follows immediately from (i). I
result follows. I
Let n be the disjoint union of sets ~l,"" Llm , where each ILlil = l, and
let G be the subgroup of Sym(n) preserving this partition. Then G is the Exercises
wreath product of the permutation groups Sym(l) and Sym(m), and we write
G = Sym(l) I Sym(m). It has a normal subgroup H ~ Sym(l)m that fixes all h-,Let G act on n and let a E nand 9 E G. Prove that GOtg = g-lGOtg.
of the Lli, and G/H ~ Sym(m). See, for example, Section 5.9 of [Hal59] for t·"

details of the theory of wreath products of permutation groups. 2. Find generators of the wreath product Sym(l) I Sym(m).
We shall now give brief proofs of some standard results pertaining to tran- 3. Let G be a group of order pn for p prime, and let N be a normal subgroup
sitivity and primitivity. .of G with N =1= 1. By adding up the sizes of the conjugacy classes of G that
are contained in N, prove that N n Z(G) =1= 1. In particular, if n > 0, then
PROPOSITION 2.29 Z(G) =1= 1.
\",

4.,Let H be a subgroup of a group G and let HG be the core of H in G, as

(i) A 2-fold transitive action is primitive. defined in Example 2.4. Prove that H G ::::: H and that, if K is any normal
subgroup of G with K::::: H, then K::::: HG.
(ii) Let G act on n, and let N:::! G. Then the orbits of Nn are blocks under
the action ofG. Calculate the conjugacy classes of the dihedral group D 2n . (The cases when
is odd and even are different.)
,'11,

(iii) If Gn is primitive, and Nn is nontrivial, then s» is transitive. that a normal subgroup of a group G is a union of conjugacy classes
ofG. '
26 Handbook of Computational Group Theory Eackground Material 27

7. Let H = Alt(O) and G = Sym(O), and let 9 E H. Show that either COROLLARY 2.32 A minimal normal subgroup N of a finite group G
CH(g) = Ca(g) and CIa (g) is the union of two equally-sized conjugacy is isomorphic to a direct product of isomorphic simple groups. In particular,
classes of H, or ICa(g) : CH(g)1 = 2 and CIa (g) = CIH(g). Show that if N is abelian, then it is an elementary abelian p-group for some prime p.
Alt(5) has five conjugacy classes, with orders 1, 12, 12, 15, 20, and deduce
that Alt(5) is a simple group.
2.3.2 Series
Let G be an O-group, and let

G = Go t> G1 t> G z t> ... t> G; = {I}

2.3 Series be a finite series of O-invariant subgroups in which each group is normal in
the preceding group. Such a series is called an (O-invariant) subnormal series
2.3.1 Simple and characteristically simple groups
of G. It is called an O-composition series if each factor GdG i+! is O-simple.
Let G be a group and let 0 be a subgroup of Aut(G). We call G an O-group for We shall be mainly interested in the cases 0 = {I}, which is an ordinary
short. A subgroup H of G is called O-invariant if gOi E H for all 9 E H, a E O. composition series with simple factor groups, and the case 0 = Inn(G), where
So, if 0 = Inn(G), then an O-invariant subgroup is the same thing as a normal an O-composition series is called a chief series for G, and the factors GdGi+1
subgroup. An Aut(G)-invariant subgroup is called a characteristic subgroup. are characteristically simple groups.
For example, the centre Z(G) of G is characteristic in G. We write H charG . It is clear that finite groups have finite composition series and finite chief
to mean that H is characteristic in G. shies. For an infinite group, such finite series mayor may not exist, and there
If H <I K and K <I G, then it is not necessarily true that H <I K. However, are generalizations which allow infinite series. But for any group G and any
H char K and K <lG imply H <lG, and H char K and K charG imply H charG. o : :; Aut(G), we have the Jordan-Holder theorem; see, for example, Section
A subgroup H of G is called subnormal in G if there exists a finite chain f.10 of [Sco64].
H = H o <I HI <I ... <I u; = G. r-

An O-group G is called O-simple if IGI > 1, and G has no proper nontrivial THEOREM 2.33 Let G be an O-group, and let
O-invariant subgroups. So a simple group is the same as an Inn(G)-simple
group. The celebrated classification of finite simple groups was completed G = Go e- G1 e- Gz t> ... e- c; = {I}
around 1980. The finite nonabelian simple groups comprise a number of
infinite families of groups, including Alt(n) for n ~ 5, the classical groups and and
the exceptional groups of Lie type, together with the 26 sporadic groups. We H= H o e- HI e- Hz e- ... e- tt, = {I}
shall assume that the reader has some basic familiarity with these groups, and ~e two finiteO-composition series of G. Then r = s and there is a bijection
refer to [CCN+85] or [Car72] for any detailed information that we may need if{.: [1..r] -t [1..r] such that Gi-IiG i and Hj-Ii H j are O-isomorphic for
about them. 1/:::; i :::; r, where j = <p(i).
An Aut(G)-simple group is called characteristically simple. The following
result, proved in Theorem 5.20 of [Rot94], describes the finite examples. We generally refer to the collection of isomorphism types of composition
fIJ.ctors or chief factors of a finite group G simply as the composition/chief
THEOREM 2.31 A finite group is characteristically simple if and only if 1~l::tors of G. This collection is technically a multiset, because it can have
r§peated elements.
it is isomorphic to a direct product of one or more isomorphic simple groups.

The simple groups in this theorem can be either abelian, in which case the The derived series and solvable groups
group itself is elementary abelian, or nonabelian. ,;.
For g, hE G, we define the commutator [g, h] := g-Ih-Igh. If Hand K are
A normal subgroup N of a group G is called a minimal normal subgroup of
.saogroups of G, then we define [H, K] := ([h, k]1 hE H, k E K). If there is
G if N =f: 1 and there is no normal subgroup M of G with 1 < M < N. A
aserious danger of confusion with the same notation [g, h] used to mean the
minimal normal subgroup must be characteristically simple, and so we have
'Ordered list of elements g, h E G, then we shall write comm(g, h) in place of
the following corollary.
i'lg.,hj. The followinglemma lists some elementary properties of commutators.
28 Handbook of Computational Group Theory Background Material 29

Recall that, for x,y E G, x Y is defined to be y-lxy. We denote the inverse of where G[i] := [G[i-l], G[i-l]] for i ::::: 1. (Wer:have used G[i] rather than the
x Y by «:», more standard G(i) to avoid clashing with our use of G(i) in Chapter 4 in the
context of stabilizer-chains of permutation groups.)
Each factor G[i-l] jG[i] in the series is abelian. Ifthere is an r such that
LEMMA 2.34 Let x, y, z be elements of a group G and let H, K be sub- G[r] = {1} or G[r] is perfect, then the series terminates at G[r] and is finite
groups of G. Then: of length r. If it terminates with afr] = {1}, then we say that G is solvable
(or soluble) with derived length r, A group is solvable if and only if it has
(i) [XV, z] = [x, z]Y[y, z] and [x,yz] = [x, z][x, yV; a normal series with abelian factors. A finite group is solvable if and only
(ii) [H, K] is a normal subgroup of (H, K); if it has a subnormal series with cyclic factors or, equivalently, if all of its
composition factors are cyclic of prime order.
(iii) [H,K] = [K,H]; If G[i] = G[i+l] for some i, then we define G[oo] := G[i] , in which case G[oo]
iil the unique smallest normal subgroup of G with solvable quotient group.
(iv) H:5 Na(K) ~ [H,K]:5 K; More generally, a group with a finite subnormal series with (possibly infi-
nite) cyclic factors is called polycyclic. Equivalently, a polycyclic group is a
(v) [[x, y-l], z]Y[[y, Z-l], x]Z[[z, X-I], y]" = 1. solvablegroup in which each factor G[i-l] jG[i] is finitely generated. Polycyclic
groups are important in computational group theory because there are effi-
The proof is left as an exercise. For (ii), notice that (i) implies [x, z]Y = ~ient algorithms for computing within them. Indeed, algorithms for polycyclic
[xy,z][y,zt l, and deduce that H :5 Na([H,K]). Show K :5 Na([H,K]) groups is the subject of Chapter 8.
similarly. For (iii), use [x, y]-l = [V, x].
The following result will be used for computing commutator subgroups.
~.3.4 Central series and nilpotent groups
A normal series
PROPOSITION 2.35 If H = (X) and K = (Y) are subgroups of a
group G, then [H,K] = (Z(H,K»), where Z = {[x,y] I x E X,y E Y}. G = Go ~ Gl ~ G2 ~ ... ~ Gr = {1}
of G is called a central series if Gi-dGi :5 Z(GjGi) for 1 :5 i :S r. A group
is called nilpotent if it has such a series, and the (nilpotency) class of G is
PROOF Let N = (Z(H,K»). By Lemma 2.34 (ii), [H,K] ~ (H,K), so
defined to be the length r of the shortest central series for G. Clearly every
N:5 [H,K]. By Lemma 2.34 (i), we have [x-l,y] = [x,y]_"-" and [x,y-l] = nilpotent group is also solvable.
[x,y]-Y-" for all x,y E G. So, if x E X, Y E Y, then [x-l,y], [X,y-l] and The lower central series of a group G is the descending series
[x-l,y-l] all lie in N. Furthermore, it follows from Lemma 2.34 (i) that
[x,z], [V, z] EN=? [XV, z] E N and [x,V], [x, z] EN=? [x,yz] E N. It
then follows by a straightforward induction argument on word length that
[h,k] EN for all h E (X) = H and all k E (Y) = K, and so [H,K] :5 N, in which each 'Yi+l(G) := [G,'}'i(G)]. It is unfortunate that the lower central
which proves the result. I series begins with G = /'1 (G) while the derived series begins with G = G[O]
but this is standard notation, so we are stuck with it. The series terminates at
In particular, [G,G], the derived subgroup or commutator subgroup of G, /'r(G), and so has length r -1, if /'r{G) = [G,/'r{G)]. Since /'i-l{G)/'Yi{G) :5
is characteristic in G and is the smallest normal subgroup of G with abelian Z{Gh;{G)) for all i > 1, it is actually a central series for G if and only if it
quotient group. terminates with Gr = 1 for some r. It is not hard to show that this is the
base if and only if G is nilpotent of class r - 1.
The upper centralseries of a group G is the ascending series
DEFINITION 2.36 A nontrivial group is called perfect if [G, G] = G.
{1} = Zo(G) ~ Zl(G) = Z(G) ~ Z2(G) ~ ... ~ Zi(G) ~ Zi+l(G) ~ ... ,
So, for example, nonabelian simple groups are perfect. The derived series
of G is the descending series where each Zi+l{G) for i ::::: 1 is defined by Zi+l(G)jZi{G) = Z(GjZi(G)).
lfhe series terminates at Zr(G), and so has length r, if GjZr{G) has trivial
a = G[O] l>- a[l] l>- ... l>- a[i-l] l>- ali] l>- ... , a
centre. It is a centralseries for if and only if it terminates with Zr (G) = a
30 Handbook of Computational Group Theory Background Material 31

for some r. It is not hard to show that this is the case if and only if G is straightforward to show that a direct product of nilpotent groups is nilpotent,
nilpotent of class r . so (ill) implies (i).
It is straightforward to show by induction that, if (i) implies (iv) is also proved by induction on IGI. For H < G, either
Z(G) ::; H and we apply induction to G/Z(G), or H < HZ(G) ::; Na(H).
This completes the proof. I
is any central series of G, then G« ~ 'Yi+l(G) and Gi ::; Zr-i(G) for 0::; i ::; r.
This accounts for the names lower and upper central series. 2.3.5 The Bode of a finite group
There are a number of characterizations of finite nilpotent groups, which The socle of a finite group plays an important role in the more advanced
we shall list shortly. First we prove: .structural computations in finite groups that will be discussed in Chapter 10,
and we present its basic properties here. These are well-known, but it was
LEMMA 2.37 Let Hand K be normal subgroups of a group G with not easy to find a convenient reference containing all of the properties that
HnK = 1. Then (H,K)!?:! H x K. we require, so we include proofs here.

PROOF Using Lemma 2.34 (iv), we have [H, K] ::; H n K = 1, so the DEFINITION 2.39 The soele soc(G) of a group G is the subgroup
map (h, k) -7 hk defines an isomorphism from H x K to H K = (H, K). I generated by all minimal normal subgroups of G.

It is customary, in the situation described in the lemma, to say that (H, K)

is equal to (rather than is isomorphic to) the direct product ofits subgroups H PROPOSITION 2.40 Let G be a finite group. Then soc(G) is the direct
and K, and we shall follow that custom. Those who like to avoid ambiguities product M 1 x ... x Mm of some minimal normal subgroups M; of G, where
prefer to call (H, K) the internal direct product of Hand K. each M; is itself a direct product of isomorphic simple groups. Furthermore,
the nonabelian M, are the only nonabelian minimal normal subgroups of G.

THEOREM 2.38 Let G be a finite group. Then the following are equiv-
alent: PROOF Let N :::) G be maximal subject to being a direct product of
minimal normal subgroups of G. If N =I- soc(G), then there is a minimal
(i) G is nilpotent; normal subgroup M of G not contained in N. Then minimality of M gives
M n N = 1 and hence, by Lemma 2.37, (M, N) = M x N, which contradicts
(ii) I Sylp(G)1 = 1 for all primes p dividing IGI; the maximality of N. So N = soc(G), which proves the first assertion, and
the fact that the M; are direct products of isomorphic simple groups follows
(iii) G is the direct product of its Sylow p-subgroups;
from Theorem 2.31.
(iv) H < Na(H) for all proper subgroups H < G of G; Let M be any minimal normal nonabelian subgroup of G. If M is not
equal to any of the M, then, by Lemma 2.37 again, [M, M i ] = 1 for all i, so
(v) All maximal subgroups of G are normal in G. [M,soc(G)] = 1. Since M is nonabelian, this contradicts M ::; soc(G). I

PROOF By Sylow's theorems, (ii) is equivalent to all Sylow p-subgroups COROLLARY 2.41 If G, M, are as in the proposition, and Mk is non-
being normal in G and, under this assumption, (iii) can be proved by repeated abelian for some k, then the simple direct factors of Mk are permuted transi-
application of Lemma 2.37. Clearly (iii) implies (ii), so (ii) and (iii) are ~ively under the action of G by conjugation on soc(G).
equivalent. Clearly (iv) implies (v). If I Sylp(G) I > 1 for some p, and H
is a maximal subgroup of G that contains Na(P) with P E Sylp(G) , then
Proposition 2.20 yields H = Na(H). Hence (v) implies (ii). PROOF By the proposition applied to M k in place of G, the simple direct
By Exercise 1 of Subsection 2.2.3, a nontrivial finite group P of prime factors are the only minimal normal subgroups of Mk, so they are permuted
power order has nontrivial centre. It follows easily by induction on IFI that under conjugation by G. If S is one of these factors, then (Sa) = Mk by
the upper central series of Preaches P, and hence that P is nilpotent. It is minimality of Mk , so the action on the simple factors is transitive. I
32 Handbook of Computational Group Theory "''B,rLck,qro'una Material 33

COROLLARY 2.42 The simple direct factors of the nonabelian mini- Exercises
mal normal subgroups of a finite group G are precisely the nonabelian simple
subnormal subgroups of G. f" Prove that subgroups, quotient groups, and d~ect products of sol.vable,
;, nilpotent, or polycyclic groups are solvable, nilpotent, or polycyclic, re-
h'
.' "spectively.
PROOF Clearly each simple factor of a minimal normal subgroup is
subnormal in G, so suppose that S is a nonabelian simple subnormal subgroup 2.' Prove that if N::::! G and N and GIN are both solvable, then G is solvable.
of G. If S ::::! G then the result is clear. Otherwise G has a proper normal Does this remain true with 'nilpotent' in place of 'solvable'?
subgroup N containing S and, by induction, we may assume that S is a
Prove that if N ::; Z(G) and GIN is nilpotent then G is nilpotent.
direct factor of a minimal normal subgroup of N. Now soc(N) char N implies
that soc(N) ::::! G, so the conjugates of Sunder G are all minimal normal Use Lemma 2.34 (v) to prove the three subgroups lemma: if H, K, L ::;
subgroups of soc(N). Hence (SG) is the direct product of these conjugates G with [[H,K],L] = [[K,L],H] = 1 then [[L,H],K] = 1. Deduce that
and is a minimal normal subgroup of G, which proves the result. I }',lYi(G),'Yi(G)] ::;'Yi+j(G) for all i,j ~ 1.
.' that the final assertion in Proposition 2.40 may be false for abelian
2.3.6 The Frattini subgroup of a group direct factors M, of soc( G).
6:.' Use the Frattini lemma (Proposition 2.21) together with Proposition 2.44
DEFINITION 2.43 The Frattini subgroup if>(G) of a group G is the . .) to prove that, if G is a finite group and P E Sylp(if>(G)), then P ::::! G. In
intersection of its maximal subgroups. particular, the Frattini subgroup of a finite group is nilpotent.
Let G be a finite p-group, and suppose that Glif>(G) has order pd. Prove
Clearly if>(G) is a characteristic subgroup of G. Its principal property is that that IXI = d for any irredundant generating set X of G. (Hint: Since X
it consists of the so-called nongenerators of G. We shall only be interested
generates Glif>(G) modulo if>(G), we have IXI ~ d. If IXI > d, then some
in this result for finite groups; see, for example, Section 10.4 of [Ha159] for a element x E X is redundant as a generator of G I if> (G), and so we can
more general treatment.
write x = WU, where W is a word over X \ {x} and U E if> (G). Now use
Proposition 2.44 to show that G = (X \ {x}).)
PROPOSITION 2.44 If X is a subset of a finite group G with G =
(X, if> (G)), then G = (X).

PROOF Otherwise (X) is contained in a maximal subgroup M of G, but

then if>(G) :::; M and so G = (X, if> (G)) :::; M, contradiction. I Presentations of groups
Weinclude here a rapid treatment of the basic theory of free groups and group
We can describe if>(G) very precisely in the case when G is a finite p-group. presentations. The reader could consult [Joh98] for a more leisurely account
This topic will arise in Section 9.4 in connection with the p-quotient algorithm. of the same material.
Let G be a finite p-group for some prime p. By Theorem 2.38, all maximal
subgroups M of G are normal in G and so they must have index p in G. So,
for all x,y E G, we have xP , [x,y] E M, and hence xP , [x,y] E if>(G). In other Free groups
words, G I if> (G) is elementary abelian.
~. free group on a set X is roughly speaking the largest possible group that is
On the other hand, if N ::::! G with GIN elementary abelian, then the inter- generated by X. This idea is made precise by the universal property described
section of the maximal subgroups of GIN is trivial, so if> (G) ::; N. Hence we in the following definition.
have:

6DEFINITION 2.46 A group F is free on the subset X of F if, for any

PROPOSITION 2.45 If G is a finite p-group, then if>(G) is equal to the group G and any map e : X -+ G, there is a unique group homomorphism
smallest normal subgroup N of G such that GIN is elementary abelian. 0' : F -+ G with B'(x) = B(x) for all x E X.
34 Handbook of Computational Group Theory Background Material 35

