This page intentionally left blank

CAMBRIDGE STUDIES IN ADVANCED MATHEMATICS 133

Editorial Board
B . B O L L O B Á S , W . F U L T O N , A . K A T O K , F . K I R W A N ,
P. SARNAK, B. SIMON, B. TOTARO

LINEAR ALGEBRAIC GROUPS AND

FINITE GROUPS OF LIE TYPE

Originating from a summer school taught by the authors, this concise treatment
includes many of the main results in the area. An introductory chapter describes the
fundamental results on linear algebraic groups, culminating in the classification of
semisimple groups. The second chapter introduces more specialized topics in the
subgroup structure of semisimple groups, and describes the classification of the
maximal subgroups of the simple algebraic groups. The authors then systematically
develop the subgroup structure of finite groups of Lie type as a consequence of the
structural results on algebraic groups. This approach will help students to understand
the relationship between these two classes of groups.
The book covers many topics that are central to the subject, but missing from
existing textbooks. The authors provide numerous instructive exercises and examples
for those who are learning the subject as well as more advanced topics for research
students working in related areas.

g u n t e r m a l l e is a Professor in the Department of Mathematics at the University

of Kaiserslautern, Germany.
d o n n a t e s t e r m a n is a Professor in the Mathematics Section at the École
Polytechnique Fédérale de Lausanne, Switzerland.
CAMBRIDGE STUDIES IN ADVANCED MATHEMATICS

Editorial Board:
B. Bollobás, W. Fulton, A. Katok, F. Kirwan, P. Sarnak, B. Simon, B. Totaro

All the titles listed below can be obtained from good booksellers or from Cambridge University Press. For a
complete series listing visit: http://www.cambridge.org/mathematics

Already published
85 J. Carlson, S. Müller-Stach & C. Peters Period mappings and period domains
86 J. J. Duistermaat & J. A. C. Kolk Multidimensional real analysis, I
87 J. J. Duistermaat & J. A. C. Kolk Multidimensional real analysis, II
89 M. C. Golumbic & A. N. Trenk Tolerance graphs
90 L. H. Harper Global methods for combinatorial isoperimetric problems
91 I. Moerdijk & J. Mrčun Introduction to foliations and Lie groupoids
92 J. Kollár, K. E. Smith & A. Corti Rational and nearly rational varieties
93 D. Applebaum Lévy processes and stochastic calculus (1st Edition)
94 B. Conrad Modular forms and the Ramanujan conjecture
95 M. Schechter An introduction to nonlinear analysis
96 R. Carter Lie algebras of finite and affine type
97 H. L. Montgomery & R. C. Vaughan Multiplicative number theory, I
98 I. Chavel Riemannian geometry (2nd Edition)
99 D. Goldfeld Automorphic forms and L-functions for the group GL(n,R)
100 M. B. Marcus & J. Rosen Markov processes, Gaussian processes, and local times
101 P. Gille & T. Szamuely Central simple algebras and Galois cohomology
102 J. Bertoin Random fragmentation and coagulation processes
103 E. Frenkel Langlands correspondence for loop groups
104 A. Ambrosetti & A. Malchiodi Nonlinear analysis and semilinear elliptic problems
105 T. Tao & V. H. Vu Additive combinatorics
106 E. B. Davies Linear operators and their spectra
107 K. Kodaira Complex analysis
108 T. Ceccherini-Silberstein, F. Scarabotti & F. Tolli Harmonic analysis on finite groups
109 H. Geiges An introduction to contact topology
110 J. Faraut Analysis on Lie groups: An Introduction
111 E. Park Complex topological K-theory
112 D. W. Stroock Partial differential equations for probabilists
113 A. Kirillov, Jr An introduction to Lie groups and Lie algebras
114 F. Gesztesy et al. Soliton equations and their algebro-geometric solutions, II
115 E. de Faria & W. de Melo Mathematical tools for one-dimensional dynamics
116 D. Applebaum Lévy processes and stochastic calculus (2nd Edition)
117 T. Szamuely Galois groups and fundamental groups
118 G. W. Anderson, A. Guionnet & O. Zeitouni An introduction to random matrices
119 C. Perez-Garcia & W. H. Schikhof Locally convex spaces over non-Archimedean valued fields
120 P. K. Friz & N. B. Victoir Multidimensional stochastic processes as rough paths
121 T. Ceccherini-Silberstein, F. Scarabotti & F. Tolli Representation theory of the symmetric groups
122 S. Kalikow & R. McCutcheon An outline of ergodic theory
123 G. F. Lawler & V. Limic Random walk: A modern introduction
124 K. Lux & H. Pahlings Representations of groups
125 K. S. Kedlaya p-adic differential equations
126 R. Beals & R. Wong Special functions
127 E. de Faria & W. de Melo Mathematical aspects of quantum field theory
128 A. Terras Zeta functions of graphs
129 D. Goldfeld & J. Hundley Automorphic representations and L-functions for the general linear group, I
130 D. Goldfeld & J. Hundley Automorphic representations and L-functions for the general linear group, II
131 D. A. Craven The theory of fusion systems
132 J. Väänänen Models and games
 Linear Algebraic Groups and
Finite Groups of Lie Type

GUNTER MALLE
University of Kaiserslautern, Germany

DONNA TESTERMAN
École Polytechnique Fédérale de Lausanne, Switzerland
 cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Tokyo, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9781107008540

C G. Malle and D. Testerman 2011

This publication 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 2011

Printed in the United Kingdom at the University Press, Cambridge

A catalogue record for this publication is available from the British Library

Library of Congress Cataloguing in Publication data

ISBN 978-1-107-00854-0 Hardback

Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate.
 Contents

Preface page ix
List of tables xiii
Notation xiv

PART I LINEAR ALGEBRAIC GROUPS 1

1 Basic concepts 3
1.1 Linear algebraic groups and morphisms 3
1.2 Examples of algebraic groups 6
1.3 Connectedness 9
1.4 Dimension 13
2 Jordan decomposition 15
2.1 Decomposition of endomorphisms 15
2.2 Unipotent groups 18
3 Commutative linear algebraic groups 20
3.1 Jordan decomposition of commutative groups 20
3.2 Tori, characters and cocharacters 22
4 Connected solvable groups 26
4.1 The Lie–Kolchin theorem 26
4.2 Structure of connected solvable groups 27
5 G-spaces and quotients 30
5.1 Actions of algebraic groups 30
5.2 Existence of rational representations 33
6 Borel subgroups 36
6.1 The Borel fixed point theorem 36
6.2 Properties of Borel subgroups 39
vi Contents

7 The Lie algebra of a linear algebraic group 44

7.1 Derivations and differentials 44
7.2 The adjoint representation 49
8 Structure of reductive groups 51
8.1 Root space decomposition 51
8.2 Semisimple groups of rank 1 53
8.3 Structure of connected reductive groups 57
8.4 Structure of semisimple groups 59
9 The classification of semisimple algebraic groups 63
9.1 Root systems 63
9.2 The classification theorem of Chevalley 68
10 Exercises for Part I 74

PART II SUBGROUP STRUCTURE AND

REPRESENTATION THEORY OF
SEMISIMPLE ALGEBRAIC GROUPS 81
11 BN-pairs and Bruhat decomposition 83
11.1 On the structure of B 83
11.2 Bruhat decomposition 90
12 Structure of parabolic subgroups, I 95
12.1 Parabolic subgroups 95
12.2 Levi decomposition 98
13 Subgroups of maximal rank 104
13.1 Subsystem subgroups 104
13.2 The algorithm of Borel and de Siebenthal 107
14 Centralizers and conjugacy classes 112
14.1 Semisimple elements 112
14.2 Connectedness of centralizers 116
15 Representations of algebraic groups 121
15.1 Weight theory 121
15.2 Irreducible highest weight modules 125
16 Representation theory and maximal subgroups 131
16.1 Dual modules and restrictions to Levi subgroups 131
16.2 Steinberg’s tensor product theorem 134
 Contents vii

17 Structure of parabolic subgroups, II 140

17.1 Internal modules 140
17.2 The theorem of Borel and Tits 145
18 Maximal subgroups of classical type simple algebraic
groups 149
18.1 A reduction theorem 149
18.2 Maximal subgroups of the classical algebraic groups 155
19 Maximal subgroups of exceptional type algebraic groups 166
19.1 Statement of the result 166
19.2 Indications on the proof 168
20 Exercises for Part II 172

PART III FINITE GROUPS OF LIE TYPE 179

21 Steinberg endomorphisms 181
21.1 Endomorphisms of linear algebraic groups 181
21.2 The theorem of Lang–Steinberg 184
22 Classification of finite groups of Lie type 188
22.1 Steinberg endomorphisms 188
22.2 The finite groups GF 193
23 Weyl group, root system and root subgroups 197
23.1 The root system 197
23.2 Root subgroups 200
24 A BN-pair for GF 203
24.1 Bruhat decomposition and the order formula 203
24.2 BN-pair, simplicity and automorphisms 209
25 Tori and Sylow subgroups 218
25.1 F -stable tori 218
25.2 Sylow subgroups 225
26 Subgroups of maximal rank 229
26.1 Parabolic subgroups and Levi subgroups 229
26.2 Semisimple conjugacy classes 232
27 Maximal subgroups of finite classical groups 236
27.1 The theorem of Liebeck and Seitz 237
27.2 The theorem of Aschbacher 240
viii Contents

28 About the classes C1F , . . . , C7F and S 244

28.1 Structure and maximality of groups in CiF 244
28.2 On the class S 246
29 Exceptional groups of Lie type 250
29.1 Maximal subgroups 250
29.2 Lifting result 254
30 Exercises for Part III 263
Appendix A Root systems 268
A.1 Bases and positive systems 268
A.2 Decomposition of root systems 272
A.3 The length function 276
A.4 Parabolic subgroups 278
Exercises 281
Appendix B Subsystems 282
B.1 The highest root 282
B.2 The affine Weyl group 285
B.3 Closed subsystems 286
B.4 Other subsystems 290
B.5 Bad primes and torsion primes 292
Exercises 296
Appendix C Automorphisms of root systems 297
Exercises 300
References 301
Index 305
 Preface

These notes grew out of a summer school on “Finite Groups and Related
Geometrical Structures” held in Venice from September 5th to September
15th 2007. The aim of the course was to introduce an audience consisting
mainly of PhD students and postdoctoral researchers working in finite group
theory and neighboring areas to results on the subgroup structure of linear
algebraic groups and the related finite groups of Lie type.
As will be seen in Part I, a linear algebraic group is an affine variety which
is equipped with a group structure in such a way that the binary group op-
eration and inversion are continuous maps. A connected (irreducible) linear
algebraic group has a maximal solvable connected normal subgroup such
that the quotient group is a central product of simple algebraic groups, a so-
called semisimple algebraic group. Thus, one is led to the study of semisimple
groups and connected solvable groups. A connected solvable linear algebraic
group is the semidirect product of the normal subgroup consisting of its
unipotent elements with an abelian (diagonalizable) subgroup (for example,
think of the group of invertible upper triangular matrices). While one cannot
expect to classify unipotent groups, remarkably enough this is possible for
the semisimple quotient.
The structure theory of semisimple groups was developed in the middle of
the last century and culminated in the classification of the semisimple linear
algebraic groups defined over an algebraically closed field, a result essentially
due to Chevalley, first made available via the Séminaire sur la classification
des groupes de Lie algébriques at the Ecole Normale Supérieure in Paris,
during the period 1956–1958 ([15]). Analogously to the work of Cartan and
Killing on the classification of the complex semisimple Lie algebras, Cheval-
ley showed that the semisimple groups are determined up to isomorphism
by a set of combinatorial data, based principally upon a root system (as for
the semisimple Lie algebras) and a dual root system. Moreover, the set of
x Preface

possible combinatorial data does not depend on the characteristic of the un-
derlying field. Part I of this text is devoted to developing the tools necessary
for describing this classification. We have followed the development in two
very good texts ([32] and [66]) on linear algebraic groups, and we often refer
to these books for the proofs which we have omitted. Our aim is to give the
reader a feel for the group-theoretic ingredients of this classification, without
going into the details of the underlying algebraic geometric foundations, and
then to move on to the material of Parts II and III, which should perhaps
be seen as the distinguishing feature of this text.
The 20- to 30-year period following the classification was a productive time
in “semisimple” theory, during which many actors, notably Borel, Bruhat,
Springer, Steinberg and Tits, played a role in the further study of these
groups. The conjugacy classes, endomorphisms, representations, and sub-
group structure were among the topics of consideration, with the principal
aim of reducing their classification and the description of their structure to
combinatorial data related to the root system and the Weyl group of the
ambient group. Part II of this book treats some of these subjects. In partic-
ular, we describe the Tits BN -pair for a semisimple linear algebraic group
and obtain a Levi decomposition for the associated parabolic subgroups;
we discuss conjugacy classes of semisimple elements and their centralizers;
we describe the parametrization of the irreducible representations of these
groups and their automorphism groups. We leave the discussion of the gen-
eral endomorphisms of simple algebraic groups to Part III, where these are
used to construct the finite groups of Lie type.
In the last chapters of Part II, we turn to more recent developments in the
theory of semisimple algebraic groups, where we describe the classification of
their maximal positive-dimensional subgroups. These results can be seen as
an extension of the fundamental work of Dynkin on the maximal subalgebras
of the semisimple complex Lie algebras. It is the subject of several very long
research articles by Liebeck, Seitz and others and was completed in 2004.
In the course of the classification of finite simple groups, attention turned
to the analogues of algebraic groups over finite fields. These so-called fi-
nite groups of Lie type were eventually shown to comprise, together with
the alternating groups, almost all the finite simple groups. Steinberg found
a unified approach to constructing not only the well-known finite classical
groups, but also their twisted “Steinberg variations”, as well as further seem-
ingly sporadic examples, the Suzuki and Ree groups, as fixed point subgroups
of certain endomorphisms of simple linear algebraic groups defined over Fp .
There does not yet seem to be a generally accepted terminology for such
endomorphisms, and we call them “Steinberg endomorphisms” in this text.
 Preface xi

Part III is devoted to the definition and study of these finite groups. We
first classify the endomorphisms whose fixed point subgroups are finite, fol-
lowing the work of Steinberg, and hence are able to describe the full set of
finite groups “of Lie type” thus obtained. The theorem of Lang–Steinberg
then provides the necessary machinery for applying the results of Parts I
and II to the study of these fixed point subgroups. For example, we give the
proof of the existence of a BN -pair for these groups, which allows one to
deduce a formula for their order. Furthermore, we study their Sylow sub-
groups and touch on some other aspects of the subgroup structure. Finally,
we return to the question of the maximal subgroups, this time sketching a
proof of Aschbacher’s reduction theorem for the maximal subgroups of the
finite classical groups and indicating how it has been applied (by Kleidman,
Liebeck and others) and how it must still be applied if one hopes to deter-
mine the maximal subgroups of the finite classical groups. We conclude with
a discussion of what is known about the maximal subgroups of the excep-
tional finite groups of Lie type, including work of Liebeck, Saxl, Seitz and
others. We then come full-circle and sketch the proof of a result which en-
ables one to lift certain embeddings of finite groups of Lie type to embeddings
of algebraic groups, where one can apply the more complete information of
Part II.

The course is not self-contained in several aspects. First, in order to keep

the size manageable we assume the reader to be familiar with some ba-
sic notions of affine and projective algebraic varieties. In the development
of the general theory of algebraic groups we include those proofs of a more
group theoretical nature, or which just use the basic notions of connectedness
and dimension, and refer to the standard texts for the others which require
deeper methods from algebraic geometry, like properties of morphisms, tan-
gent spaces, etc. Secondly, we do not explain the Steinberg presentation of
semisimple algebraic groups, although some of its consequences are men-
tioned and needed in the text. Also, while we have included an appendix
with a self-contained development of the basic theory of root systems and
Weyl groups, as far as it is relevant for the development in the main text,
we haven’t repeated the proof of the classification of indecomposable root
systems, which has already been laid out in many texts. In any case we
give references to the results we need, and some statements form part of the
exercises.
We hope that this text will be useful to doctoral students and researchers
who are working in areas which rely upon a general knowledge of the groups
xii Preface

of Lie type, without needing to understand every detail of the proof of the
classification of semisimple groups. In particular, Parts II and III should give
a good overview of much of what is known about the subgroup structure of
these groups and to a lesser extent their conjugacy classes and representation
theory. The numerous exercises are intended to supplement and illustrate the
theory, and should help the book fulfill its objective of serving as the basis
for a first-year graduate level course.

We began working on this project at the Mathematisches Forschungsin-

stitut at Oberwolfach even before the start of the summer school, and then
continued at various places, including the EPFL (Lausanne), the Isaac New-
ton Institute at Cambridge, and the Banff International Research Station.
We thank all these institutions for providing an inspiring atmosphere and
enough fresh air, and for their hospitality. The second author would also
like to acknowledge the support of the Swiss National Science Foundation
through grants numbers PP002-68710 and 200021-122267.
We are grateful to Clara Franchi, Maria Silvia Lucido, Enrico Jabara,
Mario Mainardis and John van Bon for organizing and inviting us to teach
at the Venice summer school. We also thank Stephen Clegg, Kivanc Ersoy,
Andreas Glang, Daniele Toller and Pinar Urgurlu, for providing TeX-files of
their notes taken during our classes, made available to us shortly afterwards,
and which constituted the basis for this manuscript. We thank Meinolf Geck
for many clarifying discussions on various topics, Olivier Brunat and Ulrich
Thiel for a careful reading of a preliminary version which lead to various
improvements, and Thomas Gobet, Claude Marion and Britta Späth for
proofreading parts of the manuscript.
Finally, we would like to dedicate this volume to Maria Silvia Lucido,
who died in an accident half a year after the summer school. We were both
impressed by her enthusiasm, energy and joyfulness during our brief acquain-
tance.
 Tables

9.1 Dynkin diagrams of indecomposable root systems 67

9.2 Isogeny types of simple algebraic groups 72
11.1 Graph automorphisms of Dynkin diagrams 89
13.1 Extended Dynkin diagrams 108
14.1 Bad primes and torsion primes of root systems 117
18.1 Irreducible triples 158
18.2 Maximal subgroups in C1 ∪ . . . ∪ C4 ∪ C6 in SLn 163
18.3 Maximal subgroups in C1 ∪ . . . ∪ C4 ∪ C6 in Sp2n , n ≥ 2 163
22.1 Finite groups of Lie type 193
23.1 Automorphisms of irreducible Weyl groups 198
23.2 q |ω| for twisted groups of Lie type 201
24.1 Orders of finite groups of Lie type 208
24.2 Centers of groups of simply connected type 211
24.3 Non-generic Schur multipliers 214
26.1 Some centralizers in E6 (q) 235
27.1 Finite classical groups 236
27.2 Normalizers of extraspecial subgroups R 242
28.1 The classes CiF in SLn (q) 245
28.2 Groups H ∈ CiF lying in members K ∈ S 246
A.1 Relation between two roots 274
A.2 The two-dimensional root systems 275
B.1 Highest roots of indecomposable root systems 285
B.2 Long and short roots in indecomposable root systems 291
B.3 Highest short roots of indecomposable root systems 295
 Notation

We have tried to conform to standard notation whenever that exists. There

are a few key notions for which different conventions exist in the literature.
For us, a reductive group is not necessarily connected, while a semisimple
group always is. A simple algebraic group is a non-trivial semisimple group
with no proper positive dimensional normal subgroup. A root system which
cannot be decomposed into an orthogonal union of subroot systems will be
called indecomposable (sometimes the term irreducible is used in the lit-
erature). There does not seem to be an accepted standard notation for the
various orthogonal groups. Here, the full isometry group of a non-degenerate
quadratic form is denoted by GO, and its connected component of the iden-
tity by SO. In particular, in characteristic 2, SO is not the intersection of
GO with the special linear group SL. We have chosen the name Steinberg
endomorphisms for what some authors call (generalized) Frobenius maps,
to acknowledge Steinberg’s role in the study of these endomorphisms. Asch-
bacher gave a first subdivision of natural subgroups of classical groups over
finite fields into classes which he called Ci . Later, many of these classes were
redefined by various authors. We follow the notation of Liebeck and Seitz
in their paper [50] for these classes of subgroups. Finally, we write Zn for
the cyclic group of order n since Cn is already used for one of the root sys-
tems. We write N = {1, 2, . . .} for the set of positive natural numbers and
set N0 := N ∪ {0}.
 PART I

LINEAR ALGEBRAIC GROUPS

In this part we introduce the main objects of study, linear algebraic groups
over algebraically closed fields.
We assume that the reader is familiar with basic concepts and results from
commutative algebra and algebraic geometry. More specifically, the reader
should know about affine and projective varieties, their associated coordinate
ring, their dimension, the Zariski topology, and basic properties thereof.
In Chapter 1 we define our main objects of study. The examples which
will guide us throughout the text are certain subgroups or quotient groups
of the isometry group of a finite-dimensional vector space equipped with a
bilinear or quadratic form. We state the important result which says that
any linear algebraic group is a closed subgroup of some group of invertible
matrices over our fixed field, which is nearly obvious for all of our examples
(the proof will be given in Chapter 5). In Chapter 2, we show that the
Jordan decomposition of a matrix results in a uniquely determined Jordan
decomposition of elements in a linear algebraic group. This in turn gives
us the notion of semisimple and unipotent elements in these groups. We
establish the important result that any group consisting entirely of unipotent
elements is conjugate to a subgroup of the upper unitriangular matrices.
Chapter 3 is devoted to the structure theory of commutative linear alge-
braic groups. In particular, Theorem 3.1 focuses attention on groups consist-
ing entirely of unipotent elements or of semisimple elements. Theorem 3.2
classifies the connected one-dimensional linear algebraic groups. While one
can say something about the structure of connected commutative groups
consisting entirely of unipotent elements, these will not play a role in this
text. Hence we turn at this point to commutative groups consisting entirely
of semisimple elements and introduce the notion of a torus, its character
2

group and its cocharacter group. These will play a crucial role in the classifi-
cation of semisimple groups. We turn in Chapter 4 to the structure theory of
connected solvable groups, for which the prototype is the group of upper tri-
angular invertible matrices. Indeed, the Lie–Kolchin theorem (Theorem 4.1
and Corollary 4.2) shows that any such group is isomorphic to a closed sub-
group of the group of upper triangular matrices. The importance of closed
connected solvable subgroups will become apparent in Chapter 6.
But before defining these so-called Borel subgroups, we must extend our
theory to cover group actions and in particular quotient groups; this is the
content of Chapter 5. The results on homogeneous spaces prepare the terrain
for establishing the main result of Chapter 6, the Borel fixed point theorem,
Theorem 6.1, some of whose many applications we discuss. We can also
finally define the radical of a linear algebraic group and establish its con-
nection with Borel subgroups. In these two chapters, we omit some essential
geometric arguments and notions. In particular, we do not prove results on
complete varieties but restrict ourselves only to projective varieties.
The last three chapters of this part are devoted to introducing the combi-
natorial data which classifies semisimple algebraic groups and to establish-
ing structural results and the classification theorem. The most important
ingredient of the data is a root system, which is obtained via the adjoint
representation of the group, acting on its tangent space; this theory is de-
scribed in Chapter 7. Theorem 8.17 is the main structural result on reductive
groups and we study in detail the case of the group SL2 in order to sketch a
proof of this result. The final chapter describes the classification of semisim-
ple algebraic groups in terms of the data mentioned above. We conclude by
explaining where our standard examples appear in this classification.
 1
Basic concepts

Throughout, k denotes an algebraically closed field of arbitrary characteris-

tic.

1.1 Linear algebraic groups and morphisms

Recall that a subset X of k n of the form
X = X(I) = {(x1 , . . . , xn ) ∈ k n | f (x1 , . . . , xn ) = 0 for all f ∈ I}
for some ideal I
k[T1 , . . . , Tn ] is called an algebraic set. Taking complements
of algebraic sets as open sets defines a topology on k n , the Zariski topology.
An affine algebraic variety is an algebraic set together with the induced
Zariski topology. (We will often omit the word “algebraic”.) For X ⊆ k n an
affine algebraic variety, let I
k[T1 , . . . , Tn ] denote the (radical) ideal of poly-
nomials vanishing identically on X. The quotient ring k[X] = k[T1 , . . . , Tn ]/I
is called the coordinate algebra or algebra of regular functions on X since it
can be naturally identified with the algebra of all polynomial functions on
X with values in k.
If X ⊆ k n , Y ⊆ k m are affine varieties, their cartesian product X × Y is
naturally an algebraic set in k n+m , hence possesses the structure of an affine
variety. We will always consider the product X ×Y equipped with the Zariski
topology, not with the product topology, which in general is different. Note
that k[X × Y ] ∼ = k[X] ⊗k k[Y ].
A map ϕ : X → Y between two affine varieties X, Y , which can be defined
by polynomial functions in the coordinates, is called a morphism of affine
varieties. Note that morphisms are continuous in the Zariski topology. A
morphism ϕ : X → Y induces functorially a k-algebra homomorphism ϕ∗ :
k[Y ] → k[X] via ϕ∗ (f ) := f ◦ ϕ for f ∈ k[Y ].
4 Basic concepts

X@
ϕ
/Y
@
@ f
ϕ∗ (f ) @ 
k

In fact, the above defines a contravariant equivalence between the category

of affine varieties with morphisms of varieties and the category of finitely
generated reduced k-algebras with k-algebra homomorphisms, the so-called
affine k-algebras, see [32, §1.5].
We can now define our main object of study.

Definition 1.1 A linear algebraic group is an affine algebraic variety

equipped with a group structure such that the group operations (multi-
plication and inversion)

µ : G × G −→ G, i : G −→ G,
(g, h) −→ gh, g −→ g −1 ,

are morphisms of varieties. (Recall our convention on the topology on G×G.)

Example 1.2 The base field k provides two natural examples of algebraic
groups:

(1) The additive group G = (k, +) of k is defined by the zero ideal I = (0)
in k[T ], and addition is given by a polynomial; hence G is an algebraic
group, with coordinate ring k[G] = k[T ]. The group G is called the
additive group, noted Ga .
(2) The multiplicative group G = (k × , ·) of k can be identified with the set
of pairs {(x, y) ∈ k 2 | xy = 1} (where multiplication is componentwise,
again given by polynomials), which is the algebraic set defined by the
ideal I = (XY − 1)
k[X, Y ]. So here k[G] = k[X, Y ]/(XY − 1) ∼ =
k[X, X −1 ]. The group G is called the multiplicative group and noted
Gm .

It is not immediately obvious from the above definition that the general
linear group
GLn := {A ∈ k n×n | det A = 0}

of invertible n × n-matrices over k is an algebraic group, since the determi-

nant condition is not a closed condition. But as for Gm above, GLn can be
identified with the closed subset

{(A, y) ∈ k n×n × k | det A · y = 1},

 1.1 Linear algebraic groups and morphisms 5

with componentwise multiplication, via A → (A, det A−1 ). Clearly multi-

plication is a polynomial map, and by Cramer’s rule, the same holds for
inversion. Thus GLn is a further (and very important) example of a linear
algebraic group. Its coordinate ring is given by
k[GLn ] = k[Tij , Y | 1 ≤ i, j ≤ n]/(det(Tij )Y − 1)
∼ k[Tij | 1 ≤ i, j ≤ n]det(T ) ,
= ij

the localization of k[Tij | 1 ≤ i, j ≤ n] at the multiplicative set generated by

det(Tij ). Note that GL1 = Gm .
Maps between linear algebraic groups should preserve not only the group
structure, but also the structure as an affine variety:

Definition 1.3 A map ϕ : G1 → G2 of linear algebraic groups is a mor-

phism of linear algebraic groups if it is a group homomorphism and also a
morphism of varieties, that is, the induced map ϕ∗ : k[G2 ] → k[G1 ] is a
k-algebra homomorphism.

Example 1.4 (1) If G ≤ GLn is a closed subgroup then the natural em-
bedding G → GLn is a morphism of linear algebraic groups.
(2) The determinant map det : GLn → Gm , A → det A, is a group homo-
morphism and clearly also a morphism of varieties, so a morphism of
algebraic groups.

Proposition 1.5 Kernels and images of morphisms of algebraic groups

are closed.

For the proof of the above statement, we will make use of the following
property of morphisms of varieties, which will also be used in subsequent
chapters (see [66, Thm. 1.9.5] or [26, Cor. 2.2.8]):

Proposition 1.6 Let ϕ : X → Y be a morphism of varieties. Then ϕ(X)

contains a non-empty open subset of ϕ(X).

Proof of Proposition 1.5 Let ϕ : G → H be a morphism of algebraic groups.

Since ker(ϕ) = ϕ−1 (1) and ϕ is a continuous map, ker(ϕ) is closed. By
Proposition 1.6, ϕ(G) contains a non-empty open subset of ϕ(G); then ϕ(G)
is closed, by Exercise 10.3(d).

It is clear that any closed subgroup of GLn inherits the structure of a

linear algebraic group. In fact, the converse is also true:

Theorem 1.7 Let G be a linear algebraic group. Then G can be embedded

as a closed subgroup into GLn for some n.
6 Basic concepts

The proof of this crucial characterization will be given as a corollary to

Theorem 5.5. For example, the map


1 c
Ga −→ GL2 , c→ ,
0 1
defines an embedding of the additive group Ga as a closed subgroup in GL2 ;
note that this map is in fact an isomorphism of algebraic groups onto its
image.

1.2 Examples of algebraic groups

We introduce some further important examples of linear algebraic groups
which will show up throughout the text. We start with three natural sub-
groups of GLn . Clearly the group of invertible upper triangular matrices
  
∗. ∗
Tn := .. invertible = {(aij ) ∈ GLn | aij = 0 for i > j},
0 ∗
its subgroup of upper triangular matrices with 1’s on the diagonal
 
1. ∗
Un := .. = {(aij ) ∈ Tn | aii = 1 for 1 ≤ i ≤ n},
0 1
and the group of diagonal invertible matrices
  
∗. 0
Dn := .. invertible = {diag(a1 , . . . , an ) | ai = 0 for 1 ≤ i ≤ n},
0 ∗
are closed subgroups of GLn , hence linear algebraic groups.
Recall that a group is called nilpotent if its descending central series de-
fined by
C 0 G := G, C i G := [C i−1 G, G] for i ≥ 1,
eventually reaches 1. It is not hard to see that Un is a nilpotent group,
with C n−1 (Un ) = 1. (One uses the filtration of Un by normal subgroups
Vm = {(aij ) ∈ Un | aij = 0 for 1 ≤ j − i ≤ m}, for 1 ≤ m ≤ n − 1.)
Furthermore, the derived series of a group G is defined by
G(0) := G, G(i) := [G(i−1) , G(i−1) ] for i ≥ 1.
If there exists some d with G(d) = 1, then G is called solvable, and the min-
imal such d is the derived length of G. Clearly G(i) ≤ C i G, so any nilpotent
group is solvable.
 1.2 Examples of algebraic groups 7

In our example, Tn is solvable, with T(1) n = Un . (Un is generated by

elementary matrices, all of which can be written as commutators.) We will
see later on (Corollary 4.2) that Tn is in some sense the prototype of a
connected solvable linear algebraic group.
We now define the various families of classical groups as groups of isome-
tries of non-degenerate bilinear or quadratic forms on finite-dimensional vec-
tor spaces. Recall that k is assumed to be algebraically closed.

The special linear groups

The special linear group
 
SLn := (aij ) ∈ k n×n | det(aij ) = 1

of n × n-matrices of determinant 1 is a closed subgroup of GLn , with coor-

dinate ring
k[SLn ] = k[Tij | 1 ≤ i, j ≤ n]/(det(Tij ) − 1).

As k is algebraically closed, we clearly have GLn = Z(GLn ) · SLn .

The symplectic groups  


0 Kn 0 . .1
For n ≥ 1 let J2n := where Kn := . . The symplectic
−Kn 0 1 0
group in dimension 2n is the closed subgroup
 
Sp2n = A ∈ GL2n | Atr J2n A = J2n

of GL2n ; so it is the group of invertible linear transformations of the even-

dimensional vector space k 2n leaving invariant the non-degenerate skew-
symmetric bilinear form with Gram matrix J2n (a so-called symplectic form).
Here, it is no longer so easy to explicitly write down the coordinate ring.
One can show that Sp2n is generated by transvections (see [79, 8.5]), and
hence Sp2n ≤ SL2n , and that for n = 1, any matrix of determinant 1 is
symplectic. So Sp2 = SL2 , while for all n ≥ 2, Sp2n is a proper subgroup of
SL2n .
The conformal symplectic group is the closed subgroup of GL2n defined as
 
CSp2n := A ∈ GL2n | Atr J2n A = cJ2n for some c ∈ k × ,

the group of transformations leaving J2n invariant up to a non-zero scalar.

It contains Sp2n as a closed normal subgroup.
8 Basic concepts

The odd-dimensional orthogonal groups

First assume that char(k) = 2. For n ≥ 1 the orthogonal group in (odd)
dimension 2n + 1 is defined by
 
GO2n+1 = A ∈ GL2n+1 | Atr K2n+1 A = K2n+1
with K2n+1 as above. Thus, this is the group of invertible linear transfor-
mations leaving invariant the non-degenerate symmetric bilinear form with
Gram matrix K2n+1 .
If char(k) = 2, skew-symmetric and symmetric bilinear forms coincide, and
the previous construction just yields the symplectic group in dimension 2n.
For arbitrary k the orthogonal groups have to be defined using the quadratic
form
f : k 2n+1 −→ k, f (x1 , . . . , x2n+1 ) := x1 x2n+1 +x2 x2n +· · ·+xn xn+2 +x2n+1 ,
on k 2n+1 associated to K2n+1 . The group of isometries
GO2n+1 = {A ∈ GL2n+1 | f (Ax) = f (x) for all x ∈ k 2n+1 }
of f is the odd-dimensional orthogonal group over k. (For char(k) = 2 this
defines the same group as before.)
Again there is a conformal version
 
CO2n+1 := A ∈ GL2n+1 | ∃c ∈ k × : f (Ax) = cf (x) for all x ∈ k 2n+1 ,
the odd-dimensional conformal orthogonal group, containing GO2n+1 as a
closed normal subgroup.

The even-dimensional orthogonal groups

For even dimension 2n ≥ 2 the orthogonal group is defined using the quadra-
tic form
f : k 2n −→ k, f (x1 , . . . , x2n ) := x1 x2n + x2 x2n−1 + · · · + xn xn+1 ,
on k 2n associated to K2n . The group of isometries
GO2n = {A ∈ GL2n | f (Ax) = f (x) for all x ∈ k 2n }
of f is the even-dimensional orthogonal group over k. For char(k) = 2 we can
also obtain this as the group of invertible linear transformations leaving in-
variant the non-degenerate symmetric bilinear form with Gram matrix K2n :
 
GO2n = A ∈ GL2n | Atr K2n A = K2n .
The even-dimensional conformal orthogonal group is defined as before as
 
CO2n := A ∈ GL2n | ∃c ∈ k × : f (Ax) = cf (x) for all x ∈ k 2n .
 1.3 Connectedness 9

Our choice of symmetric, skew-symmetric and quadratic forms above may

seem a bit arbitrary. In fact, any non-degenerate symmetric bilinear form
leads to the same group up to conjugacy, and similarly for non-degenerate
skew-symmetric bilinear forms, respectively quadratic forms (see for example
[2, §7]), but for the choices made above, certain natural subgroups have a
particularly nice shape, as will become apparent later.
As a final example, let G be a finite group. Then G has a faithful permuta-
tion representation, that is, there is an embedding G → Sn into a symmetric
group Sn for some n. Moreover, Sn → GLn via the natural permutation
representation. Combining these two homomorphisms we get an embedding
G → GLn whose image is a closed subgroup (i.e., the set of zeros of a finite
set of polynomial functions). Therefore, any finite group can be considered
as a linear algebraic group, with the discrete topology.

1.3 Connectedness
We now recall a topological notion which will play a crucial role in the study
of linear algebraic groups.

Definition 1.8 A topological space X is called irreducible if it cannot be

decomposed as X = X1 ∪ X2 where Xi is a non-empty proper closed subset
for i = 1, 2.

In view of the importance of this concept, we present some further ele-

mentary characterizations of irreducibility:

Proposition 1.9 The following are equivalent for an affine algebraic va-
riety X:

(i) X is irreducible.
(ii) Every non-empty open subset of X is dense.
(iii) Any two non-empty open subsets of X intersect non-trivially.
(iv) The vanishing ideal I of X is a prime ideal.
(v) k[X] is an integral domain.

Proof (i)⇔(ii): Indeed, if X1 ⊆ X is open then X = X 1 ∪ (X \ X1 ). Next,

(ii)⇔(iii) is obvious. The equivalence (i)⇔(iv) is shown in [32, Prop. 1.3C].
Finally, the equivalence of (iv) and (v) is well known.

Furthermore, we need the following basic properties (see [32, Prop. 1.3A,
1.3B and 1.4] and also Exercise 10.1):
10 Basic concepts

Proposition 1.10 Let X, Y be affine varieties. Then we have:

(a) A subset Z of X is irreducible if and only if its closure Z is irreducible.

(b) If X is irreducible and ϕ : X → Y is a morphism then ϕ(X) is irre-
ducible.
(c) If X, Y are irreducible then X × Y is irreducible.
(d) X has only finitely many maximal irreducible subsets Xi , and X = Xi .
In other words, every variety is a finite union of its maximal irreducible
subsets.

The maximal irreducible subsets in the preceding statement are called the
irreducible components of X. Note that by (a) irreducible components are
necessarily closed.

Definition 1.11 A topological space X is said to be connected if it cannot

be decomposed as a disjoint union X = X1  X2 , where the Xi ’s are non-
empty closed subsets.

Note that any irreducible set is connected; the converse is not true in
general. See Exercise 10.2.

Example 1.12 Let’s next look at some linear algebraic groups.

(1) Ga and Gm are connected by Proposition 1.9(v) since k[Ga ] = k[T ] and
k[Gm ] = k[T, T −1 ] are integral domains.
(2) GLn is connected since k[GLn ] = k[Tij ]det(Tij ) is an integral domain,
being a localization of the polynomial ring k[Tij ].

The next result gives a first example of how the Zariski topology on a linear
algebraic group allows one to deduce group theoretic structural results.

Proposition 1.13 Let G be a linear algebraic group.

(a) The irreducible components of G are pairwise disjoint, so they are the
connected components of G.
(b) The irreducible component G◦ containing 1 ∈ G is a closed normal sub-
group of finite index in G.
(c) Any closed subgroup of G of finite index contains G◦ .

Proof (a) Let X, Y be two irreducible components of G. Assume that g ∈

X∩Y . Since multiplication by g −1 is a morphism of G onto itself, g −1 X, g −1 Y
are irreducible by Proposition 1.10(b) and 1 ∈ g −1 X ∩ g −1 Y . Therefore,
without loss of generality we may assume that 1 ∈ X ∩ Y . Now, µ(X × Y ) =
XY is irreducible by Proposition 1.10(b), (c). As X = X · 1 ⊆ XY and
 1.3 Connectedness 11

Y = 1 · Y ⊆ XY , by the maximality of X, Y we get X = XY = Y .

Therefore, distinct components are disjoint.
(b) As G◦ is an irreducible component, so is (G◦ )−1 , and 1 ∈ G◦ ∩ (G◦ )−1 .
Thus, by part (a), G◦ = (G◦ )−1 and similarly G◦ · G◦ ⊆ G◦ , therefore G◦ is
a subgroup of G. For g ∈ G, g −1 G◦ g is again an irreducible component, as
an isomorphic image of G◦ and 1 ∈ G◦ ∩ g −1 G◦ g. Again by part (a) we get
G◦ = g −1 G◦ g, so G◦ is normal.
Let X be any irreducible component of G. If g ∈ X then 1 ∈ g −1 X and
so g −1 X = G◦ . It follows that X = gG◦ , that is, the components are cosets
of G◦ . Since by Proposition 1.10(d) there are finitely many components,
|G : G◦ | is finite.
(c) Let H ≤ G be a closed subgroup of finite index. Then H ◦ ≤ G◦ ≤ G.
Now, |G : H ◦ | = |G : H| · |H : H ◦ | is finite by part (b). Thus G◦ = gH ◦
is a finite disjoint union of the closed cosets of H ◦ and, being connected, is
equal to H ◦ . So G◦ ≤ H.
Thus for linear algebraic groups, the concepts of irreducibility and con-
nectedness coincide. We will henceforth refer to the connected (irreducible)
components of a linear algebraic group G as simply the components of G.
Example 1.14 (1) Let G be a linear algebraic group, H a proper closed
subgroup of finite index. Then G is not connected by Proposition 1.13(c).
In particular, for a finite algebraic group G we always have G◦ = 1.
(2) Recall the orthogonal group GO2n+1 from Section 1.2 and consider
the morphism det : GO2n+1 → Gm . Clearly, im(det) ⊆ {±1}. Since
−I2n+1 ∈ GO2n+1 we have im(det) = {±1}. Thus, if char(k) = 2,
then GO2n+1 is not connected by Proposition 1.10(b) since the im-
age is not connected. Our arguments show that in fact GO2n+1 ∼ =
ker(det) × −I2n+1 .
Similarly, one can show that GO2n is not connected in any character-
istic by exhibiting a closed subgroup of index 2. (See Exercise 10.8 for
char(k) = 2. In characteristic 2, the determinant has to be replaced by
the so-called pseudodeterminant, which we will not define here, see for
example [30, p. 124–131].)
Definition 1.15 For n ≥ 2, the special orthogonal group SOn := GO◦n is
the connected component of the identity in GOn .
It can be shown that when n is odd or char(k) = 2, SOn is the kernel of the
determinant, so SOn = GOn ∩ SLn ; when n is even and char(k) = 2, SOn is
the kernel of the pseudodeterminant. Thus SOn is of index 2 in GOn unless
n is odd and char(k) = 2 see for example [26, §1.7] or [30, §14]. Analogously,
12 Basic concepts

the even-dimensional conformal orthogonal group CO2n is not connected,

and we have |CO2n : CO◦2n | = 2.
The following fact from algebraic geometry allows one to establish the
connectedness of some algebraic groups (see [26, Thm. 2.4.6]):
Proposition 1.16 Let G be a linear algebraic group and ϕi : Yi → G,
i ∈ I, a family of morphisms from irreducible affine varieties Yi such that
1 ∈ Gi := ϕi (Yi ) for all i ∈ I. Then H := Gi | i ∈ I is a closed and
connected subgroup. Moreover, there exist n ∈ N and (i1 , . . . , in ) ∈ I n such
that H = G±1 ±1
i1 · · · Gin .

Example 1.17 (1) Let’s see an
application

 of the above criterion.

Clearly,
1 ∗ ∼ 1 0 ∼
SL2 = U2 , U− with U2 = = Ga , U −
2 = = Ga .
2
0 1 ∗ 1
Both U2 and U− 2 are closed and connected by Example 1.12(1); therefore
SL2 is connected by Proposition 1.16.
This argument can be extended to show that SLn is connected for
all n ≥ 2. Similarly, one sees that Tn , Un and Dn are connected for all
n ≥ 1 (see Exercise 10.7).
(2) This example shows that in general centralizers of elements need not
be connected, even in a connected
group.Let G = SL2 over a field of
1 1
characteristic char(k) = 2, and g = ∈ G. Then
0 1

  

a b  −1 0
CG (g) = a, b ∈ k, a 2
= 1 = H  H,
0 a  0 −1
with

  
1 b 
H := b ∈ k .
0 1 

Since H is a closed subgroup of index 2 in CG (g), this centralizer is

disconnected by Proposition 1.13(c).
As another application of our connectedness criterion we prove the follow-
ing result on commutators which will be needed later on:
Proposition 1.18 Let H, K be subgroups of a linear algebraic group G
where K is closed and connected. Then [H, K] is closed and connected.
Proof For h ∈ H, define ϕh : K → G by g → [h, g]. Since ϕh is a
composition of multiplication and inversion, it is a morphism. Moreover,
1 = ϕh (1) ∈ ϕh (K) for all h. Now, [H, K] = ϕh (K) | h ∈ H, therefore
[H, K] is closed and connected by Proposition 1.16.
 1.4 Dimension 13

In particular, for G a connected group the derived subgroup [G, G] and

more generally all terms in the derived series and in the descending central
series are closed and connected.

1.4 Dimension
Another fundamental invariant of algebraic varieties is their dimension.
For an irreducible variety X, the coordinate ring k[X] is an integral domain
by Proposition 1.9(v). Let k(X) be the field of fractions of k[X]. We define
the dimension of X by dim(X) := trdegk (k(X)), the transcendence degree of
k(X) over k. Equivalently, the dimension of X equals the maximal length of
descending chains of prime ideals in k[X]. If X is a reducible affine algebraic
variety, then according to Proposition 1.10(d) it can be decomposed as a
finite union X = X1 ∪ · · · ∪ Xt of its irreducible components Xi , and we set
dim(X) := max{dim(Xi ) | 1 ≤ i ≤ t}.
For G a linear algebraic group, dim(G) = dim(G◦ ) since G is the union
of the finitely many cosets gG◦ of the irreducible subgroup G◦ , by Propo-
sition 1.13, and dim(G) = dim(gG◦ ) = dim(G◦ ) by the above definition. In
particular, dim(G) = 0 if and only if G is a finite algebraic group.
Dimension behaves well with respect to morphisms in the following sense
(see [26, Cor. 2.2.9] or [32, Thm. 4.3]):

Proposition 1.19 Let ϕ : X → Y be a morphism of irreducible varieties

with ϕ(X) dense in Y . Then there exists a non-empty open subset U ⊆ Y
with U ⊆ ϕ(X) such that

dim ϕ−1 (y) = dim(X) − dim(Y ) for all y ∈ U.

In particular, for short exact sequences of linear algebraic groups this gives:

Corollary 1.20 Let ϕ : G1 → G2 be a morphism of linear algebraic groups.

Then
dim (im(ϕ)) + dim (ker(ϕ)) = dim(G1 ).

Proof Every fiber of ϕ is a coset of ker(ϕ), hence of the same dimension.

Now apply Proposition 1.19 to X = G◦1 , Y = im(ϕ)◦ .

Example 1.21 We compute the dimension of some linear algebraic groups.

(1) dim(Ga ) = 1, as k[Ga ] = k[T ] with field of fractions k(Ga ) = k(T ).

(2) dim(Gm ) = 1, since again k(Gm ) = k(T ).
14 Basic concepts

(3) dim(GLn ) = n2 since the field of fractions of k[GLn ] = k[Tij ]det(Tij ) is

just the rational function field k(Tij | 1 ≤ i, j ≤ n).
(4) dim(SLn ) = n2 − 1, by applying Corollary 1.20 to the (surjective) deter-
minant map det : GLn → Gm , with kernel SLn .
For inductive purposes, the following result is very helpful:
Proposition 1.22 If Y is a proper closed subset of an irreducible variety
X, then dim(Y ) < dim(X).
Proof Let Y1 ⊆ Y be an irreducible component. The embedding ϕ : Y1 →
Y → X induces a surjective homomorphism ϕ∗ : k[X] → k[Y1 ]. Since Y1 is
irreducible, k[Y1 ] is an integral domain by Proposition 1.9(v), so ker(ϕ∗ ) is
a prime ideal of k[X], non-zero since Y1 ⊂ X is proper. Now any chain of
prime ideals in k[Y1 ] can be lifted to a chain in k[X] through ker(ϕ∗ ), hence
of greater length since X is irreducible.
 2
Jordan decomposition

As a first step towards a structure theory of linear algebraic groups, we

investigate the properties of single elements.

2.1 Decomposition of endomorphisms

Recall the additive Jordan decomposition for endomorphisms: If V is a finite-
dimensional vector space over k and a ∈ End(V ) an endomorphism of V ,
then there exist unique s, n ∈ End(V ) such that s is semisimple, i.e., diag-
onalizable, n is nilpotent, a = s + n and sn = ns. Moreover, s = P (a) and
n = Q(a) for polynomials P, Q ∈ T · k[T ]. We now derive a multiplicative
version of this.
Definition 2.1 An endomorphism u ∈ End(V ) of a finite-dimensional
vector space V is called unipotent if u − 1 is nilpotent.
Equivalently, u is unipotent if and only if all its eigenvalues are equal to 1.
Note that over a field of characteristic char(k) = p > 0, u is unipotent if and
i i
only if it has p-power order, since then (u − 1)p = up − 1 for all i ≥ 0.
The additive Jordan decomposition can be used to obtain a multiplicative
version:
Proposition 2.2 For g ∈ GL(V ), there exist unique s, u ∈ GL(V ) such
that g = su = us, where s is semisimple and u is unipotent.
Proof Let s, n be the semisimple and nilpotent parts of g, as in the additive
Jordan decomposition above. Since g is invertible, so is s and we may set
u = 1 + s−1 n. As n is nilpotent and sn = ns, s−1 n = u − 1 is also nilpotent.
Therefore u is unipotent and su = s(1 + s−1 n) = s + n = g. If g = su is any
such decomposition, where u = 1 + n with n nilpotent and commuting with
16 Jordan decomposition

s, then g = s + sn is the unique additive Jordan decomposition of g, so s

and u are uniquely determined.

Definition 2.3 Let a ∈ GL(V ). We call s, respectively n, u as above the

semisimple, respectively nilpotent, unipotent part of a.

In order to transfer this definition to an arbitrary linear algebraic group

G we embed it as a closed subgroup in some GL(V ) (see Theorem 1.7). We
will first need a result on the “local finiteness” of certain automorphisms
of k[G]. Each x ∈ G defines a morphism G → G given by g → gx. We
denote the corresponding k[G]-algebra homomorphism by ρx : k[G] → k[G],
so ρx (f )(g) = f (gx) for f ∈ k[G], g ∈ G. This defines an action of G as
abstract group on k[G]. We then have the following:

Proposition 2.4 Let G be a linear algebraic group and V a finite-dimen-

sional subspace of k[G]. Then there exists a finite-dimensional G-invariant
subspace X of k[G] containing V . In particular, k[G] is a union of finite-
dimensional G-invariant subspaces. Moreover, the restriction to any such
finite-dimensional subspace X affords a morphism of algebraic groups ρ :
G → GL(X).

Proof It suffices to prove the statement for the case V = f , a one-dimen-

 
sional subspace of k[G]. Let µ∗ (f ) = i∈I fi ⊗gi , so that ρx (f ) = i gi (x)fi .
Hence the finite-dimensional subspace generated by the {fi | i ∈ I} contains
ρx (f ), for all x ∈ G. The subspace X generated by {ρx (f ) | x ∈ G} is
therefore a finite-dimensional G-invariant subspace of k[G]. It is clear from
the above construction that the coordinates of ρx are polynomial functions
in x. Hence the map x → (ρx )|X affords a morphism of algebraic groups
G → GL(X).

We can now return to the Jordan decomposition in G. Given an arbitrary

vector space V over k, we say that an element x ∈ End(V ) is locally finite
if V is a union of finite-dimensional x-stable subspaces and we say that x
is locally semisimple, respectively locally nilpotent, if its restriction to any
finite-dimensional x-stable subspace is semisimple, respectively nilpotent.
One can then show that any locally finite endomorphism of V has a Jordan
decomposition as above. Moreover, by Proposition 2.4, we see that ρx is a
locally finite endomorphism of k[G], for all x ∈ G. We are now ready to
prove:

Theorem 2.5 (Jordan decomposition) Let G be a linear algebraic group.

(a) For any embedding ρ of G into some GL(V ) and for any g ∈ G, there
 2.1 Decomposition of endomorphisms 17

exist unique gs , gu ∈ G such that g = gs gu = gu gs , where ρ(gs ) is

semisimple and ρ(gu ) is unipotent.
(b) The decomposition g = gs gu = gu gs is independent of the chosen embed-
ding.
(c) Let ϕ : G1 → G2 be a morphism of algebraic groups. Then ϕ(gs ) = ϕ(g)s
and ϕ(gu ) = ϕ(g)u .
Sketch of proof For g ∈ G, ρg is an invertible locally finite linear trans-
formation of k[G]. Since ρg is an algebra homomorphism, one can deduce
that (ρg )s and (ρg )u are also algebra homomorphisms. In particular, f →
((ρg )s (f ))(1) defines a homomorphism k[G] → k, that is, a point gs in G,
and similarly for (ρg )u . It remains to show that (ρg )s = ρgs and (ρg )u = ρgu .
See [66, 2.4.8] for the details.
We give a proof of (b) and (c) only for the case k = Fp : We may assume
that G ≤ GLn (Fp ). As Fp is the union ∪i≥1 Fpi of finite fields, any g ∈
G lies in GLn (Fq ) for some power q of p, so has finite order. Then gs is
diagonalizable, hence of order prime to p, while gu has only eigenvalues 1, so
is of p-power order. Since g = gs gu it follows that gs is the p -part of g and
gu is the p-part of g. By the uniqueness of the p- and p -parts of elements of
finite groups, the claim follows.
See [32, Thm. 15.3] for a proof in the general case.
Definition 2.6 Let G be a linear algebraic group. The decomposition
g = gs gu = gu gs in Theorem 2.5 is called the Jordan decomposition of g ∈ G,
and g is called semisimple, respectively unipotent, if g = gs , respectively
g = gu . We write
Gu := {g ∈ G | g is unipotent},
Gs := {g ∈ G | g is semisimple}.
for the subsets of unipotent, respectively semisimple elements of G. If G
consists entirely of unipotent elements then we say that G is a unipotent
group.
We note that the term “semisimple group” is reserved for a different con-
cept; see Definition 6.14.
Example 2.7 We have already encountered groups with G = Gu and with
G = Gs :

  
1 a 
(1) Ga ∼= a ∈ k is a unipotent group and, more generally, so
0 1 
is Un for all n ≥ 2.
18 Jordan decomposition

(2) Gm , and more generally Dn = {diag(t1 , . . . , tn ) | ti = 0} for n ≥ 1,
consist of semisimple elements.

Let G be a linear algebraic group. Then the set Gu of unipotent elements

is closed (see Exercise 10.12), while Gs is, in general, neither closed nor open.
For example, in GLn the set of semisimple elements is dense:

Example 2.8 We will observe that Gs is

“badly
 behaved”. Let G = GL2
a 1
and g ∈ G \ Gs . Then g is conjugate to for some a ∈ k × . Define
0 a


at 1
ϕ : Gm → GL2 by ϕ(t) = . Then ϕ is a morphism and ϕ(t)
0 at−1
−1
is semisimple for at = at , hence for t = ±1. So, ϕ(Gm \ {±1}) ⊆ Gs .
But Gm \ {±1} is open in Gm , and dense by Proposition 1.9(ii) since Gm is
irreducible. Hence Gm \ {±1} = Gm and

ϕ(Gm ) = ϕ(Gm \ {±1}) ⊆ ϕ(Gm \ {±1}) ⊆ Gs ,

so all non-semisimple elements of G lie in Gs . Hence Gs = G.

2.2 Unipotent groups

We now prove the basic structure result for unipotent groups. Recall from
Theorem 1.7 that any linear algebraic group has an embedding into some
GLn , so we may and will restrict attention to unipotent matrix groups.

Proposition 2.9 Let G ≤ GLn be a unipotent group. Then there exists

g ∈ GLn such that g −1 Gg ≤ Un .

Proof Write V = k n . We will argue by induction on n.

The only unipotent element of GL1 is the identity, so the claim is clear
for n = 1. Now suppose that n > 1. If there exists a G-invariant proper
subspace 0 =

W < V, then by choosing an appropriate basis we may assume
∗ ∗
that G ≤ .
0 ∗
The G-invariance of W induces natural homomorphisms ϕ : G → GL(W )
and Φ : G → GL(V /W ). Since dim(W ) and dim(V /W ) are both smaller than
 2.2 Unipotent groups 19

dim(V ), by induction we get (up to a change of basis, so up to conjugation)

⎧⎛ ⎞⎫
⎪
⎪ 1. ∗ ⎪
⎪
⎪
⎪⎜ . . ∗ ⎟⎪ ⎪
⎨ ⎜ 0 1 ⎟⎬
G≤ ⎜ ⎜
⎟ = Un
⎪
⎪⎝ 1. ∗ ⎟ ⎪
⎪
⎪
⎩ 0 . . ⎠⎪⎪
⎪
0 1 ⎭
as claimed.
If, on the other hand, G acts irreducibly on V , then the elements of G
generate the full endomorphism algebra End(V ) by Burnside’s double cen-
tralizer theorem [39, Thm. 1.16]. Let g ∈ G. Since any element of G, being
unipotent, has trace n we find tr((g − 1)h) = tr(gh) − tr(h) = 0 for all h ∈ G.
Therefore, tr((g − 1)x) = 0 for all x ∈ End(V ). Choosing for x matrices with
only one non-zero entry one easily sees that this is only possible if g − 1 = 0,
that is g = 1 and so G = 1, contradicting the irreducibility of G on V .
As Un is a nilpotent group (see Section 1.2) Proposition 2.9 immediately
implies the following:
Corollary 2.10 A unipotent linear algebraic group is nilpotent, hence solv-
able.
Over k = Fp , unipotent elements are p-elements, thus the above statement
can be seen as an algebraic group analogue of the fact that finite p-groups
are nilpotent.
 3
Commutative linear algebraic groups

The Jordan decomposition of elements leads to a similar decomposition of

abelian algebraic groups. The case when all elements are semisimple will be
of particular importance to us later.

3.1 Jordan decomposition of commutative groups

The structure of commutative algebraic groups is described by the following
result:

Theorem 3.1 Let G be a commutative linear algebraic group.

(a) The sets Gs and Gu are closed subgroups of G.

(b) The product map π : Gu × Gs → G is an isomorphism of algebraic
groups.
(c) If G is connected then so are Gs and Gu .

Proof (a) Choose an embedding of G into some GL(V ) according to The-

orem 1.7. Let g, g  ∈ G be two (commuting) elements. Commuting en-
domorphisms can be simultaneously trigonalized, so if g, g  are semisimple
(unipotent) so are gg  and g −1 . Therefore Gs , Gu are subgroups of G.
Note that Gu is closed by Exercise 10.12. Also, Gs consists of commuting
semisimple endomorphisms, so can be simultaneously diagonalized. Let Vi ,
1 ≤ i ≤ t, denote the various common eigenspaces of Gs on V , that is, Vi =
{v ∈ V | gv = ϕi (g)v for all g ∈ Gs } for suitable morphisms ϕi : Gs → Gm ,
t
and V = i=1 Vi . Since Gu commutes with Gs , it preserves all Vi . So G
stabilizes each Vi , therefore by Proposition 2.9 there exists a basis such that
Gu acts by upper unitriangular matrices on each Vi , while Gs acts by scalars.
 3.1 Jordan decomposition of commutative groups 21

So with respect to this basis, G ≤ Tn and Gs = Dn ∩ G, where n := dim(V ).

Hence Gs is closed.
(b) The product map π : Gs × Gu → G, π(s, u) = su, is a group homo-
morphism since G is commutative. It is surjective by Jordan decomposition
(Theorem 2.5) and injective by the uniqueness in Theorem 2.5. Moreover, π
is a morphism of varieties since multiplication µ : G × G → G is a morphism.
It remains to show that the inverse map
π −1 : G −→ Gs × Gu , π −1 (g) = (gs , gs−1 g),
is also a morphism. For this it suffices to notice that the map
   
d1 ∗ d1 0
ψ : g → gs , that is . .. −→ . .. ,
0 dn 0 dn
is a morphism.
(c) As G is assumed connected and G → Gs , g → gs , is a morphism, its
image Gs is connected by Proposition 1.10(b). Similarly G → Gu , g → gu =
ggs−1 , is a morphism so has a connected image.
Compare this again to the situation for finite groups: a finite abelian group
is the direct product of its Sylow subgroups.
The structure of one-dimensional connected groups is very restricted (see
[66, Thm. 3.4.9]):
Theorem 3.2 If G is a connected linear algebraic group of dimension 1
= Ga or G ∼
then G ∼ = Gm .
About the proof The proof of this seemingly innocent result is rather diffi-
cult, in particular the case where G consists of unipotent elements. We only
prove here that G is commutative and either G = Gs or G = Gu .
First, we claim that G is commutative. For g ∈ G define a morphism
ϕg : G → G by ϕg (x) := x−1 gx. As ϕg (G) is irreducible, ϕg (G) is also an
irreducible closed subvariety of G. So, either, ϕg (G) is of dimension 1 and
then ϕg (G) = G by Proposition 1.22, or ϕg (G) is of dimension 0.
If ϕg (G) is of dimension 0 for all g ∈ G then ϕg (G) = {g} by connectedness
and hence G is commutative.
So we may assume that ϕg (G) is dense in G for some g ∈ G. We fix an
embedding of G in some GLn . Now ϕg (G) contains a non-empty open subset
of its closure by Proposition 1.6. Thus G \ ϕg (G) is contained in a closed
proper subset, so of dimension 0 by Proposition 1.22, and hence finite.
As h ∈ ϕg (G) has characteristic polynomial equal to the characteristic
polynomial of g and there is only a finite number of elements in G \ ϕg (G),
22 Commutative linear algebraic groups

there exist finitely many characteristic polynomials p1 , . . . , pt for elements

of G. Now
t
G= ((zeros of pi ) ∩ G)
i=1

is a finite union of closed sets. As G is connected, we get t = 1 and hence

p1 = (T − 1)n (for 1 ∈ G). Therefore G is unipotent, and, by Corollary 2.10,
solvable. It follows from Proposition 1.18 that [G, G] is a closed connected
proper subgroup of G, so [G, G] = 1 by Proposition 1.22, and G is commu-
tative, a contradiction.
By Theorem 3.1, Gs and Gu are closed connected subgroups of G. If
either is proper, then it is of dimension 0 by Proposition 1.22, hence trivial
by connectedness. Thus either G = Gu or G = Gs as claimed.

See Definition 3.3 and Proposition 3.9 for the remaining step in the case
that G = Gs .

3.2 Tori, characters and cocharacters

We now turn to the study of connected groups containing only semisimple
elements. By Theorem 3.2, any such group of dimension 1 is isomorphic
to Gm . This motivates the following definition:

Definition 3.3 A linear algebraic group is called a torus if it is isomorphic

to a direct product Gm × · · · × Gm of copies of Gm , that is, to Dn for some
n ≥ 0.

Tori are best studied via their characters:

Definition 3.4 Let G be a linear algebraic group. A character of G is a

morphism of algebraic groups χ : G → Gm . The set of characters of G is
denoted by X(G). Note that it can naturally be considered as a subset of
k[G].
A cocharacter of G is a morphism of algebraic groups γ : Gm → G. The
set of cocharacters of G is denoted by Y (G).

Clearly, X(G) is an abelian group (usually written additively) with respect

to
(χ1 + χ2 )(g) := χ1 (g)χ2 (g) for χ1 , χ2 ∈ X(G), g ∈ G.

Similarly, if G is commutative then Y (G) is an abelian group with respect

 3.2 Tori, characters and cocharacters 23

to
(γ1 + γ2 )(x) := γ1 (x)γ2 (x) for γ1 , γ2 ∈ Y (G), x ∈ Gm .

Example 3.5 (Characters and cocharacters of Dn )

Let G = Dn , an n-dimensional torus. If n = 1 then G = Gm , so X(G) =
End(Gm ) = {t → tj | j ∈ Z} ∼ = Z by Exercise 10.11.
For arbitrary n, define χi ∈ X(G) by χi (g) := ti if g = diag(t1 , . . . , tn ) ∈
G, 1 ≤ i ≤ n. Then any χ ∈ X(G) can be uniquely decomposed as χ =
χa1 1 · · · χann for some ai ∈ Z, so X(Dn ) ∼ = Zn . Note that here the characters
±1
are just the monomials in k[Dn ] = k[T1 , . . . , Tn±1 ], and hence form a basis
of the coordinate ring.
Similarly, for (a1 , . . . , an ) ∈ Zn the map Gm → Dn , t → diag(ta1 , . . . , tan ),
is a morphism, hence a cocharacter. Conversely, by restricting a cocharacter
γ : Gm → Dn to the ith diagonal position one sees that any cocharacter has
the above form. Therefore Y (Dn ) ∼ = Zn .
As a result, for an n-dimensional torus T we have X(T ) ∼ = Y (T ) ∼ = Zn .
Now, for χ ∈ X := X(T ), γ ∈ Y := Y (T ), the composite

χ ◦ γ : Gm → Gm , t → χ(γ(t)),

is an endomorphism of Gm . So by Exercise 10.11 it acts by raising to some

integer power χ, γ ∈ Z, that is, χ(γ(t)) = tχ,γ .

Proposition 3.6 Let T be a torus with character group X, cocharacter

group Y . The map  ,  : X × Y → Z is a perfect pairing between X and Y ,
that is, any homomorphism X → Z is of the form χ → χ, γ for some γ ∈ Y ,
and any homomorphism Y → Z is of the form γ → χ, γ for some χ ∈ X.
This defines group isomorphisms Y ∼= Hom(X, Z) and X ∼ = Hom(Y, Z).

Proof We may assume that T = Gm × · · · × Gm (n copies) and X =

{χa1 1 . . . χann | (a1 , . . . , an ) ∈ Zn } with χi as defined above. Any homomor-
phism ϕ : X → Z is uniquely determined by ϕ(χi ) =: di ∈ Z since
 n 
! "
n "
n
ai
ϕ χi = ai ϕ(χi ) = a i di .
i=1 i=1 i=1
⎛ ⎞
td 1 0
Let γ ∈ Y with γ(t) := ⎝ ..
.
⎠ for t ∈ Gm . Then
0 td n
!
n
n
tχ,γ = χ(γ(t)) = χai i (γ(t)) = t i=1 ai d i
= tϕ(χ) .
i=1
24 Commutative linear algebraic groups

Therefore, ϕ(χ) = χ, γ for all χ ∈ X.

Thus the map Y → Hom(X, Z), γ → −, γ, is surjective. For injectivity,
assume χ, γ = 0 for all χ ∈ X, then

tdi = χi (γ(t)) = tχi ,γ = t0 = 1

for all t ∈ Gm and all i. Thus di = 0 for all i, that is, γ = 0. Hence
Y ∼= Hom(X, Z). Similarly one argues that X ∼
= Hom(Y, Z).
Definition 3.7 Let T be a torus. For a closed subgroup H ≤ T we define

H ⊥ := {χ ∈ X(T ) | χ(h) = 1 for all h ∈ H},

a subgroup of X(T ). Similarly, for a subgroup X1 ≤ X(T ) we let

#
X1⊥ := {t ∈ T | χ(t) = 1 for all χ ∈ X1 } = ker(χ),
χ∈X1

a closed subgroup of T .

The following properties of these constructions will be needed later (see

Exercise 10.15 for a proof):

Proposition 3.8 Let T be a torus.

(a) For any closed subgroup H ≤ T , restriction defines an isomorphism

X(T )/H ⊥ ∼
= X(H).

(b) For any subgroup X1 ≤ X(T ), X1⊥⊥ /X1 is a finite p-group, where p =
char(k); in particular, X1⊥⊥ = X1 if X(T )/X1 has no p-torsion.

The following can be shown using properties of characters (see [32, §16]
or [66, 3.2.7]):

Proposition 3.9 Any closed connected subgroup of Dn is a torus.

This yields the missing part in the proof of Theorem 3.2 in case G = Gs :
embed G into some GLn and diagonalize.
The following crucial result shows the “rigidity” of tori: their normaliz-
ers are only a finite bit larger than their centralizers. It will later allow us
to define the Weyl group, a finite group which plays an important role in
controlling the structure of semisimple groups (see [66, Thm. 3.2.9] for a
proof):

Theorem 3.10 Let G be a linear algebraic group and T ≤ G be a torus.

Then, NG (T )◦ = CG (T )◦ , and NG (T )/CG (T ) is finite.
 3.2 Tori, characters and cocharacters 25

Example 3.11 (To illustrate Theorem 3.10) Let G = GLn , m ≤ n, and

⎧⎛ ⎞⎫
⎪
⎪ ∗. ⎪
⎪
⎨⎜ . . ⎟⎬
⎜ ∗ ⎟
T = ⎝
⎪ 1 . ⎠⎪ = {diag(t1 , . . . , tn ) | ti = 1 for m + 1 ≤ i ≤ n},
⎪
⎩ .. ⎪ ⎭
1
an m-dimensional torus in G, which is isomorphic to Gm × · · · × Gm (m
copies). Then
⎧⎛ ⎞ ⎫

⎨ ∗. .
⎪  ⎪
⎬
⎜ ⎟
⎠  A ∈ GLn−m ∼
. 0
CG (T ) = ⎝ ∗ = Dm × GLn−m ,
⎪
⎩  ⎪
⎭
0 A 

and NG (T ) ∼
= NGLm (Dm ) × GLn−m where NGLm (Dm ) = M is the set of all
monomial m × m-matrices. As Dm has finite index in M , M ◦ = Dm . Also,
GLn−m is connected, so we get
NG (T )◦ ∼
= Dm × GLn−m ∼
= CG (T ) = CG (T )◦ ,
and NG (T )/CG (T ) ∼
= M/Dm ∼
= Sm is a symmetric group, hence indeed
finite.
 4
Connected solvable groups

We extend our structure results from commutative to solvable groups G. For

this, however, connectedness of G is crucial.

4.1 The Lie–Kolchin theorem

If G ≤ GL(V ) is unipotent, then by Proposition 2.9 it always stabilizes a
one-dimensional subspace of V . This is true for a larger class of groups.
Recall from Proposition 1.18 that commutator subgroups are closed and
connected if the original group is.

Theorem 4.1 (Lie–Kolchin) Let G be a connected solvable subgroup of

GL(V ), V = 0. Then G has a common eigenvector on V , that is, there
exists a one-dimensional subspace of V stabilized by G.

Proof We may assume that G is closed since if G is solvable then so is its

closure G ([4, §2.4]).
We will use induction on n = dim V and on the derived length d of G (see
Section 1.2).
For n = 1 the result is trivially true. Now assume that n > 1. If d = 1 then
G is commutative and the result follows by the proof of Theorem 3.1(a).
Now assume that d ≥ 2. If there exists a G-invariant proper
subspace

∗ ∗
0 = W < V of V , choose an appropriate basis such that G ⊆ .
0 ∗


ϕ(g) ∗
Then more precisely, any g ∈ G can be represented by
0 ψ(g)
where ϕ : G → GL(W ) is the canonical restriction morphism, and ψ : G →
GL(V /W ). Now, ϕ(G) is solvable, connected, acting on a vector space of
 4.2 Structure of connected solvable groups 27

smaller dimension dim W < dim V , so by the inductive hypothesis there

exists a common eigenvector w ∈ W < V .
Therefore, we may assume that G acts irreducibly on V . As G := [G, G]
is a closed, connected solvable subgroup of derived length d − 1, by induction
there exists a common eigenvector v ∈ V of G . Since G
G, gv is again a
common eigenvector of G for all g ∈ G.
Let W denote the span of all common eigenvectors of G . By the observa-
tion above, W is G-invariant and W = 0. As G acts irreducibly, necessarily
W = V . Therefore, V has a basis consisting of common eigenvectors of G
and so G acts diagonally, implying that G is commutative.
Now for fixed y ∈ G all conjugates x−1 yx by elements x ∈ G lie in
G , hence act diagonally with the same eigenvalues as y. Therefore, there
are only finitely many possibilities for x−1 yx. It follows that the morphism
ϕy : G → G , x → x−1 yx, has finite image. So, by the connectedness of G,
ϕy (G) = {y}, implying G ≤ Z(G). Since G acts irreducibly on V , Z(G)
acts by scalars by Schur’s lemma. As all elements of G have determinant 1,
it follows that G is finite. But since it is connected, G = 1 and hence G is
commutative, in contradiction to our assumption that d ≥ 2.

Corollary 4.2 Let G be a connected, solvable subgroup of GLn . Then, G

is conjugate to a subgroup of Tn .

Proof Apply Theorem 4.1 to inductively construct a basis with respect to

which G is triangular.

Remark 4.3 The assertion of the above corollary becomes false if G is

no longer assumed to be connected. Indeed, consider any finite non-trivial
solvable subgroup of GLn , n > 1, which acts irreducibly.

4.2 Structure of connected solvable groups

We have seen that connected solvable groups are isomorphic to subgroups of
Tn . This has strong implications for the structure of such groups. We have
a natural split exact sequence
π
1 −→ Un −→ Tn −→ Dn −→ 1

where π is the natural morphism with

   
t1 . ∗ t1 . 0
π .. = .. .
0 tn 0 tn
28 Connected solvable groups

This sequence is also split as an exact sequence of algebraic groups; the

inclusion Dn → Tn is a morphism and a section to π.
Now let G ≤ Tn be a closed connected subgroup. The restriction of π
to G has kernel Gu = G ∩ Un , the closed normal subgroup of G consisting
of all unipotent elements. The image T := π(G) is then a closed connected
subgroup of Dn , hence a torus by Proposition 3.9. So restriction yields the
exact sequence
π
1 −→ Gu −→ G −→ T −→ 1.

Since the torus T is abelian, it follows that [G, G] ≤ Gu .

This almost proves the first part of the following structure result for con-
nected solvable groups:

Theorem 4.4 Let G be a connected, solvable linear algebraic group. Then:

(a) Gu is a closed, connected, normal subgroup of G and [G, G] ≤ Gu .

(b) All maximal tori of G are conjugate and if T is any such maximal torus,
then G = Gu
T and NG (T ) = CG (T ).

Here a maximal torus of G is a subtorus of G which is maximal with respect

to inclusion. Note that by the theorem, maximal tori are all of the same
dimension, hence of maximal possible dimension. The semidirect product
of two algebraic groups X and N is defined as for abstract groups, where
we require that X act as a group of algebraic group automorphisms of N .
An algebraic group H is isomorphic to the semidirect product of two closed
subgroups X and N if N  H, N ∩ X = 1 and the product map N
X → H
is an isomorphism of algebraic groups.

Proof of part (a) It remains to show that Gu is connected. Set G̃ := G/G ,

where G = [G, G]. We will see later (in Proposition 5.7) that this factor
group by a closed normal subgroup has a natural structure of linear algebraic
group in such a way that the natural surjection ϕ : G → G̃ is a morphism
of linear algebraic groups.
Since G̃ is commutative and connected, G̃ = G̃u × G̃s with G̃u con-
nected by Theorem 3.1. Then, ϕ(Gu ) ≤ G̃u by Theorem 2.5, and there-
fore, Gu ≤ ϕ−1 (G̃u ). On the other hand, by our above considerations before
Theorem 4.4, ker(ϕ) = G ≤ Gu . If g ∈ ϕ−1 (G̃u ), with Jordan decomposi-
tion g = gs gu , then ϕ(gs ) = ϕ(g)s = 1. Therefore, gs ∈ ker(ϕ) ≤ Gu , so
gs = 1, g = gu and it follows that ϕ−1 (G̃u ) = Gu . Now G = ker(ϕ) ≤ Gu is
connected by Proposition 1.18, so contained in G◦u . Thus

|Gu : G◦u | = |Gu /G : G◦u /G | = |ϕ(Gu ) : ϕ(G◦u )| = |G̃u : ϕ(G◦u )|.
 4.2 Structure of connected solvable groups 29

Hence ϕ(G◦u ) is a closed subgroup of finite index in the connected group G̃u ,
so of index 1. Hence, Gu is also connected.
The proof of part (b) is more difficult and requires results about actions
of diagonalizable groups (see [32, Thm. 19.3]).
Finally, the assertion on the normalizer follows from the Frattini argument:
if g = ut ∈ NG (T ), with u ∈ Gu and t ∈ T , then uT u−1 = T . But, for s ∈ T ,
usu−1 = (u su−1 s−1 )s lies in T if and only if u su−1 s−1 = 1, that is, u
centralizes s.
Note the similarity of the second part with the Schur–Zassenhaus The-
orem for finite groups ([2, §18.1]). Indeed, G is already known to be an
extension of the unipotent group Gu by a torus π(G), all of whose elements
are semisimple. The theorem asserts that a complement to Gu exists and
that furthermore all complements are conjugate. If k is the algebraic clo-
sure of a finite field Fp , then unipotent elements have p-power order, while
semisimple elements have order prime to p. Thus Gu is the analogue of a
normal Sylow p-subgroup of G, and any maximal torus is a p -complement.
In that situation, the assertion in (b) is just as in the Schur–Zassenhaus
Theorem.
If G is not connected, anything may happen. For example, take G =
NGL2 (D2 ), an extension of D2 by an element of order 2, and assume that
char(k) = 2. Then, although G is solvable, Gu is not a subgroup, in particular
not normal, and the maximal torus D2 is not self-normalizing.
Corollary 4.5 Let G be a connected, solvable linear algebraic group. Then
any semisimple element of G lies in a maximal torus and any unipotent
element of G lies in a connected unipotent subgroup of G.
The result for unipotent elements follows from Theorem 4.4(a); for semi-
simple elements it follows from the omitted proof of Theorem 4.4(b).
Finally, we need a result on centralizers of tori in connected solvable groups
which will later on be generalized to arbitrary connected groups (see Theo-
rem 6.19) (for the proof see [66, Cor. 6.3.6]):
Proposition 4.6 Let G be connected solvable, S ≤ G a torus. Then CG (S)
is connected.
 5
G-spaces and quotients

One aspect of the theory of linear algebraic groups which has been missing
up to now is that of a quotient group. We need to first see how to give the
structure of variety to a quotient and it will become clear that we cannot
limit ourselves to affine varieties. Thus, we begin by recalling some basic
aspects of the general theory of varieties and morphisms.

5.1 Actions of algebraic groups

In group theory, it is often helpful to consider actions of groups, for example
the action of a group on itself by conjugation. We will find it necessary to
consider actions of linear algebraic groups on affine and projective varieties.
For this recall that projective n-space Pn may be defined as the set of
equivalence classes of k n+1 \ {(0, 0, . . . , 0)} modulo the diagonal action of
k × by multiplication. Taking common zeros of a collection of homogeneous
polynomials in k[T0 , T1 , . . . , Tn ] as closed sets defines a topology on Pn . A
projective variety is then a closed subset of Pn equipped with the induced
topology.
The k-algebra of regular functions on an affine variety here needs to be
replaced by a sheaf of functions, as follows. First, for X an irreducible affine
variety and x ∈ X, let I(x)
k[X] be the ideal of functions vanishing at
x and let Ox be the localization of k[X] with respect to the prime ideal
$
I(x). Then setting OX (U ) = x∈U Ox , for U ⊆ X open, defines a sheaf of
functions OX on X. Clearly the fields of fractions of OX (U ) and k[X] agree.
More generally, if X is reducible, say X = X1 ∪ . . . ∪ Xt with irreducible
components Xi , then setting

OX (U ) := {f : U → k | f |U ∩Xi ∈ OXi (U ∩ Xi )},

 5.1 Actions of algebraic groups 31

for U ⊆ X open, defines a sheaf on X. One can show that OX (X) = k[X],
the usual k-algebra of regular functions on X.
n
Now consider the covering Pn = i=0 Ui , where Ui consists of the points
in Pn with non-zero ith homogeneous coordinate. Then Ui can be identified
with affine n-space k n via


x0 xi−1 xi+1 xn
(x0 , x1 , . . . , xn ) → ,..., , ..., .
xi xi xi xi
Moreover, the induced topology on the sets Ui is precisely the topology of
the affine space k n . For each x ∈ Ui , we have the ring of functions Ox as
above, using the identification of Ui with k n . Then for U ⊆ Pn , set O(U ) :=
$
x∈U Ox . The sheaf of functions for an arbitrary projective variety is defined
by restriction of the given sheaf on Pn . It is then clear that the only globally
defined functions on Pn are the constant functions. This is more generally
true for an arbitrary projective variety.
The dimension of an irreducible projective variety is the dimension of any
affine open subset. For a general projective variety, one simply takes the
maximum dimension of the irreducible components of the variety. (See [66,
Prop. 1.2.4, §1.8.1].)
Now a morphism of varieties is a continuous map ϕ : X → Y between two
(affine or projective) varieties such that for all V ⊆ Y open and U = ϕ−1 (V )
and all f ∈ OY (V ), we have f ◦ ϕ ∈ OX (U ).
Henceforth, we will include projective varieties when we consider a general
variety, though we continue to limit ourselves to linear algebraic groups,
which are always affine varieties. We point out that Propositions 1.6, 1.19,
and 1.22 remain valid in the setting of projective varieties.
Important examples of projective varieties are provided by partial flag
varieties of an n-dimensional vector space V over k. Recall that a flag in
V is a chain of subspaces 0  V1  · · ·  Vr = V . One can equip the
set of flags having a fixed sequence of dimensions (dim V1 , . . . , dim Vr ) with
the structure of a projective variety, the partial flag variety. The case where
dim Vi = i for all i is said to be the flag variety of V . (See for example [26,
§3.3] for details.)

Definition 5.1 Let G be a linear algebraic group. A variety X is a G-space

if there exists a group action G × X → X, (g, x) → g.x, of G on X which
is also a morphism of varieties. Let X and Y be G-spaces. A morphism of
G-spaces is a morphism of varieties ϕ : X → Y such that ϕ(g.x) = g.ϕ(x)
for all g ∈ G, x ∈ X. A G-space X is said to be homogeneous if the action
of G is transitive on X.
32 G-spaces and quotients

Proposition 5.2 Let X be a G-space.

(a) For each x ∈ X, the stabilizer Gx := {g ∈ G | g.x = x} is a closed
subgroup.
(b) The fixed point set X G := {x ∈ X | g.x = x for all g ∈ G} is closed.
Proof By the definition of G-spaces, for any x ∈ X the orbit map
ϕ : G −→ X, g −→ g.x,
is a morphism of varieties. Hence ϕ−1 (x) = Gx is closed, so assertion (a)
follows. Part (b) is Exercise 10.18.
Example 5.3 We describe two important sources for G-spaces.
(1) Equip G with the structure of a G-space by considering the action of G
on itself by conjugation. Then Proposition 5.2 shows that centralizers of
elements and the center Z(G) are closed.
(2) Suppose that V is a finite-dimensional vector space over k. A morphism
ϕ : G → GL(V ) (of algebraic groups ) is called a rational representation
of G, and the space V is called a (rational ) kG-module. Then, V is a
G-space via the action (g, v) → ϕ(g)v. Furthermore, G also acts on the
associated projective space P(V ) via (g, v) → ϕ(g)v. With this action
P(V ) becomes a G-space.
In contrast to the situation for stabilizers considered above, it is not true,
in general, that orbits are always closed. But at least we have the following
weaker statement:
Proposition 5.4 Let X = ∅ be a G-space.
(a) Every orbit G.x is open in its closure.
(b) Orbits of minimal dimension are closed.
Proof By Proposition 1.6 (which continues to hold in this more general
setting) applied to the orbit map G → X, g → g.x, the orbit G.x contains
an open subset Y of its closure. As G.x is the union of G-translates of Y ,
(a) follows.
For (b), note that for x ∈ X, g.G.x is closed and contains G.x. Therefore,
G.x ⊆ g.G.x. Since the same is true for g −1 , we get g.G.x = G.x for all
g ∈ G. It follows that G.x is a union of G-orbits.
Choose x ∈ X with dim G.x minimal. If G.x is not closed, we claim that
the union of G-orbits G.x\G.x is of strictly smaller dimension, contradicting
minimality. To see this, let Y ⊆ G.x be an irreducible component intersecting
G.x. Then G.x ∩ Y is open in Y ; hence (G.x \ G.x) ∩ Y is closed in Y and so
 5.2 Existence of rational representations 33

by Proposition 1.22 has smaller dimension. (As pointed out above, the proof
of the latter result goes through in that case as well.)

5.2 Existence of rational representations

One expects that a transitive group action of an algebraic group G will
correspond to the action of G on a coset space G/H, for a subgroup H ≤ G.
In order to make this identification, one must define a structure of variety
on the coset space G/H.
Now, if H ≤ G is a closed subgroup, and I
k[G] denotes the ideal of
functions vanishing on H, then it is easily seen that
H = {x ∈ G | ρx (I) ⊆ I},
with ρx the right translation by x defined in Section 2.1.
The next theorem (see [32, §11.2]) is the first main ingredient in the con-
struction of quotients:
Theorem 5.5 (Chevalley) Let H ≤ G be a closed subgroup of a linear
algebraic group G. Then there exists a rational representation ϕ : G →
GL(V ) and a one-dimensional subspace W ≤ V such that H = {g ∈ G |
ϕ(g)W = W }.
Proof The ideal I
k[G] of functions vanishing on H is finitely generated,
say I = (Fj | j ∈ J) for some finite set J. Proposition 2.4 then implies that
there exists a finite-dimensional G-invariant subspace X of k[G] containing
the Fj , j ∈ J, and a corresponding rational representation ρ : G → GL(X).
Then M := X ∩ I is H-invariant. Conversely, if x ∈ G with ρx (M ) = M ,
then, since M generates the ideal I we have
ρx (I) = ρx (M )ρx (k[G]) = M k[G] = I,
so H = {x ∈ G | ρx (M ) = M }.
If d = dim M , we then set V = ∧d X, the dth exterior power of X, with
the representation ϕ : G → GL(V ) induced by the natural (rational) G-
action. Then the one-dimensional subspace W = ∧d M of V is clearly ϕ(H)-
invariant.
Now assume that ϕ(g)(W ) = W for some g ∈ G. Let w1 , . . . , wd be a basis
of M and wl+1 , . . . , wl+d be a basis of g(M ). Then ϕ(g)(w1 ∧ · · · ∧ wd ) ∈ W
by assumption. But by the choice of the second basis, it is also a multiple of
wl+1 ∧ · · · ∧ wl+d . But this then implies that ρg (M ) = M , so g ∈ H. Hence
W ≤ V satisfy the assertions of the theorem.
34 G-spaces and quotients

Let’s point out two important consequences. The first is the characteriza-
tion of linear algebraic groups already stated in Theorem 1.7:
Corollary. Any linear algebraic group can be embedded as a closed subgroup
into GLn for some n.

Proof Choosing H = 1 in Theorem 5.5, we obtain a faithful rational repre-

sentation ρ : G → GL(V ) ∼
= GLn , n = dim(V ), of our linear algebraic group
G in GLn , hence an embedding as a closed subgroup.

Secondly, for arbitrary H ≤ G, and V as in Theorem 5.5, let v ∈ P(V )

be the point in P(V ) corresponding to the line v stabilized by H. Set
X = G.v ⊆ P(V ). Then X is a homogeneous G-space and the action provides
a surjective map ϕ : G → X, g → g.v, with fibers the cosets of H, which
induces a bijection ϕ̄ : G/H → X.
Using this bijection ϕ̄, we can endow G/H with the structure of a variety.
Indeed, the orbit closure G.v ⊆ P(V ) is a closed subset of P(V ), so it is a pro-
jective variety. By Proposition 5.4(a), G.v is open in its closure. Therefore,
G.v is what is called a quasi-projective variety.
We call G/H, endowed with this structure of a variety, the quotient space
of G by H. Then, by construction, the natural map π : G → G/H is a
morphism of varieties. It can be shown that the variety structure on G/H
obtained in this way is independent of the chosen rational representation
[32, 12.4]. Moreover by Proposition 1.19 we have:

Proposition 5.6 Let H ≤ G be a closed subgroup of a linear algebraic

group G. Then dim(G/H) = dim(G) − dim(H).

In the case of a normal subgroup, even more is true (see [4, Thm. 6.8]):

Proposition 5.7 Let H be a closed normal subgroup of a linear algebraic

group G. Then:

(a) G/H is an affine variety.

(b) G/H with the usual group structure is a linear algebraic group.

About the proof. We apply Theorem 5.5 to obtain a morphism ϕ : G →

GL(V ) and a one-dimensional subspace L ⊆ V whose stabilizer is H. Now
H acts on L by scalar multiplications, so there is an associated character
H → k × which defines this action. Given a character χ ∈ X(H), set

Vχ = {v ∈ V | ϕ(h)v = χ(h)v for all h ∈ H}.

Then as H is normal in G, G permutes the Vχ , one of which contains L. As

 5.2 Existence of rational representations 35

the sum of the Vχ is direct there are only finitely many χ for which Vχ = 0.
We may then replace V by this sum and assume henceforth that V is the
direct sum of Vχ , with H acting as scalar multiplications on each Vχ .
Let W = {x ∈ End(V ) | x(Vχ ) ⊆ Vχ for all χ ∈ X(H)}, which is naturally

isomorphic to χ End(Vχ ). Now GL(V ) acts via conjugation on End(V ) and
ϕ(G) ≤ GL(V ) stabilizes W . Thus we have a group homomorphism ψ : G →
GL(W ), which is a rational representation. We claim that ker ψ = H. Indeed,
if g ∈ H then ϕ(g) acts as a scalar on each Vχ and so conjugating by ϕ(g) is
the identity on W . Conversely, if ψ(g) = 1, then ϕ(g) stabilizes each Vχ and
commutes with End(Vχ ), which implies that ϕ(g) acts as a scalar on each
Vχ . In particular ϕ(g) stabilizes the line L and by Theorem 5.5, g ∈ H. We
now have a group isomorphism G/H ∼ = im(ψ). To see that this is indeed an
isomorphism of algebraic varieties, refer to [4, Thm. 6.8].
Remarks 5.8 (a) In the situation of Proposition 5.7 let π : G → G/H be
the canonical surjection. One checks that π ∗ : k[G/H] → k[G] is injective.
Hence k[G/H] may be viewed as a subalgebra of k[G]. One can show that
k[G/H] ∼
= {f ∈ k[G] | ρx (f ) = f for all x ∈ H},
that is, k[G/H] is precisely the algebra of H-invariant functions on G.
(b) It follows from Proposition 5.7 that the quotient G/G considered in
the proof of Theorem 4.4 is a linear algebraic group.
Example 5.9 Let G = GLn , H = Z(G) = {tIn | t ∈ k × }. Then the
projective general linear group PGLn := GLn /Z(GLn ) is a linear algebraic
group of dimension
dim(PGLn ) = dim(GLn ) − dim(Z(GLn )) = n2 − 1
(see Exercise 10.17 for a direct proof).
Similarly one obtains the projective conformal symplectic group
PCSp2n := CSp2n /Z(CSp2n )
as well as the projective conformal orthogonal group and its connected com-
ponent of the identity
PCO2n := CO2n /Z(CO2n ), PCO◦2n = CO◦2n /Z(CO◦2n )
as linear algebraic groups.
 6
Borel subgroups

As seen in Section 4.2, the structure of connected solvable linear algebraic

groups is well-understood. We intend to exploit this by studying a particular
family of connected solvable subgroups of an arbitrary linear algebraic group.
By the Lie–Kolchin Theorem any connected solvable subgroup G ≤ GLn
can be embedded in Tn . In particular G stabilizes a flag F : 0 = V0 ⊂
V1 ⊂ · · · ⊂ Vn = k n of subspaces. Moreover, for G an arbitrary closed
subgroup of GLn , it is clear that the stabilizer GF ≤ G of any such flag
is a solvable group, and the quotient variety G/GF is a quasi-projective
variety, i.e., an open subset of a projective space. (See the remarks at the
beginning of Chapter 5 and Proposition 5.4.) Now if we choose F such that
its G-orbit is of minimal dimension, then this orbit is in fact closed (loc. cit.)
and so a projective space. We will obtain such minimal-dimensional orbits
by choosing GF of maximal possible dimension among flag stabilizers. This
leads us to our definition of Borel subgroups (see Definition 6.3 below).

