ENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS

EDITED BY G.-C ROTA

Editorial Board
R. S. Doran, M. Ismail, T.-Y. Lam, E. Lutwak, R. Spigler

Volume 76

Geometry of Sporadic Groups I

Petersen and tilde geometries
 ENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS
4 W. Miller, Jr. Symmetry and separation of variables
6 H. Mine Permanents
11 W. B. Jones and W. J. Thron Continued fractions
12 N. F. G. Martin and J. W. England Mathematical theory of entropy
18 H. O. Fattorini The Cauchy problem
19 G. G. Lorentz, K. Jetter and S. D. Riemenschneider Birkhoff interpolation
21 W. T. Tutte Graph theory
22 J. R. Bastida Field extensions and Galois theory
23 J. R. Cannon The one-dimensional heat equation
25 A. Salomaa Computation and automata
26 N. White (ed.) Theory ofmatroids
27 N. H. Bingham, C. M. Goldie and J. L. Teugels Regular variation
28 P. P. Petrushev and V. A. Popov Rational approximation of real functions
29 N. White (ed.) Combinatorial geometries
30 M. Pohst and H. Zassenhaus Algorithmic algebraic number theory
31 J. Aczel and J. Dhombres Functional equations containing several variables
32 M. Kuczma, B. Chozewski and R. Ger Iterative functional equations
33 R. V. Ambartzumian Factorization calculus and geometric probability
34 G. Gripenberg, S.-O. Londen and O. Staffans Volterra integral and
functional equations
35 G. Gasper and M. Rahman Basic hypergeometric series
36 E. Torgersen Comparison of statistical experiments
37 A. Neumaier Intervals methods for systems of equations
38 N. Korneichuk Exact constants in approximation theory
39 R. A. Brualdi and H. J. Ryser Combinatorial matrix theory
40 N. White (ed.) Matroid applications
41 S. Sakai Operator algebras in dynamical systems
42 W. Hodges Model theory
43 H. Stahl and V. Totik General orthogonal polynomials
44 R. Schneider Convex bodies
45 G. Da Prato and J. Zabczyk Stochastic equations in infinite dimensions
46 A. Bjorner, M. Las Vergnas, B. Sturmfels, N. White and G. Ziegler Oriented
matroids
47 E. A. Edgar and L. Sucheston Stopping times and directed processes
48 C. Sims Computation with finitely presented groups
49 T. Palmer Banach algebras and the general theory of *-algebras
50 F. Borceux Handbook of categorical algebra I
51 F. Borceux Handbook of categorical algebra II
52 F. Borceux Handbook of categorical algebra III
54 A. Katok and B. Hassleblatt Introduction to the modern theory of dynamical
systems
55 V. N. Sachkov Combinatorial methods in discrete mathematics
56 V. N. Sachkov Probabilistic methods in discrete mathematics
57 P. M. Cohn Skew Fields
58 Richard J. Gardner Geometric tomography
59 George A. Baker, Jr. and Peter Graves-Morris Pade approximants
60 Jan Krajicek Bounded arithmetic, propositional logic, and complex theory
61 H. Gromer Geometric applications of Fourier series and spherical harmonics
62 H. O. Fattorini Infinite dimensional optimization and control theory
63 A. C. Thompson Minkowski geometry
64 R. B. Bapat and T. E. S. Raghavan Nonnegative matrices and applications
65 K. Engel Sperner theory
66 D. Cvetkovic, P. Rowlinson and S. Simic Eigenspaces of graphs
67 F. Bergeron, G. Labelle and P. Leroux Combinatorial species and tree-like
structures
68 R. Goodman and N. Wallach Representations of the classical groups
70 A. Pietsch and J. Wenzel Orthonormal systems and Banach space geomery
ENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS

Geometry of Sporadic Groups I

Petersen and tilde geometries

A. A. IVANOV
Imperial College, London

CAMBRIDGE
UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge, United Kingdom
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge, CB2 2RU, UK www.cup.cam.ac.uk
40 West 20th Street, New York, NY 10011-4211, USA www.cup.org
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
Ruiz de Alarcon 13, 28014 Madrid, Spain

© Cambridge University Press 1999

This book is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.

First published 1999

A catalogue record of this book is available from the British Library

Library of Congress Cataloguing in Publication data

Ivanov, A. A.
Geometry of sporadic groups 1, Petersen and tilde geometries / A.A. Ivanov.
p. cm.
Includes bibliographical references and index.
Contents: v. 1. Petersen and tilde geometries
ISBN 0 521 41362 1 (v. 1 : hb)
1. Sporadic groups (Mathematics). I. Title.
QA177.I93 1999
512'.2-dc21 98-45455 CIP

ISBN 0 521 41362 1 hardback

Transferred to digital printing 2002

 Contents

Preface page ix
1 Introduction 1
1.1 Basic definitions 2
1.2 Morphisms of geometries 5
1.3 Amalgams 7
1.4 Geometrical amalgams 9
1.5 Universal completions and covers 10
1.6 Tits geometries 11
1.7 ^4/£7-geometry 16
1.8 Symplectic geometries over GF(2) 17
1.9 From classical to sporadic geometries 19
1.10 The main results 21
1.11 Representations of geometries 23
1.12 The stages of classification 26
1.13 Consequences and development 33
1.14 Terminology and notation 42
2 Mathieu groups 49
2.1 The Golay code 50
2.2 Constructing a Golay code 51
2.3 The Steiner system 5(5,8,24) 53
2.4 Linear groups 56
2.5 The quad of order (2,2) 59
2.6 The rank 2 T-geometry 62
2.7 The projective plane of order 4 64
2.8 Uniqueness of 5(5,8,24) 71
2.9 Large Mathieu groups 74
2.10 Some further subgroups of Mat24 76
vi Contents
2.11 Little Mathieu groups 81
2.12 Fixed points of a 3-element 85
2.13 Some odd order subgroups in Mat24 87
2.14 Involutions in Mat^ 90
2.15 Golay code and Todd modules 95
2.16 The quad of order (3,9) 97
3 Geometry of Mathieu groups 100
3.1 Extensions of planes 101
3.2 Maximal parabolic geometry of MatiA 102
3.3 Minimal parabolic geometry of Mat24 106
3.4 Petersen geometries of the Mathieu groups 112
3.5 The universal cover of ^(Mat2i) 117
3.6 &(Mat23) is 2-simply connected 122
3.7 Diagrams for Jf (Mat24) 124
3.8 More on Golay code and Todd modules 130
3.9 Diagrams for Jf(Mat22) 132
3.10 Actions on the sextets 138
4 Conway groups 141
4.1 Lattices and codes 141
4.2 Some automorphisms of lattices 147
4.3 The uniqueness of the Leech lattice 150
4.4 Coordinates for Leech vectors 153
4.5 Cou C02 and C03 158
4.6 The action of Co\ on A4 160
4.7 The Leech graph 163
4.8 The centralizer of an involution 169
4.9 Geometries of Co\ and C02 173
4.10 The affine Leech graph 178
4.11 The diagram of A 189
4.12 The simple connectedness of ^{Co2) and ^{Co\) 193
4.13 McL geometry 198
4.14 Geometries of 3 • 1/4(3) 203
5 The Monster 210
5.1 Basic properties 211
5.2 The tilde geometry of the Monster 216
5.3 The maximal parabolic geometry 218
5.4 Towards the Baby Monster 222
2
5.5 £6(2)-subgeometry 224
5.6 Towards the Fischer group M(24) 227
 Contents vii
5.7 Identifying M(24) 231
5.8 Fischer groups and their properties 236
5.9 Geometry of the Held group 242
5.10 The Baby Monster graph 244
5.11 The simple connectedness of &(BM) 256
5.12 The second Monster graph 259
5.13 Uniqueness of the Monster amalgam 265
5.14 On existence and uniqueness of the Monster 268
5.15 The simple connectedness of &(M) 271
6 From Cn- to Tn-geometries 272
6.1 On induced modules 273
6.2 A characterization of 0(3 • Sp4(2)) 276
6.3 Dual polar graphs 280
6.4 Embedding the symplectic amalgam 285
6.5 Constructing T -geometries 288
6.6 The rank 3 case 290
6.7 Identification of J(n) 293
6.8 A special class of subgroups in J(n) 295
6.9 The f(ri) are 2-simply connected 297
6.10 A characterization of f(ri) 301
6.11 No tilde analogues of the ^/tygeometry 303
7 2-Covers of P -geometries 307
7.1 On P -geometries 307
7.2 A sufficient condition 313
7.3 Non-split extensions 315
7.4 ^(3 23 • Co2) 318
7.5 The rank 5 case: bounding the kernel 321
7.6 ^(3 4371 • BM) 327
7.7 Some further s-coverings 330
8 7-groups 332
8.1 Some history 333
8.2 The 26-node theorem 335
8.3 From Y-groups to 7-graphs 337
8.4 Some orthogonal groups 340
8.5 Fischer groups as 7-groups 345
8.6 The monsters 351
9 Locally projective graphs 358
9.1 Groups acting on graphs 359
viii Contents
9.2 Classical examples 362
9.3 Locally projective lines 367
9.4 Main types 370
9.5 Geometrical subgraphs 374
9.6 Further properties of geometrical subgraphs 379
9.7 The structure of P 383
9.8 Complete families of geometrical subgraphs 386
9.9 Graphs of small girth 389
9.10 Projective geometries 392
9.11 Petersen geometries 394
Bibliography 398
Index 406
 Preface

Sporadic simple groups are the most fascinating objects in modern alge-
bra. The discovery of these groups and especially of the Monster is
considered to be one of the most important contributions of the classi-
fication of finite simple groups to mathematics. Some of the sporadic
simple groups were originally realized as automorphism groups of cer-
tain combinatorial-geometrical structures like Steiner systems, distance-
regular graphs, Fischer spaces etc., but it was the epoch-making paper
[Bue79] by F. Buekenhout which brought an axiomatic foundation for
these and related structures under the name "diagram geometries". Build-
ings offinitegroups of Lie type form a special class of diagram geometries
known as Tits geometries. This gives a hope that diagram geometries
might serve as a background for a uniform treatment of all finite simple
groups.
If G is a finite group of Lie type in characteristic p, then its Tits
geometry ^(G) can be constructed as the coset geometry with res-
pect to the maximal parabolic subgroups which are maximal over-
groups of the normalizer in G of a Sylow p-subgroup (this normalizer
is known as the Borel subgroup). Thus ^(G) can be defined in abs-
tract group-theoretical terms. Similar abstract construction applied to
sporadic simple groups led to maximal [RSm80] and minimal [RSt84]
parabolic geometries, most naturally associated with the sporadic sim-
ple groups. Notice that besides the parabolic geometries there are a
number of other nice diagram geometries associated with sporadic
groups.
Tits geometries are characterized by the property that all their rank 2
residues are generalized polygons. Geometries of sporadic groups besides
the generalized polygons involve c-geometries (which are geometries of
vertices and edges of complete graphs), the geometry of the Petersen

IX
x Preface
graph, tilde geometry (a triple cover of the generalized quadrangle of
order (2,2)) and a few other rank 2 residues.
In the mid 80's the classification project of finite Tits geometries at-
tracted a lot of interest, motivated particularly by the revision program
of the classification of finite simple groups (see [Tim84]). It was natural
to extend this project to geometries of sporadic groups and to try to char-
acterize such geometries by their diagrams. For two classes of diagrams,
namely

and

the complete classification under the flag-transitivity assumption was

achieved by S.V. Shpectorov and the author of the present volume
[ISh94b]. Geometries with the above diagrams are called, respectively,
Petersen and tilde geometries. A complete self-contained exposition of
the classification of flag-transitive Petersen and tilde geometries is the
main goal of the two volume monograph of which the present is the first
volume.
To provide the reader with an idea what sporadic group geometries
look like we present the axioms for the smallest case.
A Petersen geometry of rank 3 is a 3-partite graph ^ with the partition

which possesses the following properties. For a vertex x e <$ let res(x)
denote the subgraph in ^ induced on the set of vertices adjacent to x.
For xt e <S\ 1 < i < 3, the following hold:

is the incidence graph of vertices and edges of the Petersen graph

 Preface xi
res(x2) is the complete bipartite graph

res(x3) is the incidence graph of seven points and seven lines of the
Fano plane

The geometry ^ as above is flag-transitive if its automorphism group

acts transitively on the set of maximal complete subgraphs (such a
subgraph contains one vertex from each part).
It was shown by S.V. Shpectorov in [Sh85] that there exist exactly two
flag-transitive Petersen geometries of rank 3. Their automorphism groups
are isomorphic respectively to the automorphism group Aut Matn of the
Mathieu group Mat2i and to a non-split extension of Aut Mat2i by a
subgroup of order 3. This was the first step in the classification project
of Petersen and tilde geometries.
Our strategy of classification, first implemented in [Sh85], is based on
analysis of amalgams of maximal parabolic subgroups and calculation of
the universal covers and consists of two principal and rather independent
steps.
Step 1. To describe all known pairs (^, G) where ^ is a Petersen or tilde
geometry and G is a flag-transitive automorphism group of ^, calculate
the universal cover of ^ and determine its flag-transitive quotients.
Step 2. To show that the amalgam of maximal parabolic subgroups,
corresponding to a flag-transitive action on a Petersen or tilde geometry
xii Preface
3tf, is isomorphic to such an amalgam corresponding to a pair (^, G)
described in Step 1. By a standard principle this means that ^ is a
flag-transitive quotient of the universal cover of CS.
The main goal of the present volume is to realize Step 1. The local
analysis of amalgams needed for Step 2 will be given in the second
volume. Here we also discuss various applications and implications of
the classification of flag-transitive Petersen and tilde geometries.
In Chapter 1 we start with a review of the main notions and prin-
ciples concerning the diagram geometries and their flag-transitive au-
tomorphism groups. Then we formulate and discuss the results of the
classification project for flag-transitive Petersen and tilde geometries. In
Chapter 2 we prove the existence and uniqueness of the (binary) Go-
lay code and the Steiner system S(5,8,24). Our approach is a mixture
of the approach of Conway (in [Con71]) who constructs the Golay
code as the quadratic residue code over GF(23) and the approach in
Liineburg (in [Liin69]) who treats the Steiner system 5(5,8,24) as an
extension of the projective plane of order 4. The approach provides us
with a strong background to define the Mathieu groups and to study
their subgroup structure. In Chapter 3 we define and study geometries
of the Mathieu groups. We refer to computer calculations performed
independently in [Hei91] and [ISh89a] to claim the simple connectedness
of the tilde geometry ^(Mati*). The simple connectedness proofs for the
Petersen geometries ^(Matii) and ^(Matii) which we present here are
basically the original ones from [Sh85] and [ISh90a]. In Chapter 4 we
follow [Con69] and [KKM91] to establish the existence and uniqueness
of the Leech lattice. This approach immediately gives the order and basic
properties of the automorphism group of the Leech lattice. We present
a detailed study of the action of Co\ on A4 and of an orbital graph
associated with this action. This graph is the collinearity graph of the
tilde geometry ^(Coi). We present the simple connectedness proofs for
<&(Coi) and &(Coi), originally given in [Sh92] and [Iv92a], respectively.
At the end of Chapter 4 we discuss geometries of certain subgroups in
the Conway group Co\. In Chapter 5 we prove the simple connectedness
of the tilde geometry ^(M) of the Monster. We start with an amalgam
Jt similar to the amalgam of maximal parabolics associated with the
action of the Monster on its tilde geometry and consider a faithful com-
pletion G of Jt. We define a number of subgroups in G associated with
certain subgeometries in ^(M). Applying the simple connectedness of
these subgeometries originally established in [Iv92c], [Iv94] and [Iv95]
 Preface xiii
we identify in G the subgroups 3 • M(24) and 2 • BM. By considering the
subgeometry in ^(M) formed by the fixed points of an element of order
7, we construct the tilde geometry ^(He) of the Held group. We define
a graph T on the set of Baby Monster subgeometries in ^(M) (called
the second Monster graph) and study its local properties. We apply the
triangulability of T proved in [ASeg92] to establish the simple connect-
edness of ^(M). In Chapter 6 we follow [ISh93a] to construct an infinite
family of tilde geometries associated with some non-split extensions of
symplectic groups over GF(2). In the last section of Chapter 6 we follow
[ISh90a] to prove the non-existence of tilde analogues of the exceptional
C3-geometry ^(Alt-j). In Chapter 7 we construct the Petersen geometries
associated with the non-split extensions
3-AutMar22, 3 23 -Co 2 , 34371 • BM
and prove their 2-simple connectedness following [Sh92] and [ISh93b]. In
Chapter 8 we discuss the identification proof of 7555 with the Bimonster.
In this proof the simple connectedness of ^(M) plays an essential role.
In Chapter 9 we consider locally projective graphs and show how the
classification of theflag-transitivePetersen geometries implies description
of a class of locally projective graphs of girth 5. Originally this reduction
was proved in [Iv88], [Iv90] (see also a survey [Iv93a]). In this volume we
do not treat the fourth Janko group J4 and its Petersen geometry ^{J^),
and refer the reader to [IMe93] where the group and its geometry are
constructed and characterized starting with very basic principles.

I would like to thank S.V. Shpectorov for the fruitful cooperation on

the classification project for flag-transitive Petersen and tilde geometries
which led to its completion. I am grateful to B. Baumeister, S. Hobart,
G. Glauberman, C.E. Praeger, C. Wiedorn who read various parts of pre-
liminary versions of the volume and suggested a number of corrections.
I am glad to acknowledge that many suborbit diagrams presented in the
volume have been computed by D.V. Pasechnik.
 1
Introduction

In this introductory chapter we collect basic definitions, formulate main

results and discuss some of the motivations and consequences. In Sec-
tion 1.1 we start with an informal review of classical geometries in order
to motivate the general notion of geometry as introduced by J. Tits in
the 50's. In Section 1.2 we discuss morphisms of geometries and two
of their most important special cases, coverings and automorphisms.
Our main interest is in flag-transitive geometries. By a standard princi-
ple a flag-transitive geometry ^ can be uniquely reconstructed from its
flag-transitive automorphism group G and the embedding in G of the
amalgam si (defined in Section 1.3) of maximal parabolic subgroups
corresponding to the action of G on (S. In Section 1.4 we formulate a
condition under which an abstract group G and a subamalgam si in G
lead to a geometry. In Section 1.5 we formulate the most fundamental
principle in the area which relates the universal cover of a flag-transitive
geometry ^ and the universal completion of the amalgam of maximal
parabolic subgroups corresponding to a flag-transitive action on #. In
Section 1.6 we discuss parabolic geometries of finite groups of Lie type.
These geometries belong to the class of so-called Tits geometries charac-
terized by the property that all rank 2 residues are classical generalized
polygons. We formulate the local characterization of Tits geometries
which shows a special role of C3-geometries. We also formulate a very
useful description offlag-transitiveautomorphism groups of classical Tits
geometries due to G. Seitz. A very important non-classical Tits geometry,
known as the ;4/£7-geometry, is discussed in Section 1.7. In Section 1.8 we
apply the characterization of Tits geometries to Cn(2)-geometries which
play a very special role in our exposition. In Section 1.9 we mimic the
construction of Cn(2)-geometries of symplectic groups to produce a rank
5 tilde geometry of the Monster group. In Section 1.10 the classification

1
2 Introduction
result for flag-transitive Petersen and tilde geometries is stated, which
shows in particular that the Monster is strongly characterized as a flag-
transitive automorphism group of a rank 5 tilde geometry. In Section 1.11
we introduce and discuss a very important notion of natural representa-
tions of geometries. Section 1.12 contains a brief historical essay about
the classification of flag-transitive Petersen and tilde geometries. In Sec-
tion 1.13 we present some implications of the classification including the
identification of Y -groups. In the final section of the chapter we fix our
terminology and notation concerning groups, graphs and geometries. The
terminology and notation are mostly standard and we start using them
in the earlier sections of the chapter without explanations.

1.1 Basic definitions

We start this chapter with a brief and informal review of the geometries of
classical groups in order to motivate the general definition of geometries.
Let G be a finite classical group (assuming the projective version).
The group G itself and its geometry can be defined in terms of the
natural module which is an n-dimensional vector space V = Vn(q) over
the Galois field GF(q) of order q. Here q is a power of a prime number p
called the characteristic of the field. There is a sesquilinear form *F on V
which is either trivial (identically equal to zero) or non-singular and the
elements of G are projective transformations of V which preserve *F up to
multiplication by scalars. If *F is trivial then G is just a projective linear
group associated with V. If *F is non-singular, it is symplectic, unitary
or orthogonal and G is the symplectic, unitary or orthogonal group of a
suitable type determined by n, q and the type of *F. We have introduced
the trivial form in the case of linear groups in order to treat all classical
groups uniformly.
For a subspace W of V we can consider the restriction of *F to W.
The subspaces on which *F restricts trivially play a very special role and
they are called totally singular subspaces of V with respect to *F. Clearly
every subspace of a totally singular subspace is also totally singular and
in the case of linear groups all subspaces are totally singular. If *F is a
non-singular form then by the Witt theorem all maximal totally singular
subspaces have the same dimension known as the Witt index of XF.
The geometry 9 — &(G) of a classical group G is the set of all
proper totally singular subspaces in the natural module V with respect
to the invariant form *F together with a symmetrical binary incidence
relation * under which two subspaces are incident if and only if one of
 1.1 Basic definitions 3
them contains the other one. In the case of a linear group we obtain
the projective geometry associated with the natural module and in the
remaining cases we obtain various polar spaces.
By the definition every element of a classical group geometry is incident
to itself which means that the relation * is reflexive. One can consider
^ as a graph on the set of elements whose edges are pairs of incident
elements. Since two subspaces of the same dimension are incident if and
only if they coincide, one can see (ignoring the loops) that the graph
is multipartite. Two vertices are contained in the same part if and only
if they have the same dimension as subspaces of V. It is natural to
define the type of an element to be the dimension of the corresponding
subspace. The Witt theorem and its trivial analogue for the case of linear
groups imply that every maximal set of pairwise incident elements of ^ (a
maximal clique in graph-theoretical terms) contains exactly one element
of each type. This construction suggests the definition of geometry as
introduced by J. Tits in the 1950s.
Geometries form a special class of incidence systems. An incidence
system is a quadruple (0,*,£,/) where ^ is the set of elements, * is
a binary reflexive symmetric incidence relation on 9 and t is a type
function which prescribes for every element from ^ its type which is an
element from the set / of possible types; two different elements of the
same type are never incident. We will usually refer to an incidence system
(^, *,£,/) simply by writing ^, assuming that *, t and / are clear from
the context. The number of types in an incidence system (that is the size
of /) is called the rank. Unless stated otherwise, we will always assume
that / = {l,2,...,n} for an incidence system of rank n and write &l for
the set of elements of type i in ^, that is for t~x(i).
An incidence system ^ of rank n can be considered (ignoring loops) as
an n-partite graph with parts &1 ,...,&". An incidence system is connected
if it is connected as a graph.
A set O of pairwise incident elements in an incidence system is called a
flag. In this case |O| and t(Q>) are the rank and the type of®, respectively.
If ^ is an incidence system of rank n over the set I of types then n — |O|
and / \ £(O) are the corank and the cotype of <I>, respectively. Let O be a
flag in an incidence system ^. The residual incidence system res^(O) of O
in ^ (or simply residue) is the quadruple (^$, *<&,t®,Iq>) where
?* = {x | x e *, x * y for every yGO}\<D,
Jo = / \ t(Q>), *o is the restriction of * to &® and t$> is the restriction of t
to ^o> The notion of residue corresponds to that of link, more common
4 Introduction
in topology. For a flag consisting of a single element x its residue will
be denoted by res^(x) rather than by res#({x}). It is easy to see that one
can construct an arbitrary residue inductively, producing at each step the
residue of a single element.

Definition 1.1.1 A geometry is an incidence system (^,*,t,I)for which the

following two conditions hold:
(i) every maximalflagcontains exactly one element of each type;
(ii) for every i, j € t(^) the graph on &l U <3i in which two elements
are adjacent if they are incident in & is connected, and a similar
condition holds for every residue in & of rank at least 2.
The graph on the set of elements of a geometry ^ in which two distinct
elements are adjacent if they are incident in ^ is called the incidence graph
of ^. The incidence graphs of geometries of rank n are characterized as
n-partite graphs with the following properties: (i) every maximal clique
contains exactly one vertex from each part; (ii) the subgraph induced by
any two parts is connected and a similar connectivity condition holds
for every residue of rank at least 2. It is easy to see that a residue of a
geometry is again a geometry.
Let (^i,*i,£i,/i) and (^25*2^2^2) be two geometries whose sets of
elements and types are disjoint. The direct sum of ^1 and ^2 is a
geometry whose element set is ^1 U &2, whose set of types is I\ UI2,
whose incidence relation and type function coincide respectively with *,-
and U when restricted to <3\ for i = 1 and 2 and where every element
from ^1 is incident to every element from ^2-
The above definitions of residue and direct sum have the following
motivation in the context of geometries of classical groups. Let G be a
classical group with a natural module V and the invariant form XF. Let
<g = <g(G) be the geometry of G as defined above. Let W be an element
of 9 that is a totally singular subspace of V with respect to *F. It is
easy to see that ies&(W) is the direct sum of two geometries resj(W)
and resJ(W), where the former is the projective geometry of all proper
subspaces of W and the latter is formed by the totally singular subspaces
containing W and can be described as follows. Let
W1- = {v I v e 7,*F(t?,w) = 0 for every w € W}
be the orthogonal complement of W. Then W < W1- and *¥ induces
on [ / = W^/W a non-singular form ¥'. The elements of resJ(P^) are
the subspaces of U totally singular with respect to *F' with the incidence
 1.2 Morphisms of geometries 5
relation given by inclusion. So resJ(W) is the geometry of the classical
group having U as natural module and *¥' as invariant form. Certainly
resg(W) or resJ(W) or both can be empty and one can easily figure out
when this happens. In any case the observation is that the class of direct
sums of geometries of classical groups is closed under taking residues.
By introducing geometries of classical groups we started considering
the totally isotropic subspaces of their natural modules as abstract el-
ements preserving from their origin in the vector space the incidence
relation and type function. It turns out that in most cases the vector
space can be uniquely reconstructed from the geometry and moreover
the geometry itself to a certain extent is characterized by its local prop-
erties, namely by the structure of residues. The theory and classification
of geometries can be developed quite deeply without making any as-
sumption on their automorphism groups. But our primary interest is in
so-called flag-transitive geometries to be introduced in the next section.

1.2 Morphisms of geometries

Let J f and ^ be geometries (or more generally incidence systems). A
morphism of geometries is a mapping cp : J f —> ^ of the element set of
Jf7 into the element set of ^ which maps incident pairs of elements onto
incident pairs and preserves the type function. A bijective morphism is
called an isomorphism.
A surjective morphism cp : J f -» ^ is said to be a covering of ^ if for
every non-empty flag Q> of J f the restriction of cp to the residue res^ (<D)
is an isomorphism onto res&((p(Q})). In this case ^ is a cover of ^ and
^ is a quotient of Jf. If every covering of ^ is an isomorphism then ^
is said to be simply connected. Clearly a morphism is a covering if its
restriction to the residue of every element (considered as a flag of rank 1)
is an isomorphism. If xp : & -> ^ is a covering and §? is simply connected,
then xp is the universal covering and 9 is the universal cover of <3. The
universal cover of a geometry exists and it is uniquely determined up to
isomorphism. If <p : 2tf —• ^ is any covering then there exists a covering
X : 9 —> 2tf such that xp is the composition of % and cp.
A morphism cp : J f —> 9 of arbitrary incidence systems is called an
s-covering if it is an isomorphism when restricted to every residue of rank
at least s. This means that if O is a flag whose cotype is less than or
equal to s, then the restriction of cp to res^(<E) is an isomorphism. An
incidence system, every s-cover of which is an isomorphism, is said to be
s-simply connected. The universal s-cover of a geometry exists in the class
6 Introduction
of incidence systems and it might or might not be a geometry. In the
present work we will mainly use the notion of s-covers either to deal with
concrete morphisms of geometries or to establish s-simple connectedness.
For these purposes we can stay within the class of geometries. It must
be clear that in the case s = n — 1 "s-covering" and "covering" mean the
same thing.
An isomorphism of a geometry onto itself is called an automorphism.
By the definition an isomorphism preserves the types. Sometimes we will
need a more general type of automorphisms which permute types. We
will refer to them as diagram automorphisms.
The set of all automorphisms of a geometry ^ obviously forms a
group called the automorphism group of ^ and denoted by A u t ^ . An
automorphism group G of ^ (that is a subgroup of Aut ^ ) is said to be
flag-transitive if any two flags Q>i and Q>2 in ^ of the same type (that is with
t(Q>\) = t(Q>2)) are in the same G-orbit. Clearly an automorphism group is
flag-transitive if and only if it acts transitively on the set of maximal flags
in <$. A geometry ^ possessing a flag-transitive automorphism group is
called flag-transitive.
A flag-transitive geometry can be described in terms of certain sub-
groups and their cosets in a flag-transitive automorphism group in the
following way. Let ^ be a geometry of rank n and G be a flag-transitive
automorphism group of (§. Let O = {xi,X2,...,xn} be a maximal flag in
^ where x\ is of type i. Let Gf = G(XJ) be the stabilizer of Xt in G.
The subgroups G\, G2, ..., Gn are called the maximal parabolic subgroups
or just maximal parabolics associated with the action of G on ^ . When
talking about n maximal parabolic subgroups associated with an action
on a rank n geometry we will always assume that the elements which
they stabilize form a maximal flag. By the flag-transitivity assumption G
acts transitively on the set & of elements of type i in ^ . So there is a
canonical way to identify ^ with the set of right cosets of G,- in G by
associating with y e&1 the coset Gth such that x1- = y. This coset consists
of all the elements of G which map xt onto y (assuming that action is
on the right). Now with y as above let z be an element of type j which
corresponds to the coset Gjk. By the flag-transitivity assumption y and z
are incident if and only if there is an element g in G which maps the pair
(xt, Xj) onto the pair (y, z). It is obvious that g must be in the intersection
Gih n Gjk and each element from the intersection can be taken for g.
Thus y and z are incident if and only if the cosets G\h and Gjk have a
non-empty intersection. Notice that if the intersection is non-empty, it is
a right coset of Gt n Gj. In this way we arrive at the following.
 1.3 Amalgams 1
Proposition 1.2.1 Let & be a geometry of rank n over the set I = {1,2,..., n}
of types and G be a flag-transitive automorphism group of (S. Let O =
{xi,X2,...,xn} be a maximal flag in & and Gi = G(x,-) be the stabilizer of
Xi in G. Let &(G) be the incidence system whose elements of type i are the
right cosets of Gi in G and in which two elements are incident if and only
if the intersection of the corresponding cosets is non-empty. Then &(G) is
a geometry and the mapping
rj : y\-* Gfh
(where y G &l and xf = y) establishes an isomorphism of $ onto &(G). •

1.3 Amalgams
Discussions in the previous section and particularly (1.2.1) lead to the
following.

Definition 1.3.1 A (finite) amalgam si of rank n is a finite set H such

that for every 1 < i < n there are a subset Hi in H and a binary operation
*i defined on Ht such that the following conditions hold:
(i) {Hu *i) is a group for \<i<n;
(ii) H = ( X i Ht;
(iii) fl-=i K j=Q;
(iv) if x,y e Hi n Hj for 1 < i < j < n then x *,- y = x *j y.
We will usually write si = {Hi | 1 < i < n] for the amalgam si as in
the above definition. Whenever x and y are in the same Hi their product
x *,- y is defined and it is independent of the choice of i. We will normally
denote this product simply by xy. Since B := P|"=1 Hi is non-empty,
one can easily see that B contains the identity element of (Hi9 *i) for
every 1 < i < n. Moreover, all these identity elements must be equal. The
reader may notice that a more common definition of amalgams in terms
of morphisms is essentially equivalent to the above one.
If (G,*) is a group, Hu...,Hn are subgroups of G and *i,...,* w are
the restrictions of * to these subgroups, then si = {Ht \ 1 < i < n} is
an amalgam. This is the most important example of an amalgam, but
at the same time it is not very difficult to construct an example of an
amalgam which is not isomorphic to a family of subgroups of a group.
The amalgam si as above is said to be isomorphic to an amalgam
sif = {H[ | 1 < i < n} if there is a bijection of H onto H' which induces
an isomorphism of (Hu *,) onto (H-9 *•) for every 1 < i < n.
8 Introduction
Definition 1.3.2 A group G is said to be a completion of an amalgam
si = {H( | 1 < i < n} if there is a mapping cp of H into G such that
(i) G is generated by the image of (p,
(ii) for every 1 <i <n the restriction of (p to Hi is a group homomor-
phism with respect to *, and the group operation in G.
If cp is injective then the completion G is said to be faithful.
Thus an amalgam si is isomorphic to a family of subgroups of a
group if and only if si possesses a faithful completion. If G is a faithful
completion of si then we will usually identify si and its image in G.
There is a completion U(si) of si known as the universal completion,
of which any completion is a homomorphic image. The group U(si)
has the following definition in terms of generators and relations: the
generators are all the elements of H; the relations are all the equalities
of the form xyz~l = 1 where x and y are (possibly equal) elements
contained in Hi for some i and z = x *,- y. It is easy to see that U(si)
is a completion of si with respect to the mapping xp which sends every
x e H onto the corresponding generator of U(si). Moreover, if G is an
arbitrary completion of si with respect to a mapping cp then there is a
unique homomorphism x • U(si) —• G such that cp is the composition of
xp and X- Finally, si possesses a faithful completion if and only if U(si)
is a faithful completion.
Let G, ^ and the G, be as in (1.2.1). The amalgam si = {Gt\ 1 <i <n}
is called the amalgam of maximal parabolic subgroups in G associated with
the flag O. The geometry ^(G) should be denoted by ^(G, si) since its
structure is determined not only by G by also by the amalgam si and by
the embedding of si in G. We can reformulate (1.2.1) as follows.

Proposition 1.3.3 Let G be a flag-transitive automorphism group of a ge-

ometry & of rank n and si = {G,- | 1 < i < n} be the amalgam of maximal
parabolic subgroups associated with a maximal flag. Let @(G,si) be the
incidence system whose elements of type i are the right cosets of Gt in G
and in which two elements are incident if and only if the intersection of the
corresponding cosets is non-empty. Then <& and ^(G, si) are isomorphic. •

Notice that by the above proposition the residues of ^ are uniquely

determined by the amalgam si. That is, res^(xj) is isomorphic to ^(G,, sii)
where sit = {G, nGj \ 1 < j <n, j j= i}.
For a subset J ^ / = {l,2,...,w} let Gj = f]ieJ Gt be the elementwise
stabilizer in G of the flag {xt \ i e J}. The subgroup Gj is a parabolic
 1.4 Geometrical amalgams 9
subgroup of rank r where r = |/| — |/|. For i,j e I we write GJ ; instead of
G{tj}. The parabolic subgroups of rank n— 1 are the maximal parabolics.
The parabolic subgroups of rank 1 are known as minimal parabolics and
the subgroup B = G/ is called the Borel subgroup. We will usually write
Pi to denote the minimal parabolic G/\{,-} and Ptj to denote the rank 2
parabolic G A{lJ} .

1.4 Geometrical amalgams

In view of (1.3.3) the following question naturally arises.

Q. Let G be a group, Gi,G2,..., Gn be subgroups of G and si — {Gt |

1 < i < n} be the amalgam formed by these subgroups. Under what
circumstances is the incidence system ^ = &(G,s/) a geometry and
the natural action of G on ^ flag-transitive?

Below we discuss the answer to this question as given in [Ti74].

The set O = {Gi,G2,..., Gn} is a flag in ^ since each G\ contains the
identity element and O is a maximal flag since for 1 < i < n and g € G
either Gtg = Gt or Gtg n Gt = 0. A set *F = {GI-1fci,...,G^lAm} is a flag in
# if and only if G^/z, n GIfc/zfc ^ 0 for all j,k with 1 < j,k < m (which
implies particularly that ij =fc ik). We say that the flag *F is standard if
the intersection p|J =1 Gify is non-empty and contains an element h, say.
In this case *F = {G^,..., G^}*, which means that ^ is the image under h
of a subflag in O. This shows that every standard flag is contained in a
standard maximal flag and G acts transitively on the set of standard flags
of each type. Clearly G cannot map a standard flag onto a non-standard
one. Thus the necessary and sufficient condition for flag-transitivity of
the natural action of G on ^ is absence of non-standard flags.
The proof of the following result uses elementary group theory only
(compare Sections 10.1.3 and 10.1.4 in [Pasi94]).

Lemma 1.4.1 The incidence system &(G, si) does not contain non-standard
flags if and only if the following equivalent conditions hold:
(i) if J, K, L are subsets of I and g, h, f are elements in G such that
the cosets Gjg, GKK GLJ have pairwise non-empty intersection, then
Gjgr\GKhC\GLf^0;
(ii) for ij G / and J c / \ {ij} if g e Gj and Gt n Gjg £ 0 then
Gj n Gi n Gjg j* 0. •
10 Introduction
One may notice that, in general, existence of non-standard flags in
^(G, s/) depends not only on the structure of $4 but also on the structure
of G.
The connectivity condition in (1.1.1 (ii)) is also easy to express in terms
of parabolic subgroups. By the standard principle the graph on & U &
is connected if and only if G is generated by the subgroups G, and Gj.
This gives the following.

Lemma 1.4.2 The incidence system &(G, stf) satisfies the condition (ii) in
(1.1.1) if and only iffor every 2-element subset {i,j} ^ / the subgroups G*
and Gj generate G. •
Finally let K be the kernel of the action of G on y(G,stf). It is
straightforward that K is the largest subgroup in the Borel subgroup
B = P|"=1 Gj, which is normal in Gj for all i with 1 < i < n (equivalently,
normal in G). In particular the action of G on ^(G, J / ) is faithful if and
only if the Borel subgroup contains no non-identity subgroup normal in
G.

1.5 Universal completions and covers

The fact that the structure of residues in &(G, J / ) is determined solely by
j / plays a crucial role in the description of the coverings of ^(G, stf).
Let ^ be a geometry, G be a flag-transitive automorphism group of
^ and J / = {Gj | 1 < i < n} be the amalgam of maximal parabolic
subgroups associated with the action of G on (§. Then on the one hand
^ = ^(G, J / ) and on the other hand G is a faithful completion of srf. Let
G' be another faithful completion of sd and let
cp:Gf -• G

be an j/-homomorphism, i.e. a homomorphism of Gr onto G whose

restriction to J / is the identity mapping. As usual we identify srf with its
images in G' and G. The following result is straightforward.

Lemma 1.5.1 In the above terms the mapping of(S{G',s^) onto

induced by cp is a covering of geometries. •
In the above construction we could take G to be the universal com-
pletion U(jtf) of J / . The following result of fundamental importance was
proved independently in [Pasi85], [Ti86] and an unpublished manuscript
by S.V. Shpectorov.
 1.6 Tits geometries 11
Proposition 1.5.2 Let & be a geometry, G be a flag-transitive automorphism
group of y and stf be the amalgam of maximal parabolic subgroups asso-
ciated with the action of G on &. Then &(U(s/)9s#) is the universal cover

By the above proposition a flag-transitive geometry ^ is simply con-

nected if and only if a flag-transitive automorphism group G of ^ is the
universal completion of the amalgam of maximal parabolic subgroups
associated with the action of G on <§.
We also present a condition for 2-simple connectedness of a geometry.

Proposition 1.5.3 Let ^ be a geometry, G be a flag-transitive automor-

phism group of y and gft = {P,; \ 1 <i < j <n} be the amalgam of rank 2
parabolics associated with the action of G on C3. Then <$ is 2-simply con-
nected (as an incidence system) if and only if G is the universal completion
of Si. •

1.6 Tits geometries

In view of (1.3.3) and Section 1.4 a flag-transitive geometry can be
constructed starting with a group G and an amalgam si of which G is
a faithful completion. In these terms the classical geometries possess the
following very natural description.
Let G be a classical group defined over a field of characteristic p. Let
S be a Sylow p-subgroup of G and B = NG(S). Let Gi,...,Gn be those
maximal subgroups of G which contain B and
stf = {Gt\l<i< n}

be the corresponding amalgam in G. Then the classical geometry ^(G)

defined in terms of totally singular subspaces in the natural module of
G is isomorphic to y(G,stf). This observation shows that the natural
module is not needed for defining ^(G) and enables one to associate
geometries with exceptional groups of Lie type as well. We believe that
this was the main motivation of J. Tits for introducing the notion of
geometries. The geometry ^(G, si) will be called the parabolic geometry
of G.
Let us discuss residues of ^(G). Similarly to the case of geometries of
classical groups, the class of direct sums of parabolic geometries of Lie
type groups is closed under taking residues. Let us consider the smallest
12 Introduction
non-trivial residues, the residues of rank 2. We know that the residue
J^ij of type {ij} is isomorphic to

As mentioned, J^y is either the direct sum of two geometries of Lie

type groups of rank 1 or the parabolic geometry of a Lie type group of
rank 2. In the former case Jf,;- is a complete bipartite graph while in the
latter case it is a classical generalized m-gon for m > 3.

Definition 1.6.1 A generalized m-gon of order (s, t) is a rank 2 geometry

£ in which the elements of one type are called points, the elements of the
other type are called lines, such that
(i) every line is incident to s + 1 points, every point is incident to t + \
lines,
(ii) the incidence graph of Z has diameter m and its girth (the length
of the shortest cycle) is 2m.
If Z is a generalized m-gon, then the geometry in which the roles
of points and lines are interchanged is a generalized m-gon dual to Z.
Sometimes we do not distinguish generalized m-gons from their duals,
and identify them both with their incidence graphs.
If G is a Lie type group of rank 2 whose Weyl group is isomorphic
to the dihedral group Dim of order 2m, then the parabolic geometry of
G is a generalized m-gon. The generalized m-gons arising in this way are
called classical.
Notice that a complete bipartite graph is a generalized 2-gon (also
called a generalized digon). A generalized 3-gon (a generalized triangle)
is the same as a projective plane. In this case s = t is the order of the
plane.
Let £ be a generalized m-gon of order (s, t). If s = t = 1 then £ is
the ordinary m-gon, which clearly exists for every m. If s > 1 and t > 1
then Z is said to be thick. The class of thick generalized m-gons (also
called generalized polygons) is rather restricted, as follows from the Feit
- Higman theorem [FH64].

Theorem 1.6.2 Finite thick generalized m-gons exist if and only if m e

{2,3,4,6,8}. •
For every m e {3,4,6,8} there exists a classical generalized m-gon.
Since generalized triangles are just projective planes, there are many non-
classical ones. Similarly there are many non-classical generalized 4-gons
 1.6 Tits geometries 13
(also called generalized quadrangles.) So far no non-classical examples of
generalized 6-gons (hexagons) or 8-gons (octagons) have been constructed.

Spherical Coxeter diagrams :

An : o o o ••• o o o

Cn: o-

Dn: o-

E6:

£7: o-

F4:

H3 :

H4:

Let A be the diagram of ^(G), which is a graph on the set / of types

with nodes i and j being joined by an edge with multiplicity (mi; — 2) (or
just by a simple edge labelled by my) if the rank 2 residues in 0(G) of type
{i,j} are generalized my-gons. This means particularly that there is no
edge between i and j if the residues of type {i, j} are generalized digons.
Then A is the Dynkin diagram of the Lie algebra associated with G and
also the Coxeter diagram of the Weyl group W of G. This means that
W has the following presentation in terms of generators and relations:

W = (eh i e I | e} = 1, (e«ejp = 1).

14 Introduction
Since G is finite, W is a finite Coxeter group, i.e. the diagram A is
spherical. Recall that a Coxeter group is a Weyl group of a finite group
of Lie type if and only if its diagram is spherical and each mi; is from
the set {2,3,4,6}.
In a similar way one can associate with an arbitrary (flag-transitive)
geometry ^ the diagram A(&) whose nodes are the types of ^ and the
edge joining i and j symbolizes the residues of type {i,j} in ^. Under the
node i it is common to put the index qt such that the number of maximal
flags containing a flag of type I \{i} is q\ + 1. Sometimes above the node
we write the corresponding type but usually the types on the diagram are
assumed to increase rightward from 1 up to the rank of the geometry. A
geometry all of whose rank 2 residues are generalized polygons is called
a Tits geometry.
Throughout the book all geometries are assumed to be locally finite
which means that all the indices qt are finite. If all the indices are greater
than 1 then the geometry is said to be thick.
If Jf is the residue in ^ of an element of type i then the diagram A( Jf)
can be obtained from A(^) by omitting the node i and all the edges
incident to this node. Notice that if <&' —> ^ is a 2-covering of geometries
then y and ^ have the same diagram.
It turns out that many properties of ^(G) can be deduced from
its diagram and in many cases the diagram of ^(G) (including the
indices) specifies ^(G) up to isomorphism. Without going into details,
this important and beautiful topic can be summarized as follows.
Existence in G of the Weyl group W as a section imposes on ^ an
additional structure known as a building. The buildings of spherical type
(i.e. with underlying geometries having spherical diagrams) were classified
in [Ti74] by showing that they are exactly the parabolic geometries of
finite groups of Lie type. Later in [Ti82] it was shown that under certain
additional conditions the structure of a building can be deduced directly
from the condition that all rank 2 residues are generalized polygons.
That is the following result was established.

Theorem 1.6.3 Let & be a Tits geometry of rank n>2. Then & is covered
by a building if and only if every rank 3 residue in & having diagram

is covered by a building. •
 1.6 Tits geometries 15
We should emphasize again that in view of the main result of [Ti74]
the buildings of spherical type are exactly the parabolic geometries of
finite groups of Lie type.
We formulate another important result from [Ti82].

Theorem 1.6.4 Every building of rank at least 3 is 2-simply connected. •

If G is a finite group of Lie type then its parabolic geometry

is a building and hence it is 2-simply connected by the above theorem.
In view of (1.5.3) this means that G is the universal completion of the
amalgam of rank 2 parabolic subgroups associated with the action of
G on ^(G). This reflects the fact that G is defined by its Steinberg
presentation. In fact every Steinberg generator is contained in one of the
minimal parabolic subgroups associated with a given maximal flag and
for every Steinberg relation the generators involved in the relation are
contained in a parabolic subgroup of rank at most 2. Thus the Steinberg
presentation is in fact a presentation for the universal completion of the
amalgam of rank 2 parabolics.
The last important topic we are going to discuss in this section is the
flag-transitive automorphism groups of parabolic geometries of groups
of Lie type. Let G be a Lie type group in characteristic p, & = &(G) be
the parabolic geometry of G, B be the Borel subgroup and U — OP(B).
An automorphism group H of ^ is said to be classical if it contains
the normal closure UG of U in G. In this case if G is non-abelian, then
H contains the commutator subgroup of G. The following fundamental
result [Sei73] (see Section 9.4.5 in [Pasi94] for the corrected version) shows
that up to a few exceptions the flag-transitive automorphism groups of
classical geometries are classical.

Theorem 1.6.5 Let & be the parabolic geometry of a finite group of Lie
type of rank at least 2 and H be a flag-transitive automorphism group of
&. Then either H is classical or one of the following holds:
(i)& is the projective plane over GF(2) and H = Frob];
(ii)& is the projective plane over GF(S) and H = Frob913);
(iii)^ is the ^-dimensional projective GF(2)-space and H = Alt-];
(iv) & is the generalized quadrangle of order (2,2) associated with Sp4(2)
and H^Alt6;
(v) & is the generalized quadrangle of order (3,3) associated with
and H is one of24 : Alt5, 2 4 : Sym5 and 2 4 : Frob45;
16 Introduction
(vi) & is the generalized quadrangle of order (3,9) associated with 1/4(3)
and H is one o/L 3 (4).2 2 ; L4(3).23 and L 3 (4).2 2 ;
(vii) ^ is £/ie generalized hexagon of order (2,2) associated with G2(2)
and H S G2(2)' ^ t/ 3 (3);
(viii) ^ is £/ze generalized octagon of order (2,4) associated with 2 F 4 (2)
and H ^ 2 F 4 (2)' (the Tits group). D

1.7
Let us discuss the exceptional rank 3 residues from (1.6.3). By (1.6.2)
there do not exist any thick generalized 5-gons, so as long as we are
interested in thick locally finite Tits geometries we should not worry
about the H3 -residues. On the other hand there exists a thick flag-
transitive C3-geometry which is not covered by a building. This geometry
was discovered and published independently in [A84] and [Neu84] and
can be described as follows.
Let Q be a set of size 7 and G = Alt-? be the alternating group of Q. Let
n be a projective plane of order 2 having Q as set of points. This means
that n is a collection of seven 3-element subsets of Q such that any two
of the subsets have exactly one element in common. Let G\ = Alte be
the stabilizer in G of an element a e Q. Let G2 be the stabilizer in G of a
line of n containing a, so that G2 = (Sym3 x Sym^f where the superscript
indicates that we take the index 2 subgroup of even permutations. Finally
let G3 be the stabilizer of n in G, so that G3 = L3(2) is the automorphism
group of n.
Let s/ = {Gi,G2,G3} and 9 = ^ ( G , J / ) . Then ^ is a Tits geometry
with the following diagram:
C3(2) : o o
31
' 2 2 2
If {xi,x2jx3} is a maximal flag in ^ where x,- is of type i then res#(x3)
is canonically isomorphic to n9 res^(x2) is the complete bipartite graph
K^3 and res^(xi) is the (unique) generalized quadrangle of order (2,2)
associated with Sp4(2) on which G\ = Alts acts flag-transitively (1.6.5
(iv)). Notice that G, n Gj = Sym4 for 1 < i < j < 3 and B ^ D8.
The C3-geometry ^(Alt-j) was characterized in [A84] by the following
result (see also [Tim84], p. 237).

Theorem 1.7.1 Let <& be a flag-transitive C^-geometry such that the residue
of an element of type 1 is a classical generalized quadrangle and the residue
of an element of type 3 is a Desarguesian (classical) projective plane.
 1.8 Symplectic geometries over GF(2) 17
Then & is either a classical Tits geometry (a building) or isomorphic to
•
It is implicit in the above theorem that ^(Alt-j) is simply connected.
In [Yos96] with use of the classification of finite simple groups it was
shown that ^(Alt-j) is the only non-classical flag-transitive C3-geometry.
This result together with some earlier work on Tits geometries implies
that ^(Alti) is the only thick non-classical Tits geometry with spherical
diagram of rank at least 3.

1.8 Symplectic geometries over GF(2)

In this section we apply the results on classical geometries discussed
in the preceding sections to symplectic geometries over the field GF(2)
of two elements. These geometries will play an important role in our
subsequent exposition.
Let V be a vector space of dimension 2n,n>l, over GF{2) and let *F
be a non-singular symplectic form on V. If {v\,...,v\,v\,...,vfy is a basis
of V then up to equivalence *F can be chosen to be

(here and elsewhere <5,-j is the standard Kronecker symbol). A subspace

U of V is totally singular with respect to *F if Y(M, w) = 0 for all u, w G U.
Since *F is symplectic, every 1-dimensional subspace is totally singular. All
maximal totally singular subspaces have dimension n and Vn — (v\,...,vl)
is one of them.
Let ^ be the set of all non-zero totally singular subspaces in V with
respect to *P, * be the incidence relation on ^ with respect to which two
subspaces are incident if one of them contains the other one (we say that
* is defined by inclusion), t be the mapping from ^ into I = {l,2,...,n}
which prescribes for a subspace its dimension. Then ^ = (^, *,t,I) is a
geometry. Let Vu 1 < i < n, be subspaces in Vn such that dim V\ = i and
Vt is contained in Vj whenever i < j . Then O = {Ki,..., Vn} is a maximal
flag in ^ .
For 1 < i < j < n put O i; = O \ {Vu Vj}. Then res^(<Di;) = J^t U Jffj9
where 2tf\ is the set of /-dimensional totally singular subspaces in V
incident to every subspace from Q>tj. If i is less than k := j — 1 then every
subspace U € Jfj contains Vk while every W e J^t is contained in Vk,
which means that U < W and hence res&(Q>ij) is a generalized digon.
If i = j — 1 and j < n then ^f ,• and ffl) correspond to all 1- and 2-
dimensional subspaces in the 3-dimensional GF(2)-space F)+i/K/_2 and
18 Introduction

is the projective plane of order 2. Finally, if i = n — 1 and j = n

then Jf t and Jtfj correspond to 1- and 2-dimensional subspaces in the
4-dimensional space V^_2/Vn-2 that are totally singular with respect to
the symplectic form on this space induced by *F. In this case res^(<Di;) is
the generalized quadrangle of order (2,2) associated with Sp^(2). Since
all indices in *& are equal to 2, we observe that *§ is a Cn-geometry with
the following diagram, where as usual the types increase rightward from
1 to n:

Let H be the group of those linear transformations of V which preserve

x
¥. Then H = Sp2n(2), the action of H on 9 is flag-transitive and H is
the automorphism group of <§. The flag-transitivity follows from the
transitivity of H on the set of maximal totally singular subspaces and
from the fact that the stabilizer in H of such a subspace induces on the
subspace its full automorphism group Ln(2). The geometry constructed
above will be denoted by ^(Sp2n(2)).
Let G be any flag-transitive automorphism group of 9 = &(Sp2n(2)).
Since H is simple for n > 3 we conclude from (1.6.5) that either G = H
or n = 2 and G ^ Alt6.
Let B, Pi and Pfj be the Borel subgroup, minimal and rank 2 parabolics
in G associated with O (here 1 < i < j < n). Then B is a Sylow 2-subgroup
in G and P i ; = (Pi,Pj). If G = H then P;y acting on res^(O,;) induces the
automorphism group of the residue isomorphic to Sym^ x Symi, Li(2) or
Sp4(2) depending on i and j . The kernel Qtj of this action is contained
in B which is a 2-group. This shows that Q i; = 6)2(^7). Thus G and its
parabolic subgroups of rank 1 and 2 satisfy the following conditions:

(WP1) G is a group generated by its subgroups Pu 1 < i < n, where

n> 2;
(WP2) for 1 < i < j < n the intersection B := P{ n P, is a 2-group,
which is independent of the particular choice of i and j ;
(WP3) Pi/O2(Pi) ^ Syms for 1 < 1 < n;
(WP4) if P o = (P,-,P;-> for 1 < i < j < n and Qtj = O2(Pij) then B is
a Sylow 2-subgroup of Ptj and
Sym3 x 5^m3 if j — i> 1,
L3(2) ifi = j-landj<n,
!
Sp4(2) o r A / t 6 ifi = n-l,j = n;
(WP5) if AT < 5 and JV is normal in Pt for all 1 < i < n, then AT = 1
(the identity subgroup).
 1.9 From classical to sporadic geometries 19
By (1.6.4) the geometry ^ is 2-simply connected and by (1.5.3) this im-
plies that G is the universal completion (and in fact the only completion)
of the amalgam <& = {P i; | 1 < i < j < n} of rank 2 parabolics. But in
fact the conditions (WP1)-(WP5) show that (G;{P, | 1 < i < n}) is a
so-called weak parabolic system and by Theorem 3.2 in [Tim84] we have
the following.

Theorem 1.8.1 Let G be a group, Pit 1 < i < n, n > 2, be subgroups in G

and suppose that conditions (WP1)-(WP5) hold. Let Gt = (Pk\l <k<
n, k j= i) and srf = {G*< | 1 < i < n). Then &(G,jtf) is a Tits geometry of
rank n with diagram Cn(2) on which G acts faithfully and flag-transitively
so that one of the following holds:
(i) ^ ( G , J / ) 2* &(Sp2n(2)) and either G ^ Spln{2), or n = 2 and G ^
Alt6;
(ii) ^ ( G , s/) ^ <g(Alt7) and G ^ Alt7. •

We formulate a related result from [Tim84] which we will often use.

Theorem 1.8.2 In the notation of (1.8.1) suppose that the condition (*) in
(WPA) is changed to
x S
3"H3 if} ~ i > 1,
otherwise.
Then ^(G,A) is the protective GF(2)-space of rank n and either G =
L n+ i(2) or n = 3 and G ^ Alt7. •

1.9 From classical to sporadic geometries

Let us mimic the construction of geometries of symplectic groups for the
largest sporadic simple group, the Monster.
Let G = M be the Monster group. Let B be a Sylow 2-subgroup of
G, whose order is 2 46 . There are exactly five subgroups Pu...,Ps in M
which contain B as a maximal subgroup. If we put P;7 = (Pt,Pj) for
1 < i < j < 5, then the conditions (WP1)-(WP5) from the previous
section hold with (*) in (WP4) being changed to the following (where n
is assumed to be 5).

{ Sym3 x Symi

L3(2)
if j — i> 1,

ifi = ; - l a n d ; < n ,
3 • Sp4(2) or 3 • Alt6 if i = n - 1, j = n.
Here 3-Sp4(2) and 3-Alte are non-split extensions by subgroups of order
3 of Sp4{2) ^ Sym6 and Alt6, respectively. In fact P4s/O2(P45) = 3-Sp4(2)
20 Introduction
in the Monster group. The rank 2 geometry ^(^45, {P^Ps}) is a triple
cover ^(3 • Sp4(2)) of the generalized quadrangle ^(Sp4(2)) of order (2,2).
This cover is denoted by the diagram
rri

2
' 2 2

and called the tilde geometry of rank 2.

Let G, = (Pfc I 1 < k < 5, k + i), J / = {G« | 1 < i < 5} and
^(M) = 0(G, J / ) . It was shown in [RSt84] that ^(M) is a geometry on
which M acts flag-transitively and which has the following diagram:

Thus the Monster group M acts flag-transitively on a geometry

whose local properties (rank 2 residues) are similar to those of
In view of (1.8.1) it is natural to pose the following question: "To what
extent is the Monster characterized by the property that it is a flag-
transitive automorphism group of a tilde geometry of rank 5?"
Notice that ^(M) contains, as residues, tilde geometries of rank 3 and
4 associated with sporadic simple groups MatiA and Co\. It also contains
a number of other nice subgeometries and one such subgeometry can be
described in the following way.
Let T be a Baby Monster involution in M, in which case CM(T) = IBM
where BM is the Baby Monster sporadic simple group. Then a subset
of the set of elements in ^(M) fixed by T (here we do not define this
subset precisely) forms a geometry &(BM) on which BM = CM(T)/(T)
actsflag-transitivelyand which has the following diagram:

Here the rightmost edge denotes the geometry ^(Syms) of edges and
vertices (the left and the right type on the diagram) of the Petersen
graph. The vertices of the Petersen graph are the 2-element subsets of a
set of size 5 with two vertices subsets being adjacent if they are disjoint.
In other terms let Si = Dg and S2 = Sym^ x 2 be subgroups in S = Syms
such that Si DS2 = 22. Then <0(Sym5) s 9{S9{SUS2}). By (1.6.2) there
are no thick generalized 5-gons. Speaking informally the Petersen graph
is as close as one can get to such a 5-gon in terms of girth (which is 10)
and diameter (which is 6).
We say that @(BM) is a Petersen geometry of rank 5. As residues
&(BM) contains Petersen geometries of rank 3 and 4 associated with
sporadic simple groups Main and C02.
 1.10 The main results 21
1.10 The main results
The main aim of this monograph is to present the detailed exposition of
the classification of flag-transitive Petersen and tilde geometries whose
completion was announced in [ISh94b]. In this section we formulate and
briefly discuss the final result of the classification.

Theorem 1.10.1 There exist exactly eight flag-transitive Petersen geometries

of rank at least 3, whose diagrams and full automorphism groups are the
following:

P3 : o o ?—o Aut Mat22, 3-AutMat22;

P4 : o o o £—o Mat23, Co2, 3 23 • Co2, J4;

P5 : o o o o £—o BM, 3 4371 • BM. •

2 2 2 2 1
In what follows we will write ^(G) for a geometry of which G is the
commutator subgroup of the automorphism group.
The geometries %(Mat22\ &(Co2) and &(BM) in (1.10.1) are not 2-
simply connected and their universal 2-covers are, respectively,
Main), &(323 • Co2) and ^(3 4 3 7 1 • BM).

Theorem 1.10.2 There exist an infinite family of flag-transitive tilde geome-

tries (which contains one geometry of rank n for every n > 2) and four
exceptionals. Every flag-transitive tilde geometry is 2-simply connected. The
diagrams and full automorphism groups are the following:
T3 : o o ~ o Mat24, He;

T4 : o o o ^ o Co\\

M;

Tn : o o ••• o o ~ o 3©2. Sp2n(2). D

22 Introduction
It follows from the classification that whenever H is a flag-transitive
automorphism group of a Petersen or tilde geometry ^, of rank at least
3, then either H = Aut ^, or ^ is a rank 3 Petersen geometry and H is
the commutator subgroup of Aut ^.

Geometry Subgeometry

&(Mat22)

Mat22) »(3 • Sp4(2))

Co2) ^(3 7 • Sp6(2))

^(3 4371 • 5M)

<$(Mat 2 2 )

<S{BM)

The Petersen and tilde geometries are closely related to each other and
also to Cn(2)-geometries. Every Petersen geometry of rank n > 3 contains
either a Cn_i(2)-geometry or a tilde geometry of rank n — 1. In addi-
tion some tilde geometries contain Petersen geometries as subgeometries
(see table above). These mutual embeddings between Petersen and tilde
geometries were used essentially in their classification.
 1.11 Representations of geometries 23
By (1.10.2) the Monster group is indeed rather strongly characterized
by the property that it is a flag-transitive automorphism group of a rank
5 tilde geometry, since we have the following.

Theorem 1.10.3 Let G be a group for which conditions (WP\)-(WP5) in

Section 1.8 hold for n = 5 with (*) in (WP4) being changed to (#) from
Section 1.9. Then G is a flag-transitive automorphism group of a rank 5
tilde geometry and either G = M or G = 3 155 • Spio(2). •
The next characterization involves terms more common for the Mon-
ster group.

Theorem 1.10.4 The Monster is the only group which is generated by three
subgroups C, N and L which satisfy the following:
(i) C ~ 2++24.Coi and O2(C) contains its centralizer in C;
(ii) N/02(N) 3* Sym3 x Mat1Af L/O2(L) ^ L3(2) x 3 • Sym6;
(iii) [N : i V n C ] = 3, [L :LDN] = [L : L n C ] = 7, [L :LnNnC] =
21. •
In the above statement the subgroups C, N and L correspond to the
stabilizers in M of pairwise incident elements in ^(M) of type 1, 2 and
3, respectively.
In what follows P -geometry or T-geometry will mean, respectively,
flag-transitive Petersen or tilde geometry. In case we want to specify
the rank n, we talk about Pn- and Tn-geometries. Thus all our P- and
T-geometries are flag-transitive unless explicitly stated otherwise.

1.11 Representations of geometries

We say that a geometry ^ of rank n belongs to a string diagram if all
rank 2 residues of type {i,j} for \i — j \ > 1 are generalized digons. In this
case the types on the diagram usually increase rightward from 1 to n.
The elements which correspond, respectively, to the leftmost, the second
left, the third left and the rightmost nodes on the diagram will be called
points, lines, planes and hyperplanes:
X Y Z
o o o • • •o o
points lines planes hyperplanes
A graph on the set of points of ^ in which two points are adjacent if
and only if they are incident to a common line is called the collinearity
graph of <&.
24 Introduction
When constructing the geometry ^ associated with a classical group
G we started by treating the (totally singular) subspaces in the natural
module of G as abstract elements. We have seen in Section 1.6 that
in many cases ^ is uniquely determined by the structure of its rank 2
residues. This means that in these cases there is a strong possibility of
reconstructing from 0 the vector space V. It is natural to look for a more
direct way to recover from ^ the vector space V.
In a more general setting the question can be posed in the following
form. Given a geometry ^ and a vector space F, is it possible to define
a mapping cp from the element set of ^ onto the set of proper subspaces
of F, such that dim cp(x) is uniquely determined by the type of x and
whenever x and y are incident, either cp(x) < (p(y) or cp(y) < (p{x)l
This question leads to a very important and deep theory of presheaves
on geometries which was introduced and developed in [RSm86] and
[RSm89]. A special class of the presheaves, described below, has played
a crucial role in the classification of P- and T-geometries.
Let ^ be a geometry with elements of one type called points and
elements of some other type called lines. Unless stated otherwise, if ^
has a string diagram, the points and lines are as defined above. Suppose
that ^ is of GF(2)-type which means that every line is incident to exactly
three points. Let P and L denote, respectively, the point set and the line
set of (S. In order to simplify the notation we will assume that every line
is uniquely determined by the triple of points it is incident to. Let V
be a vector space over GF(2). A natural representation of (the point-line
incidence system associated with) ^ is a mapping cp of P UL into the set
of subspaces of V such that
(NR1) V is generated by Im cp,
(NR2) dim p = 1 for p e P and dim / = 2 for / e L,
(NR3) if / € L and {p, q, r} is the set of points incident to /, then
{(p(p), cp(q), cp(r)} is the set of 1-dimensional subspaces in cp(l).
If ^ possesses at least one natural representation then it possesses the
universal natural representation cpo such that any other natural represen-
tation is a composition of q>o and a linear mapping. The &F(2)-vector-
space underlying the universal natural representation (considered as an
abstract group with additive notation for the group operation) has the
presentation
= (vp, p e P | 2vp = 0; vp + vq= vq+vp for p,q eP;

vp + vq+vr=0, if {p,q,r} = leL)

 1.11 Representations of geometries 25
and the universal representation itself is defined by

cpo : p >-» vp for p G P

and
<Po-l*-> (vP,vq,vr) for {p,q,r} = / € L.

In this case F(^) will be called the universal representation module of

@. The following statement is rather obvious.

Lemma 1.11.1 Let & = &(Ln(2)) be the projective space of rank n—\ over
GF(2). Then V(&) is the n-dimensional natural module for Ln(2). D

The next result, which is also rather standard, was originally proved in
[Ti74] in a more general form.

Lemma 1.11.2 Let & = ^(Sp 2n (2)) be the symplectic geometry of rank n.
Then V(^) is the (2n + l)-dimensional orthogonal module of Sp2n(2) =
O2w+i(2). a
Natural representations of geometries usually provide a nice model
for geometries and "natural" modules for their automorphism groups.
Besides that, in a certain sense natural representations control extensions
of geometries. Below we explain this claim.
Let ^ be a geometry of rank at least 3 with a string diagram such
that the residue of type {1,2} is a projective plane of order 2, so that the
diagram of ^ has the following form:
X
2 2

Let G be a flag-transitive automorphism group of ^. Let xi be a point of

^ (an element of type 1), G\ be the stabilizer of xi in G and Jtif = res^(xi).
Then the points and lines of ^f are the lines and planes of ^ incident to
xi. Let <2i be the kernel of the action of G\ on Jtif. Then clearly G\/Q\
is a flag-transitive automorphism group of 3tf. Let R\ be the kernel of
the action of Q\ on the set of points collinear to xi (incident with xi to
a common line) and suppose that U = Qi/Ri is non-trivial. Let x2 be
a line incident to xi and {xi9y\9zi} be the points incident to x2. Since
every q e Q\ stabilizes both xi and x2, it either stabilizes x2 pointwise
permutes y\ and z\. Moreover the latter possibility must hold for some
q since G is flag-transitive and U =fc 1. Thus U is an elementary abelian
2-group and the module U* dual to U is generated by 1-dimensional
26 Introduction
subspaces, one for every point in J4f. Now if X3 is a plane incident to x\,
then the points and lines of ^ incident to X3 form a projective plane n of
order 2. It is easy to deduce from the flag-transitivity and the condition
U =f= 1 that the stabilizer of X3 in G induces the automorphism group
1,3(2) of n. In its turn this implies that the action induced by Q\ on the
set of points incident to x$ is of order 4 and we have the following.

Proposition 1.11.3 In the above notation if U = Q1/R1 is non-trivial, then

it is an elementary abelian 2-group and the module U* dual to U supports
a natural representation ofJ4? = res^(xi), in particular U* is a quotient of
the universal representation module F p f ) . •

When we follow an inductive approach to classification of geometries,

we can assume that J f and its flag-transitive automorphism groups are
known and we are interested in geometries ^ which are extensions of J f
by the projective plane edge in the diagram. Then the section Q1/R1 is
either trivial or related to a natural representation of Jf. In particular
this section is trivial if Jf7 does not possess a natural representation. In
practice it often happens that in this case there are no extensions of J f
at all.
For various reasons it is convenient to consider a non-abelian version of
natural representations. The universal representation group of a geometry
^ with 3 points on every line has the following definition in terms of
generators and relations:

R(9) = (zp, P€P\z 2p = l zpzqzr = 1 if {p,q,r} = / € L).

It is easy to observe that V(<&) = R(<S)/[R(<&\R(<&)]. Notice that gen-
erators zp and zq of R(&) commute whenever p and q are collinear. It
is straightforward from this observation that R(#(L n (2))) = F(^(L n (2))).
Less trivial but still not difficult to prove is the equality R(&(Sp2n(2))) =
F(^(S/?2n(2))). There are geometries whose universal representation
groups are non-abelian. In particular the geometries ^{JA\ @(BM) and
^(M) have non-trivial representation groups while their representation
modules are trivial.

1.12 The stages of classification

Our interest in P- and T-geometries originated from the classification
of distance-transitive graphs of small valencies. In [FII86] within the
classification of distance-transitive graphs of valency 7 we came across
the intersection arrays i(l) and i(2):
 1.12 The stages of classification 27

168 336 336 84

A graph T(Mat2i) with intersection array i(l) was discovered in [Big75].

The vertices of this graph are the blocks of the Steiner system S(5,8,24)
missing a pair of points; two vertices are adjacent if the corresponding
blocks are disjoint. The automorphism group of Y(Mat2i) is isomorphic
to Aut Matu where Mat22 is the sporadic Mathieu group of degree 22.
Every graph with the intersection array i(2) must be a 3-fold antipodal
cover of a graph with intersection array i(l). An example F(3 • Mat2i) of
such a graph was constructed in [FII86]; its automorphism group is a
non-split extension of Aut Mat22 by a normal subgroup of order 3.
Let H = Mat22 or 3 • Main. It can be deduced directly from the
intersection arrays i(l) and i(2) (Section 4 in [Iv87]) that T(H) contains
a Petersen subgraph. Let &(H) be the incidence system whose elements
of type 3, 2 and 1 are respectively vertices, edges and Petersen subgraphs
in T(H) and where incidence relation is via inclusion. Then ^(H) is a
geometry with the diagram
p . o P
3
* 2 2 1

and ^(3 • Mat2i) is a cover of ^(Mat22).

It was proved in [Sh85] that &(Mat22) and ^(3 • M^22) are the only Pa-
geometries (recall that in the present volume all P- and T-geometries are
flag-transitive by definition). In [Sh85] for the first time ever the strategy
for classification of geometries in terms of their diagrams based on
consideration of amalgams of parabolic subgroups and their completions
was applied. By now this is a commonly accepted strategy for studying
groups and geometries.
By the result of the classification of P3-geometries, every flag-transitive
automorphism group of a Pn-geometry for n > 4 must involve Main as
a section. Using this clue rank 4 geometries y(Mat2?>\ ^(Coi), ^{J^) and
a rank 5 geometry &(BM) were constructed in [Iv87]. The geometries
y(Mat23) and ^(J 4 ) as well as ^{Mat2i) were mentioned in [Bue85]. The
point residues in ^{Matii) and ^(Coi) are isomorphic to ^(Matii) while
in ^{J^) they are isomorphic to ^(3 • Matii)-
Let ^ be a Pw-geometry for n > 3 and G be a flag-transitive automor-
phism group of <§. The derived graph A = A(^) has the elements of type
28 Introduction
n in ^ as vertices and two vertices are adjacent if they are incident to a
common element of type n — 1. The graph A is locally projective of type
(n, 2) (as defined below) with respect to G, the girth of A (the length of
a shortest cycle) is 5 and unless ^ = y(Mat2?), the kernel Gi(x) at every
vertex x is non-trivial. A graph F is said to be a locally projective graph
of type (n, q) with respect to a subgroup G in its automorphism group if
G is vertex-transitive and for a vertex x the stabilizer G(x) of x in G acts
on the set T(x) of neighbours of x in F as Ln(q) (possibly extended by
some outer automorphisms) acts on the set of 1-dimensional subspaces
of an n-dimensional GF(g)-space. The kernel of the action of G(x) on
T(x) is denoted by Gi(x).
Locally projective graphs of girth 4 with non-trivial kernels at vertices
were classified in [CPr82] and it was believed for a while that no such
graphs of girth 5 exist. In [Iv88] and [Iv90] the classification problem of
locally projective graphs of girth 5 with non-trivial kernels at vertices was
reduced to the classification of P -geometries. This brought an additional
interest in P -geometries and their derived graphs.
The local analysis needed for the classification of P4-geometries was
carried out in [Sh88]. It was shown that the amalgam of maximal
parabolic subgroups associated with a flag-transitive action on a P4-
geometry is isomorphic to one of five amalgams s/i = {G{ | 1 < i < 4},
1 < j < 5. Here s/1, stf1 and J / 3 are realized in the actions of Mat2^
C02 and J4 on P -geometries associated with these groups. For k = 1 and
2 the amalgam j / 3 + f c possesses a morphism onto s/k whose restriction
to G3+fc is an isomorphism onto Gk for 2 < i < 4 and whose restriction
to G\+k is a homomorphism with kernel of order 3. More precisely G\+k
is the universal completion of the amalgam {G\ n Gk | 2 < i < 4}, so
that 01{G\Jrk/02{G\+k)) S 3 • Mat22 while O2(G\/O2(Gkl)) S Mat22. This
means that every geometry ^ corresponding to stf3+k (if it exists) has
point residues isomorphic to ^(3 • Mat22) and the universal cover of ^ is
the universal 2-cover of ^{Mat2i) or &(Co2) for k = 1 or 2, respectively.
Thus the main result of [Sh88] reduces the classification of P4-
geometries to calculation of the universal 2-covers of ^(Mat2^\ &(Co2)
and ^{JA)- The former of the geometries was treated in [ISh90a]. This
geometry contains a subgeometry 3tf = &(Altj). Using the simple con-
nectedness result for the subgeometry it was shown that the geometry
itself is simply connected. Furthermore, if ^ is a proper 2-cover of
<&(Mat23) (i.e. a 2-cover which is not a cover) then a connected com-
ponent of the preimage of ^f in ^ is a T3-geometry which possesses a
morphism onto <&(Alti). Using coset enumeration with a group given in
 1.12 The stages of classification 29
terms of generators and relations it was shown that no tilde analogue of
^(Alt-j) exists and hence no proper 2-cover of ^{Mati^) exists either.
The universal 2-cover of ^{Coi) was determined in [Sh92]. First, by
triangulating cycles in the collinearity graph of the geometry the latter
was proved to be simply connected. Let cp : ^ —• ^(Coi) be a proper 2-
covering of geometries. Then a connected component tf of the preimage
in 5 of a subgeometry 3tf = <&(Sp6(2)) in &(Co2) is a T3-geometry
and cp induces its morphism onto Jf. The amalgam S) of maximal
parabolics in a flag-transitive automorphism group of 2tf (particularly
in the action on tf of its stabilizer in a flag-transitive automorphism
group of ^) is specified up to isomorphism. It is possible to write down
an explicit presentation for the universal completion U{Si) by modifying
the Steinberg presentation for Spe(2). By means of coset enumeration
on a computer it was shown that JJ{2) = 37 • Spe(2). Based on this
result a 2-cover ^(3 23 • Coi) of ^{Coi) was constructed and its 2-simple
connectedness was established.
The geometry ^{J^) contains the ^-geometry ^(MatiA) as a subgeom-
etry and the simple connectedness question for y{J*) heavily depends on
that for @(Mat24). First the amalgams of maximal parabolics associated
withflag-transitiveactions on T3 -geometries have been classified and then
by means of coset enumeration on a computer the universal completions
of these amalgams were found. The result (presented in an unpublished
preprint [ISh89b]) was the complete list of T3-geometries: ^(3 7 • Sp6(2))
as in the above paragraph and the sporadic geometries ^{Mati^) and
&(He) constructed in [RSt84]. In the case when the stabilizer of a point
induces Sym^ on the corresponding residue an independent classification
was achieved in [Hei91]. Earlier it was shown in [Row89] and indepen-
dently in [Tim89] that in this case the order of the Borel subgroup is
either 29, realized in ^(3 7 • Sp6(2)) or 210, realized in <g(Mat1A) and <0(He).
If the stabilizer of a point induces Alts on the corresponding residue, then
the geometry must have a 1-covering onto the exceptional C3-geometry
^(Alt7). It was proved in [ISh90a] (see also [GM93]) that tilde analogues
of 9(Alt7) do not exist.
After <&(Mat24) was proved to be simply connected, the simple connect-
edness question for y{J*) was attacked in [Iv92b]. In that paper instead
of trying to triangulate cycles in the collinearity graph of the geometry
a different graph X called the intersection graph of subgeometries was
considered. The vertices of X are the ^(Mat24)-subgeometries in ^(JA)
with two of them being adjacent if they have the maximal possible num-
ber (namely 7) of common points. Using the simple connectedness of
30 Introduction
the subgeometries it was shown that every covering cp of ^(JA) induces
a covering q>1 of £ with respect to which all triangles are contractible.
Finally it was shown that every cycle in X can be triangulated, which
means that cpl and cp must be isomorphisms. About the same time the
triangulability of Z was established in [ASeg91] within the uniqueness
proof for J4.
The idea of studying triangulability of cycles in intersection graphs
of various families of simply connected subgeometries turned out to be
rather fruitful. In [Iv92d] the simple connectedness of the T4-geometry
&(Coi) was established using its simply connected subgeometry ^(Coi).
The geometry was constructed in [RSt84] based on the maximal parabolic
geometry J f {Co\) of the same group constructed in [RSm80]. The sim-
ple connectedness of 3tf(Co{) was shown in [Seg88]. The Baby Monster
geometry &(BM) contains C4(2)-subgeometries ^(Spg(2)) and also F 4 -
buildings associated with the groups 2Ee(2). Consideration of the inter-
section graph E with respect to the latter family of subgeometries led to
the simple connectedness proof for ^(BM) in [Iv92c]. Detailed informa-
tion about the structure of Z from [Seg91] has played an important role
in the proof. Finally the simple connectedness of the T5-geometry ^(M)
of the Monster was shown in [Iv91a] via consideration of the intersection
graph with respect to the ^(fJM)-subgeometries. The triangulability of
that graph was established in [ASeg92].
A question which for a while looked rather intractable is the one
about the universal 2-cover of the Baby Monster geometry <&(BM). If
^ —• g(BM) is a proper 2-covering, then a connected component J^
of the preimage in ^ of a ^(5pg(2))-subgeometry from &(BM) must
be a ^-geometry possessing a morphism onto y(Sp%(2)). Motivated by
this observation all T-geometries which possess morphisms onto Cn(2)-
geometries were classified in [ISh93a]. It turned out that there is one
family of such geometries containing one Tn-geometry for every n > 2.
The full automorphism group of this Tn-geometry is a non-split extension
of an elementary abelian 3-group of rank [n2]2 = (2n - l)(2 n ~ 1 - l)/3
by Sp2n(2). Thus the only possibility for Jf is to be isomorphic to
^(3 3 5 • 5p8(2)). It was decided to try to construct a 2-cover of <g(BM)
similarly to the way ^ ( 3 2 3 • C02) was constructed. The following question
turned out to be crucial for the construction. Let G5 = 2 5+10+10+5 .L 5 (2)
be the stabilizer in BM of an element of type 5 in &(BM) and E =
2'2E(,(2)2 be the stabilizer of an F4-subgeometry. Is there always an
 1.12 The stages of classification 31
element q € 02(Gs) such that q e E \ £'? The affirmative answer to
this question was given in [ISh93b] and was independently checked in
[Wil93] using computer calculations. This enabled to construct a proper
2-cover ^ of &(BM). After that a very tight bound on the order of
the automorphism groups of any such 2-cover was established and it
became possible to deduce that ^ = ^(3 4371 • BM) is in fact the universal
2-cover.
Let us turn to the local structure (the structure of amalgams of maximal
parabolics) in Pn- and Tn-geometries for n > 4. In [Sh88] important
information on the structure of the subgroup Gn, including a bound on its
order, was deduced for the case of P -geometries. Later it was realized that
these results can be extended word for word to the case of T-geometries.
The structure of amalgams of maximal parabolic subgroups of flag-
transitive Tn-geometries was studied in [Row91], [Row92] and [Par92].
In [Row91] it was shown that if G is a flag-transitive automorphism
group of a T4-geometry such that the Borel subgroup of the action of
G\ on the 73-residue has order 210, then the Borel subgroup of G has
order 221 (which is the order of a Sylow 2-subgroup of Co\) or 225. In
[Row92] it was shown that if G is a flag-transitive automorphism group
of a T5-geometry and the Borel subgroup of the action of G\ on the
residual T4-geometry has order 221, then the Borel subgroup of G is of
order 246 (which is the order of a Sylow 2-subgroup of the Monster).
Proceeding by induction and assuming that all P- and T-geometries of
smaller rank are known, in view of (1.11.3), certain information on the
possible structure of a G\ -parabolic can be deduced from the knowledge
of natural representations of residual P- and T-geometries.
The universal natural representations of P- and T-geometries were
studied even before their importance for the local analysis was noticed.
It was shown in [ISh89a] that ^(Matr^) has no natural representation,
V(g{Mat22)) is the 11-dimensional Golay code module while F(^(3 •
Matu)) is the direct sum of V(&(Mat22)) and the natural 6-dimensional
GF(4)-module for S Ue(2) (the latter contains 3 • Mat2i)> It was shown in
[RSm89] that V(^(Mat22)) is also the universal representation module
for &(Mat24). In [ISh94a] the universal representation module for <&(Co2)
was identified with a 23-dimensional section of A = A/2A where A is the
Leech lattice. Using this result it was not difficult to show in [Iv92a] that
A itself is the universal representation module for ^(Coi). The equality
V(&(Coi)) = A was independently proved in [Sm92]. After that it was
shown that P- and T-geometries associated with "large" sporadic groups
32 Introduction
do not have natural representations. In [ISh90b] it was proved that there
are no such representations for ^(J4)- In [ISh94a] &(BM) was proved
to have no natural representation. Since &(BM) is a subgeometry in
^(M), the latter has no natural representations either (the question about
existence of such representations was posed in [Str84]). The universal
representations of the T-geometries from the symplectic series were
determined in [Sh93], where the equality V(&(323 • Co2)) = V(<g(Co2))
was also established. The latter equality was used in [ISh94a] to show
that ^(3 4371 BM) does not have a natural representation. For a long time
the geometry &(He) was known to possess a natural representation in
a 51-dimensional irreducible GF(2)-module [MSm82]. It was established
by B. McKay (private communication) using computer calculation that
V(y(He)) is in fact 52-dimensional.
Let G be one of J4, BM and M, and ^ = ^(G). Then the elements
of type i in ^ are certain elementary abelian 2-subgroups in G of rank
i with the incidence relation defined via inclusion. From this description
it is immediate that G is a quotient of R(<g). The non-triviality of the
representation group of ^ explains in a sense why the original proofs
for the triviality of V(<&) were rather complicated. It turned out to be
easier to work with the whole representation group R(@) and to show
the equality [R(9),R(9)] = R(&) (which is of course equivalent to the
triviality of V(<&)). Recently the precise structure of R(<&) was determined
in [IPS96] and [ISh97]: R(&(G)) is J4, 2 • BM and M for G s J4, BM
and M, respectively.
Let si = {Gj; | 1 < i < n) be the amalgam of maximal parabolics
associated with a flag-transitive action of a group G on a P- or T-
geometry ^. Since the lists in (1.10.1) and (1.10.2) are known to be closed
under taking universal covers, in order to complete the classification it
is sufficient to show that si is isomorphic to the amalgam associated
with an action on a geometry from these lists. This statement was proved
in [ShSt94] for the case when ^ is a T-geometry having ^(3 7 • Sp6(2))
as a residue. Using the results from [Sh88], their generalizations for
T-geometries and the results from [Row91], [Row92], [Par92], one can
deduce some detailed information on the structure of a Gn-parabolic,
which restricts considerably the possibilities for its chief factors. Let us
assume that the residue Jf = res^(xi) is known (as well as its natural
representations) and let Q\ and JRI be as in (1.11.3). Since the dual
of Qi/Ri supports a natural representation of tf we have only a few
possibilities for the structure of this factor. In particular Q\ = R\ if
^f does not have natural representations (which happens when Jf is
 1.13 Consequences and development 33

one of <Z(Mat23), <0{JA\ &(BM), <g(34m • BM) and ^(M)). The equality
Q\= R\ must also hold in the case J f = &(He) since otherwise the Borel
subgroup of G would be of order at least 2 61 which contradicts [Row91].
On the other hand the equality Qi = R\ leads to a contradiction with
the information on the chief factors of Gn. Thus Qi/Ri is non-trivial and
we are left with the following four possibilities: (a) ^ is a T4-geometry
and Gi ~ 2n.Mat24', (b) ^ is a P 5 -geometry and G\ ~ 2+ +22 .Co 2 ; (c)
^ is a P5-geometry and Gi ~ (2++22 x 323).Co2; (d) ^ is a T5-geometry
and G\ ~ 2++24.Coi. For the former three cases the uniqueness of the
amalgam $4 was proved in an unpublished work of S.V. Shpectorov while
for the latter case it was established in [Iv92a].

1.13 Consequences and development

In this section we discuss some results which were proved either using
the classification of P- and T-geometries or under the inspiration of this
classification.

Maximal parabolic geometries

Let G be one of the following sporadic simple groups: Mat24, Cou

M, M(24), Mat22, Co2, BM and J 4 . Let 3f(G) be the maximal parabolic
geometry of G as introduced in [RSm80] with one of the following
diagrams:

6
n

o
2 6
P

o o 6-
2 2 6
34 Introduction

Jf (Co2) :

Jf(BM) :

Jf(J 4 ) :

Here o -—o denotes the geometry of 2- and 1-element subsets in

a set of size 5 in which two subsets of different size are incident if they
are disjoint. The elements of type 1 in the geometry with the diagram
o o are the maximal totally isotropic subspaces in the natural
symplectic module V of Sp^(2\ the elements of type 2 are the cosets of
the hyperplanes in V, an element S of type 1 and an element H + v of
type 2 are incident if S < H. The semidirect product V : Sp^{2) induces
on this geometry a flag-transitive action.
Let si be the amalgam of maximal parabolic subgroups associated
with the action of G on Jf(G). If G = Mat2^ Co\ or M then si contains
the amalgam 88 of maximal parabolic subgroups associated with the
action of G on the T-geometry ^(G) and if G = Mat2i, Co2, BM
or J\ then si contains the amalgam of maximal parabolic subgroups
associated with the action of G on the P-geometry ^(G). Furthermore
in both cases & generates the universal completion of si. By (1.10.1)
and (1.10.2) G is the universal completion of & and hence it is also the
universal completion of si. The simple connectedness of the geometry
Jf(M(24)) was established in [Iv95]). Thus we have the following result
which was proved in [Ron82] for G = Mat24 and in [Seg88] for G = Co\
and which answers the question posed at the end of [RSm80].
 1.13 Consequences and development 35

Theorem 1.13.1 Let G be one of the following groups:

Mat24, Cou M, M(24), Mat22, Co2, BM, J4.

Then the maximal parabolic geometry Jt(G) of G is simply connected. •

Locally projective graphs of girth 5

As we mentioned in the previous section, our interest in P -geometries

originated from the classification problem of locally projective graphs of
girth 5. As a direct consequence of the reduction results in [Iv88], [Iv90]
together with (1.10.1) we have the following.

Theorem 1.13.2 Let T be a locally projective graph of type (n, q), n > 2,
with respect to a group G. Suppose that the girth ofT is 5 and G\(x) ^ 1
at every x € T. Then q — 2 and one of the following holds:
(i) F is the derived graph A(&) of a P-geometry & of rank n and G
is a flag-transitive automorphism group of&, where & is one of the
following y(Sym5), %(Mat22), <$(3 • Mat22\ ^{Co2\ $(323 • Co2\
9(J4), &(BM) and &(34m • BM);
(ii) n = 5, G = J4, the vertices ofY are the imprimitivity blocks of size
31 of G on the vertex set of A = A(^(J4)) in which two blocks are
adjacent in T if their union contains a pair of vertices adjacent in
A. •

The locally projective graphs of girth 5 with G\(x) = 1 were studied in

[IP98] (see (9.11.6)) and it turns out that the graph A(<g(Mat23)) occupies
a very specific position in the class of such graphs.

Uniqueness of sporadics and their extensions

The existence of the geometry ^(G) for G = J4, BM or M was proved

starting with very basic local properties of the group G. The information
on G needed is the structure of the centralizer C = CG(T) of a central
involution 1 in G and the fact that T is conjugate in G to involutions from
O2(C)\{T}. The maximal parabolic geometries of J4 and M were predicted
to exist in [RSm80] even before the groups themselves were constructed.
For this reason the classification of P- and T-geometries immediately
provides a uniform uniqueness proof of G as a group possessing the
properties needed to deduce the existence of ^(G) [Iv91b].
36 Introduction
Theorem 1.13.3 Let G be a non-abelian simple group containing an invo-
lution T such that C := CG(T) is of the shape
2*_+12.3 -Aut Mat229 2^22.Co2 or 21+24.Co1.
Suppose that CG(O2(C)) < O2(C) and that TG n O2(C) + {T}. Then in
each of the three cases G is uniquely determined up to isomorphism and is
isomorphic to J4, BM or M, respectively. •
Other uniqueness proofs can be found in [Nor80], [ASeg91] for J4,
[LSi77], [Seg91] for BM and [Nor85], [GMS89] for M.
Within the classification of P- and T-geometries and their natural
representations we have proved some interesting facts about linear repre-
sentations and non-split extensions of sporadic groups. As an illustration
we formulate the following result which can be deduced from [ISh93b].

Theorem 1.13.4 Let F be afield whose characteristic is not 2. Then BM

has a unique faithful representation of dimension 4371 over F which is
irreducible. If the characteristic of F is not 3 then the extension ofBM by
the corresponding 4311-dimensional F-module always splits, and for F =
GF(3) there is a unique non-split extension. •
Generators and relations

The classification of P- and T-geometries enabled us to obtain char-

acterizations of certain sporadic simple groups, stronger than the char-
acterization by the centralizer of an involution. The groups were proved
to coincide with the universal completions of certain of their subamal-
gams. This provided us with presentations of the groups involved (the
geometric presentations as they were termed in [Iv91a]). In the case of J4
the geometric presentation was proved in [Iv92b] to be equivalent to a
presentation for J4 conjectured by G. Stroth and R. Weiss in [StW88].
In the case ofBM and M the result establishes the correctness of the so-
called Y -presentations for these groups. The Y -presentations ([CCNPW],
[CNS88], [Nor90]) describe groups as specific factor groups of Coxeter
groups with diagrams having three arms originating in a common node.
The most famous is the 7555 diagram (below).
After the announcement of the geometric presentation of M at the
Durham symposium "Groups, Combinatorics and Geometry" in July
1990 [Iv92a] S.P. Norton [Nor92] proved its equivalence to the corre-
sponding Y -presentation. This resulted in the proof of the following
theorem conjectured by J.H. Conway [Con92].
 1.13 Consequences and development 37
Theorem 1.13.5 The Coxeter group corresponding to the Y555 diagram sub-
ject to a single additional relation
(abiCiab2C2abiC3)10 = 1

is isomorphic to the wreath product M\2 of the Monster group and a group
of order 2 (this wreath product is known as the Bimonster). •

/i a o /:

The correctness of the Y -presentation for BM is proved in [Iv94].

Construction of 34

As a consequence of the classification of P- and T-geometries we

have the following. Let G be one of the sporadic simple groups Mat23,
Mat24, He, C02, Co\, J4, BM and M. Then G is the universal completion
(and because of its simplicity it is the unique faithful completion) of
the amalgam si = \G\\\ 1 < i < n} of maximal parabolic subgroups
associated with the action of G on ^(G). In addition, unless G = Mat24
or He, the isomorphism type of si is uniquely determined by the chief
factors of the G, and by the indices fey = [G, : G, n Gj\. This means
that G can be defined as "a faithful completion of an amalgam si with
given chief factors and indexes fc,/\ Hence (at least in principle) one can
start from this definition to establish the existence of G and to deduce all
its properties including the simplicity. This would give an independent
construction of G together with its uniqueness proof. In its full extent
the approach was realized in [IMe93] for the fourth Janko group J4.
38 Introduction
In [IMe93] the amalgams Ji = {Mi,M 2 ,M3} such that
Mx ~ 2n.Mat24, M2 ~ 21O.L5(2), M 3 - [215].(Sym5 x L3(2))
and

[M2 : M12I = 31, [M3 : M 13 ] = 5, [M3 : M 23 ] = 10, [M23 : B] = 3

where M i ; = M,- n M7 and B = Mi n M 2 n M 3 have been studied.
An amalgam with these properties is contained in J4, where M\ is the
stabilizer of a ^(Ma£24)-subgeometry in ^(J4), M 2 is the stabilizer of a
vertex in the graph T as in (1.13.2 (ii)) and M 3 is the stabilizer of a
Petersen subgraph in F. These subgroups are universal completions of
their intersections with the amalgam of maximal parabolic subgroups
corresponding to the action of J4 on ^(c/4).
It was shown that up to isomorphism there exists exactly one amalgam
J( as above. After that it was shown that J( has a faithful completion.
This was done by constructing an isomorphic embedding of J( into
GLi 333 (C) (notice that 1333 is the dimension of the smallest faithful
complex representation of J4). Next it was shown that if G is a completion
of M, then the subgroup M\ when it acts naturally on the set of its
right cosets in G has exactly seven orbits, whose lengths /,-, 1 < i < 7,
were explicitly calculated (so that these lengths are independent of the
particular choice of the completion G). Thus every completion G of M
has the same order \M\\ • Yli=i U which turns out to be
221-33-5-ll3-23-29-31-37-43.
Moreover G is non-abelian and simple and CQ(Z) ~ 2^_+12.3 • Aut Matu
for an involution z in G. This means that G is the fourth Janko group
J4 according to the standard definition of the latter. Originally J4 was
constructed on a computer as a subgroup of GLn 2 (2) [Nor80].

Extended dual polar spaces

Another interesting class of geometries admitting flag-transitive actions

of sporadic simple groups is formed by extended (classical) dual polar
spaces. An extended dual polar space (EDPS for short) of order (s, t) has
diagram
c
1 s t t t t

where the leftmost edge denotes the geometry of all 1- and 2-element
 1.13 Consequences and development 39
subsets of an (5 + 2)-element set with respect to the natural incidence
relation; the residue of a point is a classical thick dual polar space.
Flag-transitive EDPS's of rank 3 were classified in [BH77], [DGMP],
[WY90] and [Yos91]. The representations of rank 3 flag-transitive EDPS's
were described in [Yos92]. Rank 4 EDPS's with the property that the sta-
bilizer of a point in a flag-transitive automorphism group acts faithfully
on the residue of that point were classified in [Yos94]. Further progress
in the classification of flag-transitive EDPS's was achieved in [Iv95],
[ISt96], [IMe97], [Iv98b], [Iv98a] and [Iv97]. An EDPS g is called affine
if it possesses a flag-transitive automorphism group G which contains a
normal subgroup T acting regularly on the set of points of S. In this case
T is called the translation group of $ with respect of G. The following
characterization of affine EDPS's is proved in [Iv98a].

Theorem 1.13.6 Let £ be an affine EDPS of rank n > 3 and 9 be the

residue in $ of a point. Then

(i) s = 2, so that 3} is the dual polar space associated with Sp2n-i{2)

or U2n-2(2),
(ii) there is a 2-covering $ —> $ where $ is an EDPS which is 2-simply
connected and affine,
(iii) the translation group of $ (as above) with respect to its full auto-
morphism group is the universal representation group R(2) of the
residual dual polar space 2. •

Thus the above result reduces the classification of affine EDPS's to

the calculation of the universal representation groups of the dual polar
spaces associated with <Sp2m(2) and l/2m(2). The precise structure of these
representation groups is known only for m = 2 and 3 but some partial
results are available also for larger m (see [Iv98a] for details).

In [Iv97] it is shown that there are exactly 19 flag-transitive EDPS's

which are not 2-covered by affine ones.

Theorem 1.13.7 Let $ be an EDPS of rank at least 3 which possesses

a flag-transitive automorphism group whose Borel subgroup is finite. Then
one of the following holds:

(i) there is a 2-covering $ —> $ where $ is affine;

(ii) $ is isomorphic to one of the 19 exceptional EDPS's whose diagrams
and full automorphism groups are given in the table.
40 Introduction

, U4(2).2

o <L_o = =o Sp6(2) x 2,

3 • t/4(3).22, l/4(3).22

c McL.2
1 3 9

c HS.2, Suz.2
1 9 3

c Sp8(2), 3 • M(22).2, M(22).2

1 2 2 2

c Co 2 x 2, Co 2
1 2 4 4

c M(24)
1 4 2 2

c M(24)
1 3 3 3

c M
1 9 3 3

c •o M(23) D
1 2 2 2

In what follows ${E) will denote the EDPS for which E is the commu-
tator subgroup of the automorphism group. In case we need to specify
the order (s,t) of the EDPS, we write #SAE)' Let $ be an exceptional
EDPS from (1.13.7), G be the automorphism group of $ and H be a
subgroup of G. Unless $ = S(HS) the action of H is flag-transitive if
and only if H contains the commutator subgroup of G; AutifS is the
only flag-transitive automorphism group of ${HS).
The classification strategy implemented in [Iv97] is the one developed
within the classification project for P- and T-geometries. This strategy is
based on studying the amalgams of maximal parabolics, representations
of residual geometries and the simple connectedness question. In addition
 1.13 Consequences and development 41
the following direct connections (A), (B) and (C) between EDPS's and
P- and T-geometries have played essential roles in the classification.
(A) Let S be an EDPS of rank n > 4 in which the residual rank 3
EDPS's are all isomorphic to £(U4(2)\ so that 5 = t = 2. Let A be the
graph on the set of elements of type 2 in $ in which two elements are
adjacent if they are incident to a common element of type 3 but not to
a common element of type 1. Then every connected component of A is
the derived graph of a Petersen geometry of rank n — 1. In this way the
EDPS's *(Af(22)), /(3-M(22)) and *(Af(23)) are related to P-geometries
<${Mat22\ 9(3 • Mat22) and &(Mat23), respectively.
(B) The EDPS £4a(M(2A)) is closely related to a geometry jf(M(24))
with the same automorphism group and diagram

The point residues in Jf (M(24)) are isomorphic to the T3-geometry

g(Mat24)- In [Iv95] the simple connectedness of Jf (M(24)) was proved
first and then use this result in the simple connectedness proof for
A>2(M(24)). It is worth mentioning that there is a geometry j f (3 • M(24))
with diagram
o o -o o
2 2 2 2
possessing a morphism onto J4?(M(24)) whose simple connectedness was
also established in [Iv95].
(C) The EDPS S(M) of the Monster group was constructed in [BF83]
and [RSt84]. The simple connectedness proof for S(M) in [IMe97] re-
duces the problem to the simple connectedness of the T5-geometry of
the Monster. That is, we show that the universal completion U of the
amalgam of maximal parabolic subgroups corresponding to the action of
M on S(M) contains subgroups C, N and L as in (1.10.4) which implies
that U = M. It was noted in [Iv96] that 3-local and 2-local parabolics
in the Monster are related via a subgeometry Jf (224 • Co\) in S(M) with
diagram

acted on flag-transitively by C/Z(C). In [IMe97] it was shown that the

universal cover of S(M) also contains Jf (224 • Co\) as a subgeometry. At
this stage we applied the main result of [ISt96] which proves the simple
connectedness of the EDPS ^(M(24)) which is also a subgeometry
42 Introduction
in S(M). It is worth mentioning that the geometry ^f(224 • Co\) is not
simply connected since it possesses a double cover [ISh98] (which might
or might not be universal).

1.14 Terminology and notation

In this section we fix our terminology and notation concerning groups
and their actions on graphs and geometries.
Let G be a group. Then 1 is the identity element, G# = G \ {1}, Z(G)
is the centre, G — [G, G] is the commutator subgroup and Aut G is the
automorphism group of G. Let g be an element in G (written g € G) and
if be a subgroup of H (written H < G). Then (g) is the subgroup of G
generated by g, CG(g) and CG(H) are the centralizers in G of g and if,
NG(H) is the normalizer of H in G. If NG(H)/CG(H) s Aut If then H
is said to be fully normalized in G. By G/H we denote the set of right
cosets of H in G. In the case where H is normal in G (written H < G), by
G/H we also denote the corresponding factor group. The core of H in G,
denoted by coreG(H) is the largest normal subgroup of G contained in
H, so that coreG(H) is the intersection of the conjugates of H in G. The
smallest normal subgroup in G which contains H is the normal closure of
if in G. Let p and g be different primes. Then OP(G) is the largest normal
subgroup in G which is a p-group (i.e. whose order is pk for some fe),
Op,<z(G) is the preimage in G of Oq(G/Op(G))9 OP(G) is the smallest normal
subgroup in G with the property that the corresponding factor group is a
p-group. By pn we denote the elementary abelian group of this order, so
that p also denotes the cyclic group of order p. When we write [pn], we
mean a group of order pn; 2\+2n denotes the extraspecial group of order
2 2n+1 of type e e {+,—}. Recall that a maximal abelian subgroup of 2£1+2n
has rank n + 1 if e = + and n if s = —. The symmetric and alternating
groups of degree n we denote by Symn and A/£n, respectively. If we want
to specify the underlying set X of size n, we write Sym(X) and Alt{X\
respectively. In the case where if is a group, naturally identified with a
subgroup in Symn (say if if = Symm x Symn-m for some 1 < m < n), then
ife denotes the subgroup of even permutations in H, i.e. He = H n 4/rn.
If we write G ~ Ai? we mean that G has a normal subgroup isomorphic
to (and identified with) A such that G/A is isomorphic to B (written
G/A = B). When we write G ~ A-B we mean that G does not split over
A, i.e. there is no subgroup if in G such that H n A = 1 and H = B.
Finally by writing G = A : 5 we mean that G is a semidirect product
of A and B with respect to a homomorphism of B into Aut A which
 1.14 Terminology and notation 43
is either clear from the context or irrelevant. If the homomorphism is
trivial we have the direct product Ax B.
If V is a vector space of dimension n over the field GF(q) of q elements,
where q = pa for a prime p, then TL(V), GL(V) and SL(V) denote
respectively the group of semilinear, linear and linear with determinant 1
bijections of V onto itself. A group G is said to be a linear group of V if
SL(V)< G<TL{V\
in which case V is said to be the natural module of G. When interested in
groups only we write TLn(q\ GLn(q) and SLn(q). By ATLn(q\ AGLn(q)
and ASLn(q) we denote the affine groups which are semidirect products of
V (considered as an elementary abelian p-group) and the corresponding
linear group. The actions of the groups TLn(q), GLn{q) and SLn(q) on
the set of subspaces of the underlying vector space V are denoted by
PTLn(q\ PGLn(q) and Ln(q), respectively. By PZLn(q) we denote the
extension of Ln(q) by the field automorphisms. A group H is said to be
a projective linear group if H is the image in PTLn(q) of a linear group
of V. The (doubly transitive) permutation action of a projective linear
group on the set of 1-dimensional subspaces of the underlying vector
space is said to be the natural permutation representation of the projective
linear group. By Sp2m(q) and Vm(q1^2) we denote the projective special
symplectic and unitary groups. By Oem(q) we denote the orthogonal group
of dimension m and type e e {+, —}; its commutator subgroup is denoted
by <£(«).
Let X be a set on which a group G acts by permutations (i.e. there
is a homomorphism of G into the symmetric group of £ clear from
the context). The image of x G I under g e G is denoted by xg. If
S = {x,y,...} is a subset of I then G[S] = G[x,y9...] and G(S) = G(x,y,...)
denote the setwise and the elementwise stabilizers of S in G. If g e G(X)
then g is said to fix X and if g e G[X] then g is said to stabilize X.
Similar terminology applies to subgroups of G. If H < G[E] then HE
denotes the permutation group induced by H in S. Suppose that G acts
transitively on X. Then an orbit of G on the set of ordered pairs of
elements of S is called an orbital. The orbitals containing (x, y) and (y, x)
are said to be dual to each other. An orbital which coinsides with its
dual is called self-dual.
For a set Z let 2Z be the power set of Z, i.e. the set of all subsets of S.
The symmetric difference operator
XAY =(X\JY)\(XCiY)
44 Introduction
s
provides 2 with a GF(2)-vector-space structure. If G is a permutation
group on £ then 2 s is the permutational GF(2)-module of G. Let F be
the binary function on 2 s taking values in GF(2) and defined by

3 if \XnY\ is even,
1 otherwise.

Then F is non-singular and bilinear and will be called the parity form
on 2E. The parity form is invariant under every permutation group G
on Z and if G = Sym(L) then F is the unique invariant form which is
non-singular and bilinear. A subset of £ will be called even if it contains
an even number of elements.
A partition of E is a set Sf = {Si,..., Sn} of subsets of E such that every
d e l belongs to exactly one S*. If 9~ = {Ti,..., Tm} is a partition of 2
such that every S,- is the union of some T) then we say that &~ refines
Sf and that £f is refined by F. Sometimes we identify a subset S c H
with the partition {S,E\S}. We write (nj1^2...) for a multiset of integers
in which n\ appears k\ times, nj appears &2 times etc., where normally
n\ > «2 > ... and nt is written instead of n\.
We will follow the terminology and notation concerning actions of
groups on geometries as introduced in Section 1.3 with the following
addition. Let ^ be a geometry of rank n with a string diagram on which
types increase rightward from 1 to n and let x, be an element of type
i where 1 < i < n. Then res^(xi) and res^(x;) are the residues in 9 of
a flag of type {l,2,...,i} and a flag of type {i,i+ l,...,n} containing x,-,
respectively. If G is a (flag-transitive) automorphism group of ^ and
G, = G(XJ) is the stabilizer of x, in G then G, denotes the action induced
by Gi on res^(x,), G\ denotes the kernel of the action of G,- on res^(x,-)
for {s,S} = {+,-}.
Let F be a graph, which is assumed to be undirected and locally finite.
The latter means that every vertex is adjacent to a finite number of other
vertices. The vertex set of T will be denoted by the same letter T and we
will write E(T) and Aut T for the edge set of T and for its automorphism
group, respectively. The vertices incident to an edge will be called the
ends of the edge. For a positive integer s an s-arc (or an arc of length s)
in F is a sequence (xo,xi,...,xs) of s+1 vertices such that {x,-_i,x,-} G E(T)
for 1 < i < s and x* ^= x,_2 for 2 < i < s. Such an arc is said to originate
at xo, to terminate at xs and to join xo and xs. If xo = xs then the arc
is called a cycle of length s or simply an s-cycle. The girth of a graph is
the length of its shortest cycle. A graph whose vertices are the edges of
 1.14 Terminology and notation 45
F with two of them adjacent if they are incident to a common vertex of
F is called the line graph of F.
The graph is connected if any two its vertices can be joined by an arc.
A graph F is called n-partite or multipartite if its vertex set possesses a
partition { F | 1 < i < n} whose members are called parts such that if
{x,y} e E(T) with x e P and y e Tj then i £ j . 2-Partite graphs are also
called bipartite. It is a standard fact that a graph is bipartite if and only
if it does not contain cycles of odd length.
For x9y e F let d(x,y) denote the distance between x and y in the
natural metric of F, that is, the number of edges in a shortest arc joining
x and y. If S and A are subsets in the vertex set of F then the distance
d(E,A) between S and A is the minimum among the distances d(x,y)
for x G S and y € A. The diameter d of F is the maximum of distances
between its vertices. Put
r,-(x) = {y\ye T,d(x9y) = i} for 0 < i < d.

We will usually write F(x) instead of Fi(x).

The number of vertices adjacent to x, that is |F(x)|, is called the valency
of x. If k = |F(x)| is independent of the choice of x € F then F is called
regular of valency k.
Let 3 be a subset in the vertex set of F. The subgraph of F induced on
S has S as set of vertices and its edges are all the edges of F with both
ends contained in S. A subgraph in which any two vertices are adjacent
is called a clique; a subgraph in which no two vertices are adjacent is
called a coclique. Let A be a graph. Then a graph F is said to be locally
A if for every x € F the subgraph in F induced by Fi(x) is isomorphic
to A.
A cycle (xo,xi,...,x s ) is said to be non-degenerate if for every i,j with
0 < i < j < 5, the distance d(xt, Xj) in F is equal to the distance between
x, and Xj in the cycle, which is
min {j — i9i + s — j}.
Let S be an induced subgraph in F. Then S is said to be geodetically
closed if whenever x,y eE with d(x,y) = i, all arcs of length i joining x
and y in F are contained in S. If in addition S contains all arcs of length
1 + 1 joining x and y then S is called strongly geodetically closed. A graph
F can be considered as a 1-dimensional simplicial complex [Sp66]. The
fundamental group of F is by definition the fundamental group of the
corresponding complex.
If F and F' are graphs then a surjective mapping q> : V —• F is called
46 Introduction
a covering of graphs if for every xr G V the restriction of cp to V(xf) is a
bijection onto T((p(x')). Let x e F . Then for every xr e tp"1!*) a n d an s-
arc X = (xo = x,*i,...,xs) originating at x there is a unique s-arc X~l(xf)
originating at xr which maps onto X. Here X~l(xr) = (XQ = x\x\,...,x's)
and xj is the unique vertex in q>~l(xi) Pi F^x-^) for 1 < i < s. If X is a
cycle then X~l(xf) might or might not be a cycle. If X" 1 ^') is a cycle for
every xr e cp~l{x) then X is said to be contractible with respect to cp. A
covering is characterized by the subgroup in the fundamental group of
the graph generated by the contractible cycles [Sp66].
Suppose that X = (xo, xi, •••> *s = *o) is an s-cycle and for 0 < i < j < s
let Y = (yo = Xi,y\,...,yt = Xj) be an arc joining x,- and Xj. Then
we say that X splits into Xi = (xo,...xuyu...,yt-i9Xj,Xj+u...9xs) and
%2 = (xuXi+i,...,Xj,yt-u...,yuXi) (in this case X is the sum of Xi and X2
modulo 2). It is easy to see that if both X\ and X2 are contractible then
X is also contractible. If in its turn X\ splits into cycles X3 and X4, then
we say that X splits into the cycles X2, X3 and X4. Thus inductively we
can define the splittings of X into any number of cycles. If X splits into a
set of triangles then X is said to be triangulable. A graph is triangulable
if every of it cycles is triangulable. If Y is triangulable and every triangle
in F is contractible with respect to cp then cp must be an isomorphism and
hence T is triangulable if and only if its fundamental group is generated
by the triangles. The following sufficient condition for triangulability is a
straightforward generalization of Lemma 5 in [Ron81a].

Lemma 1.14.1 Let T be a graph of diameter d and suppose that for every
i, 2 <i < d, the following two conditions hold:
(i) if y G Ti(x) then the subgraph in T induced by T(x) Pi Ti-i(y) is
connected;
(ii) ify,z e Ti(x) and z e T(y) then i(r(x)nr M (j/),r(x)nr w (z)) <
1.
Then T is triangulable, which means that its fundamental group is generated
by the triangles. The condition (ii) is implied by the following:
(iii) if y £ T/(x) then every vertex from T(x) \ Ti-i(y) is adjacent to a
vertex from Fi_i (y) n T(x). •
Let G < Aut F be an automorphism group of F. Then G is said to be,
respectively, vertex-transitive, edge-transitive or s-arc-transitive if it acts
transitively on the vertex set, edge set or set of s-arcs. In these respective
cases F is also called vertex-transitive, edge-transitive and s-arc-transitive.
 1.14 Terminology and notation 47
If the action of a group is s-arc-transitive but not (5 + l)-arc-transitive,
we will say that it is strictly s-arc-transitive. Notice that if G acts 1-arc-
transitively on F then F can be identified with a self-dual orbital of the
action of G on the vertex set of the graph. For x e F let G(x) be the
stabilizer of x in G. The permutation group G(x)r(x) is known as the
subconstituent of G on F. For an integer i define

G,(x) = f l G(y)
d(x,y)<i

which clearly is a normal subgroup of G(x). Then G(x)r^ is abstractly

isomorphic to the factor group G(x)/Gi(x). If {x,y} is an edge, we put
Gi(x,y) = Gi(x)nGi(y).
Let F be a graph and G be a vertex-transitive automorphism group of
F. Suppose that G preserves on F an imprimitivity system &. Define A to
be a graph whose vertices are the imprimitivity blocks from 0& and two
such blocks Bi, B2 are adjacent if there is {x,y} e E(T) such that x G B\
and y € #2- Then A is said to be constructed from F by factorizing over
the imprimitivity system £&.
The standard double cover 2 • F of F is a graph on 2 • F := {(x, a ) | x G
F,a € {0,1}} with vertices (x,a) and (}/,/}) being adjacent if and only if
{x,y} € E(T) and a =£ /?. Then cp : (x,a)i-^xisa covering of 2 • F onto
F. If C is a cycle of length m in F then q>~l{C) is a disjoint union of two
cycles of length m when m is even and it is a cycle of length 2m if m is
odd. This means that a cycle in F is contractible with respect to q> if and
only if it has even length. In particular F and 2 • F have the same girth if
the girth of F is even. Furthermore, if F is bipartite and connected then
2 • F consists of two connected components, each isomorphic to F. If F is
non-bipartite and connected then 2 • F is bipartite and connected. If G is
a vertex-transitive automorphism group of F then the group G generated
by the automorphisms g : (x, a) 1—• (xg, a) for every g e G together with
the automorphism d : (x, a) i-> (x, 1 — a) is isomorphic to G x 2 and it acts
vertex-transitively on 2 • F. The pairs {(x,0), (x, 1)} form an imprimitivity
system of G and F can be reconstructed from 2 • F by factorizing over
this system.
The action of a group G < Aut F on F is distance-transitive if for every
0 < i < d the group G acts transitively on the set
rt = {(x9y)\x,yer9d(x9y) = i}.
A graph which possesses a distance-transitive action is called a distance-
transitive graph. If F is distance-transitive then for every i, 0 < i < d,
48 Introduction
the parameters
a = irwoo n r(x)|, a,- = |r,oo n r(x)|, 6, = 11V1GO n r(x)|
are independent of the choice of the pair x, y € F satisfying d(x,y) = i.
Clearly in this case F is regular and Ci + at + bi = |F(x)| = k is the valency
of F. The sequence
i(F) = {bo = k, b\,..., bd—i j Ci = 1, c2,..., Cd}
is called the intersection array of the distance-transitive graph F. If we
put kt = |F,(x)| for 1 < i < d then

Ci'C2' ... * Ci '

To represent the decomposition of a distance-transitive graph with respect

to a vertex we draw the following distance diagram:
ax a2
k

We draw similar diagrams for non-distance-transitive actions. Let G

be a group acting on a graph F and x be a (basic) vertex. The suborbit
diagram (with respect to x) consists of ovals (or circles) joined by curves
(or lines). The ovals represent the orbits of G(x) on the vertex set of F.
Inside the oval which represents an orbit Z, (the £ r oval) we show the
size of S, or place its name explained in the context. Next to the Z r oval
we show the number nt (if non-zero) of vertices in E, and adjacent to a
given vertex yi € S,-. On the curve joining the £,- and Z;-ovals we put
the numbers fyj and n7, (called valencies.) Here ni; (appearing closer to
the £r-oval) is the number of vertices in E, adjacent to yi. Clearly
\Lt\ - ntj = \Zj\ • nji

and we draw no curve if ntj = njt = 0. Normally we present the valencies

n, and ntj as sums of lengths of orbits of G(x, yt) on the vertices in S, and
E, adjacent to yt. When the orbit lengths are unknown or irrelevant, we
put the valencies into square brackets. Generally the suborbit diagram
depends on the orbit of G on F from which the basic vertex x is taken.
Even if a graph is not necessarily distance-transitive, we use the notation
ct, at, bt if the corresponding parameters are independent of the choice
of a pair of vertices at distance i.
 2
Mathieu groups

In this chapter we construct the Mathieu groups and study their basic
properties. We construct the largest Mathieu group Matu as the automor-
phism group of the (binary extended) Golay code defined in Section 2.1.
In Section 2.2 we construct a Golay code as the quadratic residue code
over GF(22>). In Section 2.3 we show that a minimal non-empty subset in
a Golay code has size 8 (called an octad). Moreover the set of all octads
in a Golay code forms the block set of a Steiner system of type S(5,8,24).
The residue of a 3-element subset of elements in a Steiner system of type
5(5,8,24) is a projective plane of order 4. In Section 2.4 we review some
basic properties of the linear groups and in Sections 2.5 and 2 6 we define
the generalized quadrangle of order (2,2) and its triple cover which is the
tilde geometry of rank 2. In Section 2.7 we prove uniqueness of the pro-
jective plane of order 4 and analyse some properties of the plane and its
automorphism group. This analysis enables us to establish the uniqueness
of the Steiner system of type S(5,8,24) in Section 2.8. The Mathieu group
Mat24 of degree 24 is defined in Section 2.9 as the automorphism group
of the unique Golay code. The uniqueness proof implies rather detailed
information about Mat24 and two other large Mathieu groups Mat^ and
Mat22- In Section 2.10 we study the stabilizers in Mat24 of an octad, a
trio and a sextet. In Section 2.11, analysing dodecads in the Golay code
and their stabilizers in Mat24, we introduce the little Mathieu groups.
In Sections 2.12, 2.13 and 2.14 we classify the subgroups in Mat24 of
order 2 and 3 and determine octads, trios and sextets stabilized by such
a subgroup. In Section 2.15 we study the action of the large Mathieu
groups on the Golay code and on its cocode. Finally in Section 2.16
we describe the generalized quadrangle of order (3,9) in terms of the
projective plane of order 4.

