Introduction to Representation Theory

Konstantin Ardakov
October 2020

1. Representations of finite groups

Group representation theory is a marriage of Group Theory and Linear Algebra.

Its aim is to study all the ways in which a given group can arise as a group of
symmetries of some vector space.

Notation 1.1. Throughout this course, k will denote a field, and all vector spaces
are understood to be k-vector spaces. Recall that the general linear group of a
k-vector space V is the group GL(V ) of invertible k-linear transformations V → V
with the operation of composition.

Definition 1.2. Let G be a finite group and let V be a finite dimensional vector
space over k. A representation of G on V is a group homomorphism
ρ : G → GL(V ).
The degree of the representation is dim V .

If we need to be more precise about the ground field of scalars, we will say “ρ is
a k-representation of G”. Here are some examples from Geometry.

Example 1.3.
(a) The cyclic group G = hgi of order 2 acts on V = k by negation: ρ(g) = −1
gives a representation of G of degree 1.
(b) Let G be the symmetry group of a triangle: G = D6 = {e, r, r2 , s, sr, sr2 } and
let k = R. Then G acts by R-linear transformations on the plane V = R2 . This
gives rise to a representation
ρ : G → GL(V )
where ρ(r) is the rotation through 2π/3 and ρ(s) is the reflection in the y-axis.
(c) The symmetry group of the regular n-gon G = D2n acts on V = R2 by R-linear
transformations, giving a natural representation of G of degree 2.
(d) Let k = R, let X ⊂ R3 be the set of vertices of a cube centred at the origin,
and let G be the stabiliser of X in the rotation group SO3 (R). Then from your
Prelims M1 course you know that G is isomorphic to the symmetric group S4 .
In this way, we obtain a degree 3 representation
ρ : S4 → GL(R3 )
of this symmetric group.

One may view representations as a “linearisation” of the notion of a group action

on a set. This gives a large class of examples of representations, as follows.
1
2

Definition 1.4. Let X be a finite set. The free vector space on X, is the set
( )
X
kX := ax x : ax ∈ X
x∈X

of formal linear combinations of members of X with coefficients ax lying in k. The

addition and scalar multiplication operations on kX are the evident ones.

Note that X is naturally a basis for kX.

Definition 1.5. Let X be a finite set equipped with a left action of the finite group
G. Each g ∈ G defines a permutation ρ(g) : X → X, given by ρ(g)(x) = g · x. This
permutation extends uniquely to an invertible linear map ρ(g) : kX → kX:
!
X X
ρ(g) ax x = ax g · x.
x∈X x∈X

Since g · (h · x) = (gh) · x for all g, h ∈ G and all x ∈ X one checks easily that
ρ(g)ρ(h) = ρ(gh) in GL(kX) for all g, h ∈ G. Thus ρ : G → GL(kX) is a represen-
tation, called the permutation representation (associated with X).

Here is an example from Galois Theory.

Example 1.6. Let k = Q be the field of rational numbers and let F be a finite field
extension of k (so that dimk F < ∞). Let G be the group of all field automorphisms
of F . Then every g ∈ G is a k-linear map g : F → F , and one of the main results
of Galois Theory tells us that G is a finite group. Then the inclusion G ,→ GL(F )
gives a k-linear representation of G of degree dimk F .
√ 2πi
For example, F could be the splitting field Q( 2, e 3 ) of the polynomial x3 − 2.
Then G is the symmetric group S3 and this gives a Q-linear representation ρ : S3 →
GL(F ) of degree 6.

The study of representations of Galois groups form a major part of modern

Algebraic Number Theory, for example through the Langlands Programme.

Definition 1.7. The representation ρ : G → GL(V ) is faithful if ker ρ = {1}.

One reason group representations are interesting to a group-theorist is the fol-

lowing. Given a representation ρ : G → GL(V ), it is either faithful, or not. If it
is faithful, then G is isomorphic to its image in GL(V ) which, using matrix rep-
resentations below, is easy to compute with. Otherwise ker ρ is a proper normal
subgroup of G and G/ ker ρ is isomorphic to ρ(G).

Definition 1.8. Let G be a finite group. A matrix representation is a group

homomorphism ρ : G → GLn (k), where GLn (k) = Mn (k)× denotes the group of
invertible matrices under matrix multiplication.

There is a close connection between representations and matrix representations.

Recall the following from Part A Linear Algebra.
 3

Definition 1.9. Let B := {v1 , . . . , vn } be a basis for V and let φ : V → V be a

linear map. The matrix of φ with respect to B is B [φ]B = (aij )ni,j=1 where
n
X
φ(vj ) = aij vi for all j = 1, . . . , n.
i=1

Remark 1.10. Let V be a vector space V with basis B.

∼
=
(a) The map φ 7→ B [φ]B gives an isomorphism of groups GL(V ) −→ GLn (k).
(b) Every representation ρ : G → GL(V ) gives rise to a matrix representation
ρB : G → GLn (k) given by
ρB (g) := B [ρ(g)]B for all g ∈ G.
(c) Conversely, every matrix representation σ : G → GLn (k) defines a repre-
sentation σ : G → GL(k n ) on the space k n of column vectors, by letting
σ(g) : k n → k n be the k-linear map given by
σ(g)(v) = σ(g)v for all g ∈ G, v ∈ k n
where σ(g)v means the multiplication of the n × n matrix σ(g) by the n × 1
column vector v. We will sometimes abuse notation and simply wrote σ for σ.

Example 1.11. Let G = S3 act on X = {e1 , e2 , e3 } by permutations of indices.

We obtain a degree 3 permutation representation ρ : G → GL(kX). Since
(123) · e1 = e2 , (123) · e2 = e3 , (123) · e3 = e1
we see that  
0 0 1
ρX ((123)) = 1 0 0 .
 

0 1 0
Similarly, one can calculate
 
0 1 0
ρX ((12)) = 1 0 0 .
 

0 0 1
Warning : in this course, permutations in Sn are multiplied from right to left,
in contrast to the conventions in Prelims M1. Thus, for example, we have in S3
(123) · (12) = (13).
This way of writing and multiplying permutations is necessary to ensure that the
natural action of Sn on {1, · · · , n} given by σ · n = σ(n) is a left action: it satisfies
the axiom g · (h · x) = (gh) · x as opposed to a right action (x · g) · h = x · (gh).

Definition 1.12. Let ρ : G → GL(V ) and σ : G → GL(W ) be two representations.

A homomorphism, also known as an intertwining operator, is a linear map
ϕ:V →W
4

such that
σ(g) ◦ ϕ = ϕ ◦ ρ(g) for all g ∈ G.
We say that ϕ is an isomorphism if it is bijective.

Definition 1.13. Two matrix representations ρ1 : G → GLn (k) and ρ2 : G →

GLn (k) are said to be equivalent if there exists A ∈ GLn (k) such that
ρ2 (g) = A ρ1 (g) A−1 for all g ∈ G.

If ρ1 and ρ2 are two equivalent matrix representations, then the equality of

products of matrices ρ2 (g)A = Aρ1 (g) translates to an equality of linear maps
ρ2 (g) ◦ A = A ◦ ρ1 (g)
in GL(k n ), which means that the representations ρ1 and ρ2 of G from Remark
1.10(b) are isomorphic. The converse is also true.

Definition 1.14. Let ρ : G → GL(V ) be a representation, and let U be a linear

subspace of V .
(a) We say that U is G-stable if ρ(g)(u) ∈ U for all g ∈ G and u ∈ U .
(b) Suppose that U is G-stable. The subrepresentation of ρ afforded by U is
ρU : G → GL(U )
given by ρU (g)(u) := ρ(g)(u) for all g ∈ G and w ∈ U .
(c) Suppose that U is G-stable. The quotient representation of ρ afforded by U is
ρV /U : G → GL(V /U )
given by ρV /U (g)(v + U ) := ρ(g)(v) + U for all g ∈ G and v + U ∈ V /U .

The following Lemma is easy to verify directly.

Lemma 1.15. Let ϕ : V → W be a homomorphism between representations

ρ : G → GL(V ) and σ : G → GL(W ).
(a) ker ϕ is a G-stable subspace of V .
(b) Im ϕ is a G-stable subspace of W .
(c) There is a natural isomorphism
∼
=
V / ker ϕ −→ Im ϕ
between the G-representations ρV / ker ϕ and σIm ϕ .

Lemma 1.15(c) is called the First Isomorphism Theorem for representations.

There are also Second and Third Isomorphism Theorems – see Remark 2.6 below.

Definition 1.16. Let G be a group. The trivial representation of G on a vector

space V is 1 : G → GL(V ), given by
1(g)(v) = v for all g ∈ G, v ∈ V.
 5

Example 1.17. Suppose that k is the finite field Fp and let G = hgi be the cyclic
group of order p. Let ρ : G → GL2 (k) be the matrix representation given by
!
i 1 i
ρ(g ) = for each i ∈ Z.
0 1
( ! !)
1 0
Let v1 = , v2 = be the standard basis for V = k 2 . Then U := hv1 i
0 1
is a G-stable subspace, because ρ(g i )(v1 ) = v1 for all i. The subrepresentation
ρU : G → GL(U ) and quotient representation ρV /U : G → GL(V /U ) in this case
are both trivial. However of course ρ itself is not trivial!

Definition 1.18. The representation ρ : G → GL(V ) is irreducible or simple if

• V is not the zero vector space, and
• if U is a G-stable subspace of V , then either U = {0} or U = V .

The irreducible representations are the atoms of representation theory. The

major goal of Representation Theory is to find all irreducible representations of a
given group, up to isomorphism.

Definition 1.19. Let ρ : G → GL(V ) be a representation and let U be a G-stable

subspace. A G-stable complement for U in G is a G-stable subspace W such that
V = U ⊕ W.

Recall from Linear Algebra that this means that U + W = V and U ∩ W = {0}.

Example 1.20. Consider the permutation representation of G = S3 afforded by

kX, where X = {e1 , e2 , e3 } as in Example 1.11. Then
U := he1 + e2 + e3 i
is a G-stable subspace, with G fixing every vector in U . So, U is a trivial subrep-
resentation of V . Now let
W := {a1 e1 + a2 e2 + a3 e3 : a1 + a2 + a3 = 0};
this is a G-stable complement to U in V — provided char(k) 6= 3. Let B = {v1 , v2 }
where v1 := e1 − e2 and v2 := e2 − e3 so that B is a basis for W . Then the degree 2
matrix representation σ := (ρW )B : G → GL2 (k) afforded by W is determined by
! !
0 −1 −1 1
σ((123)) = and σ((12)) = .
1 −1 0 1

We now come to the first non-trivial result of Representation Theory.

Theorem 1.21 (Maschke). Let G be a finite group and suppose that |G| = 6 0 in
k. Let U be a G-stable subspace of a finite-dimensional G-representation V . Then
U admits at least one G-stable complement W in V .
6

Proof. By picking a basis for U and extending it to a basis for V we can find some
linear complement Z for U in V :

V = U ⊕ Z.

This Z will certainly not be G-stable in general. We will now use an averaging
argument to replace Z with a G-stable one. To this end, let π : V → V be the
projection map along the decomposition V = U ⊕ Z, so that

π(u + z) = u for all u ∈ U, z ∈ Z.

Define a new linear map ϕ : V → V by the rule

1 X
ϕ(v) := ρ(x)π(ρ(x)−1 (v)) for all v ∈V;
|G|
x∈G

here, we use the hypothesis that the number |G| is invertible in the field k. We
now check directly that ϕ is a homomorphism of representations. To do this, it is
helpful to use the following notation:

g · v := ρ(g)(v) for all g ∈ G, v ∈ V.

Observe that since ρ is a group homomorphism, this defines a group action of G on

the vector space V as in Prelims M1. Fix g ∈ G and v ∈ V . Then
X
|G|ϕ(g · v) = x · π(x−1 · (g · v)).
x∈G

In this finite sum, make the substitution y −1 = x−1 g. As x runs over all elements
of G, so does y. Since x is then equal to gy, we obtain
X X
|G|ϕ(g · v) = (gy) · π(y −1 · v) = g · y · π(y −1 · v) = g · |G|ϕ(v).
y∈G y∈G

Cancelling |G| we deduce that ϕ is a homomorphism of representations as claimed.

