Advanced Abstract Algebra

Shivam Nadimpalli

Last updated: November 16, 2019

1
Advanced Abstract Algebra Fall 2018

Contents
0 Hello! 3

I Topics in Group Theory 4

1 Groups Actions 5
1.1 Definition of a group action . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Orbits and stabilizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Orbit-Stabilizer Theorem and Burnside’s Lemma . . . . . . . . . . . . . . . . 8
1.4 Sylow Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 The Transfer Homomorphism & Applications 11

2.1 The transfer function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Properties of the transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Making the transfer prettier . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 The transfer homomorphism in action . . . . . . . . . . . . . . . . . . . . . . 13

3 Solvable and Nilpotent Groups 19

3.1 Revisiting linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Solvable and Nilpotent Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Two special subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 An application to finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Free Groups 28
4.1 Construction of free groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Cayley graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3 Fun with free groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

II Representation Theory of Finite Groups 34

5 Basic Notions 35
5.1 What is a representation? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Maps between representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.3 Maschke’s Theorem and Schur’s Lemma . . . . . . . . . . . . . . . . . . . . . 37

6 Characters and Burnside’s Theorems 41

6.1 What is a character? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.2 Orthogonality relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3 Algebraic Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4 Two Theorems of Burnside . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

7 More Character Theory 50

7.1 Induced characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.2 Frobenius groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.3 The Second Orthogonality Relation . . . . . . . . . . . . . . . . . . . . . . . . 55
7.4 More Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2
Advanced Abstract Algebra Fall 2018

0 Hello!
These notes are based on the Advanced Abstract Algebra course taught by Dr. Péter
Hermann, as a part of the Budapest Semesters in Mathematics, Spring 2018. The course
attempts to give an idea of some basic methods in finite and infinite group theory. All
errors in these notes are my responsibility; please email any comments or typos to nadim-
[email protected].

Notation

Below we keep a running track of notation used in class. Occasionally, especially in the
section on representation theory, we might write g φ instead of φ(g).
O(g) The order of a group element g
G(x) Orbit of x under action of G
Gx Stabilizer of x under action of G
CG (g) Centralizer of g in G
Z(G) Center of G
Ca Conjugacy class of a in G
NG (H) Normalizer of H ≤ G in G
Sylp (G) Sylow p-subgroups of G
τG→A Transfer homomorphism from G to abelian subgroup A of finite index
[a, b] Commutator of a, b = a−1 b−1 ab
[A, B] h[a, b] | a ∈ A, b ∈ Bi
G0 Commutator subgroup of G; equivalently [G, G]
φ(G) Frattini subgroup of G
F (G) Fitting subgroup of G
F (X) Free group generated by X
Γ(G; X) Directed, colored Cayley graph of group G generated by X
Hom(V, W ) {f : V → W | f is a linear map}
End(V ) Equivalent to Hom(V, V )
GL(V ) The group of invertible endomorphisms of V
GLn (C) Invertible n × n matrices with entries from C
χ Will usually stand for the character of a representation
χG Induced characters, see corresponding section for more

3
Part I

Topics in Group Theory

Advanced Abstract Algebra Fall 2018

1 Groups Actions
Group actions are probably the most central topic in group theory: if a group G acts on a set
X, then it captures some information about the symmetries of X. Almost every important
topic that we will talk about in this course will be formulated in the language of group
actions.
As a motivating example, suppose we want to count the number of its symmetries, namely
the number of edge preserving bijections on its vertex set.

Instead of attempting to count the number of valid transformations, it is much easier to

study a group that “acts” on its vertices.
We now give the precise definition of a group action, and introduce related concepts.

1.1 Definition of a group action

Definition 1.1.1 (group action)

Let X 6= ∅ be a set, and G a group. Then G acts on X if:
1. (∀α ∈ X)(∀c ∈ G) : c(α) ∈ X
2. (∀α ∈ X)(∀c, d ∈ G) : cd(α) = c(d(α))
3. (∀α ∈ X) : e(α) = α

Example 1.1.2
Consider the trivial action of a group G on X: (∀α ∈ G)(∀x ∈ X) : α(x) = x.

We now look at an alternative definition of group actions. In most situations, we will be

using the one above, but the latter sheds light on the importance of symmetric groups, and
occasionally comes in handy.

Definition 1.1.3 (group action)

Group G acts on a set X 6= ∅ if there exists a homomorphism G → Sym(X), where
Sym(X) is the group of permutations of the set X. We define for each g ∈ G a
function as follows: fg : X → X; β 7→ g(β).

We immediately become informal and say that “G acts on X” where G is understood to be

a group, and X a nonempty set.
Given a set of n elements, what’s the most natural group you can think of that acts on it?
Chances are you are thinking of the symmetric group on n elements, namely Sn . In order

5
Advanced Abstract Algebra Fall 2018

to capture how far away a group action is from a permutation, we introduce the following
notion:

Definition 1.1.4 (faithful action)

Let G act on X. This action is faithful if (∀e 6= a ∈ G)(∃β ∈ X) : a(β) 6= β.

Put in words, a group action is said to be faithful if for every element of the group, not all
the elements of the underlying set are fixed points. Thus, if an action is faithful, then every
group element moves at least one element of the set. All this will make more sense after the
next proposition:

Proposition 1.1.5
Consider G acting on a set X. Then fg is a bijection.

Proof. Injectivity of fg is easy to show. To show surjectivity, for γ ∈ X, note that

fg (g −1 (γ)) = γ. We conclude that fg is a bijection from X to X.
Thus, you can think of group actions as shadows of the permutation groups on the sets.
This leads us to the next proposition:

Proposition 1.1.6
The group action of G on X is faithful if and only if the corresponding homomorphism
from G → Sym(X) is injective.

The proof is omitted as it should be easy to see; if it isn’t, then reread the earlier section
until it is.

1.2 Orbits and stabilizers

Associated to a group action are the fundamental concepts of orbits and stabilizers.

Definition 1.2.1 (orbits and stabilizers)

Let G act on X. Suppose g ∈ G and x ∈ X. Then:
1. Gx := {c ∈ G | c(x) = x} is the stabilizer of x in G.
2. G(x) := {g(x) | g ∈ G} is the orbit of x in G. We will also occasionally write
Ox for the orbit of x.
3. If G(x) = X, then the action is said to be transitive.
4. If Gx = G, then x is said to be a fixed point of the action.

So the orbit of an element is the set of all the elements it can “reach” via the action of some
element of the group; if it can reach any element of the set, then the action is said to be
transitive. The stabilizer of the element, which should be easy to remember, is the set of
group elements that fix (stabilize) the element; if a set element is unmoved by all elements
of the group, then it is said to be a fixed point of the action.

6
Advanced Abstract Algebra Fall 2018

Proposition 1.2.2 (stabilizer forms subgroup)

Let G act on X. Then for all x ∈ X, Gx ≤ G.

Proposition 1.2.3 (orbits partition the set)

Let G act on X. The orbits of this action partition X.

Important: Keep in mind that orbits do not necessarily have the same size!
It follows from Proposition 2.2.3 that the orbits of two elements are either disjoint or the
same, and that you can write |X| = |O1 | + . . . + |Ok |. We might refer to this in the future
as the “orbit decomposition” of X.

Some fundamental group actions

We now look at a group G acting on itself and its subgroups, cosets, etc.

Permutations of group elements

Let G act on itself, and define for all g, h ∈ G, g(h) = gh. This action is called the
permutation action of G on itself.
Exercise. Check that this is a valid group action! Is this action faithful?

Permutations of a subgroup

Given a subgroup H ≤ G, denote the set of its left cosets by X. We can define a permutation
action on the X. For every aH ∈ X, g ∈ G, define g(aH) = gaH. This is well-defined
because we’re only working with coset representatives.
What’s the kernel of this action? Indeed, ker = {c ∈ G | ∀a ∈ G, c(aH) = aH}. However,
for a given a ∈ G, we have:

c(aH) = aH ⇐⇒ a−1 ca ∈ H ⇐⇒ ca ∈ aH ⇐⇒ c ∈ aHa−1

aHa−1 , and this is precisely the kernel of the above action.

T
And so, c ∈ a∈G

Conjugation on an element

Let G act on itself, and define for all g, h ∈ G, g(h) = ghg −1 . This action is called the
conjugation action of G on itself. This leads us naturally to the next definition:

Definition 1.2.4 (centralizer of an element)

The centralizer of g ∈ G, written CG (g) is the stabilizer of g under the conjugation
action.

7
Advanced Abstract Algebra Fall 2018

Since stabilizers are subgroups, it follows that CG (g) ≤ G.

We have a special name for the intersection of all centralizers of group elements: it’s called
the center of the group (written Z(G)), and is (in words) the subgroup consisting of elements
that commute with all other group elements.

Proposition 1.2.5 (center of a group)

T
Define the center of a group Z(G) = g∈G CG (g). We have the following:
1. Z(G) E G
2. If G/Z(G) is cyclic, then G is abelian

What are the orbits of this action? They are precisely the conjugacy classes of G! For
a ∈ G, we might denote the conjugacy classes of a in G by Ca . Again, important to note
that the sizes of the conjugacy classes need not be the same; we will see later, however, that
they must divide the size of the group.
It is worthwhile to note that for any nontrivial group G, the conjugacy class of e = {e}.
Thus, any nontrivial group has at least two conjugacy classes.

Conjugation on a subgroup

We can similarly define conjugation on a subgroup: for H ≤ G, g(H) = gHg −1 . The

stabilizer of H ≤ G under this action also has a special name:

Definition 1.2.6 (normalizer of a subgroup)

Given H ≤ G, NG (H) = {g ∈ G | gHg −1 = H}; alternatively, NG (H) is precisely
the stabilizer of H under the conjugation action.

The following proposition follows naturally:

Proposition 1.2.7 (normalizer of normal subgroup is whole group)

H E G ⇐⇒ NG (H) = G.

1.3 Orbit-Stabilizer Theorem and Burnside’s Lemma

Theorem 1.3.1: Orbit-Stabilizer

Let G act on X, β ∈ X. Then |G(β)| = |G : Gβ |. In particular, the stabilizers of
elements in the same orbit have the same size.
Proof. We know that Gβ ≤ G, and so can consider the left cosets of Gβ . Note that each
coset simply specifies an element of G(β). Thus, we have a bijection between G(β) and
the left cosets of Gβ .
We can finally return to our original motivating example:

8
Advanced Abstract Algebra Fall 2018

Example 1.3.2 (symmetries of the cube)

Without loss of generality, take G ≤ S8 , and consider its action on the vertices of
the cube above. Note that G(1) = {1, 2, . . . , 8}. From the Orbit-Stabilizer theorem,
we have |G(1)| = |G : G1 | =⇒ |G| = 8|G1 |.

As G1 ≤ G, we can consider the action of G1 on the vertices of the cube, i.e.

we are looking at those transformations of the cube that fix vertex 1. Under
this action, the orbit of vertex 2 is given by: G1 (2) = {2, 4, 5}. This is because
the connectivity of (1, 2) must be preserved. Again, by Orbit-Stabilizer, we have:
|(G1 )2 | =: |G1,2 | =⇒ |G1 | = 3|G1,2 |.

Now, consider the action of G1,2 on the cube: G1,2 (3) = {3, 6}, and so, by Orbit-
Stabilizer, |G1,2 = 2|G1,2,3 |. However, |G1,2,3 | = 1, as fixing those three vertices fixes
the rest of the cube. Thus, |G| = 8 × 3 × 2 = 48.

Now, what if one wants to count the number of orbits of a group action? Indeed, the
lemma-that-was-not-Burnside’s (Cauchy got there first) allows us to do so easily.

Lemma 1.3.3 (Burnside’s Lemma)

Assume G acts on X, G and X both finite. Define f : G → N as follows:

(∀ g ∈ G)f (g) = |{β ∈ X | g(β) = β}|

1
P
Then the number of distinct orbits of X under G is |G| g∈G f (g).

Proof. The proof uses a double-counting argument. We consider the total number of
pairs of the form (β, g) where β ∈ X, g ∈ G, and g(β) = β.
For a given g ∈ G, he number of such pairs is equal to f (g), as each pair corresponds to
a point that is fixed by g. However, for each β ∈ X, the number of such pairs is given
by |Gβ |. Thus we have:
 
X X X 1 X X 1
f (g) = |Gβ | = |G| = |G|  
|G(β)| |T |
g∈G β∈X β∈X T orbit β∈T

where the inner sum in the final expression is equal to 1. The desired result then
follows.

1.4 Sylow Theorems

Is the converse of Lagrange’s Theorem true? Namely, if the d||G|, then does there exist a
subgroup of order d in G? This turns out to be false: the simplest example of this is the

9
Advanced Abstract Algebra Fall 2018

group A4 , of order 12, which has no subgroup of order 6.

However, Sylow discovered that the converse is true when d is a prime power; he also
discovered special relations between subgroups whose order was the highest prime power
dividing the order of the group. All this is stated below:

Theorem 1.4.1: Sylow Theorems

Let G be a finite group, p a prime number such that pk divides |G|. Then:
S1. There exists a subgroup of order pk in G. Moreover, the number of such sub-
groups is congruent to 1(mod p).
S2. Let pn be the highest power of p that divides |G|. Subgroups of order pn are called
Sylow p-subgroups of G (Sylp (G) for short). Then all Sylow p-subgroups are
conjugates of each other, and are hence isomorphic.
S3. Let R be a Sylow p-subgroup of G, and H ≤ G such that |H| = pk . Then there
exists d ∈ G such that H ≤ dRd−1 .

I was lazy and did not take the proof down. Several proofs can be found online or in the
textbook for the course (“A Course in the Theory of Groups” by Robinson). Plenty of group
actions used in the proofs.

Corollary 1.4.2 (S1∗ )

Suppose |G| = pk m such that pk+1 does not divide |G|. Then |Sylp (G)| | m.

Proof. Consider the action of conjugation by elements of G on Sylp (G). Then from S2,
all the Sylp (G) subgroups are conjugates of each other, and so there is only one orbit. It
then follows that the size of this orbit, namely |Sylp (G)| must divide |G| = pk m, but as
|Sylp (G)| ≡ 1 mod p, the number |Sylp (G)| and p are co-prime, so |Sylp (G)| | m.

10
Advanced Abstract Algebra Fall 2018

2 The Transfer Homomorphism & Applications

In this section, we create a monster. . .

2.1 The transfer function

Definition 2.1.1 (transfer homomorphism)

Let GSbe a group, A ≤ G such that |G : A| is finite. Suppose A is abelian. Let
G = i∈I ci A. We can consider the action of G on the index set I of the coset
representatives as follows: for x ∈ G, x · ci A = cx(i) A where xci = cx(i) ax,i for some
ax,i ∈ A.
Then the transfer function is given by:
Y
τ (x) = ax,i
i∈I

What happens if, in the above definition, |G : A| is not finite? Then there are infinitely
many cosets of A in G, and hence, τ (x) is an infinite product, which is not a nice thing.
What happens if A is not abelian? Then clearly the product is not well defined.
It turns out that the transfer plays an important role in the study of unsolvable groups and
aids in the characterization of groups in which all Sylow p-subgroups are cyclic. When we
get to Riesz representations, we will probably see some nifty applications of the transfer.

