Group theory for Maths, Physics and Chemistry

students

Arjeh Cohen
Rosane Ushirobira
Jan Draisma

October 18, 2007

2
Contents

1 Introduction 5
1.1 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Permutation Groups 17
2.1 Cosets and Lagrange . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Quotient groups and the homomorphism theorem . . . . . . . . . 18
2.3 Loyd’s puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Action of groups on sets . . . . . . . . . . . . . . . . . . . . . . . 28

3 Symmetry Groups in Euclidean Space 33

3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Isometries of n-space . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 The Finite Subgroups of O(2, R) . . . . . . . . . . . . . . . . . . 40
3.4 The Finite Subgroups of O(3, R) . . . . . . . . . . . . . . . . . . 42

4 Representation Theory 51
4.1 Linear representations of groups . . . . . . . . . . . . . . . . . . 51
4.2 Decomposing Displacements . . . . . . . . . . . . . . . . . . . . . 59

5 Character Tables 63
5.1 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Orthogonality of irreducible characters . . . . . . . . . . . . . . . 64
5.3 Character tables . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.4 Application: symmetry of vibrations . . . . . . . . . . . . . . . . 77
5.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

6 Some compact Lie groups and their representations 81

6.1 Some examples of Lie groups . . . . . . . . . . . . . . . . . . . . 81
6.2 Representation theory of compact Lie groups . . . . . . . . . . . 84
6.3 The 2-dimensional unitary group . . . . . . . . . . . . . . . . . . 86
6.4 The 3-dimensional orthogonal group . . . . . . . . . . . . . . . . 88

3
4 CONTENTS
Chapter 1

Introduction

1.1 Symmetry
Group theory is an abstraction of symmetry
Symmetry is the notion that an object of study may look the same from different
points of view. For instance, the chair in Figure 1.1 looks the same as its
reflection in a mirror that would be placed in front of it, and our view on the
wheel depicted next to the chair doesn’t change if we rotate our point of view
over π/6 around the shaft.
But rather than changing viewpoint ourselves, we think of an object’s sym-
metry as transformations of space that map the object ‘into itself’.
What do we mean when we say that an object is symmetric? To answer this
question, consider once more the chair in Figure 1.1. In this picture we see a
plane V cutting the chair into two parts. Consider the transformation r of three-
dimensional space that maps each point p to the point p0 constructed as follows:

Figure 1.1: Bilateral and rotational symmetry. From: [8].

5
6 CHAPTER 1. INTRODUCTION

let l be the line through p and perpendicular to V . Denoting the distance of a

point x to V by d(x, V ), p0 is the unique point on l with d(p0 , V ) = d(p, V ) and
p 6= p0 . The map r is called the reflection in V . Now apply r to all points in the
space occupied by the chair. The result is a new set of points, but because of
the choice of V it equals the old space occupied by the chair. For this reason, we
call the chair invariant under r, or say that r is a symmetry (transformation)
of the chair.
Transforming any point in space to its reflection in the mirror W and next
rotating it over an angle of π around the axis V ∩W , gives the same result as the
reflection r in V . This is an illustration of what we observed about symmetry:
the change of view on the chair from the original chair to its mirror image, was
‘neutralized’ to the original look, but then on a transformed chair (transformed
by means of r). Symmetry then is the phenomenon that the image chair is
indistinguishable from the original chair.
Next, consider the wheel depicted in Figure 1.1. It also has a symmetry,
but of a different type: the rotation s around the wheel’s axis m over 2π/6
moves all points occupied by the wheel to other points occupied by it. For this
reason we call s a symmetry of the wheel. This type of symmetry is called
rotational symmetry. Note that the rotations around m over the angles 2kπ/6
are symmetries of the wheel as well, for k = 1, 2, 3, 4, 5. The wheel also has
reflection symmetry: the reflection t in the plane W perpendicular to m and
cutting the wheel into two thinner wheels of the same size also maps the wheel
to itself.
Symmetry registers regularity, and thus records beauty. But it does more
than that, observing symmetry is useful. Symmetry considerations lead to effi-
ciency in the study of symmetric objects. For instance,

• when designing a chair, the requirement that the object be symmetric may
reduce drawings to half the chair;

• when studying functions on the plane that respect rotational symmetry

around a ‘centre’ p in the plane, we can reduce the number of arguments
from the (usual) two for an arbitrary function on the plane to one, the
distance to p.

Symmetry conditions
A few more observations can be made regarding the previous cases.

1. A symmetry of an object or figure in space is a transformation of space

that maps the object to itself. We only considered the space occupied by
the chair and the wheel, respectively, and not the materials involved. For
example, if the left legs of the chair are made of wood and the right ones
of iron, one could argue that the chair is not symmetric under reflection
in V . Usually, we shall speak about symmetries of sets of points, but this
subtlety should be remembered.
1.1. SYMMETRY 7

Figure 1.2: The Hexagon

2. Symmetry transformations are bijective, i.e., they have an inverse trans-

formation. In the case of r, it is r itself. In the case of s, it is rotation
about m over −2π/6, or, which amounts to the same, 10π/6.

3. Two transformations can be composed to obtain another transformation;

if both translations are symmetries of some object, then so is their com-
position. In the case of s, we see that s ◦ s (s applied twice) is a symmetry
of the wheel as well; it is rotation over the angle 2 · (2π/6). Also, t ◦ s
(first apply s, then t) is a symmetry of the wheel. Usually, we will leave
out the composition sign ◦ and write ts instead of t ◦ s, and s2 for ss.

4. There is one trivial transformation which is a symmetry of any object,

namely the identity, sending each point to itself. It is usually denoted by
e (but also by 0 or 1, see below).

Groups
The transformations under which a given object is invariant, form a group.
Group theory was inspired by these types of group. However, as we shall see,
‘group’ is a more general concept. To get a feeling for groups, let us consider
some more examples.

Planar groups
The hexagon, as depicted in Figure 1.2, is a two-dimensional object, lying in
the plane. There are lots of transformations of the plane leaving it invariant.
For example, rotation r around 0 over 2π/6 is one of them, as is reflection s in
the vertical line l through the barycentre. Therefore, the hexagon has at least
8 CHAPTER 1. INTRODUCTION

Figure 1.3: A frieze pattern. From: [11].

the following symmetries:

e, r, r2 , r3 , r4 , r5 , s, sr, sr2 , sr3 , sr4 , sr5 .

In fact, these are all symmetries of the hexagon. Thus, each is a (multiple)
product of r and s. These two elements are said to generate the group of
symmetries.
Exercise 1.1.1. Verify that the elements listed are indeed distinct and that

r6 = s2 = e, and srs = r5 .

Exercise 1.1.2. Show that sr is a reflection. In which line?

Figure 1.3 shows the palmette motif, which is very frequently used as a dec-
oration on the upper border of wallpaper. Imagine this pattern to be repeated
infinitely to the left and right. Then the translation t over the vector pointing
from the top of one palmette to the top of the second palmette to its right is a
symmetry of the pattern, and so is the reflection r in the vertical line l through
the heart of a given palmette. With these two, all symmetries of the frieze
pattern can be listed. They are:

e, t, t2 , t3 , . . . and r, rt, rt2 , rt3 , . . . .

We find that the symmetry group of the pattern is infinite. However, when
ignoring the translational part of the symmetries, we are left with essentially
only two symmetries: e and r.
Exercise 1.1.3. Find frieze patterns that have essentially different symmetry
groups from the one of Figure 1.3.
A pattern with a more complicated symmetry group is the wallpaper pattern
of Figure 1.4, which Escher designed for ‘Ontmoeting’ (Dutch for ‘Encounter’).
It should be thought of as infinitely stretched out vertically, as well as hori-
zontally. The translations s and t over the vectors a (vertical from one nose
1.1. SYMMETRY 9

Figure 1.4: Wallpaper pattern for Ontmoeting. From: M.C.Escher, 1944.

to the next one straight above it) and b (horizontal, from one nose to the next
one to its right) leave the pattern invariant, but this is not what makes the
picture so special. It has another symmetry, namely a glide-reflection g, which
is described as: first apply translation over 21 a, and then reflect in the vertical
line l equidistant to a left oriented and a neighbouring right oriented nose. Note
that g 2 = s.

Exercise 1.1.4. List all symmetries of the pattern in Figure 1.4.

Space groups
Having seen symmetries of some 2-dimensional figures, we go over to 3-dimen-
sional ones.
The cube in Figure 1.5 has lots of symmetries. Its symmetry group is gen-
erated by the rotations r3 , r4 , r40 whose axes are the coordinate axes, and the
reflection t through the horizontal plane dividing the cube in two equal parts.

Exercise 1.1.5. What is the size of the symmetry group of the cube?

Closely related is the methane molecule, depicted in Figure 1.6. Not all of
the symmetries of the cube are symmetries of the molecule; in fact, only those
that map the hydrogen atoms to other such atoms are; these atoms form a
regular tetrahedron.

Exercise 1.1.6. Which of the symmetries of the cube are symmetries of the
methane molecule?

Finally consider the cubic grid in Figure 1.7. Considering the black balls to
be natrium atoms, and the white ones to be chlorine atoms, it can be seen as the
structure of a salt-crystal. It has translational symmetry in three perpendicular
10 CHAPTER 1. INTRODUCTION

Figure 1.5: The cube.

Figure 1.6: The methane molecule. From: [7].

1.1. SYMMETRY 11

Figure 1.7: Structure of a crystal

directions along vectors of the same size. To understand its further symmetry,
consider all those symmetries of the crystal that leave a given natrium atom
fixed. Such a symmetry must permute the 6 chlorine atoms closest to the
given natrium atom. These atoms form a regular octahedron, as depicted in
Figure 1.8, and therefore we may say that the symmetry group of salt crystals
is generated by that of the regular octahedron and three translations.

Exercise 1.1.7. This exercise concerns symmetries of one-dimensional figures.

Considere therefore the real line R with the Euclidean metric, in which the
distance between x, y ∈ R is |x − y|. A map g from R to itself is called an
isometry if |g(x) − g(y)| = |x − y| for all x, y ∈ R.

1. Show that any isometry of R is either of the form sa : x 7→ 2a−x (reflection

in a) or of the form ta : x 7→ x + a (translation over a).

2. Prove that {e} and {e, s0 }, where e denotes the identity map on R, are
the only finite groups of isometries of R.

To describe symmetries of discrete subsets of R, we restrict our attention to

groups G whose translational part {a ∈ R | ta ∈ G} equals Z.

3. Show that if such a group G contains s0 , then

G = {ta | a ∈ Z} ∪ {t a2 | a ∈ Z}.

4. Conclude that there are only two ‘essentially different’ groups describing
the symmetries of discrete subsets of R having translational symmetry.
12 CHAPTER 1. INTRODUCTION

Figure 1.8: The octahedron.

Exercise 1.1.8. Analyse the symmetry groups of discrete subsets of the strip
R × [−1, 1] in a manner similar to that of Exercise 1.1.7. Compare the result
with your answer to Exercise 1.1.3.

1.2 Basic notions

Formal definition of a group
Having seen some examples of groups, albeit from the narrow point of view of
symmetry groups of figures in the plane or in three-dimensional space, we are
ready for some abstract definitions that should reflect the experiences we had
in those examples.
Definition 1.2.1. A group is a set G, together with an operation ·, which maps
an ordered pair (g, h) of elements of G to another element g · h of G, satisfying
the following axioms.
1. The operation is associative, i.e., for all g, h, k ∈ G we have g · (h · k) =
(g · h) · k.
2. G contains an identity element, i.e., an element e that satisfies e · g =
g · e = g for all g ∈ G.
3. Each element of G has an inverse, i.e., for each g ∈ G there is an h ∈ G
such that g · h = h · g = e. This element is denoted by g −1 .
The cardinality of G is called the order of the group, and often denoted by
|G|; it may be infinite. If the operation is not only associative but commutative
as well (meaning g · h = h · g for all g, h ∈ G), then G is called an Abelian group.
1.2. BASIC NOTIONS 13

Exercise 1.2.2. Prove that a group G cannot have more than one identity.
Also, the notation g −1 for the inverse of g seems to indicate the uniqueness of
that inverse; prove this.
Let us give some examples of groups.
Example 1.2.3.
1. All sets of transformations found in §1.1 form groups; in each case, the
composition ◦ serves as operation ·. Some of them are Abelian; like the
the chair, and some aren’t, like those of the cube.
2. The real numbers R form a group with respect to addition +, the unit
element being 0. They do not form a group with respect to ordinary
multiplication, as 0 does not have an inverse. Leaving out 0 we do obtain
an Abelian group (R \ {0}, ·), the unit element being 1.
3. Also Q and Z form additive groups. Again Q \ {0} is a multiplicative
group. However, Z cannot be turned into a multiplicative group, the only
invertible elements being {±1}.
4. Let X be a set. The bijections of X form a group. Composition is again the
group operation. The group inverse coincides with the ‘functional’ inverse.
The group is often denoted by Sym(X). If X is finite, say |X| = n, then
|Sym(X)| = n!. If two sets X and Y have the same cardinality, then the
groups Sym(X) and Sym(Y ) are essentially the same. For finite sets of
cardinality n, the representative Sym({1, ..., n}) of this class of groups is
denoted by Sym(n) or Sn .
Its element are also called permutations. For X = {1, 2, 3}, we have

Sym(X) = {e, (1, 2), (2, 3), (1, 3), (1, 2, 3), (3, 2, 1)}.

This notation is explained in Section 2.3.

5. Let V be a vector space. The set of all linear transformations of V , often
denoted by GL(V ), together with composition of maps, is a group. By
fixing a basis of V , its element can be written as invertible matrices.
Let us make notation easier.
Definition 1.2.4. Usually, we leave out the operation · if it does not cause
confusion. However, if the operation is written as +, it is never left out. For g
in a group G we write g n for the product

g · g · ... · g.
| {z }
n

Our first lemma concerns the inverse of a product.

Lemma 1.2.5. For two elements g, h in a group G we have (gh)−1 = h−1 g −1 ,
and (g n )−1 = (g −1 )n . For the latter we also write g −n .
14 CHAPTER 1. INTRODUCTION

This is very natural, as Weyl points out in [11]:

‘When you dress, it is not immaterial in which order you perform

the operations; and when in dressing you start with the shirt and
end up with the coat, then in undressing you observe the opposite
order; first take off the coat and the shirt comes last.’

Exercise 1.2.6. Show that the notation for fractions which is usual for integers,
does not work here: hg could stand for g −1 h as well as for hg −1 , and the latter
two may differ.

With the methane molecule in Figure 1.6, some of the symmetries of the
cube form a smaller group, namely the symmetry group of the tetrahedron. We
found an example of a subgroup.

Definition 1.2.7. A subgroup of a group G is a subset H of G satisfying the

following conditions.

1. H is non-empty.

2. H is closed under the operation, i.e., for all g, h ∈ H we have gh ∈ H.

3. H is closed under taking inverses. This means that, for all h ∈ H, we have
h−1 ∈ H.

Exercise 1.2.8. Prove that a subgroup of a group contains the group’s identity.
Prove that for a subset H of G to be a subgroup of G it is necessary and sufficient
that H be non-empty and that gh−1 ∈ H for all g, h ∈ H.

Exercise 1.2.9. Prove that for a subset H of a finite group G to be a sub-

group, it is necessary and sufficient that H be non-empty and closed under the
operation of G.

Generation
The following lemma is needed to formalize the concept of a subgroup generated
by some elements.

Lemma 1.2.10. The intersection of a family of subgroups of a group G is a

subgroup of G.

Definition 1.2.11. As a consequence of Lemma 1.2.10, for a subset S of G

the intersection of all groups containing S is a subgroup. It is denoted by hSi
and called the subgroup generated by S. Instead of h{g1 , . . . , gn }i we also write
hg1 , . . . , gn i. The elements g1 , . . . , gn are called generators of this group.

Some groups are generated by a single element. For example, Z is gener-

ated by 1, and the symmetry group of the triquetrum depicted in Figure 1.9 is
generated by a rotation around its center over 2π/3.
1.2. BASIC NOTIONS 15

Figure 1.9: The triquetrum.

(0,1) (1,1)

(0,0) (1,0)

Figure 1.10: The flat torus.

Definition 1.2.12. If there exists an element g ∈ G for which G = hgi, then G

is called a cyclic group.
Consider the integers modulo n, where n is a positive integer. In this set,
which is denoted by Z/nZ for reasons that will become clear later, two integers
are identified whenever their difference is a multiple of n. This set forms a
group, addition modulo n being the operation. The order of Z/nZ is n, and it
is a cyclic group. In fact, one could say that it is the cyclic group of this order,
as all such groups are the same in a sense to be made precise later. Thus the
symmetry group of the triquetrum (Figure 1.9) is ‘the same’ as Z/3Z.
Definition 1.2.13. Let g be an element of a group G. The order of hgi is called
the order of g. It may be infinite.
Exercise 1.2.14. Prove that a cyclic group is Abelian.
Exercise 1.2.15. Consider the flat torus T . It can be seen as the square
S = [0, 1] × [0, 1] with opposite edges glued together, such that the arrows in
Figure 1.10 match. (Thus, the set is really in bijection with [0, 1) × [0, 1).)
The torus has the following group structure: given two points p = (p1 , p2 ) and
q = (q1 , q2 ) on the torus, their sum corresponds to the point r on S with the
property that r − (p1 + q1 , p2 + q2 ), the latter term of which may not be in S,
has integral entries.
Determine which elements of T have finite order, and which elements have
infinite order.
16 CHAPTER 1. INTRODUCTION
Chapter 2

Permutation Groups

The realization of a group by means of symmetries (more generally, bijections

of a set) is closely related to the structure of its subgroups. Let us first focus
on the latter aspect.

2.1 Cosets and Lagrange

Recall the notion of subgroup from Definition 1.2.7.

Definition 2.1.1. For subsets S and T of a group G, we write ST for the set

{st | s ∈ S, t ∈ T }.

For s ∈ G and T ⊂ G we also write sT for {s}T and T s for T {s}.

Definition 2.1.2. For a subgroup H of G the sets of the form gH with g ∈ G

are called left cosets of H in G, and those of the form Hg are called right cosets
of H in G. The set of all left cosets of H in G is denoted by G/H.

It would be logical to denote the set of right cosets of H in G by H \ G.

Indeed, this is sometimes done, but this meaning of \ should not be confused
with the usual set-theoretic one. We shall only use G/H in these notes.

Example 2.1.3.

1. G = Z, the additive group, and H = hmi = mZ, the subgroup generated

by m, for some m ∈ Z. The set of left cosets is {H, 1 + H, 2 + H, . . . , m −
1 + H}. Since x + H = H + x, left cosets coincide with right cosets.

2. G = S3 = {(1), (12), (13), (23), (123), (132)}, the symmetry group of an

equilateral triangle (check). Take the subgroup H = h(123)i. Then the
set of left cosets is {H, (12)H}, where (12)H = {(12), (13), (23)}. Observe
that (12)H = H(12) = G \ H, so left and right cosets coincide.

17
18 CHAPTER 2. PERMUTATION GROUPS

3. G = Rn , the (additive) translation group of a real vector space and H is a

linear subspace of G. Then the left cosets are the affine subspaces parallel
to H.

4. A left coset need not be a right coset. Take for instance G = S3 and
H the cyclic subgroup h(12)i. Then (13)H = {(13), (123)} and H(13) =
{(13), (132)}.

It is easy to verify that left cosets are equivalence classes of the equivalence
relation
g ∼ h ⇔ g −1 h ∈ H,
and similarly for right cosets. So we have the following lemma.

Lemma 2.1.4. The left cosets of a subgroup H in the group G form a partition
of the group G. For g ∈ G, the map h 7→ gh is a bijection between H and the
left coset gH.

A similar statement holds for right cosets.

Exercise 2.1.5. Show that (13)h(12)i is not a right coset of h(12)i in S3 , and
that (12)h(123)i is a right coset of h(123)i in S3 .

