Representation Theory

Tudor Ciurca

June 22, 2020

Contents
1 Some Non-Commutative Algebra 3
1.1 Absolute Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Artinian Rings and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The Radical and Socle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Idempotents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Introduction to Modular Representation Theory 12

2.1 The question of semisimplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 First Approaches to Modular Representations . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Indecomposable Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Brauer Theory 22
3.1 Brauer Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Projective Indecomposable Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1
Introduction
These are loose notes covering the course on Modular Representation Theory taught by Professor
Martin Liebeck in 2019/2020. Some extra material was added in by the author to aid explanations
and for the sake of completeness. On the other hand some material is missing, since it was treated
as prerequisite material in the course. Namely we assume the reader is somewhat familiar with
standard representation theory of finite groups over C, and that they know basic technical results
in algebraic number theory such as Dedekind domains, discrete valuation rings and completions.
Some homological algebra is also assumed but we will cover most of the required material with
some easy exercises left for the reader.

The official course website is [Link] mwl/m4p72/ , where you can hope-
fully find problem sheets and other material. Another useful resource is the Stanford course
[Link] , which is taught in a slightly different way, but it is helpful
to gain many perspectives.

2
1 Some Non-Commutative Algebra
1.1 Absolute Basics
In this section R will be reserved for a (possibly non-commutative) ring R with unity. We quickly
establish basic facts about non-commutative algebra and convince you that these structures still
have interesting results.

Definition 1.1. The opposite ring Rop is defined as having the same additive group structure as
R, and we denote Rop = {x∗ : x ∈ R} as sets. However the multiplication is defined as

x∗ · y ∗ = (y · x)∗

for any x, y ∈ R.

This definition does not make much sense for commutative rings. Convince yourself that if R
is commutative, then R ∼ = Rop .

Example 1.2. For an abelian group M we have the ring End(M ) of group endomorphisms, where
multiplication is given by composition.

Definition 1.3. 1. A left R-module M is an abelian group together with a ring morphism ρ :
R → End(M ). Usually we omit ρ and realize the elements of R as maps on M . You should
think of these as scalars in the sense of vector spaces. In other words it is an assignment of
self-maps r : M → M for every element r ∈ R, so that the following two properties hold:

• r(x + y) = rx + ry for all r ∈ R and x, y ∈ M

• (rs)(x) = r(s(x)) for all r, s ∈ R and x ∈ M

2. A right R-module M is an abelian group together with a ring morphism ρ : Rop → End(M ).
In other words it is an assignment of self-maps r : M → M for every element r ∈ R, so that
the following two properties hold:

• r(x + y) = rx + ry for all r ∈ R and x, y ∈ M

• (rs)(x) = s(r(x)) for all r, s ∈ R and x ∈ M

We will mostly focus on left R-modules since there is great symmetry between the left and right
modules of a ring. A special class of left and right modules are the left and right ideals.

Definition 1.4. 1. A left ideal of R is an additive subgroup I < R that is closed under left
multiplication from R. In other words, for all r ∈ R and x ∈ I we have rx ∈ I. More
succinctly a left ideal is a left submodule of R.

2. A right ideal of R is a right submodule of R.

Definition 1.5. Let N ⊆ M be left R-modules. We may form the quotient additive group M/N
which attains a natural structure of a left R-module in the following way. If φ : M → M/N is the
quotient map and r ∈ R, φ(s) = s + N ∈ M/N then r(φ(s)) := φ(r · s).

Definition 1.6. Let I be a left ideal in R and let M be a left R-module. Then we can create the
left submodule IM ⊆ M described by IM := {im ∈ M : m ∈ M, i ∈ I}.

3
Definition 1.7. Let N ⊆ M be left R-modules. We say N is a maximal left submodule of M if
whenever N ⊆ P as left submodules of M we have P = M or P = N , and N 6= M . If I ⊂ R is a
left ideal we say I is maximal if I is a maximal left submodule of R.

Definition 1.8. A simple left R-module is a nonzero left R-module M which has only 0 and M
as left R-submodules. A semisimple left R-module is a left R-module M which is isomorphic to a
direct sum of simple left R-modules. For our purposes a semisimple left R-module will be a finite
direct sum.

Proposition 1.9. Let N ⊆ M be left R-modules. There is a one-to-one inclusion preserving

correspondence between left submodules of M containing N and left submodules of M/N . Moreover,
N is a maximal left submodule of M if and only if M/N is a simple left R-module.

Proof. Let φ : M → M/N be the quotient map. Starting from a left submodule P ⊇ N we may
form the left submodule φ(P ) ⊆ M/N , since the image of a left submodule under a morphism of
left R-modules is a left submodule. Starting from a left submodule Q ⊆ M/N we may form the
left submodule φ−1 (Q) ⊆ M . Note that we can pull back left submodules through morphisms of
left R-modules.

We have to show that these two operations are mutually inverse. Firstly φ−1 (φ(P )) = P since
if we have φ(x) ∈ φ(P ) ⊆ M/N for some x ∈ M then there is some p ∈ P so that x − p ∈ N ⊆ P .
As a result x ∈ P as required. Secondly φ(φ−1 (Q)) = Q from purely set-theoretic reasons as φ is
surjective. It is also due to set-theoretic reasons that this bijection is inclusion preserving.

The last part then follows by noting that the left submodule M ⊆ M corresponds to the left
submodule M/N and the left submodule N corresponds to the left submodule 0 ⊆ M/N .

Lemma 1.10. Let M be a simple nonzero left R-module, and m ∈ M a nonzero element. Then
M = Rm := {rm ∈ M : r ∈ R}. As a result there is a surjective map of left R-modules φ : R → M
sending r 7→ rm ∈ M for all r ∈ R, so that M ∼
= R/ker(φ). Moreover ker(φ) must be a maximal
left ideal by Proposition 1.9.

Proof. Rm ⊆ M is a left submodule by construction, and it is nonempty because m is nonzero.

Therefore we must have Rm = M by simplicity. The last part is a straightforward conclusion, so
we have classified all simple left modules.

Proposition 1.11. Let R be a ring and x ∈ R a non-unit. Then there is some maximal left ideal
M containing x.

Proof. A direct application of Zorn’s lemma. If this makes you uncomfortable there is another
proof for finite dimensional K-algebras where K is a field.

Now we need some facts about semisimple modules.

Proposition 1.12. A left R-module M is semisimple if and only if it is a finite sum of simple left
submodules of M .

Proof. If M is semisimple we have M ∼

Ln
= i=1 Mi for some simple left R-modules M1 . . . Mn , so
clearly M is the sum of the simple left submodules given by the preimage of the Mi under the
isomorphism.

4
 Conversely suppose M = ni=1 Mi for M1 . . . Mn ⊂ M simple left submodules. Suppose by
P
induction that the converse statement is true for sums of n − 1 simple left submodules, where
n ≥ 2. The base case is considering M = M1 as a simple left module so it is clear. Then
M1 ∩ Mk = 0 for all k = 2 . . . n otherwise P
M1 ∩ Mk 6= 0 =⇒ M 1 = Mk by simplicity, and so we
n P n
can use theP
∼ Pnon M = i=2 Mi . Now M1 ∩ i=2 Mi = 0 and so by definition
inductive hypothesis
n
M = M1 ⊕ i=2 Mi . However i=2 Mi is semisimple by induction so we are done, because clearly
the direct sum of semisimple left modules is semisimple.

Proposition 1.13. Let M be a semisimple left R-module with left submodule N ⊂ M . Then N
and M/N are also semisimple.
Pn
Proof. We may assume N is nonzero otherwise it is easy. Write M = i=1 Mi for simple left
submodules M1 . . . Mn ⊂ M . Then for each k = 1 . . . n we have
P either Mk ∩ N = 0 or Mk ⊂ N
by simplicity. As a result if N is nonzero it is clear that N = rk=1 Mk for some r = 1 . . . n after
reordering and so N is semisimple. On the otherLrhand, we may have supposed
Ln that M1 . . . Mn are
∼ Ln ∼ ∼
distinct. In this case M = i=1 Mi and N = i=1 Mi and so M/N = i=r+1 Mi is semisimple.

We finish with a remark regarding invertible elements.

Definition 1.14. Let x ∈ A be an element in a ring. We say x is invertible if there exists some
y ∈ A called the inverse of x so that yx = xy = 1.

1.2 Artinian Rings and Modules

The Artinian condition is quite a strong condition on the size of a ring. However all rings we will
be concerned with in representation theory (Group algebras of finite groups) are Artinian.

Definition 1.15. A left R-module M is left-Artinian if we cannot form an infinite proper descend-
ing chain of left submodules M1 ) M2 ) . . . in M . We say a ring R is Artinian if it is left-Artinian
as a left R-module.

Example 1.16 (Group algebras of finite groups). Let G be Pa finite group and K a field. The group
algebra KG is defined as the set of all finite formal sums g∈G kg g of G-elements with coefficients
in K. Addition is done formally and multiplication is inherited from the multiplication of the group
and extended linearly.

KG is is a finite dimensional K-algebra. Any finite dimensional K-algebra is left-Artinian since

a descending chain of left ideals I1 ) I2 ) . . . in KG is also a descending chain of K-vector spaces.

Proposition 1.17. Let M be a left-Artinian R-module, and N ⊂ M a left submodule. Then N

and M/N are also left-Artinian.

Proof. If N is not left-Artinian then we may form an infinite proper descending chain of left
submodules M ⊃ N ) N1 ) . . . which contradicts the fact that M is left-Artinian. Suppose
M/N is not left-Artinian and form an infinite proper descending chain of left submodules M/N )
P1 ) . . . . By Proposition 1.9 this lifts to an infinite proper descending chain of left submodules
M ) φ−1 (P1 ) ) . . . if φ : M → M/N is the quotient map, contradicting the fact that M is
Artinian.

5
Theorem 1.18 (Jordan-Hölder Theorem for Modules). Suppose M is a nonzero left R-module
with composition series 0 = M0 ⊂ M1 ⊂ . . . ⊂ Mn = M for some finite n ∈ N so that Ni+1 /Ni is a
simple left R-module for all i = 0 . . . n−1. We call n the composition length and {Ni+1 /Ni }i=0...n−1
the composition factors. Then any other composition series for M has the same composition length
and composition factors up to permutation.

Proof. We prove by induction on the composition length n. For n = 1 we are dealing with a simple
left R-module and it is obvious. Let n ≥ 2 and suppose we proved the Theorem for all 1 . . . n−1. Let
0 = M0 ⊂ M10 ⊂ . . . ⊂ Mm 0 = M be another composition series, with m ≥ n necessarily by induc-

tion. We deal with the case Mm−10 = Mn−1 first, where we have equality as left submodules of M .
Here by induction we find that m = n and {M1 /M0 . . . Mn−1 /Mn−2 } ∼ {M10 /M0 . . . Mm−1
0 0
/Mm−2 }
as composition factors. Moreover Mn /Mn−1 ∼ 0 0
= Mm /Mm−1 and so we conclude.

0
Now suppose Mm−1 0
6= Mn−1 . Clearly Mm−1 0
+Mn−1 = M , and now we show that Mm−1 ∩Mn−1
is a maximal left submodule of Mn−1 . This follows by the second isomorphism theorem for left
0
Mm−1 +Mn−1
modules since then M 0 Mn−1 ∼
= = MM is simple. Take a composition series
∩Mn−1
m−1 M0 n−1
0
n−1
0 = K0 ⊂ K1 ⊂ . . . ⊂ Kr = Mm−1 0 ∩ Mn−1 which exists, for example, by considering the se-
0 0 0
ries 0 = Mm−1 ∩ M0 ⊂ Mm−1 ∩ M1 ⊂ . . . ⊂ Mm−1 ∩ Mn−1 which becomes a composition series after
potentially removing some terms. This then extends to a composition series for Mn−1 . By induc-
tion we see that n − 1 = r + 1 and {K1 /K0 . . . Kr /Kr−1 , Mn−1 /Kr } ∼ {M1 /M0 . . . Mn−1 /Mn−2 }
as composition factors.

0
We can similarly prove that Mm−1 ∩ Mn−1 is a maximal left submodule of Mm−10 and so we
0
can form a composition series 0 = K0 ⊂ K1 ⊂ . . . ⊂ Kr ⊂ Mm−1 of composition length n − 1. We
conclude then that m = n, and for the composition factors:

{M1 /M0 . . . Mn−1 /Mn−2 , M/Mn−1 } ∼ {K1 /K0 . . . Kr /Kr−1 , Mn−1 /Kr ∼ 0
= M/Mm−1 ,

M/Mn−1 ∼ 0
= Mm−1 0
/Kr } ∼ {K1 /K0 . . . Kr /Kr−1 , Mm−1 0
/Kr , M/Mm−1 }∼
∼ {M10 /M0 . . . Mm−1
0 0
/Mm−2 0
, M/Mm−1 }

Note that every Artinian module has a composition series by definition.

Corollary 1.1. Every left-Artininan module is left-Noetherian.

Proof. Let M be left-Artinian and suppose that we may form an infinite ascending proper chain
of left submodules 0 = M0 ( M1 ( . . . . For every m ∈ N we can take a composition series
0 = K0 ⊂ K1 ⊂ . . . ⊂ Kr = M/Mm for M/Mm . Then this lifts to a proper chain of left submodules
0 = M0 ⊂ M1 ⊂ . . . ⊂ M m ⊂ K f1 ⊂ . . . ⊂ Kfr = M . This can be refined for arbitrarily large m
to give arbitrarily large composition series for M . However this is in direct contradiction to the
Theorem, and so we are done.

