Notes Group Theory in Physics

MA PH 464 - Group Theory in
Physics
Lecture Notes
MA PH 464 - Group Theory in
Physics
Lecture Notes
Vincent Bouchard
University of Alberta
March 30, 2020

v
These are lecture notes for MA PH 464 offered at the University of Al-
berta.
Motivation
Symmetry is the language of modern physics. When confronted with a phys-

ical system, the theoretical physicist’s main tool to formulate a mathematical
description of the system is to study its symmetries. Those correspond to
transformations of the physical system that leaves the physics invariant.
The mathematical language for studying transformations and symmetries
is group theory. This is why group theory is omnipresent in modern theoret-
ical physics. We study what kind of transformations leave a physical system
invariant; those transformations generally form an abstract group (the sym-
metry group of the system), and then we use the language of group theory to
describe the physical system.
Groups are defined abstractly. But what they consist in is a set of sym-
metries (of an object, a theory, etc.) that is closed under composition and
contains inverse operations. Most symmetries of nature are of that sort, which
is why group theory appears universally.
But in fact in physics what we are interested in is how groups act on
something: an object, a theory. This is what representation theory is about: it
represents the elements of a group as acting on something. While in geometry,
one may ask what the symmetries of a particular object are, in representation
theory, the question is turned around: given a group, what kind of objects does
it act on? This is the central theme of representation theory.
In particular, in physics groups of symmetries act on the space of solu-
tions of a physical system. This is given by a representation of the symmetry
group. Understanding the details of this representation tells us a lot about the
properties of the possible solutions (or particles, fields, etc.)
Therefore, by studying group theory and representation theory, we will
learn a lot about physical entities. We will see how the spin of a particle arises
naturally in terms of representations, how particles in the Standard Model
transform according to representations of the gauge group SU (3) × SU (2) ×
U (1), how these representations naturally give rise to the idea of Grand Unified
Theories, etc. Cool stuff!
vi
References
These lecture notes are not meant to be original. They are heavily influenced
by already existing lecture notes and textbooks on group theory in physics.
The main references that I have been using are:
• Anthony Zee, Group Theory in a Nutshell for Physicists, Princeton Uni-

versity Press, 2016.
• Dimitri Vedensky, lecture notes on Group Theory.
• Jock McOrist, lecture notes on Representation Theory.
• Eugene A. Lim, lecture notes on Symmetry in Physics.
vii
Contents
Motivation vi
References vii
1 Basic concepts of group theory 1

1.1 Groups . . . . . . . . . . . . . . . . . . . . . 1
1.2 Subgroups . . . . . . . . . . . . . . . . . . . . 3
1.3 Direct product . . . . . . . . . . . . . . . . . . . 7
1.4 Multiplication table . . . . . . . . . . . . . . . . . 8
1.5 Presentation. . . . . . . . . . . . . . . . . . . . 10
1.6 Cosets . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Conjugacy classes . . . . . . . . . . . . . . . . . . 13
1.8 The symmetric group Sn . . . . . . . . . . . . . . . 15
1.9 Normal subgroups . . . . . . . . . . . . . . . . . 21
1.10 Quotient groups . . . . . . . . . . . . . . . . . . 23
1.11 Homomorphisms and isomorphisms . . . . . . . . . . . 25
1.12 Fun stuff . . . . . . . . . . . . . . . . . . . . . 28
2 Representation theory 29
2.1 Representations . . . . . . . . . . . . . . . . . . 30
2.2 Properties of representations . . . . . . . . . . . . . . 33
2.3 Unitary representations . . . . . . . . . . . . . . . 37
2.4 Semisimplicity . . . . . . . . . . . . . . . . . . . 41
2.5 Schur’s lemmas . . . . . . . . . . . . . . . . . . 43
2.6 The great orthogonality theorem . . . . . . . . . . . . 45
2.7 Characters . . . . . . . . . . . . . . . . . . . . 48
2.8 Orthogonality for characters . . . . . . . . . . . . . . 49
2.9 Reducibility and decomposition. . . . . . . . . . . . . 52
2.10 An example: S3 . . . . . . . . . . . . . . . . . . 55
2.11 An example: the regular representation . . . . . . . . . . 58
2.12 Real, pseudoreal and complex representations . . . . . . . 59
3 Applications 65
3.1 Crystallography . . . . . . . . . . . . . . . . . . 65
3.2 Quantum Mechanics . . . . . . . . . . . . . . . . . 69
3.3 Coupled harmonic oscillators. . . . . . . . . . . . . . 74
viii
CONTENTS ix
3.4 The Lagrangian . . . . . . . . . . . . . . . . . . 78
4 Lie groups and Lie algebras 82

4.1 Lie groups . . . . . . . . . . . . . . . . . . . . 82
4.2 Rotations in two and three dimensions . . . . . . . . . . 88
4.3 Lie algebras . . . . . . . . . . . . . . . . . . . . 93
4.4 SU (2) . . . . . . . . . . . . . . . . . . . . . . 97
4.5 General remarks . . . . . . . . . . . . . . . . . . 101
5 Representation theory of Lie groups 103

5.1 Tensor representations of SO(3) . . . . . . . . . . . . 103
5.2 Representations of Lie algebras and the adjoint representation . 107
5.3 The highest weight construction for su(2) . . . . . . . . . 112
5.4 Projective and spin representations . . . . . . . . . . . 119
5.5 Tensor representations of SU (N ) . . . . . . . . . . . . 123
5.6 The Standard Model of particle physics and GUTs . . . . . . 129
5.7 Representations of the Lorentz group. . . . . . . . . . . 133
5.8 Unitary representations of the Poincare group . . . . . . . 139
5.9 Classification of simple Lie algebras . . . . . . . . . . . 143
Chapter 1
Basic concepts of group the-

ory
1.1 Groups
Objectives
You should be able to:
• Recall the definition of a group.
• Recognize various groups that occur frequently in physics and mathe-
matics.
• Show in particular examples that the group axioms are satisfied.

• Distinguish between discrete, finite and continuous groups.
• Determine whether a group is abelian.
1.1.1 Definition
In physics, we think of symmetries in terms of transformations of a system
(think of rotations), that we can compose with each other. What we want to
do here is capture the abstract properties that are shared by all such symmetry
transformations. This is what the abstract concept of a group does. Basically,
a group is a set with a binary operation (composition) that satisfies a bunch
of axioms.
Definition 1.1.1 Group. A group is a set G with a binary operation ·,
which we call composition or multiplication, that satisfies the following axioms:
• Closure: For all a, b ∈ G, a · b ∈ G.
• Associativity: For all a, b, c ∈ G, (a · b) · c = a · (b · c).
• Identity: There exists a group element e ∈ G, known as the identity, such

that e · a = a · e = a for all a ∈ G.
• Inverses: For all a ∈ G, there exists an element a−1 ∈ G such that
a · a−1 = a−1 · a = e.
1
CHAPTER 1. BASIC CONCEPTS OF GROUP THEORY 2
It is easy to show a few elementary properties of groups:

1. The identity element of a group is always unique.
2. For each element a ∈ G, there is a unique inverse a−1 ∈ G.
3. In a group, a · b = a · c implies b = c, and b · a = c · a implies b = c, as
expected.
Checkpoint 1.1.2 Prove these three properties.
We can further categorize groups according to some of their properties:
Definition 1.1.3 Abelian groups. If a · b = b · a for all a, b ∈ G, i.e.
composition is commutative, then we say that G is abelian. We say that it is
non-abelian otherwise. ♦
Definition 1.1.4 Finite groups. If G has a finite number of elements, we
say that the group is finite. We call |G| the order of G. ♦
Definition 1.1.5 Continuous and discrete groups. If we write G = {gα },
then the index α can be either discrete or continuous. Note that a discrete
group can well be infinite. ♦
1.1.2 Examples
You already know many examples of groups. Let us look at a few of them:
Example 1.1.6 Rotations. Rotations in two-dimensional space form a
group, denoted by SO(2), which is an abelian continuous group. Rotations in
three-dimensional space also form a group, which is denoted by SO(3), which
happens to be non-abelian. Similarly, rotations in n dimensions form the non-
abelian group SO(n).
Example 1.1.7 R and Z under addition. The set R with the operation
of addition forms an abelian continuous group. If we restrict to the subset of
integers Z, we still get an infinite abelian group, but it is now discrete.
Example 1.1.8 The trivial group. The group with only one element {1}
forms a group under multiplication, but it is a rather boring one. It is called
the trivial group.
Example 1.1.9 The cyclic group ZN . The two square roots of 1, {1, −1},
form the group Z2 under multiplication, which is an abelian discrete group of
order 2. More generally, the N ’th roots of unity, {e2πik/N }k=0,...,N −1 form the
group ZN under multiplication, which is an abelian discrete group of order N .