Definition 2.1.6. If the number of left cosets of a subgroup H in a group G is

finite, this number is called the index of H in G and denoted by [G : H].

Example 2.1.7. In the previous examples, we have [Z : hmi] = m and

[S3 : h(123)i] = 2 .

Theorem 2.1.8 (Lagrange’s Theorem). If G is a finite group and H a subgroup,

then |H| divides |G|.

Corollary 2.1.9. If G is a finite group and g ∈ G, then the order of g divides

that of G.

Exercise 2.1.10. If H and K are subgroups of a finite group G, then H ∩ K

|H||K|
is also a subgroup of G. Prove that |HK| = |H∩K| .

2.2 Quotient groups and the homomorphism the-

orem
If two groups G and H and some of their elements are in the picture, we shall
always make it clear to which of the two groups each element belongs. This
enables us to leave out the operation symbol throughout, as the product of two
elements of H is automatically understood to denote the product in H and not
in G. Also, we shall use e for the unit element of both groups.
Next, we will see the functions between two groups that preserve the struc-
ture of groups.
2.2. QUOTIENT GROUPS AND THE HOMOMORPHISM THEOREM 19

Definition 2.2.1. Let G and H be groups. A homomorphism from G to H is

a map φ : G → H satisfying

φ(g1 g2 ) = φ(g1 )φ(g2 ) for all g1 , g2 ∈ G,

and φ(eG ) = eH . If, in addition, φ is a bijection, then φ is called an isomorphism.

An automorphism of G is an isomorphism from G to itself. By Aut(G) we denote
the set of all automorphisms of G.

If there exists an isomorphism from a group G to a group H, then G is

called isomorphic to H. this fact is denoted by G ∼= H, and it is clear that ∼
=
is an equivalence relation. We claimed earlier that the symmetry group of the
triquetrum (Figure 1.9) is ‘the same as’ Z/3Z. The precise formulation is, that
both groups are isomorphic.

Definition 2.2.2. For a group homomorphism φ : G → H the subset

{g ∈ G | φ(g) = e}

of G is called the kernel of φ, and denoted by ker(φ). The set

φ(G) = {φ(g) | g ∈ G} ⊆ H

is called the image of φ, and denoted by im(φ).

Exercise 2.2.3. Let φ : G → H be a homomorphism. Then φ is injective if

and only if ker φ = {e}.

Example 2.2.4. For g ∈ G, we define φg : G → G by φg (h) = ghg −1 . Then

φg is an element of Aut(G), called conjugation with g. An automorphism of G
is called an inner automorphism if it is of the form φg form some g; it is called
outer otherwise.

Exercise 2.2.5. Let φ be a group homomorphism from G to H. Prove that

ker(φ) and im(φ) are subgroups of G and H, respectively. Show that for every
g ∈ G we have g ker(φ)g −1 ⊆ ker(φ).

Exercise 2.2.6. Prove:

1. Aut(G) is a group with respect to composition as maps;

2. the set Inn(G) = {φg | g ∈ G} is a subgroup of Aut(G). Here φg denotes

conjugation with g.

Example 2.2.7. For G = GL(V ), the group of invertible linear transformations

of a complex vector space V , the map g 7→ det(g) is a homomorphism G → C∗ .
Its kernel is the group SL(V ).

Example 2.2.8. The map Z → Z/mZ, x 7→ x + mZ, is a homomorphism of

additive groups. Its kernel is mZ.
20 CHAPTER 2. PERMUTATION GROUPS

Example 2.2.9. There is an interesting homomorphism between SL(2, C) and

the Lorentz group L, which is the group of all linear transformations of the
vector space R4 that preserve the Lorentz metric

|x| := x20 − x21 − x22 − x23 .

To x ∈ R4 we associate a 2 × 2-matrix ψ(x) as follows:

 
x0 + x3 x1 − ix2
ψ(x) = ,
x1 + ix2 x0 − x3

so that |x| = det(ψ(x)). Now, the map ϕ : SL(2, C) → L given by ϕ(A)(x) =

ψ −1 (Aψ(x)A∗ ) is a group homomorphism. Here A∗ = (āji )ij if A = (aij )ij .
Indeed, ψ is a linear isomorphism from R4 onto
  
4 x y 
ψ(R ) = x, y, z, u ∈ C and x̄ = x, ȳ = z, ū = u ,
z u 

and one checks that this space is invariant under M 7→ AM A∗ for any A ∈
GL(2, C). Also, for A ∈ SL(2, C), ϕ(A) preserves the metric due to the multi-
plicative properties of the determinant:

|ϕ(A)(x)| = det(Aψ(x)A∗ ) = det(A) det(ψ(x)) det(A∗ ) = det(ψ(x)) = |x|.

Exercise 2.2.10. Show that ϕ in Example 2.2.9 satisfies ϕ(AB) = ϕ(A)ϕ(B),

and thus finish the proof that ϕ is a homomorphism SL(2, C) → L.

Exercise 2.2.11. Determine the subgroups of (Z/2Z) ⊕ (Z/2Z). Is (Z/2Z) ⊕

(Z/2Z) isomorphic to Z/4Z?

Exercise 2.2.12. Take the group Z/4Z, the subgroup h(1234)i of S4 and the
multiplicative subgroup generated by i in C. Verify that they are all isomorphic.

As we have seen before, it is not always the case that left cosets and right
cosets of a subgroup H are the same. When that happens, we say that H is
normal.

Definition 2.2.13. A subgroup H of G is called normal if every left coset of

H in G is a right coset of H in G, or, equivalently, if gHg −1 ⊂ H for all g ∈ G.
We denote it by H E G.

If H is normal, then we can define an operation on the set G/H of all left
cosets of H in H, as follows:

(gH) · (kH) = (gk)H.

This operation turns G/H into a group, called the quotient group of G by H.
The map G 7→ G/H, g 7→ gH is a homomorphism G → G/H.
2.2. QUOTIENT GROUPS AND THE HOMOMORPHISM THEOREM 21

Exercise 2.2.14. Let H be a normal subgroup of the group G (cf. Exercise

2.2.13) Verify that the operation on G/H is well-defined, i.e., does not depend
on the choice of representatives g and k. Verify that G/H does indeed become
a group with this operation. Note that one really needs normality of H.

Exercise 2.2.15. Prove that Inn(G) (cf. Example 2.2.4) is a normal subgroup
of Aut(G) (cf. Definition 2.2.1).

Example 2.2.16.

1. The subgroup h(12)(34), (13)(24)i is normal in S4 .

2. Let G be the group of Euclidean motions in R3 (see Chapter 3). Write an

element of G as  
a v
0 1
with a ∈ SO(3) and v ∈ R3 . This presentation is convenient as the oper-
ation in G is now given by the multiplication of 4 × 4 matrices. Consider
the subgroup N of all translations; thus, N consists of all matrices
 
I v
.
0 1

Then N is a normal subgroup of G.

Lemma 2.2.17. The kernel of a group homomorphism φ : G → H is a normal

subgroup of G.

See Exercise 2.2.5 for the proof.

Exercise 2.2.18. Prove that any subgroup of index 2 is normal.

Theorem 2.2.19. Let φ : G → H be a group homomorphism. Then G/ ker φ ∼

=
im φ.

Proof. Let N := ker φ. We claim that π : G/N → im φ, gN 7→ φ(g) is an

isomorphism. Indeed, it is well-defined: if g 0 N = gN , then g −1 g 0 ∈ N so that
φ(g −1 g 0 ) = e, hence φ(g 0 ) = φ(g). Also, π is a homomorphism:

π((gN )(kN )) = π((gk)N ) = φ(gk) = φ(g)φ(k) = π(gN )π(kN ).

Finally, if π(gN ) = e, then φ(g) = e, hence gN = eN ; it follows that ker(π) =

{eN }, and Exercise 2.2.3 shows that π is injective. It is also surjective onto its
image, hence it is an isomorphism.

Theorem 2.2.20 (First Isomorphism Theorem). Let G be a group, and H, K

normal subgroups of G such that H ⊇ K. Then K is a normal sugroup of H,
H/K a normal subgroup of G/K, and G/H ∼ = (G/K)/(H/K).
22 CHAPTER 2. PERMUTATION GROUPS

A B C D

E F G H

I J K L

M N O

Figure 2.1: Initial state of Sam Loyd’s 15-puzzle.

Proof. For any g ∈ G, we have gKg −1 ⊆ K, so a fortiori this holds for

g ∈ H. Hence, K is normal in H. For any g ∈ G and h ∈ H, we have
(gK)(hK)(gK)−1 = (ghg −1 )K ∈ H/K as H is normal in G. Hence, H/K is
normal in G/K.
Consider the map π : G → (G/K)/(H/K), g 7→ (gK)(H/K). It is clearly a
surjective homomorphism, and its kernel equals {g ∈ G | gK ∈ H/K} = H. By
Theorem 2.2.19, G/H ∼= (G/K)/(H/K) as claimed.
Theorem 2.2.21 (Second Isomorphism Theorem). Let G be a group, H a
subgroup of G, and K a normal subgroup of G. Then HK is a subgroup of G,
K a normal subgroup of HK, H ∩ K is a normal subgroup of H, and HK/K ∼=
H/(H ∩ K).
Exercise 2.2.22. Prove the second Isomorphism Theorem.

2.3 Loyd’s puzzle

A Messed-Up Puzzle
Suppose that a friend borrows your 15-puzzle just after you solved it, so that it
is in the initial state depicted in Figure 2.1. You leave the room to make some
coffee and when you come back your friend returns the puzzle to you, sighing
that it is too difficult for him, and asking you to show him how to solve it. At
this time, the puzzle is in the state of Figure 2.2. Confidently you start moving
around the plastic squares; for readers unfamiliar with the puzzle a possible
sequence of moves is shown in Figure 2.3. As time passes, you start getting
more and more nervous, for your friend is watching you and you don’t seem to
be able to transform the puzzle back into its initial state. What is wrong?
To answer this question, we shall look at some basic properties of permuta-
tions.
2.3. LOYD’S PUZZLE 23

A B C D

E F G H

I J K L

M O N

Figure 2.2: Messed up state of the 15-puzzle.

A B C D A B C D A B C D

E F G H E F G H E F G H

I J K L I J K I J K

M O N M O N L M O N L

Figure 2.3: Possible moves of the 15-puzzle.

24 CHAPTER 2. PERMUTATION GROUPS

Permutations and their Signs

Recall the definition of a permutation and the symmetric groups Sn . Permuta-
tions can be represented in several ways, three of which we shall discuss here.

Two-row notation: for a permutation π : {1, . . . , n} → {1, . . . , n} one writes

a 2 × n matrix, in which both rows contain the numbers 1, . . . , n. The
image of i is the number that appears below i in the matrix. An example
is  
1 2 3 4 5
π= ∈ S5 .
5 4 1 2 3

Disjoint cycle notation: To explain this notation, consider the permutation

π from the previous example. We have π(1) = 5, π(5) = 3 and π(3) equals
1 again. The permutation that satisfies these equations and that fixes the
other elements is denoted by

(1, 5, 3) = (5, 3, 1) = (3, 1, 5).

A permutation of this form is called a 3-cycle. But π does not fix 2 or 4;

indeed, π(2) = 4 and π(4) = 2. We find that

π = (1, 5, 3)(2, 4) = (2, 4)(1, 5, 3),

i.e., π is the product of the 2-cycle (2, 6) and the 3-cycle (3, 5, 1). These
are called disjoint, because each fixes the elements occurring in the other,
and this fact makes them commute. Any permutation can be written as a
product of disjoint cycles, and this factorization is unique up to changes
in the order. In this notation, it is usual not to write down the 1-cycles.
Leaving these out, it is not always clear to which Sn the permutation
belongs, but this abuse of notation turns out to be convenient rather than
confusing.
Permutation matrix notation: consider once again the permutation π ∈ S5 .
With it we associate the 5 × 5-matrix Aπ = (aij ) with
(
1 if π(j) = i,
aij =
0, otherwise.

It can be checked that Aπσ = Aπ Aσ , where we take the normal matrix

product in the right-hand side. Hence Sn can be seen as a matrix group,
multiplication being the ordinary matrix multiplication.
Diagram notation: in this notation, a permutation is represented by a dia-
gram as follows: draw two rows of dots labeled 1, . . . , n, one under the
other. Then draw a line from vertex i in the upper row to vertex π(i) in
the lower. In our example, we get the picture in Figure 2.4.
2.3. LOYD’S PUZZLE 25

1 2 3 4 5

1 2 3 4 5

Figure 2.4: The diagram notation.

Exercise 2.3.1. Show that the 2-cycles in Sn generate that whole group. Can
you give a smaller set of 2-cycles that generates the whole group? How many
do you need?

Exercise 2.3.2. Determine the order of a k-cycle. Determine the order of π =

(2, 6)(1, 5, 3). Give and prove a general formula for the order of a permutation,
given in disjoint cycle notation, in terms of the lengths of the cycles.

Related to permutation matrices we have the following important definition.

Definition 2.3.3. For π ∈ Sn we define the sign of π by

sgn π := det Aπ .

The sign of a permutation π is either 1 or −1. In the former case, π is called an

even permutation, in the latter an odd permutation.

Exercise 2.3.4. Show that sgn : Sn → {±1} is a group homomorphism.

The sign of a permutation can be interpreted in many ways. For example,

in the diagram notation, the number of intersections of two lines is even if and
only if the permutation is even.

Exercise 2.3.5. Prove that the sign of a k-cycle equals (−1)k+1 .

Exercise 2.3.6. Prove that the sign of a permutation equals (−1)N where N
is the number of crossings of the lines drawn in the diagram notation of the
permutation.

The even permutations in Sn form a subgroup, called the alternating group

and denoted by An . It is the kernel of the homomorphism sgn : Sn → {±1} of
Exercise 2.3.4.
26 CHAPTER 2. PERMUTATION GROUPS

1 2 3 4

5 6 7 8

9 10 11 12

13 14 15 16

Figure 2.5: Numbering of the positions.

No Return Possible
Let us return to Sam Loyd’s 15-puzzle. It can be described in terms of states
and moves. The Figures 2.1 and 2.2 depict states of the puzzle. Formally, a
state can be described as a function

s : {1, . . . , 16} → {A, B, . . . , O, ∗}.

Here, s(p) is the square occupying position p in the state s and ∗ stands for the
empty square. The positions are numbered as in Figure 2.5.
For p ∈ {1, . . . , 16}, we denote by M (p) ⊆ S16 the set of possible moves in a
state s with s(p) = ∗. A move m ∈ M (p) should be interpreted as moving the
square from position j to position m(j), for j = 1, . . . , 16. For example

M (1) = {(1, 2), (1, 5)},

M (2) = {(1, 2), (2, 3), (2, 6)}, and
M (6) = {(2, 6), (5, 6), (6, 7), (6, 10)}.

Moves can be combined to form more general transformations. The set of all
possible transformations starting in a state with the empty square in position
p is denoted by T (p), and it is defined as the set of all products (in S16 ) of the
form mk mk−1 . . . m1 , where

mj+1 ∈ M (mj ◦ mj−1 ◦ . . . ◦ m1 (p))

for all j ≥ 0. This reflects the condition that mj+1 should be a possible move,
after applying the first j moves to s. In particular, in taking k = 0, we see
that the identity is a possible transformation in any state. Also, if t ∈ T (p) and
u ∈ T (t(p)), then u ◦ t ∈ T (p). Finally, we have t−1 ∈ T (t(p)).
Moves and, more generally, transformations change the state of the puzzle.
Let s ∈ S with s(p) = ∗ and let t ∈ T (p). Then t changes the state s to the
2.3. LOYD’S PUZZLE 27

Figure 2.6: Ordering of the non-empty squares in each state.

state t · s, which is defined by

(t · s)(q) = s(t−1 (q)).

In words, the square at position q after applying t is the same as the square at
position t−1 (q) before that transformation.
Now that we captured Sam Loyd’s 15-puzzle in a comfortable notation, we
can formalize our question as follows: is there a t0 ∈ T (16) that transforms the
messed up state s1 of Figure 2.2 into the initial state s0 of Figure 2.1, i.e., that
satisfies
t0 · s1 = s0 ?
The answer is no, and we shall presently make clear why. In each state, order
the non-empty squares according to the path in Figure 2.6. That is, the j-th
non-empty square one encounters when following this path in a particular state
s is called the j-th square in s, for j = 1, . . . , 15. Now consider any position
p, any t ∈ T (p) and some s ∈ S with s(p) = ∗. Then we define a permutation
πt ∈ S15 by

πt (i) = j ⇔ the i-th square in state s is the j-th square in state t · s.

Note that this is independent of the choice of s. We have that

πu◦t = πu ◦ πt (2.1)

whenever the left-hand side makes sense, i.e., whenever u ∈ T (t(p)).

We see that the t0 we would like to find equals (14, 15), and then πt0 also
equals (14, 15), considered as a permutation in S15 . This is an odd permutation.
Now consider the move m depicted in Figure 2.7. It interchanges the squares in
position 6 and 10 in Figure 2.5, so the move equals (6, 10). The corresponding
permutation πm is the cycle (10, 9, 8, 7, 6). A moment’s thought leads to the
28 CHAPTER 2. PERMUTATION GROUPS

6 7 8

10 9

Figure 2.7: Moves correspond to odd cycles.

conclusion that πm is an odd cycle for any move m, hence an even permutation.
As a consequence of this and equation (2.3), a sequence of such moves will also
be mapped to an even permutation by π. Therefore, such a transformation
cannot possibly be (14, 15).
You conclude that your friend must have cheated while you were away mak-
ing coffee, by lifting some of the squares out of the game and putting them back
in a different way.

Exercise 2.3.7. Show that the set {πt | t ∈ T (p) for some p} forms a subgroup
of S15 .

Exercise 2.3.8. Prove that the set

{t ∈ T (16) | t(16) = 16}

is a subgroup of A16 . Use this to give another argument why your friend must
have cheated.

2.4 Action of groups on sets

Although the transformations in the 15-puzzle do not quite form a group, their
‘acting’ on the puzzle states reflects an important phenomenon in group theory,
namely that of an action.

Definition 2.4.1. Let G be a group and M a set. An action of G on M is a

map α : G × M → M satisfying

1. α(e, m) = m for all m ∈ M , and

2. α(g, α(h, m)) = α(gh, m) for all m ∈ M and g, h ∈ G.

2.4. ACTION OF GROUPS ON SETS 29

If the action α is obvious from the context, we leave α out and write gm for
α(g, m). Also, for subsets S ⊂ G and T ⊂ M we write ST for {α(g, m) | g ∈
S, m ∈ T }. For {g}T we write gT and for S{m} we write Sm.

Remark 2.4.2. Given an action α, we have a homomorphism G → Sym(M )

given by g 7→ (m 7→ α(g, m)). Conversely, given a homomorphism φ : G →
Sym(M ), we have an action α : G × M → M given by

α(g, m) = φ(g)m.

In other words, an action on M is nothing but a homomorphism G → Sym(M ).

Let us give some examples of a group acting on a set.

Example 2.4.3.

1. The group Sn acts on {1, . . . , n} by π · i = π(i), and on the set of all

two-element subsets of {1, . . . , n} by π({i, j}) = {π(i), π(j)}.

2. The group GL(3, R) acts on R3 by matrix-vector multiplication.

3. Consider the group of all motions of the plane, generated by all trans-
lations (x, y) 7→ (x + α, y) for α ∈ R in x-direction and all inflations
(x, y) 7→ (x, βy) for β ∈ R∗ . This acts on the solutions of the differential
equation
d
( )2 (y) = −y.
dx
Example 2.4.4. Consider G = S4 and put a = {12, 34}, b = {13, 24}, c =
{14, 23}. Here 12 stands for {1, 2}, etc. There is a natural action (as above) on
the set of pairs, and similarly on the set {a, b, c} of all partitions of {1, 2, 3, 4}
into two parts of size two. Thus, we find a homomorphism S4 → Sym({a, b, c}).
It is surjective, and has kernel a group of order 4. (Write down its nontrivial
elements!)