6.1 The Borel fixed point theorem

The principal ingredient for the study of Borel subgroups is the following
fixed point theorem:
Theorem 6.1 (Borel fixed point theorem) Let G be a connected, solvable
linear algebraic group acting on a non-empty projective G-space X. Then
there exists x ∈ X such that g.x = x for all g ∈ G.
Sketch of proof We argue by induction on dim(G). If dim(G) = 0 then
G = 1, so all x ∈ X are fixed points. Now assume that dim(G) > 0. By
Proposition 1.18, G := [G, G] is connected, of smaller dimension by Propo-
sition 1.22 (since G is solvable). Thus, the set Y of fixed points of G on X
 6.1 The Borel fixed point theorem 37

is non-empty. By Proposition 5.2(b), Y is closed in X, hence also projective,

and it is stabilized by G since G is normal. Thus, we may replace X by Y .
Then, all stabilizers Gx , for x ∈ X, contain G , hence are normal in G. So
G/Gx is an affine variety, by Proposition 5.7. By Proposition 5.4(b) there
exists x ∈ X whose orbit is closed, hence projective. Then the canonical
morphism G/Gx → G.x is bijective, with G/Gx affine irreducible and G.x
projective. One must now appeal to a geometric argument based upon [32,
Lemma 21.1], to see that the above conditions force G/Gx to be reduced to
one point, that is, x is a fixed point for G.

Remarks 6.2 (a) The conclusion of the theorem in fact holds more gener-
ally when X is just assumed to be a complete variety (see [32, Thm. 21.2]).
(b) The Lie–Kolchin Theorem (Theorem 4.1) asserts that for G a con-
nected, solvable subgroup of GL(V ), there exists 0 = v ∈ V such that
g.v ∈ v for all g ∈ G, that is, G, acting on P(V ), has a fixed point. Borel’s
fixed point theorem can hence be used to give a short proof of this assertion.

Definition 6.3 Let G be a linear algebraic group. A Borel subgroup of

G is a closed, connected, solvable subgroup B of G which is maximal with
respect to all these properties.

Clearly, Borel subgroups exist: just take a closed, connected, solvable sub-
group of maximal possible dimension. In fact, there is not much choice (see
[32, §21.3]):

Theorem 6.4 Let G be a linear algebraic group. Then:

(a) All Borel subgroups of G are conjugate.

(b) If G is connected, G/B is projective for any Borel subgroup B of G.

Proof of (a) assuming (b) Firstly, we consider the case G = GLn . The
group Tn of upper triangular matrices is a closed, connected, solvable sub-
group of G. If H is any closed, connected, solvable subgroup of G, then
Corollary 4.2 gives H ≤ g −1 Tn g for some g ∈ G. Therefore, Tn is maximal
among closed, connected, solvable subgroups of G, hence a Borel subgroup,
and any Borel subgroup is conjugate to Tn .
For the general case we can replace G by G◦ as Borel subgroups of G◦ are
Borel subgroups of G. Let H be a Borel subgroup of G. Assuming G/B is
projective for a fixed Borel subgroup B, H acts on the projective space G/B
via gB → hgB. By Theorem 6.1, there exists a fixed point, that is, there
exists g ∈ G such that HgB = gB. Therefore, g −1 Hg ≤ B. By maximality
of H, we conclude that H = gBg −1 .
38 Borel subgroups

As in the case of solvable groups we say that a subtorus T ≤ G is a

maximal torus of G if it is maximal among subtori with respect to inclusion.

Corollary 6.5 Let G be a linear algebraic group. Then all maximal tori of
G are conjugate.

Proof Let T1 , T2 be maximal tori of G. Since Ti is connected solvable, there

exists a Borel subgroup Bi of G with Ti ≤ Bi , for i = 1, 2. By Theorem 6.4,
there exists g ∈ G such that B2 = B1g . Therefore, T1g ≤ B2 , hence T2 and T1g
are two maximal tori of the connected solvable group B2 . By Theorem 4.4,
there exists b ∈ B2 such that T2 = T1gb .

Definition 6.6 The rank of a linear algebraic group G is the dimension

of a maximal torus of G, denoted by rk(G).

Note that by Corollary 6.5, as for connected solvable groups, maximal tori
are the tori of maximal dimension so the rank is well-defined.

Example 6.7 Let’s find Borel subgroups, maximal tori and the rank of
classical groups.

(1) For G = GLn , Tn is a Borel subgroup (see the proof of Theorem 6.4),
and Dn is a maximal torus of Tn , hence of G. Therefore, rk(GLn ) =
dim(Dn ) = n.
(2) For G = SLn , if we repeat the argument given for GLn , we observe that
Tn ∩ SLn is a Borel subgroup and Dn ∩ SLn is a maximal torus of SLn .
The exact sequence
det
1 −→ Dn ∩ SLn −→ Dn −→ Gm −→ 1

then shows with Corollary 1.20 that rk(SLn ) = n − 1.

(3) For the special orthogonal group SO2n ,
⎧⎛ ⎞ ⎫
⎪ t1 . 0  ⎪
⎪
⎪ .  ⎪
⎪
⎨⎜ . ⎟ ⎬
⎜ tn ⎟ ×
T := ⎜ −1 ⎟  ti ∈ k ≤ SO2n
⎪
⎪ ⎝ tn . ⎠ ⎪
⎪
⎪
⎩ . . −1  ⎪
⎭
0 t1 

is a connected subgroup isomorphic to Gnm , hence a torus of dimension

dim(T ) = n. Therefore rk(SO2n ) ≥ n. Now let T1 ≥ T be a maximal
torus of SO2n . Choose s ∈ T with distinct eigenvalues (which is clearly
possible since k is infinite). Then

T1 ≤ CSO2n (T ) ≤ CGL2n (T ) ≤ CGL2n (s) = D2n .

 6.2 Properties of Borel subgroups 39

But it is easily checked from the definition of GO2n in Section 1.2 that
D2n ∩ GO2n = T , so T = T1 is a maximal torus and rk(SO2n ) = n.
Analogously one shows that rk(Sp2n ) = rk(SO2n+1 ) = n (see Exer-
cise 10.19).


A1 A2
(4) Let B := SO2n ∩T2n . A short calculation shows that ∈ SO2n
0 A3
1 Kn A3 = Kn , and A3 Kn A2 = −A2 Kn A3 , that is, A3 =
if and only if Atr tr tr
−tr tr −1
Kn A1 Kn , and A3 Kn A2 = Kn A1 A2 is skew-symmetric. (Recall that,
by our definition, SO2n = GO◦2n in any characteristic.) In particular, B
is the product of

  
In Kn S 
 S ∈ k n×n
skew-symmetric
0 In
with {diag(A, Kn A−tr Kn ) | A ∈ Tn } ∼
= Tn , of dimension
% & % &
n
2 +
n+1
= n2 , if n ≥ 2
dim(B) = %n+1& 2

2 = n2 , if n = 1.
We claim that B is a Borel subgroup of SO2n . Let B1 ≥ B be a Borel
subgroup of SO2n . Then B1 lies in a Borel subgroup of GL2n , so it sta-
bilizes a complete flag on V = k 2n . The only one-dimensional subspace
invariant under B is v1 , generated by the first standard basis vector.
Thus, this must also be fixed by B1 . But then B1 also fixes v1 ⊥ , a sub-
space of codimension 1. Now applying induction to the action of B1 , B
on v1 ⊥ /v1  one sees that B1 can only fix the standard flag, hence is
contained in T2n . It follows that B1 = B, as claimed.
Analogously one shows that Sp2n ∩ T2n , respectively SO2n+1 ∩ T2n+1
is a Borel subgroup of Sp2n , respectively SO2n+1 (see Exercise 10.19).

6.2 Properties of Borel subgroups

Here we collect some further results which illustrate the importance of Borel
subgroups and which will be needed later on.
Proposition 6.8 Let G be connected, B a Borel subgroup of G. Then:

(a) An automorphism of G which fixes B pointwise is the identity.

(b) Z(G)◦ ⊆ Z(B) ⊆ CG (B) = Z(G).
Proof For (a) let σ be an automorphism of G which centralizes B. Then
the morphism ϕ : G → G, x → σ(x)x−1 , sends B to 1, so factors through
40 Borel subgroups

G/B. By Theorem 6.4(b), G/B is projective. Now the image of a projective

variety under a morphism is closed (see for example [32, Prop. 6.1(c) and
Thm. 6.2]), so ϕ(G) is closed in G, hence affine. Now for any two distinct
points in this affine variety there exists a function f separating them, whence
f ◦ ϕ would be a globally defined non-constant function on the projective
variety G/B, which does not exist. Thus, ϕ(G) = 1 as claimed.
For (b), Z(G)◦ is a closed connected solvable subgroup and hence lies in
some Borel subgroup of G, and so by Theorem 6.4(a) in all of them. The
second inclusion is obvious, as is Z(G) ⊆ CG (B). Finally, for x ∈ CG (B),
conjugation by x satisfies the assumptions of (a), whence x ∈ Z(G).
Corollary 6.9 Let G be a connected linear algebraic group with nilpo-
tent Borel subgroups. Then G is nilpotent. In particular, any connected two-
dimensional group is solvable.
This is an easy consequence of Proposition 6.8, see Exercise 10.20.
Theorem 6.10 Let G be a connected linear algebraic group with Borel
subgroup B. Then G = g∈G g −1 Bg.
For GLn we may take B = Tn , and then the theorem follows from the
existence of Jordan normal forms. As GLn = SLn · Z(GLn ), this also applies
to SLn . For the general case see [32, §22.2].
Note that, in contrast, for a finite group G we can never have G =
−1
g∈G g Hg for a proper subgroup H < G.
As an immediate consequence we obtain:
Corollary 6.11 Let G be a connected linear algebraic group. Then:
(a) Every semisimple element of G lies in a maximal torus.
(b) Every unipotent element of G lies in a closed connected unipotent sub-
group.
(c) The maximal, closed, connected unipotent subgroups of G are all conju-
gate and they are of the form Bu where B is a Borel subgroup of G.
Proof Every g ∈ G lies in a Borel subgroup B of G by Theorem 6.10. If g is
semisimple then it lies in a maximal torus of B, hence of G, by Corollary 4.5.
If g is unipotent in B, then u ∈ Bu which is closed by Theorem 4.4(a). If
U ≤ G is closed connected unipotent, it is solvable by Corollary 2.10, so
contained in a Borel subgroup B, hence in its unipotent normal subgroup
Bu again by Theorem 4.4(a), which shows (c).
A further important result asserts that Borel subgroups of connected
groups are self-normalizing. This is easy to see for GLn , taking the Borel
 6.2 Properties of Borel subgroups 41

subgroup Tn . In the general case let NG (B) be the normalizer of a Borel

subgroup B of the connected group G. Then clearly B is also a Borel sub-
group of NG (B)◦ , and it is normal, so NG (B)◦ = B by Theorem 6.10. For the
proof of the following stronger assertion one must use properties of complete
varieties, see [32, Thm. 23.1].
Theorem 6.12 Let G be a connected linear algebraic group and B a Borel
subgroup of G. Then NG (B) = B.
If N, N   G are closed connected (and solvable) normal subgroups, then
N N  is closed connected normal (and solvable) by Proposition 1.16. This
gives meaning to the following definition:
Definition 6.13 The maximal closed connected solvable normal subgroup
of a linear algebraic group G is called the radical R(G) of G.
By Theorem 4.4, R(G)u is a normal connected unipotent subgroup of
the connected solvable group R(G). Any closed connected normal unipotent
subgroup of G is solvable by Corollary 2.10, hence contained in R(G), hence
in R(G)u . Thus R(G)u is the maximal closed connected normal unipotent
subgroup of G, the so-called unipotent radical Ru (G) of G. Clearly Ru (G) ≤
R(G) ≤ G◦ .
In the case where k = Fp , the unipotent radical is the largest connected
normal subgroup consisting entirely of p-elements, so the analogue of the
maximal normal p-subgroup Op (G) for a finite group G.
Definition 6.14 A linear algebraic group G is called reductive if Ru (G) =
1. It is called semisimple if it is connected and R(G) = 1.
Unipotent groups can have a very complicated structure, see for exam-
ple Exercise 10.16. On the other hand, connected reductive and semisimple
groups can be classified; this will be discussed in Section 9.2.
It is easy to see that in general, for a connected group G, G/R(G) is
semisimple, and G/Ru (G) is reductive (see Exercise 10.21).
Definition 6.15 The semisimple rank of a reductive group G is defined
to be rkss (G) := rk(G/R(G)). We will see later (Corollary 8.22) that for
connected reductive groups rkss (G) = rk([G, G]).
The radical of G is controlled by its Borel subgroups (see Exercise 10.22):
$
Proposition 6.16 Let G be connected. Then R(G) = ( B B)◦ , where B
runs over all Borel subgroups of G.
Example 6.17 (1) If G is semisimple, T a torus, then G × T is reductive.
42 Borel subgroups

(2) If G is solvable connected, then by Theorem 4.4 Ru (G) = Gu and

R(G) = G. So a connected reductive solvable group is a torus, a semisim-
ple solvable group
is trivial.
  
A ∗ 
(3) The subgroup  A ∈ GL2 of GL3 is not reductive, since
0 ∗


I2 ∗
is a non-trivial connected normal unipotent subgroup.
0 1
(4) G = GLn is reductive, since by Proposition 6.16 and Example 6.7(1)
   
−
∗. ∗ ∗. 0
R(G) ≤ Tn ∩ Tn := .. ∩ .. = Dn ,
0 ∗ ∗ ∗
hence Ru (G) = R(G)u = 1. But GLn is not semisimple, as Z(GLn ) =
{tIn | t ∈ k × } ∼
= Gm is a connected solvable normal subgroup. In fact,
R(GLn ) = Z(GLn ) by Proposition 6.20, so PGLn is semisimple.
(5) G = SLn is semisimple, since as in (4), SLn is reductive and
R(G) = Z(G)◦ = ({tIn | t ∈ k × } ∩ SLn )◦ = {tIn | tn = 1}◦ = 1
by Proposition 6.20(a) below.
Using properties of Borel subgroups, the result of Theorem 3.10 on cen-
tralizers of tori can be strengthened:
Proposition 6.18 Let G be a connected linear algebraic group and T ≤ G
a maximal torus. Then CG (T )◦ is nilpotent and T is its maximal torus.
Proof Set C = CG (T )◦ . The central torus T of C is contained in any
Borel subgroup B of C and is a maximal torus thereof. Hence B/T ∼ = Bu
is a nilpotent group, and since T is central in B, B is also nilpotent. By
Corollary 6.9 this implies that C = B is nilpotent.
We can say something as well about centralizers of arbitrary subtori, see
[66, Thm 6.4.7]:
Theorem 6.19 Let S be a torus of a connected linear algebraic group G.
(a) The centralizer CG (S) is connected.
(b) If B is a Borel subgroup of G containing S, then CG (S) ∩ B is a Borel
subgroup of CG (S) and all Borel subgroups of CG (S) are obtained in this
way.
About the proof of (a). For c ∈ CG (S) let B be a Borel subgroup of G
containing c (which exists by Theorem 6.10). Set X = {xB ∈ G/B | x−1 cx ∈
B}. Then one can show that X is a closed subvariety of G/B, and hence
 6.2 Properties of Borel subgroups 43

is a projective variety. Now S acts on X via left multiplication and so by

Theorem 6.1, there exists a fixed point xB ∈ X with x−1 Sx ≤ B. That is,
there exists a Borel subgroup containing both S and c. Thus, we are reduced
to the case of a connected solvable group, for which the result is true by
Proposition 4.6.
It is quite difficult to exploit the fact that a group is reductive, that is,
has no non-trivial connected unipotent normal subgroup. At least, we can
show:
Proposition 6.20 Let G be connected reductive. Then:
(a) R(G) = Z(G)◦ is a torus.
(b) R(G) ∩ [G, G] is finite.
(c) [G, G] is semisimple.
Proof (a) R(G) is connected solvable, so R(G) = R(G)u
T , where T is
a torus, by Theorem 4.4(b). Since G is reductive, R(G)u = Ru (G) = 1,
hence R(G) = T is a torus. Clearly Z(G)◦ ⊆ R(G). By Theorem 3.10, the
factor group G/CG (R(G)) = NG (R(G))/CG (R(G)) is finite. By Proposi-
tion 1.13(c), CG (R(G)) contains G◦ , so we have equality CG (R(G)) = G
since G is connected, and hence R(G) ⊆ Z(G)◦ .
(b) Choose an embedding G → GL(V ) according to Theorem 1.7. The
torus R(G) = Z(G)◦ acts diagonally on V . The common eigenspaces give a

decomposition V = χ Vχ , where χ runs over X(R(G)) (so t.v = χ(t)v for
v ∈ Vχ , t ∈ R(G)). Now G ≤ CGL(V ) (R(G)) by (a), so G stabilizes each of

the Vχ . Thus it consists of block diagonal matrices: G ≤ χ GL(Vχ ) and so

[G, G] ≤ χ SL(Vχ ). Thus [G, G] ∩ R(G) consists of matrices
⎛ ⎞
χ1 (t)Id1
⎜ χ2 (t)Id2 ⎟
⎝ ⎠
..
.

with di := dim Vχi and χi (t)di = 1. But there exist only finitely many such
matrices.
(c) By Proposition 1.18, [G, G] is connected. As R([G, G]) is a charac-
teristic normal subgroup of [G, G], R([G, G]) ≤ R(G). Thus R([G, G]) ≤
(R(G) ∩ [G, G])◦ = 1 by part (b), hence [G, G] is semisimple.
We also have that G = R(G)[G, G], but the proof of this statement requires
considerably more work (see Corollary 8.22 later).
 7
The Lie algebra of a linear algebraic group

To go further in our investigation of reductive groups we require the concept

of the Lie algebra of a linear algebraic group. This allows one to “linearize”
many questions. Here, we assume that the reader is familiar with the notion
of a Lie algebra over a field. We will write gln for the Lie algebra of n × n-
matrices over k with the Lie bracket [X, Y ] = XY − Y X.

7.1 Derivations and differentials

Let A be a k-algebra.
Definition 7.1 A k-linear map D : A → A with D(f g) = f D(g) + D(f )g
for all f, g ∈ A is called a derivation of A.
An easy calculation shows that if D1 , D2 are derivations of A, then so is
[D1 , D2 ] := D1 ◦ D2 − D2 ◦ D1 .
This endows Derk (A), the k-vector space of derivations of A, with the struc-
ture of a Lie algebra.
We now take A = k[G], the coordinate ring of a linear algebraic group G.
Then, for x ∈ G we have an action λx : k[G] → k[G] on k[G] via (λx .f )(g) :=
f (x−1 g) for f ∈ k[G], g ∈ G.
Definition 7.2 The Lie algebra of G is the subspace
Lie(G) := {D ∈ Derk (k[G]) | Dλx = λx D for all x ∈ G}
of left invariant derivations of k[G], a Lie subalgebra of Derk (k[G]).
Example 7.3 Let’s compute the Lie algebras of one-dimensional connected
groups.
 7.1 Derivations and differentials 45

(1) Let G = Ga , with k[G] = k[T ]. Any derivation D ∈ Derk (k[G]) is

determined by its value D(T ) =: p(T ) ∈ k[T ] on T . For x ∈ G we have
λx T = T − x. Therefore λx f (T ) = f (T − x) for f ∈ k[T ]. So D is left
invariant if and only if
(λx D)(T ) = λx p(T ) = p(T − x) equals (Dλx )(T ) = D(T − x) = p(T )
for all x ∈ G. Thus, p(T ) is constant and D = a dT d
for a = p(1) ∈ k.
∼
Therefore Lie(G) =  dT k = k is the unique one-dimensional (commu-
d

tative) Lie algebra over k.

(2) Let G = Gm , with k[G] = k[T, T −1 ]. Again, D ∈ Derk (k[G]) is deter-
mined by D(T ) =: p(T ). Here, for x ∈ G we have λx T = x−1 T , so
λx f (T ) = f (x−1 T ) for f ∈ k[G]; hence left invariance gives p(T ) = aT
for some a ∈ k, so D = aT dT d
. Thus in this case as well we have
∼
Lie(G) = T dT k = k.
d

We now give a second, more geometric construction for the Lie algebra of
an algebraic group G.
For an affine variety X we define the tangent space of X at x ∈ X by
Tx (X) := {δ : k[X] → k linear | δ(f g) = f (x)δ(g)+δ(f )g(x) for f, g ∈ k[X]}
(the k-vector space of point derivations at x). In case X is a linear algebraic
group G, then as G acts homogeneously on itself by left translation, the
tangent space at any group element g ∈ G is naturally isomorphic to T1 (G).
Thus, we will consider the tangent space T1 (G). Now
Θ : Lie(G) → T1 (G), Θ(D)(f ) := D(f )(1),
is a k-linear map. In fact, a straightforward calculation shows this to be an
isomorphism of vector spaces, see [32, Thm. 9.1]. By this, the Lie algebra
structure on Lie(G) can be transported to T1 (G). With this alternative in-
terpretation one can establish the following (see [66, Thm. 4.3.7, Cor. 4.4.6,
§4.1.7] for (a) and (b), [32, §5.1] for (c)):

Theorem 7.4 Let G, G1 , G2 be linear algebraic groups. Then:

(a) Lie(G) = Lie(G◦ ).

(b) dim G = dim G◦ = dim(Lie(G)).
(c) Lie(G1 × G2 ) ∼
= Lie(G1 ) ⊕ Lie(G2 ) as Lie algebras.
Example 7.5 For a more elaborate example, let G = GLn . We claim that
Lie(GLn ) ∼
= gln . For this, define
gln −→ Derk (k[G]), X −→ DX ,
46 The Lie algebra of a linear algebraic group
n 1
by DX (Tij ) := l=1 Til Xlj . (Recall that (k[G] = k[Tij , det(Tij ) ].) It is
straightforward to check that DX is a derivation. Now for g ∈ G write
−1
gij for the (i, j)th entry of g −1 . Then
"
−1
DX (λg Tij ) = DX ( gim Tmj )
m
"
−1
= gim DX (Tmj )
m
" " ""
−1 −1
= gim Tml Xlj = gim Tml Xlj ,
m l m l

while
"
λg (DX Tij ) = λg ( Til Xlj )
l
"
= Xlj λg (Til )
l
" " ""
−1 −1
= Xlj gim Tml = Xlj gim Tml .
l m l m

Hence DX is a left invariant derivation of k[G]. Moreover, one can check

that the map X → DX defines an injective Lie algebra homomorphism. Since
dim(gln ) = n2 = dim(GLn ), the given map is surjective by Theorem 7.4(b)
and we have the desired isomorphism.

Furthermore, the tangent space approach allows one to complete the func-
torial connection between linear algebraic groups and their Lie algebras.

Definition 7.6 Let ϕ : X → Y be a morphism of affine varieties. The

differential dx ϕ of ϕ at x ∈ X is the map dx ϕ : Tx (X) → Tϕ(x) (Y ) defined
by dx ϕ(δ) := δ ◦ ϕ∗ for δ ∈ Tx (X).
ϕ∗
k[Y ] / k[X]
F
F
F
F δ
dx ϕ(δ) F# 
k

If ϕ : G → H is a morphism of algebraic groups we set dϕ := d1 ϕ, the

differential at 1. Taking differentials behaves functorially:
ϕ ψ
Proposition 7.7 Let X → Y → Z be morphisms of affine varieties, and
x ∈ X. Then:

(a) dx (ψ ◦ ϕ) = dϕ(x) ψ ◦ dx ϕ.
 7.1 Derivations and differentials 47

(b) If X = G1 and Y = G2 are linear algebraic groups and ϕ is a mor-

phism of algebraic groups, then dϕ : Lie(G1 ) → Lie(G2 ) is a Lie algebra
homomorphism.
(c) If X = G1 and Y = G2 are linear algebraic groups and ϕ is a morphism
of algebraic groups, then ϕ is an isomorphism of algebraic groups if and
only if ϕ and dϕ are both bijective.

The proof is given in [32, §5.4, Thm. 9.1] and [66, Cor. 5.3.3].

Example 7.8 (Computation of some differentials)

(1) Consider the morphism µ : G × G → G, (x, y) → xy. For f ∈ k[G], write

 
µ∗ (f ) = fi ⊗ gi ∈ k[G] ⊗ k[G], so that f (xy) = f (x)gi (y), for all
 i
x, y ∈ G. In particular, we have f = fi (1)gi = gi (1)fi . Now let
(X, Y ) ∈ T1 (G × G) ∼= T1 (G) ⊕ T1 (G) and set dµ(X, Y ) = Z. Then
"
Z(f ) = dµ(X, Y )(f ) = (X, Y )(µ∗ f ) = (X, Y )( fi ⊗ gi ),

which by the product rule for derivatives is equal to X(fi )gi (1) +

fi (1)Y (gi ). But this is precisely the action of X + Y on f . So we have
dµ(X, Y ) = X + Y .
(2) The differential of the inversion morphism i : G → G, g → g −1 , is given
by di(X) = −X for X ∈ T1 (G). (See Exercise 10.25.) This fact and the
previous example somewhat justify the statement that consideration of
the Lie algebra “linearizes” problems.
(3) Let char(k) = p > 0. Consider the bijective morphism


1 ap
ϕ : Ga −→ U2 , a→ .
0 1

Let us identify the map dϕ : Lie(Ga ) → Lie(U2 ). We have k[U2 ] =

k[T12 ] ∼
= k[Ga ] = k[T ]. Now note that


1 ap
ϕ∗ (T12 )(a) = T12 (ϕ(a)) = T12 ( ) = ap for a ∈ Ga ,
0 1

so ϕ∗ (T12 ) = T p . Thus for D ∈ Lie(Ga ) we obtain

dϕ(D)(T12 ) = D(ϕ∗ (T12 )) = D(T p ) = pT p−1 D(T ) = 0.

Hence dϕ is not injective, and so ϕ is not an isomorphism of alge-

braic groups by Proposition 7.7(c). Indeed, on the coordinate rings,
ϕ∗ : k[T12 ] → k[T ], T12 → T p , is not surjective.
48 The Lie algebra of a linear algebraic group

We now consider Lie algebras of closed subgroups H ≤ G. If H is defined

by an ideal I  k[G], then by definition k[H] = k[G]/I. Any D ∈ Lie(G)
with DI ⊆ I naturally defines a derivation of k[H] = k[G]/I, and similarly
δ ∈ T1 (G) with δI = 0 naturally defines an element of T1 (H). With these
identifications we have:

Theorem 7.9 Let H ≤ G with vanishing ideal I  k[G]. Then:

(a) Lie(H) = {D ∈ Lie(G) | DI ⊆ I}, and T1 (H) = {δ ∈ T1 (G) | δI = 0}.

(b) If H G is normal, then Lie(H) is an ideal in Lie(G). Moreover, the dif-
ferential of the canonical projection G → G/H is the canonical projection
Lie(G) → Lie(G)/Lie(H) and thus induces an isomorphism Lie(G/H) ∼ =
Lie(G)/Lie(H).

The proof of the first part is a direct calculation, see [32, Lemma 9.4], and
uses the identification of Lie(H) with T1 (H). For the second part see [32,
Cor. 10.4A and Sec. 11.5].

Example 7.10 (1) Lie(SLn ) = sln := {A ∈ gln | trA = 0} (see Exer-

cise 10.26(b)).
(2) Lie(PGLn ) ∼= Lie(GLn )/Lie(Z(GLn )) = gln /{cIn | c ∈ k} by Theo-
rem 7.9(b).

We have the additive Jordan decomposition in gln . For X ∈ gln write

Xs , Xn for the semisimple, respectively nilpotent part of X. For any closed
subgroup G ≤ GLn we may embed Lie(G) into gln by Theorem 7.9(a) and
so define semisimple and nilpotent parts of elements of Lie(G). These satisfy
the expected compatibility properties (see [32, Thm. 15.3]):

Proposition 7.11 Let G ≤ GLn be a linear algebraic group, X ∈ Lie(G) ⊆

gln . Then we have:

(a) Xs , Xn ∈ Lie(G) and [Xs , Xn ] = 0.

(b) Xs , Xn are independent of the chosen embedding G → GLn .
(c) If ϕ : G1 → G2 is a morphism of linear algebraic groups, with differential
dϕ : Lie(G1 ) → Lie(G2 ), then (dϕ)(X)s = dϕ(Xs ) and (dϕ)(X)n =
dϕ(Xn ) for X ∈ Lie(G1 ).

Compare this result with the Jordan decomposition in G (Theorem 2.5).

 7.2 The adjoint representation 49

7.2 The adjoint representation

An important use of the Lie algebra is that it defines for each linear alge-
braic group G a natural rational representation G → GL(Lie(G)) on its Lie
algebra, as follows: For x ∈ G define Intx : G → G by Intx (y) = xyx−1 .
Then d Intx : Lie(G) → Lie(G) is a Lie algebra automorphism. We write
Ad x := d Intx for x ∈ G.
This defines a representation

Ad : G −→ GL(Lie(G)), x → Ad x,

the adjoint representation of G:

Theorem 7.12 Let G be a linear algebraic group. Then:

(a) Ad : G → GL(Lie(G)) is a rational representation of G.

(b) dAd = ad, where (ad(X))(Y ) = [X, Y ] for X, Y ∈ Lie(G).
(c) If G is connected reductive, then ker(Ad ) = Z(G).

See [32, Prop. 10.3, Thm. 10.4] for (a) and (b). The statement of (c)
will follow from Theorem 8.17, see Exercise 10.32. Thus, if G is connected
reductive, Ad is almost faithful: only the center of G is lost.
The following is easily verified: if H ≤ G is closed, then Lie(H) (as a Lie-
subalgebra of Lie(G) according to Theorem 7.9) is Ad G (H)-invariant and
the restriction of the adjoint representation of G to H, acting on Lie(H), is
just the adjoint representation of H, i.e., Ad G |H = Ad H on Lie(H).

Example 7.13 Let G = GLn , so Lie(G) ∼ = gln by Example 7.5 and for

X ∈ gln we have the derivation DX ∈ Lie(G) given by DX (Tij ) = l Til Xlj .
Via the vector space isomorphism Θ defined above, we can associate a point

derivation δX to X, with δX (Tij ) = ( l Til Xlj )(In ) = Xij .
For g = (gij ) ∈ G, we evaluate (Intg )∗ (Tij ): For x = (xij ) ∈ G,

"
−1
(Intg )∗ (Tij )(x) = Tij (gxg −1 ) = gil gmj xlm ,
l,m

 −1
so (Intg )∗ (Tij ) = l,m gil gmj Tlm . Thus Ad g = dIntg : Lie(G) → Lie(G)
50 The Lie algebra of a linear algebraic group

satisfies
(Ad g)(δX )(Tij ) = δX ((Intg )∗ Tij )
"
−1
= δX ( gil gmj Tlm )
l,m
"
−1
= gil gmj δX (Tlm )
l,m
"
−1
= gil gmj Xlm = (gXg −1 )ij .
l,m

So (Ad g)(δX ) = δgXg−1 and Ad g for GLn is just conjugation by g on gln ,

and by the remarks preceding this example, the same holds for any closed
subgroup G ≤ GLn .
We conclude this discussion of the adjoint representation of G by stating
a result ([66, Cor. 5.4.7]) which will be crucial in the investigation of the
structure of reductive groups. This concerns the action of a torus on Lie(G).
Proposition 7.14 Let G be a connected linear algebraic group and S ≤ G
a subtorus of G. Then (under the identification in Theorem 7.9)
Lie(CG (S)) = CLie(G) (S) := {X ∈ Lie(G) | Ad (s)X = X for all s ∈ S}.
Example 7.15 We verify the previous statement for G = GLn . Let m ≥ 1
and let S be the (n − m + 1)-dimensional subtorus {diag(t1 , . . . , tn ) | ti =
tj for 1 ≤ i, j ≤ m}. Then
⎧⎛ ⎞ ⎫
⎪
⎪ A 0 ··· 0  ⎪
⎪
⎨⎜ 0 tm+1 0⎟  ⎬
CG (S) = ⎜ ⎝.. . ⎟  A ∈ GL , t , . . . , t ∈ k ×
.
. .. ⎠ 
.. m m+1 n
⎪
⎪ ⎪
⎪
⎩ . ⎭
0 0 · · · tn
The ideal of k[G] = k[Tij , det(Tij )−1 ] defining CG (S) is
(Tij | i = j, i > m or j > m).
So Lie(CG (S)) can be identified with the set of matrices
⎧⎛ ⎞ ⎫
⎪
⎪ X 0 ··· 0  ⎪
⎪
⎨⎜ 0 am+1 0⎟   ⎬
⎜. . ⎟  X ∈ gl , a , . . . , a ∈ k
⎪ ⎝. .. .. ⎠  m m+1 n
⎪
⎪
⎩ .
. ⎪
⎭
0 0 · · · an
which is precisely Cgln (S).
The proposition cannot be generalized by replacing the torus by an arbi-
trary closed subgroup H of G. See Exercise 10.27(d).
 8
Structure of reductive groups

It turns out that the best way to investigate reductive algebraic groups is
via their adjoint action on the Lie algebra. We start by decomposing the Lie
algebra according to the action of a maximal torus. Then, as a preparation
for the case of an arbitrary semisimple group we consider the smallest non-
solvable groups in some detail.

8.1 Root space decomposition

Let G be a linear algebraic group. Let T ≤ G be a maximal torus and assume
that dim(T ) ≥ 1, which is the case for example if G = 1 is reductive. Write
g := Lie(G) for the Lie algebra of G. The image of T under the adjoint
representation Ad T ≤ GL(g) is a set of commuting semisimple elements, so
can be simultaneously diagonalized. For χ ∈ X(T ) write

gχ = {v ∈ g | (Ad t)(v) = χ(t)v for all t ∈ T }


for the common T -eigenspaces in g. Then g = χ∈X(T ) gχ .

Definition 8.1 The set of non-zero characters with non-zero eigenspace

Φ(G) := {χ ∈ X(T ) | χ = 0, gχ = 0}

occurring in the above decomposition is called the set of roots of G with

respect to T , and W := NG (T )/CG (T ) is the Weyl group of G with respect
to T (also denoted WG (T ) when necessary).

Note that the Weyl group is finite by Theorem 3.10.

Example 8.2 We determine Φ(G) for some connected reductive groups G.

52 Structure of reductive groups

(1) Let G = GLn , with Lie algebra gln (see Example 7.5). By Example 7.13
the adjoint action of G on gln is by conjugation. Let T = Dn . Then
Lie(T ) is the set of diagonal matrices in gln by Exercise 10.26. Let Eij ∈
gln be the n × n-matrix whose (l, m) entry is δil δjm . A straightforward
calculation gives

Ad (diag(t1 , . . . , tn ))(Eij ) = ti t−1

j Eij .

So the character χij : T → Gm , χij (diag(t1 , . . . , tn )) = ti t−1

j , is a root

of G whenever i = j. Moreover, gln = i=j Eij  ⊕ Lie(T ) and T acts
trivially on Lie(T ). So G has the set of roots

Φ(G) = {χij | 1 ≤ i, j ≤ n, i = j},

of order |Φ(G)| = n(n − 1). Note that all gα , α ∈ Φ(G), are one-
dimensional. The Weyl group is

W = NG (T )/T = (monomial matrices)/T

∼
= permutation matrices ∼
= Sn ,

the symmetric group on n letters.

(2) Let G = SLn . Recall from Example 7.10(a) that Lie(SLn ) = ∼ sln . By
Example 7.13, the action of G on Lie(G) is given by conjugation. Then,
with respect to the maximal torus Dn ∩ SLn , as above we obtain Φ(G) =
{χij | 1 ≤ i, j ≤ n, i = j}. The details are left to the reader.

By transport of structure, the Weyl group W acts naturally on the char-

acter group X(T ) as well as on the cocharacter group Y (T ) via

(w.χ)(t) := χ(tw ) for all w ∈ W, χ ∈ X(T ), t ∈ T,

w−1
w
(w.γ)(c) := (γ(c)) := γ(c) for all w ∈ W, γ ∈ Y (T ), c ∈ Gm .

These actions can easily be seen to be faithful, and moreover they are com-
patible with the pairing  ,  : X(T ) × Y (T ) → Z defined in Example 3.5
(see Exercise 10.30):

Lemma 8.3 For all w ∈ W , χ ∈ X(T ), γ ∈ Y (T ) we have

w.χ, γ = χ, w−1 .γ.

We now claim that W stabilizes the set of roots Φ(G) ⊆ X(T ). Indeed, for
n ∈ NG (T ), α ∈ Φ(G) and v ∈ gα , (Ad n)(v) is again a common eigenvector
 8.2 Semisimple groups of rank 1 53

for all elements of T , since

(Ad tAd n)(v) = (Ad nAd (n−1 tn))(v)
= Ad n(α(n−1 tn)v)
= α(n−1 tn)(Ad n)(v) for all t ∈ T.
As there exists t ∈ T such that α(n−1 tn) = 1, we have established the
following:
Proposition 8.4 Let G be reductive with maximal torus T , set of roots
Φ and Lie algebra g. Then for all α ∈ Φ, n ∈ NG (T ) and w = nCG (T ) ∈
NG (T )/CG (T ) we have (Ad n)(gα ) ⊆ gw.α . In particular, Φ is W -stable.

8.2 Semisimple groups of rank 1

Let G be a connected non-solvable group. Then rk(G) ≥ 1 and dim(G) ≥ 3
by Corollaries 2.10 and 6.9. Assume throughout this section that rk(G) = 1.
Let B be a Borel subgroup of G and T ≤ B a maximal torus, so dim(T ) = 1
and T ∼= Gm . Now the Weyl group W = NG (T )/CG (T ) acts faithfully as
a group of algebraic group automorphisms on T ∼ = Gm , so |W | ≤ 2 by
Exercise 10.11. We now require [66, Prop. 7.1.5] whose proof is geometric
and will be omitted here:
Proposition 8.5 Let G be a rank 1 connected non-solvable linear algebraic
group with Borel subgroup B. Then W has order 2 and dim G/B = 1.
This result has an immediate application to the study of semisimple groups
of rank 1 ([32, Thm. 25.3]):
Proposition 8.6 Let G be a semisimple linear algebraic group of rank 1.
Then there exists a surjective morphism ϕ : G → PGL2 , with finite kernel;
in particular dim(G) = 3.
About the proof. Being semisimple, G is non-solvable by Example 6.17(2).
By Theorem 6.4 and the previous proposition, G/B is a projective variety
of dimension 1. In fact, one can show that G/B ∼ = P1 . Moreover, G acts as
automorphisms of G/B (via left translation) which induces a morphism of
algebraic groups ϕ : G → Auta (P1 ) ∼
= PGL2 .
Let g ∈ ker(ϕ). Then ghB = hB for all h ∈ G which implies that g ∈ B h
$
for all h ∈ G. So (ker ϕ)◦ ⊆ ( h B h )◦ = R(G) = 1, by Proposition 6.16. In
particular ker ϕ is finite. We have already established that dim G ≥ 3, so
3 ≤ dim(G) = dim(ϕ(G)) ≤ dim(PGL2 ) = 3,
54 Structure of reductive groups

whence dim(G) = 3, and ϕ is surjective by Corollary 1.20 and Proposi-

tion 1.22.

We are thus led to consider carefully the structure of the rank 1 group
PGL2 = GL2 /{aI2 | a ∈ k × }. For this, we investigate the action of PGL2
on its Lie algebra Lie(PGL2 ). Write X̄ for the image of X ⊆ GL2 under
the natural projection π : GL2 → PGL2 . By dimension considerations one
sees that T2 is a Borel subgroup of PGL2 and T := D2 is a maximal torus
contained in T2 . A second Borel subgroup containing T is the image of the
lower triangular matrices. By Example 7.10 we have

Lie(PGL2 ) ∼
= gl2 /{aI2 | a ∈ k}.

Assume in what follows that char(k) = 2. Then we have gl2 /{aI2 | a ∈ k} ∼ =

sl2 (the 2 × 2-matrices of trace 0, see Exercise 10.26). As in Example 7.13
one checks that the adjoint representation of PGL2 acting on Lie(PGL2 ) is
simply the natural action of PGL2 on sl2 by conjugation. We determine the


c
root space decomposition. Fix an isomorphism ϕ : Gm → T , c → .
1
Then







0 1 0 c 0 0 0 0
Ad ϕ(c) = , Ad ϕ(c) = −1 ,
0 0 0 0 1 0 c 0


1 0
and is fixed by Ad ϕ(c). Define β, γ ∈ X(T ) by
0 −1

 

c c
β = c, γ = c−1 .
1 1

As X(T ) is written additively we have γ = −β, so sl2 decomposes as the

direct sum
sl2 = (sl2 )0 ⊕ (sl2 )β ⊕ (sl2 )−β

of common eigenspaces for Ad T , with

'
( '
( '
(
1 0 0 1 0 0
(sl2 )0 = , (sl2 )β = , (sl2 )−β = .
0 −1 0 0 1 0

Note that X(T ) = β.

In fact, the character β “occurs intrinsically” in PGL2 . That is, there
exists a subgroup of PGL2 whose Lie algebra corresponds to the second
subspace in the above decomposition: Consider the image of U2 in PGL2 .
 8.2 Semisimple groups of rank 1 55

This is the unipotent radical of our chosen Borel subgroup T2 and is therefore
normalized by T . Fix an isomorphism


1 c
u : Ga −→ U2 , u(c) = .
0 1

Then tu(c)t−1 = u(β(t)c) for all c ∈ k, t ∈ T . Finally, W = NG (T )/T is of



1 0 0 1
order two, with representatives , .
0 1 1 0
A similar calculation goes through when char(k) = 2.

We now consider G an arbitrary semisimple group of rank 1; for this,

we will refine the statement of Proposition 8.6 and obtain structural infor-
mation which resembles that which we found above for PGL2 . Let T be a
maximal torus contained in a Borel subgroup B of G. Let U = Bu and
choose n ∈ NG (T ) \ CG (T ), whence n2 ∈ CG (T ) and ntn−1 = t−1 for t ∈ T
(see Proposition 8.5). Write g := Lie(G).

Proposition 8.7 In the notation introduced above, we have:

(a) dim(G) = 3, dim(B) = 2, dim(U ) = 1, CG (T ) = T and U ∩ nU n−1 = 1.

(b) There is a unique α ∈ X(T ) \ {0} such that gα = Lie(U ) and g−α =
Lie(nU n−1 ), and g = Lie(T ) ⊕ gα ⊕ g−α .

Proof By Propositions 8.5 and 8.6, we have dim(G) = 3 and dim(G/B) = 1.

Since G is non-solvable, B is not a torus (see Corollary 6.9) and so dim(U ) =
1. By Proposition 6.18 and Theorem 6.19(a), CG (T ) is nilpotent and con-
nected and so lies in some Borel subgroup of G. As dim(B) = 2, either
CG (T ) = T or CG (T ) is a Borel subgroup of G. But in the latter case,
G would be solvable by Corollary 6.9, hence CG (T ) = T . This shows all
statements in (a) but the last.
By Theorem 3.2, U is isomorphic to Ga . Let u : Ga → U be an isomor-
phism. As T normalizes U , it acts by automorphisms. By Exercise 10.10
the automorphisms of Ga are simply multiplications by non-zero scalars, so
there exists α ∈ X(T ) such that tu(c)t−1 = u(α(t)c) for all t ∈ T and c ∈ k.
Moreover, α is non-trivial as CG (T ) = T . Now we differentiate the action of
T on U to obtain

Ad (t)Y = α(t)Y, for all t ∈ T, Y ∈ Lie(U ).

In particular, Lie(U ) lies in the α-eigenspace gα for T acting on g. As conjuga-

tion by n sends U to nU n−1 , Lie(nU n−1 ) then lies in g−α by Proposition 8.4
56 Structure of reductive groups

and the definition of n. The assertion of (b) now follows by comparing di-
mensions.
Finally, Lie(U ∩ nU n−1 ) ⊆ Lie(U ) ∩ Lie(nU n−1 ) = gα ∩ g−α = 0, so the
group U ∩nU n−1 is finite by Theorem 7.4(b), unipotent and normalized by T .
Hence this group lies in CG (T ) = T by Exercise 10.4. So U ∩nU n−1 = 1.
The above considerations finally enable one to establish the following clas-
sification of rank 1 semisimple groups ([66, Thm. 7.2.4]):
Theorem 8.8 Any semisimple group of rank 1 is isomorphic to SL2 or
PGL2 .
Example 8.9 For later use, let’s have a closer look at the case of SL2 . By
Example 8.2(2) we get the same root space decomposition
sl2 = (sl2 )0 + (sl2 )α + (sl2 )−α
= D2 ∩ SL
of Lie(SL2 ) = sl2 with respect to T
 2 as for the action of PGL2 ,
c
where now α ∈ X(T ) is defined by α = c2 . Note that α = 2X(T )
c−1
in this case.
Remark 8.10 In our calculations for PGL2 above we also found two roots
±β and a Weyl group of order two interchanging these two. That is, the
groups SL2 and PGL2 have the same Weyl group and “isomorphic root
systems”.
We finish this section with the following extension of our results on groups
of rank 1:
Proposition 8.11 Let G be connected reductive of semisimple rank 1,
T ≤ G a maximal torus, Z = Z(G) and G = [G, G]. Then:
(a) CG (T ) = T and thus Z ≤ T .
(b) G is semisimple of rank 1 with maximal torus T ∩ G , and G = G Z.
Proof By Proposition 6.20, the radical R(G) = Z(G)◦ is a torus, so con-
tained in T , G is semisimple and G ∩ R(G) is finite. As G/R(G) is assumed
to be semisimple of rank 1 we have dim(G/R(G)) = 3 by Proposition 8.6,
and G/R(G) is isomorphic to SL2 or PGL2 by Theorem 8.8. Now observe
that a maximal torus of SL2 or PGL2 is self-centralizing. So T and CG (T )
have the same image in G/R(G) under the canonical epimorphism, and hence
T ≤ CG (T ) ≤ T R(G) = T .
Since the product map R(G) × G → G has finite kernel we have
dim G ≤ dim G − dim R(G) = dim G/R(G) = 3.
 8.3 Structure of connected reductive groups 57

On the other hand, G is non-solvable (since G/R(G) and thus G is), so

dim G = 3 by Corollary 6.9. Now let T1 be a maximal torus of G . Comparing
ranks we get that T1 R(G) is a maximal torus of G, thus, after conjugating
we may assume T = T1 R(G). Using again that maximal tori of SL2 or PGL2
are self-centralizing we conclude that T ∩ G = T1 .

8.3 Structure of connected reductive groups

We start with the following characterization of the unipotent radical of a
general connected group, whose proof relies crucially on the structure results
for groups of semisimple rank 1 ([32, Thm. 26.1]); although it is easy to check
in the case of classical groups, its proof is rather involved:

Theorem 8.12 Let G be a connected linear algebraic group with maximal

torus T . Then Ru (G) is the identity component of the intersection of the
unipotent parts of the Borel subgroups which contain T .

Note that one inclusion is given by Proposition 6.16. For a reductive group,
in particular, we obtain:

Corollary 8.13 Let G be connected reductive.

(a) If S is a subtorus of G, then CG (S) is connected and reductive.

(b) If T is a maximal torus of G, then CG (T ) = T .

Proof Part (a) follows directly from the preceding theorem, applied to a
maximal torus T of G containing S, and Theorem 6.19. The second assertion
then follows from (a) with S = T , Proposition 6.18 and Example 6.17(2).

Example 8.14 Let G = GLn , S ≤ G a torus of dimension m = dim(S) (so

without loss of generality S ≤ Dn ). Then CG (S) ∼
= S × GLn−m is connected
reductive, and CG (S) = S if m = n (see Example 3.11).

We now study general reductive algebraic groups via their adjoint action
on the Lie algebra. Let G be a non-solvable connected reductive algebraic
group. We keep the notations T , g = Lie(G), Φ and W as in Section 8.2.
Let α ∈ Φ; then Tα := (ker α)◦ ≤ T is a subtorus of T by Proposition 3.9,
of codimension 1, and we let Cα := CG (Tα ) be its centralizer.

Proposition 8.15 Let G be non-solvable connected reductive. Then all Cα ,

α ∈ Φ, are connected reductive and non-solvable, and G = Cα | α ∈ Φ.
58 Structure of reductive groups

Proof The Cα , α ∈ Φ, are connected reductive by Corollary 8.13(a). By

Proposition 1.16, they generate a closed connected subgroup H of G. More-
over, by Proposition 7.14, Lie(Cα ) = Lie(CG (Tα )) = Cg (Tα ), which contains
both the root space gα and Lie(T ) (as Lie(T ) = Lie(CG (T )) = Cg (T )). Since
these subspaces span g, we have that Lie(H) = Lie(G), hence dim(H) =
dim(G) and so H = G by Proposition 1.22.
The reductive group Cα is solvable if and only if it is a torus. But we just
saw that the dimension of the centralizer of Tα is bigger than dim(T ).
Since Cα is connected reductive, non-solvable and Tα < R(Cα ), we see
that Cα is of semisimple rank 1 and so Proposition 8.11 shows that its
derived subgroup Gα := [Cα , Cα ] is semisimple of rank 1 with maximal
torus T1 := T ∩ Gα . Moreover:
Proposition 8.16 For α ∈ Φ we have Lie(Cα ) = Lie(T ) ⊕ gα ⊕ g−α and
dim gα = 1.
Proof We may decompose Lie(Cα ) = Lie(T ) ⊕ h with h a two-dimensional
T -invariant subspace which contains the root space gα of g. By Proposi-
tion 8.7(b) we have Lie(Gα ) = Lie(T1 ) ⊕ Lie(Gα )β ⊕ Lie(Gα )−β for some
β ∈ X(T1 ) \ {0}. Comparing with the T1 -eigenspaces on Lie(Cα ), we see
that h = gβ ⊕ g−β , whence h = gα1 ⊕ gα2 as T -module, with αi ∈ X(T ),
and α1 |T1 = β, α2 |T1 = −β. Since both αi have Tα in their kernel, they are
proportional. As one of them must be α the claim follows.
We are now ready to prove the first major structure theorem for reductive
groups (see also [66, Thm. 8.1.1, Cor. 8.1.2, 8.1.3] or [32, Thm. 26.3]):
Theorem 8.17 Let G be a connected reductive group, T ≤ G a maximal
torus of G, g = Lie(G) and Φ = Φ(G). Then:

(a) g = Lie(T ) ⊕ α∈Φ gα with dim gα = 1 for all α ∈ Φ, and Lie(T ) = g0 .
(b) dim G = dim Lie(G) = |Φ| + rk(G).
(c) For each α ∈ Φ there exists a morphism of algebraic groups uα : Ga →
G, which induces an isomorphism onto uα (Ga ) such that tuα (c)t−1 =
uα (α(t)c), for all t ∈ T, c ∈ k. If u is a morphism with the same proper-
ties, there is a unique a ∈ k × with u (c) = uα (ac) for c ∈ k. Moreover,
im(duα ) = gα .
(d) Uα := im(uα ) is the unique one-dimensional connected unipotent sub-
group of G normalized by T with Lie(Uα ) = gα .
(e) For w ∈ W with preimage n ∈ NG (T ) we have nUα n−1 = Uw.α .
(f) [Cα , Cα ] = Uα , U−α .
(g) G = T, Uα | α ∈ Φ.
 8.4 Structure of semisimple groups 59
$
(h) Z(G) = α∈Φ ker α.

Proof Part (a) and thus (b) follows from Propositions 7.14 and 8.16. Now let
uα : Ga → Gα be the morphism u defined in the proof of Proposition 8.7,
relative to the torus T1 . Then for all t ∈ T , c ∈ k, we have tuα (c)t−1 =
uα (α(t)c) as required, since T = T1 ·ker(α). The argument that im(duα ) = gα
is as in the proof of loc. cit. Let Uα := im(uα ). Suppose that u : Ga → G is
a homomorphism as in (c), then clearly V := im(u ) ≤ Cα . Since

tu (c)t−1 u (c)−1 = u (α(t)c)u (−c) = u ((α(t) − 1)c) for t ∈ T, c ∈ k,

we even have that V ≤ Gα , a one-dimensional connected unipotent subgroup

normalized by T . Then T1 V is a Borel subgroup of Gα , containing T1 . But by
Proposition 8.7 there are precisely two such Borel subgroups of Gα , namely
T1 Uα and T1 U−α . (For, any Borel subgroup containing T gives rise to a
corresponding weight space in Lie(Gα ).) Since T acts via α, we must have
V = Uα . Thus u−1 
α ◦ u is an automorphism of Ga , and hence given by
multiplication by a non-zero scalar. We now have the statement of (c).
If V is a one-dimensional closed connected unipotent subgroup normalized
by T , with Lie(V ) = gα , then the action of T on V must be given by the
character α and so V ≤ Cα = CG (Tα ), in which case the conclusion of (d)
follows as above.
For n ∈ NG (T ), nUα n−1 is a one-dimensional closed connected unipotent
subgroup normalized by T and T acts by wα. So the uniqueness statement
of (d) gives the result of (e). The statement of (f) follows from Theorem 8.8
applied to the rank 1 group Gα ; (g) follows from (f) and Proposition 8.15.
Finally, by (c) and (g), the right-hand side of (h) lies in Z(G). As Z(G) ≤
CG (T ) ≤ T by Corollary 8.13, the other inclusion also holds by part (c).

Definition 8.18 The one-dimensional subgroup Uα , α ∈ Φ, in Theorem

8.17(d) is called the root subgroup of G (with respect to T ) associated to α.
Similarly, the one-dimensional subspace gα of g is called a root subspace.

8.4 Structure of semisimple groups

Let’s now specialize to the situation of semisimple groups. For this, we study
the action of W in some more detail. For α ∈ Φ, let Cα = CG (Tα ) as in the
previous section. Since Tα lies in the center of Cα , the canonical epimor-
phism Cα → C̄α := Cα /Tα induces a bijection WCα (T ) ∼ = WC̄α (T /Tα ). By
Proposition 8.5, this latter Weyl group is of order 2. Recall that CCα (T ) =
60 Structure of reductive groups

T = CG (T ), by Corollary 8.13. Choose nα ∈ NCα (T ) \ CCα (T ), and let sα

be the image of nα in
NCα (T )/CCα (T ) ≤ NG (T )/CG (T ) = W.
We saw in Section 8.1 that sα acts naturally on X = X(T ) and on Y = Y (T ).
More precisely we have:
Lemma 8.19 Let α ∈ Φ. Then:
(a) There exists a unique α∨ ∈ Y such that sα .χ = χ − χ, α∨ α for all
χ ∈ X. In particular α, α∨  = 2.
(b) We have sα .γ = γ − α, γα∨ for all γ ∈ Y .
(c) im(α∨ ) ⊆ [Cα , Cα ].
Proof For χ ∈ X and t ∈ T we have
(idX − sα ).χ(t) = χ(t)/(sα .χ)(t) = χ(t)/χ(tsα ) = χ(tt−sα ).
But tt−sα lies in the rank 1 group Gα = [Cα , Cα ], isomorphic to SL2 or
PGL2 by Theorem 8.8. Let T1 := T ∩ Gα , a maximal torus of Gα by Propo-
sition 8.11(b). Since α|T1 is a root of T1 , we have 2X(T1 ) ⊆ α|T1  according
to Example 8.9 (in case Gα ∼ = SL2 ), respectively our calculations for PGL2
in Section 8.2, so 2χ|T1 = cχ α|T1 for some cχ ∈ Z. Thus
2χ(tt−sα ) = cχ α(tt−sα ) = cχ (idX − sα ).α(t) = 2cχ α(t),
showing that (idX − sα ).χ = cχ α. This defines a homomorphism X → Z,
χ → cχ . Evaluation on T1 , where sα acts by inversion, shows that cα = 2.
By Proposition 3.6 there exists a unique α∨ ∈ Y such that cχ = χ, α∨ ,
whence sα .χ = χ − χ, α∨ α and α, α∨  = cα = 2, thus proving (a).
For χ ∈ X we have by Lemma 8.3 and part (a)
χ, sα .γ = sα .χ, γ = χ − χ, α∨ α, γ
= χ, γ − α, γχ, α∨  = χ, γ − α, γα∨ .
Since this is true for all χ, Proposition 3.6 yields that sα .γ = γ − α, γα∨ .
For (c) note that α∨ (c)nα α∨ (c)−1 n−1α ∈ [Cα , Cα ] for c ∈ k × . But (a)
and (b) give that sα .α = −α , so nα α (c)−1 n−1
∨ ∨ ∨ ∨
α = α (c). So we have
∨ ∨
im(α ) = im(2α ) ⊆ [Cα , Cα ].
We say that α∨ is the coroot corresponding to α and set
Φ∨ := {α∨ | α ∈ Φ}.
Recall that an element s ∈ GL(V ), where V is a finite-dimensional real
vector space, is called a reflection along v ∈ V , if v is an eigenvector of s
 8.4 Structure of semisimple groups 61

with eigenvalue −1, and s fixes a hyperplane of V pointwise. Lemma 8.19

shows that the elements sα act as reflections on X(T ) ⊗Z R. Let’s note the
following result on the sα for future use (this is Exercise 10.31, or see [66,
Thm. 8.2.8]):
Proposition 8.20 The Weyl group W of a connected reductive group G
is generated by the sα , α ∈ Φ.
The structure of semisimple groups is now given as follows:
Theorem 8.21 Let G be a semisimple group, T, Φ, Uα as in Theorem 8.17.
Then:
(a) G = Uα | α ∈ Φ.
(b) G = [G, G].
(c) G has only finitely many minimal non-trivial closed connected normal
subgroups G1 , . . . , Gr . Moreover, [Gi , Gj ] = 1 for all i = j and Gi ∩

i=j Gj is finite.
(d) G = G1 · . . . · Gr and each Gi is a simple algebraic group.
Here a non-trivial semisimple algebraic group is called simple if it has no
non-trivial proper closed connected normal subgroups.
$
Proof Let Z = α∈Φ ker α. Then by Theorem 8.17(c) and (g), Z lies in
the center of G and hence is finite. Thus the roots Φ span a subgroup of
finite index in X(T ) and in particular span the vector space X(T ) ⊗Z R. We
claim that S := im(α∨ ) | α ∈ Φ equals T . Assume not. Then there exists
0 = χ ∈ X(T ) such that χ(S) = 1. Thus χ ◦ α∨ = id, whence χ, α∨  = 0
for all α ∈ Φ. Hence χ is fixed by all reflections sα by Lemma 8.19(a). This
contradicts the fact that the α ∈ Φ span X(T ) ⊗Z R. The assertion of (a)
now follows from Lemma 8.19(c) and Theorem 8.17(f) and (g). As shown in
the proof of Theorem 8.17, Uα ≤ [Cα , Cα ], so (b) follows.
If G = 1 has no proper non-trivial closed connected normal subgroups,
then G itself is a simple group and the statements of (c) and (d) are clear.
So assume the contrary and let H be a proper non-trivial closed connected
normal subgroup of G. As the radical of H is normal in G, it must be trivial
and so H is semisimple. We now decompose Φ into Φ1 := {α ∈ Φ | Uα ≤ H}
and Φ2 := Φ \ Φ1 ; as H is proper non-trivial and semisimple, Φ1 , Φ2 = ∅
by (a).
We claim that the subgroup H  := Uβ | β ∈ Φ2  commutes with H. For
this, let TH be a maximal torus of H; as all such are conjugate in H and
H is normal in G, we may assume that TH ≤ T . For β ∈ Φ2 we have from
Theorem 8.17(c) that tuβ (c)t−1 uβ (−c) = uβ ((β(t) − 1)c) ∈ H for all t ∈ TH
62 Structure of reductive groups

and c ∈ k. Hence β ∈ Φ2 implies that TH ≤ ker β, so TH centralizes Uβ .

Next, for α ∈ Φ1 set
uy (c) := [uβ (y), uα (c)]uα (c) = uβ (y)uα (c)uβ (−y) for c, y ∈ k.
Then uy (c) ∈ H for all y and tuy (c)t−1 = uy (α(t)c) for all t ∈ TH . Thus the
map Ga → G, c → uy (c), satisfies the conditions of Theorem 8.17(c), and
by the uniqueness result there exists ay ∈ k × such that uy (c) = uα (ay c) for
all c ∈ k. This defines a morphism of affine 1-space a : k → k, y → ay , with
a0 = 1, whose image does not contain 0. This can only be satisfied if a is the
constant function 1 and hence Uβ and Uα commute. Thus [H, H  ] = 1 and
G = H · H  by (a).
Finally we note that H ∩H  must be finite as it is a closed normal subgroup
of H which contains no Uα . One now proceeds by induction on the dimension
of G, using the fact that H and H  are both semisimple.
Finally we can improve the statement of Exercise 10.21.
Corollary 8.22 Let G be connected reductive. Then
G = [G, G]R(G) = [G, G]Z(G)◦ ;
in particular, rkss (G) = rk([G, G]) and rk(G) = rkss (G) + dim Z(G).
Proof We already argued that G/R(G) is semisimple (see Exercise 10.21).
Thus by Theorem 8.21(b)
G/R(G) = [G/R(G), G/R(G)] = [G, G]R(G)/R(G).
This result can be used to show that ker(Ad ) = Z(G), as claimed in
Theorem 7.12(c), see Exercise 10.32.
 9
The classification of semisimple
algebraic groups

The aim here is to achieve a classification of semisimple algebraic groups in

terms of combinatorial data. It is clear from the previous section that the
set of roots plays an essential role in the structure of reductive groups. We
now formalize this concept.

9.1 Root systems

Let G be a connected reductive group and T ≤ G a maximal torus. Then
associated to this we have a finite set of roots Φ ⊂ X := X(T ) with the finite
Weyl group W acting faithfully on X, preserving Φ (see Proposition 8.4).
Recall the group Y = Y (T ) of cocharacters of T and the pairing  ,  :
X × Y → Z defined in Section 3.2. We identify X and Y with subgroups
of E := X ⊗Z R and E ∨ := Y ⊗Z R, respectively, and denote the induced
pairing on E × E ∨ also by  , . The actions of W on X and on Y may be
extended to actions on E and E ∨ . Recall the reflections sα ∈ W introduced
in Section 8.4.
We first axiomatize the combinatorial properties satisfied by these data.

Definition 9.1 A subset Φ of a finite-dimensional real vector space E is

called an (abstract) root system in E if the following properties are satisfied:

(R1) Φ is finite, 0 ∈
/ Φ, Φ = E;
(R2) if c ∈ R is such that α, cα ∈ Φ, then c = ±1;
(R3) for each α ∈ Φ there exists a reflection sα ∈ GL(E) along α stabilizing
Φ;
(R4) (crystallographic condition) for α, β ∈ Φ, sα .β − β is an integral mul-
tiple of α.
64 The classification of semisimple algebraic groups

The group W = sα | α ∈ Φ is called the Weyl group of Φ. The dimension

of E is called the rank of Φ.
Since Φ is finite, generates E and is stabilized by W , the Weyl group of
an abstract root system is always finite. Thus, there exists a positive definite
W -invariant bilinear form ( , ) on E, which is unique up to non-zero scalars
on each irreducible W -submodule of E. We’ll always assume such a form
to have been chosen so that we can speak of lengths of vectors and angles
between them.
Let’s return to our reductive group G with root system Φ. We have seen
in Lemma 8.19 that sα acts as a reflection on E = X ⊗Z R. In fact, we have:
Proposition 9.2 Let G and Φ ⊂ X be as above. View Φ as a subset of
E := X ⊗Z R. Then Φ, together with {sα | α ∈ Φ} is an abstract root system
in ΦR with Weyl group W .
Moreover, if G is semisimple then ΦR = E.
Proof We first verify the axioms (R1)–(R4) of a root system. The first axiom
is clear (by our restriction of Euclidean space). (R2) follows from the fact that
Ccα = Cα for all c ∈ R× . (R3) is just Proposition 8.4 and Lemma 8.19. The
final axiom (R4) will be proved in Part II when we discuss the representation
theory of semisimple groups. See Lemma 15.4 and Example 15.5. The fact
that the sα generate the finite group W is Proposition 8.20.
The last statement was established in the proof of Theorem 8.21(a).
Abstract root systems can be classified. A crucial ingredient in this clas-
sification is the notion of a base:
Definition 9.3 Let Φ be an abstract root system in E. A subset ∆ ⊆ Φ
is called a base of Φ if it is a vector space basis of E and any β ∈ Φ is

an integral linear combination β = α∈∆ cα α with either all cα ≥ 0 or all
cα ≤ 0.
If ∆ is a base of Φ, then the subset
Φ+ := {α ∈ Φ | α is a non-negative linear combination over ∆}
is called the system of positive roots of Φ with respect to the base ∆.
We develop the basic theory of abstract root systems in the appendix.
As one of the first results, one obtains that bases exist and are essentially
unique:
Proposition 9.4 Let Φ be an abstract root system.
(a) There exists a base ∆ of Φ.
 9.1 Root systems 65

(b) Let ∆1 , ∆2 ⊆ Φ be two bases of Φ. Then there exists a unique w ∈ W

such that w(∆1 ) = ∆2 .
(c) If ∆ is a base, then W = sα | α ∈ ∆. Furthermore, for every α ∈ Φ
there is w ∈ W such that wα ∈ ∆.
See Proposition A.7 for (a), Theorem A.22 for (b) and Proposition A.11
for (c).
Example 9.5 We classify the two-dimensional abstract root systems. Let
E = R2 and Φ ⊂ E an abstract root system with α, β ∈ Φ linearly indepen-
dent. With respect to this basis, the reflections in α, respectively β, have the
form



−1 a 1 0
sα = , sβ = ,
0 1 b −1
with a, b ∈ Z by (R4). Then


ab − 1 −a
w := sα sβ =
b −1
lies in the finite Weyl group W of Φ, so its trace ab − 2 ∈ Z is the sum
of two roots of unity. This implies that ab ∈ {0, 1, 2, 3, 4}. Moreover, w is
not diagonalizable when ab = 4, so not of finite order. There remain four
possibilities for ab, and in fact all lead to root systems. Writing ψ for the
angle between α and β, we essentially have the following four cases (note
that, with Exercise 10.35(b) or Proposition A.1, ab above is just 4 cos(ψ)2 ):
(1) α ⊥ β, then Φ = {±α, ±β} with W = Z2 × Z2 the Klein four group.
(2) ψ = 2π/3, then Φ+ = {α, β, α + β}. This gives the root system of SL3
already encountered in Example 8.2(2), of type A2 (see Table 9.1 below).
Here W = S3 .
(3) ψ = 3π/4, then Φ+ = {α, β, α + β, 2α + β}. This gives the root system of
Sp4 , see Exercise 10.29(a), of type B2 = C2 , with W the dihedral group
of order 8.
(4) ψ = 5π/6, then Φ+ = {α, β, α + β, 2α + β, 3α + β, 3α + 2β}, with W the
dihedral group of order 12. This root system is said to be of type G2 . It
does not occur as a root system of a classical algebraic group.
In all cases, ∆ = {α, β} is a base of Φ.
A root system can be recovered from a base by Proposition 9.4(c), so
in order to describe a root system, it is sufficient to give its base. This is
most conveniently done with the associated Dynkin diagram which we now
describe. Its underlying graph has one node for each element of the base ∆,
66 The classification of semisimple algebraic groups

and two nodes corresponding to α, β ∈ ∆ are joined by an edge of multiplicity

mα,β as follows:
⎧
⎪
⎪0 if |(Zα + Zβ) ∩ Φ | = 2,
+
⎪
⎪
⎨1 if |(Zα + Zβ) ∩ Φ+ | = 3,
mα,β =
⎪
⎪ 2 if |(Zα + Zβ) ∩ Φ+ | = 4,
⎪
⎪
⎩
3 if |(Zα + Zβ) ∩ Φ+ | = 6.
(By Example 9.5 these are the only possibilities. Note that |(Zα+Zβ)∩Φ+ | =
o(sα sβ ), the order of the product of the corresponding simple reflections.)
The resulting graph is called the Coxeter diagram associated to Φ or to
W . It does not determine Φ uniquely. So in addition, whenever α, β are of
different lengths and joined by at least one edge, then we put an arrow on this
edge, pointing towards the shorter of the two (see Table 9.1 for examples).
This is called the Dynkin diagram of Φ. It can be shown that the base can
essentially be recovered from its Dynkin diagram (that is, up to changing
lengths in different connected components of the diagram), see [33, 11.1].
There is an obvious notion of isomorphism of root systems, and then one
has that two root systems are isomorphic if and only if their Dynkin diagrams
agree.
A root system Φ with base ∆ for which there exists a partition ∆ =
∆1  ∆2 into non-empty mutually orthogonal subsets is called decomposable;
by Proposition A.14 this is the case if and only if
Φ = (Z∆1 ∩ Φ)  (Z∆2 ∩ Φ) .
If Φ = ∅ and no such decomposition exists, Φ is said to be indecomposable.
It can easily be seen that a root system is indecomposable if and only if its
Weyl group acts irreducibly on the ambient real vector space (see Propo-
sition A.16). A root system is indecomposable if and only if its associated
Dynkin diagram is connected. For example, the root system (1) in Exam-
ple 9.5 is decomposable, while the other three are indecomposable.
It can be shown (see Exercise 10.33 or [32, Cor. 27.5]) that simple alge-
braic groups have indecomposable root systems (and conversely). Thus, as
a first step in the determination of simple groups one needs a classification
of indecomposable root systems (see [9, VI, §4] or [33, Thm. 11.4]):
Theorem 9.6 Let Φ be an indecomposable root system in some real vector
space E ∼
= Rn . Then up to isomorphism Φ is of one of the following types:
An (n ≥ 1), Bn (n ≥ 2), Cn (n ≥ 3), Dn (n ≥ 4), E6 , E7 , E8 , F4 , G2 ,
with corresponding Dynkin diagrams as shown in Table 9.1.
 9.1 Root systems 67

Table 9.1 Dynkin diagrams of indecomposable root systems

1 2 3 n 1 2 n−1 n
An Bn >
n≥1 n≥2
n−1
1 2 n−2 1 2 n−1 n
Dn Cn <
n≥4 n≥3

n 1 3 4 5 6
E6
1 2 1 2 3 4
G2 < F4 > 2

1 3 4 5 6 7 1 3 4 5 6 7 8
E7 E8

2 2

The numbering of nodes is as in [9, Pl. I–IX] and [33, p.58].

A root system or Dynkin diagram is said to be simply laced if all roots have
the same length, or equivalently, if all mα,β ∈ {0, 1}. Thus the indecompos-
able, simply laced Dynkin diagrams are those of types An , Dn , E6 , E7 , E8 .
Finite reflection groups arising from root systems only satisfying (R1)–
(R3), but not necessarily the crystallographic condition (R4), are the so-
called finite Coxeter groups. Interestingly, in addition to the Weyl groups
arising from Theorem 9.6 above, only the two-dimensional dihedral groups
and two further indecomposable cases, denoted H3 and H4 , in dimension 3,
respectively 4, occur. For more information on this and on the indecompos-
able root systems and their Weyl groups, see for example [34, §2.7–2.11].

Definition 9.7 We will often want to consider decomposable root systems.

We will use the following notation: if Φ is the orthogonal union of indecom-
posable root systems Φ1 , . . . , Φt , then we will refer to the root system Φ as
of type Φ1 Φ2 . . . Φt . If several of the Φi are isomorphic, we will write (Φi )j
for the union Φi  · · ·  Φi (j copies).

Example 9.8 We identify the root system of G = SLn and determine a

base and the corresponding positive roots. Here T = Dn ∩ SLn is a maximal
torus and B = Tn ∩ SLn is a Borel subgroup (see Example 6.7(2)). Then
Φ = {χij | i = j} where χij (diag(t1 , . . . , tn )) := ti t−1
j by Example 8.2(2). A
68 The classification of semisimple algebraic groups

base of Φ is given by

∆ := {χi,i+1 | 1 ≤ i ≤ n − 1}.

Indeed,
χij = χi,i+1 + χi+1,i+2 + . . . + χj−1,j for i < j,

so Φ+ = {χij | i < j} and Φ = Φ+  −(Φ+ ). In particular we have


3 j = i + 1,
|(Zχi,i+1 + Zχj,j+1 ) ∩ Φ | =
+
2 j > i + 1,

from which we see that the root system of SLn is of type An−1 .

9.2 The classification theorem of Chevalley

The aim here is to achieve a classification of semisimple algebraic groups in
terms of combinatorial data.
One might now hope that semisimple groups are already determined up
to isomorphism by their root systems. But unlike the case of simple complex
Lie algebras, there exist non-isomorphic simple algebraic groups having the
same root system, e.g., as we have seen SL2 , PGL2 both have root system
of type A1 (see Section 8.2). The additional piece of combinatorial data will
be provided by the coroots and the coroot lattice.
Let G be semisimple, T, W, Φ, X be as in the previous section. In Lemma
8.19 we constructed for each α ∈ Φ a unique cocharacter α∨ ∈ Y := Y (T ),
the coroot corresponding to α, such that α, α∨  = 2, and moreover im(α∨ )
is a maximal torus of the rank 1 semisimple group [Cα , Cα ].

Example 9.9 Let’s compute the coroots for SL2 and PGL2 .


(1) G = SL2 . The character group is generated by χ, where χ t
t−1 = t.


We have seen in Example 8.9 that Φ = {±α}, where α t t−1 = t2 . So


ZΦ = 2χ and X = Zχ. The coroot α∨ is given by α∨ : t → t t−1 ,
thus ZΦ∨ = Y .


(2) G = PGL2 . By Section 8.2 we have Φ = {±β}, where β t
1 = t; in
 9.2 The classification theorem of Chevalley 69

particular, ZΦ = X. The coroot β ∨ is then given by



∨ t t2
β :t→ = .
t−1 1

In this case ZΦ∨ = 2Y .

We are led to introduce the following combinatorial structure:

Definition 9.10 A quadruple (X, Φ, Y, Φ∨ ) is called a root datum if

(RD1) X ∼= Zn ∼ = Y , with a perfect pairing  ,  : X × Y → Z as in

Proposition 3.6;
(RD2) Φ ⊆ X, Φ∨ ⊆ Y are abstract root systems in ZΦ ⊗Z R, respectively
ZΦ∨ ⊗Z R;
(RD3) there exists a bijection Φ → Φ∨ such that α, α∨  = 2; and
(RD4) the reflections sα of the root system Φ, respectively s∨ ∨
α of Φ are
given by
sα .χ = χ − χ, α∨ α for all χ ∈ X,

sα∨ .γ = γ − α, γα∨ for all γ ∈ Y.

It follows easily that whenever (X, Φ, Y, Φ∨ ) is a root datum then the Weyl
groups of Φ and of Φ∨ are isomorphic via sα → sα∨ (see Exercise 10.34).
The preceding definition is justified by the following result:

Proposition 9.11 Let Φ be the root system of a connected reductive group

G with respect to the maximal torus T , with Weyl group W , and set Φ∨ =
{α∨ | α ∈ Φ}. Then (X(T ), Φ, Y (T ), Φ∨ ) is a root datum.

Proof We already saw in Proposition 9.2 that Φ is an abstract root system

in ΦR ≤ XR := X(T ) ⊗Z R. Furthermore in Lemma 8.19 we constructed
a surjective map Φ → Φ∨ , α → α∨ , satisfying properties (RD3) and (RD4)
with respect to the pairing coming from Proposition 3.6. Suppose that for
α, β ∈ Φ we have α∨ = β ∨ . Then

sα sβ (χ) = χ + χ, α∨ (α − β) for all χ ∈ X(T ).

Since α − β, α∨  = 0 we have (sα sβ − idX )2 = 0, so all eigenvalues of the

map sα sβ : XR → XR are 1. But this is a transformation of finite order on
the real vector space XR and so sα sβ = 1 and α = β. Then Φ∨ is an abstract
root system by Exercise 10.35.
70 The classification of semisimple algebraic groups

Example 9.12 Let G be a connected reductive algebraic group with root

datum (X, Φ, Y, Φ∨ ) with respect to the maximal torus T , and let T  ≤ T
be a subtorus. Then T  has root datum (X  , ∅, Y  , ∅), with

Y  := {γ ∈ Y | γ(Gm ) ≤ T  } = Y (T  )

and
X  := {χ|T  | χ ∈ X} ∼
= X/Ann(Y  ) = X(T  ),

where, for any submodule Z ≤ Y we set

Ann(Z) := {χ ∈ X | χ, γ = 0 for all γ ∈ Z}.

There is a natural notion of isomorphism of root data (see [41, II.1.13]).

Then the following fundamental result classifies semisimple groups (see [66,
9.6.2,10.1.1]):

Theorem 9.13 (Chevalley Classification Theorem) Two semisimple linear

algebraic groups are isomorphic if and only if they have isomorphic root data.
For each root datum there exists a semisimple algebraic group which realizes
it. This group is simple if and only if its root system is indecomposable.

The groups with root system of type An , Bn , Cn or Dn are called groups

of classical type; the remaining simple groups are called groups of exceptional
type.
One can determine precisely the possible root data. This involves intro-
ducing one further finite group.
By (R1) and Proposition 9.2, ZΦ is of finite index in X, and ZΦ∨ is of
finite index in Y , when G is a semisimple group. For simple groups G with
root system Φ not of type D2n , the root datum, and hence the isomorphism
type of G, is determined up to isomorphism by Φ and the index |X : ZΦ|.
In general, let Ω = Hom(ZΦ∨ , Z). Restriction gives a natural homomor-
phism
X∼ = Hom(Y, Z) → Hom(ZΦ∨ , Z) = Ω

which is injective. Thus we may view ZΦ ⊆ X ⊆ Ω.

The finite group Λ := Λ(Φ) := Ω/ZΦ does not depend on X and is called
the fundamental group of the root system Φ. The root data with fixed root
system Φ are classified by subgroups X of Ω satisfying ZΦ ⊆ X ⊆ Ω (up to
automorphisms of Ω stabilizing Φ) hence by subgroups X/ZΦ ≤ Ω/ZΦ of
the fundamental group.
 9.2 The classification theorem of Chevalley 71

Definition 9.14 Let G be a semisimple algebraic group, with X, Φ, Ω as

above. Then Λ(G) := Ω/X is called the fundamental group of G. If X = Ω,
so Λ(G) = 1, then G is said to be simply connected ; if X = ZΦ then G is
said to be of adjoint type. We write Gad , Gsc for an algebraic group with
given root system Φ of adjoint, respectively simply connected type. (See
Remark 9.17 for the origin of this terminology.)

A surjective homomorphism ϕ : G → H of algebraic groups with finite

kernel is called an isogeny. If such a morphism ϕ exists, one says that G and
H are isogenous. If G is connected then ker ϕ is central by Exercise 10.4, and
if moreover G is reductive, then ker ϕ lies in all maximal tori. The various
semisimple algebraic groups G with fixed root system Φ are called the isogeny
types corresponding to Φ due to the following result:

Proposition 9.15 Let G be semisimple with root system Φ. Then there

exist natural isogenies
π π
Gsc −→
1
G −→
2
Gad

from a simply connected group Gsc and to an adjoint group Gad , each with
root system Φ, with ker(π1 ) ∼
= Λ(G)p , ker(π2 ) ∼
= (Λ(Gad )/Λ(G))p , where
p = char(k), and such that dπi is an isomorphism for i = 1, 2.

Proof for the case that char(k) is prime to |Λ(Φ)| Let T be a maximal to-
rus of G with corresponding root system Φ. Then Z := Z(G) ≤ CG (T ) = T ,
$
so Z = α ker(α). Since Z lies in the kernel of the adjoint representation, G
acts via G/Z on Lie(G), so the roots of G and G/Z are the same under the
natural inclusion X(T /Z) ≤ X(T ) induced by the surjection T → T /Z (see
Exercise 10.13). In particular, for any subgroup S ≤ Z, we have inclusions
ZΦ ≤ X(T /Z) ≤ X(T /S) ≤ X(T ). In fact, as gcd(char(k), |Λ|) = 1, Propo-
sition 3.8 implies that X(T )/ZΦ ∼ = Z. Hence, X(T /Z)/ZΦ = 0 and G/Z is
of adjoint type; so we take π2 : G → G/Z to be the natural surjection.
On the other hand, starting with Gsc and Λ1 ≤ X(Tsc )/ZΦ the funda-
mental group of G, then with S = Λ⊥ 1 ≤ Z(Gsc ) the quotient Gsc /S has
root datum with fundamental group Λ1 . With π1 : Gsc → Gsc /S we then
have G ∼ = Gsc /S. The assertion that dπi is an isomorphism follows from
Theorem 7.9(b).

For the general case, the claim follows from the isogeny theorem [66,
Thm. 9.6.5] which asserts that any morphism of root data induces an isogeny
of corresponding semisimple groups.
Table 9.2 gives a list of the possible isogeny types for simple groups and
an identification with various classical groups (see e.g. [73, p.45]).
72 The classification of semisimple algebraic groups

Table 9.2 Isogeny types of simple algebraic groups

Φ Λ(Φ) Gsc Gad in between

An−1 , n ≥ 2 Zn SLn PGLn SLn /Zd (d|n)
Bn , n ≥ 2 Z2 Spin2n+1 SO2n+1 −
Cn , n ≥ 2 Z2 Sp2n PCSp2n −
Dn , n ≥ 3 odd Z4 Spin2n PCO◦2n SO2n
Dn , n ≥ 4 even Z2 × Z2 Spin2n PCO◦2n SO2n , HSpin2n
G2 1 G2 −
F4 1 F4 −
E6 Z3 (E6 )sc (E6 )ad −
E7 Z2 (E7 )sc (E7 )ad −
E8 1 E8 −

Example 9.16 (1) In Example 9.9 we saw that X = ZΦ for PGL2 , so

PGL2 is of adjoint type. On the other hand for SL2 , X = Ω, so SL2 is
simply connected.
(2) For Φ of type E6 we have |Ω/ZΦ| = 3, so there exist two isogeny types
of groups of type E6 .
(3) The groups SO2n with root system Φ of type Dn are neither adjoint nor
simply connected. Here, the fundamental group Λ of Φ is of order 4, and
SO2n corresponds to a subgroup of Λ of order 2. If n is even, then Λ
contains two other subgroups of order 2 which correspond to isomorphic
simple groups, the so-called half-spin groups HSpin2n . If n = 4 both
are also isomorphic to SO8 . (These isomorphisms are induced by graph
automorphisms of the Dynkin diagram of type Dn , see Theorem 11.12
below.) The groups of adjoint type Dn are obtained as the quotient
PCO◦2n := CO◦2n /Z(CO◦2n ) of the conformal orthogonal group (see Sec-
tion 1.2) modulo its one-dimensional central torus, or also as the quotient
SO2n /Z(SO2n ).
(4) The orthogonal groups of simply connected type are the so-called spin-
groups Spinn . Their smallest faithful matrix representation is of dimen-
sion 2s , with s =  n−1
2 . They are best constructed and studied as
subgroups of the units in Clifford algebras.

For more information on the various types of classical groups, see for
example the books by Dieudonné [21], Grove [30], or Goodman and Wallach
[27] (for k of characteristic 0).

Remark 9.17 Steinberg has shown that Gsc is the universal perfect central
extension, in the category of abstract groups, of any semisimple group with
 9.2 The classification theorem of Chevalley 73

root system Φ [73, Thm. 10]. There is no such universal central extension
in the category of algebraic groups since there exist bijective morphisms of
simple groups which are not isomorphisms. These isogenies will play a cru-
cial role in Part III. Over the field k = C of complex numbers the groups Gsc
are in fact simply connected in the complex topology, and for G of arbitrary
type, Λ(G) is the topological fundamental group [73, Thm. 13]. Finally we
note that the group Gad is isomorphic to its image under the adjoint repre-
sentation, which explains the denomination. (It is not however the case that
the image of the adjoint representation of any group is necessarily of adjoint
type. See Exercise 10.37.)
There is another universal property of groups of simply connected type
which will be needed later on:
Proposition 9.18 Let G be semisimple of simply connected type. Then
whenever π : H1 → H2 is an isogeny of semisimple groups with dπ an
isomorphism, any isogeny ϕ : G → H2 lifts to an isogeny ψ : G → H1 such
that ϕ = π ◦ ψ.
Proof Let G̃ := {(g, h) ∈ G × H1 | ϕ(g) = π(h)}, a closed subgroup of
G × H1 . The projection pr1 : G̃ → G onto the first factor is a surjective
morphism with finite central kernel 1 × C, where C = ker(π), hence its
restriction to G1 := G̃◦ is an isogeny.
pr2
G1 ⊆ G × H1 /H
r8 1
r
ψ
r
pr1
r π
 r r 
G / H2
ϕ

In particular G1 is semisimple and as pr1 has central kernel, it induces an

isomorphism of root systems and so G1 has the same type root system as
G. By Chevalley’s Classification Theorem 9.13 this shows that G1 is also of
simply connected type and thus pr1 |G1 is injective. Moreover, the differential
dpr1 (which is just the projection onto Lie(G)) has kernel ker(dπ) which is
0 by assumption, hence dpr1 is an isomorphism. By the criterion in Propo-
sition 7.7(c), pr1 |G1 is an isomorphism, so we may take ψ = pr2 ◦ (pr1 |G1 )−1
which is surjective since H1 is connected.
 10
Exercises for Part I

Throughout we take k to be an algebraically closed field.

Exercise 10.1 Let X be a topological space. Show the following assertions

of Proposition 1.10:

(a) Z ⊂ X is irreducible if and only if its closure Z̄ is irreducible.

(b) Let f : X → Y be a continuous map to a topological space Y . If X is
irreducible then so is the image f (X).

Exercise 10.2 Show that the set {(x, y) ∈ k 2 | xy = 0} is not irreducible

but is connected in the Zariski topology.

Exercise 10.3 Let G be a linear algebraic group.

(a) Show that CG (x) is a closed subgroup, for x ∈ G.

(b) Show that Z(G) is a closed subgroup of G.
(c) If H is a subgroup of G, then so is its closure H̄.
(d) If H is a subgroup of G containing a non-empty open subset of its closure,
then H is closed.

[Hint: For (c) first show that H H̄ ⊆ H̄. For (d), first show that if U, V ⊆ G are
dense open subsets, then U V = G.]

Exercise 10.4 Let G be a linear algebraic group and let N  G be a finite

normal subgroup. Let H ≤ G be a closed connected subgroup. Show that N
lies in CG (H), the centralizer of H in G. (See also Exercise 10.18.)
[Hint: For x in N consider the orbit map H → N , h → hxh−1 .]

Exercise 10.5 Show that CSp2n = Sp2n · Z(CSp2n ).

Exercise 10.6 Show that a non-trivial connected nilpotent group has a

center of dimension at least 1.
 Exercises for Part I 75

[Hint: Look at the last non-trivial term in the descending central series and use
Proposition 1.18.]

Exercise 10.7 Show that each of the groups Tn , Un , Dn and SLn is

connected.
Exercise 10.8 Show that GO2n is not connected when char(k) = 2.
[Hint: Determine GO2 explicitly to find an element of determinant −1.]
%n+1& %n&
Exercise 10.9 Show that dim(Tn ) = 2 , dim(Un ) = 2 , dim(Dn ) = n.
Exercise 10.10 Show that the (algebraic group) automorphisms of Ga
are the multiplications by non-zero elements of k.
Exercise 10.11 Show that
End(Gm ) := {ϕ : Gm → Gm | ϕ a morphism of algebraic groups} ∼
= Z;
conclude that the group of algebraic group automorphisms of Ga is Z2 .
Exercise 10.12 Let G be a linear algebraic group.
(a) Show that the set Gu of unipotent elements in G is closed.
(b) Show, by example, that the set Gs of semisimple elements in G is not
necessarily closed nor open.
(c) Show that the conjugacy class of a semisimple element in GLn is closed.
[Hint: For (a) consider the characteristic polynomials of unipotent elements.]

Exercise 10.13 Let ϕ : S → T be a surjective homomorphism of tori.

Show that ϕ induces a natural injective group homomorphism ϕ∗ : X(T ) →
X(S).
Exercise 10.14 (Dedekind’s Lemma) Let G be a group, K a field. Then
any finite subset of Hom(G, K × ) is linearly independent.
[Hint: See [32, 16.1].]