Example 1.1.10 The unitary group U (1). Complex numbers of norm 1,
namely eiθ with θ ∈ [0, 2π), form an abelian continuous group under multipli-
cation called U (1).
Example 1.1.11 Addition of integers mod N . Addition of integers mod
N generates an abelian discrete group of order N . The elements of the set are
{0, 1, . . . , N − 1}, with the operation of addition mod N . This has in fact the
same abstract group structure as ZN defined above.
Example 1.1.12 The symmetric group Sn . The set of invertible map-
pings f : S → S of a finite set S with n elements forms a group, which is
denoted Sn and called symmetric group (or permutation group). The ele-
ments of the group Sn are permutations of n objects. There are n! such distinct
permutations, so Sn has order n!, and is non-abelian for n ≥ 3.
Example 1.1.13 The general linear groups GL(n, R) and GL(n, C). The
set of invertible linear maps f : V → V of a vector space V forms a continuous
group, known as the general linear group of V and denoted by GL(V ). If
V = Rn , we write GL(n, R); if V = Cn , we write GL(n, C). We can also think
of those as the sets of invertible n × n matrices with the operation of matrix
multiplication (with either real or complex entries).
Example 1.1.14 The special linear groups SL(n, R) and SL(n, C). The
set of n × n matrices with unit determinant under matrix multiplication also
forms a continuous group, since det(AB) = det(A) det(B). This is called the
special linear group SL(n, R) for matrices with real entries, and SL(n, C)
for matrices with complex entries.
Example 1.1.15 The Lorentz and Poincare groups. Lorentz trans-
formations in Minkowski space form a continuous group. If we also include
translations, we obtain the Poincare group.
1.2 Subgroups
Objectives
• Determine whether a subset of a group is a subgroup.
• Construct the orthogonal and unitary groups as subgroups of the general
linear groups.
• Determine the centre of a group.
• Determine the cyclic subgroup generated by an element of a group.
1.2.1 Definition
The notion of a subgroup is rather obvious. But these are important in physics,
for instance in the context of symmetry breaking. Suppose that a physical
system is symmetric under a certain group of transformation. Then one could
think of adding a perturbation to the system, for instance an extra term in the
Lagrangian or Hamiltonian, that lowers the symmetry group to a subgroup of
symmetries. This is captured by the concept of a subgroup.
The idea of subgroups is just like subspaces for vector spaces, which consist
of subspaces that are vector spaces in their own right.
Definition 1.2.1 Subgroups. A subset H of a group G is a subgroup if it
is a group in its own right. We write H ⊂ G. ♦
Since H is a group in its own right, it follows that any subgroup of G
contains the identity element e ∈ G. According to the definition, the group
G is a subgroup of itself. Also, for any G, the subset {e} containing only
the identity element is a subgroup, known as the trivial subgroup. Those two
subgroups are rather boring; we call them improper. We are interested in
studying proper subgroups.
1.2.2 Examples
Some examples of subgroups:
Example 1.2.2 Rotations in 3D around a fixed axis. Consider only

rotations in three dimensions around a fixed axis. This gives a subgroup
SO(2) ⊂ SO(3).
Example 1.2.3 Permutations that leave some objects fixed. Consider
only permutations of n objects that leave n − m objects fixed. This gives a
subgroup Sm ⊂ Sn , with m < n.
Example 1.2.4 The alternating group An . We have not defined the parity
of a permutation yet, but if we consider the subgroup of permutations that are
even, we get the so-called alternating subgroup An ⊂ Sn .
Example 1.2.5 Cyclic subgroup in Zn . Consider the cyclic group Zn ,
represented as the n’th roots of unity under multiplication. Let m be an
integer that divides n. Then there are a subset of n’th roots of unity that are
also m’th roots of unity; those form the subgroup Zm ⊂ Zn . For instance, if
we write Z4 = {1, −i, −1, i}, then we have a subgroup Z2 = {1, −1} ⊂ Z4 .
Example 1.2.6 Integers and reals. The real numbers R form a group
under addition. If we restrict to the integers Z, then we get a subgroup Z ⊂ R
under addition.
Example 1.2.7 The subgroup mZ. The set of all multiples of a positive
integer m, denoted by mZ, is a subgroup of Z with respect to addition. In fact,
all subgroups of Z are of this form.
We can in fact construct many subgroups of GL(n, R) and GL(n, C). Recall
that we can think of GL(n, R) and GL(n, C) either as groups of real and
complex invertible matrices (respectively), or as groups of invertible linear
transformations on Rn and Cn . We can construct subgroups by restricting to
matrices of special types, or by looking at linear transformations that preserve
special structures on Rn or Cn .
Example 1.2.8 The special linear groups. Since SL(n, R) and SL(n, C)
consist of n × n matrices with determinant 1, they form subgroups of GL(, nR)
and GL(n, C) respectively.
Example 1.2.9 The orthogonal group O(n). We can start with GL(n, R),
and consider the subset of orthogonal matrices, which we denote by O(n).
Recall that a matrix M ∈ GL(n, R) is orthogonal if M T M = I.
The subset O(n) is a subgroup of GL(n, R), which we now check. First, the
identity I ∈ O(n). Second, if A, B ∈ O(n), then AB ∈ O(n), since (AB)T AB =
B T AT AB = B T B = I. Therefore the subset O(n) is closed under matrix
multiplication. Finally, if A ∈ O(n), then its inverse A−1 is also orthogonal, and
hence in O(n). Indeed, since A−1 = AT , we have (A−1 )T A−1 = (AT )T AT =
AAT = I.
Orthogonal matrices also have a special interpretation in terms of linear
transformations of Rn . We want to look at linear transformations that preserve
special structures on Rn . In this case, we are interested in preserving the
Euclidean scalar product on Rn . Recall that given two column vectors u, v ∈
Rn , the Euclidean scalar product is given by
uT v = u1 v1 + u2 v2 + . . . + un vn .
If we apply a linear transformation M : Rn → Rn , to get new vectors u0 = M u
and v 0 = M v, then
(u0 )T v 0 = (M u)T (M v) = uT M T M v.
Thus (u0 )T v 0 = uT v for all u, v ∈ Rn if and only if M T M = I. In other
words, M is orthogonal, i.e. in O(n). Therefore, O(n) can be thought of as the
subgroup of invertible linear transformations of Rn that preserve the Euclidean

scalar product.
These transformations preserve length and angle. Indeed, the length square
of a vector u is |u|2 = uT u, thus if u0 = M u with M orthogonal, then |u0 |2 =
|u|2 , and the length is preserved. But then, since uT v = |u||v| cos θ with θ the
angle between the vectors u and v, it follows from (u0 )T v 0 = uT v and the fact
that |u0 | = |u| and |v 0 | = |v| that θ0 = θ, i.e. the angle between the vectors is
also preserved.
But what transformations preserve length and angle? Rotations and re-
flections! So O(n) is the group generated by rotations and reflections in n
dimensions.
Example 1.2.10 The special orthogonal group SO(n). Consider n × n
orthogonal matrices M . Then M T M = I, and det(M T M ) = det(M )2 =
det(I) = 1, and hence det(M ) = ±1. It is easy to convince yourself that
rotations correspond to the subgroup of transformations with det(M ) = 1
(which is the component of O(n) that is connected to the identity). We call
this subgroup the special orthogonal group SO(n) ⊂ GL(n, R). This is the
group of rotations in n dimensions.
Checkpoint 1.2.11 Check that SO(n) is a subgroup of O(n).
Example 1.2.12 The unitary group U (n) and special unitary group
SU (n). We can do a similar construction for GL(n, C). Thinking of it as the
group of n × n invertible matrices with complex entries, we can first restrict
to the subset of unitary matrices, namely matrices M ∈ GL(n, C) such that
M † M = I. Here we defined the adjoint (or Hermitian conjugate) of a matrix
M as M † = (M ∗ )T , where M ∗ denotes the matrix M with the entries complex
conjugated. Following the exact same steps as in the orthgonal case for real
matrices, one can show that the subset of unitary matrices is a subgroup, which
we call the unitary group U (n) ⊂ GL(n, C). As for orthogonal matrices,
(det M )2 = 1, and if we restrict to matrices with det M = 1, we get the
special unitary group SU (n) ⊂ GL(n, C).
We can also think of unitary matrices as linear transformations of Cn . What
kind of transformations are those? They are the transformations that preserve
the standard inner product on Cn . Recall that given two vectors u, v ∈ Cn , we
define the inner product as
(u∗ )T v = u∗1 v1 + u∗2 v2 + . . . + u∗n vn .
Thus, just as for orthogonal transformations, it is easy to see that unitary

transformations preserve the inner product on Cn . So the unitary subgroup
U (n) ⊂ GL(n, C) can be understood as the subgroup of invertible linear trans-
formations on Cn that preserve the standard inner product.
Example 1.2.13 The Lorentz group and its generalizations. Recall
that the orthogonal group O(n) can be thought of as the subgroup of invertible
linear transformations of Rn that preserve the Euclidean scalar product. But
instead of looking at the Euclidean notion of distance on Rn , we could have
started with a different scalar product. For instance, we could take R4 to be
Minkowski space, that is, equipped with the Minkowski scalar product:
u · v = −u0 v0 + u1 v1 + u2 v2 + u3 v3 .
Note the different sign for the first term, which corresponds to the time direc-
tion in Minkowski space. We can look at the subset of linear transformations
of R4 that leave this notion of scalar product invariant. It forms a subgroup,
which is denoted by O(1, 3) ⊂ GL(4, R). This is nothing but the Lorentz
group of special relativity! Indeed, one can check that elements of O(1, 3)
are Lorentz transformations. So one can think of Lorentz transformations as
"rotations in Minkowski space".
(More precisely, O(1, 3) is obtained by composing Lorentz transformations
with parity and time reversal transformations, just as O(n) is obtained by
composing reflections with rotations. Lorentz transformations correspond to
the component of O(1, 3) connected to the identity.)
In general, if we start with Rn with a scalar product of the form
u · v = −u1 v1 − . . . − up vp + up+1 vp+1 + . . . + un vn ,
the subset of linear transformations that preserve this inner product is denoted
by O(p, n − p), and is a subgroup of GL(n, R).
Example 1.2.14 The symplectic group Sp(2n, R). Another interesting
subgroup that plays an important role in physics is the symplectic
group. Let

0 I
x, y ∈ R2n , and J be the 2n × 2n symplectic matrix J = , where I
−I 0
is the n × n identity matrix. The subset of GL(2n, R) that leave the “anti-
symmetric bilinear form” xT Jy invariant is a subgroup of GL(2n, R), which is
called the sympletric group and is denoted by Sp(2n, R). It is for instance
fundamental in the more geometrical treatment of Hamiltonian mechanics.
Many of those groups will play a fundamental in physics. In particular, we
will see that SO(3) and SU (2) are intimately related, and so are SO(1, 3) and
SL(2, C).
1.2.3 A few universal subgroups