49
50 Mathieu groups
2.1 The Golay code
Let X be a finite set of elements. A binary linear code %> based on X is a
subspace of the power set 2X (considered as a GF(2)-vector-space).
In general a linear code over GF(q) is a triple (V,X,^) where V is a
GF(g)-vector-space, X is a basis of V and ^ is a subspace in F. It is
obvious that in the case q = 2 this definition is equivalent to the above
one. In what follows unless explicitly stated otherwise, when talking about
codes we always mean binary linear codes.
The size of X is called the length of a code ^ based on X. A code
is even (respectively doubly even) if the number of elements in every
non-empty subset in ^ is even (respectively divisible by 4). The minimal
weight of # is the number of elements in a smallest non-empty subset
in ^ . The dual code <&* of ^ is the orthogonal complement of ^ with
respect to the parity form:

<T = {A | A e 2X, \A n B\ is even for all B € # } .

Since |AnJB| = \(\A\ + \B\ — \AAB\), every doubly even code is contained
in its dual.

If ^ = %>* then the code is called self-dual. The following characteriza-

tion of self-dual codes is immediate since the parity form is non-singular
and bilinear.

Lemma 2.1.1 A code ^ is self-dual if it is totally singular with respect to

the parity form and dim ^ = \X\/2. •

By the above lemma the length of a self-dual code is always even.

Definition 2.1.2 A code <$ is called a Golay code (extended binary Golay
code) if ^ is self-dual of length 24 with minimal weight greater than or
equal to 8.

By (2.1.1) a Golay code is 12-dimensional and we will denote it by ^n-

We will show that up to isomorphism there exists a unique Golay

code whose automorphism group is the sporadic Mathieu group Mat24
and that the Golay code is doubly even. We start in the next section by
constructing a Golay code as the quadratic residue code over GF(23).
 2.2 Constructing a Golay code 51
2.2 Constructing a Golay code
In this section we give an elementary construction of a Golay code.
That is, we will construct such a code as the quadratic residue code over
GF(23). The construction is elementary since quadratic residue codes
over GF(q) can be constructed for all q = — 1 mod 8.
Consider the field GF(23) of size 23 whose elements will be denoted by
integers from 0 to 22 so that the field operations are the usual addition
and multiplication modulo 23. The set GF(23)* of non-zero elements in
GF(23) is the union of the set Q of squares and the set N of non-squares,
where
Q = {1,2,3,4,6,8,9,12,13,16,18},

N = {5,7,10,11,14,15,17,19,20,21,22}.
Let 9 be the projective line over GF(23) considered as the union of
GF(23) and the formal symbol oo. The group L = 1^(23) acts on & by
means of transformations
ax + b
cx + d
where ad — bce Q. We introduce three elements of L:
t : x»—• x + 1; s:x\-+2x\ T : X I - > - 1/X.

Let T and 5 be the subgroups of L generated by t and s, respectively.

The following statement is immediate.
Lemma 2.2.1
(i) T is cyclic of order 23 acting regularly on GF(23) = 9 \ {oo};
(ii) S is cyclic of order 11, if acts regularly on Q and N, normalizing
T;
(iii) S is the elementwise stabilizer in L of the pair {oo, 0} ;
(iv) the stabilizer L(oo) of oo in L is the semidirect product of T and S
and it acts transitively on the set of unordered pairs of elements of
GF (23);
(v) T is an involution which normalizes S, maps oo onto 0 and Q onto
N. •
The following result is also rather standard, but we present a brief
proof for the sake of completeness.

Lemma 2.2.2
(i) L is generated by t, s and T;
52 Mathieu groups
(ii) L acts doubly transitively on 0* and has order 24 • 23 • 11;
(iii) L acts transitively on the set of 3-element subsets of 0>, and the
elementwise stabilizer in L of every such subset is trivial;
(iv) the setwise stabilizer in L of a 3-element subset of 8P is cyclic of
order 3 acting fixed-point freely on 0*.

Proof. Since L(oo) acts transitively on 9 \ {oo} and T does not fix oo
we have (i) and (ii). By the double transitivity every L-orbit on 3-element
subsets of 0 contains a triple {oo,0,a} for some a G GF(23)*. Under the
action of S the triples of this shape split into two orbits depending on
whether a e Q or a e N. Since T stabilizes {oo, 0} and permutes Q and JV,
these two orbits are fused and we obtain (iii). Finally (iv) follows from
(iii) and the order of L. •
For a e GF(23) put Na = {n + a \ n e N} U {a} and let Jf = {Na \ a e
GF(23)}.

Lemma 2.2.3
(i) L(oo) preserves Jf as a whole and acts on Jf as it acts on GF(23)f
in particular L(oo) acts transitively on the set of unordered pairs of
subsets in Jf';
(ii) every element b 6 GF(23) is contained in exactly 12 subsets from

(iii) any two subsets from Jf have intersection of size 6.

Proof. It is straightforward that Nla = iVa+i and N^ = N2a, which imply

(i). Since L(oo) acts transitively on GF(23) and preserves Jf, the number
of subsets in Jf containing a given element b e GF(23) is independent of
the choice of b and (ii) follows. Now counting in two ways the number
of configurations (a, {A, B}) where a e GF(23), A, B e JT and a G A n B,
we obtain (iii). •
Let ^ be the code based on 0 generated by Jf and the whole set 0.

Lemma 2.2.4 ^ is stable under L.

Proof. By (2.2.3 (i)) # is stable under L(oo). Hence by (2.2.2) it

is sufficient to show that ^ is stable under T. Clearly 0>x = 0 and
N% = 0>ANo. Now one can check directly or consult [MS77], p. 492 for
a general argument that for a ^ 0 we have

Nl = N_1/aANoA0>.
 2.3 The Steiner system S(5,8,24) 53
Lemma 2.2.5
(i) <& is totally singular with respect to the parity form;
(ii) dim <€ = 12;
(iii) a non-empty subset in <$ has size at least 8;
(iv) ^ is a Golay code.
Proof. By (2.2.3 (ii), (iii)) the intersection of any two subsets from
Jf U {^} has an even number of elements and (i) follows. The code ^
is a faithful GF(2)-module for the cyclic group T of order 23 and 9
generates in ^ a 1-dimensional submodule. On the other hand 11 is the
smallest number m such that 2m — 1 is divisible by 23. Hence V/ffl is at
least 11-dimensional. By (i) # is at most 12-dimensional and we obtain
(ii).
In view of (i) in order to prove (iii) we have to show that <& does not
contain subsets of size 2, 4 and 6. Let D be a subset in # of size d.
(a) Let d = 2. Then by (2.2.2 (ii)) * contains all 2-element subsets of 9
and hence all even subsets. This is impossible since the dimension of
<€ is only 12.
(b) Let d = 4 and £ be a 3-element subset of D. Let g be an element of
order 3 in L which stabilizes £ as a whole (compare (2.2.2 (iv))). Since
g acts fixed-point freely on 9 it cannot stabilize D as a whole. Hence
\DADg\ = 2 which is impossible by (a).
(c) Let d = 6. Let E and F be any two 3-element subsets in D and g be an
element from L which maps E onto F (2.2.2 (iii)). Then DnDg contains
F and must be of even size. Hence it is of size 4 or 6 and in the latter
case Dg = D. If the intersection is of size 4 then |DADg| = 4 which
is impossible by (b). Thus D must be stable under all such elements
g. This means that the setwise stabilizer of D in L acts transitively on
the set of 3-element subsets of D. Since the number of such subsets is
divisible by 5, by (2.2.2 (ii)) this contradicts the Lagrange theorem.
Now (iv) follows from (i), (ii), (iii) and (2.1.1). •

2.3 The Steiner system 5(5,8,24)

In this section £P is an arbitrary set of size 24 and #12 is a Golay
code based on 9. Eventually we will show that #12 is unique up to
isomorphism, in particular it is isomorphic to the code constructed in the
previous section.
Let ^12 be the set of cosets of #12 in 2^, so that ^12 is a 12-dimensional
54 Mathieu groups
GF(2)-space. For 0 < i < 24 let &x be the set of /-element subsets in ^ ,
\&*i\ = (2f), and let %>n(i) be the set of cosets in ^12 having non-empty
intersection with &\. If D and E are distinct subsets from 2^ contained
in the same coset from #12 then DAE is a non-empty subset from #12
and hence its size is at least 8. This immediately gives the following.

Lemma 2.3.1 Let D and E be distinct subsets of 0* contained in the same

coset from * i 2 , such that D G &u E e 0>j with 0 < ij < 4. Then 1 = 7 = 4
and DDE = 0. D

Since there can be at most six pairwise disjoint 4-element subsets of 9

we obtain the next result.

Lemma 2.3.2 7/0 < i < j < 4 then *i 2 (i) n %n(j) = 0; 1*12(01 = (2?)
I*i2(4)| > (244) /6. n

Now one can easily verify the following remarkable equality:

i + r : ) + r: ) + r: +{>- -
Since the right hand side in the above equality is the total number of
cosets in *, and the summands in the left hand size are the lower bounds
for |*i2(0l> i = °> 1>2>3 and 4, given by (2.3.2), we have the following.

Lemma 2.3.3 * ^ is the disjoint union of the <€n(i) for i — 0,1,2,3 and 4;
I*i2(4)| = f 244 J /6, which means that for every S G ^ 4 there is a partition
of 0> into six subsets S{ = S, S2,..., S6 from 0>4 such that St U Sj € * /or
1 < 1 < j < 6. •

By the above lemma the minimal weight of a Golay code is exactly 8.

A subset of size 8 in a Golay code will be called an octad. A partition of
SP into six 4-element subsets such that the union of any two is an octad
will be called a sextet. The elements from ^ 4 will be called tetrads. In
these terms by (2.3.3) every tetrad is a member of a unique sextet.

Lemma 2.3.4 Every element F e ^ 5 is contained in a unique octad.

Proof. Let S be a tetrad contained in F and {Si = S, S2,..., S^} be the

unique sextet containing S. Let {x} = F \ S and let j , 2 < j < 6, be such
that x e Sj. Then 0 = Si U Sj is the octad containing F. If there were
 23 The Steiner system S(5,8,24) 55

another octad Or containing F then OAO' would be a non-empty subset

in #12 of size at most 6, which is impossible. •

Definition 2.3.5 Let t, k, v be integers with 1 <t <k <v. A Steiner system
of type S(t,k,v) is a pair (9C,$) where 3C is a set ofv elements and 0b is a
collection ofk-element subsets of 9C called blocks such that every t-element
subset of 9C is contained in a unique block.

Lemma 2.3.6
(i) The minimal weight of a Golay code #12 is 8 and the subsets of
size 8 (called octads) are the blocks of a Steiner system of type
5(5,8,24);
(ii) #12 is generated by its octads as a GF(2)-vector-space.

Proof. Immediately from (2.3.4) and (2.3.5) we have (i). Let @ be

the subspace in #12 generated by the octads. To prove (ii) it is sufficient
to show that \2*/2\ < 2 12 . Let E c ^ . We claim that the coset of 2
containing E contains a subset of size at most 4. In fact, suppose that
|E| > 5 and D is a 5-element subset in E. Then the symmetric difference
of E and the (unique) octad containing D is smaller than E and the
claim follows by induction. Also it is clear that the tetrads from a sextet
are equal modulo Of and the result follows from the equality given after
(2.3.2). •

By the above lemma, to prove the uniqueness of the Golay code it

is sufficient to establish the uniqueness of the Steiner system of type
5(5,8,24).

It is easy to see that the number of blocks in a Steiner system of type

S(t,k,v) is (?)/(f). The next two lemmas are standard and easy to prove.

Lemma 2.3.7 Let (£98) be a Steiner system of type S(t,k,v) with t > 2
and Y be an m-element subset of 9C where m <t. Let

<%(Y) = {B\Y \B e@,Y <= £ } .

Then {£\Y98(Y)) is a Steiner system of type S(t-m,k-m,v-m) called

the residual system of(&,&) with respect to Y. U

Lemma 2.3.8 Let (3T, 3d) be a Steiner system of type S(2, n + 1, n2 + n + 1)

and let <& be the incidence system whose points are the elements of£, whose
56 Mathieu groups
lines are the elements of & and where incidence relation is via inclusion.
Then & is a projective plane of order n. •

Thus the residue of a 3-element subset in a Steiner system of type

S(5,8,24) is a projective plane of order 4. In the next section we prove
uniqueness of this plane and study its basic properties to be prepared for
the uniqueness proof for the Steiner system of type S(5,8,24).

2.4 Linear groups

In this section we summarize some standard properties of linear and
projective linear groups. We refer the reader to [AB95], [Tay92] and
Section 9.3 in [BCN89] for proofs and further details.
Let V — Vn(q) be an n-dimensional GF(q)-vector-space where n > 2
and q = pm where p is a prime. Let G be a linear group on V and put
G° = Gn GL{V). First of all we have the following.

Lemma 2.4.1 The group SLn(q) is perfect and the group Ln(q) is non-
abelian and simple, unless (n,q) = (2,2) or (2,3); GL2(2) = Sym^ and
GL2(3) S 2l+2 : Sym3. •

Let 0* = £?(V) be the projective geometry of rank n — 1 associated

with V, i.e. the set of all proper subspaces of V with type function being
the dimension and incidence relation defined via inclusion. Let &l be the
set of i-dimensional subspaces in 9. Then \0*\ = ["] where the latter is
the <?-ary Gaussian binomial coefficient:

(qn ~
lllq

An isomorphism between two projective geometries is also called a

collineation; a correlation is a product of a collineation and a diagram
automorphism. Let W be a. hyperplane in V and w be a non-zero vector
in W. A transvection t = t(w, W) with centre w and axis W is a linear
transformation defined as follows:

v ifveW,
v+w otherwise.
 2.4 Linear groups 57
Lemma 2.4.2 Let {Vi | 1 < i < n — 1} be a maximal flag in £P, where
Vt e 0>\ Gt be the stabilizer of Vt in G and B = f|"=1 Gt be the Borel
subgroup. Then
(i) the Gi are pairwise different maximal subgroups in G,
(ii) V\ is the only proper subspace of V stabilized by Gi,
(iii) Gi induces linear groups on both Vi and V/Vu
(iv) the G-orbit containing a given pair (U, W) of subspaces in V is
uniquely determined by the dimensions of U, W and U n W,
(v) if Gt HGj <H <G then H = Gkfork = i or j ,
(vi) B = NG(S) where S is a Sylow p-subgroup in G and the index ofS
in B divides (q — l)n * m,
(vii) G induces non-equivalent doubly transitive actions on £Pl and on
0>n-\ n

The next lemma contains important information on the structure of the

parabolics G,-. Notice that since 9 possesses a diagram automorphism,
we have Gf = Gn-i.

Lemma 2.4.3 For 1 < i < (n — l)/2 let [/„_,- be an (n — i)-dimensional

subspace of V with trivial intersection with Vi and Li be the stabilizer of
Un-i in Gt. Then
(i) Gi = Qi : Lt where Qt = Op(Gi),
(ii) ifi>2 then Li contains a characteristic subgroup Kt = SL(Vi) x
SL(Un-i) and Qt is a GF(q)Ki-module isomorphic to the dual of
Vi (8) £/„_;,
(iii) Li contains a characteristic subgroup K\ = SL(Un-\) and Q\ is a
GF(q)K\-module isomorphic to the dual of Un-\,
(iv) if Ri is the kernel of the action of Gi on res^(F;) then Qi < Rif
Ri < G° and Ri/Qi is a cyclic group whose order divides q — 1. •

The next two lemmas provide further details on the structure of G\. We
follow notation introduced in (2.4.3); in addition for an element U G 2P1
incident to V\ let L(U) denote the set of elements from 0*1 other then V\
incident to U, so that L(U) is a g-element subset of 0"x.

Lemma 2.4.4 The subgroup GifiG 0 induces PGLn-i(q) on res^(Fi). IfUi,

Ui are different elements from 0*1 incident to V\ then Q\ induces a regular
action of order q on both L(U\) and L(U2) and an action of order q2 on
•
58 Mathieu groups
Lemma 2.4.5 Let N ^ 1 be a normal subgroup in G\. Then N contains Q\.
If N is not contained in R\ then either N := NR\/R\ contains Ln-i(q), or
(n, q) = (3,2) and N 1*3, or (n, q) = (3,3) and N^22. •

In order to simplify the notation put Vo to be the zero subspace in V

and Vn to be the whole space V. Let {ei9e29...9en} be a basis of V such
that Vt = (ei,e2,...,ei) for 0 < i < n. Let /\l V be the i-th exterior power
of V turned into a GF(^f)G-module, 0 < i < n. Recall that f\l V is of
dimension (") and has a basis

{ekl A ek2 A... A eki \ 1 < h < k2... < k < n).
This shows that the elements of 0>l correspond to certain 1-dimensional
subspaces of /\l V so that Vt corresponds to (e\ A ei A ... A ei). We will
use the following characterization of exterior powers which generalizes
Lemma 1 in [CPr82] and Lemma 2.10 in [IMe93].

Proposition 2.4.6 Let G be a linear group of an n-dimensional GF(q)-space

V where q — pm. Let W be a GF(q)G-module. Suppose that for some
i, 1 < i < n — 1, there is an injective mapping cp of 0>l into the set of
1-dimensional subspaces of W such that
(i) W is generated by the image of cp,
(ii) cp commutes with the action of G,
(iii) if Eu~>,Eq+i are the subspaces from £Pl contained in F*+i and con-
taining Vt-i, then (<p(Ei) | 1 < i < q+l) is a 2-dimensional subspace
in W which is the natural module for the chief factor S L(Vi+i / Vt-i)

Then W is isomorphic to /\* V.

From the above proposition it is straightforward to deduce the structure

of the permutational GF(2)-module of Ln(2) acting on the set of 1-
dimensional subspaces (on the set of non-zero vectors) of the natural
module.

Lemma 2.4.7 Let V be the natural module of G = SLn(2) = Lw(2) and

2P be the projective geometry of V. Let P be the point set of £P (the set
of 1-dimensional subspaces) and Vn = V > Vn-\ > ... > V\ be a maximal
flag in £? where V\ is identified with the set of points it is incident to. Let
W be the power set of P. Then W, as a GF(2)G-module, possesses the
decomposition
W = W1 0 We,
 2.5 The quad of order (2,2) 59

where W1 = {Q,P} and We consists of the even subsets of P. Moreover,

We is uniserial,

We = W1>W2> ... > Wn-i >Wn = {0},

where W\ is scanned by the images under G ofP\Vt and Wi/Wi+i = f\l V,

1 < i < n - 1. D

2.5 The quad of order (2,2)

In this section we present a description of the classical generalized quad-
rangle (or simply quad) ^(5p4(2)) = J*2(2) of order (2,2) and study its
basic properties.
Let Q be a set of size 6. Let 5^ = (P,L) be a point-line incidence
system whose point set P is the set of all 2-element subsets of Q and
three such points form a line if they are pairwise disjoint. Thus the line set
L is the set of partitions of Q, into three pairs. It is an easy combinatorial
exercise to check that £f is a generalized quadrangle of order (2,2).
The symmetric group H of Q isomorphic to Syme acts naturally on
Sf. Moreover the points can be identified with the transpositions in
H and the lines can be identified with the fixed-point free involutions
(which are products of three pairwise commuting transpositions). In these
terms a point and a line are incident if and only if they commute and
the action of H on Sf is by conjugation. It is well known [Tay92] that
H = Syme possesses an outer automorphism T which maps transpositions
onto fixed-point free involutions. Since T is an automorphism of H it
maps commuting pairs of involutions onto commuting ones. This means
that T preserves the incidence in Sf and hence it performs a diagram
automorphism. It is well known and straightforward to check that (if, r)
acts distance-transitively on the incidence graph F of Sf and that F has
the following distance diagram:
2
^

By the construction F is bipartite with parts P and L. Let E1 and H2 be

graphs on P and L, respectively, in which two vertices are adjacent if they
are at distance 2 in F. In other terms S 1 is the collinearity graph of Sf.
Since Sf possesses a diagram automorphism, H1 and S 2 are isomorphic.
The group H acts distance-transitively on El for i = 1 and 2 and the
distance diagram of El is the following:
60 Mathieu groups

The vertices of H1 are the 2-element subsets of Q and two vertices

are adjacent if they are disjoint. For a e f i there are five vertices in H1
which contain a and these vertices form a coclique (a maximal induced
subgraph with no edges). Such a coclique will be called standard. Define
a coclique in S 2 to be standard if it is an image of a standard coclique
in S 1 under a diagram automorphism.

Lemma 2.5.1 For i = 1 and 2 we have the following:

(i) every 5-vertex coclique in S 1 is standard;
(ii) H = Sym6 is the full automorphism group ofE1;
(iii) the stabilizer in H of a standard 5-vertex coclique is isomorphic to
Sym$ and acts transitively on the vertex set ofE?~l and on the set
of standard 5-vertex cocliques in S 3 " 1 .

Proof. Since S 1 = S 2 we can assume that i = 1. Then (i) is an

elementary combinatorial exercise. Let A be the automorphism group of
S 1 . There are six standard cocliques and by (i) they are permuted by A.
Let Z be a standard coclique and B be the stabilizer of £ in A. Then a
vertex from H1 \ Z is adjacent to exactly three vertices in Z and different
vertices are adjacent to different triples. This shows that the action of B
on £ is faithful, hence

\A\<\B\'6<\Sym5\'6 =

and (ii) follows. Finally it is an easy exercise to check (iii). •

1
Notice that if Z is a standard 5-coclique in S ' for / = 1 or 2 stabi-
lized by B = Sym5 then the subgraph in H1' induced on 3* \ E is the
Petersen graph and B acts on it as the full automorphism group. Such
subgraphs will be called standard Petersen subgraphs. The following result
is straightforward.

Lemma 2.5.2 For i = 1 and 2 every 2-path (x,y,z) in B1' with d(x,z) = 2 is
contained in exactly two non-degenerate 5-cycles and in a unique standard
Petersen subgraph. •

Let Ve(2) = 2a considered as a 6-dimensional GF(2)-space. Then it

is easy to see that Ve(2) contains exactly two proper if-submodules:
 2.5 The quad of order (2,2) 61
Fi(2) = {0,Q} and Vs(2) = {X \ X c ft, \X\ = 0 mod 2} (of dimension 1
and 5, respectively) with V\(2) < F5(2).
Let (?) be a mapping of P UL into the set of subspaces of F5(2) defined
as follows: if p 6 P (here p is a 2-element subset of ft) then cp(p) = Q \ p
and if / = {p,q,r} € L then cp(/) = (q>(p)9(p(q),(p(r)). It is easy to check
that (p(p) + <p(q) + <p(r) = 0 and hence cp is a natural representation of <9*\
Consider V4(2) = Fs(2)/Fi(2) and let x be the natural homomorphism
of Vs(2) onto F4(2). The parity form induces on V4(2) a non-singular
symplectic form *F. The composition of cp and % is a natural representation
of Sf in F4(2). Moreover the images of points are all 1-dimensional
subspaces in V4(2) and a 2-dimensional subspace in V4(2) is an image
of a line if and only if it is totally singular with respect to *¥. Hence
Sf is the classical generalized quadrangle ^(Sp4(2)) = 82(2) and H is
the set of all linear transformations of VA(2) which preserve *F, reflecting
the remarkable isomorphism Sp^(2) = S^m6. Notice that V$(2) is the
orthogonal module for Sp^(2) = O5(2) and by (1.11.2) q> is the universal
natural representation of ^(Sp^{2)). Applying (1.6.5) or analysing the
maximal subgroups in Syme one can see that Sym^ = Sp4(2) and Alte are
the only flag-transitive automorphism groups of ^(Sp4(2)). The following
lemma is easy to deduce from elementary properties of symmetric groups.

Lemma 2.5.3 Let S = S 1 be the collinearity graph of& = ^(Sp4(2)) and

G be a flag-transitive automorphism group of ^. Let x e S be a vertex,
G(x) be the stabilizer of x in G, K(x) = O2(G(x)) and Z(x) = Z(G(x)).
Then G = Alt6 or G = Sym6 = Sp4(2), and the following assertions hold:
(i) G(x) = Sym4 x 2 = 23 : Sym3 if G = Sym6 and G(x) = Sym4 =
22 :5ym 3 ifG = Alt6;
(ii) G(x) induces on res^(x) the natural action of Sym3 with kernel
K(x);
(iii) G(x) induces on S(x) the group Sym4 in its action of degree 6 with
kernel Z(x);
(iv) the pointwise stabilizer in G(x) of any two lines from res^(x) is
contained in Z(x);
(v) K(x) is elementary abelian of order 22 or 23 and K(x) n G(y) = 1
foryeZ2(x);
(vi) Z(x) = 1 if G = Alte and Z(x) is of order 2 acting fixed-point
freely on E2(x) if G = Sym^;
(vii) every subgroup of index 15 in G is the stabilizer of either a vertex
or a triangle in S. •
62 Mathieu groups
There are 16 quadratic forms / on V^(2) associated with *F in the sense
that

[Tay92]; 10 of these forms are of plus and 6 are of minus type. If / is of

minus type then the set of 1-subspaces totally singular with respect to /
is a standard coclique in S1. This reflects another important isomorphism
Sym5 ^ Oj(2).
Let B be a Sylow 2-subgroup in G = Sym6i so that B = D8 x 2.
There are exactly two proper subgroups in G (say P\ and P2) properly
containing B. Here Px ^ P2 = Sym4 x 2 and P, = NG(Ri) for i = 1,2
where R{ and #2 are two elementary abelian subgroups of order 8 in B.
Then in accordance with the standard principle we have the following.

Lemma 2.5.4 ^(G, {Pi, P2}) = 9(Sp4(2)). •

2.6 The rank 2 T-geometry

The incidence graph F of the generalized quadrangle &(Sp4(2)) possesses
a distance-transitive antipodal triple cover F. The graph F is known as
the Foster graph ([BCN89], Theorem 13.2.1) and it has the following
distance diagram:

Define ^(3 • Sp4{2)) to be a geometry whose points are the vertices in

one of the parts of F, whose lines are the vertices in another part and the
incidence relation corresponds to the adjacency in F. Then ^(3 • Sp4(2))
is the rank 2 tilde geometry (or simply T2-geometry) with the diagram

The collinearity graph S of ^(3 • Sp4(2)) is an antipodal triple cover of

the collinearity graph S of <&(Sp4(2)) and it has the following distance
diagram:

The following information comes from Theorem 13.2.1 in [BCN89].

 2.6 The rank 2 T-geometry 63
Lemma 2.6.1 Let G be the full automorphism group of&(3 • Sp4(2)). Then
(i) G acts distance-transitively on S,
(ii) 02(G) is a perfect central extension of Alte by a subgroup Y of
order 3,
(iii) every element from G\O2(G) inverts Y and G/Y s Sym6 S Sp4(2),
(iv) H < G acts flag-transitively on ^(3-Sp4(2)) if and only ifH contains
02(G),
(v) ^(3 • Sp4(2)) possesses a diagram automorphism T" and (G9T)/Y =
•
An explicit incidence matrix of T can be found in [Ito82]. We will give
constructions of 9(3 • Sp4(2)) in terms of the projective plane of order
4 (2.7.13), in terms of the Steiner system S(5,8,24) (2.10.2 (v)) and in
terms of subamalgams in a group 3 6 : Syme (6.2.2). The last construction
will also provide us with a characterization of ^(3 • Sp4(2)) and of its
automorphism group. The following lemma specifies ^(3 • 5p4(2)) as a
coset geometry.

Lemma 2.6.2 Let G be as in (2.6.1) and let B be a Sylow 2-subgroup of

G. Then B = D% x 2; there are exactly two subgroups P\ and Pi such
that B <Pt <G and PtnY = 1. Moreover, Px ^ P2 = Sym4 x 2 and
} •
It follows directly from the distance diagrams of S and S that every
non-degenerate 5-cycle is contractible with respect to the natural antipo-
dal covering cp : S —> S. Since the fundamental group of the Petersen
graph is clearly generated by its 5-cycles, we have the following.

Lemma 2.6.3 If U is a (standard) Petersen subgraph in S, then ^

is a disjoint union of three isomorphic copies of H (called the standard
Petersen subgraphs ofE). •

The next lemma is also a direct consequence of the distance diagram

ofH.

Lemma 2.6.4 Every 2-path (x9%T) in S with d(x,)t) = 2 is contained

in exactly two non-degenerate 5-cycles and in a unique standard Petersen
subgraph. •

Proposition 2.6.5 The fundamental group of S is generated by the cycles

of length 3 and by the non-degenerate cycles of length 5.
64 Mathieu groups
Proof. The distance diagram of S shows that if n is the length of a non-
degenerate cycle then n £ {3,5,6,7,8}. Hence, proceeding by induction,
it is sufficient to show that every non-degenerate n-cycle for 6 < n < 8
is decomposable into shorter cycles. Let x £ S, y £ Ss(x), S(y) = {z, |
1 < i < 6}. We assume that the {zi,z,+i} are edges for i = 1,3,5 and that
z6 £ S 4 (x), which forces z5 £ H3(x). Put {w;} = S(x) n S(z,-) for 1 < i < 4
and let Ciy denote the 6-cycle (y, zl? M,-, X, M7, Z ; , y) for 1 < i < j < 4. Since
the {z,,z,+i} are edges, it is easy to see that for i = 1 and 3 the cycle
Qj+i is degenerate. Let ©be the unique (standard) Petersen subgraph
containing the 2-path (y,z\,u\). Since ©has diameter 2, valency 3 and
does not contain triangles, it contains z5 as well as one of z3 and z4.
Without loss of generality we assume that z3 £ 0 . In this case d(u\9Zi) = 2
for i = 3 and 5, which implies that C13 and all 7-cycles are degenerate.
The cycle C14 is the sum of the cycles C13 and C34, both degenerate.
In view of the obvious symmetry we conclude that every 6-cycle splits
into triangles and pentagons. By similar arguments it is easy to prove
decomposability of the 8-cycles and we suggest this as an exercise. •

As a direct consequence of (2.6.3), (2.6.4) and (2.6.5) we obtain the

following.

Corollary 2.6.6 The subgroup of the fundamental group ofE corresponding

to the covering q> : S —> S is generated by the cycles of length 3 and by
the non-degenerate cycles of length 5. •

2.7 The projective plane of order 4

Let II = (P,L) be a projective plane of order 4 where P is the set of
points and L is the set of lines. We follow [Beu86] to show that II is
unique up to isomorphism. Since there exists the Desarguesian plane
formed by 1- and 2-dimensional subspaces in a 3-dimensional GF(4)~
space we only have to show that up to isomorphism there exists at most
one projective plane of order 4. Recall that the automorphism group of
the Desarguesian plane of order 4 is isomorphic to PFL3(4).
First of all we have that |P| = \L\ = 4 2 + 4 + 1 = 21. As usual we
identify a line of II with the set of five points it is incident to. A subset
of points is called independent if every line intersects it in at most two
points. Dually a family of lines will be called independent if every point
is on at most two lines from the family. We are going to describe the
maximal independent sets of points.
 2.7 The projective plane of order 4 65
An easy counting argument gives the following.

Lemma 2.7.1 There are exactly (21 • 20 • 16 • 9)/4! = 2520 independent

4-sets of points in II. •
Let Q = {quq2,<l3, q*} be an independent 4-set of points. Every pair of
points in Q determine a unique line containing these points and since Q
is independent, different pairs determine different lines. Thus the set M
of lines intersecting Q in two points has size 6. Two lines from M have
a common point outside Q if and only if these lines are determined by
disjoint pairs of points in Q. Hence there are three points, say p\,P2,P3,
outside Q which are intersections of lines from M and these points
correspond to partitions of Q into disjoint pairs (see the figure below).
a b

Pi

It is easy to calculate that the union of lines in M contains exactly 19

points and hence there are exactly 2 points, say a and b, which are missed
by every line from M. The following lemma is an easy combinatorial
exercise (see also [Beu86]).

Lemma 2.7.2 The set I = {pi,p2,P3,a,b} is a line ofU. •

Lemma 2.7.3 The following assertions hold:

(i) Q = Q U {a, b} is a maximal independent set of points, called, a
hyperoval,
(ii) every independent 4-set of points in U is in a unique hyperoval and
there are exactly 168 hyperovals in II,
(iii) ifA = QU {puP2,Pi} and B = MU {/}, where I = {puP2,Pi,a,b},
then (A,B) is a Fano subplane in U with respect to the incidence
relation induced from II,
66 Mathieu groups
(iv) every independent 4-set of points in II is in a unique subplane and
there are exactly 360 Fano subplanes in II,
(v) a subset of points in U is the point set of a Fano subplane if and only
if it is the symmetric difference of a hyperoval and a line intersecting
the hyperoval in 2 points.
Proof. An independent set containing Q must not contain points
outside Q on lines from M. Thus such a set is contained in Q = Qu{a,b).
By (2.7.2) the line / containing a and b misses Q and this immediately
implies that Q is independent and (i) follows. By the construction Q
(which is an arbitrary 4-set) is in a unique hyperoval. By (2.7.1) there
are 2520 independent 4-sets in n and every hyperoval contains 15 of
them. Hence the total number of hyperovals is 2520/15 = 168 and
we obtain (ii). It is straightforward to check the axioms of the Fano
plane to see (iii). An arbitrary Fano subplane in II must contain an
independent 4-set, say Q, the lines M intersecting Q in 2 points, the
intersections of the lines from M outside Q and the line through these
intersection points. By (2.7.2) such a subplane can be uniquely constructed
starting with an independent 4-set. Since every Fano subplane contains 7
independent 4-sets, (iv) follows. Finally (v) follows directly from the above
constructions. D
We proceed with the uniqueness proof for II. Let Q be a hyperoval
and let ^ be the generalized quadrangle constructed starting with Q as
in Section 2.5. Let S 1 and H2 be the point graph and the line graph of
^, respectively, so that the vertices of H1 are the 2-element subsets of Q
and the vertices of S 2 are partitions of Q into three disjoint pairs. For
i = 1,2 let &l be the set of (standard) 5-cocliques in S1'. Then |©'| = 6
and the cliques in 0 1 are identified with the elements of Q. Notice that
every vertex of 31' is in exactly two cocliques from ©*.
Let Li be the set of lines intersecting Q in i points. For a point p e Q
each of the five points in Q \ {/?} determines a line from L2 incident to
p and these lines are pairwise different since Q is independent. Hence all
five lines containing p are in L2 and we have L = Lo U L2. Put PQ = Q
and P2 = P\Q. Then
|Pol = ILol = 6, \P2\ = \L2\ = 15.

Lemma 2.7.4 There exists a unique bijective mapping q> of P U L onto

S 1 U 0 1 U S 2 U 0 2 satisfying the following :
(i) cp restricted to PQ = Q IS the identity mapping onto &1 = Q;
 2.7 The projective plane of order 4 67

(ii) cp{L2) = S 1 , cp(P2) = S 2 , <p(L0) = 0 2 ;

(iii) two distinct elements x,y G P U L are incident in U if and only if
either cp(x) G 3* and <p(y) G S3~* are incident in &, or cp(x) G S 1 is
contained in <p(y) G 0 ' , i = 1,2.

Proof. We are going to construct cp satisfying the required conditions.

By (i) q> is defined on Po. If / G L2 we put cp(l) = / n Q G B 1 . Then
/ is incident to p G Q if and only if q>(l) is contained in the coclique
corresponding to p. Let q G P2- Since L\ is empty, every line incident
to q either is disjoint from Q or intersects Q in exactly two points and
since two lines in II intersect in a single point the latter intersections are
pairwise disjoint. Hence there are three lines, say l\, l2, h, containing q
and intersecting Q in two points. Thus {Qn/, | 1 < z < 3} is a partition of
Q and we define this partition to be the image of q under q>. In this case
/ G L2 and q G P2 are incident if and only if (p(l) and cp(q) are incident
in ^ . Let m e Lo and let {ri,...,r 5 } be the points (from P2) incident
to m. We claim that ^(r f ) and q>(rj) are non-adjacent in S 2 whenever
i ^ j . In fact if they are adjacent then there is a pair a G S 1 adjacent
to them both. In this case cp~l{(x) is a line from L2 incident to r, and
rj. This is a contradiction since m is the unique line incident to r\ and
ry. Hence {<p(ri),...,(p(r5)} is a (standard) coclique in S 2 which we define
to be the image of m under cp. Now it is easy to see that <p is bijective,
the conditions (i) - (iii) are satisfied and the uniqueness follows from the
construction. •

Lemma 2.7.5 The following assertions hold:

(i) let II and IT be projective planes of order 4, let Q and Q! be hy-
perovals in U and IT, respectively, and let xp be a bisection of Cl
onto Q!; then there is a unique isomorphism ofU onto IT whose
restriction to Q coincides with xp;
(ii) all projective planes of order 4 are isomorphic;
(iii) G = Aut II acts transitively on the set of hyperovals in U and the
stabilizer H of a hyperoval is isomorphic to Sym^;
(iv) H acting on P has two orbits with lengths 6 and 15;
(v) there is a correlation rofU such that (H,T> =

Proof, (i) follows directly from (2.7.4) while (ii) and (iii) are straight-
forward from (i). It is easy to see that in the terms introduced before
(2.7.4) the set L o is a hyperoval in the dual of II. Let T be a correlation
of II which sends PQ onto Lo (such a correlation exists by (i)). Since H is
68 Mathieu groups
also the stabilizer of Lo in G, T normalizes H. It is easy to observe that
T induces a diagram automorphism of the generalized quadrangle ^ of
order (2,2) and (v) follows. •
Thus up to isomorphism the projective plane II of order 4 is unique
and it is Desarguesian. By (2.7.5 (iii)) and (2.7.3 (ii)) we have
| A u t n | = 168-6! = 2 7 - 3 3 - 5 - 7
which is of course the order of PTL^A).
The following result is easy to deduce from the order of Aut n , (2.7.3
(iv)) and (2.7.5 (iii)).

Lemma 2.7.6 The group G = Aut II acts transitively on the set of Fano
subplanes in II. The stabilizer F of such a subplane is isomorphic to 1^(2) x
2 and F acting on P has two orbits with lengths 1 and 14. •
Let us discuss a relationship between II and the affine plane Q> of
order 3. Recall that an affine plane of order n is the rank 2 geometry of
elements and blocks of a Steiner system of type 5(2, n, n2). An affine plane
of order q is formed by vectors and cosets of 1-dimensional subspaces in
a 2-dimensional GF(g)-space.
Let T be a triple of independent points in II. Let /i, I2 and h be the
lines intersecting T in two points and let A = (l\ U h U h) \ T. Then A is
of size 9 and every line intersecting A in at least two points intersects it
in three points. It is easy to check that the set B of lines intersecting A
in three points has size 12 and that O = (A, B) is an affine plane of order
3. This construction goes back to [Edge65] and it is called the deleting
procedure.

Lemma 2.7.7 Every affine subplane of order 3 in H can be constructed by

the deleting procedure.
Proof. Let A a P, B c L be such that O = (A,B) is an affine plane
of order 3 with respect to the incidence relation induced from II. Let
h, h, h be parallel lines in <D. If these lines are independent then 3>
can be obtained by the deleting procedure starting with the triple of
their pairwise intersection points. Suppose this is not the case and that
p e h n l2 n /3. Let {<?,} = /, \(AU {p}) and {puP2,P3} = h n A. Then for
i = 1, 2 and 3 the line containing q\ and p, must intersect I3 in #3, which
is impossible. •
As an implication of the proof of (2.7.7) we see that every affine
 2.7 The projective plane of order 4 69
subplane Q> of order 3 in II can be constructed by the deleting procedure
starting with exactly four different triples Ti,..., T4 corresponding to the
classes of parallel lines in <£. It is clear that T, U T) for i ^ j is an
independent set of size 6, that is, a hyperoval. Moreover, by (2.7.1), (2.7.3
(ii)) and (2.7.5 (iii)) T\ U T2, T\ U T3 and T\ U T4 are all the hyperovals
containing T\. So we have the following.

Lemma 2.7.8
(i) There are exactly 280 affine subplanes of order 3 in U and the
automorphism group ofU permutes them transitively;
(ii) the symmetric difference of 2 hyperovals in H intersecting in 3 points
is a hyperoval •

Let us turn to the automorphism group G = PFL 3 (4) of II. Let

G1 ^ PGL 3 (4) and G2 ^ L3(4) be normal subgroups in G. The cosets
of G2 in G1 are indexed by the non-zero elements of GF(4) and the
unique proper coset of G1 in G contains the field automorphism. Hence
G/G2 = Sym^. The preimage in G of a subgroup of order 2 in G/G2 is
PSL 3 (4).

Lemma 2.7.9 Let tf be the set of hyperovals and !F be the set of Fano
subplanes in II. Then
(i) G2 acting on J f has three orbits, each of length 56 with stabilizers
isomorphic to Alt^,
(ii) G2 acting on 3F has three orbits, each of length 120 with stabilizers
isomorphic to L3(2),
(iii) G/G2 = Sym3 permutes the orbits of G2 in M and the orbits of G2
on 3F in the natural way,
(iv) there is a unique bisection between the orbits of G2 on &? and the
orbits of G2 on &* which is stabilized by G.

Proof. By (2.7.5 (iii)) and (2.7.6) G acts transitively on ^f and on &

with stabilizers isomorphic to Syme and L3(2) x 2, respectively. By the
fundamental theorem of projective geometry G2 is transitive on triples
of independent points and we know that each such triple is in exactly
three hyperovals. Hence G2 has on ^f one or three orbits. In the former
case the stabilizer in G2 of a hyperoval is a normal subgroup of index
6 in Sym^. Since there are no such subgroups we obtain (i). By (i) and
(2.7.3 (v)) there are one or three orbits of G2 on 3F. Since there are no
70 Mathieu groups
index 6 subgroups in L^l) x 2, (ii) follows. Finally (iii) follows from the
paragraph before the lemma and implies (iv). •
Applying (2.4.1), (2.4.2), (2.4.3) to the case (n,q) = (3,4) we obtain the
following.

Lemma 2.7.10 Let I be a line of H identified with the set of points it is

incident to and let p G /. Then the following assertions hold:
(i) G2(l) is elementary abelian of order 2 4 , it acts transitively on P\l;
O2(G2(p)) has order 2 4 ; two points from P\p are in the same orbit
of O2(G2(p)) if and only if they belong to the same line passing
through p; O2(G2(p)) n G2(/) is of order 22 and its set of orbits on
P \ I coincides with that ofO2(G2(p));
(ii) G2[l] induces on I the natural action of Alts — ^ ( 4 ) and it acts
transitively by conjugation on the set of non-identity elements of

(iii) Gl{l) is an extension of G2(l) by a group of order 3 acting fixed-

point freely on G2(l);
(iv) the elements from G[l] \ Gx[l] induce odd permutations on I and
invert G\l)/G\l) S 3, so that G[l] S 24.(3 x Alts)2 and G[l]
induces on I the natural action of Syms = PFL2(4);
(v) G2 = L3(4) is non-abelian and simple and all involutions in G2 are
conjugate. •

Let us discuss the structure of the stabilizer in G of an affine subplane

Q> of order 3 in EL It follows directly from the deleting procedure that
the elementwise stabilizer of O is trivial. It is a standard fact that the
automorphism group F of O is isomorphic to 3 2 : GL2(3) = 3 2 : 2i+2 :
Sym^. In particular F possesses a unique homomorphism onto Sym^ with
kernel 3 2 : 2l + 2 . Since G2 is transitive on triples of independent points,
by (2.7.7) it is transitive on the affine subplanes of order 3 in II. Now
comparing the order of G and the number of subplanes given by (2.7.8)
we obtain the following.

Lemma 2.7.11
(i) G2 acts transitively on the set of affine subplanes of order 3 in II
with stabilizer isomorphic to 3 2 : 2l + 2 ;
(ii) the stabilizer of Q> in G induces on Q> its full automorphism group
isomorphic to 3 2 : GL2(3). •
 2.8 Uniqueness of S(5,8,24) 71
The automorphism group of O induces Sym* on the set of four parallel
classes of lines in O and hence also on the set of triples of points in II
from which Q> can be constructed by the deleting procedure.
The next lemma contains a standard result and we present a sketch of
the proof for the sake of completeness.

Lemma 2.7.12 The group SL^(4) is a non-split extension of 1^(4) by a

centre of order 3. Every element from 1X3(4) \ GL3(4) inverts the centre
ofSL3(4).

Proof. Let D = 3 2 : 2l + 2 be the stabilizer in G2 of an affine subplane

of order 3. It is easy to check that D/O?>(D) acts transitively on four
elementary abelian factor groups of 0?>(D) having order 3. This shows
that D does not have faithful GF(2)-representations of dimension less
than 8. On the other hand the GF(2)-dimension of the natural module
of SL3(4) is 6. Hence 51^(4) does not split over its centre (of order 3).
The second statement in the lemma is obvious. •
Since the full preimage H in 5L 3 (4) of the stabilizer in G2 of a hyperoval
contains a Sylow 3-subgroup of the latter, H does not split over the centre
of 5L 3 (4). This, (2.7.12), (2.7.9) and (6.2.2) give the following.

Lemma 2.7.13 The full preimage H in 1X3(4) of the stabilizer in G of a

hyperoval satisfies Hypothesis 6.2.1 so that H is the automorphism group
of the rank 2 T-geometry. •

The subgroup H = 3 • Sym^ preserves in the natural module of 1X3(4)

a code known as the hexacode. The natural module of 1X3(4) considered
as a GF(2)-module for H will be called the hexacode module. Let us
discuss the orbits of H on the vectors of the hexacode module.

Lemma 2.7.14 The subgroup H as in (2.7.13) acting on the non-zero vectors

of the natural module 0/1X3(4) has two orbits with lengths 18 and 45 and
stabilizers isomorphic to Syms and Sym* x 2, respectively. •

2.8 Uniqueness of 5(5,8,24)

Let ( ^ , ^ ) be a Steiner system of type 5(5,8,24). In this section we
follow [Lun69] to show that ( ^ , ^ ) is unique up to isomorphism. Since
we have an example formed by the octads of the Golay code constructed
in Section 2.2, all we have to show is that there is at most one possibility
72 Mathieu groups
for the isomorphism type of such a system. Although we do not assume
a priori that (^, @) comes from a Golay code, the blocks from 31 will be
called octads.
For a 3-element subset 7 in 9 put
#,- = J*,(7) = {B | £ e &, \B n 71 = i}.

Lemma 2.8.1 |£,-| = 21, 168, 360 and 210/or i = 3, 2, 1 and 0, respectively.
Proof. Since ^4 = 0, proceeding by induction for a given i we can
assume that \0t}\ is known for j > i. Let nt denote the number of triples
(B9K,L), where B e @, K <= B nY, L <^ B\Y, such that \K\ = i and
\L\ = 5 — i. Since K U L is contained in a unique octad, we have

By the inductive assumption we can calculate the number of triples with

\B n 71 > i. Since every octad from 0t\ corresponds to exactly (®~J.J
triples, it is straightforward to calculate |^,-|. •
By (2.3.7) and (2.3.8) 11(7) := (0>\Y,{B\Y \ B e ^3}) is a projective
plane of order 4.

Lemma 2.8.2 Let A c ^ \ 7. 77ien ^1 = 5 \ 7 /or some B e @2 if and

only if A is a hyperoval in 11(7).
Proof. Recall that two different octads have at most four elements in
common. Let B e @2 and B' ^ 0^^. Then \B n Bf n 7 | = 2 and hence
|B Pi 5' \ 71 < 2, which shows that £ \ 7 is an independent set of points
in 11(7). Since the size of B \ 7 is 6, it is a hyperoval by (2.7.3). Clearly,
different octads from $2 correspond to different hyperovals. On the other
hand by (2.7.3 (ii)) and (2.8.1) | ^ l is exactly the number of hyperovals
in n ( 7 ) and hence the result. •

Lemma 2.8.3 Let BUB2 e & with \B\ n B2\ = 4. Then B{AB2 e &.
Proof. Let yx e Bx \B2, y2 e B2\BU y3 € JJi nB 2 and 7 = {yuy2,ys}-
Then BUB2 e @2(Y) and in view of (2.8.2) B{\Y and B2 \ 7 are
hyperovals in 11(7) intersecting in three points. By (2.7.8) A := B\AB2\Y
is a hyperoval and by (2.8.2) there is an octad B3 e @2{Y) such that
A = £3 \ 7. For i = 1 and 2 we have \B3nBt\Y\ = 3 and hence
\B3 n £jf n 71 < 1, which is possible only if £3 n 7 = £1AB 2 n 7. Hence
•
 2.8 Uniqueness of 5(5,8,24) 73
Lemma 2.8.4 Let A c & \ y . 77ien
(i) an octad B e @\ with A = B\Y exists if and only if A is the point
set of a Fano subplane in 11(7),
(ii) A € J^o if and only if A is the symmetric difference of a pair of
lines in U(Y).

Proof. Choose B\ e @3 and B2 € ^ 2 so that B\\Y is a line intersecting

the hyperoval B2\Y in two points. Then |J?i n B2\ = 4 and by (2.8.3)
B3 := Bi AB2 is an octad. Clearly B3 e St\ and by (2.7.3 (v)) B3 \ Y is
the point set of a Fano subplane in 11(7). Also by (2.7.3 (v)) every Fano
subplane can be obtained as the symmetric difference of a hyperoval and
a line intersecting the hyperoval in two points. By (2.7.3 (iv)) and (2.8.1)
| ^ i I is exactly the number of Fano subplanes in 11(7), hence (i) follows.
Let #1,2*2 £ ^ 3 - Since any two lines in 11(7) intersect in a single point,
we have |2?i n B2\ = 4 and by (2.8.3) B3 = B{AB2 is an octad. Clearly
B3 e @o and B3 = (B\ \ Y)A(B2 \ 7 ) . It is easy to check that different
pairs of lines in 11(7) have different symmetric differences. By (2.8.1)
IJh I is exactly the number of unordered pairs of lines in 11(7) and (ii)
follows. •

Lemma 2.8.5 Let Bx and B2 be different octads. Then \B\ n £ 2 | e {0,2,4}.

Proof. Without loss of generality we can assume that 0 < |2*in2?2| < 4.
Let yi e Bi n J52, y2,y3 eBx\B2 and 7 = {yuyi,y*}- Then Bx e @3(Y),
B2e@i(Y). Hence #1 \ 7 is a line, while B2 \ 7 is a subplane in 11(7).
Now it is an easy exercise to show that a line intersects a Fano subplane
in one or three points, which implies the result. •
As a direct consequence of (2.8.5) we obtain the following.

Lemma 2.8.6 Let B e J i ( 7 ) and let Bf be an arbitrary octad. Then

BHY <= £' if and only if\BnB'\Y\ is odd. •

Now we are ready to prove the central result of the section.

Proposition 2.8.7 Let (&,&) and (&',&) be Steiner systems of type

S(5,8,24). Let 7 <^2P,Y' <^ &' with \Y\ = \Yf\ = 3. Let xp be a collin-
eation ofU(Y) onto 11(7'). Then there exists a unique isomorphism (p of
{0>,$) onto (&',&) such that the restriction of(ptoU(Y) coincides with xp.

Proof. We are going to construct the isomorphism cp satisfying the

required condition. The action of cp on 9 \ 7 is determined by that of
74 Mathieu groups
xp. By (2.8.2) and (2.8.3) xp(A) = B'\Yf for some Br e Sf if and only if
A = B \ Y for some B e i Let BUB29B3 be octads from # i ( 7 ) such
that BtnY ± Bj n 7 for i ^ 7. It is easy to see, arguing as in the proof
of (2.8.1), that such a triple exists. Let B[,B'2,Bf3 be octads from &X(Y')
such that B't\Y' = ip(Bi \ Y). If q> is an isomorphism of Steiner systems,
then (p(Bi Pi Y) = J5- Pi 7 ' for 1 < 1 < 3 and this condition specifies q>
uniquely. We claim that, defined in this way, cp is in fact an isomorphism.
Let B € @ and let B' e @f be such that Bf \ Y1 = xp(B \ Y). By (2.8.6),
for 1 < i < 3 Bi? n Y c B if and only if |B n £* \ Y\ is odd. Since t/; is a
bijection of &> onto ^ the latter condition holds if and only if \BfnBl\Yf\
is odd, in which case B[ DY' c Bf by (2.8.6). Hence q>(B) = Br and the
claim follows. •
In view of (2.3.6) the above proposition immediately implies the fol-
lowing corollary as well as substantial information on the automorphism
group of the Golay code which will be discussed in the next section.

Corollary 2.8.8 Up to isomorphism there are a unique Steiner system of

type S(5,8,24) and a unique Golay code. •
Notice that the above uniqueness proof could be slightly simplified
if we would assume that the Steiner system comes from a Golay code.
In this case (2.8.3) and (2.8.5) would be immediate. Also it is worth
mentioning that for B,B' €@ the equality BnY = B' nY holds if and
only if B \ Y and Bf \ Y are in the same orbit on {B \ Y \ B e &} of
the L3(4)-subgroup in Autll(Y) = PTLi(4). Using this observation one
could (if one wished to) obtain an explicit model of the Steiner system
of type S(5,8,24).

2.9 Large Mathieu groups

Let (^,^) be the unique Steiner system of type S(5,8,24), * 1 2 be the
unique Golay code generated by the octads from &. The automorphism
group of #12 (equivalently of the Steiner system) is known as the Mathieu
group Mat24 of degree 24.

Lemma 2.9.1 Let G = Mat24 and Y be a 3-element subset of 0*. Then

(i) G acts transitively on the set of Z-element subsets of 3? and G[Y] =

(ii) G(Y) £ L3(4) and G[Y]/G(Y) £ Sym3,

(iii) G acts 5-fold transitively on 0*,
 2.9 Large Mathieu groups 75

(iv) |G| = 2 10 - 3 3 • 5 • 7 • 11 • 23,

(v) G contains a subgroup isomorphic to L2(23).

Proof, (i) is an immediate consequence of (2.8.7). Since G[Y] =

Autll(Y) acts transitively on the set of hyperovals in 11(7), it is easy
to see that G[Y] acts transitively on 7 . On the other hand by (2.7.10) if
xp is a homomorphism of PTLs(4) onto a transitive subgroup of Sym^
then Im xp = Sym^ and ker xp = Li(4). Hence we have (ii). By (i) and (ii)
the action of G on 0> is 3-fold transitive and by (2.4.2 (vii)) G(Y) acts
doubly transitively on g? \ Y which implies (iii). By (i) we have
24
3

and (iv) follows. Finally, ^n is isomorphic to the Golay code constructed

in Section 2.2 which is invariant under L2(23). •
Let 0 = 7 0 ci 7i a Y2 <= 7 3 = 7 <= 0> where 7, is of size i and let
-i denote the elementwise stabilizer of 7, in Mat24.

Lemma 2.9.2
(i) Mat24-i acts (5 - i)-fold transitively on 9 \ Yt.
(ii) Mat2i = L 3 (4); \Mat22\ = 2 7 -3 2 -5-7-ll; \Mat23\ = 2 7 -3 2 -5-7-ll-23.
(iii) Mat24-t is non-abelian and simple for 0 < i < 3.

Proof. By (2.9.1 (i), (iii), (iv)) we obtain (i) and (ii). To prove (iii)
we proceed by induction. For i = 3 the result follows from (2.7.10).
Suppose that 0 < i < 2, that Mat24-t-\ is simple and N is a proper
normal subgroup in Mat24-i. Since the action of Mat24-t on & \ Yt is
doubly transitive and hence primitive, the action of N on this set is
transitive. If N Pi M a ^ - i - i ^ 1, then by the simplicity of Mat24-i-\ we
have N > Ma£24-i-i which implies N = Mat24-i. This shows that the
action of N on 0> \ Yt is regular, in particular \N\ = 24 — i. It is well
known that if L is a regular normal subgroup in a doubly transitive
group H then L is elementary abelian of order pa, say, and CH(L) = L.
In particular \H\ < pa • \GLa(p)\. In the situation considered 24 — i is a
prime power only if i = 1 but in this case we also reach a contradiction
since \Mat2i\ > 23 • 22. •

The groups Mat24, Mat2?> and Mat22 are sporadic simple groups called
the Mathieu groups. Two more Mathieu groups will appear later as
subgroups of Mat24.
76 Mathieu groups

Lemma 2.9.3 The setwise stabilizer G[Yj\ of Y2 in G = Mat24 contains

Mat22 as an index 2 subgroup and elements from G[Y2\ \ Mat22 induce
outer automorphisms of Mat22-

Proof. By (2.9.1 (iii)) G[Y2] contains Mat22 properly. Since Matu acts
transitively on 0> \ Y2 there exists g e G[Y2] \ Mat22 which stabilizes Y3
as a whole. Then g induces on L3(4) = Mat2\ < Mat22 a conjugate of
the field automorphism which is outer. •
In fact GIY2] is the full automorphism group Kx\\Mat22 of Mat22 while
Mat23 and Mat24 are perfect.
By (2.3.7) the residual system of (&,$) with respect to Yt is a Steiner
system of type 5(5 —1,8 — i, 24—i). Since the residue of Y3 is the projective
plane of order 4, it is easy to see that Mat23 and Aut Mat22 are the full
automorphism groups of the residual systems of Y\ and Y2, respectively.
By arguments similar to those in Section 2.8 one can prove the following
[Lun69].

Lemma 2.9.4 For i = 0, 1, 2 and 3 a Steiner system of type 5(5 — i, 8 —

i, 24 — i) is isomorphic to the residue of Yt in (0>, $). •

2.10 Some further subgroups of Mat24

We continue to use notation and terminology introduced in Sections 2.8
and 2.9. Till the end of this chapter and throughout the next chapter G
denotes the Mathieu group Mat24.

Lemma 2.10.1 Let B e@,Gh = G[B] and Qb = G(B). Then

(i)there are 759 octads and G permutes them transitively,
(ii)Qb is elementary abelian of order 2 4 and it acts regularly on &*\B,
(iii)Gb induces on B the alternating group Alt(B) = Alts,
(iv) Gb/Qb induces the full automorphism group L42) of Qb and Alt% =
1-4(2).
(v) G,

Proof. Since every 5-element subset of 9 is in a unique octad, (i) is

implied by (2.9.1 (iii)). We assume that Y a B. Then B \ Y is a line in
U(Y) and (ii) follows from (2.7.10) and (2.9.1 (ii)). Since two octads share
at most four elements, Gb induces a 5-fold transitive action on B and by
(2.7.10) the elementwise stabilizer of Y in this action is isomorphic to
 2,10 Some further subgroups of Mat24 77
Alts. This gives (iii). Gb induces a non-trivial action on Qb since so does
its intersection with Mat2\ by (2.7.10). Since Alts is simple the action
is faithful, and we observe (iv) comparing the orders of Alt% and L^{2).
Finally, since Qb acts regularly on 9 \ B, for p e 0* \ B the subgroup
Gb n G(p) is a complement to Qb in Gb and (v) follows. •
By (2.3.3) and the uniqueness of the Steiner system of type 5(5,8,24),
every 4-element subset S of & is contained in a unique sextet which is a
partition of 9 into six 4-element subsets Si = S, S2,..., S$ called tetrads,
such that Si U S, is an octad for 1 < i < j < 6. Notice that S U S, for
2 < i < 6 are all the octads containing S.

Lemma 2.10.2 Let I = {Si,S2,...,S6} be a sextet, Gs be the stabilizer of I,

in G, Ks be the kernel of the action of Gs on the set of tetrads in £ and
Qs = O2(GS). Then
(i) there are 1771 = ( 244 j /6 sextets and G permutes them transitively,
(ii) Gs induces the natural action of Sym§ on the tetrads in X,
(iii) Ks induces the natural action of Alt* on the elements in each S,-,
1 < i < 6,
(iv) Qs is elementary abelian of order 2 6 and Ks is an extension of Qs
by a group Xs of order 3 which acts on Qs fixed-point freely,
(v) Gs/Qs is isomorphic to the automorphism group of the rank 2 tilde
geometry <&(3 • Sym^),
(vi) Gs is isomorphic to the full preimage in ATL^(4) of the stabilizer
in PFL3(4) of a hyperoval in the corresponding protective plane of
order 4, so that Qs is the hexacode module for Gs/Qs,
(vii) ifl<i<j<k<6 then Qs acts faithfully on St U Sj U Sk.

Proof. The 5-fold transitivity of the action of G on 9 implies (i) and

also the transitivity of Gs on the tetrads in E. We assume that Y a S\
and let {p} = S\ \ Y. Let H be the setwise stabilizer of Si in G and let F
be the intersection of H with the setwise stabilizer of 7 , isomorphic to
PTL 3 (4). Clearly H is contained in Gs and induces on the points of Si the
symmetric group Sym^. On the other hand, F is the stabilizer in PFL3(4)
of the point p in 11(7), so that F = P2L 3 (4). Since S, U {p}, 2 < i < 6,
are all the lines in H(Y) passing through p, we see from (2.7.10 (iv)) that
F induces Syms on these lines and hence (ii) follows. The subgroup Ks is
contained in H and its intersection with F is the elementwise stabilizer
in PFL3(4) of the lines passing through p. By (2.7.10) this intersection
is an extension of an elementary abelian group R of order 2 4 by an
78 Mathieu groups
order 3 group Xs acting on R fixed-point freely. By (2.9.1 (ii)) Xs acts
transitively on Y. Since p is an arbitrary element from S\ we obtain (iii).
Now Qs induces on each S,- an elementary abelian group of order 2 2 and
the kernel is of order 2 4 . Hence Qs is elementary abelian of order 2 6 . The
subgroup Xs acts fixed-point freely both on R and on Qs/R, hence (iv)
follows. Clearly the image of Xs in Gs/Qs is normal and the action of Xs
on Qs induces on the latter the structure of a 3-dimensional GF(4)-space.
Let Iii denote the projective plane of order 4 associated with the dual
of this space. For 1 < i < 6 let Rt be the elementwise stabilizer of St
in Qs. Then 0 = {Ri9R2,...,Rt} is the set of points of 111 and by (ii)
the setwise stabilizer of ©in the automorphism group of rii induces on
0 the action of Syme which shows that 0 is independent and hence it
is a hyperoval in Iii. Moreover by (2.7.5) Gs/Qs is the full preimage in
FL3(4) of the stabilizer of this hyperoval in Autlli. Now (v) follows from
(2.7.13). Since Xs acts fixed-point freely on Qs, by the Frattini argument
NGS(XS) is a complement to Qs in Gs. This observation and (v) imply (vi).
If 1 < i < j < k < 5 then {£,-,£/, *k} is a GF(4)-basis of Iii and (vii)
follows. •

Since & is the set of all 8-element subsets in the Golay code
whenever B± and B2 are disjoint octads, the complement B3 of their
union is an octad (disjoint from B\ and B2). A triple of pairwise disjoint
octads is called a trio.

Lemma 2.10.3 Let T = {BUB2,B^} be a trio. Let Gt be the stabilizer of

T in G, Kt be the kernel of Gt acting on the set of octads in T and let
Qt = 02{Gt). We adopt for B = B\ the notation introduced in (2.10.1) and
let 3~(B) be the set of trios containing B.

(i) \^~(B)\ = 15 and the action of Gb on ZT(B) is doubly transitive with

kernel Qb,
(ii) there are 3795 trios and G permutes them transitively,
(iii) for i = 2 and 3 the subgroup Kt acts on Bi as an elementary abelian
group of order 2 3 extended by 1.3(2),
(iv) Gt induces Sym^ on the octads in T,
(v) Qt is elementary abelian of order 2 6 and Qb f) Qt is a hyperplane in
Qb,
(vi) there is a subgroup Xt of order 3 in Gt which permutes the octads
in T transitively, such that Gt is the semidirect product of Qt and
NG,(Xt) = Sym3 x L3(2),
 2.10 Some further subgroups of MatiA 79
(vii) Gt is isomorphic to the full preimage in AYL^A) of the stabilizer in
PYLT,{A) of a Fano subplane in the corresponding projective plane
of order 4,
(viii) Qt, as a module for Gt/Qt = 1^(2) x Sym^, is isomorphic to the
tensor product of the natural module D\ for L2(2) = Sym^ and the
natural module Difor 1^(2),
(ix) there is a bijective mapping % of ^~(B) onto the set of hyperplanes
in Qb such that with T as above £ 2 and B3 are the orbits of x(T)
on 0>\B.

Proof. We assume that Y a B = Bu so that B e J*3. Then the

octads J32 and £3 are in 3t§. Moreover the lines in 11(7) of which Bt
is the symmetric difference intersect in a point on the line / := B \ Y.
On the other hand the symmetric difference of 2 lines intersecting in a
point on / misses B. An easy calculation now shows that there are 30
octads disjoint from B and hence |^%B)| = 15. In Autll(Y) the setwise
stabilizer D of / induces Syms on the points on / and for such a point
p the stabilizer of p in D induces Sym^ on the lines passing through p
other than /. This shows that G& and even its intersection with G[Y] act
transitively on the octads disjoint from B. Since 15 is an odd number,
Qb is in the kernel of the action of Gb on 3T(B) and Gb/Qb = L4(2)
acts on &~(B) as it acts on the cosets of a parabolic subgroup. By (2.4.2)
the action is doubly transitive. An element from Qb either stabilizes each
of B2 and £3 or switches them. So we have (i) and (ii). The subgroup
Qb nKt is a hyperplane in Qb. Hence the image of Kt in Gb/Qb = ^4(2)
is of index 15 and by (2.4.3) it is isomorphic to 2 3 :1,3(2). This gives (iii).
Let {Si, S2,..., S6} be a sextet. Then {Si U S2, S3 U S4, S5 U S6} is a trio and
(iv) follows from (ii) and (2.10.2 (ii)). The subgroup Kt contains Qt and
induces on B the action of 2 3 : 1^(2) with kernel Qb C\Kt of order 2 3 so
we have (v). Thus Gt = 26.L3(2).S^m3. Since L3(2) does not possess an
outer automorphism of order 3 there is a subgroup Xt in Gt whose image
in GtIQt is normal. Then Xt induces a GF(4)-vector-space structure on
the commutator [Xt,Qt] and this structure is preserved by Gt/Qt. The
L3(2)-factor of Kt acts faithfully on Qt n Qb and on Qt/(Qt n Qb). Since
L3(2) is not involved in PrL 2 (4), this implies [XuQt] = Qt\ so that the
action of Xt on Qt is fixed-point free. An element inverting Xt can be
found inside Qb, in particular commuting with the L3(2)-factor of Kt,
which implies (vi). The action of Xt on Qt defines on the latter a GF(4)-
structure so Gt is a subgroup in ^4FL3(4). Now it is easy to check that
the GF(4)-subspaces in Qt having non-trivial intersection with Qt n Qb
80 Mathieu groups
form a Fano subplane in the projective plane of order 4 associated with
Qt. In view of (v), (vi) and (2.7.6) this implies (vii) and (viii). Finally (ix)
is immediate from (iii) and (v). •

In what follows B, Z = {5i,5 2 ,...,S 6 } and T = {2*1,2*2,2*3} are typical

octad, sextet and trio and unless explicitly stated otherwise we adopt
for them and their stabilizers in G notation as in (2.10.1), (2.10.2) and
(2.10.3), although a priori we do not assume any relationship between B,
Z and T.

Lemma 2.10.4 Let B be an octad and let 0,- denote the set of octads
intersecting B in exactly i elements. Then
(i) 10OI = 30, Gb acts transitively on 0O with stabilizer 23.23.L3(2)
which is an index 2 subgroup in the normalizer in Gb of a hyperplane
in Qb,
(ii) |0 4 | = 280, Gh acts transitively on 0 4 with stabilizer 26.3.Sym3,
contained in a conjugate of Gt,
(iii) 1021 = 448, Gb acts transitively on (92 with stabilizer isomorphic to
Sym6.
Proof. We obtain (i) directly from (2.10.3 (i), (vii)). If B,Bf e & and
S := B n B' is of size 4, then we can choose notation so that B = S U S2,
Bf = S U S3 where {S = SuS2,...,S6} is the sextet containing S. Since G
acts transitively on the set of sextets and the stabilizer of a sextet induces
Syme on the set of the tetrads in the sextet, the transitivity assertion
follows. Thus the number of pairs of octads intersecting in 4 points is
the product of the number of sextets (which is 1771) and the number
of pairs of 2-element subsets of a 6-element set (of tetrads in a sextet)
intersecting in a single element (which is 60). Since G acts transitively
on the set of octads and there are 759 of them, we can calculate \G^\.
There are 759 octads altogether and 311 of them are in 0 8 U (90 U 0 4 .
By (2.8.5) the remaining 448 octads are in (92. Gb induces Alts on the
elements in B and hence it acts transitively on the set of 2-element subsets
of B. The stabilizer in Alts of such a subset is Sym6. Hence there are
16 = 448/(|) octads intersecting B in a given 2-element subset. Let us
show that Qb acts regularly on these 16 octads. Suppose to the contrary
that a non-trivial element q € Qb stabilizes an octad Br intersecting B in
2 elements. By (2.10.1 (ii)) q fixes all elements inside B and no elements
outside B. Hence q induces on Bf an odd permutation, which contradicts
(2.10.1 (iii)). •
 2.11 Little Mathieu groups 81
Lemma 2.10.5 Let £ be a sextet and B = Si U S2. Then there is a unique
2-dimensional subspace U in Qb (such that Gb n Gs = NGb(U) and Sj,
3<j<6) are the orbits ofUon0>\B.

Proof. The subgroup H := Gb n Gs is the stabilizer in Gb of the

partition {Si,S2} of B. Since [Gb : # ] = 35, if contains g& and H/Qb
is a parabolic subgroup in JU(2) = Gb/Qb- Comparing the orders, we
conclude that H = NGb(U) for a 2-dimensional subspace U in Qb. Let 7
be a 3-element subset in Si and {p} = Si \ Y. Then {p} US,-, 2 < i < 6, are
the lines in IT( Y) containing p. By (2.7.10 (i)) the elementwise stabilizer W
of {p}uS2 in G(Y) is of order 22 and Sj, 3 < j < 6, are the orbits of W on
g?\B. Comparing the orders, we obtain the equality W = Qb<~}Qs, which
shows that W is normalized by H. Since U is the unique 2-dimensional
subspace in Qb normalized by H, we have W = U and the result follows.
•
Now by (2.10.3 (iii), (ix)) and (2.10.5) we have the following.

Lemma 2.10.6 In terms of (2.10.3) let F = 23 : L3(2) be the action induced

by Kt on B. Let S be a tetrad contained in B and let Z be the sextet
containing S. Then Z refines T if and only if S is an orbit on B of a
subgroup of index 2 in O2(F). •

2.11 Little Mathieu groups

From the construction of the binary Golay code in Section 2.2 we know
that ^12 contains 12-element subsets which will be called dodecads. By
now we know that #12 contains the empty set, the set ^ , 759 octads and
759 complements of octads. Since 212 - 2 • 759 - 2 = 2576 we have the
following.

Lemma 2.11.1 #12 contains exactly 2576 subsets of size greater than 8 and
less than 16. •

We are going to show that all of the remaining subsets are dodecads.
Clearly the complement of a dodecad is a dodecad. If a dodecad should
contain an octad then their symmetric difference would be of size 4,
which is impossible in the Golay code.

Lemma 2.11.2 A dodecad never contains an octad. •

82 Mathieu groups
Clearly, the symmetric difference of two octads intersecting in two
points is a dodecad.

Lemma 2.11.3 Let D be a dodecad.

(i) If B\, B2, B3, B4 are pairwise distinct octads such that B1AB2 =
B3AB4 = D then Bir\B2iL B3C\ B4,
(ii) there are exactly 66 ways to present D as the symmetric difference
of 2 octads intersecting in 2 elements,
(iii) there are exactly 2576 dodecads in ^n and G = Mat24 permutes
them transitively.

Proof. For Bu 1 < i < 4, as in (i) suppose that B± n B2 = B3 n B4 = X.

Then X is contained in B\ n B3. Suppose that B\ n B3 = X. Then
B3 = I U ( D \Bi) = B2, which is a contradiction. Hence Bi CiB3 is of size
4 and it contains X properly. Then B5 := B\AB3 is an octad completely
contained in D, which contradicts (2.11.2) and hence (i) follows. Now we
see that the number of presentations of D as the symmetric difference of
two octads is at most the number of 2-element subsets in £? \ D, which
is 66. Thus we can produce at least (759 • 448)/(2 • 66) = 2576 different
dodecads as symmetric differences of pairs of octads. Now (2.11.1) implies
(ii). By (2.10.4 (iii)) G acts transitively on the pairs of octads intersecting
in two elements and (iii) follows. •
As a direct consequence of (2.11.1) and (2.11.3 (iii)) we have the
following.

Proposition 2.11.4 The Golay code ^12 is doubly even. •

We are going to study the setwise stabilizer in G = Mat24 of a dodecad

D. First let us define a certain structure on D. Let Q denote the set of
6-element subsets of D (blocks) which are intersections of D with octads.
If an octad B intersects D in six points then D is the symmetric difference
of B and the octad B' = DAB, also intersecting D in six points. This
and (2.11.3) imply

Lemma 2.11.5 J is of size 132 = 6 6 - 2 and it is closed under taking

complements. •

Two octads never share a 5-element subset and the same is certainly
true for the blocks from ± Since 132 = (l*} / (6\ every 5-element
subset of D is in a unique block from 2L and we have the following.
 2.11 Little Mathieu groups 83
Lemma 2.11.6 3f = (D,£) is a Steiner system of type 5(5,6,12). •

Let Main denote the stabilizer in Mat24 of a dodecad D. Let 0 =

y 0 c 7j c Y2 c 7 3 c D where Y* is of size i. Let Matn-i denote the
elementwise stabilizer of Yt in Main. Let C\ and Ci be disjoint blocks
of the Steiner system 3> of type S(5,6,12) defined on D.

Lemma 2.11.7

(i) Matn permutes transitively the blocks of 3ft

(ii) the setwise stabilizer H of C\ in Matn is isomorphic to Sym$ and
it induces on C\ and Ci two faithful inequivalent 6-fold transitive
actions, an element from Matn which maps C\ onto Ci induces an
outer automorphism of H,
(iii) Matn acts faithfully and 5-fold transitively on D,
(iv) Matn-i acts (5 — i)-fold transitively on D\ Yif
(v) |Mat 1 2 |=2 6 -3 3 -5-ll;|Matii|=2 4 -3 2 -5-ll;|Mat 1 0 |=2 4 -3 2 -5;
\Mat9\ = 2 3 - 3 2 .

Proof. By (2.10.4 (iii)) we have (i). Let X be the unique 2-element

subset of 9 \ D such that B = Ci U X and Bf = C2UX are octads. By
(2.10.4 (iii)) the subgroup of G which stabilizes each of B and Bf as a
whole is isomorphic to Sym^ and induces faithful 6-fold transitive actions
on C\ and Ci. Let 5 be a 4-element subset of C\ and let {Si = S, S2,..., S^}
be the sextet containing S. Since S is in four blocks of Sf9 it is easy to
see that for 2 < i < 6 the intersection |S,- n D\ is of size 2 or 0. This
means that a 4-element subset of C\ corresponds to a partition of C2
into three 2-element subsets and we have (ii). Every 5-element subset E
of D is contained in a unique block C of 3>. By (i) Matn is transitive
on blocks of 2 and by (ii) the stabilizer of C acts 6-fold transitively on
its points. So we have (iii) and (iv). The elementwise stabilizer of E is
contained in the stabilizer of C and obviously must be trivial. In view of
(iv) this means that \Matn-i\ = (12 — i) •... • 8 and we obtain (v) by direct
calculations. •

By (2.11.3) there is a correspondence between 2-element subsets of

9 \ D and pairs of complementary blocks of <3. A subset X corresponds
to a pair {Ci, C2} if and only if C\ UX and C2 U l are octads. Since two
distinct octads intersect in no, two or four points, it is easy to see the
following.
84 Mathieu groups
Lemma 2.11.8 Suppose that X and Z are 2-element subsets of 0* \ D
corresponding to the pairs {C\9C2} and {C^C*} of complementary blocks,
respectively. Then up to renaming the blocks the following hold:

(i) if\Xr\Z\ = l then \d n C3\ = \C2 n C3\ = 3,

(ii) if\XDZ\=O then |Ci n C 3 | = 4 and \C2 n C 3 | = 2. D

We are going to study in more detail the structure of the groups

Matn-t for 0 < i < 3.

Lemma 2.11.9
(i) Maty = 3 2 : Q% is the stabilizer in L3(4) of an affine subplane of
order 3 in U(Y),
(ii) Mat\o is an index 2 subgroup in AutSym$ = PTL2(9) in which
Maty is the normalizer of a Sylow 3-subgroup,
(iii) Mat\i is a non-abelian simple group, it acts 3-fold transitively on
£P\D with point stabilizer isomorphic to L2(ll),
(iv) Mat12 is non-abelian and simple; it contains two conjugacy classes
of Matn-subgroups permuted by an outer automorphism of Matn
realized by an element of Mat24 which maps D onto

Proof. The residual system of 3f with respect to Yt is a Steiner system

of type S(5 — i,6 — i, 12 — i) by (2.3.5). For i = 3 we obtain an affine
subplane of order 3 in the residual projective plane 11(7) of order 4.
By (2.11.7 (v)) and (2.7.11) Mat9 is the full stabilizer of this subplane in
Mat2i = L3(4) and (i) follows. Let Df = 0> \ D and & be the Steiner
system of type 5(5,6,12) defined on Df. Let Matw.2 denote the setwise
stabilizer of Y2 in Matn which clearly contains Matw with index 2. Since
Y2 corresponds to a partition of Df into 2 blocks of Q)1 we obtain (ii)
from (2.11.7 (ii)). It is easy to see that Matn acts transitively on Df which
is of size 12. Y\ is contained in 11 2-element subsets of D. These subsets
determine a collection $ of 22 blocks of 3>' forming 11 complementary
pairs. By (2.11.8) any 2 blocks from different pairs have intersection of
size 3. For p e Df let £{p) = {B\{p} \ B e g,p G B). Then (Df \ {/?}, £{pj)
is the unique 2-(ll, 5,2)-design whose automorphism group is isomorphic
to L 2 (ll) [BJL86]. Computing the orders we see by (2.11.7 (v)) that the
latter is the stabilizer of p in Matn and (iii) follows. Since Mat\2 acts
5-fold transitively on the set D of size 12 with point stabilizer being
Matn which is simple, Matn is also simple (see the proof of (2.9.2 (iii))).
The stabilizer in Matn of any point from 9 is isomorphic to Matn but
 2.12 Fixed points of a 3-element 85
the stabilizers of p G D and p' € D' are conjugate only in the extension
of Matn by an element from MatiA which permutes D and Df. •

The groups Matn and Matn are two further sporadic simple Mathieu
groups. The group Mat\o is an index 2 subgroup in Aut5ym6 = PTLi{9\
distinct from two other such subgroups Sym^ and PGL,2(9). Let Matn-2
denote the setwise stabilizer in Mat24 of a pair of complementary do-
decads. By (2.11.9 (iv)) it contains Matn with index 2 and it induces on it
an outer automorphism. In fact Matn-2 is the full automorphism group
of Matw
We observed that the residual system of 2> with respect to Y* is a
Steiner system of type 5(5 — i, 6 — i, 12 — i). It is not difficult to show using
(2.11.9) that MatiQ.2, Matn and Matn are the full automorphism groups
of these systems for i = 2, 1 and 0, respectively. Each of these systems is
unique up to isomorphism ([Wit38], [HP85], [BJL86]).

Lemma 2.11.10 For 0 < i < 3 any Steiner system of type 5(5—i, 6—i, 12—i)
is isomorphic to the residual system of Q) with respect to Yj. •

2.12 Fixed points of a 3-element

In this section we calculate the normalizer in G of a subgroup Xs of
order 3 and analyse the octads, sextets and trios fixed by Xs.
We start by studying the orbits of G on the set of 6-element subsets
of &. Since such a subset might or might not be contained in an octad,
clearly there are at least two orbits. On the other hand G acts on 9
5-transitively and hence all 5-element subsets form a single orbit and if D
is such a subset then G[D] induces on D the full symmetric group Sym$.
There is a unique octad B which contains D and in terms of (2.10.1) we
have the following.

Lemma 2.12.1 Let D be a 5-element subset of 0* and B be the unique octad

which contains D. Then
(i) G[D]/G(D) * Sym5,
(ii) G(D) is the elementary abelian group Qb of order 2 4 extended by a
fixed-point free subgroup Xs of order 3,
(iii) G(D) acts transitively on B \ D with Qb being the kernel and on
£P\B with Xs being the stabilizer of a point. •
86 Mathieu groups
By the above lemma the G-orbit which contains a given 6-element
subset E depends only on whether or not E is contained in an octad and
the following lemma holds.

Lemma 2.12.2 IfE is a 6-element subset of &, then G[E]/G(E) = Sym6.

Moreover, ifE is contained in an octad then G(E) is elementary abelian of
order 24, otherwise G(E) is of order 3. •

If E is contained in an octad then the exact structure of G[E] follows

from (2.10.1).

Lemma 2.12.3 If E is a 6-element subset of 0* which is not contained in

an octad, then there is a sextet X = {Si,...,$6} such that G[E] = NGs(Xs)
where Gs is the stabilizer of 2 in G and Xs is a Sylow 3-subgroup in
O2AGs)- I" particular G[E] = NG(XS) = NGs(Xs) ^ 3 • Sym6 is the auto-
morphism group of the rank 2 T-geometry.

Proof. Let Z be a sextet and let Gs and Xs be defined as above. Then

by (2.10.2) for every 1 < i < 6 the subgroup Xs stabilizes S,- as a whole
and it fixes exactly one element in S,-. Thus the set E of elements in 9
fixed by Xs is of size 6. Since NGS(XS) clearly stabilizes £ as a whole, the
result follows from (2.12.2) and (2.10.2 (iv)). •

Let £ be a 6-element subset of 0> not contained in an octad and

Xs = G(E), so that Xs acts fixed-point freely on &\E. Let D be a
5-element subset in E. Then the unique octad Bf which contains D is
stabilized by Xs and hence Br = D U T where T is an orbit of Xs on
9 \ D which gives the following.

Lemma 2.12.4 There is a bisection % between the elements in E and the

orbits of Xs on 0>\E such that (E \ {p}) U x(p) is an octad for every
peE. •

Let E = {pi,...,p6} and let Ti,..., T6 be the orbits of Xs on 0> \ E

indexed so that Bt := (E \ {pt}) U Tt is an octad. For 1 < i < j < 6
the octads Bt and Bj have four elements in common and hence their
symmetric difference Btj = {puPj} U Ti U T) is also an octad.

Lemma 2.12.5 The subgroup Xs stabilizes exactly 21 = 6 + 15 octads,

namely the octads from the sets {Bt | 1 < i < 6} and {Btj \ 1 < i < j < 6}.
These sets are the orbits of NG(XS).
 2.13 Some odd order subgroups in Mat24 87
Proof. Let Bf be an octad stabilized by Xs. Then Xs fixes 5 or 2
elements in Br and in any case Xs stabilizes a 5-element subset in Bf.
But one can check that every 5-element subset of 9 stabilized by Xs is
already in one of the 21 octads counted. •
Let us turn to the trios stabilized by Xs. Since Xs is not fixed-point
free on ^ , whenever it stabilizes a trio, it stabilizes every octad in the
trio. Thus all we have to do is to decide how many trios can be formed
by the octads stabilized by Xs.

Lemma 2.12.6 There are exactly 15 trios stabilized by Xs and permuted

transitively by NG(XS). These trios are indexed by the partitions of E into
3 subsets of size 2, in particular {£12, £34, #56 } is such a trio. •

Again, since Xs is not fixed-point free on ^ , whenever Xs stabilizes

a sextet, it stabilizes at least one tetrad in the sextet. Thus in order to
describe the sextets stabilized by Xs we have to decide for each 4-element
subset stabilized by Xs which sextet it belongs to. The result is in the
following.

Lemma 2.12.7 There are exactly 16 sextets stabilized by Xs, namely the
sextet E = {{pi} U Tt | 1 < i < 6} and the sextets Z i ; containing the tetrads
E \ {puPj}, Tt U {pj}, Tj U {pt} for 1 < i < j < 6. Moreover NG(XS)
stabilizes X and permutes transitively the EI;-. •

The final result in this section can be deduced for instance by compar-
ing the centralizers of Xs in the Golay code and Todd modules dual to
each other (Section 2.15).

Lemma 2.12.8 There are exactly 20 dodecads stabilized by Xs, namely the
dodecads

for l < i < j < k < 6 . These dodecads are transitively permuted by
NG(XS). D

2.13 Some odd order subgroups in Mat24

In this section we determine the conjugacy classes of subgroups of order
3 in G and calculate the normalizer of a subgroup of order 7. First we
recall a useful general result from [Alp65].
88 Mathieu groups
Lemma 2.13.1 Let H be a finite group, F, K be subgroups of H with
K <F. Put
Jf = {h~lKh | h e H9h~lKh < F}
and let J f i , Jf^,..., «3Tm be the orbits ofF acting on Jf by conjugation. Put
A = {Fh\he H,Fhk = Fh for all keK}
and let Ai,A2,..., A/ be the orbits of NH(K) on A. Then I = m and under a
suitable ordering
|A,| = [NH(Kt) : NF(Kt)]
for 1 < i < m, where Kt e Xx (notice that NH(Ki) ^ NH(K)). D
In the previous section we discussed a subgroup Xs of order 3 in G
such that NG(XS) = NGS(XS) — 3 • Sym^. The subgroups in G conjugate to
Xs will be called 3a-subgroups. Let Xt be as in (2.10.3). Since NGt(Xt) =
Symi x L3(2), Xt is not a 3a-subgroup. The G-conjugates of Xt will be
called 3fc-subgroups. We will show that every subgroup of order 3 in G is
either 3a or 3b and that NG(Xt) — NGt(Xt)- We start with a preliminary
lemma.

Lemma 2.13.2
(i) All subgroups of order 3 in L^(4) are conjugate,
(ii) if E is a set of size 6 and D = Sym(E) = Sym^, then D has
two classes of subgroups of order 3, say 3A and 3B, so that 3A-
subgroups are generated by 3-cycles; these two classes are fused in
AutD.
Proof. If D is the stabilizer in 1^(4) of an affine subplane in the
projective plane of order 4, then 0^{D) is a Sylow 3-subgroup in L3(4)
and (i) follows from (2.7.11 (i)). In (ii) everything except possibly the
fusion is obvious. Since the product of two non-commuting transpositions
is always a 3^4-subgroup, the classes are not stable under an outer
automorphism of D. •

Lemma 2.13.3 The following assertions hold:

(i) every subgroup of order 3 in G which fixes an element from £P is a
3a-subgroup,
(ii) all 3a-subgroups in NGS(XS) other than Xs are conjugate,
(iii) all 3b-subgroups in NGS(XS) are conjugate and every subgroup of
order 3 in G is either 3a or 3b,
(iv) NG(Xt) = NGt(Xt).
 2.13 Some odd order subgroups in Mat-n 89
Proof. Put JV = NGS(XS). Every subgroup of order 3 in G which is
not fixed-point free on & is contained in G(Y) = 1^(4) for a 3-element
subset Y of ^ . Since G is 5-fold transitive on ^ , (i) follows from (2.13.2
(i)). Since JV = Gs/Qs any two subgroups of order 3 are conjugate in
JV if and only if they are conjugate in Gs. By (2.12.7) and (2.13.1) there
are exactly two classes of 3a-subgroups in Gs and we obtain (ii). For
U = 3 A and 3B let C(U) be the set of subgroups of order 3 in JV,
which maps onto the class of (/-subgroups in N/Xs = Sym(E) where
E is the 6-element subset of & fixed by Xs. Clearly C(3A) and C(3B)
are unions of conjugacy classes of subgroups in JV. Since every subgroup
from C(3A) fixes an element from E, it must be a 3a-subgroup by (i).
Hence by (ii) every 3a-subgroup in JV is contained in Xs U C(3A), all
subgroups in C(3A) are conjugate and C(3B) consists of 3h-subgroups.
Since JV is the automorphism group of the rank 2 T-geometry, we know
from (2.6.1) that it possesses an outer automorphism T which induces an
outer automorphism of JV/XS = Sym$. By (2.13.2 (ii)) T permutes C(3A)
and C(3B). In view of (ii) this means that all subgroups in C(3B) are
conjugate in JV and (iii) follows. Thus G acts transitively by conjugation
on the set of pairs (A,B) where A and B are respectively 3a- and 3b-
subgroups and [A,B] = 1. Let us calculate the number of 3a-subgroups
commuting with a given 3fc-subgroup, say Xt. When acting on ^ , Xt has
eight orbits of length 3, in particular it does not stabilize an octad. Hence
if E is the union of any two X r orbits, then £ is a 6-element subset not
contained in an octad. Clearly Xt commutes with the 3a-subgroup which
is the elementwise stabilizer of E. This shows that Xt commutes with
= 28 3a-subgroups. Since |C(3i4)| = |C(3B)| = 60 we obtain

28 4 2 7 IA7
=2 -3 • 7 = \NGt

and (iv) follows. •

Lemma 2.13.4 Let Xt be a 3b-subgroup in G. Then

(i) Xt fixes 1 sextets transitively permuted by NQ(Xt),
(ii) there are 15 trios fixed by Xt; NG{XI) fixes one of them and per-
mutes transitively the remaining ones.
Proof. It is easy to check (for instance diagonalizing the corresponding
matrices) that all non-central subgroups of order 3 in SL3(4) are conju-
gate and the centralizer in the natural module of such a subgroup is of
order 22. By (2.10.2 (vi)) and (2.10.3 (vii)) this means that for x = t or s
90 Mathieu groups
if A is a subgroup of order 3 in Gx but not in 02,3(GX), then CQX(A) = 2 2 .
We know that NGS(XS) contains 60 3fc-subgroups forming a conjugacy
class and (i) follows from (2.13.1). It is easy to see that Gt contains 3
conjugacy classes of subgroups of order 3. By (2.12.6) one of these classes
consists of 3a-subgroups and it is easy to observe that such a subgroup
can be found in the L3(2)-factor of a complement to Qt. Now (2.13.1)
and straightforward calculations imply (ii). •

Lemma 2.13.5 Let S < Gt be a subgroup of order 7. Then NG(S) =

iVGt(5) = Frob^j x Sym^. In particular S is not fully normalized in G.
Proof. Acting on g? the subgroup S has three orbits of length 7 and
three fixed elements. Hence S fixes at most three octads and, since S < Gu
it fixes exactly three octads forming the trio stabilized by S. Hence the
result follows directly from (2.10.3). •

2.14 Involutions in Mat24

In this section we study subgroups of order 2 (or rather involutions) in
G. We determine the conjugacy classes of involutions, their centralizers
and also octads, trios and sextets fixed by a given involution. For this
purpose it is helpful to know the G-orbits on octad-sextet pairs.
Let B be an octad and S be a sextet. Let v = v(#,Z) be the multiset
consisting of \B n S,-| for 1 < i < 6. We assume that \B n St\ >\Bn Sj\ if
i <j.

Lemma 2.14.1 The G-orbit containing the pair (B, H) is uniquely determined
by the multiset v = v(B,Z) and one of the following holds:
(i) v = (4204), Gh n Gs contains both Qb and Ks and has order 2 10 • 3 2 ,
(ii) v = (3 I 5 ), GbnGs^ (Sym3 x Sym5)e,
(iii) v = (2 0 ), GbHGs~ [26].Sym3 and \Qb n Gs\ = 2.
4 2

Proof. Since the union of any two tetrads in £ is an octad and any
two octads have four, two or no elements in common, it is easy to see
that the possibilities for the multiset v are those given in the lemma, (i)
follows directly from (2.10.5). In case (ii) clearly Qb<Gbn G[BnS{\ acts
transitively on 9 \ B and the transitivity assertion follows. Furthermore,
GbC\Gs = Gbn G[B n Si] n G(p) where {p} = Si\B and we obtain
(ii). In case (iii) H := Gb n G[B n Si] contains Qb and H/Qb 3* Sym6
acts transitively by conjugation on the set of non-trivial elements of Qb.
Hence H acts doubly transitively on 9 \ B which implies the transitivity
 2.14 Involutions in Mat24 91
assertion. There is exactly one involution in Qb which stabilizes S\\B
as a whole and this involution generates Qb n Gs. Since Z is uniquely
determined by any of the tetrads Si, S2, S3 and S4 the number n of sextets
S with v(£,2;) = (2402) for a fixed £ is given by

and (iii) follows.

-G) •
By (2.10.1) all involutions from Qb are conjugate in G&. The G-
conjugates of the involutions from Qb will be called 2a-involutions.

Lemma 2.14.2 Let q be a la-involution contained in Qb and let C = Co(q).

Then
(i) C = CGb(q) = 2l+6
(ii) an involution which fixes an element from 0* is a la-involution,
(iii) q fixes 71 octads, 99 trios and 91 sextets.

Proof. Since £ is the set of elements from 0> fixed by q, Cc(q) < Gb
and (i) follows from (2.10.1). Let r be an involution which fixes an element
from 9. Then there is a 5-element subset D of 9 which r stabilizes as a
whole (and fixes an element in D). Without loss of generality we assume
that B is the unique octad containing D. Since G& induces Alts on B, r
fixes 4 or 8 elements in B and in any case r is contained in G(Y) = L3(4)
for a 3-element subset Y g f i a n d (ii) follows from (2.7.10 (v)). By (2.10.4
(iii)) Qb acts fixed-point freely on (9 2. By (2.10.3) an octad from (9§ is an
orbit on 9 \ B of a hyperplane R in Qb. Clearly such an octad is fixed
by q if and only if q e R. By (2.10.5) if B' e (94 then Bf \ B is an orbit
of a subgroup U of order 2 2 in Qb. Then Bf is fixed by g if and only if
q e U. These observations and basic properties of G& show that q fixes
1 + 2 - 7 + 8-7 = 71 octads. If T = {Bi,B 2 ,B 3 } is a trio stabilized by q,
then since q is an involution, g stabilizes at least one of the octads in T,
say B\. Furthermore, if B\ = B3, then B2 n B | = 0 and since g fixes B
elementwise, this means that B2 n £ = 0. Hence either £1 = 5 or every
octad in T is fixed by q. Now an easy calculation shows that q fixes 99
trios. Since Qb is the kernel of the action of G& on the set of sextets X
such that v(5,S) = (4 2 ,0), q fixes each of these 35 sextets. By (2.14.1)
if v(B,li) = (2402) then Z is fixed by exactly one involution from Qb.
Since there are 840 sextets with v(B,E) = (2402) (transitively permuted
by Gb) and 15 involutions in Qb, altogether we have 91 sextets fixed
by q. •
92 Mathieu groups
By (2.7.13) and (2.10.2 (vi)) Gs/Qs acting on the set of involutions in
Qs by conjugation has two orbits with lengths 18 and 45 and if t is
contained in the former of the orbits then Cofo) — 26.Syms. Hence t is
not a 2a-involution. The G-conjugates of t will be called 2b-involutions.

Lemma 2.14.3 Let t be a 2b-involution and C = CG(t). Then

(i) every involution in G is either 2a or 2b,
(ii) all 2b-involutions in Gb are conjugate and there are 630 of them,
(iii) there is a unique sextet L(t) such that t G Qs where Qs = O2(GS)
and Gs is the stabilizer of £(£) in G,
(iv) C = CGs(t) = 2 6 : Sym5 and C acts transitively on 0>,
(v) t fixes 15 octads, 75 trios and 51 sextets.

Proof. Since Gb contains a Sylow 2-subgroup of G, every involution in

G is a conjugate of an involution from Gb. If s G Gb is an involution which
is not a 2<z-involution, then by (2.14.2 (ii)) 5 acts fixed-point freely on 0*
and particularly on B. This means that the image of s in Gb/Qb = L$(2)
is one of the 105 transvections. Let p G & \ B and let H = Gbn G(p) be
the corresponding complement to Qb in Gb. Then every element r G 0>\B
is identified with the unique element q G Qb such that pq = r. In this way
p is identified with the identity element, Qb acts on 0> \ B by translation
and H acts by conjugation. Let so be a transvection in H with centre r
and axis R. Then CH(so) coincides with the stabilizer in H of the pair
(r,R) and it permutes transitively the involutions in R \ {r}. If q G Qb,
then soq is an involution if and only if so centralizes q, i.e. if q G R. The
elements so and r induce the same action o n ^ \ ( 5 U i ^ ) which means
that sr is a 2a-involution. On the other hand if u is an involution from
R\{r}, then sou acts fixed-point freely on ^ . By the transitivity of CH(so)
on the involutions in R \ {r} all these involutions are conjugate in Gb,
hence they are 2fe-involutions and we obtain (i) and (ii).
Let a be the partition of B into the orbits of s$. There is a unique orbit
of length 3 of C#(so) on the set of 2-dimensional subspaces in Qb. This
orbit consists of the subspaces containing r and contained in R. These
3 subspaces correspond to the partitions of B into 2 4-sets refined by
a. Let u be an involution from R \ r, so that t = s$u is a 2ft-involution.
Then t acts on K as M and on Qb\R as wr. Hence 1/ = (r, w) is the
unique 2-dimensional subspace in Qb whose orbits on £P \ B are unions
of the orbits of t on this set. Let L(r) = {Si, S2,..., S$} be the partition of
0* such that {Si,S2} is the partition of B which corresponds to U and Si,
3 < i < 6, are the orbits of U on 0> \ B. Then I(r) is a sextet by (2.10.5)
 2.14 Involutions in Matu 93

and by the construction t stabilizes every tetrad in £(£)• We claim that

is the unique sextet with this property. Let L' = {S[, Srl9..., Sf6} be a sextet,
let v = v (£,£') be the multiset as in (2.14.1) and suppose that every 5/ is
stabilized by t. If v = (4204) then E = £(£) by the above construction of
Y,(t). Notice that the set S/ n B, if non-empty, must be a union of subsets
from G. By this observation v ^= (3 I 5 ) and if v = (2402) then we can
assume that {S[ n £ | 1 < i < 4} = c. Let A be the set of all sextets £'
such that {5/ | 1 < i < 4} = a and let M be the set of all 2Mnvolutions
in s0Qb- Then |A| = 8, \M\ = 6 and (Qb,H) acts transitively on both A
and M. Let ©be a graph on A U M in which I ' e A is adjacent to
t e M whenever t stabilizes every tetrad in Lr. By the above mentioned
transitivity, if ©contains at least one edge then every S' e A is adjacent
to at least 3 involutions in M, which is impossible since \Qb n G's\ = 2 by
(2.14.1 (iii)) where G's is the stabilizer of V in G. Hence S' = S(t) and
(iii) follows. By (2.7.14) we obtain C = 26 : 5^m5. Let II be the projective
plane of order 4 formed by the subgroups in Qs normal in Ks. Then by
(2.10.2) and the proof of (2.7.5) Gs/Ks stabilizes a hyperoval Q in II and
a hyperoval Q* in the dual of II. The points on Q* are the kernels of Qs
on the Si for 1 < i < 6. The subgroup CKS/KS is the stabilizer in Gs/Ks
of a point on Q which is the normal closure of t in Ks. By (2.5.1 (iii)) C
permutes transitively the tetrads in Z(f). In addition C contains Qs which
acts transitively on every tetrad in Z(t), so (iv) follows. By (2.13.1), (ii)
and (iv) there are 15 octads stabilized by t and transitively permuted by
Co(t). Clearly they are exactly the octads refined by Z(t). Since t is an
involution, whenever it stabilizes a trio T, it stabilizes at least one octad
in T. If t stabilizes every octad in T, then by the above T is one of the
15 trios refined by Z(t). Consider the trios containing B and stabilized
by t. By (2.10.3 (ix)) the trios containing B are in a bijection with the
hyperplanes in Qb. Since tQb/Qb is a transvection, t fixes 7 hyperplanes
and 3 of them correspond to trios refined by E(£). Hence altogether t
stabilizes 15 + 4 1 5 = 75 trios. Since there are 15 octads refined by a
given sextet and t is an involution, whenever t stabilizes a sextet E', it
stabilizes at least one octad, refined by I/. By (2.10.5) the sextets refining
B are in a bijection with the 2-dimensional subspaces in Qb. Since tQb/Qb
is a transvection, it stabilizes 11 2-dimensional subspaces. One of these
subspaces, say U, corresponds to H(t). Let £' corresponds to one of the
remaining 10 subspaces, say to W. Since U and W are stabilized by a
transvection, (U, W) is a hyperplane which determines a trio containing
B and refined by Z(t) and S'. Hence there are 1 + (10 • 15)/3 = 51 sextets
stabilized by t. D
94 Mathieu groups
In the table below we summarize the normalizers, the numbers of
elements in ^ , octads, trios and sextets stabilized for various subgroups
of order 2 and 3 in

Class Normalizer £P Octads Trios Sextets

2a 2*_+6 : L3(2) 8 71 99 91

2b 2 6 : Sym5 0 15 75 51

3a 3-Sym6 6 21 15 16

3b Sym3 x L3(2) 0 0 15 7

Lemma 2.14.4 Let X be a subgroup of order 3 and s be an involution in

G such that [Y,s] = 1. Then one of the following holds:
(i) X is a 3a~subgroup and s is a 2a-involution,
(ii) X is a 3b-subgroup and s is a 2b-subgroup.
Proof. One can observe from the above table that whenever s is an
involution, a Sylow 3-subgroup of CQ{S) is of order 3 and hence all these
subgroups are conjugate. With X and s as above if s is a 2a-involution
then X stabilizes the octad formed by the elements of 9 fixed by s. Hence
X is a 3a-subgroup. On the other hand if X is a 3a-subgroup, then all
involutions in CQ(X) = 3 • Alte are conjugate. •
The following lemma describes the distribution of involutions inside
Qb, Qt and Qs.

Lemma 2.14.5
(i) Qb is 2a-pure,
(ii) Qt contains 21 2a-involutions and 42 2b-involutions,
(iii) Qs contains 45 2a-involutions and 18 2b-involutions.

Proof. The 2a-involutions are central while 2fc-involutions are not.

For x = b, t and s the subgroup Gx contains a Sylow 2-subgroup of G
 2.15 Golay code and Todd modules 95
and hence every involution from an odd-length orbit of Gx on Q$ must
be a 2a-involution. In addition Qs contains a 2Mnvolution by (2.14.3
(iii)) and we obtain (i) and (iii). In view of the above to prove (ii) it is
sufficient to show that Qt contains a 2b-involution. It follows from the
proof of (2.13.4 (ii)) that there is a 3fr-subgroup X in Gt not contained
in 02,3{Gt). Then CQt(X) is of order 2 2 and it is 2b-pure by (2.14.4). •

2.15 Golay code and Todd modules

As above let #12 be the unique Golay code based on a set 9 of size
24 and #12 = 1®Mn be the cocode. When considering #12 and #12
as GF(2)-modules for Mat24 and its subgroups we will call them 12-
dimensional Golay code and Todd modules. The modules #12 and %>n are
dual to each other. In fact, if V is a 1-dimensional subspace in #12 then
the orthogonal complement VL of V in 2^ with respect to the parity
form has dimension 23 and Vn < VL since #12 is totally singular. This
means that the image of VL in <£\2 is a hyperplane.
There is a 1-dimensional submodule V\ = {$,&} in #12; the quotient
^n/V\ is called the 11-dimensional Golay code module and is denoted
by #11. Dually %>\2 contains a submodule ^ n of codimension 1 known
as the 11-dimensional Todd module. Since G does not stabilize non-zero
vectors in ^12, both #12 and #12 are indecomposable.
Arguing as in (2.2.5 (ii)) one can see that if H is a subgroup of
Mat24 whose order is divisible by 23, then # n and # n are irreducible
GF(2)iJ-modules.
We can now describe the orbits of G = Mat24 on the non-zero vectors
in ^ n , « n , *i2 and $n. By (2.3.3), (2.11.3) and (2.11.9) we have the
following.

Lemma 2.15.1 The actions of G on #12 and #12 are indecomposable and
(i) G acting on ^f2 has four orbits: the octads, the complement of
octads, the dodecads and the set 3? itself
(ii) G acting on ^ has two orbits indexed by the partitions of & into
an octad and its complement and by the partitions of 0* into two
dodecads; the stabilizers are Gb = 2 4 : £4(2) and Matn-2, respec-
tively,
(iii) G acting on %f2 has four orbits: %?n(i) for i = 1,2,3 and 4 with
stabilizers Mat23, AutMat22, PrL 3 (4) = Mat2i.Symi and Gs =
2 6 : 3 • Sym$ (the vectors in %>n(4) are naturally indexed by the
sextets),
96 Mathieu groups

(iv) the submodule %>n of<£n contains the zero vector, the orbit
(the pairs from &) and the orbit %>n{$) (the sextets). •

Let Tt be a graph on the set of vectors in Vn in which two vectors are

adjacent if their sum is contained in #12(1). Then Tt is a quotient of the
24-dimensional cube, %>n • Mat24 acts distance-transitively on Tt and the
distance diagram of the graph is the following:

Let F g be the graph on the set of vectors of ^12 in which two vectors
are adjacent if their sum is an octad. The suborbit diagram corresponding
to the action of #12 : Matu on F g is the following:
280 280
759 1/^^X30 30/^"^ 1 759,
1 (759) (759) 1

Let 7i c 7 2 c 7 3 c ^ where Y, is of size i and let Mat^,

4) be the setwise stabilizers in Mat24 of Yu Y2 and 73, respectively.
We are going to specify the structure of ^ n and ^ n considered as
GF(2)-modules for these subgroups of Mat2*.
By the above remark # n and # n are irreducible under

Lemma 2.15.2

(i) Ma*23 acting on (^f1 has three orbits with lengths 253 (the octads
containing Y\), 506 (the octads not containing Y\) and 1288 (the
dodecads containing Y\) with stabilizers 2 4 : Alt-j, Alts and Matn,
(ii) Ma^3 acting on <€\ x has three orbits with lengths 23 (pairs con-
taining Y\), 253 (pairs not containing Y\) and 1771 (the sextets)
and with stabilizers Mat22, P2L 3 (4) and 2 4 : (3 x Alt5)2. •
 2.16 The quad of order (3,9) 97

As a GF(2)-module for AxxiMatii, %>n is indecomposable; it contains

a 1-dimensional submodule W\ generated by the image of Y2. The
quotient %>n/W\ is irreducible, called the 10-dimensional Todd module
and denoted by #10. Dually # n has a Aut Mat22-submodule #10 of
codimension 1 known as the 10-dimensional Golay code module.
Lemma 2.15.3
(i) AutMat22 acting on ^Jo has three orbits with lengths 77 (octads
containing Y2), 330 (octads disjoint from Y2) and 616 (dodecads
containing Y2) and with stabilizers 2 4 : Sym^, 2 4 : 2^(2) and

(ii) AutMat22 acting on %>f0 has three orbits with lengths 22, 231 and
770 with stabilizers PZL 3 (4), 2 5 : Sym5 and 24.(Sym31 Sym2). •
As a GF(2)-module for PFL3(4) # n is indecomposable; it contains a
2-dimensional submodule W2 generated by the pairs contained in Y3. The
quotient ^ n / W2 is irreducible, called the 9-dimensional Todd module and
denoted by §9. Dually ^ n has a PFL3(4)-submodule ^9 of codimension
2 known as the 9-dimensional Golay code module. #9 is isomorphic to
the module of Hermitian forms in a 3-dimensional GF(4)-space.
Lemma 2.15.4 Let II = n(7 3 ). Then
(i) PTL3(4) acting on #9 has three orbits with lengths 21 (the lines of
H), 210 (the pairs of lines) and 280 (the affine subplanes of order
3),
(ii) PFL3(4) acting on #9 has three orbits with lengths 21 (the points
of II), 210 (the pairs of points) and 280 (the affine subplanes of
order 3). •

2.16 The quad of order (3,9)

In this section we establish a relationship between the projective plane II
of order 4 and the generalized quadrangle of order (3,9) associated with
the group 1/4(3) = Q^"(3). This relationship is reflected in (1.6.5 (vi)).
We consider II as the residue H(Y) of the Steiner system ( ^ , ^ ) , where
Y = {a, b, c] is a 3-element subset of 0*. For x = a,b and c put

Then Jfa, Jfb9 Jfc are the orbits of G(Y) ^ L3(4) on the set of hyperovals
in II of length 56 each (2.7.9). If H and H' are two hyperovals then by
(2.8.5) |ff n Hr\ is 0 or 2 if if and Hf are from the same G(Y)-orbit and
it is 1 or 3 otherwise.
98 Mathieu groups
Let H e Jfx for x = a,b or c. Then by counting arguments as in the
proof of (2.8.1) one can see that 2tfx contains 45 hyperovals intersecting
H in 2 points and 10 hyperovals disjoint from H. Let Tx be the graph on
Jfx in which 2 hyperovals are adjacent if they are disjoint. This graph is
known as the Gewirtz graph.
Let Q> be an affine subplane of order 3 in II and Ti,..., T4 be the
triples of points in II from which <t> can be constructed by the deleting
procedure (the paragraph before (2.7.8)). Then the set
•**(*) = {D y :=TiUTj\l<i<j<4}

consists of hyperovals such that Dy and DM are in the same G(Y)-orbit

if and only if {i,j} n {k,l} = 0. Hence J^(Q>) contains a single edge of
Tx for x = a,b and c. By (2.7.11) this means that the action of G(Y) on
the Gewirtz graph is edge-transitive. In fact it is well known and easy to
check that the action is distance-transitive and the distance diagram is
the following:

By the basic axiom of (^,^) a 3-element subset from 0>\Y is contained

in a unique hyperoval from 3tfx for x = a,b and c. This means that for
H e Jfa the set Tb(H) of hyperovals in Tb intersecting H in three
elements has size 20 and it is in the natural bijection with the set of
3-element subsets in H. Hence two vertices in Tb(H) are adjacent in Tb
if and only if they correspond to disjoint subsets in H. Let A be the
graph on 3f?a U #?b in which two hyperovals are adjacent if they either
are disjoint or intersect in 3-elements and let F be the setwise stabilizer
of A in G[Y] = Autn, so that F = P£L3(4). Then the action of F on
A is vertex-transitive and it is easy to deduce from the above that the
suborbit diagram is the following:
 2.16 The quad of order (3,9) 99
One can see from the diagram that if K is a maximal clique in A then
\K | = 4 and there is an affine subplane <J> such that
K =tf{<t>)n *
By the remark after (2.7.11) one can see that the stabilizer of O in F
induces on K a transitive action of Dg. This and the suborbit diagram of
A show that A is the point graph of a generalized quadrangle of order
(3,9) on which F induces a flag-transitive action. It is known (5.3.2 (iii)
in [PT84]) that every such quadrangle is isomorphic to the classical one
associated with the group 1/4(3) = Q^(3) and we have the following.

Proposition 2.16.1 Let H be a projective plane of order 4 and Fo = Autll.

Let J^at #eh, #ec be the orbits of the L3(4)-subgroup in Fo on the set of
hyperovals in II, let A = 34Ta U J*fb and let F ^ PEL 3 (4) be the setwise
stabilizer of A in Fo. Let % be a rank 2 incidence system whose points
are the hyperovals in A, whose lines are the affine subplanes of order 3 in
II, a hyperoval H and a subplane O being incident if H € ^f (O). Then
tfl is a generalized quadrangle of order (3,9) on which F induces a flag-
transitive action. Furthermore, Ql is isomorphic to the classical generalized
quadrangle associated with the group 1/4(3) = Qjf (3). •
 3
Geometry of Mathieu groups

In this chapter we study geometries of the Mathieu groups. Construction

of the Mathieu groups as automorphism groups of Steiner systems which
are extensions of the projective plane of order 4 (for large Mathieu
groups) and of the affine plane of order 3 (for little Mathieu groups)
leads to geometries discussed in Section 3.1. In Section 3.2 we construct
and study the geometry J^(Mat24) whose elements are octads, trios and
sextets with the incidence relation defined via refinement. We investigate
this geometry via the octad graph Y whose vertices are the octads and
two of them are adjacent if they are disjoint. The geometry Jf (Ma^)
belongs to a locally truncated diagram and this reflects the fact that
F does not contain a complete family of geometrical subgraphs. In
Section 3.3 we follow a strategy developed in Chapter 2 to construct a
graph of smaller valency with a complete family of geometrical subgraphs
and the same abstract automorphism group MatiA- In this way we
obtain the rank 3 tilde geometry ^(Mat24). In Section 3.4 considering
the actions on the octad graph of Mat23 and Aut Matu we construct the
P-geometries ^{Matii) and ^(Matii) of rank 4 and 3, respectively. In
Section 3.5 we show that ^(Matu) possesses a triple cover ^(3 • Matii)
which is simply connected. In Section 3.6 we establish the 2-simple
connectedness of ^{Mat^i). In Section 3.7 we calculate the suborbit
diagrams corresponding to the action of Mat24 on its maximal parabolic
geometry. In Section 3.8 we analyse the structure of the Golay code as
GF(2)-module for the maximal parabolics associated with the action of
Mat24 on Jf (Mat24). In Section 3.9 we calculate the suborbit diagrams
of the maximal parabolic geometry of AutMat22- In Section 3.10 we
calculate the lengths of orbits of the parabolic subgroups of Mat24 on
the set of sextets.

100
 3.1 Extensions of planes 101
3.1 Extensions of planes
In the previous chapter the large Mathieu groups have been constructed
as automorphism groups of extensions of the projective plane of order
4 and the little Mathieu groups as automorphism groups of extensions
of the affine plane of order 3. These constructions can be interpreted in
terms of diagram geometries as follows.
Let Sf = (0>,@) be the Steiner system of type 5(5,8,24). Define
$(Mati4) to be an incidence system of rank 5 whose elements of type i
are the i-element subsets of 9 for 1 < i < 4, the elements of type 5 are
the octads from & and the incidence is by inclusion.

Lemma 3.1.1 The incidence system $ = ${Mat24) is a geometry with the

diagram

£{Mat2A) : o o o o o

The group Mat24 acts on ^ ( M a ^ ) flag-transitively with the stabilizer of

an element of type i being Main, AutMat22, P^L3(4), 2 6 : 3 : Syms and
2 4 : L4(2)for i = 1,2,3,4 and 5, respectively.