Exercise 10.15 Let T be a torus, H ≤ T a closed subgroup, and X1 ≤

X(T ) a subgroup. Show the following:
(a) X(H) forms a basis of k[H].
(b) Restriction defines an isomorphism
X(H) ∼
= X(T )/H ⊥ .
(c) If H1 ≤ H is a subgroup of finite index, then H1⊥ /H ⊥ is finite.
(d) X1⊥⊥ /X1 is a finite p-group, where p = char(k); in particular, X1⊥⊥ = X1
if X(T )/X1 has no p-torsion.
76 Exercises for Part I

[Hint: By Example 3.5, X(T ) is a basis of k[T ]. The embedding ϕ : H → T

defines an epimorphism k[T ] → k[H], which is just restriction. Now restrictions
of characters are again characters. Use Exercise 10.14. For (c) note that X(T ) is
a finitely generated Z-module. For (d) let x ∈ X1⊥⊥ \ X1 . Apply the elementary
divisor theorem to X1 , x and again use that characters are linearly independent.]

Exercise 10.16 The purpose of this exercise is to point out that the struc-
ture of connected unipotent groups is difficult to classify.
(a) Show that
⎧⎛ ⎞  ⎫
⎨ 1 a b  ⎬

H1 = ⎝0 1 a⎠  a, b ∈ k
⎩  ⎭
0 0 1
is a closed connected two-dimensional commutative unipotent algebraic
group.
(b) Show that H1 ∼ = Ga × Ga as algebraic groups if and only if k is a field
of characteristic different from 2.
(c) Show that
⎧⎛ ⎞ ⎫
⎪ 1 t u s  ⎪
⎪
⎨⎜  ⎪
⎬
0 1 0 u ⎟
H2 = ⎜ ⎝
⎟  s, t, u ∈ k ≤ Sp4
⎠ 
⎪
⎪ 0 0 1 −t  ⎪
⎪
⎩  ⎭
0 0 0 1
is a three-dimensional connected unipotent algebraic group.
(d) Determine [H2 , H2 ]; conclude that H2 is abelian if and only if char(k) =
2.
[Hint: For (b) note that if char(k) = 2, H1 has elements of order 4.]

Exercise 10.17 The goal of this exercise is to see that PGLn , the projec-
tive linear group in dimension n, is a linear algebraic group.
Set PGLn = GLn /Z, where Z = {cIn | c ∈ k × }, as abstract group. Let V
be the n-dimensional vector space over k on which GLn naturally acts and
V ∗ = Hom(V, k) the dual space. Consider the action of GLn on V ⊗ V ∗ . This
defines a group homomorphism ρ : GLn → GLn2 .
(a) Show that ρ is a morphism of algebraic groups and hence its image is a
closed subgroup H of GLn2 .
(b) Show that ker ρ = Z.
Hence we conclude that PGLn is isomorphic, as an abstract group, to the
closed subgroup H of GLn2 and thus can be given the structure of a linear
algebraic group.
 Exercises for Part I 77

Exercise 10.18 Let G be a linear algebraic group, X a G-space.

(a) Show that the fixed point set X g := {x ∈ X | g.x = x} is closed for any
g ∈ G. Conclude that X G := {x ∈ X | g.x = x for all g ∈ G} is closed.
(b) Let Y and Z be closed subsets of X and set TranG (Y, Z) := {x ∈ G |
x.Y ⊆ Z} (the transporter of Y into Z). Show that TranG (Y, Z) is a
closed subset of G.
(c) Show that for H a closed subgroup of G, NG (H) is a closed subgroup.

[Hint: For (a), consider the morphism ψ : X → X × X, x → (x, g.x), and use that
the diagonal in X × X is closed. For (b), consider the maps ϕy : G → X given by


x → x.y. Then TranG (Y, Z) = y∈Y (ϕ−1 y (Z)).]

Exercise 10.19 (Maximal tori and Borel subgroups of classical groups)

(a) Show that rk(Sp2n ) = rk(SO2n+1 ) = n.

(b) Show that Sp2n ∩ T2n (respectively SO2n+1 ∩ T2n+1 ) is a Borel subgroup
of Sp2n (respectively of SO2n+1 ).

Exercise 10.20 Show that a connected linear algebraic group with nilpo-
tent Borel subgroup is solvable. Conclude that any two-dimensional con-
nected group is solvable.
[Hint: Consider a counterexample of minimal dimension and use Exercise 10.6 and
Proposition 6.8.]

Exercise 10.21 Show that for a connected group G, G/R(G) is semisim-

ple, and G/Ru (G) is reductive.

Exercise 10.22 (A semisimplicity criterion)

(a) Let G be a linear algebraic group. Show that the radical R(G) is equal
$
to ( B B)◦ , where B runs over all Borel subgroups of G.
(b) Show that Sp2n is semisimple.

Exercise 10.23 (Algebraic groups consisting of semisimple elements)

(a) Let G be a connected linear algebraic group whose elements are semisim-
ple. Show that G is a torus.
(b) Give an example of an algebraic group (closed subgroup of GLn ) which
consists of semisimple elements but which is not conjugate to a subgroup
of Dn , the group of diagonal matrices.

[Hint: For (a) use Exercise 10.20.]

78 Exercises for Part I

Exercise 10.24 Let V be the three-dimensional vector space on which

SL3 naturally acts, with a fixed basis e1 , e2 , e3 . Let P = StabSL3 (e1 , e2 ).
Let
S = {g ∈ P | g.ei = λg,i ei , for some λg,i ∈ k × , 1 ≤ i ≤ 3}

(so S is a torus of P ). Find NP (S), NP (S)◦ , CP (S), and CP (S)◦ and verify
that NP (S)/CP (S) is finite.

Exercise 10.25 Let G be a linear algebraic group. Show that the inversion
morphism i : G → G, g → g −1 , on G has differential di(X) = −X, for
X ∈ Lie(G).
[Hint: Apply Example 7.8(1) and Proposition 7.7(a) to the morphism µ ◦ (i, id) :
G × G → G.]

Exercise 10.26 In this exercise, we identify the Lie algebra of various

algebraic groups.

(a) The Lie algebra of Tn , as well as that of Un and Dn , may be identi-

fied with a subalgebra of gln . Show that it equals the set of all upper
triangular, strictly upper triangular, diagonal matrices, respectively.
(b) Show that the Lie algebra of SLn may be identified with the Lie-sub-
algebra sln of gln of n × n trace zero matrices.
(c) Show that Lie(PGLn ) ∼ = Lie(SLn ) if char(k) does not divide n.
Exercise 10.27 Assume k to be a field of characteristic p > 0, G = SL3 ,
and let
⎛ ⎞
1 t tp
ϕ : Ga → G, ϕ(t) = ⎝0 1 0⎠.
0 0 1
(a) Show that ϕ defines an isomorphism of algebraic groups Ga ∼
= im(ϕ).
Set H = im(ϕ).
(b) Determine Lie(H), as a subalgebra of gl3 .
(c) Show that CG (H) is not equal to

CG (Lie(H)) := {g ∈ G | Ad (g)X = X for all X ∈ Lie(H)}.

(d) Show that if char(k) = 3, then Lie(CG (H)) is not equal to

CLie(G) (H) := {X ∈ Lie(G) | Ad (h)X = X for all h ∈ H}.

Exercise 10.28 (a) Let G be a linear algebraic group. Show that Z(G) ≤
ker(Ad ).
(b) Determine ker(Ad ) for each of the following groups: GLn , Un , Tn .
 Exercises for Part I 79

(c) This example shows that ker(Ad ) may be larger than Z(G). Let k be a
⎧⎛ ⎞ ⎫
⎨ a 0 0  ⎬
field of characteristic p > 0. Let G = ⎝0 ap b ⎠  a ∈ k × , b ∈ k .
⎩ ⎭
0 0 1 
Show that Z(G)  ker(Ad )  G.

Exercise 10.29 Let G = Sp2n . Then Lie(G) = {X ∈ gl2n | X tr J2n =

−J2n X}, where J2n is as in the definition of symplectic groups in Section 1.2.
Let T = D2n ∩ Sp2n be a maximal torus of Sp2n (see Exercise 10.19).

(a) In the case n = 2, find the roots and root subspaces of Lie(G). For each
root α, exhibit a one-dimensional closed subgroup Uα ≤ G whose Lie
algebra is the corresponding root subspace.
(b) Generalize to the case of arbitrary n. Show that the root system of Sp2n
is of type Cn . Conclude that dim Sp2n = 2n2 + n.

Exercise 10.30 Let T be a maximal torus of a connected reductive group

G, with character group X, cocharacter group Y and Weyl group W .

(a) Show that W acts faithfully on both X and Y .

(b) Prove that w.χ, γ = χ, w−1 .γ for all w ∈ W, χ ∈ X, γ ∈ Y .

Exercise 10.31 Show that the Weyl group W of a connected reductive

group G is generated by the sα , α ∈ Φ.
[Hint: Argue by induction on dim(G). For w ∈ W , with preimage n ∈ NG (T ),
consider the homomorphism ψ : T → T , t → ntn−1 t−1 . If ψ is not surjective, then
S := ker(ψ)◦ is a non-trivial subtorus of T with n ∈ C := CG (S). If C < G, use
induction for the maximal torus C ∩ T of C. If C = G, then S ≤ Z(G) is normal,
and induction applies to G/S. If ψ is surjective, use Exercise 10.13 and argue that
there exists x ∈ X(T ) ⊗Z R such that (w − 1)x = α. Then show that sα x = wx;
hence sα x has eigenvalue 1 and we can apply the argument of the first case to this
element.]

Exercise 10.32 Show that for a connected reductive group G, ker(Ad ) =

Z(G).
[Hint: By Exercise 10.28 we have Z(G) ≤ ker(Ad ). If ker(Ad ) is finite, use Ex-
ercise 10.4. For G semisimple, apply Theorem 8.17 to see that a maximal torus
of any simple factor Gi of G acts non-trivially on some root space in g. For the
general case, R(G) = Z(G)◦ ≤ ker(Ad ), and Ad induces the adjoint represen-
tation of G/R(G) with kernel Z(G/R(G)). Since G/R(G) is semisimple, we have
Z(G/R(G))◦ = 1.]

Exercise 10.33 Let G be semisimple, G = G1 · · · Gr its decomposition

80 Exercises for Part I

into simple components as in Theorem 8.21(d). Then the root system Φ

of G decomposes into an orthogonal disjoint union Φ = Φ1  . . .  Φr of
indecomposable root systems Φi , such that Gi has root system Φi .
In particular, simple algebraic groups have indecomposable root systems.
Exercise 10.34 Let (X, Φ, Y, Φ∨ ) be a root datum. Then the Weyl groups
of Φ and of Φ∨ are isomorphic via sα → sα∨ .
[Hint: Define an action of the Weyl group W of Φ on Y by: w.γ is the unique
element such that χ, w.γ = w−1 .χ, γ for all χ ∈ X. Now mimic the proof of
Lemma 8.19 to see that this sends sα to sα∨ .]

Exercise 10.35 (A realization of the dual root system inside XR ) Let

X, respectively Y , be the character, respectively cocharacter, group of a
connected reductive group, Φ ⊂ X the root system with Weyl group W and
{α∨ | α ∈ Φ} ⊂ Y the coroots as constructed in Lemma 8.19. We write
V := XR = X ⊗Z R.
(a) The pairing between X, Y induces a natural isomorphism YR ∼ = V ∗ :=
Hom(V, R).
(b) Let ( , ) : V × V → R be a W -invariant scalar product on V . Show that
the reflection sα along α is given by sα (x) = x − 2(x, α)α.
(c) The scalar product ( , ) induces an isomorphism i : V ∗ → V , via x, y =
(x, i(y)) for x ∈ X, y ∈ Y , which satisfies i(α∨ ) = 2α/(α, α) for α ∈ Φ.
(d) Show that Φ̃ := {i(α∨ ) | α ∈ Φ} is a root system in V , and hence Φ∨ is
a root system in YR .
[Hint: For (c) compare α∨ and 2α/(α, α) using Lemma 8.19.]

Exercise 10.36 Show that

[GLn , GLn ] = SLn , [CSp2n , CSp2n ] = Sp2n , [CO◦n , CO◦n ] = SOn .
[Hint: Compare the root systems of the two groups in each case and apply Propo-
sition 6.20(c) and Table 9.2. For GLn , this is also easy to show by direct matrix
calculation.]

Exercise 10.37 Let G = Sp2 , g = Lie(G), char(k) = 2, and Ad : G →

GL(g) be the adjoint representation. Show that Ad (G) is a simply connected
simple algebraic group of type A1 .
 PART II

SUBGROUP STRUCTURE AND

REPRESENTATION THEORY OF
SEMISIMPLE ALGEBRAIC GROUPS

In this part we develop the basic structure and representation theory of

semisimple and more generally of reductive algebraic groups. The aim of
this part is twofold. First, we want to provide sufficient information in order
to be able to formulate the classification results on maximal subgroups of
simple algebraic groups presented in Chapters 18 and 19 and to sketch the
important ingredients in their proof. Secondly, we prepare the notions and
results necessary for the investigation of the finite groups of Lie type in
Part III.
It has already been indicated in Chapter 8 that the structure and classifi-
cation of reductive groups is somehow controlled by their maximal connected
solvable subgroups, the Borel subgroups. This will become even more appar-
ent here. The crucial starting point, on which most of the development in
this part will be based, is the structure result for Borel subgroups of reduc-
tive groups in Theorem 11.1. From this, we first derive the so-called Bruhat
decomposition, which is of a more topological nature, and then also the
more combinatorial BN-pair structure of connected reductive groups. The
latter leads to the notion of parabolic subgroups and their Levi decomposi-
tion, which are studied in Chapter 12. In the framework of classical groups,
parabolic subgroups arise very naturally as the stabilizers of isotropic sub-
spaces in the natural matrix representation.
Chapter 13 is devoted to the study of subsystem subgroups, that is, sub-
groups normalized by a maximal torus. These comprise the Levi subgroups
already encountered before, but also the centralizers of semisimple elements.
82

We derive the classification of subsystem subgroups by the algorithm of

Borel and de Siebenthal and also treat the case of small characteristic. In
Chapter 14 we then go on to consider properties of centralizers and con-
jugacy classes. We give a parametrization of semisimple conjugacy classes,
and prove some basic properties of regular semisimple elements. We intro-
duce the important concepts of bad primes and torsion primes and indicate
their relation to connectedness of centralizers.
In Chapters 15 and 16 we give an introduction to the representation theory
of reductive algebraic groups over their underlying field, the so-called weight
theory up to Steinberg’s tensor product theorem. We also study two impor-
tant ingredients for the determination of maximal subgroups: the character-
ization of the self-dual irreducible modules in terms of their highest weight
and certain properties of restrictions to Levi subgroups.
In Chapter 17 we return to the structure theory of parabolic subgroups to
study their action on the unipotent radical, which gives rise to an important
class of naturally occurring representations, the so-called internal modules.
Moreover, we sketch in Theorem 17.10 the proof of the important result of
Borel and Tits that the normalizer of any non-trivial unipotent subgroup
in a connected reductive group is contained in a proper parabolic subgroup,
and give some applications to the study of maximal subgroups.
In Chapter 18 we state the reduction result for maximal subgroups of
simple algebraic groups of classical type, sketch its proof and give some
indications on the complete classification of maximal subgroups of positive
dimension. The maximal subgroups of simple groups of exceptional type have
to be investigated by different methods, as we explain in Chapter 19.
Throughout this part, we consider algebraic groups defined over a fixed
algebraically closed field k of characteristic p ≥ 0.
 11
BN-pairs and Bruhat decomposition

We fix the following notation. Let G be a connected reductive algebraic

group, T ≤ G a maximal torus with character group X and Weyl group W .
Let Φ be the root system of G with respect to T . For each α ∈ Φ there is an
associated root subgroup Uα ≤ G and a reflection sα ∈ W (see Sections 8.3
and 8.4).
We start by refining the assertion that G = T, Uα | α ∈ Φ from Theo-
rem 8.17(g). There will be two such results, the first giving a product decom-
position of the Borel subgroup B, the second a partition of G into double
cosets with respect to B. This will also lead to the determination of the
algebraic automorphism group of a semisimple group.

11.1 On the structure of B

Our first main result here not only gives structural information on Borel
subgroups but also provides a strong connection between Borel subgroups
containing the given maximal torus T and bases ∆ of the root system with
respect to T .
Theorem 11.1 Let G be connected reductive, B ≥ T a Borel subgroup of
G.

(a) There exists a base ∆ of Φ with positive system Φ+ ⊆ Φ such that

!
B=T· Uα
α∈Φ+

for any fixed order of the factors Uα ; moreover, we have uniqueness of

expression with respect to the product in the fixed order.
(b) If G is semisimple then |∆| = dim(X ⊗Z R) = dim(T ) = rk G.
84 BN-pairs and Bruhat decomposition

(c) W = sα | α ∈ ∆.

(d) G = T, Uα | α ∈ ±∆.

Proof of (b)–(d) Parts (b) and (c) are just Proposition 9.4(c) together with
Proposition 9.2. For (d) note that Uα , U−α  contains a preimage of the sim-
ple reflection sα , for α ∈ ∆, so H := Uα | α ∈ ±∆ contains preimages of all
w ∈ W by (c). Hence Uβ ≤ H for all β ∈ Φ by Proposition 9.4(c) and The-
orem 8.17(e). Then (d) follows from Theorem 8.17(g). See [66, Prop. 8.2.1]
for a proof of (a).

Definition 11.2 In the situation of Theorem 11.1, ∆ is called the set of

simple roots with respect to T ≤ B, and {sα | α ∈ ∆} are called the simple
reflections.

Remark 11.3 Recall from Theorem 8.17(f) and Proposition 8.15 that the
subgroup Uα , U−α  is non-solvable, so it follows from Theorem 11.1(a) that
Uα ⊆ B for α ∈ Φ− .

Example 11.4 We verify Theorem 11.1 in the case of G = SLn . Here,

T = Dn ∩ SLn is a maximal torus contained in the Borel subgroup B =
Tn ∩ SLn . By Example 9.8 the root system of G with respect to T is given
by Φ = {χij | i = j} with χij (diag(t1 , . . . , tn )) = ti t−1
j .
Thus, for i = j we have Uχij = Uij where

Uij := In + Eij .

Now, clearly B can be decomposed as

⎧⎛ ⎞ ⎫
⎨ ∗ ... ∗
⎪ ⎪
⎬ !
⎜ . . ⎟
B= ⎝ . . .
. ⎠ ∈ SLn = T · Uij .
⎪
⎩ ⎪
⎭
0 ∗ i<j

Moreover, with respect to the base ∆ = {χi,i+1 | 1 ≤ i ≤ n − 1} the set of

positive roots equals Φ+ = {χij | i < j}, by Example 9.8. So indeed
!
B=T · Uα
α∈Φ+

as claimed in Theorem 11.1, and |∆| = n − 1 = dim(T ) = rk(G). Finally,

the algorithm of Gaussian elimination shows that SLn is generated by the
Uα together with diagonal matrices.

We investigate the structure of B and Ru (B) a bit further; for this, let’s
choose and fix isomorphisms uα : Ga → Uα , for all α ∈ Φ.
 11.1 On the structure of B 85

Proposition 11.5 Let G be reductive, and let U ≤ Ru (B) be a T -stable

subgroup. Then U is the product of the root subgroups Uα it contains, hence
closed and connected. In particular, the Uα , α ∈ Φ+ , are the minimal non-
trivial T -invariant subgroups of Ru (B).
Proof Let U ≤ Ru (B) be T -stable. Since U is unipotent, U  := [U, U ] is a
T -stable subgroup of smaller dimension, hence a product of root subgroups
by induction. Thus we may argue in the abelian group U/U  . Clearly, if
uα (c) ∈ U with c = 0, then tuα (c)t−1 = uα (α(t)c) ∈ U for all t ∈ T , so
Uα ≤ U since α = 0.
Assume the result fails. Then by Theorem 11.1(a) there is 1 = u ∈ U which

is a product u = α∈M uα (cα ), for some M ⊆ Φ+ , such that Uα ⊆ U for
all α ∈ M . Take u such that |M | is minimal, and let β, γ ∈ M be different.
Since β, γ are linearly independent, there is t ∈ T with β(t) = 1, γ(t) = 1.
Then modulo U  we have that
! ! !
tut−1 u−1 ≡ uα (α(t)cα ) uα (−cα ) ≡ uα (cα (α(t) − 1))
α∈M α∈M α∈M

does not involve uβ any more, but a non-trivial element from Uγ occurs in
the expression. So it is a non-trivial product of smaller length, contradicting
our minimal choice.
The following consequence will be used in the investigation of maximal
rank subgroups:
Corollary 11.6 Let G be connected reductive with maximal torus T and
H ≤ G a connected reductive subgroup normalized by T . Then
H = T ∩ H, Uα | Uα ≤ H.
Proof Note that HT is a reductive subgroup of G. Let BH be a Borel
subgroup of HT containing T . It lies in a Borel subgroup B of G, hence
so does U := Ru (BH ). Application of Proposition 11.5 shows that U =

α∈M Uα for some subset M ⊆ Φ. But then HT = T, Uα | α ∈ ±M  by
Theorems 8.17(g) and 11.1. As HT is reductive, we have U ≤ [HT, HT ] ≤ H
and the claim follows.
Example 11.7 (Root system and root subgroups of SO2n ) Let G = SO2n
with Borel subgroup B = G ∩ T2n and maximal torus T = G ∩ D2n (see
Example 6.7). Here,
Uij := I2n + Eij − E2n−j+1,2n−i+1 , 1 ≤ i < j ≤ 2n − i,
are one-dimensional connected subgroups of the unipotent radical Ru (B) =
86 BN-pairs and Bruhat decomposition

G ∩ U2n normalized by T . By the description of B in Example 6.7(4) we see

that Ru (B) is the product of the Uij . Thus, the Uij are the positive root
subgroups with respect to B and T , by Proposition 11.5.
A short calculation shows that T acts via the character i − j on Uij ,
where
−1
 i : T → Gm , diag(t1 , . . . , tn , t−1
n , . . . , t1 ) → ti .

Now i = −2n−i+1 on T , so we see that Φ+ = {i ± j | 1 ≤ i < j ≤ n},

with base ∆ := {1 − 2 , 2 − 3 , . . . , n−1 − n , n−1 + n }. Computation of
|(Zα + Zβ) ∩ Φ+ | for α, β ∈ ∆ now shows that the root system Φ of SO2n
has type Dn .

The following result essentially describes multiplication in the unipotent

radical of B (see [32, Lemma 32.5], and [73, Thm. 8]).

Theorem 11.8 (Commutator formula) Given a root system Φ, with a fixed

total ordering compatible with addition, there exist integers cmn
αβ such that for
any connected reductive group G with root system Φ over an algebraically
closed field k, the morphisms uα can be chosen so that for all roots α = ±β
we have
!
[uα (t), uβ (u)] = umα+nβ (cmn m n
αβ t u ) for all t, u ∈ k,
m,n>0

where the product is over all integers m, n > 0 such that mα + nβ ∈ Φ, taken
according to the chosen ordering.

The cmn
αβ above are called structure constants. Note that all unipotent
elements of a connected reductive group G are contained in its derived sub-
group, which is semisimple, and also that any semisimple group is a quotient
of a group of simply connected type by a subgroup consisting of semisimple
elements. So it suffices to prove the above statement for semisimple groups
of simply connected type. The above, together with more precise versions of
Theorem 8.17(c) and (e), yields the very important Steinberg presentation
of a semisimple group G, see [73, §6] or [13, Thm. 12.1.1].
From now on, we fix a total ordering on Φ and, for each α ∈ Φ, an
isomorphism uα : Ga → Uα , c → uα (c), for which Theorem 11.8 holds.

Example 11.9 We determine the commutator relations for SL3 and Sp4 .

(1) For the group G = SL3 , with B and T as in Example 9.8, by Exam-
ple 11.4 the root groups of G with respect to T are

Uα = I3 + E12 , Uβ = I3 + E23 , Uα+β = I3 + E13 ,

 11.1 On the structure of B 87

for the roots α = χ12 and β = χ23 . A direct computation gives

[uα (t), uβ (u)] = uα+β (tu)
and all other commutators between root subgroups corresponding to
positive roots are trivial.
(2) For G = Sp4 , we have by Exercise 10.19 that B = G ∩ T4 is a Borel
subgroup with maximal torus T = G ∩ D4 . We write (t1 , t2 ) to denote
the element diag(t1 , t2 , t−1 −1
2 , t1 ) ∈ T . A similar calculation to the one in
Example 11.7 shows that the root subgroups of G in B are given by
Uα = I4 + E12 − E34 , Uβ = I4 + E23 ,
Uα+β = I4 + E13 + E24 , U2α+β = I4 + E14 ,
for the roots α, β ∈ X(T ) with α(t1 , t2 ) = t1 t−1 2
2 , β(t1 , t2 ) = t2 . In partic-
ular the root system of Sp4 is of type C2 (compare with Exercise 10.29).
A direct computation gives
[uα (t), uβ (u)] = uα+β (−tu)u2α+β (−t2 u),
[uα (t), uα+β (u)] = u2α+β (−2tu),
and all other commutators between root subgroups corresponding to
positive roots are trivial. It becomes apparent from these relations that
the structure of the unipotent radical of B in characteristic 2 is consid-
erably different from the case of odd (or zero) characteristic. See also
Example 13.16 where the effect of this on maximal rank subgroups will
emerge.
We can now describe the automorphism group of a semisimple group G.
Let’s fix a pair T ≤ B consisting of a maximal torus of G contained in a
Borel subgroup, with corresponding root system Φ, set of positive roots Φ+
and base ∆. We first prove a general result about endomorphisms which will
also be used in Part III in the investigation of Steinberg endomorphisms.
Lemma 11.10 Let G be connected reductive with T , B, Φ+ as above. Let
σ : G → G be an endomorphism stabilizing T and B. Then there exists a
permutation ρ of Φ+ stabilizing ∆ such that for all α ∈ Φ+ the following
conditions hold:
(a) There exists a positive integer qα , equal to 1 or to a power of char(k) > 0
such that σ(ρ(α)) := ρ(α) ◦ σ|T = qα α.
(b) There exists aα ∈ k × such that σ(uα (c)) = uρ(α) (aα cqα ) for all c ∈ k.
If moreover σ is an automorphism of algebraic groups, then qα = 1 for all
α ∈ Φ+ .
88 BN-pairs and Bruhat decomposition

Proof As σ stabilizes both B and T , it permutes the minimal T -stable

subgroups of the unipotent radical Ru (B) of B, viz., the root subgroups
Uα , α ∈ Φ+ , see Proposition 11.5. This induces a permutation ρ of Φ+ , via
σ(Uα ) = Uρ(α) , which must preserve ∆ since any positive system contains a
unique base (see Proposition A.7). In particular for any α ∈ Φ+ there exists
an endomorphism ν : Ga → Ga such that σ(uα (c)) = uρ(α) (ν(c)) for c ∈ k.
For t ∈ T we have tuα (c)t−1 = uα (α(t)c) by Theorem 8.17(c). Application
of σ yields
% &
σ(tuα (c)t−1 ) = σ(t)σ(uα (c))σ(t)−1 = uρ(α) ρ(α)(σ(t)) ν(c)
on the left-hand side, and
% &
σ(uα (α(t)c)) = uρ(α) ν(α(t)c)
on the right, whence
ν(α(t)c) = ρ(α)(σ(t)) ν(c) for all c ∈ k, t ∈ T.
Thus, ν is a monomial, of degree qα say. Clearly, this defines a group mor-
phism only if qα is a power of char(k), showing (b). If σ is an isomorphism,
then so is ν, but then qα = 1 by Exercise 10.10.
Note that as T is stabilized by σ, by composition σ also defines a linear
map X(T ) → X(T ), χ → χ ◦ σ. Putting c = 1 above we get
σ(ρ(α))(t) = ρ(α)(σ(t)) = ν(α(t))/ν(1) = α(t)qα
for all t ∈ T , which shows (a).
Let’s now write Auta (G) for the group of (algebraic group) automorphisms
of G, Inn(G) for the subgroup of inner automorphisms, and let ΓG denote
the subgroup of Auta (G) of automorphisms of G which fix a pair T ≤ B
consisting of a maximal torus of G contained in a Borel subgroup.
Theorem 11.11 Let G be semisimple. Then:

(a) Auta (G) = Inn(G).ΓG .

(b) The elements of ΓG induce diagram automorphisms of the Dynkin dia-
gram of G, which determine them uniquely modulo Inn(G) ∩ ΓG .
In particular, Outa (G) := Auta (G)/Inn(G) is finite.
Proof Let σ ∈ Auta (G). Since G acts transitively on its Borel subgroups by
conjugation, and B acts transitively on its maximal tori, there is g ∈ G such
that σ composed with conjugation by g fixes both T and B. This proves the
first part.
 11.1 On the structure of B 89

For part (b), let σ ∈ ΓG . Then by Lemma 11.10 it induces a permutation ρ

of the positive system Φ+ such that σ(ρ(α)) = α, that is, the homomorphism
of X(T ) induced by σ permutes the set of positive roots. Exercise C.1 then
shows that σ must induce a diagram automorphism of the Dynkin diagram
with respect to the base ∆ in Φ+ .
For the uniqueness assertion it suffices to show that any σ ∈ ΓG which
induces the identity on ∆ is contained in Inn(G). Any such σ fixes all root
subgroups Uα = uα (Ga ) for α ∈ ∆, thus by Lemma 11.10 there exist aα ∈ k ×
such that σ(uα (c)) = uα (aα c) for c ∈ k. As ∆ is linearly independent, there
exists s ∈ T such that α(s) = aα , for all α ∈ ∆. Replacing σ by s−1 σ we may
assume by Theorem 8.17(c) that σ acts as the identity on all Uα , α ∈ ∆.
Furthermore, as σ acts trivially on ∆, for t ∈ T we have α(σ(t)) = α(t)
$
for α ∈ ∆, so α(σ(t)t−1 ) = 1. But G is semisimple, so α∈∆ ker(α) is finite.
Hence the morphism T → T , t → σ(t)t−1 , has finite, connected image,
forcing σ(t) = t for all t ∈ T . Thus σ centralizes Tα , hence stabilizes Cα =
CG (Tα ) for all α ∈ ∆. As it centralizes the Borel subgroup T Uα of the latter,
it must be trivial on Cα by Proposition 6.8. Since G is generated by the Cα
(see Proposition 8.15), the claim follows.
The non-trivial groups of diagram automorphisms of connected Dynkin
diagrams are collected in Table 11.1. It then follows from the previous result
that the outer (algebraic) automorphism group of a simple algebraic group
is solvable, of order at most 6.

Table 11.1 Graph automorphisms of Dynkin diagrams

Φ An (n ≥ 2) Dn (n ≥ 5) D4 E6
ΓG Z2 Z2 S3 Z2

Theorem 11.11 is complemented by the following existence result for au-

tomorphisms (see [15, 23.7, Cor. 3] or [73, Cor. on p.156]):
Theorem 11.12 (Chevalley) Let G be semisimple of simply connected or
adjoint type, ρ a permutation of the set of simple roots ∆ of G inducing a
symmetry of the Dynkin diagram. Then there exists an automorphism σ ∈
Auta (G) with
σ(uα (c)) = uρ(α) (c) for all α ∈ ∆, c ∈ k.
In particular, Outa (G) is isomorphic to the group of graph automorphisms
of the Dynkin diagram of G.
90 BN-pairs and Bruhat decomposition

Proof for type An−1 For An−1 , n ≥ 3, there is a unique non-trivial graph
automorphism of the Dynkin diagram, which sends αi to αn−i , for 1 ≤ i ≤
n−1 (with the simple roots labeled as in Table 9.1). An easy calculation with
the root subgroups in Example 11.4 shows that
 theproduct of the transpose-
0 . .1
inverse automorphism with conjugation by . · diag(1, −1, 1, −1, . . .) is
1 0
an automorphism of SLn with the required property. Factoring out the center
we obtain a corresponding automorphism on the adjoint group PGLn .
The automorphisms in Theorem 11.12 are called graph automorphisms
of G. For SO2n , of type Dn , elements of GO2n \ SO2n induce a non-trivial
graph automorphism of order 2 (see Exercise 20.1). The exceptional graph
automorphisms for D4 , of order 3, and for E6 , of order 2, are realized inside
suitable larger groups, see Examples 12.12 and 13.9.
Remark 11.13 There exist further abstract group automorphisms of a
simple algebraic group G, but which are not invertible as morphisms. For
example, any field automorphism of the underlying field k extends to an
automorphism of G. Also, if G is of type B2 or F4 in characteristic 2, or of
type G2 in characteristic 3, and k is perfect, there exists a bijective endo-
morphism which comes from the non-trivial symmetry of the corresponding
Coxeter diagram (see [73, Thm. 29 and Cor.]). These will be important in
Part III.

11.2 Bruhat decomposition

We now discuss the partition of a connected reductive group G into double
cosets of a Borel subgroup. Throughout, we fix a maximal torus T lying in a
Borel subgroup B, with root system Φ, positive roots Φ+ and base ∆ deter-

mined by B. We write Ru (B) = α∈Φ+ Uα according to Theorem 11.1. For
any w ∈ W = NG (T )/T let ẇ denote an arbitrary fixed preimage in NG (T ).
We first need to study the multiplication of double cosets with respect to B.
Lemma 11.14 Let α ∈ ∆ be a simple root with corresponding simple
reflection s ∈ W . Then for all w ∈ W we have
B ẇB · B ṡB ⊆ B wsB
˙ ∪ B ẇB.
Proof It can easily be checked inside the homomorphic image Uα , U−α  of
SL2 that ṡ(Uα \ {1})ṡ ⊆ T Uα ṡUα . So clearly
ṡUα ṡ ⊆ T Uα ṡUα ∪ Uα ⊆ B ṡB ∪ B.
 11.2 Bruhat decomposition 91

Moreover, by Theorem 11.1 we have B = ( β∈Φ+ \{α} Uβ )·Uα ·T . Recall that
v̇Uβ v̇ −1 = Uvβ for all v ∈ W and β ∈ Φ by Theorem 8.17(e), and sα .β ∈ Φ+
for α = β ∈ Φ+ by Lemma A.8. Now first assume that wα ∈ Φ+ . Then
B ẇB · B ṡB = B ẇUα ṡB = BUwα wsB
˙ = B wsB.
˙
On the other hand, if wα ∈ −Φ+ then wsα = w(−α) = −wα ∈ Φ+ . So,
with v := ws we have B v̇B · B ṡB = B ẇB by the above. Then
B ẇB · B ṡB = B vsB
˙ ṡB ⊆ B v̇(B ṡB ∪ B) = B v̇B ṡB ∪ B v̇B = B ẇB ∪ B wsB
˙
as claimed, where the inclusion follows from the first part of the proof.
We now introduce a crucial combinatorial group theoretical structure
which is particularly well adapted to reductive algebraic groups.
Definition 11.15 A pair B, N of subgroups of a group G is called a BN -
pair (for G) if the following axioms are satisfied:
(BN1) G is generated by B and N .
(BN2) B ∩ N is a normal subgroup of N .
(BN3) The group W := N/(B ∩ N ) is generated by a set S of involutions.
(BN4) If ṡ ∈ N maps to s ∈ S under the natural homomorphism N → W ,
and n ∈ N , then BnB · B ṡB ⊆ BnṡB ∪ BnB.
(BN5) If ṡ is as before then ṡB ṡ = B.
The group W is called the Weyl group of the BN-pair.
Theorem 11.16 (Tits) Let G be a connected reductive algebraic group with
Borel subgroup B and N := NG (T ) for some maximal torus T ≤ B. Then
B, N is a BN -pair in G whose Weyl group is equal to that of G.
Proof By Theorem 4.4 and Corollary 8.13(b), B ∩ N = NB (T ) = CB (T ) =
T and so B∩N is normal in N , giving (BN2). Moreover W = N/(B∩N ) is the
Weyl group of G, hence generated by the set S of simple reflections by The-
orem 11.1(c), whence we have (BN3). By Theorem 8.17(g), G is generated
by T and the Uα for α ∈ Φ. But for α ∈ Φ+ , s˙α Uα s˙α −1 = U−α ≤ B, N  by
Theorem 8.17(e), showing (BN1). We just proved (BN4) in Lemma 11.14.
Finally, writing α for the root of a simple reflection s ∈ S, ṡB ṡ contains
ṡUα ṡ = U−α , which does not lie in B by Remark 11.3. This shows (BN5).
Let’s put Φ− := −Φ+ whenever Φ+ is a positive system inside Φ, and
!
Uw− := Uα
α∈Φ+ ,w.α∈Φ−

for any w ∈ W (note that by Theorem 11.8 this is a subgroup of G). With
92 BN-pairs and Bruhat decomposition

this we have the following extension of Theorem 8.17(g) (see [32, Thm. 28.3
and 28.4]):
Theorem 11.17 (Bruhat decomposition) Let G be a group with a BN-pair.
Then
)
G= B ẇB
w∈W

for any choice of preimages ẇ ∈ N mapping to w ∈ W = B/N .

If G is connected reductive, T ≤ B a maximal torus in a Borel subgroup
of G, W the Weyl group of G with respect to T , then more precisely, every
g ∈ G can be uniquely written as g = uẇb, where b ∈ B, w ∈ W and u ∈ Uw−
with respect to the positive system Φ+ ⊆ Φ determined by T ≤ B.
Proof of the double coset decomposition We first show that B v̇B ∩ B ẇB =
∅ if v = w. Write v as a product of minimal length (v) in the generators
from S. We argue by induction on (v), which we may assume to be not
larger than (w). If (v) = 0, then v = 1. Now B = B ẇB implies ẇ ∈ B ∩ N ,
so w = 1. If (v) > 0, we may write v = v  s with s ∈ S and (v  ) < (v).
Then B v̇  ṡ = B v̇ ⊆ B v̇B = B ẇB. By (BN4) this gives
B v̇  ⊆ B ẇB ṡ ⊆ B wsB
˙ ∪ B ẇB.
So B v̇  B = B wsB
˙ or B v̇  B = B ẇB. As (v  ) < (v) induction yields that
either v = ws or v = w. The second is not possible since (v  ) < (v) ≤ (w)
 

by assumption. So v = v  s = w.
On the other hand, an easy induction on (w) shows from (BN4) that the
union of the B ẇB, where w runs over W , is closed under multiplication.
Since it contains B and N , it is all of G by (BN1).
The proof of the uniqueness assertion uses in an essential way that G is
reductive.
Corollary 11.18 In the notation of Theorem 11.17, there exists a unique
element w0 ∈ W such that w0 (∆) = −∆. Moreover w02 = 1 and B ẇ0 ∩B = T .
Proof The first part follows directly from the simple transitivity of W on
the set of bases, see Proposition 9.4(b). Now assume that g ∈ B ẇ0 ∩ B, so
g = ẇ0 utẇ0 = t u for some u, u ∈ U = Ru (B), t, t ∈ T . Thus
ẇ0−1 t u = uẇ0 tẇ0 .
As Uw−0 = U by the defining property of w0 , the uniqueness statement in
Theorem 11.17 shows that u = 1 = u, whence g = t ∈ T . The other
inclusion is clear since ẇ0 normalizes T .
 11.2 Bruhat decomposition 93

The element w0 above is called the longest element of W with respect to S

since its length (w0 ) as introduced in the proof of the Bruhat decomposition
can be shown to be maximal possible (see also Proposition A.21 for another
characterization of the length function on W ). The subgroup B ẇ0 is called
the Borel subgroup opposite to B.
Corollary 11.19 Let G be reductive. Then the intersection of any two
Borel subgroups of G contains a maximal torus.
Proof Since all Borel subgroups and maximal tori of G lie in G◦ , we may
assume that G is connected. Let B, B1 be Borel subgroups, so B1 = gBg −1
for some g ∈ G (by Theorem 6.4). Choose a maximal torus T ≤ B, with
Weyl group W . Then g = bẇb with b, b ∈ B, w ∈ W , as in Theorem 11.17;
hence
B1 = gBg −1 = bẇb B(b )−1 ẇ−1 b−1 = bẇB ẇ−1 b−1 .
Now ẇT ẇ−1 = T , so bẇT ẇ−1 b−1 = bT b−1 ≤ B is a maximal torus in
B ∩ B1 .
We end this chapter with the following topological aspect of the Bruhat
decomposition which will be needed for the investigation and construction
of representations in Section 15.2.
Theorem 11.20 Let G be connected reductive, T ≤ B a maximal torus
inside a Borel subgroup, with corresponding positive system Φ+ , U − := Uα |
α ∈ Φ− . Then the product map π : U − × B → G is a bijective morphism of
U − × B onto a dense open subset of G (called the big cell).
Proof By definition of U − and of w0 we have U − = w˙0 U w˙0 , so the product
map U − × B = w˙0 U w˙0 × B → G is a bijective morphism onto the sub-
set w˙0 U w˙0 B by the uniqueness of expression in the Bruhat decomposition
(Theorem 11.17). Now
dim(U − × B) = dim(U − ) + dim(B) = |Φ− | + rk(G) + |Φ+ | = dim(G)
by Theorem 8.17(b). Moreover, U − × B is irreducible, so the closure of its
image is an irreducible subset of G of the same dimension, hence equal to G
by Proposition 1.22.
It remains to show that the image is open; we use the fact that it is a
translate of the coset B w˙0 B = U w˙0 B by w˙0 . As B w˙0 B is the B-orbit of
w˙0 B (in the action of G on G/B by left multiplication), Proposition 5.4(a)
implies that B w˙0 B is open in its closure. But we have just observed that the
closure is G/B. So B w˙0 B is open in G/B. This then implies that B w˙0 B is
open in G because the quotient map is continuous.
94 BN-pairs and Bruhat decomposition

It can be shown that the map π above is even an isomorphism onto its
image, see [32, 28.5].
 12
Structure of parabolic subgroups, I

We now come to a natural collection of subgroups of a reductive group,

the parabolic subgroups; in fact, one can define parabolic subgroups in any
group having a BN-pair. It will turn out that these are precisely the closed
subgroups containing a Borel subgroup of G. Moreover, their structure as
a semidirect product of a reductive group, the so-called Levi complement,
with a unipotent normal subgroup enables one to argue inductively for many
questions concerning the subgroup structure and representation theory of re-
ductive groups. We’ll see that Levi complements can alternatively be char-
acterized as the centralizers of subtori.

12.1 Parabolic subgroups

Let G be a connected reductive algebraic group, T ≤ G a maximal torus
contained in a Borel subgroup B of G. Let Φ be the root system of G, ∆
the set of simple roots with respect to T ≤ B, S = {sα | α ∈ ∆} the
corresponding set of generating reflections of the Weyl group W = NG (T )
(see Theorem 11.1(c)).
We define a collection of natural subgroups of W . For a subset I ⊆ S,
WI := s ∈ I is called a standard parabolic subgroup of W . A parabolic
subgroup of W is any conjugate of a standard parabolic subgroup. We let
∆I := {α ∈ ∆ | sα ∈ I} and
"
ΦI := Φ ∩ Zα
α∈∆I

be the corresponding parabolic subsystem of roots.

Proposition 12.1 Let I ⊆ S. Then ΦI is a root system in RΦI with base
∆I and Weyl group WI .
96 Structure of parabolic subgroups, I

This is proved in Proposition A.25.

The WI are, up to conjugation, precisely the subspace centralizers of W
in its natural reflection representation on XR := X ⊗Z R, see Corollary A.29.
The group G now contains similar natural subgroups:

Proposition 12.2 Let G be connected reductive, T ≤ B a maximal torus

in a Borel subgroup of G, with root system Φ and set of simple reflections S.

(a) Let I ⊆ S. Then PI := BWI B = w∈WI B ẇB is a closed, connected,

self-normalizing subgroup of G which contains B.
(b) The PI are mutually non-conjugate; in particular, PI = PJ implies I =
J.
(c) PI = T, Uα | α ∈ Φ+ ∪ ΦI .

Moreover, all overgroups of B in G arise in this way.

Proof (a) By Lemma 11.14 we have B ẇB ·B ṡB ⊆ PI for all w ∈ WI and all
s ∈ I. Hence, by an easy induction, PI ≤ G. Now assume that g ∈ NG (PI ).
Then B, B g are two Borel subgroups of PI , hence conjugate by some p ∈ PI
by Theorem 6.4(a). Then gp ∈ NG (B) = B, by Theorem 6.12, so g ∈ PI .
In (b), if PJ = PIg with g ∈ G then B, B g are Borel subgroups of PJ ,
whence g ∈ PJ as before, so PI = PJ . Next, if PI = PJ for I, J ⊆ S then
by (a) we must have WI = WJ . This implies that WI = WI∪J , so we may
assume that I ⊆ J. But the elements of S are reflections on XR , so the fixed
space of WI on XR has codimension at most |I|. On the other hand, by
Corollary 8.22 and the proof of Theorem 8.21, the fixed space of W = WS
on XR has codimension rkss (G), which equals |∆| by Theorem 11.1(b), so
we must in fact have equality. Thus WI = WJ implies that I = J.
For (c), write Φ± ±
I := ΦI ∩ Φ . Since B ⊆ PI we have Uα ⊆ PI for all
α ∈ Φ+ . Since ΦI is a root system by Proposition 12.1, the longest element
− −
w ∈ WI satisfies w(Φ+ I ) = ΦI (see Corollary 11.18). Thus, for β ∈ ΦI
−1
there exists α ∈ ΦI such that wα = β. Then Uβ = Uwα = ẇUα ẇ ⊆ PI ,
+

which proves the inclusion “⊇”. For the converse note that if α ∈ ∆ is the
simple root corresponding to s ∈ S, then we may choose ṡ ∈ U±α . So
T, Uα | α ∈ ΦI  contains preimages of all s ∈ I, hence of all w ∈ WI ,
which in view of the decomposition in (a) gives the other inclusion. The
connectedness of PI now follows by Proposition 1.16.
Finally, all overgroups of B are of the form PI for some I ⊆ S by Exer-
cise 20.3.

Note that all statements except for (c) and the connectedness of PI remain
valid for arbitrary BN-pairs, see Exercise 20.3.
 12.1 Parabolic subgroups 97

Definition 12.3 The PI (I ⊆ S) are called standard parabolic subgroups of

G. A parabolic subgroup of G is any subgroup containing a Borel subgroup.
Note that by Proposition 12.2 these are just the conjugates of standard
parabolic subgroups.

Example 12.4 Let G = SLn . Then T = Dn ∩ SLn is a maximal torus

contained in the Borel subgroup B = Tn ∩ SLn , with root system Φ = {χij |
1 ≤ i, j ≤ n, i = j}, set of positive roots Φ+ = {χij | i < j}, and base
∆ = {χi,i+1 | i < n} (see Example 9.8). Furthermore,
⎛ ⎞
Ii−1
⎜ 0 1 ⎟
si := ⎜
⎝
⎟ T ∈ NG (T )/T
⎠
−1 0
In−i−1

is the simple reflection corresponding to the simple root χi,i+1 . We iden-

tify W = NG (T )/T with Sn via si → (i, i + 1), so S = {si | i < n} =
{(12), (23), . . . , (n − 1, n)}. Let I = S \ {sa } for some a < n. The associated
maximal parabolic subgroup of W is WI = s1 , . . . , sa−1 , sa+1 , . . . , sn−1  ∼
=
Sa × Sn−a . And so the corresponding parabolic subgroup of G consists of
block diagonal matrices

 
A ∗ 
Pa := PI = A ∈ GLa , B ∈ GLn−a ∩ SLn ,
0 B 

the stabilizer in the natural representation of G of the subspace generated

by the first a standard basis vectors.

More generally, parabolic subgroups in GLn and in SLn are just the sta-
bilizers of flags of subspaces in their natural n-dimensional representation
space. In fact, the parabolic subgroups of all classical groups (groups with
root system of type An , Bn , Cn or Dn ) have an interpretation in terms of
the natural module for the group. We give a precise statement and proof in
the next section (see Proposition 12.13).
The BN-pair setting also allows one to give an easy proof that simple
algebraic groups (as defined in Theorem 8.21) are simple modulo center as
abstract groups; for this recall from the discussion before Theorem 9.6 that
the Weyl group of a simple algebraic group is irreducible.

Proposition 12.5 Let G be a simple linear algebraic group. Then G/Z(G)

is simple as an abstract group.

Proof Let H  G be a normal subgroup and B some Borel subgroup of G.

98 Structure of parabolic subgroups, I

Since H is normal in G, the product BH is a subgroup of G containing B.

Then
)
BH = BWI B = B ẇB for some I ⊆ S
w∈WI

by Proposition 12.2. We claim that also WI is normal in W . For this, let

s ∈ S \ I. For any w ∈ WI , B ẇB ⊆ BH, so there exists b ∈ B with ẇb ∈ H
by the normality of H. Thus
˙ ṡ ∈ (B swsB
swb ˙ ∪ B swB)
˙ ∩ BH
by Lemma 11.14. By the choice of s, clearly sw ∈ / WI , so sws ∈ WI , for all
s ∈ S \ I, w ∈ WI . Since S generates W by Theorem 11.1(c) this shows that
WI  W as claimed.
Now first assume that I = S. Then WI centralizes the non-zero subspace
$
s∈I ker(s − id) of XR , while W acts irreducibly on XR . But a normal sub-
group of an irreducible group which fixes a non-zero vector is trivial, whence
I = ∅ and WI = 1. Then BH = B, so H ≤ B. As H is normal in G, it also
lies in all conjugates of B. Since
*# +◦
B g = R(G) = 1
g∈G

by Proposition 6.16, H is finite, and hence H ≤ Z(G) by Exercise 10.4.

Otherwise, if I = S then BH = G, so B/(B ∩ H) ∼ = BH/H = G/H. Now,
since B is solvable, B/(B ∩ H) is solvable while G/H is perfect since G is.
Thus G/H = 1 and hence H = G.

12.2 Levi decomposition

We continue the investigation of the structure of parabolic subgroups of a
connected reductive group G. For I ⊆ S define
!
UI := Uα = Uα | α ∈ Φ+ \ ΦI 
α∈Φ+ \ΦI

with ΦI as above. In this situation,

LI := T, Uα | α ∈ ΦI  ≤ PI
is a complement to UI in the parabolic subgroup PI :
Proposition 12.6 Let I ⊆ S. Then Ru (PI ) = UI , and LI is a complement
to UI , so PI = UI
LI . In particular, LI is reductive with root system ΦI .
 12.2 Levi decomposition 99

Furthermore, all closed complements to UI are conjugate to LI in PI and

LI = CG (Z(LI )◦ ).

Observe the special case when I = ∅, so PI = B, UI = Ru (B) and LI = T

where this is just Theorem 4.4.

Proof We see that UI  PI and PI = UI , LI  by the commutator relations

(Theorem 11.8) and the generation property of PI in Proposition 12.2(c). To
$
show that LI ∩ UI = 1 let Z := ( α∈ΦI ker α)◦ ≤ T and L := CG (Z). Note
that LI ≤ L by Theorem 8.17(c). We claim that L = LI .
By Corollary 8.13(a), L is connected reductive, so by Corollary 11.6 it is
generated by T and the Uβ it contains. Now assume that Uβ ≤ L. Then
the restriction of β to Z is trivial by Theorem 8.17(c), so β ∈ Z ⊥ (see Def-
$
inition 3.7). The identity component Z is of finite index in α∈ΦI ker α =
ΦI ⊥ , so some multiple of β lies in ΦI ⊥⊥ (see Exercise 10.15). Further-
more, by Proposition 3.8(b), ΦI ⊥⊥ /ΦI  is a finite group, so some non-zero
multiple of β even lies in ΦI . By (R2) this shows that β ∈ ΦI , so L = LI .
But then UI ∩ LI is normal in the reductive group LI = L and unipo-
tent, whence trivial, so PI = UI
L. As LI is reductive, we moreover have
Z(LI )◦ = Z, so LI = CG (Z(LI )◦ ), and Z is a maximal torus of R(PI ).
In general, if L ≤ PI is any closed complement to UI , then by the first
part L ∼= LI as abstract groups, hence also Z  := Z(L )◦ ∼= Z(LI )◦ = Z by
 ∼
Theorem 8.17(h). Since Ru (PI ) = UI we have Ru (L ) = Ru (PI /UI ) = 1 and
so Ru (Z  ) = 1. Thus, Z  is a torus by Example 6.17(2). Now tori of different
dimensions cannot be isomorphic as abstract groups (count elements of fixed
finite order), so Z  is a torus of the same dimension as Z, hence also a
maximal torus of R(PI ). But all maximal tori of R(PI ) are conjugate, by
Theorem 4.4(b), so L = CPI (Z  ) is conjugate to LI = CPI (Z).

Definition 12.7 The decomposition PI = UI
LI in the preceding propo-

sition is called the Levi decomposition of the parabolic subgroup PI , and LI
is called the (standard) Levi complement of PI . The conjugates of standard
Levi complements are called Levi subgroups of G.

Corollary 12.8 In the setting of Proposition 12.6, PI = NG (UI ) for all

I ⊆ S.

Proof Clearly, PI ≤ N := NG (UI ) = NG (Ru (PI )), so N = PJ is parabolic

for some subset J ⊇ I of S by Proposition 12.2. Now for s = sα ∈ J \ I we
have s ∈ N and Uα ≤ UI , so ṡUα ṡ−1 = U−α ≤ UI , which is not the case, so
J = I, N = PI .
100 Structure of parabolic subgroups, I

In contrast to the situation for parabolic subgroups in Proposition 12.2(b),

Levi subgroups LI for different subsets I ⊆ S may be conjugate in G:

Example 12.9 Let G = SLn and Pa the parabolic subgroup of G defined

in Example 12.4, where 1 ≤ a ≤ n − 1. It has Levi complement

 
A 0 
La = A ∈ GLa , B ∈ GLn−a ∩ SLn ,
0 B 
and clearly La is conjugate to Ln−a in G.
More generally, for a flag 0 = V0 ⊂ V1 ⊂ · · · ⊂ Vr = k n with dim Vi /Vi−1 =
ni , so n = n1 + · · · + nr , the stabilizer in GLn is a parabolic subgroup (see
Proposition 12.13) with Levi complement GLn1 × · · · × GLnr consisting of
block diagonal matrices, where GLni ∼ = GL(Vi /Vi−1 ).
The preceding proposition shows in particular that Levi subgroups L are
centralizers of their central torus Z(L)◦ . In fact, the converse is also true:

Proposition 12.10 Let G be connected reductive, Z ≤ G a torus. Then

CG (Z) is a Levi subgroup of G.

Proof Let C := CG (Z), a connected reductive group by Proposition 8.13.

A Borel subgroup BC of C lies in some Borel subgroup B of G. As Z ≤
Z(C)◦ ≤ BC ≤ B, Z is contained in a maximal torus T of B, hence of
G. Clearly, T ≤ C. So by Theorem 8.17(g), C = T, Vα | α ∈ ΦC , where
ΦC denotes the root system of C with respect to T , and Vα are the T -root
subgroups of C. In particular, for α ∈ ΦC we have Lie(Vα ) ⊆ gα = 0, so
ΦC ⊆ Φ. Moreover, Vα is the unique one-dimensional subgroup of C with
Lie(Vα ) = (Lie(C))α = gα . Hence by Theorem 8.17(d), Vα = Uα , the T -root
subgroup of G corresponding to α.
Now clearly Uα ≤ C if and only if Z ≤ ker α, whence C = T, Uα | Z ≤
ker(α). So ΦC = Φ ∩ α | Z ≤ ker(α), the intersection of Φ with a subspace
of RΦ. By an elementary property of root systems, see Corollary A.29, ΦC
is a parabolic subsystem of Φ, and thus C is the Levi complement of the
corresponding parabolic subgroup of G.

For later use let’s also point out the following (see Exercise 20.4):

Corollary 12.11 Let I ⊆ S, WI = I, LI the corresponding standard Levi

subgroup. Then there is a natural isomorphism NG (LI )/LI ∼
= NW (WI )/WI .
Example 12.12 From the Dynkin diagram one sees that a simple group G
of type E6 has a Levi subgroup L with derived subgroup L = [L, L] of type
D4 . By Corollary 12.11 the normalizer of L in G can be computed inside
 12.2 Levi decomposition 101

the Weyl group; it is an extension of L by the symmetric group S3 , which

necessarily leaves L invariant. An element σ therein of order 3 induces the
triality automorphism σ of the root system D4 of order 3, hence a corre-
sponding triality graph automorphism of L whose existence was asserted in
Theorem 11.12:

See also Example 13.9 for another realization of triality.

Similarly, the non-trivial graph automorphism of E6 may be realized by
embedding this group as a Levi subgroup inside E7 with non-trivial normal-
izer.

We now consider the case of parabolic subgroups in classical groups and

show that the situation of Examples 12.4 and 12.9 generalizes. Let V be
a finite-dimensional vector space over k equipped with the 0 form β, a
symplectic (i.e., non-degenerate skew-symmetric) bilinear form β, or a non-
degenerate quadratic form Q (with associated bilinear form βQ ). In the
case where V of odd dimension is equipped with a quadratic form and
char(k) = 2, we say the quadratic form is non-degenerate if the radical
of the form βQ is a 1-space w with Q(w) = 0. Recall that v ∈ V is called
isotropic if β(v, v) = 0, and in the orthogonal case it’s said to be singular if
Q(v) = 0. A subspace W of V is said to be totally isotropic if β(v, w) = 0 (or
βQ (v, w) = 0) for all v, w ∈ W , and in the orthogonal case it is said to be to-
tally singular if Q(v) = 0 for all v ∈ W . Note that (totally) singular implies
(totally) isotropic and that the two notions are equivalent if char(k) = 2.
When the space is equipped only with a bilinear form, we may use the term
“singular” in place of “isotropic”.
Let Isom(V ) denote the full group of isometries of V and Cl(V ) ≤ Isom(V )
be the group SL(V ), Sp(V ) or SO(V ) respectively, as introduced in Sec-
tion 1.2.

Proposition 12.13 The parabolic subgroups of a simple classical group

Cl(V ) are precisely the subgroups of the form

P = StabCl(V ) (0 ⊂ V1 ⊂ . . . ⊂ Vr ⊂ V ),

where Vi is a totally isotropic (totally singular if V is equipped with a quadra-

tic form) subspace of V for 1 ≤ i ≤ r.
102 Structure of parabolic subgroups, I

Proof To simplify the exposition, throughout this proof we will use the
terms “singular” and “totally singular” for isotropic vectors and subspaces
in the case where the space is equipped with a symplectic form or the zero
form.
Given a flag F of totally singular subspaces, refine this to obtain a maximal
flag inside a maximal totally singular subspace of V . By Witt’s lemma (see [2,
§20]), all maximal singular subspaces, and moreover all maximal flags inside
a given maximal singular subspace are Isom(V )-conjugate. By the explicit
constructions in Example 6.7, there exists a Borel subgroup of G stabilizing
a maximal flag of a maximal singular subspace of V . So the stabilizer of F
contains a Borel subgroup and hence is parabolic by Proposition 12.2.
In addition to the above remarks, we see that the explicit constructions of
Borel subgroups in Example 6.7 also show that the Borel subgroups of a sim-
ple classical group Cl(V ) act indecomposably on V . Indeed, one first checks
that except when V is an odd-dimensional orthogonal space and char(k) = 2,
the unipotent radical of the given Borel subgroup B has a one-dimensional
fixed point space on V ; as Cl(V ) acts irreducibly this also follows easily from
the theory of highest weights, see Corollary 15.10. In the exceptional case,
one quotients out by the radical of the bilinear form and passes to the sym-
plectic group, where B acts indecomposably. Then the result follows from
the fact that the radical of the form has no B-invariant complement in V .
Now let P ≤ Cl(V ) be a parabolic subgroup with Levi decomposition
P = QL, containing a fixed Borel subgroup B. We first show that if P
stabilizes a flag of subspaces on V with Q acting as the identity on the
quotient spaces, then P is the full stabilizer in G of this flag. Let F be the
flag 0 = V0 ⊂ V1 ⊂ . . . ⊂ Vr+1 = V with P ≤ StabG (F) and with Q
acting as the identity on each of the quotients Vi /Vi−1 . By Proposition 12.2,
StabG (F) is a parabolic subgroup, containing B, say P̂ . By Proposition 12.6,
Ru (P̂ ) ≤ Q. Now let U ≤ P̂ be the subgroup of elements which act as the
identity on all quotients in F. Then U is a closed unipotent, normal subgroup
of P̂ , and so U ◦ ≤ Ru (P̂ ). But Q ≤ U and so Q ≤ U ◦ ≤ Ru (P̂ ) ≤ Q, whence
Q = Ru (P̂ ). By Corollary 12.8, this implies P = P̂ .
To complete the proof, we must find a flag of totally singular subspaces
of V stabilized by P such that Q acts as the identity on the quotients. By
Proposition 2.9, Q has a non-zero fixed point on any representation space,
so P stabilizes the flag 0 = V0 ⊂ V1 ⊂ . . . ⊂ Vr+1 = V , where Vi /Vi−1 :=
(V /Vi−1 )Q , for i ≥ 1. If V is equipped with the trivial form (so G = SL(V )),
this flag satisfies the criteria.
Now suppose V is equipped with a non-degenerate form Q, and assume
for now that char(k) = 2 if V is equipped with a quadratic form. Set W :=
 12.2 Levi decomposition 103

V Q . Note that W ∩ W ⊥ = 0, else W is a non-degenerate subspace and

P ⊆ Isom(W ) × Isom(W ⊥ ), contradicting the indecomposability of B on V .
Hence X := W ∩ W ⊥ = 0 and P stabilizes X, a totally singular subspace
of V . If X ⊥ = X, then X is a maximal totally singular subspace of V and
P ⊆ Stab(0 ⊂ X ⊂ V ). As the dual space X ∗ is isomorphic as Q-module to
V /X ⊥ = V /X, Q also acts trivially on V /X and we have the desired result.
If X ⊥ = X, P acts on the non-trivial quotient space X ⊥ /X on which Q has
a non-trivial fixed point, and we proceed by induction on dim V .
Finally, look at the case of G = SO(V ) and p = 2. Set W = V Q ; if
W ∩W ⊥ = 0, we proceed as above to get a contradiction. So X := W ∩W ⊥ =
0. If this space contains non-zero singular vectors, then set U := {v ∈ X |
Q(v) = 0}, a totally singular subspace, stabilized by P , fixed pointwise by
Q. The remainder of the argument goes through as in the preceding case.
Now suppose W ∩W ⊥ contains no non-zero singular vectors, so dim X = 1.
If dim V = 2n, so that the bilinear form is non-degenerate, P acts on X ⊥ /X
a non-degenerate symplectic space. Argue using the quadratic form that
the kernel of this action is trivial. So P embeds in Sp2n−2 and we have a
contradiction by comparing the dimensions of Borel subgroups.
Now suppose dim V = 2n + 1. Then X = rad(βQ ), and in V /X, we
have the direct sum decomposition W/X ⊕ W ⊥ /X = W/X ⊕ (W/X)⊥ ,
stabilized by P , and hence by a Borel subgroup of Isom(V /X), contradicting
the indecomposability of the latter.
We end our discussion of Levi complements by an observation on isogeny
types.
Proposition 12.14 Let G be semisimple of simply connected type. Then
for any Levi subgroup L of G, the derived subgroup [L, L] is again of simply
connected type.
Proof Let T ≤ G be a maximal torus. Assume that L is the standard
Levi subgroup corresponding to I ⊆ S and let T  := T ∩ [L, L], a maximal
torus of [L, L]. We have to show that Hom(ZΦ∨ 
I , Z) ≤ X(T ). For this, let
 ∨ ∨
χ ∈ Hom(ZΦI , Z). Extend it to χ ∈ Hom(ZΦ , Z) by setting χ(α) = 0 for
α ∈ ∆ \ ∆I . Since G is simply connected, χ ∈ X(T ). But then χ = χ|T  ∈
X(T  ) as claimed.
See Exercise 20.7 for one generalization of this statement, and Theo-
rem 14.16(b) for a generalization in a different direction.
We’ll have more to say about the structure of unipotent radicals of para-
bolic subgroups in Chapter 17, once we’ve prepared the necessary tools from
representation theory.
 13
Subgroups of maximal rank

Let G be a semisimple linear algebraic group. Many naturally defined closed

subgroups of G are normalized by a maximal torus T of G, for example
those containing a maximal torus. The following are some examples of such
subgroups:

• parabolic subgroups containing T ;

• Levi complements of parabolic subgroups;
• centralizers CG (s) of elements s ∈ T .

The aim of this chapter is a classification of all semisimple such subgroups

in terms of data from the root system of G.

13.1 Subsystem subgroups

We are led to make the following definition:

Definition 13.1 A subsystem subgroup of a connected reductive group G

is a semisimple subgroup normalized by a maximal torus of G.

One family of subsystem subgroups of a group G arises from certain sub-

sets of the root system of G.

Definition 13.2 Let Φ be a root system. A subset Ψ ⊆ Φ is said to be

closed if

(C1) for all α, β ∈ Ψ we have sα .β ∈ Ψ, and

(C2) for α, β ∈ Ψ with α + β ∈ Φ, we have α + β ∈ Ψ.

The subset Ψ is called p-closed if it satisfies (C1) and moreover

 13.1 Subsystem subgroups 105

(C2p) for α, β ∈ Ψ, m, n > 0, we have mα+nβ ∈ Ψ whenever in a semisimple

group of type Φ over a field k of characteristic p the structure constant
cmn
αβ (from the commutator formula in Theorem 11.8) is non-zero in k.

Note that by Theorem 11.8 the integers cmn αβ only depend on the root
system, not on the isogeny type of a corresponding semisimple group G.
It is clear from (C1) that closed and p-closed subsets are root systems in
their own right, in the appropriate Euclidean spaces.

Example 13.3 If ∆I is a subset of a base ∆ of Φ, then the corresponding

parabolic subsystem ΦI is a closed subset. Also, for Φ indecomposable, any
subsystem consisting of long roots can be seen to be closed (see Exercise B.2).

It can be seen from the classification of two-dimensional root systems that

a closed subsystem is p-closed for all p ≥ 0. In order to discuss the converse,
one needs a knowledge of the values of the structure constants cmnαβ occurring
in the commutator relations, as given for example in [13, Thm. 5.2.2]. From
this information, the following is easily verified:

Proposition 13.4 Let Φ be an indecomposable root system, Ψ ⊆ Φ a p-

closed subset. Then Ψ is closed unless possibly if Ψ contains some short roots
and either Φ is of type Bn , Cn , F4 and p = 2, or Φ is of type G2 and p = 3.

We now discuss the relationship between subsystem subgroups of a reduc-

tive group and p-closed subsystems of its root system.

Proposition 13.5 Let G be connected reductive with root system Φ and

H ≤ G a subsystem subgroup. Then the root system of H can be naturally
regarded as a subsystem of Φ, and as such it is char(k)-closed.

Proof By assumption H is normalized by a maximal torus T of G, so H̃ :=

HT is connected reductive. Let B1 be a Borel subgroup of H̃ containing T ,
and B a Borel subgroup of G containing B1 . The root subgroups in B1 are
one-dimensional connected unipotent T -invariant subgroups of B, hence of
the form Uα for some α ∈ Φ by Proposition 11.5. This defines an embedding
of the root system Ψ of H = [H̃, H̃] with respect to T ∩ H into Φ.
Clearly Ψ satisfies (C1). Now let α, β ∈ Ψ+ (without loss: see Exer-
cise A.5). Then Uα , Uβ ≤ B1 , hence also [Uα , Uβ ] ⊆ B1 . Again by Propo-
sition 11.5 and Theorem 11.8, it follows that B1 will contain all Uγ where
γ = mα + nβ with cmn αβ = 0 in k. Thus Ψ is char(k)-closed.

Conversely, we obtain the following generalization of Proposition 12.6 for

Levi complements:
106 Subgroups of maximal rank

Theorem 13.6 Let G be connected reductive with root system Φ, and Ψ ⊆

Φ a p-closed subset, where p = char(k). Then G(Ψ) := T, Uα | α ∈ Ψ is a
connected reductive subgroup with root system Ψ and Weyl group W (Ψ) :=
sα | α ∈ Ψ.
Proof As remarked above, the p-closed subset Ψ is a root system in RΨ. Let
∆1 be a base of Ψ with positive system Ψ+ . By the commutator relations
 
in Theorem 11.8, [Uα , Uβ ] ⊆ m,n>0 Umα+nβ , so B1 := T α∈Ψ+ Uα is a
connected solvable subgroup of G(Ψ) by p-closedness. Let N1 ≤ NG(Ψ) (T )
so that N1 /T = sα | α ∈ Ψ = W (Ψ) is the Weyl group of Ψ. We claim
that B1 , N1 form a BN-pair in G(Ψ) with Weyl group W (Ψ). (Note that at
this point we don’t yet know that G(Ψ) is reductive.)
Axioms (BN1)–(BN3) are immediate. Now let w ∈ W (Ψ) and s = sα with
α ∈ ∆1 . Using that v̇Uβ v̇ −1 = Uvβ for v ∈ W = NG (T )/T we have
B1 ẇB1 ṡB1 ⊆ B1 ẇUα ṡB1 .
If wα ∈ Ψ+ then this lies in B1 ẇṡB1 . Otherwise wsα = −wα ∈ Ψ+ , so using
U−α = ṡUα ṡ ⊆ T Uα ṡUα ∪ Uα from the proof of Lemma 11.14 we have
˙ −α B1 ⊆ B1 ws(U
B1 ẇUα ṡB1 = B1 wsU ˙ α ṡUα ∪ Uα )B1 ⊆ B1 ẇB1 ∪ B1 wsB
˙ 1,
showing (BN4). Finally, ṡB1 ṡ contains U−α , which together with Uα gen-
erates the non-solvable group Uα , U−α , so cannot be contained in B1 , as
required for (BN5).
We next claim that B1 is a Borel subgroup of G(Ψ). The weak form of the
Bruhat decomposition in Theorem 11.17, valid for any BN-pair, gives
)
G(Ψ) = B1 ẇB1 .
w∈W (Ψ)

Now let g ∈ G(Ψ) \ B1 , so g = bẇb for some b, b ∈ B1 and w ∈ W (Ψ) \ {1}.

Let α ∈ Ψ+ with β := wα ∈ −Ψ+ , which exists by Proposition A.21. Then
B1 , g contains ẇUα ẇ−1 = Uβ , hence the non-solvable group Uβ , U−β .
Thus B1 is a maximal connected solvable subgroup of G(Ψ), that is, a Borel
subgroup.
Hence B1 can be embedded into a Borel subgroup B of G, with maximal
torus T . The unipotent radical Ru (G(Ψ)) is then a T -invariant subgroup of
B1 , hence of B, hence generated by suitable root subgroups Uβ , β ∈ Φ, by
Proposition 11.5. Assume that Uβ ≤ Ru (G(Ψ)) ≤ B1 . Then β ∈ Ψ by the
definition of B1 and Theorem 11.1(a), so U−β ≤ G(Ψ). But Uβ , U−β  has

trivial radical, so Ru (G(Ψ)) = 1. Finally, as B1 = T α∈Ψ+ Uα is a Borel
subgroup of G(Ψ), it follows from Theorem 11.1 that Ψ is the root system
of G(Ψ).
 13.2 The algorithm of Borel and de Siebenthal 107

The preceding two results can be summarized as follows:

Corollary 13.7 Let G be connected reductive with maximal torus T . The

map Ψ → [G(Ψ), G(Ψ)] establishes a bijection between char(k)-closed subsets
Ψ ⊆ Φ of the root system with respect to T , and subsystem subgroups of G
normalized by T .

We have the following generalization of Corollary 12.11 (see Exercise 20.8):

Proposition 13.8 Let G be semisimple with a maximal torus T with Weyl

group W , H ≤ G a subsystem subgroup normalized by T with Weyl group
WH . Then there is a natural isomorphism NG (HT )/HT ∼
= NW (WH )/WH .

Example 13.9 Let G be a simple algebraic group of type F4 , and H the

subsystem subgroup of type D4 generated by the long root subgroups (cor-
responding to the closed subsystem consisting of all long roots). It can be
calculated with Proposition 13.8 that the normalizer NG (H) is an exten-
sion of H by the symmetric group S3 ; as in Example 12.12 this realizes
the full group of graph automorphisms of a group of type D4 described in
Theorem 11.12.

13.2 The algorithm of Borel and de Siebenthal

The preceding results reduce the question of determining all subsystem sub-
groups to the purely combinatorial task of finding all p-closed subsets of the
root system. Clearly it is sufficient to do this in the indecomposable case. For
the case of closed subsets the algorithm described by Borel and de Sieben-
thal does precisely this. In order to formulate it we need a further ingredient
from the theory of root systems, see Proposition B.5:

Proposition 13.10 Let Φ be an indecomposable root system with base ∆.

Then there exists a unique root α0 ∈ Φ+ with the following property: Writing
 
α0 = α∈∆ nα α, then for every root β = α∈∆ cα α ∈ Φ+ we have cα ≤ nα
for all α ∈ ∆.

Definition 13.11 The height of a root β = α∈∆ cα α ∈ Φ is ht(β) :=

α∈∆ cα . The root α0 above is called the highest root of Φ with respect to ∆
(since, clearly it has the largest height of all roots in Φ). The extended Dynkin
diagram of an indecomposable root system Φ is the diagram obtained from
the set ∆ ∪ {−α0 } by the same procedure by which the ordinary Dynkin
diagram is obtained from ∆.
108 Subgroups of maximal rank

See Table B.1 for the list of highest roots in indecomposable root systems.
The extended Dynkin diagrams for the indecomposable root systems are
shown in Table 13.1 (see [34, §4.7]). It is clear from this list that the highest
root is always a long root, if Φ has two root lengths. This can also be shown
without using the classification of root systems, see Corollary B.7.

Table 13.1 Extended Dynkin diagrams

1 2 3 n 0 1 2 n−1 n
Ãn C̃n > <
n≥2 n≥2

n−1
0
0 2 3 n−2
D̃n
0 n≥4

2 3 n−1 n 1 n
B̃n >
n≥3 1 2 0
G̃2 <
1
0 1 3 4 5 6 7
0 1 2 3 4 Ẽ7
F̃4 >
2
1 3 4 5 6
Ẽ6
1 3 4 5 6 7 8 0
Ẽ8
2
2
0

The following result (see Theorem B.18), which was first shown in [5, §7]
in the context of compact Lie groups exhibits the role played by the extended
Dynkin diagrams in the study of subsystem subgroups:
Theorem 13.12 (Borel–de Siebenthal) Let Φ be an indecomposable root

system with base ∆ and highest root α0 = α∈∆ nα α with respect to ∆.
Then the maximal closed subsystems of Φ up to conjugation by W are those
with bases:
(1) ∆ \ {α} ∪ {−α0 }, for α ∈ ∆ with nα a prime, and
(2) ∆ \ {α} for α ∈ ∆ with nα = 1.
It is thus easy to write down explicitly the possibilities for maximal closed
subsystems in any of the various types of indecomposable root systems, see
Exercise 20.9.
 13.2 The algorithm of Borel and de Siebenthal 109

The algorithm of Borel–de Siebenthal for the determination of all closed

subsystems of a root system Φ now proceeds as follows. For any proper
subset of ∆ ∪ {−α0 }, corresponding to a subdiagram of the extended Dynkin
diagram, form the extended Dynkin diagram of each indecomposable part
of that subdiagram and repeat the process. At any stage of the process, the
set of nodes of the current diagram is a subset J ⊆ Φ and Ψ := ZJ ∩ Φ is a
closed subset of Φ.

Example 13.13 We start with a root system of type E8 . The correspond-

ing extended Dynkin diagram is as shown in Figure 13.1.

1 3 4 5 6 7 8 0
Ẽ8

Figure 13.1 The extended Dynkin diagram for E8 .

Removal of the first node leads to the Dynkin diagram of type D8 , so the
E8 -root system contains a closed subsystem of type D8 . Since the highest
root for E8 is given by α0 = 2α1 + 3α2 + 4α3 + 6α4 + 5α5 + 4α6 + 3α7 + 2α8
(see Table B.1), this is in fact maximal in E8 by Theorem 13.12.

0 7 0 3
2 3 4 5 6 2
D̃8 D̃4

1 8 1 4

Figure 13.2 The extended Dynkin diagrams for D8 and D4 .

The corresponding extended Dynkin diagram D̃8 is shown in Figure 13.2.

Removing the node with label 4, we reach a diagram of type D4 D4 . Therefore
the E8 root system has a closed subsystem of type D4 D4 . Clearly, this is not
maximal inside E8 , but it is so inside the root system of type D8 , since the
corresponding coefficient in the highest root for D8 equals 2.
We now take the extended Dynkin diagram of the D4 D4 diagram. We see
that there is a subsystem subgroup with root system of type (A1 )4 of each of
110 Subgroups of maximal rank

the D4 subgroups and hence E8 has a subsystem subgroup with root system
of type (A1 )8 . In particular we have inclusions of subsystem subgroups of
E8 as follows:
(A1 )8 ≤ D4 D4 ≤ D8 ≤ E8 .

Here, by abuse of notation, we identify the group with its root system.

We next discuss the extent to which closed subsystems account for all
subsystem subgroups.

Theorem 13.14 Let G be a simple algebraic group with root system Φ.

The algorithm of Borel–de Siebenthal described above gives all subsystem
subgroups of G, except when

(1) Φ is of type Bn , Cn , F4 and char(k) = 2, or

(2) Φ is of type G2 and char(k) = 3.

Proof By Corollary 13.7 subsystem subgroups correspond bijectively to p-

closed subsystems of Φ. By Proposition 13.4 these are even closed in Φ unless
Φ and p are as in the exceptions. The result then follows from Theorem 13.12
by induction.

In the exceptional cases there exist subsystem subgroups corresponding

to non-closed subsystems; the latter can also be enumerated, see Proposi-
tion B.21 and Example B.22, and we arrive at the following:

Proposition 13.15 Let G be a simple algebraic group with root system

Φ and assume further that (Φ, p) are as in (1) or (2) of Theorem 13.14.
Then any non-closed p-closed subsystem Ψ of Φ contains short roots and if
Ψ contains all short roots of Φ, one of the following holds:

(1) (G, char(k)) = (Cn , 2) and Ψ has type Dn ,


(2) (G, char(k)) = (Bn , 2) and Ψ has type Bm1 · · · Bmr with mi = n,
(3) (G, char(k)) = (F4 , 2) and Ψ has type C4 or D4 , or
(4) (G, char(k)) = (G2 , 3) and Ψ has type A2 .

We illustrate the exceptional behavior for small primes in one example:

Example 13.16 Let G = Sp4 , with root system of type C2 , with base
{α, β}, where α is short. Here U±β ×U±(2α+β)  is a subsystem subgroup of
type A1 A1 corresponding to the subdiagram of the extended Dynkin diagram
of G where the middle node is removed, see Table 13.1.
Now consider the short roots, α, α+β. If p = 2 we have [Uα , Uα+β ] = 1 and
indeed the group generated by the root subgroups corresponding to the short
 13.2 The algorithm of Borel and de Siebenthal 111

roots is all of G. However, if p = 2 we have [U±α , U±(α+β) ] = 1 (see Exam-

ple 11.9(2)) and the subgroup U±α  × U±(α+β)  is also a proper subsystem
subgroup of type D2 = A1 A1 , which arises from the fact that the group SO4
is naturally a subgroup of the symplectic group Sp4 in characteristic 2.
 14
Centralizers and conjugacy classes

We now consider properties like generation, conjugacy, classification, con-

nectedness and dimension of centralizers in connected reductive groups. It
turns out that the situation is easiest for semisimple elements.

14.1 Semisimple elements

Recall from Corollary 6.11(a) that every semisimple element of a connected
group lies in some maximal torus. More precisely we have:

Proposition 14.1 Let G be connected, s ∈ G semisimple, T ≤ G a maxi-

mal torus. Then s ∈ T if and only if T ≤ CG (s)◦ . In particular, s ∈ CG (s)◦ .

Proof As T is abelian, s ∈ T if and only if T ≤ CG (s), which is equivalent

to T ≤ CG (s)◦ as T is connected.

We remark that in contrast, for u ∈ G unipotent, u may not be in CG (u)◦ .

See Exercise 20.10 for an example in Sp4 over a field of characteristic 2.
We now determine the structure of centralizers of semisimple elements:

Theorem 14.2 Let G be connected reductive, s ∈ G semisimple, T ≤ G a

maximal torus with corresponding root system Φ. Let s ∈ T and Ψ := {α ∈
Φ | α(s) = 1}. Then:

(a) CG (s)◦ = T, Uα | α ∈ Ψ.

(b) CG (s) = T, Uα , ẇ | α ∈ Ψ, w ∈ W with sw = s.

Moreover, CG (s)◦ is reductive with root system Ψ and Weyl group W1 =

sα | α ∈ Ψ.
 14.1 Semisimple elements 113

Proof The inclusion “⊇” is clear in both (a) and (b) (see Theorem 8.17(c)
for the action of s on Uα ). Now let g ∈ CG (s). By the Bruhat decomposition
in Theorem 11.17, g can be uniquely expressed as a product g = u1 ẇtu2
with u1 ∈ Uw− , w ∈ W , t ∈ T and u2 ∈ U = Ru (B). Now

sgs−1 = su1 s−1 · ẇ · sw ts−1 · su2 s−1 = g = u1 ẇtu2


implies u1 , u2 ∈ CG (s), s = sw . But, if u = α∈Φ+ uα (cα ) ∈ U with cα ∈ k
then
!
sus−1 = uα (α(s) cα ),
α∈Φ+

so u ∈ CG (s) implies that α(s) = 1 whenever cα = 0, by Theorem 11.1.

This shows the inclusion “⊆” in (b). Now T, Uα | α ∈ Ψ is a connected
subgroup of CG (s) of index at most |W |, hence the identity component of
CG (s), which completes the proof of (a).
As Ψ satisfies (C1) (by Lemma 8.19(a)) and (C2) from Definition 13.2, it is
a closed subsystem of Φ. The last claim then follows from Theorem 13.6.

Corollary 14.3 Let G be connected reductive. Then up to conjugation,

there exist only finitely many different centralizers of semisimple elements
in G.

Proof Let s ∈ G be semisimple. After conjugation we may assume that

s ∈ T . Then by Theorem 14.2 the centralizer CG (s) is determined by Ψ ⊆ Φ
and the subgroup {w ∈ W | sw = s}. But for both of these, there are just
finitely many possibilities.

Example 14.4 (1) Let G = GLn , s = diag(t, u, . . . , u) ∈ G with t = u.

Then CG (s) = GL1 × GLn−1 , a Levi subgroup of the parabolic subgroup
⎧⎛ ⎞ ⎫

⎪ ∗ ∗. . .∗
⎪
⎨  ⎪
⎪
⎬
⎜ 0 ⎟
⎜ . ⎟  A ∈ GLn−1 .
⎪
⎪
⎩
⎝ .. A ⎠  ⎪
⎪
⎭
0 

(2) In characteristic char(k) = 2 the image s of diag(1, −1) in G = PGL2 is

a non-central semisimple element whose centralizer (see Exercise 20.14)
is not contained in a proper Levi subgroup; such semisimple elements are
called quasi-isolated . In this example at least CG (s)◦ is a torus, hence a
proper Levi subgroup.
(3) Let G = Sp4 , char(k) = 2, s = diag(1, −1, −1, 1). Then CG (s) = Sp2 ×
Sp2 , which is a subgroup of maximal semisimple rank, hence not even
114 Centralizers and conjugacy classes

CG (s)◦ lies inside a proper Levi subgroup. Such semisimple elements are
called isolated .

Remark 14.5 There is a criterion established by Carter which allows one

to decide combinatorially via the root datum which maximal rank subgroups
actually do occur as centralizers of semisimple elements, see for example [35,
2.12]. Deriziotis [20] has shown that up to conjugation the root systems of
centralizers of semisimple elements always have a base contained in ∆ ∪
{−α0 }, see Exercise 20.13.

The latter result relies on the following parametrization of semisimple

classes (see Exercise 20.12):

Proposition 14.6 The set of semisimple conjugacy classes of a connected

reductive group G is in bijection with the orbits of its Weyl group on a
maximal torus.

General conjugacy classes of a reductive group G can be described as

follows. Let g ∈ G, with Jordan decomposition g = us. If h ∈ CG (g), then h
centralizes both s and u, by the uniqueness of Jordan decomposition. Hence
CG (g) = CH (u), where H := CG (s) is reductive by Theorem 14.2. Thus,
the centralizer of g is known once the centralizers of unipotent elements
in certain reductive (but not necessarily connected) subgroups are known.
Moreover we have:

Proposition 14.7 Let G be connected reductive, g ∈ G with Jordan de-

composition g = su. Then u ∈ CG (s)◦ .

Proof Let B be a Borel subgroup containing g, and T ≤ B a maximal

torus containing s (see Theorem 6.10 and Corollary 4.5). Then u lies in
CB (s)∩Ru (B), a unipotent subgroup of B normalized by T , hence connected
by Proposition 11.5.

In particular, for s semisimple, |CG (s) : CG (s)◦ | is prime to char(k). See

Proposition 14.20 below for a much stronger assertion.
Unfortunately, there is no systematic description of unipotent elements
and their centralizers comparable to that given above for semisimple ele-
ments. At present, the conjugacy classes and centralizers of unipotent ele-
ments in simple algebraic groups have to be determined case-by-case, using
the natural matrix representation (and refined Jordan canonical forms) for
the classical groups and explicit computations with root elements for the
exceptional types. In principle this then allows one to parametrize arbitrary
 14.1 Semisimple elements 115

conjugacy classes. See Springer and Steinberg [67], Steinberg [75], and Spal-
tenstein [64] for further information.
We conclude our investigation of conjugacy classes by introducing an im-
portant type of element, which in a sense comprises most elements in a
connected group (see Corollary 14.10 below).
Definition 14.8 An element x of a linear algebraic group G is called
regular if dim CG (x) is smallest possible among all elements of G.
Proposition 14.9 Let G be connected, x ∈ G. Then dim CG (x) ≥ rk(G).
Proof Let B be a Borel subgroup of G containing x (according to Theo-
rem 6.10), with unipotent radical U = Ru (B). As B/U is abelian by Theo-
rem 4.4(a), the conjugacy class [x]B of x in B is contained in the coset xU ,
whence of dimension at most dim U = dim B − dim T , where T denotes a
maximal torus of B, hence of G. Consider the surjective morphism B → [x]B ,
g → gxg −1 , with fibers the cosets of CB (x), of dimension dim CB (x). This
shows that dim[x]B = dim B − dim CB (x) by Proposition 1.19, so
dim B − dim CB (x) = dim[x]B ≤ dim B − dim T = dim B − rk(G),
whence dim CG (x) ≥ dim CB (x) ≥ rk(G) as claimed.
Regular elements play an important role in the study of reductive groups.
For semisimple elements, we have the following characterization:
Corollary 14.10 Let G be connected reductive with maximal torus T and
root system Φ. For s ∈ T the following are equivalent:
(i) s is regular;
(ii) α(s) = 1 for all α ∈ Φ;
(iii) CG (s)◦ = T .
Moreover, regular semisimple elements are dense in G.
Proof By Theorem 14.2(a) the conditions (ii) and (iii) are equivalent. As
dim T = rk(G), any element with CG (s)◦ = T is necessarily regular by
Proposition 14.9, so (ii), (iii) imply (i). Now ker(α) is a (closed) subgroup
of T of codimension 1 for any α ∈ Φ, and |Φ| is finite, so the complement
of ∪α ker(α) is non-empty open in T and thus dense by Proposition 1.9.
In particular, again by Theorem 14.2(a), the bound in Proposition 14.9 is
attained and so (i) implies (iii).
Now let B denote a Borel subgroup of G containing T , with unipotent
radical U . Let Treg denote the open set of regular semisimple elements of
G in T . We claim that Treg · U consists of regular semisimple elements of
116 Centralizers and conjugacy classes

G, and thus regular semisimple elements (of G) are dense in B. Indeed, let
su ∈ Treg U with Jordan decomposition tv, where t is semisimple and v ∈ U .
As U acts transitively by conjugation on the set of maximal tori of B, the
semisimple part t of su is conjugate to s, hence also regular in G. By the
first part, CB (t)◦ is a torus, so v = 1 by Proposition 14.7, and hence su = t
is regular semisimple.
Since G is the union of conjugates of B (see Theorem 6.10), regular
semisimple elements are also dense in G.

For example in GLn , regular elements are those whose characteristic poly-
nomial and minimal polynomial agree. This shows that GLn and SLn also
do contain regular unipotent elements. The proof of this fact for other types
of semisimple groups is much more delicate, see [71, Thm. 4.6].

14.2 Connectedness of centralizers

We have seen in Example 14.4 that, in general, centralizers of semisimple ele-
ments need not be connected. In order to investigate the extent to which this
fails, we’ll need the following important notions related to closed subsystems
of root systems:

Definition 14.11 Let Φ be a root system. A prime r is called bad for Φ if

ZΦ/ZΨ has r-torsion for some closed subsystem Ψ ⊆ Φ.
Let Φ∨ denote the dual root system (see Exercise 10.35 or Definition A.3).
The prime r is said to be a torsion prime for Φ if ZΦ∨ /ZΨ∨ has r-torsion
for some closed subsystem Ψ ⊆ Φ.