On the other hand, if u ∈ U then
1 X 1 X
ϕ(u) = x · π(x−1 · u) = x · (x−1 · u) = u
|G| |G|
x∈G x∈G

because U is G-stable, so π(x−1 · u) = x−1 u for all x ∈ G. So the restriction of ϕ to

U is the identity map. On the other hand, since U is G-stable and since π(V ) = U ,
the definition of ϕ shows that ϕ(V ) ⊆ U . As ϕ(U ) = U we actually have ϕ(V ) = U ,
so Im(ϕ) = U . Let W := ker ϕ, a G-stable subspace of V by Lemma 1.15(a). Then
dim W + dim U = dim V by the Rank-Nullity Theorem, whereas if v ∈ W ∩ U then
0 = ϕ(v) = v. So, V = U ⊕ W and W is a G-stable complement to U in V . 

Remark 1.22. Maschke’s Theorem fails if the characteristic of our ground field k
happens to divide |G|, see Example 1.17.
 7

Definition 1.23. Let ρ : G → GL(V ) be a representation. We say that ρ is

completely reducible if there exist G-stable subspaces U1 , . . . , Um of V such that
V = U1 ⊕ · · · ⊕ Um
and the subrepresentation of G afforded by each Ui is irreducible, or if V = {0}.

Corollary 1.24. Let G be a finite group and suppose that char(k) - |G|. Then
every finite dimensional representation ρ : G → GL(V ) of G is completely reducible.

Proof. Proceed by induction on dim V , the case dim V = 0 being true by defini-
tion. Let U1 be a G-stable non-zero subspace of V of smallest possible dimension.
Clearly U1 is irreducible. Then U1 admits a G-stable complement W by Maschke’s
Theorem, Theorem 1.21. Now dim W < dim V so by induction, W = U2 ⊕ · · · ⊕ Um
for some G-stable irreducible subspaces U2 , · · · , Um . Hence V = U1 ⊕ U2 ⊕ · · · ⊕ Um
is also completely reducible. 

2. The group ring

We observed in the proof of Theorem 1.21 that every representation ρ : G →

GL(V ) determines a left action of the group G on V via
g · v := ρ(g)(v) for all g ∈ G, v ∈ V.
We already noticed that sometimes when working a representation ρ it is cumber-
some to keep writing ρ(g)(v) when really one can get away with just g · v. On the
other hand, since V is a k-vector space it also carries an action of the ground field
of scalars k, (λ, v) 7→ λv. This may make one suspect that in fact there is a module
hiding in the background, and this is in fact the case. Recall the following from
Part A Rings and Modules.

Definition 2.1. A ring is an abelian group R equipped with an associative multi-

plication R × R → R written (a, b) 7→ a · b, which satisfies the distributive laws
a · (b + c) = a · b + a · c and (a + b) · c = a · c + b · c for all a, b, c ∈ R
and which admits an identity element 1 ∈ R such that
1·a=a=a·1 for all a ∈ R.

In algebraic geometry and algebraic number theory, most rings one encounters
are commutative. However in this course rings are most definitely not commutative
in general!

Definition 2.2. Let G be a finite group. The group ring of G (with coefficients in
k) is the vector space kG from Definition 1.5, with the following multiplication:
!   !
X X X X
ax x ·  by y  = ax bx−1 g g.
x∈G y∈G g∈G x∈G
8

It is easy to check that with this multiplication kG is an associative ring with

identity element equal to 1 (the identity element of the group, thought of as a formal
linear combination with precisely one non-zero coefficient). The group G becomes
embedded into the group ring via g 7→ g: this embedding respects multiplication
and thereby realises G as a subgroup of the group of units kG× of the ring kG. We
will identify G with its image in kG× . In this way, one should think of the ring kG
as being an “envelope” of the group G.

Example 2.3. Let G = hxi be a cyclic group of order n. Then kG has G =

{1, x, · · · , xn−1 } as a basis, so it is generated by k and x as as a ring, and k commutes
with x. Define a ring homomorphism ϕ : k[T ] → kG by setting ϕ(f (T )) = f (x) for
each f (T ) ∈ k[T ]. Then ϕ is surjective, and ker ϕ = hT n − 1i. Hence by the First
Isomorphism Theorem for Rings,
kG ∼
= k[T ]/hT n − 1i.

If in addition the ground field k contains a primitive n-th root of unity ζ, then
the polynomial T n − 1 factors into a product of distinct linear factors (T − 1)(T −
ζ)(T − ζ 2 ) · · · (T − ζ n−1 ). Then the Chinese Remainder Theorem implies that
kG ∼
= k[T ]/hT n − 1i ∼
=k × k × ··· × k.
| {z }
n times
Next, recall the following important definition from Part A Rings and Modules.

Definition 2.4. Let R be a ring. An R-module is an abelian group M equipped

with a left R-action R × M → M which satisfies the axioms
r · (m + n) = (r · m) + (r · n)
(r · s) · m = r · (s · m)
1·m = m
for all r, s ∈ R, m, n ∈ M .

Strictly speaking, what we have defined is at Definition 2.4 is usually called a

left module. But we will not consider right modules in this course, so all modules
we will be concerned with are left modules.

Proposition 2.5. Let V be a vector space and let G be a group.

(a) Suppose that ρ : G → GL(V ) is a representation. Then V becomes a left
kG-module via the action
!
X X
ax x · v = ax ρ(x)(v) for all ax ∈ k, v ∈ V.
x∈G x∈G

(b) Suppose that V is a left kG-module. Then ρ : G → GL(V ), defined by

ρ(g)(v) := g · v for all g ∈ G, v ∈ V
is a representation of G.
 9

(c) These recipes set up a bijection between the set of representations ρ : G →

GL(V ) and the set of kG-module structures kG × V → V on V .

Important Convention: From now on, we will use the bijective correspon-
dence provided by Proposition 2.5 to freely convert a representation ρ : G → GL(V )
into the corresponding kG-module V , and vice versa. We will also sometimes say
“V is a representation of G” when we actually mean “V is a kG-module”.

Remark 2.6. The correspondence between G-representations and kG-modules al-

lows us to apply all general theorems about modules from Part A Rings and Mod-
ules to the study of G-representations. For example, the First, Second and Third
Isomorphism Theorems, and the Correspondence Theorem hold automatically for
G-representations.

Example 2.7. Let A be a ring. The free A-module of rank 1 is the abelian group
A equipped with the left-multiplication action of A:

a · b = ab for all a, b ∈ A.

A-submodules of this A-module are called left ideals.

Definition 2.8. If A = kG, the representation ρ : G → GL(kG) corresponding to

the free kG-module of rank 1 is called the left regular representation.

The left regular representation coincides with the permutation representation of

G on kG from Definition 1.5 associated with the action of G on X = G by left
multiplication.
All of the concepts introduced in §1 above for representations have immediate
generalisations in the language of modules. For example, an A-module M is said
to be irreducible, or simple, if M is non-zero, and if whenever N is an A-submodule
of M we must have N = {0} or N = M itself. A homomorphism of representations
is simply a map of kG-modules, also known as a kG-linear map.
Generalising Definition 1.23, we say that an A-module V is completely reducible
if it is either the zero module, or it is equal to a direct sum of finitely many simple
submodules.

Definition 2.9. Let A be a ring. We say that A is semisimple if the free A-module
of rank 1 is completely reducible.

Maschke’s Theorem, Theorem 1.21 gives us the first important example of a

semisimple ring.

Example 2.10. Let G be a finite group such that |G| 6= 0 in k. Then the group
ring kG is semisimple.
10

Definition 2.11. Let V be an A-module. We say that V is cyclic if it can be

generated by a single element v: V = A · v. The annihilator of v ∈ V is the left
ideal
annA (v) := {a ∈ A : av = 0}.

Clearly every simple module is also cyclic.

Lemma 2.12. Every cyclic A-module V is isomorphic to a quotient module of the

free A-module of rank 1: if V = A · v then
V ∼
= A/ annA (v).

Proof. The map ϕ : A → V given by a 7→ a · v is an A-module homomorphism.

It is surjective by hypothesis. So by the First Isomorphism Theorem for modules,
V = Im ϕ ∼ = A/ ker ϕ. However ker ϕ = annA (v). 

The following trivial result has profound consequences.

Lemma 2.13. Let V, W be simple A-modules. Then every non-zero A-linear map
ϕ : V → W is an isomorphism.

Proof. We know that ker ϕ is an A-submodule of V and that Im ϕ is an A-submodule

of W by the module-theoretic version of Lemma 1.15. Since ϕ is non-zero, ker ϕ is
not all of V and Im ϕ is not the zero module. Since V and W are both simple, we
conclude that ker ϕ is zero and Im ϕ = W . Hence ϕ is bijective, and therefore an
isomorphism. 

Proposition 2.14. Let A be a semisimple ring. Then A has only finitely many
simple A-modules, up to isomorphism.

Proof. Write A = V1 ⊕ · · · ⊕ Vr for some simple A-submodules Vi of A. Let V be

a simple A-module, pick a non-zero vector v ∈ V and consider the A-module map
ϕ : A → V from Lemma 2.12. Let ϕi : Vi → V be the restriction of ϕ to Vi , so that
if a = a1 + · · · + am is the decomposition of a ∈ A with ai ∈ Vi for each i, then
ϕ(a) = ϕ1 (a1 ) + ϕ2 (a2 ) + · · · + ϕr (ar ).
If ϕi is the zero map for all i, then it follows that ϕ is also the zero map. So we
see that ϕi 6= 0 for some index i. But now Lemma 2.13 implies that this ϕi must
be an isomorphism. So, V is isomorphic to one of the irreducible representations
in the list V1 , V2 , · · · , Vr . 

Here is a striking application of our new viewpoint on representations.

Theorem 2.15. Let G be a finite group such that |G| 6= 0 in k. Then G has only
finitely many irreducible representations, up to isomorphism.

Proof. The ring kG is semisimple by Maschke’s Theorem, Theorem 1.21. Now

apply Proposition 2.14. 
 11

Definition 2.16. For a finite group G, we will write rk (G) to denote the number
of isomorphism classes of irreducible k-representations of G.

A big motivating question for us will be: how do we compute rC (G) effectively?

3. Structure of semisimple algebras

We begin with some very general ring theory.

Definition 3.1. The centre of the ring A is

Z(A) := {z ∈ A : az = za for all a ∈ A}.

The centre is always a commutative unital subring of A.

Definition 3.2. Let A be a ring and let V be an A-module. The endomorphism ring
of V , EndA (V ), is the set of all A-module homomorphisms ϕ : V → V , equipped
with pointwise addition of homomorphisms, and composition as multiplication.

Note that whenever V is an A-module, it also becomes an EndA (V )-module

via f · v := f (v) for all f ∈ EndA (V ) and v ∈ V . The two actions (of A and
EndA (V )) on V commute pointwise, by definition. This observation shows that
the action of any central element z ∈ Z(A) on V is necessarily by an A-module
endomorphism.

Definition 3.3. Let A be a ring. The opposite ring to A, Aop , has A as the
underlying abelian group, but with the following new multiplication:

a ? b := ba for all a, b ∈ Aop .

Lemma 3.4. Let A be a ring.

(a) For each a ∈ A, right-multiplication by a defines an A-module endomorphism
ra : A → A of the free A-module of rank 1, given by b 7→ ra (b) := ba.
(b) Every A-module endomorphism ω : A → A is of this form.
(c) The map Aop → EndA (A) given by a 7→ ra is an isomorphism of rings.

Proof. Problem Sheet 2. 

Definition 3.5. We say that A is a k-algebra if it contains k as a central subfield.

If, in addition, A is a semisimple ring, we say that A is a semisimple k-algebra. A
homomorphism of k-algebras is a k-linear ring homomorphism.

Theorem 3.6 (Schur’s Lemma). Suppose that k is algebraically closed. Let V

be a simple module over a finite dimensional k-algebra A. Then every A-module
endomorphism of V is given by the action of some scalar λ ∈ k, so that

EndA (V ) = k1V .
12

Proof. By Lemma 2.12, V is isomorphic to a quotient module of A, so V is itself

finite dimensional as a k-vector space. Let ϕ : V → V be an A-module endomor-
phism; then in particular it is a k-linear map so it has at least one eigenvalue λ ∈ k
since k is algebraically closed. This means that ϕ − λ1V : V → V has a non-zero
kernel, and is therefore not an isomorphism. So it must be the zero map, by Lemma
2.13. But then ϕ = λ1V is the action of λ ∈ k, as required. 

Definition 3.7. Let A be a k-algebra and let V be an A-module with EndA (V ) = k1V .
Then by Theorem 3.6, every element z ∈ Z(A) must act on V by a scalar, which
we denote by zV . The ring homomorphism
Z(A) → k, z 7→ zV
is called the central character of V .