Corollary 1.2. Let N ⊂ M be a left submodule of a left R-module. Then in fact M is left-
Artinian if and only if N and M/N are Artinian. Moreover, M is left-Artinian if and only if it
has a composition series.

Proof. Direct applications of the Theorem.

6
Lemma 1.19. If A, B are left-Artinian R-modules then so is A ⊕ B.

Proof. Apply Corollary 1.2 to the left modules A ⊂ A ⊕ B and A ⊕ B/A ∼

= B.

Proposition 1.20. Let R be an Artinian ring. A left R-module M is finitely generated if and only
if it is Artinian.

Proof. If M is a finitely generated left module we have a surjection of left R-modules f : Rn → M for
some n ∈ N. However a finite direct sum of Artinian modules is again Artinian, and M ∼ = Rn /ker(f )
is a quotient of an Artinian module, and is therefore an Artinian left module.

Conversely if M is left-Artinian choose a composition series 0 = M0 ⊂ . . . ⊂ Mn = M for

M . Each Mk+1 /Mk is simple, hence generated by some single mk+1 ∈ Mk+1 /Mk over R, for all
k = 0 . . . n − 1. After lifting all m0 . . . mn−1 to M we see that they generate M over R so we can
create a surjection of left R-modules f : Rn → M .

1.3 The Radical and Socle

Definition 1.21. The radical Rad(M ) of a left R-module M is the intersection of all maximal
left R-submodules of M . In particular, the radical Rad(R) of a ring R is the intersection of all
maximal left ideals of R. Rad(M ) ⊆ M is itself a left submodule of M .

Lemma 1.22. Let R be a ring. If x ∈ Rad(R) then 1 − x is a unit in R.

Proof. If 1−x is not a unit then it must be contained in some maximal left ideal I of R. However x ∈
Rad(R) ⊂ I by definition and so 1 = (1 − x) + x ∈ I which gives us the necessary contradiction.

If you’ve encountered enough algebra you may know that the radical measures the failure of
some algebraic structure to be semi-simple, and is related to nilpotency, as we shall now see.

Proposition 1.23. Let R be an Artinian ring. Then we have

1. Rad(R) = {x ∈ R : xM = 0 for all simple left modules M }.

2. Rad(R) is the largest nilpotent 2-sided ideal of R.

3. R is semi-simple if and only if Rad(R) = 0.

4. R/Rad(R) is semi-simple and Rad(R) is the smallest 2-sided ideal with semi-simple quotient.

Proof. 1. For any maximal left ideal I of R we have that R/I is a simple left module and
so if x(R/I) = 0 then x ∈ I. This immediately implies Rad(R) ⊇ {x ∈ R : xM =
0 for all simple left modules M }. On the other hand suppose x ∈ Rad(R) and M is a simple
left R-module. Then M ∼ = R/I for some maximal left ideal I of R. However x ∈ I and so
x(R/I) = xM = 0. We did not need the Artinian assumption.

2. From [1] it follows that Rad(R) is 2-sided, and in fact we get the symmetric definition
\
Rad(R) = I
I⊂R maximal right ideal

[Proof 1 of Nilpotency] Since R is Artinian we have a descending chain Rad(R) ) Rad(R)2 )

. . . which must terminate, and so there is some k ∈ N with Rad(R)k = Rad(R)k+1 . Assume

7
 Rad(R)k 6= 0. Let S := {I ⊂ R nonzero left ideal : Rad(R)I = I} which is non-empty
because it contains Rad(R)k . Since R is Artinian we may choose a minimal nonzero left
ideal J ∈ S. Then there is some x ∈ J so that Rad(R)k x 6= 0 since Rad(R)k J = J.
Now Rad(R)k x ∈ S because Rad(R)k+1 x = Rad(R)k x. However Rad(R)k x ⊆ J and so
Rad(R)k x = J by the minimality assumption on J.

In particular J = Rad(R)k x ⊂ . . . ⊂ Rad(R)x ⊂ Rx ⊂ J and so Rad(R)x = Rx. Therefore

there is some y ∈ Rad(R) so that yx = x, but then (1 − y)x = 0 in J with 1 − y a unit as
y ∈ Rad(R). Conclude that x = 0 which gives us a contradiction unless Rad(R)k = 0.

[Proof 2 of Nilpotency] By the Jordan-Hölder Theorem Rad(R) is an Artinian module with

some composition series 0 = A0 ⊂ . . . ⊂ Ar = Rad(R). From [1] we see that Rad(R) annihi-
lates simple left modules and so Rad(R)(Ai+1 /Ai ) = 0 for all i = 0 . . . r − 1. Conclude that
Rad(R)r+1 = 0.

Let J be a nilpotent 2-sided ideal, so that J r = 0 for some r ∈ N. Let M be a simple left
module. Then JM = 0 necessesarily, otherwise JM = M and then 0 = J r M = M . Since J
annihilates all simple left modules we have J ⊆ Rad(R) by [1].
3. Suppose Rad(R) = 0. We may assume R has some maximal left ideal, otherwise R is simple
and we are done. We will carry an induction on the composition length for R. The idea is
that we can create a composition series for R by intersecting maximal left ideals.

We may form a proper descending chain of left ideals I1 ) I1 ∩ I2 ) . . . by intersecting maxi-

mal left ideals of R. This chain converges to 0 since we assumed R is Artinian and Rad(R) = 0.
Then there is some n ∈ N with I1 ∩ . . . ∩ In−1 6= 0 and I1 ∩ . . . ∩ In = 0. Then necessarily
In + I1 ∩ . . . ∩ In−1 = R and so we see that R ∼= In ⊕ (I1 ∩ . . . ∩ In−1 ) with (I1 ∩ . . . ∩ In−1 ) a
simple left module. One can show easily that R ) I1 ) I1 ∩ I2 ) . . . ) (I1 ∩ . . . ∩ In ) = 0 is a
composition series for R and so we are done by induction.

Suppose ∼ Ln Si for simple left modules S1 . . . Sn . Then

L R is semisimple and write R = i=1
Ij := i6=j Si are all maximal left submodules of R, and clearly ∩ni=1 Ij = 0 and so we can
conclude that Rad(R) = 0.
4. We first show that Rad(R/Rad(R)) = 0. It is important to note that since Rad(R) is a
2-sided ideal, R/Rad(R) attains the structure of a ring, and so we can use condition [1] on it.
Let r +Rad(R) ∈ Rad(R/Rad(R)) and M a simple left R-module. Then Rad(R)M = 0 by [1]
and so M attains the structure of a simple left R/Rad(R)-module. Thus (r + Rad(R))M = 0
by [1], and so r ∈ Rad(R) since M was an arbitrary simple left R-module. Conclude that
R/Rad(R) is semisimple by [3].

Let J be a 2-sided ideal so that R/J is semisimple. This is equivalent to Rad(R/J) = 0 by [3].
We must show that Rad(R) ⊆ J. Let M be a an arbitrary simple left R-module. Then for
any x ∈ Rad(R) we have xM = 0. Now if N is a simple left R/J-module it naturally attains
the structure of a simple left R-module through the quotient. As a result (x + J)N = 0.
However N was also arbitrary and so x + J ∈ Rad(R/J) = 0 which is equivalent to x ∈ J.

8
 Some results proved above can be extended to finitely generated left modules in general.

Lemma 1.24. Any finitely generated left module over a semisimple ring is semisimple.

Proof. Let R be a semisimple ring. If M is a finitely generated left R-module then there is a
surjection of left R-modules f : Rn → M for some n ∈ N and so M ∼ = Rn /ker(f ). Now Rn is
semisimple because finite direct sums of semisimple modules is semisimple, and so ker(f ) and M
are semisimple by Proposition 1.13.

Proposition 1.25. Let R be Artinian and M a finitely generated left R-module. Then

1. M is semisimple if and only if Rad(M ) = 0.

2. Rad(M ) is the smallest submodule with semisimple quotient.

3. Rad(M ) = Rad(R)M

Proof. 1. Adapt the proof of Proposition 1.23[3] to modules. Remember that finitely generated
left modules over Artinian rings are equivalent to Artinian modules.

2. We give a very simple natural proof. Let S be the set of maximal left submodules of M and
for a given left submodule N ⊂ M let SN be the set of maximal left submodules of M/N .
By virtue of Proposition 1.9 we have a surjective operator qN : S → SN sending a maximal
left submodule of M to its restriction modulo N . This becomes a bijection when we restrict
qN to S N ⊂ S the set of maximal left submodules of M containing N . Then we compute
\ \ \
Rad(M/N ) = m= qN (n) = ( n)/N ⊂ Rad(M )/N ∩ Rad(M )
m∈SN n∈S N n∈S N

By [1] it follows that if we set N = Rad(M ) we get Rad(M/N ) = 0 above andT so M/Rad(M )
is semisimple. Conversely
T by [1] if M/N is semisimple then Rad(M/N ) = ( n∈S N n)/N = 0
and so Rad(M ) ⊂ ( n∈S N n) ⊂ N as required.

3. One sees that M/Rad(R)M attains a natural structure of left R/Rad(R)-module. Now
R/Rad(R) is semisimple whilst M/Rad(R)M is finitely generated because M is Artinian and
so M is finitely generated. By Lemma 1.13 M/Rad(R)M is semisimple so by [2] we have
Rad(M ) ⊂ Rad(R)M .

We show Rad(R)M ⊂ Rad(M ). Given x ∈ Rad(R) we know that x annihilates all simple left
R-modules. In particular for a maximal left submodule N ⊂ M we have that M/N is simple
and so x(M/N ) = 0. Thus xM ⊂ N but N was an arbitrary maximal left submodule and so
xM ⊂ Rad(M ) as required.

Definition 1.26 (Radical series). Let R be a left-Artinian ring, and M a finitely generated left
R-module. We define Rad0 (M ) = M and Radk+1 (M ) = Rad(Radk (M )) for all k ≥ 0. The radical
series for M is M = Rad0 (M ) ⊃ Rad1 (M ) ⊃ . . . .

Corollary 1.3. Take the radical series M = Rad0 (M ) ⊃ Rad1 (M ) ⊃ . . . as in the definition.
Then the quotients Radk+1 (M )/Radk (M ) are semisimple and Radk (M ) = Radk (R)M for k ≥ 0.

9
 P
Definition 1.27. Let M be a left R-module. The socle of M is Soc(M ) = ( N ⊂M simple N ) ⊂M
the sum of all simple left submodules of M .

Proposition 1.28. Let R be left-Artinian and M a finitely generated left R-module. Then

1. Soc(M ) is the largest semisimple left submodule of M .

2. Soc(M ) = {m ∈ M : Rad(M )m = 0}.

Proof. 1. Soc(M ) is the sum of the simple left submodules of M . In particular Soc(M ) is a finite
sum of simple left submodules of M because the sum of simple left submodules of M forms
an ascendng chain 0 ⊂ M1 ⊂ M1 + M2 ⊂ . . . in M which must terminate. By Proposition
1.12, Soc(M ) is semisimple. This proposition also implies that any semisimple submodule
N ⊂ M is a finite sum of simple left submodules of M , and so clearly N ⊂ Soc(M ).

2. For N ⊂ M a simple left submodule we necessarily have Rad(R)N = 0 because the radical
Rad(R) is the annihilator of simple left R-modules. As a result Rad(R)Soc(M ) = 0 and so
Soc(M ) ⊂ {m ∈ M : Rad(M )m = 0}.

Suppose m ∈ M so that Rad(R)m = 0. Then Rad(R)hmi = 0 = Rad(hmi) and so the

submodule hmi ⊂ M is semisimple. By [1] it follows that m ∈ hmi ⊂ Soc(M ).

Definition 1.29 (Socle series). Let R be left-Artinian and M a finitely generated left R-module.
The socle series 0 = Soc0 (M ) ⊂ Soc1 (M ) ⊂ . . . is defined by Soc1 (M ) = Soc(M ) and if πk : M →
M/Sock (M ) is the quotient map, we define Sock+1 (M ) = πk−1 (Soc(M/Sock (M ))) for all k ≥ 0.
Sock+1 (M )
Corollary 1.4. It follows that Sock (M )
= Soc(M/Sock (M )) for all k ≥ 0 and so these successive
quotients are semisimple.

1.4 Idempotents
Definition 1.30. An element e ∈ A is idempotent if e2 = e. We say e is primitive if whenever
e = e1 + e2 for idempotents e1 , e2 , we have e1 = 0 or e2 = 0. An orthogonal decomposition of e is a
sum e = e1 + . . . + en with ei idempotent for all i = 1 . . . n and ei ej = 0 for all distinct pairs ei , ej .

Proposition 1.31. Let A be a ring and I ⊂ A a nilpotent 2-sided ideal. Suppose e ∈ A/I is an
idempotent. Then there is some idempotent f ∈ A so that e ∈ f + I. Moreover, if e is primitive
then so is f .