Proof. Since every 4-element subset of 9 is contained in an octad, it

is easy to see that every flag is contained in a maximal one. Since the
incidence relation is via inclusion, $ belongs to a string diagram. Let Y\ c
Y2 c Y3 c Y4 c B be a maximal flag in g where | Yt\ = i and B e ®. Then
the elements of type 4 incident to Y3 are in the natural correspondence
with the elements in & \ Y3 and the elements of type 5 incident to Y3
are the blocks from ^3(^3). Hence resj(y 3 ) = {9 \ Y3,@(Y3)) which is
the projective plane 11(73) of order 4. The remaining rank 2 residues
in S are even more obvious. Since Mat24 acts transitively on & and
the stabilizer of an octad induces Alts on the elements in the octad, the
flag-transitivity of Mat24 on $ follows. The structure of stabilizers follows
from the definition of Mat23 and A u t M a ^ , together with (2.9.1) and
(2.10.2) (notice that the stabilizer of Y4 is of index 6 in the stabilizer of
the sextet determined by Y4). •

Let Of = (D, J ) be the unique Steiner system of type <S(5,6,12). Define

^(Matn) to be an incidence system of rank 5 whose elements of type i
are the i-element subsets of D for 1 < i < 4, the elements of type 5 are
the blocks from J and the incidence relation is by inclusion. The proof
of the following statement is similar to that of (3.1.1).
102 Geometry ofMathieu groups
Lemma 3.1.2 The incidence system 3F(Mat12) is a geometry with the dia-
gram
c Aff
* *
1 1 1 2 3

The group Matn acts on !F flag-transitivity with the stabilizer of an el-

ement of type i being Matn, Mat\o.2, Matg.Sym^, Mat^.Sym^ and Sym^
for i= 1,2,3,4 and 5, respectively. •
Let 7i c 7 2 c y 3 c 7 4 c Q be a maximal flag in 3F. The uniqueness
results for 5^, 3f and their residues can be reformulated in the following
way.

Lemma 3.1.3 Each of the following geometries is characterized by its

diagram including indices: $ = $(Mat24), res^(7i), resJ(Y2), ^(Matn),
resjr(Yi), resJ-CYi). •

3.2 Maximal parabolic geometry of Mat24

In this section we study the geometry tf(Mat24) whose elements are
sextets, trios and octads of the Steiner system Sf in which two elements
are incident if one refines the other one as partitions. It is convenient to
define first a graph associated with the geometry.
Let F = T(Mat24) be the octad graph which is a graph on the set of
octads in which two octads are adjacent if they are disjoint. Then the
octads contained in a trio form a triangle while the octads refined by
a sextet induce a 15-vertex subgraph (called a quad) isomorphic to the
collinearity graph of the generalized quadrangle

Lemma 3.2.1
(i) F has diameter 3 and two of its vertices are at distance 1, 2 and
3 if and only if as octads they have intersection of size 0, 4 and 2,
respectively,
(ii) Mat24 acts distance-transitively on T and the distance diagram of
Tis
1 3

(iii) every pair of vertices at distance 2 is contained in a unique quad

and every quad is strongly geodetically closed,
 3.2 Maximal parabolic geometry of Matu 103
(iv) whenever x is a vertex and T is a triangle, there is a unique vertex
in T nearest to x.

Proof. First of all the action of G on F is vertex-transitive. Let B

be an octad. By (2.10.4) 0 8 = {B}, &0, &4 and 02 are the orbits of Gb
on the vertex set of F. By definition &o consists of the octads adjacent
to B. Let B\, #2 be a pair of octads from (9$. For i = 1 and 2 let H,-
be the hyperplane in Qb such that Bt is an orbit of Ht (compare (2.10.3
(ix)). Then Bx n B2 = 0 if Hi = H2 and |Bi n B2\ = 4 otherwise. Hence
a\ = l,&i= 28 and O4 is the set of vertices at distance 2 from B. Since
Gfr acts transitively on ©4 this implies C2 = 3. Let £3 e O4 and (7 be
the 2-dimensional subspace in Qb such that B$\B is an orbit of U and
let Bi, Hi be as above. Then either U < Hi, in which case B\ is refined
by the sextet determined by B n £3, i.e. £1 is contained in the quad
containing B and £3 or 1/ n Hi is of order 2 and £1 n £3 is an orbit of
this intersection. This implies (i) and (ii). Now (iii) and (iv) follow directly
from the distance diagram of F. •
Because of the properties of F in (3.2.1 (iii), (iv)) it is a so-called near
hexagon [ShY80].
For a vertex x e F define nx to be the geometry whose points are the
triangles containing x, whose lines and planes are 3- and 7-element subsets
of points whose setwise stabilizers in G(x) contain Sylow 2-subgroups
of the latter. Since there are 35 quads containing x and transitively
permuted by G(x), it is easy to see that 3 triangles containing x form a
line in nx if and only if they are contained in a common quad. By (2.10.3
(ix)) and (2.10.5) the trios containing B correspond to the hyperplanes
and the sextets refining B correspond to the 2-dimensional subspaces in
Qb. This shows that nx is isomorphic to the rank 3 protective geometry
associated with the dual of 02(G(x)).
Let T = {x,y,z} be a triangle in F. A quad H containing T determines
a line /X(Z) in nx and a line ly(L) in ny. The mapping \pxy defined by

for every quad L containing T is the unique mapping of the residue of

T in nx onto the residue of T in ny which commutes with the action of
G(x,y).
The local projective space structures nx enable us to introduce the
notion of geometrical subgraphs. A subgraph S in F is called geometrical
if
104 Geometry ofMathieu groups
(a) whenever S contains a pair of adjacent vertices it contains the unique
triangle containing this pair,
(b) if x e S then the set *F of triangles containing x and contained in S
is a subspace in nX9
(c) G[S] n G(x) = GPP] n G(x).

The geometrical subgraph as above will be denoted by S(x,^). Since

G(x) induces the automorphism group of nx and in view of the mapping
\pxy, arguing as in Section 9.5, one can easily show that if S(x,xF) exists
then it is unique. Moreover, a quad is a geometrical subgraph S(x,xF)
where *F is a line in nx.

Lemma 3.2.2 Let *¥ be a plane in nx. Then the geometrical subgraph

S(x, *¥) does not exist.

Proof. Suppose to the contrary that E^x,^) exists. Since G acts on

F vertex-transitively and G(x) induces the automorphism group of nx,
for every x' e T and every plane *¥' in nx> the geometrical subgraph
H(x', *F') exists. Let JK be the set of all such geometrical subgraphs. Let
E be a quad containing x, so that 2 = S(x,€> ) for a line Q> of nx. Let *F,-,
1 < i < 3, be the planes in nx containing O. Then B(x,*F,-), 1 < i < 3,
are the geometrical subgraphs from Jt containing Z. Since G(x) n
induces Sym?> on {^Fj | 1 < i < 3} we conclude that G[Z] induces
on {S(x,xFI) | 1 < i < 3}, which is impossible, since G[E] = 26 : 3 •
does not possess a homomorphism onto Sym^. •

One may notice a similarity between the local structure of T(Mat24)

and that of the dual polar graph of &(Sps(2)); the difference is that the
latter graph contains a complete family of geometrical subgraphs.

Define ^f = ^f {Mat^A) to be the rank 3 geometry whose elements of

type 1, 2 and 3 are the sextets, trios and octads, where two elements
are incident if one refines another one as partitions of £P. Equivalently
the elements of Jrif are the quads, triangles and vertices of Y and the
incidence relation is via inclusion. It is easy to deduce from the above
that #e is in fact a geometry and to prove the following.

Lemma 3.2.3 Let {xi,X2,X3} be a maximal flag in $? where Xj is of type i.

Then r e s ^ x i ) ^ &(Sp4(2)), res^(x 2 ) = K3J and res^(x 3 ) is the geometry
of 1- and 2-dimensional subspaces in a ^-dimensional GF(2)-space. •
 3.2 Maximal parabolic geometry of Mat24 105
By the above lemma 3tfp(Mat24) belongs to the following diagram:

The leftmost node is a convention to indicate that the residue of an

element of type 3 (that is of an octad) is a truncation of the rank 3
projective geometry over GF(2). The elements corresponding to this node
can be defined locally (in the residue) and by (3.2.2) it is not possible to
define them globally.

The following result was established in [Ron81a].

Proposition 3.2.4 The geometry #?(Mat24) is simply connected.

Proof.^ Let (p : 3tf -* Jf be the universal covering of Jtf =^3^(Mat24)

and let F be a graph on the set of elements of type 3 in 34?, in which
two elements are adjacent if they are incident to a common element of
type 2. Since cp is a covering, the vertices in F incident to a given element
of type 2 form a triangle and the vertices incident to a given element of
type 1 induce a quad (the collinearity graph of <&(Sp4(2))). Since a pair
of elements of type 3 in ffl are incident to at most one element of type
2, arguing as in (6.3.3), we conclude that cp induces a covering of F onto
F = T(Mat24) (denoted by the same letter cp) and in view of (3.2.1 (iii)) all
cycles of length 3, 4 and 5 in F are contractible with respect to cp. Since
the diameter of F is 3, to prove the proposition it is sufficient to show that
every non-degenerate cycle in F of length 6 or 7 is decomposable into
shorter cycles. Let x e F, y e F3(x) and {z, | 1 < i < 15} = T(y) n T2(x).
For 1 < ij < 15, i £ j , let Q, = (x9ui9zi9y9Zj9Uj9x) be a 6-cycle. It
is easy to see that the decomposability of Qj is independent of the
particular choice of ut and Uj. Moreover, whenever C{zuzj) and C(zJ9Zk)
are decomposable, so is Cik. By (2.10.4 (iii)) G(x,y) ^ Sym6 ^ Sp4(2).
This means that G(x, y) acts primitively on the point set of ny and in
view of (3.2.1 (iv)) also on T(y) n F2(x). Thus to show that all 6-cycles
are decomposable it is sufficient to show that at least one such cycle
is decomposable. Let (y,z\,ui) be a 2-path, where u\ e F(x), and £ be
the unique quad containing this path. Then there is another path, say
(y9z29ui), joining y and u\ in Z. This shows that Cn is decomposable into
shorter cycles and hence so is any other 6-cycle. Finally, by (3.2.1 (iv)),
every cycle of length 7 splits into a triangle and two (possibly degenerate)
6-cycles, so the result follows. •
106 Geometry ofMathieu groups
3.3 Minimal parabolic geometry of Mat24
By (3.2.2) the octad graph T = T(Mat24) does not contain a complete
family of geometrical subgraphs. In Section 9.5 we will learn in general
how to construct in such circumstances a graph of smaller valency with a
complete family of geometrical subgraphs with the same abstract group
of automorphisms. We start with a helpful lemma.

Lemma 3.3.1 Let Z be a sextet identified with the quad in T induced by

the octads refined by Z, x G S, O x be the line in nx such that Z = 3(x,0 x )
and let ¥ x be a plane containing O x . Then G[L] = Gs, G(L) = Ks and
G(X)nG\*¥x]=Qs.

Proof. Everything except the last equality follow from (2.10.2) and the
definition of the octad graph. The group G(x) induces L^(2) on nx and
the stabilizer in L^(2) of the line 0>x induces Sym^ x Sym^ on the set of
points and planes incident to <&x. On the other hand G(x)nG[Q>x] induces
Syni4 x 2 on S. Hence G(E) permutes transitively the three planes in nx
containing O x and the result follows. •
Let Q be a graph whose vertices are the pairs (x, *F) where x E T and
*F is a plane in nX9 two such vertices (x, *F) and (x'5 *F') are adjacent if x
and x' are adjacent in T, the triangle determined by the edge {x,x'} is
contained in both *F and *F' and !/>**'(¥) = *F'. By (3.2.2) and a statement
analogous to (9.6.4) Q is connected of valency 14 and G = MatiA acts
naturally on Q.
We use notation as in (3.3.1); in addition let T = {x,y,z} be a triangle
contained in O x . Let Q>y and *¥y be the images under \pxy of Q>x and *FX
and let 0>z and *FZ be defined similarly.
Then R = {(x,*Fx),()>,*Fj,),(z, *FZ)} is a triangle in Q and every edge of
Q is in a unique triangle. Put

H2 = G[T] n G[O X UQ>yU O 2 ] n G\x¥x U *Py U *FZ].

Lemma 3.3.2
(i) # i = Gs n G((x5^Fx)); H2 = Gsn G[R],
(ii) H\C[H2 contains a Sylow 2-subgroup ofGs, in particular it contains
Qs,
(iii) Hi/Q s is a complement to Ks/Qs in (GsnG(x))/Qs and H2/Qs is a
complement to Ks/Qs in (Gs n G[T])/QS,
 3.3 Minimal parabolic geometry of Mati4 107

(iv) (HUH2) = G8,

(v) the subgraph ©in Q induced by the images of(x,x¥x) under Gs is
a geometrical subgraph isomorphic to the collinearity graph of
9(3 • SP4(2)).

Proof. Since £ is a geometrical subgraph, we have G(x) nG[I] =

G(x) n G[Q>X] and (i) follows. Now (ii) and (iii) follow directly from
(3.3.1). Since Gs/Qs does not split over Ks/Qs, we obtain (iv). By (2.10.2
(v)) Gs/Qs is the automorphism group of ^(3 • Sp4(2)). By (i), (ii) and
(iii) {HiKs/KS9H2Ks/Ks} is the amalgam of maximal parabolics of the
action of Gs/Ks S Sp4(2) on #(Sp 4 (2)). By (2.6.2) this implies that
{GS/QS9 {Hi/Qs,H2/Qs}} is isomorphic to 9(3 • Sp4(2)) and (v) follows. •

Below we present the suborbit diagram of the action of G = Mati4 on

Q computed by D.V. Pasechnik.

Let ^ = y(Mat24) be the rank 3 incidence system whose elements of

type 1 are the geometrical subgraphs in Q as in (3.3.2 (v)), the elements
of type 2 and 3 are the triangles and vertices of Q; the incidence relation
is via inclusion. We obtain the following result directly from (3.3.2).
108 Geometry ofMathieu groups
Lemma 3.3.3 &(Mat24) is a geometry with the diagram

2 2 2

and G = Mat24 is a flag-transitive automorphism group of ^{Mat24). •

Let us analyse the amalgam of maximal parabolics corresponding to

the action of G on <$. We continue to use notation as in (3.3.1) and (3.3.2);
in addition assume that x = B, T = {Bi = B,B2,B3}, B = Si U S2. Then
Jf == {Gs, Gu Gb} is the amalgam of maximal parabolics corresponding to
the action of G on Jf(Mat24) and associated with the flag {L, T,x}. Put
Gx = G[0], G2 = G[R], G3 = G((x,*Fx)). Then j * = {GUG2,G3} is the
amalgam of maximal parabolics corresponding to the action of G on ^
and associated with the flag {©,K,(x,^x )}.

Lemma 3.3.4 The following assertions hold:

(i) G t = Gs,
(ii) GT, = CGb(t) = CG(T) where T is an involution from Qb^Qt^Qs, so
1 6
that G 3 ^ 2 * : L 3 (2) and [Gb : G 3 ] = 15,
(iii) G2 = Nat(E) where E = 2 2 fs the normal closure of % in 02^{Gt)y
so that G2 = 2 6 : (Sym3 x Sym4) and [Gt : G2] = 7.

Proof, (i) follows directly from (3.3.2 (iv)). There is a bijection cp of

the set of planes in nx onto Qf such that <p(*F) is the unique involution
in Qb which fixes elementwise every triangle in *F. By (2.14.2 (i)) this
implies that there is a mapping x from the vertex set of Q onto the set
of 2a-involutions in G such that G(v) = CG{X(V)) for v € Q. Let 1/ be the
subgroup of order 2 2 in (^ which corresponds to the partition {Si,^}
of B (2.10.5). It is easy to see that U = Qb nQs = Qb nQt C\QS and
(pi^x) e U which implies (ii). Since G2 = G[R], we have G2 = NG(E)
where E is generated by the images under x of the vertices in R. Let X
be a subgroup of order 3 in O23(Gt). Since X permutes transitively the
vertices in R, it acts transitively by conjugation on the set of generators
of E. Since % e Qt and X acts on Qt fixed-point freely, E = 2 2 and (iii)
follows. •

By the above lemma we could first define si as an abstract subamalgam

in & and then define ^ as the coset geometry ^(G, si).

There is yet another way to construct the minimal parabolic geometry

^ of Mat24 from its maximal parabolic geometry Jf. The point is that
 3.3 Minimal parabolic geometry of Mat24 109
the elements of type 1 in both tf and ^ are the sextets. Moreover, an
element of either of these two geometries can be identified with the set of
sextets it is incident to, so that the incidence relation is via inclusion. In
their turn the sextets are identified with the quads in the octad graph, so
an element of type 2 or 3 in ^ is the set of quads containing a triangle
or a vertex of F, respectively. In order to define the collections of sextets
corresponding to elements of type 2 and 3 in ^ it is convenient first to
describe the orbits of G on the pairs of sextets.
Let Z = {Si,S2,...,S6} and Z' = { S ^ , . . . ^ } be distinct sextets. For a
tetrad S let fi(S) be a non-increasing sequence consisting of the non-zero
values from the set {\S n Sj\ | 1 < j < 6} and let /* = JN(2,S') be the
lexicographically largest among the sequences jtf(S/) for 1 < i < 6. For a
sequence /4> let n(fio) be the number of sextets Z' with //(Z, Z') = /*0 for
a given sextet Z.

Lemma 3.3.5 The G-orbit containing (Z, Zr) is uniquely determined by fi =

/i(Z,Z') and one of the following holds:

(i) \i = (2,2), n(fi) = 90, there is a unique trio refined by Z and Z',
(ii) /a = (3,1), n(n) = 240, there is a unique octad refined by Z and Z',
(iii) /x = (2,1,1), n(fi) = 1440, tfeere are no octads refined by Z and Z ; .

Proof. We claim that \i — (1,1,1,1) is not possible. Suppose the

contrary. Then \i(S[) = (1,1,1,1) for every 1 < i < 6. We can assume
without loss of generality, that for 1 < j < 4 we have \S[ n Sj\ = 1 and
that |S£ n Si | = |S£ n S 5 | = 1, in which case \(S[ U S£) n (Si U S5)| = 3, a
contradiction with (2.8.5). Hence \i is one of the sequences in (i), (ii) and
(iii). We assume that \i = fi(S[) = (\S[ n Si|,..., |S{ n Sr|) where r = 2 or 3.
Let B = Si US2. Suppose first that r = 2, so we are in case (i) or (ii). Since
B is an octad containing S[, B \ S[ is another tetrad from Z' which we
assume to be S'2. It is easy to see that in this case ti(S'2) = \x. Since Gb n Gs
induces {Sym$ x Sym4)e on the set of elements in B, it is easy to see that it
acts transitively on the set of tetrads S with | S n S i | = |SnS2| = 2 and on
the set of such tetrads with \SnS\\ = 3, |SnS 2 | = 1. Hence the transitivity
assertions for (i) and (ii) follow. Let U and V be the subgroups of order
2 2 in Qb corresponding to the partitions {Si,S2} and {S{,S2}, respectively.
It follows in particular that U n U' is of order 2 in case (i) and trivial
in case (ii). Thus in case (i) if an orbit of U and an orbit of Uf have a
non-empty intersection then the intersection is of size 2; Z and Z' refine
110 Geometry of Mathieu groups
the trio containing B and corresponding to the hyperplane (U, Uf) in Qb\
H(SD = \i for 1 < i < 6 which gives n((2,2)) = 90. In case (ii) if an orbit
of U and an orbit of Uf have a non-empty intersection, the intersection
is of size 1. Hence ju(S/) = (1,1,1,1) for 3 < i < 6 and n((3,1)) = 240.
Let us turn to the case (iii). Clearly Gb n Gs acts transitively on the set of
triples Y with \Y nS\\=2 and |Y Pi S2| = 1. In addition Qb stabilizes B
elementwise and permutes transitively the elements in 0>\B. This implies
the transitivity assertion in (iii). Since the total number of sextets is 1771,
by (i) and (ii) we obtain n((2,1,1)) = 1440. •

It follows from the proof of the above lemma that Gs acts transitively
on the set of tetrads S with fi(S) = (2,2), (3,1) and (2,1,1).

Definition 3.3.6 The sextet graph is a graph on the set of sextets with two
sextets S and II being adjacent j / > ( S , S ' ) = (2,2).

Let SC be the sextet graph. Then the vertices of 3C can be considered

as quads in the octad graph and in these terms two quads are adjacent
if they intersect in a triangle. Also, in view of (2.15.1 (iii), (iv)) the vertex
set of SE can be identified with a G-orbit on the set of non-zero vectors
in # n . Let S, S' be adjacent sextets as in the proof of (3.3.5). Then
S" = S\ AS[ is of size 4 and the sextet S" determined by S" is adjacent to
both S and S'. Thus there is a binary operation (denoted by *) defined
on the pairs of adjacent vertices in 9E such that x * y is a vertex adjacent
to x and y. Here x * y is the sum of x and y9 considered as vectors in
# n . A clique (a complete subgraph) K in % will be called *-closed if
whenever K contains x and y, it contains x * y. Since a *-closed clique
is the set of non-zero vectors in a subspace in # n , it contains 2n — 1
vertices for an integer n. Clearly a *-closed clique of size 3 is of the form
{x,y,x * y}, where x and y are adjacent vertices in 9C. We are going to
classify all *-closed cliques in 3C and start with the following.

Lemma 3.3.7 For every clique in the sextet graph 3C the corresponding
quads in the octad graph contain a common vertex.

Proof. It is sufficient to show that whenever three quads pairwise have

common triangles then all three of them have a common vertex. Let Si,
S2 be quads such that Si n S2 = T where T is a triangle and let x e T.
Let S3 be a quad such that Tt = S; n S3 is a triangle for i = 1 and
2. Then by (3.2.1) Tt contains a vertex, say yu adjacent to x. Since S3
 3.3 Minimal parabolic geometry of Mat24 111

contains y\ and j2> it contains the path (yi,x,y2). Hence x e Z3 and the
result follows. •

By the above lemma in order to classify the cliques in 9C it is sufficient

to consider the quads containing a given vertex in the octad graph.
The 35 quads containing x = B can be identified with the lines in nx or,
equivalently, with the 2-dimensional subspaces in Qb. Two such subspaces
l/i and U2 are adjacent (as vertices of X) if U\ n U2 is 1-dimensional,
equivalently if (Uu U2) is a hyperplane. In this case U\ * U2 is the unique
2-subspace other than U\ and l/ 2 which contains UiDfy and is contained
in(Ul9U2).

Lemma 3.3.8 A maximal clique K in 9£ has size 7, it is *-closed and G[K]

induces on K the natural action of L$(2). The group G acting on the set of
maximal cliques in X has two orbits tfv and Xt such that \jfv\ — 3 • |Jf\|.
Moreover,

(i) if K G Jfy then there are a unique vertex x in the octad graph
and a plane *¥x in nx such that K corresponds to the lines of nx
contained in ¥*,
(ii) ifKe Jft then there is a unique triangle T in the octad graph (a
trio) such that K corresponds to the quads containing T.

Proof. It follows from the general description of the cliques in the

Grassmann graphs [BCN89] or can be checked directly, that a maximal
set of pairwise intersecting lines in nx is of size 7 and consists either of
the lines containing a given point or of the lines contained in a given
plane. In view of (3.3.7) this implies the result. •

Now directly from (3.3.2), (3.3.3) and (3.3.8) we obtain the following
combinatorial characterization

Lemma 3.3.9 Let Q) be a rank 3 incidence system whose elements of type 1

are the vertices ofSC (the sextets), the elements of type 2 are the *-closed
cliques of size 3 and the elements of type 3 are the cliques from the G-orbit
tfv, where the incidence relation is via inclusion. Then Q) is isomorphic to

On the following page we present the suborbit diagram of the sextet

graph with respect to the action of Mat24. It is straightforward to deduce
this diagram from the diagram Ds(Mat24) proved in (3.7.3).
112 Geometry of Mathieu groups

By (3.3.9) the elements of type i in (S(Mat24) can be identified with

certain f-dimensional subspaces in ^ n so that the incidence relation is
via inclusion. This defines a natural representation of ^(Mati^) in ^ n .
The following result was established in [RSm89].

Proposition 3.3.10 The respesentation of ^(Matu) in %?n is universal and

it is the only representation which is invariant under the action of Ma
In fact ^ n is the universal representation group of ^(Mat24) [IPS96].
In was shown in [Hei91] and independently in [ISh89a] using coset
enumeration on a computer that Mat24 is the universal completion of the
amalgam of maximal parabolics corresponding to its action on
By the standard principle (1.5.2) this is equivalent to the following.

Proposition 3.3.11 The geometry {3{Mat24) is simply connected. •

3.4 Petersen geometries of the Mathieu groups

In this section we construct a rank 4 P-geometry &(Mat23) possessing
the Mathieu group Mat23 as a flag-transitive automorphism group and
a rank 3 P-geometry (S(Mat22) as a residue. The latter residue possesses
Aut Mat22 as a flag-transitive automorphism group.
Let Y\ be a 1-element subset of ^ and let H = Mat23 be the stabilizer of
Y\ in G = Mat24. Since the stabilizer in G of an octad B acts transitively
both on B and on & \ B, we conclude that if, acting on the vertex set of
the octad graph F, has two orbits F° and F 1 with lengths 253 and 506,
consisting of the octads containing and not containing Y\, respectively.
It is easy to deduce from (2.10.1) that the stabilizer in H of an octad
from F° or F 1 is isomorphic to 2 3 : L3(2) or Alt% = L4(2), respectively.
 3.4 Petersen geometries of the Mathieu groups 113
Let F 1 denote also the subgraph in F induced by that orbit. We
assume that x = B is a vertex-octad contained in F 1 . Then for a trio
T = {B\ = £,£2, £3} containing B exactly one of B2 and £3 is contained
in T1. Hence the valency of F 1 is 15 and H(x) = L4(2) acts doubly
transitively on F*(x) as on the point set of nx. Let nx denote also the
projective space having F*(x) as its point set and preserved by H(x). Let
Z be a quad containing x. Then by the remark before (2.5.2) Z n F° is
a standard 5-coclique and Z n F 1 is a Petersen subgraph. It is easy to
deduce from the structure of the octad graph that EOF 1 is a geometrical
subgraph, corresponding to a line in nx. We are going to show that F 1
contains a complete family of geometrical subgraphs.
Let Y2 be a 2-element subset of 9 containing Y\ and disjoint from
B. Let F = AutMat22 and F^ = Mat22 be the setwise and elementwise
stabilizers of Y2 in G, respectively. Let F 2 be the subgraph in F induced
by the octads disjoint from Y2. One can easily deduce from (2.8.1) that
|r 2 | = 330. Since Gb acts doubly transitively on &\B9 both F and
F^ act transitively on F 2 with stabilizers isomorphic to 23 : L3(2) x 2
and 23 : 1,3(2), respectively. The direct factor of order 2 in F[B] is the
unique subgroup W of order 2 in Qt, such that Y2 is a W-orbit. Let
T = [B\ = #,£2, £3} be a trio containing B and corresponding to a
hyperplane D in Qb. Then Y2 c Bt for i = 2 or 3 if and only if W < D.
This shows that F 2 n Fx(x) is a plane in nx corresponding to W, so that
if(x)nif[F 2 nr 1 (x)]=F b (x)
and hence F 2 induces in F 1 a geometrical subgraph corresponding to a
plane in nx. Thus F 1 contains a complete family of geometrical subgraphs
and by (9.8.1) we obtain the following.

Lemma 3.4.1 Let ^ ( M a ^ ) be a rank 4 incidence system whose elements

of type 1 and 2 are the geometrical subgraphs in F 1 of valency 7 and 3,
respectively, whose elements of type 3 and 4 are the edges and vertices of
F 1 with the incidence relation via inclusion. Then ^(Mat2z) is a geometry
with the diagram
P
2 2 2 1

and Mat23 is a flag-transitive automorphism group of this geometry. •

The geometry ($(Mat2i) can be defined directly in terms of the Steiner

system 5(5,8,24): the elements of type 1 are the 1-element subsets X of
9 \ Yi, the elements of type 2 are the 3-element subsets Z of 0> \ Yi,
114 Geometry ofMathieu groups

the elements of type 3 are the pairs {2*1,2*2} of disjoint octads both
missing Y\ and the elements of type 4 are the octads £3 disjoint from Y\.
The incidence relation between these elements is given by the following
conditions where Z is the sextet containing Y\ U Z :

IcZ; Xn(B1UB2) = 0;

Z refines Bx and B2\ I refines B3; B3 e {BUB2}.

It is an easy combinatorial exercise to check that this definition of
<&(Mat23) is equivalent to the one given in (3.4.1). Notice that G[B{\ n
G[B2] n G[Yi] induces on 9 \ (B\ U 5 2 U Y{) the natural action of L3(2)
on the point set of a projective plane n of order 2. By the above and in
view of (2.10.6) an element Z of type 2 is incident to {2*1,2*2} if and only
if Z is a line of n.
The description of 0(Maf 23) in terms of the Steiner system 5(5,8,24)
enables us to define an important subgeometry. Let B be an octad
containing Y\. Define a rank 3 subgeometry $f in ^(Mat23) containing
the elements of type 1 and 2 which are subsets of B and the elements
{2*1,2*2} of type 3 such that {2*, 2*1,2*2} is a trio. The incidence relation is
induced by that in <&(Mat2i).

Lemma 3.4.2 The subgeometry Sf is isomorphic to the C^geometry

the group G[B] n G[Y]] = 2 4 : Altj acts flag-transitively on ¥ with
G(B) = 2 4 being the kernel.
Proof. A trio T containing B corresponds by (2.10.3) to a hyperplane
in Qb and T induces on B \ Y\ a structure n(T) of a projective plane
of order 2. The ^4/£7-subgroup in Gb/Qb = L^(2) acts flag-transitively
on the projective geometry associated with Qb (compare (1.6.5)). In
particular it permutes transitively the 15 trios containing B. Thus the
corresponding 15 projective plane structures n(T) form an ,4/t7-orbit.
Now the result follows directly from the definition of the ^4/t7-geometry
given in Section 1.7. •
For i = 1,2,3 and 4 let Ht be the stabilizer in H = Mat23 of an element
of type i in ^(Mat23). In view of the description of &(Mat2?) in terms of
the Steiner system we have the following:
Hi S Mat22; H2 s 2 4 : (Alt5 x 3)2; H3 s 2 4 : L 3 (2); H4 = Alt*.
Here H2 is the stabilizer in Gs of an element from 0*. The structure of
H3 deserves a further comment.
 3.4 Petersen geometries of the Mathieu groups 115
Lemma 3.4.3 if3 = Aut^4GL3(2), in particular O2(H3) is an indecompos-
able module for H3/O2(H3) S L3(2).
Proof. Let {#2, £3} be the element of type 3 stabilized by if3. Put
B = Bi = 0>\(B2UB3). For i = 2 and 3 Kt induces the group ^GL3(2) ^
23 : L3(2) on the elements in 2*,- and the kernel, which is elementary
abelian of order 23, acts regularly on B. Hence KtnH = AGL3(2) acts
faithfully on £,. Since B2 and B3 are orbits of a hyperplane D in <2&, an
element from D stabilizes both B2 and £3 while an element from Qb\D
switches them. This shows that if3 = (KtnH,Qb) and since if3 stabilizes
B, Qb < if3. On the other hand H[B] ^ 24 : Alt-, and H[B]/Qb ^ Alt-,
actsflag-transitivelyon the projective space associated with Qb. It follows
from elementary properties of Alt-j that the stabilizer of a point and the
stabilizer of a plane are non-conjugate subgroups isomorphic to L3(2)
(and conjugated in Sym-j). Since L3(2) ^ (Kt nH)Qb/Qb < H[B]/Qb =
Alti and Kt D H normalizes the hyperplane 2), it does not normalize
subgroups of order 2 and hence Qb is indecomposable under H3/Qb.
Thus an element from Qb\D permutes the classes of complements to D
in Kt n H. Since there are two classes of such complements [JP76], if3 is
the automorphism group of ^4GL3(2). •
The residue in ^(Mat23) of an element of type 1 is a rank 3 P-geometry
with the diagram
P
-o
2 2 1
denoted by &(Mat22). This geometry can also be described directly in
terms of the Steiner system S(5,8,24). Specifically, the elements of type
1 are sextets Z = {Si,52,...,S6} such that Y2 c St for some 1 < i < 6,
the elements of type 2 are the pairs {2*1,2*2} of disjoint octads, both
disjoint from Y2 and the elements of type 3 are the octads B3 disjoint
from Y2. The incidence between these elements is given by the following
conditions:
S refines £1 and B2\ E refines B3\ B3 e {BUB2}.
There is a natural bijection cp between the set of 2-element subsets Z of
9 \ Y2 and the set of elements of type 1 in ^(Mat22) where cp(Z) is the
unique sextet containing Y2 UZ. An element B3 of type 3 in <g(Mat22)
is incident to <p(Z) if and only if Y2 and Z are different orbits of a
subgroup of order 2 in G(B3) = 24.
The stabilizer in Mat23 of an element of type 1 in ^(Mat23) induces on
the group F^ = Mat22 and it is clear from the above description
116 Geometry ofMathieu groups
that ^(Mat2i) also admits Aut Matii as a flag-transitive automorphism
group.
For i = 1,2 and 3 let FJ7 and Ft be the stabilizers in F b and F of an
element of type i in ^{Matri). Then from the above we can easily deduce
the following:
F\ * 24 : 5ym5; F2b s 24 : 5ym4; F\ S 23 : L3(2);

F2 s 24 : (Sym4 x 2); F3 = 23 : L3(2) x 2.

The residue of an element of type 1 in the >4/£7-subgeometry Sf in
& (Mat23) is a subgeometry 2 in ^(Mflfe) isomorphic to the generalized
quadrangle of order (2,2). This subgeometry can be described in the
following way. Let B be an octad containing Y2. Then 2 consist of the
elements of type 1 contained in B and of the elements {1*1,2*2} of type 2
such that {2*, 2*i, 1*2} is a trio. The stabilizers g b and Q of 2 in F b and F
are the elementwise and the setwise stabilizers of Y2 in G& isomorphic to
24 : Alt6 and 24 : Sym6, respectively. Clearly O2(Q) = Qb is the kernel of
the action of Q on 2. We formulate this as follows.

Lemma 3.4.4 Every octad B containing Y2 corresponds to a rank 2 sub-

geometry 2 in (§{Mat22) isomorphic to the generalized quadrangle
9(S(2)) •
Recall that if # is a P -geometry of rank n then the derived graph A(#)
of ^ is a graph on the set of elements of type n in 9 in which two
elements are adjacent if they are incident to a common element of type
n — 1. In these terms the subgraphs in the octad graph F induced by the
orbits T1 and T2 are the derived graphs A(<g(Mat23)) and A(<g(Mat22)),
respectively. It is well known [BCN89] and can be easily deduced from
(3.2.1) that these two graphs are distance-transitive with the following
respective distance diagrams.
2 6
1
15 1 /7TM4
15 1 / 210
CM2 9

We conclude this section with the following.

Lemma 3.4.5 (§(Mat22) is a subgeometry of&(Mat2A).

 3.5 The universal cover o 111

Proof. We consider &(Mat24) as described in (3.3.9). Then the set of

elements of type 1 in ^ ( M a ^ ) is a subset of the set of elements of type
1 in ^(Mat24). Let B be an element of type 3 in (§(Mat22), i.e. an octad
disjoint from Y2. Let W be the setwise stabilizer of Y2 in Qb = G(B),
so that W is the unique subgroup of order 2 in Qb, such that Y2 is
a W-orbit. If E is a sextet which refines B then Z is determined by a
tetrad which is an orbit on 9 \ B of a subgroup U of order 22 in Qb.
Furthermore, Y2 is contained in a tetrad of £ if and only if W < U. In
terms of (3.3.8) this means that the set L of elements of type 1 incident
to B is a maximal clique in the sextet graph from the orbit Xv. Hence
L is an element of type 3 in &(Mat24). Clearly B is uniquely determined
by L. In a very similar way one can show that an element of type 2 in
&(Mat22) is uniquely determined by the set M of sextets incident to it
and that M is an element in ^ ( M a ^ ) . O
3

By (3.4.5) the sextet graph contains the collinearity graph

as a subgraph. The suborbit diagram of the latter graph as given above
is easy to deduce from the diagram Dp(Mat22) proved in (3.9.6).

3.5 The universal cover of (§(Mat22)

In this section we show that ^(Mat22) possesses a triple cover
which is simply connected.
We start by proving a result established by S.V. Shpectorov in [Sh85]
which gives an upper bound 1898 for the number of elements of type 3
in a P-geometry of rank 3.
Let ^ be a P -geometry of rank 3 and A = A(^) be the derived graph
of ^. An element y of type 2 in ^ corresponds to an edge in A whose
ends are the elements of type 3 incident to y. We claim that different
118 Geometry ofMathieu groups
elements of type 2 correspond to different edges. In fact, suppose that y\
and >>2 are different elements of type 2 incident to the same pair {xi,X2}
of elements of type 3. Since y\ and yi are lines in the projective plane
res^(xi), there is a point z in this plane (an element of type 1 in ^)
incident to both y\ and yi. Since res^(z) is the geometry of edges and
vertices of the Petersen graph, which does not contain multiple edges,
we reach a contradiction. Thus the elements of type 2 in ^ are identified
with the edges of A. Similarly an element z of type 1 corresponds to
a subgraph S (which is not necessarily an induced subgraph) formed
by the vertices and edges incident to z. Clearly S is isomorphic to the
Petersen graph. Since a point in a projective plane is uniquely determined
by the set of lines it is incident to, different elements of type 1 correspond
to different subgraphs. Throughout this section when talking about a
Petersen subgraph in A we always mean a subgraph formed by vertices
and edges incident to an element of type 1.
For x e A let nx denote the projective plane structure having A(x) as
the point set and dual to res^(x). A triple {u9v,w} of vertices from A(x)
is a line of nx if there is a Petersen subgraph S containing x, such that
{u,v,w} = S(x). If (y,x,z) is a 2-arc in A then the points y and z in nx
determine a unique line and hence there is a unique Petersen subgraph
E(y, x, z) in which (y, x, z) is a 2-arc. Dually, if S and S' are Petersen
subgraphs containing a common vertex x, then S(x) and S'(x) are lines
in nx which must have a point in common. Hence whenever two Petersen
subgraphs have a common vertex, they have a common edge incident to
this vertex.
As usual, for a vertex x of A and an integer i by A,-(x) we denote the
set of vertices at distance i from x in A and A(x) = Ai(x). In addition
put
A*(x) = {x} U Ai(x) U... U A,(x)
In this section the meaning of the parameters bt and ct slightly differs
from that in the rest of the book. To wit, we put
bt = max {|A,-+i(x) n A(y)\ \ x e A, y e A,(x)},

ct = min {|A,-_i(x) n A(y)\ | x € A, y e A,-(x)}.

The following lemma generalizes some well-known properties of
distance-transitive graphs.

Lemma 3.5.1 Let 1 <i < d where d is the diameter of A. Then

 3.5 The universal cover of &{Mat22) 119
(i) bt + ct<7 and ct > 1,
(ii) ifl<j<i then bj > bt and ct > cjf
(iii) |A|+i(x)| < (bi/cM) • |A|(x)|.
Proof. Since \A(y)\ is the number of points in the protective plane ny of
order 2, which is 7, (i) follows. Let X G A J G A,-(X) and z G A,_;(x)nA;(};).
Then
A, +1 (x) n AGO <= A y + i(z) n AGO,
Aj-i(z) Pi AGO <= A,_i(x) n AGO,
and we obtain (ii). Let D be the set of edges {u,v} such that u e Af(x),
v G A,-+i(x). Then every vertex from A,-(x) is incident to at mostft,-edges
from D while every vertex from A,-+i(x) is incident to at least c,+i edges
from D. Hence

and (iii) follows. D

Lemma 3.5.2 bi < 4.

Proof. Let (x,y,z) be a 2-arc such that z e A2(x) and S = S(x,y,z).
Since E is of diameter 2, S(z) c A2(x) and |A3(x) n A(z)| < 7 - |S(z)| =
4. D

Lemma 3.5.3 If y G A^x) f/ien tftere 15 a Petersen subgraph S containing

y such that E(y) ^ A4(x). Furthermore \E(y) n A3(x)| > 2, in particular
c4 >2.
Proof. Let (x, M, Z, V, y) be a 4-arc in A joining x and y. Put S =
E(z,v,y) and O = E(X,M,Z). Since E and O have a common vertex z, they
have a common edge, say {z,w}. Since both E and O have diameter 2,
w G A2(x) n A2()0 and E c A4(x). Both E(j) n S(z) and S(y) n S(w) are
non-empty while S(z) n S(w) is empty since there are no triangles in E.
Hence \E(y) n A3(x)| > 2. •

Lemma 3.5.4 bs < 1.

Proof. Let y G As(x) and suppose to the contrary that z\ and z2 are dif-
ferent vertices from A(>0nA6(x). Let a G A(^)nA4(x). Let A = E(zi,y,z2),
&i = E(a, y, Zi) for i = 1 and 2 and let E be the Petersen subgraph con-
taining a and contained in A4(x) whose existence is guaranteed by (3.5.3).
Since a Petersen subgraph has diameter 2, for i = 1 and 2 the subgraph
120 Geometry ofMathieu groups

®; intersects S in an edge which joins a to a vertex x, in H(a) n A4(x). By

(3.5.3) |S(a) n A3(x)| > 2 and hence xi = x 2 . This means that the 2-arc
(x\,a,y) is contained in Oi and in <1>2, so Oi = <1>2. Since {z\,y} is an
edge of Oi and {y,z2} is an edge of <X>2, we obtain €> i = O 2 = A and
hence {y, a] is an edge in A. Since a was taken to be an arbitrary vertex
from A(y) n A4(x), we have A(y) n A4(x) c A(y). Since the valency of A
is 3 and zuz2 € A(y) n A6(x) we must have |A(y) n A4(x)| = 1, which is
impossible, since |A(y) n A4(x)| > c5 > c4 and c4 > 2 by (3.5.3). D
Now we are ready to establish the upper bound.

Proposition 3.5.5 A P-geometry of rank 3 (possibly not flag-transitive)

contains at most 1898 elements of type 3.
Proof. Using (3.5.1), (3.5.2) and (3.5.3) we obtain |A0(x)| = 1, |A(x)| =
7, |A2(x)| < 42, |A3(x)| < 168, |A4(x)| < 336 and |A5(x)| < 672. By
(3.5.1 (iii)), (3.5.3) and (3.5.4) |A I+ i(x)| < \ • |A,(x)| for i > 5. Hence
Zf=6|Aj(x)| < |A5(x)| < 672 and the result follows. •
The existence of a triple cover ^(3 • Mat 22 ) of ^(Matri) can be estab-
lished using the following result (see [Maz79] and references therein).

Lemma 3.5.6 There exists a group F ~ 3 • Aut Matii having a normal

subgroup Y of order 3, such that F/Y = AutMat 2 2 . The commutator
subgroup F^ of F is a perfect central extension of F^ = Matii by Y and
F b = C~(7). If D^ is the preimage in F of a subgroup Mat2\ = L3(4) in
F b , then 5 b = SL3(4) and an element from Nj(Db)\D^ induces on & a
field automorphism. •
Let F ~ 3 • Aut Main and F b ^ 3 • Mat2i be as in (3.5.6) and let
(p : F —• F be the canonical homomorphism. Let si = {FJ7 | 1 < i < 3}
be the amalgam of maximal parabolics corresponding to the action of
F b on ^(Mat22y For £ = 1,2 and 3 let FJ7 = (p~l{F\) be the preimage
of Ff in F^. Since a Sylow 3-subgroup of F\ is of order 3 and every
such subgroup is inverted in its normalizer, we conclude that F\ splits
over Y and since O3(F\) = p\ we obtain the following direct product
decomposition:
ff = O\F\) x y,
where O3(FJ>) = *f• Let si = {O3(Ff) | 1 < i < 3}. It is easy to check
that for 1 < i < j < 3
O\F\) n O\F)) = O\F\ n F)) ^Ffn F)
and hence (p induces an isomorphism of si onto si.
 3.5 The universal cover of ^(Mat22) 121

Lemma 3.5.7 Let <&(3 • Mat2i) be the coset geometry ^ ( F b , s/\ Then cp
induces a (3-fold) covering of&(3-Mat22) onto &(Mat22) which is universal.
Proof. Since q> induces an isomorphism of si onto si9 it induces a
covering
q> : $(F\^) -> 3?(F b ,s/) ^ ^(Ma£ 22 )
(denoted by the same letter cp). Notice that ^(3 • Mat22) is connected
since F17 does not split over Y. Since ^(3 • Mat22) is a triple cover of
^(Mat22), it contains 990 = 330 • 3 elements of type 3. A flag-transitive
(in particular the universal) cover of ^(3 • Mat22) has 990 • n elements of
type 3, where n is either an integer or infinity. Since 990-2 > 1898, (3.5.5)
implies the simple connectedness of ^(3 • Mat22). D
We could define ^(3Mat 2 2 ) to be the coset geometry ^ ( F , {JV~(O3(FJ>)) |
1 < i < 3}) and this shows that it admits F as a flag-transitive automor-
phism group.
The derived graph of ^(3 • Mat22) is a triple antipodal cover of the
derived graph of ^(Mat22) and it has the following distance diagram:
2 2 1 2 2

O 7 1 /^~\
(7J
6 1 /"""N 4
(42)
1 /^-N 4
(l68)
2 X~"\ 4
(336)
4 /^"""N 1
(336)
4 ^"N
(84)
1 6 /""^v 1
M4)
7 /"""N
[2J

Let J = &(Sp4(2)) be the subgeometry in <g(Mat22) and g = 2 4 : Sym6

be the stabilizer of 1 in F. Recall that Q is the stabilizer in G = Mat24 of
an octad 5 and a 2-element subset Y2 in B. In particular 02(<2) = Gb acts
regularly on SP \ B and hence for p € SP \ B the subgroup Q(p) = Sym^
is a complement to Q& in Q. Since g(p) acts naturally on the 6-element
set B \ 7 2 , it is easy to see that B \ Y2 is a hyperoval in I1(Y2 U {/?}) and
is its full stabilizer in G[Y2 U {p}] ^ PrL 3 (4).

Lemma 3.5.8 Let J fte the preimage of 2, in &(3 • Mat2i) and Q be the
preimage of Q in F. Then 2, is the rank 2 tilde geometry and Q induces the
automorphism group of 2.
Proof. It is clear that O2(Q) = O2(Q) = 2 4 . Let Q(p) be the preimage
of Q(p) in F, which is clearly a complement to O2(Q) in Q. By (3.5.6),
(2.7.13) and in view of the discussion before the lemma, Q(p) = 3 • Sym^
is the automorphism group of the rank 2 T-geometry. Let {5ci,x2} be a
flag in J and Qt be the stabilizer of 3c,- in F, i = 1,2. Then O2(Q) < Qt
and Qi/O2(Q) = Sym* x 2 is a complement to Y in the stabilizer in F of
5c,-. Now the result follows directly from (2.6.2). •
122 Geometry ofMathieu groups
It is clear that the stabilizer of 1 in F^ = 3 • Matu is isomorphic to
4
2 : 3 • Alts and it induces on 1 a flag-transitive action.

3.6 ^(Mat22») is 2-simply connected

Let 3tf = ^(Mar23) be the rank 4 P-geometry of the Mathieu group
H = Mat23 with the diagram
P
2 2 2 1

Let si = {Hj; | 1 < i < 4} and 3d = {Ptj | 1 < i < j < 4} be the amalgams
of maximal and rank 2 parabolics corresponding to the action of H
on 2tf and associated with a maximal flag {xi,X2,X3,X4} where x, is of
type L In this section we follow [ISh90a] to show that ^f is 2-simply
connected by proving that H coincides with the universal completion U
of the amalgam 38. First we show that U is a completion and hence the
universal completion of s/. After that we prove the simple connectedness
of JT.
For k = 2,3 and 4 the residue res^(xfc) is 2-simply connected which
means that Hk is the universal completion of the amalgam 0^ = {Py |
1 < i < j < 4,i ^ k,j ^= k} of rank 2 parabolics corresponding to the
action of Hk on resjf (x^). Since the amalgam J / 1 = {Hk | 2 < k < 4}
contains ^ and is generated by the elements in 38, we have the following.

Lemma 3.6.1 U is the universal completion of the amalgam <s/1. •

The amalgam 3} = {^34,^24,^23} is the amalgam of maximal parabolics

corresponding to the action of H\ = Matu on res^(xi) = ^{Matii) and
by (3.5.6) the universal completion of Q) is isomorphic to 3

Lemma 3.6.2 Let D be the subgroup in U generated by 2. Then D = H± =

Proof. We know that D is isomorphic either to H\ or to the universal

completion of 2 which is 3Mat22. Let Sf = ^(Altj) be the subgeometry in
^f as in (3.4.2) and S = 24 : Alt7 be the stabilizer of Sf in H. Assuming
without loss of generality that {xi,X2,X3} is a maximal flag in S?> let
Si = SnHt denote the stabilizer of x, in S. Let S = {S,-nS/ | 1 < i < j < 3}
be the amalgam of minimal parabolics corresponding to the action of S
on Sf. Since StOSj < Pku whenever {i,j9k,l} = {1,2,3,4}, $ is contained
in 3$ and hence also in U. Let E be the subgroup in U generated by
 3.6 ^(Matiz) is 2-simply connected 123
S and for i = 1,2,3 let Tt be the subgroup in E generated by St n Sj
and SiHSk where {i,j,k} = {1,2,3}. Since Hk is the universal completion
of @{k) for k = 2 and 3 we have T2 ^ S2 and T3 ^ S3. If £ = i*i
then Ti ^ Si and if D ^ 3 • Ma£22, then by (3.5.8) T{ ^ 24 : 3 • 4/t 7 .
In the latter case 5" = &(E,{TU T2, T3}) is a flag-transitive rank 3 tilde
geometry possessing a morphism onto Sf = ^(Alt-j) which commutes
with the flag-transitive action. By (6.11.4) there are no such geometries
and the result follows. •
Since ^(Alt-j) is simply connected by (1.7.1), the proof of (3.6.2) has
the following implication.

Lemma 3.6.3 In the above terms E^S^l4 : Alt7 and F^ ¥ ^ 9(Alt7).

•
By (3.6.2) U contains si and hence it is the universal completion of si.
This means that the universal 2-cover of ^f coincides with its universal
cover.

Lemma 3.6.4 The geometry tf = ^(Mati?) is simply connected.

Proof. Let \p : #? —• ffl be the universal covering of ffl. We are going
to show that ^f and #? have the same number of elements of type 1,
which is 23. By (3.6.3) a connected component of the preimage in ^f
of the subgeometry Sf = ^{Alt-i) is isomorphic to Sf. Let & be a graph
on the elements of type 1 and 2 in Jf in which 2 distinct elements are
adjacent if they are incident in tf and let 0 be the analogous graph
associated with #?. Since \p is a covering of geometries, it induces a
covering of 0 onto 0 (denoted by the same letter \p). Recall that the
elements of type 1 and 2 in 3tf are the elements and the 3-element
subsets of 9 \ Y\9 respectively, with the incidence relation via inclusion.
Let x be an element of type 1 in Jf and x = \p(x) (where x is also
considered as an element from 0>\Y\). Since res^(x) = ^(Mat22), there
are 231 • (3 — 1) = 462 2-arcs in 0 originating in x and, since \p is a
covering, the same number of 2-arcs in 0 originate in x. Let
C = (x, {x, y, a}, y, {x, y, b}9 x)

be a 4-cycle in 0, where y,a,b are distinct elements from 0>\Yi. There

is an octad B which contains Y\ U {x,y,a,b}. Hence C is contained in
the ^04/^7)-subgeometry associated with B and C is contractible with
respect to xp. Since C was taken to be an arbitrary 4-cycle containing
124 Geometry ofMathieu groups

x, we have |©2(x)|^= |®2(x)| = 462/21 = 22. This implies in particular

that for any j / , z 6 ®2(3c) there is an element {5c,y,z} of type 2 in ^f. Let
y,z e ®2(x) and let
C = (x, {x, y9 a}, y, {y, z, a}, z, {x, z, a}, x)

be a 6-cycle where a e 0* \ (Y\ U {x,)>,z}). Then C is contained in the

^(,4/f7)-subgeometry determined by the octad containing Y\ U {x,y,z9a}.
Hence C is contractible, which means the following. Whenever 2 elements
of type^ 1 in j f are joined by a 4-arc in 0 , they are joined by a 2-arc.
Since ©is connected, in view of the above this means that altogether
there are 23 elements of type 1 in j f and the result follows. •

Now combining (3.6.2) and (3.6.4) we obtain the final result of the
section.

Proposition 3.6.5 The geometry &(Mat23) is 2-simply connected. •

3.7 Diagrams for Jf (Mat24)

Consider the maximal parabolic geometry 3tf = J^(Mat24) as a 3-partite
graph with the partition {Jf&, Jfft9 Jfs} where #?x is the set of octads, trios
and sextets for x = b,t and s, respectively. In this section we calculate
the suborbit diagrams of #C with respect to the action of G = Mat2A-
For x,y e {b, t, s}, by Ny we denote the orbit of Gx on J^y of length N.
It turns out that in all cases x is uniquely determined by the pair (N, y),
so that there is no need to mention x explicitly. The suborbit diagram
with the base vertex taken from J^fx will be denoted by Dx(Mat2A). In
Dx(Mat24) the valencies of Ny will always be given as sums of lengths
of orbits of Gx n G(Y) for Y e Ny. The diagrams Dx(Mat2A) together
with similar diagrams for the maximal parabolic geometry of Aut Ma*22
(to be calculated in Section 3.9) are of crucial importance for studying
geometries of larger sporadic groups, especially of J4. We identify the
elements of ^f with vertices, triangles and quads in the octad graph T
and when talking about distances we mean the distances in F. As usual
B, T = {#i,£ 2 ,£ 3 } and Z = {Si,S2,...,S6} are typical octad, trio and
sextet, respectively.

To calculate the diagrams Dx(Mat24) it will be helpful to analyse the

action of Qx on the J^fy. Since the distribution of classes of involutions in
Qx is given in (2.14.5) and for an involution its permutation character on
 3.7 Diagrams for 125
J^y can be taken from (2.14.2 (iii)) and (2.14.3 (v)), it is straightforward
to calculate the number of orbits of Qx on Jt?y. In order to find out
how many Qx-orbits are contained in a given Gx-orbit, we will use the
following rather obvious lemma.

Lemma 3.7.1 Let F be a group acting transitively on a set Q> of size n = 2r-
m where m is odd and suppose that \O2(F)\ = 2s. Then there is t < min{r,s}
such that every C>2(F)-orbit on O has length 2l and F/C>2(F) permutes these
orbits transitively. In particular the maximal number of O2(F)-orbits is m
and if F(x) is the stabilizer in F of x e ^ then F(x)O2(F)/O2(F) is the
stabilizer in F/02(F) of the O2(F)-orbit containing x. •

We start with the easiest diagram Db(Mat24)-

Lemma 3.7.2 The following assertions hold:

(i)Gb has four orbits on J^b with lengths 1, 30, 280 and 448,
(ii)Gb has three orbits on Jtft with lengths 15, 420 and 3360,
(iii)Gb has three orbits on 3tfs with lengths 35, 840 and 896,
(iv) I / I G 840s then there is a unique octad Bf adjacent to E such that
(GbnGs)Ks = GsnG[B'l
(v) the diagram Db(Mat24) is as given below.

Db(Mat24)
126 Geometry ofMathieu groups
Proof. Part (i) follows from (3.2.1). It is easy to deduce from the
proofs of (3.2.1) and (3.2.4) that if B' e rt(B) for i = 1,2 and 3
then Gh n G[Br] acts transitively on T(Bf) n Tj(B) for j = i - 1, i
and i 4- 1. This immediately implies (ii) together with the valencies
between the Nt and Afc. By (2.14.1) and straightforward calculations
we obtain (iii). To prove (iv) notice that by (2.14.1 (iii)) there is a
unique octad B' adjacent to £ and disjoint from B. Specifically, if
\B n S,-1 = 2 for 1 < i < 4, then Bf = S5 U S6. Hence Gb n Gs <
G[Bf] and since Gfc n Gs ~ [26].Sym3, it is sufficient to show that
\Gb nKs\ < 23. Since B n Si is not stabilized by an element of order
3 from Ks, Gb C)KS < Qs. Let JR, be the kernel of the action of Qs
on St. Since Qs/Rt are the points on a hyperoval on the GF(4)-space
dual to Qs (2.10.2), ^ n R 2 n R 3 = 1. On the other hand for i = 1,
2 and 3 the stabilizer of Si n B in the action of order 22 induced
by Qs on S, is of order 2, hence \Gb HKS| < 23 which implies that
(Gb n Gs)/(Gb nKs) = Sym4 x 2 and (iv) follows. Now the valencies of Ns
are straightforward from the possible shapes of the multiset v in (2.14.1).
The information on the stabilizers Gb n Gs contained in (2.14.1) and (iv)
shows that the orbit under Gb n G[E] of an element adjacent to £ is
uniquely determined by the orbit of Gb containing this element. This
gives (v). •

Ds(Mat2A)
 3.7 Diagrams for 127

Lemma 3.7.3 For II = {S{,S£,...,S£} € Jf, \ S let G

's be the
stabilizer of
V in G, K's = O2,3(G'S) and Q's = O2(G'S). Then

(i) Gs has four orbits on 3fs with lengths 1, 90, 240 and 1440,
(ii) Gs has three orbits on J^fb with lengths 15, 360 and 384,
(iii) L(Z') := (Gs n Grs)K's/K's is the stabilizer in GfJKfs of a point from
if II e 90s and L(S') is the stabilizer of a line in this
residue if II e 240s U 1440s,
(iv) Gs has four orbits on jft with lengths 15, 180, 720 and 2880,
(v) the diagram Ds(Mat2^) is as given above.

Proof. Part (i) follows from (3.3.5) while (ii) follows from (2.14.1). Let
A be the 6 x 6-matrix whose (i,y)-entry is \St n Sj|. It is easy to deduce
from the proof of (3.3.5) or otherwise that A is of the form

220000 \ 310000 \ ' 200011

220000 130000 020011
002200 001111 002011
002200 001111 or 000211
000022 001111 111100
000022 001111 111100

if \i = (2,2), (3,1) or (2,1,1), respectively. From this it is straightforward

to calculate the valencies between the Ns and Afo (but not necessarily
the decompositions of the valencies into sums of orbit lengths). Direct
calculation with the data in (2.14.5), (2.14.2 (iii)) and (2.14.3 (v)) shows
that Qs has 105 orbits on Jts \ I . Since 90 = 45 • 2, 240 = 15 • 2 4 and
1440 = 45 • 2 5 , by (3.7.1) we conclude that Qs has 45, 15 and 45 orbits
on Ns for N = 90, 240 and 1440, respectively. Since Ks/Qs < Gs/Qs,
the action of Ks/Qs on the set of Qs-orbits on Ns is either trivial or
fixed-point free. In view of (2.12.7) we conclude that the action is trivial
if and only if N = 240. Hence Ks has exactly 15 orbits on Ns for N = 90,
240 and 1440. By (3.7.1) and (2.5.3 (vii)) L(Z') is the stabilizer in GfJK's
of an element a from res^(S'). The matrices A given above now show
that a is a point if II e 90s and a line otherwise, so (iii) follows. By (iii)
G's acts transitively on the set of octads adjacent to II and contained
in Nb unless N = 360 in which case there are 2 orbits with lengths 1
and 6.
128 Geometry ofMathieu groups
Recall that B e Nb for N = 15, 360 and 384 if v(£,Z) = (4 2 0 4 ); (2402)
and (3 I 5 ), respectively, where v(£,S) is the multiset as in (2.14.1). For a
trio T = {BUB29B3} clearly the set k(T) = {v(Bi,T,) | 1 < i < 3} is an
invariant of the Gs-orbit containing T. Certainly every trio is refined by
some sextet. Thus, considering for each of the 3 matrices A given above all
possible partitions of the set of columns into 3 pairs and summing up the
pairs of columns, we obtain the following exhaustive list of possibilities
for k(T):

k{ = {(4204), (4204), (4204)}, k2 = {(4204), (2402), (2402)},

h = {(2402), (2402), (2 4 0 2 )}, k4 = {(2402), (3 I 5 ), (3 I 5 )}.

For 1 < i < 4 let 0 , denote the set of trios T such that k(T) = kt. Within
the analysis of the partitions of columns of the matrices A we also obtain
the numbers of trios in ®, adjacent to a sextet 2/ depending on i and on
the Gs-orbit containing £'. In view of (iii) these numbers (as on Ds(Mat24)
below) show that for every i the subgroup G's acts transitively on the set
of trios in ©; adjacent to E\ This implies that for every 1 < i < 4 the set
©, is a Gs-orbit. In addition the number of octads in Nb adjacent to a
given T G 0 ; is readily seen from the shape of Xt. Let us determine the
sizes of the ©,. Since v(£,E) = (4204) if and only if B (as a vertex of T)
is contained in the quad Z, we conclude that 0 i consists of the trios (the
triangles in F) contained in £ while 02 consists of the trios intersecting
D in a single vertex. This gives |©i| = 15, |©2| = 180. All the 15 trios
adjacent to an octad from 384& are contained in © 4 while every trio from
©4 is adjacent to 2 octads from 384&. Hence |®4| = 2880 and since the
total number of trios is 3795, we have |©3 | = 720 and (iv) follows. To
complete the proof of (v) it remains to show that if T G © 3 = 720t then
the stabilizer of T in Gs permutes transitively the octads in 360& adjacent
to T and we suggest this as an exercise. •

Lemma 3.7.4 The following assertions hold:

(i) Gt has three orbits on &\ with lengths 3, 84 and 672,
(ii) Gt has four orbits on J^s with lengths 7, 84, 336 and 1344,
(iii) L(Z) := (Gt n GS)KS/KS is the stabilizer in Gs/Ks of a point from
resjf (L) if Z G lt U 336t and L(Z) is the stabilizer of a line from
this residue if^e 84, U 1344,,
(iv) Gt has four orbits on jft\T with lengths 42, 56, 1008 and 2688,
(v) the diagram Dt(Mat2A) is as given below.
 3.7 Diagrams for 3tf?(Mat24) 129

Dt(Mat24)
Proof, (i) is dual to (3.7.2 (ii)) while (ii) is dual to (3.7.3 (iv)). Notice
that B G Nb for N = 3, 84 and 672 if and only if the distance between
T and B is 0, 1 and 2, respectively. Similarly, Z e 7S if Z contains T,
Z € 84S if Z intersects T in a vertex, Z e 336S if every octad in T is at
distance 1 from Z and Z e 1344S if exactly 1 octad in T is at distance 1
from Z. This shows that a sextet from ls or 84S is adjacent, respectively,
to 3 or 1 octads in 3& while a sextet from 336S or 1344S is adjacent,
respectively, to 3 or 1 octads from 84S. To prove (iii) consider the action
of Qs on Jf t (compare Ds(Mat24)). By (2.14.5), (2.14.2 (iii)) and (2.14.3
(v)) Qs has 150 orbits on J^t. Since 15 = 15-2°, 180 = 45-2 2 , 720 = 45-2 4
and 2880 = 45 • 2 6 , by (3.7.1) Qs fixes 15, elementwise and has 45 orbits
on Nt for N = 180, 720 and 2880. In view of (2.12.6) we conclude that
Ks fixes 15, elementwise and acts fixed-point freely on the set of orbits
of Qs on Nt for N = 180, 720 and 2880. Hence Ks has 15 orbits on each
of the Nt. By (2.5.3 (vii)) and the above established partial information
on the valencies between the Ns and Nb on Dt(Mat24) we obtain (iii) and
easily calculate the remaining valencies between the Ns and N&.
Consider the action of Gt on JtPt \ T. For T = {B[,B'2,Bf3} ± T put
co(T, V) = em where e = d(T, T ) (the distance in the octad graph) and
m is the number of pairs (i,j) such that d(Bi,B'j) = e. By (3.2.1 (iii), (iv)),
for every 1 < i < 3 there is a unique octad in V nearest to Bt and every
4-cycle in T is contained in a quad. In view of these observations it is
easy to see that co(T, Tf) is one of the following:
a > i = 0 1 ; a>2 = l 3 ; C93 = l 1 ; a>4 = 2 3 .
130 Geometry of Mathieu groups
Let Qt be the set of V e Jf t \ T such that co(T, V) = cot for 1 < i < 4.
Since Gb acts doubly transitively on the set of trios containing B, Qi is
a Grorbit of length 42. Since every 4-cycle in F is in a unique quad,
co(T, T') = a>2 if and only if T and T' are disjoint but contained in a
common quad. Since Gs permutes transitively the pairs of disjoint trios
in S, Q2 is a Gt-orbit of length 56. Let L G 1344S, so that exactly 1 octad
from T is at distance 1 from E. Then 3 trios from Z are in Q3 and 12
are in Q4. On the other hand by (iii) Gs D Gt acting on the set of trios
contained in £ has 2 orbits with lengths 3 and 12. Hence Q3 and Q4 are
Grorbits. It is easy to calculate that |Q3| = 1008, so that |Q4| = 2688 and
(iv) follows. Now the rest of Dt(Mat2A) is straightforward except possibly
for the transitivity of the stabilizer in Gt of T" G 2688t on the octads
contained in V and we suggest this as an exercise. •

3.8 More on Golay code and Todd modules

In this section we analyse the structure of 11-dimensional Golay code
#11 and Todd # n modules as GF(2)-modules for G&, Gt and Gs. In order
to simplify the notation, we put X = # n , Y =%>n-
Let x = b, tors. Since X and Y are GF(2)-modules, Qx acts trivially on
each irreducible composition factor of Gx in X or Y. In particular every
minimal Gx-submodule in X or Y is contained in Cx(Qx) or Cy(Qx\
respectively. In addition, since X and Y are dual to each other, Cx(Qx)
is dual to Y/[Y,QX] and CY(QX) is dual to X/[X,QX]. So it is natural to
calculate first the centralizers of the Qx in X and 7. We start with the
following.

Lemma 3.8.1 If x = b, t or s then Qx does not stabilize pairs of comple-

mentary dodecads.
Proof. Let {D, D'} be a pair of complementary dodecads stabilized by
Qx. Then a subgroup of index at most 2 in Qx stabilizes D. It is easy
to deduce from (2.10.1), (2.10.2) and (2.10.3) that for every subgroup of
index 2 in Qx a union of size 12 of its orbits always contains an octad, a
contradiction with (2.11.2). •

Lemma 3.8.2 For x — b, t and s put Ax = CX(QX) and Bx = CY(QX).

Then Gx acts irreducibly on Ax and Bx and the following hold:
(i) \Ab\=2,Bb^/\2Qb,
(ii) in terms of (2.10.3 (viii)) At = £>i and Bt is the dual of D2,
(iii) As is the natural symplectic module ofGs/Ks = Sp^(2) and \BS\ = 2.
 3.8 More on Golay code and Todd modules 131
Proof. The dimensions of the Ax and Bx are straightforward from
(2.15.1), (3.8.1) and the diagrams Dx(Mat24). The sextets in 35S are indexed
by the 2-dimensional subspaces in Qb. Applying (3.3.8) and (2.4.6) we
obtain (i). The non-zero vectors in At are indexed by the octads in T and
their sum is zero, hence At = D\. If B = Bi then D2 = Qb n Qt. The non-
zero vectors from Bt are indexed by 2-dimensional subspaces in D2 and
(ii) follows. The non-zero vectors of As are indexed by the octads in the
quad E. The sum of the vectors corresponding to octads in a triangle is
zero. Hence As supports a natural representation of res^(Z) = ^(Sp4(2))
and (iii) follows by (1.11.2). •
By (3.8.2), if Z = X or Y then CZ(QX) and [Z, Qx] are, respectively, the
only minimal and the only maximal proper Gx-submodules, in particular
Cz(Qx) < [Z,QX]. In addition [X,QX]/CX(QX) and [Y,QX]/CY(QX) are
dual to each other.

Lemma 3.8.3 For x = b, t and s put Cx = [X,QX]/CX(QX)- Then Gx acts

irreducibly on Cx and the following hold:
(i) Ch is dual to Qh,
(ii) Ct is dual to Qt,
(iii) Cs is isomorphic to Qs.

Proof. By (3.8.2) the dimension of Cx is 4, 6 and 6 for x = b, t and s,

respectively. It is easy to see that Cb is generated by the images of octads
from 30b and these images are indexed by the trios containing B. Since
the setwise stabilizer in Qb of such a trio is a hyperplane in Qb, (i) follows.
For x = t and s let Ux be a Sylow 3-subgroup in 02,3(GX). Comparing
the dimensions of the centralizers of Ux in 2^, CX(QX) and CY(QX),
we conclude that Cx, as a module for Nx := NGX(UX), is isomorphic to
[UX,X]. We claim that the action of N on Cx is faithful. In fact for
x = t it is immediate from the dimension of the centralizer in X of
an element of order 7 and for x = s it follows from the fact that Nx
does not split over Ux. Hence Cx is an irreducible 3-dimensional GF(4)-
module for Nx. It is well known and easy to check that every subgroup
in PFL3(4) isomorphic to Sym^ or L^(2) x 2 stabilizes a hyperoval or
a Fano subplane in the corresponding projective plane of order 4. This
shows that Cx is isomorphic to Qx or to its dual. Since a hyperplane in
Cb is contained in Q, (i) implies (ii). From the diagram Ds(Mat24) we
observe that [Y,QS]/CY(QS) (which is dual to Cx) is generated by the
images of sextets from 90s and these images are indexed by the *-closed
132 Geometry ofMathieu groups
triangles in the sextet graph. Since the elementwise stabilizer in Qs of
such a triangle is a hyperplane in Qs, (iii) follows. •
Thus we have the following main result of the section.

Lemma 3.8.4 For Z =%>n or <€\\ and x = b, t or s

KCZ(QX)<[Z,QX]<Z

is the only composition series of Z as a module for Gx; the composition

factors are as in (3.8.2) and (3.8.3). •

In terms of (3.8.4) let Xs be a Sylow 3-subgroup in Gs. Then clearly

Z=CZ(XS)®[Z,XS]
and using (3.8.2) and (3.8.3) one can easily show the following.

Lemma 3.8.5 Cz(Xs) is an reducible indecomposable 5-dimensional mod-

ule for GS/O2,3(GS) = Sp4(2). Moreover Cgn(Xs) contains the natural 4-
dimensional symplectic submodule while C%n(Xs) contains a 1-dimensional
submodule. •

3.9 Diagrams for

An element of type 2 in ^(Matu) is a pair {2*1,2*2} °f disjoint octads,
both disjoint from Y2. Such an element determines a unique octad

B = 0>\(B1UB2)
containing Y2. In its turn B determines a ^(S/>4(2))-subgeometry con-
taining {2*1,2*2}. Thus an element of type 2 is contained in a unique such
subgeometry. The maximal parabolic geometry j f (Mat22) introduced in
[RSm80] can be defined as follows. The elements of type 1 and 3 to-
gether with the incidence between them are as in <3(Mat22)\ the elements
of type 2 are the ^(Sp4(2))-subgeometries with an element of type 1 or 3
being incident to an element of type 2 if in ^(Mat22) it is incident to an
element of type 2 in the subgeometry. Notice that an element of type 1
is incident to an element of type 2 in a subgeometry if and only if it is
contained in the subgeometry. In order to distinguish between ^(Mat22)
and 3tf(Mat22\ the elements of type 1, 2 and 3 in the latter geometry will
be called pairs, hexads and octets, respectively. The diagram of 3tf?(Mat22)
is the following:
 3.9 Diagrams for J^{Mat22) 133
octets

pairs a- 6 hexads

In this section we calculate the suborbit diagrams of ^f as a 3-partite

graph with the partition

J T = jfp u Jffc U Jfo

where j f p, Jf h and Jf 0 are the pairs, hexads and octets, respectively. The
diagrams are with respect to the group F = Aut Mat22 acting naturally
on Jf. If {Fp,Fh,F0} is the amalgam of maximal parabolics associated
with the action, then

Fp * 25 : Sym5, Fh^ 24 : 5ym6, F o s 2 x 23 : L3(2).

A typical pair, hexad or octet will be denoted by P, if or 0, respectively.

For x = p, h or o put Qx = C>2(FX).
To calculate the diagrams it is helpful to know the conjugacy classes
of involutions in F and their distributions inside the Qx.

Lemma 3.9.1 The group F = AutMat22 has 3 classes of involutions: 2a,

2b and 2c; an involution from F is contained in F^ if and only if it is a
2a-involution. Furthermore,
(i) a 2a-involution s can be chosen so that s E Qh and Cf(s) = CFh(s) =
2 4 : (Sym4 x 2), s fixes 23 pairs, 13 hexads and 26 octets,
(ii) a 2b-involution t can be chosen so that t G Qo and Cp{t) = Fo, t
fixes 35 pairs, 21 hexads and 50 octets,
(iii) a 2c-involution u can be chosen so that u G Qp and Cp{u) —
CFp(u) = 25 : Frob\, u fixes 11 pairs, 5 hexads and 10 octets.

Proof. It is clear that the classification of involutions in F (resp. in

F b ) is equivalent to classification of the orbits of CG(T) on the set of 2-
element subsets of 9 stabilized (resp. fixed) by x for various involutions
T in G = Mat24. In view of this observation the classes of involutions
in F and the corresponding centralizers are immediate from (2.14.2) and
(2.14.3). The number of pairs stabilized by an involution v G F follows
directly from the cyclic shape of v on 9 \ Y2. Let x be a 2a-involution in
G and let B be the octad formed by the elements of 3Pfixedby T. Then
134 Geometry ofMathieu groups
the set of octads stabilized by T consists of B, 14 octads disjoint from B
which are orbits of the hyperplanes in Qb containing T and 56 octads Bf
such that \B C\B'\ = 4 and B' is refined by a sextet containing an orbit on
0> \ B of a subgroup of order 4 on Qb containing T (compare the proof
of (2.14.2)). Using this description it is straightforward to calculate the
numbers of hexads and octets stabilized by the involutions 5 and t. The
octads stabilized by u (which is a 2b-involution in G) are exactly those
refined by the sextet which contains the tetrad Y2UP where P is the pair
stabilized by Fp. Since 5 of these octads contain Y2 and 10 do not, (iii)
follows. •

Lemma 3.9.2
(i) Qh is la-pure,
(ii) Qo contains 1 and 8 2a- and Ib-involutions, respectively,
(iii) Qp contains 15, 10 and 6 2a-, 2b- and 2c-involutions, respectively.

Proof. Since Qh is the natural symplectic module for Fh/Qh = Sp4(2),

(i) follows. The structure of Fo implies that Fo/Qo acting on the set of
involutions in Qo by conjugation has three orbits with lengths 1, 7 and 7.
One of the orbits of length 7 is formed by the involutions contained in
Qo C\F^. By (3.9.1) a 2c-involution never commutes with an element of
order 3 and hence (ii) follows. One can deduce from (2.7.14) or otherwise
that the orbits of Fp/Qp on the set of involutions in Qp are of lengths 15,
10 and 6 (in particular Qp is indecomposable). The former of the orbits
consists of the involutions contained in QPC\F^. Since an involution from
the orbit of length 10 is centralized by a 3-element, (iii) follows. •

In Chapter 7 we will make use of the following result.

Lemma 3.9.3 For x = p, h and o put F% = F b D Fx. Then the following

assertions hold:
(i) C>2(FP) acting on 0> \ Y2 has one orbit of length 2 (the pair) and
five orbits of length 4, OiiF^) fixes every element in the pair and
acts transitively on every O2(Fp)-orbit of length 4,
(ii) O2(Fh) and 02{F\i) has the same orbits on &\Y2, namely, one orbit
of length 16 (the complement of the hexad) and three orbits of
length 2,
(iii) 02CF0) and O2(F£) have the same orbits on 0* \ Y2, namely, one
orbit of length 8 (the octet) and seven orbits of length 2,
(iv) Fx = NF(O2(FX)).
 3.9 Diagrams for 3tf?(Mat22) 135

Proof. In view of the definition of &(Mat22) in terms of subsets of

&\ Y2 the assertions (i)-(iii) follow directly from (3.9.1) and (3.9.2).
These assertions immediately imply (iv). •
For x = p, h and o by Dx(Mat22) we denote the suborbit diagram
corresponding to the action of F on Jf(Mat22) with the base point taken
from #ex.

Lemma 3.9.4 The following assertions hold:

(i) Fh has two orbits on J^h\H with lengths 16 and 60; if Hf e Nh
then \(H n H') \ Y2\ = 0 and 2 for N = 16 and 60, respectively,
(ii) Fh has three orbits on Jf0 with lengths 30, 60 and 240,
(iii) Fh has three orbits on Jtfp with lengths 15, 96 and 120,
(iv) (Fh n Fp)Qp/Qp is isomorphic to Sym4, Alts cind Sym^ x Sym2 for
P taken from 15P, 96P and 120p, respectively,
(v) the diagram Dh(Mat22) is as given below.

Dh(Mat22)

Proof. Suppose that Y2 G B so that H = B is a hexad and consider B

as a vertex of the octad graph T. Then Fh is the stabilizer of Y2 in G^,
in particular it contains <2&. By (2.10.4) and its proof Qb has 15 orbits of
length 2 on T(B) and these orbits are indexed by the hyperplanes in Q&.
If B' e Ti(B) for i = 2 or 3, then the Q^-orbit of B' (of length 4 or 16) is
uniquely determined by Bf OB (of size 4 or 2). Hence F& acts transitively
136 Geometry of Mathieu groups
on the set of 30 octets in T(B) and the orbit under Fh of Bf e rt(B) for
i = 2 or 3 is uniquely determined by i and \B'n Y2\. This means that there
are 60 hexads and 60 octets in T2(B), 16 hexads and 240 octets in T3(B)
and Fh acts transitively on each of these 4 sets. From this information
we easily deduce (i), (ii) and the valencies between the Nh and Mo on
Dh(Mat22).
Since the actions of Fh on B \ Y2 and 9 \ B are doubly transitive while
the action of Qb on 9 \ B is transitive, we conclude that the orbit of a
pair under Fh is uniquely determined by the size of its intersection with
H which implies (iii). Dualizing (iii) we obtain that Fp has 3 orbits on Jf /,
with lengths 5, 32 and 40. By (3.9.1) and (3.9.2) Qp has 17 orbits on jHfh
and it is easy to deduce from (3.7.1) that Qp fixes 5/, elementwise, has 2
orbits on 32/, and 10 orbits on 40/,. Since Syms = Fp/Qp has a single class
of subgroups of index 5 (isomorphic to Sym4) and a unique subgroup of
index 2 (isomorphic to Alts), to complete (iv) we have to show that for
P e 120p (Fh n FP)QP/QP is isomorphic to Sym$ x Sym2 rather than to
Alt4. If H1 is a hexad from 16/, and D is the stabilizer of Hf in Fh, then
D is a complement to Qb in Fh and hence it permutes transitively the 15
pairs in H'\Y2. This means that a pair from 120p is incident in ffl(Mat22)
to 2 = 16 • 15/120 hexads from 16/, and (iv) follows from the obvious fact
that AH4 does not have orbits of length 2 in the natural action of Sym5
of degree 5. Now using the divisibility it is straightforward to reconstruct
the remainder of Dh(Mat22). •

Lemma 3.9.5 The following assertions hold:

(i) Fo has three orbits on &?h wftfe lengths 7, 14 and 56 consisting of

the hexads intersecting O in no, four and two elements, respectively,
(ii) for a hexad H we have (FonFh)Qh/Qh = Sym4 and FonFh stabilizes
a pair in H if and only if H € 14/, U 56/,,
(iii) Fo has four orbits on Jf 0 \ O with lengths 1, 42, 112 and 168,
(iv) ifOf e No then FO[O']QO/QO is isomorphic to Sym4, Ds, Alt4 and
D 8 for N = 7, 42, 112 and 168, respectively,
(v) ifOf G7 O U 112 O then Fo[Of] stabilizes a hexad incident to O'',
(vi) Fo has four orbits on Jtfp with lengths 1, 28, 84 and 112,
(vii) if P e Np then (Fo n FP)QP/QP is isomorphic to Sym^ x Sym2,
Sym3 x Sym2, Dg and Sym4for N = 7, 28, 84 and 112, respectively,
(viii) the diagram D0(Mat22) is as given below.
 3.9 Diagrams for 137

D0(Mat22)

Proof, (i) is dual to (3.9.4 (ii)). By (3.9.1), (3.9.2) and (3.7.1) Qh has
15 orbits on No for N = 30, 60 and 240 (the diagram Dh(Mat22)) hence
(Fo n Fh)Qh/Qh = Sym4 for every hexad H. If H e 56& then Fo n Fh
stabilizes the pair H nO and if H G 14/, then Fo n Fh stabilizes the
pair H\((H n O ) U Y2). On the other hand if H e lh then Fo n Fh
permutes transitively the 3 pairs incident to both O and if, so we have
(ii). Part (iii) follows from the distance diagram of A = A(^(Mat22))
given before (3.4.5). The proof of (vi) is similar to that of (ii). Since on
the distance diagram of A we have c\ = a* = 1, if O' e l0 U 1120 then
Fo[Of] stabilizes an octet 0" adjacent to 0' in the derived graph. Hence
it also stabilizes the hexad & \ (Of U 0") and (v) follows. The group Fo
acts triply transitively on the elements in O; Fo/Qo = L^il) permutes
doubly transitively the 7 orbits of Z(FO) o n ^ \ ( O U Y2) and hence Qo
stabilizes each of these 7 orbits as a whole. This implies that Fo acts
transitively on the pairs in 0; on the pairs intersecting 0 in one element;
on the orbits of Z(F0) o n ^ \ ( O U Y2) and on the remaining pairs in the
latter set, so that (vi) follows. The proof of (vii) is similar to that of (ii).
Now with this information in hand it is straightforward to reconstruct

Every statement in the next lemma either is dual to a statement in

(3.9.4) and (3.9.5) or can be deduced by similar methods.
138 Geometry ofMathieu groups

Dp(Mat22)

Lemma 3.9.6 The following assertions hold:

(i) Fp has three orbits on J^h with lengths 5, 32 and 40,
(ii) if H G Np then (Fp n Fh)Qh/Qh is isomorphic to Sym*, Alts and
Sym4 for N = 5, 32 and 40, respectively,
(iii) Fp has three orbits on Jt?p with lengths 30, 40 and 160,
(iv) Fp has four orbits on J^o with lengths 10, 40, 120 and 160,
(v) if O e No then (Fp D F0)Q0/Q0 is isomorphic to Sym*, Sym4, Dg
and AlU for N = 10, 40, 120 and 160, respectively,
(vi) the diagram Dp(Mat22) is as given above. •

3.10 Actions on the sextets

In this section we describe the orbits of a few subgroups of G on J^s,
that is on the set of sextets.

Lemma 3.10.1 Let Y be a 3-element subset of&. Then G[Y] s PTL3(4)

acting on Jt?s has three orbits with lengths 21, 630 and 1120 consisting
of the sextets Z such that Y intersects i tetrads in £ for i = 1, 2 and 3,
respectively.

Proof. It follows from (2.10.2) that the orbit under Gs of a 3-element

subset X is uniquely determined by the multiset {\X n S,-| | 1 < i < 6}.
 3.10 Actions on the sextets 139
Now the result follows from the obvious duality and straightforward
calculations. •

Lemma 3.10.2 Let {Di,D 2 } be a complementary pair of dodecads and

R = Mat\2-2 be the setwise stabilizer in G of this pair. Then R acting on
tfs has three orbits with lengths 396, 495 and 880.

Proof. For a sextet Z = {Si,S2,...,S6} put mn = \Dj n St\ for j = 1,2,

1 < i < 6, and suppose that m is the maximum of the m;i. Then
m e {4,3,2} and we are going to show that each value of m corresponds
to a single orbit Qm of R on 3tfs. Notice that because Golay subsets
always have even intersection, all the values m;i have the same parity.
Suppose first that m = 4. By (2.11.2) Dj does not contain octads and
hence for 7 = 1 and 2 at most one of the values m;i can be equal to 4
and in order to meet the total balance exactly one of the m/,- must be 4.
Since the action on D\ of its stabilizer in R is 5-fold transitive, Q4 is an
£-orbit and |Q 4 | = (*42) = 495.
If m = 3 then for 7 = 1 and 2 exactly three of the m7, are equal to
3 and three to 1. By (2.11.9 (iii)) the stabilizer in R of an element from
D 2 induces on D\ a 3-fold transitive action of Mat\\ and hence Q3 is an
K-orbit and |Q 3 | = (x32) • 12 -\ = 880.
Finally, if m = 2 then all the m;i are equal to 2. For 1 < i < 6
put A = Stn D\ and B = Sid D2. By (2.11.3) A determines a partition
{Ti, T2} of D such that both A U Ti and i U T 2 are octads. If B a Tk
then Tfc \ B is a tetrad in E, which is impossible since m = 2. Hence
£ intersects both Ti and T2. Since the elementwise stabilizer of A in R
induces two inequivalent actions of Syme on T\ and T2, Q2 is an jR-orbit
and |Q 2 | = (x22) • 6 • 6 • £ = 396. •

Lemma 3.10.3 The subgroup G3 = 2++6 : L3(2) acting on ^fs has six orbits
with lengths 1, 28, 56, 336, 448 and 896.

Proof. By (3.3.4) G3 has index 15 in G&, it contains Qb and

is the stabilizer of a point in the natural action of Gb/Qb = 1-4(2) on
the rank 3 projective geometry n of the proper subgroups in Qb. Thus a
G3-orbit on J f s is a union of 2^-orbits and is contained in a G^-orbit.
By (2.14.1) and (3.7.2) Gb acting on Jfs has three orbits ©1, 0 2 and
©3 with lengths 35, 810 and 896; g^-orbits on ©, have lengths 1, 8 and
16, respectively. Furthermore, if 0 , is the set of Q^-orbits on ©,, then
— £4(2) acts on ®j as it acts on the set of lines in n, on the
140 Geometry ofMathieu groups
incident point-hyperplane pairs in n and on 3-element subsets of B for
i = 1,2 and 3, respectively. This and elementary geometric arguments
show that Gi/Qb has two orbits on ®i with lengths 7 and 28, three orbits
on ®2 with lengths 7, 42 and 56, and acts transitively on @3. The latter
follows from the 3-fold transitivity of G3/Qb = 23 : L3(2) on the octad B
stabilized by G/>. •
 4
Conway groups

The largest Conway sporadic simple group is the quotient over the centre
of order 2 of the automorphism group of the Leech lattice (the unique
even unimodular lattice of dimension 24 without roots). In Section 4.1
we recall some standard results concerning construction of lattices from
binary codes. In Section 4.2 we discuss some symmetries of the lat-
tices coming from the binary code construction. In Section 4.3 we follow
[Con69] to prove the uniqueness of the Leech lattice A; the proof immedi-
ately gives the order of the automorphism group COQ of A. In Section 4.4
we introduce the standard coordinate system for the Leech lattice and
describe explicitly the Leech vectors of length up to 8. In Section 4.5 we
discuss the action of COQ on the Leech lattice modulo 2 (denoted by A)
as well as on the Leech vectors of small length. In this way we introduce
the sporadic Conway groups Coi, C02 and C03. In Sections 4.6 and 4.7
we study the action of Co\ on the images in A of the Leech vectors of
length 8 and calculate the suborbit diagram of the Leech graph which
is the smallest orbital graph of this action. In Section 4.8 we study the
structure of the centralizer in Co\ of a central involution which we use
in Section 4.9 to construct the tilde geometry y(Co\) and the Petersen
geometry ^{Coi). In order to establish the simple connectedness of these
geometries in Section 4.12, in Sections 4.10 and 4.11 we study the afBne
Leech and the shortest vector graphs in terms of their suborbit diagrams.
In Sections 4.13 and 4.14 we discuss some further geometries possessing
natural descriptions in terms of the Leech lattice.

4.1 Lattices and codes

Let n be a positive integer, let R n be an n-dimensional Euclidean vector
space and for x,y e R n let (x,y) denote the inner product of x and

141
142 Conway groups
y. A lattice of dimension n is a subset L c R n with the property that
there exists a basis 08 of R" such that L consists of all integral linear
combinations of vectors from 08. In this case 38 is said to be a basis of
L. A lattice L cz F S R" and a lattice L' c F' S R n are isomorphic if
there is an isomorphism q> : V —> V of Euclidean spaces which sends L
onto L'.
Let L be a lattice of dimension n. The dual lattice L* of L is defined as
follows:
L* = {y\ye R n ,(y,x) € Z for every x e L}.

A lattice L is integral if L ^ L*, which means that the inner product of

any two vectors from L is an integer; L is even if (x,x) e 2TL for every
x e L. Since

() <
every even lattice is integral. Let 36 be a basis of an integral lattice L and
let A denote the Gram matrix of 08. The absolute value of the determinant
of A is independent of the choice of 08 and it is called the discriminant of
L, written as disc L. It is known [Ebe94] that disc L = |L*/L|. A lattice
L is unimodular if L* = L, in which case disc L = 1. We will write x 2 for
the (squared) length of a vector x, that is for (x, x). If x 2 = 2 then x is
called a root. We are mainly interested in even unimodular lattices. It is
known [Ebe94] that the dimension of such a lattice is divisible by 8. A
Leech lattice is an even unimodular lattice of dimension 24 which does
not contain roots. We will see in due course that there exists a unique
Leech lattice which has a certain remarkable group of automorphisms.
Let L and M be integral lattices of dimension n and suppose that
M ^ L. Then clearly L ^ M* so that L corresponds to a subgroup in
the finite abelian group M*/M. We will discuss a family of lattices which
contain specific sublattices and show that this family possesses a natural
description in terms of binary codes.
Let 0tn be a basis of R n consisting of pairwise orthogonal roots. This
means that 0tn is of size n, a2 = 2 for every a e 0tn and (a, b) = 0 for all
a,b€0$n with a ± b. Then for x G R n we have

x =

Let i f = &{@tn) be the lattice having ^ n as a basis. It is easy to see the

following.
 4.1 Lattices and codes 143
Lemma 4.1.1 The lattice ££ = J£f(^n) is n-dimensional and even with
discriminant 2n. •
For 8 = 0 or 1 put

maa ma G Z, 7 ma = & mod 2

Then if o is a sublattice of index 2 in if and if i is the only proper coset

of ifo in S£. We will study the lattices which contain ifo. In order to
understand these lattices it is helpful to have a description of the duals
5£* and J?Q, which is rather straightforward and given in the lemma
below.

Lemma 4.1.2 The following assertions hold:

(i) i£n is a lattice with the basis {\a \ a G 0tn);
(ii) x,y G JSf* are in the same coset of S£ if and only if for every a G @tn
we have (x,a) = (y,a) mod 2;
(iii) xe£?l\&* if and only if (x, a) G Z + \ for every aeMn. U
The above lemma possesses an easy reformulation in terms of coor-
dinates of vectors in the basis 0tn. To wit, x G if * if and only if every
coordinate of x is either an integer, or a half integer, x, y G S£* are in the
same coset of ^ if and only if x and y have the same set of non-integer
coordinates (which are half integers), x G Z£\ \ ^* ^ a n d on ty ^ every
coordinate of x is an odd integer divided by 4.
For a vector x G S£* put
X(x) = {a\a€@n, (x,a)G2Z+l},
so that a G X(x) if and only if the corresponding coordinate of x in the
basis Mn is not integral. By (4.1.2 (ii)) we have X(x) = X(y) if and only
if x and y are in the same coset of Se in £e\ Since |JSf*/JSf| = disc S£ =
2n = 12^n |, the mapping
<^> : x i—• X(x)

induces a bijection of <£*!<£ onto 2^» and for I g | n w e have

= i£ + ex, where e* = - Y^ a.

Lemma 4.1.3 Let x,y G «£f*,

144 Conway groups
(i) <p(x + y) = cp(x)Acp(y) = X(x)AX(y),
(ii) (x,y) e TL if and only if\X(x)C\X(y)\ e 2Z, i.e. if and only ifX(x)
and X(y) are orthogonal with respect to the parity form,
(iii) if X ^0tn then elx = \\X\, in particular ex is a root if and only if
1*1=4.
(iv) ifxeg** then x2 = e2x{x) + 21 for some I > 0.
Proof. Let z = x + y and a e ^n- Then (z,a) = (x,a) + (y,a) and
hence a € X(z) if and only if a is contained in exactly one of the sets
X(x) and X(y). So X(z) = X(x)AX(y) and (i) follows. To prove (ii) put
x = ex(x) + 0 and y = ex(y) + b for a, b G J£\ Then

(x, j ) = (eX(x), eX{y)) + (^(JC), 6) + (eX(y), a) + (a, b).

Since e ^ ) , ^ ) G J^7*, the last three terms on the right hand side of the
above equality are integers and, since (eX{x),ex{y)) = j\X(x)nX(y)\, (ii)
follows. The assertion (iii) comes by direct calculations. In (iv) if we put
x = eX(x) + afor ae J£, then
x2 = e2x{x) + 2(eX(x),a) + a2,
where a2 is even since ££ is even. In addition it is easy to see that for
X c 0tn and a e 0tn
(ex,a)= min \(x,a)\
xe£?+e
and hence (iv) follows. •
For * c 2®n put

(the ^-construction in [CS88]).

Lemma 4.1.4 The set 5£A{^€) is a lattice if and only if <& is a (binary
linear) code. If ^ is a code then
(i) S£A{^€) is integral if and only if%> is contained in its dual,
(ii) S£A^€) is even if and only if%> is doubly even,
(iii) ££A{%>) is unimodular if and only if^ is self dual.

Proof. If £fA{^) is a lattice then it is closed under addition and by

(4.1.3 (i)) this happens exactly when %> is closed under taking symmetric
differences, i.e. when ^ is a code. On the other hand if ^ is a code then
is closed under addition, and it is always closed under negation.
 4.1 Lattices and codes 145
A
Hence £^ (^) is a subgroup of finite index in S£* which means that it is
a free abelian group of rank n. This implies that S£A^€) is a lattice. Now
(i) holds by (4.1.3 (ii)) while (ii) holds by (4.1.3 (iii), (iv)). If <€ is a code
then the index of SeA(^) in S£* equals the index of * in 2*\ In view of
(4.1.1) SeA{^) is unimodular if and only if it is of index 2"/2 in <£* and
by (2.1.1) the index of * in 2^» is 2 n/2 if and only if <€ is self-dual. •
The above lemma reduces the classification of even unimodular lattices
containing ££(@tn) to that of doubly even self-dual codes based on 0tn.
Notice that all these lattices contain roots, since so does £?(&tn). Let us
turn to the lattices which contain if o but do not contain if starting with
those contained in if*. For X ^ 0tn the coset S£ + ex splits into two
ifo-cosets ifo + ex and if i + ex- A lattice which contains ifo but does
not contain S£ may contain at most one of these two if o-cosets. Let
* c 2®n and let p be a {0, l}-valued function on <€. Put

(B-construction in [CS88]).

Lemma 4.1.5 The set ifB (#,/?) is a lattice if and only ifV is a code and
for all X,Y e% we have

P(XAY) = (fi(X) + P(Y) + \XdY\) mod 2.

is a lattice then
B
(i) if (#,j?) is integral if and only if the code <& is contained in its
dual, in which case P is a linear function on <€,
(ii) if 5 (#,/?) is even if and only if^ is doubly even,
(iii) if ifB(#,j?) is integral then disc if*(#,j?) > 4 with the equality
holding if and only if^ is self dual,
(iv) i/if 5 (^,j?) is integral then we can change the signs of some of the
vectors in 0tn so that P becomes the zero function.
Proof. Since
ex + eY = eXAY + y ^ a
aexnY
and the latter sum is contained in ifo if \X n Y \ is even and it is in ifi
otherwise, we obtain the condition for <£B{$, P) to be a lattice. The proofs
of (i) and (ii) are analogous to proofs of (i) and (ii) in (4.1.4). Since ifo
is of index 2 in 5£, its discriminant is 2 n+2 and if <6 c #*, then the order
146 Conway groups
of # is at most 2"/2 and (iii) follows. Let us turn to (iv) and suppose that
ifB (#,/?) is integral which means that # is totally singular with respect
to the parity form and /? is linear. Put # 0 = {X \ X e ^,p(X) = 0}.
If # 0 = <$ then we are done, otherwise there is Y ^ 0tn such that
Y-1 n # = #o- Now it is easy to see that the change of signs of vectors in
Y transfers /? to the zero function. •
By (4.1.5 (iv)), if # ^ #*, then we can (and will) assume that the signs
of vectors in 0tn are chosen in such a way that /? is the zero function and
write £eB(%) instead of J£?5(#, j8), so that

Let M be an even unimodular lattice which contains if o and does not

contain J2\ Since |if/ifo| = I^o/-2H = 2, we have |M/(M n if*)| < 2.
On the other hand disc(M n if*) > 4 by (4.1.5 (iii)). Hence we must
have \M/(M n if*)| = 2 and disc(M n if*) = 4. By (4.1.5 (ii), (iii)) and
in view of the above notational convention the latter equality and the
fact that M is even imply that Mn&* = if B (#), where <€ is a doubly
even code (of length n). Since disc if B (^) = 4, there are three proper
cosets of £?*{<$) in if*(#)*. Since if B (#) <= j£T, clearly if <= if B (#)*.
The vector \e®n (whose coordinates in the basis 0tn are all equal to \) is
contained in S£*§ \ <£* and since ^ is doubly even, (^e@n,ex) e Z for all
X e <€. This shows that the proper cosets of £?*{<£) in &B(<gy are the
ones containing

&u if0 + ^ B , and ifi + ^ n .

Since M is unimodular, it must contain one of these cosets and not if i,
since in that case it would contain the whole of 5£.

Lemma 4.1.6 If x € S£t + \e&n for e = 0 or 1, then

x2 = Q +e\ mod 2.
Proof. It is straightforward to check that (\e&n)2 = | , {\e&n — a)2 =
| + 1 for a e 0tn and (\e®n,x) e 2TL for x € &Q. Since ifi + \e®n =
ifo — Q + \egin9 the result follows. •
For e = | mod 2 put
 4.2 Some automorphisms of lattices 147
(C-construction in [KKM91]). By (4.1.6) and the arguments before that
lemma we have the following.

Lemma 4.1.7 If ^ is a self-dual doubly even code of length n then

is an n-dimensional even unimodular lattice and up to isomorphism every
even unimodular lattice which contains ££§ and does not contain <£ can be
obtained in this way. •
Using (4.1.3 (iii)) it is not difficult to check that for n > 24 <£c{$) does
not contain roots if and only if the minimal weight of ^ is greater than
4. In view of (2.8.8) and (2.11.4) this gives the following.

Lemma 4.1.8 Let ^n be the unique Golay code. Then &c{%>n) is a Leech
lattice and up to isomorphism it is the unique Leech lattice which contains

4.2 Some automorphisms of lattices

Recall that an automorphism of a lattice L c R" is a linear transfor-
mation of R n (i.e. an element of GL n (R)) which stabilizes L as a whole.
In this section we show that ££ and j£?o have the same automorphism
group isomorphic to 2n : Symn and also calculate the stabilizers of <£
in the automorphism groups of the lattices 5£A^€\ J^ B (#) and <£c{$)
(depending on the code c€).
Let 0tn, <£ and i?o be as in the previous section. It is easy to see that
& = {±a \ae@n}
is the set of roots in J£? and such a subset in Rw will be called a frame.
Let M be an integral lattice of dimension n and !F be a frame. Then
M is said to be of type A, B or C with respect to IF if the following
conditions hold:
type 4 : i f c M,
type B: $£ n M = jg?0 and M c jjf,
type C: <£ n M = ^0 and M D JST ^ M.
The following proposition is a reformulation of (4.1.4), (4.1.5) and
(4.1.7).

Proposition 4.2.1 Let M be an integral lattice in R n which is of type A, B

or C with respect to a frame # \ Suppose also that in the type C case M
148 Conway groups
is even and unimodular. Let

Then %> is a code and possibly after changing the signs of some vectors in
9tnt we have M = &*(<£), J ^ ( # ) or i ? c ( # ) , respectively. U

Lemma 4.2.2 Let D = Aut if. Then D^2®" : Sym(@n) ^ 2" : Symn.
Proof. Let 3F be the frame formed by the roots in ££. Since 3F contains
the basis 0tn of R n , the action of D on & is faithful. Let us say that two
roots in ?F are equivalent if they are scalar multiples of each other. The
group D preserves this equivalence relation (with n classes of size 2 each)
and induces on the set !F of equivalence classes a subgroup of Sym(&).
Furthermore, the kernel of the action of D on !F is an elementary abelian
2-group of rank at most n. On the other hand each permutation of 0tn
can be extended to a linear transformation of R n (which stabilizes if)
and for a subset Y £ 0tn the mapping which sends a € 01n to y(Y,a) • a,
where y(Y,a) — — 1 if a e Y and y(Y,a) = 1 otherwise, defines a linear
transformation of R" which stabilizes every equivalence class in 3F as a
whole. •

Lemma 4.2.3 The lattices ££ and ifo have the same group of automor-
phisms.

Proof. It follows from the proof of (4.2.2) that every automorphism

from D = Autif stabilizes ifo- So in order to prove the equality it is
sufficient to show that there is a canonical (i.e. basis independent) way
to reconstruct if from if o. Let A be the set of vectors which are sums of
pairs of non-collinear vectors from 3F. It is easy to check that A is in fact
the set of all vectors of length 4 in ifo. Define on A a graph (denoted
by the same letter A) in which two vectors are adjacent if they are equal,
coUinear or orthogonal. Let us say that two vectors from A are equivalent
if they have the same support in the basis 0tn. This equivalence relation
can be described in the internal terms of ifo as follows: two vectors are
equivalent if and only if in the graph A they are adjacent to the same
set of vectors. Now it is sufficient to observe that the roots from 3F are
halves of sums of pairs of equivalent but not coUinear vectors from A.D
We would like to describe the stabilizers of S£ in the automorphism
groups of the lattices SeA(^\ ^B(^) and ^c{^) in terms of the code #
and its automorphism group.
 4.2 Some automorphisms of lattices 149
n
Let V = V = R be two n-dimensional Euclidean vector spaces, 0tn
and 0t'n be bases in V and V consisting of pairwise orthogonal roots
and let i f = &(9tn) and <£' = JSf (^ n ) be the lattices formed by integral
linear combinations of vectors from 0tn and @l'n, respectively. By (4.2.2)
the linear transformations of V onto V which map i f onto S£' are
parametrized by the pairs (Y,o") where Y ^ 0tn and c is a bijection of
^ n onto 0t'n. If T(Y,(T) is the transformation corresponding to such a pair
and a € 0tn, then

where y(Y,a) = - 1 if a G Y and y(Y,a) = 1 if a $ Y.

Lemma 4.2.4 Let %> be a code based on Mn and <€' be a code based on
0t'n. Let M = <eA{%) and Mr = J^(<T) be the lattices obtained by A-
construction using <$ and %>', respectively. Then
(i) T(Y,CT) maps M onto Mf if and only if a maps %1 onto W',
(ii) Aut &A(<g) n Aut i f s 2®» : Aut <€.

Proof. Let X e <€ so that S£ + ex<^M. Then

=<?' + eG{x) - Yl "! = &' + effiX)

a'eo{Y)C\(j{X)

and this coset belongs to Mr if and only if G(X) G %>'. So (i) follows and
immediately implies (ii). •

Lemma 4.2.5 Let <6 be a code based on 0tn and <€' be a code based on
0l'n. Let M = <£*($) and Mr = ^B{^) be the lattices obtained by B-
construction using *$ and %>', respectively. Then
(i) maps M onto Mr if and only if a maps % onto <€' and
T(Y,CT)
Y e%*,
(ii) Aut <£B{$) n Aut S£ ^ T : Aut <€.

Proof. Let X e % so that if 0 + ex £ M. Then

a'eo{Y)C\o{X)

and this coset belongs to W if and only if G(X) G <€' and \<j(Y)n<r(X)\ =
\Y C\X\ is even. The latter condition holds for all X G ^ if and only if
Y G T. So (i) follows and implies (ii). •
150 Conway groups

Lemma 4.2.6 Let ^ be a doubly even self-dual code based on 0tn and <€'
be a code based on ®fn. Let M = S£c{(€) and W = &c(<$f) be the lattices
obtained by C-construction using <& and <&', respectively. Then
(i) maps M onto M' if and only if
T(Y,(T) G maps <& onto <€' and
Y e V = *,
(ii) Aut S£ci$) n Aut SB ss <g : Aut <e.

Proof. Let JV = Mn&* = S£B($) and JV' = M'n£>'* = S£BW). Since

T ( 7 , ( T ) maps JS? onto jSf', it maps j£f onto if'*. Hence if x(Y,a) maps M
onto M', it maps JV onto JV'. By (4.2.5) the latter happens exactly when
a maps * onto «" and 7 G #* (in our case ^* = #). Assuming that
these conditions hold let us show that T(Y,<T) maps M onto Mf. Since ^
is doubly even and self-dual, by the argument before (4.1.6) we conclude
that M \ N is the only coset of JV in N* which contains even vectors only
and does not contain i f i. Similarly Mf \ Nf is the only coset of JV' in JV'*
which contains even vectors only and does not contain £P'V Since T(Y,<T)
maps <ifi onto S£\ and preserves the lengths of vectors, it indeed sends
M onto M' and (i) follows. Now (ii) is immediate with the remark that
<$* = <$ because of the self-duality assumption. •

Since the Golay code #12 has no subsets of size 4, all roots in ^
are contained in the frame # \ Hence !F is the only frame for which
S£\Cn) is of type A and by (4.2.4) Aut J ^ ( # i 2 ) = 2 24 : Mat24.
The stabilizer in Aut if c (#12) of the frame 3F is isomorphic to 2 12 :
Mat24 by (4.2.6) and we will see in the next section that it is a proper
subgroup in Aut if c (#12).
In view of the above discussion and by (4.2.6) we have the following.

Proposition 4.2.7 Let A be a Leech lattice and ^ be a frame. Then A

cannot be of type A or B with respect to 3F and if A is of type C then
A = i f c ( # i 2 , ) / o r a basis $24 consisting of roots from # \ Furthermore, if
3F1 is another frame for which A is of type C then the automorphism group
of A contains an element which maps 3F onto $*'. •

4.3 The uniqueness of the Leech lattice

In this section we follow the brilliant article [Con69] by J.H. Conway to
show that the Leech lattice is unique up to isomorphism.
Let A be a Leech lattice. It can be deduced from the general theory of
integral lattices that for every r the number JVr of vectors of length r in
 4.3 The uniqueness of the Leech lattice 151
A is the same for all Leech lattices and can be computed explicitly. The
situation is the following.
Let L be an integral lattice. The theta function ©L of L is a power
series in a formal variable q defined as follows:

xeL r=0

where Nr is the number of vectors of length r in L.

The following fundamental result is known as Hecke's theorem ([Ser73],
[Ebe94]).

Theorem 4.3.1 Let L be an even unimodular lattice of dimension n.Ifq in

the above expression for ©L(#) is replaved by e2niz where z is a variable
taking values in the complex upper half plane then ®L(Z) is a modular form
of weight n/2, which means that

^ ) for

Thus in the case of a Leech lattice A the theta function ®A(Z) is a

modular form of weight 12. It is known [Ser73] that the space of modular
forms of weight 12 is 2-dimensional. On the other hand ®\(z) satisfies
two additional conditions: No — 1 (true for all lattices) and N2 = 0 (since
there are no roots in A). These two conditions turn out to be independent
and they are satisfied by a unique modular form of weight 12. Hence
this unique form is the theta function of a Leech lattice. The coefficient
of q2m in the power series expansion of this form is
65 520
( ( ) ())
where <7n(m) is the sum of 11th powers of the divisors of m and x(m) is
the Ramanujan function defined by

m=l m=l

Let Am denote the set of vectors of length 2m in A. Then |Am| = N2m and
by the above for a given m we can calculate the size of Am explicitly. In
particular we have the following.

Proposition 4.3.2 Let Abe a Leech lattice and let Am be the set of vectors
152 Conway groups

of length 2m in A. Then |A 0| = 1, |Ai| = 0,

|A2| = 196560, |A 3 | = 16773 120 and |A4| = 398034000.

Let A = A/2A be the Leech lattice A taken modulo 2, so that A is an

elementary abelian 2-group of rank 24. If X e A and M ^ A then X and
M denote the images in A of X and M, respectively.

One may notice a similarity between (2.3.1) and the following lemma.

Lemma 4.3.3 Let X and v be distinct vectors in A with X = v such that

X e A,-, v € A; for 0 < i,j < 4. Then either X = —v, or i — j = 4 and
(X,v) = 0.

Proof. Clearly X and —X have the same image in A, so we assume

that X ^ — v. Replacing X by —A, if necessary we can assume that (X, v) is
non-negative. Since both X and v have length at most 8 we have

(X-v)2 = X2 -2(X,v) + v2 < 16

with the equality holding if and only if X and v are orthogonal vectors
from A4. Since X = v and X ^ —v, there is a non-zero vector /a in A such
that X — v = 2fi. Since A is a Leech lattice the length of \i is at least 4
and hence

(X - v)2 = 4/i2 > 16.

So (X - v)2 = 16 and the result follows. •

A maximal set of pairwise orthogonal 1-dimensional subspaces in a 24-
dimensional Euclidean space is obviously of size 24 and each 1-subspace
contains exactly two vectors of any given positive length (in particular
of length 8). In view of this observation (4.3.3) implies the following.

Lemma 4.3.4 / / 0 < i < j < 4 then A, n A7- = 0, |A,-| = £|A,-| and
|A4| > ^ | A 4 | . •

By direct calculation with numbers in (4.3.2) one can easily check the
following equality:

t , |A2l , |A3l , |A4l = ?24

+ +
2 "*" 2 48
Since the right hand side is exactly the order of A we have the following.
 4.4 Coordinates for Leech vectors 153
Lemma 4.3.5 A is the disjoint union of the A,/or i = 0,2,3 and 4; IA4I =
^ IA41, which means that for every k G A4 there is a unique R 24 -basis Q)
consisting of pairwise orthogonal vectors from A4 such that k G {+a | a G
<$} and whenever a and /} are distinct vectors from 3f then a + /? = l\i for
some fi G A2, in particular \$)\ = \. •

Let k G A4 and let 2 be as in the above lemma. Put ^24 = {a\2ae @}

and <F = {±a | a G ^24}. Then ^ 2 4 is a basis of R 2 4 consisting of
pairwise orthogonal roots and by (4.3.5) the sum of any two (possibly
equal) vectors from & is contained in A. This immediately implies that
£ A and by (4.2.7) we obtain the following.

Proposition 4.3.6 Let A be a Leech lattice. Then

(i) A ss <£c(<€n\
(ii) there is a one-to-one correspondence between the set A4 and the set
of frames for which A is of type C,
(iii) AutA acts transitively on the set A4 of size |A 4 |/48 = 8292 375 =
3 6 • 5 3 • 7 • 13 with stabilizer isomorphic to 2 12 : Mat24,
(iv) |Aut A| = 2 22 • 3 9 • 5 4 • 72 • 11 • 13 • 23. •

4.4 Coordinates for Leech vectors

In the remainder of the volume A is the unique Leech lattice in R 24 ,
the vectors in A will be called Leech vectors. In order to carry out more
or less explicit calculations in A it is convenient to choose a basis 9 in
R 2 4 such that A = J^ c (^i 2 ) with respect to ^ 2 4 = {4a \ a e 0>). In this
case the coordinates of a Leech vector are integral. Moreover, a vector
k G R 2 4 whose coordinates {k(a) | a G £?} in the basis 9 are integral is a
Leech vector if and only if for m = 0 or 1 the following three conditions
hold (we assume that ^12 is based on £?):
(Al) k(a) = m mod 2 for every a G ^ ,
(A2) {a I k(a) = m mod 4} G « 1 2 ,
4
(A3) £ a e ^ ( < 0 =

Here m = 0 if k G ^(^24)* and m = 1 otherwise.

Notice that if k9 v e A and k G A* then

a n d
'•=^
154 Conway groups
In this chapter when talking about a Golay code and a Steiner system
we always mean the code #12 based on 0> and the system formed by the
octads in #12, respectively.

We write G for the automorphism group of A (also denoted by .0

and Co0). Let G\ — G n Aut <£ (#24). By (4.2.6) G\ consists of the
transformations induced by the mappings

T(Y,(T) :a^y{Y,a)'G(a) for a e ^ ,

where a G Aut #12, 7 G * i 2 = ^ 2 , y(7,a) = - 1 if a € 7 andy(Y,a) = 1

otherwise. Thus G\ is the semidirect product of

and

Li = {r(0,a) | a G Aut #12} = Mat2*.

with respect to the natural action.

We are going to describe the orbits of G\ on A2, A3 and A4. For this
purpose we represent the coordinates of a Leech vector X in the basis
9 by a triple (N(X),P(X),X(X)) where N(X) is the multiset of absolute
values of coordinates of X, P(X) is an ordered partition (Pni,Pn2,...,Pni)
of 9 such that a G Pnj if and only if \X(a)\ = nj for 1 < i < j < I, and
finally X(X) is a subset of 9 such that X(a) is negative if and only if
a G X(X). It is clear that the coordinates of X (and hence X itself) are
uniquely determined by the triple (N(X),P(X),X(X)). The multiset N(X) is
called the shape of /I. Notice that if X and v are in the same G\-orbit,
then they have the same shape; if they are in the same Q\-orbit, then
P(X) = P(v) as well. In the case of short vectors it often happens that the
shape of X uniquely determines the G\ -orbit containing X. It is common
to denote by A" the vectors in At for which n is the maximum of the
absolute values of coordinates. If X G A" then i and n are determined
by the shape of X, which means that A" is a union of G\ -orbits. If there
more than one G\-orbit in A?, we denote these orbits by A^a,Af,....

Lemma 4.4.1 The orbits of G\ = 2 12 : Mat2A on A2, A3 and A4, the shapes
of vectors they consist of their lengths and the corresponding stabilizers
are as given in the table below.
 4.4 Coordinates for Leech vectors 155

A, Orbits Shapes Lengths Stabilizers

A2 A; (42022) (-)•* [210] : Aut Mat22

(31 23 ) 24 • 2 12 Mat23
to to

7
8 16
(2 0 ) 759 • 2 [2 ] : (24 : L4(2))
5

A3 Ai (5 I 23 ) 24 • 2 12 Mat23
8 15 8 4
Af (42 0 ) 759 • 16 • 2 [2 ] • L 4 (2)

A3 (33121) ( 2 3 4 )-2 1 2 iTL 3 (4)

A§ (212012) 2576 • 2 11 [2] x Matx2

A4 AS (80 23 ) 24-2 [211] : Mat-n

At (62 7 0 16 ) 759 • 8 • 2 7 [25] : (24 : Alt-,)

Al (5 32121) C 4 )-3-2' 2 P2L 3 (4)

Af (44020) (?)•* [28] : (26 : 3 : Sym5)

Af (4228014) 759-(!26)-29 [23] : (2 x 2 3 : L3(2))

Af (42 1 2 0 n ) 2576 • 12 • 2 12 Mat 11

A3 (35119) (24)-212 24 : (Symi x SynisY

Af (21608) 759 • 2 11 [2] : 2 4 : L4(2)

Af (21608) 759-15-2 1 1 [2].2],.+6 : L3(2)

Proof. Given i e {2,3,4} we first determine the possible shapes of

vectors in A,-. If X G A, and N(X) = {n\lnk^...n\l) is the shape of A, then
fci + k2 +... + ki = 24 and kin\ + k2n\ +... + kin} = 16 • I
By (Al) the numbers n ; have the same parity and in the even case by (A2)
the sum s of the kj with n ; = 2 mod 4 is the size of a Golay subset (i.e.
a subset from #12), that is s € {0,8,12,16,24}. Having these conditions
156 Conway groups
it is not difficult to list the possible shapes as in the third column of the
table. We will see below that for every shape in the table the signs can
be chosen so that (A3) is satisfied.
If N(X) is as above then P(X) = {Pni,Pn2,...,Pni) and \Pni\ = kt for
1 < i < I. We claim that for every shape in the table Gi/Qi ^ A u t # i 2 =
MatiA acts transitively on the set of corresponding ordered partitions.
In fact for shapes (42022), (3 I 23 ), (5 I 23 ), (33121), (80 23 ) ? (5 32121), (44020)
and (35119) this follows from the 5-fold transitivity of Mat24 on 0> (2.9.1
(iii)), for shapes (28016), (42 8 0 15 ), (62 7 0 16 ), (4228014) and (21608) from the
transitivity of Mat24 in the set of octads and the double transitivity of
the stabilizer of an octad B on B and on 0* \ B (2.10.1) and finally for
shapes (212012) and (42 1 2 0 n ) from the transitivity of Mat24 on the set of
dodecads and the (5-fold) transitivity of the stabilizer of a dodecad D on
D a n d o n ^ \ D (2.11.7).
Now it remains to analyse the possibilities for the signs of coordinates.
For a multiset N from the table and an ordered partition P corresponding
to N put

O = <&(N,P) = {X\Xe A,N(X) = N,P(X) = P}.