There are a few particular subgroups that are rather universal. In some way,
what we are doing now parallels what we you did in linear algebra for vec-
tor spaces. We want to define the analog of the span of a subset, and the
commutator of two elements, but in the context of abstract groups.
Definition 1.2.15 Subgroup generated by a subset of a group. Let S
be a subset of finite group G. The subgroup generated by S, denoted by
hSi, is the union of S and all inverses and products of the elements in S
In particular, if S consists of a single element S = {a} with a ∈ G, then we
write hai for the subgroup generated by a, which consists in the collection of
integer powers of a. Note that because G is finite, we know that at some point
ak = e for some integer k. hai is called the cyclic subgroup generated by
a. ♦
In fact, based on this, we can define the notion of cyclic groups.
Definition 1.2.16 Cyclic group. A cyclic group is a group generated
by a single element. In other words, it is equal to one of its cyclic subgroups
G = hgi for some element g ∈ G, called a generator of G. ♦
Given a particular element of a group, we can also look at all elements that
commute with it.
Definition 1.2.17 Centralizer and the centre of a group. Let a ∈ G.
The set of elements of G that commuate with a, denoted by CG (a), is called
the centralizer of a in G. The set of elements of G that commute with all
elements of G is called the centre of G, and denoted by Z(G). ♦
It is easy to to show that both CG (a) and Z(G) are subgroups of G. More-
over, Z(G) is abelian by definition, and, clearly, G is abelian if and only if
Z(G) = G, since all its elements commute with each other.
Remark 1.2.18 The definitions that we are making look very similar to the
analogous definitions for vector places, but with 0 replaced by the identity
element e. Indeed, the definitions that we are proposing reduce to the previous
propositions for vector spaces when the group operation is considered to be
vector space addition. In this case, the identity element for addition in a
vector space is the zero vector, so e would be replaced by 0, as needed.
We can also define the “commutator” of two elements of a group G. But we
need to keep in mind that we only have one operation to work with, namely
group multiplication. The commutator should somehow measure how non-
abelian a group G is.
Definition 1.2.19 The commutator subgroup of a group. Let a, b ∈ G.
The commutator [a, b] of a and b is defined by
[a, b] = a−1 · b−1 · a · b.
If a and b commute, then [a, b] = e, thus we can think of the commutator as

measuring how far from being abelian a group is. The subgroup generated by
all commutators of elements G is called the commutator subgroup of G. ♦
If G is abelian, then [a, b] = e for all a, b ∈ G. The implication goes in the
other direction as well. Thus a group G is abelian if and only if its commutator
subgroup is the trivial subgroup {e} ⊂ G.
1.3 Direct product

Objectives
• Calculate the direct product of two groups.
• Determine when a group is the direct product of two of its subgroups.
We can also take products of groups to construct bigger groups. The natural
operation is called direct product. This is analogous to the notion of direct
sums of vector subspaces, where vectors are written as sums of vectors of
different subspaces, such that the only common vector in the distinct subspaces
is the zero vector.
Here we will first take the point of view where we study when a group is a
direct product of two of its subgroups. We will then see how we can use the
Cartesian product of two sets to naturally construct direct products of groups.
Definition 1.3.1 The direct product. A group G is said to be the direct
product of its subgroups H1 and H2 , which is denoted by G = H1 × H2 , if
the following conditions are satisfied:
1. All elements of H1 commute with all elements of H2 ;
2. The group identity is the only common element to H1 and H2 ;
3. Every g ∈ G can be written as g = h1 h2 for some h! ∈ H1 and h2 ∈ H2 .

♦
We can use this definition to construct bigger groups by taking direct prod-
ucts of groups:
Proposition 1.3.2 Let G and H be groups. The Cartesian product G × H can
be given a group structure by defining the operation
(g, h) ? (g 0 , h0 ) = (gg 0 , hh0 )
for all g, g 0 ∈ G and h, h0 ∈ H. The resulting group is the direct product of the
subgroups (G, e) and (e, H), regarded as subgroups of G × H.
Checkpoint 1.3.3 Prove this proposition.
Example 1.3.4 The Klein four-group, aka Z2 × Z2 . An example of a
direct product is the group Z2 × Z2 (also called the “Klein four-group”). Using
the construction above, one can think of this group as consisting of the four
elements (1, 1), (1, −1), (−1, 1), (−1, −1), with the operation being component-
wise multiplication. Note that this group is different from the cyclic group of
order 4 Z4 ; for instance, in Z4 = {1, −i, −1, i} only two elements (1 and −1)
square to the identity, while in Z2 × Z2 all four elements square to the identity.

This notion of direct product is what is needed to make sense, for instance,
of the gauge group of the Standard Model in particle physics, which is given
by SU (3) × SU (2) × U (1).
1.4 Multiplication table

Objectives
• Construct the multiplication table of a finite group.
1.4.1 The idea

So far we studied various examples of groups by giving explicit realizations of
them. But in the end groups can be understood entirely abstractly, without
referring to a particular realization of it. In this section we will focus on finite
groups.
To specify a finite group all that we need to know is the number of elements
in the set, and the result of multiplying these elements together. This is neatly
encoded in the form of a multiplication table. Suppose that G is a finite
group of order n. Then its multiplication table will be a square n×n array with
rows and columns labelled by elements of the group, and entries corresponding
to the products of these elements:
Table 1.4.1 An example of a multiplication table
··· gj ···
.. ..
. .
gi gi gj
.. ..
. .
1.4.2 The rearrangement theorem

Before we construct multiplication tables, an interesting fact to note is that
each element of a group can only appear once and only once in each row, and
same for each column. This is sometimes formally stated as the “rearrangement
theorem”:
Theorem 1.4.2 Rearrangement theorem. Let G = {g1 , . . . , gn } be a finite

group, and pick an element gk ∈ G. Then
gk G = {gk g1 , . . . , gk gn }
contains each element of the group once and only once.

Proof. Suppose that two elements of gk G are equal: gk gi = gk gj for i 6= j.
Then multiply by gk−1 on the left to get gi = gj . This is a contradiction, since
i 6= j. Thus all elements of gk G must be distinct. Since |gk G| = n = |G|, it
follows that all elements of G must appear once and only once in gk G. Thus
the sets G and gk G are identical up to reordering of elements, hence the name
of the theorem.
1.4.3 Groups of order 2 and 3

With this under our belt let us start by constructing the multiplication table
for a group of order two. Let us write G = {e, a}, where e is the identity
element. We know that e2 = e, ea = a and ae = a. Since each element can
appear once and only once in each row and column of the multiplication table,
it follows that we must have a2 = e, to get:
Table 1.4.3 The multiplication table for the group of order 2
e a
e e a
a a e
Let us now move to groups with three elements G = {e, a, b}. We know by
definition of the identity that
Table 1.4.4 Constructing the multiplication table for the group of
order 3
e a b
e e a b
a a
b b
Further, the product ab cannot be equal to a (resp. b), since otherwise we
could multiply by a−1 (resp. b−1 ) on the left to conclude that b = e (resp.
a = e). Thus we must have ab = e. We then complete the table using the
requirement that each element appears once and only once in each row and
column (this is like sudoku! :-) ), to conclude that
Table 1.4.5 Multiplication table for the group of order 3
e a b
e e a b
a a b e
b b e a
Again, this is the unique possibility. Hence there is a unique abstract group
with three elements. Further, it is abelian, since the table is symmetric about
its diagonal.
Going to higher order however we lose uniqueness. For groups with four
elements, there are two distinct multiplication tables (corresponding to the
groups Z4 and Z2 × Z2 , see Example 1.3.4). And for higher order groups there
are many possibilities. I will not construct multiplication tables for groups of
higher order, but if you like this kind of thing feel free to do it yourself as an
exercise. Try for instance the symmetric group S3 , which has order 3! = 6.
Lots of fun!
1.5 Presentation
Objectives
• Describe a group in terms of its presentation.
• Recall the general properties of the modular group.
If you have constructed the multiplication table for S3 as an exercise, you
have probably come to the realization that writing down multiplication tables
becomes quite annoying rather quickly for higher order groups. There has to be
a better way of encoding the abstract group structure of finite groups without
having to write down the multiplication table each time. One such way is to
write down the “presentation” of a group.
The idea is to specify a set of generators S of the group, which is a subset
of elements of the group from which all other elements can be obtained by group
multiplication, and the essential relations R that the generators satisfy. We
write the presentation as hS|Ri. Without getting into too much detail, let
me simply give a few examples of presentations.
Example 1.5.1 Z2 , Zn and Z2 × Z2 . As a basic example, consider the group
Z2 of two elements. Its presentation would be given as:
ha|a2 = ei.
Similarly, the cyclic group with n elements Zn would be written as:
ha|an = ei.
The direct product Z2 × Z2 would be written as:
ha, b|a2 = b2 = e, ab = bai.

Example 1.5.2 The modular group. As a more interesting example, let
us look at the so-called modular group, which is important in various areas
theory and conformal field theory. Consider 2 × 2
of physics, suchas string
a b
matrices M = with integer entries and such that det M = 1. This
c d
group is denoted by SL(2, Z), the special linear group of 2 × 2 matrices with
integer entries. If we also impose the identification M = −M in SL(2, Z) we
end up with the group P SL(2, Z), the projective special linear group of 2 × 2
matrices with integer entries, which is also known as the modular group. This
group is important because it is the group of linear fractional transformations
of the upper half of the complex plane, where it acts as:
az + b
z 7→ , (1.5.1)
cz + d
for z ∈ C with =(z) > 0.
This group looks rather complicated, but in fact one can show that all such
2×2 matrices can be obtained by repeatedly multiplying together two matrices:

0 1 1 1
S= , T = .
−1 0 0 1
In terms of fractional linear transformations, what that means is that the

general transformation (1.5.1) can be obtained by repeatedly composing the
two transformations:
1
S : z 7→ − , T : z 7→ z + 1.
z
Moreover, it is clear that S 2 = I, where I is the identity matrix (the identity
element in the group), and one can check easily that (ST )3 = I as well. Thus
we can write the presentation of P SL(2, Z) as:
hS, T |S 2 = I, (ST )3 = Ii.
1.6 Cosets
Objectives
• Determine the cosets of a subgroup of a finite group.
• State and prove Lagrange’s theorem.
• Use Lagrange’s theorem to rule out when a subset cannot be a subgroup

of a group.
1.6.1 Definition
Now that we understand subgroups, we can talk about cosets. In fact we have
already seen the idea of a coset when we stated the rearrangement theorem, but
now instead of taking the whole group G, we will consider subgroups H ⊂ G.
Definition 1.6.1 Cosets. Let H = {e, h1 , . . . , hr } be a subgroup of G, and
pick an element g ∈ G. Then the set:
Hg = {eg, h1 g, . . . , hr g}
is a right coset of H. Similarly, the set
gH = {ge, gh1 , . . . , ghr }
is a left coset of H. Cosets are generally not subgroups of G; they will be

only if g ∈ H ⊂ G. ♦
Example 1.6.2 The cosets of Z2 ⊂ Z4 . Consider the group Z4 rep-
resented as G = {1, −i, −1, i} under multiplication. Consider the subgroup
H = {1, −1}. Then the left cosets are:
1 · H = {1, −1}, i · H = {i, −i},
−1 · H = {−1, 1}, −i · H = {−i, i}.
So there are really two distinct cosets, given by H itself and the set {i, −i}.
Since the group is abelian, the left cosets are the same as the right cosets. In
general however left cosets and right cosets are not the same. We also notice
that all elements of G appear exactly once in the distinct cosets; as we will see,
this is a general result.
Cosets will turn out to be very useful in physics. Often we need to deal with
the fact that many different mathematical quantities give the same physical
observable. For instance, the electromagnetic potential is not uniquely fixed;
add to it the gradient of any function, and you get the same electric and
magnetic fields. Similarly, in quantum mechanics, any two states in a Hilbert
space that only differ by a phase give the same physical system. The way
to deal with this kind of redundancy mathematically is to define equivalence
classes of objects that give the same physics. It turns out that cosets are
perfectly suited for that; under suitable conditions, we can see the cosets as
being equivalence classes, hence from a physics point of view what we are often
really interested in are the cosets of a given group, rather than the group itself.
Thus what we would like to see is whether we can upgrade the cosets to
groups. It is clear from the definition that cosets of a subgroup H ⊂ G are
generally not subgroups of G themselves. However, what we can do is look at
the “set of cosets” all together; these in fact form a group, which will be called
the quotient group. This will be the group of equivalence classes formed by
the cosets, which is often the relevant group of objects that we are interested
in physically. But we are jumping ahead of ourselves; we will come back to
quotient groups shortly.
1.6.2 Cosets and Lagrange’s theorem

Going back to cosets, the cool thing about them is that, given a subgroup,
there is some sort of decomposition of a group into cosets of the subgroup,
as you can see in the example above. This is because of the following simple
Lemma:
Lemma 1.6.3 Two cosets of a subgroups are either identical as sets or have
no common elements.
This lemma will be important to make sense of cosets as equivalence classes,
since two elements cannot be in two distinct but not identical equivalence
classes.
Proof. Let H = {e, h1 , . . . , hr } be a subgroup of G. Clearly, two cosets either
have no common elements or have at least one common element. Let us show
that if they have a common element, then they are identical.
Suppose that the left cosets generated by gi and gj , i 6= j, have a common
element, say gi hm = gj hn for some m and n. Then gj−1 gi = hn h−1 m , therefore
gj−1 gi ∈ H. It thus follows that the left coset gj−1 gi H is in fact equal to
H itself as a set, since multiplying by an element of H on the left is only
rearranging terms (this is the statement of the Rearrangement Theorem 1.4.2
for the subgroup H). Therefore gj−1 gi H = H as sets, and multiplying by gj on
the left, we get gi H = gj H, that is, the two cosets are identical as sets.
We conclude that left cosets are either identical as sets or contain not
common elements. The same proof goes through for right cosets.
This lemma allows us to prove an important theorem about subgroups of
finite groups, known as Lagrange’s theorem:
Theorem 1.6.4 Lagrange’s theorem. The order of a subgroup H ⊂ G of
a finite group G divides the order of G, that is, |H| divides |G|.
Proof. First, since the identity element e is in the subgroup H, and to construct
the cosets of H we multiply by all elements of G, we see that all elements of
G must appear in at least one coset of H. But since two cosets are either
identical as sets or have no common elements, it follows that all elements of G
must appear in exactly one distinct coset.
Furthermore, one can prove that all cosets have the same number of ele-
ments, which is equal to the order of the subgroup |H|. Consider a coset gH for
some g ∈ G. Suppose that two elements of the coset are identical, ghi = ghj
for some hi 6= hj . Multiplying by g −1 on the left, we get hi = hj , which is a
contradiction. Therefore all elements of gH must be distinct, and hence the
number of elements in gH is the same as the number of elements in H.
Putting this together, we conclude that the distinct cosets are partitioning
the group G into non-intersecting bins of size equal to |H|. It follows that the
number of distinct cosets times |H| must be equal to |G|, that is, |H| divides
|G|.
Definition 1.6.5 The index of a subgroup. The number of distinct cosets
of a subgroup H ⊂ G is called the index of the subgroup. ♦
Lagrange’s theorem is in fact quite powerful to rule out when a subset
cannot be a subgroup of a finite group. For instance, Lagrange’s theorem
implies the following corollaries:
Corollary 1.6.6 More than half implies abelian. If more than half of the
elements of a finite group commute with each other, then the group is abelian.
Proof. We know that the centre of a group is a subgroup. Thus, by Lagrange’s
theorem, we know that if the order of the centre of the group is larger than
half of the order of the group, then the centre must be the whole group, i.e.
the group is abelian.
Corollary 1.6.7 Groups of prime order. All finite groups whose order
are prime numbers are cyclic, and have no proper subgroups.
Proof. By Lagrange’s theorem, if a group G has order given by a prime number,
then its only subgroups must have either order one (the trivial subgroup) or the
same order as the group itself (the group itself). Thus it cannot have proper
subgroups. Moreover, if you pick any element g ∈ G, then it generates a cyclic
subgroup hgi ⊆ G. But since G has no proper subgroups, then hgi is either
the trivial subgroup if g is the identity element, or it is the whole group G for
any other element g ∈ G. That is, G is a cyclic group, as it is equal to its
non-trivial cyclic subgroups.
1.7 Conjugacy classes

Objectives
• Determine the conjugacy classes of a finite group.
1.7.1 Definition
We have already discussed briefly the idea of equivalence classes. Given a group
G, the idea is to define an equivalence relation, which we generally denote by
≡, between elements of the group, which is reflexive (a ≡ a), symmetric (if
a ≡ b then b ≡ a) and transitive (if a ≡ b and b ≡ c then a ≡ c). Once we have
an equivalence relation, we can define equivalence classes, which are subsets of
elements of the group that are equivalent according to ≡.
There is a particularly interesting equivalence relation, known as conjuga-

tion, that partitions any finite group into equivalence classes, known as conju-
gacy classes.
Definition 1.7.1 Conjugacy classes. We say that two elements a, b ∈ G
are conjugate if there exists an element g ∈ G such that a = gbg −1 . This is
an example of an equivalence relation. Then we can form equivalence classes,
that is, classes of elements of G that are conjugate to each other. The set of
all elements conjugate to one another is called a conjugacy class. ♦
Conjugacy classes are composed of elements that somehow behave in a sim-
ilar way. What “similar” means here depends on the context. As we will see,
for the symmetric group Sn , conjugacy classes will be comprised of permuta-
tions that have the same cycle structure. In the group of three-dimensional
rotations SO(3), we can characterize rotations by specifying an axis of rotation
and an angle of rotation. Then one can show that all rotations that have the
same angle belong to the same conjugacy class.
As we will see, conjugacy classes are very useful in the study of groups,
especially in representation theory.
1.7.2 Properties of conjugacy classes

It is clear that conjugacy classes are disjoint, since an element cannot be in two
different conjugacy classes. So we can partition a group G into its conjugacy
classes.
It is also interesting to note that all elements in a given conjugacy class
have the same order.
Definition 1.7.2 Order of an element of a group. We define the order
of an element a ∈ G to be the smallest integer n such that an = e. ♦
With this definition we can prove the following lemma:
Lemma 1.7.3 All elements of a given conjugacy class have the same order.
Proof. Suppose that a ∈ G has order n. If b is conjugate to a, that is b = gag −1
for some g ∈ G, then
bn = (gag −1 )n = (gag −1 )(gag −1 ) · · · (gag −1 ) = gan g −1 = geg −1 = e.
It follows that bn = e, and hence the order of b is ≤ n.

To finish the proof, we need to argue that there cannot be an integer m < n
such that bm = e, so that the order of b is precisely n. Suppose that there is
such a m < n. Then
bm = gam g −1 = e.
Multiplying by g −1 on the left and g on the right, we get that am = e. But n
must be the smallest integer such that an = e, therefore this is a contradiction.
It follows that the order of b is equal to n.
We can also prove the following interesting result:
Lemma 1.7.4 Each element in the centre of a group G constitutes a conjugacy
class by itself. And hence the identity is always a conjugacy class by itself, and
for an Abelian group G, each element forms its own distinct class.
Proof. Let h ∈ Z(G) be an element of the centre of G. Then for all g ∈ G, we
have hg = gh, and hence for all g ∈ G, ghg −1 = h, therefore the only element
conjugate to h is h itself.
The two corollaries follow directly, since the identity is always an element
of the centre of a group, and if G is abelian, then Z(G) = G.
1.8 The symmetric group Sn

Objectives
• Sketch the proof of Cayley’s theorem.
• Use standard notations (permutations, cycles) to describe elements of the
symmetry group Sn .
• Calculate the parity of a permutation.
• Describe conjugacy classes in Sn in terms of cycle structures and integer
partitions.
Many of the concepts that we just studied abstractly become alive when
we study the symmetric group Sn , which is the group of permutations of n
elements. Everything becomes clear and explicit, since we kind of all known
intuitively how permutations work.
1.8.1 Cayley’s theorem

But in fact studying the symmetric group Sn is fundamental, because of what
is known as Cayley’s theorem, which basically states that any finite group G
of order n is isomorphic (that is, identical as abstract group) to a subgroup of
the symmetric group Sn . More precisely, the statement of Cayley’s theorem is
that G is a subgroup of the group of permutations of the elements of G, which
is Sn since G has order n. So from this point of view, studying symmetric
groups and their subgroups basically means studying all finite groups.
Cayley’s theorem is quite deep, since it says that we can realize any abstract
finite group as a subgroup of a permutation group. In other words, it connects
abstract group theory to the very concrete world of permutations. Isn’t it cool?
As we will see, Cayley’s theorem gives an example of a nice representation
which exists for all finite groups, and is called the regular representation.
Why is Cayley’s theorem true? Without going through a formal proof,
let us see how it goes. Consider the multiplication table of a finite group G =
{g1 , . . . , gn }. The i’th row of the table is given by the elements {gi g1 , . . . , gi gn }.
By the Rearrangement Theorem Theorem 1.4.2, this is the same set of elements
as G, but in a different order. Thus we can assign to the i’th row an element
of the permutation group of n elements, that is an element of Sn , which acts
as {g1 , . . . , gn } 7→ {gi g1 , . . . , gi gn }. As a result, to every element gi ∈ G we
can assign a permutation si ∈ Sn ; we have constructed a map from G to a
subgroup of Sn , and one can show that this map is a group isomorphism. This
is the essence of Cayley’s theorem.
Example 1.8.1 Z3 as a subgroup of S3 . As an example, consider the
group of three elements G = Z3 with multiplication table Table 1.4.5. We see
that the map assigns to the element e ∈ G the identity permutation e ∈ S3 ;
to a ∈ G it assigns the cyclic permutation 1 7→ 2, 2 7→ 3, 3 7→ 1 in S3 ; and to
b ∈ G it assigns the cyclic permutation 1 7→ 3, 2 7→ 1, 3 7→ 2. Thus we have
identified G = Z3 with the alternating subgroup A3 ⊂ S3 consisting of even
permutations.
1.8.2 Notation
Now let us study the symmetric group Sn in more detail. First we introduce
standard notation. An element of Sn corresponds to a permutation of n ele-
ments. It is customary to denote it by its action on the n elements {1, . . . , n}.
For instance, the cyclic rotation of the n elements, which takes 1 to 2, 2 to 3,
etc., would be denoted by

1 2 ··· n − 1 n
π= .
2 3 ··· n 1
Generically, a permutation π that takes 1 to π(1), 2 to π(2), and so on, would

be denoted by

1 2 ··· n−1 n
π= .
π(1) π(2) · · · π(n − 1) π(n)
Example 1.8.2 The symmetric group S3 . As an example, consider S3 .