Using these tools, we will now focus on semisimple rings. Recall from Proposition
2.14 that such a ring admits only finitely many simple modules up to isomorphism.

Notation 3.8. Until the end of §3, A will denote a fixed semisimple ring, and
V1 , · · · , Vr will denote a complete list of representatives for the isomorphism classes
of simple A-modules. We also fix a decomposition
ni
r M
M
(3.1) A= Li,j
i=1 j=1

of the A-module A into a direct sum of simple left ideals Li,j , where Li,j ∼
= Vi for
each i and j.

We must have n1 , · · · , nr ≥ 1. This is because each Vi occurs as a direct sum-

mand of A at least once: this follows from Proposition 2.14. Warning: the left
ideals Li,j are not unique, in general!

Proposition 3.9. Let A be a finite dimensional semisimple k-algebra and suppose

that k is algebraically closed. Then dim Z(A) 6 r.

Proof. By Schur’s Lemma, Theorem 3.6, we have EndA (Vi ) = k1Vi for all i =
1, · · · , r. So we can define a k-linear map ψ : Z(A) → k r by ψ(z) := (zV1 , zV2 , · · · , zVr ).
Suppose that ψ(z) = 0 for some z ∈ Z(A); then zVi = 0 for all i; we will show that
z = 0. Consider the decomposition of 1 ∈ A along the decomposition (3.1):
ni
r X
X
1= ei,j for some ei,j ∈ Li,j .
i=1 j=1

P ni
r P ni
r P
P
Then z = z1 = zei,j = zVi ei,j . But zVi = 0 for all i, so z = 0. So, ψ
i=1 j=1 i=1 j=1
r
is injective and dim Z(A) 6 dim k = r. 
ni
L
Definition 3.10. For each i = 1, . . . , r, define Bi := Li,j .
j=1
 13

Lemma 3.11. Each Bi is a two-sided ideal of A, and A = B1 ⊕ · · · ⊕ Br .

Proof. It follows from (3.1) that A = B1 ⊕ · · · ⊕ Br and that each Bi is a left ideal
of A; it will therefore be enough to show that each Bi is also a right ideal in A.
Fix a ∈ A and consider Li,j ⊆ Bi . Let i0 6= i and 1 6 j 0 6 ni0 be another
pair of indices, and consider the projection ϕ : A  Li0 ,j 0 along our decomposition
(3.1). The restriction of ϕ ◦ ra : A → Li0 ,j 0 to Li,j is an A-module homomorphism
from Li,j to Li0 ,j 0 . Because i0 6= i, these two modules are not isomorphic, so this
restriction must be the zero map by Lemma 2.13. Varying i0 and j 0 , we see that the
projection of Li,j a onto each Bi0 with i0 6= i is equal to zero. Therefore Li,j a ⊆ Bi .
But since Bi is equal to the sum of all of the Li,j , Bi a ⊆ Bi for all a ∈ A. 

Lemma 3.12. Let R be a k-algebra and suppose that R = S1 ⊕ · · · ⊕ Sr for some

non-zero two-sided ideals S1 , · · · , Sr . Then dim Z(R) ≥ r.

Proof. Write 1 = e1 + · · · + er for some ei ∈ Si . Let a ∈ R and fix i = 1, . . . , r.

Since Si is a left ideal, aei ∈ Si . Since a = ae1 + · · · + aer , we see that aei is the
component of a in Si in the decomposition R = S1 ⊕ · · · ⊕ Sr . Similarly since Si is a
right ideal, ei a is the component of a in Si in this decomposition. Hence aei = ei a
for all i and all a ∈ R, which shows that each ei is central.
r
ej = e2i .
P
If i 6= j then ei ej ∈ Si ∩ Sj = {0} so ei ej = 0. Hence ei = ei · 1 = ei
j=1
We have shown that {e1 , · · · , er } forms a set of pairwise orthogonal idempotents:

(3.2) ei ej = δi,j ei for all i, j = 1, · · · , r.

r
P
Now suppose that λi ei = 0 for some λi ∈ k. Multiply this equation by ej and
i=1
use (3.2) to get λj ej = 0. If ej = 0 then for all a ∈ Sj we have a = aej = 0,
contradicting the assumption that Sj 6= {0}. So ej 6= 0 for all j = 1, . . . , r, this
shows that {e1 , · · · , er } is linearly independent over k. Hence r 6 dim Z(R). 

Theorem 3.13. Let A be a finite dimensional semisimple k-algebra and suppose

that k is algebraically closed. Then r = dim Z(A).

Proof. By Proposition 3.9 we have r ≥ dim Z(A). By Lemma 3.11 we have A =

B1 ⊕ · · · ⊕ Br for some two-sided ideals Br , so Proposition 3.12 gives the reverse
inequality r 6 dim Z(A). 

We have now related the number of isomorphism classes of simple modules over
our semisimple k-algebra to the centre of A, which motivates the question of cal-
culating dim Z(kG). Computing the centre of group rings is very easy.

Definition 3.14. For a finite group G, let s(G) denote the number of conjugacy
classes of G.
14

Proposition 3.15. Let G be a finite group and let C1 , · · · , Cs be the conjugacy

classes of G. For each i = 1, . . . , s, define the conjugacy class sum of Ci to be
X
C
ci := x ∈ kG.
x∈Ci

Then {C
c1 , · · · , C
cs } is a basis for Z(kG) as a vector space, so

dim Z(kG) = s(G).

Proof. Problem Sheet 1. 

Corollary 3.16. Let G be a finite group, and let k be an algebraically closed field
with |G| =
6 0 in k. Then rk (G) = s(G).

Proof. By Theorem 1.21, kG is a semisimple k-algebra with dim Z(kG) = s(G) by

Proposition 3.15. Now apply Theorem 3.13. 

We were able to obtain this striking result without really understanding the
internal structure of the ring kG properly. With a little additional effort, this
structure will become completely transparent at least in the case where the ground
field k is algebraically closed and where Maschke’s Theorem applies.
We continue with the notation established at (3.8) above.

Lemma 3.17.
(a) Each Bi is a ring with identity element ei .
(b) A is isomorphic to the product of the rings (Bi , ei ):
A∼
= B1 × · · · × Br .
(c) Each Bi is itself a semisimple ring, with unique simple module Vi .

Proof. (a) Lemma 3.11 implies that Bi is an additive subring of A, stable under
multiplication. We saw in the proof of Lemma 3.12 that for any a ∈ A, aei = ei a is
the Bi -component of a along the decomposition A = B1 ⊕· · ·⊕Br . So aei = ei a = a
for all a ∈ Bi which says that ei is an identity element in Bi .
(b) The isomorphism sends a ∈ A to (ae1 , · · · , aer ) ∈ B1 × · · · × Br .
(c) Fix ` = 1, . . . , ni and suppose that U is a Bi -submodule of Li,` . Then
 
Mr
A·U = Bj  · U 6 U
j=1

because by (3.2), Bj · U 6 Bj · Bi = Bj ej · ei Bi = 0 if j 6= i, and Bi · U 6 U as U is

a Bi -submodule. Hence U is also an A-submodule of Li,` , so U is either zero or all
of Li,` because Li,` is a simple A-module. We’ve shown that Li,` , and hence also
Vi , are simple Bi -modules.
ni
L
Since Bi = Li,j , it is a semisimple ring. The proof of Proposition 2.14 now
j=1
shows that Vi is the only simple Bi -module, up to isomorphism. 
 15

Warning: even though each Bi is an additive subgroup of A stable under

multiplication, it is not a unital subring when r ≥ 2, because its identity element
ei is then not equal to the identity element 1 in A.

Proposition 3.18. Let B be a semisimple ring with exactly one simple module
n times
z }| {
∼
V , up to isomorphism. Suppose that B = V ⊕ · · · ⊕ V as a left B-module, and let
D := EndB (V ). Then there is a ring isomorphism
B∼
= Mn (Dop ).

Proof. By Lemma 3.4(c), we know that B is isomorphic to EndB (B)op . Since B =

V n as a left B-module, we have EndB (B) ∼
= Mn (D). Hence, B ∼ = EndB (B)op ∼
=
Mn (D)op ∼= Mn (Dop ). See By Problem Sheet 2 for more details. 

Theorem 3.19 (Artin-Weddernburn). Suppose that k is an algebraically closed

field and that A is a finite dimensional semisimple k-algebra. Then there exist
positive integers n1 , n2 , · · · , nr and a k-algebra isomorphism
∼
=
A −→ Mn1 (k) × · · · × Mnr (k).

Proof. By Lemma 3.17, we may assume that r = 1, so that A has exactly one
simple module V up to isomorphism. Then A ∼
= Mn (Dop ) where D := EndA (V ),
by Lemma 3.18. But D ∼= k by Schur’s Lemma, Theorem 3.6. 

Corollary 3.20. Suppose that k is algebraically closed. Let G be a finite group

such that |G| =6 0 in k and let V1 , · · · , Vr be a complete list of pairwise non-
isomorphic simple kG-modules.
(a) kG ∼= V1dim V1 ⊕ V2dim V2 ⊕ · · · ⊕ Vrdim Vr as a kG-module.
r
(dim Vi )2 .
P
(b) |G| =
i=1

Proof. (a) kG is a semisimple ring by Maschke’s Theorem, Theorem 1.21, so kG ∼=

Mn1 (k) × · · · × Mnr (k) by the Artin-Wedderburn Theorem, Theorem 3.19. The
matrix algebra Mn (k) acts on the space of column vectors k n naturally by left
multiplication, k n is then a simple Mn (k)-module and Mn (k) is isomorphic to a
direct sum of precisely n copies of this simple module by Problem Sheet 1. So,
ni = dim Vi for each i = 1, . . . , r.
(b) This is immediate from part (a). 
16

4. Multilinear algebra

The regular representation of G afforded by kG gives a good way to theoretically

bound the number of possible irreducible representations. But it does not really
help us to find them. We harness the full power of linear algebra here to find many
new representations from old. Throughout §4, G will denote a fixed finite group.
The following observation will be useful for constructing G-representations.

Lemma 4.1. Let V be a vector space and let G × V → V be a G-action on the set
V . Then this extends to a kG-module structure on V if and only if the G-action
on V is linear, which means that the following condition holds:
g · (v + λw) = (g · v) + λ(g · w) for all g ∈ G, v, w ∈ V, λ ∈ k.

Definition 4.2. Let V, W be G-representations. The (external) direct sum is the

vector space
V ⊕ W := V × W
which is again a G-representation via
g · (v, w) = (g · v, g · w) for all g ∈ G, v ∈ V, w ∈ W.

We will usually identify V and W with their images {(v, 0) : v ∈ V } and {(0, w) :
w ∈ W } inside V × W , respectively. The sum of these images is all of V × W and
their intersection is {(0, 0)}, so this notation V ⊕ W is consistent with the one
you’ve met in Linear Algebra.

Definition 4.3. Let V be a G-representation. The dual representation is the space

V ∗ of all linear functions f : V → k on V . The group G acts on V ∗ via
(g · f )(v) := f (g −1 · v) for all g ∈ G, f ∈ V ∗ , v ∈ V.

This action is linear so we get a new G-representation V ∗ of the same dimension

as V . Note the inverse appearing in this definition! Without it, the axiom g·(h·f ) =
(gh) · f for a G-action does not hold.

Definition 4.4. Let V, be G-representations. The vector space Hom(V, W ) of all

linear maps from V to W admits a linear G-action given by
(g · f )(v) = g · f (g −1 · v) for all g ∈ G, f ∈ Hom(V, W ), v ∈ V.

In the case where W is the trivial 1-dimensional representation, we recover the

dual space Hom(V, k) = V ∗ , so Definition 4.4 is a generalisation of Definition 4.3.

Lemma 4.5. Let V be a finite dimensional G-representation. The natural biduality

isomorphism from Part A Linear Algebra
τ : V → (V ∗ )∗ , τ (v)(f ) := f (v) for all f ∈ V ∗, v ∈ V
is an isomorphism of G-representations.
 17

Definition 4.6. Let V and W be two vector spaces, let {v1 , · · · , vm } be a basis
for V and let {w1 , · · · , wn } be a basis for W . The tensor product of V and W ,

V ⊗ W,

is the free vector space (see Definition 1.4) on the set of formal symbols

{vi ⊗ wj : 1 6 i 6 m, 1 6 j 6 n}.
m
P n
P
If v = λi vi and w = µj wj are elements of V and W respectively, we define
i=1 j=1
the elementary tensor
m X
X n
(4.1) v ⊗ w := λi µj (vi ⊗ wj ) ∈ V ⊗ W
i=1 j=1

Remarks 4.7.
(a) From the definition we see that dim V ⊗ W = (dim V )(dim W ).
(b) The elementary tensors span V ⊗ W .
(c) Not every element of V ⊗ W is an elementary tensor v ⊗ w.