Proof. Suppose by induction there is an idempotent ek ∈ A/I k so that ek +I = e. Let u ∈ A/I k+1 so
that u+I k = ek . Then define ek+1 = 3u2 −2u3 . Firstly ek+1 +I k = 3e2k −2e3k = 3ek −2ek = ek . Now
e2k+1 = 9u4 + 4u6 − 12u5 so e2k+1 − ek+1 = (3u2 − 2u3 )(3u2 − 2u3 − 1) = (3 − 2u)(u2 − u)2 (−2u − 1) =
(u2 − u)2 (2u − 3)(2u + 1). Since u2 − u ∈ I k we have e2k+1 − ek+1 ∈ I 2k ⊂ I k+1 so we conclude ek+1 is
idempotent in A/I k+1 . Suppose e is primitive but we can write ek = a + b in A/I k for idempotents
a, b. Then e = a + b in A/I where a, b are idempotents and so a ∈ I or b ∈ I by primitivity.
However I is a nilpotent 2-sided ideal and so it cannot contain any nonzero idempotents. For the
same reason there is some r ∈ N so that I r = 0 and so er =: f ∈ A is the lifted idempotent we were
searching for.

10
Proposition 1.32. Let I ⊂ A be a nilpotent 2-sided ideal in a ring A. Suppose 1 = e1 + . . . + en is
an orthogonal decomposition of 1 in A/I. Then we get an orthogonal decomposition 1 = f1 +. . .+fn
in A with ei ∈ fi + I for all i = 1 . . . n. Moreover, if ei is primitive then so is fi for all i.

Proof. We prove this by induction on n, where the base case n = 1 is clear. Write E = e2 +. . .+en =
1 − e1 . Then E is idempotent in A/I since (1 − e1 )2 = 1 − 2e1 + e21 = 1 − e1 . Lift e1 to some
idempotent f1 ∈ A so that e1 = f1 + I. Write F = 1 − f1 . Then F is idempotent and E = F + I.

We may form the ring F AF := {F aF : a ∈ A} ⊂ A where F is the identity. Now F IF is

a nilpotent 2-sided ideal of F AF since I was originally nilpotent and 2-sided. Consider the map
π : F AF → A/I induced by the quotient. The kernel is I ∩ F AF = F IF . Clearly I ∩ F AF ⊃ F IF ,
and I ∩ F AF ⊂ F IF since F aF ∈ I then F F aF F = F aF ∈ F IF . Now this is a ring morphism
with image E(A/I)E where E = F + I is the identity. Recall that E = e2 + . . . + en and this
holds in E(A/I)E since Eej E = ej for all j = 2 . . . n. We apply the induction hypothesis to
this orthogonal decomposition of the identity to get f2 . . . fn ∈ F AF orthogonal idempotents with
ej = fj + F IF for all j = 2 . . . n, and F = f2 + . . . fn . Then it clearly holds that 1 = f1 + . . . fn
and these idempotents are orthogonal in A since f1 F = F f1 = 0 and so f1 fj = fj f1 = 0 for all
j = 2 . . . n. The last part of the proposition follows from the first proposition.

11
2 Introduction to Modular Representation Theory
2.1 The question of semisimplicity
To motivate modular representation theory we need to deal with the following theorem.

Theorem 2.1 (Maschke). Let G be a finite group and K a field of characteristic p so that p 6 | |G|.
Then every finitely generated left KG-module is semisimple.

Proof. Let V be a finitely generated left KG-module. Then V is necessarily finite-dimensional over
K. We therefore prove by induction on dimK (V ). The base case is clear since every 1-dimensional
KG-module must be simple.

Take a left proper nonzero submodule W ( V . Then V ∼ = W ⊕ Q for some complement Q as

K-vector spaces. Let ρ : V → Q be the induced projection map of K-spaces. we wish to turn this
into a map of left KG-modules. To this end we define L := |G| g∈G gρg −1 : V → V and define
1 P

Q0 = Im(L) as a K-subspace of PV . Firstly Q0 is a left KG-submodule since given h ∈ G we have

1 −1 1 −1
P
hLv = |G| g∈G hgρg v = |G| g∈G hgρ(gh) hv = Lhv for all v ∈ V . Also for any v ∈ V we
1 −1
P
have v = Lv + |G| g∈G g(1 − ρ)g v with v − Lv ∈ W necessarily since W is G-invariant. As a
result V ∼
= W ⊕ Q0 as a direct sum of left KG-modules. However both W, Q0 are semisimple by the
induction hypothesis and so we conclude.

Definition 2.2. A representation ρ : KG → GL(V ) is modular if char(K)| |G|.

Maschke’s theorem makes it easy to study representations in the non-modular case since we
are reduced to studying irreducible representations. For modular representations we have to work
harder. One direction to take is to study semisimple Artinian rings in more detail, since by Chapter
1 we know that finitely generated left modules over semisimple Artinian rings are semisimple. We
begin by studying endomorphisms of semisimple modules.

Lemma 2.3 (Schur). Let K be an algebraically closed field, and R a K-algebra. Let S be a simple
left R-module. Then EndR (S) ∼
= K consists of the scalar maps in K.

Proof. For any nonzero s ∈ S we have Rs = S. Take l ∈ EndR (S). Considering l as a K-linear
map on the vector space S, we see that l has a nonzero eigenvector v ∈ S with eigenvalue λ since K
is algebraically closed. If λ = 0 then necessarily lS ( S is a proper left submodule since rlS = lrS
for all r ∈ R. As a result we must have lS = 0 and so l = 0.

Suppose λ 6= 0. Then 0 6= hvi ⊂ S is a left submodule since for any r ∈ R we have rv =

λ−1 rλv = λ−1 rlv = λ−1 lrv. It follows that Rv = S is the eigenspace for l, so that l acts via scalar
multiplication by λ on S.

Lemma 2.4. Let A be a ring and let EndlA (A) be the A-endomorphisms of A as a left A-module.
Then EndlA (A) ∼
= Aop as rings.

Proof. The ring structure on EndlA (A) is given by composition and addition. Choose multiplica-
tions ∗ for Aop and · for A, and identify A = Aop as sets. We define the map φ : Aop → EndlA (A)
where φ(a)(x) = xa for all a, x ∈ A. This is a ring morphism because φ(a ∗ b)(x) = φ(b · a)(x) =
x · b · a = φ(a)(φ(b)(x)) for all a, b, x ∈ A. Also φ(a + b)(x) = x(a + b) = xa + xb = φ(a)(x) + φ(b)(x)
for all a, b, x ∈ A. Finally φ(1) is the identity and φ(0) is the zero map clearly. It suffices to define

12
an inverse map and show that it too is a ring map.

Define ψ : EndlA (A) → Aop where ψ(f ) = f (1) for all f ∈ EndlA (A). This will be an inverse to φ
since ψ(φ(a)) = φ(a)(1) = a for all a ∈ A and φ(ψ(f ))(x) = xψ(f ) = xf (1) = f (x) for all x ∈ A and
f ∈ EndlA (A). Now ψ is a ring morphism because ψ(f ◦ g) = f (g(1)) = g(1) · f (1) = ψ(f ) ∗ ψ(g) for
all f, g ∈ EndlA (A). Also ψ(f + g) = (f + g)(1) = f (1) + g(1) = ψ(f ) + ψ(g) for all f, g ∈ EndlA (A).
Finally ψ(0) = 0(1) = 0 and ψ(Id) = Id(1) = 1.

Lemma 2.5. Let R be a ring and M ∼

Ln
= i=1 Mi a semisimple left R-module with M1 . . . Mn simple.
Then EndR (M ) ∼ = X := {(ρij )i,j=1...n : ρij ∈ HomR (Mj , Mi )} as rings, the latter being matrices of
left R-homomorphisms.
Ln
LnWe will be rough in the details. Let Ij : Mj → i=1 Mi denote the standard inclusion and
Proof.
πj : i=1 Mi → Mj the standard projection for all j = 1 . . . n. We define a map φ : EndR (M ) → X
by φ(f ) = (πi ◦ f ◦ IjP )i,j=1...n for allPnf ∈ EndR (M ). We also define a map ψ : X → EndR (M )
n
by ψ((ρij )i,j=1...n ) = ( i=1 ρi1 , . . . , i=1 ρin ) for all (ρij )i,j=1...n ∈ X. We leave it to the reader to
show that theses two maps are ring morphisms and mutually inverse.

Definition 2.6. Let K be a field. For distinction we define a K-algebra to be simple if it has no
proper nonzero 2-sided ideals.

A simple algebra A is clearly semisimple as a ring, because the radical Rad(A) is a 2-sided ideal
which must necessarily be 0. Note that the proof of Proposition 1.23[1] and the fact that Rad(A)
is 2-sided did not require the Artinian hypothesis.

Example 2.7. Let K be a field. Let Mn (K) denote the n × n matrices with K-coefficients. Then
Mn (K) with the obvious ring structure is a simple K-algebra. Given a nonzero matrix m ∈ Mn (K)
we show that the 2-sided ideal hmi generated by m is in fact Mn (K) entirely. This will be sufficient.

Suppose the entry mij is nonzero. By using projection and scalar maps given by matrix multi-
plication in Mn (K) on both sides we can produce the elementary matrix Ei,j which has entries 1 at
row i, column j and is 0 elsewhere. Using other elementary matrices we can produce all elementary
matrices, and thus Mn (K) entirely.

Theorem 2.8 (Wedderburn). Let K be an algebraically closed field and let A be a simple K-algebra.
Then A ∼
= Mn (K) for some n ∈ N.

Proof. Take a simple left submodule S ⊂ A. Let I be the set of simple left submodules P of A
isomorphic to S as left A-modules. Then define the semisimple left submodule M = s∈I s. This
is semisimple since A
Lis semisimple as a left module and M is a left submodule of A. As a result
we may write M = i=1 Si for S1 . . . Sn simple. Now EndA (Si ) ∼
∼ n
= K for all i = 1 . . . n by Schur’s
lemma and EndA (M ) ∼ = X := {(ρij )i,j=1...n : ρij ∈ HomA (Sj , Si )} by Lemma 2.5. However by
construction Si ∼
= S for all i = 1 . . . n so it follows that X ∼ = Mn (K) by using Schur’s lemma and
∼
the fact that HomA (Sj , Si ) = EndA (S) for all i, j = 1 . . . n.

Now we show that M is a 2-sided ideal of A. It is certainly a left ideal. For x ∈ A we have that
Si x is a left ideal of A for all i = 1 . . . n. suppose there is some s ∈ Si so that sx = 0. Then we know
As = Si by simplicity and so Si x = Asx = 0. Otherwise the map ρ : Si → Si x defined by ρ(s) = sx
for all s ∈ Si is injective, and therefore an isomorphism onto its image Si x. It follows that Si x ⊂ M
in either case for all i = 1 . . . n. As a result M x ⊂ M but x ∈ A was arbitrary. Hence M is a 2-sided

13
nonzero ideal of A, so by the simplicity of A as a K-algebra we have M = A. Finally by Lemma
2.4 we have Aop ∼ = EndlA (A) = EndA (M ) ∼ = Mn (K), and A ∼= (Aop )op ∼
= Mn (K)op ∼ = Mn (K) is
clear by using the transpose morphism.

Lemma 2.9. Let U ∼

Lr
= i=1L Ui be a left A-module and suppose HomA (Ui , Uj ) = 0 for all distinct
pairs i, j. Then EndA (U ) = ri=1 EndA (Ui ) as rings.

Proof. This is a special case of Lemma 2.5. We have EndA (U ) ∼ = X := {(ρij )i,j=1...n : ρij ∈
HomA (Uj , Ui )}. However ρij = 0 for all ρij ∈ HomA (Uj , Ui ) whenever i, j L
are distinct, and so
every matrix in EndA (U ) is diagonal and we have naturally that EndA (U ) = ri=1 EndA (Ui ).

Theorem 2.10 (Artin-Wedderburn). Let K be an algebraically closed Qr field. Let A be an Artinian

K-algebra which is semisimple as a left A-module. Then A ∼ = i=1 Mni (K) for some postive
integers n1 . . . nr .

Proof. Write A ∼
Lt
= i=1 Ti as a direct sum of simple left A-modules. Suppose T1 . . . Tr are repre-
sentatives for isomorphism classes of simple left A-modules among the T1 . . . Tt . Define Si as the
sum in A of all simple left A-submodules isomorphic to Ti , for each i = 1 . . . r. Then for distinct
i, j we have HomA (Si , Sj ) = 0 since given ρj,i ∈ HomA (Si , Sj ) we see that Si is a semisimple left
A-module and so Si ∼ = ker(ρj,i ) ⊕ Im(ρj,i ). However Im(ρj,i ) is a left A-submodule of Sj so all of
its composition factors must be Tj . If Im(ρj,i ) is nonzero it will contribute a Tj composition factor
to Si , which will give a contradiction. Therefore Im(ρj,i ) = 0 and so HomA (Si , Sj ) = 0.

∼ Lr ∼ op ∼
Lr op ∼
LrWe have A =op i=1 Si . Now using Lemma 2.9 we see that A = EndA (A) = ( i=1 EndA (Si )) =
( i=1 Mni (K)) , where L the last isomorphism
Lr is achieved as in LWedderburn’s Theorem using
Schur’s lemma. Now ( ri=1 Mni (K))op ∼ = ( i=1 Mni (K)op ) ∼ =
r
i=1 M ni (K) as required, since
taking opposite rings respects products and the opposite of a matrix ring is itself, via the transpose
morphism.