Note to self: Stare at the above definition until it makes sense!

2.2 Properties of the transfer

In what follows, we take A ≤ G to be an abelian group with finite index. WeSmight also
sometimes refer to the set of coset representatives {c1 , . . . , ck } such that G = i ci A as a
(left) traversal for A in G; we will refer to the index set {1, . . . , k} by I.

Proposition 2.2.1 (transfer is a homomorphism)

The transfer function τ : G → A is a homomorphism.

Proof. Let x, y ∈ G, and consider the action of xy on the index set of coset representatives
I as described in Defiition 2.1.1. We thus have for i ∈ I:

xy(ci ) = x(y(ci )) = x(cy(i) ay,i ) = (xcy(i) )ay,i = cx(y(i)) (ax,y(i) ay,i )

However, we also have:

xy(ci ) = cxy(i) axy,i = cx(y(i)) axy,i

And so we can write:

axy,i = ax,y(i) ay,i

11
Advanced Abstract Algebra Fall 2018

It is then easy to see that τ (xy) = τ (x)τ (y), because A is abelian and as y(i) must range
over all of the index set (because left multiplication is a bijection).

Proposition 2.2.2 (transfer doesn’t depend on traversal)

Let A ≤ G as described in Definition 2.1.1. Then τ does not depend on the choice of
coset representatives of A in G.

Proof. Let {di } be an alternative set of coset representatives for A in G. For every i ∈ I,
we thus have di A = ci A such that di = ci αi for some αi ∈ A. Consider the action of
x ∈ G on some i in the index set. We thus have:

xdi = dx(i) a∗x,i = cx(i) αx(i) a∗x,i

But we also have:

xdi = xci αi = cx(i) ax,i αi
And so we can write:
−1
cx(i) αx(i) a∗x,i = cx(i) ax,i αi =⇒ a∗x,i = αx(i) ax,i αi

Again, as A isQ abelian, Q

and x(i) must also range over the set of all coset representa-
tives, we have a∗x,i = ax,i and so the transfer is independent of the choice of coset
representatives.

2.3 Making the transfer prettier

We will now try to find a nice form for the transfer for arbitrary elements of G. Why are
we doing this? Because, as we will see shortly, with a clever choice of coset representatives,
computing the transfer will not be as difficult as it appears.
S
Let {c1 A, . . . , cn A} be the cosets of A in G such that G = i ci A, where A ≤ G is taken to
be an abelian group. Consider the action of G on its cosets by left multiplication. Let x ∈ G
be an arbitrary group element. Then, note that under the action of x, we can decompose
the set of left cosets into disjoint cycles.
Let {xai A, x2 ai A, . . . , xni −1 ai A} be one such cycle, where ni is the first power of x in
Pk
ai A. It follows, then, that if we have k such disjoint cycles, then |G : H| = n = i=1 ni .
We can use this fact to obtain a very nice traversal to compute the image of x under τ :
{ai xj | 1 ≤ i ≤ k, 1 ≤ j ≤ ni }.
Why is this a nice traversal? Consider the “contribution” to τ (x) made by the second cycle,
namely {a2 A, xa2 A, . . . , xn2 −1 a2 A}. We have:

x(a2 A) = xa2 A =⇒ xa2 = xa2 (e)

..
.
x(xn2 −1 a2 A) = a2 A =⇒ x(xn2 −1 a2 ) = a2 (a−1 n2
2 x a2 )

And so we have a contribution to τ (x) equal to e × e × . . . × (a−1 n2 −1 n2

2 x a2 ) = a2 x a2 .

12
Advanced Abstract Algebra Fall 2018

By extending the above argument to the k cycles, we can write:

k
Y
τ (x) = a−1 ni
i x ai
i=1

Pk
where i=1 = n, and for each i = 1 : k, ni is the smallest positive integer such that
−1 ni
ai x ai ∈ A. It is important to note that in the above construction, we are presupposing
that we know the disjoint cycles that the cosets are decomposed into.

Example 2.3.1 (subgroup contained in center)

Let A ≤ G, A abelian with |G : A| = n < ∞. We continue with our notation earlier
from this section.

Note that if x ∈ Z(G), we have the following nice form for τ (x):
Y Y P
τ (x) = (c−1 ni
i x ci ) = x ni = x i ni = x n
i i

Now, if A ≤ Z(G), then note that elements of the form c−1 ni

i x ci are always elements
ni
of A, and as A ∈ Z(G), we can rewrite each expression as x , from which it follows
that τ (x) = xn for all x ∈ G.

2.4 The transfer homomorphism in action

This rest of this section is entirely devoted to applications of the transfer.

2.4.1 Burnside’s transfer theorem

Lemma 2.4.1 (normalizer controls fusion)

Let G be a finite group, p a prime, and R ∈ Sylp (G). Assume R abelian. Let a, b ∈ R
such that b = y −1 ay for some y ∈ G. Then there exists c ∈ NG (R) such that
c−1 ac = b.
Proof. As R is abelian, R ≤ CG (a) and also R ≤ CG (b). Thus, we have:

y −1 Ry ≤ y −1 CG (a)y = CG (y −1 ay) = CG (b)

where the equality above is easy to prove (but is important). So both R and y −1 Ry
are subgroups of CG (b), and so R ∈ Sylp (CG (b)) and y −1 Ry ∈ Sylp (CG (b)). By Sylow’s
Theorem, there exists d ∈ CG (b) such that d−1 Rd = y −1 Ry, and so:

R = (dy −1 )R(yd−1 ) =⇒ yd−1 ∈ NG (R) =⇒ (yd−1 )−1 a(yd−1 ) = dbd−1 = b

13
Advanced Abstract Algebra Fall 2018

Theorem 2.4.2: Burnside’s transfer theorem

Suppose G is a finite group, p a prime, and R ∈ Sylp (G). Assume R ≤ Z(NG (R)).
Then there exists N / G such that |N | = |G : R|.

N in the above theorem is called a normal p-complement in G, and so the above theorem
is also sometimes called “Burnside’s normal p-complement theorem”. Before proving the
above theorem, we make a few remarks.
First, what does the seemingly-grotesque condition R ≤ Z(NG (R)) mean? Indeed, if the
condition holds, then for any g ∈ G, either g −1 Rg 6= R, or for every r ∈ R, g −1 rg = r. So
in a way, it is an all-or-nothing condition.
Finally, can we say anything about G/N ? Indeed, we have N ≤ G =⇒ RN ≤ G. From
HW1 Problem 1(b), and as gcd(|R|, |N |) = 1, we can write:

|R||N | |G|
|RN | = =
|R ∩ N | 1

which means that RN = G. And so, by the third isomorphism theorem we have RN/N ∼
=
R/R ∩ N , which in turn implies G/N ∼
= R.
Now that we have gotten all this out of the way, let’s return to the proof of Burnside’s
transfer theorem:
Proof. Let r ∈ R. We have τ (r) = j yj−1 rnj yj where j nj = |G : R|. Note that both
Q P

rnj and yj−1 rnj yj are elements of R, and so, by Lemma 2.4.1, we have the following:

(∀j)(∃ cj ∈ NG (R)) : c−1 nj −1 nj

j r cj = yj r yj

However, as R ≤ Z(NG (R)), rnj commutes with cj , and so

rnj = c−1 nj −1 nj
j r cj = yj r yj

Thus we have:
Y
τ (r) = rnj = r|G:R| . =⇒ Imτ ⊇ {r|G:R| | r ∈ R} = R
j

with the last equality following from HW2 Problem 5, as |G : R| is co-prime to p.

Now, when do the conditions for the above theorem hold? As an example, it does when G
is finite and p is the smallest prime divisor of |G|, and all the Sylp (G) are cyclic. Is this
too much to ask? No, because, if you remember, if the order of a group is prime, then it is
cyclic. We make all this rigorous in the next proposition.

Proposition 2.4.3 (Burnside’s transfer theorem isn’t unreasonable)

Let G be a finite group, and p be the smallest prime dividing |G|, and all Sylp (G) are
cyclic. Then, if R ∈ Sylp (G), then R ≤ Z(NG (R)).

Proof. Consider α : NG (R) → Aut(R), x 7→ conjugation by x on R. Then, note that

R ≤ Z(NG (R)) if and only if α is a trivial homomorphism.

14
Advanced Abstract Algebra Fall 2018

Now, for all prime q such that q | |NG (R)|, let Pq ∈ Sylq (NG (R)). We make the following
claim:
Claim. α Pq is trivial.
Recall that |R| = pn for some n. We have two possible cases:
Case 1: p 6= q. Then, as the number of automorphisms is equal to the number of
generators, we have |Aut(R)| = pn − pn−1 = pn−1 (p − 1).
Now, as R finite, and R = hci for some c ∈ R, we have the following series of deductions:

R = hci ⇐⇒ R = hck i
⇐⇒ O(c) = O(ck ) = pn
O(x)
⇐⇒ pn =
gcd(O(c), k)
n
⇐⇒ gcd(p , k) = 1
⇐⇒ (pn − p) coprime to p

Now, recall that is gcd(|A|, |B|) = 1, then any homomorphism from A to B must be
trivial, and so, it follows that α Pq is trivial.
Case 2: p = q. Then note that Pq ∈ Sylp (G), and so Pq is cyclic, and hence, abelian.
It follows that the automorphism is trivial.
αk
Now, if A is a finite group such that |A| = pα 1 . . . pk , then hRi ∈ Sylpi (A)i = A. As the
1

restriction of the homomorphism α to each of the Sylow q-subgroups of NG (R) is trivial,

we can conclude that its restriction to NG (R) is trivial, from which the desired result
follows.

2.4.2 An application to infinite groups

We introduce several new notions in this subsection.

Definition 2.4.4 (commutator)

Let a, b ∈ G. Then [a, b] = a−1 b−1 ab is the commutator of a, b.

What’s the first thing we can say about the commutator? Indeed, if the commutator of two
elements is trivial, then they commute! Namely, [a, b] = e ⇐⇒ ab = ba.

Definition 2.4.5 (commutator subgroup)

The subgroup generated by the commutators of elements of G is called the commutator
subgroup, denoted by G0 . Thus, G0 = h[a, b] | a, b ∈ Gi.

Why do we care about this? Well, firstly, it’s what we call a characteristic subgroup, i.e.
is invariant under all automorphisms. Moreover, it plays an important role when we try to
obtain abelian factor groups, because of the next two propositions:

Proposition 2.4.6 (commutator subgroup is normal)

G0 E G.

15
Advanced Abstract Algebra Fall 2018

Proposition 2.4.7 (commutator subgroup ⊆ kernel of abelian quotient)

If N / G then G/N abelian if and only if G0 ⊆ N .

Proof. G/N abelian ⇐⇒ abN = baN ⇐⇒ a−1 b−1 ab ∈ N ⇐⇒ G0 ⊆ N .

We now move on to a somewhat interesting proposition. It’s kind of obvious if G is finite,
but is pretty cool that it holds even when G is infinite.

Proposition 2.4.8
Suppose |G : Z(G)| is finite. Then the commutator subgroup G0 is finitely generated.
Sn
Proof. Z(G) having finite index means that G = i=1 Z(G)ci . Thus, any two elements
a, b ∈ G can be written as a = z1 ci and b = z2 cj for appropriately chosen z1 , z2 , ci , cj .
Thus:
[a, b] = [z1 ci , z2 cj ] = c−1 −1
i cj ci cj

and so there are only finitely many commutator elements. Thus, we can conclude that
the commutator subgroup is finitely generated.
In fact, we can actually prove a stronger version of the above proposition, but first a lemma
is needed:

Lemma 2.4.9
If G is finitely generated, and H ≤ G with |G : H| finite, then H is finitely generated.

Proof. Without loss of generality, assume G = ha1 , . . . , an i where each ai has its inverse
Sl
among the aj -s. Now, as H has finite index in G, we have G = j=1 cj H. Take c1 = e,
and note that we can define ai cj ∈ cai (j) H which means ai cj = cai (j) hi,j .
Claim. H = hhi,j | 1 ≤ i ≤ n, 1 ≤ j ≤ li.
Indeed, let y ∈ H. We then have:

y = ai1 ai2 . . . ait−1 ait c1

= ai1 ai2 . . . ait−1 cait (1) hit ,1
..
.
Y 
= cai1 ai2 ...ait (1) hp,q
Y 
= cy(1) hp,q
Y
= hp,q

and so we’re done.

Proposition 2.4.10
Suppose |G : Z(G)| is finite. Then the commutator subgroup G0 is finite.

Proof. The proof idea is as follows: X is a finite group if N / X and X/N are both finite.
Consider G0 ∩ Z(G) / G0 .

16
Advanced Abstract Algebra Fall 2018

Claim. G0 /(G0 ∩ Z(G)) and G0 ∩ Z(G) are both finite.

Proof. Recall the 3rd iso. theorem: H ≤ A, N / A =⇒ HN/N ∼
= H/(H ∩ N ). And
so we have:

G0 /(G0 ∩ Z(G)) ∼
= G0 Z(G)/Z(G) ≤ G/Z(G) which is finite by assumption

All that is left to show is that G0 ∩ Z(G) is finite.

Indeed, note that it is necessarily abelian (≤ Z(G)), and that a) |G0 : G0 ∩ Z(G)| is
finite, b) G0 is finitely generated. The latter two together imply G0 ∩ Z(G) is finitely
generated.
Now, note that if A is a finitely generated abelian group, then A 3 a − α1k1 . . . αm km

for some set of generators {αi }. Then, A is finite if and only if O(αi ) finite for every
i.
Claim. For z ∈ G0 ∩ Z(G), O(z) < ∞.
Proof. Consider the transfer τ : G → Z(G). Let z ∈ G0 ∩ Z(G). Then τ (z) ∈
Imτ ∼= G/ ker τ . However, as Imτ ≤ Z(G) which is abelian, it follows that G/ ker τ
is abelian. It follows then that ker τ ⊇ G0 =⇒ τ (z) = e.
However, from Example 2.3.1, z ∈ Z(G) =⇒ τ (z) = z |G:Z(G)| . However, as
|G : Z(G)| is finite, we have z |G:Z(G)| = e =⇒ O(z) is finite.

And so, we can conclude that G0 ∩ Z(G) is finite. Note that we used Lemma 2.4.9 in
the above argument.
Now as G0 /(G0 ∩ Z(G)) and (G0 ∩ Z(G)) are both finite, it follows directly that G0 is
finite.
Recall the definition of a torsion-free group: G is said to be torsion-free if for all e 6= g ∈ G,
O(g) = ∞.