It may appear that this definition depends on the choice of bases for V and W .
To address this, we have the following

Lemma 4.8. Let {v10 , · · · , vm

0
} and {w10 , · · · , wn0 } be other bases for V and W ,
respectively. Then

X 0 := {vi0 ⊗ wj0 : 1 6 i 6 m, 1 6 j 6 n}

is a basis for V ⊗ W .

Proof. The elementary tensors in V ⊗ W distribute, in the sense that

(v+v 0 )⊗(w+w0 ) = (v⊗w)+(v⊗w0 )+(v 0 ⊗w)+(v 0 ⊗w0 ) for all v, v 0 ∈ V, w, w0 ∈ W

and also that (λv) ⊗ w = λ(v ⊗ w) = v ⊗ (λw) for all v ∈ V, w ∈ W, λ ∈ k.

Using this observation, we can now write each vi as a linear combination of
{v10 , · · · , vm
0
} and each wj as a linear combination of {w10 , · · · , wn0 }, and see that
the original basis vectors vi ⊗ wj for V ⊗ W all lie in the span of X 0 . Therefore
X 0 spans V ⊗ W as a vector space, but being a set of size at most mn, it must be
linearly independent and is therefore a basis. 

Note that there is a canonical map

⊗ : V × W → V ⊗ W, (v, w) 7→ v ⊗ w

which is bilinear, in the sense that for all λ, µ ∈ k, v1 , v2 ∈ V, w1 , w2 ∈ W we have

(λv1 + v2 ) ⊗ (µw1 + w2 ) = λµ(v1 ⊗ w1 ) + λ(v1 ⊗ w2 ) + µ(v2 ⊗ w1 ) + (v2 ⊗ w2 ).

In fact, this map is universal with respect to this property, in the following sense.
18

Lemma 4.9 (Universal Property of Tensor Product). Let V and W be vector

spaces. Then for every bilinear map b : V × W → U to some third vector space U ,
there is a unique linear map eb : V ⊗ W → U such that

b(v, w) = eb(v ⊗ w) for all v, w ∈ V.

Proof. Fix bases {v1 , · · · , vn } for V and {w1 , · · · , wm } for W . Let b : V × W → U

be a bilinear map, and define b̃ : V ⊗ W → U to be the unique linear map which
P
sends the basis vector vi ⊗ wj ∈ V ⊗ W to b(vi , wj ). Let v = i λi vi ∈ V and
P P
w = j µj wj ∈ W . Then b̃ sends the elementary tensor v ⊗ w = λi µj vi ⊗ wj to
i,j
 
X X X
λi µj b(vi , wj ) = b  λ i vi , µj wj  = b(v, w)
i,j i j

because b is bilinear. So this linear map eb : V ⊗ W → U satisfies the required

property. If c : V ⊗ W → U is another linear map such that b(v, w) = c(v ⊗ w) for
all v ∈ V, w ∈ W , then c(vi ⊗ wj ) = b(vi , wj ) = eb(vi ⊗ wj ) shows that c agrees with
eb on a basis for V ⊗ W , so c = eb. 

Definition 4.10. Let V and W be finite dimensional kG-modules. Define a G-

action on the tensor product V ⊗ W by setting

g · (v ⊗ w) := (g · v) ⊗ (g · w) for all g ∈ G, v ∈ V, w ∈ W.

This defines the tensor product representation V ⊗ W .

It is straightforward to check that this defines a linear G-action on V ⊗ W in the

sense of Lemma 4.1, and therefore a G-representation.

Lemma 4.11. Let V and W be finite dimensional kG-modules. Then there is an

isomorphism of kG-modules
∼
=
V ∗ ⊗ W −→ Hom(V, W ).

Proof. For every f ∈ V ∗ and w ∈ W , we have a linear map b(f, w) : V → W

given by b(f, w)(v) := v 7→ f (v)w. The resulting map b : V ∗ × W → Hom(V, W ) is
bilinear, so by Lemma 4.9 it extends to a linear map

α : V ∗ ⊗ W → Hom(V, W )

given by α(f ⊗ w)(v) := f (v)w for all f ∈ V ∗ , w ∈ W, v ∈ V . Let {v1 , · · · , vn } be a

basis for V and let {v1∗ , · · · , vn∗ } be the corresponding dual basis for V ∗ . Then we
define a linear map
n
X
β : Hom(V, W ) → V ∗ ⊗ W, f 7→ vi∗ ⊗ f (vi ).
i=1
 19

Let f ∈ Hom(V, W ) and v ∈ V ; then

n n
α(vi∗ ⊗ f (vi ))(v) = vi∗ (v)f (vi )
P P
(α ◦ β)(f )(v) = α(β(f ))(v) =
 n i=1
 i=1
P ∗
= f vi (v)vi = f (v)
i=1

which shows that α ◦ β = 1Hom(V,W ) . We leave the verification that β ◦ α = 1V ∗ ⊗W

and that α is a homomorphism of kG-modules to Problem Sheet 2. 

It turns out that the tensor square V ⊗ V is reducible whenever dim V ≥ 2.

Definition 4.12. Suppose that char(k) 6= 2 and let V be a finite dimensional

vector space.
(a) For each v, w ∈ V , define
1
(v ⊗ w + w ⊗ v) ∈ V ⊗ V.
vw :=
2
The symmetric square of V is the following subspace of V ⊗ V :
S 2 V := h{vw : v, w ∈ V }i.
(b) For each v, w ∈ V , define
1
v ∧ w := (v ⊗ w − w ⊗ v) ∈ V ⊗ V.
2
The alternating square of V is the following subspace of V ⊗ V :
Λ2 V := h{v ∧ w : v, w ∈ V }i.

Note that vw = wv in S 2 V and that v ∧ w = −w ∧ v in Λ2 V for all v, w ∈ V .

Lemma 4.13. Let dim V = n and suppose that char(k) 6= 2.

(a) V ⊗ V = S 2 V ⊕ Λ2 V .
(b) dim S 2 V = n(n+1)
2 and dim Λ2 V = n(n−1)
2 .
(c) If V is a G-representation, then so are S 2 V and Λ2 V via
g · (vw) = (g · v)(g · w) and g · (v ∧ w) = (g · v) ∧ (g · w)
for all g ∈ G, v, w ∈ V .

Proof. (a) Let S2 := hσi be the cyclic group of order 2. Since char(k) 6= 2, the group
ring kS2 admits orthogonal idempotents e1 := 1+σ 2 ∈ kS2 and e2 := 1−σ 2 ∈ kS2 ,
which give rise to the ideal decomposition from Lemma 3.11
kS2 = kS2 e1 ⊕ kS2 e2 = ke1 ⊕ ke2 .
It follows that every kS2 -module M admits an even-odd decomposition
M = e1 M ⊕ e2 M = {m ∈ M : σm = m} ⊕ {m ∈ M : σm = −m}.
Now S2 acts linearly on V ⊗ V as follows:
σ · (v ⊗ w) = w ⊗ v for all v, w ∈ V.
20

Then S 2 V = e1 · (V ⊗ V ) is the even part of V ⊗ V and Λ2 V = e2 · (V ⊗ V ) is the

odd part of V ⊗ V . The even-odd decomposition then implies V ⊗ V = S 2 V ⊕ Λ2 V .
(b) If {v1 , . . . , vn } is a basis for V then {vi ⊗ vj : 1 6 i, j 6 n} spans V ⊗ V , so
{e1 ·(vi ⊗vj ) : 1 6 i, j 6 n} spans S 2 V = e1 ·(V ⊗V ). Now e1 ·(vi ⊗vj ) = vi vj = vj vi ,
so already {vi vj : 1 6 i 6 j 6 n} spans S 2 V , and therefore
n(n + 1)
dim S 2 V 6 .
2
Similarly, {e2 · (vi ⊗ vj ) = vi ∧ vj : 1 6 i < j 6 n} spans Λ2 V , and therefore
n(n − 1)
dim Λ2 V 6 .
2
But dim V ⊗ V = n2 , so V ⊗ V = S 2 V ⊕ Λ2 V implies the result.
(c) We have two groups acting on V ⊗ V , namely G and S2 . Now
σ · (g · (v ⊗ w)) = σ(g · v ⊗ g · w) = g · w ⊗ g · v = g · (w ⊗ v) = g · (σ · (v ⊗ w))
for any v, w ∈ V and g ∈ G. This means that these two actions commute pointwise.
In particular, the G-action preserves every S2 -submodule of V ⊗ V . Hence S 2 V
and Λ2 V inherit a linear G-action from V ⊗ V as claimed. 

Using similar ideas, it is possible to find proper kG-submodules of the tensor cube
V ⊗ V ⊗ V , one for each irreducible representation of kS3 . Similarly we obtain a
decomposition of the nth tensor power V ⊗n of V as a direct sum of kG-submodules
S λ (V ), one for each irreducible representation λ of the symmetric group Sn . This
construction V 7→ S λ (V ) is called a Schur functor.
 21

5. Character theory

We will now specialise to the case k = C. It turns out that G-representations

are determined by what appears to be very little information at all, namely the
knowledge of the character of that representation.

Definition 5.1. Let ρ : G → GL(V ) be a complex representation of G. The

character of ρ is the function
χρ : G → C, g 7→ tr ρ(g).
The degree of a character χρ is the degree of the representation ρ.
We will also write χV to denote the character of the representation afforded by
a CG-module V , when the CG-module structure on V is understood. This notion
only depends on the isomorphism class of the CG-module V , and the isomorphism
class of the representation ρ.

Definition 5.2. A function f : G → C is said to be a class function if it is constant

on conjugacy classes of G:
f (xgx−1 ) = f (g) for all g, x ∈ G.
We will denote the space of all class functions on G by C(G).

Lemma 5.3. The character χV of any finite dimensional kG-module V is a class

function.

Proof. If ρ : G → GL(V ) is the corresponding representation, then the linear

endomorphism ρ(g) of V is conjugate to ρ(xgx−1 ) in GL(V ). But conjugate linear
maps have the same trace: tr(ABA−1 ) = tr((AB)A−1 ) = tr(A−1 (AB)) = tr(B)
for any A, B ∈ GL(V ). 

Definition 5.4. Let G be a finite group, let {g1 , · · · , gs } be a set of representatives

for the conjugacy classes of G and let V1 , · · · , Vr be a complete list of representatives
for the isomorphism classes of simple CG-modules. The character table of G is the
r × s array with (i, j)-th entry being given by χVi (gj ).

Recall from Corollary 3.16 that in fact r = rC (G) is equal to s = s(G), so the
character table is always square. If a representation ρ is known, computing its
character is usually straightforward.

Example 5.5.
(a) The character table of the cyclic group of order 3, G = {1, x, x2 } is
1 x x2
1 1 1 1
χ 1 ω ω2
χ2 1 ω2 ω
22

2πi
where ω := e 3 is a primitive cube root of unity.
(b) Let G = S3 . In addition to the trivial character, we have the sign character
 : S3 → {±1} ⊂ C× , defined by
(
1 if σ is even,
(σ) =
−1 if σ is odd.
We also have the two-dimensional irreducible representation W of S3 from
Example 1.20. Since 12 + 12 + 22 = 6 = |S3 |, we have found all the characters,
and the character table of S3 is
1 (123) (12)
1 1 1 1
 1 1 −1
χW 2 −1 0

We record some basic facts about characters that we used in Example 5.5.

Lemma 5.6. Let ρ : G → GL(V ) be a finite dimensional representation.

(a) χV (1) = dim V .
(b) χV (g) = χV (1) if and only if ρ(g) = 1.
(c) If dim V = 1 then χ is a group homomorphism.
(d) If G is abelian and V is irreducible, then dim V = 1.

Proof. Problem Sheet 3. 

Characters of degree 1 are called linear characters. We have the following im-
portant consequence of our study of the group ring CG from §3.