It has 3! = 6 elements; the identity permutation (π1 ), three permutations
corresponding to a single exchange of two elements (π2 , π3 , π4 ), and two per-
mutations that permute the three elements (π5 , π6 ). In the notation introduced
above, the six elements are

1 2 3 1 2 3 1 2 3
e = π1 = , π2 = , π3 = ,
1 2 3 2 1 3 3 2 1

1 2 3 1 2 3 1 2 3
π4 = , π5 = , π6 = .
1 3 2 3 1 2 2 3 1
To construct the multiplication table for S3 , we need to multiply elements in

S3 , where by multiplication here we mean composition of permutations. For
instance, the composition π4 ◦ π5 can be constructed as follows. Start with π5 ;
it takes 1 7→ 3, 2 7→ 1, 3 7→ 2. Then, combining with the action of π4 , we get
1 7→ 3 7→ 2, 2 7→ 1 7→ 1, and 3 7→ 2 7→ 3, which is π2 . Therefore, π4 ◦ π5 = π2 .
In fact, we see explicitly that S3 is non-Abelian. Consider π5 ◦ π4 . We
obtain 1 7→ 1 7→ 3, 2 7→ 3 7→ 2 and 3 7→ 2 7→ 1, which is π3 . Therefore
π5 ◦ π4 = π3 , hence π4 ◦ π5 6= π5 ◦ π4 .
1.8.3 Cycles
Since we are dealing with permutations of set of n elements, it is clear that
after applying a given permutation a certain number of times, we will come
back to the initial elements. In other words, consider for instance the element
1. A given permutation π will take 1 → π(1). Applying it again, we will get
1 → π(1) → π(π(1)). After a number of application, say r, we will necessarily
get back π r (1) = 1. The numbers 1, π(1), π 2 (1), . . . , π r−1 (1), that are reached
from 1 by π form what we call a “cycle”. More precisely:
Definition 1.8.3 Cycle. Let π ∈ Sn , i ∈ {1, . . . , n} and let r be the
smallest positive integer such that π r (i) = i. Then the set of r distinct elements
{π k (i)}r−1k=0 is called a cycle of π of length r, or an r-cycle generated by i.
We denote a given cycle by (i π(i) . . . π r−1 (i)).
It is clear that a given permutation π breaks up the set of n elements
{1, . . . , n} into disjoint cycles. We can then denote the permutation π by the
cycle decomposition of {1, . . . , n} that it implies. ♦
Example 1.8.4 The cycle structure of permutations in S3 . Consider

the permutation π5 for S3 as in the previous example. Start with 1. Then
π5 (1) = 3, π52 (1) = π5 (3) = 2, and π53 (1) = π5 (2) = 1. Thus π5 decomposes
{1, 2, 3} into a single cycle of length 3. In cycle notation, we write π5 = (132).
We could have written π5 = (321) or π5 = (213) as well; those are the same
permutations. In this notation the numbers defining the cycles can be moved
around cyclically without changing anything. Similarly, π6 , which is the other
permutation that exchanges the three elements, has cycle decomposition π6 =
(123).
Consider instead the permutation π2 . Start with 1. Then π2 (1) = 2, and
π22 (1) = π2 (2) = 1. Thus 1 generates a 2-cycle given by (12). There is another
cycle however. Start with 3. Then π2 (3) = 3, hence 3 generates a 1-cycle given
by (3). Therefore, the set of three elements is broken up into two cycles by
π2 , that is π2 = (12)(3). Similarly, we get π3 = (13)(2), π4 = (1)(23), and
π1 = (1)(2)(3).
With the notion of cycles we can define permutations that are “cyclic”:
Definition 1.8.5 Cyclic permutations. If π ∈ Sn has a cycle of length
r and all other cycles of π have only one element, then π is called a cyclic
permutation of length r. ♦
Example 1.8.6 Cyclic permutations in S3 . In the example of S3 , we then
have that π5 and π6 are cyclic permutations of length 3, while π2 , π3 , π4 are
cyclic permutations of length 2. Thus in S3 all permutations are cyclic. But
for instance in S4 , we have the permutation π = (12)(34), which is not cyclic.

A particularly useful type of cyclic permutation is of length two:
Definition 1.8.7 Transposition. A cyclic permutation of length 2 (i.e. a
permutation that only exchanges two elements) is called a transposition. ♦
Example 1.8.8 Transpositions in S3 . To come back to S3 , there are three
transpositions, namely π2 , π3 and π4 .
Now that we understand cycles, we can easily compute composition of
permutations. Given two permutations π1 and π2 , the composition π1 ◦ π2 can
be computed easily by “multiplying cycles”.
This is easier understood in terms of an example. Consider the product
of cycles (15)(234)(253) in S5 . To find the corresponding permutation, we
consider each element of {1, . . . , 5}, and, starting from the far right of the
product of cycles, we keep track of what the element becomes under the action
of the cycles. In our case, consider first 1. The first two cycles on the right
leave it invariant, while the last cycle sends its to 5. Thus 1 → 5. Consider
2. The first cycle sends it to 5, the second cycle then leaves 5 invariant, and
then the last cycle sends it to 1. Thus 2 → 1. Consider 3. Then we get
3 → 2 → 3 → 3. Consider 4. We get 4 → 4 → 2 → 2. Finally, for 5 we get
5 → 3 → 4 → 4. The corresponding permutation is then:

1 2 3 4 5
π= .
5 1 3 2 4
(A consistency check here is that each number 1 to 5 only appear once in the
final result of the permutation, which must be the case.) Equivalently, we
can write the resulting permutation in terms of its cycle structure, which is
π = (1542)(3). Hence we get that
(15)(234)(253) = (1542)(3).
This is how we multiply cycles.
Example 1.8.9 Multiplication of cycles vs composition of permu-

tations. Let us just show in an example that multiplication of cycles and
composition of permutations are equivalent. Consider π1 = (123)(45) and
π2 = (1)(2)(345) in S5 . Suppose that we want to compute the composition
π1 ◦ π2 . The resulting permutation is easily calculated to be

1 2 3 4 5
π1 ◦ π2 = .
2 3 5 4 1
We claim that this is equivalent to the product of cycles (123)(45)(1)(2)(345).

When we consider products of cycles, we can remove cycles of length 1, since
they do not do anything in the product. So we want to compute (123)(45)(345).
Under this product, we have 1 → 1 → 1 → 2, 2 → 2 → 2 → 3, 3 → 4 → 5 → 5,
4 → 5 → 4 → 4, 5 → 3 → 3 → 1, which is indeed the permutation above.
In fact, a few minutes of thought will make it clear that the way we defined
products of cycles is precisely equivalent to composition of the corresponding
permutations.
A very interesting result for the symmetric group is that any permutation
can in fact be written as the product of a number of transpositions. This is
just saying that every permutation can be done in steps, where in each step we
only exchange two objects. Let’s see how it goes. We start by studying how a
cycle can be written as a product of transpositions:
Proposition 1.8.10 Decomposition of cycles into products of trans-
positions. An r-cycle (i1 i2 . . . ir ) can be decomposed into the product of r − 1
transpositions:
(i1 i2 . . . ir ) = (i1 ir )(i1 ir−1 ) · · · (i1 i3 )(i1 i2 ).
Note however that the decomposition is not unique; in particular, since the
square of any transposition is the identity, we could insert the square of any
transposition in the product on the RHS without changing the end result.
Proof. The proof simply involves multiplying the transpositions on the RHS.
Start with i1 . Under the product of transpositions, we get i1 → i2 → . . . → i2 .
Consider i2 : we get i2 → i1 → i3 → . . . → i3 . Iterating, we get that ik → ik+1
for all k = 1, . . . , r − 1, and ir → i1 . Thus the resulting permutation is the
cyclic permutation of length r that has cycle structure (i1 i2 . . . ir ).
Definition 1.8.11 Parity. We define the parity of the cycle as being even
if the number of transpositions in the decomposition is even, and odd if the
number if transpositions is odd. Note that even though the decomposition is
not unique, the parity of the decomposition is uniquely defined. ♦
We can then apply these results to permutations themselves, not just to
individual cycles. We obtain:
Proposition 1.8.12 Decomposition of permutations into products of
transpositions. Any permutation can be decomposed as a product of transpo-
sitions (and length 1 cycles). The parity (even or odd, according to whether the
decomposition has an even or odd number of transpositions) of the permutation
is unique.
In fact, the parity of a permutation can be determined directly from its
cycle structure, without having to work out its decomposition into transposi-
tions, since each r-cycle can always be decomposed into the product of r − 1
transpositions.
Example 1.8.13 Decomposition of a permutation in S6 . As an exam-

ple, consider the permutation π = (145)(23)(6) in S6 . Looking at the cycle
structure, the 3-cycle is even (can be decomposed into two transpositions), the
2-cycle is of course odd (it is a transposition itself), and the 1-cycle does not
contribute to the parity. Thus π is odd, since 2+1 = 3. In fact, a decomposition
is easy to find following Proposition 1.8.10. We get:
π = (15)(14)(23)(6).
Of course, this decomposition is not unique (for instance (15)(14)(24)(24)(23)(6)

would also work), but its parity is uniquely defined.
Definition 1.8.14 The alternating group. The set of even permutations
of n elements is denoted by An and is called the alternating group. ♦
Clearly, An is a subgroup of Sn , since the product of two even permutations
is even, it contains the identity, and the inverse of an even permutation is also
even.
1.8.4 Conjugacy classes