Corollary 2.4.11
Let G be a torsion-free group. Suppose C ≤ G such that C is cyclic and |G : C| is
finite (which means that C is necessarily infinite). Then G is cyclic.

Proof. First, it is easy to see that G is finitely generated. If C = hti, then G =

S|G:C|
i=1 Cxi =⇒ G = hx1 , . . . , x|G:C| , ti. Now if G were abelian, then this, together
with the earlier fact that G is finitely generated, would easily imply that G is isomorphic
to Z.
Claim. G is abelian.
Proof. We show that G is abelian by showing that |G : Z(G)| is finite. Let the index
of C in G be n, and let {x1 , . . . , xn } be the set of coset representatives of C in G
whose union is G. Without loss of generality, take x1 = e. Define C2 = C ∩ hx2 i. For
each i = 3 : n, set Ci := Ci−1 ∩ hxi i ≤ C.
Now, note that for each i, Ci 6= 1. Suppose that this were the case. Then note
that Cxi , Cx2i , . . . are all distinct, which is a contradiction. Thus, we can write that
|C : Ci | is finite for each i.
By iterating the above argument, we can show that Cn has finite index in C, i.e. has
finite index in G. However, Cn ≤ Z(G) and so Z(G) has finite index in G.

17
Advanced Abstract Algebra Fall 2018

The desired result will follow from the next proposition.

Proposition 2.4.12
If A is a finitely generated abelian group, then A ∼
= B ⊕ Z+ ⊕ . . . ⊕ Z+ where B is a
finite group.

We first make the following observation: F = hx1 i ⊕ . . . ⊕ hxn i where hxi i is infinite for all
i, is free in the category of abelian groups Ab. That is, if f : {x1 , . . . , xn } → G where G
abelian, then there exists a unique homomorphism ψ such that ψ(xj ) = f (xj ).
1. If A is abelian, and  : A  F a surjective homomorphism, then A = ker  ⊕ A1 where
A1 ∼= F.
Proof. Note that for every xj , there exists aj ∈ A such that (aj ) = xj ; let ψ : F → A
be determined by ψ(xj ) = aj . It’s easy to see that  ◦ ψ will give you the identity
homomorphism.
Let Ai = Im (ψ). It is obvious that ker  ∩ Im (ψ) = e = 0. It’s also straightforward
to show that any element can be written as a unique combination of elements from
Im (ψ) and ker , and so we’re done.
2. If our finitely generated abelian group is torsion-free, then we don’t have to worry
about the direct component B, i.e. it disappears from expression in Proposition 2.4.12.
Proof. We proceed by induction on the number of generators. Suppose A = ha1 , . . . , an i.
When n = 1, life is good. Suppose true for n − 1 generators, then simply quotient out
by han i. This might not be good, as the quotient might not be torsion-free.
Instead, we do the following: let C = {b ∈ A | ∃t 6= 0 : tb ∈ han i} which is the radical of
the subgroup (you may have seen something similar in commutative algebra/algebraic
number theory). You can check that A/C is torsion-free and is generated by n − 1
elements. Then, by the inductive hypothesis, A/C ∼ = Z+ ⊕ . . . ⊕ Z+ =: F .
Consider the natural surjection η : A → F and then by the previous proposition, we
have A = K ⊕ L where L ∼ = F . We get K ∼ = A/L which is finitely generated. We also
∼
have K = C. The rest of it is to show that K is cyclic. Since C is finitely generated
there exists some positive integer s such that C = {b ∈ A | sb ∈ han i}. Therefore we
have a homomorphism ν : C → han i by ν(c) = sc. As C is torsion-free ν is injective,
thus C is isomorphic to the image of ν, a subgroup of the cyclic group han i.

We can finally return to the general case. Suppose A is finitely generated and abelian. Then
we define B to be the torsion subgroup of A, namely the set of elements of finite order. If
you mod out by B, then it’s easy to check that A/B will be torsion free.
So all that remains to be shown is that B is finite. We have A = A1 ⊕ B where A1 ∼
= A/B
is torsion free and B ∼
= A/A1 is finitely generated. Thus B is finite.

18
Advanced Abstract Algebra Fall 2018

3 Solvable and Nilpotent Groups

We start this section by doing some matrix calculations and linear algebra review. We do
this because later we will try to generalize the argument below to solvable and nilpotent
groups (to be defined shortly). As you will see in §3.1, working with transformations as
functions is usually much cleaner than working directly with matrices.
As a bit more of a teaser, solvable and nilpotent groups are closest in spirit to abelian
groups. In a way, they can be viewed as a kind of upper bound of non-abelianness; abelian
groups sit right between nilpotent groups and solvable groups.

3.1 Revisiting linear algebra

Recall the commutator from §2.4.2. Consider the group of upper semitriangular matri-
ces of size 2 over the same field:
  
a c
G= | a, b, c ∈ F, ab 6= 0
0 b

We make the following two observations:

 −1  
a c 1/a −c/ab
=
0 b 0 1/b

and:  
1 z
A−1 B −1 AB = for some z ∈ F
0 1
where A−1 B −1 AB is the commutator of two matrices A, B. We say that the commutators
of two elements of this group are upper unitriangular matrix. However, two upper
unitriangular matrices commute, as we have:
    
1 z1 1 z2 1 z1 + z2
=
0 1 0 1 0 1

from which it follows that G0 , namely the commutator subgroup, is abelian.

Now, we could think of the above matrices as members of Hom(V ), where V is a 2-
dimensional vector space over the field F. More specifically, we can think of a matrix
from the above discussion as a linear transformation f ∈ GL(2). Next, suppose {b1 , b2 } is a
basis of V ; write Vn = Span(bn ). Then note that V = V1 ⊕ V2 .
We can now look at upper semitriangular matrices in a new light: they correspond to
precisely those f ∈ GL(2) s.t. f (V1 ) ⊆ V1 , as they only scale along the first entry of
any vector. And so, we say that V1 is f -invariant, where f is a linear transformation
corresponding to an upper semitriangular matrix.
Now, if f ∈ G0 , then x = v1 +v2 for v1 ∈ V1 , v2 ∈ V2 , and so f (x) = f (v1 +v2 ) = f (v1 )+f (v2 ).
However, as f (v1 ) = v1 (since we’re working with upper unitriangular matrices), and as
f (v2 ) = (w1 , v2 ) where w1 = v2 z, we have f (x) = v1 + (w1 + v2 ) = x + w1 , and so each
vector x is mapped to the sum of itself and some other vector.

19
Advanced Abstract Algebra Fall 2018

3.2 Solvable and Nilpotent Groups

Solvable groups, as you may have guessed, have to do with the solvability of polynomials.
Indeed, if you have studied any Galois theory, then you may know that for an equation to
be solvable its Galois group must be solvable.

Definition 3.2.1 (solvable group)

G is solvable if G(1) := G0 , G(2) := (G(1) )0 , . . . terminates at 1, i.e. there exists n
such that G(n) = 1.

When will the above sequence terminate? Recall that G ≥ G(1) ≥ G(2) ≥ . . ., and so if G is
finite, then it must terminate. The above series is sometimes called the derived series for
G.

Proposition 3.2.2 (subgroup of solvable group is solvable)

If G is a solvable group, H ≤ G =⇒ H is solvable.

Proof. Look at H ≥ H (1) ≥ H (2) ≥ . . . Note that H (k) ≤ G(k) , and so, as G is solvable,
there must exist some n ∈ N such that H (n) = 1.

Proposition 3.2.3 (quotient of solvable group is solvable)

If G is a solvable group, N / G =⇒ G/N is solvable.

Proof. We claim that (G/N )(k) = G(k) N/N . Indeed, we have:

(G/N )0 = h[Na , Nb ] | Na , Nb ∈ G/N i

= h(N a)−1 (N b)−1 N aN b | a, b ∈ Gi
= hN a−1 b−1 ab | a, b ∈ Gi
= hN [a, b] | a, b ∈ Gi
= N G0 /N = G0 N/N

We can repeat this argument for higher values of k, from which the desired result follows.

We can actually obtain a somewhat stronger result:

Proposition 3.2.4
Let N / G. Then G is solvable if and only if both N and G/N are solvable.

Proof. The forward direction follows from Propositions 3.2.2 and 3.2.3. As for the reverse
direction, we know that G/N is solvable, and so there exists k ∈ N such that (G/N )(k) =
N/N . Thus we can write N/N = G(k) N/N =⇒ G(k) ≤ N .
Now, as N is solvable, there exists l ∈ N such that N (l) = 1, and so (G(k) )(l) = G(k+l) = 1.
We conclude that G is solvable.
We conclude this subsection with some examples of solvable groups.

20
Advanced Abstract Algebra Fall 2018

Example 3.2.5 (examples of solvable groups)

1. Abelian groups
2. Dn /hai where Dn is the dihedreal group of an n-gon, and hai is the rotation
subgroup.
3. Sn solvable if and only if n ≤ 4. (Why? Because Sn solvable iff An solvable.
A1 , A2 , A3 clearly solvable, for A4 you can use the Sylow Theorems!)
4. All finite groups of prime power order (Why? Because they have nontrivial
center)
5. The group of upper unitriangular matrices is solvable.

Just as solvable groups were defined as those with terminating derived series, nilpotent
groups are defined as those with terminating “lower central series”.

Definition 3.2.6 (nilpotent group)

G is called nilpotent if G = γ1 (G) ≥ γ2 (G) ≥ . . . terminates at 1, where γ1 (G) =
G, γ2 (G) = G0 , and γn+1 (G) = [G, γn (G)] with [A, B] denoting the subgroup of the
form h[a, b] | a ∈ A, b ∈ Bi.

We might occasionally write γn instead of γn (G) when G is understood. Some facts that’ll
come in handy later:
1. G0 = [G, G]
2. [A, B] / hA, Bi
3. For N ≤ G, N / G ⇐⇒ [G, N ] ⊆ N

Proposition 3.2.7 (properties of lower central series)

For all n:
1. γn / G
2. γn+1 ≤ γn
3. γn ⊇ G(n)

Corollary 3.2.8 (nilpotent implies solvable)

If G is nilpotent, then G is solvable.

Proof. This follows from Proposition 3.3.2.

Example 3.2.9
The upper unitriangular matrices are nilpotent.

Proposition 3.2.10
Let G be nilpotent, H < G. Then H  NG (H).

21
Advanced Abstract Algebra Fall 2018

Proof. As G is nilpotent, we have G ≥ γ2 ≥ . . . ≥ γn = 1. As H is a proper subgroup,

there must exist a unique j such that γj+1 ⊂ H but γj * H. Thus, there exists some
x ∈ γj /H. Now, from the definition of γj+1 , it follows that for all y ∈ H, [x, y] ∈ γj+1 ⊂
H, i.e. x−1 y −1 xy ∈ H. As y ∈ H, it follows that x−1 y −1 x ∈ H, and so x ∈ NG (H).
We conclude that H  NG (H).
Recall that Ca denotes the conjugacy class of a in G. As the conjugacy classes are orbits
of elements under the conjugation action (see §1.3.3 for a review), they partition our group
G, and moreover, the order of each conjugacy class must divide the order of G. All this will
come in handy as we study group of prime power order.

Proposition 3.2.11 (groups of prime power order are nilpotent)

Let p be a prime, and G be a group such that |G| = pk for some k. Then G is
nilpotent.

Proof. First, we make the following claim:

Claim. |Z(G)| > 1.
Proof. If G is abelian, then Z(G) = G and we’re done. If G is not abelian, then look
at the conjugacy classes of G:
n
X
|G| = |Ci |
i=1

An element is in the center if and only if its conjugacy class is a singleton (consisting
of only the element itself). Now, Ce = {e}, and so, the as |G| is divisible by p, there
must be at least (p − 1) singleton conjugacy classes for the above expression to be
valid. We conclude that the center is nontrivial.
We then proceed using induction on |G|. Let G = G/Z(G). As Z(G) is nontrivial, we
have |G| < pk and so our inductive hypothesis holds, i.e. G is nilpotent.
Thus, there exists n such that γn (G) = 1 =⇒ γn (G)Z(G)/Z(G) = 1. Finally:

γn (G)Z(G) ⊆ Z(G) =⇒ γn (G) ⊆ Z(G) =⇒ [G, γn+1 (G)] = 1

In the above argument, we used a simple property of commutator of quotient groups;

make sure to prove it yourself later.

Lemma 3.2.12 (product of nilpotent is nilpotent)

If A, B are nilpotent groups, then A × B is nilpotent.

Proof. This is because γk (A × B) = γk (A) × γk (B).

Corollary 3.2.13
If A1 , . . . , Al are nilpotent, then A1 × . . . × Al is nilpotent.

Before we proceed, we pause to recollect some simple facts about normal groups:
1. A, B / G =⇒ AB / G
2. A, B / G with A ∩ B = 1, then for all a ∈ A, b ∈ B, we have ab = ba

22
Advanced Abstract Algebra Fall 2018

3. A, B / G with A ∩ B = 1, then AB ∼
=A×B

Lemma 3.2.14 (Frattini’s argument)

If G is a finite group with N / G, R ∈ Sylp (N ), then G = N · NG (R) = NG (R) · N .

Proof. Let a ∈ G. Note that a−1 Ra ∈ Sylp (N ) since a−1 Ra ≤ N as N / G. Then, from
Sylow’s Theorems (S2? S3? I forget) there exists some n ∈ N such that

a−1 Ra = n−1 Rn =⇒ n−1 aRa−1 n = R =⇒ a−1 n ∈ NG (R) =⇒ a ∈ NG (R) · N

Theorem 3.2.15
Let G be a finite group. Then G is nilpotent if and only if all Sylow-p subgroups are
normal (and hence, unique).

We only give the forward direction of the proof of the above theorem, the reverse direction
is easily found online or in Robinson’s book. Maybe I will write it up later.
Proof. Let G be nilpotent, and R ∈ Sylp (G). Now, note that NG (R) / NG (NG (R)).
Then, by the Frattini Argument (3.3.9), we have:

NG (NG (R)) = NG (R) · NNG (NG (R)) (R)

But the latter is just NG (R). Then, by Proposition 3.3.5, we have NG (R) = G and so
R / G.

3.3 Two special subgroups

From the previous two sections, we can see that nilpotence is quite a nice property; on the
other hand, solvability is not as nice. Our next project is to try to characterize solvable
groups in terms of nilpotent groups arising from matrix groups.

3.3.1 The Frattini Subgroup

Recall the definition of a maximal subgroup:

Definition 3.3.1 (maximal subgroup)

max
H < G if H  G and there exists no S  G such that H  S  G.

The intersection of all the the maximal subgroups of a given group is called the Frattini
subgroup of that group.