Proposition 5.7. Let χ1 , · · · , χr be the complete list of characters of the irre-

ducible complex representations of the finite group G. Then

χ1 (1)2 + · · · + χr (1)2 = |G|

Proof. Suppose that the simple kG-module Vi affords the character χi . Then
χi (1) = dim Vi by Lemma 5.6(a); now use Corollary 3.20(b). 

Definition 5.8. Let N be a normal subgroup of the finite group G and let ρ :
G/N → GL(V ) be a representation. The inflated representation of G

ρ̇ : G → GL(V )

is defined by ρ̇(g) := ρ(gN ) for all g ∈ G.

Definition 5.9. Let G be a finite group. The derived subgroup, G0 , is the subgroup
of G generated by all commutators [x, y] := xyx−1 y −1 in G:

G0 := hxyx−1 y −1 : x, y ∈ Gi.
 23

Lemma 5.10. Let G be a finite group. Then G has precisely |G : G0 | distinct

complex linear characters.

Proof. Problem Sheet 3. 

Example 5.11. Let G = A4 be the alternating group of order 12. We know that
A4 has a normal subgroup of order 4, called the Klein four-group
V4 := {1, (12)(34), (14)(23), (13)(24)}.
Since A4 /V4 has order 3, it must be a cyclic group of order 3, hence abelian, so
A04 6 V4 and |A04 | ∈ {1, 2, 4}. No subgroup of order 2 in V4 is normal in A4 . So, A04
has to be V4 : it cannot be the trivial group since A4 is non-abelian.
Conclusion: A4 admits 3 distinct linear characters, inflated from A4 /V4 ∼ = C3 .

Definition 5.12. Let G be a finite group. The inner product on class functions
h−, −i : C(G) × C(G) → C
is defined as follows:
1 X
hϕ, ψi := ϕ(g)ψ(g)
|G|
g∈G

It is routine to verify that this is indeed a complex inner product on C(G), which
means that the following properties are satisfied for all ϕ, ψ ∈ C(G) and λ ∈ C:
• hλϕ, ψi = λhϕ, ψi and hϕ, λψi = λhϕ, ψi
• h−, −i is additive in both variables,
• hϕ, ψi = hψ, ϕi,
• hϕ, ϕi ≥ 0 with equality if and only if ϕ = 0.
We can now state our next big theorem.

Theorem 5.13 (Row Orthogonality). Let ϕ, ψ be irreducible characters of the

finite group G. Then (
1 if ϕ = ψ
hϕ, ψi =
0 if ϕ 6= ψ.

Theorem 5.13 has the following striking consequence.

Corollary 5.14. Let V and W be two finite dimensional kG-modules. Then V is

isomorphic to W if and only if χV = χW .

Proof. Let χ1 , · · · , χr be the complete list of characters of the irreducible complex

representations of G, and suppose that Vi is the simple kG-module with character
χi . By Maschke’s Theorem, Theorem 1.21, we know that V is a direct sum of
simple kG-modules. Since V1 , · · · , Vr are the only possible simple kG-modules (up
to isomorphism) we can find non-negative integers a1 , · · · , ar such that
V ∼
= V1a1 ⊕ · · · ⊕ Vrar .
24

We say that ai is the multiplicity of Vi in V . Passing to characters, we have

χV = a1 χ1 + · · · + ar χr

and now Theorem 5.13 tells us that we can recover ai from χV as follows:
r
X r
X
hχi , χV i = hχi , aj χj i = aj δi,j = ai .
j=1 j=1

Now, if χV = χW and W ∼ = V1b1 ⊕ · · · ⊕ Vrbr as a kG-module, then for each

i = 1, . . . , r, ai = hχi , χV i = hχi , χW i = bi . Therefore V ∼
= V1a1 ⊕ · · · ⊕ Vrar =
V1b1 ⊕ · · · Vrbr = W and hence V ∼ = W . The converse implication is trivial. 

Here is an another consequence of Theorem 5.13.

Corollary 5.15. The irreducible characters of G form an orthonormal basis for

C(G).

Proof. We know that hχi , χj i = δi,j by Theorem 5.13, so the χi ’s are pairwise or-
thogonal elements of the inner product space C(G) by Lemma 5.3. Now dim C(G) =
s(G) = rC (G) = r by Corollary 3.16, so {χ1 , · · · , χr } forms a basis for C(G). 

Let us see how to use Theorem 5.13 to compute character tables: we can complete
the analysis that we started in Example 5.11 above, but first, some notation.

Definition 5.16. Let G be a finite group and let g ∈ G.

(a) g G denotes the conjugacy class of g in G:

g G := {g x : x ∈ G} where g x := x−1 gx.

(b) CG (g) denotes the centraliser of g in G:

CG (g) = {x ∈ G : gx = xg}.

Lemma 5.17. For any g ∈ G we have |g G | · |CG (g)| = |G|.

Proof. Apply the Orbit-Stabiliser Theorem from Prelims M1 to the conjugation

action of G on itself: the orbit of g ∈ G is its conjugacy class, and the stabiliser of
g is precisely the centraliser CG (g). 

Example 5.18. Let G = A4 . Then A04 = V4 and G has 3 distinct linear characters
by Example 5.11. The representatives for the conjugacy classes in A4 are 1, g2 :=
(12)(34), g3 := (123) and g4 := (132). So, A4 has exactly 4 conjugacy classes, and
therefore rC (G) = s(G) = 4 because the character table is square by Corollary 3.16.
 25

So, the character table of A4 looks as follows:

g 1 g2 g3 g4
G
|g | 1 3 4 4
|CG (g)| 12 4 3 3
χ1 1 1 1 1
χ2 1 1 ω ω2
χ3 1 1 ω2 ω
χ4 d a b c
2πi
where ω = e 3 . Proposition 5.7 tells us that 12 + 12 + 12 + d2 = |G| = 12. Then
Lemma 5.6(a) implies that d = 3 is the only possibility. Now, by Theorem 5.13:

0 = |G|hχ1 , χ4 i = 1 · 1 · 3 + |g2G | · 1 · a + |g3G | · 1 · b + |g4G | · 1 · c = 3 + 3a + 4b + 4c

and

0 = |G|hχ2 , χ4 i = 1 · 1 · 3 + |g2G | · 1 · a + |g3G | · ω · b + |g4G | · ω 2 · c = 3 + 3a + 4ωb + 4cω 2

and

0 = |G|hχ3 , χ4 i = 1 · 1 · 3 + |g2G | · 1 · a + |g3G | · ω 2 · b + |g4G | · ω · c = 3 + 3a + 4ω 2 b + 4cω.

Solving these linear equations we see that a = −1 and b = c = 0. So, the full
character table of A4 is

g 1 g2 g3 g4
G
|g | 1 3 4 4
|CG (g)| 12 4 3 3
χ1 1 1 1 1
χ2 1 1 ω ω2
χ3 1 1 ω2 ω
χ4 3 −1 0 0.

We now start working towards the proof of Theorem 5.13.

Definition 5.19. Let V be a CG-module. The invariant submodule of V is

V G := {v ∈ V : g · v = v for all g ∈ G}.

Clearly V G is the largest subspace of V on which G acts trivially. How do we

calculate the dimension of V G ? Using a special case of Theorem 5.13:

Proposition 5.20 (Fixed Point Formula). Let G be a finite group and let V be a
finite dimensional CG-module. Then
1 X
dim V G = χV (g) = h1, χV i.
|G|
g∈G
26

1
g ∈ CG. Then ge = eg = e for all g ∈ G so e2 = e; this is
P
Proof. Let e := |G|
g∈G
why e is called the principal idempotent of CG. We have

V = e · V ⊕ (1 − e) · V.

Now if g ∈ G then g · (e · v) = (ge) · v = e · v so e · V 6 V G ; on the other hand if

v ∈ V G then g · v = v for all g ∈ G, so |G|e · v =
P
g · v = |G|v whence v = e · v
g∈G
and v ∈ e · V . We have shown that

e · V = V G.

The action of e ∈ CG on V is a linear map eV : V → V which is an idempotent

with image e · V . So, writing ρ : G → GL(V ) for the representation afforded by V ,
1 X 1 X
dim V G = dim e · V = tr(eV ) = tr ρ(g) = χV (g). 
|G| |G|
g∈G g∈G

Before we give the proof of Theorem 5.13, we need to compute the characters
of the representations obtained through various multilinear constructions from §4.
Observe that the vector space of class functions C(G) is in fact a commutative ring,
via pointwise multiplication of functions:

(ϕψ)(g) := ϕ(g)ψ(g) for all g ∈ G.

Proposition 5.21. Let G be a finite group and let V , W be finite dimensional

CG-modules. Then we have the following equalities in C(G):
(a) χV ∗ = χV ,
(b) χV ⊕W = χV + χW ,
(c) χV ⊗W = χV χW ,
(d) χHom(V,W ) = χV χW ,
(e) χS 2 V (g) = 21 (χV (g)2 + χV (g 2 )) for all g ∈ G,
(f) χΛ2 V (g) = 21 (χV (g)2 − χV (g 2 )) for all g ∈ G.

Proof. Fix g ∈ G. By Problem Sheet 1, the action gV ∈ GL(V ) of g on V is

diagonalisable. Fix a basis of gV -eigenvectors {v1 , · · · , vn } for V with corresponding
eigenvalues λ1 , · · · , λn , and fix a basis of gW -eigenvectors {w1 , · · · , wm } for W with
eigenvalues µ1 , · · · , µm . Then
n
X m
X
χV (g) = tr(gV ) = λi and χW (g) = tr(gW ) = µj .
i=1 j=1

(a) Let {v1∗ , · · · , vn∗ } be the dual basis for V ∗ relative to {v1 , · · · , vn }. Then

(g · vi∗ )(vj ) = vi∗ (g −1 · vj ) = vi∗ (λ−1 −1 −1 ∗

j vj ) = λj δi,j = (λi vi )(vj )

for all i, j = 1, · · · , n shows that

g · vi∗ = λ−1 ∗
i vi for all i = 1, · · · , n.
 27

But each λi is a root of unity by Problem Sheet 1, so g · vi∗ = λi vi∗ and hence
n
X
χV ∗ (g) = tr(gV ∗ ) = λi = tr(gV ) = χV (g).
i=1

(c) By our definition of V ⊗ W — see Definition 4.6 — the elementary tensors

{vi ⊗ wj : 1 6 i 6 n, 1 6 j 6 m} form a basis for V ⊗ W . Using Definition 4.10,

g · (vi ⊗ wj ) = (g · vi ) ⊗ (g · wj ) = (λi vi ) ⊗ (µj wj ) = λi µj (vi ⊗ wj ).

In other words, the elementary tensors {vi ⊗ wj : 1 6 i 6 n, 1 6 j 6 m} form a

basis of eigenvectors for the g-action on V ⊗ W , with vi ⊗ wj having eigenvalue
λi µj . Therefore

n Xm n
! m 
X X X
χV ⊗W (g) = λi µj = λi  µj  = χV (g)χW (g).
i=1 j=1 i=1 j=1

(d) By Lemma 4.11, there is a natural isomorphism of G-representations

∼
=
V∗⊗W −→ Hom(V, W ).

Isomorphic representations have the same character. Then (a) and (c) give

χHom(V,W ) = χV ∗ ⊗W = χV ∗ χW = χV χW .

(b,e,f) We leave these to Problem Sheet 3. 

Proposition 5.22. Let V and W be finite dimensional CG-modules. Then

(a) HomCG (V, W ) = Hom(V, W )G and
(b) hχV , χW i = dim HomCG (V, W ).

Proof. (a) Let f ∈ Hom(V, W ). Then f is fixed by the G-action if and only if

g · f (g −1 · v) = f (v) for all g ∈ G, v ∈ V

or in other words, if and only if

gW ◦ f = f ◦ gV for all g ∈ G.

Definition 1.12 tells us that this is the same as f ∈ HomCG (V, W ).

(b) The key idea is to apply the Fixed Point Formula to the G-representation
Hom(V, W ) from Definition 4.4. By Proposition 5.20 and Proposition 5.21(d),
1 X 1 X
dim Hom(V, W )G = χHom(V,W ) (g) = χV (g)χW (g) = hχV , χW i. 
|G| |G|
g∈G g∈G

We can now prove the Row Orthogonality Theorem.

28

Proof of Theorem 5.13. Let V and W be the simple CG-modules whose characters
are ϕ = χV and ψ = χW , respectively. Since V and W are simple, Schur’s Lemma
(Theorem 3.6) and Lemma 2.13 together tell us that
(
1 if V ∼
=W
dim HomCG (V, W ) =
0 if V  W.