Example 14.12 Let Φ be a root system of type G2 , with base {α, β}

where α is short (see Example 9.5). Then Ψ1 := ±{α, 3α + 2β} is a closed
subsystem of type A1 A1 , and clearly |ZΦ/ZΨ1 | = 2; furthermore, Ψ2 =
±{β, 3α + β, 3α + 2β} is a closed subsystem of type A2 with |ZΦ/ZΨ2 | = 3,
so 2 and 3 are certainly bad primes for G2 .
In the dual root system, Ψ∨ ∨ ∨ ∨ ∨ ∨
1 = ±{α , α + 2β } still has |ZΦ /ZΨ2 | = 2,
∨ ∨ ∨ ∨ ∨ ∨ ∨
but Ψ2 = ±{β , α + β , α + 2β } now spans ZΦ , so we only find the
torsion prime 2. Note that Ψ∨ ∨
2 is not a closed subsystem of Φ .

Proposition 14.13 The bad primes, respectively torsion primes, for the
indecomposable root systems are as given in Table 14.1. The bad primes and
torsion primes for a decomposable root system are those of its indecomposable
components.
 14.2 Connectedness of centralizers 117

Table 14.1 Bad primes and torsion primes of root systems

Φ bad primes torsion primes

An − −
Bn (n ≥ 3), Dn (n ≥ 4) 2 2
Cn (n ≥ 2) 2 −
G2 2, 3 2
F4 , E6 , E7 2, 3 2, 3
E8 2, 3, 5 2, 3, 5

Proof This can be verified from the criteria involving coefficients of highest
roots in Corollary B.28 and Proposition B.32; see also Exercise B.4.

Definition 14.14 Let G be a connected reductive group. A prime r is

called bad for G if it is bad for the root system of G. It is called a torsion
prime for G if the fundamental group of some subsystem subgroup of G has
r-torsion.

Here, the fundamental group of a reductive group H is by definition that

of its semisimple part [H, H], see Definition 9.14.

Proposition 14.15 The torsion primes of a connected reductive group G

are precisely the torsion primes of its root system Φ together with the prime
divisors of the order of its fundamental group.

Proof This is Exercise 20.15.

The fundamental groups for the various simple groups can be read off
from Table 9.2. Comparison with that list shows that for simple groups G
the torsion primes for G are those of its root system Φ, except possibly for
prime divisors of n + 1 for type An , and for 2 for type Cn .
We can now characterize an important situation in which centralizers of
semisimple elements are always connected (see [14, Thm. 3.5.6] for (a), [74,
Thm. 0.1] for (b)):

Theorem 14.16 (Steinberg) Let G be connected reductive such that the

derived group [G, G] is simply connected, and s ∈ G a semisimple element.
Then:

(a) CG (s) is connected.

(b) If the order of s in G/Z(G) is finite, but not divisible by any torsion
prime of G then [CG (s), CG (s)] is again simply connected.
118 Centralizers and conjugacy classes

Note that part (b) generalizes Proposition 12.14: by Proposition 12.6 any
Levi subgroup L of G is the centralizer of its connected center Z(L)◦ ; by
Exercise 20.11 we can choose a semisimple element s ∈ Z(L)◦ of order prime
to all torsion primes with L = CG (s).
Corollary 14.17 Let G be connected reductive with simply connected de-
rived subgroup [G, G], and s1 , . . . , sr ∈ G mutually commuting semisimple
elements of finite order. Let n be the number of si whose order is not prime
to all torsion primes of G. If n ≤ 2 then there exists a maximal torus of G
containing all si .
Proof Assume that s1 , . . . , sr−2 have order not divisible by any torsion
prime of G, and for i = 1, . . . , r put Gi := CGi−1 (si ), with G0 := G. By
Theorem 14.16(a) and (b) applied inductively we see that Gi is connected
and [Gi , Gi ] is simply connected for i = 1, . . . , r − 2. Furthermore, Gr−1 is
still connected and contains sr . Now any maximal torus of Gr−1 containing
sr (see Corollary 6.11(a)) is as claimed.
Corollary 14.18 Let G be a semisimple group, r = char(k) a prime which
is not a torsion prime for G. Then the maximal rank of an elementary abelian
r-subgroup of G equals rk(G).
Proof Let R be an elementary abelian r-subgroup of G and consider the
natural isogeny π : Gsc → G from a simply connected group of the same
type (see Proposition 9.15). Then ker(π) is a subgroup of the fundamental
group of G, so of order prime to r. Hence, the preimage π −1 (R) ≤ Gsc has
a Sylow r-subgroup R̃ isomorphic to R. Since r is not a torsion prime, by
Corollary 14.17 there exists a maximal torus T ≤ Gsc with R̃ ≤ T . Since T ∼=
(k × )l , with l = rk(G), we see that R ∼
= R̃ has r-rank at most l. Conversely,
T clearly contains an elementary abelian r-subgroup of rank l.
Example 14.19 We illustrate Theorem 14.16 and its corollaries.
(1) For G = GLn or SLn the derived group G = SLn (by Exercise 10.36)
is simply connected, and indeed, in both groups any set of commuting
semisimple elements can be simultaneously diagonalized, hence embed-
ded into a maximal torus.
(2) The assumption on torsion primes in Theorem 14.16(b) is necessary:
Let G = G2 with char(k) = 2, a group of simply connected type,
with root system Φ with respect to a maximal torus T ≤ G. Then
the closed subsystem Ψ1 of type A1 A1 defined in Example 14.12 satis-
fies |ZΦ/ZΨ1 | = 2. Let H denote the subsystem subgroup of G of type
A1 A1 corresponding to Ψ1 and containing T (see Theorem 13.6). As G
 14.2 Connectedness of centralizers 119

is also of adjoint type, X(T ) = ZΦ, hence H has |X(T ) : ZΨ1 | = 2,

while its fundamental group is of order 4 (see Table 9.2). So it is a quo-
tient of the simply connected group SL2 × SL2 of type A1 A1 by a central
subgroup of order 2. Thus |Z(H)| = 2, and by maximality of Ψ1 in Φ,
H is the centralizer of Z(H), and it is not of simply connected type.
On the other hand, when char(k) = 3, the closed subsystem Ψ2 from
Example 14.12 belongs to the simply connected centralizer SL3 of a
semisimple element of order 3, consistent with the fact that 3 is not a
torsion prime for G2 .
(3) The assumption that G is simply connected in Corollary 14.17 is neces-
sary: the quaternion group Q8 of order 8 has a faithful two-dimensional
representation over k when char(k) = 2, embedding it into SL2 . The im-
age of Q8 in the quotient PGL2 = SL2 /Z(SL2 ) is elementary abelian of
order 4, hence clearly does not embed into a (one-dimensional) maximal
torus of PGL2 . Note that the group PGL2 is not simply connected (see
Table 9.2)

Cases where the conclusion of Corollary 14.18 fails for torsion primes lead
to interesting maximal subgroups, see Theorem 29.3 below.
For semisimple groups, Theorem 14.16 can be generalized as follows: the
extent to which centralizers of semisimple elements can fail to be connected
is controlled by the fundamental group.

Proposition 14.20 Let G be a semisimple algebraic group, π : Gsc → G

the natural isogeny from a simply connected group of the same type as G.
Then for any semisimple s ∈ G, the group of components CG (s)/CG (s)◦ is
isomorphic to a subgroup of ker(π) ≤ Z(Gsc ). Moreover, if s is of finite order
then the exponent of CG (s)/CG (s)◦ divides the order of s.

Proof This follows from Theorem 14.16(a), see Exercise 20.16.

Example 14.21 (Some isolated elements in E6 ) Let G be simple of simply

connected type E6 and assume char(k) = 3. The subsystem subgroup H of
G corresponding to the subset J = {0, 1, 2, 3, 5, 6} of the extended Dynkin
diagram of E6 (see Table 13.1) has type A32 .
1 3 4 5 6

0
120 Centralizers and conjugacy classes

An application of Proposition 13.8 yields that NG (H)/H ∼ = S3 . Direct

calculation with the root datum shows that H is a quotient of SL33 by a
central subgroup of order 3, so has center C3 × C3 . By Theorem 13.12, H
is a subsystem subgroup for a maximal subsystem, so it is the connected
component of the centralizer C := CG (s) of any semisimple element s ∈
Z(H) \ Z(G). In fact, C = C ◦ = H is connected by Theorem 14.16. In
particular, NG (H)/H acts fixed point freely, hence transitively, on the six
elements of Z(H) \ Z(G), so there is a unique conjugacy class of elements of
order 3 in G with centralizer H.
The image H̄ = H/Z(G) of H in the group Ḡ = G/Z(G) of adjoint type
has center of order 3. Let 1 = s ∈ Z(H̄), with centralizer C = CḠ (s). Then,
as above we have C ◦ = H̄, and C/C ◦ is a subgroup of the fundamental
group Z3 by Proposition 14.20. Since clearly the elements of order 3 in
= S3 must act trivially on Z(H̄), we see that C/C ◦ ∼
NḠ (H̄)/H̄ ∼ = Z3 .
Again, the quotient NḠ (C)/C ∼ = Z2 has just one orbit on the elements of
Z(H̄) \ Z(Ḡ), of length 2.
 15
Representations of algebraic groups

In Chapter 18 we will consider the subgroup structure of the classical al-

gebraic groups SLn , Sp2n and SOn . Hence we are interested in morphisms
ρ : G → GLn of algebraic groups, that is rational representations of a linear
algebraic group G. As with finite groups, the representation theory of linear
algebraic groups is a tool in the study of a wide variety of questions, not only
those directly concerned with the subgroup structure of classical groups.
We begin by developing the theory necessary for describing the irreducible
representations of semisimple groups. It turns out that the irreducible rep-
resentations are parametrized by “abstract weights” as in the representation
theory of complex semisimple Lie algebras.

15.1 Weight theory

Let ρ : G → GL(V ) be a rational representation of the linear algebraic
group G. The following result shows that if we are interested in irreducible
representations, we may as well assume that G is semisimple:

Proposition 15.1 Let ρ : G → GL(V ) be an irreducible rational represen-

tation of a connected linear algebraic group G. Then:

(a) Ru (G) ≤ ker(ρ).

(b) [G/Ru (G), G/Ru (G)] acts irreducibly on V .

Proof By Proposition 2.9, the fixed point space V ρ(Ru (G)) of the unipotent
radical Ru (G) of G is a non-trivial G-invariant subspace of V , so all of V by
irreducibility. That is, Ru (G) ≤ ker(ρ).
So ρ induces an irreducible representation ρ̄ : Ḡ → GL(V ) of the con-
nected reductive group Ḡ := G/Ru (G). Moreover, as G acts irreducibly on
122 Representations of algebraic groups

V , R(Ḡ) ≤ Z(Ḡ) acts by scalars by Schur’s Lemma. As Ḡ = R(Ḡ)[Ḡ, Ḡ] by

Corollary 8.22, the semisimple group [Ḡ, Ḡ] (see Proposition 6.20(c)) also
acts irreducibly on V .

This seems to lead us to concentrate on semisimple groups; however for

inductive arguments (where we pass to Levi factors of parabolic subgroups),
it will be convenient to consider the more general setting of reductive groups.
So let G be a connected reductive algebraic group. Let T be a maximal torus
of G contained in a Borel subgroup B of G, Φ the root system with respect
to T and Φ+ a positive system corresponding to B with base ∆ ⊂ Φ+ . Let
U = Uα | α ∈ Φ+  and U − = U−α | α ∈ Φ+ . Let W = NG (T )/T be the
Weyl group of G.
Let ρ : G → GL(V ) be a rational representation. As ρ(T ) is a set of
commuting semisimple elements in GL(V ), ρ(T ) is a diagonalizable subgroup
and V can be decomposed as a direct sum of common eigenspaces for T ,
,
V = Vχ , where Vχ = {v ∈ V | ρ(t)v = χ(t)v for all t ∈ T }.
χ∈X(T )

Definition 15.2 Let χ ∈ X(T ) be such that Vχ = 0. Then χ is said to

be a weight of V (or a weight of the corresponding representation ρ) with
respect to the torus T and Vχ a T -weight space of V . (If it is clear in the
given context, we may omit the reference to the torus T .) The multiplicity
of χ in V is dim Vχ .

Thus, for example, the weights in the adjoint representation of G are just 0
and the roots of G. The multiplicity of a non-zero weight (that is, a root) is 1
and the multiplicity of the 0 weight is the rank of G by Theorem 8.17(a). The
weights of the natural representation of GLn with respect to the maximal
torus Dn of diagonal matrices are just the n coordinate functions, again with
one-dimensional weight spaces.
We consider the action of certain elements and subgroups on weight spaces
of V , starting with NG (T ). As seen in Section 8.1, the Weyl group W acts
on X(T ): for w ∈ W and χ ∈ X(T ) we have w.χ(t) = χ(ẇ−1 tẇ), where
ẇ ∈ NG (T ) is an arbitrary preimage of w under the natural map NG (T ) →
NG (T )/T = W . This leads to the following generalization of Proposition 8.4:

Lemma 15.3 Let w ∈ W . Then ẇ(Vχ ) = Vw.χ ; in particular weights in

the same W -orbit have equal multiplicities.

Proof Let v ∈ Vχ and t ∈ T . Then

t(ẇv) = ẇ(ẇ−1 tẇ)v = ẇχ(ẇ−1 tẇ)v = (w.χ)(t)ẇv

 15.1 Weight theory 123

so ẇv ∈ Vw.χ . Hence ẇ(Vχ ) ⊆ Vw.χ and ẇ−1 (Vw.χ ) ⊆ Vχ , and the result
follows.
Recall from Theorem 8.17(g) that a connected reductive group G is gener-
ated by a maximal torus and its root subgroups Uα , for α ∈ Φ. We describe
the action of the root groups Uα on the Vχ .
Lemma 15.4 Let G, T , Φ be as above, ρ : G → GL(V ) a rational repre-
sentation. Let α ∈ Φ and λ ∈ X(T ) with T -weight space Vλ . Then for all
v ∈ Vλ we have
"
Uα .v ⊆ v + Vλ+mα ,
m∈N

so Uα Vλ ⊆ m∈N0 Vλ+mα .
Proof Fix an isomorphism uα : Ga → Uα , c → uα (c). Since G acts ra-
tionally on V , uα (c).v is given by a polynomial function in c with respect
to some basis for V and coordinates with respect to that basis. Collecting
n
terms with equal c-power, we find vi ∈ V such that uα (c).v = i=0 ci vi , for
some n ≥ 0. For t ∈ T we have on the one hand
"
n
tuα (c)t−1 .v = λ(t−1 )tuα (c).v = λ(t−1 ) ci t.vi ,
i=0

but (using Theorem 8.17(c)) also

"
n
tuα (c)t−1 .v = uα (α(t)c).v = (α(t)c)i vi .
i=0

Since this is true for all c ∈ k, the coefficients at equal powers of c have
to agree: λ(t−1 )t.vi = α(t)i vi , that is, t.vi = (λ + iα)(t)vi for all i. Hence
vi ∈ Vλ+iα for all i, and setting c = 0 we also find that v0 = v.

As a consequence, Uα , U−α  preserves the submodule m∈Z Vλ+mα of V .
Example 15.5 Let G be connected reductive with root system Φ. For
α ∈ Φ let nα ∈ Uα , U−α  as defined in Section 8.4. Applying the above
result to the adjoint action of G on its Lie algebra g, we see that nα preserves

the sum of weight spaces m gβ+mα for all β ∈ Φ, so sα .β = β + mα for
some m ∈ Z. This shows that the root system axiom (R4) is satisfied for Φ,
as claimed in Proposition 9.2; that is, we have sα .β − β an integral multiple
of α.
We continue our study of rational representations ρ : G → GL(V ) of a
connected reductive group G, where V = 0. By Lie–Kolchin (Theorem 4.1),
124 Representations of algebraic groups

there exists v + ∈ V \ {0} such that v +  is invariant under the image ρ(B)
of the Borel subgroup B of G. Any such vector v + is called a maximal vector
of V (with respect to B). As v +  is stabilized by any maximal torus T of
B, v + ∈ Vλ for some λ ∈ X(T ).
Proposition 15.6 In the notation introduced above, set V  = Gv + . Then
the weights of V  are of the form
"
λ− mα α with all mα ∈ N0 .
α∈∆

Moreover, λ has multiplicity 1 in V  , and V  has a unique maximal proper

kG-submodule.
Proof For α ∈ Φ, u ∈ Uα , v ∈ Vλ , we have
"
u.v ∈ v + Vλ+mα
m>0

by Lemma 15.4. So by successive applications of elements from U − = Uα |

α ∈ Φ−  we obtain
"
U − .v + ⊆ v + + Vλ+µ .
0=µ∈N0 Φ−

On the other hand, Uα v + = v + for α ∈ Φ+ . So the weights occurring


in U − Bv +  are of the form λ − α∈∆ mα α with all mα ≥ 0 and the
multiplicity of the weight λ is 1. But by Theorem 11.20, U − B is dense in
G, so U − Bv +  = U − Bv +  = Gv +  = V  . For the final statement of the
proposition, let X be a proper kG-submodule of V  . Then v + ∈ / X, and so
λ is not a weight of X. Hence, if we set M to be the sum of all proper kG-
submodules of V  , then M is proper and clearly the unique maximal proper
kG-submodule of V  .
We will see that the weight of a maximal vector must satisfy certain con-
ditions. To describe this we first need the following notions:
Definition 15.7 We say that λ ∈ X(T ) is dominant if λ, α∨  ≥ 0 for all

α ∈ ∆. Given µ, ν ∈ X(T ), we write µ ≤ ν if ν − µ = α∈Φ+ nα α, with
nα ∈ N0 for all α.
For example, if G = T is a torus, then Φ = ∅ and all λ ∈ X(T ) are
dominant. As the following result shows, it is enough to understand the
dominant weights (in terms of the action of the Weyl group on X(T )):
Proposition 15.8 For λ ∈ X(T ), there exists a unique dominant λ+ ∈
X(T ) such that λ ∈ W.λ+ . If λ is dominant, then wλ ≤ λ for all w ∈ W .
 15.2 Irreducible highest weight modules 125

See Lemma B.4 for the proof.

Proposition 15.9 Let ρ : G → GL(V ) be a rational representation of a
connected reductive group G. Let v + be a maximal vector of V (with respect
to B) of weight λ, for some λ ∈ X(T ). Then λ is dominant.
Proof By Proposition 15.8 there exists w ∈ W such that µ := wλ is domi-
nant, and moreover w µ ≤ µ for all w ∈ W . Thus
λ = w−1 (wλ) = w−1 µ ≤ µ = wλ.
So wλ − λ is a sum of positive roots. As Gv +  is a G-invariant subspace
of V , the set of weights occurring in this subspace is invariant under the
action of W . (See Lemma 15.3.) So wλ is also a weight in Gv + . But by

Proposition 15.6 weights in Gv +  are of the form λ− α∈∆ cα α with cα ≥ 0.
So we must have wλ = λ and hence λ is itself dominant.

15.2 Irreducible highest weight modules

The following result shows that the situation for irreducible representations
of connected reductive linear algebraic groups is very similar to that for
semisimple complex Lie algebras:
Corollary 15.10 Let G be connected reductive, V an irreducible rational
kG-module. Then all maximal vectors of V have the same weight λ, which is

dominant, dim Vλ = 1, and all weights in V are of the form λ − α∈∆ cα α
with cα ∈ N0 .
Proof Let v + be a maximal vector for V , of weight λ. Then λ is dominant by
Proposition 15.9, and V  := Gv +  is a kG-submodule, hence all of V . If v is a
maximal vector for some other weight µ, then µ ≤ λ by Proposition 15.6, but
also λ ≤ µ, and hence λ = µ. The remaining statements are also immediate
from Proposition 15.6.
Definition 15.11 Let V be a rational kG-module generated by a maximal
vector v + ∈ V (with respect to the fixed Borel subgroup B), and assume
that v + ∈ Vλ for λ ∈ X(T ). Then V is called a highest weight module of G,
with highest weight λ.
Let H be an algebraic group and V1 and V2 two rational kH-modules
with corresponding rational representations ρi : H → GL(Vi ), i = 1, 2.
We then define an action of H on V1 ⊗ V2 as follows: for vi ∈ Vi we set
g(v1 ⊗ v2 ) := gv1 ⊗ gv2 . It is easy to see that this extends to an action on
126 Representations of algebraic groups

V1 ⊗ V2 . We then obtain a homomorphism ρ : H → GL(V1 ⊗ V2 ) whose

coordinate functions are algebraic expressions in terms of the coordinate
functions of ρ1 and ρ2 . Hence ρ is a rational representation of H.
Proposition 15.12 Let V1 , V2 be highest weight modules of highest weights
µ1 , respectively µ2 . Then V1 ⊗ V2 has an irreducible subquotient of highest
weight µ1 + µ2 .
Proof Let vi be maximal vectors for Vi , i = 1, 2. Then a straightforward
calculation shows that v1 ⊗ v2 is a maximal vector of V1 ⊗ V2 of weight
µ1 + µ2 . Hence, by Proposition 15.6, the kG-submodule G(v1 ⊗ v2 ) has an
irreducible quotient of highest weight µ1 + µ2 .
We now show existence and uniqueness of irreducible highest weight mod-
ules of a given dominant weight. While uniqueness is rather straightforward,
the existence statement is more difficult to prove; it is solved by explicitly
constructing, for semisimple groups G, a kG-module with a maximal vector
of weight λ within the infinite-dimensional vector space k[G].
Assume that G is semisimple, V an irreducible highest weight module for
G of highest weight λ, and v + ∈ V a maximal vector. Let V  denote the
sum of the weight spaces for weights different from λ. Then V = v +  ⊕ V 
by Corollary 15.10. Let ϕ+ ∈ V ∗ := Hom(V, k) be defined by ϕ+ (v + ) = 1,
ϕ+ |V  = 0, and set
fλ : G → k, x → fλ (x) := ϕ+ (x.v + ).
Note that fλ ∈ k[G] since G acts rationally on V and ϕ+ is a morphism.
Lemma 15.13 The function fλ satisfies fλ (ub) = λ(b) for all u ∈ U − ,
b ∈ B, where λ is considered as a character of B via the natural epimorphism
B → T.
Proof By the definition of ϕ+ we have
fλ (ub) = ϕ+ (ub.v + ) = λ(b) ϕ+ (u.v + ) = λ(b) ϕ+ (v + + v  ) = λ(b)
for some v  ∈ V  , as claimed.
Let λ ∈ X(T ) be a weight of a rational representation ρ : G → GL(V ).
Then as in Example 15.5, sα .λ − λ is an integral multiple of α. With respect
to a W -invariant non-degenerate bilinear form ( , ) on the Euclidean space
(λ,α)
E = X(T )⊗Z R, sα acts as the reflection in α. So we have sα .λ = λ−2 (α,α) α
which shows that the number 2(λ, α)/(α, α) is an integer for all α ∈ Φ. Now
any element λ ∈ E with this property is called an abstract (integral) weight
in the sense of root systems. Since Φ spans E by Proposition 9.2, the set of
 15.2 Irreducible highest weight modules 127

abstract weights is a lattice in E. From now on we will identify this lattice

with Ω = Hom(ZΦ∨ , Z) via λ → (γ ∨ → λ, γ ∨  = 2 (λ,γ)
(γ,γ) ), for all abstract
weights λ ∈ E and for all γ ∨ ∈ ZΦ∨ . As explained in Section 9.2, X(T ) can
be viewed as a sublattice of Ω of finite index.
The notion of dominant weight and the partial order ≤ from Defini-
tion 15.7 extend to Ω in an obvious way.
Definition 15.14 For G semisimple with base ∆ = {α1 , . . . , αl } we let
{λ1 , . . . , λl } ⊆ Ω denote the dual basis of ∆∨ = {α1∨ , . . . , αl∨ } with respect
to the perfect pairing  ,  : X(T ) × Y (T ) → Z from Proposition 3.6 (that
is, λi , αj∨  = δij ). The λi are called the fundamental dominant weights of T
with respect to ∆. (The notation i is also used in the literature.)
l
Thus λ ∈ Ω is dominant if and only if λ = i=1 ai λi with ai ∈ N0 .
We start by constructing highest weight modules for multiples of the fun-
damental dominant weights:
Lemma 15.15 Let G be semisimple, λm a fundamental dominant weight.
Then there exists an irreducible highest weight module for G of highest weight
am λm for some am ∈ N.
Proof Let P ≤ G denote the standard parabolic subgroup of G correspond-
ing to the subset ∆\{αm }, see Proposition 12.2. By Theorem 5.5 there exists
a rational representation ρ : G → GL(V ) with v ∈ V such that P is the sta-
bilizer of the line v. Since P contains the Borel subgroup B, v is a maximal

vector of V , say with dominant weight λ = i ai λi , ai ≥ 0. But P also con-
tains representatives s˙i of the simple reflections si ∈ S, i = m, which thus
also stabilize v, hence fix λ by Lemma 15.3. But
si (λi ) = λi − αi , si (λj ) = λj − λj , αi∨ αi = λj for j = i,
so we have ai = 0 for i = m, whence λ = am λm . The kG-submodule of V
generated by v now has highest weight λ, and by Proposition 15.6 it has an
irreducible quotient of the same highest weight.
To continue, we need a property of the field of fractions k(G) := Frac(k[G])
of the coordinate ring k[G]. Note that k[G] is an integral domain by Propo-
sition 1.9, so it embeds into k(G):
Proposition 15.16 Let G be a connected linear algebraic group. Then k[G]
is integrally closed in k(G).
See for example [73, Lemma 70] for the proof.
We are now ready to prove the following existence and uniqueness result
(see [15, Exposés 15,16]).
128 Representations of algebraic groups

Theorem 15.17 (Chevalley) Let G be connected reductive.

(a) Two irreducible kG-modules V1 , V2 of highest weights µ1 , µ2 , are iso-
morphic if and only if µ1 = µ2 .
(b) For each dominant weight λ ∈ X(T ), there exists an irreducible kG-
module with highest weight λ.
Proof (a) It is clear that an isomorphism of kG-modules V1 → V2 will send
maximal vectors to maximal vectors of the same weight. Hence, V1 ∼ = V2
implies that µ1 = µ2 . Now assume that V1 , V2 both have highest weight λ.
Let vi be a maximal vector of Vi , of weight λ, for i = 1, 2. Then v := v1 ⊕ v2
is a maximal vector of the kG-module V := V1 ⊕ V2 , also of weight λ. By
Proposition 15.6, λ has multiplicity 1 in V0 := Gv. As the vi are not
proportional to v, they cannot lie in V0 . So Vi ∩ V0 is a proper submodule of
Vi , hence 0 by irreducibility, for i = 1, 2. Thus the projection V → Vi maps
V0 onto a non-zero submodule of Vi , so onto all of Vi . But then V1 ∼
= V0 ∼= V2
as claimed.

For (b), first assume that G is semisimple. Let λ = α∈∆ cα λα ∈ X(T ) be
dominant, with λα the fundamental dominant weight corresponding to the
simple root α, so cα ≥ 0. Let aα > 0 such that there exists a highest weight

module Vα of highest weight aα λα , as in Lemma 15.15, and set a := α aα .
Now µ := aλ is a non-negative integral linear combination of the aα λα , so
- ⊗cα dα
Proposition 15.12 and an easy induction show that α∈∆ Vα has an
irreducible subquotient of highest weight µ, where dα = a/aα . Let fµ ∈
k[G] be the corresponding function constructed in the discussion preceding
Lemma 15.13.
Recall the dense open subset O := U − B of G from Theorem 11.20. We
define f : O → k by f (xz) := λ(z) for x ∈ U − , z ∈ B. Then f ∈ OG (O),
so f ∈ Frac(OG (O)) = k(G), as pointed out at the beginning of Section 5.1.
Since µ = aλ we then have fµ = f a . But k[G] is integrally closed in k(G)
by Proposition 15.16, so f ∈ k[G]. By Proposition 2.4, f lies in a finite-
dimensional subspace of k[G] which is invariant under the action of G by right
multiplication. We claim that f generates a submodule of k[G] of highest
weight λ. Indeed, for xy ∈ U − B, z ∈ B we have
ρz (f )(xy) = f (xyz) = λ(yz) = λ(z)λ(y) = λ(z)f (xy),
so ρz (f ), λ(z)f ∈ k[G] agree on the dense open set O, hence on all of G.
Thus f is a maximal vector of weight λ. We obtain an irreducible quotient
of highest weight λ from Proposition 15.6.
Now for G connected reductive, let G := [G, G], Z := Z(G)◦ , so G = ZG
by Corollary 8.22. If λ ∈ X(T ) is dominant, then λ := λ|T  ∈ X(T  ), with
 15.2 Irreducible highest weight modules 129

T  := T ∩ G a maximal torus of G , is again dominant, so by the previous

step there exists an irreducible highest weight module V := L(λ ) for G .
Note that elements z of the central subgroup G ∩ Z ≤ Z(G ) act by scalars
on the irreducible module V , hence necessarily by λ(z). Thus we may define
an action of Z on V by z.v := λ(z)v for z ∈ Z, v ∈ V , to obtain a kG-module
structure on V . Clearly a maximal vector of V with respect to G is also a
maximal vector with respect to G, of weight λ.
Definition 15.18 The irreducible kG-module with highest weight λ con-
structed in Theorem 15.17 is denoted L(λ). (The notation V (λ) is also used,
for example in [32].)
Example 15.19 Let G be a simple algebraic group, B ≤ G a Borel sub-
group with maximal torus T , Φ+ the corresponding set of positive roots and
α0 ∈ Φ+ the highest root (from Proposition 13.10). Let Ad be the adjoint
representation of G. By definition the weights of T acting on Lie(G) are
given by 0 and the roots of G. Now by Theorem 11.8, [Uβ , Uα0 ] = 1 for all
β ∈ Φ+ . This implies that Ad (Uβ ) fixes Lie(Uα0 ) = gα0 pointwise, whence
any v0 ∈ gα0 \ {0} is a maximal vector for Ad (B) of weight α0 .
Thus, if M is the unique maximal kG-submodule of Gv0 , then Gv0 /M
is an irreducible kG-module of highest weight α0 . In fact, in most cases, the
submodule Gv0  is already an irreducible kG-module (see [31] or [60, 1.9];
the case of SLn is Exercise 20.17):
Theorem 15.20 Let G be a simple algebraic group over an algebraically
closed field k of characteristic p. Then Lie(G) is an irreducible module under
the adjoint representation of G, except in the following cases, where 0 = M ⊂
Lie(G) is a proper kG-submodule of Lie(G):
(1) M ⊆ Z(Lie(G)) and the root system of G has type An , respectively Dn ,
E6 , E7 and p | n + 1, respectively p = 2, p = 3, p = 2.
(2) The root system of G has type Bn , Cn , F4 , G2 and p = 2, 2, 2, 3 respec-
tively and M contains all root subspaces belonging to short roots in Φ.
In the cases where Lie(G) is reducible, one has precise information about
the highest weights of the remaining composition factors.
Example 15.21 (1) Let G = G2 (see Example 9.5 for its root system).
The highest root is 3α + 2β, with 3α + 2β, α∨  = 0, 3α + 2β, β ∨  = 1.
So 3α + 2β is in fact one of the fundamental dominant weights of G2 .
When p = 3, Lie(G) has a seven-dimensional ideal containing all root
vectors corresponding to short roots. The highest such (with respect to
the partial order on weights) is the highest short root 2α + β; here we
130 Representations of algebraic groups

have 2α +β, α∨  = 1 and 2α +β, β ∨  = 0. So Lie(G) has a composition

factor with the second fundamental dominant weight as highest weight.
These are in fact the only composition factors of Lie(G) in this case.
(2) Let G = SL2 acting on its natural module V with fixed basis {x, y}. Let
T = D2 ∩ SL2 , B = T2 ∩ SL2 and Φ = {±α}. Then x is invariant under
the Borel subgroup B and so x is a maximal vector of V with respect to
B. Now X(T ) = Zλ, where λ(diag(c, c−1 )) = c. As
diag(c, c−1 )x = cx = λ(diag(c, c−1 ))x,
x is of weight λ. Moreover,
diag(c, c−1 )y = c−1 y = λ(diag(c, c−1 ))−1 y
shows that the other weight of V is −λ. As α(diag(c, c−1 )) = c2 we have
α = 2λ, so −λ = λ − α.
More generally, it is straightforward to show that the natural module
for SLn is irreducible with highest weight λ1 , when numbering the base
of Φ such that αi (diag(t1 , t2 , . . . , tn )) = ti t−1
i+1 . The remaining weights
of the representation are λ1 − α1 − · · · − αj for 1 ≤ j ≤ n − 1. (See
Example 9.8; we have taken αi = χi,i+1 .)
Over fields of characteristic 0, much more is known about the irreducible
representations of semisimple groups. For example, there is the Weyl degree
formula giving the dimension of an irreducible module of a given highest
weight, and formulas which allow one to calculate the dimensions of individ-
ual weight spaces (Freudenthal’s formula and Kostant’s formula). These are
precisely those coming from the representation theory of semisimple complex
Lie algebras. See [33, §24] for precise statements.
When char(k) is positive and large with respect to the weight λ (in some
precise sense), these same formulas remain valid. Even when char(k) is small,
they give upper bounds for the dimensions of irreducible modules and their
weight spaces.
Finally, we mention the Lusztig conjecture which predicts the dimensions
of irreducible modules whose highest weights satisfy certain geometric con-
ditions, and assuming char(k) is not too small relative to Φ, in terms of
the Kazhdan–Lusztig polynomials of the affine Weyl group associated to G.
Andersen, Jantzen and Soergel have shown that for each root system Φ,
there exists an (unknown) bound n(Φ), such that the conjecture holds for
all groups with this root system if char(k) > n(Φ). We refer the reader to
[36, §3] for an expository account of this theory.
 16
Representation theory and maximal
subgroups

In the previous chapter, we described the parametrization via highest weights

of the irreducible representations of a semisimple linear algebraic group.
There are many further questions which can be raised, for example, on exten-
sions of simple modules and on the validity of dimension formulas in positive
characteristic. In this section, we consider three aspects of the representation
theory of a semisimple algebraic group G which play a particularly important
role in the study of the maximal subgroups of the classical algebraic groups,
namely, a criterion for the existence of a G-invariant bilinear form, the re-
striction of an irreducible representation to a Levi subgroup and Steinberg’s
tensor-product decomposition of certain irreducible highest weight modules.

16.1 Dual modules and restrictions to Levi subgroups

A faithful rational representation ρ : G → SLn identifies G with a subgroup
of SLn . When is this subgroup maximal? Certainly, if ρ is not an irreducible
representation, then G is contained in the stabilizer of a proper non-trivial
subspace and so if G is maximal it must be the full stabilizer. Now suppose
ρ is irreducible and that G stabilizes some non-degenerate form on k n , and
thus lies in the isometry group of this form. Then if G is maximal it must be
the full isometry group of the form. So we are led to ask when G stabilizes
a non-degenerate form on a given irreducible representation. Now G stabi-
lizes a non-degenerate form if and only if the corresponding kG-module is
isomorphic to its dual. When this is the case, the corresponding representa-
tion ρ : G → SL(V ) has image lying in one of the classical groups Sp(V ) or
SO(V ); in particular if ρ(G) = Sp(V ), SO(V ) then ρ(G) is not a maximal
subgroup of SL(V ). We will see in Chapter 18 that in the large majority of
132 Representation theory and maximal subgroups

cases, ρ(G) is a maximal subgroup of the smallest classical group containing

it.
Note that for V a kG-module, the dual V ∗ := Hom(V, k), with the natural
induced G-action

(gf )(v) := f (g −1 v) for g ∈ G, f ∈ Hom(V, k), v ∈ V,

is again a rational kG-module, irreducible if V is irreducible.

Fix a Borel subgroup B of G and a maximal torus T of B. Let Φ+ be the
corresponding set of positive roots. We consider the case where V (and hence
V ∗ ) is an irreducible kG-module, so V = L(λ) for some dominant λ ∈ X(T )
(see Theorem 15.17). In order to decide when L(λ)∗ is isomorphic to L(λ),
we will determine its highest weight and apply Theorem 15.17(a).

Proposition 16.1 Let G be a connected reductive algebraic group and let

λ ∈ X(T ) be dominant. Then L(λ)∗ ∼
= L(−w0 (λ)), where w0 ∈ W is the
longest element.

Proof We have w0 (Φ+ ) = Φ− by Corollary 11.18, and if ẇ0 ∈ NG (T ) is

a preimage of w0 , then ẇ0 B ẇ0−1 = B − = T, Uα | α ∈ Φ− . Let v + be
a maximal vector (with respect to B) of the irreducible kG-module L(λ).
Write V = v +  ⊕ V  , with V  the sum of the weight spaces for weights other
than λ, and let f ∈ L(λ)∗ denote the linear form with f (v + ) = 1, f |V  = 0,
as in Section 15.2. We claim that f is a maximal vector of L(λ)∗ for B − of
weight −λ. Indeed, if g ∈ B − then g = tu for some t ∈ T , u ∈ Uα | α ∈ Φ− 
by Theorem 11.1, and hence
(gf )(v + ) = f (g −1 v + ) = f (u−1 t−1 v + ) = f (u−1 λ(t−1 )v + )
= λ(t−1 )f (u−1 v + ) = λ(t−1 )f (v + + v  ) = λ(t−1 )f (v + )
for some v  ∈ V  by Lemma 15.4. Furthermore, for v ∈ V  , (gf )(v) =
f (g −1 v) = f (u−1 t−1 v) = 0 = f (v) as u−1 t−1 v is a sum of weight vectors of
weight strictly less than λ. So gf = λ(t−1 )f , i.e., f is a maximal vector for
B − of weight −λ.
But then ẇ0−1 f is a maximal vector for B:

B ẇ0−1 f = ẇ0−1 B − ẇ0 (ẇ0−1 f ) = ẇ0−1 B − f ∈ ẇ0−1 f .

Moreover, ẇ0−1 f is of weight w0−1 (−λ) = −w0 (λ). The result now follows
from Corollary 15.10.

Remark 16.2 In view of the previous result, it is interesting to know for

which Weyl groups −w0 = id, as in this case all irreducible kG-modules are
self-dual.
 16.1 Dual modules and restrictions to Levi subgroups 133

It is known that w0 = −id (on X(T )) when Φ is of type A1 , Bn , Cn ,

Dn (n even), E7 , E8 , F4 , G2 , see Corollary B.23. (These are precisely the
irreducible Weyl groups which contain −id.) In the remaining irreducible
cases, −w0 induces a non-trivial graph automorphism of the Dynkin diagram
as indicated below:

An : Dn :

E6 : 1 3 4 5 6
2

= L(λ)∗ ∼
For example, for type E6 and λ = λ1 + λ6 , L(λ) ∼ = L(−w0 (λ)).

For the purposes of studying embeddings of reductive groups H < G, in

particular when G is a classical group and H acts irreducibly on the natural
module for G, it will also be useful to know something about the restriction
of irreducible representations to Levi complements of parabolic subgroups.
The result we state will enable us to apply inductive arguments in the study
of irreducible representations.
Let ∆ denote the base of Φ with respect to B. For some subset I ⊆ S =
{sα | α ∈ ∆}, let P denote the corresponding parabolic subgroup of G
with Levi decomposition P = QL, where Q = Ru (P ) and L = LI , as in
Section 12.2. We consider the restriction of an irreducible representation of
G to the Levi subgroup L.

Proposition 16.3 Let G be connected reductive, λ ∈ X(T ) a dominant

weight, and write L(λ)Q for the fixed point space of Q on the irreducible
kG-module L(λ). Then:

(a) L(λ)Q = ⊕γ∈N0 ∆I L(λ)λ−γ .

(b) L(λ)Q is the irreducible kL-module with highest weight λ.

Proof Recall from Proposition 12.6 that Q = Uα | α ∈ Φ+ \ ΦI . Set

V = L(λ)Q . For γ ∈ Z∆I and α ∈ Φ+ \ ΦI , Corollary 15.10 implies that
λ − γ + α is not a weight of the module L(λ) and hence by Lemma 15.4, Uα
fixes pointwise the weight space L(λ)λ−γ . Thus, L(λ)λ−γ ⊆ V .
Now L normalizes Q and hence stabilizes the subspace V ; in particu-
lar T stabilizes V and so V is a sum of T -weight spaces. For each weight
134 Representation theory and maximal subgroups

µ ∈ X(T ) with Vµ = 0, ⊕γ∈Z∆I Vµ+γ is a non-trivial L-invariant submod-

ule of V (again by Corollary 15.10). Let U = Ru (B). By the Lie–Kolchin
Theorem 4.1, the unipotent subgroup L ∩ U must have a non-trivial fixed
point v on ⊕γ∈Z∆I Vµ+γ . But then v is fixed by the whole of U = (L ∩ U )Q
(see Proposition 12.6), hence a maximal vector, of weight λ. That is, for all
µ with Vµ = 0, there exists γ ∈ Z∆I with µ + γ = λ, so µ = λ − γ. Thus
V ⊆ ⊕γ∈Z∆I L(λ)λ−γ and (a) follows, using Corollary 15.10.
The above argument shows that V = ⊕γ∈N0 ∆I L(λ)λ−γ has a unique 1-
space of fixed points under the action of L ∩ U , namely the space L(λ)λ .
But L ∩ U is the unipotent radical of a Borel subgroup of L. Thus the sum
of all simple kL-submodules of V , that is the socle of the restriction V |L ,
is irreducible. As in the proof of Corollary 15.10, we see that the socle is
the kL-submodule generated by any v + ∈ Vλ \ {0}: soc(V |L ) = Lv + . Now
as L(λ) is an irreducible kG-module, we have as in the proof of Propo-
sition 15.6 Gv +  = U − T v +  = L(λ), where U − = w˙0 U w˙0 −1 with w0
the longest element of the Weyl group. Writing U − = Q− (U − ∩ L), where
Q− = Uα | α ∈ Φ− \ ΦI , we see that for all γ ∈ N0 ∆I , the weight
space L(λ)λ−γ is contained in (U − ∩ L)v +  ⊆ Lv +  = soc(V |L ). Hence
V = ⊕γ∈N0 ∆I L(λ)λ−γ ⊆ soc(V |L ), which is irreducible, establishing (b).

Example 16.4 Let G be semisimple, with simple roots ∆ and fundamental

dominant weights {λα | α ∈ ∆}. For I ⊂ S, let ∆I be the corresponding
base of ΦI . Then {λα | α ∈ ∆I } is the dual basis of ∆∨ I in ZΦI ⊗Z R, so
these are the fundamental dominant weights of T  := T ∩ LI with respect

to ∆I , where LI = [LI , LI ]. Thus if λ = α∈∆ aα λα ∈ X(T ) is dominant,

then its restriction to T  equals λ|T  = α∈∆I aα λα . So the highest weight

of L(λ)Q as a kLI -module in Proposition 16.3 is just α∈∆I aα λα .

16.2 Steinberg’s tensor product theorem

Let ρ : G → GL(V ) be a rational representation which embeds G as a
maximal subgroup into GL(V ). If G leaves invariant a tensor product de-
composition V = V1 ⊗ V2 , then it lies in the stabilizer of this decomposition,
so is equal to it by maximality. Thus, it is important for the study of maxi-
mal subgroups to know which irreducible representations are tensor decom-
posable. A partial answer in positive characteristic is given by Steinberg’s
tensor product theorem, which is a major result in the representation theory
of semisimple groups.
Let G be a semisimple algebraic group with root system Φ with respect to
 16.2 Steinberg’s tensor product theorem 135

the maximal torus T ≤ G. As already mentioned in Section 11.1 Steinberg

has given a presentation of G in terms of the root subgroups Uα = {uα (t) |
t ∈ k}, α ∈ Φ; a description can be found in [73, §6]. Let τ be an auto-
morphism of the field k. Then it is immediate from this so-called Steinberg
presentation that it induces an automorphism of the abstract group G such
that uα (t) → uα (t)τ := uα (τ (t)), for all α ∈ Φ, t ∈ k.
For the classical matrix groups as defined in Section 1.2, such an auto-
morphism can be obtained by componentwise application of τ to all matrix
entries, as can be seen from an explicit determination of the root subgroups
with respect to the maximal torus consisting of diagonal matrices, see Ex-
amples 11.4 and 11.7.
Here, we will be interested in the case of positive characteristic only. So
throughout this section k will be an algebraically closed field of characteristic
char(k) = p > 0. Then we obtain the following special case:
Theorem 16.5 (Chevalley, Steinberg) Let G be a semisimple algebraic
group over k with char(k) = p, with root system Φ with respect to a maximal
torus T . Then the Frobenius automorphism Fp : k → k, c → cp , of k induces
an endomorphism F : G → G of algebraic groups by
uα (c) → F (uα (c)) := uα (Fp (c)) = uα (cp ) for all α ∈ Φ, c ∈ k.
Moreover, F (t) = tp for all t ∈ T .
For the classical groups constructed in Section 1.2 the desired map can
be obtained by raising all matrix entries to their pth power. In general, the
claim follows from the form of the relations in the Steinberg presentation
in [73, p.66], see [66, Thm. 9.4.3]. Such endomorphisms play a crucial role
in the construction of the finite groups of Lie type and will be discussed in
more detail in Chapter 3.
Now given a rational representation ρ : G → GL(V ) and i ≥ 1, we can
i
define a new rational representation on V by ρ(p ) := ρ ◦ F i , with F from
i
Theorem 16.5. The corresponding kG-module is denoted V (p ) , the action
being given by
i
ρ(p ) (g)v = ρ(F i (g))v for g ∈ G, v ∈ V .
The behavior of highest weights under this construction is easily described:
Proposition 16.6 Let G be semisimple, λ ∈ X(T ) a dominant weight.
Then L(λ)(p) ∼
= L(pλ) as kG-modules.
Proof Since B is F -invariant, a maximal vector v of L(λ) is also maximal
for L(λ)(p) . Moreover, F (t).v = tp .v = λ(t)p .v for t ∈ T by Theorem 16.5,
136 Representation theory and maximal subgroups

so v is of weight pλ. Clearly, if V is an irreducible kG-module, then so is

V (p) . Thus, L(λ)(p) is the irreducible highest weight module of highest weight
pλ.

We now relate the representations of G to those of its Lie algebra g :=

Lie(G) as follows. If ρ : G → GL(V ) is a rational representation, then its dif-
ferential dρ : g → gl(V ) is a Lie algebra representation of g, i.e., a Lie algebra
homomorphism to gl(V ). The irreducible constituents of g-representations
obtained in this way are called restricted .
Note that g acts trivially on V (p) : indeed, we have ρ(p) |T = ρ ◦ F , where
F : T → T , t → tp , is the p-power map by Theorem 16.5. Thus dρ(p) |Lie(T ) =
dρ ◦ dF |Lie(T ) , but clearly dF = 0, so Lie(T ) lies in the kernel of dρ(p) .
Similarly, since the action of F on the root group Uα , for α ∈ Φ, is simply the
composition of a fixed isomorphism Ga → Uα with the p-power map on Ga ,
we see that Lie(Uα ) also lies in ker dρ(p) , for all α ∈ Φ. Then Theorem 8.17(a)
shows that dρ(p) = 0.
Clearly, if V1 ≤ V is a G-invariant subspace, then it is also g-invariant,
hence if dρ is irreducible, then so was ρ. The converse does not hold in
general. But there is the following crucial lifting result:

Theorem 16.7 (Cline–Parshall–Scott) Let G be semisimple of simply con-

nected type. Then every irreducible restricted g-module has a (unique) exten-
sion to a rational irreducible kG-module.

Here ρ : G → GL(V ) is an extension of ϕ : g → gl(V ) if dρ = ϕ. The proof

of this result is not very difficult, see [16, Thm. 1], but it needs the notion
of restricted universal enveloping algebra, which we did not introduce. This
allows one to construct a first tensor decomposition:

Lemma 16.8 Let V be an irreducible rational kG-module, V1 ≤ V an

irreducible g-submodule. Then V ∼
= V1 ⊗ V2 as kG-modules, with V1 , V2 irre-
ducible as kG-modules and with trivial action of g on V2 .

Proof By Theorem 16.7, V1 has a structure of irreducible kG-module. Then

the space V2 := Homg (V1 , V ) of g-invariant homomorphisms from V1 to
V can be made into a kG-module by (g.ϕ)(v) := g.ϕ(g −1 v) for ϕ ∈ V2 ,
g ∈ G and v ∈ V1 . The k-linear map V1 ⊗ V2 → V , (v, ϕ) → ϕ(v), is then
a kG-homomorphism, non-zero by assumption and thus surjective by the
irreducibility of V . As dim V2 ≤ dim V / dim V1 we see that in fact it must be
an isomorphism.
Now clearly, if V2 has a G-invariant proper non-zero subspace, then so has
the tensor product V1 ⊗ V2 , which is not the case, thus V2 is irreducible for
 16.2 Steinberg’s tensor product theorem 137

G. Differentiating the action of G on V2 shows that X ∈ g acts on V2 by

(X.ϕ)(v) = Xϕ(v) − ϕ(Xv). But V2 consists of g-homomorphisms, so it is a
trivial g-module.

To continue, we need one further piece of notation. Let Φ be the root

system relative to T with positive system Φ+ and base ∆ ⊂ Φ determined
by B.

Definition 16.9 A dominant weight λ ∈ X(T ) is said to be p-restricted if

λ, α∨  < p for all α ∈ ∆. An irreducible kG-module with highest weight λ
is called restricted if λ is p-restricted, and the corresponding representation
is called a restricted representation.

Note that there are only finitely many p-restricted weights for a fixed
semisimple group G and a fixed prime p.

Lemma 16.10 Let G be semisimple of simply connected type with maximal

torus T . If λ ∈ X(T ) is p-restricted and dλ = 0 then λ = 0.

Proof Let’s write T = Gm × · · · × Gm (l factors). Clearly the differentials of

the coordinate functions form a basis of Hom(Lie(T ), k) = Hom(k l , k). Since
G is simply connected, we have Ω = X(T ), so the fundamental dominant
weights {λi } form another Z-basis of X(T ), whence their differentials also
form a basis of Hom(Lie(T ), k).
 
Now write λ = i ai λi , so 0 = dλ = i ai dλi by the formula for the
differential of multiplication in Example 7.8(1), which shows that ai ≡ 0
(mod p) for all i by the above. As λ is p-restricted, this implies that ai = 0
for all i.

The following result shows that restricted kG-modules behave particularly

well with respect to the differential:

Corollary 16.11 (Curtis) Let λ ∈ X(T ) be p-restricted. Then L(λ) is an

irreducible g-module.

Proof Let V1 be an irreducible g-submodule of L(λ). By Lemma 16.8 there

is a tensor decomposition of L(λ) into irreducible kG-modules, of highest
weights λ , µ say: L(λ) ∼
= L(λ ) ⊗ L(µ) with V1 ∼
= L(λ ). But then λ = λ + µ
by Proposition 15.12, so µ is necessarily also p-restricted. Now Lie(T ) acts
by dµ on a maximal vector of V2 = L(µ), but g acts trivially by Lemma 16.8,
so dµ = 0. By Lemma 16.10 this forces µ = 0, and so L(λ) ∼ = L(λ ) ∼
= V1 is
an irreducible g-module.
138 Representation theory and maximal subgroups

Now each dominant weight λ ∈ X(T ) can be uniquely expressed as

λ = µ0 + pµ1 + · · · + pa µa ,

where a ∈ N0 and µi ∈ X(T ) is a p-restricted (dominant) weight for each

0 ≤ i ≤ a. The main result of this section shows that there is an analogous
decomposition of the corresponding irreducible kG-module L(λ):

Theorem 16.12 (Steinberg) Let G be semisimple of simply connected type,

let λ ∈ X(T ) be dominant, and write λ = µ0 + pµ1 + · · · + pa µa , with a ∈ N0
and p-restricted weights µi ∈ X(T ) for each 0 ≤ i ≤ a. Then

L(λ) ∼
a
= L(µ0 ) ⊗ L(µ1 )(p) ⊗ · · · ⊗ L(µa )(p )

as kG-modules.

Proof Write λ = µ0 + pµ (so µ = µ1 + · · · + pa−1 µa ). Then L(λ) is a

kG-composition factor of V := L(µ0 ) ⊗ L(µ )(p) ∼ = L(µ0 ) ⊗ L(pµ ) by Propo-
sitions 15.12 and 16.6. As g acts trivially on L(pµ ), all g-composition factors
of V are isomorphic to the irreducible g-module L(µ0 ) (see Corollary 16.11),
so L(µ0 ) is an irreducible g-submodule of L(λ). By Lemma 16.8 we have
L(λ) ∼= L(µ0 ) ⊗ L(ν) for some dominant weight ν, so λ = µ0 + ν. Uniqueness
of the highest weight shows that ν = pµ , so L(λ) ∼= L(µ0 )⊗L(pµ ) ∼ = L(µ0 )⊗
L(µ )(p) . Now an easy induction, applied to µ , completes the proof.

Thus, in positive characteristic, the investigation of all rational irreducible

kG-modules is reduced to the study of the finitely many restricted ones.

Example 16.13 We use the tensor product theorem to illustrate that

dimensions of irreducible highest weight modules may depend on the under-
lying characteristic.

(1) Let G = SLn and let X be the natural n-dimensional module for G, of
highest weight λ1 (as mentioned in Example 15.21(2)). Then by Propo-
sition 15.12 the highest weight in V = X ⊗ X is 2λ1 , occurring with
multiplicity 1. In particular V possesses a kG-composition factor with
highest weight 2λ1 . If p > 2, this weight is p-restricted and it can be
shown that the corresponding highest weight module is precisely the
symmetric square S 2 (X) of X and V = S 2 (X) ⊕ ∧2 (X). If however
p = 2, the irreducible module with highest weight 2λ1 is simply the
“twist” of the natural module: L(2λ1 ) ∼= L(λ1 )(2) = X (2) by Proposi-
tion 16.6. Nevertheless, ∧ (X) is an irreducible kG-module and V has
2

three composition factors: X (2) and two factors isomorphic to ∧2 (X).

 16.2 Steinberg’s tensor product theorem 139

(2) Consider the kSL3 -module with highest weight λ = 2λ1 +3λ2 , where λi is
the fundamental dominant weight corresponding to the simple root αi . If
p > 3, then λ is p-restricted and dim L(λ) = 39 if p = 5 and dim L(λ) =
42 if p = 5. (See [54].) If however p = 3, the p-adic expansion of λ is
λ = µ0 + 3µ1 where µ0 = 2λ1 and µ1 = λ2 , so L(λ) ∼ = L(2λ1 ) ⊗ L(λ2 )(3)
in this case by Theorem 16.12. Thus, by the previous example and by
Proposition 16.1 and Remark 16.2, L(λ) ∼ = S 2 (X)⊗X ∗ (3) , where X is the
natural three-dimensional module for SL3 . Hence, dim L(λ) = 6 · 3 = 18.
If p = 2, then Theorem 16.12 gives L(λ) ∼ = L(λ2 ) ⊗ L(λ1 + λ2 )(2) . Using
Example 15.19, one can show that L(λ1 + λ2 ) is isomorphic to the Lie
algebra of SL3 , and so dim L(λ) = 3 · 8 = 24.
 17
Structure of parabolic subgroups, II

We continue the investigation of the structure of a parabolic subgroup by

now looking at its unipotent radical. We will see that the conjugation action
of a Levi factor on the unipotent radical induces a family of representations
of the Levi factor. Now that we know something about the representation
theory of reductive groups, we will use this to obtain more information about
the unipotent radical.

17.1 Internal modules

Let G be a connected reductive group with maximal torus T contained in
a Borel subgroup B. Let Φ be the root system of G with respect to T , and
∆ ⊆ Φ the base determined by B. Let Φ+ be the corresponding set of positive
roots, with set of simple reflections S. So in particular, B = T, Uα | α ∈ Φ+ 
by Theorem 11.1(a).
For a subset I ⊆ S, recall the parabolic subsystem ΦI of Φ. Let P = BWI B
be the corresponding standard parabolic subgroup with unipotent radical
Q := Ru (P ) and Levi decomposition
P = QL = Uα | α ∈ Φ+ \ ΦI  · T, Uβ | β ∈ ΦI 
from Proposition 12.6. Note that L is reductive by Proposition 12.6, so [L, L]
is semisimple by Corollary 8.22.
Definition 17.1 Let I ⊆ S with corresponding set of simple roots ∆I . For
β ∈ Φ+ , write
" "
β= cα α + dγ γ.
α∈∆I γ∈∆\∆I

We say that β is of level dγ , and we call (dγ )γ∈∆\∆I the shape of β. For
 17.1 Internal modules 141

j ≥ 1, set
Qj = Uβ | β ∈ Φ+ of level ≥ j ≤ Q.
Example 17.2 Let G = SLn , the simply connected simple group of type
An−1 , with root system base {α1 , . . . , αn−1 } and corresponding set of posi-
tive roots Φ+ (see Example 9.8).
Let P be the standard parabolic subgroup corresponding to the subset
I = {sα1 } of S with unipotent radical
.  "
n−1 "
n−1 /

Q = Uβ  β = ai αi ∈ Φ+ , ai = 0
i=1 i=2

and Levi complement L = U±α1 , T . Then the roots of level i for 1 ≤ i ≤

n − 2 are αj + · · · + αj+i−1 for 2 ≤ j ≤ n − i and α1 + α2 + · · · + αi+1 .
We obtain the following filtration of unipotent radicals:
Proposition 17.3 Let P ≤ G be a parabolic subgroup of the connected
reductive group G, with unipotent radical Q. Then with the notation as in
the definition above, we have for all j ≥ 1:
(a) Qj  P .

(b) Qj /Qj+1 ∼
= β Ga , where the product runs over β ∈ Φ+ of level j.
Proof (a) Recall the commutator formula from Theorem 11.8:
!
[Uα , Uβ ] ⊆ Umα+nβ for all roots α = ±β.
m,n>0

Let β ∈ Φ of level at least j and α ∈ ΦI ∪ Φ+ . Then in particular, for

+

x ∈ Uα , xUβ x−1 lies in the product of root subgroups for roots of level at
least j. Since Qj is a product of root subgroups for T , Qj is also normalized
by T and (a) follows.
For (b), let’s choose a total order on the positive roots so that ht(α) >
ht(β) implies that α > β. For each α ∈ Φ, fix an isomorphism uα : Ga →
Uα . Then by Theorem 11.1(a) any element u of Qj may be written as a

product β uβ (cβ )y over β ∈ Φ+ of level j, with y ∈ Qj+1 and the cβ
uniquely determined by u. Thus, as Qj /Qj+1 is abelian by the commutator
 
formula, the map (cβ )β → β uβ (cβ )Qj+1 defines an isomorphism β Ga ∼ =
Qj /Qj+1 , with β as before.
As Qj  P for all j, L acts on the quotient Qj /Qj+1 . It is immediate to
check that the rule cuβ (t)Qj+1 := uβ (ct)Qj+1 , for β ∈ Φ+ \ ΦI of level j and
c ∈ k, defines a k-vector space structure on Qj /Qj+1 which commutes with
the action of L.
142 Structure of parabolic subgroups, II

Now for a shape S of level j let VS := ( β of shape S Uβ )Qj+1 /Qj+1 . From
the commutator relations it is then easy to verify:

Proposition 17.4 Let S be a shape of level j ≥ 1.

(a) As k-vector space, VS is isomorphic to the direct sum of the Uβ , for β

of shape S.
(b) VS is normalized by L, hence a kL-submodule of Qj /Qj+1 .

The subgroups Qj are closely related to the descending central series of Q.

Recall (from Section 1.2) that C i−1 Q denotes the ith term of the descending
central series for Q. Then one has [3, Lemma 4]:

Proposition 17.5 Let G be a simple algebraic group and assume that

(G, char(k)) = (Bn , 2), (Cn , 2), (F4 , 2), (G2 , 2), (G2 , 3). Then for any para-
bolic subgroup P = QL of G we have Qj = C j−1 Q for all j ≥ 1.

About the proof Essentially this follows easily from an inductive argument
and the commutator relations, Theorem 11.8. The exceptions occur when
certain structure constants cmn
αβ appearing in the commutator formula are
divisible by char(k).

In fact one has very good control over the kL-modules which occur in the
quotients Qj /Qj+1 .

Theorem 17.6 (Azad–Barry–Seitz) Let G be a simple algebraic group and

assume that (G, char(k)) = (Bn , 2), (Cn , 2), (F4 , 2), (G2 , 2), (G2 , 3). Then for
any parabolic subgroup P = QL of G we have:

(a) There exists a unique root βS of shape S having maximal height with
respect to this property.
(b) For each shape S, VS is an irreducible kL-module of highest weight βS .
(c) If S = S  , then VS ∼
= VS  as kL-modules.
(d) Qj /Qj+1 is the direct sum of the VS , for the shapes S of level j, hence
a completely reducible kL-module.

Sketch of proof The assertion of (a) is just Exercise A.7. As all roots of
shape S and of the same length are conjugate under the Weyl group WI of
L, by Lemma A.31, this already shows that VS is irreducible when there is
only one root length. By avoiding the small characteristic configurations ex-
cluded in the theorem, it is straightforward to show this in general. Part (b)
then follows from the classification of highest weight modules given in Theo-
rem 15.17(a) since clearly any non-identity element in the root group UβS is
 17.1 Internal modules 143

a maximal vector of weight βS for the Levi factor L. Part (c) is now imme-
diate by Theorem 15.17(a). We have already shown in Proposition 17.4(b)
that Qj /Qj+1 decomposes as a direct sum of the VS , whence (d).

Remark 17.7 Under the assumptions of Theorem 17.6, Azad, Barry and
Seitz also show that the Levi complement L has finitely many orbits on each
VS . In the proof, they apply Richardson’s result [59, Thm. E] to see that
there exist a finite number of L-orbits on Q = Q1 and so on Q1 /Q2 , and
therefore on the VS occurring in Q1 /Q2 as well. For the general case, one
realizes Qj /Qj+1 as the commutator quotient of the unipotent radical of a
parabolic subgroup of a certain reductive subgroup of G. See [3, Thm. 2(f)]
for the details.

Definition 17.8 The kL-modules VS appearing in the previous theorem

are called internal modules for L.

Example 17.9 We illustrate some types of internal modules which may

occur.

(1) Let G = SLn , the simply connected group of type An−1 , with the
parabolic subgroup P as in Example 17.2. The Levi complement has
the form L = U±α1 , T  = [L, L]Z(L)◦ , where Z(L)◦ is an (n − 2)-
dimensional torus and [L, L] is a rank 1 semisimple group, so isomorphic
to SL2 or PGL2 by Theorem 8.8 (in fact isomorphic to SL2 by Proposi-
tion 12.14). Let’s consider some of the levels worked out in Example 17.2.
At level 1, for shape S1 = (1, 0, . . . , 0) we have VS1 = Uα2 · Uα1 +α2 .
This is easily seen to be the natural two-dimensional module for [L, L].
For each i > 1, VSi = Uαi+1 is the trivial module for the shape Si =
(0, . . . , 0, 1, 0, . . . , 0).
At level i for i ≥ 2, let S = (1 , . . . , n−2 ), j = 0 or 1 for all j and
n−2
j=1 j = i be a shape of level i. Then if 1 = 1, the corresponding
simple module VS is the natural two-dimensional module for [L, L] and
if 1 = 0, then VS is the trivial module.
(2) For a more elaborate example, let G = SO2n (see Definition 1.15) and B
the Borel subgroup of G consisting of upper triangular matrices exhibited
in Example 6.7(4). Recall the basis {v1 , . . . , v2n } of the orthogonal space
on which G naturally acts, such that the quadratic form with respect to
this basis is given by f (x1 , . . . , x2n ) = x1 x2n + x2 x2n−1 + · · · + xn xn+1 .
Then set V1 = v1 , . . . , vn , a totally singular subspace of V , and let
P = StabG (V1 ), a parabolic subgroup (see Proposition 12.13). Then one
144 Structure of parabolic subgroups, II

checks that P has Levi factor

  
A 0 
L=  A ∈ GLn ,
0 Kn A−tr Kn 

where Kn is as in Section 1.2, and unipotent radical

 
In Kn S 
Q=  S ∈ k n×n
skew-symmetric .
0 In
Let’s consider the action of L on Q. We have



−1

A 0 I n Kn S A 0 I AKn SB −1
= n ,
0 B 0 In 0 B 0 In
where B = Kn A−tr Kn . Hence the action of L on Q is equivalent to
the natural action of GLn on the space of skew-symmetric matrices:
S → ASAtr , for A ∈ GLn and S ∈ k n×n skew-symmetric.
Let V be the natural n-dimensional kGLn -module (viewed as column
vectors written with respect to a fixed basis). A skew-symmetric matrix
A defines a bilinear map ϕA : V ⊗ V → k via v ⊗ w → v tr Aw. As
A is skew-symmetric, ϕA induces a linear map ϕ̄A : ∧2 V → k. The
map A → ϕ̄A defines a kGLn -module isomorphism between the space of
skew-symmetric matrices and Hom(∧2 V, k) = (∧2 V )∗ . It turns out that
this is an irreducible module for GLn . Hence Q is irreducible, Qj = 1
for all j > 1 and Q = VS for a unique shape S = (1), that is, coefficient
1 of the unique simple root lying outside of the root system of L.
(3) In this example we look at some internal modules for G = SO8 , with
root system of type D4 and Dynkin diagram

3
1 2
4

Fix a Borel subgroup B and a maximal torus T ≤ B, such that

{α1 , α2 , α3 , α4 } is the corresponding base of Φ.
Let P be the standard parabolic subgroup of G corresponding to
the subset {α1 , α3 , α4 }, with Levi decomposition P = QL, where Q =
Ru (P ). Then [L, L] = U±α1 , U±α3 , U±α4 . Here, Q1 = Q and Q2 =
Uα1 +2α2 +α3 +α4 .
Now Q2 is one-dimensional and so must be the trivial module for
[L, L], with highest weight 0.
As there is a unique shape of roots of level 1, Theorem 17.6 implies
 17.2 The theorem of Borel and Tits 145

that Q1 /Q2 is irreducible. Set BL = Uα1 , Uα3 , Uα4 , T , a Borel subgroup

of L. By Theorem 11.8, [Uαi , Uα1 +α2 +α3 +α4 ] = 1 for i = 1, 3, 4. So
Uα1 +α2 +α3 +α4 Q2 /Q2 is a BL -stable line in Q1 /Q2 . Therefore xQ2 is a
maximal vector in the irreducible kL-module Q1 /Q2 for any non-identity
element x ∈ Uα1 +α2 +α3 +α4 .
Note that the weight α1 + α2 + α3 + α4 restricted to the maximal
torus im(αi∨ ) of U±αi , for i = 1, 3, 4, is precisely the fundamental
dominant weight λ (for the group SL2 ). In particular, each of these
three subgroups is isomorphic to SL2 and [L, L] is an image of the rank 3
group SL2 × SL2 × SL2 . The quotient Q1 /Q2 as module for [L, L] is then
L(λ) ⊗ L(λ) ⊗ L(λ).

17.2 The theorem of Borel and Tits

The internal modules arising in Theorem 17.6 are important when consid-
ering embeddings of reductive groups. To illustrate this, we first need the
following fundamental result on normalizers of unipotent subgroups (see [32,
Prop. 30.3] or [29, Thm. 3.1.1]):

Theorem 17.10 (Borel–Tits) Let G be a connected reductive linear alge-

braic group. Let U ≤ G be a closed unipotent subgroup of G, which lies in
a Borel subgroup of G. Then there exists a parabolic subgroup P of G with
U ≤ Ru (P ) and NG (U ) ≤ P .

Sketch of proof Let U ≤ G be a closed unipotent subgroup lying in a Borel

subgroup B of G. Set U0 := U and define inductively Ni+1 = NG (Ui ) and
Ui+1 = Ui .Ru (Ni+1 ) for all i ≥ 0. Then U = U0 ≤ U1 ≤ U2 ≤ . . . and
N1 ≤ N2 ≤ . . . are chains of closed subgroups of G. (By Exercise 10.18,
normalizers are closed.)
We claim that each Ui lies in a Borel subgroup of G. This is already the
case for U0 , so we assume it is true for Ui . Thus, the set of fixed points
of Ui acting by left multiplication on G/B is non-empty and closed, and
stabilized by the connected solvable group Ru (Ni+1 ) ≤ NG (Ui ). Hence by
Theorem 6.1, there exists gB which is fixed by Ru (Ni+1 ) as well as by Ui .
So Ui+1 = Ui .Ru (Ni+1 ) lies in the Borel subgroup gBg −1 as claimed. Now,
for all i, as

Ui+1 /Ui = Ui .Ru (Ni+1 )/Ui ∼

= Ru (Ni+1 )/(Ui ∩ Ru (Ni+1 ))

is connected, either dim(Ui+1 /Ui ) ≥ 1 or Ui+1 /Ui = 1. Since dim(G) is

146 Structure of parabolic subgroups, II

finite, the sequence U ≤ U1 ≤ . . . stabilizes at some point, say Ul = Ul+j for

all j ≥ 1 and hence Nl+1+j = Nl+1 for j ≥ 1.
Set P (U ) := Nl+1 and V := Ul . Then, by our previous argument, V lies in
a Borel subgroup and V ⊇ Ru (NG (V )). The difficult step in the proof, which
is given in [32, Prop. 30.3], shows that, under the established conditions, V
is connected, and hence U ≤ V = Ru (NG (V )) and P (U ) is a parabolic
subgroup of G, and the result follows.
Corollary 17.11 Let U ≤ G be a closed connected unipotent subgroup.
Then there exists a parabolic subgroup P of G such that U ≤ Ru (P ) and
NG (U ) ≤ P .
Proof Since U is a closed connected unipotent, hence solvable subgroup of
G, U lies in a Borel subgroup. The result then follows from Theorem 17.10.

We now state a consequence which is useful when considering inclusions

of one reductive group in another.
Corollary 17.12 Let H, G be connected reductive algebraic groups with H
a closed subgroup of G. Let PH be a parabolic subgroup of H. Then there
exists a parabolic subgroup PG of G with Ru (PH ) ≤ Ru (PG ) and PH ≤ PG .
Proof The subgroup Ru (PH ) is a closed connected unipotent group. So, by
Corollary 17.11 there exists a parabolic subgroup PG of G such that
PH ≤ NG (Ru (PH )) ≤ PG
and Ru (PH ) ≤ Ru (PG ).
This can be used to study embeddings of connected reductive algebraic
groups H ≤ G. Let PH be a parabolic subgroup of H, and P the parabolic
subgroup of G with QH = Ru (PH ) ≤ Ru (P ) = Q and PH ≤ P , whose
existence is asserted by Corollary 17.12. Let PH = QH LH and P = QL be
Levi decompositions of PH , respectively P . Then we obtain an embedding
of reductive groups LH → L via

= QLH /Q ≤ P/Q ∼
LH ∼ = L.
Now L acts on each of the quotients Qj /Qj+1 in Q (defined in Section 17.1),
with LH stabilizing the image of QH in this quotient. Using Theorem 17.6,
we obtain a composition series for QH as kLH -module and this must be
compatible with the restrictions of the kL-modules Qj /Qj+1 to LH . This
can be used to provide precise information about the embeddings LH → L
and QH ≤ Q.
 17.2 The theorem of Borel and Tits 147

Example 17.13 Let us apply this discussion to a known embedding of

reductive groups, namely G2 ≤ G, where G is semisimple with root system
of type B3 and char(k) = 2, 3. (The group G2 has an irreducible seven-
dimensional representation on which it fixes a non-degenerate quadratic
form.) Let T be a maximal torus of G2 and B ≤ G2 a Borel subgroup contain-
ing T . Let Φ+ be the corresponding set of positive roots with base {α1 , α2 }
(ordered as in the Dynkin diagram in Table 9.1; see Example 9.5(4) for the
root system). Let P ≤ G2 be the standard parabolic subgroup containing B
corresponding to the simple root α1 , so P has a Levi factor L = U±α1 , T 
and unipotent radical Q = Uα2 Uα1 +α2 U2α1 +α2 U3α1 +α2 U3α1 +2α2 . By Theo-
rem 17.6, Q1 /Q2 is a four-dimensional irreducible kL-module and Q2 is a
one-dimensional trivial kL-module.
Now let PG be a parabolic subgroup of G with P ≤ PG and Q ≤ Ru (PG ) :=
QG as in Corollary 17.12. Let BG ≤ PG be a Borel subgroup of G and Φ+ G the
corresponding set of positive roots with base {β1 , β2 , β3 }. As Q ≤ QG , we
know that dim QG ≥ 5 and that a Levi factor LG of PG acting on QG must
have a composition factor of dimension at least 4 on some (QG )j /(QG )j+1 .
This restricts the possibilities for PG = PI to one of the following: ∆I =
{β1 , β3 } or ∆I = {β2 , β3 }. If we further assume that PG is minimal with
respect to P ≤ PG , then using Proposition 12.13 and the representation
theory of SL2 , one can see that in the second case, L would act irreducibly
on the five-dimensional LG -module QG , and therefore would not admit the
required composition series of Q. Hence we conclude that the parabolic PG
corresponds to the subset ∆I = {β1 , β3 }.

The above corollary also has the following important consequence for the
study of maximal subgroups of a reductive group.

Corollary 17.14 Let G be a connected reductive algebraic group. Let H ≤

G be maximal among proper closed subgroups of G (with respect to inclusion).
Then either H ◦ is reductive or H is a parabolic subgroup of G.

Proof Suppose H ◦ is not reductive. Then Ru (H ◦ ) is a non-trivial closed

connected unipotent subgroup of G. By Corollary 17.12, there exists a proper
parabolic subgroup P of G with Ru (H ◦ ) ≤ Ru (P ) and NG (Ru (H ◦ )) ≤ P .
So, H ≤ NG (Ru (H ◦ )) ≤ P . But since H is maximal, we have H = P .

We conclude our discussion of the Borel–Tits Theorem with a further

corollary, which shows that the hypotheses of Corollary 17.11 may be relaxed.

Corollary 17.15 Let G be a connected reductive group and U ≤ G a

unipotent subgroup of G. Then U lies in a Borel subgroup of G.
148 Structure of parabolic subgroups, II

Proof The proof is by induction on dim(G). If dim(G) = 1 then G is solvable

by Theorem 3.2 so it is a Borel subgroup of G containing U .
Suppose dim(G) > 1. Since U is unipotent, it is nilpotent by Corol-
lary 2.10; in particular Z(U ) = 1. Let 1 = u ∈ Z(U ) and set U1 = u,
the closure of the subgroup generated by u. Then U ≤ NG (u) ≤ NG (U1 ).
By Theorem 6.10, u lies in some Borel subgroup B of G. So, u ≤ B; hence
U1 ≤ B and in fact U1 ≤ Ru (B), by Theorem 4.4. By Theorem 17.10 there
exists a parabolic subgroup P of G with U ≤ NG (U1 ) ≤ P and U1 ≤ Ru (P ).
Now U1 = 1, so Ru (P ) = 1, hence P = G. The connected reductive group
P/Ru (P ) has dimension less than dim(G). By induction, U.Ru (P )/Ru (P )
lies in a Borel subgroup of P/Ru (P ), of the form H/Ru (P ) for some closed
connected solvable subgroup H of G with Ru (P ) ≤ H. Now there exists a
Borel subgroup B1 of G with B1 ≤ P and Ru (P ) ≤ B1 . So B1 /Ru (P ) is
a closed connected solvable subgroup of P/Ru (P ) and hence lies in some
conjugate of H/Ru (P ). But then B1 ≤ H x for some x ∈ G whence H is a
Borel subgroup of G and U ≤ H as required.
Remarks 17.16 (a) With U as in Corollary 17.15, Theorem 17.10 applies
to produce a parabolic subgroup P of G with U ≤ Ru (P ) and NG (U ) ≤
P . In fact, as shown in [7, Cor. 3.9], P is canonical in the sense that any
automorphism of G which stabilizes U must stabilize P .
(b) We now see that maximal unipotent subgroups of G are unipotent
radicals of Borel subgroups.
 18
Maximal subgroups of classical type simple
algebraic groups

Let V be a finite-dimensional vector space over k equipped with the 0

form, a non-degenerate symplectic bilinear form (and dim V ≥ 4), or a
non-degenerate symmetric bilinear form and associated quadratic form (and
dim V ≥ 7). In particular, when char(k) = 2 and V is equipped with a
non-zero form, then dim V is even. Let Isom(V ) denote the full group of
isometries of V , so GL(V ), Sp(V ), GO(V ), respectively. Let Cl(V ) denote
the group SL(V ), Sp(V ), SO(V ), respectively. (Recall that by our definition
SO(V ) = GO(V )◦ .) Set G = Cl(V ) for the remainder of this section; so G
is a simple algebraic group (see Table 9.2).

18.1 A reduction theorem

We first establish a reduction theorem for the study of the closed connected
subgroups of simple classical groups.
Before stating this result, we recall the relationship between homogeneous
representations, tensor products of representations and central products of
groups, which will be used. For H a group, a kH-module V is said to be
homogeneous if it is the direct sum of isomorphic simple kH-modules. For
representations ρi : Hi → GL(Vi ) of groups Hi , i = 1, 2, we write H1 ⊗ H2 ≤
GL(V1 ⊗ V2 ) for the central product generated by the images of the Hi ,
letting ρi (Hi ) act trivially on V3−i .
Proposition 18.1 Let K be a field, H a group and V = V1 ⊕· · ·⊕V1 (r > 1
summands) a homogeneous finite-dimensional KH-module, with V1 an abso-
lutely irreducible KH-module. Then there is a tensor product decomposition
V = V1 ⊗ V2 with dim V2 = r, such that H ≤ GL(V1 ) ⊗ 1. Moreover,
CGL(V ) (H) = 1 ⊗ GL(V2 ), and CGL(V ) (CGL(V ) (H)) = GL(V1 ) ⊗ 1.
150 Maximal subgroups of classical type simple algebraic groups

Proof As V1 is absolutely irreducible, V2 := HomKH (V1 , V ) is an r-dimen-

sional K-vector space. Then the K-linear map

ψ : V1 ⊗ V2 → V, v ⊗ ϕ → ϕ(v),

defines an isomorphism, since indeed both sides have the same dimension,
and surjectivity is easily verified. Endowing V2 with the trivial H-action, ψ
even becomes an isomorphism of KH-modules, as

ψ(h(v ⊗ ϕ)) = ψ(hv ⊗ ϕ) = ϕ(hv) = hϕ(v) = hψ(v ⊗ ϕ)

for all h ∈ H, v ∈ V1 , ϕ ∈ V2 .
Now clearly 1 ⊗ GL(V2 ) ≤ CGL(V ) (H). Conversely, if g ∈ CGL(V ) (H)
then g acts on the embeddings of V1 into V , so on V2 via g : V2 → V2 ,
(gϕ)(v) = gϕ(v). This action is faithful, since an element g in the kernel
satisfies gϕ(v) = ϕ(v) for all ϕ ∈ V2 , v ∈ V1 . Thus CGL(V ) (H) embeds into
1 ⊗ GL(V2 ). The last statement follows as before by replacing H acting on
V1 by GL(V2 ) acting on V2 .

Corollary 18.2 Let K be a field, H = H1 · · · Hr a commuting product

of normal subgroups Hi  H, V a finite-dimensional absolutely irreducible
KH-module.

(a) There exist absolutely irreducible KHi -modules Vi such that V = V1 ⊗

· · · ⊗ Vr , with Hi acting trivially on Vj for j = i.
(b) If V is a self-dual KH-module, then Vi is self-dual for Hi , for 1 ≤ i ≤ r.

Proof Clearly it suffices to consider the case when r = 2, so H = H1 H2 .

Then the socle of V |H1 is an H-invariant submodule of V , so all of V by irre-
ducibility, whence V |H1 is completely reducible. Moreover, H2 must permute
the irreducible summands transitively, and since [H1 , H2 ] = 1, they must all
be H1 -isomorphic, that is, V |H1 is homogeneous. Then Proposition 18.1 ap-
plies to show that V = V1 ⊗ V2 and H2 ≤ CGL(V ) (H1 ⊗ 1) = 1 ⊗ GL(V2 ).
t
For (b) let’s write V |H1 = i=1 V1 for some absolutely irreducible KH1 -
t
module V1 . Since V is self-dual, we also have V |H1 = i=1 V1∗ , hence V1 ∼ =
V1∗ as KH1 -modules, as claimed.

As a final preparation we state a result about forms on tensor products

([50, Prop. 2.2]) which gives a kind of converse to Corollary 18.2(b), see
Exercise 20.24:

Proposition 18.3 Let fi be non-degenerate bilinear forms on the finite-

dimensional k-vector spaces Vi , i = 1, 2, and set V := V1 ⊗ V2 . Then:
 18.1 A reduction theorem 151

(a) There is a unique bilinear form f on V such that

f (v1 ⊗ v2 , w1 ⊗ w2 ) = f1 (v1 , w1 )f2 (v2 , w2 ) for all vi , wi ∈ Vi .

(b) The form f is symmetric if and only if f1 , f2 are either both symmetric
or both skew-symmetric.
(c) If char(k) = 2 then there is a unique quadratic form Q on V , with
associated bilinear form f , such that Q(v1 ⊗ v2 ) = 0 for all vi ∈ Vi , and
Q is preserved by Sp(V1 ) ⊗ Sp(V2 ).

The first weak reduction result for closed connected subgroups of classical
groups is now the following:

Proposition 18.4 Let H be a closed connected subgroup of G = Cl(V ).

Then one of the following holds:

(1) H ≤ StabG (X) with X ≤ V a proper non-zero subspace which is either

totally singular or non-degenerate, or p = 2 and X is non-singular of
dimension 1 in an orthogonal space;
(2) V = V1 ⊗ V2 and H lies in a subgroup of the form Cl(V1 ).Cl(V2 ) acting
naturally on V1 ⊗ V2 . Here V is equipped with the product form and
dim Vi ≥ 2 for i = 1, 2; or
(3) H is a simple algebraic group acting irreducibly on V and V |H is tensor
indecomposable.

Proof If V |H is reducible then choose 0 = X ≤ V minimal with respect to

being H-invariant. Then X ∩ X ⊥ is an H-invariant subspace, so the mini-
mality of X implies that either X ∩ X ⊥ = 0 and X is non-degenerate, or
X ∩ X ⊥ = X and X is totally isotropic. Moreover in the latter case, if V is
an orthogonal space and char(k) = 2, then the set of singular vectors in X
is again H-invariant and by minimality either X is totally singular or X is
a one-dimensional space of non-singular vectors, as in (1).
So, we may and will assume now that V |H is irreducible. Then H is re-
ductive (see Proposition 15.1). Thus we have H = [H, H]Z(H)◦ by Corol-
lary 8.22. Now, H acts irreducibly, so by Schur’s Lemma Z(H) acts by scalars
on V . Thus, Z(H) ≤ Z(Cl(V )) which is finite, whence H is semisimple and
H = [H, H].
By Theorem 8.21(d), H = H1 · · · Ht is a central product of simple algebraic
groups Hi , and so V |H = V1 ⊗ · · · ⊗ Vt with Vi an irreducible representa-
tion of Hi , by Corollary 18.2(a). Moreover, if V is self-dual, then so are the
kHi -modules Vi by Corollary 18.2(b), and thus Hi ≤ Cl(Vi ) for the corre-
sponding classical group. Since Vi is a faithful module for the simple group
152 Maximal subgroups of classical type simple algebraic groups

Hi , dim Vi ≥ 2. Hence if t > 1, H stabilizes a tensor product decomposition

on V as in the statement of (2).
We now assume that t = 1, that is, H is simple. Again, if V |H is a
tensor product of irreducible kH-modules, we have H as in (2); otherwise
we have (3).
Note that the above result gives no information about finite subgroups
of G.
In order to formulate a more powerful reduction theorem for arbitrary (not
necessarily connected) closed subgroups H of a classical group G = Cl(V ),
we first need to define five classes C1 , . . . , C5 of natural subgroups:
Class C1 (Subspace stabilizers). Here H ∈ C1 if H = StabG (X) where X is
a proper non-zero subspace of V , with X totally singular or non-degenerate
or non-singular of dimension 1 if G = SO(V ) with p = 2.

Class C2 (Stabilizers of orthogonal decompositions). Here H ∈ C2 if H is

t
the stabilizer of an orthogonal decomposition V = i=1 Vi with Vi mutually
orthogonal and isometric subspaces of V . Thus H = StabG (V1 ⊥ . . . ⊥Vt ) ∼=
(Isom(V1 )  St ) ∩ G, where we write St for the symmetric group on t letters.

Class C3 (Stabilizers of totally singular decompositions). Here G = Sp(V )

or SO(V ) and H ∈ C3 if H is the stabilizer of a direct sum decomposition
V = X ⊕ X  with X, X  maximal totally isotropic (singular if V is equipped
with a quadratic form) subspaces of V . In particular, dim V = 2m is even
and H ∼ = GLm .2 or GLm , the latter happens when m is odd and G = SO(V ).
(The element of order 2 exchanges the two spaces X and X  ; the condition of
lying in the connected component of Isom(V ) shows that this element does
not lie in G when m is odd and G = SO(V ).)

Class C4 (Stabilizers of tensor product decompositions). Here, there are two

cases:
(a) V = V1 ⊗ V2 with V1 not isometric to V2 and dim(Vi ) > 1. The central
product Cl(V1 ).Cl(V2 ) acts naturally on V and H = NG (Cl(V1 ).Cl(V2 )).
-t
(b) V = i=1 Vi with t > 1 where Vi are mutually isometric with dim(Vi ) ≥
2 and
! t
! t 
H = NG ( Cl(Vi )) = ( Cl(Vi )).St ∩ G,
i=1 i=1

where i Cl(Vi ) acts naturally on ⊗i Vi and the Cl(Vi ) are simple.
 18.1 A reduction theorem 153

Class C5 (Finite local subgroups). Recall that a subgroup of a finite group

is called local if it normalizes a non-trivial r-subgroup for some prime r. Let
r be a prime with r = char(k). Let R be an extra-special r-group of order
r2m+1 (and we will write R = r2m+1 ), or when r = 2 we may take R to be
22m+1 or a central product of such a group with a cyclic group of order 4,
i.e., Z4 ◦ 22m+1 .
Every faithful irreducible k-representation of R has dimension rm ; more-
over, these representations are uniquely determined by the action of Z(R).
(See [28, 5.4, 5.5].) Take ρ : R −→ GLrm (k) to be such a representation and
V the corresponding kR-module. Now Z(R) is cyclic of order r or of order
4, if R = r2m+1 , respectively R = Z4 ◦ 22m+1 ; a generator of Z(R) acts as
a scalar ω on V and as ω −1 on the dual V ∗ . So V can be self-dual only if
Z(R) has order 2. In particular, if r is odd or if r = 2 and R = Z4 ◦ 22m+1 ,
then R fixes no non-degenerate form on V , while in fact if R = 22m+1 ,
then ρ(R) ≤ SOrm (k) or ρ(R) ≤ Sprm (k). We consider the representation
ρ : R −→ Cl(V ). Then H ∈ C5 if H = NCl(V ) (ρ(R)). Here the structure of
H/Z(G) is described in Table 27.2 in Part III.
Definition 18.5 For a classical group G = Cl(V ), write C(G) := C1 ∪ C2 ∪
C3 ∪ C4 ∪ C5 . For H ≤ G a closed subgroup, write H ∞ for H ◦ if H is infinite
and for the last term of the derived series of H if H is finite.
Recall that an abstract group H is said to be almost simple if there is a
simple group S such that S ≤ H ≤ Aut(S). A finite group H = 1 is called
quasi-simple, if H is perfect and H/Z(H) is simple. The subnormal quasi-
simple subgroups of a finite group H are called the components of H (see [2,
§31], for example).
We can now state the reduction theorem ([50, Thm. 1]):
Theorem 18.6 (Liebeck–Seitz) Let G = Cl(V ) be a classical algebraic
group and H a closed subgroup of G. Then one of the following holds:
(1) there exists M ∈ C(G) such that H ≤ M , or
(2) H.Z(G)/Z(G) is almost simple and H ∞ acts irreducibly on V . Moreover,
if H is infinite, H ◦ acts tensor indecomposably on V .
Sketch of proof Let H ≤ G be a closed subgroup not satisfying the conclu-
sion. We consider the action of H on the natural kG-module V . If V |H is
reducible, then we get H ≤ M ∈ C1 by the first paragraph in the proof of
Proposition 18.4, so V |H is irreducible.
Step 1: V |H ◦ is homogeneous.
Since H stabilizes the socle of V |H ◦ , this socle is V . Suppose that V |H ◦ is
154 Maximal subgroups of classical type simple algebraic groups

not homogeneous, that is V |H ◦ ∼ = V1 ⊕ · · · ⊕ Vt and Vj ∼= Yj ⊕ · · · ⊕ Yj ,

for some set of pairwise non-isomorphic irreducible kH ◦ -modules Y1 , . . . , Yt ,
with t > 1. The irreducibility of the kH-module V forces H to permute the
V1 , . . . , Vt transitively. If G = SL(V ), then H ≤ StabG (V1 ⊥ . . . ⊥ Vt ), the
latter subgroup lying in C2 , contradicting our assumption on H.
So we have G = Sp(V ) or SO(V ). Consider V1 ∼ = Y1 ⊕ · · · ⊕ Y1 as kH ◦ -
module, and assume for the moment that Y1 is not a self-dual kH ◦ -module.
The non-degenerate form induces an isomorphism V ∼ = V ∗ and, via this
⊥ ∼ ∗ ∼
isomorphism, we obtain V /Y1 = Y1 . As Y1 = Y1 , we have V1 ⊆ Y1⊥ , which
∗

in turn implies that V1 is totally isotropic. Moreover, if G = SO(V ) and

p = 2, then the irreducibility of Y1 implies that Y1 is either totally singular
or a non-singular 1-space. In the latter case, we have Y1 the trivial module
for H ◦ and hence Y1 ∼ = Y1∗ , contradicting our assumption. So in all cases
we have Y1 , and hence V1 , totally singular, and by the transitivity of H on
the Vj , all Vj are totally singular. Finally, we have V /V1⊥ ∼ = V1∗ ∼ = V1 , so,
⊥
relabeling if necessary, we get V = V1 ⊕ V2 and V1 ⊕ V2 is a non-degenerate
subspace of V . Working inside (V1 ⊕ V2 )⊥ we find (again after relabeling)
that V |H ◦ = (V1 ⊕ V2 ) ⊥ . . . ⊥ (Vt−1 ⊕ Vt ). If t = 2, then V |H ◦ = V1 ⊕ V2 is a
decomposition into two totally isotropic subspaces, so H lies in a subgroup
M ∈ C3 and if t > 2, H lies in a subgroup M ∈ C2 , contradicting our
assumption on H.
Now we turn to the case where Y1 ∼ Y ∗ and so Vj ∼
= Vj∗ for all j. Then
⊥ ∼ ∗ ∼
 = 1⊥
V /V1 = V1 = V1 and so V1 ⊆ ( j>1 Vj ) . Then H preserves the decom-
position V1 ⊥ . . . ⊥ Vt , with each Vj non-degenerate and so H ≤ M ∈ C2 ,
again contradicting our assumption on H.
Step 2: V |H ◦ is irreducible or H is finite.
Suppose that V |H ◦ is reducible (and homogeneous by Step 1), that is,
V |H ◦ ∼
= Y ⊕ · · · ⊕ Y (r > 1 summands), for some simple H ◦ -module Y .
If dim Y > 1, we apply Proposition 18.1 to find that V = V1 ⊗ V2 with

H ◦ ≤ GL(V1 ) ⊗ 1 = CGL(V ) (CGL(V ) (H ◦ )).

