Rep N Control

Finite Group Representations
for the
Pure Mathematician
by
Peter Webb
Preface
This book is a teaching tool intended for students who are studying finite group
representation theory beyond the level of a first course in character theory. It started as
notes for courses given at the second-year graduate level at the University of Minnesota
and it is appropriate for a year-long course at this level.
Most of the students who attend such a course do not go on to be specialists in group
representation theory (for otherwise the audience would be much smaller!) but instead
go on to pursue interests in other areas of pure mathematics, which might perhaps be
combinatorics, topology, number theory or commutative algebra. These students need a
solid, well-balanced grounding in representation theory that enables them to apply the
theory to their own situation when the occasion demands it. It is probably irrelevant for
them to be presented with overly deep material whose main function - at least at present
- is to serve the internal workings of the subject. Thus topics have been selected and in
some instances standard results have been omitted (sometimes just the proofs) where they
appear to me not to have applications. This book is written primarily for those students.
Their needs have provided the main criterion for whether or not to include certain material,
and I have also tried to aim the level of exposition at them. The explanations are intended
to be full, but not overly lengthy. For the students who go on to be specialists in the area
there is no shortage of more advanced texts - but they may find it helpful to start with
this one!
One feature of the treatment is that wherever possible we develop the subject over
an arbitrary ring. The philosophy behind this is that representations over many different
rings do arise in practice in pure mathematics, and we need to be able to handle them.
These rings might be finite fields arising in the context of homology groups of spaces, or
they might be local rings arising from questions in commutative algebra, or they might be
rings of algebraic integers. It is my intention that the reader should come away from this
book no less confident in characteristic p than in characteristic 0. The questions that arise
in positive characteristic or over rings of integers are genuinely harder than over fields of
characteristic zero, but at least the reader should come away with some idea of what can
be done and how to do it, and what cannot be done in this context.
The exercises at the ends of sections are an important part of the book. The benefit
of learning actively by having to apply the theory to calculate with examples and solve
problems cannot be underestimated. I have tried to provide a good number of exercises:
some of them are easy and some are more challenging. In some instances I use the exercises
as a place to present extensions of results that appear in the text. Sometimes the exercises
indicate how the subject may be developed further.
It is supposed that the reader has already studied the material in a first course in
algebra and is familiar with the properties of groups, rings, field extensions and with linear
algebra. More specifically the reader should know about Sylow subgroups, solvable and
nilpotent groups, as well as the examples that are introduced in a first group theory course,
such as the dihedral, symmetric, alternating and quaternion groups. The reader should
also be familiar with tensor products, Noetherian properties of commutative rings, the
structure of modules over a principal ideal domain, and the first properties of ideals, as
well as Jordan and rational canonical forms for matrices.
Many people have read sections of this book, worked through the exercises and made
important comments and I wish to thank them all. These include Cihan Bahran, Dave
Benson, Daniel Hess, John Palmieri, Sverre Smaland many others.
Table of Contents
1
2
3
4
5
6
7
8
9
10
11
12
A
B
Representations, Maschkes theorem and semisimplicity

The structure of algebras for which every module is semisimple
Characters
The construction of modules and characters
Induction and restriction
Symmetric and Exterior powers
The construction of character tables
More on induction and restriction: theorems of Mackey and Clifford
Representations of p-groups in characteristic p and the radical
Projective modules for finite-dimensional algebras
Projective modules for group algebras
Changing the ground ring: splitting fields and the decomposition map
Brauer characters
Indecomposable modules: vertices, sources and Green correspondence
Blocks
Appendix: Discrete valuation rings
Appendix: Some character tables
1. Representations, Maschkes Theorem and Semisimplicity

In this section we present some basic definitions and examples to do with group representations. We then prove Maschkes theorem, which states that in many circumstances
representations are completely reducible. We conclude by describing the properties of
semisimple modules.
Definitions and examples

Informally, a representation of a group is a collection of invertible linear transformations of a vector space (or, more generally, of a module for a ring) that multiply together
in the same way as the group elements. The collection of linear transformations thus establishes a pattern of symmetry of the vector space which copies the symmetry encoded
by the group. Because symmetry is observed and understood so widely, and is even one of
the fundamental notions of mathematics, there are applications of representation theory
across the whole of mathematics, as well as in other disciplines.
For many applications, especially those having to do with the natural world, it is
appropriate to consider representations over fields of characteristic zero such as C, R or Q
(the fields of complex numbers, real numbers or rational numbers). In other situations, that
might arise in topology or combinatorics or number theory for instance, we find ourselves
considering representations over fields of positive characteristic such as the field with p
elements Fp , or over rings that are not fields such as the ring of integers Z. Many aspects
of representation theory do change as the ring varies, but there are also parts of the theory
which are similar regardless of the field characteristic, or even if the ring is not a field. We
will develop the theory independently of the choice ring where possible so as to be able to
apply it in all situations and to establish a natural context for the results.
Let G denote a finite group, and let R be a commutative ring with a 1. If V is an Rmodule we denote by GL(V ) the group of all invertible R-module homomorphisms V V .
In case V
= Rn is a free module of rank n this group is isomorphic to the group of all
non-singular n n matrices over R, and we denote it by GL(n, R) or GLn (R), or in case
R = Fq is the finite field with q elements by GL(n, q) or GLn (q). We point out also that
unless otherwise stated, modules will be left modules and morphisms will be composed
reading from right to left, so that matrices in GL(n, R) are thought of as acting from the
left on column vectors.
A (linear) representation of G (over R) is a group homomorphism
: G GL(V ).
In a situation where V is free as an R-module, on taking a basis for V we may write
each element of GL(V ) as a matrix with entries in R and we obtain for each g G a
matrix (g). These matrices multiply together in the manner of the group and we have
Printed Oct. 13, 2014
a matrix representation of G. In this situation the rank of the free R-module V is called
the degree of the representation. Sometimes by abuse of terminology the module V is also
called the representation, but it should more properly be called the representation module
or representation space (if R is a field).
To illustrate some of the possibilities that may arise we consider some examples.
(1.1) Examples. 1. For any group G and commutative ring R we can take V = R and
(g) = 1 for all g G, where 1 denotes the identify map R R. This representation is
called the trivial representation, and it is often denoted simply by its representation module
R. Although this representation turns out to be extremely important in the theory, it does
not at this point give much insight into the nature of a representation.
2. A representation on a space V = R of rank 1 is in general determined by specifying
a homomorphism G R . Here R is the group of units of R, and it is isomorphic to
GL(V ). For example, if G = hgi is cyclic of order n and k = C is the field of complex
numbers, there are n possible such homomorphisms, determined by g 7 e
2ri
n
where
0 r n 1. Another important example of a degree 1 representation is the sign

representation of the symmetric group Sn on n symbols, given by the group homomorphism
which assigns to each permutation its sign, regarded as an element of the arbitrary ring R.
3. Let R = R, V = R2 and G = S3 . This group G is isomorphic to the group of
symmetries of an equilateral triangle. The symmetries are the three reflections in the lines
that bisect the equilateral triangle, together with three rotations.
Positioning the triangle at the origin of V and labelling the three vertices of the triangle
Peter Webb
as 1, 2 and 3 we get a representation

() 7
(1, 2) 7
(1, 3) 7
(2, 3) 7
(1, 2, 3) 7
(1, 3, 2) 7
1 0
0 1
1
1
0 1
1 0

0
1
1 1
0 1
1
1
0 1
1 1
1
0
where we have taken basis vectors in the directions of vertices 1 and 2, making an angle of
2
3 to each other. In fact these matrices define a representation of degree 2 over any ring
R, because although the representation was initially constructed over R the matrices have
integer entries, and these may be interpreted in every ring. No matter what the ring is,
the matrices always multiply together to give a copy of S3 .
4. Let R = Fp , V = R2 and let G = Cp = hgi be cyclic of order p generated by an
element g. We see that the assignment
r
(g ) =
1 0
r 1
is a representation. In this case the fact that we have a representation is very much
dependent on the choice of R as the field Fp : in any other characteristic it would not work,
because the matrix shown would no longer have order p.
We can think of representations in various ways. One of them is that a representation
is the specification of an action of a group on an R-module, as we now explain. Given
a representation : G GL(V ), an element v V and a group element g G we get
another module element (g)(v). Sometimes we write just g v or gv for this element. This
rule for multiplication satisfies
g (v + w) = g v + g w
(gh) v = g (h v)
1v =v
for all g G, v, w V and , R. A rule for multiplication G V V satisfying

these conditions is called a linear action of G on V . To specify a linear action of G on V
is the same thing as specifying a representation of G on V , since given a representation

we obtain a linear action as indicated above, and evidently given a linear action we may
recover the representation.
Another way to define a representation of a group is in terms of the group algebra.
We define the group algebra RG (or R[G]) of G over R to be the free R-module with
the elements of G as an R-basis, and with multiplication given on the basis elements by
group multiplication. The elements of RG are the (formal) R-linear combinations of group
elements, and the multiplication of the basis elements is extended to arbitrary elements
P
using bilinearity of the operation. A typical element of RG may be written gG ag g
where ag R, and symbolically
(
gG
ag g)(
hG
bh h) =
X X
(
ag bh )k.
kG gh=k
More concretely, we may exemplify the definition by listing some elements of QS3 . The
elements of S3 such as (1, 2) = 1 (1, 2) are also elements of QS3 (they appear as basis
elements), and () serves as the identity element of QS3 (as well as of S3 ). In general,
elements of QS3 may look like (1, 2) (2, 3) or 51 (1, 2, 3) + 6(1, 2) 71 (2, 3). Here is a
computation:
((1, 2, 3) + (1, 2))((1, 2) (2, 3)) = (1, 3) + () (1, 2) (1, 2, 3).
Having defined the group algebra, we may now define a representation of G over
R to be a unital RG-module. The fact that this definition coincides with the previous
ones is the content of the next proposition. Throughout this text we may refer to group
representations as modules (for the group algebra).
(1.2) PROPOSITION. A representation of G over R has the structure of a unital
RG-module; conversely, every unital RG-module provides a representation of G over R.
Proof. Given a representation : G GL(V ) we define a module action of RG on V
P
P
by ( ag g)v =
ag (g)(v).
Given a RG-module V , the linear map (g) : v 7 gv is an automorphism of V and
(g1 )(g2 ) = (g1 g2 ) so : G GL(V ) is a representation.
We have defined the group algebra without saying what an algebra is! For the record,
an (associative) R-algebra is a ring A with a 1, equipped with a (unital) ring homomorphism
R A whose image lies in the center of A. The group algebra RG is indeed an example
of an R-algebra.
The group algebra gives another example of a representation, called the regular representation. In fact for any ring A we may regard A itself as a left A-module with the
action of A on itself given by multiplication of the elements. We denote this left A-module
by A A when we wish to emphasize the module structure, and this is the (left) regular
Peter Webb
representation of A. When A = RG we may describe the action on RG RG by observing

that each element g G acts on RG RG by permuting the basis elements in the fashion
g h = gh. Thus each g acts by a permutation matrix, namely a matrix in which in every
row and column there is precisely one non-zero entry, and that non-zero entry is 1. The
regular representation is an example of a permutation representation, namely one in which
every group element acts by a permutation matrix.
Regarding representations of G as RG-modules has the advantage that many definitions we wish to make may be borrowed from module theory. Thus we may study
RG-submodules of an RG-module V , and if we wish we may call them subrepresentations
of the representation afforded by V . To specify an RG-submodule of V it is necessary to
specify an R-submodule W of V that is closed under the action of RG. This is equivalent
to requiring that (g)w W for all g G and w W . We say that a submodule W
satisfying this condition is stable under G, or that it is an invariant submodule or invariant
subspace (if R happens to be a field). Such an invariant submodule W gives rise to a
homomorphism W : G GL(W ) that is the subrepresentation afforded by W .
(1.3) Examples. 1. Let C2 = {1, 1} be cyclic of order 2 and consider the representation
: C2 GL(R2 )

1 0
1 7
0 1

1 0
1 7
0 1

There are just four invariant subspaces,
namely {0}, h 10 i, h 01 i, R2 and no others. The

representation space R2 = h 10 i h 01 i is the direct sum of two invariant subspaces.

2. In Example 4 of 1.1 above, an elementary calculation shows that h 01 i is the only
1-dimensional invariant subspace, and so it is not possible to write the representation space
V as the direct sum of two non-zero invariant subspaces.
We make use of the notions of a homomorphism and an isomorphism of RG-modules.
Since RG has as a basis the elements of G, to check that an R-linear homomorphism f :
V W is in fact a homomorphism of RG-modules, it suffices to check that f (gv) = gf (v)
for all g G we do not need to check for every x RG. By means of the identification
of RG-modules with representations of G (in the first definition given here) we may refer to
homomorphisms and isomorphisms of group representations. In many books the algebraic
condition on the representations that these notions entail is written out explicitly, and two
representations that are isomorphic are also said to be equivalent.
If V and W are RG-modules then we may form their (external) direct sum V W ,
which is the same as the direct sum of V and W as R-modules together with an action
of G given by g(v, w) = (gv, gw). We also have the notion of the internal direct sum
of RG-modules and write U = V W to mean that U has RG-submodules V and W
satisfying U = V + W and V W = 0. In this situation we also say that V and W are
direct summands of U . We already met this property in (1.3) above, whose first part is
an example of a representation that is a direct sum of two non-zero subspaces; however,

the second part of (1.3) provides an example of a subrepresentation that is not a direct
summand.
Semisimple representations
We come now to our first non-trivial result, and one that is fundamental to the study
of representations over fields of characteristic zero, or characteristic not dividing the group
order. The result says that in this situation representations fall apart remarkably easily,
so that they can be understood by taking direct sums of smaller representations. We do
now require the ring R to be a field, and in this situation we will often use the symbols F
or k instead of R.
(1.4) THEOREM (Maschke). Let V be a representation of the finite group G over a
field F in which |G| is invertible. Let W be an invariant subspace of V . Then there exists
an invariant subspace W1 of V such that V = W W1 as representations.
Proof. Let : V W be any projection of V onto W as vector spaces, i.e. a linear
transformation such that (w) = w for all w W . Since F is a field, we may always
find such a projection by finding a vector space complement to W in V , and projecting
off the complementary factor. Then V = W Ker() as vector spaces, but Ker() is not
necessarily invariant under G. Consider the map
=
1 X
gg 1 : V V.
|G|
gG
Then is linear and if w W then

(w) =
1 X
g(g 1w)
|G|
gG
1 X 1
gg w
|G|
gG
1
|G|w
|G|
= w.
Since furthermore (v) W for all v V , is a projection onto W and so V =

W Ker( ). We show finally that Ker( ) is an invariant subspace by verifying that
Peter Webb
is an F G-module homomorphism: if h G and v V then

(hv) =
1 X
g(g 1hv)
|G|
gG
1 X
h(h1 g)((h1 g)1 v)
|G|
gG
= h (v)
since as g ranges over the elements of G, so does h1 g. Now if v Ker( ) then hv

Ker( ) also (since (hv) = h (v) = 0) and so Ker( ) is an invariant subspace.
Because the next results apply more generally than to group representations we let
A be a ring with a 1 and consider its modules. A non-zero A-module V is said to be
simple or irreducible if V has no A-submodules other than 0 and V . We see immediately
that a non-zero module is simple if and only if it is generated by every non-zero vector it
contains. Furthermore, the simple A-modules are exactly those of the form A/I for some
maximal left ideal I of A: every such module is simple, and given a simple module S with
a non-zero element x S the A-module homomorphism A S specified by a 7 ax is
surjective with kernel a maximal ideal I, so that S
= A/I. Since all simple modules appear
inside A in this way, we may deduce that if A is a finite dimensional algebra over a field
there are only finitely many isomorphism types of simple modules, these appearing among
the composition factors of A when regarded as a module.
A module that is the direct sum of simple submodules is said to be semisimple or
completely reducible. We saw in (1.3) two examples of modules, one of which was semisimple
and the other of which was not. Every module of finite composition length is somehow built
up out of its composition factors, which are simple modules, and we know from the Jordan
Holder theorem that these composition factors are determined up to isomorphism, although
there may be many composition series. The most rudimentary way these composition
factors may be fitted together is as a direct sum, giving a semisimple module. In this case
the simple summands are the composition factors of the module and their isomorphism
types and multiplicities are uniquely determined. There may, however, be many ways to
find simple submodules of a semisimple module so that the module is their direct sum.
We will now relate the property of semisimplicity to the property that appears in
Maschkes theorem, namely that every submodule of a module is a direct summand. Our
immediate application of this will be an interpretation of Maschkes theorem, but the
results have application in greater generality in situations where R is not a field, or when
|G| is not invertible in R. To simplify the exposition we have imposed a finiteness condition
in the statement of each result, thereby avoiding arguments that use Zorns lemma. These
finiteness conditions can be removed, and we leave the details to an exercise at the end of
this chapter.
In the special case when the ring A is a field and A-modules are vector spaces the
next result is familiar from linear algebra.
(1.5) LEMMA. Let A be a ring with a 1 and suppose that U = S1 + + Sn is

an A-module that can be written as the sum of finitely many simple modules S1 , . . . , Sn .
If V is any submodule of U there is a subset I = {i1 , . . . , ir } of {1, . . . , n} such that
U = V Si1 Sir . In particular,
(1) V is a direct summand of U , and
(2) (taking V = 0), U is the direct sum of some subset of the Si , and hence is necessarily
semisimple.
Proof. Choose a subset I of {1, . . . , n} maximal subject to the condition that the sum
L
W = V ( iI Si ) is a direct sum. Note that I = has this property, so we are indeed
taking a maximal element of a non-empty collection of subsets. We show that W = U .
If W 6= U then Sj 6 W for some j. Now Sj W = 0, being a proper submodule of Sj ,
so Sj + W = Sj W and we obtain a contradiction to the maximality of I. Therefore
W = U . The consequences (1) and (2) are immediate.
(1.6) PROPOSITION. Let A be a ring with a 1 and let U be an A-module. The
following are equivalent.
(1) U can be expressed as a direct sum of finitely many simple A-submodules.
(2) U can be expressed as a sum of finitely many simple A-submodules.
(3) U has finite composition length and has the property that every submodule of U is a
direct summand of U .
When these three conditions hold, every submodule of U and every factor module of U
may also be expressed as the direct sum of finitely many simple modules.
Proof. The implication (1) (2) is immediate and the implications (2) (1) and
(2) (3) follow from Lemma 1.5. To show that (3) (1) we argue by induction on the
composition length of U , and first observe that hypothesis (3) passes to submodules of U .
For if V is a submodule of U and W is a submodule of V then U = W X for some
submodule X, and now V = W (X V ) by the modular law (exercise 1.2). Proceeding
with the induction argument, when U has length 1 it is a simple module, and so the
induction starts. If U has length greater than 1, it has a submodule V and by condition
(3), U = V W for some submodule W . Now both V and W inherit condition (3) and
are of shorter length, so by induction they are direct sums of simple modules and hence
so is U .
We have already observed that every submodule of U inherits condition (3), and so
satisfies condition (1) also. Every factor module of U has the form U/V for some submodule
V of U . If condition (3) holds then U = V W for some submodule W that we have just
observed satisfies condition (1), and hence so does U/V since U/V
= W.
Peter Webb
We now present a different version of Maschkes theorem. The assertion remains

correct if the words finite-dimensional are removed from it, but we leave the proof of this
to the exercises.
(1.7) COROLLARY. Let R be a field in which |G| is invertible. Then every finitedimensional RG-module is semisimple.
Proof. This combines Theorem 1.4 with the equivalence of the statements of Proposition 1.6.
This theorem puts us in very good shape if we want to know about the representations
of a finite group over a field in which |G| is invertible for example any field of characteristic zero. To obtain a description of all possible finite-dimensional representations we
need only describe the simple ones, and then arbitrary ones are direct sums of these.
The following corollaries to Lemma 1.5 will be used on many occasions when we are
considering modules that are not semisimple.
(1.8) COROLLARY. Let A be a ring with a 1, and let U be an A-module of finite
composition length.
(1) The sum of all the simple submodules of U is a semisimple module, that is the unique
largest semisimple submodule of U .
(2) The sum of all submodules of U isomorphic to some given simple module S is a
submodule isomorphic to a direct sum of copies of S. It is the unique largest submodule
of U with this property.
Proof. The submodules described can be expressed as the sum of finitely many submodules by the finiteness condition on U . They are the unique largest submodules with
their respective properties since they contain all simple submodules (in case (1)), and all
submodules isomorphic to S (in case (2)).
The largest semisimple submodule of a module U is called the socle of U , and is denoted
Soc(U ). There is a dual construction called the radical of U , denoted Rad U , that we will
study in Chapter 6. It is defined to be the intersection of all the maximal submodules of
U , and has the property that it is the smallest submodule of U with semisimple quotient.
(1.9) COROLLARY. Let U = S1a1 Srar be a semisimple module over a ring
A with a 1, where the Si are non-isomorphic simple A-modules and the ai are their multiplicities as summands of U . Then each submodule Siai is uniquely determined and is
characterized as the unique largest submodule of U expressible as a direct sum of copies
of Si .
Proof. It suffices to show that Siai contains every submodule of U isomorphic to Si . If
T is any non-zero submodule of U not contained in Siai then for some j 6= i its projection
10
to a summand Sj must be non-zero. If we assume that T is simple this projection will be

an isomorphism T
= Sj . Thus all simple submodules isomorphic to Si are contained in
ai
the summand Si .
Summary of Chapter 1
1. Representations of G over R are the same thing as RG-modules.
2. Semisimple modules may be characterized in several different ways. They are
modules that are the direct sum of simple modules, or equivalently the sum of simple
modules, or equivalently modules for which every submodule is a direct summand.
3. If F is a field in which G is invertible, F G-modules are semisimple.
4. The sum of all simple submodules of a module is the unique largest semisimple
submodule of that module: the socle.
Exercises for Chapter 1

1. In Example 1 of 1.3 prove that there are no invariant subspaces other than the
ones listed.
2. (The modular law.) Let A be a ring and U = V W an A-module that is the
direct sum of A-modules V and W . Show by example that if X is any submodule of U
then it need not be the case that X = (V X) (W X). Show that if we make the
assumption that V X then it is true that X = (V X) (W X).
3. Suppose that is a finite-dimensional representation of a finite group G over C.
Show that for each g G the matrix (g) is diagonalizable.
4. Let
1 : G GL(V )
2 : G GL(V )
be two representations of G on the same R-module V that are injective as homomorphisms.

(We say that such a representation is faithful.) Consider the three properties
(a) the RG-modules given by 1 and 2 are isomorphic,
(b) the subgroups 1 (G) and 2 (G) are conjugate in GL(V ),
(c) for some automorphism Aut(G) the representations 1 and 2 are isomorphic.
Show that (a) (b) and that (b) (c). Show also that if Aut(G) is an inner
automorphism (i.e. one of the form conjugation by g for some g G) then 1 and 1
are isomorphic.
5. One form of the JordanZassenhaus theorem asserts that for each n, GL(n, Z)
(that is, Aut(Zn )) has only finitely many conjugacy classes of subgroups of finite order.
Peter Webb
11
Assuming this, show that for each finite group G and each integer n there are only finitely
many isomorphism classes of representations of G on Zn .
6. (a) Write out a proof of Maschkes theorem in the case of representations over C
along the following lines.
Given a representation : G GL(V ) where V is a vector space over C, let ( , ) be any
positive definite Hermitian form on V . Define a new form ( , )1 on V by
(v, w)1 =
1 X
(gv, gw).
|G|
gG
Show that ( , )1 is a positive definite Hermitian form, preserved under the action of G,
i.e. (v, w)1 = (gv, gw)1 always.
If W is a subrepresentation of V , show that V = W W as representations.
(b) Show that any finite subgroup of GL(n, C) is conjugate to a subgroup of U (n, C)
(the unitary group, consisting of n n complex matrices A satisfying AAT = I). Show
that any finite subgroup of GL(n, R) is conjugate to a subgroup of O(n, R) (the orthogonal
group consisting of n n real matrices A satisfying AAT = I).
7. Let U = S1 Sr be an A-module that is the direct sum of finitely many simple
modules S1 , . . . , Sr . Show that if T is any simple submodule of U then T
= Si for some i.
8. Let : U V be a homomorphism of A-modules. Show that : (Soc U ) Soc V ,
and that if is an isomorphism
then restricts to an isomorphism Soc U Soc V .
2
2

9. Let G = hx, y x = y = 1 = [x, y]i be the Klein four-group, R = F2 , and consider
the two representations 1 and 2 specified on the generators of G by
1 1 0
1 0 1
1 (x) = 0 1 0 , 1 (y) = 0 1 0
0 0 1
0 0 1
and
2 (x) = 0
0
0 0
1 1,
0 1
2 (y) = 0
0
0 1
1 0.
0 1
Calculate the socles of these two representations. Show that

semisimple.
10. Let G = Cp = hxi and R = Fp for some prime p
representations 1 and 2 specified by
1 1 0
1 1
1 (x) = 0 1 1
and 2 (x) = 0 1
0 0 1
0 0
neither representation is
3. Consider the two
1
0.
1
Calculate the socles of these two representations and show that neither representation is
semisimple. Show that the second representation is nevertheless the direct sum of two
non-zero subrepresentations.
12
11. (a) Using 1.6 show that if A is a ring for which the regular representation A A is
semisimple, then every finitely generated A-module is semisimple.
(b) Extend the result of part (a), using Zorns lemma, to show that if A is a ring for
which the regular representation A A is semisimple, then every A-module is semisimple.
12. Let U be a module for a ring A with a 1. Show that the following three statements
are equivalent.
(1) U is a direct sum of simple A-submodules.
(2) U is a sum of simple A-submodules.
(3) every submodule of U is a direct summand of U .
[Use Zorns lemma to prove a version of Lemma 1.5 that has no finiteness hypothesis and
then copy Proposition 1.6. This deals with all implications except (3) (2). For that, use
the fact that A has a 1 and hence every (left) ideal is contained in a maximal (left) ideal,
combined with condition (3), to show that every submodule of U has a simple submodule.
Consider the sum of all simple submodules of U and show that it equals U .]
13. Let RG be the group algebra of a finite group G over a commutative ring R with
1. Let S be a simple RG-module and let I be the anihilator in R of S, that is

I = {r R rx = 0 for all x S}.
Show that I is a maximal ideal in R.

[This question requires some familiarity with standard commutative algebra. We conclude
from this result that when considering simple RG modules we may reasonably assume that
R is a field, since S may naturally be regarded as an (R/I)G-module and R/I is a field.]
14. Let k be an infinite field of characteristic 2, and G = hx, yi
= C2 C2 be the
non-cyclic group of order 4. For each k let (x), (y) be the matrices
(x) = ( 1 0cr1 1 ) ,
(y) =
0
1
regarded as linear maps U U where U is a k-vector space of dimension 2 with basis

{e1 , e2 }.
(a) Show that defines a representations of G with representation space U .
(b) Find a basis for Soc U .
(c) By considering the effect on Soc
U , show
that any kG-module homomorphism
a 0
: U U has a triangular matrix =
.
b c
(d) Show that if U
= U as kG-modules then = . Deduce that kG has infinitely
many non-isomorphic 2-dimensional representations.
15. Let V be an A-module for some ring A and suppose that V is a sum V =
V1 + +Vn of simple submodules. Assume further that the Vi are pairwise non-isomorphic.
Show that the Vi are the only simple submodules of V and that V = V1 Vn is their
direct sum.
Peter Webb
13
2. The structure of algebras for which every module is semisimple.

In this chapter we present the ArtinWedderburn structure theorem for semisimple
algebras and its immediate consequences. This theorem is the ring-theoretic manifestation
of the module-theoretic hypothesis of semisimplicity that was introduced in Section 1, and
it shows that the kind of algebras that can arise when all modules are semisimple is very
restricted. The theorem applies to group algebras over a field in which the group order is
invertible (as a consequence of Maschkes theorem), but since the result holds in greater
generality we will assume we are working with a finite-dimensional algebra A over a field
k.
The first result is basic, and will be used time and time again.
(2.1) THEOREM (Schurs Lemma). Let A be a finite dimensional algebra over a
field k and let S1 and S2 be simple A-modules. Then HomA (S1 , S2 ) = 0 unless S1
= S2 ,
in which case the endomorphism ring EndA (S1 ) is a division ring. If k is algebraically
closed, then every A-module endomorphism of S1 is multiplication by some scalar. Thus
EndA (S1 )
= k in this case.
Proof. Suppose : S1 S2 is a non-zero homomorphism. Then 0 6= (S1 ) S2 ,
so (S1 ) = S2 by simplicity of S2 and we see that is surjective. Thus Ker 6= S1 , so
Ker = 0 by simplicity of S1 , and is injective. Therefore is invertible, S1
= S2 and
EndA (S1 ) is a division ring.
If k is algebraically closed, let be an eigenvalue of . Now ( I) : S1 S1 is a
singular endomorphism of A-modules, so I = 0 and = I.
In many results that follow we will make the assumption that k is algebraically closed,
and this is invariably to make sure that division rings EndA (S) that are larger than k do not
arise. Sometimes this is a significant point, but very often it is no more than a convenience
to simplify the formulation of results so that numbers such as dim End(S) do not appear.
In these cases we try to indicate the general result as well as the simplified version. We
comment also that k may have the property that EndA (S) = k for all simple A-modules
S without being algebraically closed. If the field k has this property it is called a splitting
field for A. The theory of splitting fields will be developed in Chapter 9, but for the
moment it suffices know that algebraically closed fields are always splitting fields. In many
of the results that follow, the hypothesis that k is algebraically closed can be replaced by
the requirement that k is a splitting field.
The next result is straightforward and seemingly innocuous, but it has an important
consequence for representation theory. It is the main tool in recovering the structure of an
algebra from its representations. We use the notation Aop to denote the opposite ring of
A, namely the ring that has the same set and the same addition as A, but in which there
is a new multiplication given by a b = ba.
14
(2.2) LEMMA. For any ring A with a 1, EndA (A A)

= Aop .
Proof. The inverse isomorphisms are
7 (1)
(a 7 ax) x.
There are several things here that need to be checked: that the second assignment does
take values in EndA (A A), that the morphisms are ring homomorphisms, and that they are
mutually inverse. We leave most of this to the reader, observing only that under the first
homomorphism a composite is sent to ()(1) = ((1)) = ((1)1) = (1)(1), so that
this is indeed a homomorphism to Aop .
Observe that the proof of Lemma 2.2 establishes that every endomorphism of the
regular representation is of the form right multiplication by some element.
A ring A with 1 all of whose modules are semisimple is itself called semisimple. By
Exercise 11 of Chapter 1 it is equivalent to suppose that the regular representation A A is
semisimple. It is also equivalent if A is a finite dimensional algebra over a field to suppose
that the Jacobson radical of the ring is zero, but we will not deal with this point of view
until Chapter 6.
(2.3) THEOREM (ArtinWedderburn). Let A be a finite-dimensional algebra over
a field k with the property that every finite-dimensional module is semisimple. Then A is
a direct sum of matrix algebras over division rings. Specifically, if
AA
= S1n1 Srnr
where the S1 , . . . , Sr are non-isomorphic simple modules occuring with multiplicities n1 , . . . , nr

in the regular representation, then
A
= Mn1 (D1 ) Mnr (Dr )
where Di = EndA (Si )op . Furthermore, if k is algebraically closed then Di = k for all i.
More is true: every such direct sum of matrix algebras is a semisimple algebra. Each
matrix algebra over a division ring is a simple algebra (namely one that has no 2-sided
ideals apart from the zero ideal and the whole ring), and it has up to isomorphism a
unique simple module (see the exercises). Furthermore, the matrix algebra summands are
uniquely determined as subsets of A (although the module decomposition of A A is usually
only determined up to isomorphism). The uniqueness of the summands will be established
in Proposition 3.22.
Proof. We first observe that if we have a direct sum decomposition U = U1 Ur
of a module U then EndA (U ) is isomorphic to the algebra of r r matrices in which
Peter Webb
15
the i, j entries lie in HomA (Uj , Ui ). This is because any endomorphism : U U

may be writen as a matrix of components = (ij ) where ij : Uj Ui , and when
viewed in this way endomorphisms compose in the manner of matrix multiplication. Since
n
HomA (Sj j , Sini ) = 0 if i 6= j by Schurs lemma, the decomposition of A A shows that
n
EndA (A A)
= EndA (S1 1 ) EndA (Srnr )
op
op
and furthermore EndA (Sini )
= Mni (Di ). Evidently Mni (Di )op
= Mni (Di ) and by
Lemma 2.2 we identify EndA (A A) as Aop . Putting these pieces together gives the matrix
algebra decomposition. Finally, if k is algebraically closed it is part of Schurs lemma that
Di = k for all i.
(2.4) COROLLARY. Let A be a finite-dimensional semisimple algebra over a field k.

In any decomposition
AA
= S1n1 Srnr
where the Si are pairwise non-isomorphic simple modules we have that S1 , . . . , Sr is a

complete set of representatives of the isomorphism classes of simple A-modules. When k
is algebraically closed ni = dimk Si and dimk A = n21 + + n2r .
Proof. All isomorphism types of simple modules must appear in the decomposition
because every simple module can be expressed as a homomorphic image of A A (as observed
at the start of this chapter), and so must be a homomorphic image of one of the modules
Si . When k is algebraically closed all the division rings Di coincide with k by Schurs
lemma, and EndA (Sini )
= Mni (k). The ring decomposition A = Mn1 (k) Mnr (k) of
Theorem 2.3 immediately gives dimk A = n21 + + n2r .
We obtained this decomposition by identifying A with End(A A)op in such a way
that an element a A is identified with the endomorphism right multiplication by a,
n
by Lemma 2.2. From this we see that right multiplication of an element of Sj j by an
element of Mni (k) is 0 if i 6= j, and hence Sini is the unique summand of A (in the initial
decomposition of A) containing elements on which Mni (k) acts in a non-zero fashion from
the right. We deduce that Mni (k)
= Sini as left A-modules, since the term on the left
is isomorphic to the quotient of A by the left submodule consisting of elements that the
summand Mni (k) annihilates by right multiplication, the term on the right is an image of
P
this quotient, and in order to have dimk A = i dimk Sini they must be isomorphic. Hence
dimk Mni (k) = n2i = dimk Sini = ni dim Si ,
and so dim Si = ni .
16
Let us now restate what we have proved specifically in the context of group representations.
(2.5) COROLLARY. Let G be a finite group and k a field in which |G| is invertible.
(1) As a ring, kG is a direct sum of matrix algebras over division rings.
(2) Suppose in addition that k is algebraically closed. Let S1 , . . . , Sr be pairwise nonisomorphic simple kG-modules and let di = dimk Si be the degree of Si . Then di
equals the multiplicity with which Si is a summand of the regular representation of
G, and |G| d21 + + d2r with equality if and only if S1 , . . . , Sr is a complete set of
representatives of the simple kG-modules.
Proof. This follows from Maschkes Theorem 1.4, the ArtinWedderburn Theorem
2.3 and Corollary 2.4.
Part (2) of this result provides a numerical criterion that enables us to say when we
have constructed all the simple modules of a group over an algebraically closed field k in
P
which |G| is invertible: we check that d2i = |G|. While this is an easy condition to verify,
it will be superseded later on by the even more straightforward criterion that the number
of simple kG-modules (with the same hypotheses on k) equals the number of conjugacy
P 2
classes of elements of G. Once we have proved this, the formula
di = |G| allows the
degree of the last simple representation to be determined once the others are known.
(2.6) Example. Over R we have constructed for S3 the trivial representation, the sign
representation that is also of dimension 1 but not the same as the trivial representation, and
a 2-dimensional representation that is simple because visibly no 1-dimensional subspace of
the plane is invariant under the group action. Since 12 +12 +22 = |S3 | we have constructed
all the simple representations.
At this point we make a deduction about representations of finite abelian groups.
Looking ahead to later results, we will obtain a partial converse of this in Theorem 4.4,
and in Theorem 5.10 we will obtain an extension to fields in which |G| is not invertible.
A more detailed description of representations of abelian groups when |G| is not invertible
follows from Example 8.6.
(2.7) COROLLARY. Let G be a finite abelian group. Over an algebraically closed
field k in which |G| is invertible, every simple representation of G has degree 1 and the
number of non-isomorphic simple representations equals |G|. In particular we may deduce
that every invertible matrix of finite order, with order relatively prime to the characteristic
of k, is diagonalizable.
Proof. We know that kG is semisimple, and because kG is a commutative ring the
matrix summands that appear in Theorem 2.3 must all have size 1, and the division rings
that appear must be commutative. In fact, since we have supposed that k is algebraically
closed, the division rings must all be k. This means that the degrees of the irreducible
Peter Webb
17
representations are all 1, and so the number of them must be |G| since this is the sum of
the degrees.
A matrix of finite order gives a representation of the cyclic group it generates on the
space of vectors on which the matrix acts and, by invertibility of the order, the representation is semisimple. It is a direct sum of 1-dimensional spaces by what we have just shown.
On choosing basis vectors to lie in these 1-dimensional spaces, the matrix is diagonal.
We do not have to exploit the theory we have developed to show that a matrix of finite,
invertible order is diagonalizable over an algebraically closed field. A different approach
is to consider its Jordan canonical form and observe that all Jordan blocks have size 1,
because blocks of size 2 or more have order that is either infinite or a multiple of the field
characteristic.
1. Endomorphism algebras of simple modules are division rings.
2. Semisimple algebras are direct sums of matrix algebras over division rings.
3. For a semisimple algebra over an algebraically closed field, the sum of the squares
of the degrees of the simple modules equals the dimension of the algebra.

1. Let A be a finite-dimensional semisimple algebra. Show that A has only finitely
many isomorphism types of modules in each dimension. [This is not in general true for
algebras that are not semisimple: we saw in Chapter 1 Exercise 14 that k[C2 C2 ] has
infinitely many non-isomorphic 2-dimensional representations when k is an infinite field of
characteristic 2.]
2. Using Exercises 4 from Chapter 1, Exercise 1 from Chapter 2 and Maschkes
theorem, show that if k is any field of characteristic 0 then for each natural number m,
GLn (k) has only finitely many conjugacy classes of subgroups of order m. [In view of the
comment to Exercise 1, the same is not true when when k = F2 .]
3. Let D be a division ring and n a natural number.
(a) Show that the natural Mn (D)-module consisting of column vectors of length n is
a simple module.
(b) Show that Mn (D) is semisimple and has up to isomorphism only one simple
module.
(c) Show that every algebra of the form
Mn1 (D1 ) Mnr (Dr )
18
is semisimple.
(d) Show that Mn (D) is a simple ring, namely one in which the only 2-sided ideals
are the zero ideal and the whole ring.
4. Prove the following extension of Corollary 2.4:
THEOREM. Let A be a finite-dimensional semisimple algebra, S a simple A-module
and D = EndA (S). Then S may be regarded as a module over D and the multiplicity of
S as a summand of A A equals dimD S.
5. Using the fact that Mn (k) has a unique simple module up to isomorphism, prove
the Noether-Skolem theorem: every algebra automorphism of Mn (k) is inner, i.e. of the
form conjugation by some invertible matrix.
6. Show that for any field k we have Mn (k)
= Mn (k)op , and in general for any division
ring D that given any positive integer n, Mn (D)
= Dop .
= Mn (D)op if and only if D
7. Let A be a ring with a 1, and let V be an A-module. An element e in any ring is
called idempotent if and only if e2 = e.
(a) Show that an endomorphism e : V V is a projection onto a subspace W if and
only if e is idempotent as an element of EndA (V ). (The term projection was defined at
the start of the proof of 1.4. It is a linear mapping onto a subspace that is the identity on
restriction to that subspace.)
(b) Show that direct sum decompositions V = W1 W2 as A-modules are in bijection
with expressions 1 = e + f in EndA (V ), where e and f are idempotent elements with
ef = f e = 0. (In case ef = f e = 0, e and f are called orthogonal.)
(c) A non-zero idempotent element e is called primitive if it cannot be expressed as a
sum of orthogonal idempotent elements in a non-trivial way. Show that e EndA (V ) is
primitive if and only if e(V ) has no (non-trivial) direct sum decomposition. (In this case
e(V ) is said to be indecomposable.)
(d) Suppose that V is semisimple with finitely many simple summands and let e1 , e2
EndA (V ) be idempotent elements. Show that e1 (V )
= e2 (V ) as A-modules if and only if
e1 and e2 are conjugate by an invertible element of EndA (V ) (i.e. there exists an invertible
A-endomorphism : V V such that e2 = e1 1 ).
(e) Let k be a field. Show that all primitive idempotent elements in Mn (k) are
conjugate under the action of the unit group GLn (k). Write down explicitly any primitive
idempotent element in M3 (k). (It may help to use Exercise 3.)
8. Prove the following theorem of Burnside: let G be a finite group, k an algebraically
closed field in which |G| is invertible, and let : G GL(V ) be a representation over
k. By taking a basis of V write each endomorphism (g) as a matrix. Let dim V = n.
Show that the representation is simple if and only if there exist n2 elements g1 , . . . , gn2 of
G so that the matrices (g1 ), . . . , (gn2 ) are linearly independent, and that this happens if
and only if the algebra homomorphism kG Endk (V ) is surjective. (Note that itself is
generally not surjective.)
Peter Webb
19
9. (We exploit results from a basic algebra course in our suggested approach to this
question.) Let G be a cyclic group of order n and k a field.
(a) By considering a homomorphism k[X] kG or otherwise, where k[X] is a polynomial ring, show that kG
= k[X]/(X n 1) as rings.
(b) Suppose that the characteristic of k does not divide n. Use the Chinese Remainder
Theorem and separability of X n 1 to show that when kG is expressed as a direct sum of
irreducible representations, no two of the summands are isomorphic, and that their degrees
are the same as the degrees of the irreducible factors of X n 1 in k[X]. Deduce, as a special
case of Corollary 2.7, that when k is algebraically closed all irreducible representations of
G have degree 1.
(c) When n is prime and k = Q, use irreducibility of X n1 + X n2 + X + 1 to
show that G has a simple module S of degree n 1, and that EndkG (S)
= Q(e2i/n ).
n1
(d) When k = R and n is odd show that G has 2 simple representations of degree
2 as well as the trivial representation of degree 1. When k = R and n is even show that G
has n2
simple representations of degree 2 as well as two simple representations of degree
2
1. If S is one of the simple representations of degree 2 show that EndkG (S) = C.
10. Let U be a module for a semisimple finite dimensional algebra A. Show that if
EndA (U ) is a division ring then U is simple.
11. Let H be the algebra of quaternions, that has a basis over R consisting of elements
1, i, j, k and multiplication determined by the relations
i2 = j 2 = k 2 = 1, ij = k, jk = i, ki = j, ji = k, kj = i, ik = j.
You may assume that H is a division ring. The elements {1, i, j, k} under multiplication form the quaternion group Q8 of order 8, and it acts on H by left multiplication, so
that H is a 4-dimensional representation of Q8 over R.
(a) Show that EndRQ8 (H)
= H, and that H is simple as a representation of Q8 over
R. [Consider the image of 1 H under an endomorphism.]
Lt
(b) In the decomposition RQ8 = i=1 Mni (Di ) predicted by Corollary 2.5, compute
the number of summands t, the numbers ni and the divisions rings Di . Show that RQ8
has no simple representation of dimension 2. [Observe that there are four homomorphisms
Q8 {1} R that give four 1-dimensional representations. Show that, together with
the representation of dimension 4, we have a complete set of simple representations.]
(c) The span over R of the elements 1, i H is a copy of the field of complex numbers
C, so that H contains C as a subfield. We may regard H as a vector space over C by letting
elements of C act as scalars on H by multiplication from the right. Show that with the
action of Q8 from the left and of C from the right, H becomes a left CQ8 -module. With
respect to the basis {1, j} for H over C, write down matrices for the action of the elements
i, j Q8 on H. Show that this 2-dimensional CQ8 -module is simple, and compute its
endomorphism ring EndCQ8 (H).
(d) Show that C R H
= M2 (C).
12. Let k be a field of characteristic 0 and suppose the simple kG-modules are
Pr
2
S1 , . . . , Sr with degrees di = dimk Si . Show that
i=1 di |G| with equality if and
only if EndkG (Si ) = k for all i.
20
3. Characters
Characters are an extremely important tool for handling the simple representations
of a group. In this chapter we will see them in the form that applies to representations
over a field of characteristic zero, and these are called ordinary characters. Since representations of finite groups in characteristic zero are semisimple, knowing about the simple
representations in some sense tells us about all representations. Later, in Chapter 10, we
will study characters associated to representations in positive characteristic, the so-called
Brauer characters.
Characters are very useful when we have some specific representation and wish to
compute its decomposition as a direct sum of simple representations. The information we
need to do this is contained in the character table of the group, which we introduce in
this chapter. We also establish many important theoretical properties of characters that
enable us to calculate them more easily and to check that our calculations are correct.
The most spectacular of these properties is the orthogonality relations, which may serve
to convince the reader that something extraordinary and fundamental is being studied.
We establish numerical properties of the character degrees, and a description of the center
of the group algebra that aids in decomposition the group algebra as a sum of matrix
algebras. This would be of little significance unless we could use characters to prove
something outside their own area. Aside from their use as a computational tool, we use
them to prove Burnsides pa q b theorem: every group whose order is divisible by only two
primes is solvable.
Peter Webb
21
The character table

Assume that : G GL(V ) is a finite-dimensional representation of G over the field
of complex numbers C or one of its subfields. We define the character of to be the
function : G C given by
(g) = tr((g)),
the trace of the linear map (g). For example, the 2-dimensional representation of S3
we considered in Chapter 1 has character given on the group elements by forming the
representing matrix and taking the trace as follows:

1 0
()
7
7 2
0 1
0 1
(1, 2) 7
7 0
1 0
1 0
(1, 3) 7
7 0
1
1

1 1
(2, 3) 7
7 0
0 1
0 1
(1, 2, 3) 7
7 1
1
1

1 1
(1, 3, 2) 7
7 1
1 0
We say that the representation and the representation space V afford the character ,
and we may write or V when we wish to specify this character more precisely. The
restriction to subfields of C is not significant: we will see in Chapter 9 that representations
of a finite group over a field characteristic 0 may always be realized over C.
In the next result we list some immediate properties of characters. The converse of
part (7) will be established later on in Corollary 3.9. As it is, part (7) provides a useful
way to show that representations are not isomorphic, by showing that their characters are
different.
(3.1) PROPOSITION. Let be the character of a representation of G over C and
let g, h G. Then
(1) (1) is the degree of , namely, the dimension of the representation space of ;
(2) if g has order n then (g) is a sum of nth roots of 1 (including roots whose order
divides n);
(3) |(g)| (1), with equality if and only if (g) is scalar multiplication;
(4) (g) = (1) if and only if (g) = 1, i.e. g lies in the kernel of ;
(5) (g 1 ) = (g), the complex conjugate;
(6) (hgh1 ) = (g);
22
(7) if V and W are isomorphic CG-modules then V = W as functions on G.

Proof. (1) is immediate because the identity of the group must act as the identity
matrix, and its trace is the degree of , since this is the dimension of the representation
space of .
(2) Recall from Corollary 2.7 or the comment after it that (g) is diagonalizable, so
that (g) is the sum of its eigenvalues 1 , . . . , d , where d is the degree of . (In fact, it
is sufficient for now to let 1 , . . . , d be the diagonal entries in the Jordan canonical form
of (g), without knowing that this form is diagonal.) These eigenvalues are roots of unity
since g has finite order, and the roots of unity have orders dividing n.
(3) Each root of unity has absolute value 1, so adding d of them and applying the
triangle inequality repeatedly we get |(g)| |1 | + + |d | = d. The only way we can
have equality is if 1 = = d so that (g) is scalar (since it is diagonalizable).
(4) If (g) = 1 then (g) = d. Conversely, if (g) = d then by part (3), (g) is
multiplication by a scalar . Now (g) = d = d, so = 1.
1
(5) If (g) has eigenvalues 1 , . . . , d then (g 1 ) has eigenvalues 1
1 , . . . , d , and
= i for each i since these are roots of unity. Thus (g 1 ) = 1 + + d = (g).
1
i
(6) This results from the fact that tr(ab) = tr(ba) for endomorphisms a and b, so that
(hgh1 ) = tr(hgh1 ) = tr((h)(g)(h1)) = tr(g) = (g).
(7) Suppose that V and W are the representations of G on V and W , and that we
have an isomorphism of CG-modules : V W . Then V (g) = W (g) for all g G,
so that W (g) = trW (g) = tr(V (g)1 ) = trV (g) = V (g).
As an application of Proposition 3.1 part (4), we see that certain normal subgroups,
namely the kernels of representations, are determined by knowing the characters of the
representations. We will see in an exercise that all normal subgroups of a finite group
may be found from knowledge of the characters of representations, as intersections of the
kernels of simple characters. This means that whether or not a group is simple may be
easily read from this information on characters.
We see in Proposition 3.1 part (6) that characters are constant on conjugacy classes,
so that in listing values of characters on group elements we only need take one element
from each conjugacy class. The table of complex numbers whose rows are indexed by
the isomorphism types of simple representations of G, whose columns are indexed by the
conjugacy classes of G and whose entries are the values of the characters of the simple
representations on representatives of the conjugacy classes is called the character table of
G. It is usual to index the first column of a character table by the (conjugacy class of the)
identity, and to put the character of the trivial representation as the top row. With this
convention the top row of every character table will be a row of 1s, and the first column
will list the degrees of the simple representations. Above the table it is usual to list two
rows, the first of which is a list of representatives of the conjugacy classes of elements of
G, in some notation. The row underneath lists the value of |CG (g)| for each element g in
the top row.
Peter Webb
23
(3.2) Example. We present the character table of S3 . We saw at the end of Chapter
2 that we already have a complete list of the simple modules for S3 , and the values of
their characters on representatives of the conjugacy classes of S3 are computed from the
matrices that give these representations.
S3
ordinary characters
g
|CG (g)|
() (12) (123)
6
2
3
1
sign
2
1
1
1 1
2
0
1
1
1
TABLE: The character table of S3 .

We will see that the character table has remarkable properties, among which are
that it is always square, and its rows (and also its columns) satisfy certain orthogonality
relations. Our next main goal is to state and prove these results. In order to do this we
first introduce three ways to construct new representations of a group from existing ones.
These constructions have validity no matter what ring R we work over, although in the
application to the character table we will suppose that R = C.
Suppose that V and W are representations of G over R. The R-module V R W
acquires an action of G by means of the formula g (v w) = gv gw, thereby making
the tensor product into a representation. This is what is called the tensor product of
the representations V and W , but it is not the only occurrence of tensor products in
representation theory, and as the other ones are different this one is sometimes also called
the Kronecker product. The action of G on the Kronecker product is called the diagonal
action. To do things properly we should check that the formula for the diagonal action
does indeed define a representation of G. This is immediate, but the fact that we can make
the definition at all is special for finite groups and group algebras: it does not work for
algebras in general.
For the second construction we form the R-module HomR (V, W ). This acquires an
action of G by means of the formula (gf )(v) = gf (g 1v) for each R-linear map f : V W
and g G. It is worth checking that the negative exponent in the formula is really necessary
so that we have a left action of G. Thus HomR (V, W ) becomes an RG-module.
The third construction is the particular case of the second in which we take W to be
the trivial module R. We write V = HomR (V, R) and the action is (g f )(v) = f (g 1 v)
for each f : V R and g G. This representation is called the dual or contragredient
representation of V . It is usually only considered when V is free (or at least projective) of
finite rank as an R-module, in which case we have V
= V as RG-modules. The exponent
1 in the action of G is necessary to make V a left RG-module. Without this exponent
we would get a right RG-module.
24
If R happens to be a field and we

have bases v1 , . . . , vm for V and w1 , . . . , wn for
W then V W has a basis {vi wj 1 i m, 1 j n} and V has a dual basis
v1 , . . . , vm . With respect to these bases an element g G acts on V W with the matrix
that is the tensor product of the two matrices giving its action on V and W , and on V it
acts with the transpose of the inverse of the matrix of its action on V . The tensor product
of two matrices is not seen so often these days. If (apq ), (brs ) are an m m matrix and an
n n matrix their tensor product is the mn mn matrix (cij ) where if i = (p 1)n + r
and j = (q 1)n + s with 1 p, q m and 1 r, s n then cij = apq brs . For example,

a b
c d
e
g
f
h
ae
ag
=
ce
cg
af
ah
cf
ch
be
bg
de
dg
bf
bh
.
df
dh
If : V V and : W W are endomorphisms, then the matrix of

: V W V W
is the tensor product of the matrices that represent and (provided the basis elements
vi wj are taken in an appropriate order). We see from this that tr( ) = tr()tr().
In the following result we consider the sum and product of characters, which are
defined in the usual manner for functions by the formulas
(V + W )(g) = V (g) + W (g)
(V W )(g) = V (g) W (g).
(3.3) PROPOSITION. Let V and W be finite-dimensional representations of G over
a field k of characteristic zero.
(1) V W has character V + W .
(2) V W has character V W .
(3) V has character V (g) = V (g 1 ) = V (g), the complex conjugate.
(4) Let M and N be representations of G over any ground ring R. Suppose that, as an Rmodule, at least one of M and N is free of finite rank, then HomR (M, N )
= M R N
as RG-modules. When R = k is a field of characteristic zero this representation has
character equal to M N .
Proof. (1), (2) and (3) are immediate from 3.1 and the subsequent remarks.
As for (4), we define an R-module homomorphism
: M R N HomR (M, N )
f v 7 (u 7 f (u) v),
this being the specification on basic tensors. We show that if either M or N is free as an
R-module of finite rank then is an isomorphism.
Peter Webb
25
Suppose that M is R-free, with basis u1 , . . . , um . Then M is R-free with basis

u
1 , . . . , u
m where u
i (uj ) = i,j . Any morphism f : M N is determined by its values
Pm
f (ui ) on the basis elements by means of the formula f (u) = i=1 u
(u)f (ui ) and it follows
Pm
Pm i
that f = ( i=1 u
i f (ui )) so that is surjective. If ( i=1 u
i vi ) = 0 for certain
elements vi N then applying this map to uj gives vj , so we deduce vj = 0 for all j, and
Pm
i vi = 0. Thus is injective
hence i=1 u
If N is free as an R-module with basis v1 , . . . , vn , let pj : N R be projection onto
Pn
component j. Any morphism f : M N equals ( i=1 pi f vi ), so is surjective. If
Pn
Pn
( i=1 gi vi ) = 0 for certain gi M then for all u M , i=1 gi (u)vi = 0, which
implies that gi (u) = 0 for all i and u. This means that gi is the zero map for all i, so that
Pn
i=1 gi vi = 0 and is injective.
We now observe that is a map of RG-modules, since for g G,
(g(f w)) = (gf gw)
= (v 7 (gf )(v) gw)
= (v 7 g(f (g 1v)w))
= g(v 7 f (v)w)
= g(f w).
Finally the last formula for characters follows from parts (2) and (3)
Of course, if R is taken to be a field in part (4) of 3.3 then the argument can be
simplified. Since M and N are always free, only one of the two arguments given is needed,
and after showing either that is surjective or injective, the other follows by observing
that the dimensions on the two sides are equal.
Orthogonality relations and bilinear forms

We start with some preliminary constructions that will be used in the proof of the
orthogonality relations for characters. A fundamental notion in dealing with group actions
is that of fixed points. If V is an RG-module we define the fixed points

V G = {v V gv = v for all g G}.
This is the largest RG-submodule of V on which G has trivial action.
(3.4) LEMMA. Over any ring R, HomR (V, W )G = HomRG (V, W ).

Proof. An R-linear map f : V W is a morphism of RG-modules precisely if it
commutes with the action of G, which is to say f (gv) = gf (v) for all g G and v V , or
in other words gf (g 1v) = f (v) always. This is exactly the condition that f is fixed under
the action of G.
26
The next result is an abstraction of the idea that was used in proving Maschkes
theorem, where the application was to the RG-module HomR (V, V ). We will use this
idea a second time in proving the orthogonality relations for characters. The projection
operator about to be described appears throughout representation theory.
(3.5) LEMMA. Let V be an RG-module where R is a ring in which |G| is invertible.
Then
1 X
g :V VG
|G|
gG
is a map of RG-modules which is projection onto the fixed points of V . In particular, V G

is a direct summand of V as an RG-module. When R is a field of characteristic zero we
have
1 X
g) = dim V G
tr(
|G|
gG
where tr denotes the trace.

P
1
Proof. Let : V V denote the map multiplication by |G|
gG g. Clearly is a
linear map, and it commutes with the action of G: if h G and v V we have
(hv) = (
1 X
gh)v
|G|
gG
= (v)
1 X
=(
hg)v
|G|
gG
= h(v)
since as g ranges through the elements of G so do gh and hg. The same equations show
that every vector of the form (v) is fixed by G. Furthermore, if v V G then
(v) =
1 X
1 X
gv =
v=v
|G|
|G|
gG
gG
so is indeed projection onto V G .

There is one more ingredient we describe before stating the orthogonality relations
for characters. We define an inner product on characters, but this does not make sense
without some further explanation, because an inner product must be defined on a vector
space and characters do not form a vector space. They are, however, elements in a vector
space, namely the vector space of class functions on G.
A class function on G is a function G C that is constant on each conjugacy class of
G. Such functions are in bijection with the functions from the set of conjugacy classes of G
Peter Webb
27
to C, a set of functions that we may denote Ccc(G) where cc(G) is the set of conjugacy classes
of G. These functions become an algebra when we define addition, multiplication and scalar
multiplication pointwise on the values of the function. In other words, ()(g) = (g)(g),
( + )(g) = (g) + (g) and ()(g) = (g) where , are class functions and C. If
G has n conjugacy classes, this algebra is isomorphic to Cn , the direct sum of n copies of
C, and is semisimple. We have seen in Proposition 3.1 that characters of representations
of G are examples of class functions on G.
We define a Hermitian form on the complex vector space of class functions on G by
means of the formula
1 X
h, i =
(g)(g).
|G|
gG
Since the functions and are constant on conjugacy classes, this can also be written
h, i =
=
1
|G|
g[cc(G)]
g[cc(G)]
|G|
(g)(g)
|CG (g)|
1
(g)(g)
|CG (g)|
where [cc(G)] denotes a set of representatives of the conjugacy classes of G, since the
number of conjugates of g is |G : CG (g)|. As well as the usual identities that express
bilinearity and the fact that the form is evidently Hermitian, it satisfies
h, i = h, i
where (g) = (g) is the class function obtained by complex conjugation. If and
happen to be characters of a representation we have (g) = (g 1 ), is the character of
the contragredient representation, and we obtain further expressions for the bilinear form:
h, i =
1 X
(g 1 )(g)
|G|
gG
1 X
(g)(g 1)
=
|G|
gG
1 X
(g)(g),
|G|
gG
where the second equality is obtained by observing that as g ranges over the elements of
G, so does g 1 .
With all this preparation we now present the orthogonality relations for the rows of
the character table. The picture will be completed once we have shown that the character
table is square and deduced the orthogonality relations for columns in 3.16 and 3.17.
28
(3.6) THEOREM (Row Orthogonality relations).

(1) If is the character of a simple representation over C then h, i = 1.
(2) If and are the characters of non-isomorphic simple representations over C then
h, i = 0.
Proof. Suppose that V and W are simple complex representations affording characters
and . By 3.3 the character of HomC (V, W ) is . By 3.4 and 3.5
dim HomCG (V, W ) = tr(
1 X
g) in its action on HomC (V, W )
|G|
gG
1 X
(g)(g)
=
|G|
gG
= h, i.
Schurs lemma asserts that this number is 1 if V

6 W.
= W , and 0 if V
=
Consequences of the orthogonality relations

We will describe many consequences of the orthogonality relations, and the first is
that they provide a way of determining the decomposition of a given representation as
a direct sum of simple representations. This procedure is similar to the way of finding
the coefficients in the Fourier expansion of a function using orthogonality of the functions
sin(mx) and cos(nx).
(3.7) COROLLARY. Let V be a CG-module. In any expression
V = S1n1 Srnr
in which S1 , . . . , Sr are non-isomorphic simple modules, we have
ni = hV , i i
where V is the character of V and i is the character of Si . In particular ni is determined
by V independently of the choice of decomposition.
(3.8) Example. Let G = S3 and denote by C the trivial representation, the sign
representation and V the 2-dimensional simple representation over C. We decompose the
4-dimensional representation V V as a direct sum of simple representations. Since the
values of the character V give the row of the character table
V : 2
1,
Peter Webb
29
V V has character values

V V : 4
0 1.
Thus
1
(4 1 + 0 + 2 1 1) = 1
6
1
hV V , i = (4 1 + 0 + 2 1 1) = 1
6
1
hV V , V i = (4 2 + 0 2 1 1) = 1
6
hV V , C i =
and we deduce that

V V
= C V.
(3.9) COROLLARY. For finite-dimensional complex representations V and W we
have V
= W if and only if V = W .
Proof. We saw in 3.1 that if V and W are isomorphic then they have the same
character. Conversely, if they have the same character they both may be decomposed as a
direct sum of simple representations by 1.7, and by 3.7 the multiplicities of the simples in
these two decompositions must be the same. Hence the representations are isomorphic.
The next result is a criterion for a representation to be simple. An important step in
studying the representation theory of a group is to construct its character table, and one
proceeds by compiling a list of the simple characters which at the end of the calculation will
be complete. At any stage one has a partial list of simple characters, and considers some
(potentially) new character. One finds the multiplicity of each previously obtained simple
character as a summand of the new character, and subtracts off the these simple characters
to the correct multiplicity. What is left is a character all of whose simple components are
new simple characters. This character will itself be simple precisely if the following easily
verified criterion is satisfied.
(3.10) COROLLARY. If is the character of a complex representation V then h, i
is a positive integer, and equals 1 if and only if V is simple.
Pr
Proof. We may write V
= S1n1 Srnr and then h, i = i=1 n2i is a positive
integer, which equals 1 precisely if one ni is 1 and the others are 0.
30
(3.11) Example. We construct the character table of S4 , since it illustrates some

techniques in finding simple characters.
S4
ordinary characters
g
|CG (g)|
() (12) (12)(34) (1234) (123)

24 4
8
4
3
1
sign
2
3a
3b
1
1
1 1
2
0
3 1
3
1
1
1
2
1
1
1
1
0
1
1
1
1
1
0
0
TABLE: The character table of S4 .

Above the horizontal line we list representatives of the conjugacy classes of elements of S4 ,
and below them the orders of their centralizers. The first row below the line is the character
of the trivial representation, and below that is the character of the sign representation.
There is a homomorphism : S4 S3 specified by (12) 7 (12), (34) 7 (12),
(23) 7 (23), (123) 7 (123) which has kernel the normal subgroup h(12)(34), (13)(24)i.
(One way to obtain this homomorphism is to identify S4 as the group of rotations of
a cube and observe that each rotation gives rise to a permutation of the three pairs of
opposite faces.) Any representation : S3 GL(V ) gives rise to a representation of
S4 obtained by composition with , and if we start with a simple representation of S3 we
will obtain a simple representation of S4 since is surjective and the invariant subspaces
for and are the same. Thus the simple characters of S3 give a set of simple characters
of S4 obtained by applying and evaluating the character of S3 . This procedure, which
in general works whenever one group is a homomorphic image of another, gives the trivial,
sign and 2-dimensional representations of S4 .
There is an isomorphism between S4 and the group of rotations of R3 that preserve
a cube. One sees this from the fact that the group of such rotations permutes the four
diagonals of the cube. This action is faithful and since every transposition of diagonals
may be realized through some rotation, so can every permutation of the diagonals. Hence
the full group of such rotations is isomorphic to S4 . The character of this action of S4 on
R3 is the fourth row of the character table. To compute the traces of the matrices that
represent the different elements we do not actually have to work out what those matrices
are, relying instead on the observation that every rotation of R3 has matrix conjugate to
a matrix
cos sin 0
sin cos 0
0
0
1
Peter Webb
31
where is the angle of rotation, and conjugate matrices have the same trace. Thus, for
example, (12) and (123) must act as rotations through and 2
3 , respectively, so act via
matrices that are conjugates of
1
2 23 0
1 0 0
0 1 0
3
and
12 0
2
0 0 1
0
0
1
which have traces 1 and 0.

There is an action of S4 on C4 given by the permutation action of S4 on four basis
vectors. Since the trace of a permutation matrix equals the number of points fixed by the
permutation, this has character
:
4 2
0 0 1
and we compute
2
1
4
+ + 0 + 0 + = 1.
24 4
3
Thus = 1 + where is the character of a 3-dimensional representation:
h, 1i =
1 0.
Again we have
9
1 1 1
+ + + +0 =1
24 4 8 4
so is simple by 3.10, and this is the bottom row of the character table.
There are other ways to complete the calculation of the character table. Having
computed four of the five rows, the fifth is determined by the facts that it is orthogonal to
the other four, and that the sum of the squares of the degrees of the characters equals 24.
Equally we could have constructed the bottom row as where is the other character
of degree 3 and is the sign character.
h, i =
Our next immediate goal is to prove that the character table of a finite group is square,
and to deduce the column orthogonality relations. Before doing this we show in the next
two results how part of the column orthogonality relations may be derived in a direct way.
Consider the regular representation of G on CG, and let CG denote the character of
this representation.
(3.12) LEMMA.
CG (g) =
|G|
0
if g = 1
otherwise.
Proof. Each g G acts by the permutation matrix corresponding to the permutation

h 7 gh. Now
CG (g) equals the number of 1s down the diagonal of this matrix, which
equals |{h G gh = h}|.
32
We may deduce an alternative proof of 2.5 (in case k = C), and also a way to do the
computation of the final row of the character table once the others have been determined.
(3.13) COROLLARY. Let 1 , . . . , r be the simple complex characters of G, with
degrees d1 , . . . , dr . Then hCG , i i = di , and hence
Pr
(1)
d2i = |G|, and
Pi=1
r
(2)
i=1 di i (g) = 0 if g 6= 1.
Proof. Direct evaluation gives
hCG , i i =
1
|G|i (1) = di
|G|
and hence CG = d1 1 + + dr r . Evaluating at 1 gives (1), and at g 6= 1 gives (2).
The center of the group algebra

It is an immediate deduction from the fact that the rows of the character table are
orthogonal that the number of simple complex characters of a group is at most the number
of conjugacy classes of elements in the group. We shall now prove that there is always
equality here. For any ring A we denote by Z(A) the center of A.
(3.14) LEMMA.
(1) For any commutative ring R, Z(Mn (R)) = {I
(2) The number of simple complex characters of G

R}
= R.
equals dim Z(CG).
Proof. (1) Let Eij denote the matrix which is 1 in place i, j and 0 elsewhere. If
X = (xij ) is any matrix then
Eij X = the matrix with row j of X moved to row i, 0 elsewhere,
XEij = the matrix with column i of X moved to column j, 0 elsewhere.
If X Z(Mn (R)) these two are equal, and we deduce that xii = xjj and all other entries
in row j and column i are 0. Therefore X = x11 I.
(2) In 2.3 we constructed an isomorphism
CG
= Mn1 (C) Mnr (C)
where the matrix summands are in bijection with the isomorphism classes of simple modules. On taking centers, each matrix summand contributes 1 to dim Z(CG).
Peter Webb
33
(3.15) LEMMA. Let x1 , . . . , xt be representatives of the conjugacy classes of elements

of G and let R be any ring. For each i let xi RG denote the sum of the elements in the
same conjugacy class as xi . Then Z(RG) is free as an R-module, with basis x1 , . . . , xt .
P
Proof. We first show that xi Z(RG). Write xi =
yxi y, where denotes
conjugacy. Then
X
X
gxi =
gy = (
gyg 1 )g = xi g
yxi
yxi
since as y runs through the elements of G conjugate to xi , so does gyg 1, and from this it
follows that xi is central.
P
Next suppose gG ag g Z(RG). We show that if g1 g2 then ag1 = ag2 . Suppose
P
P
that g2 = hg1 h1 . The coefficient of g2 in h( gG ag g)h1 is ag1 and in ( gG ag g) is
ag2 . Since elements of G are independent in RG, these coefficients must be equal. From
this we see that every element of Z(RG) can be expressed as an R-linear combination of
the xi .
Finally we observe that the xi are independent over R, since each is a sum of group
elements with support disjoint from the supports of the other xj .
(3.16) THEOREM. The number of simple complex characters of G equals the number
of conjugacy classes of elements of G.
Proof. In 3.14 we showed that the number of simple characters equals the dimension
of the center, and 3.15 we showed that this is equal to the number of conjugacy classes.
We conclude that the character table of a finite group is always square. From this we
get orthogonality relations between the columns of the character table.
(3.17) COROLLARY (Column orthogonality relations). Let X be the character table
of G as a matrix, and let
|C (x )|
0
G
C=
0
..
.
|CG (x2 )|
..
.
.
|CG (xr )|
..
where x1 , . . . , xr are representatives of the conjugacy classes of elements of G. Then

X T X = C,
where the bar denotes complex conjugation.
Proof. The orthogonality relations between the rows may be written XC 1 X T = I.
Since all these matrices are square, they are invertible, and in fact (XC 1 )1 = X T =
1
CX . Therefore X T X = C.
34
A less intimidating way to state the column orthogonality relations is

r
X
|CG (g)| if g h,
i (g)i (h) =
0
if g 6 h.
i=1
The special case of this in which g = 1 has already been seen in Corollary 3.13.
Algebraic integers and divisibility of character degrees

So far we have established that the degrees of the irreducible complex characters of G
have the properties that their number equals the number of conjugacy classes of G, and the
sum of their squares is |G|. We now establish a further important property, which is that
the character degrees all divide |G| a big restriction on the possible degrees that may
occur. It is proved using properties of algebraic integers, whose properties we now develop.
They will be used again in the proof of Burnsides theorem 3.24, but not elsewhere in this
text.
Suppose that S is a commutative ring with 1 and R is a subring of S with the same
1. An element s S is said to be integral over R if f (s) = 0 for some monic polynomial
f R[X], that is, a polynomial in which the coefficient of the highest power of X is 1.
We say that the ring S is integral over R if every element of S is integral over R. An
element of C integral over Z is called an algebraic integer. We summarize the properties
of integrality that we will need.
(1)
(2)
(3)
(4)
(3.18) THEOREM. Let R be a subring of a commutative ring S.

The following are equivalent for an element s S:
(a) s is integral over R,
(b) R[s] is contained in some R-submodule M of S which is finitely generated over
R and such that sM M .
The elements
of S integral over R form a subring of S.

{x Q x is integral over Z} = Z.
Let g be any element of a finite group G and any character of G. Then (g) is an
algebraic integer.
In this theorem R[s] denotes the subring of S generated by R and s.
Proof. (1) (a) (b). Suppose s is an element integral over R, satisfying the equation
sn + an1 sn1 + + a1 s + a0 = 0
where ai R. Then R[s] is generated as an R-module by 1, s, s2 , . . . , sn1 . This is because

the R-span of these elements is also closed under multiplication by s, using the fact that
s sn1 = an1 sn1 a1 s a0 ,
Peter Webb
35
and hence equals R[s]. We may take M = R[s].