Using Proposition 5.22, we conclude that always we have

hϕ, ψi = hχV , χW i = dimCG Hom(V, W ) ∈ {0, 1}.

Suppose that χV = χW . Then χV = χW so,

1 X (dim V )2
hχV , χW i = ||χV ||2 = |χV (g)|2 ≥ >0
|G| |G|
g∈G

because χV (1) = dim V by Lemma 5.6(a). So in this case necessarily hχV , χV i = 1.

Suppose now χV 6= χW . Then V cannot be isomorphic to W as isomorphic
representations have the same characters, so hϕ, ψi = dim HomCG (V, W ) = 0. 

Next we come to another consequence of the Row Orthogonality Theorem, which

will turn out to be very useful for completing character tables.

Theorem 5.23 (Column Orthogonality). Let G be a finite group, let χ1 , · · · , χr

be the irreducible characters of G and let g, h ∈ G. Then

r
(
X |CG (g)| if g is conjugate to h,
χi (g)χi (h) =
i=1
0 otherwise.

Proof. Let {g1 , · · · , gr } be a complete list of representatives for the conjugacy

classes of G, recalling from Corollary 3.16 that the number s(G) of these classes
equals the number r = rC (G) of isomorphism classes of simple CG-modules.
Say and g ∈ gjG and h ∈ gkG for some j, k. Since the characters χi are class
functions by Lemma 5.3, we may assume that g = gj and h = gk .
We have already seen in Example 5.18 that the character table viewed as a matrix
(χi (gj )) does not quite form a unitary matrix, because we have the remember the
weights |gjG |/|G| when computing the inner product of two rows of this matrix. To
fix this, we introduce a fudge factor as follows: define for each i, j = 1, · · · , r
q
xi,j := χi (gj ) · cj where cj := |gjG |/|G|.
 29

We compute that for any i, k = 1, · · · , r we have

r
P r
P
xi,j xk,j = χi (gj )cj χk (gj )cj
j=1 j=1
r
χi (gj )χk (gj )c2j
P
=
j=1
r
1
|gjG |χi (gj )χk (gj )
P
= |G|
j=1
1
P
= |G| χi (x)χk (x)
x∈G
= hχi , χk i.

Since hχi , χk i = δi,k by the Row Orthogonality theorem, Theorem 5.13, this means
that the r × r matrix X := (xi,j ) is unitary:

X · X T = I.

So X is the left-inverse of X T in GLr (C). It is also the right-inverse, by Part A

Linear Algebra; applying complex conjugation to X T · X = I, we obtain
T
X · X = I.

Hence for any j, k = 1, · · · , r, we have

r r
T X X
(X · X)j,k = xi,j xi,k = χi (gj )cj χi (gk )ck = δj,k .
i=1 i=1

Divide both sides by cj ck : by Lemma 5.17 we have 1/c2j = |G|/|gjG | = |CG (gj )|. 

Example 5.24. Let G be the symmetric group S4 . The conjugacy class represen-
tatives are g1 = 1, g2 = (12)(34), g3 = (123), g4 = (12), g5 = (1234), with conjugacy
classes of sizes 1, 3, 8, 6, 6 respectively. We know from Problem Sheet 0 that V4 is
a normal subgroup of G with S4 /V4 being isomorphic to the symmetric group S3 .
This gives us three irreducible characters 1 e, e
, χ
g W obtained by inflation from S3
using Example 5.5(b):

g 1 g2 g3 g4 g5
G
|g | 1 3 8 6 6
|CG (g)| 24 8 3 4 4
1e 1 1 1 1 1

e 1 1 1 −1 −1
χg W 2 2 −1 0 0
χ4 d4 α4 β4 γ4 δ4
χ5 d5 α5 β5 γ5 δ5

Since rC (G) = s(G) = 5 by Corollary 3.16, we are missing two irreducible char-
acters χ4 and χ5 of degrees d4 and d5 , say. Then d24 + d25 = 24 − 12 − 12 − 22 = 18
by Theorem 5.23 with g = h = 1 (or Corollary 3.20(b)). The only solution to this
30

equation in positive integers is d3 = d4 = 3. Now apply Theorem 5.23 to the first

pair of columns and then to the second column to obtain

1 + 1 + 4 + 3α4 + 3α5 = 0, 12 + 12 + 22 + |α4 |2 + |α5 |2 = 8.

Hence α4 + α5 = −2 and |α|2 + |α|2 = 2, and in this situation Problem Sheet 0,

Question 4 tells us that α4 = α5 = −1. Next, Theorem 5.23 applied to the third
column tells us that

12 + 12 + (−1)2 + |β4 |2 + |β5 |2 = 3

whence β4 = β5 = 0. Similar considerations show that γ5 = −γ4 = δ4 = −δ5

and that |γ4 | = 1. Now, g4 = (12) has order 2 so it acts with eigenvalues ±1 in
any representation. Hence γ4 , the sum of these eigenvalues, is a real number, so
γ4 ∈ {1, −1}. Without loss of generality, we may assume that γ4 = 1, otherwise we
can swap χ4 with χ5 .
We conclude that the complete character table of S4 is
g 1 g2 g3 g4 g5
|g G | 1 3 8 6 6
|CG (g)| 24 8 3 4 4
1e 1 1 1 1 1

e 1 1 1 −1 −1
χg W 2 2 −1 0 0
χ4 3 −1 0 1 −1
χ5 3 −1 0 −1 1

6. Induced representations

The Row and Column orthogonality theorems, Theorems 5.13 and Theorem 5.23
are powerful tools but they are only good for completing character tables and not for
constructing them. We will now introduce a powerful new technique of constructing
representations of G using representations of its proper subgroups. This technique
is called induction.
We return to the setting of kG-modules over a general field k for now.

Definition 6.1. Let H be a finite group and let V be a kH-module. The vector
space of H-coinvariants VH of V is

VH := V /k{h · v − v : h ∈ H, v ∈ V }.

Equivalently, this is the largest quotient kH-module of V which is isomorphic to a

trivial module.

Whenever W is a vector space, the set kG ⊗ W is again a vector space which is

isomorphic to a direct sum of |G| copies of W . The proof of the following Lemma
can be safely left as an exercise.
 31

Lemma 6.2. Let H be a subgroup of the finite group G, let W be a kH-module.

(a) There is a linear G-action on kG ⊗ W given on elementary tensors by
g · (x ⊗ w) := gx ⊗ w for all g, x ∈ G, w ∈ W.
We call this the left G-action.
(b) There is a linear H-action on kG ⊗ W given on elementary tensors by
h ? (x ⊗ w) := xh−1 ⊗ h · w for all h ∈ H, x ∈ G, w ∈ W.
We call this the right H-action.
(c) The two actions commute pointwise:
g · (h ? u) = h ? (g · u) for all g ∈ G, h ∈ H, u ∈ kG ⊗ W.

Corollary 6.3. Let H be a subgroup of the finite group G and let W be a kH-
module. Then the space of H-coinvariants (kG ⊗ W )H of kG ⊗ W with respect to
the right H-action is a kG-module.

Proof. The action of G on kG ⊗ W preserves the subspace of (H − 1) ∗ (kG ⊗ W ):

for each u ∈ kG ⊗ W , g ∈ G, h ∈ H we have
g · (h ? u − u) = h ? (g · u) − (g · u) ∈ (H − 1) ∗ (kG ⊗ W )
by Lemma 6.2(c). So, the G-action on kG ⊗ W from Lemma 6.2(a) descends to the
right H-coinvariants (kG ⊗ W )H = (kG ⊗ W )/(H − 1) ∗ (kG ⊗ W ). 

Definition 6.4. Let H be a subgroup of G and let W be a kH-module.

(a) The induced kG-module IndG
H W is the space of H-coinvariants

IndG
H W := (kG ⊗ W )H

of kG ⊗ W under the right H-action from Lemma 6.2.

(b) If σ : H → GL(W ) is the representation of H afforded by W , we call the
representation of G afforded by IndG
H W the induced representation and denote
it by IndG
H σ.
(c) We will write g ⊗ w to denote the image of g ⊗ w ∈ kG ⊗ W in the quotient
vector space IndG
H W = (kG ⊗ W )H .

Lemma 6.5. Let g ∈ G and w ∈ W . Then

(a) gh ⊗ w = g ⊗ h · w for all h ∈ H, and
(b) g · (x ⊗ w) = gx ⊗ w for all x ∈ G.

What does IndG

H W look like as a vector space?

Lemma 6.6. Let x ∈ G. Then xkH ⊗ W is an H-stable subspace of kG ⊗ W under

the right H-action, and there is a linear isomorphism
∼
=
α : W −→ (xkH ⊗ W )H , w 7→ x ⊗ w,
so that (xkH ⊗ W )H = x ⊗ W .
32

Proof. The first assertion is an easy exercise. By Lemma 4.9, there is a linear map
β : xkH ⊗ W → W given on elementary tensors by β(xh ⊗ w) = h · w for all h ∈ H
and w ∈ W . Now for any y ∈ H we have

β(y ? (xh ⊗ w)) = β(xhy −1 ⊗ y · w) = (hy −1 ) · (y · w) = h · w.

Hence β is zero on (H − 1) ? (xkH ⊗ W ), so it descends to a well-defined linear map

β : (xkH ⊗ W )H → W, xh ⊗ w 7→ h · w.

Then using Lemma 6.5(a), we compute that for every h ∈ H and w ∈ W ,

α(β(xh ⊗ w)) = α(h · w) = x ⊗ h · w = xh ⊗ w.

Since xkH ⊗ W is spanned by vectors of the form xh ⊗ w, this shows that α ◦ β =

1(xkH⊗W )H . Checking β ◦ α = 1W is easy. Hence α is an isomorphism. 

Proposition 6.7. Let {x1 , · · · , xm } be a complete set of left coset representatives

for H in G so that G = x1 H ∪ · · · ∪ xm H.
(a) There is a vector space decomposition

IndG
H W = (x1 ⊗ W ) ⊕ (x2 ⊗ W ) ⊕ · · · ⊕ (xm ⊗ W ).

(b) We have dim IndG

H W = |G : H| dim W .

Proof. By Lemma 6.6, xkH ⊗ W is an H-stable subspace of kG ⊗ W for the right

H-action for any x ∈ G. Since G is the disjoint union of the left cosets xi H,
m
M
kG ⊗ W = xi kH ⊗ W.
i=1

The operation of taking H-coinvariants commutes with direct sums up to isomor-

phism. Taking H-coinvariants and applying Lemma 6.6 gives
m m
∼
M M
IndG
H W = (kG ⊗ W )H = (xi kH ⊗ W )H = xi ⊗ W,
i=1 i=1

where xi ⊗ W ∼
= W for each i. Now take dimensions. 

Example 6.8. There is a kG-linear isomorphism k(G/H) ∼ H 1 between the

= IndG
H 1.
permutation module k(G/H) from Definition 1.5 and the induced module IndG

Definition 6.9. Let H be a subgroup of G and let V be a kG-module. Then V is

also a kH-module by restricting the action of kG to its subring kH; we denote the
resulting kH-module ResGHV.

In categorical terms (see Part C Category Theory), our next result tells us that
“induction is left-adjoint to restriction”.
 33

Proposition 6.10. Let H be a subgroup of the finite group G, let W be a kH-

module and let U be a kG-module. Then there is a linear isomorphism
∼
=
Φ : HomkG (IndG
H W, U ) −→ HomkH (W, ResG
H U)

given by Φ(α)(w) = α(1 ⊗ w) for all α ∈ HomkG (IndG

H W, U ) and all w ∈ W .

G
Proof. The map W → ResG H IndH W given by w 7→ 1 ⊗ w is kH-linear. Hence,
given a kG-linear α : IndGH W → U , we can view α as being a kH-linear map
G G G
ResH IndH W → ResH U by restriction, and then precompose it with w 7→ 1 ⊗ w
to obtain the kH-linear map Φ(α). This shows that Φ is well-defined.
We will now construct a map Ψ in the opposite direction. Take some kH-linear
G
map β : W → ResG H U and define Ψ(β) : IndH W → U by setting

Ψ(β)(g ⊗ w) := g · β(w) for all w ∈ W, g ∈ G.

To see that this is well-defined, we must show that gh · β(w) = g · β(h · w) for all
g ∈ G, h ∈ H, w ∈ W . But this follows immediately from the hypothesis that β is
kH-linear. We check that Ψ(β) is kG-linear as follows: for all g, x ∈ G, w ∈ W ,

Ψ(β)(g · (x ⊗ w)) = Ψ(β)(gx ⊗ w) = (gx) · β(w) = g · (x · β(w)) = g · Ψ(β)(x ⊗ w).