We consider even and odd cases separately, starting with the latter
one. Thus assume first that all the integers in N are odd (so that there
are five possibilities for N from the table). Let /io be a vector such that
N(fio) = N, P(/io) = P and /io(a) = 3 mod 4 for all a € SP. Then one
easily checks that in each of the five cases we have J2ae& ^(a) — 4 mod 8,
which means that /io is a Leech vector and hence /ioGO. Since /io has
been chosen so that |/io(fl)| = \iM)(b)\ implies juo(a) = /M)(b), the stabilizer
of /io in G\ is contained in the complement L\. The isomorphism type
of this stabilizer as in the last column of the table follows directly from
(2.9.1), the definition of Mat23 and (2.10.1). For an arbitrary vector X e Q>
put Y = X(fio) AX(A). Then X can be obtained from /i 0 by changing signs
in the coordinates in Y, i.e. X = fiQYtl\ On the other hand Y is the set
of coordinates of X equal to 1 modulo 4 and by (A2) we have Y £ ^n.
Hence ?(Y,1) G Qu which shows that Q\ acts regularly on O and the
analysis of the odd case is completed.
Let us turn to the even case which is slightly more delicate. Let N be an
even multiset from the table, P = (Pg? ^6> P4, P2, PQ) be the corresponding
partition (where some of the P ; can be empty) and let O = O(iV,P)
be as above. Notice that by (A2) P^ U P2 is a Golay set. Let /io be a
vector such that N(fio) = N, P(/io) = P and no(a) < 0 if and only if
 4.4 Coordinates for Leech vectors 157
8 15 12 n
a e P6- Then juo is a Leech vector unless N = (42 0 ) or (42 0 ). We
postpone the analysis of these two cases and assume that JJQ is a Leech
vector. Then /io is stabilized by ?(7,(r) € G\ if and only if Y ^ p0 and a
stabilizes the partition P. Thus if no is stabilized by such a T(Y,<7) then
it is stabilized by both ?(Y, 1) e Qi and ?(0,<T) G L\. Hence G\(fio) is the
semidirect product of Qi(^o) (consisting of the Golay subsets contained
in P o ) and Li(/*o) which is the stabilizer of P in Li. The structure
of Gi(jUo) is given in the last column of the corresponding row in the
table in the form [Qi(fio)] : £i(/4))- It is not difficult to deduce these
structures from the properties of Mat24 and %>n contained in Chapter 2.
A vector X = $Y'l) belongs to O if and only if \Y n (P6 U P2)\ is even.
This enables us to calculate the size of O. Comparing this size with
[G\ : Gi(jUo)] we conclude that Q\ : Li(/io) is transitive on <D in all cases
under consideration except for the case JV = (21608). In the latter case
the orbit h\a containing fio does not contain all the Leech vectors of
this shape. This is seen by the following argument. Let X e O((2 16 0 8 ),P)
and \X(X)\ = 2. Then fio cannot be mapped onto X by an element from
G\. In fact, Pi is the complement of an octad B and we know ((2.8.5)
and (2.11.4)) that a Golay set cannot intersect it in two elements. Let us
calculate the stabilizer in G\ of such a vector L Clearly Qi(/l) is of order 2
and?(£, 1) is its only non-identity element. Let c be the unique involution
in Li which stabilizes B elementwise and X(k) setwise (compare (2.10.1
(ii))). We claim that Gi(/l)gi = C-(c)gi. In fact, if T(Y,<T) stabilizes k
then either a(X{X)) = X(X) and Y k Po or <J(X(X)) n X(k) = 0 and Y is
an octad such that Y nP2 = X(X)Uc(X(X)). Notice that by (2.10.5) for a
2-element subset Z e P2\ X(X) an octad which intersects P2 in X(X) U Z
exists if and only if Z is an orbit of c. Since c is uniquely determined by
any of its orbits on P2, the claim follows. Now direct calculations show
that Aja U A f contains all Leech vectors of the shape (21608).
Let us turn to the pair of multisets left out before. If N(X) = (42 1 2 0 n )
and X(X) consists of a single element from P2, then X is a Leech vector.
We calculate the stabilizer Gi(X). By (2.11.2) Qi(X) = 1. We claim that
G\{X)Q\ = L\(P)Q\. In fact L\(X) is the stabilizer in L\ of the partition
(P4,X(X\P2 \X(1),PO) (this stabilizer is isomorphic to L 2 (ll) by (2.11.9
(iii))). Let B be an octad such that \B n P2\ = 2, X(X) c B, P4 n J5 = 0
(such an octad exists by (2.15.1)) and let a be an element which stabilizes
(P4,P2,Po) and maps X(X) onto 5\(Bn(P 0 UX(A)). Then X{B,G) stabilizes
X, which means that G\{X)Q\ acts transitively on {±a \ a G P 2} and the
claim follows. By (2.11.7) G{(X) ^
158 Conway groups
8 15
Finally let N(X) = (42 0 ) and suppose that X(X) consists of a single
element from P2. Then X is a Leech vector. Let us calculate its stabilizer in
G\. There are exactly 15 octads disjoint from P4UP2 and hence Q\{1) = 2 4 .
Clearly L\(k) is the stabilizer in L\ of the partition {P^X(X\P2\X{X\Po)
which is isomorphic to Alti by (2.10.1). Arguing as in the previous
paragraph we show that G\(X) acts transitively on {±a \ a G P2}, which
implies that G\{k) ~ 24.L^(2). A more detailed analysis shows that the
extension does not split. •

4.5 Cou C02 and C03

Our nearest goal is to show that G acts transitively on A2 and A3. Notice
that Leech vectors with all their coordinates in the basis 9 divisible by 4
form the sublattice i f 0(^24) and those with all their coordinates divisible
by 2 form the sublattice A n ^(^24)* • We should emphasize that 9 is a
basis of R 2 4 and not a basis of the lattice A. By (4.2.6) and (4.3.6) G{
is the intersection of G and Aut <£($24)- In view of (4.2.3) this means
that G\ is the stabilizer in G of the sublattice J£?o(^24) and also of the
sublattice A n J£?(^24)*- Since G\ is a proper subgroup in G, by (4.3.6 (iii))
these two sublattices are not stable under G and we have the following.

Lemma 4.5.1 Let Q = J^o(^24) or Q = A n ^(^24)*• Suppose that M is

an orbit of G\ on the set of Leech vectors such that M c Q and Q = {mv |
m e Z,v G M}. Then the orbit of M under G contains a vector outside
Q. D

Lemma 4.5.2 G acts transitively on A2.

Proof. By (4.4.1) A£ A^ and Aj are the orbits of Gi on A2. The vectors

in A4 are the shortest vectors in i?o(^24) and hence A4 generates
over the integers. The orbit Aj is contained in (A n J&?(#24)*)
Since the Golay code is generated by its octads as a GF(2)-space, A%
generates A Pi ^(^24)* over the integers. Now the result is immediate
from (4.5.1). •

Lemma 4.5.3 G acts transitively on A3.

Proof. By (4.4.1) the orbits of Gi on A 3 are A|, Af, A^ and A\. The
second and the last of the orbits are contained in A n ^(^24)* • Since the
Golay code is generated by its octads as well as by its dodecads as a
GF(2)-space, each of A^ and A^ generates A n &(^24)* over the integers.
 4.5 Coi, Co2 and Co3 159
Suppose that the action of G on A3 is not transitive. Then in view of
the above and by (4.5.1) the orbit of A3 under G is one of the following
four sets: A^ U A^; A^ U A^; Aj U A^ U A^; and A^uA^U A\. But the size
of each of the four sets is divisible by a prime number which does not
divide the order of G given in (4.3.6 (iv)). Hence we get the result. •
The set A® consists of a single element which we will denote by Io- By
(4.3.5) Gi is the stabilizer of Io in G. Let K be the kernel of the action
of G on A = A/2A.

Lemma 4.5.4 K = (r(^, 1)) is of order 2 and G does not split over K.

Proof. Since G\ = G(Io), clearly K < G\. Since ? ( ^ , 1) multiplies every

(Leech) vector by minus 1, it is contained in K. Every normal subgroup
in G\ which properly contains (r(^, 1)) contains the whole of Q\. We
claim that Q\ is not contained in K. Let a,b G 0>, a =^= b, and v be such
that v(a) = v(b) = 4 and v(c) = 0 for c e 0> \ {a,b}. Then (A1)-(A3) are
satisfied and hence v is a Leech vector. It is clear that there is a subset
X € ^12 which contains a and does not contain b. Since

( v v )

is not a Leech vector, the claim follows. By (2.15.1) #12 is indecomposable

under MatiA and hence G does not split over K. •
The action induced by G on A is the first Conway sporadic simple
group denoted by Co\. By (4.3.6 (iv)) and (4.5.4) we have

|Coi| = 2 21 • 3 9 • 5 4 • 7 2 • 11 • 13 • 23.

The stabilizers in G of vectors from A2 and A3 are the second and

the third Conway sporadic simple groups denoted by C02 and C03,
respectively. By (4.5.2) and (4.5.3) we have Cot = |G|/|A,-|. Hence (4.3.2)
and (4.3.6 (iv)) give

|Co 2 | = 2 1 8 • 3 6 • 5 3 - 7 - 1 1 -23,

|Co3|=210-37-53-7-ll-23.

For the remainder of the chapter G will denote the first Conway group
Co\ isomorphic to the action induced by G on A. The image G\ of G\ in
G is the semidirect product of Qi = <2i/(?(^, 1)) which is the irreducible
11-dimensional Golay code module and the (bijective) image L\ of L\
in G. The elements ?(7,o-) and ? ( ^ \ Y,<r) have the same image in Q\
160 Conway groups
and we denote this image by T(Z,GT) where Z is either Y or &\ Y. Since
neither C02 nor C03 contains ? ( ^ , 1), they map isomorphically onto their
images in Co\ and will be identified with these images.
Directly from (4.3.5), (4.3.6 (iii)), (4.5.2) and (4.5.3) we obtain the
following.

Lemma 4.5.5 The group G = Co\ acting on A # has three orbits A2, A3
and A4 with lengths 98280, 8 386 560 and 8 292375 and stabilizers Co2,
Co-x, and G\ = 2n.Mat24, respectively. •

Let 9 be a mapping of A onto GF{2) such that

X2 = 2 • 9{X) mod 4

for X G A. Then 0 induces on A a quadratic form (denoted by the same

letter 0), which is clearly preserved by G, here

0(1) = 1 if I e A3 and 0(1) = 0 otherwise.

Let /? denote the bilinear form on A associated with 9:

For a quadratic form on a 24-dimensional GF(2)-space the numbers of

isotropic and non-isotropic vectors are known [Tay92]; comparing these
numbers with the numbers in (4.5.5) we have

Lemma 4.5.6 The form 9 is the only non-trivial quadratic form on A pre-
served by G. •

4.6 The action of Co\ on A4

In this section we study the action of G = Co\ on the set A4. By (4.3.6
(iii)) this action is transitive and G\ is the stabilizer of the element
Io € A4 such that A | = {Io}- Thus the action under consideration is
of G on the cosets of G\ or equivalently of G on the cosets of G\. For
every element v G A4 the 48 vectors in A4 which map onto v under the
natural homomorphism xp : A —• A belong to 24 pairwise orthogonal
lines (1-dimensional subspaces) in R 2 4 . Furthermore, Q(v) := 6>2(G(v))
is the image in G of the kernel of the action of G(v) on these 24 lines.
Notice that Qi = <2(Io)-
 4.6 The action of Co\ on A4 161
Lemma 4.6.1 The group G\ acting on A4 has six orbits A4 for a =
8,6,5,4a, 4b and 4c. The preimages of these orbits in A4, their lengths
and element stabilizers are as given in the table below.

Orbits Preimages Lengths Stabilizers

A! 1 [211] : Mat24
1
A! A 6 -1- * A 2 a 2 6 • 759 [25] : (24 : L4(2))

A3 2" • (?) iTL 3 (4)

AJ- i_A4a 2•1771 [210] : (26 : 3 • Sym6)

/Y
48 4
K? AAf + ^Af 2 7 • 15 • 759 [24].2i+6 : L3(2)

2 11 • 1288 M«, 12 .2

/n the second column corresponding to an orbit N by writing ^M + jL

we mean that v G N is the image of m vectors from the orbit M of G\ on
A4 and of I vectors from the orbit L.
Proof. If A4 is an orbit of G\ on A4 then A4 (which is the image of
A4 in A) is an orbit of G\ on A4. Furthermore, for V G A J the set

is an imprimitivity block of G\ on A4 and G\(v) is the image in G of

the setwise stabilizer of \p~l(v) n A4 in G\. We say that two vectors in
A4 are equivalent if they have the same image in A. We know that each
equivalence class consists of 48 vectors and that every vector is equivalent
to its negative.
We adopt the following strategy of the proof. For every a G {8,6,5,
4a, 4b, 4c} we choose a representative v e A J and find 48 vectors in A4
equivalent to v. This will show in particular that every vector fi e A4 for
p € {3,2a, 2b} is equivalent to a vector from A4 with a as above. As in
the proof of (4.4.1) we represent the coordinates of a Leech vector X by
the triple (N(A),P(A),X(A)).
For a = 8 everything is clear. Let v G A4 so that i ^ v ) U Piiv) is an
octad B of #12 and let X(v) = P6(v). If \i G K% p ± v, X(ji) = P 6(fi)
162 Conway groups

and P6(fi) U P2(fi) = B then X := ±(v + /*) G A\ with P 2(A) = B and

X(A) = P6(v) U P(>{ii). Since every vector is equivalent to its negative this
gives 16 vectors in A\ equivalent to v. Let co be a vector from A\a such
that P2(co) = 0> \ B and X(co) is a Golay set disjoint from B. Then
X := I( V + co) e A32 with P3(A) = P6(v) and X(^) = X(co). Since there
are 32 possibilities for X(co) (including the empty set), we obtain 32
vectors in A^a equivalent to v. It is now easy to see that the stabilizer
of ip^iv) H A4 in G\ is the semidirect product of the stabilizer in Q\ of
the pair {v,-v} (of order 26) and L\[B] = 2 4 : L4(2) which gives the
structure of Gi(v) as in the table. Notice that gi(v) = [QuO2(Li[B])]
(3.8.4) and that T(£, 1) = T(0> \ B, 1) is the only non-trivial element in
Qi n Q(v).
Let v G A4 and v(a) = 3 mod 4 for every a £ ^ . If /i G A4, fi =fc v,
H{a) = 3 mod 4 for all a G ^ and P5(/z) U P 3 (M) = Ps(v) U P3(v) then
j(v — //) G A4 and in this way we obtain 6 vectors in A | equivalent to v.
If \x G Aj with /i(a) = 3 mod 4 for all a e ^ and P3(v) U P5(v) U P3()u)
is an octad then ^(v — fi) e A\. Since there are 21 octads containing a
given 3-element subset of ^ , we obtain 42 vectors in A\ equivalent to v.
The stabilizer of xp~l(v)C\ A\ in G\ is contained in L\ and coincides with
the stabilizer of P5(v) U P3(v), isomorphic to PI"X3(4).
Let v G A\a, X(v) = 0 and Z = {Si = P 4 (v),S 2 ,...,S 6 } b e t h e s e x t e t
containing P4(v). If \i e Af1 with P4(ju) = St for some i, 1 < i < 6,
and |X(/i)| even, then ^(v + fi) is contained in A\ if 1 = 1 and in
A4 if i =fc 1. Thus all the 48 vectors equivalent to v are in A\a. The
stabilizer of \p~l{v)C\A\a in G\ is the semidirect product of the subgroup
{T(Y, 1) I \Y n P 4 (v)| is even} of index 2 in Qx and the stabilizer in Li
of the sextet E. This implies the structure of Gi(v) as given in the table.
Notice that Qi(v) = [QuO2(Li[l])] and that
(6i n 6(v)) # = {T(7 , l) 1 y = st u s,-, 1 < i < j < 6}.
Let v G A ^ so that B = P2(v)is an octad and let X(v) be of size 1 and
contained in P4(v). Let c be the unique involution in L\(B) = 2 4 which
stabilizes P4(v) as a whole and let R\ = P 4 (v),# 2 ,..-,#8 be the orbits of c
on 0> \ B. If pi is a vector from A f such that P2(/x) = B, P4(fi) = P4(v)
and X(fi) = P4(v) \ X(v) then i(v - ^) G A^. In addition if P2(n) = B, for
2 < i < 8 we have P4(fi) = Ru \X(ti)C\P4(n)\ = 1 and £iUK,U(X(/i)nP2Gu))
is an octad, then ^(v +fi) e A% and altogether we obtain 32 vectors in Af
equivalent to v. Let co be the vector from Af such that P2(a>) = 9 \ B
and X(co) = P4(v). Then X := \{y + co) G A^ with P3(X) = X(v). Similar
results will be achieved if instead of co we consider its image under ?(7,1)
 4.7 The Leech graph 163
where Y is one of the 7 octads disjoint from P2(v) U P4(v). Including the
negatives this gives 16 vectors in A^ equivalent to v. Comparing this
with the information in the table from (4.4.1) we conclude that Qi(v) is
of order 24, Qt n Q(v) = 1 and Gi(v)Qi/Qi = CLl(c)Qi/Qi.
Finally let v € Af so that D = P2(v) is a dodecad and suppose that
\X(v)\ = 2 with P4(v) <= JSf(v). Recall that by (2.11.3) for every 2-element
subset T in 0>\D there is a unique partition {E1(T),E2(T)} of D such
that Ei(T)U T and £ 2 (T)U T are both octads. Let n be a vector from A^c
such that P2Qi) = P2(v), P4(/i) ^ P4(v), P 4 00 s * 0 0 and X(ji) n P2(^) =
X(v)A£,(P 4 (v) U P2(ju)) f o r i = 1 or 2. Then A := ±(v - n) e A\ with
P2{X) = £,(P4(v)UP4(/i))UP4(v)UP4(ju). Thus (including v and its negative)
we obtain 24 vectors \i in Af with Pi{ix) = P2OO equivalent to v. Now
let w b e a vector from Af such that P2(co) = 0>\ P2(v), X(co) = P4(v)
and P4(co) = X(v) n P2(v). Then /I := \{v + a>)eA32 with P3(A) = P4(v)
and X(a) = 1 mod 4 for all a e 0*. In this way we obtain the remaining
24 vectors from A^c equivalent to v. It is clear that Q\(v) = 1 and that
Gi(v)<2i/gi < Sgi/<2i where 5 = Matn.2 is the stabilizer in L\ of the
partition {D,0>\D}. On the other hand Gi(v) contains Gi(v) = Matn
with index 48 which implies that Gi(v) = S. •
We summarize (4.6.1) and its proof in the following.

Lemma 4.6.2 For cc = 4a, 6, 4b, 4c and 5, respectively, the following two
assertions hold:
(i) QQQ) = Qx acts on AJ wit/i orbits of length 2, 2 6 , 2 7 , 2 11 and 2 11 ,
(ii) G1/Q1 = Mat 24 acts on the set of Qi-orbits in A4 as it acts on
the set of sextets, octads, elements of type 3 in ^(Mat 2 4 ), 3-element
subsets of & and complementary pairs of dodecads.
Ifv€ Af then Q(^o)(^Q(v) is of order 2 4 with non-identity elements being
T(B, l)for the octads B refined by the sextet corresponding to the Q\-orbit
ofv;ifv£ AJj then Q(Io) n Q(v) is of order 2 containing T(B, 1) where B
is the octad corresponding to the Q\-orbit of v; Q(lo) n Q(v) is trivial in
the remaining cases. •

4.7 The Leech graph

For a € {8,6,5,4a,4b, 4c} and v € A4 let AJ(v) denote the image of
A4 under an element g e G such that 11 = v. Since AJ is an orbit of
Gi = G(Io)> this definition is independent of the particular choice of
g with the above property. In this section we study a graph T on A4
164 Conway groups
(called the Leech graph) such that F(v) = A ^ v ) . We will use this graph
to construct a rank 4 tilde geometry &(Coi) associated with the Conway
group Co\. In fact T is the collinearity graph of ^(Co\). We are going
to sketch the calculation of the suborbit diagram of the Leech graph T
based at Io given below.

Since a Sylow 2-subgroup in G\ has order 2 21 , directly from (4.6.1) we

obtain the following.

Lemma 4.7.1 If v,ju e A4 with v ^ Ji and \G(v) D G(p)\ is divisible by 2 16

then v and p, are adjacent in T. •
In view of (4.6.2) there is a mapping Q of K\a onto the set of sextets
which commutes with the action of G\. Recall that the sextet graph
defined in (3.3.6) is the collinearity graph of

Lemma 4.7.2 Let v,ju e K\a with v =£ ji. Then

(i) Gi(v) acting on A%*\{v} has 4 orbits Oi, <E>2, $3 and O4 with lengths
1, 180, 480 and 2880, respectively,
(ii) ju is adjacent to v in Y if and only if p, G <I>i U O2,
(iii) the orbit of Ji under Q(v) = O2(G(v)) has length 2 6 if Ji € ^3 and
2 7 ifjXeQ*.
Proof. Consider the action of Q\ on F(Io) = A^. Each orbit has
length 2, corresponds to a sextet X and the kernel is [gi,O2(Li[S])]
 4.7 The Leech graph 165

which is dual to the subgroup of order 2 in the irreducible Todd module

^ n . Since different sextets correspond to different subgroups in the Todd
module, the kernels at different orbits are different. This implies that
the Gi(v)-orbit of Jx is uniquely determined by the (Gi(v)Qi/<2i)-orbit
of Q(JL) and by (3.7.3) we get (i). Furthermore, Q>\ U {v} is a Qi-orbit
and £(®2) consists of the sextets adjacent to Q(V) in the sextet graph.
If Ji e Oi U <D2, then |Gi(v) n G\{Ji)\ is divisible by 2 18 , hence v and ji
are adjacent by (4.7.1) and we have established the "if" part of (ii). By
(4.6.1) and (4.6.2) we have Gi(v) - 2 4+6 .2 6 .3-Sym 6 , |GiHGi(v)l = 2 10 and
|Qi nQ(v)| = 2 4 . By the obvious symmetry IG(v)nGi(v)| = 2 10 and hence
(G(v)nGi)Qi/Qi = O 2 (Li(v))Gi/Qi. By the proof of (3.7.3) the orbits of
02(Li(v)) on £(<X>3) and g(®4) are of length 2 4 and 2 5 , respectively. Since
every orbit of Q(v) on F(v) must be of length 2, v is not adjacent to
vertices from O3 U Q>4 and we obtain the "only if" part of (ii). In view
of (3.8.2) it is easy to see that g i n Q(v) equals to C ei (O 2 (Li[e(v)])) and
it is generated by the elements T(B, 1) for the octads B refined by Q(V).
The matrices given in the proof of (3.7.3) show that for JL e O3 U $4
there is an octad refined by Q(V) which has odd intersection with a tetrad
from Q(JI). Hence the orbits of Q(v) n G\ on O3 and O4 are of length 2 5
and 2 6 , respectively. If <X>i = {1} then I is fixed by G\ n G(v) and by the
proof of (i) {Io>^} is an orbit of Q(v). By (i) 1 is not adjacent to vertices
from O3 U O4 and hence the latter set is disjoint from its image under an
element from Q(v) \G\. In view of the above this gives (iii). •

By the above lemma, for every edge {v, p] of T there is a unique vertex
I such that T = {v,ju,I} is a triangle and T \ {a} is an orbit of Q(oc)
for every a € T. Such triangles will be called lines. It follows from the
tables in (4.4.1) and (4.6.1) that I is the unique element in A # stabilized
by G(v,Ji) which implies the following.

Lemma 4.7.3 If T = {v, ji, 1} is a line then v + / 2 + ! = 0. •

In the notation introduced in the proof of (4.7.2) let Ji e 0>t for i = 3

or 4. Then by (4.6.1) and (4.7.2 (iii)) Ji e A64(v) if i = 3 and Ji e Kf (v) if
i = 4. This gives the valencies between A4a and A4 for a = 6 and 4b as
on the diagram.

In order to calculate the remaining data on the suborbit diagram we

adopt the following strategy. Notice that the subdegrees of G on A4 are
pairwise different, hence ju e A^v) if and only if v e A%(p). In order to
calculate the valencies on the diagram, for all a,/? € {8,6,5,4a,4b,4c} we
166 Conway groups
have to determine the number n^ of vertices in AJ^v) contained in A%
for a given vertex v e A4. Since the orbitals are self-paired, n^ is equal
to the number of vertices in A\a n A^(v). Hence the numbers na can be
calculated using the following lemma.

Lemma 4.7.4 Let v be a preimage ofv in A4 and fii, ...,/M8 be the preimages
ofjie A4a in A4a such that \IIAM — —\*>\ for 1 < i < 24. Then the multiset
consisting of the values £(/*,-, v) for 1 < i < 24 is the shape ofv in the
basis {^fij \ I < i < 24}. By (4.6.1) this multiset determines the orbital
containing {v9ji}. •
Below we apply this lemma for the remaining cases. We start with two
remarks concerning the action of G\ on A4a.

Lemma 4.7.5 The group L\ acting on A^a has two orbits O\ and O2 such
that g(Oi) = g(Af) for i = 1 and 2.
Proof. Clearly L\ acts transitively on g(A4a). Let vi and vi be vectors
from A%a such that |X(vi)| and |X(v2)| have different parities. Then vi ^ V2
and since L\ preserves multisets of coordinates, vi and V2 are in different
L\ -orbits. D

Lemma 4.7.6 Let B be an octad, R be an orbit ofQi on A\a and Z = Q(R),

where Z = {Si,S2,...,Se} is a sextet. Then T(B, 1) is not contained in Q\(R)
if and only if n(B,X) = (3 I 5 ) in the notation of (2.14.1).
Proof. Let vi and V2 be vectors from A\a such that R = {vi,V2} and
suppose that v[(B)1) = v2. Then P4(vi) = P4(v2) = St for some 1 < i < 6,
|Z(vi)| and |X(v2)| have different parities and hence |BnSj| must be odd.
Now the result is immediate from (2.14.1). •

Lemma 4.7.7 Let v e A*}. Then G\(v) has four orbits on A\a with lengths
35, 35, 1680 and 1792 which are contained in AJ(v) for a = 4a, 6, 4b and
5, respectively.
Proof. Let B be the octad such that Li(v) = Li[B]. Then by (2.14.1)
and (3.7.2) Gi(v) acting on ^(A^a) has 3 orbits $ 1 , $ 2 and $ 3 with lengths
35, 840 and 896 containing the sextets such that A*(£,E) = (4204), (2402)
and (3 I 5 ), respectively. Let *¥i be the preimage of *F,- in A%a for 1 < i < 3.
We claim that the kernel of Q\ on its orbit R contains <2i(v) if and only
if R c ^F1# By the proof of (4.6.1) Qx(vf = {T(BU 1) | |Bi n B\ e {0,8}}.
A sextet from ¥ 1 refines B and hence Qi(v) is in the kernel of R when
 4.7 The Leech graph 167
R c ¥ 1 . If R c *F3 then T(£, 1) is not in the kernel by (4.7.6). Suppose
that gi(v) is in the kernel of R if R c *F2. Then by (4.7.6) the action of
T(B, 1) on A\a coincides with that of T(B\, 1) where \B\ n B\ = 0 which is
impossible and the claim follows. Thus the actions of Gi(v) on ¥ 2 and
¥ 3 are transitive. On the other hand we know that v is adjacent in T to
exactly 35 vertices in A\a. Hence there are 2 Gi-orbits in *Fi of length 35
each.
Let v be a preimage of v in A4 so that P2(v) U Peiy) = B and suppose
that X(v) = P6(v). Let X e A\a with P6(v) c PA(X) <= P2(v) U P6(v) and
X(X) = P6(v). Then 2 € ¥ 1 and ±(v,vl) = 6, which shows that one of the
Gi(v)-orbits in *Fi is contained in A^v) (the other one is known to be
contained in Af). Now let X e Af with P6(v) c P4(A), |P 4 (A)nP 2 (v)| = 2
and X(A) n (P2(v) U P6(v)) = P6(v). Then 1 G ^ 3 , |(v,A) = 5 and we
have ¥ 3 cz A^v). By similar calculations we can deduce that ¥2 c A ^
but we can also apply a different argument. In the notation of (4.7.2) a
sextet from ^(^2) is adjacent in the sextet graph to a sextet from (KO3)
(the diagram Ds(Mat24)) and hence v must be adjacent to some vertices
in A4*. Since we are left with only one Gi(v)-orbit to locate, the result
follows. •

Lemma 4.7.8 Let v be a vector from A4 such that v(a) = 3 mod 4 for all
ae^ and put Y = P5(v) U P3(v). Then
(i) Gi(v) acting on ^(A^a) has three orbits A\, A2 and A3 with lengths
21, 630 and 1120,
(ii) Gi(v) acting on Kf has six orbits with lengths 21, 21, 630, 630,1120
and 1120 contained in A%(v)for a = 6,5,4b, 5,4c and 5, respectively.
Proof. By (3.10.1) we obtain (i) and also that A,- consists of the sextets
Z such that Y intersects i tetrads in Z. Let A,- be the preimage of A,- in K\a
for 1 < i < 3. It was established in the proof of (4.6.1) that G\(v) = Li(v)
and by (4.7.5) A,- n O\ and A,- n O2 are the orbits of Gi(v) on A,-. Now
considering representatives ji from the orbits A,- n O ; , 1 < i < 3, j = 1,2,
and calculating the inner products of v with the vectors ^ , 1 < k < 48,
from the preimage of Ji in A\a and using (4.7.4) we obtain (ii). •

Lemma 4.7.9 Ifv e Af, then

(i) Gi(v) = Mat\2.2 acting on ^(A^a) has three orbits ©4, ©3 and © 2
with lengths 495, 880 and 396, respectively,
(ii) Gi(v) acting on A\a has four orbits with lengths 495, 495, 792 and
1760, contained in A%(v)for a = 4b, 4c, 4c and 5, respectively.
168 Conway groups

Proof. Let Di = P2(v), which is a dodecad, and let D2 = &\D\ be the

complementary dodecad. By the proof of (4.6.1) the action of Gi(v) on
Q(A^) is similar to that of the stabilizer F in L\ of the partition {D\,D2}.
Hence by (3.10.2) we obtain (i) and also that © m consists of the sextets
£ = {Si,S2,...,S6} such that
m = max|D ; nS,-|.

Let © m be the preimage of 0 m in A4a. Then by (4.7.7) © 3 must be a single

Gi(v)-orbit contained in A^v). By (4.7.4) and straightforward calculations
we see that ©4 consists of two Gi(v)-orbits contained in A 4b(v) and A ^ v ) ,
respectively. Similarly (4.7.4) shows that © 2 is contained in A 4c(v) and
it only remains to show that Gi(v) acts on © 2 transitively. Let A be a
vector from A\c such that P2(X) = P2(v), P4(A) ± P 4(v), P4(X) c X{X) and

B := {X(X) O P2{X)) U P4(v) U P4(A)

is an octad. Then by the proof of (4.6.1), or otherwise, one can see that
I = v. Let n e A\a with P4(v) U PA{X) <= pA(ji) c p 4 ( v ) U PA(X) U P2(v),
|P4(/x)nP 2 (^)nX(^)| = 1 and X(n) = 0. Then Ji e © 2 . It can be shown (the
proofof (3.10.2)) that there is 5 = T(B,G) e G\ such that P4(//)<T = P4(/i)
and v*5 = L Then |B n P4(/^)| = 3 and hence for fi\ — \ib we have Ji\ ^Ji
and the result follows. •

Lemma 4.7.10 Ifv € Af, then

(i) Gi(v) fcas six orfcite on g(A44a) with lengths 7, 28, 56, 336, 448 and
896,
(ii) Gi(v) ftas dgfa orbits on K\a with lengths 7, 7, 56, 56, 56, 672, 896
and 1792 contained in A4(v) for a = 4a, 4b,4b, 6,4b,4b,4c and 5,

Proof, (i) is immediate from the proofs of (4.6.1) and (3.10.3). Let
{Qi I 1 < i < 6} be the set of Gi(v)-orbits on ^(A^a) assuming that
|n,-| < |O/| for i < j and let Q/ be the preimage of Q, in A\a. By (4.7.8)
Q6 is a Gi(v)-orbit contained in A|(v); by (4.7.9) Q5 is a Gi(v)-orbit
contained in A4C(v). By (4.7.7) and its proof v is adjacent in T to 56
vertices from AJJ. Furthermore, if Ji is one of them and {v,p,,l} is a line
then I € A4b. This shows that Q3 consists of 2 Gi(v)-orbits contained in
A^(v) and A4b(v), respectively. By the paragraph after (4.7.3) Qi consists
of 2 Gi(v)-orbits contained in A ^ v ) and A4b(v), respectively. By the
above Q2 and Q 4 are contained in A^(v). It only remains to show that
 4.8 The centralizer of an involution 169
Gi(v) is transitive on each of these 2 sets and we suggest this as an
exercise. •

4.8 The centralizer of an involution

A clique in the Leech graph F will be called *-closed if together with an
edge it contains the line containing this edge. By (4.7.3) if L is a *-closed
clique then its vertices form the set of non-zero vectors of a subspace
V(L) in A, particularly \L\ = 2a — 1 where a is the dimension of V(L).
This means that the vertices and *-closed cliques contained in L form
the projective geometry of V(L) with respect to the incidence relation
defined via inclusion.

Lemma 4.8.1 If L is a maximal clique in F, then L is *-closed, it contains

15 vertices and G[L] induces on L a transitive action. There are 2 orbits
££v and !£t of G on the set of maximal cliques in F and \S£V\ = 3 • \^t\-
Proof. By (4.7.2) a maximal clique L containing Io is of the form
{Io} U (Q~{(K)) where K is one of the maximal cliques in the sextet
graph described in (3.3.8). Thus the maximal cliques in F containing Io
are in two G\ -orbits bijectively corresponding to the classes Jf*y and Jft
of maximal cliques in the sextet graph. Since \Jfv\ = 3 • \tft\ and G
acts transitively on the vertex set of F, there are two G-orbits, JSf„ and
<£t on the set of maximal cliques in Y so that if Io € L e j£?a then
Q(L \ {Io}) £ J^oc for a = v or t. Also because of the transitivity of G on
the vertex set of F, G[L] is transitive on the set of vertices in L. By the
above, whenever L contains an edge incident to Io> it contains the line
determined by this edge. Hence the transitivity of G[L] on L implies that
L is *-closed. •
It is clear that a *-closed clique of size 3 is a line. A *-closed clique of
size 7 is the intersection of a clique from <£v and a clique from <£t; such
a clique containing Io is of the form Io U Q~{(M) where M is (the set of
sextets incident to) an element of type 2 in ^{Mat24).

Lemma 4.8.2 For i = 1, 2 and 3 let I/,- be the subspace in A generated by

k\, K\ U K\a and K\ U A^ U Kf U Kf U K\c, respectively. Then
l<Ui<U2<U3<A
is the only composition series of A as a module for G\ and the following
assertions hold:
170 Conway groups

(i) t7i
(ii) U2/U1 is the centralizer of Q\ in A/U\,
(iii) I/2/I/1 = Vn and U3/U2 = #11,
(iv) Ui = U3, Ui = l/2.
Proof. For i = 1 and 2 let M,- be a *-closed clique in the Leech
graph of size 2i+1 — 1 containing Io. Then 7(Af,-) is an (i + l)-space in
A containing U\ so that V(Mi)/U\ is an i-space in Ui/U\. Furthermore
p(Mt \ {Io}) is a sextet (an element of type 1 in ^ ( M a ^ ) ) if i = 1
and it is a *-closed clique of size 3 in the sextet graph (an element of
type 2 in ^(MatiA)) if i = 2. This means that C/2/C/1 supports a natural
representation of & (Mat24) invariant under Mat-u- Hence U2/U1 = ^ n
by (3.3.10). Since G\ preserves the quadratic form 6 on A as in (4.5.6), the
remaining assertions are straightforward from (4.4.1) and (4.6.1), noticing
that C/f = I/3 and C/f = l/2, as stated in (iv). •

The following result will play an essential role in Section 5.6.

Lemma 4.8.3 Let E2 = {v e A$,\X(v)\ = 1}, E4 = {v e A44a,X(v) = 0},

£ 3 = {v e A\9X(v) = P3(v)} and E = {0} U E2 U £4.
(i) Li = Ma^24 stabilizes E\ setwise for i = 2, 3 and 4,
(ii) Li = Mat24 acts on £, as it acts on the set of ordered pairs of
element subsets from &>, on the 1-element subsets of & and on the
set of sextets, for i = 2, 3 and 4, respectively,
(iii) E is a complement to U\ in 1/2,
(iv) E3=E±nAl

Proof. The assertions (i) and (ii) are immediate. A Leech vector and
its negative have the same image in A; if vi, V2 € £4 and P^vi), P4(v2) are
tetrads from the same sextet then vi = V2; by the proof of (3.7.3) in any
two sextets there are tetrads intersecting in at least two elements. Using
these facts it is easy to see that E is closed under addition which implies
(iii) since |£| = 211. Finally (iv) follows directly from the definition of the
invariant quadratic form 6. •

Lemma 4.8.4 Let B be an octad, *F be the set of sextets which refine B,

*¥ be the preimage 0 / $ in A%> and W = O 2 (Li[£]). Then
(i) W is the kernel of the action of Li[B] on x¥,
(ii) [Qi, W] is the kernel of the action of Q\ on ¥
 4.8 The centralizer of an involution 171
Proof. By (2.14.1) W is the kernel of U[B] on $ and by (4.7.5) we
obtain (i). If B\ is an octad such that \B n B\\ € {0,8}, then B\ has even
intersection with the tetrads in every sextet from *F and T(B\, 1) is in the
kernel of Q\ on *F by (4.7.6). The elements T(J?I,1) as above generate the
commutator subgroup [Q\9 W] of order 25. Since the action of Q\ on *F
is non-trivial and by (3.8.4) Qi/[Qi,W] is irreducible as a module for
Li[B], (ii) follows. D
By (2.15.1) Gi, acting on Qf by conjugation, has two orbits with
lengths 759 and 1288 containing elements T(Y,1) where Y is an octad
and dodecad, respectively (recall that T(Y, 1) = z(0> \ Y, 1)). By (4.6.2) if
Y is a dodecad then Io is the only element JL € A4 such that T(Y, 1) €
Qifi) := O2(G(/2)).
For an octad B and <5 = T(£, 1) put
O = 9(5) = {[1 I AI e A4,S € Q(p)}
and let D be the setwise stabilizer of <D in G. Then by the above paragraph
and (4.6.2) we have the following.

Lemma 4.8.5 The action ofD onQ> is transitive and D = CG(S). The set O
consists oflo, the 35 Q\-orbits on A^a corresponding to the sextets which
refine B and the Q\-orbit on K\ which corresponds to B in the sense of
(A.62 (ii)). In particular
|O| = 1 + 2 - 35H-26 = 135
and O n T(I0) = ¥ in the notation of (4.8.4). •
Let R be the kernel of the action of D on O and D = D/R, V\ be the
subspace in A generated by <D and V2 = R/Z(R). We continue to follow
notation introduced in (4.8.4).

Lemma 4.8.6 The following assertions hold:

(i) 5^Q+(2),
(ii) R = [QU W] : W, Z(R) = [R,R] = (T(B, 1)) and R S 2l+\
(iii) V\ = Ch(R) and V\ is the natural module for D,
(iv) vf consists of €> ,64 elements from A* and 56 elements from K\,
(v) V2 = V* where a is a diagram automorphism of D.
Proof. Let 2f be an incidence system whose elements of type 1, 2,
3 and 4 are the vertices, the lines, the cliques from ££t and the cliques
from JS?,,, respectively, contained in O. A clique from S£t and a clique
172 Conway groups
from ££v are incident if their intersection is (a *-closed clique) of size
7 and the remaining incidences are via inclusion. Let {xi,X2,X3,X4} be
a maximal flag in S> where x,- is of type i and xi = Io- Let U be the
subspace in A generated by *F. Then by (4.8.2) V /JJ\ is the centralizer
of W in Ui/U2 = « n and by (3.8.2) U/Ui = /\2 W. This shows that
the *-closed cliques of size 2' — 1 containing Io and contained in O are
in a bijection with the totally singular subspaces of dimension i — 1 in
U/U\ with respect to the unique quadratic form of plus type on U/U\
preserved by L\[B]/W ^ L4(2) ^ Q£(2). This shows that res^(xi) is
the projective rank 3 geometry over GF(2) and also that for a *-closed
clique N of size 7 such that Jo e N ^ Q> there are a unique element of
type 3 and a unique element of type 4 in 3f which contain N. By the
latter observation, for j = 3 and 4 the residue res^(x;) is isomorphic to
the rank 3 geometry of proper subspaces of V(XJ). Hence 3) is a Tits
geometry with the following diagram:

o-
2

By (1.6.3) & is the parabolic geometry of Qf (2) and by (1.6.5) the latter
group is contained in D. Since Qg~(2) is the automorphism group of its
parabolic geometry (i) follows. By basic properties of 3) the stabilizer of
Io in D acts faithfully on x¥. This gives the first equality in (ii) because of
(4.8.4). By (3.8.2) (T(B, 1)) is the centralizer of W in [Qu W] and also the
commutator subgroup of [Qu W] and W. Hence noticing that [Qu W]
is elementary abelian of order 25 we obtain the remaining equalities in
(ii). By the construction V\ supports a natural representation of {& and
hence it is the 8-dimensional natural module for D. From (4.4.1) and
the table therein it is easy to see that R, besides the elements in O,
stabilizes 56 elements in A^ which are images of vectors from A2 with
P4 ^ B and 64 elements in A2 which are images of vectors from A2 with
P2 = B. Thus (iii) and (iv) follow. Since V2 = R/Z(R) is 8-dimensional it
is isomorphic either to V\ or to its image under a diagram automorphism
of D. Since [Qu W]/Z(R) is a 4-dimensional subspace in V2 normalized
by the stabilizer D n G\ of the vector Io from V\ we have the latter
possibility as stated in (v). •
 4.9 Geometries of Co\ and C02 173
We will use the same letter <X> to denote the subgraph in T induced
by Q>. Let K be the orthogonal form (of plus type) on V\ preserved by
D. Then <X> = V\ n A4 is the set of isotropic and V\ n A2 is the set of
non-isotropic vectors in V\ with respect to K and we have the following.

Lemma 4.8.7 The following assertions hold:

(i) ifk eAandleVi then KQ) = 0 if X1 = 0 mod 8 and K(1) = 1 if

X2 = 4 mod 8,
(ii) D acts transitively on FiPiA2 with stabilizer isomorphic to Sp$(2). •

By (4.8.7) Q> is a graph on the set of vectors in V\ isotropic with respect

to K in which two such vectors are adjacent if they are perpendicular.
Hence the suborbit diagram of O with respect to the action of D is the
following:
1+36 35

d> 70 1 / 7 ~ T ^ 32 35

Comparing this diagram with the suborbit diagram of the Leech graph,
we observe that whenever v, Ji are non-adjacent vertices in G>, r(v)riT(p) c
O. On the other hand, since (v,p) is non-singular, V\ = (v,p) © (v,/*)-1,
where by (4.8.7) the latter summand is (F(v) n T(p)). This gives the
following.

Lemma 4.8.8 Let v G A4 and p, e A$(v). Let V\(v9p) = (v,p,T(v) n T(p))

(a subgroup in A) and Q>(v,Ji) be the subgraph in T induced by the vertices
contained in Vi(v,p). Then

(i) Vi(v,p) = Vf and O(v,/i) = O g for some g € G,

(ii) K!(v,/i) = (v,/i,(/ 2 (v)nl/ 2 (/i)). •

4.9 Geometries of Co\ and C02

The maximal parabolic geometry of Co\ can be defined in terms of
the Leech graph and some of its subgraphs. To wit, let 3tf(Co\) be an
incidence system of rank 4, whose elements of type 1, 2, 3 and 4 are the
vertices, triangles, maximal cliques from the class 5£t and all images under
G of the subgraph O defined before (4.8.5), respectively. The incidence
relation is via inclusion.
174 Conway groups
Lemma 4.9.1 The incidence system 3^{Co\) is a geometry with the diagram

and G = Co\ acts on ^(Co{) flag-transitively.

Proof. Since the incidence relation is via inclusion, it is easy to see that
J^(Coi) is a geometry with a string diagram. By (3.3.8) and (4.8.1) we
identify the residue in J^(Coi) of an element of type 1 (say of l 0 ) with
the maximal parabolic geometry 3#?(Mat24) of the Mathieu group. Then
by (4.8.5) the residue of an element of type 4 (say of O) is isomorphic
to the geometry of totally isotropic 1-spaces, 2-spaces and one class of
4-spaces in the natural module of D =fig"(2). In other terms this residue
is a truncation of the Tits geometry of the latter group and the result
follows. •
If {Hi | 1 < i < 4} is the amalgam of maximal parabolics associated
with the action of G on J^(Co\\ then

Hi £ Gi S 211 : Mat24, H2 S 24+12.(3 • Sym6 x Sym3),

H3 £ 22+12.(L4(2) x Sym3), H4 = D

In order to obtain a diagram without fake nodes, similarly to the

case of Mat24 we define the minimal parabolic geometry ^(Co\\ whose
elements of type 1 and 2 are the same as in 3tf(Co\). The elements of
type 3 are the *-closed cliques of size 7 in T while the elements of type
4 are the cliques from S£v. The incidence relation is via inclusion.

Lemma 4.9.2 ^(Co\) is a rank 4 tilde geometry with the diagram

2 2 2 2

and G = Co\ acts on it flag-transitively.

Proof. First notice that a clique from ifv is contained in a unique
subgraph which is an element of type 4 in 3tf{Co\) while a clique from
££t is contained in three such subgraphs. By (3.3.9) the residue in ^(Coi)
of an element of type 1 is the tilde geometry &(Mat24) while the residue
 4.9 Geometries of Co\ and C02 175
of an element of type 4 is the rank 3 projective GF(2)-geometry of proper
*-closed subgraphs contained in the corresponding maximal clique. •
Let {Gi I 1 < i < 4} be the amalgam of maximal parabolics associated
with the action of G on ^(Coi). Then
Gi S Hi S 2 11 : Mat 24 , G2=H2 = 2 4+12 .(3 • Sym6 x Sym3),

G3 S 2 2+12+3 .(L 3 (2) x Sym3), G4 = 2 1+4+4+6 .L 4 (2)

(notice that G4 is the preimage in D of a maximal parabolic subgroup in
D which is the stabilizer of a maximal totally isotropic subspace in the
natural module).
Directly by the definition of the Leech graph we obtain the following.

Lemma 4.9.3 The Leech graph is the collinearity graph of both

and @(Co\) and each of the geometries possesses a natural representation
in A. •
For v € A and j = 0,2,3 or 4 put
A,(v) = Aj + v = {X 11 e A, v + I e A,-},
so that Aj = A/(6).
Let us fix a vector fio € A2 and let F denote the stabilizer G(/2o) of Jlo
in G, which is isomorphic to C02 by (4.5.5) and will be identified with
in view of the remark before that statement.

Lemma 4.9.4 For j = 2,3 and 4 the action ofF on A4nA/(/2o) w transitive
with stabilizers isomorphic to
2 10 : AutMat229 Mat2i and 2 5 : 2 4 : L4(2) £ 2 ^ : L4(2),

Proof. Notice that for j = 4, 3 and 2 we have A^ = A2 n Afe(Io) for

k = 2, 3 and 4, respectively. Since G acts transitively on both A2 and A^
in view of this observation the result is immediate from (4.4.1) and the
table therein. D
We study the subgraph 0 = ©(/*o
) of the Leech graph T induced by
^ n A2(/io). First notice that
| e | = IA4I • |AJ| • IA2I"1 = 4 6 5 7 5 .
In order to work with the coordinates for Leech vectors introduced in
Section 4.4, it is convenient to assume that Jio € A^ (so that IQ £ ®). If
176 Conway groups

in addition we assume that X(fio) = 0 and put {a,b} — P4OU0) then an

element X € A4 is contained in 0 if and only if X is the image of a vector
X € A4 such that X(a)-\-X(b) = 8. There are exactly two choices for X when
X is fixed (in view of the transitivity of F on 0 it is sufficient to check this
for X = lo)- Furthermore, Fi := F(Xo) = G\(Jio) is the semidirect product
Qiifio) : Li(po) where Qi(Jio) consists of the elements T ( Y , 1 ) for Golay
sets Y containing P4O10) and Li(juo) = Li[a,b] = AutMat2i- By (2.15.3)
this shows that 2i(/2o) = #2(^1) is isomorphic to the 10-dimensional
Golay code module.

Lemma 4.9.5 The group F\ acting on ©has five orbits 0 D A | for cc =

8,6,5, 4a and 4b. Their lengths and element stabilizers are given in the table
below. The group F\/02(F\) = AutMa£22 acts on the set of O2{F\)-orbits
in © n A ^ / or a = 4a, 6,4b and 5 as it acts on the elements in 3tfp, 3tfh and
3^o of the geometry ^(Matji) and on the elements of the Steiner system
5(3,6,22), respectively.

Orbits Lengths Stabilizers

®DAl 1 [210] :AutMaf 2 2

0r\A64 2 5 -77 [25] : (24 : Sym6)

&r\A54 2 10 • 22 P2L 3 (4)

2-231 [29] : (25 : Sym5)

0nAf 2 6 • 330 [24].(2 x 2 3 : L3(2))

Proof. We have seen that X e ©if and only if X is the image of X e A4

with X(a) + A(b) = 8. This immediately shows that ©n K\ c = 0, 0 n A^
consists of the images of A e h\ with P 6 (^) ^ {a,b} c P6(A) u P 2 (^),
for fS = 4a and 4b the set 0 n A { consists of the images of X with
X(a) = X(b) = 4 and ©n A | consists of the images of X with Ps(X) c=
{a,b} cz Ps(X)UPi(X). The transitivity of F\ on © n A^ can be established
in the following way. Choose an appropriate pair A,v e AJ such that
X(a) + X(b) = v(a) + v(ft) = 8 a n d l = v; calculate the stabilizer H of
{/l,v} in F\ and observe that the order of H equals the order of F\
divided by the size of ©n A4 (notice that F = C02 splits over the centre
of G = COQ). Along these lines we also obtain the structure of element
 4.9 Geometries of Co\ and C02 111
stabilizers. For this purpose (4.4.1), its proof and the table there are very
useful. •

Lemma 4.9.6 The suborbit diagram of the graph 0 corresponding to the

action of F is as given after the proof

Proof. It is straightforward to calculate the diagram using (4.9.5),

methods and results from Section 4.7, particularly (4.7.4) and the dia-
grams for Jt(Mat22) calculated in Section 3.9. •
15

Let ^(Coi) be the subgeometry in ^(Co\) formed by the elements

(*-closed cliques) contained in ©.

Lemma 4.9.7 is a rank 4 Petersen geometry with the diagram

2 2 1
and F = C02 acts on -transitively.
Proof. Using (3.4.5) and the remark after its proof, we identify the
residue in &(Co2) of an element of type 1 with the rank 3 Petersen
geometry <&(Mat22)> For an element of type 4 its residues in (&(Mat22)
and &(Mat24) are the same. Finally, since F acts transitively on the vertex
set of 0 and Fi acts flag-transitively on the residue of Io i n ^(£#2), we
conclude that the action of F on <§(Co2) is flag-transitive. •
Let {Ft I 1 < i < 4} be the amalgam of maximal parabolics associated
with the action of F = Co2 on ^(Co 2 ). Then

Fi S 2 10 : Aut Mat22, F2 = 2 4+10 .(5ym 5 x Sym3),

178 Conway groups
1+7+6
F3 s 2 .(L 3 (2) x 2), F4 S 2 1+4+6 .L 4 (2).

Let 5 be an octad containing Ptifio) and 5 = T(B, 1). By (4.8.7 (ii))

CF{$) = CG(^) n G(jiio) is the preimage in D = CG(S) of the stabilizer of
po in 5 , so that CF(8) s 2^+8.Sp6(2). By (4.8.7 (i)) V := «(5) n 0 is the
set of isotropic vectors in V\ — CjJifiiiD)) orthogonal to Jx^ with respect
to the bilinear form associated with K. Furthermore the elements of type
1, 2 and 3 in ^{Coi) contained in *F correspond to totally singular 1-,
2- and 3-dimensional subspaces in the natural orthogonal module pjj-
for S/?6(2) = OT(2) and hence they form a subgeometry 0t in ^(Coi)
isomorphic to the classical C3(2)-geometry #(Sp$(2)) with the diagram

The residue of Ao in 0t is the 5p4(2)-subgeometry in

22) as in (3.4.4). This can be summarized as follows.

Lemma 4.9.8 Let d be an involution in F = C02 conjugate to an involution

from the orbit of length 11 of F\/02{F\) on O2(F\). Then the elements of
type 1, 2 and 3 in ^(Coi) which are pointwise fixed by O2(Cp(S)) form
a subgeometry 01 isomorphic to #(Sp6(2)) on which CF(S) = 2++8.Sp6(2)
induces a flag-transitive action of Spe(2). •
Notice that the orthogonal complement of Jio in V\ with respect to
the quadratic form K as in (4.8.7) supports the universal 7-dimensional
natural representation of 0t.
Let & be &(Coi) or &(Co2) and let Z be the collinearity graph of
3F (that is T or @). Any two points in !F are on at most one line and
every triangle in S is contained in a plane. Hence 3F is simply connected
if and only if the fundamental group of S is generated by triangles.
The latter can be established directly using the suborbit diagram of IF
and this strategy was realized in [Sh92] for the case of ^(Coi). Here we
follow the strategy developed in [Iv92d] which enables us to deal with
a smaller number of cycles. In order to implement the strategy we need
some further information about Co\ and its action on A.

4.10 The affine Leech graph

The semidirect product A : Co\ acts naturally on A. By (4.5.5) the orbit
of a pair I, v e A under this action is uniquely determined by the value
i such that l + ve A,-, where i = 0,2,3 or 4. Let A,- denote also the
 4.10 The affine Leech graph 179
corresponding orbital graph on A, so that A;(v) is the set of elements in
A adjacent to v in the graph A*. Let p^ be the intersection parameter
defined by

for k G Afc(v). Since all the orbitals are self-paired i.e., X G A^(v) if and
only if v + 1 G Afe, we have the following:

Pkij = Pkji a n
d |Afc| • pkij = |A_/1 • p?ki.

We are particularly interested in the graph A2 and call it the affine

Leech graph. In this section we calculate the suborbit diagram of A2.
As above Io denotes the element in A4 stabilized by G\ = Q\ : L\, so
that {Io} = AJ. By the tables in (4.4.1) and (4.6.1) and in view of (4.5.6)
we obtain the following result which gives the valencies of A4 on the
diagram of A2 (as well as on the diagrams of A3 and A4).

Lemma 4.10.1

A3(Io) = A^ U A^ U A^ U A|,

A 4 (I 0 ) = A22 u A^ u Kf u Kf
11178+37950

8386560 ^ A3

48 600

98280

A4

48 576

Lemma 4.10.2 The intersection parameters of A2 are as on the above dia-

gram.
180 Conway groups

Proof. By (4.5.6) the graph A3 is the graph on the set of vectors of the
24-dimensional GF(2)-space A in which two vectors are adjacent if their
sum is non-singular with respect to the non-degenerate quadratic form 6
of plus type. This graph is well known to be strongly regular [BCN89],
which means that p33 = p33. On the other hand by (4.10.1) we have

P33 = \*4\ + |A||, p234 = \A\\ • IA4I/IA2I,

and hence

Now the remaining intersection parameters follow from the obvious

identities in which they are involved. •
To complete calculating the diagram there remains to show that G(v)
acts transitively on A2OO n A2 and on A2(v) n A3 if v G A2 and that G(v)
has two orbits on A2OO n A3 if v G A3.

Lemma 4.10.3 The following assertions hold:

(i) the subgraph in A2 induced by A* has the following suborbit diagram
with respect to the action of G\

(ii) G\ acts on this subgraph vertex- and edge-transitively with the ver-
texwise stabilizer of an edge being isomorphic to 2 9 : 1^(4);
(iii) ifv\9V2 G A* and V2 G A2(vi), then v\ + vi G A2.
Proof. The orbits of g i on A2 are all of length 2 and there is
an obvious mapping from the set of these Q\-orbits onto the set of
2-element subsets of 9 which commutes with the action of G\. Two
elements from A2 are adjacent in A2 if their Qi-orbits are different and
correspond to intersecting pairs from ^ , which gives (i). Let fio and
Jlo be as introduced in the paragraph before (4.9.5), so that fio G A2
with P4Guo) = {a,b} and X(/JQ) = 0. Let c G & \ {a,b} and let fiU fi2
be vectors in A2 such that P*{ii\) = {b,c}, X(JJ,\) = 0, P4GU2) = {a,c},
{c}- Then Jiujii ^ A2 and fio — fi\ — fa, which means that
 4.10 The affine Leech graph 181

T = {A2O,/2I,JU2} is a triangle in the graph A2. In particular {/2o,£i}

is an edge in the subgraph induced by K\. It is easy to check that
G(Xo,Jio,jii) is the semidirect product (Qi(/io) n <2i(£i)) : Li(a,fr,c), where
(2i(/io) n Gi(^i) consists of the elements r(7,1) where Y is a Golay set
containing {a,b,c} and L\(a,b,c) = 1^(4). In view of (2.15.4) (ii) follows.
Now (iii) is straightforward. •

Lemma 4.10.4 Let T = {/io, A^fe} ^ A* and juo + £i + fe = 0. For i = 1

and 2 put [7(0 = G(Jio,jii) and S(i) = A2(/io) n A2(ju,) n A4. Then
(i) l/(i) acts transitively on S(0 with stabilizer isomorphic to 29 : Ls(4),
(ii) [7(1) = [7(2) and 3(1) = S(2).

Proof. By (4.10.3 (ii)) G\ acts transitively on the set of ordered pairs

(6b,rj) of elements from A* with co G A2(*7) and hence (i) follows. It is
clear that [7(1) = G(/2o,/ii) = G(/io,/ii + juo) = [7(2). Hence in view of (i)
in order to prove the second equality in (ii) it is sufficient to show that
S(l) n 3(2) ^ 0, but since K\ = A2 n A 2 (I 0 ), we have l 0 € S(l) n 3(2). •

In view of the above lemma we put U = [7(0 and B = B(0 for i = 1

or 2. Taking i = 1 we observe that an element I € A4 is contained in B if
and only if it is the image of X e A4 such that X(a)+k(b) = A(fc)H-A(c) = 8.
By (4.4.1) and (4.6.1) we conclude that B n A4 consists of the images of
the X with X(a) = X(b) = X(c) = 4 for a = 4a, X(a) = X(c) = 2, X(b) = 6
for a = 6, X(a) = X(c) = 3, X(b) = 5 for a = 5, and S n A j is empty for
P = 4b and 4c. Arguing as in (4.9.6) we obtain the following.

Lemma 4.10.5 The suborbit diagram o / S corresponding to the action of

U is the following:
1 5 21

42

The action of U\ = G\ n [7 = 2 9 : 1^(4) on B is faithful and transitive on

EnA%for a = 4a, 6 and 5 wit/i O2([7i) having orbits of lengths 2, 2 4 and
2 9 , respectively. •

Lemma 4.10.6 77ie group U is isomorphic to Ue(2) and it acts on B as on

the set of maximal totally singular subspaces in the natural ^-dimensional
GF(4)-module.
182 Conway groups
Proof. Let Y = {a,b9c}. Recall that Q is a mapping from A\a onto the
set of sextets which commutes with the action of G\. Then Q(EC\ K\a) are
the sextets containing Y in a single tetrad and they are identified with
elements from the set 9 \ Y which is the point set of the projective plane
H(Y) of order 4 as in (2.9.1). Let B be an octad containing Y, so that B
corresponds to a line in H(Y), let 8 — T(B, 1) and D = CG(<5). Then by
(4.8.7 (i)) T is (the non-zero vectors in) a minus 2-space in V\ = C^(02(D))
and T : = S n O ( < 5 ) is the set of isotropic vectors orthogonal to T. Hence
the vertices and lines of T contained in Y correspond to 1- and 2-
dimensional totally singular subspaces in the 6-dimensional orthogonal
space T1 of minus type. By (4.8.7 (ii)) Cv(5) contains O2(D) ^ 2}_+8 and
induces on T a flag-transitive action of Q^"(2) = (74(2). This shows that
the vertices and lines contained in S together with the images of T under
U form (with respect to the incidence relation defined via inclusion) a
C3-geometry 3F with the diagram

on which 17 induces a flag-transitive action. By [Ti82] & = &(U6(2)) and

the result follows. •
Since p\2 = 4600 is exactly the index of U in G(jiio) = C02 we have the
following.

Lemma 4.10.7 If v € A2 then G(v) acts transitively on A2(v) n A2 with

stabilizer isomorphic to 1/6(2). •

It is easy to observe that if T is as in (4.10.4) then the setwise stabilizer

of T in G\ is of the form 2 9 : PTL?,(A) and it induces on T the natural
action of Sym?>. On the other hand the automorphism group of the C3-
geometry associated with U is PTU^l) = C/6(2) : Sym^ and hence we
have the following.

Lemma 4.10.8 Let JLQ be an element from A2, Ji\ be an element from A2 n
A2(/io) and J12 = /2o + Ah- £e* 71 = {juo, jui, JU2}» (7 be the setwise stabilizer
of T in G and S = A4 n A 2 (T). Then
(i) T is a triangle in the affine Leech graph and T ^ A2,
(ii) U induces on T the natural action of Sym?>,
(iii) U is the full stabilizer in G of the subgraph in the Leech graph T
induced by S,
(iv) 17 ^ Aut S ^ PTU6(2) - U6(2).Sym3. D
 4.10 The qffine Leech graph 183
Notice that the stabilizer in F = Co 2 of the subgraph 3 coincides with
the stabilizer of /io in U and it is isomorphic to PLU^(2) ~ £/6(2).2.
Before continuing the calculation of the parameters of the affine Leech
graph we would like to mention an important subgeometry of the ge-
ometry 9F(176(2)) as in the proof of (4.10.6). Let K s PTL3(4) be a
complement to O 2 (Gi[T]) in Gi[T] = 29 : PFL 3 (4) and let co be an
involution in K which realizes the field automorphism. Then co induces
on T an action of order 2. Furthermore, CK(co) = L3(2) x 2 and the
centralizer of co in O2(G2[T]) is of order 2 6 . Hence by (19.9) in [ASei76]
C^{co) = 5p6(2) x 2. Using (4.10.5) and the paragraph before that lemma
it is not difficult to identify the subgraph in 3 induced by the vertices
fixed by co with the dual polar graph of Spe(2) with the following suborbit
diagram:

Lemma 4.10.9 The following assertions hold:

(i) up to conjugation G[T] contains a unique involution co which com-
mutes in G[T] with an element of order 7, and induces on T an
action of order 2,
(ii) Cd(co) * Sp6(2) x 2,
(iii) the points, lines and quads in 3 fixed by co form a geometry iso-
morphic to the C^-geometry <&(Sp6(2)),
(iv) co is conjugated in G to an element T(B,1) from Q\ where B is an
octad.

Proof. By (19.9) in [ASei76] co is as in the paragraph before the

lemma, which gives (i), (ii) and (iii). By (4.10.7) G acts transitively on
the triples {vo, vi, V2} such that v, e A2 n A2(v;) for 0 < i < j < 2 and
vo + vi + V2 = 0. Let vo € A2 with X(vo) = ^Mvo), B be an octad disjoint
from P3(vo) and vi be the image of vo under T(B, 1). The v2 := vo+vi is in
A2 (with P2(v2) = B and X(v2) = 0). By the construction T ( 5 , 1) permutes
vo and vi, stabilizes v2, and the centralizer of ?(£, 1) in G(vo,vi,v2) is
isomorphic to Alt% (in particular has order divisible by 7). By (i) co and
T(JB, 1) are conjugate and (iv) follows. •
184 Conway groups

Let us continue the analysis of the affine Leech graph. For d G 2P

let cod be a vector in A3 such that Ps(cod) = {d} and X(<x>d) = 0 and
put

= {11 e A42\ d e PA(H\X{PL) = 0},

= {fie A32\ P3Qi) = XQi) + {<*}},

= {fieA22\de P2Ui)9x<ji) = 0},

{fie A32\ P3(ii) = {d},\XQi)\ = 7}.

Notice that if /x G *F4(d) then P3(/i) UX(fi) is an octad. Then |*Fi(d)| =
\V2(d)\ = 23, \V3(d)\ = \W4(d)\ = 253 and Gx(<od) = U(d) s Mat23
acts transitively on V,-(d) for 1 = 1,2,3 and 4. In addition if [i\ e
Ti(d), /i 2 € V2 (d) with P4(iUi) = {d} U P3O12) then m + p2 = cod
and similarly if fi3 e x¥3(d), m G ^ ( d ) with ^2(^3) = ^3(^4) U X(/z4)
then fi3 + 114 = ctf</. Hence if ^(d) is the union of the ^ ( d ) , then
¥(d) c A2 n A2(c^). It is easy to see that different vectors from *F(d)
have different images in A and since |*F(d)| = 552 = pl2 we have the
following.

Lemma 4.10.10 ¥(d) = A2nA2(cbd); the action ofGi(cod) = Li(d) = Mat23

on ^iid) is similar to its action on 0>\ {d} for i = 1,2 and to the action
on the set of octads containing d for i = 3,4. •

There is an equivalence relation on *F(d) with classes of size 2 with

respect to which 2 distinct vectors //, v are equivalent if \i + v = a>d> Let Z
be a graph on *F(d) such that /z, v are adjacent if they are not equivalent
and Ji + v G A3. One can check that whenever {jtfi,/^} and {vi, V2} are
2 distinct equivalence classes there is a bijection a of {1,2} onto itself
such that fii and v; are adjacent if and only if j = a(i) (so that £ is a
double cover of the complete graph on 276 vertices). In view of this rule
all the adjacencies in Z are determined by the following 3 conditions:
^i(d) is a coclique; \i G ^i(d) and v G ^3{d) are adjacent exactly
when P4(/z) n Pi{v) — {d}; \iu\ii G ^ ( d ) are adjacent exactly when
I^M^i) n ^2(^2)1 = 2. It is straightforward to reconstruct the suborbit
diagram of £ based at \i G ^ ( d ) , with respect to Gi(a)^,/i) = Mat22
(one might find useful the diagrams for jff(Mat24) and 3tif(Mat22) from
Chapter 3).
 4.10 The affine Leech graph 185
16 16

Now for a 3-element subset D in let cap be a vector from A3 such

that P3(coD) = X(coD) = D and put

*2(D) = <= P3(coD)},

= {/* G A| = 4},

\X(ii)\ = 7},

e A22 n
<D6(D) = = 9}.
Let 0>(D) be the union of the *,(D). Then |Oi(D)| = |O2(D)| = 3,
|*3(D)I = I*4(D)I = 105, \Q5(D)\ = |* 6 (D)| = 168, GI(COD) = LX[D] *
PFL 3 (4) acts transitively on each of the 0/(1)) and we have the following
result similar to (4.10.10).

Lemma 4.10.11 O(D) = A2 n A2(cbD); the action ofGi(coD) = Li[D] =

PFL 3 (4) on 0>i(D) is similar to its action on the elements in D for i = 1,2,
on the maximal flags in H(D) for i = 3,4, on the hyperovals in H(D) for
i = 5,6. •

Recall that by (2.8.2) and (2.9.1), for an octad B intersecting D in two

elements, H = B \ D is a hyperoval in U(D) and two such hyperovals
Hi = Bi \ D and H2 = B2 \ D are in the same orbit of Li(D) = L3(4)
if and only if Bx n D = B2 n D. This shows that if /x G 3>I(D) then
186 Conway groups

acting on O(D) has exactly two orbits of length /

for / = 1, 2, 56, 105 and 112.

Lemma 4.10.12 Let w e A3 and /2, v G A2 n A2(co) with Ji — v^ {0,a>}.

Then
(i) G((b) acts transitively on A2
(ii) G(a>,£,v) = E/4(3),
(iii) G(a), /2) is £/ie sporadic McLaughlin group McL of order 2 8 • 3 6 • 5 3 •
7-11.
Proof. Since G is transitive on A3 we can assume that cb = (b&. For
the same reason there is g e G which maps a>d onto co^ and hence
gG\(coD)g~1 is a subgroup of G(ct^) which acts on *F(d) as G(a>r>) acts on
O(D). Comparing the orbit lengths of G(a)d) on *F(d) given in (4.10.10)
and of G(COD) on O(D) given in (4.10.11), we immediately conclude that
G(cod) acts transitively on the 276 equivalence classes of vectors in *F(d).
Now, comparing the suborbit diagram of E and the orbit lengths of
G(coD,fi) on O(D) (for \i G Oi(D)) given in the paragraph after (4.10.11),
in view of the above we conclude that for every v e ^(d) the action
of G(a>d,v) on Z(v) (which is the set of vertices adjacent to v in S) is
transitive. Since Z is not bipartite we obtain (i).
Let fi be the vector in *Fi(d) at which the suborbit diagram of E is
based, let v be a vector from ^ ( d ) which is not equivalent to \i and let
Y = P4O/) r)Pi(v). Then one can see from the description of E and from
its suborbit diagram that the vectors in E(/x) n E(v) are indexed by the
hyperovals constituting two orbits of L\{Y) on the set of all hyperovals
in II(Y); two vertices are adjacent if and only if the corresponding
hyperovals have no or three common elements. Hence !(/*) n E(v) is the
point graph of the generalized quadrangle % as in (2.16.1) and by (2.16.1)
it is the classical one associated with the group 1/4(3). By the above we
have

It is easy to see that the action of G(co^,/i,v) on E(/i) n E(v) is faithful

and hence by (1.6.5 (vi)) it contains £/4(3). Since |G(a)<*,/x,v)| = |C/4(3)|
we obtain (ii).
It is an easy combinatorial exercise to check that the subgraph induced
by E(//) is a strongly regular graph known as the McLaughlin graph.
The group G(a>d,n) induces on it a rank 3 action corresponding to the
following suborbit diagram:
 4.10 The affine Leech graph 187
30 56

© 112 1/ "\ 81 56

Alternatively one can (as we do) consider (iii) as a definition of the

McLaughlin group McL. •

Lemma 4.10.13 Let Q be the McLaughlin graph and let H = McL act
naturally on Q. Let x9y be a pair of vertices at distance 2 in Q. Then
H(x,y) = Li(4) acting on Q \ {x,y} has three orbits of length 56 and one
of length 105.

Proof. We identify Q with the subgraph in E induced by £(/*). Then we

can assume that \i = nu x = H2 + OJD, y = /*3 + O>D, where {^1,^2,^3} =
Oi(D). Then H(x,y) contains G(coD,fiu 112,1*3) = L3(4) and by (4.10.12)
H(x,y) = G(COD, fi\,\xi, \ii) and the orbit lengths follow from (4.10.11). •
The graph £ represents the two-graph on 276 vertices as in [GS75].
The geometry whose elements are vertices, edges, triangles and maximal
cliques (of size 6) in L is a doubly extended generalized quadrangle of
order (3,9) with the diagram
c c
o o o ^o
1 1 3 9
whose full automorphism group is isomorphic to C03 x 2; the residue of
a vertex is the exceptional EDPS from (1.13.7) associated with McL.
Now there remains to show that G(cb) has two orbits on A3 n
with lengths given on the diagram.

Lemma 4.10.14 For fj e A3 n ^((o) suppose that

(i) Y = A 2 nA 2 (a>)nA 2 0;)f 0,
(ii) G[cb,rj] =/= G(cb,rj).
Let K be the elementwise stabilizer of T in G and A be the automorphism
group of the subgraph in At for i = 2,3 or 4 induced by T. Then \G(cb, fj)\ <
i\K\-\A\. •

Proof. It is clear that G[cb9fj] is contained in G[T] and the order of

the latter group is at most |K| • \A\. •

We assume that co = cbd. If e G 9 \ {d} then a>d — coee A2 so that we

can put fj = cbe. Clearly there is an element in L\ which permutes a>d
and coe. Hence (ii) in (4.10.14) holds and T as in (4.10.14 (i)) is the image
188 Conway groups

in A of the union of the T := ¥,-(d) n ¥,(<?) for i = 1,2,3 and 4. It is

straightforward to check that T4 = 0, while

Y3 = {11 e A2 I {d,e} s P20i)9XQi) = 0},

so that |T| = 1 + 22 + 77 = 100. If we assume that the base vertex \i
in the suborbit diagram of £ is in Ti, then besides 11 the set T contains
the 22-orbit of Gi(co,fi) = Mat22 in the neighbourhood of \x and the
77-orbit at distance 2 from \i. The vectors in Y2 are identified with
the elements in 9 \ {d, e] while that in Y3 with the octads containing
{d,e}; the adjacency relation is via inclusion. Thus (Y2, Y3) is the residual
Steiner system 5(3,6,22). By (2.9.3) and the remark after its proof the
stabilizer of \i in the automorphism group of the subgraph in A2 induced
by T is isomorphic to Li[d,e] = Aut Mat22. It is not difficult to show
that the elementwise stabilizer of Y in G is trivial and we suggest this as
an exercise. Accepting this fact, by (4.10.14) we obtain the following.

Lemma 4.10.15 If I is the length of the orbit of coe under G(co^) then
8
r 11170

and the equality holds if and only if G(cdd, cbe) is transitive on the set Y of
size 100. •
Let D, coD, cbD be as in (4.10.11) and suppose that D = {d,e,f}
so that d e D. If X = cod - coD then N(X) = (84 2 0 21 ), P8(A) = {d},
P4(X) = {e9f}9 X(k) = 0. Hence if v e A^ with P4(v) = {d, e] and X(v) = 0,
then X — 2v e A*9 which shows that 1 e A2 and hence a>/> € A3((bd).
Using the description of the sets *F(d) and <D(D) it is straightforward
to check that Y := A2 n A(&d) n A(G>D) is of size 4 (in particular coe
and a>D are in different G(d)^)-orbits), consisting of the images of vectors
IM, 1 < i < 4, such that \ii e A$ with P4(m) = {d,e}9 X(JJLI) = 0,
Vi = cod-Hu 1*3 = (OD+HU IH = cod-H3. In fact {/JLU/M} = Hfi(d)n9i(D)9
{^2^3} = ^i(d) n 0>2(D). Since ]X\ + Jx2 = o>^, the elementwise stabilizer
K of Y in G is contained in G(cbd) and by (4.10.12 (ii)) K ^ I74(3). The
subgraph in A2 induced by Y has two edges {jui, JU4} and {£2, £3}, so
its automorphism group is D%. Furthermore, if G contains an element h
which induces on Y an element of order 4 then h permutes (&d and a>x>.
By an obvious generalization of (4.10.14) we obtain the following.
 4.11 The diagram of A 189

Lemma 4.10.16 If m is the length of the orbit of COD under G(cbd) then
\Co3\
m> THQ := = 37950.
4-11/4(3)1
•
Since coe and cop are in different G(cod)-otbits and p\3 = k + wo where
/o and mo are as in (4.10.15) and (4.10.16), we have the following.

Lemma 4.10.17 If (o € A3 then G(co) has two orbits Q\ and O2 on A3 n

A2(a>d) and the following hold, where Y = A2 n A2(ct>) n A2(^/):
(i) 1/*/ G Qi tn^n |T| = 100, G[c&,rj] induces the full automorphism
group of the subgraph in A3 induced by T and G(co, fj) is the Higman-
Sims sporadic group HS of order 29 • 3 2 • 5 3 • 7 • 11,
(ii) iffjeQ2 then Y is of size 4 and G(a>9Fj) - t/4(3).22. •
We consider (i) in the above lemma as a definition of the Higman-Sims
group, the subgraph on 100 vertices is known as the Higman-Sims graph.

4.11 The diagram of A

In this section we study the subgraph A in A2 induced by the neighbours
of the zero element. In other terms A is a graph on A2 in which v and
ft are adjacent exactly when v — Ji e A2. We call A the shortest vector
graph. We use the suborbit diagram of the affine Leech graph to deduce
the suborbit digram of A with respect to the action of G, as given below.
275+2025

47104 ^ A3(juo)

46 575 ^ A4(Jk)

2464
190 Conway groups
We follow notation introduced before (4.9.5): m> G A|, F = G(po) =
Co2, X(fio) = 0, {a, b) = P4(/4)). For v G A = A2 we put A->(v) = AnA/(v).
Directly from (4.10.1), (4.10.12 (iii)), (4.10.6) and (4.10.17) we obtain
the following.

Lemma 4.11.1 For j = 2, 3 and 4 the action of F on A-'(po) is transitive

with stabilizers isomorphic to Ue(2), McL and 2 10 : Aut Main, respectively.
•
Let pi G A2(po), p2 = Po + Pi, T = {po,pi,p 2 } and f = T U {0}.
Then by (4.10.8) T is a triangle in A and it will be said to be a singular
triangle. Thus every edge of A is in a unique singular triangle and the
set of all singular triangles is clearly preserved by G. In terms introduced
before (4.9.5) we have the following.

A4(Po) := A2(0) U A4O10) = (A2G10) n A4(6)) + Po = ©+ Po-

Let S = 0(po)n®(pi). Then by (4.10.4) we have S = A 4 (0)nA 2 (T* \{0})
and hence H + p, = A^fii)n A 2 (T* \ {p,-}) for i = 0, 1 and 2. Thus in view
of (4.10.5) and (4.10.8) we have the following

Lemma 4.11.2 Let pi G A2(po). Then

(i) A 2 (p0 n A4(Po) = S + Po (o/siz^ 891),
(ii) A2(po) n A 2 (pi) = {p2} U (S + p 2 ) where p 2 = po + Pi- a

Lemma 4.11.3 Let rj G A3(po). Then

(i) F(f/) = McL acts transitively on the set A2(po) n A2(f/) o/size 275
wif/i stabilizer isomorphic to 1/4(3),
(ii) t/ie subgraph in A induced by A2(po) n A2(fy) is t/ie complement of
the McLaughlin graph (which is connected of valency 162J,
(iii) i/pi G A2(po) then F(pi) acts transitively on the set A 3 (po)nA 2 (pi)
of size 2816.

Proof. Let a> G A3, po, ^ € A2 n A2(a>). Then po — i/ G A2 if and only

if rj is at distance 2 from po in the graph S defined after (4.10.10). In
the latter case a> + po G A3(po), rj + po G A2(po) n A2(G> + po) and (i)
is immediate from (4.10.12), while (ii) follows from the definition of the
McLaughlin graph, given in the proof of (4.10.12). Finally (iii) is a direct
consequence of (i). •

Lemma 4.11.4 Let v G A4(po). Then

 4.11 The diagram of A 191
10
(i) for j = 2, 3 and 4 tfie subgroup F(v) = 2 : AutMat22 acts
transitively on the set A2(po) n A->(v) of size 88, 2048 and 2464
with stabilizer isomorphic to 2 9 : L3(4), Ma*22 and 2 5 : 2 4 : Sym^,
respectively,
(ii) r/ie subgraph induced by A2(po) n A2(v) is connected,
(iii) euery p € A2(po) w adjacent to a vertex from A2(po) n A2(v).
Proof, (i) For 7 = 2 the result is by (4.10.3). We assume that v = Io+po-
Then A4(po)nA2(v) = (®nA!j)+po and the case j = 4 follows from (4.9.5)
and the table therein. With v as above let co € A\ withP 3 (G>) ^ PA{IM)) and
X(co) = P3(co). Then cb € A2(v) n A3(jtio) and F(v,co) = Li{a,b) S Maf22
and the case j = 3 follows. The assertion (ii) is immediate from (4.10.3),
its proof and the diagram there. If Ji\ € A2(/2o) n A2(v), then by (4.11.2)
and (4.10.5) F(v,/ii) = 2 9 : L3(4) has three orbits, say Q 2 , Q 3 , ^4, on
A2(/io) n A2(jui) \ {po + pi} with lengths 42, 512, 336. By the diagram in
(4.10.3) and the divisibility condition we have Q& <z A2(po) n Ak(v) for
k = 2, 3 and 4, so that (iii) follows. •

Lemma 4.11.5 Let fj e A3(po). Then F(fj) = McL acting on A3(po)nA2(f/)

has two orbits: (Afe(po) Pi A2(f7)) + fj for k = 2 and 4 with lengths 275 and
2025 and stabilizers isomorphic to 1/4(3) and Matii, respectively.
Proof. By (4.11.4) A4(po) n A2(f/) is of size

2025 = 2048 • |A4(Po)| • l A 3 ^ " 1 .

If 9 is the quadratic form as in (4.5.6) then for a e A we have 0(po, a) = 1
if and only if a € A3(po). Hence every singular triangle which intersects
A3(po) must intersect it in two vertices. Since 275 + 2025 = 5|A2(fy)|, the
result follows. D
Now by (4.11.1), (4.11.2), (4.11.3), (4.11.4) and (4.11.5) we have the
complete suborbit diagram of A. In what follows we will make use of the
following.

Lemma 4.11.6 Let fj e A3(po), p € A2(po) n A2(fy). Then F(vj,p) S l/3(4)

acting on A4(po) H A2(fj) has three orbits with lengths 162, 567, 1296 and
stabilizers isomorphic to L3(4), 2 4 : Alt^, Alti, respectively.
Proof. It follows from the paragraph after (4.10.10) that if v e
A 4 (po)nA 2 0J) then G(f/,po,v) ^ Mat12 has three orbits on A2(po) n A2(fj)
with lengths 22, 77 and 176. Hence the result follows from the obvious
duality. D
192 Conway groups
The following useful statement can be checked directly.

Lemma 4.11.7 Let \i € A2 and ji =fc p®. Then Ji e AJ(Jio) for j = 2, 3 and
4 if and only if(fi,fio) = ±2, ±1 and 0, respectively. •

We have deduced the suborbit diagram of A from that of the affine

Leech graph. In a similar way (using (4.11.2), (4.11.3), (4.11.6) and sug-
gesting the reader fill in some minor details) we obtain the following.

Lemma 4.11.8 The subgraph II in A induced by A2(/2o) has the following

suborbit diagram with respect to the action of F — G(JLQ) = C02:
336

•
There is an equivalence relation on the vertex set of II with classes of
size 2 with respect to which two distinct vertices are equivalent if together
with po they form a singular triangle. Let II be the quotient of n with
respect to this equivalence relation. Then immediately from (4.11.8) we
obtain the following.

Lemma 4.11.9 The group F = C02 induces on II a rank 3 action and the
distance diagram ofU is the following:

891
©
U6(2)2 210 : PZL 3 (4) C74(3).22

•
2
Let I T be a graph on A (//o) in which Y\\ and rj2 are adjacent if
f}2 G A4(*/i). Then I I ' and I I are the point graphs of the exceptional
extended dual polar spaces S(Co2 x 2) and <f (C02) with the diagram
 4.12 The simple connectedness of^(Co2) and &(Coi) 193
from (1.13.7) whose full automorphism groups are isomorphic to C02 x 2
and C02, respectively. Notice that the residue of an element of type 4 in
$(Co2 x 2) is the universal (double) cover of the corresponding residue
in ${Coi). The following result was established in [Ron81a].

Lemma 4.11.10 The geometries $(Coi) and S(Co2 x 2) are simply con-
nected and S(Co2 x 2) is the universal 2-cover of $(Co2). •

4.12 The simple connectedness of <g(Co2) and 9(Coi)

In this section we establish the simple connectedness of the P4-geometry
^(Coi) and the 2-simple connectedness of the T4-geometry &(Co\). We
will see in (7.4.8) that ^(Coi) is not 2-simply connected.
Let cp : & -> & be the universal covering of ^ = ^(Coi), F be the
universal completion of the amalgam of maximal parabolics associated
with the action of F = C02 on 2F and let 0 be the collinearity graph of J5".
Then F acts naturally on & and on 0 . Furthermore (the last paragraph
in Section 4.9), cp induces a covering <p\ of 0 onto the collinearity graph
0 = 0(jUo) of <&(Co2) and every triangle from 0 is contractible with
respect to <p\.
Let Sf be the ^(Sp6(2))-subgeometry of &(Co2) as in (4.9.8). By (1.6.4)
¥ is simply connected and hence every connected component Sf of the
preimage of Sf in 3F is isomorphic to Sf^ which implies the following.

Lemma 4.12.1 If&(^) and®{5f) are the subgraphs in 0 and 0 induced

by the point-sets of Sf and SPt respectively, then q>\ induces an isomorphism
ofe(S?) onto Q(ST) and F[®(&)] S F [ 0 ( ^ ) ] S 2^ 8 .5p 6 (2). •

Let S = 0(po)n0(/2i) be the subgraph in 0 as in (4.10.5) (so that S is

isomorphic to the dual polar graph of Ut(2)) and assume (the proof of
(4.10.6)) that Q := S n 0 ( ^ ) is a quad in S. This means that the subgraph
in S induced by Q is the point graph of the generalized quadrangle of
order (2,4). By (4.10.8) F[E] ^ (76(2).2 (the extension of l/6(2) by a field
automorphism). Since the dual polar space of Ue(2) is simply connected
by (1.6.4), we have the following.

Lemma 4.12.2 Let Ebe a connected component of the preimage ofE in 0

such that Q : = S n 0 ( 5 ^ ) is non-empty. Then
(i) cp\ induces an isomorphism ofE onto S,
194 Conway groups
(ii) Q is a quad in S,
(iii) F[E] s F[E] s l/6(2).2. D

Let x be a vertex of ©contained in Q (we can assume that x = AQ, SO

that F(x) = G{ n F ^ 2 10 : Aut Mat 22 ). Then by (4.10.5) 02(F(x) n F[S])
has index 2 in 02(F(x)), which shows the following.

Lemma 4.12.3 The graph II as in (4.11.9) is a graph on the set of images

of E under F in which two such images Ef and E" are adjacent if they are
distinct and there is y G Er n S" such that E" is the image of E' under an
element from C>2(F(y)). •

Let II be a graph on the set of images of E (as in (4.12.2)) under F

in which two such images Ef and S" are adjacent if they are distinct
and there is y G E' n E" such that E" is the image of Ef under an
element from 02{F(y)). By (4.12.2) and the paragraph before (4.12.3) the
valency of IT is the number of vertices in S (which is 891) and hence cp
induces a covering cpi of II onto II. We are going to show that cpi is an
isomorphism.
The definition of II in (4.12.3) shows that there is a bijection a from
the set of neighbours of S in II and the vertex set of S such that
<r(S') = y if and only if S' is the image of S under an element from
O2(F(y)). Comparing the suborbit diagram of S in (4.10.5) and the
suborbit diagram of II in (4.11.9), we obtain the following.

Lemma 4.12.4 Let E' and E" be distinct vertices adjacent to E in II. Then
Er and E" are adjacent in II if and only if the distance between c(S') and
a(E") in E is 1 or 2. •

Let ®(S?) and @(S?) be as in (4.12.1) while Q and Q are as in (4.12.2).

Let E and E be the subgraphs of II and II induced by the images of S
and S under F [&(£?)] and F[©(^)], respectively.

Lemma 4.12.5 The following assertions hold:

(i) both Z and £ are cliques of size 28,

(ii) every triangle in U is contractible with respect to q>2.

Proof. In terms of (4.8.7) and (4.10.6) F[&(^)] is the stabilizer in

D ^ 2}+8.Q;j"(2) of a non-isotropic 1-subspace (fio) in Vi while F[E]nF[Q]
 4.12 The simple connectedness of&(Co2) and &(Coi) 195
is the stabilizer of a minus 2-space (/io,Pi)- Hence
|Z| = [Sp6(2) : U4(2).2] = 28

and F[&(^)] acts on 2 doubly transitively. Furthermore, if x E

then F [&(£?)] D F(x) contains a Sylow 2-subgroup of F(x) and hence
it contains O2(F(x)). This implies that c~ 1 ( x ) *s contained in S and the
latter is a clique. By (4.12.1) and (4.12.2) £ is a clique isomorphic to £
and (i) follows. It is easy to see that the restriction of o to 2 \ {3} is
a bijection onto Q. In view of this observation (ii) follows from (i) and
(4.12.4). •
In the next chapter we will make use of the following result.

Lemma 4.12.6 Let S = F[I] ^ 2^s.Sp6(2) (the stabilizer in F ^ Co2 of

a &(Sp6(2))-subgeometry from &(Co2)). Then
(i) S has three orbits: £, Ei and E2 on the vertex set ofH with lengths
28, 2016 and 256, respectively, where Si is the set of vertices at
distance 1 from Z,
(ii) O2(S) acts transitively on Z 2 with kernel Z(O2(S)) so that the sta-
bilizer is of the form 2.Spe(2).
Proof. Clearly £ is an orbit of S. If y € E then the set S \ {y} corre-
sponds to the points incident to a plane n in the residue of y in S(Co2)
isomorphic to ^(Ue(2)). Let u € n(j;)\£, so that u corresponds to a point
outside 7i. It is well known and easy to check that the stabilizer of n in
Ue(2).2 permutes transitively the points outside n. This shows that S is
transitive on the set £1 of vertices at distance 1 from E. One can see from
the intersection diagram of the collinearity graph S of ^(Ue(2)) (4.10.5)
that there are exactly 1 vertex in n adjacent to u in S and 10 vertices at
distance 2. By (4.12.4) this implies that u is adjacent to exactly 12 vertices
from 2 and hence |Zi| = 2016. Thus in order to prove (i) it remains to
show that S is transitive on the set Z2 := n \ (2 U Di) of size 256. By
(4.11.4) Fi ^ 2lo.Aut Mat22 has 3 orbits on II with lengths 44, 1024 and
1232. If we assume that the element of type 1 stabilized by Fi is contained
in the subgeometry stabilized by S, then T := Fi n S ~ 210.24Sym6 and
it is easy to see (compare the diagram Dh(Mat22)) that T has 6 orbits on
II with lengths 12, 32, 1024, 16, 256 and 960. In view of the above this
immediately shows that L2 is an 5-orbit. Since Oi(S) = ([O2(T)9Q],Q)
where Q is a complement in O2(T) to O2(F\), (ii) is easy to deduce
from (4.11.4). •
196 Conway groups
Below we present the orbit diagram of II with respect to the action
ofS.

2l++*.U4(2).2 [29]Sp4(2)

The following result was proved in [Sh92].

Proposition 4.12.7 The geometry ^(Coi) is simply connected.

Proof. By (4.12.5 (ii)) it is sufficient to show that II is triangulable.
Let a G II, p e n 2 (a); we put <£ = II(a) n U(p) and identify 0> with the
subgraph in II induced by this set. Then by (4.11.9) O and II(a)\O are the
orbits of F(a, /?) = l/4(3).2 with lengths 324 and 567, respectively. Since
the diameter of II is 2, in order to apply (1.14.1) all we have to show is
that O is connected and that every vertex from II(a) \ Q> is adjacent to a
vertex from O. Let y e O and assume that y = S. Then by (4.12.4) <r(a)
and G(P) are at distance 3 in S and 5 G H(y) is contained in O if and
only if G(S) is at distance 1 or 2 from both o(<x) and o(P). For a pair
x, y of vertices at distance k in S let p^- denote the number of vertices
at distance i from x and at distance j from y. The numbers p^ can be
calculated from the intersection parameters of S in (4.10.5) [BI84]. In
view of the above
<D(y) = p\2 + plx + pl2 = 21 + 21 + 105 = 147.
Since the valency of O is larger than one third of its size, there are at most
two connected components. By (4.11.3 (ii)) and (4.11.8) 02(F(oc,p)) ^
1/4(3) (which is the only index 2 subgroup in F(a, /?)) has two orbits, Oi
and Q>2, on O of length 162 each and the action on O, is similar to the
action of 1/4(3) on the set of vertices at distance 2 from a given vertex in
the McLaughlin graph. Since the valency of O is 147, by (4.10.13) Q>i can
not be a connected component of Q> and hence the latter is connected.
Since the valency of O is less than the valency of the subgraph induced
by II(a) and F(a, /?) is transitive on II(a) \ O, every vertex from the latter
set is adjacent to a vertex from O and the result follows. D
Let us turn to the simple connectedness question for ^ = &(Coi). Let
xp :<&-+<& be the universal covering, G be the universal completion of
the amalgam of maximal parabolics corresponding to the action of G on
^ and F be the coUinearity graph of <S. Then by the remark in the last
 4.12 The simple connectedness of<g(Co2) and <&(Co{) 197
paragraph of Section 4.9 \p induces a covering \p\ of T onto the Leech
graph F and the triangles in Y are contractible with respect to \p\.
Let ©= ®(/2o) be the subgraph in T defined after the proof of (4.9.4)
and let H = ®(po) n ®(/2i) be as in (4.10.5). Since ©is the collinearity
graph of the geometry ^(Co2) which is simply connected by (4.12.7), we
have the following.

Lemma 4.12.8 Let & be a connected component of the preimage of ©in

F. Then the restriction £ of xp\ to © is an isomorphism onto © and if
S = (THE) then G[0] ^ G[@] ^ Co2 and G[S\ ^ G[H] ^ l/6(2).S);m3. D

If F = G[0] then F[B] s l/6(2).2 has index 3 in G[E] ^ C76(2).S3>m3

and hence we have the following.

Lemma 4.12.9 The shortest vector graph A is the graph on the set of images
of ©under G in which two such images are adjacent if their intersection
is an image o / S . •

Let A be the graph on the set of images of © under G in which

two images are adjacent if their intersection is an image of S. Then
by (4.12.8) and the remark before (4.12.9) the valency of A is twice the
number (which is 2300) of images of S under F. Hence A and A have
the same valency 4600 and xp induces a covering xp2 of A onto A.
We assume that !o £ a and let Ao be the preimage of lo in a. Let *F
and *F be the subgraphs in A and A induced by the images of © and
©containing lo and >lo, respectively. Since G[®] is transitive on ©and
G[0] is transitive on ©, *F and *F consist of the images of ©and ©under
Gi = G(I 0 ) and G(I 0 )^This shows that |*F| = \¥\ = 552 and hence \p2
induces a bijection of *F onto XF. Furthermore, if ©',©" € *F and ©'n ©"
is an image of S, then, since S is connected, the preimages of ©' and ©"
in *¥ intersect in an image of S and hence xp2 induces an isomorphism
of *F onto XF. One can see from the suborbit diagram of A that F has
two orbits on the set of triangles in A (one of the orbits is formed by
the singular triangles). On the other hand *F is just the subgraph in A
induced by A^ and from its suborbit diagram in (4.10.3) we observe that
*F contains both singular and non-singular triangles, which gives

Lemma 4.12.10 Every triangle in A is contractible with respect to \p2. •

Proposition 4.12.11 The geometry ^(Co{) is 2-simply connected.

198 Conway groups
Proof. By (1.6.4) and (3.3.11) every residue in ^(Co\) is 2-simply
connected and hence we only have to show that ^{Co\) is simply con-
nected. By (4.12.10) in order to prove the simple connectedness of ^(Co\)
it is sufficient to show that A is triangulable. We apply (a version of)
(1.14.1). If Ji is at distance 2 from po in A, then the subgraph induced by
A20*o) n A2(/2) is connected by (4.11.3 (ii)) if £ € A3(p0) and by (4.11.4
(ii)) if p G A4(/2o). Hence the quadrangles in A are triangulable. Let
C = {po,pi,/i2,/23,/i4} be a non-degenerate 5-cycle in A. We say that C
is of type 1 if fy G A4(p;+2) for some 0 < i < 4 (where addition of
indices is modulo 5). Suppose that C is of type 1 and assume without
loss of generality that Jii G A4(/2o). In this case by (4.11.4 (iii)) there is a
vertex v G A2(juo)nA2(p2) which is adjacent to JI4. Hence C splits into the
triangle {po,v,j&4} and the quadrangles {jk^ufav}, fefe^v}. Thus
the 5-cycles of type 1 are triangulable. Suppose now that C is not of
type 1 which means that /22,p3 € A3(po). Then by (4.11.5) the third vertex
ju5 = jx2 + ju3 in the singular triangle containing the edge {fi^Ti-i} is con-
tained in Afe(juo) for k = 2 or 4. If k = 2 then C splits into a triangle and
two quadrangles while if k = 4 then C splits into the triangle { f e / ^ P s }
and two 5-cycles of type 1: {foyfiufafalk} and {po9jh,jJi39l*5,Jk} where
JU6 € A2(/2o) n A2(ju5). In any case C is triangulable and so is A. •

4.13 McL geometry

In this section we discuss a geometry related to the Petersen graph and
associated with the McLaughlin group. By (4.11.4 (i)) if rj e A3(/2o) then
G(Jio,rj) = McL acts transitively on the set A2(fy) n A4(Jk) of size 2025
with stabilizer isomorphic to Mat22> We analyse this action in further
detail calculating in the Leech lattice, rather than in the Leech lattice
modulo 2.
Let /io be as above (i.e. m> € A^, X(fio) = 0, ^4(^0) = {a,b}), let
rj £ Al with Pi(ri) = X(rj) = {a}, co = fio + rj, so that co e A3 with
X(a>) = 0, P5(o>) = {*}- Put Mc = G(jM>9ri) = G(jM>,<x>), M{ = Mc n Gi.
Since ^ G A3(po), by (4.11.1), M c ^ G(/io,q) = McL and it is easy to see
that M[ = Li(a,fc) = Mar 22 . Let v0 G A^ with P4(v0) = {a,b}, X(v0) = {a}
so that v0 = Ao - AA) for Ao G A^ with P8(^o) = {b} and X(i 0 ) = 0. Then
v0 G A4(/io) H A2(f7) and hence by (4.11.7) the set Q (of size 2025) of
images of vo under Mc is the following:
Q = {v I v G A 2 ,(v, jio) = O,(v,if) = 2}.
Using (4.4.1) and the table therein it is straightforward to check that M\
 4.13 McL geometry 199

has the following three orbits on ft \ {vo}:

Qx = {v | v e A i X ( v ) = P2(v) n {a,b} = 0},

Q2 = {v | v e A22,{a,b} s P2(v),I(v) = {a,c},c + b\

Q3 = {v | v e Aiv(fl) = -l,v(fc) = l,X(v) c Pi(v),|X(v)| = 7}.

Then v € Q* can be identified with the octet P2OO if i = 1, with the hexad
^2(v) \ {«,fc} having a distinguished element X(v) \ {a} if 1 = 2 and with
the hyperoval X(v) \ [a] in the residual projective plane Il(P3(v) U {a,b})
if i = 3.
Let Q also denote the graph on this set which is invariant under the
action of Mc and such that fi(vo) = Qi. In other terms two vectors from Q
are adjacent if they are perpendicular. Then straightforward calculations
using the diagrams for ^f (Mat2i) in Section 3.9 show that the suborbit
diagram of Q with respect to the action of Mc is the following:

2 4 : Alt5

Two vertices in Qi are adjacent if the corresponding octets are disjoint.

Hence the subgraph induced by fii is isomorphic to the derived graph
A(&(Mat22)) of the P3-geometry ^(Matri). Thus the elements of type 1,
2 and 3 in ^(Matu) are identified with the Petersen subgraphs, edges
and vertices in the subgraph induced by Qi so that the incidence relation
is via inclusion and M\ is a flag-transitive automorphism group of
). This means that M\ — Mc(vo) acts transitively on the set of
Petersen subgraphs in fii = Q(vo) and if ©1 is such a subgraph then
M\ [©1] = 2 4 : Sym$. We are going to construct in Q a family of locally
Petersen subgraphs.
200 Conway groups
Recall that the complement T(n) of the triangular graph of order n is
a graph on the set of 2-element subsets of an n-element set in which two
such subsets are adjacent if they are disjoint. The Petersen graph is T(5)
while f (6) is the point graph of ^(Sp4(2)). It is easy to see that t(n)
is locally T(n — 2), in particular T(7) is locally Petersen. The incidence
system J^(Alt-j) whose elements are the triangles, edges and vertices in
T(l) with the incidence relation via inclusion is a geometry with the
following diagram:

The groups Altj and Sym-j are flag-transitive automorphism groups of

34?(Alti) with vertex stabilizers isomorphic to Syms and Syms x 2, respec-
tively.
Consider rj' e \ \ with {a,b} n P2(rjf) = {b}, X(rjf) = P2(rj') and put
ca' = ^ + nf. Then co' e A43 with P4(co') = {a}, b e P2(co') and X(co') =
P2(co') \ {b}. Let Q' be a graph on

in which two vectors are adjacent if they are perpendicular.

It follows from (4.4.1) and its proof that Gi^o,*?') is a semidirect
product of the subgroup in <2i formed by the elements ?(Y, 1) for the
Golay sets Y disjoint from P4((o')UP2(cof) and of Li(a9b)nLi[P2(co')] =
Alt7.

Lemma 4.13.1 Let 0 ' = {// \ yl e A^,X(^) = P4(fi') s X(co')}. Then

(i) the subgraph in Q! induced by & is isomorphic to T(7),
(ii) GiC/io,^') induces on & an action isomorphic to Altj with kernel
24.

Proof. The mapping £ : // \-> P4{n') establishes a bijection of & onto

the set of 2-element subsets of the set X(cof) of size 7. Furthermore,
(n\li") = 0 if and only if £(//) n £(ti") = 0, which gives (i). The assertion
(ii) is immediate by the paragraph before the lemma. •

Since (/Jo,*/') = (Mo,*/) = —1> by (4.11.7) there is an element g 6

G(juo) which maps rjf onto rj. Clearly such a g maps Qr onto Q. Since
Mc = G(fio,rj) acts transitively on the vertex set of Q, the element g
can be chosen in such a way that vo is contained in the image © of
®' under g. By (4.13.1) ® = f (7) which is locally Petersen. Hence
Mc[& PiQi] ^ 2 4 : Sym5 and by (4.13.1) this is exactly the stabilizer in
 4.13 McL geometry 201
Gi(fio,ri') of a vertex of 0'. This means that G\(^Y\')g = 24 : Alt-j is the
full stabilizer of ©in M c.
Define &(McL) to be an incidence system of rank 4 whose elements
of type 1 are the images of 0 under Mc, elements of type 2, 3 and
4 are the triangles, edges and vertices of Q and the incidence relation
is via inclusion. Then the residue of vo is isomorphic to ^(Matii) and
the residue of ©is isomorphic to &(Alti). Since the incidence relation
is via inclusion, &(McL) belongs to a string diagram and we have the
following.

Lemma 4.13.2 ^(McL) is a geometry with the diagram

P
2 2 1 1
c
and M induces on ^(McL) aflag-transitiveaction. •
The geometry <&(McL) was first constructed in an unpublished work
of A. Neumaier (see [Bue85], [ISh88]). It has been shown in [BIP98] that
&(McL) is simply connected and that it possesses a 2-cover associated
with a non-split extension 323 • McL.
Let us define in <&(McL) a subgeometry. By (4.11.6) a subgroup 174(3) in
Mc = McL has three orbits on Q with lengths 162, 567 and 1296. We are
interested in the orbit of length 567 and particularly in the subgraph in
Q induced by this orbit. Consider /ii € \\ with X(m) = 0, {a,b} c P2(^i)
and put U = G(co9fio,fii), U\ = UnGi. Since fi\ e A2(no) D A2(rj),
by (4.10.12 (ii)) we have U = l/4(3). It is easy to see that U\ is the
stabilizer in M{ = Li(a,b) = Mat^i of the hexad Pi{ii\) \ {a,b}, so that
U\ = 24 : Alts (which is the stabilizer in M\ of the subgeometry J as in
(3.4.4)), in particular the set *F of images of vo under U is of size 567
and in fact

We calculate the orbits of U\ on *F \ {/lo} in the way we have calculated

the orbits of M\ on Q \ {XQ} and obtain the following four orbits:

¥ 2 = {v I v e Q2,P2(v) n p 2 on) = {b}}9

= {v I v e Q2, |P2(v) n p 2 0n)| = 3,X(v) s P 2 ( W )},

^4 = {v I v G Q3,P3(v)nP2(^i) = {b}},
202 Conway groups
of length 30, 96,120, 320, respectively. Having the suborbits it is straight-
forward to calculate the suborbit diagram of the subgraph in Q induced

Notice that *Fi is the set of vertices in the derived graph of

incident to the edges (elements of type 2) contained in the
subgeometry 2L stabilized by U\.
Let £ be the sextet used in the definition of the element © of type
1 in &(McL) and assume that /M^i) = S\U S2. Then the subgraph
T induced by ® n *F is the point graph of the generalized quadrangle
^(Sp4(2)). Let ^(l/ 4 (3)) be a geometry whose elements of type 1, 2 and
3 are the vertices of *F, the triangles in *F and images of T under U,
respectively; the incidence relation is via inclusion. Then ^(1/4(3)) is a
GAB, a (geometry which is almost a building) described in [Kan81] with
the following diagram:

The residue of {Ao} (as an element of type 3) can be identified with

the subgeometry 1 in ^(Matu) and the residue of T (as an element
of type 1) is the geometry of vertices and triangles in T. The universal
cover of #(t/ 4 (3)) is an infinite affine building ([Kan81], [Ti82]). An outer
automorphism of 174(3) performs a diagram automorphism of ^(1/4(3))
permuting the sets of elements of type 1 and 3.
There is another geometry associated with *F. Let Q = 02(1/1), 0>i be
an orbit (of length 16) of Q in ¥2* $>i be the set (of size 10) of vertices in
*Fi adjacent to vertices in Oi and €> = {Ao} U <X>i U O2. Then the subgraph
induced by <D is the Schlafli graph, i.e. the line graph of the generalized
quadrangle of order (2,4). The geometry <f( 1/4(3)) whose elements are the
images of O under U, the triangles and the vertices of *F is the extended
 4.14 Geometries of 3 • C74(3) 203
dual polar space with the diagram

o-
1 4 2

as in (1.13.7) with the automorphism group of the form l/4(3).22.

4.14 Geometries of 3 l/4(3)

Let Jtf = ^f(Coi) and ^ = ^(Coi) be the maximal parabolic and
tilde geometries of G = Coi, respectively. Let {yi | 1 < i < 4} and
{x, | 1 < i < 4} be maximal flags in J f and ^ , respectively, such that
*i = 3>i = ^o> *2 = )>2, *3 c )>3> *4 <= y* (here the yt and the x, are
considered as vertices, lines and *-closed cliques in the Leech graph). Put
Hi = G(yt) and G,- = G(xt), Qt = O2(Gi) for 1 < i < 4, so that Gi = fflf
G2 = H2, G3 < ff3, G4 < H4 (Section 4.9).

Lemma 4.14.1 The group G\ = 2 11 : Mat2A contains two conjugacy classes

of subgroups of order 3 with representatives Xs and Xt such that
(i) NGl(Xs) ^ 2 4+1 .3 • Sym6 and iVGl(Xs)O2(ifi n H 2 ) = Hx n if2,
(ii) N Gl (X t ) S 23.(5>;m3 x L3(2)) anrf N G l (X t )0 2 (H! n H 3 ) = Hi n H 3 ,
(iii) X s and Xr are not conjugate in G.

Proof, (i) and (ii) follow from (2.12.3), (2.13.3) and (3.8.2) with the
remark that Xs and Xt map, respectively, onto 3a- and 3b-subgroups in
Gi = MatiA- The elements of order 3 in 3a- and 3fc-subgroups in the
complement L\ = Mat24 have different characters in the 24-dimensional
real representation of G = Coo in the vector space containing the Leech
lattice, which implies (iii). •
By the above lemma and (2.13.1) for a = s and t ATG(Xa) acts transitively
on the set O(Xa) of vertices in the Leech graph fixed by Xa.
The parabolic G2 = 2 4+12 .(3 • Syme x Sym^) induces 3 • Sym^ on
resj(x 2 ) = ^(3 • Sym6) and the kernel G^~ is of order 2 17 • 3. Let X
be a Sylow 3-subgroup of G^~.

Lemma 4.14.2 NGl(X)/X ^ 2 4+1 .3 • Sym6.

Proof. By the Frattini argument NQ2{X)Q2 = G2 and hence all

we have to show is that CQ2(X) is the natural symplectic module for
Gj/O 2 , 3 (Gj) ^ Sp4(2). Since [Q{ : Q{ n Q2] = 2; g i is the irreducible
Golay code module ^ n for Gi = Mat24 and Gi 2 Qi/Qi is the stabilizer of
204 Conway groups
a sextet in Gi, we conclude by (3.8.4) that there are three chief factors of
G2/Q2 — 3 • Sym$ = 3 • Sp4(2) inside Q2; two isomorphic to the hexacode
module and one to the natural symplectic module. Since X acts faithfully
on Q2 and XQ2/Q2 is in the centre of O2(G2)/Q2 the result follows. •
Since X is contained in G^~, it fixes resj(*2) elementwise. Let u be an
element of type j in r e s j f e ) for j = 3 or 4. Then res^(u) is a projective
space of rank j — 2 over GF(2), which contains *2 and hence there is a
unique element xp(u,X) of type j—l incident to u which X fixes pointwise.
Here y)(u,X) is the centralizer of X in the subgroup of order 2j which
represents u in A. Furthermore xp(u,X) and tp(t;,X) are incident if and
only if u and v are incident, and we have the following.

Lemma 4.14.3 Let *F = xV(x2, ^0 &e t/ie subgeometry in & formed by the
elements \p(u,X) for all u e resj(x2). Then *¥ is isomorphic to res^(*2)
(i.e. to the rank 2 tilde geometry &(3 • Sym^)) and NG2(X) induces on *F
its full automorphism group with kernel of order 2 5 . •
If w is an element of type 1 in *F, then by the above lemma G(w) n
NG2(X) has index 45 in NG2(X) and since G(w) is a conjugate of Gu by
(4.14.1) we have the following.

Lemma 4.14.4 There is an element g e G which conjugates X onto Xs and

maps x¥(x2,X) onto a subgraph *F(;y,Xs) contained in O(XS), isomorphic to
the point graph of&(3 • Sym^) (here y = x\). •
Let us look more closely at the subgraph <J>(XS) in the Leech graph
induced by the set of vertices fixed by Xs. We identify Xs with a subgroup
of type 3a in the complement L\ = Ma£24 to Q\ in G\. Concerning the
action of Xs on 9 we follow the notation introduced in Section 4.9.
Recall that Q denotes the mapping of r ( I 0 ) = Aj 1 onto the set of sextets
which commutes with the action of G\.

Lemma 4.14.5
(i) The subgraph O(XS) is of valency 32; O(I s )nAJ f l = O 0 UOi where
Q(OO) = 2 and Q(®I) = {S y | 1 < i < j < 6} (in terms of'{212.1)),
(ii) NG{(XS) acts transitively on Q>\.
Proof. Directly from (2.12.7) we obtain (i) together with the remark
that Q>j consists of one or two ATG1(Xs)-orbits for 7 = 0 and 1. Without
loss of generality we assume that the subgraph ^(>;,XS) as in (4.14.4)
contains IQ- Since NG(y)(Xs) induces the (flag-transitive) automorphism
 4.14 Geometries of 3 • £/4(3) 205
group of the rank 2 tilde geometry associated with ^(y, X), we obtain
(ii) together with the remark that ^ ( y , X ) n A^a c o 1# •
By the remark after (4.14.1) NG(XS) is transitive on (the vertex set of)
<D(X5) and by (4.14.5) it has two orbits, say Qo and Qi, on the set of
lines contained in O(XS) with |Qi| = 15 • |Q 0 |. Although O2(NG(XS)/XS)
is still to be identified with 3 • 1/4(3), let ^(3 • 1/4(3)) denote the incidence
system of rank 3 whose elements of type 1, 2 and 3 are the vertices in
O(XS), the lines from Qi and the images ofx¥(y,Xs) under NG(XS), with
the incidence relation via inclusion.

Lemma 4.14.6 ^(3 • £/4(3)) is a geometry with the diagram

2 2 2
and NGI(XS) induces on it a flag-transitive action with kernel Xs.
Proof. The group NG^XS) acts on the set g(®i) of lines from Qi
incident to lo as Sym^ = Sp$(2) acts on the point set of ^(Sp4(2)). Then
a subset of size 3 in Q(Q>\) is a line of ^{Sp^(2)) if and only if its stabilizer
in NGI(XS) has order 2 9 • 3. On the other hand ^(y^s) contains three
lines from Qi containing Io and by (4.14.3) N G ^ G O O P Q is of order 2 9 • 3.
Since the vertices and the lines (from Qi) contained in *F(y,-3Q form the
rank 2 tilde geometry, the result follows. •
We are going to construct another geometry on 0>(Xs). Let Bt =
(E \ {pt}) U Tt for some i, 1 < i < 6, so that Bi belongs to the orbit of
length 6 ofNLl(Xs) on the set of octads fixed by Xs. Let d = T(BU 1) be the
corresponding involution in Q\ and let <t>(S) be the set of vertices x in the
Leech graph such that S e 0 2 (G(x)), D = CG(S), R = O2(D), D = D/R
and Vi = CA(R). Then by (4.8.6) and (4.8.7) K s 2}+\ D s Q+(2) and
the subgraph in the Leech graph induced by O(<5) is the point graph of
the parabolic geometry offig"(2).

Lemma 4.14.7 The subgraph in the Leech graph induced by O(<5)

is isomorphic to the Schldfli graph and (D n NG(XS))/XS = 2 x l/4(2).2
induces the full automorphism group of the subgraph.
Proof. Without loss of generality we assume that Xs is contained
in a complement K ^ Alt% ^ L4(2) in CGl(S) ^ 2 1+4+4+6 .L 4 (2). Since
CGl(S) n NGl{Xs) has index 6 in NGl(Xs), we conclude that NK(XS) =
{Sym$ x Symif. This means that Xs acts fixed-point freely on the natural
module of K and has 4-dimensional centralizer in the exterior square
206 Conway groups
of the natural module. By (4.8.6) this means that NR(XS) = (6) and
NG(Xs)nO2(CGl(S)) is of order 2 5 . Now the result follows either by direct
calculation in the Leech graph or by analysing subgroups of order 3 in
Q+(2). •
By the above lemma T := [Xs, Vi] is a minus 2-space in V\ and by the
proof of (4.10.6) T is as in (4.10.4). Hence we have the following ((4.10.5)
and the diagram therein).

Lemma 4.14.8 If T = [Xs, V{\ with T* = {po, pi,fe}' then®(5) n O(Z S)

is a quad in the subgraph S of the Leech graph induced by AiiT) n A4. •
Define S(3 • 1/4(3)) to be the incidence system of rank 3 whose elements
of type 1 are the images of O(<5) n O(XS) under NG(XS), the elements of
type 2 are the lines in the orbit €l\ and the elements of type 3 are the
vertices in O(XS), and the incidence is via inclusion.

Lemma 4.14.9
(i) (f (3 • t/4(3)) is an extended dual polar space with the diagram
c
1 4 2
on which NG(XS) induces a flag-transitive automorphism group,
(ii) NG(XS)/XS ~ 3 • L/4(3).22,
(iii) S{2> • 1/4(3)) possesses a 3-fold covering onto the geometry S(U^(3))
constructed at the end of Section 4.13.
Proof. The elements of type 1 and 2 incident to Xo (considered as
an element of type 3) are indexed by the octads Bi9 1 < i < 6, and the
sextets L;fc, 1 < j < k < 6, respectively. Furthermore the element indexed
by Bi and the element indexed by E/fc are incident if and only if i € {./,&}.
Hence (i) follows directly from (4.14.5) and (4.14.7). The classification of
flag-transitive extended dual polar spaces of rank 3 achieved in [DGMP]
and [Yos91] (the table after (1.13.7)) implies that NG(XS)/XS must be
isomorphic either to l/3(4).22 or to 3 • l/4(3).22. By (4.14.3) NG(XS)/XS
contains a subgroup of the form 25.3.Syme and hence the latter possibility
holds, which implies (ii) and (iii). •
Using (4.4.1) it is not difficult to list all the vectors in O(XS) and to
describe the action of NGl(Xs) on the set of these vectors which gives
the suborbit diagram of the action of NG(XS) on O p Q , which earlier
appeared in [Yos92], p. 159. We omit the details of the calculations and
summarize the result in the following.
 4.14 Geometries of 3 • I/4(3) 207
5
Lemma 4.14.10 The group NGl(Xs)/Xs = 2 .Sym6 acting on the set O(XS)
of vectors in the Leech graph fixed by Xs has nine orbits with lengths
1, 2, 30, 60, 96, 192, 320, 360 and 640 which are contained in A^ for
a = 8,4a, 4a, 6,6,5,4c, Ab and 5, respectively. The suborbit diagram of the
graph of valency 30 on 0>(Xs) is as given below. •
5 3+9 5

1+2

Notice that OI(NG(XS)/XS) stabilizes every element of type 3 in

(74(3)) and the geometry whose elements are the orbits of 03(NG(Xs)/Xs)
on the element set of ^(3 • U4(3)) with the induced incidence relation and
type function is isomorphic to Kantor's GAB ^(1/4(3)) discussed at the
end of the previous section.