Example 2.4.5. The group SL(2, C) acts on the complex projective line P1 (C)
as follows. Denote by π : C2 \ {0} → P1 (C) the map sending a pair of homoge-
neous coordinates to the corresponding point. The linear action of SL(2, C) on
C2 by left multiplication permutes the fibers of π, hence induces an action on
P1 (C). On the affine part of P1 (C) where the second homogeneous coordinate
is non-zero, we may normalize it to 1, and the action on the first coordinate x
is given by  
a b ax + b
x= .
c d cx + d
Note, however, that this affine part is not invariant under the group.

Lemma 2.4.6. Let G be a group. Then G acts on itself by conjugation, that

is, the map α : G × G → G defined by α(h, g) = hgh−1 is an action.
30 CHAPTER 2. PERMUTATION GROUPS

Proof. We check the two conditions of Definition 2.4.1. First,

α(e, g) = ege−1 = g

for all g ∈ G, and second,

α(h, α(k, g)) = h(kgk −1 )h−1 = (hk)g(hk)−1 = α(hk, g).

The following lemma allows us to partition a set M on which a group G acts

into orbits.
Lemma 2.4.7. Let G be a group acting on a set M . Then the sets Gm = {gm |
g ∈ G} for m ∈ M partition the set M . The set Gm is called the orbit of m
under G.
Definition 2.4.8. An action of a group G on a set M is said to be transitive if
it has only one orbit, or, stated differently, Gm = M for any m ∈ M .
To define symmetry in a general context, we look at sets that are mapped
to themselves by a given group.
Definition 2.4.9. Let G be a group acting on M . A subset T ⊆ M is called
invariant under S ⊆ G if ST ⊆ T . In this case, the elements of S are called
symmetries of T .
Definition 2.4.10. Let G be a group acting on M and let m ∈ M . Then the
set
Gm := {g ∈ G | gm = m}
is called the stabilizer of m.
Lagrange’s theorem has an important consequence for problems, in which
one has to count the number of symmetries of a given set. It can be formulated
as follows.
Theorem 2.4.11. Let G be a group acting on a set M and let m ∈ M . Then
Gm is a subgroup of G, and the map

f : Gm → G/Gm , g · m 7→ gGm

is well defined and bijective. As a consequence, if G is finite, then

|G| = |Gm||Gm |.

Remark 2.4.12. One must distinguish the elements of G from their actions on
M . More precisely, the map sending g ∈ G to the map α(g, ·) : M → M is a
homomorphism, but need not be an isomorphism.
We now want to make clear how the action of groups can help in classification
and counting problems, and we shall do so by means of the following problem.
2.4. ACTION OF GROUPS ON SETS 31

1 2 1 2 1

2
3

3 4 3 4 4 5

Figure 2.8: Three graphs, two of which are isomorphic.

Given is a finite set X with |X| = v. What is the number of non-

isomorphic simple graphs with vertex set X?
This natural, but seemingly merely theoretical, question has strong analogue
in counting all isomers with a given chemical formula, say C4 H10 (see [6]).
Of course, the question is formulated rather vaguely, and the first step in
solving it is to formalize the meanings of the words graph and non-isomorphic.
A simple graph is a function
 
X
f: → {0, 1}.
2

Here X


2 stands for the set of all unordered pairs from X. If the function f
takes the value 1 on such a pair {x, y}, this means that the pair is an edge in
the graph. Often, graphs are associated with a picture like the ones in Figure
2.8. The set of all graphs on X is denoted by Graphs(X).
The first two graphs in Figure 2.8 are essentially the same; to make them
exactly the same one has to relabel the vertices. Such a relabeling is in fact a
permutation of X. Formally, this leads to an action
of Sym(X) on Graphs(X)
in the following way. First of all Sym(X) acts on X 2 by

π · {x, y} = {π(x), π(y)}.

Furthermore, we define π · f by the commutativity of the following diagram.
X

f
2 −−−−→ {0, 1}
 

π·y

yId
X


2 −−−−→ {0, 1}
π·f

X


This diagram tells us that e ∈ 2 is an edge in the graph f if and only if π · e
is an edge in the graph π · f . In formula,
π · f (x) = f (π −1 x).
32 CHAPTER 2. PERMUTATION GROUPS

Graphs that can be obtained from each other by the action of some π ∈
Sym(X) are called isomorphic. Now our original question can be rephrased
very elegantly as follows.

How many orbits does the action of Sym(X) have on Graphs(X)?

The following lemma answers this question in general.

Lemma 2.4.13 (Cauchy-Frobenius). Let G be a finite group acting on a finite

set M . Then the number of orbits of G on M equals
1 X
fix(g).
|G|
g∈G

Here fix(g) denotes the number of m ∈ M with g · m = m.

Exercise 2.4.14. Prove Lemma 2.4.13.

To apply this lemma to counting graphs, we need to know for some π ∈

Sym(X), the number of f ∈ Graphs(X) that are fixed by π. The
function f is
fixed by π if and only if f is constant on the orbits of hπi in X
2 . This reflects
the condition that π should map edges to
edges and non-edges to non-edges.
Denoting the number of orbits of hπi on X 2 by o(π), we find that the number
of non-isomorphic graphs on X equals
1 X
2o(π) ,
v!
π∈Sym(X)

where v = |X|.

Exercise 2.4.15. Consider the permutation π = (1, 2)(3, 4, 5) on {1, 2, 3, 4, 5}.

Calculate the number o(π). Try to find a general formula in terms of the cycle
lengths of π and the number v. Observe that o(π) only depends on the cycle
type.

Example 2.4.16. For v = 4, we have o(e) = 6, o((12)) = 4, o((123)) = 2,

o((1, 2, 3, 4)) = 2 and o((12)(34)) = 4; the conjugacy clases of each of these
permutations have sizes 1, 6, 8, 6 and 3, respectively. Hence, the number of
non-isomorphic graphs on 4 vertices equals

(1 · 26 + 6 · 24 + 8 · 22 + 6 · 22 + 3 · 24 )/24 = 11.

Can you find all?

Chapter 3

Symmetry Groups in
Euclidean Space

This chapter deals with groups of isometries of n-dimensional Euclidean space.

We set up a framework of results that are valid for general n, and then we
restrict our attention to n = 2 or 3. In these two cases, we classify the finite
groups of isometries, and point out which of these are crystallographic point
groups, i.e., preserve a 2- or 3-dimensional lattice, respectively. First, however,
we consider an example that motivates the development of this theory.

3.1 Motivation
In low dimensions, there are few finite groups of isometries and this enables us,
when given a two- or three-dimensional object, to determine its symmetry group
simply by enumerating all axes of rotation and investigating if the object admits
improper isometries such as inversion and reflections. As an example, suppose
that we want to describe the symmetry group G of the CH4 molecule depicted
in Figure 3.1. The hydrogen atoms are arranged in a regular tetrahedron with
the C-atom in the center.
As a start, let us compute |G|. Consider a fixed H-atom a. Its orbit Ga
consists of all four H-atoms, as can be seen using the rotations of order 3 around
axes through the C-atom and an H-atom. Applying Lagrange’s theorem we find
|G| = 4|Ga |. The stabilizer of a consists of symmetries of the remaining 4 atoms.
Consider one of the other H atoms, say b. Its orbit under the stabilizer Ga has
cardinality 3, and the stabilizer (Ga )b consists of the identity and the reflection
in the plane through a, b and the C-atom, so that we find

|G| = |Ga||Ga | = |Ga||Ga b||(Ga )b | = 4 · 3 · 2 = 24.

Only half of these isometries are orientation-preserving.

To gain more information on the group G, we try to find all rotation axes:

33
34 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

C
H
H
H
Figure 3.1: The CH4 molecule.

Figure 3.2: A regular tetrahedron inside a cube.

1. There are 4 three-fold axes, namely those through an H-atom and the
center of the opposite side of the tetrahedron.

2. There are three two-fold axes, namely those through the centers of two
perpendicular sides.

Later on, we shall see that this information determines the orientation pre-
serving part of G uniquely; it is denoted by T . From the fact that G does
not contain the inversion, but does contain other improper isometries, we may
conclude that G is T ∪ i(W \ T ), a group that can be described by Figure 3.2.
Here we see a regular tetrahedron inside a cube. Half of the proper isometries
of the cube are isometries of the tetrahedron, and the improper isometries of
the tetrahedron may be obtained as follows: take any proper isometry of the
cube which is not a isometry of the tetrahedron, and compose it with the in-
version i. Note that T ∪ i(W \) and W are isomorphic as abstract groups, but
their realizations in terms of isometries are different in the sense that there is
no isometry mapping one to the other. This point will also be clarified later.
3.2. ISOMETRIES OF N -SPACE 35

Exercise 3.1.1. Consider the reflection in a plane through two H-atoms and
the C-atom. How can it be written as the composition of the inversion and a
proper isometry of the cube?
The finite symmetry group of the tetrahedron and the translations along
each of three mutually orthogonal edges of the cube in Figure 3.2 generate the
symmetry group of a crystal-like structure. For this reason, the group G is
called a crystallographic point group.

3.2 Isometries of n-space

Consider the n-dimensional Euclidean space Rn with the distance function
X
d(x, y) = kx − yk = (x − y, x − y)1/2 = ( (xi − yi )2 )1/2 .
i

Definition 3.2.1. An isometry is a map g : Rn → Rn such that

d(g(x), g(y)) = d(x, y) for all x, y ∈ Rn .

The set of all isometries is a subgroup of Sym(Rn ), which we denote by AO(n, R).
The following definition and lemmas will explain the name of this group.
Definition 3.2.2. For a fixed a ∈ Rn , the isometry ta : x 7→ x + a is called the
translation over a.
Lemma 3.2.3. For any g ∈ AO(n, R), there exists a unique pair (a, r) where
a ∈ Rn and r ∈ AO(n, R) fixes 0, such that g = ta r.
Proof. Set a := g(0). Then t−1 −1
a g(0) = 0, so ta g is an isometry fixing 0, and it
is clear that a = g(0) is the only value for which this is the case.
Lemma 3.2.4. Let r ∈ AO(n, R) be an isometry fixing 0. Then r is an orthog-
onal R-linear map.
Proof. Let x, y ∈ Rn , and compute

(rx, ry) = (−d(rx, ry)2 + d(rx, 0)2 + d(ry, 0)2 )/2

= (−d(x, y)2 + d(x, 0)2 + d(y, 0)2 )/2 = (x, y),

where we use that r is an isometry fixing 0. Hence, r leaves the inner product
invariant. This proves the second statement, if we take the first statement for
granted. For any z ∈ Rn , we have

(z, rx + ry) = (z, rx) + (z, ry) = (r−1 z, x + y) = (z, r(x + y)),

so that rx + ry − r(x + y) is perpendicular to all of Rn , hence zero. Similarly,

one finds r(αx) = αrx for α ∈ R. This proves the first statement.
36 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

Exercise 3.2.5. Prove that an isometry is bijective. (The assertion is used

in the second part of Definition 3.2.1.) Hint: If r is an isometry, then t−r0 r
is an isometry fixing 0, so it suffices to prove the assertion for r with r0 = 0.
Now the proof of Lemma 3.2.4 uses r−1 , so we need another proof. Set v =
r(λx + µy) − λr(x) − µr(y) and consider (v, v). Using bilinearity of the inner
product and orthogonality of r, it can be shown that (v, v) = 0.
The group of all orthogonal linear maps Rn → Rn is denoted by O(n, R).
An isometry is thus an orthogonal linear map, followed by a translation; hence
the name affine orthogonal transformation.
For a group G, we write

T (G) := {a ∈ Rn | ta ∈ G};

this is an additive subgroup of Rn , which we shall identify with the group of all
translations in G.
Using the linearity of isometries fixing 0, one can prove the following.
Proposition 3.2.6. For a subgroup G ⊆ AO(n, R), the map R : G → O(n, R)
defined by
g = ta R(g) for some a ∈ Rn
is well-defined, and a group homomorphism. Its kernel is T (G), and this set is
invariant under R(G).
Proof. Lemma 3.2.3 and 3.2.4 show that R is well-defined. Let r1 , r2 ∈ O(n, R)
and a1 , a2 ∈ Rn , and check that

ta1 r1 ta2 r2 = ta1 +r1 a2 r1 r2 .

This proves that R is a homomorphism. Its kernel is obviously T (G). Let

r ∈ R(G) and a1 ∈ Rn such that ta1 r1 ∈ G, and let a2 ∈ T (G). Then, for
x ∈ Rn ,

ta1 r1 ta2 r1−1 t−a1 x = ta1 r1 ta2 (r1−1 x − r1−1 a1 )

= ta1 (x − a1 + r1 a2 )
= tr1 a2 x.

This shows shows that T (G) is invariant under R(G).

We are interested in certain groups that act discretely.
Definition 3.2.7. A subset V ⊆ Rn is called discrete if for every point p ∈ V
there is an open neighbourhood U ⊆ Rn such that

U ∩ V = {p}.

Definition 3.2.8. A discrete subgroup of Rn is called a lattice, and the dimen-

sion of its R-linear span is called its rank.
3.2. ISOMETRIES OF N -SPACE 37

Lemma 3.2.9. Let L ⊆ Rn be a lattice. Then L is a closed set.

Proof. As 0 is an isolated point, there is an  > 0 such that kvk ≥  for all
v ∈ L \ {0}. But then kv − wk ≥  for all distinct v, w ∈ L. Hence, any sequence
(vn )n in L with vn − vn+1 converging to 0, is eventually constant.

Proposition 3.2.10. Let a1 , . . . , ak ∈ Rn be linearly independent over R. Then

the set X
L= Zai
i

is a lattice. Conversely, every lattice is of this form.

Proof. To prove that an additive subgroup of Rn is a lattice, it suffices to prove

that 0 is an isolated point. For L as in the lemma, let m be the minimum of
the continuous function X
x→k xi ai k
i

on the unit sphere {x ∈ Rk | kxk = 1}. As the ai are linearly independent,

m > 0. For any non-zero x ∈ Zk , we find
X X xi
k xi ai k = kxkk ai k ≥ kxkm ≥ m,
i i
kxk

proving the first statement.

For the converse, we proceed by induction on the rank of the lattice. Clearly,
a lattice of rank 0 is of the desired form. Now suppose that every lattice of rank
k − 1 is of that form, and let L ⊆ Rn be a lattice of rank k. First let us show
that L is a closed subset of Rn .
Being discrete and closed, L has only finitely many points in any compact
set. In particular, one can choose a v1 ∈ L of minimal non-zero norm. Let
π : Rn → v1⊥ denote the orthogonal projection along v1 . Then π(L) is an
additive subgroup of v1⊥ . Let v ∈ L be such that π(v) 6= 0. After subtracting an
integer multiple of v1 from v, we may assume √ that kv − π(v)k ≤ kv1 k/2. Using
kvk ≥ kv1 k and Pythagoras, we find kπ(v)k ≥ 3/2kv1 k. This proves that π(L)
is discrete, hence a lattice. It has rank k − 1, so there exist v2 , . . . , vk ∈ L whose
images under π are linearly independent, and such that

k
X
π(L) = Zπ(vi ).
i=2

Now v1 , . . . , vk are linearly independent. For v ∈ L we can write

k
X
π(v) = ci π(vi )
i=2
38 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

Pk
with ci ∈ Z. Then v− i=2 ci vk ∈ L is a scalar multiple of v1 , and by minimality
of kvk the scalar is an integer. Hence,

k
X
L= Zvi .
i=1

Definition 3.2.11. A subgroup G ⊆ AO(n, R) is said to act discretely if all its

orbits in Rn are discrete.

If G ⊆ AO(n, R) is a group that acts discretely, then T (G) does so, too. In
general, this does not hold for R(G), as the following example shows.

Example 3.2.12. Let α ∈ R be an irrational multiple of π, and define r ∈

O(3, R) by
 
cos α sin α 0
r=  − sin α cos α 0 .
0 0 1
Consider the group G generated by g = te3 r. From g n = tne3 rn it is clear that
G acts discretely, and that R(G) = {rn | n ∈ Z}. As r has infinite order, we
have T (G) = {0}, and R(G) does not act discretely: the R(G)-orbit of (1, 0, 0)T
is an infinite set in the circle {x ∈ R3 | x3 = 0, kxk = 1}.

This phenomenon does not occur if we require T (G) to have the maximal
possible rank.

Theorem 3.2.13. Let G be a subgroup of O(n, R) leaving invariant a lattice L

of rank n. Then G is finite.

Proof. Let a1 , . . . , an ∈ L be linearly independent. As L is discrete, it contains

only finitely many vectors of norm kai k, for each i. Hence, the G-orbits Gai
n
are finite. As
Sn the ai span R , the permutation representation of G on the finite
set X := i=1 Gai is faithful. Hence, G is embedded into the finite group
Sym(X).

Definition 3.2.14. A subgroup G ⊆ AO(n, R) for which T (G) is a lattice of

rank n, is called a crystallographic group in n dimensions. A group G ⊆ O(n, R)
leaving invariant a lattice of rank n, is called a crystallographic point group in
n dimensions.

Corollary 3.2.15. Let G ⊆ AO(n, R) be a crystallographic group. Then R(G)

is a crystallographic point group. Any crystallographic point group is finite.

Proof. By Proposition 3.2.6, T (G) is invariant under R(G). Now apply Theorem
3.2.13.
3.2. ISOMETRIES OF N -SPACE 39

This corollary allows for the following approach to the classification of crys-
tallographic groups in O(n, R): first classify all finite subgroups of O(n, R), then
investigate for each of them whether it leaves invariant a lattice of rank n. Fi-
nally, for each pair (R, L) of a crystallographic point group R and a lattice L of
rank n left invariant by R, there may be several non-equivalent groups G with
T (G) = L and R(G) = R. Among them is always the semi-direct product L o R:
the subgroup of AO(n, R) generated by L and R.
In this setting, ‘classification’ means ‘up to equivalence under AO(n, R)’, i.e.,
the groups G and hGh−1 for some h ∈ AO(n, R) are considered the same. This
is a finer equivalence relation than isomorphism, as is shown in the following
example.
Example 3.2.16. The groups C2 and D1 (see Section 3.3) are isomorphic as
abstract groups, but not equivalent under AO(2, R). The same holds for the
groups W and T ∪ i(W \ T ) (see Section 3.4).
Exercise 3.2.17. Show that the symmetry group G of Figure 1.4 is not the
semi-direct product of R(G) and T (G).
In the next sections, we shall carry out this classification in the cases n = 2
and 3, up to the point of determining the crystallographic point groups. The
determinant of an orthogonal map turns out to be a useful tool there.
Definition 3.2.18. An isometry r ∈ O(n, R) is called proper or orientation
preserving if det r = 1. The set of all proper isometries in O(n, R) is denoted
by SO(n, R). An isometry g ∈ AO(n, R) is called proper if R(g) is proper.
Otherwise, it is called improper.
Exercise 3.2.19. Prove that, in any subgroup G ⊆ AO(n, R) containing im-
proper isometries, the proper ones form a subgroup of index 2.
Exercise 3.2.20. Consider the following map:

ra : v 7→ v − 2(v, a)a,

where a ∈ Rn has norm 1 (that is, ||a|| = 1). Prove that

1. it belongs to O(n, R);
2. fixes each vector in the hyperplane a⊥ ;
3. ra (a) = −a;
4. det(ra ) = −1;
5. ra has order 2;
Such an element is called a reflection.
Exercise 3.2.21. Show that any finite subgroup G ⊆ AO(n, R) fixes a point
in Rn . Hint: consider the ‘average’ of an arbitrary orbit.
40 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

Exercise 3.2.22. Let G be a finite subgroup of the group GL(n, R) of all invert-
ible n × n-matrices. Show that G leaves invariant a positive definite symmetric
bilinear form, and that it is hence conjugate to a subgroup of O(n, R). Hint: let
(., .) denote the standard inner product, and define

1 X
hx, yi := (gx, gy)
|G|
g∈G

for x, y ∈ Rn . Show that h., .i is symmetric and positive definite. Let A be the
matrix with entries aij = hei , ej i, and B ∈ GL(n, R) such that B T B = A. Show
that BGB −1 ⊆ O(n, R).