Moreover, H normalizes the product of derived subgroups

(GL(V1 ) ⊗ 1) (1 ⊗ GL(V2 )) = SL(V1 ) ⊗ SL(V2 )

and hence in the case G = SL(V ), we have H ≤ M ∈ C4 .

When G = Sp(V ) or SO(V ), a long argument (to be found in the proof
of [50, Lemma 3.3]) shows that H ≤ M ∈ C4 as before. So in all cases, we
have H ≤ M ∈ C(G), contradicting our assumption. So dim Y = 1, and
H ◦ ≤ Z(G) and so H is finite.
 18.2 Maximal subgroups of the classical algebraic groups 155

Step 3: H is finite.
Assume that H is infinite, so H ◦ = 1. By Step 2 we have that V |H ◦ is
irreducible. As in the proof of Proposition 18.4, we see that H ◦ is reductive,
indeed semisimple as Z(H ◦ ) must lie in Z(G). Thus H ◦ = H1 · · · Hr , a
commuting product of simple algebraic groups. If r > 1, then V |H ◦ is a tensor
product of irreducible Hi -modules by Corollary 18.2 and H ≤ M ∈ C4 . So
we must have that r = 1 and H ◦ = H1 acts irreducibly on V , and indeed
tensor indecomposably (else H ≤ M ∈ C4 ). So finally we have H as in (2) of
the theorem, contradicting our assumption on H.
Step 4: H is finite with no components and H ≤ Z(G).
By Step 3, H is finite. As H is not contained in a subgroup lying in C(G)
and does not satisfy the conclusion (2) of the result, the above arguments
show that any non-trivial normal subgroup of H acts irreducibly on V , and
so H ≤ Z(G). Let E := E1 ◦ · · · ◦ Et be the product of the components of H,
a characteristic subgroup of H. Suppose E = 1. Arguing as in Step 3 above,
replacing H ◦ by E, we see that either H lies in a member of C4 , or t = 1
and E = E1 acts irreducibly on V . In the latter case CG (E) ≤ Z(G), so
HZ(G)/Z(G) ≤ Aut(EZ(G)/Z(G)) and so HZ(G)/Z(G) is almost simple
and H ∞ ≥ E acts irreducibly on V as in conclusion (2). Thus we have E = 1.
Step 5: Final contradiction.
So finally we have that H is finite and has no components. Set H̄ :=
HZ(G)/Z(G) = 1. Then all minimal normal subgroups of H̄ are elemen-
tary abelian, that is, direct products of cyclic groups of equal prime order.
Fix one such prime r, and let Q be a Sylow r-subgroup of the preimage in
H of the product of all minimal normal r-subgroups of H̄, a characteristic
subgroup of H. Let R denote its subgroup generated by all elements of order
r, respectively of order 4 if r = 2. Then R is a normal r-subgroup of H.
Then as above, V |R is irreducible. In particular, r = p = char(k) by
Proposition 2.9. In this case the irreducibility of V |R forces every character-
istic abelian subgroup of R to be cyclic. Then R has precisely the structure
of the r-groups described in C5 , by [2, 23.9]. Hence H ≤ M ∈ C5 , a final
contradiction.

18.2 Maximal subgroups of the classical

algebraic groups
We turn now to the determination of the maximal positive-dimensional sub-
groups of the classical algebraic groups; let G = Cl(V ) be as defined in the
156 Maximal subgroups of classical type simple algebraic groups

previous section. We must discuss the maximality of the subgroups in the

various families in Theorem 18.6(1), excluding however the finite groups in
the family C5 , as well as of those subgroups satisfying condition (2) of loc. cit.
It will be convenient to define a subclass of the family of subgroups satisfying
Theorem 18.6(2):

Class C6 (Normalizers of classical groups). Here H ∈ C6 if H = NG (Sp(V ))

or H = NG (SO(V )) in G = SL(V ), and H = NG (SO(V )) in G = Sp(V ) if
p = 2.

We let S k denote the remaining positive-dimensional subgroups, that is:

Class S k . Here H ∈ S k if HZ(G)/Z(G) is almost simple, H ◦ acts irreducibly

and tensor-indecomposably on V , and if H is of classical type, V is not the
natural module for H.

Let H1 , H2 ∈ C1 ∪ C2 ∪ C3 ∪ C4 ∪ C6 ∪ S k and suppose H1 < H2 . It turns

out that the most elusive configurations are those where H2 ∈ S k and H1 ∈
C4 ∪ S k . In that case H1◦ is semisimple and acts irreducibly on V , and if H2 is
of classical type, V is not the natural module for H2 . Moreover if H1◦ = H2◦ ,
then H1◦ ≤ H1 ≤ NG (H2◦ ) and so H1 corresponds to some subgroup of the
finite (by Theorem 11.11) group NG (H2◦ )/H2◦ . So finally we may assume
H1◦ < H2◦ .
Thus in order to complete the classification of the maximal positive-dimen-
sional closed subgroups of the classical algebraic groups, we must determine
all triples (X, Y, V ) satisfying the following conditions:

(i) Y is a simple algebraic group,

(ii) X is a semisimple closed proper subgroup of Y ,
(iii) V is a non-trivial tensor indecomposable kY -module,
(iv) X acts irreducibly on V , and
(v) if Y is a classical group then V is not the natural module for Y .

If V is a non-restricted kY -module, then by Theorem 16.12, V ∼

i
= M (p )
for some restricted irreducible kY -module M . As X acts irreducibly on V ,
it must also act irreducibly on M , and hence we may assume that

(iii)* V is a non-trivial tensor indecomposable restricted kY -module.

Also, as our aim is to obtain a complete list of triples (X, Y, V ), we will

assume further that

(i)* Y is a simple algebraic group of simply connected type.

 18.2 Maximal subgroups of the classical algebraic groups 157

The condition (i)* implies that the character group of a maximal torus of
Y coincides with the lattice of abstract weights Ω, and thus in particular, by
Theorem 15.17(b), there exist irreducible highest weight modules V for any
dominant weight from Ω. Finally, the condition (v) corresponds to the fact
that we are interested in the maximality of NG (X) in the smallest classical
group containing it.
The main results of Dynkin [23, 24] (which cover the case when char(k) =
0) and Seitz [60] and Testerman [80] (which cover the case when char(k) =
p > 0) are summarized in the following theorem. For the purposes of this
theorem, we will take p = ∞ when char(k) = 0. Before stating the result, let
us introduce some additional notation. Let {ωi | 1 ≤ i ≤ rk(X)}, respectively
{λj | 1 ≤ j ≤ rk(Y )}, be a set of fundamental dominant weights with respect
to a fixed choice of maximal torus and Borel subgroup of X, respectively Y ,
where we label Dynkin diagrams as in Table 9.1.

Theorem 18.7 (Dynkin, Seitz, Testerman) Let Y and V satisfy conditions

(i)*, (iii)* and (v), and let ρ : Y → SL(V ) be the corresponding rational
representation of Y . Let X be a subgroup of Y satisfying (ii). Then V |X is
irreducible if and only if one of the following occurs:

(1) the triple (ρ(X), ρ(Y ), V ) (or (ρ(X), ρ(Y ), V ∗ )) belongs to one of the
infinite series given in Table 18.1; or
(2) the triple is among a specified finite list of additional triples, each cor-
responding to a fixed embedding of simple algebraic groups and a kY -
module V of fixed p-restricted highest weight.

Remarks 18.8 (a) Recall that V is a restricted kY -module and hence

all of the parameters a, b, ai are non-negative integers, strictly less than p.
Moreover, when there is a congruence relation modulo p, it is understood
that char(k) > 0.
(b) The majority of the examples of Table 18.1 above correspond to nat-
ural embeddings of X in the group Y . For example, the first two config-
urations Cn < A2n−1 and Bn < A2n arise from the natural embeddings
of a symplectic or orthogonal group in the linear group. The embeddings
An < A(n2 +n−2)/2 and An < A(n2 +3n)/2 correspond to the action of SLn+1
on the exterior square or symmetric square of its natural module.
The embeddings in groups of type Dm all correspond to a usual decompo-
sition of an orthogonal space into a direct sum of non-degenerate subspaces,
each corresponding to the natural module of one of the factors of the sub-
group X, with one exception: the embedding Bn A1 < Dn+3 arises from a
158 Maximal subgroups of classical type simple algebraic groups

Table 18.1 Irreducible triples

X<Y V |X V |Y conditions
Cn < A2n−1 , n ≥ 2 aω1 aλ1 a≥2
Cn < A2n−1 , n ≥ 2 aωj + bωj+1 , aλj + bλj+1 a+b=p−1>1
j<n a = 0 if j = n − 1
Bn < A2n , n ≥ 3 ωj , 2 ≤ j < n λj p=2
Bn < A2n , n ≥ 2 2ωn λn p=2
Dn < A2n−1 , n ≥ 4 ωj , 2 ≤ j < n − 1 λj p=2
Dn < A2n−1 , n ≥ 4 ωn−1 + ωn λn−1 p=2
An < A(n2 +n−2)/2 , n ≥ 3 ω1 + ω3 λ2 p=2
An < A(n2 +3n)/2 , n ≥ 2 2ω1 + ω2 λ2 p=2
Bn < Dn+1 , n ≥ 3 aωn aλn , aλn+1 a>0
Bn < Dn+1 , n ≥ 3 aωi + bωn aλi + bλj a + b + n ≡ i (p)
1≤i<n j = n, n + 1 a=0=b
Bn−j Bj < Dn+1 , n ≥ 2 ωn−j + ωn λn , λn+1
X → Bn−j Bj < Dn+1 , ωn−j + ωn λn , λn+1 p ≥ 5 if
n≥2 some πi (X) = A1
Bn A1 < Dn+3 , n ≥ 2 ω
nj+ 3ωn+1 λn+2 , λn+3 p≥5
X →  B n1 · · · B nj < Dm , i=1 ωn1 +...+ni λm−1 , λm p=2
m = n1 + . . . + nj + 1 n−1 n−1 n−1
Dn < C n ji=1 ai ωi + an−1 ωn i=1 ai λ i p = 2, i=2 ai = 0
X →  B n1 · · · B nj < B m , i=1 ωn1 +...+ni λm p=2
m = n1 + . . . + nj

decomposition of the 2(n+3)-dimensional space into a direct sum of the nat-

ural module for the Bn and the irreducible module with p-restricted highest
weight 4ωn+1 (the five-dimensional irreducible p-restricted module for A1 ).
Finally, the last two embeddings are maximal rank configurations corre-
sponding to maximal rank subsystem subgroups generated by short roots.
See Theorems 13.12 and 13.14, as well as Proposition 13.15.
(c) The symbol X → M occurs when M is a central product and the
notation means that either X projects surjectively to each of the simple
factors of M or some factor is of type B2 and the projection is an A1 acting
irreducibly on the spin module for the B2 . Moreover, in order to ensure that
V |X is irreducible, it may be necessary for the projections to involve field
twists.
(d) The specified list of additional triples comprises approximately 45
examples, depending on how one counts the examples arising from graph
automorphisms of Y .
(e) In [80], there were three question marks in the final result, due to the
absence of an existence proof for particular embeddings in the exceptional
 18.2 Maximal subgroups of the classical algebraic groups 159

groups F4 and E6 . The existence was later established in [81] and hence
there are no remaining question marks in the above result.
(f) The above result in case Y is of type Am and char(k) > 0 was estab-
lished independently by Suprunenko in [77].
(g) Note that there are essentially three types of examples:
(I) Positive characteristic analogues of characteristic 0 configurations. For
example, in characteristic 0, the naturally embedded X = Sp2n in
Y = SL2n = SL(W ) acts irreducibly on the symmetric powers S a (W )
for all a; in characteristic p > 0, X acts irreducibly on S a (W ) for all
1 < a < p. This is the first family of examples in the table.
(II) For a fixed embedding X < Y , families which exist only in positive
characteristic, for each positive characteristic p. For example, taking
X and Y as in (I), the second family of examples in the table has no
characteristic 0 analogue.
(III) Examples which occur for a fixed prime characteristic. For example,
the adjoint representation of the group SL3 induces a seven-dimen-
sional irreducible representation when p = 3. (See Theorem 15.20(1).)
Moreover, SL3 preserves a non-degenerate orthogonal form on the
corresponding module (see [11, Table 2]). Hence, in characteristic 3,
we have an embedding X := PGL3 < Y := SO7 = SO(W ). It
turns out that the symmetric square S 2 (W ) is irreducible for Y when
char(k) = 0, but has a 21-dimensional quotient when char(k) = 3.
One can show that the restriction of this quotient module to the sub-
group X is an irreducible kX-module, hence giving an example. This
particular example, which is unique with respect to one further con-
dition (see Proposition 18.9 below) is among the isolated examples
mentioned in the statement of the theorem.
The relatively short length of this list of triples justifies the following
statement: If ρ : H → SL(V ) is a rational irreducible, tensor indecomposable
representation of a simple algebraic group H, then, most of the time, ρ(H)
is maximal among closed connected subgroups of G = Cl(V ), the smallest
classical group containing ρ(H). In particular, the following result is deduced
directly from the classification given by Theorem 18.7 (see [60, Cor.4]).
Corollary 18.9 Let H, ρ and G be as above. If Y is a closed connected sub-
group of G with ρ(H) < Y < G, then with just two exceptions, Y is maximal
among closed connected subgroups of G. The exceptions are as follows:
(1) when p = 3, there is an embedding of irreducible subgroups A2 < G2 <
B3 < SO27 , and
160 Maximal subgroups of classical type simple algebraic groups

(2) when p = 2, there is an embedding of irreducible subgroups D4 < C4 <

F4 < Cl(V ), where dim V = 26.

Remark 18.10 Note that the exception of (2) above is not listed in [60,
Cor. 4]. Moreover, the precise statement concerning the type of the classical
group Cl(V ) is not at this time a published result. The three groups do fix
a symplectic form on V ; see Remark 16.2.

We conclude this section with a discussion of how the results of this chapter
are applied to determine the triples (X, Y, V ). As a first step we classify the
triples where X is the group SL2 or PGL2 . We will apply the following
lemma.

Lemma 18.11 Let H = SL2 , with maximal torus T , and let V be a ra-
tional irreducible kH-module. Then the T -weight spaces of V are all one-
dimensional.

Proof Let α be a generator of the root lattice of H with respect to T and

λ ∈ X(T ) the corresponding fundamental dominant weight. Let µ ∈ X(T )
be the highest weight of V . With the p-adic expansion µ = a1 q1 λ+· · ·+at qt λ,
for some 1 ≤ ai < p and for distinct p-powers q1 , · · · , qt , we have

V = L(a1 λ)(q1 ) ⊗ · · · ⊗ L(at λ)(qt )

by Steinberg’s tensor product theorem (Theorem 16.12). Moreover, Exer-

cise 20.18(d) shows that the weights of the (ai + 1)-dimensional module
L(ai λ) are ai λ, ai λ − α, . . . , ai λ − ai α, and hence the weights of L(ai λ)(qi )
are
ai qi λ, qi (ai λ − α), . . . , qi (ai λ − ai α),

and each is of multiplicity 1.

Now a basis of V consisting of weight vectors is obtained by tensoring
weight vectors of the L(ai λ)(qi ) . The weight vectors thus obtained have
weights of the form

"
t "
t
qi (ai λ − mi α) = µ − mi qi α,
i=1 i=1

for some 0 ≤ mi ≤ ai . Two such weights are equal only if the coefficients of
q1 , . . . , qt are the same. Thus, there is a one-dimensional space of vectors of
weight µ − (m1 q1 + · · · + mt qt )α, for each choice of mi with 0 ≤ mi ≤ ai ,
which establishes the result.
 18.2 Maximal subgroups of the classical algebraic groups 161

For the remainder of the section we assume the triple (X, Y, V ) satisfies
the conditions (i)*, (ii), (iii)*, (iv) and (v) above. Fix a maximal torus TX
of X and a maximal torus TY of Y with TX ≤ TY . Thus, when X = SL2 or
PGL2 , Lemma 18.11 shows that V must have one-dimensional weight spaces
with respect to TY as well. This leads to a short list of possibilities for λ,
the highest weight of V ; for example, if Y has type Bn , λ = λ1 or λn , or
if Y has type E6 , then λ = λi for i = 1 or 6. At this point, case-by-case
considerations lead to two isolated examples, one in G2 and another in D5 ,
as well as the configurations in Table 18.1 where X is diagonally embedded
in a central product of a certain number of copies of A1 , when p = 2. See
[60, Thm. 7.1] for the proof.
Now one is in a position to proceed by induction on rk(X). In order to
do so, we will want to consider the action of a Levi subgroup of X on the
irreducible module V . Choose a parabolic subgroup PX of X; by Corol-
lary 17.12, there exists a parabolic subgroup PY of Y , with PX ≤ PY and
QX := Ru (PX ) ≤ QY := Ru (PY ). Let LX , respectively LY , be Levi sub-
groups of PX , respectively PY . The following result plays a crucial role in
applying the induction hypothesis:

Proposition 18.12 With notation as above, we have that the fixed point
spaces V QX and V QY are equal.

Proof As LX ≤ PY , LX stabilizes V QY ≤ V QX . But by Proposition 16.3,

we know that V QX is an irreducible kLX -module, and since V QY = 0 by
Proposition 2.9, we get the result.

With PX and PY as above, write [LY , LY ] = L1 · · · Lt , a central product of

simple algebraic groups and let πi be the projection of [LY , LY ] onto Li . Then
by Corollary 18.2 the irreducible LY -module V QY (which is equal to the
irreducible LX -module V QX ) is a tensor product M1 ⊗ · · · ⊗ Mt , where Mi is
an irreducible kLi -module. By Proposition 16.3, Mi has a restricted highest
weight. Moreover, the closed subgroup πi ([LX , LX ]) must act irreducibly
on Mi for all i. Choose PX such that [LX , LX ] is simple. Then for each
i such that Mi is non-trivial and πi ([LX , LX ]) < Li , we obtain a triple
(π([LX , LX ]), Li , Mi ) which satisfies the hypotheses (i), (ii), (iii)*. In fact,
Proposition 12.14 shows that the condition (i)* is satisfied, since Y , and
so [LY , LY ], is simply connected. As rk([LX , LX ]) < rk(X), our inductive
hypothesis applies and we will have a list of these smaller rank configurations.
If we choose PX to be a maximal parabolic subgroup of X, the induction will
determine the highest weight of V |X , in terms of the highest weight of the
kY -module V , up to one missing coefficient of the fundamental dominant
162 Maximal subgroups of classical type simple algebraic groups

weight corresponding to the unique simple root of the root system of X

which is not in the root system of LX .
In case Y is a classical type algebraic group, one also considers the action
of X on W , the natural module for Y . Here one shows that either W |X is
again an irreducible kX-module or the highest weight of the kY -module V is
explicitly known (see [60, 5.1]). In the latter case, it is fairly straightforward
to then determine the groups X which can act irreducibly on the known
module V .
Various further techniques come into play to determine explicitly the high-
est weight of the kY -module V . Dynkin’s methods differ significantly from
those used in [60] and [80]. In characteristic 0, representations of reductive
groups are completely reducible and the dimensions of irreducible modules
and of the weight spaces in these modules can be calculated (see the remarks
at the end of Section 15.2), while in characteristic p > 0, representations are
not completely reducible, the extension theory of the simple modules is not
completely understood and the dimensions of the irreducible modules can at
best be bounded above and below.
At this point, we have determined which of the members of the class S k
are in fact maximal in the smallest classical group Cl(V ) containing them.
For if H ∈ S k is not maximal, then H < Y for some Y ∈ C6 or Y ∈ S k .
The first possibility is ruled out by the minimality of Cl(V ) and the second
possibility occurs on the list of triples. To decide the type of the minimal
classical group containing H, if p = 2, we can apply Proposition 16.1 and
[73, Lemma 79].
What remains now is comparatively easy; we must decide which of the sub-
groups H in the families C1 , . . . , C4 , C6 are indeed maximal. This is straight-
forward when one applies the structure theory of G and the known action of
H on V . It turns out that the majority of the subgroups in C1 ∪C2 ∪C3 ∪C4 ∪C6
are indeed maximal. We give the result in the cases SL(V ) and Sp(V ), the
orthogonal case being slightly more technical; we refer the reader to [60,
Thm. 3] for a discussion of the connected case.

Proposition 18.13 For Cl(V ) ∈ {SL(V ), Sp(V )}, the positive-dimensio-

nal members in C1 ∪ . . . ∪ C6 which are maximal inside Cl(V ) are as given in
Tables 18.2 and 18.3.

Here, Pm denotes the parabolic subgroup corresponding to the set I = S \

{sm }, where the simple reflections are numbered as in the Dynkin diagrams
in Table 9.1. The subgroup H of G = Cl(V ) is obtained as the intersection
with G of the normalizer in GL(V ) of the group in the second column.
 18.2 Maximal subgroups of the classical algebraic groups 163

Table 18.2 Maximal subgroups in C1 ∪ . . . ∪ C4 ∪ C6 in SLn

class of H structure conditions rkss (H)

C1 Pm 1≤m≤n−1 n−2
C2 GLm  St n = mt, t ≥ 2 t(m − 1)
C4 GLn1 ⊗ GLn2 n = n1 n2 , 2 ≤ n1 < n2 n1 + n2 − 2
(⊗ti=1 GLm ).St n = mt , m ≥ 3, t ≥ 2 t(m − 1)
C6 Spn n even n/2
SOn p=2 n/2

Table 18.3 Maximal subgroups in C1 ∪ . . . ∪ C4 ∪ C6 in Sp2n , n ≥ 2

class of H structure conditions rkss (H)

C1 Pm 1≤m≤n n−1
Sp2m × Sp2(n−m) 1 ≤ m < n/2 n
C2 Sp2m  St n = mt, t ≥ 2 n
C3 GLn .2 p=2 n−1
C4 Sp2n1 ⊗ SOn2 p = 2, n = n1 n2 , n2 ≥ 3 n1 + n2 /2
(⊗ti=1 Sp2m ).St 2n = (2m)t , t ≥ 3, pt odd tm
C6 GO2n p=2 n

Proof First consider the subgroups H = Stab(X) of the family C1 . If X ≤ V

is a totally isotropic subspace, then by Proposition 12.13, H is a parabolic
subgroup of G. By Proposition 12.2 any subgroup K ≤ G with H ≤ K
is also a parabolic subgroup and the maximal proper parabolic subgroups
are precisely those corresponding to subsets of ∆ of the form ∆ \ {α}, for
some simple root α. Conversely, again by Proposition 12.13 these subgroups
are precisely stabilizers of totally isotropic subspaces. For G = Sp2n , the
subgroups H = Sp2m × Sp2(n−m) have maximal semisimple rank, so by
Theorem 13.14 and Proposition 13.15, we see that the only possible proper
containment of H in a subgroup of Sp(V ) would occur when p = 2 and the
subgroup is of type Dn . But again by Theorem 13.14, we see that there is no
such subgroup of Dn . Thus the listed members of C1 are certainly maximal.
On the other hand, since the members of all other families act irreducibly
on the natural module, none of them can be contained in any member of C1 .
Secondly, the members H of C2 are maximal rank subgroups and as above
one determines the possible maximal rank overgroups. Then one argues that
none of these contain the full normalizer of H ◦ and so H is maximal. The
members of C3 have a simple component whose rank is larger than the rank of
164 Maximal subgroups of classical type simple algebraic groups

all simple components of any member of C2 . The members K of the remaining

classes are such that K ◦ acts irreducibly on V , hence none of these can lie
in a member of C2 .
For the members of C3 , one notes again that these are of maximal rank.
This time however, there is a maximal rank overgroup, namely the maximal
rank Dn subgroup of Sp(V ), by Proposition 13.15. Hence we must exclude
p = 2 in order to obtain a maximal subgroup. Now since the connected
component of a C3 subgroup acts reducibly on V , no member of the remaining
classes can be contained in such a subgroup.
The groups H ∈ C4 fix a non-degenerate bilinear form of the same type
as G, by Proposition 18.3. So they aren’t contained in members of class C6 ,
except in the following cases: when G = SLn and m = 2, then SL2 = Sp2 fixes
a skew-symmetric form, so H lies in a symplectic or orthogonal subgroup of
G, by Proposition 18.3; similarly, when G = Sp2n and pt is even, then H fixes
a quadratic form; and when G is symplectic and p = 2, then Sp2n1 · SOn2 is
contained in the group SO2n by Proposition 18.3(c). We must now determine
whether H is contained in a member Y of S k . If this is the case, then the
triple (H ◦ , Y, V ) is covered by Theorem 18.7. In fact, by inspection one sees
that the only candidate is the embedding Bn A1 < Dn+3 of Table 18.1. But
clearly in that case, the subgroup H ◦ does not act via the tensor product of
the natural modules of the two factors on V .
Finally, the groups in C6 are such that H ◦ acts tensor indecomposably,
hence do not lie in any member of C4 .

Remarks 18.14 (a) The above considerations, based upon the reduction
theorem 18.6, do not treat the odd-dimensional orthogonal groups defined
over fields of characteristic 2. Indeed, while these are isomorphic as abstract
groups to the symplectic groups acting on a space of dimension one less,
Theorem 9.13 and the identification of the different simple algebraic groups
with certain classical groups (Table 9.2) shows that the groups SO2n+1 and
Sp2n are not isomorphic as algebraic groups. (See Exercise 20.25 for further
details.) Nevertheless the natural map (homomorphism of abstract groups)
ϕ : SO2n+1 → Sp2n is indeed a bijective morphism of algebraic groups. Then
the fact that ϕ is continuous, together with Proposition 1.5, shows that it
suffices to determine the maximal closed positive-dimensional subgroups of
Sp2n in order to determine those of SO2n+1 .
(b) The classification of the maximal positive-dimensional closed sub-
groups of the simple algebraic groups of types An , Bn , Cn and Dn fol-
lows from the above considerations. Let Gsc be a simply connected simple
algebraic group with the given root system. Then there exists an isogeny
 18.2 Maximal subgroups of the classical algebraic groups 165

π : Gsc → Cl(V ), for one of the classical groups Cl(V ). The maximal closed
positive-dimensional subgroups of Gsc are then inverse images of those of
Cl(V ). Moreover, the maximal closed positive-dimensional subgroups of any
homomorphic image ϕ(Gsc ) of Gsc are images of the corresponding sub-
groups of Gsc , since by Proposition 1.5 images of closed subgroups under
morphisms are closed.
 19
Maximal subgroups of exceptional type
algebraic groups

We now consider the case of the exceptional type algebraic groups. The
classification of the maximal closed connected subgroups was obtained by
Dynkin [23] in the case where char(k) = 0. The case of positive characteristic
is covered by three lengthy articles of Seitz [61] and Liebeck–Seitz [48, 53].

19.1 Statement of the result

We require some further notation before stating the results. In what follows
we write AGLn for the affine general linear group (semi-direct product of
GLn with the group of translations). By abuse of notation, we will write Φ for
a semisimple algebraic group with root system of type Φ; if the subgroup is
a subsystem subgroup corresponding to the p-closed subset Ψ ⊂ Φ, we will
write Ψ for those subsystem subgroups generated by long root subgroups
and Ψ̃ for subsystem subgroups generated by short root subgroups. We also
write W (Ψ) for the Weyl group of Ψ. Finally, we will continue to adopt the
convention p = ∞ when char(k) = 0.
The first result classifies the subgroups which are maximal among proper
closed connected subgroups.

Theorem 19.1 Let G be an exceptional algebraic group defined over an

algebraically closed field of characteristic p. Let X < G be a closed subgroup.
Then X is maximal among proper closed connected subgroups of G if and
only if one of the following holds:

(1) X is a maximal parabolic subgroup.

 19.1 Statement of the result 167

(2) X is a maximal rank subsystem subgroup as in the table below:

G X
G2 A1 Ã1 , A2 , Ã2 (p = 3)
F4 B4 , A1 C3 (p = 2), C4 (p = 2), A2 Ã2
E6 A1 A5 , A2 A2 A2
E7 A1 D6 , A7 , A2 A5
E8 D8 , A1 E7 , A8 , A2 E6 , A4 A4

(3) X and G are as in the following tables:

G X simple
G2 A1 (p ≥ 7)
F4 A1 (p ≥ 13), G2 (p = 7)
E6 A2 (p = 2, 3), G2 (p = 7), C4 (p = 2), F4
E7 A1 (2 classes, p ≥ 17, 19 respectively), A2 (p ≥ 5)
E8 A1 (3 classes, p ≥ 23, 29, 31 respectively), B2 (p ≥ 5)

G X semisimple, non-simple

F4 A1 G2 (p = 2)
E6 A2 G2
E7 A1 A1 (p = 2, 3), A1 G2 (p = 2), A1 F4 , G2 C3
E8 A1 A2 (p = 2, 3), G2 F4

The second result classifies the maximal closed positive-dimensional sub-

groups.

Theorem 19.2 Let G be an exceptional algebraic group of adjoint type,

defined over an algebraically closed field of characteristic p. Let M < G be
a closed subgroup. Then M is maximal among positive-dimensional closed
subgroups of G if and only if one of the following holds:

(1) M is a maximal parabolic subgroup.

(2) M = NG (X) where X is connected reductive of maximal rank and
the pair (X, M/X) is as in the table below, where Ti denotes an i-
dimensional subtorus of G:
168 Maximal subgroups of exceptional type algebraic groups

G X M/X
G2 A1 Ã1 , A2 , Ã2 (p = 3) 1, Z2 , Z2
F4 (p = 2) B4 , D4 , A1 C3 , A2 Ã2 1, S3 , 1, Z2
F4 (p = 2) B4 , C4 , D4 , D̃4 , A2 Ã2 1, 1, S3 , S3 , Z2
E6 A1 A5 , (A2 )3 , D4 T2 , T6 1, S3 , S3 , W (E6 )
E7 A1 D6 , A7 , A2 A5 , (A1 )3 D4 , 1, Z2 , Z2 , S3 ,
7
(A1 ) , E6 T1 , T7 PSL3 (2), Z2 , W (E7 )
E8 D8 , A1 E7 , A8 , A2 E6 , 1, 1, Z2 , Z2 ,
2
(A4 ) , (D4 ) ,2
Z 4 , S3 × Z 2 ,
(A2 )4 , (A1 )8 , T8 GL2 (3), AGL3 (2), W (E8 )

(3) G = E7 , p = 2 and M = (Z22 × D4 ).S3 .

(4) G = E8 , p = 2, 3, 5 and M = A1 × S5 .
(5) M = NG (X) for X as in (3) of Theorem 19.1.
(6) G = E8 , p = 2, M = NG (X), where X = A1 G2 G2 , with A1 G2 maximal
closed connected in F4 , and M/X = Z2 .
Remark 19.3 When M is as in (5) of the above theorem, the index |M : X|
is 1 or 2. In all cases where X has a factor of type A2 , M induces a non-trivial
graph automorphism of this factor.

19.2 Indications on the proof

The proofs of the above theorems in positive characteristic were established
in two phases: the first phase (in [61] and [48]) proved Theorems 19.1 and 19.2
under some mild assumptions on char(k), while the second phase (in [53])
completed the proof of these theorems. The bulk of the work in the second
phase was in showing that no “new” examples occur in small characteristics.
We will not go into the second phase here, but rather outline the strategy
used in the first phase under the additional assumption that X is connected.
We assume the exceptional group G to be simply connected, since any
maximal subgroup of G must contain Z(G) and hence under the natural
isogeny of Proposition 9.15, we obtain an image which is a maximal sub-
group. (Clearly, any maximal subgroup of an image of G is the image of a
maximal subgroup of G.)
We establish a first easy reduction result:
Proposition 19.4 In the above setting, let X be a proper closed connected
 19.2 Indications on the proof 169

subgroup of G which is maximal among such subgroups. Then either X is

semisimple or X is a maximal proper parabolic subgroup of G.
Proof If X is not reductive, then by Corollary 17.14, X is a parabolic
subgroup. So we may assume X is reductive. If Z(X)◦ = 1, it is a torus by
Proposition 6.20, and Proposition 12.10 then implies that CG (Z(X)◦ ) is a
proper Levi subgroup and hence X lies in a proper parabolic subgroup of G.
The result then follows from the maximality assumption.
Now by Proposition 12.2 the maximal proper parabolic subgroups of G
correspond to maximal proper subsets of a base of the root system of G.
Thus we will assume from now on that X is a semisimple group. In addition,
we have the following information about Z(X):
Lemma 19.5 Let X be a semisimple maximal proper closed connected
subgroup of G. Then either Z(X) ≤ Z(G) or X is a maximal rank subgroup
of G.
Proof As Z(X) ≤ CX (T ) = T for some maximal torus T of X, Z(X)
consists of semisimple elements. Now suppose Z(X) ≤ Z(G) and let s ∈
Z(X) \ Z(G). Then by maximality X = CG (s)◦ , which by Proposition 14.1
is a proper maximal rank connected subgroup of G.
As maximal rank semisimple subgroups of G are well-understood (see
Chapter 13), it is straightforward to determine which of these subgroups are
maximal, leading to the cases in Theorem 19.1(2). We shall therefore assume
henceforth that X is semisimple of rank strictly less than rk(G). We can then
show the following:
Lemma 19.6 Consider Lie(X) ≤ Lie(G). Then:
(a) X = StabG (Lie(X))◦ , and
(b) CLie(G) (X) := {v ∈ Lie(G) | Ad (x)v = v for all x ∈ X} ≤ Z(Lie(G)).
Proof of (a) In the adjoint action of X on Lie(G), X stabilizes Lie(X) and
so X ≤ StabG (Lie(X))◦ . By maximality either X = StabG (Lie(X))◦ or
StabG (Lie(X))◦ = G. In the latter case, we apply Theorem 15.20 to see that
either Lie(X) ⊆ Z(Lie(G)) or Lie(X) contains all root subspaces for short
roots of G. In the first case, dim(Lie(X)) = 1, which is impossible as X is
semisimple. In the second case, we are reduced to two possible configurations:
G is of type G2 , p = 3 and dim(Lie(X)) = 7 (see Example 15.21(1)) or G is
of type F4 , p = 2 and dim(Lie(X)) = 26. But there is no semisimple group
of dimension 7, respectively 26 and of rank less than 2, respectively 4. Hence
X = StabG (Lie(X))◦ as required for (a).
170 Maximal subgroups of exceptional type algebraic groups

For the proof of (b), one must use the fact that the centralizer in G of a
semisimple, or nilpotent, element in Lie(G) contains a maximal torus of G,
or has a non-trivial unipotent radical. See [61, 1.3] for details.
With these lemmas in place, and after fixing some additional notation,
we introduce the main tool used in the study of the embedding X < G.
Let T be a maximal torus of G, TX a maximal torus of X with TX ≤ T
and ΦX the corresponding root system of X. Let BX be a Borel subgroup
of X containing TX and let ∆X be the corresponding base of ΦX . For α ∈
ΦX , let α∨ : Gm → TX be the coroot as in Lemma 8.19. For c ∈ k × ,

set γ(c) = α∨ (c), the product taken over all roots α ∈ Φ+ X . Then im(γ)
is a one-dimensional torus in X and we shall first consider the action of

this torus on Lie(X). Let Lie(X) = Lie(TX ) ⊕ α∈ΦX Lie(X)α and choose
vα ∈ Lie(X)α \ {0}.

Lemma 19.7 With the above notations we have:

(a) γ(c)h = h for all h ∈ Lie(TX );

(b) γ(c)vβ = c2 vβ for all β ∈ ∆X and c ∈ k × .

Proof For (a), we simply note that im(γ) ≤ TX and TX acts trivially on
itself via conjugation, hence TX fixes pointwise its Lie algebra. For (b), let

β ∈ ∆X , c ∈ k × . Then we note that γ(c)vβ = ( α∨ (c))vβ = cr vβ where
 ∨
 ∨
α∈Φ+ β, α . Setting ρ = 2 β, 2ρ. By
1
r = α∈Φ+ α , we have r =
X X
Exercise A.1 {α∨ | α ∈ Φ+ X } forms a positive system in the root system
∨
ΦX . Now by Exercise 20.26 we know that ρ is the sum of the fundamental
dominant weights relative to the abstract root system Φ∨ X and ΦX = ΦX .
∨∨

So using the isomorphism i of Exercise 10.35, we see that r = 2β, ρ = 2 as

claimed.

Finally we indicate one of the key tools in the analysis which follows:

Proposition 19.8 There exists a base ∆ of the root system of G with

respect to T such that γ(c) acts as c0 or c2 on each of the root spaces Lie(G)α ,
for α ∈ ∆.

About the proof For β ∈ Φ, the root system of G with respect to T , let
rβ ∈ Z such that γ(c)v = crβ v for all v ∈ Lie(G)β ; that is, rβ = β, γ. One
first shows that all TX -weights on Lie(G) are integral linear combinations
of elements of ΦX . See [61, 2.3] for this proof. Then apply Lemma 19.7 to
see that the rβ must be even integers. Choosing an appropriate ordering on
X(T ) one finds a base ∆ of Φ such that rβ ≥ 0 for all β ∈ ∆ and so for all
 19.2 Indications on the proof 171

β ∈ Φ+ (see Section A.1). Now set

J := T, Uβ , U−β | β ∈ ∆ and rβ = 0 or 2.
Then J is the Levi factor of a suitable parabolic subgroup of G and it is
clear that
J = T, Uβ , U−β | β ∈ Φ and rβ = 0 or 2.
Lemma 19.7 and Theorem 8.17 imply that Lie(X) ≤ Lie(J); so
Z(J)◦ ≤ CG (Lie(X))◦ ≤ StabG (Lie(X))◦ = X
by Lemma 19.6. In particular Z(J)◦ ≤ CX (Lie(X))◦ . Now CX (Lie(X)) =
ker(Ad X ) and as we are assuming X to be semisimple, Exercise 10.32 implies
that ker(Ad X ) = Z(X). Hence CX (Lie(X))◦ = 1, so Z(J)◦ = 1 and J = G
by Proposition 12.6 as required.
We can now explain the strategy of the proof of Theorem 19.1. By Proposi-
tion 19.8, there are 2|∆G | possibilities for the |∆G |-tuple of integers (rβ )β∈∆G
as defined in the proof of loc. cit. For each of these possibilities, one can
list the precise weights of im(γ) on Lie(G). Now there will be a maximum

such weight, indeed given by β∈∆G nβ rβ , where the highest root in ΦG

is β∈∆G nβ β. As there are a finite number of irreducible kX-modules for
which the im(γ) weights are less than or equal to this highest weight, one ob-
tains a collection of possible X-composition factors of Lie(G)|X . Moreover,
in many cases the precise dimensions of these irreducible modules are known,
or at least upper and lower bounds for their dimensions. Thus, one has a
finite list of possible composition series (up to permutation of the factors)
for the module Lie(G)|X . For each of these composition series, one computes
the precise set of weights of im(γ) on Lie(G) and compares this set of weights
with the given set of weights obtained from the choice of (rβ )β∈∆G . For a
fixed pair of groups X, G, one thus reduces to a finite number of choices for
(i) the tuples (rβ )β∈∆G , and
(ii) the corresponding set of composition factors of Lie(G)|X .
The next part of the proof is that which required the assumptions on
char(k) in [61] and [48]. Many of the configurations remaining after the
above analysis correspond to a composition series containing a large number
of trivial composition factors. If char(k) is large enough, this often con-
tradicts the conclusion of Lemma 19.6(b). The consideration of the small
characteristic configurations required a detailed analysis of the remaining
possible X-composition series of Lie(G).
 20
Exercises for Part II

Let k be an algebraically closed field of characteristic p ≥ 0. We will take

the numbering of Dynkin diagrams of irreducible root systems as given in
Table 9.1.

Exercise 20.1 (Existence of graph automorphisms)

(a) Show how to reduce the proof of Theorem 11.12 on the existence of
graph automorphisms to the case of simple groups of simply connected
type.
(b) Verify the details of the proof for type SLn , n ≥ 3.
(c) Show that a suitable element of GO2n induces a non-trivial graph auto-
morphism of SO2n , n ≥ 2.

[Hint: For (c) consider the element given in Example 22.9(2).]

Exercise 20.2 Let G be a group with a BN-pair, with W = N/(B ∩ N )

generated by a set of involutions S. For w ∈ W write (w) for the length of
a shortest expression w = s1 · · · sr with si ∈ S. Show the following:

(a) If s ∈ S, w ∈ W with (ws) ≥ (w) then B ẇB · B ṡB ⊆ B wsB.

˙
(b) If s ∈ S, w ∈ W with (ws) ≤ (w) then B ẇB · B ṡB has non-empty
intersection with B ẇB.
(c) If (ws) < (w), then ṡ ∈ B ẇ−1 B ẇB.

[Hint: For (a), reason by induction on (w) and apply Theorem 11.17; for (b)
apply (a). The result of (c) is a corollary of (a) and (b).]

Exercise 20.3 We continue using the notation of Exercise 20.2. For w ∈ W

let w = s1 · · · sr , with si ∈ S, be a shortest expression, so that r = (w), and
let J := {s1 , . . . , sr } ⊆ S. Show the following:
 Exercises for Part II 173

(a) We have B, ẇ = B, ẇ−1 B ẇ = PJ .

(b) All overgroups of B in G are of the form PI for some I ⊆ S.
(c) Conclude that all PI are self-normalizing and that distinct PI cannot be
conjugate.
[Hint: For (a), apply the preceding exercise. For (b) let P be a subgroup of G
containing B. Any element in P can be written in the form g = bẇb , with b, b ∈ B,
w ∈ W , by Theorem 11.17. We conclude that ẇ ∈ P . Thus, P is a union of double

cosets P = w∈M B ẇB for some M ⊆ W . Now apply (a).]

Exercise 20.4 Let G be connected reductive, I ⊆ S and LI the corre-

sponding standard Levi subgroup of G. Then there is a natural isomorphism
∼
NG (LI )/LI −→ NW (WI )/WI .
[Hint: Let g ∈ NG (LI ), then T g is a maximal torus in LI , hence LI -conjugate
to T . Thus gl ∈ NG (T ) for some l ∈ LI . Show that this defines the required
isomorphism.]

Exercise 20.5 Let G be a connected reductive algebraic group with root

datum (X, Φ, Y, Φ∨ ) with respect to the maximal torus T .
(a) Let T  ≤ T be a subtorus with root datum (X  , ∅, Y  , ∅) (see Exam-
ple 9.12). Show that the Levi subgroup CG (T  ) has corresponding root
datum (X, Ψ, Y, Ψ∨ ) with Ψ := Φ ∩ Ann(Y  ).
(b) Let L be a standard Levi subgroup of G with root datum (X, Ψ, Y, Ψ∨ ).
Show that the torus Z(L)◦ has the root datum (X  , ∅, Y  , ∅), where Y  =
Ann(Ψ), X  = X/Ann(Y  ).
Here, for Z ≤ X we set Ann(Z) := {γ ∈ Y | χ, γ = 0 for all χ ∈ Z}.
Exercise 20.6 Let G be a simple algebraic group with simply laced Dynkin
diagram. Let ∆ denote a base of its root system. Show that the isomorphisms
uα : Ga → Uα , α ∈ ∆, may be chosen such that [uα (t), uβ (u)] = uα+β (±tu)
whenever α, β ∈ ∆ with α + β ∈ Φ.
[Hint: The statement holds for SL3 by Example 11.9. Now use Example 12.9 to
show it for SL4 and then induction for the general case.]

Exercise 20.7 Prove the following generalization of Proposition 12.14:

Let G be semisimple, LI ≤ G the standard Levi subgroup corresponding to
I ⊆ S. If gcd(Λ(G), Λ(ΦI )) = 1 then [L, L] is of simply connected type.
Exercise 20.8 Let G be semisimple with a maximal torus T with Weyl
group W , H ≤ G a subsystem subgroup normalized by T with Weyl group
WH . Then there is a natural isomorphism NG (HT )/HT ∼
= NW (WH )/WH .
[Hint: Apply the arguments from Exercise 20.4 to the subgroup HT .]
174 Exercises for Part II

Exercise 20.9 Let Φ be one of the indecomposable root systems of type

An , Bn , Cn or Dn .

(a) Determine the proper closed subsystems of Φ of rank equal to the rank
of Φ.
(b) Let G = Sp2n , a simple algebraic group of type Cn , acting on the nat-
ural 2n-dimensional module V preserving a symplectic bilinear form.

For n1 , . . . , nt ∈ N with ni = n, there exist subspaces Vi ≤ V with
dim Vi = 2ni and V = V1 ⊥ · · · ⊥ Vt . Hence the group Sp2n1 ×· · ·×Sp2nt
is naturally embedded in G; let H denote the corresponding closed sub-
group of G.
(i) Show that H is a maximal rank subgroup of G, for all characteris-
tics char(k).
(ii) Find a closed subsystem Ψ of the root system of G such that H is
the subsystem subgroup associated to this subsystem.

[Hint: (a) You should find no example for An , n − 1 for type Bn , (n − 1)/2 for
type Cn and (n − 2)/2 for type Dn .]

Exercise 20.10 Show that the unipotent element

⎛ ⎞
1 1 0 0
⎜0 1 0 0⎟
u := ⎜
⎝0
⎟ ∈ Sp4
0 1 1⎠
0 0 0 1

of G = Sp4 over a field of characteristic 2 is not contained in the connected

component CG (u)◦ of its centralizer.

Exercise 20.11 Let G be connected, S ≤ G a torus. Show that the set of

s ∈ S such that CG (s) = CG (S) is dense open in S.
[Hint: Embed G ≤ GL(V ) and decompose V into weight spaces for S. Have a look
at the proof of Corollary 14.10.]

Exercise 20.12 Let G be connected reductive with maximal torus T .

(a) Show that NG (T ) controls G-fusion in T , that is, if s, t ∈ T are G-

conjugate, they are conjugate by an element of NG (T ).
(b) Deduce that there is a natural bijection between the set of semisimple
classes of G and the set T /W of W -orbits on T .

[Hint: For (a), use the uniqueness of expression in the Bruhat decomposition.]
 Exercises for Part II 175

Exercise 20.13 Let G be a simple algebraic group, with maximal torus

T , root system Φ and Weyl group W over the algebraic closure k of a finite
field Fp . Show the following:

(a) There is a (non-canonical) isomorphism e : Qp /Z → k × , where Qp

denote the additive group of rational numbers with denominator prime
to p.
(b) There is an isomorphism of groups

τ : (Y ⊗ Qp )/Y −→ Y ⊗ Qp /Z −→ Y ⊗ k × −→ T,

defined by γ ⊗ r → γ ⊗ e(r) → γ(e(r)), which is W -equivariant with

respect to the natural W -actions on Y and T .
(c) The semisimple conjugacy classes of G are in bijection with W -orbits on
(Y ⊗ Qp )/Y , that is, with elements of (Y ⊗ Qp )/Wa , where Wa = Y.W
denotes the affine Weyl group (of W acting on Y ), see Section B.2.
(d) Conclude that any semisimple element of G is conjugate to some s ∈ T
such that the root system Ψ of CG (s)◦ (as in Theorem 14.2) has a basis
∆1 ⊆ ∆ ∪ {−α0 }.

[Hint: (c) Use Proposition 14.6. (d) By Proposition B.12 every Wa -orbit on Y ⊗
Qp ⊂ Y ⊗ R contains a point x from the fundamental alcove A. Then for s = τ (x),
for α ∈ Φ we have α(s) = 1 if and only if sα , x = 0, that is, if and only if sα∨
fixes x, that is, x lies on the wall of A corresponding to α∨ .]

Exercise 20.14 Assume char(k) = 2 and let s be the image in G = PGL2

of diag(1, −1) ∈ GL2 . Show that CG (s) has two connected components. In
particular, centralizers of semisimple elements in connected groups are not
necessarily connected.

Exercise 20.15 Show that the torsion primes of a reductive group G are
precisely the torsion primes of its root system Φ together with the prime
divisors of the order of its fundamental group.
[Hint: First show that Λ(G) ∼ = Y /ZΦ∨ by applying Hom(−, Z) to the exact se-
quence defining Λ(G). Now, if G ≤ G is a subsystem group with root system Ψ,
the torsion of Y /ZΨ∨ is the torsion of Y /ZΦ∨ together with that of ZΦ∨ /ZΨ∨ .]

Exercise 20.16 Let G be a semisimple algebraic group and s ∈ G semisim-

ple.

(a) Show that CG (s)/CG (s)◦ is isomorphic to a subgroup of ker(π), where

π : Gsc → G denotes the natural isogeny from a simply connected group
with the same root system as G.
176 Exercises for Part II

(b) If s is of finite order, the exponent of CG (s)/CG (s)◦ divides the order
of s.
(c) If s is of finite order prime to | ker(π)| then CG (s) is connected.

[Hint: Fix a preimage ŝ ∈ Gsc of s. For g ∈ CG (s) with preimage ĝ, ĝŝĝ −1 = zŝ
for some z ∈ ker(π). Then g → z = [ĝ, ŝ] is a well-defined group homomorphism
with the required properties. For (b), use that [ĝ, ŝ] is central, so [ĝ, ŝ]n = [ĝ, ŝn ]
for all n.]

Exercise 20.17 Show that the adjoint representation of SLn is irreducible

if and only if char(k) does not divide n.
[Hint: Think about the weights of the adjoint representation and the possible
highest weights of composition factors. This establishes Theorem 15.20(1) in case
G = SLn .]

Exercise 20.18 Let G = SL2 (K), the group of 2 × 2, determinant 1

matrices over an arbitrary (not necessarily algebraically closed) field K. Then
G acts naturally on the polynomial ring K[X, Y ] in two indeterminates X, Y
via


a b
(X i Y j ) = (aX + cY )i (bX + dY )j ,
c d

extended by linearity. Let Vd ⊂ K[X, Y ] be the subspace of homogeneous

polynomials of degree d. We thus have a representation ρ : G → GLd+1 (K)
of G.

(a) Verify that for K of characteristic 0 or for char(K) > d, Vd is irreducible.

(b) Verify that if char(K) = p > 0, X p , Y p  is a G-invariant subspace of Vp .
(c) Show that if K is of characteristic p > 0 and K = F2 , then Vp is not
completely reducible as a KG-module.
(d) Now take K = k, an algebraically closed field of characteristic p ≥ 0. As
weights for the maximal torus consisting of diagonal matrices in G are
all multiples of the weight λ = 12 α (where α is the unique positive root in
the root system for G), the irreducible modules for G are parametrized
by positive integers. Show that Vd  L(dλ) and that if char(k) = p > 0,
all restricted irreducible rational kSL2 -modules are of the form Vd , for
d < p. In particular, SL2 has a unique restricted irreducible rational
representation of dimension m + 1 for all m < p, and the weights of the
corresponding module are m, m − 2, . . . , −(m − 2), −m.

Exercise 20.19 Let G be a connected reductive algebraic group with max-

imal torus T and Weyl group W . Let B be a Borel subgroup containing T and
 Exercises for Part II 177

∆ the corresponding set of simple roots in the root system Φ. Let λ ∈ X(T )
be a dominant weight.

(a) Show that Wλ := StabW (λ) is the subgroup generated by the sα , α ∈ ∆,

such that λ, α∨  = 0; that is Wλ is the Weyl group of the Levi subgroup
corresponding to the subset {α ∈ ∆ | λ, α∨  = 0}.

(b) Let Wλ be as in (a). Show that dim(L(λ)) ≥ µ |W : Wµ |, where the
sum is taken over all dominant weights µ for which the multiplicity of µ
in L(λ) is non-zero.

[Hint: For (a), let v + ∈ L(λ) be a maximal vector with respect to the Borel
subgroup B. Then B ≤ StabG (v + ), so StabG (v + ) is a parabolic subgroup of
G. Now use the structure of parabolic subgroups (or see Corollary A.29). For (b),
use Lemma 15.3 and Proposition 15.8.]

Exercise 20.20 Let char(k) = 2 and let G = SL2 . Fix a maximal torus T
of G, a Borel subgroup B containing T and a simple root corresponding to
the choice of B. Let λ ∈ X(T ) be the unique fundamental dominant weight.
Let F : k → k be the 2-power map.

(a) Show that the adjoint representation of G equips Lie(G) with the struc-
ture of an indecomposable kG-module, having a composition series 0 ⊂
V1 ⊂ Lie(G) = V , with V1 ∼
= L(0) and V /V1 ∼= L(λ)(2) .
(b) Show that V2 , as in Exercise 20.18, is an indecomposable kG-module
with a composition series 0 ⊂ W1 ⊂ V2 , such that W1 ∼ = L(λ)(2) and
∼
V2 /W1 = L(0).

Exercise 20.21 Let G be a simple algebraic group of type G2 defined

over the field k and assume char(k) = 0. Let V = Lie(G), the irreducible
kG-module of highest weight λ2 , the second fundamental dominant weight
with respect to a choice of maximal torus and positive system of roots. (See
Example 15.21(1).) Show that the kG-module V ∧ V ∼ = Lie(G) ⊕ L(3λ1 ).
[Hint: You may use the fact that kG-modules are completely reducible when
1
char(k) = 0 (see [33, 14.3]) and that dim(V (a1 λ1 + a2 λ2 )) = 5! (a1 + 1)(a2 +
1)(a1 + a2 + 2)(a1 + 2a2 + 3)(a1 + 3a2 + 4)(2a1 + 3a2 + 5), which follows from the
Weyl degree formula, see [33, §24].]

Exercise 20.22 Let G be a simple algebraic group of type D4 . For each

of the maximal parabolic subgroups P of G, with unipotent radical Q, show
that each of the internal modules Qj /Qj+1 , for j ≥ 1 is irreducible and
determine their highest weights as modules for the Levi factor. (See also
Example 17.9(3).)
178 Exercises for Part II

Exercise 20.23 Let G = SL3 and let W be the three-dimensional vector

space on which G naturally acts with fixed ordered basis {e1 , e2 , e3 }. Let
BG = StabG (ke1 ⊂ ke1 + ke2 ⊂ V ), a Borel subgroup of G. Set V = ∧2 W ,
a six-dimensional vector space equipped with the natural G-action inherited
from the action of G on W ⊗ W . Let ρ : G → GL6 be the corresponding
rational representation.
(a) Show that im ρ < SL6 .
(b) Let PG = StabG (ke1 ), a proper parabolic subgroup of G, with BG <
PG . Find P ≤ SL6 , a parabolic subgroup such that ρ(PG ) ≤ P and
ρ(Ru (PG )) ≤ Ru (P ).
Exercise 20.24 Let fi be non-degenerate bilinear forms on the finite-
dimensional k-vector spaces Vi , i = 1, 2, and set V := V1 ⊗ V2 . Show the
following:
(a) There is a unique bilinear form f = f1 ⊗ f2 on V defined by
f (v1 ⊗ v2 , w1 ⊗ w2 ) := f1 (v1 , w1 )f2 (v2 , w2 ) for vi , wi ∈ Vi .
(b) The form f is symmetric if and only if f1 , f2 are either both symmetric
or both alternating.
(c) If char(k) = 2 then there is a unique quadratic form Q on V , with
associated bilinear form f , such that Q(v1 ⊗ v2 ) = 0 for all vi ∈ Vi , and
Q is preserved by Sp(V1 ) ⊗ Sp(V2 ).
[Hint: See [44, §4.4].]

Exercise 20.25 Let char(k) = 2 and let V be a three-dimensional vector

space over k, with fixed basis {e, w, f }. Define a quadratic form on V by
Q(ae + bw + cf ) := b2 + ac and let SO3 be the corresponding orthogonal
group, as in Section 1.2. Show that w is the radical of the associated
bilinear form and that ϕ : SO3 → SL(V /w) defines a morphism of algebraic
groups whose image is the group Sp(V /w). Show that ker(ϕ) = 1 and
so ϕ : SO3 → Sp(V /w) is an isomorphism of abstract groups. Now use
Proposition 7.7 to see that ϕ is not an isomorphism of algebraic groups.
(One can also see directly that the inverse is not a polynomial map.)
Exercise 20.26 Let Φ be a root system in a Euclidean space E, with base

∆ ⊂ Φ and corresponding set of positive roots Φ+ . Set ρ = 12 α∈Φ+ α.
Show that ρ is the sum of the fundamental dominant weights with respect
to the base ∆.
[Hint: Use Lemma A.8 to deduce that sβ (ρ) = ρ − β and hence ρ, β ∨  = 1 for all
β ∈ ∆.]
 PART III

FINITE GROUPS OF LIE TYPE

In this part we introduce and study finite analogues of the simple linear alge-
braic groups which were the subject of the first two parts. The construction
of the various classical groups over algebraically closed fields in Section 1.2
generalizes in a straightforward way to give versions over arbitrary base
fields. It is much less obvious how to obtain versions of the simple excep-
tional groups. This was first achieved by Chevalley [15] who showed how to
construct analogues of all simple algebraic groups as automorphism groups
of simple Lie algebras over any base field. Still this approach falls short of
producing all versions in which we will be interested. For example, over an
algebraically closed field there is just one class of non-degenerate orthogo-
nal forms up to similarity in any dimension, while over arbitrary fields there
may be many, with corresponding non-isomorphic isometry groups. Also, the
isometry groups of unitary forms do not arise in Chevalley’s setup.
Shortly after Chevalley’s construction, Steinberg [68] presented a varia-
tion of this by considering fixed points of field automorphisms composed
with algebraic group automorphisms. This allows one to recover the uni-
tary groups, for example. But this wasn’t yet the end of the story. In 1960
M. Suzuki discovered an infinite series of finite simple permutation groups
which at that time did not seem to have any relation with algebraic groups.
It was recognized in the same year by Steinberg how his construction could
be generalized to yield these as subgroups of the four-dimensional symplectic
group over an algebraically closed field of characteristic 2.
We will follow this approach of Steinberg, which now seems to be the best
way to recover via a general construction all of the versions described above:
given a field K, the various analogues of linear algebraic groups are obtained
from algebraic groups over the algebraic closure k of K by means of a Galois
180

descent. The situation becomes particularly simple when K = Fq is a finite

field, since then the absolute Galois group Gal(Fq /Fq ) is profinite cyclic,
topologically generated by the Frobenius automorphism F with respect to
the base field Fq . The finite groups of Lie type then occur as the fixed point
groups of suitable variants of this Frobenius automorphism. More precisely,
they arise as fixed point groups of endomorphisms some power of which is the
standard Frobenius. In most but not all cases, such endomorphisms are ob-
tained by composing the Frobenius with an automorphism of the underlying
algebraic group.
This point of view allows one to answer many questions on the families
of finite groups of Lie type by interpreting them as questions on the F -fixed
points of corresponding structures in algebraic groups and transferring the
results from there.
In Chapter 21 we describe the setting for this method of Galois descent
over finite fields by introducing the notion of Steinberg endomorphism (which
is sometimes also called Frobenius endomorphism or generalized Frobenius
map) and formulate the important theorem of Lang–Steinberg. Chapter 22
is devoted to the study of properties of Steinberg endomorphisms and their
classification. The fixed point groups under such endomorphisms are what
we call the finite groups of Lie type, the central topic of this part.
We begin the structural investigation in Chapter 23 by exhibiting a root
system and root subgroups for these groups. In Chapter 24 we first derive
a Bruhat decomposition from which, using some results from the invariant
theory of finite reflection groups, we compute the orders of the finite groups
of Lie type. We then show that these groups also possess a BN-pair, as do
their algebraic counterparts, which enables us to study simplicity and auto-
morphism groups. The structure of tori, Sylow subgroups, their centralizers
and normalizers is the subject of Chapter 25. In Chapter 26 we investigate
subgroups of maximal rank such as parabolic subgroups, Levi subgroups and
centralizers of semisimple elements.
The last three chapters are devoted to the study of maximal subgroups
of finite groups of Lie type, using the corresponding results for simple al-
gebraic groups in the previous part. We present the reduction theorems of
Aschbacher for the classical groups and of Liebeck–Seitz for the exceptional
groups and comment on the current status of the determination of all max-
imal subgroups.
 21
Steinberg endomorphisms

In this section, we introduce the notion of Steinberg endomorphisms on linear

algebraic groups over the algebraic closure of a finite field and define the finite
groups of Lie type. We present the crucial theorem of Lang–Steinberg and
several of its consequences, which will be essential for most of the results to
come. Throughout, the base field k is the algebraic closure of a finite field of
characteristic char(k) = p.

21.1 Endomorphisms of linear algebraic groups

We start by looking at certain endomorphisms of the algebraic group GLn :
Example 21.1 Let k = Fq where q = pf . The map Fq : k → k, t → tq , is
a field automorphism of k which fixes Fq pointwise. In fact the Galois group
Gal(k/Fq ) is generated (as a profinite group) by this Frobenius automor-
phism.
Letting Fq act on the matrix entries, this induces a group homomorphism
Fq : GLn −→ GLn , (aij ) −→ (aqij ),
with fixed point group
(GLn )Fq := {g ∈ GLn | Fq (g) = g} = GLn (Fq ),
the general linear group over Fq which we will henceforth denote by GLn (q).
We call Fq the standard Frobenius of GLn with respect to Fq .
This example may be generalized as follows: Let V be an affine algebraic
variety over k = Fq defined by a set of polynomials I ⊆ Fq [T1 , . . . , Tn ] with
coefficients in Fq . We then say that V is defined over Fq . The Frobenius
automorphism Fq : k → k, t → tq , of k which acts naturally on k[T1 , . . . , Tn ]
182 Steinberg endomorphisms

via the coefficients, then leaves I invariant. Thus Fq also acts on the set V
of common zeros of I in k n , by
Fq : V → V, (v1 , . . . , vn ) −→ (v1q , . . . , vnq ).
This induced map Fq is called the Frobenius morphism of V with respect to
the Fq -structure given by I. As in the example above, we write
V Fq := V (Fq ) := {v ∈ V | Fq (v) = v}
for the Fq -fixed points of V . Note that, as Fq is induced by an element of
the Galois group of k/Fq , it is a bijective map.
Thus, if a closed subgroup G ≤ GLn is defined by equations over Fq , this
gives rise to a Frobenius morphism Fq : G → G, (aij ) → (aqij ), with respect
to this Fq -structure, which clearly is a morphism of algebraic groups, with
finite fixed point group G(Fq ) = GFq ≤ GLF n = GLn (q). Note, however,
q

that this bijective group homomorphism is not an isomorphism of algebraic

groups.
Still, this approach for the construction of finite fixed point groups is not
yet general enough to cover all situations we want to treat.
Example 21.2 Let’s consider again G = GLn , but now with the endo-
morphism
F : GLn −→ GLn , (aij ) −→ (aqij )−tr .
Thus F is the composite of the previous Fq with the map sending a matrix
to the transpose of its inverse. Note that these two maps commute. Then
2
certainly F 2 : GLn → GLn , (aij ) → (aqij ), is the standard Frobenius map
Fq2 with respect to Fq2 . So, the fixed points under F satisfy
2 F
GF ≤ GF = GLnq = GLn (q 2 ).
2

Here, the fixed point group GUn (q) := GF is the so-called general unitary
n = GUn (q) ∩ SLn (q ), the special
group over Fq2 . We also set SUn (q) := SLF 2

unitary group. (Note that, despite the suggestive notation, GUn (q) cannot
be obtained as a subgroup of GLn (q). Some justification for this notation will
be given in Example 22.11.) The above definition shows that it is the group
of invertible n × n-matrices over Fq2 leaving invariant the non-degenerate
sesquilinear form
."n "
n / "n
 ,  : Fnq2 × Fnq2 → Fq2 , x i ei , y i ei = xi yiq ,
i=1 i=1 i=1

where {ei | 1 ≤ i ≤ n} denotes the standard basis of Fnq2 .

 21.1 Endomorphisms of linear algebraic groups 183

Definition 21.3 An endomorphism F : G → G of a linear algebraic group

G such that for some m ≥ 1 the power F m : G → G is the Frobenius
morphism with respect to some Fpa -structure of G is called a Steinberg en-
domorphism of G. We write GF for the group of fixed points of F on G.

Let’s note that since some power of a Steinberg endomorphism is induced

by a Galois automorphism of k, Steinberg endomorphisms are always auto-
morphisms of abstract groups, but not of algebraic groups in general. See
Exercise 30.1 for a statement on arbitrary endomorphisms of linear algebraic
groups.

Example 21.4 Steinberg endomorphisms on Gm and Ga .

(1) By Exercise 10.11 the only algebraic group endomorphisms of Gm are of

the form σm : c → cm , for some m ∈ Z. The standard Frobenius on Gm
maps all c ∈ Gm to some fixed power which is a power of p = char(k).
Clearly, some power of σm is of this form if and only if m itself is a
power of p. Thus, the only Steinberg endomorphisms on Gm are of the
form F : Gm → Gm , c → cq , for some q = pf , with fixed point group
∼ ×
GFm = Fq , the multiplicative group of a finite field.
(2) Similarly, it is easily seen that in characteristic p > 0 the only endomor-
n i
phisms of Ga are of the form σh : c → h(c), where h = i=0 ai T p ∈
k[T ], that is, h is a so-called additive polynomial . Some power of σh is
i
the standard Frobenius endomorphism on Ga if and only if h = T p . So,
the only Steinberg endomorphisms on Ga are of the form F : Ga → Ga ,
c → cq , with GF a = Fq .
+

In particular, for Ga and Gm all Steinberg endomorphisms are Frobenius

maps.

Examples 21.1 and 21.2 have already shown two instances of Steinberg
endomorphisms of GLn . Both of them fall into the second case of the funda-
mental dichotomy for endomorphisms of simple algebraic groups proved by
Steinberg ([72, Thm. 10.13]):

Theorem 21.5 (Steinberg) Let G be a simple linear algebraic group, σ :

G → G an endomorphism of G. Then precisely one of the following holds:

(1) σ is an automorphism of algebraic groups, or

(2) the group Gσ := {g ∈ G | σ(g) = g} is finite.

The second case occurs if and only if σ is a Steinberg endomorphism.

184 Steinberg endomorphisms

(Clearly the statement does not generalize to semisimple groups!) The

automorphisms occurring in (1) have been studied in Theorem 11.11: up
to inner automorphisms, these come from the graph automorphisms on the
Dynkin diagram of G. From now on, we will be interested exclusively in the
endomorphisms of the second type.
Definition 21.6 Let G be a semisimple algebraic group, F : G → G a
Steinberg endomorphism. Then the finite group of fixed points GF is called
a finite group of Lie type.

21.2 The theorem of Lang–Steinberg

The crucial tool for transferring results from algebraic groups G to finite
groups GF of fixed points under a Steinberg endomorphism F is the theorem
of Lang–Steinberg (see [72, Thm. 10.1] or [76, Thm. J]):
Theorem 21.7 (Lang–Steinberg) Let G be a connected linear algebraic
group over Fp with a Steinberg endomorphism F : G → G. Then the mor-
phism
L : G → G, g → F (g)g −1 ,
is surjective.
Proof for F a standard Frobenius map Since F is standard there exists an
embedding G → GLn as a closed subgroup such that F is given by F :
(aij ) → (aqij ) for some power q of the characteristic of k.
The differential dF of F is 0 (see Example 7.8(3)), so by Proposition 7.7(a)
and Example 7.8(1),(2), d1 L = dF − 1 = −1 is an isomorphism. Now for any
x ∈ G, the morphism L := L ◦ ρx satisfies
L (g) = L ◦ ρx (g) = L(gx) = F (gx)x−1 g −1 = F  (g)g −1 for g ∈ G,
with F  = ρL(x) ◦ F . By Proposition 7.7(a) its differential equals
dx L ◦ d1 ρx = d1 L = dF  − 1 = −1,
hence also is an isomorphism. This shows that dx L is an isomorphism for all
x ∈ G. Now by [66, Thm. 4.3.3(ii)] there exists x ∈ G such that
dim L(G) = dim(dx L)(Lie(G)) = dim Lie(G) = dim(G),
and thus L(G) = G as G is connected. Hence L(G) contains a dense open
subset of G by Proposition 1.6.
Now, for arbitrary x ∈ G consider the map Lx : G → G, g → F (g)xg −1 .
 21.2 The theorem of Lang–Steinberg 185

As above, we can conclude that Lx (G) contains a dense open subset of G.

Since G is connected, by Proposition 1.9 there exists y ∈ L(G) ∩ Lx (G), that
is, y = F (g1 )g1−1 = F (g2 )xg2−1 for some g1 , g2 ∈ G. But then x = L(g2−1 g1 )
lies in the image of L.

Note that the assumption of G being connected is crucial here, otherwise

the conclusion fails. For example, take 1 = G ≤ GLn (p), a finite, hence dis-
connected algebraic group, and F the standard Frobenius map, then clearly
L is a constant map.
In the remainder of this chapter we derive various direct consequences
of this central result. The first concerns the semidirect product G
F 
of an algebraic group with the cyclic subgroup generated by a Steinberg
endomorphism.

Corollary 21.8 Let G be connected with a Steinberg endomorphism F :

G → G. Then, in the semidirect product G
F , the coset G.F of F consists
of a single conjugacy class, that is, G.F = F G . In particular, GgF and GF
are G-conjugate for any g ∈ G.

Proof This is Exercise 30.4.

Definition 21.9 Let H be a group, σ an (abstract group) automorphism

of H. We say that h1 , h2 are σ-conjugate if there exists an x ∈ H with h2 =
σ(x)h1 x−1 . The equivalence classes for this relation are called σ-conjugacy
classes of H.

Note that h1 , h2 ∈ H are σ-conjugate if and only if the elements h1 σ, h2 σ ∈

Hσ are conjugate (by an element of H) in the semidirect product H
σ
of H with σ. So indeed σ-conjugacy is an equivalence relation. Clearly, for
σ acting trivially we recover the usual conjugacy classes of H.

Lemma 21.10 Let G be a linear algebraic group with Steinberg endomor-

phism F : G → G, H a closed connected normal F -stable subgroup. Then
the quotient map induces a natural bijection from F -conjugacy classes of G
to F -conjugacy classes of G/H.

Proof Clearly the natural map π : G → G/H induces a surjection from the
set of F -conjugacy classes of G to those of G/H. Now let g1 , g2 ∈ G have
F -conjugate images, so g2 H = F (x)g1 x−1 H for some x ∈ G. Then g2 =
F (x)g1 hx−1 for some h ∈ H. Now the Lang–Steinberg Theorem, applied to
the endomorphism F  = g1−1 F : H → H, g → g1−1 F (g)g1 , of the connected
group H yields that h = F  (y)y −1 = g1−1 F (y)g1 y −1 for some y ∈ H, whence
g2 = F (x)F (y)g1 y −1 x−1 = F (xy)g1 (xy)−1 are F -conjugate, as claimed.
186 Steinberg endomorphisms

Let’s prove a further useful consequence of the Lang–Steinberg Theorem,

which has many applications.

Theorem 21.11 Let G be a connected linear algebraic group with a Stein-

berg endomorphism F : G → G, acting transitively on a non-empty set V
with a compatible F -action F : V → V (i.e., F (g.v) = F (g).F (v) for all
g ∈ G, v ∈ V ). Then:

(a) F has fixed points on V , i.e., V F = ∅.

(b) If the stabilizer Gv is closed for some v ∈ V , then for any v ∈ V F there
is a natural 1–1 correspondence:

{GF -orbits on V F } ←→ {F -classes in Gv /G◦v }.

Proof (a) Let v ∈ V . By transitivity of the action there exists some g ∈ G

with g.F (v) = v. By the theorem of Lang–Steinberg we may write g =
x−1 F (x) for some x ∈ G. Hence, F (x.v) = F (x).F (v) = x.v, so x.v ∈ V F .
(b) We have Gg.v = gGv g −1 for all g ∈ G, hence since conjugation is an
isomorphism of algebraic groups, by transitivity all stabilizers Gv of v ∈ V
are closed. By part (a), there exists v ∈ V F . Its stabilizer Gv is a closed
subgroup, containing G◦v with finite index, and both are F -stable.
Now, if w ∈ V F , then by transitivity there exists g ∈ G such that w = g.v.
Since v, w ∈ V F we have g.v = w = F (w) = F (g).F (v) = F (g).v, so
g −1 F (g) ∈ Gv . Now if g.v = h.v = w for h ∈ G then h−1 g ∈ Gv , whence
g −1 F (g) and h−1 F (h) are F -conjugate in Gv . By Exercise 30.5 this map
V F /GF → Gv , g → g −1 F (g), induces a bijection from GF -orbits in V F to
F -classes in Gv . By Lemma 21.10 the latter are in natural bijection with
F -classes in Gv /G◦v .

Let’s apply this result to the action of G on itself by conjugation:

Corollary 21.12 Let G be connected reductive, F : G → G a Steinberg

endomorphism. Then there exists a pair T ≤ B consisting of an F -stable
maximal torus T contained in an F -stable Borel subgroup B of G. All such
pairs T ≤ B are GF -conjugate.

Proof By Theorems 6.4 and 4.4 the group G acts transitively by conjugation
on

V := {(B, T ) | T a maximal torus in a Borel subgroup B of G},

compatibly with F . So by Theorem 21.11(a) there exists an F -stable pair

v = (B, T ) in V . Moreover, the stabilizer of v is just Gv = NG (B, T ) :=
NG (B) ∩ NG (T ). We claim that this is a closed, connected subgroup of G.
 21.2 The theorem of Lang–Steinberg 187

As NG (B) = B by Theorem 6.12, we have Gv = NB (T ). By Theo-

rem 4.4(b), NB (T ) = CB (T ) ≤ CG (T ) = T , the last equality by Corol-
lary 8.13(b), as G is connected. So Gv = T is closed connected, and hence
all F -stable pairs in V are GF -conjugate by Theorem 21.11(b).
Definition 21.13 A maximal torus of G as in Corollary 21.12 is called
maximally split with respect to F .
For the standard Frobenius map F on G = GLn , the F -fixed points of
maximally split tori of G are again isomorphic to a product of copies of the
multiplicative group (now of the finite field), but this need not be true for
general Steinberg endomorphisms, as the next example shows.
Example 21.14 The F -structure of maximally split tori in GLn .
(1) Let G = GLn with Fq : G → G, (aij ) → (aqij ), the standard Frobenius
map. Then Tn ≤ G is an Fq -stable Borel subgroup, Dn ≤ Tn is an
Fq -stable maximal torus, so Dn is a maximally split torus of GLn with
q ∼ × × × n
respect to Fq . Clearly, we have DFn = Fq × · · · × Fq = (Fq ) of order
|Dn | = (q − 1) .
Fq n
−tr w
(2) Let G = GL  n , F = Fq σ with Fq as before and σ : (aij ) → ((aij ) )
0 . .1
with w = . . Then again Tn is an F -stable Borel subgroup, and
1 0
Dn ≤ Tn is an F -stable maximal torus, hence maximally split in GLn
with respect to F . Let’s compute its structure: For t = diag(t1 , . . . , tn )
we have
−q
F (t) = F (diag(t1 , . . . , tn )) = diag(t−q
n , . . . , t1 ).
−q −q
So, F (t) = t if and only if t1 = t−q
n , t2 = tn−1 , . . . , tn = t1 . In particular,
2
tqi = ti for all 1 ≤ i ≤ n, and tq+1
m+1 = 1 if n = 2m + 1 is odd. Hence,
 
× m
∼ (F 2 ) (q 2 − 1)m if n = 2m,
DF n =
q
× m × ×
of order
(Fq2 ) × Fq2 /Fq (q − 1) (q + 1) if n = 2m + 1.
2 m

Note that here GF ∼= GUn (q) is again isomorphic to the general unitary
group by Corollary 21.8, since we have changed the Frobenius map from
Example 21.2 only by the inner element w.
 22
Classification of finite groups of Lie type

We study the Steinberg endomorphisms more closely via their action on the
character group and on the Weyl group and present their classification which
in turn gives a classification of the finite groups of Lie type. As for semisimple
algebraic groups, this classification can be given in combinatorial terms via
so-called complete root data. We then comment on the construction of the
various finite groups of Lie type.

22.1 Steinberg endomorphisms

Let G be a connected reductive linear algebraic group, F : G → G a Stein-
berg endomorphism. Recall from Corollary 21.12 that there exists an F -
stable maximal torus T contained in an F -stable Borel subgroup B of G.
As T is F -stable, so is NG (T ), so F naturally acts on the Weyl group W =
NG (T )/T of G. Similarly, F also acts on the character group X := X(T )
and the cocharacter group Y := Y (T ) via

F (χ)(t) := χ(F (t)) for χ ∈ X, t ∈ T,

F (γ)(c) := F (γ(c)) for γ ∈ Y, c ∈ k × .

Let’s consider these actions in some more detail:

Lemma 22.1 Let T be a torus with character group X, and F : T → T a

Steinberg endomorphism. Then:

(a) There exists δ ∈ N and a power r of p = char(k) such that F δ |X = r idX

on X.
(b) We have T F ∼ = X/(F − 1)X.
 22.1 Steinberg endomorphisms 189

Proof Let’s first consider the case where T = Dn ≤ GLn and F = Fq is the
standard Frobenius map on GLn . Then
F (χ)(t) = χ(F (t)) = χ(tq ) = χ(t)q = (qχ)(t) for all χ ∈ X, t ∈ T,
so F |X = q idX as claimed. Moreover, T F ∼ = (F× n
q ) (see Example 21.14(1)).
In the general situation, as F is a Steinberg endomorphism, there exists
an embedding T → GLn such that some power F  of F is the restriction
to T of a standard Frobenius on GLn . Let T1 be a maximal torus of GLn
containing the torus T . This is conjugate to the maximal torus Dn , so T1 =
Dgn for some g ∈ GLn . Now g has entries in some finite subfield of k, so
there is a positive power F δ of F  which fixes g. Then conjugation by g
defines an isomorphism T1 ∼ = Dn which commutes with F δ . By the first
part, F δ acts by multiplication with some power r of p on X(Dn ), so on
X(T1 ). Restriction of characters from T1 to T is F -equivariant, with F -
stable kernel T ⊥ (see Proposition 3.8(a)), so F δ also acts by the scalar r on
X = X(T ) ∼ = X(T1 )/T ⊥ , proving (a).
For (b), again apply Proposition 3.8(a) to the closed subgroup T F ≤ T to
obtain that X(T F ) ∼= X/(T F )⊥ . Now we claim that
t ∈ TF ⇐⇒ F (t) = t ⇐⇒ (F − 1)χ(t) = 1 for all χ ∈ X.
The implication from left to right is clear. Conversely, if F (t) = t then clearly
there is χ ∈ X with χ(F (t)t−1 ) = 1. So we have T F = ((F − 1)X)⊥ . As
(F −1)X contains (F δ −1)X = (r −1)X, it is of index prime to p in X, so we
have (T F )⊥ = (F − 1)X by Proposition 3.8(b). Finally, as |T F | is prime to p
and k contains all p -roots of unity, X(T F ) = Hom(T F , k × ) ∼
= Hom(T F , C× ),
F
and the latter is isomorphic to T . Thus indeed

= X(T F ) ∼
TF ∼ = X/(T F )⊥ ∼
= X/(F − 1)X.
Write Φ ⊂ X for the root system of G, with positive system Φ+ with
respect to T and B. For α ∈ Φ let’s choose isomorphisms uα : Ga → Uα
onto the root subgroups (see Theorem 8.17(c)). We set XR := X ⊗Z R.
Proposition 22.2 Let G be a connected reductive algebraic group with
Steinberg endomorphism F : G → G, and T , B, X, Φ as above.
(a) There exists a permutation ρ of Φ+ and, for each α ∈ Φ+ , a positive
integral power qα > 1 of p = char(k) and aα ∈ k × such that F (ρ(α)) =
qα α and F (uα (c)) = uρ(α) (aα cqα ) for all c ∈ k.
(b) There exists δ ≥ 1 such that F δ |X = q δ idX and F = qφ on XR for
some positive fractional power q of p and some φ ∈ Aut(XR ) of order δ
inducing ρ−1 on Φ+ .
190 Classification of finite groups of Lie type

In particular both F and ρ induce graph automorphisms of the Coxeter dia-

gram of Φ.

Proof Part (a) is just Lemma 11.10 applied to the endomorphism F . For
part (b), by Lemma 22.1 there exists some minimal δ ∈ N such that F δ |X =
1
r idX . Thus all eigenvalues of F on X ⊗Z C have absolute value q := r δ > 1, a
fractional power of p. Then with φ := 1q F : XR → XR we get φδ = q −δ F δ =
idXR , so φ has finite order and permutes the roots in the same way as ρ−1
by (a).
By Lemma 11.10, ρ stabilizes the set of simple roots ∆ ⊆ Φ+ . Hence, in
its action on the Weyl group F stabilizes the set of simple reflections of W
and thus both F and ρ induce graph automorphisms of the Coxeter diagram
of Φ.

Note that F δ fixes all the one-dimensional submodules of X generated

by the roots, so it also fixes all simple reflections in W , hence acts trivially
on W .
In order to obtain further information on the qα we need to assume that
F permutes the simple components of the derived group [G, G] transitively,
which is the case for example if G is simple.

Lemma 22.3 In the situation of Proposition 22.2 assume that G is simple.

Then qα is constant on roots of the same length.
Moreover, either all qα = q, in which case ρ preserves root lengths, or else
G is of type B2 , G2 or F4 with p = 2, 3, 2, respectively, ρ interchanges long
and short roots, qα qβ = q 2 and qβ /qα = p for all long roots α and all short
roots β.

Proof We first show that qα is constant on roots of a given length. As G is

simple its root system Φ is indecomposable, see Exercise 10.33. Then any two
roots α, β ∈ Φ of the same length lie in the same W -orbit by Corollary A.18,
so β = wα for some w ∈ W . Let w1 ∈ W such that F (w1 ) = w, then by
Proposition 22.2(a)

qα F (ρ(β)) = qα qβ β = qα qβ wα = qβ wF (ρ(α)) = qβ F (w1 ρ(α)).

Thus ρ(β) and w1 ρ(α) are proportional roots with positive proportionality
factor qα /qβ , so they are equal by axiom (R2) whence qα = qβ is constant
on roots of the same length.
Since G is simple, W is irreducible, so there is a unique W -invariant scalar
product on XR up to scalar multiples, which must hence be fixed by φ, up
to a scalar. If qα is constant on Φ+ , then by Proposition 22.2 we necessarily
 22.1 Steinberg endomorphisms 191

have qα = q, and φ ◦ ρ = id on Φ+ . Thus φ is a permutation of the roots,

and it preserves quotients of root lengths, so the same holds for ρ.
Since ρ induces a graph automorphism of the Coxeter diagram, G must
be of type B2 , G2 or F4 if ρ does not preserve root lengths. This case is
discussed in Exercise 30.6.

Definition 22.4 We’ll say that a Steinberg endomorphism F : G → G of a

connected reductive group G is Fq -split if there exists an F -stable maximal
torus T ≤ G such that F (t) = tq for all t ∈ T , or equivalently, if F acts as
q id on X(T ).
It is called twisted if it is not split, and the product of an Fq -split endo-
morphism with an (algebraic group) automorphism of G (see Theorem 11.11
for a description of these). Finally, we call F very twisted if some power of
F induces the non-trivial graph automorphism on the Coxeter diagram of
an irreducible component of G of type B2 , G2 or F4 .

It turns out that a Steinberg endomorphism of a simple algebraic group

is already essentially determined by q > 0 and φ ∈ Aut(XR ) as follows (see
[72, 11.7]):

Theorem 22.5 (Classification of Steinberg endomorphisms) Let G be a

simple simply connected algebraic group, F : G → G a Steinberg endomor-
phism. Then F is uniquely determined, up to inner automorphisms of G, by
q and the Coxeter diagram automorphism ρ|∆ in Proposition 22.2.
Conversely, for every pair (q, ρ), with q an integral power of char(k) and
ρ a Dynkin diagram automorphism of ∆, there exists a Steinberg endomor-
phism F of G with parameters q, ρ. For ρ the non-trivial Coxeter diagram
automorphism of ∆ of type B2 , G2 , respectively F4 , there exists a Stein-
berg endomorphism with associated graph automorphism ρ if and only if
q 2 = 22f +1 , 32f +1 , respectively 22f +1 , for some f ∈ N0 .

The uniqueness essentially follows from the classification of simple alge-

braic groups via their root data, see Theorem 9.13. Let’s discuss the question
of existence of the various Steinberg endomorphisms.

Example 22.6 Let G be simple of simply connected type and let Fp : G →

G be the endomorphism from Theorem 16.5 which acts as p-powering on a
maximal torus and on the parameters of the associated root subgroups. It
can be shown that this is an Fp -split Steinberg endomorphism, so the same
is true for all of its powers Fq = Fpf .
Now let ρ be a permutation of the set of simple roots which induces a
symmetry of the Dynkin diagram and σ the graph automorphism of G in-
192 Classification of finite groups of Lie type

duced by ρ whose existence was asserted in Theorem 11.12, which permutes

the root subgroups for simple roots according to ρ. Then Fp commutes with
σ, so the composition F = Fq σ has the property that F δ = Fpf δ is Fqδ -split.
Thus, as Fq acts trivially on the root system, F is a Steinberg endomorphism
which induces ρ on ∆.

More explicitly, for the groups of type An , n ≥ 2, we have already exhibited

in Example 21.2 a twisted Steinberg endomorphism inducing the non-trivial
graph automorphism, with fixed point group the unitary group, and we will
construct the two different types of Steinberg endomorphisms for SO2n (of
type Dn ) in Example 22.9 below.
The following lifting result allows one to pass to simple groups of arbitrary
isogeny type:

Proposition 22.7 Let G be semisimple, π : Gsc → G the natural isogeny

from a simply connected group of the same type. Then every isogeny σ : G →
G can be lifted to an isogeny σsc : Gsc → Gsc such that π ◦ σsc = σ ◦ π.

Proof Take H1 = Gsc , H2 = G and ϕ = σ ◦ π in Proposition 9.18.

Thus, all Steinberg endomorphisms of semisimple groups are induced by

Steinberg endomorphisms of simply connected groups.

Example 22.8 Conversely, a Steinberg endomorphism F : Gsc → Gsc of a

simply connected group descends to an epimorphic image G = π(Gsc ) under
an isogeny π if ker(π) is F -stable. Since ker(π) is a subgroup of the fundamen-
tal group, in the case of simple groups the latter condition can only possibly
fail when G is of type D2n , ker(π) has order 2 and F acts non-trivially on the
Dynkin diagram (see Table 9.2). Indeed, the graph automorphism of order 2
of D2n interchanges the two subgroups of order 2 of the fundamental group
which belong to the half-spin groups (see Example 9.16(3)). So the Steinberg
endomorphism of Gsc inducing a non-trivial graph automorphism does not
descend to the half-spin groups.
Also, the triality graph automorphism of D4 of order 3 (see Example 12.12)
acts non-trivially on the fundamental group, hence there do not exist Stein-
berg endomorphisms associated to triality graph automorphisms on SO8 or
on HSpin8 .

Thus, finally, by Examples 22.6 and 22.8 the existence question in Theo-
rem 22.5 is only about the very twisted automorphisms of groups of excep-
tional type, which we will not discuss here. Note that there necessarily q is
a non-integral power of 2 or 3. See [73, §11] for this case.
 22.2 The finite groups GF 193

22.2 The finite groups GF

Let G be a simple linear algebraic group with Steinberg endomorphism
F : G → G inducing a graph automorphism of order δ on the Dynkin
diagram such that F δ = q δ id on XR . If the root system of G has type R, the
corresponding finite group of fixed points GF is sometimes denoted δ R(q)
(an abuse of notation since it does not specify the isogeny type of G). The
groups of type 2 B2 are called Suzuki groups since they were discovered by
M. Suzuki [78] as a particular class of Zassenhaus groups; the groups of
types 2 G2 and 2 F4 were first constructed by R. Ree [57, 58] and are hence
called Ree groups. In the case where F induces a non-trivial diagram au-
tomorphism, the group GF is sometimes also called a twisted group of Lie
type.
Table 22.1 shows the various possibilities for the finite groups of Lie type
GF according to Theorem 22.5 and Proposition 22.7, and the identification
of some of them with classical groups over finite fields.