Lemma 4.14.11 C^(XS) is a 12-dimensional irreducible module for NG(XS)/

Xs, the restriction to this module of the quadratic form 9 is non-zero and
03(NG(XS)/XS) acts on this module fixed-point freely.
Proof. By (3.8.2)-(3.8.4) the centralizers of Xs in the irreducible Golay
code and Todd modules are 5-dimensional. In view of (4.8.2) this shows
that C^(XS) is 12-dimensional. By (4.14.10) 0>(Xs) contains lo and some
vertices from A4, and hence the restriction of 6 to C^(XS) is non-trivial.
In order to check the irreducibility one could for instance determine all
the vectors in A fixed by Xs and describe the orbits of NG(XS) on these
vectors. •
Arguing as in the above lemma it is easy to check that C^(Xt) is
8-dimensional.
Let us consider the intersection of 9(XS) and the point set 0 = ®(/io)
of the subgeometry ^{Coi) in ^. We assume that Xs stabilizes 0, or,
equivalently, that {a, b) := P^{^) is contained in E. The intersection can
208 Conway groups
be calculated either by removing from O(XS) the vertices which are not
contained in ©or by determining the vertices in 0 fixed by X s (where
the latter can be considered as a subgroup in F\).
The result is certainly independent of the particular way of calculating
and is summarized in the following.

Lemma 4.14.12 The following assertions hold:

(i) the group NFl(Xs)/Xs ^ 24.(Sym4 x 2) acting on the set 0 n O ( I s )
has six orbits with lengths 1, 2, 12, 24, 32 and 64 contained in A4
for a = 8,6,4a, 4b, 6 and 5, respectively,
(ii) Np(Xs) preserves on ®nO(X 5 ) an imprimitivity system with classes
of size 3 and the graph obtained by factorizing over this system is
the point graph of the generalized quadrangle of order (2,4),
(iii) NF(XS)/XS = 2 x £/4(2).2 and the suborbit diagram of the subgraph
in the Leech graph induced by ©n O(X S) is as given below. •

1+1

Notice that the non-empty intersections with © of the subgraphs

realizing the elements of type 3 in ^(3 • t/4(3)) are isomorphic to the line
graph of the Petersen graph while the non-empty intersections with ©of
the subgraphs realizing elements of type 1 in <f (3 • 1/4(3)) are isomorphic
to the point graph of the generalized quadrangle of order (2,2).
To the end of the section we discuss thefixedpoints in the Leech graph
of a subgroup of order 7. Let S be a subgroup of order 7 in G\. Then
by (2.13.5) and its proof S stabilizes exactly three elements in ^ , say a,
b and c, and exactly three octads forming a trio, say T = [BuB2,B{\.
Thus without loss of generality we may assume that S < L\ nH^. Since S
centralizes exactly three non-zero elements in the irreducible Golay code
module <2i> namely the ones corresponding to B\9 B2 and £3, we have
the following.
 4.14 Geometries of 3 • l/4(3) 209

Lemma 4.14.13 NG^S) = iVGinH3(S) is the direct product of a subgroup

isomorphic to Sym* and N^ajj^iS) = Frotf. •
Since a Sylow 3-subgroup X in NLl(a,b,c)(S) fixes a, b and c, by (4.14.1)
we can assume that X = Xs.
Using (4.4.1) it is straightforward to find the set S of vertices in the
Leech graph fixed by S and to establish the following result applying
(2.13.1).

Lemma 4.14.14 The following assertions hold:

(i) the set E is of size 14 consisting of Xo> three vertices from A4, six
vertices from Af* and 4 vertices from A4,
(ii) the graph on E in which two vertices v,ju € H are adjacent if
v 6 A4(p) is isomorphic to the point-line incidence graph of the
protective plane of order 2,
(iii) NG(S) induces on E the natural action of AutL^(2) with kernel
NLl(a,b9c)(S) = Frob37, CG(S)/S ^ L3(2) has two orbits on E of
length 1 each, in particular S is fully normalized in G,
(iv) W := C^(5) and W is generated by the vertices from E,
(v) W, as a module for NQ(S), is isomorphic to the direct sum of the
natural module for CQ(S)/S = 1^(2) and its dual; the restriction to
W of the quadratic form 9 from (4.5.6) is non-trivial,
(vi) S is contained in O(XS) where Xs is a Sylow 3-subgroup of
NLl(a,b,c)(S)- •
 5
The Monster

The Monster is the largest among the sporadic simple groups. It was
predicted to exist independently by B. Fischer and R. Griess in 1973
and it was constructed by R. Griess in 1980. According to its standard
definition the Monster is a simple group M which contains a subgroup
Z\ of order 2 such that

which means that Q\ := O2(G\) is an extraspecial group of order 2 25 (i.e.

Z\ = Z(<2i), and Q\/Z\ is elementary abelian of order 224) and such that
G\/Qi = Co\ acts on Q\/Z\ as it acts on the Leech lattice A modulo
2. Since by (4.9.3) the tilde geometry ^(Co\) of the Conway group Co\
possesses a natural representation in A, this means that the elements of
type i of y(Co\) are realized by certain subgroups of order 2 I+1 in Q\
which contain Z\ so that the incidence relation is via inclusion. Let Z2
and Z3 realize incident elements of type 1 and 2, respectively, and put
Gt = NG(Zi\ Qt = O2(Gi), Gt = Gi/Qu i = 2,3. Then G2 n G3 contains a
Sylow 2-subgroup of G\,
G2 = Symi x Mat24, G3 = L3(2) x 3 • Sym6
and
[G2 : G12] = 3, [G3 : G23] = [G3 : G13] = 7, [G3 : G123] = 21
(here as usual G12 = G\ n G2 etc.) In this chapter we study a group G
generated by an amalgam M = {Gi, G2, G3} such that the structure of the
Gt and the intersection indices are as above. We will call Jl the Monster
amalgam. We will show that G acts flag-transitively on a tilde geometry
^(M) of rank 5. Then we construct a number of subgroups of G associ-
ated with certain subgeometries of ^(M). Some of the subgroups involve

210
 5.1 Basic properties 211
other sporadic simple groups. We will determine the structure of these
subgroups and establish the simple connectedness of the corresponding
subgeometries of ^(M). Then we apply the triangulability of a graph
on the set of Baby Monster involutions in the Monster group estab-
lished in [ASeg92] to prove the simple connectedness of ^(M). We start
by studying some basic properties of Jt and constructing the geometry

5.1 Basic properties

We start with a couple of definitions. Let Jl = {Gi,G2,G3} be an
amalgam of rank 3, put Qt = 02{Gi\ G{ = Gi/Qu 1 < i < 3, Gtj = GtOGj,
Ttj = Qt n Qj for 1 < i < j < 3, Gm = GinG2n G3, Zx = Z(6i).

Definition 5.1.1 The amalgam Jl is called the Monster amalgam if the

following hold:
(i) Q\ is an extraspecial group of order 225;
(ii) G\ = Co\ acts on Q\/Z\ as it acts on the Leech lattice A modulo
2;
(iii) G123 contains a Sylow 2-subgroup of G\;
(iv) G2 = Sym3 x Mat24 and G3 = L3(2) x 3 • Sym6;
(v) [G2 : G 12 ] = 3 ; [G 3 : G 23 ] = [G 3 : G 13 ] = 7; [G 3 : G 123 ] = 2 1 ;
(vi) for 1 < i < j < 3, we have Qt n Qj ± Qt-

Notice that condition (iv) can be deduced from the other conditions
together with certain information on subgroups in Co\ and Mat24 con-
taining Sylow 2-subgroups.
In this chapter M is a Monster amalgam and G is a faithful completion
oiM.
Let rj : Q\ -> A be the homomorphism commuting with the action of
G\ whose existence is guaranteed by (5.1.1 (ii)), so that Z\ = ker rj. In
what follows we identify a subgroup in A with the set of its non-trivial
elements. There are a quadratic and the associated bilinear mappings
of Qi/Zi onto Z\ invariant under the action of G\ which are defined,
respectively, by p •—• p2 and (p,q) •—> \p,q] for p,q e Q\. By (4.5.6) there
is a unique Co\ -invariant quadratic form on A (denoted by 9) and hence
we obtain the following.
212 The Monster
Lemma 5.1.2 Let p,q e Q\ \Z\, rj(p) e A,-, rj(q) e A/ and rj(pq) e A*.
Then
(i) p is an involution if i = 2 or 4 and it is an element of order 4 (/*
i = 3,
(ii) p and q commute if and only ifi + j + k is even. •
For a subspace A in A let Ax denote the orthogonal complement of A
with respect to 6. For S < Q\ put
S± = ri-HriiSH
Since 0 is non-singular, when Z i < S we have ±|S| • {S^ = \QX\. By (5.1.2)
we have the following.

Lemma 5.1.3 Let q\ and q2 be different elements of Q\ \ Z\ having the

same image under rj (i.e. S := (quq2) = *\~lQ) for some 1 G A), and
let q € Q\. Then for i = 1 and 2 we have q~xqtq = q\ if q £ S1 and
q~^qiq = qi-\ otherwise. •
Recall that an element of type i in the T-geometry ^(Co\) of the
Conway group Co\ is an i-dimensional subspace A in A such that A cz A4
and A < A1 (compare the definition of ^{Co\) given before (4.9.2)).
Hence f/"1^) is elementary abelian of order 2I+1. We will identify the
elements of y(Co\) with their preimages under rj.

Lemma 5.1.4 Gn is the normalizer in G\ of an element Z^ of type 1 in

Proof. Since [G2 : Gn] = 3, we have Q2 < Gn and by (5.1.1 (iv))

Gn/Qi — 2 x MatiA- Since Gn contains a Sylow 2-subgroup of G\ it
contains Q\. On the other hand 6162/82 is non-trivial by (5.1.1 (vi)).
Then the structure of G12/62 implies that 8182/62 has order 2 and
hence Tyi is of index 2 in 81 • Since 81 is extraspecial, Tn contains Z\.
Since Gyi normalizes Tn, it also normalizes Z2 := T^, where Z 2 has
order 4 and Z\ < Z2. We know that Gn contains a Sylow 2-subgroup of
G\ but by (4.5.5) A4 is the only orbit of Co\ on A# of odd length and
hence r\(Z-i) € A4(i.e. Z2 is an element of type 1 in ^(Co\)). Since Gn
involves a chief factor isomorphic to Mat24 we have Gn = NG1(Z2) and
the result follows. •
Without loss of generality from now on we assume that rj(Z2) = Io =
 5.1 Basic properties 213
Lemma 5.1.5 G13 is the normalizer in G\ of an element Z3 of type 2 in
incident to Z2.

Proof. Since [G3 : G i3 ] = 7, we have Q3 < G i3 and X := Gi 3 /g 3 =

Syni4 x 3 -Sym^. Since Q\ is not contained in Q3, Q1Q3/Q3 is a non-trivial
normal 2-subgroup in X and the structure of X immediately shows that
Q1Q3/Q3 is of order 2 2 . Since G13 contains a Sylow 2-subgroup of G\, it
contains Q\ and hence T13 has index 2 2 in Q\ and it contains Zi, since
Qi is extraspecial. Thus G13 normalizes Z 3 := Tfe which is a subgroup of
order 2 3 in Qi containing Z\. Furthermore, Z 3 is normalized by a Sylow
2-subgroup of G\ which normalizes Z2. By (4.5.5), for i = 2 and 3 the
size of A,- is divisible by 8 and hence Io = n{Zi) < v(z3) ^ A4. Finally,
by (4.6.1) A!\a is the only non-trivial suborbit of Co\ on A4 whose length
is not divisible by 4 and hence rj(Z3) is a triangle in the Leech graph.
Since G13 contains a Sylow 2-subgroup of G\ and has a factor group
isomorphic to Synt3 x 3 • Synt^, we have G13 = NG1(Z3) and the result
follows. •

By (4.5.5) and (5.1.4) we have Gn/Qi = 2 11 : Mat24 (the semidirect

product of the irreducible Golay code module ^ n and the Mathieu group
Mat24). By (4.8.2) Q1/Z1, as a module for Gn/Qu is uniserial containing
a chief factor isomorphic to # n , a chief factor isomorphic to ^ n and two
1-dimensional chief factors, namely Z2/Z1 and Qi/Z^. This implies that
Z2 = Z((>2) and in particular Z2 is normal in G2. In a similar way one
can see that Z3 = Z(<23). Furthermore, since G13 contains Q\, by (5.1.3)
G3 does not centralize Z3 and the structure of G3 immediately implies

Lemma 5.1.6 The following assertions hold:

(i) G2/CG2(Z2) s Sym 3 ; G3/CG3(Z3) ^ L 3 (2);
(ii) G13 = NciZi) = N G l (Z 3 ); G23 = NG3(Z2) = iVG2(Z3);
(iii) Z2 and Z3 are the normal closures ofZ\ in G2 and G3, respectively;
(iv) if{ij9k} = {1,2,3} then Gt = (GiJ9Gik).

Proof. Since both G13 and G23 have index 7 in G3 we obtain (i), (ii)
and (iii) by the arguments before the lemma. In (iv) for i = 1 the result
follows from (5.1.4) and (5.1.5) since the Leech graph is connected; the
remaining cases are immediate from (i) and (ii). •
Let G\ and G^~ be the preimages in G2 of the direct factors Mat24 and
Sym3 of G2, respectively (notice that by (5.1.6) we have G j = CG 2 (Z2)).
214 The Monster
We consider the chief factors of G\ inside Q2. By (4.8.7)

is the unique subgroup in Q\ containing Z2 and normal in G j such that

R2/Z2 is the irreducible Todd module for G2/Q2' The quotient Q2/R2
involves two irreducible chief factors, both isomorphic to the irreducible
Golay code module. Since G j < G2, this shows that R2 < G2. Let X be
a Sylow 3-subgroup of G^. By (5.1.6) X acts on Z2 fixed-point freely
and by the Frattini argument NG2(X)Q2 = G2. By (4.8.7) Rj; = R2 and
hence R2 is elementary abelian. Since G2/Q2 acts irreducibly on R2/Z2,
since the dimension of R2/Z2 is 11 which is an odd number and since
X/Q2 < G2/Q2, X centralizes R2/Z2 and we have the following.

Lemma 5.1.7 If X is a Sylow 3-subgroup of G2, then, as a module for

CQ2(X)> the subgroup R2 possesses the decomposition R2 = Z2 © CR2(X)
where CRl(X) S * n . D

Let us analyse the structure of Q2/R2. Since Q\ is extraspecial, the

commutator subgroup of Z^ = T\2 is exactly Z\. Since X does not
normalize Zi, it does not normalize Z^. If d is a generator of X then
Z2/R2 and (Z2Y/R2 are two abelian normal subgroups in Q2/R2 which
have trivial intersection and factorize Q2/R2' Thus we have the following.

Lemma 5.1.8 Q2/R2 is an elementary abelian 2-group of rank 22 and as a

module for G2 = ^^^13 x Mat24 it is isomorphic to Z2 ® ^11. In particular
CR2(X) = CQ2(X). •

The structure of G2 specified above can be expressed by writing

G2 - 22+n+22.(Sym3 x Mat24).

Lemma 5.1.9 Let Y be a subgroup of order 2 in Z3, such that Y =£ Z\,

and suppose that g € G3 conjugates Z\ to Y. Then

(i) Q\ is independent of the particular choice of g and Z(Q\) = Y,

(ii) ifY <Z2 then Q\ < G2 and Qi n Q\ = R2.

Proof, (i) follows directly from (5.1.6 (i)). If Y < Z2 then by (5.1.6 (i))
the element g can be chosen from G2 in which case Q\ < G2. We know
that on the one hand G2 normalizes R2 and on the other hand Q\ and
Qi(Y) are two extraspecial groups of order 2 25 with different commutator
 5.1 Basic properties 215
subgroups Z\ and Y. Hence their intersection is an elementary abelian 2-
group of rank at most 13, which is exactly the rank of R2. Since R2 < Q\
and Ri is normalized by G2, (ii) follows. •
Let G^" and Gj be the preimages in G3 of the direct factors 3
and L3(2) of G3, respectively. Then by (5.1.6) we have G$ = CGl{Z3) and
hence G3" < Gj. In the next lemma we analyse the chief factors of G3
inside Q3.

Lemma 5.1.10 Let R3 = coreo^Ri) and T3 be the normal closure of R^

in G3. Then
1<Z3<R3<T3<Q3

is the only chief series of G3 inside Q3. Furthermore, Q3/T3 = Z3 ®H

where H is the hexacode module for G$/Q3 ^ 3 • Sym6, T3/R3 ^ Z\ <g> V
where V is the natural symplectic module of G^/O2,3(G^) = Sp4(2) and
R3/Z3 s H* (here U* denotes the dual of U).

Proof. Put S = G\ n G^~, so that S/Q2 = 26 : 3 • Sym6 is the stabilizer

of a sextet in G%/Q2 = Mat2*. We call H := O2(S/Q2) the hexacode
module for S/O2(S) to distinguish it from its dual. By (4.8.7), (3.8.2),
(3.8.3), (3.8.4) there are seven non-trivial chief factors of S inside O2(S),
one isomorphic to H*, three isomorphic to H and three isomorphic to
the natural symplectic module V of S/C>2i3(S) = Sp4(2). In particular
CR2/Z3(O2(S)) is the only chief factor isomorphic to if*, which shows that
the preimage K3 of CR2/z3(O2(S)) in Q3 is normal in G3. By (3.8.4) JR3 does
not split over Z3. Since R3/Z3 is the only chief factor of S isomorphic to
H*, and R3 < R2, R3 is contained in every conjugate of £2 in G3. On the
other hand K3 is a maximal Gi23-submodule in R2 and in view of (5.1.6
(iv)) K2 cannot be normalized by G13, since it is already normalized by
G12, and Gi is irreducible on Q\. Hence #3 = coreG3(^2).
By (5.1.3) Z%- = Qif)G$ and by (4.8.7) Z^/Rj- is isomorphic to H
(which is dual to JR3/Z3) as a G%-module. Hence by (4.8.7) and (3.8.4)
R^/R3 involves exactly two chief factors of G3": ^2/^3 and R3/R2, both
isomorphic to V. Let T3 be the normal closure of £3- in G3. Then all
chief factors of G3" in T3/.R3 are isomorphic to V. Let £ be a Sylow
7-subgroup in G3. Then G3 = (G23,£) and hence E does not normalize
R2. This shows that E acts non-trivially on T3/R3 and on the other
hand it centralizes the action of G3"/O2,3(G3~) on T3/R3. We have noticed
that Q3 involves only three chief factors of Gj isomorphic to V. This
shows that T3/R3 is elementary abelian and that as a module for G3 it
216 The Monster
is isomorphic either to Z3 <g> V or to Z\ <g> V. Since G23 normalizes #2 as
well as the subgroup Z2 of order 4 in Z3 the latter possibility holds. In a
similar way we identify the structure of Q3/T3 (suggesting the reader fill
in the details). •

Thus the structure of G3 as specified above can be expressed in the

following way:

G3 - 2 3+6+12+18 .(L 3 (2) x 3 • Syme).

5.2 The tilde geometry of the Monster

We follow the notation introduced in the previous section. Let A be a
graph (called the first Monster graph) on the set of right cosets of G\
in G in which two such cosets are adjacent if their intersection is a
coset of G12. The group G acts on A by right translations; let vo be the
coset containing the identity (i.e. G\ itself). Then G(vo) = G\ and for
v = Gig we have G(v) = g~lGig. Put Qv = O2(G(v)\ Zv = Z(G(v)). Since
[G2 : G12] = 3, for every coset of G2 in G there are exactly three vertices
in A which intersect it in a coset of G12; furthermore these three vertices
form a triangle called a line. The action of G\ on the set L(vo) of lines
containing vo is similar to its action on the set of cosets of G12 in G\. By
(5.1.4) this means that Q\ is the kernel of the action and G\/Q\ = Co\
acts on L(vo) as it acts on A4. Let

be the line formed by the vertices intersecting G2 in cosets of G12. Then

by (5.1.6) G2 induces on / the natural action of Symi with kernel G3"
and ZVo = Z\, ZVl9 ZV2 are the subgroups of order 2 in Z2. This gives the
following (where as usual A(i?o) denotes the set of vertices of A adjacent
to vo).

Lemma 5.2.1 Let % be the set of subgroups X of order 2 in Q\ such that

rj(X) e A4. Then
(i) the mapping q> \v\-±Zv establishes a bijection of A(t?o) onto 9C,
(ii) a triangle {u,v, w} in A is a line if and only if (ZU,ZV,ZW) is of
order 4. •

By the above lemma the orbits of g i on A(i;0) are of length 2 and such
an orbit together with VQ forms a line. Thus if / = {VQ9V\9V2} is as above,
 5.2 The tilde geometry of the Monster 217

then {vuvi} is an orbit of Q\ and the kernel Q\(vi,V2) is Tn = Z^, so

we have the following.

Lemma 5.2.2 If {u\,U2} is an orbit of Qi on A(t;o) then Q\(u\,U2) =

{(p(u\\cp(u2))L; in particular, kernels at different orbits are different. •

Lemma 5.2.3 Let M, V G A(t;o). Then u and v are adjacent in A if and only
ifrj(p(u) and rj(p(v) are either equal or adjacent vertices of the Leech graph.

Proof. Let Z be the subgraph in A induced by the images of vo under

G3. Then by (5.1.6) \L\ = 7, X contains the line / and G3 induces on E
a doubly transitive action of L$(2). Hence S is a clique. Furthermore,
by (5.1.6 (iii)) and (5.2.1 (i)) {(p(v) \ v G Z} is the set of subgroups of
order 2 in Z3 and by (5.1.5) r\(£i) is a line in the Leech graph. Since G\
acts transitively on the vertices and lines of the Leech graph this proves
the "if" part of the statement. Thus A(VQ) n A(v\) contains qrl(Ri n 3£)
where ,R2 is defined after (5.1.6) and 3T, (p are as in (5.2.1). By (5.2.1 (i)) if
u e A(i>o)nA(i;i) then Zu G QiCiQVl and by (5.1.9 (ii)) the latter intersection
is exactly R2 which proves the "only if" part of the statement. •

Recall that a clique N in A is said to be *-closed if together with every

edge it contains the unique line containing the edge. By (5.2.1) and (5.2.3)
we have the following.

Lemma 5.2.4 Let N be a clique in A containing VQ. Then

(i) rjcp(N \{VQ}) is a clique in the Leech graph,

(ii) N is *-closed if and only if N = cp~lri~l(L) U {i?o} for a ^-closed
clique L in the Leech graph. D

Notice that if AT is a *-closed clique of size 2l — 1 in A containing t?o

then Q\ normalizes the subgroup Z(N) of order 2l in Q\ generated by
r\~l(N) and induces on Z(N) an elementary abelian group of order 2l~l.
By the above we have the following result analogous to (4.8.1).

Lemma 5.2.5 If N is a maximal clique in A, then N is *-closed of size 31

and {(p(u) I u G N} is the set of subgroups of order 2 in an elementary
abelian group Z(N) of order 25 contained in Qwfor every w G N. Further-
more G[N] acts on N as GL(Z(N)) = L$(2) acts on the set of subgroups
of order 2 in Z(N). There are two orbits, Jfv and Jft, of G on the set of
maximal cliques in A with \Jfv\ = 3 • \Jft\ and for a = v or t whenever N
218 The Monster

is a clique from «yfa containing vo we have rjcp(N \ {vo}) G if a , where J£?a

is as in (4.8.1). •

Define <&(M) to be an incidence system of rank 5 whose elements of

type 1, 2, 3, 4 and 5 are the vertices, lines, *-closed cliques of size 7,
*-closed cliques of size 15 and the maximal cliques from the orbit Jfvy
respectively, and the incidence relation is via inclusion. Then vo, I and
£ (as in the proof of (5.2.3)) are pairwise incident elements in ^(M)
stabilized by G\, Gi and G3, respectively. The residue of v0 is isomorphic
to the tilde geometry ^(Co\) on which G = Co\ induces a flag-transitive
action. If N is an element of type 5 in 9(M), so that N e Jfv, then the
residue of N is the projective geometry of the proper subspaces in Z(N)
and by (5.2.5) G[N] induces the full automorphism group of this residue.
Finally, since the incidence relation is via inclusion, it is easy to see that
is a geometry with a string diagram and we have the following.

Proposition 5.2.6 The geometry ^(M) is a rank 5 tilde geometry with the
diagram

the group G induces on ^(M) a flag-transitive action and A is the collinear-

ity graph of <&(M). •

5.3 The maximal parabolic geometry

In this section we construct the maximal parabolic geometry J^(M) for
the group G. A crucial step is to show that G contains a subgroup D of
the form 210+16.Q£)(2) containing Qx such that (D n Gi)/gi = 0^(5) ^
2++8.Qjj~(2) where S is a central involution in G\ = Co\ as in Section 4.8.
As above, A is the first Monster graph (i.e. the collinearity graph of
^(M)), vo is a vertex of A such that G(v0) = Gu (p : v \-+ Zv = Z(G(v))
is a bijection of A(i;o) onto a set 3C of subgroups of order 2 in Q\ as in
(5.2.1), rj is a mapping of 9C onto the vertex set A4 of the Leech graph F,
commuting with the action of G\9 such that the fibres of rj are the orbits
of Q\. Then u9v e A(i;0) are adjacent if and only if r\cp(u) and rjq>(v) are
either equal or adjacent in F. Furthermore, we assume that {vo, 1^2} is
a line stabilized by G2 and that r\(p(vi) = r\cp(v2) = Io-
Let u be a vertex from A(i;o) such that rjcp(u) e A4 for a = 6,4fc,4c
or 5. Then by (5.2.3) u is at distance 2 from v\. By (5.1.4) G12/61 =
Gi(Io) = 2n.Mat24 and by (5.2.2) Qx(vi) j* Qi(u). Hence G(i; 0 ^i) = G+
 5.3 The maximal parabolic geometry 219
l l
acts transitively on cp~ Y]~ (A\) for every a as above. Let A^tfi) be
the orbit of u under G(v\). By the above this orbit is independent
of the particular choice of u. For an arbitrary w e A we put A^w)
to be the image of A^tfi) under an element of G which maps v\
onto w.

Lemma 5.3.1 Let u e A(i>o), v = rjq>(u) and suppose that v € A£ for

a = 6,Ab, Ac or 5.
(i) [Zu,ZVl] =Z\ if a — 5 and [Zu,ZVl] = 1 otherwise,
(ii) the orbit of u under QVl has length 2 8 , 2 9 , 2 13 or 2 13 , respectively,
(iii) t/iere are exactly four orbits of G\ on A2(uo), namely the orbits

(iv) */ w € A2(i?o) ffow Gi n G(w) acts transitively on A(vo) n A(w),

(v) i/ w G A^t^o) ^ w f/iere is a unique vertex v adjacent to both VQ
and w and [ZW,Z\\ = Zv,
(vi) ifwE A^ivo) for a = 6,4b or Ac then Z w < G\ and Z w ^ Qi.

Proof. Part (i) follows directly from (4.8.2 (iv)) and (5.1.3). Let B be the
orbit of v under QVl n Gx. Since 02{Gn) = (QVl n Gi)(6i n G ri ), by (4.6.2)
we have |B| = 2 6 , 2 7 , 2 11 and 2 11 for a = 6, 46, 4c and 5, respectively.
Let {w,wi} = (p~lr\~x(v). We claim that Qyi n Qi contains an element ^
which maps M onto u\. By (5.1.3) and (5.1.9 (ii)) an element q from Q\
possesses these properties if rj(q) e A*a n A|(v). Since the orbitals of the
action of Co\ on the Leech graph are self-paired, such a q exists if and
only if A4a(v) Pi A | ^= 0 or, equivalently, if there is a vertex in A | adjacent
to v in F. One can see from the suborbit diagram of T in Section 4.7 that
such a vertex exists for every a under consideration. Hence the orbit S
of u under QVl n G\ is twice as long as S. Since u is adjacent to vo and
not adjacent to V2 whenever g is an element from QVl \ G\ (which maps
vo onto V2) we have S f i S g = 0 and hence the length of the orbit of u
under QVl is four times the size of S and (ii) follows. Now (iii), (iv), (v)
and (vi) follow immediately from (i), (ii) and their proof. •

For u e A(i;0) n A^(i?i) put

A = A(vi9u) = (Zw\we {vl9u,

By (4.8.8) the vertices Io = *l<p(vi) and v := rj(p(u) determine a subgraph

O = O(lo? v) in the Leech graph which is induced by the vertices fixed by
02{CQ1(S)) for a central involution S in G\ = C01 and O generates in A
a subspace Fi = V\(Xo,v) of dimension 8.
220 The Monster
Lemma 5.3.2 The following assertions hold:
(i) A<(ZVl,Zu,QVinQu);
(ii) A is an elementary abelian 2-group and the index of An QVl in A
is at most 2;
(iii) AHQi = (rj-l(Ji) \Jie^)and\Ar)Q1\= 2 9;
(iv) whenever u,v e A(i;o) are such that r\(p(u) and rjcp(v) are distinct
vertices in Q> at distance 2 from each other, we have A = A(u,v);
(v) A = A(vo,v)for some v € A^(t;0);
(vi) \A\ = 2 1 0 .

Proof. By (5.2.1) if w G A(vi)nA(u) then Z w < QVl nQu and (i) follows.
By (i) the subgroups Zw taken for all w e A(t;i) n A(u) generate in A
an elementary abelian subgroup centralized by ZVl and ZM. By (5.1.3)
and (4.8.2) we have [ZVl,Zu] = 1 and (ii) follows. By (5.1.9) we have
<2i n QVl = #2 while in terms of (4.8.2) we have R2 = r\~l{U2Q^)y Hence
(iii) and (iv) follow from (4.8.8 (ii)) and (i). Since the graph which is the
complement of <D is connected (compare the diagram before (4.8.8)), by
(iv) A = A(w,v) for some w e A(v\) n A(u). By the definition of A it is
normalized by G(v\9u) and by (5.3.1 (iv)) G(vi,u) acts transitively on the
set A(vi) n A(w) which contains vo. Hence (v) follows. By (v), (ii) and (iii)
either \A\ = 2 10 or A < Q\. In the latter case by the obvious symmetry
A must be contained in QVl which is impossible since Zu is not in QVl.
Hence (vi) follows. •
Let *F be the connected component containing vo of the subgraph
induced by the vertices u G A such that A = A(u, v) for some v e A^(M)
and put D = G\*¥]. Then D is vertex-transitive on ¥ and by (5.3.2 (iii))
¥(tfo) = <p-lr\-l(Q>) (in particular V is of valency 270 = 2 • 135). Let &
be a geometry whose elements are the vertices, the lines and the ^-closed
cliques of size 7 and 31 contained in *F. A clique Nv E Jfv and a clique
Nt e Jft have different type and they are incident if and only if Nv D Nt
is a *-closed clique of size 15; the remaining incidences are via inclusion.

Lemma 5.3.3 The following assertions hold:

(i) !F is the natural parabolic Ds-geometry of 0,^(2);
(ii) D induces the full automorphism group of 3F';
(iii) D - 210+16.Q+0(2) and A = Z(O2(D)).
Proof. By (5.3.2 (iii)) the residue of vo in <F is the parabolic geometry
of Qg~(2) as in the proof of (4.8.6) what particularly implies that for a
*-closed clique J of size 15 and a € {v,t} there is a unique Na € JV<X
 5.3 The maximal parabolic geometry 221
such that J c: Na <= *F. Hence the residue in 3F of a maximal clique N
is isomorphic to the rank 4 projective GF(2)-geometry of the *-closed
cliques contained in N. This shows that 3F is a Tits geometry of type
D5(2) and (i) follows from (1.6.3). By (5.3.2 (iii), (iv)) D n Gx is the full
preimage in Gi of Gi [<X>] which acts flag-transitively on TQS^(VQ). In view
of the vertex-transitivity of D on *F this shows that the action of D on 3F
is flag-transitive and (ii) follows from (1.6.5). Let K be the kernel of the
action of D on 3F. Then K is contained in D(vo) and the latter is of the
form 2j.+24.2j.+8.Qj(2). It is easy to see that A is contained in K and that
^4 is the natural module for D/K = Cl£0(2). Hence K/i4 is of order 2 16
and it involves two irreducible chief factors for D(vo)/02(D(vo)) = Qg"(2).
Since one of these factors, namely (K n g i M M , is not normalized by D
we obtain (iii) with the remark that K/A is the spin module for D/K. D
By the above lemma *F is the graph on the non-zero isotropic vectors
in the natural module of $2J)(2) in which two vectors are adjacent if their
sum is an isotropic vector. The intersection diagram of *F with respect to
the action of D/K = Q+0(2) is the following:
1+140 135
270

Comparing the diagram and (5.3.1 (ii)) we observe that *F n A2(^o) is an

orbit of Qi on Af (t?o).
Let Jff(M) be an incidence system of rank 5 whose elements of type
1, 2, 3, 4 and 5 are the vertices, lines, *-closed cliques of size 7, *-closed
cliques of size 31 from the orbit Jft and the images under G of the
subgraph *F stabilized by D; the incidence relation is via inclusion. Then
by (4.9.1) and (5.3.3) we have the following.

Lemma 5.3.4 The incidence system ffl(M) is a geometry with the diagram
n

and the group G acts on Jtf\M) flag-transitively with D ~ 2 10+16 .Qj ) (2) as
the stabilizer of an element of type 5. •

Notice that for i = 1, 2 and 3 the set of elements of type i in

coincides with the set of elements of type i in J^(M), so that G\,
222 The Monster
and G3 are also maximal parabolics of the action of G on J^(M). A
clique Nv e Jfv (an element of type 5 in ^(M)) is contained in a unique
element of type 5 in J f (M) while a clique Nt e Jft (an element of
type 4 in Jf(M)) is in three such elements. If we put G5 = G[NV] and
H4 = G[Nt] then
G5 ~ 2 5+5+10+5+1+10 .L 5 (2), H4 ~ 2 5+10+20 .(L 5 (2) x Sym3).
Finally the stabilizer G4 of an element of type 4 in ^(M) has index 31 in
H4 and
G4 - 2 4+1+2+12+8+8+4 .(L 4 (2) x Sym3).

5.4 Towards the Baby Monster

As in the previous section let A = Z(02(D)) where D = H5 is the
stabilizer in G of the subgraph *F which is an element of type 5 in
& = J^(M). Then an element n e r e s ^ ( ^ ) can be identified with the
subgroup Z n in A generated by the Zu taken for all u G II. In this way
the elements vo, I and Z of type 1, 2 and 3 in the corresponding maximal
flag are identified with Z\9 Z2 and Z3, respectively. For a subgroup
in 4 the terms "isotropic", "non-isotropic", "orthogonal complement"
etc. are with respect to the orthogonal form of plus type preserved by
D/OiiD) = Qxo(2). Let Y\ be a non-isotropic subgroup of order 2 in
A contained in the orthogonal complement of Z3. We will study the
centralizer of Y\ in the group G; the quotient of this centralizer over Y\
will eventually be identified with the sporadic simple group known as the
Baby Monster.
In view of (5.3.3 (iii)) and the standard properties of the orthogonal
groups [Tay92] we have the following.

Lemma 5.4.1 Let Bo = CD{Yi) and Bo = B0/Yi. Then Bo ^ 2 9+16 .Sp 8 (2)
and A/Y\ is the orthogonal module for BO/O2(BO) = Sp8(2) = ^ ( 2 ) . •
Since Z\ < Z 3 the subgroup Y\ is in the orthogonal complement of
Z\. By (5.3.2 (iii)) the orthogonal complement of Z\ is A n Q\. Hence
^i ^ 61 and since Y\ is non-isotropic, rj(Yi) e A2. Hence without loss
of generality we can assume that 77(Yi) = /2o where //o is as introduced
before (4.9.5), i.e. ^ € A^,XQIQ) = 0 and P4(/*o) = {a,b}-

Lemma 5.4.2 Let B\ = CGl(Yi) and B\=Bi/Yi. Then

22
(i) BX S 2l+ .Co2 and &
 5.4 Towards the Baby Monster 223

(ii) B\ acting on A(t>o) has three orbits, Qj := rj"1 (A4 n A;(/io)) for
j = 2, 3 and 4,
(iii) i / u G Q ; then [YUZU] = 1 (equivalently Yi < G(M)>) I/and on/y 1/
j = 2 or 4,
(iv) /or ; = 2, 3 and A if u e O, fften 5i(w)Qi/6i S 2 10 : Aut Mat 22 ,
Mar23 and [25] : (24 : L4(2)), respectively,
(v) z / M G Q 4 ^ n [ e M : C Q u ( y 1 ) ] > 2 6 ,
(vi) t;oflwdQ4 are in different orbits of CG{Y\) on A.

Proof. Part (i) follows directly from (4.5.5) and (5.1.3) and (ii), (iii),
(iv) from (4.4.1) and the table therein. Let u e Q4 and let g be an
element in G\ such that ug = v\ (recall that rj(vi) = Io). Then fil e A|.
By (5.1.9) g i n QVl is of order 2 13 and hence QVlQi/Qi = 0 2 (Gi(Io))
(where Gi(I 0 ) = 2 11 : Mat 24 ). We can see from the table in (4.4.1) that
the orbits of 0 2 (Gi(Io)) on A^ are of length 2 6 and we obtain (v). Since
[& : CQl(Yi)] = 2 by (i), we have (vi) by (v). •

Let B2 = CG2(Yi). In terms of (5.4.1) Z2Yi/Y\ is a 2-dimensional

isotropic subspace in the 9-dimensional orthogonal space A/Y\. Since
B0/02(B0) = 5pg(2), the normalizer of Z 2 in B o permutes transitively
the three subgroups of order 2 in Z 2 . This means that B 2 contains an
element which maps vo onto vi. Let B = (BuBi) and let 0 be the
subgraph in A induced by the images of vo under B. We call B := B/Y\
the Baby Monster group and denote it also by BM. Since B < CQ{Y{)
and B\ = C G l (ri) we have 5(i?0) = 5i. By (5.4.2 (ii)—(vi)) we have the
following.

Lemma 5.4.3 0(i?o) = ^2 and 7y(Q2) is the point set of a

subgeometry in

Let &(BM) denote the subgeometry in ^(M) formed by the elements

contained in 0 with respect to the induced incidence relation and type
function.

Lemma 5.4.4 <g(BM) is a rank 5 Petersen geometry with the diagram

P
2 2 2 2 1

and BM acts on &(BM) flag-transitively.

Proof. By (5.4.3) the residue of v0 in <#(BM) is isomorphic to &(Co2)

and by (5.2.4) an element N of type 5 in &(BM) is a maximal clique in 0 .
224 The Monster

Since B is 1-arc-transitive on 0 , the stabilizer of N in B acts transitively

on its vertex set. Furthermore the stabilizer of rj(N) in CQ^Y^QI/QI
induces L4(2) on the element set of rj(N) and since N e (^(Yi))-1, Q\ n
CGy{Y\) induces on N an elementary abelian group of order 2 4 . Hence
Bi[N]/Bi(N) ^ 24.L4(2) and so B[N]/B(N) s L5(2). Since the incidence
relation is via inclusion, &(BM) belongs to a string diagram and the
result follows. •
Let {Bi | 1 < i < 5} be the amalgam of maximal parabolic subgroups
associated with the action of BM on $(BM). Then from the structure of
the maximal parabolics in the group G we can deduce the following:

B{ ~ 2l_+22.Co29 B2 ~ 22+10+20.(Sym3 x Aut Mat22),

B3 ~ 23.[232].(L3(2) x Sym5), B4 ~ 24.[230].(L4(2) x Sym2\

B5 ~ 2 5+10+10+5 .L 5 (2).
Notice that because of the choice of the maximal flag in ^(M) we have
B t = CGi(Y1)/Y1 for l < i < 5.
Let *F be the subgraph of A introduced before (5.3.3) and isomorphic
to the point graph of the parabolic geometry of Qi~0(2). Then *¥ n 0
consists of the points contained in the orthogonal complement of 7i in
A. This complement is clearly an 8-dimensional non-singular symplectic
space and we have the following.

Lemma 5.4.5 The elements in &(BM) of type 1, 2, 3 and 4 which are

contained in *F Pi 0 form a C^{2)-subgeometry &(Sps(2)) whose stabilizer
in B is the subgroup Bo = 29+16.Spz(2) as in (5.4.1). The residue ofvo in
this subgeometry is the <&(Sp6(2))-subgeometry in &(Co2) as in (4.9.8). •
Since B\ acts transitively on the set of lines in &(BM) incident to vo,
we have the following.

Lemma 5.4.6 Let x2 be an element of type 2 in &(M) which contains VQ.

Then Co^iYi) £ CGl(Yx) if and only ifnq>(x2) e A2(Jio)nA4 = n(Q2). D

5.5 2E6(2)-subgeometry
Let Y2 be a subgroup of order 4 in A generated by Y\ and a non-isotropic
subgroup Y{ which is not perpendicular to Y\ (so that Y2 is a minus
2-space). We assume that Y2 is contained in the orthogonal complement
 5.5 2E6(2)-subgeometry 225

of Z3 and that rj(Y2) = (A), Mi, £2) where fi\ G A* such that JX2 := /io + Pi
is contained in A2 (say ^4(^1) = {b,c}, c j= a and X(/*i) = 0).
In view of the choice of Y2, by (5.3.3) and standard properties of the
orthogonal groups we have the following.

Lemma 5.5.1 Let E4 = ND(Y2), E4 = E 4 /Y 2 and T be the subgraph

induced by the vertices y such that Zy is in the orthogonal complement of
Y2 in A. Then E4 = 28+16.(fi£~(2) x 3).2 and T is isomorphic to the point
graph of the natural parabolic geometry of $Jjj"(2). •
The suborbit diagram of T with respect to the action of E4/O2,3(E4) =
Qjj"(2).2 is the following:
54 1 / T ~ ~ T \ 32 27
©•
Directly from (4.10.8) we obtain the following.

Lemma 5.5.2 Let Ex = NGl(Y2) and Ex = Ex/Y2. Then O 2 (£i) = rj-^T1)

(where T is as in (4.10.4); and Ex S 21+20.U6(2).Sym3. •

Let £2 = NG2(Y2). Since Z2 is in the orthogonal complement of Y2,

the normalizer of Z2 in E4 permutes transitively the three subgroups of
order 2 in Z2. This means that E2 contains an element which maps vo
onto v\. Put E = (Ei,£2), E = E/Y2 and let S be the subgraph in A
induced by the images of vo under E.

Lemma 5.5.3 3(t>o) = V~X(^4 n ^i{fio) n A2GU1)) and the action of E\ on

this set is transitive.
Proof. By the paragraph before the lemma S(uo) contains v\ and hence
it is non-empty. By (5.5.2) (or rather by (4.10.8)) E\/CGX(Y2) = Sym3 =
Aut Y2. This means that S is also the orbit of vo under E n CG(Y2). Since
CG(Y2) = CG(Yi) n CG(Y{), by (5.4.2 (ii), (iv)) we conclude that S(i?0) is
contained in rj~1(A4 D A2(Jio) n A2(p,i)). Finally by (4.10.5) the action of
E\ on the latter set is transitive and the result follows. •
Define y^E^l)) (just a name so far) to be the incidence system of
rank 4 such that for 1 < i < 3 the elements of type i are the elements of
type i in ^(M) contained in S and the elements of type 4 are the images
under E of the subgraph Y, the incidence relation being is via inclusion.

Proposition 5.5.4 The following assertions hold:

226 The Monster
2
(i) ^( E^(2)) is a Tits geometry with the diagram

2 2 4 4

and E induces a flag-transitive automorphism group of^(2Ee(2));

(ii) £ ^2E6(2)Sym3.
Proof. Since the incidence relation is via inclusion, ^(2Ee(2)) belongs
to a string diagram. By the proof of (4.10.6) the residue of vo is the
C3-geometry of Ue(2) on which E\ induces a flag-transitive action while
the residue of T is the C3-geometry of Q^(2) on which £4 acts flag-
transitively. Hence (i) follows. Since the C3-residues in ^(2E^(2)) are
buildings, by [Ti82] the geometry itself is a building of 2 £ 6 (2). By (1.6.5)
and (5.5.2) £ is the full automorphism group of ^( 2 £6(2)) and (ii) follows.
•
Below we present the suborbit diagram of S with respect to the action
of £. Notice that T contains vo, 54 vertices from E(t;o) and 64 vertices
from Efj^o)- Furthermore, by (4.10.5) it is easy to observe that in the
notation of (5.3.1) S^tfo) is contained in Al2(vo) for i = 5 and 6.

Let {Et I 1 < i < 4} be the amalgam of maximal parabolics associated

with the action of E on ^( 2 £6(2)). Then £1 and £4 are as above and we
have
£1 = 2i+2O.l76(2).S^m3, E2 s 2 2 + 9 + 1 8 .(PrL 3 (4) x

£ 3 ^ 2 3+4+12+12 .(L 3 (2) x 3.Sym5), £4 S 28+16.(Q^"(2) x 3).2

and because of the choice of the maximal flag in ^f(M), Et is the quotient
over Y2 of the normalizer of Y2 in H\ = G\9 H2 = G2, if3 = G3 and
if5 = D for i = 1,2,3 and 4, respectively.
 5.6 Towards the Fischer group M(24) 227
2
There is an involution co e E\ O (E) such that Q(co) = F4(2) x 2
[ASei76]. The element co induces on Y2 an action of order 2 so that
(Y2,co) = Z)g. The vertices in S fixed by co form a subgraph which is
the collinearity graph of the natural parabolic geometry of F^(2). The
suborbit diagram of this subgraph is presented below.
1+28 1+14+112

5.6 Towards the Fischer group M(24)

Let Xs be a Sylow 3-subgroup in 02,3(63). Then Xs is contained in Gj
and in G\ and it maps onto a conjugate of the subgroup in G\ introduced
under the same name in (4.14.1). We are going to study the normalizer of
Xs in the group G and the connected component A(XS) of the subgraph
in A induced by the vertices which are fixed by Xs. Notice that if IT
is the neighbourhood of vo in A(XS) then in terms of (4.14.10) we have
rj(U) = 3>(XS). Let F denote the setwise stabilizer of A(XS) in NG(XS). We
are going to identify F = F/Xs with the largest Fischer 3-transposition
group M(24). Our first result follows directly from (4.14.9) and (4.14.11).

Lemma 5.6.1 NGl(Xs)/Xs s l1^12 : 3 • l/4(3).22. •

Let X be a Sylow 3-subgroup in G2. Then by (3.8.2) and the Frattini
argument we obtain the next result.

Lemma 5.6.2 Let N = NGl(X). Then N s 2n:(Sym3 x Mat24) and O2(N)

is isomorphic to the irreducible Todd module <€\\. •
Since X is contained in G^~, it fixes res^(/) = ^(MatiA) elementwise
(where / is the element of type 2 in ^ = ^(M) stabilized by G2). Arguing
as in the paragraph before (4.14.3) we observe that for every element u of
type i in resj(/) there is a unique element xp(u,X) of type i — 2 incident to
228 The Monster
u which X fixes pointwise (where 3 < i < 5) and we obtain the following
analogue of (4.14.3).

Lemma 5.6.3 The subgeometry ^(/,X) in @ formed by the elements xp(u,X)

taken for all u e resJ(Z) is isomorphic to res^(Z) = ^(Mati^) and NQ2{X)
induces on ^F(/, X) the natural action of Ma
Recall that the sextet graph is the collinearity graph of both &(Mat24)
and

Lemma 5.6.4 There is an element g € G whichconjugates X onto Xs

and maps the set of vertices in ^(hX) onto a subgraph ^(/ g ,X s ) of A(XS)
isomorphic to the sextet graph.
Proof. Let u be an element of type 4 in resJ(Z) and v = tp(u,X). Then
v is of type 2 and X fixes res^(t;) which means that X is conjugate to
a subgroup in Gi2. By (5.6.2) and (3.3.4) the order of NG(X) n G(v) is
222 • 33. Hence the result follows from (4.14.1), (4.14.11) and the remark
after the proof of that lemma. •
By the paragraph before (4.14.6) NGI(XS) has 2 orbits on the set of
planes (*-closed cliques of size 7) contained in A(XS) and containing vo,
furthermore one of the orbits is 15 times as long as the other one. This
means that F has 2 orbits on the set of planes contained in A(XS). Let
^(M(24)) be an incidence system of rank 4 whose elements of type 1
and 2 are the vertices and lines of &(M) contained in A(XS), the elements
of type 3 are the planes from the long orbit of F on the set of planes
contained in A(XS), the elements of type 4 are the images under F of the
subgraph *F(/g,Xs) as in (5.6.4) and the incidence relation is via inclusion.

Lemma 5.6.5 ^(M(24)) is a geometry with the diagram

2 2 2 2

and F induces aflag-transitiveautomorphism group o

Proof. Since the incidence relation is via inclusion, ^(M(24)) belongs
to a string diagram. By (5.6.3) and the paragraph before that lemma
the residue of an element of type 4 is isomorphic to the tilde geometry
^(Ma*24). The residue of an element of type 1 is isomorphic to the
geometry ^(3 • 1/4(3)) by (4.14.6) and the paragraph before that lemma.
The flag-transitivity is straightforward. •
 5.6 Towards the Fischer group M(24) 229
Let {Kt | 1 < i < 4} be the amalgam of maximal parabolics associated
with the action of F on ^ ( M (24)). Then under a suitable choice of the
maximal flag of #(M) we have Kt = NGi(Xs)/Xs for 1 < i < 3 and
K 4 = NG2{X)/X where X is a Sylow 3-subgroup in 02,3(62) and we have
the following:
Ki S 2*+12 : 3 • l/4(3).22, X 2 S 2 2+1+4+8+2 .(Sym 3 x Sym6),

K3 3* 2 8+6+4 .(L 3 (2) x 5ym3), K 4 = 211+1.Maf24.

The subgraph x F(/ g ,X s ) in A(XS) which realizes an element of type 4 in
^ ( M (24)) is the collinearity graph of 3tf?(Mat24). An element of type 2 in
J^{Mat2A) (a trio) is realized by a 7-vertex complete subgraph © 2 and
one can see that © 2 is a plane of ^(M) contained in the short orbit of
F on the set of planes in A(XS). An element of type 3 in 3tf(Mat24) (an
octad) is realized by a 35-vertex subgraph © 3 whose vertices are indexed
by the 2-dimensional subspaces in a 4-dimensional GF(2)-vector-space
with two subspaces being adjacent if their intersection is 1-dimensional.
Let J f (M(24)) be an incidence system of rank 4 whose elements of type
1 and 4 are as in ^ ( M (24)), whose elements of type 2 and 3 are the
images under F of © 2 and ©3, respectively, and the incidence relation is
via inclusion. Then J f (M(24)) is a geometry with the following diagram:

The action of F on ^f(M(24)) is flag-transitive and if {Ft \ 1 < i < 4}

is the amalgam of maximal parabolic subgroups corresponding to the
action of F on ^f (M(24)) then Fi = Ku F4 = K4 while
F2 = 2 3+12 .(L 3 (2) x Syme\ F3 s 2 7+8 (L 4 (2) x Sym3).

Let us consider the intersection of A(XS) and the subgraph *F intro-

duced before (5.3.3) which realizes an element of type 5 in J^(M). Since
^(vo) = cp~lY\~l(Q>) and we can choose Q> to be equal to O(<5) as in the
paragraph before (4.14.7), we can assume that rj(*¥(VQ) C\ A(XS) C\ A(VQ))
realizes an element of type 1 in ^(3 • C/4(3)) as (4.14.9). By (4.14.8), in
this case [A, Xs] is a minus 2-subspace in A which can be identified with
Y2 as in the first paragraph of Section 5.5. Hence ND(XS) is contained in
E4 = ND(Y2), ND(XS)/XS ^ 28.Q^(2).2 and A(XS) n *F coincides with the
subgraph Y as in (5.5.1).
Define S(M(2A)) to be an incidence system of rank 4 whose elements of
type 1 are the images under F of the subgraph T as above, the elements
230 The Monster
of type 2 are the planes from the longer orbit of F on the set of planes
contained in A(XS), the elements of type 3 and 4 are the lines and points
of ^(M) contained in A(XS) and the incidence relation is via inclusion.
Then by the preceding paragraph and (4.14.9) we obtain the following

Lemma 5.6.6 <^(M(24)) is an extended dual polar space with the diagram
c
1 4 2 2
on which F induces a flag-transitive action. •

By the construction {ND(Xs)/Xs,K3,K2,Ki} is the amalgam of maxi-

mal parabolics associated with the action of F on ${M(24)).
In the next lemma we specify the structure of O2CF4).

Lemma 5.6.7 Let Q = O2CF4). Then Q is elementary abelian and as a

module for F4/Q = Mat24 it is isomorphic to the 12-dimensional Todd
module.

Proof. Recall that F4 = K4 = N/X where N = NGl(X) and X is a

Sylow 3-subgroup in G^ S 22+n+22.Sym3. Let P = O2(N), P = PX/X
and Q be a Sylow 2-subgroup in the preimage of Q in N. Then Q = Q,
P is a subgroup of index 2 in Q and by (5.6.2) P is the irreducible Todd
module for F4/Q. Since P is an irreducible submodule of codimension
1 in <2, it is easy to see that Q is abelian. Furthermore, P is the only
faithful submodule of F4/Q in Q. Hence in order to prove the lemma
it is sufficient to show that F4/Q has an orbit of length 24 in Q \ P on
which it acts as on the base set 9 of the Golay code (2.15.1). We consider
N as a subgroup in G2. Since NGl(X)/NG-(X) = Mati4 has no normal
2-subgroups, Q is contained in one of the three Sylow 2-subgroups of G^".
Since G12 ~ 21f24.2n.Mat24 and O2(Gi2) is a Sylow 2-subgroup in G^~
we can assume without loss of generality that Q < O2(Gi2). Moreover,
since P < R2 < Qu 02(Gn)(N n Gn) = Gn and Gi 2 /O 2 (Gi 2 ) = Ma*24
acts irreducibly on O2(Gi2)/6i, we conclude that Q < Q\. Now given an
element from Q\ we have to decide in what case it normalizes and does
not centralize X. Recall that R2 = i / " 1 ^ U Af) and P = CR2(X) is a
complement to Z 2 in R2. This means that (# 2 , X) = (Z2, X) xP = Alt4 xP.
Hence if Y is a Sylow 3-subgroup in G^~ then P = C^2(7) if and only if
Y is one of the four Sylow 3-subgroups in (Z2,X). On the other hand
G2 contains 2 24 = [G^ : N] Sylow 3-subgroups and R2 contains 2 22
complements to Z2. Hence for each complement to Z2 in R2 there are
 5.7 Identifying M(24) 231
exactly four Sylow 3-subgroups Y in G^~ such that P = CR2(Y) and Z 2
acts transitively on the set of these four subgroups by conjugation. In
particular [NG2(P) : N] = 4 and an element g from Qi normalizes P if and
only if Xq < (Z 2 ,X) and in this case q normalizes and does not centralize
a subgroup of order 3 in (Z 2 ,X) if and only if it does not centralize
Z 2 , i.e. if q £ Z^. Hence without loss of generality we can assume that
rj(P) coincides with the complement E to U\ = f/(Z2) in l/ 2 = ^(#2) as
in (4.8.3). In terms of (4.8.3) let_F = Y\~\EI\ Then by (4.8.3) NGn(P)
stabilizes V setwise. Since £3 ^ A2 we have V & Z^ and by (4.8.3 (iv))
V c p-L. By the Frattini argument NGl2(P)/NG-(P) s Mat 24 and by
(4.8.3 (ii)) the latter group induces on £3 the natural action of degree 24.
Finally, since the stabilizer of an element in this action (isomorphic to
Mat23) has no subgroups of index 8 or less we conclude that N/X has
an orbit of length 24 on VX/X and the result follows. •

Corollary 5.6.8 A subgroup X of order 3 in Gj can be chosen in such a

way that for

the set r\~l(Ei) is contained in NQ2{X) \ CQ2{X) and it maps onto an orbit
of length 24 ofNGl(X)/X on O2(NGl(X)/X). •

5.7 Identifying Af (24)

In this section we study the geometry ffl = J^(M(24)) and the action of F
(still to be identified with M(24)) on ^f. Let {yt \ 1 < i < 4} be a maximal
flag in Jf so that {Ft = F(yi) \ 1 < i < 4} is the amalgam of maximal
parabolics associated with the action of F on Jf. Put Rt =
Ftj = Ft n Fj9 Rtj = O2(Fij) for 1 < ij < 4.

Lemma 5.7.1 For i = 1, 2 and 3 we have #,4 =

Proof. Comparing the shapes of F, and F4 we obtain the following:

F14 - 2i_+12.25.3 • Sym6 - 212.26.3 • Sym6;

F24 - 2 3+12 .(L 3 (2) x (2 x Sym4)) - 212.26.(L3(2) x Sym3);

F 34 - 2 7+8 .(L 4 (2) x 2) - 212.24.L4(2).

By (2.10.1), (2.10.2) and (2.10.3) the action of F I 4 /JR, 4 on Ri4/R4 is irre-

ducible and the result follows. •
232 The Monster
Let F = F(M(24)) be a graph on the set of elements of type 4 in J^
in which two such elements are adjacent if they are incident in Jf to
a common element of type 3. For z G «f of type 1, 2 and 3 let T[z]
denote the subgraph induced by the vertices incident to z in Jf. (Recall
that for z € F of type 4 T(z) is the neighbourhood of z in the graph
F.) When talking about octads, trios and sextets we mean those from

Lemma 5.7.2 The following assertions hold:

(i) F acts on T vertex- and edge-transitively;
(ii) there is a mapping K from F(j/4) onto the set of octads which com-
mutes with the action of F4;
(iii) if B is the octad which corresponds to y^, then K~1(B) is an orbit
of length 2 of R4 and Tlys] = {y4,K~1(B)} is a triangle in F ;
(iv) T\y2] is the point graph of ^(S/?4(2)) = r e s ^ t e ) and K(T\y2] n
r()>4)) is the set of octads contained in the trio which corresponds
to y2;
(v) T{yi] is the point graph of (the dual of) #(174(3)) = resj,(j/i) and
K(T[yi] Pi r(^4)) is the set of octads refined by the sextet which
corresponds to y\ ;
(vi) the valency of F is 2 • 759.
Proof. Part (i) follows from the flag-transitivity of the action of F on
Jf. By (5.7.1) the subgroup R4 induces on T\y2] an action of order 8.
By the basic properties of the generalized quadrangle of order (2,2) this
implies that R4 induces on T(y4) n r\y2] an action of order 4. By (5.6.7)
R4 is the 12-dimensional Todd module which is indecomposable. Hence
R4 acts faithfully on FO^) with orbits of length 2 and the kernels at these
orbits correspond to one of the two orbits of length 759 of F4/.R4 on the
dual of R4 (which is the Golay code module). This implies (ii) and (iii).
For i — 1 and 2 there is a unique orbit of JF^/JR,- on T\yi\ (with length 30
and 6, respectively) on which R^/Ri acts with orbits of length 2. Hence
(iv) and (v) follow from (iii). Finally (vi) is a direct consequence of (ii)
and (iii). •
In our further considerations a crucial role is played by the observation
that the geometry ^(1/4(3)) and the geometry dual to ^(1/4(3)) have the
same sets of elements of type 2 and 3 (Section 6.13). By this observation
r[)>i] contains the Schlafli graph as a subgraph. More specifically the
following holds. Let E = {Si,...^6} be the sextet which corresponds to
yu so that {Si U S/ | 1 < i < j < 6} is the image of T(y4) n T\yi] under
 5.7 Identifying Af(24) 233
K. Then for every k, 1 < k < 6 there is a unique Schlafli subgraph ©in
r[yi] containing y* which is an element of type 1 in (f( 1/4(3)) such that
K(®(y4)) = {Sk U Si \ 1 < i < 6,i ^ fc}. Notice that S (and hence yx as
well) is uniquely determined by S&.

Lemma 5.7.3 The graph T = T(M(24)) contains a family 9> of Schlafli

subgraphs with the following properties:
(i) for every 4-element subset S of the base set 0> of the Golay code
associated with y4 there is a unique 0 G ^ which contains y4, such
that K(&(y4)) is the set of (five) octads containing S;
(ii) the group F acts transitively on 9* and

F [ 0 ] ~ 2*_+12.3 : (l/4(2).2 x 2);

(iii) ^14/02,3(^14) = Syme has a unique orbit Q, of length 6 on the

involutions in O2)3(Fi4)/O2,3(F1) s 2 5 ;
(iv) 0 is (the connected component containing y4 of) the subgraph of
induced by the vertices fixed by an involution from Q.
Proof. The assertions (i) and (ii) follow from the paragraph before
the lemma. Since 02,3(Fi4)/02,3(Fi) *s °f order 2 5 while by the basic
properties of the Schlafli graph we have F 4 [ 0 ] / F ( 0 ) S 24.Sym5 (4.14.7)
we obtain (iv). Since 02,3(Fi4)/02,3(Fi) involves the natural symplectic
module of Sym^ = Sp4(2) we obtain the uniqueness of Q stated in (iii).n

Recall that the Todd module ^12 = R4 is the quotient of the power set
2^ over the Golay code #12 and that it is generated by the images of the
subsets of size at most 4 (2.3.3). By (5.7.1) and (2.15.1 (iii)) we have the
following.

Lemma 5.7.4 The unique non-identity element s in Z(R\) is contained in

R4 and it is the image in R4 = %>n of a tetrad from the sextet which
corresponds to y\ and Fu = CF^S) for *' = 2,3 and 4. •

For a subset Y of & let &n(Y) and ^n[Y] denote the subspaces in
generated by the images of the subsets contained in Y and of the
subset having even intersection with Y, respectively.

Lemma 5.7.5 Let z e T(y4) and K(Z) = B (an octad). Then

(i) O 2 (F 4 H F(z)) = (*4 n F(z))(O2(F(z)) n F 4 )),
(ii) \R4n02(F(z))\=21,
234 The Monster
(iii) R4 n O2(F(z)) = Z(O2(F4 n F(z)) S « 12 (B),
(iv) *

Proof. By (5.7.2 (iii)) we have F 4 n F(z) ~ 2 n .2 4 .L 4 (2) and applying

(5.7.1) for the case i = 3 we obtain (i) which implies (ii). Since both
R4 and 02(F(z)) are abelian £4 n C>2(F(z)) is contained in the centre of
02(^4 n F(z)) which it is easy to identify with <6\2(B\ which gives (iii).
By (5.7.2 (iii)) R4 n F(z) is a hyperplane in R4 normalized by F 4 n F(z)
and by (2.15.1 (i)) this hyperplane is either &n[B] or %12[0>\B]. Since
the former does not contain *i 2 (B) = JS4 n <>2(F(z)) (iv) follows. D

Lemma 5.7.6 Jn terms 0/(5.7.5) suppose that z e T[ys\. Then

(i)
(ii) i / H is F 3 or F 4 n F ( z ) t/ien H acting on Z(# 3 )) # has four orbits
with lengths 8, 28, 56 and 35 consisting of the images in R4 = %>n
of the i-element subsets of B for i = 1, 2, 3 and 4, respectively,
(iii) t/ie action induced by H on its orbit of length 8 in Z(R^ is iso-
morphic to Alt%.

Proof. By (5.7.2 (iii)) we have the factorization F 34 = (F4nF(z))jR4 and

(i) follows from (5.7.5 (iii)). Since R4 is abelian the above factorization
also implies that F3 and F 4 n F(z) have the same orbits on Z(R{j^ and
hence (ii) and (iii) follow. •
Let t\912, t-i and £4 be distinct elements in 9 which we identify with the
corresponding involutions in the orbit of length 24 of F 4 on R4. Let Yt
be the union of the tj for 1 < j < i and we identify Yt with the subgroup
(of order 2l) in R4 generated by the corresponding involutions. Let ®,- be
the connected component containing y4 of the subgraph in T induced by
the vertices u such that Yt < 02{F{u)).

Lemma 5.7.7 The following assertions hold:

(i) a vertex z G F()>4) is contained in 0* if and only if Yi is contained
in the octad K(Z);
(ii) the subgroup F[&t] is contained in NpiYt) and it acts on 0 , vertex-
and edge-transitively;
(iii) the valency of 0 , is 2 • 253, 2 • 77, 2 • 21 and 2 • 5 for i = 1,2,3 and
4, respectively;
(iv) NF4(Yi)/R4 is isomorphic to Mat23, AutMat 22 , PTL3(4) and
26.3.Syni5 for i = 1,2,3 and 4, respectively.
 5.7 Identifying M(24) 235
Proof. Let z e T\y^\ and t = tj for some j , 1 < j < 4. Then by (5.7.5
(iv)) t e F(z) if and only if t € K(Z). On the other hand by (5.7.5 (iii))
and (5.7.6 (i)) if t e K(Z) then t e Z(R3) = JR4 n O2(F(z)), hence (i) and
(iii) follow. Since £4 < NF4(Yi) and NF4(Yt) acts transitively on the set
of octads containing Yt (2.10.1 (iii)) the action of NF4(Yt) on ©,-(^4 ) is
transitive. By (5.7.6 (ii)) NF3(Yi)Ri = F3 and since ©j is connected (by the
definition), we have Ft®,] = (N F4(Yi\NF3(Yi)) and the action of F[@,]
on &i is vertex- and edge-transitive, which gives (ii). Finally (iv) follows
directly by (2.15.1 (iii)). •

Lemma 5.7.8 The subgraph ©

4 belongs to the family Sf of Schldfli sub-
graphs as in (5.7.3).

Proof. Let E = {S\,...9Se} be the sextet corresponding to y\ and

assume that Y4 = Si. Then by (5.7.4) s := t\tit^U is the unique non-
identity element in Z(R\). Since s is also the unique element in Y4 which
is the image in R4 of a 4-element subset of &>, in view of (5.7.4) we
have NF4(Y4) < CF4(s) = Fu. By (5.7.6 (ii)) we also have NF3(Y4) <
CFs(s) = F13. Hence we have ^[©4] < Fi which implies © 4 ^ F[yi]
in view of (5.7.7 (ii)) and the flag-transitivity of F\ on res^(yi). Since
F[©4] = {N F4{Y4\NFi(Y4)), the subgroup Y4 fixes © 4 elementwise. Let
v e T(y4) n T\yi]. Then K(V) = Sk U S/ for some fe,/, 1 < fe < / < 6
and by (5.7.5 (iii)) tm e F(v) if and only if k = 1 (equivalently if
74 ^ K(V)), independently of the choice of m € {1,2,3,4}. Hence Y4
induces on T\y\] n T(y4) an action of order 2. Therefore © 4 is fixed
by an involution from an orbit of length 6 of ^14/02,3(^14) ~ Sym6 o n
O2,3(^i4)/O2,3(^i) = 2 5 . By (5.7.3 (iii), (iv)) we obtain the desired inclusion
04ey. •
Let y = ^~(M(24)) be an incidence system of rank 6 in which the
elements of type i are the images under F of the subgraphs © * for
i = 1,2,3 and 4, the elements of type 5 are the images of F ^ ] under F,
the elements of type 6 are the vertices of F and the incidence relation is
via inclusion.

Lemma 5.7.9 The incidence system &~ is a geometry with the diagram

c*
1 1 1 4 4 2

and F induces on it a flag-transitive action.

236 The Monster
Proof. Since the incidence relation is via inclusion it is easy to see that
2T is a geometry with a string diagram. The elements in resJ-(^4) are the
subsets of 9 of size at most 4 and the octads. Since the incidence relation
is via inclusion, this residue is isomorphic to the geometry S(Mat2^) as
in (3.1.1). By (5.7.8) resJ-(®4) is isomorphic to the geometry of triangles
and vertices of the Schlafli graph (i.e. to the generalized quadrangle of
order (2,4)), hence the diagram is as given above. The flag-transitivity
follows by (5.7.7 (ii)). •
Now we are in a position to apply the geometric characterization of
the largest Fischer 3-transposition group achieved in [Mei91].

Proposition 5.7.10
Let «^~6 be a geometry with diagram as in (5.7.9) and Me be a flag-
transitive automorphism group of 3~e> Then one of the following holds:
(i) M^ is the largest Fischer 3-transposition group M(24) of order
2 2 2 • 3 1 6 • 5 2 • 7 3 • 11 • 13 • 17 • 23 • 29
or the commutator subgroup (of index 2) of M(24);
(ii) Me is the unique non-split extension 3 • M(24) or the commutator
subgroup of 3 • M(24).
In each of the cases (i) and (ii) the geometry &~e is uniquely determined up
to isomorphism. •

5.8 Fischer groups and their properties

Some intermediate results in the proof of (5.7.10) will play an important
role in our subsequent exposition and we discuss these steps in the
present section. Thus let F and F be as in the previous section. Let
{si | 1 < i < 6} be a maximal flag in ^~, where Si is of type i. For
3 < i < 5 let ZTi be a geometry whose diagram coincides with that of
res^(s6-i); let 3~e be a geometry whose diagram coincides with that of
ZT and for 3 < j < 6 let M ; be a flag-transitive automorphism group
of 2Tj. The elements from the left to the right on the diagram of &)
will be called points, lines etc. Let 11/ be the collinearity graph of Fj.
Recall that a group G is said to be a 3-transposition group if it contains
a conjugacy class D of involutions which generates G and such that the
product of any two involutions from D has order at most 3. In this case
the transposition graph of G is a graph on D in which two involutions
are adjacent if their product is of order 2.
 5.8 Fischer groups and their properties 237
The geometry 3~3 is a flag-transitive C3-geometry in which the residue
of a point is the unique generalized quadrangle of order (2,4), which
is classical, and the residue of a plane is the unique projective plane of
order 4; by [A84] and [Ti82] we have the following.

Lemma 5.8.1 The geometry 2T3 is isomorphic to the natural parabolic ge-
ometry of Ut(2), and M3 contains 1/6(2) as a normal subgroup. •

Since F(s6) n F(s3) ^ 2 1 2 JTL 3 (4), F(s3) induces on res£(s 3 ) the full
automorphism group of the latter geometry isomorphic to U^(2).Sym3.
The graph IT3 is strongly regular with the following suborbit diagram
with respect to the action of Ue(2).Sym3:
3+48 135
180 ! S N 128 45
(•>
1
2 +\(U4(2) x 3).2 24+8.[33].22 (l/ 4 (2) x 3).2

The point graph II 4 of ^4 is locally II3 and an important step in the

proof of (5.7.10) is the following characterization of locally 113-graphs
given in [Mei91] (with [BH77] and [DGMP] being credited).

Lemma 5.8.2 The geometry ^4 is uniquely determined up to isomorphism

and M 4 is either the Fischer 3-transposition group M(22) or the extension
M(22).2 of M(22) by an outer automorphism; II 4 is the transposition graph
ofM(22). D

The suborbit diagram of Il 4 with respect to the action of M(22).2 is

the following:
180 567
693 1S \ 512 126
(•>
2.l/6(2).2 22+8.l/4(2).2

In its turn II5 is locally II4 and the next step in the proof is the
following (Proposition 6.2 in [Mei91]).

Lemma 5.8.3 The geometry 3~$ is uniquely determined up to isomorphism

and M5 is the Fischer 3-transposition group M(23); II5 is the transposition
graph of M(23). •

The suborbit diagram of II5 with respect to the action of M(23) is the
following:
238 The Monster
693

2.M(22) 22.U6(2) Q7(3)

Finally 116 is locally IT5 and (5.7.10) (which is Proposition 6.3 in
[Mei91]) completes the picture. In (5.7.10 (i)) 116 is the transposition
graph of M(24) while in (5.7.10 (ii)) it is a 3-fold antipodal cover of the
transposition graph with the following suborbit diagram with respect to
the action of 3 • M(24) (it is straightforward to deduce from this diagram
that of the transposition graph of M(24)).
3510 28431 3510
v 31 671 1 f N 28 160 1080 ,

2xM(23) 22.M(22) Q+(3).2 2.M(22) M(23)

It follows from the above result that the geometries ^"4, 3T^ and ^ 6
as in (5.7.10) are 2-simply connected, a result, originally established in
[Ron81a]. An independent proof of (5.8.2), (5.8.3) and (5.7.10) based on
computer calculations can be found in [BW92a] (see also [BW92b]). In
[Pase94] it was shown that the flag-transitivity assumptions in (5.8.2),
(5.8.3) and (5.7.10) can be removed.
Since a Sylow 2-subgroup of F(s6) = F4 = 2n.Mat24 is of order 2 22 ,
by (5.7.10) we have either F ^ Af(24) or F ^ 3 • M(24). We are going to
show that the former of the possibilities holds by constructing a triple
cover of F = «T(M(24)).
As in the beginning of Section 5.6 let F be the setwise stabilizer in
NG(XS) of the connected component A(XS) containing vo of the subgraph
induced by the vertices fixed by Xs, so that F = F/Xs. For 1 < i < 4 let
Kt and Ft be the preimages in F of Kt and Fu respectively. Then by the
paragraph after the proof of (5.6.5) we have the following:
Ki=Fi~ 3.2i+12 : 3 • C/4(3).22; £ 4 = F4 ~ 2n.(Sym3 x Mat24);
17 3+12
K2 ~ [2 ].(Sym3 x 3 • Sym6); F2 ~ 2 .(L 3 (2) x 3 • Sym6);
K3 - 2 8+6+3 .(L 3 (2) x Sym3 x Sym3); % - 2 6+8 .(L 4 (2) x Sym3 x Sym3).
Since F2 is the normalizer in G3 of a Sylow 3-subgroup from O2j3(G3),
we observe that F2 (and hence F as well) does not split over Xs. Since
F\ n F2 contains a section 3 • Sym^ F\ does not split over Xs either. On
the other hand K3, F3 and K4 = F4 split over Xs. More precisely we
have the following. The number of maximal flags in ^(M(24)) is odd
and hence there is an element q in the Borel subgroup of the action of
 5.8 Fischer groups and their properties 239
F on ^(M(24)) which inverts Xs. It is clear that q is also in the Borel
subgroup of the action of F on 3tf(M(24)). Let H be one of K3, F3 and
K4 = F4. Then Xs is a direct factor of Cg(X s ) and

H := (O 3 (Cg(X s )), 4)

is a complement to X s in H. Furthermore, K3 n K4 is a complement to

Xs in K 3 OK 4 . Put ^ = {Xi,X 2 ,X 3 ,X 4 } and St = {Fi,F 2 ,F 3 ,F 4 }.

Lemma 5.8.4 Let 9(3 • M(24)) = 9(F9s/), 3^(3 • Af(24)) = 9(F9S(). Then
(i) ^(3 • M(24)) is a geometry with the diagram

on which F acts faithfully and flag-transitively;

(ii) J^(3 • M(24)) is a geometry with the diagram

on which F acts faithfully and flag-transitively.

Proof. Since 9(M(7A)) and jf(ht(2A)) are geometries on which F
acts flag-transitively, it is easy to check that the amalgams s/ and 0&
satisfy the conditions in (1.4.1). Hence 9(3 • M(24)) and Jf(3 • M(24))
are geometries; their diagrams follow from the paragraph before the
lemma. •
It is obvious that the natural homomorphism cp : F —• F induces
1-coverings of 9 = 9(3 • M(24)) onto 9(M(24)) and of J f = JT(3 • M(24))
onto 34?(M(24)). We will denote these 1-coverings by the same letter cp.
Let F = F(3 • M(24)) be a graph on the set of elements of type 4 in
3tf in which two vertices are adjacent if they are incident to a common
element of type 3. Let {yt \J. < i < 4} be a maximal flag in jtf such that
cp(yi) = yt. For z € Jtf* let F[z] denote the subgraph in F induced by the
vertices incident to z. Since the residue of % in Jtf* and the residue of y4 in
J f (M(24)) are both isomorphic to Jf(Mat24), we observe that <p induces a
covering of F onto F(M(24)). Furthermore, F[yi] is the coUinearity graph
of the geometry ^ ( 3 - (74(3)) as in (4.14.6) with the suborbit diagram given
in (4.14.10). The morphism (p induces also a 1-covering of the residue of
yi in 5f isomorphic to ^(3 • U4(3)) onto the residue of yi in J f (M(24))
isomorphic to ^((7 4 (3)). It is easy to check that the Schlafli graph does
not possess connected triple covers which are vertex- and edge-transitive.
In view of (4.14.9) this means that cp induces a covering of the geometry
240 The Monster

$(3 • (74(3)) associated with the residue of y\ in 3tf onto the geometry
S(U4(3)) associated with the residue of y\ in Jf(M(24)). This gives the
following.

Lemma 5.8.5 The graph T contains a family £f of Schldfli subgraphs and

under (p a member of £f maps isomorphically onto a subgraph from the
family ^ as in (5.7.3). •

For 1 < i < 4 let % be the subgroup in F4 such that ?iXs/Xs = Y,

(where the YJ are as in the paragraph before (5.7.7)) and let ©; be the
connected component containing y4 of the subgraph in f induced by the
vertices u such that % < O2(F(u)). Then ©, maps onto ®, and by (5.8.5)
©4 maps onto © 4 isomorphically. Let &" =^f(3 • M(24)) be an incidence
system whose elements are the vertices of T and the images under F of
r[?3], ©4, ©3, ©
2 and ©1 with respect to the incidence relation defined
via inclusion. Then ZT is a geometry with the diagram as in (5.7.9) and <p
induces a covering of $~ onto «^"(M (24)). Now by (5.7.10) we obtain the
following.

Proposition 5.8.6 The geometry ^(3 • M(24)) is the universal 2-cover of

?S3-M(24) and F £ M(24).

By (5.8.1), (5.8.2) and (5.8.3) it is easy to deduce the shapes of the

maximal parabolics associated with the action of F on ^~(M(24) (or
equivalent^ of F on «T(3 • Af(24))):

F(si) - 2 x M(23), F(s2) - ( 2 x 2 - M(22)).2,

F(s3) - (2 x 2 2 • U6(2)).Sym39 F(s4) - 2 1+12 .(3 x l/4(2)).22,

F(s5) = F3 - 2 7+8 .(L 4 (2) x Sym3), F(s6) = F4 - 2nMat14.

Proposition 5.8.7 Let !F be one of the following geometries:

Jf(M(24)), 9(3 - M(24)), »(M(24)) and S(M(24)). Then & is simply con-
nected.
Proof. Let & = Jff(M(24)) and let xp : # -^ & be the universal
covering. Let T be a graph on the set of elements of type 4 in & in which
two elements are adjacent if they are incident to a common element of
type 3. For z* G ^ let F[z] be the subgraph in Y induced by the vertices
incident to 'z and let {% \ 1 < i < 4} be a maximal flag in #" such that
 5.8 Fischer groups and their properties 241
xp(yt) = yt. Then xp induces a covering of F onto F = F(M(24)) (denoted
by the same letter xp) and the restriction of xp to T{yi] is an isomorphism
onto T[y\]. Since T[y\] contains a Schlafli subgraph ©from the family
£f as in (5.7.3), there is a subgraph ©in F which maps isomorphically
onto 0. Let &" be an incidence system whose elements are the vertices
of F and the connected components of the preimages of the subgraphs
in F which realize elements of 3~ = ^~(M(24)); the incidence relation is
via inclusion. Then ©is an element of type 4 in ^ , which shows that
&~ is a geometry and that xp induces its covering onto 3T. By (5.8.6) we
have either j r = ^ r o r ^ = ^ = ^"(3 • M (24)). In the latter case f
must be F(3 • M(24)) but since T{y\] is a proper triple cover of T\y\] this
is impossible. Hence xp is an isomorphism. Almost the same argument
shows that Jf(3 • M(24)) is simply connected.
By the above paragraph F is the universal completion of {Ft \ 1 <
i < 4} and F is the universal completion of {FuFi^F^F^}. We claim
that F is also the universal completion of the amalgam {Kt | 1 < i < 4}.
First of all K\ = Fx and K4 = F4. For i = 2 or 3 let P, = O2(Fi). Then
Pi < Ki and it is easy to check that Ft is the unique completion of the
amalgam {NFl(Pi),KhNF4(Pi)}. Hence a completion of {Kt \ 1 < i < 4}
must also be a completion of {Ft \ 1 < i < 4} and the claim follows. In
a similar way one can show that F is also the universal completion of
the amalgam {KuK2,k3,K4} and hence both 9(M(2A)) and 9(3 • M(24))
are simply connected. Finally the residue in ^(M(24)) of an element of
type 4 is isomorphic to ^(MatiA) and it is simply connected by (3.3.11).
Hence K4 is the universal completion of {K4 n fC* | 1 < i < 3} and F is
also the universal completion of {Ki,K2,Ks}. Since the latter amalgam
is a subamalgam of the amalgam of maximal parabolics associated with
the action of F on ^(M(24)) (the paragraph after (5.6.6)), this implies
that S(M(24)) is simply connected. •

As in the beginning of Section 5.6 let Xs be a Sylow 3-subgroup of

#2,3(63). Since the subgraph A(XS) is connected by the definition, (5.8.6)
implies the following.

Lemma 5.8.8 The subgroup of the group G generated by NG{(XS) and

NG2(XS) is isomorphic to 3 • M(24). •

Let % be the preimage of Yt in F4 as introduced after (5.8.5), 1 < i < 4.

Then F(st) (which is the preimage of F(st) in F) is (contained in) the
normalizer of 7, in F. It is easy to see that Y\ and Y2 are conjugate in
242 The Monster
G to subgroups Y\ and Y2 as in Section 5.4 and Section 5.5, respectively,
and we have the following.
Lemma 5.8.9 The Baby Monster group B contains the Fischer group M(23)
and E ^2E6(2).Sym3 contains M(22).2. •

5.9 Geometry of the Held group

In this section we study the normalizer in the group G of a subgroup of
order 7. More specifically we analyse the subgroup in G generated by the
normalizers in G\ and G2 of a subgroup S of order 7 from G12. We start
with the following.

Lemma 5.9.1 Let S be a Sylow 1-subgroup in Gn and let E be the subgraph

of A induced by the vertices fixed by S. Then S is of order 1 and
(i) S is of valency 28,
(ii) there is a subgroup T = Frob] in Gn containing S such that
NGi(S) = NGi(T)for i = 1 and 2,
(iii) NGl(S)/T ^ 2i_+6.AutL3(2), S is fully normalized in G{ and

(iv) S is not fully normalized in G2 and NGl(S)/T ^ CGl(T) ~

[2s].Sym3 x Sym3.
Proof. A vertex u e H(i?o) is fixed by S if and only if r\(u) is fixed by
the image S of S in G\ = Co\ and hence (i) follows from (4.14.14 (i)).
By (4.14.14 (iv)) CQ{(S) is generated by the subgroups r\~l{Ji) taken for
all elements JL G A4 fixed by S. From (4.14.14 (v)) it is easy to deduce
that CQ{(S) is extraspecial of order 2 7 and of plus type. This implies in
particular that Z\ is the kernel of the action of CQl(S) on S(i;0). Let
K be the kernel of the action of NGl(S) on E(v0) and put T = O2(K).
Then T ^ Frob] by (4.14.14 (iii)) and NGl(S) = NGl(T). Since T fixes
/, it is contained in G\ and by the Frattini argument it is centralized
by a subgroup of order 3 from G^ and hence (ii) follows. Now (iii) is
immediate from (4.14.14 (iii)) and (iv) follows from (2.13.5), (5.1.8) and
(3.8.4). D
By (4.14.14 (iii)) and since CQ^S) induces a non-trivial action on S,
we conclude that NGl(S) acts transitively on E(vo) while CGl(S) has
two orbits of length 14 each. Let H = (CGl{T),CG2(T)) and S c be the
subgraph in A induced by the images of vo under H. By the definition
Hc is connected and it is contained in S. Since T is contained in G\ and
normalized by if, it fixes Ec elementwise. Furthermore by (5.9.1) and the
above discussions we obtain the following.
 5.9 Geometry of the Held group 243

Lemma 5.9.2 The following assertions hold:

(i) Hc is of valency 14 and it contains the line I stabilized by G2I
(ii) H acts on S c vertex- and edge-transitively;
(iii) there are seven lines incident to vo and contained in S c ; CGl(T)
induces on the set of these lines the natural action of L^l). •

Let Si be a Sylow 7-subgroup of G3. Then S\ is contained in Gj and

by (5.1.10) we have the following.

Lemma 5.9.3 There is a unique subgroup T\ in G3 containing S\ such

that NG3(Si) = NG3(Ti); SI is not fully normalized in G3 and CGz(T{) =
NG3(Si)/Ti = 2 6 : 3 • Sym6 where O2(CG3(T1)) is the hexacode module. D

Since Si < Gj9 it fixes resJ(S) elementwise (where £ is the element of

type 3 in ^ = ^(M) stabilized by G3). Recall that the elements of type
i in resJ(Z) are the *-closed cliques in A containing Z of size 21 — 1 for
1 = 4 and 5, respectively. Clearly, in such a clique Si fixes a vertex if i = 4
and a line if i = 5. It is easy to see (the paragraph before (5.6.3)) that
these vertices and lines form a subgeometry of 9 isomorphic to resJ(E)
and we have the following.

Lemma 5.9.4 Let *F be the set of vertices of A fixed by S\ and contained

in elements from resJ(E). Then the vertices of *¥ together with the lines
contained in *F form the rank 2 tilde geometry ^(3-Sym^) on which CGl(T\)
induces the full automorphism group with kernel O2(CGl(T\)) = 2 6 . •

By the above lemma Si fixes an incident vertex-line pair (v'Jf) and

hence Si is a Sylow 7-subgroup of a G-conjugate of G12. Thus there is an
element g € G which conjugates Si onto S and maps (1/, /') onto (VQ, I).
Then *¥g is a subgraph in S c isomorphic to the point graph of9(3'Synts).
Since T\ fixes *F elementwise, it is easy to see that Tf — T. Hence *Pg is
stabilized by CG*(T) = 2 6 : 3 • Sym6. By (5.9.2 (iii)) there is a structure
n of a projective plane of order 2 on the set of lines containing vo and
contained in Ec which is preserved by CGl(T). Since CG^(T) n CGl(T)
contains a Sylow 2-subgroup of CGl(T) we see that the lines containing
vo and contained in ^ g form a line in n.
Let &(He) be the incidence system of rank 3 whose elements of type 1,
2 and 3 are the images of *Fg under H, the lines contained in S c and the
vertices of S c , the incidence relation being via inclusion. Then by (5.9.2
(iii)), (5.9.4) and the above paragraph we have the following.
244 The Monster
Lemma 5.9.5 ^(He) is a rank 3 tilde geometry with the diagram

2 2 2

on which H induces a flag-transitive automorphism group. •

Notice that by (5.9.1 (iii)) an element from NGl(T) \ CGl(T) induces
an outer automorphism of H (this element is not an automorphism of
y(He)). Since it is well known and easy to check that Mat24 does not
possesses outer automorphisms we conclude that &(He) is not isomor-
phic to ^(MatiA)- The flag-transitive rank 3 tilde geometries have been
classified in [Hei91], giving the following.

Proposition 5.9.6 The group H is the sporadic simple Held group He of

order

and the geometry &(He) is simply connected. •

By (4.14.14 (vi)) the subgroup Xs as in Section 5.6 is conjugate to a
Sylow 3-subgroup in T and we have the following.

Corollary 5.9.7 The Held group He is a subgroup in the Fischer group

M(24). •

5.10 The Baby Monster graph

We follow notation introduced in Section 5.4 as follows: Y\ is a non-
isotropic subgroup in the orthogonal complement of Z3 in A; B =
(CG1(Y\),CG2(YI)); A ( Y I ) is the subgraph in A induced by the images of
vo under B and B = B/Y\ is the action induced by B on A(Yi). Let Y2 be a
minus 2-space containing Y\ and contained in the orthogonal complement
of Z 3 in A ^ 2 10 as in Section 5.5. Put E = (NGl(Y2lNG2(Y2)) n B and
E = E/Yi. Then E < B and by (5.5.4) E ^ 2-2£6(2).2. The subgraph
S defined before (5.5.3) is the subgraph of A( Yi) induced by the images
of vo under E. By (5.5.4) S is isomorphic to the point graph of the
natural parabolic geometry of E with the suborbit diagram as given in
Section 5.5. By (5.5.3) and (4.10.8) £1 is the full stabilizer of E(v0) in B{
and hence E is the full stabilizer of 3 in B.
Define the Baby Monster graph Q to be the graph whose vertices are
the images of 3 under B, where two such distinct images Si and H2 are
adjacent if there is a A-vertex u in 3i n S2 such that Si is the image of
S2 under an element from O2(B(u)).
 5.10 The Baby Monster graph 245
+22 2+20
Comparing the structures of Bx ^ 2!f .Co2 and Ex s 2 .(76(2).2
we observe that
[O2(Bl):(O2(B1)nE1)]=2
and hence for every vertex u € Sthere is a unique Q-vertex adjacent to
S which is S g for g € O2(2?i) \ E\. It is well known (and easy to deduce
from the suborbit diagram of S) that the action of E on the vertex set
of 3 is primitive, which gives the following.

Lemma 5.10.1 The valency ofQ is 3 968 055 which is the number of vertices
in 3 and also the index of E\ in E. •

Let z denote S as a vertex of Q. By the paragraph before (5.10.1) we

have a bijection
£ : Q(z) -> S
which commutes with the action of E\ (here as usual Q(z) denotes the set
of neighbours of z in Q). Let S2 denote the graph on S in which u, w € S
are adjacent if u e S^w). The suborbit diagram of Sf, as given below,
can be deduced from that of S using some standard relations between
the parameters of symmetric association schemes ([BI84] and [BCN89])
and these calculations were kindly performed for us by D.V. Pasechnik.

2-891 672

43008

[2916]
i r 7
^ V ^ - ^ i 2"-891) [22596]
[21 504]

[18710]

24948

B3(t>o)
246 The Monster
Lemma 5.10.2 Let a,b be vertices from Q(z) such that £(a) and £(b) are
adjacent either in E or in E\. Then a and b are adjacent in £1

Proof. Let Y be the subgraph of S defined in (5.5.1). Then the stabilizer

of T in E is of the form 2 9+16 .Q^(2).2. By the definition Y is induced by
the vertices y e A such that Zy is in the orthogonal complement of Y2 in
A. If y G Y then since D contains O2(G(y)), D also contains an element
which maps z onto £~l(y). This shows that Ye = {z} U ^ ( Y ) is the orbit
of z under Bo := CD(Yi)/Yi = 2 9+16 .Sp 8 (2). By the above the action of
Bo on Te is similar to the action of Sp%(2) on the set of cosets of Ojj"(2).2.
The latter action is doubly transitive (of degree 120) and in view of the
paragraph after the proof of (5.5.4) we obtain the result. •

Let Kz = Yi/Y\. Then Kz is the centre (of order 2) of E and it coin-

cides with the kernel of the action of E on S (equivalently on Q(z)). For
an arbitrary vertex d e Q put Kd = Z{B(d)\ so that Kd is the kernel of
the action of B(d) on Q(d).

Lemma 5.10.3 Let e e Q(z).Then

(i) [Ke,Kz] = 1,
(ii) ifu = {(e) then KeZu = KZZU,
(iii) in terms of (5.10.2) ifL = {Kv\v€ Ye) then L = Z(O2(BO)) S 2 9 .

Proof. Since Kz fixes Q(z) elementwise, it is contained in B(e) and,

since Ke = Z(B(e)\ (i) follows. The definition of the adjacency relation
in Q implies that Ke = Yf/Yi for an element q e 02(B(u)) < Qu. Since
Qu is extraspecial with centre Zu, this gives (ii). In order to see (iii) it is
sufficient to prove that L is isomorphic to the orthogonal complement of
7i in A. D
We need some further properties of the action of E =2Ee(2).Symi on
S (the suborbit diagram in Section 5.5). Let ${11${2)) be the C3-geometry
with the diagram

on which E(vo)/02(E(vo)) induces the full automorphism group. If u e E

then E(u) = 2+~20.l/6(2).Sym3 and the centre of E(u) contains a unique
non-identity element (a root element) which we denote by y(u). The
following result is rather standard.

Lemma 5.10.4 Let X be an orbit of E(v0) on E\ {VQ} and u e X. Then

 5.10 The Baby Monster graph 247

for X being, respectively, S(i?o), B^ivo), S ^ o ) and ^(vo) the following

assertions hold:
(i) the product y(v$) • y(u) has order 2, 2, 4 and 3;
(ii) O2(E{v0)) acts on X with orbits of length 2, 2 6 , 2 11 and 2 2 1 ;
(iii) E(vo)/02(E(vo)) acts on the set of orbits of O2(E(VQ)) on X as it
acts on the set of points, planes, points in ^(Ue(2)) and on a 1-
element set. •

We will need some information about involutions in E.

Lemma 5.10.5 Let T be an involution in E =2E6(2).Syni3, such that

T • y(u) is of order 2 or 4 for every M G S . Then
(i) T 6 E \ Bf and the conjugacy class of E containing % is uniquely
determined,
(ii) C := Q ( T ) ^ F4(2) x 2,
(iii) C has two orbits S (2) and S (4) on S with lengths 69 615 and 3 898 440
and stabilizers of the shape [216].Sp6(2) and [215].t/4(2).2, respec-
tively,
(iv) ifuE 3 ^ for i = 2 or 4, then x • y(u) is of order i,
(v) the subgraph of S induced by S (4) is connected and it is at distance
1 from S<2).

Proof. Using the computer package [GAP] it is straightforward to

calculate the structure constants p^ of the group association scheme of
E. The structure constant has the following meaning. If Q , Q and C, are
conjugacy classes of E then p^ is the number of ways in which a given
element a e C& can be represented as a product b • c where b € Q and
c e Cj. These calculations give everything except (v) and the structure of
the stabilizers. The remaining information is easy to deduce, comparing
the suborbit diagrams of the collinearity graphs of ^( 2 E 6 (2)) and ^(F 4 (2))
given in Section 5.5. •

The quotient E/Kz is a subgroup in E and for u e S the subgroup in

E generated by y(u) is the image of (Kz,K^-i^).

Lemma 5.10.6
(i) Two distinct vertices a,b G Q(z) are adjacent in Q if and only if
£(b) and £(a) are adjacent either in S or in 3^, (equivalently if
y(£(a)) and y(£(b)) commute);
248 The Monster

(ii) if n = (d,e,f) is a 2-path in Q such that d and f are not adja-

cent then the B-orbit containing n is uniquely determined by the
isomorphism type of (Kd,Ke,Kf)/Ke, which is either De = Sym^ or

(iii) E = B(z) has two orbits, Q?2(z) and Q%(z), on the set Q2(z) of
vertices at distance 2 from z, if d G Q2(z) and e G Q(z)nQ(d) then
(Kd,Kz,Ke)/Ke S D2ifor i = 3 and 4;
(iv) if y G Q2(z) f/i^n #(z,}0 flcte transitively on Q,(z)nQ(y).
Proof. Part (i) follows from (5.10.2), (5.10.3 (i)) and (5.10.4). Notice
that r 1 (S 3 (»o)) £ O ^ r 1 ^ ) ) and rH^vo)) s ^ ( r 1 ^ ) ) . Since E(i;0)
acts transitively on S3(t?o) and on S^i^o) we obtain (ii), (iii) and (iv). •
As a corollary of (5.10.6 (i)) and the proof of (5.10.2) we have the
following.

Lemma 5.10.7 Let u G Sffao)- Then E,~l{u) is contained in a unique 120-

vertex complete subgraph Ye which contains z and £~* (*>())• The setwise
stabilizer ofYe in B is the stabilizer of a &(Sps(2))-subgeometry in &(BM)
as in (5.4.5). •
Let S be the graph on the vertex set of S in which u and v are adjacent
if u G E(v) U B^u), so that the edge set of B is the union of the edge sets
of B and S^. Then by (5.10.6 (i)) £ establishes an isomorphism of Q(z)
onto B.
The group E contains M(22).2 as a subgroup (5.8.9). The following
result has been established in [Seg91] (see also Section 5 in [ISa96] and
Section 8 in [Iv92c]).

Lemma 5.10.8 A subgroup in E isomorphic to M(22).2 acting on B has

four orbits L,-, i = 1,2,3,4, with lengths
3510, 142155, 694980, 3127410
and stabilizers isomorphic, respectively, to
2-U6(2).2, 2 10 .AutMat 22 , 21.Sp6(2), 2.(2 9 .PSL 3 (4)).
Furthermore, if w G E; for i = 2, 3 and 4, then w is adjacent in B to 22,
126 and 1 + 2 1 vertices from Si, respectively. •

Lemma 5.10.9 Let u G B3(t?o). Then the subgraph in B induced by B(t?o) n

B(M) is isomorphic to the transposition graph of U^(2) with the suborbit
diagram given after (5.8.1).
 5.10 The Baby Monster graph 249
Proof. Notice that the action of £ on the vertex set of S is equivalent to
its action by conjugation on the set {y(w) | w € S}. Let H = E(vo)nE(u).
Then by (5.10.4 (ii)) if is a (Levi) complement to O2(E(v0)) in E(v0), in
particular H = Ue(2).Syni3 and by the above remark H = C^ivo)) n
C^(y(u)). One can readily see from the suborbit diagrams of S and B^
that n := S(t>0) n S(u) = E%(v0) n S^(u) is of size 693. On the other
hand by the remark at the beginning of the proof, the set II is a union
of some conjugacy classes of H. Since |II| is odd, at least one of the
conjugacy classes must contain central involutions. By [ASei76] the only
class of central involutions in H is the class of 3-transpositions of size
693, contained in the simple subgroup 1/6(2). Since w, v € S are adjacent
if and only if y(w) and y(v) commute, the subgraph induced by II is
exactly the transposition graph of Ue(2). •
Notice that the elements of the C3-geometry ^((/6(2)) can be realized
by the maximal cliques (of size 21), by the 5-vertex cliques contained in
more than one maximal clique and by the vertices of the transposition
graph of 176(2).

Lemma 5.10.10 Let a,b € Q, b e £l\{a) and *F be the connected component

of the subgraph induced by fi(a) n Q(b). Then
(i) *F is locally the transposition graph of Us(2),
(ii) *F is isomorphic to the transposition graph of M(22),
(iii) ¥ = Q(fl) n Q(fc) and B(a, b) s M(22).2,
(iv) every vertex adjacent to a is at distance at most 2 from b,
(v) if(d,e,f,h) is a 3-path in Q such that (Kd,Ke,Kf)/Ke ^ Sym3 then
the distance from d to h is at most 2.
Proof. First assume that z e Q(a) n Q(b) and that £(a) = vo. Then
by the proof of (5.10.6) £(b) € E3(v0) and (i) follows from (5.10.9) and
the fact that t, induces an isomorphism of the subgraph induced by Q(z)
onto S (5.10.6 (i)). By (5.10.6 (iv)) the stabilizer of ¥ in B(a,b) acts
transitively on the vertex set of *F. Let Of be a geometry formed by the
maximal cliques (of size 22) in *F, by the cliques of size 6 contained in
more than one maximal clique, by the edges and the vertices of XF. Then
by the proof of (5.10.9) B(a,z,b) induces on ^(z) an action isomorphic
to C76(2).2. Hence ty is flag-transitive and by (5.8.2) we obtain (ii). From
(ii) we deduce that |*F| = 3 510 and that the stabilizer of *F in B(a,b)
is isomorphic to M(22).2. Now assume that a = z. Then by (5.10.8) the
image of *F in S is the orbit Zi and by the last sentence of (5.10.8) every
vertex in Q(z) is adjacent to a vertex from *F, which gives (iii). Since
250 The Monster
every vertex from Q(a) is at distance at most 1 from *F = Q(a) n Q(fr),
it is at distance at most 2 from b and (iv) follows. Finally (v) is a direct
consequence of (iv) and (5.10.6 (iii)). •
If y e £l\{z) then by the above lemma there is more than one common
neighbour of y and z and in view of (5.10.6) this means that (Ky,Kz) =
Sym^.
Below, II will denote the graph on 2300 vertices introduced before
(4.11.9) on which C02 induces a rank 3 action with the suborbit diagram
given in (4.11.9).
Let Q(Zi) be the subgraph in Q induced by the images of z under
#i = 2++22.Co2. If y G Q(Zi) then the^preimage of Ky in B is a conjugate
Yf of Y2 under an element b e B. Clearly Y2fc < Qi and rj(Yf) is
a singular triangle in the shortest vector graph containing r\(Y\) (this
triangle corresponds to a vertex in the graph II). On the other hand
Y2Z1 contains besides Y2 exactly one image Y% of Y2 under b £ B\ (in
fact b can be taken from O2CB1)), so that Y^/Yi = K^-iM, Hence every
vertex of n corresponds to an edge in fi(Zi) and we obtain the following.

Lemma 5.10.11 The following assertions hold:

(i) the subgraph fl(Zi) contains 4600 vertices;
(ii) every orbit 0/02(^1) on Q(Zi) is of length 2;
(iii) there is a mapping 5 from the set of O2(B\)-orbits on Q(Zi) onto
the vertex set ofH which commutes with the action of B\. •
The subgroups Ky taken for all y € Q(Zi) generate 02(^1), and since
the latter is extraspecial of type 2^_+22 and has different kernels at different
orbits on fi(Zi), we have the following.

Lemma 5.10.12 The following assertions hold:

(i) the group £1 acts on Q(Zi) with suborbits 1, 1, 2-891 and 2-1408;
(ii) for any two vertices d,f contained in Q(Zi), the subgroup (Kd,Kf)
is either abelian (of order 22) or dihedral of order 8 with centre
Zi. •

Lemma 5.10.13 The following assertions hold:

(i) Q(z) n Q(Z0 = £-l({vo} U 3fa>));
(ii) u,v G Q(Zi) are adjacent in Q if and only if the distance in U
between S(u) and S(v) is 0 or 1;
(iii) I / M G fi(Zi) is not adjacent to z then u €
 5.10 The Baby Monster graph 251
e
Proof. Consider the intersection O of Q(Zi) and the subgraph Y as
in (5.10.7). By (5.10.7) B(z,£~l(vo)) acts transitively on the set of images
of such a subgraph under B which contain s := {Z^~1{VQ)}. Hence this
intersection consists of the images of the edge e under / = B\ n B[Ye].
By (5.4.5) the group I is the full preimage in B\ of the stabilizer in
B1/02(B1) ^ Co2 of a ^(Sp 6 (2))-subgeometry in <&(Co2) as in (4.9.8).
From standard properties of ^(Spe(2)) it is easy to deduce that e has
28 images under / and / induces on the set of these images the doubly
transitive action of Sp6(2) on the cosets of Q^"(2).2 and that £(0>\ {z}) c
{vo} U S(i;o). Since Ye is a complete subgraph, in view of (5.10.12 (i)) we
obtain (i) and (ii), and the latter implies that the diameter of Q(Zi) is 2.
Now (iii) follows from (5.10.6) and (5.10.12 (i)). •
By (5.10.6 (ii)) and (5.10.13 (ii)) every vertex y e Q\(z) is contained in
an image of Q(Zi) under an element from B(z).

Lemma 5.10.14 Let y e Q(Zi) n O%(z). Then

1
(i)
(ii)
(iii) Q(z) n &(y) c Q(Zi) and |Q(z) n Q(y)| = 648.
Proof. It is clear that 5(z) nB\ = B{Z^~1{VQ)) is the centralizer of
Zi in 5(z). On the other hand by (5.10.12) Zi is the centre of (Kz,Ky)
which implies (i). By (5.10.11 (ii)) and (5.10.13 (i)) B(z,y) n O2(Bl) has
index 2 in O2(B\) and hence (ii) follows from the suborbit diagram of
IT. Finally (iii) is a direct consequence of (i), (ii), (5.10.13 (i)) and the
suborbit diagram of II. •
The following information can be deduced either from the construction
of the GAB ^(1/4(3)) in [Kan81] or by means of calculations in the Leech
lattice. Recall that in ^(£/ 6 (2)) a point is incident to 21 planes and a
plane is incident to 27 points.

Lemma 5.10.15 Let V ^ U4(3).22 be a subgroup in Aut#(l/ 6 (2)) =

U6(2)Sym3. Then
(i) V has 2 orbits on the set o/891 points of 9(U6(2)) with lengths
324 and 567 and stabilizers PEL3(4) and 25.Syme,
(ii) V has 2 orbits on the set of 693 planes of <g(U6(2)) with lengths
126 and 567 and stabilizers l/4(2).2 x 2 and 25.Sym^
(iii) the point-plane incidence graph has the following diagram with re-
spect to the orbits of V:
252 The Monster

•
Notice that the subgraph in the point-plane incidence graph of
g(Us(2)) induced by the union of the two 567-orbits is the point-plane
incidence graph of ^(1/4(3)) while the subgraph induced by the 567-orbit
on points and the 126-orbit on planes is the point-plane incidence graph

Lemma 5.10.16 Let y e flg(z) n O(Zi). Then B(z,y) S 2^+20.C/4(3).22

acting on Q(z) has eight orbits 0it 1 < i < 8, with lengths 1, 2 • 324,
2 • 567, 2 6 • 126, 2 6 • 567, 2 11 • 324, 2 11 • 567 and 2 21 so that { " H W ) = ^i>
#2 u (p3, r 1 ^ ^ ) ) = o* U <P5, r H s ^ ) ) = <p6 u <p7,
^8 am* #2 = Q(z) n Q(y).
Proof. Since £(z, y) does not contain Kz (which is the kernel of the
action of B(z) on Q(z)), we conclude that O2(B{z,y)) and O2(B{z, i~l(vo)))
have the same orbits on Q(z). Hence the result follows from (5.10.4 (ii),
(iii)) and (5.10.15). •
Lemma 5.10.17 In terms of (5.10.16) let uj € Ojfor 1 < j < 8. Then
(i) if j = l, 2, 3, 5, 6 or 7, then Uj is at distance at most 1 from &2,
(ii) there is a vertex x e (92 such that (Kx,KUi) = Syms,
(iii) wn/ess j = 4 t/ie orbit 0/ is at distance at most 2 from y.
Proof. For j = 1 and 2 the assertion (i) is obvious. For j = 3 (i) follows
from the fact that the subgraph in S induced by S(t;o) is connected, while
for j = 5 it follows from the diagram given in (5.10.15). Let a € S(uo),
b G S^uo) and let a and 5 be the orbits under 02(£(i?o)) °f fl a n d ^>
respectively. Then in view of (5.10.4 (iii)) a and b are points of ^([/6(2)).
One can see from the suborbit diagram of S that whenever a and b are
adjacent in this graph, one has a = b which implies (i) for j = 6. Similarly
one can see from the suborbit diagram of H2 that whenever a and b are
adjacent in S2 the vertices a and b are adjacent in the collinearity graph
of y(U6(2)). Since Ei(b) = O 2 (£i)£i(b) a n d the collinearity graph of
&(U6(2)) is connected, we deduce (i) for ; = 7. Let c e E3(i;o). Then by
the suborbit diagram of S we have |3(i>o) n S3(c)| = 891 which is the
number of orbits of 02(E(vo)) on 3(t;o). Since E(VQ9C) is a complement to
 5.10 The Baby Monster graph 253

02(E(vo)) in E(vo), we conclude that (92 n ^(^(c)) is non-empty, which

implies (ii) in view of (5.10.4 (i)). Finally, since (92 = Q(z) C\ Q(y), (iii)
follows from (i), (ii) and (5.10.10 (v)). •

Lemma 5.10.18 Let u e E62(v0), b = ^(u), *F = Q(b) n Q(Zi), Ye be as

in (5.10.7) and N = (Kw | w e x¥). Then
(i) \V\ = 56 and ¥ = Ye n Q(Zi),
(ii) JV is elementary abelian of order 2 7 and ft contains Z\,
(iii) ffee orftit o/fr wnder O2(Bi) has length 2 6 .
Proof. Since Ye is a complete subgraph, b is adjacent to each of
the 56 vertices in Ye \ ^(Sf(t; 0 )). By (5.10.13 (i)) and (5.10.14 (iii)) ¥ is
contained in {z}u£~l({vo}uE(vo)). By the suborbit diagrams of S and E\
we see that in B u is adjacent to exactly 54 vertices from H(i;0) and hence
(i) follows. By (i) the subgroup N is the orthogonal complement in A of
(ZuKb) which gives (ii). By (5.10.4 (ii)) the orbit of ft under Biz^'^vo))
is of length 2 6 . From (5.10.7) and the suborbit diagram of S we observe
that there are 693 images of Ye under B which contain {z, ^(vo)} and
these images are transitively permuted by B[z,^(vo)]. Since the latter
group is contained in B\ and contains 02(^1), we conclude that O2(#i)
stabilizes Ye and we obtain (iii). •
In terms of the above lemma put S = B\ n 5[T e ]. Then S is the
full preimage in £1 of the stabilizer S S 2^.Sp6(2) of a &(Sp6(2))-
subgeometry from ^(Co2)). Let R denote the action induced by S on
Ye \ Q ( Z i ) . Then by (5.10.18 (iii)) we have R ^ 26.5p6(2). Since O2(R)
is the natural symplectic module for R/02(R) = Spe(2), the latter group
permutes transitively the 63 non-identity elements in O2(R). Let \p de-
note the natural homomorphism of 02(2*1) onto C>2(R). By the above
observation if q e 02(^1) and \p(q) ^ 1 then the length of the orbit of q
under S is divisible by 63. By (4.12.6 (i)) S has 3 orbits Z, Si and S 2 on
the vertex set of II with lengths 28, 2016 and 256, respectively, and the
vertices in S2 are at distance 2 from E in IT. Let E, Zi and E2 be the
preimages of these orbits under <5 (5.10.11). Then Z = Ye nQ(Zi) and by
(5.10.13 (ii), (iii)) the vertices from Z2 are at distance 2 from Z. The latter
implies that Z2 £ ^ ( z ) - Furthermore since 256 is not divisible by 63 we
have xp(Ky) = 1 for y e Z 2 . This means that for every b e Ye \ Q(Zi) the
subgroups Ky and Kb commute. In terms of (5.10.16) b £ G2 and hence
b and y are at distance at least 2. On the other hand b is adjacent to z
and z is at distance 2 from y, hence the distance between b and y in Q is
at most 3. Since [Ky,Kb] = 1 the distance must be 3 by (5.10.6 (iii)) and
by (5.10.17 (iii)) b e (9*.
254 The Monster
Lemma 5.10.19 The group B(z) acts transitively on the set Cls(z) of vertices
at distance 3 from z. Furthermore
(i) a vertex y £ £l\{z) is adjacent to exactly 8064 = 2 6 • 126 vertices in
£l?>{z) transitively permuted by B(z,y),
(ii) ifuen3(z)then[KX9Ku] = l,
(iii) ifve Q(Zi) n Qj(z) and u £ Q3(z) Pi Q(i>), tfeen

Proof. Parts (i) and (ii) follow from the paragraph before the lemma.
In terms of that paragraph in order to prove (iii) we have to analyse the
structure of B\ n B(b) n B(y). We know that the latter is contained in
S ~ 2i +22 .2i +8 .Sp 6 (2). By (5.10.18 (iii)) we have [O 2 (£i) : O2(B1)nB(b)] =
2 6 and it is easy to see that L is the set of vertices in O(Zi) fixed by
O2(Bi)nB(b). In view of (4.12.6 (ii)) this means that B(b)nO2(S) induces
on E2 a regular action of degree 2 9 and (iii) follows. •

Lemma 5.10.20 The diameter ofQ is 3.

Proof. We claim that whenever (xo,xi,X2 = z,X3,X4) is a 4-path in
Q, the distance between xo and X4 is at most 3. Clearly we can assume
that x o ,x 4 £ Q?2(z) UQ^(z). If x 0 € Q|(z) then z G Q|(xo) and by (5.10.10
(iv)) X3 is at distance at most 2 from xo and the claim follows. Hence we
assume that x o ,x 4 e H\{z) and also that £(xi) = v0. Then by (5.10.17 (iii))
unless A := Q(x4) n fl(z) is contained in (P4, there is a vertex in A which
is at distance at most 2 from xo. We show that the inclusion A c ®4 is
not possible. Since x 4 € Q^(z), by (5.10.13 (i)) and (5.10.14 (iii)) A is of
size 648 and contained in ^ -1 (S(w)) for a vertex w £ S. We are going
to show that for every w e S the intersection S(w) n {(©4 ) is of size less
than 648. One can see directly from the suborbit diagram of S given in
Section 5.5 that S(w) n S^uo) is of size less than 648 unless w e S(t;o)
(recall that by (5.10.16) {(0 4 ) c 2|(oo)). If w € S(t;0) then by (5.10.15
(iii)) and the diagram therein we observe that w £ S(^o) is adjacent in
S to 0 or 192 = 6-32 vertices in £($4). This completes the proof of the
claim and also of the lemma. •

Lemma 5.10.21 Let u £ Q3(z) and x be the involution generating Ku.

Then
(i) u is adjacent to 69615 vertices in Qi(z) and to 3 898440 vertices in
Oft*).
(ii) B(u) n B(z) = C B ( 2 ) (T) 3 F4(2) x 22.
 5.10 The Baby Monster graph 255
Proof. By (5.10.19 (ii)) T G B(Z). By (5.10.19), (5.10.20) and the obvious
duality we have

and hence (compare (5.10.6 (iii)) if y is the generator of Kv for v G Q(z),

then the product T 7 is of order 2 if v G 03(1*) and of order 4 if v G Q^")-
By (5.10.5 (ii)) we have C* ( Z ) (T) S F4(2) X 2 2 and from (5.10.5 (iii), (iv))
we deduce the sizes of Q(z)nQ2(w) and Q ( Z ) D Q 3 ( M ) , which gives (i). Now
straightforward calculations give the equality

|Q3(z)| = [B(z) : CB(Z)(T)] = 23 113 728.

Since B(u) D B(z) is obviously contained in CB(Z)(T) the equality proves
(ii). •
By (5.10.6 (i)), (5.10.10), (5.10.14), (5.10.19), (5.10.20) and (5.10.21) we
obtain the main result of the section.

Proposition 5.10.22 The group B is the Baby Monster sporadic simple group
BM of order

2 4i . 313. 56 . 72 . 1 1 . 1 3 . 1 7 . 19 . 23 • 31 • 47

and the suborbit diagram of Q with respect to the action of B is as given

below, in particular Q has 13 571955000 vertices.

142155+694980

M(22).2

2 1+20 .U 4 (3).2 2
1+1134+36288+1161216+2097152
256 The Monster

5.11 The simple connectedness of

The simple connectedness of <g(BM) can be proved following the strat-
egy developed in Section 4.12 using the properties of &(BM) and its
subgeometries established in Sections 5.4, 5.5 and 5.10. In this section we
outline the proof.
Let #" = &(BM) be the P -geometry of the Baby Monster, O =
{xi,...,x 5 } be a maximal flag in & and @ = {Bt \ 1 < i < 5} be the
amalgam of maximal parabolics associated with the action of B = BM
on #", where Bt = B(xt) ((5.4.4) and the paragraph after its proof), and
let ©be the collinearity graph of # \ Let q> : 3F —• 3F be the universal
covering, B be the universal completion of the amalgam ^ which by
(1.5.2) is a flag-transitive automorphism group of J*. Let O = {xi,...,X5}
be a maximal flag in & such that (p(6) = O and let ^ = {%; | 1 < i < 5}
be the amalgam of maximal parabolics associated with the action of the
group B on 3F. Then cp induces a homomorphism xp : B —> B whose
restriction to & is an isomorphism onto ^ , and also a covering x • © —• ©
of graphs where ©is the collinearity graph of J*\

Let Sf be the ^(Sp8(2))-subgeometry in 3F containing X4, S ~

29+16.S/?8(2) be the stabilizer of ^ in B and I be the subgraph in
©induced by the images of *i under S. Then ^ = {S n B t \ 1 < i < 4}
is the amalgam of maximal parabolics associated with the action of S
on Sf (the kernel of the action is 02(5)) and E is isomorphic to the
collinearity graph of Zf. By (1.6.4) 9* is simply connected and hence S
is the universal (and in fact the only) completion of <g, which gives the
following.