Definition 3.3.2 (Frattini subgroup) T

Let G be a finite group. Then φ(G) = max H is called the Frattini subgroup of G.
H<G

23
Advanced Abstract Algebra Fall 2018

Note that the Frattini subgroup is also a characteristic subgroup, i.e. is invariant un-
der all automorphisms of the group. Recall that we’ve encountered another characteristic
subgroup before, namely the commutator subgroup of a group.

Lemma 3.3.3
char
1. C ≤ B / A =⇒ C / A.
char char char
2. C ≤ B ≤ A =⇒ C ≤ A.

Proposition 3.3.4 (Frattini subgroup is nilpotent)

Let G be a finite group. Then φ(G) is a nilpotent group.

Proof. Let R ∈ Sylp (φ(G)). Note that for Sylow p-groups, being normal is the same as
being a characteristic group. Using the Frattini argument, we have G = φ(G) · NG (R).
Now, if NG (R) = G, then we’re done by the definition of maximal subgroups. If NG (R) 
max
G, then there exists NG (R) ≤ M < G. By definition, φ(G) ≤ M =⇒ G = φ(G) ·
NG (R) ≤ M which is a contradiction.

Proposition 3.3.5
If N / G, then N is nilpotent if and only if N φ(G)/φ(G) is nilpotent.

Proof. Note that nilpotence is preserved by homomorphism, so the forward direction of

the above proposition follows trivially. We prove the reverse direction below.
We need to show that if R ∈ Sylp (N ), then R / N , i.e. R / G, i.e. NG (R) = G.
Claim. (Rφ(G))/φ(G) ∈ Sylp ((N φ(G))/φ(G)).
Proof. From the third isomorphism theorem, we have:

|(Rφ(G))/φ(G)| = |R/(R ∩ φ(G))|

and the latter divides |R| = ps . Note that we also have

|N φ(G)|
|(N φ(G))/φ(G) : (Rφ(G))/φ(G)| = |N φ(G) : Rφ(G)| =
|Rφ(G)|

Some fun calculations follow:

|N ||φ(G)|
|N φ(G)| |N ∩φ(G)|
= |R||φ(G)|
|Rφ(G)|
|R∩φ(G)|
|N |
|R|
= |N ∩φ(G)|
|R∩φ(G)|

Note that the thing above divides |N : R| but is coprime to p. It follows then that the
whole fraction thing in the last step above is coprime to p, and so we are done.

24
Advanced Abstract Algebra Fall 2018

By the claim and assumption, we have:

char
(Rφ(G))/φ(G) / (N φ(G))/φ(G) / G/φ(G)

And so, by the lemma, we have:

(Rφ(G))/φ(G) / G/φ(G) =⇒ Rφ(G) / G

The Frattini Argument then gives us G = Rφ(G) · NG (R) = NG (R) · Rφ(G) and so
G = NG (R)φ(G).So, if NG (R) were not the whole group G, there would exist a maximal
max
M such that NG (R) ≤ M ≤ G (this is from the argument in the previous proof), so
G = NG (R)φ(G) ≤ M , which is a contradiction.
In what follows G is taken to be a finite group unless stated otherwise.

Proposition 3.3.6
If N / G then φ(G) ≤ φ(G).
max char
Proof. We need, for every M ≤ G, φ(G) ≤ M . Now, note that for φ(G) ≤ N / G,
max
it follows that φ(N ) ≤ G. Suppose φ(N ) 6≤ M . So there exists M < G. Then, as
φ(N ) · M ≤ G, we would have φ(N )M = G =⇒ N = N ∩ G = N ∩ φ(N ) · M , but this
must mean N = φ(N ) · (N ∩ M ).
max
Now, if N ∩M 6= N , there exists maximal H < N such that N ∩M ⊆ H. As φ(N ) ≤ H,
we have N ⊆ H. This is a contradiction, so N ∩ M = N , and so N ⊆ M .
However, this contradicts φ(N ) ≤ M , which was our original assumption, and so we’re
done.
The argument in the first paragraph above is commonly called the “modular law”. I restate
it below for reference:

Lemma 3.3.7 (Modular Law)

If A, B, C ≤ G such that B ≤ A. Then A ∩ BC = B(A ∩ C).

3.3.2 The Fitting Subgroup

Lemma 3.3.8
G finite, A, B / G such that A, B are both nilpotent. Then AB is still a nilpotent
normal subgroup.

Proof. Proving that AB / G is omitted, as it should be easy to do. Now, we want to

show that AB is nilpotent. Let p be a prime. Let Sylp (A) = {R} and Sylp (B) = {S}.
Claim. Sylp (AB) = RS.

25
Advanced Abstract Algebra Fall 2018

char char
Proof. First, note that R < A / G =⇒ R / G. Similarly, S < B ≤ G =⇒ S / G.
These two facts together imply RS / G =⇒ RS / AB.
|R||S|
Now, |RS| = |R∩S| = pv for some v. We compute |AB : RS|.

|A||B|/|A ∩ B| |A : R||B : S|
|AB : RS| = =
|R||S|/|R ∩ S| |A ∩ B : R ∩ S|
The above is known to be an integer, so the denominator must divide the numerator
in the last expression. Co-primality then follows.

The desired result follows immediately.

We define fitting subgroups (with which we become more familiar in the homeworks) before
moving on to an application.

Definition 3.3.9 (fitting subgroup)

By F (G) we denote the fitting subgroup of G, which is defined as the greatest nilpotent
normal subgroup of G.

3.4 An application to finite groups

Warning: I was late to class this day and my notes will almost certainly contain errors in
this section. Sorry about this, and please let me know of any mistakes!

Proposition 3.4.1
Let G be finite and solvable, then F (G) ≥ φ(G). Write G = G/φ(G). Then F (G) =
F (G)/φ(G).

We state some additional facts below:

1. F (G) = R1 × R2 × . . . such that Ri is an elementary abelian pi subgroup.
2. CG (F (G)) = F (G). The solvability assumption plays a role here.
3. G/F (G) ∼
= G/F (G), and the latter acts on F (G) by conjugation faithfully.

Corollary 3.4.2 (a weak structure theorem)

G/F (G) ∼
= G∗ ≤ ×i GL(Fpi ).

Proposition 3.4.3
If |R| = pk where p is prime, then φ(R) = hR0 , rp | r ∈ Ri.

Proof. We first make the following claim, from which it follows that H in the given claim
is a normal subgroup of R.
max
Claim. If |R : H| = p, then H < R if and only if |R : H| = p.

26
Advanced Abstract Algebra Fall 2018

Proof. We know that R is nilpotent, and that |R : H| = p. Recall that as R is

nilpotent, if H  R then NR (H) H, so as H maximal, we have NR (H) = H, so
H / R. By the correspondence theorem, we have subgroups of R/H in correspondence
with subgroups H ≤ L ≤ R, but as H maximal, L = H or R. So, R/H only has
trivial subgroup, which means |R/H| = p.
Before, we proceed, another claim.
Claim. Given an index set I, and G / Ni for i ∈ I, G/ i∈I Ni ∼
T
= S ≤ ×i∈I G/Ni .
Proof. Consider Tthe map given by g 7→ (. . . , gNi , . . .). The kernel of this map is
given by ker = i∈I Ni . Use the first isomorphism theorem to conclude the above
result.
From the above claim, we have
max
H ≤ RH ∼
\
R/φ(R) = R/ =S≤× max R/H
H ≤ R

which is elementary abelian. This gives us φ(R) ⊇ R0 3 rp . So, φ(R) ⊇ hR0 , rp i. We

write L = hR0 , rp i.
Claim. φ(R/L) = 1.
Proof. Note R/L ∼
= Zp × Zp . . . × Zp . So, φ(R/L) ∼
= φ(Zp × . . . × Zp ) = 1.
From the above claim, we can conclude that φ(R) ⊆ L. With this, we are done.
One last proposition before we move on to free groups:

Proposition 3.4.4
If G is finite and solvable, then CG (F (G)) ≤ F (G).

I wasn’t present in class for the lecture that covered the above two propositions, so I might
have missed a few things. Sorry!

27
Advanced Abstract Algebra Fall 2018

4 Free Groups

Figure 4.1. Cayley graph of F ha, bi, the free group on 2 generators.
Free groups are foundational to many aspects of group theory. For instance every group is
the factor group of some free group. Informally, free groups are those in which no relations
hold between the elements besides the standard assumption that an element multiplied by
its inverse is the identity.
A familiar example of a free group is the group of integers. Another way to think of free
groups (which you may have seen before if, like me, you major in computer science)1 is as
words or strings over some set of generators or an alphabet.
Along the way, we will also encounter Cayley graphs, and will take a closer look at Cayley
graphs of free groups.

4.1 Construction of free groups

We give two constructions of free groups, one using a set of generators, and another using
a universal property.
• Given a set of symbols S, we can consider words over this set to be products of
elements of S ∪ S −1 . We say that S is a generating set for these words.
• We define the empty word  to be the word with no symbols.
• If, in a word, an element of S lies immediately next to its inverse, the word may be
simplified by omitting the pair.
• A word that cannot be simplified further is said to be reduced.

Definition 4.1.1 (free group)

Given a set S, the free group with generating set S, written FS , is the group of reduced
words over S. The group operation is given by concatenation.

We will sometimes write just F for the free group when the generating set S is understood.
We will also sometimes write F = hSi. It is easy to check that the above construction is
indeed a group, and is left as an exercise.
1 Lookup regular languages and automatic groups if interested in applications of group theory in the
theory of computation.

28
Advanced Abstract Algebra Fall 2018

We now give an alternative definition of free groups, which at first seems quite different from
the one above. With some reflection, however, it is easy to see that they are equivalent.
This is the more category-theoretical point of view, and free objects in a category (whatever
these animals are) appear everywhere in math.

Definition 4.1.2 (universal property definition of free groups)

F = hSi if for all f : X → A, where A is any group, there exists a unique homomor-
phism f : F → A a homomorphism that makes the following diagram commute:

S F
f
f
A

In the above diagram, the unnamed mapping denotes the inclusion from S into F . So you
can alternatively say that f S = f .

Proposition 4.1.3 (free groups are a bit like vector spaces)

Two free groups are isomorphic if and only if the cardinalities of their generator sets
are the same.

The reverse direction is fairly obvious and follows easily from the universal property con-
struction of free groups. The forward direction, however, is not, and we might talk about it
later, possibly in a week or two.

4.2 Cayley graphs

Cayley graphs are graphs associated with groups, and their construction as well as their
properties are quite intuitive. Hence I won’t be very formal in this section.
Suppose G is a group and S is a generating set. The Cayley graph Γ = Γ(G; S) is a colored
directed graph constructed as follows:
• Each elementg ∈ G is assigned a vertex: the vertex set V (Γ) of Γ is identified with G.
• Each generator s ∈ S is assigned a color, which we will (abusing notation) call s.
• For any g ∈ G, s ∈ S, the vertices corresponding to elements g, gs are joined by a
directed edge (g, gs) of color s. This gives us the edge set E(Γ).
Do we know whether a given graph is a Cayley graph? Indeed, given Γ, it is a Cayley graph
if:
1. Γ is connected as an ordinary graph.
2. G∗ ≤ Aut(Γ) such that G∗ acts regularly on V (Γ).
3. For a given vertex, and for each color x, there is exactly one outgoing edge from the
vertex colored by x; and similarly, exactly one incoming edge to the vertex colored by
x.

29
Advanced Abstract Algebra Fall 2018

4.3 Fun with free groups

The following example is important to keep in mind.

Example 4.3.1 (Cayley graphs of free groups)

Γ is a Cayley graph of F(X) if and only if the three conditions of Proposition 4.1.2
met, and if:
1. Γ is a tree as an ordinary graph
2. There exists no x ∈ X such that O(x) = 2

Now, what do Cayley graphs of subgroups of free groups look like?

The above question motivates the remainder of this section. Let G = hXi, H ≤ G. We
want to get Γ(H; Y = ?) from Γ(G; X), where Y is the generating set of H (and is unknown
to us at the moment).
First we require a tool from set theory: the (infamous) Zorn’s lemma. We will then use
group actions stuff to obtain a Cayley graph for H ≤ G; next time we will show that it is
indeed the Cayley graph of H. The argument given below is sometimes called a contraction
argument.

Lemma 4.3.2 (Zorn’s lemma)

Let S denote a partially ordered set with the property that for any totally ordered
subset T ⊆ S, there exists some u ∈ S such that for all t ∈ T , t ≤ u. There there
exists m ∈ S such that m is maximal in S, i.e. @ m ∈ S m ≤ s and m 6= s.

Now, look at the induced action of H ≤ G on V (Γ) such that for each g ∈ V (Γ) and for
each h ∈ H, we have h(g) := hg. The orbits of H acting on V (Γ) = G, then, are precisely
the right cosets of H.

Definition 4.3.3 (partial transversal)

T is a partial transversal if T ≤ Γ as a subgraph, such that for each g ∈ G, |Hg ∩
V (T )| ≤ 1, and T is connected (as an ordinary graph).

As we want to use Zorn’s Lemma, our partially ordered set (poset, for short) will be the set
of all partial transversals of Γ. Let us denote this set by T . It is a poset with respect to
containment.
So Zorn’s Lemma says that there exists a maximal partial transversal, T0 . We will show
that T0 is “complete” in the sense that it intersects each coset Hg for all g ∈ G exactly
once.
Claim. For every g ∈ G, we have |Hg ∩ V (T0 )| = 1.
Note that h(T0 ) is a maximal transversal too, as left multiplication is an automorphism.
We thus obtain a graph corresponding to H; all that remains is construction of a colored
and directed edges of the new graph. You can check that this subgraph is indeed a Cayley
graph for H.

30
Advanced Abstract Algebra Fall 2018

The above discussion essentially is Schreier’s original proof of the Nielsen-Schreier theorem,
which is stated below for completeness. Another proof, which uses some algebraic topology
(fundamental groups and covering spaces), can be found here. Yet another proof can be
found here.

Theorem 4.3.4: Nielsen-Schreier theorem

Every subgroup of a free group is free.

What is cool is that if we know the index of H in G in the above discussion, then we know
the size of the generating set of H! This is called Schreier’s index formula, and we prove it
below:

Proposition 4.3.5 (Schreier’s index formula)

Let |X| = n < ∞, and let G be the free group generated by X. If H ≤ G such
that |G : H| = k < ∞, then H is free with respect to some generating set Y , where
|Y | = (n − 1)k + 1.

Proof. Note that deg(h) = 2|Y | for all h ∈ H. This follows directly from how the Cayley
graph is constructed. By T we will denote our maximum partial transversal as in the
discussion on the preceding page. Then V (T ) = k, since it intersects each of our cosets
in exactly 1 vertex. Moreover, we have E(T ) = k − 1.
Now, in our original graph, the degree of each point p was 2n. Then, should the total
number of edges incident to any one of our cosets be 2n? No, since here we are double
counting the number of edges incident to p that are connected to points within the same
coset as p.
We divide the obtained expression divided by 2 (since we don’t add inverses to the size
of a generating set), and so we’re done.
Recall the universal-property definition of a free group. It is easy then to see that free
groups have lots of homomorphisms, and hence, lots of normal subgroups; what is more,
the next theorem states that they not only have several normal subgroups, but have several
normal subgroups of finite index.