Table 22.1 Finite groups of Lie type

Φ δ GF
sc GF
ad other isogeny types
An−1 1 SLn (q) PGLn (q) Zd/e .PSLn (q).Ze
e|d := (n, q − 1)
Bn 1 Spin2n+1 (q) SO2n+1 (q) −
Cn 1 Sp2n (q) PCSp2n (q) −
Dn 1 Spin+2n (q) (PCO◦2n )+ (q) SO+ +
2n (q), HSpin2n (q)
G2 1 G2 (q) −
F4 1 F4 (q) −
E6 1 (E6 )sc (q) (E6 )ad (q) −
E7 1 (E7 )sc (q) (E7 )ad (q) −
E8 1 E8 (q) −
An−1 2 SUn (q) PGUn (q) Zd/e .PSUn (q).Ze
e|d := (n, q + 1)
Dn 2 Spin−
2n (q) (PCO◦2n )− (q) SO−2n (q)
D4 3 3
D4 (q) −
E6 2 (2 E6 )sc (q) (2 E6 )ad (q) −
B2 2 2
B2 (q ) = Suz(q 2 ), q 2 = 22f +1
2
−
G2 2 2
G2 (q 2 ), q 2 = 32f +1 −
F4 2 2
F4 (q 2 ), q 2 = 22f +1 −

Example 22.9 We construct the two types of even-dimensional orthogonal

groups.
194 Classification of finite groups of Lie type

(1) Recall the orthogonal group GO2n with respect to the quadratic form

f (x1 , . . . , x2n ) := x1 x2n + · · · + xn xn+1

2n
on V = Fq . Clearly GO2n is stable under the standard Frobenius
Fq
map Fq : GL2n → GL2n . We write GO+ 2n (q) := GO2n and call it the
Fq
general orthogonal group of plus type. Similarly, SO+2n (q) := SO2n (where
◦
SO2n = GO2n ). Note that the subspace of V spanned by the first n
standard basis vectors is totally singular for SO+2n (q).
(2) Now let
⎛ ⎞
In−1
⎜ 0 1 ⎟
g := ⎜⎝
⎟ ∈ GO2n .
⎠
1 0
In−1
As det(g) = −1 we have g ∈ GO2n \ SO2n when char(k) = 2. In fact,
this also holds when char(k) = 2. An easy calculation shows that g
interchanges the root subgroups Un−1,n and Un−1,n+1 of SO2n (see Ex-
ample 11.7), hence the simple roots n−1 − n and n−1 + n , and fixes
all other simple roots.
Thus F  := gF : GO2n → GO2n is a Steinberg endomorphism which
induces the non-trivial graph automorphism of order 2 on the Dynkin
F
diagram of SO2n of type Dn . The fixed point group GO− 2n (q) := GO2n , is
called the general orthogonal group of minus type, and SO− F
2n (q) := SO2n
the special orthogonal group of minus type, or non-split orthogonal group.
Equivalently, these non-split groups can be obtained as follows: When
char(k) = 2 let G be the isometry group of the Fq -anisotropic quadratic
2
form f (x1 , x2 ) := x21 − wx22 on V = Fq , where w ∈ F× q is a fixed non-
square. Since all non-degenerate quadratic forms on V are equivalent
(see [2, §21], for example), we have G ∼ = GO2 and G◦ ∼ = SO2 by Defini-
tion 1.15. The standard Frobenius map Fq : GL(V ) → GL(V ) stabilizes
G. But it is easily seen that V Fq = F2q contains no non-zero isotropic
vector for f , so (G◦ )Fq is not conjugate to the special orthogonal group
SO+ 2 (q) on V
Fq
. In fact, we have (G◦ )Fq ∼
= SO−
2 (q). More generally, let
G be the isometry group of the quadratic form

f (x1 , . . . , x2n ) := x1 x2n + · · · + xn−1 xn+2 + x2n − wx2n+1

on V = Fq , isomorphic to GO2n . Then GFq ∼ ◦ Fq ∼

2n
= GO− 2n (q) and (G ) =
−
SO2n (q). A similar construction, starting from any two-dimensional Fq -
anisotropic quadratic form, works when char(k) = 2.
 22.2 The finite groups GF 195

See the books of Dieudonné [21] or Grove [30], for example, for more
information on the various types of finite classical groups.
The various possibilities for the finite groups of Lie type GF can again
be encoded in a purely combinatorial way. The root datum (X, Φ, Y, Φ∨ )
of a semisimple group G is obtained by fixing a maximal torus T of G.
If F : G → G is a Steinberg endomorphism, we may assume T to be F -
stable. Then its Weyl group W = NG (T )/T is also F -stable, and this defines
a semidirect product W F . Any element wF in the coset W.F of W in
W F  still stabilizes T , hence the root datum. By Corollary 21.8 it also
defines a G-conjugate, hence isomorphic, finite group of fixed points. So, by
Theorem 22.5, (G, F ) is determined up to isomorphism by the root datum
of G, the coset W φ and q, where φ ∈ Aut(XR ) stabilizes Φ ⊂ X and also
Φ∨ ⊂ Y , hence normalizes W , in the following sense: if (G, F ), (G , F  ) both
correspond to (X, Φ, Y, Φ∨ , W φ) and the same q, there is an isomorphism
σ : G → G such that F  ◦ σ = σ ◦ F .

Definition 22.10 G := (X, Φ, Y, Φ∨ , W φ) as above is called a complete

root datum.

The complete root datum G, together with q, determines (G, F ) and hence
the finite group GF up to isomorphism by Corollary 21.8. We write GF =:
G(q). Thus, G stands for a whole family of finite groups of Lie type, for
example {SLn (q)}, {SUn (q)} or {E8 (q)}, where q runs over all prime powers.
It is not true, though, that groups GF for different complete root data
are necessarily non-isomorphic. For example we have SL3 (2) ∼ = PGL3 (2)
(non-isomorphic but isogenous algebraic groups may lead to isomorphic fi-
nite groups), Sp2n (2f ) ∼
= SO2n+1 (2f ) (different root systems may lead to
isomorphic finite groups), or Sp4 (3) ∼ = SU4 (2) (here even the underlying
characteristic differs). But this happens not very often and all such “excep-
tional” isomorphisms are known, see Remark 24.9.

Example 22.11 (Complete root data for GLn (q) and GUn (q))

(1) Let G ≤ GLn be a connected reductive subgroup stable under the stan-
dard Frobenius map Fq on GLn , and assume that Fq acts trivially on the
Weyl group W of G. Then φ is trivial and the corresponding complete
root datum is G = (Γ, W ), where Γ is the root datum of G.
(2) Let G = GLn , Fq : G → G the standard Frobenius map, and σ : G → G,
(aij ) → ((aij )−tr )w , with w as in Example 21.14(2). Let F := Fq σ, so
GF = GUn (q). Then T := Dn is an F -stable maximal torus, and σ
acts on X(T ) as −w, hence F acts as −wq on X(T ). Thus, in this case
196 Classification of finite groups of Lie type

G = (Γ, −W ) is the complete root datum for the family {GUn (q)} (with
Γ the root datum of GLn and W its Weyl group).
This is sometimes expressed by saying that GUn (q) is obtained from
GLn (q) by replacing q by −q, or “GUn (q) is Ennola-dual to GLn (q)”.
 23
Weyl group, root system and root subgroups

We already saw how the structure of semisimple algebraic groups is con-

trolled by their root system and its Weyl group. We now show how the root
system of a connected reductive group with a Steinberg endomorphism gives
rise to a root system and then also to root subgroups of the corresponding
group of fixed points.

23.1 The root system

In order to investigate the action of Steinberg endomorphisms on the root
system and its Weyl group we first need another consequence of the Lang–
Steinberg Theorem. In general taking fixed points under F is not exact, that
is, it does not preserve short exact sequences.

Example 23.1 Consider the exact sequence of algebraic groups

1 −→ Z(SLn ) −→ SLn −→ PGLn −→ 1

with fixed points

1 −→ Z(SLn (q)) −→ SLn (q) −→ PGLn (q)

under the standard Frobenius map. Here, the last arrow is in general not
surjective, the image PSLn (q) has index gcd(n, q − 1) in PGLn (q), see Sec-
tion 24.1.

This phenomenon cannot occur if the kernel is connected, see Exercise 30.7:

Proposition 23.2 Let G be a linear algebraic group with a Steinberg endo-

morphism F : G → G, and H an F -stable closed connected normal subgroup
of G. Then the natural map GF /H F → (G/H)F is an isomorphism.
198 Weyl group, root system and root subgroups

Applied to the normalizer NG (T ) of an F -stable maximal torus T of a

connected reductive group G with Steinberg endomorphism F : G → G, this
shows that the F -fixed points on W satisfy

W F = (NG (T )/T )F ∼
= NGF (T )/T F .

Note that in general, though, NGF (T ) = NGF (T F ); for example we might

have T F = 1 while T = 1, which happens when T is maximally split and
q = 2, so T F = (F×2)
dim T
. In order to study the fixed point group W F we
need the following result about automorphisms of Weyl groups. For this,
recall the standard parabolic subgroup WI of W , for I a subset of the set of
simple reflections S of W .

Lemma 23.3 Let W be the Weyl group of a root system with set of simple
reflections S ⊂ W and φ an automorphism of W stabilizing S.

(a) For each φ-orbit I ⊆ S, WIφ = sI  for a (unique) involution sI ∈ WI .

(b) The group of fixed points W φ is generated by {sI | I ⊆ S a φ-orbit}.

Proof As φ stabilizes S, it induces a graph automorphism on the Coxeter

diagram of W . So according to the Classification Theorem 9.6, there are
only the following two possibilities for I: either I = {s1 , . . . , sr } are mutually
commuting and φ permutes them cyclically, in which case sI := s1 · · · sr is
as claimed, or I = {s1 , s2 } with o(s1 s2 ) = m > 2. In the latter case an easy
computation in the dihedral group WI = s1 , s2  shows that WIφ = sI ,
where sI := s1 s2 s1 · · · (m factors). For (b) (and a proof of (a) not referring
to the classification of root systems) see Lemma C.1.

The proof shows that the element sI can also be described as being the
longest element of the parabolic subgroup WI . The various possibilities for
W φ when W is irreducible and φ is non-trivial are displayed in Table 23.1.
It turns out that W φ is again a Coxeter group, and even a Weyl group
except for W of type F4 with φ acting non-trivially; in the latter case, W φ
is isomorphic to the dihedral group I2 (8) of order 16, which is a real, but
non-rational reflection group.

Table 23.1 Automorphisms of irreducible Weyl groups

Φ A2n−1 A2n Dn D4 E6 B2 G2 F4
o(φ) 2 2 2 3 2 2 2 2
Wφ Cn Bn Bn−1 G2 F4 A1 A1 I2 (8)
 23.1 The root system 199

We next introduce a root system for a finite group of Lie type. Let G be
a semisimple algebraic group with a Steinberg endomorphism F : G → G.
As we saw in Section 22.1, fixing an F -stable torus inside an F -stable Borel
subgroup B of G with associated root system Φ induces an action of F on the
set of corresponding root subgroups Uα , α ∈ Φ+ , and hence a permutation
ρ of Φ+ . As in Proposition 22.2 we write F |XR = qφ with φ ∈ Aut(XR ) of
finite order δ. Let XRφ denote the fixed space of φ on XR , i.e., the eigenspace
for the eigenvalue 1. Then the homomorphism

1" i
δ−1
π : XR −→ XRφ , χ→ φ (χ),
δ i=0

clearly is the identity on XRφ , hence a projection. Let S ⊂ W denote the set
of simple reflections of W .

Lemma 23.4 I ) | I ⊆ S an F -orbit, w ∈ W }. Then:

Let Ω := {w(Φ+ F

(a) Ω defines a partition of Φ. If ∼ denotes the associated equivalence rela-

tion, then α ∼ β if and only if π(α), π(β) are positive multiples of each
other.
(b) Each ω ∈ Ω is contained in either Φ+ or in Φ− .

For the proof see Lemma C.2.

Definition 23.5 For each equivalence class ω for the equivalence relation
∼ defined in Lemma 23.4 let αω denote the vector of maximal length among
the {π(α) | α ∈ ω}. The set ΦF := {αω | ω ∈ Ω} is called the root system of
GF . We let ∆F := {αω | ω ⊆ ∆} ⊆ ΦF .

By Theorem C.5 the set ΦF satisfies axioms (R1)–(R3) of an abstract

root system in XRφ , with base ∆F and Weyl group W φ (in its action on
XRφ ) with set of simple reflections SF := {sI | I ⊆ S an F -orbit}. For Φ
indecomposable, ΦF also satisfies the integrality axiom (R4) unless Φ is of
type F4 and ρ is non-trivial. Apart from this exceptional case, the type of
the root system ΦF can then be read off from Table 23.1. These facts will
not be needed in the sequel (see also Exercise C.2).

Example 23.6 Let’s consider G of type D4 with F : G → G inducing the

triality automorphism (see Example 12.12). Here
Φ+ = {α1 , α2 , α3 , α4 , α1 + α2 , α2 + α3 , α2 + α4 , α1 + α2 + α3 ,
α1 + α2 + α4 , α2 + α3 + α4 , α1 + α2 + α3 + α4 , α1 + 2α2 + α3 + α4 }
with the simple roots numbered as in the Dynkin diagram in Table 9.1. The
200 Weyl group, root system and root subgroups

second simple root is fixed, while the other three are permuted cyclically by
triality. Let’s set
1
µ := π(α1 ) = (α1 + α3 + α4 ), ν := π(α2 ) = α2 .
3
Then π(α1 +α2 ) = π(α2 +α3 ) = π(α2 +α4 ) = µ+ν, π(α1 +α2 +α3 ) = 2µ+ν,
π(α1 + α2 + α3 + α4 ) = 3µ + ν and π(α1 + 2α2 + α3 + α4 ) = 3µ + 2ν. Thus
here ΦF is a root system of type G2 , with base {µ, ν} (see Example 9.5).

23.2 Root subgroups

Let G be connected reductive with Steinberg endomorphism F : G → G. We
next construct root subgroups in GF .
Proposition 23.7 For any equivalence class ω ∈ Ω the group
Uω := Uα | α ∈ ω
is F -stable connected unipotent and has a product decomposition
!
Uω = Uα
α∈ω

with uniqueness of expression, for any fixed order of the α ∈ ω.

Proof After conjugation by some w ∈ W F we may assume that ω = Φ+ I
for some φ-orbit I ⊆ S. Then Uω is just the unipotent radical UI of a
Borel subgroup of the standard Levi subgroup LI corresponding to I, see
Section 12.2. The product decomposition thus follows from Theorem 11.1.
Since all Uα are connected, so is Uω . It is clearly F -stable, being generated
by an F -stable set of subgroups (see Proposition 22.2(a)).
The fixed point subgroups UωF , ω ∈ Ω, are called the root subgroups of
GF . We next determine the orders of these root subgroups. For this, recall
the qα attached to α ∈ Φ+ from Proposition 22.2.
Proposition 23.8 Let U be an F -stable subgroup of Ru (B) normalized by

T . Then |U F | = α∈Σ qα , where Σ := {α ∈ Φ+ | Uα ≤ U }.
Proof We use induction on |Σ|. Let U  := [U, U ], the derived subgroup of U
and Σ := {α ∈ Φ+ | Uα ≤ U  } the corresponding roots. By Proposition 11.5
both U and U  are generated by the Uα they contain, and moreover the
abelian group U/U  is naturally isomorphic to the direct product of the Uα
with α ∈ Σ1 := Σ\Σ . Clearly U  is normalized by T , F -stable and of strictly
 23.2 Root subgroups 201

smaller dimension than U , so by induction |(U  )F | = α∈Σ qα . Furthermore,
as U  is connected, |U F | = |(U  )F | · |(U/U  )F | by Proposition 23.2, so we

need to show that |(U/U  )F | = α∈Σ1 qα .
The Steinberg endomorphism F acts on the direct product of the Uα
according to Proposition 22.2(a). Considering a single ρ-orbit in Σ1 at a
time we are reduced to showing that the number of solutions t1 , . . . , tn ∈ k
to the system
c1 tq11 = t2 , c2 tq22 = t3 , . . . , cn tqnn = t1 ,

equals i qi , where ci ∈ k × and qi are positive powers of char(k). Solving
the first n − 1 equations for the ti , i ≥ 2, in terms of t1 we are left with

the single equation ctr1 = t1 , with c ∈ k × and r = i qi an integral power
of char(k), which indeed has precisely r solutions in the algebraically closed
field k.
Corollary 23.9 Let G be simple, ω ∈ Ω. Then |UωF | = q |ω| .
Proof By Lemma 22.3 either all qα = q for all α ∈ Φ+ , or ω = {α, β}
with α long, β short and qα qβ = q 2 . In both cases, the claim follows from
Propositions 23.7 and 23.8.
The values of q |ω| for the non-trivial automorphisms of indecomposable
root systems are shown in Table 23.2. The underlying diagrams are those for
the Weyl group W F from Table 23.1, respectively the Coxeter group I2 (8).

Table 23.2 q |ω| for twisted groups of Lie type

q2 q2 q2 q3 q2 q2 q2 q
2 2
A2n > A2n−1 <
n≥1 n≥2

q q q q2
2
Dn >
n≥4

q3 q q q q2 q2
3 2
D4 < E6 >

q4 q6 q4 q2
2 2 2
B2 G2 F4 <
202 Weyl group, root system and root subgroups

Example 23.10 Let’s compute the structure of some root subgroups.

(1) Let G be semisimple, F : G → G a Steinberg endomorphism with trivial
action on the Weyl group. Then XRφ = XR , the root system ΦF equals
Φ, and the root subgroups of GF are just the UαF ∼
= F+
q , for α ∈ Φ (see
Example 21.4(2)).
(2) Let G = SL3 and F : G → G the Steinberg endomorphism with GF =
SU3 (q) as in Example 21.14(2). Here Φ = {±α, ±β, ±(α + β)}, and
ρ interchanges α with β and fixes α + β. Thus δ = 2, π(α + β) =
α + β = 2π(α) = 2π(β), and the equivalence classes under ∼ are just
±{α, β, α + β}. For ω = {α, β, α + β} we have
Uω = Uα Uβ Uα+β = U3 ,
the unipotent radical of the Borel subgroup T3 of G. For our choice of
F we get
⎧⎛ ⎞  ⎫
⎨ 1 a b  ⎬

UωF = UF = ⎝0 1 −aq ⎠  a, b ∈ Fq2 , bq + b = −aq+1
3
⎩  ⎭
0 0 1 

as (unique positive) root subgroup of SU3 (q) for the root αω = α + β.

It has order q 3 , since for any value of a ∈ Fq2 , there are q solutions to
xq + x = −aq+1 in Fq2 (see also Corollary 23.9 and Table 23.2 for 2 A2 ).
(3) For G of type D4 with F : G → G inducing the triality automorphism
as in Example 23.6 we have UµF ∼ = F+ F ∼ +
q 3 and Uν = Fq (see Table 23.2 for
3
D4 ).
The previous example shows that the root subgroups of GF are, in general,
not commutative. More precisely, this happens for orbits of types A2 , B2 and
G2 .
Using the Chevalley commutator formula (Theorem 11.8) for the algebraic
group, an analogue for the root subgroups in GF can be shown (see [73, Cor.
of Lemma 62]):
Proposition 23.11 (Commutator formula in GF ) For all roots αω = −αν
in the root system ΦF of GF we have
!
[UωF , UνF ] ⊆ UµF
µ

where the product runs over the orbits µ ∈ Ω such that αµ = mαω + nαν for
some m, n > 0, in any fixed order.
 24
A BN-pair for GF

We now derive a Bruhat decomposition for the finite groups of Lie type by
investigating the fixed points of a Steinberg endomorphism on the Bruhat
decomposition for algebraic groups. This leads to a first order formula for
groups of Lie type. A second, factored form can be derived from this using
some standard results on the invariant theory of finite reflection groups. It
turns out that the orders in a family of groups of Lie type belonging to the
same complete root datum are given by a polynomial in q.
The Bruhat decomposition also gives rise to a BN-pair structure for finite
groups of Lie type. Using general properties of BN-pairs this then allows
us to conclude that most groups of Lie type derived from simple algebraic
groups of simply connected type are quasi-simple, that is, simple modulo
their center. Furthermore, we are able to determine their automorphism
groups.

24.1 Bruhat decomposition and the order formula

Recall the double coset decomposition of a connected reductive group with
respect to a Borel subgroup proved in Theorem 11.17. We obtain the follow-
ing finite analogue:

Theorem 24.1 (Bruhat decomposition in GF ) Let G be connected reduc-

tive with Steinberg endomorphism F : G → G, T ≤ B an F -stable maximal
torus in an F -stable Borel subgroup of G, W the Weyl group of G with respect
to T . Then
)
GF = (Uw− )F ẇB F
w∈W F
204 A BN-pair for GF

with uniqueness of expression. Here, for w ∈ W F , ẇ denotes a fixed preimage

in NGF (T ).
Proof Theorem 11.17 gives the Bruhat decomposition
)
G= Uw− ẇB
w∈W

with uniqueness. Now B and T are F -stable, so F acts on W = NG (T )/T

and it permutes the Uw− . Note that by Proposition 23.2 for each w ∈ W F we
may choose an F -stable representative ẇ ∈ NGF (T ). Now if g = uẇb ∈ G is
F -stable, with w ∈ W , u ∈ Uw− and b ∈ B, then
uẇb = g = F (g) = F (u)F (ẇ)F (b)
with v = F (w) ∈ W , F (u) ∈ Uv− and F (b) ∈ B. By uniqueness of the
decomposition, this implies that w ∈ W F and so F (ẇ) = ẇ, and F (b) = b.
Also, Uw− is clearly F -stable for w ∈ W F , so F (u) = u. It follows that
u ∈ (Uw− )F , w ∈ W F and b ∈ B F , so
)
GF = (Uw− )F ẇB F .
w∈W F

As a first consequence we obtain a result on generation:

Corollary 24.2 With the notation as in Theorem 24.1, the F -stable repre-
sentatives ẇ of w ∈ W F may be chosen in U F , (U − )F , where U = Ru (B)
and U − = U w˙0 is the unipotent radical of the opposite Borel subgroup. In
particular GF = U F , (U − )F , T F .
Proof Let I ⊆ S be an F -orbit on S, ν ∈ Ω the corresponding simple root
with root subgroups Uν ≤ U and U−ν = Uνw˙0 ≤ U − . By Corollary 23.9 there
exists 1 = u ∈ U−ν F
. By the Bruhat decomposition in Theorem 24.1 applied
to the reductive group Uν , U−ν F , this may be written as u = u1 ẇu2 with
ui ∈ UνF and ẇ a preimage in NGF (T ) of w ∈ W F . Since Uν ∩ U−ν = 1
we have w = 1, so w = wI by Lemma 23.3, whence w˙I ∈ U F , (U − )F .
Since W F is generated by the wI we may choose such representatives for
all w ∈ W F . The Bruhat decomposition for GF then shows that GF =
U F , (U − )F , T F .
We now determine the orders of the finite groups of Lie type. Recall the
length function  on W introduced in the proof of the Bruhat decomposi-
tion Theorem 11.17. According to Proposition A.21 we have the alternative
characterization
(w) = |{α ∈ Φ+ | w.α ∈ Φ− }| for w ∈ W.
 24.1 Bruhat decomposition and the order formula 205

Proposition 24.3 Let G be simple with Steinberg endomorphism F : G →

G. Let T ≤ B be an F -stable maximal torus in an F -stable Borel subgroup,
Φ the root system associated to B with set of positive roots Φ+ and Weyl
group W . Then
" "
q (w) = q |Φ | |T F |
+
|GF | = |B F | q (w) .
w∈W F w∈W F

Proof The Bruhat decomposition

)
GF = (Uw− )F ẇB F
w∈W F

from Theorem 24.1 implies that

" "
|GF | = |(Uw− )F | · |B F | = |(Uw− )F | · |T F | · |U F |
w∈W F w∈W F

(using that B F = U F · T F ). Now we have the factorization U = Ru (B) =

 −

α∈Φ+ Uα by Theorem 11.1 while Uw = α∈Φ+ ,w.α∈Φ− Uα . Clearly, {α ∈
Φ+ | w.α ∈ Φ− } is φ-invariant for w ∈ W F , so by Corollary 23.9
|w.α∈Φ− }|
|U F | = q |Φ |
|(Uw− )F | = q |{α∈Φ
+ +
and = q (w)

by the formula for (w) above.

Similarly, a formula for |GF | in terms of the qα from Proposition 22.2 can
be deduced for arbitrary connected reductive groups G, see Exercise 30.9.

The polynomial w∈W x
(w)
∈ Z[x] is called the Poincaré polynomial
of the Weyl group W . In order to obtain a factorization, we need another
interpretation of this polynomial in terms of the invariant theory of W in
its reflection representation on V := X ⊗Z R. It is a fundamental property
of finite reflection groups, i.e., finite linear groups generated by complex
reflections, that their invariants form a polynomial ring:

Theorem 24.4 (Shephard–Todd, Chevalley) Let W ≤ GL(V ) be a finite

reflection group on a real vector space V of dimension n. Then:

(a) The invariants S(V )W of W in the symmetric algebra S(V ) of V form

a polynomial algebra.
(b) Let f1 , . . . , fn denote algebraically independent generators of S(V )W ,
homogeneous of degrees di := deg fi . Then the multiset {di | 1 ≤ i ≤ n}
is uniquely determined by W .
(c) We have d1 · · · dn = |W | and d1 + · · · + dn = N + n, where N denotes
the number of reflections in W .
206 A BN-pair for GF

See [9, V, Thm. 5.3] for (a), [9, V, §5.1, Cor.] for (b), and [9, V, Cor. 5.3
and Prop. 5.3] for (c).
The d1 , . . . , dn are called the degrees of the reflection group W . Note that
for W the Weyl group of a root system Φ we have N = |Φ|/2 = |Φ+ | since
by axiom (R2), Proposition A.21 and Exercise A.6 there are precisely two
roots for each reflection in W .
Returning to our algebraic group G, the Steinberg endomorphism F acts
on V as a scalar times an automorphism φ of finite order δ, which normalizes
W (see Proposition 22.2). Thus φ also acts naturally on the ring of invari-
ants S(V )W , and the algebraically independent homogeneous generators fi
may in fact be chosen to be eigenvectors of φ. We denote the corresponding
eigenvalues by i . Thus, the i are roots of unity of order dividing δ. Clearly
all i = 1 if F acts as a scalar on V . Using that the Poincaré polynomial
is the Hilbert series of the coinvariant algebra of S(V ) one obtains (see [72,
Thm. 2.1 and Cor. 2.9]):
Proposition 24.5 Let W ≤ GL(V ) be a finite reflection group on a real
vector space V of dimension n, φ ∈ NGL(V ) (W ) and fi , 1 ≤ i ≤ n, homo-
geneous generators of S(V )W of degrees di which are eigenvectors of φ with
eigenvalues i . Then the i are also the eigenvalues of φ on V and we have
" !n
xdi − i
x(w) = .
i=1
x − i
w∈W φ

This leads to the following factorization of the order formula from Propo-
sition 24.3 (see [72, Thm. 11.16]):
Corollary 24.6 Let G be connected reductive, F : G → G a Steinberg
endomorphism. Then

!
rk(G)
|GF | = q |Φ |
+
(q di − i ),
i=1

where the di are the degrees of W and i are as defined above.

Proof for the case that G is simple We evaluate the formula in Proposi-
tion 24.3. By Lemma 22.1(b) we have |T F | = |X/(F −1)X| = | det XR (F −1)|
by the elementary divisor theorem. Using that | det(φ)| = 1 since φ is an au-
tomorphism of finite order, and that the characteristic polynomial of φ is
real, so equal to the one of φ−1 , this equals
!
n
|T F | = | det XR (qφ − 1)| = | det XR (q − φ−1 )| = (q − i ).
i=1
 24.1 Bruhat decomposition and the order formula 207

Hence the claim follows with Proposition 24.5.

It ensues that the order of GF is already determined by the underlying

complete root datum G = (X, Φ, Y, Φ∨ , W φ), together with q. We call

!
rk(G)
|G| := x|Φ |
+
(xdi − i ) ∈ Z[x]
i=1

the order polynomial of the complete root datum G. The preceding corollary
can then be rephrased as saying that |G(q)| = |G|(q), i.e., the order of
GF is the order polynomial evaluated at q. Note that deg |G| = dim G by
Theorems 8.17(b) and 24.4(c).
A further immediate consequence of the above formula is the fact that
|G(q)| is independent of the isogeny type of G, e.g., |SLn (q)| = |PGLn (q)|,
and |E6 (q)sc | = |E6 (q)ad |. We also get a divisibility property of order poly-
nomials, see Exercise 30.10:

Corollary 24.7 Let G be connected reductive with Steinberg endomorphism

F : G → G, H ≤ G a closed connected reductive F -stable subgroup, with
complete root data G, respectively H. Then |H| divides |G| in Z[x].

Example 24.8 Let’s look at some examples of order polynomials.

% &
(1) Let G = GLn , F the standard Frobenius map. Here |Φ+ | = n2 by
Example 9.8. The Weyl group acts on V = XR as the symmetric group
in its natural permutation representation, so the invariants S(V )W are
generated by the elementary symmetric polynomials, of degrees di = i,
1 ≤ i ≤ n. Since F acts as a scalar on X, i = 1 for all i. Therefore
Corollary 24.6 gives the well-known formula

n !
n
|GLn (q)| = q ( 2 ) (q i − 1).
i=1

(2) For G = SLn with the standard Frobenius map, the Weyl group acts
via the deleted permutation module on XR , hence the invariants are
generated by the elementary symmetric polynomials of degree at least 2.
We obtain
n !
n
|SLn (q)| = q ( 2 ) (q i − 1) = |PGLn (q)|.
i=2

(3) For the special orthogonal group G = SO8 , with root system of type
D4 the degrees of W are (d1 , d2 , d3 , d4 ) = (2, 4, 6, 4) and N = d1 + d2 +
208 A BN-pair for GF

d3 + d4 − n = 12. If F acts trivially on the Weyl group, so i = 1 for

1 ≤ i ≤ 4, then

8 (q)| = q (q − 1)(q − 1) (q − 1).

| SO+ 12 2 4 2 6

For the triality graph automorphism on D4 , φ is of order 3, with eigen-

values (1 , 2 , 3 , 4 ) = (1, ζ3 , 1, ζ32 ) on suitable homogeneous generating
invariants of degrees (2, 4, 6, 4), where ζ3 = exp(2πi/3) is a primitive
third root of unity. So

| 3D4 (q)| = q 12 (q 2 − 1)(q 4 − ζ3 )(q 6 − 1)(q 4 − ζ32 )

= q 12 (q 2 − 1)(q 6 − 1)(q 8 + q 4 + 1).

The list of order polynomials for the various types is given in Table 24.1
(see for example [13, Prop. 10.2.5 and Thm. 14.3.2]).

Table 24.1 Orders of finite groups of Lie type

GF |GF |
n
SLn (q) q ( 2 ) (q 2 − 1)(q 3 − 1) · · · (q n − 1)
n
SUn (q) q ( 2 ) (q 2 − 1)(q 3 + 1) · · · (q n − (−1)n )
2
SO2n+1 (q) q n (q 2 − 1)(q 4 − 1) · · · (q 2n − 1)
2
Sp2n (q) q n (q 2 − 1)(q 4 − 1) · · · (q 2n − 1)
2
SO+2n (q) q n −n (q 2 − 1)(q 4 − 1) · · · (q 2n−2 − 1)(q n − 1)
2
SO−2n (q) q n −n (q 2 − 1)(q 4 − 1) · · · (q 2n−2 − 1)(q n + 1)
G2 (q) q 6 (q 2 − 1)(q 6 − 1)
3
D4 (q) q 12 (q 2 − 1)(q 6 − 1)(q 8 + q 4 + 1)
F4 (q) q 24 (q 2 − 1)(q 6 − 1)(q 8 − 1)(q 12 − 1)
E6 (q) q 36 (q 2 − 1)(q 5 − 1)(q 6 − 1)(q 8 − 1)(q 9 − 1)(q 12 − 1)
2
E6 (q) q 36 (q 2 − 1)(q 5 + 1)(q 6 − 1)(q 8 − 1)(q 9 + 1)(q 12 − 1)
E7 (q) q 63 (q 2 − 1)(q 6 − 1)(q 8 − 1)(q 10 − 1)(q 12 − 1)(q 14 − 1)(q 18 − 1)
E8 (q) q 120 (q 2 − 1)(q 8 − 1)(q 12 − 1)(q 14 − 1)(q 18 − 1)(q 20 − 1)(q 24 − 1)
·(q 30 − 1)
2
B2 (q 2 ) q 4 (q 2 − 1)(q 4 + 1) (q 2 = 22f +1 )
2
G2 (q 2 ) q 6 (q 2 − 1)(q 6 + 1) (q 2 = 32f +1 )
2
F4 (q 2 ) q 24 (q 2 − 1)(q 6 + 1)(q 8 − 1)(q 12 + 1) (q 2 = 22f +1 )

Remark 24.9 It is easy to see from the formulas in Table 24.1 that most of
the time the defining prime occurs to a much higher power in |GF | than any
other prime. From this, and using Zsigmondy’s Theorem (see Theorem 28.3),
it is straightforward to check for coincidences of orders among the (generally
 24.2 BN-pair, simplicity and automorphisms 209

simple, see Theorem 24.17) groups GF /Z(GF ) for G simple of simply con-
nected type. Apart from those coming from an isomorphism of the underlying
root system (like A3 ∼= D3 or B2 ∼= C2 , for example), or from the isogeny
between type Bn and Cn in characteristic 2 (viz. Sp2n (2f ) ∼
= SO2n+1 (2f ), see
Remark 18.14(a)) there are only the following three cases of isomorphisms
PSL2 (4) ∼
= PSL2 (5), PSL2 (7) ∼
= PSL3 (2), PSU4 (2) ∼
= PSp4 (3),
and the following infinite series of order coincidences between non-isomorphic
groups:
|PSp2n (q)| = |SO2n+1 (q) | for q odd, n ≥ 3
(see for example [44, §2.9]).

24.2 BN-pair, simplicity and automorphisms

The Bruhat decomposition allows us to exhibit a BN-pair in GF (see Defi-
nition 11.15):
Theorem 24.10 Let G be connected reductive with Steinberg endomor-
phism F : G → G, T ≤ B an F -stable maximal torus in an F -stable Borel
subgroup of G, N := NG (T ). Then B F , N F is a BN-pair in GF with Weyl
group W F .
Proof The Bruhat decomposition in Theorem 24.1 shows that GF is gen-
erated by B F and N F , whence (BN1). Next, B F ∩ N F = (B ∩ N )F = T F is
normal in N F = NGF (T ), showing (BN2). By Proposition 23.2 we have
N F /(B F ∩ N F ) = N F /T F ∼
= (N/T )F = W F ,
which is generated by the set of involutions {sI | I ⊆ S a φ-orbit} by
Lemma 23.3. For (BN4) let w ∈ W F , and sI corresponding to a φ-orbit
I ⊆ S, so sI = s1 · · · sr for suitable si ∈ I. Then an easy induction, using
(BN4) for the group G, shows that
B ẇB · B ṡI B ⊆ B wvB.
˙
v∈WI

A double coset B wvB ˙ on the right-hand side is F -stable if and only wv is,
that is, if v ∈ W F . But by Lemma 23.3 the only F -stable elements in WI
are 1 and sI , whence taking F -fixed points we find (BN4).
Finally, let s = sI with preimage ṡ ∈ N F and set U := Us− . Then
!
ṡU ṡ = U−α | α ∈ Φ+ , s(α) ∈ Φ−  ≤ Uα ≤ B w˙0 ,
α∈Φ−
210 A BN-pair for GF

so B ∩ ṡU ṡ ≤ B ∩ B w˙0 = T by Corollary 11.18. On the other hand, U is F -

stable, so U F contains non-trivial unipotent elements by Proposition 23.8.
This shows that ṡB F ṡ ≥ ṡU F ṡ cannot be contained in B F , thus (BN5)
holds.
This means that all general results on groups with a BN-pair which were
derived in Sections 11.2 and 12.1 also hold for the group GF . In particular,
there is a natural notion of parabolic subgroups and Levi subgroups of GF ,
coming from its BN-pair structure. We will investigate the relation between
these and the F -fixed points of parabolic subgroups of G later in Section 26.1.
For the moment, let’s just observe the following:
Corollary 24.11 In the situation of Theorem 24.10, let U = Ru (B). Then
U F is a Sylow p-subgroup of GF , with normalizer NGF (U F ) = B F .
Proof By Proposition 12.2, NGF (U F ) is a parabolic subgroup of GF . By
(BN5), it does not contain any simple reflection, so it equals B F . By Propo-
sition 23.2 we have
|B F : U F | = |B F /U F | = |(B/U )F | = |T F |.
Since T consists of semisimple elements, this cardinality is prime to the field
characteristic p. So the p-subgroup U F of GF has index prime to p in its
normalizer. By elementary group theory this implies that U F is a Sylow
p-subgroup (see, for example [2, (9.10)]).
One may be tempted to expect that for G a simple algebraic group, GF
is again close to being a (finite) simple group. In fact, most of the argument
in the proof of Proposition 12.5 only depends on the BN-pair axioms, and
we obtain:
Lemma 24.12 Let G be a simple algebraic group with Steinberg endomor-
phism F : G → G. Any normal subgroup H of GF either satisfies B F H = GF
or H ≤ Z(GF ).
Proof Let T ≤ B denote an F -stable maximal torus inside an F -stable
Borel subgroup of G. Since GF has a BN-pair by Theorem 24.10, the argu-
ments given in the proof of Proposition 12.5 apply to the present situation to
show that any proper normal subgroup H of GF either lies in B F or satisfies
B F H = GF .
If H ≤ B F , then it is contained in all GF -conjugates of B F . The set
of simple roots ∆ with respect to B is F -stable, hence so is the unique
element w0 sending it to −∆ (the longest element of W ). But B ∩ B ẇ0 = T
by Corollary 11.18. So any proper normal subgroup H ≤ B F of GF lies
 24.2 BN-pair, simplicity and automorphisms 211

in the abelian group T F . Let U := Ru (B). Then [U F , H] ≤ U F ∩ H ≤

U ∩ T = 1, whence H ≤ Z(B F ), and similarly H ≤ Z((B w˙0 )F ). Since
GF = B F , (B F )w˙0  by Corollary 24.2 this shows that H ≤ Z(GF ).
Corollary 24.13 Let G be a simple algebraic group with Steinberg endo-
morphism F : G → G. Then Z(GF ) = Z(G)F .
In particular, for G of simply connected type, Z(GF ) is as given in Ta-
ble 24.2. Here, an entry d stands for a cyclic group of that order, while
(2, q − 1)2 denotes a group Zd × Zd , with d = (2, q − 1).

Table 24.2 Centers of groups of simply connected type

GF Z(GF ) GF Z(GF )
SLn (q), n ≥ 2 (n, q − 1) 2
B2 (22f +1 ) 1
SUn (q), n ≥ 3 (n, q + 1) 2
G2 (32f +1 ) 1
Spin2n+1 (q), n ≥ 3 (2, q − 1) G2 (q) 1
Sp2n (q), n ≥ 2 (2, q − 1) 3
D4 (q) 1
Spin+2n (q), n ≥ 4 even (2, q − 1)2 2
F4 (22f +1 ) 1
Spin+2n (q), n ≥ 5 odd (4, q − 1) F4 (q) 1
Spin−2n (q), n ≥ 4 even (2, q − 1) E6 (q) (3, q − 1)
Spin−2n (q), n ≥ 5 odd (4, q + 1) 2
E6 (q) (3, q + 1)
E7 (q) (2, q − 1)
E8 (q) 1

Proof We have Z(G)F ≤ Z(GF ) ≤ T F , the latter inclusion by the previous

proof. Now assume that t ∈ T F \ Z(G)F . Then by Theorem 8.17 there exists
α ∈ Φ with α(t) = 1. But then also F i (α)(t) = α(F i (t)) = α(t) = 1 for all i,
so t acts fixed-point freely on the root subgroup Uω , with ω the equivalence
class of α as in Lemma 23.4. Thus, t ∈/ Z(GF ) by Corollary 23.9.
By Proposition 9.15 the center Z(G) of a simple group G of simply con-
nected type is isomorphic to the p -part of the fundamental group Λ of the
root system of G, as given in Table 9.2. The center Z(GF ) can then be
computed by the formula in the first part.
Corollary 24.14 Let G be a simple algebraic group with Steinberg endo-
morphism F : G → G. If GF is perfect, then GF /Z(GF ) is a finite simple
group.
Proof By Lemma 24.12 any normal subgroup H of GF lies in Z(GF ) or
satisfies B F H = GF . In the latter case
GF /H = B F H/H ∼
= B F /(B F ∩ H),
212 A BN-pair for GF

with the right-hand side being solvable since B is, while GF is perfect, so
H = GF . The result follows.
This criterion leads one to ask under which conditions GF is perfect, and
how this can be verified. For this, let’s set G1 := GF
u , the normal (in fact,
characteristic) subgroup of GF generated by its unipotent elements.
Theorem 24.15 (Steinberg) Let G be a simply connected semisimple linear
algebraic group with Steinberg endomorphism F : G → G. Then GF = G1 ,
that is, GF is generated by its unipotent elements.
Proof Let B ≤ G be an F -stable Borel subgroup of G with F -stable
maximal torus T and unipotent radical U , and U − = U w˙0 the unipotent
radical of the opposite Borel subgroup. By Corollary 24.2 we have that
GF = U F , (U − )F , T F . Since T F normalizes both U F and (U − )F , we con-
clude that G1 = U F , (U − )F .
Thus it remains to show that T F ≤ U F , (U − )F . Now for α ∈ ∆ let
Gα = Uα , U−α  and Tα = T ∩Gα , a one-dimensional torus (see Section 8.3).
Since G is simply connected, the coroots {α∨ : k × → Tα ≤ T | α ∈ ∆}

generate Y (T ), so T = α∈∆ Tα . Collecting according to F -orbits we obtain

the F -invariant decomposition T = I⊆S TI , where I runs over F -orbits in
 
S and TI := α∈∆I Tα , so T F = I⊆S TIF .
We are thus reduced to the case that S is a single F -orbit. Moreover, by
Exercise 30.2 we may assume that G is simple. But then Φ has root system
of type A1 , A2 , B2 or G2 . In the second case, GF = SU3 (q) and the claim
can be checked by direct computation, as in Exercise 30.11. In the other
three cases, W F = {1, w = wS }. Let t ∈ T F and t1 ∈ T with t21 = t. Then
ẇt1 ẇ−1 = t−1 −1
1 , so t1 ẇt1 = ẇt. Recall from Corollary 24.2 that we may
choose ẇ ∈ G1 . If char(k) = 2 then ẇ and ẇt are semisimple elements of GF
which are G-conjugate, so also GF -conjugate by Theorem 26.7(c) below. In
particular, as we chose ẇ ∈ G1 , ẇt ∈ G1 and so t ∈ G1 . If char(k) = 2 this
also holds, since then |T F | is odd, so t1 may already be chosen in T F .
Example 24.16 Let G = SL2 with Steinberg endomorphism F , so GF =
SL2 (q) for some prime power q by the classification of Steinberg endo-
morphisms. We claim that G1 = GF u  is perfect for q > 3 and equal to
F
G . Indeed, by Theorem 24.15 SL2 (q) is generated by its root subgroups
F
U±α = {u±α (c) | c ∈ Fq }, where



1 c 1 0
uα (c) := , u−α (c) := ,
0 1 c 1
so GF = G1 . Since q > 3 there exists a ∈ F×
q \ {±1}. Set t := diag(a, a
−1
)∈
 24.2 BN-pair, simplicity and automorphisms 213

GF . Then tu±α (c)t−1 = u±α (a±2 c), so by the choice of a the map
F
U±α −→ U±α
F
, u±α (c) → [t, u±α (c)] = u±α ((a±2 − 1)c),
is surjective. Thus U±αF
≤ [GF , GF ], and so GF is perfect. Note that SL2 (q)
is solvable for q ≤ 3.
Similarly using Example 23.10(2) one can check that SU3 (q) is perfect for
q > 2 (see Exercise 30.11).
This example generalizes to yield perfectness and hence simplicity modulo
center in the simply connected case.
Theorem 24.17 (Tits) Let G be a simply connected simple linear algebraic
group with Steinberg endomorphism F : G → G. Then, unless GF is one of
SL2 (2), SL2 (3), SU3 (2), Sp4 (2), G2 (2), 2 B2 (2), 2 G2 (3), 2 F4 (2),
G1 = GF is perfect, and thus GF /Z(GF ) is simple.
Proof when F is not very twisted and q > 3 If F is not very twisted, then
an orbit ω of F on the set of simple roots of G either consists of mutually
orthogonal roots, so is of type A1 × · · · × A1 , or it is of type A2 . In the first
case, the various U±α , for α ∈ ω, commute pairwise and are permuted
transitively by F , so
U±α | α ∈ ωF ∼
m
= U±β F
for m = |ω| and any β ∈ ω, by Exercise 30.2. But Gβ := U±β  is a central
quotient of SL2 by Section 8.2 and Theorem 8.17(f), so an application of
Example 24.16 shows that (Gβ )1 is perfect, and similarly when ω is of type
A2 (see Exercise 30.11). Now by Proposition A.11 any root subgroup of
GF = G1 is conjugate to a root subgroup corresponding to a simple root.
So [G1 , G1 ] contains all root subgroups of G1 , hence a Sylow p-subgroup by
Corollary 24.11, and thus equals G1 . Now apply the previous theorem and
Corollary 24.14.
A slight variation of the above argument applies when q ≤ 3. The groups
of type 2 B2 and 2 G2 have to be treated differently. Then, the groups of type
2
F4 can again be handled as before, see [82].
Remark 24.18 The groups SL2 (2), SL2 (3), SU3 (2), 2 B2 (2) occurring in
the exceptions of Theorem 24.17 are solvable, the groups Sp4 (2) ∼ = S6 ,
G2 (2) ∼ = Aut(PSU3 (3)), 2 G2 (3) ∼= Aut(PSL2 (8)) are almost simple, and
2
F4 (2) contains a normal simple subgroup of index 2 which does not occur
elsewhere in the classification of finite simple groups, the so-called Tits group
(see [82, 4.3]).
214 A BN-pair for GF

Remark 24.19 The fact that GF in Theorem 24.17 is perfect and S =

GF /Z(GF ) is simple implies by definition that GF is a so-called covering
group of S. In fact, whenever GF is perfect then, except for the finitely
many cases in Table 24.3 it is the full covering group of the simple group
S, that is, Z(GF ) as given in Table 24.2 is the Schur multiplier M (S) of S
(see e.g. [29, Table 6.1.3]). Here O7 (3) = Spin7 (3)/Z(Spin7 (3)), which is also
isomorphic to the derived subgroup of SO7 (3) (see Proposition 24.21 below).

Table 24.3 Non-generic Schur multipliers

S M (S) S M (S) S M (S)

PSL2 (4) Z2 PSU4 (3) Z12 × Z3 2
B2 (8) Z 2 × Z2
PSL2 (9) Z6 PSU6 (2) Z 6 × Z2 G2 (3) Z3
PSL3 (2) Z2 Sp6 (2) Z2 G2 (4) Z2
PSL3 (4) Z12 × Z4 O7 (3) Z6 F4 (2) Z2
PSL4 (2) Z2 SO+8 (2) Z2 × Z2 2
E6 (2) Z 6 × Z2
PSU4 (2) Z2

Some of the exceptional multipliers are “explained” by exceptional auto-

morphisms, see Remark 24.9. It may be noted that the generic Schur mul-
tiplier always has order prime to the characteristic p, while the exceptional
part has order a power of p.

In order to investigate the general, non-simply connected case, we need the

following elementary result (see Proposition 23.2 for a related statement):

Lemma 24.20 Let H be a group, Z ≤ H a central subgroup, and F : H →

H an automorphism normalizing Z. Then the short exact sequence

1 −→ Z −→ H −→ H/Z −→ 1

induces a long exact sequence

1 −→ Z F −→ H F −→ (H/Z)F −→ (L(H) ∩ Z)/L(Z) −→ 1,

where L : H → H is defined by L(h) := F (h)h−1 .

For the proof, see Exercise 30.12.

Proposition 24.21 Let G be a simple algebraic group, π : Gsc → G the

natural isogeny from a group of simply connected type with central kernel
from Proposition 9.15. Let F be a Steinberg endomorphism on Gsc normal-
izing Z := ker(π). Then:
 24.2 BN-pair, simplicity and automorphisms 215

(a) GF /π(GF ∼
sc ) = Z/L(Z) and
F ∼
(b) π(Gsc ) = Gsc /Z F .
F

F ∼
In particular, if GF F F F F
sc is perfect then π(Gsc ) = [G , G ] = Gsc /Z , and
F F
[Gad , Gad ] is simple.
Proof We apply Lemma 24.20 with H = Gsc . As Gsc is connected the map
L : Gsc → Gsc , g → F (g)g −1 , is surjective by Theorem 21.7 and we obtain
the exact sequence
π
1 −→ Z F −→ GF
sc −→ G −→ Z/L(Z) −→ 1,
F

which yields (a) and (b). Clearly, if GF F

sc is perfect, then so is its image π(Gsc ),
which must hence be contained in [G , G ]. On the other hand, G /π(GF
F F F
sc )
F F F
is abelian by (a), so we have equality [G , G ] = π(Gsc ).
Finally, if G = Gad is of adjoint type, then ker(π) = Z(Gsc ) (see Sec-
tion 9.2), so [GF , GF ] ∼
= GF sc /Z
F
= GF F
sc /Z(Gsc ) (see Corollary 24.13) is
simple by Theorem 24.17.
Example 24.22 Let G = PGLn with standard Frobenius map F so that
GF = PGLn (q). This is a solvable group for (n, q) ∈ {(2, 2), (2, 3)}. Other-
wise, using that Gsc = SLn we see from Proposition 24.21 that the derived
subgroup equals
[PGLn (q), PGLn (q)] ∼
= SLn (q)/Z(SLn (q)) = PSLn (q) = G1 .
Since |SLn (q)| = |PGLn (q)|, this has index |Z(SLn )F | = |Z(SLn (q))| =
gcd(n, q − 1) in GF and is simple (compare with the situation over k where
PGLn ∼ = SLn /Z(SLn ) as abstract groups).
The automorphisms of the simple group [GF F F
ad , Gad ] induced by Gad are
called diagonal (since for Gad = PGLn they are induced by conjugation
with diagonal matrices).
Further automorphisms of finite groups of Lie type arise as follows: Let G
be semisimple. The endomorphism Fp : G → G from Theorem 16.5 is an Fp -
split Steinberg endomorphism of G (see Example 22.6) and thus all Fq -split
Steinberg endomorphisms, where q = pf , are obtained as powers F = Fpf .
Then clearly Fp stabilizes GF ; the induced automorphism Fp : GF → GF
and its powers are called field automorphisms of GF . Furthermore, as stated
in Example 22.6 the graph automorphisms σ : G → G from Theorem 11.12
induced by symmetries of the Dynkin diagram of G can be chosen so as to
commute with Fp , so restrict to automorphisms σ : GF → GF ; these are
called the graph automorphisms of GF . Clearly, any automorphism of GF
descends to the characteristic subgroup [GF , GF ].
216 A BN-pair for GF

In fact, these already generate the full automorphism group. Let’s first
consider the case of groups of rank 1:
Proposition 24.23 Any automorphism of S = PSL2 (pf ) is the product of
an inner, a diagonal and a field automorphism.
This is an easy exercise, see for example [37, Aufg. 11.15].
Theorem 24.24 (Steinberg) Let Gsc be simple of simply connected type,
F : Gsc → Gsc a split Frobenius map, G = Gad the adjoint type image
under the natural isogeny, and assume that GF sc is perfect (and hence S :=
[GF , GF ] is simple). Then any element of Aut(S) is the product of an inner,
a diagonal, a field and a graph automorphism.
Proof for the case that all roots of G have the same length Let T ≤ B be
a maximal torus inside a Borel subgroup of G with root system Φ and set
of simple roots ∆. As pointed out above, we may assume that F = Fpf
for the endomorphism Fp from Theorem 16.5, for some f ≥ 1, and T, B
are F -stable and F acts trivially on Φ. Let U = Ru (B) be the unipotent
radical of B. Then P := U F is a Sylow p-subgroup of GF with normalizer
B F by Corollary 24.11. By Proposition 24.21, GF /S consists of semisimple
elements, so P is a Sylow p-subgroup of S. Thus, by Sylow’s Theorem any
σ ∈ Aut(S) can be multiplied by an inner automorphism so as to stabilize P .
We claim that we may modify σ further so as to also stabilize P − :=
(U − )F = (U ẇ0 )F , a second Sylow p-subgroup of S. Indeed, σ(P − ) must be
S-conjugate to P , by an element g = uẇb ∈ S with w ∈ W F , u ∈ (Uw− )F
and b ∈ B F (see Theorem 24.1); as B normalizes U we have σ(P − ) =
uẇP ẇ−1 u−1 . Now
uẇP ẇ−1 u−1 ∩ P = σ(P − ) ∩ σ(P ) = σ(P − ∩ P ) ≤ σ(U − ∩ U ) = 1
by Corollary 11.18, whence ẇP ẇ−1 ∩ P = 1 by conjugation by u ∈ (Uw− )F ≤
U F = P . If α ∈ Φ+ is not made negative by w, then ẇUαF ẇ−1 ≤ P , so w
must send all positive roots to negative ones and thus w = w0 . So σ(P − ) =
uw˙0 P w˙0 −1 u−1 = uP − u−1 . Changing σ by the inner automorphism induced
by u (which stabilizes P ) then gives σ(P − ) = P − .
Since σ fixes B F , it permutes the minimal parabolic subgroups above B F ,
which are of the form B F ∪ B F s˙α B F for some α ∈ ∆ by Proposition 12.2
(and the subsequent remark). Now U − ∩ B ṡα B = U−α , so σ induces a
permutation on the set {U−α F
| α ∈ ∆}, and similarly a permutation ρ on
{Uα | α ∈ ∆}. Note that these two permutations must agree since Uα , U−β
F

commute for distinct α, β ∈ ∆, by the commutator formulas in Theorem 11.8,

while [UαF , U−α
F
] = 1.
 24.2 BN-pair, simplicity and automorphisms 217

Furthermore, ρ induces a graph automorphism of the Dynkin diagram

since α, β are linked in the Dynkin diagram if and only if [UαF , UβF ] = 1 (recall
that we assume that all roots have the same length). By Theorem 11.12 and
the remarks preceding this theorem there exists a graph automorphism of
GF corresponding to the inverse permutation ρ−1 ; its composition with σ
fixes the UαF , α ∈ ±∆.
So we now have σ(uα (1)) = uα (aα ) for all α ∈ ∆ and suitable aα ∈ Fq .
We claim that we may change σ by a diagonal automorphism so that aα = 1
for α ∈ ∆. Indeed, since ∆ is linearly independent, there exists t ∈ T with
α(t) = aα for all α ∈ ∆. Now F (α(t)) = F (aα ) = aα = α(t) for all α ∈ ∆,
so F (χ)(t) = χ(t) for all χ ∈ Z∆ = X (since G is of adjoint type), whence
t ∈ T F . Replacing σ by its composition with the diagonal automorphism
induced by t ∈ GF we thus obtain that σ(uα (1)) = uα (1) for α ∈ ∆.
The case of rank 1 is settled by Proposition 24.23. Otherwise, let α1 , α2 ∈
∆ be linked in the Dynkin diagram, so that α3 = α1 + α2 ∈ Φ. Then
σ(uαi (c)) = uαi (fi (c))
for suitable functions fi : Fq → Fq , 1 ≤ i ≤ 3, with f1 (1) = f2 (1) = 1.
Since uαi is additive, so is fi . Moreover, by the commutator relations in
Theorem 11.8 and Exercise 20.6 we have
[uα1 (c1 ), uα2 (c2 )] = uα3 (±c1 c2 ),
so application of σ gives that
f3 (±c1 c2 ) = ±f1 (c1 )f2 (c2 )
for all c1 , c2 ∈ Fq , so f3 (c1 c2 ) = f1 (c1 )f2 (c2 ). Choosing c1 = 1 we get that
f3 (c2 ) = f2 (c2 ), and similarly f3 (c1 ) = f1 (c1 ), hence f1 = f2 = f3 is also
multiplicative. So in fact fi is a field automorphism of Fq . Composing σ with
the inverse field automorphism of S, we reach that σ(uα (c)) = uα (c) for all
α ∈ ∆, c ∈ Fq .
Thus for α ∈ ∆, σ stabilizes UαF , U−α F
 and is the identity on UαF . By
F
Proposition 24.23 this forces σ to also centralize U−α , hence some represen-
tative s˙α of the simple reflection sα ∈ W . As W is generated by the sα , and
F

U F , (U − )F are products of suitable W F -conjugates of the UαF , σ centralizes

S by Theorem 24.15.
See [73, Thm. 30] or [13, Thm. 12.5.1] for the case of different root lengths.
If F is twisted, the statement must be modified in so far as no graph au-
tomorphisms do arise, see [73, Thm. 36] for the statement and its proof in
that case.
 25
Tori and Sylow subgroups

We now turn to the subgroup structure of the finite groups of Lie type,
first discussing maximal tori and then some aspects of Sylow subgroups. Let
G be connected reductive with Steinberg endomorphism F , B an F -stable
Borel subgroup of G. Recall that p = char(k). By Corollary 24.11 (or by the
order formula in Corollary 24.6) U F = Ru (B)F is a Sylow p-subgroup of GF .
Some results on the structure of U F were obtained in Section 23.2; see also
Theorem 26.5 below. Here we will be concerned with Sylow subgroups of GF
for the other prime divisors of its order. Again, there is a close connection
to the structure of the algebraic group G.

25.1 F -stable tori

Let Φd (x) ∈ Z[x] be the dth cyclotomic polynomial over Q, i.e., the monic
irreducible polynomial whose zeros are the primitive dth roots of unity, so

that xm − 1 = d|m Φd (x). Then we may factorize the order formula from
Corollary 24.6 in Z[x] to obtain

!
rk(G)
!
|G| = x|Φ |
(xdi − i ) = x|Φ |
+ +
Φd (x)a(d)
i=1 d≥1

for suitable integers a(d) ≥ 0. Thus the irreducible factors of the generic
order |G| are the Φd (x) with a(d) > 0, so the Φd (q) can in some sense be
considered as the “generic primes” dividing |G|(q) = |GF |. In this section
we will substantiate this statement. In order to do this, we first need more
precise information on the classification of F -stable maximal tori and the
orders of their F -fixed points.
 25.1 F -stable tori 219

Proposition 25.1 Let G be connected reductive with Steinberg endomor-

phism F : G → G, T ≤ G an F -stable maximal torus with Weyl group W .
There is a natural bijection
 
GF -classes of F -stable 1−1
←→ φ-conjugacy classes in W .
maximal tori of G
Proof Apply Theorem 21.11 with G acting on the set of maximal tori of
G. The stabilizer NG (T ) is closed, and NG (T )◦ = T by Theorem 3.10 and
Corollary 8.13, so the GF -classes of F -stable conjugates of T are in natural
bijection with the F -classes (hence the φ-classes, where F |X(T ) = qφ) in
NG (T )/T = W .

More precisely, according to the proof of Theorem 21.11, if the conjugate

gT g −1 of T is F -stable, then it corresponds to the element w := g −1 F (g)T ∈
NG (T )/T = W . We’ll write Tw := gT g −1 for such an F -stable maximal torus
of G corresponding to the φ-class of w ∈ W .
We next compute the orders of F -fixed points of maximal tori.

Proposition 25.2 Let G be connected reductive with Steinberg endomor-

phism F : G → G, T ≤ G an F -table maximal torus. Then

|T F | = | det XR (F − 1)| = det XR (q − (φ)−1 ),

where XR := X(T ) ⊗Z R, and F |X = qφ.

Proof By Lemma 22.1 we have T F ∼ = X(T )/(F −1)X(T ). The first equality
is thus a direct consequence of the elementary divisor theorem. With F |X =
qφ we get det(F − 1) = det(φ) det(q − (φ)−1 ). Since φ has finite order, all its
eigenvalues are roots of unity. So | det(φ)| = 1, and moreover all eigenvalues
of q − (φ)−1 have real part at least q − 1 > 0, whence the real number
det(q − (φ)−1 ) is positive.

The parametrization in Proposition 25.1 will now allow us to relate the

orders of all maximal tori to the characteristic polynomials of elements on
the character group of one fixed reference torus, but this translation (conju-
gation) introduces an element w ∈ W .

Proposition 25.3 In the situation of Proposition 25.1 let X denote the

character group of the F -stable maximal torus T . Then we have:

(a) NGF (Tw )/TwF ∼ = CW (wφ).

(b) TwF ∼= X/(wF − 1)X.
(c) |TwF | = | detX⊗R (wF − 1)| = detX⊗R (q − (wφ)−1 ).
220 Tori and Sylow subgroups

Proof Note that T = T1 with the identity element 1 ∈ W in our above

parametrization. For Tw = gT1 g −1 with w = g −1 F (g)T1 ∈ W we have
NG (Tw ) = gNG (T1 )g −1 , and NGF (Tw )/TwF ∼
= (NG (Tw )/Tw )F by Proposi-
tion 23.2. An element gng −1 Tw ∈ NG (Tw )/Tw , with n ∈ NG (T1 ), is F -stable
if and only if
ng −1 F (g)T1 = g −1 F (g)F (n)T1 .

Writing n̄ for the image of n in NG (T1 )/T1 = W this is the case if and only
if n̄w = wφ(n̄), so if and only if n̄ lies in the centralizer CW (wφ), giving (a).
For (b) note that F sends gtg −1 ∈ gT1 g −1 = Tw to gwF (t)w−1 g −1 , so
it acts on Tw like wF acts on T1 . The claim is then just the assertion of
Lemma 22.1(b). Part (c) is Proposition 25.2.

In particular, since F acts on X(Tw ) as wF acts on X(T ), we see that if

(G, F ) has complete root datum (X, Φ, Y, Φ∨ , W φ) with respect to T , then
the torus Tw has complete root datum (X, ∅, Y, ∅, wφ).
Recall that NGF (Tw ) = NGF (TwF ) in general, see the remarks after Propo-
sition 23.2.

Example 25.4 We describe the classes of maximal tori in GLn (q).

(1) For G = GLn with Fq : G → G the standard Frobenius map with fixed
point group GFq = GLn (q) we let T = Dn denote the maximally split
torus of diagonal matrices. For g ∈ GLn with
⎛ ⎞
0 ··· 0 1
⎜ 1 0. ⎟
g −1 Fq (g) = w := ⎜
⎝ 0 ...
⎟
.. ⎠ ∈ GLn
0 0 1 0

let Tw = gT g −1 , an Fq -stable maximal torus of G. By Proposition 25.3,

Fq acts on Tw like F = wFq acts on T . Now for t = diag(t1 , . . . , tn ) ∈
T = Dn we have F (t) = diag(tqn , tq1 , . . . , tqn−1 ). So
  

t 1 . 0  2 n−1 n
∼
= F×
q q q q
TF = ..  t2 = t1 , t3 = t1 , . . . , t n = t1 , t1 = t1 qn ,
0 tn 

as predicted by the order formula |T F | = | det(wq − 1)| = q n − 1 from

Proposition 25.2. Now w is the preimage of an n-cycle in the Weyl group
W = NG (T )/T ∼ = Sn of G, so its centralizer in W is cyclic of order n.
According to Proposition 25.3(a) we then have NGLn (q) (T ) = T F · Zn .
The generators of this cyclic torus of GLn (q) are sometimes called Singer
cycles.
 25.1 F -stable tori 221

Let’s construct this subgroup even more explicitly. As Fqn /Fq is a

field extension of degree n, the additive group F+ ∼ n
q n = (Fq ) is an n-
×
dimensional vector space over Fq . Multiplication by a ∈ Fqn is an in-
vertible linear map on this vector space, thus any choice of basis defines
an embedding F× ∼
q n → GLn (q). Also, Gal(Fq n /Fq ) = Zn acts by linear
maps on this space. Putting these together, we have obtained an embed-
ding of the semidirect product

F×
q n
Gal(Fq n /Fq ) → GLn (q)

whose image, by comparing with the structure obtained above, is all of

NGF (T ).
(2) More generally, let n1 + · · · + nr = n be a partition of n and GLn1 × · · · ×
GLnr ≤ GLn a Levi complement of a corresponding parabolic subgroup
(see Example 12.9). In each factor GLni choose a maximal F -stable torus
Ti with |TiF | = q ni − 1 as in (1) above, and let T := T1 × · · · × Tr ≤ GLn .
As dim Ti = ni , T is a maximal, F -stable torus of G with
!
r !
r
|T F | = |TiF | = (q ni − 1),
i=1 i=1

corresponding under the parametrization in Proposition 25.1 to the class

of permutations v ∈ Sn of cycle shape (n1 , . . . , nr ).

As a consequence of our description of F -stable maximal tori, we can now

derive a formula for their number:

Theorem 25.5 (Steinberg) Let G be connected reductive with Steinberg

endomorphism F : G → G. Then the number of F -stable maximal tori of G
is q 2N where N := |Φ+ |.

Proof By Proposition 25.1 the GF -classes of F -stable maximal tori of G

are in bijection with the φ-conjugacy classes in the Weyl group W , where
F |XR = qφ. The GF -class of a maximal torus Tw parametrized by the F -class
of w ∈ W contains |GF : NGF (Tw )| distinct conjugates. Thus the number in
question equals
" 1
|GF : NGF (Tw )|,
|[wφ]|
w∈W

where [wφ] denotes the φ-class of w. Here, |NGF (Tw )| = |CW (wφ)| · |TwF |
by Proposition 25.3(a). Now by Proposition 25.3(c) and using that |[wφ]| =
222 Tori and Sylow subgroups

|W |/|CW (wφ)| (see Exercise 30.3), the above quantity equals

1 " |GF | |GF | " 1

=
|W | |TwF | |W | | det(wF − 1)|
w∈W w∈W
|GF | " 1
= ,
|W | q l det(1 − (qwφ)−1 )
w∈W

where l = dim XR = dim(Tw ) = rk(G). By Molien’s formula (see Exer-

cise 30.15) this can be rewritten as

|GF | ! q N ! q di − i
l l
1 N +d1 +...+dl −l
= −1 −di = q = q 2N ,
q l i=1 1 − −1
i q −di q l
i=1
1 −  i q

where we’ve also used Theorem 24.4(c) and Corollary 24.6. (Note that the
multiset {(di , i )} is invariant under complex conjugation, since φ is a real
endomorphism.)

Torus orders are products of cyclotomic polynomials, which naturally leads

to the following notion:

Definition 25.6 Let G be a connected reductive group with a Steinberg

endomorphism F : G → G. An F -stable torus S ≤ G is called a d-torus
if its generic order equals |S| = Φd (x)a for some a ≥ 0, where S denotes
the complete root datum corresponding to (S, F ). So in particular |S F | =
|S|(q) = Φd (q)a . We call S a Sylow d-torus of G if |S| = Φd (x)a(d) is the
precise power of Φd (x) dividing the generic order |G|.

In view of Lemma 22.1(b) a torus S is a d-torus if and only if the charac-

teristic polynomial of φ on its character group (as well as on its cocharacter
group) is a power of Φd (x).
The structure of d-tori is very simple:

Proposition 25.7 Let S be a d-torus of G of rank r. Then S is a direct

product of r/ϕ(d) F -stable tori of generic order Φd , where ϕ(d) = deg(Φd ).

Proof By our previous remark Φd (φ) = 0 on the character group X of S.

We may thus define an action of the ring R := Z[ζ] on X by letting ζ act
as φ, where ζ denotes a primitive dth root of unity. This endows X with the
structure of a projective module of finite rank over R. Since R is the ring of
integers in Q(ζ), hence a Dedekind domain, X is a direct sum of projective
submodules of rank 1 over R, hence of φ-stable submodules of rank ϕ(d)
over Z. (See, for example [25, Chap. 2] for the theory of Dedekind domains.)
 25.1 F -stable tori 223

Since ζ acts as φ we have by Lemma 22.1

SF ∼
= X/(F − 1)X ∼
= X/(qζ − 1)X ∼
= X/(q − ζ −1 )X

as R-modules.
So from now on we may assume that X is a projective R-module of rank 1,
hence can be identified with an ideal of R. Then multiplying by its inverse
we obtain X/(q − ζ −1 )X ∼= R/(q − ζ −1 )R, whose order equals the norm of
−1
the principal ideal (q − ζ )R, so that |S F | = Φd (q) in this case. If n ∈ N
with nX ⊆ (q − ζ −1 )X, then q − ζ −1 divides n in R. As Z[ζ] = Z[x]/(Φd )
this implies n ≡ q − x (mod Φd ), so n ≡ 0 (mod Φd (q)). Hence S F is cyclic
of order Φd (q).

Example 25.8 For G = GLn with F = Fq the standard Frobenius,

T = Dn is a maximally split torus. Since |T F | = (q − 1)n and |GLn (q)| =
n 
q ( 2 ) i=1 (q i − 1) by Example 24.8(1), we see that T is a Sylow 1-torus.
n

More generally, maximally split tori always contain a Sylow 1-torus, and
conversely an F -stable maximal torus containing a Sylow 1-torus is maxi-
mally split. Tori always possess Sylow d-tori (see Exercise 30.16):

Lemma 25.9 Let (T, F ) be a torus with corresponding complete root datum
T = (X, ∅, Y, ∅, φ). Then S = (X  , ∅, Y  , ∅, φ), with Y  = kerY (Φd (φ)) and
X  = X/Ann(Y  ), is the complete root datum of the Sylow d-torus of T .

We now show why the name Sylow tori is justified. For this we need a
further fundamental result from the theory of finite reflection groups (see
[65, 3.4 and 6.2]):

Theorem 25.10 (Springer) Let W ≤ GL(V ) be a finite complex reflection

group, φ ∈ NGL(V ) (W ) and ζ ∈ C a root of unity. For w ∈ W let

V (wφ, ζ) := {v ∈ V | wφ v = ζv}

denote the ζ-eigenspace of wφ. Let a(ζ) := maxw∈W dim V (wφ, ζ). Then we
have:

(a) For all w1 , w2 ∈ W with dim V (w1 φ, ζ) = dim V (w2 φ, ζ) = a(ζ) there
exists w ∈ W with w.V (w1 φ, ζ) = V (w2 φ, ζ).
(b) For any w ∈ W there exists w ∈ W with V (wφ, ζ) ≤ V (w φ, ζ) and
such that dim V (w φ, ζ) = a(ζ) is maximal.

This implies the following result (see [10, Thm. 3.4], and also Exercise 30.17
for the case of GLn ):
224 Tori and Sylow subgroups

Theorem 25.11 (Generic Sylow Theorems) Let G be a connected reductive

group, F : G → G a not very twisted Steinberg endomorphism and d ≥ 1.
(a) There exist Sylow d-tori in G and they are all GF -conjugate.
(b) Any d-torus of G is contained in some Sylow d-torus.
Proof Let W denote the Weyl group of G with respect to an F -stable
maximal torus T with character group X and cocharacter group Y , such
that F |X = qφ. By the remark after Proposition 25.3 the complete root data
of F -stable maximal tori of G are of the form (X, ∅, Y, ∅, wφ) for w ∈ W .
By Example 9.12 the complete root datum of an arbitrary F -stable torus
S is then of the form S = (X/Ann(YS ), ∅, YS , ∅, wφ) for some wφ-stable
sublattice YS of Y , and by Proposition 25.3(c) its generic order is given
by the characteristic polynomial of wφ on YS ⊗Z R. Now first note that by
Molien’s formula in Exercise 30.15(d) there exist F -stable maximal tori in G
whose generic order is divisible by the same power of Φd as the generic order
of G, and so by Lemma 25.9 there exist d-tori of that order, which proves
the first half of (a).
Write V := Y ⊗Z C and VS := YS ⊗Z C ≤ V and let ζd be a prim-
itive dth root of unity. Then S is a d-torus if and only if the character-
istic polynomial of wφ on Y (S) is a power of Φd , hence if and only if

VS = i:gcd(i,d)=1 VS (wφ, ζdi ).
By Proposition 25.1, conjugacy of tori is translated into φ-conjugacy in the
Weyl group. Thus we may rephrase assertions (a) and (b) as assertions about
elements of W as follows: all maximal eigenspaces V (wφ, ζd ) are conjugate
under W , and the ζd -eigenspace of any element wφ is contained in a ζd -
eigenspace of maximal possible dimension. But these are just the statements
of Theorem 25.10.
Example 25.12 For G = GLn or SLn with the standard Frobenius map,
a maximal torus parametrized by w ∈ W = Sn (as in Proposition 25.1)
contains a Sylow d-torus if and only if the cycle decomposition of the per-
mutation w contains a disjoint d-cycles, where n = ad + b with b < d, since
according to Example 25.4(2) these are precisely the tori whose generic order
is divisible by the maximal possible power of Φd .
For the very twisted groups of types 2 B2 , 2 G2 or 2 F4 excluded in the
statement above, the right “generic primes” to consider are divisors of the
order polynomial over a suitable quadratic extension of Z. With this modi-
fication the assertions of Theorem 25.11 remain true in these cases as well,
see Broué–Malle [10, 3F] for details, and for an analogue of the third Sylow
Theorem.
 25.2 Sylow subgroups 225

25.2 Sylow subgroups

We now discuss the structure of Sylow subgroups for true primes, their cen-
tralizers and normalizers. We give statements and proofs only for not very
twisted Steinberg endomorphisms, but analogous statements hold after suit-
able modifications in the other cases as well.
Consider a prime r such that r divides |GF | and r = char(k) = p. (For the

defining prime r = p see Corollary 24.11.) As |GF | = q N d Φd (q)a(d) this
implies that r|Φd (q) for some d with a(d) > 0. Let us set

e := er (q) := the order of q modulo r,

so that r|q e −1, but r does not divide q d −1 for any positive integer d < e. We
require the following standard number theoretical lemma (see [56, Lemma
5.2], for example).

Lemma 25.13 If r|Φd (q), r prime, then d = er (q) rf for some f ≥ 0.

Theorem 25.14 Let G be connected reductive with a not very twisted

Steinberg endomorphism F : G → G, r = char(k) a prime dividing |GF |
with r  |W φ|. Then:

(a) There exists a unique d such that a(d) > 0 and r|Φd (q). Then any Sylow
d-torus of G contains a Sylow r-subgroup of GF .
(b) The Sylow r-subgroups of GF are homocyclic with a(d) cyclic factors of
order |Φd (q)|r .

In particular, in this situation all Sylow r-subgroups of GF are abelian.

Recall that a group is homocyclic if it is the direct product of isomorphic

cyclic groups. So in the situation of Theorem 25.14 the Sylow r-subgroups
are isomorphic to Zm × · · · × Zm (a(d) times), where m = |Φd (q)|r denotes
the r-part of Φd (q).

Proof By Corollary 24.6 the generic order |G| of (G, F ) is a product of

a(d)
cyclotomic factors Φd , where d divides some δdi , with δ the order of φ on
XR and di the degrees of W . By Lemma 25.13, if r|Φe1 (q), r|Φe2 (q), with
e1 < e2 , then r|e2 . Thus, if a(e1 ) > 0, a(e2 ) > 0, that is, if both Φe1 , Φe2
divide |G| then r|δdi for some i, whence it divides |W φ| = δd1 · · · dn (see
Theorem 24.4(c)), which was excluded.
a(d)
Now let S be a Sylow d-torus of G, of generic order Φd . Then the claim
in (b) follows from Proposition 25.7.

For general primes r the Sylow r-subgroups of GF may be non-abelian,

226 Tori and Sylow subgroups

so we cannot expect to embed them into tori. But it turns out that Sylow
r-subgroups do embed in the normalizer of a maximal torus.
Definition 25.15 An algebraic group S is supersolvable if there exists a
series 1 = S0
S1
· · ·
Sn = S of closed normal subgroups Si
S such
that Si /Si−1 is either cyclic or a torus for all i = 1, . . . , n.
For example, finite p-groups, having non-trivial center, are supersolvable,
hence so are finite nilpotent groups.
Theorem 25.16 Let G be a connected reductive group, F : G → G a
Steinberg endomorphism, S ≤ G an F -stable supersolvable subgroup such
that each Si
S is F -stable and consists of semisimple elements. Then there
exists an F -stable maximal torus T ≤ G with S ≤ NG (T ).
Proof We argue by induction on the length of the chain 1 = S0
· · ·
Sn =
S. If S = 1, then the existence of an F -stable maximal torus T is guaranteed
by Corollary 21.12. Secondly, if S = 1 but CG (S1 ) = G, then S1 ≤ Z(G),
so S1 lies in every maximal torus of G by Corollary 8.13(b). In this case we
may argue with a shorter chain in G/S1 by induction.
Otherwise, G1 := CG (S1 )◦ is a proper connected F -stable subgroup of G,
reductive by Corollary 8.13(a) (if S1 is a torus) respectively by Theorem 14.2
(if S1 is cyclic), and the maximal tori of G1 are maximal tori of G by the
same references. Now, by the previous case there exists an F -stable maximal
torus of G1 which is normalized by S. This is a maximal torus of G.
Corollary 25.17 Let r be a prime, r = p. Then any Sylow r-subgroup of
GF lies in NGF (T ) for some F -stable maximal torus T ≤ G.
Proof As the Sylow r-subgroups are nilpotent and consist of semisimple
elements, and finite nilpotent groups are supersolvable, this follows immedi-
ately from Theorem 25.16 by taking F -fixed points.
Example 25.18 Let G = GLn , F the standard Frobenius map with re-
spect to Fq . The normalizer in GF of the maximally split torus Dn of G
equals NGF (Dn ) = DF F
n .W = Dn .Sn (see Example 8.2(1)). It can be checked
from the order formula that for q ≡ 1 (mod 4) this contains a Sylow 2-
subgroup of GLn (q). On the other hand, if q ≡ 3 (mod 4) then the order
of NGF (Dn ) is not divisible by the full 2-part of the order of GLn (q), so in
particular it cannot contain a Sylow 2-subgroup. In this case, though, the
normalizer of a Sylow 2-torus contains a Sylow 2-subgroup.
We now show that centralizers of Sylow subgroups are strongly related to
centralizers of Sylow tori.
 25.2 Sylow subgroups 227

Theorem 25.19 Let G be simple with a not very twisted Steinberg endo-
morphism F : G → G, let r be a prime divisor of |GF |, r = p. Then every
semisimple element g ∈ GF which centralizes a Sylow r-subgroup of GF lies
in an F -stable torus containing a Sylow d-torus of G, where d = er (q). In
particular, g centralizes a Sylow d-torus.

Proof for r > 3 Let g ∈ G be semisimple, centralizing a Sylow r-subgroup

of G, and set C := CG (g). We proceed in several steps, where for the first
two G may be an arbitrary connected reductive group.
(1) Assume for a moment that the index |C : C ◦ | is prime to r. Then (C ◦ )F
contains a Sylow r-subgroup of GF . The generic orders of G and C ◦ are both
monic polynomials, with the second dividing the first, see Corollary 24.7.
Hence, all cyclotomic polynomials Φi with r|Φi (q) dividing the generic order
of G also divide the generic order of C ◦ , to the same power. In particular, a
Sylow d-torus of C ◦ is a Sylow d-torus of G. Thus, g centralizes a Sylow d-
torus of G. By Proposition 14.1 we even have g ∈ C ◦ , so g ∈ Z(C ◦ ), whence
g is contained in a maximal torus containing a Sylow d-torus.
(2) Now assume that the order of g is prime to r. By Proposition 14.20
the exponent of C/C ◦ divides the order of g, so is prime to r. Hence the
claim follows from the previous consideration.
(3) Next assume that g has r-power order. Let π : Gsc → G denote the nat-
ural isogeny from a group of simply-connected type (see Proposition 9.15).
Again by Proposition 14.20 the factor group C/C ◦ is isomorphic to a sub-
group of ker(π) ≤ Z(Gsc ), hence we are done by step (1) if the order of
the fundamental group is prime to r. This argument is valid for arbitrary
semisimple groups G.
Consulting the list of possible orders of fundamental groups for simple
groups in Table 9.2 we see, as r > 3, that the only cases not covered by the
above argument are G of type An−1 with r dividing n.
For G of type An−1 we use the explicit knowledge of the Sylow structure
of GF . First assume that GF is untwisted. Then, by Corollary 25.17 a Sylow
r-subgroup R of GF is contained in the normalizer of a Sylow d-torus T of
order Φd (q)a of GF , where 0 ≤ n − ad < d, with normalizer quotient Wd =
NGF (T )/CGF (T ) a wreath product Zd Sa (see Exercise 30.17). The action of
Wd on T F is given as follows: By Proposition 25.7, T = T1 ×· · ·×Ta is a direct
product of F -stable d-tori Ti ∼ = T1 of generic order Φd . The Zd -factors of
Wd now act as field automorphisms of Fqd /Fq on the cyclic subgroups TiF of
T F of order Φd (q), while the symmetric group Sa permutes the coordinates,
see Example 25.4. Now the r-elements in F× qd
are not contained in a proper
subfield, by the definition of er (q), hence not centralized by non-trivial field
228 Tori and Sylow subgroups

automorphisms. In particular, the normalizer quotient Wd acts faithfully

on the Sylow r-subgroup of T , so every r-central element of R is already
contained in the maximal torus T , as claimed. A similar reasoning applies if
GF is a unitary group.
(4) Finally, in the general case, let g = g1 g2 be the decomposition of g
into its r- and r -part. Let C1 := CG (g1 ), the centralizer in G of the r-part
of g1 . By step (3) its connected component C1◦ contains a Sylow d-torus of
G. Since CG (g) ≥ CC1◦ (g2 ) and C1◦ is reductive by Theorem 14.2, the result
now follows from (2) applied to the r -element g2 of C1◦ .
In the case r ≤ 3, some further configurations have to be considered in
step (3) of the above proof, see [56, Thm. 5.9]. A suitably modified statement
remains true for very twisted F , see [56, Thm. 8.2].
We end this section by citing the following extension of Corollary 25.17 to
normalizers of Sylow subgroups ([56, Thm. 5.14]; see also [29, Thm. 4.10.3]):
Theorem 25.20 Let G be a semisimple group, F : G → G a not very
twisted Steinberg endomorphism, r > 3 a prime different from p, R ∈
Sylr (GF ), d = er (q). Then there exists a Sylow d-torus S of G such that
NGF (R) ≤ NGF (S).
About the proof By Corollary 25.17, R ≤ NG (T ) for some F -stable maxi-
mal torus T of G. As r ≥ 5 one can show (using that T F is not a so-called
failure of factorization module for NG (T )/T , see Cabanes [12]) that R con-
tains a unique maximal elementary abelian toral r-subgroup A which, more-
over, lies in a unique Sylow d-torus S of G. So any g ∈ NGF (R) stabilizes A,
hence stabilizes S, so lies in NGF (S).
Example 25.21 This example is to show that the conclusion of Theo-
rem 25.20 can fail for r ≤ 3. Let G = SL2 , F the standard Frobenius with
GF = SL2 (q). Assume q ≡ 3, 5 (mod 8), and let r = 2. Then a Sylow 2-
subgroup R of GF is a quaternion group Q8 and NGF (R) ∼ = SL2 (3). But the
normalizers NGF (T ) ∼
= Zq−1 .2 or Zq+1 .2 of the two distinct F -stable maxi-
mal tori (up to GF -conjugation) clearly cannot contain such a subgroup.
Similar examples exist for all Sp2n (q), q ≡ 3, 5 (8) and r = 2, and for
r = 3 in SL3 (q), SU3 (q) and G2 (q). See [56, Thm. 5.14, 5.19] for the precise
statements, and [56, Thm. 8.4] for the case of Suzuki and Ree groups.
See [29, Thm. 4.10.6] for a completely different approach, due to Asch-
bacher, to the Sylow 2-subgroups of groups of Lie type over fields of odd
characteristic.
 26
Subgroups of maximal rank

We transfer results in Chapter 12 on parabolic subgroups and Levi comple-

ments in a reductive algebraic group G to the finite group GF of fixed points
under a Steinberg endomorphism and compare with the BN-pair structure
in GF . We also derive several results on conjugacy classes and centralizers
in GF from the corresponding statements for G in Chapters 13 and 14.

26.1 Parabolic subgroups and Levi subgroups

Let G be a connected reductive group, T ≤ B a maximal torus inside a Borel
subgroup and N := NG (T ). By Theorem 11.16, B, N form a BN-pair in G,
with Weyl group W generated by a distinguished set of involutions S. Now
assume in addition that we are given a Steinberg endomorphism F : G → G
and choose B, T both F -stable. Then we observed in Theorem 24.10 that
B F , N F form a BN-pair in GF , with Weyl group W F generated by a set SF
of involutions in natural bijection with the F -orbits on S (see Lemma 23.3).
We compare the BN-pairs in G and GF . Let I ⊆ S and PI the correspond-
ing parabolic subgroup of G with standard Levi complement LI . Then by
the double coset decomposition in Proposition 12.2(a) PI is F -stable if and
only if I is F -invariant, i.e., if I is a union of F -orbits in S, so corresponds to
a subset IF ⊆ SF of the set of F -orbits on S. We write PIF for the parabolic
subgroup of GF defined by IF .

Proposition 26.1 Let G be connected reductive with a Steinberg endomor-

phism F : G → G and let I ⊆ S be F -invariant.

(a) There is a unique GF -conjugacy class of F -stable G-conjugates of PI ,

and we have PIF = PIF .
230 Subgroups of maximal rank

(b) There is a unique PIF -conjugacy class of F -stable Levi complements to

UI = Ru (PI ) in PI , and we have PIF = UIF
LFI .

Proof The first claim follows exactly as in the special case I = ∅, that
is, PI = B, in Corollary 21.12. Indeed, PI is self-normalizing by Proposi-
tion 12.2(a), so there is a unique GF -conjugacy class of F -stable G-conju-
gates of PI by Theorem 21.11(b). Taking F -fixed points in the double coset
decomposition for PI in Proposition 12.2 we obtain that PIF = PIF .
For part (b) we use that the standard Levi complement LI of PI is gen-
erated by an F -stable subset of root groups and so is F -stable, and that all
Levi complements in PI are PI -conjugate (see Proposition 12.6). It is clear
by this same result that UIF ∩ LFI ≤ UI ∩ LI = {1}. Now let p ∈ PI . Then
F

p = ul for unique u ∈ UI and l ∈ LI . Since both UI , LI are F -invariant, this

implies that u and l are both F -stable, whence PIF = UIF
LF I .

That is, the standard parabolic subgroups of the BN-pair GF arise as the
F -fixed points of F -stable standard parabolic subgroups of G. Note however
that in the finite case it is no longer true in general that all complements
to UIF in PIF are conjugate to LFI . Over small finite fields, the central torus
of LI used in the proof of Proposition 12.6 may have a trivial fixed point
group, so that the argument given there fails in this case; see the following
example.

Example 26.2
Let G = SL4 , char(k)

 = 2, and F the 2-power map on
A ∗ 
G. Set P = A ∈ GL3 ∩ G. One complement of the normal
0 ∗ 
unipotent group U F in P F is the subgroup

  
A 0 
L1 :=  A ∈ GL3 .
F
0 det(A)−1 

A second non-conjugate complement is obtained as the image of a four-

dimensional indecomposable representation of the group SL3 (2) ∼ = PSL2 (7),
over the field F2 , in which SL3 (2) stabilizes and acts irreducibly on a 3-space.
(This representation is obtained by decomposing the natural permutation
representation of PSL2 (7) acting on the cosets of a Borel subgroup.)

Let’s turn to the parametrization of F -stable Levi subgroups L of G.

They are not necessarily contained in F -stable parabolic subgroups of G.
Indeed, already in the case I = ∅, where L is a maximal torus, we saw that
not all F -stable G-conjugates of L lie in an F -stable Borel subgroup (see
Proposition 25.1).
 26.1 Parabolic subgroups and Levi subgroups 231

Example 26.3 Let G = GLn with the Steinberg endomorphism F : G →

G from Example 21.14(2) such that GLF n = GUn (q). Recall from Exam-
ple 12.4 the set S = {s1 , . . . , sn−1 } of simple reflections generating the Weyl
group of G.
⎧⎛ ⎞ ⎫
⎪  ⎪
⎨ ∗ ∗ ∗  ⎬
⎜ ⎟ 
(1) The parabolic subgroup PI := ⎝ 0 A ∗ ⎠  A ∈ GLn−2 of G
⎪
⎩ 0 0 ∗  ⎪
⎭
corresponding to I = {s2 , . . . , sn−2 } is F -stable, with fixed point group
⎧⎛ ⎞ ⎫
⎪  ⎪
⎨ ∗ ∗ ∗  ⎬
⎜ ⎟ 
PI = ⎝ 0 A ∗ ⎠  A ∈ GUn−2 (q) ∩ GUn (q)
F
⎪
⎩ 0 0 ∗  ⎪
⎭

∼ ×
and F -stable Levi complement LI with F -fixed points LF I = Fq 2 ×
GUn−2 (q).
(2) On the other hand, for n ≥ 3 the maximal parabolic subgroup

  
∗ ∗ 

PJ :=
0 A  A ∈ GLn−1

of G corresponding to J = {s2 , . . . , sn−1 } has no F -stable G-conjugate,

since J is not F -invariant. Nevertheless, its Levi complement

  
∗ 0 
 ∼
LJ :=
0 A  A ∈ GLn−1 = Gm × GLn−1

possesses F -stable G-conjugates. This is most easily seen by replacing F

by the Steinberg endomorphism F  as in Example 21.2 (which replaces
GF by a G-conjugate, see Corollary 21.8). Then LJ is F  -stable.
As WJ ∼ = Sn−1 is self-normalizing in W ∼ = Sn , so is LJ in G by Corol-
lary 12.11. So there is just one GF -conjugacy class of F -stable conjugates
of LJ ; the fixed point group is isomorphic to F× ×
q 2 /Fq ×GUn−1 (q). It is not
contained in any proper parabolic subgroup of GF , since, for example,
its order is larger than that of any proper Levi complement of GF .
See Exercise 30.20 for geometric interpretations of the two subgroups

PIF and LF J appearing above.