(b) (a). Suppose R[s] M = Rx1 + Rxn with sM M . Thus for each i we
Pn
have sxi = j=1 ij xj for certain ij R. Consider the n n matrix A = sI (ij )
with entries in S, where I is the identity matrix. We have Ax = 0 where x is the vector
(x1 , . . . , xn )T , and so adj(A)Ax = 0 where adj(A) is the adjugate matrix of A satisfying
adj(A)A = det(A) I. Hence det(A) xi = 0 for all i. Since 1 R M is a linear
combination of the xi we have det(A) = 0 and so s is a root of the monic polynomial
det(X I (ij )).
(2) We show that if a, b S are integral over R then a + b and ab are also integral
over R. These lie in R[a, b], and we show that this is finitely generated as an R-module.
We see from the proof of part (1) that each of R[a] and R[b] is finitely generated as an
R-module. If R[a] is generated by x1 , . . . , xm and R[b] is generated by y1 , . . . , yn , then
R[a, b] is evidently generated as an R-module by all the products xi yj . Now R[a, b] also
satisfies the remaining condition of part (b) of (1), and we deduce that a + b and ab are
integral over R.
(3) Suppose that ab is integral over Z, where a, b are coprime integers. Then
a n1
a n
a
+ cn1
+ + c1 + c0 = 0
b
b
b
for certain integers ci , and so
an + cn1 an1 b + + c1 abn1 + c0 bn = 0.
Since b divides all terms in this equation except perhaps the term an , b must also be a
factor of an . Since a and b are coprime, this is only possible if b = 1, and we deduce that
a
b Z.
(4) (g) is the sum of the eigenvalues of g in its action on the representation which
affords . Since g n = 1 for some n these eigenvalues are all roots of X n 1 and so are
integers.
In the next result we identify Z with the subring Z 1, which is contained in the centre
Z(ZG) of ZG.
(3.19) PROPOSITION. The center Z(ZG) is integral over Z. Hence if x1 , . . . , xr are
representatives of the conjugacy classes of G, xi ZG is the sum of the elements conjugate
Pr
to xi , and 1 , . . . , r C are algebraic integers then the element i=1 i xi Z(CG) is
integral over Z.
Proof. It is the case that every commutative subring of ZG is integral over Z, using
condition 1(b) of 3.18, since such a subring is in particular a subgroup of the finitelygenerated free abelian group ZG, and hence is finitely generated as a Z-module.
We have seen in 3.15 that the elements x1 , . . . , xr lie in Z(ZG), so they are integral
Pr
over Z, and by part (2) of 3.18 the linear combination i=1 i xi is integral also. (We note
that the xi are in fact a finite set of generators for Z(ZG) as an abelian group, but we did
not need to know this for the proof.)
36
Let 1 , . . . , r be the simple representations of G over C with degrees d1 , . . . , dr and

characters 1 , . . . , r . Then each i : G Mdi (C) extends by linearity to a C-algebra
homomorphism
i : CG =
r
M
i=1
Mdi (C) Mdi (C)
projecting onto the ith matrix summand. The fact that the group homomorphism i
extends to an algebra homomorphism in this way comes formally from the construction
of the group algebra. The fact that this algebra homomorphism is projection onto the ith
summand arises from the way we decomposed CG as a sum of matrix algebras, in which
each matrix summand acts on the corresponding simple module as matrices on the space
of column vectors.
(3.20) PROPOSITION. Fixing the suffix i, if x Z(CG) then i (x) = I for some
C. In fact
i (x) =
1
tr(i (x)) I
di
where tr is the trace and I is the identity matrix. Writing x =

i (x) =
1 X
ag i (g) I.
di
gG
ag g we have
gG
Proof. Since x is central the matrix i (x) commutes with the matrices i (g) for all
g G. Therefore by Schurs lemma, since i is a simple complex representation, i (x) = I
for some scalar that we now compute. Evidently =
1
di tr(I).
Substituting i (x) into
the right hand side and multiplying both sides by I gives the first expression for i (x).
Replacing x by the expression in the statement of the proposition we obtain
=
X
1
tr(i (
ag g))
di
gG
1 X
ag tr(i (g))
=
di
gG
1 X
ag i (g)
di
gG
which gives the last expression for i (x).
Peter Webb
37
(3.21) THEOREM. The degrees di of the simple complex representations of G all

divide |G|.
P
1
Proof. Let x =
)g. This element is central in CG by 3.15, since the
gG i (g
coefficients of group elements are constant on conjugacy classes. By 3.20
i (x) =
1 X
i (g 1 )i (g) I
di
gG
|G|
I
di
the second equality arising from the fact that hi , i i = 1. Now x is integral over Z1 by 3.19
since the coefficients i (g 1 ) are algebraic integers, so i (x) is integral over i (Z 1) = Z I.

|G|
|G|.
is
integral
over
Z
and
hence
Z.
We
deduce
that
d
Thus |G|
i
di
di
The matrix summands of the complex group algebra
Given a set of rings with identity A1 , . . . , Ar we may form their direct sum A = A1
Ar , and this itself becomes a ring with componentwise addition and multiplication.
In this situation each ring Ai may be identified as the subset of A consisting of elements
that are zero except in component i, but this subset is not a subring of A because it does
not contain the identity element of A. It is, however, a 2-sided ideal. Equally, in any
decomposition of a ring A as a direct sum of 2-sided ideals, these ideals have the structure
of rings with identity.
Decompositions of a ring as direct sums of other rings are closely related to idempotent
elements in the center of the ring. We have seen idempotent elements introduced in the
exercises to Chapter 2, and they will retain importance throughout this book. An element
e of a ring A is said to be idempotent if e2 = e. It is a central idempotent element if it
lies in the center Z(A). Two idempotent elements e and f are orthogonal if ef = f e = 0.
An idempotent element e is called primitive if whenever e = e1 + e2 where e1 and e2 are
orthogonal idempotent elements then either e1 = 0 or e2 = 0. We say that e is a primitive
central idempotent element if it is primitive as an idempotent element in Z(A), that is,
e is central and has no proper decomposition as a sum of orthogonal central idempotent
elements.
We comment that the term idempotent element is very often abbreviated to idempotent, thereby elevating the adjective to the status of a noun. This does have justification,
in that idempotent element is unwieldy, and we will usually conform to the shorter usage.
38
(3.22) PROPOSITION. Let A be a ring with identity. Decompositions

A = A1 Ar
as direct sums of 2-sided ideals Ai biject with expressions

1 = e1 + + er
as a sum of orthogonal central idempotent elements, in such a way that ei is the identity
element of Ai and Ai = Aei . The Ai are indecomposable as rings if and only if the ei
are primitive central idempotent elements. If every Ai is indecomposable as a ring then
the subsets Ai and also the primitive central idempotents ei are uniquely determined;
furthermore, every central idempotent can be written as a sum of certain of the ei .
Proof. Given any ring decomposition A = A1 Ar we may write 1 = e1 + + er
where ei Ai and it is clear that the ei are orthogonal central idempotent elements.
Conversely, given an expression 1 = e1 + + er where the ei are orthogonal central
idempotent elements we have A = Ae1 Aer as rings.
To say that the ring Ai is indecomposable means that it cannot be expressed as a
direct sum of rings, except in the trivial way, and evidently this happens precisely if the
corresponding idempotent element cannot be decomposed as a sum of orthogonal central
idempotent elements.
We now demonstrate the (perhaps surprising) fact that there is at most one decomposition of A as a sum of indecomposable rings. Suppose we have two such decompositions,
and that the corresponding primitive central idempotent elements are labelled ei and fj ,
so that
1 = e1 + + er = f1 + + fs .
We have
ei = ei 1 =
s
X
ei fj ,
j=1
and so ei = ei fj for some unique j and ei fk = 0 if k 6= j, by primitivity of ei . Also

r
X
fj = 1 fj =
ek fj
k=1
so that ek fj 6= 0 for some unique k. Since ei fj 6= 0 we have k = i and ei fj = fj . Thus

ei = fj . We proceed by induction on r, starting at r = 1. If r > 1 we now work with the
P
P
ring A k6=i ek = A k6=j fk in which the identity is expressible as sums of primitive
P
P
central idempotent elements k6=i ek = k6=j fk . The first of these expressions has r 1
terms, so by induction the ek s are the same as the fk s after some permutation.
If e is any central idempotent and the ei are primitive then eei is either ei or 0 since
e = eei + e(1 ei ) is a sum of orthogonal central idempotents. Thus
r
r
X
X
e=e
ei =
eei
k=1
is a sum of certain of the ei .
k=1
Peter Webb
39
Notice that it follows from Proposition 3.22 that distinct primitive central idempotents
must necessarily be orthogonal, a conclusion that is false without the word central. The
primitive central idempotents, and also the indecomposable ring summands to which they
correspond according to Proposition 3.22, are known as blocks. We will study blocks in
detail in Chapter 12, and also at the end of Chapter 9. In the case of the complex group
algebra CG, that is a direct sum of matrix rings, the block idempotents are the elements
that are the identity in one matrix summand and zero in the others. We shall now give a
formula for these block idempotents. It is the same (up to a scalar) as the formula used
in the proof of Theorem 3.21.
(3.23) THEOREM. Let 1 , . . . , r be the simple complex characters of G with degrees
d1 , . . . , dr . The primitive central idempotent elements in CG are the elements
di X
i (g 1 )g
|G|
gG
where 1 i r, the corresponding indecomposable ring summand of CG having a simple

representation that affords the character i .
Proof. Using the notation of 3.20 we have that the representation i which affords i
yields an algebra map i : CG Mdi (C) that is projection onto the ith matrix summand
in a decomposition of CG as a sum of matrix rings. For any field k the matrix ring Mn (k)
is indecomposable, since we have seen in 3.14 that Z(Mn (k))
= k and the only non-zero
idempotent element in a field is 1. Thus the decomposition of CG as a direct sum of matrix
rings is the unique decomposition of CG as a sum of indecomposable ring summands.
The corresponding primitive central idempotent elements are the identity matrices in the
various summands, and so they are the elements ei CG such that i (ei ) = I and
j (ei ) = 0 if i 6= j. From the formula of 3.20 and the orthogonality relations we have
(using the Kronecker )
j (
di X
di X
i (g 1 )g) =
i (g 1 )j (g) I
|G|
|G|dj
gG
gG
di
hi , j i I
dj
di
= i,j I
dj
=
= i,j I,
so that the elements specified in the statement of the theorem do indeed project correctly
onto the identity matrices, and are therefore the primitive central idempotent elements.
40
While the identity matrix is a primitive central idempotent element in the matrix
ring Mn (k), where k is a field, it is never a primitive idempotent element if n > 1 since
it is the sum of the orthogonal (non-central) primitive idempotent elements I = E1,1 +
+ En,n . Furthermore, removing the hypothesis of centrality we can no longer say that
decompositions of the identity as a sum of primitive idempotent elements are unique;
indeed, any conjugate expression by an invertible matrix will also be a sum of orthogonal
primitive idempotent elements. Applying these comments to a matrix summand of CG, the
primitive idempotent decompositions of 1 will never be unique if we have a non-abelian
matrix summand which, of course, happens precisely when G is non-abelian. It is
unfortunately the case that in terms of the group elements there is in general no known
formula for primitive idempotent elements of CG lying in a non-abelian matrix summand.
Burnsides pa q b theorem
We conclude this chapter with Burnsides remarkable pa q b theorem, which establishes
a group-theoretic result using the ideas of representation theory we have so far developed,
together with some admirable ingenuity. In the course of the proof we again make use of
the idea of integrality, but this time we also require Galois theory at one point. This is
needed to show that if is a field element that is expressible as a sum of roots of unity,
then every algebraic conjugate of is again expressible as a sum of roots of unity. We
present Burnsides theorem here because of its importance as a theorem in its own right,
not because anything later depends on it. In view of this the proof (which is fairly long)
can be omitted without subsequent loss of understanding.
Recall that a group G is solvable if it has a composition series in which all of the
composition factors are cyclic. Thus a group is not solvable precisely if it has a nonabelian composition factor.
(3.24) THEOREM (Burnsides pa q b theorem). Let G be a group of order pa q b where
p and q are primes. Then G is solvable.
Proof. We suppose the result is false, and consider a group G of minimal order subject
to being not solvable and of order pa q b .
Step 1. The group G is simple, not abelian and not of prime-power order; for if it were
abelian or of prime-power order it would be solvable, and if G had a normal subgroup N
then one of N and G/N would be a smaller group of order p q which was not solvable.
Step 2. We show that G contains an element g whose conjugacy class has size q d
for some d > 0. Let P be a Sylow p-subgroup, 1 6= g Z(P ). Then CG (g) P so
|G : CG (g)| = q d for some d > 0, and this is the number of conjugates of g.

Step 3. We show that there is a simple non-identity character of G such that q6 (1)
and (g) 6= 0. To prove this, suppose to the contrary that whenever 6= 1 and q 6 (1)
Peter Webb
41
then (g) = 0. Let R denote the ring of algebraic integers in C. Consider the orthogonality
relation between the column of 1 (consisting of character degrees) and the column of g:
1+
(1)(g) = 0.
6=1
Then q divides every term apart from 1 in the sum on the left, and so 1 qR. Thus
1
q 1 R. But q 1 Q and
so q Z by 3.18, a contradiction. We now fix a non-identity
character for which q6 (1) and (g) 6= 0.
Step 4. Recall that the number of conjugates of g is q d . We show that
q d (g)
(1)
is an algebraic integer. To do this we use results 3.15 and 3.20. These imply that if
P
g = hg h CG is the sum of the elements conjugate to g and is a representation
affording the character then g Z(CG) and
(g) =
1 X
(h) I
(1)
hg
q d (g)
I,
(1)
where I is the identity matrix. Now by 3.19 this is integral over (Z) = Z I, which proves
what we want.
Step 5. We deduce that (g)
(1) is an algebraic integer. This arises from the fact that

q6 (1). We can find , Z so that q d + (1) = 1. Now
(g)
q d (g)
=
+ (g)
(1)
(1)
is a sum of algebraic integers.

Step 6. We show that |(g)| = (1) and put = (g)/(1). We consider the algebraic
conjugates of , which are the roots of the minimal polynomial of over Q. They are all
algebraic integers, since and its algebraic conjugates are all roots of the same polynomials
over Q. Thus the product N () of the algebraic conjugates is an algebraic integer. Since
it is also the constant term of the minimal polynomial of , it is rational and non-zero.
Therefore 0 6= N () Z by 3.18.
Now (g) is the sum of the eigenvalues of (g), of which there are (1), each of which is
a root of unity. Hence by the triangle inequality, |(g)| (1). By Galois theory the same
is true and a similar inequality holds for each algebraic conjugate of (g). We conclude
that all algebraic conjugates of have absolute value at most 1. Therefore |N ()| 1.
The only possibility is |N ()| = 1 and || = 1, so that |(g)| = (1), as was to be shown.
42
Step 7. We will obtain a contradiction by considering the subgroup

H = {h G |(h)| = (1)}
where is the simple non-identity character introduced in Step 3. We argue first that H
is a normal subgroup. If the eigenvalues of (h) are 1 , . . . , n then, since these are roots
of unity, |1 + + n | = n if and only if 1 = = n . Thus |(h)| = (1) if and only if
(h) is multiplication by some scalar, and from this we see immediately that H is a normal
subgroup. It also implies that H/ Ker is abelian. From Step 6 we see that H contains
the non-identity element g, so simplicity of G forces H = G. Since is not the trivial
representation Ker 6= G, so simplicity of G again forces Ker = 1, so that G must be
abelian. However G was seen not to be abelian in Step 1, and this contradiction completes
the proof.
1. There is a tensor product operation on RG-modules whose result is an RG-module.
2. Characters are class functions.
3. The character of a direct sum, tensor product or dual is the sum, product or
complex conjugate of the characters.
4. The characters of the simple representations form an orthonormal basis for class
functions with respect to a certain bilinear form.
5. The character table is square and satisfies row and column orthogonality relations.
The number of rows of the table equals the number of conjugacy classes in the group.
6. The conjugacy class sums form a basis for the center of RG.
7. The simple character degrees are divisors of |G|. The sum of their squares equals
|G|.
8. There is a formula for the primitive central idempotents in CG.
9. Every group of order pa q b is solvable.
Peter Webb
43

We assume throughout these exercises that representations are finite dimensional.
1. Suppose that V is a representation of G over C for which V (g) = 0 if g 6= 1. Show
that dim V is a multiple of |G|. Deduce that V
= CGn for some n. Show that if W is any
representation of G over C then CG C W
= CGdim W as CG-modules.
2. (a) By using characters show that if V and W are CG-modules then (V C W )
=

V C W , and (CG CG) = CG CG as CG-modules.
(b) If k is any field and V , W are kG-modules, show that (V k W )
= V k W ,

and (kG kG) = kG kG as kG-modules.
3. Consider a ring with identity that is the direct sum (as a ring) of non-zero subrings
A = A1 Ar . Suppose that A has exactly n isomorphism types of simple modules.
Show that r n.
4. Let g be any non-identity element of a group G. Show that G has a simple complex
character for which (g) has negative real part.
5. Show that if every element of a finite group G is conjugate to its inverse, then every
character of G is real-valued.
Conversely, show that if every character of G is real-valued, then every element of G
is conjugate to its inverse.
[The quaternion group of order 8 in its action on the algebra of quaternions provides
an example of a complex representation that is not equivalent to a real representation,
but whose character is real-valued (see Chapter 2 Exercise 11). In this example, the
representation has complex dimension 2, but there is no basis over C for the representation
space such that the group acts by matrices with real entries. A real-valued character does
not necessarily come from a real representation.]
6. (Jozsef Pelikan) While walking down the street you find a scrap of paper with the
following character table on it:
1
1
1
1
2 1
3
1
3
1
All except two of the columns are obscured, and while it is clear that there are five rows
you cannot read anything of the other columns, including their position. Prove that there
is an error in the table. Given that there is exactly one error, determine where it is, and
what the correct entry should be.
7. A finite group has seven conjugacy classes with representatives c1 , . . . , c7 (where
c1 = 1), and the values of five of its irreducible characters are given by the following table:
c1 c2 c3 c4 c5 c6 c7
1
1
1 1
1
1
1
1
1
1 1 1 1 1
4
1 1 0
2 1
0
4
1 1 0 2
1
0
5 1
0 1
1
1 1
44
Calculate the numbers of elements in the various conjugacy classes and the remaining
simple characters.
8. Let g G.
(a) Use Proposition 3.1 to prove that g lies in the center of G if and only if |(g)| =
|(1)| for every simple complex character of G.
(b) Show that if G has a faithful simple complex character (one whose kernel is 1)
then the center of G is cyclic.
9. Here is a column of a character table:
g
1
1
0
1
1
1+i 11
2
1i 11
2
0
1
0
(a) Find the order of g.

(b) Prove that g 6 Z(G).
(c) Show that there exists an element h G with the same order as g but not conjugate
to g.
(d) Show that there exist two distinct simple characters of G of the same degree.
10. Let A be a semisimple finite dimensional algebra over a field and let 1 = e1 + +en
be a sum of primitive central idempotents in A.
(a) If f A is a primitive idempotent (not necessarily central), show that there is a
unique i so that ei f 6= 0, and that for this i we have ei f = f .
(b) Show that for all A-modules V and for all i, ei V is an A-submodule of V and
Ln
V = i=1 ei V .
(c) Show that if S is a simple A-module there is a unique i so that ei S 6= 0, and that
for this i we have ei S = S.
(d) Show that for each i there is a simple module Si so that ei S 6= 0, and that if A is
assumed to be semisimple then Si is unique up to isomorphism.
(e) Assuming that A is semisimple and f A is a primitive idempotent (not necessarily
central), deduce that there is a simple module S which is unique up to isomorphism, such
that f S 6= 0.
11. (a) Let 1 = e1 + + en be a sum of primitive central idempotents in CG and let
V be a CG-module. Show that ei V is the largest submodule of V that is a direct sum of
copies of the simple module Si identified by ei Si 6= 0.
(b) Let V be any representation of S3 over C. Show that the subset

V2 = {2v (1, 2, 3)v (1, 3, 2)v v V }
Peter Webb
45
of V is the unique largest CS3 -submodule of V that is a direct sum of copies of the simple
2-dimensional CS3 -module.
12. Let G = hxi be cyclic of order n.
(a) Write down a complete set of primitive (central) idempotents in CG.
(b) Let Cn be an n-dimensional space with basis v1 , . . . , vn . Let g : Cn Cn be the
linear map of order n specified by gvi = vi+1 , 1 i n 1, gvn = v1 , so that g has matrix
0 0 1
.
0
1 ..
.
T =. .
..
..
..
.
0 1 0
Let = e2i/n be a primitive nth root of unity. Show that for each d the vector
ed =
n
X
ds vs
s=1
is an eigenvector of T , and that e1 , . . . , en is a basis of Cn .

13. (a) Let A be a finite dimensional semisimple algebra over a field. Show that the
center Z(A) is a semisimple algebra.
(b) Let G be a finite group and k a field in which |G| is invertible. Show that the
number of simple representations of kG is at most the number of conjugacy classes of G.
14. Let x be an element of G.
(a) If is a character of G, show that the function g 7 (xg) need not be a class
function on G.
(b) Show that the fact that the elements of CG specified in Theorem 3.23 are orthogonal idempotents is equivalent to the validity of the following formulas, for all x G and
for all of the simple characters i of G:

di X
0
if i 6= j,
1
i (g )j (xg) =
i (x) if i = j.
|G|
gG
15. Let G permute a set and let R denote the permutation representation of G
over R determined by . This means R has a basis in bijection with and each element
g G acts on R by permuting the basis elements in the same way that g permutes .
(a) Show that when H is a subgroup of G and = G/H is the set of left cosets of H
in G, the kernel of G in its action on R is H if and only if H is normal in G.
(b) Show that the normal subgroups of G are precisely the subgroups of the form
Ker i1 Ker it where 1 , . . . , n are the simple characters of G. Use Proposition
3.1 to deduce that the normal subgroups of G are determined by the character table of G.
(c) Show that G is a simple group if and only if for every non-trivial simple character
and for every non-identity element g G we have (g) 6= (1).
46
4. The Construction of Modules and Characters

In this chapter we describe the most important methods for constructing representations and character tables of groups. We describe the characters of cyclic groups, and
show also that simple characters of direct products are products of characters of the factors. This allows us to construct the simple characters of abelian groups and, in fact, all
degree 1 characters of any finite group. We go on to describe the construction of representations by induction from subgroups, that includes the special case of permutation
representations. Having obtained new characters of a group in these ways, we break them
apart using orthogonality relations so as to obtain characters of smaller representations.
An important tool in the process is Frobenius reciprocity. We conclude the section with a
description of symmetric and exterior powers
Cyclic groups and direct products

We start with the particular case of complex characters of cyclic groups, noting that
some properties of simple representations of cyclic groups over more general fields were
already explored in Exercise 9 from Section 2. For representations of cyclic groups over
fields where the characteristic divides the order of the group see 6.1, 6.2, 8.6 and 11.9.

(4.1) PROPOSITION. Let G = hx xn = 1i be a cyclic group of order n, and let
n C be a primitive nth root of unity. Then the simple complex characters of G are the
n functions
r (xs ) = nrs
where 0 r n 1.
Proof. We merely observe that the mapping
xs 7 nrs
is a group homomorphism
G GL(1, C) = C
giving a 1-dimensional representation with character r , that must necessarily be simple.
These characters are all distinct, and since the number of them equals the group order we
have them all.
Peter Webb
47
We next show how to obtain the simple characters of a product of groups in terms of
the characters of the groups in the product. Combining this with the last result we obtain
the character table of any finite abelian group. We describe a construction that works over
any ring R. Suppose that 1 : G1 GL(V1 ) and 2 : G2 GL(V2 ) are representations of
groups G1 and G2 . We may define an action of G1 G2 on V1 R V2 by the formula
(g1 , g2 )(v1 v2 ) = g1 v1 g2 v2
where gi Gi and vi Vi . When R is a field we may choose bases for V1 and V2 , and
now (g1 , g2 ) acts via the tensor product of the matrices by which g1 and g2 act. It follows
when R = C that
V1 V2 (g1 , g2 ) = V1 (g1 )V2 (g2 ).
(4.2) THEOREM. Let V1 , . . . , Vm and W1 , . . . , Wn be complete lists of the simple
complex representations of groups G1 and G2 . Then the representations Vi Wj with 1
i m and 1 j n form a complete list of the simple complex G1 G2 representations.
Remark. Theorem 4.2 is false in general when the field over which we are working is
not algebraically closed (see Exercise 15 to Section 4). The theorem is an instance of a
more general fact to do with representations of finite-dimensional algebras A and B over an
algebraically closed field k: the simple representations of A k B are precisely the modules
S k T , where S is a simple A-module and T is a simple B-module. This is proved in
[CRI, Theorem 10.38] and [Lemma 6.8 of R. Steinberg, Lectures on Chevalley Groups,
Yale University Notes, 1967.]. The connection between this result for abstract algebras
and group representations is that the group algebra R[G1 G2 ] over any commutative ring
R is isomorphic to RG1 R RG2 , which we may see by observing that this tensor product
has a basis consisting of elements g1 g2 with gi Gi that multiply in the same way as
the elements of the group G1 G2 .
Proof. We first verify that the representations Vi Wj are simple using the criterion
of 3.10:
hVi Wj , Vi Wj i =
=
=
1
|G1 G2 |
1
|G1 ||G2 |
1
|G1 |
= 1.
(g1 ,g2 )G1 G2
Vi Wj (g1 , g2 )Vi Wj (g1 , g2 )
Vi (g1 )Vi (g1 )Vj (g2 )Vj (g2 )
(g1 ,g2 )G1 G2
g1 G1
Vi (g1 )Vi (g1 )
1 X
Vj (g2 )Vj (g2 )
|G2 |
g2 G2
48
The characters of these representations are distinct, since by a similar calculation if (i, j) 6=
(r, s) then hVi Wj , Vr Ws i = 0. To show that we have the complete list, we observe that
if dim Vi = di and dim Wj = ej then Vi Wj is a representation of degree di ej and
n
m
m X
n
X
X
X
2
2
e2j = |G1 ||G2 |.
di
(di ej ) =
i=1
i=1 j=1
j=1
This establishes what we need, using 2.5 or 3.13.
Putting the last two results together enables us to compute the character table of any
finite abelian group. To give a very small example, let

G = hx, y x2 = y 2 = [x, y] = 1i
= C2 C2 .
The character tables of hxi and hyi are
hyi
hxi
g
|CG (g)|
1
2
x
2
1
2
1 1
1 1
g
|CG (g)|
1
2
y
2
1
2
1 1
1 1
TABLE: Two copies of the character table of C2 .

and the character table of C2 C2 is
hxi hyi
g
|CG (g)|
1
4
x y
4 4
xy
4
1 1
2 1
1 2
2 2
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
TABLE: The character table of C2 C2 .
We immediately notice that this construction gives the character table of C2 C2 as the
tensor product of the character tables of C2 and C2 , and evidently this is true in general.
(4.3) COROLLARY. The character table of a direct product G1 G2 is the tensor
product of the character tables of G1 and G2 .
We may see from this theory that all simple complex characters of an abelian group
have degree 1 (a result already shown in Corollary 2.10), and that in fact this property
characterizes abelian groups. We will give a different argument later on in Theorem 5.10
that shows that over any algebraically closed field the simple representations of abelian
groups all have degree 1.
Peter Webb
49
(4.4) THEOREM. The following are equivalent for a finite group G:

(1) G is abelian,
(2) all simple complex representations of G have degree 1.
Proof. Since the simple representations over C of every finite cyclic group all have
degree 1, and since every finite abelian group is a direct product of cyclic groups, the last
result shows that all simple representations of a finite abelian group have degree 1.
Pr
Conversely, we may use the fact that |G| = i=1 d2i where d1 , . . . , dr are the degrees of
the simple representations. We deduce that di = 1 for all i r = |G| every conjugacy
class has size 1 every element is central G is abelian.
Another proof of this result may be obtained from the fact that CG is a direct sum of
matrix algebras over C, a summand Mn (C) appearing precisely if there is a simple module
of dimension n. The group and hence the group ring are abelian if and only if n is always
1.
Lifting (or inflating) from a quotient group

Whenever we have a group homomorphism G H and a representation of H we
get a representation of G: regarding the representation of H as a group homomorphism
H GL(V ) we simply compose the two homomorphisms. The resulting representation
of G is call the lift or inflation of the representation of H. By this means, copies of the
character tables of quotient groups of G all appear in the character table of G, because the
lift of a simple representation is always simple. This observation, although straightforward,
allows us to fill out the character table of a group very rapidly, provided the group has
normal subgroups.
As an instance of this, we may construct the part of the character table of any finite
group that consists of characters of degree 1 by combining the previous results with the
next one, which is formulated so as to be true over any field.
(4.5) PROPOSITION. The degree 1 representations of any finite group G over any
field are precisely the degree 1 representations of G/G , lifted to G via the homomorphism
G G/G .
Proof. We only have to observe that a degree 1 representation of G over a field k is a
homomorphism G GL(1, k) = k that takes values in an abelian group, and so has kernel
containing G . Thus such a homomorphism is always a composite G G/G GL(1, k)
obtained from a degree 1 representation of G/G .
50
(4.6) Example. Neither implication of 4.4 holds if we do not assume that our representations are defined over an algebraically closed field of characteristic prime to |G|, such as
C. We have seen in examples before now that over R the 2-dimensional representation of
the cyclic group hx x3 = 1i, in which x acts as rotation through 2
3 , is simple since there
is no 1-dimensional subspace stable under the group action. We need to pass to C to split
it as a sum of two representations of degree 1. It is also possible to find a non-abelian group
all of whose simple representations do have degree 1: we shall see in Proposition 6.3 that
this happens whenever G is a p-group and we consider representations in characteristic p.
Induction and Restriction
We now consider how to construct representations of a group from representations of
its subgroups. Let H be a subgroup of G and V an RH-module where R is a commutative
ring with 1. We define an RG-module
V G
H = RG RH V
with the action of G coming from the left module action on RG:
X
X
x(
ag g v) = (x
ag g) v
gG
gG
where x, g G, ag R and v V . We refer to this module as V induced from H to G,

G
is used for this
and say that V G
H is an induced module. In many books the notation V
induced module, but for us this conflicts with
the
notation
for
fixed
points.

We denote the set of left cosets {gH g G} by G/H.
(4.7) PROPOSITION. Let V be an RH-module and let g1 H, . . . , g|G:H| H be a list

of the left cosets G/H. Then
|G:H|
M
G
gi V
V H =
i=1

as R-modules, where gi V = {gi v v V } RG RH V . Each gi V is isomorphic
to V as an R-module, and in case V is free as an R-module we have
rankR V G
H = |G : H| rankR V.
If x G then x(gi V ) = gj V where xgi = gj h for some h H. Thus the R-submodules

gi V of V G
H are permuted under the action of G. This action is transitive, and if g1 H
then StabG (g1 V ) = H.
L|G:H|
|G:H|
Proof. We have RGRH =

as right RH-modules. This
i=1 gi RH = RH
is because in its right action H permutes the group element basis of RG with orbits
Peter Webb
51
g1 H, . . . , g|G:H| H. Each orbit spans a right RH-submodule R[giH] = gi RH of RG and

so RG is their direct sum. Each of these submodules is isomorphic to RHRH as right
RH-modules via the isomorphism specified by gi h 7 h for each h H. Now
RG RH V = (
=
|G:H|
M
i=1
|G:H|
M
i=1
|G:H|
M
i=1
gi RH) RH V
(gi RH RH V )
gi RH V
and as R-modules gi RH RH V
= RH RH V
=V.
We next show that with its left action on RG RH V coming from the left action on
RG, G permutes these R-submodules. If x G and xgi = gj h with h H then
x(gi v) = xgi v
= gj h v
= gj hv,
so that x(gi v) gj V . We argue that we have equality using the invertibility of
x. For, by a similar argument to the one above, we have x1 gj V gi V , and so
gj V = xx1 (gj V ) x(gi V ). This action of G on the subspaces is transitive since
given two subspaces gi V and gj V we have (gj gi1 )gi V = gj V .
Now to compute the stabilizer of g1 V where g1 H, if x H then x(g1 V ) =
g1 (g11 xg1 ) V = g1 V , and if x 6 H then x gi H for some i 6= 1 and so x(g1 V ) =
gi V . Thus StabG (g1 V ) = H.
The structure of induced modules described in the last result in fact characterizes
these modules, giving an extremely useful criterion for a module to be of this form that
we will use several times later on.
(4.8) PROPOSITION. Let M be an RG-module that has an
R-submodule V with
the property that M is the direct sum of the R-submodules {gV g G}. Let H = {g
G gV = V }. Then M
= V G
H.

We are using the notation {gV g G} to indicate the set of distinct possibilities for
gV , so that if gV = hV with g 6= h we do not count gV twice.
Proof. We define a map of R-modules
RG RH V M
g v 7 gv
52
extending this specification from the generators to the whole of RG RH V by R-linearity.

This is in fact a map of RG-modules. The R-submodules gV of M are in bijection with
the cosets G/H, since G permutes them transitively, and the stabilizer of one of them is
H. Thus each of RG RH V and M is the direct sum of |G : H| R-submodules g V and
gV respectively, each isomorphic to V via isomorphisms g v v and gv v. Thus on
each summand the map g v 7 gv is an isomorphism, and so RG RH V M is itself
an isomorphism.
(4.9) EXAMPLES. 1. Immediately from the definitions we have R G

1 = RG R R =
RG, so that RG is induced from the identity subgroup.
2. Permutation modules: suppose that is a G-set, that is a set with an action of G
by permutations. We may form R, the free R-module with the elements of as a basis,
and it acquires the structure of an RG-module via the permutation action of G on this
L
basis. This is the permutation module determined by . Now R = R is the direct
sum of rank 1 R-submodules, each generated by a basis vector. In case G acts transitively
on these are permuted transitively by G. If we pick any and let H = StabG then
H is also the stabilizer of the space R and R
= R G
H by Proposition 4.8. This shows
that permutation modules on transitive G-sets are exactly the modules that are induced
from the trivial module on some subgroup.
Transitive permutation modules may be realized as quotient modules of RG, and
also as submodules of RG. The induced module R G
H = RG RH R is a quotient of
RG = RG R R from the definition of the tensor product. To realize this permutation
P
module also as a submodule of RG, write H :=
hH h RG. Then H generates a
G
submodule of RG isomorphic to R H . This may be proved as an application of Proposition
4.8, and it is Exercise 16(a) at the end of this section.
The general function of induced representations is that they are a mechanism that
relates the representations of a group to those of a subgroup. When working over C they
provide a way of constructing new characters, and with this in mind we give the formula
for the character of an induced representation. If is the character of a representation V
G
of a subgroup H, let us simply write G
H for the character of V H . We will also write
[G/H] to denote some (arbitrary) set of representatives of the left cosets of H in G. With
this convention, [G/H] is a set of elements of G, not a set of cosets.
(4.10) PROPOSITION. Let H be a subgroup of G and let V be a CH-module with
character . Then the character of V G
H is
G
H (g) =
=
1
|H|
(t1 gt)
tG
t1 gtH
t[G/H]
t1 gtH
(t1 gt).
Peter Webb
53
Proof. The two formulas on the right are in fact the same, since if t1 gt H and
h H then (th)1 gth H also, and so {t G t1 gt H} is a union of left cosets of
H. Since (t1 gt) = ((th)1 gth) the terms in the first sum are constant on the cosets
of H, and we obtain the second sum by choosing one representative from each coset and
multiplying by |H|.
Using the vector space decomposition of 4.7 we obtain that the trace of g on V G
H is
the sum of the traces of g on the spaces t V that are invariant under g, where t [G/H].
This is because if g does not leave t V invariant, we get a matrix of zeros on the diagonal
at that point in the block matrix decomposition for the matrix of g. Thus we only get a
non-zero contribution from subspaces t V with gt V = t V . This happens if and only
if t1 gt V = 1 V , that is t1 gt H. We have
X
G
(g)
=
trace of g on t V.
H
t[G/H]
t1 gtH
Now g acts on t V as
g(t v) = t(t1 gt) v = t (t1 gt)v
and so the trace of g on this space is (t1 gt). Combining this with the last expression
gives the result.
We see in the above proof that g leaves invariant t V if and only if t1 gt H,
or in other words g tHt1 . Thus StabG (t V ) = tHt1 . Furthermore, if we identify
t V with V by means of the bijection t v v, then g acts on t V via the composite
homomorphism
ct1
hgiH
GL(V )
where is the homomorphism associated to V and ca (x) = axa1 is the automorphism of
G that is conjugation by a G.
(4.11) Example. To make clearer what the terms in the expression for the induced
character are, consider G = S3 and H = h(123)i, the normal subgroup of order 3. To avoid
expressions such as (()) we will write the identity element of S3 as e. We may take the
coset representatives [G/H] to be {e, (12)}. If is the trivial character of H then
e
(12)
G
) = 2,
H (e) = (e ) + (e
G
H ((12)) = the empty sum = 0,
e
(12)
G
) = 2.
H ((123)) = ((123) ) + ((123)
Recalling the character table of S3 we find that G

H is the sum of the trivial character
and the sign character of S3 .
Induction is a very important tool in the construction of character tables, and in representation theory generally. Induced representations are rather hard to understand from
54
first principles, so we now describe some formalism that will enable us to compute with
them more easily. The companion notion to induction is that of restriction of representations. If H is a subgroup of G and W is a representation of G we denote by W G
H the
representation of H obtained by letting the elements of H the way they do when regarded
as elements of G. In other words, we just forget about the elements of G not in H. Restriction and induction are a particular case of the following more general situation. Whenever
we have a (unital) homomorphism of rings A B, an A-module V and an B-module
W , we may form the B-module B A V and the A-module W B
A . On taking A = RH
and B = RG we obtain the induction and restriction we have been studying. There is, in
fact, a further operation we mention at this point, that of coinduction: we may form the
B-module HomA (B, V ), which acquires the structure of a left B-module because of the
right action of B on itself. We are going to prove that induction and coinduction are the
same in the case of group rings, so that we will not need to consider coinduction for long.
(4.12) LEMMA. Let A B be a homomorphism of rings, V an A-module and W a
B-module.
(1) (Left adjoint of restriction) HomB (B A V, W )
= HomA (V, W A ).
(2) (Right adjoint of restriction) HomA (W A , V ) = HomB (W, HomA (B, V )).
(3) (Transitivity of induction) If : B C is another ring homomorphism then
C B (B A V )
= C A V.
Proof. In the case of (1) the mutually inverse isomorphisms are
f 7 (v 7 f (1 v))
and
(b v 7 bg(v))g.
In the case of (2) the mutually inverse isomorphisms are
f 7 (w 7 (b 7 f (bw)))
and
(w 7 g(w)(1))g.
In the case of (3) the mutually inverse isomorphisms are
c b v 7 c(b) v
and
c 1 vc v.
There is checking to be done to show that morphisms are indeed well-defined homomorphisms of A-modules and B-modules, and that maps are mutually inverse, but it is all
routine.
Peter Webb
55
It is worth knowing that 4.12(1) and (2) are instances of a single formula to do with
bimodules. An (A, B)-bimodule T is defined to have the structure of both a left A-module
and a right B-module, and such that these two module actions commute: for all a A,
b B, t T we have (at)b = a(tb). The basic adjoint relationship between A-modules V
and B-modules W is an isomorphism
HomA (T B W, V )
= HomB (W, HomA (T, V ))
that is given by mutually inverse maps
f 7 (w 7 (t 7 f (t w)))
and
(t w 7 g(w)(t))g.
Supposing that we have a ring homomorphism A B, one such bimodule T is the set B
with left action of A given by left multiplication after applying the homomorphism to B,
and right action of B given by right multiplication. We denote this bimodule A BB . There
is a similarly defined (B, A)-bimodule B BA on which B acts by left multiplication and A
acts by right multiplication after applying the homomorphism to B. We have isomorphisms
of A-modules
HomB (B BA , V )
= V B
A = A BB B V
whereas B BA A W and HomA (A BB , W ) are by definition the induction and coinduction
of W . Applying the single adjoint isomorphism in the two cases of these bimodules yields
the relationships of 4.12(1) and (2)
(4.13) COROLLARY. Let H K G be subgroups of G, let V be an RH-module
and W an RG-module.
(1) (Frobenius reciprocity)
G
HomRG (V G
H , W ) = HomRH (V, W H )
and
G
HomRG (W, V G
H ) = HomRH (W H , V ).
(2)
(3)
(4)
(5)
G
G
(Transitivity of induction) (V K
H ) K = V H as RG-modules.
K
G
(Transitivity of restriction) (W G
K ) H = W H as RH-modules.
G
G
V G
H R W = (V R W H ) H as RG-modules.
(Induced and coinduced are the same) V G

H = HomRH (RG, V ) as RG-modules.
Proof. The first isomorphism of (1) and part (2) follow from the relationships in 4.12,
in the case of the ring homomorphism RH RG. Part (3) also holds in this generality
56
and is immediate. The second isomorphism in (1) as well as (4) and (5) are special for
group representations.
Part (4) is the isomorphism
(RG RH V ) R W
= RG RH (V R W )
and it is not a corollary of 4.12. Here the mutually inverse isomorphisms are
(g v) w 7 g (v g 1 w)
and
(g v) gwg (v w).
We prove (5) by exhibiting mutually inverse isomorphisms
RG RH V
= HomRH (RG, V )
given by
g v 7 g,v
where g,v (x) = (xg)v if x Hg 1 and is 0 otherwise. Here g, x G and v V . In the
opposite direction the map is
X
g (g 1 )
g[G/H]
where the sum is taken over a set of representatives for the left cosets of H in G. We must
check here that g,v is a homomorphism of RH-modules, does specify a map on the tensor
product, that is well defined, and that the two morphisms are mutually inverse. We have
shown that induced and coinduced modules are the same.
Finally the second isomorphism of part (1) follows from 4.12(2), using (5).
In the case of representations in characteristic zero all of these results may be translated into the language of characters. In this setting the second Frobenius reciprocity
formula, which used the fact that coinduced modules and induced modules are the same,
becomes much easier. If our interest is only in character theory, we did not need to read
about left and right adjoints, coinduction and bimodules. By analogy with the notation
G
K
G
K for the character of an induced representation V K , let us write H for the
character of V K
H.
Peter Webb
57
(4.14) COROLLARY. Let H K G be subgroups of G, let be a complex

character of H and a character of G.
(1) (Frobenius reciprocity)
G
h G
H , iG = h, H iH
and
G
h, G
H iG = h H , iH .
In fact all four numbers are equal.

G
G
(2) (Transitivity of induction) ( K
H ) K = H .
K
G
(3) (Transitivity of restriction) ( K ) H = G
H.
G
G
(4) G
=
(
.
H
H
H
Proof. In (1) we write h , iG and h , iH to denote the inner product of characters
of G and H, respectively. The four parts are translations of the first four parts of 4.13
into the language of characters. In part (1) we use the fact that the inner products are the
dimensions of the Hom groups in 4.13(1). We may deduce the second formula from the
first here because the inner product is symmetric.
Frobenius reciprocity for complex characters is equivalent to saying that if and
are simple characters of G and H respectively then the multiplicity of as a summand of
G
G
H equals the multiplicity of as a summand of H .
At a slightly more sophisticated level we may interpret induction, restriction and
Frobenius reciprocity in terms of the space Ccc(G) of class functions introduced in Chapter
3, that is, the vector space of functions cc(G) C where cc(G) is the set of conjugacy
classes of G. Since each conjugacy class of H is contained in a unique conjugacy class
of G we have a mapping cc(H) cc(G) and this gives rise by composition to a linear
cc(G)
map G
Ccc(H) that on characters is the restriction operation we have already
H: C
cc(H)
defined. We may also define a linear map G
Ccc(G) that on characters sends a
H: C
character of H to the character G
H . It would be possible to define this on arbitrary
class functions of H by means of the explicit formula given in 4.10, but the trouble with
this is that transitivity of induction is not entirely obvious. It is perhaps easier to observe
that the characters of simple representations of H form a basis of Ccc(H) . We have defined
G
G
H on these basis elements, and we may define H on arbitrary class functions so that
it is a linear map. With these definitions the formulas of 4.14 hold for arbitrary class
functions. We may also interpret Frobenius reciprocity within this framework. The inner
products h , iG and h , iH provide us with the notion of the transpose of a linear
map between the vector spaces Ccc(H) and Ccc(G) . Now Frobenius reciprocity states that
induction and restriction are the transpose of each other. We know that the characters
of simple modules form orthonormal bases of these vector spaces. Taking matrices with
respect to these bases, the matrix of induction is the transpose of the matrix of restriction.
(4.15) Examples. 1. Frobenius reciprocity is a most useful tool in calculating with
induced characters. In the special case that V and W are simple representations over C of
58
H and G, respectively, where H G, it says that the multiplicity of W as a summand of

G
V G
H equals the multiplicity of V as a summand of W H . As an example we may take
both V and W to be the trivial representations of their respective groups. As explained
in 4.9, C G
H is a permutation module. We deduce from Frobenius reciprocity that as
representations of G, C is a direct summand of C G
H with multiplicity one.
n
2
1
1

2. Let G = hx, y x = y = 1, yxy = x i = D2n , the dihedral group of order 2n.
Suppose that n is odd. We compute that the commutator [y, x] = xn2 , and since n is odd
we have G = hxn2 i = hxi
= Cn and G/G
= C2 . Thus G has two complex characters of
degree 1 that we denote 1 and 1.
Let ns denote the degree 1 character of hxi specified by ns (xr ) = nrs where n =
2i
e n . Then ns G
hxi has values given in the following table:
D2n , n odd
ordinary characters
1
2n
x
n
x2
n
1
1a
1
1
1
1
1
1
ns G
hxi
g
|CG (g)|
(1 s
n1
2 )
n1
2
y
2
n
1
1
n1
ns + ns n2s + n2s n 2
n1
+ n 2
1
1
0
TABLE: The character table of D2n , n odd.

We verify that
G
hs G
hxi , 1iG = hs , 1 hxi ihxi = 0

if n6 s, using Frobenius reciprocity (or a direct calculation), and hence the characters s

must be simple when n 6 s, because otherwise they would be a sum of two characters of
degree 1, that must be 1 or 1, and evidently this would not give the correct character
values. For 1 s n1
they are distinct, and so we have constructed n1
+ 2 = n+3
2
2
2
simple characters. This equals the number of conjugacy classes of G, so we have the
complete character table.
Peter Webb
59
Symmetric and Exterior Powers

As further ways of constructing new representations from old ones we describe the
symmetric powers and exterior powers of a representation. If V is a vector space over a
field k its nth symmetric power is the vector space
S n (V ) = V n /I
where V n = V V with n factors, and I is the subspace spanned by tensors of
the form (( vi vj ) ( vj vi )), where all places in the
two basic tensors are the same except for two of them, where the elements vi , vj V are
interchanged. We write the image of the tensor v1 vn in S n (V ) as a (commutative)
product v1 vn , noting that in S n (V ) it does not matter in which order we write the terms.
A good way to think of S n (V ) is as the space of homogeneous polynomials of degree n in a
polynomial ring. Indeed, if u1 , . . . , ur is any basis of V and we let k[u1 , . . . , ur ]n denote the
vector space of homogeneous polynomials of degree n in the ui as indeterminates, there is
a surjective linear map
V n k[u1 , . . . , ur ]n
ui1 uin 7 ui1 uin
(extended by linearity to the whole of V n ). This map contains I in its kernel, so there is
induced a map
S n (V ) k[u1 , . . . , ur ]n .
1
2
r
This is now an isomorphism since, modulo I, the tensors ua
ua
ua
r
1
2
Pr
n
where
, and they map to the monomialsthat form a basis of
i=1 ai = n span V
dimk k[u1 , . . . , ur ]n . As is well-known, dimk k[u1 , . . . , ur ]n = n+r1
.
n
The nth exterior power of V is the vector space
n (V ) = V n /J
where J is the subspace spanned by tensors (( vi vj )+( vj vi ))
and ( vi vi ) where vi , vj V . We write the image of v1 vn in
n (V ) as v1 vn , so that interchanging vi and vj changes the sign of the symbol,
and if two of vi and vj are equal the symbol is zero. If the characteristic of k is not 2
the second of these properties follows from the first, but for the sake of characteristic 2 we
impose it anyway. By an argument similar to the one used for symmetric powers we see
n

that
(V ) has as anbasis {ui1 uin 1 i1 < < in r}, and its dimension is
r
n . In particular, (V ) = 0 if n > dim V .
Suppose now that a group G acts on V and consider the diagonal action of G on V n .
The subspaces of relations I and J are preserved by this action, and so there arise actions
of G on S n (V ) and n (V ):
g (v1 v2 vn ) = (gv1 )(gv2 ) (gvn )
g (v1 vn ) = (gv1 ) (gvn ).
60
Because we substitute the expressions for gvi into the monomials that form the bases of
S n (V ) and n (V ), we say that G acts on these spaces by linear substitutions. With these
actions we have described the symmetric and exterior powers of the representation V .
(4.16) Example. Consider the representation of G = hx x3 = 1i on the vector space
V with basis {u1 , u2 } given by
xu1 = u2
xu2 = u1 u2 .
Then S 2 (V ) has a basis {u21 , u1 u2 , u22 } and
x u21 = u22
x (u1 u2 ) = u2 (u1 u2 ) = u1 u2 u22
x u22 = (u1 u2 )2 = u21 + 2u1 u2 + u22 .
Similarly 2 (V ) has basis {u1 u2 } and

x (u1 u2 ) = u2 (u1 u2 ) = u1 u2 .
The symmetric and exterior powers fit into a more general framework where we consider tensors with different symmetry properties. There is an action of the symmetric
group Sn on the n-fold tensor power V n given by permuting the positions of vectors in a
tensor, so that for example if , , are vectors in V then
(1, 2)( ) = ,
(1, 3)( ) = .
From the above very convincing formulas and the fact that (1, 2, 3) = (1, 3)(1, 2) we deduce
that
(1, 2, 3)( ) =
which is evidence that if Sn then
(v1 vn ) = v1 (1) v1 (2) v1 (n) ,
a formula that is not quite so obvious. With this action it is evident that S n (V ) is the
largest quotient of V n on which Sn acts trivially, and when char(k) 6= 2, n (V ) is the
largest quotient of V n on which Sn acts as a sum of copies of the sign representation.
We define the symmetric tensors or divided powers to be the fixed points (V n )Sn , and
when char k 6= 2 we define the skew-symmetric tensors to be the largest kSn -submodule of
V n that is a sum of modules isomorphic to the sign representation. Thus

symmetric tensors = {w V n (w) = w for all Sn },

skew-symmetric tensors = {w V n (w) = sign()w for all Sn }.
Peter Webb
61
When we let G act diagonally on V n the symmetric tensors, the skew-symmetric

tensors, as well as the subspaces I and J defined earlier remain invariant for the action of
G. We easily see this directly, but at a more theoretical level the reason is that the actions
of G and Sn on V n commute with each other (as is easily verified), so that V n acquires
the structure of a k[G Sn ]-module, and elements of G act as endomorphisms of V n as
a kSn -module, and vice-versa. Every endomorphism of the kSn -module V n must send
the Sn -fixed points to themselves, for example, and so the symmetric tensors are invariant
under the action of G. One sees similarly that the other subspaces are also invariant under
the action of G.
We remark that, in general, the symmetric power S n (V ) and the symmetric tensors
provide non-isomorphic representations of G, as do n (V ) and the skew-symmetric tensors.
This phenomenon is investigated in the exercises at the end of this chapter. However these
pairs of kG-modules are isomorphic in characteristic zero, and we now consider in detail the
case of the symmetric and exterior square. Suppose that k is a field whose characteristic
is not 2. In this situation the only tensor that is both symmetric and skew-symmetric is 0.
Any degree 2 tensor may be written as the sum of a symmetric tensor and a skew-symmetric
tensor in the following way:
X
ij vi vj =
1X
1X
ij (vi vj + vj vi ) +
ij (vi vj vj vi ).
2
2
We deduce from this that

V V = symmetric tensors skew-symmetric tensors
as kG-modules. The subspace I that appeared in the definition S 2 (V ) = (V V )/I is
contained in the space of skew-symmetric tensors, and the subspace J for which 2 (V ) =
(V V )/J is contained in the space of symmetric tensors. By counting dimensions we see
that dim I + dim J = dim V V and putting this together we see that
I = skew-symmetric tensors,
J = symmetric tensors, and
V V = I J.
From this information we see on factoring out I and J that
S 2 (V )
= symmetric tensors,
2 (V )
= skew-symmetric tensors,
and we have proved the following result.
62
(4.17) PROPOSITION. Suppose V is a representation for G over a field k whose

characteristic is not 2. Then
V V
= S 2 (V ) 2 (V ),
as kG-modules, where G acts diagonally on V V and by linear substitutions on S 2 (V )
and 2 (V ).
One application of this is that when constructing new representations from an existing
representation V by taking tensor products, the tensor square will always decompose (away
from characteristic 2) giving two smaller representations.
Suppose now that k = C. If is the character of a representation V we write S 2
and 2 for the characters of S 2 (V ) and 2 (V ).
(4.18) PROPOSITION. Let be the character of a representation V of G over C.
Then
1
S 2 (g) = ((g)2 + (g 2 ))
2
1
2 (g) = ((g)2 (g 2 )).
2
Proof. For each g G, V G
hgi is the direct sum of 1-dimensional representations of
the cyclic group hgi, and so we may choose a basis u1 , . . . , ur for V such that g ui = i ui
for scalars i . The monomials u2i with 1 i r and ui uj with 1 i < j r form a basis
for S 2 V , and so the eigenvalues of g on this space are 2i with 1 i r and i j with
1 i < j r. Therefore
2
S (g) =
r
X
2i +
i=1
i j
1i<jr
1
((1 + + r )2 + (21 + + 2r ))
2
1
= ((g)2 + (g 2 )).
2
Similarly 2 V has a basis ui uj with 1 i < j r, so the eigenvalues of g on 2 V are

i j with 1 i < j r and
X
2 (g) =
i j
1i<jr
1
((1 + + r )2 (21 + + 2r ))
2
1
= ((g)2 (g 2 )).
2
Peter Webb
63
There is a formula due to Molien for the generating function of characters of the
symmetric powers of V . We present Moliens theorem in exercise 14.
The Construction of Character Tables

We may now summarize some major techniques used in constructing complex character tables. The first things to do are to determine
the conjugacy classes in G,
the abelianization G/G ,
the 1-dimensional characters of G.
We construct characters of degree larger than 1 as
natural representations of G,
representations lifted from quotient groups,
representations induced from subgroups,
tensor products of other representations,
symmetric and exterior powers of other representations,
contragredients of other representations.
As a special case of the induced representations, we have permutation representations.
The representations obtained by these methods might not be simple, so we test them for
simplicity and subtract off known character summands using the
orthogonality relations
which are assisted in the case of induced characters, by
Frobenius reciprocity.
The orthogonality relations provide a check on the accuracy of our calculations, and also
enable us to complete the final row of the character table. The facts that the character
degrees divide |G| and that the sum of the squares of the degrees equals |G| also help in
this.
64
1. The characters of representations obtained by induction, tensor product, symmetric
and exterior powers are all useful in constructing character tables and there are formulas
for these characters.
2. The character table of G1 G2 is the tensor product of the character tables of G1
and G2 .
3. The degree 1 characters of G are precisely the characters of the simple representations of G/G .
4. The simple complex characters of a cyclic group of order n are the n homomorphisms to the group of nth roots of unity.
5. Induced characters may be decomposed using Frobenius reciprocity.
6. An induced module may be identified by the fact that it is a direct sum of subspaces
that are permuted by G.

In these exercises every module is supposed to be a finite dimensional vector space
over the ground field.
1. (a) Compute the character table of the dihedral group D2n when n is even.
(b) Compute the character table of the quaternion group Q8 that was described in
Chapter 2 Exercise 11.
(c) Use this information to show that the posets of normal subgroups of D8 and Q8
are isomorphic.
[The groups D8 and Q8 provide an example of non-isomorphic groups whose character
tables are the same.]
2. Let G be the non-abelian group of order 21:

G = hx, y x7 = y 3 = 1, yxy 1 = x2 i.
Show that G has 5 conjugacy classes, and find its character table.
3. Find the character table of the following group of order 36:

G = ha, b, c a3 = b3 = c4 = 1, ab = ba, cac1 = b, cbc1 = a2 i.
[It follows from these relations that ha, bi is a normal subgroup of G of order 9.]
4. Compute the character table of the symmetric group S5 by the methods of this
chapter. To help in doing this, consider especially the decomposition of the permutation
representation on 5 symbols, the symmetric and exterior square of the summands, as well
as tensor product with the sign representation.
Peter Webb
65
5. Let k be any field, H a subgroup of G, and V a representation of H over k. Show

G
that V G
H = (V H ) . Deduce from this that kG = (kG) and (more generally) that
permutation modules are self-dual (i.e. isomorphic to their dual).
6. Let k be any field, and V any representation of G over k. Prove that V kG is
isomorphic to a direct sum of copies of kG.
7. The tensor product V = R3 R R3 R R3 is a vector space of dimension 27 with basis
the tensors ei ej ek where e1 , e2 , e3 is a standerd basis for R3 . The symmetric group S3
acts on V by permuting the positions of the suffixes, so for instance (1, 2) (e3 e1 e2 ) =
e1 e3 e2 .
(a) Find the multiplicity of each simple representation of S3 in a decomposition of V as
a direct sum of simple representations. [Observe that V is a permutation representation.]
(b) Give also the decomposition of V as a direct sum of three subspaces consisting of
tensors with different symmetry properties under the action of S3 . What are the dimensions
of these subspaces? Find a basis for each subspace. [Use the result of Chapter 3 Exercise
11.]
(c) The Schur algebra SC (3, 3) may be defined to be the endomorphism ring HomCS3 (V, V ).
Show that SC (3, 3) is semisimple and find the dimensions of its simple representations.
8. Let V be a representation of G over a field k of characteristic zero. Prove that the
symmetric power S n (V ) is isomorphic as a kG-module to the space of symmetric tensors
in V n .
9. Let U, V be kG-modules where k is a field, and suppose we are given a nondegenerate bilinear pairing
h , i:U V k
that is G-invariant, that is, hu,vi = hgu, gvi for all u U , v V , g G. If U1 is a
subspace of U let U1 = {v V hu, vi = 0 for all u U1 } and if V1 is a subspace of V let
V1 = {u U hu, vi = 0 for all v V1 }.
(a) Show that V
= U as kG-modules, and that there is an identification of V with
U so that h , i identifies with the canonical pairing U U k.
(b) Show that if U1 and V1 are kG-submodules, then so are U1 and V1 .
(c) Show that if U1 U2 are kG-submodules of U then
U1 /U2
= (U2 /U1 )
as kG-modules.
(d) Show that the composition factors of U are the duals of the composition factors
of U .
10. Let be a finite G-set and k the corresponding permutation module, where
k is a field. Let h , i : k k k be the symmetric bilinear form specified on the
elements of as
n
1 if 1 = 2 ,
h1 , 2 i =
0 otherwise.
66
(a) Show that this bilinear form is G-invariant, i.e. h1 , 2 i = hg1 , g2 i for all g G.
(b) Deduce from this that k is self-dual, i.e. k
= (k) .
11. Let V be a kG-module where k is a field, and let h , i : V V k be the
canonical pairing between V and its dual, so hv, f i = f (v).
(a) Show that the specification hv1 vn , f1 fn i = f1 (v1 ) fn (vn ) determines a non-degenerate bilinear pairing h , i : V n (V )n k that is invariant both
for the diagonal action of G and the action of Sn given by permuting the positions of the
tensors.
(b) Let I and J be the subspaces of V n that appear in the definitions of the symmetric
and exterior powers, so S n (V ) = V n /I and n = V n /J. Show that I (defined in
question 9) equals the space of symmetric tensors in (V )n , and that J equals the space
of skew-symmetric tensors in (V )n (at least, when char k 6= 2).
(c) Show that (S n (V ))
= STn (V ), and that (n (V ))
= SSTn (V ), where STn
denotes the symmetric tensors, and in general we define the skew-symmetric tensors
SSTn (V ) to be J .
12. Let G = C2 C2 be the Klein four group with generators a and b, and k = F2
the field of two elements. Let V be a 3-dimensional space on which a and b act via the
matrices
1 0 0
1 0 0
1 1 0 and 0 1 0 .
0 0 1
1 0 1
Show that S 2 (V ) is not isomorphic to either ST 2 (V ) or ST 2 (V ) , where ST denotes the

symmetric tensors. [Hint: Compute the dimensions of the spaces of fixed points of these
representations.]
13. (Artins Induction Theorem) Let Ccc(G) denote the vector space of class functions
on G and let C be a set of subgroups of G that contains a representative of each conjugacy
class of cyclic subgroups of G. Consider the linear mappings
resC : Ccc(G)
and
indC :
HC
Ccc(H)
HC
Ccc(H) Ccc(G)
whose component homomorphisms are the linear mappings given by restriction

cc(G)
G
Ccc(H)
H: C
and induction
cc(H)
G
Ccc(G)
H: C
Peter Webb
67
(a) With respect to the usual inner product h , iG on Ccc(G) and the inner product
L
on HC Ccc(H) that is the orthogonal sum of the h , iH , show that resC and indC are
the transpose of each other.
(b) Show that resC is injective.
[Use the fact that Ccc(G) has a basis consisting of characters, that take their information
from cyclic subgroups.]
(c) Prove Artins induction theorem: In Ccc(G) every character can be written as a
rational linear combination
X
=
aH, G
H
where the sum is taken over cyclic subgroups H of G, ranges over characters of H and
aH, Q.
[Deduce this from surjectivity of indC and the fact that it is given by a matrix with integer
entries. A stronger version of Artins theorem is possible: there is a proof due to Brauer
which gives an explicit formula for the coefficients aH, ; from this we may deduce that
when is the character of a QG-module the that arise may all be taken to be the trivial
character.]
(d) Show that if U is any CG-module then there are CG-modules P and Q, each a
direct sum of modules of the form V G
H where H is cyclic, for various V and H, so that
n
Q
for
some
n,
where
U
is
the
direct sum of n copies of U .
Un P
=
14. (Moliens Theorem) (a) Let : G GL(V ) be a complex representation of G, so
that V is a CG-module, and for each n let S n (V ) be the character of the nth symmetric
power of V . Show that for each g G there is an equality of formal power series
S n (V ) (g)tn =
n=0
1
.
det(1 t(g))
Here t is an indeterminate, and the determinant that appears in this expression is of a

matrix with entries in the polynomial ring C[t], so that the determinant is a polynomial
in t. On expanding the rational function on the right we obtain a formal power series that
is asserted to be equal to the formal power series on the left.
[Choose a basis for V so that g acts diagonally, with eigenvalues 1 , . . . , d . Show that on
P
both sides of the equation the coefficient of tn is equal to i1 ++id =n 1i1 did .]
(b) If W is a simple CG-module we may write the multiplicity of W as a summand
n
of S (V ) as hS n (V ) , W i and consider the formal power series
MV (W ) =
X
i=0
Show that
MV (W ) =
hS n (V ) , W itn .
1 X W (g 1 )
.
|G|
det(1 t(g))
gG
68
(c) When G = S3 and V is the 2-dimensional simple CS3 -module show that
MV (C) =
1
(1
t2 )(1
t3 )
= 1 + t2 + t3 + t4 + t5 + 2t6 + t7 + 2t8 + 2t9 + 2t10 +
t3
MV () =
= t3 + t5 + t6 + t7 + t8 + 2t9 + t10 +
2
3
(1 t )(1 t )
t(1 + t)
MV (V ) =
= t + t2 + t3 + 2t4 + 2t5 + 2t6 + 3t7 + 3t8 + 3t9 + 4t10 +
(1 t2 )(1 t3 )
where C denotes the trivial module and the sign representation. Deduce, for example,
that the eighth symmetric power S 8 (V )
= C2 V 3 .
15. Show that every simple representation of C3 C3 over R has dimension 1 or 2.
Deduce that if V is a simple 2-dimensional representation of C3 over R then V V is not
a simple R[C3 C3 ]-module.
16. Let H be a subgroup of G.
P
(a) Write H = hH h for the sum of the elements of H, as an element of RG. Show
that RG H
= R G
H as left RG-modules. Show also that RG H equals the fixed points
of H in its action on RG from the right.
(b) More generally let : H R be a 1-dimensional representation of H (that is,
:= P
a group homomorphism to the units of R). Write H
hH (h)h RG. Show that
= as RG-modules.
RG H
H
17. Let H be a subgroup of G and V an RH-module. Show that if V can be generated
by d elements as an RH-module then V G
H can be generated by d elements as an RGmodule.
18. Compute the character tables of the alternating groups A4 and A5 using the
following procedure. You may assume that A5 is a simple group that is isomorphic to
the group of rotations of a regular icosahedron, and that A4 is isomorphic to the group of
rotations of a regular tetrahedron.
(a) Compute the conjugacy classes by observing that each conjugacy class of even
permutations in Sn is either a single class in An or the union of two classes of An , and
that this can be determined by computing centralizers of elements in An and comparing
them with the centralizers in Sn .
(b) Compute the abelianization of each group, and hence the 1-dimensional representations.
(c) Obtain further representations using the methods of this section. We have natural
3-dimensional representations in each case. It is also helpful to consider d induced representations from the Sylow 2-subgroup in the case of A4 , and from the subgroup A4 in the
case of A5 .
Peter Webb
69
5. More on Induction and Restriction: Theorems of Mackey and Clifford

The results in this chapter go more deeply into the theory. They apply over fields of
arbitrary characteristic, and even over arbitrary rings in the case of Mackeys decomposition formula. We start with this formula, which is a relationship between induction and
restriction. After that we explain Cliffords theorem, which shows what happens when a
simple representation is restricted to a normal subgroup. These results will have many
consequences later on. In this chapter we will see the consequence of Cliffords theorem
that simple representations of p-groups are induced from 1-dimensional representations of
subgroups.
For Mackeys formula we need to consider double cosets. Given subgroups H and K
of G we define for each g G the (H, K)-double coset

HgK = {hgk h H, k K}.
If is a left G-set we use the notation G\ for the set of orbits of G on , and denote a
set of representatives for the orbits by [G\]. Similarly if is a right G-set we write /G
and [/G]. We will use all the time the fact that if is a transitive G-set and then
= G/ StabG (), the set of left cosets of the stabilizer of in G.

(5.1) PROPOSITION. Let H, K G.
(1) Each (H, K)-double coset is a disjoint union of right cosets of H and a disjoint union
of left cosets of K.
(2) Any two (H, K)-double cosets either coincide or are disjoint. The (H, K)-double cosets
partition G.
(3) The set of (H, K)-double cosets is in bijection with the orbits H\(G/K), and also
with the orbits (H\G)/K under the mappings
HgK 7 H(gK) H\(G/K)
HgK 7 (Hg)K (H\G)/K.