For the remainder of this chapter K will be an algebraically closed field and G a finite group.
Studying endomorphisms of semisimple rings has also led us to the following important combina-
torial argument for counting irreducible representations in the non-modular case.

Corollary 2.1. Suppose Char(K) = 0. PLet V1 . . . Vn be representatives for the isomorphism classes
of simple left KG-modules. Then |G| = ni=1 dimK (Vi )2 .

This argument necessarily fails in the modular case. One thing we can know in all cases is the
number of irreducible representations. This is Brauer’s theorem, covered in the next subsection.
For now we prove a more general result hinting at how we may find the simple left modules of an
Artinian K-algebra. This will directly imply the above corollary.

Lemma 2.11. Let A be an Artinian ring. Then the quotient map π : A → A/Rad(A) realizes
every left A/Rad(A)-module as an A-module, and this induces a bijection between the simple left
A/Rad(A)-modules and simple left A-modules up to respective isomorphism classes.

Proof. Given a simple left A-module S, we see that Rad(A) annihilates S and so naturally S
becomes a simple left A/Rad(A)-module. For the same reason every morphism of left A-modules
f : S1 → S2 when S1 , S2 are simple, gives rise to a morphism of simple left A/Rad(A)-modules.
This is sufficient to establish the bijection.

14
Lemma 2.12. Let R = A×B be a product of Artinian rings. Then R is Artinian and the simple left
R-modules are precisely the simple left A-modules and the simple left B-modules with the R-action
induced by the projections π1 : A × B → A and π2 : A × B → B.

Proof. A, B are Artinian so we can get composition series 0 = A0 ⊂ . . . ⊂ An = A and 0 = B0 ⊂

. . . ⊂ Bm = B. Then clearly 0 = A0 ⊂ . . . ⊂ An ⊂ A × B1 ⊂ . . . ⊂ A × B is a composition series
for A × B and so A × B is Artinian.
Certainly we can get distinct simple left R-modules in the manner described in the lemma. A
simple left R-module is isomorphic to A × B/M for some maximal left ideal M ⊂ R. However M
splits as e1 M ⊕ e2 M where e1 = (1, 0), e2 = (0, 1) are the orthogonal idempotents for R. As a result
e1 M = A or e2 M = B since otherwise we may construct the bigger proper ideal M ( A ⊕ e2 M ( R
or M ( e1 M ⊕ B ( R. This means that R/M ∼ = A/e1 M ⊕ B/e2 M and one of A/e1 M or B/e2 M
is zero, which concludes the lemma.

Lemma 2.13. Let A = Mn (K) be a matrix algebra over a field K and n ≥ 1. Then there is
precisely one nontrivial simple left A-module S, and EndA (S) ∼
= A.

Proof. Every simple ring is semisimple as a left module over itself, since the radical is a proper
2-sided ideal. In this particular case,
Ln if we consider A as a left A-module we can consider the column
decomposition A := Mn (K) ∼ = i=1 Ci , where the columns Ci are left A-submodules. Now the
columns C1 . . . Cn are simple and mutually isomorphic, and so we are done, since every nontrivial
simple left A-module is isomorphic to A/M for some maximal left ideal M of A, and so it must be
isomorphic to the column C1 .

Corollary 2.2. Let A = ri=1 Mni (K) be a product of matrix rings. Then there are precisely r
Q
simple left A-modules, each given by an ni -dimensional K-space which is acted upon by the Mni (K)-
component of A.

Proof. We clearly get at least r non-isomorphic simple left A-modules in this manner. Then by the
2 lemmas above we conclude that these are all of them.

We will also use direct sum notation for products of rings.

2.2 First Approaches to Modular Representations

Theorem 2.14 (Clifford). Suppose V is a simple left KG-module and N / G a normal subgroup.
Then V becomes a semisimple left KN -module with the natural action from the inclusion KN ⊂
KG.

Proof. To show that V is a semisimple left KN -module it suffices to show that Soc(V ) = V . Let
W be a simple left KN -submodule of V . Then for every g ∈ G we have that gW is a simple
left KN -module, since for every h ∈ N we have ngW = g(g −1 ng)W with g −1 ng ∈ N and so
ngW ⊂ gW . Moreover gW is isomorphic to W with −1
P inverse given by multiplication by g .
Now
P V is a simple KG-module and so necessarily g∈G gW = V . This proves the theorem as
g∈G gW = V ⊂ Soc(V ).

We can use this result with some basic group theory to deal with the case when G is a p-group.

Definition 2.15. We say G is a p-group if |G| = pn for some n ≥ 1. Alternatively, by Cauchy’s

theorem, G is a p-group if and only if for all elements x ∈ G we have ord(x) = pnx for some nk ≥ 0.

15
Lemma 2.16. Let G be a p-group. Then there exists a normal subgroup N / G, possibly zero, so
that |G : N | = p.

Proof. Let |G| = pk for some k ≥ 1. We prove this by induction on k, where the base case k = 1
is clear. Recall that the center Z(G) / G is a normal subgroup. We show that the center of a
p-group is nonzero. Let g1P. . . gr be representatives for conjugancy classes of G not in the center.
Then |G| = pk = |Z(G)| + ri=1 |CG (gi )|. We claim p||CG (gi )| for all i = 1 . . . r, from which we will
conclude p||Z(G)|. Since |Z(G)| ≥ 1, we will have Z(G) 6= 0.

Let h ∈ G and suppose ha gi h−a = hb gi h−b . Then ha−b gi hb−a = 1 and necessarily p|b − a
otherwise since ord(h) is a power of p, we may apply Euclid’s algorithm to conclude 1 = hgh−1 .
This shows that p||CG (gi )| for all i = 1 . . . r since h either contributes a multiple of p or nothing to
the conjugancy class of gi . We may suppose G is non-abelian otherwise the lemma is easy. Then
Z(G) is a proper normal subgroup of G so by induction we have a normal subgroup H / G/Z(G)
with [G/Z(G) : H] = p. Let π : G → G/Z(G) be the quotient map. Then π −1 (H) / G is a normal
subgroup with [G : π −1 (H)] = p as required.

Corollary 2.3. If G is a p-group and Char(K) = p then the only simple left KG-module is the
trivial 1-dimensional module Ktriv .

Proof. We prove again by induction on k where |G| = pk . The case k = 1 is clear since we know
fully the representation theory of finite abelian groups. Now let N / G be a normal subgroup of
index p, and suppose ρ : KG → V is an irreducible representation. By Clifford’s theorem we
know ρ|KN : KN → V realizes V as a semisimple left KN -module. By induction all simple left
KN -modules are trivial and so V must be a trivial left KN -module. As a result we are reduced
to considering the action of G/N ∼= Cp on V , which we know to be trivial also from the base case.
This concludes the result.

More generally we can quotient by normal p-subgroups when considering a p-modular represen-
tation. In other words, normal p-subgroups always lie in the kernel of p-modular representations.

Definition 2.17. Let G be a finite group. An element x ∈ G is called a p-element if ord(x) = pnx
for some nk ≥ 1. Otherwise we say x is p-regular. Note that the identity e ∈ G is always p-regular
for all prime p.

Lemma 2.18. Let G be a finite group and g ∈ G. Given a prime p, there are elements u, x ∈ G
so that g = ux = xu, u is a p-element and x is p-regular.

Proof. Write ord(g) = pa m for some m coprime to p and a ∈ N0 . Then pa is coprime to m too
so there exist by Euclid’s algorithm elements r, s ∈ Z so that rpa + sm = 1. As a result we write
a a
g = g sm g p r and let u = g sm and x = g p r . Now u is clearly a p-element, and x is p-regular since
xm = 1 with m coprime to p. Moreover u, x commute as they are powers of g.

Lemma 2.19. Let G be a finite group and K a field. We write [KG, KG] = hab−ba : a, b ∈ KGi ⊂
KG and let x1 . . . xt be representatives for conjugancy classes of G. Then {x1 + [KG, KG] . . . xt +
[KG, KG]} is a basis for KG/[KG, KG].

Proof. Firstly we have the simpler description [KG, KG] = hab − ba : a, b ∈ Gi ⊂ G since the
commutator bracket is bilinear over K and G forms a K-basis for KG. Now for any h, g ∈ G we
have g − h−1 gh = (gh)h−1 − h−1 (gh) ∈ [KG, KG], and so {x1 + [KG, KG] . . . xt + [KG, KG]} spans

16
KG/[KG, KG]. Suppose now that ti=1 ki xi ∈ [KG, KG] for some k1 . . . kt ∈ K. Then we may
P
define for each i = 1 . . . t a K-linear map φi : KG → K sending
(
1 g ∼ xi
φi (g) =
0 otherwise

for all g ∈ G. Then given g, h ∈ G we have φi (gh − hg) = φi (g(hg)g −1 ) − φi (hg) = 0 necessarily
since ∼ is an equivalence. As a resultPφ1 . . . φt vanish on [KG, KG] and induce K-linear maps
on KG/[KG, KG]. However 0 = φi ( tj=1 kj xj ) = ki for all i = 1 . . . t and so indeed {x1 +
[KG, KG] . . . xt + [KG, KG]} are linearly independent and form a basis for KG/[KG, KG].

Lemma 2.20. Let A, B be K-algebras. Then [A ⊕ B, A ⊕ B] ∼ = [A, A] ⊕ [B, B] as direct sums of

K-algebras. Moreover given n ∈ N, we have [Mn (K), Mn (K)] = Mn0 (K) which we use to denote
the zero trace matrices.

Proof. The first claim is a simple consequence of the bilinearity of the commutator bracket. For the
second claim, we clearly have [Mn (K), Mn (K)] ⊂ Mn0 (K) due to the relation T r(AB) = T r(BA) for
matrices A, B ∈ Mn (K). Now if i, j are distinct integers among 1 . . . n we may form the elementary
matrix Eij = [Ei,j , Ej,j ]. We can also form the matrix Eii − Ejj = [Ei,j , Ej,i ]. These matrices span
Mn0 (K) and so we are done.

Lemma 2.21. Let K be an algebraically closed field of characteristic p and m ∈ Mn (K) a matrix
for some n ≥ 1. Then T r(mp ) = T r(m)p .

Proof. We show that m and mp have the same eigenspaces. Suppose (m − λI)r v = 0. Then
0 = (m − λI)pr v = (mp − λp I)r v. One can proceed by the Jordan Canonical Form Theorem.
Alternatively, if λ1 . . . λn are the eigenvalues for m with multiplicity then observe that λp1 . . . λpn are
the eigenvalues for mp with multiplicity, and so
n n
λpi
X X
p
T r(m ) = =( λi )p = T r(m)p
i=1 i=1

just as required.
n
Lemma 2.22. Let K be a field of characteristic p. Suppose x ∈ K with xp = 1 for some n ≥ 0.
n
Then x = 1. Also if xp = 0 for some n ≥ 0 then x = 0 in the same scenario.
n n
Proof. As polynomials in z we have z p −1 = (z−1)p by using the binomial theorem with exponent
n
p repeatedly and the fact that p = 0 in K. Then clearly 1 is the only root of z p − 1 by the division
n n
algorithm for polynomials over fields. For the second part we have (x + 1)p = xp + 1 = 1 which
gives x + 1 = 1 by the first part, and so x = 0.

Theorem 2.23 (Brauer). Let G be a finite group and K an algebraically closed field of character-
istic p. The number of isomorphism classes of simple left KG-modules is the number of p-regular
conjugancy classes in G.

Proof. Let A =L KG/Rad(KG). Then A is semisimple as a left KG-module so by Artin-Wedderburn

we have A ∼ r ∼
Lr = i=1 Mni (K) as K-algebras, for some positive integers n1 . . . nr . Now [A, A] =
0 0
i=1 Mni (K) where for each i = 1 . . . r we write Mni (K) for the zero trace matrices. As a result
dimK (A/[A, A]) = r. Let x1 . . . xt be representatives for conjugancy classes of G. Then we are
also aware that KG/[KG, KG] has a basis {x1 + [KG, KG] . . . xt + [KG, KG]}. Let π : KG → A

17
denote the quotient map and denote S := π −1 ([A, A]). Then in fact S = [KG, KG] + Rad(KG)
simply because π is a ring morphism.
m
Now we get into the meat of the proof. Consider S0 := {x ∈ KG : xp ∈ [KG, KG] for some m ≥
m
0}. We claim that S = S0 . We show S0 ⊂ S, so select an element x ∈ S0 and suppose xp ∈
m m
[KG, KG]. Then π(x)p ∈ [A, A] and so we amy write π(x)p = ri=1 mi where mi ∈ Mn0i (K)
P
for all i = 1 . . . r. Then since A ∼ r
L
m
= i=1 Mni (K) as a direct sum of K-algebras we may write
π(x) = i=1 ni where npi = mi for all i = 1 . . . r. Moreover
Pr
Lr T r(n i ) = 0 for all i since Char(K) = p,
and so we can use Lemma 2.21. As a result π(x) ∈ i=1 Mni (K) ∼0
= [A, A], and so by definition
x ∈ π −1 ([A, A]) = S. Now we show that S ⊂ S0 . Firstly [KG, KG] ⊂ S0 is clear by definition. Also
Rad(KG) ⊂ S0 since Rad(KG) is a nilpotent 2-sided ideal. Thus [KG, KG] + Rad(KG) = S ⊂ S0
since S0 is a K-linear space. We conclude S = S0 .