Theorem 4.3.6: free groups are residually finite

If G is free, e 6= a ∈ F , then there exists N / F such that:
1. a ∈
/N
2. |F : N | is finite

Proof. Let p be a prime. Let a = xm mr

i1 . . . xir , where xij ∈ X, be a word in reduced form
1

(i.e. iu 6= iu+1 for all u, and mj 6= 0 for all j). Pick n such that pn - m1 m2 . . . mr .
Consider (r + 1) × (r + 1) matrices over Zpn . Let I denote the (r + 1) × (r + 1) identity
matrix, and Es,s+1 be the matrix whose (s, s + 1)th entry is 1, all others are 0.
Let G := {(r + 1) × (r + 1) upper unitriangular matrices over Zpn }. For each xj having
nonzero exponent in a, we define gj as follows:
Y
gj = (I + Eµ,µ+1 )
µ:iµ =j

31
Advanced Abstract Algebra Fall 2018

We want to show that a does not lie in the kernel of the above homomorphism.
Claim. gim1 1 . . . gimr r 6= I.
Proof. First, note the following for our special matrices: Es,t Ev,z = Es,z if t = v,
and Es,t Ev,z = 0 if t 6= v. Now, consider our situation: look at Eµ,µ+1 Eν,ν+1 . If
µ + 1 = ν, then iµ = j = iν = iµ+1 which is a contradiction. From this, we can see
that matrices of the form I + Eµ,µ+1 commute. More precisely:

(I + Eµ,µ+1 )(I + Eν,ν+1 ) = I + Eµ,µ+1 + Eν,ν+1

Thus, we can write:

Y X
gim1 1 . . . gimr r = (I + m1 Eµ,µ+1 )(some blah) = I + λEs,t
µ1 :iµ1 =i1 s,t

The proof can be completed easily.

Why are we doing this? Because this allow the rank (i.e. the size of the generating set)
of a free group to be well defined. But before doing that, we talk about a property that is
shared by several groups; in particular by finitely generated free groups.

Definition 4.3.7 (hopfian group)

G is a hopfian group if for all 1 6= N / G, G/N ∼
6 G.
=

Example 4.3.8
1. Any finite group is hopfian.
2. Let G = H (1) × H (2) × . . ., then if we take N = H (1) , the quotient group is still
an infinite group that is isomorphic to to G.

Being hopfian is a very nice property, as is evident from the next lemma, whose proof should
be filled out as an exercise.

Lemma 4.3.9
G is hopfian if and only if all surjective endomorphisms of G are automorphisms.

Theorem 4.3.10
If G is finitely generated and it is residually finite, then G is hopfian.

Proof. We use the definition of hopfian groups given by Lemma 4.3.9. Suppose the
contrary, i.e. there exists homomorphism  : G  G such that ∃ e 6= a ∈ ker . As
G is finitely generated, the number of homomorphisms from G to itself is finite, as
homomorphisms are uniquely determined by the images of the generators.
Now, consider the following set H = {all homomorphisms f : G → G/N } where N / G
such that a ∈/ N and |G : N | is finite. Such an N exists as G is residually finite. Again,
as G is finitely generated, H is finite.
Suppose η ∈ H. Then we have η ◦  : G → G/N . It follows that η ◦  ∈ H.

32
Advanced Abstract Algebra Fall 2018

Claim. If η1 , η2 ∈ H, then η1 6= η2 =⇒ η1 ◦  6= η2 ◦ .
Proof. This follows from the fact that  is surjective.
Now, we are done. Why? Because η ◦ (a) = e, as a ∈ ker . So, for all homomorphisms
in H (as it is finite and composition by  preserves membership in H) from G → G/N ,
a 7→ e. If we can find a homomorphism that does otherwise, then we have a contradiction,
as we are done. Indeed, the natural homomorphism mapping a 7→ aN does not map a
to the identity.

Corollary 4.3.11
If F is free on X, |X| < ∞, then F is hopfian.

And finally, we have:

Corollary 4.3.12 (rank of a free group is well defined)

Let F be free on X, |X| = n < ∞. Let Z ⊂ F such that hZi = F . If |Z| ≤ n, then
|Z| = n and F is free on Z too.

Proof. There exists f : X  Z, and so, by definition of free groups, there exists f :
F → F such that f X = f . Thus, Imf = F , i.e. f is a surjective endormorphism. By
Corollary 4.3.9, f is an automorphism of F .
What happens if, in the above corollary, you have one infinite group of generators and one
finite group of generators? Take surjective map, and you run into a contradiction!
What if both are ordinals? Suppose, as an example, one has generating set of size C, and
the other ω. This is unsurprisingly not possible, as on one hand, we must have our group
to have cardinality C, and simultaneously must also have cardinality ω.
With this, we end our chapter on free groups.

33
Part II

Representation Theory of Finite

Groups
t
Advanced Abstract Algebra Fall 2018

5 Basic Notions
The goal of group representation theory is to study groups via their actions on vector spaces.
Consideration of groups acting on sets leads to such important results as the Sylow theorems.
By studying actions on vector spaces even more detailed information about a group can be
obtained. This is the subject of representation theory.
Important. In these notes, only representations over the complex field C will be considered.
Moreover, all groups are assumed to be finite unless stated otherwise.

5.1 What is a representation?

By GL(V ) we denote the group of all nonsingular linear transformations of a vector space V
over C. Alternatively, it is the set of all bijective linear transformations on V with respect
to composition.

Definition 5.1.1 (representation of a group)

Let G be a group and V a vector space (linear space) of finite dimension over C. A
homomorphism φ : G → GL(V ) will be called a representation of G on V .

Using basic linear algebra, it is clear that Definition 5.1.1 is equivalent to:

Definition 5.1.2 (representation of a group)

Let G be a group and n a positive integer. A homomorphism φ : G → GLn (C) will
be called a representation of G of degree n.

where GLn (C) denotes the group of all nonsingular n × n matrices over C.
Why two definitions? Because sometimes, as we will see, working concretely with some
special classes of matrices leads to some really deep results regarding abstract groups, e.g.
Burnside’s theorems, which we will learn about later.
I sometimes refer to the abstract definition of a representation (Definition 5.1.1) as a repre-
sentation, and the concrete one (Definition 5.1.2) as a matrix representation to keep things
clear.
Here are three examples of degree one representations:

Example 5.1.3
We look at representations of cyclic groups over C. First, note that GL1 (C) ∼=
C ∗ . The isomorphism is given by f 7→ f (1). Think of this visually using polar
representations of complex numbers. Then, the following are representations:
1. Z/2Z → C ∗ given by m 7→ (−1)m
2. Z/4Z → C ∗ given by m 7→ im
3. More generally, Z/nZ → C ∗ given by m 7→ e(2πim/n)

35
Advanced Abstract Algebra Fall 2018

Definition 5.1.4 (G-invariant subspaces)

Let φ : G → GL(V ) be a representation of G, and let W be a subspace in V (written
W ≤ V ); W is called G-invariant if φ(g)(W ) ⊆ W (or, equivalently, φ(g)(W ) = W )
holds for every g ∈ G.

Naturally, given a representation (V, ρ) of a group G, if W ⊆ V is a G-invariant subspace,

then we have a natural representation (W, ρ) of G whose degree will be the dimension of
W . Thus, if a representation V has a G-invariant subspace, then you can, in a way, obtain
a “smaller” representation of G.
Those representations which cannot be reduced in the above sense have a special name:

Definition 5.1.5 (irreducible representations)

A representation φ : G → GL(V ) is irreducible if {0} and V 6= {0} are the only
G-invariant subspaces.

Irreducible representations are to representation theory what prime numbers are to the
integers. We will learn more about this when we see Maschke’s Theorem.

Example 5.1.6
If the vector space V has dimension 1, then any representation of G on V is irreducible.

5.2 Maps between representations

More generally, we can consider morphisms between representations, and if a morphism

is bijective, then we say that two representations are isomorphic/equivalent. I also came
across the term intertwining operators for such maps.

Definition 5.2.1 (equivalent representations)

Let φ1 : G → GL(V1 ) and φ2 : G → GL(V2 ) be representations of the same group G.
They are called equivalent (written as φ1 ∼ φ2 ) if there is some f ∈ HomC (V1 , V2 )
such that f is bijective and for every g ∈ G, φ2 (g) = f ◦ φ1 (g) ◦ f −1 (or, equivalently,
f ◦ φ1 (g) = φ2 (g) ◦ f ) holds.

Essentially what the above definition is saying is that given φ1 and φ2 two representations
of G over V1 and V2 respectively, then they are equivalent if there’s some isomorphism
f : V1 → V2 such that φ2 (g) = f ◦ φ1 (g) ◦ f −1 . In pictures, we have that the following
diagram
φ1 (g)
V1 V1
f f
φ2 (g)
V2 V2
commutes for all g ∈ G.

36
Advanced Abstract Algebra Fall 2018

Intuitively, this is equivalent to saying that “two representations are equivalent if and only
if they describe the same representation but in different bases”. The map f above is nothing
more than a change of basis transformation.

Proposition 5.2.2
If φ1 and φ2 are equivalent representations of G, then φ1 is irreducible ⇐⇒ φ2 is
irreducible.

I omit the proof as it’s very straightforward.

The following table helps me remember everything that we have learned up until this point:
Groups Vector Spaces Representations
Subgroup Subspace G-invariant subspace
Simple group 1-dimensional subspace Irreducible representation
Direct product Direct sum Direct sum
Isomorphism Isomorphism Equivalent representations

5.3 Maschke’s Theorem and Schur’s Lemma

We begin this section with a motivating example.

Example 5.3.1 (Representations of S3 )

We take G = S3 the symmetric group on 3 elements, and n = dimC V = 3. Take
ψ : a 7→ ψ(a) ∈ C3×3 such that
(
1 a(j) = i
(ψ(a))i,j =
0 otherwise

where i, j denote the labels of the three elements that S3 permutes. So we have:
 
0 0 1
a = (1, 2, 3) 7→ 1 0 0
0 1 0
What’s an invariant subspace of this representation? Indeed, if you look at the
action of the above matrix on an arbitrary vector, it simply permutes the order of
the components of the vector. We have the following G-invariant subspace:
  
 λ 
W1 := λ | λ ∈ C
λ
 

Are there any other G-invariant subspaces? Let’s look at the orthogonal complement
of W1 .
        
 1 1   α1 X 
W2 := α  1  + β −2 | α, β ∈ C = α2  ∈ C3 | αi = 0
−2 1 α3
   
i

37
Advanced Abstract Algebra Fall 2018

Observe that W1 ⊥ W2 , and that G = W1 ⊕ W2 = C3 . We can thus obtain two

representations of G on W1 and W2 , say ψ1 and ψ2 respectively. We say that ψ =
ψ1 ⊕ ψ2 . We now try to obtain a matrix representation for ψ2 (a) which arises from
the “restriction” of the original ψ(a) to W2 .
      
0 0 1 1 1 1
1 0 0  1  = (−1)  1  + (−1) −2
0 1 0 −2 −2 1
    
0 0 1 1 1
1 0 0 −2 = (1)  1 
0 1 0 1 −2
Thus, we obtain:  
0 0 1  
1 −1 1
0 0
−1 0
0 1 0

Important. Above, we decomposed a degree 3 representation of S3 into two irreducible

representations of degree 1 and 2 respectively. Working with an infinite group can change
the rather nice situation we have in the above example. We will see more of this in the
homework.
A natural question to ask then is whether all representations can be decomposed into the
direct sum of irreducible representations? Surprisingly, the answer is yes! And this leads us
to Maschke’s Theorem.

Theorem 5.3.2: Maschke’s Theorem

Let φ : G → GL(V ) be a representation of G and assume that W is a G-invariant
subspace in V . Then there exists a G-invariant direct complement to W in V , i.e. a
G-invariant subspace U in V such that V = W ⊕ U .

Proof. Let π be a projection on V such that Im π = W (a linear transformation π is

called a projection if π 2 = π; it is easy to show that V = Im π ⊕ Ker π holds for any
projection). Set
1 X
π ∗ := φ(h)−1 πφ(h) .
|G|
h∈G
∗ ∗
We show that π is also a projection with Im πP = W , and that U := Ker π ∗ is G-
invariant. For assume w ∈ W , then π (w) = |G| h∈G φ(h)−1 πφ(h)(w).
∗ 1

Now, as W is G-invariant, then we have φ(h)(w) = w, and as π is a projection, we have

π(w) = w, and again, by G-invariance of W , we have φ(h)−1 (w) = w. Thus, we have
π ∗ (w) = w. It follows then that W ⊆ Im π ∗ . On the other hand, let v ∈ V and h ∈ G.
Then, πφ(h)−1 (v) ∈ W , and as W is G-invariant, we have φ(h)πφ(h)−1 (v) ∈ W . It
follows then that π ∗ (v) ∈ W , and so Im π ∗ ⊆ W . We conclude that Im π ∗ = W .

38
Advanced Abstract Algebra Fall 2018

Next, pick arbitrary g ∈ G. Then we have:

!
∗ −1 1 X
φ(g)π φ(g) = φ(g) −1
φ(h) πφ(h) φ(g)−1
|G|
h∈G
1 X
= φ(hg −1 )−1 πφ(hg −1 )
|G|
h∈G
1 X
= φ(h)−1 πφ(h)
|G|
h∈G
∗
=π

and so φ(g)π ∗ = π ∗ φ(g) holds for each g ∈ G. At last assume u ∈ U and g ∈ G; then
π ∗ φ(g)(u) = φ(g)π ∗ (u) = φ(g)(0) = 0 i.e φ(g)(u) ∈ U . This is because π ∗ sends φ(g)(u)
to 0, but ker π ∗ = U . Thus U is G-invariant.
Important. Maschke’s Theorem goes horribly wrong in the case of infinite groups, as we
will see on the homework.
Before moving on, we quickly define the direct sum of representations.

Definition 5.3.3
Let φi : G → GL(Vi ) be representations of G (i = 1, 2 . . . , k). The (direct) sum of
Lk
these representations is φ = φ1 + φ2 + · · · + φk where φ : G → GL( i=1 Vi ) is defined
by
φ(g)(v1 , v2 , . . . , vk ) := (φ1 (g)(v1 ), . . . , φk (g)(vk )) .
(It follows that φ1 + φ2 + · · · + φk is also a representation of G .)

The real power of Maschke’s Theorem is the following corollary.

Corollary 5.3.4
Every representation is equivalent to the direct sum of irreducible representations.

and now it should make sense why we earlier called irreducible representations to be the
prime numbers of representation theory.
The other basic theorem in this section is Schur’s Lemma, which is also of great use in
representation theory.