3.3 The Finite Subgroups of O(2, R)

In nature, as well as in art and many other areas, many plane figures with
certain symmetries exist. One may think of flowers (seen from above), frieze
patterns on picture frames, Escher’s plane tilings and decorative wall papers.
In order to describe these symmetries, we investigate the finite subgroups of
O(2, R).
In two dimensions, we have two different types of isometries.

Lemma 3.3.1. The linear map rφ : R2 → R2 with matrix

 
cos φ − sin φ
sin φ cos φ

with respect to the standard basis is an element of SO(2, R), called rotation
around 0 over φ. Conversely, any element of this group equals some rφ . Any
improper isometry in O(2, R) is a reflection in a line, see Exercise 3.2.20.

The proof of this lemma is straightforward. Let us describe two classes of

finite subgroups of O(2, R).

1. The cyclic group of order n, denoted by Cn , and generated by r2π/n .

2. The dihedral group of order 2n, denoted by Dn , and generated by r2π/n

and the reflection in a fixed line.

In fact, these are the only two classes, as we shall now prove. Let G be a
finite subgroup of O(2, R). There are two possibilities:

1. G ⊆ SO(2, R). By Lemma 3.3.1, it consists of rotations rφ . Let φ > 0

be minimal such that rφ ∈ G. As G is finite, φ = 2π/n for some integer
n > 0. If θ is any other angle such that rθ ∈ G, then θ = kφ + σ for some
integer k and angle σ ≥ 0 and < φ. By minimality of φ, σ = 0. Hence G
is generated by r2π/n , and equal to Cn .
3.3. THE FINITE SUBGROUPS OF O(2, R) 41

2. G contains a reflection g in a line l. Then G ∩ SO(2, R) is a subgroup of

index 2 in G, and by the first case generated by r2π/n for some integer
n > 0. Now G is generated by the g and r2π/n , and conjugate to Dn .

We have thus proved the following theorem.

Theorem 3.3.2. Let G be a finite subgroup of O(2, R). Then G is conjugate

to one of the groups Cn or Dn for some integer n ≥ 1.

Exercise 3.3.3. Show that O(n, R) is generated by reflections. Hint: show first
that it is generated by elements that fix an (n − 2)-dimensional subspace of Rn
pointwise, and that O(2, R) is generated by reflections.

Using this classification, we can find which of the finite groups are crystal-
lographic point groups.

Theorem 3.3.4. The following groups are crystallographic point groups:

C1 , C2 , C3 , C4 , C6 ,
D1 , D2 , D3 , D4 , D6 .

Moreover, any crystallographic point group G in two dimensions is conjugate to

one of the above.

Proof. For the first part, one only needs to construct invariant lattices for D4
and D6 (see 3.3.5), as the other groups are contained in one of these see.
For the second statement, suppose that G is a crystallographic point group.
According to Theorem 3.2.13, G must be finite. The lattice L is invariant under
the subgroup K of O(2, R) generated by G ∪ {i}, where i denotes the inversion.
As the inversion commutes with all elements of G, K is finite (either equal to
G or twice as large). Therefore, according to Theorem 3.3.2, K is conjugate to
Cn or Dn for some n; as K contains the inversion, n must be even. So either
n ≤ 6, in which case G is of one of the types mentioned in the theorem, or
n ≥ 8. Suppose that the latter were the case and choose a vector c in the
lattice of shortest possible length. Now r2π/n c ∈ L and c − r2π/n c ∈ L. A
little calculus shows that the latter vector is shorter than c because the rotation
angle too small (for n = 8, this can be seen from Figure 3.3), and we arrive at
a contradiction.

Exercise 3.3.5. Show that D4 and D6 are crystallographic point groups.

Note that we have only classified all crystallographic point groups, and not
the crystallographic groups. There are 17 of the latter, and only 10 of the
former; see [2].
42 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

c-hc

hc

Figure 3.3: Why n cannot not be 8.

3.4 The Finite Subgroups of O(3, R)

In this section, we shall repeat the classification of §3.3, but now for three
dimensions.
Lemma 3.4.1. Any element g ∈ SO(3, R) fixes a line, i.e., with respect to some
orthonormal basis g has the matrix
 
1 0
,
0 M

where M is the matrix of an element of SO(2, R).

Proof. Let λ1 , λ2 , λ3 be the eigenvalues of g. As g leaves the norm invariant,
all have norm 1. As g is proper, we have λ1 λ2 λ3 = 1. Now either all are real,
in which case at least one of them is 1, or we may assume that λ1 is real and
λ3 = λ2 . But then λ2 λ3 = |λ2 |2 = 1, so λ1 = 1.
An element of SO(3, R) is called a rotation, the line fixed by it its axis and
the angle corresponding to M its rotation angle. We present some classes of
finite subgroups of SO(3, R).

1. Fix an axis l and consider the proper rotations around l over the angles
2kπ/n for k = 0, . . . , n − 1. They form a group, denoted by Cn , due to its
analogy to the class denoted by Cn in the two-dimensional setting.
2. Start with the group Cn of rotations around l and choose a second axis m
perpendicular to l. Consider the group generated by Cn and the rotation
around m over π. It is twice as large as Cn , and denoted by Dn0 . Note that
it contains only proper isometries. The proper symmetries of an n-prism
form a group of type Dn0 . Note that D10 is conjugate to C2 .
3.4. THE FINITE SUBGROUPS OF O(3, R) 43

3. The five platonic solids (tetrahedron, cube, octahedron, dodecahedron

and icosahedron) with vertices on the unit sphere yield examples of finite
subgroups of SO(3, R). However, only three distinct classes arise:
• T , the group of all proper rotations leaving a tetrahedron invariant.
• W , the group of all proper rotations that are symmetries of a cube.
• P , the group of all proper rotations leaving a dodecahedron invariant.
The reason is that the octahedron and the cube are polar figures: it is
easily seen that the proper symmetries of a cube are exactly those of the
octahedron which has its vertices in the middles of the faces of the cube.
A similar statement holds for the dodecahedron and the icosahedron.

Theorem 3.4.2. Any finite subgroup of SO(3, R) is conjugate to one of the

following.

Cn for n ≥ 1,
Dn0 for n ≥ 2,
T, W, P.

The following proof is due to Euler.

Proof. Let G be a finite subgroup of SO(3, R). Consider the set S defined by

S := {p ∈ R3 | ||p|| = 1, there exists g ∈ G such that g 6= e and g(p) = p}.

Count the number N of pairs (g, p), with e 6= g ∈ G and p on the unit sphere,
that satisfy g(p) = p. On one hand, each g 6= e fixes exactly two anti-podal
points, hence this number is

N = 2(|G| − 1).

On the other hand, for each p ∈ S, the number of g 6= e fixing p equals |Gp | − 1.
Hence X
N= (|Gp | − 1).
p∈S

If gp = p and g 6= e, and h ∈ G, then (hgh−1 )hp = hp and hgh−1 6= e, so S is

invariant under G. Partition S into orbits of G. The value of |Gp | is constant
on such an orbit o, namely |G|/|o|, due to Lagrange’s theorem. Hence we find
X
2(|G| − 1) = |o|(|G|/|o| − 1),
o

where the sum is taken over all orbits of G in S. Hence, dividing both sides by
|G|, we get
2 X |o|
2− = (1 − ).
|G| o
|G|
44 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

Group type Order Number of orbits on S Their sizes rotation axes

Cn n 2 1, 1 1×n
Dn0 2n 3 2, n, n 1 × n, n × 2
T 12 3 4, 4, 6 4 × 3, 3 × 2
W 24 3 8, 6, 12 4 × 3, 3 × 4, 6 × 2
P 60 3 12, 20, 30 6 × 5, 10 × 3, 15 × 2

Table 3.1: The finite groups of proper rotations in space.

Hence, if we set ao := |G|/|o| for each orbit o on S, then we have

X
(1 − 1/ao ) < 2.
o

As each s ∈ S has non-trivial stabilizer, each orbit in S has size at most |G|/2;
hence each ao is at least 2. Now Exercise 3.4.3 leads to Table 3.1. That is, it
provides the numerical data for it; it remains to check that this data determines
the group up to conjugation.

Exercise 3.4.3. Consider the following function in GAP.

extend:=function(a,s)

#Pre-condition: a is a non-empty non-decreasing list of integers >=2,

#and s equals (sum i: (1-1/a[i])). This procedure ‘extends’ a in all
#possible ways, under the condition that this sum remains smaller than
#2.

local b,g,o;
if s >= 2 then return false;
else g:=2/(2-s); o:=List(a,x->g/x);
Print("Orbit sizes:",o," |G|:",g,"\n");
b:=a[Length(a)];
while extend(Concatenation(a,[b]),s+(1-1/b)) do
b:=b+1;
od;
return true;
fi;
end;
Explain why the function call extend([2,2],1) results in an endless loop. On
the other hand, when called as extend([2,3],1/2+2/3), the function does halt,
and prints
Orbit sizes:[ 6/5, 4/5 ] |G|:12/5
Orbit sizes:[ 6, 4, 4 ] |G|:12
Orbit sizes:[ 12, 8, 6 ] |G|:24
Orbit sizes:[ 30, 20, 12 ] |G|:60
3.4. THE FINITE SUBGROUPS OF O(3, R) 45

Figure 3.4: The orbits of W on the points on rotation axes.

Compare this with Table 3.1.

Exercise 3.4.4. In Figure 3.4 we see a cube with vertices on the unit sphere.
Its group of proper symmetries contains non-trivial rotation axes through the
points drawn in that picture; they form the set S in the proof of Theorem 3.4.2.
The points in the different orbits are given different shades of grey. Make similar
pictures for the group types Cn , Dn0 and T . Describe the symmetry axes for the
icosahedron.
Now let us try to include improper rotations. Let G be a finite subgroup of
O(3, R) containing improper rotations. Then H := G ∩ SO(3, R) has index 2 in
G. If G contains the inversion i, then we have a disjoint union
G = H ∪ iH.
Note that this is isomorphic to H × C2 as an abstract group. Now assume that
G does not contain i. Then the set
G0 := H ∪ i(G \ H)
is a subgroup of SO(3, R) in which H has index 2, and one can recover G by
G = H ∪ i(G0 \ H).
Conversely, if G0 , H ⊆ SO(3, R) are finite subgroups, and H has index 2 in G0 ,
then this equation defines a finite subgroup of O(3, R), which is isomorphic to
G0 as an abstract group, but not equivalent to it.
This enables us to classify all finite subgroups of O(3, R).
Theorem 3.4.5. Any finite subgroup of O(3, R) is conjugate to one of the
following:
Cn , Cn ∪ iCn , Cn ∪ i(C2n \ Cn ), for n ≥ 1,
Dn0 , Dn0 ∪ iDn0 , Dn ∪ 0
i(D2n \ Dn0 ), Cn ∪ i(Dn0 \ Cn ), for n ≥ 2,
T, W, P, T ∪ iT, W ∪ iW, P ∪ iP, T ∪ i(W \ T ).
46 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

We do not give all details of the proof, but restrict ourselves to the following
remarks.

1. Note that D10 and C2 describe the same class; for this reason the second
list starts with n = 2.

2. The last group in the list is made possible by the fact that a group of
type W has a subgroup of type T of index 2. This can be seen as follows:
consider the group of proper symmetries of a cube with center 0, and
look at the lines between anti-podal vertices of the cube; they are axes of
rotations of order 3 leaving the cube invariant; these proper rotations form
the symmetry group of a regular tetrahedron with center 0 and points on
those axes.

In applications, one wants to determine the (finite) symmetry group of a

given object. The following procedure, based on Table 3.1 and the list in The-
orem 3.4.5, can then be helpful.

1. List all rotation axes and their orders. They determine the subgroup H
of G of all proper rotations.

2. Check whether G contains any improper rotations. If not, then G = H.

3. Suppose that G contains improper rotations. Check whether i ∈ G. If so,

then G = H ∪ iH.

4. Suppose that G contains improper rotations, but not the inversion. Then,
according to the list of Theorem 3.4.5, there is only one possibility for G,
unless H is of type Cn .

5. Suppose that H = Cn and that G contains improper rotations, but not

the inversion. Then if G is cyclic, it is of type Cn ∪i(C2n \Cn ). Otherwise,
it is of type Cn ∪ i(Dn0 \ Cn ).

Exercise 3.4.6. Consider the molecule depicted in Figure 3.5. Identify the
class of its symmetry group in the list of Theorem 3.4.5.

Exercise 3.4.7. Consider the full symmetry group of the tetrahedron, including
its improper rotations. Which class in the list of Theorem 3.4.5 does it belong
to?

Exercise 3.4.8. Determine the symmetry group of the CH4 molecule depicted
in Figure 3.1.

Exercise 3.4.9. Consider the ethane molecule in staggered configuration as

depicted in Figure 3.6. What is the type of its symmetry group?

Exercise 3.4.10. Determine the symmetry group of the N4 P4 -molecule, de-

picted in Figure 3.7.
3.4. THE FINITE SUBGROUPS OF O(3, R) 47

H H

H
Figure 3.5: The NH3 - molecule.

C
H H

H H
C

H
Figure 3.6: The ethane molecule in staggered configuration.
48 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE

P
P
N N

N N
P
P
Figure 3.7: The ethane molecule in staggered configuration.

Most of the above examples are taken from [2].

Finally, the question arises which of the finite subgroups of O(3, R) are crys-
tallographic point groups.

Theorem 3.4.11. Let G be a crystallographic point group in 3 dimensions.

Then G is conjugate to one of the following groups:

Cm , Cm ∪ iCm m = 1, 2, 3, 4, 6
0 0 0
Dm , Dm ∪ iDm m = 2, 3, 4, 6
Cm ∪ i(C2m \ Cm ) m = 1, 2, 3
0 0 0
Dm ∪ i(D2m \ Dm ) m = 2, 3
0
Cm ∪ i(Dm \ Cm ) m = 2, 3, 4, 6
T, W, T ∪ iT, W ∪ iW, T ∪ i(W \ T )

Like in the two-dimensional case, a crystallographic point group may leave

several lattices invariant. Indeed, there are 230 mutually non-equivalent crys-
tallographic groups in 3 dimensions, while there are only 32 crystallographic
point groups, see [2].
The proof runs like that of Theorem 3.3.4. For the sake of compatibility
with other texts on crystallographic groups, especially texts on the applications
in physics and chemistry, we include table 3.4. It is organized as follows: in
the first column, one finds lattices classified up to their symmetry groups. The
second column contains the groups leaving that lattice invariant (but no lattice
with a smaller symmetry group), in our notation. The last row for each lattice
is the full symmetry group of that lattice. There is only one exception to this
rule: a trigonal lattice is really hexagonal. The third and fourth column contain
two commonly used sets of symbols for them.
3.4. THE FINITE SUBGROUPS OF O(3, R) 49

Type of Lattice Our notation Schoenflies International

Triclinic C1 C1 1
C1 ∪ iC1 Ci = S2 1
Monoclinic C2 C2 2
C1 ∪ i(C2 \ C1 ) Cs = C1h m
C2 ∪ iC2 C2h 2/m
Orthorhombic D20 D2 = V 222
C2 ∪ i(D20 \ C2 ) C2v mm2
D20 ∪ iD20 D2h = Vh mmm
Trigonal C3 C3 3
D30 D3 32
C3 ∪ iC3 C3i = S6 3
C3 ∪ i(D30 \ C3 ) C3v 3m
D30 ∪ iD30 D3d 3m
Tetragonal C4 C4 4
D40 D4 422
C4 ∪ iC4 C4h 4/m
C2 ∪ i(C4 \ C2 ) S4 4
C4 ∪ i(D40 \ C4 ) C4v 4mm
D40 ∪ i(D40 \ D20 ) D2d = Vd 42m
D40 ∪ iD40 D4h 4/m m m
Hexagonal C6 C6 6
D60 D6 622
C6 ∪ iC6 C6h 6/m
C3 ∪ i(C6 \ C3 ) C3h 6
C6 ∪ i(D60 \ C6 ) C6v 6mm
D3 ∪ i(D60 \ D30 ) D3h 6m2
D60 ∪ iD60 D6h 6/m m m
Isometric T T 23
W O 432
T ∪ iT Th m3
T ∪ i(W \ T ) Td 43m
W ∪ iW Wh m3m

Table 3.2: A Dictionary of Crystallographic Group Names.

50 CHAPTER 3. SYMMETRY GROUPS IN EUCLIDEAN SPACE
Chapter 4

Representation Theory

4.1 Linear representations of groups

An abstract group can be represented in various ways. In Chapter 2, we en-
countered permutation representations; in the present chapter we will start the
theory of linear representations. In this chapter, G denotes a group and K a
field of characteristic zero. All vector spaces will be over K.

The notion
We introduce the concept of a linear representation of G.
Definition 4.1.1. A (linear) representation of a group G on a vector space V
is a homomomorphism
ρ : G → GL(V ).
If ρ is injective, it is called faithful.
After a choice of basis of V , we can view a representation as a homomorphism

G → GL(n, K),

where GL(n, K) is the group of invertible matrices with entries in K; such a map
is called matrix representation of G.
If ρ : G → GL(V ) is a representation of G on V , then we often write gv
instead of ρ(g)v, if no confusion is possible. Also, G is said to act linearly on V ,
and V is called a G-module. The dimension of the representation is by definition
the dimension of V . Note that (g, v) 7→ ρ(g)v is indeed an action of G on V in
the sense of Definition 2.4.1.
Exercise 4.1.2. To experience the use of matrix representations of groups, try
to guess which orthogonal linear map is the product ρx ρy ρz . Here ρx stands
for rotation over π about the x-axis, and similarly for y and z. To check your
guess, represent the rotations by matrices and compute the product.

51
52 CHAPTER 4. REPRESENTATION THEORY

Example 4.1.3. Let G be the cyclic group Cn of order n with generator c. It

has one-dimensional complex representations rl (l ∈ Z) defined by

rl : Cn → GL(1, C), ck 7→ (ekl2πi/n ).

Note that rl is faithful if and only if n and l are co-prime.

Example 4.1.4. Recall (2.3) the permutation matrix presentation of the sym-
metric group S5 given in Chapter 2: to each permutation π ∈ S5 , we associate
the 5 × 5 matrix Mπ with Mπ ei = eπ(i) . This is a matrix representation of S5 .
The same thing works for S3 . The permutation (1, 2) can be represented by
the matrix  
0 1 0
M(1,2) = 1 0 0 ,
0 0 1
and (1, 2, 3) can be represented by
 
0 0 1
M(1,2,3) = 1 0 0 .
0 1 0

We know that (1, 2)(1, 2, 3) = (2, 3) and this can be seen from the matrices as
well:  
1 0 0
M(1,2) M(1,2,3) = 0 0 1 .
0 1 0
Exercise 4.1.5. Prove that the correspondence given above between S3 and
GL(n, K) is a group homomorphism, that is, a matrix representation of S3 .
Example 4.1.6. Recall from 3.3 the dihedral group Dn of order 2n. It contains
the cyclic group Cn of order n with generator c as a subgroup of index 2 and has
a reflection a (of order 2) with aca = c−1 . The group Dn has a two-dimensional
real representation r : Dn → GL(2, R) determined by
   
k cos(2kπ/n) − sin(2kπ/n) 0 1
c 7→ , a 7→ .
sin(2kπ/n) cos(2kπ/n) 1 0

Example 4.1.7. An important example of a linear representation of a finite

group G is the regular representation. Let G = {g1 , . . . , gn } and take V to be
the vector space of dimension n and basis {eg1 , . . . , egn }. Define ρ : G → GL(V )
by ρ(gi )(egj ) = egi gj , extended linearly to all of V .
Example 4.1.8. The above examples can be summarized in a more general
construction. Let π : G → Sn be a permutation representation. Then π gives
rise to a linear representation Mπ : G → GL(V ), where V is a complex vector
space with (formal) basis ei (i = 1, . . . , n), in the following way:

Mπ (g)ej = eπ(g)j (g ∈ G; j ∈ {1, . . . , n}).