Proof. (1) If hgk HgK and k1 K then hgk k1 = hg(kk1 ) HgK so that the
entire left coset of K that contains hgk is contained in HgK. This shows that HgK is a
union of left cosets of K, and similarly it is a union of right cosets of H.
1
(2) If h1 g1 k1 = h2 g2 k2 Hg1 K Hg2 K then g1 = h1
Hg2 K so that
1 h2 g2 k2 k1
Hg1 K Hg2 K, and similarly Hg2 K Hg1 K. Thus if two double cosets are not disjoint,
they coincide.
(3) In this statement G acts from the left on the left cosets G/K, hence so does H
by restriction of the action. We denote the set of H-orbits on G/K by H\(G/K). The
mapping
{double cosets} H\(G/K)
HgK 7 H(gK)
70
is evidently well-defined and surjective. If H(g1 K) = H(g2 K) then g2 K = hg1 K for some
h H, so g2 Hg1 K and Hg1 K = Hg2 K by (2). Hence the mapping is injective.
The proof that double cosets biject with (H\G)/K is similar.
In view of (3) we denote the set of (H, K)-double cosets in G by H\G/K. We denote
a set of representatives for these double cosets by [H\G/K].
(5.2) Example. Consider S2 = {1, (12)} as a subgroup of S3 . We have
S2 \S3 /S2 = {{1, (12)}, {(123), (132), (13), (23)}},
while, for example,
[S2 \S3 /S2 ] = {1, (123)}.
S3 acts transitively on {1, 2, 3} with StabS3 (3) = S2 , so as S3 -sets we have
S3 /S2
= {1, 2, 3}.
Thus the set of orbits on this set under the action of S2 is
S2 \(S3 /S2 ) {{1, 2}, {3}}.
We observe that these orbits are indeed in bijection with the double cosets S2 \S3 /S2 .
This example illustrates the point that when computing double cosets it may be
advantageous to identify G/K as some naturally occurring G-set, rather than as the set
of left cosets.
In the next result we distinguish between conjugation on the left and on the right:
g
x = gxg 1 and xg = g 1 xg. Later on we will write cg (x) = gx, so that cg : H gH is
the homomorphism that is left conjugation by g, and cg 1 (x) = xg .
(5.3) PROPOSITION. Let H, K be subgroups of G and g G an element. We have
isomorphisms
HgK/K
= H/(H gK) as left H-sets
and
H\HgK
= (H g K)\K
as right K-sets.
Thus the double coset HgK is a union of |H : H gK| left K-cosets and |K : H g K|
right H-cosets. We have
X
|G : K| =
|H : H gK|
g[H\G/K]
and
|G : H| =
g[H\G/K]
|K : H g K|.
Peter Webb
71
Proof. HgK is the union of a single H-orbit of left K-cosets. The stabilizer in H of
one of these is

StabH (gK) = {h H hgK = gK}

= {h H hg K = K}

= {h H hg K}
= H gK.
Thus HgK/K
= H/(H gK) as left H-sets and the number of left K-cosets in HgK
equals |H : H gK|. By summing these numbers over all double cosets we obtain the total
number of left K-cosets |G : K|.
The argument with right H-cosets is similar.
We next introduce conjugation of representations, a concept we have in fact already
met with induced representations. Suppose H is a subgroup of G, g G and V is a
representation of H. We define a representation gV of gH by specifying that gV = V as a
set, and if gh gH then ghv = hv. Thus if : H GL(V ) was the original representation,
cg 1
the conjugate representation is the composite homomorphism gH H GL(V ) where

cg 1 ( gh) = h.
L
When studying the structure of induced representations V G
H=
g[G/H] g V , the
g
subspace g V is in fact a representation for H; for
ghg 1 (g v) = ghg 1 g v = gh v = g hv.
When g V is identified with V via the linear isomorphism g v 7 v, the action of gH
on V that arises coincides with the action we have just described on gV .
(5.4) THEOREM (Mackey decomposition formula). Let H, K be subgroups of G
and V a representation for K over a commutative ring R. Then
M
G
H
(V G
)
( g(V K
=
K
H
H g K )) H gK
g[H\G/K]
as RH-modules.
Proof. We have V G
K=
The terms
x[G/K] x
V . Consider a particular double coset HgK.
x[G/K]
xHgK
xV
form an R-submodule invariant under the action of H, since it is the direct sum of an orbit
of R-submodules permuted by H. Now

StabH (g V ) = {h H hg V = g V }

= {h H g 1 hg StabG (1 V ) = K}
= H gK.
72
Therefore as a representation for H this subspace is (g V ) H

H gK by Proposition 4.8.
As observed before the statement of this theorem we have g V
= g(V K
H g K ) as a
g
representation of H K. Putting these expressions together gives the result.
As an application of Mackeys theorem we consider permutation modules arising from
multiply transitive G-sets. We say that a G-set is n-transitive (or, more properly, the
action of G on is n-transitive) if has at least n elements and for every pair of ntuples (a1 , . . . , an ) and (b1 , . . . , bn ) in which the ai are distinct elements of and the bi
are distinct elements of , there exists g G with gai = bi for every i. For example,
Sn acts n-transitively on {1, . . . , n}, and one may show that An acts (n 2)-transitively
on {1, . . . , n} provided n 3. Notice that if G acts n-transitively on then it also acts
(n 1)-transitively on .
(5.5) LEMMA. Let be a G-set with at least n elements (where n 1) and let
. Then G acts n-transitively on if and only if G acts transitively on and
StabG () acts (n 1)-transitively on {}.
Proof. If G acts n-transitively then G also acts transitively, and if a2 , . . . , an and
b2 , . . . , bn are two lists of n 1 distinct points of , none of them equal to , then there
exists g G so that g() = () and g(ai ) = bi for all i. This shows that StabG () acts
(n 1)-transitively on {}.
Conversely, suppose G acts transitively on and StabG () acts (n 1)-transitively
on {}. Let a1 , . . . , an and b1 , . . . , bn be two lists of n distinct points of . We may
find u, v G so that ua1 = and v = b1 , by transitivity of G on . The elements
, ua2 , . . . , uan are distinct, as are the points , v 1 b2 , . . . , v 1 bn , so we can find g
StabG so that guai = v 1 bi when 2 i n. Now vguai = bi for 1 i n and this
shows that G acts n-transitively on .
(5.6) PROPOSITION. Whenever is a G-set the permutation module C may be
written as a direct sum of CG-modules
C = C V
for some module V . Suppose that || 2, so V 6= 0. The representation V is simple if
and only if G acts 2-transitively on . In that case, V is not the trivial representation.
Proof. Pick any orbit of G on . It is isomorphic as a G-set to G/H for some subgroup
H G and so C[G/H] is a direct summand of C, with character 1 G
H . Since
h1, 1 G
H iG = h1, 1iH = 1
by Frobenius reciprocity we deduce that C is a summand of C[G/H] and hence of C.
Peter Webb
73
In the equivalence of statements that forms the third sentence, neither side is true if
G has more than one orbit on , so we may assume = G/H. The character of C is
1 G
H , and we compute
G
G
G
h1 G
H , 1 H iG = h(1 H ) H , 1iH
X
=h
( g1) H
H gH , 1iH
g[H\G/H]
h1 H
H gH , 1iH
g[H\G/H]
h1, 1iH gH
g[H\G/H]
g[H\G/H]
= |H\G/H|,
using Frobenius reciprocity twice and Mackeys formula. Now |H\G/H| is the number of
orbits of H (the stabilizer of a point) on G/H. By Lemma 5.5 this number is 2 if G acts
2-transitively on , and otherwise it is greater than 2 (since || 2 was a hypothesis).
Writing C[G/H] = S1 Sn as a direct sum of simple representations we have
G
h1 G
H , 1 H iG n,
and we get the value 2 for the inner product if and only if there are 2 simple representations
in this expression, and they are non-isomorphic. This is equivalent to requiring that V is
simple, because it could only be the trivial representation if G acts trivially on G/H, which
our hypotheses exclude. In any case we deduce that V is not the trivial representation.
(5.7) EXAMPLE. Let = {1, . . . , n} acted upon transitively by Sn and also by An .
Then C
= C V where V is a simple representation of Sn , which remains simple on
restriction to An provided n 4.
We now turn to Cliffords theorem, which we present in a weak and a strong form.
The weak form is used as a step in proving the strong form a little later, and as a result
in its own right it only has force in a situation where |G| is not invertible in the ground
ring. In these versions of Cliffords theorem we make the hypothesis that the ground ring
is a field, but this is no loss of generality in view of Exercise 13 from Chapter 1.
(5.8) THEOREM (Weak form of Cliffords theorem). Let k be any field, U a simple
kG-module and N a normal subgroup of G. Then U G
N is semisimple as a kN -module.
Proof. Let V be any simple kN -submodule of U G
N . For every g G, gV is also
1
a kN -submodule since if n N we have n(gv) = g(g ng)v gV , using the fact that
74
N is normal. Evidently gV is also simple, since if W were a kN -submodule of gV then

P
g 1 W would be a submodule of V . Now gG gV is a non-zero G-invariant subspace of
P
the simple kG-module U , and so gG gV = U . As a kN -module we see that U G
N is the
G
sum of simple submodules, and hence U N is semisimple by the results of Chapter 1.
The kN -submodules gV that appear in the proof of 5.8 are isomorphic to modules we
have seen before. Since N G, the conjugate module gV is a representation for gN = N .
The mapping
g
V gV
v 7 gv
is an isomorphism of kN -modules, since if n N the action on gV is n v = g 1 ngv and

the action on gV is n(gv) = g(g 1 ngv). Recall also that these modules appeared when we
described induced modules. Part of Cliffords theorem states that the simple module U is
in fact an induced module.
(5.9) THEOREM (Cliffords theorem). Let k be any field, U a simple kG-module
a1
ar
and N a normal subgroup of G. We may write U G
N = S1 Sr where the Si
are non-isomorphic simple kN -modules, occurring with multiplicities ai . (We refer to the
summands Siai as the homogeneous components.) Then
(1) G permutes the homogeneous components transitively;
(2) a1 = a2 = = ar and dim S1 = dim S2 = = dim Sr ; and
(3) if H = StabG (S1a1 ) then U
= S1a1 G
H as kG-modules.
Proof. The fact that U G
N is semisimple and hence can be written as a direct sum as
claimed follows from 5.8. We observe that, by Corollary 1.9, the homogeneous component
Siai is characterized as the unique largest kN -submodule that is isomorphic to a direct sum
of copies of Si . If g G then g(Siai ) is a direct sum of isomorphic simple modules gSi ,
and so by this characterization must be contained in one of the homogeneous components:
a
g(Siai ) Sj j for some j. Since U = g(S1a1 ) g(Srar ), by counting dimensions we have
P
a
g(Siai ) = Sj j . Thus G permutes the homogeneous components. Since gG g(S1a1 ) is a
non-zero G-invariant submodule of the simple module U , it must equal U , and so the action
on the homogeneous components is transitive. This establishes (1), and (2) follows since
a
for any pair (i, j) we can find g G with g(Siai ) = Sj j , so ai = aj and dim Si = dim Sj .
Finally, (3) is a direct consequence of Proposition 4.8.
For now, we give just one application of Cliffords theorem, which is Corollary 5.11.
In the proof of 5.11 we will need a fact about representations of abelian groups that so far
we have only proved when |G| is invertible in Corollary 2.7 (and over C in Theorem 4.4).
Peter Webb
75
(5.10) THEOREM. Let k be any algebraically closed field. If G is abelian then every
simple kG-module has dimension 1.
Proof. Consider a simple kG-module S and let g G. In its action on S, g has
an eigenvalue , with non-zero eigenspace S . Since all elements h G commute with g
we have hS = S (by the argument that if v S then gv = v, so g(hv) = h(gv) =
hv = hv, so hv S ; but also the action of h is invertible). Thus S is a kG-submodule
of S, so S = S by simplicity of S. It follows that every element g G acts by scalar
multiplication on S, and such a simple module S must have dimension 1.
As a consequence of the last result and Proposition 4.5, over an algebraically closed
field the degree 1 representations of any group G are the same as the representations of
G/G , lifted to G via the quotient homomorphism G G .
The next result about simple representations of p-groups is true as stated when k
has characteristic p, but it has no force in that situation because (as we will see in the
next section) the only simple representation of a p-group in characteristic p is the trivial
representation. We are thus only really interested in the following result over fields of
characteristic other than p, and in particular over fields of characteristic 0.
(5.11) COROLLARY. Let k be any algebraically closed field and G a p-group. Then
every simple module for G has the form U G
H where U is a 1-dimensional module for some
subgroup H.
Proof. We proceed by induction on |G|. Let : G GL(S) be a simple representation
of G over k and put N = Ker . Then S is really a representation of G/N . If N 6= 1 then
G/N is a group of smaller order than G, so by induction S has the claimed structure as
a representation of G/N , and hence also as a representation of G. Thus we may assume
N = 1 and G embeds in GL(S).
If G is abelian then all simple representations are 1-dimensional, so we are done.
Assume now that G is not abelian. Then G has a normal abelian subgroup A that is not
central. To construct this subgroup A, let Z2 (G) denote the second centre of G, that is,
the preimage in G of Z(G/Z(G)). If x is any element of Z2 (G) Z(G) then A = hZ(G), xi
is a normal abelian subgroup not contained in Z(G).
We apply Cliffords theorem:
a1
ar
S G
A = S1 Sr
a1
a1
and S = V G
K where V = S1 and K = StabG (S1 ). We argue that V must be a
simple kK-module, since if it had a proper submodule W then W G
K would be a proper
submodule of S, which is simple. If K 6= G then by induction V = U K
H where U is
K
G
G
1-dimensional, and so S = (U H ) K = U H has the required form.
76
We show finally that the case K = G cannot happen. For if it were to happen then
S
S1a1 , and since A is abelian dim S1 = 1. The elements of A must therefore act via
scalar multiplication on S. Since such an action would commute with the action of G,
which is faithfully represented on S, we deduce that A Z(G), a contradiction.
G
A=
The above result is evidently useful if we are constructing the character table of a
p-group, because it says that we need look no further than induced characters. We note
that the conclusion of Corollary 5.11 also applies to supersolvable groups, which again
have the property, if they are not abelian, that they have a non-central normal abelian
subgroup.
A representation of the form U G
H for some subgroup H and with U a 1-dimensional
representation of H is said to be monomial. A group G all of whose irreducible complex
representations are monomial is called an M-group. Thus p-groups (and also supersoluble
groups) are M-groups.
1. The Mackey formula: induction followed by restriction is a sum over double cosets
of restriction followed by conjugation followed by induction.
2. A permutation representation is 2-transitive if and only if the complex permutation
module has two summands.
3. Cliffords theorem: the restriction of a simple module to a normal subgroup is
semisimple, and the module is induced from the stabilizer of a homogeneous component.
4. For a p-group over an algebraically closed field, every simple module is induced
from a 1-dimensional module.
Peter Webb
77

1. Let k be any field, and g any element of a finite group G.
(a) If K H G are subgroups of G, V a kH-module, and W a kK-module, show that
g
g
H
g
H g
H
g
( gV ) gH
K = (V K ) and ( W ) gK = (W K ). [This allows us to put conjugation before,
between, or after restriction and induction in Mackeys formula.]
(b) If U is any kG-module, show that U
= g U by showing that one of the two mappings
g
1
U U specified by u 7 gu and u 7 g u is always an RG-module isomorphism. [Find
which one of these it is.]
2. Let H and K be subgroups of G with HK = G and H K = 1. Show that for any
G
kH-module U the module U G
H K is a direct sum of copies of the regular representation
kK.
3. Let k be a field. Show by example that it is possible to find a subgroup H of a
group G and a simple kG-module U for which U G
H is not semisimple.
4. The following statements generalize Lemma 3.5 and Maschkes theorem. Let H be
a subgroup of G and suppose that k is a field in which |G : H| is invertible. Let V be a
kG-module.
(a) Show that
X
1
g :VH VG
|G : H|
g[G/H]
is a well-defined map that is a projection of the H-fixed points onto the G-fixed points.
In particular, this map is surjective.
(b) Show that if V G
H is semisimple as a kH-module then V is semisimple as a kG-module.
5. Let H be a normal subgroup of G and suppose that k is a field of characteristic p.
(a) Let p6 |G : H|. Show that if U is a semisimple kH-module then U G
H is a semisimple
kG-module.

(b) Let p |G : H|. Show by example that if U is a semisimple kH-module then it need
not be the case that U G
H is a semisimple kG-module.
6. Let H be a subgroup of G of index 2 (so that H is normal in G) and let k be a
field whose characteristic is not 2. The homomorphism G {1} k with kernel H is a
1-dimensional representation of G that we will call . Let S, T be simple kG-modules and
G
let U, V be simple kH-modules. You may assume that U G
H and V H are semisimple
(this is proved as Exercise 5(a)). Let g G H.
(a) Show that S G
H is the direct sum of either 1 or 2 simple kH-modules.
(b) Show that U G
H is the direct sum of either 1 or 2 simple kG-modules.
In the following questions, notice that
G
G
S G
H H = S (k H ) = S (k ) = S (S ).
For some parts of the questions it may help to consider

G
HomkH (S G
H , T H )
and
G
HomkG (U G
H , V H ).
78
(c) Show that the following are equivalent:

(i) S is the induction to G of a kH-module,
(ii) S G
H is not simple,
(iii) S = S .
(d) Show that the following are equivalent:
(i) U is the restriction to H of a kG-module,
(ii) U G
H is not simple,
(iii) U = g U .
G
(e) Show that S G

H and T H have a summand in common if and only if S = T or
S
= T .
G
(f) Show that U G

H and V H have a summand in common if and only if U = V or
g
U = V.
(g) We place an equivalence relation 1 on the simple kG-modules and an equivalence
relation 2 on the simple kH-modules:
S 1 T S
= T or S
=T
U 2 V U
= V or U
= g V.
G
Show that induction G
H and restriction H induce mutually inverse bijections between the
equivalence classes of simple kG-modules and of simple kH-modules in such a way that an
equivalence class of size 1 corresponds to an equivalence class of size 2, and vice-versa.
(h) Show that the simple kG-modules of odd degree restrict to simple kH-modules,
and the number of such modules is even.
(i) In the case where G = S4 , H = A4 and k = C, show that there are three equivalence
classes of simple characters under 1 and 2 . Verify that SA44 and SA44 give mutually inverse
bijections between the equivalence classes.
7. Let G = GL(3, 2) be the group of 3 3 invertible matrices over k = F2 and let
a b
H ={ c d
e f
0
0 a, b, c, d, e, f, 1 F2 , (ad bc) 6= 0}
1
You may assume from group theory that |G| = 168. Let V be the natural 3-dimensional
space of column vectors on which G-acts.
(a) Show that |H| = 24, so that |G : H| = 7.
(b) Show that V is simple as a kG-module.
(c) Show that as a kH-module V G
H has a simple socle with trivial H action, and such
that the quotient of V by the socle is a simple 2-dimensional module.
G
(d) Show that dim HomkG (k G
H , V ) = 1 and dim HomkG (V, k H ) = 0.

(e) Show that k G
H is not semisimple, thereby showing that even if p 6 |G : H| it need
not be the case that the induction of a simple module is semisimple when H is not
normal.
Peter Webb
79
G
(f) Show that dim HomkG (k G
H , V ) = 0 and dim HomkG (V , k H ) = 1. Show that
V 6
= V . Show that k G
H is the direct sum of the trivial module k and a 6-dimensional
module that has socle V and socle quotient V .
8. Find the complete list of subgroups H of the dihedral group D8 such that the
2-dimensional simple representation over C can be written U G
H for some 1-dimensional
representation U of H. Do the same thing for the quaternion group Q8 .
9. Compute the character tables of the generalized quaternion group of order 16

Q16 = hx, y x8 = 1, x4 = y 2 , yxy 1 = x1 i
and the semidihedral group of order 16:

SD16 = hx, y x8 = y 2 = 1, yxy 1 = x3 i.
80
6. Representations of p-groups in characteristic p and the radical

The study of representations of a group over a field whose characteristic divides the
group order is more delicate than the case of ordinary representation theory. Modules no
longer need be semisimple, and to determine the isomorphism type of a module we have to
do more than count multiplicities of simple direct summands. In this chapter we begin the
task of assembling techniques specifically aimed at understanding the structure of modules
where we have to go beyond direct sum decompositions. We introduce the radical and
socle series of modules, which are a fundamental tool. We first focus on representations of
p-groups in characteristic p because in this situation specific things may be said with very
little background preparation, and which have impact on representations of all groups.
We will show that there is only one simple module in this situation, and that the regular
representation is indecomposable, identifying its radical as the augmentation ideal. In spite
of the title of this section, some of our deductions are to groups that are not p-groups.
Cyclic p-groups
We start by describing representations of cyclic p-groups because it is possible to give
the complete picture in an elementary fashion. In the first proposition we reduce their
study to that of modules for a principal ideal domain. When G is cyclic of order N we
have already made use of an isomorphism between kG and k[X]/(X N 1) in Exercise 9
from Section 2. When N is a power of p we can express this slightly differently.
n
(6.1) PROPOSITION. Let k be a field of characteristic p and let G = hg g p = 1i
n
be cyclic of order pn . Then there is a ring isomorphism kG
= k[X]/(X p ), where k[X] is
the polynomial ring in an indeterminate X.
Proof. We define a mapping
n
G k[X]/(X p )
g s 7 (X + 1)s .
Since
(X + 1)p = X p + p( ) + 1 1 (mod(X p )),
this mapping is a group homomorphism to the unit group of k[X]/(X p ), and hence it
extends to a linear map
n
kG k[X]/(X p )
that is an algebra homomorphism. Since g s is sent to X s plus terms of lower degree, the
n
n
images of 1, . . . , g p 1 form a basis of k[X]/(X p ). The mapping therefore gives a bijection
n
between a basis of kG and a basis of k[X]/(X p ), and so is an isomorphism.
Peter Webb
81
Direct sum decompositions of modules are the first consideration in describing their
structure. We say that a module U for a ring A is indecomposable if it cannot be expressed
as a direct sum of two modules except in a trivial way, that is, if U
= V W then either
V = 0 or W = 0. When A is an algebra over a field, by repeatedly expressing summands of
a module as further direct sums we can express any finite dimensional module as a direct
sum of indecomposable direct summands. It is useful to know, but we will not prove it until
Theorem 11.16, that for each module these summands are determined up to isomorphism,
independently of the choice of direct sum decomposition. This is the content of the KrullSchmidt Theorem. We point out that, since we need not be in characteristic zero when
group representations are semisimple, indecomposable modules need not be simple. This
is the point of introducing the new terminology! An example of an indecomposable module
that is not simple was given 1.3, and we will see many more examples.
A module over a ring is said to be cyclic if it can be generated by one element. We now
exploit the structure theorem for finitely-generated modules over a principal ideal domain,
which says that such modules are direct sums of cyclic modules.
(6.2) THEOREM. Let k be a field of characteristic p. Every finitely-generated
n
k[X]/(X p )-module is a direct sum of cyclic modules Ur = k[X]/(X r ) where 1 r pn .
The only simple module is the 1-dimensional module U1 . Each module Ur has a unique
composition series, and hence is indecomposable. From this it follows that if G is cyclic of
order pn then kG has exactly pn indecomposable modules, one of each dimension i with
1 i pn , each having a unique composition series.
n
Proof. The modules for k[X]/(X p ) may be identified with the modules for k[X] on
n
which X p acts as zero. Every finitely-generated k[X]-module is a direct sum of modules
n
k[X]/I where I is an ideal. Hence every k[X]/(X p )-module is a direct sum of modules
n
n
k[X]/I on which X p acts as zero, which is to say (X p ) I. The ideals I that satisfy this
n
last condition are the ideals (a) where a X p . This forces I = (X r ) where 1 r pn ,
and k[X]/I = Ur .
The submodules of Ur must have the form J/(X r ) where J is some ideal containing
r
(X ), and they are precisely the submodules in the chain
0 (X r1 )/(X r ) (X r2 )/(X r ) (X)/(X r ) Ur .
This is a composition series, since each successive quotient has dimension 1, and since it
is a complete list of submodules, it is the only one. If we could write Ur = V W as
a non-trivial direct sum, then Ur would have at least 2 composition series, obtained by
taking first a composition series for V , then one for W , or vice-versa. Hence each Ur is
indecomposable and we have a complete list of the indecomposable modules. The only Ur
that is simple is U1 , which is the trivial module.
The final identification of the indecomposable kG-modules comes from the isomorphism in Proposition 6.1.
82
A module with a unique composition series is said to be uniserial. It is equivalent to

say that its submodules are linearly ordered by inclusion, and the equivalence of these and
n
other conditions is explored in the exercises. We see from the description of k[X]/(X p )modules that Ur has a basis 1 + (X r ), X + (X r ), . . . , X r1 + (X r ) so that X acts on Ur
with matrix
1 0
.
.. ..
.
.
1
Translating now to modules for kG where G is a cyclic p-group, the generator g acts on
Ur as X + 1, which has matrix
1
1
1
..
..
1
.
1
Thus we see that the indecomposable kG-modules are exactly given by specifying that the
generator g acts via a matrix that is a single Jordan block, of size up to pn . It is helpful
to picture Ur using a diagram
X=g1
y
X=g1
y
Ur = .
..
X=g1
y
that may be interpreted by saying that the vertices are in bijection with a basis of Ur , and
the action of X or g 1 is given by the arrows. Where no arrow is shown starting from a
particular vertex (as happens in this case only with the bottom vertex), the interpretation
is that X and g 1 act as zero.
Peter Webb
83
Simple modules for p-groups and normal p-subgroups

The following seemingly innocuous result has profound consequences throughout the
rest of this book.
(6.3) PROPOSITION. Let k be a field of characteristic p and G a p-group. The only
simple kG-module is the trivial module.
Proof. We offer two proofs of this.
Proof 1. We proceed by induction on |G|, the induction starting when G is the identity
group, for which the result is true. Suppose G 6= 1 and the result is true for p-groups of
smaller order. There exists a normal subgroup N of G of index p. If S is any simple
kG-module then by Cliffords theorem S G
N is semisimple. By induction, N acts trivially
on S. Thus S is really a representation of G/N that is cyclic of order p. We have just
proved that the only simple representation of this group is the trivial representation.
Proof 2. Let S be any simple
kG-module and let 0 6= x S. The subgroup of S

generated by the elements {gx g G} is invariant under the action of G, it is abelian
and of exponent p, since it is a subgroup of a vector space in characteristic p. Thus it is a
finite p group acted on by G. Consider the orbits of G on this finite group. Since G is a
p-group the orbits all have size a power of p (or 1), because the size of an orbit is the index
of the stabilizer of an element in the orbit. The zero element is fixed by G, and we deduce
that there must be another element fixed by G since otherwise the other orbits would all
have size pn with n 1, and their union would not be a p-group. Thus there exists y S
fixed by G, and now hyi is a trivial submodule of S. By simplicity it must equal S.
As an application of this we can give some information about the simple representations of arbitrary finite groups in characteristic p. For this we observe that in every finite
group G there is a unique largest normal p-subgroup of G, denoted Op (G). For if H and
K are normal p-subgroups of G then so is HK, and thus the subgroup generated by all
normal p-subgroups of G is a again a normal p-subgroup, that evidently contains all the
others.
(6.4) COROLLARY. Let k be a field of characteristic p and G a finite group. Then
the common kernel of the action of G on all the simple kG-modules is Op (G). Thus the
simple kG-modules are precisely the simple k[G/Op (G)]-modules, made into kG-modules
via the quotient homomorphism G G/Op (G).
Proof. Let H be the kernel of the action of G on all simple kG-modules, that is,

H = {g G for all simple S and for all s S, gs = s}.
By Cliffords theorem, if S is a simple kG-module then S G

Op (G) is semisimple. Therefore,
by 6.3, Op (G) acts trivially on S, so that Op (G) H. We show that H contains no
84
element of order prime to p. For, suppose h H were to have order prime to p. Then
G
kG G
hhi would be a semisimple khhi-module that is the direct sum of modules S hhi with
S a simple kG-module. Since h acts trivially on all of these, it must act trivially on kG,
which is a contradiction. Therefore H is a p-group, and since it is normal, Op (G) H.
We therefore have equality. The last sentence is immediate.
(6.5) Examples. When G is nilpotent (and this applies, of course, if G is abelian) it
is a direct product G = H K where H is a Sylow p-subgroup of G, and so Op (G) = H.
Thus the simple kG-modules may be identified with the simple kK-modules when k has
characteristic p. For a different example, let k be a field of characteristic 2, and consider
the representations of A4 over k. Since O2 (A4 ) = C2 C2 , the simple kA4 modules are
the simple C3 = A4 /O2 (A4 )-representations, made into representations of A4 . Now kC3
is semisimple, and if k contains a primitive cube root of unity (i.e. if F4 k) there are
three 1-dimensional simple representations, on which the generator of C3 acts as 1, or
2.
Radicals and socles

At this point we examine further the structure of representations that are not semisimple, and we work in the context of modules for a ring A, that is always supposed to have a
1. At the end of section 1 we defined the socle of an A-module U to be the sum of all the
simple submodules of U , and we showed (at least in the case that U is finite-dimensional)
that it is the unique largest semisimple submodule of U . We now work with quotients and
define a dual concept, the radical of U . We work with quotients instead of submodules,
and use the fact that if M is a submodule of U , the quotient U/M is simple if and only if
M is a maximal submodule of U . We put
\

Rad U = {M M is a maximal submodule of U }.
In our applications U will always be Noetherian, so provided U 6= 0 this intersection will

be non-empty and hence Rad U 6= U . If U has no maximal submodules (for example,
if U = 0, or in more general situations than we consider here where U might not be
Noetherian) we set Rad U = U .
(6.6) LEMMA. Let U be a module for a ring A.
(1) Suppose that M1 , . . . , Mn are maximal submodules of U . Then there is a subset
I {1, . . . , n} such that
U/(M1 Mn )
=
which, in particular, is a semisimple module.
M
iI
U/Mi
Peter Webb
85
(2) Suppose further that U has the descending chain condition on submodules. Then
U/ Rad U is a semisimple module, and Rad U is the unique smallest submodule of U
with this property.
Proof. (1) Let I be a subset of {1, . . . , n} maximal with the property that the
T
T
quotient homomorphisms U/( iI Mi ) U/Mi induce an isomorphism U/( iI Mi )
=
L
T
Mn and argue by contradiction. If it
iI U/Mi . We show that
iI Mi = M1
T
were not the case, there would exist Mj with iI Mi 6 Mj . Consider the homomorphism
f :U (
M
iI
U/Mi ) U/Mj
whose components are the quotient homomorphisms U U/Mk . This has kernel Mj
T
iI Mi , and it will suffice to show that f is surjective, because this will imply that the
larger set I {j} has the same property as I, thereby contradicting the maximality of I.
T
To show that f is surjective let g : U U/ iI Mi U/Mj and observe that
T
( iI Mi ) + Mj = U since the left-hand side is strictly larger than Mj , which is maxT
imal in U . Thus if x U we can write x = y + z where y iI Mi and z Mj . Now
T
T
g(y) = (0, x + Mj ) and g(z) = (x + iI Mi , 0) so that both summands U/ iI Mi and
U/Mj are contained in the image of g and g is surjective. Since f is obtained by composing
T
L
g with the isomorphism that identifies U/ iI Mi with iI U/Mi , we deduce that f is
surjective.
(2) By the assumption that U has the descending chain condition on submodules,
Rad U must be the intersection of finitely many maximal submodules. Therefore U/ Rad U
is semisimple by part (1). If V is a submodule such that U/V is semisimple, say U/V
=
proj.
S1 Sn where the Si are simple modules, let Mi be the kernel of U U/V Si .
Then Mi is maximal and V = M1 Mn . Thus V Rad U , and Rad U is contained
in every submodule V for which U/V is semisimple.
We define the radical of a ring A to be the radical of the regular representation
Rad A A and write simply Rad A. We present some identifications of the radical that are
very important theoretically, and also in determining what it is in particular cases.
(6.7) PROPOSITION. Let A be a ring. Then,

(1) Rad A = {a A a S = 0 for every simple A-module S}, and
(2) Rad A is a 2-sided ideal of A.
(3) Suppose further that A is a finite-dimensional algebra over a field. Then
(a) Rad A is the smallest left ideal of A such that A/ Rad A is a semisimple A-module,
(b) A is semisimple if and only if Rad A = 0,
(c) Rad A is nilpotent, and is the largest nilpotent ideal of A.
86
(d) Rad A is the unique ideal U of A with the property that U is nilpotent and A/U
is semisimple.
Proof. (1) Given a simple module S and 0 6= s S, the module homomorphism
A A S given by a 7 as is surjective and its kernel is a maximal left ideal Ms . Now if
a Rad A then a Ms for every S and s S, so as = 0 and a annihilates every simple
module. Conversely, if a S = 0 for every simple module S and M is a maximal left ideal
then A/M is a simple module. Therefore a (A/M ) = 0, which means a M . Hence
T
a maximalM M = Rad A.
(2) Being the intersection of left ideals, Rad A is also a left ideal of A. Suppose that
a Rad A and b A, so a S = 0 for every simple S. Now a bS a S = 0 so ab has the
same property that a does.
(3) (a) and (b) are immediate from 6.6. We prove (c). Choose any composition series
0 = An An1 A1 A0 = A A
of the regular representation. Since each Ai /Ai+1 is a simple A-module, Rad A Ai Ai+1
by part (1). Hence (Rad A)r A Ar and (Rad A)n = 0.
Suppose now that I is a nilpotent ideal of A, say I m = 0, and let S be any simple
A-module. Then
0 = I m S I m1 S IS S
is a chain of A-submodules of S that are either 0 or S since S is simple. There must be
some point where 0 = I r S 6= I r1 S = S. Then IS = I I r1 S = I r S = 0, so in fact
that point was the very first step. This shows that I Rad A by part (1). Hence Rad A
contains every nilpotent ideal of A, so is the unique largest such ideal.
Finally (d) follows from (a) and (c): these imply that Rad A has the properties stated
in (d); and, conversely, these conditions on an ideal U imply by (a) that U Rad A, and
by (c) that U Rad A.
Note that if I is a nilpotent ideal of A then it is always true that I Rad(A) without
the assumption that A is a finite dimensional algebra. The argument given to prove part
3c of Proposition 6.7 shows this.
For any group G and commutative ring R with a 1, the ring homomorphism
: RG R
g 7 1 for all g G
is called the augmentation map. As well as being a ring homomorphism it as a homomorphism of RG-modules, in which case it expresses the trivial representation as a homomorphic image of the regular representation. The kernel of is called the augmentation ideal,
P
and is denoted IG. Evidently IG consists of those elements gG ag g RG such that
P
gG ag = 0. We now show that when k is a field of characteristic p and G is a p-group
this construction gives the radical of kG.
Peter Webb
87
(6.8) PROPOSITION. Let G be a finite group and R a commutative ring with a 1.

(1) Let R denote the trivial RG-module. Then IG = {x RG x R = 0}.

(2) IG is free as an R-module with basis {g 1 1 6= g G}.
(3) If R = k is a field of characteristic p and G is a p-group then IG = Rad(kG). It
follows that IG is nilpotent in this case.
Proof. (1) The augmentation map is none other than the linear extension to RG of
the homomorphism : G GL(1, R) that is the trivial representation. Thus each x RG
acts on R as multiplication by (x), and so will act as 0 precisely if (x) = 0.
(2) The elements g 1 where g ranges through the non-identity elements of G are
linearly independent since the elements g are, and they lie in IG. We show that they span
P
P
IG. Suppose gG ag g IG, which means that gG ag = 0 R. Then
X
gG
ag g =
gG
ag g
gG
ag 1 =
16=gG
ag (g 1)
is an expression as a linear combination of elements g 1.

(3) When G is a p-group and char(k) = p we have seen in 6.3 that k is the only simple
kG-module. The result follows by part (1) and 6.7.
Working in the generality of a finite-dimensional algebra A again, the radical of A
allows us to give a further description of the radical and socle of a module. We present
this result for finite-dimensional modules, but it is in fact true without this hypothesis.
We leave this stronger version to the exercises.
(6.9) PROPOSITION. Let A be a finite-dimensional algebra over a field k, and U a
finite-dimensional A-module.
(1) The following are all descriptions of Rad U :
(a) the intersection of the maximal submodules of U ,
(b) the smallest submodule of U with semisimple quotient,
(c) Rad A U .
(2) The following are all descriptions of Soc U :
(a) the sum of the simple submodules of U ,
(b) the largest semisimple submodule of U ,

(c) {u U Rad A u = 0}.
Proof. Under the hypothesis that U is finitely generated we have seen the equivalence
of descriptions (a) and (b) in 1.8 and 6.6. Our arguments below actually work without
the hypothesis of finite generation, provided we assume the results of Exercises 11 and 12
from Chapter 1. The reader who is satisfied with a proof for finitely generated modules
can assume that the equivalence of (a) and (b) has already been proved.
88
Let us show that the submodule Rad A U in (1)(c) satisfies condition (1)(b). Firstly
U/(Rad AU ) is a module for A/ Rad A, which is a semisimple algebra. Hence U/(Rad AU )
is a semisimple module and so Rad A U contains the submodule of (1)(b). On the other
hand if V U is a submodule for which U/V is semisimple then Rad A (U/V ) = 0 by
6.7, so V Rad A U . In particular, the submodule of (1)(b) contains Rad A U . This
shows that the descriptions in (1)(b) and (1)(c) are equivalent.
To show that they give the same submodule as (1)(a), observe that if V is any maximal
submodule of U , then as above (since U/V is simple) V Rad A U , so the intersection of
maximal submodules of U contains Rad A U . The intersection of maximal submodules of
the semisimple module U/(Rad A U ) is zero, so this gives a containment the other way,
since they all correspond to maximal submodules of U . We deduce that the intersection
of maximal submodules of U equals Rad A U .

For the conditions in (2), observe that {u U Rad A u = 0}, this is the largest submodule of U annihilated by Rad A. It is thus an A/ Rad A-module and hence is semisimple.
Since every semisimple submodule of U is annihilated by Rad A, it is the unique largest
such submodule.
(6.10) Example. Consider the situation of 6.1 and 6.2 in which G is a cyclic group of
order pn and k is a field of characteristic p. We see that Rad Ur
= Ur1 and Soc Ur
= U1
n
for 1 r p , taking U0 = 0.
We now iterate the notions of socle and radical: for each A-module U we define
inductively
Radn (U ) = Rad(Radn1 (U ))
Socn (U )/ Socn1 (U ) = Soc(U/ Socn1 U ).
It is immediate from 6.9 that
Radn (U ) = (Rad A)n U

Socn (U ) = {u U (Rad A)n u = 0}
and these submodules of U form chains
Rad2 U Rad U U
0 Soc U Soc2 U
that are called, respectively, the radical series and socle series of U . The radical series of
U is also known as the Loewy series of U . The quotients Radn1 (U )/ Radn (U ) are called
the radical layers, or Loewy layers of U , and the quotients Socn (U )/ Socn1 (U ) are called
the socle layers of U .
The next corollary is a deduction from 6.9, and again it is true without the hypothesis
that the modules be finite-dimensional.
Peter Webb
89
(6.11) COROLLARY. Let A be a finite-dimensional algebra over a field k, and let

U and V be finite-dimensional A-modules. Then for each n we have Radn (U V ) =
Radn (U ) Radn (V ) and Socn (U V ) = Socn (U ) Socn (V ).
Proof. One way to see this is to use the identifications Radn (U V ) = (Rad A)n

(U V ) and Socn (U V ) = {(u, v) U V (Rad A)n (u, v) = 0}.
The next result can be proved in various ways; it is also a consequence of Theorem
7.14 in the next section.
(6.12) COROLLARY. Let k be a field of characteristic p and G a p-group. Then the

regular representation kG is indecomposable.
Proof. If kG = U V is the direct sum of two non-zero modules then Rad kG =
Rad U Rad V where Rad U 6= U and Rad V 6= V , so the codimension of Rad kG in kG
must be at least 2. We know from 6.8 that Rad kG has codimension 1, a contradiction.
(6.13) PROPOSITION. Let A be a finite-dimensional algebra over a field k, and U
an A-module. The radical series of U is the fastest descending series of submodules of U
with semisimple quotients, and the socle series of U is the fastest ascending series of U
with semisimple quotients. The two series terminate, and if m and n are the least integers
for which Radm U = 0 and Socn U = U then m = n.
Proof. Suppose that U2 U1 U0 = U is a series of submodules of U with
semisimple quotients. We show by induction on r that Radr (U ) Ur . This is true when
r = 0. Suppose that r > 0 and Radr1 (U ) Ur1 . Then
r1
Radr1 (U )/(Radr1 (U ) Ur )
= (Rad (U ) + Ur )/Ur Ur1 /Ur
is semisimple, so Radr1 (U ) Ur Rad(Radr1 (U )) = Radr (U ). Therefore Radr (U )

Ur . This shows that the radical series descends at least as fast as the series Ui . The
argument that the socle series ascends at least as fast is similar.
Since A is a finite-dimensional algebra we have (Rad A)r = 0 for some r. Then

Radr U = (Rad A)r U = 0 and Socr U = {u U (Rad A)r u = 0} = U , so the two series
terminate. By what we have just proved, the radical series descends at least as fast as
the socle series and so has equal or shorter length. By a similar argument (using the fact
that the socle series is the fastest ascending series with semisimple quotients) the socle
series ascends at least as fast as the radical series and so has equal or shorter length. We
conclude that the two lengths are equal.
90
The common length of the radical series and socle series of U is called the Loewy
length of the module U , and from the description of the terms of these series we see it is
the least integer n such that (Rad A)n U = 0.
We conclude this section by mentioning without proof the theorem of Jennings on the
Loewy series of kG when G is a p-group and k is a field of characteristic p and summarize
its implications. For proofs, see Bensons book [Ben]. Jennings constructs a decreasing
series of subgroups

r (G) = {g G g 1 modulo Radr (kG)},
that is sometimes called the Jennings series of G. This series of subgroups has the properties
1) [r , s ] r+s ,
2) g p ip for all g i ,
3) r /2r is elementary abelian.
Furthermore, we may generate r as
(p)
r = h[r1 , G], r/p i,

(p)
where 1 = G, r/p is the least integer greater than or equal to r/p, and r is the set
of pth powers of elements of r . Evidently the first term in this series is 1 (G) = G and
we may see that the second term 2 (G) is the Frattini subgroup of G (the smallest normal
subgroup of G for which the quotient is elementary abelian). After that the terms need to
be calculated on a case-by-case basis.
For each i 1 let di be the dimension of i /i+1
as a vector space over Fp , and choose
any elements xi,s of G such that the set {xi,s i+1 1 s di } forms a basis for i /i+1 .
Q i,s
Let x
i,s = xi,s 1 kG. There are |G| products of the form x
i,s , where the factors are
taken in some predetermined order, and 0 i,s p 1. The weight of such a product
P
is defined to be
ii,s . Jennings theorem states that the set of products of weight w is
contained in Radw (kG), and forms a basis of Radw (kG) modulo Radw+1 (kG). Thus the
set of all these products is a basis for kG compatible with the powers of the radical.
For example, if we take an element x of order 4 and an element y of order 2 that
generate the dihedral group of order 8, so D8 = hx, y x4 = y 2 = 1, yxy = x1 i, we have
1 = D8 , 2 = hx2 i, 3 = 1. We may choose x1,1 = x, x1,2 = y, x2,1 = x2 , and note that
x
2 = x2 . Now the products x
1,1 x
22,1 y1,2 = x
1,1 +22,1 y1,2 , where 0 i,s 1, form a
basis of kD8 that is compatible with the powers of the radical. In this special case, these
elements may be simplified as x
i yj with 0 i 3 and 0 j 1.
When doing calculations with group rings of p-groups in characteristic p the basis
given by Jennings is often to be preferred over the basis given by the group elements. This
basis gives a description of the powers of the radical in group-theoretic terms, and it allows
us to deduce a result about the socle series as well. Since each element x
i,s of weight i
Peter Webb
91
contributes factors of weights 0, i, 2i, 3i, . . . , (p 1)i in Jennings basis , the total number
of products of weight w is the coefficient of tw in
2
(1 + t + t + + t
p1 d1
) (1 + t + + t
2(p1) d2
Y (1 tip ) di
i1
(1 ti )
P
and this equals the polynomial w0 (dim Radw (kG)/ dim Radw+1 (kG))tw . We see from
this that the dimensions of the factors in the radical series of kG are symmetric, in that
they are the same if taken in reverse order. Jumping ahead of ourselves for a moment and
using the fact that kG
= kG (see Chapter 8) and also using Exercise 6 to this section,
we may see that when G is a p-group and k is a field of characteristic p the terms of the
radical series and the socle series of kG coincide, although these terms appear in the two
series in the opposite order.
n
1. The group ring of Cpn over a field k of characteristic p is isomorphic to k[X]/(X p ).

The indecomposable modules are all cyclic and uniserial.
2. The only simple module for a p-group in characteristic p is the trivial module.
3. When k is a field of characteristic p, simple kG modules are the same as simple
k[G/Op (G)]-modules.
4. The radical of a finite dimensional algebra over a field is the largest nilpotent ideal
of the algebra.
5. The radical series of a module is the fastest descending series with semisimple
factors, and the socle series is the fast ascending such series.
6. When G is a p-group and k is a field of characteristic p the radical of kG is the
augmentation ideal. The radical series and socle series of kG coincide and the ranks of the
factors are given in terms of the Jennings series of G.
92

1. Let A be a ring. Prove that for each n, Socn A A is a 2-sided ideal of A.
P
2. Let G = gG g as an element of kG, where k is a field.
(a) Show that the subspace kG of kG spanned by G is an ideal.
(b) Show that this ideal is nilpotent if and only if the characteristic of k divides |G|.
(c) Deduce that if kG is semisimple then char(k)6 |G|.
(d) Assuming instead that G is a p-group and char(k) = p, show that kG = Soc(kG), the
socle of the regular representation.
3. Prove that if N is a normal subgroup of G and k is a field then Rad(kN ) =
kN Rad(kG).
[Use the descriptions of the radical in Proposition 6.7 and also Cliffords theorem.]
4. Suppose that U is an indecomposable module with just two composition factors.
Show that U is uniserial.
5. Show that the following conditions are equivalent for a module U that has a
composition series.
(a) U is uniserial (i.e. U has a unique composition series).
(b) The set of all submodules of U is totally ordered by inclusion.
(c) Radr U/ Radr+1 U is simple for all r.
(d) Socr+1 U/ Socr U is simple for all r.
6. Let U be a finitely generated kG-module and U its dual. Show that for each n

Socn (U ) = {f U f (Radn (U )) = 0}
and

Radn (U ) = {f U f (Socn (U )) = 0}.
n
n+1
Deduce that Socn+1 (U )/ Socn (U )
(U )) as kG-modules. [Hint:
= (Rad (U )/ Rad
recall Exercise 9 to Chapter 4.]
7. Show that for each RG-module U , U/(IG U ) is the largest quotient of U on which
G acts trivially. Prove also that U/(IG U )
= R RG U .
[The first sentence means that G does act trivially on the given quotient; and if V is any
submodule of U such that G acts trivially on U/V , then V IG U . By analogy with the
notation for fixed points, this largest quotient on which G acts trivially is called the fixed
quotient of U and is denoted UG := U/(IG U ).]
The next five exercises give a direct proof of the result that is part of Proposition 6.8,
that for a p-group in characteristic p the augmentation ideal is nilpotent.
8. Show that if elements g1 , . . . , gn generate G as a group, then (g1 1), . . . , (gn 1)
generate the augmentation ideal IG as a left ideal of kG.
[Use the formula (gh 1) = g(h 1) + (g 1).]
9. Suppose that k is a field of characteristic p and G is a p-group. Prove that each
element (g 1) is nilpotent. (More generally, every element of IG is nilpotent.)
Peter Webb
93
10. Show that if N is a normal subgroup of G then the left ideal

RG IN = {x y x RG, y IN }
of RG generated by IN is the kernel of the ring homomorphism RG R[G/N ] and is in

fact a 2-sided ideal in RG.
[One approach to this uses the formula g(n 1) = ( gn 1)g.]
Show that (RG IN )r = RG (IN )r for all r.
11. Show that if a particular element (g 1) appears n times in a product
(g1 1) (gr 1)
then
(g1 1) (gr 1) (g 1)n x
modulo kG (IG )
for some x kG, where G denotes the commutator subgroup.

[Use the formula (g 1)(h 1) = (h 1)(g 1) + (ghg 1 h1 1)hg.]
Show that if G is a p-group and k a field of characteristic p then IGr kG IG for some
power r.
12. Prove that if G is a p-group and k is a field of characteristic p then (IG)r = 0 for
some power r.
13. Let k be a field of characteristic p. Show that the Loewy length of kCpn , the group
algebra of the direct product of n copies of a cycle of order p, is n(p 1) + 1.
14. The dihedral group of order 2n has a presentation

D2n = hx, y x2 = y 2 = (xy)n = 1i.
Let k be a field of characteristic 2. Show that when n is a power of 2, each power (ID2n )r
of the augmentation ideal is spanned modulo (ID2n )r+1 by the two products (x 1)(y
1)(x 1)(y 1) and (y 1)(x 1)(y 1)(x 1) of length r. Hence calculate the
Loewy length of kD2n and show that Rad(kD2n )/ Soc(kD2n ) is the direct sum of two
kD2n -modules that are uniserial.
15. When n 3, the generalized quaternion group of order 2n has a presentation
n1
n2
= 1, y 2 = x2 , yxy 1 = x1 i.
Q2n = hx, y x2
Let k be a field of characteristic 2. Show that when r 1 each power (IQ2n )r of the
augmentation ideal is spanned modulo (IQ2n )r+1 by (x 1)r and (x 1)r1 (y 1). Hence
calculate the Loewy length of kQ2n .
16. Let H be a subgroup of G and let IH be the augmentation ideal of RH, which
we may regard as a subset of RG. Show that RG IH
= IH G
H as RG-modules, and
G
that RG/(RG IH) = R H as RG-modules. Show also that RG/(RG IH) is the largest
quotient of RG on which H acts trivially when acting from the right.
94
17. (a) Let G be any group and IG ZG the augmentation ideal over Z. Prove that
IG/(IG)2
= G/G as abelian groups.
[Consider the homomorphism of abelian groups IG G/G given by g 1 7 gG . Use
the formula ab 1 = (a 1) + (b 1) + (a 1)(b 1) to show that (IG)2 is contained in
the kernel, and that the homomorphism G/G IG/(IG)2 given by gG 7 g 1 + (IG)2
is well defined.]
(b) For any group G write d(G) for the smallest size of a set of generators of G as a
group, and if U is a ZG-module write d(U ) for the smallest size of a set of generators of
U as a ZG-module. Use Exercise 8 to show that d(G/G ) d(IG) d(G) with equality
when G is a p-group. [For the final equality use properties of the Frattini subgroup of G.]
(c) If now R is any commutative ring with 1 and IG RG is the augmentation ideal
of G over R, show that IG/(IG)2
= R Z G/G as R-modules. When G is a p-group and
R is a field of characteristic p, show again that d(G) = d(IG).
18. Let be a transitive G-set and k a field. Let k be the corresponding permutation
P
module. There is a homomorphism of kG-modules : k k defined as ( a ) =
P
P
a . Let =
k.
(a) Show that every kG-module homomorphism k k is a scalar multiple of .
G
(b) Show that the fixed points of G on k are k
= k .

(c) Show that () = 0 if and only if char k ||, and that if this happens then
Rad k and the trivial module k occurs as a composition factor of k with multiplicity
2.
(d) Show that if () 6= 0 then is a split epimorphism and 6 Rad k.
(e) Show that kG is semisimple if and only if the regular representation kG has the
trivial module k as a direct summand (i.e. k is a projective module).
19. Let be a transitive G-set for a possibly infinite group G and let R be the
corresponding permutation module. Show that is infinite if and only if (R)G = 0 and
deduce that G is infinite if and only if (RG)G = 0.
20. Let Ur be the indecomposable kCp -module of dimension r, 1 r p, where k is
a field of characteristic p. Prove that Ur
= S r1 (U2 ), the (r 1) symmetric power.
[One way to proceed is to show that if Cp = hgi then (g 1)r1 does not act as zero on
S r1 (U2 ) and use the classification of indecomposable kCp -modules.]
21. Let G = SL(2, p), the group of 2 2 matrices over Fp that have determinant 1,
where p is a prime. The subgroups
P1 =

1
0

Fp ,
P2 =

0
1

Fp
have order p. Let U2 be the natural 2-dimensional module on which G acts. When
0 r p 1 prove that S r (U2 ) is a uniserial Fp P1 -module, and also a uniserial Fp P2 module, but that the only subspaces of S r (U2 ) that are invariant under both P1 and P2
are 0 and S r (U2 ). Deduce that S r (U2 ) is a simple Fp G-module.
Peter Webb
95

1 0
Show further when p is odd that the matrix
acts as the identity on S r (U2 )
0 1
if and only if r is even, and hence that we have constructed (p+1)/2 simple representations
of

1 0
P SL(2, p) := SL(2, p)/
.
0 1
[Background to the question that is not needed to solve it: |G| = p(p2 1); both P1 and P2
are Sylow p-subgroups of G. In fact, the simple modules constructed here form a complete
list of the simple Fp SL(2, p)-modules.]
22. Let g be an endomorphism of a finite-dimensional vector space V over a field k
of characteristic p, and suppose that g has finite order pd for some d. Show that as a
khgi-module, V has an indecomposable direct summand of dimension at least pd1 + 1.
[You may assume the classification of indecomposable modules for cyclic p-groups in characteristic p.]
Deduce that if such an endomorphism g fixes pointwise a subspace of V of codimension 1 then g has order p or 1.
[An endomorphism (not necessarily of prime-power order) that fixes a subspace of codimension 1 is sometimes referred to as a reflection in a generalized sense.]
23. Let k be a field of characteristic p and suppose N is a normal subgroup of G that
is a Sylow p-subgroup of G. Show that Rad kG = kG Rad kN .
[Use questions 3 and 10, and show that k[G/N ] is the largest semisimple quotient of kG.]
24. Let A be a finite dimensional algebra over a field and let U be an arbitrary Amodule. As in the text, we define Rad U to be the intersection of the maximal submodules
of U .
(a) Use exercises 11 and 12 from Chapter 1 to show that Proposition 6.9 holds without
the hypothesis of finite generation. That is, show that Rad U = Rad A U , and that Rad U
is the smallest submodule of U with semisimple quotient. Show also that Soc U , defined as
the sum of the simple submodules of U , is the largest semisimple submodule of U , and it is
the set of elements of U annihilated by Rad A. [It is a question of copying the arguments
from 6.9. Note that the property of A being used is that A/ Rad A is semisimple.]
(b) Show that if U 6= 0 then U has a non-zero finite-dimensional homomorphic image.
[Use the fact that Rad A is nilpotent.]
(c) Show that each proper submodule of U is contained in a maximal submodule.
25. Let A be a finite dimensional algebra over a field and let U be an A-module. Write
(U ) for the Loewy length of U .
(a) Suppose V is a submodule of R. Show that (V ) (U ) and (U/V ) (U ).
Show by example that we can have equality here even when 0 < V < U .
(b) Suppose that U1 , . . . , Un are submodules of U for which U = U1 + + Un . Show
that (U ) = max{(Ui) 1 i n}.
26. (This exercise extends the theorem of Burnside presented in Chapter 2 Exercise
8 to non-semisimple algebras.) Let A be a finite dimensional algebra over a field k, let
96
V be an A-module so that the action of A on V is given by an algebra homomorphism

: A Endk (V ) and let I = Ker .
(a) Show that if V is simple then A/I is a semisimple ring with only one simple module
(up to isomorphism).
(b) Assuming that k is algebraically closed, show that is surjective if and only if V
is simple.
(c) If A = kG is a group algebra, k is algebraically closed and dim V = n, show that
V is simple if and only if there exist n2 elements g1 , . . . , gn2 of G so that (g1 ), . . . , (gn2 )
are linearly independent.
27. Let k be a field of characteristic p and let G = Cpn be a cyclic group of order
n
p . Suppose that H G is the subgroup of order pt , for some t n, and for each r with
1 r pt write Ur,H for the indecomposable kH-module of dimension r.
(a) Show that Ur,H G

H = U|G:H|r,G as kG-modules, so that indecomposable modules
induce to indecomposable modules. [Exploit the fact that if V is a cyclic module then so
is V G
H , by Chapter 4 Exercise 17.]
(b) Write r = apnt + b where 0 b < pnt . Show that
nt
b
p
Ur,G G
H = (Ua+1,H ) (Ua,H )
nt
b
nt
nt
to show
as kH-modules. [If G = hxi, H = hxp i use the fact that xp 1 = (x 1)p
pnt
1 that is non-zero on Ur,G is the ath power, and use this
that the largest power of x
to identify the summands in a decomposition of Ur,G G
H .]
Peter Webb
97
7. Projective modules for finite-dimensional algebras

In previous sections we have seen the start of techniques to describe modules that are
not semisimple. The most basic decomposition of such a module is one that expresses it
as a direct sum of modules that cannot be decomposed as a direct sum any further. These
summands are called indecomposable modules. We have also examined the notions of radical series and socle series of a module, which are series of canonically defined submodules
that may shed light on submodule structure. We combine these two notions in the study of
projective modules for group rings, working at first in the generality of modules for finite
dimensional algebras over a field. In this situation the indecomposable projective modules
are the indecomposable summands of the regular representation. We will see that they are
identified by the structure of their radical quotient. The projective modules are important
because their structure is part of the structure of the regular representation. Since every
module is a homomorphic image of a direct sum of copies of the regular representation,
by knowing the structure of the projectives we gain some insight into the structure of all
modules.
Characterizations of projective and injective modules

Recall that a module M over a ring A is said to be free if it has a basis; that is, a
subset {xi i I} that spans M as an A-module, and is linearly independent over A. To
L
say that {xi i I} is a basis is equivalent to requiring M = iI Axi with A
= Axi
via an isomorphism a 7 axi for all i. Thus M is a finitely generated free module if and
only if M
= An for some n. These conditions are also equivalent to the condition in the
following proposition:
(7.1) PROPOSITION. Let A be a ring and M an A-module. The following are

equivalent
for a subset {xi i I} of M :
(a) {xi i I} is a basis of M ,

(b) for every module N and mapping of sets : {xi i I} N there exists a unique
module homomorphism : M N that extends .

Proof. The proof is standard, and we merely remind the reader. If {xi i I} is
P
P
a basis, then given we may define ( iI ai xi ) = iI ai (xi ) and this is evidently
the unique module homomorphism extending
. Conversely if condition (b) holds we may

construct the free module F with {xi i I} as a basis and use the condition to construct
a homomorphism from M F that is the identity on {xi i I}. The fact just shown
that the free module also satisfies condition (b) allows us to construct a homomorphism
F M that is again the identity on {xi i I}, and the two homomorphisms have
composites in both directions
that are the identity, since these are the unique extensions
of the identity map on {xi i I}.
98
We define a module homomorphism f : M N to be a split epimorphism if and only

if there exists a homomorphism g : N M so that f g = 1N , the identity map on N . Note
that a split epimorphism is necessarily an epimorphism since if x N then x = f (g(x)) so
that x lies in the image of f . We define similarly f to be a split monomorphism if there
exists a homomorphism g : N M so that gf = 1M . Necessarily a split monomorphism
is a monomorphism. We are about to show that if f is a split epimorphism then N is
(isomorphic to) a direct summand of M . To combine both this and the corresponding
result for split monomorphisms it is convenient to introduce short exact sequences. We
say that a diagram of modules and module homomorphisms LM N is exact at M
if Im = Ker . A short exact sequence of modules is a diagram 0 LM N 0

that is exact at each of L, M and N . Exactness at L and N means simply that is a
monomorphism and is an epimorphism.
(7.2) PROPOSITION. Let 0 LM N 0 be a short exact sequence of

modules over a ring. The following are equivalent:
(1) is a split monomorphism,
(2) is a split epimorphism,
(3) there is a commutative diagram
0 L
k
0 L
LN
where 1 and 2 are inclusion into the first summand and projection onto the second
summand,
(4) for every module U the sequence
0 HomA (U, L) HomA (U, M ) HomA (U, N ) 0
is exact,
(4 ) for every module U the sequence
0 HomA (N, U ) HomA (M, U ) HomA (L, U ) 0

is exact.
In any diagram such as the one in (3) the morphism is necessarily an isomorphism. Thus
if any of the listed conditions is satisfied it follows that M
= L N.
Proof. Condition (3) implies the first two, since the existence of such a commutative
diagram implies that is split by 1 and is split by 1 2 , and it also implies the last
Peter Webb
99
two conditions because the commutative diagram produces similar commutative diagrams
after applying HomA (U, ) and HomA (, U ).
Conversely if condition (1) is satisfied, so that = 1L for some homomorphism
: M L, we obtain a commutative diagram as in (3) on taking the components of
to be and . If condition (2) is satisfied we obtain a commutative diagram similar to
the one in (3) but with a homomorphism : L N M in the wrong direction, whose
components are and a splitting of . We obtain the diagram of (3) on showing that in
any such diagram the middle vertical homomorphism must be invertible.
The fact that the middle homomorphism in the diagram must be invertible is a consequence of both the five lemma and the snake lemma in homological algebra. We leave
it here as an exercise.
Finally if (4) holds then on taking U to be N we deduce that the identity map on N
is the image of a homomorphism : U M , so that 1N = and is split epi, so that
(2) holds. Equally if (4 ) holds then taking U to be L we see that the identity map on L
is the image of a homomorphism : M U , so that 1L = and (1) holds.
In the event that and are split, we say that the short exact sequence in Proposition
7.2 is split. Notice that whenever : M N is an epimorphism it is part of the short exact
sequence 0 Ker M N 0, and so we deduce that if is a split epimorphism then

N is a direct summand of M . A similar comment evidently applies to split monomorphisms.
(7.3) PROPOSITION. The following are equivalent for an A-module P .
(1) P is a direct summand of a free module.
(2) Every epimorphism V P is split.
(3) For every pair of morphisms
P
y
V
where is an epimorphism, there exists a morphism : P V with = .

(4) For every short exact sequence of A-modules 0 V W X 0 the corresponding
sequence
0 HomA (P, V ) HomA (P, W ) HomA (P, X) 0
is exact.
Proof. This result is standard and we do not prove it here. In condition (4) the
sequence of homomorphism groups is always exact at the left-hand terms HomA (P, V ) and
HomA (P, W ) without requiring any special property of P (we say that HomA (P, ) is left
exact). The force of condition (4) is that the sequence should be exact at the right-hand
term.
100
We say that a module P satisfying any of the four conditions of 7.3 is projective.
Notice that direct sums and also direct summands of projective modules are projective.
An indecomposable module that is projective is an indecomposable projective module,
and these modules will be very important in our study. In other texts the indecomposable
projective modules are also known as PIMs, or Principal Indecomposable Modules, but we
will not use this terminology here.
We should also mention injective modules, which enjoy properties similar to those of
projective modules, but in a dual form. We say that a module I is injective if and only if
whenever there are morphisms
I
x
with a monomorphism, then there exists a morphism : V I so that = . Dually

to Proposition 7.3, it is equivalent to require that every monomorphism I V is split; and
also that HomA ( , I) sends exact sequences to exact sequences. When A is an arbitrary
ring we do not have such a nice characterization of injectives analogous to the property
that projective modules are direct summands of free modules. However, for group algebras
over a field we will show in Corollary 8.13 that injective modules are the same thing as
projective modules, so that in this context they are indeed summands of free modules.
Projectives by means of idempotents

One way to obtain projective A-modules is from idempotents of the ring A. If e2 =
e A then A A = Ae A(1 e) as A-modules, and so the submodules Ae and A(1 e) are
projective. We formalize this with the next result, which should be compared with 3.22 in
which we were dealing with ring summands of A and central idempotents.
(7.4) PROPOSITION. Let A be a ring. The decompositions of the regular representation as a direct sum of submodules
AA
= A1 Ar
biject with expressions 1 = e1 + + er for the identity of A as a sum of orthogonal

idempotents, in such a way that Ai = Aei . The summand Ai is indecomposable if and
only if the idempotent ei is primitive.
Proof. Suppose that 1 = e1 + + er is an expression for the identity as a sum of
orthogonal idempotents. Then
AA
= Ae1 Aer ,
Peter Webb
101
for the Aei are evidently submodules of A, and their sum is A since if x A then x =
P
xe1 + + xer . The sum is direct since if x Aei j6=i Aej then x = xei and also
P
P
x = j6=i aj ej so x = xei = j6=i aj ej ei = 0.
Conversely, suppose that A A = A1 Ar is a direct sum of submodules. We
may write 1 = e1 + + er where ei Ai is a uniquely determined element. Now
ei = ei 1 = ei e1 + + ei er is an expression in which ei ej Aj , and since the only such
expression is ei itself we deduce that
n
ei if i = j,
ei ej =
0 otherwise.
The two constructions just described, in which we associate an expression for 1 as a

sum of idempotents to a module direct sum decomposition and vice-versa, are mutually
inverse, giving a bijection as claimed.
If a summand Ai decomposes as the direct sum of two other summands, this gives
rise to an expression for ei as a sum of two orthogonal idempotents, and conversely. Thus
Ai is indecomposable if and only if ei is primitive.
In Proposition 3.22 it was proved that in a decomposition of A as a direct sum of indecomposable rings, the rings are uniquely determined as subsets of A and the corresponding primitive central idempotents are also unique. We point out that the corresponding
uniqueness property need not hold with module decompositions of A A that are not ring
decompositions. For an example of this we take A = M2 (R), the ring of 2 2-matrices
over a ring R, and consider the two decompositions

1 0
0 0
0 1
1 1
A
=A
A
.
AA = A
0 0
0 1
0 1
0 0
The submodules here are all different. We will see later that if A is a finite-dimensional
algebra over a field then in any two decompositions of A A as a direct sum of indecomposable
submodules, the submodules are isomorphic in pairs.
We will also see that when A is a finite-dimensional algebra over a field, every indecomposable projective A-module may be realized as Ae for some primitive idempotent e.
For other rings this need not be true: an example is ZG, for which it is the case that the
only idempotents are 0 and 1 (see the exercises to Chapter 8). For certain finite groups (an
example is the cyclic group of order 23, but this takes us beyond the scope of this book)
there exist indecomposable projective ZG-modules that are not free, so such modules will
never have the form ZGe for any idempotent element e.
(7.5) Example. We present an example of a decomposition of the regular representation in a situation that is not semisimple. Many of the observations we will make are
consequences of theory to be presented in later sections, but it seems worthwhile to show
that the calculations can be done by direct arguments.
Consider the group ring F4 S3 where F4 is the field of 4 elements. The choice of F4
is made because at one point it will be useful to have all cube roots of unity available,
102
but in fact many of the observations we are about to make also hold over the field F2 . By
Proposition 4.5 the 1-dimensional representations of S3 are the simple representations of
S3 /S3
= C2 , lifted to S3 . But F4 C2 has only one simple module, namely the trivial module,
by Proposition 6.3, so this is the only 1-dimensional F4 S3 -module. The 2-dimensional
representation of S3 constructed in Chapter 1 over any coefficient ring is now seen to be
simple here, since otherwise it would have a trivial submodule; but the eigenvalues of the
element (1, 2, 3) on this module are and 2 , where F4 is a primitive cube root of 1,
so there is no trivial submodule.
Let K = h(1, 2, 3)i be the subgroup of S3 of order 3. Now F4 K is semisimple with
three 1-dimensional representations on which (1, 2, 3) acts as 1, and 2 , respectively. In
fact
F4 K = F4 Ke1 F4 Ke2 F4 Ke3
where
e1 = () + (1, 2, 3) + (1, 3, 2)
e2 = () + (1, 2, 3) + 2 (1, 3, 2)
e3 = () + 2 (1, 2, 3) + (1, 3, 2)
are orthogonal idempotents in F4 K. We may see that these are orthogonal idempotents
by direct calculation, but it can also be seen by observing that the corresponding elements
2i
of CK with replaced by e 3 are orthogonal and square to 3 times themselves (Theorem
2i
2i
3.23), and lie in Z[e 3 ]K. Reduction modulo 2 gives a ring homomorphism Z[e 3 ] F4
that maps these elements to e1 , e2 and e3 , while retaining their properties. Thus
F4 S3 = F4 S3 e1 F4 S3 e2 F4 S3 e3
and we have constructed modules F4 S3 ei that are projective. We have not yet shown that
they are indecomposable.
We easily compute that
(1, 2, 3)e1 = e1 ,
(1, 2, 3)e2 = 2 e2 ,
(1, 2, 3)e3 = e3
and from this we see that K F4 ei = F4 ei for all i. Since S3 = K (1, 2)K we have
F4 S3 ei = F4 ei F4 (1, 2)ei , which has dimension 2 for all i. We have already seen that
when i = 2 or 3, ei is an eigenvector for (1, 2, 3) with eigenvalue or 2 , and a similar
calculation shows that the same is true for (1, 2)ei . Thus when i = 2 or 3, F4 S3 ei has
no trivial submodule and hence is simple by the observations made at the start of this
example. We have an isomorphism of F4 S3 -modules
F4 S3 e2 F4 S3 e3
e2 7 (1, 2)e3
(1, 2)e2 7 e3 .
Peter Webb
103
P
On the other hand F4 S3 e1 has fixed points F4 gS3 g of dimension 1 and so has two
composition factors, which are trivial. On restriction to F4 h(1, 2)i it is the regular representation, and it is a uniserial module.
We see from all this that F4 S3 = 11 2 2, in a diagrammatic notation. Thus the
2-dimensional simple F4 S3 -module is projective, and the trivial module appears as the
unique simple quotient of a projective module of dimension 2 whose socle is also the trivial
module. These summands of F4 S3 are indecomposable, and so e1 , e2 and e3 are primitive
P
idempotents in F4 S3 . We see also that the radical of F4 S3 is the span of gS3 g.
Projective covers, Nakayamas lemma and lifting of idempotents
We now develop the theory of projective covers. We first make the definition that an
essential epimorphism is an epimorphism of modules f : U V with the property that
no proper submodule of U is mapped surjectively onto V by f . An equivalent formulation
is that whenever g : W U is a map such that f g is an epimorphism, then g is an
epimorphism. One immediately asks for examples of essential epimorphisms, but it is
probably more instructive to consider epimorphisms that are not essential. If U V
is any epimorphism and X is a non-zero module then the epimorphism U X V
constructed as the given map on U and zero on X can never be essential. This is because
U is a submodule of U X mapped surjectively onto V . Thus if U V is essential then
U can have no direct summands that are mapped to zero. One may think of an essential
epimorphism as being minimal, in that no unnecessary parts of U are present.
The greatest source of essential epimorphisms is Nakayamas lemma, given here in
a version for modules over non-commutative rings. Over an arbitrary ring a finiteness
condition is required, and that is how we state the result here. We will see in the exercises
that when the ring is a finite-dimensional algebra over a field, the result is true for arbitrary
modules without any finiteness condition.
(7.6) THEOREM (Nakayamas Lemma). If U is any Noetherian module, the homomorphism U U/ Rad U is essential. Equivalently, if V is a submodule of U with the
property that V + Rad U = U , then V = U .
Proof. Suppose V is a submodule of U . If V 6= U then V M U where
M is a maximal submodule of U . Now V + Rad U M and so the composite V
U U/ Rad U has image contained in M/ Rad U , which is not equal to U/ Rad U since
(U/ Rad U )/(M/ Rad U )
= U/M 6= 0.
104
When U is a module for a finite-dimensional algebra it is always true that every

proper submodule of U is contained in a maximal submodule, even when U is not finitely
generated. This was the only point in the proof of 7.6 where the Noetherian hypothesis
was used, and so in this situation U U/ Rad U is always essential. The details are left
to the exercises.
The next result is not at all difficult and could usefully be proved as an exercise.
(7.7) PROPOSITION.
(a) Suppose that f : U V and g : V W are two module homomorphsms. If two of
f , g and gf are essential epimorphisms then so is the third.
(b) Let f : U V be a homomorphism of Noetherian modules. Then f is an essential epimorphism if and only if the homomorphism of radical quotients U/ Rad U V / Rad V
is an isomorphism.
(c) Let fi : Ui Vi be homomorphisms of Noetherian modules, where i = 1, . . . , n. The
fi are all essential epimorphisms if and only if
M
M
fi :
Ui
Vi
i
is an essential epimorphism.
Proof. (a) Suppose f and g are essential epimorphisms. Then gf is an epimorphism
also, and it is essential because if U0 is a proper submodule of U then f (U0 ) is a proper
submodule of V since f is essential, and hence g(f (U0)) is a proper submodule of S since
g is essential.
Next suppose f and gf are essential epimorphisms. Since W = Im(gf ) Im(g) it
follows that g is an epimorphism. If V0 is a proper submodule of V then f 1 (V0 ) is a
proper submodule of U since f is an epimorphism, and now g(V0 ) = gf (f 1 (V0 )) is a
proper submodule of S since gf is essential.
Suppose that g and gf are essential epimorphisms. If f were not an epimorphism
then f (U ) would be a proper submodule of V , so gf (U ) would be a proper submodule of
W since gf is essential. Since gf (U ) = W we conclude that f is an epimorphism. If U0 is
a proper submodule of U then gf (U0 ) is a proper submodule of W , since gf is essential,
so f (U0 ) is a proper submodule of V since g is an epimorphism. Hence f is essential.
(b) Consider the commutative square
U
U/ Rad U
V / Rad V
where the vertical homomorphisms are essential epimorphisms by Nakayamas lemma. Now
if either of the horizontal arrows is an essential epimorphism then so is the other, using
Peter Webb
105
part (a). The bottom arrow is an essential epimorphism if and only if it is an isomorphism;
for U/ Rad U is a semisimple module and so the kernel of the map to V / Rad V has a direct
complement in U/ Rad U , which maps onto V / Rad V . Thus if U/ Rad U V / Rad V is
an essential epimorphism its kernel must be zero and hence it must be an isomorphism.
(c) The map
(i Ui )/ Rad(i Ui ) (i Vi )/ Rad(i Vi )
induced by fi may be identified as a map
M
M
(Vi / Rad Vi ),
(Ui / Rad Ui )
i
and it is an isomorphism if and only if each map Ui / Rad Ui Vi / Rad Vi is an isomorphism. These conditions hold if and only if fi is an essential epimorphism, if and only if
each fi is an essential epimorphism by part (b).
We define a projective cover of a module U to be an essential epimorphism P U ,
where P is a projective module. Strictly speaking the projective cover is the homomorphism, but we may also refer to the module P as the projective cover of U . We are justified
in calling it the projective cover by the second part of the following result, which says that
projective covers (if they exist) are unique.
(7.8) PROPOSITION.
(1) Suppose that f : P U is a projective cover of a module U and g : Q U is an
epimorphism where Q is a projective module. Then we may write Q = Q1 Q2 so
that g has components g = (g1 , 0) with respect to this direct sum decomposition and
g1 : Q1 U appears in a commutative triangle
Q1
g1
y
where is an isomorphism.
(2) If any exist, the projective covers of a module U are all isomorphic, by isomorphisms
that commute with the essential epimorphisms.
Proof. (1) In the diagram
g
y
106
we may lift in both directions to obtain maps : P Q and : Q P so that

the two triangles commute. Now f = g = f is an epimorphism, so is also an
epimorphism since f is essential. Thus is an epimorphism. Since P is projective
splits and Q = Q1 Q2 where Q2 = Ker , and maps Q1 isomorphically to P . Thus
g = (f |Q1 , 0) is as claimed with = |Q1 .
(2) Supposing that f : P U and g : Q U are both projective covers, since Q1
is a submodule of Q that maps onto U and f is essential we deduce that Q = Q1 . Now
: Q P is the required isomorphism.
(7.9) COROLLARY. If P and Q are Noetherian projective modules over a ring then
P
= Q if and only if P/ Rad P
= Q/ Rad Q.
Proof. By Nakayamas lemma P and Q are the projective covers of P/ Rad P and
Q/ Rad Q. It is clear that if P and Q are isomorphic then so are P/ Rad P and Q/ Rad Q,
and conversely if these quotients are isomorphic then so are their projective covers, by
uniqueness of projective covers.
If P is a projective module for a finite-dimensional algebra A then Corollary 7.9 says
that P is determined up to isomorphism by its semisimple quotient P/ Rad P . We are
going to see that if P is an indecomposable projective A-module, then its radical quotient
is simple, and also that every simple A-module arises in this way. Furthermore, every
indecomposable projective for a finite dimensional algebra is isomorphic to a summand
of the regular representation (something that is not true in general for projective ZGmodules, for instance). This means that it is isomorphic to a module Af for some primitive
idempotent f A, and the radical quotient P/ Rad P is isomorphic to (A/ Rad A)e where
e is a primitive idempotent of A/ Rad A satisfying e = f + Rad A. We will examine this
kind of relationship between idempotent elements more closely.
In general if I is an ideal of a ring A and f is an idempotent of A then clearly
e = f + I is an idempotent of A/I, and we say that f lifts e. On the other hand, given an
idempotent e of A/I it may or may not be possible to lift it to an idempotent of A. If, for
every idempotent e in A/I, we can always find an idempotent f A such that e = f + I
then we say we can lift idempotents from A/I to A.
We present the next results about lifting idempotents in the context of a ring with a
nilpotent ideal I, but readers familiar with completions will recognize that these results
extend to a situation where A is complete with respect to the I-adic topology on A.
(7.10) THEOREM. Let I be a nilpotent ideal of a ring A and e an idempotent in
A/I. Then there exists an idempotent f A with e = f + I. If e is primitive, so is any
lift f .
Proof. We define idempotents ei A/I i inductively such that ei + I i1 /I i = ei1
for all i, starting with e1 = e. Suppose that ei1 is an idempotent of A/I i1 . Pick any
Peter Webb
107
element a A/I i mapping onto ei1 , so that a2 a I i1 /I i . Since (I i1 )2 I i we have

(a2 a)2 = 0 A/I i . Put ei = 3a2 2a3 . This does map to ei1 A/I i1 and we have
e2i ei = (3a2 2a3 )(3a2 2a3 1)
= (3 2a)(1 + 2a)(a2 a)2
= 0.
This completes the inductive definition, and if I r = 0 we put f = er .