Lemma 5.3.5 (Schur’s Lemma)

Let φ1 : G → GL(V1 ) and φ2 : G → GL(V2 ) be irreducible representations of G.
Assume that f ∈ HomC (V1 , V2 ) such that for every g ∈ G, f ◦ φ1 (g) = φ2 (g) ◦ f holds.
Then
1. If φ1 and φ2 are not equivalent then f = 0.
2. If φ1 = φ2 := φ then f = λI, a scalar multiple of the identity.

Proof. (0) We claim that ker f and Im f are G-invariant subspaces in V1 and V2 re-
spectively. Indeed, suppose v1 ∈ ker f . Then f ◦ φ1 (g)(v1 ) = φ2 (g) ◦ f (v1 ) =

39
Advanced Abstract Algebra Fall 2018

0 =⇒ φ1 (g)(v1 ) ∈ ker f , so it is G-invariant. Similarly suppose f (v2 ) ∈ Im f . Then

φ2 (g) ◦ f (v2 ) = f ◦ φ1 (g)(v2 ) ∈ Im f just because we are composing with f .
As φ1 , φ2 are irreducible the following cases can only occur: (i) Ker f = 0 , Im f = V2 ;
or (ii) Ker f = V1 and Im f = 0 .
1. Under the assumption in (1) f would be bijective in case (i), implying that φ1 and
φ2 are equivalent; a contradiction.
2. To prove (2) let w be an eigenvector of f , i.e. w 6= 0 and f (w) = βw with some
0 6= β ∈ C . We can do so as C is algebraically closed. As a scalar transformation
commutes with all linear transformations, f ∗ := f − βI also satisfies the assump-
tions of the lemma. Since 0 6= w ∈ ker f ∗ , our f ∗ must be zero by (ii); thus
f = βI.

Important. Note that only the second part of Schur’s Lemma relies on the fact that our
underlying field is C; more specifically, it relies on the fact that C is algebraically closed and
hence guarantees the existence of an eigenvector. The first part holds for representations
over arbitrary fields.
Remark. Recall that we have taken G finite. Assume that φ1 : G → GL(V1 ) and φ2 :
G →PGL(V2 ) are irreducible representations of G and f0 ∈ HomC (V1 , V2 ) is arbitrary. Then
f := h∈G (φ2 (h))−1 f0 (φ1 (h)) meets the requirements of 5.2.4.
For let g ∈ G then:
!
X
φ1 −1 φ2 φ1 −1 φ1 −1 φ2
(g ) fg = (g ) (h ) f0 h g φ2
h∈G
X X
φ1 −1
= (g ) (hφ1 )−1 f0 hφ2 g φ2 = ((hg)φ1 )−1 f0 (hg)φ2
h∈G h∈G
X
= (hφ∗ 1 )−1 f0 hφ∗ 2 =f.
h∗ ∈G

40
Advanced Abstract Algebra Fall 2018

6 Characters and Burnside’s Theorems

We quickly set up the notation that we will be using during this section.
1. For a matrix M we’ll denote by Mi,j the (i, j)-th element of M .
2. For an n × n matrix A we denote by Tr(A) the trace of A, i.e. the sum of its elements
in the main diagonal. (As known from linear algebra, Tr(A) equals the sum of the
eigenvalues of A . It is also of great importance that Tr(M −1 AM ) = Tr(A) holds for
any A and M .)
3. Let E (k,`) denote the matrix (of a given size) with its (k, `)-th element 1 and all others
0; i.e. (
(k,`) 1 if i = k and j = `
Ei,j =
0 otherwise

4. The standard notation (

1 if i = j
δi,j =
0 6 j
if i =
will also be used.

6.1 What is a character?

Characters will be of great use, as we will see later in the course. To every representation V
of G we will attach a so-called character. It will turn out that the characters of irreducible
representations of V will determine the representation V completely.
Recall Schur’s Lemma. The following proposition will later motivate the definition of a
character.

Proposition 6.1.1
Assume that φ1 : G → GLn1 (C) , φ2 : G → GLn2 (C) are irreducible representations
of G . Then
−1
P
1. If φ1 and φ2 are not equivalent then h∈G φ1 (h) a,b · φ2 (h)c,d = 0 for all
1 ≤ a, b ≤ n1 and 1 ≤ c, d ≤ n2 .
P −1 |G|
2. If φ1 = φ2 := φ then h∈G φ1 (h) a,b · φ2 (h)c,d = δa,d δb,c n for all 1 ≤
a, b, c, d ≤ n (n := n1 = n2 ) .

Proof. We shall make use of matrix representations. Take f0 := E (b,c) of size n1 ×n2 ; this
is the same f in the remark after Schur’s Lemma in §5. We then compute the (a, d)-th

41
Advanced Abstract Algebra Fall 2018

φ1 (h)−1 E (b,c) φ2 (h) :

P
element of h∈G
!
X X
φ1 (h)−1 E (b,c) φ2 (h) = (φ1 (h)−1 E (b,c) φ2 (h))a,d =
h∈G a,d h∈G

X n1
X n1
XX n2
X
(i) φ1 (h)−1
a,i · (E
(b,c)
φ2 (h))i,d = φ1 (h−1 )a,i · E (b,c) i,j φ2 (h)j,d =
h∈G i=1 h∈G i=1 j=1
XX n1 X
φ1 (h−1 )a,i δi,b φ2 (h)c,d = φ1 (h−1 )a,b φ2 (h)c,d .
h∈G i=1 h∈G

ThusP(1) is proved by Schur’s Lemma. As for (2), we know again from Schur’s Lemma
that h∈G (h−1 )φ E (b,c) hφ = λI . Since
!
1 1 X
−1 (b,c)
λ = Tr(λI) = Tr φ(h )E φ(h) =
n n
h∈G
1 X 1 X 1
Tr(φ(h−1 )E (b,c) φ(h)) = Tr(E (b,c) ) = |G|δb,c ,
n n n
h∈G h∈G

we have by (i)
!
1 X
−1 (b,c)
|G|δa,d δb,c = δa,d λ = φ(h )E φ(h) =
n
h∈G a,d
X
φ(h−1 )a,b φ(h)c,d .
h∈G

The role played by the trace of a matrix above is crucial, and when dealing with representa-
tions, the trace of the representation of a group element is given by the character associated
with that representation.

Definition 6.1.2 (character of a representation)

Let φ : G → GLn (C) be a representation of G . The character of φ is the function
χφ : G → C defined by χφ (g) := Tr (φ(g)) .

Proposition 6.1.3 (properties of characters)

Let χ be the character of some representation φ of G . Then for every g, h ∈ G
1. χ(h−1 gh) = χ(g) ;
2. equivalent representations have the same character;
3. χ(g −1 ) = χ(g) .

Proof. Note that (1) follows from the remark at the beginning of this section, (2) is
an immediate consequence of (1). To prove (3) let g ∈ G. Assume that the minimal
polynomial of φ(g) is p(x) ; then p(x) divides xk −1 where k = o(g) . Thus the eigenvalues
of φ(g) are k-th roots of unity. It follows that the characteristic polynomial of φ(g) is

42
Advanced Abstract Algebra Fall 2018

c(x) = det (In −φ(g)) = (x−ε1 )(x−ε2 ) . . . (x−εn ) (where εi k = 1 , n = the degree of φ).
Thus the characteristic polynomial of (φ(g))−1 is xn c( x1 ) = (x − ε11 )(x − ε12 ) . . . (x − ε1n ) ,
hence χ(g −1 ) = Tr((φ(g))−1 ) = ε11 + ε12 +· · ·+ ε1n = ε1 +ε2 +· · ·+εn = ε1 + ε2 + · · · + εn =
Tr(φ(g)) = χ(g) .

6.2 Orthogonality relations

The properties of characters naturally lead to the following”

Corollary 6.2.1 (The First Orthogonality Relation)

Assume that φ1 , φ2 are irreducible representations of G and their character is χ1 and
χ2 , respectively. Then
(
1 X 0 if φ1  φ2
χ1 (g)χ2 (g) =
|G| 1 if φ1 ∼ φ2
g∈G

Proof. Let n1 and n2 denote the degree of φ1 and φ2 .

1 X 1 X
χ1 (g)χ2 (g) = χ1 (g −1 )χ2 (g) =
|G| |G|
g∈G g∈G
n1 n 2
1 X X X 1 X X
( φ1 (g −1 )i,i )( φ2 (g)j,j ) = ( φ1 (g −1 )i,i φ2 (g)j,j ) .
|G| i=1 j=1
|G| i,j
g∈G g∈G

Assume that φ1  φ2 , then for every i , j the sum g∈G φ1 (g −1 )i,i φ2 (g)j,j is zero by
P
6.1.1 (1) . If φ1 ∼ φ2 then we can assume that φ1 = φ2 := φ (with χ1 = χ2 := χ ,
n1 = n2 := n) , and the assertion will follow from 6.1.1 (2) :

1 X 1 X X
χ(g)χ(g) = ( φ(g −1 )i,i φ(g)j,j ) =
|G| |G| i,j
g∈G g∈G
n
1 X |G| 1 X |G| 1 |G|
δi,j = = · · n = 1.
|G| i,j n |G| i=1
n |G| n

Definition 6.2.2
Let |G| = n and take a vector space V of dimension n such that the elements of some
basis B are labelled by the elements of G : B = {bh | h ∈ G} . We define the regular
representation R of G as R : G → GL(V ) such that for every g ∈ G and bh ∈ B
(bh )g R := bhg .

Thus, given any group, we can obtain a regular representation. This, together with Cayley’s
Theorem, makes the study of representations of symmetric groups very important.

43
Advanced Abstract Algebra Fall 2018

Lemma 6.2.3
R is a representation of G . Let % denote its character; then
(
|G| if g = 1
%(g) =
0 if g 6= 1

Proof. The first statement is clear since any linear transformation is uniquely determined
by its action on a basis. To calculate % consider R in matrix form; each R(g) being a
permutation matrix %(g) equals the number of 1-s in the main diagonal of R(g) , i.e. the
number of basis vectors bh satisfying bhg = bh , i.e. the number of group elements h
satisfying hg = h .
We know by Maschke’s Theorem that every representation is equivalent to a sum of irre-
ducible representations; therefore R ∼ φ1 + φ2 + . . . + φm with φi irreducible. It is easy
to see (e.g. by taking the matrix form of the representations) that the character of a sum
equals the sum of the corresponding characters, thus % = χφ1 + χφ2 + . . . + χφm . Since
equivalent representations have the same character this can be written (by collecting the
equivalent φj -s) as
% = m1 χ1 + m2 χ2 + . . . + mc χc reg
where χ1 , χ2 , . . . , χc already belong to pairwise nonequivalent (irreducible) representa-
tions, and m1 , m2 , . . . , mc are positive integers. To be able to compute these mi -s the
following definition will be useful.

Definition 6.2.4 (inner product)

Let f1 , f2 be mappings (functions) from G to C . Their inner product is a scalar
defined as
1 X
(f1 , f2 ) := f1 (g) · f2 (g) .
|G|
g∈G

Proposition 6.2.5
The inner product is bilinear , i.e.
1. (f1 + f2 , f3 ) = (f1 , f3 ) + (f2 , f3 ) ;
2. (f1 , f2 + f3 ) = (f1 , f2 ) + (f1 , f3 ) ;
3. (f1 , αf2 ) = α(f1 , f2 ) (for each α ∈ C);
4. (αf1 , f2 ) = α(f1 , f2 ) (for each α ∈ C).

In terms of the inner product the first orthogonality relation can be formulated as follows.

Assume that φ1 , φ2 are irreducible representations of G and their character is χ1 and χ2 ,

respectively. Then
(
1 if φ1 ∼ φ2
(χ1 , χ2 ) = ortog I.
0 if φ1  φ2 .

44
Advanced Abstract Algebra Fall 2018

Proposition 6.2.6
Keeping the notation of (reg), mi = χi (1) , i.e. the degree of the representation
belonging to χi .

Proof. Let’s use 5.3.8 and (ortog I.):

c
X c
X c
X
(χi , %) = (χi , mk χk ) = (χi , mk χk ) = mk (χi , χk ) = mi .
k=1 k=1 k=1

1
On the other hand, by 5.3.6 , (χi , %) = |G| χi (1) · %(1) = χi (1) .

6.3 Algebraic Integers

We quickly define a few algebraic structures that will come in later when we talk about
algebraic integers.

Definition 6.3.1 (F-algebra)

Let F be a field and A a set with three operations + , · and . (multiplication of elements
of A by elements of F). (A , + , · , . ) is called an F–algebra (or an algebra over F) if
(A , + , . ) is a vector space over F , (A , + , · ) is a ring, and for any a , b ∈ F and
λ ∈ F λ. (a · b) = (λ. a) · b = a · (λ. b) holds.

Thus, a F-algebra A is nothing more than a possibly non-commutative ring equipped with
an injective ring homomorphism F ,→ A whose image is the “copy of F”. Importantly, we
have 1F 7→ 1A .

Definition 6.3.2 (group algebra)

Let G be a group. We denote by C[G] the set of formal linear combinations as follows:
 
X 
C[G] = λg g | λg ∈ C
 
g∈G

By taking the obvious definitions of (component-wise) addition and taking scalar mul-
tiples, C[G]becomes a vector space over C. Defining multiplication in C[G] by
  !  
X X X X X
 λg g  λh h := λg λh gh =  λg λh  a
g∈G h∈G g,h∈G a∈G g,h∈G:gh=a

C[G] becomes an algebra over C . It is called the group algebra of G over C.

Note that we can have an alternative interpretation of group algebras:

X
λa · a ↔ f : G → C f (a) := λa
a∈G

45
Advanced Abstract Algebra Fall 2018

Then, note that multiplication can be thought of as convolution:

X
f ∗ g(c) := f (a)g(a−1 c)
a∈G

We now look at a generalization of vector spaces, called modules.

Definition 6.3.3 (F-module)

Let G be a group, F a field and V a vector space over F. Then V is called an F[G]–
module if for each a ∈ F[G] and v ∈ V an element (v)a ∈ V is defined satisfying:
1. (v1 + v2 )a = (v1 )a + (v2 )a for vi ∈ V , a ∈ F[G])
2. (λv)a = λ(v)a (for v ∈ V and a ∈ F[G])
3. (v)[ab] = ((v)a)b (for v ∈ V , a, b ∈ F[G])
4. (v)1 = v (for v ∈ V ).

Intuitively speaking, given a ring R, a R-module is simply an algebraic structure where you
can add two elements, and scale by elements of R. Then, note that a vector space is simply
a module whose commanding ring is a field.

Definition 6.3.4 (algebraic integer)

Let β be an algebraic number (in C) ; then β is called an algebraic integer if all
coefficients of its minimal polynomial are integers.

Proposition 6.3.5
1. Let β ∈ C ; then β is an algebraic integer iff there exists some monic polynomial
p(x) ∈ Z[x] such that p(β) = 0 .
2. The set of all algebraic integers form a subring in C .