Now let us study conjugacy classes of Sn . Recall that conjugacy classes are
defined as being subsets of elements that are conjugate to each other, where
two elements x, y ∈ G are conjugate if y = gxg −1 for some g ∈ G.
The main result for Sn is the following theorem:
Theorem 1.8.15 Conjugacy classes and cycle structure. Two permu-
tations in Sn are conjugate if and only if they have the same cycle structure,
meaning that they have the same number of cycles of equal length.
To clarify the notation, if we go back to S3 , then the three permuations
(12)(3), (13)(2) and (1)(23) have the same cycle structure, hence are conjugate.
The two permutations (123) and (132) are then also conjugate, and the identity
permutation (1)(2)(3) is of course only conjugate to itself. Thus there are three
conjugacy classes in S3 .
Proof. We will only show that conjugate permutations have the same cycle
structure, and leave the other direction as an exercise. Consider two permu-
tations π, σ ∈ Sn . We want to check whether σ ◦ π ◦ σ −1 has the same cycle
structure as π. If π and σ are given by the permutations:

1 2 ... n 1 2 ... n
π= , σ= ,
π(1) π(2) . . . π(n) σ(1) σ(2) . . . σ(n)
then it follows that

σ(1) σ(2) ... σ(n)
σ ◦ π ◦ σ −1 = ,
σ(π(1)) σ(π(2)) . . . σ(π(n)),
since σ −1 takes the element σ(i) and sends it to i. Thus we can see the
permutation σ ◦π ◦σ −1 as applying σ to the symbols in the cycle decomposition
of π. The result will generally be different from π, but its cycle structure will
necessarily be the same.
The argument may become clearer with a simple example. Consider π =
(123)(4) ∈ S4 , and σ = (12)(34) ∈ S4 . The corresponding permutations are

1 2 3 4 1 2 3 4
π= , σ= .
2 3 1 4 2 1 4 3
The conjugate permutation is then

−1 2 1 4 3
σ◦π◦σ = ,
1 4 2 3
which has cycle structure σ ◦ π ◦ σ −1 = (214)(3). It is indeed the same cycle

structure as π. In fact, it could have been obtained directly by simply applying
σ to the symbols in the cycle decomposition π = (123)(4), which gives directly
σ ◦ π ◦ σ −1 = (214)(3).
It follows from Theorem 1.8.15 that all one has to do to find the conjugacy
classes in Sn is to construct the possible cycle structures. This in turn is
equivalent to partitioning the set {1, . . . , n} into disjoint subsets of various
lengths. What we now show is that this problem can be reformulated into the
problem of finding integer partitions of the positive integer n.
Consider a permutation π ∈ Sn and its cycle decomposition. Let νk be the
number of k-cycles in its decomposition. We introduce a new notation which
gives the cycle structure of the permutation (i.e. the number of cycles of each
length): we write (1ν1 , 2ν2 , . . . , nνn ) to denote that the cycle decomposition of
π has ν1 cycles of length 1, ν2 cycles of length 2, and so on.
The total number of elements Pn in the cycle decomposition must of course be
n. This means that the sum k=1 kνk = n. We can rewrite this sum as:
(ν1 + ν2 + . . . + νn ) + (ν2 + ν3 + . . . + νn ) + (ν3 + . . . + νn ) + . . . + (νn ) = n.

Pn
We then introduce the notation λj := k=j νk , so that the sum above becomes
λ1 + λ2 + λ3 + . . . + λn = n.
It is also clear that λ1 ≥ λ2 ≥ . . . ≥ λn ≥ 0, since λ1 − λ2 = ν1 , λ2 − λ3 = ν2 ,

and so on, and the νi are by definition non-negative integers.
The result of this combinatorial analysis is that the cycle structure of per-
mutations in Sn are in one-to-one correspondence with decreasing
Pn sequences
of non-negative integers λ1 ≥ λ2 ≥ . . . ≥ λn such that k=1 λk = n. Such
a decreasing sequence of non-negative integers is called a partition of n (we
usually omit the λk ’s that are zero).
To be precise, we only showed how the λk ’s can be constructed out of the
cycle structure, that is out of the νi ’s. But it works the other way around
as well. Given a partition of n, that is a sequence of λk , then we define
νk = λk − λk+1 , with νn = λn . νk then gives us the number of cycles of length
k in the cycle structure associated to the partition.
Example 1.8.16 Conjugacy classes in S4 and partitions. This con-
struction will become clearer with an example. Let us construct the conjugacy
classes in S4 , corresponding to the different cycle structures in S4 . This corre-
sponds to different partitions of 4. The five distinct partitions of 4 are easily
found to be 4, 3 + 1, 2 + 2, 2 + 1 + 1 and 1 + 1 + 1 + 1. What this tells us
directly is that there are five distinct conjugacy classes in S4 .
For fun, let us find out what kind of cycle structures these partitions corre-
spond to. Consider first the partition 4, which means λ1 = 4, λ2 = λ3 = λ4 = 0.
To get the cycle structure, we compute ν1 = λ1 − λ2 = 4, ν2 = ν3 = ν4 = 0.
Thus the corresponding cycle structure consists in 4 cycles of length 1. There
is in fact a unique permutation of S4 with this cycle structure, which is the
identity permutation π = (1)(2)(3)(4).
Consider the partition 3 + 1. We get ν1 = 2, ν2 = 1, ν3 = ν4 = 0. So the
cycle structure consists in 2 cycles of length 1, and one cycle of length 2. The
corresponding permutations are transpositions, and there are 6 of them in S4 .
Consider the partition 2 + 2. We get ν1 = 0, ν2 = 2, ν3 = ν4 = 0. The

cycle structure is two cycles of length 2. There are 3 such permutations in S4 .
Consider 2 + 1 + 1. We get ν1 = 1, ν2 = 0, ν3 = 1 and ν4 = 0. The cycle
structure is one cycle of length 3 and one cycle of length 1, which corresponds
to cyclic permutations of length 3. There are 8 such permutations in S4 .
Finally, 1 + 1 + 1 + 1, which gives ν1 = ν2 = ν3 = 0, ν4 = 1. The cycle
structure is a single cycle of length 4, corresponding to a cyclic permutation of
length 4. There are 6 such permutations in S4 .
The result is that there are five conjugacy classes in S4 , corresponding to five
possible cycle structures, corresponding to the five partitions of 4. Counting the
number of elements in each conjugacy classes, we end up with 1+6+3+8+6 =
24 = 4!, which is indeed the order of S4 .
For those of you who like combinatorics, it is also fun to count in general
the number of permutations of Sn of a given cycle structure. We will leave
that as an exercise, but P the result is the following. Consider a cycle structure
(1ν1 , 2ν2 , . . . , nνn ) with k kνk = n. The number of permutation in Sn with
this cycle structure is precisely:
n!
Qn νj
.
j=1 j νj !
Let us check that this formula is consistent with what we wrote above in S4 .
Consider for instance the cycle structure given by one cycle of length 3 and
one cycle of length 1. The number of permutations with that cycle structure
4!
is 3·1!·1·1! = 8, as written above. Similarly, the number of permutations with
4!
one cycle of length 2 and two cycles of length 1 is 2·1!·1 2 ·2! = 6, and so on and
so forth.
Remark 1.8.17 Note that partitions of an integer n can be encoded diagram-
matically into Young diagrams. A Young diagram is a finite collection of
boxes arranged in left-justified rows, with the rows “weakly decreasing” (i.e.,
such that each row has the same or shorter length than the row right above
it). It is then clear that Young diagrams with n boxes contain precisely the
same information as integer partitions of n, with the integers in the partition
given by the length of the rows of the Young diagram. Young diagrams are
useful for studying representations of the symmetric group, in which case they
are upgraded to Young tableaux.
1.9 Normal subgroups

Objectives
• Prove whether a given subgroup is normal or not.
• Recall the definition of simple groups.
Let us go back to the idea of cosets. We know that the cosets are not
subgroups of G, except for the one that includes the identity element. What
we want to do now is define a group structure on the “set of cosets”. That is,
the elements of the new group will be the cosets themselves.
To get a group structure, given two cosets, we want to be able to “multiply
them”.
Definition 1.9.1 Product of subsets. Given any two subsets S1 and S2 of
G, we define the product of S1 and S2 as being the subset
S1 S2 = {s1 s2 | s1 ∈ S1 and s2 ∈ S2 }.
That is, we multiply each element of S1 with each element of S2 . ♦

Now we want to give the set of cosets a group structure under this operation
of multiplication of cosets. It turns out that this is only possible if the left cosets
of H are equal to its right cosets. Or, equivalently, if all elements conjugate to
elements of H are also in H. So we give such subgroups a special name:
Definition 1.9.2 Normal subgroups. A subgroup H of G is called normal
if Hg = gH for all g ∈ G. Equivalently, it is normal if gHg −1 = H (as sets)
for all g ∈ G. Such subgroups are also called self-conjugate, since given any
element h ∈ H, all its conjugate elements are also in H. ♦
In fact, a subgroup is normal if and only if for all h ∈ H, all the elements
conjugate to h are also in H. That is, gHg −1 ⊆ H for all g ∈ G. This may
seem like a slighly weaker condition than gHg −1 = H for all g ∈ G, but in fact
the latter is implied by the former for any group, as one can prove. Therefore,
to prove that a subgroup is normal, it is sufficient to show that all elements
conjugate to elements of H are also in H, which is often taken as the definition
of a normal subgroup.
Checkpoint 1.9.3 Prove that gHg −1 ⊆ H for all g ∈ G implies that gHg −1 =
H for all g ∈ G.
Lemma 1.9.4 A normal subgroup is the union of conjugacy classes.
A subgroup H ⊆ G is normal if and only if it is the union of conjugacy classes.
Proof. This is also rather clear. Suppose H is normal. Pick a h ∈ H. Then,
for any g ∈ G, ghg −1 ∈ H, and hence the conjugacy class of h is included in
H. Since this must true for all h ∈ H, we conclude that H is the union of the
conjugacy classes of elements in H.
Conversely, if H is the union of conjugacy classes, then we know that the
elements conjugate to any element in H are also in H, and hence H is normal.