Suppose that e is primitive and that f can be written f = f1 + f2 where f1 and
f2 are orthogonal idempotents. Then e = e1 + e2 , where ei = fi + I, is also a sum of
orthogonal idempotents. Therefore one of these is zero, say, e1 = 0 A/I. This means
that f12 = f1 I. But I is nilpotent, and so contains no non-zero idempotent.
We will very soon see that in the situation of 7.10, if f is primitive, so is e. It depends
on the next result, which is a more elaborate version of 7.10.
(7.11) COROLLARY. Let I be a nilpotent ideal of a ring A and let 1 = e1 + + en
be a sum of orthogonal idempotents in A/I. Then we can write 1 = f1 + + fn in
A, where the fi are orthogonal idempotents such that fi + I = ei for all i. If the ei are
primitive then so are the fi .
Proof. We proceed by induction on n, the induction starting when n = 1. Suppose
that n > 1 and the result holds for smaller values of n. We will write 1 = e1 + E in A/I
where E = e2 + +en is an idempotent orthogonal to e1 . By Theorem 7.10 we may lift e1
to an idempotent f1 A. Write F = 1 f1 , so that F is an idempotent that lifts E. Now
F is the identity element of the ring F AF which has a nilpotent ideal F IF . The composite
homomorphism F AF A A/I has kernel F AF I and this equals F IF , since clearly
F AF I F IF , and if x F AF I then x = F xF F IF , so F AF I F IF . Inclusion
of F AF in A thus induces a monomorphism F AF/F IF A/I, and its image is E(A/I)E.
In E(A/I)E the identity element E is the sum of n 1 orthogonal idempotents, and this
expression is the image of a similar expression for F + F IF in F AF/F IF . By induction,
there is a sum of orthogonal idempotents F = f2 + +fn in F AF that lifts the expression
in F AF/F IF and hence also lifts the expression for E in A/I, so we have idempotents
fi A, i = 1, . . . , n with fi +I = ei . These fi are orthogonal: for f2 , . . . , fn are orthogonal
in F AF by induction, and if i > 1 then F fi = fi so we have f1 fi = f1 F fi = 0.
The final assertion about primitivity is the last part of 7.10.
(7.12) COROLLARY. Let f be an idempotent in a ring A that has a nilpotent ideal
I. Then f is primitive if and only if f + I is primitive.
Proof. We have seen in 7.10 that if f + I is primitive, then so is f . Conversely, if f + I
can be written f + I = e1 + e2 where the ei are orthogonal idempotents of A/I, then by
applying 7.11 to the ring f Af (of which f is the identity) we may write f = g1 + g2 where
the gi are orthogonal idempotents of A that lift the ei .
108
We now classify the indecomposable projective modules over a finite-dimensional algebra as the projective covers of the simple modules. We first describe how these projective
covers arise, and then show that they exhaust the possibilities for indecomposable projective modules. We postpone explicit examples until the next section, in which we consider
group algebras.
(7.13) THEOREM. Let A be a finite-dimensional algebra over a field and S a simple
A-module.
(a) There is an indecomposable projective module PS with PS / Rad PS
= S, of the form
PS = Af where f is a primitive idempotent in A.
(b) The idempotent f has the property that f S 6= 0 and if T is any simple module not
isomorphic to S then f T = 0.
(c) PS is the projective cover of S, it is uniquely determined up to isomorphism by this
property and has S as its unique simple quotient.
(d) It is also possible to find an idempotent fS A so that fS S = S and fS T = 0 for
every simple module T not isomorphic to S.
Proof. Let e A/ Rad A be any primitive idempotent such that eS 6= 0. It is possible
to find such e since we may write 1 as a sum of primitive idempotents and some term in
the sum must be non-zero on S. Let f be any lift of e to A, possible by 7.12. Then f is
primitive, f S = eS 6= 0 and f T = eT = 0 if T
6 S since a primitive idempotent e in the
=
semisimple ring A/ Rad A is non-zero on a unique isomorphism class of simple modules.
We define PS = Af , an indecomposable projective module. Now
PS / Rad PS = Af /(Rad A Af )
= (A/ Rad A) (f + Rad A) = S,
the isomorphism arising because the map Af (A/ Rad A) (f + Rad A) defined by
af 7 (af + Rad A) has kernel (Rad A) f . The fact that PS is the projective cover of
S is a consequence of Nakayamas lemma, and the uniqueness of the projective cover was
dealt with in 7.8. Any simple quotient of PS is a quotient of PS / Rad PS , so there is only
one of these. Finally we observe that if we had written 1 as a sum of primitive central
idempotents in A/ Rad A, the lift of the unique such idempotent that is non-zero on S is
the desired idempotent fS .
(7.14) THEOREM. Let A be a finite-dimensional algebra over a field k. Up to
isomorphism, the indecomposable projective A-modules are exactly the modules PS that
are the projective covers of the simple modules, and PS
= PT if and only if S
= T . Each
projective PS appears as a direct summand of the regular representation, with multiplicity
equal to the multiplicity of S as a summand of A/ Rad A. As a left A-module the regular
representation decomposes as
M
A
(PS )nS
=
simple S
Peter Webb
109
where nS = dimk S if k is algebraically closed, and more generally nS = dimD S where

D = EndA (S).
In what follows we will only prove that finitely generated indecomposable projective
modules are isomorphic to PS , for some simple S. We leave the general case to the exercises.
Proof. Let P be an indecomposable projective module and write
P/ Rad P
= S1 Sn .
Then P S1 Sn is a projective cover. Now
PS1 PSn S1 Sn
is also a projective cover, and by uniqueness of projective covers we have
P
= PS 1 PS n .
Since P is indecomposable we have n = 1 and P
= PS 1 .
Suppose that each simple A module S occurs with multiplicity nS as a summand
L
nS
of the semisimple ring A/ Rad A. Both A and
are the projective cover of
simple S PS
A/ Rad A, and so they are isomorphic. We have seen in Corollary 2.4 that nS = dimk S
when k is algebraically closed, and in exercise 4 to section 2 that nS = dimD S in general.
(7.15) THEOREM. Let A be a finite-dimensional algebra over a field k, and U an
A-module. Then U has a projective cover.
Again, we only give a proof in the case that U is finitely generated, leaving the general
case to the exercises.
Proof. Since U/ Rad U is semisimple we may write U/ Rad U = S1 Sn , where
the Si are simple modules. Let PSi be the projective cover of Si and h : PS1 PSn
U/ Rad U the projective cover of U/ Rad U . By projectivity there exists a homomorphism
f such that the following diagram commutes:
PS 1 PS n
.
yh
U/ Rad U
Since both g and h are essential epimorphisms, so is f by 7.7. Therefore f is a projective

cover.
110
We should really learn more from 7.15 than simply that U has a projective cover: the
projective cover of U is the same as the projective cover of U/ Rad U .
(7.16) Example. The arguments that show the existence of projective covers have a
sense of inevitability about them and we may get the impression that projective covers
always exist in arbitrary situations. In fact they fail to exist in general for integral group
rings. If G = {e, g} is a cyclic group of order 2, consider the submodule 3Z e + Z (e + g)
of ZG generated as an abelian group by 3e and e + g. We rapidly check that this subgroup
is invariant under the action of G (so it is a ZG-submodule), and it is not the whole of ZG
since it does not contain e. Applying the augmentation map : ZG Z we have (3e) = 3
and (e + g) = 2 so (3Z e + Z (e + g)) = 3Z + 2Z = Z. This shows that the epimorphism
is not essential, and so it is not a projective cover of Z. If Z were to have a projective
cover it would be a proper summand of ZG by Proposition 7.8. On reducing modulo 2 we
would deduce that F2 G decomposes, which we know not to be the case by Corollary 6.12.
This shows that Z has no projective cover as a ZG-module.
The Cartan matrix

Now that we have classified the projective modules for a finite-dimensional algebra
we turn to one of their important uses, which is to determine the multiplicity of a simple
module S as a composition factor of an arbitrary module U (with a composition series). If
0 = U0 U1 Un = U
is any composition series of U , the number of quotients Ui /Ui1 isomorphic to S is determined independently of the choice of composition series, by the JordanHolder theorem.
We call this number the (composition factor) multiplicity of S in U .
(7.17) PROPOSITION. Let S be a simple module for
with projective cover PS , and let U be a finite-dimensional
(1) If T is a simple A-module then
n
dim HomA (PS , T ) = dim EndA (S)
0
a finite-dimensional algebra A
A-module.
if S
= T,
otherwise.
(2) The multiplicity of S as a composition factor of U is

dim HomA (PS , U )/ dim EndA (S).
(3) If e A is an idempotent then dim HomA (Ae, U ) = dim eU .
We remind the reader that if the ground field k is algebraically closed then dim EndA (S) =
1 by Schurs lemma. Thus the multiplicity of S in U is just dim HomA (PS , U ) in this case.
Peter Webb
111
Proof. (1) If PS T is any non-zero homomorphism, the kernel must contain Rad PS ,
being a maximal submodule of PS . Since PS / Rad PS
= S is simple, the kernel must be
Rad PS and S = T . Every homomorphism PS S is the composite PS PS / Rad PS

S of the quotient map and either an isomorphism of PS / Rad PS with S or the zero map.
This gives an isomorphism HomA (PS , S)
= EndA (S).
(2) Let
0 = U0 U1 Un = U
be a composition series of U . We prove the result by induction on the composition length

n, the case n = 1 having just been established. Suppose n > 1 and that the multiplicity
of S in Un1 is dim HomA (PS , Un1 )/ dim EndA (S). The exact sequence
0 Un1 U U/Un1 0
gives rise to an exact sequence
0 HomA (PS , Un1 ) HomA (PS , U ) HomA (PS , U/Un1 ) 0
by 7.3, so that
dim HomA (PS , U ) = dim HomA (PS , Un1 ) + dim HomA (PS , U/Un1 ).
Dividing these dimensions by dim EndA (S) gives the result, by part (1).
(3) There is an isomorphism of vector spaces HomA (Ae, U )
= eU specified by 7
(e). Note here that since (e) = (ee) = e(e) we must have (e) eU . This mapping
is injective since each A-module homomorphism : Ae U is determined by its value
on e as (ae) = a(e). It is surjective since the equation just written down does define a
module homomorphism for each choice of (e) eU .
Again in the context of a finite-dimensional algebra A, we define for each pair of simple
A-modules S and T the integer
cST = the composition factor multiplicity of S in PT .
These are called the Cartan invariants of A, and they form a matrix C = (cST ) with rows
and columns indexed by the isomorphism types of simple A-modules, called the Cartan
matrix of A.
(7.18) COROLLARY. Let A be a finite-dimensional algebra over a field, let S and T
be simple A-modules and let eS , eT be idempotents so that PS = AeS and PT = AeT are
projective covers of S and T . Then
cST = dim HomA (PS , PT )/ dim EndA (S) = dim eS AeT / dim EndA (S).
If the ground field k is algebraically closed then cST = dim HomA (PS , PT ) = dim eS AeT .
While it is rather weak information just to know the composition factors of the projective modules, this is at least a start in describing these modules. We will see later on
in the case of group algebras that there is an extremely effective way of computing the
Cartan matrix using the decomposition matrix.
112
1. Direct sum decompositions of A A as an A-module (with indecomposable summands) correspond to expressions for 1A as a sum of orthogonal (primitive) idempotents.
2. U U/ Rad U is essential.
3. Projective covers are unique when they exist. For modules for a finite dimensional
algebra over a field they do exist.
4. Idempotents can be lifted through nilpotent ideals.
5. The indecomposable projective modules for a finite dimensional algebra over a field
are exactly the projective covers of the simple modules. Each has a unique simple quotient
and is a direct summand of the regular representation. Over an algebraically closed field
PS occurs as a summand of the regular representation with multiplicity dim S.

1. Let A be a finite-dimensional algebra over a field. Show that A is semisimple if
and only if all finite-dimensional A-modules are projective.
f
2. Suppose that we have module homomorphisms U V W . Show that part of

Proposition 7.6(a) can be strengthened to say the following: if gf is an essential epimorphism and f is an epimorphism then both f and g are essential epimorphisms.
3. Let U and V be arbitrary modules for a finite-dimensional algebra A. Use the
results of exercise 24 of Chapter 6 to show the following.
(a) Show that the quotient homomorphism U U/ Rad U is essential.
(b) Show that a homomorphism U V is essential if and only if the homomorphism of
radical quotients U/ Rad U V / Rad V is an isomorphism.
(c) Show that U has a projective cover.
(d) Show that every indecomposable projective A-module is finite-dimensional, and hence
isomorphic to PS for some simple module S.
(e) Show that every projective A-module is a direct sum of indecomposable projective
modules.
4. In this question U, V and W are modules for a finite-dimensional algebra over a
field and PW is the projective cover of W . Assume either that these modules are finitedimensional, or the results from the last exercise.
(a) Show that U W is an essential epimorphism if and only if there is a surjective
homomorphism PW U so that the composite PW U W is a projective cover of W .
In this situation show that PW U must be a projective cover of U .
(b) Prove the following extension and converse to Nakayamas lemma: let V be any
submodule of U . Then U U/V is an essential epimorphism V Rad U .
5. Let PS be an indecomposable projective module for a finite-dimensional algebra
over a field. Show that every homomorphic image of PS
Peter Webb
113
(a) has a unique maximal submodule, and

(b) is indecomposable.
6. Let A be a finite-dimensional algebra over a field, and suppose that f, f are
primitive idempotents of A. Show that the indecomposable projective modules Af and
Af are isomorphic if and only if dim f S = dim f S for every simple module S.
7. Let A be a finite-dimensional algebra over a field, and suppose that Q is a projective
A-module. Show that in any expression
Q = PSn11 PSnrr
where S1 , . . . , Sr are non-isomorphic simple modules, we have
ni = dim HomA (Q, Si )/ dim EndA (Si ).
8. Let A be a finite-dimensional algebra over a field. Suppose that V is an A-module,
and that a certain simple A-module S occurs as a composition factor of V with multiplicity
1. Suppose that there exist non-zero homomorphisms S V and V S. Prove that S
is a direct summand of V .
9. Let G = Sn , let k be a field of characteristic 2 and let = {1, 2, . . . , n} permuted
transitively by G.
(a) When n = 3, show that the permutation module k is semisimple, being the direct
sum of the one-dimensional trivial module and the 2-dimensional simple module.
(b) When n = 4 there is a normal subgroup V S4 with S4 /V
= S3 , where V =
h(1, 2)(3, 4), (1, 3)(2, 4)i. The simple kS4 -modules are precisely the two simple kS3 -modules,
made into kS4 -modules via the quotient homomorphism to S3 . Show that k is uniserial
with three composition factors that are the trivial module, the 2-dimensional simple module and the trivial module.
[Use Exercise 18 from Chapter 6.]
10. Show by example that if H is a subgroup of G it need not be true that Rad kH
Rad kG.
[Compare this result with Exercise 3 from Chapter 6.]
11. Let A be a finite-dimensional algebra over a field and f A a primitive idempotent. Show that there is a simple A-module S with f S 6= 0, and that S is uniquely
determined up to isomorphism by this property.
114
8. Projective modules for group algebras

We focus in this section on facts about group algebras that are not true for finite
dimensional algebras in general. The results are a mix of general statements and specific
examples describing the representations of certain types of groups. At the beginning of
the section we summarize the properties of projective modules for p-groups and also the
behaviour of projective modules under induction and restriction. Towards the end we show
that the Cartan matrix is symmetric (then the field is algebraically closed) and also that
projective modules are injective. In the middle we describe quite explicitly the structure of
projective modules for many semidirect products and we do this by elementary arguments.
It shows that the important general theorems are not always necessary to understand
specific representations, and it also increases our stock of examples of groups and their
representations.
Because of the diversity of topics it is possible to skip certain results in this section
without affecting comprehension of what remains. For example, the reader who is more
interested in the general results could skip the description of representations of specific
groups between 8.6 and 8.10.
The behaviour of projective modules under induction, restriction and tensor

product
We start with a basic fact about group algebras of p-groups in characteristic p.
(8.1) THEOREM. Let k be a field of characteristic p and G a p-group. The regular representation is an indecomposable projective module that is the projective cover of
the trivial representation. Every finitely generated projective module is free. The only
idempotents in kG are 0 and 1.
Proof. We have seen in 6.12 that kG is indecomposable and it also follows from
7.14. By Nakayamas lemma kG is the projective cover of k. By 7.13 and 6.3 every
indecomposable projective is isomorphic to kG. Every finitely generated projective is a
direct sum of indecomposable projectives, and so is free. Finally, every idempotent e kG
gives a module decomposition kG = kGe kG(1 e). If e 6= 0 then we must have
kG = kGe, so kG(1 e) = 0 and e = 1.
Peter Webb
115
The next lemma is a consequence of the fact that induction is both the left and the
right adjoint of restriction, as shown in Lemma 4.12. Without using that language we give
a direct proof.
(8.2) LEMMA. Let H be a subgroup of G.
(1) If P is a projective RG-module then P G
H is a projective RH-module.
(2) If Q is a projective RH-module then Q G
H is a projective RG-module.
Proof. (1) As a RH-module,
RG H
=
RHg
= (RH)|G:H| ,
g[H\G]
which is a free module. Hence a direct summand of RGn on restriction to H is a direct

summand of RH |G:H|n , and so is projective.
(2) We have
G
H
G
(RH) G
H = (R 1 ) H = R 1 = RG
so that direct summands of RH n induce to direct summands of RGn .
We now put together 8.1 and 8.2 to obtain an important restriction on projective
modules.
(8.3) COROLLARY. Let k be a field of characteristic p and
let pa be the exact power

of p that divides |G|. If P is a projective kG-module then pa dim P .
Proof. Let H be a Sylow p-subgroup of G and P a projective kG-module. Then P G

H
is projective by 8.2, hence free as a kH-module by 8.1, and of dimension a multiple of
|H|.
We are about to describe in detail the structure of projective modules for some particular groups that are semidirect products, and the next two results will be used in our
proofs. The first is valid over any commutative ring R.
(8.4) PROPOSITION. Suppose that V is any RG-module that is free as an R-module
and P is a projective RG-module. Then V R P is projective as an RG-module.
Proof. If P P
= RGn then V RGn
= V P V P and it suffices to show that
V RGn is free. We offer two proofs of the fact that V RG
= RGrank V . The first is
G
G
that V RG
= V (R G
1 ) = (V R) 1 = V 1 , with the middle isomorphism coming
from Corollary 4.13. As a module for the identity group, V is just a free R-module and so
G rank V
V G
= RGrank V .
1 = (R 1 )
116
The second proof is really the same as the first, but we make the isomorphism explicit.
Let V triv be the same R-module as V , but with the trivial G-action, so V triv
= Rrank V as
RG-modules. We define a linear map
V RG V triv RG
v g 7 g 1 v g
which has inverse gw g w g. One checks that these mutually inverse linear maps
are RG-module homomorphisms. Finally V triv RG
= RGrank V .
In the calculations that follow we will need to use the fact that for representations
over a field, taking the tensor product with a fixed representation preserves exactness.
(8.5) LEMMA. Let 0 U V W 0 be a short exact sequence of kG-modules
and X another kG-module, where k is a field. Then the sequence
0 U k X V k X W k X 0
is exact. Thus if U is a submodule of V then (V /U ) k X
= (V k X)/(U k X).
Proof. Since the tensor products are taken over k, the question of exactness is independent of the action of G. As a short exact sequence of vector spaces 0 U V W 0
is split, so that V
= U W with the morphisms in the sequence as two of the component inclusions and projections. Applying k X to this we get V k X
= (U k X) (W k X)
with component morphisms given by the morphisms in the sequence 0 U k X
V k X W k X 0. This is enough to show that the sequence is (split) exact as a
sequence of vector spaces, and hence exact as a sequence of kG-modules.
Another approach to the same thing is to suppose that U is a submodule of V and
take a basis v1 , . . . , vn for V such that v1 , . . . , vd is a basis for U and let x1 , . . . , xm be a
basis for X. Now the vi k xj with 1 i n and 1 j m form a basis for V k X, and
the same elements with 1 i d and 1 j m form a basis for U k X. This shows
that U k X is a submodule of V k X, and the quotient has as a basis the images of the
vi k xj with d + 1 i n and 1 j m, which is in bijection with a basis of W k X.
Peter Webb
117
Projective modules for semidirect products of a p-group and a p -group

Before continuing with the general development of the theory we describe in detail
the projective modules for groups that are a semidirect product with one of the terms a
Sylow p-subgroup, since this can be done by direct arguments with the tools we already
have at hand. The general theory will continue with Proposition 8.11.
(8.6) EXAMPLE. We first describe the projective kG-modules where k is a field of
characteristic p and G = H K, where H is a p-group and K has order prime to p.
Following this example we will consider projective modules for the semidirect products
H K and K H and the observations we will make when we do this also apply to the
direct product. However, it is useful to consider the simpler example first.
In this example the only tensor products that appear are tensor products k over the
field k and we supress the suffix k from the notation. We make use of the following general
isomorphism (not dependent on the particular hypotheses we have here):
k[H K]
= kH kK
as k-algebras,
which arises because kG has as a basis the elements (h, k) where h H, k K, and
kH kK has as a basis the corresponding elements h k. These two bases multiply
together in the same fashion, and so we have an algebra isomorphism.
Let us write kK = S1n1 Srnr , where S1 , . . . , Sr are the non-isomorphic simple
kK-modules, bearing in mind that kK is semisimple since K has order relatively prime to
p. Since H = Op (G), these are also the non-isomorphic simple kG-modules, by Corollary
6.4. We have
kG = kH kK = (kH S1 )n1 (kH Sr )nr
as kG-modules, and so the kH Si are projective kG-modules. Each does occur with
multiplicity equal to the multiplicity of Si as a summand of kG/ Rad(kG), and so must
be indecomposable, using 7.14. We have therefore constructed all the indecomposable
projective kG-modules, and they are the modules PSi = kH Si .
Suppose that 0 P1 Pn = kH is a composition series of the regular representation of H. Since H is a p-group, all the composition factors are the trivial representation,
k. Because k Si preserves exact sequences, the series 0 P1 Si Pn Si = PSi
has quotients k Si = Si , which are simple, and so this is a composition series of PSi .
There is only one isomorphism type of composition factor.
We can also see this from the ring-theoretic structure of kG. Assuming that k is
algebraically closed (to make the notation easier) we have EndkG (Si ) = k for each i and
Lr
kK
= i=1 Mni (k) as rings where ni = dim Si . Now
kG = kH kK
=
r
M
i=1
Mni (kH)
118
as rings, since kH Mn (k)

= Mn (kH). The latter can be proved by observing that the
two sides of the isomorphism have bases that multiply together in the same way. The
projective kG-modules PSi = kH Si are now identified in this matrix description as
column vectors of length ni with entries in kH.
We next observe that
Rad kG = Rad(kH) kK
=
r
M
Mni (Rad(kH)).
i=1
The first equality holds because the quotient

(kH kK)/(Rad(kH) kK)
= (kH/ Rad kH) kK
= k kK
is semisimple, since kK is semisimple, so that
Rad(kH kK) Rad(kH) kK.
On the other hand Rad(kH)kK is a nilpotent ideal of kHkK, so is contained in the radiLr
cal, and we have equality. To look at it from the matrix point of view, i=1 Mni (Rad(kH))
Lr
is a nilpotent ideal with quotient i=1 Mni (k), which is semisimple, so again we have correctly identified the radical. Computing the powers of these ideals we have
Rad kG = Rad (kH) kK
=
n
r
M
Mni (Radn (kH))
i=1
for each n.
As a very specific example, suppose that H is cyclic of order ps . Then kH has a
unique composition series by Theorem 6.2, with terms Pj = Rad(kH)j kH. We see from
the above discussion and the fact that Rad kG = Rad(kH) kK that the terms in the
composition series of PSi are Pj Si = Rad(kG)j PSi . Thus the radical series of PSi is in
fact a composition series, and it follows (as in Exercise 5 of Chapter 6) that PSi also has
a unique composition series, there being no more submodules of PSi other than the ones
listed.
We move on now to describe the projective kG-modules where k is a field of characteristic p and where G is a semidirect product with Sylow p-subgroup H, doing this first
when G has the form G = H K and afterwards doing it when G = K H. Before this
we give a module decomposition of the group ring of a semidirect product that holds in
all cases.
Peter Webb
119
(8.7) LEMMA. Let R be a commutative ring and G = H K a semidirect product.

There is an action of RG on RH such that H acts on RH by left multiplication and K
acts by conjugation. With this action, RH

= R G
K = RG K as RG-modules, where
P
K = xK x. We have a tensor decomposition of RG-modules
G
RG
= (R G
K ) R (R H )
G
where the structure of R G
K has just been described, and R H = RK with H acting
trivially.
Proof. We may take R G

K = RG RK R to have as R-basis the tensors h 1 where
h H. The action of an element x H on such a basis element is x(h 1) = xh 1, and
the action of an element k K is given by k(h 1) = kh 1 = kh k 1 1 = khk 1 1
since the tensor product is taken over RK. Note here that although kh 1 is defined,
it is not one of our chosen basis elements (unless k = 1), whereas khk 1 1 is one of
them. We see from this that the R-linear isomorphism RH R G
K given by h 7 h 1 is
in fact an isomorphism of RG-modules with the specified action on RH, and indeed this
specification does give an action.
The identification of R G
K with RG K was seen in 4.9 and in Exercise 16 from
section 4, but we give the argument here. The submodule RG K is spanned in RG by
the elements G K = HK K = H K. However the elements hK are independent in RG
as h ranges over H, since they have disjoint supports, so these elements form a basis of
RG K. The basis elements are permuted transitively by G and the stabilizer of K is K,
so the RG K
= R G
K.
Since RH R RK has an R-basis consisting of the basic tensors h k here h H
k K and every element of G is uniquely written in the form hk with h H, k K, the
mapping RH R RK RG specified by h k 7 hk is an R-linear isomorphism. This
map commutes with the action of G, since if uv G with u H, v K then
uv(h k) = (uv) h (uv) k = u(v h) vk
since v acts by conjugation of H and u acts trivially on K. This is mapped to u(v h)vk =
(uv)(hk) RG so the map commutes with the action of G.
We now suppose that k is a field of characteristic p and G = H K where H is a
Sylow p-subgroup. We shall see that several of the properties of projective modules that we
identified in the case of a direct product G = H K still hold for the semidirect product,
but not all of them.
120
(8.8) PROPOSITION. Let k be a field of characteristic p and let G = H K where

H is a p-group and K has order prime to p.
P
1
(a) Let eK = |K|
xK x. Then the indecomposable projective Pk has the form
Pk
= kGeK
= kH
= k G
K
(b)
(c)
(d)
(e)
where kH is taken to have the kG-module action where H acts by multiplication and
K acts by conjugation, as in Lemma 8.7.
Rad(kG) is the kernel of the ring homomorphism kG kK given by the quotient
homomorphism G K, so that kG/ Rad(kG)
= kK as rings, and also as kG-modules.
Furthermore, Rad(kG) is generated both as a left ideal and a right ideal by Rad(kH) =
IH, the augmentation ideal of kH.
The simple kG-modules are precisely the simple kK-modules, regarded as kG-modules
via the quotient homomorphism G K. If S is any simple kG-module with projective
cover PS then PS
= Pk S
= S G
K as kG-modules.
For each simple kG-module S with projective cover PS , the radical series of PS has
terms Radn (PS ) = (IH)n S with radical layers ((IH)n1 /(IH)n ) S, where IH is
taken to have the restriction of the action on RH given in Lemma 8.7. Consequently
all of the indecomposable projective modules have the same Loewy length.
There is an isomorphism of kG-modules Pk kK
= kG.
Proof. (a) Since |K| is invertible in k we have kGeK = kG K (with the notation
of Lemma 8.7), and since eK is an idempotent this is a projective kG-module, and it has
the stated identifications by Lemma 8.7. This module is indecomposable as a kH-module
since H is a p-group, by Theorem 8.1, so it is indecomposable as a kG-module. Since its
unique simple quotient is represented by 1 H, and K conjugates this trivially, kGeK is
the projective cover of the trivial module.
(b) Since H = Op (G) the simple kG-modules are precisely the simple kK-modules,
by 6.4. Now G acts on these via the ring surjection kG kK, so the kernel of this map
acts as zero on all simple modules and hence is contained in Rad kG. But also kK is a
semisimple ring, so the kernel equals Rad kG.
The fact that the kernel of the ring homomorphism kG kK (and hence Rad kG)
has the description kG IH is shown in Exercise 10 of Chapter 6, and we also give the
argument here. We know that Rad kH = IH from 6.8. Taking coset representatives
F
G = g[G/H] gH and writing k[gH] for the span in kG of the elements gh, h H, we
have that
X
X
M
kG IH =
k[gH] IH =
gIH =
gIH,
g[G/H]
g[G/H]
g[G/H]

the sum being direct since each term gIH has basis {g(h 1) h H} with support in the
coset gH. The span of the elements of K inside kG is a space complementary to kG IH
Peter Webb
121
that is mapped isomorphically to kK. From this we see that kG IH is the kernel of the
homomorphism kG kK. From the fact that H is a normal subgroup we may calculate
directly that kG Rad(kH) = kG Rad(kH) kG = Rad(kH) kG. We may also show this
by observing that kG Rad(kH) is the kernel of a ring homomorphism, hence is a 2-sided
ideal and so kG Rad(kH) = kG Rad(kH) kG. We could have argued with Rad(kH) kG
just as well, and so this also is equal to kG Rad(kH) kG.
(c) The fact that the simple kG-modules are the same as the simple kK-modules was
observed at the start of the proof of (b). If S is a simple module, by 8.4 Pk S is projective.
Tensoring the epimorphism Pk k with S gives an epimorphism Pk S S, so Pk S
contains PS as a summand. We show that Pk S equals PS . Now
Rad(Pk S) = Rad(kG) (Pk S) = Rad(kH) kG (Pk S) = IH (Pk S).
We show that this equals (IH Pk ) S. The reason for this is that IH is spanned by
elements h 1 where h H, and if x s is a basic tensor in Pk S then
(h 1) x s = h (x s) 1 (x s)
= hx hs x s
= hx (h 1)s + (h 1)x s
= (h 1)x s
since H is a normal p-subgroup which thus acts trivially on S. Note that this argument
does not depend on the first module in the tensor product being Pk . To continue the
argument, we have
(IH Pk ) S = (IH kG Pk ) S = (Rad(kG) Pk ) S = Rad(Pk ) S.
Now Rad(Pk ) has codimension 1 in Pk , so Rad(Pk ) S has codimension dim S in Pk S.
From all this it follows that Pk S is the projective cover of S, and so is isomorphic to
G
PS . We have k G
K S = S K by 4.13.
(d) We have seen in the proof of part (c) that Rad(PS ) = (IH Pk )S. It was observed
there that the validity of the equation Rad(Pk S) = (IH Pk )S did not depend on upon
the particular structure of Pk . Thus we see by induction that Radn (PS ) = ((IH)n Pk )S.
Using the identification of Pk as kH that we established in (a) gives the result.
(e) This is immediate from 8.7 and part (a).
Note in 8.8(a) that when Pk is identified with kH using the module action from 8.7,
it is not being identifed with the subring kH of kG, which does not immediately have the
structure of a kG-module.
Proposition 8.8 allows us to give very specific information in the case of groups with
a normal Sylow p-subgroup that is cyclic. This extends the description of the projectives
that we already gave in 8.6 for the case when the Sylow p-subgroup is a direct factor (as
happens, for instance, when G is cyclic). We recall from Chapter 6 that a uniserial module
is one with a unique composition series, and that some equivalent conditions to this were
explored in the exercises to Chapter 6.
122
(8.9) PROPOSITION. Let k be a field of characteristic p and let G = H K where

H = hxi
= Cpn is cyclic of order pn and K is a group of order relatively prime to p. Let
: G GL(W ) be the 1-dimensional representation of kG on which H acts trivially and K
acts via its conjugation action on H/hxp i. Thus if y K conjugates x as y x = xr then (y)
is multiplication by r. If S is any simple kG-module then the projective cover PS is uniserial
n
with radical quotients Radi PS / Radi+1 PS given as S, W S, W 2 S, . . . , W p 1 S =
S.
Proof. We know from 6.2 that the powers IH s are a complete list of the kHsubmodules of Pk , and since they are also kG-submodules in the action described in 8.7
we have a complete list of the kG-submodules of Pk . We thus have a unique composition
series for Pk as a kG-module.
For each element y K as in the statement of the theorem the action of y on IH/(IH)2
is multiplication by r, as the following calculation shows:
y(x 1) = yx 1
= xr 1
= (x 1)(xr1 + + x + 1 r) + r(x 1)
r(x 1) (mod IH 2 ).
More generally for some IH 2 ,
y(x 1)s = (xr 1)s
= (r(x 1) + )s
= r s (x 1)s + sr(x 1)s1 +
r s (x 1)s
(mod IH s+1 ),
and so y acts on the quotient IH s /IH s+1 as multiplication by r s . One way to describe
this is that IH s /IH s+1 = W s , the s-fold tensor power. Thus by 8.8(d) the radical layers
of PS are of the form W s S. These are simple because W has dimension 1, so the
radical series of PS is its unique composition series, by Exercise 5 from Chapter 6.
An algebra for which all indecomposable projective and injective modules are uniserial
is called a Nakayama algebra, so that we have just shown that k[H K] is a Nakayama
algebra when k has characteristic p, H is a cyclic p-group and K has order prime to p.
At least, we have shown that the projectives are uniserial, and the injectives are uniserial
because they are the duals of the projectives. We will see in Corollary 8.13 that for group
algebras over a field, injective modules and projective modules are, in fact, the same thing.
In Proposition 11.8 we will give a complete description of the indecomposable modules
for a Nakayama algebra. It turns out they are all uniserial and there are finitely many of
them.
Peter Webb
123
Let us continue further with our analysis of the composition factors of the indecomposable projective module PS in the situation of Proposition 8.9. Observe that the
isomorphism types of these composition factors occur in a cycle that repeats itself: for
each element y K the map x 7 y x = xr is an automorphism of Cpn and there is a least
positive integer fy such that r fy 1 (mod pn ). This fy divides pn 1, and letting f be
the l.c.m. of all fy with y K we put pn 1 = ef . Then the modules k, W, W 2 , W 3 , . . .
give rise to f different representations. They repeat e times in Pk , except for k which
appears e + 1 times. A similar repetition occurs with the composition factors W i S of
PS .
As a specific example of this, consider the non-abelian group

G = hx, y x7 = y 3 = 1, yxy 1 = x2 i = C7 C3
of order 21 over the field F7 . For the element y we have r = 2 and fy = 3, so that f = 3
and e = 2. There are three simple F7 G-modules by Proposition 8.8, all 1-dimensional,
which we will label k1 , k2 , k4 . The element x acts trivially on all of these modules, and y
acts trivially on k1 , as multiplication by 2 on k2 and as multiplication by 4 on k4 . In the
previous notation, W = k2 , W 2 = k4 and W 3 = k1 . The three projective covers are
uniserial with composition factors as shown.
Pk1
k1
|
k2
|
k4
|
= k1
|
k2
|
k4
|
k1
Pk2
k2
|
k4
|
k1
|
= k2
|
k4
|
k1
|
k2
Pk4
k4
|
k1
|
k2
|
= k4
|
k1
|
k2
|
k4
FIGURE: Indecomposable projectives for F7 G, where G is non-abelian of order 21.

We next consider the projective modules for groups that are a semidirect product
of a p-group and a group of order prime to p, but with the roles of these groups the
opposite of what they were in the last example. We say that a group G has a normal
p-complement if and only if it has a normal subgroup K G of order prime to p with
|G : K| a power of p. Necessarily in this situation, if H is a Sylow p-subgroup of G then
G = K H by the Schur-Zassenhaus theorem. In this situation we also say that G is a
p-nilpotent group, a term that has exactly the same meaning as saying that G has a normal
124
p-complement. To set this property in a context, we mention that a famous theorem of

Frobenius characterizes p-nilpotent groups as those groups G with the property that for
every p-subgroup Q, NG (Q)/CG (Q) is a p-group. This is not a result that we will use here,
and we refer to standard texts on group theory for this and other criteria that guarantee
the existence of a normal p-complement.
(8.10) THEOREM. Let G be a finite group and k a field of characteristic p. The
following are equivalent.
(1) G has a normal p-complement.
(2) For every simple kG-module S, the composition factors of the projective cover PS are
all isomorphic to S.
(3) The composition factors of Pk are all isomorphic to k.

Proof. (1) (2): Let G = K H where p6 |K| and H is a Sylow p-subgroup of G. We
show that kH, regarded as a kG-module via the homomorphism G H, is a projective
module. In fact, since kK is semisimple we may write kK = k U for some kK-module
G
G
G
U , and now kG = kK G
K = k K U K . Here k K = kH as kG-modules (they are
permutation modules with stabilizer K) and so kH is projective, being a summand of kG.
Now if S is any simple kG-module then S k kH is also projective by 8.4, and all its
composition factors are copies of S = S k k since the composition factors of kH are all k
(using 6.3 and 8.5). The indecomposable summands of S k kH are all copies of PS , and
their composition factors are all copies of S.
(2) (3) is immediate.
(3) (1): Suppose that the composition factors of Pk are all trivial. If g G is any
t
element of order prime to p (we say such an element is p-regular) then Pk G
hgi = k for
some t, since khgi is semisimple. Thus g lies in the kernel of the action on Pk and if we
put

K = hg G g is p-regulari
then K is a normal subgroup of G, G/K is a p-group and K acts trivially on Pk . We show
that K contains no element of order p: if g K were such an element, then as Pk G
hgi is
a projective khgi-module, it is isomorphic to a direct
sum of copies of khgi by 8.1, and so

g does not act trivially on Pk . It follows that p6 |K|, thus completing the proof.
In Example 7.5 we have already seen an instance of the situation described in 8.10.
In that example we took G = S3 = K H where H = h(1, 2)i and K = h(1, 2, 3)i and we
worked with a field k of characteristic 2, which for a technical reason was F4 . Note that
if V is the 2-dimensional simple kS3 -module then V kH
= V V since V is projective.
We see from this that the module S kH that appeared in the proof of 8.10 need not be
indecomposable.
Peter Webb
125
Symmetry of the group algebra

Group algebras are symmetric, a term that will be defined before Theorem 8.15,
and this has a number of important consequences for group representation theory. Some
of these consequences may also be deduced in a direct fashion from weaker conditions
than symmetry and we present these direct arguments first. Over a field k, the following
properties of the dual U = Homk (U, k) are either well known or immediate.
(1)
(2)
(3)
(4)
(8.11) PROPOSITION. Let k be a field and U a finite dimensional kG-module. Then

U
= U as kG-modules,
U is semisimple if and only if U is semisimple,
U is indecomposable if and only if U is indecomposable, and
a morphism f : U V is a monomorphism (epimorphism) if and only if f : V U
is an epimorphism (monomorphism).
The first part of the next proposition has already been proved in two different ways in
the exercises to Chapter 4. The proof given here is really the same as one of those earlier
proofs, but it is presented a little differently.
(8.12) PROPOSITION. Let k be a field. Then
(1) kG
= kG as kG-modules, and
(2) a finitely generated kG-module P is projective if and only if P is projective as a
kG-module.

Proof. (1) We denote the elements of kG dual to the basis elements {g g G} by
g, so that g(h) = g,h k, the Kronecker . We define an isomorphism of vector spaces
X
gG
kG kG
X
ag g 7
ag g .
gG
To see that this is a kG-module homomorphism we observe that if x G then

(x
g )(h) = g(x1 h) = g,x1 h = xg,h = x
cg(h)
for g, h G, so that x
g=x
cg.

(2) Since P
= P as kG-modules it suffices to prove one implication. If P is a
n
summand of kG then P is a summand of (kGn )
= kGn , and so is also projective.
126
From part (1) of Proposition 8.12 we might be led to suppose that P

= P whenever
P is finitely generated projective, but this is not so in general. We will see in Corollary
8.16 that if PS is an indecomposable projective with simple quotient S then PS
= (PS )
if and only if S
= S .
We now come to a very important property of projective modules for group algebras
over a field, which is that they are also injective.
(8.13) COROLLARY. Let k be a field.
(1) Finitely generated projective kG-modules are the same as finitely generated injective
kG-modules.
(2) Each indecomposable projective kG-module has a simple socle.
The result is true without the hypothesis of finite generation. It may be deduced
from the finitely generated case and Exercise 3(f) from Section 7 which implies that every
projective kG-module is a direct sum of indecomposable projectives (and, dually, every
injective kG-module is a direct sum of indecomposable injectives).
Proof. (1) Suppose that P is a finitely generated projective kG-module and that there
are morphisms
P
x
with injective. Then in the diagram
is surjective, and so by projectivity of P there exists f : P V such that f = .

Since f = we see that P is injective.
To see that all finitely generated injectives are projective, a similar argument shows
that their duals are projective, hence injective, whence the original modules are projective,
being the duals of injectives.
(2) One way to proceed is to quote Exercise 6 of Chapter 6 which implies that Soc(P )
=
(P / Rad P ) . If P is an indecomposable projective module then so is P and P / Rad P

is simple. Thus so is Soc(P ).
Alternatively, since homomorphisms S P are in bijection (via duality) with homomorphisms P S , if P is indecomposable projective and S is simple then P is also
indecomposable projective and
dim HomkG (S, P ) = dim HomkG (P , S )
n
= dim End(S ) if P is the projective cover of S ,

0
otherwise.
Peter Webb
127
Since dim End(S ) = dim End(S) this implies that P has a unique simple submodule and
Soc(P ) is simple.
An algebra for which injective modules and projective modules coincide is called selfinjective or quasi-Frobenius, so we have just shown that group rings of finite groups over
a field are self-injective. An equivalent condition on a finite dimensional algebra A that
it should be self-injective is that the regular representation A A should be an injective
A-module.
(8.14) COROLLARY. Suppose U is a kG-module, where k is a field, for which there
are submodules U0 U1 U with U1 /U0 = P a projective module. Then U
= P U for
some submodule U of U .
P U0 . Thus
Proof. The exact sequence 0 U0 U1 P 0 splits, and so U1 =
P is isomorphic to a submodule of U , and since P is injective the monomorphism P U
must split.
We will now sharpen part (2) of 8.13 by showing that Soc PS
= S for group algebras,
and we will also show that the Cartan matrix for group algebras is symmetric. These
are properties that hold for a class of algebras called symmetric algebras, of which group
algebras are examples. We say that a finite dimensional algebra A over a field k is a
symmetric algebra if there is a non-degenerate bilinear form ( , ) : A A k such that
(1) (symmetry) (a, b) = (b, a) for all a, b A,
(2) (associativity) (ab, c) = (a, bc) for all a, b, c A.
The group algebra kG is a symmetric algebra with the bilinear form defined on the
basis elements by

1 if gh = 1,
(g, h) =
0 otherwise,
as is readily verified. Notice that this bilinear form may be described on general elements
a, b kG by (a, b) = coefficient of 1 in ab. Having learned that group algebras are symmetric it will be no surprise to learn that matrix algebras are symmetric. When A = Mn (k)
is the algebra of n n matrices over a field k, the trace bilinear form (A, B) = tr(AB)
gives the structure of a symmetric algebra.
We will use the bilinear form on kG in the proof of the next result. Although we only
state it for group algebras, it is valid for symmetric algebras in general.
(8.15) THEOREM. Let P be an indecomposable projective module for a group algebra kG. Then P/ Rad P
= Soc P .
Proof. We may choose a primitive idempotent e kG so that P
= kGe as kG-modules.
We claim that Soc(kGe) = Soc(kG) e, since Soc(kG) e Soc(kG) and Soc(kG) e kGe
so Soc(kG) e kGe Soc(kG) = Soc(kGe), since the last intersection is the largest
128
semisimple submodule of kGe. On the other hand Soc(kGe) Soc(kG) since Soc(kGe) is
semisimple so Soc(kGe) = Soc(kGe) e Soc(kG) e.
Next, Hom(kGe, Soc(kG)e) and e Soc(kG)e have the same dimension by 7.17(3), and
since Soc(kG)e is simple, by 8.13, this is non-zero if and only if Soc(kG)e
= kGe/ Rad(kGe)
by Theorem 7.13. We show that e Soc(kG)e 6= 0.
If e Soc(kG)e = 0 then
0 = (1, e Soc(kG)e)
= (e, Soc(kG)e)
= (Soc(kG)e, e)
= (kG Soc(kG)e, e)
= (kG, Soc(kG)e e)
= (kG, Soc(kGe)).
Since the bilinear form is non-degenerate this implies that Soc(kGe) = 0, a contradiction.
Recall that for any RG-module U we have defined the fixed points of G on U to be
U := {u U gu = u for all g G}. We also define the fixed quotient of G on U to be
UG := U/{(g 1)u 1 6= g G}. Then U G is the largest submodule of U on which G acts
trivially and UG is the largest quotient of U on which G acts trivially.
G
(8.16) COROLLARY. Let k be a field.

(1) If P is any projective kG-module and S is a simple kG-module, the multiplicity of S
in P/ Rad P equals the multiplicity of S in Soc P . In particular
dim P G = dim PG = dim(P )G = dim(P )G .
(2) For every simple kG-module S, (PS )
= PS .
Proof. (1) This is true for every indecomposable projective module, hence also for
every projective module. For the middle equality we may use an argument similar to the
one that appeared in the proof of 8.13 (2).
(2) We have seen in the proof of 8.13 that (PS ) is the projective cover of (Soc PS ) ,
and because of 8.15 we may identify the latter module as S .
From this last observation we are able to deduce that, over a large enough field, the
Cartan matrix of kG is symmetric. We recall that the Cartan invariants are the numbers
cST = multiplicity of S as a composition factor of PT
where S and T are simple. The precise condition we require on the size of the field is that
it should be a splitting field, and this is something that is discussed in the next section.
Peter Webb
129
(8.17) THEOREM. Let k be a field and let S, T be simple kG-modules. The Cartan
invariants satisfy
cST dim EndkG (T ) = cT S dim EndkG (S).
If dim EndkG (S) = 1 for all simple modules S (for example, if k is algebraically closed)
then the Cartan matrix C = (cST ) is symmetric.
Proof. We recall from 7.18 that
cST = dim HomkG (PS , PT )/ dim EndkG (S)
and in view of this we must show that dim HomkG (PS , PT ) = dim HomkG (PT , PS ). Now
HomkG (PS , PT ) = Homk (PS , PT )G
= (PS k PT )G
by 3.3 and 3.4. Since PS k PT is projective by 8.4, this has the same dimension as
(PS k PT )G
= HomkG (PT , PS ),
= (PS k PT )G
using 8.16.
We conclude this section by summarizing some further aspects of injective modules.
We define an essential monomorphism to be a monomorphism of modules f : V U
with the property that whenever g : U W is a map such that gf is a monomorphism
then g is a monomorphism. An injective hull (or injective envelope) of U is an essential
monomorphism U I where I is an injective module. By direct arguments, or by taking
the corresponding results for essential epimorphisms and projective covers and applying
the duality U 7 U , we may establish the following properties for finitely-generated kGmodules.
The inclusion Soc U U is an essential monomorphism.
g
Given homomorphisms W V U , if two of f , g and f g are essential monomorphisms

then so is the third.
A homomorphism f : V U is an essential monomorphism if and only if f |Soc V :
Soc V Soc U is an isomorphism.
U I is an injective hull if and only if I U is a projective cover. Injective hulls
always exist and are unique. From Theorem 8.15 we see that S PS is the injective
hull of the simple module S.
The multiplicity of a simple module S as a composition factor of a module U equals
dim Hom(U, PS )/ dim End(S).
130
1. When G is a p-group and k is a field of characteristic p, the regular representation
kG is indecomposable.
2. The property of projectivity is preserved under induction and restriction.
3. Tensor product with a projective modules gives a projective module.
4. When k is a field of characteristic p we have an explicit description of the projective
kG-modules when G has a normal Sylow p-subgroup, and also when G has a normal pcomplement.
5. G has a normal p-complement if and only if for every simple module S the composition factors of PS are all isomorphic to S.
6. Projective kG-modules are the same as injective kG-modules.
7. kG is a symmetric algebra. Soc PS
= S always. The Cartan matrix is symmetric.
1. Prove that if G is any finite group then the only idempotents in the integral group
ring ZG are 0 and 1.
[If e is idempotent consider the rank of the free abelian group ZGe and also its image
under the homomorphism ZG Fp G for each prime p dividing |G|, which is a projective
Fp G-module. Show that rankZ ZGe is divisible by |G|. Deduce from this that if e 6= 0 then
e = 1.]
2. (a) Let H = C2 C2 and let k be a field of characteristic 2. Show that (IH)2 is a
P
one-dimensional space spanned by hH h.
(b) Let G = A4 = (C2 C2 ) C3 and let F4 be the field with four elements. Compute the
radical series of each of the three indecomposable projectives for F4 A4 and identify each
of the quotients
Radn PS / Radn+1 PS .
Now do the same for the socle series. Hence determine the Cartan matrix of F4 A4 .
[Start by observing that F4 A4 has 3 simple modules, all of dimension 1, which one might
denote by 1, and 2 . This exercise may be done by applying the kind of calculation that
led to Proposition 8.9.]
(c) Now consider F2 A4 where F2 is the field with two elements. Prove that the 20 1
dimensional F2 -vector space on which a generator of C3 acts via
is a simple
1 1
F2 C3 -module. Calculate the radical and socle series for each of the two indecomposable
projective modules for F2 A4 and hence determine the Cartan matrix of F2 A4 .
3. Let G = H K where H is a p-group, K is a p -group, and let k be a field of
characteristic p. Regard kH as a kG-module via its isomorphism with Pk , so H acts as
usual and K acts by conjugation.
Peter Webb
131
(a) Show that for each n, (IH)n is a kG-submodule of kH, and that (IH)n /(IH)n+1 is a
kG-module on which H acts trivially.
(b) Show that
Pk = kH IH (IH)2 (IH)3
is the radical series of Pk as a kG-module.
(c) Show that there is a map
IH/(IH)2 k (IH)n /(IH)n+1 (IH)n+1 /(IH)n+2
x + (IH)2 y + (IH)n+1 7 xy + (IH)n+2
that is a map of kG-modules. Deduce that (IH)n /(IH)n+1 is a homomorphic image of

(IH/(IH)2)n .
(d) Show that the abelianization H/H becomes a ZG-module under the action g xH =
gxg 1 H . Show that the isomorphism IH/(IH)2 k Z H/H specified by (x 1) +
(IH)2 7 1 xH of Chapter 6 Exercise 17 is an isomorphism of kG-modules.
4. The group SL(2, 3) is isomorphic to the semidirect product Q8 C3 where the
cyclic group C3 acts on Q8 = {1, i, j, k} by cycling the three generators i, j and k.
Assuming this structure, compute the radical series of each of the three indecomposable
projectives for F4 SL(2, 3) and identify each of the quotients
Radn PS / Radn+1 PS .
[Use Chapter 6 Exercise 15.]

5. Let G = P S3 be a group that is the semidirect product of a 2-group P and
the symmetric group of degree 3. (Examples of such groups are S4 = V S3 where
V = h(1, 2)(3, 4), (1, 3)(2, 4)i, and GL(2, 3)
= Q8 S3 where Q8 is the quaternion group of
order 8.)
(a) Let k be a field of characteristic 2. Show that kG has two non-isomorphic simple
modules.
(b) Let e1 , e2 , e3 F4 S3 be the orthogonal idempotents that appeared in Example
7.5. Show that each ei is primitive in F4 G and that dim F4 Gei = 2|P | for all i.
[Use the fact that the F4 Gei are projective modules.]
(c) Show that if e1 = () + (1, 2, 3) + (1, 3, 2) then F4 S4 e1 is the projective cover of the
trivial module and that F4 S4 e2 and F4 S4 e3 are isomorphic, being copies of the projective
cover of a 2-dimensional module.
(d) Show that F4 Gei
= F4 h(1, 2, 3)iei G
h(1,2,3)i for each i.
6. Let A be a finite-dimensional algebra over a field k, and let A = Homk (A, k) be the
vector space dual. Regarding A as a right module AA gives (AA ) the structure of a left
A-module via the action (af )(b) = f (ba) where a A, b AA and f (AA ) . Similarly,
regarding A as a left module A A gives (A A) the structure of a right A-module. In fact,
132
both A and A are (A, A)-bimodules. Consider a k-bilinear form ( , ) : A A k.

It determines, and is determined by, a linear map : A A , where (b)(a) = (a, b).
Consider three conditions such a form may satisfy:
(1) (symmetry) (a, b) = (b, a) for all a, b A,
(2) (associativity) (ab, c) = (a, bc) for all a, b, c A.
(3) (ba, c) = (a, cb) for all a, b, c A.
(a) Show that if any two of conditions (1), (2) and (3) hold, then so does the third.
(b) Show that is a map of left A-modules if and only if (2) holds.
(c) Show that is a map of right A-modules if and only if (3) holds.
(d) Show that the following are equivalent (an algebra satisfying any of these conditions is called a Frobenius algebra):
(i) there is a non-degenerate form on A satisfying (2),
(ii) A
= A as left A-modules,
(iii) A
= A as right A-modules,
(iv) there is a non-degenerate form on A satisfying (3).
(e) For a Frobenius algebra (as in (d)), show that the left A-module A A is injective.
(f) Show that A is a symmetric algebra (that is, there is a non-degenerate form on A
satisfying both (1) and (2)) if and only if A
= A as (A, A)-bimodules.
7. Let A be a finite-dimensional algebra over a field k and suppose that the left regular
representation A A is injective. Show that every projective module is injective and that
every injective module is projective.
8. Let S and T be simple kG-modules, with projective covers PS and PT , where k is
an algebraically closed field.
(a) For each n prove that
HomkG (PT , Socn PS )
= HomkG (PT / Radn PT , Socn PS )
n
= HomkG (PT / Rad PT , PS ).

(b) Deduce Landrocks theorem: the multiplicity of T in the nth socle layer of PS equals
the multiplicity of S in the nth radical layer of PT .
(c) Use Exercise 6 of Chapter 6 to show that these multiplicities equal the multiplicity of
T in the nth radical layer of PS , and also the multiplicity of S in the nth socle layer of
PT .
9. Let U be an indecomposable kG-module, where k is a field, and let Pk be the
projective cover of the trivial module. Prove that
dim((
gG
g) U ) =
1 if U
= Pk ,
0 otherwise.
For an arbitrary finite dimensional module V , show that dim((

plicity with which Pk occurs as a direct summand of V .
gG
g) V ) is the multi-
Peter Webb
133
P
[Observe that kGG = PkG = k gG g. Remember that Pk is injective and has socle
isomorphic to k.]
10. Let U be a finite-dimensional kG-module, where k is a field, and let PS be an
indecomposable projective kG-module with simple quotient S. Show that in any decomposition of U as a direct sum of indecomposable modules, the multiplicity with which PS
occurs is equal to
dim HomkG (PS , U ) dim HomkG (PS / Soc PS , U )
dim EndkG (S)
and also to
dim HomkG (U, PS ) dim HomkG (U, Rad PS )

.
dim EndkG (S)
11. Let k be an algebraically closed field of characteristic p and suppose that G has
a normal p-complement, so that G = K H where H is a Sylow p-subgroup of G. Let
S1 , . . . , Sn be the simple kG-modules with projective covers PSi .
Ln
(a) Show that there is a ring isomorphism kG
=
i=1 Mdim Si (EndkG (PSi )) where the
right hand side is a direct sum of matrix rings with entries in the endomorphism rings of
the indecomposable projectives.
[Copy the approach of the proof of Wedderburns theorem.]
(b) For each i, show that if PSi
= Pk Si then EndkG (PSi )
= kH as rings.
[Show that dim EndkG (PSi ) = |H|. Deduce that the obvious map EndkG (Pk ) EndkG (Pk
Si ) is an isomorphism.]
(c) Show that if PSi
6 Pk Si then EndkG PSi has dimension smaller than |H|.
=
12. (a) Show that F3 A4 has just two isomorphism types of simple modules, of dimensions 1 and 3, and that the simple module of dimension 3 is projective.
[Eliminate modules of dimension 2 by observing that a projective cover of such a module
must have dimension at least 6.]
(b) Show that F3 A4
= F3 C3 M3 (F3 ) as rings.
13. Let k be a field of characteristic p and let G = H K where H is a p-group and
K has order prime to p. Show that Radn (kG)
= Radn (kH) G
H as kG-modules.
14. Let D30 be the dihedral group of order 30. By using the fact that D30
= C5 D6
has a normal Sylow 5-subgroup, show that F5 D30 has three simple modules of dimensions
1, 1 and 2. We will label them k1 , k and U , respectively, with k1 the trivial module. Use
the method of Proposition 8.9 to show that the indecomposable projectives have the form
Pk1
k1
|
k
|
= k1
|
k
|
k1
Pk
k
|
k1
|
= k
|
k1
|
k
U
|
U
|
PU = U
|
U
|
U
Finite Group
3 2
Deduce that the Cartan matrix is 2 3

0 0
rings.
Representations
134
0
0 and that F5 D30
= F5 D10 M2 (F5 C5 ) as
5
Peter Webb
135
9. Changing the ground ring: splitting fields and the decomposition map
We examine the relationship between the representations of a fixed group over different
rings. Often we have have assumed that representations are defined over a field that is
algebraically closed. What if the field is not algebraically closed? Such a question is
significant because representations arise naturally over different fields, which might not be
algebraically closed, and it is important to know how they change on moving to an extension
field such as the algebraic closure. It is also important to know whether a representation
may be defined over some smaller field. We introduce the notion of a splitting field, showing
that such a field may always be chosen to be a finite extension of the prime field.
After proving Brauers theorem that over a splitting field of characteristic p the number of non-isomorphic simple representations equals the number of conjugacy classes of
elements of order prime to p, we turn to the question of reducing representations from
characteristic 0 to characteristic p. The process involves first writing a representation in
the valuation ring of a p-local field and then factoring out the maximal ideal of the valuation ring. This gives rise to the decomposition map between the Grothendieck groups
of representations in characteristic 0 and characteristic p. We show that this map is welldefined, and then construct the so-called cde triangle. This provides a very effective way
to compute the Cartan matrix in characteristic p from the decomposition map.
In the last part of this section we describe in detail the properties of blocks of defect
zero. These are representations in characteristic p that are both simple and projective.
They always arise as the reduction modulo p of a simple representations in characteristic
zero, and these are also known as blocks of defect 0. The blocks of defect zero have
importance in character theory, accounting for many zeroes in character tables, and they
are also the subject of some of the deepest investigations in representation theory.
136
Some definitions
Suppose that A is an algebra A over a commutative ring R and that U is an A-module.
If R R is a homomorphism to another commutative ring R we may form the R -algebra
R R A, and now R R U becomes an R R A-module in an evident way. In this section
we study the relationship between U and R R U . When we specialize to a group algebra
A = RG we will identify R R RG with R G.
We will pay special attention to two particular cases of this construction, the first being
when R is a subring of R . If U is an A-module, we say that the module V = R R U
is obtained from U by extending the scalars from R to R ; and if an R R A-module V
has the form R R U we say it can be written in R. In this situation, when U is free as
an R-module we may identify U with the subset 1R R U of R R U , and an R-basis of
U becomes an R -basis of R R U under this identification. In case A = RG is a group
ring, with respect to such a basis of V = R R U the matrices that represent the action
of elements g G on V have entries in R, and are the same as the matrices representing
the action of these elements on U (with respect to the basis of U ). Equally, if we can find
a basis for an R G-module V so that each g G acts by a matrix with elements in R then
RG preserves the R-linear span of this basis, and this R-linear span is an RG-module U
for which V = R R U . Thus an R -free module V can be written in R if and only if V
has an R -basis with respect to which G acts via matrices with entries in R.
The second situation to which we will pay particular attention arises when R = R/I
for some ideal I in R. In this case applying R R to a module U is the same as reducing
U modulo I. If V is an R R A-module of the form R R U for some A-module U we
say that V can be lifted to U , and that U is a lift of V . Most often we will perform this
construction when R is a local ring and I is the maximal ideal of R.
Splitting fields
We start by considering the behaviour of representations over a field. It is often a
help to know that a representation can be written in a small field.
(9.1) PROPOSITION. Let F E be fields where E is algebraic over F and let A be
a finite-dimensional F -algebra. Let V be a finite-dimensional E F A-module. Then there
exists a field K with F K E, of finite degree over F , so that V can be written in K.
Proof. Let a1 , . . . , an be a basis for A and let at act on V with matrix (atij ) with
respect to some basis of V . Let K = F [atij , 1 t n, 1 i, j d]. Then [K : F ] is finite
since K is an extension of F by finitely many algebraic elements, and A acts by matrices
with entries in K.
Peter Webb
137
Let A be a finite-dimensional F -algebra, where F is a field. A simple A-module U

is said to be absolutely simple if and only if E F U is a simple E F A-module for all
extension fields E of F . We say that an extension field E of F is a splitting field for A if
and only if every simple E F A-module is absolutely simple. If A is a group algebra, we
say that E is a splitting field for G, by extension of the terminology.
The kind of phenomenon that these definitions are designed to address is exemplified
by cyclic groups. If G = hgi is cyclic of order n then g acts on CG as a direct sum of
2i
1-dimensional eigenspaces with eigenvalues e n . Since these lie outside Q (if n 3), the
regular representation of QG is a direct sum of simple modules but some of them have
2i
dimension greater than 1. On extending scalars to a field containing e n these simple
modules decompose as direct sums of 1-dimensional modules. Thus Q is not a splitting
2i
field for G if n 3, but any field containing Q(e n ) is a splitting field since the simple
modules are now 1-dimensional and remain simple on extension of scalars.
We will use several times the fact that if we let F be any field, then F is a splitting field
for the matrix algebra Mn (F ). In fact, if E F is a field extension then E F Mn (F )
=
Mn (E), an isomorphism that is most easily seen by observing that Mn (F ) has a basis
consisting of the matrices Eij that are non-zero only in position (i, j), where the entry is
1. Thus E F Mn (F ) has a basis consisting of the elements 1 F Eij . Since these multiply
together in the same fashion as the corresponding basis elements of Mn (E) we obtain the
claimed isomorphism. Every simple Mn (F )-module is isomorphic to the module of column
vectors of length n over F , and on extending the scalars to E we obtain column vectors
of length n over E, which is a simple module for E F Mn (F ). This shows that F is a
splitting field for Mn (F ).
As another example, the prime field Fp is a splitting field for every p-group, since the
only simple Fp G-module here is Fp , which is absolutely simple.
(9.2) PROPOSITION. Let U be a simple module for a finite-dimensional algebra A
over a field F . The following are equivalent.
(1) U is absolutely simple.
(2) EndA (U ) = F .
(3) The matrix algebra summand of A/ Rad A corresponding to U has the form Mn (F ),
where n = dim U .
Proof. (2) (3): If the matrix summand of A/ Rad A corresponding to U is Mn (D)
for some division ring D then D = EndA (U ). The hypothesis is that D = F so the
matrix summand is Mn (F ) and since U identifies as column vectors of length n we have
n = dim U .
(3) (1): The hypothesis is that A acts on U via a surjective ring homomorphism
A Mn (F ) where U is identified as F n . Now if E F is an extension field then E A
acts on E U = E n via the homomorphism E A E Mn (F )
= Mn (E), which is
138
also surjective. Since E n is a simple Mn (E)-module it follows that E F U is a simple

E F A-module.
(1) (2): We prove this implication here only in the situation where F is a perfect
field, so that all irreducible polynomials with coefficients in F are separable. The result
is true in general and is not difficult but requires some technicality that we wish to avoid
(see [CR]). This implication will not be needed for our application of the result.
Suppose that EndA (U ) is larger than F , so there exists an endomorphism : U U
that is not scalar multiplication by an element of F . Let be a root of the characteristic
polynomial of in some field extension E F : in other words, is an eigenvalue of .
Then 1E : E F U E F U is not scalar multiplication by , because if it were the
minimal polynomial of over F would be a factor of (X )n where n = dimF U and by
separability of the minimal polynomial we would deduce F . Now 1E 1U
EndEA (E F U ) is a non-zero endomorphism with non-zero kernel, and since E F U is
simple this cannot happen, by Schurs lemma.
The next theorem is the main result about splitting fields that we will need for the
process of reduction modulo p to be described later in this section.
(9.3) THEOREM. Let A be a finite-dimensional algebra over a field F . Then A has
a splitting field of finite degree over F . If G is a finite group, it has splitting fields that
are finite degree extensions of Q (in characteristic zero) or of Fp (in characteristic p).
Proof. The algebraic closure F of F is a splitting field for A, since by Schurs lemma
condition (2) of Proposition 9.2 is satisfied for each simple F F A-module. By Proposition
9.1 there is a finite extension E F so that every simple F F A-module can be written
in E. The simple E F A-modules U that arise like this are absolutely simple, because if
K E is an extension field for which K E U is not simple then K E U is not simple,
where K is an algebraic closure of K, and since K contains a copy of F , F F U cannot
be simple since F is a splitting field for A, a contradiction.
It is also true that every simple E F A-module is isomorphic to one of the simple
modules U that arise in this way from the algebraic closure F . For, if V is a simple E F Amodule let e2 = e E F A be an idempotent with the property that eV 6= 0 but eV = 0
for all simple modules V not isomorphic to V . Let W be a simple F F A-module that
is not annihilated by e. We must have W
= F E V since V is the only possible simple
module that would give a result not annihilated by e.
Group algebras are defined over the prime field Q or Fp (depending on the characteristic), and by what we have just proved QG and Fp G have splitting fields that are finite
degree extensions of the prime field.
Peter Webb
139
We see from the above that every simple representation of a finite group may be
written over a field that is a finite degree extension of the prime field. In characteristic
zero this means that every representation can be written in such a field, by semisimplicity.
It is not always true that every representation of a finite group in positive characteristic
can be written in a finite field, and an example of this is given as exercise 8 to this section.
Some other basic facts about splitting fields are left to the exercises at the end of this
section. Thus, if A is a finite dimensional algebra over a field F that is a splitting field
for A and E F is a field extension, it is the case that every simple E F A-module
can be written in F (exercises 3 and 4). It is also true for a finite dimensional algebra
that no matter which splitting field we take, after extending scalars we always have the
same number of isomorphism classes of simple modules (exercise 4). Thus in defining the
character table of a finite group, instead of working with complex representations we could
have used representations over any splitting field and obtained the same table.
In positive characteristic the situation is not so straightforward. It is usually not the
case that indecomposable modules always remain indecomposable under all field extensions
(those that do are termed absolutely indecomposable), even when all fields concerned are
splitting fields. We can, however, show that once we are working over a splitting field, the
indecomposable projective modules remain indecomposable under field extension (exercise
7). The consequence of this is that the Cartan matrix does not change once we have a
splitting field, being independent of the choice of the splitting field. Just as when we speak
of the character table of group we mean the character table of representations over some
splitting field, so in speaking of the Cartan matrix of a group algebra we usually mean the
Cartan matrix over some splitting field.
In the case of group algebras there is a finer result about splitting fields than 9.3. It
was first conjectured by Schur and later proved by Brauer as a deduction from Brauers
induction theorem. We state the result, but will not use it and do not prove it. The
exponent of a group G is the least common multiple of the orders of its elements.
(9.4) THEOREM (Brauer). Let G be a finite group, F a field, and suppose that F
contains a primitive mth root of unity, where m is the exponent of G. Then F is a splitting
field for G.
2i
This theorem tells us that Q(e m ) and Fp () are splitting fields for G, where is a
primitive mth root of unity in an extension of Fp . Often smaller splitting fields than these
can be found, and the determination of minimal splitting fields must be done on a caseby-case basis. For example, we may see as a result of the calculations we have performed
earlier in this text that in every characteristic the prime field is a splitting field for S3
the same is in fact true for all the symmetric groups. However, if we require that a field
2i
be a splitting field not only for G but also for all of its subgroups, then Q(e m ) and Fp ()
are the smallest possibilities, since as we have seen earlier that a cyclic group of order n
requires the presence of a primitive nth root of 1 in a splitting field.
Again we will not use it, but it is important to know the following theorem about field
extensions. For a proof see [CRI, p. 139 Section 6 Exercise 6].
140
(9.5) THEOREM (Noether-Deuring). Let A be a finite-dimensional algebra over a

field F and let E F be a field extension. Suppose that U and V are A-modules for which
E F U
= E F V as E F A-modules. Then U
= V as A-modules.
The number of simple representations in positive characteristic