It follows directly from the definition that any group G that is generated . ".Let us fix our set X and set A = Ax. Call two words v,w over A directly
by a set X is an epimorphic image of, and hence isomorphic to a quotient of, ~~uivalent if one can be obtained from the other by insertion or deletion of
a free group on X. (We shall see shortly that a free group on X exists.) a;csubword xx- l with x E A. For example, if X = {x,y}, then xxx-ly and
Uniqueness (up to isomorphism) of a free group on a set X is straightforward iliY are directly equivalent, as are xy and y-lyxy. Let >- be the equivalence
to prove, as is the fact that it depends only on IXI. ~elation on A* generated by direct equivalence. That is, for v,w E A*, v"" w
itand only if there is a sequence v = vo, VI, . . • ,Vr = W of elements of A *,
~th r 2: 0, such that Vi and Vi+! are directly equivalent for 0 :::; i < r.
PROPOSITION 2.47 . ltWe denote the equivalence class of wE A* under r - by [w]. Let Fx be the
(i) Two free groups on the same set X are isomorphic. ~et of equivalence classes of r-«, It is clear that Ul "" VI and U2 "" Vz imply
(ii) Free groups on Xl and X 2 are isomorphic if and only if IXll = IX21. .. z '" VI Vz so we can define a well-defined multiplication on Fx by setting
][uz] = [Ul U2] and, since the operation of concatenation of words is clearly
PROOF (i) follows from the 'if' part of (ii), so we just need to prove (ii). dative, this multiplication is associative. Furthermore, [e] is an identity
Suppose that Fl,F2 are free on X l,X2 with IXll = IX2J, let ii : Xi -t Fi ent and [XlXZ'" x r ] multiplied by [iIi;:-l ... X2lX 1l] is equal to [e], so Fx
be the inclusion maps, for i = 1,2, and let 7 : Xl -t X 2 be a bijection. By a group under this multiplication.
the definition applied to the maps i27 : Xl -t F 2 and i 1 7 - l : X2 -t F l , we
find group homomorphisms 81 : Fl -t F2 and 82 : F2 -t F l that restrict to 'HEOREM 2.49 For any set X, Fx is a free group on the set [X] :=
the i27 and i17- l on Xl and X 2, respectively. So (h8 l : Fl -t Fl restricts ~] I x E X}, and the map x t-+ [x] defines a bijection from X to [X].
to the identity on Xl, and by the uniquenesspart of the definition, 8281 is
the identity on Ft. Similarly 8182 is the identity on F2 , so 81 , (}2 are mutually
inverse and hence are isomorphisms. ROOF Let F = Fx and A = Ax. Let G be a group and let <p
Conversely suppose that Ft,F2 are free on X l,X2, and Ft and F 2 are -t G be a map. We extend <p to a map <p' : A * -t G as follows. Let
isomorphic. Let G be a group of order 2. Then there are 21 x i l distinct maps i= XpX~2. ··x~" E A*, where each Xi E X and €i = ±1, and let gi = <p(Xi)
from Xi to G for i = 1,2, and hence by the definition there are also 21x ;J 1:::; i:::; r, Then we define <p'(w) = gPg~2 ... g~" and of course <p'(e) = 1Q.
homomorphisms from Fi to G. So, for finite Xl and X 2, Fl isomorphic to F2 ce v '" w implies ip' (v) = <p' (w), <pI induces a well-definedmap 8' : Fx -+ G,
implies that IXll = IXzl. The result is also true for infinite generating sets . ich is clearly a homomorphism.
because, for infinite X, the cardinality of a free group on X is equal to that . ~IT'0 show that Fx is free on [X], for a given map () : [X] -+ G, we define
of X, but we shall only be interested in finite X in this book, so we omit the : X -t G by setting <p(x) = 8([x]), and then the map 8' defined above is a
details.' I momorphism extending 8, and is clearly the unique such homomorphism.
;To show that the map z t-+ [x] defines a bijection from X to [X], let G be
The cardinality IXI of X is known as the rank of a free group F on X. y group with IGI 2: lXI, and let <p : X -+ G be an injection. Then, since
To prove existence, we need to construct a free group on a set X. For any ([xl) = <p(x) for x E X, x t-+ [x] must define a bijection, as claimed. I
set X, we define the set X-I to be equal to {(x; -1) I x EX}, but we shall
write X-I rather than (x, -1). For y = X-I E X-I, we define y-l to be equal It is customary to write w rather than [w] for elements of F := Fx, and
to x. We define Ax (or just A when X is understood) to be X U X-I. e shall do that from now on. This renders the equality of words ambiguous.
resolve this, we shall always write WI =F W2 to mean that [WI] = [wz],
ereas WI = Wz will mean that WI and W2 are equal as words. The second
DEFINITION 2.48 For a set A, we define A* to be the set of all strings of Theorem 2.49 enables us to identify X with [X] in Fx, which we
XlX2 ... Xr with each Xi E A. The number r is the length of the string, and for I do, and we shall also follow the custom of referring to Fx as the free
a E A", we denote the length of a by lal. The empty string over A of length up on X. Since Fx is generated by X, it follows from Proposition 2.47
ois denoted by eA, or just bye. d Theorem 2.49 that any free group on X is generated by X. This can also
proved directly from the definition of a free group; see Exercise 1, below.
Elements of A * are also known as words in A * or words over A. A subword word in A* is called reduced, if it does not contain xx- l as a substring
or substring of w = Xl ... Xr E A* is the empty word, or any word of the form any x E A. We have not used quite the same definition of free groups as
XiXi+! ... xj, with 1 :::; i :::; j :::; r, It is called a prefix of w if it is empty or t given in Chapter 1 of [Joh98], for example, which defines the elements of
i = 1, and it is called a suffix of w if it is empty or j = r. ..to be reduced words over A rather than equivalence classes of words. We
36 Handbook of Computational Group Theory l1ackground Material 37

chose our definition because, as we shall see later in Chapter 12, it enables us sentation itself, which can be thought of as just a set or sequence X together
to regard group presentations as a special case of monoid presentations. The :With a set R of words in A", and the group G that is defined by the pre-
following proposition shows that the two definitions are essentially the same. sentation. If we wish to emphasize this distinction, then we shall denote the
presentation by {X I R}, and the group defined by it by (X I R).
PROPOSITION 2.50 If X is any set, then each equivalence class in F x
contains exactly one reduced word. THEOREM 2.52 =
Let X be a set, let A Ax, let R be a subset of A", and
lita be a group. For any map 0 : X -+ G, we extend fJ to a map 0 : A -+ G
~yputting fJ(x- l ) = O(X)-l for all X E X. Suppose that fJ: X -+ G is a map
PROOF Let "'x be defined as above. Since xx- l <x E: for all x E
'oj with the property that, for all W = Xl ... X r E R, we have O(Xl) ... O(X r ) = 1G.
A = Ax, it is clear that each equivalence class contains a reduced word.
Then there exists a unique group homomorphism 0' : (X I R) -+ G for which
Suppose that a class contains two distinct reduced words, U and v. Then
there is a sequence U = Uo, Ul, ... ,Ur = v of elements of A * such that Ui
:?t~.t:(xN) =O(x) for all X E X, where (X I R) =
FIN as in the definition
and Vi are directly equivalent for 0 ::; i < r, Choose this sequence to make . ,~1Jdve.
the sum of the IUil as small as possible. It is clear that directly equivalent :' Conversely, if there is a group homomorphism 01 : (X I R) -+ G for which
elements of A * differ in length by two, and they cannot both be reduced ~~;(:xN) = O(x) for all z E X, then we must have O(Xl)'" O(x r ) = 1G for all
words, so we must have r > 1. Choose i such that IUil is maximal. Then fu= Xl"
·Ci
'X r E R.
0< i < 1', IUi-ll = IUi+l1 = IUil- 2, and both Ui-l and Ui+l are derived from
Ui by deleting a substring of the form xx- l. If these two substrings of Ui are lIPROOF Since (X I R) is generated by the elements xN for X E X, the
disjoint, then we can reverse the order of the substitutions and obtain another li'Omomorphism 0' is certainly unique if it exists. By the definition of the free
sequence with IUil = lUi-II - 2, contrary to the minimality condition on the group F = Fx , there is a homomorphism Oil : F -+ G with B"(x) = B(x) for
sequence. On the other hand, if the two substrings xx- l are not disjoint, then .llilhx E X. The hypothesis of the theorem then says that W E ker(O") for all
either they are equal, or they are the substrings xx- l, X-IX of a substring in E R, and hence N = (R F ) ::; ker(O"), and so Oil induces a homomorphism
XX-lX of Ui. In both of these cases, we have Ui-l = Ui+l, so we can shorten Q,L:(X I R) = FIN -+ G with the required property.
the sequence and again contradict its minimality condition. I tt<:.Conversely, if 01 exists as described, then we must have 0'(wN) = Oil (w) for
ll)ILw E F, and so OI(W) = 1G for all wEN. In particular,
If W = XIX2 ... X r E A* then, for any i with 1 ::; i ::; 1', W is conjugate in Fx
to the cyclic permutation Wi := XiXi+1 ... X rXIX2 ... Xi-l of w. The words w 8(XI) ... 8(x r ) = 0' (Xl) ···0' (x r ) = 8' (Xl'" xr ) = 1G
and Wi are said to be cyclic conjugates of one another. The word W is said to
be cyclically reduced if it is reduced and Xl t= X r . This is equivalent to all of I
its cyclic conjugates being reduced words. Clearly every W is conjugate in F
to a cyclically reduced word. ;We shall further abuse notation by using v to represent the image vN of v
AX I R) for v E A*. We shall use U =G v to mean that u,v E A* represent
e same element of G.
2.4.2 Group presentations
In general, if G is a group generated by a set X, A = Ax, and w E A * with
=G 1G, then we call the word W a relator of G. The elements of R in a group
DEFINITION 2.51 Let X, A = Ax be defined as above, and let R be §sentation G = (X I R) are clearly relators of G, and these are known as
a subset of A *. We define the group presentation (X I R) to be equal to the ing relators. If WI, W2 E G with Wl =G W2, then we call Wl =G W2 a
quotient group FIN, where F = Fx is the free group on X, and N is the Iltion in G. Of course, this is equivalent to wlwi l being a relator of G. It
normal closure (R F ) of R in F. ften convenient to write down group presentations in the alternative form
(X I R), where R is a subset of A* x A*, but the element (WI, W2) of R
Informally, (X I R) is the 'largest' group G that is generated by X in whic~' sually written as WI = W2. We formally define this group to be equal to
all of the strings v E R represent the identity element. This idea is expressed I R), where the elements of R are obtained from those of R by replacing
l
more precisely by the universal property proved next. This property could be ':::: W2 by Wl wi • The elements of R are called defining relations of G.
used as an alternative definition of a group presentation. ior example, the group (x, y I x 2, y3, (xy)2), which is dihedral of order 6,
Notice that this definition does not distinguish properly between the pre- ld be written as (x, y I x 2 = 1, y2 = y-l, xyx = y-l ).
38 Handbook of Computational Group Theory Background Material 39

If the hypotheses of Theorem 2.52 are satisfied and the resulting map B' : In particular, for each 'f E R there is a corresponding word r, and then
(X I R) --+ G is an isomorphism, then we often, in practice, identify the set f== 1G/N implies that r E N, so in the group G we have r =G W n for some
X with its image in Gunder B, and say that (X I R) is a presentation of G on Word W r E (Y U y-l)*. Let R be the set {rw;l I 'f E R}.
the generators X of G. For example, if G = ((1,2), (1,2,3)) = Sym(3), then ;" For each y E Y and x E X, we have x-Iyx E N, so x-Iyx =G wx y for some
we might say that (x, y I x 2 = 1, y2 = y-l, xyx = y-l) is a presentation of ~ord Wxy E (Y U y-l)*. Let T be the set {x-Iyxw;i I x E X, Y E Y}.
G on the generators x = (1,2), y = (1,2,3) of G.
Using that slight abuse of notation, we can formulate the following version PROPOSITION 2.55 With the above notation, (XUY I Ru5uT) is
of Theorem 2.52, which is computationally useful for testing whether a map a presentation of G.
sp from a group G to a group H is a homomorphism.

,PROOF Let F be the group defined by the above presentation. To avoid

THEOREM 2.53 Let (X I R) be a presentation of a group G, and let gonfusion, let us denote the generators of F mapping onto X E X or y E Y by x
B: X --+ H be a map from X to a group H. Extend B to B : X-I --+ H by
and fj, respectively. We have chosen the sets R, 5, T to be words that evaluate
putting B(x- l) = B(x)-l for all x E X.
to the identity in G, so Theorem 2.52 tells us that the mapping x --+ x, fj --+ Y
Then B extends to a homomorphism 8' : G --+ H if and only B(Xl) ... 8 (x r ) = ";~,n"N:'Q a homomorphism 8 ; F --+ G. In fact 8 is an epimorphism, because
1H for all w = Xl ... Xr E R. This extension is unique if it exists.
""l,p"rlv G is generated by Xu Y.
Let K be the subgroup (fj lyE Y) of F. Then 8(K) == N. Since,
PROOF The uniqueness of the extension follows from the fact that X assumption, (Y I 5) is a presentation of N, and the relators of 5 are
generates G. The statement that (X I R) is a presentation of a group G relators of K, Theorem 2.52 implies that the map y --+ fj induces a
really means that there is an isomorphism sp : (X I R) --+ G, and that we Iibinomorphism N --+ K. But this homomorphism is then an inverse to BK,
are identifying (X I R) with its image under sp, Similarly, the statement so 8K : K --+ N is an isomorphism.
8(xd" ·8(xr ) = 1H for all w = Xl' "X r E R means 8(<P(Xl)) .. ·B(<p(xr ) ) = ,¥/Wenow wish to assert that the relators in T imply that K:::1 F. This is not
1H for all for all w = Xl ... Xr E R. By Theorem 2.52, this is true if and quite as clear as it may at first sight appear! For it to be clear, we would need
only if B<p extends to a homomorphism 'ljJ : (X I R) --+ H. Since ip is an t'bkn6w that xfjx- l E K for all x E X, Y E Y, in addition to x-Ifjx E K. But
isomorphism, the existence of 'ljJ is equivalent to the existence of the required t!ltefact that N:::!G tells us that each x E X induces an automorphism of N by
homomorphism 'ljJ<p-1 ; G --+ H. I ,~oIljugation, which implies that, for each x EX, we have N = (w x y lyE Y).
hce N is isomorphic to K via 8K, the corresponding statement is true in K
'each x. So the fact that X-I conjugates each word w x y into K implies the
2.4.3 Presentations of group extensions
'ired property xfjx- l E K for all x E X,y E Y.
So we do indeed have K:::1 F. Now, by a similar argument to the one above
DEFINITION 2.54 A group G with a normal subgroup N is called a r BK , the fact that (X I R) is a presentation of GIN implies that the induced
(group) extension of N by GIN. It is called a split extension if there is a momorphism 8F / K : FIK --+ GIN is an isomorphism. So, if 9 E ker(8),
subgroup C of G with NC = G and N n C = 1. In that case, C is called a 'Il gK E ker(8F / K ) implies that 9 E K, but then 9 E ker(8K ) = 1, so 8 is
complement of N in GIN. lisomorphism, which proves the result. I
Suppose that the group G has a normal subgroup N, and that we have the special case of a split extension, we can choose our inverse images x
presentations (Y I 5) of Nand (X I R) of GIN on generating sets Y and 'to lie in a complement of N in G, in which case we get W r = 1 for all
X, respectively. Here we shall describe a general recipe for constructing 'R. In the even more special case of trivial action of the complement on
presentation of G as an extension of N by GIN. ,', e have a direct product, and we get the following well-known result as a
For each x E X, choose x E G with xN = x, and let <:Jllary.

X ;= {x I x E X}.
;,~OLLARY 2.56 Let (X I R) and (Y IS) be presentations of groups
- --1
Then, for any word wE (X U X )*, we can define w E (X U X-l)* with H. Then (X U Y I R U 5 U T) is a presentation of G x H, where
=
wN w, by substituting x or X-I for each x or X-I occurring in w. ,[x,y] I X E X,y E Y}.
40 Handbook of Computational Group Theory Background Material 41

2.4.4 Tietze transformations in R then we can apply common substring replacements to replace them by
There are a number of simple manipulations of a group presentation that do
~-l ~r v, thereby eliminating them. We can then apply TT4 and eliminate x
from X and w from R. We shall refer to this combination of transformations
not change the group or at least the isomorphism class of the group defined
as eliminating a generator. One practical problem with doing this is that
by the presentation. These are known as Tietze transformations of the pre-
if v is long, then the common substring replacements may render the total
sentation. There are four of these, which come in mutually-inverse pairs. As
relator length of the presentation considerably greater than it was before the
before, we denote X UX-I by A.
~limination.
TTl Adjoin an extra relator in (X I R) to R;
TT2 Remove a redundant relator w from R where (R F ) = ((R \ {w} t); Exercise
1. Prove from the definition of a free group F on a set X that F is generated
TT3 For any word wE A*, adjoin an extra element z, not already in X, to
X, and adjoin the element wx- 1 to R. byX.

TT4 If there is a generator z E X and a relator w E R such that there is

- exactly one occurrence of z or x-I in w, and neither x nor x-I occurs
in any v E R \ {w}, then remove z from X and w from R.

It is almost obvious that transformations of type (i) or (ii) do not change 2.5 Presentations of subgroups
the group defined by the presentation, and it is not hard to prove that (iii) and Let G be a group generated by a set X, and let A = XUX-l. Then G ~ FIN
(iv) replace this group by an isomorphic group. We leave this as an exercise for some N, where F = Fx is the free group on X, and we shall assume that
for the reader, who could alternatively refer to [Joh98]. G = FIN. Let H = EIN be a subgroup of G, and let T be a set of reduced
There are some combinations of these transformations that are used very words in A* that form a right transversal of E in F. We shall always assume
frequently, and we shall mention a few of these here. that T contains the empty word E as the representative of the coset E. Hence
We can always replace a relator in a presentation by its inverse or by a the images of T in G form a right transversal of H in G. For a word w E A *,
conjugate of itself. This is immediately clear from the definition of a group we denote the unique element of T n Ew by w.
presentation, but it could also be regarded as an application of TTl followed
by TT2. We shall feel free to carry out such replacements without com-
ment. In particular, we often need to replace a relator XIX2 ... Xr by a cyclic 2.5.1 Subgroup presentations on Schreier generators
conjugate. This facility also enables us to assume that defining relators are Let w = XIX2 ... »; E A* and, for 0::; i ::; r, let ti := XIX2 ... Xi, where to = c.
cyclically reduced words. Then we have
A slightly less trivial instance is when we have two relators uv, uw E R,
where u, v, w E A* and u =f:. c. The common substring u is called a piece. If
G = FIN as before, then uv,uw E N implies v-1w E N, and so we can apply Y of E by
Define the subset
TTl to adjoin v-1w to R. But uv, v-1w EN implies uw E N, so we can apply
TT2 to remove uw from R. The combined effect is to replace uui by v-1w. Y:= {txtx- 1 It E T,x E X,tx =f:.F tx- 1}.
Let us call this combination a common substring replacement. In practical
Note that the final condition tx :f.F tx -1 simply has the effect of omitting the
applications, we might want to do this if Iv-1wl < luwl, because this shortens
iaentity element from Y. If x E X, t E T, and u = tx- 1, then t = ux, so
the presentation. Since we can always replace relators by cyclic conjugates, we
~k'"'-lu-l = (uxt- 1)-1 = (uxux-1)-l, which is an element of y-l (unless it
can apply common substring replacement whenever two elements of R have a
gquals IE). Since ti = ti-lXi for 1 ::; i ::; r, we see that each of the bracketed
common substring u, regardless of whether or not u is a prefix of ther~lij.tQ~~/.
~~\'ms
cr., tiXi+lti=t\ in the equation (*) is either equal to IE, or is an element of
TT4 is more useful when applied in combination with common substHnif'
replacement. Suppose that we have x E X and w E R such that there is r
.'}':'
A

n y-l. If w represents an element of H, then t r = W = e, and so we have:

exactly one occurrence of z or X-I in w. By replacing r by its inverse and a
=
cyclic conjugate if necessary, we can assume that w xv, where v does not .~(rHEOREM 2.57 (Schreier) The subgroup E of F is generated by Y.
contain x or X-I. If there are occurrences of x or x- 1 in any other relators &imilarly the subgroup H of G is generated by the images of in G. Y
42 Handbook of Computational Group Theory Background Material 43

The set Y is called the set of Schreier generators of E with respect to the DEFINITION 2.59 Let < be a given well-ordering of a set A. Then,
group F. the associated lexicographical (dictionary) ordering <L of A", is defined by
We now let Y be a set of new symbols corresponding to Y, where we denote Il<la2'" am < bsbz :': bn (with ai, bj E A) if, for some k :::: 0, ai = bi for
1
the element of Y corresponding to txtx- E Y by Ytx. To ease the notation, ~i ~ i ~ k, and either k = m < n or ak+! < bk+!'
we will let Ytx denote the empty string e when tx =F tx. The map t.p : Y -+ F
The lexicographical ordering satisfies the u < v :::} uw < vw property, but
defined by cp(Ytx) := txtx -1 extends to a homomorphism from the free group
it is not a well-ordering for IAI > 1 because, if a, b E A with a > b, then
F y on Y to F, and the theorem above says that im(cp) = E. For t E T,
---1 aib >L ai+1b for all i :::: O. The shortlex ordering, however, which we shall
X E X, we denote the element of y-1 that t.p maps to tx- 1tx-1 by Ytx-l. now define, is a well-ordering.
We define a map p: A* -+ Fy by
'bEFINITION 2.60 Let < be a given well-ordering of a set A. Then, the
associated shortlex (also called lenlex) ordering <8 of A * is defined by u <8 V
if either lui < lvi, or if lui = Ivl and u <L V.
where w = XlX2 ... x r , as in the equation (*) above. Of course, p and t.p
depend on the choice of the transversal T. It is straightforward to check that , If T is a Schreier transversal and w = t E T, then all of the bracketed terms
p(uxx- 1v) =Fy p(uv) for any u, v E A*, x E A, and so p induces a well- in'the equation (*) are equal to IE, and so p(t) = c. H w = tx with t E T,
defined map, also denoted by p, from F to Fy. This map is called a rewriting z E X and tx f.F tx, then all of the bracketed terms are trivial except for the
map. last, and so p(tx) = Ytx and, using Lemma 2.58, we have
More generally, for t E T and w = X1X2 ... Xr E A*, we define p(t, w) in the
same way as for p(w), but with to = t; that is,

!' So the composite pt.p is equal to the identity homomorphism on Fy. But

this implies that t.p is a monomorphism, so we have proved the well-known

Nielsen-Schreier theorem:
where t; = tX1X2'" Xi for 0 ::::; i ::::; r; so p(W) = p(c, w).
'THIW.Rl~M 2.61 1fT is Schreier transversal, then the subgroup E of F
LEMMA 2.58 (i) p(uv) = p(u)p(u, v) for all u, v E A*. a free group on the generating set Y.
(ii) p(U)-l =Fy p(u,u- 1) for all u E A*.
Now suppose that G = (X I R) is a presentation of G, and G = FIN with
PROOF (i) is clear from writing down the definition of p(uv). For (ii),
N = (RF ) . The following theorem will be used later, in Subsection 5.3.1, in
an algorithm to compute a presentation of the subgroup H of G.
note that p(uu- 1 ) =Fy IFy and use (i). I
It follows from (i) of this lemma that, if u E E, then u =c and p(uv) = THEOREM 2.62 Assume that T is a Schreier transversal of E in F,
p(u)p(v), so p restricts to a homomorphism from E to Fy. and define S:= {p(twt- 1 ) I t E T,w E R}. Then the group H' := (Y I S)
is isomorphic to H, and an isomorphism is induced by the map cp : Fy -+ E
The transversal T is called a Schreier transversal if it is prefix-closed; that is,
defined above.
if all prefixes of all elements of T lie in T. To prove that Schreier transversals
exist, let < be a well-ordering of A * with the property that u < v :::} uw < vw
for all u, v, w E A *, and choose the elements of T to be the least elements ltPROOF We saw above that ip and piE are inverse isomorphisms. Since
of their cosets under this ordering (see the exercises below). Examples of each twr 1 with t E T, w E R lies in N ::::; E, we have t.pp(twt-1) =F twr 1 E
orderings of A * with this property are shortlex orderings, which are defined N, and so by Theorem 2.52 t.p induces an epimorphism Ij5 from H' to H.
below. They are defined with respect to a given well-ordering of A and so, i'Now N = (R F ) is generated by conjugates uwu- 1 of elements w E R by
if we assume the axiom of choice, then they exist for all sets A; but in this ements u E A*. For any such u, we have u =F vt, with vEE and t E T,
book, we are only concerned with finite sets A. and then uwu- 1 =F v(twt- 1)v- 1 and so p(uwu- 1) =Fy p(v)p(twt- 1)p(vt 1
44 Handbook of Computational Group Theory Background Material 45

is a conjugate of an element of 8. It follows that p(z) E (8 FY ) for all zEN, THEOREM 2.63 With the notation just defined, let 8 1 := {p(twt- 1) I
and hence piE induces a homomorphism p : H = E / N -t HI. t E T,w E R}, and let 8 2 := {pcp(y)y- 1 lYE Y}. Then the group H I :=
We saw above that pcp is the identity homomorphism on Fv , and hence pcp =
(Y I 8) with 8 8 1 U 8 2 is isomorphic to H, and an isomorphism is induced
is the identity on HI. So Tj5 is an isomorphism, which proves the result. I by the map ip : Fy -t E.