Lemma 5.11.1 Let W be the subamalgam in 0$ which maps isomorphically

onto the subamalgam <£ in $, let Sf be the subgroup in B generated by <£
and let E be the subgraph in 0 induced by the images ofx\ under S. Then
the restriction ofxp to S is an isomorphism onto S and the restriction of x
to Z is an isomorphism onto Z. •

Considering Z(O2(S)) as the orthogonal module for S/O2(S) = 5p8(2),

let G be a non-isotropic involution in Z(C>2(S)) which is perpendicular to
Z(O2(B4)). Put

9 = {CBl((7),CB2((7),CB3((7),CS((7)},

let E be the subgroup in B generated by 2 and let S be the subgraph in

©induced by the images of xi under E. Then by (5.5.4) E = 2 2 £ 6 (2).2,
 5.11 The simple connectedness of &(BM) 257
2f is the amalgam of maximal parabolics associated with the natural
action of E on its Tits geometry $ = ^( 2 #6(2)) and E is the collinearity
graph of $ with the suborbit diagram given after the proof of (5.5.4). By
(1.6.4) $ is simply connected and we have the following.

Lemma 5.11.2 Let 3) be the subamalgam in $ U S which maps isomor-

phically onto the subamalgam @ in 31 U S, let E be the subgroup in B
generated by Sf and S be the subgraph in ©induced by the images ofx\
under E. Then the restriction ofxp to E is an isomorphism onto E and the
restriction of % to E is an isomorphism onto E. •

Let Q be the graph whose vertices are the images of S under B in

which two such distinct images Si and S2 are adjacent if there is a
©-vertex u in their intersection such that H2 is the image of Hi under
an element from 02(B(u)). Then since the restriction of xp to B\ is an
isomorphism onto B\ and by (5.11.2) we obtain the following.

Lemma 5.11.3 The covering cp induces a covering rj : Q —• Q of graphs. •

Arguing as in the proof of (5.10.2) one can easily show that the images
of H (considered as a vertex of Q) under S induce in Q a complete
subgraph on 120 vertices and by (5.10.6 (i)) we have the following.

Lemma 5.11.4 Every triangle in Q is contractible with respect to the cov-

ering r\ : Q —> Q. •

Proposition 5.11.5 The following assertions hold:

(i) the Baby Monster graph is triangulable;
(ii) the geometry &(BM) is simply connected.

Proof. It is easy to check the conditions in (1.14.1) using (5.10.10),

(5.10.17) and (5.10.21) in view of (5.10.5 (v)). By (i) and (5.11.4) rj is an
isomorphism which forces cp to be an isomorphism and proves (ii). •
In the remainder of the section we study the action on the Baby
Monster graph Q of the subgroup B\ = 21+22.Co2 of the Baby Monster.
One can see immediately from the suborbit diagram of the Baby
Monster graph that whenever u and v are distinct vertices of Q, KU^=KV.
This enables us to identify u e Q with the non-identity element in Ku.
In this case B(u) = CB(u) and u,v e Q are adjacent if and only if the
product uv is a conjugate of the involution in the centre of B\.
258 The Monster

Lemma 5.11.6 The subgroup B\ = 2++22.Co2 acting on the vertex set of

the Baby Monster graph Q has five orbits denoted by

Q(2a), Q(2rf), Q(4a), Q(4d\ Q(6a)

with lengths
2300-2, 56925 • 26, 46 575-212, 56925-216, 2300 • 222
and stabilizers of shape
2 2+20 .C/ 6 (2).2, [2 26 ].Sp 6 (2), [2 22 ].Aut Mat 2 2 , [2 16 ].Sp 6 (2), 2
respectively. Here when writing n - 2m for the length of an orbit Q(fa) we
mean that Q\ = O2CB1) acting on Q(ia) has n orbits of length 2m each.
Proof. If we put Q(2a) = fi(Zi), then by (5.10.11) the properties of
Q(2a) are as stated in the lemma. Consider the B\ -orbits at distance 1
from Q(2a). If u = £~l{v$) then {z,u} is the orbit of z under Q\ and
since B\ = CB(zu\ we have B(z,u) = B(z)nB\ and hence by (5.10.4)
the orbits of B(z) n Bx on Q(z) \ Q(2a) are the i~\X) for X = 3f (t\>),
S|(«?o) and 23(i;o). Let Q(2d), Q(4d) and Q(6a) denote the orbits of £i
containing ^~\X) for X as above. By (5.10.2) if w e £-\X\ then w
is adjacent to u only if Z = Sf(i;0). Hence by (5.10.4 (ii)) and (5.10.18
(iii)) the orbits of gi on Q(2d), Q(4d) and Q(6a) are of length 26, 212
and 222, respectively. In particular these 3 orbits are different and we
have the following property. Whenever w is at distance 1 from Q(2a),
the subgroup B\(w) acts transitively on the set U(w) := Q(2a) nQ(vv).
By (5.10.18), if w e ^ ( S ^ o ) ) then U(w) has size 56, consists of 28 Qi-
orbits and B\(w) induces on the set of these orbits the doubly transitive
action of Spt(2) on the cosets of l/4(2).2. If w G ^"^S^i^o)) then one
can see from the suborbit diagrams of S and Hi? that Tl(w) is of size
44 and 2?i(w) preserves on this set an imprimitivity system with classes
of size 2. Using (5.10.4 (iii)) it is not difficult to see that B\(w) induces
on the set of imprimitivity blocks the natural 3-fold transitive action of
AutMat22> Finally, by the suborbit diagrams of S and H2 we observe that
z is the only vertex in Q(2a) adjacent to a vertex w G ^(E^vo)), which
completes the description of the B\ -orbits at distance 1 from Q(2a) and
leaves us with 56 925 • 216 vertices whose distance from Q(2a) is at least 2.
We are going to show that the remaining vertices form a single B\ -orbit.
By (5.10.21 (i)) a vertex w G £li{z) is adjacent to 3 898440 vertices in
Q^(z) while every vertex from Q^(z) is contained in a unique image Q(Z{)
of Q(Zi) = Q(2a) under an element from B{z). By the paragraph before
 5.12 The second Monster graph 259
(5.10.19), if w is at distance 1 from Q(Z{), then it is adjacent to exactly
56 vertices in O(Z{). Hence w is at distance 1 from 69615 = 3 898440/56
images of Q(2a) under B(z). Since there is a natural bijection between
the images of Q(2a) under B(z) and the edges incident to z, one can see
from (5.10.21 (i)) that w is at distance more than 1 from Q(2a) if and
only if u e O^iw). This in particular shows that Bi acts transitively on
the set of pairs (z, w) where z e Q(2a), w is at distance 3 from z and at
distance more than 1 from Q(2a). Moreover by (5.10.5 (iii)) the stabilizer
of such a pair is of the form [216].l/4(2).2 (recall that £(z, w) ^ F4(2) x 22).
Hence in order to calculate the length of the orbit of w under B\ it is
sufficient to calculate the number of vertices in Q(2a) at distance 3 from
w. Comparing the suborbit diagrams of S and its subgraph induced by
£(Q(z) n Q3(w)) given in Section 5.5, we easily calculate that the set E of
vertices in £(Q(z) n Q.i(w)) adjacent to £(u) in S is of size 27. A more
detailed analysis shows that £~l(L) U {z} contains all the vertices from
Q(2a) which are at distance 3 from w. Alternatively one can calculate the
structure constants of the group association scheme of BM as in Section
3 in [ISh93a]. It can be checked that Bi(w) induces on ^(L) U {z} the
doubly transitive action of Sp6(2) on the cosets of l/4(2).2. Since the
length of the B\ -orbit containing w turns out to be exactly 56925 • 216
this completes the proof. •
The notation in the above lemma has the following interpretation. If
T = zu is the involution in the centre of B\ and v € Q(ia) then the product
TV is contained in the conjugacy class ice of BM as in [CCNPW]. The
fact that the product is of order i is easily seen from the above proof.
We sketch the proof of the following result (see [ISh93a] for the details).

Lemma 5.11.7 In terms of (5.11.6) let u e Q(ia) and let B(u)' S 2-2£6(2)
be the commutator subgroup of B(u). Then Qi(u) is contained in B(u)f if
and only if ice = Id or Ad.
Proof. It is not difficult to see that B\(u) is not contained in B{u)r. Then
the result follows from the following facts. If fa = Id or Ad then the quo-
tient B\(u)/Q\(u) does not contain subgroups of index 2 while if ioc = 2a,
4a or 6a then the quotient B\(u)/O2{B\(u)) is not a section in B(u)r. •

5.12 The second Monster graph

Let us turn back to the group G which is a faithful completion of a
Monster amalgam, its tilde ^(M) and maximal parabolic J^(M) geome-
260 The Monster
tries and the first Monster graph A which is the collinearity graph of
both 9(M) and Jf(M). Recall that 0 is the^ subgraph of A induced
by the images of vo under B. By (5.10.22) B is an extension of the
Baby Monster sporadic simple group BM by a group of order 2. As
in Section 5.5 let S be the subgraph of A induced by the images of vo
under E = 22'2E6(2).Sym3. J^otice that S is contained in ©and that
the intersection of E with B is of index 3 in E. Let F = F(G) be the
second Monster graph defined as follows. The vertices of F are the images
of 0 under the elements of G; two such images are adjacent if their
intersection is an image of S. By the above there are exactly three images
of 0 under £, which gives the following.

Lemma 5.12.1 The valency of the second Monster graph F is twice the
number of vertices in the Baby Monster graph Q. •
Let t denote 0 considered as a vertex of F. The triple T = {t = t\, ti, £3}
of images of t under E forms a triangle. This triangle and its images
under G will be called lines. Thus there is a natural bijection between the
lines containing t and the vertices of the Baby Monster graph. The group
G acts naturally on F with B being the stabilizer of t. For a vertex x of F
put Lx — Z(G(x)) (a subgroup of order 2) and for a subset X of vertices
in F put Lx = (Lx \ x G X). Then in terms of Section 5.4 Lt = Y\ and
LT — Y2 (elementary abelian of order 22). This gives the following.

Lemma 5.12.2 Let {x, y} be an edge of F. Then

(i) there is a unique line X = {x,y,z} containing {x,y},
(ii) Lx, Ly and Lz are the subgroups of order 2 of the elementary abelian
subgroup Lx of order 22,
(iii) there is a mapping a : T(t) —• Q which commutes with the action of
B and for ueQthe set a~x{u) U {t} is a line,
(iv) Lt fixes T(t) elementwise. •
Let x,y € F(t), X and Y be the lines containing {t,x} and {t,y},
respectively, and suppose that X ^ Y. Let u = a(x) and v = a(y). Since
Kz = Y2/Y1 (see the paragraph before (5.10.3)) we have Ku = Lx/Lt and
Kv = Ly/Lt, which shows that x and y could be adjacent only if either
v G Q(w) or v G Q3(w). We will see below that in the latter case x and y
are not adjacent.
Let D ~ 210+16.Q]h0(2) be as in Section 5.3 and let O be the subgraph of
F induced by the images of t under D. Then by (5.4.1) and the definition
 5.12 The second Monster graph 261
of Y2 we observe that O is a graph on the set of 496 non-isotropic
vectors in A = Z(02(D)) (which is the natural module for D/Oi{D)) in
which 2 vectors are adjacent if they are not perpendicular. Using some
standard properties of strongly regular graphs associated with classical
groups [BvL84] or by means of straightforward calculations one can see
the following.

Lemma 5.12.3 The following assertions hold:

(i) O is a strongly regular graph with the intersection diagram
1+119 128

© —i^y1—®
240 ! r~ \ 119 H2 /" \

(ii) D induces on Q> a rank 3 action of Q^Q(2) on the cosets of Sp%(2);

(iii) the subgraph induced by O(t) is a double antipodal cover of the
complete graph on 120 vertices with the suborbit diagram

together with a matching which joins the antipodal vertices;

(iv) a,b e O(t) are antipodal vertices of the graph in (iii) if and only if
t = a-\- b, equivalently if {t, a, b} is a line in P. •
Define Q to be a graph on T(t) in which two vertices x and y are
adjacent if they are adjacent in T and a(y) G Q(G(X)) (notice that the
latter inclusion implies that a(x) =^= cr(y)).

Lemma 5.12.4 The mapping o induces a covering of graphs Q —> Q (which

we denote by the same letter a). If X = {z,u,v} is a triangle in Q where
u = {-^(tfo), then X is contractible with respect to a if and only if v G

Proof. It follows basically by the definition that <T(®(0) is the complete

120-vertex subgraph Ye in the Baby Monster graph as in the proof of
(5.10.2). Since Te n (THSfo)) and Ye n ^l(E62(v0)) are of size 56 and 64,
respectively, the claim follows directly from the intersection diagram in
(5.12.3 (iii)). •
Next we are intesested in the suborbit diagram of Q with respect to
the action of G(t) - 2 • BM. Let {zjf} = G~\Z) SO that E := G(t,z) ^
22-2E6(2) is a subgroup of index 6 in E. The action of G(t) on T(t)
is equivalent to its action by conjugation on the set {Ly \ y e T(t)}.
262 The Monster
Alternatively this action is similar to that of B = G(t)/Lt on the cosets
of B(z)' = E/Lt s 2-2£6(2). Let fta(z) be an orbit of B(z) on ft \ {z} and
fta(z) be the preimage of fta(z) in ft.

Lemma 5.12.5 / / fta(z) = ft(z) or ft^(z) then fta(z) consists of two E-

orbits; i/fta(z) = ft^(z) or ft3(z) then E acts transitively on fta(z).
Proof. Let y € fta(z) and {y,y'} = G~l{y)> First of all it is easy to
see from the suborbit diagram of the Baby Monster graph in (5.10.22)
that in no cases is B(z,y) contained in the commutator subgroup B(z)f of
B(z) which means that E acts transitively on fta(z) and hence there are at
most two E-orbits on fta(z). Furthermore the number of such orbits is 1
or 2 depending on whether the index nzy of B(z)r C\B(y)r in B(z,y) is 2 or
4, respectively. Again one can see from the structure of stabilizers given
on the suborbit diagram of the Baby Monster graph in (5.10.22) that if
y e ft(z) or y e ft^(z) then B(z,y) s 22+20.l76(2).2 or B(z,y) s M(22).2,
respectively, and nzy = 2 since there are no subgroups of index 4 in
B(z,y). Alternatively the fact that ft(z) consists of two £-orbits follows
from (5.12.4). If y e ft3(z), then B(z,y) ^ Kz x Ky x F where Ky is
not in B(z)\ Kz is not in B(y)' and F ^ F4(2) (5.10.5 (i)) which shows
that nzy = 4. Similarly one can show that nzy = 4 if y e ft*(z) ((5.12.8)
below). •
In view of (5.12.5) and (5.10.22) in order to complete the suborbit
diagram of ft with respect to the action of B(t) = 2 • BM it remains to
determine the number of vertices in £l\(z') adjacent to a given vertex from
ft^(z). In terms of (5.10.8) in the graph S a vertex from S2 is adjacent
to 891 vertices from Zi and a vertex from Z3 is adjacent to 24948 such
vertices. Comparing these numbers with the suborbit diagrams of S and
S2 we conclude that in ft a vertex from Cl\(z) is adjacent to 694980
vertices from ft^fz) and to 142155 vertices from £l\(zr). Thus we obtain
the diagram as given below.

Lemma 5.12.6 If y € ft3(z) then a vertex u G {?,?'} and a vertex v e

{y,y'} are not adjacent in T.
Proof. Let F be the stabilizer of y in E so that F = L j x F where
T = {t,!,y} and F = F4(2). By (5.12.5) F acts transitively on {y,yf} and
hence an element from F conjugates Is- onto L~. Since F is non-abelian
simple and Lt commutes with L-, we conclude that L~ performes the
conjugation which implies that (L~, L~) = D% and the result follows from
(5.12.2 (ii)). " •
 5.12 The second Monster graph 263
69615

44352

By the paragraph after (5.12.2) the above lemma shows that two
vertices from T(t) are adjacent in Q if and only if they are adjacent in F
and not on a line containing t.
For ioc = 2a, 2b, 3a, 4a, 4b, 6a let F*a(z) denote the set of images under
G(z) of the vertices from Q(z) U {z}, O(z'), Q^ 2 )* ^ ( ^ X ^ ( ^ X ^ ( z ' ) 5
respectively. For g e G let F ia (zf) denote the image of Tia(z\) under g.
Notice that in this case T2a(t) = F(t) (the meaning of the notation will
be explained after (5.14.1)).

Lemma 5.12.7 Let u e F ia (t), N = (Lt,Lu) and v € T(t) n F(M). Then

(i) if ice = 2a then N = 22 and all subgroups of order 2 in N are
conjugate to Lt,
(ii) if ion = 2b then N = 2 2 and N contains a conjugate of Z\,
(iii) if ia = 3a then N = D6,
(iv) if ioc = 4a then N = D% and Z(N) is a conjugate of Z\,
(v) if ioc = 4b then N = D% and Z(N) = Lv, in particular v is uniquely
determined,
(vi) ifioi = 6a then N = Dn and Z(N) = Lv, in particular v is uniquely
determined.
In particular T2b(t), F3fl(t), F4a(f), T4b(t) and T6a(t) are the orbits of G(t)
on T2(t).
264 The Monster

Proof. Since L~Lt = L~,LU and in view of (5.12.2), (5.12.5) the result
follows from (5.10*3), (5.10.6) and (5.10.19). •
Let Z be the subgraph of F induced by the images of t under G\. Then
the mapping w i—• Lu establishes a bijection between the vertex set of Z
and the set of subgroups in Q\ conjugate to Y\ in G\. One may observe
that G\ acting on X preserves an equivalence relation with classes of size
2 with respect to which u, v e L are equivalent if and only if
LUZ\ = LVZ\.
The set of equivalence classes is in a correspondence with the set of
vectors in A2 and using (5.12.7) we obtain the following.

Lemma 5.12.8 Let Z be the subgraph ofT induced by the images oft under
G\. Then there is a mapping (p : Z —• A2 which commutes with the action of
G\, the fibres of cp are of size 2 and the quotient ofY* over these fibres is the
shortest vector graph A as in Section 4.11. Furthermore, if JLQ = cp(t) and
{t,tf} = cp-\jiol then {f} U q>-\A4(P«)) <= Y2h{t\ <p~\t?{]*>)) ^ T 4 «(0;
ifs e {Utr} or s e <p-l(A3(po)) then G(t,s) = G(t) D Gx ^ 2 2+22 .Co 2 or
21+22.McL, respectively. •

Let u e T 3a and Xs = O3((Lt9Lu}). Then by the paragraph before (5.8.9)

one can see that Xs is as in (5.8.8). Let F = (NGl(Xs),NG2(Xs)) s 3M(24).
Then using the results established in Section 5.8 together with (5.12.7) it
is not difficult to deduce the following.

Lemma 5.12.9 Let H be the subgraph ofT induced by the images oft under
F = 3 • M(24). Then II is the antipodal triple cover of the transposition
graph ofM(24). Furthermore, U2(t)UU4(t) <= T3a(t) and U3(t) <= T6a(t).
Ifs e U3(t) or s e U4(t) then G(t,s) = G(i) n F S 2 • M(22) or M(23),
respectively. •
Now by (5.12.7), (5.12.8) and (5.12.9) we obtain the following.

Lemma 5.12.10 Let u e Ti<x(t). Then

(i) ifia = 2a then u is adjacent to t and G(t,u) = 2 2 - 2 £ 6 (2),
(ii) ifioL = 2b then \T(u) n T(t)\ = 9200 and G(t,u) S 2 2+22 .Co 2 ,
(iii) ifioc = 3a then \T(u) n T(t)\ = 31 671 and G(t,u) ^ M(23),
(iv) ifia = 4a then \T(u) n r(t)| = 550 and G(t,u) S 2 1+22 .McL,
(v) ifia = 4b then \T(u) n T(t)\ = 1 and G(t,u) ^ 2.F4(2),
(vi) if ioc = 6a then \T(u) n T(t)\ = 1 and G(t, u) S 2 • M(22). •
 5.13 Uniqueness of the Monster amalgam 265
5.13 Uniqueness of the Monster amalgam
In this section we show that all Monster amalgams are isomorphic. The
following result was established in [GMS89], Proposition (2.6).

Lemma 5.13.1 There exist exactly two isomorphism types of groups G\ ~

2i+24.Coi satisfying (5.1.1 (i), (ii)), where Qx = O2(G{).

Proof. We assume that the reader is familiar with standard properties

of extraspecial groups and their automorphism groups [Gri73]. First
we identify the image G\ of G\ in the automorphism group A of Q\
isomorphic to 2 24 • Oj4(2). By (4.5.6) Gi = Co\ preserves on A = Qi/Zi
a unique quadratic form 6, which shows that the subgroup G\/02(A)
in A/O2{A) is uniquely determined up to conjugation and hence G\ is
uniquely determined in A (up to conjugation). Let V be a 212-dimensional
vector space over the complex numbers, L = GL(V) and cp : Q\ —• L
be the unique faithful irreducible representation of Q\. Then NL((P(QI))
realizes all automorphisms of q>(Q\) = Q\ and hence up to conjugation
L contains a unique subgroup G ^ which satisfies the hypotheses of the
lemma.
Now it is clear that G\ is a perfect central extension of G\ by a group
of order 2. Let G\ be the largest perfect central extension of G\ by
an elementary abelian 2-group Y. We claim that Y is of order 4. Let
Y3 be the commutator subgroup of Q\ := O2(G\). Then the mapping
(#i>#2) •-* [^1^2] for qi,q2 G Q\ defines a bilinear map from A = Q\/Y
onto Y3 invariant under the action of Co\. By (4.5.6) there is a unique
such non-zero form, which shows that Y3 is of order at most 2. Next we
observe that Q1/Y3 is a GF(2)-module for Co\ which is an extension of A
by some trivial submodules. Since A is self-dual and the first cohomology
group of A was proved to be trivial in [Gri82], lemma (2.11), we conclude
that Q1/Y3 contains a submodule which maps isomorphically onto A.
Factorizing over this submodule we obtain a perfect central extension of
Co\ by Y/Y3. It is well known that Coo is the universal perfect central
extension of Co\, which shows that Y is of order at most 4. Let q>\
and q>2 be homomorphisms of Coo and G ^ onto Co\ and Gf* be the
subgroup in the direct product Coo x G(/} consisting of the pairs (a, b)
with a e Co0, b e G(t1} and (p\(a) = <p2(&). Then it is easy to see that Gf] is
a perfect central extension of G\ by a group Y of order 4. Let Y\9 Y2, Y3
be the subgroups of order 2 in Y, where Y3 is as above and Y\ = Z(Coo).
Then G ^ / T i = G^\ Gf := Gf/Y2 is the second group satisfying the
266 The Monster
hypotheses of the lemma and G^p := Gf*/Y$ is a (non-split) extension
of Co\ by an elementary abelian group of order 225. •
Consider the action of H = Co\ on its tilde geometry &(Coi). Let
ft ^ 2n.Mat24 and ft = 24+12.(3 • Sym6 x 5ym3) be the stabilizers of a
point and a line incident to the point, respectively. Let J be a Sylow 3-
subgroup in 02,3(ft) anc* & = CO2(H2)(J)- Since 02(ft) is the irreducible
Golay code module for ft/02(ft) = Mat24 and 6>2(ft) n ft has index
2 in 02(/Ji), it follows from (3.8.5) that U is the natural 4-dimensional
symplectic module for S := H2/R = Sp*{2\ where £ is the largest
solvable normal subgroup in ft.
We follow notation introduced in the proof of (5.13.1) and in addition
put G{4) = Co0. For 0 < i < 4 let Hf be the preimage of ft in
G^ (with respect to the natural homomorphism), J® be a Sylow 3-
subgroup in 02,3(^2}) anc* ^ (l) = Q>2(H(I))(^(l))* ^ e a s s u m e that the natural
homomorphism sends H$\ J w and 17^ onto ft, J and U, respectively.

Lemma 5.13.2 For exactly one i G {1,2} wefawetfatf U® is a 5-dimensional

indecomposable module for S.
Proof. Notice that in terms of Section 4.14 J is a conjugate of
OI(NJJ(XS)) and by (4.14.11) J acts on A fixed-point freely. This shows
that l/(0) is an extension of U by the subgroup Y which is the centre
of Gf\ This shows that l/(0) is centralized by the preimage of R in Gf\
Furthermore, since S does not preserve a non-zero quadratic form on U,
we conclude that l/|0) is abelian and hence can be considered as a GF(2)-
module for S. We assume that l/(0 is a quotient of l/(0) for 1 < i < 3.
The first cohomology group of £7 is 1-dimensional [JP76] and since V
is self-dual, this means that U^ is decomposable. We claim that U^ is
indecomposable. Since J acts fixed-point freely on A, the definition of
G(x3) implies that U™ is isomorphic to ZjW. Let H[4) £ 2n.Mat24 be the
preimage of ft in G(4). Then O2(H[4)) is the Golay code module and by
a straightforward calculation we see that U^ is indecomposable. Hence
one of U^ and U^ is decomposable and the other one is not. •

Lemma 5.13.3 In a Monster amalgam M = {Gi,G2, G2} the isomorphism

type of G\ is uniquely determined.
Proof. Let if2 be the preimage of ft in Gi, J be a Sylow 3-subgroup
in O2,3(if2) and U = CO2{H2){J\ By (5.13.1) and (5.13.2) it is sufficient
to show that U is indecomposable with respect to the action of S. We
 5.13 Uniqueness of the Monster amalgam 267
adapt our notation so that J contains the subgroup X of order 3 as in
(5.6.2). Then A := CGl{X)/X ^ 2n.Mat24 and O2(A) is the irreducible
Todd module for A/O2(A) = Mat24. Let J be the image of J in A. Then
NA(J)/O2(A) = 3 • 5p4(2) and by (3.8.5) COl{A)0) is an indecomposable
5-dimensional GF(2)-module for Sp4(2). •
In the proof of the next lemma we make use of some argumentation
in [Th79].

Lemma 5.13.4 The subamalgam {G\,G2} of the Monster amalgam is

uniquely determined up to isomorphism.

Proof. By (5.13.3) the isomorphism type of G\ is uniquely determined.

It is easy to deduce from the proof of (5.1.4) that up to conjugation
in G\ there is a unique subgroup Z2 in Q\ whose normalizer in G\
contains a Sylow 2-subgroup of the latter and by the statement of (5.1.4)
we have G\2 = NG^ZI). Thus G\2 (as a subgroup in G\) is uniquely
determined up to conjugation. Comparing (5.6.2) and the structure of
G2 as given after (5.1.8) we conclude that G2 is generated by G\2 and a
subgroup X of order 3 which normalizes Gj = CGX{Z2) ~ 22+n+22.Mat24,
which is inverted by every element from G2 \ G\, and by (5.6.2) X can
be chosen so that CQ^X) ~ 2n.Mat24. Thus to prove the lemma it is
sufficient to show that the image of X in the outer automorphism group
of Gj is uniquely determined. Suppose that there are two possibilities
X\ = (x\) and X2 = (x2) for X. By (5.1.8) there are exactly three
elementary abelian subgroups of order 211 in Q2/R2 which are normal in
G2/R2 (one of these subgroups is (Q\ n G2)/R2). Moreover, these three
subgroups are transitively permuted by X\ and X2. Hence without loss
of generality we can assume that x^1 normalizes each of these three
subgroups. For i = 1 or 2 put Nt = CG+(Xt). Then by (5.6.2) and our
assumption Nt ~ 2n.Mat24 and O2(Nt) is the irreducible Todd module
for Ni/O2(Ni) ^ Mat24. Hence O2(Ni) < R2. It is well known that the
first cohomology group of the irreducible Golay code module is trivial,
which enables us to assume that R2N\ = R2N2. Since N,- is perfect we
have
Ni = (K2N1)' = (R2N2)' = N2
1
and hence xix^ centralizes JVi = N2. Since N\R2/R2 = Mat24 acts
irreducibly on (Q\ n G2)/R2, we conclude that xix^1 centralizes (Q\ n
G2)/R2 as well as its images under Xt. Finally Z\ is the commutator
subgroup of Qi Pi G2 and hence x\x^} centralizes Z\. Since x\x~^ also
268 The Monster
normalizes Z2 we conclude that the element xix^1 induces the trivial
automorphism of Gj and the result follows. •

Proposition 5.13.5 All Monster amalgams are isomorphic.

Proof. Let JK = {G\, G2, G3} be a Monster amalgam. Then by (5.13.4)
the isomorphism type of the subamalgam {Gi, G2} is uniquely determined.
By (5.1.5) the subgroup Z3 is uniquely determined up to conjugation in
G12 and by (5.1.6) we have G13 = NGl(Z3) and G23 = NG2(Z3). Hence
the subamalgam M = {Gi3,G23} in {Gi,G2} is determined uniquely up
to conjugation and clearly G3 is generated by this subamalgam. Hence
in order to specify the isomorphism type of Jt it is sufficient to indicate
the kernel K of the homomorphism onto G3 of the universal completion
U of the amalgam @. Let G^ = CGl2(Z3). Then G$ ~ 23+6+12+18.3 • Sym6,
G^ is normal in both G13 and G23 (hence it is normal in U). Furthermore
Gn/Gj ^ G23/G$ ^ Sym4, Gi23/G^ ^ D8 and Z 3 is the centre of G^\
It is clear that K n G ] " = l, which means that K < CV{G^) and K
is a complement to Z3 in CJJ(G^). We claim that there is at most one
such complement. Suppose to the contrary that there are two different
complements, say K\ and K2. Then by the homomorphism theorem
Cu(G^)/Z3{Kx DK2) = Z3, which means that if we put L = {G$,Ki DK2)
then the quotient V := U/L is isomorphic to the semidirect product of
Z3 and L3(2\ the latter being the subgroup of the automorphism group
of Z3 (in fact the whole automorphism group) generated by the images
of G13 and G23. So we have that U is generated by its subamalgam
{G13/L, G23/L}. In [Sh88] a very nice lemma was proved asserting that
the semidirect product 23 : L3(2) is never generated by a subamalgam
{PuPi} with Pi = P2 = Sym4 and P\ n P2 = D8. This contradicts
our assumption on existence of two complements. Hence K is uniquely
determined and the result follows. •

5.14 On existence and uniqueness of the Monster

As we have already mentioned, the Monster group M was predicted
to exist in 1973 independently by B. Fischer and R.L. Griess. During
the 70's many properties of the hypothetical group were established. In
particular it was shown that M involves many sporadic groups known by
that time and that the centralizers in M of certain elements of order 2, 3
and 5 involve new sporadic simple groups. These sporadic groups (now
known as the Fischer Baby Monster BM, the Thompson group Th and
 5.14 On existence and uniqueness of the Monster 269
the Harada-Norton group HN) were constructed before the Monster
itself was proved to exist.
In [Gri76] the number 196883 was proved to be a lower bound on
the dimension of a faithful complex representation of M. In [Th79]
J. Thompson proved the uniqueness of the Monster under certain ad-
ditional assumptions, the most crucial one being existence of a faithful
C-module V of dimension 196 883. The proof consists of two principal
steps. In the first step effectively it was shown that the subamalgam
{Gi,G2} in the Monster amalgam is uniquely determined up to isomor-
phism. In the second step it was shown that this amalgam possesses (up
to conjugation) at most one isomorphism into GL(V).
The Monster was constructed by R.L. Griess in [Gri82] as a subgroup
of GL(V) where V is a 196 883-dimensional vector space over the complex
numbers. He started with a rather explicit description of the action of
G\ on V which could be the restriction to G\ of the action of M on
V, in other words he realized G\ ~ 21+24.Co\ as a subgroup in GL(V).
He then found an additional element a e GL(V) which normalizes
Z2 < G\ and together with Gn = ^Gii^i) generates in GL(V) a subgroup
G2 ~ 22+n+22.(SymixMat24) containing Gn with index 3. In constructing
the element a as well as in the identification step a crucial role was played
by a non-associative algebra B (the Griess algebra) which is preserved
by G\ and G2. The existence of this algebra was earlier pointed out
by S.P. Norton who had calculated the values of the (hypothetical at
that time) character of degree 196883. In the stage of identification,
the reductions of V over primes p > 5 were considered. By studying
these reductions it was shown that G\ is the full centralizer of Z\ in
the subgroup M of GL(V) generated by G\ and G2. Finally, application
of results on characterization of groups by their involution centralizers
completed the identification of M with the Monster.
Later, Griess' construction was modified in different directions. In
[Con85] the subgroup G2 was taken as a starting point. This group was
explicitly constructed in terms of a so-called Parker loop. This description
turned out to be convenient enough to define the action of G2 on V. It
was shown that CG-S^I) possesses inside GL(V) three different extensions
to Gi and in a certain sense these extensions correspond to the involutions
in Z2. The mutual consistency of these extensions was shown and again
an isomorphism of the amalgam {Gi, G2} into GL{V) was constructed. In
addition a vector from V (called a transposition vector) was pointed out,
which has only finitely many images under M = {GuGi). The action of
M on the set of these images was proved to be faithful and this implies
270 The Monster
the finiteness of M. In [Ti85] it was proved that the full automorphism
group of the Griess algebra is finite and that G\ is the full centralizer
of Z\ in the automorphism group of this algebra. An exposition of the
construction of the Monster can be found in Chapter 10 of [A94].
In [Iv93b] it was shown that the embedding of the amalgam {Gi,G2}
into GL(V) can be easily extended to an embedding of the whole Monster
amalgam M. That is, it was shown that the subgroup in GL(V) generated
by JVGl(Z3) and NG2(Z3) is of the form G3 - 23+6+12+18.(L3(2) x 3 • Sym6).
Thus a Monster group which possesses a 196 883-dimensional represen-
tation was constructed in [Gri82] but at that time it was not known that
every Monster group possesses such a representation. This was proved by
S.R Norton in [Nor85]. He considered the action of a Monster group M
on the conjugacy class of its involutions (2a-involutions) called the Baby
Monster involutions with centralizers IBM and established the following.

Proposition 5.14.1 Let Y be the set of Baby Monster involutions in M.

Then
\T\ = 97239461142009186000

and for t e T the subgroup M(t) = CM{t) = 2 • BM acting on T \ {t}

has eight orbits Tl0C(t) where ioc = 2a, 2b, 3a, 3c, 4a, 4b, 5a or 6a with
stabilizers of the form 22-2E6(2), 22+22.Co2, Af (23), Th, 2 1+22 .McL, 2.F4(2),
HN or 2 • M(22), respectively. The centralizer algebra corresponding to the
action of M onY has a primitive idempotent of dimension 196883. •

Notation in (5.14.1) has the following meaning: if u e Tia(t) then (in

terms of [CCNPW]) the product t • u belongs to the conjugacy class
ia of the Monster; this notation is consistent with that in (5.12.7). It
follows from the general theory of centralizer algebras that the primitive
idempotent in (5.14.1) is a C-module for M.
Let F = T(M) be the graph on the set F as in (5.14.1) in which t
is adjacent to the vertices in T2a(t). Then T(M) is the second Monster
graph as defined in Section 5.12. In [Nor85] S.P. Norton determined the
orbits of M on the set of all triples {x,y,z} of vertices of the second
Monster graph such that x and y are adjacent. The lengths of these orbits
enabled him to calculate the structure constants of the centralizer algebra
of the action of M on Y and these constants in turn provide the ranks of
the primitive idempotents of the centralizer algebra. The paper [Nor85]
contains extremely important information on the structure of the second
Monster graph but most of the information is given without proof.
 5.15 The simple connectedness of &(M) 271
An independent uniqueness proof for the Monster M was given in
[GMS89]. The assumption was that M is a finite group containing invo-
lutions with centralizers of the form 2++24.Coi and 2 • BM. A graph on
the set of involutions of the Baby Monster type was defined which corre-
sponds to the second Monster graph. The conditions on the centralizers
turned out to be strong enough to reconstruct the structure of the graph.
The use of the information on the second Monster graph established in
[GMS89] was essential in [ASeg92] to prove the following.

Proposition 5.14.2 The second Monster graph T(M) is triangulable. •

5.15 The simple connectedness of

In this section we establish the simple connectedness of the tilde geometry
^ = ^(M) of the Monster group. Let cp : ^ —• ^ be the universal covering
and M be the universal completion of the amalgam of maximal parabolics
associated with the action of M on ^(M). Then M is a flag-transitive
automorphism group of G and cp induces a homomorphism x • M —> M.
Then both M and M are faithful completions of the same Monster
amalgam so we can define second Monster graphs T(M) and T(M) as
at the beginning of Section 5.12 and observe that x induces a morphism
xp : T(M) -+ T(M). By (5.12.4) and (5.12.6) every triangle in T(M) is
contractible with respect to xp. Since T(M) is triangulable by (5.14.2), we
conclude that xp (and hence q> as well) is an isomorphism. Notice that
by (5.12.10) we know that every quadrangle in T(M) is also contractible
with respect to xp, so we only need to know that the fundamental group
of T(M) is generated by the cycles of length 3 and 4. In any case since
by (5.13.5) all Monster amalgams are isomorphic, we have the following.

Proposition 5.15.1 Let J( = {Gi, G2, G3} be a Monster amalgam and G be

a faithful completion of J(. Then G is the Monster sporadic simple group
of order
246 . 330 . 59 . 76 . U 2 . 1 3 3 . 17 . 19 . 23 • 29 • 31 • 41 • 47 • 59 • 71

(so that G\ — CG(ZI)), in particular the tilde geometry <&(M) of the Mon-
ster group M is simply connected. •
 6
From Cn- to TVgeometries

In this chapter we construct an infinite series of flag-transitive tilde ge-

ometries possessing morphisms onto Cn-geometries of symplectic groups
over GF(2). For every n > 2 the series contains one T-geometry f(ri) of
rank n, whose automorphism group is isomorphic to

We also prove that these geometries are 2-simply connected. In Section 6.3
we review some known facts about the dual polar graphs associated with
Cn-geometries of the symplectic groups. In Section 6.4 we consider the
semidirect product W(n) : Sp2n(2) where W(n) is the GF(3)-module
induced from a 1-dimensional non-trivial module of 0^(2) < Sp2n(2).
Using the technique presented in Section 6.1, we show that up to con-
jugation in its automorphism group the semidirect product contains
two subamalgams isomorphic to the amalgam of minimal parabolics
associated with the action of Spin{2) on its Cn-geometry. One of these
subamalgams generates a complement to W(n) while the other one
leads to a Tn-geometry f(n) constructed in Section 6.5. Section 6.6
is devoted to a detailed analysis of the rank 3 geometry /(?>). This
analysis enables us to identify in Section 6.7 the automorphism group
J(n) of f{ri). Since O?,(J(rij) is a submodule of the induced mod-
ule W(n), it contains a family of subgroups of order 3 permuted
doubly transitively by J(n)/03(J(n)) = Sp2n(2). Analysing this family
in Section 6.9 we show that the geometries f{n) are 2-simply con-
nected. In Section 6.10 we characterize the geometries f(n) in terms
of special coverings of the dual polar graphs of the Cw-geometries of
symplectic groups. Finally, in Section 6.11 we show that there are no
flag-transitive T3-geometries possessing morphisms onto the C3 -geometry

272
 6.1 On induced modules 273
6.1 On induced modules
We start with a more detailed (but still a bit informal) discussion of the
strategy for constructing 1-covers of T- and P-geometries implemented
in this chapter.
Let ^ be a T- or P-geometry of rank n and O = {xuX2,...,xn} be a
maximal flag in ^. Let G be a flag-transitive automorphism group of ^,
si — {Pt | 1 < i < n} be the amalgam of minimal parabolic subgroups
in G, where Pt is the stabilizer of the flag 0> \ {*,-}. Let B = f|"=i pi
be the Borel subgroup and Rt = O2(Pi) n B for 1 < i < n. Notice that
Rt = O2(Pt) unless i = n and ^ is a P-geometry. We assume that ^ is
a 2-local geometry of G in the sense that B is a non-trivial 2-group and
NG(Ri) = Pt for 1 < i < n.
We consider a group G which possesses a homomorphism onto G with
kernel W being an elementary abelian 3-group. Moreover, as a GF(3)-
module for G = G/W, the kernel W is induced from a 1-dimensional
module Wo of a subgroup X of G, so that the kernel Xo of Wo is of
index 2 in X. In some cases G splits over W, in some cases it does not.
The crucial step in the construction is to classify in G all subamalgams
si = {Pi: | 1 < i < n) such that the restriction of cp to si is an
isomorphism onto si (we assume that cp{Pi) = Pi). We classify these
subamalgams up to conjugation in the automorphism group of G.
Let si be such an amalgam and B = C\"=iPu so that B maps iso-
morphically onto B under <p. Since W has odd order, B is a Sylow
2-subgroup in the full preimage of B in G and the choice of B is unique
up to conjugation.
For every i, 1 < i < n, there is a unique subgroup Rt in # which maps
isomorphically onto Rf. Since P; = NciRt), it is clear that P, is contained
in N-g(Ri) and by the Frattini argument we have the following.

Lemma 6.1.1 Nd(Ri) nW = CV(#0 and A^RO/CV^) s Pf. •

Thus Pi is a complement to Cw{Ri) in N^(Ri) and if CVW) = 0 then

pt = N*(Ri) is uniquely determined.
For each subamalgam si with the prescribed properties we consider
the subgroup G in G generated by si. For 1 < i < n let Gt denote the
subgroup in G generated by all the P ; except for Pt. Define ^ to be the
geometry whose elements of type i are all the cosets of G; in G and two
cosets are incident if they have a non-empty intersection. In the final
step of the construction we show that ^ is a geometry which possesses a
274 From Cn- to Tn-geometries
1-covering onto ^ and specify the diagram of ^ . The covering is proper
unless G is isomorphic to G. In the case of such an isomorphism G is
a complement to W in G and this is certainly impossible if G does not
split over W.
Thus in order to realize the strategy outlined, we should be able
to construct non-split extensions of groups by induced modules and
to calculate centralizers of various subgroups in such modules. In the
remainder of the section we discuss the necessary machinery.
It is well known that if G is a finite group and U is a GF(p)-module
for G then the non-split extension U • G exists if and only if the second
cohomology group H2(G, U) is non-trivial. By the Eckmann-Shapiro
lemma (Shapiro lemma in [Bro82]) the cohomology group of the induced
module is isomorphic to that of the original module. This gives the
following.

Lemma 6.1.2 Let G be a group, X be a subgroup in G. Let Wo be a

1-dimensional GF(3)-module for X and W be the GF(3)-module for G
induced from Wo. Then a non-split extension W • G exists if and only if
there exists a non-split extension Wo • X. •

In the concrete situations we consider, the kernel Xo of Wo in X is

a non-abelian simple group and the question about non-split extensions
Wo - X reduces to consideration of the 3-part of the Schur multiplier of
X0.
Let us turn to the calculation of the centralizers. We consider a slightly
more general situation. Let G be a group, X, Y be subgroups of G, Xo
be a subgroup of index 2 in X and IF be a field, whose characteristic is
not 2. Let Wo be a 1-dimensional IF-space, turned into an X-module by
the following rule: the elements from Xo centralize Wo and every element
from X \ Xo inverts Wo. Let W be the module for G induced from Wo.
We are interested in the dimension of Cw(Y).
Since W is an induced module, it possesses the direct sum decomposi-
tion

W = G) Wu

where the W\ are 1-dimensional F-spaces indexed by the cosets from

& = G/X. The group G acting on W permutes the subspaces W\ in the
way it permutes the cosets in ^ \ If io denotes the coset X-l, then Wt0 and
Wo are isomorphic as X-modules. For an arbitrary coset i = X • g e 3~
 6.1 On induced modules 275
8
let Xt = X be the stabilizer in G of this coset. Then the elements from
Xo}1 := XQ centralize W\ while every element from Xt \ Xo,,- inverts Wt.
Let T be an orbit of 7 on . f and WT = 0 l € T Wt.

Lemma 6.1.3 CwT{Y) is an W-subspace in Wj, whose dimension is 1 if

Y f]Xt< XOii for every i G T and 0 otherwise.
Proof. LetQ = (J i e T wf be the set of all non-zero elements contained
in the subspaces W\. Since WT is a direct sum of the Wi9 for every
element w G WT there is a uniquely determined subset Q(w) ^ Q such
that |Q(w) n VF/| < 1 for every i € T and w is equal to the sum of all
elements in Q(w). Moreover, w G CwT(Y) if and only if 7 stabilizes Q(w)
as a whole. Suppose first that Y n Xt is not contained in Xo?l for some
(and hence for all) i € «^". Then Y n Xf contains an element inverting
WTf. Hence if u e W\ then the orbit of u under Y contains — u. Thus
in this case there are no non-zero elements in WT centralized by Y. If
Y nXi < Xo,i, then Y n X; centralizes Wi and for every non-zero element
u e Wi the sum a{u) of all its images under Y is a non-trivial element
from CV r (Y). Furthermore, for u,v G W^ and A,/i G IF we have
XG(U) + /x(j(f) = o(hi + )Ui^).

Finally, since T is a 7-orbit, every orbit of Y on Q intersects wf and

the result follows. •

Definition 6.1.4 An orbit TofYonF will be called untwisted if Y nX, <

Xo,j for every i G T and twisted otherwise.
By (6.1.3) we have the following.

Lemma 6.1.5 In the above notation Cw(Y) is an W-subspace in W whose

dimension is equal to the number of untwisted orbits ofY on ZT = G/X. •
Let us give a reformulation of the above result in terms of complex
characters. Notice that the number of untwisted orbits has nothing to do
with the field IF and it is uniquely determined by the triple (X,Xo, Y) of
subgroups in G.

Lemma 6.1.6 In the above terms assume that F = C Let xo be the char-
acter of Wo and x = Xo be the induced character of W. Then

•
276 From Cn- to Tn-geometries
When constructing the covers we will always take IF to be GF(3), but
the above lemma shows that the calculations of the centralizers can be
carried out in the complex number field (provided that the corresponding
characters are known). Finally (6.1.5) implies a purely group-theoretical
condition for the triviality of Cw(Y).

Lemma 6.1.7 In the above terms Cw(Y) is trivial if and only if for every
geGwe have Y * Pi X J= Y * n Xo. •

6.2 A characterization of 9(3 Sp4(2))

In this section we apply the strategy outlined in the previous section to
construct the rank 2 T-geometry ^(3 • Sp4(2)) as a 1-cover of the gen-
eralized quadrangle &(Sp4(2)) of order (2,2). This construction provides
us with a characterization of the rank 2 T-geometry as well as with a
background for construction of the infinite series of T-geometries.

Hypothesis 6.2.1 G is a group having a normal subgroup Y of order 3 such

that G/Y = Syme. The centralizer CofY in G is a perfect group having
index 2 in G. This means that C is a non-split central extension ofAlte by
Y and Y is inverted by the elements ofG which map onto odd permutations
ofSym6.

We will prove a sequence of lemmas which are of independent interest

and imply the following.

Proposition 6.2.2 There exists a unique (up to isomorphism) group G which

satisfies Hypothesis 6.2.1, and G is the automorphism group of the rank 2
T-geometry 9(3 • Sp4(2)).

First observe that by (2.6.1) the automorphism group G of the rank 2

T-geometry satisfies the conditions of Hypothesis 6.2.1.
We are going to construct a group which contains every group satisfying
Hypothesis 6.2.1 as a subgroup. Let G = Sym6 = Sp4(2)\ X be a
subgroup in G isomorphic to Syms and Xo be the commutator subgroup
of X isomorphic to Alts. Let Q = {1,2,3,4,5,6} denote the set of cosets
of X in G. Let X, be the stabilizer in G of the coset i and Xo,i be
the commutator subgroup of Xt (which is the intersection of Xt with
the commutator subgroup of G, isomorphic to Alte). We assume that
Xi=X.
 6.2 A characterization of&(3 • Sp4(2)) 277
Let Wo be a 1-dimensional GF(3)-vector-space on which X acts by the
following rule: the elements from Xo centralize Wo while every element
from X\Xo inverts Wo. Let W be the GF(3)-module for G induced from
Wo. Then W possesses the direct sum decomposition

into 1-dimensional submodules permuted naturally by G. Moreover W\

and Wo are isomorphic as X-modules. Furthermore, Xt and Xo,t are the
normalizer and the centralizer of W\ in G, respectively, for 1 < i < 6.
Notice that whenever an odd element g € G normalizes a subspace Wu
g inverts W\.
When restricted to G = Alt6 the module W becomes the permutational
module of G acting on Q. The centralizer of G in W is a 1-dimensional
submodule Z. A generator z of Z is the sum over the orbit under G
of a non-zero element from W\ for some i e Q. It is clear that Z is a
submodule for G, inverted by every odd element (i.e. by every element
from G \ GO.

Lemma 6.2.3 Let G be a group which possesses a homomorphism cp onto G

whose kernel is elementary abelian of order 36 and as a GF(3)-module for
G = Sym$ the kernel is isomorphic to the module W defined above. Then
G is isomorphic to the semidirect product W : G = 36 :
Proof. The 3-part of the Schur multiplier of Alts is trivial and hence
by (6.1.2) G splits over the kernel of q> and the result follows. •

Lemma 6.2.4 Let G be a group satisfying Hypothesis 6.2.1. Then G is

isomorphic to a subgroup of the group G defined above.
Proof. By the construction there is an action of G on W. By Hypoth-
esis 6.2.1 there is a surjective homomorphism xp of G onto G. Define the
action of G on W so that g E G acts as \p(g) e G (in particular Y is the
kernel of the action). Let G be the semidirect product of W and G with
respect to this action. Then Y and Z are normal subgroups of order
3 in G. Each of these two subgroups is centralized by the commutator
subgroup of G and is inverted by every element outside the commutator
subgroup. Let U = (Y,Z). Then U is elementary abelian of order 9 and
besides Y and Z it contains two "diagonal" subgroups of order 3, which
we denote by U\ and I/2. Then for i = 1 and 2 the subgroup Ut is normal
in G and l/jO W = l/jOG = 1. Hence both W and G map isomorphically
278 From Cn- to Tn-geometries
onto their images in G, := G/17,-. It is easy to check that G,- satisfies the
hypothesis of (6.2.3) and hence G,- = G which means that G contains a
subgroup isomorphic to G. Notice that if a is the automorphism of G
which centralizes G and inverts every element in W, then <r(l/,-) = t/3_,-
for i = 1 and 2. D

Lemma 6.2.5 Let si = {Pi,P2} &e the amalgam of minimal (which are also
maximal) parabolic subgroups corresponding to the action of G = Syme
on &(Sp4(2)). Then every group G satisfying Hypothesis 6.2.1 is gener-
ated by its subamalgam si which maps isomorphically onto si under the
homomorphism of G onto G.

Proof. We have P1^P2 = Sym4 x 2 and B = Px n P 2 = As x 2 is a

Sylow 2-subgroup of G. Let #, = O2(Pi). Then #, is elementary abelian of
order 2 3 , it contains odd elements of G and P, = No(Ri) for i =1 and 2.
Let i/> be a homomorphism of G onto G and let £ be a Sylow 2-subgroup
in \p~l(B). Then the restriction tpi of xp to JB is an isomorphism onto B.
Let Rt = tpf^R,) for i = 1 and 2. Since Rt contains odd elements of G, we
have Cy(Ri) = 1. Now the Frattini argument (compare (6.1.1)) implies
that the subamalgam si = {N~(R\),N~(R2)} maps isomorphically onto
si. Since G does not split over Y, this subamalgam generates the whole
of G. •

Lemma 6.2.6 The group G (up to conjugation in its automorphism group)

contains at most two subamalgams which map isomorphically under cp :
G —> G onto the subamalgam si of minimal parabolic subgroups associated
with the action of G on ^(Sp4{2)). One of these subamalgams generates a
complement to W in G.

Proof. Let si = {PuP2}, B = P{ nP2, Ri = 02{Pi\ so that Pt = NG(Ri)

for i = 1 and 2. Let si = {Pi,P2} be the subamalgam in G which
maps isomorphically onto si under cp. Since G = W : G, at least one
such subamalgam is contained in the complement to W. Notice that
B := Pi DP2 is a Sylow 2-subgroup in the full preimage of B in G. Hence
B is uniquely determined up to conjugation and without loss of generality
we assume that B is contained in the complement G to W in G. The
restriction q>\ of cp to B is an isomorphism onto B. For i = 1 and 2 let
Rt = q>Yl(Ri) and Nt = iVg(^)- Then Rt < JV, and by Frattini argument
Nif n W = Cw(Ri) and Ni/Cw(Ri) = Pi- Thus we have to calculate the
centralizers in W of R\ and R2.
 6.2 A characterization of<0(3 • Sp4(2)) 279
Without loss of generality we assume that Pi is the stabilizer in G of
the partition Q = {1,2} U {3,4} U {5,6} and P 2 is the stabilizer in G of the
pair {1,2}. Then the orbits of R\ on ft are the pairs forming the partition.
The transposition (1,2) is contained in R\ and it stabilizes the subspace
Wt for i > 3. Since this transposition is an odd element, it inverts every
subspace Wt which it normalizes. Since the transposition (3,4) is also in
Ru it is clear that all orbits of R\ on ft are twisted.
The orbits of R2 on ft are {1,2} and {3,4,5,6}. Since the transposition
(1,2) is contained in R2 as well, the orbit of length 4 is twisted. On the
other hand since every odd element from #2 switches the elements in the
orbit of length 2, this orbit is untwisted.
Thus Pi = JVi is uniquely determined and it is contained in the
complement G to W in G. But N2 is an extension of P2 by the subgroup
Cw{Ri) of order 3. Since P2 contains the transposition (3,4), it induces a
non-trivial action on Cw(Ri). This implies that N2/R2 is an elementary
abelian group of order 3 2 , extended by an involutory fixed-point free
automorphism. By the definition P2 is a subgroup of index 3 in N2
containing the Sylow 2-subgroup B of N2 which maps isomorphically
onto P2 under cp. It is easy to see that there are exactly four subgroups
of index 3 in N2 containing J5, say Sj for 1 < j < 4. One of them,
say S4, has non-trivial intersection with W and for this reason cannot
map isomorphically onto P2. Each of the remaining three subgroups
does map isomorphically onto P2 and hence forms together with Pi a
subamalgam which maps isomorphically onto s/. We can assume that 53
is contained in the complement G to W in G and hence {Pi, S3} generates
this complement. Let a be the automorphism of G which commutes with
the complement G and inverts every element from W. Then it is easy
to see that <x(S/) = S$-j for j = 1 and 2. Hence up to conjugation in
the automorphism group of G there are two subamalgams J / which
map isomorphically onto srf and one of them generates a complement
to W. •
Now (6.2.2) follows from (6.2.4), (6.2.5) and (6.2.6). As a corollary we
observe that the subamalgams {Pi,Si} and {Pi,S3} are not conjugate
in the automorphism group of G since they generate non-isomorphic
subgroups.
Notice that G = Sym^ contains two conjugacy classes of subgroups
isomorphic to Syms fused in the automorphism group of G. So there are
two possibilities for the module W which are equivalent with respect to
the automorphism group. On the other hand the outer automorphism of
280 From Cn- to Tn-geometries
G performs a duality of the geometry ^(Sp 4 (2)). In what follows we will
need some information on the action of the parabolics Pi and P2 on the
module W irrelevant to the above duality. This information is contained
in the following two lemmas. The former one is a reformulation of (6.2.6).

Lemma 6.2.7 Consider ^(Sp^l)) as a geometry of 1- and 2-dimensional

totally singular subspaces of a ^-dimensional GF(2)-space V with respect
to a non-singular symplectic form. Let Pi be the stabilizer in G = Sp4(2) =
Syme of a totally singular (3 — i)-dimensional subspace and let Rt = 02(^0
for i = 1, 2. Let W be a GF(3)-module for G induced from the unique
non-trivial 1-dimensional module Wo for a subgroup X = 0^(2) = Syms
in G which stabilizes in V a quadratic form. Then Cw(Ri) is trivial and
Cw{Ri) is 1-dimensional. •

Lemma 6.2.8 In terms of (6.2.7) let G = W : G. Then, up to conjugation

in the automorphism group of G, there are exactly two subamalgams in G
which map isomorphically onto the subamalgam $4 of minimal parabolic
subgroups associated with the action of G on &(Sp4(2)). Moreover, one
of the subamalgams generates a complement to W while the other one
generates a subgroup isomorphic to the automorphism group 3 • Sp^{2) of
the rank 2 T-geometry.
Proof. By (6.2.6) up to conjugation there are two subamalgams J / in G
with the prescribed properties and one of them generates a complement to
W. The automorphism group G of ^(3 • Sp4(2)) satisfies Hypothesis 6.2.1
and by (6.2.4) G is a subgroup of G. Finally by (6.2.5) G is generated
by a subamalgam J / which maps isomorphically onto J / and the result
follows. •

6.3 Dual polar graphs

In this section we introduce notation to be used till the end of the chapter
and discuss some further properties of Cw(2)-geometries.
Let V be a 2n-dimensional GF(2)-space, n > 3, and *¥ be a non-singular
symplectic form on V. If {v\9...,v\9v\9...,v*} is a (symplectic) basis of V
then we can take

Let 9 = y(Sp2n(2)) be the Cn(2)-geometry associated with the pair ( F , ¥ )

(Section 1.8). Since n > 3, by (1.6.5) G = Sp2n(2) is the only flag-transitive
automorphism group of (3.
 6.3 Dual polar graphs 281
Let F be the dual polar graph of ^ whose vertices are the elements
of type n in ^ and two vertices are adjacent if they are incident to a
common element of type n — 1. In other terms the vertices of F are
maximal (n-dimensional) totally singular subspaces in V and two such
subspaces are adjacent if their intersection is (n — l)-dimensional.
The following result can be found in [BCN89], Theorem 9.4.3.

Lemma 6.3.1 With F and G as above the following assertions hold:

(i) F is distance-transitive of diameter n with the following intersection

numbers: a = at = [}] 2 , bt = 2 ([J] 2 - [|] 2 ) for 0 < i < n;
(ii) G acts distance-transitively on F ;
(iii) for x, y G F we have d(x9 y) = i for 0 <i <n if and only if
dim (xDy) = n — i;
(iv) every edge ofT is in a unique triangle and whenever T is a triangle
and x is a vertex there is a unique vertex y G T such that d(x, T) =
d(x,y);
(v) ifxeT then the subgraph induced on Tn(x) is connected. •

Let U be an element in ^ of type n — i where 0 < i < n — 1. Define

r(C7) to be the subgraph of T induced on the vertices which are incident
to (which means contain) U. It is easy to see that T(U) consists of a
single vertex if i = 0, otherwise it is isomorphic to the dual polar graph
of the geometry ^(Sp2i(2)) associated with the pair (U^/U,*?') where ¥ '
is the form induced by *F on UL/U. The subgraph T(U) will be called a
geometrical subgraph of type n—i in F. The geometrical subgraphs of type
n — 2 are isomorphic to the point graph of the generalized quadrangle
of order (2,2) and will be called quads. The whole graph F can be
considered as a geometrical subgraph of type 0. Since F(I7) ^ T(W)
if and only if W < U9 it is easy to see that the mapping t/ •—> T(U)
establishes an isomorphism of ^ onto the geometry whose elements of
type j are geometrical subgraphs in F of type j , I < j < n, and the
incidence relation is via inclusion.
Since a geometrical subgraph of type n — i is isomorphic to the dual
polar graph associated with &(Sp2i(2)) and since by (6.3.1 (i)) the inter-
section parameters c\ and at are independent of n we have the following.

Lemma 6.3.2 If x, y G F with d(x, y) = i then T(x D y) is the unique

geometrical subgraph of type n — i containing x and y. Every geometrical
subgraph in F is strongly geodetically closed. •
282 From Cn- to Tn-geometries

Proposition 6.3.3 The geometry &(Sp2n(2)) is 2-simply connected.

Proof. Let ^ be the incidence system of rank n which is the universal
2-cover of ^ = ^(Sp2n(2)), let <p : 5 -> ^ be the universal 2-covering and
let F be the graph on the set of elements of type n in ^ in which two such
elements are adjacent if they are incident to a common element of type
n — 1. Let {Fi,..., Vn} be a maximal flag in ^ where Vt is of type z. Since
res^(Fn) is a projective geometry, it is 2-simply connected. In addition any
two elements of type n in ^ are incident to at most one common element
of type n—1. From this it is easy to conclude that (p induces a covering of
F onto F (which we denote by the same letter cp). Since cp is a 2-covering,
the subgraph in F induced by the vertices incident to an element of
type n — 2 maps isomorphically onto a quad in F. In view of (6.3.2) this
implies that the cycles in F of length 3, 4 and 5 are contractible with
respect to q>. Thus to prove the proposition it is sufficient to show that
the fundamental group of F is generated by its cycles of length 3 and 4.
The latter is equivalent to the statement that every non-degenerate cycle
in F can be decomposed into triangles and quadrangles. We proceed by
induction and assume that the statement is true for all geometries under
consideration of rank less than n. By (6.3.2) every non-degenerate cycle
of length 2i or 2i + 1 is contained in a geometrical subgraph of type n — i
and hence unless i = n it is decomposable by the induction hypothesis.
Let C = (yo,yi9...9yn-uyn,yn+u~,y2n = yo) be a non-degenerate cycle of
length 2n, which means that yn G Tn(yo\ yn-i,yn+i € Tn-i(y0). We claim
that there is a vertex z £ Fn_2(yo) which is adjacent to both yn-\ and
yn+i. In fact, considering the vertices in C as subspaces in V which are
maximal isotropic with respect to *¥, we can put
z = (yn-i n yn+u(yn-\ n yn+i^ n y0)
and it is straightforward to check that z possesses the required properties.
Hence C can be decomposed into a quadrangle and two cycles of length
2n — 2. Now let D = (yo,-9y2n+i) be a non-degenerate cycle of length
2n + 1. Then by (6.3.1 (iv)) the unique vertex z adjacent to both yn and
yn+i is contained in Tn-i(yo)9 which shows that D is decomposable into
a triangle and two cycles of length 2n. •
Certainly the above proposition is nothing but a special case of (1.6.4)
and we present a proof for the sake of completeness and to illustrate on
an easy example the technique of decomposing cycles.
As above let O = {Vi,V2,..., Vn} be a maximal flag in ^. Let si —
{Pi | 1 < i < n} be the amalgam of minimal parabolic subgroups in G
 6.3 Dual polar graphs 283
associated with <D, & = {Gt | 1 < i < n} be the amalgam of maximal
parabolics and # = {P |; | 1 < i < j < n} be the amalgam of rank 2
parabolics. Let B = P|"=1 P,- be the Borel subgroup. For 1 < fe < n let
#(*) = {pfj | 1 < i < j < n,k £ {ij}}. By (6.3.3) ^ is 2-simply connected.
Since every rank 2 parabolic P*, is contained in at least one of Gn_2, Gn_i
and Gn, we have the following.

Lemma 6.3.4 In the above terms G = Sp2n(2) is the universal completion

of the amalgam <& as well as of the amalgam {Gn_2, Gw_i,Gw}. If k = n
or n — 1 then the parabolic Gk is the universal completion of the amalgam

Let Qt = O2(Gi) and Rt = O2(Pi) for 1 < i < n. Then Qt is the kernel of
the action of G, on res^(F;) and Gt/Qi s Lj(2) x Sp2n-2i(2) for 1 < i < n
where Li(2) and Spo(2) are assumed to be the identity groups.
The element x = Vn is a vertex of Y and Gw = G(x) is its stabilizer
in G. We need some more detailed information on the structure of this
parabolic. As usual let Gj(x) denote the elementwise stabilizer in G of all
the vertices which are at distance at most i from x in F.

Lemma 6.3.5 The following assertions hold:

(i) ify e rn(x) then H := G(x)nG(y) is isomorphic to Ln(2); x can be
considered as the natural module for H, in which case y is the dual
of the natural module; G(x) is the semidirect product of Qn and H;
(ii) G2(x) = l ;
(iii) Qn is elementary abelian of order 2^n2+n^2;
(iv) Qn is a quotient of the permutational GF(2)-module ofH acting on
the set of non-zero vectors in x;
(v) Qn/Gi(x) is isomorphic to x as a module for H;
(vi) Gi(x) is isomorphic to /\2 x as a module for H.

Proof. The group Gn = G(x) induces the full automorphism group

Ln(2) of the projective space res^(x). The kernel of the action is contained
in the Borel subgroup which is a 2-group. Hence the kernel is exactly
Qn. In terms of the symplectic basis we can put x = (t?},...,^) and
y = (i£ ...,!>;*) in which case it is clear that G(x) n G(y) = GL(x) ^ Ln{2)
is a complement to Qn in Gn and (i) follows. One can see from the
intersection numbers of F that any two vertices in F have at most three
common neighbours. If g E G2(x) and z G F 3 (x) with zg ^ z then seven
vertices in F2(x) n F(z) must be common neighbours of z and z g which
284 From Cn- to Tn-geometries
is impossible. Hence G2(x) = G$(x) and (ii) follows from (9.1.4). Thus Qn
acts faithfully on {x} U T(x) U ^ ( x ) i.e. on the union of quads containing
x. Since Qn stabilizes every such quad as a whole, by (2.5.3 (v)) Qn is
elementary abelian. By (i) and (6.3.1 (ii)) Qn acts regularly on Tn(x) whose
size is 2("2+n)/2 and (iii) follows.
Let y e Tn(x). Then there is a bijection xp between the non-zero vectors
in x and the vertices in T(y) n Tn(x). To wit, for u G x # we have

V>(u) == (ynM±>w + Mi)> where wi G j ; \ ( j n I T 1 ) .

It is straightforward to check that the transvection t(w, {u)-1) is contained
in Qn and maps y onto I/)(M). By (6.3.1 (v)) the subgraph induced by Fn(x)
is connected and hence such transvections taken for all non-zero vectors
u in x generate a subgroup acting transitively on Fn(x). This subgroup is
clearly Qn and (iv) follows.
Considering the transvections t(u, (w)-1), it is easy to observe that
Qn/G\(x) is non-trivial. Let n x be the protective space dual to res^(x),
so that the points of IT* are the triangles in T containing x. For such
a triangle T let K(T) be the elementwise stabilizer of T in Qn. Then
K(T) has index 2 in Qn. Triangles Tu T2, T3 form a line in IIX if and
only if they are contained in a common quad. By (2.5.3 (iv)), in this case
K(Ti) nK(T2) < K(T3). Hence the dual of Qn/Gx{x) supports a natural
representation of II* and (v) follows from (1.11.1).
By (ii), (iii) and (v) G\(x) is non-trivial and it acts faithfully on ^(x).
By (2.5.3 (vi)) the dual of G\(x) is generated by subgroups of order 2
indexed by the quads containing x, i.e. by the lines of n x . By (v) such
subgroups corresponding to quads containing a given vertex z € Fi(x)
generate the natural module of (G(x) n G(z))/O2(G(x) n G(z)) ^ Ln_i(2),
so (vi) follows from (2.4.6). •
Notice that in terms of (2.4.7) Qn is isomorphic to the quotient
W/(W\W3) of the permutational GF(2)-module W of H acting on
1-dimensional subspaces of x.
We formulate the following direct consequence of (6.3.5) and (2.4.7
(vi)).

Lemma 6.3.6 Let E be the quad in F stabilized by Gn-2. Let H be the

full preimage in Gn-2 of the subgroup of index 2 in Gn-2/Qn-2 isomorphic
to Ln-2(2) x Alte. Then Qn contains elements which are not contained in
H. •
 6.4 Embedding the symplectic amalgam 285
6.4 Embedding the symplectic amalgam
We follow the notation introduced in the previous section. In addition
let A be the set of quadratic forms of minus type on V associated with
*F in the sense that if / G A then

for v, u G V (notice that since the characteristic is 2 there is no difference

between plus and minus). The group G acts transitively on A; the sta-
bilizer O(f) of / G A in G is the orthogonal group 0^(2) containing a
subgroup Q(/) of index 2 which is the non-abelian simple group
We have

The following result (p. xii in [CCNPW]) enables one to distinguish

the elements from Q(/) and those from O(f) \ Q(/).

Lemma 6.4.1 An element g G O(f) is contained in Q(f) if and only if the

dimension of Cy(g) is even. •
Recall that Q\ is the kernel of the action of G\ on V\ = V^/V\. Since
V\ is 1-dimensional, V\ is (2n — 2)-dimensional, *¥ induces on this space
a non-singular symplectic form *F and G\/Q\ = Sp2n-i(2) acts as the full
stabilizer of *F. Let c be the transvection with centre w and axis (w)1,
then it is easy to see that c is in the centre of Q\.

Lemma 6.4.2 Let a be a non-identity element in Q\. Then

j - si / \ \ In —I Jif a=c;
dim Cv(a) = < .
(^ 2n — 2 otherwise.

Proof. Since a acts trivially on V\9 we have va—v G V\ for every v G Fj1.
Suppose first that a acts trivially on Vf~. Then va - v G (Ff1)1- = V\ for
every v G V. Hence a = c and V(a) = Fj1. Now suppose that a acts non-
trivially on Fj1. Then X = Cy(a) n Fj 1 has dimension 2n — 2. Suppose
that a centralizes a vector v G F \ Ff1. Then for every u G F 1 \ X we
have *F(i;a, ua) = ^(i;, u-\-w)^= *F(t;, u) where w is the non-zero element
from V\. This contradiction shows that Cy(a) = X. •
Let /o G A and O(/0) = O^"n(2) be the stabilizer of / 0 in G. Let JF0
be a 1-dimensional GF(3)-space on which 0(/o) acts by the following
rule: all elements from Q(/o) centralize Wo and every element from
286 From Cn- to Tn-geometries

0(/o) \ ft(/o) inverts Wo. Then Wo becomes the unique non-trivial 1-

dimensional GF(3)-module for O(/ 0 ). Let W = W(n) be the GF(3)-
module for G induced from Wo. Then W possesses the direct sum
decomposition

feA
where Wj is a 1-dimensional subspace whose normalizer and centralizer
in G are O(f) and fi(/), respectively.
Let G = W : G be the semidirect product of W = W(n) and G
with respect to the natural action and q> : G —• G be the canonical
homomorphism. We are going to classify the subamalgams in G which
map isomorphically onto J / under q>. By Section 6.1 in order to classify
these subamalgams we have to describe the centralizers in W of the
subgroups Rt for 1 < i < n. We proceed by induction on n noticing that
W(2) is the module for Sp4(2) = Syme as in Section 6.2 and we assume
that n > 3.
Our nearest goal is to describe Cw(Qi). Since G\ is the stabilizer in
G of a 1-dimensional subspace in V, by Witt's theorem (Theorem 7.4
in [Tay92]), G\ has exactly two orbits on A. One of these orbits, say
Ai, contains all forms from A vanishing on V\ and the other orbit, say
A2, contains the remaining forms. We claim that c G O(f) for every
/ G A2. In fact, c acts trivially on Vf- and for v € V \ V^ we have
f(vc) = f(v + w) = f(v) + /(w) + W&w). Since both /(w) and Y(i>, w)
are equal to 1, f(xc) = f(x) and the claim follows. By (6.4.1) and (6.4.2)
c £ Q(/) and hence we have the following.

Lemma 6.4.3 Every Qi-orbit in A2 is twisted. •

Now let / G Ai. If a e Q\ then fa coincides with / on V^~. But one

can easily check that there is a unique form g G Ai distinct from / which
coincides with / on V^- and the element c switches / and g. This and
(6.4.2) imply the following.

Lemma 6.4.4 Every Q\-orbit on Ai is untwisted and of length 2. •

Every form / G Ai induces a quadratic form / of minus type on V\

and g = / if and only if g = / or g = fc. Hence G\ = G\/Q\ acts on
the set of pairs {/,/c} as it acts on the set of minus forms defined on
V\ and associated with *F. Hence Cw(Qi) is a module for G\ induced
from a 1-dimensional GF(3)-module for its subgroup O(f) = 0^_ 2 (2).
 6.4 Embedding the symplectic amalgam 287
Since G\ n 0(f) is a maximal parabolic subgroup in 0(f) (the stabilizer
of the 1-dimensional subspace V\ isotropic with respect to / ) , it contains
a Sylow 2-subgroup of 0(/). This means that Ai is a twisted orbit for
G\ and the above mentioned 1-dimensional GF(3)-module for 0{f) is
non-trivial. This gives the following.

Lemma 6.4.5 As a GF(3)-module for Gi = Gi/Qi S Sp2n-2(2) the central-

izer Cw(Qi) is isomorphic to the induced module W(n— 1). In particular
dim CwiQi) = 2"-2(2w"1 - 1). •
For 1 < j < n - 1 let Nj = f]{=1 Gt so that Gi = JVi > N2 > ... >
Nn-i = Pn. It is easy to see that 02(iVi+i) > O2(iV,-) and the image of
JVf+i in Ni/02(Nt) = Sp2n-2i(2) is a maximal parabolic subgroup which
is the stabilizer of a 1-dimensional subspace in the (2n — 2i)-dimensional
space V^/Vj for 1 < i < n — 2. By (6.4.5) this enables us to calculate the
dimensions of the CV(02W-)) inductively and we arrive at the following
lemma (recall that Rt = O2(P,))-

Lemma 6.4.6 dim Cw(O2(Ni)) = 2n-i-l(2n~i - 1) and in particular

dim Cw(Rn) = 1. •
Notice that C^(O2(iVn_2)) is the 6-dimensional module W(2) associated
with Nn-2/O2(Nn-2) = Sp4(2) and hence the latter equality in the above
lemma is consistent with (6.2.7).
An important role in the subsequent construction will be played by
the following.

Lemma 6.4.7 The parabolic Gn_i normalizes but does not centralize C =
Cw(Rn). °
Proof. Let U = Fw_i, S be the set of forms in A vanishing on U and

/el

Then by the arguments before (6.4.6) Cw(Rn) ^ E. Since the forms in 2

are of minus type, U is a maximal totally singular subspace to each of
them. We claim that Qn_i acts transitively on Z. Recall that an element
from G is contained in Qn-\ if and only if it acts trivially both on U and
on U^/U. Let f,g G i . Then both / and g vanish on U and equal 1 on
U1- \ U. There is a linear transformation a of V which maps / onto g. By
Witt's theorem there is a linear transformation b of V which preserves
288 From Cn- to Tn~geometries
g and realizes (a\u±)~l on (I/- 1 ) 0 . Then ab maps / onto g (and hence
preserves the form *F) and acts trivially on U^. Hence ab e Qn-\ and the
claim follows.
Since X is a Qn-\-orbit, dim CsiQu-i) < 1. On the other hand C <
CE(Qn-i) since g n _i < Rn. Hence C = CE(Qn-\)' Since £ is invariant
under Gn_i and Qw_i is normal in Gn_i, we conclude that C is invariant
under Gn_i. Finally by the arguments as before (6.4.6) the action of Gn_i
on C is non-trivial. D

Lemma 6.4.8 Cw(Qn) = 0.

Proof. Since Vn is maximal totally isotropic with respect to *F, for
any two forms / and g from A their restrictions to Vn have radicals of
codimension 1. Hence these restrictions are equivalent in the sense that
there is a linear transformation of Vn which maps one restriction onto
the other. Now arguments similar to those from the proof of the previous
lemma show that there is an element in Gn which maps / onto g. Hence
Gn is transitive on A and the orbits of Qn on A are either all twisted or all
untwisted. On the other hand there are twisted orbits since the element
c from the centre of Q\ acts trivially on Vn and hence c G Qn. •
Now we are ready to prove the main result of the section.

Proposition 6.4.9 The group G=W : G = W(n) : Sp2n(2) contain^ (up to

conjugation in its automorphism group) at most two subamalgams si which
map isomorphically onto si under cp. One of the subamalgams is contained
in the complement G to W.
Proof. Let si = {Pt | 1 < iI < n} be as stated and B be the Borel
subgroup of si. Then B is uniquely determined up^to conjugation since
it is a Sylow 2-subgroup of G. We assume that B is contained in the
complement G to W. Let Ri be the preimage of Ri in B for 1 < i < n. For
1 < i < n—1 Rt contains Qn whose centralizer in W is trivial by (6.4.8) and
hence Pi = N~(Ri) is uniquely determined. By (6.4.6) dim Cw(Rn) = 1 and
arguing as in the proof of (6.2.6) we see that there are three candidates
for Pn. One of then, say Si, is contained in the complement to W while
two others, say S2 and S3, are permuted by the automorphism of G which
commutes with G and inverts W. •

6.5 Constructing T-geometries

We use the notation as in the previous section. Let si = {Pi | 1 < i < n}
denote the subamalgam in G which maps isomorphically onto ^ under
 6.5 Constructing T-geometries 289
the canonical homomorphism cp and such that, in terms of that of the
proof of (6.4.9) Pn = S2. For k = n and n — 1 put

h = (Pi \l<i<n,i^ k).

Lemma 6.5.1 Under q> the subamalgam 2 = {Jn>Jn-\} maps isomorphically

onto the subamalgam 3) = {Gn, Gn_i}.
Proof. The preimage Qn of Qn in B is contained and normal in P, for
every 1 < i < n — 1. Hence Jn < N^(Qn) and by (6.4.8) the restriction of
cp to Jn is an isomorphism onto Gn. This implies that (P,- | 1 < i < n — 2)
(which is in Jn n Jn-\) maps isomorphically onto Gn n Gw_i. This means
that all we have to show is that Jn-\ maps isomorphically onto Gn_i.
Let C = Cw(Rn) and F = (Gn-uC). By (6.4.7) F = C : Gn_i and since
Qn-i < Rn, C is centralized by g n _i. Since C is centralized by Rn and
normalized by Gn_i, it is easy to see that C is centralized by G+_x and
hence the latter is normal in F. Then F/G^_x is an elementary abelian
group of order 9 extended by a fixed-point free involution. Now it is
easy to see that Jn-\ and Pn have the same image in F/G+_x isomorphic
to Sym^. This shows that Jn_i maps isomorphically onto Gn_i and the
result follows. •

Lemma 6.5.2 Let

Then the quotient <Sfw,n-i/02('S>x,n-i) is isomorphic to the automorphism group

of the geometry <g(3 • Sp4(2)).
Proof. Let Q\ be the preimage of Q\ in B. Then by (6.4.5)

Nd(Qi)/Qi = CwiQO : {G./Q,) s W(n - 1) : 5p2w_2(2).

Clearly Pn and Pn_i are contained in iVg(Qi). Thus proceeding by induc-

tion on n, we reduce the calculations into the group W(2) : Sp4(2) and
apply (6.2.8). •
As above let T be the dual polar graph of ^. Every element U e & can
be identified with the geometrical subgraph r(L7) so that the incidence
relation is via inclusion. Let T® = r(Vj). Then T (n) = {x} is a vertex,
r ( n - 1 ) is a triangle, r (n ~ 2) is the point graph of the generalized quadrangle
of order (2,2) and for 1 < i < n—3 T® is the dual polar graph of resJ(Ff).
Every edge of r is in a unique triangle which represents an element of
type n — 1 incident to both ends of the edge. The parabolic Gn is the
290 From Cn- to Tn-geometries
stabilizer of x in G and it induces the natural doubly transitive action of
Ln(2) on the set of triangles containing x. For 1 < i < n — 2 the subgraph
T(l) can also be defined as the one induced on all the images of x under
G, and since ^ is connected we have G, = (Gn n G,, Gn_i n G,-).
Let J = J(n) be the subgroup of G = G(n) generated by 3) = {Jn, J n - i } .
By (6.5.1) the restriction (p\ of the canonical homomorphism q> to 3) is
an isomorphism onto 3) = {Gn, G n _i}. Define J, to be the subgroup in J
generated by q>Yl(Gn n G,-) and qh[\Gn-\ n G,-) for 1 < i < n - 2. Let f be
the graph whose vertices are the cosets of Jn in J and two vertices are
adjacent if they intersect in a common coset of J n _i. Then by (6.5.1) the
restriction of cp to J induces a covering xp of T onto T. Let F (n) = {x}
be the vertex Jn • 1 and F^" 1 ^ be the triangle formed by the vertices
intersecting Jn-\ • 1. Then obviously \p(x) = x and ^ ( F ^ " ^ ) = F ^ " ^ . Let
T® be the subgraph in F induced by all the images of x under J,. Because
of (6.5.1) F (l) can equivalently be defined as the connected component
containing x of \p~l(T®).
Let / = #{ri) be the geometry whose elements are all the vertices and
triangles of F as well as all the images under J of the subgraphs T®
for 1 < i < n — 2. The incidence relation is via inclusion and the type
function is inherited from (S.

Proposition 6.5.3 # = #(ri) is a T -geometry of rank n, possessing a 1-

covering onto (3y and J = J(n) acts on / faithfully and flag-transitively.
Proof. The only claim we still have to prove is that / is a I -
geometry. For this we have to show that F^n~2^ is the point graph of
#(3 • Sp4(2)). Let L be the kernel of the action of Gn_2 on resJ(F w _ 2 ).
Then L is the largest subgroup of Gn n Gn_i which is normal in Gw_2
and Gn-2/L ^ G+_2 ^ Sp4(2). Let L = (p\l{L). Then L is the largest
subgroup of Jn n Jn-\ normal in Jw_2. Then Jn-i/L = S n , n _iL/L and
Sn,n-i HL = O2(Sn,n-i). Hence by (6.5.2) we have Jn-2/L ^ 3 • Sp4(2). D
It follows from the construction that {Jt | 1 < i < n} is the amalgam
of maximal parabolics which corresponds to the action of J on t/.

6.6 The rank 3 case

Let $ = #(ri) and J = J(n) be as in (6.5.3). In this and the next
two sections we specify the structure of J and show that # is 2-simply
connected.
 6.6 The rank 3 case 291
Recall that, if J f is a geometry, we write ^f' for the set of elements
of type i in Jtf* and, if © is a flag in J^, then resjf (©)' denotes the
set of elements of type i incident to ©. Let K be the kernel of the
homomorphism of J onto G = Sp2n(2), which is the restriction of q>.
Then K = J n W, in particular K is an elementary abelian 3-group. It
is easy to deduce from the proof of (6.5.3) that Gn-i n K is of order 3.
In view of the flag-transitivity this implies that for every element x of
type n — 2 in $ the intersection Kx of K and the stabilizer of x in J is a
subgroup of order 3.

Lemma 6.6.1 K = (Kx \ x e f n ~ 2 ) .

Proof. It is clear that L = (Kx | x € #n~2) is normal in J and

the image in J/L of the amalgam {Jn_2, Jn-u Jn} is isomorphic to the
subamalgam {Gn_2, Gn_i, GM} in G. By (6.3.4) J / L = G = Sp2n(2) and the
result follows. •

We are intending show that in order to generate K it is sufficient to

take the subgroups Kx for all elements x of type n — 2 incident to a
fixed element of type n. For this purpose we use the following result
established in [Hei91] and independently in [ISh89b] (in both cases
computer calculations were used).

Proposition 6.6.2 Let ^ be a T-geometry of rank 3 and ¥ be a flag-

transitive automorphism group oftf*. Suppose that !F possesses a l-covering
onto &(Sp6(2)) which commutes with the action ofF. Then F is isomorphic
to a non-split extension 3 7 • Spe(2)f Oi(F) is isomorphic to the E-j-lattice
taken modulo 3 and F/Oi(F) = Spe(2) acts irreducibly on Os(F) as a
subgroup of Cox(E7) S Sp6(2) x 2. •

This immediately gives

Corollary 6.6.3 The geometry /(?>) is simply connected and J(3) is isomor-
phic to the group F from (6.6.2). •

In the remainder of the section we deal with the case n = 3 only.

Let y be an element of type 3 in / stabilized by J3 in J. Since J3 maps
isomorphically onto a maximal parabolic in Sp$(2) and J3 = 2 6 : L3(2) by
(6.3.5), where O2(h) is a direct summand of the permutational module
of ^3/02(^3) = L3(2) acting naturally and doubly transitively on the
elements of type 1 incident to y. For x G / 1 let Kx be the subgroup of
292 From Cn- to Tn~geometries
order 3 in K as in (6.6.1). Then J3 stabilizes as a whole the set
1
By = {Kx I x G r
of seven such subgroups indexed by the elements of type 1 incident
to y. The subgroup OiiJi) normalizes every Kx G By. Moreover, q G
02{Ji) inverts Kx if and only if q projects onto an odd element in
J(x)/O2,3(J(x)) = Symt and by (6.3.6) there are such elements in 02C/3).
Since Ji/OiiJi) permutes the subgroups in By doubly transitively, this
shows that different subgroups in By have different centralizers in 02(«/3).
Hence the subgroups in By are linearly independent and hence they
generate the whole of K.

Lemma 6.6.4 Let n = 3 and ye,/3. Then the subgroups in the set
By = {Kx I x G res/();) 1 } are linearly independent and generate K. •
The set By of subgroups from K as in (6.6.4) will be called the special
basis ofK associated with y. Of course, in order to obtain a basis in the
ordinary sense one should choose a non-trivial vector from each of the
subspaces in By, but most of our arguments are independent of such a
choice (a similar convention will be assumed for n > 4). In particular
for every cyclic subgroup M G K there is a well defined support of M
with respect to By which is the set of elements from x G res/(j;)1 such
that a vector from Kx has a non-zero component in a decomposition
of m G M # in the basis consisting of vectors from the Kx. By (6.6.2)
there is a quadratic form f on K which is preserved by J/K and one
may observe that in the basis By the form / can be written as a sum of
squares.
Let x and z be elements of type 1 in / which are in the same K-
orbit. Then, since K is abelian, we have Kx = Kz. On the other hand
K-orbits on f1 are indexed by the elements of 91. Since G = Spe(2) acts
primitively on the set ^ 1 of size 63, we conclude that the set
B = {Kx I x G Z 1 }
contains 63 subgroups and we know that 7 of them are in By.
The structure of 02^3) implies that for any subset in By of even
size there is exactly one element in 02(^3) which inverts every subspace
from the subset and centralizes the remaining subspaces. In addition
there is a unique protective plane structure n(y) on By which is pre-
served by Ji/OiiJi). This specifies K as a module for J3 and enables
one to calculate the orbits of J3 on the set of 1-dimensional sub-
 6.7 Identification ofJ(n) 293
spaces in K. Clearly B \ By is a union of some of these orbits. The
calculations (which are quite elementary) show that only one union
of orbits has size 56, which is an orbit itself. To wit, for a subgroup
from this orbit its support with respect to By is a complement of a
line in n(y). For further reference we state this fact explicitly in the
following.

Lemma 6.6.5 Let n = 3, y e f3 and z e f1 so that Kz g By. Then the

support of Kz with respect to By is the complement of a line in n(y). •

This lemma provides us with a rule for rewriting one special basis in
terms of another one. Let z e #2 and y, / , y" be the elements of type
3 incident to z. Then / = {Kx \ x € resz(z)1} is of size 3. It is clear
that / = By n By n By> and that / is a line in %{y\ %{y') and n{y"). For
i = 1 and 2 let v\ be distinct elements of type 1 incident to y' but not
to z and a, be non-trivial elements from KVr Since KVi is orthogonal to
every subspace in /, by (6.6.5) the support of at in the basis By is the
complement of / in n(y). Since a\ and ai are orthogonal, a\a2 has two
non-zero coordinates in the basis By. Finally, the set D of elements in K
whose support is n(y)\l has size 24 and it is closed under taking inverses.
Thus 23 subgroups of order 3 in K have support n(y) \ I. On the other
hand 23 such subgroups are in (By' \ By) U {By \By) and hence we have
the following.

Lemma 6.6.6 Let n = 3, z e #2 and {y9y\y"} = res/(z)3. Let KVl

and KV2 be distinct subgroups from By \ By. Let a\ and ai be non-trivial
elements from KVl and KV2, respectively. Then for i = 1 and 2 the support
of a\ with respect to the special basis By is the 4-element set By \ By; the
product a\a2 in this basis has exactly two non-zero components. Moreover,
every subgroup of order 3 in K whose support is By \ By is contained in
ByUBy. •

6.7 Identification of J(n)

We follow notation introduced in Section 6.5 and assume that n > 4.
By (6.6.2) if u e fn~3 then res^(u) is isomorphic to f(3) which is the
T-geometry of the group 37 • Sp6(2). This implies in particular, that the
subgroups in {Kx \ x e res/(w)n~2} generate in K an elementary abelian
subgroup of order 37. Moreover, by (6.6.4), if y is an element of type n
294 From Cn- to Tn-geometries

incident to u then
{Kx | x G re S/ ({ W ^}r 2 }
is a special basis for this subgroup.

For an element y of type n in / let

By = {Kx | x G re S / (y) w - 2 }.

Let y and yf be elements of type n which are incident to a common

element z of type n — 1. We are intending to express the non-trivial
elements from every subgroup Kv e By in terms of elements contained
in the subgroups from By. If v is incident to z then Kv G By and the
expression is obvious. So suppose that v and z are not incident. Since
r e s / ( / ) is a projective geometry, in this case there is a unique element
u of type n — 3 which is incident to both v and z. Since u is incident to
z, it is also incident to y and / . For pairwise incident elements u, z and
y of type n — 3, n — 1 and n, respectively, let JB(u, z, y) denote the set of
subgroups Kx for all elements x of type n — 2 which are incident to u
and y but not to z. It is easy to see from the diagram of / that B(u,z,y)
contains four different subgroups. Now (6.6.6) gives the following.

Proposition 6.7.1 In the above terms suppose that n > 4. Let z G #n~l
and {y,y',y"} = res/(z) n . For i = 1 and 2 let v\ be an element of type
n — 2 incident to yr but not to z, let ut be the unique element of type
n — 3 incident to both vt and z (so that KVi G B(uu z, y')) and let a\ be a
non-trivial element from KVi. Then
(i) a\ is a product of four non-trivial elements taken from different
subgroups in B(u\,z,y),
(ii) ifui + u2 then B(uuz,yf) n JJ(ii 2 ,z,/) = 0,
(iii) ifu\ = ui then a\a2 is a product of exactly two non-trivial elements
taken from different subgroups in B(u\9z,y),
(iv) if d is any product of four non-trivial elements taken from different
subgroups in B(ui,z,y), then d G Kx for some Kx G B(u\,z9yf) U
B(uuz9y"). •

Proposition 6.7.2 In the above terms if y G $n and Ky = (Kx | x G

2
) then K = Ky.

Proof. By (6.7.1) if y' is incident with y to a common element of

type n — 1 then Ky < Ky and hence Ky = Ky because of the obvious
 6.8 A special class of subgroups in J(n) 295
symmetry. Now the result immediately follows from the connectedness
of/. •
Now we are ready to specify the structure of J(n).

Proposition 6.7.3 If n > 3 then J(n) is a non-split extension of G(n) =

Sp2n(2) by an elementary abelian 3-group K(n) of rank o(ri) = [^h- The
action of J(n)/K(n) on K(n) is irreducible.

Proof. In view of (6.6.2) we can assume that n > 4. Let y be an

element of type n in # stabilized by Jn in J. By (6.7.2) K is generated by
the order 3 subgroups from the set By = {Kx | x e res/(>;)n~2} whose size
is exactly a(n). The subgroup Jn maps isomorphically onto the parabolic
Gn in G. From the basic properties of the latter group it is easy to deduce
(6.3.6) that for every Kx e By there is an element in OiiJn) which inverts
Kx. On the other hand Jn/O2(Jn) — Ln(2) acts primitively on the set
of subgroups in By. Hence different subgroups from By have different
centralizers in 02(Jn). This implies that the subgroups in By are linearly
independent and that the action of Jn (and hence also of J/K) on K is
irreducible.
There remains to prove that J does not split over K. Let t be an
element of type 1 incident to y and stabilized by J\ and let Q =
Then by the previous paragraph J\/Q is some extension of
by an elementary abelian 3-group of rank a(n — 1). The latter group
is generated by the subgroups Kx for x e res/({y,t})n~2. Therefore
dim CK(Q) > o(n — 1). On the other hand N := J\ nJn acting on the set
of subgroups from By has two orbits consisting of the subgroups Kx € By
with x incident and non-incident to y, respectively. Since Q is normal in
N and acts faithfully on K, it cannot centralize subspaces in both orbits,
so dim CK(Q) = (r(n — l). Hence J\ = Nj(n)(Q) and if J were split over K,
J\ would split over O?,(J{). Now the non-splitness follows by induction
since for n = 2 we have a non-split extension 3 • Sym^ = 3 • Sp^{2). •

6.8 A special class of subgroups in J(n)

We start by constructing a family E = E(n) of subgroups of order 3 in
K = K(n) such that J/K acts doubly transitively on $ by conjugation
with stabilizer isomorphic to 0^,(2). After that we show that a simi-
lar family of subgroups must exist in the automorphism group of the
universal 2-cover of f.
296 From Cn- to Tn-geometries
Lemma 6.8.1 There is a non-singular quadratic form on K preserved by
J/K.

Proof. For y e fn let us define a quadratic form Xy to be the sum of

squares in the basis By. This form is obviously invariant under the action
of the stabilizer of y in J. Since $ is connected, to prove the lemma it
is sufficient to show that Xy' = Xy whenever y and yf are incident to a
common element of type n — 1. But the equality easily follows from the
rewriting rules given in (6.7.1). •

Corollary 6.8.2 The subgroup K, considered as a GF(3)-module for J/K,

is self-dual •

Lemma 6.8.3 There is a family $ of 1-dimensional subspaces in K, such

that J/K acting on K by conjugation preserves $ as a whole and induces
on it a doubly transitive action of Sp2n(2) on the cosets of 0^,(2).

Proof. By (6.8.2) it is sufficient to indicate the required family of

subspaces in the module dual to K. Recall that J was constructed as a
subgroup of the semidirect product W : G where G = Sp2n(2) and W is
a GF(3)-module for G induced from a non-trivial 1-dimensional module
for a subgroup 0 in G isomorphic to 0^(2). So K is a submodule in W.
Since W is an induced module, it possesses a direct sum decomposition
into 1-dimensional subspaces Wf indexed by the cosets of O in G and G
induces on the set of these subspaces a doubly transitive action. Then the
desired set of hyperplanes in K is formed by the kernels of the projections
of K onto the subspaces Wf. •

Let us calculate the support with respect to the special basis By of

a subspace E e S. Let O and Jn be the stabilizers in J of £ and y,
respectively. Then the orbits of O n Jn on By are the same as the orbits of
q>(OnJn) on the set of elements of type n—2 incident to the image of y in
^ = ^(5p2«(2)). Hence the calculations of the orbits can be carried out in
the latter geometry. In terms of Section 6.4 we can assume that the image
of y in ^ is the maximal totally singular subspace Vn and cp(O) is the
stabilizer 0(f) = 0^(2) of a quadratic form / of minus type associated
with the symplectic form *F. Then By is in the natural bijection with
the set of codimension 2 subspaces in Vn. Let U = U(f) be the unique
subspace of codimension 1 in Vn which is totally singular with respect
to / . Then by Witt's theorem the orbit under 0(f) n Gn of a subspace X
of codimension 2 in Vn depends on whether or not U contains X. Let
 6.9 The f(n) are 2-simply connected 297
1
£ = £,{y,E) be the unique element from resy(y)"~ which maps onto U
under the 1-covering of # onto (S. Thus the following two sets are the
orbits of OnL on By:
ei(y, I) = {Kx | x e re S/ (£)"- 2 }, ®2(y, {) =

Lemma 6.8.4 /n t/ie afrot;e terms ®i()>, £) is t/ie support of E € $ with

respect to the basis By.

Proof. Since 0 nJn acts transitively on ®t(y, <!;) for i = 1 and 2, the
support is either one of these orbits or the whole of By. Suppose that
®2(y,£) is contained in the support. Then E is not orthogonal (in the
sense of (6.8.1)) to any subspace Kx € ®2{y>£)> Let {y9y',y"} be the set
of elements of type n incident to £. It is clear that ®i(/,£) = ®i(y,£),
hence the support of E with respect to By contains ®i{y\ £) and E is
not orthogonal to any Kx e ®2(/, £) (and similarly for / ' ) • Let u be an
element of type n — 3 incident to <J and let w\, W2, W3, W4 be the elements
of type n — 2 incident to u and y but not to £. Then by the assumption
made, £ is generated by a product e^eietf where ^j is a non-trivial
element from KWi for 1 < i < 4, and / is an element orthogonal to
(Kw. I 1 < i < 4). By (6.7.1 (iv)) the subgroup / generated by the product
e\e2ej1e^1 is contained in ®2(/,£) U ©2(/',£). Since / is orthogonal to
E this is a contradiction. •

6.9 The </(n) are 2-simply connected

The 2-simple connectedness of / ( 3 ) follows from (6.6.2) so we assume in
this section that n > 4.
Let £f = {Stj I 1 < i < j < n) be the amalgam of rank 2 parabolic
subgroups associated with the action of J on /. Let J be the universal
completion of Sf. Then there is a homomorphism e : J ^> J such that the
composition 5 of 8 and the restriction of q> to J is a homomorphism onto
G = Sp2n(2) which maps 5^ onto the amalgam # of rank 2 parabolics
associated with the action of G on ^.

Lemma 6.9.1 J is a flag-transitive automorphism group of a rank n T-

geometry f possessing a 2-cover co onto f. Moreover, s is an isomorphism
if and only if co is an isomorphism.
Proof. For k = n and n— 1 let J& be the subgroups in J generated by the
subamalgams Sf^ = {Stj | 1 < i < j < n,k £ {Uj}}. The subamalgams
298 From Cn- to Tn-geometries
n
and <?( -V map under <5 isomorphically onto the subamalgams
and (g(n~1\ respectively. Hence by (6.3.4) each mapping in the following
sequence is an isomorphism:

Let r be a graph whose vertices are the cosets of Jn in J with two vertices
being adjacent if they intersect a common coset of Jn-\. Then the above
isomorphisms induce the following sequence of graph coverings:
f -> f -> r.
Define , / to be the geometry whose elements are the vertices and the
triangles in F and the connected components of the full preimages of
the subgraphs representing the elements of # in F (equivalently of the
elements of ^ in F), where the incidence relation is via inclusion and
the type function is the obvious one. Then $ possesses 1-coverings onto
# and ^. Now arguments as in the proof of (6.5.3) show that $ is a
T-geometry possessing a 2-cover co onto #. •
Let {Jt | 1 < i < n} be the amalgam of maximal parabolic subgroups
corresponding to the action of J on /. We assume that e and 3 map Jf
onto Ji and G,, respectively. The elements of / will be denoted by letters
with hats; the same letter without a hat will denote the image in #. A
similar convention will be applied to the elements of J.
Let K be the kernel of S. Then clearly e maps K onto K. It is easy to
see from (6.9.1) that Jw_2 OK is of order 3 and hence with every element
x of type n — 2 in / we can associate a unique subgroup K^ of order 3
in K which stabilizes 5c. Moreover, Kq maps onto Kx under e. Let u be
an element of type n — 3 in / . Then by (6.6.2) both rest(w) and resj(w)
are isomorphic to the rank 3 T-geometry of the group 37 • Spe(2) and we
have the following.

Lemma 6.9.2 In the above notation the group

is elementary abelian of order 37 and it maps isomorphically onto its image

in K. In particular K^ and Kq commute whenever x and z* are incident to
a common element of type n — 3. •
For an element y of type n in # define
 6.9 The f(n) are 2-simply connected 299
In view of (6.9.2), arguing as in (6.7.1) and (6.7.2) we observe that the
subgroups in R^ generate the whole of K. Moreover, if yf is an element
of type n incident with y to a common element of type n — 1 then there is
a canonical way to express elements from the subgroups in £~> in terms
of elements from the subgroups in B^ as in (6.7.1).

Lemma 6.9.3 The homomorphism s : J —> J is an isomorphism if and only

if any two subgroups from B^ commute.

Proof. The "only if" part is obvious so suppose that the subgroups
in B^ pairwise commute. Since K is generated by these subgroups, we
conclude that K is elementary abelian of rank at most the number of
subgroups in £K Since the latter number is the rank of K, this means that
£ restricted to K is an isomorphism onto K. Since J/K = J/K = Sp2n(2)
the result follows. •
Whenever two elements of type n — 2 in f are incident to a common
element of type n they are always incident to a common element of type
n — 4. Thus by the above lemma the isomorphism of J and J for n = 4
would imply the isomorphism for the higher ranks. But we are going to
prove the isomorphism uniformly on n by constructing in K a family
$ of subgroups of order 3 similar to the family $ in K constructed in
Section 6.8.
For an element £ of type n — 1 in / put

Since res^(£) is a projective geometry, any two elements of type n — 2

incident to { are incident to a common element of type n — 3. Hence
is an elementary abelian 3-group of rank 2n~l — 1 which is the number
of elements of type n — 2 incident to £ and it maps isomorphically onto
its image Y(£) in K, where
Y(0 = (Kx\xe re S/ (£r 2 >.
By (6.8.4) every subgroup E e $ is contained in Y(£) for a unique
element ^ = £(y,E) of type n — 1 incident to y. In the above terms let
E(y) be the unique subgroup in Y(£) which maps onto E under £ (here
£ is the unique element of type n — 1 incident to y which maps onto £)
and let S(y) be the set of all subgroups obtained in this way. Clearly
S{y) maps bijectively onto S.
300 From Cn- to Tn-geometries
Lemma 6.9.4 The family S{y) is independent of the particular choice ofj.
Proof. Because of the connectedness it is sufficient to show that
whenever y' is an element of type n incident with y to a common element
? of type n — 1, we have S(yf) — $(j). In order to prove the latter equality
it is sufficient to show that every subgroup E e S(yr) is contained in
S{y). Thus let E e S(yr) and £ be the element of type n — 1 incident to
y' such that E is the unique preimage in Y(£) of a subgroup E e S. If
7 = £, (equivalently if y is incident to <!;), then E € S{y) by the definition.
Thus we assume that ? ^ {. Let ? be a non-trivial element from E. Then
? = d\d2...ds, where s = 2n~1 — 1 and the d\ are non-trivial elements taken
from different subgroups in the set

Since the subgroups in B(£) commute pairwise, the factors d, can be

rearranged in an arbitrary way. Let di be the unique element of type
n — 2 incident to both £ and ?. Let ivi,..., wt be the elements of type n — 3
incident to %, where t = 2n~2 — 1. Then for 1 < i < t the element w, is
incident to three elements of type n — 2 incident to ^, one of them is ?i
and two others we denote by t^ and tJii+i- In this case

Since the subgroups in B(£) commute pairwise, without loss of generality

we can assume that dj e K~ for 1 < j < s. Then by the analogue of
(6.7.1 (iii)) the product ^21^21+1 is equal to the product /21/21+1 of two
non-trivial elements taken from different subgroups in the set IK Since
Kg belongs to 2K as well, we conclude that £ = fifi—fs, where the /* are
non-trivial elements taken from different subgroups in iK Then fifi-fs
is the decomposition of e in the basis By. By (6.8.4) there is an element v
of type n — 1 incident to y such that the /, are non-trivial elements taken
one from each subgroup from {Kx \ x G res/(v)n~2}. Hence e e Y(v) and
? is the unique preimage of e in Y (?) where ? is the element incident to
y which maps onto v. Thus E e S(y) and the result follows. •
By the above lemma we can denote the family $(y) simply by S.

Lemma 6.9.5 $ is a conjugacy class of subgroups in J; J acting on S

induces the doubly transitive action of Sp2n(2) on the cosets of 0^(2) and
K is in the kernel of the action.
 6.10 A characterization of f(n) 301
Proof. Let £, be the element of type n - 1 in / stabilized by Jn-\.
Since the latter group maps isomorphically onto Jn-\ and Y(£) maps
isomorphically onto Y(£), it is easy to see that Jn-\ stabilizes as a whole
the set of subgroups from $ contained in Y(£). Let y be the element of
type n stabilized by Jn. It is easy to see from the above that Jn stabilizes
S = S{y) as a whole. Now the result follows from (6.9.4). •
Since the subgroups in $ have order 3, none of them can be inverted by
an element from K. Hence by (6.9.5) the subgroups from $ are contained
in the centre of K. Thus in order to prove that K is abelian, it is sufficient
to show that K is generated by the subgroups in S. The subgroup Y(£)
for an element £ of type n — 1 contains subgroups from $ as well as
subgroups Kq which are known to generate K. Thus the statement we
need is a direct consequence of the following lemma whose proof is very
similar to that of (6.7.3).

Lemma 6.9.6 The stabilizer of £ in J acts irreducibly on Y(£). •

Thus K is abelian, J is isomorphic to J and we have the main result
of the section.

Proposition 6.9.7 The geometries f(ri) are 2-simply connected for all n>3.
•

6.10 A characterization of f(ri)

We start with an elementary but important result about the point graph
of the rank 2 T-geometry. Let 0 and 0 be tiie point graphs of ^(Sp4(2))
and ^(3 • Sp4(2)), respectively, and let \i : 0 —• 0 be the corresponding
covering of graphs. The following result can be deduced directly from
the intersection diagram of 0.

Lemma 6.10.1 The subgroup of the fundamental group of 0 associated

with fi is generated by the cycles of length 3 and by the non-degenerate
cycles of length 5; it does not contain cycles of length 4. •
Let T be the dual polar graph of ^ and T be the graph on the set of
elements of type n i n / and \p : T —• F be the corresponding covering
of graphs. Then F is a near n-gon with quads. A quad ©is the subgraph
induced by the vertices incident to a given element of type n — 2. Every
cycle in F whose length is 4 or 5 is contained in a unique quad. Every
connected component of tp"^©) is the point graph 0 of the rank 2
T-geometry.
302 From Cn- to Tn-geometries
Let F be the fundamental group of F. Let F(4) be the subgroups of
F generated by the cycles of length 4 and F(3,5) be the subgroups of
F generated by the cycles of length 3 and by the non-degenerate cycles
of length 5. It follows directly from the 2-simple connectedness of ^
that F(4) together with F(3,5) generates the whole of F. By the previous
paragraph we have the following.

Lemma 6.10.2 The subgroups of the fundamental group of T associated

with \p contains F(3,5) and it does not contain F(4). •

Let x • r —> F be the covering of T associated with F(3,5). Let # be

the geometry whose elements are vertices and triangles of Y together with
the connected components of full preimages of subgraphs in T which
represent the elements of ^ with respect to the natural incidence relation
and type function. It is clear that F(3,5) is normal in F and it is also
normalized by the action of G on F. Hence every element g G G can be
lifted to an automorphism of P. It is clear that all these liftings form
a flag-transitive automorphism group J of / which commutes with / ,
and the action induced by J on ^ coincides with G. By (6.10.2) F(4)
is not contained in F(3,5) and in view of flag-transitivity and (6.10.1),
every connected component 0 of x~{(®) * s the point graph of the rank
2 T-geometry. Since x *s a covering of graphs, this shows that / is a
flag-transitive T-geometry. Also by (6.10.2) we have a covering X : T —• T
which induces a 2-covering of # onto $. By (6.9.7) X is an isomorphism
and we have

Lemma 6.10.3 The subgroup of the fundamental group ofT associated with
\p is exactly F(3,5). •

Let Jf be a rank n geometry with the following diagram:

where the rightmost edge indicates that for an element z of type n — 2 the
residue resj,(z) is either ^(Sp4(2)) or ^(3 • Sp4(2)), possibly depending on
the choice of z. Suppose that Jf possesses a 1-covering v onto ^ . Let A
be the graph on the set of elements of type nin Jif in which two elements
are adjacent if they are incident to a common element of type n — 1.
Then v induces a covering co : A —• T and every connected component
of co-l(®) is the point graph of either &(Sp 4(2)) or 9(3 • Sp4(2)). This
shows that the subgroup of F associated with co contains F(3,5) and by
 6.11 No tilde analogues of the Alt-]-geometry 303
(6.10.3) there is a covering S : F —• A which induces a covering of J f
onto #. This gives

Proposition 6.10.4 Let Jf be a geometry with the above diagram, possessing

a 1-cover onto &. Then Jtf is a quotient of J'. In particular every T-
geometry of rank n (maybe not flag-transitive) possessing a 1-cover onto
<g(Sp2n(2)) is a quotient of f(n). •

An example of a quotient T-geometry of / can be constructed as

follows. Let L be a non-trivial subgroups of K which intersects Kx
trivially for every element x of type n — 2 in / . Let / be the geometry
whose elements are the orbits of L on / with the type function and
incidence relation induced by those in #. Then it is easy to see that ^ is
a T-geometry.

6.11 No tilde analogues of the 4/r7-geometry

In this section we show that there are no geometries ^ satisfying the
following.

Hypothesis 6.11.1 <§ is a flag-transitive T-geometry of rank 3, G is a flag-

transitive automorphism group of & and there is a 1-covering cp : ^ —•
^(Alt-j) which commutes with the action of G (i.e. the fibres of q> are
unions of G-orbits).

If ^ and G satisfy the above hypothesis then G induces on ^(Alti) its

unique flag-transitive automorphism group which is Alt-]. Since 3 • Alte is
the only flag-transitive automorphism group of the rank 2 T-geometry
which possesses a homomorphism onto Alte, it is easy to see that ^ and
G satisfy the following.

Hypothesis 6.11.2 ^ is a rank 3 T-geometry; G is a flag-transitive auto-

morphism group of& such that the amalgam & = {Gi,G2, G3} of maximal
parabolic subgroups satisfies the following:

Gx S 3 • Alt6; G2 ^ (Sym3 x Sym4)e; G3 ^ L3(2).

(Here G2 is isomorphic to the setwise stabilizer in Alt-] of a 3-element

subset.)

Thus the non-existence of geometries satisfying Hypothesis 6.11.1 will

follow from the non-existence of geometries satisfying Hypothesis 6.11.2.
304 From Cn- to Tn-geometries
Lemma 6.11.3 Let & and G satisfy Hypothesis 6.11.1. Then the isomor-
phism type of & is uniquely determined.
Proof. Since ^ is a T-geometry, it is easy to see that G i; := Gz n Gj =
Sym4 for 1 < i < j < 4 and that B := GinG2DG3 = D 8 . First, we observe
that all subgroups Sym4 in G\ are conjugate in the automorphism group of
Gj. Notice that G2 contains three classes of Sym^subgroups and the outer
automorphism group of G2 induces Sym^ on the set of these classes. Since
all automorphisms of Sym4 are inner this shows that the subamalgam
{G2, G3} is uniquely determined. Hence to complete the proof we have
to show that there is a unique way to adjoin G\ to this subamalgam
intersecting it in the subamalgam {G12, G13}. The outer automorphism
group of the latter amalgam has order 2 and it is represented by the
automorphism which centralizes G12 and acts on G13 \ G12 by means of
conjugation by the non-trivial element from the centre of B. But this
automorphism can be realized inside the normalizer of {G12, G13} in the
automorphism group of G\. •
It is easy to see that the unique amalgam 31 from (6.11.3) possesses
a homomorphism (in the obvious sense) on the amalgam of maximal
parabolic subgroups corresponding to the action of Alt-] on ^(Altj). This
shows that a simply connected geometry satisfying Hypothesis 6.11.2
possesses a 1-covering onto ^(Alt-j) commuting with the action of its
automorphism group (i.e. satisfies Hypothesis 6.11.1).
We are going to describe a presentation for the universal completion
of the amalgam & as in (6.11.3), which is due to S.V. Tsaranov (private
communication). All the facts claimed below can be easily verified by
coset enumeration with a computer.
(a) The Coxeter group of the diagram

a\ a2 a^

is isomorphic to Sym4 x 2. To eliminate the centre, one can add the

relation (aia2ai)3 = 1.
(b) The Coxeter group of the diagram

b\ b2 ft3

is isomorphic to Alts x 2. To eliminate the centre we put {b\b2b?)5 = 1.

 6.11 No tilde analogues of the Alt-]-geometry 305
(c) The presentation

c\ c2

(c 3 cic 2 ) 3 = (c 4 c 2 ci) 3 = 1

c3 c4

defines the group L3(2).

(d) The presentation

(d3did2)3 = 1,
5
(d{d3d4)

defines the non-split extension 3

(e) The presentation

e3

(e3e2ei)3 = (e4e2ei)3 = 1,

e4

defines the group {Sym4 x Sym^f. Notice that the relation (e2e3e4)3 = 1
reduces the corresponding Coxeter group Cox(A2) = Z 2 : Sym3 to the
group Sym4 = 2 2 :
306 From Cn- to Tn-geometries
Now let us consider the group F defined by the following presentation:
n
(/4/1/2)3 = (/3/2/0 3 = 1,

(/5/2/0 3 = (/1/4/5)5 = 1,

(/2/5/4)5 = {hhhf = 1.
n n
It is easy to see from the above that F is the universal completion of the
unique amalgam & from (6.11.3). On the other hand coset enumeration
on a computer implementing the Todd-Coxeter algorithm shows that
F = Alt-]. Since the latter group does not contain 3 • Alte this shows that
& has no faithful completions and we obtain the following.

Proposition 6.11.4 There are no geometries & satisfying Hypothesis 6.11.1

or 6.11.2. •
 7
2-Covers of P -geometries

In this chapter we construct the univesal 2-covers ^(3-Mat22), ^(3 23 'C02)

and ^(3 4371 • BM) of the P-geometries <${Mat22\ &(Co2) and f(J5M),
respectively. The construction goes as follows. We start with a group G
acting flag-transitively on a P -geometry ^ and consider an extension G
of G by an elementary abelian 3-group W. We show that G contains a
subamalgam si which maps isomorphically onto the amalgam of rank 2
parabolics associated with the action of G on c§. Then the subgroup G in
G generated by si is a flag-transitive automorphism group of a geometry
^ which possesses a 2-covering onto ^. So in a sense our construction
strategy is similar to that in Chapter 6, the difference being that here G
does not split over W. We identify the structure (particularly the order) of
G by establishing an upper bound on the size of a flag-transitive 2-cover
of <&. This bound also enables us to prove the 2-simple connectedness of
^. We start by reviewing some properties of a generic P -geometry and
of the geometries <Sf(Mat22)9 &(Co2) and

7.1 On P -geometries
Let 9 be a P -geometry of rank n > 3, i.e. a geometry with the diagram

n- o ° o ^ o

Let O = {xi,x2,...,xn} be a maximal flag in ^, where x, is of type i. Let

G be a flag-transitive automorphism group of <8. For 1 < i < n let G*
and Pi be the maximal and minimal parabolics of type i associated with
the action of G on ^, so that G, = G(xi) and Pt is the intersection of the
Gj for 1 < j < n, j j= i. Let

307
308 2-Covers of P-geometries
be the Borel subgroup and let Qi denote the kernel of the action of G\
on res#(x,).
Let A = A(^) be the derived graph of ^, which is a graph on the set
of elements of type n in &, in which two such elements are adjacent
if they are incident to a common element of type n — 1. In particular
xn is a vertex of A. Let n = n(xn) denote the projective space dual to
res#(xn). This means that the points of n are the elements of type n — 1
in the residue of xn. Then Gn = Gn/Qn is a flag-transitive automorphism
group of n and by (1.6.5) either Gn = Ln(2) or n = 3 and G3 = Frob].
In any case the action of Gn on the point set of n is primitive and this
immediately gives the following.

Lemma 7.1.1 If {x9 y) is an edge of A then there is a unique element of

type n—1 incident to both x and y. •

The above lemma enables us to identify the elements of type n—1

with the edges of A so that the vertex-edge incidence in A corresponds
to the incidence in ^. In this way the set A(xn) of vertices adjacent to
xn is naturally identified with the set of points in n. In fact it is easy
to show [Sh85] that the possibility 63 = Frob* cannot be realized in a
flag-transitive P -geometry, which means that (with respect to the action
of G) A is a locally projective graph of type (w,2) (see Section 9.1 for the
definitions).
For an element y e <S let A|>] denote the subgraph in A induced by
the vertices incident to y. By the above, if y is of type n o r n - 1 then
A\y] is a vertex or an edge, respectively. Using the diagram of ^ and the
fact that in a projective geometry every element is uniquely determined
by the set of points it is incident to, it is easy to check the following.