Our next aim is to show that over a splitting field of characteristic p, the number
of non-isomorphic simple representations of a group G equals the number of conjugacy
classes of p-regular elements. Several proofs of this result are available, the first appearing
in a paper of Brauer from 1932. The proof we shall present is also due to Brauer, coming
from 1956. This proof is appealing because it is technically elementary, and could have
appeared earlier in this text once we knew that the radical of a finite-dimensional algebra
is nilpotent.
We start with some lemmas. These have to do with a finite-dimensional algebra A
over a field of characteristic p, and we will write

S = linear span in A of {ab ba a, b A}
n
T = {r A r p S for some n > 0}.
(9.6) LEMMA. T is a linear subspace of A containing S.
Proof. We show first that if a, b A then (a + b)p ap + bp (mod S). To prove

this we use a modification of the familiar binomial expansion argument, but we must be
careful because S need not be an ideal of A, and so it might not be true, for instance,
that aabab aaabb (mod S). On expanding (a + b)p we obtain a sum of terms that
are products of length p. Letting a cyclic group of order p permute these products by
operating on the positions in the product, two terms are fixed (namely ap and bp ) and the
remaining terms all occur in orbits of length p, such as aabab, baaba, abaab, babaa, ababa
when p = 5. The difference of any two of these terms can be expressed as a commutator,
such as aabab baaba = (aaba)b b(aaba), and so lies in S. It follows that all the terms
in an orbit of length p are equal modulo S, and so their sum lies in S, since p = 0.
We next observe that if a, b A then
(ab ba)p (ab)p (ba)p = a(b(ab)p1 ) (b(ab)p1 )a 0 (mod S),
P
so that commutators lie in T . We deduce that S T since if i ci is a linear combination
of commutators then
X
X p p
(
i ci )p
i ci 0 (mod S)
using both formulas we have proved, so that in fact pth powers of elements of S lie in S.
Peter Webb
141
The proof will be completed by showing that T is a linear subspace. Let a, b T , so

n
that ap S, bp S with m n, say. Then

m
(a + b)p p ap + p bp 0 (mod S),

showing that T is closed under taking linear combinations. Here we used the fact that
m
mn
ap = (ap )p
S because pth powers of elements of S lie in S.
(9.7) LEMMA. If A = Mn (F ) is a matrix algebra where F is a field then S = T =

matrices of trace zero.
Proof. Since tr(ab ba) = 0 we see that S is a subset of the matrices of trace
zero. On the other hand when i 6= j every matrix Eij (zero everywhere except for a 1
in position (i, j)) can be written as a commutator: Eij = Eik Ekj Ekj Eik , and also
Eii Ejj = Eij Eji Eji Eij . Since these matrices span the matrices of trace zero we
deduce that S consists exactly of the matrices of trace 0. Now S T A and S has
codimension 1 so either T = S or T = A. The matrix E11 is idempotent and does not lie
in T , so T = S.
(9.8) PROPOSITION. Let A be a finite-dimensional algebra over a field of characteristic p that is a splitting field for A. The number of non-isomorphic simple representations
of A equals the codimension of T in A.
Proof. Let us write T (A), S(A), T (A/ Rad(A)), S(A/ Rad(A)) for the constructions
S, T applied to A and A/ Rad(A). Since Rad(A) is nilpotent it is contained in T (A). Also
(S(A) + Rad(A))/ Rad(A) = S(A/ Rad(A))
n
is easily verified. We claim that T (A)/ Rad(A) = T (A/ Rad(A)). For, if ap S(A) then
n
(a + Rad(A))p (S(A) + Rad(A))/ Rad(A) = S(A/ Rad(A)) and this shows that the
n
left-hand side is contained in the right. Conversely, if (a + Rad(A))p S(A/ Rad(A))

n
then ap S(A) + Rad(A) T (A) so T (A/ Rad(A)) T (A)/ Rad(A).
Now A/ Rad(A) is a direct sum of matrix algebras. It is apparent that both S and
T preserve direct sums, so the codimension of T (A/ Rad(A)) in A/ Rad(A) equals the
number of simple A-modules, and this equals the codimension of T (A) in A.
142
Let p be a prime. An element in a finite group is said to be p-regular if it has order

prime to p, and p-singular if it has order a power of p. The only element that is both
p-regular and p-singular is the identity.
(9.9) LEMMA. Let G be a finite group and p a prime. Each element x G can be
uniquely written x = st where s is p-regular, t is p-singular and st = ts. If x1 = s1 t1 is
such a decomposition of an element x1 that is conjugate to x then s is conjugate to s1 ,
and t is conjugate to t1 .
Proof. If x has order n = where is a power of p and is prime to p then we
may write 1 = + for integers , and put s = x and t = x . If x = st = s1 t1 is
a second such decomposition then s1 commutes with x and hence commutes with s and t
1
which are powers of x. Similarly t1 commutes with s and t. Thus s1
and now
1 s = t1 t
1
1
s1 s is p-regular and t1 t is p-singular, so these products equal 1, and s1 = s, t1 = t. If
x1 = gxg 1 then x1 = gsg 1 gtg 1 is a decomposition of x1 as a product of commuting
p-regular and p-singular elements. Hence s1 = gsg 1 and t1 = gtg 1 by uniqueness of the
decomposition.
(9.10) LEMMA. Let F be a field and G a group. Then S is the set of elements of
F G with the property that the sum of coefficients from each conjugacy class of G is zero.
Proof. S is spanned by elements ab ba with a, b G. Now ab ba = a(ba)a1 ba
is the difference of an element and its conjugate. Such elements exactly span the elements
of F G that have coefficient sum zero on conjugacy classes.
We come now to the result that is the goal of these lemmas.
(9.11) THEOREM. Let F be a splitting field of characteristic p for a finite group G.
The number of non-isomorphic simple F G modules equals the number of conjugacy classes
of p-regular elements of G.
Proof. We know that the number of simple F G modules equals the codimension of
T in F G. We show this equals the number of p-regular conjugacy classes by showing that
if x1 , . . . , xr is a set of representatives of the conjugacy classes of p-regular elements of G
then x1 + T, . . . , xr + T is a basis of F G/T .
If we write x = st where s is p-regular and t is p-singular, s and t commute, then
n
n
n
n
n
n
(st s)p = sp tp sp = sp sp = 0 for sufficiently large n, so that s + T = st + T .
The elements g + T , g G do span F G/T , and now it follows from the last observation
that we may throw out all except the p-regular elements and still have a spanning set. We
show that the set that remains is linearly independent.
P
P
P pn pn
n
Suppose that
i xi T . Then ( i xi )p =
i xi S for sufficiently high
pn
n
p , and there is always a sufficently large n so that xi = xi for all i, since the xi are
Peter Webb
143
P pn
p-regular. Now
i xi S. But x1 , . . . , xr are independent modulo S by Lemma 9.10
pn
so i = 0 for all i, and hence i = 0 for all i. This shows that x1 + T, . . . , xr + T are
linearly independent.
(9.12) COROLLARY. Let F be a splitting field of characteristic p for a finite groups
G1 and G2 . The simple F [G1 G2 ]-modules are precisely the tensor products S1 S2 where
Si is a simple F Gi -module, i = 1, 2, and the action of G1 G2 is given by (g1 , g2 )(s1 s2 ) =
g1 s1 g2 s2 . Two such tensor products S1 S2 and S1 S2 are isomorphic as F [G1 G2 ]modules if and only if Si
= Si as F Gi -modules, i = 1, 2.
We have commented before, after Theorem 4.2, that this kind of result is a special case
of a more general statement about the simple modules for a tensor product of algebras,
and this can be found in [CRI, Theorem 10.38]. The general argument is not hard, but we
can use our expression for the number of simple representations of a group to eliminate
half of it.
Proof. We verify that the modules S1 S2 are simple. This is so since the image of
each F Gi in EndF (Si ) given by the module action is the full matrix algebra EndF (Si ),
by Burnsides theorem that was presented as Exercise 8 in Section 2, and which follows
from the ArtinWedderburn theorem. The image of F [G1 G2 ] in EndF (S1 S2 ) contains
EndF (S1 ) EndF (S2 ) and so by counting dimensions it is the whole of EndF (S1 S2 ).
This implies that S1 S2 is simple.
On restriction to G1 , S1 S2 is a direct sum of copies of S1 , and similarly for G2 ,
so S1 S2
= S1 as F G1 -modules and
= S1 S2 as F [G1 G2 ]-modules if and only if S1
S2
= S2 as F G2 -modules.
We conclude by checking that this gives the right number of simple modules for
G1 G2 . By Brauers theorem this is the number of p-regular conjugacy classes of G1 G2 .
Since (g1 , g2 ) is p-regular if and only if both g1 and g2 are p-regular, and this element is
conjugate in G1 G2 to (g1 , g2 ) if and only if g1 G1 g1 and g2 G2 g2 , the number of
p-regular classes in G1 G2 is the product of the numbers for G1 and G2 .
144
Reduction modulo p and the decomposition map

We turn now to the theory of reducing modules from characteristic zero to characteristic p, for some prime p. This is a theory developed principally by Richard Brauer.
There is inherent interest in studying the relationships between representations in different
characteristics, but aside from this our more specific goals include a remarkable way to
compute the Cartan matrix of a group algebra, and a second proof of the symmetry of this
matrix. After that we study the simple characters whose degree is divisible by the order
of a Sylow p-subgroup of G (the so-called blocks of defect zero). In the next chapter the
same ideas will be used in a proof that the Cartan matrix is non-singular.
There will be three rings in the set-up for reducing modules to characteristic p. We
list them as a triple (F, R, k) where F is a field of characteristic zero equipped with a
discrete valuation, R is the valuation ring in F with maximal ideal (), and k = R/() is
the residue field of R, which is required to have characteristic p. A quick introduction to
valuations and valuation rings is given in an appendix. Such a triple is called a p-modular
system. We may find such systems by taking F to be a finite extension of Q the complex
numbers will not work and this is one of the reasons that we have studied representations
over arbitrary fields.
Given a finite group G, if both F and k are splitting fields for G we say that the
triple is a splitting p-modular system for G. If F contains a primitive mth root of unity,
where m is the exponent of G, then necessarily R and k also contain primitive mth roots of
unity because roots of unity always have valuation 1, and according to Brauers Theorem
9.4 both F and k are then splitting fields. If we do not wish to use Brauers theorem we
may still deduce from 9.3 the existence of splitting p-modular systems where F is a finite
extension of Q and k is a finite field.
We start by studying representations of a finite group over a discrete valuation ring
R. In Propositions 9.15 and 9.16 we assume R is complete, but in other results this is not
necessary, and sometimes all we need is that R is a principal ideal domain with the field
k as a factor ring. We comment also that in the next few results nothing specific about
group representations is used, except for the fact that the group algebra of G over a field
of characteristic zero is semisimple. Many results apply in the generality of an order in a
finite-dimensional semisimple algebra.
(9.13) LEMMA. Let R be a discrete valuation ring with maximal ideal () and residue
field k = R/(). Let G be a finite group.
(1) If S is a simple RG-module then S = 0.
(2) The simple RG-modules are exactly the simple kG-modules made into RG-modules
via the surjection RG kG.
(3) For each RG-module U , U Rad(U ), and in particular RG Rad(RG).
Peter Webb
145
(4) For each RG-module U we have (Rad U )/U = Rad(U/U ).

Proof. (1) S is an RG-submodule of S, so S = S or 0. Since Rad R = () the
R-module homomorphism S S/S is essential by Nakayamas lemma, so that S 6= S.
Therefore S = 0.
(2) This follows from (1) since kG = RG/()G and ()G annihilates the simple RGmodules.
(3) This again follows from (1) since if V is a maximal submodule of U then U/V
is simple so that U V , and it follows that U is contained in all of the maximal
submodules of U and hence in their intersection.
(4) Rad U is the intersection of kernels of all the homomorphisms from U to simple
modules. These homomorphisms all factor through the quotient homomorphism U
U/U , and so Rad U is the preimage in U of the radical of U/U , which is what we have
to prove.
(9.14) COROLLARY. Let R be a discrete valuation ring with maximal ideal () and
residue field k = R/(). Let G be a finite group. Let P and Q be finitely generated
projective RG-modules. Then P
= Q as RG-modules if and only if P/P
= Q/Q as
kG-modules.
Proof. If P/P
= Q/Q as kG-modules then by 9.13 the radical quotients of P and
Q are isomorphic, P/ Rad P
= Q/ Rad Q. Now P and Q are projective covers of their
radical quotients, by Nakayamas lemma 7.6, so P
= Q by uniqueness of projective covers.
The converse implication is trivial.
In the next pair of results we see that some important properties of idempotents and
projective modules that we have already studied in the case of representations over a field,
continue to hold when we work over a complete discrete valuation ring. It is a crucial
hypothesis that the discrete valuation ring be complete. The idea of the proofs is the same
as for the corresponding results over a field.
(9.15) PROPOSITION. Let R be a complete discrete valuation ring with maximal
ideal () and residue field k = R/(). Let G be a finite group. Every expression 1 =
e1 + + en as a sum of orthogonal idempotents in kG can be lifted to an expression 1 =
e1 + + en in RG, where the ei RG are orthogonal idempotents with ei + () RG = ei .
Each idempotent ei is primitive if and only if its lift ei is primitive.
Proof. The proof is very like the proofs of 7.10, 7.11 and 7.12. We start by showing
simply that each idempotent e kG can be lifted to an idempotent e RG. Consider the surjections of group rings (R/( n ))G (R/( n1 ))G for each n 2. Here
( n1 )G/( n )G is a nilpotent ideal in (R/( n))G and so by Theorem 7.9, any idempotent
en1 + ( n1 )G (R/( n1 ))G can be lifted to an idempotent en + ( n )G (R/( n))G.
146
Starting with an element e1 RG for which e1 + ()G = e we obtain a sequence e1 , e2 , . . .

of elements of RG that successively lift each other modulo increasing powers of (), and
so is a Cauchy sequence in RG. (The metric on RG comes from the valuation on R by
taking the distance between two elements to be the maximum of the distances in the coordinate places.) This Cauchy sequence represents an element e RG, since R is complete.
Evidently e is idempotent, because it is determined by its images modulo the powers of
() and these are idempotent. It also lifts e.
The argument that sums of orthogonal idempotents can be lifted now proceeds by
analogy with the proof of 7.11, and the assertion that e is primitive if and only if e is
primitive is proved as in 7.12.
We have seen before in Example 7.16, working with Z as our ground ring, that ZGmodules do not always have projective covers. It is also the case that indecomposable
projective ZG-modules do not always have the form ZGe where e is idempotent. To give
an example of this phenomenon will take us too far afield, but we have seen in exercise 1
of section 8 that the only non-zero module of the form ZGe is ZG itself, since 0 and 1 are
the only idempotents. By using facts from number theory one can show, for example, that
when G is cyclic of order 23 there is an indecomposable non-free projective ZG-module.
In the next result we show that when the ground ring R is a complete discrete valuation
ring such examples can no longer be found.
We will be examining the relationship between RG-modules and their reductions modulo the ideal (). To facilitate this we introduce some terminology. Working over a principal ideal domain R, an RG-module L is called an RG-lattice if it is finitely-generated
and free as an R-module. (In more general contexts an RG-lattice is merely supposed
to be projective as an R-module, but since projective modules are free over a principal
ideal domain we do not need to phrase the definition that way here. We also sometimes
encounter the term (maximal) Cohen-Macaulay module instead of lattice.)
The reason we introduce lattices is that we can reduce them modulo ideals of R. Given
an ideal I of R and an RG lattice L evidently V = L/(I L) is an (R/I)G-module. We
say that V is the reduction modulo I of the lattice L, and also that L is a lift from R/I to
R of V .
Not every (R/I)G-module need be liftable from R/I to R. For example, taking R = Z
and I = (p) with p 3, the group GL(2, p) has a faithful 2-dimensional representation
over Fp that cannot be lifted to Z. Such a lifted representation would have to be faithful
also, and on extending the scalars from Z to R would provide a faithful 2-dimensional
representation over R. It is well known that he only finite subgroups of 2 2 real matrices
are cyclic and dihedral, and when p 3, GL(2, p) is not one of these, so there is no such
faithful representation. On the other hand the 2-dimensional faithful representation of
GL(2, 2) (which is dihedral of order 6) over F2 does lift to Z.
Peter Webb
147
(9.16) PROPOSITION. Let R be a complete discrete valuation ring with maximal

ideal () and residue field k = R/(). Let G be a finite group.
(1) For each simple RG-module S there is a unique indecomposable projective RG-module
PS that is the projective cover of S. It has the form PS = RG
eS where eS is a primitive
idempotent in RG for which eS S 6= 0.
(2) The kG-module PS /( PS )
= PS is the projective cover of S as a kG-module and PS
is the projective cover of PS as an RG-module. Furthermore, S is the unique simple
quotient of PS . Thus for simple kG-modules S and T , PS
= PT if and only if S
= T.
(3) Every finitely-generated RG module has a projective cover.
(4) Every finitely-generated indecomposable projective RG-module is isomorphic to PS
for some simple module S.
(5) Every finitely generated projective kG-module can be lifted to an RG-lattice. Such a
lift is unique, and necessarily projective.
Proof. (1) Let eS kG be a primitive idempotent for which eS S 6= 0 and let
eS RG be an idempotent that lifts eS (as in 9.15), so that eS S = eS S 6= 0. We
define PS = RG
eS . Then PS is projective, and it is indecomposable since eS is primitive.
Furthermore PS /( PS ) = kGeS , and defining this module to be PS it is a projective
cover of S as a kG-module by 7.13.
(2) Now PS / Rad(PS )
= PS / Rad(PS ) by part (4) of 9.13, and this is isomorphic
to S. Thus each of the three morphisms PS PS , PS S and PS S is essential,
by Nakayamas lemma, and so they are all projective covers . Since S is the radical
quotient of PS it is the unique simple quotient of this module. This quotient determines
the isomorphism type of PS by the uniqueness of projective covers.
(3) Let U be a finitely-generated RG-module. Then U/ Rad U is a kG-module by 9.13,
and it is semisimple, so U/ Rad U
= S1 St for various simple modules Si . Consider
the diagram
PS1 PSt
y
U
U/ Rad U
where the vertical arrow is the projective cover of S1 St as an RG-module. By

projectivity we obtain a homomorphism PS1 PSt U that completes the triangle
and it is an essential epimorphism by Proposition 7.7. Thus it is a projective cover.
(4) Let P be a finitely-generated projective RG-module. By part (3) it has a projective
cover, of the form : PS1 PSt P . Since P is projective must split, and there
is a monomorphism : P PS1 PSt with = 1P . Since is an essential
epimorphism must be an epimorphism also, so it is an isomorphism. If we suppose that
P is indecomposable, then t = 1 and P
= PS1 .
(5) It is sufficient to prove the assertion for indecomposable projective modules. The
indecomposable projective kG-modules all have the form PS for some simple module S, and
148
we have seen that such a module lifts to PS , which is a lattice and is the projective cover of
PS . Suppose that L is any RG-lattice for which L/L
= PS . Since L Rad L the radical
quotient of L is S. The projective cover morphism PS S factors as PS L S, giving
an isomorphism on radical quotients. It follows that PS L is surjective by Nakayamas
lemma, and since the ranks of PS and L are the same, this map is an isomorphism.
We next examine the relationship between RG-modules and F G-modules where R is
a principal ideal domain and F is its field of fractions. Given an RG-lattice L we may
regard L as a subset of F R L. In this situation the F G-module F R L may be written
in R, according to the terminology introduced at the start of this chapter. Conversely, if
U is an F G-module, a full RG-lattice U0 in U is defined to be an RG-lattice U0 U that
has an R basis which is also an F -basis of U . In this situation U
= F R U0 and we say
that U0 is an R-form of U . Thus an F G-module that has an R-form can be written in R.
We now show that every finitely-generated F G-module has an R-form.
(9.17) LEMMA. Let R be a principal ideal domain with field of fractions F , and let
U be a finite-dimensional F -vector space. Any finitely-generated R-submodule of U that
contains an F -basis of U is a full lattice in U .
Proof. Let U0 be a finitely-generated R-submodule of U that contains an F -basis of U .
Since U0 is a finitely-generated torsion-free R-module, U0
= Rn for some n, and it has an
R-basis x1 , . . . , xn . Since U0 contains an F -basis of U it follows that x1 , . . . , xn span U over
F . We show that x1 , . . . , xn are independent over F . Suppose that 1 x1 + n xn = 0
for certain i F . We may write i = abii where ai , bi R, since F is the field of fractions
Q
of R. Now clearing denominators we have ( bi )(1 x1 + n xn ) = 0 which implies that
Q
( bi )i = 0 for each i since x1 , . . . , xn is an R-basis. This implies that i = 0 for all i
and hence that n = dim U and x1 , . . . , xn is an F -basis of U .
The kind of phenomenon that the last result is designed to exclude is exemplified by
considering subgroups
of R generated by elements that are independent over Q, such as
the subgroup h1, 2i

= Z2 . This is a free abelian group, but its basis is not an R-basis for
R. According to the last lemma, such a phenomenon would not occur if R were the field
of fractions of Z; indeed, the finitely-generated subgroups of Q are all cyclic.
(9.18) COROLLARY. Let R be a principal ideal domain with field of fractions F ,
and let U be a finite-dimensional F G-module. Then there exists an RG-lattice U0 that is
an R-form for U .
Proof.
Let u1 , . . . , un be any basis for U and let U0 be the R-submodule of U spanned
by {gui i = 1, . . . , n, g G}. This is a finitely-generated R-submodule of U that contains
an F -basis of U . Since R is a principal ideal domain with field of fractions F , by the last
result U0 is a full RG-lattice in U , which is what we need to prove.
Peter Webb
149

We should expect that much of the time when p |G| an F G-module U will contain
various non-isomorphic full sublattices. To show how such non-isomorphic sublattices may
come about, consider an indecomposable projective RG-module P . It often happens that
F R P is not simple as an F G-module. Writing F R P = S1 St as a direct
sum of simple F G-modules and taking a full RG-lattice in each Si , the direct sum of these
lattices will not be isomorphic to P , because P is indecomposable. In this manner we may
construct non-isomorphic R-forms of F R P .
For another specific example, consider a cyclic group G = hgi of order 2 and let
(F, R, k) be a 2-modular system. The regular representation F G contains the full lattice
R 1 + R g that is indecomposable since its reduction kG is indecomposable. It also
contains the full lattice R(1 + g) + R(1 g), which is a direct sum of RG-modules and
hence is decomposable.
The following result is crucial to the definition of the decomposition map, which will
be given afterwards.
(9.19) THEOREM (Brauer-Nesbitt). Let (F, R, k) be a p-modular system, G a finite
group, and U a finitely-generated F G-module. Let L1 , L2 be full RG-lattices in U . Then
L1 /L1 and L2 /L2 have the same composition factors with the same multiplicities, as
kG-modules.
Proof. We observe first that L1 + L2 is also a full RG-lattice in U , by Lemma 9.17,
so by proving the result first for the pair of lattices L1 and L1 + L2 and then for L2 and
L1 + L2 we see that it suffices to consider the case of a pair of lattices, one contained in
the other. We now assume that L1 L2 .
As R-modules, L1 and L2 are free of the same rank, and so L2 /L1 is a torsion module.
Hence L2 /L1 has a composition series as an R-module, and hence also as an RG-module,
because every series of RG-modules can be extended to give a composition series of Rmodules. By working down the terms in a composition series, we see that it suffices to
assume that L1 is a maximal RG-submodule of L2 , and we now make this assumption.
Since L2 /L1 is a simple RG-module we have L2 L1 , by 9.13, and we consider the

chain of sublattices L2 L1 L2 L1 . We must show that L2 /L2 and L1 /L1 have
the same composition factors. The composition factors of L1 /L2 are common to both
L2 /L2 and L1 /L1 , and we will complete the proof by showing that L2 /L1
= L2 /L1 .
In fact, the map
L2 L2 /L1
x 7 x + L1
is a surjection with kernel L1 .
150
We now define the decomposition matrix for a group G in characteristic p. Suppose

that (F, R, k) is a splitting p-modular system for G. The decomposition matrix D is the
matrix with rows indexed by the simple F G-modules and columns indexed by the simple
kG-modules whose entries are the numbers
dT S = multiplicity of S as a composition factor of T0 /T0
where S is a simple kG-module, T is a simple F G-module and T0 is a full RG-lattice in T .
By the theorem of Brauer and Nesbitt these multiplicities are independent of the choice
of full lattice T0 in T . Although it would be possible to define a decomposition matrix
without the assumption that the p-modular system should be splitting, this is never done,
because apart from the inconvenience of having possibly more than one decomposition
matrix, the important relationship that we will see between the Cartan matrix and the
decomposition matrix in Corollary 9.26 does not hold without the splitting hypothesis.
Examples. 1. At the moment the only technique we have to compute a decomposition
matrix is to construct all simple representations in characteristic 0, find lattice forms for
them, reduce these modulo the maximal ideal and compute composition factors. A better
method, in general, is to use Brauer characters as described in Chapter 10. The approach
we have for now does at least allow us to construct the decomposition matrices for S3 in
characteristic 2 and characteristic 3. They are:
1 0
1 0
1 0 and 0 1 .
0 1
1 1
At the end of Chapter 2 we observed that S3 has three simple representations over a field
of characteristic 0: the trivial representation, the sign representation and a 2-dimensional
representation that we have exhibited as a ZS3 -lattice. These three representations index
the rows of the decomposition matrices. In characteristic 2 the simple representations are
the trivial module and the 2-dimensional irreducible with the same matrices as in characteristic 0, but with the entries interpreted as lying in F2 . The latter representation is
simple because S3 permutes the three non-zero vectors in the representation space transitively. In characteristic 3 the simple representations are the trivial representation and
the sign representation. We may show that these lists are complete in various ways. For
example, we could use Theorem 9.11. Another approach is to say that all 1-dimensional
representations are representations of the abelianization C2 of S3 , giving one such representation in characteristic 2 and two in characteristic 3. Now use a count of dimensions of
the simple modules together with Theorem 7.14 to show that the lists are correct.
It is now a question of calculating composition factors of the reductions from characteristic 0. The 1-dimenional representations remain simple on reduction. The 2-dimensional
representation has been given as a Z-form, and it suffices to compute composition factors
when the matrices are interpreted as have entries in F2 and F3 . Over F2 this representation
is simple, as has been observed.
Peter Webb
151
2. When G is a p-group and k has characteristic p, the decomposition matrix has a

single column and an entry for each ordinary simple character, that entry being the degree
of the character. This is a consequence of Proposition 6.3.
3. The third example is sufficiently important that we state it as a separate result. It
describes the situation when |G| is invertible in k, so that both F G and kG are semisimple.
(9.20) THEOREM. Let (F, R, k) be a splitting p-modular system for the group G
and suppose that |G| is relatively prime to p. Then each simple F G-module reduces to
a simple kG-module of the same dimension. This process establishes a bijection between
the simple F G-modules and the simple kG-modules, so that when the two sets of simple
modules are taken in corresponding order, the decomposition matrix is the identity matrix.
The bijection preserves the decomposition into simples of tensor products, Hom spaces as
well as induction and restriction of modules.
Proof. If necessary we may extend (F, R, k) to a larger p-modular system in which
R is complete by completing R with respect to its maximal ideal and replacing F by the
field of fractions of the completion (k remains unchanged in this process). Since F and
k are splitting fields, distinct simple modules remain distinct simple modules under field
extension, and their properties under taking tensor products, Hom groups, induction and
restriction do not change. We thus assume R is complete.
Let T be a simple F G-module with full RG-lattice T0 . Then T0 /T0
= S1 Sn
for various simple kG-modules Si , since kG is semisimple. For the same reason these
modules are projective, so by 9.16 they lift to projective RG-lattices S1 , . . . , Sn that are
the projective covers of S1 , . . . Sn . Thus the projective cover of T0 is a homomorphism
S1 Sn T0 , and this is an isomorphism since the R-ranks of the two modules
are the same. We deduce that T
= (F R S1 ) (F R Sn ), and so n = 1 since
T is simple. Thus every reduction of a simple module is simple. Equally, every simple
kG-module is a composition factor of the reduction of some simple F G-module, since it is
a composition factor of the reduction of F G, and so every simple kG-module does appear
as the reduction of a simple F G-module.
For any F G-modules U and V with R-forms U0 and V0 , U0 R V0 is an R-form of
U F V and
(U0 R V0 )/(U0 R V0 )
= (U0 /U0 ) k (V0 /V0 ).
From this we can see that if U F V decomposes in a certain way as a direct sum of
simple modules, on reduction modulo it gives rise to a corresponding decomposition of
(U0 /U0 ) k (V0 /V0 ) as a direct sum of simple modules. In a similar way we see that
HomF (U, V ) and Homk (U0 /U0 , V0 /V0 ) decompose in a corresponding fashion, as do the
induction and restriction of corresponding modules.
152
Coming out of the proof of the last result we see that in the situation where |G| is
relatively prime to p, an0 RG-module L is projective if and only if it is projective as an
R-module, and furthermore that for each F G-module U , all R-forms of U are isomorphic
as RG-modules. We leave the details of this as an exercise.
When |G| is divisible by p the decomposition matrix cannot be the identity, because
as a consequence of Theorem 9.11 it is not even square. We state without proof a theorem
which says that when G is p-solvable the decomposition matrix does at least contain the
identity matrix as a submatrix of the maximum possible size. A group G is said to be
p-solvable if it has a chain of subgroups
1 = Gn G1 G0 = G
so that each factor Gi /Gi+1 is either a p-group or a group of order prime to p. A proof
can be found in [CRI, 22.1]
(9.21) THEOREM (Fong, Swan, Rukolaine). Let (F, R, k) be a splitting p-modular
system for a p-solvable group G. Then every simple kG-module is the reduction modulo
() of an RG-lattice.
The cde triangle

It is conceptually helpful to express the decomposition matrix as the matrix of a linear
map, and to this end we introduce three groups, which are instances of Grothendieck
groups. These groups should properly be defined in an abstract fashion after which we
would prove that they have a certain structure under the hypotheses in force. For our
purposes it is more direct to skip the abstract step and define the Grothendieck groups in
terms of this structure.
Let (F, R, k) be a splitting p-modular system for a finite group G, and suppose that
F and R are complete with respect to the valuation. We define
G0 (F G) = the free abelian group with the isomorphism types
of simple F G-modules as a basis,
G0 (kG) = the free abelian group with the isomorphism types
of simple kG-modules as a basis,
K0 (kG) = the free abelian group with the isomorphism types
of indecomposable projective kG-modules as a basis.
Thus G0 (F G) has rank equal to the number of conjugacy classes of G, and both G0 (kG)
and K0 (kG) have rank equal to the number of p-regular conjugacy classes of G. If T is a
Peter Webb
153
simple F G-module we write [T ] for the corresponding basis element of G0 (F G). Similarly
if S is a simple kG-module we write [S] for the corrsponding basis element of G0 (kG), and
if P is an indecomposable projective kG-module we write [P ] for the corresponding basis
element of K0 (kG). Extending this notation, if U is any kG-module with composition
factors S1 , . . . , Sr occurring with multiplicities n1 , . . . , nr in some composition series of U ,
we write
[U ] = n1 [S1 ] + + nr [Sr ] G0 (kG).
The fact that this is well defined depends on the JordanHolder theorem. There is a similar
interpretation of [V ] G0 (F G) if V happens to be an F G-module. This time because V
is semisimple it is the direct sum of its composition factors, and so if V = T1n1 Trnr
we put [V ] = n1 [T1 ] + + nr [Tr ] G0 (F G). In the same way if P = P1n1 Prnr
where the Pi are indecomposable projective kG-modules we put
[P ] = n1 [P1 ] + + nr [Pr ] K0 (kG).
Since simple F G-modules biject with their characters, we may identify G0 (F G) with the
subset of the space of class functions Ccc(G) consisting of the Z-linear combinations of the
characters of the simple modules as considered in Chapter 3. Such Z-linear combinations of
characters are termed virtual characters of G, so G0 (F G) is the group of virtual characters
of F G.
We now define the homomorphisms of the cde triangle, which is as follows:
G0 (F G)
e
K0 (kG)
G0 (kG)
The definition of the homomorphism e on the basis elements of K0 (kG) is that if PS is an

indecomposable projective kG-module then e([PS ]) = [F R PS ]. As observed in 9.16 the
lift PS is unique up to isomorphism, and so this map is well-defined. The decomposition
map d is defined thus on basis elements: if V is a simple F G-module containing a full
RG-lattice V0 , we put d([V ]) = [V0 /V0 ]. By Theorem 9.19 this is well-defined, and in
fact the formula works for arbitrary finite-dimensional F G-modules V , not just the simple
ones. This definition means that the matrix of d is the transpose of the decomposition
matrix. The homomorphism c is called the Cartan map and is defined by c([PS ]) = [PS ],
where on the left the symbol [PS ] means the basis element of K0 (kG) corresponding to
the indecomposable projective PS , and on the right [PS ] is an element of G0 (kG). From
P
the definitions we see that the matrix of c is the Cartan matrix: [PT ] = simple S cST [S]
for each simple kG-module T .
154
(9.22) PROPOSITION. c = de.

Proof. It is simply a question of following through the definitions of these homomorphisms. If PS is an indecomposable projective kG-module then e[PS ] = [F R PS ]. To
compute d[F R PS ] we choose any full RG-sublattice of F R PS and reduce it modulo
(). Taking PS to be that lattice, its reduction is PS and so de([PS ]) = [PS ] = c([PS ]).
To investigate the properties of the cde triangle we study the relationship between
homomorphisms between lattices and between their reductions modulo ().
(9.23) PROPOSITION. Let (F, R, k) be a p-modular system. Let U, V be F Gmodules containing full RG-lattices U0 and V0 .
(1) HomRG (U0 , V0 ) is a full R-lattice in HomF G (U, V ).
(2) HomRG (U0 , V0 ) = HomRG (U0 , V0 ) as a subset of HomRG (U0 , V0 )
(3) Suppose that U0 is a projective RG-lattice. Then
HomRG (U0 , V0 )/ HomRG (U0 , V0 )
= HomRG (U0 , V0 /V0 )
= HomkG (U0 /U0 , V0 /V0 ).

Proof. (1) We should explain how it is that HomRG (U0 , V0 ) may be regarded as a
subset of HomF G (U, V ). The most elementary approach is to take R-bases u1 , . . . , ur for
U0 and v1 , . . . , vs for V0 . These are also F -bases for U and V . Any RG-homomorphism
U0 V0 can be represented with respect to these bases by a matrix with entries in R.
Regarding it as a matrix with entries in F , it represents an F G-module homomorphism
U V.
To see that HomRG (U0 , V0 ) is in fact a sublattice of HomF G (U, V ), we observe that
HomRG (U0 , V0 ) HomR (U0 , V0 )
= Rrs where r = dim U , s = dim V . The latter is a free
R-module, so HomRG (U0 , V0 ) is an R-lattice since R is a principal ideal domain. We show
that it is full in HomF G (U, V ). Using the bases for U0 , V0 , let : U V be an F G-module
P
homomorphism. Then (ui ) =
ji vj with ji F . Choose a R so that aji R
for all i, j. Then a : U0 V0 , showing that belongs to F HomRG (U0 , V0 ). Therefore
HomRG (U0 , V0 ) spans HomF G (U, V ) over F .
(2) The map V0 V0 given by x 7 x is an RG-isomorphism so the morphisms
U0 V0 are precisely those that arise as composites U0 V0 V0 , which are in turn

the elements of HomRG (U0 , V0 ).
(3) Consider Hom(U0 , V0 ) Hom(U0 , V0 /V0 ). Its kernel is Hom(U0 , V0 ), which
equals Hom(U0 , V0 ). Since U0 is projective, the map is surjective, and it gives rise to
the first isomorphism. For the second, all homomorphisms : U0 V0 /V0 contain U0
in the kernel, and so factor as U0 U0 /U0 V0 /V0 . The correspondence of and

provides the isomorphism.
Peter Webb
155
(9.24) COROLLARY. Suppose U0 and V0 are full RG-lattices in U and V , and U0

is projective. Then
dimF HomF G (U, V ) = dimk HomkG (U0 /U0 , V0 /V0 ).
Proof. Both sides equal rankR HomRG (U0 , V0 ) by parts (1) and (3) of the last result.
We now identify the map e as the transpose of d.
(9.25) THEOREM. Let (F, R, k) be a splitting p-modular system for G and suppose
that R is complete with respect to its valuation.
(1) Let S be a simple kG-module and let T be a simple F G-module containing a full
RG-lattice T0 . The multiplicity of T in F R PS equals the multiplicity of S as a
composition factor of T0 /T0 .
(2) With respect to the given bases of G0 (F G), G0 (kG) and K0 (kG) the matrix of e is
D and the matrix of d is DT , where D is the decomposition matrix.
The given bases of the Grothendieck groups are the bases whose elements are the
symbols [T ], [S] and [PS ] where T is a simple F G-module and S is a simple kG-module.
Proof. (1) Applying the last corollary to the full RG-lattice PS of F R PS we obtain
dimF HomF G (F R PS , T ) = dimk HomkG (PS , T0 /T0 ). The left side equals dimF EndF G (T )
times the multiplicity of T in F R PS , and the right side equals dimk EndkG (S) times the
multiplicity of S as a composition factor of T0 /T0 . The splitting hypothesis implies that
the endomorphism rings both have dimension 1, and the result follows.
(2) We have already observed when defining d that its matrix is DT . The entries eT S
in the matrix of e are defined by
X
e([PS ]) = [F R PS ] =
eT S T,
T
so that eT S is the multiplicity of T in F R PS . By part (1), eT S = dT S .

We comment that part (1) of the above theorem gives a second proof of the BrauerNesbitt Theorem 9.19 that the decomposition numbers (dT S = the multiplicity of S as a
composition factor of T0 /T0 ) are defined independently of the choice of lattice T0 , since
they have just been shown to be equal to quantities that do not depend on this choice.
It also shows that the decomposition numbers are independent of the choice of p-modular
system (F, R, k), provided it is splitting.
(9.26) EXAMPLE. This result allows us to compute the characters of the indecomposable projective RG-modules PS (or more properly the characters of the F G-modules
F R PS ). Using the decomposition matrices for S3 that were previously computed we see
that in characteristic 2,
P1 = 1 +
P2 = 2 ,
and in characteristic 3
156
P1 = 1 + 2
P = + 2 .
We now have a second proof of the symmetry of the Cartan matrix, but perhaps more
importantly an extremely good way to calculate it. The effectiveness of this approach will
be increased once we know about Brauer characters, which are treated in the next section.
(9.27) COROLLARY. Let (F, R, k) be a splitting p-modular system for G. Then the
Cartan matrix C = DT D. Thus C is symmetric.
Examples. 1. When G = S3 the Cartan matrices in characteristic 2 and in characteristic 3, expressed as a product DT D, are

1 0

1 0

1 1 0
2 0
1 0 1
2 1
1 0 =
and
0 1 =
.
0 0 1
0 1
0 1 1
1 2
0 1
1 1
The decomposition matrices were calculated earlier in this chapter. In characteristic 2 the
Cartan matrix follows from Example 7.5, and in characteristic 3 it follows from Proposition
8.9. The decomposition matrix factorization provides a new way to compute the Cartan
matrices.
2. Results 9.25 and 9.27 fail without the hypothesis that the p-modular system is
splitting. An example of this is provided by the alternating group A4 with the 2-modular
system (Q2 , Z2 , F2 ). The Cartan matrix of F2 A4 was computed in Exercise 2 of Chapter
8, and we leave it as a further exercise to compute the matrices of d and e here.
The equality of dimensions that played the key role in the proof of Theorem 9.25 can
be nicely expressed in terms of certain bilinear pairings between the various Grothendieck
groups. On the vector space of class functions on G we already have defined a Hermitian
form and on the subgroup G0 (F G) it restricts to give a bilinear form
h ,
i : G0 (F G) G0 (F G) Z
specified by h[U ], [V ]i = dim HomF G (U, V ) when U and V are F G-modules. We also have
a pairing
h , i : K0 (kG) G0 (kG) Z
specified by h[P ], [V ]i = dim HomkG (P, V ) when P is a projective kG-module and V is a

kG-module. By Proposition 7.17 this quantity depends only on the composition factors
of V , not on the actual module V , and so this pairing is well-defined. We claim that
each of these bilinear pairings is non-degenerate, since in each case the free abelian groups
have bases that are dual to each other. Thus if U and V are simple F G-modules we have
h[U ], [V ]i = [U],[V ] , and if S and T are simple kG-modules we have h[PS ], [T ]i = [S],[T ].
The equality of dimensions that appeared in the proof of 9.25 can now be expressed as
follows. If x K0 (kG) and y G0 (F G) then he(x), yi = hx, d(y)i. This formalism is an
expression of the fact that e and d are the transpose of each other.
Peter Webb
157
Blocks of defect zero

Blocks of a group algebra were introduced in Chapter 3 as the ring direct summands
of that algebra, or equivalently the primitive central idempotents to which the summands
correspond. They will be studied in a much more complete fashion in Chapter 12. Before
that we study the properties of a special kind of block, known as a block of defect zero.
Taking a splitting p-modular system (F, R, k) for a group G, a block of defect zero may
be defined as a ring summand of kG having a projective simple module. We will see that
such a block can have only one simple module, it is a matrix algebra, and there is a unique
ordinary simple character that reduces to it. These simple representations, over k and
over F , are also called blocks of defect zero, by abuse of terminology. The notion of the
defect of a block will be explained in Chapter 12, and for now we must accept this term
as nothing more than a name.
We will find that blocks of defect zero have a useful consequence in predicting zeros
in the character table of G: such a simple complex character (of degree is divisible by the
p-part of |G|) is zero on all elements of G of order divisible by p. It is easy to identify from
the character table and it happens quite often that the zeros in a character table arise from
blocks of defect zero. This will be explained in Corollary 9.30. Blocks of defect zero are also
useful because they remain simple on reduction mod p, which is helpful when calculating
decomposition matrices. They arise naturally in the context of representations of groups
of Lie type in defining characteristic in the form of the Steinberg representation. Blocks
of defect zero are one of the main ingredients in one of the most significant conjectures in
the representation theory of finite groups: Alperins weight conjecture.
We will make the hypothesis in 9.28 that R should be complete, but this is just a
convenience and the assertions are still true without this hypothesis. The result can be
proved without this hypothesis by first completing an arbitrary p-modular system.
(9.28) THEOREM. Let (F, R, k) be a splitting p-modular system in which R is complete, and let G be a group of order pd q where q is prime to p. Let T be an F G-module of
dimension
n, containing a full RG-sublattice T0 . The following are equivalent.
(1) pd n and T is a simple F G-module.
(2) The homomorphism RG EndR (T0 ) that gives the action of RG on T0 identifies
EndR (T0 )
= Mn (R) with a ring direct summand of RG.
(3) T is a simple F G-module and T0 is a projective RG-module.
(4) The homomorphism kG Endk (T0 /T0 ) identifies Endk (T0 /T0 )
= Mn (k) with a
ring direct summand of kG.
(5) As a kG-module, T0 /T0 is simple and projective.
Proof. (1) (2) Suppose that (1) holds. We will use the formula obtained in Theorem
3.23 for the primitive central idempotent e associated to T , namely
n X
e=
T (g 1 )g
|G|
gG
158

where T is the character of T . Observe that n/|G| R because pd n, and also T (g 1 )
lies in R since it is a sum of roots of unity, and roots of unity (in some extension ring if
necessary) have valuation 1, so lie in R. Thus e RG. It follows that RG = eRG (1
e)RG as a direct sum of rings.
The homomorphism : F G EndF (T ) that expresses the action of G on T identifies
eF G with the matrix algebra EndF (T ), and has kernel (1 e)F G. The restriction of to
RG takes values in EndR (T0 ) because T0 is a full RG-sublattice of T . The kernel of this
restriction is (1 e)F G RG = (1 e)RG which is a direct summand of RG. We will
show that this restricted homomorphism is surjective to EndR (T0 ) and from this it will
follow that eRG
= EndR (T0 )
= Mn (R), a direct summand of RG.
As an extension of the formula in Theorem 3.23 for the primitive central idempotent
corresponding to T , we claim that if EndF (T ) then
=
n X
tr((g 1 ))(g).
|G|
gG
To demonstrate this it suffices to consider the case = (h) where h G, since these
elements span EndF (T ). In this case
n X
n X
tr((g 1 )(h))(g) = (h)
tr((g 1 h))(h1 g)
|G|
|G|
gG
gG
= (h)(e)
= (h)
using the previously obtained formula for e. This shows that the claimed formula holds
when = (h), and hence holds in general.
Finally we may see that the restriction of is a surjective homomorphism RG
P
n
1
EndR (T0 ), since any EndR (T0 ) is the image under of |G|
))g RG.
gG tr((g
This completes the proof of this implication.
(2) (3) Certainly T0 is projective as a module for EndR (T0 ) since it identifies with
the module of column vectors for this matrix algebra. Assuming (2), we have that T0 is
a projective RG-module, since RG acts via its summand eRG which identifies with the
matrix algebra. Furthermore, F G acts on T
= F R T0 as column vectors for a matrix
algebra over F , so T is a simple F G-module.
(2) (4) The decomposition RG = eRG(1e)RG with eRG
= Mn (R) is preserved
on reducing modulo (), and we obtain kG = ekG (1 e)kG where e is the image of e
in kG. Furthermore ekG
= Mn (k) because it is the reduction module () of Mn (R), and
the action of kG on T0 /T0 is via projection onto ekG.
(3) (5) Since T0 is a direct summand of a free RG-module it follows that T0 /T0
is a direct summand of a free kG-module, and hence is projective. Furthermore, we claim
that T0 /T0 is indecomposable. To see this, write T0 /T0 = PS1 PSt where the PSi
are indecomposable projectives, so that T0 /T0 is the projective cover of S1 St as
Peter Webb
159
an RG-module. It follows that T0

= PS1 PSt and T
= F R PS1 F R PSt .
Hence t = 1 since T is simple. By Theorem 9.25 the column of the decomposition matrix
corresponding to S1 consists of zeros except for an entry 1 in the row of T . Since C = DT D,
the multiplicity of S1 as a composition factor of PS1 is 1. We know from Theorem 8.15
that Soc PS1
= S1 , so that S1 occurs as a composition factor
= S1 and also PS1 / Rad PS1
of PS1 with multiplicity at least 2 unless PS1 = S1 . This shows that T0 /T0 is simple.
(4) (5) This is analogous to the proof (2) (3).
(5) (1) Since T0 /T0 is projective its dimension is divisible by pd by Corollary 8.3,
and this dimension equals rankR T0 = dim T . If T were not simple as an F G-module we
would be able to write T = U V , and taking full RG-lattices U0 , V0 the composition
factors of T0 /T0 would be the same as those of U0 /U0 V0 /V0 . Since T0 /T0 is in fact
simple, this situation cannot occur, and T is simple.
Part (5) of Theorem 9.28 appears to depend on the F G-module T , but this is not really
the case. If P is any simple projective kG-module, it can be lifted to an RG-module P and
taking T = F R P we have a module with a full RG-lattice T0 for which T0 /T0
= P.
Thus every simple projective kG-module is acted on by kG via projection onto a matrix
algebra direct summand of kG, and the module T just constructed is always simple.
Notice in Theorem 9.28 that since the full RG-sublattice T0 is arbitrary, every full RGsublattice of T is projective. Since such a full lattice T0 is the projective cover of T0 /T0 ,
which is a simple module defined independently of the choice of T0 , all such lattices T0 are
isomorphic as RG-modules.
Looking at the various equivalent parts of Theorem 9.28, one might be led to suspect
that if k is a field of characteristic p and S is a simple kG-module of dimension divisible
by the largest power of p that divides |G| then S is necessarily projective, but in fact
this conclusion does not always hold. It is known that McLaughlins simple group has an
absolutely simple module in characteristic 2 of dimension 29 7 [Thackray]. The module
is not projective and the 2-part of the group order is 27 .
We present the consequence of Theorem 9.28 for character tables in 9.30, with a
preliminary lemma about character values of projective modules before that.
(9.29) LEMMA. Let (F, R, k) be a p-modular system and G a finite group. Let P
be a projective RG-module and the character of F R P . Then (1) is divisible by the
order of a Sylow p-subgroup of G, and if g G has order divisible by p then (g) = 0.
Proof. Since P / P is a projective kG-module it has dimension divisible by the order
of a Sylow p-subgroup of G (Corollary 8.3), and this dimension equals the rank of P , that
in turn equals (1) = dim F R P .
Consider now an element g G of order divisible by p. To show that (g) = 0 it
suffices to consider P as an Rhgi-module, and as such it is still projective. We may suppose
that P is an indecomposable projective Rhgi-module.
160
Let us write g = st where s is p-regular, t is p-singular and st = ts, as in Lemma 9.9, so

hgi = hsi hti. As in Example 8.6 we can write P / P = S khti where S is a simple khsimodule. Since khsi is semisimple and S is projective as a khsi-module, we can lift S to a
S = S. Now P / P = Skhti = Sk hgi = S hgi .
projective Rhsi-module S for which S/
hsi
hsi
hgi
This lifts to S , which is a projective Rhgi-module. It is the projective cover of P / P ,

hsi
hgi
hgi
so by uniqueness of projective covers P
= S hsi . We deduce that = S hsi . It follows
that (g) = 0 from the formula for an induced character, since no conjugate of g lies in
hsi.
(9.30) COROLLARY. Let T be a simple F G-module, where F is a splitting field for

G of characteristic 0, and let T be the character of T . Let p be a prime, and suppose
that the highest power of p that divides |G| also divides the degree T (1). Then if g is any
element of G of order divisible by p we have T (g) = 0.
Proof. This combines Lemma 9.29 with Theorem 9.28. If F does not initially appear
as part of a p-modular system, we may replace F by a subfield that is a splitting field
and which is a finite extension of Q, since QG has a splitting field of this form and by the
argument of 9.3 it may be chosen to be a subfield of F . We may write T in this subfield
without changing its character T and the hypothesis about the power of p that divides
T (1) remains the same. Take the valuation on F determined by a maximal ideal p of the
ring of integers for which p Z = (p), and complete F with respect to this valuation to
get a splitting, complete, p-modular system. We may now apply Theorem 9.28.
The character table information given by the last corollary can be observed in many
examples. Thus the simple character of degree 2 of the symmetric group S3 is zero except
on the 2-regular elements, and the simple characters of S4 of degree 3 are zero except on
the 3-regular elements. All of the zeros in the character table of A5 may be accounted for
in this way, and all except one of the zeros in the character table of GL(3, 2). It is notable
that in order to prove this result about representations in characteristic zero we have used
technical machinery from characteristic p.
Peter Webb
161
1. Every finite dimensional algebra over a field has a splitting field of finite degree.
2. If two modules are isomorphic after extending the ground field, they were originally
isomorphic.
3. If k is a splitting field of characteristic p, the number of simple kG-modules equals
the number of p-regular conjugacy classes in G.
4. Idempotents lift from the group ring over the residue field to the group ring over a
complete discrete valuation ring.
5. Projective covers exist over RG when R is a complete discrete valuation ring. The
indecomposable projective RG-modules are the projective covers PS , where S is simple.
6. When R is a principal ideal domain with field of fractions F , every F G-module
can be written in R.
7. The decomposition map is well defined and C = DT D. The decomposition matrix
also computes the ordinary characters of the projectives PS .

8. If k is a splitting field of characteristic p where p 6 |G|, the representations of G
over C and over k are the same in a certain sense.
9. Blocks of defect zero are identified by the fact that the degree of their ordinary
character is divisible by the p-part of |G|. Such characters vanish on elements of order
divisible by p. They remain irreducible on reduction mod p, where they give a projective
module and a matrix ring summand of kG.
162

1. Let E = Fp (t) be a transcendental extension of the field with p elements and let F
be the subfield Fp (tp ). Write = tp F , so that tp = 0. Let A = E, regarded as an
F -algebra.
(a) Show that A has a simple module that is not absolutely simple.
(b) Show that E is a splitting field for A, and that the regular representation of E F A
is a uniserial module. Show that Rad(E F A) 6= E F Rad(A).
[Notice that E F Rad(A) is always contained in the radical of E F A when A is a
finite-dimensional algebra, being a nilpotent ideal.]
(c) Show that A is not isomorphic to F G for any group G.
2. Let G be a cyclic group, F a field and S a simple F G-module. Show that E =
EndF G (S) is a field with the property that E F S is a direct sum of modules that are all
absolutely simple.
3. Let A be a finite-dimensional algebra over a field F and suppose that F is a splitting
field for A. Let E F be a field extension. Prove that Rad(E F A)
= E F Rad(A).
[The observation at the end of question 1(b) may help here.]
4. Let A be a finite-dimensional F -algebra.
(a) Suppose that F is a splitting field for A, and let E F be a field extension. Show
that every simple E F A-module can be written in F . Deduce that A and E F A have
the same number of isomorphism classes of simple modules.
[Bear in mind the result of question 3.]
(b) Show that if E1 and E2 are splitting fields for A then E1 F A and E2 F A have
the same number of isomorphism classes of simple modules.
[Assume that E1 and E2 are subfields of some larger field.]
5. Let A be a finite-dimensional F -algebra and let E F be a field extension.
(a) Suppose that S is an A module with the property that E F S is an absolutely
simple E F A-module. Show that S is absolutely simple as an A-module.
(b) Suppose now that E is a splitting field for A and that every simple A-module
remains simple on extending scalars to E. Show that F is a splitting field for A.
6. Let A be a finite-dimensional F -algebra and E F a field extension.
(a) Show that if U V is an essential epimorphism of A-modules then E F U
E F V is an essential epimorphism of E F A-modules.
(b) Show that if P U is a projective cover then so is E F P E F U .
7. Let A be a finite-dimensional F -algebra where F is a splitting field for A. Let P be
an indecomposable projective A-module. Show that if E F is any field extension then
E F P is indecomposable and projective as an E F A-module. Show further that every
indecomposable projective E F A-module can be written in F .
Peter Webb
163
8. Let G = C2 C2 be generated by elements a and b, and let E be a field of

characteristic 2. Let t E be any element, which may be algebraic or transcendental over
F2 . Let : G GL2 (E) be the representation with
(a) =
1
0
1
1
(b) =
1 t
0 1
Show that this representation is absolutely indecomposable, meaning that it remains indecomposable under all field extensions. Show also that this representation cannot be written
in any proper subfield of F2 (t).
9. Let (F, R, k) be a p-modular system and suppose that R is complete. Let L and
M be RG-lattices, where G is a finite group.
(a) Show that L is a projective RG-module if and only if L/L is a projective kGmodule.
[Consider the projective cover of L.]
(b) Deduce that if L/L
= M/M as kG-modules and that L/L is a projective kGM
as
RG-modules.
In other words, projective kG-modules lift uniquely
module then L
=
to RG-lattices.
10. Let (F, R, k) be a p-modular system and G a finite group. Show that if U = U1 U2
is a finite-dimensional F G-module and L is a full RG-lattice in U then L U1 , L U2 are
full RG-lattices in U1 and U2 , but that it need not be true that L = (L U1 ) (L U2 ).
[Consider the regular representation when G = C2 .]
11. Let U be the 2-dimensional representation of S3 over Q that is defined by requiring
that with respect to a basis u1 , u2 the elements (1, 2, 3) and (1, 2) act by matrices

0 1
1 1
and
1
0
1
1
Let U0 be the ZS3 -lattice that is the Z-span of u1 and u2 in U .

(a) Show that U0 /3U0 has just 3 submodules as a module for (Z/3Z)S3 , namely 0,
the whole space, and a 1-dimensional submodule. Deduce that U0 /3U0 is not semisimple.
(b) Now let U1 be the Z-span of the vectors 2u1 + u2 and u1 + u2 in U . Show
(for example, by drawing a picture in which the angle between u1 and u2 is 120 , or else
algebraically) that U1 is a ZS3 -lattice in U , and that it has index 3 in U0 . Write down
matrices that give the action of (1, 2, 3) and (1, 2) on U1 with respect to the new basis.
Show that U1 /3U1 also has just 3 submodules as a (Z/3Z)S3 -module, but that it is not
isomorphic to U0 /3U0 . Identify U0 /U1 as a (Z/3Z)S3 -module.
(c) Prove that U1 is the unique ZS3 -sublattice of U0 of index 3.
(d) Show that the Z3 S3 -sublattices of Z3 S U0 are totally ordered by inclusion.
12. Let (F, R, k) be a splitting p-modular system and G a finite group. Let T be an
F G-module with the property that every full RG-sublattice of T is indecomposable

and
n
n
n+1
projective. Show that T is simple of dimension divisible by p , where p |G|, p
6 |G|.
164
13. Let (F, R, k) be a splitting p-modular system for the group G and suppose that
|G| is relatively prime to p. Let L be a finitely generated RG-module. Show that L
is projective if and only if it is projective as an R-module. Show further that for each
finite-dimensional F G-module U , all R-forms of U are isomorphic as RG-modules.
14. Consider the cde triangle for A4 with the 2-modular system (Q2 , Z2 , F2 ). Compute
the matrices DT and E of the maps d and e with respect to the bases for the Grothendieck
groups described in this section. Verify that E 6= D, but that the Cartan matrix does
satisfy C = DT E and is not symmetric. [Compare Chapter 8 Exercise 2. You may assume
that Q2 contains no primitive third root of 1. This fact follows from the discussion of roots
of unity at the start of the next section, together with the fact that F2 does not contain a
primitive third root of 1.]
15. Give a proof of the following result by following the suggested steps.
THEOREM. Let E F be a field extension of finite degree and let A be an F algebra. Let U and V be A-modules. Then
E F HomA (U, V )
= HomEF A (E F U, E F V )
via an isomorphism F f 7 ( F u 7 F f (u)).
(a) Verify that there is indeed a homomorphism as indicated.
(b) Let x1 , . . . , xn be a basis for E as an F -vector space. Show that for any F -vector
Pn
space M , each element of E F M can be written uniquely in the form i=1 xi F mi
with mi M .
Pn
Pn
(c) Show that if an element i=1 xi fi E F HomA (U, V ) maps to 0 then i=1 xi
fi (u) = 0 for all u U . Deduce that the homomorphism is injective.
(d) Show that the homomorphism is surjective as follows: given an E F A-module
Pn
homomorphism g : E F U E F V , write g(1F u) = i=1 xi fi (u) for some elements
fi (u) V . Show that this defines A-module homomorphisms fi : U V . Show that g is
Pn
the image of i=1 xi fi .
16. Let E F be a field extension of finite degree and let A be an F -algebra.
(a) Let 0 U V W 0 be a short exact sequence of A-modules. Show that
the sequence is split if and only if the short exact sequence 0 E F U E F V
E F W 0 is split.
(b) Let U be an A-module. Show that U is projective if and only if E F U is projective
as an E F A-module.
(c) Let U be an A-module, S an absolutely simple A-module and let LS (U ) denote the
largest submodule of U that is a direct sum of copies of S, as in Corollary 1.8. Show that
LEF S (E F U )
= E F (U G ).
= E F LS (U ). Deduce for fixed points that (E F U )G
Prove a similar result for the largest quotient of U that is a direct sum of copies of S and
hence a similar result for the fixed quotient of U .
Peter Webb
165
(d) Show that if F is assumed to be a splitting field for A and U is an A-module then
Rad(E F U )
= E F Rad(U ) and Soc(E F U )
= E F Soc(U ). Explain why this does
not contradict the conclusion of question 1.
[Hint: use the most promising of the equivalent conditions of Propositions 7.2 and 7.3
in combination with the result of question 15.]
17. Let R be a complete discrete valuation ring with residue field k of characteristic
p. Let g GL(n, k) be an n n-matrix with entries in k. Suppose g has finite order s
and suppose that both R and k contain primitive sth roots of unity. Show that there is
an n n-matrix g GL(n, R) of order s whose reduction to k is g.
18. Show that, over a splitting 2-modular system (F, R, k), the dihedral group D30
has seven 2-blocks of defect zero and two further 1-dimensional ordinary characters. Hence
find the degrees of the simple kD30 -modules. Find the decomposition matrix and Cartan
matrix in characteristic 2. Show that kD30
= kC2 M2 (k)7 as rings.
166
10. Brauer characters

After the success of ordinary character theory in computing with representations over
fields of characteristic zero we wish for a corresponding theory of characters of representations in positive characteristic. Brauer characters provide such a theory and we describe
it in this chapter, starting with the definition and continuing with their main properties, which are similar in many respects to those of ordinary characters. We will arrange
them in tables that satisfy orthogonality relations, although of a more complicated kind
than what we saw with ordinary characters. We will find that the information carried by
Brauer characters is exactly that of the composition factors of a representation, but not
more. They are very effective if that is the information we require, but they will not tell
us anything like the range of complicated possibilities that we find with representations
in positive characteristic. Using Brauer characters is, however, usually the best way to
compute a decomposition matrix, and hence a Cartan matrix in view of Corollary 9.27.
We obtain at the end of this section the theoretical deduction that the Cartan matrix is
invertible.
Brauer characters and computation of the decomposition matrix

To define Brauer characters for a finite group G we start with a p-modular system
(F, R, k) and assume that both F and k contain a primitive ath root of unity, where a is
the l.c.m. of the orders of the p-regular elements of G (the elements of order prime to p).
If we wish to define the Brauer character of a kG-module U where k or F do not contain a
primitive ath root of unity, we first extend the scalars so that the p-modular system does
have this property. Let us put
F = {ath roots of 1 in F }
k = {ath roots of 1 in k.}
The polynomial X a 1 is separable both in F [X] and k[X] since its formal derivative
d
a
a1
is not zero and has no factors in common with X a 1, so both
dX (X 1) = aX
F and k are cyclic groups of order a. Also F R since roots of unity have value 1
under the valuation. We claim that the quotient homomorphism R R/() = k gives
. This is because X a 1 reduces to the
an isomorphism F k , which we write
polynomial that is written the same way, and so its linear factors over F must map to the
over F and
complete set of linear factors over k. These linear factors have the form X
X over k, so we obtain a bijection between the two groups of roots of unity.
Let g G be a p-regular element, and let : G GL(U ) be a representation over k.
Then (g) is diagonalizable, since khgi is semisimple and all eigenvalues of (g) lie in k,
being ath roots of unity. If the eigenvalues of (g) are 1 , . . . , n we put
1 + +
n,
U (g) =
Peter Webb
167
and this is the Brauer character of U . It is a function that is only defined on the p-regular
elements of G, and takes values in a field of characteristic zero, which we may always take
to be C.

0 1
Example. Working over F2 , the specification (g) =
provides a 2-dimensional
1 1
representation U of the cyclic group hgi of order 3. The characteristic polynomial of this
matrix is t2 + t + 1 and its eigenvalues are the primitive cube roots of unity in F4 . These
lift to primitive cube roots of unity in C, and so U (g) = e2i/3 + e4i/3 = 1. It is very
tempting in this situation to observe that the trace of (g) is 1, which can be lifted to
1 C, and thus to suppose that U (g) = 1; however, this supposition would be incorrect.
We list the immediate properties of Brauer characters.
and
(1)
(2)
(3)
(4)
(5)
(6)
(10.1) PROPOSITION. Let (F, R, k) be a p-modular system, let G be a finite group,

let U , V , S be finite-dimensional kG-modules.
U (1) = dimk U .
U is a class function on p-regular conjugacy classes.
U (g 1 ) = U (g) = U (g).
UV = U V .
If 0 U V W 0 is a short exact sequence of kG-modules then V = U +W .
In particular, U depends only on the isomorphism type of U . Furthermore, if U has
P
composition factors S, each occurring with multiplicity nS , then U = S nS S .
(so U = U
/ U
) and the ordinary character of U
is
If U is liftable to an RG-lattice U
U , then U (g) = U (g) on p-regular elements g G.
Proof. (1) In its action on U the identity has dimk U eigenvalues all equal to 1. They
all lift to 1 and the sum of the lifts is dimk U .
(2) This follows because g and xgx1 have the same eigenvalues.
(3) The eigenvalues of g 1 on U are the inverses of the eigenvalues of g on U , as are
the eigenvalues of g on U (since here g acts by the inverse transpose matrix). The lifting
lifts then
1
of roots of unity is a group homomorphism, so the result follows since if
lifts 1 .
(4) If g is a p-regular element then U and V have bases u1 , . . . , ur and v1 , . . . , vs
consisting of eigenvectors of g with eigenvalues 1 , . . . , r and r , . . . , s , respectively. Now
the tensors ui uj form a basis of eigenvectors of U V with eigenvalues i j . Their lifts
j since lifting is a group homomorphism, and P
j = (P
P
are d
i j = i
i,j i
i i )(
j j )
so that UV (g) = U (g)V (g).
(5) If g is a p-regular element then khgi is semisimple so that V
= U W as khgimodules. It follows that the eigenvalues of g on V are the union of the eigenvalues on V
and on W (taken with multiplicity), and from this V (g) = U (g) + W (g) follows. If
U
= U1 we may consider the sequence 0 U1 U 0 0 to see that U = U1 . The
final sentence follows by an inductive argument.
168
then g acts on U = U
/ U
(6) If g is p-regular and acts with eigenvalues 1 , . . . , n on U

with eigenvalues 1 + (), . . . , n + (). Since U (g) is the sum of the lifts of these last
quantities we have (g) = 1 + n = U (g).
We arrange the Brauer characters in tables but, unlike the case of ordinary characters,
there are now two significant tables that we construct: the table of values of Brauer characters of simple modules, and the table of values of Brauer characters of indecomposable
projective modules. By Theorems 9.11and 7.14, if F and k are splitting fields, both these
tables are square. We will eventually establish that they satisfy orthogonality relations
that generalize those for ordinary characters, but we first present some examples.
Examples. 1. Let G = S3 . We have seen that in both characteristic 2 and characteristic 3 the simple representations of S3 lift to characteristic zero, and so the Brauer
characters of the simple modules form tables that are part of the ordinary character table
of S3 . The indecomposable projective modules for a group always lift to characteristic
zero, but if we do not have some information such as the Cartan matrix or the decomposition matrix it is hard to know a priori what their characters might be. In the case of
S3 we have already computed this information, and we now present the tables of Brauer
characters of the simple and indecomposable projective modules.
S3
ordinary characters
S3
Brauer simple p = 2
S3
Brauer projective p = 2
g
|CG (g)|
() (12) (123)
6 2
3
g
|CG (g)|
() (123)
6
3
g
|CG (g)|
() (123)
6
3
1
sign
2
1 1
1 1
2 0
1
2
1
1
2 1
1
2
2
2
2 1
1
1
1
TABLE: Character tables of S3 : ordinary and characteristic 2.