This gives another way of proving that a subgroup is normal: it will be if
and only if it "does not break conjugacy classes", i.e. is the union of conjugacy
classes.
Before we study examples of normal subgroups, let us introduce one more
piece of jargon:
Definition 1.9.5 Simple groups. A group that does not have proper normal
subgroups is called simple. ♦
Simple, isn’t it? All right, let us now look at some examples!
Example 1.9.6 The alternating group is a normal subgroup of the
symmetric group. Let us show explicitly that the alternating group A3 is a
normal subgroup of S3 . Recall from Definition 1.8.14 that A3 consists of the
even permutations π1 = (1)(2)(3), π5 = (132) and π6 = (123). To show that it
is normal, one could show that all left and right cosets agree. But let us avoid
such lengthy calculations and simply argue that the elements of gHg −1 must
be in H for all g ∈ G. So what we need to show is that all permutations in
gHg −1 are even, in which case they are in H = A3 . Assume that h ∈ H is
even. If g is even, then ghg −1 is certainly even. If g is odd, then g −1 is also
odd, and hence ghg −1 is even. Therefore H is a normal subgroup. In fact, this
argument works for all alternating subgroups An ⊂ Sn , which are all normal.

Example 1.9.7 Subgroups of abelian groups are always normal. All

subgroups of abelian groups are necessarily normal, since ghg −1 = gg −1 h = h
for all g ∈ G and h ∈ H. Equivalently, each element in an Abelian group
constitutes a conjugacy class by itself, and hence all subgroups are unions of
conjugacy classes. Thus all subgroups are normal. As an example, the group
H = {1, −1} under multiplication is a normal subgroup of G = {1, i, −1, −i}.

Example 1.9.8 The centre of a group is normal. The centre Z(G)
of a group G is always normal. Indeed, for all g ∈ G and h ∈ Z(G), we
have ghg −1 = gg −1 h = h, since h must commute with all elements of G.
Equivalently, each element in the centre of a group constitutes a conjugacy
class by itself, and hence the centre of a group is always a union of conjugacy
classes. Therefore it is normal.
Example 1.9.9 Normal subgroups of a direct product. Given a direct
product G1 × G2 , the subgroups (e, G2 ) and (G1 , e) are normal. Indeed, it is
easy to check that these subgroups are self-conjugate.
Example 1.9.10 A normal subgroup of the Rubik’s cube group. An-
other interesting example of a normal subgroup is the subgroup C0 of the 3 × 3
Rubik’s cube group consisting of all moves that leave the position of every block
fixed but can change the orientation of the blocks. Why is it normal? Well, if
you first do an arbitrary move, than do a move in C0 which only changes the
orientiation of the blocks, and then undo the first move, you will end up with
all the block in their original position, but with their orientation potentially
changed. In other words, you will end up with an element of C0 . Therefore we
conclude that C0 is a self-conjugate subgroup.
1.10 Quotient groups

Objectives
• Given a normal subgroup H of a group G, construct the quotient group
G/H.
By definition, for normal subgroups left and right cosets agree, so we can
talk about cosets without specifying left or right. Given a normal subgroup
H ⊂ G, we can then define a group structure on the set of cosets. This is
called the quotient group:
Theorem 1.10.1 Quotient groups. If H is a normal subgroup of G, then
the collection of all distinct cosets of H, denoted by G/H, is a group (with
the operation being product of subsets defined in Definition 1.9.1), called the
quotient group of G by H. The order of the quotient group is the number
of distinct cosets, which is given by |G/H| = |G|/|H|, also called the index of
the subgroup H ⊂ G.
Proof. Consider the collection of all distinct cosets of H in G, which we denote
by G/H. We want to give it a group structure. We define group multiplication
as being the multiplication of sets defined in Definition 1.9.1. Let us first
show that the multiplication of two cosets yield another coset. Consider the
left cosets aH and bH for a, b ∈ G. Elements of (aH)(bH) have the form
ah1 bh2 = a(h1 b)h2 for h1 , h2 ∈ H. Since H is normal, we know that there
exists a h3 ∈ N such that h1 b = bh3 . Thus we can write ah1 bh2 = (ab)h3 h2 ,
which is an element of the coset (ab)H. Hence multiplication of sets defines a
group operation (aH)(bH) = abH.

Then we can prove that this operation defines a group structure. The
identity element is the coset eH = H. For any g ∈ G, the inverse of gH is the
coset g −1 H, since (gH)(g −1 H) = gg −1 H = H. Associativity is also clear:
((aH)(bH))(cH) = (abH)(cH) = abcH = (aH)(bcH) = (aH)((bH)(cH)).

Remark 1.10.2 The reason for this construction to be called a quotient group
comes from division of integers. Suppose you consider 12/4 = 3. You can
understand this calculation of the quotient as taking 12 objects, and dividing
them into disjoint classes of 4 objects; the result is 3 disjoint classes. Here
we are doing the same thing, but instead of only talking about the number of
things in a set, we have a group structure, so we start with a group, partition
the set into subsets, and equip the resulting partition with the structure of a
group.
Example 1.10.3 Trivial examples. Note that the quotient group G/G is
clearly isomorphic to the trivial group (since it has only one element, which is
the identity); and G/hei is isomorphic to G, since the cosets of the subgroup
hei are just the elements of G themselves.
Example 1.10.4 The parity quotient group for permutations. Con-
sider the normal subgroup A3 ⊂ S3 . Using the notation of Definition 1.8.14,
there are two cosets, consisting of the even permutations E = {π1 , π5 , π6 } and
the odd permutations O = {π2 , π3 , π4 }, so the order of the quotient group is 2.
We know that there is only one abstract group of order 2, with multiplication
table given in Table 1.4.3, and using the group operation on cosets we can
check that indeed, E · E = O · O = E, E · O = O · E = O, since composing
two even permutations or two odd permutations gives and even permutation,
while composing and odd and an even permutation gives an odd permutation.

Example 1.10.5 Finite cyclic groups as quotients. A fundamental
example of quotient groups consists in the construction of finite cyclic groups
as quotients. Start with the integers Z under addition. Let H = mZ be the
subgroup of multiples of the positive integer m. Since Z is abelian, mZ is
necessarily normal. We then construct the quotient group Z/mZ. What do
elements of Z/mZ look like? Consider any integer k ∈ Z. Recalling that the
group operation here is addition, the coset kZ generated by the integer k is
k + mZ, that is k plus mutiples of m. Since any two k1 and k2 that differ
by a mutiple of m generate the same coset, we see that there are exactly m
disjoint cosets k + mZ, generated by k ∈ {0, 1, . . . , m − 1}. It follows that the
quotient Z/mZ is a finite group of order m. Product of cosets (in the sense of
Definition 1.9.1)
(k1 + mZ) · (k2 + Z) = k1 + k2 + mZ = k + mZ,
where k = k1 + k2 (mod m).

The group Z/mZ is often denoted by Zm , and we can write its elements
simply as {0, 1, . . . , m − 1}, each of which representing the corresponding coset,
with the understanding that the group operation on Zm is addition modulo m.
This group if of course finite of order m and cyclic; it is generated by 1 (the
other elements are just the “powers of 1”, with power here meaning repeated
application of the operation of addition, i.e. 2 = 1 + 1, 3 = 1 + 1 + 1, etc.). In
fact, it can be shown that every cyclic group of order m is isomorphic to Zm ,
so it is of crucial importance. This is also why we can use the same notation
Zm as we used for the group of roots of unity; those are the same abstract
groups.
Example 1.10.6 SO(2) as a quotient group. Another interesting example
is constructed similarly. Start with the real numbers R, and consider the
normal subgroup consisting of integers Z. The cosets have the form a + Z for
a ∈ R. It is clear that two a1 and a2 that are related by the addition of an
integer generate the same cosets, hence the cosets are indexed by a ∈ [0, 1).
The quotient group R/Z is the group of these cosets, with operation given by
adding the cosets a1 + a2 + Z, with a1 + a2 being understood as addition such
that if the result is greater than one, than we substract one so that the result
is always between 0 and 1.
Let us now argue that R/Z is isomorphic to the group of complex numbers
of absolute value 1 under multiplication, also called U (1). Consider the map-
ping f : R/Z → U (1) given by f (a+Z) = e2πia . Under this mapping, the group
operation on cosets (addition of numbers) is mapped to multiplication of com-
plex numbers. We have not defined group homomorphisms and isomorphisms
yet (we will revisit this example in Example 1.11.5), but this mapping is an
isomorphism, as it is a bijection, and it preserves the group structure. Thus we
can think of R/Z and U (1) as being the same abstract group. Further, from
the point of view of complex numbers of absolute value 1, we can understand
2πa as an angle, and we see that U (1) is isomorphic to the group of rotations
in the complex plane (SO(2)). The group operation on U (1), multiplication
of exponentials, is then mapped to composition of rotations in SO(2). All in
all, we obtain that the quotient group R/Z is the same abstract group as the
group of two-dimensional rotations SO(2)!
1.11 Homomorphisms and isomorphisms

Objectives
• Show when a given mapping between groups is a group homomorphism,
isomorphism and automorphism.
• Find the kernel of a group homomorphism.