4.1. LINEAR REPRESENTATIONS OF GROUPS 53

The regular representation of Example 4.1.7 is Mπ with π : G → Sym(G)

the left regular permutation representation of G.
The matrix representation of Example 4.1.4 is Mπ , where π is the natural
permutation representation of Sn (actually, for n = 5, 3, respectively).
Example 4.1.9. An important class of constructions of new representations
from known ones arises from function space constructions. Let r : G → GL(V )
be a linear representation of G on V . Consider the vector space F (V ), of all
maps V → K with pointwise addition (f + g)x = f (x) + g(x) and pointwise
scalar multiplication (λf )x = λf (x) (where λ ∈ K, x ∈ V , f, f 0 : V → K). Then
r induces the following representation r∗ : G → GL(F (V )):

r∗ (g)(f )x = f (r(g)−1 x) (x ∈ V, f ∈ F (V ), g ∈ G).

As done before, we usually unwind the heavy notation by dropping r wherever

possible, and write (gf )(x) = f (g −1 x) instead.
Note that F (V ) has the dual space V ∗ of K-linear functions as an invariant
subspace.
Exercise 4.1.10. Let R[x]≤3 be the linear space of all polynomials in x of degree
≤ 3. For a ∈ R we consider the R-linear representation ρa : R → GL(R[x]≤3 )
defined as ρa (r)p = p(x − ar) (r ∈ R, p ∈ R[x]≤3 ). Find a corresponding matrix
representation.

Equivalence and Subrepresentations

Just as with groups, we are mainly interested in representations up to isomor-
phism.
Definition 4.1.11. Let r and s be representations of a group G on vector
spaces V and W , respectively. A linear map T : V → W is said to intertwine r
and s if s(g)T = T r(g) for all g ∈ G. In this case, T is called a homomorphism
of G-modules. The space of all homomorphisms of G-modules is denoted by
HomG (V, W ).
The representations r and s are called equivalent if there exists a linear
isomorphism T : V → W intertwining them. In this case, T is an isomorphism
of G-modules.
It is straightforward to prove that this really defines an equivalence relation
∼ on representations.
Exercise 4.1.12. Consider the linear representations rl of Cn defined in Exam-
ple 4.1.3. For which pairs (l, m) are the representations rl and rm equivalent?
Exercise 4.1.13. Consider the linear representations ρa of Exercise 4.1.10.
Show that ρa ∼ ρb if and only if ab 6= 0.
In Example 4.1.9, we saw that V ∗ is a invariant under the action of G on
F (V ). This gives rise to the following definition.
54 CHAPTER 4. REPRESENTATION THEORY

Definition 4.1.14. Let r be a representation of a group G on V . A subspace

W is invariant if r(g)W ⊂ W for all g ∈ G. In this case, r|W : G → GL(W )
given by r|W (g) = r(g)|W , is called a subrepresentation of r, and W is called a
sub-G-module of V .
If G is clear from the context, we leave it out and call W a submodule of V .
Example 4.1.15. Suppose that T : V → W is a G-module homomorphism
of G on V and W . Then ker T is a G-invariant subspace of V and im T an
invariant subspace of W .
Example 4.1.16. Consider the two-dimensional matrix representation ρ : R →
GL(R2 ) defined by  
1 t
Mt = .
0 1
Then it is not hard to see that the 1 dimensional subspace spanned by (1, 0) is
an invariant subspace.
Example 4.1.17. Let V be a finite-dimensional vector space over K, and let
ρ : G → GL(V ) be a representation of a group g. Let P (V ) be the set of all poly-
nomial functions on V , i.e., the set of all elements of F (V ) (see Example 4.1.9)
that can be written as a K-linear combination of products of elements of V ∗ .
The subspace P (V ) of F (V ) is invariant under G, and so are its homogeneous
components P (V )d , defined by

P (V )d = {f ∈ P (V ) | f (λx) = λd f (x) for all x ∈ V and λ ∈ K}.

We find that P (V )0 is the one-dimensional space of constant functions, and

P (V )1 = V ∗ has dimension n := dim V . Choosing a basis x1 , . . . , xn of V ∗ , we
find that P (V )d is precisely the K-linear span of monomials xa1 1 · · · xann where
(a1 , . . . , an ) ∈ Nn satisfies a1 + . . . + an = d. This implies dim P (V )d = n+d−1


d ;
prove this!
To make all this more explicit, suppose that n = 2, and consider an element
A ∈ GL(V ) whose matrix is  
a b
c d
with respect to the basis e1 , e2 dual to x1 , x2 .
We compute the matrix of A on P (V )1 with respect to x1 , x2 . We have
 
−1 −1 d −b
A = (ad − bc) ,
−c a
so
Ax1 (e1 ) = x1 (A−1 e1 ) = (ad − bc)−1 x1 (de1 − ce2 ) = (ad − bc)−1 d
Ax1 (e2 ) = x1 (A−1 e2 ) = (ad − bc)−1 x1 (−be1 + ae2 ) = −(ad − bc)−1 b
Ax2 (e1 ) = x2 (A−1 e1 ) = (ad − bc)−1 x2 (de1 − ce2 ) = −(ad − bc)−1 c
Ax2 (e2 ) = x2 (A−1 e2 ) = (ad − bc)−1 x2 (−be1 + ae2 ) = (ad − bc)−1 a
4.1. LINEAR REPRESENTATIONS OF GROUPS 55

from which we conclude Ax1 = (ad − bc)−1 (dx1 − bx2 ) and Ax2 = (ad −
bc)−1 (−cx1 + ax2 ). Hence A has matrix
 
d −c
(ad − bc)−1
−b a

on the basis x1 , x2 of P (V )1 . Note that this is the tranposed inverse of A.

Finally, let us compute the matrix of A on P (V )2 with respect to the basis
x21 , x1 x2 , x22 . Now

Ax21 = (Ax1 )2 = (ad − bc)−2 (dx1 − bx2 )2

= (ad − bc)−2 (d2 x21 − 2bdx1 x2 + b2 x22 )
A(x1 x2 ) = (Ax1 )(Ax2 ) = (ad − bc)−2 (dx1 − bx2 )(−cx1 + ax2 )
= (ad − bc)−2 (−cdx21 + (ad + bc)x1 x2 − abx22 )
Ax22 = (Ax2 )2
= (ad − bc)−2 (−cx1 + ax2 )2
= (ad − bc)−2 (c2 x21 − 2acx1 x2 + a2 x22 ),

so the matrix sought for is

d2 c2
 
−cd
−2 
(ad − bc) −2bd ad + bc −2ac .
b2 −ab a2

In conclusion, we have found a matrix representation of GL(2, K) on K3

sending A to the 3 × 3 matrix just computed. Can you determine its kernel?

Exercise 4.1.18. Find the invariant subspaces of the matrix representation

ρ : R → GL(R2 ) defined by
 t
e et − 1

Mt = .
0 1

Lemma 4.1.19. Suppose that G is finite, and that ρ is a representation of

G on a vector space V . Let U be a sub-G-module of V . Then there exists a
sub-G-module W of V such that V = U ⊕ W .

Proof. Let W 0 be any vector space complement of U in V , and let π 0 be the

projection of V onto U with kernel W 0 . Define a new map
1 X
π := ρ(g)π 0 ρ(g −1 )
|G|
g∈G

on V . We claim that π is a projection commuting with all ρ(h), h ∈ G. To see

that it does indeed commute with ρ(h), consider

1 X
πρ(h) = ρ(g)π 0 ρ(g −1 h);
|G|
g∈G
56 CHAPTER 4. REPRESENTATION THEORY

by the dummy transformation g 7→ hg this equals

1 X
ρ(hg)π 0 ρ((hg)−1 h)
|G|
g∈G
1 X
= ρ(h)ρ(g)π 0 ρ(g −1 h−1 h)
|G|
g∈G
1 X
= ρ(h) ρ(g)π 0 ρ(g −1 )
|G|
g∈G

= ρ(h)π,
as claimed. To verify that π is a projection, we first note that ππ 0 = π 0 ; indeed,
both are zero on W 0 and the identity on U . Compute
1 X
π2 = π ρ(g)π 0 ρ(g −1 )
|G|
g∈G
1 X
= ρ(g)ππ 0 ρ(g −1 )
|G|
g∈G
1 X
= ρ(g)π 0 ρ(g −1 )
|G|
g∈G

= π,
where the second equality follows from the fact that π commutes with each
ρ(g). This concludes the proof that π is a projection. This implies that V =
im(π)⊕ker(π), where both im(π) and ker(π) are sub-G-modules of V by Exercise
4.1.15. Now note that im(π) = im(π 0 ) = U , so that we may take W = ker(π)
to conclude the proof of the lemma.
Remark 4.1.20. Let ρ be a linear representation of G on a finite-dimensional
complex vector space V . Suppose that V is endowed with a Hermitian inner
product (., .) satisfying
(ρ(g)(v), ρ(g)(w)) = (v, w) ∀v, w ∈ V, g ∈ G.
Now if U is a G-invariant subspace of V , then the orthogonal complement U ⊥
of U in V is an invariant subspace of V complementary to U .
It is important note that, if G is finite, one can always build an invariant
Hermitian inner product (.|.) from an arbitrary one (., .) as was done in Exercise
3.2.22 for positive definite inner products:
X
(v|w) = (ρ(g)(v), ρ(g)(w)).
g∈G

Thus, the above argument gives an alternative proof of the lemma in the case
where V is a finite-dimensional complex vector space.
4.1. LINEAR REPRESENTATIONS OF GROUPS 57

Exercise 4.1.21. Show that in Exercises 4.1.10 and 4.1.18, there are invariant
subspaces that have no invariant complements.

Definition 4.1.22. A G-module V is called reducible if it has sub-G-modules

other than 0 and V ; it is called irreducible otherwise. If V is the direct sum of
irreducible sub-G-modules, then V is called completely reducible.

This terminology is also used for the corresponding representations. By def-

inition, each irreducible representation is completely reducible. Exercise 4.1.18
gives an example of a reducible completely reducible representation. In Exer-
cise 4.1.27 we shall see that not every reducible representation is completely
reducible.

Exercise 4.1.23. Consider the cyclic group Cn of order n with generator c. It

has a two-dimensional real representation defined by
 
k cos(2kπ/n) − sin(2kπ/n)
r : Cn → GL(2, R), c 7→ .
sin(2kπ/n) cos(2kπ/n)

Show that it is irreducible for n > 2. If we replace R by C, we obtain a complex

representation. Show that this representation is completely reducible.

Exercise 4.1.24. Show that any irreducible representation of an Abelian group

G over an algebraically closed field is one-dimensional.

For finite groups, irreducible representations are the building blocks of all
representations.

Theorem 4.1.25. For a finite group G, every finite-dimensional G-module is

completely reducible.

Proof. Proceed by induction. Suppose that the stament holds for all representa-
tions of dimension smaller than n, and let V be an n-dimensional representation.
If V is irreducible, then we are done. Otherwise there exists a sub-G-module U
with 0 ( U ( V . By Lemma 4.1.19, U has a invariant complementary sub-G-
module W . As both dim U and dim W are smaller than n, we may apply the
induction hypothesis, and find that U is the direct sum of irreducible sub-G-
modules U1 , . . . , Uk of U and W is the direct sum of irreducible sub-G-modules
W1 , . . . , Wl . But then

V = U1 ⊕ . . . ⊕ Uk ⊕ W1 ⊕ . . . ⊕ Wl .

Remark 4.1.26. The representation of R in Example 4.1.16 is reducible, but

it is not the direct sum of irreducibles. Exercise 4.1.21 also gives examples of
this phenomenon. This shows that we need at least some condition on G (like
finiteness) for Theorem 4.1.25 to hold.
58 CHAPTER 4. REPRESENTATION THEORY

Exercise 4.1.27. Let p be a prime, and consider the cyclic group Cp with
generator c. The map Cp → GL2 (Z/pZ) defined by
 
1 k
ck 7→
0 1

is a matrix representation of Cp over Z/pZ. The one-dimensional subspace

spanned by (1, 0)T is invariant, but it has no invariant complement.
In general, representation theory of a group G over a field of characteristic
p tends to be much harder if p divides |G| than otherwise.

An important result about irreducible representations known as Schur’s

Lemma.

Lemma 4.1.28 (Schur’s Lemma). Suppose that K is algebraically closed, and

let V be an irreducible finite-dimensional G-module over K. Then any homo-
morphism of G-modules V → V is a scalar.

In other words, a matrix commuting with all matrices of an irreducible ma-

trix representation is a multiple of the identity matrix.

Proof. Let T : V → V be a homomorphism of G-modules, and let λ ∈ K be

an eigenvalue of T . Then T − λI is also a homomorphism of G-modules. By
Example 4.1.15, its kernel is a sub-G-module of V , and it is non-zero as λ is an
eigenvalue of T . As V is irreducible, ker(T − λI) equals V .

Exercise 4.1.29. Compute the set of all 2 × 2 matrices commuting with Cn

in one of the 2-dimensional complex representations given in Example 4.1.6.
Conclude that the representation is reducible.

Exercise 4.1.30. Prove that the following converse of Schur’s lemma holds:
Let r : G → GL(V ) be a completely reducible representation of a group G. For
r to be irreducible, it suffices that any linear map A : V → V that commutes
with all r(g), g ∈ G, is a scalar.

Exercise 4.1.31. Suppose V is a vector space of dimension n with a subspace

W of dimension k where 1 < k < n, and let G be the group of elements in
GL(V ) that leave W invariant. Then the embedding r : G → GL(V ) is clearly
a faithful reducible representation of G on V .

1. Establish that r is not completely reducible.

2. Show that the only G-module homomorphisms A : V → V are scalar

multiplications.

3. Conclude that the converse of Schur’s lemma does not necessarily hold if
G is not completely reducible, cf. Exercise 4.1.30.
4.2. DECOMPOSING DISPLACEMENTS 59

1 2

Figure 4.1: A system with 3 point masses and springs.

1 2

Figure 4.2: A displacement.

4.2 Decomposing Displacements

Consider the system of Figure 4.1. It√depicts three √ point masses (labelled with
P = {1, 2, 3}) at positions (− 12 , − 12 3), (− 12 , 12 3), (0, 1) in two-dimensional
space E = R2 , and these masses are connected with springs of equal lengths.
The system can be thought of as vibrating (in two dimensions) about its equi-
librium state, in which the point masses form an equilateral triangle.
A displacement of the system is a function f : P → R2 which attaches to
each point mass a velocity vector. Displacements can be depicted as in Figure
4.2. The set of all displacements is an R-linear space, which we denote by Γ.
Its dimension is clearly 3 × 2 = 6, i.e., there are 6 degrees of freedom.
We would like to decompose Γ into subspaces in a manner that is consis-
tent with the symmetry group D3 that acts on the system. To be precise, we
60 CHAPTER 4. REPRESENTATION THEORY

3 3

(1,2,3)

1 2 1 2

Figure 4.3: The action of (1, 2, 3) on a displacement.

have a permutation representation π : D3 → Sym(P ). The map is in fact an

isomorphism, and we shall identify D3 with Sym(P ) = S3 . For example, the
permutation (1, 2) corresponds to reflection in the line bisecting the edge 12 and
meeting point 3.
The group D3 also acts linearly on E. The corresponding linear representa-
tion D3 → GL(E) is given by
   √ 
−1 0 1 −1 3
(1, 2) 7→ and (1, 2, 3) 7→ √ .
0 1 2 − 3 −1

The space Γ becomes a D3 -module by setting

(gf )(i) = gf (g −1 i), i ∈ P, f ∈ Γ, g ∈ G.

Our goal can now be formalized as follows: try to write Γ as a direct sum of
irreducible submodules. Theorem 4.1.25 tells us that this is possible.
Here, we shall try to solve this problem ‘by hand’, without using too much
representation theory. In this way, the reader will appreciate the elegant char-
acter arguments which are to be given in Example 5.2.23.
As a first step, we can write each displacement in a unique way as a purely
‘radial’ displacement and a purely ‘tangental’ displacement (see Figure 4.4), and
the spaces Γrad and Γtan of radial and tangental displacements, respectively, are
three-dimensional submodules of Γ.
The submodule Γrad contains a one-dimensional submodule Γ1rad , consisting
of displacements with the ‘same’ velocity vector in all three directions, such as
the one depicted in Figure 4.5. To be precise: this space consists of all radial
displacements f for which (1, i)(f (1)) = f (i).
4.2. DECOMPOSING DISPLACEMENTS 61

3 3

1 2 1 2

Figure 4.4: A radial (left) and a tangental (right) displacement.

1 2

Figure 4.5: A special radial displacement.

62 CHAPTER 4. REPRESENTATION THEORY

3 3

1 2 1 2

Figure 4.6: Two more special radial displacements.

The space Γ1rad must have an invariant complement in Γrad , and looking at
the pictures we find that the space Γ2rad of radial displacements spanned by the
two in Figure 4.6 is also invariant. Note that this space consists of all radial
displacements for which (1, 3)f (1) + (1, 2)f (1) + f (1) = 0.
One may wonder if the two-dimensional module Γ2rad can be decomposed
further, but this is not the case, as we shall see in Example 5.2.23. The rep-
resentation Γtan can be decomposed in a similar fashion. However, instead of
going into details rightaway, we will first treat more representation theory, and
come back to this application in Example 5.2.23.

Notes
This section is based on a similar (but more difficult) example in [10]. If your
interest is aroused by this example, you are encouraged to read the pages in
that book concerned with the displacemements of a tetrahedral molecule.
Chapter 5

Character Tables

In this chapter, G denotes a finite group and K an algebraically closed field of

characteristic 0, e.g. C.

5.1 Characters
An extremely useful notion regarding matrices is the trace. The trace of a matrix
A = (aij )n×n is tr(A) = a11 + · · · + ann .
Exercise 5.1.1. Prove that for two matrices A and B, one has tr(AB) =
tr(BA).
Exercise 5.1.2. Show that in the case of a permutation representation π : G →
Sn , the number tr(Mπ (g)) (see Example 4.1.8) is equal to the number of fixed
points of the permutation π(g).
Definition 5.1.3. Let r be a representation of a group G on a finite-dimensional
vector space V . The character of r is the function χr defined on G by

χr (g) = tr(r(g)) (g ∈ G).

If r is irreducible, then the character χr is also called irreducible.

Thus, the character of r on g is the sum of the eigenvalues of r(g) counted
with multiplicities.
Remark 5.1.4. If G has finite order n, and ρ : G → GL(V ) is a finite-
dimensional representation, then ρ(g)n = I for all g ∈ G. Hence, over the
algebraic closure of K, each ρ(g) is diagonalizable, and all its eigenvalues are
n-th roots of unity. If K = C, then all eigenvalues have absolute value 1.
We next give some properties of complex characters.
Lemma 5.1.5. For any complex representation r : G → GL(V ) of a finite
group, we have:

63
64 CHAPTER 5. CHARACTER TABLES

1. χr (e) = tr(I) = dim V ;

2. χr (g −1 ) = χr (g);
3. χr (ghg −1 ) = χr (h) for all g, h ∈ G.
Proof. The first part is easy. As for the second part, we have seen that all the
eigenvalues λ1 , . . . , λn of r(g) have absolute value 1. Therefore:
X X
χr (g −1 ) = tr(r(g −1 )) = tr(r(g)−1 ) = λ−1
i = λi = tr(r(g)) = χr (g).
The last part follows directly from Exercise 5.1.1.
The last part states that characters are constant on conjugacy classes of G.
Example 5.1.6. Take G = GL(2, K). The trace of an arbitrary element
 
a b
A=
c d
is a+d. This is the character of the identity, which in this case is a representation
(often called the natural representation).
But, as we have seen in Example 4.1.17, there are also linear actions on
P (V )1 and on P (V )2 . We read off the corresponding characters as traces of the
matrices found in 4.1.17. They are (a + d)/(ad − bc) on P (V )1 , and (a2 + ad +
bc + d2 )/(ad − bc)2 on P (V )2 .
Lemma 5.1.7. Let V1 , . . . , Vs be representations of G with characters respec-
tively, χ1 , . . . , χs . Then the character of the representation V1 ⊕ · · · ⊕ Vs is the
sum of characters χ1 + · · · + χs .