As a final observation on the above theorem, note that by Lemma 2.58 we ,PROOF Let w = X1X2'" z; E A* and pew) = YtOX 1YtlX2 ... Yt~_lX~ with
can express the elements p(twr 1 ) of 8 as the ti defined as in the equation (*). Since ti = ti-1 Xi for 1 S; i S; r, our
) IHlsumption tx =G cp(Yt",)tx tells us that CP(Yti_1X;) =G ti_1xit;1 for 1 S; i S; r,
p(twc 1) = p(t)p(t, w)p(tw, C 1) = p(t)p(t, w)p(t, C 1) = p(t)p(t, w)p(t)-l. and so we have
We saw earlier that the fact that T is a Schreier transversal implies that
pet) = 5, so in fact we have
In particular, if WEE, then cpp(w) =G w. From this equation, it follows that
8 = {p(t,w) It E T,w E R}.
cp(s) E N for all s E 8 1 U 8 2 and so ip induces a epimorphism Tj5: HI -t H.
This form of 8 is more convenient for use in the algorithm in Subsection 5.3.1 It follows exactly as in Theorem 2.62 that piE induces a homomorphisrri
for computing subgroup presentations. p ; H = E / N -t HI. The inclusion of 82 in the set of relators of HI ensures
that p(cp(y)) = H' Y for all Y E Y, and so pcp is the identity on HI, and hence
cp is an isomorphism, which proves the result. I
2.5.2 Subgroup presentations on a general generating set
We shall now describe a generalization of the results in Subsection 2.5.1, which As in the final paragraph of Subsection 2.5.1, the element p(twt- 1) of 8 1 is
can be used to find a presentation of a subgroup on a given generating set, equal to p(t)p(t, w)p(t)-l in Fv . It is no longer necessarily true that pet) = 5,
rather than on the set of Schreier generators. This subsection could be omitted but since p(t)p(t,w)p(t)-l is a conjugate of p(t,w), the theorem still remains
on a first reading. true if we replace 8 1 by
As before, we assume that G = <X I R) and G = F / N with F = Fx
and N = (RF), and that H = E/N is a subgroup of G. Let T be any right {p(t,w) It E T,w E R}.
transversal of E in F (not necessarily a Schreier transversal) consisting of
reduced words and containing 5 and, as before, denote the unique element of
T n Ew by w for w E A ". Let Y be a set disjoint from X, and suppose that
Exercises
we have an epimorphism ip : Fy -t E, and that, for t E T and x E X, we are
given elements Ytx E F y satisfying tx =G cp(Yt",)tx for all t E T, x E X. Note
Show that the transversal T of E in F defined as the set of least elements
that these conditions are satisfied by the ep and Ytx defined in Subsection 2.5.l.
in their cosets under a well-ordering of A * satisfying U < v :::} uw < vw for
For t E T, x EX, we use Ytx-1 to denote the inverse of the element Yux of
all w E A * is indeed a Schreier transversal.
Fy that satisfies ux =G ep(Yux)t. We define the rewriting map p : A* -t Fy
exactly as before: for any w E A *, we put
Show that the shortlex ordering <8 of A * associated with a given well-
ordering < of A is a well-ordering and satisfies u < v :::} uw < vw and wu <
wv for all wE A*.
where w = XlX2 ... X r as in the equation (*) in Subsection 2.5.1.
Again the images of equivalent words under p are equal, so we can thinksof Let A = {x,x- l,y,y- l } and write down the first 20 elements of A* in the
p as mapping F to Fy, and Lemma 2.58 is true as before, so again piE i~a shortlex ordering of A* defined with respect to the ordering z < Y < u:' <
X-I of A.
homomorphism. In Subsection 2.5.1, we had PCP(Ytx) = p(txtx -1) = Ytx for
all Ytx E Y, but this is no longer necessarily the case. So, in Theorem 2.62,
the relators in the set 8 2 in the theorem below freely reduce to the empty Show that the nor~al closure of the subgroup (y) of the free group on
word, and can be omitted. {x,y} is free on {yx' liE Z}.
46 Handbook of Computational Group Theory 47

= It follows that a finitely generated free abelian group is isomorphic to zn

for some n E No, and we shall generally use zn as our prototype free abelian
2.6 Abelian group presentations group. Similarly, any finitely generated abelian group with n generators is
isomorphic to a quotient group of zn.
The theory that we have just developed for group presentations can be recast There is an alternative construction of the free abelian group on X as an
in the framework of abelian groups rather than general groups, but much of ordinary group presentation. For a set X, we define [X, X] to be the set of
the material becomes significantly easier in the abelian case. In particular, a formal commutators
free abelian group is just a direct sum of copies of an infinite cyclic group. We
shall quickly run through this basic theory here. We shall, however, delay our
treatment of the structure theorem for finitely generated abelian groups until
Section 9.2, because it is more convenient to discuss this at the same time as' This is a subset of the set A
to:{
xof words over Ax, where Ax := X U X-I.
we describe the related algorithms.
;PROPOSITION 2.66 If F is a free group on a set X, then (X I [X, X] )
DEFINITION 2.64 An abelian group F ab is free abelian on the subset isa free abelian group on X.
X of Fab if, for any abelian group G and any map 8 : X ~ G, there is a
unique group homomorphism 8' : Fab ~ G with 8'(x) = 8(x) for all z EX. 'PROOF Let 8 : X ~ G be a map, with G an abelian group. By the
The set X is known as a free basis of Fab. definition of the free group on X, 8 extends to a homomorphism 8' : F ~ X.
Since G is abelian, [X, X] and hence N := ([X, X]F) is contained in the kernel
Again it follows directly from the definition that any abelian group gener- of 8', so 8' induces a homomorphism (X I [X, X] ) = FIN ~ G, and the result
ated by a set X is an epimorphic image of, and hence isomorphic to a quotient follows. I
of, a free abelian group on X.
Exactly as in Proposition 2.47, we can show that free abelian groups on Xl ~.'We can now define an abelian group presentation.
and X2 are isomorphic if and only if IXII = IX21, and we call IXI the rank of
a free abelian group on X. nrEF'Il\rI1~IC)N 2.67 Let X be a set, let R be a subset of FJcb, and let N
We generally prefer to use additive rather than multiplicative notation when the subgroup of Fj(b generated by R. Then the abelian group presentation
discussing abelian groups, and we shall do so here. X I R) is defined to be equal to the quotient group Fj(bIN.
For an arbitrary set X, we define FJi-b to be the direct sum EB"'EXZ", of
infinite cyclic groups Z'" generated by z. As usual, we identify the groups Z'" analogous result to Theorem 2.52 holds for presentations of abelian
with the corresponding component subgroups of the direct sum, and so an but it is not used so frequently as Theorem 2.52 itself.
element 9 E FJcb has the form 2:"'EX a",x for a", E Z, where all but finitely Next we relate abelian group presentations to the abelianizations of ordinary
many of the a", are zero. (In this book, X will virtually always be finite, but 9foup presentations. For a set of words R ~ Ax, we let R+ denote the
there is no reason not to consider general X at this stage.) corresponding set of elements of Fj(b obtained by writing each element of R
Many books define the free abelian group on X to be the group FJcb, and ~dditively. So, for example, if R = {X 2yx-l y3x4, y3 x-4 x y}, then R+ =
we shall now show that this is equivalent to our definition. L5x + 4y, -3x + 4y}.
",~ ;

PROPOSITION 2.65 If X is a set, then the group FJcb is free abelian i;PROPOSITION 2.68 Let G = (X I R) be a group presentation. Then
;i?!~i
onX.
(i) G/[G,G]~(XIRU[X,X]) and

PROOF For an abelian group G and a map 8 : X ~ G, we define } G/[G,G] ~ Ab(X I R+}.
8' : FJcb ~ G by 8'(g) = 2:"'EX a",8(x), where 9 = 2:"'EX a",x with all
but finitely many of the a", equal to zero. It is easy to see that 8' is a ROOF Let F be the free group on X. Then G = FIN with N = (RF ) ,
homomorphism extending 8. I . d (X I R U [X, X]) = FjM with M = ((R U [X,X))F). Since each pair
 48 Handbook of Computational Group Theory 49
l3:ackground Material

of generators commutes in FjM, FjM is abelian, and so (FjN)j(MjN) ~ irreducible if and only if M j N is a chief factor of G:
FjM is an abelian quotient of G and hence [G,G] ::::; MjN. But M == In this book, mainly in Chapter 7, we shall attempt to cover only those parts
F,
(R [X,X]F) == NL, where L = ([X,X]F), so MjN is generated by commu- of,computational representation theory that have applications to the analysis
tators, and hence [G,G] = MIN, which proves (i). ,fthe structure of finite groups. This does not require that the reader have
By Proposition 2.66, FIL is a-free abelian group on the set X, and so ;:very deep knowledge of representation theory, but it does involve certain
FIM ~ (FjL)/(MjL). An element of R F has the form U W for u E R, ~ppics that are not always covered by the most elementary books on t~e
W E F, and U
W
=
u[u,w] with [u,w] E [F,F] C L (since FIL is abelian), s'ubject such as representations over finite fields rather than over C, so It
so MIL = NLIL is generated by the images of the elements u E R, which is/not easy to find references at the appropriate level. Two possibilities are
proves (ii). I the,books by Isaacs [Isa76] and Rotman [Rot02], which we shall use here for
references to theoretical results.
Exercise 3 in Section 2.5 shows that subgroups of finitely generated groups
are not themselves finitely generated in general. The following result shows
that this property does hold for finitely generated abelian groups, from which ~:7~1 The terminology of representation theory
r,·!·

it follows that all finitely generated abelian groups have finite abelian group get us briefly review the basic definitions and results from the representation
presentations. theory of finite groups. Let K be a commutative ring with 1, and let G be
,. . . group. The group ring KG of Gover K is defined to be the ring of
PROPOSITION 2.69 Any subgroup H of a finitely generated abelian formal sums
group G is finitely generated. (L::>gg I r gE K}
gEG

PROOF Let G be generated by Xl, ... ,Xn . We use induction on n, the ~ith the obvious addition and multiplication inherited from that of G. In fact
result being clear for n = o. Let K := (X2, ... , x n ) . By induction, K n H is KG is an associative algebra with 1 and thus it is a ring with 1 and a module
finitely generated, whereas Hj(KnH) ~ HKIK::::; GIK is cyclic, and hence bver K. It is also known as the group algebra of Gover K.
finitely generated. The result follows. I :' Let M be a right (unital) KG-module. We shall write m·x (m E M,
f.E KG) to represent the module product in M, but when x EK and we are
thinking of M primarily as a K-module, then we may write xm rather than
m·x. Since K is commutative, this does not cause any problems. From the
module axioms, and the fact that (m·g)·g-l = m for m EM, g E G, we see
that multiplication by a group element 9 E G defines an automorphism of M
2.7 Representation theory, modules, extensions, as a K-module. So we have an associated action cP : G -t AutK(M), and we
derivations, and complements shall sometimes use the group action notation m g as an alternative to m·g.
Conversely, if M is a K-module, then any action cp : G -t AutK(M) can be
This section concerns group representation theory, together with the basic used to make M into an KG-module.
theory of group extensions of M by G in the case when M is abelian.
'. We shall always assume that M is finitely generated and free as a K -module,
Computation with group representations is a significant subtopic within and so, after fixing on a free basis of M, we can identify M with Kd for
CGT. Some of the methods in this area, particularly those related to group ?9med. Then, using the same free basis of M, AutK(M) can be identified
representation theory for its own sake involve some rather advanced theory. with the group GL(d,K) of invertible d x d matrices over K. So the action
Even for computations that are concerned only with the group-theoretical homomorphism cp is cp : G -t GL(d, K), which is the standard definition of a
structure of finite groups, some of the more sophisticated algorithms r i"!:lpresentation of G of degree dover K.
some familiarity with representation theory. ' 7<11;lir,lf,
•. '",?'
According to basic results from representation theory, two KG-modules are
The basic reason for this is that if a finite group G has normal subgroups ~bmorphic if and only if the associated representations CPI, CP2 are equivalent,
N < M for which MIN is an elementary abelian p-group for some prime Which means that they have the same degree and there exists a E GL(d,K)
p, then the conjugation action of G on M gives rise to a representation of with a·CP2(g) = CPI(g)·a for all g.E G.
G1M over the field of order p, and properties of that representation translate .'JWhen K is a field, a KG-module and its associated representation is called
into group-theoretical properties of G. For example, the representation is ;simple or irreducible if it has no proper nonzero KG-submodules. (A slightly
50 Handbook of Computational Group Theory : Background Material 51

different definition of irreducibility is normally used when K = Z, for exam- hThe semidirect product is an extension of M by G, using the maps M -+,
ple.) , G KM and G I>< M -+ G defined by m t-t (IG, m) and (g, m) t-t g. It is a split
Since we are dealing only with finite-dimensional modules, any such mod- extension, with complement { (g, 1M) I 9 E G}.
ule M has a composition series, (that is, an ascending series of submodules Conversely, if the group E has a normal subgroup M with a complement
with simple factor modules), and the Jordan-Holder theorem (Thereom 8.18 G then any e E E can be written uniquely as e = gm for 9 E G, m EM, and
of [Rot02)) asserts that any two such series have isomorphic factors, counting =
gmhn ghmhn, so we have:
multiplicity, and so we can refer to the composition factors of M.
For any KG-module M, we can define the K-algebra EndKG(M) of endo- PROPOSITION 2.70 Any split extension E of M by G is isomorphic
morphisms of M (= KG-homomorphisms from M to M). This is also known to'the semidirect product G I>(rp M, where the action <p of G on M is defined
as the centralizing algebra of M and its associated representation. It contains ~y the conjugation action of a complement of M in E on M.
the scalar automorphisms, which form a subalgebra isomorphic to K. When
K is a field and M is a simple KG-module, then Schur's lemma (Theorem l>In general, different complements could give rise to different actions ip,
8.52 [Rot02] or Lemma 1.5 of [Isa76)) says that EndKG(M) is division ring. However, if M is abelian, then the actions coming from different complements
This can be noncommutative in general, but in this book we shall be par- ike the same. We shall assume for the remainder of this subsection that M is
ticularly concerned with the case when K is a finite field, in which case a abelian and use additive notation for M.
well-known theorem of Wedderburn (Theorem 8.23 of [Rot02]) tells us that We shall also assume that M is a K-module for some commutative ring
EndKG(M) is a field, and it can be regarded as an extension field of K. K with 1. This is no loss of generality, because any abelian group can be
When K is a field, the KG-module M is called absolutely irreducible if it is regarded as a Z-modulejust by defining n·m = nm for nEZ, mE M. In the
irreducible and remains irreducible when regarded as an LG-module for any case when M is an elementary abelian p-group, we can take K to be the field
extension field L of K. By Theorem 9.2 of [Isa76], M is absolutely irreducible ]E'p.
if and only if EndKG(M) consists of scalars only. As we saw in Subsection 2.7.1, an action <p : G -+ AutK(M) of G on the
So, if K is a finite field and L = EndKG(M), then we can use the action of K-module M corresponds to endowing M with the structure of a KG-module,
L on M to make M into an LG-module with dimL(M)IL: KI = dimK(M), and so we can talk about the semidirect product G I>< M = G 1><'1' M of the
and M is absolutely irreducible as an LG-module. In particular, there is a KiG·module M with G. The multiplication rule in G I>< M, using additive
finite extension L of K for which all irreducible LG-modules are absolutely ,notation in M, is (g,m)(h,n) = (gh,m h +n). A

irreducible, and such an L is called a splitting field for G. ,:\A general left transversal of the subgroup M := {(IG' m) I m EM}
isomorphic to M in G I>< M has the form T x = {(g, X(g)) I 9 E G}, for a
2.7.2 Semidirect products, complements, derivations, and first map x : G -+ M. Then T x is a complement of M in G I>< M if and only if
cohomology groups , ,X(g))(h,X(h)) = (gh,X(gh)) for all g,h E G or, equivalently,
This and the following subsection contain a very brief description of the first X(gh) = X(g)h + X(h) Vg, hE G. (t)
and second cohomology groups of groups acting on modules, insofar as they
are relevant to the (computational) study of group extensions. For a more '.~If M is a KG-module, then a map X : G -+ M is called a derivation or
complete treatment, the reader may consult Chapter 10 of [Rot02], particu- !,qrossed homomorphism or a l-cocycle if (t) holds. Notice that by putting
larly Sections 10.2 and 10.3. But, in common with the majority of published ':;= IG in (t), we see that X(IG) = OM for any derivation X.
material on this topic, the account in [Rot02] is in terms of left modules, ,We denote the set of such derivations by Zl(G,M). By using the obvious
whereas ours uses right modules, so there will be some differences, such as in intwise addition and scalar multiplication, we can make Zl (G, M) into a
the definitions of cocycles. 'module. We have proved:
We defined the notion of a (split) extension of one group by another in
Subsection 2.54. Let G and M be groups, and suppose that we are given:ai:
homomorphism <p : G -+ Aut(M). We define the semidirect product of M by ROPOSITION 2.71 If M is a KG-module, then the set T x defined
G using ip to be the set G x M endowed with the multiplication (g, m) (h, n) = ve is a complement of Min G I>< M if and only if X E Zl(G,M).
(gh, mhn), for g, hE G, m, n E M, where, as usual, m h is an abbreviation for
mrp(h). The standard notation for a semidirect product is G I>< M or G 1><'1' M. otice that for a fixed m E M, {(g,OM)(l,m) = (g,m-m Y ) I 9 E G} is
)complement of M in G I>< M, and so 9 t-t m - mY is a derivation. Such a
52 Handbook of Computational Group Theory 53

map is called a principal derivation or l-coboundary. The set of all, principal group extension E of M by G in which the given KG-module M is the same
derivations is denoted by B 1(G, M) and forms a K-submodule of ZI(G, M). as the KG-module defined as above by conjugation within E.
,;,Given E as above, the elements {g I 9 E G} form a transversal of Min G.
DEFINITION 2.72 The first cohomology groupH~(G, M) isJhequotient
K-module Zl(G, M)I B 1 (G, M). For g,h E G, we have fj1. =g'lvr(g,h), for some function 7: G x G -7 M,
where the associative law in E implies that, for all g, h, kEG,
From the discussion above, it follows that HI (G, M) is in one-one corre-
spondence with the set of conjugacy classes of complements of M in G I>< M.
7(g, hk) + 7(h, k) = 7(g, h)k + 7(gh, k).
The following result tells us that derivations are uniquely determined by '. A function 7 : G x G -7 M satisfying this identity is called a 2-cocycle,
their action on a generating set of a group. ;~nd the additive group of such functions forms a K-module and is denoted
byZ2(G,M).
PROPOSITION 2.73 Suppose that G = (X) and let X : X -7 M be a \;Conversely, it is straightforward to check that, for any 7 E Z2(G, M), the
map. If X extends to a derivation X : G -7 M, then: . oup E = {(g,m) I 9 E G, m E M} with multiplication defined by
(i) X(x- 1 ) = _X(X)"'-' for all x E X. (g, m)(h, n) = (gh, 7(g, h) + m h + n)
(ii) Let 9 E G with 9 = xt ... x~~, where each c; = ±1. Then
,(1, KG-module extension of M by G that defines the 2-cocycle r when we
I oose 9 = (g,O).
X(g) = Lc;X(XkYi A general transversal of M in E has the form fj = is. X(g») for a function
;=1 : G -7 M, and it can be checked that this transversal defines the 2-cocycle
where g; = x~::~ ... x~~ or x~~ ... x~~, when c; = 1 or -1, respectively. + cx ' where Cx is defined by cx(g, h) = X(gh) - X(g)h - X(h). A 2-cocycle
the form Cx for a function X : G -7 M is called a 2-coboundary, and the
additive group of such functions is a K-module and is denoted by B2(G, M).
PROOF By (t), we have OM = X(xx- 1 ) = X(X)"'-' +X(x- 1 ) , which proves Two KG-module extensions E 1 and E2 of a KG-module M by G are said
(i). (ii) is proved by repeated use of (t)· I 10; be equivalent if there is an isomorphism from El to E 2 that maps Ml to
2 and induces the identity map on both M and on G. From the above
In general, given any map X : X -7 M, we can use (i) and (ii) of the above cussion, it is not difficult to show that the extensions corresponding to the
proposition to extend X to a derivation Fx -7 M, where Fx is the free group ocycles 71 and 72 are equivalent if and only if 71 - 72 E B 2(G, M) and,
onX. particular, an extension E splits if and only if its corresponding 2-cocycle
E B 2 (G , M ).
2.7.3 Extensions of modules and the second cohomology group
Let E be any extension of an abelian group M (regarded as subgroup of E)
EFINITION 2.75 The second cohomology group H2 (G, M) is the quo-
by a group G. So we have an epimorphism p : E -7 G with kernel M. For nt K-module Z2(G,M)IB2(G,M).
9 E G, choose gEE with peg) = 9 and, for m E M, define m g := mg. Since
M is abelian, this definition is independent of the choice of g, and it defines So H2 (G, M) is in one-one correspondence with the equivalence classes of
an action of G on M. In general, this action makes M into a ZG-module, -module extensions of M by G.
but if M happens to be a module over a commutative ring K with 1, and the Checking directly whether a given 2-cocycle is a 2-coboundary can be diffi-
conjugation actions of 9 E G define K-automorphisms of M, then M becomes It, but there is an alternative approach to deciding whether a KG-module
a KG-module. In particular, this is true with K = IF'p in the case wnen':tti~t: ension E splits, which we shall now describe.
an elementary abelian p-group. Suppose that we have a finite presentation (X I R) of G, and let us identify
with the group FIN defined by the presentation, where F is the free group
X and N = (RF ) . For each x E X, choose an element x E E with
DEFINITION 2.74 Let G be a group and M a KG-module for some )=a z. Then there is a unique homomorphism 8: F -7 E with 8(x) = x
commutative ring K. We define a KG-module extension of M by G to be a r all x E X and, since pO( x) =a z, pO is the natural map from F to
54 Handbook of Computational Group Theory 55