Lemma 7.1.2 The following assertions hold:

(i) ifl<i<n — 2 then A[x,-] is isomorphic to the derived graph of the

P-geometry resj(xj), in particular A[xw_2] is a Petersen subgraph;
(ii) A[z] a A[y] if and only if z and y are incident elements of type i
and j , respectively, and 1 < j < i < n;
(iii) A[xj] n A(xn) is the point set of a subspace in n of (projective)
dimension n — i—l, in particular A[XJ] is of valency 2n~i — 1;
(iv) i / O is the point set of a subspace in n of dimension n — i— l,for
1 < i < n — 1, then there is a unique element y of type i incident to
xn, such that 0 = A(xn) n A[y];
 7.1 On P-geometries 309
(v) ifue A2(xn) then there is a unique element y of type n — 2 which
is incident to both xn and u (equivalently such that the Petersen
subgraph A[y] contains both xn and u). •
In terms of Sections 9.6 and 9.8 the above lemma says that the
subgraphs A[y] taken for all y e & form a complete family of geometrical
subgraphs in the locally projective graph A. One can also observe from
the lemma that the elements of ^ can be identified with the subgraphs
A\y] so that the incidence relation is via inclusion and the type of a
subgraph is determined by its valency. Since ^ is a geometry, the graph
A and all the subgraphs A\y] are connected and, since G is a flag-transitive
automorphism group of ^, for every 1 < i < n — 2 the action of Gj on
A[x,] is 1-arc-transitive, which implies the following.

Lemma 7.1.3 G = (Gn,Gn_i) and for every 1 < i < n — 2 we have Gt =

(GiHGn, GiDGn-i). •
Sometimes it is convenient to study 2-coverings of P -geometries in
terms of their derived graphs.

Lemma 7.1.4 Let A be the derived graph of a P-geometry CS. Let cp : A —• A

be a covering of graphs such that for every element y of type n — 2 in
& every connected component of cp~l{A\y]) is isomorphic to the Petersen
graph. Then A is the derived graph of a P-geometry & which possesses a
2-covering onto (S.
Proof. We define ^ to be the incidence system whose elements are
the connected components of the preimages in A of the subgraphs A[y]
taken for all y G ^ , the incidence relation is via inclusion and the type
function is induced by that in <&. We claim that ^ is a P -geometry. Since
the incidence relation is via inclusion ^ belongs to a string diagram;
since (p is a covering of graphs, it is easy to see that for xn € (p~l{xn) we
have res-(xn) = res#(xw). Finally, by the hypothesis of the lemma, if y is
an element of type n — 2 in ^ then res±(y) is the Petersen graph geometry
and the claim follows. It is clear that q> induces a 2-covering of ^ onto
^. •
Let K be the kernel of the action of Gn_2 on resj(xn_2). Then K is
the largest subgroup in Gn n Gn-i which is normal in both Gn_2 n Gn
and Gn_2 n Gn-\ and Gn-2/K is a flag-transitive automorphism group of
the Petersen graph geometry isomorphic to Syms or Alts. The following
lemma describes some 2-covers of ^ in group-theoretical terms.
310 2-Covers of P-geometries
Lemma 7.1.5 Let G be a group possessing a homomorphism xp onto G.
Suppose that
(a) G contains a rank 2 subamalgam 8ft = {Gn, Gw_i} which generates
G and such that the restriction xpo ofxp to & is an isomorphism onto
the subamalgam {G n ,G n _i} in G.
Put
K=xp»1(K) and Gn-2 = (N~n(KlN~nJK)).

Suppose also that

(b) Gn.2/k * Gn-i/K.
Then G is a flag-transitive automorphism group of a P-geometry which
possesses a 2-cover onto <&.

Proof. Let A be a graph on the set of (right) cosets of Gn in G in

which two such cosets are adjacent if they intersect a common coset of
Gn-\. Then G acts 1-arc-transitively on A and by (a) xp induces a covering
cp : A —> A of graphs. Let xn denote Gn considered as a vertex of A and
let II be the subgraph in A induced by the images of xn under Gn_2.
Then by (b) the restriction of cp to n is an isomorphism onto A[xw_2]
and the result follows directly from (7.1.4). •
Notice that the vertices of the Petersen graph can be identified with
the transpositions in Syms so that two vertices are adjacent if and only
if the corresponding transpositions commute.

Lemma 7.1.6 Suppose that Qn is finite and non-trivial Then the following
assertions hold:
(i) there is q e Qn such that Qn = (q, Qn n Qn-2) and acting on the
Petersen graph resj(x n _2) the element q induces the transposition
from G+_2 which corresponds to xn;
(ii) Gn_2 induces Syms on resj(x n _2);
(iii) Qn = O2(Gn);
(iv) Qn-i has index 2 in O2(Gw_i) and Qn < Qn_i.

Proof. It is easy to see that Qn is the kernel of the action of Gn

on A(xn). By the definition Qn stabilizes the Petersen subgraph A[y]
whenever y is an element of type n — 2 incident to xn. Suppose that Qn
fixes every vertex in such a Petersen subgraph. Then by (7.1.2 (v)) Qn acts
trivially on A2(xn) which implies the triviality of Qn by (9.1.4). Hence Qn
induces on A[xn_2] an action of order 2 generated by the transposition
 7.1 On P-geometries 311

corresponding to xn (recall that A[xn_2] is isomorphic to the Petersen

graph). Hence (i) follows and immmediately implies (ii). We have seen
that Qn induces a 2-group on A2(xw). By (9.1.5) Qn is a 2-group itself.
Since O2(Gn/Qn) = 1 (iii) follows. Since all automorphisms of L n _i(2) are
inner, we see from the diagram of ^ that Gn-i/Qn-i — Ln-\(2) x 2 and
in view of (iii) we obtain (iv). •
Each of the five P -geometries we have constructed so far, namely

ViMatn), 9(3 • Matn), 9(Mat^)9 9(Co2) and 9(BM)9

contains a subgeometry which is a Cm- or Tm-geometry, namely

9(Sp4(2))9 9(3-Sp4(2))9 9(Altn)9 $(Sp6(2)) and

((3.4.2), (3.4.4), (3.5.8), (4.9.8) and (5.4.5) and the table in Section 1.10). We
are going to present a systematic way to construct such subgeometries.
Recall that the edge graph of the Petersen graph is an antipodal
distance-transitive graph with the intersection diagram

Thus there is an equivalence relation on the set of edges of the Petersen

graph with classes of size 3.
With 9 and A as above let S be a graph on the set of edges of A in
which two edges are adjacent if they are incident to a common element
y of type n — 2 and if they are equivalent in the Petersen subgraph
A[y]. For any two distinct elements u and v of type n — 2 the Petersen
subgraphs A[w] and A[v] have at most one common edge and hence
every edge of Z is contained in a unique triangle which corresponds to
a Petersen subgraph in A. Let E c denote the connected component of S
which contains xn-\. Let Sf = «5^(xn_i) be a geometry whose elements of
type i, 1 < i < n — 1, are the non-empty intersections S c n A[y], where y
is an element of type i in ^ ; the incidence relation is via inclusion.

Lemma 7.1.7 In the above terms suppose that n > 4. Then

(i) i/resj(x w _ 3 ) = ^(Mat22) then <? is a Cn-i~geometry with the dia-
gram
312 2-Covers of P-geometries
(ii) i/resj(x w _ 3 ) = ^(3 • Mat22) then £f is a Tn-i-geometry with the
diagram

2 2 2 2 2

In both cases the stabilizer of Sf in G induces on 9 a flag-transitive

action.

Proof. Since the incidence relation is via inclusion Of belongs to a

string diagram. It is easy to observe that there is a natural isomorphism
between the residue of xn_i in Sf and res^(xw_i), so that both the residues
are projective geometries of rank n — 2 over GF(2). Finally it is an easy
combinatorial exercise to check that in cases (i) and (ii) the geometry
formed by L c n A[xw_3] coincides with the subgeometry as in (3.4.4) and
(3.5.8), respectively. •
Recall that by the main result of [Sh85] every flag-transitive P-
geometry of rank 3 is isomorphic either to ^(Mat22) or to ^(3 • Mat22) so
that the cases (i) and (ii) in the above lemma cover all the possibilities.
To the end of the section we discuss the properties of some concrete P-
geometries. The description of the flag-transitive automorphism groups
of P -geometries comes most naturally from the complete classification
(Preface) which gives all the pairs (&, G) where ^ is a P -geometry and
G is a flag-transitive automorphism group of ^ . In particular cases
the flag-transitive automorphism groups can be determined by ad hoc
analysis of subamalgams in the amalgam of maximal parabolics in the
full automorphism group. We suggest the reader check the details and
formulate the final result.

Lemma 7.1.8 Let G be one of Mat22, Mat23, Co2 and BM, and &(G) be
the P-geometry on which G acts flag-transitively. Let H be a flag-transitive
automorphism group of ^(G). Then either G = H, or G = Mat22 and
H = AutMat22. O

Lemma 7.1.9 Let & be isomorphic to &(Mat22), &(Co2) or <g(BM) and G

be the full automorphism group of <$. Then

Gt = NG(Qi) for i = n-l and n.

Proof. For ^ = <0(Mat22) the result is by (3.9.3 (iv)). For ^ = &(Co2)

using (4.9.5) and the table therein it is not difficult to check that the sets
of vertices in the collinearity graph 0 fixed by Q3 and Q4 are exactly the
 7.2 A sufficient condition 313
*-closed cliques of size 7 and 15 which represent X3 and X4, respectively.
In the case of the Baby Monster BM we have G\ = CG(Z(Q\)) which
enables us to identify xt with Z(Q,) for 1 < i < 5 and immediately gives
the result. •

7.2 A sufficient condition

In this section we formulate and prove a sufficient condition for existence
of proper 2-covers of a flag-transitive P -geometry in terms of non-
split extensions of its automorphism group. We start with a preliminary
technical result.
Let F = Syms act naturally on the Petersen graph II and $ = {E\,E2}
be a subamalgam in F consisting of the stabilizers in F of a vertex and
an edge from II which are incident. This means that E\ = 2 x Sym^
E2 = D8 and Ex n E2 = 22.

Lemma 7.2.1 Let F be a group possessing a homomorphism onto F = Syms

whose kernel is an elementary abelian 3-group U. Suppose that U is the
centre of the full preimage of Ff = Alts in F. Let $ be a subamalgam in
F which maps isomorphically onto S. Then $ generates in F a subgroup
isomorphic to
Proof. Since the 3-part of the Schur multiplier of Alts is trivial, it
is easy to show that F contains a normal subgroup H isomorphic to
Alts. Consider the image of $ in F/H. It is easy to see that this image
generates a subgroup of order 2 and the result follows. •
Now we are ready to prove the main result of the section.

Proposition 7.2.2 Let & be a P-geometry of rank n > 3 and G be a flag-

transitive automorphism group of & such that
(a) Gi = NG(Qi) for i = n-l and n.
Let G be a group which possesses a homomorphism \p onto G whose kernel
W is an elementary abelian 3-group. Suppose further that
(i) G does not split over W,
(ii) the preimage of Qn in G does not centralize non-identity elements
in W,
Then there is a subgroup G of G such that the restriction of xp to G is a
non-injective homomorphism onto G. Moreover, G acts faithfully and flag-
transitively on a P-geometry <& which possesses a 2-covering onto <S.
314 2-Covers of P-geometries
Proof. First notice that the condition (a) implies that Qn is non-trivial
and that
(b) Gn O Gn_! = NGn{Qn-l) = No^iQn).
The minimal parabolic subgroup Pn is a 2-group which contains B
with index 2. Let Pn be a Sylow 2-subgroup in the full preimage of
Pn in G. Then the restriction xp\ to xp to Pn is an isomorphism onto
Pn. Put & = xpiHQi) for i = n, n - 1, n - 2 and G, = AT~(Q,) for
i = n,n-L By (ii) and (7.1.6 (iv)) Cw(Qn) =jCV(8n-i) = 1 and hence
by the Frattini argument (compare (6.1.1)) G; = G, for i = n,n — 1.
Also by (b) Gn Pi Gn_i = Gn n Gn_i and hence the restriction tpo of t/?
to the subamalgam 0& = {Gn, Gn_i} in G is an isomorphism onto the
subamalgam & = {Gn, Gn_i} in G. Let G be the subgroup in G generated
by J^. We are going to show that G acts faithfully and flag-transitively
on a P-geometry which possesses a 2-covering onto (3. By (7.1.3) the
restriction of xp to G is a homomorphism onto G. This restriction must
have a non-trivial kernel, since otherwise G, would be a complement to
VF in G, which is impossible, since by (i) G does not split over W. We
are going to show that condition (b) from (7.1.5) holds.
Let Gw_2 = (Fn,Fn_i>, where Fj = % H ^ ) with F/ = G^nGj for j = n
and n — 1 and xp2 be the restriction of xp to Gn_2. Then U := Gn_2 n W is
the kernel of xp2. Notice that Qn-i := t/;^"1(gw_2) is contained and normal
in both Fn and F n _i. Hence Qn-i is normal in Gn_2 and U < CwiQn-i)-
Let K be the kernel of the action of Gn_2 on resj(x n _2). Then K
is the largest subgroup in Fn n Fn_i normal in both F n and Fn_i and
^ = {Fn/K,Fn^i/K} is the subamalgam in G w _ 2 /X = S^ms (7.1.6 (ii))
consisting of the stabilizers of incident vertex and edge in the Petersen
graph resj(x n _ 2 ) (i.e. ^ ^ { 2 x Sym^Ds}). Let K ^xp^iKy, Then K is
the largest subgroup in Fn n Fn_i normal in both F n and Fn-\ and 1^2
induces an isomorphism of Q) — {Fn/K,Fn-i/K} onto Of. We claim that
the induced homomorphism

is an isomorphism. The kernel of \p-$ equals the kernel U of \pi and in

order to apply (7.2.1) it is sufficient to show that U is centralized by the
commutator subgroup of Gn-2/K (isomorphic to Alts).
Let q G Qn be as in (7.1.6 (i)) and q = xp^l(q). Since Cw(Qn) = 1 and
Qn = (?>6« n QB-2), the element q inverts every element of Cw{Qn-i)
and hence of U as well. On the other hand q induces on resj(x n _2) the
transposition from Gn-2/K = Syms which corresponds to xn. Consider-
 7.3 Non-split extensions 315
ing other elements of type n incident to xn-2> we conclude that every
transposition from Gn-2/K inverts U and hence every even permutation
centralizes U. Thus (7.2.1) gives Gn-2/K = Sym5, completing the proof.
•
The reader might notice that we did not quite follow the strategy
outlined in Section 6.1 and instead of embeddings of the amalgam si =
{Pi | 1 < i < n} we consider embeddings of the amalgam & = {Gn, Gn_i}.
But in fact these embeddings are related since 38 contains si.

7.3 Non-split extensions

In this section we construct some extensions which satisfy the conditions
in (7.2.2).
Let ^ be one of the geometries ^(Mat2i), &(Co2) and &(BM), G be the
full automorphism group of ^ (isomorphic, respectively, to Aut Matu,
C02 and BM). Let X be a subgroup in G defined as follows. If G =
AutMat22 then X is the stabilizer of an element from &\ Y2 (Section 3.4)
and X ^ PIL 3 (4). If G = Co2 then X is the stabilizer of the subgraph S
in the coUinearity graph of ^(Coi) isomorphic to the dual polar graph of
U6(2) (the remark after (4.10.8)) and X ^ PZC76(2). Equivalently X is the
stabilizer in C02 of a vertex in the rank 3 graph IT as in (4.11.9). Finally
if G = BM then X is the stabilizer of a vertex of the Baby Monster graph
(Section 5.10), so that X = 2 -2 E6(2).2. Let Xo denote the commutator
subgroup (of index 2) in X.
Let Wo be a 1-dimensional GF(3)-module for X whose kernel is Xo.
This means that the elements from Xo centralize and the elements from
X \ Xo invert every element of Wo. Let W be the GF(3)-module for G
induced from the module Wo of X. Notice that the dimension of W is
22, 2300 and 13 571955000 (the latter being the number of vertices in
the Baby Monster graph) for G = Aut Mat22, C02 and BM, respectively.

Lemma 7.3.1 In the above terms we have the following:

(i) there exists a group X which is a non-split extension of X by a
subgroup of order 3 isomorphic to Wo as a GF(3)-module for X;
(ii) there exists a group G which possesses a homomorphism \p onto G
such that ker xp is an elementary abelian 3-group isomorphic to W
as a GF(3)-module for X and such that G does not split over ker xp.
Proof. Let U be the natural GF(4)-module of X, so that X = X/Z(X)
acts faithfully on the projective geometry of U and the dimension of U
316 2-Covers of P-geometries
is 3 , ^ and 27 for X ££ PEL 3 (4), PZ(/ 6 (2) and 2 -2 £6(2).2, respectively.
Let X be the preimage of X in the group of semilinear transformations
of U.Jt is a standard fact (proved in (2.7.12) for the case X = PZL 3 (4))
that X is a non-split extension of X by the multiplicative group of GF(4)
which is centralized by Xo and inverted by the elements from X \Xo (the
latter elements act semilinearly but not linearly on U). This gives (i) and
implies (ii) by (6.1.2). •

Since the 3-part of the Schur multiplier of Xo is of order 3 the extension

X in (7.3.1 (i)) is unique.

Lemma 7.3.2 With <99 G, X and W as above let Gn be the stabilizer in G

of an element of type n in & and Qn = O2(Gn). Then Cw(Qn) = 1.

Proof. We discuss the cases G = Aut Mat22, C02 and BM separately.

In the case G = AutMat22 we apply (6.1.7) by showing that for every

g G G the intersection X g n Q3 is not contained in X% D Q3. In the case
considered X g is the stabilizer in G of an element from 9 \ Y2 and
X^ = X8 n G where G = Mat2i is the commutator subgroup of G. On
the other hand Q'3 := Q3 n G has index 2 in Q3 and by (3.9.3 (iii)) g'3 and
Qi have the same orbits on 9 \ Y2, which shows that for every c e 0* \ Y2
we have [G(c) n 63 : G(c) n Q'3] = 2 and the result follows.

In the case G = Co2 we apply (6.1.6). The induced character x of

G is the permutational character of G on the cosets of Xo = 1^6(2)
minus the permutational character of G on the cosets of X. The suborbit
diagrams of these two permutational actions are given in (4.11.8) and
(4.11.9), respectively. Since the former of the actions has rank 5 while the
latter has rank 3, we conclude that x involves two irreducible characters.
Looking at the character table of Co2 in [CCNPW] we deduce that the
irreducibles are the ones of degree 23 and 2277. The character table of
G4 ^ (2 1+6 x 24).L4(2) can be found in GAP computer package [GAP].
Along with the character table comes the fusion map of the classes of G4
into the classes of G. This information enables us to calculate the class
function on Q4:

(la) 1 , (2a) 15+7 °, (2b) 1+15+21 °, (2c) 840, (4fc)336, (4c)560,

where (2a) 15+7 ° means that Q4 contains 85 elements from the Co2-class
2a and that these elements form two G4-classes of size 15 and 70. Now it
is straightforward to check that the restriction of x to Q4 is zero, which
 7.3 Non-split extensions 317
gives the result. An alternative proof making use of (6.1.7) can be found
in [Sh92].
The case G = BM is more complicated technically (but not conceptu-
ally). The induced character of the action of BM on the complex analo-
goue of W has been calculated by D.G. Higman in [Hig76]. The character
is the sum of three irreducible characters of degree 4371, 63 532485 and
13 508418 144. Thus if the class function on Qs were known, to check the
claim would be a matter of straightforward calculations. The character
table and the fusion pattern of G5 ~ 2 5+10+10+5 .L 5 (2) are not available
in GAP now, but they probably will be in due course. Meanwhile the
claim has been checked by two different methods. In [Wil92] and [Wil93]
R.A. Wilson has checked the condition in (6.1.7) by explicit calculations
with elements of the Baby Monster represented by 4370 x 4370 matrices
over GF(2). A different strategy was implemented in [ISh93b]. It follows
directly from (5.11.7) that for Gi 2* 2^ +22 .Co 2 and Qi = O2(Gi) the cen-
tralizer C := Cw(Qi\ as a module for G\ = Co 2 , possesses a direct sum
decomposition

where C ^ and C ^ are isomorphic to the module W (of dimen-

sion 2300) from the case G = C02 and the module C (3) (of dimen-
sion 46575) is induced from a 1-dimensional non-trivial module of
G12/61 = 210.AutMat22 (the direct summands correspond to the or-
bits Q(2a), fi(6a) and Q(4a) of G\ on the vertex set of the Baby Monster
graph). We have checked that O2(Gi5) has trivial centralizer in C®.
For i = 1 and 2 this follows from the proof of the lemma in the case
G = C02 and for i = 3 the result can be achieved by similar calculations,
since the corresponding induced character (computed by GAP) has three
irreducible components whose degrees are 23, 2277 and 44 275.
Finally we have shown (Lemma 3.10 in [ISh93b]) that in every module
of G5 = 1,5(2) induced from any 1-dimensional module of any subgroup
the subgroup, O2(Gi5/<2s) = 2 4 has a non-trivial centralizer. This shows
that if Q5 had a non-trivial centralizer in W9 then O2(Gi5) would have a
non-trivial centralizer as well, which is not the case. •
Now by (7.1.5), (7.2.2), (7.3.1) and (7.3.2) we obtain the main result of
the section.

Corollary 7.3.3 Let & be one of the geometries ^(Mat 2 2 ), &(Co2) and
Let G be the full automorphism group of & (isomorphic to
318 2-Covers of P-geometries

AutMat22, C02 and BM, respectively). Then there exist a geometry 9

which possesses a 2-covering cp : 9 —• 9, and a flag-transitive automor-
phism group G of 9 which possesses a homomorphism xp onto G, and the
kernel K of xp is a non-trivial elementary abelian 3-group. Moreover, K
as a module for G is a submodule of the module W induced from a non-
trivial 1-dimensional module of a subgroup X in G isomorphic to PEL3(4),
PE176(2) and 2 -2 E6(2).2, respectively. D

In (7.3.3) if ^ = 9(Mat22) then by (3.5.5) the order of K is at most

(and hence exactly) 3. This shows that cp in (7.3.3) is a 3-fold covering
which is universal by (3.5.7) and hence 9 = 9(3- Mat22).

1A 9(323 • Co2)
In this section we identify the structure of the group G as in (7.3.3) in
the case 9 = 9(Co2).
Throughout the section 9 is a P -geometry of rank 4, G is a flag-
transitive automorphism group of 9 and it is assumed that 9 possesses
a non-bijective 2-covering cp onto # := &(Co2) and cp commutes with
the action of G. This means that the action G which G induces on 9
coincides with the unique flag-transitive automorphism group of # which
is Co2 (7.1.8).
By (4.12.7) &(Co2) is simply connected and hence cp is not a covering.
Let xp denote the natural homomorphism of G onto G induced by cp and
let K be the kernel of xp.
Let O = {xi,...,X4} be a maximal flag in &9 where xt is of type i, and
Gt = G(xt) be the stabilizer of x,- in G. Let x,- be the image of x, in #
and Gt be the image of G* in G. Since res^(xj) is 2-simply connected for
i = 2, 3 and 4 and cp is not a covering, we conclude that K n G, is trivial
for i = 2, 3, 4 and non-trivial for i — 1. By (3.5.7) the universal 2-cover
of res^(xi) = <g(Mat22) is its triple cover ^(3 • Mat22). Thus we have the
following.

Lemma 7.4.1 res^(xi) = 9(3 • Mat22) and KnGi is of order 3. •

By the above lemma, for every element v of type 1 in ^ the subgroup
Kv := K n G(v) is of order 3. Since 9(Co2) is simply connected, the
subgroups Kv taken for all elements v of type 1 in 9 generate the whole
of K. For an arbitrary element z in 9 put

Bz = { K v \ v e res^(z)1}, K z = ( K v \ v €
 7.4 ^ ( 3 2 3 • Co2) 319

Let Sf = y ( x 3 ) and ST = &(x3) be the subgeometries of <S and # defined

as in (7.1.7) with respect to the elements X3 and X3, respectively. Let S
be the stabilizer of Sf in G and S be the stabilizer of ST in G. By (4.9.8)
we have
&^<g(SP6(2)) and S s 2i +8 .Sp 6 (2).
By (7.1.7 (ii)) and (7.4.1) Sf is a T-geometry of rank 3 possessing a
1-cover onto ^(Spe(2)) which commutes with the action of S. By (6.10.4)
^S/(3) and S/O 2 (S) ^ 3 7 • Sp6(2),
in particular K#> := K n S is an elementary abelian 3-group of rank 7.
By (6.6.4) this gives the following.

Lemma 7.4.2 In the above terms Ky> is elementary abelian of order 3 7 and
it coincides with KX3. In particular, if u and v are elements of type 3 from
the same T-subgeometry, then Ku—Kv. •
In what follows, for a subgroup Kf in K the statement "rk(K') = n"
will mean that Kr is an elementary abelian group of order 3 n .

Lemma 7.4.3 rk(X X4 ) = 15.

Proof. The set BX4 has size 15. Since res^(x4) is a protective geometry,
every two elements of type 1 incident to X4 are incident to a common
element of type 3. Hence the subgroups in BX4 commute pairwise and
it remains to show that they are linearly independent. The parabolic G4
acts primitively on BX4 and Q4 is the kernel of the action. It is easy to
see that the quotient of Q4 over the centralizer of KX4 in Q4 has order
more than 2. Hence different subgroups in BX4 have different centralizers
in Q4. In particular they are linearly independent and the result follows.
•

Lemma 7.4.4 Let {z\ = x^z^z?*) = res^({x 2? x 4 }) 3 . Then KX4 = (KZi \ 1 <
i < 3).
Proof. The elements z\9 zi9 z?> are three hyperplanes in res^(x4) con-
taining a common line (which is x 2 ). Hence every point (an element of
type 1 in the residue) is incident to zt for i = 1, 2 or 3. •

Lemma 7.4.5 Let ^4 be the element of type 4 other than X4 incident to

X3. Then KX4 n Ky4 = KX2, so that the quotient of (KX4,Ky4) over its
commutator subgroup has rank 23.
320 2-Covers of P-geometries
Proof. Let E = BM \ By4, so that BX4 = BX3 U E, and let X be the
subgroup in KX4 generated by the subgroups in E. Since G4 induces the
full automorphism group of res#(x4), G(x^)nG(y^) = G34 acts transitively
on BX3 and on E. Since different subgroups from BX4 have different
centralizers in Q4, this implies that KX3 and X are the only proper
subgroups in KX4 normalized by G(x4) n G(y4). Hence either KX4 n Ky4 =
KX3 or KM = Ky4. In the latter case since ^ is connected, we immediately
obtain K — KX4, in particular K is an elementary abelian 3-group of
rank 15. Since the shortest orbit of G\/Q\ = AxxiMatu on the set of
hyperplanes of Q\ has length 22, G has no faithful GF(3)-representations
of dimension less than 22. Since K cannot be centralized by G either, the
result follows. •

Lemma 7.4.6 Let U be the subgroup ofK generated by the subgroups Ku

taken for all elements u G res^(x2)3. Then U = (KX4,Ky4).

Proof. By (7.4.4) both KX4 and Ky4 are contained in U, so it is sufficient

to show that (KX4,Ky4) contains U. With zi, 22, 23 as in (7.4.4) let z\9 Z4
and zs be the elements of type 3 incident to both X2 and ^4. There are
exactly five T-subgeometries in ^ containing elements of type 3 incident
to X2 and the elements z, for 1 < i < 5 are in pairwise different such
subgeometries. Hence whenever u is an element of type 3 incident to *2,
£f(u) = S?(zi) for some 1 < i < 5. By (7.4.2) this means that Ku = KZi
and hence Ku < (KZi \ 1 < i < 5) < (KX4,Ky4). a

Lemma 7.4.7 rk(K) = 23.

Proof. Let x, y be elements of type 4 in ^ incident to a common

element z of type 3 and suppose that z is incident to x%. Then by
(7.4.6) (Kx,Ky) = U = (KX49Ky4). We are going to show that K = U.
Let w be an element of type 4 in ^ . Since ^ is connected, there is a
sequence of vertices wo,wi,...,ws = w such that {wo,wi} = {x^y^} and
for 0 < i < s — 1 the elements w, and wi+i are distinct and incident to
a common element of type 3. In this case for every i, 0 < i < s — 2,
the vertices w,, w,+i and Wj+2 are incident to common element of type
2 which means (7.4.6) that (KWi,KWi+l) = (KWi+l,KWi+2). This shows that
Kw < U. Since this is true for every w G ^ 4 , K = U. By (7.4.3) and
(7.4.5) KXl < KX3 is in the centre of U = K. Since K is generated by the
subgroups Kv taken for all elements of type 1, we conclude that K is
abelian. •
It is well known and easy to check that C02 has a unique faithful
 7.5 The rank 5 case: bounding the kernel 321
GF(3)-representation of dimension less than or equal to 23, which is the
unique faithful irreducible section L of the Leech lattice taken modulo
3. In the present context the isomorphism between K and L can be
established along the following lines.
Let N be the GF(3)-module for G = C02 induced from the unique
non-trivial 1-dimensional module of the group G\ = 210.AutMat22> Then

where the Nu are 1-dimensional. Let S = 2+~8.Sp6(2) be the stabilizer

in G of the C3-subgeometry SP. Put M = ©Me^i NU9 so that M is a
63-dimensional S-submodule in N. Then M possesses a unique homo-
morphism onto the 7-dimensional S-module which is the Evlattice taken
modulo 3. Let Mo be the kernel of this homomorphism and No be the
smallest G-submodule in N which contains Mo. Then by the arguments
almost identical to those for (7.4.2) - (7.4.7) one can show that N/No has
dimension at most 23. On the other hand both K and L are quotients of
N/No and hence K ^ L.

Proposition 7.4.8 Let ^ be a P -geometry of rank 4 and G be a flag-

transitive automorphism group of *&. Suppose that & possesses a non-
bijective 2-cover onto (&(Co'i) which commutes with the action of G. Then
the kernel K of the homomorphism of G onto C02 is an elementary abelian
3-group of rank 23 isomorphic to the unique faithful irreducible Co2-section
in the Leech lattice taken modulo 3 and G does not split over K. •

Thus in terms of (7.3.3) G = 3 23 Co2 and the corresponding P-geometry

will be denoted by ^ ( 3 2 3 • Co 2 ).

7.5 The rank 5 case: bounding the kernel

Let ^ be a P -geometry of rank 5, G be a flag-transitive automorphism
group of ^ and suppose that ^ possesses a non-bijective 2-covering cp
onto # = <S(BM) which commutes with the action of G. In this case
by (7.1.8) the action G induced by G on ^ is the only flag-transitive
automorphism group of the latter geometry which is the Baby Monster
group BM. Let xp be the natural homomorphism of G onto G. For x e&
or x € G we write 5c to denote (p(x) or \p(x), respectively.
Let €> = {x\,...,xs} be a maximal flag in ^, where Xi is of type i,
Gt = G(xt) be the stabilizer of xt in G, so that @ = {Gt \ 1 < i < 5} is
322 2-Covers of P-geometries
the amalgam of maximal parabolic subgroups associated with the action
of G on ^. As usual we put Gtj = G, n Gj for 1 < i < j < 5. Let K
be the kernel of xp and K,- = K fl G, for 1 < i < 5. Then the elements
of # can be identified with the orbits of K on ^ with respect to the
induced incidence relation and type function. Therefore the restriction
of cp to res^(x;) associates with every element in the residue its orbit
under Ku In particular the restriction of cp to res^(xj) is an isomorphism
if and only if Kt = 1. Since res^(xj) is 2-simply connected for i = 3,4
and 5, we have K3 = K4 = K5 = 1. On the other hand by (5.11.5 (i))
y(BM) is simply connected and since cp is non-bijective, some of the
Ki must be non-trivial. Since res^(x,)~ is 2-simply connected for every
1 < i < 5, K\ must be non-trivial. Hence the restriction of cp to res^(xi)
is a non-bijective 2-covering onto res^(xi) = ^{Co^). By (7.4.8), this gives
the following:

Lemma 7.5.1
(i) res^(xi) = ^ ( 3 2 3 • C02) and K\ is elementary abelian of order 3 2 3 ;
(ii) res^(x 2 ) + = 9(3 • Mat22) and K2 is of order 3. •

Lemma 7.5.2 If v is an element of type 2 in <& then Kv := K n G(v) is of

order 3. Moreover, the subgroups Kv taken for all elements v of type 2 in
& generate K.

Proof. By (7.5.1) we only have to prove the statement about the

generation. It is clear that L := (Kv | v € ^2 ) is normal in G and
the image in G/L of the amalgam & is isomorphic to the amalgam i%
associated with the action of G on #. Since ^ is simply connected, this
shows that G/L = G and hence L = K. •
For an arbitrary element z in ^ put
Bz = {Kv\ve res^(z)2}.
If L is a subgroup in K then the statement "rk (L) = n" will mean that
L is an elementary abelian 3-group of order 3". In this case a set of n
subgroups of order 3 in L which generate L will be called a basis of L.
In these terms (7.5.1 (i)) can be reformulated as follows.

Lemma 7.5.3 rk(KXl) = 23 and as a module for Gi/0 2 ,3(Gi) = Co2, KXl
is a section of the Leech lattice modulo 3. •
 7.5 The rank 5 case: bounding the kernel 323
Let ST = &(x4) and Sf = Sf{x4) be the subgeometries of # and <S
respectively defined as in (7.1.7). Let S and S be the stabilizers of SP and
¥ in G and G, respectively. Then by (5.4.5) we have

&^<g(Sps(2)) and S s 2 9+16 .Sp 8 (2).

By (7.5.1 (ii)) and (7.1.7 (ii)) ¥ is a rank 4 T-geometry. The restriction

of q> to 5^ is a 1-cover onto £P. Hence by (6.10.4)

• 5 ^ / ( 4 ) and 5/O 2 (S) s 3 35 • Spg(2).

By (6.7.2) we have the following.

Lemma 7.5.4 Let £f = ^(^4) be the rank 4 Tsubgeometry in & corre-

sponding to X4, S be the stabilizer of ¥ in G and K#> = S OK. Then
tk(Ky) = 35 and BM is a basis ofKy. •

Lemma 7.5.5 Let i = 3, 4 or 5. TTien rk(Kx.) = [y 2 CwWcfc is 7, 35 or 155,

respectively) and BXi is a basis of KXi.

Proof. If i = 3 or 4 then the result follows immediately from (7.5.4),

so suppose that i = 5. Since res^(x5) is a projective space over GF(2),
any two elements, say v\ and vi of type 2 incident to x$ are incident to
a common element of type 4. By (7.5.4) K^ and KV2 commute. To prove
linear independence, consider the action of Qs = 02(65) on KX5. Since Qs
is the kernel of the action of G5 on res^(xs), Qs normalizes each Kv with
v e res^(x 5 ) 2 . Let Sf = ^ ( x 4 ) . By (7.5.4), Q5 normalizes K&. Comparing
the orders of Qs (which is 230) and O2(5) (which is 225) we observe that
Qs induces on KX5 an action of order at least 2 5 . Since G5 acts primitively
on res^(x5)2, we conclude that BX5 is a basis of KX5. •
It follows directly from the proof of the above lemma that different
subgroups in BX5 have different centralizers in

Lemma 7.5.6 If 3 < i < 5 then BXl nBXi is a basis ofKXl nKXi, in particular
rk(KXl C\KXi) = 3, 7 and 15 for i = 3, 4 and 5, respectively.

Proof. Let D, = flx. n BXl, E{ = B x . \ BXl and let Jf,- and Yt be the
subgroups in KXi whose bases are Dt and £,-, respectively. Then rk(Xj) = 3,
7 and 15 while rk(Yi) = 4, 28 and 140 for i = 3, 4 and 5, respectively.
Clearly X; is contained in KXl n KXi for 3 < i < 5. The parabolic G,
induces the full automorphism group of the projective space res#(x;)~
and the kernel iV, of this action contains C^Gs). This means that Gu
324 2-Covers of P -geometries
acts transitively on D, and on E\ and in view of the remark after the
proof (7.5.5) different subgroups in Bx. have different centralizers in N,-.
This shows that Xt and Y* are the only subgroups in KXi normalized by
G\i- Hence either KXl C\KX. = Xt or KXl contains KXr If i = 4 or 5 then
the latter is impossible since the rank of KXl is only 23 by (7.5.3). If KXl
were to contain KX3 then since Y3 < Y4, it would contain the whole KX4,
and we have seen that this is impossible. •

Lemma 7.5.7 Let ys be the element of type 5 incident to X4 other than x$.
ThenKX5C\Ky5 = KX4.
Proof. It is clear that KX4 is contained in KX5 nKys. Let Z be the
subgroup in KX5 whose basis is BX5 \ BX4. Since G5 induces the full
automorphism group of res^(xs), in view of the remark after the proof
of (7.5.5) KX4 and Z are the only subgroups in KXs normalized by G45.
Hence if KX4 ^ KX5 nKy5 then KX5 = Kys. Since ^ is connected, the latter
equality implies that KX5 = K (7.5.2), which is impossible since KXl is not
contained in KX5, by (7.5.6). •

Lemma 7.5.8 In terms of (7.5.7), KXl < (KX5,Ky5).

Proof. By (7.5.6) and (7.5.7) rk(KXl CiKX5) = rk(KXl DKys) = 15 and
rk(XXl HKX5nKy5) = rk(KXl nKX4) = 7. Hence rk(KXl n (KX5,Ky5)) >
15 + 15 - 7 = 23 which is the rank of KXl by (7.5.3). •

Lemma 7.5.9 Let y\ be an element of type 1 other than x\ incident to X2.

Then the subgroups KXl and Kyi commute.
Proof. First we analyse the structure of KXl as a module for a certain
subgroup in G(xi) n G(yi). By (7.5.3) G\ acts on KXl as C02 acts on a
faithful section of the Leech lattice modulo 3. If L is the kernel of this
action then
L = KXl x O2(Gi) s 3 23 x 2^ +22 .
By (7.5.1 (ii)) and in view of the structure of the maximal parabolic sub-
groups associated with the action of C02 on ^(Co2), we have Gn/Qi =
3.210.Aut Mat22- Let H be the preimage in G\2 of the unique subgroup of
index 2 in G12/Qu so that H := HL/L = 210.Mat22- Since there are only
three elements of type 1 incident to xi, H is contained in G(xi) n G(y\).
We claim that KXl = KX2 0 [XXl, O2(H)] and [KXl, O2(H)] is an irreducible
22-dimensional GF(3)-module for H. This can be seen either by restrict-
ing to H the action of C02 on the Leech lattice modulo 3 or directly,
 7.5 The rank 5 case: bounding the kernel 325
since the shortest orbit of H/Oi{H) on the hyperplanes in O2C&) has
length 22. In view of this decomposition and since KX2 is contained in
Kyi, in order to prove the lemma it is sufficient to find an element in
KXl \KX2 which commutes with Kyr
By (7.5.7) and (7.5.5) if ys is an element of type 5 other than xs incident
to X4 then KM is contained in the centre of (KX59Kys) and by (7.5.8) the
latter group contains both KXl and Kn. Hence KM C\KM (which is of
rank 7 by (7.5.6)) commutes with Kn and the result follows. •
Let us introduce some notation. Put

Ui(x5) = (Kx\xe res^(x 5 ) 1 ),

U2(x5) = (KX5,Ky | {x 5 ,y} = TQS^V)5 for some v e <#4),

so that U\(x5) and 1/2(^5) are the subgroups in K generated by the
subgroups Kx taken for all elements of type 1 incident to X5 and for
all elements of type 5 incident with X5 to a common element of type 4,
respectively.
Since TQS^(XS) is a projective space, any two elements of type 1 incident
to X5 are incident to a common element of type 2. By (7.5.9) this shows
that U\(x5) is abelian.

Lemma 7.5.10 In the above terms f/i(xs) = L^tes).

Proof. By (7.5.8) if y$ is an element of type 5 other than X5 incident to
X4 then KXl < (KX5,Kys) and hence Ui(xs) < Uzixs). Let ys be as above
and a be an element of type 2 incident to ^5. Since resyiys) is a projective
space there is an element x of type 1 in this residue which is incident to
both a and X4 (a line and a hyperplane always have a common point).
It is clear that x is incident to X5. Hence Ka < Kx < U\(xs). Since Kys is
generated by all such subgroups Ka, we have ^2(^5) < Ui(xs). •
Let A be the derived graph of ^ whose vertices and edges will be
identified with the elements of type 5 and 4 in ^ , respectively. For a pair
x, y of vertices in A let d(x,y) be the distance in A between these vertices.
As usual for a vertex y let At(y) denote the set of vertices at distance i
from y.
For a vertex y of A put
Vi{y) = (Kz I z € V(A)9 d(z,y)<i).
Then V0(y) = Ky and Vx{y) = Ux{y) = U2(y) by (7.5.10).
326 2-Covers of P-geometries
Lemma 7.5.11 rk{Vi(y)/V0(y)) < 248.
Proof. By (7.5.10), V\(y) is generated by 31 subgroups Kx for x G
res^y) 1 , each of them has rank 23 and intersects Vo(y) in a subgroup of
rank 15 (7.5.6). Hence Tk(Vi{y)/V0(y)) < 31 • 8 = 248. •

Lemma 7.5.12 Let z e Ai(y). Then tk(Vi(z)/(V0(y)9 V0(z))) < 128.

Proof. Let v be the unique element of type 4 in ^ which is incident to
both y and z. The group V\(z) is generated by the subgroups Kx taken
for all x e res^(z)1. If such an x is incident to v, then Kx < (Vo(y), Vo(z))
by (7.5.8). There are exactly 16 elements x which are not adjacent to v9
each of them giving contribution of rank at most 8 (indeed, by (7.5.3)
the rank of Kx is 23, while the rank of Kx nKz is 15 by (7.5.6)). •
If w is a vertex at distance 2 from y in A then Kw is contained in
V\(z) for the unique vertex z adjacent to both w and x. Hence V\(y)
and the subgroups V\(z) taken for all vertices z adjacent to y generate
V2(y). Since there are exactly 31 such vertices z, by (7.5.12) we obtain the
following.

Corollary 7.5.13 The quotient V2(y)/V\(y) is generated by at most

31 • 128 = 3968 subgroups of order 3. •
By the above result if V2(2) is abelian then rk(V2(y)/Vi(y)) < 3968.

Lemma 7.5.14 Let y be a vertex of A, u be an element of type 3 incident

to y. Let {1^1,1^2,^3} be the elements of type 4 incident to both u and y; and
for i = 1, 2, 3 let z\ be the element of type 5 other than y incident to vt.
Then
Vi(y) = (V0(y\ F0(z!), V0(z2), V0(z3)).
Proof. In the residue of y the elements v\,v2 and v3 are three hyper-
planes having the subspace u of codimension 2 in common. Therefore,
every element x of type 1 incident to y is incident to zt for some i,
1 < i < 3. By (7.5.8), Kx < (V0(y), Vo(zi)). •

Lemma 7.5.15 V3(y) = V2(y).

Proof. Let z € A2(y). By (7.1.2 (v)) we can find an element u of type 3
in ^ which is incident to both y and z. Let {v\,v29V3} be the elements of
type 4 incident to both u and z. For i = 1, 2 and 3 let z\ be the element
of type 5 other than y incident to vt. Since res^(w)+ is the geometry of
 7.6 ^(3 4 3 7 1 • BM) 327
edges and vertices of the Petersen graph, and since the Petersen graph
has diameter 2, we have
Fi(z) < (Vo(z), V0(z1), F0(z2), F0(z3)) < V2(y),
where the first of the inclusions is forced by (7.5.14). •
Since A is connected, we immediately obtain the following.

Corollary 7.5.16 For every y e&5 we have K = V2(y). O

Lemma 7.5.17 K is abelian.

Proof. By (7.5.16), K = V2(y). Let us show that V0(y) is in the centre
of K. Let z be a vertex at distance at most 2 from y in A and a be a
vertex which is at distance at most 1 from both y and z. By the remark
before (7.5.10), V\(a) is abelian. Hence Vo(y) and Vo(z) commute. By
(7.5.15) the subgroups VQ(Z) taken for all vertices z with distance at most
2 from y in A generate K. Hence Vo(y) is in the centre of K. Again using
the fact that K is generated by the subgroups Vo(y) for all the vertices y
of A, K is abelian. •

Proposition 7.5.18 Let & be a P-geometry of rank 5 and G be a flag-

transitive automorphism group of (§. Suppose that & possesses a non-
bijective 2-cover onto the geometry &(BM) which commutes with the action
ofG and let K be the kernel of the action of G on &(BM). Then K is an
elementary abelian 3-group of rank at most 4371.
Proof. By (7.5.17), K is abelian. By (7.5.5), (7.5.11) and (7.5.13),
rk(K) < 155 + 248 + 3968 = 4371.
•
7.6 ^(3 4 3 7 1 • BM)
Let ^ be the geometry as in (7.3.3) possessing a 2-covering onto ^(BM).
Then ^ and G satisfy the hypothesis of (7.5.18) and hence the intersection
of G and W (which is the kernel of the homomorphism of G onto BM)
is of rank at most 4371.

Lemma 7.6.1 In the above terms let K = GnW. Then

(i) rk(K) = 4371,
(ii) K is an irreducible GF(3)-module for BM,
(iii) G does not split over K.
328 2-Covers of P-geometries
Proof. We are going to analyse the structure of K as a module for
G/K = BM. Let 5c be an element of type 1 in ^ and H be the stabilizer of
x in G. Then by (7.5.1) H f)K is of rank 23. The quotient H := H/(HDK)
can be naturally identified with the stabilizer in BM of an element of
type 1 from <#(BM\ so that H s 2l+22.Co2. By (7.5.3) Q := 02(H) is the
kernel of the irreducible action of H on K D H. Let Z be the centre of
Q which is also the centre of H. Put
K{=CK(Q), K2 = CK(Z)n[K,Q], K3 = [K,Z],
Then clearly K = Xi 0 X2 © £3 as a module for 5 . Since Xi contains
K n if, by the above rk(Ki) > 23. Since H acts faithfully on K, K3 must
be non-trivial. Since Q is isomorphic to 2++22, it must act faithfully on K3,
the rank of the latter is at least 2048 = 211 which is the dimension of the
unique faithful irreducible GF(3)-representation of Q. From the structure
of &(BM) we know that Z is conjugate in BM to a non-central subgroup
Z' in Q which means that CK(Z) and CK(Z') have the same rank. By
definition K\ is centralized by both Z and Z'. On the other hand since
Z is in the centre of H while Zf is not in the centre, we observe that
CK3(Z) is trivial while CK^Z') is non-trivial. Since the ranks of CK(Z)
and CK(Z') are isomorphic, K2 must be non-trivial. Since Q is normal in
H and it induces on K2 an elementary abelian 2-group Q/Z, Clifford's
theorem implies that rk(X2) is at least the length of the shortest orbit of
Co2 = H/Q on the set of hyperplanes in Q/Z. Since Q/Z is self-dual,
this gives rk(K2) > 2300. Now summing up we obtain
rk(X) = rk(Xi) + rk(X2) 4- rk(K3) > 23 + 2300 + 2048 = 4371.
Since this lower bound meets the upper bound from (7.5.18), (i) follows.
By the above it is clear that K does not involve trivial composition
factors. For a faithful submodule L in K put L2 = CL(Z) PI [L,Q] and
L3 = [L9Z], Then by arguments as in the above paragraph both L2
and L3 must be non-trivial of rank at least 2300 and 2048, respectively.
Since the sum of these numbers exceeds half of the rank of K, L
cannot be proper and we obtain (ii). From the proof of (i) we see that
K\ = CK{Q). Since H = NBM(Q) we conclude that N := N~(Q)/Q is the
full automorphism group of the residual geometry res-(x) = ^(3 23 • Co2).
If G should split over K, N would split over K\, but we know that this
is not the case. •
Thus G ^ 34371 BM and the corresponding P-geometry ^ from (7.3.3)
will be denoted by ^(3 4371 • BM).
 7.6 <g(34m - BM) 329
Proposition 7.6.2 ^(3 4371 • BM) is the universal 2-cover of^(BM).
Proof. By the construction ^ := ^(3 4371 • BM) possesses a 2-cover
onto # := @(BM). Thus to prove the proposition it is sufficient to show
that G is the universal completion of the amalgam # of rank 2 parabolic
subgroups associated with the action of BM on §. Let H be the universal
completion of # and % be the homomorphism of H onto BM. Since ^
possesses a 2-covering onto # which commutes with the action of G,
# is also the amalgam of rank 2 parabolic subgroups associated with
the action of G on <S. Hence there is a homomorphism rj of H onto G,
such that x is the composition of ff and the homomorphism \p of G onto
BM. All proper residues in ^ are 2-simply connected and hence there
is a subamalgam Of in H which maps isomorphically under t/> onto the
amalgam of maximal parabolic subgroups in G associated with its action
on CS. Thus we can construct a P -geometry ffl acted on flag-transitively
by H and possessing a covering onto (S. To wit, the elements of #? are
the cosets in H of the subgroups constituting Q). By the construction Jf
possesses a 2-cover onto # which commutes with the action of H. Now
by (7.5.18) the kernel of x h a s order at most 34371 and xp must be an
isomorphism. D
Let K be as in (7.6.1). Using the technique developed in this chapter
(Section 6 in [ISh93b] for details) one can show that K is the unique
faithful GF(3)-module of BM of dimension 4371 or less. In particular
if Mp is the module obtained by taking modulo p the J3M-module
M of dimension 4371 over the rationals, then K = M3. We conclude
the section with the following result concerning cohomology of certain
representations of BM.

Lemma 7.6.3 Let G be a group possessing a homomorphism onto BM with

kernel isomorphic to the module Mp as above for p > 3. Then either G
splits over Mp or G is the automorphism group of the geometry
4m
• BM).
Proof. One can easily check that for every p ^ 2 the centralizer
in Mp of Qs is trivial. Then the result follows from a straightforward
generalization of (7.2.2), for the case when the characteristic of W is
more than 3, and (7.6.2). •
Let G be the non-split extension of BM by the GF(3)-module W
induced from the non-trivial 1-dimensional module of X = 2-2Ee(2).2, so
that G contains the automorphism group G of ^(3 4371 • BM). Then G/K
330 2-Covers of P-geometries
splits over W/K since G/K is a complement. Similar observations can
be made for the extensions of C02 and AutMat22-

7.7 Some further s-coverings

It has been shown in [Wie97] and [BIP98] that the 2-coverings
^(3 2 3 • Coi) -> <Z(Co2) and ^(3 4 3 7 1 • BM) -> ^(BM)
induce 1- and 2-coverings of certain subgeometries in ^{Coi) and
We have seen in Section 4.13 that C02 contains the McLaughlin group
McL as a subgroup and that the latter acts flag-transitively on a geometry
&(McL) with the diagram
P
2 2 1 1
It can be seen that the parabolic subgroups associated with the action
of McL on &(McL) are contained in the parabolic subgroups associated
with the action of Co2 on ^ ( C ^ ) - Let xp be the natural homomorphism
of 3 23 • Co2 onto Co2.

Proposition 7.7.1 [BIP98] The full preimage of McL under xp, which is a
non-split extension of the form 3 23 • McL, acts flag-transitively on a geom-
etry <g(323 • McL), and xp induces a 2-covering <&(323 • McL) -+ &(McL). •
23
The residues in ^ ( 3 • McL) of elements of type 1 and 4 (isomorphic
to J f (3 • Alt-]) and <S(Mat22)) are the universal (triple) covers of the
corresponding residues in &(McL) (isomorphic to J^(Altj) and <S(Mat22)->
respectively). We do not know whether or not ^ ( 3 2 3 • McL) is simply
connected.
Also in Section 4.13 we have indicated in <&(McL) a subgeometry
#(I74(3)) with the diagram

2 2 2

on which a subgroup of C02 isomorphic to l/4(3).22 induces a flag-

transitive action.

Proposition 7.7.2 [BIP98] The full preimage ofU^O).!2 under xp, which is
a non-split extension of the form 3 23 • l/4(3).22, acts flag-transitively on a
geometry ^ ( 3 2 3 • C/4(3)) with the diagram
 7.7 Some further s-coverings 331
23
There exists a covering of&(3 • 1/4(3)) onto the residue of an element of
type 1 in the geometry &(3 • M(24)) as in (5.8.4). •

We have observed in Section 5.5 that BM contains a subgroup iso-

morphic to 1M(2) which is the automorphism group of a Tits geometry
with the diagram

Let xp denote the natural homomorphism of G = 3 4371 • BM onto BM.

Proposition 7.7.3 [Wie97] Let si be the amalgam of minimal parabolics

associated with the action ofF4(2) on ^(F^(2)). Then there is a subamalgam
si in G such that the restriction ofxp to si is an isomorphism onto si. The
subgroup in G generated by A is a non-split extension of the form 3833-F4(2)
which acts flag-transitively on a geometry ^(3 8 3 3 • F^{2)) with the diagram

and xp induces a 1-covering ^(3 8 3 3 • F4(2)) -> ^(F 4 (2)). •

833
It is not known up to now whether or not ^(3 • ^ ( 2 ) ) is simply
connected.
 8
7-groups

In early stages of studying the Monster group M in the 70's, B. Fischer

had noticed that M can be generated by 15 involutions with pairwise
products of order 2 or 3 corresponding to the following Coxeter diagram
and found a few nice non-Coxeter relations satisfied by the involutions
in the Monster.

Around 1980 J.H. Con way conjectured that 16 involutions satisfying

the Coxeter relations of the 7555 diagram given on the next page together
with the so-called "spider" relation
(abiCiab2c2ab3C3)10 = 1
constitute a presentation for a group called the Bimonster, which is the
wreath product M I 2 of the Monster and a group of order 2. Many
people contributed to the proof of Conway's conjecture which has been
completed in 1990 (1.13.5).

332
 8.1 Some history 333
A crucial role in the proof of the Y555 theorem was played by the
simple connectedness result for the tilde geometry of the Monster. In this
chapter we review the original proof of the Y555 theorem and present an
alternative proof based on an inductive approach to 7 -groups.

8.1 Some history

We start with the above Coxeter diagram known as the 7555 diagram
and the following relation known as the "spider" relation:
{ab\C\ab2C2ab?>c?)m = 1.
For 2 < p, q, r < 5 define Ypqr be the quotient over the spider relation of
the Coxeter group, whose generators are a and
first p terms from &i,ci,di,ei,/i,
first q terms from b2,C29d2,e2,f2,
first r terms from b^c^d^e^f^
and whose Coxeter relations correspond to the subdiagram of the 7555
diagram induced by the generators. A homomorphic image of the group
Ypqr will be called a 7pgr-group and the Coxeter generators of Ypqr will
usually be identified with their images in a 7pgr-group. If Z is a Ypqr-
group and x,y,...,z are some Coxeter generators of Z (or rather of Ypqr)
then Z [x, y,..., z\ denotes the subgroup in Z generated by all its Coxeter
generators except for x,y,...,z. In these terms if x is the terminal node of
the left arm of the Coxeter diagram of Ypqr and p > 3 then Ypqr [xj is a
334 Y-groups
If min{p,q,r} < 2 then we define Ypqr as YPiqiri[x,...,z\ where p\ =
min{2,p}, q\ = min{2,g}, r\ = min{2,r} and x,...,z are the nodes in the
Coxeter diagram of YPiqiri whose removal gives the Coxeter diagram of
Ypqr. Suppose that p — l,q,r > 2 and that x is the terminal node of the
left arm of the Coxeter diagram of Ypqr. Then a YMr-group Z is said to
be strong if Z\x\ = Y^-\)qr.
If p, q, r > 2 then every defining relation of Ypqr has even length which
implies the following.

Lemma 8.1.1 Suppose that 2 < p,q,r < 5, that Z is a Ypqr-group and that
O2(Z) = Z . Then the direct product of Z and a group of order 2 is also a
D

The structure of the groups Ypqr is given in the following table.

pqr Ypqr [Ypqr • Y(p-l)qr]

321 2 x SP6(2) 56
421 2-Q+(2):2 240
7
331 2 .(2 x SP6(2)) 128
431 2 x Sp&(2) 255
441 nro(2) = 2 528
222 3 5 : Q5(3) : 2 243
322 2 x Q7(3) 728
422 2 • Il+(3) : 2 2160
332 2 x 2 - M(22) 28160
432 2 x Af (23) 31671
442 3 • Af (24) 920808
333 2 x 2 2 -2 £ 6 (2) 2370830336
433 2 x 2 • BM 27143910000
443 2xM 97239461142009186000
444 M\2 \M\ ~ 1054
 8.2 The 26-node theorem 335
The groups above Y442 have been identified by means of coset
enumeration on a computer in [CNS88], the group Y442 has been iden-
tified by D.Z Djokovic also by coset enumeration on a computer and a
computer-free identification, achieved by studying a certain hyperbolic
reflection group, can be found in [CP92]. The group Y333 has been iden-
tified using double coset enumeration performed by S.A. Linton ([Lin89]
and [Soi91]). The isomorphism type of Y443 was proved by combining
the results in [Nor90], [Nor92] and in [Iv91a], [Iv92a] (see also [Con92]).
The group 7433 has been identified in [Iv94]. It has been proved in
[Soi89] that the isomorphism Y443 = 2 x M implies the isomorphism
Y444 = M12. An independent characterization of Fischer groups as 7 -
groups can be found in [Vi97]. The groups Yp22, p > 5, were identified
in [Pr89] with certain orthogonal groups over GF(3) (we do not present
these results here). If q > 3, r > 2 then Y5qr = 74^r, while Y^i and
higher Y-groups collapse to a group of order 2 ((8.5.4) and (8.5.5)). It is
worth mentioning that Ypqr maps isomorphically onto its natural image
in Y555 except for the groups 7 m and 7422 which lose their centres of
order 2.

8.2 The 26-node theorem

In this section we discuss the 26-node theorem proved in [CNS88] and
related results.

Theorem 8.2.1 The group Y555 contains a set of 26 involutions including

the set of 15 generators from the 7555 diagram, which satisfy the Coxeter
relations given by the incidence graph Z of the projective plane of order
3. The subgroup in Y555 which conjugates the vertex set of Z onto itself
induces the full automorphism group oflL isomorphic to Li(3) : 2. •

Notice that the generators of M discovered by B. Fischer correspond

to a subgraph of E.

By the 26-node theorem (8.2.1) 7555 is a quotient of the Coxeter group,

whose diagram is the projective plane of order 3. L.H. Soicher [Soi91]
found a very simple hexagonal relation which characterizes 7555 as such
a quotient.

Theorem 8.2.2 Let E be the incidence graph of the projective plane of order
3 and let (u,v,w,x,y,z) be a cycle of length 6 in S. Let C be the Coxeter
336 Y-groups
group of Z subject to the single additional relation
(uxvywz)4 = 1.
Then C £ Y555. •
By (1.13.5) Y555 is isomorphic to the Bimonster group B := M I 2
whose commutator subgroup D is isomorphic to the direct product of
two copies of the Monster M. This shows that if p is a vertex of I
then p is not contained in D and CB(P) = (p) x M. Let ©denote the
set of involutions in £ non-adjacent to p. Then every involution from ©
commutes with p and the subgraph of L induced by ©is the incidence
graph of the affine plane of order 3. It turns out that the subgroup in
Cs(p) generated by the involutions from 0 is isomorphic to the Monster.
In [Mi95] the involutions from ©are described explicitly in terms of the
action of M on the Moonshine module so that the Coxeter relations can
be checked, although it is not clear how difficult it would be to check a
non-Coxeter relation, say the hexagonal relation.
The following very elegant characterization of Y555 proved in [CP92]
was used originally for showing that the Bimonster is a quotient of Y555.

Theorem 8.2.3 Let G be a group (finite or infinite) which contains a sub-

group A = Syms such that C := CQ{A) = Symn, and if B is a 1-point
stabilizer in the natural permutational action of C then B = Ag for some
g G G. Suppose that no proper subgroup of G possesses the same property.
Then either G = Symn or G is a quotient of Y555. •
In [Nor90] S.P. Norton, using the 26-node theorem in a crucial way,
has determined subgroups of Y555 of the shapes
21+26(224 : COl) and (210+16 x 210+16).Q+(2)
which correspond to some maximal parabolic subgroups associated with
the action of M on its maximal 2-local parabolic geometry Jf(M).
It was realized during the Durham Symposium on Groups and Combi-
natorics in July 1990 that the simple connectedness result for the 2-local
tilde geometry ^(M) of the Monster [Iv92a] can be used to identify Y555
with the Bimonster. Using his earlier results on Y -groups Simon Norton
proved in the course of the symposium that the derived subgroup of Y443
is generated by a subamalgam {C,N,L} with
, N

2 3+6+12+16 .(L 3 (2) x 3 • Sym6)

 8.3 From Y-groups to Y-graphs 337
and [N :NnC] = 3, [L : LnC] = [L : LnN] = 7. As shown in [Iv92a],
up to isomorphism there is only one such amalgam which consists of
parabolic subgroups of the action of M on ^(M) and M is the only
completion of this amalgam. That is how the proof of the 7555 theorem
was achieved.
After that, Y433 remained the only unknown case in the identification
problem for Y -groups. The original proof of the Y555 theorem reduces
the problem (via the 26-node theorem and the simple connectedness of
^(M)) to the triangulation problem for the second Monster graph. It
turns out that there is a more direct way to associate a graph with a
Y -group and to reduce the identification problem to the triangulation
problem for that graph. This approach was realized in [Iv94] for Y433. The
identification problem was reduced to the triangulation problem of the
graph ft as in Section 5.12. The triangulability of ft was proved in [Iv92c]
within the simple connectedness proof for the Petersen type geometry of
the Baby Monster. After that it became possible to apply an inductive
approach to identify all the Y -groups. This approach, which gives an
alternative proof of the 7555 theorem, is discussed in the remainder of
the chapter.

8.3 From Y-groups to Y-graphs

We start this section with a definition. Let A be a graph and G be a
vertex- and edge-transitive automorphism group of A. Let H be another
graph and H be an automorphism group of S which is also assumed to
be vertex- and edge-transitive. As usual, for a vertex a 6 S by B,-(a) we
denote the set of vertices at distance i from a and write 3(a) instead of
Si (a), while if (a) denotes the stabilizer of a in H. Then (3, if) is said to
be weakly locally (A, G) if for every a € 3 there is an isomorphism
p«:(A,G)-+(S(a),ff(a))
of permutation groups such that whenever {x,y} is an edge of A,
{(pa(x), (pa(y)} is an edge of 3. Notice that if (3,H) is weakly locally
(A,G) then H is a transitive extension of G ([Su86], p. 545). Identifying
A and B(a) via (pa we can say that the subgraph in 3 induced by 3(a) is
a union of some orbitals of the action of G on A and this union contains
the orbital formed by the edges of A. When H and G are clear from the
context we simply say that 3 is weakly locally A.
Suppose that Z is a Yp^r-group, where p > 2, x is the terminal node of
the left arm of the Coxeter diagram of Ypqr and y is the node adjacent to
x. We are dealing with the left arm just in order to simplify the notation.
338 Y-groups
Define a Y-graph F = T(Z,x) to be a graph on the set of right cosets
in Z of the subgroup Z[x\ in which two cosets Z[x\gu Z[x\g2 are
adjacent if there is an element h\ in the former coset and an element hi
in the latter coset such that h^ = x^i- In other terms the edges of F are
the images under the natural action of Z of the pair e := {Z [x\,Z [x\x}.
If Z(e) is the elementwise stabilizer of the edge e then
Z(e) = Z[x\nZ[x\x.
It is obvious that the latter group contains Z[:x,j/] and the Y -graph F
will be called correct if Z(e) = Z[x,y\.
Let a = Z[x\, P = Z[x\x, y = Z[x\xy9 H = (x,y) = Sym3 and
suppose that F is correct. Then Z\x\ = Z(a) acts on F(a) as it acts on
the cosets of Z[x,yJ. Furthermore, since {xyf = 1 and y e Z[x\ we
have
yx = Z [xjxyx = Z Lx
which shows that T := {a,/J,y} is a triangle in F on which H induces
the natural action. The images of T under Z will be called Y-triangles.
Thus the action of Z(a) on F(a) is similar to its action on the vertex
set of A := F(Z[xJ,y) and two vertices in F(a) are adjacent whenever
the corresponding vertices in A are adjacent. This shows that F(Z,x)
is weakly locally T(Z[x\,y) (notice that Z\x\ is a 7(p_i^r-group). We
summarize the most important case of this observation in the following.

Lemma 8.3.1 Suppose that Z is a strong Ypqr-group where p — l,q9r > 2

and that T(Z,x) is correct. Then T(Z9x) is weakly locally T(Y(p-\)qr9y). •

Suppose that both F(Z, x) and T(Ypqr, x) are correct. This is the case, for
instance, when F(Z,x) is correct and Yj^-i^rL^J i s a maximal subgroup
of Y(P-i)qr. Then the natural homomorphism
q> : Ypqr - » Z
induces a covering
xp :
of graphs such that the Y-triangles are contractible with respect to xp.
This gives the following

Lemma 8.3.2 Suppose that Z is a strong Ypqr-group and that both T(Z,x)
and r(Ypqr9x) are correct. Suppose further that the Y-triangles in F(Z,x)
generate the fundamental group ofT(Z,x). Then Z = Ypqr. •
 8.3 From Y-groups to Y-graphs 339
The next lemma shows that in some cases examples of Y -groups can
be constructed via their Y -graphs.

Lemma 8.3.3 Let y be the terminal node of the left arm of the Y(p-i)qr
diagram, where p — l,q,r > 2 and z be the node adjacent to y. Let S be
a graph and Z be a vertex- and edge-transitive automorphism group of E
and suppose that the following conditions hold for a being a vertex of S.

(i) r(Yip-i)qr,y) is correct;

(ii) (S,Z) is weakly locally (r(Y(p-i)qr,y\Y(p-i)qr) and q>a is the corre-
sponding isomorphism;
(iii) if ft = (pa(Y{p-i)qr[y\) then the setwise stabilizer in Z o/{a,/?} is
the direct product of Z (a) n Z(/?) and a group of order 2 generated
by an element x;
(iv) the setwise stabilizer in Y(p-i)qr of {Y(p-\)qr\y\, Y(p-\)qr\y\y} is the
direct product (y) x Y^p-\)qr\y,z\ and (y) is the centre of this sta-
bilizer.

Then Z is a strong Ypqr-group.

Proof. The Coxeter generators of Z are x and the set K of (the images
under cpa of) the Coxeter generators of Y(p-i)qr. By (ii) the generators in K
satisfy the Coxeter relations and the spider relation. By (iii) x commutes
with all the generators in K except for y. The product xy induces an
action of order 3 on the triangle T = {a,/},y} where y = (pa(Y(p-i)qr[y\y)-
Hence a := (xy)3 is in the elementwise stabilizer L of this triangle. By (i)
L = Y(p-i)qr [y, z\, by (iii) and (iv) a is in the centre of L and this centre
is trivial by (iv). Hence (xy)3 = 1 and the result follows. •

Our inductive approach to Ypqr is the following. We consider a group Z

acting vertex- and edge-transitively on a graph S and we show eventually
that S is T(Ypqr,x) where x is the terminal node of the left arm of the
Ypqr diagram. First we show that S is weakly locally T(Y(p-i)qr,y) where
y is the node adjacent to x. Then we check the conditions in (8.3.3) and
conclude that Z is a strong Yp^-group. Finally we show that the Y-
triangles generate the fundamental group of H and conclude from (8.3.2)
that Z = Ypqr. On the last step we often use (1.14.1). In some cases we
will be able to show that the covering of S under consideration induces
another covering of graphs which is known to be an isomorphism. For
this we use the strategy introduced in [Iv94].
340 Y-groups
8.4 Some orthogonal groups
In this section we identify Y222, Y322 and 1422- The diagram of Y222 is
afBne of extended E6-type and hence the generators, when subject to the
Coxeter relations only, produce a group isomorphic to Z 6 : Gs(3).2. By
carrying out explicit calculations in the latter group we will find out the
effect of adjoining the spider relation.
Let H = Cox(Y222) be the Coxeter group and let V = R 6 be a 6-
dimensional real vector space with the natural inner product ( , ). Then
H can be realized as a group of affine transformations of V in the
following way.
Let %J be the set of Coxeter generators of H. It follows from the general
theory of (affine) Coxeter groups ([Ebe94], [Hum90]) that there exists an
(essentially unique) system {r(x) | x € $/} of roots in V such that the
angle between r(x) and r(y) is 120 or 90 degrees whenever x and y are
adjacent or non-adjacent nodes, respectively. To wit, if {ij,k} = {1,2,3}
then we choose {r(x) | x € W \ {c,}} to be a fundamental system of roots
in an ^-lattice F and
r(a) = r(Cj) + r(ck) + 2r(fc) + 2r(bj) + 2r(bk) + 3r(a)
is the longest root in F with respect to this fundamental system.
For a root r e V let l(r) be the reflexion with respect to the hyperplane
orthogonal to r. Let L denote the group of orthogonal transformations
generated by the reflexions /(r(x)) for all x e * Then L is the Coxeter
group Cox(Ee), which is known to be isomorphic to members of a number
of series of classical groups over fields of characteristic 2 and 3:
L s Ojr(2).2 s 174(2).2 S Sp4(3).2 S Q5(3).2.
In this chapter we will mainly use the isomorphism L = Qs(3).2. The
roots r(x) generate the ^-lattice T. Consider the semidirect product
T : L with respect to the natural action. Then every element g eT : L
can be uniquely represented by a pair (l(g)j(g)) where l(g) is a linear
transformation of V and t(g) is a translation. In this case the action of
g on V is given by the following:
g : v •-» /(g) • 1; + f(g)
where 1? € F and /(g)• t; is the image of v under /(g).
The Coxeter generators of H can be chosen in the following way:
for every x G ^ w e take l(x) = /(r(x)). Furthermore f(x) = 0 if x =£ c3
and £(c3) = r(ci). Then all the Coxeter relations are satisfied, the group
generated is F : L and we obtain the required realization of H.
 8.4 Some orthogonal groups 341
Now let us analyse the effect of adjoining the spider relation. This
relation can be written in the form cp10 = 1 where q> = ab\C\ab2C2ab3c3.
We have chosen our generators so that t((p) = r(c3). Then l(cp10) = l((p)10
and
t(cpi0) = l(cp)9 • r(c3) + ... + l(q>) • r(c3) + r(c3).
One can easily calculate the roots l{(py • r(c3) as given below in the basis
{r(x) | x G <3f \ {ci}}, which is another fundamental system of roots in F :

' • r(c3)

0
1 r{b3)
2 -r(bi)-r(a)-r(b2)-r(c2)
3 r{bx) + 2r(a) + 2r(b2) + r(c2) + 2r(b3) + r(c3)
4 -2r(bi) - 3r(a) - 2r(b2) - r(c2) — 2r(b3) - r{c3)
5 rih) + r{a) + r(b2) + r(c2) + r(b3) + r(c3)
6 -ribO-ria)-^)-^)
7 r(c2)

9 -r(bl)-2r{a)-r(b2)-r(c2)-r(c3)
10 r(c3)

The above table shows that the orbit of r(c3) under the subgroup
generated by l(<p) is of length 10 and it is easy to see that the vectors
from this orbit generate the lattice T. This gives

Lemma 8.4.1 /(<p)10 is the identity element of L. Q

10
The expression for t(q> ) shows that this element is equal to the sum
of all the rows in the above table which gives
t(cp10) = -3r(bi) - 3r(a) - r(b2) + r(c2) + r(b3) + 2r(c3).
Let K be the kernel of the homomorphism H —• Yni- By (8.4.1) the
image of K in H/T = L is trivial and hence K < F. Since F is abelian,
K is generated by the images of t(cp10) under conjugation by elements of
L.

Lemma 8.4.2 K = (t((plo),3T).

342 Y-groups
Proof. By the definition K contains t(cp10). Let us show that K also
contains 3F. In fact, a direct calculation shows that
t(cplo)-l(c2)'t(cplo) =
Of course, the images of 3r(c2) under L generate 3F. To show the reverse
inclusion it is sufficient to show that l(x) • t((p10) G (t((pm), 3F) for every
x G Q) \ {ci}. Since l(x) is the reflexion associated with r(x), we have the
following:
l(x)-t(<pw) = t(<p10)-(t((pm),r(xMx).
One can check that for x G <& \ {c\\ the inner product (t(cplo),r(x)) is
divisible by 3 and hence the result follows. •
Since t((p) does not belong to 3F, we have the following.

Proposition 8.4.3 T/K ^ 3 5 and Y211 ^ 3 5 : Q5(3).2. •

There is an orthogonal form on 03(7222) and 7222/03(7222) is the full
automorphism group of this form. Then F(7222,ci) is a graph on the
set of all vectors in a 5-dimensional GF(3)-space W with a non-singular
orthogonal form, such that v,w G W are adjacent if (v + w) is a plus
vector, which means that the orthogonal complement (v + w)1- contains a
2-dimensional totally singular subspace. It is straightforward to calculate
that the suborbit diagram of F(7222,ci) is the following

Let us turn to 7322- Let W be a 7-dimensional GF(3)-space with a non-

singular quadratic form and Z = 2 x Cli(3) be the full automorphism
group of this form. Let S be a graph on the set of non-zero isotropic
vectors in W in which two such vectors are adjacent if their inner product
is 1. Direct calculations show that the suborbit diagram is the following
 8.4 Some orthogonal groups 343
72 72

If a G 2 then Z(a) ^ 3 5 : Q5(3) : 2 ^ y222 and O 3 (Z(a)) acts regularly

on S(a) which immediately shows that S is weakly locally F( Y222,ci). It
is easy to check that the remaining conditions in (8.3.3) are also satisfied
and hence Z is a strong Y322-group. From the above diagram we see that
all triangles in S are Y -triangles. Direct calculations in the orthogonal
module W enable one to check the conditions in (1.14.1). Thus S is
triangulable, hence Z s Y322 and S ^ F(Y 322 ,di) by (8.3.2).
For {i,j,k} = {1,2,3} the nodes a, bu c,, du bj, Cj, bk on the Y555 diagram
induce a spherical ^-diagram, so that the corresponding Coxeter group
is isomorphic to Spe(2) x 2 and its centre is generated by the following
element [CNS88]:

For i = 2 and 3 put Xt = Y322 [cj and let Z* be the subgraph in F( Y322, d\)
induced by the images of Y322[diJ under Xt. The Coxeter diagram of Xt
is spherical of type E-j and since all the Coxeter generators in Y322 are
pairwise different, either Xt = Sp6(2) x 2 or Xt = Sp6(2). In the latter
case I Lj I = 28 and Xt acts on E; doubly transitively. By observing that
F(Y 322 ,di) does not contain cliques of size 28, or otherwise one concludes
that Xt = Sp(>(2) x 2 and the suborbit diagram of Z, with respect to the
action of Xt is the following:

Comparing the above diagram with the diagram of F(Y 322 ,di), we
immediately deduce that the centres of X2, X 3 and Y322 coincide, and in
terms of the above paragraph / i 2 3 = / i 3 2 .

Lemma 8.4.4 If q,r > 2 then the element / i 2 3 = / i 3 2 is in the centre of

344 Y-groups
Proof. A Coxeter generator of Y322 commutes with fuj since the latter
element generates the centre of 7322. On the other hand di and higher
terms clearly commute with /132 and the result follows. •

Permuting the indices p, q, r we obtain obvious analogues of (8.4.4)

(compare the centres of 7-groups in the table).
By the diagram of F( 7322,^2) given below and the list of maximal
subgroups in Clj(3) [CCNPW], we have the following.

Lemma 8.4.5 The graph F( 7322,^2) is the unique orbital graph of valency
288 o/Q7(3) acting on the cosets ofSp^(2) and every subgroup in Q7P) of
index 3159 is isomorphic to Spe(2). •

216

The last group to be considered in this section is 7422. Let W be an

8-dimensional GF(3)-space with a non-singular quadratic form of plus
type. The automorphism group of this form is 2 • Qg~(3) : 2 2 [CCNPW].
Let Z be a subgroup of index 2 in the automorphism group which
contains a subgroup H = 2 x Q7(3) trivially intersecting the centre.
Then Z ~ 2 Qg~(3).22 in the atlas notation. Let O be the orbit of
Z on the set of non-isotropic vectors in W such that H stabilizes a
vector from O and let S be a graph on O in which two vectors are
adjacent if their inner product is 1. Then 3 has the following suborbit
diagram.
 8.5 Fischer groups as Y-groups 345
243 243
1+240

Performing some easy calculations in the orthogonal module W, we check

that a triangle of S is contained in 72 complete subgraphs on 4 vertices.
In view of the suborbit diagram of F( Y322, c{) this shows that S is weakly
locally r(Y322,ci). It is straightforward to check the conditions in (8.3.3)
and to conclude that Z is a strong Y422-group. Finally the conditions in
(1.14.1) hold, which show the isomorphism between Z and Y422.

8.5 Fischer groups as Y-groups

In this section we identify Y332, Y432 and Y442 with the groups 2 • M(22),
M(23) and 3 • M(24), respectively, and also discuss the group Y333.
In this and next sections by A(M(23)) and A(3 • M(24)) we denote,
respectively the transposition graph of the Fischer group M(23) and
the triple cover of the transposition graph of the Fischer group M(24).
These graphs were introduced in Section 5.8 under the names II5 and
116, respectively. The suborbit diagrams of A(M(23) and A(3 • M(24) with
respect to the actions of M(23) and 3 • M(24) can also be found in
Section 5.8.

The following result was proved in [Ron81a].

Proposition 8.5.1 The graph A(M(23)) and the graph A(3 • M(24)) are
triangulable. D

Consider the group Y332. By (8.4.4) (/213) is central in both Y332 and
^332 [d\\ and hence it is in the kernel of the action of Y332 on F(7332,^1).
Consider the action of Z := 2 • M(22) (the non-split extension) on the
cosets of a subgroup isomorphic to £27(3). One of the orbital graphs (we
denote it by S) with respect to this action has the following suborbit
diagram
346 Y-groups
288

In view of the above diagram and (8.4.5) we conclude that S is weakly

locally F(7322,^2). Now it is easy to check the conditions in (8.3.3) and
to conclude that Z is a strong 7332-group.
One can see from the suborbit diagram of A(M(23)) that the stabilizer
in M(23) of a vertex a e A = A(M(23)) is isomorphic to Z and its
actions on the vertex-set of S and on A2(a) are similar. Furthermore
the subgraph in A induced by A2(a) is also an orbital of valency 3159.
Noticing that the stabilizer in Z of a triangle in S is isomorphic to Sym-j
while the stabilizer in M(23) of a triangle in A(M(23)) is of the form
[211]. 1/4(2) (in particular it does not involve Symj), we have the following

Lemma 8.5.2 The subgraph in A induced by A2(a) and the graph S with
the above suborbit diagram are two different orbitals of valency 3159 of
the action of 2 • M(22) on the cosets ofQ-j(3). •

Using (8.5.2) and calculating in the graph A(M(23)) it is not difficult

to check that the conditions in (1.14.1) are satisfied for S, which gives
the isomorphism Y332/{f213) = 2 • M(22). Finally (8.1.1) completes the
identification of 7332.
Noticing that the Coxeter diagram of 733i is affine of type £7, it is
not difficult to identify 7332|_C3_|/(/i23,/2i3) with a maximal subgroup in
M(22) of the form 2 6 : Sp6(2). The subdegrees of M(22) acting on the
cosets of 2 6 : Sp6(2), as calculated in [ILLSS], are the following:

1, 135, 1260, 2304, 8640, 10080, 45 360, 143 360, 241920 2 .

Since 7332 ^3^3J has index 2304 in 2 6 : Sp^(2) the above subdegrees
show that F( 7332,03) is correct and that it is isomorphic to the unique
orbital graph of valency 2304 of the action of M(22) on the cosets of
2 6 : SP6(2).
 8.5 Fischer groups as Y-groups 347
Let us turn to Y432. Put Z = M(23) and let S be the complement of
A(M(23)). Then the vertex stabilizer Z(a) of the action of Z on S is
isomorphic to 2 • M(22) which is the index 2 commutator subgroup of
Y332. The suborbit diagram of S is
1+3159+21840
x 28160 1 Z' \ 3159 25 344
1 1 28160

2.M(22) Q7(3) 22.U6(2)

and by (8.5.2) S is weakly locally F( 7332,^1). Checking the conditions in

(8.3.3), we conclude that Z is a Y432-group.
The natural homomorphism q> : 7432 := Y^/Cfcn) -> % induces a
covering
xp : r(Y432,ei) -> 3
of graphs with respect to which the Y -triangles are contractible. Our
nearest goal is to show that xp induces a covering of A = A(M(23)).
Let P = (Si,82,53) be a 2-path in r(Y432,ei), P = (si,s2,s3) be its
image in S and suppose that s\ and S3 are adjacent in A. Since xp is a
covering of graphs the stabilizer of P in Y432 maps isomorphically onto
the stabilizer H\ of P in Z. On the other hand the suborbit diagram of S
and (8.4.5) show that H\ = Sp(>(2). Without loss of generality we assume
thatJSi,32} = {Ym\ei\9YAn\ei\ei}9 so that {sus2} = {Z|eiJ,Z|.eiM}.
Let 2 be the set of images of Si under 7432^2] and E be the set of images
of si under Z [^J- Comparing the isomorphism Y422 = 2 • £1^(3) : 2 and
the list of maximal subgroups in M(23) or otherwise one concludes that
Z[d2\ = Qf(3) : 2 and hence |Z| = 1080. Thus the restriction of \p to
S is either a bijection, or has fibers of size 2. In either of the cases we
can assume without loss of generality that P c I , P c I and by the
above sentence the stabilizer of {51,53} in 7432^2] has index at most
2 in the stabilizer if2 of {^1,53} in ZL^J- From the suborbit diagram
of r(Y422,ei) we see that H2 = 2 x 2 • £74(3) : 2. Thus the stabilizer of
{si,S3} in 7432 contains a subgroup isomorphic to Spt(2) and a subgroup
isomorphic to 2 • 1/4(3). On the other hand the stabilizer in M(23) of an
edge in A, isomorphic to 22 • Ue(2) (a non-split extension), is generated
by any two of its subgroups isomorphic to Sp6(2) and 2 • 1/4(3). Hence
the stabilizer of {Si,S3} in Y432 maps isomorphically onto the stabilizer
of {si,S3} in Z which shows that xp induces a covering
348 Y-groups
of graphs. Here the vertex set of A is that of F( Y ^ e i ) and the edges
are the images of {si,S3} under Y432. Since % is a covering of graphs,
the restriction of xp to S must be a bijection and hence x induces an
isomorphism of the subgraph in A induced by £ onto the subgraph in
A induced by Z. The latter graph is the antipodal folding of F( 7422,^1)
and it is of rank 3. Since this subgraph obviously contains triangles and
M(23) acts transitively on the set of triangles in A, we conclude that all
the triangles in A are contractible with respect to x> But then x and xp must
be isomorphisms since by (8.5.1) the triangles generate the fundamental
group of A. Application of (8.1.1) completes the identification of Y432.
Analysing the list of maximal subgroups of M(23) [CCNPW] it is
not difficult to identify Y432LC3J with 2 x Sp%(2). Consider the action
of Y432 on r(Y432,C3). By (8.4.4) /231 is in the kernel of the action. The
stabilizer in Y432 of the edge e := {Y432LC3J, Y432LC3 JC3} obviously contains
?432L^3?c3j — Symg. On the other hand the subdegrees of the action of
M(23) on the cosets of Sp8(2) were calculated in [ILLSS]. The only
non-trivial subdegree which divides the index 130560 of Symg in Sp%(2)
is 13056 and the corresponding 2-point stabilizer is isomorphic to Sym\$.
Thus r(Y432,C3) is not correct but in fact there is a way to "correct"
the situation by adjoining an additional generator. Let H = Symio be
the stabilizer in Y432 of the edge e. Then the subdiagram of the Coxeter
diagram of Y432 which is the Coxeter diagram of ?432L^3>c3j c a n be
extended to that of H by adjoining a node adjacent to e\ or to d^. Since
/213 is in the centre of 7432 the extra node (denote it by /1) must be
adjacent to e\. Since H has no outer automorphisms, f\ commutes with
C3. We claim that f\ also commutes with b^. This claim can be checked
by noticing that every edge of r(Y432,c3) is contained in 210 = [Sym\o :
Syme x Sym*] triangles [ILLSS] and that b^ is involved in the expression
for the element /213 and the latter commutes with f\. Thus Y432 is a
Y532-group. Furthermore, /1 commutes with Y^L^iJ = 2 x 2 - M(22),
the latter subgroup is self-centralized in Y432 and by (8.4.4) its centre is
(/123, /213). Since e\ has product of order 3 with both /1 and /123 we
conclude that the latter two elements are equal.

Lemma 8.5.3 Y532 = Y432.

Proof. Suppose that Y532L/1J is a proper subgroup in Y532 and consider

the action of Y532 := Y532/(/23i) on r(Y 5 3 2 ,/i). Then the structure of Y432,
Y332 and Y232 show that the elementwise stabilizers of a vertex, an edge
and a triangle in Y543 are isomorphic to M(23), 2 • M(22) and
 8.5 Fischer groups as Y-groups 349
respectively. Hence F(Y532,/i) is weakly locally the complement S of
A(M(23)) with the suborbit diagram given in this section. If the diameter
of r(Y532,/i) is 1 then the action of Y532 on the vertex set of the graph
is doubly transitive and it is easy to show that this is not possible. On
the other hand from the suborbit diagram of S we see that the number
of vertices at distance 2 from a given vertex is at most
31671 -3510/25 345 < 4500.
Comparing this estimate with the indices of maximal subgroups in M(23)
we conclude that there is only one vertex at distance two. Since the action
of M(23) on S is primitive, this gives a contradiction. •
By the above lemma and the paragraph before it, we obtain the
following

Corollary 8.5.4
Ifq>3andr>2 then Y5qr = Y4qr and /1 = / 123 = /132. °
Consider Y632 with the obvious meaning. Then the generator corre-
sponding to the terminal node of the left arm of the Coxeter diagram
commutes with /123 and the order of its product with /1 divides 3. By
Corollary 8.5.4 this gives the following.

Corollary 8.5.5
Y632 and higher Y-groups collapse to a group of order 2. •

In order to identify Y442 consider the action of Z := 3 • M(24) on

A := A(3 • M(24)). If a G A then Z(a) S 2 x Af (23) S Y432 and a can be
identified with the unique non-trivial element in the centre of Z(a) (this
element is an involution which maps onto a 3-transposition in M(24)).
In these terms if /} € Aj(a) then the product a/? is of order 2, 3, 6 and 3
for i = 1, 2, 3 and 4, respectively.
Let p e A2(a). Then Z(a) n Z(fi) s Q+(3) : 2 s Y432[d2\- Since
the commutator subgroup Z' of Z acts distance-transitively on A, we
conclude that Z(a) nZ(/?) is not contained in the direct factor M(23) of
Z(a). Since all subgroups in Af (23) isomorphic to Qg~(3) : 2 are conjugate,
this specifies the action of Z(a) on A2(a) and in particular shows that
this action is similar to the action of Y432 on the vertex set of F( 7432,^2)-
Since Y432Lc2^2j — ^jf(2) : 2 is a maximal subgroup of index 28431 in
Y432M2J = £*8~(3) : 2, we conclude that 1^7432,^2) is correct of valency
28431. The suborbit diagram of A shows that the subgraph in A induced
350 Y-groups
by A2(a) is also an orbital of valency 28431. We claim that they are
different orbitals. Indeed, by (8.5.4) the stabilizer in Y432 = Y532 of a
triangle in F( 7*32,^2) contains Y5321.^2, C2, <fej — Sym^ while the stabilizer
in Z of a triangle in A is of the form 23.U^(2) and does not involve Symg.
Hence the claim follows.
Notice that the set
0(jS) = { / I y e A2(a) n A(jS)} = A2(a) n A(jSa)
is an orbit of length 28431 of Z(a) n Z(j8) on A2(a) containing vertices
which are at distance 2 from /? in A. It follows from [PS97] that the
action of M(23) on the cosets of QjJ~(3) : 2 has subdegree 28431 with
multiplicity one, which means that if

cp : (r(r432,<*2), ^32) -> (A 2 (a)),Z(a))

is an isomorphism of permutation groups which sends 7432 [d2\ onto /?,
then ©(/?) is the image under cp of the set of vertices adjacent to T ^ L ^ J
in F(7432,^2). Thus we have the following.

Lemma 8.5.6 Let S be a graph on the set of vertices of A = A(3 • M(24))

in which two vertices are adjacent if they are at distance 2 in A. Then S is
weakly locally F( 7432,^2)- E
Now it is easy to see that the conditions in (8.3.3) are satisfied and
hence Z = 3 • M(24) is a 7442-group (by (8.5.4) it is also a 7552-group).
Our next goal is to show that the natural homomorphism
q> : Y442 —• Z

induces a covering of A. Let P = (81,32,53) be a 2-path in F(7442, [e\\\

P = (si,S2,S3) be its image in S and suppose that s\ and S3 are adjacent
in A. Since q> induces a covering of F( 7442, ei) onto S, the stabilizer of P
in 7442 is isomorphic to Qjj"(2) : 2 which is the stabilizer of P in Z . Let
£ be the set of images of T^L^iJ (considered as a vertex of F( 7442,^1))
under T ^ L ^ J and let £ be the set of images of Z[e\\ under Z\ei\.
Since T ^ L ^ J =Z[e2\ = 7 432 = 2 x Af(23), £ maps bijectively onto 2.
Furthermore 7442 [e2 J acts on S with kernel of order 2 and the induced
action is isomorphic to that of M(23) on the vertex set of A(M(23)).
Without loss of generality we assume that P cz S in which case it follows
from the suborbit digram of A(M(23)) that the stabilizer of {Si,S3} in
Y442L^2J is of the form 23.Ue(2). Since the stabilizer of {suss} in Z,
isomorphic to 2 x 2 • M(22) is generated by its subgroups isomorphic to
 8.6 The monsters 351
3
Qg"(2) : 2 and 2 .(76(2), we conclude that the stabilizer of {suh} in Y442
maps isomorphically onto the stabilizer of {si,S3} in Z which implies
that cp induces a covering
X ' A-* A
of graphs. The subgraph in A induced by E maps isomorphically onto the
subgraph in A induces by Z and both these subgraphs are isomorphic to
A(M(23)). Since the latter graph contains triangles and Z acts transitively
on the set of triangles in A, we conclude that the triangles are contractible
with respect to x- Since A is triangulable by (8.5.1), both x and w are
isomorphisms and hence Y442 = 3 • M(24). Now analysing the maximal
subgroups in M(24) or otherwise one can check that Y442 [c3\ = Qfo(2) : 2.
Finally, let us discuss the group 7 333 . By (8.4.4) (fnsjmjsn) is central
in 7333 and {f2i3,hn) is contained in 1^333 L^iJ- Consider the action of
Z :=2E6(2) on the cosets of M(22) = 1W(/213,/312>. The intersection
numbers of the centralizer algebra of this action have been calculated
in [ISa96]. These calculations show in particular that there is an orbital
graph H of valency 694 980 with edge stabilizer isomorphic to a maximal
subgroup of M(22) isomorphic to 2 6 : Spe(2). Furthermore, every edge
of S is in exactly 13 644 triangles. Since
13 644 = 1260 + 2304 + 10080
is the only decomposition of the number of triangles on an edge into the
lengths of suborbits of M(22) on the cosets of 2 6 : Sp^(2)9 given above,
we conclude that S is weakly locally F(7332^3). It is easy to check the
conditions in (8.3.3). Hence Z is a Y333-group. A possible way to identify
Z with Y333/(/i23,/2i3,/3i2) would be to show that the fundamental
group of S is generated by the Y -triangles. But this seems to be far too
difficult, since the structure of S is rather complicated and there are many
classes of cycles in this graph. By this reason we refer to the original
identification of 7333 which follow from the double coset enumeration
performed by S.A. Linton ([Lin89], [Soi91]).

8.6 The monsters

In this section we identify 7433, 7443 and 7444 with 2x2- BM,2x M and
M12, respectively.
Let Z = 2 • BM be the non-split extension of the Baby Monster BM
by a centre of order 2, introduced in the first paragraph of Section 5.12
under the name B. Let Q be the graph introduced in the paragraph before
352 Y-groups
(5.12.5). Then the vertex set of Q is the conjugacy class of involutions in
Z with centralizers of the form 22-2E6(2) [CCNPW]. The group Z acts
on Q by conjugation, the centre of Z is the kernel and the induced action
is similar to that of BM on the cosets of 2-2E^(2). The suborbit diagram
of this action is given in Section 5.12. The vertex a' antipodal to a is
the product of a and the involution in the centre of Z. By (5.12.7) the
product ocp has order 2, 2, 3, 4, 6 and 4 if j? is contained in fi(a), Q(a'),
Q^(a), £l\{u), &\{a!) and O^a), respectively. Notice that by joining in Q
the antipodal vertices we obtain the graph isomorphic to the subgraph
of the second Monster graph induced by the vertices adjacent to a given
vertex.

The group Z(cc) = 22-2Ee(2) (the commutator subgroup of 7333) acts

on Clli^) a s i t a c t s o n the cosets of its subgroup 2M(22) (the commutator
subgroup of Y332). Hence this action is similar to the action induced by
Y333 on r(Y333,di). The graph F(7333,^1) has valency 694980 and one can
see from the diagram of Q that this is also the valency of the subgraph
of Q induced by Q^(a). By [ISa96] the subdegree 694980 appears with
multiplicity one in the action of 2Et(2) on the cosets of M(22). This means
that for P £ f2|( a ) the subgroup Z(a) HZ(/J) has exactly two orbits of
length 694980 on Q^(a), namely O|(a)nQ(j8) and 0(jB) := Q|(a)nQ(j8a).
If y e ®(j8) then fiy is of order 3 and hence y e Q^O 8)- W e c l a i m that
r(Y333,di) is isomorphic to the graph on Q|(a) in which /? is adjacent to
©(/?). In fact, the stabilizer in Y333/(/2i3,/3i2) of a triangle in F(Y333,di)
contains Sym& while if H is the stabilizer in BM of a triangle in Q then
H/C>2(H) = 1/4(2) which does not involve Sym% and the claim follows.
Let S be a graph on the set of vertices of Q in which a and /?
are adjacent if /? e Q ^ ) - ^Y the above paragraph S is weakly locally
F(Y333,di) and checking the remaining conditions in (8.3.3) we conclude
that Z = 2 • BM is a 7433-group. Notice that we have realized the
Coxeter generators of Y433 by involutions inside 2 • BM. By (8.1.1) the
direct product 2 x 2 - BM is also a 1433-group.
We claim that the homomorphism

q> : ?433 := W ( / 2 1 3 J312} - BM

induces a covering x of Q. Let P = (suh,h) be a 2-path in r(Y433,ei)

which maps onto a 2-path P = (^1,52,53) in H such that s\ and S3 are
adjacent in Q. Since cp induces a covering of F( 1433,^1) onto S, the
stabilizer of P in Y433 maps isomorphically onto the stabilizer of P in
BM and the latter is the edge stabilizer of the subgraph in Q induced by
 8.6 The monsters 353
6
a), isomorphic to 2 : Sp6(2). We assume that Si = Y433UiJ> so that
si = BM\e\\. Let E be the set of images of s\ under Y433L^2j and let S
be the set of images of s\ under BM[d2\. Since
= Y432/(f213) s M(23),

E maps bijectively onto Z. Furthermore the action of Y433 L^2J on £ is

similar to that of M(23) on A(M(23)). Assuming without loss of generality
that P c £ we conclude from the suborbit diagram of A(M(23)) that
the stabilizer of {suh} in Y433IAI is of the form 2 2 • l/6(2). Finally the
stabilizer of {si,S3} in BM, isomorphic to 22+20.Us(2) is generated by
its subgroups 2 6 : Sps(2) and 2 2 • U^(2) which implies that (/> induces
a covering # : Q —• Q of graphs. The subgraph in Q induced by X
maps isomorphically onto the subgraph in Q induced by £ (both these
subgraphs are isomorphic to A(M(23))). Thus the triangles in Q are
contractible with respect to x- It has been proved in [Iv92b] and [Iv94]
that Q is triangulable. Hence both x and cp are isomorphisms and Y433 =
BM.
Now let Z be the Monster group M and T be the second Monster
graph as in Section 5.12. Then T is a graph on the conjugacy class
of 2a- (Baby Monster) involutions in the Monster with two involutions
being adjacent if their product is again a 2a-involution. If a € T then
Z(a) = 2 • BM (the commutator subgroup of Y433) and the subgraph in T
induced by F(a) is the graph Q as above together with a matching which
joins pairs of antipodal vertices. This shows that Z has two orbits on the
triangles in F. Every edge is contained in a unique triangle from one of
the orbits (we call them short triangles) and in 3 968 055 triangles from
another orbit (we call then long triangles). The suborbit diagram of F
has been calculated in [Nor85] and we will use the following result from
that paper.

Lemma 8.6.1 Let a e T and p e F 3a (a). Then Z(a)nZ(jS) acts transitively

on the set of vertices y € F3a (a) Pi F2a(j8) with stabilizer isomorphic to
•
Let S be a graph on the vertex set of F in which a and j8 are adjacent
if P e F 3a (a). By (5.12.10 (iii)) the isomorphism

G : F433 := W ( / 2 i 3 > -> Z(a)

induces an isomorphism of the permutation group (F(Y433,d3), Y433) onto
the permutation group (H(a),Z(a)). We denote the latter isomorphism
354 Y-groups
by the same letter a. We claim that whenever u and v are adjacent
vertices in 1^7433,^3), a(u) and a(v) are adjacent vertices in S. First the
stabilizer of {u,v} in Y433 is isomorphic to Sp%(2) and by (5.12.10) we have
a(u) e rc (cr(i;)) where C is 2a, 3a or 4b. Let us first exclude the latter
possibility. Without loss of generality we assume that u = Y433 [^3] and
v = ^433L^3j^3- Then the isomorphism o sends the Coxeter generators of
Y433 M3J i n t o Z(<J(U)) and by (5.12.10 (iii)) the images of the generators are
contained in r(a)nF(cr(w)). The image under a of d$ maps a(u) onto a(v)
and commutes with the images of the Coxeter generators of Y433LC3, d$\
which shows that G(U) and a(v) have at least 9 common neighbours in F
and by (5.12.10 (v)) o(v) $ T4b(a(u)). Suppose that a(v) e T2a(a(u)). Then
by (8.6.1 (iii)) r(7 4 33,d 3 ) maps isomorphically onto the subgraph in F
induced by S(a). We know from the previous section that the stabilizer
in Y433 of a triangle in 1^(7433,^3) is isomorphic to Sym\o- Clearly this
triangle is a long triangle in F, but if H is the stabilizer in Z of a long
triangle then H/OiiH) = U^(2) and the latter group does not involve
Symio. This contradiction shows that a(u) and o(v) are adjacent in S.
Hence S is weakly locally F(7433,d3). It is easy to check the conditions
in (8.3.3) and to conclude that Z = M is a 7443-group.

We claim that the homomorphism

q> : 7443 := Ym/{hn) -> M

induces a covering % : F —> F of graphs with respect to which all long

triangles are contractible. Consider a 2-path P = (Si,S2,S3) in F(7443,^i)
which maps onto a 2-path P = (si,S2,S3) in S such that s\ and S3 are
adjacent in F. Then by (8.6.1) and since cp induces a covering of F(7443, e\)
onto H, the stabilizer of P in Y443 is isomorphic to Sp%{2). Assume without
loss of generality that si = 7443 [eij and that S3 is contained in the orbit
of s\ under 7443 \ei\ = 2 • BM. Then the suborbit diagram of Q given in
Section 5.12 shows that the stabilizer of {51,53} in 74431/2J i s isomorphic
to 23+20.(76(2). Finally since the stabilizer of {51,53} in Z, isomorphic to
22-2Ee(2) is generated by its subgroups isomorphic to 23^20.Ue(2) and
Sps(2) we conclude that cp indeed induces a covering % : F -^ F of
graphs. The subgraph in F induced by the images of si under 7443 \ei\ is
isomorphic either to Q or to the subgraph in F induced by F(a) and in
any case the long triangles are contractible with respect to x- By (5.14.2)
F is triangulable. In [Iv94] using this result it was shown that the long
triangle already generate the fundamental group of S. Hence 7443 = M
and in view of (8.1.1) we have 7443 = 2 x M .
 8.6 The monsters 355

Let us turn to Y444. As above, let M be the Monster group, let D be

the direct product of two copies of M:
D = {(g, h) I g,fc G M}; (gl,fcO • (g2,fc2)= (gig2, h2h)
and define an action of D on M by (g, / I ) : W H gm/i for every m e M
and (g,/i) G D. In this way we realize D as the group generated by the
left and right regular representations of M. Let T be the permutation on
M acting by T :WIH> m"1. Then T can be considered as a permutation of
D via (g,/i)T = (/i" 1 ^" 1 ), in particular T normalizes D and permutes its
direct factors. Let Z = (D,T) be the Bimonster. If Z ( l ) is the stabilizer
in Z of the identity element of M then Z ( l ) = (T) X Mr where Mr =
{(g,h) G D | g = h~1} and hence every orbit of Z ( l ) on M is of the form
CKJC where C is a conjugacy class of M and C = {g~l | g G C}.
Let S be a graph on M in which m\ and m2 are adjacent if and only
if mim^1 is an element of type 3a in M. Since the class of 3a-elements is
closed under taking inverses, Z acts on S vertex- and edge-transitively.
If t is of type 3a then the stabilizer in Z of the triple T = {1, t, r1} is
<T> x NM,((t)) = 2 x 3 - M (24),
which shows that the elementwise stabilizer in Z of the edge {l,t}
is isomorphic to 3 • M(24) while the setwise stabilizer is of the form
3 • M(24) x 2.
Let A = Alt4 be a subgroup in M with the normalizer of the form
(Alt4 x Qfo(2)) : 2 [Nor98]. Then all the elements of order 3 in A are of
type 3a and hence by choosing t\ and ti to be suitable such elements,
we obtain a triangle T\ = {1, ti, ^2} in S whose elementwise stabilizer is
isomorphic to Of0(2) : 2 (notice that T\ is fixed by the product of an
element in the normalizer of A which inverts both t\ and £2 and the
element T). Hence S is weakly locally T(Y^^d-i) and by checking the
conditions in (8.3.3) we conclude that Z is a Y^-group.
Let xp : F(Y444,ei) —> S be the covering of graphs induced by the
homomorphism of Y444 -> Z and let ©be a graph on M in which m\
and m2 are adjacent if m\rrql is an element of type 2b in M. We are going
to show that xp induces a covering % of 0 and that certain triangles in 0
are contractible with respect to %. Notice that the elementwise stabilizer
in Z of an edge of 0 is isomorphic to 2 x 2^_+24.Coi. Consider in M the
stabilizer P of an element of type 2 in ^(M), so that

p ^ 2 2+11+22 .(Sym 3 x Mat1A)

and let s be an element of order 3 in O2,3(P). Then CP(s) 2* 2 n .(3 xMat 2 4 ),

356 Y-groups

s is of type 3a and the centre S of O2(P) is 2fr-pure of order 4. Hence

for si e sS\ {s} we obtain a triple T2 = {l,s,si} such that {s, l,si} is a
2-arc in S and {s,si} is an edge in 0 . The elementwise stabilizer of T2 in
Z contains Mat24> Since y> is a covering, there is a pair of vertices {s,Si}
in r(Y444,ei) which maps onto an edge of 0 and whose stabilizer in Y444
contains Mat2^.
Put H = Z[ei,\ and let E be the set of images under H of the identity
element of M. Then by the previous subsection H = Y444 [e^J = 2 x M.
Furthermore if £ is the vertex-set of a connected component of the
subgraph in F( Y444, e\) induced by \p~x(L) then the subgraph in F( Y444, e?)
induced by Z and the subgraph in S induced by £ are isomorphic
to F(7443,^1) and H acts on Z with kernel of order 2. Now without
loss of generality we can assume (in terms of the previous paragraph)
that s,si € Z and s,s\ e Z. Then the setwise stabilizer of {s,s\} in H
(isomorphic to the stabilizer of {s, s\} in Y444) is of the form 2 x 21+11.Co2.
Since the stabilizer in Z of an edge in 0 is generated by any two of its
subgroups isomorphic to Mat24 and 2 x 22+22.Co2, we conclude that the
stabilizer in Y444 of {s,si} maps bijectively onto the stabilizer in Z of
{s,s\} and hence \p induced a covering / : 0 —> 0 of graphs. Notice that
the vertex-set of 0 is that of F( 7444,^3) and the edges are the images
under Y444 of the pair {S,Si}.
IHs clear that the covering x induces an isomorphism of the subgraph
in 0 induced by X onto the subgraph in 0 induced by I . This means
that every triangle in Z is contractible with respect to %. Such a triangle
is formed for instance by the non-identity elements from C. Thus we
conclude that whenever z\,z2,zs are elements of type 2b in M such that
Z1Z2Z3 = 1 and z\ e O2(CM(ZJ)) for 1 < i,j < 3, then the triangle in 0
induced by {l,zi,z2} is contractible with respect to x>
In order to show that x is a n isomorphism we apply the result from
[IPS96] that M is the universal representation group of its tilde geometry
^(M). A direct factor M of D acts regularly on 0 and hence 0 can be
considered as a Cayley graph of M so that the corresponding generators
are the 2fc-involutions. Let
<5 : © - >©
the covering of 0 with respect to the subgroup in its fundamental group
generated by the images under M of the triangles {l,zuz2} such that
zi,Z2,Z3 := Z1Z2 are 2fc-involutions and z, € O2(CM(ZJ)) for 1 < i,j < 3.
Let M be the group of all liftings of elements of M to automorphisms
of 0 . It is clear that the subgroup of deck transformations acts regularly
 8.6 The monsters 357
on each fiber and hence M acts regularly on 0 . This means that ©is
a Cayley graph of M with respect to generators t(z\ one for every 2b-
involution z in M. Since ©is undirected the generators are involutions
and since the triangle {l,zi,Z2} as above is contractible with respect to 6,
the corresponding generators satisfy the equality t(z\)t(z2)t(z?) = 1. The
following result was proved in [IPS96].

Lemma 8.6.2 Let M be a group generated by involutions t(z), one for every
Ib-involution z in the Monster M such that t(zi)t(z2)t(z$) = 1 whenever z\,
Z2, Z3 are 2b-involutions in M such that z\ G 02(CM(ZJ)) for 1 < i9j < 3
and ziz 2 z 3 = 1. Then 'M = M. •
By (8.6.2) and the paragraph before it 5 is an isomorphism. Hence x
is an isomorphism as well and Y444 = M\2.
 9
Locally projective graphs

In this chapter we study locally projective graphs. Let F be a graph

and G be a vertex-transitive automorphism group of F. Then F is said
to be a locally projective graph with respect to G if for every x e F
the subconstituent G(x)r(x) is a projective linear group in its natural
permutation representation. Incidence graphs of certain truncations of
classical geometries are locally projective graphs with respect to their
full automorphism groups. These examples can be characterized in the
class of all locally projective graphs by the property that their girth is a
small even number. We present a proof of this characterization based on
the classification of Tits geometries and observe how a class of sporadic
Petersen geometries naturally appear in this context via locally projec-
tive graphs of girth 5. In Section 9.1 we review some basic results on
2-arc-transitive actions of groups on graphs. In Section 9.2 we discuss
examples of locally projective graphs coming from classical geometries.
Locally projective lines and their characterizations are discussed in Sec-
tion 9.3. In Section 9.4 we analyse the possibilities for the action of the
vertex stabilizer G(x) on the set of vertices at distance 2 from x. These
possibilities determine the main types of locally projective graphs. In a
locally projective graph there are virtual projective space structures de-
fined on neighbourhoods of vertices. These virtual structures lead to the
notion of geometrical subgraphs introduced in Section 9.5. Analysis of
geometrical subgraphs enables us to specify further the structure of ver-
tex stabilizers in Section 9.7. In Section 9.8 we show that if F contains a
complete family of geometrical subgraphs then aflag-transitivegeometry
with a nice diagram is associated with the graph. In case F does not con-
tain a complete family of geometrical subgraphs a procedure described
in Section 9.6 enables us to associate with F a locally projective graph
of smaller valency with a complete family of geometrical subgraphs and

358
 9.1 Groups acting on graphs 359
the same abstract group of automorphisms. In the remaining sections of
the chapter we consider locally projective graphs of small girth g. In the
case g = 4 we obtain parabolic geometries of orthogonal groups, in the
case g = 6 the projective geometries and in the case g = 5 the Petersen
geometries.

9.1 Groups acting on graphs

In this chapter we consider pairs (F, G) satisfying the following hypothesis.
Hypothesis LP. F is a graph and G is a 2-arc-transitive automorphism
group ofT. There are an integer n > 2, a prime power q = pm and a group
H, satisfying
SL(V) <H< TL(V\
where V is an n-dimensional GF(q)-space, such that for every x € F the
action of G(x) on T(x) is similar to the action of H on the set of 1-
dimensional subspaces of V. In other terms the subconstituent G(x)r(x) is
a projective linear group of V in its natural permutation representation, in
particular
Ln(q) < G(x)r(x) < PTLn(q).

If F satisfies the above hypothesis for a subgroup G in its automorphism

group, it is said to be a locally projective graph of type (n,q) (with respect
toG).
In this section we prove a few standard results concerning actions of
groups on graphs. We start with the following elementary lemma.

Lemma 9.1.1 Let T be a connected graph and e = {x,y} be an edge of

F. Let K\ and K2 be subgroups of the automorphism group ofT such that
K\ stabilizes x and acts transitively on T(x) while K2 stabilizes y and acts
transitively on T(y). Then the action ofK = (K\,K2) is edge-transitive and
it is vertex-transitive if and only if T is not bipartite.
Proof. Let Q be the orbit of K on the edge set of T which contains
e. We will prove that every edge is contained in Q by induction on the
distance from e. Every edge at distance 0 from e is either in the K\- or
in the ^2-orbit containing e and hence it is in Q. Let / = {u,v} be an
edge at distance s > 0 from e in F. Without loss of generality we assume
that (xo = x,xi,...,xs = u) is the shortest among the arcs joining a vertex
360 Locally projective graphs
from e with a vertex from / , which means in particular that xi ^ y. Let
h be an element from K\ which maps xi onto y. Then /* is at distance
5 — 1 from e and it is contained in ft by the induction hypothesis. Hence
/ is also in Q. If F is bipartite then both K\ and K2 preserve each part as
a whole and hence K cannot be vertex-transitive. Let z be a vertex of F.
If there is an arc of even length joining x and z then it is easy to see that
x and z are in the same orbit of K. Suppose that F is not bipartite. Then
it contains a cycle of odd length. Since K is edge-transitive, there is such
a cycle which contains e, say xo = x,xi = )>,..., X2H-1 = x- I*1 this case
(xi,X2,...,X2f+i) is an arc of even length joining y and x. By the above
observation this implies the vertex-transitivity of the action of K on F.D
The following elementary result is quite important.

Lemma 9.1.2 Let T be a graph and G < Aut F. Suppose that every vertex of
F has valency at least 2. Then the following two conditions are equivalent:
(i) G acts 2-arc-transitively on F ;
(ii) G is vertex-transitive and for every x e F the subconstituent G(x) r(x)
is a doubly-transitive permutation group.
Proof, (i) —• (ii) Let a, x € F. Since the valencies of both a and x are
at least 2, there are 2-arcs p = (b, a, c) and q = (y, x, z). Let g e G be such
that pg = q. Then ag — x and we have vertex-transitivity. Let {y\,z\) be
an ordered pair of vertices in F(x) and q\ = ()>i,x,zi). Then an element
of G which maps q\ onto q stabilizes x and maps (yuz\) onto (y,z). This
implies the double transitivity of G(x) r(x) .
(ii) —• (i) Let p = (b,a,c) and q = (y,x9z) be 2-arcs in F. By the
vertex-transitivity there is an element g € G such that ag = x. Then
(bg,cg) is a pair of distinct vertices from F(x) = T(ag). Since G(x) r(x) is
doubly transitive, there is h G G(x) which maps (bg,cg) onto (y,z). Then
psh _ q a n ( j t ^ action of G is 2-arc-transitive. •

The next result gives a necessary condition for (s + l)-arc-transitivity

of an action which is known to be s-arc-transitive.

Lemma 9.1.3 Let T be a graph, G be an automorphism group of F which

acts s-arc-transitively for s > 0. Suppose that every vertex ofT has valency
at least 2. Then the following two conditions are equivalent:
(i) for an s-arc ps = (xo,xi,...,x s ) its elementwise stabilizer G(ps) in G
acts transitively on F(x s ) \ {x s _i};
(ii) G acts (s + l)-arc-transitively on F.
 9.1 Groups acting on graphs 361
Proof, (i) —> (ii) Let xs+i be a vertex in T(xs) \ {x s _i}, let ps+i =
(x o ,...,x s ,x s+ i) and let q = (yo,yu-,ys+i) be an arbitrary (s + l)-arc
in F. Since G is s-arc-transitive, there exists g € G such that yf — x,-
for 0 < i < s. Since G(ps) is transitive on F(x s ) \ {x s _i}, there exists
h € G(ps) which maps y*+1 onto x s +i. Then gh maps q onto ps+i and
(s + l)-arc-transitivity follows.
(ii) —• (i) Since the valency of every vertex in F is at least 2, every s-arc
is contained in an (s + l)-arc and the s-arc-transitivity is implied by the
(s+l)-arc-transitivity. Let x£+1 e F(x s )\{x s _i} andp^ + 1 = (x o ,...,x s ,x^ +1 ).
Then an element in G which maps pfs+1 onto ps+\ is contained in G(ps) and
maps x^+1 onto xs+\. Hence G(ps) acts transitively on T(xs) \ {x s _i}. •

Lemma 9.1.4 Let G act vertex-transitively on a connected graph T and

suppose that for an integer i > 0 and a vertex x € T the subgroup Gj(x)
acts trivially on r,+i(x), that is G/(x) = G I+ i(x). Then G,(x) = 1.

Proof. Since G acts vertex-transitively on F, the hypothesis implies

that Gt(z) = G,-+i(z) for every z eT. Let y be an arbitrary vertex from
F(x). Since Gj(x) = Gj+i(x), G/(x) fixes every vertex at distance at most
i from y which means that G,(x) < Gi(y). By the connectivity of F we
obtain that Gt(x) < Gi(z) for every z G F, which means that G*(x) = 1. •

Lemma 9.1.5 Let G act 1-arc-transitively on a connected graph F and

{x,y} e E(T). Suppose that G(x) is finite. Then
(i) if G i ( x ) r ^ is a p-group for a prime number p, then G\(x) is a
p-group,
(ii) every composition factor of G(x) is isomorphic either to a compo-
sition factor of G(x)/Gi(x) or to a composition factor of G\(x)/
Gi(x,y).

Proof. By (9.1.4) if Gt = G I+i then G,(x) = 1. Hence for some n the

following is a normal series of G(x):

In order to prove (i) it is sufficient to show that G,-(x)/G,-+i(x) is a p-group

for 1 < i < n — 1. By the definition G,-(x)/G,-+i(x) is the action induced by
G,(x) on F,+i(x). For every u G F l+ i(x) there is a 2-arc (w,v,u) such that
w € F;_i(x) and v e F/(x). Then G;(x) < Gi(vv) and by the hypothesis
G,(x)r(M) < Gi(w)r(M) is a p-group. Hence G,(x) r ' +l(x) is a p-group as well
and (i) follows. The assertion (ii) can be proved in a similar way. •
362 Locally projective graphs
Lemma 9.1.6 Let G act distance-transitively on F and suppose that the
action ofG is strictly s-arc-transitive. Then the girth ofT is at most 2(s+l).
Proof. Suppose that the girth of F is greater than 2(s + 1). Then for
every i < s + 1 there is a natural bijection between the pairs of vertices at
distance i and i-arcs in F. Hence in this case distance-transitivity would
imply the (s + l)-arc-transitivity. •
The following lemma shows that regular generalized polygons appear
as extremal cases in s-arc-transitive actions along with the Moore graphs.
Recall that a Moore graph is a regular graph of valency k > 3 and girth
Id + 1 where d is the diameter of the graph.