Since everything in sight is linear, this defines the required linear map
G
Ψ : HomkH (W, ResG
H U ) → HomkG (IndH W, U ).

We now show Φ and Ψ are mutually inverse. If α : IndG

H W → U is kG-linear, then

Ψ(Φ(α))(g ⊗ w) = g · Φ(α)(w) = g · α(1 ⊗ w) = α(g ⊗ w) for all g ∈ G, w ∈ W

which shows Ψ(Φ(α)) = α, and similarly for any β ∈ HomkH (W, ResG
H U ) we have

Φ(Ψ(β))(w) = Ψ(β)(1 ⊗ w) = 1 · β(w) = β(w) for all w∈W

which shows that Φ(Ψ(β)) = β. 

We will now again specialise to the case k = C.

Definition 6.11. Let H be a subgroup of the finite group G.

(a) Let ψ be a character of G afforded by the CG-module V . Then the restricted
character is the character ResG G
H ψ of the CH-module ResH V .
(b) Let ϕ be a character of H afforded by the CH-module W . Then the induced
character is the character IndG G
H ϕ of the CG-module IndH W .

Since there are now two groups in play, we will denote the inner product on C(G)
from Definition 5.12 by using a subscript:

h−, −iG : C(G) × C(G) → C.

34

Corollary 6.12 (Frobenius Reciprocity). Let ϕ be a character of H and let ψ be

a character of G. Then
hIndG G
H ϕ, ψiG = hϕ, ResH ψiH .

Proof. Let U be the CG-module whose character is ψ, and let W be the CH-module
whose character is ϕ. Applying Proposition 5.22 twice, we have
hIndGH ϕ, ψiG = dim HomCG (IndG
H W, U ), and
hϕ, ResG
H ψiH = G
dim HomCH (W, ResH U ).
Now apply Proposition 6.10. 

Example 6.13. Let U be a simple CG-module and consider the trivial character
H 1 is the character of the free CG-module of rank 1 by
1 of H := {1}. Then IndG
Example 6.8. Then Frobenius Reciprocity tells us that

H 1, χU iG = h1, ResH χU iH = dim U.

hIndG G

In other words, U appears in CG precisely dim U times. We knew this already —

see Corollary 3.20(a).

Since ψ(h) is the trace of the action of an element h ∈ H on U , it is clear that

ResGH ψ is simply the restriction of the class function ψ ∈ C(G) to H; the resulting
function on H is still constant on H-conjugacy classes and in this way we have a
natural linear map
ResG
H : C(G) → C(H).

Corollary 6.12 suggests that there is linear map in the opposite direction
IndG
H : C(H) → C(G)

which is an adjoint for ResG

H in the sense of Part A Linear Algebra. To see that this
is indeed the case, we must compute the character of our induced module IndG H W.
We start this calculation with the following Lemma.

Lemma 6.14. Let H be a subgroup of the finite group G. Let {x1 , · · · , xm } be a

complete set of left coset representatives for H in G so that G = x1 H ∪ · · · ∪ xm H.
For each g ∈ G, write
gxi H = xg·i H for all i = 1, . . . , m
for some permutation i 7→ g · i of {1, · · · , m}. Define
Fix(g) := {i ∈ {1, · · · , m} : g · i = i}
to be the fixed-point set of g in its action on {1, · · · , m}. Then for every finite
dimensional CH-module W , we have
(a) g · (xi ⊗ W ) ⊆ xg·i ⊗ W for all i = 1, · · · , m, and
(b) (IndG χW (x−1
P
H χW )(g) = i gxi ).
i∈Fix(g)
 35

Proof. (a) Using Lemma 6.5, we have

g · (xi ⊗ w) = gxi ⊗ w = xg·i (x−1 −1
g·i gxi ) ⊗ w = xg·i ⊗ (xg·i gxi ) · w ∈ xg·i ⊗ W

for any i = 1, · · · , m and any w ∈ W .

(b) Let ρ : G → GL(IndG H W ) be the representation afforded by the CG-module
G
IndH W . Then by Proposition 6.7(a) we have a decomposition of vector spaces
IndG
H W = (x1 ⊗ W ) ⊕ (x2 ⊗ W ) ⊕ · · · ⊕ (xm ⊗ W )

and by part (a), ρ(g) permutes the direct summands xi ⊗ W in the same way as g
permutes the numbers {1, · · · , m} (or equivalently, in the same way as g permutes
the left cosets G/H). Writing down the matrix of ρ(g) with respect to a choice
of basis for IndG
H W which respects the direct sum decomposition, we see that the
block matrices on the diagonal corresponding to the indices i that are not fixed by
g are all zero. However if g · i = i then ρ(g) preserves xi ⊗ W , and its restriction to
this subspace has trace equal to the trace of the action of x−1i gxi ∈ H on W . 

It turns out that there is a more invariant way of expressing the result of this
calculation, which does not depend on the choice of coset representatives.

Definition 6.15. For each ϕ : H → C, define its extension by zero to G to be the

function ϕ◦ : G → C which agrees with ϕ on H and which is zero on G\H.

Theorem 6.16. Let H be a subgroup of the finite group G and let W be a finite
dimensional CH-module. Then for all g ∈ G we have
1 X ◦ −1
(IndG
H χW )(g) = χW (x gx).
|H|
x∈G

Proof. Fix g ∈ G. Using the notation of Lemma 6.14, we note that i ∈ Fix(g) ⇔
gxi H = xi H ⇔ x−1
i gxi ∈ H. So, we can rewrite Lemma 6.14(b) as follows:
X m
X
(IndG
H χW )(g) = χW (x−1
i gxi ) = χ◦W (x−1
i gxi ).
i∈Fix(g) i=1

Note that χ◦W (h−1 yh)= χ◦W (y)

for any h ∈ H and any y ∈ G because χW is a
class function on H and because both H and G\H are stable under conjugation by
H. Therefore χ◦W ((xi h)−1 g(xi h)) = χ◦W (x−1
i gxi ) for any h ∈ H, so
X m X
X m
X
χ◦W (x−1 gx) = χ◦W ((xi h)−1 g(xi h)) = |H| χ◦W (x−1
i gxi ). 
x∈G i=1 h∈H i=1

Remark 6.17. It could very well happen that g G ∩ H = ∅. In this case, Theorem
6.16 tells us that (IndG
H χW )(g) = 0.

Corollary 6.18. Let H be a subgroup of G. For each ϕ ∈ C(H), define

1 X ◦ −1
(IndG
H ϕ)(g) := ϕ (x gx).
|H|
x∈G
36

Then IndG G
H : C(H) → C(G) is left adjoint to the map ResH : C(G) → C(H) which
sends ψ ∈ C(G) to its restriction ψ|H : H → C:

hIndG G
H ϕ, ψiG = hϕ, ResH ψiH for all ϕ ∈ C(G), ψ ∈ C(H).

Proof. The formula holds true when ϕ and ψ are characters of representations of
H and G, respectively, by Frobenius Reciprocity (Corollary 6.12) together with
Theorem 6.16. The result follows because characters of representations span class
functions, by Corollary 5.15. 

As an exercise, prove Corollary 6.18 directly. For practical calculations, the

following reformulation of Theorem 6.16 will be useful.

Corollary 6.19. Let H be a subgroup of G. Let g ∈ G, and let {h1 , · · · , h` } be a

complete set of representatives for the conjugacy classes in H contained in g G ∩ H:

g G ∩ H = hH H
1 ∪ · · · ∪ h` .

Then for every finite dimensional CH-module W ,

`
|G| X |hH i |
χIndG W (g) = χW (hi ).
H |H| i=1 |g G |

Proof. Consider the set S := {x ∈ G : x−1 gx ∈ H}. We can rewrite it as follows:

[
S= {x ∈ G : x−1 gx = y}.
y∈g G ∩H

For a fixed y = x−1 G

0 gx0 ∈ g ∩H, there is a bijection CG (g) → {x ∈ G : x
−1
gx = y}
given by z 7→ zx0 . Applying Theorem 6.16, we obtain

χW (x−1 gx) =
P P̀ P
|H|χIndG
H W
(g) = |CG (g)|χW (y)
x∈S i=1 y∈hH
i

|hH
P̀
= |CG (g)| i |χW (hi ).
i=1

Now divide by |H| and apply Lemma 5.17. 

Next, we specialise to the case where we have a normal subgroup N of G. In

this case, conjugation by any x ∈ G preserves N .

Definition 6.20. Let N be a normal subgroup of G, let s ∈ G and let ϕ : N → C.

The x-twist of ϕ is the function

ϕx : N → C, h 7→ ϕ(x−1 hx).

Since conjugation by x ∈ G is an automorphism of N , it permutes the conjugacy

classes of N , so ϕx ∈ C(N ) whenever ϕ ∈ C(N ). When ϕ ∈ C(N ), ϕx only depends
on xN ∈ G/N . This defines a permutation action of G/N on C(N ), via

xN · ϕ = ϕx for all xN ∈ G/N, ϕ ∈ C(N ).

 37

Proposition 6.21. Let ϕ be a character of the normal subgroup N of G. Then

(a) ϕx is another character of N , and
G
(b) ResGN IndN ϕ = ϕ
x1
+ · · · + ϕxm , where {x1 , · · · , xm } is a complete set of left
coset representatives for N in G.

Proof. Let W be the CN -module with character ϕ, let h ∈ N and fix i = 1, · · · , m.

Since N is normal, hxi N = xi (x−1
i hxi N ) = xi N shows that multiplication by h
fixes each point of G/N . So, Lemma 6.14 implies that xi ⊗ W is stable under
the action of N , so xi ⊗ W is another CN -module. In fact, the decomposition of
G
ResGN IndN W into a direct sum of vector subspaces xi ⊗ W from Lemma 6.7 is a
decomposition into a direct sum of CN -modules, so
m
X
G
ResG
N IndN χW = χxi ⊗ W.
i=1

But χxi ⊗ W = χxWi by the proof of Lemma 6.14(b). 

We can now give a construction of irreducible characters using induction.

Corollary 6.22. Let ϕ be an irreducible character of the normal subgroup N of

G, such that ϕx 6= ϕ for all x ∈ G\N . Then the character IndG
N ϕ is irreducible.

Proof. We use Frobenius Reciprocity, Corollary 6.12, to compute || IndG 2

N ϕ|| :

hIndG G G G
N ϕ, IndN ϕiG = hResN IndN ϕ, ϕiN .
m
G
But ResG x1
+ · · · + ϕxm by Proposition 6.21, where G =
S
N IndN ϕ = ϕ xi N . We
i=1
x
may assume that x1 = 1; our assumption ϕ 6= ϕ for all x ∈ G\N then implies that
ϕxi 6= ϕ for i ≥ 2. Because each ϕxi is irreducible, hϕ, ϕiN = 1 and hϕxi , ϕiN = 0
for i ≥ 2 by Theorem 5.13. Hence || IndG 2
N ϕ|| = 1 and now we can use Corollary
G
5.15 to deduce that IndN ϕ is irreducible. 

Using Corollary 6.22, we can now find all irreducible characters of the dihedral
groups D2n . We only treat the case where n = 2m + 1 is odd, and leave the even
case as an exercise.

Example 6.23. Let G be the dihedral group of order 4m + 2 for some m ≥ 1.

Then G has m irreducible characters of degree 2, and two linear characters.

Proof. Write G = hr, s : srs−1 = r−1 i and note that N := hri is a normal cyclic
subgroup of G of order 2m + 1. By Lemma 5.10, N has 2m + 1 linear characters
{ϕi : 0 6 i 6 2m} where ϕ : N → C× sends r to some fixed primitive (2m + 1)th
root of unity ζ, say. We calculate

(ϕi )s (rj ) = ϕi (s−1 rj s) = ϕi (r−j ) = ζ −ij = ϕ2m+1−i (rj ) for all i, j = 0, · · · , 2m.
38

So, the trivial character 1 = ϕ0 of N is fixed under the conjugation action of G on

C(N ), and the other irreducible characters {ϕ1 , · · · , ϕ2m } break up into m orbits
of size 2. For each i = 1, · · · , m, the induced character
χi := IndG
N ϕ
i

is irreducible by Corollary 6.22, and its restriction back to N is equal to ϕi +ϕ2m+1−i

by Proposition 6.21(b). This implies that χ1 , · · · , χm are pairwise distinct degree
2 characters of G, by Proposition 6.7(b). Inflation from the cyclic group G/N of
order 2 gives two distinct linear characters of G — see Definition 5.8. Since
m
X
χi (1)2 + 12 + 12 = 4m + 2 = |G|,
i=1

Corollary 3.20(b) says that we have found all of the irreducible characters of G. 