G. Hence ker(pO) = N. It follows that ker(B) ::::: N in any case and, by Example 2.8
Theorem 2.52, ker(O) = N if and only if 0 induces i} : G -+ E, in which case ,LetM = C2 = (t), G = O2 X O2 = (x,y) with G acting trivially on M. It is
pi} is an isomorphism and im(O) = (x I x E X) is a complement of Min E. not difficult to check that there are eight equivalence classes of extensions of
Furthermore, ker(O) = N if and only if O(w) = IE for all wE R, so we have M by G (in fact H2(G,M) ~ O2 X O2 X O2), in which each of x2, y2 and [x,y]
proved the following lemma. can be equal to 1M or t. However, there are only four isomorphism classes of
gJ:Qups E that arise, namely O2 x O2 X O2,04 X O2 , D s , and Qs. 0
LEMMA 2.76 If x E E are chosen with p(x) =G x and 0 : F -+ E is ;,Suppose that a : E 1 -+ E 2 is a group isomorphism mapping M to M, and
defined by B(x) = x, then the elements x generate a complement of M in E ; dueing J.t E AutK(M). Then, since El/M ~ E2/M ~ G, a also induces
if and only if O(w) = IE for all w E R. JE Aut(G). We have m 9P, = mP,9 v for all m E M, 9 E G, because both
pressions result from applying a to fJ-'-lmfJ, where fJ E E 1 maps onto 9 E G.
The elements O(w) for w E R always lie in ker(p) = M. Let {XX(x) I x E
X } be another choice of the inverse images of x under p, where X : X -+ M
;!:r:,
~'ilEFINITION 2.79 Let M be a KG-module. If J.t E AutK(M) and
is a map, and let Ox: F -+ E be the associated homomorphism with Ox(x) = v
FE Aut(G) are isomorphisms satisfying m 9P, = mP,9 for all m E M and
XX(x) for x E X. If we use (i) and (ii) of Proposition 2.73 to extend X to
leG, then (v, J.t) is called a compatible pair.
X: F -+ M, then a simple calculation shows that Bx(w) = B(w)x(w) for all
w E R. Since the elements in these equations all lie in M, we can switch to
Lrhis expression was introduced by Robinson [Rob8I]. It was first used
additive notation and write them as Ox(w) = B(w) + X(w). So we have the
following result. :c9Illputational group theory by M.J. Smith in Section 4.2 of [Smi94], in
miection with the computation of the automorphism groups of solvable
oups; we shall return to that theme in Section 8.9.
PROPOSITION 2.77 With the above notation, the elements {XX(x) I ['he set Comp(G, M) of compatible pairs forms a group under composition,
x EX} generate a complement of M in E if and only if x(w) = -B(w) for d is a subgroup of Aut(G) x AutK(M). If r E Z2(G, M) and (v, J.t) is a
all wE R. mpatible pair, then we can define r(v,p,) by the rule

We shall use this result later, in Section 7.6, to help us determine compu-
r(v,p,) (g, h) = r(gV- 1, hV-1)P,
tationally whether an extension splits.
'all g, hE G. It is straightforward to check that r(v,p,) E Z2(G, M). Indeed,
In particular, from the case E = G ~ M, and x = (x,D) for all x E X, mapping r f-7 r(v,p,) defines an automorphism of Z2(G, M) that fixes
Proposition 2.71 yields: (G, M) setwise, and so it induces an automorphism of H2(G, M).
o Comp(G, M) induces a group of automorphisms of H 2 (G, M). From the
THEOREM 2.78 If M is a KG-module with G = (X I R), then a map 9ve discussion, it can be shown that the isomorphism classes of KG-module
X: X -+ M extends to a derivation G -+ M if and only if x(w) = OM for all ~nsions of M by G (where we are restricting attention to isomorphisms that
w E R, where X : F -+ M is defined by (i) and (ii) of Proposition 2.73. 'M) correspond to the orbits of Comp(G,M) on H 2(G,M).

This last result is the analogue of Theorem 2.52 for derivations. It can of
course be proved directly from the definition of derivations, without involving
semidirect products. It will be used later in Section 7.6 to help us to compute UG = On and K = JF q , then show that the smallest splitting field for G
Zl(G,M). containing L is lFq" where r is minimal with qr ::::: 1 (mod n).
:et M be an KG-module, where G is finite with IGI = n.
2.7.4 The actions of automorphisms on cohomology groups (i) Let X E Zl(G, M). By considering E 9EG X(gh) for h E G, show that
Two KG-module extensions E 1 and E 2 of M by G can be isomorphic as pX E Bl(G,M).
groups without being equivalent as extensions. This remains true even if we ..) Let r E Z2(G,M). By considering E 9EG r (g, hk ) for h,k E G, show
restrict our attention to isomorphisms that map M to M. that nr E B 2 (G, M ).
56 Handbook of Computational Group Theory 57

(iii) Deduce that if the map m 1-7 nm of M is an automorphism of M, then If a is algebraic over F, then the set of all / E F[x] with lea) = 0 forms
HI(G,M) = H 2(G,M) = O. This holds, for example, when M is finite ,ideal, and hence a principal ideal (P) of F[x], where p can be chosen to be
with IMI coprime to n. i nic- Then p is called the minimal polynomial of a over F. Since F has no

~f6 divisors, p must be irreducible over F.

3. Calculate Comp(G, M) and its action on H 2(G, M) in Example 2.8. "lip is the minimal polynomial over F of a E K, then (P) is the kernel of
fK',ring homomorphism r : F[x] -7 K in which r(7) = 7 for all 7 E F and
4. Show that Comp(G, M) induces a naturally defined action on HI (G, M).
) = a. Hence im(r) ~ F[xJl(P). Since p is irreducible, (P) is a maximal
5. Let M be a KG-module defined via 'P: G -7 AutK(M). of F[x] and so F[x)/(P) is a field and
(i) If 'P is a faithful action, then show that Comp(G, M)~NAutK(M)('P(G)). im(r) = F(a) = {/(a) I f E F[x], deg(f) < deg(p) },
(ii) If 'P is trivial, then show that Comp(G,M) = Aut(G) x AutK(M).
[F(a) :F] = deg(p).
6. If M is a KG-module defined via 'P : G -7 AutK(M), and v E Aut (G), n the other hand, if F is any field and p E F[x) is an irreducible polyno-
then we can define a KG-module MV via the action g 1-7 'P(gV). ,then K := F[x]/(P) is a field, which can be thought of as an extension
(i) Verify that M 1-7 M V defines an action of Aut(G) on the set of isomor- of degree deg(p) by identifying F with its natural image in K. If we
phism classes of KG-modules. e ato be the image x + (P) of x in K, then pea) = O. Hence a is a root
jin K, and p factorizes in K as (x - a)q(x) for some q E K[x].
(ii) Check that (1J,J.L) E Comp(G,M) exactly when J.L is a KG-module
isomorphism from M to MV. Hence, for v E Aut(G), there exists (v,/1) EO ore generally, if / E F[x) has an irreducible factor p of degree greater
Comp(G, M) if and only if M ~KG M V. ····1, then p factorizes nontrivially in the extension field K = F[x)/(P).
epeating this construction, we can show by a straightforward induction
ent on the largest degree of an irreducible factor of t, that there is an
sian field K of F, with [K: F] ~ n!, in which / factorizes into deg(f)
factors.
K has this property, and no proper subfield of K containing F has this
2.8 Field theory perty, then K is called a splitting field of / over F.
For a detailed treatment of the material in this section, see any book on is an important result that if K and K' are two splitting fields of f over
abstract algebra. J~~n there is a field isomorphism from K to K' that fixes every element of
~ere is a very brief outline of the proof. Use induction on the minimum
K: F] and [K' : F], let a E K, a' E K' be roots of the same irreducible
2.8.1 Field extensions and splitting fields or p of f with deg(p) > 1, observe that F(a) ~ F[x)/(P) ~ F(a'), and
IT F is a subfield of K, then K is said to be a (field) extension of F. The apply the inductive hypothesis to K and K' regarded as splitting fields
degree [K: F] of the extension is defined to be the dimension of K as a vector ,pvE)r F[xJl(P).
space over F. he characteristic char(F) of a field F is defined to be the smallest integer
If F ~ K ~ L, then we have [L : F] = [L: K][K : F]. This is proved by · 0 such that nIp = Op, or zero if there is no such integer n > O. So the
showing that for bases [Ui 11 ~ i ~ [K: F]] and [Vj 11 ~ j ~ [L: K]] of K iliar fields Q, lR and <C all have characteristic zero. If char(F) > 0 then,
s
over F and Lover K, [UiVj 11 i ~ [K:F], 1 ~ j ~ [L:K]] is a basis of L e F has no zero divisors, char(F) must be a prime p.
over F. is easily shown that a polynomial f E F[x) has repeated roots (that is,
An element a E K is said to be algebraic over F if f(a) = 0lgr:~9~e ted linear factors) if and only if gcd(f, /') 1= 1, where t' is the derivative
polynomial f E F[x]. Otherwise a is transcendental over F. If [K: F] is] · '" If F ~ K, then gcd(f, /') = I in F[x) if and only if gcd(f, f') = I in
finite, then all elements of K satisfy polynomials of degree at most [K: F] ],;so we can use this condition in F[x) to check whether f has repeated
over F, and so are algebraic over F. ts in its splitting field.
In any case, F(a) is defined to be the subfield of K generated by F and f is irreducible then gcd(f, 1') = I if and only if f' 1= 0, which is certainly
aj that is, the intersection of all subfields of K that contain F and a. The · e when char (F) = O. If char(F) = p > 0, then t' = 0 if and only if f
elements of F(a) are all quotients f(a)/g(a) with f,g E F[x], 9 =j:. O. lynomial in x p •
58 Handbook of Computational Group Theory 59

2.8.2 Finite fields To prove the proposition, let the finite subgroup H in question have order
-If H had exponent m < n, then the polynomial x m - 1 would be satisfied
The main result on finite fields is that all finite fields have prime power order
:allelements of H and hence have n > m distinct roots, which is impossible.
and that for each positive prime power q = pn, there is, up to isomorphism,
a unique' finite field of order q. We shall outline the proof of this fact in
iI has exponent n, and the result follows from the lemma.
n element a of multiplicative order q-l in lFq is called a primitive element
this subsection. Although it is only defined up to isomorphism, and can be
. Clearly lFq = F(a) with F = lFp , and so the minimal polynomial f of
constructed in different ways, it is customary to regard 'the' finite field of
qer F must be of degree n, where q = pn. An irreducible polynomial of
order q as a fixed object, and to denote it by lFq.
ee n overF, of which the roots are primitive elements of lFq is called a
Let K be a finite field. Then we must have char(K) = p > 0, and the subset
itive polynomial,
F := {nlK In E Z, 0 ::; n < p} forms a subfield of K, which is isomorphic
to the field lFp of integers modulo p. In particular, all fields of order p are (This meaning is distinct from and unconnected with the meaning of a
mitive polynomial over Z as one in which the greatest common divisors
isomorphic to lFp . . •
The degree n:= [K:F] must be finite, and then q:= IKI = pn IS a prime the coefficients is 1. This clash of meanings is unfortunate, but since the
cept of a greatest common divisors of field elements is trivial, there is
power, so all finite fields have prime power order.
bably little danger of confusion.)
The multiplicative group K# of K\ {OK} has order q-l, and hence a q - l ==
lK for all a E K#, and a q - a = OK for all a E K. So K contains q distinct t is easily verified that the map x -7 x P defines an automorphism of lFq of
roots of the polynomial x q - x E F[x]. Clearly no proper subfield of K can rn (it is called the Frobenius automorphism, and generates the automor-
have this property, so K is a splitting field of x q - x over F. It now follows sm group of lFq, but we shall not need that fact). So, if f is a primitive
from the uniqueness of splitting fields that all fields of order q are isomorphic. ynomial of degree n over F with root a E lFq, then the n elements in the
pi
To prove the existence of a field of order q for any prime power q = s", let Ta 10::; i < n} are all roots of fin lFq. Hence lFq is a splitting field of f.
K be the splitting field of x q - x over F := lFp . The derivative of x q - x is 'w is a primitive element of lFq then, for 0 ::; k < q - 1, w k is primitive if
-1, which is nonzero so, as we saw earlier, x q - x has q distinct roots in K. ~only if gcd(k, q-l) = 1, and so the total number of primitive elements is
It is easily checked that, if a, (3 are roots of x q - x in K then so are a ± (3, ~~1), where cP is the Euler Phi-function. Each primitive polynomial f has
a(3, and a / (3 if (3 i= OK, so the set of these roots form a subfield of K which,' ots in lFq so there are a total of cP(q-l)/n primitive polynomials.
by minimality of the splitting field, must be equal to K itself. SO IKI = q, as
required. Conway polynomials
hough lFq is unique up to isomorphism, it can, and often does, arise as
PROPOSITION 2.80 Any finite subgroup of the multiplicative group of splitting field of many different irreducible polynomials of degree n over
a field K is cyclic. In particular, the multiplicative group of a finite field is <.For computational purposes, it is useful to agree on a standard primitive
cyclic. Iynomial, so that different computer algebra systems can use the same
resentation of the elements of lFq. Unfortunately, there appears to be no
The proof of this depends on the following result from group theory. The . ural mathematical way of choosing such a standard polynomial.
exponent of a group G is defined to be the least common multiple of the orders .he standard that has been generally agreed upon is known as the Conway
of its elements or, equivalently, the least n > 0 such that gn = 1 for all g E G. nomial for lFq. (This is an unfortunate choice of name, because there is
ither meaning of Conway polynomial in knot theory!) They were originally
LEMMA 2.81 If G is a finite abelian group of exponent n, then G has oduced by Richard Parker, who also computed many examples. To define
an element of order n. , we first need to define an ordering on the set of all polynomials of degree
er F = lFp , and it is here that an apparently arbitrary choice had to be
e.
PROOF Let n = p~lp~2 '" P;' for distinct primes Pi. Then G must h~~e e order lFp itself by 0 < 1 < 2 < ... < P - 1. Then the polynomial
elements 9i of order p~i for each i. Since G is abelian, 9192 ... 9r hasoider-nj
as required. I
Note that the final step in the proof of the lemma is not necessarily true apped onto the word a n-lan-2'" alaO, and the resulting words are
for nonabelian groups G, and indeed the lemma itself is not true in general.: ered lexicographically using the above ordering of lFp .
60 Handbook of Computational Group Theory

The Conway polynomial for JFp is defined to be the least primitive polyno-
mial of degree 1 under this ordering. In other words, it is x - a, where a is
the smallest primitive element in JFp. For nonprime fields, there is an extra
hapter 3
condition.
Itturns out (see exercises below) that JFp~isasubfieldof JFpn if and only if
m divides n. For compatibility between the Conway polynomial f of the field .epresenting Groups on a
JFpn and its subfields, it is required that if a is a root of the Conway polynomial
f of JFpn then, for all proper divisors m of n, at with t := (pn _l)/(pm -1) omputer
should be a root of the Conway polynomial of JFp~ .
It is a slightly tricky exercise in modular arithmetic to show that a primitive
polynomial f exists that satisfies this property, where we may assume by
induction that the Conway polynomials are already defined and satisfy the
property for all proper subfields of lFpn. A proof can be found in the thesis of
W. Nickel [Nic88]. We can then define the Conway polynomial f of JFp" to be \'hare now ready to embark upon our study of computational group theory!
the least polynomial under the ordering defined above that has the property. any of the methods in CGT depend on whether the group is represented as
For example, for q = 2,4,8,16,32,3,9,27,5,25, the Conway polynomials oup of permutations, a group of matrices, or by means of a presentation
are respectively x+1, x 2+x+l, x 3+x+l, x 4+x+l, x 5+x2+1, x+l, x 2+2x+2, .. generators and relators. The most fundamental algorithms in CGT
x 3+2x+l, x+3, x 2+4x+2. aIly apply just to one specific representation type: for example, coset
One disadvantage of this definition is that there is nothing much better than oration can be applied only to a group defined by means of a finite
brute-force algorithms available for calculating Conway polynomials, which entation. Most of the later chapters in the book will be concerned with
become rapidly impractical as the field grows larger. On the other hand, they ithms for one particular representation type of groups.
only have to be computed once, and some of the calculations can be done .the current chapter, however, we shall start with some general consid-
in parallel. Some recent computations are described by Heath and Loehr ions about the different ways of representing groups on computers, and
in [HL04]. For example, they have been computed for fields up to order p97 ut the properties of a group and its elements that we might hope to be
for p ::; 11, p43 for p ::; 23, 673 1, 1276 • Complete lists are available on [Con]. .to compute in various situations. We shall then go on to introduce a few
calgorithms and techniques, such as choosing random elements of groups,
ch can be studied independently of the representation type.
Exercises