S3
Brauer simple p = 3
S3
g
|CG (g)|
() (12)
6 2
g
|CG (g)|
() (12)
6 2
1
sign
1 1
1 1
1
sign
3 1
3 1
TABLE: Character tables of S3 : characteristic 3.
Peter Webb
169
2. When G = S4 the ordinary character table is

S4
ordinary characters
g
|CG (g)|
() (12) (12)(34) (1234) (123)

24 4
8
4
3
1
sign
2
3a
3b
1
1
1 1
2
0
3 1
3
1
1
1
2
1
1
1
1
0
1
1
1
1
1
0
0
as seen in Example 3.11. In characterstic 2, S4 has two simple modules, namely the trivial
module and the 2-dimensional module of S3 , made into a module for S4 via the quotient
homomorphism S4 S3 . Both of these lift to characteristic zero, and so the Brauer
characters of the simple 2-modular simple representations are
S4
Brauer simple p = 2
g
|CG (g)|
() (123)
24
3
1
2
1
2
1
1
which is the same as for S3 . The Brauer characters of the reductions modulo 2 of the
ordinary characters of S4 are
S4
Reductions from characteristic 0 to 2
g
|CG (g)|
() (123)
24
3
1
sign
2
3a
3b
1
1
2
3
3
1
1
1
0
0
We see from this that the sign representation reduces modulo 2 to the trivial representation, and reductions of the two 3-dimensional representations each have the 2-dimensional
representation and the trivial representation as composition factors with multiplicity 1.
This is because the corresponding Brauer characters are expressible as sums of the simple
170
Brauer characters in this way, and since the simple Brauer characters are visibly independent there is a unique such expression. This expression has to be the expression given by
the composition factor multiplicities in the manner of Proposition 10.1 part (5). It follows
that the decomposition and Cartan matrices for S4 at the prime 2 are
1 0
1 0

1 0
1 0
4 2
1 1 0 1 1
.
D = 0 1 and C =
0 1 =
2 3
0 0 1 1 1
1 1
1 1
1 1
1 1
Knowing the Cartan matrix and the simple Brauer characters we may now compute the
Brauer characters of the indecomposable projective representations.
In characteristic 3 the trivial representation and the sign representation are distinct
1-dimensional representations, and we also have two non-isomorphic 3-dimensional representations that are the reductions modulo 3 of the two 3-dimensional ordinary representations. This is because these 3-dimensional representations are blocks of defect zero, and
by Theorem 9.28 they remain simple on reduction modulo 3. This constructs four simple
representations in characteristic 3, and this is the complete list by Theorem 9.11 because
S4 has four 3-regular conjugacy classes. Thus the table of Brauer characters of simple
modules in characteristic 3 is
S4
Brauer simple p = 3
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
1
sign
3a
3b
1
1
1 1
3 1
3
1
1
1
1
1
1
1
1
1
and each is the reduction of a simple module from characteristic zero. The remaining 2dimensional ordinary representation has Brauer character values 2, 0, 2, 0 and since this
Brauer character is uniquely expressible as a linear combination of simple characters,
namely the trivial Brauer character plus the sign Brauer character, these two 1-dimensional
modules are the composition factors of any reduction modulo 3 of the 2-dimensional representation. We see that the decomposition and Cartan matrices for S4 in characteristic 3
are
1 0 0 0
2 1 0 0
0 1 0 0
1 2 0 0
D = 1 1 0 0 and C = DT D =
.
0 0 1 0
0 0 1 0
0 0 0 1
0 0 0 1
This information now allows us to compute the Brauer characters of the projective representations.
Peter Webb
171
Orthogonality relations and Grothendieck groups

In these examples we exploited the fact that the Brauer characters of the simple
representations turned out to be independent. In fact they always are, and we prove this
as a consequence of orthogonality relations for Brauer characters. The development is
similar to what we did with ordinary characters, the extra ingredient being that some of
the modules we work with must be projective, since these can be lifted from characteristic
p to characteristic 0. We start with an expression for dimensions of homomorphism spaces
in terms of character values.
(10.2) PROPOSITION. Let (F, R, k) be a p-modular system and let G be a finite
group. Suppose that P and U are finite-dimensional kG-modules and that P is projective.
Then
X
1
dim HomkG (P, U ) =
P (g 1 )U (g).
|G|
pregular gG
Proof. We may assume without loss of generality that R is complete. For, if necessary,
replace R by its completion at (), and let F be the field of fractions of R. Making this
change does not alter the residue field k or the Brauer characters of representations, so the
equation we have to establish is unaltered. Assuming that R is complete we may now lift
projective modules from kG to RG.
We make use of the isomorphism HomkG (U, V )
= HomkG (U k V , k) whenever U
and V are finite-dimensional kG-modules, which holds since both sides are isomorphic to
(U k V )G , using results 3.3 and 3.4. Now HomkG (P, U )
= HomkG (P U , k) and P U
d
is a projective kG-module by 8.4. Thus it lifts to a projective RG-lattice P
k U and we
have
dim HomkG (P, U ) = dim HomkG (P U , k)
d
= rank HomRG (P
k U , R)
d
= dim HomF G (F R (P
k U ), F )
1 X
(g 1 )k (g).
=
d
F R (P
kU )
|G|
gG
1 X
(g 1 ).
)
d
F
(P
U
R
k
|G|
gG
We claim that
d
F R (P
kU )
(g
)=
P (g 1 )U (g) if g is p-regular,
0
otherwise,
and from this the result follows. If g is not p-regular it has order divisible by p and the
d
character value is zero by Proposition 9.29, since P
k U is projective. When g is p-regular
172
we calculate the character value by using the fact that it depends only on the structure
of P and U as khgi-modules. Since khgi is a semisimple algebra both P and U are now
c is isomorphic to
projective, and they lift to Rhgi-lattices whose tensor product P R U
d
P
k U as Rhgi-modules, since both of these are projective covers as Rhgi-modules of
P k U . From this we see that
F
d (g
R P k U
1
) = (F P )(F U
c ) (g )
1
= F P (g 1 )F U
c (g )
= P (g 1 )U (g)
as required.
It is convenient to interpret the formula of the last proposition in terms of an inner
product on a space of functions, in a similar way to what we did with ordinary characters.
Let preg(G) denote the set of conjugacy classes of p-regular elements of G, so that
preg(G) cc(G) where the latter denotes the set of all conjugacy classes of G. Since
Brauer characters are constant on the conjugacy classes of p-regular elements we may
regard them as elements of the vector space Cpreg(G) of functions
preg(G) C.
We define a Hermitian form on this vector space by
h, i =
1
|G|
(g)(g)
pregular gG
and just as with the similarly-defined bilinear form on Ccc(G) we note that
h, i = h, i
where (g) = (g) is the complex conjugate of (g). If and are the Brauer characters
of representations we have (g) = (g 1 ) so that h, i = h, i = h , i = h , i.
With the notation of this bilinear form the last result now says that if P and U are
finite-dimensional kG-modules with P projective then
dim HomkG (P, U ) = hP , U i.
(10.3) THEOREM (Row orthogonality relations for Brauer characters). Let G be a
finite group and k a splitting field for G of characteristic p. Let S1 , . . . , Sn be a complete
list of non-isomorphic simple kG-modules, with projective covers PS1 , . . . , PSn . Then the
Brauer characters S1 , . . . , Sn of the simple modules form a basis for Cpreg(G) , as do also
Peter Webb
173
the Brauer characters PS1 , . . . , PSn of the indecomposable projective modules. These
two bases are dual to each other with respect to the bilinear form, in that
hPSi , Sj i = i,j .
The bilinear form on Cpreg(G) is non-degenerate.
Proof. Since PSi has Si as its unique simple quotient we have HomkG (PSi , Sj ) = 0
unless i = j, in which case HomkG (PSi , Sj )
= EndkG (Si ). Because k is a splitting field,
this endomorphism ring is k and so hPSi , Sj i = i,j . Everything follows from this and
the fact that the number of non-isomorphic simple modules equals the number of p-regular
Pn
Pn
conjugacy classes of G. Thus if i=1 j Sj = 0 we have j = hPSi , i=1 j Sj i = 0,
which shows that the Sj are independent, and hence form a basis. By a similar argument
the PSi also form a basis. The matrix of the bilinear form with respect to these bases is
the identity matrix and it is non-degenerate.
This result says that the rows of the matrices of Brauer characters of simples and projectives are orthogonal to each other, provided that entries from each column are weighted
by the reciprocal of the centralizer order of the element that parametrizes the column, this
being the number of conjugates of the element divided by |G|. We will give examples of this
later. The result implies, of course, that the Brauer characters of the simple kG-modules
are linearly independent as functions on the set of p-regular conjugacy classes of G, a fact
we observed and used in the earlier examples. It has the consequence that the information
contained in a Brauer character is exactly that of composition factor multiplicities and we
see this as part (c) of the next result. Since there is the tacit assumption with Brauer
characters that we might have extended the field of definition to include roots of unity, we
describe the behaviour of composition factors under field extension.
(10.4) COROLLARY. Let E k be a field extension and let U and V be finitedimensional kG-modules.
(a) If S and T are non-isomorphic simple kG-modules then E k S and E k T have no
composition factors in common.
(b) U and V have the same composition factors as kG-modules if and only if E k U and
E k V have the same composition factors as EG-modules.
(c) U and V have the same composition factors if and only if their Brauer characters U
and V are equal.
Proof. (a) As in Theorem 7.13, let fS kG be an idempotent such that fS S = S and
fS T = 0. Now fS (E k S) = E k S and fS (E k T ) = 0, so fS acts as the identity on
all composition factors of E k S, but as zero on all composition factors of E k T . Thus
E k S and E k T can have no composition factors in common.
174
(b) The implication from left to right is immediate since extending scalars is exact.
Conversely, the composition factors of E k U are precisely the composition factors of the
E k S where S is a composition factor of U . Since non-isomorphic S and T give disjoint
sets of composition factors in E k S and E k T , if the composition factors of E k U
and E k V are the same, the composition factors of U and V that give rise to them must
be the same.
(c) By part (b) we may extend k if necessary to assume that it is a splitting field. We
know from part (5) of 10.1 that U is the sum of the Brauer characters of the composition
factors of U , and similarly for V , so if U and V have the same composition factors then
U = V . Conversely, the composition factors of U are determined by U since by Theorem
10.3 the Brauer characters of simple modules form a basis for Cpreg(G) . Hence if U = V
then U and V must have the same composition factors.
Computing the Brauer character of a module and expressing it as a linear combination
of simple characters is a very good way of finding the composition factors of the module,
and some examples of this appear in the exercises. As another application we may deduce
that if S is a simple kG-module then S
= S if and only if the Brauer character S takes
real values, since this is the condition that S = S = S .
It also follows from Theorem 10.3 that the Brauer characters of the indecomposable
projective modules are independent functions on preg(G), and so Brauer characters enable us to distinguish between projective modules. Since the Brauer characters of a module
are determined by its composition factors this has the following consequence.
(10.5) COROLLARY. Let P and Q be finitely generated projective kG-modules,
where k is a splitting field of characteristic p. Then P and Q are isomorphic if and
only if they have the same composition factors. The Cartan matrix is invertible. The
decomposition matrix has maximum rank.
Pn
Proof. Write P = PSa11 PSann and Q = PSb11 PSbnn so that P = i=1 ai PSi
Pn
and Q =
b . Then P
= Q if and only if ai = bi for all i, if and only if
i=1
Pn
Pni PSi
i=1 ai PSi =
i=1 bi PSi since the PSi are linearly independent, if and only if P = Q .
By the last corollary this happens if and only if P and Q have the same composition factors.
We claim that the kernel of the Cartan homomorphism c : K0 (kG) G0 (kG) is zero.
Any element of K0 (kG) can be written [P ] [Q] where P and Q are projective modules,
and such an element lies in the kernel if and only if P and Q have the same composition
factors. This forces P
= Q, so that the kernel is zero. It now follows that the Cartan
homomorphism is an isomorphism.
For the final assertion about the decomposition matrix we use the fact that C = DT D.
Thus rank C rank D. The number of columns of D equals the number of columns of C,
and this number must be the rank of D.
Peter Webb
175
A finer result than this is true. The determinant of the Cartan matrix over a splitting field of characteristic p is known to be a power of p, and the decomposition map
d : G0 (F G) G0 (kG) is, in fact, a surjective homomorphism of abelian groups. The
assertion in 10.5 that its matrix has maximum rank implies only that the cokernel of the
decomposition map is finite. These stronger results may be proved by a line of argument that originates with the induction theorem of Brauer, a result that we have omitted.
The surjectivity of of the decomposition map is also implied by the Fong-Swan-Rukolaine
theorem 9.21 in the case of p-solvable groups.
We may write the row orthogonality relations in various ways. The most rudimentary
way, for simple kG-modules S and T , is the equation
n
X
1
1 if S
= T,
PT (g 1 )S (g) =
|G|
0 otherwise.
pregular gG
We can also express this as a matrix product. Let be the table of Brauer character
values of simple kG-modules, the table of Brauer character values of indecomposable
projective modules, and let B be the diagonal matrix whose entries are |CG1(g)| as g ranges
through the p-regular classes. The row orthogonality relations may now be written as
BT = I.
Note that the independence of the Brauer characters of simple modules and of projective
modules is equivalent to the property that and are invertible matrices, and also that
= C where C is the Cartan matrix. From this we now deduce the column orthogonality
relations for Brauer characters, which say that each column of is orthogonal to the
remaining columns of , and has product with the corresponding column of equal to
the order of the centralizer of the group element that indexes the column.
(10.6) PROPOSITION (Column orthogonality relations for Brauer characters). With
the above notation,
|C (x )|
0
G
T
=
0
..
.
|CG (x2 )|
..
.
.
|CG (xn )|
..
where x1 , . . . , xn are representatives of the p-regular conjugacy classes of elements of G.

Thus
n
X
S (g 1 )PS (h) = |CG (g)| if g and h are conjugate,
0
otherwise.
simple S
1
Proof. We take the equation BT = I and multiply on the left by and on the
1
right by (T )1 to get B = (T )1 . Inverting both sides gives B 1 = T . We
finally take the complex conjugate of both sides, observing that B is a real matrix.
176
The orthogonality relations can be used to determine the composition factors of a

representation in a similar way to the procedure with ordinary characters. The idea is that
we obtain the composition factor multiplicity of a simple kG-module S in another module U
as hPS , U i (assuming k is a splitting field). However this possibility is less useful than in
characteristic zero because we need to know the Brauer characters of the indecomposable
projective PS to make it work. Usually we would only have this information once we
already have fairly complete information about the simple modules, so that we know the
decomposition matrix, the Cartan matrix and hence the table of Brauer characters of
projectives. Such an approach is less useful in constructing the table of simple Brauer
characters. By contrast, in characteristic zero we can test for simplicity of a character
and subtract known character summands from a character of interest without complete
character table information.
The orthogonality relations can also be used to decompose a module that is known
to be projective into its projective summands, and here they are perhaps more useful.
Working over a splitting field, the idea is that if P is a projective kG-module and S is a
simple kG-module, then the multiplicity of PS as a direct summand of P equals hP , S i
(assuming k is a splitting field).
EXAMPLE. We present an example in which the orthogonality relations for Brauer
characters are used to find the composition factors of a module, and also the decomposition
of a projective module into indecomposable projective summands. We have already seen
that the table of simple Brauer characters of S4 in characteristic 3 is
S4
Brauer simple p = 3
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
1
sign
3a
3b
1
1
1 1
3 1
3
1
1
1
1
1
1
1
1
1
and the Cartan matrix C = DT D is

S4
Cartan matrix p = 3
1 sign 3a 3b
1
sign
3a
3b
2
1
0
0
1
2
0
0
0
0
1
0
0
0
0
1
Peter Webb
177
Thus the table of Brauer characters of projectives is

S4
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
1
sign
3a
3b
3
1
3 1
3 1
3
1
3
3
1
1
1
1
1
1
In these tables of Brauer characters of projective modules we are writing S = PS for

the Brauer character of the projective cover of S. Consider the first of the 3-dimensional
simple modules, which we denote 3a. It is a block of defect zero, being the reduction mod
3 of a 3-dimensional simple module in characteristic zero, and hence is projective. The
Brauer character of its tensor square is
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
3a3a = 3a 3a
Taking first the inner products hPS , 3a3a i with the Brauer characters of projectives we
obtain the numbers
27 1 3 1
+ + + =2
24 4 8 4
27 1 3 1
+ =1
24 4 8 4
27 1 1 1
+ =1
24 4 8 4
27 1 1 1
+ =1
24 4 8 4
which means that 3a 3a has composition factors 1 (with multiplicity 2), , 3a, 3b where
is the sign representation. On the other hand 3a 3a is itself projective, and we get more
information if we take the inner products h3a3a , S i with the simple Brauer characters,
giving numbers
1 1 1
9
+ + + =1
24 4 8 4
1 1 1
9
+ =0
24 4 8 4
27 1 1 1
+ =1
24 4 8 4
27 1 1 1
+ = 1.
24 4 8 4
178
This shows that 3a 3a

= P1 3a 3b as kG-modules.
The spaces of functions defined on conjugacy classes of G that we have been considering are very closely related to the Grothendieck groups previously defined. We know that
as S ranges through the isomorphism classes of simple kG-modules the Grothendieck group
G0 (kG) has the elements [S] as a basis, the group K0 (kG) has the elements [PS ] as a basis,
and the space Cpreg(G) has the Brauer characters S and also the Brauer characters PS
as bases. Similarly when T ranges through the isomorphism classes of simple F G-modules
the Grothendieck group G0 (F G) has the elements [T ] as a basis, and the space Ccc(G) has
the ordinary characters T as a basis. There are thus group homomorphisms
K0 (kG) Cpreg(G)
G0 (kG) Cpreg(G)
G0 (F G) Ccc(G)
defined on the basis elements [P ] K0 (kG), [S] G0 (kG) and [T ] G0 (F G), by

[P ] 7 P ,
[S] 7 S ,
and
[T ] 7 T
and in fact these same formulas hold whenever P is an arbitrary finitely-generated projective module and S, T are arbitrary finitely-generated modules. Extending scalars from Z
to C they define isomorphisms of vector spaces
C Z K0 (kG)
= Cpreg(G)
C Z G0 (kG)
= Cpreg(G)
C Z G0 (F G)
= Ccc(G) .
There is further structure that we have not mentioned yet, which is that there is a
multiplication defined on the three Grothendieck groups by [U ] [V ] := [U V ], the same
formula working for all three groups but with the modules U and V interpreted suitably
over F G or kG and projective or not as appropriate. Since tensor product over the ground
field preserves exact sequences, this definition makes sense for arbitrary (finitely generated)
modules. This multiplication makes G0 (F G) and G0 (kG) into rings with identity, the
identity element being the class of the trivial module. We see that the isomorphisms just
given preserve the product structure.
(10.7) PROPOSITION. After extending scalars to C, the Grothendieck groups C Z
G0 (F G) and C Z G0 (kG) are semisimple rings.
Proof. The isomorphisms to Ccc(G) and Cpreg(G) are ring isomorphisms, and the
latter algebras are direct sums of copies of C, which are semisimple.
Peter Webb
179
This means that after extending scalars to C we may identify the cde triangle with
the following diagram of vector spaces
Ccc(G)
e
Cpreg(G)
Cpreg(G)
where, as before, the maps c, d, e have matrices C, DT and D, respectively.

(10.8) PROPOSITION. Let (F, R, k) be a splitting p-modular system for G. With
this interpretation of the cde triangle, if Cpreg(G) then e() is the function that is
the same as on p-regular conjugacy classes and is zero on the other conjugacy classes.
If Ccc(G) then d() is the restriction of to the p-regular conjugacy classes. The
Cartan homomorphism c is an isomorphism, the decomposition map d is surjective, and
e is injective. The image of e is the space of class functions that are non-zero only on
p-regular classes.
Proof. The description of e() follows from Proposition 9.29 and the definition of
e, whereas the description of d comes from part (6) of Proposition 10.1. The Cartan
homomorphism is an isomorphism because the Cartan matrix is non-singular, and because
the decomposition matrix D has maximal rank d is surjective and e is injective. The image
of e is a space whose dimension is the number of p-regular conjugacy classes of G, and it
is contained in the space of maps whose support is the set of p-regular conjugacy classes,
so we must have equality.
1. Brauer characters take values in C. They satisfy similar properties to ordinary
characters, except that they are only defined on p-regular elements of G.
2. Two Brauer characters U and V are equal if and only if U and V have the same
composition factors.
3. The tables of Brauer characters of simple modules and of projective modules satisfy
orthogonality relations.
180

1. Let H be a subgroup of G. If Cpreg(H) is a function defined on the set of
p-regular conjugacy classes of elements of H we may define an induced function G
H on
the p-regular conjugacy classes of elements of G by
G
H (g) =
=
1
|H|
(t1 gt)
tG
t1 gtH
(t1 gt).
t[G/H]
t1 gtH
This is the same as the formula in Proposition 4.10 that was used to define induction of
ordinary class functions.
(a) Let U be a finite-dimensional kH-module with Brauer character U , where k is
G
G
a field of characteristic p. Prove that UG
= U G
H , and that e( H ) = e() H and
H
G
d( G
H ) = d() H if is a class function.
preg(G)
(b) Similarly define the restriction G
and show that similar
H where C
formulas hold.
(c) Using the Hermitian forms defined on these spaces of functions, show that h G
H
, i = h, G
i
always
holds.
H
2. Let (F, R, k) be a splitting p-modular system for G. Let P and Q be finitelygenerated projective RG-modules such that F R P
= F R Q as F G-modules. Show that
P = Q.
3. The simple group GL(3, 2) has order 168 = 8 3 7. The following is part of its
ordinary character table (the numbers that label the conjugacy classes of elements in the
top row indicate the order of the elements):
GL(3, 2)
ordinary characters
g
|CG (g)|
1
3
?
6
7
8
1 2 4
168 8 4
1 1 1
3 1 1
3 7a 7b
3 7 7
1
6 2 0
1 1
7 1 1 1
8
1 1 1
Here = 7 + 72 + 74 where 7 = e2i/7 . Note that 2 =

1 and
= 2.
(a) Obtain the complete character table of GL(3, 2).
(b) Compute the table of Brauer characters of simple F2 [GL(3, 2)]-modules.
Peter Webb
181
(c) Find the decomposition matrix and Cartan matrix of GL(3, 2) at the prime 2.
(d) Write down the table of Brauer characters of projective F2 [GL(3, 2)]-modules.
(e) Determine the direct sum decomposition of the module 83 (where 8 and 3 denote
the simple F2 [GL(3, 2)]-modules of those dimensions shown in the table), as a direct sum
of indecomposable modules.
(f) Determine the composition factors of 3 3 and 3 3 .
[Note that 3 can be taken to be the natural 3-dimensional F2 [GL(3, 2)]-module. One
approach is to use the orthogonality relations.]
4. It so happens that GL(3, 2)
= P SL(2, 7). In this exercise we regard this group as
P SL(2, 7) and assume the construction of the simple modules over F7 given in exercise 21
of Chapter 6.
(a) Construct the table of Brauer characters of simple F7 [P SL(2, 7)]-modules.
[It will help to observe that all elements of order prime to 7 in SL(2, 7) act semisimply
on F27 by Maschkes theorem. From this it follows that there is a unique element of order
2 in SL(2, 7), namely I, since its eigenvalues must both be 1, and it is semisimple.
We may deduce that an element of order 8 in SL(2, 7) represents an element of order 4
in P SL(2, 7), and its square represents an element of order 2, since I represents 1 in
P SL(2, 7). We may now determine the eigenvalues of 7-regular elements in their action
on the symmetric powers of the 2-dimensional module.]
(b) Compute the decomposition and Cartan matrices for P SL(2, 7) in characteristic
7. Show that the projective cover of the trivial module, P1 , has just four submodules,
namely 0, P1 and two others.
5. Let (F, R, k) be a p-modular system for some prime p. For the following two
statements, show by example that the first is false in general, and that the second is true.
(a) Suppose that A is an invertible matrix with entries in R and let A be the matrix
with entries in k obtained by reducing the entries of A modulo (). Then the eigenvalues
of A are the lifts of the eigenvalues of A.
(b) Suppose that A is an invertible matrix with entries in R and let A be the matrix
with entries in k obtained by reducing the entries of A module (). Suppose further that
A has finite order, and that this order is prime to p. Then the eigenvalues of A are the
lifts of the eigenvalues of A. Furthermore, the order of A is the same as the order of A.
6. (Modular version of Moliens Theorem. Copy the approach of Chapter 4 Exercise 14
and assume Exercise 5.) Let G be a finite group and let (F, R, k) be a splitting p-modular
system for G, for some prime p.
(a) Let : G GL(V ) be a representation of G over k, and for each n let S n (V ) be
d
the Brauer character of the nth symmetric power of V . For each p-regular g G let (g)
be a matrix with entries in R of the same order as (g) that reduces modulo () to (g).
Show that for each p-regular g G there is an equality of formal power series
n=0
S n (V ) (g)tn =
1
d
det(1 t(g))
182
Here t is an indeterminate, and the determinant that appears in this expression is of a

matrix with entries in the polynomial ring C[t], so that the determinant is a polynomial
in t. On expanding the rational function on the right we obtain a formal power series that
is asserted to be equal to the formal power series on the left.
[Choose a basis for V so that g acts diagonally, with eigenvalues 1 , . . . , d . Show that on
P
both sides of the equation the coefficient of tn is equal to i1 ++id =n 1i1 did .]
(b) If W is a simple kG-module we may write the multiplicity of W as a composition
factor of S n (V ) as hPW , S n (V ) i and consider the formal power series
MV (W ) =
X
i=0
Show that
MV (W ) =
1
|G|
hPW , S n (V ) itn .
gpreg(G)
PW (g 1 )
.
d
det(1 t(g))
(c) When G = S3 and V is the 2-dimensional simple F2 S3 -module show that

1 + t3
= 1 + t2 + 2t3 + t4 + 2t5 + 3t6 + 2t7 + 3t8 + 4t9 + 3t10 +
2
3
(1 t )(1 t )
t(1 + t)
MV (V ) =
= t + t2 + t3 + 2t4 + 2t5 + 2t6 + 3t7 + 3t8 + 3t9 + 4t10 +
(1 t2 )(1 t3 )
MV (k) =
Show that S 8 (V ) is the direct sum as a F2 S3 -module of three copies of V and a 3dimensional module whose composition factors are all trivial. [Note that the ring of
1
invariants (S (V ))G has series (1t2 )(1t
3 ) in this case, that is not the same as MV (k).]
(d) When G = GL(3, 2) and V is the natural 3-dimensional F2 GL(3, 2)-module, use
the information from Exercise 3 to show that
1 t + t4 t7 + t8
(1 t)(1 t3 )(1 t7 )
t(1 t3 + t4 + t5 )
MV (V ) =
(1 t)(1 t3 )(1 t7 )
t2 (1 + t t2 + t5 )
MV (V ) =
(1 t)(1 t3 )(1 t7 )
t4
MV (8) =
(1 t)(1 t3 )(1 t7 )
MV (k) =
7. Show that when k is an arbitrary field of characteristic p, the number of isomorphism classes of simple kG-modules is at most the number of p-regular conjugacy classes
in G.
8. Let k be a splitting field for D30 in characteristic 3. Using the fact that D30
=
C3 D10 has a normal Sylow 3-subgroup, show that kD30 has four simple modules of
Peter Webb
183
dimensions 1, 1, 2 and 2 and that the values of the Brauer characters on the simple
modules are the same as the ordinary character table of D10 . We will label the simple
modules k1 , k , U a and U b respectively. Using the method of Proposition 8.9, show that
the indecomposable projectives have the form
Pk1
k1
|
= k
|
k1
Pk
k
|
= k1
|
k
PUa
U a
U b
|
|
= U a PUb = U b
|
|
U a
U b
Deduce that the Cartan matrix is
2
1
C=
0
0
1
2
0
0
0
0
3
0
0
0
0
3
and that kD20

= kD6 M2 (kC3 ) M2 (kC3 ) as rings. Compute the decomposition matrix
and calculate a second time the Cartan matrix as DT D.
184
11. Indecomposable modules

We recall from Chapter 6 that a module is said to be indecomposable if it cannot
be expressed as the direct sum of two non-zero submodules. We have already considered
many indecomposable modules, but mainly only those that are projective or simple. An
exception to this in Chapter 6 is that we classified in their entirety the indecomposable
modules for a cyclic p-group over a field of characteristic p. To understand indecomposable
modules in general might be considered one of the goals of representation theory, because
such understanding would enable us to say something about all finite dimensional representations, as they are direct sums of indecomposables. We will explain in this chapter
that this goal is not realistic, but that nevertheless there is much that can be said. We try
to give an overview of the theory, much of which has as its natural context the abstract
representation theory of algebras. We can only go into detail with a small part of this
general theory, focusing on what it has to say for group representations, and the particular
techniques that are available in this case.
Our first task is to understand the structure of the endomorphism rings of indecomposable modules. This leads to the Krull-Schmidt theorem, which specifies the extent to
which decompositions as direct sums of indecomposable modules are unique. We go on to
describe in detail the indecomposable modules in a particular situation: the case a group
with a cyclic normal Sylow p-subgroup. After describing the theory of relative projectivity
we discuss representation type, from which the conclusion is that most of the time we
cannot hope to get a description of all indecomposable modules. Vertices, sources, Green
correspondence and the Heller operator complete the topics covered.
Peter Webb
185
Indecomposable modules, local rings and the Krull-Schmidt theorem

We start with an extension of Proposition 7.4.
(11.1) PROPOSITION. Let U be a module for a ring A with a 1. Expressions
U = U1 Un
as a direct sum of submodules biject with expressions 1U = e1 + + en for the identity
1U EndA (U ) as a sum of orthogonal idempotents. Here ei is obtained from Ui as
the composite of projection and inclusion U Ui U , and Ui is obtained from ei as
Ui = ei (U ). The summand Ui is indecomposable if and only if ei is primitive.
Proof. We must check several things. Two constructions are indicated in the statement
of the proposition, whereby given a direct sum decomposition of U we obtain an idempotent
decomposition of 1U , and vice-versa. It is clear that the idempotents constructed from a
module decomposition are orthogonal and sum to 1U . Conversely, given an expression
1U = e1 + + en as a sum of orthogonal idempotents, every element u U can be
written u = e1 u + + en u where ei u ei U = Ui . In any expression u = u1 + un
with ui ei U we have ej ui ej ei U = 0 if i 6= j so ei u = ei ui = ui , and this expression
is uniquely determined. Thus the expression 1U = e1 + + en gives rise to a direct sum
decomposition.
We see that Ui decomposes as Ui = V W if and only if ei = eV + eW can be
written as a sum of orthogonal idempotents, and so Ui is indecomposable if and only if ei
is primitive.
(11.2) COROLLARY. An A-module U is indecomposable if and only if the only
non-zero idempotent in EndA (U ) is 1U .
Proof. From the proposition, U is indecomposable if and only if 1U is primitive,
and this happens if and only if 1U and 0 are the only idempotents in EndA (U ). This
last implication in the forward direction follows since any idempotent e gives rise to an
expression 1U = e + (1U e) as a sum of orthogonal idempotents, and in the opposite
direction there simply are no non-trivial idempotents to allow us to write 1U = e1 + e2 .
The equivalent conditions of the next result are in fact satisfied by the endomorphism
ring of an indecomposable module, but we first present them in abstract. The connection
with indecomposable modules will be presented in Corollary 11.5.
(1)
(2)
(3)
(4)
(5)
(6)
186
(11.3) PROPOSITION. Let B be a ring with 1. The following are equivalent.

B has a unique maximal left ideal.
B has a unique maximal right ideal.
B/ Rad(B) is a division ring.
The set of elements in B that are not invertible forms a left ideal.
The set of elements in B that are not invertible forms a right ideal.
The set of elements in B that are not invertible forms a 2-sided ideal.
Proof. (1) (3) Let I be the unique maximal left ideal of B. Since Rad(B) is the
intersection of the maximal left ideals, it follows that I = Rad(B). If a B I then
Ba is a left ideal not contained in I, so Ba = B. Thus there exists x B with xa = 1.
Furthermore x 6 I, so Bx = B also and there exists y B with yx = 1. Now yxa = a = y
so a and x are 2-sided inverses of one another. This implies that B/I is a division ring.
(1) (6) The argument just presented shows that the unique maximal left ideal I is
in fact a 2-sided ideal, and every element not in I is invertible. This implies that every
non-invertible element is contained in I. Equally, no element of I can be invertible, so I
consists of the non-invertible elements, and they form a 2-sided ideal.
(3) (1) If I is a maximal left ideal of B then I Rad(B) and so corresponds to
a left ideal of B/ Rad(B), which is a division ring. It follows that either I = Rad(B) or
I = B, and so Rad(B) is the unique maximal left ideal of B.
(4) (1) Let J be the set of non-invertible elements of B and I a maximal left ideal.
Then no element of I is invertible, so I J. Since J is an ideal, we have equality, and I
is unique.
(6) (4) This implication is immediate, and so we have established the equivalence
of conditions (1), (3), (4) and (6).
Since conditions (3) and (6) are left-right symmetric, it follows that they are also
equivalent to conditions (2) and (5), by analogy with the equivalence with (1) and (4).
We will call a ring B satisfying any of the equivalent conditions of the last proposition
a local ring. Any commutative ring that is local in the usual sense (i.e. it has a unique
maximal ideal) is evidently local in this non-commutative sense. As for non-commutative
examples of local rings, we see from Proposition 6.8 part (3) that if G is a p-group and k
is a field of characteristic p then the group algebra kG is a local ring. This is because its
radical is the augmentation ideal and the quotient by the radical is k, which is a division
ring, thus verifying condition (3) of Proposition 11.3.
We have seen in Corollary 11.2 a characterization of indecomposable modules as modules whose endomorphism ring only has idempotents 0 and 1. We now make the connection
with local rings.
Peter Webb
187
(11.4) PROPOSITION.
(1) In a local ring the only idempotents are 0 and 1.
(2) Suppose that B is an R-algebra that is finitely generated as an R-module, where R is
a complete discrete valuation ring or a field. If the only idempotents in B are 0 and
1 then B is a local ring.
Proof. (1) In a local ring B, any idempotent e other than 0 and 1 would give a nontrivial direct sum decomposition of B = Be B(1 e) as left B-modules, and so B would
have more than one maximal left ideal, a contradiction.
(2) Suppose that 0 and 1 are the only idempotents in B, and let () be the maximal
ideal of R. Just as in the proof of part (1) of Proposition 9.13 we see that annihilates
every simple B-module, and so B Rad(B). This implies that B/ Rad(B) is a finitedimensional R/()-algebra. If e B/ Rad(B) is idempotent then by the argument of
Proposition 9.15 it lifts to an idempotent of B, which must be 0 or 1. Since e is the image
of this lifting, it must also be 0 or 1. Now B/ Rad(B)
= Mn1 (1 ) Mnt (t ) for
certain division rings i , since this is a semisimple algebra, and the only way this algebra
would have just one non-zero idempotent is if t = 1 and n1 = 1. This shows that condition
(3) of the last proposition is satisfied.
We put these pieces together:
(11.5) COROLLARY. Let U be a module for a ring A.
(1) If EndA (U ) is a local ring then U is indecomposable.
(2) Suppose that R is a complete discrete valuation ring or a field, A is an R-algebra,
and U is finitely-generated as an R-module. Then U is indecomposable if and only if
EndA (U ) is a local ring. In particular this holds if A = RG where G is a finite group.
Proof. (1) This follows from 11.2 and 11.4.
(2) From 11.2 and 11.4 again all we need to do is to show that EndA (U ) is finitelygenerated as an R-module. Let Rm U be a surjection of R-modules. Composition with
this surjection gives a homomorphism EndA (U ) HomR (Rm , U ), and it is an injection
since Rm U is surjective (using the property of Hom from homological algebra that
it is left exact and the fact that A-module homomorphisms are a subset of R-module
homomorphisms). Thus EndA (U ) is realized as an R-submodule of HomR (Rm , U )
= U m,
which is a finitely generated R-module. Since R is Noetherian, the submodule is also
finitely-generated.
188
The next result is a version of the Krull-Schmidt theorem. We first present it in

greater generality than for group representations.
(11.6) THEOREM (Krull-Schmidt). Let A be a ring with a 1, and suppose that U
is an A-module that has two A-module decompositions
U = U1 Ur = V1 Vs
where for each i, EndA (Ui ) is a local ring and Vi is an indecomposable A-module. Then
r = s and the summands Ui and Vj are isomorphic in pairs when taken in a suitable order.
Proof. The proof is by induction on max{r, s}. When this number is 1 we have
U = U1 = V1 , and this starts the induction.
Now suppose max{r, s} > 1 and the result is true for smaller values of max{r, s}.
For each j let j : U Vj be projection onto the jth summand with respect to the
Ps
decomposition U = V1 Vs , and let j : Vj U be inclusion. Then j=1 j j = 1U .
Now let : U U1 be projection with respect to the decomposition U = U1 Ur
and : U1 U be inclusion so that = 1U1 . We have
1U1 = (
s
X
j=1
j j ) =
s
X
j j
j=1
and since EndA (U1 ) is a local ring it follows that at least one term j j must be
invertible. By renumbering the Vj if necessary we may suppose that j = 1, and we
write = 1 1 . Now (1 1 )(1 ) = 1U1 and so 1 : U1 V1 is split mono and
1 1 : V1 U1 is split epi. It follows that 1 (U1 ) is a direct summand of V1 . Since V1
is indecomposable we have 1 (U1 ) = V1 and 1 : U1 V1 must be an isomorphism.
We now show that U = U1 V2 Vs . Because 1 is an isomorphism, 1

is one-to-one on the elements of U1 . Also 1 is zero on V2 Vs and it follows
that U1 (V2 Vs ) = 0, since any element of the intersection is detected by its
image under 1 , and this must be zero. The submodule U1 + V2 + + Vs contains
V2 + + Vs = Ker 1 and so corresponds via the first isomorphism theorem for modules
to a submodule of 1 (U ) = V1 . In fact 1 is surjective and so U1 + V2 + + Vs = U . It
follows that U = U1 V2 Vs .
We now deduce that U/U1
= V2 Vs . It follows by induction
= U2 Ur
that r = s and the summands are isomorphic in pairs, which completes the proof.
Peter Webb
189
Note that the proof of 11.6 shows that an exchange lemma property holds for the
indecomposable summands in the situation of the theorem. After the abstraction of general
rings, we state the Krull-Schmidt theorem in the context of finite group representations,
to make things clear.
(11.7) COROLLARY. Let R be a complete discrete valuation ring or a field and G
a finite group. Suppose that U is a finitely-generated RG-module that has two decompositions
U = U1 Ur = V1 Vs
where the Ui and Vj are indecomposable RG-modules. Then r = s and the summands Ui
and Vj are isomorphic in pairs when taken in a suitable order.
Proof. We have seen in Corollary 11.5 that the rings EndRG (Ui ) are local, so that
Theorem 11.6 applies.
Groups with a normal cyclic Sylow p-subgroup

Before developing any more abstract theory we present an example where we can
describe explicitly all the indecomposable modules. The example will help to inform our
later discussion. Let k be a field of characteristic p and suppose that G has a normal
cyclic Sylow p-subgroup H. Thus by the Schur-Zassenhaus theorem G = H K where
H = hxi
= Cpn is cyclic of order pn and K is a group of order relatively prime to p. In this
situation we have seen in Proposition 8.9 that all the indecomposable projective modules
are uniserial, and we also gave a description of their composition factors. Since projective
and injective modules are the same thing for a group algebra over a field, it follows that
all the indecomposable injective modules are uniserial too. Such an algebra, where the
indecomposable projective and injective modules are all uniserial is called a Nakayama
algebra.
(11.8) PROPOSITION. Let A be a finite dimensional Nakayama algebra over a field.
Then every indecomposable module is uniserial and is a homomorphism image of an indecomposable projective module. It follows that the homomorphic images of indecomposable
projectives form a complete list of indecomposable modules.
Proof. Let M be an indecomposable A-module and let be its Loewy length. Consider
any surjective homomorphism Ad M where Ad is a free module of some rank d. At
least one indecomposable summand P of Ad must have image with Loewy length in
M (otherwise M would have shorter Loewy length than ). Let U be the image of this
P M , so U is a uniserial submodule of M of Loewy length . Consider now the injective
190
hull U I of U . Since U has a simple socle the injective module I is indecomposable. By

the property of injectivity we obtain a factorization U M I. The image of M in I
can have Loewy length no larger than , and since this is the Loewy length of U the image
must have Loewy length equal to . Because I is uniserial and has a unique submodule of
each Loewy length, it follows that U and M have the same image in U . Composing with
the inverse of the isomorphism from U to its image in I, we get a homomorphism M U
that splits U M , so that U is a direct summand of M . Since M is indecomposable,
U = M.
We have just established that M is a homomorphic image of an indecomposable projective. Equally, any such module is indecomposable since it has a unique simple quotient,
so we have a complete list of indecomposables.
(11.9) COROLLARY. Let k be a field of characteristic p and suppose that G = H K
where H = hxi
= Cpn is cyclic of order pn and K is a group of order relatively prime to
p. The number of isomorphism classes of indecomposable kG-modules equals pn lk (K)
where lk (K) is the number of isomorphism classes of simple kK-modules.
Proof. We have seen in Proposition 8.9 that each indecomposable projective is uniserial of Loewy length pn , so has pn non-isomorphic homomorphic images. Since there are
lk (K) simple kG-modules and hence indecomposable projectives, the result follows.
From these results we can immediately describe all the indecomposable modules for
cyclic groups or, indeed,abelian groups with a cyclic Sylow p-subgroup, the dihedral group
D2p in characteristic p when p is odd, or for groups such as the non-abelian group of order
21 in characteristic 7 (as well as other more complicated groups). For the groups just
mentioned the subgroup K is abelian of order prime to p, so lk (K) = |K| and the number
of isomorphism classes of indecomposable modules happens to equal |G|. Thus when p is
odd, D2p has 2p indecomposable modules and when p is 7, C7 C3 has 21 indecomposable
modules: they are the homomorphic images of the three projectives shown after 8.9.
Peter Webb
191
Relative projectivity
The relationship between the representations of a group and those of its subgroups
are one of the most important tools in representation theory. In the context of modular
representations of groups this relationship shows itself in a refinement of the notion of
projectivity, namely relative projectivity. We will use it to give a criterion for actual
projectivity and to determine the group rings of finite representation type. Finally we will
describe the theory of vertices and sources of indecomposable modules, which appear as
part of Green correspondence.
Let H be a subgroup of G and R a commutative ring with 1. An RG-module is said to
be H-free if it has the form V G
H for some RH-module V . It is H-projective, or projective
relative to H, if it is a direct summand of a module of the form V G
H for some RH-module
V.
For example, the regular representation RG
= R G
1 is 1-free, and projective modules
are 1-projective. If R is a field then every 1-projective module is projective, but if R is not
a field and V is an R-module that is not projective as an R-module, then V G
1 is 1-free
but not projective as an RG-module. Every RG-module is G-projective.
In order to investigate relative projectivity we first deal with some technicalities. We
P
have seen the pervasive importance of the group ring element gG g at every stage of the
development of representation theory. As an operator on any representation of G it has
image contained in the G-fixed points. We now consider something more general, and for
H
a subgroup H of G and an RG-module U we define the relative trace map trG
U G.
H :U
To define this we choose a set of representatives g1 , . . . , gn of the left cosets of H in G,
Pn
so G = g1 H gn H. If u U H we define trG
H (u) =
i=1 gi u. To complete the
picture we mention that when H G there is an inclusion of fixed points that we denote
g
G
resG
U H . When H G and g G there is also a map cg : U H U ( H) given by
H :U
u 7 gu. These operations behave like induction, restriction and conjugation of modules.
(11.10) LEMMA. Let U be an RG-module and let K H G and L be subgroups
of G.
H
(1) The homomorphism trG
U G is well-defined.
H :U
H
G
(2) trG
H trK = trK ,
G
G
resH
K resH = resK ,
g
H
cg trH
K = trg K cg ,
g
H
cg resH
K = resg K cg and
H
trH
H = 1 = resH .
H
(3) trH
: K|.
K resK is multiplication by |H
P
g
G G
H
(4) (Mackey formula) resL trH = g[L\G/H] trL
Lg H resLg H cg .
Proof. (1) We could have chosen different coset representatives of H in G, in which

case the different set we might have picked would have the form g1 h1 , . . . , gn hn for certain
192
Pt
elements h1 , . . . , ht in H. Now the definition of trG
would be trG
(u) = i=1 gi hi u, but
H
H
Pt
this equals i=1 gi u as before, since u is fixed by all of the hi .
(2) Many of these formulas are quite straightforward and we only prove the first and
the third. For the first, if we take left coset representatives h1 , . . . , hm of K in H and left
coset representatives g1 , . . . , gn of H in G then the set of all elements gi hj is a set of left
coset representatives for K in G and their sum is (g1 + gn )(h1 + + hm ). Now
H
G H
trG
K (u) = (g1 + gn )(h1 + + hm )u = (g1 + gn )trK (u) = trH trK (u).
For the third formula, the elements g h1 , . . . , g hm are a set of left coset representatives
Pm g
Pm
g
H
for K in g H and so trg H
K cg (u) =
i=1 hi gu =
i=1 ghi u = cg trK (u).
(3) Continuing with the notation in use, if u is fixed by H then
g
H
trH
K resK (u)
m
X
hi u =
i=1
m
X
u = mu
i=1
where m = |H : K|.
(4) Considering the action from the left of L on the cosets G/H, for any g G the
L-orbit containing the coset gH consists of the cosets agH where a lies in L and ranges
through representatives of the cosets L/(L g H). This is because the cosets in the L-orbit
g
are the agH with a L, and a1 gH = a2 gH if and only if a1
2 a1 StabL (gH) = L H,
which happens if and only if a1 and a2 lie in the same coset of L g H in L. Thus on
partitioning the cosets G/H into L-orbits we see that g[L\G/H] {ag a [L/(L g H)]}
is a set of left coset representatives for H in G. Hence if u U H ,
X
X
X
trG
agu =
trL
Hu =
Lg H cg u.
gL\G/H] a[L/(Lg H)]
g[L\G/H]
The most important situation where we will use the relative trace map is when the
RG-module U is a space of homomorphisms HomR (X, Y ) between RG-modules X and
Y . In this situation HomR (X, Y )H = HomRH (X, Y ) for any subgroup H, so that the
relative trace map from H to G is a homomorphism of R-modules trG
H : HomRH (X, Y )
HomRG (X, Y ).
In the following result it would have been technically correct to insert resG
H in several places, but since this operation is simply the inclusion of fixed points it seems more
transparent to leave it out.
(11.11) LEMMA. Suppose that : U V and : W X are homomorphisms
G
of RG-modules and that : V G
H W H is an RH-module homomorphism. Then
G
G
G
G
(trH ) = trH ( ) and (trH ) = trH ( ).
Proof. Let g1 , . . . , gn be a set of left coset representatives for H in G and u U .
Pn
Pn
1
1
G
Then (trG
=
u) =
Similarly
H )(u)
i=1 gi (gi P
i=1 gi (gi u) = trH ()(u).
P
n
n
1
1
G
G
(trH )(u) = i=1 gi (gi u) = i=1 gi (gi u) = trH ()(u).
Peter Webb
193
(11.12) COROLLARY. Let U be an RG-module and let H be a subgroup of G.

(1) The image of trG
H : EndRH (U ) EndRG (U ) is an ideal.
G
(2) The map resH : EndRG (U ) EndRH (U ) is a ring homomorphism.
Proof. The first property is immediate from 11.11, which implies that the image of
is closed under composition with elements of EndRG (U ). The second statement is not
a corollary, but it is immediate because resG
H is the inclusion map.
trG
H
It will help us to consider the adjoint properties of induction and restriction of modules
in detail. We have seen in Corollary 4.13 that when H G, U is an RH-module and V is
G
an RG-module, we have HomRG (U G

H , V ) = HomRH (U, V H ). There may be many such
isomorphisms, but there is a choice that is natural in U and V . This means that whenever
U1 U2 is an RG-module homomorphism the resulting square
HomRG (U1 G
H , V )
x
HomRG (U2 G
H , V )
HomRH (U1 , V G
H)
x
HomRH (U2 , V G
H)
commutes, as does the square
G
HomRG (U G
H , V1 ) HomRH (U, V1 H )
y
y
G
HomRG (U G
H , V2 ) HomRH (U, V2 H )
whenever V1 V2 is a homomorphism of RG-modules. In this situation we say that

G
the operation G
H : RH-modules RG-modules is left adjoint to H : RG-modules
RH-modules. (These operations are in fact functors.) We say also that G
H is right
G
adjoint to H , and that this relationship is called an adjunction.
G
We have also seen in Corollary 4.13 that HomRG (V, U G

H ) = HomRH (V H , U ) in the
same circumstances. Again, if this isomorphism can be given naturally in both U and V
(meaning that the corresponding squares commute) then induction G
H is right adjoint to
G
restriction H . It is, in fact, the case for representations of finite groups that induction is
both the left and right adjoint of restriction.
In Proposition 11.13 we will need to know in detail about the natural isomorphisms
that arise in these adjunctions, and we now describe them. They depend on certain
distinguished homomorphisms called the unit and counit of the adjuntions. For each RHmodule U we will define RH-module homomorphisms
G
: U U G
H H
G
: U G
H H U
194
and for each RG-module V we will define RG-module homomorphisms

G
: V V G
H H
G
: V G
H H V.
In the language of category theory, and are the unit and counit of the adjuntion that
G
shows that G
H is left adjoint to H , and , are the unit and counit of the adjuntion that
G
shows that G
H is right adjoint to H .
To define these homomorphisms, choose a set of left coset representatives {g1 , . . . , gn }
Ln
G
of H in G with g1 = 1. For each RH-module U let : U U G
H H =
i=1 gi U be the
G G
inclusion into the summand 1 U , so (u) = 1 u, and let : U H H U be projection
onto this summand. We see that is a monomorphism, is an epimorphim and their
composite is the identity. If V is an RG-module we define the RG-module homomorphisms
Pn
P
: V V G
G
and : V G
G
V by (v) = i=1 gi gi1 v and ( x x u) =
H
H
H
H
P
G G
G G
x xu. In fact, regarding trG
H : HomRH (V, V H H ) HomRG (V, V H H ) we have
G
= trG
H where has domain V regarded as an RH-module. Similarly = trH , and this
shows that and are defined independently of the choice of coset representatives. We
see that is a monomorphism, is an epimorphism and their composite is multiplication
by |G : H|.
We now construct the natural isomorphism
G
HomRG (U G
H , V ) = HomRH (U, V H ).
Given an RG-module homomorphism : U G

H V we obtain an RH-module homomorphsim
G
H
G
G
U U G
H H V H
and given an RH-module homomorphism : U V G

H we obtain an RG-module homomorphism
G
H
G G
U G
H V H H V.
We may check that these two constructions are mutually inverse, and are natural in U and
V.
We next construct a natural isomorphism
G
HomRG (V, U G
H ) = HomRH (V H , U )
whenever U is an RH-module and V is an RG-module. Given an RG-module homomorphism : V U G

H we obtain an RH-module homomorphism
G G
V G
H U H H U
and given an RH-module homomorphism : V G

H U we obtain an RG-module homomorphism
H
G
G
V V G
H H U H .
Again we check that these operations are natural and mutually inverse.
Peter Webb
195
(11.13) PROPOSITION. Let G be a finite group with a subgroup H. The following

are equivalent for an RG-module U .
(1) U is H-projective.
(2) Whenever we have homomorphisms
U
where is an epimorphism and for which there exists an RH-module homomorphism

G
U G
H V H making the diagram commute, then there exists an RG-module homomorphism U V making the diagram commute.
G
(3) Whenever : V U is a homomorphism of RG-modules such that G
H : V H
U G
H is a split epimorphism of RH-modules, then is a split epimorphism of RGmodules.
(4) The surjective homomorphism of RG-modules
G
U G
H H = RG RH U U
x u 7 xu
is split.
G
(5) U is a direct summand of U G
H H .
(6) (Higmans criterion) 1U lies in the image of trG
H : EndRH (U ) EndRG (U ).
Proof. (1) (2) We first prove this implication in the special case when U is an
induced module T G
H . Suppose we have a diagram of RG-modules
T G
H
y
W
G
G
and a homomorphism of RH-modules : T G
H H V H so that = . Under the
G
G
adjoint correspondence corresponds to the composite T T G
H H W H and we
have a commutative triangle of RH-module homomorphisms
V G
H
y .
W G
H
196
By the adjoint correspondence (and its naturality) this corresponds to a commutative

triangle of RG-module homomorphisms
T G
H
()G
H
which proves this implication for the module T G

H.
G
Now consider a module U that is a summand of T G
H , and let U T H U be
G
inclusion and projection. We suppose there is a homomorphism : U G
H V H so that
G
= . The homomorphism : T H W has the property that = () and so by
what we proved there is a homomorphism of RG-modules : T G
H V so that = .
Now = = so that : U V is an RG-module homomorphism that makes the
triangle commute.
(2) (3) This follows immediately on applying (2) to the diagram
1
y .U
U
G
(3) (4) We know that : U G
H H U is split as an RH-module homomorphism
G
by : U U G
H H . Applying condition (3) it splits as an RG-module homomorphism.
(4) (5) and (5) (1) are immediate.
G
(5) (6) We may prove that 1UG
G = trH () by direct computation. Writing
H H
G
U G
H H = V1 V2 where V1 = U , we can represent as a matrix

f11 f21
=
f12 f22
where fij : Vi Vj . Then

trG
H ()
trG
H f11
trG
H f12
trG
H f21
trG
H f22
1 0
0 1
G
and from this we see that for every summand of U G
H H (and in particular for U ) the
G
identity map on that summand is in the image of trH .
G
G
(6) (5) Write 1U = trG
H for some morphism : U H U H . Now corresponds
by the adjoint correspondence to the composite homomorphism
H
G G
G
U U G
H H U H H .
We claim that G
H splits : for
G
G
G
G
G
G
H = trH () H = trH ( H ) = trH () = 1U .
Peter Webb
197
Condition (6) of Proposition 11.13 is in fact equivalent to the surjectivity of trG

H :
is
an
ideal
in
End
(U
),
and
so
EndRH (U ) EndRG (U ) This is because the image of trG
RG
H
it equals EndRG (U ) if and only if it contains 1U . Conditions (2), (3) and (4) are modeled on
conditions associated with the notion of projectivity. There are dual conditions associated
with the notion of an injective module, obtained by reversing the arrows and interchanging
the words epimorphism and monomorphism. These conditions are also equivalent to the
ones listed in this result. In fact, the notion of relative projectivity in the context of group
algebras of finite groups is the same as that of relative injectivity.
(11.14) PROPOSITION. Suppose that H is a subgroup of G and that |G : H| is
invertible in the ring R. Then every RG-module is H-projective.
Proof. For any RG-module U we may write 1U =
(6) of the last result.
1
trG 1 ,
|G:H| H U
thus verifying condition
(11.15) COROLLARY. Suppose that H is a subgroup of G for which |G : H| is

invertible in the ring R, and let U be an RG-module. Then U is projective as an RGmodule if and only if U G
H is projective as an RG-module.
Proof. We already know that if U is projective then U G
H is projective, no matter
G
what subgroup H is. Conversely, if U H is projective it is a summand of a free module
G
RH n . Since U is H-projective it is a summand of U G
H H , which is a summand of
n
RH n G
H = RG . Therefore U is projective.
EXAMPLES. 1. This criterion for projectivity would have simplified matters when we
were considering the projective modules for groups of the form G = H K in Chapter 8. In
this situation we saw that RH becomes an RG-module where H acts by left multiplication
and K acts by conjugation. If K has order prime to p and R is a field of characteristic p (or
a discrete valuation ring with residue field of characteristic p) it follows from the corollary
that RH is projective as an RG-module, because on restriction to H it is projective and
the index of H in G is invertible. On the other hand, we may also regard RK as an
RG-module via the homomorphism G K, and if now H has order prime to p then RK
is a projective RG-module, because it is projective on restriction to RK and the index of
K in G is prime to p.
2. In the exercises to Chapter 6 the simple Fp SL(2, p)-modules were considered.
The goal of Exercise 21 of Chapter 6 was to show that the symmetric powers S r (U2 )
are all simple Fp SL(2, p)-modules when 0 r p 1, where U2 is the 2-dimensional
space on which SL(2, p) acts as invertible transformations of determinant 1. The order
of SL(2, p) is p(p2 1) and so a Sylow p-subgroup of this group is cyclic of order p. In
Exercise 20 of Chapter 6 one shows that on restriction to a certain Sylow p-subgroup,
S r (U2 ) is indecomposable of dimension r + 1 when 0 r p 1. From the classification of
indecomposable modules for a cyclic group of order p we deduce that S p1 (U2 ) is projective
198
as a module for the Sylow p-subgroup. It follows from the last corollary that S p1 (U2 ) is
projective as an Fp SL(2, p)-module. This module is thus a simple projective Fp SL(2, p)module, or in other words a block of defect zero, which therefore lifts to characteristic
zero. Start from information in characteristic p we have thus deduced the existence of an
ordinary simple character of SL(2, p) of degree p.
Representation type of algebras

In trying to understand the representation theory of a ring we might hope, where
possible, to be able to describe all the indecomposable modules, since this would allow us
to construct all modules up to isomorphism by taking direct sums. The possibility of such a
description depends on there being, in some sense, sufficiently few indecomposable modules
for a classification to be a reasonable goal, and for a description of the indecomposable
modules to make any sense or have any utility. Unfortunately for the majority of group
rings we encounter in positive characteristic both a classification of the indecomposable
modules and a way of organizing the classification so that it can be understood seem
to be unreasonable expectations. On the other hand, there are special cases where the
indecomposable modules can indeed be classified.
We say that a ring A has finite representation type if and only if there are only finitely
many isomorphism classes of indecomposable A-modules, and otherwise we say that A has
infinite representation type. We have seen in Theorem 6.2 that if G is a cyclic p-group and
k is a field of characteristic p, then kG has finite representation type. Our immediate goal
in this section is to characterize the groups for which kG has finite representation type.
We first reduce it to a question about p-groups.
(11.16) PROPOSITION. Let R be a discrete valuation ring with residue field of
characteristic p or a field of characteristic p, and let P be a Sylow p-subgroup of a finite
group G. Then RG has finite representation type if and only if RP has finite representation
type.
Proof. Since |G : P | is invertible in R, by Proposition 11.14 every indecomposable RGmodule is a summand of some module T G
P , and we may assume that T is indecomposable,
G
G
since if T = T1 T2 then T P = T1 P T2 G
P , and by the Krull-Schmidt theorem the
G
indecomposable summands of T P are the indecomposable summands of T1 G
P together
G
with the indecomposable summands of T2 P . If RP has finite representation type then
there are only finitely many modules T G
P with T indecomposable, and these have only
finitely many isomorphism types of summands by the Krull-Schmidt theorem. Hence RG
has finite representation type.
G
Conversely, every RP -module U is a direct summand of U G
P P and hence is a
direct summand of some module V G
P . If U is indecomposable, we may assume V is
Peter Webb
199
indecomposable. Now if RG has finite representation type there are only finitely many
isomorphism types of summands of modules V G
P , by the Krull-Schmidt theorem, and
hence RP has finite representation type.
We have already seen in Theorem 6.2 that cyclic p-groups have finite representation
type over a field of characteristic p, so by Proposition 11.16 groups with cyclic Sylow psubgroups have finite representation type. We will show the converse in Theorem 11.18,
and as preparation for this we now show that k[Cp Cp ] has infinitely many non-isomorphic
indecomposable modules, where k is a field of characteristic p. Apart from the use of this
in establishing infinite representation type, it is useful to see how indecomposable modules
may be constructed, and in the case of C2 C2 it will lead to a classification of the
indecomposable modules.
We will first describe infinitely many modules of different dimensions for k[Cp Cp ],
and after that we will prove that they are indecomposable. Let G = Cp Cp = haihbi. For
each n 1 we define a module M2n+1 of dimension 2n + 1 with basis u1 , . . . , un , v0 , . . . , vn
and an action of G given as follows:
a(ui ) = ui + vi1 , b(ui ) = ui + vi
a(vi ) = vi ,
b(vi ) = vi
where 1 i n
where 0 i n.
It is perhaps easier to write this as

(a 1)ui = vi1 , (b 1)ui = vi
(a 1)vi = 0,
(b 1)vi = 0
where 1 i n
where 0 i n
and to describe M2n+1 diagrammatically:
a1
v0
u1
b1
a1
v1
u2
b1
a1
un
b1
vn
We may check that this is indeed a representation of G by verifying that

(a 1)(b 1)x = (b 1)(a 1)x = 0
and
(a 1)p x = (b 1)p x = 0
for all x M2n+1 , which is immediate. This is sufficient to know that we have a representation of G since the equations (a 1)(b 1) = (b 1)(a 1) and (a 1)p = 0 = (b 1)p
are defining relations for kG as a k-algebra with generators a, b.
We now show that M2n+1 is indecomposable.
200
(11.17) PROPOSITION. The quotient EndkG (M2n+1 )/ Rad EndkG (M2n+1 ) has dimension 1. Thus EndkG (M2n+1 ) is a local ring and M2n+1 is indecomposable.
Proof. We will show that EndkG (M2n+1 )/I has dimension 1 for a certain nilpotent
ideal I. Such an ideal I must be contained in the radical, being nilpotent, and the fact
that it has codimension 1 will then force it to equal the radical. This will prove the result.
Observe that
Soc(M2n+1 ) = Rad(M2n+1 ) = kv0 + kvn .
The ideal I in question is HomkG (M2n+1 , Soc(M2n+1 )), and this squares to zero since if
: M2n+1 Soc(M2n+1 ) then Rad(M2n+1 ) Ker and so Soc(M2n+1 ) = 0.
We now show that I has codimension 1. It is easy to get an intuitive idea of why this
is so, but not so easy to write it down in technically correct language. The intuitive idea is
that the basis elements of M2n+1 lie in a string as shown diagrammatically, each element
related to those on either side by the action of a 1 and b 1. To within elements of A,
any endomorphism must send this string to another string of elements that are similarly
related. If the endomorphism shifts the string either to the left or the right, then part
of the string must fall off the end of the module, or in other words be sent to zero. The
connection between adjacent basis elements then forces the whole shift to be zero, so that
no shift to the left or right is possible. From this we deduce that, modulo I, endomorphisms
are scalar multiples of the identity.
We now attempt to write down this intuitive idea in formal terms. If is any endomorphism of M2n+1 then (Soc(M2n+1 )) Soc M2n+1 so induces an endomorphism
of M2n+1 / Soc(M2n+1 ). We show that is necessarily a scalar multiple of the identity. To
establish this we will exploit the equations
(a 1)ui = (b 1)ui1
when 2 i n
and also the fact that (a 1) and (b 1) both map ku1 + + kun injectively into
Soc(M2n+1 ). Applying to the last equations we have for 2 i n,
((a 1)ui ) = (a 1)(ui ) = ((b 1)ui1 ) = (b 1)(ui1 ).
It follows from this that is completely determined once we know (u1 ), since then
2 ) by injectivity of a 1 on the span of the ui ,
(a 1)(u2 ) = (b 1)(u1 ) determines (u
3 ) similarly, and so on.
(a 1)(u3 ) = (b 1)(u2 ) determines (u
Suppose that
(u1 ) 1 u1 + + r ur (mod Soc(M2n+1 ))
where the i are scalars and r 6= 0. Multiplying both sides by b1 and using the equations
(a 1)(ui ) = (b 1)(ui1 ) as before, as well as injectivity of multiplication by a 1 on
the span of the ui , we see that
(u2 ) 1 u2 + + r ur+1 (mod Soc(M2n+1 ))
Peter Webb
201
and inductively
(unr+1 ) 1 unr+1 + + r un (mod Soc(M2n+1 )).
If it were the case that r > 1 then the equation
(b 1)(unr+1 ) = (a 1)(unr+2 ) = 1 vnr+1 + + r vn
would have no solution, since no such vector where the coefficient of vn is non-zero lies in
the image of a 1.
We conclude that r = 1 and (ui ) 1 ui (mod Soc(M2n+1 )) for some scalar 1 , for
all i with 1 i n. Thus
1 1M2n+1 : M2n+1 Soc(M2n+1 )
and so EndkG (M2n+1 )/ HomkG (M2n+1 , Soc(M2n+1 )) has dimension 1.
We can now establish the following:
(11.18) THEOREM (D.G. Higman). Let k be a field of characteristic p. Then kG
has finite representation type if and only if Sylow p-subgroups of G are cyclic.
Proof. By Proposition 11.16 it suffices to show that if P is a p-group then kP has
finite representation type if and only if P is cyclic. We have seen in Theorem 6.2 that kP
has finite representation type when P is cyclic. If P is not cyclic then P has the group
Cp Cp as a homomorphic image. (This may be proved using the fact that if (P ) is
the Frattini subgroup of P then P/(P )
= (Cp )d for some d and that P can be generated
by d elements. Since P cannot be generated by a single element, d 2 and so (Cp )2 is
an image of P .) The infinitely-many non-isomorphic indecomposable k[Cp Cp ]-modules
become non-isomorphic indecomposable kP -modules via the quotient homomorphism, and
this establishes the result.
Even when the representation type is infinite, the arguments that we have been using
still yield the following result.
(11.19) THEOREM. Let k be a field of characteristic p. For any finite group G, the
number of isomorphism classes of indecomposable modules that are projective relative to
a cyclic subgroup is finite.
We now turn to group rings of infinite representation type, namely the group rings
in characteristic p for which the Sylow p-subgroups of the group are not cyclic. We might
expect that, even though the technical difficulties may be severe, a classification of indecomposable modules is about to be revealed. Perhaps the surprising thing about infinite
202
representation type is that in some sense, most of the time, a classification of indecomposable modules is not merely something that is beyond our technical capabilities, it is rather
something that may never be possible in any meaningful sense, because of inherent aspects
of the problem.
Discussion of such matters raises the question of what we mean by a classification.
There are many instances of classification in mathematics, but whichever one we are considering, we might reasonably understand that it is a description of some objects that is
simpler than the objects themselves and that allows us to identify in some reasonable way
some significant aspects. Thus it would be inadequate to say that we had a parametrized
indecomposable modules by using the same set of indecomposable modules to achieve the
classification, because this would provide no simplification. It would also be inadequate
to put the set of isomorphism classes of indecomposable modules in bijection with some
abstract set of the same cardinality. Although we could certainly do this, and for some
people it might be a classification, it would provide no new insight. Those who maintain
that it is always possible to classify perhaps have such possibilities in mind. In reality it
might be very difficult or impossible to classify in any meaningful sense because there are
simply too many objects to classify: if we were to put the objects in a list in a book, the list
would simply be too long and structureless to have any meaning. Of course, just because
we cannot see how to make sense of a set of objects does not mean it cannot be done.
However, it remains the case that for most group algebras in positive characteristic no
one has been able to provide any reasonable classification of the indecomposable modules.
Furthermore, we will present reasons why it would be hard to do so.
Infinite representation type divides up into two possibilities: tame and wild. For
the tame group algebras we can (in principle) classify the indecomposable modules. For
the wild algebras no one can see how to do it. Before we address these general questions we will describe the indecomposable representations of C2 C2 in over a field k
of characteristic 2, since this exemplifies tame representation type. In constructing infinitely many indecomposable modules for Cp Cp we already constructed some of the
indecomposable k[C2 C2 ]-modules, but now we complete the picture. As before we let
G = C2 C2 = hai hbi and we exhibit the modules diagrammatically by the action of

a 1 and b 1 on a basis. Here are the indecomposable modules:
Peter Webb
kG =
W2n+1 =
W1 = M1 = ()
M2n+1 =
Ef,n =
E0,n =
E,n
203
and for n 1
In these diagrams each node represents a basis element of a vector space, a southwest
arrow emanating from a node indicates that a 1 sends that basis element to the basis
element at its tip, and similarly a southeast arrow indicates the action of b 1 on a
basis element. Where no arrow in some direction emanates from a node, the corresponding
element a 1 or b 1 acts as zero.
The even-dimensional indecomposable representations Ef,n require some further explanation. They are parametrized by pairs (f, n) where f k[X] is an irreducible monic
polynomial and n 1 is an integer. Let the top row of nodes in the diagram correspond
to basis elements u1 , . . . , un , and the bottom row to basis elements v1 , . . . , vn . Let
(f (X))n = X mn + amn1 X mn1 + + a0 .
The right-most arrow starting at umn that has no terminal node is supposed to indicate
that (b 1)umn = vmn+1 where
vmn+1 = amn1 vmn1 a1 v2 a0 v1 ,
so that with respect to the given bases b 1 has matrix
0
1
0
..
.. ..
...
.
.
.
.
0
0
1
a0
amn1
204
The significance of this matrix is that it is an indecomposable matrix in rational canonical

form with characteristic polynomial f n .
(11.20) THEOREM. Let k be a field of characteristic 2. The k[C2 C2 ]-modules
shown are a complete list of indecomposable modules.
Proof. We describe only the strategy of the proof, and refer to the exercises at the end
of this section and [Ben] for more details. The first step is to use the fact that the regular
P
representation is injective, with simple socle spanned by gG g. If U is an indecomposable
P
P
module for which ( gG g)U 6= 0 then there is a vector u U with ( gG g)u 6= 0. The
homomorphism kG U specified by x 7 xu is a monomorphism, since if its kernel were
P
non-zero it would contain gG g, but this element does not lie in the kernel. Since kG
is injective, the submodule kGu is a direct summand of U , and hence U
= kG since U is
indecomposable. From this we deduce that apart from the regular representation, every
P
indecomposable module is annihilated by
gG g, and hence is a module for the ring
P
kG/( gG g), which has dimension 3 and is isomorphic to k[, ]/(2 , , 2 ), where
corresponds to a 1 kG and to b 1 kG.
Representations of this ring are the same thing as the specification of a vector space
U with a pair of linear endomorphisms , : U U that annihilate each other and square
to zero. The classification of such pairs of matrices up to simultaneous conjugacy of the
matrices (which is the same as isomorphism of the module) was achieved by Kronecker in
the 19th century, and he obtained the indecomposable forms that we have listed.
The modules M2n+1 and W2n+1 have become known as string modules and the Ef,n
as band modules, in view of the form taken by the diagrams that describe them. More
complicated classifications, but along similar lines, have been achieved for representations
of the dihedral, semidihedral and generalized quaternion 2-groups in characteristic 2. For
dihedral 2-groups, all the modules apart from the regular representation are string modules
or band modules (see [RinD2])).
Provided the field k is infinite, k[C2 C2 ] has infinitely many isomorphism types
of indecomposable modules in each dimension larger than 1. They can nevertheless be
grouped into finitely many families, as we have seen intuitively in their diagrammatic
description. As a more precise version of this idea, consider the infinite-dimensional k[C2
C2 ]-module M with diagram
M=
and basis u1 , u2 , . . . , u1 , v2 , . . . This module has an endomorphism that shifts each of

the two rows one place to the right, specified by (ui ) = ui+1 and (vi ) = vi+1 , so that
M becomes a (k[C2 C2 ], k[X])-bimodule, where the indeterminate X acts via . As
k[X]-modules we have M
= k[X] k[X]. Given an irreducible polynomial f k[X] and
Peter Webb
205
an integer n 1 we may construct the k[C2 C2 ]-module M k[X] k[X]/(f n), which is a
module isomorphic to (k[X]/(f n))2 as a k[X]-module, and which is acted on by k[C2 C2 ]
as a module isomorphic to Ef, n. This construction accounts for all but finitely many of
the indecomposable k[C2 C2 ]-modules in each dimension.
With our understanding improved by the last example, we now divide infinite representation type into two kinds: tame and wild. Let A be a finite-dimensional algebra over
an infinite field k. We say A has tame representation type if it has infinite type and for each
dimension d there are finitely many (A, k[X])-bimodules Mi that are free as k[X]-modules
so that all but finitely many of the indecomposable A-modules of dimension d have the
form Mi k[X] k[X]/(f n) for some irreducible polynomial f and integer n. If the bimodules
Mi can be chosen independently of d (as happens with representations of C2 C2 ) we say
that A has domestic representation type, and otherwise it is non-domestic.
We say that the finite-dimensional algebra A has wild representation type if there
is a finitely-generated (A, khX, Y i)-bimodule M that is free as a right khX, Y i)-module,
such that the functor M khX,Y i from finite-dimensional khX, Y i-modules to finitedimensional A-modules preserves indecomposability and isomorphism type. Here khX, Y i
is the free algebra on two non-commuting variables, having as basis the non-commutative
monomials X a1 Y b1 X a2 Y b2 where ai , bi 0.
In view of the following theorem it would have been possible, over an algebraically
closed field, to define wild to be everything that is not finite or tame. We state the next
three results without proof, since they take us outside the scope of this book.
(11.21) THEOREM (Drozd [Dro]; Crawley-Boevey [CB]). Let A be a finite-dimensional
algebra over an algebraically-closed field. Then A has either finite, tame or wild representation type.
When A has wild representation type, the idea is that phenomena that occur with
representations of khX, Y i also appear with A-modules. Thus A has at least as many isomorphism types indecomposable modules as khX, Y i does. The indecomposable khX, Y imodules are quite diverse, and in some sense it is a hopeless task to try to classify them.
It is known that the theory of finite-dimensional khX, Y i-modules is undecidable, meaning that there exists a sentence in the language of finite-dimensional khX, Y i-modules
that cannot be decided by any Turing machine (an account of this result can be found in
[Prest]). It is also the case in a heuristic sense that the khX, Y i-modules are as badly behaved as those of any algebra: it is possible to embed the category of A-modules and their
homomorphisms, for any algebra A, into the category of khX, Y i-modules (see [Brenner,
Gabriel]).
In the context of group algebras the division into finite, tame and wild representation
type is given by the following theorem, in which we include the result of D.G. Higman
already proven.
206
(11.22) THEOREM (Bondarenko, Brenner, Drozd, Higman, Ringel). Let k be an

infinite field of characteristic p and let G be a finite group with Sylow p-subgroup P . Then
kG has finite representation type if and only if P is cyclic, and tame representation type
if and only if p = 2 and P is dihedral, semidihedral or generalized quaternion. In all other
cases kG has wild representation type.
We include C2 C2 as a dihedral group in this theorem. The first step in the proof
of this theorem we have already seen, and it is to identify the group algebras of finite
representation type. Next, certain group algebras were established as being wild. This is
implied by the following result, which also serves as an example of the kind of phenomenon
we may expect with wild algebras.
(11.23) THEOREM (Brenner [Bre1]). Let P be a finite p-group having either Cp Cp
(p odd), C2 C4 or C2 C2 C2 as a homomorphic image, let k be a field of characteristic p,
and let E be any finite-dimensional algebra over k. Then there exists a finite-dimensional
kP -module M such that EndkP (M ) has a nilpotent ideal J and a subalgebra E isomorphic
to E, with the property that the quotient map sends E isomorphically to EndkP (M )/J.
A theorem of Blackburn implies that if P is a 2-group that is not cyclic and does not
have C2 C4 or C2 C2 C2 as a homomorphic image, then P is dihedral, semidihedral
or generalized quaternion. Groups with these as Sylow 2-subgroups were the only groups
whose representation type was in question at this point. The representation type in these
cases has been decided by classifying explicitly the indecomposable modules, and it was
done by Bondarenko, Drozd and Ringel. A later approach can be found in the work of
Crawley-Boevey [CB].
Peter Webb
207
Vertices, sources and Green correspondence

Having just given an impression of the difficulty of classifying indecomposable modules
in general, we now explain some positive techniques that are available to understand them
better.
(11.24) THEOREM. Let R be a field or a complete discrete valuation ring, and let
U be an indecomposable RG-module.
(1) There is a unique conjugacy class of subgroups Q of G that are minimal subject to
the property that U is Q-projective.
(2) Let Q be a minimal subgroup of G such that U is Q-projective. There is an indecomposable RQ-module T that is unique up to conjugacy by elements of NG (Q) such
G
that U is a summand of T G
Q . Such a T is necessarily a summand of U Q .
Proof. (1) We offer two proofs of this result, one employing module-theoretic techniques, and the other a ring-theoretic approach. Both proofs exploit similar ideas, in which
the Mackey formula is a key ingredient.
First proof: we start by supposing that U is both H-projective and K-projective where
G
G G
H and K are subgroups of G. Then U is a summand of U G
H H and also of U K K , so
it is also a summand of
M
G G G
H
K
G
U G
( g((U G
H H K K =
H ) K g H )) K gH ) K
g[K\G/H]
G
( g(U G
K g H )) K gH
g[K\G/H]
using transitivity of restriction and induction, and now U must be a summand of some
module induced from one of the groups K gH. If both H and K happen to be minimal
subject to the condition that U is projective relative to these groups, we deduce that
K gH = K, so K gH. Similarly H g K for some g and so H and K are conjugate.