Lemma 9.1.7 Let G act s-arc-transitively on a graph F of girth g and

suppose that every vertex of F has valency at least 3. Then
(i)g>2s-2,
(ii) if g = 2s — 2 then F is a generalized (s — l)-gon and the action of
G onT is distance-transitive,
(iii) if g = 2s — 1 then F is a Moore graph of diameter s — 1 and the
action of G onY is distance-transitive.
Proof. Let (xo,xi,...,xg_i,xg = *o) be a shortest cycle in F. For
0 < t < g let G(pt) denote the elementwise stabilizer in G of the arc
pt = (xo,xu...,xt).
(i) Let t be the least integer greater than or equal to (g + l)/2. Since
the girth of F is g there is a unique arc of length g — t joining xt and
xg. Hence G(pt) stabilizes xt+\. By (9.1.3) this means that G cannot act
(t + l)-arc-transitively on F.
(ii) Let t = g/2 = s — 1. By (9.1.3) G(pt) acts transitively on the vertices
in T(xt) \ {xt-i} and one of these vertices, namely xt+u is in Ff_i(xo).
Hence all the vertices adjacent to xt are in Ff_i(xo). This means that F
does not contain cycles of odd length {i.e. F is bipartite) and the diameter
of F is t = 5 — 1. So the result follows.
(iii) Let t = (g — l)/2. Then G(pt) acts transitively on the vertices in
T(xt) \ {xt-i} and one of these vertices, namely x t+ i, is in Ft(xo). Hence
every vertex other than xt_i which is adjacent to xt is in Tt(xo). This
means that F is a distance-transitive Moore graph. •

9.2 Classical examples

In this section we describe a few infinite families of locally projective
graphs associated with classical geometries and present some motivations
for the general interest in locally projective graphs of small girth.
 9.2 Classical examples 363
s#n(q). Consider an (n + l)-dimensional GF(g)-vector-space V and
form a bipartite graph stf\(q) whose vertices are 1- and n-dimensional
subspaces of V with the adjacency relation defined via inclusion. Then
s^\(q) is the incidence graph of the symmetric 2-design having the 1-
dimensional subspaces of V as elements and the n-dimensional subspaces
as blocks. The parameters of this design are v = \n+1 , k = ["] ,
X = "y1 . If G is an extension of PGLn+\(q) by a contragredient au-
tomorphism then G acts 2-arc-transitively on s4\(q\ G(x)r(x) = PGLn(q)
and |Gi(x,j;)| = q. From the description of s#\{q) as the incidence graph
of a symmetric 2-design it is easy to see that it is distance-transitive of
diameter 3 with the following intersection array:

In particular the girth of s#\(q) is 4.

^in-iio)' Now let V be a (2n — l)-dimensional GF(g)-space. The
vertices of ^2«-2(^) a r e a ^ (n ~ 1)" anc^ w-dimensional subspaces of V
with the adjacency relation defined via inclusion. This graph is known
as the q-analogue of the double cover of the odd graph. The extension G
of PGL,2n-i(q) by a contragredient automorphism acts 3-arc-transitively
on ^2n-2(^) w i t h G(x)r(x) = PGLn(q) and with Gi(x) containing a
section isomorphic to SLn-\(q). The action is distance-transitive and the
intersection numbers are the following:

c a - i = c2i = [ [ ] q ; b 2 i ^ = b2i = [n~\ q\

In particular the girth of ^in-iiQ) ^s 6-

@n(q)- In this case V is a 2n-dimensional GF(^f)-space equipped with
a non-singular quadratic form / of Witt index n. The maximal totally
singular subspaces of V have dimension n and they are partitioned into
two classes in such a way that whenever two such subspaces intersect in
an (n — l)-dimensional subspace, they are from different classes. Every
(n — l)-dimensional totally singular subspace is contained in exactly two
maximal ones (from different classes). The vertices of @n{q) are the
maximal totally singular subspaces of V, with two subspaces adjacent if
their intersection is of dimension (n— 1). In view of the above this means
that the graph is bipartite. The extension G of the Lie type group Dn(q) by
a diagram automorphism (that is, the group of all linear transformations
364 Locally projective graphs
of V preserving / ) acts 2-transitively on @n(q) with G(x)r(x) = PGLn(q)
and Op(Gi(x)), as a GF(g)-module for G(x)/Op(G\(x)\ is isomorphic to
A2 Vn(q). The graph is distance-transitive of diameter n with parameters

The above three families of graphs possess a uniform description in

terms of truncations of the corresponding classical geometries. To wit,
the vertices of the graphs are the elements corresponding to the black
nodes in the respective diagrams with the adjacency relation induced by
the incidence relation in the geometry.
1 2 3 n-2 n-l n
q q q q

1 2 n-l n 2n-3 2n-2

Q ** # Q
q q q

1 2 3 n-3 n-2 n-l

o o o

The graphs $£\(q) and stf\{q) are both isomorphic to the incidence
graph of the projective plane over GF(q) (i.e. to the corresponding gen-
eralized triangle), and we will denote it by s&i(q). Its full automor-
phism group isomorphic to Aut L^(q) acts strictly 4-arc-transitively and
distance-transitively.
There are two more series of rank 2 Lie type geometries possessing
diagram automorphisms. The corresponding graphs are locally projective
lines.

The vertices of &2(q) are the totally isotropic 1- and 2-dimensional

subspaces of a 4-dimensional GF(g)-space with respect to a fixed non-
singular symplectic form. So &2(q) is the generalized quadrangle of
symplectic type. The diagram automorphism exists if and only if q is
even (that is, a power of 2). The graph J^(2m) is strictly 5-arc-transitive
and distance-transitive.

q q

The graph &2{q) is the generalized hexagon associated with the Lie type
 9.2 Classical examples 365
group Giiq). The diagram automorphism exists if and only if q is a power
of 3, and G2(3m) is strictly 7-arc-transitive and distance-transitive.
Another series of locally projective graphs is associated with the Lie
type groups of type F^q) and in the above terms can be described by
the following diagram:

q q q q

In order for ^4(4) to be vertex transitive we need a diagram auto-

morphism. Such an automorphism exists if and only if q is even (that is
a power of 2). The graph ^4(4) is not distance-transitive for any q and
its girth is 8.
The following very elegant result characterizing the graphs $#\{q) and
9n{q) was proved in [CPr82].

Theorem 9.2.1 Let (r, G) be a pair satisfying Hypothesis LP with G\(x) j=

1 and suppose that the girth ofT is 4. Then one of the following holds:
(i) T = stf\(q) and Ln+i(q).{r) < G < AutLn+i(g), where T is a dia-
gram automorphism;
(ii) T s 2n(q) and QJI(«).(T) < G < AutQ%n(q), where T is a diagram
automorphism;
(iii) F = Kmim is the complete bipartite graph, m = ["] and Ln(q) x
Ln(q) <G< Aut (Ln(q) x Ln{q)). * U
Later in this chapter we will extend the above characterization to
graphs of girth 5 and 6. In the girth 6 case this will provide us with a
characterization of the graphs ^\n_2(q). In the girth 5 case a class of
sporadic Petersen type geometries arises. Before proceeding to this let us
discuss another motivation for the interest in locally projective graphs,
especially in those of small girth.
The motivation comes from a general problem of bounding the order
of the vertex stabilizer G(x) for a group G acting 2-arc-transitively on a
graph T in terms of the subconstituent G(x)r(x). For an arbitrary doubly
transitive permutation group H it is possible to produce a 2-arc-transitive
action with G(x)r(x) ^ H and Gi(x) + 1. For this we take T to be the
complete bipartite graph Km>m (where m is the degree of H) and G to
be the wreath product H 12. However, in all these cases G\(x,y) = 1.
The situation when G\(x,y) =fc 1 turns out to be much more specific
as is shown by the following result known as the Thompson-Wielandt
theorem (see [Wei79a] for the proof of an improved version of it).
366 Locally projective graphs
Theorem 9.2.2 Suppose G acts 2-arc-transitively on F and G\(x,y) j= 1 for
{x,y} e E(T). Then G\(x,y) is a p-group for a prime number p. •
A non-trivial element from G\(x,y) is called an elation. It is easy
to see from (9.2.2) and (9.1.4) that if G acts 2-arc-transitively on F and
contains elations then the point stabilizer G(x, y) r(x) in the subconstituent
Q(xjr(x) j s p-local (i.e. it has a non-trivial normal p-subgroup). Using a
case-by-case analysis of the known doubly transitive permutation groups
with p-local point stabilizers, the following result was proved (the survey
[Wei81a]).

Theorem 9.2.3 Let G act 2-arc-transitively on F and suppose that for an

edge {x,y} ofT, Gi(x,y) ^= 1. Then F is locally projective with respect to
the action of G. •
Thus the problem of bounding the order of G(x) for a 2-arc-transitive
action was reduced to the case of projective subconstituents. A possible
way to solve this problem is to find a constant c such that Gc(x) = 1 for
every 2-arc-transitive action of G on F with a projective subconstituent.
(Although there is no a priori reason at all for such a constant to exist.)
The solution along these lines was announced in [Tro91] and some
particular cases are now published in [Tro92] and [Tro94]. The result
says that the constant c exists and c = 6 works.
The next problem of great interest is to describe all the possibilities
for the point stabilizer G(x) coming from 2-arc-transitive actions with
projective subconstituents. For this type of problem it is a standard
strategy to consider actions on trees.
The formalism is the following. Let G act 1-arc-transitively on a graph
F. Then there always exist a tree f and an automorphism group G of f
such that the actions of G on F and of G on F are "locally isomorphic".
We can define f and G in the following manner. Let x E F, {x,y} e E(T),
G(x) be the stabilizer of x in G and G[x,y] be the setwise stabilizer of
{x,y}. Because of 1-arc-transitivity, G[x,y] n G(x) = G(x,y) is of index 2
in G[x,y]. We define G to be the universal completion of the amalgam
jtf = {G(x), G[x,y]}, that is, the free amalgamated product of G(x) and
G[x,)>] over the common subgroup G(x,y). Now define the vertices of
f to be the right cosets of G(x) in G and declare two vertices adjacent
if and only if there exists a right coset of G[x, j;] in G which intersects
them both. Since F is connected, G(x) and G[x,y] generate G. This means
that there are a covering q> : f -> F of graphs and a homomorphism
xp : G —> G of groups such that the fibres of cp are the orbits on f of the
 9.3 Locally projective lines 367
kernel of xp and the kernel consists of all the elements from G which are
deck transformations with respect to cp. In other words the vertices of f
can be identified with the arcs in F originating at x with two such arcs
adjacent in f if and only if one of them can be obtained from the other
one by deleting the terminal vertex. This shows the meaning of the local
isomorphism between the actions of G on F and of G on f. In particular
G acts s-arc-transitively on F if and only if G does so on f.
In view of the above, as long as we are only concerned about local
properties of a 2-arc-transitive action such as the structure of vertex
stabilizers, the underlying graph can be assumed to be a tree. Eventually
we are interested in actions of finite groups on finite graphs, so only
locally finite actions (with finite vertex stabilizers) have to be considered.
In certain circumstances it is much more convenient to work with actions
on trees, basically because there are no cycles to cause problems. On the
other hand it is known [Ser77] that a locally finite 1-arc-transitive action
is always locally isomorphic to an action on a finite graph. Although a
construction procedure exists for producing a finite graph and an action
on it which is locally isomorphic to a given action, the resulting graph is
usually rather large. However, in practice it often happens that important
actions can be realized on surprisingly small graphs. For instance it turns
out (see (9.3.2) in the next section) that for s > 4 every action which
is strictly s-arc-transitive is locally isomorphic to such an action on a
graph of girth 2(s — 1). At the same time a graph of valency at least
3 is never s-arc-transitive if its girth is less than 2(s — 1) and every s-
arc-transitive graph of girth 2(s — 1) is a generalized (s — l)-gon (9.1.7).
The fact that every s-arc-transitive action for s > 4 is locally isomorphic
to an action on a generalized (s — l)-gon has played an important role
in the classification of such actions, particularly in specification of the
corresponding vertex stabilizers ([Wei79b], [DGS85]).
It is believed [Iv93a] that every 2-arc-transitive action with a projective
subconstituent can be realized on a graph with small girth, say up to
8. We consider this to be a motivation for the particular interest in
(9.2.1) and in its generalizations. The fact that sporadic groups and their
geometries appear in this generalization is also quite remarkable.

9.3 Locally projective lines

In this section we present a brief survey of what is known about locally
projective graphs of type (2,g), also known as locally projective lines.
Thus we consider pairs (F, G) where F is a graph, G is a 2-arc-transitive
368 Locally projective graphs
automorphism group of F and

for every vertex x € F. In view of the discussions in the previous section,

when considering local properties of the action of G on F we can always
assume that the latter is a tree (of valency q + 1).
The graphs which are locally projective lines are of particular im-
portance because of the following result proved in [Wei81b] using the
classification of finite doubly transitive permutation groups.

Theorem 9.3.1 Let G act s-arc-transitively on a graph F and s > 4. Then

s < 7 and F is locally a projective line with respect to the action of G. •

It turns out that every 4-arc-transitive action is locally isomorphic to

an action on a classical generalized polygon ([Wei79b], [DGS85]).

Theorem 9.3.2 Let G act strictly s-arc-transitively for 4 < s < 1 on a

graph F. Then s ^ 6 and the action is locally isomorphic either to an s-
arc-transitive action on the generalized (s — \)-gon s#2{q)> &i{2m), ^2(^m)
for s = 4,5,7, respectively, or to a 4-arc-transitive action of PGLi{9) on
•
Notice that a group which acts s-arc-transitively on a classical gener-
alized (s — l)-gon contains the corresponding simple group of Lie type
of rank 2 (1.6.5).
The actions on vertex-transitive classical generalized polygons were
characterized in [Wei85] in the context of distance-transitive graphs.

Theorem 9.3.3 Let G act on a graph F distance-transitively and s-arc-

transitively for s > 4. Then F is isomorphic to one of the following:
(i) the generalized polygon ^2{q), @i{2m) or ^ ( 3 m ) ;
(ii) the incidence graph of the rank 2 tilde geometry ^(3 • Sjp4(2));
(iii) a cubic distance-transitive graph on 102 vertices with the automor-
phism group isomorphic to L2(17). •

Some particular cases of the above theorem were known long be-
fore [Wei85]. Specifically it was proved in [Glea56] that every distance-
transitive generalized triangle is isomorphic to s0i{q), in [Hig64] that
every distance-transitive generalized quadrangle is isomorphic to J^(2m)
and in [Yan76] that every distance-transitive generalized hexagon is iso-
morphic to
 9.3 Locally projective lines 369
By (9.3.2) all local properties of an s-arc-transitive action for s > 4 can
be checked for suitable classical generalized polygon. For s = 4 and 5 all
calculations are elementary and for s = 7 one can use a model of &2{q)
given in [Kan86].
We will make use of the following result of this type. Let V2{q) denote
the natural 2-dimensional GF(g)-module for the group SL2(q) and let
V\(q) denote the trivial 1-dimensional GF(g)-module for this group.

Lemma 9.3.4 Let G act on a graph T strictly s-arc-transitively for s > 4,

so that F is locally a projective line over GF(q). Let p be the characteristic
of the field and let Wt = O p (G,(x))/G I+ i(x)/or i > 1. Then G(x)/Op(G(x))
contains a characteristic subgroup K isomorphic to SL2(q). Moreover, if
Wt is considered as a GF(p)K~module then the following hold:

(i) ifs = 4 then Wx ^ V2(q) and W2 = 1;

(ii) ifs = 5 then Wi S V2(q), W2 = Vi(q) and W3 = l;
(iii) ifs = 7 then W{^W2 = V2(q), W3 ^ Vi(q) and W4 = 1.
In particular \G\(x,y)\ = qs~3 and if y,z G T(x) then G\(x,y) induces on
T(z) \ {x} a regular action whose kernel coincides with G2(x). •

For the sake of completeness we present the following result whose

proof can be achieved by completely elementary methods.

Lemma 9.3.5 Let G act strictly 3-arc-transitively on a graph which is locally

a projective line. Then G\(x,y) = 1. •

The following well-known characterization of distance-transitive Moore

graphs (Section 6.7 in [BCN89]) is in a certain sense analogous to (9.3.3).

Lemma 9.3.6 Let T be a Moore graph of valency k > 3 and diameter

d. Then d = 2 and k e {3,7,57}. / / in addition G = AutT acts on T
distance-transitively then one of the following holds:
(i) k = 3, F is the Petersen graph, G = Syms acts 3-arc-transitively on
T, G(x) r ^ ss Sym3 s L2(2) and Gi(x) s 2;
(ii) k = l,Y is the Hoffman-Singleton graph, G ^ PEL 3 (5) acts 3-arc-
transitively on Y, G(x) r(x) = Sym-j and Gi(x) = 1. •

It is clear that a 3-arc-transitive graph of girth 4 must be a complete

bipartite graph. Thus by (9.1.7), (9.3.3), (9.3.6) we have the following
result.
370 Locally projective graphs
Lemma 9.3.7 Let G act s-arc-transitively on Tfor s > 3 and suppose that
the girth gofT is less than or equal to 2s—I. Then one of the following
holds:
(i) g = 4, s = 3 and T is complete bipartite;
(ii) g = 2s — 2, s > 4 and T is a classical generalized (s—\)-gon srfiifi),
@2{2m) or y2{yn)for s = 4, 5 or 7, respectively;
(iii) g = 5, s = 3 and T is either the Petersen graph or the Hoffman-
Singleton graph. D

In the above lemma distance-transitive Moore graphs together with

vertex-transitive classical generalized polygons appear as extremal cases
of s-arc-transitive graphs of small girth. Notice that the Petersen graph
is locally projective while the Hoffman-Singleton graph is not.
Later in this chapter we will observe that the way in which sporadic
Petersen geometries are built from the Petersen graph is similar to the way
classical geometries are built from generalized polygons. The important
property of the Petersen graph which allows us to build complicated
geometries on its base is the non-triviality of the kernel at a vertex in the
full automorphism group (9.3.6).

9.4 Main types

In this section we consider some basic properties of locally projective
graphs of type (n, q) for n > 3. By considering the action of G(x) on
T2(x) for a vertex x of the graph we will distinguish the main types of
such graphs.
Let r be a graph which is locally projective of type {n,q\ n > 3,
q = pm, with respect to a group G of its automorphisms. The action of
G(x) on F(x) induces on the latter the structure of a projective geometry
of rank (n — 1) which we denote by nx. To wit, a subset in T(x) of size
[ J ] is an element of type i in nx if and only if its setwise stabilizer in
G(x) contains a Sylow p-subgroup of G(x). The incidence relation is via
inclusion. In particular the points of nx are the vertices of T(x). Let Lx
and Hx denote the sets of lines and hyperplanes of nX9 respectively (these
two sets coincide when n = 3).
For a vertex y e T(x) let nx(y) denote the set of subspaces of nx
containing y (where y is considered as a point of nx). Let Lx(y) and
Hx(y) denote the sets of lines and hyperplanes of nx which contain y.
For y,z € T(x) let lx(y,z) denote the unique line of nx which contains
both y and z. The set Lx(y) can be naturally treated as the point set of a
 9.4 Main types 371
projective geometry of rank (n — 2). This geometry (which is the residue
of y in nx) will also be denoted by nx(y). In what follows we sometimes
identify projective geometries with their point sets.
Let P denote the action of Op(G(x)) on ^ ( x ) which is abstractly
isomorphic to 0p(G(x))/G2(x). We will see below that P is non-trivial
whenever G\(x) ^ 1. Let D denote the set of orbits of P on T2(x).
Let G(x) denote the permutation action induced by G(x) on S and use
similar notation for subgroups in G(x), so that P is the identity group.
Our nearest goal is to determine the possibilities for G(x) (compare
[Wei78]).

Proposition 9.4.1 Let (F, G) satisfy Hypothesis LP for some n > 3 and
q = pm. Suppose that Gi(x) ^ 1. Let P = Op(G(x))/G2(x) and let Z
denote the set of orbits of P on ^ ( x ) . Then for S e Z we have \S\ = q
and P induces on S an elementary abelian group (of order q).
Let s be the integer such that G acts strictly s-arc-transitively on Y and
let g denote the girth of T. Then one of the following cases (1) and (2)
holds.
(1) s = 3, Ln(q)xLn.i(q)< G(x) < PTLn(q)xPTLn-1(q), a projective
geometry Sx of rank (n — 2) over GF(q) is associated with x so that
G(x) acts flag-transitively on nx x Sx, there is a mapping (px of 2*
onto the point set of dx which commutes with the action of G(x)
and for y G F(x) the restriction of (px to ny(x) is a collineation.
Moreover, either
(1.1) g = 4, F is the complete bipartite graph Xm>m where m= ["]q, or
(1.2) g > 4, Z = {S(u9oc) \uenx, a e dx}, and S(u,a) cz T(y) exactly
when u = y.
(2) s = 2, Ln(q) < G(x) < PTLn(q), for every vertex y G T(x) there is
an isomorphism \p = xpxy of nx(y) onto ny(x) which commutes with
the action on G(x,y) and for all edges one of the following subcases
occurs:
(2.1) \p is a collineation and either
(2.1-i) g = 4, E = {S(l) | / e Lx), and S(l) cz T(y) exactly when
I e Lx(y), or
(2.1.ii) g > 4, I = {S(uJ) \uenxj € Lx(u)} and S(uJ) c T(y)
exactly when u = y;
(2.2) xp is a correlation (n > 4) and either
372 Locally projective graphs

(2.2.i) g = 4, Z = {S(ft) | ft G if x }, and S(h)

ft € #*()>), or
(2.2.ii) g > 4, I = {S(u,ft) | M G 7ix,ft G H X (M)}, ami S(u,ft)
exactly when u = y.

Proof. Let H = G(x,y)r^y\ Since G(x) r ^ contains as a normal sub-

group the group Ln(q) in its natural doubly transitive action, by (2.4.3
(i)) H = A : £, where ^4 = OP(H) is an elementary abelian group of
order qn~l and SLn-i(q) < B < TLn-i(q) with £ acting faithfully and
irreducibly on A. This implies in particular that every nontrivial normal
subgroup of H contains A. By (2.4.4) if S is an orbit of A on T(y) \ {x}
then S = l\ {x} for a line / G Ly(x) and the action induced by A on S is
of order g.
Let M = Gi(x)r(y\ It is standard that M is non-trivial. Indeed, if
M = 1 then since Gi(x) <! G(x) and G(x)F2(x) is transitive, we obtain that
Gi(x)r2(x) = 1. But the latter implies Gi(x) = 1 by (9.1.4), a contradiction.
Hence M is a nontrivial normal subgroup in H and M > A by the above
paragraph. Let z G T(x) \ {y} and suppose that N := Gi(x,z)r(>;) ^ 1.
Then by (9.2.2) N is of a prime power order. Since N is normal in
M and J5 acts faithfully on A, N must intersect A properly and hence
the primes in (9.2.2) and in the present proposition are the same. Since
Op(G(x)/Gi(x)) = 1 this implies that Pr{y) = A and the first paragraph
of the proposition is proved.
Furthermore, since Ln-i(q) < HLy{x) < PTLn-i(q\ and M < H, the
action of M on Ly(x) is either (1) transitive or (2) trivial. Consider
possibility (1). In this case Gi(x) is transitive on T(y) \ {x} and applying
(9.1.3) one can see that s > 3. Second, by (2.4.1) the group Ln-\(q) is
simple except for the cases (n, q) = (3,2) and (3,3). Hence we conclude
that either (a) ML^*) > Ln-i{q) or (b) (n,q) = (3,2) and ML*W = 3 or
(c) (n,q) = (3,3) and M L ' (x) = 22.
We show that the last two cases cannot be realized. With this end in
mind, let us consider the centralizer C(x) of the group Gi(x) in G(x)
and its complete preimage C(x) in G(x). We claim that G\{y) < C(x).
Indeed, G t (x)< Gi^y^Gxiy)^ G(x,y) and Gi(x)nGi(y) = Gi(x,y) <P;
therefore Gi(x) n Gi(y) = 1, i.e. G(x,y) > Gi(x) x Gi(y). Since Gi(j;)r(x)
is non-trivial and C(x) < G(x), we conclude that Ln{q) < C(x)r(x). The
factor group G(x9y)/(C(x)nG(x,y)) is isomorphic to 2 in case (b) and to
3 in case (c). At the same time, by what we said above and since Gi(x)
is abelian in both the cases under consideration, we have G(x) = C(x) in
case (b) and [G(x) : C(x)] < 2 in case (c), a contradiction.
 9.4 Main types 373

By this we see that G\(x) contains Ln_i(g), and since the centre of the
latter group is trivial, G(x) > G\(x) x C(x). From this it is straightforward
to conclude that
Ln(q) x Lw_!(4) < G{x) < PTLn(q) x
Therefore Ln-\(q) < G\(x) < PTLn-\(q) and for y e T(x) the group
G\(x) induces on Ly(x) the natural doubly transitive action. For another
vertex z e T(x) the action is either (a) isomorphic or (b) dual and there
is an isomorphism £*z between ny(x) and %z(x) which commutes with
the action of G\(x). In case (a) it is a collineation and in case (b) it
is a correlation. Since G(x)r(x) is doubly transitive, the form of £*z is
independent of the particular choice of the pair (y,z). For a third vertex
u e F(x) the isomorphism £*M can be realized as a product of £*z and £*u.
Since the product of two collineations as well as of two correlations is a
collineation we see that £*z must be a collineation for all y,z e T(x). So
we can define a single projective geometry dx of rank (n—2) and an action
of G(x) on this space so that for every y € T(x) there is a collineation
Xy of 7iy(x) onto dx which commutes with the action of G(x,y). The
points of Sx are equivalence classes of orbits from X, where two orbits
are equivalent if their setwise stabilizers in G\(x) coincide. So we have
the desired mapping cpx of Z onto (the point set of) Sx. Let (u,y,x,z,vi)
and (u,y,x,z,V2) be two 4-arcs in F such that ly(x,u) \ {x} is equivalent
to Zz(x, v\) \ {x} but not to lz(x, vj) \ [x] with respect to the above defined
equivalence relation. Then clearly these two 4-arcs are in different orbits
of G and so G acts strictly 3-transitively on F. If the girth of F is 4 it
must be complete bipartite by (9.1.7). Otherwise different pairs (u,a) with
u e T(x) and a € dx determine different orbits S(u,oc) G Z. Thus, case (1)
is completely settled.
Let us turn to case (2). Here MLy^ = 1, and hence G\(x) = 1. From
this it follows that Ln(q) < G(x) < PTLn(q). Furthermore, Ln_i(<?) <
G(x,y) < PTLn-i(q) induces natural doubly transitive actions on Lx(y)
and on Ly(x). Hence there is an isomorphism xpxy between nx(y) and
ny(x) commuting with the action of G(x,y). In this case xpxy can be
a collineation as well as a correlation. Certainly the type of xpxy is
independent of the choice of the edge {x,y}. If the girth of F is at least
5, then we arrive at situations (2.1.i) or (2.2.ii), respectively.
Suppose that F is of girth 4. We consider here the case of collineation
(the correlation case can be treated quite analogously). Consider the set
Y = {(u,l) | u e T(x),l e Lx(u)}. A pair (uj) e Y determines a unique
orbit from E, namely xpxu(l) \ {x}. Since the girth of F is 4 some pairs
374 Locally projective graphs
correspond to the same orbits from E and we have an equivalence relation
on Y preserved by the action of G(x). The group G(x) induces on Y an
action of Ln(q) < G(x)/Gi(x) < PYLn(q) on the cosets of a premaximal
parabolic subgroup with respect to the natural projective geometry. So
it follows from (2.4.2 (v)) that a proper subgroup of G{x)/G\(x) which
contains the stabilizer of (w, /) e Y must be either the stabilizer of u or
the stabilizer of /. From this we easily see that the orbits from E are in
a bijection with the lines of nx, thus leading to subcase (2.1.i). •
By (9.2.1) if r corresponds to the subcase (1.1), (2.1.1) or (2.2.i) in the
above proposition then it is isomorphic to Km,m, S>n(q) or srfln{q\ respec-
tively. One may notice that the graphs ^In-iiv) a n ( * ^4(Q) correspond
to the subcase (1.2).
We formulate explicitly the following results which are implicit in the
proof of (9.4.1).

Corollary 9.4.2 If s = 2 then for every edge {x,y} ofT there is a unique
isomorphism xpxy of nx(y) onto ny(x) which commutes with the action of
G(x, y). The type of xpxy (i.e. whether it is collineation or correlation) is
independent of the choice of {x,y}. •

Corollary 9.4.3 If s = 3 and {x,y} is an edge of F then there is a

collineation x$ of ny(x) onto dx commuting with the action of G(x,y). If
{y,x,z} is a 2-arc in F then £yz = (zj)" 1 /^ is a collineation of ny(x) onto
nz(x) which commutes with the action of G(x,y,z). •

For the following corollary compare (2.4.3).

Corollary 9.4.4 If s = 2 then G\(x)/Ov(G(x)) is a cyclic group whose

order divides q — l;ifs = 3 then G\(x)/Op(G(x)) contains a characteristic
subgroup K isomorphic to SLn^\(q) and Op{G(x))/G\(x,y) is a natural
module for K. •

9.5 Geometrical subgraphs

For the remainder of the chapter we assume that F is a locally projective
graph of type {n,q\ n > 3, q = pm, with respect to a subgroup G in
the automorphism group of F such that Gi(x) ^ 1 for x e F. Let s be
the integer such that the action of G on F is strictly s-arc-transitive. By
(9.4.1) either s = 3 or s = 2 and F is of collineation (case (2.1)) or of
correlation (case (2.2)) type.
 9.5 Geometrical subgraphs 375
Let x e F . The action of G(x) induces on F(x) a projective geometry
structure KX. If s = 3 then the action of G(x) induces on the set of
orbits of Op(G(x)) on F2(x) a structure of a direct product nx x Sx of
projective geometries. So far these structures are defined virtually in the
sense that they were not attached to the combinatorial structure of F.
On the other hand, if F is one of the classical examples stf\{q\ ^l(q),
@n(q) and ^4(4), then with every element of the original geometry we
can associate a subgraph in F induced by the vertices which are incident
to this element in the geometry. In those cases the structures nx and
nx x dx are realized by the subgraphs of this type passing through x. We
will attempt to find a similar system of subgraphs in an arbitrary locally
projective graph. It turns out that such a system does not necessarily
exist, but if it does, then it is unique.

Definition 9.5.1 Let F be a locally projective graph of type (n, q) with

respect to G < Aut F. A connected subgraph E of F will be called a geo-
metrical subgraph if the following conditions hold:
(Gl) for every vertex x € S the intersection E D T(x) is a subspace in
n x;
(G2) if s = 2 then for every x e E the subgroup G(x) n G[3 n F(x)]
stabilizes E setwise;
(G3) if s = 3 then for every x e E the subgroup G(x) n G[E n F2(x)]
stabilizes E setwise.
We start by discussing some properties of geometrical subgraphs in
the case s = 2.

Lemma 9.5.2 Let E be a geometrical subgraph inF.Ifs = 2 then for every

edge {x,y} ofE we have E n T(y) = \pxy(E n F(x)).

Proof. By (Gl) ^ : = S f i T(x) is a subspace in nx containing y and

O := S n T(y) is a subspace in ny containing x. It is easy to deduce from
(G2) that O must be stable under G(x,y) n G[¥] and that *F must be
stable under G(x,y) n G[<D]. It follows from (9.4.1) and (2.4.2 (ii)) that
this is possible only if O = xpxyi^) and the result follows. •
Assuming that we are still in the case s = 2, let x be a vertex of F, *F
be a subspace in nx and suppose that there exists a geometrical subgraph
S in F containing x such that S n F(x) = VF. Since S is connected, for
every z eE there exists an arc (x0 = x,xi,...,xr = z) in S joining x with
376 Locally projective graphs
z. Put ¥(*,) = S n T(xi). Then we have the following (9.5.2):
(*) V(xo) = % x M , x w € V(Xi) for 1 < i < t - 1
and

*(*) = v ^ P F f e - i ) ) for 1 < i < t.

Now let X = (xo = x,xi,...,xt) be an arbitrary arc in F, originating at x.
Suppose (*) holds for a family of subspaces ¥, in nXi, 0 < i < t. Then
we say that X transfers *F to ^(xt). Notice that the subspaces ¥(*,) (if
they exist) are uniquely determined by X and XF. The following result is
a direct consequence of (9.5.2).

Lemma 9.5.3 In the case s = 2 let x be a vertex of T, *¥ be a subspace

in nx and suppose that there exists a geometrical subgraph 3 containing x
such that S n F(x) = x¥. Let z be a vertex in F. Then z is contained in S
if and only if there exists an arc in F joining x and z which transfers *F to
a subspace *F(z) in nz. If such an arc exists then *F(z) = S Pi F(z). •
The following lemma gives necessary and sufficient conditions for
existence of geometrical subgraphs in the case 5 = 2.

Lemma 9.5.4 Let s = 2, x e T and *F be a subspace in nx. Then a

geometrical subgraph S containing x such that S Pi T(x) = *¥ exists if and
only if F does not contain cycles through x transferring *F to subspaces in
ftx different from *F. Moreover, if S exists then it is unique.
Proof. Suppose that T contains a cycle (xo = x,x\,...,xt = x) which
transfers ¥ to a subspace ¥ ' in nx and ¥ ' ^ ¥ . Then a vertex « e f \ *
must be in S by (9.5.3) and it must not be in H since S 0 F(x) = *F.
Hence S does not exist in this case. Suppose that F does not contain
cycles passing through x which transfer *F to different subspaces in nx.
By vertex-transitivity this is true for every vertex of F. Let H(x, *F) be
the set of vertices in F defined as follows: z e S^^F) if and only
if there exists an arc X = (xo = x,xi,...,xt = z) in F which joins x
with z and transfers *F to a subspace *F(X) in nz. We claim that *F(X)
depends only on z but not on X. Let t be minimal with the property
that there is another arc Y = (y0 = x,yu...,ys = z) with *F(Y) ^ ¥(X).
In view of vertex-transitivity t is independent of x and xf_i ^= ys-\ by
the minimality assumption. Let r be the largest index such that xr = yq
for some 0 < q < s. Then Z = (z = xt,xf_i,...,xr = yq9yq+u».9ys = z) is
 9.5 Geometrical subgraphs 377
a cycle which transfers ^(X) to ^(Y), a contradiction. Hence ^(X) is
independent of the choice of the arc and *F(X) = S(x,^) n F(z). Since
S(x,^) is defined only in terms of x and *F, it is stable under G(x)nG[*F].
Finally if z e 3(x, *F) and ¥(z) = 3(x, xF)nr(z) the connectivity of 3(x, *F)
implies that H(x,xF) = S(z,*F(z)). This means that Hfo^F) is stable under
G(z) n GPF(z)] and hence S(x,^) is a geometrical subgraph with the
required properties. The uniqueness claim is by the construction. •
In what follows, if s = 2, x is a vertex of F and *F is a subspace in
nx then E(x,xF) will denote the geometrical subgraph containing x such
that S n F(x) = *F. This subgraph may or may not exist.
Let us consider geometrical subgraphs in the case s = 3.

Lemma 9.5.5 Suppose that 5 = 3 and let S be a geometrical subgraph in

F. Then for every 2-arc (y9x,z) in 3 we have S Pi F(z) = £JZ(3 Pi T(y)).
Proof. Since S is a geometrical subgraph containing (y,x,z), 0 :=
SnF(y) is a subspace in ny containing x and A := SnF(z) is a subspace
in nz containing x. The action of G\{x) on dx contains Ln-\(q) and hence
by (2.4.2 (ii)) %*(0) and xJ(A) are the only subspaces in dx stabilized
by Gi(x) n G[O] and G\(x) D G[A], respectively. Since both of these
subgroups are contained in G[3nF2(x)], we have %*(<!>) = Xz(A) and the
result follows. •
Still assuming that 5 = 3, consider a geometrical subgraph 3 in F. For
an edge {x, y} in 3 put *F = 3 n F(x) and <I> = 3 n T(y). Let z e 3 and let
(XQ = x, xi,..., xt = z) be an arc joining x and z in 3 (here xi may or may
not be equal to y). Let A(x,-) = 3 Pi F(x;). Then we have the following
properties (9.5.5):
(**) A(x0) = ¥, XM,X,-+I e A(xt) for 1 < i < t - 1, A(xt) = ^Xl(O)

and A(xI+1) = £*_lXw(A(x,_i)) for 1 < i < t - 1.

In the case x\—y the mapping £*Xl in (**) is assumed to be the identity.
Similarly to the case s = 2 let X = (xo = x,xi,...,xt) be an arbitrary arc
originating in x, but in this case we assume that t is even. Suppose that
for every i, 0 < i < t, there is a subspace A(x,-) in nx. such that (**) holds.
Notice that if the A(x,) exist, then they are uniquely determined by X, *¥
and O. In this case we will say that X transfers *F to A(xt) with respect
to O.
Now it is easy to prove the following analogue of (9.5.3) and (9.5.4).
378 Locally projective graphs

Lemma 9.5.6 In the case s = 3 let {1/1,1*2} be an edge in F. For i = 1 and

2 let *F,- be a subspace in nUi containing W3_,-. Then

(i) a geometrical subgraph S such that S n F(WJ) = ¥ ; /or 1 = 1 and 2

exists if and only if for j = 1 and 2 there is no arc in F transferring
*¥j with respect to ¥ 3 - / to a subspace in nUl different from *Fi or
to a subspace in nU2 different from ¥2,
(ii) if S as above exists then a vertex zofT is contained in S 1/ and
only if for i = 1 or 2 there is an arc in F which transfers ¥,- with
respect to ¥ 3 - , to a subspace A in nz, and if such an arc exists then
A = s n r(z). D

In what follows, if s = 3, {x, y} is an edge in r , *F is a subspace in nx

containing y and O is a subspace in ny containing x, then S^Y, O) denotes
the geometrical subgraph containing {x, j } such that S Pi F(x) = *F and
S n T(y) = O. Similarly to the case 5 = 2 such a subgraph may or may
not exist. Notice that *F can be taken to be the whole space nx or just a
point (similarly for $>), in particular E(nx, x) = {x} U F(x).

Definition 9.5.7 We will say that T contains a complete family of geomet-

rical subgraphs if either

(i) 5 = 2 and for every vertex x and every subspace *F in nx the geo-
metrical subgraph S(x, l F) exists, or
(ii) 5 = 3 and for every edge {x,y}, every subspace *F in nx containing
y and every subspace O in ny containing x the geometrical subgraph
H(^,O) exist.

We formulate two direct consequences of (9.5.4) and (9.5.6).

Corollary 9.5.8 Suppose that T is a tree which is locally projective with

respect to the action of G < Aut F. Then F contains a complete family of
geometrical subgraphs. •

Corollary 9.5.9
(i) Let s = 2 and F be of collinearity type. Let *¥ and *F' be sub-
spaces in nx and suppose that the geometrical subgraphs E(x,*F)
and S ( x , ^ ) exist. Then S(x,^ / n x F / ) exists and is equal to S(x,*F)fl
H(x, x¥f). In particular S(x, x F) contains S(x, ^ r ) whenever *P con-
tains * ' .
 9.6 Further properties of geometrical subgraphs 379
(ii) Let s = 3. Let {x,y} e E(T), *F and ¥ ' be subspaces in nx and O
and O' be subspaces in ny. Suppose that S^P,®) and H^F',®') exist.
Then B0Fn¥ / ,®n® / ) exists and is equal to 30F,®)n30F',®'). in
particular S^F,®) contains S^F',®') whenever *F contains *¥' and
® contains ®'. D

9.6 Further properties of geometrical subgraphs

Let S be a geometrical subgraph of F and let H be the setwise stabilizer
of S in G. If x e E then the stabilizer H(x) of x in H induces on the
set E(x) of vertices adjacent to x in S the natural permutation action of
Ln(q) of degree ["] = (qn — l)/(q — 1) (the case n = 1 is also included),
possibly extended by outer automorphisms. In particular the action of H
on S is edge-transitive. By (9.1.1) this means that for an edge e = {x,y}
in S the group if + = (H(x)9 H(y)) acts edge-transitively on S and either
H+ = H, or [if : H+] = 2 and S is bipartite. In any case we can
redefine S as the subgraph induced on the set of images under H+ of the
vertices on e = {x,y}. This enables us to give a group-theoretical version
of the necessary and sufficient conditions for existence of geometrical
subgraphs.

Lemma 9.6.1
(i) Let 5 = 2, {x,y} be an edge ofT, W be a subspace in nx containing
y and<S> = ipxyi^). Put
H+ = (G{x) n
Then the geometrical subgraph ^(x,^) exists if and only if

(ii) Let s = 3, {x,y} be an edge ofT, W be a subspace in nx containing

y and O be a subspace in ny containing x. Put
H+ = (G(x) n c m n G[xxymi G(y) n G[®] n
*e geometrical subgraph H(^, O) exists z/ and on/y i/
H+ n G(x) = G(x) n GPP] n
Proof. We consider the case (i); the case (ii) can be proved analogously.
Suppose that the geometrical subgraph H(x,xF) exists and let F be the
setwise stabilizer of S. Then H+ defined as in the lemma is contained
in F and G(x) nG^] is the stabilizer of x in F. Hence H+ n G(x) =
380 Locally projective graphs

G(x) n GPF]. If S(x,¥) does not exist then by (9.5.4) F contains a cycle
(x = xo,*!,...,** = *) which transfers ^ to a different subspace in nx.
This implies that there are subspaces ^(x,) in nXi for 0 < i < t for
which (*) holds and *F(xr) ^ *F. Then arguing as in (9.1.1) we can show
that for every i with 0 < i < t the subgroup H+ contains an element
which maps x onto x, and conjugates G(x) Pi G\*¥] to G(XJ) n G[¥(x,)].
Hence H+ contains G(x) n G\*¥(xt)] for ¥(x,) ^ ¥ and by (2.4.2 (i))
H + n G(x) = G(x). D
A vertex-transitive geometrical subgraph is locally projective with re-
spect to its setwise stabilizer. The following characterization of vertex-
transitive geometrical subgraphs is quite straightforward.

Lemma 9.6.2 Let T be a locally projective graph with respect to a group

G. A geometrical subgraph S is acted on vertex-transitively by its setwise
stabilizer in G if and only if one of the following holds:
(i) 5 = 2 and F is of collineation type;
(ii) s = 2, F is of correlation type, E = S(x, *F) and *¥ is of dimension

(iii) s = 3 and S = S(^,O) where *F and <!> have the same dimension^

If G acts s-arc-transitively on F and S is a vertex-transitive geometrical

subgraph in F, then it is easy to see that the action on S of its setwise
stabilizer is t-arc-transitive for t > s. In some cases geometrical subgraphs
happen to have higher degrees of transitivity than the original graph.
Some of those cases are described in the following lemma.

Lemma 9.6.3
(i) Suppose that s = 2, F is of collineation type and G\(x) ^ 1 for x €
F. Let *¥ be a line in nx and suppose that S = S(x, *F) exists. Then
the action on S of its setwise stabilizer is strictly t-arc-transitive for
t>3.
(ii) Suppose that s = 3 and Gi(x,y) ^ 1 for an edge {x,y} of F.
Let *F € Lx(y) and O € Ly(x) and suppose that S = S(^,O)
exists. Then the action on S of its setwise stabilizer is strictly t-arc-
transitive for t>4.

Proof, (i) Let y e *¥ and O = xp^). By (9.4.1) O \ {x} is an orbit

of Op(Gi(x)) on F2(x) and O \ {x} = S n T(y) by (9.5.2). So the result
follows directly from (9.1.3).
 9.6 Further properties of geometrical subgraphs 381
(ii) Let z e *F \ {y} and A = £*Z(O). Then by (9.4.1) A \ {x} is an orbit
of Gi(x,y) on F 2 (x) and A \ {x} = S n F(z) by (9.5.5). So again it is
sufficient to apply (9.1.3). •

For the remainder of the section let us assume that either s = 3, or

s = 2, and F is of collineation type. Let {x, y} be an edge of F, *F be
a d-dimensional subspace in nx which contains y, d > 2. Let O b e a
d-dimensional subspace in ny containing x if s = 3 and put <X> = i/^OF)
if 5 = 2. Let 2 = S0F,<D) if 5 = 3 and S = S(x, x F) if s = 2. Put
Hi = G[x,y] n G[¥ U O] and H 2 = G(x) n G^F] n G[^(0>)] if s = 3 and
if2 = G(x) D G[^P] if s = 2. If the geometrical subgraph S exists then
H = (ifi,if 2 ) is the setwise stabilizer of S in G with H\ and if2 being the
stabilizers of {x, y) and x, respectively. Independently of the existence of
S define Q = Q(F, d) to be a graph whose vertices are the (right) cosets
of Hi in G with two such vertices adjacent if they intersect a common
coset of if i in G. Clearly, in this case the edges of Q are indexed by the
cosets of if i in G. It is possible to describe Q(F, d) in combinatorial terms
as follows.
If s = 2, then the vertices of Q(F, d) are all pairs (z, A) where z is a
vertex of F and A is a d-dimensional subspace in nz with (z, A) being
adjacent to (z', A') if and only if z' G A and A' = xpZZ'(A).
If s = 3, then the vertices of Q(F, d) are all triples (z, A, a) where z is
a vertex of F, A is a d-dimensional subspace in nz and a is a (d — 1)-
dimensional subspace in <5X. This vertex is adjacent to a similar vertex
(z', A',oO if and only if z ; e A, z € A', ^'(A) = a' and tfz,{K) = a.
It is clear that G acts 2-arc-transitively on Q and with respect to this
action Q is a locally projective graph of type (d,q). Let Qc denote the
connected component of Q containing (x, *F) and let co be the mapping
of Q onto F defined by

co : (z, A) t—• z if s = 2,

co : (z, A, a) i-> z if s = 3.

Lemma 9.6.4 In the above notation the following three conditions are equiv-
alent:
(i) S exists;
(ii) ifH is defined to be (ifi,if 2 ) then H n G[x,)>] = H{;
(iii) ft is disconnected and the restriction of co to ftc is an isomorphism
onto S.
382 Locally projective graphs
Proof. To simplify the notation we only consider the case s = 2. The
equivalence of (i) and (ii) is just the vertex-transitive version of (9.6.1 (i)).
The definition of Q and (9.5.4) imply that S exists if and only if (*,¥)
and (x, *F') are in distinct connected components of Q whenever *F ^ *F',
i.e. if and only if (iii) holds. •
Let us consider in more detail the situation when s = 2 and F is of
collineation type. If S does not exist then H n G(x) contains H2 as a
proper subgroup. Since H2 is maximal in G(x) this means that H contains
the whole of G(x). Since H contains H\ as well and F is connected, this
means that H = G and Q is connected. For x € F the set co-1 (x) is
an imprimitivity system of the action of G on Q and |ct> 1(JC)| = r ^ 1
where n — 1 is the rank of nx. The graph F can be reconstructed from
Q by factorizing over this imprimitivity system. Thus when S does not
exist we can study Q instead of F since both these graphs are locally
projective with respect to the same abstract group G. The next lemma
shows that if Q satisfies a certain minimality condition, then it contains
a complete family of geometrical subgraphs.

Lemma 9.6.5 Let s = 2, F be of collineation type and let d be the smallest

number such that S(x, *¥) does not exist for any d-dimensional subspace *¥
in nx. Then Q = Q(F, d) is connected and contains a complete family of
geometrical subgraphs.

Proof. If / < d then the graph ©= Q(F,/) possesses the following

description in terms of F. The vertices of ©are triples (z, O, A) where
z G F, <D and A are / - and d-dimensional subspaces in 7i2, respectively,
with Q> a A and this vertex is adjacent to a similar vertex (z'9<&9A')
if and only if z' e O, O' = xpZZ'(Q>) and A' = \pzz>(A). Since fi(F,/) is
disconnected by the minimality assumption (9.6.4), it is clear that ©is
also disconnected. •

Consider the action of Hi on the connected component Qc of Q

containing (*,¥). H2 acts on the set of vertices adjacent to (*,¥) as
it acts on the point set of *F. By (9.4.4) the elementwise stabilizer of
these vertices induces on the set of vertices adjacent to (y, O) a p-group
extended by a cyclic group whose order divides q — 1. Since the action of
G on Q is 2-arc-transitive, by (9.2.2) the elementwise stabilizer in H2 of
the vertices at distance 1 from the edge {(x9x¥),(y,Q>)} induces a p-group
on Q c . On the other hand, if m is the codimension of *P in nx, then the
elementwise stabilizer of *F in H\ induces on the set of subspaces in nx
 9.7 The structure of P 383
containing *¥ an action which contains Lm(q) as a section. This means
that Qc cannot be the whole of Q if m is at least 2, and we have the
following result.

Lemma 9.6.6 Suppose that s = 2 and that T is of collineation type. If for a

vertex x e T and a subspace *F from nx the geometrical subgraph S(x, *F)
does not exist, then *F is a hyperplane. •
In the next section we show that S(x,*F) always exists when *F is a
line.
It should be mentioned that we know only one example of a locally
projective graph which does not contain a complete family of geometrical
subgraphs. This is a graph of valency 31 related to the fourth sporadic
Janko simple group J4, which is locally projective of type (5,2). This
graph appeared in (1.13.2 (ii)).

9.7 The structure of P

In this section we restrict ourselves to the situation when either s = 3,
or s = 2, and T is of collineation type. Here we study the action of
Op(G(x)) on Tjix). We will deduce important information about this
action from analysis of vertex-transitive geometrical subgraphs which
are locally projective lines (i.e. of valency q + 1).
We use the notation introduced before (9.4.1). In particular Z stays for
the set of orbits of P = Op(G(x))/G2(x) on T 2 (x). For an orbit S e l
let P(S) be the elementwise stabilizer of S in P. By (9.4.1) P/P(S) is
elementary abelian of order q. This means that the whole of P is an
elementary abelian p-group.
We consider first the case s = 3 and assume as usual that n > 3.
Directly from (9.4.4) we have the following.

Lemma 9.7.1 Let s = 3 and G\(x,y) = 1. Then P is elementary abelian

of order qn~l and it is the natural GF(q)-module for the characteristic
subgroup of G\(x)/P isomorphic to SLn-\(q). •

Lemma 9.7.2 Let s = 3 and G\{x,y) ± l. Let x e T, ¥ e L x, u\ = y,u2,U3

be distinct points on *F and let a be a point of 5X. Then
P(S(uu a)) n P(S(u2, a)) < P(S(u3, a)).
Proof. Let O be the line from Ly(x) such that x£(O) = a. We can and
will assume that the geometrical subgraph S = S(^,O) exists. In fact,
384 Locally projective graphs
if it does not exist, then instead of F we can consider its covering tree
(which contains a complete family of geometrical subgraphs by (9.5.8))
and instead of G the free amalgamated product of G(x) and G[x,y]. The
local properties of the action of G on T including the one stated in the
lemma are preserved when we switch to the covering tree. By (9.6.3 (ii))
the setwise stabilizer H of S in G induces on S a 4-arc-transitive action.
Clearly P stabilizes 2 as a whole and it stabilizes every vertex adjacent
to x in S. Now S(uu a) is exactly the set of vertices other than x adjacent
to ut in 2 and the result follows by applying (9.3.4) to the action of P on
S. •
Let K be the characteristic subgroup in Gi(x)/Op(G(x)) isomorphic
to SLn-i(q) (compare (9.4.4)). We are going to specify P as a GF(p)K-
module.

Lemma 9.7.3 As a GF(p)-module for K = SLn-i(q) the group P is the

direct sum of I copies of the natural GF(q)-module and I < n.

Proof. We will identify subgroups in Op(G(x)) with their images in

P. Let Z = {zi,...,z/} be a non-empty subset of T(x) and put R(Z) =
flLi Gi(x,zt). Since P fixes T(x) elementwise, R(Z) is a K-submodule in
P. Let y,z e F(x) with y ^= z and Q be a submodule in P. Since the
natural module of K is irreducible, either G\(x,y) C\Q = G\(x,z) n Q
or (G\(x,y) n Q, Gi(x,z) C\Q) = Q. Let Z as above be maximal with
the property that R(Z) is not contained in Gi(x,y). Then R(Z) is a
complement to Gi(x,y) in P. Hence P is the direct sum of natural
GF(q)K-modules. Let Z = {zi,z2,...,zn} be a maximal set of independent
points in nx, which means that Z is not contained in any hyperplane of
nx. Applying (9.7.2) it is easy to show that R(Z) = G2(x) and the result
follows. •
Let us consider the subgroup of G(x) which commutes with the action
of K on P. Since Gi(x) is normal in G(x) and K is characteristic
in Gi(x)/Op(G(x)), the group G(x)/Op(G(x)) acts on K. Let F be the
kernel of this action, that is the centralizer of K in G(x)/Op(G(x)).
Since G(x)r(x) contains a normal subgroup isomorphic to Ln(q) and
AutSLn_i(g) does not have sections isomorphic to Ln(q)9 we conclude
that F contains a section isomorphic to Ln(q). This means that the
centralizer in G(x)/Op(G(x)) of the action of K on P contains a section
isomorphic to Ln(q). On the other hand the centralizer of the action of
K on its natural module is isomorphic to the multiplicative group of the
 9.7 The structure of P 385
field GF(q) and the centralizer of its action on / copies of the natural
module is isomorphic to GLi(q) by Schur's lemma. Since SLn(q) is not
involved in SLi(q) for / < n, the equality holds in (9.7.3) and we have the
following.

Lemma 9.7.4 Ifs = 3 then P is isomorphic to the tensor product of the nat-
ural GF(q)-module for SLn_i(q) and the natural GF(q)-module for SLn(q).
In particular \P\ = qn2~n. •

Let us turn to the situation when s = 2 and T is of collineation type.

Lemma 9.7.5 Let s = 2, T be of collineation type and G\(x) j= 1. Let P*

denote the dual of P. Then for every incident point-line pair (y,^) in nx
there is a subgroup P*(y^x¥) of order q in P* such that
(i) the subgroups P*(y,x¥) taken for all incident point-line pairs in nx
generate P*,
(ii) for a fixed y0 e T(x), P*(y0) := (P*(y o ,¥) I yo € *F) is the natural
GF(q)-module for the SLn(q)-section ofG(x,yo),
(iii) for a fixed *¥0 € Lx, P*(*¥o) := <P*(y,¥o) I y € *F0> is either the
natural GF(q)-modulefor the SL2(q)-section ofG(x)r^G\^Q] or the
trivial 1-dimensional GF(q)-module for this section.

Proof. We define Pm(y,*¥) to be the dual of P(S()>,*F)). Then (i) and

(ii) follow directly from (9.4.1). As in the case s = 3 we assume that
F contains a complete family of geometrical subgraphs. Let *Fo G Lx,
S = B(x,*F0) and H be the stabilizer of S in G. By (9.6.3 (i)) the action of
H on S is strictly t-arc-transitive for t > 3. Let y = u\,U29u$ be distinct
points on ^o- Then Siut^o) contains all vertices other than x adjacent to
ut in S. Clearly P is contained in H and it stabilizes every vertex adjacent
to x in S. If t = 3 then P(S(u I 5 ¥)) = P ( S ( M ; , * F ) ) for all i,j9 1 < ij < 3
by (9.3.5) and if t > 4 then P{S{uu^)) n?(S(M 2 ,*)) < P ( S ( M 3 , ^ ) ) by
(9.3.4). This implies the result. •
By (9.7.5) and (2.4.6) we have the following.

Lemma 9.7.6 In the notation of (9.1.5) i/P*(^F 0 ) is 1-dimensional then P*

is the exterior square of the natural SLn(q)-module, in particular \P\ =
2

Lemma 9.7.7 In the notation of (9.7.5) if the module P *(¥<>) is 2-dimen-

sional then \P\ > q1{n~l\
386 Locally projective graphs
Proof. Let y and z be distinct points on ^o- We are going to show
that the action of G\(x,y) on T(z) coincides with the action of G\(x) on
this set. This will immediately imply the result. Since Gi(y) is non-trivial,
it acts transitively on ¥0 \ {y} and hence the action of G(x,y,z) on T(y)
coincides with the action of G(x,y) n G[^o] on this set. On the other
hand G(x,y)C)G\K¥oi normalizes in P/G\(x,y) a unique proper subspace,
namely the dual of P(S(y,^ 0 )). Since P induces on 5(y,^o) U S(z,^F0)
an action of order q2 we observe that G\(x,y) induces on S(z, *Fo) an
action of order q, in particular G\(x,y) ^ 1. But applying the obvious
symmetry between y and z to the above observation, we obtain that the
dual of R(5(z,xFo)) is the only proper subspace in P/G\(x,z) normalized
by G(x,y,z), so the result follows. •
We conclude the section with the following.

Lemma 9.7.8 Suppose that either s = 3, or s = 2, and T is of collineation

type. Then T contains geometrical subgraphs which are locally projective
lines.
Proof. If the claim fails then the graph Q = Q(F, 1) defined before
(9.6.4) is connected. The action of G on Q is strictly t-arc-transitive for
3 < t < 7. By the construction the subgroup H2 is the vertex stabilizer
of the action of G on £1 By (9.3.4) and since Q is connected, we have
\OP(H2)\ = ql~2. On the other hand, OP(H2) contains Op{G(x)) and by
(9.7.4), (9.7.6) and (9.7.7) the latter has order at least q2{n~l). In addition
OP(H2) induces on T(x) an action of order q2{jl~2\ Since n > 2 this is a
contradiction. •

9.8 Complete families of geometrical subgraphs

If F is a classical locally projective graph, that is s/l(q)9 s$2n(q), Q)n{(\)
or ^4(4), then T contains a complete family of geometrical subgraphs.
Moreover, every geometrical subgraph of T is induced by the vertices
incident to a certain flag in the underlying classical geometry. Conversely,
if F is an arbitrary locally projective space which contains a complete
family of geometrical subgraphs, then some of these subgraphs can be
considered as elements of a diagram geometry associated with P. In this
section we specify the diagrams of geometries arising in this way.
Let us start with the case when s = 2 and T is of collineation type.
Suppose that nx is of rank n — 1 and that for every subspace *F in nx
the geometrical subgraph S(x,xF) exists. We also include the degenerate
 9.8 Complete families of geometrical subgraphs 387
cases. That is, if *F = {y} is a point, then S(x, *F) is the edge {x, y} and,
if *F is empty, then H(x,^) is just {x}. Define J»f (F) to be the geometry
of rank n whose set of elements of type i consists of the geometrical
subgraphs S(x,*F) for all vertices x € F and all subspaces *F in nx of
dimension n — i. Two elements are adjacent if one of them contains the
other one considered as subgraphs in F. This means in particular that
the elements of type n — 1 and n are the edges and the vertices of F.

Lemma 9.8.1 Let T be a locally protective graph of type (n,q) with respect
to the action of G. Suppose that s = 2 and that F is of collineation type. If
F contains a complete family of geometrical subgraphs, then G acts flag-
transitively on the above defined geometry &(T) and the geometry has the
following diagram:
1 2
Q Q • • •

where X is the geometry of edges and vertices of a geometrical subgraph

S(x, x F) where *F is a line in nx.

Proof. Let <D0 = {S,. | 1 < h < ... < im < n) be a flag in
where Etj is of type ij. Let x be a vertex in S,m. Since the incidence
relation is by inclusion, x is contained in HI; for 1 < j < m and without
loss of generality we can assume that E,-m = {x}. This means that the
subspaces ^ = T(x) n Etj for 1 < j < m — 1 form a flag in nx and
hence there is a maximal flag { ^ | l < f e < n — 1 } containing it. Let
0>i = {Eix^k) | 1 < k < n - 1} U {x}. Then by (9.5.9) €> i is a maximal
flag in ^(F) containing €>o . It is clear from the above that ^(F) belongs
to a string diagram and that the residue of x is isomorphic to nx. Let
A = {Ej | 1 < j < n — 2} be a flag in ^(F) where S 7 is of type j . Then
Sw_2 = H(x, ^ ) where ^ is a line in nx and the residue of A is formed by
the edges and vertices of Sn_2 with the natural incidence relation. Finally
it is clear that the action of G on ^(F) is flag-transitive. •

In order to get closer to the classical diagrams we need F to be

bipartite. If F is not bipartite, then instead of F we can consider its
standard double cover 2 • F. If F is not bipartite then 2 • F is bipartite
and connected, otherwise it is a disjoint union of two copies of F. The
action of 2 x G on 2 • F is locally isomorphic to the action of G on F, in
particular both actions are strictly 5-arc-transitive for the same s. In fact
2 • F shares more properties with F. For instance if the girth g of F is
even then g is also the girth of 2 • F ; 2 x G acting on 2 • F preserves an
388 Locally projective graphs
equivalence relation whose classes have size 2 and intersect both parts.
We will use this property to recognize locally projective spaces which are
standard double covers of other locally projective spaces.
Let F be as in (9.8.1) and assume in addition that F is bipartite with
parts F 1 and F 2 . Let us modify the geometry ^(F) as follows. We exclude
the edges from the element set of the geometry and consider the vertices
from F 1 as elements of type n — 1 and the vertices from F 2 as elements of
type n. The incidence relation is as it used to be, except that two vertices
from distinct parts are incident if and only if they are adjacent in F. Let
us denote the geometry obtained in this way by J"f (F). The proof of the
following result is similar to that of (9.8.1).

Lemma 9.8.2 Let F be as in (9.8.1) and in addition assume that F is

bipartite. Let G+ be the subgroup of index 2 in G preserving the parts of
F. Then G+ acts flag-transitively on the above defined geometry J f (F). The
geometry 34?(T) is described by the following diagram:
1 2 3 n-3 n-2 n-1
O O O • •

where X stays for the rank 2 geometry realized by the vertices of the (bi-
partite) geometrical subgraph S ( x , ^ ) where *¥ is a line in nx. The elements
from G \ G + perform diagram automorphisms of J4f(T) permuting the types
n—\ and n. •

Now let s = 3 and suppose that F is bipartite with parts F 1 , F 2

and that F contains a complete family of geometrical subgraphs. Let us
associate with F an incidence system ^(F) of rank 2n—2 by the following
rule. For 1 < i < n — 1 the elements of type i in ^ are the subgraphs
H(7ix, O) where x £ F 1 and O is a subspace of dimension n — 1 — i
in Sx; for n < i < 2n — 2 the elements of type i are the geometrical
subgraphs H(TTX, O) where x G F 2 and <I> is a subspace of dimension i — n
in dx. This means in particular that the elements of type n — 1 and n
are the subgraphs {x} U F(x) for x € F1 and x G F 2 , respectively. Let
Si = S(7cx,Oi) and H2 = E(ny,<f}2). If x and y are in the same part then
Si and S2 are incident if and only if either Si ^ S2 or S2 £ Si. If
x and y are in different parts then Si and S2 are incident if and only
 9.9 Graphs of small girth 389
+
if Hi n H2 ^ 0. As above let G be the index 2 subgroup in G which
preserves the parts of F.

Lemma 9.8.3 Suppose that s = 3, F is bipartite and contains a complete

family of geometrical subgraphs. Then the above defined incidence system
y(T) is a geometry on which G + acts flag-transitively; the diagram of^(T)
is
1 2 n-2 n-1 X n n+1 2w-3 2w-2
o o • •• o o o o •• • o o
q q q q q q q q

where X is the geometry of rank 2 whose incidence graph is S0F,<X>) where

W and O are lines. Every element from G\G+ performs a diagram auto-
morphism of&(r) permuting the type i and 2n—l — i for 1 < i < n — 1.

Proof. It is an easy exercise to check that ^ = ^(F) is a geometry and

G + is a flag-transitive automorphism group of ^ . Let S = {x} U F(x) be
an element of type n — 1 in ^ . If Si G res^(S) and H2 G resJ(E), then
Hi = E(nX9Q>i) for Oi G ny and H2 = E(ny,Q>2) for y G F(x) and O 2 G nx.
Since x e Si n S2, the elements Si and S2 are incident and res^(S) is
the direct sum of the res|(S) for s G {+,—}. In addition it is easy to see
from the above that res|(S) = 5X and res^(S) = nx. Since the elements of
type n have similar residues, we conclude that ^ has a string diagram as
above and all we have to do is to specify the residue of a flag of cotype
{n — 1, n}. Notice that an element 3 = {y} U T(y) of type n is incident
to S if and only if {x9y} G E(T). Let © 1 and © 2 be incident elements of
type n — 2 and n + 1, respectively. Then © 1n © 2 contains an edge {x,y}9
say, and ©i = S(7r x,^P) for *F G Ly and ® 2 = S(7ry,O) for Q> G Lx. An
element 0 = {z} U F(z) of type n — 1 or n is incident to both © 1 and
©2 if and only if z e © 1 n @ 2. On the other hand by (9.5.9) the latter
intersection is exactly S(*F, <D) and the result follows. •

9.9 Graphs of small girth

Let F be locally projective of type (n, q), n > 3, with respect to a group
G with Gi(x) ^ 1 and Gi(x,y) ^ 1 in the case s = 3. In this section if
s = 2 then F will be assumed to be of collineation type. Then by (9.7.8)
F contains a family of geometrical subgraphs of valency q + 1 which
are locally projective lines. Suppose that the girth g of F satisfies the
inequality g < 2s + 1. By (9.1.6) this is always the case when the action
of G on F is distance-transitive. We are going to show that under these
390 Locally projective graphs
circumstances a shortest cycle is contained in a geometrical subgraph of
valency q + 1.

Lemma 9.9.1 Under the above assumptions if the girth g of T satisfies

g < 2s + 1 then a shortest cycle in F can be found inside a geometrical
subgraph of valency q + 1 which is a locally projective line.
Proof. We consider different values of g separately. Let k = [ J] be
the valency of F.
g = 4. If s = 3 then by (9.3.7) F is complete bipartite and the claim
is obvious. Let s = 2 and (x9y9z) be a 2-arc (which is contained in a
4-cycle). Let ¥ = ly(x,z), 0> = xpyx(?¥) and 0 = T(x) n F(z). We know
that 0 contains y and at least one extra vertex and that 0 is a union of
G(x, y, z)-orbits. There are exactly three orbits of G(x, y) n G\*¥] on F(x),
namely {y}, O\ {y} and F(x)\O. Since Gi(x) acts transitively on *F\ {x},
the orbits of G(x,y,z) on F(x) are the same. If there are no 4-cycles in
S(x,O), then 0 = {y} U (F(x) \ O) and
l) _ fc(fc-l)
" k-q
which is not an integer if n > 3 and q>2.
g = 5. Since n > 3 the case s = 3 is impossible by (9.3.7) and we assume
that s = 2. As above let £ be the set of orbits of P = Op(G(x))/G2(x) on
F2(x). Since F contains cycles of length 5, there are vertices u,v e F2(x)
which are adjacent. Let Si,S2 e S so that u e S\, v G S2. Since there are
no triangles in F, we have Si ^ 52. Moreover, if Si a T(y) and S2 c F(z)
for y,z G F(x), then y ^ z since otherwise (y9u9v9z) would be a 3-cycle.
Since P acts regularly on Si and S2, we conclude that every vertex from
Si is adjacent to exactly one vertex from S2 and hence P(Si) = P(S2).
Thus we obtain a non-trivial equivalence relation on E, defined by
Si - S2 if and only if P(Si) = P(S2),
which is invariant under G(x). Let S = S(w, /) G Z for / € Lx(w) and T
be the union of P-orbits equivalent to S. Then G[S] = G(x)nG{w)nG[l]
and G[T] contains G[S] as a proper subgroup. By (2.4.2 (v)) either
G[T] = G(x,w) or G[T] = G(x) n G[l] and since clearly the former is
impossible we conclude that S(w,l) ~ S(t,m) if and only if / = m, in
particular the cycle (x,y,u,v,z) is contained in the geometrical subgraph
S ( x , n where V = lx(y,z).
 9.9 Graphs of small girth 391
g = 6. In this and the next cases we have s = 3. Let (u, y, x, z) be a 3-arc
in F contained in a 6-cycle and let 0 = F(z) n T2(u). Then ©contains x
and at least one extra vertex and ©is a union of G(u, y, z, x)-orbits. Let
¥ = ly(u,x) and ¥ ' = ^Z(^F). Then G(y,x,z) D G\¥] has three orbits on
T(z), namely {x}, *F'\{x} and F(z)\¥'. Since Gi(x,y) acts transitively on
*F\{x}, the orbits of G(u,y,x,z) on T(z) are the same. Suppose that there
are no 6-cycles containing (u,y,x,z) and contained in 2(^,0), where
# = lx(y,z)- Then ©= {x} U (F(z) \ *¥') and similarly to the case g = 4
we achieve a contradiction with the integrality of

g = 7. Let X = (x,y,u,z) be a 3-arc in F (contained in a 7-cycle) and

let S = S(/3;(x,«),/M();,z)) be the unique geometrical subgraph of valency
q + 1 which contains X. Then G(X) acting on F(z) \ {M} has exactly two
orbits, ©i and ©2 , with length q and k—q—1 such that ®IU{M} = F(z)nS.
If S does not contain 7-cycles then the vertices in © 1 are all in F^x).
Let Q be the connected component containing z of the subgraph in F
induced on F3(x). Then the valency k\ of Q is k — q — 1 and its girth is
at least the girth of F which is 7. Hence Q contains at least

1 +fci +ki(k1 - 1) + M*i - I)2 = k\ -k\ + kx + 1

vertices and without loss of generality we can assume that

Let us produce an upper bound on |Q n T2(y)\. We claim that an

element h e G\(x) n G(z) fixes Q elementwise. In fact, suppose that
v, w G F3(x) Pi F(z) and i;* = w. Let (v,t,s,x) be the shortest arc joining
v and x. Since h e Gi(x), sh = s and (z^^Sjt^WjZ) is a 6-cycle, a
contradiction to g = 7. Since Q is connected, this implies that h must fix
it elementwise. Let F = Gi(x) n Gi(^) n G(z), S be the orbit of Op(G(y))
on F2(y) containing z and *¥ = ly(u,x). Since by (9.7.4) Op(G(y))/G2(y)
is the tensor product of the natural module for SLn(q) and the natural
module for SLn-i(q), it is easy to show that F fixes elementwise an
orbit T of Op(G(y)) on T2(y) if and only if T c F(w) for w G T and
T = <^W(S). This shows that F fixes exactly 42 vertices in T2(y) n F3(x),
which contradicts the above established lower bound on |Q n F2(y)|. •
By (9.6.3), (9.3.7) and the above lemma we have the following.
392 Locally projective graphs
Proposition 9.9.2 Let T be a locally projective graph of type (n,q), n>3,
with respect to a group G. Suppose that G\(x,y) j= 1 in the case 5 = 3,
Gi(x) ^ 1 and F is of collineation type in the case s = 2. Suppose also
that g <2s+l. Then a geometrical subgraph E in F of valency q + 1 is
( s + \)-arc transitive of girth g, so that one of the following holds:
(i) g =
(ii) g = 5, s = 2, q = 2 and S is the Petersen graph;
(iii) g = 6, s = 3 and 3 = sfiiq). •

By (9.2.1) in case (i) of the above lemma F is either complete bipartite

or isomorphic to 3>n{q). We will show later in this chapter that in case
(iii) F is isomorphic to ^\n_2{q) and in case (ii) F can be constructed
from the derived graph of a P -geometry.

9.10 Projective geometries

In this section we classify the pairs (F, G) which satisfy (9.9.2 (iii)). If
(F, G) satisfies the conditions and F is not bipartite then it corresponds
to the pair (2 • F, 2 x G) which satisfies the same conditions (since the
girth of F is even) and the graph is bipartite. On the other hand the pair
(2 • F, 2 x G) is specified by the property that 2 x G preserves on 2 • F
an imprimitivity system with classes of size 2. We will show that if (F, G)
satisfies (9.9.2 (iii)) and F is bipartite then F = ^\n_2{qy By (1.6.5) and
since G\(x,y) ^ 1 this will imply that Lm-iiq) ^ G + . Since L2n-i(q) does
not preserve on ^2n-2(#) a n equivalence system with classes of size 2,
there are no non-bipartite examples at all.
Thus we assume that (F, G) satisfies (9.9.2 (iii)) and F is bipartite.
By (9.8.3) and (1.6.3) to establish the isomorphism F ^ ^2ln_2{q) it
is sufficient to show that F contains a complete family of geometrical
subgraphs.

Lemma 9.10.1 Suppose that (F, G) satisfies (9.9.2(iii)). Then F contains a

complete family of geometrical subgraphs.

Proof. Let {x,y} e E(T). In view of (9.5.9) it is sufficient to show

that for every subspace O in ny the geometrical subgraph E(nX9 <D) exists.
Certainly we may assume O to be a proper subspace. Let A = %J(O)
be the image of O in dx (compare (9.4.3)) and let Gi(x, A) denote the
largest subgroup in Gi(x) which stabilizes elementwise every orbit S G S
whose image q>x(S) in dx is contained in A. Let 0 be the connected
 9.10 Protective geometries 393
component containing x of the subgraph in T induced on the vertices
fixed by Gi(x,A). We claim that 0 = S(TCX,O). Since Op(G(x))/G2(x)
is the tensor product on natural modules (9.7.4), it is easy to see the
following: if v G r 2 (x) and S is the orbit of Op(G(x)) containing v,
then v is fixed by Gi(x,A) if and only if q>x(S) G A. Thus in order
to establish the claim it is sufficient to show that whenever v is at
an even distance from x in ®, there is a subspace M in d v (whose
dimension is equal to the dimension of A) such that Gi(x,A) = G\(v,M).
Moreover, since ®is connected, it is sufficient to consider the case when
v G F2(x)n©and in view of the obvious symmetry it is sufficient to show
that Gi(x, A) < Gi(v,M). Thus let v € r2 (x)n©and let {y} = r ( x ) n r » .
We first show that Gi(x,A) < G\{v). Let W = (x,y,v,u) be a 3-arc and
let S be the unique geometrical subgraph of valency q + 1 which contains
W. Then cpx(E n r 2 (x)) G A and hence Gi(x, A) fixes S n F2(x). Since
S ^ sfiiq), by (9.3.4 (i)) Gi(x,A) < G(S) < G(M), which means that
G1(x,A)<G1(v).
As above let O be the largest subspace in ny fixed elementwise by
Gi(x, A). Then <£\{tf} is contained in ^ ( x ) and the image M of <1> in Sv is
a subspace whose dimension is equal to the dimension of A. Let z e Tjiv),
z £ T(y) and <pv(z) e M. Let us show that Gi(x,A) < G(z). Towards this
end consider a 3-arc U = (z,w,v,y) where {w} = T(z) D F(i;) and let S'
be the unique geometrical subgraph of valency q + 1 which contains U.
Then <pv(3(y) \ {v}) = <pv(z) G M and hence Gi(x, A) fixes every vertex
adjacent to y and contained in 3. Let us show that Gi(x,A) fixes every
vertex at distance at most 2 from y in 3'. We have noticed that every
vertex t adjacent to y in 3' is fixed by Gi(x, A). The vertex t is contained
in r 2 (x) U {x} and it is fixed by Gi(x, A). We have shown in the previous
paragraph that in this case Gi(x,A) < G\(t) and hence Gi(x,A) indeed
fixes every vertex at distance at most 2 from y in 3. Since 3' = stfiiq)
this implies that Gi(x,A) < G(3\ in particular Gi(x,A) < G(z). Thus
Gi(x,A) < Gi(x,M) and the result follows. •

In view of the discussion at the beginning of the section we obtain the

following.

Proposition 9.10.2 Let T be a locally protective graph of type (n,q), n > 3,

with respect to a group G, such that the action ofGonT is 3-arc-transitive
and Gi(x,>>) =/= 1 for {x,y} G E(T). Suppose that the girth ofT is 6. Then
r = ^L-2(«) and L2n-i(q) < G+. U
394 Locally projective graphs
9.11 Petersen geometries
In this section we consider the pairs (F, G) corresponding to (9.9.2 (ii))
and show that there is a P -geometry ^ on which G acts as a flag-transitive
automorphism group and F either is the derived graph A(^) or can be
obtained from A(^) by factorizing over an imprimitivity system of G. The
following result is immediate from (9.9.2 (ii)), (9.8.1) and the definition
of the derived graph of a P -geometry.

Lemma 9.11.1 Suppose that the pair (F,G) satisfies (9.9.2(ii)) and that F
contains a complete family of geometrical subgraphs. Then Jf (F) defined
before (9.8.1) is a P-geometry of rank n, G acts on J^(T) as a flag-transitive
automorphism group and F is the derived graph of J4?(T). •
Suppose now that F fails to contain a complete family of geometrical
subgraphs. By (9.6.6) if x € F and *¥ is a subspace in nx then H(x, *¥)
exists unless T is a hyperplane (i.e. unless the dimension of *¥ is n — 1)
and by (9.7.8) *P is not a line, which means that n > 4. Furthermore by
(9.6.5) the graph Q = Q(F,n — 1) is connected, locally projective of type
(n — 1,2) with respect to the action of G, and Q contains a complete
family of geometrical subgraphs. Notice that F can be obtained from
Q by factorizing over an imprimitivity system of G with blocks of size
2n — 1. Thus to achieve our goal we have to show that the pair (Q, G)
also satisfies (9.9.2 (ii)) which means that the girth of Q is 5.

Lemma 9.11.2 Under the above assumptions the girth o / Q is 5.

Proof. Let {x,y} e £(F), K = {*PX = {y},^2 ^ n - i } be a maximal
flag in nx containing {y} where *F/ is 7-dimensional. Let <I>7 = xpxyi^fj)
for 1 < j < n — 1, so that L = {Oi = {x},O2,...,On_i} is a maximal
flag in Uy containing {x}. Then a = (Xj^V-i) and fi = (y,O n _i) are
adjacent vertices in Q and S = H(x,xF2) = E(y,Q>2) is a geometrical
Petersen subgraph in F. Since q = 2,it follows from (2.4.2 (iii)) that G(S)
induces on r e s j j ^ ) the full automorphism group of the latter residue
isomorphic to Lw_2(2). Let Q be the stabilizer in G(S) of K (equivalently
the stabilizer in G(S) of L). Then Q is a Sylow 2-subgroup of G(H) and
NG(Z)(Q) = g. Put i^ = O2(G(x,y)). Since q = 2 and F is of collineation
type, we have G{x,y)/R = L n _i(2), G[x,y]/R = L n _i(2) x 2 and hence
there is an element T in G[x,};] \ G(x,y) such that T2 e R and KT = L.
This means that T stabilizes the edge {a,jS} of Q and normalizes g.
Let C = (x, y, z, M, u, x) be a 5-cycle contained in S. Since 02(G(x)) is
a non-trivial subgroup in R, the element T can be chosen to induce on
 9.11 Petersen geometries 395
C the permutation (x,y)(z9v)(u). Let a be an element in G[S] which
induces on G the permutation (x)(y,v)(z9u). Since G(S) induces the full
automorphism group of res+(¥2), w e c a n choose <r to stabilize K and
hence to normalize Q. In this case G stabilizes the vertex a of Q. The
element (TG)5 fixes C elementwise and since S is the Petersen graph, it
is easy to see that (TCT)5 G G(S). By the construction (TG)5 normalizes Q
and since Q is self-normalized in G(S) we have (TG)5 e Q. Hence (TG)5
stabilizes the vertex a and the subgraph in Q induced on the set of images
of a under (T, G) is a 5-cycle. •

The following lemma handles the graphs of correlation type.

Lemma 9.11.3 Let Y be a locally projective graph of type (n,q), n > 4,

with respect to a group G acting strictly 2-arc-transitively with G\(x) ^ 1.
If Y is of correlation type then g ^ 5.

Proof. Let u and v be adjacent vertices from ^ ( x ) and let u e S(y9 h),
v e S(z,k) in the notation of (9.4.1). Then arguing as in the proof of (9.9.1)
one can show that h = fc. On the other hand G(x, y, u) acts transitively
on the vertices in h other than y and hence for every z e h\{y} there is
a vertex v e S(z, h) which is adjacent to u. Let X = X(h) be the subgraph
of T induced on the union of the S(y, h) for all y e h. Then X is of
valency "7* o n # | "T1 vertices and its girth is at least the girth of
F which is 5. Since n > 4 this is impossible. •

Thus we have established the main result of the chapter.

Proposition 9.11.4 Let T be a locally projective graph of type (n,q), n>3,

with respect to a group G such that G\(x) ^ 1 and suppose that the girth
of F is 5. Then q = 2 and there is a P-geometry & of rank m on which
G acts as a flag-transitive automorphism group and either m = n and Y is
the derived graph A(&), or m = n—l and Y can be obtained from A(^) by
factorizing over an imprimitivity system of G with blocks of size 2n — 1. •

The classification of flag-transitive P -geometries together with (9.11.4)

provides us with the complete classification of locally projective graphs
of type (n,q), n>39 with Gi(x) j= 1 and girth 5, as stated in (1.13.2).

It was shown in [Iv90] using the classification of P3 -geometries in

[Sh85] that a distance-transitive graph which is either the derived graph
of a P -geometry ^ or a quotient of A(^) must be the derived graph
396 Locally projective graphs
of a P3-geometry. This together with (9.1.6), (9.2.1), (9.2.3), (9.3.3), (9.9.2),
(9.11.1) and (9.11.3) gives the following.

Theorem 9.11.5 Let G be a group acting 2-arc-transitively and distance-

transitively on a graph F so that G\(x,y) ^ 1 for {x,y} € E(T). Then T
is locally projective of type (n, q) with respect to G and F is isomorphic to
one of the following graphs:
(i) the point-hyperplane incidence graph stfln(q);
(ii) the q-analogue ^In-ifa) of the double cover of the odd graph;
(iii) the orthogonal graph @n(q);
(iv) the generalized polygon $i(q), q = 2W, or ^i^q), q = 3 m ;
(v) the incidence graph of the rank 2 tilde geometry ^(3 • Sp4(2));
(vi) a cubic distance-transitive graph on 102 vertices with the automor-
phism group isomorphic to 1^(17);
(vii) the derived graph of the P^-geometry c&(Mat2i) or @(3 • Mat?}). •
In the above context it is natural to ask for the classification of locally
projective graphs of girth 5 with G\(x) = 1. In [IP98] the classification
problem has been reduced to analysis of a family of Cayley graphs
defined as follows.
For n > 3 let T be the group freely generated by the involutions from
the set
D = {tt\ 1 < i < R] 4 }.
Suppose that the structure II of an (n — l)-dimensional projective GF(4)-
space is defined on D, so that {tu-Js} is a line. Let A = PGLn(4) be
a subgroup in the automorphism group of II and G be the semidirect
product of T and A with respect to the natural action. Let R be the
normal closure in G of the element t i ^ ^ . Let W(n) be the Cayley
graph of T/R with respect to the (bijective) image in T/R of the
generating set D. Then it can be seen that W(n) is a locally projective
graph of type (n,4). If T/R is abelian (equivalently if £i£2^5*4 € R) then
the girth of W(n) is 4, otherwise the girth is 5. The main result of [IP98]
is the following.

Theorem 9.11.6 Let Y be a locally projective graph of type (n,q) with

n > 3, and of girth 5, with respect to a subgroup G of automorphisms of
F. Suppose that Gi(x) = 1. Then one of the following holds:
(i) n = 4, q = 2, F is the derived graph of the P-geometry &(Mat23);
(ii) n>3,q — 4 and F is a quotient of the graph W(n); moreover
 9.11 Petersen geometries 397
(a) W(3) is of girth 5 and has exactly 220 vertices,
(b) the regular subgroup T of the automorphism group of W(n) sat-
isfies [T,T,T] = 1, both T/[T,T] and [T,T] are elementary
abelian 2-groups of rank less than [J]4 and [^4, respectively, in
particular W(n) is finite. •
We conjecture that W(n) is of girth 5 for all n > 3. Notice that a
geometrical subgraph of valency 5 in W(3) is isomorphic to the so-
called Wells graph. The Wells graph is a 2-fold antipodal cover of the
folded 5-cube. The automorphism group 2l +4 : Alts of the Wells graph is
isomorphic to involution centralizers in the sporadic simple Janko groups
J2 and J3.
 Bibliography

[Alp65] J.L. Alperin, On a theorem of Manning, Math. Z. 88 (1965), 434^35.

[AB95] J.L. Alperin and R.B. Bell, Groups and Representations, Springer,
New York, 1995.
[A83] M. Aschbacher, Flag structures on Tits geometries, Geom. Dedic. 14
(1983), 21-32.
[A84] M. Aschbacher, Finite geometries of type C3 with flag-transitive
automorphism group, Geom. Dedic. 16 (1984), 195-200.
[A94] M. Aschbacher, Sporadic Groups, Cambridge Univ. Press, Cambridge, 1994.
[A97] M. Aschbacher, ^-Transposition Groups, Cambridge Univ. Press,
Cambridge, 1997.
[ASeg91] M. Aschbacher and Y. Segev, The uniqueness of groups of type J 4 ,
Invent. Math. 105 (1991), 589-607.
[ASeg92] M. Aschbacher and Y. Segev, Extending morphisms of groups and
graphs, Ann. Math. 135 (1992), 297-324.
[ASei76] M. Aschbacher and G.M. Seitz, Involutions in Chevalley groups over
fields of even order, Nagoya Math. J. 63 (1976), 1-91.
[BI84] E. Bannai and T. Ito, Algebraic Combinatorics I, Association Schemes,
Benjamin-Cummings Lect. Notes, Benjamin, Menlo Park, Calif., 1984.
[BIP98] B. Baumeister, A.A. Ivanov and D.V. Pasechnik, A characterization of
the Petersen-type geometry of the McLaughlin group, Math. Proc. Comb.
Phil. Soc. 18 (2000).
[Ben80] D.J. Benson, The simple group J4, Ph.D. Thesis, Cambridge Univ., 1980.
[BJL86] T. Beth, D. Jungnickel and H. Lenz, Design Theory,
Cambridge Univ. Press, Cambridge, 1986.
[Beu86] A. Beutelspacher, 21 — 6 = 15: A connection between two distinguished
geometries, Amer. Math. Monthly 93 (1986), 29-41.
[Big75] N.L. Biggs, Designs, factors and codes in graphs, Quart. J. Math. Oxford
(2)26(1975), 113-119.
[BW92a] J. van Bon and R. Weiss, A characterization of the groups Fi'22, ^'23
and Fi2A, Forum Math. 4 (1992), 425^32.
[BW92b] J. van Bon and R. Weiss, An existence lemma for groups generated by
3-transpositions, Invent. Math. 109 (1992), 519-534.
[BCN89] A.E. Brouwer, A.M. Cohen and A. Neumaier, Distance-Regular Graphs,
Springer Verlag, Berlin, 1989.
[BvL84] A.E. Brouwer and J.H. van Lint, Strongly regular graphs and partial

398
 Bibliography 399

geometries, in: Enumeration and Design, D.M. Jackson and S.A. Vanstone
eds., Academic Press, Toronto, 1984, pp. 85-122.
[Bro82] K.S. Brown, Cohomology of Groups, Springer Verlag, Berlin, 1982.
[Bue79] F. Buekenhout, Diagrams for geometries and groups, J. Combin. Theory,
All (1979), 121-151.
[Bue81] F. Buekenhout, The basic diagram of a geometry, in: Geometries and
Groups, Lect. Notes Math. 893, Springer Verlag, Berlin, 1981, pp. 1-29.
[Bue85] F. Buekenhout, Diagram geometries for sporadic groups, In: Finite
Groups-Coming of Age, Proc. 1982 Montreal Conf., J. McKay ed.,
Contemp. Math. 45, AMS, Providence, R.I., 1985, pp. 1-32.
[BF83] F. Buekenhout and B. Fischer, A locally dual polar space for the
Monster, unpublished manuscript around 1983.
[BH77] F. Buekenhout and X. Hubaut, Locally polar spaces and related rank 3
groups, J. Algebra, 45 (1977), 391^34.
[BP95] F. Buekenhout and A. Pasini, Finite diagram geometries extending
buildings, in: Handbook of Incidence Geometry, F. Buekenhout ed., Elsevier,
Amsterdam, 1995, pp. 1143-1254.
[Chi94] K. Ching, On 3-transitive graphs of girth 6, J. Combin. Theory (B) 62
(1994), 316-322.
[CPr82] P.J. Cameron and C.E. Praeger, Graphs and permutation groups with
projective subconstituents, J. London Math. Soc. 25 (1982), 62-74.
[Con69] J.H. Conway, A characterization of Leech's lattice, Invent. Math. 7
(1969) 137-142.
[Con71] J.H. Conway, Three lectures on exceptional groups, in: Finite Simple
Groups, M.B. Powell and G. Higman eds., Acad. Press, New York, 1971,
pp. 215-247.
[Con85] J.H. Conway, A simple construction of the Fischer-Griess monster
group, Invent. Math. 79 (1985), 513-540.
[Con92] J.H. Conway, 7555 and all that, in: Groups, Combinatorics and Geometry,
Durham 1990, M. Liebeck and J. Saxl eds., London Math. Soc. Lect. Notes
165, Cambridge Univ. Press, Cambridge, 1992, pp. 22-23.
[CCNPW] J.H. Conway, R.T. Curtis, S.R Norton, R.A. Parker and R.A. Wilson,
Atlas of Finite Groups, Clarendon Press, Oxford, 1985.
[CNS88] J.H. Conway, S.R Norton and L.H. Soicher, The bimonster, the group
Y555 and the projective plane of order 3, in: Computers in Algebra,
M.C. Tangora ed., Marcel Dekker, 1988, pp. 27-50.
[CP92] J.H. Conway and A.D. Pritchard, Hyperbolic reflexions for the bimonster
and 3Fi24, in: Groups, Combinatorics and Geometry, Durham 1990,
M. Liebeck and J. Saxl eds., London Math. Soc. Lect. Notes 165,
Cambridge Univ. Press, Cambridge, 1992, pp. 23^5.
[CS88] J.H. Conway and N.J.A. Sloane, Sphere Packing, Lattices and Groups,
Grundlehren Math. Wiss., 290, Springer, Berlin, 1988.
[DGMP] A. Del Fra, D. Ghinelli, T. Meixner and A. Pasini, Flag-transitive
extensions of Cn geometries, Geom. Dedic. 37 (1991), 253-273.
[DGS85] A. Delgado, D.M. Goldschmidt and B. Stellmacher, Groups and
Graphs: New Results and Methods, Birkhauser Verlag, Basel, 1985.
[Ebe94] W. Ebeling, Lattices and Codes, Vieweg, Wiesbaden, 1994.
[Edge65] W. L. Edge, Some implications of the geometry of the 21-point plane,
Math. Z. 87 (1965), 348-362.
[FII86] LA. Faradjev, A.A. Ivanov and A.V. Ivanov, Distance-transitive graphs
of valency 5, 6 and 7, Europ. J. Combin. 7 (1986), 303-319.
 400 Bibliography
[FH64] W. Feit and D. Higman, The nonexistence of certain generalized
polygons, J. Algebra 1 (1964), 114^131.
[Glea56] A.M. Gleason, Finite Fano planes, Amer. J. Math. 78 (1956), 797-807.
[GS75] J.M. Goethals and J.J. Seidel, The regular two-graph on 276 vertices,
Discrete Math., 12 (1975), 143-158.
[GM93] G. Grams and T. Meixner, Some results about flag-transitive diagram
geometries using coset enumeration, Ars Combinatoria 36 (1993), 129-146.
[Gri73] R.L. Griess, Automorphisms of extra special groups and nonvanishing
degree 2 cohomology, Pacific J. Math. 28 (1973), 355^21.
[Gri76] R.L. Griess, The structure of the "Monster" simple group, in: Proc. of
the Conference on Finite Groups, W.R. Scott and R. Gross eds, Academic
Press, New York, 1976, pp. 113-118.
[Gri82] R.L. Griess, The Friendly Giant, Invent. Math. 69 (1982), 1-102.
[GMS89] R.L. Griess, U. Meierfrankenfeld and Y. Segev, A uniqueness proof for
the Monster, Ann. Math. 130 (1989), 567-602.
[Hei91] St. Heiss, On a parabolic system of type M24, J. Algebra 142 (1991),
188-200.
[Hig64] D.G. Higman, Finite permutation groups of rank 3, Math. Z. 86 (1964),
145-156.
[Hig76] D.G. Higman, A monomial character of Fischer's Baby Monster, in:
Proc. of the Conference on Finite Groups, W.R. Scott and R. Gross eds,
Academic Press, New York, 1976, pp. 277-283.
[HP85] D.R. Hughes and F.C. Piper, Design Theory, Cambridge Univ. Press,
Cambridge, 1985.
[Hum90] J.E. Humphreys, Reflection Groups and Coxeter Groups, Cambridge
Univ. Press, Cambridge, 1990.
[Ito82] T. Ito, Bipartite distance-regular graphs of valency three, Linear Algebra
and Appl. 46 (1982), 195-213.
[Iv87] A.A. Ivanov, On 2-transitive graphs of girth 5, Europ. J. Combin. 8 (1987),
393-420.
[Iv88] A. A. Ivanov, Graphs of girth 5 and diagram geometries related to the
Petersen graphs, Soviet Math. Dokl. 36 (1988), 83-87.
[Iv90] A.A. Ivanov, The distance-transitive graphs admitting elations, Math.
USSR Izvestiya, 35 (1990), 307-335.
[Iv91a] A.A. Ivanov, Geometric presentation of groups with an application to
the Monster, In: Proc. ICM-90, Kyoto, Japan, August 1990, Springer
Verlag, Berlin, 1991 pp. 385-395.
[Iv91b] A. A. Ivanov, A geometric approach to the question of uniqueness for
sporadic simple groups, Soviet Math. Dokl. 43 (1991), 226-229.
[Iv92a] A.A. Ivanov, A geometric characterization of the Monster, in: Groups,
Combinatorics and Geometry, Durham 1990, M. Liebeck and J. Saxl eds.,
London Math. Soc. Lect. Notes 165, Cambridge Univ. Press, Cambridge,
1992, pp. 46-62.
[Iv92b] A.A. Ivanov, A presentation for J4, Proc. London Math. Soc. 64 (1992),
369-396.
[Iv92c] A.A. Ivanov, A geometric characterization of Fischer's Baby Monster,
J. Algebraic Combin. 1 (1992), 43-65.
[Iv92d] A.A. Ivanov, The minimal parabolic geometry of the Conway group Co\
is simply connected, in: Combinatorics 90: Recent Trends and Applications,
North-Holland, Amsterdam, 1992, pp. 259-273.
[Iv93a] A.A. Ivanov, Graphs with projective subconstituents which contain short
cycles, in: Surveys in Combinatorics 1993, K. Walker ed., London Math.
 Bibliography 401
Soc. Lect. Notes 187, pp. 173-190, Cambridge Univ. Press, Cambridge,
1993, pp. 173-190.
pv93b] A.A. Ivanov, Constructing the Monster via its Y-presentation, In:
Combinatorics, Paul Erdos is Eighty, vol. 1, Bolyai Soc. Math. Studies,
Budapest, 1993, pp. 253-270.
[Iv94] A.A. Ivanov, Presenting the Baby Monster, J. Algebra 163 (1994), 88-108.
pv95] A.A. Ivanov, On geometries of the Fischer groups, Europ. J. Combin., 16
(1995), 163-183.
[Iv96] A.A. Ivanov, On the Buekenhout-Fischer geometry of the Monster, in:
Moonshine, the Monster and Related Topics, C. Dong and G. Mason eds.,
Contemp. Math. 193 AMS, 1996, pp. 149-158.
pv97] A.A. Ivanov, Exceptional extended dual polar spaces, Europ. J. Combin.,
18 (1997), 859-886.
pv98a] A.A. Ivanov, Affine extended dual polar spaces, In: Trends in
Mathematics, A. Pasini ed., Birkhauser Verlag, Basel, 1998, pp. 107-121.
pv98b] A.A. Ivanov, Flag-transitive extensions of dual polar spaces, Doklady
Math. 58 (1998), 180-182.
PLLSS] A.A. Ivanov, S.A. Linton, K. Lux, J. Saxl and L.H. Soicher,
Distance-transitive representations of the sporadic groups, Comm. Algebra
23 (1995), 3379-3427.
PMe93] A.A. Ivanov and U. Meierfrankenfeld, A computer free construction of
J4, preprint, 1993, submitted to J. Algebra
[IMe97] A.A. Ivanov and U. Meierfrankenfeld, Simple connectedness of the
3-local geometry of the Monster, J. Algebra, 194 (1997), 383-407.
PPS96] A.A. Ivanov, D.V. Pasechnik and S.V. Shpectorov, Non-abelian
representations of some sporadic geometries, J. Algebra 181 (1996),
523-557.
PP98] A.A. Ivanov and C.E. Praeger, On locally projective graphs of girth 5,
J. Algebraic Combin. 7 (1998), 259-283.
[ISa96] A.A. Ivanov and J. Saxl, The character table of 2E6(2) acting on the
cosets of Fi22, in: Advanced Studies in Pure Math. 24 (1996), 165-196.
pSh88] A.A. Ivanov and S.V. Shpectorov, Geometries for sporadic groups related
to the Petersen graph. I, Comm. Algebra 16 (1988), 925-954.
[ISh89a] A.A. Ivanov and S.V. Shpectorov, Geometries for sporadic groups
related to the Petersen graph. II, Europ. J. Combin. 10 (1989), 347-362.
pSh89b] A.A. Ivanov and S.V. Shpectorov, On geometries with the diagram
o ~ o o ••• o o, preprint, 1989.
pSh90a] A.A. Ivanov and S.V. Shpectorov, The P -geometry for M23 has no
nontrivial 2-coverings, Europ. J. Combin. 11 (1990), 373-379.
pSh90b] A.A. Ivanov and S.V. Shpectorov, P -geometries of J4-type have no
natural representations, Bull. Soc. Math. Belgique (A) 42 (1990), 547-560.
pSh93a] A.A. Ivanov and S.V. Shpectorov, An infinite family of simply
connected flag-transitive tilde geometries, Geom. Dedic. 45 (1993), 1-23.
pSh93b] A.A. Ivanov and S.V. Shpectorov, The last flag-transitive P -geometry.
Israel J. Math. 82 (1993), 341-362.
pSh94a] A.A. Ivanov and S.V. Shpectorov, Natural representations of the
P-geometries of Co2-type, J. Algebra 164 (1994), 718-749.
pSh94b] A.A. Ivanov and S.V. Shpectorov, Flag-transitive tilde and Petersen type
geometries are all known, Bull Amer. Math. Soc. 31 (1994), 173-184.
pSh94c] A.A. Ivanov and S.V. Shpectorov, Application of group amalgams to
algebraic graph theory, in: Investigations in Algebraic Theory of
 402 Bibliography
Combinatorial Objects, Kluwer Acad. Publ., Dordrecht, NL, 1994, pp.
417-441.
[ISh97] A.A. Ivanov and S.V. Shpectorov, The universal non-abelian
representation of the Petersen type geometry related to J4, J. Algebra, 191
(1997), 541-567.
[ISh98] A.A. Ivanov and S.V. Shpectorov, A new cover of the 3-local geometry of
Cou Geom. Dedic. 73 (1998), 237-244.
[ISt96] A.A. Ivanov and G. Stroth, A characterization of 3-local geometry of
M(24), Geom. Dedic. 63 (1996), 227-246.
[J76] Z. Janko, A new finite simple group of order 86,775,571,046,077,562,880
which possesses M24 and the full covering group of M22 as subgroups,
J. Algebra 42 (1976), 564-596.
[JLPW] Ch. Jansen, K. Lux, R. Parker and R. Wilson, An Atlas of Brauer
Characters, Clarendon Press, Oxford, 1995.
[JP76] W. Jones and B. Parshall, On the 1-cohomology of finite groups of Lie
type, in: Proc. of the Conference on Finite Groups, W.R. Scott and R. Gross
eds, Academic Press, New York, 1976, pp. 313-327.
[Kan81] W.M. Kantor, Some geometries that are almost buildings, Europ.
J. Combin. 2 (1981), 239-247.
[Kan86] W.M. Kantor, Generalized polygons, SCABs and GABs, Lect. Notes
Math. 1181, Springer Verlag, Berlin, 1986, pp. 79-158.
[Kit84] M. Kitazume, A 2-local geometry for the Fischer group Fi24, J. Fac. Sci.
Univ. Tokyo, Sec. 1A, Math. 31 (1984), 59-79.
[KKM91] M. Kitazume, T. Kondo and I. Miyamoto, Even lattices and doubly
even codes, J. Math. Soc. Japan 43 (1991), 67-87.
[LSi77] J.S. Leon and C.C. Sims, The existence and uniqueness of a simple group
generated by 3,4-transpositions, Bull. Amer. Math. Soc. 83 (1977),
1039-1040.
[Lin89] S. Linton, The maximal subgroups of the sporadic groups Th, Fi24 and
Fi'1A and other topics, Ph.D. Thesis, Cambridge Univ. 1989.
[Liin69] H. Luneburg, Transitive Erweiterungen endlicher Permutationgruppen,
Lect. Notes Math. 89, Springer Verlag, Berlin, 1969.
[MS77] FJ. MacWilliams and N.J.A. Sloane, The Theory of Error-Correcting
Codes, North-Holland, Amsterdam, 1977.
[MSm82] G. Mason and S.D. Smith, Minimal 2-local geometries for the Held
and Rudvalis sporadic groups, J. Algebra, 79 (1982), 286-306.
[Maz79] P. Mazet, Sur le multiplicateur du groupe de Mathieu M22, C.R. Acad.
Sci. Paris 289 (1979), 659-661.
[Mei91] T. Meixner, Some polar towers, Europ. J. Combin. 12 (1991), 397-417.
[Mi95] M. Miyamoto, 21 Involutions acting on the Moonshine module,
J. Algebra 175 (1995), 941-965.
[Neu84] A. Neumaier, Some sporadic geometries related to PG(3,2), Arch. Math.
42 (1984), 89-96.
[Nor80] S.P. Norton, The construction of J 4 , in: Proc. Symp. Pure Math. No. 37,
B. Cooperstein and G. Mason eds., AMS, Providence, R.I., 1980,
pp. 271-278.
[Nor85] S.P. Norton, The uniqueness of the Fischer-Griess Monster, in: Finite
Groups-Coming of Age, Proc. 1982 Montreal Conf., J. McKay ed.,
Contemp. Math. 45, AMS, Providence, R.I., 1985, pp. 271-285.
[Nor90] S.P. Norton, Presenting the Monster? Bull. Soc. Math. Belgique (A) 42
(1990), 595-605.
 Bibliography 403
[Nor92] S.P. Norton, Constructing the Monster, in: Groups, Combinatorics and
Geometry, Durham 1990, M. Liebeck and J. Saxl eds., London Math. Soc.
Lect. Notes 165, Cambridge Univ. Press, Cambridge, 1992, pp. 63-76.
[Nor98] S.R Norton, Anatomy of the Monster: I, in: The Atlas of Finite Groups:
Ten Years on, R. Curtis and R. Wilson eds., London Math. Soc. Lect. Notes
249, Cambringe Univ. Press, Cambridge 1998, pp. 198-214.
[Par92] C. Parker, Groups containing a subdiagram o o ~ o, Proc.
London. Math. Soc. 65 (1992), 85-120.
[Pase94] D.V. Pasechnik, Geometric characterization of the sporadic groups Fi'22,
Fi23 and Fi24, J. Combin. Theory (A) 68 (1994), 100-114.
[Pasi85] A. Pasini, Some remarks on covers and apartments, in: Finite
Geometries, C.A. Baker and L.M. Batten eds., Marcel Dekker, New York,
1985, pp. 233-250.
[Pasi94] A. Pasini, Diagram Geometries, Clarendon Press, Oxford, 1994.
[PT84] S.E. Payne and J.A.Thas, Finite Generalized Quadrangles, Pitman Publ.
Ltd, London, 1984.
[PS97] C.E. Praeger and L.H. Soicher, Low Rank Representations and Graphs for
Sporadic Groups, Cambridge Univ. Press, Cambridge, 1997.
[Pr89] A.D. Pritchard, Some reflection groups and their quotients, Ph. D. Thesis,
Cambridge, Univ. 1989.
[Ron80] M.A. Ronan, Coverings and automorphisms of chamber systems, Europ.
J. Combin. 1 (1980), 259-269.
[Ron81a] M.A. Ronan, Coverings of certain finite geometries, in: Finite
Geometries and Designs, Cambridge Univ. Press, Cambridge, 1981,
pp. 316-331.
[Ron81b] M.A. Ronan, On the second homotopy group of certain simplicial
complexes and some combinatorial applications, Quart. J. Math. (2) 32
(1981) 225-233.
[Ron82] M.A. Ronan, Locally truncated buildings and M 24 , Math. Z. 180 (1982),
489-501.
[Ron87] M.A. Ronan, Embeddings and hyperplanes of discrete geometries,
Europ. J. Combin. 8 (1987), 179-185.
[RSm80] M.A. Ronan and S. Smith, 2-Local geometries for some sporadic
groups, in: Proc. Symp. Pure Math. No. 37, B. Cooperstein and G. Mason
eds., AMS, Providence, R.I., 1980, pp. 283-289.
[RSm86] M.A. Ronan and S.D. Smith, Universal presheaves on group
geometries, and modular representations, J. Algebra 102 (1986), 135-154.
[RSm89] M.A. Ronan and S.D. Smith, Computation of 2-modular sheaves and
representations for L4(2), A7, 3$6 and M24, Comm. Algebra 17 (1989),
1199-1237.
[RSt84] M.A. Ronan and G. Stroth, Minimal parabolic geometries for the
sporadic groups, Europ. J. Combin. 5 (1984), 59-91.
[Row89] P. Rowley, On the minimal parabolic system related to M24, J. London
Math. Soc. 40 (1989), 40-56.
[Row91] P. Rowley, Minimal parabolic systems with diagram
o o o ~ o, J. Algebra 141 (1991), 204^-251.
[Row92] P. Rowley, Pushing down minimal parabolic systems, in: Groups,
Combinatorics and Geometry, Durham 1990, M. Liebeck and J. Saxl eds.,
London Math. Soc. Lect. Notes 165, Cambridge Univ. Press, Cambridge,
1992, pp. 144-150.
 404 Bibliography
[Row94] P. Rowley, Sporadic group geometries and the action of involutions, J.
Austral Math. Soc. (A) 57 (1994), 35-48.
[GAP] M. Schonert et al., GAP: Groups, Algorithms and Programming, Lehrstuhl
D fur Mathematik, RWTH, Aachen, 1995.
[Seg88] Y. Segev, On the uniqueness of the Co\ 2-local geometry, Geom. Dedic.
25 (1988), 159-219.
[Seg91] Y. Segev, On the uniqueness of Fischer's Baby Monster, Proc. London
Math. Soc. (3) 62 (1991), 509-536.
[Sei73] G. Seitz, Flag-transitive subgroups of Chevalley groups, Ann. Math. 97
(1973), 27-56.
[Ser73] J.-P. Serre, A Course in Arithmetic, Springer, New York, 1973.
[Ser77] J.-P. Serre, Arbres, amalgams, SL2, Asterisque 46, Soc. Math, de France,
Paris, 1977.
[Sh85] S.V. Shpectorov, A geometric characterization of the group M22, in:
Investigations in Algebraic Theory of Combinatorial Objects, VNIISI,
Moscow, 1985, pp. 112-123 [In Russian, English translation by Kluwer
Acad. Publ., Dordrecht, NL, 1994]
[Sh88] S.V. Shpectorov, On geometries with diagram Pn, preprint, 1988. [In
Russian]
[Sh92] S.V. Shpectorov, The universal 2-cover of the P-geometry &(Co2), Europ.
J. Combin. 13 (1992), 291-312.
[Sh93] S.V. Shpectorov, Natural representations of some tilde and Petersen type
geometries, Geom. Dedic. 54 (1995), 87-102.
[ShSt94] S.V. Shpectorov and G. Stroth, Classification of certain types of tilde
geometries, Geom. Dedic. 49 (1994), 155-172.
[ShY80] E. Shult and A. Yanushka, Near n-gons and line systems, Geom. Dedic.
9 (1980) 1-72.
[Sm92] S.D. Smith, Universality of the 24-dimensional embedding of the .1
2-local geometry, Comm. Algebra, 22(13) (1994), 5159-5166.
[Soi89] L.H. Soicher, From the Monster to the Bimonster, J. Algebra 111 (1989),
275-280.
[Soi91] L.H. Soicher, More on the group Y555 and the projective plane of order
3, J. Algebra 136 (1991), 168-174.
[Sp66] E.H. Spanier, Algebraic Topology, McGraw-Hill, New York, 1966.
[Str84] G. Stroth, Parabolics in finite groups, in: Proc. Rutgers Group Theory
Year, 1983 - 1984, Cambridge Univ. Press, Cambridge 1984, pp. 211-224.
[Str96] G. Stroth, Some sporadic geometries, in: Groups, Difference Sets and the
Monster, Walter de Gruyter, Berlin, 1996, pp. 99-116.
[StW88] G. Stroth and R. Weiss, Modified Steinberg relations for the group J4,
Geom. Dedic. 25 (1988), 513-525.
[Su86] M. Suzuku, Group Theory II. Springer Verlag, 1986.
[Tay92] D.E. Taylor, The Geometry of the Classical Groups, Heldermann Verlag,
Berlin, 1992.
[Th79] J.G. Thompson, Uniqueness of the Fischer-Griess Monster, Bull. London
Math. Soc. 11 (1979), 340-346.
[Tim84] KG. Timmesfeld, Tits geometries and revisionism of the classification of
finite simple groups of Char. 2-type. in: Proc. Rutgers Group Theory Year,
1983 - 1984, Cambridge Univ. Press, Cambridge 1984, pp. 229-242.
[Tim89] F.G. Timmesfeld, Classical locally finite Tits chamber systems of rank 3,
J. Algebra, 124 (1989), 9-59.
[Ti74] J. Tits, Buildings of Spherical Type and Finite BN-pairs, Lect. Notes Math.
386, Springer-Verlag, Berlin 1974.
 Bibliography 405

[Ti82] J. Tits, A local approach to buildings, in: The Geometric Vein

(Coxeter-Festschrift), Springer Verlag, Berlin, 1982, pp. 519-547.
[Ti85] J. Tits, Le Monstre, in: Seminar Bourbaki expose no 620, 1983/84,
Asterisque 121-122 (1985), 105-122.
[Ti86] J. Tits, Ensembles ordonnes, immeubles et sommes amalgamees, Bull. Soc.
Math. Belg. A38 (1986) 367-387.
[Tro91] V.I. Trofimov, Stabilizers of the vertices of graphs with projective
suborbits, Soviet Math. Dokl 42 (1991), 825-827.
[Tro92] V.I. Trofimov, Graphs with projective suborbits, Math. USSR Izvestiya
39 (1992) 869-894.
[Tro94] VI. Trofimov, Graphs with projective suborbits. The case of small
characteristics, I and II, Math. Izvestiya Russian Acad. Sci. 58 (1994), No 5,
124-173 and No 6, 137-156.
[Vi97] M.-M. Virotte Ducharme, Some 7-groups, Geom. Dedic 65 (1997), 1-30.
[Wei78] R. Weiss, Symmetric graphs with projective subconstituents, Proc. Amer.
Math. Soc. 72 (1978), 213-217.
[Wei79a] R. Weiss, Elations of graphs, Ada Math. Acad. Sci. Hungar., 34 (1979),
101-103.
[Wei79b] R. Weiss, Groups with a (B, N)-pair and locally transitive graphs,
Nagoya Math. J., 74 (1979), 1-21.
[Wei81a] R. Weiss, s-Transitive graphs, in: Algebraic Methods in Graph Theory,
North-Holland, Amsterdam, 1981, pp. 827-847.
[Wei81b] R. Weiss, The nonexistence of 8-transitive graphs, Combinatorica, 1
(1981), 309-311.
[Wei85] R. Weiss, Distance-transitive graphs and generalized polygons, Arch.
Math. 45 (1985), 186-192.
[WY90] R. Weiss and S. Yoshiara, A geometric characterization of the groups
Suz and HS, J. Algebra 133 (1990), 182-196.
[Wie97] C. Wiedorn, A tilde geometry for F4(2), Cont. Algebra and Geom. 38
(1997), 337-342.
[Wil89] R.A. Wilson, Vector stabilizers and subgroups of Leech lattice groups,
J. Algebra 127 (1989), 387-408.
[Wil92] R.A. Wilson, The action of a maximal parabolic subgroup on the
transpositions of the Baby Monster, Proc. Edinburgh Math. Soc. 37 (1992),
185-189.
[Wil93] R.A. Wilson, A new construction of the baby Monster and its
applications. Bull. London Math. Soc. 25 (1993), 431-437.
[Wit38] E. Witt, Uber Steinersche Systeme, Abh. Math. Sem. Univ. Hamburg 12
(1938), 265-275.
[Yan76] A. Yanushka, Generalized hexagons of order (t, t), Israel J. Math. 23
(1976), 309-324.
[Yos91] S. Yoshiara, A classification of flag-transitive classical c.C2-geometries
by means of generators and relations, Europ. J. Combin. 12 (1991), 159-181.
[Yos92] S. Yoshiara, Embeddings of flag-transitive classical locally polar
geometries of rank 3, Geom. Dedic. 43 (1992), 121-165.
[Yos94] S. Yoshiara, On some extended dual polar spaces I, Europ. J. Combin. 15
(1994), 73-86.
[Yos96] S. Yoshiara, The flag-transitive C3-geometries of finite order,
J. Algebraic Combin. 5 (1996), 251-284.
 Index

amalgam, 7 extended dual polar space, 38

automorphism group, 6 affine, 39
classical, 15

Borel subgroup, 9 standard, 9

building, 14 flag-transitive, 6
frame, 147
characteristic, 2 fully normalized subgroup, 42
clique, 45
coclique, 45, 60 generalized polygon, 12
standard, 60 classical, 12
code thick, 12
binary linear, 50 geometric presentation, 36
doubly even, 50 geometry
even, 50 locally finite, 14
Golay, 50 thick, 14
length, 50 tilde, 20
self-dual, 50 Tits, 14
collineation, 56 which is almost a building, 202
completion, 8 girth, 44
faithful, 8 Golay code module, 95
universal, 8 graph
corank, 3 Baby Monster, 244
core, 42 bipartite, 45
correlation, 56 collinearity, 23
cotype, 3 connected, 45
cover, 5 derived, 27, 308
covering, 5 distance-transitive, 47
of graphs, 46 dual polar, 281
cycle, 44 edge-transitive, 46
contractible, 46 Gewirtz, 98
non-degenerate, 45 Leech, 164
line, 45
deleting procedure, 68 locally A, 45
diagram, 13 locally projective, 28, 359
diagram automorphism, 6 Monster first, 216
direct sum of geometries, 4 Monster second, 260
dodecad, 81 Moore, 362
multipartite, 45
elation, 366 octad, 102

406
 Index 407
regular, 45 octad, 54, 72
s-arc-transitive, 47 octet, 132
Schlafli, 202 orbit
sextet, 110 twisted, 275
strictly s-arc-transitive, 47 untwisted, 275
transposition, 236 orbital, 43
triangulable, 46
triangular, 200 pair, 132
vertex-transitive, 46 parabolic geometry, 11
Wells, 397 parabolic subgroup
Griess algebra, 269 maximal, 6
group minimal, 9
3-transposition, 236 parity form, 44
Bimonster, 332 Petersen geometry, 20
fundamental, 45 Petersen graph, 20
linear, 43 power set, 43
projective, 43
translation, 39 quad, 102, 281
Weyl, 12 quotient, 5

hexacode, 71 rank of an incidence system, 3

hexacode module, 71 representation
hexad, 132 group, 26
hexagonal relation, 335 module, 25
hyperoval, 65 natural, 24
permutational, 43
incidence defined by inclusion, 17 universal, 24
incidence graph, 4 residue, 3
incidence system, 3 root, 142
connected, 3
residual, 3 5-covering, 5
independent set, 64 s-simply connected, 5
intersection array, 48 sextet, 54
intersection graph of subgeometries, 29 simply connected, 5
isomorphism, 6 sporadic group
isomorphism of geometries, 5 Baby Monster, 222
Conway, 159
lattice, 142 Fischer, 236, 237
automorphism, 147 Harada-Norton, 269
basis, 142 Held, 244
dimension, 142 Higman-Sims, 189
discriminant, 142 Janko, 37
dual, 142 Mathieu, 74, 75, 83
even, 142 McLaughlin, 187
integral, 142 Monster, 271
Leech, 142 Suzuki, 40
unimodular, 142 Thompson, 269
Leech vector, 153 standard double cover, 47
shape, 154 string diagram, 23
locally projective line, 367 subconstituent, 47
subgraph
minimal weight, 50 geodetically closed, 45
Monster amalgam, 211 geometrical, 103, 375
morphism of geometries, 5 induced, 45
strongly geodetically closed, 45
natural module, 43 suborbit diagram, 48
normal closure, 42 subspace, totally singular, 2, 17
408 Index
tetrad, 54 weak parabolic system, 19
Todd module, 95 weakly locally, 337
transvection, 56 Witt index, 2
trio, 78
Y-graph, 338
universal cover, 5 Y-group, 333
universal covering, 5 strong, 334

valencies, 48