1. Find an example of a nonabelian group having no element of order equal

to its exponent.
2. If q and T are prime powers, show that lFr occurs as a subfield of JFq if and
only if q is a power of T, and in that case there is a unique subfield of lFq Representing groups on computers
isomorphic to JFr . The fundamental representation types
3. Prove that Conway polynomials can be defined for all finite fields lFq • You re are three methods commonly used to represent groups on a computer,
may want to use the fact that gcd(pm _1,pn -1) = pgcd(m,n) - 1 for all
ely, as groups of permutations of a finite set, groups of matrices over a
positive integers p, m, n. and as groups defined by a finite presentation. Most of the algorithms
ussed in this book will concern groups defined using one of these three
hods. Chapter 4 and various sections of later chapters will be devoted
mputing in permutation groups, Section 7.8 will deal specifically with
groups, and Chapters 5, 9, 12, and 13 will be concerned with groups
by a finite presentation. In addition, Chapter 8 will be concerned with
.' ps defined by a particular type of finite presentation, namely a polycyclic
.Elntation.
62 Handbook of Computational Group Theory :epresenting Groups on a Computer 63
We generally prefer our permutation groups of degree n to act on the set. Situation B. As in Situation A and, additionally, we can decide whether
{I . . n}; that is, we prefer to compute with subgroups of Sym(n). In cases two representations of group elements define the same element.
when the action is given or arises naturally on some different set, we would
Examples: a permutation group G on a finite set defined by an arbi-
normally choose to rename the points as {I .. n} before attempting any fur-
ther computations. In the case of matrix groups over a ring, we shall only be trary generating set; a matrix group G over a finite field or an algebraic
number field defined by an arbitrary generating set.
concerned with the case where we can carry out the basic operations of addi-
tion and multiplication exactly within the ring. So the ring might be a finite 3. Situation C. As in Situation B and, additionally, we have a special gen-
field, the integers or the rational numbers, or perhaps an algebraic number erating set gl, ... ,gr of G and an algorithm that can compute for every
field, but not the real or complex numbers. element 9 EGa unique word wg with g = W g (gl , ' " ,gr).
We may also occasionally want to consider other groups arising from these
basic types, such as quotient groups of permutation or matrix groups, which Example: a permutation group G with a so-called 'strong generating
cannot always be easily represented in the same way themselves. Examples set'. This 'strong generating set' is used as the special generating set
are given by P.M. Neumann in [NeuS7] of permutation groups of order sn and for G. For its definition and the corresponding algorithm to compute
degree 4n, for n ? 1, which have quotients for which the smallest degree of a unique words W g , we refer to Section 4.4.
faithful permutation representation is 22n +l . See the exercise below. In general, the algorithm for computing w g is known as a rewriting
Virtually all of the techniques that have been built up for computing within algorithm.
groups take as a starting point a finite generating set for the group or groups
concerned, and we shall assume in this book that all of our groups are defined 'rr a certain group is given to the computer, then our first and principal aim
by means of a finite generating set (or sequence). So, for example, if we were fore doing anything else with the group is usually to move into the best
trying to develop an algorithm to find a Sylow p-subgroup of a finite group, , computational situation for this group. The methods to move into a
then our aim would be to output generators for that subgroup. situation depend (again) on the representation of the given group. For
We want to be able to compute in very large finite groups, and so we
try very hard to avoid algorithms that would require computing a list of all
elements of the group, or even looping over all such elements. This is not A group G is defined by a finite presentation (X I R) and, additionally,
always possible. As we saw in Example 2.3, any finite group whatsoever has a it is known that G is finite. Then for most computational purposes we
faithful permutation representation acting by right multiplication on its own would first try to determine a faithful permutation representation for
elements, which we called the (right) regular action. Occasionally, we may G and thereby move from Situation A to B. The Todd-Coxeter coset
need to resort to using this representation, which is tantamount to listing enumeration algorithm to be discussed in detail in Chapter 5 is most
all of the group elements, although the group elements would be represented commonly used for this purpose.
by natural numbers in this situation. For example, if we need a faithful
2. A group G is given by a generating set of permutations. Then for most
permutation representation of a proper quotient of an existing permutation
computational purposes we would first compute a strong generating set
or matrix group, then the only default choice is the regular representation.
for G and thereby move from Situation B to C. Algorithms for achieving
this will be described in Section 4.4 and in Chapter 6.
3.1.2 Computational situations
,~~dSituation C is our preferred situation for computational purposes. It has
As a rough guideline, there are three basic situations that occur in practice, ~rious computational advantages. For example, the words wg used in Sit-
and which affect our ability to extract information about a group G compu- '~ll,tion C play the role of a normal form for group elements. They allow us
tationally. Of course, variations and combinations of them are also possible, read off the order of the group G (which may be finite or infinite) as the
but they serve to give an approximate idea of the possible states of affairs. rnber of possible normal forms. Also, in Situation C we can store and com-
1. Situation A. We can represent every element of G on the computer, and e elements using their corresponding normal form words, which is often a
we have methods to compute representations of inverses an~.pro1:u~ts putational advantage. Often it enables us also to test a given element of
of group elements. But we may not be able to decide wIlethettwo me superstructure (such as Sym(O) for permutation groups, or the general
representations of group elements define the same element. ear group for matrix groups) for membership in G.
In some situations, including the base and strong generating set framework
Example: a group G defined by a finite presentation (X I R). permutation and matrix groups, and the polycyclic series for polycyclic
64 Handbook of Computational Group Theory Representing Groups on a Computer 65

groups to be described in Chapter 8, this normal form is derived from a chain t,Now it turns out to be much easier to write an arbitrary element of Alt(6)
of subgroups ~a word in Xl,X2,XS,X4 than in the original generators Xl,X2· (The reason
G = G(l) J G(2) J ... J G(k + 1) = 1 for this will become clear in Section 4.4, when we define bases and strong
generating sets. In fact Xl, X2, Xs, X4 is a strong generating set with respect to
of G, and the normal form itself is 9 = 9k'" 9291, where 9i lies in a fixed ~hebase [1,3,2].) For example, (1,'5)(2,6) = x4xix2lxlXS' We could now, if
transversal U, of G(i+1) in G(i). cWbwanted to, use the definitions of Xs and X4 coming from the corresponding
.SLP elements to evaluate this element as a word in the original generators.
example, Xs = (XlX2? and
3.1.3 Straight-line programs
The most convenient generating set for the normal form words w g that we X4 = X2XS
2 -1
X2 = X2 XlX2
2( )-2
X2 = X2Xl X2 Xl X2,
-1 -1 -1

discussed above in connection with Situation C is often not the same as the
initial generating set on which the group is defined. For example, for a per-
(1,5)(2,6) = x2xllx2lxllx2xlx2xi (XlX2?'
mutation group, a strong generating set used for w g is generally a superset
of the initial generators. However, in many situations we need to be able to cidentally, this is not the shortest word possible for (1,5)(2,6), which is
relate the generators used for the w g to the initial generators. '>xlx2xllx22, but in large groups there is no satisfactory algorithm for find-
For that purpose, we shall introduce a data structure known as a a strai9ht- kshortest words for elements in the original generators. We shall discuss
line pro9ram (SLP). It also provides a space-efficient method for storing ele- is problem later, in Subsection 4.8.3.
ments in groups such as permutation or matrix groups. The terminology was In general, we proceed by defining an SLP group on generators Xi corre-
introduced in computer science as a certain type of program. Here we shall '. ding to our initial generators Xi (1 :::; i :::; k) of G, and the corresponding
use it in the following form. Juation in which Xi maps to Xi. We then define SLP elements Wl, , ui-,
We define an SLP group S of rank k to be a free group of rank k on ally with Wi = Xi for 1 :::; i :::; k, together with their evaluations Xl, , x;
generators Xl, ... , Xk. We shall call the elements of S SLPs or SLP elements. ,0, where these elements are chosen in such a way that arbitrary elements
From a purely mathematical viewpoint, there is no difference between an SLP EG can be written easily as reasonably short words in the extended gen-
group and any other free group. The difference lies in the way that the SLP , ting set {Xl, ... , x r } . In the case of permutation and matrix groups, we
elements are stored. They are not, in general, stored as words in the original nd the initial generating set to a strong generating set (see Section 4.4),
Xi, but as words in SLPs that have been previously defined. So, for example, ch enjoys this particular property. The definitions within the SLP could,
if k = 2, then rincipal, be used for rewriting group elements as words in the original gen-
rs, but we would avoid doing that if possible. We might choose to store
WI
A A ( ) 2 ,W4:= W22WS-1 W2
:= Xl, W2 := X2, Ws:= WlW2 elements and their inverses from this extended generating set explicitly as
'tmutations or matrices, but that is not essential, because they can always
defines a sequence of SLP elements WI , W2, Ws, W4. To avoid the words growing evaluated when required.
too long, they are not evaluated as words in the original generators of S, but A. s we shall explain shortly, straight-line programs are of vital importance
stored exactly as they are defined. So, in particular, a product of two existing icomputations involving group homomorphisms.
SLP elements is stored as that product, and the product is not expanded.
An evaluation ep of an SLP group S in a group G is an assignment of the
11..4 Black-box groups
generators Xi of S to elements Xi of G. By definition of a free group, this
defines a unique homomorphism <p : S -t G in which each Xi is mapped to Xi. ~jshall not be devoting too much attention to black-box groups in this book,
The storage method renders the evaluation of <p(w) for an SLP element W E S the topic is important enough to warrant a mention, if only a brief one.
very easy, and this is the motivation behind these ideas. Since wis~t~~~~ as he term 'black-box group' can be confusing, because it does not describe a
a (usually short) word in SLP elements Wi that were defined earliE!r;:w€/first icular type of group; it refers rather to a method of representing group el-
calculate the evaluation ep(Wi) E G of the Wi used to define w, and then we ts within a computer, together with some assumptions on the availability
can calculate ep(w) itself. gorithms for composing, inverting, and comparing group elements.
So, in the example above, if we take G = Alt(6) and Xl = (1,2,3), X2 :::::: .was introduced by Babai and Szemeredi in [BS84]. The definition is
(2,3,4,5,6), then WI, W2, Ws and W4 evaluate in G respectively to Xl, X2, llows. A finite alphabet A and an integer N > 0 are given. Group
X3:= (2,5)(4,6), and X4:= (4,5,6). ents are represented by strings in A* of length at most N. If we have
66 Handbook of Computational Group Theory Representing Groups on a Computer 67

strings representing g, h E G, then we can compute in constant time (using

an 'oracle') strings that represent g-l and gh, and we can decide whether
Let G be the direct product of n copies of the dihedral group of order 8.
g=G h.
Show that G has a faithful permutation representation of degree 4n. Let Zi
Note that we cannot in general decide whether a given string of length at be the central element of order 2 in the i-th copy of DB, and let K be the
most N in A* represents a group element. (Deciding that is similar to the subgroup of G generated by {ZlZ;l I 2 ::s i ::s n}. So K has order 2n - 1, and
membership testing problem.) G/ K is a central product of n copies of D8 • Show that the smallest degree
of a faithful permutation representation of G/ K is 2n +l (see [Neu87].)
In fact a black-box group is really just a description of Situation B for a
group G, but with a significant additional restriction: the stipulation that (Hint: Let the i-th copy of D s be generated by Xi, Yi. Then the image in
strings have length at most N means that a black-box group is necessarily G/ K of the subgroup generated by the Xi is abelian of order 2n and is dis-
finite, with an upper bound of I:~o IAl i on the group order. Finite permu- joint from Z (G/ K). Any larger subgroup must include two noncommuting
tation groups and matrix groups over finite fields are important examples of elements and hence must contain Z (G/ K). )
black-box groups.
Let G = (x I x n ) be cyclic of order n for some large positive integer n.
An algorithm for black-box groups lends itself readily to complexity analysis Devise a method of representing elements of G efficiently using an SLP
in terms of both time and space, where the time complexity will involve the wgroup on the single generator x.
number of calls to the oracle.
There is a significant research project in CGT to develop a full collection
of algorithms for finite groups given as black-box groups. The advantage is
that they are independent of the particular type of representation of the group.
The disadvantage is that they cannot make any use of this representation. For The use of random methods in CGT
example, this automatically rules out many of the most important algorithms
for permutation groups, which typically rely heavily on the orbit structure of M;my algorithms in CGT, and indeed in computer algebra in general, depend
the group and its subgroups. 9Jitically on making some random choices, such as choosing random elements
of groups. In the first subsection we describe the two principal types of ran-
There is one other important attraction of black-box group algorithms, g.pmized algorithms that are used in CGT, and in the second we present one
which is that if K is a normal subgroup of G, and we have an algorithm ~PlJCific method of choosing random elements from a finite group.
for testing strings for membership in K, then this algorithm can be used as
the oracle for testing whether an element of G/ K is the identity, and then
G/ K becomes a black-box group, where the representatives of 9 E G become Randomized algorithms
representatives of gK E G/K. ~i deterministic algorithm is one that depends on its input data alone. So, if
~s run repeatedly with the same input data, it will always proceed in the
Unfortunately, the black-box setting is too restrictive for most purposes.
ntical fashion. A complexity analysis of such an algorithm will provide an
One major problem is that there is no efficient black-box algorithm for de-
curate and consistent estimate of its time and space requirements.
termining the order of an element g; the only way is to test gk = 1, for
'A randomized algorithm is one that makes use of a random number genera-
k = 1,2, ..., which is likely to be too slow if Igl is large. Since there are
efficient ways of finding the order of an element in a permutation group or in t, during its execution, typically in order to choose random elements of some
a matrix group (a method for the latter is described by Celler and Leedham- up G. Such an algorithm does not depend on its input data alone, and its
Green in [CLG97]), a common recourse is to assume that the group is equipped ormance may vary, sometimes dramatically, from one run to another with
arne input.
with a further oracle for determining element orders. But then the problem
of passing to quotient groups reasserts itself! may still be possible to carry out complexity analyses of such algorithms,
he results will typically merely estimate the average running time and
Perhaps the most important achievement in this area to date is the devel- requirements of the algorithm. An analysis of this kind will need to
opment of methods for the constructive recognition of the finite nonabelian e that the random number generator being used is working properly, and
simple groups given as black-box groups. We shall return briefly to this topic ble of choosing genuinely random integers within a given range [a.. b].
later in the book, in Section 10.3. ,ractice, of course, there is no such thing as a perfect random number
68 Handbook of Computational Group Theory .' Representing Groups on a Computer 69

generator, but this is not the place to explore this interesting question, and required that € should be an input parameter.
we refer the interested reader to the lengthy treatment by Knuth in Chapter 'niThese are much less controversial than Monte Carlo algorithms, because in
3 of [Knu69]. practice they can always be made to return a correct answer just by running
There are two distinct types of randomized algorithm that arise in CGT; them repeatedly with the same €, and implementations will often do this
they have become generally known within the trade as Monte Carlo and Las . automatically. The only problem is that, if one is unlucky, then one may
Vegas algorithms. . have an unexpectedly long wait for an answer. The majority of randomized
A Monte Carlo algorithm is one which may sometimes output a wrong '. algorithms that we shall encounter in this book are of this type.
answer! It is, however, a requirement that one of the input parameters of
such an algorithm should be a real number € with 0 < € < 1, and that the
Finding random elements of groups
probability of the answer being wrong should then be less than € for any-values
of the remaining input data. The performance will depend on €, with smaller }\.gain we shall assume that we can choose random integers within a given
values of e resulting in longer running times. range. This clearly enables us to choose random elements of any indexed list
Although the first reaction of many traditional mathematicians (who may [,,[land we shall assume the availability of a function RANDoM(l) that does this
",.'
in any case be generally suspicious of the use of computers in mathematics) is r.us.
to recoil in horror at the idea of a process that has an intrinsic possibility of t3Many algorithms in CGT depend critically on the ability to generate quickly
generating a false result, these algorithms do have their uses. The best-known l111iformly distributed random elements of a group G. There are some cases
example is primality testing of integers, for which there are Monte Carlo meth- ~here this is relatively easy; for example if G = Sym(n) (see exercise below)
ods that run significantly faster that the best-known deterministic methods, :d, more generally, we shall see in Chapter 4 that it is easy for finite per-
and which are adequate, and essential in practice, for their applications to tation groups G for which a base and strong generating set are known.
cryptography. wever, in some situations we need to be able to generate a sequence of
There are other mathematicians, more in the habit of relying on the com- dom elements from a group G for which we are in Situation A or B, as
puter to help with their work, who go to the opposite extreme, and argue ned in Subsection 3.1.2 above. For finitely presented groups (Situation
that an answer delivered with an error estimate of 10- 20 , for example, should ,all that we can sensibly do is to find a random word for which the length
simply be regarded as proven correct, because even traditional mathematical 'g within a given range.
proofs written out in full detail might struggle to acquire a comparable degree Unfortunately, even for black-box groups, there is no known method of
of reliability. Indeed, with any computer calculation, there are small possibili- nerating random elements that is completely satisfactory. A method pro-
ties of error, such as hardware or compiler error, which are completely outside : sed by Babai in [Bab91] does generate elements that are guaranteed to be
the control of the typical mathematical user. nuinely uniformly distributed, but it is far too slow to be useful in prac-
For a mathematical treatment of algorithms, such as the one we are at- ceo Here, we shall describe the product replacement algorithm, proposed
tempting in this book, it seems desirable, however, to make suitable simplify- [CLGM+95], which generates a sequence of elements that is not guaran-
ing assumptions, such as the absence of hardware or compiler errors, and the d to be uniformly distributed, but which is very fast, and appears to behave
ability to generate genuinely random numbers, and then to treat the output isfactorily in those algorithms in which it has been used to date. (How-
of a Monte Carlo algorithm exactly for what it is; namely, an answer with a er, there are some serious reservations. For example, the behaviour can be
small, known probability of being wrong. satisfactory when the group is a direct product of a large number of copies
In most, if not all, of the important Monte Carlo algorithms in CGT, certain the same finite simple group. See the article of 1. Pak [Pak01] for a theoret-
answers are in fact guaranteed correct, whereas other answers come with the investigation.) We refer the reader to Section 2.2 of [Ser03] for a detailed
possibility of error. For example, in Section 4.2 we shall describe a Monte ussion and theoretical analysis of both Babai's method and the (original)
Carlo algorithm for testing whether an input permutation group on 0 is Alt(n) uct Replacement Algorithm.
or Sym(O). The answer 'yes' is guaranteed correct, whereas the answer 'no' e maintain a list [Xl,X2,'" ,xr ] of elements of G, which generate G. A
has a small chance of being wrong. In practice, such algorithms are often used .ble value of r has been found in practice to be about 10, but if the given
as filters; we want to check quickly whether the group is very large (Alt(n) generating sequence X of G has more than this number of elements,
or Sym(O)) and, if not, then we plan to analyze it further and perhaps to we must take r = IXI. It is argued in [PakOl] that a larger value of r
calculate its precise order. . be necessary for guaranteed good theoretical behaviour. If IXI < r,
A Las Vegas algorithm is one that never delivers an incorrect answer, but 'we initialize the list to contain the generators in X repeated as many
has a probability of at most € of not returning an answer at all. It is again '€is as required. In the original version, we repeatedly choose random dis-
70 Handbook of Computational Group Theory R,epresenting Groups on a Computer 71

tinct integers s, t E [1.. r], and replace Xs by xsxt 1 or xt1x s. Notice that ::"The reader should be aware that we are omitting the details of the imple-
the list still generates G after making this change. After doing this a certain inentation of SLP groups in the code for PRRANDOM; each SLP element Wi is
number of times to get us away from the initial generating set (50 times has being defined as a word in existing SLP elements, and so all of these existing
been used as a default in implementations), we start returning the new values elements would need to be stored in order to allow later evaluation of the Wi.
of X s as our random elements. llisituationslater in the book in which we do not need the SLP-elements, we
It is proved in [CLGM+95] that the elements returned will eventually be S1\.all call PRRANDOM with only one input parameter, X.
uniformly distributed random elements of G, but there is no reasonable esti- qi
mate known on how long we will have to wait for this, and it is likely that )YRRANDOM("-' X ,"-'W)
the ad hoc default waiting time of 50 changes is not enough. Another possi-
Input: List X = [XO,Xl,'" ,xr ], W = [wo,Wl,'" ,wr ]
ble problem with the algorithm is that the group elements 91,92 returned on
Output: An element of G and corresponding SLP element
successive calls of the algorithm will never be uniformly distributed in G x G.
s := RANDOM([I .. r]);
The version that we shall present here is a modification due to Leedham-
t:= RANDOM([l . . r] \ [8]);
Green, which involves the use of an accumulator Xo that is always used as the
X :;= RANDOM([I .. 2]);
return value. (As a cricket enthusiast, Leedham-Green likes to take r = 11 and
:= RANDOM({-I,I});
to call Xo the 'twelfth man', but we prefer to avoid the gender bias inherent
if X = 1
in that terminology!) There is heuristic evidence that this converges more
then Xs := xsxr; xo:= xox s;
quickly than the original version.
Ws := wswr; Wo;= wows;
To use the method, we must first call the following initialization function
else Xs;= xlxsi Xo:= XsXO;
on our generating set X. This sets up the list X = [XO,Xl,'" ,X r ], performs
ui, := wlw s; Wo;= WsWO;
the product replacement process PRRANDOM n times as initialization, and
,return Xo, Woi
returns X. In some applications, we shall need to know exactly how the
random elements returned were derived from the original generators, and so
we shall also maintain SLP elements uu, which express each element Xi as {The list [1 .. r] \ [s] in this function can be easily constructed by removing
a word in the original generators Xl, ... , X k of G. As usual, we denote the from [1 .. r] and, if s =/:. r, replacing s by r.
generators of the SLP group that evaluate to the Xi by Xi.

PRINITIALIZE(X, r, n)
Input: X = [Xl, ... , Xk] generating a black-box group, r, n E N
Output: List X and associated SLP elements W Find a method for constructing uniformly distributed random elements of
1 Let Xi (1 :::; i :::; k) be generators of an SLP group; Sym(n) in time O(n), on the assumption that random elements from a
2 for i E [1.. k] do Wi := Xi; list of length up to n can be found in constant time. (This assumption
3 for i E [k+ 1 .. r] do Xi := Xi-k; Wi:= Wi-k; 'is false for large n, but provided that n is small enough for the list to
4 Xo:= 10; Wo:= t; ,be represented on a computer in RAM, then it is a reasonable practical
5 X:= [XO,Xl,""X r ]; W:= [WO,Wl,""W r ]; 'assumption to make.) It is clear how to start: choose the image 19 under
6 for i E [1. .n] do PRRANDOM("-' X, "-'W); our random permutation 9 to be a random element of {I . . n}. But now
7 return X, W; ,e have to choose 29 from {1.. n} \ {II}, so how can we manage this by
oing only a small constant amount of reorganization of our data?