The situation here is described by the obvious generalization of Proposi-

tions 25.1 and 25.3:
Proposition 26.4 Let G be connected reductive with Steinberg endomor-
phism F : G → G, L ≤ G an F -stable Levi complement of some parabolic
subgroup of G, with Weyl group WL .
232 Subgroups of maximal rank

(a) The GF -classes of F -stable G-conjugates of L are in bijection with φ-

classes of NW (WL )/WL , with φ as in Proposition 22.2(b).
(b) Write Lw for a Levi subgroup corresponding to the φ-class of w under
the map in (a). Then NGF (Lw )/LF ∼
w = NW (WL wφ)/WL .

This is Exercise 30.21.

Let’s close our discussion of parabolic subgroups with the following exten-
sion of the theorem of Borel and Tits to finite groups of Lie type:
Theorem 26.5 (Borel–Tits) Let G be connected reductive with Steinberg
endomorphism F : G → G. Let U ≤ GF be a unipotent subgroup of GF .
Then there exists an F -stable parabolic subgroup P of G with U ≤ Ru (P )F
and NGF (U ) ≤ P F .
Proof We first claim that U lies in some F -stable Borel subgroup of G.
Indeed, U is unipotent, hence a p-group, where p is the underlying charac-
teristic of G. Thus, by Sylow’s Theorem, it lies in a Sylow p-subgroup of
GF . Corollary 24.11 shows that for any F -stable Borel subgroup B of G, B F
contains a Sylow p-subgroup of GF . Thus, U lies in a GF -conjugate of B F ,
whence of B.
Thus, we are in the situation of Theorem 17.10. As U is F -stable, all
the groups Ui , Ni , P (U ) constructed in the proof there are F -stable as well,
hence NG (U ) is contained in the F -stable parabolic subgroup P := P (U ),
with U ≤ Ru (P ).

26.2 Semisimple conjugacy classes

The Lang–Steinberg Theorem easily allows one to deduce properties of con-
jugacy classes and centralizers in the finite groups GF .
Let’s first observe the following analogue of Corollary 6.11(a):
Proposition 26.6 Let G be connected reductive with Steinberg endomor-
phism F : G → G, and s ∈ GF semisimple. Then s lies in an F -stable
maximal torus of G.
Proof By Theorem 14.2, the centralizer CG (s)◦ is connected reductive, F -
stable since s is. Thus, by Corollary 21.12 it contains an F -stable maximal
torus T ≤ CG (s)◦ . Now Proposition 14.1 shows that s ∈ T .
Note that any conjugacy class of an algebraic group G with Steinberg
endomorphism F : G → G is F m -stable for some sufficiently large m. Indeed,
embed G → GLn such that F i is the restriction to G of a standard Frobenius
 26.2 Semisimple conjugacy classes 233

map Fq on GLn . Then any s ∈ G is represented by a matrix with entries in

some finite subfield of k, hence fixed under some power Fqj of the standard
m
Frobenius map. Thus s ∈ GF with m := ij; in particular, the conjugacy
class of s is F m -stable.

Theorem 26.7 Let G be a connected reductive group, F : G → G a Stein-

berg endomorphism. Then we have:

(a) Every F -stable conjugacy class of semisimple elements of G contains an

F -stable element.
(b) Let [s]G be an F -stable semisimple conjugacy class of G. Then the GF -
◦
classes in [s]F
G correspond bijectively to the F -classes in CG (s)/CG (s) .

(c) Assume that the derived group G of G is simply connected. Then there
is a natural bijective correspondence
0 1 
F -stable semisimple 1−1 GF -classes of semisimple
←→ .
classes in G elements in GF
Proof Parts (a) and (b) follow directly from Theorem 21.11 applied to the
conjugation action of G on itself, since the centralizer CG (s) of any semisim-
ple element s ∈ G is closed by Proposition 5.2. In the situation of (c), all
centralizers CG (s) of semisimple elements are connected by Theorem 14.16,
so the claim follows from parts (a) and (b).

Example 26.8 How many semisimple conjugacy classes are there in the
general linear group GLn (q)? A conjugacy class is determined by its Jordan
canonical form, hence semisimple classes are parametrized by the character-
istic polynomial of their elements. Any monic polynomial over Fq of degree n
and with non-vanishing constant coefficient occurs as characteristic polyno-
mial, and there are precisely (q − 1)q n−1 of these, so this is the number of
semisimple conjugacy classes of GLn (q).

Steinberg found a natural generalization of this observation; its proof re-

quires the following result on the natural representation of a finite reflection
group, which is also due to Steinberg [72, Thm. 14.4] (see also [9, V, §5,
Ex. 2.3(b)]):

Lemma 26.9 Let W ≤ GL(V ) be a finite reflection group on a finite-

dimensional real vector space V , with trivial fixed space V W = 0 and σ ∈
NGL(V ) (W ). Then
1 "
det V (x idV − σw) = xdim V .
|W |
w∈W
234 Subgroups of maximal rank

Sketch of proof The sum in question is the scalar product on the coset W.σ
of the class function detV (x idV − σw) with the trivial character (see [22, I.6]
for the elementary properties of such scalar products). Now note that the
characteristic polynomial of any z ∈ GL(V ) is given by
"
n
det V (x idV − z) = (−1)i χi (z) xn−i ,
i=0

where n = dim V and χi is the character of GL(V ) on the ith exterior power
∧i (V ). Using the assumptions on W one shows that W has trivial fixed space
on all ∧i (V ), i > 0, so only the term for i = 0 in the sum contributes to that
scalar product, yielding the value xn .
Theorem 26.10 (Steinberg) Let G be a connected reductive algebraic
group, G := [G, G], l := rk G , Z := Z(G)◦ , and F : G → G a Steinberg
endomorphism. Then:
(a) The number of F -stable semisimple conjugacy classes of G equals q l |Z F |.
(b) If the derived group G is simply connected, then this is also the number
of GF -classes of semisimple elements in GF .
Proof By Proposition 14.6 the F -stable semisimple conjugacy classes of G
are in bijection with the set (T /W )F of F -stable W -orbits on an F -stable
maximal torus T of G. Thus, we are counting t ∈ T satisfying F (t) = tw for
some w ∈ W , up to W -conjugation. Hence
1 " " 1 " " 1 " w−1 F
|(T /W )F | = 1= 1= |T |.
|W | |W | |W |
t∈T w∈W w∈W
t∈T w∈W
F (t)=tw F (t)=tw
−1 −1
Now |T w F | = |Z F | · |S w F | with Z = Z(G)◦ and S = T ∩ G by Ex-
−1
ercise 30.22, and by Proposition 25.2 we have |S w F | = detV (q − φ−1 w),
where V = X(S) ⊗Z R, giving
|Z F | "
|(T /W )F | = det V (q − φ−1 w).
|W |
w∈W

Now (a) follows from Lemma 26.9, (b) follows from this by Theorem 26.7(c).

Example 26.11 (Centralizers of some isolated elements in E6 (q)) We con-

tinue Example 14.21, with G of simply connected type E6 , char(k) = 3
and H the subsystem subgroup of type A32 corresponding to the subset
{0, 1, 2, 3, 5, 6} of the extended Dynkin diagram (see Table 13.1). By The-
orem 22.5 there exists a Steinberg endomorphism F : G → G which acts
 26.2 Semisimple conjugacy classes 235

trivially on W , so F is Fq -split for some power q of char(k) and GF = E6 (q).

Then the set J is F -stable, hence so is H.
Let s ∈ Z(H)\Z(G), with centralizer C := CG (s) = H. As C is connected,
according to Theorem 26.7 there is a unique GF -class in [s]F G , and we now
assume that s ∈ G . In order to determine the centralizer of s in GF , we
F

consider the possible F -fixed points of C = H. By Theorem 21.11 there exist

three GF -classes of F -stable conjugates of H, parametrized by the conjugacy
classes in W (H) = NG (H)/H ∼ = S3 . As W (H) acts on H by permuting the
three A2 -factors, the structure of HwF , for w ∈ W (H) can be determined
using Exercise 30.2; see Table 26.1 where we’ve also given the order of the
center of HwF , which can be determined from the action of F on Z(Hw ).

Table 26.1 Some centralizers in E6 (q)

w F
Hw NG (Hw )F /Hw
F
|Z(Hw
F
)| |Z(H̄w
F
)|
() A2 (q)3 S3 gcd(3, q − 1)2 gcd(3, q − 1)
(1 2) A2 (q 2 ) · 2A2 (q) Z2 gcd(3, q 2 − 1) gcd(3, q + 1)
(1 2 3) A2 (q 3 ) Z3 gcd(3, q − 1) gcd(3, q − 1)

Note that Z(GF ) ≤ Z(HwF ) for all w. As |Z(GF )| = gcd(3, q − 1) this

shows that in the simply connected case, E6 (q) contains a class of semisimple
elements of order 3 with centralizer of type A2 (q)3 when q ≡ 1 (mod 3),
respectively of type A2 (q 2 ) · 2A2 (q) when q ≡ 2 (mod 3).
Now consider Ḡ = G/Z(G), of adjoint type, with the induced split Frobe-
nius map. Here, for 1 = s ∈ Z(H̄) the centralizer C := CḠ (s) is disconnected,
with C/C ◦ ∼= Z3 , by Example 14.21. Now, for q ≡ 1 (mod 3), F acts trivially
on C/C ◦ , so by Theorem 26.7(b) there exist three ḠF -classes of semisimple
elements in [s]FḠ
. For q ≡ 2 (mod 3), F acts non-trivially on C/C ◦ , so there
is just one F -class in C/C ◦ , hence also just one ḠF -class in [s]F
Ḡ
.
Looking again at the various F -stable conjugates of H̄ and their centers, as
given in the last column of Table 26.1, we see that for q ≡ 1 (mod 3), there
is one class with centralizer A2 (q)3 .Z3 and two with centralizer A2 (q 3 ).Z3 ,
while for q ≡ 2 (mod 3), there is one class with centralizer A2 (q 2 ) · 2A2 (q).
Taking for F the non-split Steinberg endomorphism on G instead, we ob-
tain the Ennola dual parametrization of classes, i.e., “with q suitably replaced
by −q” throughout.
 27
Maximal subgroups of finite classical groups

The aim of this chapter is to study the subgroup structure of the finite clas-
sical groups. The classical groups are defined as certain subgroups of the
isometry groups of the zero form or of non-degenerate bilinear, sesquilinear
or quadratic forms f on finite-dimensional vector spaces over finite fields.
According to the classification of such forms and their isometry groups
(which can be found for example in [2, §21]), there exist the following non-
degenerate possibilities: f is symplectic on an even-dimensional vector space,
f is quadratic of maximal Witt index on an odd-dimensional vector space,
f is quadratic on a 2n-dimensional vector space, of Witt index n or n − 1, or
f is unitary on a vector space over a field with a subfield of index 2. Here,
the Witt index of a quadratic form is the maximal dimension of a totally
singular subspace.
The various possibilities for the isometry groups are collected in Table 27.1.

Table 27.1 Finite classical groups

V form isometry group Cl(V )

Fnq zero GLn (q) SLn (q)
F2n
q symplectic Sp2n (q) Sp2n (q)
Fq2n+1 quadratic GO2n+1 (q) SO2n+1 (q)
F2n
q quadratic of Witt index n GO+2n (q) SO+
2n (q)
F2n
q quadratic of Witt index n − 1 GO−2n (q) SO−
2n (q)
Fnq2 unitary over Fq2 GUn (q) SUn (q)

Definition 27.1 We shall refer to groups in the last column of the above
table as classical groups, and we shall denote them by Cl(V ).
 27.1 The theorem of Liebeck and Seitz 237

27.1 The theorem of Liebeck and Seitz

The key to the understanding of maximal subgroups of the groups Cl(V ) will
be to relate them to algebraic groups. In all but the last case of Table 27.1,
extending scalars to k = Fq we obtain the corresponding isometry group
G over the algebraically closed field k, acting on V̄ = k d such that the
original group consists of the fixed points GF on V = V̄ F under the standard
Frobenius on V̄ with respect to Fq , with compatible Fq -structures on G and
on V̄ , as indicated:

GF SLn (q) Sp2n (q) SO2n+1 (q) SO+

2n (q) SO−
2n (q)
G SLn Sp2n SO2n+1 SO2n SO2n

This is clear from our definition of the classical algebraic groups in Sec-
tion 1.2, respectively from the construction of SO− 2n (q) in Example 22.9.
(In the case of orthogonal groups in characteristic 2 this can serve as the
definition of the finite special orthogonal groups.)
A similar construction is not possible for the classical groups SUn (q),
despite the fact that they also arise as fixed points of SLn under a Steinberg
endomorphism, as the concept of a sesquilinear form is tied to the field in
question. So it is not possible to obtain compatible Fq -structures on SLn and
the underlying natural n-dimensional k-vector space k n . The SUn (q)-case is
therefore more technical and we will not go into the proofs for this case here.
We want to apply techniques from algebraic groups to obtain results on
classical groups. For this, we need the following standard result (see for
example [66, §11.1]).

Proposition 27.2 Let V = k n and F : V → V, (v1 , . . . , vn ) → (v1q , . . . , vnq ),

the standard Frobenius map. Assume that V1 ≤ V is an F -stable subspace.
Then V1F contains a k-basis of V1 ; in particular dimFq V1F = dimk V1 .

Proposition 27.3 Let V = k n , H ≤ GL(V ) be stable under a standard

Frobenius map F of GL(V ). Then:

(a) F permutes the H-invariant subspaces of V .

(b) If H is (absolutely) irreducible on V and preserves a non-degenerate
bilinear or quadratic form on V , then the corresponding classical group
is F -stable.

Proof Assertion (a) is clear since V1 ≤ V being H-invariant implies that

F (V1 ) ≤ V is H-invariant.
In (b) let f denote an H-invariant non-degenerate bilinear form on V and
238 Maximal subgroups of finite classical groups

define
f˜ : V × V → k, f˜(v, w) := F −1 (f (F (v), F (w))).

Then f˜ is bilinear, non-degenerate and H-invariant. But H, acting irre-

ducibly, can fix at most one non-degenerate form on V , up to scalars, so
f˜ = af for some a ∈ k × . Thus f and f˜ have the same isometry group, which
is therefore F -stable. One argues similarly for quadratic forms.

We now state a refined version of the Reduction Theorem 18.6 for maximal
subgroups of classical algebraic groups from Part II which accommodates for
the presence of a Steinberg endomorphism, see [50, Thm. 1 ]. This will allow
us later to descend to the finite groups GF . The proof is very similar in that
one basically has to go through the proof of Theorem 18.6 once again, but
now paying attention to F -stability.
Recall the subgroup classes C1 , . . . , C5 from Section 18.1.

Theorem 27.4 (Liebeck–Seitz) Let G ≤ SL(V ) be a simple classical group

on V , invariant under a standard Frobenius map F . If H ≤ G is an F -stable
closed subgroup, then one of the following occurs.

(1) H is contained in some F -stable member of C1 ∪ . . . ∪ C5 , or,

(2) H/Z(H) is almost simple and H ∞ acts irreducibly on V . Moreover, if
H is infinite, then H ◦ acts tensor indecomposably on V .

Sketch of proof Let H ≤ G be as in the statement, and first assume that

there is a proper H- and F -invariant subspace 0 = V1 < V of V , which we
choose of minimal possible dimension. If G = Sp(V ) or SO(V ), then V1⊥
is also H- and F -invariant and by minimality of V1 , V1 ∩ V1⊥ ∈ {0, V1 }.
In the first case, V1 is non-degenerate, in the second it is totally isotropic.
The argument given in the proof of Proposition 18.4 shows that in the case
(G, char(k)) = (SO(V ), 2) either V1 is totally singular or else one-dimensional
non-singular. So, in any case, H is contained in an F -stable member of C1 .
Thus, we may assume that there are no non-trivial proper H- and F -
invariant subspaces in V . By Proposition 27.3(a), the socle of V |H is F -
invariant, so equal to all of V . Hence H acts completely reducibly on V .
Let V1 , . . . , Vt denote the homogeneous components of V |H . Then F must
permute the Vi , transitively since otherwise we obtain a proper F -stable
H-submodule.
Assume that t ≥ 2. Now, if G = SL(V ) then H ≤ StabG (V1 ⊕ · · · ⊕ Vt ),
which lies in class C2 and is F -stable. For symplectic and orthogonal type,
one can argue further as in the proof of Theorem 18.6 that in fact V = V1 ⊥
 27.1 The theorem of Liebeck and Seitz 239

. . . ⊥ Vt is an orthogonal decomposition, so again H will be contained in an

F -stable member of C2 .
Thus we may assume that V |H is homogeneous, and V = V1 ⊕. . .⊕Vt with
pairwise isomorphic irreducible H-modules Vi , of dimension dim(Vi ) > 1
since otherwise H ≤ Z(G), lying in an F -stable member of C1 . Let now
first G = SL(V ) and set C := CG (H). Then C is also F -invariant, and by
Proposition 18.1 there exists a decomposition V ∼ = V1 ⊗ HomH (V1 , V ) such
that C = (1 ⊗ GL(U )) ∩ G with dim(U ) = t and CG (C) = (GL(V1 ) ⊗ 1) ∩ G,
which is again F -stable. So, if t > 1, then V = U ⊗ V1 is an F -invariant
tensor decomposition of V stabilized by H, and H lies in the F -invariant
member NG (C ◦ CG (C)) = NG (SL(V1 ) ⊗ SL(U )) of C4 .
For Sp(V ) and SO(V ), the argument given in the proof of Theorem 18.6
can be adapted to show that H must be contained in an F -invariant member
of C3 ∪ C4 .
So we may assume that V |H is irreducible. Let’s first discuss the case
when H is infinite. We saw in the proof of Theorem 18.6 that then the
irreducibility assumption implies that H ◦ is semisimple, say H ◦ = H1 · · · Ht .
-t
By Corollary 18.2 we then have a tensor product decomposition V = i=1 Vi
- -
with H ◦ ≤
t
i=1 GL(Vi ) and CGL(V ) (Hj ) = i=j GL(Vi ). Since H is F -
invariant, so is H ◦ , and F permutes the Hi . Thus, if G = SL(V ) then the
double centralizers

CC(Hi ) := CG (CG (Hi )) = GL(Vi ) ∩ SL(V ) (1 ≤ i ≤ t)

-t
are permuted by F and hence F normalizes i=1 SL(Vi ), an F -invariant
member of C4 when t > 1. On the other hand, for G = Sp(V ) or G = SO(V ),
V is self-dual, hence so are the Vi by Corollary 18.2(b). So Hi ≤ Cl(Vi ) and
 
i CC(Hi ) ≤ i Cl(Vi ), an F -invariant group in C4 when t > 1. Otherwise,
t = 1 and H ◦ is simple and irreducible on V as in conclusion (2).
It remains to consider the case when H is finite. Let H1 ◦ . . . ◦ Ht be the
product of the (quasi-simple) components of H, a characteristic subgroup. By
the same argument as in the previous case, either t ≤ 1 or H lies in a member
of C4 . If t = 1 then CG (H1 ) = Z(G) by irreducibility, so H1 Z(G)/Z(G) is
simple and thus HZ(G)/Z(G) is almost simple as in (2).
Finally, if t = 0, all minimal normal subgroups of H̄ := H/Z(H) are
abelian and H contains a normal subgroup R as in class C5 . Then H lies in a
member of C5 , F -stable since maximal elements of C5 are the full normalizers
of the F -stable groups R.
240 Maximal subgroups of finite classical groups

27.2 The theorem of Aschbacher

We now study case (1) of the preceding theorem in some more detail in order
to define certain natural collections of subgroups of the finite classical groups
GF , where G = Cl(V ) with dim(V ) = n. If an F -stable subgroup H of G
is contained in some F -stable group M ∈ Ci , then clearly H F ≤ M F . What
possibilities can arise for M F ?
If M ∈ C1 then M stabilizes an F -stable non-trivial subspace 0 = V1 < V .
By Proposition 27.2 then 0 = V1F < V F . This leads to the following class of
subgroups:
Class C1F : stabilizers in GF of subspaces 0 = V1F < V F , non-degenerate
or totally singular, or non-singular of dimension 1 when (G, char(k)) =
(SO(V ), 2).
Now assume that M ∈ C2 is F -stable. Then M stabilizes an orthogonal
t
decomposition V = i=1 Vi into t 
> 1 mutually isometric subspaces Vi .
t
As F stabilizes M = G{V1 ,...,Vt } = i=1 Cl(Vi ).St , it permutes the Vi . So
M F , F  induces a permutation group ∆t on {1, . . . , t}. We may distinguish
several cases:

1. ∆t acts intransitively. Then M F stabilizes a proper F -stable subspace of

V corresponding to an orbit of ∆t , so M F lies in a group in C1F .
2. ∆t is transitive, but imprimitive. Then we may group together the Vi
according to the blocks of imprimitivity of this action to get another
member of C2 , with smaller value of t, but still ≥ 2.
3. ∆t acts primitively. Clearly, ∆t commutes with F , so the orbits of (any
power of) F form a block system. Hence, either F acts trivially, i.e., all
Vi are F -stable, or t is a prime and F acts as a t-cycle, in which case ∆t
is the cyclic group of order t generated by the image of F .

In the third case, if F acts trivially, M F lies in

Class C2F : stabilizers in GF of orthogonal decompositions V F = V1F ⊥
. . . ⊥ VtF into t ≥ 2 mutually isometric F -stable subspaces Vi , isomorphic
to Clm (q)  St where m = dim(Vi ), so n = mt,
while otherwise we are in the situation of Exercise 30.2, so M F lies in
Class C2 : field extension subgroups Clm (q t ).t with n = mt where t is prime.
F

In the case that all Vi are one-dimensional, this leads to the normalizer of
the Singer cycle as in Example 25.4(1).
 27.2 The theorem of Aschbacher 241

Now assume that M ∈ C3 is F -stable, so G = Sp(V ) or SO(V ), M fixes

a totally singular decomposition V = W ⊕ W  , and M ∼= GLm or GLm .2.
Then M F ∈ {GLm (q).a, GUm (q).a} with a ≤ 2. This gives
Class C3F : GLm (q).a or GUm (q).a with a ≤ 2, in symplectic or orthogonal
groups GF where n = 2m.
Now assume that M ∈ C4 is F -stable, the stabilizer of a tensor product
decomposition, and either V = V1 ⊗ V2 with V1 , V2 not isometric, or V =
-t
i=1 Vi with t ≥ 2 mutually isometric Vi . In the first case, M
F
falls into
Class C4F : stabilizers in GF of tensor product decompositions V F = V1F ⊗V2F
with V1 , V2 not isometric, of the form NGF (Cln1 (q)◦Cln2 (q)) where n = n1 n2
(see Proposition 18.1).
t
In the second case, M = NG ( i=1 Cl(Vi )) with F permuting the fac-
tors. Again, we look at the permutation group ∆t induced by the action of
M F , F  on the set of factors and argue as for class C2 above: if ∆t is intran-
sitive, we obtain a member of C4F . If ∆t is imprimitive, grouping together
the Vi according to the blocks, we obtain another member of C4 with smaller
t ≥ 2. Finally, if ∆t is primitive, then either F is trivial or t is prime and ∆t
is cyclic of order t. This leads to
-t
Class C4 : stabilizers of tensor product decompositions V F =
F F
 i=1 Vi
t
into t ≥ 2 isometric subspaces Vi , with M = NGF ( i=1 Clm (q)), where
F

n = mt , or

Class C4 : twisted tensor product subgroups Clm (q t ).t, where n = mt with
F

t prime.
Here, a tensor product structure is preserved over Fq , but not over Fq , like
for example (modulo scalars) PSLm (q t ) < PSLmt (q), PSLm (q 2 ) < PSUm2 (q)
or PSpm (q t ) < PSpmt (q).
Next, if M ∈ C5 is F -stable, then it has a characteristic (and hence also
F -stable) normal r-subgroup R which is extraspecial or a central product of
such with Z4 . Here the F -fixed points lie in class
Class C5F : NGF (R) with NG (R) in C5 , as in Table 27.2 (see [50, p. 430] or
[44, Tab. 4.6.A]).
This completes our discussion of F -stable members of C1 , . . . , C5 . If H ≤
G is F -stable but not contained in any F -stable member of C1 ∪ . . . ∪ C5 ,
then by Theorem 27.4 it is an almost simple subgroup (modulo Z(G)), with
H ◦ acting irreducibly on V . Let’s single out two further subclasses in this
situation:
242 Maximal subgroups of finite classical groups

Table 27.2 Normalizers of extraspecial subgroups R

G NG (R)/Z(G)
SLrm r2m .Sp2m (r)
Sp2m 22m .GO−2m (2)
SO2m 22m .GO+2m (2)

2
Class C6F : subgroups of GF which are classical on V F (respectively on V F ,
for GF a unitary group).
Examples of this type are the normalizers of Spn (q) or SO(±) n (q) inside
SLn (q) and SUn (q), the normalizer of SUn (q 1/2 ) inside SLn (q), and the nor-
malizer of SOn (q) inside Spn (q) if p = 2.
Finally, for subfields Fq < Fq of prime index, we get:
Class C7F : subfield subgroups NGF (Cln (q  )), with |Fq : Fq | prime, where
GF = Cln (q).
We can now state the principal reduction result on maximal subgroups of
finite classical groups, first proved by Aschbacher [1]; here we present the
approach by Liebeck and Seitz [50] using descent from the corresponding
algebraic groups.
Theorem 27.5 (Aschbacher, Liebeck–Seitz) Let G ≤ SL(V ) be a classical
group on the finite-dimensional k-vector space V , stable under a Steinberg
endomorphism F : SL(V ) → SL(V ), so that GF is a finite classical group on
2
V F (respectively V F ), and let H ≤ GF . Then one of the following holds:
(1) H lies in a member of C1F , . . . , C7F , or
(2) H lies in class S, that is, HZ(G)/Z(G) is almost simple, H ∞ acts
absolutely irreducibly and tensor indecomposably on V F (respectively
2
V F ), the representation cannot be realized over a proper subfield, and
if GF = SLn (q) then H fixes no non-degenerate bilinear or unitary form
on V .

Proof for GF not unitary Apply Theorem 27.4 and our previous discus-
sion. Since H is F -stable, either H ≤ M F for an F -stable member M ∈
C1F ∪ . . . ∪ C5F , or HZ(G)/Z(G) is almost simple and H ∞ acts absolutely
irreducibly (since otherwise it fixes a direct sum decomposition over Fq and
H lies in a member of C2 ). If H lies in no member of C7F , then it cannot be
F

defined over a proper subfield, and if it lies in no member of C6 , it fixes no

non-degenerate form on V F other than the one defining G.
 27.2 The theorem of Aschbacher 243

Example 27.6 Assume that GF = SLn (q) for n a prime. Then only the
classes C1F , C5F , C6F , C7F and S can occur, since for the other classes n has to
be composite (see Proposition 28.1 for arbitrary n).
In fact Liebeck and Seitz [50] prove a more general version of the re-
duction theorem for normalizers of classical groups GF inside ΓL(V F ), the
extension of GL(V F ) by the group of field automorphisms, where additional
classes C1 and C7 can arise. We shall not attempt to describe this here.
F F

This general version then allows one to obtain a similar reduction theorem
for maximal subgroups of all isogeny types of classical groups. Indeed, for
our given classical group G = Cl(V ) let Gsc be the corresponding group of
simply connected type. By Proposition 9.15 there exists a natural isogeny
π : Gsc → G which restricts to a homomorphism GF sc → G
F
with central
F F
kernel and whose image contains [G , G ] by Proposition 24.21. Now note
that π induces a bijection between maximal subgroups of GF sc and those of
F F
its image π(Gsc ) in G . In this way, the results on maximal subgroups of
[GF , GF ] can be transferred to those of GF sc .
On the other hand, if Gad denotes the corresponding group of adjoint type,
then the natural isogeny G → Gad induces a homomorphism GF → GF ad
whose image contains [GF F F F F
ad , Gad ]. Now, Gad is an extension of [Gad , Gad ] by
certain (diagonal) automorphisms, so the above-mentioned general version
of the reduction theorem applies to the maximal subgroups of GF ad as well.

Remark 27.7 (Relationship between the classes CiF and Aschbacher’s and
Kleidman–Liebeck’s classes Ci ) The classes C1F , . . . , C7F of natural subgroups
defined above do not agree completely with the classes Ci introduced by Asch-
bacher in his original paper [1]. Moreover, the latter were already modified
in the book of Kleidman and Liebeck [44] in which the structure, conjugacy
and maximality of members in these classes were determined. The following
table indicates the relationship between the classes CiF here and in [50], and
the classes Ci in [44]:

[50] C1F , C1 C2F , C2 , C3F C4 C7F , C7 C4 , S
F F F F F
C4F C5F C6F
[44] C1 C2 , C3 C4 C7 C6 C8 C5 S
 28
About the classes C1F , . . . , C7F and S

The theorem of Aschbacher in the previous chapter still leaves several im-
portant questions open:

1. What is the structure of members in the classes C1F , . . . , C7F ?

2. Which of the groups in classes C1F , . . . , C7F are actually maximal in Cl(V )?
3. What kind of groups can occur in class S?
4. Which of the groups in class S are actually maximal?

In this section we comment on the status of these questions at the time of

writing and on some of the methods used to treat them.

28.1 Structure and maximality of groups in CiF

The answer to the first question on the structure of members in CiF is rel-
atively straightforward, given their explicit geometric description. This has
been laid out in the book of Kleidman and Liebeck [44].
The second question on maximality of members in the classes CiF is al-
ready much more difficult. Again, the complete answer has been obtained by
Kleidman and Liebeck [44] at least in the case where the dimension of the
classical module V of Cl(V ) is at least 13.
It turns out that essentially all members in the classes CiF are maximal,
apart from a few well-understood situations. As an illustration we present
the result for GF = SLn (q), taken from [44, Table 3.5.A] (compare with the
corresponding Table 18.2 for the algebraic group SLn ); here, as in Table 18.2,
Pm denotes the standard parabolic subgroup of SLn corresponding to the
set S \ {sm } of simple reflections of its Weyl group:

Proposition 28.1 The maximal members in the classes C1F , . . . , C7F for
 28.1 Structure and maximality of groups in CiF 245

GF = SLn (q), n ≥ 3, are as given in Table 28.1. Here, for the classes
C2F , C2 , C4F , C4 , the maximal subgroup is obtained as the normalizer in
F F

SLn (q) of the intersection of the corresponding entry in the table with SLn (q).

Table 28.1 The classes CiF in SLn (q)

class structure conditions non-max.

C1F F
Pm 1≤m≤n−1 −
C2F GLm (q)  St n = mt, t ≥ 2 m = 1, q ≤ 4
or m = q = 2
C2
F
GLm (q t ).t n = mt, t prime −
C4F GLn1 (q) ⊗ GLn2 (q) n = n1 n 2 , 2 ≤ n 1 < n 2 n1 = q = 2
C4
F
GLm (q)  St n = mt , m ≥ 3, t ≥ 2 −
1
C5F r2m+1 .Sp2m (r) see below −
C6F Spn (q) n even −
SOn (q) nq odd −
SO±n (q) n even, q odd −
SUn (q0 ) q = q02 −
C7F GLn (q0 ) q = q0t , t prime −

1 In C F , r
= p is prime, q = pf with f odd and minimal subject to pf ≡ 1

5
(mod 2 gcd(2, r)).

About the proof It is clear from the description in Table 18.2 of the maximal
members in the classes C1 , C2 , C4 , C6 that there exist F -stable representatives,
with F -fixed points in C1F , C2F , C4F , C4 , C6F as given in Table 28.1. Moreover,
F

multiplying F by an inner element of G of order t which interchanges the

factors GLm in a group of type C2 cyclically leads to the fixed point group
of type C2 (using Exercise 30.2 and Corollary 21.8). The existence of ex-
F

traspecial subgroups with the stated normalizer is well-known. Finally, the

subgroups in class C7F clearly occur.
It then remains to work out for which values of the parameters any of the
above groups is contained in a member of some other class. For example,
since SLn (q) has a BN-pair, the parabolic subgroups in class C1F are always
maximal by Proposition 12.2.

In all classes and for all classical groups there are only three families of
examples in which maximal members of a class CiF lie in some group in
the class S. They all originate from the fact that in characteristic 2 the
246 About the classes C1F , . . . , C7F and S

symplectic groups are isomorphic to odd-dimensional orthogonal groups in

one higher dimension. The three families of examples are listed in Table 28.2
(see [44, Table 3.5.I]).

Table 28.2 Groups H ∈ CiF lying in members K ∈ S

GF class H K conditions
C4
F
SO+
2t
(q) Sp2 (q)  St Sp2t (q) q ≥ 4 even, t ≥ 4
C4
F
SO+
2t
(q) Sp2 (q)  St SO+
2t+2 (q) q ≥ 4 even, t ≥ 5 odd
F
SO+
4t
(q) C4 Sp4 (q)  St Sp4t (q) q even

Example 28.2 Let GF = SO+ 2t (q), with t ≥ 4 and q ≥ 4 even. Then the
class C4 contains the tensor product subgroup H := Sp2 (q)  St . On the
F

other hand H is also contained in the symplectic group K := Sp2t (q) as an

imprimitive subgroup from class C2F . But the 2t -dimensional spin representa-
tion of SO2t+1 (q) ∼
= Sp2t (q) embeds K as an absolutely irreducible and tensor
indecomposable subgroup, hence as a member from S, into GF = SO+ 2t (q).
This gives the first family of examples in Table 28.2. The other two are
obtained by a similar construction.

28.2 On the class S

At the time of writing this book, the two questions on groups contained in
the class S and their maximality (mentioned at the beginning of this section)
are far from being resolved, and probably a complete answer will never be
obtained. We content ourselves here with giving some first indications on how
these problems can be attacked, and where the difficulties lie. At this point
it seems unavoidable to use the classification of the finite simple groups.
Determining the members of S is tantamount to finding the degrees of
all absolutely irreducible, tensor indecomposable representations of finite
quasi-simple groups, and their invariant forms. This is far from being solved
at present. Nevertheless, for any classical group of fixed dimension, a list of
candidates for groups in S can be compiled.
The goal must be to eliminate as many candidates as possible by simple
arguments. The first very helpful observation relies on the order formulas
and is applicable in particular to embeddings of groups of Lie type in the
same characteristic. We encountered the order polynomial of a finite group
 28.2 On the class S 247

of Lie type and its factorization into cyclotomic polynomials, which in some
sense behaved like generic prime divisors of the order. There is one further
aspect to this factorization, given by the theorem of Zsigmondy (see [38, IX,
Thm. 8.3]):

Theorem 28.3 Let q > 1 be a prime power. Then there exists a prime
divisor  of q n − 1 which does not divide q i − 1 for 1 ≤ i ≤ n − 1, unless one
of:

(1) n = 2 and q = 2f − 1 is a Mersenne prime, or

(2) n = 6 and q = 2.

A prime divisor  = (n, q) of q n − 1 as in the theorem is called a primitive

prime divisor of q n − 1. Note that (n, q)|Φn (q), but does not divide Φi (q)
for i < n.

Example 28.4 As a simple application we claim that the smallest degree

d of a faithful representation of the finite symplectic group Sp2n (q) over
Fq is 2n. Indeed, any such representation defines an embedding Sp2n (q) ≤
GLd (q). Since |Sp2n (q)| is divisible by a primitive prime divisor (2n, q),
and |GLd (q)|q = (q − 1) · · · (q d − 1) (see Table 24.1), we conclude from
Theorem 28.3 that d ≥ 2n, except possibly for Sp6 (2) (where the claim can
still be shown to hold).

As a further application, we obtain the following result on values of cyclo-

tomic polynomials:
r s
Corollary 28.5 Let f = i=1 Φdi (x), g = i=1 Φei (x) ∈ Z[x] with not
necessarily distinct di , ei ∈ N. If f (q) = g(q) for some prime power q > 1,
then f = g unless one of:

(1) q = 3, f /g = (x − 1)2a /(x + 1)a with a ∈ Z, or

(2) q = 2, f /g = (x − 1)a (x + 1)b (x2 − x + 1)−b with a, b ∈ Z.

Proof Dividing both f, g by their greatest common divisor, we may assume

that they are coprime in Z[x]. Let n be maximal subject to Φn (x)|(f g).
If there exists a primitive prime divisor (n, q), then Φn (x) has to divide
both f and g, which is not possible. So, by the theorem of Zsigmondy either
q = 2 and n = 6, or n ≤ 2. In the latter case, (q − 1)a = (q + 1)b with
a + b ≥ 1 implies that q + 1 = 4 and a = 2b. If n = 6 and q = 2, then clearly
f /g = (x − 1)a (x + 1)b (x2 − x + 1)−b for some a, b ∈ Z.

Zsigmondy primes can often be used to rule out the existence of certain
embeddings of quasi-simple groups of Lie type in the same characteristic, like
248 About the classes C1F , . . . , C7F and S

for example B8 (q) ≤ E8 (q) for all q (using (16, q)). To deal with embeddings
in different characteristic, it is useful to know the minimal dimensions of
faithful representations of groups of Lie type in cross characteristic. These
turn out to be exponential in the rank of the group, and thus much larger
than dimensions of representations in the defining characteristic. We give
one example of such a result:
Proposition 28.6 (Landazuri–Seitz) Let K be a field of characteristic
r = p.

(a) Assume n ≥ 3, and let ρ : SLn (q) → GLd (K) be a representation of

SLn (q) over K with central kernel, where q = pf . Then d ≥ q n−1 − 1.
(b) Assume n ≥ 2, and let ρ : Sp2n (q) → GLd (K) be a representation of
Sp2n (q) over K with central kernel, where q = pf is odd. Then d ≥
q n−1 (q − 1)/2.
Proof For (a), let P ≤ SLn (q) be the maximal parabolic subgroup corre-
sponding to the subset I = {s2 , . . . , sn−1 } of the set of simple reflections,
with Levi decomposition

  
  
1 0  a 0 
P = U.L =  v ∈ Fq n−1

a det A = 1 .
v In−1  0 A 
Thus, the Levi complement L is isomorphic to GLn−1 (q), while the normal
subgroup U is elementary abelian of order q n−1 . Moreover, the Levi factor
L acts on U by conjugation via



−1 

−1 a 0 1 0 a 0 1 0
l.u = lul = −1 = −1
0 A v 1 0 A a Av 1
(for l ∈ L, u ∈ U ), so U is the natural module for SLn−1 (q) ≤ [L, L]. (See
Section 17.1 for the corresponding algebraic group situation.)
In particular, since n ≥ 3 we see that L acts transitively on the non-zero
vectors of U ∼
= Fn−1
q , whence it also acts transitively on the set of non-trivial
homomorphisms Hom(U, K̄ × ) \ {1U } from U into K̄ × , where K̄ denotes an
algebraic closure of K. By assumption, U is not contained in ker ρ, so ρ|U
contains one, hence by the action of L, all µ ∈ Hom(U, K̄ × ) \ {1U }. As |U | is
prime to the characteristic of K, |Hom(U, K̄ × )| = |U | = q n−1 , which shows
that d = dim ρ ≥ q n−1 − 1.
The proof of (b) is similar and is left as an exercise (see Exercise 30.23).
These so-called Landazuri–Seitz bounds show that subgroups of Lie type
in the wrong characteristic inside finite groups of Lie type are very tiny.
Indeed, if for example SLn (q) ≤ GLd (2) with q odd, then |SLn (q)| ∼ q n −1
2
 28.2 On the class S 249

by the order formula (see Example 24.8(2)), while the order of |GLd (2)| is
of the order of magnitude
2 n−1
−1)2 2n−2
2d = 2(q = 2q
by the bound for d in Proposition 28.6. So indeed these are extremely small
subgroups.
We close the chapter with some comments on our fourth question, the
maximality of the groups in class S. Here, one has to investigate the follow-
ing situation: we have embeddings H < K < GF = Cl(V )F (respectively in
2
SU(q 2 )), both H, K absolutely irreducible on V F (respectively V F ), with
HZ(G)/Z(G) and KZ(G)/Z(G) almost simple. By the classification of finite
simple groups, H, K are either sporadic, alternating, or of Lie type. More-
over, in the latter case, the characteristics of H, K may coincide or not, and
may or may not equal the characteristic of G. This leads to quite a number
of different possible configurations for (H, K, Cl(V )) to consider.
For example, if K is one of the 26 sporadic groups, the classification of
triples H < K < GF is essentially a question on maximal subgroups of
sporadic simple groups. In the situation where H is sporadic, one has to
determine all absolutely irreducible representations of sporadic groups and
their covering groups. Both are finite, but nevertheless very difficult problems
which have not yet been solved for all groups.
If K is an alternating group An , one has to decide which absolutely ir-
reducible representations of covering groups of An restrict irreducibly to
(almost simple) subgroups. The cases where H and K are of Lie type in dif-
ferent characteristics can essentially be dealt with using the Landazuri–Seitz
bounds.
This short overview should have convinced the reader that the methods
to study the groups in class S are of a representation theoretic nature; in
particular they lie outside the scope of this text.
 29
Finite exceptional groups of Lie type

In this chapter we are concerned with the determination of the maximal sub-
groups of the finite exceptional groups of Lie type. According to Table 22.1
these are the members of the ten infinite series
2
B2 (22f +1 ), 2 G2 (32f +1 ), G2 (q), 3D4 (q), 2F4 (22f +1 ),

F4 (q), E6 (q), 2E6 (q), E7 (q), E8 (q),

where f may be an arbitrary positive integer, and q an arbitrary prime

power. At first this might seem easier than the case of classical groups, since
there are just a finite number of underlying complete root data. And indeed,
complete lists of maximal subgroups have been obtained by Suzuki [78],
Cooperstein [18], Kleidman [42, 43], respectively Malle [55], for the first five
families in the above list. Thus, one may hope that eventually complete lists
will also be obtained for the remaining five families of large rank. Except for
some very “small” subgroups this has in fact been reached.
On the other hand the task is more difficult here since there is no conve-
nient natural representation in which to study these groups. For example, the
smallest faithful representation of G = E8 in its defining characteristic is the
adjoint one on its Lie algebra, of dimension dim G = 248. Representations
in any other characteristic are even much larger than this.

29.1 Maximal subgroups

Again, as for classical groups, it seems reasonable to attempt to use the
results on maximal subgroups of simple exceptional algebraic groups (see
Theorem 19.2), and study F -stability. Throughout we assume that G is an
exceptional simple algebraic group of adjoint type, so Z(G) = 1. (For the
 29.1 Maximal subgroups 251

other isogeny types the discussion before Remark 27.7 applies; by Table 9.2
this is only relevant for groups of types E6 and E7 .) The first result here is
(see [48, Thm. 1]):

Theorem 29.1 (Liebeck–Seitz) Let G be a simple exceptional algebraic

group of adjoint type, F : G → G a Steinberg endomorphism, and M ≤ G
an F -stable closed subgroup, maximal among positive-dimensional F -stable
closed subgroups. If M ◦ is simple, assume that p = 0 or p > 7. Then one of
the following holds:

(1) M is a parabolic subgroup,

(2) M is the normalizer of some connected reductive subgroup of maximal
rank,
(3) M is the normalizer of some semisimple subgroup not of maximal rank
as in Theorem 19.1(3),
(4) G = E7 , p = 2, M = (Z22 × D4 ).S3 ,
(5) G = E8 , p = 2, 3, 5, M = A1 × S5 , or
(6) G = E8 , p = 2, M = (A1 G2 G2 ).S2 .

In the next step, one needs to descend to the finite groups GF .

Let H ≤ GF be maximal. If H normalizes some non-trivial F -stable con-
nected proper subgroup D of G, then HD lies in some closed F -stable
maximal subgroup M ≤ G of positive dimension, and maximality implies
H = MF .
For the types occurring in the preceding theorem, the possible classes of
fixed point groups M F can be worked out, as seen for tori, Levi subgroups,
and so on, in the preceding chapters, by using Theorem 21.11.
So we may assume that H normalizes no non-trivial F -stable connected
proper subgroup of G. Let’s first consider the case that H normalizes a non-
trivial F -stable r-subgroup R ≤ G for some prime r, that is, H is an r-local
subgroup. If r = char(k), then H lies in a proper F -stable parabolic sub-
group of G, by Theorem 26.5, which is of positive dimension, so we have
r = char(k). Since NG (R) ≤ NG (R0 ) for any characteristic normal subgroup
R0 of R, we may (and will) assume by maximality of H that R is elemen-
tary abelian and moreover that it has no proper H-invariant non-trivial
subgroups.

Definition 29.2 Let G be a simple algebraic group. An elementary abelian

r-subgroup R of G, with r = char(k), is called a Jordan subgroup of G if it
satisfies the following conditions:

1. CG (R) (and hence NG (R)) is finite,

252 Exceptional groups of Lie type

2. R is a minimal normal subgroup of NG (R),

3. NG (R) is maximal subject to conditions 1. and 2., and
4. there is no non-trivial connected NG (R)-invariant proper subgroup of G.
Thus, by what we said above the investigation of local maximal subgroups
naturally leads us to the study of Jordan subgroups. For classical groups
these are precisely the images in Gad of extraspecial groups with normalizer
in class C5 . For simple groups of exceptional type, some further interesting
examples occur (see [8, Thm. 1] and [17, Thm. 1]). In the next statement,
the notation ra , with r a prime and a ≥ 1, denotes an elementary abelian
r-group of this order, while ra+b stands for an (unspecified) extension of an
elementary abelian r-group of order ra by one of order rb .
Theorem 29.3 (Borovik, Cohen–Liebeck–Saxl–Seitz) Let G be a simple
exceptional algebraic group of adjoint type with Steinberg endomorphism F :
G → G. Then the Jordan subgroups R in GF and their normalizers H =
NGF (R) are given as follows:
(1) G = G2 , R = 23 and H = 23 .SL3 (2),
(2) G = F4 , R = 33 , H = 33 .SL3 (3),
(3) G = E6 , R = 33 , H = 33+3 .SL3 (3), or
(4) G = E8 , R = 25 , H = 25+10 .SL5 (2), or R = 53 , H = 53 .SL3 (5).
In each case, these are unique up to GF -conjugation.
The normalizers NGF (R) of these Jordan subgroups are called exotic local
subgroups of exceptional groups of Lie type.
Sketch of proof First note that a Jordan subgroup R ≤ GF cannot be con-
tained in a torus T of G. Otherwise, CG (R) would contain T and hence
be infinite, contrary to our assumption that R is a Jordan subgroup. Now
let π : Gsc → G be the natural isogeny from a simply connected group of
the same type as G as in Proposition 9.15. If | ker(π)| is prime to r, then
π −1 (R) = R̃ × ker(π) for some subgroup R̃ ≤ Gsc isomorphic to R (so
R = π(R̃)), which itself cannot be contained in any torus of Gsc . It then fol-
lows from Corollary 14.17 that r must be a torsion prime for G and moreover
|R| ≥ r3 . For simple groups of exceptional type the condition that ker(π)
is prime to r is satisfied unless (G, r) ∈ {(E6 , 3), (E7 , 2)} (see Table 9.1). In
the latter two cases, we still have that r is torsion. Thus, by Table 14.1 we
have r = 2 for G2 , r ≤ 3 for F4 , E6 , E7 , and r ≤ 5 for E8 .
The various possibilities are now analyzed case-by-case by using informa-
tion on the centralizers of semisimple elements. Assume for example that R is
a Jordan subgroup in G = G2 and thus r = 2. Let s ∈ R \ {1}. Then one can
 29.1 Maximal subgroups 253

show that CG (s) ∼ = SL2 ◦SL2 , the subgroup constructed in Example 14.19(2).
By Example 14.19(3) both factors contain a quaternion subgroup of order 8.
Their central product in CG (s) is then extraspecial of order 32 and contains
a self-centralizing elementary abelian subgroup R of order 8, with the stated
normalizer. Similarly, for G = F4 and r = 3, respectively G = E8 and r = 5
there is s ∈ R \ {1} with centralizer SL3 ◦ SL3 , respectively SL5 ◦ SL5 , cor-
responding to maximal rank subsystems of type A2 A2 , respectively A4 A4 ,
inside which R can be constructed.

Note that there do exist maximal local subgroups which are not covered by
Theorem 29.3: they are normalizers of some positive-dimensional semisimple
subgroups, as for example in Theorem 29.1(4)!
To complete our investigation of maximal subgroups now assume that
H ≤ GF is not local, and does not normalize a proper connected subgroup
of positive dimension, that is, it is not covered by Theorem 29.1. Then its
generalized Fitting subgroup is a direct product F ∗ (H) = H1 × · · · × Ht
of non-abelian simple groups Hi . We aim to show that t = 1 (and hence
H is almost simple), so instead assume t > 1. For each i, let CC(Hi ) :=
CG (CG (Hi )) be the double centralizer of Hi .

Lemma 29.4 In the above situation the following holds:

(a) The CC(Hi ), 1 ≤ i ≤ t, mutually commute.

t
(b) If t > 1, the commuting product H ∗ := i=1 CC(Hi ) is finite.

Proof Clearly for each j = i we have Hj ⊆ CG (Hi ), so CG (Hj ) ⊇ CC(Hi )

and CC(Hj ) ⊆ CG (CC(Hi )), whence (a).
t
Thus H ∗ := i=1 CC(Hi ) is an F -stable commuting product, normalized
by H (because H normalizes each Hi , so each C(Hi ), so each CC(Hi )).
Moreover, H ∗ < G as t > 1, thus H ∗ is finite by assumption.

Now for any g ∈ Hi we have CG (g) ⊇ CG (Hi ), so Z(CG (g))◦ ⊆ CC(Hi ).

As H ∗ is finite by the lemma, we must have dim Z(CG (g))◦ = 0. Taking g
semisimple and of prime order, CG (g)◦ is a connected reductive subgroup
containing a maximal torus and with finite center, so it is semisimple and of
maximal rank. By the classification of maximal rank subgroups in Chapter 13
we see that in exceptional groups this can only happen for elements of order
2, 3 or 5 (the latter only in E8 ). So if G = E8 , only the primes 2, 3 can
divide |F ∗ (H)|, whence F ∗ (H) is solvable by Burnside’s pa q b -theorem, which
contradicts our assumption that H is non-local.
Thus t > 1 necessarily forces G = E8 , p > 5, and the Hi are 2, 3, 5-groups.
By the classification of finite simple groups, this implies that Hi is one of
254 Exceptional groups of Lie type

A5 , A6 , or PSU4 (2). Finally one can show that the only possibility is G = E8
and F ∗ (H) = A5 ×A6 , see Borovik [8]. Otherwise, t = 1 and so H ≤ NG (H1 )
is almost simple.
We have thus sketched the proof of the following reduction theorem (see
[48, Thm. 2]):

Theorem 29.5 (Borovik, Liebeck–Seitz) Let G be an exceptional simple

algebraic group, F : G → G a Steinberg endomorphism, and H < GF maxi-
mal. Then one of the following holds:

(1) H = M F for M as in Theorem 29.1,

(2) H is an exotic local subgroup as in Theorem 29.3,
(3) G = E8 , p > 5, H = (A5 × S6 ).2, or
(4) H is almost simple.

This is a first important reduction step. In order to complete the deter-

mination of all maximal subgroups of the finite exceptional groups of Lie
type, one needs to investigate groups which might occur in conclusion (4)
of the theorem. Here, two essentially different situations arise: if H is of Lie
type in characteristic p, one would like to use the results on the subgroup
structure of the corresponding simple algebraic groups and the representa-
tion theory of the finite groups as in the second part. To what extent such
embeddings arise from embeddings of algebraic groups is discussed in the
next section. One class of examples here is given by centralizers of graph-,
field- and graph-field automorphisms of G.
On the other hand, if H is not of Lie type in characteristic p, then one
can use the bounds of Landazuri–Seitz type (see Proposition 28.6) in order
to get down to a finite (but still lengthy) list of possibilities for H. This can
further be reduced by comparing the maximal r-rank of G to that of H for
various primes r.
The complete list of candidates in conclusion (4) has at present not yet
been obtained. Generally speaking, the smaller H is, the more difficult it is
to find all embeddings into GF up to conjugation. See [52] for a survey of
recent results.

29.2 Lifting result

We derived information on the subgroup structure of finite groups of Lie
type by considering fixed points of closed subgroups of an algebraic group
under a Steinberg endomorphism. Now consider the converse question: given
 29.2 Lifting result 255

a subgroup of a finite group of Lie type, does it arise as the group of fixed
points under a Steinberg endomorphism of some closed positive-dimensional
subgroup of the associated algebraic group? In particular, when considering
the configurations of Theorem 29.5(4), when H is again a finite group of
Lie type in characteristic p, one is led to ask whether H is the fixed point
subgroup of some maximal positive-dimensional closed connected subgroup?
Let’s make this question a bit more precise. Let G, H be simple algebraic
groups over k = Fp . Let F : G → G, F1 : H → H be Steinberg endomor-
phisms, so GF , H F1 are finite groups of Lie type. Let ϕ : H F1 → GF be a
group homomorphism. Then under what conditions can ϕ be lifted to an
appropriate morphism of algebraic groups, that is, when does there exist
ϕ : H → G, a morphism of algebraic groups, such that ϕ|H F1 = ϕ? The first
major result in this direction is due to Steinberg [70, Thm. 1.3]:
Theorem 29.6 (Steinberg) Let V be a finite-dimensional vector space over
Fq , q = pa . Let ϕ : H F1 → GL(V ) be an absolutely irreducible representation
of H F1 , i.e., H F1 acts irreducibly on V ⊗Fq k. Then there exists a rational
representation ϕ : H → GL(V ⊗Fq k) such that ϕ|H F1 = ϕ. In particular,
absolutely irreducible representations of finite groups of Lie type over fields
of characteristic p are also parametrized by highest weights.
Remark 29.7 In fact Steinberg also shows how to obtain a complete set
of absolutely irreducible representations for H F1 . In case F1 is a standard
Frobenius morphism, say a q-power map, then
{L(λ) | 0 ≤ λ, α∨  < q for all α ∈ ∆}
forms a complete set of non-isomorphic irreducible kH F1 -modules.
The following example shows that the condition of absolute irreducibility
is necessary.
Example 29.8 Let GF = SL2 (3) (isomorphic to a 2-fold central extension
of A4 ) with N  GF the normal Sylow 2-subgroup, so GF /N is cyclic of
order 3, generated by c. Consider the representation
⎛ ⎞
1 1 1
GF /N → GL3 (F3 ), c → ⎝0 1 1⎠ .
0 0 1
This defines an indecomposable but reducible representation of GF .
We claim that there does not exist a rational representation of G :=
SL2 (F3 ) whose restriction to GF is the above described representation. Sup-
pose the contrary and let V be the underlying F3 G-module which affords the
256 Exceptional groups of Lie type

representation. Let T ≤ B ≤ G be the subgroups of SL2 defined in Exam-

ple 6.7(2) and W = NG (T )/T . For 1 = w ∈ W there exists ẇ ∈ NG (T ) such
that ẇ ∈ N ≤ GF . So W acts trivially on all weights in the representation.
But this implies that the only weight occurring is the 0 weight. (See Exer-
cise 20.18.) So V has an F3 G-composition series 0 ≤ M1 ≤ M2 ≤ V with
one-dimensional quotients. But this implies that G is solvable, contradicting
the simplicity of G.

The goal of this section is to discuss results that generalize Theorem 29.6.
It is perhaps not surprising that one can obtain a similar result when GL(V )
is replaced by one of the classical groups SL(V ), Sp(V ) or SO(V ), making
some appropriate assumptions on the action of H F1 on the natural module
for the classical group. In fact, one can also say something for representations
inside groups of exceptional type. We just need to know that a semisimple
group is “recovered” by its action on its Lie algebra, as shown in [69, 4.2]:

Theorem 29.9 (Seligman, Steinberg) Let G be a semisimple linear alge-

braic group over k with Lie algebra g, and Ad : G → GL(g) the adjoint
representation. Then Ad (G) has finite index in Aut(g), the group of Lie
algebra automorphisms of g. In particular, Ad (G) = Aut(g)◦ .

Returning now to the question of lifting homomorphisms between finite

groups to morphisms of algebraic groups, let’s first observe the following: if
H is a simple algebraic group with a Steinberg endomorphism F1 : H → H,
and X := [H F1 , H F1 ] is perfect, then X ∼ = HscF1
/Z F1 for some group Hsc
of simply connected type and some Z ≤ Z(Hsc ) (with F1 lifted to Hsc by
Proposition 22.7) by Proposition 24.21(b), so in particular any homomor-
phism ϕ : X → GF can be regarded as a homomorphism Hsc F1
→ GF . Thus
it suffices to construct lifts of homomorphism in the case that H is simply
F1
connected and X = Hsc is perfect (note that the latter is not a serious re-
striction by Theorems 24.15 and 24.17). We start by treating a special case.
Its proof is a good illustration of the interplay of many of the methods and
concepts presented in this text.

Proposition 29.10 Let X, H and F1 be as above and let G be a simply

connected simple algebraic group of exceptional type with Steinberg morphism
F : G → G. Let ϕ : X → GF be a homomorphism such that ϕ(X) lies in
no F -stable, closed connected proper subgroup of G. Assume in addition that
p > 3 dim G. Then ϕ can be extended to a morphism ϕ : H → G of algebraic
groups with ϕ|X = ϕ.

Sketch of proof Since p > 3 dim G we have in particular p ≥ 11.

 29.2 Lifting result 257

(1) We first construct a one-dimensional subtorus of the group GL(g),

where g = Lie(G), which will play a key role in what follows. Let Ad G : G →
GL(g) denote the adjoint representation of G. By Exercise 30.24, there exists
J ≤ X with J ∼ = SL2 (p). Now let S be the subgroup of J corresponding
to the group of diagonal matrices in SL2 (p). So S ≤ J is isomorphic to
F× ×
p ; let Fp → S, c → S(c), denote such an isomorphism. As S is cyclic,
Proposition 26.6 shows that S lies in an F -stable maximal torus T of G.
Let Φ be the set of roots of G with respect to T . Fix a basis CT of Lie(T )
and for each α ∈ Φ, choose vα ∈ gα \ {0}, so that C = CT ∪ {vα | α ∈ Φ}
is a basis of g. Using this basis, identify GL(g) with the group of invertible
n × n-matrices (where n = dim g); so we have Ad G (T ) ≤ Dn , the group of
diagonal matrices.
By Theorem 29.6 and Remark 29.7, the composition factors of J on g
are realized as restrictions of restricted irreducible representations of SL2 .
As p > 3 dim G = 3 dim g, the J-composition factors on g are of dimension
strictly less than p3 . By Exercise 20.18, the action of S on any J-composition
factor is diagonalizable with weights l ∈ Z satisfying − p−1 p−1
3 < l < 3 , where
the weights are defined by S(c) → cl , for c ∈ F× p . So in the present situation
we have, for c ∈ F× p , Ad G (S(c))v = v for all v ∈ CT and Ad G (S(c))vα =
clα vα for some integers lα satisfying − p−13 < l p−1
α < 3 .
We now define a cocharacter γ ∈ Y (Dn ): for a ∈ k × , set
(i) γ(a)v = v for all v ∈ CT .
(ii) γ(a)vα = alα vα , for α ∈ Φ.
Set S := im(γ), a one-dimensional subtorus of GL(g); so γ(c) = Ad (S(c)),
for all c ∈ F×
p.
(2) We claim that S ≤ Ad G (G). Indeed, we will show that S acts as
◦
a group of Lie algebra automorphisms of g and so S = S ≤ Aut(g)◦ =
Ad G (G), by Theorem 29.9.
Let α, β ∈ Φ with [vα , vβ ] = 0. Then considering the action of the torus
T , we see that [vα , vβ ] is a scalar multiple of vα+β . Thus for all c ∈ F×
p,

clα clβ [vα , vβ ] = [γ(c)vα , γ(c)vβ ] = γ(c)[vα , vβ ] = clα+β [vα , vβ ]

so clα +lβ = clα+β for all c ∈ F× p . Using the fact that lα , lβ lie in the interval
[− p−13 , p−1
3 ] we see that l α + l β = lα+β and so a
lα +lβ
= alα+β for all a ∈ k × .
One easily checks that the action on the remaining commutators [v, v  ], for
v, v  ∈ C is also preserved by γ(a) and so S acts as a group of automorphisms
of the Lie algebra g, as claimed.
(3) We next show that the closed subgroup R = (Ad −1 ◦
G (S)) is F -stable.
Write F |X = qφ, for some positive integral power q of p and for some
258 Exceptional groups of Lie type

φ ∈ Aut(XR ), as in Proposition 22.2 (recall p ≥ 11). As p ≥ 11, Theo-

rem 15.20 shows that g is an irreducible kG-module; moreover by Exam-
ple 15.19 its highest weight is the highest root α0 , and hence invariant under
all graph automorphisms of G by the defining property of the highest root.
The highest weight of the representation Ad G ◦F of G is then just q times the
highest weight of the adjoint representation. Theorem 15.17(a) implies that
the representations Ad G ◦ F and F  ◦ Ad G are equivalent, where F  is a q-
power standard Frobenius morphism of GL(g). Thus there exists x ∈ GL(g)
such that xAd G (F (g))x−1 = F  (Ad G (g)) for all g ∈ G. Applying Theo-
rem 21.7, we get that x = y −1 F (y) for some y ∈ GL(g), which then implies
that
Ad G (F (g)) = ωAd G (g)ω −1 for each g ∈ G,
  q
where ω is the invertible semilinear map cj v j → cj vj on g with respect
to our fixed basis of g.
For t ∈ T , α ∈ Φ, and setting g = F −1 (t) in the above, we have
Ad G (t)ωvα = ωAd G (F −1 (t))vα = (α(F −1 (t))q )ωvα .
But α(F −1 (t))q = (F −1 (α))(t)q = (φ−1 (α))(t) = (ρ(α)(t)) (with ρ as in
Proposition 22.2). So Ad G (t)ωvα = ρ(α)(t)ωvα for all t ∈ T and we conclude
that ωvα = bvρ(α) for some b ∈ k × . Hence for a ∈ k × ,

ωγ(a)ω −1 vρ(α) = ωγ(a)b−1/q vα = ωalα b−1/q vα

= aqlα b−1 ωvα = aqlα vρ(α) .
Similarly, ωγ(a)ω −1 v = v, for v ∈ CT .
Taking a in the prime field we have γ(a) ∈ Ad G (S), so ω centralizes
γ(a). But then the above equality and the restriction on the lα show that
lρ(α) = lα and then applying the same equality for arbitrary a ∈ k × , we see
that ωγ(a)ω −1 = γ(a)q . Finally, we have
Ad G (F (Ad −1 −1
G (γ(a)))) = ωAd G (Ad G (γ(a)))ω
−1
= γ(a)q ≤ S
and so Ad −1 −1 ◦
G (S) is F -stable and hence R = (Ad G (S)) as well.
(4) We can now show that X acts irreducibly on g. For suppose that V
is a proper X-invariant subspace of g, so V is spanned by weight vectors
for S. We claim that weight vectors for S are in fact weight vectors for S.
For suppose two S-weights on g have equal restrictions to S. Then clα = clβ
for some α, β ∈ Φ and for some generator c ∈ F× p ; that is, c
lα −lβ
= 1, so
(p − 1)|(lα − lβ ). But − 3 ≤ lα , lβ ≤ 3 , so lα = lβ as claimed. Now set
p−1 p−1

D = xRx−1 | x ∈ X, a closed connected subgroup of G by Proposition 1.16,

which is F -stable as R is. Moreover D stabilizes the subspace V and so by
 29.2 Lifting result 259

Theorem 15.20, D is proper in G. But the group [X, S] is normal in X and

contains the subgroup J, so we have
X = [X, S] ≤ [X, RZ(G)] = [X, R] ≤ D,
which contradicts our standing assumption. Hence X acts irreducibly as
claimed.
(5) A lengthy argument relying upon detailed considerations of the rep-
resentation theory of X and G, in particular, Theorem 29.6 and Proposi-
tion 16.3, shows that the groups H and G have isomorphic root systems and
the absolutely irreducible representation Ad G ◦ϕ : X → GL(g) is the restric-
tion of a twist of the adjoint representation of H, that is, there exists a stan-
dard Frobenius endomorphism F  of H such that Ad G ◦ ϕ = (Ad H ◦ F  )|X .
We refer the reader to [62, pp.565–566] for the details.
We now have that X stabilizes two Lie brackets on g: [ , ]G coming from g
and [ , ]H coming from Lie(H). Now the existence of a Lie bracket on g shows
that HomkG (g ∧ g, g) is non-trivial. Moreover, using the theory of highest
weights and the assumption on p, one can check that for each exceptional
group G, g occurs with multiplicity 1 as a composition factor of g ∧ g and so
HomkG (g∧g, g) is a 1-space. (See Exercise 20.21 for the case G = G2 .) Hence
the two Lie brackets are scalar multiples of each other and Theorem 29.9
implies that Ad H (F  (H)) = Ad G (G).
Recall that H is simply connected. Moreover the isogeny Ad G : G →
Ad G (G) satisfies ker dAd = ker ad = 0, by the restriction on p and Theo-
rem 15.20. Hence we may apply Proposition 9.18 to obtain a morphism ψ :
H → G such that Ad H = Ad G ◦ ψ. But then Ad G (ϕ(x)) = Ad H (F  (x)) =
Ad G (ψ(F  (x))) for all x ∈ X. By Theorem 24.15, X is generated by its
unipotent elements and Ad G is bijective on unipotent elements, so we have
ϕ = ψ ◦ F  |X , whence ψ ◦ F  is the desired extension.
One of the first generalizations of Theorem 29.6 is the following result,
where the condition “absolute irreducibility” is replaced by requiring that
the subgroup stabilize no proper totally singular subspace of the natural
module in the classical group case. (See Proposition 12.13.)
Theorem 29.11 Let G be a semisimple algebraic group defined over k = Fp
with Steinberg endomorphism F : G → G, and H, X as before. Let ϕ : X →
GF be a homomorphism such that ϕ(X) lies in no proper F -stable parabolic
subgroup of G. Then there exists an integer N , depending on the maximal
dimension of a simple factor of G, such that if p > N , ϕ can be extended to
a morphism ϕ : H → G of algebraic groups with ϕ|X = ϕ. Moreover, if G is
classical, no restriction on p is necessary.
260 Exceptional groups of Lie type

Proof We prove the result by induction, taking G as a counterexample of

minimal dimension. Take N = 3m where m is the maximal dimension of
a simple factor of G; in particular, p ≥ 11. As pointed out above we may
assume that H is simply connected.
Step 1: There is no F -stable, closed connected proper subgroup D < G with
ϕ(X) ≤ D.
Indeed, suppose ϕ(X) ≤ D < G as above. If D is not reductive, then
1 = Ru (D) is an F -stable unipotent subgroup. Hence as in the proof of
Theorem 26.5, ϕ(X) ≤ D ≤ NG (Ru (D)) lies in a proper F -stable parabolic
subgroup, contradicting the assumption on ϕ(X). Hence D is reductive and
as X is perfect, ϕ(X) ≤ [D, D], a semisimple F -invariant subgroup of dimen-
sion less than dim G. So by minimality of G, we have the desired extension
of ϕ.
Step 2: We may suppose that G is of simply connected type.
Indeed, let π : Gsc → G be the natural isogeny from a group Gsc of
simply connected type and F̂ : Gsc → Gsc a Steinberg morphism with π ◦
F̂ = F ◦ π (see Proposition 22.7). Since H is simply connected, the Schur
multiplier of X has order prime to | ker π| (see [44, Thm. 5.1.4]). Thus,
by the defining property of the Schur multiplier (see [39, Cor. 11.20(a)])
the full preimage π −1 (ϕ(X)) is a direct product of the abelian group ker π
with a perfect subgroup X̂ isomorphic to ϕ(X) = [ϕ(X), ϕ(X)]. Since X̂ =
[π −1 (ϕ(X)), π −1 (ϕ(X))], X̂ is characteristic and π|X̂ is an isomorphism.
Hence, X̂ is centralized by F̂ . So assuming the result holds for G simply-
connected, there exists a homomorphism ϕ̂ : X → X̂ ≤ Gsc such that
ϕ = π ◦ ϕ̂. Now, if ϕ : H → Gsc lifts ϕ̂, then π ◦ ϕ is the desired lift of ϕ.
Now let M1 , . . . , Mt be the simple factors of the simply connected group
G.
Step 3: F acts transitively on the set {M1 , . . . , Mt }.
Let A be the product of the elements of an F -orbit on {M1 , . . . , Mt }. If
γ is the projection of G onto A, then γ ◦ ϕ is a homomorphism from X to
the F -invariant subgroup A of G. If F is not transitive, then A < G is of
smaller dimension, and ϕ can be lifted by Step 1.
Step 4: G is simple.
By Step 3 we may assume that F i (M1 ) = Mi+1 for 1 ≤ i ≤ t − 1. Then
M = {x · F (x) · F 2 (x) · . . . · F t−1 (x) | x ∈ M1 } ∼
= M1
is a closed connected F t -stable simple subgroup of G. Moreover, as ϕ(X) ≤
GF , ϕ(X) ≤ M . Note that ϕ(X) does not lie in an F t -stable proper parabolic
of M , as such a parabolic would be diagonally embedded in a subgroup
 29.2 Lifting result 261

P · F (P ) · · · F t−1 (P ) for some F t -stable proper parabolic subgroup of M1 ,

contradicting our hypothesis on X. So we now have F t : M → M with
t
ϕ(X) ≤ M F , not contained in any F t -stable parabolic of M , and p >
3 dim G ≥ dim M , so by Step 1 we necessarily have that M = G is simple.
If G is simple of simply connected type, we may extend ϕ to ϕ : H → G
by Proposition 29.10, contradicting our choice of G.

With the same proof, but looking carefully at possible X-composition

factors on Lie(G) and the corresponding weights for S, as in the proof of
Proposition 29.10, one gets an improved result, where we may take N = 13,
unless H is of type A1 , A2 , or B2 . In particular, the worst case occurs if H
is of type A1 and G has a factor of type E8 , with N = 113 (see [62, Thm. 2]
for a precise statement).
Finally we conclude this discussion with the statement of the most general
lifting result for exceptional groups established at this point in time. We first
introduce one additional notion:
Let Φ be an irreducible root system; we call an element of the root lattice
ZΦ a root difference if it is of the form α − β for some α, β ∈ Φ. Given a
sublattice L of ZΦ, we write t(L) for the exponent of the torsion subgroup
of the quotient ZΦ/L; we set

t(Φ) = max{t(L) | L a sublattice of ZΦ generated by root differences}.

Take H, F1 and X as before the statement of Proposition 29.10 and let

F1 = qφ as in Proposition 22.2. Write Φ(H) for the root system of H with
respect to a fixed maximal torus. Then we have the following result ([51,
Thms. 1, 4 and 10]):

Theorem 29.12 (Liebeck–Seitz) Let G be an exceptional algebraic group

of adjoint type over Fp with root system Φ and Steinberg endomorphism
F : G → G, such that H F1 = X ≤ GF .

(a) Assume further that X = SL3 (16), SU3 (16) and


q > t(Φ(G)) · (2, p − 1) if X = A1 (q), 2 B2 (q 2 ), or 2 G2 (q 2 ),
q>9 otherwise.

Then there exists a closed connected F -stable subgroup X of G normal-

ized by NG (X) with X ≤ X and such that X stabilizes every X-invariant
subspace of Lie(G). Moreover, if X is not of the same type as G, that is
Φ(H) and Φ are not isomorphic, then X may be chosen to be a proper
subgroup of G.
262 Exceptional groups of Lie type

(b) Assume in addition that p > 7. Then X lies in a closed connected

semisimple F -stable subgroup X of G, where each simple factor of X
is of type Φ(H).
In view of this result it is interesting to know the value t(Φ) for each of
the exceptional type root systems. This has been calculated by Lawther (see
[46, Thm. 2]):
Proposition 29.13 The values t(Φ) for Φ an indecomposable root system
of exceptional type are as follows: t(G2 ) = 12, t(F4 ) = 68, t(E6 ) = 124,
t(E7 ) = 388, t(E8 ) = 1312.
 30
Exercises for Part III

Throughout all algebraic groups are considered over k = Fp , an algebraic

closure of Fp .
Exercise 30.1 Let G be a linear algebraic group and σ : G → G a surjec-
tive endomorphism. Show the following:
(a) ker(σ) is finite and lies in Z(G).
(b) If G◦ is semisimple, then ker(σ) = 1, so σ is an automorphism of abstract
groups.
(c) If H ≤ G is connected and σ(H) ≤ H, then σ(H) = H.
[Hint: Use that dim G = dim σ(G) and Exercise 10.4 for (a). For (b), consider the
powers σ n of σ. For (c), use again Corollary 1.20.]

Exercise 30.2 Let F : G → G be a surjective endomorphism of a group

G and define
F1 : Gm → Gm , (g1 , . . . , gm ) → (g2 , . . . , gm , F (g1 ))
and
F2 : Gm → Gm , (g1 , . . . , gm ) → (F (g2 ), . . . , F (gm ), F (g1 )).
Show that (Gm )F1 ∼
= GF and (Gm )F2 ∼
m
= GF .
Exercise 30.3 Let W be a finite group and σ ∈ Aut(W ). Show that
σ-classes in W are in natural bijection with W -conjugacy classes in the
coset W.σ, and the size of such a conjugacy class is given by |[wσ]| =
|W |/|CW (wσ)|.
Exercise 30.4 Let F : G → G be a Steinberg endomorphism on the
connected group G. Then in the semidirect product G
F  we have G.F =
F G . In particular, GgF is G-conjugate to GF for all g ∈ G.
264 Exercises for Part III

Exercise 30.5 Let G be a connected group with Steinberg endomorphism

F : G → G, acting transitively on a set V with compatible F -action. Let
v ∈ V F with closed stabilizer Gv . Show that the map g.v → g −1 F (g) induces
a bijection between GF -orbits on V F and F -classes in Gv .
[Hint: Use the Lang–Steinberg Theorem for the connected group G◦v .]

Exercise 30.6 Let G be simple with Steinberg endomorphism F : G → G,

T , B, X, Φ, ρ, qα as in Proposition 22.2. Assume that qα is not constant
on Φ. Show that G is of type B2 , G2 or F4 with p = 2, 3, 2 respectively, ρ
interchanges long and short roots and qα qβ = q 2 , qβ /qα = p for all long roots
α and all short roots β.
[Hint: Let α, β ∈ Φ with α long, β short and qα = qβ . By Proposition 22.2 we have
F ◦ sρ(α) = sα ◦ F . Apply this to ρ(β) to deduce that n(ρ(β), ρ(α))qα = n(β, α)qβ .
By Table A.1 we have n(β, α) = 1 and n(ρ(β), ρ(α)) ∈ {1, 2, 3}. But qβ /qα is a
power of p.]

Exercise 30.7 Let G be a linear algebraic group with a Steinberg en-

domorphism F : G → G, and H1 , H2 two F -stable closed subgroups of G
such that H2 is connected and normal in H1 . Show that the natural map
H1F /H2F → (H1 /H2 )F is an isomorphism.

Exercise 30.8 (Tits) Let G be a group with a BN-pair (B, N ), let G1 be

a subgroup of G and B1 := B ∩ G1 , N1 := N ∩ G1 . Show the following:

(a) If (B ∩ N )B1 = B then (B1 , N1 ) is a BN-pair in G1 .

(b) If moreover G1 is normal in G, then its Weyl group can be naturally
identified with the one of G and we have G = (B ∩ N )G1 .

Exercise 30.9 Let G be connected reductive with a Steinberg endomor-

phism F : G → G, T an F -stable maximal torus contained in an F -stable
Borel subgroup of G, with root system Φ and corresponding positive system
Φ+ . Show that
"
|GF | = Q |T F | qw
w∈W F

with
! !
Q := qα and qw := qα
α∈Φ+ α∈Φ+ ,w(α)∈Φ−

with qα as in Proposition 22.2.

Exercise 30.10 Let G be connected reductive with Steinberg endomor-

phism F : G → G, H ≤ G a closed connected reductive F -stable subgroup,
 Exercises for Part III 265

with complete root data G, respectively H. Then the polynomial |H| divides
|G| in Z[x].
[Hint: You may want to use that both polynomials are monic and that the rational
function |G|/|H| takes integral values for infinitely many prime powers q.]

Exercise 30.11 Use Example 23.10(2) to show that SU3 (q) is perfect for
q>2
Exercise 30.12 (Steinberg) Let G be a group, Z ≤ G a central subgroup,
and F : G → G an automorphism normalizing Z. Then the exact sequence
π
1 −→ Z −→ G −→ G/Z −→ 1
induces a long exact sequence
π δ
1 −→ Z F −→ GF −→ (G/Z)F −→ (L(G) ∩ Z)/L(Z) −→ 1,
where L : G → G is defined by L(g) := g −1 F (g).
[Hint: Show that δ := ν ◦ L ◦ π −1 , with ν : Z → Z/L(Z) the natural map, is a
well-defined homomorphism. See for example [26, Prop. 4.2.3].]

Exercise 30.13 Let G be a simple algebraic group with Steinberg en-

domorphism F : G → G. Show that GF = [GF , GF ] T F for any F -stable
maximal torus T of G.
[Hint: Let π : Gsc → G be an isogeny from a simply connected group with central
kernel Z, let Tsc be the full preimage of T , a maximal torus of Gsc . Let G̃ := Gsc ◦Z
Tsc be the central product of Gsc with Tsc over Z and let F act componentwise on G̃.
Then π̃ : G̃ → G, (g, t) → π(g), is an epimorphism with connected kernel Tsc . Apply
Proposition 23.2 to see that π̃|G̃F is surjective onto GF and Proposition 24.21 to
conclude.]

Exercise 30.14 (Maximal tori in finite classical groups)

(a) Classify the maximal tori in GUn (q) and work out their orders.
(b) Classify the maximal tori in Sp2n (q) and work out their orders.
Exercise 30.15 (Molien’s formula) Let V be a finite-dimensional complex
vector space, W ≤ GL(V ) finite and φ ∈ NGL(V ) (W ). Denote by V W the
invariants of W in V .
1

(a) Show that tr(φ, V W ) = |W | w∈W tr(φw, V ).
(b) For a graded complex W -module ⊕i≥0 Vi and g ∈ GL(⊕Vi ) define the

graded trace grtr(g, ⊕Vi ) := i≥0 tr(g, Vi )xn ∈ C[[x]]. Now the symmet-
ric algebra S(V ) of V is a graded W -module. For g ∈ GL(V ) show that
grtr(g, S(V )) = detV (1 − gx)−1 .
266 Exercises for Part III

(c) Compute the graded trace of φ on the W -invariants S(V )W in S(V ) as

1 " 1 " 1
grtr(φ, S(V )G ) = grtr(φw, S(V )) = .
|W | |W | detV (1 − φwx)
w∈W w∈W

(d) Let f1 , . . . , fn be homogeneous generators of S(V )W which are eigenvec-

tors for φ with eigenvalues 1 , . . . , n . Conclude that
!
n
1 1 " 1
= .
i=1
1 −  i x di |W | det V (1 − φwx)
w∈W

[Hint: For (a) use that v → 1
|W | w∈W w.v defines a projection from V to V
W
.
For (b) choose a basis of V with respect to which g is triangular and compute the
trace on each homogeneous component. See [9, V, Lemma 5.3] for the case where
φ = idV .]

Exercise 30.16 Let (T, F ) be a torus with complete root datum T =

(X, ∅, Y, ∅, φ). Then S = (X  , ∅, Y  , ∅, φ), with Y  = kerY (Φd (φ)) and X  =
X/Ann(Y  ), is the complete root datum of the Sylow d-torus (S, F ) of T .
[Hint: Show that kerY (Φd (φ)) is a pure submodule of Y , then compare the generic
orders of S, T .]

Exercise 30.17 (Generic Sylow Theorem for GLn ) Let W = Sn in its

permutation representation in GLn (Q), d ∈ N, and ζd ∈ C a primitive dth
root of unity. Show the following:

(a) All ζd -eigenspaces of maximal possible dimension of elements g ∈ Sn

are conjugate.
(b) The ζd -eigenspace of any w ∈ Sn is contained in a maximal such.
(c) The stabilizer in W of a ζd -eigenspace as in (a) is isomorphic to Zd Sa ×
Sb , where n = ad + b with 0 ≤ b < d and Zd  Sa denotes the wreath
product of the cyclic group of order d with the symmetric group Sa .

Exercise 30.18 Let G be connected reductive with Steinberg endomor-

phism F : G → G, and S a Sylow d-torus of G. Then NGF (S) controls
GF -fusion in CGF (S), that is, if s, t ∈ CGF (S) are GF -conjugate, they are
conjugate by an element of NGF (S).
[Hint: Apply the Sylow Theorem 25.11 to the connected component of C/Ru (C),
where C is the centralizer of an element in CG (S). See [56, Prop. 5.11].]

Exercise 30.19 Let G be connected reductive with Steinberg endomor-

phism F : G → G. Show that the centralizers of 1-tori in G are precisely the
F -stable Levi subgroups contained in F -stable parabolic subgroups of G.
 Exercises for Part III 267

Exercise 30.20 Show that the general unitary group GUn (q) has exactly
two orbits on the lines in its natural representation on Fnq2 , and determine
the stabilizers.
[Hint: Theorem 21.11 does not apply, since here the F -structures on GLn and
on V = kn are not compatible. You should find the groups constructed in Exam-
ple 26.3.]

Exercise 30.21 Let G be connected reductive with Steinberg endomor-

phism F , L ≤ G an F -stable Levi subgroup of some parabolic subgroup of
G, with Weyl group WL .
(a) The GF -classes of F -stable G-conjugates of L are in bijection with φ-
classes of NW (WL )/WL .
(b) Write Lw for a Levi subgroup corresponding to the φ-class of w under
the map in (a). Then NGF (Lw )/LF ∼
w = NW (WL wφ)/WL .

Exercise 30.22 Let G be connected reductive with Steinberg endomor-

phism F : G → G, T ≤ G an F -stable maximal torus of G, Z := Z(G)◦ ,
G := [G, G] and S := T ∩ G , an F -stable maximal torus of G . Show that
|T F | = |Z F | · |S F |.
[Hint: We have T = SZ. Show that the intersection of the kernels of the (surjec-
tive) restriction morphisms from X(T ) to X(S) and to X(Z) is trivial. Conclude
that X(T ) ⊗ R ∼ = (X(S) ⊗ R) ⊕ (X(Z) ⊗ R) as F -modules.]

Exercise 30.23 Assume n ≥ 2, and let ρ : Sp2n (q) → GLd be a represen-

tation of Sp2n (q) with central kernel over a field k of characteristic r = p,
where q = pf is odd. Then d ≥ q n−1 (q − 1)/2.
[Hint: Let P ≤ G = Sp2n (q) be the maximal parabolic subgroup with Levi com-
plement of type CSp2n−2 (q), and consider its action on the faithful characters of
the unipotent radical U of P .]

Exercise 30.24 Let 1 = G be semisimple of simply connected type, F :

G → G a Steinberg endomorphism. Show that GF contains a subgroup
SL2 (p), where p = char(k), unless for every simple component of G there is
a power of F stabilizing it and which is very twisted. (The claim holds more
generally unless GF is a direct product of groups 2 B2 .)
[Hint: Check that there is an F -stable completely disconnected subset of the
Dynkin diagram of G, unless G is a product of groups of type A2 . In the first
case apply Proposition 12.14 and Exercise 30.2. For type A2 , use the computations
in Example 23.10.]
 Appendix A
Root systems

The structure theory for linear algebraic groups shows that many questions
on these groups can be translated into questions on the associated root
system, which are then of a purely combinatorial nature.
In this appendix we collect some basic results on root systems as they are
pertinent to our study of algebraic groups and finite groups of Lie type. Most
of these are well-known (see for example [9, §VI] or [33, Chap. III]), except
possibly for the discussion of maximal subsystems in Sections B.3 and B.4.
We do not present the proof of the classification of indecomposable root
systems, since it is not used here and is very well-documented in the literature
(see for example [9, VI, §4] or [33, §11]), nor the properties of invariant rings
of finite groups generated by reflections (see [9, V, §5] or [47]).

A.1 Bases and positive systems

Recall that an element s ∈ GL(V ), where V is a finite-dimensional vector
space, is called a (linear) reflection along α ∈ V , if α is an eigenvector of s
with eigenvalue −1, and s fixes a hyperplane of V pointwise.
The action of reflections in a Euclidean space E can be described by an
easy formula:

Proposition A.1 Assume that s ∈ GL(E) is a reflection stabilizing the

positive definite bilinear form ( , ) on E. Then for any eigenvector α for the
non-trivial eigenvalue of s we have
(v, α)
s.v = v − 2 α.
(α, α)
Proof Let H ⊂ E denote the eigenspace for the eigenvalue 1, the fixed
 A.1 Bases and positive systems 269

space of s. Then for v ∈ H we have

(v, α) = (s.v, s.α) = (v, −α),
(v,α)
so (v, α) = 0 for all v ∈ H. Thus the linear map E → E, v → v − 2 (α,α) α,
agrees with s on H as well as on α, so on E = H ⊕ αR .
Definition A.2 A subset Φ of a finite-dimensional real vector space E is
called an (abstract) root system in E if the following properties are satisfied:
(R1) Φ is finite, 0 ∈
/ Φ, ΦR = E;
(R2) if c ∈ R is such that α, cα ∈ Φ, then c = ±1;
(R3) for each α ∈ Φ there exists a reflection sα ∈ GL(E) along α stabilizing
Φ;
(R4) (crystallographic condition) for α, β ∈ Φ, sα .β − β is an integral mul-
tiple of α.
The group W = W (Φ) := sα | α ∈ Φ is called the Weyl group of Φ. The
dimension of E is called the rank of Φ.
Let Φ be a root system in E. Since Φ is finite, generates E and is stabilized
by W , the Weyl group of an abstract root system is always finite. Thus, it
stabilizes a positive definite W -invariant symmetric bilinear form ( , ) on
E, which is unique up to non-zero scalars on each irreducible W -submodule
of E. We’ll always assume such a form to have been chosen so that E is
Euclidean and we can speak of lengths of vectors and angles between them.
Using the non-degenerate bilinear form ( , ) we may identify E with
its dual E ∗ = Hom(E, R) by v → (E → R, u → v(u) := (u, v)). For
α ∈ Φ define the corresponding coroot α∨ := 2α/(α, α). Then the formula
in Proposition A.1 for an orthogonal reflection s along α can be written as
s.v = v − (v, α∨ )α = v − (v, α)α∨ = v − α∨ (v)α.
Definition A.3 Let Φ be a root system. The set Φ∨ := {α∨ | α ∈ Φ} is
called the dual root system of Φ.
It is easily seen that Φ∨ is indeed a root system in E, see Exercise A.1.
Lemma A.4 Let Φ be a root system with Weyl group W . Then for all
α ∈ Φ, w ∈ W we have
wsα w−1 = sw.α .
Proof By Proposition A.1 we have
% (w−1 .v, α) & (v, w.α)
wsα w−1 .v = w. w−1 .v − 2 α =v−2 w.α = sw.α .v
(α, α) (w.α, w.α)
270 Root systems

for all v ∈ E.

Given a total ordering “>” on E compatible with addition and scalar

multiplication by positive real numbers we say that v ∈ E is positive if
v > 0. Clearly such orderings exist: choose a basis B of E and take the
lexicographical ordering on the coefficient vectors with respect to this basis.

Definition A.5 A subset Φ+ ⊆ Φ consisting of all positive roots in Φ with

respect to some total ordering on E as above is called a system of positive
roots or positive system of Φ. A subset ∆ ⊂ Φ is called a base of Φ if it is a

vector space basis of E and any β ∈ Φ is a linear combination β = α∈∆ cα α
with either all cα ≥ 0 or all cα ≤ 0.

Note that if Φ+ ⊆ Φ is a positive system, then Φ = Φ+  −Φ+ by (R2)

and (R3). We’ll then also write Φ− := −Φ+ .

Lemma A.6 Let Φ+ ⊆ Φ be a positive system and ∆ ⊆ Φ+ minimal with

the property that any element of Φ+ is a non-negative linear combination of
elements in ∆. Then (α, β) ≤ 0 for all α, β ∈ ∆, α = β.

Proof Suppose that this fails for some α, β. By Proposition A.1 we then
have sα .β = β − cα for some c > 0, and by (R3) this has to lie in Φ,
so either sα .β or −sα .β lies in Φ+ . If sα .β ∈ Φ+ then we can write it as

sα .β = γ∈∆ cγ γ with cγ ≥ 0. Now, if cβ ≥ 1 we get
"
0 = sα .β − (β − cα) = cα + (cβ − 1)β + cγ γ
γ =β

which is absurd since all summands on the right are non-negative, and the
first is strictly positive. On the other hand, if cβ < 1 then the above equation
implies that (1 − cβ )β is a non-negative linear combination of ∆ \ {β} in
contradiction to the minimal choice of ∆. Thus −sα .β ∈ Φ+ , but then a
similar argument leads to a contradiction.

Proposition A.7 Let Φ be an abstract root system. Then every positive

system contains a unique base. Conversely, any base is contained in a unique
positive system.