Notation. For this section we set A := C[G] .

Definition 6.3.6 (center)

Let the centre of A be Z(A) := {c ∈ A | (∀a ∈ A) ca = ac} .

First, given a group algebra F[G], when does an element lie in the center of this algebra?
Indeed, z ∈ Z(F[G]) ⇐⇒ (∀g ∈ G)zg = gz. This is important and might come in handy
in several situations!
P
Is it possible to obtain a nice form for elements in the center? Indeed, let z = a∈G λa a.
Then, for g ∈ G, if we have zg = gz =⇒ g −1 zg = z, and so, we have:
X X
λa g −1 ag = λa a
a∈G a∈G

Maybe it is worth noting here that two formal linear combinations are equal if and only if
coefficients are all equal. Each element in a group algebra has a unique form as a unique
formal linear combination.

46
Advanced Abstract Algebra Fall 2018

We call ωi in the proposition below a class sum.

Proposition 6.3.7 (properties of the center)

1. Z(A) is a subalgebra of A , i.e. it is a subspace and a subring as well.
2. Let G = K1 ∪ K2 ∪ · · · ∪ Kc be the decomposition
P of G into the disjoint union of
conjugacy classes. For each Ki define ωi := a∈Ki a . Then {ωi | 1 ≤ i ≤ c}
is a C–basis of Z(A) (considered as a subspace).

Proof. (1) is obvious Pby definition. For (2) observe that c ∈ Z(A) iff cg = gc holds for
−1
each g ∈ G . Let c P h∈G αh h , then c ∈ Z(A) iff for every g ∈ G g cg = c holds, i.e.
=
−1
P
h∈G αh g hg = h∈G αh h . This just requires αg−1 hg = αh for every g, h ∈ G , which
means that c is the linear combination of the ωi -s.
On the other hand, the ωi -s are linearly independent as they do not have “common
elements”.

Theorem 6.3.8
Let φ : G → GLn (C) be an irreducible representation of G. We denote the character
of φ by χ .
1. Each φ(ωi ) is a scalar multiple of the identity, i.e. φ(ωi ) = λi In (with some
λi ∈ C) .
2. Let gi ∈ Ki ; then λi = |Ki |χ(g
n
i)
is an algebraic integer.
Pc
Proof. We have for each 1 ≤ i, j ≤ c that ωi ωj = k=1 tk (i,j) ωk (with tk (i,j) ∈ C) . Since
the ωk -s are linearly independent the coefficient tk (i,j) equals the number of solutions
of the equation xy = gk with gk ∈ Kk fixed and x ∈ Ki , y ∈ Kj . This means that
tk (i,j) ∈ N ⊂ Z . Now fix j := 1 , say, and consider the system of equations
c
X
ωi ω1 = tk (i,1) ωk ,
k=1

(for i = 1, 2, . . . , c) written in the form

c
X
(δi,k ω1 − tk (i,1) )ωk = 0 .
k=1

Now apply the extension of φ to A to get

c
X
(δi,k φ(ω1 ) − tk (i,1) )φ(ωk ) = O .
k=1

By Schur’s Lemma each φ(ωk ) = λk In (with some λk ∈ C), and obviously λk =

1 1
P |Kk |χ(gk )
n Tr(λk In ) = n Tr( a∈Kk φ(a)) = n ; so the previous equations become
c
X
(δi,k λ1 − tk (i,1) )λk = 0 .
k=1

This can be considered as a homogeneous system of linear equations for the variables
λk , k = 1, 2, . . . , c . Since χ is not the zero function (χ(1) = n is a positive integer) not

47
Advanced Abstract Algebra Fall 2018

all λk -s are zero. Thus the above homogeneous system of linear equations has nontrivial
solutions, therefore the determinant of its matrix should be zero. This determinant is a
polynomial expression of λ1 of the form p(λ1 ) where p(x) ∈ Z[x] and p(x) is monic. So
6.3.5 (1) implies that λ1 is an algebraic integer, and the argument clearly extends to any
of the other λj -s.
We quickly recollect all the facts we know about characters and algebraic integers:
1. If χ is any characters and g ∈ G, then χ(g) is an algebraic integer.
2. If χ is the character of an irreducible representation, and g ∈ K where K is a conjugacy
class in G, then χ(G)|K|
χ(e) is an algebraic integer.
As a corollary of the above two, we obtain the following:

Proposition 6.3.9
Let φ be an irreducible representation of G of degree n and denote its character by
χ . Assume that K is a conjugacy class in G such that (|K|, n) = 1 . Then χ(g)
n is an
algebraic integer for any g ∈ K .

Proof. Since χ(g) is a sum of roots of unity, χ(g) is an algebraic integer and so is χ(g)|K|
n
by the previous proposition. As |K| and n are coprimes there exist integers `, t satisfying
`|K| + tn = 1; thus χ(g) χ(g) χ(g)
n = n · 1 = n (`|K| + tn) =
χ(g)|K|
n ` + χ(g)t is an algebraic
integer.

Proposition 6.3.10
Assume that φ is an irreducible representation of G of degree n and χ is its character.
Assume that K is a conjugacy class in G such that χ(g)n is an algebraic integer (with
g ∈ K) . Then either g φ = αIn is a scalar multiple of the identity or χ(g) = 0 .

Proof. Let λ = χ(g) n ; we know that χ(g) = ε1 + ε2 + . . . + εn where ε1 , ε2 , . . . , εn

— the eigenvalues of φ(g) — are roots of unity, particularly |εi | = 1 . Hence
|λ| = |ε1 +ε2 +...+ε
n
n|
≤ 1, by the triangle inequality. Let r be the order of g and let
µ denote a primitive r-th root of unity. Let F = Q(µ). If σ ∈ Ga ` (F |Q) then
λσ is also
Q an algebraic integer and |λσ | ≤ 1 as well. Consider now the polynomial
σ
p(x) := σ∈Ga ` (F |Q) (x − λ ) . Since the coefficients of p(x) are symmetric functions
of the λσ -s, each coefficient belongs to Φ(Gal (F |Q)) = Q ; so p(x) ∈ Q[x] . On the
other hand, the coefficients are algebraic integers, as they are sums of products of the
λσ -s; hence p(x) ∈ Z[x] . Therefore the constant term of p(x) is an integer c such that
|c| = σ∈Ga ` (F |Q) |λσ | ≤ 1 , i.e. c = 0 or c = ±1 . If c = 0 then some of the λσ -s is
Q

zero, and then λ = 0 yields χ(g) = 0 . If c = ±1 then |λ| = 1 ; this can hold only if
ε1 = ε2 = . . . = εn . Since the minimal polynomial of φ(g) divides xr − 1 , the minimal
polynomial has no multiple roots. Thus the minimal polynomial of φ(g) should be x−ε1 ,
implying that φ(g) = ε1 In .

48
Advanced Abstract Algebra Fall 2018

6.4 Two Theorems of Burnside

We conclude this section with a famous group-theoretic result whose proof uses representa-
tion theory.

Theorem 6.4.1
/ K and |K| = pt , p
Let G be a finite group, K a conjugacy class in G such that 1 ∈
a prime, t ≥ 0. Then G can’t be nonabelian simple.

Proof. Suppose by way of contradiction that G is nonabelian simple, and let g ∈ K.

Let χi be the character of a nontrivial irreducible representation φi of degree ni ; then
Ker φi = 1 by the simplicity of the nonabelian group G , i.e. φi is 1–1.
Case 1: p doesn’t divide ni . Then χin(g)
i
is an algebraic integer by 6.3.9 , hence χi (g) = 0
or φi (g) is a scalar multiple of Ini . Suppose that χi (g) 6= 0 ; then 1 6= g ∈ Zφi (G) :=
{h ∈ G | φi (h) = αh Ini , αh ∈ C} . It is easy to see that Zφi (G) is a normal subgroup in
G and in our case Zφi (G) ' φi (Zφi (G)) is abelian; this is a contradiction, so χi (g) = 0
in this case.
Case 2: p divides ni . Let ni := pmi .
But we know that %(g) = 0 . Thus
c
X X
0 = %(g) = ni χi (g) = 1 + pmi χi (g)
i=1 i:p|ni

(1 standing for
Pn1 χ1 (g) where φ1 denotes the trivial representation).
Hence − p1 = i:p|ni mi χi (g) is an algebraic integer, a contradiction.

Theorem 6.4.2
Every group G of order pa q b (p, q primes) is soluble.

Proof. Induction on the order of the group. If there is some proper normal subgroup N
in G then the order of N and G/N is also the product of two prime powers, therefore N
and G/N is soluble by induction hypothesis; then G is also soluble. If G is simple and
abelian we are done; suppose therefore that G is nonabelian simple. Let Q be a Sylow
q–subgroup of G . Since groups of prime power order are soluble Q must be nontrivial;
let 1 6= g ∈ Z(Q) . Then the number of conjugates of g in G equals |G : CG (g)| , which
divides |G : Q| as Q ≤ CG (g) . But |G : Q| is a power of p , and so is |G : CG (g)| ; this
contradicts to 6.4.1 .

49
Advanced Abstract Algebra Fall 2018

7 More Character Theory

In this section, we look at some results that make use of character theory of finite groups.
I.M. Isaacs’s textbook is a good reference for this material, according to Prof. Hermann.

7.1 Induced characters

A representation is nothing but a special kind of homomorphism, and Burnside’s Theorems

were a non-simplicity criterion. Thus, we’re looking for the kernel of some homomorphism.
Turns out we can use the relations between a character and associated representations to
obtain the kernel of the representation, and this is, in a way, a second look we’ll be taking
at Burnside’s Theorems.

Proposition 7.1.1
Let ψ : G → GLn (C) be a representation of G, and let χ be the character o ψ. Then:
1. If a ∈ ker ψ, then χ(a) = n = χ(e).
2. Conversely, if a ∈ G such that χ(a) = χ(e), then a ∈ ker ψ.

Note that χ(a) = sum of all roots of the characteristic

Pn polynomial of ψ(a) = 1 +2 +. . .+n ,
where j is a O(a)th root of unity. So, if n = i=1 i =⇒ 1 = 2 = . . . = n = In by the
triangle inequality.

Definition 7.1.2 (kernel of a character)

Let ker(χ) := {c ∈ G | χ(c) = χ(e)} be the kernel of χ (= ker ψ).

Now, let φ be some representation, χφ be its character. Then, for irreducible χi ’s, we have:
h
X
χφ = ni χi
i=1

A natural question to ask, then, is whether there exists any relation between ker(χφ ) and
ker(χi )? To find out, we work with matrix representations.
Note that ∀ a ∈ G, there exists some matrix M such that:

[φ1 (a)]
[φ2 (a)]
φ(a) = M −1 · .. ·M
.
[φl (a)]
T
where ψj are irreducibles. It follows then that ker(χφ ) = i ker(χi ).
In principle, one must be able to obtain all normal subgroups of G in this fashion. Why is
this the case? To see this, let N / G. Write G = G/N . We want a representation of G with
kernel N , but this is equivalent to finding a faithful representation of G. We have met such
a representation before: it’s nothing but the regular representation of G.

50
Advanced Abstract Algebra Fall 2018

So, we have: G  G → GLm (C) where the second map is one-to-one as it is the regular
representation. Thus, we have a representation of G whose kernel is precisely N .
We finally get to our main question: how to check if a function is a character
without using the representation?

Definition 7.1.3 (irreducible characters)

Irr(G) is the set of all irreducible characters of G.

Definition 7.1.4 (class functions)

We define cf(G) := {f | f : G → C : (∀x, y ∈ G)f (y −1 xy) = f (x)} and call it the set
of class functions of G.

Proposition 7.1.5 (irreducibles form basis)

Irr(G) is a basis of cf(G).

Corollary 7.1.6
If φ ∈cf(G), then φ is a character of (some representation of ) G if and only if ∀χ ∈
Irr(G), [χ, φ] is a nonnegative integer (and in addition φ is not the constant zero
function).

And finally, we have:

Definition 7.1.7 (induced function)

Let H ≤ G and let f : H → C be some function. Write fˆ : G → C such that:
(
f (b) b ∈ H
fˆ(b) =
0 b∈/H

Also define f G : G → C as follows:

1 X ˆ −1
f G (a) = f (x ax)
|H|
x∈G

We will call f G the function induced from H to G.

It is easy to see that f G is a class function of the group G. It’s also clear that induction is
linear, that is, if H ≤ G, and f1 , f2 : H → C, then (λ1 f1 + λ2 f2 )G = λ1 f1G + λ2 f2G . But
what we are concerned about is the case when f is a class function of H, and after the
following theorem, we will focus our attention on the case when f is a character of H.

Proposition 7.1.8 (reciprocity law)

Let φ ∈ cf (G), f : H → C. Then [φ, f G ] = [φ H , f ](H) .

51
Advanced Abstract Algebra Fall 2018

Proof. We have:
1 X
[φ, f G ](G) = φ(a) · f G (a)
|G|
a∈G
1 X 1 X ˆ −1
= φ(a) · f (x ax)
|G| |H|
a∈G x∈G
!
1 1 X X
= φ(a)fˆ(x ax)
−1
|G| |H|
a∈G x∈G
!
1 1 X X
= φ(x−1 ax)fˆ(x−1 ax)
|G| |H|
a∈G x∈G
!
1 1 X X
= φ(x−1 ax)fˆ(x−1 ax)
|G| |H|
x∈G a∈G

Now, focus on the inner sum. We have a ranging over all elements of G, and hence,
x−1 ax also ranges over all elements of G. Moreover, fˆ vanishes at elements outside of
H. Thus, we have: !
1 1 X X
[φ, f G ](G) = φ(a)f (a)
|G| |H|
x∈G a∈H

However, the inner product above is equal to |H| · [φ H , f ], and so:

1 1
[φ, f G ](G) = (|G| · |H| · [φ H , f ]) = [φ H , f ]
|G| |H|

and we’re done.

Proposition 7.1.9
If f is a character of H, then f G is a character of G.

Proof. This follows from the bilinearity of the inner product and the reciprocity law.

Indeed, if f is a character of H, then we have f G (e) = |G : H|f (e). It would be nice if the
induction of an irreducible character of H turned out to be an irreducible character of G,
but unfortunately, this is not true. The converse of this, however, holds.
It is not clear at this point what motivated the Frobenius reciprocity law, but one suspicion is
that we can use the Frobenius reciprocity to prove that the Frobenius kernel is a subgroup
of a Frobenius group. We will prove this. The first question, then, is to ask, what is a
Frobenius group?

7.2 Frobenius groups

Recall that any group can be thought of as a permutation group via its regular representa-
tion. When you think of a group as a permutation group, note that a non-identity element
has zero fixed points.