We remind the reader that we precede a variable by a "-' symbol in a function ,(,you have access to GAP or MAGMA, then implement the procedures
or procedure call when the value of that variable may be be altered by the ,il,INITIALIZE and PRRANDOM (but leave out the calculations with SLP
function or procedure. ." .....' ements if you prefer). Tryout your functions on the groups GL(d,q)
The function PRRANDOM, which outputs random group element~;:'shocld .various parameter settings, and various d and q. How many iterations
be called for the first time using the lists X, W returned by PRINITIALIZE. .PRRANDoM in PRINITIALIZE are needed to ensure that the elements
Subsequently, it should be called using the altered values of X, W resulting urned by PRRANDOM appear to be random? How dependent is this
from the previous call of PRRANDOM. umber on q?
72 Handbook of Computational Group Theory presenting Groups on a Computer 73

:::
RDERBOUNDED(g, n)
3.3 Some structural calculations Input: 9 E G, n E No with Igi dividing n
'Output: Igi
In this section, we present examples of some structural. computations that if n = 1 then return 1;
can be carried out in any class of finitely generated groups for which we can for p E PRIMEDIVIsoRs(n)
find random group elements, and in which we can test membership of group do if PowER(g,nlp) = IG
elements in finitely generated subgroups. This is typically the case when we then return ORDERBoUND-ED(g, nip);
are in Situation C (see Subsection 3.1.2).

3.3.1 Powers and orders of elements

Suppose that we are in Situation B for a group G, and that we need to t H = (Y) be a subgroup of a group G = (X) with X and Y finite, and
calculate s" for some 9 E G and n E No. The obvious thing to do is just pose that we want to find the normal closure N:= (HG) of H in G. The
to multiply 9 by itself n - 1 times. This is fine for small n, but for larger ious way to proceed is to start with N = H, test each y" with y E Y,
n, there is a much quicker way, which has complexity O(log(n)) rather than A := X U X-I for membership in N, and adjoin it to Y as an extra
O(n). We calculate gn as a product of elements g2', where each individual erator if the test fails.
g2' can be calculated by repeated squaring of g. This is carried out in the f course, in general, N might not be finitely generated, in which case this
following function. cess would not terminate, but usually we will want to do this when at
Of course, if we knew in advance that we were going to need to calculate t IG: HI is finite, in which case termination is guaranteed, and the above
gn for the same 9 and for many different values of n, then we could save time cedure provides us with a deterministic algorithm.
by storing the elements g2' rather than recalculating them each time. fact, when IG: HI is large and finite, it is usually faster to adjoin a number
random conjugates of H by G to the generating set of N, and then to test
POWER(g,n) ether N is the full normal closure. The procedure NORMALCLOSURE does
Input: 9 E G, n E No ~. The random conjugates are added in batches of size n. In Section 5.1.4
Output: gn Ser03], Seress suggests the value n = 10 for permutation groups.
!..,
1 x:= IG;
2 if n mod 2 = 1 then x := xg;
3 while n > 1 ,3 The commutator subgroup, derived series, and lower central
4 do g:= g2; series
5 n:= n div 2; .H = (X) and K = (Y) be subgroups of a group G, where X and Y
6 if n mod 2 = 1 then x := xg; ,finite. By Proposition 2.35, [H, K] is the normal closure in (H, K) of the
7 return x; ,C := {[h, k] I hEX, kEY}. This can be computed by the normal
re algorithm described in Subsection 3.3.2 provided that the appropriate
A well-designed computer algebra system would use this method by default mptions on membership testing in subgroups are satisfied.
whenever the user types gn, at least for large values of n. wever, if the generating sets X and Yare large, then it may be expen-
As we mentioned in Subsection 3.1.4, finding the order of an element in a to construct the set 0, which has size IXIIYI, explicitly. In that case,
black-box group is not so easy in general and, without further information; fUight prefer to construct our random elements of the group H in NOR-
all that we can do is to test whether gn = IG for each power gn in turn. In ,'CLOSURE as random words in a smaller set of commutators [h, k], where
some situations, however, such as for finite field elements, orm'a'tl.'iCes-:'6ver ,d k are themselves random elements in Hand K. If we do this, then we
finite fields, it is possible to find an n such that the order Igl of 9 divides n, .keep changing the commutators [h, k] used, as well as the words in these
Provided that we can factorize n (which, unfortunately, may not be completely utators.
straightforward when n is very large), the following function will find the order se techniques immediately enable us to calculate the derived series and
of g in time at most O(log(n)3). wer central series of a group.
74 Handbook of Computational Group Theory" resenting Groups on a Computer 75

fa finite generating set X of G. This will of course uniquely determine

NORMALCLOSURE(X, Y, n)
G -+ H if such a ip exists, but unless F happens to be free on X, an
Input: Generating sequences X, Y of groups G, H, n E N ary assignment like this will not usually define a homomorphism at all.
Output: Generators of HO a particular situation, we might know already that ~ur ip does defi~e .a
1 Z:= Y; C:= false; morphism, .but in general we need a method for testing whether this IS
2 X:= PRINITIALIZE(X, 10, 20); ase- For example, if we are trying to compute the automorphism group
3 while not C G) of a finite group G, then we will certainly need to be able to carry
4 do Z := PRINITIALIZE(Z, 10, 10); uch a test.
(* Add some new random conjugates to Z *) e know a presentation (X I R) of G on X, then Theorem 2.53 provides us
5 for i E [1. . n] a simple test for this. We assume that cp does extend, and then evaluate
6 do g:= PRRANDOM("'X); age in H on each of the group relators in R. If these all evaluate to IH
7 h:= PRRANDOM("'Z); cp extends, and otherwise it does not. . . .
8 if h g rt (Z) then ApPEND(",Z,hg); some situations, such as when G is a large permutation group, It IS not
(* Test whether (Z) = HO *) to find a group presentation on the original generators X = [Xl,.'" Xk],
9 C:= true; one can be found efficiently on an extended sequence X' = [Xl, ... , x r ],
10 for 9 E X, h E Z . the new generators are defined from the existing ones via an SLP, as
11 do if hg rt (Z) ibed in Subsection 3.1.3 above. In that case, if we are given cp(Xi) for
12 then C := false; :::::; k, then we can simply use the expressions for each Xj with j > k as
13 break; 'lements in the earlier Xi, in turn, to compute cp(Xj). We can then carry
14 return Z; ur test for sp being a homomorphism using the presentation on XI.

2 Desirable facilities
3.4 Computing with homomorphisms ;iy, for a grouphomomorphism cp : G -+ H, we would like to be able to
rm the following operations.
The ability to compute effectively with group homomorphisms is one of t~
most important aspects of CGT. In particular, it allows us to move from on Compute the image cp(g) of any 9 E G.
representation of a group to another, possibly more convenient, represent
tion. For example, if a group is given initially as a matrix group or finite 'Por h E H, test whether h E im( cp) and, if so, find agE G with
presented group, then we might prefer to perform our computations in <p(g) = h.
isomorphic permutation group. ind cp(K) for a subgroup K of G.
We shall consider some particular types of homomorphisms, such as ma
ping to the induced action of a permutation group on a nontrivial block systei Find cp-I(K) for a subgroup K of H.
in later chapters, but there are some general ideas that are independent of t
particular representation of the group, which we can consider at this poi
The reader should bear in mind that, in certain specific situations, there some of these cases, it may be unclear what we mean by "Find". It might
happen to be more efficient solutions to these problems than the general on latively easy, for example, to find generators of cp(K) in (iii), which would
described here. s in Situation B for cp(K), but we would prefer to be in Situation C,
ing us with a rewriting algorithm and membership testing for cp(K).
3.4.1 Defining and verifying group homomorphisms at is immediately clear is that, if we can solve the rewriting problem in
ssibly by using extra generators and a SLP, then we can solve (i), and
There are two methods that are commonly used for definh'{i grou'p homom at least find generators of cp(K) for finitely generated subgroups K
phisms cp : G -+ H. The first is by using a general rule for calculating <p Similarly, if we can solve the rewriting problem in im(cp), then we can
for 9 E G. The induced action of a permutation group on a block syste ii), provided that we know inverse images for our chosen generating set
an example of this. The second method is to specify the images cp(x) E H£ ). To solve (iv), we need to be able to solve (ii) and (v), because, if
Random documents with unrelated
content Scribd suggests to you:
 The battle of Vimiero took place, and sir Arthur Wellesley naturally
declined a mission more suitable to a staff captain than a victorious
commander; but before sir Hew’s answer, exposing the false
calculations of the minister’s plans, could be received in England, a
despatch, dated the 2d of September, announced the resolution of
the government to employ an army in the northern provinces of
Spain, and directed twenty thousand men to be held in readiness to
unite with other forces to be sent from England; nevertheless, this
project also was so immature, that no intimation was given how the
junction was to be effected, whether by sea or land; nor had the
ministers even ascertained that the Spaniards would permit English
troops to enter Spain at all; for three weeks later, lord William
Bentinck, writing from Madrid, says, “I had an interview with Florida
Blanca: he expressed his surprise that there should be a doubt of
the Spaniards wishing for the assistance of the English army.” Such
also was the confusion at home, that lord Castlereagh repeatedly
expressed his fears lest the embarkation of Junot’s troops should
have “absorbed all the means of transport” in the Tagus, when a
simple reference to the transport office in London would have
satisfied him, that although the English army should also be
embarked, there would still remain a surplus of twelve thousand
tons.
When the popular cry arose against the convention of Cintra, the
generals in chief were recalled in succession, as rapidly as they had
been appointed; the despatches addressed to one generally fell into
the hands of his successor; but the plans of the ministers becoming
at last mature, on the 6th of October sir John Moore was finally
appointed to lead the forces into Spain. At this period the head of
the grand French army was already in the passes of the Pyrenees,
and the hostile troops on the Ebro coming to blows. The Spaniards
were weak and divided, and the English were forty marches from the
scene of action; yet, said the minister to sir John Moore, “there will
be full time to concert your plan of operations with the Spanish
generals before the equipment of your army can be completed.” Was
this the way to oppose Napoleon! Could such proceedings lead to
ought but disaster! It has been said, that sir Hew Dalrymple’s
negligence was the cause of this delay; that he should have had the
troops in readiness: but that general could not prudently incur the
expense of equipping for a march, an army that was likely to be
embarked; he could not, in short, divine the plans of the ministers
before they were formed; and it is evident that the error attaches
entirely to the government.
The incapacity of the Spanish generals has been already
sufficiently exposed by occasional observations in the narrative; their
faults, glaring and fatal, call for no further remark; but the exact
combinations, the energy and rapidity of the French emperor, merit
the most careful examination; his operations were not, as they have
been generally considered, a pompous display of power, to create an
appearance of conquest that was unreal, not a mere violent irruption
with a multitude of men, but a series of skilful and scientific
movements, worthy of so great a general and politician. It is true
that his force was immense, and that the Spaniards were but
contemptible soldiers; but he never neglected the lessons of
experience, nor deviated from the strictest rules of art. With
astonishing activity, and when we consider the state of his political
relations on the continent, we may add, with astonishing boldness,
he first collected ample means to attain his object, then deceiving
his enemies with regard to his numbers, position, and intentions,
and choosing his time with admirable judgment, he broke through
the weak part of their line, and seized Burgos, a central point, which
enabled him to envelop and destroy the left wing of the Spaniards
before their right could hear of his attack, the latter being itself
turned by the same movement, and exposed to a like fate. This
position also enabled him to menace the capital, to keep the English
army in check, and to cover the formation of those magazines and
stores which were necessary to render Burgos the base and pivot of
further operations.
Napoleon’s forces were numerous enough to have attacked
Castaños and Palafox, while Blake was being pursued by the first
and fourth corps; but trusting nothing to chance, he waited for
twelve days, until the position of the English army was ascertained,
the strength of the northern provinces quite broken, and a secure
place of arms established. Then leaving the second corps to cover
his communications, and sending the fourth corps into the flat
country, to coast, as it were, the heads of the English columns on his
right, and to turn the passes of the Carpentino mountains, he
caused the Spanish right wing to be destroyed, and himself
approached the capital, at a moment when not a vestige of a
national army was left, when he had good reason to think that the
English were in full retreat, when the whole of his own corps were
close at hand, and consequently when the greatest moral effect
could be produced, and the greatest physical power concentrated at
the same time to take advantage of it. Napoleon’s dispositions were
indeed surprisingly skilful; for although marshal Lefebre’s
precipitation at Zornoza, by prolonging Blake’s agony, lost six days of
promise, it is certain that reverses in battle could neither have
checked the emperor, nor helped the Spaniards.
For if Soult had been beaten at Gamonal, Napoleon was close at
hand to support the second corps, and the sixth corps would have
fallen upon the flank and rear of the Spaniards.
If the first corps had been defeated at Espinosa, the second and
fourth corps, and the emperor’s troops, would have taken Blake in
flank and rear.
If Lasnes had been defeated at Tudela, he could have fallen back
on Pampeluna; the fifth and eighth corps were marching to support
him, and the sixth corps would have taken the Spaniards in flank.
If the emperor had been repulsed at the Somosierra, the sixth
corps would have turned that position by Guadalaxara, and the
fourth corps by Guadarama.
If sir John Moore had retreated on Portugal, the fourth corps was
nearer to Lisbon than he was.
If he had overthrown Soult, the fifth and eighth corps were ready
to sustain that marshal, and Napoleon, with fifty thousand men, as
we have seen, was prepared to cut the British line of retreat into
Gallicia. In short, no possible event could have divided the emperor’s
forces, and he constantly preserved a central position that enabled
him to unite his masses in sufficient time to repair any momentary
disaster. By a judicious mixture of force and policy also, he obliged
Madrid to surrender in two days, and thus prevented the enthusiasm
which would doubtless have arisen if the capital had been defended
for any time, and the heart burnings if it had been stormed. The
second sweep that he was preparing to make when sir John Moore’s
march called off his attention from the south would undoubtedly
have put him in possession of the remaining great cities of the
Peninsula. Then the civil benefits promised in his decrees and
speeches would have produced their full effect, and the result may
be judged of by the fact that in 1811 and 12, Andalusia and Valencia
were under the able administration of marshals Soult and Suchet, as
tranquil and submissive as any department of France, and the
former even raised numerous Spanish battalions, and employed
them not only to preserve the public peace, but to chase and put
down the guerillas of the neighbouring provinces.
Sir John Moore’s talents saved the Peninsula from this great
danger, and here perhaps a military error of Napoleon’s may be
detected. Forgetting that war is not a conjectural art, he took for
granted that the English army was falling back to Portugal, and
without ascertaining that it was so, acted upon the supposition. This
apparent negligence, so unlike his usual circumspection, leads to the
notion, that through Morla he might have become acquainted with
the peculiar opinions and rash temper of Mr. Frere, and trusted that
the treacherous arts of the Spaniard, in conjunction with the
presumptuous disposition of the plenipotentiary, would so mislead
the English general, as to induce him to carry his army to Madrid,
and thus deliver it up entire and bound. It was an error; but
Napoleon could be deceived or negligent only for a moment. With
what vigour he recovered himself, and hastened to remedy his error!
How instantaneously he relinquished his intentions against the
south, turned his face away from the glittering prize, and bent his
whole force against the only man among his adversaries that had
discovered talent and decision! Let those who have seen the
preparations necessary to enable a small army to act, even on a pre-
conceived plan, say what uncontrollable energy that man possessed,
who, suddenly interrupted in such great designs, could, in the course
of a few hours, put fifty thousand men in movement on a totally new
line of operations, and in the midst of winter execute a march of two
hundred miles with a rapidity hardly to be equalled under the most
favourable circumstances.
The indefatigable activity of the duke of Dalmatia greatly
contributed to the success of the whole campaign, and it is a
remarkable circumstance, that Soult and Napoleon, advancing from
different bases, should have so combined their movements, that
(after marching, the one above a hundred, and the other two
hundred miles, through a hostile country) they effected their
junction at a given point, and at a given hour, without failure; and it
is no less remarkable that such a decided and well-conducted
operation should have been baffled by a general at the head of an
inexperienced army.

O B S E R VAT I O N S O N S I R J O H N M O O R E ’ S R E T R E AT.

When Sylla, after all his victories, styled himself a happy, rather
than a great general, he discovered his profound knowledge of the
military art. Experience had taught him that the urgent speed of one
legion, the inactivity of another, the obstinacy, the ignorance, or the
treachery of a subordinate officer, was sufficient to mar the best
concerted plan, nay, that the intervention of a shower of rain, an
unexpected ditch, or any apparently trivial accident, might determine
the fate of a whole army. It taught him that the vicissitudes of war
are so many, that disappointment will attend the wisest
combinations; that a ruinous defeat, the work of chance, often
closes the career of the boldest and most sagacious of generals; and
that to judge of a commander’s conduct by the event alone, is
equally unjust and unphilosophical, a refuge for vanity and
ignorance.
These reflections seem to be peculiarly applicable to sir John
Moore’s campaign, which has by sundry writers been so unfairly
discussed. Many of the subsequent disasters of the French can now
be distinctly traced to the operations of the British army. It can be
demonstrated that the reputation of that excellent man was basely
sacrificed at the period of his death, and that the virulent censures
passed upon his conduct have been as inconsiderate as they were
unmerited and cruel.
The nature of the commands held by sir John Moore in the years
1807-8-9, forced him into a series of embarrassments from which
few men could have extricated themselves. After refusing the charge
of the absurd expedition to Egypt in 1806, which ended, as he
judged it must do, unfavourably, he succeeded to the command of
the troops in Sicily, a situation which immediately involved him in
unpleasant discussions with the queen of Naples and the British
envoy: discussions to which the subsequent well-known enmity of
the cabinet of that day may be traced. By his frank conduct, clear
judgment, and firm spirit, he obtained an influence over the
wretched court of Palermo that promised the happiest results. The
queen’s repugnance to a reform was overcome, the ministers were
awed, and the miserable intrigues of the day were for the time put
down. The Sicilian army was reorganized, and a good military
system was commenced under the advice of the British general. This
promising state of affairs lasted but a short time; the Russian fleet
put into the Tagus, the French threatened Portugal, and Sicily was
no longer considered! Sir John Moore was ordered to quit that
island, and to assemble a large force at Gibraltar for a specific
service; but the troops to be gathered were dispersed in the
Mediterranean from Egypt to the straits, and their junction could not
be effected at all unless the English ambassador at Constantinople
should succeed in bringing a negotiation then pending between the
Turks and Russians to a happy issue. The special service in question
had two objects, 1º. to aid sir Sydney Smith in carrying off the royal
family of Portugal to the Brazils; and 2º. to take Sir John Moore’s
possession of Madeira; but neither were made known Journal, MSS.
to the general before his arrival at Gibraltar, which was not until after
Junot had taken possession of Lisbon. Sir John Moore then
(following his instructions) proceeded home, and thus our interests
in Sicily were again abandoned to the vices and intrigues of the
court of Palermo. On the passage he crossed general Spencer going
with a force against Ceuta, and soon after he had reached England,
he was despatched to Sweden, without any specific object, and with
such vague instructions, that an immediate collision with the
unfortunate Gustavus was the consequence. Having with much
dexterity and judgment withdrawn himself and his army from the
capricious violence of that monarch, sir John was superseded and
sent to Portugal, with the third rank in an army which at that time
no man had such good claims to command as himself[27].
The good fortune of England was never more conspicuous than at
this period, when her armies and fleets were thus bandied about,
and a blind chance governed the councils at home. For first a force
collected from all parts of the Mediterranean was transported to the
Baltic sea, at a time when an expedition composed of troops which
had but a short time before come back from the Baltic were sailing
from England to the Mediterranean. An army intended to conquer
South America was happily assembled in Ireland at the moment
when an unexpected event called for their services in Portugal, and
a division destined to attack the Spaniards at Ceuta arrived at
Gibraltar at the instant when the insurrection of Andalusia
fortunately prevented them from making an attempt that would have
materially aided Napoleon’s schemes against the Peninsula. Again,
three days after sir John Moore had withdrawn his army from
Sweden, orders arrived to employ it in carrying off the Spanish
troops under Romana, an operation for which it was not required,
and which would have retarded, if not entirely frustrated, the
campaign in Portugal; nor was it the least part of that fortune, that
in such long continued voyages in bad seasons, no disaster befell
those huge fleets thus employed in bearing the strength of England
from one extremity of Europe to the other.
After the convention of Cintra, sir John Moore was again placed at
the head of an army; an appointment unexpected by him, for the
frank and bold manner in which he expressed himself to the
ministers on his return from Gottingen left him little to hope; but the
personal good-will of the king, and other circumstances, procured
him this command. Thus, in a few months after he had quitted
Sweden, Moore, with an army not exceeding twenty-four thousand
men, was in the heart of Spain, opposed to Napoleon, who having
passed the Pyrenees at the head of three hundred and thirty
thousand men, could readily bring two hundred thousand to bear on
the British; a vast disproportion of numbers, and a sufficient answer
to all the idle censures passed upon the retreat to Coruña.
The most plausible grounds of accusation against sir John Moore’s
conduct rest on three alleged errors:
1st. That he divided his forces.
2dly. That he advanced against Soult.
3dly. That he made a precipitate and unnecessary retreat.
When a general, aware of the strength of his adversary, and of the
resources to be placed at his own disposal, arranges a plan of
campaign, he may be strictly judged by the rules of art; but if, as in
the case of sir John Moore, he is suddenly appointed to conduct
important operations without a plan being arranged, or the means
given to arrange one, then it is evident that his capacity or
incapacity must be judged of, by the energy he displays, the
comprehensive view he takes of affairs, and the rapidity with which
he accommodates his measures to events, that the original vice of
his appointment will not permit him to control.
The first separation of the English army was the work of the
ministers, who sent sir David Baird to Coruña. The after separation
of the artillery was sir John Moore’s act; the reasons for which have
been already stated; but it is worth while to examine what the effect
of that measure was, and what it might have been; and here it may
be observed, that, although a brigade of light six-pounders did
accompany the troops to Almeida, the road was not practicable; for
the guns were in some places let down the rocks by ropes, and in
others, carried over the difficult places; a practicable affair with one
brigade; but how could the great train of guns and ammunition
waggons that accompanied sir John Hope have passed such places
without a loss of time that would have proved more injurious to the
operations than the separation of the artillery?
The advance of the army was guided by three contingent cases,
any one of which arising would have immediately influenced the
operations; 1º. Blake on the left, or Castaños and Palafox upon the
right, might have beaten the French, and advanced to the Pyrenees.
2º. They might have maintained their position on the Ebro. 3º. The
arrival of reinforcements from France might have forced the
Spaniards to fall back upon the upper Douero, on one side, and to
the mountains of Guadalaxara on the other. In the first case, there
was no risk in marching by divisions towards Burgos, which was the
point of concentration given by the British and Spanish ministers. In
the second case, the army could safely unite at Valladolid; and in the
third case, if the division of sir David Baird had reached Toro early in
November (and this it was reasonable to expect, as that general
arrived at Coruña the 13th of October), the retrograde movement of
the Spanish armies would probably have drawn the English to the
Guadarama, as a safe and central point between the retiring Spanish
wings. Now the artillery marching from the Alemtejo by the roads of
Talavera and Naval Carnero, to Burgos, would pass over one
hundred and two Spanish leagues. To Aranda de Douero, eighty-nine
leagues. To Valladolid, ninety-two leagues. While the columns that
marched by Almeida and Salamanca would pass over one hundred
and sixteen leagues to Burgos, and ninety-eight to Valladolid.
Wherefore supposing the Spaniards successful, or even holding their
own, the separation of the artillery was an advantage, and if the
Spaniards were driven back, their natural line of retreat would have
brought them towards Madrid, Blake by Aranda to the Somosierra,
and Castaños and Palafox by Siguenza and Tarancon, to cover the
capital, and to maintain an interior communication between the
Somosierra and the Henares river. The British artillery would then
have halted at Espinar, after a march of only eighty leagues, and
Baird and Moore’s corps uniting at Salamanca early in November,
might, by a flank march to Arevalo, have insured the concentration
of the whole army.