5.2 Orthogonality of irreducible characters

The following theorem is very important in the representation theory of finite
groups.
Theorem 5.2.1. Let G be a finite group. The irreducible complex characters of
G form an orthonormal basis for the space of class functions on G, with respect
to the Hermitian inner product (· | ·).
We need a definition to make clear what this theorem states.
Definition 5.2.2. A class function on a group G is a function f : G → C that is
constant on conjugacy classes, or, equivalently, that is fixed by the linear action
of G on CG defined by
(g · f )(h) := f (ghg −1 ).
The space of all class functions is denoted by H. The Hermitian inner product
is given by
1 X
(χ | ψ) = χ(g)ψ(g)
|G|
g∈G

for χ, ψ ∈ H.
5.2. ORTHOGONALITY OF IRREDUCIBLE CHARACTERS 65

We have noticed earlier that characters are class functions. We divide the
proof of Theorem 5.2.1 into two parts: first, we will show the orthonormality
of the irreducible characters, and then their being a full basis of the space of
class functions. For our first purpose, we introduce the tensor product of two
representations.
Definition 5.2.3. Let V and W be vector spaces over K. Then V ⊗ W , called
the tensor product of V and W , is the quotient of the free vector space over K
with basis {v ⊗ w | v ∈ V, w ∈ W } by its subspace spanned by all elements of
the following forms (for v, v1 , v2 ∈ V , w, w1 , w2 ∈ W and λ ∈ K):

(v1 + v2 ) ⊗ w − v1 ⊗ w − v2 ⊗ w,
v ⊗ (w1 + w2 ) − v ⊗ w1 − v ⊗ w2 ,
(λv) ⊗ w − λ(v ⊗ w), and
v ⊗ (λw) − λ(v ⊗ w).

Exercise 5.2.4. Show that if v1 , . . . , vm and w1 , . . . , wn are bases of V and W ,

respectively, then

{vi ⊗ wj | i = 1, . . . , m, j = 1, . . . , n}

is a basis of V ⊗ W .
It is not hard to see that if U, V, and W are vector spaces over K, then
the spaces (U ⊗ V ) ⊗ W and U ⊗ (V ⊗ W ) are canonically isomorphic. We
will identify them, leave out parentheses, and write T d (V ) for the d-fold tensor
product of V with itself. The direct sum of all T d (V ) for d ∈ N (including 0,
for which T d (V ) is isomorphic to K) forms an associative algebra with respect
to the operator ⊗; this algebra is called the tensor algebra on V and denoted
by T (V ).
Definition 5.2.5. Let V be a vector space over K and let d be a natural number.
Then the quotient of T d (V ) by the subspace spanned by the set

{v1 ⊗ v2 ⊗ . . . ⊗ vd − vπ(1) ⊗ vπ(2) ⊗ . . . ⊗ vπ(d) | π ∈ Sd }

is denoted by S d (V ), and called the d-th symmetric power of V .

Just like T (V ), the direct sum of all S d (V ) for d ∈ N forms an associative
algebra, called the symmetric algebra on V and denoted by S(V ). This algebra
is commutative, and to stress this fact we usually leave out the operator ⊗ from
expressions in S(V ).
Exercise 5.2.6. Show that if v1 , . . . , vn is a basis of V , then the set

{v1m1 · · · vnmn | m1 , . . . , mn ∈ N, m1 + . . . + mn = d}

is a basis for S d (V ). Prove that S d (V ∗ ) is canonically isomorphic to P (V )d (see

Example 4.1.17).
66 CHAPTER 5. CHARACTER TABLES

Definition 5.2.7. Let r : G → GL(V ) and s : G → GL(W ) be linear represen-

tations of G. Then we define a linear representation r ⊗ s : G → GL(V ⊗ W ) of
G by
(r ⊗ s)(g)(v ⊗ w) := (r(g)v) ⊗ (s(g)w),
extended linearly to all of V ⊗ W . This representation is denoted by r ⊗ s and
called the tensor product of r and s.
Lemma 5.2.8. Let G, r, V, s, and W be as in Definition 5.2.7. Then the char-
acter of r ⊗ s satisfies χr⊗s = χr χs .
Proof. Fix bases v1 , . . . , vm and w1 , . . . , wn of V and W , respectively. Consider,
for a fixed g ∈ G, the matrix A = (aij,kl ) of (r ⊗ s)(g) with respect to the basis
vi ⊗ wj of V ⊗ W ; We have
X
(r ⊗ s)(g)(vk ⊗ wl ) = aij,kl vi ⊗ wj . (5.1)
i,j

On the other hand, by the definition of the tensor representation, the left-hand
side equals
X X X
(r(g)vk ) ⊗ (s(g)wl ) = ( rik vi ) ⊗ ( sjl wj ) = rik sjl vi ⊗ wj . (5.2)
i j i,j

Comparing the right-hand sides of equations (5.1) and (5.2) we obtain

aij,kl = rik sjl ,

and therefore
X X
χs⊗r (g) = aij,ij = rii sjj = χr (g)χs (g),
i,j i,j

concluding the proof.

Exercise 5.2.9. Let V be a G-module, and consider the linear isomorphism
σ : V ⊗ V → V ⊗ V determined by σ(x ⊗ y) = y ⊗ x for x, y ∈ V . Show that σ
has precisely two eigenspaces, with eigenvalues 1 and −1, and that both spaces
are G-invariant. Here G acts on V ⊗ V via the tensor product representation
of the linear representation of G on V (given by the G-module structure of V )
with itself.
Another important ingredient in our proof of Theorem 5.2.1 is the so-called
dual representation.
Definition 5.2.10. Let r : G → GL(V ) be a representation of the group G.
Then we write V ∗ for the dual of V (in the notation of Example 4.1.17, this is
P (V )1 ), and r∗ for the corresponding representation, given by

(r∗ (g)f )(v) := f (r(g)−1 v).

5.2. ORTHOGONALITY OF IRREDUCIBLE CHARACTERS 67

∗
Exercise 5.2.11. Show that χr = χr .

One final lemma before proceeding with the proof of Theorem 5.2.1.

Lemma 5.2.12. Let r, V, s, W, G be as in Definition 5.2.7. Define a repre-

sentation t of G in Hom(V, W ), the space of all C-linear maps from V to W
by
t(g)H := s(g)Hr(g)−1 , H ∈ Hom(V, W ).
This representation is equivalent to s ⊗ r∗ .

Proof. Define T : W ⊗ V ∗ → Hom(V, W ) by

T (w ⊗ f )(v) = f (v) · w, v ∈ V, f ∈ V ∗ , w ∈ W,

and extend it linearly. It is a matter of elementary linear algebra to see that T

is a linear isomorphism. We claim that it intertwines the representations t and
s ⊗ r∗ . The best way to see this is to do the computation yourself, so we leave
it as an exercise for the reader.

Exercise 5.2.13. Prove that the map T defined in the proof of Lemma 5.2.12
is really an intertwining map.

If we choose a basis (vj )j of V with dual basis (v j )j of V ∗ , and a basis (wi )i

of W , then the map T sends wi ⊗v j to the linear map V → W having the matrix
Eij (the matrix with 1 on position (i, j), and zeroes elsewhere) with respect to
the bases (vj )j and (wi )i .
Having completed all preparations, of which the tensor product is a very
useful tool in its own right, we proceed with the first part of the proof of this
section’s main theorem.

Proof of Theorem 5.2.1, first part. Let r, V, s, W, G be as in Definition 5.2.7 and

t as in Lemma 5.2.12, and moreover, assume that both representations r and
s are irreducible. We wish to compute the inner product (χs | χr ). It can be
written in a particularly nice form, using tensor products.
Define the linear map P : Hom(V, W ) → Hom(V, W ) by

1 X
P = t(g).
|G|
g∈G

The trace of P equals

1 X t 1 X s ∗
χ (g) = χ (g)χr (g)
|G| |G|
g∈G g∈G
1 X s
= χ (g)χr (g))
|G|
g∈G

= (χ | χr ).
s
68 CHAPTER 5. CHARACTER TABLES

Here we used Lemmas 5.2.12, 5.2.8 and Exercise 5.2.11.

We claim that P is a projection from Hom(V, W ) onto the space HomG (V, W ).
To show that it is indeed a projection, let H ∈ Hom(V, W ), and compute
1 X 1 X
P 2H = s(g)s(h)Hr(h−1 )r(g −1 )
|G| g |G|
h
1 X 1 X
= s(gh)Hr((gh)−1 )
|G| g |G|
h
1 X 1 X
= s(h)Hr(h−1 )
|G| g |G|
h
1 X
= PH
|G| g
= P H,

where in third equality the dummy h is replaced by g −1 h. This proves that P

is a projection. To see that im(P ) is contained in HomG (V, W ), let g0 ∈ G,
H ∈ Hom(V, W ), and v ∈ V , and compute
1 X
(P H)(g0 v) = s(g)Hr(g −1 g0 )v
|G| g
1 X
= s(g0 g)Hr(g −1 g0−1 g0 )v
|G| g
1 X
= s(g0 ) s(g)Hr(g −1 )v
|G| g
= g0 (P Hv),

where in the second equality, the dummy g is replaced by g0 g. Finally, if H ∈

HomG (V, W ), then
1 X
PH = s(g)Hr(g −1 )
|G| g
1 X
= s(g)s(g −1 )H
|G| g
= H,

so that im(P ) = HomG (V, W ). This proves the claim. We now distinguish two
cases.
First, suppose that r and s are not equivalent. Then, by Schur’s lemma, the
space HomG (V, W ) is zero-dimensional, so that P and, a fortiori, its trace are
0. In this case (χs | χr ) = 0.
Alternatively, suppose that r and s are equivalent. As we only wish to
compute the inner product of their characters, we may assume that they are
5.2. ORTHOGONALITY OF IRREDUCIBLE CHARACTERS 69

equal: W = V and s = r. In this case, the image of P is the linear span of the
identity map I on V . Taking any basis of Hom(V, V ) that starts with I, we see
that the matrix of P with respect to that basis has only non-zero elements on
the first row. The only contribution to the trace of P is therefore the element in
the upper left corner. Since P I = I, that element is 1. Hence (χr | χr ) = 1.

This proves that the characters are an orthonormal system in the space of
class functions on G.

Corollary 5.2.14. The number of non-isomorphic irreducible representations

of G is finite.

Proof. The dimension of H is the number of conjugacy classes of G. The irre-

ducible characters form an orthonormal, whence independent, set in this finite-
dimensional space by the above.

Lemma 5.2.15. Let r be a complex representation of G on a space V , and

write
V = V1 ⊕ · · · ⊕ Vt ,
where the Vi are irreducible submodules. Let s be an irreducible representation of
G on a space W . Then the number of i for which Vi is isomorphic to W is equal
to (χr |χs ). In particular, this number does not depend on the decomposition in
irreducibles.

Proof. If χi is the character of Vi then the character χr of V can be written as

χr = χ1 + · · · + χt . Hence

(χr |χs ) = (χ1 |χs ) + · · · + (χt |χs ),

and by the above (χi |χs ) is one if Vi ∼

= W , and zero otherwise.

Corollary 5.2.16. Two representations of a finite group are equivalent if and

only if they have the same character.

Recall that the regular representation ρ : G → GL(V ) of the group G is

defined by ρ(g)(eh ) = egh , with V the vector space of dimension |G| and basis
{eg | g ∈ G}.

Proposition 5.2.17. The character of the regular representation ρ of G satis-

fies ρ(g) = δg,e |G| (g ∈ G).

Proof. Since eg = g, the transformation ρ(e) fixes each element g of G. There-

fore, tr(ρ(e)) = n = |G|. If g 6= e, then eh 6= egh for each h ∈ G. This implies
that tr(ρ(g)) = 0.

Lemma 5.2.18. Every irreducible representation W of G is a sub-representation

of the regular representation with multiplicity dim W .
70 CHAPTER 5. CHARACTER TABLES

Proof. From Proposition 5.2.17 we have

1 X ρ 1
(χρ | χs ) = χ (h)χs (h) = |G|χs (e) = dim(W ),
|G| |G|
h∈G

and the result follows from Lemma 5.2.15.

Corollary 5.2.19. Let r1 , r2 , . . . , rs be the irreducible representations of G.

Then X
|G| = n2i
where ni is the dimension of the representation ri .

Proof. We saw in the lemma that the regular representation ρ can be written
as a sum of all irreducible representations ri of G. Moreover, each of these
representations appears ni times in the decomposition. So
n1 times ns times
z }| { z }| {
ρ = r1 ⊕ . . . ⊕ r1 ⊕ . . . ⊕ rs ⊕ · · · ⊕ rs .
Ps Ps
Therefore |G| = χρ (e) = i=1 ni χri (e) = i=1 n2i .

Suppose that we manage to construct some non-isomorphic irreducible rep-

resentations of G. Then Corollary 5.2.19 can be used to decide whether we have
already find all.
In order to finish the proof of Theorem 5.2.1, we still need to show that the
irreducible characters span H.

Lemma 5.2.20. Suppose that r : G → GL(V ) is an irreducible n-dimensional

complex representation of G, and let f ∈ H. Then the linear map
1 X
H := f (g)r(g)
|G|
g∈G

equals
1
(f | χr )I.
n
Proof. First of all, note that H verifies r(g)H = Hr(g) for all g ∈ G. Hence, as
r is irreducible, H must be multiplication by a scalar λI. Its trace equals nλ.
On the other hand, it equals
1 X
f (g)χr (g) = (f | χr ).
|G|
g∈G

Consequently,
λ = (f | χr )/n.
5.2. ORTHOGONALITY OF IRREDUCIBLE CHARACTERS 71

Proof of Theorem 5.2.1, second part. In order to prove that the irreducible char-
acters form a basis of H, it suffices to prove that the orthoplement of their linear
span in H is zero. Hence, let f ∈ H be orthogonal to all irreducible characters.
According to Lemma 5.2.20, the linear map

1 X
H= f (g)r(g)
|G|
g∈G

is zero for all irreducible representations r. But, writing an arbitrary represen-

tation as a direct sum of irreducible ones, we see that this linear map is zero for
any representation r. Let us apply this knowledge to the regular representation.
Compute
1 X 1 X
Heh = f (g)r(g)eh = f (g)egh .
|G| |G|
g∈G g∈G

Thus, for H to be zero, f must be identically zero. This concludes the proof.

Theorem 5.2.21. The number of irreducible characters equals the number of

conjugacy classes of G.

Proof. The dimension of H equals the number of conjugacy classes of G, and

the irreducible characters form a basis of this linear space.

Half of the following proposition is Exercise 4.1.24, and can be done without
character theory.

Proposition 5.2.22. The group G is Abelian if and only if all of its irreducible
representations are one dimensional.

Proof. Denote by s the number of irreducible characters of G and denote their

dimensions, as before by ni .
If G is Abelian, all conjugacy classes consist of a single element of G. Then
by the theorem above, there are |G| irreducible representations of G. Hence by
Corollary 5.2.19, all ni ’s are equal to 1. This settles one implication.
As for the converse, suppose that ni = 1 for all i. Then |G| = s and so there
are |G| distinct conjugacy classes of G. Therefore, G is Abelian.

Example 5.2.23. Recall from Section 4.2 the D3 -module Γ of displacements.

Let us employ character theory to find the decomposition of Γ into irreducible
modules. First of all, note that D3 has three conjugacy classes, with represen-
tatives (1), (12) and (123). Hence D3 has three distinct irreducible characters
by Theorem 5.2.21. Let V = R3 , and let s : D3 → GL(V ) be the linear rep-
resentation corresponding to the permutation representation of D3 on {1, 2, 3}
(see Example 4.1.8). Then

χs ((1)) = 3, χs ((1, 2)) = 1, and χs ((1, 2, 3)) = 0.

72 CHAPTER 5. CHARACTER TABLES

We have (χs |χs ) = (9 + 3 ∗ 1)/6 = 2, so that V is reducible. Indeed, the line

spanned by (1, 1, 1) ∈ V is invariant. Let r1 be the representation of D3 on this
line, and let r2 be the representation of D3 on an invariant complement. Then

χr1 ((1)) = 1, χr1 ((1, 2)) = 1, and χr1 ((1, 2, 3)) = 1

and
χr2 ((1)) = 2, χr2 ((1, 2)) = 0, and χr2 ((1, 2, 3)) = −1.
Both χr1 and χr2 have squared norm 1, so they are irreducible. The represen-
tation of D3 on E is equivalent to r2 , as can be read of from the traces of the
matrices on page 4.2.
An element of Γ can be extended in a unique way to a linear map V → E, so
that we may identify Γ with the space HomK (V, E) of K-linear maps from V to
E. By Lemma 5.2.12, HomK (V, E) is isomorphic to V ∗ ⊗E; let r : D3 → V ∗ ⊗E
be the corresponding representation. Lemma 5.2.8 shows that χr = χs χr2 .
Hence,
χr ((1)) = 6, χr ((1, 2)) = 0, and χr ((1, 2, 3)) = 0.
First, we can decompose χr = (χr1 + χr2 )χr2 = χr1 χr2 + (χr2 )2 = χr2 +
(χr2 )2 . The character (χr2 )2 has squared norm 3, so it is reducible. We have
(χr1 |(χr2 )2 ) = 1 and also (χr2 |(χr2 )2 ) = 1, so that (χr2 )2 − χr1 − χr2 is the char-
acter of a third irreducible representation, say r3 , of dimension 1. Its character
equals
χr3 ((1)) = 1, χr3 ((12)) = −1, and χ(r3 ) (123) = 1,
that is, χr3 = sgn . By Theorem 5.2.21, the characters r1 , sgn , r2 exhaust all
irreducible characters of D3 . We conclude that the character χr of D3 on Γ
decomposes into χr1 + 2χr2 + χr3 .

5.3 Character tables

Let χ1 , χ2 , . . . , χs be the characters of the distinct irreducible representations of
G. Let C1 , C2 , . . . , Cs be the distinct conjugacy classes of G, with representatives
g1 , . . . , gs , respectively.
Construct a table with rows labelled by χ1 , χ2 , . . . , χs and columns labelled
by g1 , g2 , . . . , gs ; under each gi we write the size of Ci . The (i, j)-entry of the
table is χi (gj ). It is common to take C1 = {e} = {g1 }, so that the first column
contains the dimensions of the respective characters, and also to take χ1 to be
the trivial representation, so that the first row contains only ones.
This table is called the character table of G.
Example 5.3.1. Let us check the character table of S3 . First notice that there
are three conjugacy classes: C1 = e, C2 = {(12), (13), (23)} consisting of all
transpositions and C3 = {(123), (132)} consisting of all 3-cycles. By Theorem
5.2.21, we know that there are three irreducible representations.
By convention, χ1 is the trivial character. For the second one we take the
sign representation. Recall this is the one that assigns to each cycle its sign
5.3. CHARACTER TABLES 73

(2-cycles are odd, 3-cycles are even). Since it is a 1-dimensional representation,

it is equal to its character.
Finally for the third representation, we proceed as follows. The permutation
representation π of S3 on C3 has an invariant subspace U spanned by (1, 1, 1).
Its orthogonal complement is also an invariant subspace, say V ; it consists of
all (z1 , z2 , z3 ) ∈ C3 such that z1 + z2 + z3 = 0. The latter representation has
dimension 2 and is irreducible; it is called the standard representation of S3 .
To find its character, we observe that χV = χπ − χU . But U is just the trivial
representation and χπ is equal to 3 on C1 , 1 on C2 and 0 on C3 (why? see
Exercise 5.1.2)
We have found the complete character table.