Write t = s + q so that x1 . . . xs are exactly the representatives for p-regular conjugancy classes,
after reordering. Then we claim that s = r = dimK (A/[A, A]) = dimK (KG/S) where the last
equality follows by linear algebra. Since S = S0 , we need to show that {x1 + S0 . . . xs + S0 } is a
basis for KG/S0 . Now {x1 + S0 . . . xt + S0 } clearly spans KG/S0 . Given g ∈ G use 2.18 to write
m
g = ux for commuting u, x where u is a p-element and x is p-regular. For some m ≥ 0 we have g p =
m m m m m m
up xp = xp and so 0 = g p − xp = (g − x)p which implies g − x ∈ S0 . Therefore in KG/S0
every conjugancy class is equivalent to a p-regular conjucancy P class and so {x1 + S0 . . . xs + S0 }
spans KG/S0 . We move on to linear independence. Suppose si=1 ki xi ∈ S0 for some k1 . . . ks ∈ K.
m m m
Then for some m ≥ 0 we have ( si=1 ki xi )p = si=1 kip xpi ∈ [KG, KG]. Then by group theory,
P P
M M
for some large M ≥ m we have xpi = xi for all i = 1 . . . s and so si=1 kip xi ∈ [KG, KG]. Note
P
here that [KG, KG] is closed under taking pth powers. Since {x1 + [KG, KG] . . . xt + [KG, KG]} is
M
a basis for KG/[KG, KG] then kip = 0 for all i = 1 . . . s and so necessarily ki = 0 for all i. This
concludes the theorem since s is the number of p-regular classes and r is the number of isomorphism
classes of simple left KG-modules by Corollary 2.2.

Example 2.24. Choose a prime p and let G = SL2 (p) be the group of 2 × 2 matrices with deter-
minant 1 over the field Fp . Then there are p simple left KG-modules when Char(K) = p and K is
algebraically closed, and they are V1 . . . Vp given below.

Let Vm ⊂ K[X, Y ] be the K-subspace of homogeneous polynomials of degree m − 1. Clearly

dimK (Vm ) = m for all m. Given an element
 
a b
g= ∈ SL2 (p)
c d

We define the representation ρm : KG → GL(Vm ) by setting ρm (g)F (X, Y ) = F (aX + bY, cX +

dY ) and extending this action linearly. We can show V1 . . . Vp are simple by considering
   
1 0 1 1
u= v=
1 1 0 1
an their action on Vm with the obvious monomial basis for all m = 1 . . . p. Clearly V1 . . . Vp are
non-isomorphic as they have different dimensions. The hard part is showing by means of Brauer’s
theorem that there are exactly p distinct p-regular conjugancy classes in G.

18
2.3 Indecomposable Modules
Definition 2.25. Let R be a ring and V a left R-module. We say V is indecomposable if V cannot
be written as a direct sum V = A ⊕ B for some nonzero left R-modules A, B.

Indecomposable modules are a phenomena of modular representation theory. In the nonmodular

case all finitely generated left KG-modules are semisimple. For semisimple modules, indecompos-
ability is the same as simplicity.

Example 2.26. We show an example where infinitely many isomorphism classes of finitely gener-
ated indecomposable left modules exists. This showcases their wild behaviour (in the modular case).

Let p be any prime and consider G = Cp × Cp . Write G = ha, bi for obvious generators a, b. We
construct indecomposable KG-modules of dimension 2n for all n ≥ 1, when Char(K) = p and K is
algebraically closed. Let x1 . . . xn , y1 . . . yn be a basis for V2n := K 2n . We define linear maps L1 , L2
on V2n as follows. Send L1 (xi ) = yi and L1 (yi ) = 0 for all i = 1 . . . n. Also send L2 (xi ) = yi+1 for
all i = 1 . . . n − 1 and send L2 (yi ) = L2 (xn ) = 0 for all i = 1 . . . n.

We build the representation ρ : G → GL(V2n ) by sending ρ(a) = I + L1 and ρ(b) = I + L2 .

Since L21 = L22 = L1 L2 = L2 L1 = 0 we have (I + L1 )p = (I + L2 )p = I and I + L1 commutes with
I + L2 . As a result this representation is well-defined, of dimension 2n. It remains to show that
V2n is indecomposable as a left KG-module. For this we make use of Theorem 2.30 below, so that
is suffices to show that EndKG (V2n ) is a local K-algebra. In the basis above we may write
   
In 0 In 0
L1 = L2 =
In In N In
where N is the nilpotent matrix with 1 in the diagonal one place below the main diagonal. A
K-linear endomorphism E lies in EndKG (V2n ) if and only if L1 E = EL1 and L2 E = EL2 . This
translates to E taking the form
 
A 0
E=
B A
for any n × n matrices A, B with K-entries so that AN = N A. Note that E is invertible if
and only if A is invertible. Supposing A is not invertible, we must show that E is nilpotent. The
relation AN = N A simply means that shifting entries down is the same as shifting entries to the
left in A, and so A takes a specific form where all entries along any usual diagonal are the same.
In particular since A, N commute there is a basis where they are both upper triangular, and A has
the same entries in the main diagonal. If A where not invertible, the entries in the main diagonal
would have to be zero in this basis and so A is nilpotent. We can now conclude by Lemma 2.29.

Example 2.27. Let p be any prime and consider G = Cpn for some n ≥ 1. We classify the
indecomposable KG-modules, for Char(K) = p. Let ρ : G → GLn (K) be an indecomposable repre-
sentation, and choose a generator G = hxi. The representation is determined by the image of the
generators, in this case ρ(x). Apply the Jordan Normal Form Theorem to ρ(x) and observe that
ρ(x) ∼ J(ρ(x)) must have a single Jordan block, since the Jordan blocks of J(ρ(x)) give a direct
sum decomposition of K n into indecomposable left submodules.
n n
Necessarily we have ρ(x)p = I so for any eigenvalue λ ∈ K of ρ(x) we have λp = 1. However
Char(K) = p and so λ = 1 is the only possible eigenvalue. We conclude the classification as for

19
every m ≥ 1 we get a single indecomposable KG-module of dimension n where ρ(x) is given by the
Jordan block Jn (1) up to base change.

We have a unique factorization theorem for direct sum decomposition into indecomposable
modules. We need some lemmas first.

Definition 2.28. We call a K-algebra A local if A/Rad(A) ∼

= K as K-algebras.

Lemma 2.29. An Artinian K-algebra A is local if and only if for all x ∈ A either x is invertible
or x is nilpotent.

Proof. Suppose A/Rad(A) ∼ = K. Then given x ∈ A we may write x = r + k · 1 for some r ∈ Rad(A)
and k ∈ K. If k = 0 then x = r ∈ Rad(A) is nilpotent since Rad(A) is a 2-sided nilpotent ideal.
Otherwise suppose Pk 6= 0 and rn = 0 for some n ∈ N0 . We use a geometric
Pn−1 series argument. Consider
the element y = k ( n−1
1 r i r r i Pn r i−1
i=0 (− k ) ). One sees that xy = k( k +1)y = ( i=0 (− k ) )+ i=1 (− k) = 1,
and similarly yx = 1. Therefore x is invertible with inverse y.

Now we prove the converse. A/Rad(A) is a semisimple Artinian K-algebra so by Artin-

Wedderburn ∼ Qt
Qwe have A/Rad(A) = i=1 Mni (K) for some n1 . . . nt positive integers. Now every
element in ti=1QMni (K) ∼ = A/Rad(A) is again either nilpotent or invertible. Suppose t ≥ 2. Then
(In1 , 0 . . . 0) ∈ ti=1 Mni (K) is neither nilpotent nor invertible so we get a contradiction. Thus
t = 1 and A/Rad(A) ∼ = Mn1 (K) is a simple ring. Suppose n1 ≥ 2 now, and consider the diagonal
element diag(1, 0 . . . 0) ∈ Mn1 (K). This is neither nilpotent nor invertible so we get a contradiction.
Therefore n1 = 1 and A/Rad(A) ∼ = K as K-algebras.

Theorem 2.30. Let A be an Artinian K-algebra and M a left A-module. Then M is indecompos-
able if and only if EndA (M ) is local.

Proof. Suppose M is an indecomposable as a left A-module. Let ρ ∈ EndA (M ) and k ∈ K an

eigenvalue of ρ. Consider the generalized k-eigenspace of ρ given by Mρk := {m ∈ M : (ρ − kI)n m =
0 for some n ∈ N}. Then M = k∈K Mρk as a consequence of the Jordan Normal Form Theorem
L
applied to ρ. Now this is a decomposition as left A-modules since ρ is an A-endomorphism. Since
M is indecomposable we must have that ρ has a single eigenvalue k. As a consequence of the
Jordan Normal Form Theorem we have ρ ∼ Jn (k) where n = dimK (M ). Therefore k = 0 =⇒ ρ
is nilpotent and k 6= 0 =⇒ ρ is invertible. Convlude that EndA (M ) is local.

Suppose EndA (M ) is local. Given a proper decomposition M ∼= N ⊕ L as left A-modules, let

π1 ∈ EndA (M ) be the projection onto N . Then π1 is neither invertible nor nilpotent and so we
get a contradiction.

Theorem 2.31 (Krull-Schmidt). Let K be an algebraically closed field and let A be an LArtinian
K-algebra. Let M be a left A-module. Suppose we have direct sum decompositions M ∼
= i=1 Ui ∼
r
=
Ls
j=1 V j with U i , Vj indecomposable for all i = 1 . . . r and j = 1 . . . s. Then r = s and {U1 . . . Ur } ∼
{V1 . . . Vs }.

Proof. We prove this by induction on r, where the base case r = 1 is clear. Take M ∼ = i=1 Ui ∼
Lr
=
L s
j=1 jV as in the statement and assume the statement is true for all direct sum decompositions
into less than r indecomposables. Consider the inclusions iU j : Uj → M for j = 1 . . . r and
ij : Vj → M for j = 1 . . . s. Consider also the projections πjU : M → Uj for j = 1 . . . r and
V

πjV : M → Vj for j = 1 . . . s. Define ρj = π1U ◦ iVj ◦ πjV ◦ iU 1 : U1 → U1 for each j = 1 . . . s.

20
Now EndK (U1 ) is local so ρj is either nilpotent or invertible for each j = 1 . . . s. One verifies
that ρ1 + . . . + ρs = idU1 which is invertible so at least one of the ρj is invertible. Then it is an
isomorphism so π1U ◦ iVj : Vj → U1 is surjective and πjV ◦ iU 1 : U1 → Vj is injective. One applies
V U U V 2
the same trick to j = πj ◦ i1 ◦ π1 ◦ ij : Vj → Vj since j = j and so necessarily j is invertible,
hence
Lr an∼isomorphism.
L This is is sufficient to establish U1 ∼
= Vj . Then one naturally obtains
i=2 Ui = k6=j Vk to which we apply the inductive hypothesis and complete the proof.

21
3 Brauer Theory
3.1 Brauer Characters
In this chapter assume again that K is an algebraically closed field and G is a finite group.
Definition 3.1. Given a representation ρ : G → GLn (K) its character is defined as the map
χ : G → K sending χ(g) = T r(ρ(g)). This is well-defined on isomorphism classes of representations
since the trace is invariant under base change.
This character map is used in classical representation theory over C to classify irreducible rep-
resentations. In the Modular case it is not so useful, and we need to use a more sophisticated
character called the Brauer character, defined shortly. However we will make use of this ”standard
character” to prove linear indpendence of Brauer Characters. It also turns out that these standard
characters are solely determined by their values on p-regular conjugancy classes of G.

Proposition 3.2. Let ρ : KG → EndK (V ) be a representation with associated standard character

χV . Then χV is solely determined by its values on p-regular conjugancy classes of G.

Proof. For any element g ∈ G we can write g = ux for commuting u, x where u is a p-element and x is
p-regular. Then we claim χV (g) = χ(x). Let ρ : KG → EndK (V ) be the associated representation.
Since ρ(u), ρ(x) commute, they are simultaneously triangularizable by linear algebra. But ρ(u) is
a p-element so all of its eigenvalues will be 1, and therefore we may write ρ(u) = I + N where
N is a nilpotent, strictly upper triangular matrix. Now χV (g) = T r((I + N )ρ(x)) = T r(ρ(x)) +
T r(N ρ(x)) = T r(ρ(x)) = χV (x) as required, since N ρ(x) is again strictly upper triangular and
hence nilpotent.

Let O denote the ring of all algebraic integers in Q. Let p be a prime and choose a maxi-
mal ideal p 3 p in O. Then O/p is a field of characteristic p which is algebraically closed. Let
Up := {z ∈ C : z m = 1 for some m ∈ N with p 6 |m} denote the group of p-regular roots of unity in
C. From the quotient map π : O → O/p we get an induced map π ∗ : Up → (O/p)× of multiplicative
groups by restriction. This is clear as π ∗ (z) = 0 =⇒ p|N (z) for any root of unity z, however
N (z) = 1 gives us a contradiction.

Proposition 3.3. The map π ∗ : Up → (O/p)× is a group isomorphism.