Plate VIII
To Face Page 516
London. Published March 1828, by John Murray, Albermarle
Street.
 Thus, in the three anticipated cases, the separation of the artillery
was prudent, and promised to be advantageous. There was indeed a
fourth case, that which really happened. All the Spanish armies were
dispersed in an instant!—utterly effaced! but sir John Moore could
not have divined such a catastrophe, while his ears were ringing
with the universal clamour about the numbers and enthusiasm of
the patriots; and if he had foreseen even a part of such disasters, he
would never have advanced from Portugal. With the plans of the
Spanish government he was unacquainted; but he was officially
informed that above one hundred and forty thousand Spanish
soldiers were between him and a feeble dispirited enemy; and as the
intercepted letter from the governor of Bayonne stated that
reinforcements would only arrive between the 18th of October and
the 18th of November, it was reasonable to suppose that the French
would not commence offensive operations before the latter period,
and that ample time would be afforded to concentrate the English
troops under the protection of the Spanish armies. If sir John Moore
could have suspected the delusion under which the British
government acted, the incredible folly of the central junta and the
Spanish generals, or the inaccuracy of the military agents, if he
could have supposed that the Spanish armies were weak in
numbers, weaker in spirit, and destitute of food and clothing; or
that, while the Spanish authorities were pressing him to advance,
they would wantonly detain sir David Baird’s troops seventeen days
on board the transports; if he could have imagined all this,
undoubtedly his arrangements ought and would have been different;
his army would have been kept together, and the road through
Coria, however difficult, would have been preferred to a divided
march.
The dangerous and absurd position of the Spanish armies, and the
remote situation of the British troops in October, may be explained
by the annexed diagram. Lisbon being taken as a centre, and the
distance a between Lisbon and Coruña being the radius, let a circle
passing through Madrid be described. Let the tangential line c be
drawn perpendicular to the radius a, meeting the secant b at
Sanguessa.
The extreme right of the Spaniards was posted at Sanguessa.
Castaños was at Calahorra, and Blake was near Durango, but the
main body of the French was at Vittoria; and not only divided the
Spaniards, but was actually twenty-five miles nearer to Burgos and
Valladolid, (the points of concentration for Moore’s and Baird’s corps)
than either Castaños or Blake, and seventy-five miles nearer than
Palafox.
The 10th, the emperor struck the first blow, by beating Belvedere
and seizing Burgos. Sir David Baird did not march from Coruña until
the 12th, and did not bring up the whole of his troops to Astorga
before the 4th of December; hence it is clear, that whatever road the
artillery had taken, the British army could not have averted the ruin
of the Spaniards. Let us suppose the troops assembled at Salamanca
on the 13th of November. They must have advanced either to
Valladolid or to Madrid. If to Valladolid, the emperor was at Burgos
with the imperial guards, ten or twelve thousand cavalry, and a
hundred pieces of artillery. The first corps was within a day’s march,
the second and fourth corps within three marches, and the sixth
corps within two marches. Above a hundred thousand French
soldiers could, therefore, have been concentrated in three days; and
it is to be observed that sir John Moore never had twenty-five
thousand in the field.
It is said, he might have gone to Madrid: in that case the
separation of the artillery was a decided advantage, and the
separation of Baird’s corps (which was not the general’s
arrangement) was the error. The army could not have marched from
Salamanca to Madrid in less than seven days; on the 21st of
November then, twenty-four thousand British soldiers could have
been collected in the capital; but the fourth French corps, which
reached Segovia the 1st of December, would have cut off their
communication with Portugal, and the emperor with forty thousand
men was at Aranda de Douero. Castaños was defeated on the 23d;
the remnants of his army were only at Guadalaxara about the 1st of
December, and the sixth corps was in full pursuit of them. The
English general must then have done one of three things; advanced
to the relief of Castaños’s retreating army, joined St. Juan at the
Somosierra, or retreated across the Tagus. In the first case, the
emperor would have forced the Somosierra, and uniting with the
fourth corps, have placed sixty thousand men upon sir John Moore’s
rear; in the second case, the sixth and fourth corps, turning both his
flanks, would have effected a junction behind the Somosierra, and
cut him off from Madrid, while Napoleon, with forty thousand men,
assailed him in front. To retreat over the Tagus was to adopt the
southern provinces for a new base of operations, and might have
been useful if the Spaniards would have rallied round him with
enthusiasm and courage; but would they have done so when the
emperor was advancing with his enormous force? After-experience
proves that they would not. The duke of Dalmatia, in 1810, with an
army very inferior to that under Napoleon, reached the gates of
Cadiz without a serious blow being struck to oppose him, and at this
time the people of the south were reckless of the opportunity
procured for them by sir John Moore’s march on Sahagun; but, it has
been said, that twenty-four thousand British troops acting vigorously,
could have checked the emperor, and raised the courage of the
Spaniards. To such an observation I will oppose a fact. In 1815,
Napoleon crossed the Sambre with one hundred and fifteen
thousand men, and the two hundred and ten thousand regular
troops in his front, among which were more than thirty thousand
English, could with difficulty stop his progress after four days’
fighting, in three of which he was successful.
If sir John Moore, at a subsequent period, was willing to risk the
danger of a movement on the capital, it was because he was
misinformed of the French strength, and the Spaniards were
represented to be numerous and confident; he was also
unacquainted with the defeat at Tudela. His object was, by assisting
Castaños, to arouse the spirit of the patriots: and nothing more
strongly evinces his hardihood and prompt judgment, for, in his
letter to Mr. Frere, he distinctly stated the danger to be incurred, and
carefully separating the military from the political reasons, only
proposed to venture the army if the envoy was satisfied that the
Spanish government and people would answer to such an appeal,
and that the British cabinet would be willing to incur the risk for such
an object. If he did not follow up his own proposal, it was because
he had discovered that the army of Castaños was, not simply
defeated, but destroyed; because the Somosierra had been forced
by a charge of cavalry, and because the passes of the Guadarama,
on his line of march to Madrid, were seized by the enemy before his
own army could be concentrated.
Why then did he not retreat into Portugal? Because Napoleon,
having directed the mass of his forces against the capital, the British
army was enabled to concentrate; because Madrid shut her gates;
because Mr. Frere and the Spanish authorities deceived him by false
information; because the solemn declaration of the junta of Toledo,
that they would bury themselves under the ruins of that town rather
than surrender, joined to the fact that Zaragoza was fighting
heroically, seemed to guarantee the constancy and vigour of that
patriotic spirit which was apparently once more excited; because the
question was again become political, and it was necessary to satisfy
the English people, that nothing was left undone to aid a cause
which they had so much at heart; and, finally, because the peculiar
position of the French army at the moment, afforded the means of
creating a powerful diversion in favour of the southern provinces.
These are the unanswerable reasons for the advance towards
Sahagun. In the details of execution, that movement may be liable
to some trifling objections; perhaps it would have been better to
have carried the army on the 21st at once to Carrion and neglected
Sahagun and Saldanha; but in its stratagetical and political character
it was well conceived and well timed, hardy and successful.
The irritating interference that sir John Moore was called upon to
repel, and the treachery and the folly, equal in its effects to
treachery, that he was obliged to guard against, have been
sufficiently dwelt upon already; but before discussing the retreat
from Astorga, it may be of some military interest to show that the
line of Portugal, although the natural one for the British army to
retire upon, was not at this period necessarily either safe or useful,
and that greater evils than those incurred by a retreat through
Gallicia would probably have attended a retrograde march upon
Lisbon.
The rugged frontier of Portugal lying between the Douero and the
Tagus, is vulnerable in many points to an invading army of superior
force. It may be penetrated between the Douero and Pinhel, and
between Pinhel and Guarda, by roads leading into the valleys of the
Zezere and the Mondego. Between the Sierra de Estrella and the
Sierra de Gata, by the road from Alfayates to Sabugal and
Penamacor, or that by Guarda and Coria. Again, it may be pierced
between the Sierra de Gata and the Tagus by Idanha Velha, Castello
Branco, and Sobreira Formosa; and from the Tagus to the Guadiana,
a distance of about twenty leagues, the Alentejo presents an open
country without any strong fortress, save Lalippe, which may be
disregarded and passed without danger.
Sir John Moore commenced his forward movement from
Salamanca on the 12th of December, and at that period, the fourth
corps being at Talavera de la Reyna, was much nearer to Lisbon than
the British army was, and the emperor was preparing to march on
that capital with the sixth corps, the guards, and the reserve. He
could, as the duke of Berwick did, penetrate by both sides of the
Tagus, and what was to prevent him from reaching Lisbon before the
British force, if the latter had retreated from Salamanca? he marched
on a shorter line and a better road; he could supply his troops by
requisitions, a system that, however fatal it may be in the end, is
always advantageous at first. Sir John Moore must, from a scanty
military chest, have purchased his supplies from a suspicious
peasantry, rendered more distrustful by the retreat. In Lisbon, sir
John Craddock commanded six thousand infantry and two hundred
and fifty-eight cavalry; but the provisional government, who had
only organized a few ill-composed battalions, were so inactive, that it
was not until the 8th of December that a proclamation, calling on
the people to arm, was issued. In the arsenal there were scarcely
musquets and equipments for eight thousand men, and the new
levies were only required to assemble when Portugal should be
actually invaded. Sir Robert Wilson, indeed, having with great
activity organized about two thousand of the Lusitanian legion,
marched in the middle of December from Oporto; but this was all
that could be opposed to an army more numerous, more favourably
situated for invasion, and incomparably better commanded than that
with which Massena invaded the country in 1810. Thus it may be
affirmed, that if a retreat from Lisbon was advisable, before
Napoleon took Madrid, it was not a safe operation after that event,
and it is clear that sir John Moore neither lightly nor injudiciously
adopted the line of Gallicia.
The arguments of those who deny the necessity of falling back,
even behind the Esla, are scarcely worth notice; a simple reference
to the numbers under the emperor, and the direction of his march, is
sufficient to expose their futility; but the necessity of the continued,
and as it has been unjustly called, the precipitate retreat to Coruña,
may not be quite so obvious. The advance to Sahagun was intended
to create a diversion, and give the Spaniards an opportunity of
making head in the south; but although it succeeded in drawing
away the enemy, the Spaniards did not make any head. The central
junta displayed no energy or wisdom; a few slight demonstrations by
the marquis of Palacios, on the side of the Sierra Morena, and by the
duke of Infantado on the side of Cuenca, scarcely disturbed the first
corps which remained in La Mancha; ten thousand men were
sufficient to maintain Madrid in perfect tranquillity, and a part of the
fourth corps even marched from Talavera by Placentia on
Salamanca. By the letters of Mr. Stuart, and the reports of his own
spies, sir John Moore was informed of all these disheartening
circumstances; but the intelligence arrived slowly and at intervals,
and he, hoping that the Spaniards would finally make an effort,
announced his intention to hold the Gallicias; but Mr. Stuart’s
correspondence deprived him of that hope; and the presence of the
emperor, the great amount of his force, and the vehemence with
which he pressed forward, confirmed the unhappy truth that nothing
could be expected from the south.
Sir John Moore could not with twenty-three thousand men
maintain himself against the whole French army, and until he
reached Astorga his flanks were always exposed. From thence,
however, he retreated in comparative security; but the natural
strength of the country between that town and Coruña misled
persons of shallow judgment, who have since inconsiderately
advanced many vague accusations, such as that passes where a
hundred men could stop an army were lightly abandoned; that the
retreat was a flight, and the general’s judgment clouded by the
danger of his situation. There might be some foundation for such
observations if military commanders were like prize-fighters, bound
to strike always at the front; but as long as armies are dependent for
their subsistence and ammunition upon lines of communication, the
safety of their flanks and rear must be considered as of
consequence. Sir John Moore was perfectly aware that he could fight
any number of men in some of the mountainous positions on the
road to Coruña; but unless he could make a permanent defence,
such battles would have been worse than useless, and a permanent
defence was impossible, inasmuch as there were none but
temporary magazines nearer than Coruña, and there were neither
carriages of transport, nor money to procure them; a severe winter
had just set in, and the province being poor, and the peasantry
disinclined to aid the troops, few resources could be drawn from the
country itself, neither was there a single position between Astorga
and Coruña which could be maintained for more than a few days
against a superior force, for that of Rodrigatos could Appendix, No. 13,
be turned by the old road leading to Villa Franca, Villa sect. 2.
Franca itself by the valley of the Syl, and from thence the whole line
to Coruña might be turned by the road of Orense, which also led
directly to Vigo, and until he reached Nogales, sir J. Moore’s
intention was to retire to Vigo. The French could have marched
through the richest part of Gallicia to St. Jago and Coruña on the
left, or from the Asturias, by the way of Mondonedo, on the right. If
it be asked why they did not do so? the answer is prompt. The
emperor having quitted the army, the jealousies and
misunderstandings usual between generals of equal rank impeded
the operations. A coolness subsisted between marshal Ney and the
duke of Dalmatia, and without entering into the grounds of their
difference it is plain that, in a military point of view, the judgment of
the latter was the soundest. The former committed a great error by
remaining at Villa Franca instead of pushing his corps, or a part of it,
(as recommended by Soult) along the valley of Orense to St. Jago de
Compostella. The British army would have been lost if the sixth
corps had reached Coruña before it; and what would have been the
chances in the battle if three additional French divisions had been
engaged?
Granting, therefore, that the troops could have been nourished
during the winter, Villa Franca, Nogales, Constantino, and Lugo,
were not permanently defensible by an army whose base of
operations was at Coruña. Hence it was that sir John Moore resolved
to regain his ships with the view to renew the war in the south, and
Hannibal himself could have done no more. Nor was the mode of
executing the retreat at all unbecoming the character of an able
officer.
Lord Bacon observes, that “honourable retreats are no ways
inferior to brave charges, as having less of fortune, more of
discipline, and as much of valour.” That is an honourable retreat in
which the retiring general loses no trophies in fight, sustains every
charge without being broken, and finally, after a severe action, re-
embarks his army in the face of a superior enemy without being
seriously molested. It would be honourable to effect this before a
foe only formidable from numbers, but it is infinitely more creditable,
when the commander, while struggling with bad weather and worse
fortune, has to oppose veterans with inexperienced troops, and to
contend against an antagonist of eminent ability, who scarcely
suffers a single advantage to escape him during his long and
vigorous pursuit. All this sir John Moore did, and finished his work by
a death as firm and glorious as any that antiquity can boast of.
 Put to lord Bacon’s test, in what shall the retreat to Coruña be
found deficient? something in discipline perhaps, but that fault does
not attach to the general. Those commanders who have been
celebrated for making fine retreats were in most instances well
acquainted with their armies; and Hannibal, speaking of the elder
Scipio, derided him, although a brave and skilful man, for that, being
unknown to his own soldiers, he should presume to oppose himself
to a general who could call to each man under his command by
name: thus inculcating, that, unless troops be trained in the peculiar
method of a commander, the latter can scarcely achieve any thing
great. Now sir John Moore had a young army suddenly placed under
his guidance, and it was scarcely united, when the superior numbers
of the enemy forced it to a retrograde movement under very
harassing circumstances; he had not time, therefore, to establish a
system of discipline, and it is in the leading events, not the minor
details, that the just criterion of his merits is to be sought for.
Was the retreat uncalled for? Was it unnecessarily precipitate?
Was any opportunity of crippling the enemy lost? Was any weakness
to be discovered in the personal character of the general? These are
the questions that sensible men will ask; the first has been already
examined, the second is a matter of simple calculation. The rear
guard quitted Astorga on the 1st of January; on the 3rd, it repulsed
the enemy in a sharp skirmish at Calcabelos; the 6th it rejoined the
main body at Lugo, having three times checked the pursuers during
the march. It was unbroken, and lost no gun, suffered no
misfortune; the whole army offered battle at Lugo for two successive
days, it was not accepted, and the retreat recommencing, the troops
reached Betanzos on the morning of the 10th, and Coruña on the
11th; thus in eleven days, three of which were days of rest, a small
army passed over a hundred and fifty miles of good road. Now
Napoleon, with fifty thousand men, left Madrid on the 22d of
December, the 28th he was at Villapando, having performed a
march, on bad roads, of a hundred and sixty-four miles in seven
days. The retreat to Coruña was consequently not precipitate, unless
it can be shown, that it was unnecessary to retreat at all beyond
Villa Franca, neither can it be asserted, that any opportunity of
crippling the enemy was lost. To fight a battle was the game of the
French marshal, and if any censure will apply to his able campaign,
it is that he delayed to attack at Lugo; victorious or beaten, the
embarrassments of his adversary must have been increased; sir
John Moore must have continued his retreat encumbered with the
wounded, or the latter must have been abandoned without succour
in the midst of winter.
At Coruña the absence of the fleet necessarily brought on a battle;
that it was honourable to the British troops is clear from the fact that
they embarked without loss after the action; and that it was
absolutely necessary to embark notwithstanding the success, is as
certain a proof how little advantage could have been derived from
any battle fought farther inland, and how prudently sir John Moore
acted in declining an action the moment he had rallied his army at
Lugo, and restored that discipline which the previous movements
had shaken; but, notwithstanding the clamour with which this
campaign has been assailed, as if no army had ever yet suffered
such misfortunes, it is certain that the nominal loss was small, the
real loss smaller, and that it sinks into nothing when compared with
the advantages gained. An army which, after Appendix, No. 26.
marching in advance or retreat above five hundred Ibid.
miles before an enemy of immensely superior force, has only lost,
including those killed in battle, four thousand men, or a sixth part of
its numbers, cannot be said to have suffered severely, nor would the
loss have been so great but for the intervention of the accidental
occurrences mentioned in the narrative. Night marches are seldom
happy; that from Lugo to Betanzos cost the army in stragglers more
than double the number of men lost in all the preceding operations;
nevertheless the reserve in that, as in all the other movements,
suffered little; and it is a fact, that the light brigades detached by the
Vigo road, which were not pursued, made no forced marches, slept
under cover, and were well supplied, left, in proportion to their
strength, as many men behind as any other part of the army; thus
accumulating proof upon proof that inexperience was the primary
and principal cause of the disorders which attended the retreat.
Those disorders were sufficiently great, but many circumstances
contributed to produce an appearance of suffering and
disorganization which was not real. The intention of sir John Moore
was, to have proceeded to Vigo, in order to restore order before he
sailed for England: instead of which the fleet steered home directly
from Coruña; a terrible storm scattered it; many ships were
wrecked, and the remainder, driving up the channel, were glad to
put into any port. The soldiers, thus thrown on shore, were spread
from the Land’s End to Dover. Their haggard appearance, ragged
clothing, and dirty accoutrements, things common enough in war,
struck a people only used to the daintiness of parade, with surprise;
the usual exaggerations of men just escaped from perils and
distresses were increased by the uncertainty in which all were as to
the fate of their comrades; a deadly fever, the result of anxiety, and
of the sudden change from fatigue to the confinement of a ship,
filled the hospitals at every port with officers and soldiers, and thus
the miserable state of sir John Moore’s army became the topic of
every letter, and a theme for every country newspaper along the
coast. The nation, at that time unused to great operations, forgot
that war is not a harmless game, and judging of the loss positively,
instead of comparatively, was thus disposed to believe the calumnies
of interested men, who were eager to cast a shade over one of the
brightest characters that ever adorned the country. Those calumnies
triumphed for a moment; but Moore’s last appeal to his country for
justice will be successful. Posterity, revering and cherishing his
name, will visit such of his odious calumniators as are not too
contemptible to be remembered with a just and severe retribution;
for thus it is that time freshens the beauty of virtue and withers the
efforts of baseness; and if authority be sought for in a case where
reason speaks so plainly, future historians will not fail to remark, that
the man whose talents exacted the praises of Soult, of Wellington,
and of Napoleon, could be no ordinary soldier.
“Sir John Moore,” says the first, “took every advantage that the
country afforded to oppose an active and vigorous Appendix, No. 16.
resistance, and he finished, by dying in a combat that must do credit
to his memory.”
Napoleon more than once affirmed, that if he Vivian’s
committed a few trifling errors they were to be Conversations at
Elba.
attributed to his peculiar situation, for that his talents
and firmness alone had saved the English army from Voice from St.
destruction. Helena

“In sir John Moore’s campaign,” said the duke of Wellington, “I can
see but one error; when he advanced to Sahagun he should have
considered it as a movement of retreat, and sent officers to the rear
to mark and prepare the halting-places for every brigade; but this
opinion I have formed after long experience of war, and especially of
the peculiarities of a Spanish war, which must have been seen to be
understood; finally, it is an opinion formed after the event.”
 APPENDIX.