S3 (1) (12) (123)

1 3 2
trivial U 1 1 1
sign 1 −1 1
standard V 2 0 −1

Proposition 5.3.2. The standard representation of Sn is irreducible for all

n > 1.

Proof. Let χ be the character of the n-dimensional linear representation corre-

sponding to the permutation representation of Sn on {1, . . . , n} (see Example
4.1.8), and let χ1 be the trivial character of Sn . We wish to show that χ − χ1 is
irreducible; to this effect, it suffices to prove that (χ|χ) = 2. Denoting by fix(σ)
the set of fixed points of σ ∈ Sn on {1, . . . , n}, we have χ(σ) = | fix(σ)|. Hence,
we find
1 X
(χ|χ) = | fix(σ)|2 .
n!
σ∈Sn

To evaluate the right-hand side of this equation we count the number of per-
mutations having precisely k fixed points (0 ≤ k ≤ n). This number is clearly
equal to
 
n
F (n − k),
k

where F (m) denotes the number of fixed-point-free permutations in Sm . A

standard inclusion-exclusion argument shows that

m
X m!
F (m) = (−1)i .
i=0
i!
74 CHAPTER 5. CHARACTER TABLES

Combining the above considerations we find

n   n−k
1 X 2 n X (n − k)!
(χ|χ) = k (−1)i
n! k i=0 i!
k=0
n n−k
X 1 X (−1)i
= k2
k! i=0 i!
k=0
=: Mn .

It is readily seen that M2 = 2. To prove that Mn = 2 for all n ≥ 2, we note

that for n ≥ 3 we have:
n
X 1
Mn − Mn−1 = (n − i)2 (−1)i ,
i=0
(n − i)!i!

and it is not hard to prove that the right-hand side is zero for n ≥ 3.

Example 5.3.3. Consider the cyclic group Cn with generator c, and let ζ ∈ C
be a primitive n-th root of unity. Define, for j = 0, . . . , n − 1, the map χj :
Cn → C by χj (ck ) := ζ jk . Then the χj are characters of Cn , and irreducible
as they are one-dimensional. By Theorem 5.2.21 there are no other characters,
so that the character table of Cn is as follows.

Cn e c c2 ... cn−1
1 1 1 ... 1
χ1 1 1 1 ... 1
χ2 1 ζ ζ2 ... ζ n−1
χ3 1 ζ2 ζ4 ... ζ 2(n−1)
· · · · ... ·
· · · · ... ·
· · · · ... ·
(n−1)2
χn 1 ζ n−1 ζ 2(n−1) ... ζ

Example 5.3.4. To compute the character table of S4 , we first count the

number of conjugacy classes. They are the identity, the 2-cycles, the 3-cycles,
the 4-cycles and the product of two distinct 2-cycles; five in total. Hence, by
Theorem 5.2.21, S4 has five distinct irreducible characters. It has three of them
in common with S3 (or, indeed, with any symmetric group; see Proposition
5.3.2), namely the trivial character, the sign and the standard character. They
are listed in the following partial character table.

S4 (1) (12) (123) (1234) (12)(34)

1 6 8 6 3
χ1 (trivial) 1 1 1 1 1
χ2 (sign) 1 −1 1 −1 1
χ4 (standard) 3 1 0 −1 −1
5.3. CHARACTER TABLES 75

By Lemma 5.2.19, the squares of the dimensions of the irreducible characters

sum up to |S4 | = 24, and we only have 1 + 1 + 9 = 11 so far. We conclude that
the two remaining irreducible characters have dimensions 2 and 3. The latter is
the product χ5 := χ2 χ4 , which is irreducible as it has norm 1. The remaining 2-
dimensional character χ3 can be found using the orthogonality relations among
characters. We then find the following character table of S4 .

S4 (1) (12) (123) (1234) (12)(34)

1 6 8 6 3
χ1 1 1 1 1 1
χ2 1 −1 1 −1 1
χ3 2 0 −1 0 2
χ4 3 1 0 −1 −1
χ5 = χ2 χ4 3 −1 0 1 −1

Exercise 5.3.5. Compute the number of elements in each conjugacy class of

S4 .
Exercise 5.3.6. Consider the permutation representation of S4 on partitions
of {1, 2, 3, 4} into two sets of size 2; that is, the action of S4 on the set

{{A, B} | |A| = |B| = 2, A ∪ B = {1, 2, 3, 4}, A ∩ B = ∅}.

Show that the character of the corresponding linear representation is χ1 + χ3 ;

this explains the character χ3 .
A more or less explicit description of the irreducible representations of Sn is
known for general n. It involves the beautiful combinatorics of Young tableaux,
a good textbook on which is Fulton’s book [5].
Exercise 5.3.7. Construct the character table of D4 .
Exercise 5.3.8. Compute the character table of the Klein group (Z/2Z) ⊕
(Z/2Z).
Example 5.3.9. Consider the subgroup T of the orthogonal group SO3 (R)
consisting of the proper rotations of the tetrahedron. We wish to compute its
character table. The conjugacy classes are those of (1), (234), (243) and (13)(24)
(make a picture that corresponds to this numbering), and their cardinalities are
1, 4, 4 and 3, respectively. Thus, there must be 4 irreducible characters, of
representations with dimensions m1 ≤ m2 ≤ m3 ≤ m4 . The sum of their
squares should equal |T | = 12. Their turns out to be only one solution, namely
m1 = m2 = m3 = 1 and m4 = 3. Hence, apart from the trivial character, there
should be two more characters of degree 1. Let r be one of them. As r is nothing
but a homomorphism from T to C∗ , the complex number r((234)) should have
multiplicative order 3; hence it is either ω = e2πi/3 or ω 2 . The image of (243) is
fixed in both cases, and so is the image of (13)(24), as χr should be orthogonal
to the trivial representation. This explains the first three lines in Table 5.1. The
last line can be obtained from the orthonormality conditions.
76 CHAPTER 5. CHARACTER TABLES

T (1) (234) (243) (13)(24)

1 4 4 3
χ1 1 1 1 1
χ2 1 ω ω2 1
χ3 1 ω2 ω 1
χ4 3 0 0 −1

Table 5.1: The character table of the tetrahedral group

Exercise 5.3.10. Find a representation corresponding to the last line of Table

5.1.

Exercise 5.3.11. Let G be a group and let H be a normal subgroup of G.

Let r be a representation of G with the property that H ⊆ ker r. Show that r
induces a representation of G/H. Conversely, any representation of G/H can
be lifted to one on G. Apply this principle to the group T , which has the Klein
4-group as a normal subgroup.

Exercise 5.3.12. Prove the column orthogonality of the character table: for
g, h ∈ G, and χ1 , . . . , χs a complete set of irreducible characters of G,
s
X
χi (g)χi (h) = |CG (h)|δg,h .
i=1

This formula is also useful for the construction of character tables.

Exercise 5.3.13. Let χ be the character of a complex linear representation of

G on V . Show that the character of G on the symmetric square S 2 (V ) of V
equals
χ(g)2 + χ(g 2 )
g 7→ .
2
What is the character of G on S 3 (V )? Use these results to find new characters
of A5 from the irreducible 4-dimensional character occurring in the standard
permutation representation of degree 5.

Exercise 5.3.14. Consider the alternating group A5 on 5 letters.

1. Write down representatives for each of its conjugacy classes.

2. Determine the sizes of the corresponding conjugacy classes.

3. Show that A5 has only one irreducible character of degree 1. (Hint: if

there were one more, then there would have to be a normal subgroup N
such that A5 /N is abelian.)
5.4. APPLICATION: SYMMETRY OF VIBRATIONS 77

4. The rotation group of the icosahedron is A5 . This gives an irreducible

representation ρ of A5 of degree 3. Write down its character.

5. Verify that the composition of ρ with conjugation by (1, 2) gives a non-

equivalent representation and determine its character.

6. Apply the symmetric square formula of Exercise 5.3.13 to ρ to find a

6-dimensional character. Show that this character splits into the trivial
character and an irreducible 5-dimensional character.

7. Conclude from the orthogonality of the character table that there is one
more irreducible character. Compute this character, and compare it with
the character of the restriction of the standard representation of S5 to A5 .

Exercise 5.3.15. Up to isomorphism, there are precisely two groups of order 8.

They have identical character tables. Exhibit these groups and their character
tables.

5.4 Application: symmetry of vibrations

There exist many spectroscopic techniques in chemistry that can be used to mea-
sure symmetry properties of molecules of a given substance. In this section, we
describe two of them: infrared (IR) spectroscopy and Raman spectroscopy. In
the former method, the substance is exposed to radiation of various frequencies,
and the intensity of the light leaving the substance is measured. If the radiation
excites the molecules from the ground state into vibration, then it is absorbed,
giving rise to a (downward) peak in the intensity (plotted as a function of the
frequency). As we have seen in Section 4.2, there are several essentially different
vibrations, corresponding to irreducible submodules of the representation of the
symmetry group of the molecule on the space of all displacements. As with Ra-
man spectroscopy—another technique whose physical details we do not discuss
here—only actual vibrations may contribute a peak to the intensity spectrum,
i.e., the (infinitesimal) isometries of three-dimensional space do not contribute.
However, not even all irreducible modules in the module of vibrations can be
observed by IR spectroscopy; due to technical properties of this technique, only
those irreducibles occurring also in the natural 3-dimensional representation do
actually contribute a peak in the spectrum. This distinguishes IR from Ra-
man spectroscopy, which allows to observe all irreducibles that occur in the
symmetric square of the aforementioned natural representation.
Hence, to predict the number of peaks using either of the above methods,
for a given molecule with n molecules, we proceed as follows.

1. Determine the symmetry group G of the molecule at hand, and compute

or look up its character table.

2. Determine the character χdis of G on the 3n-dimensional module of dis-

placements of the molecule, as well as its character χiso on the 6-dimensional
78 CHAPTER 5. CHARACTER TABLES

Cl

Cl
P
Cl Cl

Cl
Figure 5.1: The PCl5 molecule

submodule of (infinitesimal) isometries of R3 . Then

χvib := χdis − χiso

is the character of G on vibrations of the molecule.

3. Compute the character χ1 of G on R3 , and the character χ2 of G on the

symmetric square of R3 .

4. Then the number of peaks in IR spectroscopy is the number of irreducible

characters in χvib (counted with multiplicity) that also occur in χ1 , and
and the number of peaks in Raman spectroscopy is the number of irre-
ducible characters in χvib (counted with multiplicity) that also occur in
χ2 .

Example 5.4.1. We want to predict the number of peaks in IR and Raman

spectroscopy when applied to the molecule PCl5 of Figure 5.1. In these chemical
applications, it is common to use the Schoenflies notation (see Table 3.2), and
we will adhere to this tradition in the present example.

1. The symmetry group of this molecule is D3h , whose character table is as

follows (E is the identity, Ci stands for a rotation of order i, σh and σv
are reflections in a horizontal and vertical plane, respectively, and S3 is
the composition of a C3 and a σh ).
5.4. APPLICATION: SYMMETRY OF VIBRATIONS 79

D3h E C3 3C2 σh S3 σv
1 2 3 1 2 3
A01 1 1 1 1 1 1
A02 1 1 -1 1 1 -1
E0 2 -1 0 2 -1 0
A001 1 1 1 -1 -1 -1
A002 1 1 -1 -1 -1 1
E 00 2 -1 0 -2 1 0

2. Let χ1 denote the character of D3h in its natural representation on R3 ,

and let χ3 denote the character of D3h corresponding to its permutation
representation on the 6 molecules of PCl5 (see Example 4.1.8). Their
values are easily calculated to be as follows.
D3h E C3 3C2 σh S3 σv
χ1 3 0 -1 1 -2 1
χ3 6 3 2 4 1 4
As in Example 5.2.23, the character of the module of displacements equals
the dual of χ3 times χ1 . Having only real values, χ3 is self-dual, so that
we find the following values for χdis .
D3h E C3 3C2 σh S3 σv
χdis 18 0 -2 4 -2 4
This character can be decomposed into irreducible ones as follows:

χdis = 2A01 + A02 + 4E 0 + 3A002 + 2E 00

The character χiso is the sum of the characters of D3h on translations and
on rotations. The character of D3h on translations equals χ3 = E 0 + A002 ,
and the character χ4 on rotations turns out to be as follows.
D3h E C3 3C2 σh S3 σv
χ4 3 0 -1 -1 2 -1
We find that χ4 = A02 + E 00 . We conclude that

χvib = 2A01 + 3E 0 + 2A002 + E 00 .

3. We have already computed χ1 . Using Exercise 5.3.13 we find the following

values for its symmetric square χ2 .
D3h E C3 3C2 σh S3 σv
χ2 6 0 2 2 2 2
We find that χ2 = 2A01 + E 0 + E 00 , while χ1 = E 0 + A002 .
4. From the above, we conclude that only vibrations with characters E 0 or
A002 are observable with IR spectroscopy; the number of such irreducible
characters in χvib is 5, so that is the predicted number of peaks when
using this technique. Using Raman spectroscopy, only vibrations with
characters A01 , E 0 and E 00 can be observed; this leads to 6 peaks.
80 CHAPTER 5. CHARACTER TABLES

N N N Cl

Pd Pd

Cl Cl Cl N

cis−PdCl (NH ) trans−PdCl (NH )

2 3 2 2 3 2

Figure 5.2: Two configurations of PdCl2 (NH3 )2

The following exercises shows how spectroscopic techniques may or may not
be used to recognize molecules.

Exercise 5.4.2. The square planar coordination compound PdCl2 (NH3 )2 can
be prepared with two different configurations; see Figure 5.2. Is it possible
to discern the cis- and the trans-configurations using information concerning
Pd − Cl stretch vibrations in IR spectroscopy?

Exercise 5.4.3. Is it possible to deduce the symmetry of Ni(CO)4 (tetrahedron

or square plane) from the IR and/or Raman spectra of the CO-stretches?

Exercise 5.4.4. Let T be the octahedron whose 6 vertices are joined by strings.
Determine the decomposition of the 18-dimensional space of vibrations of irre-
ducibles of the group of rotational symmetries of T (of order 24). Interpret
the result. Which spaces fuse into irreducibles for the group of all (proper or
improper) symmetries of T (of order 48)?

5.5 Notes
Most of the content of this chapter can be found in Serre’s book [9]. The last
section is entirely based on Theo Beelen’s lectures in the course for which these
notes were prepared.
Chapter 6

Some compact Lie groups

and their representations

In the previous two chapters we got acquainted with the representation theory
of finite groups. In this chapter we will treat some of the representation theory
of compact Lie groups, without going into the details of defining a Lie group and
showing, for example, the existence of an invariant measure. A good reference,
on which in fact most of this chapter is based, is [1]. Some general, less detailed
remarks can be found in [9].

6.1 Some examples of Lie groups

One can think of a Lie group G as a smooth surface embedded in some Euclidean
space. Moreover, G carries the structure of a group in such a way that the
multiplication map (g, h) 7→ g ·h, G×G → G and the inverse map g 7→ g −1 , G →
G are smooth. Examples, some of which we already saw, are:

1. GL(n, R),

2. SL(n, R),

3. O(n, R),

4. SO(n, R),

5. U(n) := {g ∈ Mn (C) | g ∗ g = I}, where g ∗ stands for the conjugate

transpose of g, i.e., g ∗ = (g ji )ij if g = (gij )ij .

6. SU(n) := {g ∈ U(n) | det(g) = 1}.

The latter two groups are called the unitary group and the special unitary group.
A more intrinsic definition is the following: let (., .) be a Hermitian inner product
on a finite-dimensional complex vector space V . Then the corresponding unitary

81
82CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS

group is the group of all linear maps on V with (gv, gw) = (v, w) for all g ∈ G
and v, w ∈ V . Note that each of the real (complex) matrix groups above is
the zero set of some smooth map F from Mm (R) (Mm (C)) to some Euclidean
space Rd such that the rank of the Jacobian of F at every point of the zero
set of F is d. If, in this situation, the zero set of F is a group with respect to
matrix multiplication—as it is in the above cases—then it is automatically a
Lie group. Such Lie groups are called linear because they are given by a linear
representation. In the present chapter, this class of Lie groups suffices for our
needs.

Exercise 6.1.1. How can GL(n, R) be constructed from Mn+1 (R) via the above
construction? What about (R, +)?

Exercise 6.1.2. Show that O(n, R) ∼ = SO(n, R) o (Z/2Z). Here o means that
the subgroup at the left is a normal subgroup and the subgroup at the right is
a complementary subgroup in the sense that their product is the whole group
and their intersection is the trivial group. Such a product is called semi-direct.
If n is odd, then it is a direct product.

Exercise 6.1.3. Show that, in SU(2), any element is conjugate to a diagonal

matrix of the form
 it 
e 0
e(t) := for certain t ∈ R.
0 e−it

The theory of Lie groups is a blend of group theory and topology (surfaces,
continuity, differentiability, etc.), as might be expected. For example, one can
ask whether a given Lie group is (path-wise) connected, or whether it is compact.
A typical lemma from the theory is the statement that the connected component
of the identity element is a normal subgroup.

Exercise 6.1.4. Show that O(n, R) is not connected.

Proposition 6.1.5. The Lie group SO(n, R) is connected.

Proof. We will connect an arbitrary matrix g = (aij )ij in SO(n, R) to the iden-
tity via a continuous and piecewise smooth path in SO(n, R). The first step is
to connect it to a diagonal matrix; to this end, suppose first that there exists a
position (i, j) such that i > j and aij 6= 0. Then one can choose this position
such that, in addition, ai0 j 0 = 0 for all (i0 , j 0 ) with i0 > j 0 and either j 0 < j or
j 0 = j and i0 > i. Schematically, g has the following form:
 
T ∗ ∗ ∗ ∗
0
 ajj ∗ aji ∗

0
g= ∗ ∗ ∗ ∗
,
0 aij ∗ aii ∗
0 0 ∗ ∗ ∗
6.1. SOME EXAMPLES OF LIE GROUPS 83

where T is a (j − 1) × (j − 1) upper triangular matrix. Now let r(t), t ∈ R be

the real n × n matrix given by



 cos(t), if i0 = j 0 = j or i0 = j 0 = i,
0 0
 sin(t), if i = i and j = j,



r(t)i0 j 0 = − sin(t), if i0 = j and j 0 = i,
1, if i 6= i0 = j 0 6= j, and






 0 otherwise.

Then r(t) has the following block form:

 
I 0 0 0 0
0
 cos(t) 0 − sin(t) 0
0
r(t) =  0 I 0 0;
0 sin(t) 0 cos(t) 0
0 0 0 0 I

note that r(t) ∈ SO(n, R) for all t ∈ R. Now we have

 
T ∗ ∗ ∗ ∗
0
 ajj cos(t) − aij sin(t) ∗ aji cos(t) − aii sin(t) ∗
0
c(t) := r(t)g =  ∗ ∗ ∗ ∗
0 ajj sin(t) + aij cos(t) ∗ aji sin(t) + aii cos(t) ∗
0 0 ∗ ∗ ∗

Now R(sin(t), cos(t)) runs through all 1-dimensional subspaces of R2 as t runs

from 0 to π, so that there exists a t0 ∈ [0, π) for which (sin(t0 ), cos(t0 )) is
perpendicular to (ajj , aij ) in the standard inner product; we find that c(t0 )ij =
0, in addition to c(t0 )i0 j 0 = 0 for all (i0 , j 0 ) with i0 > j 0 and either j 0 < j or j 0 = j
and i0 > i. Moreover, the path c(t), t ∈ [0, t0 ] lies in SO(n, R) and connects g
to c(t0 ).
Repeating the above construction, we find a path connecting g to an upper
triangular matrix h. Being an element of O(n, R), h is in fact diagonal and
has only entries ±1 on the diagonal. Moreover, as det(h) = 1, the number of
−1s is even, and they can be partitioned into pairs. Each such pair can be
smoothly transformed to 1s by multiplication from the left with an appropriate
r(t), t ∈ [0, π] as above.