Proof. Suppose π ∗ (ζr ) = 1 for some rth root of unity ζr ∈ Up , where r is necessarily coprime to
p. Then 1 − ζr Q∈ p and so taking norms to Z we see that p|N (1 − ζr ). However we can compute
explicitly that r−1 i
i=1 (1 − ζr ) = r and so N (1 − ζr )|r and so this gives us a contradiction unless
ζr = 1. This proves injectivity.

Now we prove surjectivity. Given x ∈ (O/p)× , we have that x is algebraic over Fp and so Fp (x)
is a finite field extending Fp , with pN elements for N the degree of the minimal polynomial of x in
Fp [T ]. Now Fp (x) is a field so as a group |Fp (x)× | = pN − 1 and so x is a rth root of unity in Fp (x)
for some r|pN − 1 minimal. Then x is a root of T r − 1 in the field (O/p), of which there are r roots
exactly by the division algorithm for polynomials over fields, with multiplicity. Now r is coprime
to p since it was chosen to be minimal and we have the polynomial relation xps − 1 = (xs − 1)p
otherwise. As a result, it follows by injectivity of π ∗ that we get r distinct rth roots of unity from
Up , one of which must be x.

22
 Fix an identification K := Fp ∼ = O/p. Given a p-modular representation ρ : G → GLn (K)
denote Gp0 as the subset of p-regular elements in G. For any g ∈ Gp0 , ρ(g) has some eigenvalues
u1 . . . un in K, with multiplicity. Then let vi = (π ∗ )−1 (ui ) for all i = 1 . . . n denote the com-
plex roots Pnof unity given by the identification above. Finally we can define the Brauer character
φ(g) := i=1 vi . This is well defined up to action of the Galois group Gal(K : Fp ) since changing
identification above amounts to an automophism of K.

We may define Brauer characters for more general algebraically closed fields K of characteristic
p. If we construct an isomorphism π ∗ : Up → (O/m)× first, for some maximal ideal m above p, then
by general field theory there is a (non-canonical) injective field morphism i : O/m → K. Then we
consider the injective group morphism i ◦ π ∗ : Up → K × . Algebraic elements in K, in particular
eigenvalues of finite-dimensional linear transformations, will lie in the image of i ◦ π ∗ , and can be
pulled back uniquely to Up .
Proposition 3.4. Let K be an algebraically closed field of characteristic p and let V, W be finitely
generated left KG-modules with Brauer characters φV , φW , both constructed from the same injective
group morphism π : Up → K × , fixed throughout. We have the following.
1. φV (1) = dim(V )
2. φV : Gp0 → C is a class function.

3. φV (g −1 ) = φV (g) for all g ∈ Gp0 .

4. φV ⊗W = φV φW
P
5. φV = S simple nS φS where nS is the multiplicity of S as a composition factor of V and the
sum runs over all simple left KG-modules.
Proof. Let ρV , ρW be representations associated to V, W .
1. The eigenvalues of I = ρV (e) are dim(V ) copies of 1. These uniquely lift to π −1 (1) = 1 and
so indeed φV (1) = dim(V ).
2. If g, h are conjugate p-regular elements in G then ρV (g) and ρV (h) have the same eigenvalues,
and then it becomes clear that φV (g) = φV (h).
3. If rhoV (g) has eigenvalues u1 . . . ut then rhoV (g −1 ) has eigenvalues u−1 −1
1 . . . ut . Now π is a
group morphism so φV (g −1 ) = ti=1 π −1 (u−1 t −1 −1 = ( t −1
P P P
i )= i=1 π (ui ) i=1 π (ui )) = φV (g),
since inverses of roots of unity are just their complex conjugate.
4. For g a p-regular element we have ρV ⊗W (g) = ρV (g) ⊗ ρW (g) where the later is the Kronecker
product of matrices, if we choose the standard basis for the tensor product. Suppose u1 . . . un
are eigenvalues for ρV (g) and v1 . . . vm are eigenvalues
Pn Pmfor ρ−1
W (g). ThenPρnV ⊗W−1
(g) hasPeigenval-
ues u1 v1 . . . un vm and so we have φV ⊗W (g) = i=1 j=1 π (ui vj ) = i=1 π (ui ) m −1
j=1 π (vj ) =
φV (g)φW (g) as required.
5. Given a p-regular element g consider the cyclic subgroup hgi ⊂ G and its action on V . Choose
a composition series 0 = V0 ⊂ . . . Vn = V for V as a finitely generated left KG-module.
However, V is necessarily semisimple as a finitely
Lt generated left Khgi-module, and so this
∼ ni
composition series becomes a direct sum V = i=1 Si as left Khgi-modules, if S1 . . . St are
the simple leftPKG-modules. Here ni is the multiplicity of Si as a composition factor of V .
Now φV (g) = ti=1 ni φsi (g) becomes clear as Brauer characters are additive over direct sums.

23
Definition 3.5. Let g1 . . . gr be representatives for the p-regular conjugancy classes of G. Let
φ1 . . . φr be the Brauer characters for the simple left KG-modules. The r × r matrix (φi (gj ))i,j=1...r
is called the Brauer character table of G at the prime p. We also use the notation IBRp (G) =
{φ1 . . . φr } to denote the set of irreducible Brauer characters of G.

Lemma 3.6. Let G be a finite group and p a prime. Given an algebraically closed field K of
characteristic p, the standard characters of simple left KG-modules forms a basis of the p-regular
class functions with K-coefficients.

Proof. Let S1 . . . St be the simple left KG-modules for an algebraically closed field K of character-
istic p. Let ρi : KG → End Qt K (Si ) denote the corresponding representations for i = 1 . . . t. Recall
∼ ∼
that KG/Rad(KG) = i=1 Mni (K) where naturally EndK (Si ) = Mni (K) and KG/Rad(KG)
acts Qon Si via matrices in its ith component. In other words each pi is surjective with kernel
π −1 ( j6=i Mni (K)). Due to this description we can easily select elements xi ∈ EndK (Si ) for all
i = 1 . . . t so that T r(xi ) = 1. These pull back to elements X1 . . . Xt ∈ KG so that χj (Xi ) = δi,j for
all pairs i, j. It follows that the standard characters of simple left KG-modules span the space of
p-regular class functions, and are therefore linearly independent by dimension considerations. Note
that p-elements in G lie in the kernel of all representations so X1 . . . Xt can be chosen as K-linear
combinations of p-regular elements in G.

Definition 3.7. Let K be an algebraically closed field of characteristic p with a choice of injection
π : Up → K × . Suppose φV is a Brauer character for some left KG-module V , where G is a
finite group. We define φ∗V , the reduction of φV through π, as a K-valued functionP on p-regular
classes of G as follows. Given g ∈ G0p a p-regular element, we will have φV (g) = ti=1 ui for some
u1 . . . ut ∈ Up . Then we define φ∗V (g) = ti=1 π(ui ).
P

Lemma 3.8. Let K be an algebraically closed field of characteristic p with a choice of injection
π : Up → K × . Suppose φV is a Brauer character for some left KG-module V , where G is a finite
group. Then φ∗V = χV , the standard K-valued character of V .

Proof. Let π : Up → K × be a choice of injection. We show φ∗V = χV and that φ∗V is well defined.
Now π is defined via a composition Up → (O/m)× → K × . Here the first map is an isomorphism
induced by reduction modulo a maximal ideal m above p in the ring of all integers O, and therefore
also extends to an additive map if we send 0 to 0. The latter map is also induced by some non-
canonical injection of fields O/m → K, and is also additive if we send 0 to 0. This demonstrates
well-definedness, and given a p-regular element g ∈ G, let ρ : KG → EndK (V ) be the Pmassociated
representation, and u . . . u the eigenvalues of ρ(g) with multiplicity. Then φ (g) = −1
i m V i=1 π (ui )
so φ∗V (g) = m
P
i=1 ui = χV (g) as required.

Lemma 3.9. Let E = Q(a1 . . . at ) be a number field, so that a1 . . . at ∈ Q. Then there exists b ∈ E
so that ba1 . . . bat generates the ring of integers OE .

Proof. Consider the fractional ideal a := (a1 . . . at )OE . Then it is a theorem that the fractional
ideals of OE form a group, so there exists unique fractional ideal b so that ab = OE . Therefore
there exists a ∈ a, b ∈ b so that
Pab = 1. Necessarily ba1 . . . bat ∈ OE and clearly these
Pt generate the
t
ring of integers OE since a = i=1 ri ai for some r1 . . . rr ∈ OE and so 1 = ab = i=1 ri bai .

24
Lemma 3.10 (Nakayama). Let m be the maximal ideal of a commutative local ring R and let M
be a finitely generated R-module. If mM = M then M = 0. Recall that a commutative ring is local
if it has a unique maximal ideal.

M over R. Write vn = ni=1 mi vi for some

P
Proof. Let {v1 . . . vn } be a minimal generating set of P
n−1
m1 . . . mn ∈ m, since mM = M . Then vn (1 − mn ) = i=1 mi vi and 1 − mn is a unit in R. As a
result we have a smaller generating set {v1 . . . vn−1 } which leads to a contradiction unless n = 0
which implies M = 0.

Lemma 3.11. Let O denote the ring of all algebraic integers and m ⊂ O a maximal ideal. Let
V be a Q-vector space and let W be a finitely generated Om -submodule of V . Then there exist
w1 . . . wn ∈ W so that
1. W = ni=1 Om wi
P

2. w1 . . . wn are linearly independent over (Q)

Proof. W/mW is a non-empty vector space over Om /m otherwise W = mW which gives W = 0
by Nakayama’s lemma. Choose a basis v1 . . . vn for W/mW over Om /m. Lift to some elements
w1 . . . wn ∈ W and let U = ni=1 wi Om be a Om -submodule of W . Then clearly U + mW = W and
P
W/U = m(W/U ). By applying PNakayama’s lemma once more we see that W/U = 0 and so W = U
necessarily. Finally suppose ni=1 ai wi = 0 for a1 . . . an ∈ Q. Then working in the number field
E = Q(a1 . .P . an ) we see that there exists some element b ∈ OE so that ba1 . . . ban ∈ OE generate
OE . Then ni=1 bai wi = 0 with somePbai not in the ideal m, since ba1 . . . ban must also generate
O. Reducing modulo m, we see that ni=1 bai vi = 0 is a nontrivial expression for 0 which gives a
contradiction since v1 . . . vn is a basis for W/mW . This concludes the lemma.

Theorem 3.12. The irreducible Brauer characters of G are linearly independent over C in the
space of all p-regular class functions. In other words the determinant of the Brauer character table
is nonzero.

Proof. We prove this by contradiction. Let S1 . . . St be the simple left KG-modules for K an
algebraically closed field of characteristic p. Choose an injective morphism π1∗ : Up → K × and
let φ1 . . . φt be corresponding Brauer characters for the simple left KG-modules. Recall that the
Brauer characters φ1 . . . φt take values in Q and so they
P are linearly independent over C if and only
if they are linearly independent over Q. Suppose ti=1 ai φi = 0 for a1 . . . at ∈ Q. Consider the
algebraic extension E = Q(a1 . . . at ).PBy Lemma 3.9 there exists b ∈ E so that ba1 . . . bat generates
the ring of integers OE . As a result ti=1 bai φi = 0. Choose a maximalP ideal m ⊂ O above p in the
ring of integers of Q and a lift π2∗ : U → O/m. Then the expression ti=1 bai (φi )∗2 = 0 still holds in
O/m after associating OE ⊂ O and taking the quotient. Moreover one of bai is nonzero since these
generate OE , so in particular some OE -combination of ba1 . . . bat is 1. Here (φi )∗2 is the reduction
of the Brauer character through π2∗ , which is an isomorphism

Now thePmap π1∗ ◦ (π2∗ )−1 := π : (O/m)× → K × is an injective group morphism and yields the
expression ti=1 π(bai )(φi )∗1 = 0 where (φi )∗1 are the reductions of the Brauer characters through
π1∗ . Again one of the (bai ) is nonzero, but (φi )∗1 = χi for all i = 1 . . . t in fact, and we know these
to be linearly independent over K by Lemma 3.6. This yields the necessary contradiction and we
are done.

This is a very important theorem showing that Brauer characters are sufficient in computing
the composition factors of any finitely generated left KG-module.

25
Definition 3.13. Let K be a number field, and let ρ : G → GLn (C) be a complex representation.
We say ρ is realizable over OK if there is a change of basis B ∈ GLn (C) so that Bρ(g)B −1 ∈
GLn (OK ) for all g ∈ G. Let mp be a maximal ideal above p in OK . Then the induced representation
ρ ◦ B −1 : G → GLn (OK /mp ) is called the reduction modulo mp .
B ◦^

The definition of realizability extends to any subring D of C. We can then reduce modulo any
ideal of D. Our aim is to take a complex representation and realize it over some ring where a
certain integer prime p is not a unit. We will then get an induced representation over a field of
characteristic p by reduction.

Theorem 3.14 (Brauer). Let p be a prime and ρ : G → GLn (C) be a complex representation of a
finite group G. Choose a maximal ideal m above p in O. Then ρ is realizable over the localization
Om . The reduction modulo m given by ρp : G → GLn (Om /m) is p-modular, and its Brauer character
φρp is the restriction of the complex character χρp to p-regular classes in G.