Second proof: we start the same way and suppose that U is both H-projective and
G
K-projective. We may write 1U = trG
H = trK for certain EndRH (U ) and
EndRK (U ). Now
G
G
G
G
G
1U = (trG
H )(trK ) = trK ((trH )) = trK (trH ()) =
trG
K gH (cg ).
g[K\G/H]
Since U is indecomposable its endomorphism ring is local and so some term trG
K gH (cg )
must lie outside the unique maximal ideal of EndRG (U ) and must be an automorphism.
This implies that trG
K gH : EndR[K gH] (U ) EndRG (U ) is surjective, since the image of
G
trK gH is an ideal, and so U is K gH-projective.
208
We now deduce as in the first proof that if K and H are minimal subgroups relative
to which U is projective, then H and K are conjugate.
(2) Let Q be a minimal subgroup relative to which U is projective. We know that U
G
G
is a summand of U G
Q Q and hence it is a summand of T Q for some indecomposable
summand T of U G
Q . Suppose that T is another indecomposable module for which U is a
L
Q
G G
g
g
summand of T G
Q . Now T is a summand of T Q Q =
g[Q\G/Q] ( (T Q Q )) Q gQ
and hence a summand of some ( g(T Qg Q )) Q

Q gQ . For this element g we deduce that
g
U is Q Q-projective and by minimality of Q we have Q = Q gQ and g NG (Q). Now
T is a summand of gT , and since both modules are indecomposable we have T = gT . We
g 1
deduce from the fact that T is a summand of U G
T must be a summand
Q that T =
g 1
G
G
g 1
of ( U ) and hence of U , since

U = U as RG-modules.
Q
A minimal subgroup Q of G relative to which the indecomposable module U is projective is called a vertex of U , and it is defined up to conjugacy in G. We write vtx(U ) to
denote a subgroup Q that is a vertex of U . An RQ-module T for which U is a summand
of T G
Q is called a source of U , and given the vertex Q it is defined up to conjugacy by
elements of NG (Q).
We record some immediate properties of the vertex of a module.
(11.25) PROPOSITION. Let R be a field of characteristic p or a complete discrete
valuation ring with residue field of characteristic p.
(1) The vertex of every indecomposable RG-module is a p-group.
(2) An indecomposable RG-module is projective if and only if it is free as an R-module
and its vertex is 1.
(3) A vertex of the trivial RG-module R is a Sylow p-subgroup of G.
Proof. (1) We know from Proposition 11.14 that every module is projective relative
to a Sylow p-subgroup, and so vertices must be p-groups.
(2) If an indecomposable module is projective it is a summand of RG, which is induced
from 1, so it must be free as an R-module and have vertex 1. Conversely, if U has vertex 1
G
G
it is a summand of U G
1 1 , so if U is free as an R-module it is a summand of R 1 = RG
and hence is projective.
(3) Let Q be a vertex of R and P a Sylow p-subgroup of G containing Q. Then R is
L
G
G G
P
a summand of R G
Q , so R P is a summand of R Q P =
g[P \G/Q] R P gQ and hence
is a summand of R P
P gQ for some g G. We claim that for every subgroup H P ,
P
R H is an indecomposable RP -module. From this it will follow that R = R P
P gQ and
that Q = P . The only simple RP -module is the residue field k with the trivial action
(in case R is a field already, R = k), and HomRP (R P

H , k) = HomRH (R, k) = k is a
space of dimension 1. This means that R P
H has a unique simple quotient, and hence is
indecomposable.
Peter Webb
209
The vertices of indecomposable modules partition the collection of all indecomposable

RG-modules into subclasses, namely, for each p-subgroup H the modules with vertex H.
The class of modules with vertex 1 that are free over R consists of the projective modules,
and here we have identified the isomorphism types of the individual modules. Identifying
the isomorphism types of modules with cyclic vertex is also possible in many instances.
Aside from this, every other class contains infinitely many isomorphism types, and they
may be beyond our capabilities to classify.
We now lead in to Green correspondence. This allows us to reduce many questions
about indecomposable modules to a situation where the vertex of the module is a normal
subgroup. The theory will be used when we come to consider blocks in Chapter 12, and it
is very helpful in many other situations. The philosophy that many questions in modular
representation theory of finite groups are determined by normalizers of p-subgroups has
been one of the most important themes in this area.
Green correspondence gives a bijection (denoted f in what follows, with inverse g)
from isomorphism types of RG-modules with vertex a given p-subgroup Q to isomorphism
types of RNG (Q)-modules with vertex Q. It also gives an isomorphism of certain groups
of homomorphisms, but we omit this part of the statement.
(11.26) THEOREM (Green correspondence). Let R be a field of characteristic p
or a complete discrete valuation ring with residue field of characteristic p. Let Q be a
p-subgroup of G and L a subgroup of G that contains the normalizer NG (Q).
(1) Let U be an indecomposable RG-module with vertex Q. Then in any decomposition
of U G
L as a direct sum of indecomposable modules there is a unique indecomposable
summand f (U ) with vertex Q. Writing U G
L = f (U ) X, each summand of X is
projective relative to a subgroup of the form L xQ where x G L.
(2) Let V be an indecomposable RL-module with vertex Q. Then in any decomposition
of V G
L as a direct sum of indecomposable modules there is a unique indecomposable
summand g(V ) with vertex Q. Writing V G
L = g(V ) Y , each summand of Y is
projective relative to a subgroup of the form Q xQ where x G L.
(3) In the notation of parts (1) and (2) we have gf (U )
= U and f g(V )
= V.
As a preliminary to the proof, notice that if x GL then Q x Q is a strictly smaller
group than Q, since NG (Q) L so x does not normalize Q. On the other hand, L x Q
might be a group of the same size as Q, and in that case it is conjugate to Q in G (by
the element x). However, it will be important in step 1 of the proof to know that L x Q
cannot be conjugate to Q in L. The argument is that if L x Q = z Q for some z L then
z 1 x NG (Q) L so x zL = L, which contradicts x G L.
Proof. We will prove (2) before (1). Let V be an indecomposable RL-module with
vertex Q.
210
Step 1. We show that in any decomposition as a direct sum of indecomposable RLG

modules, V G
L L has a unique summand with vertex Q, the other summands being projective relative to subgroups of the form L xQ with x 6 L. To show this, let T be a
source for V , so that T L

Q = V Z for some RL-module Z. Put
G
V G
L L = V V ,
G
Z G
L L = Z Z
for certain RL-modules V and Z . Then

G
G G
G G
T G
Q L = V L L Z L L
= V V Z Z
M
L
=
( x(T Q
Lx Q )) L xQ .
x[L\G/Q]
There is one summand in the last direct sum with x L and it is isomorphic to T L
Q=
x
V Z. The remaining summands are all induced from subgroups L Q with x 6 L, and
it follows that all indecomposable summands of V and Z are projective relative to these
subgroups. This in particular implies the assertion we have to prove in this step, since
such subgroups cannot be conjugate to Q by elements of L.
Step 2. We show that in any decomposition as a direct sum of indecomposable modules, V G
L has a unique indecomposable summand with vertex Q and that the remaining
summands are projective relative to subgroups of the form Q xQ where x 6 L. To show
this, write V G
L as a direct sum of indecomposable modules and pick an indecomposable
summand U for which U G
L has V as a summand. This summand U must have vertex Q;
for it is projective relative to Q, since V is, and if U were projective relative to a smaller
group then V would be also, contradicting the fact that Q is a vertex of V . This shows
that the direct sum decomposition of V G
L has at least one summand with vertex Q.
G
Let U be another summand of V L . Then U G
L must be a summand of V , in the
notation of Step 1, and every indecomposable summand of U G
L is projective relative to a
y
G
subgroup L Q with y 6 L. Since U is a summand of T Q it is projective relative to Q,
and hence has a vertex Q that is a subgroup of Q. Since L Q it follows that U G
L has
an indecomposable summand that on restriction to Q has a source of U as a summand,

and so Q is a vertex of this summand. It follows that some L-conjugate of Q must be
contained in one of the subgroups L yQ with y 6 L. In other words zQ L yQ for
1
some z L. Thus Q z yQ where x = z 1 y 6 L. This shows that Q Q xQ and
completes the proof of assertion (2) of this theorem.
Step 3. We establish assertion (1). Suppose that U is an indecomposable RG-module
with vertex Q. Letting T be a source of U , there is an indecomposable summand V of T L
Q
G
L
G
)
=
(T
.
This
is
because
U
is
a
summand
of
T
for which U is a summand of V G

L.
Q
Q
L
This RG-module V must have vertex Q, since it is projective relative to Q, and if it
Peter Webb
211
were projective relative to a smaller subgroup then so would U be. Now U G

L is a direct
G
summand of V G
L L , and by Step 1 this has just one direct summand with vertex Q,
namely V . In fact U G
L must have an indecomposable summand that on further restriction
to Q has T as a summand, and this summand has vertex Q. It follows that this summand
must be isomorphic to V , and in any expression for U G
L as a direct sum of indecomposable
modules, one summand is isomorphic to V and the rest are projective relative to subgroups
of the form L xQ with x 6 L. This completes the proof of assertion (1) of the theorem.
Step 4. The final assertion of the theorem follows from the first two and the fact that
G
G G
U is isomorphic to a summand of U G
L L and V is isomorphic to a summand of V L L .
The use of Green correspondence is to allow us to understand the indecomposable

modules for G in terms of the indecomposable modules for a subgroup of the form NG (Q)
where Q is a p-group. The easiest situation where we may apply this result is when
the characteristic of k is p and Q is a Sylow p-subgroup of G of order p. The detailed
structure of the modules in this situation depends on the Brauer tree of the block to
which they belong, which is outside the scope of this book. We can, however, use Green
correspondence together with the information we already have about groups with a normal
Sylow p-subgroup to say how many indecomposable modules there are.
We will write lk (G) for the number of isomorphism classes of simple kG-modules,
and by Theorem 7.14 this equals the number of isomorphism classes of indecomposable
projective kG-modules. We will consider lk (NG (Q)) = lk (Q K) where K is a group of
order prime to p, and this equals lk (K) by Corollary 6.4.
(11.27) COROLLARY. Let k be a field of characteristic p and let G be a group with
a Sylow p-subgroup Q of order p. Then the number of indecomposable kG-modules is
(p 1)lk (NG (Q)) + lk (G).
Proof. By the Schur-Zassenhaus theorem we may write NG (Q) = Q K for some
subgroup K of order prime to p, and lk (NG (Q)) = lk K. By Corollary 11.9 there are
plk (NG (Q)) indecomposable kNG (Q)-modules and lk (NG (Q)) of these are projective, so
(p 1)lk (NG (Q)) indecomposable modules have vertex Q. By Green correspondence this
equals the number of indecomposable kG-modules with vertex Q. The remaining indecom-
posable kG-modules are projective, and there are lk (G) of them, giving (p1)lk (NG (Q)) +
lk (G) indecomposable modules in total.
212
EXAMPLES. When G = S4 and p = 3 we determined the simple and projective

F3 S4 -modules in Chapter 10 and there four of them. If Q is a Sylow 3-subgroup then
NS4 (Q)
= S3 , which has two simple modules in characteristic 3. Therefore the number of
indecomposable F3 S4 -modules is (3 1)2 + 4 = 8. However, examination of the Cartan
matrix of F3 S4 shows that it is a direct sum of two blocks of defect zero and a Nakayama
algebra isomorphic to F3 S3 , so we did not need to invoke Green correspondence to obtain
this result.
2
When p isodd the
group SL(2, p) has order p(p 1)(p 1) and has the subgroup
1
Q of matrices
with Fp as a Sylow p-subgroup with normalizer the matrices
0
1

with , Fp and 6= 0. Thus N (Q)

= Cp Cp1 and l(N (Q)) = p 1.
0 1
We have seen in Chapter 6 Exercise 21 that SL(2, p) has at least p-simple modules in
characteristic p, and in fact this is the complete list. It follows that Fp SL(2, p) has (p2
1) + p = p2 p + 1 indecomposable modules over any field of characteristic p.
The case of P SL(2, p) = SL(2, p)/{I} when p is odd is similar. Now N (Q) =
Cp C(p1)/2 has (p 1)/2 simple modules, so the number of indecomposable modules is
(p 1)2
p+1
p2 p + 2
+
=
.
2
2
2
This applies, for instance to the alternating group A5
= P SL(2, 5) in characteristic 5 which
52 5+2
= 11 indecomposable modules.
has
2
Further calculations with these modules will be found in the exercises.
We present in 11.29 a refinement of one part of Green correspondence known as
the BurryCarlsonPuig theorem whose proof illustrates a number of techniques that are
regularly used in this theory. The approach is ring-theoretic and it is interesting to compare
this with the module theoretic approach of our proof of the Green correspondence. Before
that we single out a result that we will use a number of times.
(11.28) LEMMA.

(1) Suppose that B is a local ring and that {Ij j J} is a family of ideals of B. If
P
1 jJ Ij then 1 Ij for some j, so that Ij = B.
(2) (Rosenbergs Lemma) Suppose that B is an R-algebra that is finitely generated as
an R-module, where R is a complete discrete valuation
ring or a field. Let e be a

primitive idempotent in B and suppose that {Ij j J} is a family of ideals of B. If
P
e jJ Ij then e Ij for some j.
Proof. (1) If 1 6 Ij for all j then every Ij is contained in the unique maximal ideal of
P
B and so 1 6 jJ Ij since 1 does not lie in the maximal ideal.
(2) The only idempotents in eBe are 0 and e since e is primitive. By Proposition 11.4
since eBe is finitely generated as an R-module, it is a local ring. It contains the family of
ideals {eIj e j J} and e lies in their sum. Therefore by part (1), e lies in one of the
ideals.
Peter Webb
213
(11.29) THEOREM (BurryCarlsonPuig). Let R be a field of characteristic p or

a complete discrete valuation ring with residue field of characteristic p. Let V be an
indecomposable RG-module, Q a p-subgroup of G, and H a subgroup of G containing
NG (Q). Suppose V G
H = M N where M is indecomposable with vertex Q. Then V has
vertex Q, from which it follows that M is the Green correspondent of V .
Proof. We only need to show that V has vertex Q because the conclusion about Green
correspondence then follows from Theorem 11.26. We will use Higmans criterion to verify
this, as well as the Mackey formula, the fact that the image of the relative trace map
between endomorphism rings is an ideal, and Rosenbergs lemma.
For each subgroup K of G we will write EK (V ) for the endomorphism ring HomRK (V, V ),
because this simplifies the notation. For each subgroup K of H the subset trH
K EK (V ) is
an ideal of EH (V ). We will put
J :=
trH
Hg Q Eg Q (V ),
g6H
and this is an ideal of EH (V ).

We show that for all EQ (V ),
G
H
resG
H trQ () trQ () (mod J).
This is a consequence of the Mackey formula, for

G
resG
H trQ () =
Q
trH
Hg Q resHg Q cg ()
g[H\G/Q]
and all of the terms in the sum lie in J except for the one represented by 1 which gives
trH
Q ().
Let e EH (V ) be the idempotent that is projection onto M . We claim that e
H
trQ EQ (V ) and also that e 6 J. These come about because Q is a vertex of M and e is
the identity in the ring eEH (V )e, which may be identified with the local ring EH (M ).
H
Thus e trH
Q EQ (M ) trQ EQ (V ) since M is projective relative to Q. If e were to lie in
J, which is a sum of ideals trH
Hg Q Eg Q (V ) with g 6 H, it would lie in one of them by
Rosenbergs Lemma, and hence M would be projective relative to H g Q for some g 6 H.
The vertex Q of M must be conjugate in H to a subgroup of H g Q, so that g Q = h Q for
some h H. This implies that g 1 h NG (Q) H, so g H, a contradiction. Therefore
e 6 J.
resG
H
Let be the composite ring homomorphism EG (V )E
H (V ) EH (V )/J where the
second morphism is the quotient map. Since it is the image of a local ring, (EG (V )) is
also a local ring. It contains the ideal (trG
Q EQ (V )). We claim that this ideal contains the
214
non-zero idempotent e + J. This is because we can write e = trH

Q () for some EQ (V )
so that
G H
G
H
trG
H (e) = trH trQ () = trQ () trQ () = e
(mod J)
by an earlier calculation.
Any ideal of a local ring containing a non-zero idempotent must be the whole ring, so
G
that (trG
Q EQ (V )) = (EG (V )). Again since EG (V ) is local, trQ EQ (V ) = EG (V ).
We have just shown that V is projective relative to Q, so Q contains a vertex of V .
The vertex cannot be smaller than Q, because otherwise by a calculation using the Mackey
formula M would have vertex smaller than Q. We conclude that Q is a vertex of V .
The Heller operator

The Heller or syzygy operator provides a way to construct new indecomposable
module from old ones, obtaining sequences of modules i M that are closely related in a
sense to do with homological algebra. It plays a role as the inverse of the shift operator in
a triangulated structure on the stable module category for a group algebra. These topics
go beyond the scope of this text, but all the same it is important to know about the Heller
operator simply as a useful tool in handling indecomposable modules.
Suppose that R is either a field or a complete discrete valuation ring. Given an RGmodule M we define M to be the kernel of the projective cover PM M , so that
there is a short exact sequence 0 M PM M 0. Since projective covers are
unique up to isomorphism of the diagram, M is well-defined up to isomorphism. We
can immediately see that M = 0 if and only if M is projective, so that the operator
is not invertible on all modules. On the other hand if we exclude certain modules from
consideration then we do find that is invertible. In case R is a field, all we need to do
is exclude the projective modules, which are also injective. We may define for any module
N a module 1 N to be the cokernel of the injective hull of N , so that there is a short
exact sequence 0 N IN 1 N 0. In the other situation where R is a complete
discrete valuation ring we restrict attention to RG-lattices. In this situation there is always
a relative injective hull N IN for any RG-lattice N that may be constructed as the
dual of the projective cover PN N , identifying N with its double dual and IN with
(PN ) . Again we define 1 N to be the cokernel, so that there is a short exact sequence
0 N IN 1 N 0. Note that since (RG)
= RG as RG-modules, the module
IN is in fact projective, as well as having an injective property relative to RG-lattices.
We state the next result for RG-lattices. If R happens to be a field, an RG-lattice is
the same as a finitely generated RG module.
Peter Webb
215
(11.30) PROPOSITION. Let R be a field or a complete discrete valuation ring and

G a finite group.
(1) Let 0 U V W 0 be a short exact sequence of RG-lattices. If W has no
non-zero projective summand and V W is a projective cover then U V is a
(relative) injective hull. If U has no non-zero projective summand and U V is a
(relative) injective hull then V W is a projective cover.
(2) For any RG-lattice M we have (1 M )
= (M ).
(3) If M is an RG-lattice with no non-zero projective summands then
1 (M )
=M
= 1 (M ).
(4) When M1 and M2 are RG-lattices we have (M1 M2 )
= M1 M2 and 1 (M1
M2 )
= 1 M1 1 M2
(5) Let M be an RG-lattice with no non-zero projective summands. Then M is indecomposable if and only if M is indecomposable, if and only if 1 M is indecomposable.
Proof. (1) The short exact sequence splits as a sequence of R modules, because W is
projective as an R-module, and so the dual sequence 0 W V U 0 is exact.
Suppose that V W is a projective cover and W has no non-zero projective summand.
Then V is projective, and if V U is not a projective cover then by Proposition 7.8
we may write V = X Y where X U is a projective cover and Y maps to zero. This
would mean that V
= (V ) Y and where Y is projective, and on dualizing we deduce
that V has a non-zero projective summand, which is not the case. Hence V U is a
projective cover, so that U V is a (relative) injective hull. The second statement follows
from the first on applying it to the dual sequence 0 W V U 0.
(2) The dual of the sequence 0 (M ) PM M 0 that computes (M )
is 0 M (PM ) (M ) 0 and it computes 1 (M ). Thus 1 (M )
= (M )
and dualizing again gives the result.
(3) These isomorphisms follow immediately from (1) since the same sequence that
constructs M also constructs 1 (M ) and the sequence that constructs 1 M also
constructs 1 (M ).
(4) This comes from the fact that the projective cover of M1 M2 is the direct sum
of the projective covers of M1 and M2 , and similarly with (relative) injective hulls.
(5) If (M ) were to decompose then so would M
= 1 (M ) by (4), so the indecomposability of M implies the indecomposability of M . The reverse implication and
the equivalence with the indecomposability of 1 M follow similarly.
216
We see from this that permutes the isomorphism types of indecomposable RGlattices, with inverse permutation 1 . Let us write i for the ith power of when i > 0
and the ith power of 1 when i < 0. When i = 0 we put 0 M = M .
Many further methods in representation theory that go beyond the scope of this book
make implicit reference to , especially the ones that have to do with homological algebra,
and the connection is the following. A key notion in homological algebra is that of a
projective resolution of a module M . This is a sequence of projective modules
d
d1
3
2
1
0
P
2 P1 P0 00
that is exact everywhere except at P0 , where its homology (in this case the cokernel
of the map d1 ) is M . When R is a field or a complete discrete valuation ring we can
construct always a minimal projective resolution, which has the property that each map
Pi Ker di1 is a projective cover. We see by induction that for a minimal projective
resolution, i M
= Ker di1 when i 1.
We will use properties of in the exercises as part of a proof that the list of indecomposable modules for k[C2 C2 ] is complete, where k is a field of characteristic 2. Before
we leave we mention its properties under the Kronecker product.
(11.31) PROPOSITION. Let R be a field or a complete discrete valuation ring and
G a finite group. For any RG-lattice M we have M R i R
= i M Qi for each i, where
Qi is a projective RG-module.
Proof. Let P2 P1 P0 0 be a projective resolution of k. We claim that
the sequence M R P2 M R P1 M R P0 0 is exact except in position
0. This is because every map in the resolution must split as a map of R-modules since
M is projective as an R-module. The homology in position 0 is M R k
= M for the
same reason. Furthermore, all of the modules M R Pi are projective RG-modules by
Proposition 8.4. Thus we have a projective resolution of M , but there is no reason why
it should be minimal. Assuming inductively that the kernel of the map in position i 1
has the form i M Qi for some projective module Qi (and this is true when i = 0), we
deduce from Proposition 7.8 that the kernel in position i has the form i+1 M Qi+1 ,
which completes the proof by induction on i when i 0. When i < 0 we can deduce the
result by duality using part (2) of Proposition 11.30.
We see from this result that up to isomorphism the modules i R where i Z, together
with the projective modules, are closed under the operations of taking indecomposable
summands of tensor products.
Peter Webb
217
Some further techniques with indecomposable modules

There are many further techniques for handling indecomposable modules that have
been developed, but at this point they start to go beyond the basic account we are attempting in this book and move into more specialist areas. Here is a pointer to some
further topics. The reader can consult more advanced texts about these topics, such as
the books by Benson [Ben].
Greens indecomposability theorem. This states for a p-group, working with
an algebraically closed field, that indecomposable modules remain indecomposable under
induction. More generally the result is true, over an arbitrary field, for modules that are
absolutely indecomposable. This means for p-groups (over algebraically closed fields)
that if a module has a certain subgroup as a vertex then the module is induced from
that subgroup. It also has the consequence for arbitrary groups that if an indecomposable
module has a certain subgroup as a vertex then its dimension is divisible by the index of
the vertex in a Sylow p-subgroup.
Blocks with cyclic defect. We have not yet defined the defect of a block (it is
done in the next section) but this concept includes the case of all representations of groups
with cyclic Sylow p-subgroups. Each block of cyclic defect is described by a tree, called the
Brauer tree, and from this tree it is possible to give a complete listing of indecomposable
modules together with their structure.
Auslander-Reiten theory. The theory of almost split sequences (also known as
Auslander-Reiten sequences) and the Auslander-Reiten quiver has been one of the fundamental tools in the abstract representation theory of algebras since they were introduced.
They provide a way to describe some structural features of indecomposable modules even
in situations of wild representation type, giving information that is more refined than the
theory of vertices, for example. In the context of group representations they have been used
in several ways, and most notable is the theory developed by Erdmann [Erd] in classifying
the possible structures of blocks of tame representation type.
The Green ring. This is a Grothendieck group of modules that has a basis
in bijection with the isomorphism classes of indecomposable modules, and it is a useful
home for constructions that extend classical results about the character ring. Notable
are induction theorems of Conlon and Dress, generalizing the theorems for the character
ring of Artin and Brauer. They have had various applications including a technique for
computing group cohomology.
Diagrams for modules. In this chapter and in some other places we have drawn
diagrams with nodes and edges to indicate the structure of representations, but nowhere
have these diagrams been defined in any general way. Evidently this is a useful approach
to representations, and the reader may wonder what kind of theory these diagrams have.
The trouble is that it is not easy to make a general definition that is sufficiently broad to
apply to all modules that might arise. If we do give a general definition we find that the
diagrams can be so complicated that they are no easier to understand than other ways of
218
looking at representations. The reader can find more about this in the work of [Alperin],
[Benson-Carlson], [Benson-Conway].
1. When R is a complete discrete valuation ring or a field the indecomposable summands in any direct sum decomposition of a finitely generated RG-module have isomorphism type and multiplicity determined independently of the particular decomposition.
2. When R is a field of characteristic p or a discrete valuation ring with residue field
of characteristic p, every RG-module is projective relative to a Sylow p-subgroup. An
RG-module is projective if and only if it is projective on restriction to a Sylow p-subgroup.
3. The representation type of kG is characterized in terms of the Sylow p-subgroups
of G.
4. Every indecomposable RG-module has a vertex and a source, which behave well
with respect to the Heller operator .
5. Green correspondence gives a bijection between isomorphism types of indecomposable RG-modules with vertex D and indecomposable R[NG (D)]-modules with vertex
D.

We will assume throughout these exercises that the ground ring R is either a complete
discrete valuation ring with residue field of characteristic p, or a field k of characteristic p,
so that the Krull-Schmidt theorem holds.
1. Write out proofs of the following assertions. They refer to subgroups H K G
and J G, an RG-module U and an RK-module V .
(a) If U is H-projective then U is K-projective.
g
(b) If U is H-projective and W is an indecomposable summand of U G
J then W is J Hprojective for some element g G. Deduce that there is a vertex of W that is contained
in a subgroup J gH.
(c) If U is a summand of V G
K and V is H-projective then U is H-projective.
(d) For any g G, U is H-projective if and only if gU is gH-projective.
(e) If U is H-projective and W is any RG-module then U W is H-projective.
2. Let A be a ring with a 1.
(1) Suppose that U is an A-module with two decompositions U = U1 U2 = V1 V2
as left A-modules corresponding to idempotent decompositions 1U = e1 + e2 = f1 + f2
in EndA (U ). Show that U1
= V2 if and only if f1 = e1 1 for some
= V1 and U2
AutA (U ).
Peter Webb
219
(2) Suppose that e, f A are idempotents. Show that Ae

= Af and A(1 e)
=
1
A(1 f ) as left A-modules if and only if f = e for some unit A .
(3) Suppose that A is Noetherian as a left module over itself and suppose I A is a
2-sided ideal of A with I Rad A. Let e, f A be idempotents for which e + I = f + I.
Show that f = e1 for some unit A . [Use part (2) and the uniqueness of projective
covers.]
3. Let U be an indecomposable module for a finite dimensional algebra A over a field.
Assuming that U is not simple, show that Soc U Rad U . Deduce that if U has Loewy
length 2 then Soc U = Rad U .
4. Let k be a field of characteristic 2, and let U be an indecomposable k[C2 C2 ]module that is not simple or projective. By considering the Loewy length of U , dim Soc U
and dim U/ Soc U show that one of the following three possibilities must occur:
(a) dim U < dim U < dim 1 U , or
(b) dim U > dim U > dim 1 U , or
(c) dim U = dim U = dim 1 U .
In cases (a) and (b) show that U = i (k) for some i. Deduce that dim U is odd and
| dim Soc U dim U/ Soc U | = 1.
In case (c) show that dim Soc U = dim U/ Soc U so that dim U is even.
5. Consider one of the even-dimensional k[C2 C2 ]-modules Ef,n (in the notation
used before Theorem 11.20) where k is a field of characteristic 2, and suppose that f 6=
0, . The goal of this question is to show that Ef,n is indecomposable. Define umn+1 =
amn1 umn1 a1 u2 a0 u1 and let : Ef,n Ef,n be the linear map specified by
(ui ) = ui+1 , (vi ) = vi+1 where 1 i mn.
(1) Show that Endk[C2 C2 ] (Ef,n ).
(2) Show that the subalgebra k[] of Endk[C2 C2 ] (Ef,n) generated by is isomorphic
to k[X]/(f n).
(3) Show that
k[] + Homk[C2 C2 ] (Ef,n, Soc(Ef,n)) = Endk[C2 C2 ] (Ef,n).
(4) Deduce that Ef,n is an indecomposable k[C2 C2 ]-module.
6. Let G = hai hbi = C2 C2 and let U be an even-dimensional indecomposable
kG-module where k is a field of characteristic 2. Suppose that multiplication by a 1
induces an isomorphism U/ Rad(U ) Soc(U ).
(1) Show that there is an action of the polynomial ring k[X] on U so that X(a 1)u =
(b 1)u = (a 1)Xu for all u U . Show that U
= khai k (U/ Rad(U )) as khai k K[X]modules.
(2) Show that invariant subspaces of U as a khai k K[X]-module are also invariant
subspaces of U as a kG-module. Show that U/ Rad(U ) is an indecomposable k[X]-module
and deduce that U/ Rad(U )
= k[X]/(f n) as k[X]-modules for some irreducible polynomial
f and integer n.
(3) Prove that U
= Ef,n as kG-modules.
220
7. Let k be a field of characteristic p and suppose that U is a kG-module with the

property that for every proper subgroup H < G, U G
H is a projective kH-module.
(1) Show that if G is a cyclic p-group of order p then U must be a projective
kG-module.
(2) Show by example that if G = C2 C2 is the Klein four-group and p = 2 then U
need not be a projective kG-module.
8. Suppose that U is an indecomposable RG-module that is Q-projective and that
G
U Q has an indecomposable summand that is not projective relative to any proper subgroup of Q. Show that Q is a vertex of U .
9. Let H be a subgroup of G and U an indecomposable RG-module that is a direct
summand of R G
H . Show that the source of U is the trivial module (for the subgroup that
is the vertex of U ).
[Because of this, the indecomposable summands of permutation modules (over a field) are
some called trivial source modules.]
10. Suppose that Q is a vertex of an indecomposable RG-module U and that H is a
subgroup of G that contains Q.
(a) For each subgroup Q H that is conjugate in G to Q, show that U G
H has an
G
indecomposable summand with vertex Q . Deduce that if U H is indecomposable then
subgroups of H that are conjugate in G to Q are all conjugate in H.
(b) Show that there is an indecomposable RH module V with vertex Q so that U is
a direct summand of V G
H.
11. Suppose that H is a subgroup of G and that V is an indecomposable RH-module
with vertex Q, where Q H. Show that V G
H has an indecomposable direct summand
with vertex Q. Show that for every p-subgroup Q of G there is an indecomposable RGmodule with vertex Q.
12. Given an indecomposable RG-module U , let X be the set of pairs (Q, T ) such
that Q is a vertex of U and T is a source of U with respect to Q. For each g G define
g
(A, T ) := (g Q, g T ). Show that this defines a permutation action of G on X, and that it
is transitive.
13. Let U be an indecomposable RG-module. Show that U and U have the same
vertex.
14. Let U be an indecomposable kG-module where k is a field. Assuming that U is
not projective, show that U/ Rad U
= Soc U .
15. Let U be an indecomposable RG-module with vertex Q, let H NG (Q) and
let f (U ) be the RH-module that is the Green correspondent of U . Show that f (U ) =
(f (U )).
16. The group G := P SL(2, 5) (which is isomorphic to A5 ) has three simple modules
over k = F5 , of dimensions 1, 3 and 5, and has Cartan matrix (with the simples taken in
the order just given)
2 1 0
1 3 0
0 0 1
Peter Webb
221
You may assume that a Sylow 5-subgroup is cyclic of order 5 and that its normalizer H is
dihedral of order 10.
(a) Show that has two orbits, each of length 4, on the non-projective kD10 -modules.
Deduce that for kG there are 8 indecomposable non-projective modules in two orbits of
length 4.
(b) Show that apart from a block of defect zero, each indecomposable projective
module for kG has Loewy length 3, and that its radical series equals its socle series.
(c) Show that each non-projective indecomposable module for kG has Loewy length
2.
(d) Identify the socle and the radical quotient of each of the 8 indecomposable nonprojective kG-modules.
17. The group G := P SL(2, 7) (which is isomorphic to GL(3, 2)) has four simple
modules over k = F7 , of dimensions 1, 3, 5 and 7, and has Cartan matrix (with the simples
taken in the order just given)
2 0 1 0
0 3 1 0
1 1 2 0
0 0 0 1
You may assume that a Sylow

is cyclic of order 7 and that its normalizer H is
7 7-subgroup
3

a non-abelian group hx, y x = y = 1, yxy 1 = x2 i of order 21.
(a) Show that has three orbits, each of length 6, on the non-projective kH-modules.
Deduce that for kG there are 18 indecomposable non-projective modules in three orbits
of length 6.
(b) Show that apart from a block of defect zero, each indecomposable projective
module for kG has Loewy length 3, and that its radical series equals its socle series.
(c) Show that each non-projective indecomposable module for kG has Loewy length
2.
(d) Identify the socle and the radical quotient of each of the 18 indecomposable nonprojective kG-modules.
18. Let k be a field of characteristic p and suppose that G = H K where H = hxi
=
Cpn is cyclic of order pn and K is a group of order relatively prime to p (as in Corollary
11.9).
(a) Let Ur be the indecomposable kH-module of dimension r, 1 r pn . If J H,
show that J is a vertex of Ur if and only if r = |H : J|q where q is prime to p. [Use
the fact, shown in the exercises to Chapter 6, that indecomposable kJ-modules induce to
indecomposable kH-modules.]
(b) Using the description of projective kG-modules in Proposition 8.9 and the description of indecomposables in 11.8, show that if V is an indecomposable kG-module then
V G
H is a direct sum of copies of a single indecomposable kH-module Ur , for some r, the
number of copies being dim S for some simple kK-module S. Let J H be a vertex of V .
Show that r = |H : J|q for some number q prime to p. Assuming that k is a splitting field
for G, show that dim V = |H : J|q for some number q prime to p.
222
19. Let H be a subgroup of G and let U , V be RG-modules. We say that an RGmodule homomorphism f : U V factors through an H-projective module if there is an
H-projective RG-module M and RG-module homomorphisms U M V so that f = .

(a) Similarly to Proposition 11.13, show that the following conditions are equivalent:
(1) f factors through an H-projective module,
(2) f factors through an induced module N G
H for some RH-module N ,
G
(3) f factors as U U G
H H V for some RG-module homomorphism , where is the
map constructed after 11.12,
G
(4) f factors as U V G
H H V for some RG-module homomorphism , where is the
map constructed after 11.12,
(5) f lies in the image of trG
H : HomRH (U, V ) HomRG (U, V ).
(b) When R is a field or a complete discrete valuation ring, use question 9 from Chapter
8 to show that the space of homomorphisms U V that factor through a projective has
dimension equal to the multiplicity of PR as an RG-module summand of HomR (U, V ).
Peter Webb
223
12. Blocks
Block theory is one of the deepest parts of the representation theory of finite groups,
and in this chapter we can only scratch the surface of the sophisticated constructions
and techniques that have been developed. We present the basic parts of the theory, in
line with one of the goals of this book, which is to have relevance for the needs of the
non-specialist in dealing with representations that naturally arise. For such purposes it
is useful to have an idea of the defect group of a block and its relation to vertices of
indecomposable modules in the block. It is useful to see the different methods used in
block theory - sometimes ring-theoretic, sometimes module-theoretic, and to be able to
relate them. After a discussion of different definitions of a block involving idempotents,
ring direct summands and equivalence classes of representations, and having presented
some examples, we move on to define the defect group in different ways, showing that
these ways are equivalent. We are able to see that the blocks of defect zero introduced in
Chapter 9 do indeed deserve the name. We conclude with Brauers First Main Theorem,
which establishes a correspondence between blocks with a certain defect group D and
blocks of NG (D) with defect group D.
The various definitions of a block

Let (F, R, k) be a p-modular system and G a finite group. Thus R is a discrete valuation ring with field of fractions F of characteristic zero and residue field k of characteristic
p. We will define a p-block of G in a moment, but whatever it is, it determines and is
determined by any of the following data:
an equivalence class of kG-modules,
an equivalence class of indecomposable kG-modules,
an equivalence class of simple kG-modules,
an equivalence class of RG-modules,
a primitive central idempotent in kG or RG,
an equivalence class of primitive idempotents in kG or RG,
an equivalence class of F G-modules,
an indecomposable 2-sided direct summand of kG,
an indecomposable 2-sided direct summand of RG,
a division of the Cartan matrix of kG into block diagonal form obtained by permuting
rows and permuting columns, with as many diagonal blocks as possible.
In view of this, a block can be defined by specifying any of these data, and there are many
possible definitions of a block. People who work with blocks often seem to have many of
these definitions in mind at the same time. Almost always we will take a block of a ring A
with identity to mean a primitive idempotent in the center Z(A). We start by exploring
some of the equivalent properties of blocks. First we recall without proof Proposition 3.22.
224
(12.1) PROPOSITION. Let A be a ring with identity. Decompositions

A = A1 Ar
as direct sums of 2-sided ideals Ai biject with expressions
1 = e1 + + er
as a sum of orthogonal central idempotent elements, where ei is the identity element of
Ai and Ai = Aei . The Ai are indecomposable as rings if and only if the ei are primitive
central idempotent elements. If every Ai is indecomposable as a ring then the Ai , and also
the primitive central idempotents ei , are uniquely determined as subsets of A, and every
central idempotent can be written as a sum of certain of the ei .
It makes sense to consider blocks for arbitrary algebras with identity, but the notion
is particularly relevant for group algebras because here we are naturally presented with an
algebra that may not be indecomposable. In a more abstract study of algebras in general
we would probably make the assumption at the start that the algebra we are studying is
indecomposable, and if we were to do this the notion of a block would be irrelevant.
(12.2) PROPOSITION. Let A be a ring with identity, let 1 = e1 + + en be a sum
of orthogonal idempotents of Z(A) and let U be an A-module. Then U = e1 U en U
as A-modules. Thus if U is indecomposable we have ei U = U for precisely one i, and
ej U = 0 for j 6= i. These summands of U satisfy HomA (ei U, ej U ) = 0 if i 6= j.
This result means that each indecomposable A-module U belongs to a unique block, or
lies in a unique block, namely the unique primitive central idempotent e for which eU 6= 0,
and this e acts as the identity on U . More generally, we say that an A-module U belongs
to e if each of its indecomposable summands belongs to e, and this is the same as requiring
that eU = U . The modules that belong to a block determine that block, since in any
expression for A as a direct sum of indecomposable modules the sum of the summands
belonging to e is necessarily the 2-sided ideal eA. When A = RG is a group algebra, the
block to which the trivial module R belongs is called the principal block.
(12.3) PROPOSITION. Let G be a finite group and (F, R, k) a p-modular system in
which R is complete.
(1) Reduction modulo () gives a surjective ring homomorphism Z(RG) Z(kG).
(2) Each idempotent of Z(kG) lifts uniquely to an idempotent of Z(RG), with primitive
idempotents corresponding to primitive idempotents under the lifting process.
P
Proof. (1) The conjugacy class sums gx g (i.e. the sum of all elements g conjugate
to x) form a basis for Z(RG) over R, and over k they form a basis for Z(kG). Reduction
modulo () sends one basis to the other, which proves surjectivity.
Peter Webb
225
(2) We have seen the lifting argument before. Since n1 Z(RG)/ n Z(RG) is a nilpotent ideal in Z(RG)/ n Z(RG) we may lift any idempotent en1 + n1 Z(RG) to an
idempotent en + n Z(RG), thereby obtaining from any idempotent e1 + Z(RG) Z(kG)
a Cauchy sequence e1 , e2 , . . . of elements of Z(RG) whose limit is the required lift. We have
also seen before that primitive idempotents correspond to primitive idempotents. In the
present situation the lift of each primitive idempotent is unique since the primitive central
idempotents of RG themselves are unique, so that there is only one primitive idempotent
of Z(RG) that reduces to each primitive idempotent of Z(kG).
Because of this, it is the same thing to study the blocks of RG and of kG since they
correspond to each other under reduction modulo . If U is a kG-module we may regard
it also as an RG-module via the surjection RG kG and if e is a block of RG with image
the block e Z(kG) there is no difference between saying that U belongs to e or that U
belongs e. We may also partition the simple F G-modules into blocks in a way consistent
with the blocks for RG and kG. Regarding RG as a subset of F G, a primitive central
idempotent e of RG is also a central idempotent of F G. We say that an F G module U
belongs to e if eU = U . Evidently each simple F G-module belongs to a unique block. We
see that if U is an RG-module and U0 U is any R-form of U (i.e. a full RG-lattice in
U ) then U0 belongs to e if and only if U belongs to e.
Examples. 1. When A is a finite-dimensional semisimple algebra over a field, the
blocks correspond to the matrix summands of A, each block being the idempotent that
is the identity element of a matrix summand. Each simple module lies in its own block,
and so there is only one indecomposable module in each block. We saw in Theorem 3.23 a
formula in terms of characters for the primitive central idempotent in CG corresponding
to each simple complex representation.
2. When G is a p-group and k is a field of characteristic p the regular representation
kG is indecomposable and the identity element is a block. There is only one block in this
situation and block theory does nothing for us if we are interested only in representations
of p-groups in characteristic p.
3. We have seen at the end of Chapter 9 that, when we have a block of defect zero
for G over a splitting p-modular system (F, R, k), there is a 2-sided direct summand of
RG that is isomorphic to a matrix algebra Mn (R), and also a matrix summand Mn (k)
of kG that is the reduction modulo of the summand of RG. There is a unique simple
kG-module in this block, and it is projective. It lifts to a unique RG-lattice, and all RGsublattices of it are isomorphic to it. A key fact that we used in identifying the features of
this situation is that the primitive central idempotent of CG corresponding to the matrix
summand in fact lies in RG. All this explains why we made reference to a block in that
situation, but not why we used the term defect zero. This too will soon be explained.
4. When G = S3 in characteristic 2 there are two blocks, since the simple module of
degree 2 is a block of defect zero, and the only other simple module is the trivial module,
which lies in the principal block. This has been seen in Example 7.5, the examples after
9.27 and Theorem 9.28. The projective cover of the trivial module as an RG-module has
226
character equal to the sum of the characters of the trivial representation and the sign
representation, and so the principal block idempotent in RG acts as the identity on this,
meaning that they are the ordinary characters in the principal block. The other ordinary
character, of degree 2, lies in the other block.
In characteristic 3, S3 has only one block since there are two simple modules (trivial
and sign) and the sign representation appears as a composition factor of the projective
cover of the trivial module. This means that it belongs to the principal block.
(12.4) LEMMA. Let e be a block of a ring A with identity. If 0 U V W 0
is a short exact sequence of A-modules then V belongs to e if and only if U and W belong
to e.
In other words, every submodule and factor module of a module that belongs to e
also belong to e, and an extension of two modules that belong to e also belongs to e.
Proof. A module belongs to e if and only if multiplication by e is an isomorphism of
that module, and this property holds for V if and only if it holds for U and W .
In what follows we characterize blocks in terms of a relation on the simple modules.
We could work with modules for an algebra over a complete discrete valuation ring R, but
since the simple modules are naturally defined over the residue field k we will assume that
A is a finite-dimensional k-algebra. The next result is an elaboration of the observation
from the last lemma that each indecomposable projective module PS lies in the same block
as the simple module S, and in fact all of the composition factors of PS lie in this block.
(12.5) PROPOSITION. Let A be a finite-dimensional algebra over a field k. The
following are equivalent for simple A-modules S and T .
(1) S and T lie in the same block.
(2) There is a list of simple A-modules S = S1 , S2 , . . . , Sn = T so that Si and Si+1
are both composition factors of the same indecomposable projective module, for each
i = 1, . . . , n 1.
(3) There is a list of simple A-modules S = S1 , S2 , . . . , Sn = T so that for each i =
1, . . . , n 1, Si and Si+1 appear in a non-split short exact sequence of A-modules
0 U V W 0 with {U, W } = {Si , Si+1 }.
Proof. The implications (3) (2) (1) are straightforward in that all the composition factors of any indecomposable module necessarily belong to the same block, and
whenever we have a non-split extension of the kind that appears in condition (3), the
middle module is uniserial and hence indecomposable.
To show that (1) (2) we will write S T to mean that the simple modules S and
T satisfy the condition of (2), which is an equivalence relation. All the composition factors
of any particular indecomposable projective module are equivalent in this sense. Suppose
that S and T lie in the same block. We need to show that S T . We may write the
Peter Webb
227
regular representation as A = P1 Pr Q1 Qs where the composition factors of

every Pi are equivalent to S, and none of the composition factors of the Qj are equivalent
to S. Let P = P1 Pr and Q = Q1 Qs so that A = P Q. In fact every
submodule of A whose composition factors are all equivalent to S is contained in P , for
if U is such a submodule then P + U has the same property, and (P + U ) Q = 0 since
composition factors of the intersection are both equivalent and not equivalent to S. Thus
A = (P + U ) A, and it follows that P + U = P so U P .
We show that P and Q are 2-sided ideals of A. Since they are already left ideals
we only need to show that they are closed under right multiplication by elements of A.
Each element x A determines a module endomorphism : A A, which is (a) = ax.
Now (P ) is a submodule of A all of whose composition factors are equivalent to S, so
(P ) P . This means that P x P , so that P is a right ideal, as we were trying to show.
Similarly Q is a right ideal of A, and hence a 2-sided ideal.
It follows that P is a direct sum of block ideals. The block ideal determined by S is
a direct summand of P , and hence every simple module in this block is equivalent to S in
the sense of condition (2).
(2) (3)
The effect of this result is that the division of the simple A-modules into blocks can
be achieved in a purely combinatorial fashion, knowing the Cartan matrix of A. This is
the content of the next corollary, in which the term block is used in two different ways.
The connection with the block matrix decomposition of the Cartan matrix is probably the
origin of the use of the term in representation theory.
(12.6) COROLLARY. Let A be a finite-dimensional algebra over a field k. On listing
the simple A-modules so that modules in each block occur together, the Cartan matrix of
A has a block diagonal form, with one block matrix for each block of the group. Up to
permutation of simple modules within blocks and permutation of the blocks, this is the
unique decomposition of the Cartan matrix into block diagonal form with the maximum
number of block matrices.
Proof. Given any matrix we may define an equivalence relation on the set of rows
and columns of the matrix by requiring that a row be equivalent to a column if and only
if the entry in that row and column is non-zero, and extending this by transitivity to
an equivalence relation. Now if we order the rows and columns so that the rows and
columns in each equivalence class come together, the matrix is in block form, and this is
the unique expression with the maximal number of blocks (up to permutation of the blocks
and permutation of rows and columns within a block). The last result implies that this
equivalence relation coincides with the division of modules into blocks.
228
Example. Suppose that G = K H where K has order prime to p and H is a pgroup. In other words, G has a normal p-complement and is termed p-nilpotent. We saw in
Theorem 8.10 that this is precisely the situation in which each indecomposable projective
module has only one isomorphism type of composition factor. Another way of expressing
this is to say that the Cartan matrix is diagonal. In view of the last results we see that
we have characterized the groups for which each p-block contains just one simple module
over a field of characteristic p as being the p-nilpotent groups.
We can describe the primitive central idempotents explicitly in this situation.
(12.7) PROPOSITION. Let G = K H be a p-nilpotent group, where K has order
prime to p and H is a p-group. Let k be a field of characteristic p. Each block of kG lies
in kK, and is the sum of a G-conjugacy class of blocks of kK.
Proof. Observe that if 1 = e1 + + en is the sum of blocks of kK then for each
i and g G the conjugate gei g 1 is also a block of kK. For this we verify that this
element is idempotent, and also that it is central in kK, which is so since if x K then
xgei g 1 = g(g 1 xg)ei g 1 = gei (g 1 xg)g 1 = gei g 1 x. Furthermore gei g 1 is primitive

in Z(kK) since if it were the sum of two orthogonal central idempotents, on conjugating
back by g 1 we would be able to deduce that ei is not primitive either.
The blocks of kK are uniquely determined, and it follows that gei g 1 = ej for some
j. Thus G permutes the blocks of kK.
For each fixed i the element f =
e=gei g 1
central in kG since if x G then xf x1 =
e Z(kK) is idempotent. It is in fact
e=gei g 1
xex1 = f , the sum again being over
the elements in the G-orbit of ei . We now show that f is primitive in Z(kG). Suppose
instead that f = f1 + f2 is a sum of orthogonal idempotents in Z(kG). Then there are
non-isomorphic simple kG-modules U1 and U2 with f1 U1 = U1 and f2 U2 = U2 . Since
f f1 = f1 and f f2 = f2 we have f U1 = U1 and f U2 = U2 .
a
a
By Cliffords theorem U1 G
K = S1 St for some integer a, where S1 , . . . , St are G-
conjugate simple kK-modules. Since f kK we have f Si = Si for all i, and this identifies
these modules exactly as the G-orbit of simple kK modules associated to f . By a similar

argument U2 G
= S b S b for some integer b, where the Si are the same modules.
K
Since K consists of the p-regular elements of G it follows that the Brauer characters of
U1 and U2 are scalar multiples of one another. Since Brauer characters of non-isomorphic
simple modules are linearly independent we deduce that U1
= U2 , a contradiction. This
shows that f is primitive in Z(kG).
Peter Webb
229
We may see the phenomenon described in the last result in many examples, of which
the smallest non-trivial one is G = S3 = K H where K = h(1, 2, 3)i and H = h(1, 2)i,
taking p to be 2 and k = F4 . By Theorem 3.23 the blocks of kK are
e1 = () + (1, 2, 3) + (1, 3, 2),
e2 = () + (1, 2, 3) + 2 (1, 3, 2), and
e3 = () + 2 (1, 2, 3) + (1, 3, 2)
where is a primitive cube root of 1 in F4 . (In case the reader expects some factors 31 ,
note that 3 = 1 in characteristic 2. In the next expression 1 has been written as 1 and
2 as 0.) In the action of G on these idempotents there are two orbits, namely {e1 } and
{e2 , e3 }. The blocks of kG are thus e1 and e2 + e3 = (1, 2, 3) + (1, 3, 2). These idempotents
have already appeared in Example 7.5, where it was calculated that kGe1 = 11 = P1 and
kG(e2 + e3 ) = 2 2 is the direct sum of two simple projective kG-modules. This structure
has also been considered in Exercise 5 of Chapter 8. From Theorem 9.28 we know the
second summand is a block of defect zero, and kG(e2 + e3 )
= M2 (k) as rings. We also see
2
that kGe1 = kC2 = k[X]/(X ) as rings. An approach to this can be found in Exercise 11
of Chapter 8.
The defect of a block: module theoretic methods

In view of the multiple meanings of the term block it will be no surprise that there
is more than one approach to the definition of a defect group of a block. We will start
with a module-theoretic approach pioneered by J.A. Green, and after that relate it to the
ring-theoretic approach that goes back to Brauer.
There are two ways to obtain the regular representation from the group ring RG. One
is to regard RG as a left RG-module via multiplication from the left; but there is a second
left action of G in which an element g G acts by multiplication from the right by g 1 .
The module we obtain in this way is isomorphic to the regular representation obtained via
left multiplication. Because these two actions commute with each other we may combine
them, and regard kG as a representation of G G with an action given by
(g1 , g2 )x = g1 xg21
where g1 , g2 G, x kG.
It will be important to consider the diagonal embedding of : G G G specified by

(g) = (g, g). Via this embedding we obtain yet another action of G on RG, which is the
action given by conjugation: g x = gxg 1 .
230
(12.8) PROPOSITION.
(1) The submodules of RG, regarded as a module for R[G G], are precisely the 2-sided
ideals of RG.
(2) Any decomposition of RG as a direct sum of indecomposable R[G G]-modules is a
decomposition as a direct sum of blocks.
(3) Regarded as a representation of G G, RG is a transitive permutation module in
GG
which the stabilizer of 1 is (G). Thus RG
as R[G G]-modules.
= R (G)
Proof. (1) The 2-sided ideals of RG are precisely the R-submodules of RG that are
closed under multiplication from the left and from the right by G. These is equivalent to
being an R[G G]-submodule
(2) This follows from (1) and 12.1.
(3) Evidently G G permutes the basis of RG consisting of the group elements. We

only need check that StabGG (1) = {(g1 , g2 ) g1 1g21 = 1} = (G).
(12.9) COROLLARY. Let R be a field of characteristic p, or (more generally) a

discrete valuation ring with residue field of characteristic p. Let G be a finite group and
let e be a block of RG. Then, regarded as a R[G G]-module, the summand eRG has a
vertex of the form (D) where D is a p-subgroup of G. Such a subgroup D is uniquely
defined to within conjugacy in G.
Proof. By 12.8 eRG is indecomposable as an R[G G]-module, and since it is (G)projective it has a vertex contained in (G). Such a subgroup is necessarily a p-group, so
has the form (D) for some p-subgroup D of G where (G) is determined up to conjugacy
in G G. If D1 is another subgroup of G for which (D1 ) is a vertex of eRG then
(D1 ) = (g1 ,g2 ) (D) for some g1 , g2 G, and so for all x D, (g1 x, g2 x) (D1 ). Thus
g1
x D1 for all x D and since D and D1 have the same order it follows that D1 = g1 D.
We define a subgroup D of G to be a defect group of the block e if (D) is a vertex of
eRG, and according to Corollary 12.9 it is defined up to conjugacy in G. If |D| = pd we
say that d is the defect of e, but often we abuse this terminology and say simply that the
defect of the block is D. According to these definitions, a block of defect 0 is one whose
defect group is the identity subgroup. It is not immediately apparent that this definition
of a block of defect zero coincides with the previous one, and this will have to be proved.
We now start to investigate the kinds of groups that can be defect groups.
Peter Webb
231
(12.10) THEOREM (Green). Let e be a p-block of G with defect group D, and let
P be a Sylow p-subgroup of G that contains D. Then D = P gP for some element
g CG (D).
Proof. Let P be a Sylow p-subgroup of G containing D. We consider
GG GG
GG
RG P
P = R (G) P P
M
=
(G)
P P
( y(R (P P )y (G) )) (P
P ) y(G)
y[P P \GG/(G)]
y[P P \GG/(G)]
P P
R (P
P ) y(G) .
P P
Now each R (P
P ) y(G) is indecomposable since P P is a p-group, as shown in the
proof of part (3) of Proposition 11.24. The summand eRG of RG on restriction to P P
has a summand that on further restriction to (D) has a source of eRG as a summand.
GG
This summand of eRG P
P also has vertex (D) (an argument that is familiar from the
P P
proof of the Green correspondence 11.26), and must have the form R (P
P ) y(G) for some
y
y GG. Thus (D) is conjugate in P P to (P P ) (G), so (D) = z((P P ) y(G))
for some z P P .

The elements 1 G = {(1, t) t G} form a set of coset representatives for (G) in
G G, and so we may assume y = (1, t) for some t G. Write z = (r, s), where r, s P .
Now

(P P ) (1,t)(G) = {(x, tx) x P and tx P }

= {(x, tx) x P P t }
(1,t)
(P P t )
and (D) = (r,s)(1,t)(P P t ). The projection onto the first coordinate here equals D =
1
1
r
(P P t ) = rP rt P = P rt P since r P . At this point though the proof is complete,
1
1 1
apart from the fact that rt1 might not centralize D. Now (1,t )(r ,s )(D) (G),
1 1
1
so that r x = t s x for all x D, and rt1 s1 CG (D). Since s P we have
1 1
D = P rt s P and this completes the proof.
(12.11) COROLLARY. If e is a p-block of G with defect group D then
D = Op (NG (D)) Op (G),
where Op (G) denotes the largest normal p-subgroup of G.
Proof. We start by observing that OP (G) is the intersection of the Sylow p-subgroups
of G; for the intersection of the Sylow p-subgroups is a normal p-subgroup since the Sylow
p-subgroups are closed under conjugation, and on the other hand Op (G) P for some
232
Sylow p-subgroup P , and hence g(Op (G)) = Op (G) gP for every element g G. Since
g
P accounts for all Sylow p-subgroups by Sylows theorem it follows that Op (G) is contained
in their intersection.
We see immediately that D Op (G) since D is the intersection of two Sylow psubgroups and hence contains the intersection of all Sylow p-subgroups.
To prove that D = Op (NG (D)), let P be a Sylow p-subgroup of G that contains a
Sylow p-subgroup of NG (D). Such a P necessarily has the property that P NG (D) is a
Sylow p-subgroup of NG (D). Now D = P gP for some g CG (D) and so in particular
g NG (D). Thus gP NG (D) = g(P NG (D)) is also a Sylow p-subgroup of NG (D), and
D = (P NG (D)) g(P NG (D)) is the intersection of two Sylow p-subgroups of NG (D).
Thus D Op (NG (D)). But on the other hand D is a normal p-subgroup of NG (D), and
so is contained in Op (NG (D)). Thus we have equality.
The condition on a subgroup D of G that appeared in Corollary 12.11, namely that
D = Op NG (D), is quite restrictive. Such subgroups have been called p-radical subgroups in
recent years, and they have also been called p-stubborn subgroups. They play an important
role in many questions about groups to do with topology, for example in studying the
properties of classifying spaces and in studying the topological properties of the partiallyordered set of p-subgroups of G. The terminology p-radical comes from the fact that when
G is a finite group of Lie type in characteristic p the p-radical subgroups are precisely
the unipotent radicals of parabolic subgroups. Thus the definition of p-radical subgroup
extends this notion to all finite groups. The name p-stubborn refers to the fact that in
certain calculations with the partially-ordered set of p-subgroups of G, many subgroups
can be ignored, but the p-stubborn ones cannot. The reader should be warned that the
term p-radical has also been used in some different senses, one of which is described in
the book by Feit [Fei].
It is immediate that Sylow p-subgroups of a group are p-radical, as is Op (G). An
exercise in group theory (presented at the end of this section) shows that other p-radical
subgroups must lie between these two extremes. It is always the case that G has a p-block
with defect group a Sylow p-subgroup (the principal block, for example, as will be seen in
Corollary 12.18) but it need not happen that Op (G) is the defect group of any block. An
example of this is S4 in characteristic 2: from the Cartan matrix computed in Chapter 10
and Corollary 12.6 we see that the only block is the principal block, and it also follows
from Corollary 12.26.
Peter Webb
233
The defect of a block: ring-theoretic methods

The advantage of the module-theoretic approach we have just taken is that it immediately establishes the notion of a defect group of a block and its uniqueness up to conjugacy,
using the theory of vertices and sources we have previously developed. In view of the fact
that blocks are rings it is not surprising that there is also a ring-theoretic approach. We
use it in the next results, showing in Corollary 12.17 that modules for a block with defect
group D are projective relative to D. As a consequence we will see that defect groups of
the principal block are Sylow p-subgroups (Corollary 12.18) and also that blocks of defect
zero are correctly named (Corollary 12.19). Before that we obtain further characterizations
of the defect group.
Our main tool will be the relative trace map. We have already been using the properties of this map in the context of endomorphism rings of modules and we will now do
something very similar in the context of blocks. It is helpful at this point to introduce a
more general axiomatic setting that includes these examples.
Let R be a commutative ring with a 1 and G a group. We define a G-algebra over R to
be an R-algebra A together with an action of G on A by R-algebra automorphisms. Thus
for each g G and a A there is defined an element g a A so that the mapping a 7 g a
is R-linear, g (ab) = g ag b always and g 1 = 1. A homomorphism of G-algebras : A B
is defined to be an algebra homomorphism so that (g a) = g (a) always holds.
Given an RG-module U we have already been using the G-algebra structure on
EndR (U ) given by g f (u) = gf (g 1u) whenever f EndR (U ), g G, u U . Another
example of a G-algebra is the group rung RG, where for x RG and g G we define
g
x = gxg 1 . In fact the RG-module structure on U is given by an algebra homomorphism
RG EndR (U ), and it is a homomorphism of G-algebras. The same holds for any block
B, and for B-modules: since B is a summand of RG as a representation of G G, B is
preserved under the G-algebra action on RG and so becomes a G-algebra in its own right.
If U is a B-module we obtain a G-algebra homomorphism B EndR (U ), since this factors
as B RG EndR (U ) where the first map is inclusion (not an algebra homomorphism,
but the action of G is preserved).
Whenever A is a G-algebra we have algebras of fixed points AH for each subgroup
H
G
H G, and as before the relative trace map trG
AG , the inclusion resG
H : A
H : A
g
AH and conjugation cg : AH A H given by a 7 g a. They satisfy the properties already
established in Lemma 11.10 as well as the analogues of 11.11 and 11.12 which we present
now.
(12.12) PROPOSITION. Let A be a G-algebra and H G.
G
G
G
(1) If a AG , b AH we have a(trG
H b) = trH (ab) and (trH b)a = trH (ba).
H
G
H
(2) The image of trG
AG is an ideal of AG . The inclusion resG
is a
H : A
H : A A
ring homomorphism.
234
G
(3) If : A B is a homomorphism of G-algebras then (trG
H (b)) = trH ((b)).
Proof. The calculations are the same as for 11.11 and 11.12, and part (3) is immediate.
For most of the arguments we will present it will be sufficient to consider G-algebras,
but sometimes a stronger property is needed which block algebras possess. We define an
interior G-algebra over R (a notion due to L. Puig) to be an R-algebra A together with
a group homomorphism u : G A where A denotes the group of units of A. As an
example, the group algebra RG is itself an interior G-algebra where u is the inclusion of
G as a subset of RG. Whenever A is an interior G-algebra and : A B is an algebra
homomorphism (sending 1A to 1B ) we find that B becomes an interior G-algebra via the
homomorphism u : G B . Thus if B is a block of G the algebra homomorphism
RG B makes B into an interior G-algebra. As a further example, if U is any RGmodule its structure is determined by an algebra homomorphism RG EndR (U ) that
expresses the action of RG. In fact, to specify a module action of G on U is the same as
specifying the structure of an interior G-algebra on EndR (U ).
Given an interior G-algebra A we obtain the structure of a G-algebra on A by letting
each g G act as a 7 ga := u(g)au(g 1). We see immediately that our three examples of
the G-algebra structure on RG, on a block, and on EndR (U ) are obtained in this way.
Why should we consider interior G-algebras? The reason here is that if A is an interior
G-algebra and U is an A-module then we can recover the structure of U as an RG-module
u
from the composite homomorphism GA EndR (U ), and without the extra property
of an interior G-algebra we cannot do this. This property is used in the next lemma, which
will be used in Corollary 12.17.
(12.13) PROPOSITION. Let A be an interior G-algebra over R, let U be an AH
module and let H be a subgroup of G. Suppose that 1A = trG
H a for some element a A .
Then regarded as an RG-module, U is H-projective.
Proof. The representation of A on U is given by a G-algebra homomorphism : A
EndR (U ). This homomorphism is a homomorphism of G-algebras and by Proposition
G
12.12 we have 1U = (1A ) = (trG
H a) = trH (a). Thus by Higmans criterion (part (6) of
11.13) U is H-projective.
Our goal now is Theorem 12.16, which makes a connection between the moduletheoretic approach to the defect group as we have defined it, and the ring-theoretic ap
proach. The proof will follow from the next two results. As before, (G) = {(g, g) g G}.
Peter Webb
235
(12.14) LEMMA. Let U be an R[G G]-module that is (H)-projective. Then
GG
(G)
is (H)-projective as a representation of (G).
GG
GG
Proof. The hypothesis says that U is a summand of some module V (H)
, so U (G)
is a summand of
GG GG
(G) =
V (H)
(G)
(H)
( x(V (G)x (H) )) (G) x(H) .
x[(G)\GG/(H)]
In this formula we may write each x G G as x = (a, b) = (a, a)(1, a1b) and now
(G) x(H) =
(a,a)
((G)
(a,a)
(a,a)
{(H, a
(1,a1 b)
(H))

1
b
h) h = a bh}
(CH (a1 b))
(a,a)
(H).
GG GG
(G) is projective relative
It follows that every summand of the decomposition of V (H)
GG
is
to a (G)-conjugate of (H), which is the same as being (H)-projective. Thus U (G)
(H)-projective.
The point about these lemmas is that we are relating the structure of a block algebra
eRG as a representation of G G to its structure as a representation of G via the isomor-
phism G (G). Note in this context that if U is an RG-module and EndRH (U )

(G)
then trG
H and tr(H) are exactly the same thing, from the definitions. This is because G
is taken to act on EndR (G) via .
(12.15) LEMMA. Let e Z(RG) be a central idempotent of RG and H G. Then
eRG is projective relative to (H) as an R[G G]-module if and only if e = trG

H a for some
element a (eRG)H .
Proof. Suppose first that eRG is (H)-projective as an R[G G]-module. Then by
GG
is (H)-projective. By Higmans criterion there is an endomorLemma 12.14 (eRG) (G)
(G)
phism of eRG as a R[(H)]-module so that the identity morphism 1eRG = tr(H) =
236
trG
H . Now
e = 1eRG (e)
(G)
= (tr(H) )(e)
X
g
=
(e)
g[G/H]
(( g e))
g[G/H]
g((g 1 eg))g 1
g[G/H]
g(e)g 1
g[G/H]
= trG
H ((e)),
using the fact in the middle that e is central. We take a = (e) and the proof in this
direction is complete.
H
1
Conversely, suppose e = trG
= a for all h H
H a for some a (eRG) . Thus hah
P
1
and e = g[G/H] gag . Now : eRG eRG specified by (x) = ax is an endomorphism of R[(H)]-modules (which is the same thing as an endomorphism of RH-modules),
since
((h, h)x) = ahxh1 = h(h1 ah)xh1 = haxh1 = (h, h)(x).
(G)
We claim that tr(H) = 1eRG . For any x eRG we have