Proof Let Φ+ be a positive system in Φ and choose ∆ ⊆ Φ+ minimal with

the property that any element of Φ+ is a non-negative linear combination
of elements in ∆. (Clearly such a subset exists.) It suffices to show that ∆

is linearly independent. For this let α∈∆ cα α = 0. With ∆ := {α ∈ ∆ |
 
cα ≥ 0} ⊆ ∆, we can rewrite this as α∈∆ cα α = − β ∈∆ /  cβ β =: v with
 A.1 Bases and positive systems 271

non-negative coefficients on both sides. Then

" "
0 ≤ (v, v) = − cα cβ (α, β) ≤ 0
α∈∆ β ∈∆
/ 

by Lemma A.6, so v = 0 and hence, since cα α ≥ 0 for all α, all cα = 0,

showing linear independence of ∆. Clearly ∆ is characterized as the set of
roots in Φ+ which are not expressible as a non-negative linear combination
of more than one element of Φ+ , so it is unique.
Conversely, given a base ∆, choose a total ordering on ∆ and define a

total ordering on E by letting α∈∆ cα α be positive if and only if the first
non-zero coefficient cα (with respect to the ordering on ∆) is positive. Then
the set of positive elements in Φ is a positive system containing ∆. Again,
uniqueness is obvious since any positive system containing the base ∆ must
consist precisely of those roots which are non-negative linear combinations
over ∆.

In particular this shows that bases of root systems exist, and we can speak
of the base in a positive system, or the positive system containing a given
base. The elements of ∆ are then also called the simple roots with respect
to Φ+ .
We next aim to show that the Weyl group acts transitively on the set of
bases. For this we need:

Lemma A.8 Let ∆ be a base of the root system Φ, and α ∈ ∆. Then α is

the only positive root made negative by sα , that is, sα (Φ+ \ {α}) ⊆ Φ+ .

Proof By (R3) we have sα (Φ+ ) ⊂ Φ. Any β ∈ Φ+ can be written as β =

 ∨
α∈∆ cα α with cα ≥ 0. Then all coefficients of sα .β = β − (β, α )α are still
non-negative, except possibly the coefficient at α. By the defining property
of a base, if sα .β ∈ Φ− , then necessarily cγ = 0 for γ = α, that is, β is a
(positive) multiple of α. By (R2) this implies that β = α.

Proposition A.9 Any two bases (respectively positive systems) of a root

system Φ are conjugate under the Weyl group W of Φ.

Proof By the existence and uniqueness assertions in Proposition A.7 it is

enough to show this for positive systems. So let Φ+ +
1 , Φ2 be two positive
systems of Φ. We induct on n := |Φ1 ∩ −Φ2 |. If n = 0 there is nothing to
+ +

show. Now assume that n > 0. Then the base ∆ of Φ+1 cannot be contained in
Φ2 , say α ∈ ∆ ∩ −Φ2 . Then Lemma A.8 shows that |sα .Φ+
+ +
1 ∩ −Φ2 | = n − 1.
+

The result now follows from the inductive hypothesis applied to sα .Φ+ +
1 , Φ2 .
272 Root systems

Definition A.10 Let ∆ be a base of the root system Φ. The height of a

 
root β = α∈∆ cα α ∈ Φ (with respect to ∆) is defined as ht(β) := α∈∆ cα .

Proposition A.11 If ∆ is a base, then W = sα | α ∈ ∆. Furthermore,

for every α ∈ Φ there is w ∈ W such that w.α ∈ ∆.

Proof Let W  := sα | α ∈ ∆, a subgroup of W . For any β ∈ Φ+ , W  .β∩Φ+

is a non-empty set of positive roots. Let γ ∈ W  .β ∩ Φ+ of smallest height,

say γ = α∈∆ cα α with cα ≥ 0. Then
"
0 < (γ, γ) = cα (γ, α),
α∈∆

whence (γ, α) > 0 for some α. We claim that γ = α. Otherwise sα .γ is

positive by Lemma A.8, but on the other hand it is obtained from γ by
subtracting a positive multiple of α, so it is of strictly smaller height than γ.
Since sα .γ ∈ W  .β ∩ Φ+ this contradicts the choice of γ. So we have shown
that γ = α ∈ ∆.
In particular the W  -orbit of any β ∈ Φ+ contains a simple root, so Φ+ ⊆
W  .∆. Since sα .α = −α we conclude that in fact also −Φ+ ⊆ W  .(−∆) ⊆
W  .∆. Now take any generator sβ of W , where β ∈ Φ. By the above we
have β = w.α for some w ∈ W  , α ∈ ∆, so sβ = sw.α = wsα w−1 ∈ W  by
Lemma A.4, whence W = W  .

The reflections in S := {sα | α ∈ ∆} along simple roots of Φ are called

the simple reflections of W (with respect to ∆). As a direct consequence,
we obtain the following integrality result, which for the first time uses ax-
iom (R4).

Corollary A.12 Let ∆ be a base of Φ. Then every α ∈ Φ is an integral

linear combination of the roots in ∆.

Proof The result is trivially true for the roots in ∆. Now let β ∈ Φ and
(β,α)
α ∈ ∆. If the claim holds for β, then so it holds as well for sα .β = β −2 (α,α) α
by (R4). Since any β ∈ Φ is of the form w.γ for some w ∈ W and γ ∈ ∆, and
w can be written as a product of reflections sα for α ∈ ∆ by Proposition A.11,
the claim follows by induction.

A.2 Decomposition of root systems

Definition A.13 A non-empty root system Φ with base ∆ is called de-
composable if there exists a non-trivial partition ∆ = ∆1  ∆2 such that
 A.2 Decomposition of root systems 273

(α1 , α2 ) = 0 for all αi ∈ ∆i , i = 1, 2; if no such decomposition exists then Φ

is said to be indecomposable.

Note that by Proposition A.1 this notion does not depend on the choice
of a W -invariant scalar product on E.

Proposition A.14 Let Φ be decomposable with corresponding partition of

a base ∆ = ∆1  ∆2 , and set Ei := R∆i , Φi := Φ ∩ Ei , i = 1, 2. Then we
have:

(a) Φi is a root system in Ei with base ∆i , for i = 1, 2,

(b) Φ = Φ1  Φ2 , and (α1 , α2 ) = 0 for all α1 ∈ Φ1 , α2 ∈ Φ2 ,
(c) W (Φ) = W (Φ1 ) × W (Φ2 ) by letting W (Φi ) act trivial on E3−i .

Proof By Proposition A.1 for α ∈ ∆1 , β ∈ ∆2 we have sβ .α = α, so

sα , sβ commute by Lemma A.4. Since W is generated by the sα , α ∈ ∆,
by Proposition A.11, any w ∈ W can be written as a commuting product
w = w1 w2 with wi ∈ Wi , where Wi := sα | α ∈ ∆i , i = 1, 2. Now
any w ∈ Wi fixes all elements of ∆3−i , and again by Proposition A.1 also
Wi .∆i ⊂ Ei . The intersection W1 ∩ W2 thus fixes the basis ∆ of E, hence is
trivial, so W = W1 × W2 .
Furthermore, by Proposition A.11 we have
Φ = W.∆ = (W1 × W2 ).(∆1 ∪ ∆2 ) = W1 .∆1 ∪ W2 .∆2 ⊂ E1 ∪ E2 .
Since ∆ is a basis of E, E = E1 ⊕ E2 , so Φi = Φ ∩ Ei = Wi .∆i , whence (b).
Clearly axioms (R1)–(R4) are satisfied for Φi ⊂ Ei , ∆i is a base of Φi , and
thus Wi = W (Φi ) by Proposition A.11, showing (a) and (c).
Induction then gives:

Corollary A.15 Any root system Φ can be decomposed uniquely (up to

reordering) into a disjoint orthogonal union Φ1  . . .  Φr of indecomposable
root systems Φi , and then W (Φ) ∼= W (Φ1 ) × · · · × W (Φr ).
The Φi in Corollary A.15 are called the indecomposable components of the
root system Φ.

Proposition A.16 A root system Φ ⊂ E is indecomposable if and only if

its Weyl group W acts irreducibly on E.

Proof If Φ is decomposable then W is reducible by Proposition A.14(c).

Conversely, if E1 < E is a non-trivial W -invariant subspace then we obtain
a W -invariant orthogonal decomposition E = E1 ⊕ E2 , with E2 = E1⊥ . We
claim that any α ∈ Φ lies in either E1 or E2 . Indeed, as sα acts semisimply on
274 Root systems

E there is a basis of E1 consisting of eigenvectors of sα . If the eigenvalue −1

occurs then, since the −1-eigenspace of sα is spanned by α, we have α ∈ E1 .
Otherwise, E1 ≤ ker(sα −1). The same reasoning applies to E2 . Since clearly
we can’t have both E1 , E2 ≤ ker(sα − 1), α ∈ Ei for some i. So ∆ ⊂ E1 ∪ E2
gives an orthogonal (and non-trivial) decomposition of ∆.

Irreducible linear groups can preserve at most one scalar product up to

scalars, hence for indecomposable root systems the W -invariant scalar prod-
uct on E is unique up to scalars.
For roots α, β ∈ Φ let’s write
(α, β)
n(α, β) := (α, β ∨ ) = 2 .
(β, β)
By the crystallographic condition (R4) and our Proposition A.1 we have
n(α, β) ∈ Z, and by elementary geometry,
4(α, β)2
n(α, β)n(β, α) = = 4 cos2 ∠(α, β) ≤ 4,
(α, α)(β, β)
where ∠(α, β) denotes the angle between α and β. Here, n(α, β)n(β, α) = 4
can only occur if cos ∠(α, β) = ±1, that is, when α, β are proportional. But
in that case β = ±α by (R2), so n(α, β) = n(β, α) = ±2. Thus, for β = ±α
we always have |n(α, β)| ≤ 3.
The various possibilities can now easily be enumerated; they are given in
Table A.1, up to interchanging α, β and up to replacing β by −β. We have
also listed the order o(sα sβ ) of the product sα sβ .

Table A.1 Relation between two roots

n(α, β) n(β, α) ∠(α, β) lengths o(sα sβ )

0 0 π/2 any 2
1 1 π/3 (α, α) = (β, β) 3
1 2 π/4 2(α, α) = (β, β) 4
1 3 π/6 3(α, α) = (β, β) 6
2 2 0 α=β 1

From this, the various two-dimensional root systems can be classified. If

∆ = {α, β} denotes a base, then (α, β) ≤ 0 by Lemma A.6 and thus either
α, β are perpendicular, or n(α, β) = −1 with n(β, α) ∈ {−1, −2, −3} and
∠(α, β) as in Table A.1. Application of the corresponding reflections gives
the possibilities depicted in Table A.2.
 A.2 Decomposition of root systems 275

Proposition A.17 Any two-dimensional root system is of one of the four

types in Table A.2, except that the two root lengths in type A1 × A1 may be
different.

Table A.2 The two-dimensional root systems

β
A1 × A1 A2 β α+β

α α

3α + 2β
B2 G2

α+β α+β 2α + β 3α + β
β 2α + β β

α α

Corollary A.18 Let Φ be indecomposable. Then:

(a) There are at most two different root lengths in Φ.
(b) All roots of Φ of the same length are conjugate under W .
Proof As W acts irreducibly on E by Proposition A.16, the W -orbit of any
α ∈ Φ generates E. Thus, for any α, β ∈ Φ there exists a W -conjugate β  of β
(hence of the same length) with (α, β  ) = 0. Now by the above considerations
(β  , β  )/(α, α) ∈ {1, 2, 3, 1/2, 1/3}, so we may rescale the scalar product such
that (α, α) = 1 for some α ∈ Φ, and all squared root lengths are among 1, 2, 3.
By what we saw before, neither 2/3 nor 3/2 is a quotient of squared root
lengths, so not both squared root lengths 2, 3 may occur.
For the second assertion, see Exercise A.3.
276 Root systems

The roots α ∈ Φ of maximal length are called long roots. If there are two
distinct root lengths, then the shorter ones are called short roots.
We now introduce a convenient description of root systems. By Proposi-
tion A.11 any root system can be recovered from a base. The base is encoded
in terms of a directed graph which we now describe. Its nodes are in bijec-
tion with the elements of a base ∆, and two different nodes corresponding
to α, β ∈ ∆ are joined by an edge of multiplicity n(α, β)n(β, α). The result-
ing graph is called the Coxeter diagram associated to Φ or W . It does not
determine Φ uniquely. So in addition, whenever α, β are of different lengths
and joined by at least one edge, then we put an arrow on this edge, pointing
towards the shorter of the two. This directed graph is called the Dynkin di-
agram of Φ. It can be shown that the base can essentially be recovered from
its Dynkin diagram (that is, up to changing lengths in different connected
components of the diagram), see [33, 11.1].
The indecomposable root systems can be classified, see Theorem 9.6 and
Table 9.1 for their Dynkin diagrams. We’ll not repeat this proof here, since
it is well documented in the literature, see for example [9, VI, §4] or [33,
Thm. 11.4].

A.3 The length function

An important tool in the study of a root system Φ and its Weyl group W is
played by the length function.

Definition A.19 Let ∆ be a base of Φ. The length (w) of w ∈ W (with

respect to ∆) is the minimal integer l such that w = sα1 · · · sαl with αi ∈ ∆.
Any such expression of w of minimal length is called reduced .

It is clear that (wsα ) = (w) ± 1 for w ∈ W , α ∈ ∆, and also (w) =

(w−1 ) since all sα are involutions.
We define a second function on W by

n(w) := |{β ∈ Φ+ | w.β ∈ Φ− }|,

the number of positive roots made negative by w ∈ W . It will turn out that
this coincides with the length (w) introduced above. For this we need:

Lemma A.20 Let α ∈ ∆ and w ∈ W . Then


n(w) + 1 if w.α ∈ Φ+ ,
n(wsα ) =
n(w) − 1 if w.α ∈ Φ− .
 A.3 The length function 277

Proof Write N (w) := (w.Φ+ ) ∩ Φ− , so that n(w) = |N (w)|. Then by

Lemma A.8 we have
% &
N (wsα ) = (wsα .Φ+ ) ∩ Φ− = w. Φ+ \ {α}  {−α} ∩ Φ− .

Thus, if w.α ∈ Φ+ then

N (wsα ) = (w.(Φ+ \ {α})  {−w.α}) ∩ Φ− = N (w)  {−w.α},

whence the result. On the other hand, if w.α ∈ Φ− then this shows that
N (wsα ) = N (w) \ {w.α}.

Proposition A.21 We have (w) = n(w) for all w ∈ W .

Proof Let w ∈ W . An easy induction, starting with w = 1 and using

Lemma A.20 shows that n(w) ≤ (w) for all w ∈ W . Assume that n(w) <
(w) and choose a reduced expression w = s1 · · · sl with si = sαi for αi ∈ ∆
and l = (w). According to Lemma A.20 there is j < l with n(s1 · · · sj+1 ) <
n(s1 · · · sj ) and s1 · · · sj .αj+1 ∈ Φ− . Since αj+1 ∈ Φ+ there is i ≤ j such that
si · · · sj .αj+1 ∈ Φ− and si+1 · · · sj .αj+1 ∈ Φ+ , so that si+1 · · · sj .αj+1 = αi
by Lemma A.8. With u := si+1 · · · sj this gives

si+1 · · · sj · sj+1 · sj · · · si+1 = usj+1 u−1 = su.αj+1 = sαi = si

by Lemma A.4, so si+1 · · · sj+1 = si · · · sj . Substituting the right-hand side

for the left in w and using s2i = 1 we find a shorter expression for w, a
contradiction. Thus n(w) = (w) as claimed.

Theorem A.22 W acts simply transitively on the set of bases (respectively

positive systems) in Φ, that is, the only element of W fixing a given base
setwise is the identity.

Proof The transitivity was already shown in Proposition A.9. Now if w ∈ W

stabilizes the base ∆, then clearly it fixes Φ+ , whence (w) = n(w) = 0 by
Proposition A.21.

Corollary A.23 Let ∆ be a base. Then there is a unique element w0 of W

with w0 (∆) = −∆. It is the element of maximal length in W , and w02 = 1.

Proof If ∆ is a base of Φ, then clearly so is −∆, so by Theorem A.22 there

exists a unique w0 ∈ W with w0 (∆) = −∆. Its length (w0 ) = n(w0 ) = |Φ+ |
is maximal by Proposition A.21. Since w02 (∆) = ∆, the last claim follows.

Definition A.24 The element w0 in Corollary A.23 is called the longest

element of W (with respect to ∆).
278 Root systems

A.4 Parabolic subgroups

We next introduce a natural collection of subsystems of a root system Φ
and corresponding subgroups of its Weyl group W , where a subset Ψ ⊂ Φ is
called a subsystem if it is a root system in RΨ.
For this, let ∆ denote a base and S = {sα | α ∈ ∆} the set of simple
reflections of W . For a subset I ⊆ S, WI := s ∈ I is called a standard
parabolic subgroup of W . A parabolic subgroup of W is any conjugate of a
standard parabolic subgroup. We let ∆I := {α ∈ ∆ | sα ∈ I} and

ΦI := Φ ∩ Z∆I

be the corresponding parabolic subsystem of roots.

Proposition A.25 Let I ⊆ S. Then ΦI is a root system in RΦI with base

∆I and Weyl group WI .

Proof Conditions (R1)–(R4) are trivially satisfied, as ΦI is a subset of Φ

with ΦI = −ΦI (for (R3) use the explicit formula for sα in Proposition A.1).
Now any root in ΦI is a Z-linear combination of roots in ∆I . As ∆I is
contained in a base, the coefficients are either all positive or all negative. Thus
∆I is a base of ΦI . By Proposition A.11, the Weyl group of ΦI is generated
by the corresponding simple reflections, that is, by I, as claimed.

We have the following generalization of Lemma A.8 (which is the case

|I| = 1):

Corollary A.26 Let I ⊆ S. Then we have:

(a) The length function I (with respect to ∆i ) on WI is the restriction of

the length function  (with respect to ∆) on W .
(b) Let wI ∈ WI be the longest element of WI . Then
−
wI (Φ+
I ) = ΦI and wI (Φ+ \ Φ+
I ) = Φ \ ΦI .
+ +

Proof Let w ∈ WI . Then certainly (w) ≤ I (w) since I ⊆ S. On the other

hand we have I (w) = nI (w) ≤ n(w) = (w) by Proposition A.21.
The first part of (b) holds by the definition of the longest element and
Proposition A.25. Since nI (wI ) = n(wI ) by (a), wI cannot make any further
roots negative, so wI (Φ+ \ Φ+
I ) ⊆ Φ . The claim now follows as wI = 1.
+ 2

We next study the action of W on E in some more detail. The set

C := {v ∈ E | (v, α) > 0 for all α ∈ ∆}

 A.4 Parabolic subgroups 279

is called the fundamental chamber of W (with respect to ∆). We write

C̄ = {v ∈ E | (v, α) ≥ 0 for all α ∈ ∆}

for the closure of C in the natural topology of E.

Theorem A.27 The closure C̄ of C is a fundamental domain for the

action of W on E.

Proof Let v ∈ E and choose w ∈ W such that |{β ∈ Φ+ | (w.v, β) < 0}| is
minimal. We claim that then w.v ∈ C̄. Otherwise, (w.v, α) < 0 for some α ∈
∆, but then (sα w.v, α) = (w.v, sα .α) = −(w.v, α) > 0, and by Lemma A.8
this contradicts our choice of w. Hence the W -orbit of any v ∈ E has a
representative in C̄.
To finish the proof it is enough to show that when v1 , v2 ∈ C̄ with w.v1 =
v2 for some w ∈ W then v1 = v2 . We argue by induction on (w), the case
(w) = 0 being trivial. If (w) > 0 then by Proposition A.21 there is some
simple root α ∈ ∆ made negative by w. By Lemma A.20 this means that
n(wsα ) = n(w) − 1. Now v1 , v2 ∈ C̄, so

0 ≤ (v1 , α) = (w.v1 , w.α) = (v2 , w.α) ≤ 0,

whence (v1 , α) = 0 and so sα .v1 = v1 by Proposition A.1. Thus wsα .v1 = v2 .

Since (wsα ) = n(wsα ) < (w) we conclude that v1 = v2 by the induction
hypothesis.

The above proof actually showed a bit more: choosing v = v1 = v2 we saw

that any w ∈ W fixing v ∈ C̄ is a product of simple reflections which fix v:

Corollary A.28 Let v ∈ C̄. Then CW (v) = WI with I = {s ∈ S | s.v = v},

a standard parabolic subgroup of W .

More generally this shows the following characterization of parabolic sub-

groups:

Corollary A.29 Let M ⊆ E. Then the pointwise stabilizer CW (M ) of M

in W is a parabolic subgroup. Conversely, any parabolic subgroup H ≤ W is
the centralizer of its fixed space, that is, H = CW (V H ).

Proof Clearly, the pointwise stabilizer of M is the same as that of its linear
span, so we may replace M by a basis of the latter. Now let v ∈ M . We may
assume that v ∈ C̄ by Theorem A.27 and then H = CW (v) is a parabolic
subgroup by the above observation. Now, by induction on |M |, the centralizer
in H of M \ {v} is a parabolic subgroup of H, hence of W , as claimed.
For the converse, note that for I ⊆ S the fixed space M := V WI of the
280 Root systems

corresponding parabolic subgroup WI satisfies dim M = dim V − |I|, since

W is generated by |S| = dim V reflections. Then WI ≤ CW (M ) = WJ for
some I ⊆ J ⊆ S by the first part, and comparison of dimensions shows that
I = J, so CW (M ) = WI .

Recall the longest element w0 of W from Definition A.24.

Proposition A.30 Let W be the Weyl group of a root system Φ of rank l.

Then −id ∈ W if and only if Φ contains a subset of l mutually orthogonal
roots.

Proof First note that if −id ∈ W then necessarily −id = w0 by Corol-

lary A.23. Now if α1 , . . . , αl ∈ Φ are mutually orthogonal, then the cor-
responding reflections s1 , . . . , sl commute. Thus by Proposition A.1 with
w := s1 · · · sl we have w.αi = −αi for i = 1, . . . , l, whence w = −id ∈ W .
Conversely, if w0 = −id let α ∈ Φ with corresponding reflection sα . The
stabilizer W1 of α in W is a parabolic subgroup of W by Corollary A.29. Any
reflection in W1 fixes α, so the root system Ψ of W1 (see Proposition A.25)
lies in E1 = α⊥ , and sα w0 ∈ W1 acts as −id on E1 . The claim now follows
by induction on the rank applied to Ψ ⊂ E1 .

Let I ⊆ S; for β ∈ Φ+ with

" "
β= cα α + dγ γ
α∈∆I γ∈∆\∆I

we call (dγ )γ∈∆\∆I the shape of β. Note that WI preserves the set of roots
of a given shape. Let’s note the following property:

Lemma A.31 Let Φ be a root system with base ∆, and I ⊆ S. Then:

(a) Any two roots in Φ+ \ ΦI of the same shape and the same length are
conjugate under WI .
(b) There is a unique root of minimal height of a given shape.

Proof Let β, γ ∈ Φ+ \ ΦI have the same shape. We may assume that β = γ

are of minimal height in their respective WI -orbits. Then (β, α) ≤ 0, (γ, α) ≤
0 for all α ∈ ∆I by Proposition A.1. As β ∈ Φ+ \Φ+ I the set {β}∪∆I consists
of linearly independent positive roots. Since β, γ have the same shape, γ lies
in the span of this set.
If (β, γ) ≤ 0, then {β, γ} ∪ ∆I would be a set of positive roots all of
whose pairwise angles are obtuse, which would force them to be linearly
independent. Thus (β, γ) > 0. Now first assume that β, γ are of the same
length. Then n(β, γ) = 1 (see Table A.1) and so sγ (β) = β − γ =: δ ∈ ΦI , a
 Exercises 281

root of the same length as β and γ. But then sδ (β) = β−n(β, δ)δ = β−δ = γ,
with sδ ∈ WI , proving (a). In general, either n(β, γ) = 1 or n(γ, β) = 1,
whence β −γ ∈ Φ, a contradiction if β, γ have the same (minimal) height.

Exercises
A.1 Let Φ be a root system with base ∆. Show that Φ∨ is a root system
with base ∆∨ := {α∨ | α ∈ ∆}.
A.2 Let Φ be a root system with dual root system Φ∨ . Then the Weyl
groups of Φ and of Φ∨ are isomorphic via sα → sα∨ .
A.3 Let Φ ⊂ E be an indecomposable root system with Weyl group W and
fix a W -invariant scalar product on E. Show the following: There are
at most two W -orbits of roots in Φ, and any two roots of the same
length are conjugate.
[Hint: First argue that given α, β ∈ Φ we can always find w ∈ W such
that α, w.β are not orthogonal, then apply Proposition A.17 and Corol-
lary A.18(a).]
A.4 Let Φ be an indecomposable root system with two different root lengths
and let p = (α, α)/(β, β) for α, β ∈ Φ with α long and β short.

pα∨ α long,
(a) The map α → extends to a homothety.
α∨ α short,

(b) A root γ = α∈∆ cα α is long if and only if p|cα for all short α ∈ ∆.
A.5 Let Φ be a root system, α, β ∈ Φ with β = −α. Show that there exists
a base ∆ of Φ with α ∈ ∆ such that β ∈ Φ+ (with respect to ∆).
A.6 Let Φ be a root system in E with Weyl group W . Show that the only
reflections in W are of the form sα for α ∈ Φ.
[Hint: You may want to use Corollary A.29.]
A.7 Let Φ be a root system with base ∆, I  S a subset of the correspond-
ing set of simple reflections, and S a shape with respect to the subset
I. Then there exists a unique root of maximal height of shape S.
[Hint: Use Lemma A.31, and show that if β is of maximal height and of
shape S, then wI (β) is of minimal height and of shape S, where wI is the
longest element of the parabolic subgroup WI .]
 Appendix B
Subsystems

In this appendix we collect various results on subsystems of an indecompos-

able root system Φ, including a classification of maximal subsystems. This
turns out to be closely related to the so-called highest root of Φ which we
introduce first. We follow mainly [74, §1] and [9, Ex. VI.4.4].

B.1 The highest root

Throughout Φ ⊂ E denotes an indecomposable root system with Weyl group
W , ∆ ⊂ Φ a base with positive system Φ+ , and ( , ) a fixed W -invariant
scalar product on E.
Recall from Section A.2 that for α, β ∈ Φ,
(α, β)
n(α, β) = (α, β ∨ ) = 2
(β, β)
can only take values 0, ±1, ±2, ±3.

Lemma B.1 Let α, β ∈ Φ with α = ±β.

(a) If (α, β) > 0 then α − β ∈ Φ;

(b) if (α, β) < 0 then α + β ∈ Φ.

Proof By assumption α, β are neither perpendicular nor proportional, hence

at least one of n(α, β), n(β, α) is equal to ±1 by Table A.1
Now if (α, β) > 0 then moreover both terms are positive. If n(α, β) = 1
then Proposition A.1 shows that sβ .α = α − β ∈ Φ, while sα .β = β − α ∈ Φ
when n(β, α) = 1. By (R2) this proves (a). The second part is an immediate
consequence by replacing β by −β in (a).
 B.1 The highest root 283

Corollary B.2 Let β = α∈∆ cα α ∈ Φ+ be a positive root. Then for all
α ∈ ∆ and all 1 ≤ j ≤ cα there exists a root γ ∈ Φ+ whose coefficient at α
equals j.

Proof We have 0 < (β, β) = α∈∆ cα (β, α), so there is α ∈ ∆ with cα > 0
and (β, α) > 0. If β ∈ ∆ there is nothing to prove. Else, β = α, so β1 :=
β − α ∈ Φ+ by the preceding result. The coefficient vector of β1 differs from
that of β only at α, by 1. A straightforward induction now completes the
proof.

Proposition B.3 Let α, β ∈ Φ with α = ±β. Then:

(a) The set of integers I := {i | α + iβ ∈ Φ} forms an interval [−q, p]

containing 0.
(b) We have p − q = −n(β, α) and thus sα .(β + pα) = β − qα.

Proof Let q := − min I, p := max I. Since 0 ∈ I we have p, q ≥ 0. Assume

that I is not the interval [−q, p]. Then there exist r, t ∈ I with t > r + 1 and
r+i ∈ / I for 1 ≤ i ≤ t − r − 1. As (β + rα) + α, (β + tα) − α ∈
/ Φ, Lemma B.1
gives

0 ≤ (α, β + rα) < (α, β + rα) + (α, (t − r)α) = (α, β + tα) ≤ 0,

which is not possible. This proves (a).

We have sα .(β + iα) = β − n(β, α)α − iα = β + (−i − n(β, α))α, so
−i − n(β, α) ∈ I, and i → −i − n(β, α) is an order reversing involution on
I. Thus, for i = p we get −p − n(β, α) = −q, showing (b).

The choice of a base ∆ of Φ induces a partial order relation on E and hence

on Φ as follows: we say that µ is smaller than λ if λ − µ is a non-negative
linear combination of positive roots.

Lemma B.4 Let v ∈ E. Then there exists a unique maximal element

v  ∈ W.v with respect to the partial order introduced before, and v  ∈ C̄.

Proof Let v  be a maximal element in the orbit W.v. We claim that v  ∈ C̄.
Indeed, if (v  , α) < 0 for some α ∈ ∆ then sα .v  = v  − 2(v  , α)/(α, α)α >
v  , contradicting the maximality of v  . By Theorem A.27 this implies the
uniqueness of v  .

The maximal element of Φ with respect to this order plays an important

role:

Proposition B.5 Let Φ be an indecomposable root system with base ∆.

Then there exists a unique maximal root α0 ∈ Φ+ in the above partial order,
284 Subsystems

that is, with the following property. Writing α0 = α∈∆ nα α, then for every

root β = α∈∆ cα α ∈ Φ+ we have cα ≤ nα for all α ∈ ∆.

Proof Let α0 = α∈∆ nα α ∈ Φ+ be maximal with respect to the partial
order. Since α0 ∈ Φ+ , all nα are non-negative. Let ∆ ⊆ ∆ be the set
of simple roots α for which nα > 0. Clearly ∆ = ∅. If ∆ = ∆ then by
Lemma A.6 the indecomposability of Φ gives the existence of α1 ∈ ∆ and
α2 ∈ ∆ \ ∆ with (α1 , α2 ) < 0. Then
" "
(α0 , α2 ) = nα (α, α2 ) = nα (α, α2 ) < 0
α∈∆ α∈∆

by Lemma A.6, so α0 + α2 ∈ Φ by Lemma B.1, which is not possible. Thus,

nα > 0 for all α ∈ ∆.

Now let β = α∈∆ cα α ∈ Φ+ be another maximal root. Then Lemma
B.1(b) shows that (β, α) ≥ 0 for all α ∈ ∆, and at least one of them is strictly

positive. We conclude that (β, α0 ) = α∈∆ nα (β, α) > 0. If α0 −β ∈ Φ, then
it’s either positive or negative, both of which contradict the maximality of
α0 , β. But then we must have β = α0 by Lemma B.1(a).

Definition B.6 The root α0 above is called the highest root of Φ with
respect to ∆ (since, clearly it has the largest height (see Definition A.10) of
all roots in Φ).

Corollary B.7 Let Φ be indecomposable with highest root α0 . Then we

have:

(a) (α0 , α0 ) ≥ (α, α) for all α ∈ Φ, that is, α0 is a long root.

(b) If α ∈ Φ+ is different from α0 then n(α, α0 ) ∈ {0, 1}.

Proof For (a) let α ∈ Φ. Since C̄ is a fundamental domain for the action
of W by Theorem A.27 we may assume that α ∈ C̄. Now α0 − α ≥ 0 by
Proposition B.5, so (α0 −α, v) ≥ 0 for all v ∈ C̄. In particular (α0 −α, α0 ) ≥ 0
and (α0 − α, α) ≥ 0, which gives (α0 , α0 ) ≥ (α0 , α) ≥ (α, α).
In (b) we have n(α, α0 ) ∈ {0, ±1} by Table A.1. But since α0 ∈ C̄ this
integer is non-negative.

The coefficients of the highest root in the indecomposable root systems

are given in Table B.1 (see for example [9, pp. 250–275]).
The coefficients of the highest root will come up again in the investigation
of bad primes and torsion primes, see Section B.5.
 B.2 The affine Weyl group 285

Table B.1 Highest roots of indecomposable root systems

1 1 1 1 1 2 2 2
An Bn >
n≥1 n≥2
1
2 2 2 1 1 2 2
Cn < Dn
n≥3 n≥4

1 2 3 2 1 1
E6
2 3 4 2 3 2
2 F4 > G2 <

2 3 4 3 2 1 2 4 6 5 4 3 2
E7 E8

2 3

B.2 The affine Weyl group

We now consider the extension of the Weyl group of a root system by the
group of affine translations in the coroots.
For v ∈ E let tv denote the corresponding affine translation

tv : E → E, u → u + v.

For α ∈ Φ and j ∈ Z we define the affine reflection

sα,j : E → E, sα,j (v) := v − ((v, α) − j) α∨ .

The following is easy to see (see Exercise B.1):

Lemma B.8 The transformation sα,j satisfies sα,j = tjα∨ sα , i.e., it is a

reflection in the hyperplane Hα,j := {v ∈ E | (v, α) = j} along α∨ .

Definition B.9 The affine Weyl group Wa of a root system Φ is the group
of affine transformations of E generated by all sα,j , for α ∈ Φ, j ∈ Z.

Let L := ZΦ∨ , the lattice generated by the coroots. Then we have the
following structure result for Wa :

Proposition B.10 The affine Weyl group is the semidirect product of the
translation subgroup corresponding to L with W .

Proof The Weyl group W normalizes L, and their intersection is trivial, so

286 Subsystems

their product is a semidirect product. By Lemma B.8 all generators of Wa

lie in L.W , but also tα∨ = sα,1 sα ∈ Wa for all α∨ ∈ Φ∨ , so L.W = Wa .
The set
A := {v ∈ E | 0 ≤ (v, α) ≤ 1 for all α ∈ Φ+ }
is called the fundamental alcove. It is easily seen to be a connected compo-
nent of the complement in E of the reflecting hyperplanes ∪α,j Hα,j . There
is another description of this set using the highest root α0 of Φ:
Lemma B.11 Let Φ be indecomposable. Then
A = {v ∈ E | 0 ≤ (v, α) for all α ∈ ∆, (v, α0 ) ≤ 1}.
Proof Write B for the set in the statement. Then clearly A ⊆ B. On the
other hand, if v ∈ B then (v, α) ≥ 0 for all α ∈ Φ+ , so also (v, α0 − α) ≥ 0
by Proposition B.5. Thus (v, α) ≤ (v, α0 ) ≤ 1 and so v ∈ A.
The lemma shows that A is the intersection of the half-spaces defined by
0 ≤ (v, α), α ∈ ∆, and (v, α0 ) ≤ 1. The corresponding hyperplanes are called
the walls of the fundamental alcove A.
Proposition B.12 Let Φ be indecomposable. Every Wa -orbit on E con-
tains a point of A.
Proof Let v ∈ E and u be in the interior of A. The orbit of v under the
translation subgroup L of Wa is clearly discrete in E. Since Wa is a finite
extension of L by Proposition B.10, the Wa -orbit of v is also discrete. So
there exists w ∈ Wa such that w.v has minimal distance to u.
If w.v ∈
/ A there is a wall of A separating u and w.v. Let s ∈ Wa denote
the reflection in that hyperplane. Then w.v, sw.v, u, s.u form a trapezoid.
Now in any trapezoid the length of any diagonal is greater than the lengths
of the two non-parallel sides, so (u − w.v, u − w.v) > (u − sw.v, u − sw.v)
and sw.v has smaller distance to u than w.v. But this contradicts the choice
of w ∈ Wa . Thus w.v ∈ A and the claim is proved.
In fact, A is even a fundamental domain for the action of Wa on E, see
for example [34, Thm. 4.8].

B.3 Closed subsystems

Definition B.13 Let Φ be a root system. A subset Ψ ⊆ Φ is said to be
closed if
 B.3 Closed subsystems 287

(C1) for all α, β ∈ Ψ we have sα .β ∈ Ψ, and

(C2) for α, β ∈ Ψ with α + β ∈ Φ, we have α + β ∈ Ψ.

Proposition B.14 A subset Ψ of a root system Φ is closed if and only

if it satisfies (C2) and it is symmetric, that is, for all α ∈ Ψ we have
Rα ∩ Ψ = {±α}.

Proof Let Ψ ⊆ Φ be symmetric with property (C2). Now for α, β ∈ Ψ,

α = ±β, we have sα .β = β − n(β, α)α. If n(β, α) is negative, then by
Proposition B.3(a) all β + iα for 1 ≤ i ≤ −n(β, α) are contained in Φ, hence
in Ψ by (C2), which shows (C1). The same argument with β replaced by −β
applies when n(β, α) > 0. The converse is clear.

The proposition shows that closed subsets satisfy (R1)–(R4), and therefore
are automatically subsystems.

Example B.15 Let Φ be a root system.

(1) If I is a subset of the set of simple reflections S of Φ, then ΦI is a closed

subset.
(2) More generally, if Ψ ⊆ Φ is any subset then Φ ∩ ZΨ is closed in Φ.
(3) Let Φ be indecomposable. Any subsystem consisting only of long roots
is closed. In particular, the subset Φl consisting of all long roots is a
closed subsystem of Φ (see Exercise B.2).
(4) On the other hand, the subset Φs consisting of all short roots in Φ
is, in general, not a closed subset although it is a subsystem (by the
classification of two-dimensional root systems).

Proposition B.16 The closed subsystems of Φ are precisely the sets Φ∩H
for H a subgroup of ZΦ of finite index.

Proof Let H ≤ ZΦ be a subgroup. Then clearly Φ ∩ H is symmetric and

satisfies (C2). For the converse we claim that for any finite subset M of
a finitely generated abelian group A not containing 0 there is a subgroup
of finite index having empty intersection with M . Indeed, let v ∈ M . By
the structure theorem for finitely generated abelian groups there exists a
generating system e1 , . . . , el of A such that v = me1 for some m ∈ N. Let n
denote the order of v, respectively n = 2 when v has infinite order. Clearly,
the subgroup mne1 , e2 , . . . , el  has finite index in A and does not contain
v. The claim now follows by induction on |M |.
Now, for Ψ a closed subset, choosing A = ZΦ/ZΨ and M the image of
Φ \ Ψ in A we obtain a subgroup H as required.
288 Subsystems

Proposition B.17 Let Φ be indecomposable, Ψ ⊂ Φ be a maximal closed

subsystem. Then either Ψ lies in a proper subspace of E, or there exists a
W -conjugate Ψ of Ψ and γ ∈ ∆ such that ∆ \ {γ} ∪ {α0 } ⊆ Ψ .

Proof Let H ≤ ZΦ of finite index such that Ψ = Φ ∩ H as in Proposi-

tion B.16. Since Ψ is supposed to be maximal we may assume, replacing H
by an overgroup if necessary, that ZΦ/H is cyclic. Choosing adapted bases of
ZΦ and H as in the previous proof we see that there is u ∈ E such that Ψ is
the set of roots α for which (u, α) ∈ Z. By (R4) this condition is not affected
by adding to u integral multiples of β ∨ , β ∈ Φ. Application of elements of
W sends Ψ to a conjugate maximal subsystem. So by Proposition B.12 we
may assume that u lies in the alcove A, that is, (u, α) ≥ 0 for α ∈ ∆ and
(u, α0 ) ≤ 1 by Lemma B.11.
Let ∆ := {α ∈ ∆ | (u, α) = 0}. Since Ψ is proper, ∆ = ∅. Now
" "
1 ≥ (u, α0 ) = nα (u, α) = nα (u, α) > 0.
α∈∆ α∈∆

Thus, if β = α cα α ∈ Ψ, then
"
(u, β) = cα (u, α) ∈ {0, 1}
α∈∆

by Proposition B.5, and this is non-zero only if

" "
cα (u, α) = nα (u, α) = 1,
α∈∆ α∈∆

that is, cα = nα for α ∈ ∆ . So Ψ ⊆ Z(∆ \ ∆ ) + Z( ∆ nα α) lies in a proper
subspace unless |∆ | = 1.

If ∆ = {γ}, then for β = α cα α ∈ Ψ with cγ = 0 we have

0 < (u, β) = cγ (u, γ) ≤ 1,

so in fact cγ (u, γ) = 1. In particular, 1 ≤ nγ (u, γ) ≤ (u, α0 ) ≤ 1, so α0 ∈

Ψ.

Theorem B.18 (Borel–de Siebenthal) Let Φ be an indecomposable root

l
system with base ∆ = {α1 , . . . , αl } and highest root α0 = i=1 ni αi with
respect to ∆. Then the maximal closed subsystems of Φ up to conjugation by
W are those with bases:

(1) ∆ \ {αi } for 1 ≤ i ≤ l with ni = 1, and

(2) ∆ \ {αi } ∪ {−α0 } for 1 ≤ i ≤ l with ni a prime.
 B.3 Closed subsystems 289

Proof For 1 ≤ i ≤ l let Φi be the subset of Φ consisting of the roots

which are linear combinations of ∆ \ {αi }. This is a closed subsystem by
Example B.15 with base ∆ \ {αi } (see Proposition A.25). If ni = 1 then by
Proposition B.5 the coefficient at αi of all roots β ∈ Φ \ Φi equals ±1. Thus
any subgroup H ≤ ZΦ containing Φi and one of these roots must also contain
αi , and thus be equal to ZΦ. Hence Φi is maximal by the characterization
of closed subsystems in Proposition B.16.
Next, for 1 ≤ i ≤ l with ni > 1 consider the closed subset Ψi of Φ
consisting of all Z-linear combinations of ∆0i := ∆ \ {αi } ∪ {−α0 } in Φ. Note
that the coefficient at αi for all roots in Ψi is divisible by ni , hence equal to
0, ±ni by Proposition B.5. So Ψi is a subsystem of Φ lying properly between
Φi and Φ, whence Φi is not maximal in this case. We claim that ∆0i is a base
l
of Ψi . Indeed, if β = j=1 cj αj ∈ Ψi with ci = 0, then it is a non-negative
or non-positive linear combination of ∆ \ {αi }. If ci < 0, then ci = −ni and

again β = −α0 + j =i (nj + cj )αj is a non-negative linear combination of
∆0i since nj + cj ≥ 0 by Proposition B.5.
To investigate maximality of Ψi , first assume that ni = cd with c, d > 1.
Then the set of all roots whose coefficient at αi is an integral multiple of c is
closed, strictly smaller than Φ (since it does not contain αi ), and contains Ψi .
By Corollary B.2 there exist positive roots whose coefficient at αi equals c,
so the last containment is also strict and Ψi is not maximal in this case. On
the other hand, if ni is prime then obviously ZΦ/ZΨi is cyclic of prime order
ni , so Ψi is maximal.
It remains to show that any maximal closed subsystem Ψ is of one of
the above two forms. By the previous proposition either Ψ is contained in a
proper subspace, and then it is contained in (and hence equal to) a parabolic
subsystem by Corollary A.29, or it contains α0 and all but one of the αi .

Example B.19 Let Φ be a root system of type E8 . The corresponding

Dynkin diagram is as shown in Figure B.1. We have also given the coefficients
nα of the highest root (see Table B.1).

2 4 6 5 4 3 2
E8 ......
−α0
3

Figure B.1 Coefficients of the highest root of E8 .

Since nα > 1 for all α ∈ ∆ there are no maximal closed subsystems as in

290 Subsystems

Theorem B.18(1). The maximal closed subsystems as in Theorem B.18(2)

are of types D8 , A8 , A4 × A4 , E6 × A2 and E7 × A1 . (See also Exercise B.3.)

B.4 Other subsystems

We investigate the subsystems of root systems which are not necessarily
closed.

Lemma B.20 Let Φ be an indecomposable root system containing roots of

two different lengths. Then Φ = Φ ∩ ZΦs , that is, Φ is generated by its short
roots.

Proof By Proposition A.11 any base ∆ of Φ contains short and long roots.
Since Φ is indecomposable, there is a short root α and a long root β in ∆
with (α, β) = 0. Then by Table A.2, Φ∩(Zα+Zβ) is of type B2 or G2 , and in
both cases, all long roots are integral linear combinations of the short ones.
Since W acts transitively on the set of long roots (see Corollary A.18(b))
this completes the proof.

Proposition B.21 Let Φ be an indecomposable root system, Ψ ⊂ Φ a

maximal subsystem which is not closed. Then one of the following occurs:

(1) there is an indecomposable parabolic subsystem Ψ1 ⊂ Φ containing short

roots such that Ψ = Ψ1  (Φ ∩ Ψ⊥ 1 ) is decomposable; or
(2) the subsystem Ψs of Ψ consisting of its short roots is of maximal rank
in Φ, Ψ∨ ∨
s is closed in Ψ , and Φ ⊂ ZΨs .

Proof First note that Ψ is of the same rank as Φ, since otherwise by Corol-
lary A.29 it is contained in the proper (closed) parabolic subsystem of the
parabolic subgroup pointwise fixing Ψ⊥ . Since Ψ is not closed, it contains
some short root by Example B.15(3). So there is an indecomposable compo-
nent Ψ1 of Ψ containing short roots. If Ψ1 is proper in Ψ, then

Ψ = Ψ1  (Ψ ∩ Ψ⊥ ⊥
1 ) ⊆ (Φ ∩ ZΨ1 )  (Φ ∩ Ψ1 ),

and by maximality we have equality, as in (1). Now suppose Ψ = Ψ1 is

indecomposable. Hence, by Lemma B.20 it is generated by its short roots,
so Ψs is of maximal rank. Again by maximality, we have Φ ∩ ZΨs = Φ since
Ψ is not closed. Since Ψ∨ ∨
s is a subsystem of long roots in Ψ , it is closed by
Example B.15(3).

Note that the conditions in Proposition B.21 are necessary, but not always
 B.4 Other subsystems 291

sufficient for a subsystem to be maximal. Still, in conjunction with Theo-

rem B.18 it can be used to determine the maximal non-closed subsystems of
each indecomposable root system in turn.
In Table B.2 we describe the subroot systems Φl , Φs of long, respectively
short roots in the various indecomposable root systems.

Table B.2 Long and short roots in indecomposable root systems

Φ An Bn Cn Dn E6 E7 E8 F4 G2
Φl An Dn An1 Dn E6 E7 E8 D4 A2
Φs − An1 Dn − − − − D4 A2
|Φ| n(n + 1) 2n2 2n2 2n(n − 1) 72 126 240 48 12

Since duality of root systems (see Definition A.3) interchanges long and
short roots, we see from the classification that type Cn is dual to type Bn ,
and all other indecomposable root systems are self-dual.

Example B.22 (Maximal non-closed subsystems)

(1) If Φ is indecomposable with only one root length, then any subsystem
is automatically closed, by Example B.15(3).
(2) Let Φ be of type Bn . The indecomposable proper parabolic subsystems
Ψ1 containing short roots are of types Br , 1 ≤ r < n, and then Ψ1 
(Φ ∩ Ψ⊥1 ) are maximal subsystems of type Br Bn−r . As for the situation
in Proposition B.21(2), Φs is of type An1 by Table B.2, without proper
subsystems of maximal rank. The only indecomposable root system of
rank n having fewer roots than Bn is An , but this does not contain a
maximal rank subsystem of type An1 . Thus, the only maximal non-closed
subsystems of Bn are of type Br Bn−r , 1 ≤ r < n.
(3) Let Φ be of type Cn . This root system is dual to that of type Bn , so
its maximal subsystems can be obtained as duals of those in Bn . The
duals of parabolic subsystems are parabolic, hence closed. According
to Theorem B.18 the closed maximal rank subsystems in type Bn are
Dr Bn−r with 2 ≤ r ≤ n, with duals Dr Cn−r in Cn . Finally the dual
of the maximal non-closed subsystem Br Bn−r from part (2) is Cr Cn−r ,
which is closed. So here the maximal non-closed subsystems are of type
Dr Cn−r , 2 ≤ r ≤ n.
(4) Let Φ be of type F4 . The indecomposable proper parabolic subsys-
tems containing short roots are of types A1 , A2 , C3 , which by Propo-
sition B.21(1) lead to subsystems A1 B3 , A2 A2 , C3 A1 , only the first of
292 Subsystems

which is not closed. Further, Φs is of type D4 , with only proper maximal

rank subsystem A41 . The only indecomposable rank 4 system (apart from
F4 ) containing a maximal rank subsystem of type D4 or A41 consisting of
long roots is B4 by Theorem B.18, so dualizing we obtain a subsystem
C4 .
(5) Let Φ be of type G2 . The only indecomposable proper parabolic subsys-
tem containing short roots is of type A1 , with corresponding maximal
subsystem of type A1 A1 which is closed. Further, Φs is of type A2 . This
has no proper maximal rank subsystems and it is obvious that any larger
subsystem equals Φ, so the only non-closed maximal subsystems of G2
are of type A2 (the short roots).

Corollary B.23 Let Φ be an indecomposable root system. Then −id ∈ W

if and only if Φ is of type

A1 , Bn , Cn , Dn (n even), E7 , E8 , F4 , or G2 .

Proof This follows from the above description of subsystems with Proposi-
tion A.30.

In the remaining cases An (n ≥ 2), Dn (n odd) and E6 , the longest element

necessarily acts as the (unique) non-trivial graph automorphism of order 2
on the Dynkin diagram.

B.5 Bad primes and torsion primes

Here we investigate two types of primes which play special roles for a given
indecomposable root system.

Definition B.24 Let Φ be a root system. A prime r is called bad for Φ if

ZΦ/ZΨ has r-torsion for some closed subsystem Ψ ⊆ Φ.

Proposition B.25 Let Ψ ⊆ Φ be closed. Then the bad primes of Ψ are

among those for Φ.

Proof If Ψ1 ⊆ Ψ is closed in Ψ, then it is also closed in Φ, and if ZΨ/ZΨ1

has r-torsion then so has ZΦ/ZΨ1 .

Let Φ be indecomposable with base ∆ = {α1 , . . . , αl } and highest root

l
α0 = i=1 ni αi . We set n0 := 1. Let

n(Φ) = max{ni | 1 ≤ i ≤ l}
 B.5 Bad primes and torsion primes 293

denote the largest coefficient of the highest root of Φ. Clearly, this is inde-
pendent of the chosen base.
Proposition B.26 Let Φ be indecomposable, Ψ ⊆ Φ a closed indecompos-
able subsystem. Then n(Ψ) ≤ n(Φ).
Proof Let ∆1 be a base of Ψ. Extend this to a basis of E and choose an
ordering on E with respect to which this basis is positive. Let Φ+ be the
positive system of Φ with respect to this ordering, with base ∆. Then all
roots in ∆1 are positive in this ordering, hence all elements of ∆1 are non-
negative integral combinations over ∆. Substituting these combinations into
the expression for the highest root of Ψ shows the claim.
Lemma B.27 Let Φ be indecomposable with ∆ = {α1 , . . . , αl } labeled such
that α1 , . . . , αr is a sequence in ∆ of minimal length r ≥ 0 subject to
(1) nr = n(Φ) is maximal among the ni , 1 ≤ i ≤ l, and
(2) (αi , αi+1 ) = 0 for i = 0, . . . , r − 1.
Then for i = 1, . . . , r we have (αi , αi ) = (α0 , α0 ) and ni = i + 1.
Proof If nr = 1 then r = 0 and there is nothing to prove. Now assume
that nr > 1 and let 0 ≤ j < r. Taking the scalar product of αj∨ with
l
α0 = i=1 ni αi we obtain
"
l
2= (α0 , α0∨ ) = ni (αi , α0∨ ) ≥ n1 (α1 , α0∨ ) and
i=1

"
l "
j+1
0 ≤ (α0 , αj∨ ) = ni (αi , αj∨ ) ≤ ni (αi , αj∨ ) for j = 1, . . . , r − 1
i=1 i=j−1

since (α0 , αj∨ ) ≥ 0 by Corollary B.7(b) and ni (αi , αj∨ ) ≤ 0 for i, j ≥ 1, i = j

∨
by Lemma A.6. Using that (αj , αj+1 ) ≤ −1 by (2) this yields
2 − n1 ≥ 0 and − nj−1 + 2nj − nj+1 ≥ 0 for j = 1, . . . , r − 1.
Adding up these equations gives 1 + nr−1 − nr ≥ 0, hence nr = nr−1 + 1
by the minimality of the sequence. Thus equality holds here, and hence in
fact in all of the inequalities above. By induction this shows that nj = j + 1.
∨
From (αi , αi±1 ) = −1 for 1 ≤ i ≤ r − 1, respectively (α1 , α0∨ ) = 1, we get
(αi−1 , αi−1 ) = (αi , αi ) as claimed.
Corollary B.28 Let Φ be an indecomposable root system with highest root

α0 = α∈∆ nα α and r a prime. The following are equivalent:
(i) r is bad for Φ.
294 Subsystems

(ii) r equals some nα .

(iii) r divides some nα .
(iv) r ≤ nα for some α ∈ ∆.

Proof The equivalence of (ii), (iii), and (iv) is a direct consequence of

Lemma B.27. Now assume that r is bad for Φ. So there is a closed sub-
system Ψ ⊆ Φ such that ZΦ/ZΨ has r-torsion. Let I ⊆ ∆ be minimal such
that Ψ ⊆ ΦI up to conjugation. Then by Theorem B.18 we have that r = nIα
for some α ∈ I, where nIα denotes the coefficient of α in the highest root of
ΦI . But nIα ≤ n(Ψ) ≤ n(Φ) by Proposition B.26 so we get (iv).
Conversely, if r = nα for some α ∈ ∆ then the maximal subsystem Ψα in
Theorem B.18(2) gives rise to r-torsion.

This characterization of bad primes could also easily be verified “by inspec-
tion” of the various indecomposable root systems, see Table B.1. Steinberg
[74, §1] has given the above proof relying on Lemma B.27 which does not
use the classification.
We now come to the second type of primes:

Definition B.29 Let Φ be a root system with dual root system Φ∨ . A

prime r is said to be a torsion prime for Φ if ZΦ∨ /ZΨ∨ has r-torsion for
some closed subsystem Ψ ⊆ Φ.

Note that here we do not require that Ψ∨ be closed in Φ∨ .


Lemma B.30 Let Φ be indecomposable with highest root α0 = α nα α.

Then α0∨ = α n∨ α∨ with n∨α = nα (α, α)/(α0 , α0 ). Moreover, if β ∈ Φ is
+
∨
α ∨ ∨ ∨ ∨
long, and β = α cα α , then cα ≤ nα for all α ∈ ∆.

Proof By definition, for β = α cα α ∈ Φ we have
" cα " (α, α) "
β∨ = 2 α= cα α∨ =: c∨ ∨
αα .
α
(β, β) α
(β, β) α

Since cα ≤ nα for all α ∈ ∆ by Proposition B.5 this shows that c∨ ∨

α ≤ nα
whenever (β, β) = (α0 , α0 ).

In particular, if Φ only contains roots of one fixed length, then α0∨ is the
highest root of Φ∨ . Else, by definition β ∈ Φ is long if and only if β ∨ ∈ Φ∨
is short, so the preceding result can be paraphrased as saying that α0∨ is
the highest short root of Φ∨ . A list of highest short roots in the case where
two different root lengths occur can readily be derived from Table B.1, see
Table B.3.
We let n∨ (Φ) := max{n∨ i | 1 ≤ i ≤ l}, the largest coefficient of the highest
 B.5 Bad primes and torsion primes 295

Table B.3 Highest short roots of indecomposable root systems

1 1 1 1 1 2 2 1
Bn > Cn <
n≥2 n≥3

1 2 3 2 2 1
F4 > G2 <

short root in the dual root system. Then we have the following analogue of
Proposition B.26:
Proposition B.31 Let Φ be indecomposable, Ψ ⊆ Φ a closed indecompos-
able subsystem. Then n∨ (Ψ) ≤ n∨ (Φ).
Proof As in the proof of Proposition B.26 we may choose compatible bases
in Ψ and Φ. Then, by induction we may assume that Ψ is a maximal subsys-

tem of Φ. Now let α̃0 denote the highest root of Ψ and write α̃0 = i ñi α̃i ,

where {α̃1 , . . . , α̃r } is the chosen base of Ψ. Expressing α̃i = j mij αj in
the base {α1 , . . . , αl } of Φ with non-negative integral coefficients, we have
"%" &
α̃0 = ñi mij αj
j i

and thus i ñi mij ≤ nj for all j by Proposition B.5. Now sums of long roots
are long, by Example B.15(3), so for any short root α̃i there is some j with
αj short and mij = 0. Thus,
max{ñi | α̃i short} ≤ max{nj | αj short},
and similarly for long roots. Now if α̃0 has the same length as α0 (hence is
long) then by Lemma B.30 we find
n∨ (Ψ) = max{ñ∨ ∨ ∨
i } ≤ max{nj } = n (Φ).

Thus we may assume α̃0 is short, whence all roots in Ψ are short, and
among the possible maximal closed subsystems described in Theorem B.18
this can only happen when Ψ = Z(∆ \ {αi }) ∩ Φ for some i, with αi long
and all other simple roots short. Let s denote the reflection in α̃0 . Then β :=
s.αi = αi − (αi , α̃0∨ )α̃0 is long, and since α̃0 is a positive linear combination
in ∆ \ {αi }, (αi , α̃0∨ ) < 0 by Lemma A.6. So the coefficients of α̃0∨ are
dominated by those of β ∨ , whose coefficients in turn are dominated by those
of α0∨ , because both β, α0 are long.
296 Subsystems

Proposition B.32 Let Φ be an indecomposable root system with highest

 ∨ ∨
short root α0∨ = α nα α of the dual root system and r a prime. The
following are equivalent:
(i) r is a torsion prime for Φ.
(ii) r equals some n∨ α.
(iii) r divides some n∨ α.
(iv) r ≤ n∨α for some α ∈ ∆.

Proof The proof is completely analogous to that of Corollary B.28, using

Proposition B.31 in place of Proposition B.26.
The lists of bad primes and torsion primes for the various indecomposable
root systems can now be extracted from the highest roots in Tables B.1
and B.3. They are given in Table 14.1.

Exercises
B.1 Let Φ be a root system and α ∈ Φ. Show that the affine transformation
sα,j is a reflection in the hyperplane Hα,j := {v ∈ E | (v, α) = j} along
α∨ and that sα,j = tjα∨ sα .
B.2 Let Φ be an indecomposable root system, Ψ a subsystem consisting
only of long roots. Then Ψ is closed in Φ.
[Hint: Use the classification of two-dimensional root systems in Proposi-
tion A.17.]
B.3 Consider the root system Φ of type E8 (see Example B.19). By Theo-
rem B.18, the subsystems Ψi generated by ∆ \ {αi } ∪ {−α0 } are not
maximal when ni is not prime. In each such case find a (base of a)
maximal subsystem containing Ψi .
B.4 Let Φ be a root system, S the set of simple reflections corresponding
to some base of Φ.
(a) Show that ZΦ/ZΦI has no torsion, for any I ⊆ S.
(b) Conclude that Φ of type An has no bad primes nor torsion primes.
(c) Determine the torsion primes for Φ of type Cn .
 Appendix C
Automorphisms of root systems

In this appendix we consider the following situation: Φ ⊂ E is a root system

in the Euclidean space E with positive system Φ+ , and F ∈ GL(E) is a
linear transformation which permutes Φ+ up to positive multiples, that is,
F stabilizes the set of half-lines {R+ α | α ∈ Φ+ }. Such automorphisms arise
naturally in the study of finite groups of Lie type viewed as groups of fixed
points under Steinberg endomorphisms in Part III.
Given these data there exists a permutation ρ of Φ stabilizing Φ+ such
that F (ρ(α)) = qα α for all α ∈ Φ and suitable qα > 0. By Proposition A.1,
F permutes the reflections {sα | α ∈ Φ}, so normalizes the Weyl group
W of Φ. Note that ρ also stabilizes the simple system ∆. Indeed, for all

β = α∈∆ cα α ∈ Φ+ we have that
" "
ρ(β) = qβ F −1 (β) = qβ cα F −1 (α) = qβ cα qα−1 ρ(α) ∈ Φ+
α∈∆ α∈∆

is a non-negative linear combination of ρ(∆), so ρ(∆) is another base in

Φ+ , hence equal to ∆ by Proposition A.7. Thus, F also permutes the set
S = {sα | α ∈ ∆} of simple reflections of W and hence induces a graph
automorphism of the corresponding Coxeter diagram.
We first investigate the centralizer of F in W . For I ⊆ S we let ∆I ⊆ ∆
denote the corresponding set of simple roots and WI = I the associated
parabolic subgroup of W . We write W F for the centralizer of F in W .

Lemma C.1 Let F ∈ GL(E) be as above.

(a) For each F -orbit I ⊆ S we have WIF = wI , where wI is the longest
element of the parabolic subgroup WI .
(b) The group of fixed points W F is generated by {wI | I ⊆ S an F -orbit}.

Proof Let wI ∈ WI denote the longest element of WI . Since I is an F -orbit,

298 Automorphisms of root systems

F normalizes WI and we have F wI F −1 ∈ WI . Since ρ also preserves Φ+ ,

−
F wI F −1 sends Φ+
I to ΦI . But then by the uniqueness of the longest element
(Corollary A.23) we must have F wI F −1 = wI , that is, wI ∈ WIF .
Now let w ∈ W F . If w = 1 then there exists α ∈ ∆ such that w.α ∈ Φ− .
But then also w.β ∈ Φ− for all β in the ρ-orbit ∆I of α. By Corollary A.26,
−
wI also sends Φ+I to ΦI but leaves all other positive roots positive, so

(wwI ) = n(wwI ) = n(w) − n(wI ) = (w) − (wI ).

Thus wwI ∈ W F has smaller length than w, and we conclude by induction
that W F is generated by the wI . Applying this to WI we obtain that WIF =
wI .
The preceding result implies in particular that the Coxeter diagram of W F
can be obtained by a suitable folding process from the one of W , according
to the graph automorphism induced by F . The various possibilities for W F
when W is irreducible and F is non-trivial are displayed in Table 23.1.
We now study the action of F on the root system. If δ denotes the order
of ρ, then F δ fixes all roots up to positive constants. To simplify exposition
we make the further assumption that all these constants agree (which is the
case for example if Φ is indecomposable). Then F = qφ for some q > 0 and
an automorphism φ ∈ GL(E) of order δ. Since W, φ is finite, we may and
will assume that the scalar product on E is W, φ-invariant. Consider the
homomorphism
1" i
δ−1
π : E −→ E φ , χ→ φ (χ);
δ i=0

this is clearly surjective and the identity on E φ , hence a projection.

Lemma C.2 I ) | I ⊆ S an F -orbit, w ∈ W }. Then:
Let Ω := {w(Φ+ F

(a) Ω defines a partition of Φ. If ∼ denotes the associated equivalence rela-

tion, then α ∼ β if and only if π(α), π(β) are positive multiples of each
other.
(b) Each ω ∈ Ω is contained in either Φ+ or in Φ− .
Proof Let w0 be the longest element of W . As F preserves the set of positive
roots up to multiples, F w0 F −1 has the same property as w0 , so equals w0 .
Hence w0 ∈ W F .
Now let α ∈ Φ+ . Then w0 .α ∈ Φ− . Write w0 = w1 · · · wr , where wj =
wIj ∈ W F for suitable F -orbits Ij ⊆ S, by Lemma C.1(b). Then there exists
1 ≤ j ≤ r such that
wj+1 · · · wr .α ∈ Φ+ and wj wj+1 · · · wr .α ∈ Φ− .
 Automorphisms of root systems 299

By Corollary A.26 all positive roots made negative by wj ∈ WI lie in Φ+ I ,

where I := Ij . Thus wj+1 · · · wr .α ∈ Φ+ I , whence α is contained in ω :=
(wj+1 · · · wr )−1 (Φ+
I ) ∈ Ω. As w0 ∈ W , the same applies to negative roots.
F

Now suppose that α, β ∈ Φ with π(β) = aπ(α) for some a > 0. By the
previous argument, w.α ∈ Φ+ I for some w ∈ W
F
and some F -orbit I ⊆ S.
So π(w.α) is a non-negative linear combination of roots in ∆I . However

π(w.β) = w.π(β) = w.(aπ(α)) = aπ(w.α),

so π(w.β) also is a non-negative linear combination of roots in ∆I . By the

definition of π, this must also hold for w.β, so w.β ∈ Φ+ I as well, whence
α, β lie in the same equivalence class w−1 (Φ+I ).
Finally, π is clearly constant on ∆I for any F -orbit I ⊆ S. Now any root
γ ∈ Φ+ I is a non-negative linear combination of roots in ∆I , so π(γ) is a
positive multiple of π(α), for α ∈ ∆I . This completes the proof of (a).
For (b) let I ⊆ S be an F -orbit and α ∈ ∆I . If w.α ∈ Φ+ for some
w ∈ W F , then also w.ρi (α) = ρi (w.α) ∈ Φ+ for all i since ρ preserves Φ+
and commutes with W F . Thus w(∆I ) ⊆ Φ+ and then also w(Φ+ I )⊆Φ .
+

Lemma C.3 Let I ⊆ S be an F -orbit. Then the restriction to E φ of the

longest element wI ∈ W F is a reflection in π(α) for all α ∈ ∆I .

Proof Let α ∈ ∆I . Since wI (∆I ) = −∆I we have

1" 1" i
δ−1 δ−1
wI .π(α) = wI .φi (α) = φ (wI .α) = −π(α).
δ i=0 δ i=0

On the other hand, if v ∈ E φ satisfies (π(α), v) = 0, then

"
δ−1 "
δ−1
i
0= (φ (α), v) = (α, v),
i=0 i=0

so (α, v) = (φi (α), v) = 0 for all i. Since wI is a product of reflections sβ ,

with β ∈ ∆I , this forces wI .v = v.

Definition C.4 For each equivalence class ω for the equivalence relation ∼
defined in Lemma C.2 we let αω denote the vector of maximal length among
the {π(α) | α ∈ ω}. We set ΦF := {αω | ω ∈ Ω} and ∆F := {αω | ω ⊆ ∆} ⊆
ΦF .

Theorem C.5 The set ΦF satisfies axioms (R1)–(R3) of an abstract root

system in E φ , with Weyl group W F and base ∆F .
300 Automorphisms of root systems

Proof Since Φ spans E, it is immediate that ΦF spans E φ . By construction,

ΦF is finite and none of the αω is zero. This shows (R1). By Lemma C.2(a)
and the definition of ΦF , if µ, ν ∈ ΦF are positive multiples of each other,
they are equal, proving (R2). Next, if µ ∈ ∆F then by Lemma C.3 there
is w ∈ W F whose restriction to E φ is the orthogonal reflection in µ. By
Lemma A.4 we obtain reflections for arbitrary µ ∈ ΦF . Then for β ∈ Φ,
w ∈ W F , we have w.π(β) = π(w.β) ∈ ΦF , which proves (R3).
Each element of Φ is a non-negative or non-positive linear combination in
∆, so by construction every element of ΦF is a non-negative or non-positive
linear combination in ∆F . Since ∆ is linearly independent, so is ∆F . Finally,
the Weyl group of ΦF is generated by the reflections in the α ∈ ∆F by
Lemma C.1.
For Φ indecomposable it can be shown case by case that ΦF also satisfies
the integrality axiom (R4) unless Φ is of type F4 and ρ is non-trivial. See
also Exercise C.2.

Exercises
C.1 (Automorphisms of root systems) Let Φ be a root system in the Eu-
clidean space V and σ ∈ GL(V ) stabilizing a positive system Φ+ ⊂ Φ.
Then σ induces a graph automorphism of the Dynkin diagram of Φ.
[Hint: First check that σ stabilizes the base ∆ in Φ+ . Then use the clas-
sification of root systems of rank 2 to see that both the number of edges
between two roots α, β ∈ ∆, as well as the direction of the arrow, is already
determined by the set (Zα + Zβ) ∩ Φ.]
C.2 Verify the structure of W F and ΦF for the cases occurring in Table 23.1.
 References

[1] M. Aschbacher, On the maximal subgroups of the finite classical groups. In-
vent. Math. 76 (1984), 469–514.
[2] M. Aschbacher, Finite Group Theory. Second edition. Cambridge Studies in
Advanced Mathematics, 10. Cambridge University Press, Cambridge, 2000.
[3] H. B. Azad, M. Barry, G. M. Seitz, On the structure of parabolic subgroups.
Comm. Algebra 18 (1990), 551–562.
[4] A. Borel, Linear Algebraic Groups. Second edition. Graduate Texts in Math-
ematics, 126. Springer-Verlag, New York, 1991.
[5] A. Borel, J. de Siebenthal, Les sous-groupes fermés de rang maximum des
groupes de Lie clos. Comment. Math. Helv. 23 (1949), 200–221.
[6] A. Borel, J. Tits, Groupes réductifs. Inst. Hautes Études Sci. Publ. Math. 27
(1965), 55–150.
[7] A. Borel, J. Tits, Eléments unipotents et sous-groupes paraboliques de groupes
réductifs. I. Invent. Math. 12 (1971), 95–104.
[8] A. V. Borovik, The structure of finite subgroups of simple algebraic groups.
(Russian) Algebra i Logika 28 (1989), 249–279, 366; translation in Algebra
and Logic 28 (1989), 163–182 (1990).
[9] N. Bourbaki, Groupes et Algèbres de Lie. IV, V, VI, Hermann, Paris, 1968.
[10] M. Broué, G. Malle, Théorèmes de Sylow génériques pour les groupes réductifs
sur les corps finis. Math. Ann. 292 (1992), 241–262.
[11] J. Brundan, Double coset density in classical algebraic groups. Trans. Amer.
Math. Soc. 352 (2000), 1405–1436.
[12] M. Cabanes, Unicité du sous-groupe abélien distingué maximal dans certains
sous-groupes de Sylow. C. R. Acad. Sci. Paris Sér. I Math. 318 (1994), 889–
894.
[13] R. W. Carter, Simple Groups of Lie Type. John Wiley & Sons, London, 1972.
[14] R. W. Carter, Finite Groups of Lie Type—Conjugacy Classes and Complex
Characters. John Wiley & Sons, New York, 1985.
[15] C. Chevalley, Classification des Groupes Algébriques Semi-simples. Collected
Works. Vol. 3. Springer-Verlag, Berlin, 2005.
[16] E. Cline, B. Parshall, L. Scott, On the tensor product theorem for algebraic
groups. J. Algebra 63 (1980), 264–267.
302 References

[17] A. M. Cohen, M. W. Liebeck, J. Saxl, G. M. Seitz, The local maximal sub-

groups of exceptional groups of Lie type, finite and algebraic. Proc. London
Math. Soc. (3) 64 (1992), 21–48.
[18] B. N. Cooperstein, Maximal subgroups of G2 (2n ). J. Algebra 70 (1981), 23–
36.
[19] C. W. Curtis, I. Reiner, Representation Theory of Finite Groups and Asso-
ciative Algebras. John Wiley & Sons, New York, 1962.
[20] D. I. Deriziotis, Centralizers of semisimple elements in a Chevalley group.
Comm. Algebra 9 (1981), 1997–2014.
[21] J. A. Dieudonné, La Géométrie des Groupes Classiques. Third edition.
Springer-Verlag, Berlin, 1971.
[22] F. Digne, J. Michel, Fonctions L des Variétés de Deligne–Lusztig et Descente
de Shintani. Mém. Soc. Math. France (N.S.) 20 (1985).
[23] E. B. Dynkin, Semisimple subalgebras of semisimple Lie algebras. Amer.
Math. Soc., Transl., II. Ser. 6 (1957), 111–243.
[24] E. B. Dynkin, Maximal subgroups of the classical groups. Amer. Math. Soc.,
Transl., II. Ser. 6 (1957), 245–378.
[25] A. Fröhlich, M. J. Taylor, Algebraic Number Theory. Cambridge Studies in
Advanced Mathematics, 27. Cambridge University Press, Cambridge, 1993.
[26] M. Geck, An Introduction to Algebraic Geometry and Algebraic Groups. Ox-
ford Graduate Texts in Mathematics, 10. Oxford University Press, Oxford,
2003.
[27] R. Goodman, N. Wallach, Representations and Invariants of the Classical
Groups. Encyclopedia of Mathematics and its Applications, 68. Cambridge
University Press, Cambridge, 1998.
[28] D. Gorenstein, Finite Groups. Chelsea Publishing Company, New York, 1980.
[29] D. Gorenstein, R. Lyons, R. Solomon, The Classification of Finite Simple
Groups, Number 3. Mathematical Surveys and Monographs, 40. American
Mathematical Society, Providence, RI, 1998.
[30] L. Grove, Classical Groups and Geometric Algebra. Graduate Studies in Math-
ematics, 39. American Mathematical Society, Providence, RI, 2002.
[31] G. Hiss, Die adjungierten Darstellungen der Chevalley-Gruppen. Arch. Math.
(Basel) 42 (1984), 408–416.
[32] J. Humphreys, Linear Algebraic Groups. Graduate Texts in Mathematics, 21.
Springer-Verlag, New York, 1975.
[33] J. Humphreys, Introduction to Lie Algebras and Representation Theory. Grad-
uate Texts in Mathematics, 9. Springer-Verlag, New York, Second printing,
1980.
[34] J. Humphreys, Reflection Groups and Coxeter Groups. Cambridge University
Press, Cambridge, 1992.
[35] J. Humphreys, Conjugacy Classes in Semisimple Algebraic Groups. Mathe-
matical Surveys and Monographs, 43. American Mathematical Society, Prov-
idence, RI, 1995.
[36] J. Humphreys, Modular Representations of Finite Groups of Lie Type. LMS
Lecture Notes Series, 326. Cambridge University Press, Cambridge, 2006.
[37] B. Huppert, Endliche Gruppen. I. Grundlehren der Mathematischen Wis-
senschaften, 134. Springer-Verlag, Berlin, 1967.
 References 303

[38] B. Huppert, N. Blackburn, Finite Groups. II. Grundlehren der Mathematis-

chen Wissenschaften, 242. Springer-Verlag, Berlin, 1982.
[39] I. M. Isaacs, Character Theory of Finite Groups. Dover, New York, 1994.
[40] J.C. Jantzen, Darstellungen halbeinfacher algebraischer Gruppen und zuge-
ordnete kontravariante Formen. Bonner math. Schr. 67 (1973).
[41] J.C. Jantzen, Representations of Algebraic Groups. Second edition. Mathe-
matical Surveys and Monographs, 107. American Mathematical Society, Prov-
idence, RI, 2003.
[42] P. Kleidman, The maximal subgroups of the Steinberg triality groups 3 D4 (q)
and of their automorphism groups. J. Algebra 115 (1988), 182–199.
[43] P. Kleidman, The maximal subgroups of the Chevalley groups G2 (q) with q
odd, the Ree groups 2 G2 (q), and their automorphism groups. J. Algebra 117
(1988), 30–71.
[44] P. Kleidman, M. W. Liebeck, The Subgroup Structure of the Finite Classical
Groups. London Mathematical Society Lecture Note Series, 129. Cambridge
University Press, Cambridge, 1990.
[45] V. Landazuri, G. M. Seitz, On the minimal degrees of projective representa-
tions of the finite Chevalley groups. J. Algebra 32 (1974), 418–443.
[46] R. Lawther, Sublattices generated by root differences, preprint.
[47] G. I. Lehrer, D. E. Taylor, Unitary Reflection Groups. Australian Mathemati-
cal Society Lecture Series, 20. Cambridge University Press, Cambridge, 2009.
[48] M. W. Liebeck, G. M. Seitz, Maximal subgroups of exceptional groups of Lie
type, finite and algebraic. Geom. Dedicata 35 (1990), 353–387.
[49] M. W. Liebeck, G. M. Seitz, Reductive Subgroups of Exceptional Algebraic
Groups. Memoirs Amer. Math. Soc., 121 (1996).
[50] M. W. Liebeck, G. M. Seitz, On the subgroup structure of classical groups.
Invent. Math. 134 (1998), 427–453.
[51] M. W. Liebeck, G. M. Seitz, On the subgroup structure of exceptional groups
of Lie type. Trans. Amer. Math. Soc. 350 (1998), 3409–3482.
[52] M.W. Liebeck, G.M. Seitz, A survey of maximal subgroups of exceptional
groups of Lie type. Groups, Combinatorics & Geometry (Durham, 2001),
World Sci. Publ., River Edge, NJ, 2003, pp. 139–146.
[53] M.W. Liebeck, G.M. Seitz, The Maximal Subgroups of Positive Dimension in
Exceptional Algebraic Groups. Memoirs Amer. Math. Soc., 802 (2004).
[54] F. Lübeck, Small degree representations of finite Chevalley groups in defining
characteristic. LMS J. Comput. Math. 4 (2001), 135–169.
[55] G. Malle, The maximal subgroups of 2 F4 (q 2 ). J. Algebra 139 (1991), 52–69.
[56] G. Malle, Height 0 characters of finite groups of Lie type. Represent. Theory
11 (2007), 192–220.
[57] R. Ree, A family of simple groups associated with the simple Lie algebra of
type (F4 ). Amer. J. Math. 83 (1961) 401–420.
[58] R. Ree, A family of simple groups associated with the simple Lie algebra of
type (G2 ). Amer. J. Math. 83 (1961), 432–462.
[59] R. W. Richardson, Finiteness theorems for orbits of algebraic groups. Nederl.
Akad. Wetensch. Indag. Math. 47 (1985), 337–344.
[60] G. M. Seitz, The Maximal Subgroups of Classical Algebraic Groups. Memoirs
Amer. Math. Soc., 67 (1987).
304 References

[61] G. M. Seitz, Maximal Subgroups of Exceptional Algebraic Groups. Memoirs

Amer. Math. Soc., 90 (1991).
[62] G. M. Seitz, D. M. Testerman, Extending morphisms from finite to algebraic
groups. J. Algebra 131 (1990), 559–574.
[63] S. D. Smith, Irreducible modules and parabolic subgroups. J. Algebra 75
(1982), 286–289.
[64] N. Spaltenstein, Classes Unipotentes et Sous-Groupes de Borel. Lecture Notes
in Mathematics, 946. Springer-Verlag, Berlin, 1982.
[65] T. A. Springer, Regular elements of finite reflection groups. Invent. Math. 25
(1974), 159–198.
[66] T. A. Springer, Linear Algebraic Groups. Second edition. Progress in Mathe-
matics, 9. Birkhäuser, Boston, 1998.
[67] T. A. Springer, R. Steinberg, Conjugacy classes. In: Seminar on Algebraic
Groups and Related Finite Groups. Lecture Notes in Mathematics, 131.
Springer-Verlag, Berlin, 1970, pp. 167–266.
[68] R. Steinberg, Variations on a theme of Chevalley. Pacific J. Math. 9 (1959),
875–891.
[69] R. Steinberg, Automorphisms of classical Lie algebras. Pacific J. Math. 11
(1961), 1119–1129.
[70] R. Steinberg, Representations of algebraic groups. Nagoya Math. J. 22 (1963),
33–56.
[71] R. Steinberg, Regular elements of semisimple algebraic groups. Inst. Hautes
Études Sci. Publ. Math. 25 (1965), 49–80.
[72] R. Steinberg, Endomorphisms of Linear Algebraic Groups. Memoirs Amer.
Math. Soc., 80 (1968).
[73] R. Steinberg, Lectures on Chevalley Groups. Notes prepared by J. Faulkner
and R. Wilson. Yale University, New Haven, Conn., 1968.
[74] R. Steinberg, Torsion in reductive groups. Advances in Math. 15 (1975), 63–
92.
[75] R. Steinberg, Conjugacy Classes in Algebraic Groups. Lecture Notes in Math-
ematics, Vol. 366. Springer-Verlag, Berlin, 1974.
[76] R. Steinberg, On theorems of Lie–Kolchin, Borel, and Lang. In: Contributions
to Algebra (Collection of Papers Dedicated to Ellis Kolchin). Academic Press,
New York, 1977, pp. 349–354.
[77] I. Suprunenko, Conditions on the irreducibility of restrictions of irreducible
representations of the group SL(n, K) to connected algebraic subgroups.
Preprint #13, (222), Ins. Mat. Akad. Nauk BSSR (1985) (in Russian).
[78] M. Suzuki, On a class of doubly transitive groups. Ann. of Math. (2) 75
(1962), 105–145.
[79] D. E. Taylor, The Geometry of the Classical Groups. Heldermann Verlag,
Berlin, 1992.
[80] D. Testerman, Irreducible Subgroups of Exceptional Algebraic Groups. Mem-
oirs Amer. Math. Soc., 75 (1988).
[81] D. Testerman, A construction of certain maximal subgroups of the algebraic
groups E6 and F4 . J. Algebra 122 (1989), 299–322.
[82] J. Tits, Algebraic and abstract simple groups. Ann. of Math. (2) 80 (1964),
313–329.
 Index

abstract root system, 63, 269 character, 22

additive group, 4 character group, 22
additive polynomial, 183 of Dn , 23
adjoint representation, 49 Chevalley commutator formula, 86
highest weight, 129 Cl(V ), 101, 236
weights of, 122 class S, 242
adjoint type, 71 class S k , 156
affine algebraic variety, 3 classes C1 , . . . , C5 , 152
affine reflection, 285 classes C1F , . . . , C7F , 240
affine Weyl group, 285 classical groups, 7
alcove, 286 classical type, 70
wall of, 286 closed subsystem, 104, 286
algebra of regular functions, 3 cocharacter, 22
algebraic set, 3 cocharacter group, 22
almost simple group, 153 of Dn , 23
Ann, 70, 173 commutator formula, 86
Auta (G), 88 for GF , 202
automorphism group complete root datum, 195
of a semisimple group, 88 for GLn (q), 195
of GF , 216 for GUn (q), 195
bad prime, 116, 292 components
of an algebraic group, 117 of a finite group, 153
base of a root system, 64, 270 of a root system, 273
big cell, 93 of an algebraic group, 11
BN-pair, 91 connected variety, 10
for GF , 209 coordinate algebra, 3
Borel fixed point theorem, 36 coroots, 60, 269
Borel subgroup, 37 of SL2 and PGL2 , 68
of classical groups, 39, 77 Coxeter diagram, 66, 276
opposite, 93 Coxeter group, 67
Borel–de Siebenthal algorithm, 109 crystallographic condition, 63, 269
Bruhat decomposition, 92 CSp2n , 7
for GF , 203 cyclotomic polynomial, 218
center of GF , 211 d-torus, 222
306 Index

degrees of a reflection group, 206 of a semisimple group, 71

derivation, 44 G-space, 31
derived length, 6 G a , Gm , 4
derived series, 6 general linear group, 4
descending central series, 6 general unitary group, 182
diagonal automorphism, 215 generic Sylow theorems, 224
differential of a morphism, 46 GLn , 4
dimension gln , 44
of a variety, 31 GLn (q), 181
of an affine variety, 13 GOn , 8
Dn , 6 GO± 2n (q), 194
dominant weight, 124 graded trace, 265
double centralizer, 239, 253 graph automorphisms
dual root system, 269 of a semisimple group, 90
Dynkin diagram, 66, 276 of An , 90
extended, 107 of D4 , 107
endomorphism of Dn , 172
locally finite, 16 of E6 , 100
locally nilpotent, 16 of GF , 215
locally semisimple, 16 GUn (q), 182
Ennola duality, 196, 235 half-spin groups, 72
exceptional type, 70 height of a root, 107, 272
exotic local subgroup, 252 highest root, 107, 284
extended Dynkin diagram, 107 highest short root, 294
F -conjugacy class, 185 highest weight, 125
F -conjugate, 185 of adjoint representation, 129
field automorphisms highest weight module, 125
of GF , 215 homocyclic group, 225
finite classical groups, 236 homogeneous G-space, 31
finite group of Lie type, 184 homogeneous module, 149
automorphism group, 216 indecomposable root system, 66, 273
BN-pair, 209 integral weight, 126
Bruhat decomposition, 203 internal module, 143
center, 211 irreducible component, 10
diagonal automorphism, 215 irreducible variety, 9
field automorphism, 215 isogeny, 71
graph automorphism, 215 lifting, 73, 192
order, 208 isogeny type, 71
root system, 199 isolated elements, 114, 119, 234
Schur multiplier, 214 Isom(V ), 101
simplicity, 213 isomorphism theorem, 70
Sylow p-subgroups, 210 isotropic vector, 101
twisted, 193
J2n , 7
finite reflection group, 205
Jordan decomposition, 17
invariant ring, 205
in the Lie algebra, 48
flag, 31
Jordan subgroup, 251
flag variety, 31
fundamental chamber, 279 kG-module, 32
fundamental dominant weight, 127 Kn , 7
fundamental group Landazuri–Seitz bounds, 248
of a root system, 70 left invariant derivation, 44
 Index 307

left translation, 44 orthogonal group, 8

length function, 276 conformal, 8
level of a root, 140 non-split, 194
Levi complement, 99 of minus type, 194
Levi decomposition, 99 of plus type, 194
Levi subgroup, 99 projective conformal, 35
Lie algebra special, 11
of an algebraic group, 44 p-closed subsystem, 104
of Ga , 45 parabolic subgroup, 97
of GLn , 45 in classical groups, 101
of Gm , 45 of a Weyl group, 95, 278
of SLn and PGLn , 48 parabolic subsystem, 95, 278
of Un , Dn and Tn , 78 PCO2n , 35
Lie algebra of derivations, 44 PCSp2n , 35
Lie algebra representation, 136 PGLn , 35
linear algebraic group, 4 Poincaré polynomial, 205
local subgroup, 153 point derivation, 45
exotic, 252 positive roots, 64, 270
locally finite endomorphism, 16 positive system, 270
locally nilpotent endomorphism, 16 primitive prime divisor, 247
locally semisimple endomorphism, 16 projective general linear group, 35
long root, 276 projective variety, 30
longest element, 93, 277
quasi-isolated elements, 113
maximal tori quasi-projective variety, 34
in GLn (q), 221 quasi-simple group, 153
maximal torus, 28, 38 quotient variety, 34
maximal vector, 124
maximally split torus, 187 radical, 41
in GLn (q), 187 rank, 38
in GUn (q), 187 of a root system, 64, 269
morphism of classical groups, 38, 77
of affine varieties, 3 rational representation, 32
of G-spaces, 31 reduced expression, 276
of linear algebraic groups, 5 reductive algebraic group, 41
of varieties, 31 Ree groups, 193
multiplicative group, 4 reflection, 60, 268
multiplicity of a weight, 122 reflection group, 205
invariant ring, 205
natural isogenies, 71 regular element, 115
nilpotent group, 6 restricted
nilpotent part, 16 module, 137
non-degenerate quadratic form, 101 representation, 137
Op (G), 41 weight, 137
opposite Borel subgroup, 93 restricted representation, 136
orbit map, 32 right translation, 16
order formula for GF , 206 root datum, 69
order on X(T ), 124 root subgroups, 59
order polynomial, 207 of SLn , 84
of GLn , 207 of SO2n , 85
of SLn , 207 of GF , 200
of SO+8 , 208 root subspace, 59
308 Index

root system standard parabolic subgroup

abstract, 63, 269 of a reductive group, 97
classification, 66 of a Weyl group, 95, 278
of GLn , 52 Steinberg endomorphism, 183
of SLn , 52 classification, 191
of SO2n , 85 Fq -split, 191
of Sp2n , 79 on Gm and Ga , 183
of GF , 199 twisted, 191
of a connected algebraic group, 51 very twisted, 191
of classical type, 70 Steinberg presentation, 86
of exceptional type, 70 Steinberg tensor product theorem, 138
rank of, 64, 269 structure constants, 86
simply laced, 67 subsystem
two-dimensional, 65, 275 closed, 104
roots, 51 of a root system, 278
height of, 107 p-closed, 104
level of, 140 subsystem subgroup, 104
positive, 64, 270 SUn (q), 182
shape of, 140, 280 supersolvable group, 226
Sn , 9 Suzuki groups, 193
Schur multiplier Sylow d-torus, 222
of finite groups of Lie type, 214 Sylow subgroups
semisimple algebraic group, 41 centralizer, 227
classification, 70 normalizer, 210, 228
semisimple endomorphism, 15 structure of, 225
semisimple part, 16 symmetric subset, 287
semisimple rank, 41 symplectic form, 7
shape of a root, 140, 280 symplectic group, 7
sheaf of functions, 30 conformal, 7
short root, 276 projective conformal, 35
simple algebraic group, 61 system of positive roots, 64, 270
of adjoint type, 71 tangent space, 45
of simply connected type, 71 tensor product theorem, 138
simple reflections, 84, 272 Tits group, 213
simple roots, 84, 271 Tn , 6
simplicity torsion prime
of GF , 213 of a root system, 116, 294
of algebraic groups, 97 of an algebraic group, 117
simply connected type, 71 torus, 22
simply laced, 67 totally isotropic subspace, 101
Singer cycle, 220 totally singular subspace, 101
singular vector, 101 triality automorphism, 101
SOn , 11
SO± Un , 6
2n (q), 194
socle of a module, 134 unipotent endomorphism, 15
solvable group, 6 unipotent group, 17
Sp2n , 7 unipotent part, 16
special linear group, 7 unipotent radical, 41
special unitary group, 182 variety
spin groups, 72 affine, 3
standard Frobenius, 181 connected, 10
 Index 309

defined over Fq , 181

dimension, 13, 31
irreducible, 9
projective, 30
quasi-projective, 34
weight
dominant, 124
fundamental dominant, 127
integral, 126
restricted, 137
weight of a representation, 122
highest, 125
restricted, 137
weight space, 122
weights
of the adjoint representation, 122
Weyl group, 51
affine, 285
of a BN-pair, 91
of a root system, 64, 269
Witt index, 236
Zariski topology, 3
Zsigmondy prime divisor, 247