Proof. We first prove an intermediate step, that ρ can be realized over Q. Since CG is semisimple,
it suffices to assume that ρ is irreducible. Q is also an algebraically closed field of characteristic
0 and so QG has the same number of simple left modules as CG. Let S1 . . . St be the simple left
QG-modules up to isomorphism. Then C ⊗Q S1 . . . C ⊗Q St are pairwise non-isomorphic simple left
CG-modules, since by Artin-Wedderburn we can realize Si ∼ = Mni (Q) and C ⊗Q Si ∼ = Mni (C) with
∼ L t
CG = i=1 Mni (C).

Let V be a finitely generated left QG-module with representation ρ : QG → EndQ (V ). Choose

a basis B = {b1 . . . bn } for V over Q and let W = hgv : g ∈ G, v ∈ BiOm be a finitely generated left
Om -submodule of V , which is G-invariant. By Lemma 3.11 there are w1 . . . wn linearly independent
over Q which generate W over Om . Then w1 . . . wn is a basis for V over Q and in this basis ρ(g)
has entries in Om for all g ∈ G. Then by definition ρ is realizable over Om . The last part of this
theorem is immediate from the definitions.

Corollary 3.1. Let G be a finite group and p a prime. Then there is a number field L and prime
ideal p0 ⊂ OL containing p so that every complex irreducible representation of G is realizable over
OL,p0 , and every simple left KG-module is realizable over k 0 := OL /p0 OL , where K is an algebraic
closure of k 0 .

Proof. Using Brauers Theorem we may find a maximal ideal m above p in O, so that all irreducible
complex representations are realizable over Om . However G is finite and there are finitely many such
representations ρ1 . . . ρn , so adjoining the entries of all matrices ρi (g) is some basis (fixed for each
representation) for all i = 1 . . . n and g ∈ G will yield a number field L1 . Naturally, p := m ∩ OL1
is a prime ideal above p and every irreducible complex representation is realizable over OL1 ,p .

Now K := O/m is an algebraically closed field of characteristic p. Let φ1 . . . φm be the irreducible

p-modular representations over the field K. Adjoin representatives is O for every entry in the matrix
φj (g) (fixing a basis for each representation) for every j = 1 . . . m and g ∈ G to get a number field
L2 . Let q := m ∩ OL2 be a prime ideal. Then k := OL2 /qOL2 is a subfield of K, and with this
identification all simple left KG-modules are realizable over k. We conclude the statement of the
corollary by taking the compositum L = L1 ∪L2 , again a number field, and considering p0 := m∩OL
an k 0 := OL /p0 OL .

26
Definition 3.15. Let G be finite group. Define IRR(G) as the set of complex irreducible characters
of G. Fix a prime p and let IBRp (G) be the associated irreducible Brauer characters at p. Let
χ ∈ IRR(G). Then as in Brauer’s Theorem, we can reduce χ to a Brauer character χ◦ and write
◦
P
χ = φ∈IBRp (G) dχ,φ φ for positive integers dχ,φ . The decomposition matrix for G at the prime p
is defined by Dp (G) = {dχ,φ }χ∈IRR(G),φ∈IBRp (G) .

Proposition 3.16. We have the following properties regarding the decomposition matrix defined
above.

1. rank(Dp (G)) = |IBRp (G)|

2. ∀φ ∈ IBRp (G) there is some χ ∈ IRR(G) so that dχ,φ 6= 0.

Proof. 1. This is a question of comparing dimensions. Clearly rank(Dp (G)) ≤ |IBRp (G)|. By
Brauer’s Theorem the reduction χ◦ of a complex character is simply the restriction of χ to
p-regular classes in G. Now the space of p-regular class functions has dimension |IBRp (G)|
and so the equality rank(Dp (G)) = |IBRp (G)| becomes immediate.

2. This follows immediately from part 1.

3.2 Projective Indecomposable Modules

We introduce yet another class of modules which will help tie in the study of simple modules with
indecomposable modules.

Definition 3.17. A left R-module F is free if we have an isomorphism F ∼

L
= i∈I R for some
index set I.

Definition 3.18. A left R-module P is projective if we can solve the following diagram.
f
V W
g
e
P
For all left R-modules V, W and morphisms of left R-modules f, g with f surjective. That is, given
the solid diagram above, we may find a morphism of left R-modules e : P → V , not necessarily
unique, so that the completed diagram commutes.

Proposition 3.19. The following are equivalent.

1. M is a projective left R-module.

2. For every surjection f : N → M of left R-modules we have N ∼

= M ⊕ ker(f ).

3. M ⊕ N is free for some left R-module N .

Proof. 1. Assume that M is a projective left R-module. Take a surjection f : N → M of left

R-modules. Then by definition there is some map e : M → N of left R-modules so that
f ◦ e = idM . Consider the injection i : ker(f ) → N and form the morphism of left R-modules
(e, i) : M ⊕ ker(f ) → N . We claim this is an isomorphism. Suppose for injectivity that
e(m) + i(n) = 0. Then applying f we see that 0 = (f ◦ e)(m) + (f ◦ i)(n) = m and so

27
 e(m) = 0 = i(n). However i injective so n = 0 and finally (m, n) = 0.

For surjectivity let n ∈ N . Then n−(e◦f )(n) ∈ ker(f ) since we observe f (n)−(f ◦e◦f )(n) =
f (n)−f (n) = 0. As a result we see that (e, i)(f (n), n−(e◦f )(n)) = (e◦f )(n)−(e◦f )(n)+n = n
as required.

2. Consider the free module RM := m∈M R indexed by M . We have a surjective map of left
L
R-modules f : RM → M where the identity em in the mth component is sent to m ∈ M . By
assumption we have RM ∼ = M ⊕ ker(f ) as required.

3. Suppose M ⊕ N ∼
L
= i∈I R is a free module. Let f : V → W be a surjective map of left
R-modules and g : L M → W a map of left R-modules. We prove first that free modules are
projective. Let h : i∈I R → W be a map of L left R-modules, and suppose h(ei ) = wi ∈ W
where ei is the identity in the ith component in i∈I R. Since f is surjective we L have elements
vi ∈ V so that f (vi ) = wi for all i ∈ I. Then we construct the map e : i∈I R → V by
sending e(ei ) = vi for all i ∈ I and this is clearly well defined by freeness and satisfies f ◦e = h.

L L
We define h = g ◦ π1 where π1 : i∈I R → M is the first projection. Let e L : i∈I R → V
be a map of left R modules so that f ◦ e = h = g ◦ π1 . If we let i : M → i∈I R be the
natural inclusion then we see that pi1 ◦ i = idM and so f ◦ e ◦ i = h ◦ i = g. Therefore the
map e ◦ i : M → V is exactly what we needed.

Proposition 3.20. Let A be an Artinian ring, and P a finitely generated left A-module. Then P
is a projective indecomposable left A-module if and only if P is a direct summand of A, so that
A∼ = P ⊕ N for some left module N .

Proof. The converse follows directly from Proposition 3.19. We know that there is some
Lr left A-
∼ k ∼
module N so that P ⊕ N = A for some k ∈ N since P is finitely generated. Let A = i=1 Ui be
the decomposition of A into indecomposable left A-submodules. Then by Krull-Schmidt we know
that P ∼
= Ui for some i = i . . . r and so we conclude.

Theorem 3.21. Let A be an Artinian K-algebra where K is algebraically closed. Let P be a

projective indecomposable left A-module. Then we have the following.

1. P/Rad(P ) is simple.

2. π : P → P/Rad(P ) establishes a bijection between simple left A-modules and projective

indecomposable left A-modules.

Proof. 1. We claim that EndK (P/Rad(P )) ∼ = EndK (P )/Rad(A)EndK (P ) as left A-modules.

Let φ ∈ EndK (P ). Then φ(Rad(P )) ⊂ Rad(P ) since φ is a morphism of left A-modules and
Rad(P ) = Rad(A)P . As a result φ descends to an endomorphism H(φ) ∈ EndK (P/Rad(P ))
and this produces a morphism of left A-modules H : EndK (P ) → EndK (P/Rad(P )).
Now Rad(A)EndK (P ) clearly lies in the kernel of this map. We wish to prove that H is
surjective, so take an endomorphism ψ ∈ EndK (P/Rad(P )) and consider the projection
π : P → P/Rad(P ) which is surjective. Consider also the map ψ ◦ π : P → P/Rad(P ).
By projectivity we get some map of left A-modules e : P → P so that π ◦ e = ψ ◦ π. It
becomes clear that e ∈ EndK (P ) and H(e) = ψ. Now P is indecomposable so necessarily
EndK (P )/Rad(A)EndK (P ) ∼ e : EndK (P )/Rad(A)EndK (P ) →
= K, but we have a surjection H

28
 EndK (P/Rad(P )) so either EndK (P/Rad(P )) = 0 or EndK (P/Rad(P )) ∼= K. We cannot
have EndK (P/Rad(P )) = 0 and so EndK (P/Rad(P )) = K which implies P/Rad(P ) is simple
by Schur’s lemma.

2. By 1, this map is well defined. We prove it is injective. Suppose P1 /Rad(P1 ) ∼

= P2 /Rad(P2 ).
Consider the projections π1 : P1 → P1 /Rad(P1 ) and π1 : P2 → P2 /Rad(P2 ). By projectivity
we get some maps of left A-modules u : P1 → P2 and v : P2 → P1 so that π2 ◦ u = π1
and π1 ◦ v = π2 . Then π2 ◦ u ◦ v = π1 ◦ v = π2 . Now Rad(P1 ), Rad(P2 ) are maximal left
submodules of P1 , P2 respectively, and so by definition they must be the unique maximal
left submodules. We also have Im(u) + Rad(P2 ) = P2 and Im(v) + Rad(P1 ) = P1 and so
necessarily Im(u) = P2 and Im(v) = P1 . This proves u, v are surjective. However P1 , P2 are
indecomposable and projective so necessarily Ker(u) = Ker(v) = 0 and so u, v are isomor-
phisms as required.

Write AL ∼ Lr
= i=1 Pi L
as a decomposition L into projective indecomposables. Then Rad(A) =
Rad(A) ri=1 Pi = r
i=1 Rad(A)P i = r
i=1 Rad(Pi ) by using properties of the radical,
since A is Artinian and each P1 . . . Pr is finitely generated. Then we have A/Rad(A) ∼ =
Lt dim(Sj ) ∼ Lr
S
j=1 j = P
i=1 i /Rad(P i ) where S 1 . . . St are the simple left A-modules. A simple
module is indecomposable, so by Krull-Schmidt we have that Sj ∼ = Pj /Rad(Pj ) for each
j = 1 . . . t after rearrangement and so necessarily t = r and we are done.

Definition 3.22. Let A be an Artinian K-algebra where K is algebraically closed. Let S, T be

simple left A-modules. Let PT be the projective indecomposable left A-module associated to T ,
called the projective cover of T . We denote by CST the multiplicity of S as a composition factor
of PT in its Jordan-Hölder composition series. We call C(A) := {CST }S,T simple left A-modules the
Cartan matrix of A.

For a finite group G and an algebraically closed field K of characteristic p of the form considered
in this chapter, we write Cp (G) = C(KG).

Definition 3.23. Let K be an algebraically closed field and let G be a finite group. A subfield
L ⊂ K is called a splitting field for G if every simple left KG-module can be realized over L.

Now we search for a method to lift modular representations to representations over fields of
characteristic 0. This is where idempotent lifting kicks in from chapter 1.

Definition 3.24. Let p be a prime and G a finite group. Suppose we have the following data

1. A splitting field F for G of characteristic 0, with a discrete valuation v.

2. The valuation ring R of F at v with unique maximal ideal πR.

3. K = R/πR a field of characteristic p, also a splitting field for G.

Then we call (F, R, K) a p-modular system for G.

Definition 3.25. Let R be a valuation ring with valuation v and let G be a finite group. Let ||v
be the absolute value
P corresponding to v. Then we define an absolute value || on the group ring
RG by setting | g∈G ag g| = maxg∈G (|ag |v ). The reader can verify that this is an absolute value.
Moreover if R is complete with respect to ||v then RG is complete with respect to ||.

29
Lemma 3.26. Let R be a valuation ring with valuation v and unique maximal ideal πR so that R
is complete with respect to the induced absolute value. Let G be a finite group. Then any idempotent
e ∈ (R/πR)G can be lifted to an idempotent f ∈ RG so that e = f + πRG, and if e is primitive
then so is f .

Proof. Suppose we have lifted to ek ∈ RG/π k RG so that e = ek + πRG by virtue of Proposition

1.31. Then we claim that {ek }k≥1 is Cauchy with respect to the absolute value || on RG. However
this is clear since ek − ej ∈ π j RG when k ≥ j. Let f be the limit of this sequence, since RG is
complete. Then f 2 = f by establishing this relation modulo π k RG for every k ∈ N. Moreover if e
was primitive then so is ek for all k ≥ 1, and in particular f is primitive.

Lemma 3.27. If 1 = e1 + . . . + en is an orthogonal decomposition of 1 in RG/πRG then we can

find an orthogonal decomposition 1 = f1 + . . . + fn in RG so that ei = fi + πRG for all i = 1 . . . n.