7. Algebraic integers and Burnside’s pα q β theorem

Definition 7.1. Let z ∈ C.

(a) We say that z is an algebraic number if z satisfies a polynomial equation with
rational coeffcients.
(b) We say that z is an algebraic integer if z satisfies a monic polynomial equation
with integer coefficients.
(c) The set of algebraic integers is denoted A.

From Part A Rings and Modules, we know that the set of algebraic numbers is
the union of all subfields of C of finite dimension as a Q-vector space.

Examples 7.2.
(a) Every integer a ∈ Z is an algebraic integer, being a root of t − a = 0.
(b) Every root of unity is an algebraic integer.
(c) If z is an algebraic number, then mz is an algebraic integer for some integer m.
(d) A rational number α ∈ Q is an algebraic integer if and only if α ∈ Z. To see
this, write α = r/s with r, s coprime integers and suppose that
 r n  r n−1 r
+ an−1 + · · · + a1 + a0 = 0
s s s
for some a0 , a1 , · · · , an−1 ∈ Z. Clearing denominators we obtain the equation
rn + an−1 rn−1 s + · · · + a1 rsn−1 + a0 sn = 0
and therefore s divides rn . Since s and r are coprime, this is only possible if
s = ±1 and then α is an integer.

We have the following remarkable and fundamental fact.

Theorem 7.3. Algebraic integers form a subring of C.

The heart of the proof of Theorem 7.3 is the following useful

 39

Proposition 7.4. Let M be a finitely generated subgroup of (C, +). Then

{z ∈ C : zM ⊆ M } ⊂ A.

Proof. Suppose zM ⊆ M . Choose a generating set {v1 , · · · , vd } for M and write

d
X
(7.1) zvi = uij vj for all i = 1, · · · , d
j=1

for some integers uij ∈ Z. Consider the characteristic polynomial g(t) := det(tI−U )
of the d × d matrix U = (uij ). Since all entries of U are integers, g(t) is a monic
polynomial with integer coefficients. We can rewrite (7.1) as a matrix equation

U v = zv

where v = (v1 , · · · , vd )T . Hence z is an eigenvalue of U , so g(z) = det(zI − U ) = 0

and z is an algebraic integer. 

Proof of Theorem 7.3. Let α, β ∈ A; we have to show that αβ ∈ A and α + β ∈ A.

Choose integers a0 , · · · , am−1 and b0 , · · · , bn−1 such that

αm + am−1 αm−1 + · · · + a1 α + a0 = 0 and β n + bm−1 β m−1 + · · · + b1 β + b0 = 0

and consider the abelian subgroup M of C generated by the set {αi β j : 0 6 i <
m, 0 6 j < n}. The monic polynomial equations satisfied by α and β ensure that

αM ⊆ M and βM ⊆ M.

Since M is an additive subgroup of C, we deduce that

(αβ)M ⊆ M and (α + β)M ⊆ M.

Hence αβ and α + β ∈ A by Proposition 7.4. 

Next we return to representation theory and extract some algebraic integers from
the complex group ring of a finite group.

Lemma 7.5. Let χ be a character of the finite group G. Then χ(g) is an algebraic
integer for all g ∈ G.

Proof. We know that χ(g) is a sum of n-th roots of unity where n is the order of
g. Since these are all algebraic integers, we can now apply Theorem 7.3. 

Recall the conjugacy class sums from Proposition 3.15.

Lemma 7.6. Let G be a finite group and let C1 , · · · , Cr be the conjugacy classes
in G. Let S be the additive subgroup of CG generated by the conjugacy class sums
c1 , · · · , C
C cr . Then S is a subring of Z(CG).
40

Proof. We know that each C ci is central in CG by Proposition 3.15, so it will be

enough to show that S is stable under multiplication. Fix i, j = 1, · · · , r. Write
! 
r X
X X X
C
ci C
cj = x  y = aijk (z)z
x∈Ci y∈Cj k=1 z∈Ck

for some aijk (z) ∈ C. Now actually aijk (z) = |{(x, y) ∈ Ci × Cj : xy = z}| ∈ N,
which implies that aijk (z) = aijk (z g ) for any z ∈ Ck and any g ∈ G. So aijk :=
aijk (z) does not depend on z. Hence
r
X
C
ci C
cj = ck ∈ S
aijk C for all i, j = 1, · · · , r.
k=1

ci s as an abelian group, we conclude that S ·S ⊆ S. 

Since S is generated by the C

We will now compute the value of the central characters of simple CG-modules
— see Definition 3.7 — on our conjugacy class sums. These values turn out to be
algebraic integers!

Theorem 7.7. Let V be a simple CG-module and let g ∈ G.

G
G acts on V by the scalar |g |χV (g) ∈ C:
(a) The conjugacy class sum gc χV (1)

G·v =
|g G |χV (g)
gc ·v for all v ∈ V.
χV (1)

(b) This scalar is an algebraic integer.

G
Proof. (a) Since V is a simple CG-module and since the conjugacy class sum z := gc
is central in CG, it acts by a scalar zV ∈ C on every simple CG-module by Schur’s
Lemma, Theorem 3.6. Now take the trace of this action to obtain

zV dim V = |g G |χV (g).

Part (a) now follows because dim V = χ(1) by Lemma 5.6(a).

(b) Let ρ : G → GL(V ) be the representation afforded by V . Then ρ extends
to a C-algebra homomorphism ρe : CG → End(V ) by Question 3(a) on Problem
Sheet 3. The restriction of this homomorphism to the centre Z(CG) is the central
character of V , so ρe(Z(CG)) ⊆ C. So ρe(S) is a finitely generated abelian subgroup
of C. But it is also a subring of C because ρe is a ring homomorphism and because
S is a subring of Z(CG) by Lemma 7.6. Hence zV · ρe(S) ⊆ ρe(S). Proposition 7.4
now tells us that zV is an algebraic integer. 

This theorem has the following interesting consequence.

Corollary 7.8. If V is a simple CG-module, then dim V divides |G|.

 41

Proof. By Theorem 5.13, we have hχV , χV i = 1. Letting g1 , · · · , gr be a complete

set of representatives for the conjugacy classes of G. Using Proposition 5.21(a), we
can rewrite this orthogonality relation as follows:
r
X |giG |χV (gi ) |G|
χV (g −1 ) · = .
i=1
χV (1) χV (1)
−1
Now χV (g ) is a sum roots of unity, which are algebraic integers by Example
7.2(a), and the other factor is an algebraic integer by Theorem 7.7(b). Therefore
|G|/χV (1) is an algebraic integer by Theorem 7.3. Since it is also a rational number,
it must be an integer by Example 7.2(d). 

We will now apply our knowledge of the value of the central characters of irre-
ducible complex G-representations on conjugacy class sums to obtain an interesting
group-theoretic application of representation theory.

Theorem 7.9 (Burnside, 1904). Let G be a non-abelian group of order pα q β where

p, q are primes. Then G is not a simple group.

We begin the proof by recalling Sylow theory from Part A Group Theory.

Definition 7.10. Let G be a finite group and let p be a prime. Write |G| = pα m
where p - m. A Sylow p-subgroup of G is a subgroup P of G of order pα .

Sylow’s three theorems are as follows.

Theorem 7.11 (Sylow, 1874). Let G be a finite group.

(a) G contains at least one Sylow p-subgroup.
(b) Any two Sylow p-subgroups are conjugate in G.
(c) The number of Sylow p-subgroups of G is congruent to 1 mod p, and this
number divides m = |G|/pα .

The first step in the proof of Burnside’s Theorem is the following Lemma.

Lemma 7.12. Let G be a group of order pα q β where p, q are distinct primes and
α, β ≥ 1. Let g be a central element of a Sylow p-subgroup P of G. Then |g G | is a
power of q.

Proof. Since P centralises g, we have P 6 CG (g). So [G : CG (g)] divides |G|/|P | =

q β . But this index equals |g G | by Lemma 5.17. 

We will actually prove the following general result about finite simple groups.

Theorem 7.13. Let G be a finite group and suppose that the size of a conjugacy
class of a non-central element g ∈ G is a power of q. Then G is not a simple group.

The proof of this statement requires a little Galois Theory.

42

Lemma 7.14. Let ζ1 , · · · , ζn be roots of unity and let α := ζ1 +···+ζ

n
n
. Suppose
also that α is an algebraic integer. Then either α = 0, or α = ζ1 = · · · = ζn .

Proof (Non-examinable). We may assume that ζ1 , · · · , ζn ∈ Q(ω) for some primi-

tive k-th root of unity ω. Let G be the Galois group of Q(ω) over Q, and consider
the norm of α, which is defined as follows:
Y
a := NormQ(ω)/Q (α) := σ(α).
σ∈G

On the one hand, because each σ(ζj ) is a root of unity, we have |σ(ζj )| = 1 for all
j = 1, · · · , n and all σ ∈ G. Hence |σ(α)| 6 1 for each σ ∈ G, so
Y
|a| = |σ(α)| 6 1.
σ∈G

On the other hand, clearly a is fixed by the action of G, and Galois Theory tells us
that this means that a ∈ Q(ω)G = Q. By our hypothesis, α is an algebraic integer,
so σ(α) is also an algebraic integer for any σ ∈ G. So, a is again an algebraic integer
by Theorem 7.3. Hence a ∈ Z by Example 7.2(d).
These two facts force a ∈ {−1, 0, 1}. If α 6= 0 then |a| = 1, so |ζ1 + · · · + ζn | = n.
By the solution to Question 5(a) on Problem Sheet 3, α = ζ1 = · · · = ζn . 

Proof of Theorem 7.13. The idea will be to examine all of the non-trivial irreducible
representations ρ2 , · · · , ρr of G and show that for at least one of these, ρi : G →
GL(Vi ) say, the linear map ρi (g) is a scalar multiple of the identity: g ∈ ρ−1 ×
i (C ).
Once we have done this, we can consider the following two normal subgroups of G:
ker ρi 6 ρ−1 ×
i (C ).

We know that ker ρi is a proper subgroup of G since ρi is non-trivial. We also

know that ρ−1 ×
i (C ) is non-trivial. So, the only way G could still be a simple group
is if ker ρi is trivial and ρ−1 × ×
i (C ) = G. But then ρi (G) 6 C , and G would be
isomorphic to a subgroup of C× which is abelian — this contradicts our hypothesis
that g ∈ G is non-central.
Let χi be the character of ρi and consider the Column Orthogonality relation
for the pair 1 and g given by Theorem 5.23:
r
X
1+ χi (1)χi (g) = 0.
i=2

If all of degrees χi (1) were divisible by q, then we would deduce from Lemma 7.5
and Theorem 7.3 that −1/q ∈ A, which contradicts Example 7.2(d). The same
reasoning shows that χi (g) 6= 0 and q - χi (1) for at least one index i.
Since |g G | is a power of q, χi (1) is coprime to |g G |. By Bezout’s Lemma we can
find integers a, b such that a|g G | + bχi (1) = 1. Hence
|g G |χi (g) χi (g)
a + b χi (g) =
χi (1) χi (1)
 43

Now |g G |χi (g)/χi (1) ∈ A by Theorem 7.7(b) and χi (g) ∈ A by Lemma 7.5, and
then Theorem 7.3 then forces χi (g)/χi (1) to also be an algebraic integer. Since it is
nonzero, Lemma 7.14 tells us that the eigenvalues of ρi (g) are all equal to some root
of unity ζ. Then ρi (g) = ζ1Vi is a scalar multiple of the identity as claimed. 

The proof of Burnside’s Theorem is now a formality.

Proof of Theorem 7.9. We may assume α, β ≥ 1. By Sylow’s Theorem, Theorem

7.11(a), G has a Sylow p-subgroup P . Since P is a non-trivial p-group (as α ≥ 1),
it must have a non-trivial central element g ∈ Z(P ). If this element is also central
in G then hgi will be a non-trivial proper normal subgroup of G (as β ≥ 1) and
we’re done. Otherwise, g is not central in G and Lemma 7.12 tells us that |g G | is
a power of q. Now apply Theorem 7.13.