This proposition implies that SO(n, R) is the connected component of the

identity in O(n, R). The following exercise will be used later on, for the repre-
sentation theory of SO(3, R).

Exercise 6.1.6. Show that SU(2) is connected. Hint: find a continuous path
from any given element of SU(2) to some e(t) as in Exercise 6.1.3.
84CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS

6.2 Representation theory of compact Lie groups

Compact Lie groups (i.e., with the property that their embedding in Euclidean
space is closed and bounded) are particularly nice, because much of the repre-
sentation theory of finite groups works just as well for these. The main tool is
the so-called Haar measure, which enables us to take ‘the average over a group’
just as we did in finite group theory. Let us state the existence and uniqueness
of this measure in a theorem.
Theorem 6.2.1. Let G be a compact real Lie group, and denote by C(G) the
linearR space of all continuous functions G → R on G. There is a unique map
f 7→ G f (g)dg, C(G) → R which is
1. linear, i.e., for all e, f ∈ C(G) and α ∈ R, we have
Z Z Z Z Z
(e+f )(g)dg = f (g)dg + e(g)dg and (αf )(g)dg = α f (g)dg,
G G G G G

2. monotonous, i.e., if e(g) ≤ f (g) for all g ∈ G, then also

Z Z
e(g)dg ≤ f (g)dg,
G G

3. left-invariant, i.e., for the action of G induced on C(G) by the left multi-
plication, we have for all h ∈ G and f ∈ C(G) that
Z Z
h · f (g)dg = f (g)dg,
G G
−1
where h · f is the function g 7→ f (h g) (h ∈ G).
4. and normalized, i.e., Z
1dg = 1.
G

This linear map is called the invariant (Haar)-integral.

Exercise 6.2.2. A finite group is a compact Lie group. What is the invariant
integral?
In order to find an invariant Haar integral on the compact Lie group T =
{e(t) | t ∈ R}, where e(t) is defined as in Exercise 6.1.3, we search for invariant
measures on e(t), that is, differential forms invariant under left multiplication.
In general, given a representation r : G → GL(V ), we can make invariant
measures on G by taking certain entries of the matrix r(g)−1 dr(g). For the
group T and the natural representation, this is e(t)−1 de(t). The 1, 1 entry of
R 2π
this matrix is e−it deit = idt. Since 0 idt = 2πi, the invariant Haar-integral,
viewed as a map assigning a real number to a function f : R/2πZ → R, is
Z 2π
1
f (t)dt.
2π 0
6.2. REPRESENTATION THEORY OF COMPACT LIE GROUPS 85

A similar but more laborious computation yields that the invariant Haar mea-
sure for SU2 (C) is
1
sin2 θ sin ψdθdψdφ,
2π 2
where
− sin θ sin ψeiφ
 
cos θ − i sin θ cos ψ
sin θ sin ψe−iφ cos θ + i sin θ cos ψ
is a general element of SU2 (C).
The result of the following exercise will be used in proving that certain
irreducible representations of SU2 (C) exhaust all irreducible representations.
Exercise 6.2.3. Show that the map sending a class function f : SU(2) → C
to (f ◦ e) : R → C, where e is the one-parameter subgroup of SU(2) defined
in Exercise 6.1.3, is a linear bijection between the space H of continuous class
functions on SU(2) and the linear space of all even, 2π-periodic continuous
functions R → C. Moreover, the norm corresponding to the invariant inner
product (6.1) below corresponds, under this linear bijection, to the normalized
L2 inner product on the space of continuous even 2π-periodic functions on R.
One could define representations of compact Lie groups in Hilbert spaces,
but here we shall restrict our attention to finite-dimensional representations.
Also, we shall only consider complex representations.
Definition 6.2.4. A finite-dimensional complex representation of the Lie group
G is a smooth homomorphism G → GL(V ), where V is a finite-dimensional
complex vector space.
In fact, one can prove that continuous homomorphisms from a Lie group G
are automatically smooth.
For finite-dimensional complex representations of compact Lie groups, al-
most all theorems that we proved in the preceding chapters for finite groups
hold, be it that at some places a sum must be replaced by an integral. To give
an idea how things work, one can try the following exercise.
Exercise 6.2.5. Let r : G → GL(V ) be a finite-dimensional representation of
the compact Lie group G, and let W ⊂ V be an invariant subspace. Show that
W has an invariant complement in V .
We will study the representations of the groups SU(2) and SO(3, R), so it is
good to verify that these are indeed compact Lie groups.
Exercise 6.2.6. In this exercise we prove that U(n), SU(n), O(n, R) and
SO(n, R) are compact Lie groups. To this end, endow the space Mn (C) with
the Hermitian inner product given by
X
((aij )ij , (bij )ij ) := aij bij .
i,j

1. Show that (a, b) = tr(ab∗ ) for all a, b ∈ Mn (C).

86CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS

2. Show that U(n) is bounded with respect to the norm corresponding to

this inner product.
2
3. Show that U(n) is the zero set of a smooth map F : Mn (C) → Rn , whose
Jacobian has rank n2 at each element of U (n).

Thus, U(n) is a Lie group by the construction of the beginning of this chapter,
and compact as it is a closed and bounded set in Mn (C). Find similar arguments
for the other groups above. Show, on the contrary, that the group O(n, C) of
complex matrices g satisfying g T g = Iis not compact for n ≥ 2.

A final remark concerns the compact analogon of the inner product on the
space of complex-valued functions on a finite group: in the case of a compact
Lie group, one should consider the space of Hilbert space of continuous class
functions with Hermitian inner product defined by
Z
(f |h) := f (g)h(g)dg. (6.1)
G

Here, the irreducible characters form an orthonormal Hilbert basis in the space
of continuous class functions.

6.3 The 2-dimensional unitary group

This section is based on section II.5 of [1].
Let r : SU(2) → GL(V ) be the standard representation of SU(2) on V = C2 ,
the action being given by matrix-column multiplication (the elements of V are
understood to be columns). As we have seen before, this action induces linear
actions on P (V )d , the space of homogeneous polynomials of degree d defined on
V , by letting

(gP )(ξ) = P (g −1 ξ) for ξ ∈ V, P ∈ P (V )d , g ∈ SU(2).

Let rd : SU(2) → GL(P (V )d ) be the corresponding representation and χd its

character.
We are ready to state the main theorem of this section.

Theorem 6.3.1. The representations rd are irreducible for all d = 1, 2, . . ..

Proof of theorem 6.3.1. We will use Exercise 4.1.30. Let A ∈ End(P (V )d ) be a

linear map that commutes with all rd (g) for g ∈ SU(2).
The space P (V )d is spanned by the polynomials Pk : ξ 7→ ξ1k ξ2d−k for k =
0, . . . , d, and is therefore (d + 1)-dimensional.
Consider the diagonal subgroup of SU(2). It consists of all matrices of the
form  
a 0
ga :=
0 a−1
6.3. THE 2-DIMENSIONAL UNITARY GROUP 87

Note that the element a ∈ C must necessarily have norm 1. We compute

(ga Pk )(ξ) = Pk (g a1 ξ)
= (ξ1 /a)k (aξ2 )d−k
= ad−2k Pk (ξ).
Hence we conclude that Pk is an eigenfunction of rd (ga ) with eigenvalue ad−2k .
Choose a0 such that none of these eigenvalues coincide. If A commutes with
rd (ga0 ), then the eigenspaces of the latter map are invariant under A. Hence,
as these eigenspaces are all one-dimensional, there are complex numbers ck for
k = 0, . . . , d such that
APk = ck Pk .
It remains to show that all ck are equal. For this purpose, consider the real
rotations st defined by  
cos t − sin t
st :=
sin t cos t
and compute
(st Pn )(ξ) = Pn (s−t ξ)
= (ξ1 cos t + ξ2 sin t)n
X n 
= cosk t · sinn−k t · Pk (ξ)
k
k

Applying A to the result, we get

X n 
A(st Pn ) = cosk t · sinn−k t · ck Pk .
k
k

First applying A and then rd (st ), and using the fact that the two commute, we
find that also
X n 
A(st Pn ) = cosk t · sinn−k t · cn Pk .
k
k
Taking t such that cos t sin t 6= 0, it follows from the fact that the Pk are linearly
independent that cn = ck for all k = 1, . . . , n, and hence that A = cn I.
We wish to prove that these are all irreducible characters. To this end we
shall use some Fourier theory, and the result of Exercise 6.2.3.
Theorem 6.3.2. The representations rd are, up to equivalence, all irreducible
representations of the group SU(2).
Proof. In view of Exercise 6.2.3, we can identify H with the space of even 2π-
periodic complex-valued functions on R. The matrix of rd (e(t)) with respect to
the basis P0 , . . . , Pd is diagonal, and its trace is
d
X
κd (t) = ei(d−2k)t .
k=0
88CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS

It is readily verified that κ0 (t) = 1 (this corresponds to the trivial representation

of the group), and that κ1 (t) = 2 cos t. Indeed, one has κd+2 (t) = κd (t) +
2 cos((d + 2)t). Hence, the functions κd for d = 0, 1, 2, . . . generate the same
space as the functions cos dt. From Fourier theory we know that this subspace
is dense in the Hilbert space of continuous even 2π-periodic functions. This
implies, using the isomorphism from Exercise 6.2.3, that the characters χd span
a dense subspace in the Hilbert space of continuous class functions on SU(2).
Any irreducible character different from all rd would have inner product 0 with
all rd , and hence be zero. This concludes the proof that the rd exhaust the set
of irreducible characters of SU(2).

6.4 The 3-dimensional orthogonal group

We will obtain the irreducible representations of SO(3, R) from those of SU(2)
by means of the following lemma, which appeared earlier as Exercise 5.3.11.
Lemma 6.4.1. Let G and H be groups, φ : G → H a surjective homomorphism.
Then the (irreducible) representations of H are in bijective correspondence with
the (irreducible) representations r of G with ker r ⊇ ker φ.
Thus, we wish to find a surjective homomorphism of SU(2) onto SO(3) and
to determine its kernel. Recall the homomorphism φ : SL(2, C) → L of Example
2.2.9. Here, L is the group of Lorentz-norm preserving linear maps from M to
itself, where M is the R-linear span of the following four complex 2×2-matrices.
     
0 1 0 −i 1 0
e0 = I, e1 = , e2 = , e3 = .
1 0 i 0 0 −1

It can be checked that in this way

det(x0 e0 + x1 e1 + x2 e2 + x3 e3 ) = x20 − x21 − x22 − x23 .

So L is the group

{B ∈ GL(M ) | det(Bx) = det(x), for all x ∈ M }.

The homomorphism φ was defined by

φ(A)(x) = AxA∗ , for all x ∈ M, A ∈ SL(2, C).

From now on, φ will denote the restriction of this homomorphism to SU(2). We
claim that the image is the group

G := {B ∈ L | Be0 = e0 and det B = 1},

which is, by its action on the invariant 3-space e⊥

0 , easily seen to be isomorphic
to SO(3, R).
First of all φ(SU(2)) ⊆ G. Indeed, if A ∈ SU(2), then A∗ = A−1 . Therefore,
φ(A)(e0 ) = e0 , since eo is just the 2 × 2 identity matrix. Also, the function
6.4. THE 3-DIMENSIONAL ORTHOGONAL GROUP 89

det ◦φ is a continuous function SU(2) → R, which can apparently take only

values in {−1, 1}. As SU(2) is connected, and det(φ(I)) = 1, the determinant
of any element in the image φ(SU(2)) is 1.
Now let us prove that G ⊆ φ(SU(2)). We follow the approach given in [4].
First we prove the following two lemmas.
Lemma 6.4.2. Let F be a subgroup of SO(3, R) with the following two proper-
ties.
1. F acts transitively on the unit sphere.
2. There is an axis such that F contains all rotations around that axis.
Then F = SO(3, R).
Exercise 6.4.3. Prove this lemma.
Lemma 6.4.4. For every non-zero x ∈ e⊥ 0 there is an A ∈ SU(2) such that
φ(A)x = ce3 for some c ∈ R and c > 0. Here e⊥ 0 denotes the orthogonal
complement of e0 with respect to the Lorentz inner product, i.e.,

e⊥
0 = {x1 e1 + x2 e2 + x3 e3 | xi ∈ R} ⊆ M.

Proof. Any such x is in fact a Hermitian matrix with trace 0. As such it has
two real eigenvalues c, −c, where c > 0, (note that x is non-zero). Let two
(necessarily orthogonal) eigencolumns of norm 1 corresponding to c and −c
be the first and second column of a matrix B, respectively. Then the usual
coordinate transformation rule gives

x = B −1 (cH)B.

We have that B ∈ U(2). Now take A = dB where d2 = 1/(det B); then A is

still unitary, since |d| = 1. Furthermore, we have

AxA−1 = (dB)x(dB)−1 = dd−1 BxB −1 = cH,

and det A = d2 det B = 1.

Now let us continue with the proof that F := φ(SU(2)) coincides with G.
The group F acts on the unit sphere S in e⊥ 2 2 2
0 defined by x1 + x2 + x3 , and by
Lemma 6.4.4 this action is transitive, since any vector in the sphere is mapped
to e3 by some group element. Hence, to apply Lemma 6.4.2 to F , we need only
check that F contains all rotations around some axis.
We leave this as an exercise to the reader:
Exercise 6.4.5. Show that φ(e(t)) is the rotation around he3 i through an angle
of 2t. Here e : R → SU(2) is the one-parameter group defined earlier.
We conclude that φ : SU(2) → G is a surjective homomorphism. Its kernel
contains {I, −I}, as is easily checked, and the following exercise shows that this
is the whole kernel.
90CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS

Exercise 6.4.6. Show that ker φ = {I, −I}. Hint: the elements of this kernel
must commute with all of M , hence in particular with e1 , e2 , and e3 .
Finally we can apply lemma 6.4.1: the irreducible representations of SO(3)
correspond to those of SU(2) for which −I is mapped to the identity transfor-
mation. More precisely, we wish to know for which d we have rd (−I) = I. For
this to be the case, it is necessary and sufficient that

((−I)P )(ξ) = P (ξ) for all ξ ∈ V, P ∈ P (V )d .

This holds if and only if d is even. Hence the irreducible representations are

Wd := P (V )2d .

Note that the dimension of Wd is 2d + 1.

In the following example, we will see another interpretation of the irreducible
representations of SO(3, R).
Example 6.4.7. Consider P (V )d the space of homogeneous polynomials of
degree d defined on the vector space V which will be of dimension three here. As
we have seen before, the group GL(3, R) acts on this space P (V )d by (A·p)(x) =
p(A−1 x), for all A ∈ GL(3, R), p ∈ P (V )d and x ∈ R3 . So does the subgroup
SO(3).
Take the subspace U of P (V )2 generated by the polynomial x21 + x22 + x23 .
This is an invariant subspace for the action of SO(3). Hence P (V )2 is not
irreducible. Actually the spaces P (V )d are not irreducible for all d ≥ 2.
To find SO(3)- invariant subspaces of P (V )d , we need to define the Laplace
∂2 ∂2 ∂2
operator ∆ on R3 : ∆ = ∂x 2 + ∂x2 + ∂x2 . The space of harmonic polynomials
1 2 3
hd is the vector space:

hd = {p ∈ P (V )d | ∆p = 0}.

As

was noticed in Example 4.1.17, the dimension of P (V )d is dim P (V )d =
d+2
d .
We now want to prove that the dimension of hd is 2d + 1. A homogeneous
polynomial p of degree d can be written as a sum of homogeneous polynomials
pk in x2 and x3 with power of x1 ’s as coefficients:
x21 xd
p(x1 , x2 , x3 ) = p0 (x2 , x3 ) + x1 p1 (x2 , x3 ) + p2 (x2 , x3 ) + · · · + 1 pd (x2 , x3 )
2 d!
d
X xk 1
= pk (x2 , x3 )
k!
k=0

where the pk ’s have degree d − k. If one applies the Laplace operator to p (you
see why we took factorials):
d−2 k d
xk1
 2
∂2

X x 1
X ∂
∆(p) = pk+2 (x2 , x3 ) + + pk (x2 , x3 )
k! k! ∂x22 ∂x23
k=0 k=0
6.4. THE 3-DIMENSIONAL ORTHOGONAL GROUP 91

Hence an element p ∈ P (V )d is in hd if and only if for all k = 0, . . . , d − 2:

 2
∂2

∂
pk+2 (x2 , x3 ) = − + pk (x2 , x3 )
∂x22 ∂x23

This formula implies that p in hd depends only on the first two homogeneous
polynomials p0 and p1 . Therefore we conclude that the dimension of hd is
precisely the number of homogeneous polynomials in two variables of degree d
and d − 1. That is, dim hd = 2d + 1.
Of course, the degree of hd reminds us of the irreducible representations Wd
of SO(3) we met before. Indeed, we have the following.

Theorem 6.4.8. The space hd of harmonic polynomials of degree d is an irre-

ducible representation of SO(3).

To see that what we just stated is true, we should first check that hd is a
representation of SO(3). We state here the following fact without proof.
Fact: The action of the Laplace operator on P (V )d commutes with the
action of SO(3).
From this we conclude that hd is a representation of SO(3).
To prove that hd is irreducible, we claim that hd = Wd . Set p a harmonic
polynomial in hd , p(x1 , x2 , x3 ) = (x2 + ix3 )d . Let R(t) be the matrix of SO(3)
defined by  
1 0 0
R(t) = 0 cos t − sin t
0 sin t cos t
Apply R(t) to p. That gives us R(t)p = exp(−idt)p. So p generates an
invariant subspace W of hd under the action of R(t).
Any element in SO(3) is conjugate to R(t). So in order to find the character
of Wd on R(t) it suffices to compute it on R(t). That is just the value of
the character χ of rd (the irreducible representation of SU(2)) at et/2 , i.e.,
P2d i(d−k)t 2d
k=0 e .
By the decomposition of hd in irreducible Ws ’s, we know that its character
is a linear combination of exp(ist). So what we saw above about the action of
R(t) on W leads us to conclude that hd = Wd .
92CHAPTER 6. SOME COMPACT LIE GROUPS AND THEIR REPRESENTATIONS
Bibliography

[1] T. Bröcker and T. tom Dieck. Representations of Compact Lie Groups.

Springer Verlag, New York, 1995.

[2] F. Albert Cotton. Chemical applications of group theory. Wiley-

Interscience, Chichester, third edition, 1990.

[3] J.D.Dixon and B. Mortimer Permutation Groups. Springer Verlag, New

York, 1991.

[4] E.Vinberg. Linear Representations of Groups. Birkhäuser, Basel, 1989.

[5] William Fulton. Young tableaux, volume 35 of London Mathematical Society

Student Texts. Cambridge university press, 1997.

[6] Juergen Hinze, editor. The permutation group in physics and chemistry,
volume 12 of Lecture notes in chemistry. Springer Verlag, Berlin, 1979.

[7] H.H. Jaffe and Milton Orchin. Symmetry in chemistry. Wiley, London,
1965.

[8] A. Nussbaum. Applied group theory; for chemists, physicists and engineers.
Prentice-Hall electrical engineering series. Prentice-Hall, Englewood Cliffs,
1971.

[9] J.P. Serre. Linear representations of finite groups. Springer Verlag, New
York, 1977.

[10] S. Sternberg. Group theory and physics. Cambridge University Press, Cam-
bridge, 1994.

[11] Hermann Weyl. Symmetry. Princeton science library. Princeton University

Press, Princeton, 1989.

93