X
(G)
(tr(H) )(x) =
(g, g), ((g 1, g 1 )x)
g[G/H]
(g, g)(g 1xg)
g[G/H]
(g, g)ag 1xg
g[G/H]
gag 1xgg 1
g[G/H]
gag 1x
g[G/H]
= ex
= x.
GG
This shows that eRG (G)
is (H)-projective, and hence since eRG is (G)-projective by
Proposition 12.8 it implies that eRG is in fact (H)-projective.
Peter Webb
237
Putting the last results together we obtain a proof of the following characterization of
the defect group in terms of the effect of the relative trace map on the interior G algebra
eRG. This characterization could have been used as the definition of the defect group.
(12.16) THEOREM. Let e be a block of RG and D a p-subgroup of G. Then D
is a defect group of e if and only if D is a minimal subgroup with the property that
D
trG
(eRG)G is surjective. Equivalently, D is a minimal subgroup with the
D : (eRG)
H
property that e = trG
D a for some a (eRG) .
Proof. From the definition, D is a defect group of e if and only if D is a minimal
subgroup of G such that eRG is (D)-projective. By Lemma 12.15 it is equivalent to say
(G)
that D is minimal such that e can be written e = tr(H) a for some a (eRG)H . With
the understanding that G acts on eRG via (G), we may write this as e = trG
H a. Since
G
H
G
D
tr(eRG) is an ideal of (eRG) , it is equivalent to say that trD : (eRG) (eRG)G is
surjective.
(12.17) COROLLARY. Let e be a block of RG with defect group D. Then every
eRG-module is projective relative to D and hence has a vertex that is a subgroup of D.
Proof. We see from Theorem 12.16 that it is possible to write e = trG
D a for some
D
a (eRG) . Thus by Proposition 12.13 every eRG-module is projective relative to D.
Note that it follows by the remarks about Greens indecomposability theorem at the
end of Chapter 11 that if |G| = |D|pa q where q is relatively prime to p, then pa divides
the dimension of every F G-module and kG-module in the block.
It is a fact (which we will not prove in general) that every block has an indecomposable module with vertex exactly the defect group D, so that the defect group may be
characterized as the unique maximal vertex of modules in the block. In the case of the
principal block the trivial module provides an example of such a module and it allows us
to deduce the next corollary.
(12.18) COROLLARY. The defect groups of the principal block are the Sylow psubgroups of G.
Proof. A defect group is a p-subgroup of G, and it must contain a vertex of the trivial
module, which is a Sylow p-subgroup by 11.24 part (3).
238
We are now in a position to show that our previous use of the term block of defect
zero in Chapter 9 is consistent with the definitions of this section. A block of defect zero
was taken to be a block with a representation satisfying any of the equivalent conditions
of Theorem 9.28. These were seen to be equivalent to the condition that over the field k
the block has a simple projective representation, and also equivalent to the condition that
over k the block is a matrix algebra.
(12.19) COROLLARY. Assume that k is a splitting field for G. A block of kG has
defect zero in the sense of this chapter if and only if it has defect zero in the sense of
Chapter 9, which can be characterized as saying that the block is a matrix algebra over k.
Proof. If the block has defect zero in the sense of this chapter its defect group is 1 and
by Corollary 12.17 every module in the block is 1-projective, or in other words projective.
Thus the block has a simple projective module and by Theorem 9.28 and the comment
immediately after its proof the block has defect zero in the sense considered there.
Conversely, suppose the block has defect zero in the sense of Chapter 9, so that
ekG is a matrix algebra over k. We will show that ekG is projective as a k[G G]module, and will make use of the isomorphism k[G G]
= kG k kG. Let : kG kG
be the algebra anti-isomorphism that sends each group element g to its inverse, so that
P
P
( g g) =
g g 1 . An element x in the second kG factor in the tensor product acts on
ekG as right multiplication by x , and it follows that e e acts as the identity on ekG
(Since e = e). Now e e is a central idempotent in kG k kG that generates the 2-sided
ideal ekG k e kG = ekG k (ekG) . This is the tensor product of two matrix algebras,
since the image of a matrix algebra under an anti-isomorphism is a matrix algebra. Such
a tensor product is again a matrix algebra, for if Eij and Fkl are two matrix algebra bases
(consisting of the matrices that are non-zero in only one place, where the entry is 1), then
the tensors Eij k Fkl are a basis for the tensor product that multiply together in the
manner of a matrix algebra basis. We see that the block ekG k e kG of kG k kG is
semisimple, and so ekG is a projective k[G G]-module. This shows that the defect group
of ekG is 1, and so ekG has defect zero as defined in this chapter.
Peter Webb
239
The Brauer morphism and correspondence of blocks

We now describe Brauers First Main Theorem, which gives a correspondence between
blocks of kG with defect group D and blocks of k[NG (D)] with defect group D. We will
also discover a number of other useful facts, such as Corollary 12.26 which says that if
G has a normal p-subgroup whose centralizer is a p-group then kG only has one block.
The major tool that we introduce to prove these things is the Brauer morphism. This was
originally defined by Brauer in the specific context of idempotents in group rings, but it
has subsequently been realized that the same construction is important more widely. It
applies whenever we have a structure with mappings like restriction, conjugation and the
relative trace map satisfying the usual identities, including the Mackey formula. We first
introduce the Brauer morphism in (not quite such) a general context and then make the
connection with the map as Brauer conceived it.
In what follows we will work over a field k of characteristic p. When U is a kG-module
and K H are subgroups of G we have already made use of the inclusion of fixed points
H
K
U K , the relative trace map trH
resH
U H and the conjugation map
K : U
K : U
g
H
H
cg : U U
for each g G specified by cg (x) = gx. We define for each subgroup
H G the Brauer quotient
U (H) = U H /
K
trH
K (U ).
K<H
G
We also define the Brauer morphism BrG
U (H) to be the composite
H :U
resG
H
H
U G U
U (H)
where the latter map is the quotient homomorphism. Here U G is simply a vector space
but U (H) has further structure: it is a k[NG (H)]-module with the action determined
g
by the maps cg : U H U H . If g NG (H) then g H = H so cg preserves H-fixed
g
K
K
K
points, and since cg trG
= trH
= trH
we see that cg permutes the terms
g K cg U
gK U
KU
H
in the sum being factored from U to produce U (H), so cg has a well-defined action on
U (H). We see also that the image of BrG
points under the action of
H lies in the fixed
P
K
NG (H). If U happens furthermore to be a G-algebra then K<H trH
K (U ) is an ideal of
U H by Proposition 12.12, the Brauer quotient U (H) is an NG (H)-algebra, and the Brauer
morphism is a ring homomorphism.
We will eventually apply these constructions to kG regarded as a G-algebra, in which
case we regard kG as a representation of G via the conjugation action and (kG)G is the
centre of kG. The Brauer quotient provides a means to express properties of the relative
trace map. We already have a characterization in Theorem 12.16 of the defect group of a
block in terms of this map. We now provide a corresponding characterization in terms of
the Brauer quotient. This is preceded by a more technical lemma.
240
(12.20) LEMMA. Let U be an kG-module where k is a field of characteristic p and

let H and J be subgroups of G.
(1) If U (H) 6= 0 then H is a p-group.
G
J
(2) If BrG
H (trJ (a)) 6= 0 for some element a U then H is conjugate to a subgroup of J.
P
Proof. (1) For any group H, if P is a Sylow p-subgroup of H then trH
U H is
P :U
1
H
surjective, because any u U H can be written u = |H:P
| trP u. Thus if H is not a p-group
P
K
then P is a proper subgroup of H and so K<H trH
= U H and U (H) = 0.
K (U ) P
G
G G
H
(2) BrH (trG
J (a)) is the image in U (J) of resH trJ (a) =
g[H\G/J] trH gJ (ga). If this
is not zero in U (J) then for some term in the sum, H gJ must not be a proper subgroup
of H, or in other words H gJ.
The next result provides two more characterizations of the defect group of a block.
Compare parts (2) and (3) with the statement from Theorem 12.16 that D is equivalently
H
a minimal subgroup such that e = trG
D a for some a (ekG) . (Such D is therefore
necessarily unique to up to conjugacy.)
(12.21) THEOREM. Let e be a block of kG and D a subgroup of G. The following
are equivalent.
(1) e has defect group D.
D
(2) e = trG
and BrG
D (e) 6= 0.
D (a) for some element a (kG)
(3) D is up to conjugacy the unique maximal subgroup of G such that BrG
D (e) 6= 0.
Proof. (1) (2) If e has defect group D then D is minimal among groups for which
D
e = trG
by Theorem 12.16. Thus certainly e = trG
D (a) for some a (kG)
D (a). Suppose
P
P
G
D
K
that BrD (e) = 0. then e K<D trK (kG) and we may write e = K<D trD
K (uK ) where
K
uK (kG) . Note that although this expression only suggests that e (kG)D , in fact
e (kG)G . Now
e = ee
X
= (trG
trD
D (a))(
K (uK ))
K<D
= trG
D (a
trD
K (uK ))
K<D
trG
D
trD
K (auK )
K<D
trG
K (auK ).
K<D
P
K
K
Thus e K<D trG
and since e is a primitive idempotent, e trG
for some
K (kG)
K (kG)
K < D, by Rosenbergs Lemma 11.28. This contradicts the minimal property of D and so
the supposition BrG
D (e) = 0 was false.
Peter Webb
241
D
(2) (3) Suppose that e = trG
and BrG
D (e) 6= 0. By Lemma
D (a) for some a (kG)
12.20, if K is any subgroup for which BrG
(e)
=
6
0
then
K
is
a
subgroup
of a conjugate of D,
K
and this shows that D is up to conjugacy the unique maximal subgroup with BrG
D (e) 6= 0.
(3) (1) Suppose that condition (3) holds and let D1 be a defect group of e. By the
implication (1) (3) we know that D1 is up to conjugacy the unique maximal subgroup
for which BrD1 (e) 6= 0. Since D has this property, D and D1 are conjugate, and D is a
defect group.
The relative trace map and the Brauer morphism have the convenient theoretical
properties we have just described, but so far it does not appear to be easy to calculate
with them in specific cases. We remedy this situation by showing that the Brauer morphism
for the module kG acted upon by conjugation has an interpretation in terms of subgroups
of G and group elements. We start with a general lemma about permutation modules.
Since we will apply this in the situation of a subgroup H of G we denote our group by
:= P
H. If is an H-set we write
w as an element of the permutation module k.
(Elsewhere we might have denoted this element , but the bar notation is already in use
here.)
(12.22) LEMMA. Let be an H-set and k the corresponding permutation module.
1, . . . ,
n where =
(1) The fixed point space (k)H has as a basis the H-orbit sums
1 n is a disjoint union of H-orbits.
P
K
(2) If H is a p-group then K<H trH
of (k)H with basis the
K ((k) ) is the subspace
P
H
K
i where |i | > 1. Thus (k)H = k[H ]
orbit sums
K<H trK ((k) ) in this
case. The Brauer quotient k(H) may be identified with k[H ], the span of the fixed
points of the H-set .
Proof. (1) For each transitive H-set i we know that (ki )H is the 1-dimensional
i . From this it follows that if = 1 n where the i are the
space spanned by
orbits of H on then (k)H = (k1 )H (kn )H has as a basis the orbit sums.
(2) Suppose H is a p-group. Since the relative trace map preserves direct sums it
suffices to assume that H acts transitively on , so
H/J for some subgroup J. We
P
P=
H
K
have = trJ (0 ) for any fixed 0 . Thus K<H trH
contains the orbit
K (k)
sums for orbits of size larger than 1. On the other hand, if = {} has size 1 and K < H
then trH
K = |H : K| = 0 since H is a p-group so that |H : K| = 0 in k.
(12.23) COROLLARY. Let G act on kG via conjugation and let H be a p-subgroup
of G. Then kG(H)
= k[CG (H)]. With this identification the Brauer morphism is a
NG (H)
ring homomorphism BrG
that truncates a group ring element
H : Z(kG) k[CG (H)]
P
P
gG g g to
gCG (H) g g.
Proof. In the conjugation action kG is a permutation module and the set of fixed
points of H on G is CG (H). The Brauer morphism may be identified as inclusion of
242
fixed points followed by projection onto the first factor in the decomposition (kG)H =
P
K
k[CG (H)] K<H trH
K ((kG) ) given in Lemma 12.22. Under this identification the Brauer
map is truncation to of support to CG (H).
In older treatments the Brauer map may be defined as the map Z(kG) k[CG (H)]
that truncates a group ring element to have support on CG (H). A disadvantage of this
direct approach is that it is less obvious that the Brauer morphism is a ring homomorphism
and has the other properties we have described. It also ignores the larger context in
which the Brauer quotient and morphism may be defined for any kG-module, and this has
significance beyond the scope of this text. When the module is kG with the conjugation
action, the interpretation of BrG
H as truncation of the support of a group ring element to
CG (H) does provide a concrete understanding of this homomorphism in the context of
blocks. We immediately see the following, for example.
(12.24) COROLLARY. Let e be a block of kG. A defect group of e is a maximal
p-subgroup D of G such that e has some part of its support in CG (D).
Proof. This follows from part (3) of Theorem 12.21 since the condition given is exactly
the requirement that BrG
D (e) 6= 0.
We illustrate with G = S3 and k = F2 where we have blocks e1 = ()+(1, 2, 3)+(1, 3, 2)
and e2 = (1, 2, 3) + (1, 3, 2). When H = h(1, 2)i we have CG (H) = H and BrG
H (e1 ) = (),
G
G
BrG
(e
)
=
0.
We
also
have
Br
(e
)
=
e
and
Br
(e
)
=
e
showing
that
h(1,
2)i and 1
1
1
2
2
H 2
1
1
G
are (respectively) the largest subgroups H (up to conjugacy) for which BrH (e1 ) 6= 0 and
BrG
H (e2 ) 6= 0. These subgroups are the defect groups of the blocks, as asserted by Corollary
12.18, Example 7.5 and the discussion after 12.7.
Notice that, as with Galois correspondence, for each containment of subgroups K H
we have the reverse containment of fixed points (kG)K (kG)H and also a containment
of centralizers CG (K) CG (H). This
means that we obtain a commutative diagram
resG
H
G
G
(kG)H kG(H) = kCG (H)
BrH : (kG)
k
yresK
y
G
BrG
K : (kG)
resG
K
(kG)K
kG(K) = kCG (K)
G
where is the inclusion of the centralizer group rings, so that BrG
K is the composite BrH .
G
G
We see from this that if BrH (x) 6= 0 for some x (kG)G then BrK (x) 6= 0 for every K
G
H, since BrG
H (x) equals BrK (x) with its support truncated to CG (H). This observation
provides a strengthening of part (3) of Theorem 12.21: if K is any subgroup of a defect
group D of a block e of kG, then BrG
K (e) 6= 0.
The next result is important and interesting in its own right. It will be used when
we come to describe the Brauer correspondence of blocks. The statement of this result is
given in terms of Op (G), the largest normal p-subgroup of G. Although Op (G) could, in
principle, be 1, the result gives no information in that case, so it is really a result about
groups with a non-trivial normal p-subgroup.
Peter Webb
243
(12.25) PROPOSITION. Let Q = Op (G) is the largest normal p-subgroup of G.

Then every block of G lies in k[CG (Q)] and is the sum of a G-orbit of blocks of CG (Q).
P
K
Proof. We show first that K<Q trQ
Rad((kG)Q ). Observe that if S is a
K (kG)
simple kG-module then Q acts trivially on S by Corollary 6.4. Thus if K < Q, a (kG)K
P
P
1
u =
and u S we have trQ
K (a) u =
g[Q/K] gag
g[Q/K] au = |Q : K|au = 0.
P
Q
K
annihilates every simple kG-module and so is contained in the
Thus
K<Q trK (kG)
P
K
is a nilpotent ideal, and so is contained in
radical of kG. It follows that K<Q trQ
K (kG)
Q
Rad((kG) ).
If e is a central idempotent of kG then both e and BrQ (e) are idempotents of (kG)Q
that map under the quotient homomorphism (kG)Q kG(Q) = k[CG (Q)] to BrQ (e).
Since the kernel of this homomorphism is nilpotent it follows that e and BrQ (e) are conjugate in (kG)Q , by Exercise 2 of Chapter 11. (The argument is that the kernel I is contained
G
in Rad((kG)Q ) so (kG)Q e and (kG)Q BrG
Q (e) are both projective covers of kG(Q) BrQ (e),
and hence are isomorphic. For a similar reason (kG)Q (1 e) and (kG)Q (1 BrG
Q (e)) are
Q
Q
Q
isomorphic so there is an automorphism of (kG) sending (kG) e to (kG) BrG
Q (e) and
G
Q
Q
Q
(kG) (1 e) to (kG) (1 BrQ (e)). Now (x) = x for some unit (kG) and it has
G
Q
1
the property that ue = uBrG
.) Since e is
Q (e) for all u (kG) . Hence e = BrQ (e)
central its only conjugate is e. Thus e = BrQ (e), and this lies in k[CG (Q)].
We can now write e = f1 + + fn as a sum of primitive central idempotents fi of
k[CG (Q)], and since e is stable under conjugation by G the f1 , . . . , fn must be a union of
G orbits in the conjugation action on the blocks of k[CG (Q)]. However the sum of a single
G orbit of the fi is already a central idempotent of kG and e is the sum of such sums, so
if e is a block it must be the sum of a single G-orbit of the fi .
(12.26) COROLLARY. If there is a normal p-subgroup Q of G for which CG (Q) is a
p-group (for example, if CG (Q) Q) then G has only one p-block.
Proof. Any block of kG must lie in k[CG (Q)] by Proposition 12.25; but this is the
group ring of a p-group that has only one block, so kG also has only one block.
The above corollary implies, for example, that if K is a subgroup of Aut(Q) where Q
is a p-group then the semidirect product Q K has only one p-block. Such is the case with
the symmetric group S4 at p = 2 on taking Q to be the normal Klein four-group, and there
are many other examples of this phenomenon including, including dihedral groups Cp C2
in characteristic p, and the non-abelian group C7 C3 of order 21 in characteristicc 7, and
so on.
We now define the Brauer correspondence of blocks. We will define it whenever b is
a block of kJ, where J is a subgroup of G satisfying HCG (H) J NG (H) for some
p-subgroup H of G. In this situation the Brauer morphism is a ring homomorphism BrG
H :
NG (H)
Z(kG) (k[CG (H)])
, and the latter ring is contained in Z(kJ) since CG (H) J,
244
and elements that are fixed under NG (H) are also fixed under J. If 1 = e1 + + en is
G
the sum of blocks of kG then 1 = BrG
H (e1 ) + + BrH (en ) is a sum of orthogonal central
G
idempotents of kJ. Thus if b is a block of kJ then b = bBrG
H (e1 ) + + bBrH (en ) is a
decomposition of b as a sum of orthogonal idempotents in Z(kJ), and since b is primitive
G
in Z(kJ) there is a unique block ei of G so that bBrG
H (ei ) = b. We write b for this block
ei , and call it the Brauer correspondent of b. There may be several blocks b of kJ with the
same Brauer corresponding block of kG: bG = bG . We see that the blocks of kG partition
the blocks of kJ by this means.
There was a choice of p-subgroup H in the definition of bG , since it would be possible
to have another p-subgroup H1 with H1 CG (H1 ) J NG (H1 ), and perhaps the definition
of bG would be different using H1 . In fact the choice of H does not matter, as we now
show.
(12.27) PROPOSITION. If b is a block of kJ, the definition of bG is independent of
the choice of the p-group H satisfying HCG (H) J NG (H).
Proof. Since H is a normal p-subgroup of J we have H Op (J) and CG (H)
CJ (Op (J)). By Proposition 12.25 every central idempotent of kJ lies in k[CG (Op (J))].
For each block ei of kG the idempotent BrG
H (ei ) thus lies in k[CG (Op (J))] and could have
been computed by truncating the support of ei to lie in CJ (Op (J)) instead of CG (H). This
G
identification of BrG
H (ei ) is independent of the choice of H and hence so is b .
(12.28) PROPOSITION. Let H be a p-subgroup of G, J a subgroup of G with
HCG (H) J NG (H) and b a block of kJ. Then bG has a defect group that contains a
defect group of b.
Proof. If D J is a defect group of b then D H by Corollary 12.11 since H is a
normal p-subgroup of J, and so CG (H) CG (D) = CJ (D) since J CG (H). Let us write
G
BrG
H (b ) = b + b1 where b and b1 are orthogonal central idempotents of kJ. We can now
apply BrJD to this, and since this map truncates to CJ (D) we get the same thing as if we
J
J
G G
J
G G
had originally applied BrG
D . Thus BrD (b ) = BrD (BrH (b )) = BrD (b) + BrD (b1 ) is a sum
G
of orthogonal idempotents in k[CJ (D)]. Since BrJD (b) 6= 0 it follows that BrG
D (b ) 6= 0,
and hence D is contained in a defect group of bG by Theorem 12.21.
(12.29) LEMMA. Let H be a subgroup of G and U a kG-module. Then
N (H)
G
G
BrG
H trH = trH
H
BrH
U (H)NG (H) .
H :U
P
H
G
G G
Proof. BrG
H trH (x) is the image in U (H) of resH trH (x) =
g[H\G/H] trH gH (gx).
g
The only terms that contribute have H H = H, which happens if and only if g NG (H),
P
N (H)
so the image equals the image of g[NG (H)/H] gx = trHG x.
Peter Webb
245
The following is a basic version of Brauers first main theorem.

(12.30) THEOREM (Brauers first main theorem). Let D be a p-subgroup of G.
The Brauer morphism induces a bijection between blocks of kG with defect group D and
blocks of NG (D) with defect group D with inverse given by the Brauer correspondent.
Proof. Let us write N for NG (D). We prove that if e is a block of kG with defect
group D then BrG
D e is a block of kN with defect group D; and if b is a block of kN with
G
defect group D then bG is a block of kG with defect group D. Furthermore (BrG
D e) = e
G
and BrG
D (b ) = b.
Let b Z(kN ) have defect D. Then by Theorem 12.16 b = trN
D (a) for some
X
K
a (kN )D = k[CG (D)]
trD
K (kN ) .
K<D
Since trN
D preserves both of the two summands on the right and b k[CG (D)] by PropoD
N
sition 12.25, we may assume a k[CG (D)] and so BrD
D (a) = a. Thus b = trD BrD (a)
trN
D (k[CG (D)]).
Let e = bG Z(kG) be the Brauer correspondent of b, so that b = bBrG
D (e) is a
G
summand of BrD (e) and e has a defect group D1 D by Proposition 12.28. We will show
that D1 = D. Now using Lemma 12.29
D
G
D
N
D
N
BrG
D (trD ((kG) )) = trD BrD ((kG) ) = trD (k[CG (D)])
and so, since BrG

D is a ring homomorphism,
G
G
G
G
D
G
D
N
BrG
D (etrD ((kG) )) = BrD (e) BrD (trD ((kG) )) = BrD (e) trD (k[CG (D)]),
D
which contains b. Thus the ideal etrG
D ((kG) ) of eZ(kG) is not nilpotent, since it has
an image under a ring homomorphism that contains a non-zero idempotent. It follows
D
that etrG
D ((kG) ) = eZ(kG), since eZ(kG) is local by Proposition 11.4. This implies,
D
G
D
G
since trG
D (e(kG) ) = etrD ((kG) ), that e lies in the image of trD and so has defect group
contained in D. This completes the argument that the defect group of e equals D.
G
We have also just seen that BrG
D (eZ(kG)) contains b, and it also contains BrD (e). It
is an image of a local ring and hence is local and contains only one non-zero idempotent. It
G
follows that b = BrG
D (e). This shows us that b 7 b is a one-to-one mapping from blocks
of kN with defect group D to blocks of kG with defect group D, and that its inverse
on one side is BrG
D . We conclude by observing that this mapping is surjective. For, if
e Z(kG) is a block with defect group D then BrG
D (e) is a sum of blocks b of kN for
which e = bG , and the blocks b have defect groups that are subgroups of D by Proposition
12.28. On the other hand every block of kN has defect group containing D, since D is a
normal p-subgroup of N . It follows that e = bG for some block of kN with defect group
D, showing that b 7 bG is surjective with the domain and codomain as specified above.
This completes the proof.
246
Examples. 1. When G = S3 and k = F2 we have seen (in the discussion after

12.7 and elsewhere) that e1 = () + (1, 2, 3) + (1, 3, 2) has defect group h(1, 2)i and e2 =
(1, 2, 3) + (1, 3, 2) has defect group 1. Here BrG
h(1,2)i (e1 ) = () kh(1, 2)i, so that the Brauer
G
correspondent () of the only idempotent in kh(1, 2)i is e1 , this giving a bijection between
the idempotents of S3 and kh(1, 2)i with defect group h(1, 2)i.
2. Let G = A5 and let k be a splitting field of characteristic 2. A Sylow 2-subgroup
P = C2 C2 has NG (P )
= A4 , and we have seen in exercise 2 from Chapter 8 that kA4 has
only one block. Thus there is only one block of kG with defect group P : it is the principal
block. The subgroups C2 have Sylow p-subgroups P as their normalizers and there are no
blocks of kP with C2 as defect group, since P is a 2-group. Hence there are no blocks of
kG with this defect group. This duplicates information we know from a different source,
to the effect that C2 cannot be a defect group by Corollary 12.11 because it is not O2 of
its normalizer. Finally there remain the blocks of defect zero of kA5 , which have defect
group 1. There is in fact just one of these, as may be seen by inspecting the character
table of A5 for characters of degree divisible by 4.
1. Blocks correspond to blocks of the Cartan matrix, indecomposable ring summands
of kG or RG, primitive central idempotents in kG or RG, and certain equivalence classes
of representations in characteristic 0 or in characteristic p.
2. The defect group of a block may also be characterized in several ways, using the
bimodule structure of the block as a ring summand, the relative trace map, the Brauer
morphism, and the vertices of modules in the block.
3. A defect group is always the intersection of two Sylow p-subgroups and is the
largest normal p-subgroup of its normalizer.
4. The principal block has Sylow p-subgroups as defect groups.
5. The Brauer correspondent bG provides a bijection between blocks of G and of
NG (D) with defect group D.
Peter Webb
247

1. Prove that if a module U lies in the principal block then so does U .
2. Use the results of Chapter 10 Exercises 3 and 5 to show that the simple group
GL(3, 2) of order 168 has two blocks in characteristic 2 and two blocks in characteristic 7.
3. Suppose that Q1 and Q2 are subgroups of G that satisfy Qi = Op (NG (Qi )) for
i = 1, 2, and let Ni = NG (Qi ) for each i. Prove that if N1 N2 then Q1 Q2 . Deduce
that Qi Op (G) for each i.
4. Let A be an algebra and let U and V be A-modules that lie in different blocks of
A. Show that
(1) HomA (U, V ) = 0, and
(2) every short exact sequence of A-modules 0 U W V 0 is split.
5. Let (F, R, k) be a complete p-modular system and G a finite group. Let the
decomposition number dT S be the multiplicity of the simple kG-module S as a composition
factor of the reduction modulo of the simple F G-module T . Show that:
(1) The simple F G-modules T for which dT S 6= 0 for some simple module S in a fixed
block of kG constitute the simple modules in a p-block of F G-modules.
(2) The simple kG-modules S for which dT S 6= 0 for some simple module T in a fixed
block of F G constitute the simple modules in a block of kG-modules.
6. Let (F, R, k) be a splitting p-modular system for G. Let e be a block idempotent
of kG with defect group D, and suppose g G is an element that centralizes a p-power
element of G that does not lie in any conjugate of D.
(1) Use the characterization of D in terms of the Brauer morphism to show that g does
not lie in the support of e.
(2) Suppose furthermore that e is a block of defect 0 corresponding to an ordinary character . Show that (g) (), the maximal ideal of R. [Use the formula in Theorem
3.23 that gives the block idempotent in RG in terms of the characters of G, and that
reduces to e.]
248
Appendix: Discrete valuation rings

Let F be a field. A (multiplicative) valuation on F is a mapping : F R0 such
that
(a) = 0 if and only if a = 0,
(ab) = (a)(b) for all a, b F , and
(a + b) (a) + (b) for all a, b F .
In many texts the theory is developed in terms of additive valuations. A suitable reference
which uses the multiplicative language is [Hua].
(A.1) Examples. 1. No matter what the field F is, we always have the valuation
(a) =
0 if a = 0
1 otherwise.
This valuation is the trivial valuation, and we generally exclude it.

2. If F is any subfield of the field of complex numbers we may take (a) = |a|, the
absolute value of a.
3. Let F = Q and pick a prime p. Every rational number a Q may be written a = rs
where (r, s) = 1. We set
p (a) =
power to which p divides r

power to which p divides s
so that
a = pp (a)
if a = 0,
if p r,
otherwise,
r
s
where (r , p) = 1 = (s , p). Now let be any real number with 0 < < 1 and put
(a) = p (a) . Often is taken to be p1 , but the precise choice of does not affect the
properties of the valuation. This valuation is called the p-adic valuation on Q.
This last is an example of a valuation which satisfies the so-called ultrametric inequality
(a + b) max{(a), (b)},

which in the case of this example comes down to the fact that if pn a and pn b
where a, b Z, then pn (a + b). We say that is non-archimedean if it satisfies the
ultrametric
inequality. The valuations in the third example are also discrete, meaning that

{(a) a K, a 6= 0} is an infinite cyclic group under multiplication. It is the case that
discrete valuations are necessarily non-archimedean.
We deduce from the axioms for a valuation that (1) = (1) = 1. Using this we
see that every valuation gives rise to a metric d(a, b) = (a b) on the field F . We
say that two valuations are equivalent if and only if the metric spaces they determine are
equivalent, i.e. they give rise to the same topologies. In Example 3 above, changing the
value of between 0 and 1 gives an equivalent valuation.
Peter Webb
249
(A.1) THEOREM (Ostrowski). Up to equivalence, the non-trivial valuations on Q

are the ones just described, namely the usual absolute value and for each prime number a
non-archimedean valuation.
More generally, if R is a ring of algebraic integers (or more generally a Dedekind
domain) with quotient field F , the non-archimedean valuations on R, up to equivalence,
biject with the maximal ideals (which are the same as the non-zero prime ideals) of R.

Let be a non-archimedean valuation on a field F . The set R = {a F (a) 1}
is a ring, called the valuation ring of , and any ring arising in this way for some is
called a valuation ring. If the valuation is discrete the ring is called a discrete valuation
ring. We set P = {a F (a) < 1}, and this is evidently an ideal of R . For example,
if F = Q and is the non-archimedean valuation corresponding to the prime p then R is
the localization of Z at p, and P is its unique maximal ideal.
(A.2) PROPOSITION. Let be a discrete valuation on a field F with valuation ring
R and valuation ideal P .
(1) An element a R is invertible if and only if (a) = 1.
(2) P is the unique maximal ideal of R , consisting of the non-invertible elements.
(3) P = () where is any element such that () generates the value group of .
(4) Every element of R is uniquely expressible a = a where a is a unit in R . In this
situation (a) = () .
(5) The ideals of R are precisely the powers Pn = ( n ). Thus R is a principal ideal
domain.
(6) F is the field of fractions of R .
Proof. The proofs are all rather straightforward. If ab = 1 where a, b R then
(ab) = (a)(b) = (1) = 1 and since (a) and (b) are real numbers between 0 and 1
it follows that they equal 1. Conversely, if (a) = 1 let b be the inverse of a in F . Since
(b) = 1 also it follows that b R and a is invertible in R . This proves part (1).
Now P is seen to consist of the non-invertible elements of R . It is an ideal, and
it follows that it is the unique maximal ideal. Defining to be an element for which
() generates the value group of we see that P . If a P is any element then
(a) = () for some and so ( a) = 1. Thus a = a for some element a R
which is a unit, and so a = a . This proves that P = (), and also the first statement
of (4). The uniqueness of the expression in (4) comes from the fact that in any expression
a = a with a a unit, necessarily is defined by (a) = () , and then a is forced to
be a.
To prove (5), if I is any ideal of R we let n be minimal so that I contains a non-zero
element a with (a) = ()n . Then by (4) we have (a) = ( n ) I, and it follows that
I = ( n ).
250
As for (6), given any element a F , either a R or (a) = ()n for some n > 0.
In the second case the element a = n a has (a ) = 1 so a R and now a = an , showing

that a lies in the field of fractions of R in both cases.
When we reduce representations from characteristic zero to positive characteristic we
need to work with algebraic number fields, that is, field extensions of Q of finite degree.
Let F be an algebraic number field, and R its ring of integers. We quote without proof
some facts about this situation. A full account may be found in [Hua], [Cas] or [FT]. A
fractional ideal in F is a finitely-generated R-submodule I of F . For any such I we put
I 1 = {x F xI F }. With this definition of inverse and with a multiplication defined
the same way as the multiplication of ideals, the fractional ideals form a group, whose
identity is R. Every fractional ideal may be written uniquely as a product I = pa1 1 pat t
where p1 , . . . , pt are maximal ideals of R and the ai are non-zero integers (which may be
positive or negative). Let us write pi (I) = ai , and let 0 < < 1. Then for each maximal
ideal p of R we obtain a discrete valuation on F by putting (a) = p (Ra) , which is called
the p-adic valuation on F .
(A.3) PROPOSITION. Let F be an algebraic number field with ring of algebraic
integers R, and let be the discrete valuation on F associated to a maximal ideal p of R.
Let Rp be the valuation ring of with maximal ideal Pp . Then Rp is the localization of R
at p, and the inclusion R Rp induces an isomorphism R/p
= Rp /Pp .
Proof. We assume the group structure of the set of fractional ideals. The localization
of R at p is
a
{ a, b R, b 6 p}
b
and this is clearly contained in Rp . Conversely, if ab Rp then p (a) p (b) and if we
a
choose x pp (b) pp (b)+1 then ab = ax
bx . Now bx 6 p, showing that b lies in the
localization.
The kernel of the composite R Rp Rp /Pp is R Pp = p. We show that this
composite is surjective. We can write any element of Rp /Pp as ab + Pp where a, b R and
b 6 p. Since p is maximal in R, R/p is a field and so there exists c R with bc1 p. Now
a
ac = ab (1 bc) Pp and ab + Pp = ac + Pp is the image of ac R. These observations
b
show that we have an isomorphism R/p
= Rp /Pp .
Given a valuation on F we may form the completion F as a metric space, which

contains F in a canonical way. The completion F acquires a ring structure extending that
of F , and in fact F is a field. The valuation extends uniquely to a valuation on F ,
If is non-archimedean then so is ,
and
and F is complete in the metric given by .
with the same value group. Thus in the case of a discrete
if is discrete then so is ,

= {a F (a)
1} with unique maximal ideal
valuation we have a valuation ring R

P = {a F (a) < 1}, and P = R () since () = () generates the value group.
Peter Webb
251
etc, but this seems excessive.) The ideals of R

are exactly
(We should properly write R
the powers Pn .
is denoted Qp and is called
When is the p-adic valuation on Q, the completion Q
with respect to is denoted Zp and is
the field of p-adic rationals. The valuation ring Q
called the ring of p-adic integers.
(A.4) LEMMA. Let be a discrete valuation on a field F with valuation ring R .
induces an isomorphism R /P n
n
The inclusion R R
= R /P for all n.
R
/P n . Its kernel is
Proof. Consider the composite homomorphism R R
Pn and the desired isomorphism will follow if we can show that this homomorphism is
we know from the construction of the completion
surjective. To show this, given a R
that there exists b R with (b a) < ()n , that is, b a Pn . Now b maps to
a + Pn .
is by definition the set of equivalence classes of Cauchy sequences
The completion R
in R . We comment that a sequence (ai ) of elements of R is a Cauchy sequence if and
only if for every n there exists a number N so that whenever i, j > N we have ai aj Pn ,
that is, ai aj (mod Pn ).
(A.5) LEMMA. Let be a discrete valuation on a field F with valuation ring R ,
. Any element of R
is uniquely expressible as a series
maximal ideal P and completion R
a = a0 + a1 + a2 2 +
where the ai lie in a set of representatives S for R /P .
. Since R
/P
Proof. Let a R
= R /P , we have a+ P = a0 + P for some uniquely
. Repeating this
determined a0 S. Now a a0 P so a = a0 + b1 for some b1 R
construction we write b1 = a1 + b2 with a1 S uniquely determined, and in general
bn = an + bn+1 with an S uniquely determined. Now a0 , a0 + a1 , a0 + a1 + a2 2 , . . .
is a Cauchy sequence in R whose limit is a, and we write this limit as the infinite series.
The last result, combined with Proposition A.3, provides a very good way to realize
. For example, in the case of the p-adic valuation on Q we may take
the completion R
= Zp may be realized as the set of infinite
S = {0, 1, . . . , p 1}. The completion Z
sequences a3 a2 a1 a0 . of elements from S presented in positions to the left of a point,
analogous to the decimal point (which we write on the line following American convention).
Thus a0 is in the 1s position, a1 is in the ps position, a2 is in the p2 s position, and so
on. Unlike decimal numbers these strings are potentially infinite to the left of the point,
whereas decimal numbers are potentially infinite to the right of the point. Addition and
multiplication of these strings is performed by means of the same algorithms (carrying
252
values from one position to the next when p is exceeded, etc.) that are used with infinite
decimals. Note that p-adic integers have the advantage over decimals that certain real
numbers have more than one decimal representation, whereas distinct p-adic expansions
always represent distinct elements of Zp .
Exercises.
1. With the description of the p-adic integers as the set of infinite sequences
a3 a2 a1 a0 .
in positions to the left of a point, where ai {0, . . . , p 1}, show that when p = 2 we
have
1 = 1111. and
1
= 10101011.
3
Find the representation of the fraction 1/5 in the 2-adic integers. What fraction does
1100110011. represent?
2. Show that the field of p-adic rationals Qp may be constructed as the set of sequences
a3 a2 a1 a0 .a1 a2 an which may be infinite to the left of the point, but must be finite
to the right of the point, where ai {0, . . . , p 1} for all i. Show that the field of rational
numbers Q is the subset of these sequences which eventually recur.
Peter Webb
253
Appendix: some character tables

In studying characters our goal has been to become proficient in the basic methods
of constructing character tables to give an understanding of how it may be done, but not
necessarily to calculate large numbers of character tables. In reality, someone who wishes
to know a character table would probably obtain it from a book or a computer source.
Nevertheless it is useful to have a list of some important character tables. We collect here
the character tables that have been studied in this book. We use the notation n = e2i/n
for a primitive nth root of unity. Where a group is isomorphic to one of the groups SL(2, p)
or P SL(2, p) we have emphasized this, because there is a special construction of simple
modules over Fp , described in Section 6 Exercise 21.
Cyclic groups.

We let Cn = hx xn = 1i.
Characteristic 0:
Cn
ordinary characters
xn1
n
g
|CG (g)|
1 x
n n
ns
(0 s n 1)
1 ns n
s(n1)
This table was described in Proposition 4.1. The representations in positive characteristic
were described for abelian groups in Example 8.6 and the results which followed.
The symmetric group S3 .
This group is isomorphic to the dihedral group D6 and also to GL(2, 2) = SL(2, 2)
Characteristic 0:
S3
ordinary characters
g
|CG (g)|
() (12) (123)
6
2
3
1
sign
2
1
1
1 1
2
0
1
1
1
254
Characteristic 2:
S3
= SL(2, 2)
Brauer simple p = 2
S3
= SL(2, 2)
g
|CG (g)|
() (123)
6
3
g
|CG (g)|
() (123)
6
3
1
2
1
1
2 1
1
2
2
2
2 1
S3
= SL(2, 2)
Decomposition matrix p = 2
S3
= SL(2, 2)
Cartan matrix p = 2
1 2
1 2
1
sign
2
1
1
0
0
0
1
1
2
2
0
0
1
Characteristic 3:
S3
Brauer simple p = 3
S3
g
|CG (g)|
() (12)
6
2
g
|CG (g)|
() (12)
6
2
1
sign
1
1
1 1
1
sign
3
1
3 1
S3
S3
Cartan matrix p = 3
1 sign
1 sign
1
sign
2
1
0
1
0
1
1
1
sign
2
1
1
2
The dihedral and quaternion groups of order 8.

If we put D8 = hx, y x4 = y 2 = 1, yxy 1 = x1 i and Q8 = hx, y x4 = 1, x2 =
Peter Webb
255
y 2 , yxy 1 = x1 i then in both cases the character table is

D8 and Q8
ordinary characters
g
|CG (g)|
1 x2 x y xy
8 8 4 4 4
1
1a
1b
1c
2
1 1 1 1 1
1 1 11 1
1 1 1 1 1
1 1 11 1
2 2 0 0 0
Construction notes: to determine the conjugacy classes it is convenient simply to consider

a list of the elements in the group {1, x, x2 , x3 , y, xy, x2y, x3 y} and explicitly calculate the
effect of conjugacy. In both cases the derived subgroup is hx2 i with a C2 C2 quotient,
so that the four 1-dimensional characters look the same in both cases. There remains
a fifth character which is determined by orthogonality relations, so the character tables
must be the same. In the case of D8 the 2-dimensional character is also obtained from the
natural construction of D8 as a the group of symmetries of a square. In the case of Q8
it is obtained from the action on the quaternion algebra, as in Chapter 2 Exercise 11. In
both cases the 2-dimensional character is induced from any linear character of a subgroup
of index 2 which does not have x2 in its kernel.
The alternating group A4 .
A4
ordinary characters
g
|CG (g)|
() (12)(34) (123) (132)

12
4
3
3
1
1a
1b
3
1
1
1
3
1
1
1
1
1
3
32
0
1
32
3
0
Characteristic 2:
A4
Brauer simple p = 2
A4
g
() (123) (132)
|CG (g)| 12 3
3
g
() (123) (132)
|CG (g)| 12 3
3
1
1a
1b
1
1a
1b
1
1
1
1
3
32
1
32
3
4
4
4
1
3
32
1
32
3
A4
A4
Cartan matrix p = 2
1 1a 1b
1 1a 1b
1
1a
1b
3
1
0
0
1
0
1
0
1
0
0
1
1
256
1
1a
1b
2
1
1
1
2
1
1
1
2
Characteristic 3:
A4
Brauer simple p = 3
A4
g
|CG (g)|
() (12)(34)
12
4
g
|CG (g)|
() (12)(34)
12
4
1
3
1
3
1
3
3
3
1
1
A4
A4
Cartan matrix p = 3
1 3
1 3
1
1a
1b
3
1
1
1
0
0
0
0
1
3
1
1
3
3
0
0
1
Construction notes for A4 : the element (123) conjugates transitively the three nonidentity elements of order two in the unique Sylow 2-subgroup h(12)(34), (13)(24)i. The
remaining group elements have order three, conjugated in two orbits by the Sylow 2subgroup, which equals the derived subgroup. There are three degree 1 characters. The
remaining character can be found from the orthogonality relations; it can be constructed
by inducing a non-trivial character from h(12)(34), (13)(24)i; it is also the character of the
realization of A4 as the group of rotations of a regular tetrahedron.
The dihedral and quaternion groups of order 16.
Put

D16 = hx, y x8 = y 2 = 1, yxy 1 = x1 i,

Q16 = hx, y x8 = 1, x4 = y 2 , yxy 1 = x1 i.
Peter Webb
257
In both cases the character table is

D16 and Q16
ordinary characters
g
|CG (g)|
1 x4 x2
16 16 8
1
1a
1b
1c
2a
2b
2c
1 1 1 1
1 1 1
1 1 1 1
1 1 1
1 1 1 1 1 1 1
1 1 1 1 1 1 1
2 2 2 0 0 0 0
2 2 0 22 0 0
2 2 0 2
2 0 0
x
8
x5 y xy
8 4 4
Construction notes: as with D8 and Q8 , find the conjugacy classes by listing the elements.
The quotient by hx4 i is a copy of either D8 or Q8 (depending on the case), and we obtain
the top 5 rows of the character table by lifting (or inflating) the characters from the
quotient group. The final two characters are obtained by inducing the characters 8 and
83 from the cyclic subgroup hxi.
The semidihedral group of order 16.

The character table of the group SD16 = hx, y x8 = y 2 = 1, yxy 1 = x3 i is
SD16
ordinary characters
g
|CG (g)|
1 x4 x2
16 16 8
1
1a
1b
1c
2a
2b
2c
1 1 1
1
1 1 1
1 1 1
1
1 1 1
1 1 1 1 1 1 1
1 1 1 1 1 1 1
2 2 2
0
0 0 0
2 2 0 i2i2 0 0
2 2 0 i 2 i 2 0 0
x
8
x5
8
y xy
4 4
Construction notes: the same as for D16 and Q16 . The quotient by hx4 i is a copy of D8 ,
and we obtain the top 5 rows of the character table by inflating (or lifting) the characters
from the quotient group. The final two characters are obtained by inducing the characters
8 and 1 from the cyclic subgroup hxi.
8
The nonabelian group C7 C3 of order 21.
258

The representations of C7 C3 = hx, y x7 = y 3 = 1, yxy 1 = x2 i in characteristic 7
were explicitly described after Proposition 8.9.
C7 C3
ordinary characters
g
|CG (g)|
1
21
1
1a
1b
3a
3b
1
1
1
3
3
x1
7
x
7
y y 1
3
3
1
1
1
1
1
3
1
1
32
2
4
3
5
6
7 + 7 + 7 7 + 7 + 7 0
73 + 75 + 76 7 + 72 + 74 0
1
32
3
0
0
Characteristic 3:
C7 C3
Brauer simple p = 3
g
1
|CG (g)| 21
1
3a
3b
C7 C3
x1
7
x
7
1
1
1
2
4 3
3 7 + 7 + 7 7 + 75 + 76
3 73 + 75 + 76 7 + 72 + 74
C7 C3
g
1
|CG (g)| 21
1
3a
3b
3
3
3
2
4 3
3 7 + 7 + 7 7 + 75 + 76
3 73 + 75 + 76 7 + 72 + 74
C7 C3
Cartan matrix p = 3
1 3a 3b
1 3a 3b
1
1a
1b
3a
3b
1
1
1
0
0
0
0
0
1
0
0
0
0
0
1
x1
7
x
7
1
3a
3b
3
0
0
0
1
0
0
0
1
Characteristic 7:
C7 C3
Brauer simple p = 7
C7 C3
g
1 y y 1
|CG (g)| 21 3 3
g
1 y y 1
|CG (g)| 21 3 3
1
1a
1b
1
1a
1b
1 1 1
1 3 32
1 32 3
7 1 1
7 3 32
7 32 3
Peter Webb
259
C7 C3
Cartan matrix p = 7
C7 C3
1 1a 1b
1 1a 1b
1
1a
1b
3a
3b
1
0
0
1
1
0
1
0
1
1
1
1a
1b
0
0
1
1
1
3
2
2
2
3
2
2
2
3
Construction notes: the two ordinary characters of degree 3 are induced from the characters
7 and 73 of the cyclic subgroup of order 7. In characteristic 3 they are blocks of defect
zero, so remain simple on reduction. In characteristic 7 the indecomposable projectives
were already constructed after 8.9, giving the Cartan matrix by a different method. In
the context of characters we can simply say that the three 1-dimensional characters are all
distinct on the 7-regular classes so give the three simple Brauer characters.
S4
ordinary characters
g
|CG (g)|
() (12) (12)(34) (1234) (123)

24 4
8
4
3
1
sign
2
3a
3b
1
1
1 1
2
0
3 1
3
1
1
1
2
1
1
1
1
0
1
1
1
1
1
0
0
Construction notes: this table was constructed in Chapter 3. It (and tables for the other
symmetric groups) can also be constructed using the combinatorial methods available for
representations of the symmetric groups, not described in this text.
Characteristic 2:
S4
Brauer simple p = 2
S4
g
|CG (g)|
() (123)
24
3
g
|CG (g)|
() (123)
24
3
1
2
1
2
1
2
8
8
1
1
2
1
260
S4
Cartan matrix p = 2
S4
1 2
1 2
1
sign
2
3a
3b
1
1
0
1
1
1
2
0
0
1
1
1
4
2
2
3
Characteristic 3:
S4
Brauer simple p = 3
S4
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
g
|CG (g)|
() (12) (12)(34) (1234)

24 4
8
4
1
sign
3a
3b
1
1
1 1
3 1
3
1
1
sign
3a
3b
3
1
3 1
3 1
3
1
1
1
1
1
1
1
1
1
S4
1
0
1
0
0
0
1
1
0
0
0
0
0
1
0
1 sign 3a 3b
1
sign
3a
3b
0
0
0
0
1
2
1
0
0
1
2
0
0
The special linear group SL(2, 3)

There is an isomorphism SL(2, 3)
= Q8 C3 . We let
a=
0
1
1
1
1
1
S4
Cartan matrix p = 3
1 sign 3a 3b
1
sign
2
3a
3b
3
3
1
1
1
0
and
y=
1
0
1
1
0
0
1
0
0
0
0
1
Peter Webb
261
with entries in F3 . These matrices have orders 4 and 3, respectively.

SL(2, 3)
ordinary characters
g
|CG (g)|
1 a2 a
24 24 4
1
1a
1b
2a
2b
2c
3
1 1 1 1
1
1 1 1 3
32
1 1 1 32
3
2 2 0 1 1
2 2 0 3 32
2 2 0 32 3
3 3 1 0
0
y
6
y2
6
ya2 y 2 a2
6
6
1
3
32
1
3
32
0
1
32
3
1
32
3
0
Characteristic 2:
SL(2, 3)
Brauer simple p = 2
SL(2, 3)
g
1 y y 1
|CG (g)| 24 6 6
g
1 y y 1
|CG (g)| 24 6 6
1
1a
1b
1
1a
1b
1 1 1
1 3 32
1 32 3
SL(2, 3)
SL(2,3)
Cartan matrix p = 2
1 1a 1b
1 1a 1b
1
1a
1b
2a
2b
2c
3
Characteristic 3:
1
0
0
0
1
1
1
0
1
0
1
0
1
1
0
0
1
1
1
0
1
8 2 2
8 23 232
8 232 23
1
1a
1b
4
2
2
2
4
2
2
2
4
262
SL(2, 3)
Brauer simple p = 3
SL(2, 3)
g
1 a2 a
|CG (g)| 24 24 4
g
1 a2 a
|CG (g)| 24 24 4
1
2
3
1
2
3
1 1 1
2 2 0
3 31
SL(2, 3)
SL(2,3)
Cartan matrix p = 3
1 2 3
1 3 5
1
1a
1b
2a
2b
2c
3
1
1
1
0
0
0
0
0
0
0
1
1
1
0
0
0
0
0
0
0
1
3 3 3
6 6 0
3 3 1
1
2
3
3
0
0
0
3
0
0
0
1
Construction notes: we first establish the isomorphism SL(2, 3)

= Q8 C3 . The order of
2
SL(2, 3) is 24, by counting ordered bases of F3 . We
show that
there is only one element
1 0
of order 2 in SL(2, 3), namely the matrix I =
. This is because if g 2 = 1
0 1
then g has minimal polynomial dividing X 2 1. If this were the characteristic polynomial
then det g = 1, which is not possible, so both eigenvalues of g must be 1. Since g is
diagonalizable, g must be the matrix I.
Next, SL(2, 3) permutes the 4 one-dimensional subspaces of F23 giving a homomorphism to S4 , with kernel {I} of order 2. The image is thus a subgroup of S4 of order 12,
so must be A4 . Since A4 has a normal Sylow 2-subgroup, the preimage of this subgroup in
SL(2, 3) is a normal subgroup of order 8, which must therefore be a Sylow 2-subgroup of
SL(2, 3). This subgroup has only one element of order 2, so it is Q8 . A Sylow 3-subgroup
is now a complement to this normal subgroup, and the isomorphism SL(2, 3)
= Q8 C3
is established.
We enumerate the conjugacy classes and centralizer orders, and find that all elements
of order 4 in Q8 are conjugate, using conjugacy within Q8 and the action of the 3-cycle.
The elements outside Q8 commute with no elements of Q8 apart from the center, so they
have centralizers of order 6. This puts them in conjugacy classes of size 4, and we are able
to complete the enumeration of conjugacy classes from this information.
Peter Webb
263
Since SL(2, 3) has A4 as a quotient, the character table of A4 lifts to SL(2, 3) giving
the characters of degree 1 and 3. Calculating the sum of the squares of the degrees shows
that the remaining three characters have degree 2. Since a2 is central it must act as a
scalar on each simple representation, so it acts as 2 in the degree 2 representations, and
column orthogonal between columns 1 and 2 shows that the entries must be 2 for the
characters of degree 2. Column orthogonality now gives the values of these characters as 0
on a. If any remaining entry in the degree 2 characters is non-zero, it gives non-zero values
in the same column for the degree two characters, by multiplying by the three degree 1
characters. Thus any zero entry here gives three zero entries on the degree 2 characters.
This is not possible because the product of that column with its complex conjugate must be
the centralizer order, which is 6. Hence the remaining entries are all non-zero, and once one
of the degree 2 characters is determined the other two are obtained by multiplying by the
degree 1 characters. We deduce from column orthogonality that each degree 2 character
must have absolute value 1 on y, y 2, ya2 and y 2 a2 . The matrices by which the element y
acts has eigenvalues taken from 1, 3 , 32 and the only possible sums of two of these with
absolute value 1 are 1 = 3 + 32 , 3 = 1 + 32 and 32 = 1 + 3 , so these must be the
three values of the degree 2 characters on y. The values on y 2 are their conjugates. The
three degree 2 characters must arise as one complex conjugate pair and one real character.
A similar argument with the possible sums of the eigenvalues of the matrix of ya2 , which
must come from 1, 3 , 32 now determines the remaining character values.
When p = 2 it is immediate that the three degree 1 characters reduce to give three
distinct simple Brauer characters. When p = 3 we have the trivial Brauer character, and
there are no more degree 1 characters because the abelianization is C3 . The degree 3
character is a block of defect zero so also gives a simple Brauer character. Finally the
degree 2 character gives a simple Brauer character, because it cannot be written as a sum
of degree 1 characters.
The tables for characteristics 2 and 3 illustrate the theory described in Proposition 8.8
and Theorem 8.10 for semidirect products of a p-group and a p -group. In characteristic
3 they also illustrate the form of the simple representations, as described in Chapter 6
Exercise 21. In characteristic 2 more detailed information about the projectives is obtained
in Chapter 8 Exercise 4. However, the calculation of these tables is straightforward and
does not require this theory.
The alternating group A5 .
264
This simple group is isomorphic to SL(2, 4) and also to P SL(2, 5).

A5
ordinary characters
g
|CG (g)|
() (12)(34) (123)
60
4
3
1
3a
3b
4
5
1
3
3
4
5
1
1
1
0
1
1
0
0
1
1
(12345)
5
(13524)
5
1
1
3
+ 5 ) (5 + 54 )
(5 + 54 ) (52 + 53 )
1
1
0
0
(52
Characteristic 2:
A5
= SL(2, 4)
Brauer simple p = 2
A5
= SL(2, 4)
g
|CG (g)|
() (123) (12345) (13524)

60
3
5
5
g
|CG (g)|
() (123)
60
3
1
2a
2b
4
1
2
2
4
1
2a
2b
4
12
0
2
2
8 1 (52 + 53 )(5 + 54 )
8 1 (5 + 54 )(52 + 53 )
4
1
1
1
1
1
1
1
1
1
5 + 54 52 + 53
52 + 53 5 + 54
1
1
A5
= SL(2, 4)
1
1
1
0
1
0
1
0
0
1
0
0
1
0
1
(13524)
5
A5
= SL(2, 4)
Cartan matrix p = 2
1 2a 2b 4
1 2a 2b 4
1
3a
3b
4
5
(12345)
5
1
2a
2b
4
0
0
0
1
0
4
2
2
0
2
2
1
0
2
1
2
0
0
0
0
1
Characteristic 3:
A5
A5
Brauer simple p = 3
g
() (12)(34) (12345)
|CG (g)| 60
4
5
1
3a
3b
4
1
3
3
4
1
1
1
0
(13524)
5
1
1
3
+ 5 )(5 + 54 )
(5 + 54 )(52 + 53 )
1
1
(52
g
() (12)(34) (12345)
|CG (g)| 60
4
5
1
3a
3b
4
6
3
3
9
2
1
1
1
(13524)
5
1
1
3
+ 5 )(5 + 54 )
(5 + 54 )(52 + 53 )
1
1
(52
Peter Webb
A5
265
A5
Cartan matrix p = 3
1 3a 3b 4
1 3a 3b 4
1
3a
3b
4
5
1
0
0
0
1
0
1
0
0
0
0
0
1
0
0
0
0
0
1
1
1
3a
3b
4
2
0
0
1
0
1
0
0
0
0
1
0
1
0
0
2
Characteristic 5:
A5
= P SL(2, 5)
Brauer simple p = 5
A5
= P SL(2, 5)
g
() (12)(34) (123)
|CG (g)| 60
4
3
g
() (12)(34) (123)
|CG (g)| 60
4
3
1
3
5
1
3
5
1
3
5
1
1
1
1
0
1
A5
= P SL(2, 5)
5
10
5
1
0
0
1
0
0
1
1
1
0
0
0
0
0
1
2
1
1
A5
= P SL(2, 5)
Cartan matrix p = 5
1 3 5
1 3 5
1
3a
3b
4
5
1
2
1
1
3
5
2
1
0
1
3
0
0
0
1
Construction notes for A5 : to compute the conjugacy classes, compute the centralizer
of each element first in S5 , then intersect that centralizer with A5 . The index in A5 is the
number of conjugates of that element. This enables us to see that the class of 5-cycles
splits into two in A5 , but the other classes of S5 do not. One of the three-dimensional
representations can be obtained via the realization of A5 as the group of rotations of the
icosahedron. Note that when computing the trace of a rotation matrix it simplifies the
calculation to choose the most convenient basis. The tensor square of this representation
contains two copies of itself, determined by taking an inner product. The remaining
three-dimensional summand is simple and new. It can also be obtained as an algebraic
266
conjugate of the first three-dimensional representation. The permutation representation on

5 symbols is 2-transitive, so decomposes as the direct sum of the trivial representation and
an irreducible 4-dimensional representation, by Lemma 5.5. The remaining 5-dimensional
representation can be found by the orthogonality relations. It is also induced from one
of the non-trivial 1-dimensional representations of A4 , as can be verified by taking the
inner product of this induced representation with the other representations of A5 so far
obtained, using Frobenius reciprocity, to see that the induced representation is new.
S5
ordinary characters
g
|CG (g)|
1
sign
4a
4b
6
5a
5b
() (12) (123) (12)(34) (1234) (123)(45) (12345)

120 12
6
8
4
6
5
1
1
4
4
6
5
5
1
1
2
2
0
1
1
1
1
1
1
0
1
1
1
1
0
0
2
1
1
1
1
0
0
0
1
1
1
1
1
1
0
1
1
1
1
1
1
1
0
0
Construction notes for S5 : the conjugacy classes are determined by cycle type, the derived subgroup is A5 and there are two 1-dimensional representations. The permutation
representation on the five symbols is the direct sum of the trivial representation and a 4dimensional simple representation 4a (by Lemma 5.5, but it can be checked using the orthogonality relations). Its tensor with the sign representation gives another 4-dimensional
simple. Its exterior square is simple of dimension 6. The symmetric square decomposes
with a new summand of dimension 5. The final character is obtained by multiplying by
the sign representation.
The general linear group GL(3, 2).
The group GL(3, 2) = SL(3, 2) is simple and is isomorphic to P SL(2, 7).
GL(3, 2)
ordinary characters
g
|CG (g)|
1
3a
3b
6
7
8
1 2 4
168 8 4
3 7a 7b
3 7 7
1 1 1 1 1 1
3 1 1 0
3 1 1 0
6 2 0 0 1 1
7 1 1 1 0 0
8 0 0 1 1 1
Peter Webb
267
Here = 7 + 72 + 74 so that = 73 + 75 + 76 . In calculating with orthogonality relations

it is helpful to know that 2 =
1 and
= 2.
Characteristic 2:
GL(3, 2)
Brauer simple p = 2
GL(3, 2)
g
|CG (g)|
g
|CG (g)|
1 3
168 3
1
3a
3b
8
8 2
1
1
16 1 1 1
16 1 1 1
8 1
1
1
1
3a
3b
8
1 3 7a 7b
168 3 7 7
1 1 1
3 0
3 0
8 1 1
GL(3, 2)
1
1
0
0
1
0
0
0
0
1
1
0
0
0
1
1
1
0
0
0
0
0
0
1
7b
7
GL(3, 2)
Cartan matrix p = 2
1 3a 3b 8
1 3a 3b 8
1
3a
3b
6
7
8
7a
7
1
3a
3b
8
2
1
1
0
1
3
2
0
1
2
3
0
Characteristic 7:
GL(3, 2)
= P SL(2, 7)
Brauer simple p = 7
GL(3, 2)
= P SL(2, 7)
g
|CG (g)|
g
|CG (g)|
1 2 4 3
168 8 4 3
1
3
5
7
7 3 1 1
14 2 21
14 2 01
7 11 1
1 2 4 3
168 8 4 3
1
3
5
7
1 1 1 1
3 1 1 0
5 111
7 11 1
GL(3, 2)
= P SL(2, 7)
GL(3, 2)
= P SL(2, 7)
Cartan matrix p = 7
1 3 5 7
1 3 5 7
1
3a
3b
6
7
8
1
0
0
1
0
0
0
1
1
0
0
1
0
0
0
1
0
1
0
0
0
0
1
0
1
3
5
7
2
0
1
0
0
3
1
0
1
1
2
0
0
0
0
1
0
0
0
1
268
Construction notes for GL(3, 2): By counting ordered bases for F32 the order of
GL(3, 2) is 168 = (23 1)(23 2)(23 22 ). It is helpful to identify certain subgroups
of GL(3, 2). We describe these by indicating the form of the matrices in the subgroups.
These matrices must be invertible and, subject to that condition, can have any field element
from F2 (i.e. 0 or 1) where is positioned. Let
1 0
B = 1

and
Then
0 , P1 =
1
1 0
P2 =

0
1 0 0

0 , U1 = 0 1 0 , L 1 =
1
1
0 0
0
1
, U2 =
0 0
1
1 0 , L2 = 0
0 1
0
0
0
1
0 0
.

B
= D8 ,
L1
= L2
= GL(2, 2)
= S3 ,
U1
= U2
= C2 C2 ,
P1 = U 1 L 1
= S4 ,
P2 = U 2 L 2
= S4
because S4 also has this semidirect product structure. The subgroup B is a Sylow 2subgroup of GL(3, 2), since it has order 8.
We show that GL(3, 2) has a single class of elements of order 2 (involutions) by
showing that all involutions in B are conjugate in GL(3, 2). Since all Sylow 2-subgroups
are conjugate in GL(3, 2), this will be sufficient. Conjugacy of involutions within B follows
because in P1 they fall into two conjugacy classes, and in P2 they fall into two different
classes, so that combining this information we see they are all conjugate in GL(3, 2). Direct
calculation shows that
1 0 0
B = CGL(3,2) ( 0 1 0 )
1 0 1
so that there are 21 = 168/8 elements of order 2, all conjugate.

The two elements of B of order 4 are conjugate in B, so there is a single class of
elements of order 4. The centralizer of such an element also centralizes its square, so is a
subgroup of B, and hence has order 4. Thus there are 42 = 168/4 elements of order 4, all
conjugate.
Sylows theorem shows that the number of Sylow 3-subgroups must be one of 1, 4,
7 or 28. The first three possibilities would imply that there is an element of order 2
which centralizes a Sylow 3-subgroup, which does not happen since the centralizer of an
involution has order 8. Thus there are 28 Sylow 3-subgroups H and NGL(3,2) (H)
= S3 of
Peter Webb
269
order 6, since this group appears as a subgroup of P1 . All elements of order 3 are thus
conjugate, and there are 56 = 168/3 of them.
Sylows theorem shows that there are 1 or 8 Sylow 7-subgroups K, and since GL(3, 2)
does not have a normal 7-cycle, there are 8. Thus NGL(3,2) (K)
= K H of order 21, where
H has order 3. The action of H on K is non-trivial because the centralizer of H contains
no element of order 7, so if K = hgi the action of a generator of H may be taken to send
g 7 g 2 7 g 4 7 g. This means the elements of order 7 fall into two conjugacy classes,
represented by g and g 1 , each of size 24.
Since 168 = 1 + 21 + 42 + 56 + 24 + 24 we have accounted for all the elements of
GL(3, 2).
We construct the three largest degree characters of GL(3, 2). The induced character
from the subgroup K H of order 21 is
g
|CG (g)|
GL(3,2)
1 KH
1 2 4 3 7a 7b
168 8 4 3 7 7
8
0 0 2
by the induced character formula. This is because no elements of order 2 or 4 can be

conjugated into K H; an element which conjugates a generator of K into K H must
normalize K, so lies in K H and hence the character is 1 on an element of order 7; the
elements of order 3 in K H lie in two conjugacy classes under this group, each of size 7.
The images of such a class under the action of GL(3, 2) partition the 56 elements of order
3 into 8 sets, two of which lie in K H. From this we see that the value of this induced
GL(3,2)
character on a 3-element is 2. Now 1 KH 1 is verified to be a simple character of
degree 7 using the orthogonality relations.
GL(3,2)
We next compute 1 P1
. To do this it is convenient to observe that GL(3, 2)

0
3
permutes the 7 non-zero elements of F2 transitively, and the stabilizer of 0 is P1 , so

1
that the induced character is the permutation character on these 7 points. We now take
typical elements of orders 2, 4 and 3 such as
0 1
1 0
0 0

0
1 0
0, 1 1
1
0 1

0
0
0, 0
1
1
1 0
0 1
0 0
and find the number of fixed points. For elements of order 7 it is clear that they act
regularly on the 7 non-zero elements of F32 , so there are no fixed points. This shows that
270
the character is
g
|CG (g)|
1 2 4 3 7a 7b
168 8 4 3 7 7
GL(3,2)
1 P1
3 1 1
GL(3,2)
and orthogonality relations show that 1 P1

1 is simple and it has degree 6.
To construct the simple character of degree 8 we exploit a slightly more complicated
combinatorial structure, namely the incidence graph of lines and planes in F32 . This graph
is, in fact, the building of GL(3, 2). The graph has 14 vertices which are the 7 linear
subspaces of F32 of dimension 1 and the 7 linear subspaces of F32 of dimension 2. We
place an edge between a 1-dimensional subspace and a 2-dimensional subspace if one is
contained in the other, obtaining a bipartite graph. These containment pairs are permuted
transitively by GL(3, 2) as are the 1-dimensional subspaces (with typical stabilizer P1 ) and
the 2-dimensional subspaces (with typical stabilizer P2 ). The stabilizer of a typical edge
is thus P1 P2 = B, and we deduce that there are |GL(3, 2) : B| = 21 edges. Regarding
this graph as a simplicial complex, the simplicial chain complex over C has the form
GL(3,2)
0 C B
GL(3,2)
C P 1
GL(3,2)
C P2
0.
The simplicial complex is, in fact, connected so that we have an exact sequence
GL(3,2)
0 H1 C B
GL(3,2)
C P 1
GL(3,2)
C P2
C0
where H1 is a CG-module which is the first homology of the complex. Because CG is

semisimple the sequence is split and we have an alternating sum formula for the character
of H1 :
GL(3,2)
GL(3,2)
GL(3,2)
H1 = 1 B
1 P1
1 P2
+1 .
GL(3,2)
We have already computed 1 P1

GL(3,2)
1 B
GL(3,2)
and 1 P2
is similar with the same answer.
The computation of
is made easier by the fact that B is a 2-group, so that the
character is only non-zero on 2-power elements. It is
g
|CG (g)|
GL(3,2)
1 B
1 2 4 3 7a 7b
168 8 4 3 7 7
21
5 1 0
The value on an involution arises because there are 5 involutions in B, which is the centralizer one of them, and so 5 coset representatives of B conjugate this involution to an
Peter Webb
271
element of B. There are two elements of order 4 in B, and they determine B as the normalizer of the subgroup they generate. Since B is self-normalizing, there is only one coset
of B whose representative conjugates an element of order 4 into B. We compute that H1
is a simple character using the orthogonality relations, and it has degree 8.
It remains to compute the two 3-dimensional characters of GL(3, 2). To do this we
can say that the two columns indexed by elements of order 7 must be complex conjugates
of each other because the elements are mutually inverse. All the characters constructed so
far are real-valued, so the two remaining characters must not be real-valued in order that
those two columns should be distinct. This means that the two remaining characters must
be complex conjugates of each other. On elements of orders 1, 2, 3 and 4 these characters
are real, so must be equal. With this information we can now determine those characters
on these columns using orthogonality relations and the fact that the sum of the squares
of the degrees of the characters is 168. Orthogonality gives an equation for the missing
complex number which is also determined in this way.
Dihedral groups.

We let D2n = hx, y xn = y 2 = 1, yxy 1 = x1 i.
D2n , n odd
ordinary characters
1
2n
x
n
x2
n
1
1a
1
1
1
1
1
1
ns G
hxi
g
|CG (g)|
(1 s
y
2
n
1
1
n1
ns + ns n2s + n2s n 2
n1
2 )
n1
2
n1
+ n 2
1
1
0
D2n , n even
ordinary characters
g
|CG (g)|
1
2n
1
1a
1b
1c
ns G
hxi
(1 s
1
1
1
1
2
n
2
1)
x
n
x2
n
1
1
1
1
1
1
1
1
s
2s
n + ns n + n2s
x2
2n
y xy
4 4
1
1 1
1 1 1
n
(1) 2 1 1
n
(1) 2 1 1
2(1)s 0 0
Construction notes for D2n : these tables were constructed in Example 4.15 and Section 4
Exercise 1.

Rep N Control

Uploaded by

Copyright:

Available Formats

Rep N Control

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rep N Control

Uploaded by

Copyright:

Available Formats

Finite Group Representations

Representations, Maschkes theorem and semisimplicity

1. Representations, Maschkes Theorem and Semisimplicity

Definitions and examples

Printed Oct. 13, 2014

Finite Group Representations

0 r n 1. Another important example of a degree 1 representation is the sign

Printed Oct. 13, 2014

as 1, 2 and 3 we get a representation

for all g G, v, w V and , R. A rule for multiplication G V V satisfying

Finite Group Representations

Printed Oct. 13, 2014

is the same thing as specifying a representation of G on V , since given a representation

Printed Oct. 13, 2014

representation of A. When A = RG we may describe the action on RG RG by observing

Printed Oct. 13, 2014

Finite Group Representations

an example of a representation that is a direct sum of two non-zero subspaces; however,

Then is linear and if w W then

Since furthermore (v) W for all v V , is a projection onto W and so V =

Printed Oct. 13, 2014

is an F G-module homomorphism: if h G and v V then

since as g ranges over the elements of G, so does h1 g. Now if v Ker( ) then hv

Printed Oct. 13, 2014

Finite Group Representations

(1.5) LEMMA. Let A be a ring with a 1 and suppose that U = S1 + + Sn is

Printed Oct. 13, 2014

We now present a different version of Maschkes theorem. The assertion remains

Printed Oct. 13, 2014

Finite Group Representations

to a summand Sj must be non-zero. If we assume that T is simple this projection will be

Exercises for Chapter 1

be two representations of G on the same R-module V that are injective as homomorphisms.

Printed Oct. 13, 2014

Calculate the socles of these two representations. Show that

Printed Oct. 13, 2014

Finite Group Representations

Show that I is a maximal ideal in R.

regarded as linear maps U U where U is a k-vector space of dimension 2 with basis

Printed Oct. 13, 2014

2. The structure of algebras for which every module is semisimple.

Printed Oct. 13, 2014

Finite Group Representations

(2.2) LEMMA. For any ring A with a 1, EndA (A A)

where the S1 , . . . , Sr are non-isomorphic simple modules occuring with multiplicities n1 , . . . , nr

Printed Oct. 13, 2014

the i, j entries lie in HomA (Uj , Ui ). This is because any endomorphism : U U

(2.4) COROLLARY. Let A be a finite-dimensional semisimple algebra over a field k.

where the Si are pairwise non-isomorphic simple modules we have that S1 , . . . , Sr is a

Printed Oct. 13, 2014

Finite Group Representations

Printed Oct. 13, 2014

Exercises for Chapter 2

Printed Oct. 13, 2014

Finite Group Representations

Printed Oct. 13, 2014

Printed Oct. 13, 2014

Finite Group Representations