Proof. Apply the previous lemma to all idempotents to get f1 . . . fn ∈ RG lifting e1 . . . en respec-
tively. Then by Proposition 1.32 1 = f1 + . . . + fn is an orthogonal decomposition upon reduction
modulo RG/π k RG for all k ≥ 1 and so we get the desired result.

Proposition 3.28. Let ALbe an Artinian ring. Then there is a one-to-one bijection betweenPr direct
sum decompositions A ∼ r
= i=1 Ai as left A-modules and orthogonal decompositions 1 = i=1 ei of
1 into idempotents of A. Under this bijection, a decomposition of A into projective indecomposable
left submodules corresponds to an orthogonal decomposition of 1 into primitive idempotents.

Proof. Given an orthogonal decomposition 1 = ri=1 ei we have A ∼

P Lr
= i=1 Aei soP we match the
idempotent ei with ∼ r
Pr the left module Aei for all i = 1 . . . r. We clearly have A = i=1 Aei from
the identity 1 = i=1 ei . Now suppose a ∈ Aei ∩ Aej . Then we may write a = xei = yej and so
aei = xe2i = xei = yej ei = 0 so a = 0. From this we get the direct sum decomposition.

Let us start now with a direct sum decomposition A = ∼ Lr Ai , and write 1 = Pr ei for
i=1 i=1
i ∈ Ai . ByP
eP the nature of the left action of A on these submodules we have that 12 = 1 =
( ri=1 ei )2 = ri=1 e2i since ei ej = 0 for all distinct pairs i, j. From this we conclude that e2i = ei for
all i = 1 . . . r since we are dealing with a direct sum decomposition, and so e1 . . . er arePidempotents.
Now we show these processes are mutually inverse. Given a ∈ A we have a · 1 = ri=1 aei with
aei ∈ Ai = Aei as required. The last part of the proposition is left to the reader as an easy
exercise.

Definition 3.29. Let R be a ring. An RG-lattice is a finitely generated left RG-module L which is
a free R-module. Let (F, R, K) be a p-modular system for G and let L be a left F G-module. Then
L0 ⊂ L is a full RG-lattice if it contains an F -basis of L and if L0 is an RG-lattice.

Theorem 3.30. Let (F, R, K) be a p-modular system for G where F is complete with respect to
the discrete valuation v. Let S be a simple left KG-module with projective cover PS . Then there
exists a projective indecomposable RG-module P cS which is an RG-lattice and PS = P
cS /π P
cS .

Proof. Let e be a primitive idempotent of KG so that KGe = PS . We lift e to a primitive

idempotent f ∈ RG so that e = f + KG and P cS := RGf is a projective indecomposable left
RG-submodule of RG. Now PS = KGe = RGf /πRGf = P cS /π P
cS as required. Also P
cS is an
RG-lattice since RG is a free R-module, so any submodule is torsion-free. Now RGf ⊂ RG is a
free R-submodule by virtue of the structure theorem for finitely generated modules over PIDs since
R is a PID as it is a discrete valuation ring.

30
Definition 3.31. For a ring R and a finite group G, let S be a simple left RG-module and M a left
RG-module. Then we write [M, S] for the composition multiplicity of S appearing in M , whenever
M has a composition series.

Proposition 3.32. Let (F, R, K) be a p-modular system for G. Let U, V be left F G-modules with
full RG-lattices U0 , V0 respectively. Then we have the following.

1. HomRG (U0 , V0 ) is naturally a free R-submodule of HomF G (U, V ) which spans it over F .

2. πHomRG (U0 , V0 ) = HomRG (U0 , πV0 ).

3. If U0 is a projective left RG-module then HomRG (U0 ,V0 ) ∼ U0

= HomKG ( πU , V0 ).
πHomRG (U0 ,V0 ) 0 πV0

U0
4. If U0 is a projective left RG-module then dimF (HomF G (U, V )) = dimK (HomKG ( πU , V0 )).
0 πV0

Proof. 1. Let x1 . . . xn ∈ U0 and y1 . . . ym ∈ V0 be R-bases respectively, since these are free R-

modules. Then {x1 . . . xn } and {y1 . . . ym } extend to F -bases for U, V respectively since U0 , V0
are full RG-lattices. Then every element in HomRG (U0 , V0 ) can be described as as an m × n
matrix with entries in R under these bases, and every element in HomF G (U, V ) can be de-
scribed as an m×n matrix with entries in F . This identifies HomRG (U0 , V0 ) ⊂ HomF G (U, V )
as a left RG-submodule. It is sufficient and necessary for these matrices to commute with
the G-actions. Since R is a discrete valuation ring (hence a PID), HomRG (U0 , V0 ) is a free
R-module by the structure theorem and the fact that it is naturally a submodule of Rmn .
Let π = (t) be the maximal ideal of R so that F is the localization of R at t. Then clearly
every matrix in HomF G (U, V ) becomes a matrix in HomRG (U0 , V0 ) once multiplied by an
appropriate power of t, and so F HomRG (U0 , V0 ) = HomF G (U, V ).

2. As in the first part, πHomRG (U0 , V0 ) can be identified with m × n matrices with entries in pi
and commuting with all G-actions. This is exactly the description of HomRG (U0 , πV0 ) since
π = (t) for some generator t ∈ R so ty1 . . . tym is an R-basis for πV0 , and we can repeat the
analysis carried out in the first part.

3. So far we can conclude HomRG (U0 ,V0 ) ∼ HomRG (U0 ,V0 )

πHomRG (U0 ,V0 ) = HomRG (U0 ,πV0 ) . Let’s look again at both modules
HomRG (U0 ,V0 )
as m × n matrices with R-entries. By reduction modulo π, every element of HomRG (U0 ,πV0 )
U0
naturally becomes an element of HomKG ( πU , V0 ). More importantly, x1 +πU0 . . . xn +πU0 ∈
0 πV0
U0 /πU0 and y1 + πV0 . . . ym + πV0 ∈ V0 /πV0 are K-bases for U0 /πU0 and V0 /πV0 respectively,
U0
which both become free K-vector spaces. As a result we can use realize HomKG ( πU , V0 ) as
0 πV0
m×n matrices in these bases. The difficult part is lifting such a matrix to some representative
U0 V0
in HomRG (U0 , V0 ). Take a map f : πU 0
→ πV 0
of left KG-modules, which naturally becomes
U0
a map of left RG-modules. Now U0 is projective and there are surjections πU : U0 → πU 0
V0
and πV : V0 → πV 0
of left RG-modules. Therefore there exists a map e : U0 → V 0 so that
πV ◦ e = f ◦ πU . This lift e is unique since it is easily checked from our descriptions that the
HomRG (U0 ,V0 ) U0
map H : Hom RG (U0 ,πV0 )
→ HomKG ( πU , V0 ) is injective, and that H(e) = f is satisfied when
0 πV0
looking at appropriate matrices. This confirms the fact that H is an isomorphism.
U0
4. By part 3 we have dimK (HomKG ( πU , V0 )) = rankR (HomRG (U0 , V0 )). Then by part 1
0 πV0
we have that rankR (HomRG (U0 , V0 )) = dimF (HomF G (U, V )) since freeness as an R module
implies linear independence over F .

31
Lemma 3.33. Let KG be a group ring and S a simple left KG-module. Let M be an Artinian left
KG-module. Then [M/Rad(M ), S] = dimK (HomKG (M, S))

Proof. Let f : M → S be a map of left KG-modules. Then f must vanish on Rad(M ) since
Rad(M ) = Rad(KG)M and Rad(KG) annihilates simple left KG-modules. As a result we have
HomKG (M, S) ∼ = HomKG (M/Rad(M ), S) to begin with. L Now M/Rad(M ) is a semisimple left
KG-module so we can write M/Rad(M ) ∼ = S [M/Rad(M ),S] ⊕ r
i=1 Si where S1 . . . Sr are simple left
KG-modules not isomorphic L to S. Now Hom(·, S) is an additive functor so HomKG (M/Rad(M ), S) ∼ =
HomKG (S, S) [M/Rad(M ),S] ⊕ ri=1 HomKG (Si , S) and we are left to make some minor calculations.

Consider a map f : Si → S for i = 1 . . . r. This map must be zero, otherwise its image is
S since S is simple, and its kernel is 0 since Si is simple, and we get Si ∼
= S. From this case
we deduce HomKG (Si , S) = 0 for all i = 1 . . . r. Now we have dimK (HomKG (S, S)) = 1 since
HomKG (S, S) ∼ = EndKG (S) ∼= K by using Schur’s Lemma. Putting everything together gives us
dimK (HomKG (M, S)) = [M/Rad(M ), S].

Lemma 3.34. Let KG be a group ring and S a simple left KG-module. Let M be an Artinian left
KG-module. Then [M, S] = dimK (HomKG (PS , M )).

Proof. Let 0 = M0 ⊂ . . . ⊂ Mn = M be a composition series for M , and let us select elements

xk ∈ Mk \Mk−1 for all k = 1 . . . n so that xk generates Mk /Mk−1 over KG. We prove the lemma
by induction on the composition length n. The base case n = 0 is vacuously true, so suppose the
statement is true for composition lengths less than n, and that n ≥ 1. Let π : M → M/Mn−1 be the
quotient map. We start with the short exact sequence 0 → Mn−1 → M → M/Mn−1 → 0 and apply
the functor Hom(PS , ·), which is exact since PS is projective (simple homological algebra exercise
for the reader) and we have that 0 → Hom(PS , Mn−1 ) → Hom(PS , M ) → Hom(PS , M/Mn−1 ) → 0
is exact. From this we deduce that dimK (Hom(PS , M )) = dimK (Hom(PS , Mn−1 )) +
dimK (Hom(PS , M/Mn−1 )), and by induction we may assume dimK (Hom(PS , Mn−1 )) = [Mn−1 , S].

There are two cases to consider when it comes to a map f : PS → M/Mn−1 . Since Rad(PS ) =
Rad(KG)PS and M/Mn−1 is simple we must have Rad(PS ) ⊂ Ker(f ) and so we get
Hom(PS , M/Mn−1 ) ∼ = Hom(S, M/Mn−1 ). If M/Mn−1 ∼ 6 S Then [Mn−1 , S] = [M, S] and
=
Hom(S, M/Mn−1 ) = 0 as seen in Lemma 3.33. If M/Mn−1 ∼ = S Then [M, S] = [Mn−1 , S] + 1 and
dimK (Hom(S, M/Mn−1 )) = 1 as seen in Lemma 3.33. We can conclude by putting everything
together and using the induction hypothesis.

What follows is the most powerful result that brings all the theory together.

Theorem 3.35. Let (F, R, K) be a p-modular system for G with both F , R complete with respect
to the discrete valuation v. Let T be a simple left F G-module with full RG-lattice T0 , and let S be
a simple left KG-module. Then [F ⊗R P cS , T ] = [T0 /πT0 , S].

Proof. Observe that [F ⊗R P cS , T ] = dimF (HomF G (F ⊗R P

cS , T )) by Lemma 3.33 since Rad(F ⊗R
cS ) = 0 as we are in characteristic 0. Then dimF (HomF G (F ⊗R P
P cS , T )) = dimK (HomKG (PS , T0 /πT0 ))
by Proposition 3.32[4] since P
cS /π P ∼
cS = PS by Theorem 3.30 and P cS is a full RG-lattice of F ⊗R P
cS as
it takes the form PS = RGf for some primitive idempotent f . Finally dimK (HomKG (PS , T0 /πT0 )) =
c
[T0 /πT0 , S] by Lemma 3.34.

Theorem 3.36. Let G be a finite group and p a prime. Then Cp (G) = Dp (G)T Dp (G),

32
Proof. Let χ ∈ IRR(G) be an irreducible complex character and φ ∈ IBRp (G) a Brauer char-
acter. Select a p-modular system (F, R, K) for G with both F , R complete with respect to the
discrete valuation v. We can choose this in a way that a simple F G-module T exists with character
χ, and a simple KG-module S exists with Brauer character φ. One way to do this is to find a
number field L and a prime ideal p ⊂ OL so that OL,p realizes all complex irreducible representa-
tions, and OL,p /pOL,p realizes all irreducible p-modular representations. This can be done through
Corollary 3.1. Then the p-modular system (Lp , (OL )p , OL /pOL ) will work for us, where we denote
OL )p := limn (OL /pn OL ).

With these choices, we have [F ⊗R P cS , T ] = [T0 /πT0 , S] directly from Theorem 3.35. Now
T0 /πT0 is by definition the reduction of T modulo p and so by definition [T0 /πT0 , S] = dχ,φ . Let α
be the standard character of F ⊗R P cS . Then we have computed α = P [F ⊗R P cS , Ti ]χi = P χi dχ ,φ
i i i
where Ti are the simple irreducible F G-representations with corresponding standard characters χ i.
Also 0 ∼ P ◦
P α = P Pφ , the Brauer character of PS since PS /pPS = PS . Thus we have Pφ = i dχi ,φ χi =
c c
d
i χi ,φ j dχi ,φj φj by definition and the fact that reduction of standar characters is additive. Here
the φj are the irreducible Brauer characters of G at p, with corresponding P simplePleft KG-modules
given by Sj . Finally we conclude by noting [PSj , Si ] = Cp (G)i,j = l dχl ,φj k dχl ,φk [Sk , Si ] =
T
P
l dχl ,φj dχl ,φi = (Dp (G) Dp (G))i,j .

33