52
Advanced Abstract Algebra Fall 2018

Question. What about permutation groups in which non-identity elements have at most
one fixed point?
Answer. This is a silly question, since you can take any group and extend the underlying
set by an element that is fixed by all other elements.
Question. The same question as above, but now ourPgroup should also be transitive. Answer.
Recall that by Burnside’s Lemma, we have |G| = a∈G |f (a)|. If each non-identity element
has exactly one fixed point, then |G| = n + |G| − 1 =⇒ n = 1.
Thus, the above question is equivalent to asking for groups in which non-identity elements
have at most one fixed point, and ∃e 6= a ∈ G such that a does not fix anything.
Such groups are called Frobenius Groups. This seemingly slight modification of the
questions surprisingly adds a great amount of structure to the groups, as we will see soon.

Definition 7.2.1 (Frobenius Groups)

A group G ≤ Sn is called a Frobenius groups if:
1. G is transitive
2. No non-trivial element fixes more than one point
3. Some non-trivial element fixes a point

Recall that if two elements are in the same orbit, then their stabilizers are conjugates of
each other.

Proposition 7.2.2
Let G be a Frobenius group, and let α ∈ Ω (the underlying set on which G acts).
Then we have:
1. 1 ≤ Gα  G.
2. Let α 6= β ∈ Ω. Then Gα ∩ Gβ = {e}.

The above proposition gives us the abstract definition of a Frobenius group. Let H = Gα ,
/ H we have cHc−1 ∩ H = {e}.
such that 1  H  G. Then, if for every c ∈ G, c ∈
As an exercise, try to reconstruct original permutation group situation from the above
definition. You can do so by considering the permutation action on cosets of H.

Definition 7.2.3 (Frobenius complement)

H ≤ G as seen above (where G is a Frobenius group) is called a Frobenius complement.
That is, H ≤ G and for all a ∈ G\H we have H ∩ aHa−1 = {e}.

Definition 7.2.4 (Frobenius kernel)

In a Frobenius group G, write F = {e} {G\ c∈G cHc−1 }. F is called the Frobenius
S S
kernel.

In what follows, G, F, H are as above. You would think that normality is the surprising
part, but in fact, if we can prove that F is a subgroup of G, normality comes for free (since
the union of conjugates of H is invariant under conjugation, F must also be invariant under
conjugation and hence normal). So in fact, it is amazing that F is a subgroup at all!

53
Advanced Abstract Algebra Fall 2018

There is no known character-free proof that the Frobenius kernel is a subgroup, and most
purely group-theoretic proofs of special cases are very complicated.

Lemma 7.2.5
Let φ ∈ cf(H) such that φ(e) = 0. Then, the following holds: φG H = φ.

Proof. Recall that φG (e) = |G : H| · φ(e) =⇒ φG (e) = 0. Now, let e 6= h ∈ H. Then we

1 −1
have: φG (h) = |H|
P
x∈G φ̂(x hx).
Now, the only thing that matters is for which x ∈ G, we have x−1 hx ∈ H, since φ̂ is 0
for elements not in H. The equivalent question then is h ∈ xHx−1 , which is equivalent
to h ∈ H ∩ xHx−1 . Now, this happens for a non-identity element of H if and only if
x ∈ H because of the special property defining Frobenius groups.
Recall that φ is a class function, so φ(x−1 hx) = φ(h). So, we have:
1 X
φG (h) = φ(x−1 hx) = φ(h)
|H|
x∈H

which is the desired result.

Let us use the above lemma to construct another function: 1H 6=  ∈ Irr(H). Then define
φ = (e) · 1H − . Note that this is a class function of H and takes value 0 at e. On the
other hand, we have φG = ((e)1H )G − G by linearity of induction.

Lemma 7.2.6
[φG , φG ] = (e)2 + 1

Proof. We first use the reciprocity law:

[φG , φG ] = [φG H , φ] = [φ, φ][(e) · 1H − , (e) · 1H − ] = (e)2 + [, ] = (e)2 + 1

where we also utilized bilinearity of inner product and orthogonality of induced charac-
ters.

Lemma 7.2.7
[φG , 1G ] = (e)

Proof. First using reciprocity law:

[φG , 1G ] = [φ, 1H ] = [(e) · 1H − , (e) · 1h − ] = (e)[1h , 1H ] − [, 1H ] = (e)

using orthogonality and bilinearity, as in the previous lemma.

The above result gives us a lot about φG as a class function. We know now that if φG is
decomposed as a linear combination of characters, then we get the coefficient of the trivial
character. This can be written as:

Corollary 7.2.8P
φG = (e) · 1G + 1G 6=χ∈Irr(G) nχ · χ where nχ ∈ Z.

54
Advanced Abstract Algebra Fall 2018

We now compute the inner product of φG with itself using the form above, and compare
with the result of Lemma 7.2.6.

X X
[φG , φG ] = (e)2 + n2χ = (e)2 + 1 =⇒ 1 = n2χ
1G 6=χ∈Irr(G) 1G 6=χ∈Irr(G)

So, there exists a unique irreducible character χ 6= 1G such that n2χ = 1. So, is nχ = 1 or
−1? Actually, nχ = −1, and this is because φG vanishes at the identity.
From all of this, the following theorem will follow by considering kernels of χ obtained as
1H 6=  → χ 6= 1G .
Let’s now return to the Frobenius kernel F .

Theorem 7.2.9
F E G.

We also gave an alternative characterization of the Frobenius kernel F as the intersection

of “extensions” of irreducible characters of the Frobenius complement H.
I was absent for the above, but it can be found very easily online and in I.M. Isaacs’s
textbook.

7.3 The Second Orthogonality Relation

Fact. |Irr(G)| = number of conjugacy classes in G = dim(cf(G)).

Let M denote the character table of G where the columns denote the conjugacy classes of
G and the rows represent irreducible characters. Then we can get (M ·M T )i,j = [χi , χj ] =⇒
M · M T = In where n is the number of irreducible characters of G.
But M · M T = In =⇒ M T M = In , that is:
n p
T
X |Ki |Kj |
δi,j = (M M )i,j = χt (a + i)χt (aj ) ·
t=1
|G|

which means:
(
|G|
X
|K| if i = j; a, b ∈ K
χ(a)χ(b) =
χ∈Irr(G)
0 if a and b are not conjugates

|G|
where |K| = |G : Ga | = |G : C + G(a)| =⇒ |K| = |CG (a)|.
Thus, we have: (
X |CG (a)| if b ∼ a
χ(a)χ(b) =
χ∈Irr(G)
0 otherwise

55
Advanced Abstract Algebra Fall 2018

7.4 More Applications

G a finite group acting on Ω, a finite set. For g ∈ G, let f (g) := |{v ∈ Ω | g(v) = v}|. Then:
P
1. g∈G f (g) = |G| (# of all distinct orbits of G on Ω)

2. If the action is transitive then g∈G f (g)2 = |G| (# of all distinct orbits of Gα on Ω)
P

We discuss the above and recast them in the language of inner products:
1. [f, 1G ] = # of all orbits of G on Ω
2. If [f, 1G ] = 1 then [f, f ] = # of all orbits of G.
Suppose G acts transitively on Ω, Gα has 2 orbits on Ω: {α} and {Ω\α}. If this is the case,
we say that G acts 2-transitively on Ω. Let Ω[2] := {(β1 , β2 ) ∈ Ω2 | β1 6= β2 }. Then G acts
on Ω[2] in the following way:
a(β1 , β2 ) = (a(β1 ), a(β2 ))

We say that the action of G on ω is 2-transitive if the above action of G on Ω[2] is transitive.

Proposition 7.4.1
G acts 2-transitively on Ω iff G acts transitively on Ω, and if the number of orbits of
Gα is precisely 2 (∀/∃α ∈ Ω).

Example 7.4.2 (2-transitive actions)

Let V be a vector space over the field F. Let G = {fA,b | A ∈ GL(V ), b ∈ V } be the
set of affine transformations of V . The action of G on V is given by fA,b (v) = A(v)+b.
This defines a 2-transitive action.

Our goal for doing all of this:

Proposition 7.4.3
G a finite group, G acting on Ω such that |Ω| = n < ∞. Suppose that the action of G
on Ω is 2-transitive, and let H ≤ G such that |G : H| < n. Then H acts transitively
on Ω.
Proof. We can rewrite the given conditions as [f, f ] = 2 and [f, 1g ] = 1. Therefore, f is
a character (which should be obvious
P if you consider the permutation matrix associated
with group elements). So f = χ∈Irr(G) nχ χ where the nχ are nonnegative integers.
Moreover, nχ = [χ, f ].
Now we have: X
f = 1G + nχ χ
1G 6=χ∈Irr(G)

But [f, f ] = 2, and [f, f ] = 1+ 1G 6=χ∈Irr(G) n2χ . So f = 1G +χ for some 1G 6= χ ∈ Irr(G).

P

We need : [f H , 1H ] = 1. By the reciprocity law we get [f H , 1H ] = [f, (1H )G ].

Consider [1G , (1H )G ]. By the reciprocity law again, it is equal to [(1G ) H , 1H ] =
[1H , 1H ] = 1 and so, 1G is a constituent character ofP (1H )G .
The last statement can be rewritten as (1H )G = 1G + 1g 6=χ∈Irr(G) tχ χ =⇒ [f, (1H )G ] =

56
Advanced Abstract Algebra Fall 2018
P
[f, 1G ] + [f, tχ χ].PThe first term [f, 1G ] = 1 because of transitivity.
Thus, we need [f, tχ χ] = 0. PNote that f = χρ for some representation ρ whose
dimension must be n. Similarly, tχ χ = χφ for some representation φ whose dimension
must be |G : H| − 1. P
Now, we evaluate
P [f, tχ χ] Pand use that f = 1G + χ which we obtained earlier. We
have [1G + χ, tχ χ] = [χ, Ptχ χ]. The dimension of χ = n − 1 since f = 1G + χ. We
also know dimension of the tχ χ is |G : H| − 1.
Now, the only way that the above inner product is greater than zero is if the coefficient
of χPin the proof is greater than 0, i.e. at least 1, but this means that χ is a constituent
of
P tχ χ. This is not possible, as dimension of χ is greater than the dimension of
tχ χ.
We now look at another application.
P
Let K be some conjugacy class in G. Suppose ωk = x∈K x. Then we have: ωA ωB =
P A,B A,B
K nK ωK where in the final sum the coefficients denote multiplicities. Hence nK ≥0
and is an integer. Let ψ be an irreducible representation of G.
So we have: λA I × λB I = K nA,B A,B
P P
K λK I =⇒ λA λB = K nK λ K .

Thus we have: P
T r(ψ̂(ωA )) T r( a∈A ψ(a)) χψ (a)|A|
λA = = =
χψ (e) χψ (e) χψ (e)

We try to get a single coefficient out of this using the second orthogonality relation. We
have:
|A||B| X A,B χ(x)|K|
χ(a)χ(b) = nK
χ(e)2 χ(e)
K

Let c ∈ C which is a conjugacy class. Then, multiplying the above expression by χ(c) we
get:
X |A||B| X X A,B χ(x)|K|χ(c)
χ(a)χ(b) 2
= nK
χ(e) χ(e)
χ∈Irr(G) χ∈Irr(G) K

Rewriting we get:
X |A||B| X A,B X
χ(a)χ(b) = nK |K| χ(x)χ(c)
χ(e)
χ∈Irr(G) K χ∈Irr(G)

Thing on the right only survives for K = C. Then we get the sum on the right to be
nA,B A,B
C |C||CG (c)| = nC |G| by the orbit-stabilizer theorem.

Rearranging everything gives us:

|A||B| X χ(a)χ(b)χ(c)
nA,B
C =
|G| χ(e)
x∈Irr(G)

[A,B]
Notation. We changed notation slightly across classes: nA,B
C is the same as nC in what
follows.

57
Advanced Abstract Algebra Fall 2018

What’s the use of all this? Well, it could be used to study the commutator subgroup,
which is the smallest normal subgroup whose factor group is abelian. The commutator
[A,B]
elements can be obtained from the expression we obtained for nC . We look at this next
by attempting to answer the following question.
Question. Given g ∈ G, is g a commutator?
Answer. Note that if g is a commutator, then g = aba−1 b−1 = a(ba1− b−1 ). So, g ∈ G
[A,A−1 ]
such that ∃A conjugacy class such that nC > 0. Here A−1 is the set of inverses of all
elements in the conjugacy class A. But this is equivalent to
X |G| [A,A−1 ]
× nC >0
|A|
A∈Conjugacy Classes

We now try to get a nicer form for the left hand side expression. Making use of the stuff
from last class:
X 1 |A||A−1 | X χ(a)χ(a−1 χ(c) X χ(c) X
|G| = |A|χ(a)χ(a−1 )
|A−1 | |G| χ(e) χ(e)
A x∈Irr(G) χ∈Irr(G) A

χ(a)χ(a−1 ) = |G|. Thus, the above condition is

P
The right-most expression is simply a∈G
equivalent to
X χ(c)
|G| >0
χ(e)
χ∈Irr(G)

and this can finally be written as

X χ(c)
> 0 ⇐⇒ c is a commutator
χ(e)
χ∈Irr(G)

What’s all the above good for? Here’s an example.

Corollary 7.4.4
If c is a commutator in G, a finite group, such that gcd(O(c), k) = 1 for some
Z 3 k > 0, then ck is a commutator.

This follows easily from the stuff we did earlier today, simply consider the diagonal matrix
representation of ρ(c) and look at its k th power. We don’t give a complete proof in class, but
intuitively it’s easy to see what’s happening. Refer to Isaac’s book for a detailed explanation.
“More or less, I guess, this is the end of the series of applications of characters and orthog-
onality relations. Maybe one more thing. . . ”
What can be said about values of group characters?
If a character is linear, then the values arising there are roots of unity. But what if it’s not
a linear character. Here’s a “very useless but funny” proposition.

58
Advanced Abstract Algebra Fall 2018

Proposition 7.4.5
If χ ∈ Irr(G), χ(e) > 1 =⇒ (∃b ∈ G) : χ(b) = 0.

Proof. Suppose the contrary. Then we have

[
G= Lc

where Lc = {ck | O(ck ) = O(c)}. Note that χ(ck ) 6= 0. What about ck ∈Lc χ(ck )?
Q
This is known to be an algebraic integer, but because of some result in Galois Theory to
do with field automorphisms it must be rations. It follows that it is an ordinary integer.
Then we have Y Y Y
|χ(ck )| ≥ 1 =⇒ ( |χ(ck )|) ≥ 1
ck ∈Lc all classes ck ∈Lc

Thus, we have Y
|χ(g)|2 ≥ 1
g∈G

But by the AM-GM inequality we have

!|G|
|χ(g)|2
P
g∈G
Y
1= ≥ |χ(g)2 | ≥ 1
|G|
g∈G

and so we have equality, and so all the elements are equal and so χ(e) = 1, giving a
contradiction.
The purpose of the above was to see a cool application of some sophisticated techniques.

59