APPENDIX.

The following five Notes, dictated by the emperor Napoleon, and signed by general Bertrand, were
found in king Joseph’s portfolio, at the battle of Vittoria.

N o. I .

OBSERVATIONS ADDRESSÉES AU GÉNÉRAL SAVARY SUR LES AFFAIRES D’ESPAGNE.

Le 13 Juillet, 1808.
1ere Observation.—Les affaires des Français en Espagne seroient dans une excellente
position si la division Gobert avait marché sur Valladolid, et si la division Frère eut occupé
San Clemente, ayant une colonne mobile à trois ou quatre journées sur la route du général
Dupont.
Le gal Gobert ayant été dirigé sur le général Dupont, le gal Frère étant avec le maréchal
Moncey, harassé et affaibli par des marches et des contremarches, la position de l’armée
Française est devenue moins belle.
2e Observation.—Le maréchal Bessières est aujourd’hui à Medina del Rio Secco avec 15
mille hommes, infanterie, cavalerie, artillerie. Le 15 ou le 16, il attaquera Bénavente, se
mettra en communication avec le Portugal, jettera les rebelles en Galice, et s’emparera de
Léon. Si toutes les opérations réussissent ainsi, et d’une manière brillante, la position de
l’armée Française redeviendra ce qu’elle était.
Si le général Cuesta se retire de Bénavente sans combattre, il se retirait sur Zamora,
Salamanque, pour venir gagner Avila et Segovia, certain qu’alors le maréchal Bessières ne
pourrait point le poursuivre, puisque, dans cette supposition, il serait menacé par l’armée de
Galice, dont l’avant-garde est réunie à Léon.
Alors il faut que le général qui commande à Madrid puisse promptement réunir 6 à 7000
hommes pour marcher sur le général Cuesta. Il faut que la citadelle de Ségovie soit occupée
par quelques pièces de canon, trois à 400 convalescens, avec six semaines de biscuit.
C’est une grande faute de n’avoir pas occupé cette citadelle, quand le major-général l’a
mandé. De toutes les positions possibles, Ségovie est la plus dangereuse pour l’armée:
capitale d’une province, assise entre les deux routes, elle ôterait a l’armée toutes ses
communications, et l’ennemi une fois posté dans cette citadelle, l’armée Française ne
pourrait plus l’en déloger. Trois ou 400 convalescens et un bon chef de bataillon, une
escouade d’artillerie, rendront le château de Ségovie imprégnable pendant bien de tems, et
assureront à l’armée l’importante position de Ségovie.
Si le général Cuesta se jette en Galice, sans combattre, sans éprouver de défaite, la
position de l’armée devient toujours meilleure; à plus forte raison, s’il est jetté en Galice
après avoir éprouvé une forte défaite.
3e Observation.—Si le maréchal Bessières, arrivé devant Bénavente, reste en présence
sans attaquer le gal Cuesta, ou s’il est repoussé, son but sera toujours de couvrir Burgos, en
tenant le plus possible l’ennemi en échec; il peut être renforcé de 3000 hommes de troupes
de ligne, qui accompagnent le roi, mais alors il n’y a point à hésiter. Si le maréchal Bessières
a fait une marche rétrograde sans bataille, il faut sur le champ lui envoyer 6000 hommes de
renforts. S’il a fait son mouvement après une bataille, où il ait éprouvé de grandes pertes, il
faudra faire de grandes dispositions: rappeller à marche forcée sur Madrid le gal Frère, le gal
Caulaincourt, le gal Gobert, le gal Vedel, et laisser le gal Dupont sur les montagnes de la
Sierra Morena, ou le rapprocher même de Madrid, en le tenant toujours, cependant, à sept
ou huit marches, afin de pouvoir écraser le gal Cuesta et toute l’armée de Galice, pendant
que le gal Dupont servira d’avant-garde pour tenir l’armée d’Andalousie en échec.
4e Observation.—Si le général Dupont éprouvait un échec, cela serait de peu de
conséquence. Il n’aurait d’autre résultat que de lui faire repasser les montagnes; mais le
coup qui serait porté au maréchal Bessières serait un coup porté au cœur de l’armée, qui
donnerait le tetanos, et qui se ferait sentir à toutes les points extrêmes de l’armée. Voilà
pourquoi il est très malheureux que toutes les dispositions ordonnées n’aient pas été suivis.
L’armée du maréchal Bessières devrait se trouver avoir au moins huite mille hommes de plus,
afin qu’il n’y eut aucune espèce de chance contre l’armée du maréchal Bessières.
La vraie manière de renforcer le général Dupont, ce n’est pas de lui envoyer des troupes,
mais c’est d’envoyer des troupes au maréchal Bessières. Le général Dupont et le général
Vedel sont suffisans pour se maintenir dans les positions qu’ils ont retranchés; et si le
maréchal Bessières avait été renforcé, et l’armée de Galice écrasée, le général Dupont
immédiatement après se trouverait dans la meilleure position, non seulement par des forces
qu’on pourrait alors lui envoyer, mais encore par la situation morale des affaires. Il n’y a pas
un habitant de Madrid, pas un paysan des vallées qui ne sente que toutes les affaires
d’Espagne aujourd’hui sont dans l’affaire du maréchal Bessières. Combien n’est-il pas
malheureux que dans cette grande affaire on se soit donné volontairement 20 chances
contre soi.
5e Observation.—L’affaire de Valence n’a jamais été d’aucun considération. Le maréchal
Moncey seul était suffisant. C’était une folie que de songer à le secourir. Si le mal Moncey ne
pouvait pas prendre Valence, 20 mille hommes de plus ne le lui auraient pas fait prendre,
parcequ’alors c’était un affaire d’artillerie, et non une affaire d’hommes: car on ne prend pas
d’un coup de collier une ville de 80 ou 100 mille âmes, qui a barricadé ses rues, mis de
l’artillerie à toutes les portes et dans toutes les maisons. Or, dans cette hypothèse, le mal
Moncey était suffisant pour former une colonne mobile, faire face à l’armée de Valence, et
faire sentir dans toute leur force les horreurs de la guerre.
 Le gal Frère ne pouvait donc rien pour faire prendre Valence, et le gal Frère pouvait
beaucoup posté à San Clemente, soit qu’il dût revenir à Madrid, soit qu’il dût prendre une
position intermédiaire pour secourir le gal Dupont.
C’était une autre erreur que de songer à faire aller le mal Moncey à Valence pour ensuite le
faire marcher en Murcie et sur Grenade. C’était vouloir fondre ce corps d’armée en détail et
sans fruit. Comme le dit fort bien le général Dupont, il valait mieux lui envoyer directement
un régiment que de lui envoyer trois dans cette direction là.
Dans les guerres civiles ce sont les points importans qu’il faut garder: il ne faut pas aller
partout. Si cependant on a dirigé le mal Moncey sur Valence, c’était à une époque où la
situation des affaires n’était pas la même; c’était lorsque l’armée de Valence pouvait envoyer
en Catalogne ou à Saragosse comme elle en menaçait.
6e Observation.—Le but de tous les efforts de l’armée doit être de conserver Madrid. C’est
là qu’est tout. Madrid ne peut être menacé que par l’armée de Galice. Elle peut l’être aussi
par l’armée de l’Andalousie, mais d’une manière beaucoup moins dangereux, parcequ’elle est
simple et directe, et que par toutes les marches que fait le gal Dupont sur ses derrières, il se
renforce. Les généraux Dupont et Vedel étaient suffisans, ayant plus de 20,000 hommes; le
mal Bessières ne l’est pas proportionnellement, vû que sa position est plus dangereuse. Un
échec que recevrait le gal Dupont serait peu de chose; un échec que recevrait le mal
Bessières serait plus considérable et se ferait sentir à l’extremité de la ligne.
Résumé.—Faire reposer et rapprocher de Madrid le gal Frère, le gal Caulaincourt, le gal
Gobert, afin qu’ils puissent arriver à Madrid avant le gal Cuesta, si celui-ci battait le mal
Bessières. Immédiatement après l’événement qui aura lieu le 15 ou le 16, prendre un part
selon les événemens qui auront eu lieu, et dans le but d’écraser l’armée ennemie en Galice.
Si le maréchal Bessières a eu grand succès, sans éprouver de grandes pertes, tout sera
bien dans la direction actuelle. S’il a un succès après avoir éprouvé beaucoup de pertes, il
faut se mettre en mésure de le renforcer. S’il se tient en observation sans attaquer, il faut le
renforcer. S’il a été défait et bien battu, il faut se concentrer et rassembler toutes ses troupes
dans le cercle de sept ou huit journées de Madrid, et étudier les dispositions dans les
différentes directions pour savoir où placer les avant-gardes, afin de profiter de l’avantage
qu’on a d’être au milieu, pour écraser successivement avec toutes ses forces les divers corps
de l’ennemi. Si on n’ordonne pas sur le champ au gal Dupont de repasser les montagnes,
c’est qu’on espère que malgré la faute faite, le mal Bessières a la confiance (qu’on partage)
qu’à la rigueur il est suffisant pour écraser l’ennemi. Le mal Bessières a eu le bon esprit de
tellement réunir toutes ses forces, qu’il n’a pas même laissé un seul homme à St. Ander.
Quelqu’ avantage qu’il y eut à laisser là un millier d’hommes, il a senti qu’un millier
d’hommes pouvait décider sa victoire.
Quant à la division du gal Verdier devant Saragosse, elle a rempli aux trois quarts son but.
Elle a désorganisé tous les Arragoniens, a porté le découragement parmi eux, les a reduits à
défendre les maisons de leur capitale, a soumis tous les environs, a bloqué la ville, et réuni
tous les moyens pour s’en emparer sans que cela devienne trop conteux.
Voilà l’esprit de la guerre d’Espagne.
 Dictated by the emperor Napoleon. Taken at the battle of Vittoria.

N o. I I .

NOTE POUR LE ROI D’ESPAGNE.

Bayonne, le Juillet, 1808.
L’armée d’Espagne a son quartier général à Madrid; voici sa composition actuelle:
1º. Corps des Pyrénées Occidentales.
Le maréchal Bessières commande le corps des Pyrénées Occidentales, qui est fort de 23
mille hommes, infanterie, cavalerie, artillerie, occupe la place de St. Sebastian, les trois
Biscays, les montagnes de St. Ander, la place de Burgos, et est chargée de combattre l’armée
ennemie des Asturies et de Galice.
Toutes les troupes sont en mouvement pour composer l’armée de la manière suivante.
┌ ┌ le 4e regt d’infanterie legère ┐
ere
│ 1 brigade, │ 15e d’infanterie de ligne │
al
│ le g Reynaud. └ 1er baton de Paris en marche │
total 3000 hom. présens sous les armes, et 6 pièces de
Division │ │
canon, ci 3000 hes
du gal ┤ (Cette brigade marche sur Bénévent.) ├ 5100 hes
Mouton. │ 2e brigade, { 2e regt d’infanterie legère │
│ le gal Rey. { 12e idem │
│ total 2100 hommes et 6 pièces de canon, ci 2100 │
(Cette brigade est à Burgos avec le roi, et doit joindre sa
└ ┘
division.)
——
À reporter 5100 hes
 De l’autre part 5100 hes
┌ Brigade d’Armagnac 1800 ┐
Division │ Brigade Gaulois 1800 │
du gal │ Brigade Sabathier 2800 │ 8400 hes
Merle. └ Brigade Ducos 2000 ┘
——
Total 8400 hes
et 16 pièces de canon.
┌ Infanterie
Garde. 1900 hes
└ et 6 pièces de canon.
(Toutes ces troupes marchent sur Bénévent.)

┌ 10e de chasseurs 450 ┐

│ e 450 │
22 id.
│ Garde 300 │
│ (Ces troupes marchent sur Bénévent.) │
│ Escadrons de dragons 200 │
(Ces escadrons sont en marche et ont
Cavalerie. ┤ ├ 1950 hes
dépassé la frontière.
│ 26e de chasseurs 450 │
(Arrivant à Bayonne sous peu de
│ │
jours.)
│ —— │
│ Total de la cavalerie 1950 hes │
└ ==== ┘

Les forces actives du maréchal Bessières sont donc de 17,000 hes. Il

n’en a guère que 15,000 pour l’affaire de Bénévent.

S’il obtenait à Bénévent et à Léon un grand succès contre l’armée de

Galice, peut-être serait-il convenable pour profiter de la victoire et de la
terreur des premiers moments de se jetter dans la Galice. Toutesfois, il
devrait d’abord prendre position à Léon, en s’emparant de la plaine,
jettant l’ennemi dans les montagnes, et interceptant au moins à Astorga la
communication de la grande route.

Garnison de Burgos.—Il y a dans le château de Burgos une garnison de

600 hes
dépôt[28]
———
À reporter 17,950 hes
———
 De l’autre part 17,950 hes
Colonne du général Bonnet.—Il y a encore à Burgos le gal de
division Bonnet, faisant partie du corps du mal Bessières: ce
gal va avoir sous ses ordres une colonne mobile de 1200 ┐
hommes, pour maintenir la tranquillité dans la ville et ses
environs. Cette colonne est composée comme il suit:
4 bataillon du 118e, formant
e
450 hes │
(Actuellement existant à Burgos.) │
3e bataillon du dépôt gal actuellement à Vittoria 450 │ 1500 hes
2 compies du 4e d’infanterie legère, formant un petit bataillon 400 ├
(En marche, ayant passé la frontière.) │
—— │
1300 hes │
Escadron de dragons (en marche) 200 │
2 pièces de canon en marche │
—— │
1500 hes │
==== ┘

Colonne d’Aranda.—Cette colonne, formée du 1er bataillon de marche, fort de

1000 hes et de 4 pièces de canon, peut se réunir au bésoin avec la colonne
1000 hes
du gal Bonnet: elles doivent assurer la communication jusqu’aux montagnes
en avant d’Aranda, ci
Colonne de Vittoria.—Le général de brigade Monthion, et le colonel Barerre,
occupent Vittoria avec une colonne composée comme il suit:
2 compagnies du 15e de ligne, formant un petit bataillon de 300 hes
Le 2e baton du 12e d’infanterie legère 600
Le 2e baton du 2e id. 600
——
(Ce qui fait en infanterie) 1500 hes
1 escadron de dragons (en marche) 200
2 pièces de canon.
——
(Tous ces corps sont en marche) 1700 ci 1700
———
À reporter 22,150 hes
———
 De l’autre part 22,150 hes
Garnison de St. Sebastian.—Le général Thouvenot commande à St. Sebastian
1000
avec mille hommes de garnison, ci
———
al 23,150
Recapitulation.—Le corps du m Bessières est de
Et 36 pièces de canon. ======
Les détachemens et troisièmes bataillons des corps qui sont aux divisions actives du mal
Bessières pourront sous 15 jours le réjoindre, vû qu’ils seront remplacés à Vittoria et à
Burgos par d’autres corps.
2º. Arragon.
Jusqu’à cette heure les troupes qui sont en Arragon, faisaient partie du corps des Pyrénées
Occidentales. Mais le corps des Pyrénées Occidentales se portant sur la Galice, il devient
indispensable d’en faire une division à part.
Aujourd’hui, ce commandement comprend Pampelune, la Navarre, et les troupes qui
forment le siège de Saragosse, sous les ordres du général Verdier.
Ces troupes sont divisées en quatre brigades, et sont composées ainsi qu’il suit:

3 regimens d’infanterie de ligne de la Vistule, ayant sous les armes 3600 hes
Les 4e, 6e et 7e bataillons de marche 1500
Le 3e bataillon du 14e provisoire 1300
Le 1er regiment supplementaire 900
Les 47e, 15e et 70e 1600
Un bataillon des gardes nationales d’élite 600
——
Total. 9500 hes
La cavalerie consiste dans un regiment de lanciers
700 } 1100
Polonais
Plus un escadron de marche 400 }
al
À Pampelune le g Dagout commande. Indépendamment d’une dépôt de 800
hommes, formant 800 la garnison de la citadelle; il a une colonne mobile
composée du 1er bataillon de marche du Portugal, du troisième bataillon du
(ci 800
118e, fort de 650 hommes, et d’un escadron de dragons, ce qui forme un
total de 1400 hommes disposibles pour se porter sur tous les points de la
Navarre, et sur
———
À reporter 11,400
 De l’autre part 11,400
Et les communications de Saragosse, pour y mettre l’ordre: ci 1,400
Artillerie 200
———
Il y a donc en cernement en Arragon et en Navarre 13,000
======
Aussitôt que Saragosse sera pris, et que le corps de l’Arragon sera constitué, il sera
nécessaire de fairer entrer au corps du marl Bessières le bataillon du 47e, celui du 15e, et les
trois bataillons du 14e provisoire; ce qui augmentera le mal Bessières de deux mille hommes,
afin de tenir les corps réunis. Il est possible qu’on fasse partir de Bayonne le 19,300 hommes
de bonnes troupes de ligne, pour se diriger sur Saragosse et enlever la prise de cette place,
si toutefois elle n’est pas encore prise.
Si Saragosse était pris, le corps du mal Bessières pourrait être renforcé de ces trois mille
hommes d’élite et de 2000 hommes du corps de Saragosse, ce qui lui ferait un corps
nombreux pour la campagne de Galice.
Indépendamment de Saragosse, les rebelles occupent la ville de Jaca et plusieurs points
dans les vallées. À tous les débouchés des vallées en France il y a un général de brigade
avec une colonne mobile. On attendra la prise de Saragosse pour entrer dans ces vallées et y
marcher dans les deux sens. En général l’esprit des vallées est bon; mais des troupes de
contrebandiers que les chefs des rebelles ont enrégimentés les vexent.
3º. Catalogne.
Le général Duhesme occupe Barcelone, qui est une place qui a deux très belles
forteresses, qui la dominent. C’est la plus grande ville de la monarchie.
Le général Duhesme a deux divisions, la division Chabran et la division Lechi, formant
11,000 hes d’infanterie, 1600 hes de cavalerie et 18 pièces de canon.
Le général Duhesme a eu plusieurs événemens; il a brulé un grand nombre de villages, et
maintenu en respect le pays à 15 lieues à la ronde.
La ville de Géronne, n’ayant pas été occupée, les insurgés de la Catalogne ont établis là
leur Junte, d’où ils donnent le mouvement au reste de la province. 2000 insurgés assiégeaint
le fort de Figuéras. On y avait heureusement laissé 300 Français: ils ont été obligés de tirer
beaucoup de coups de canon et de bruler le village.
Le gal de division Reille, avec deux bataillons Toscans; a marché sur Figuéras, l’a débloqué,
le 6 du mois, et y a fait entrer une grande quantité de vivres, dont on manquait. Le 10, il
réunissait sa division, qui arrivait de divers points de la France; il avait déjà 6000 hommes, et
il doit avoir aujourd’hui 9000 hes; il doit s’assurer de Roses et marcher sur Géronne, établir
ses communications avec le général Duhesme et ensemble pacifier la Catalogne.

Les forces réunies des généraux Duhesme et Reille s’élevent donc à 22,000 hes
Ainsi le corps des Pyrénées Occidentales est fort de 23,000
Celui d’Arragon, de 13,000
Celui de Catalogne, de 22,000
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookgate.com