• Use the first isomorphism theorem to show that a given subgroup is
normal.
Let us define what we mean more precisely when we say that two groups are
“the same” as abstract groups. We want to define the notion of isomorphisms
for groups. But to start we define homomorphisms. In a few words, those
are maps between groups that preserve the group structure. Isomorphisms are
then defined as homomorphisms that are bijective.
1.11.1 Definitions and examples

Definition 1.11.1 Group homomorphism. Let (G, ?) and (H, ·) be groups,
with corresponding group operations denoted by ∗ and · respectively. A group
homomorphism f : G → H is a map that preserves the group structure, that
is
f (a ? b) = f (a) · f (b) for all a, b ∈ G.
♦
In other words, a group homomorphism is a map between sets that is

consistent with their group structures.
Example 1.11.2 Trivial homomorphism. The most simple example of a
homomorphism is called the trivial homomorphism. Take any group (G, ?),
and let H be the group consisting of a single element {e}. Then the trivial
homomorphism is given by the mapping f (a) = e for all a ∈ G, which trivially
preserves the group structure. For any G with order > 1, it is of course not
an isomorphism, since the mapping is not one-to-one (all elements of G are
mapped into e ∈ H).
While a group homomorphism is a mapping between groups that is consis-
tent with their group structures, in general it may “lose information”, if many
group elements are mapped to the same group element. This is clear by look-
ing at the trivial group homomorphism, which maps all groups to the trivial
group with a single element. So we want to define a group isomorphism as an
homomorphism that “preserves all information about the group.”
Definition 1.11.3 Group isomorphism and automorphism. An iso-
morphism is a group homomorphism that is bijective. We say that two groups
are isomorphic, denoted by G ' H, if there is an isomorphism f : G → H.
An isomorphism of a group into itself is called an automorphism. ♦
Example 1.11.4 The exponential map. An interesting example of an
isomorphism is given by the exponential map, exp : (R, +) → (R>0 , ·). The no-
tation (R, +) denotes the group of real numbers under addition, while (R>0 , ·)
denotes the group of positive real numbers under multiplication. The mapping
exp takes any real number and outputs its exponential, which is a positive real
number. This mapping preserves the group structure (the operation in the first
group is addition, while in the second group it is multiplication). What this
means is that the exponential of the sum of two real numbers should be equal
to the product of the exponential of these real numbers. Indeed:
exp(a + b) = exp(a) · exp(b) for all a, b ∈ R.
Thus exp is a group homomorphism. Moreover, from standard properties of

the real exponential it is easy to see that the homomorphism is bijective, thus
it is an isomorphism.
Example 1.11.5 R/Z and U (1). A similar example was studied in the previ-
ous section (see Example 1.10.6). We considered the quotient group R/Z, and
define a mapping f : R/Z → U (1) by f (a + Z) = e2πia . As in the previous ex-
ample, it follows from properties of the exponential that this mapping preserves
the group structure (being coset multiplication in R/Z and multiplication of
exponentials in U (1)), and hence it is a group homomorphism. Moreover, one
can see that it is bijective, therefore it is a group isomorphism. This is what we
meant when we said that the two groups were “the same as abstract groups”.

1.11.2 First isomorphism theorem

Given an homomorphism f : G → H, we can define its kernel as being the
subset of elements of G that are mapped to the identity element in H. It is
straightforward to show that it is a subgroup of G:
Definition 1.11.6 The kernel of an homomorphism. Let G and H be
groups, and f : G → H a homomorphism. The kernel of f is
ker(f ) = {g ∈ G|f (g) = e ∈ H}.
One can show that ker(f ) is a subgroup of G. ♦

Checkpoint 1.11.7 Prove that ker(f ) is a subgroup of G.
In fact, one of the important theorems in group theory is known as the first
isomorphism theorem (also sometimes called the fundamental theorem of
homomorphisms). It is quite fundamental, as it relates the structure of the
kernel and image of any group homomorphism. The statement of the theorem
goes as follows:
Theorem 1.11.8 First isomorphism theorem. Let f : G → H be a group
homomorphism. Then ker(f ) is a normal subgroup of G, and the image f (G)
is isomorphic to the quotient group G/ ker(f ).
Proof. We know that ker(f ) is a subgroup of G. Let us show that it is normal.
Let x ∈ ker(f ), and pick any g ∈ G. We want to show that gxg −1 ∈ ker(f ).
Since f : G → H is a group homomorphism, we have:
f (gxg −1 ) = f (g)f (x)f (g −1 ) = f (g)eH f (g −1 ) = f (g)f (g −1 ) = f (gg −1 ) = f (eG ) = eH ,
where eH (resp. eG ) denotes the identity element in H (resp. in G). Thus we

conclude that gxg −1 is in the kernel of f , and hence ker(f ) is normal.
The isomorphism φ between G/ ker(f )and f (G) is simply given by sending
the coset g ker(f ) generated by g ∈ G to the image f (g) ∈ f (G) of g under the
group homomorphism. φ is of course surjective, and it is a good exercise to
show that it is also one-to-one, hence and isomorphism.
This theorem is quite deep. In particular, it implies the following result:
Corollary 1.11.9 Normal subgroups are kernels of group homomor-
phisms. Any normal subgroup N ⊂ G can be realized as the kernel of a group
homomorphism.
Proof. Given any group G and normal subgroup N ⊂ G, there is a natural
group homomorphism π : G → G/N , which sends each element of G to the
coset to which it belongs. This homomorphism is surjective, but not one-to-
one (unless N is trivial). The kernel of π consists of all elements of G that are
mapped to the identity coset of N in G, which is just a copy of N itself. Thus
one can identify ker(π) = N .
The first isomorphism theorem (and the direct corollary above, which is
also referred to as first isomorphism theorem) is useful for many reasons. In
particular, it can be used to show that a given subgroup is normal, by realizing
it as the kernel of a group homomorphism. It can also be used to construct
quotient groups as images of group homomorphisms.
Example 1.11.10 The determinant homomorphism. Consider the gen-
eral linear group GL(n, C) and define the mapping det : GL(n, C) → C∗ by
taking the determinant of the matrices (zero is not in the image since GL(n, C)
contains invertible matrices). The image C∗ here is the group of non-zero com-
plex numbers under multiplication. First, let us argue that det is a group
homomorphism. That is, we must show that it respects the group structure
(in the first group the operation is matrix multiplication, in the second group
it is product of complex numbers). For any two matrices A, B ∈ GL(n, C),
det(AB) = det(A) det(B), therefore det is a group homomorphism.
Since the identity element in the multiplicative group C∗ is 1, the kernel
of det is given by the subgroup of GL(n, C) consisting of matrices with unit
determinant, that is SL(n, C). The first isomorphism theorem then implies
that SL(n, C) is a normal subgroup of GL(n, C).
We have seen another example of group homomorphism when we discussed
parity of permutations in Sn and the definition of the alternating group An .
In this case the first isomorphism theorem can be used to show directly that
An is always a normal subgroup of Sn .
Example 1.11.11 The parity homomorphism for Sn . Consider the
group H = {1, −1} under multiplication. Consider the mapping : Sn → H
that assigns 1 to even permutations and −1 to odd permutations. Let us first
show that is a group homomorphism.
For any permutations π1 , π2 ∈ Sn , we need to show that (π1 ◦ π2 ) =
(π1 )(π2 ). What this means is that we must show that the composition of two
even or odd permutations is even, while the composition of an even and an odd
permutation is odd. But this is clear by definition of parity. If r1 (resp. r2 ) is
the number of transpositions in the decomposition of π1 (resp. π2 ), then the
number of transpositions in the decomposition of π1 ◦ π2 is r1 + r2 , and thus if
r1 and r2 are both even or odd, then r1 + r2 is even, while if r1 is even and r2
odd, or vice-versa, then r1 + r2 is odd.
Then is a group homomorphism, and its kernel is the alternating group
An consisting of even permutations. It thus follows from the first isomorphism
theorem that An is a normal subgroup of Sn , and that the quotient group
Sn /An is isomorphic to H.
1.12 Fun stuff

Let me end this section on finite groups with three interesting problems to
think about.
1. Can you prove that the binomial coefficient

m+n (m + n)!
=
m m!n!
is always an integer for any positive integers m, n ∈ Z using group theory?

2. The set of sensible orientations of a rectangular mattress on a bed forms
a group. What group is it? What strategy can you take to rotate peri-
odically your mattress between all its orientations?
3. For more fun stuff have a look at the Futurama theorem... It’s really fun!
See for instance https://en.wikipedia.org/wiki/The_Prisoner_of_Benda
and https://theinfosphere.org/Futurama_theorem, as well as https://
arxiv.org/abs/1608.04809 for an interesting generalization.
Chapter 2
Representation theory
What is representation theory? So far we have studied groups as abstract ob-

jects, defined by a set and some operation on it satisfying a bunch of axioms.
Abstraction is beautiful, but sometimes it is useful to make things more con-
crete. From a mathematical point of view, this is what representation theory
does for you.
The idea is simple. In fact we have already encountered it in the previous
section. A representation of a group is basically a concrete realization of a
group as acting on something. In this course, we will mostly deal with matrix
representations; that is, we will realize the elements of an abstract group as ma-
trices. More formally, we will define a representation as a group homomorphism
from an abstract group to a subgroup of GL(V ) for some finite-dimensional
(complex) vector space V . After choosing a basis on V , we can represent the
group elements by n × n matrices. We will call the dimension n of the vector
space the dimension of the representation.
In other words, what we are doing is shifting viewpoint. One can think
of group theory as studying the group of symmetries of a given object, or
theory, or what not. Instead, we now start with an abstract group, and ask
the question: on what kind of objects can this group act on? Representation
theory is concerned with studying all possible ways that a given group can act.
But representation theory is also more than just the study of a question of
mathematical interest. In fact, in many ways it is the essence of group theory.
Groups usually arise because they act on something. They are symmetries of
some object, or some physical theories. Thus the elements of the group are
really realized concretely, in a given context, as things that operate on some
vector space, and after choosing a basis, they can be represented as matrices.
So in many contexts, especially in physics, we encounter groups in terms of
their representations.
In fact, many physicists think of some groups as their representations.
When physicists think of the group SO(3) of rotations in three dimensions,
they will often think of the corresponding 3 × 3 rotation matrices. But SO(3)
itself, as a group, is an abstract entity; a mathematician would think of ro-
tations as abstract elements of SO(3), with the abstract group operation of
SO(3). The 3 × 3 matrices form in fact a three-dimensional representation of
SO(3) (but there are in fact many more representations of SO(3)!)
In any case, in physics groups often arise as symmetry groups of a particular
system. So one has a particular physical system, and a group of symmetry acts
on it to leave the physical observables invariant. However, the mathematical
objects describing the physical theory (or the space of solutions) may not be
invariant; for instance, while the observables in quantum field theory may be
29
CHAPTER 2. REPRESENTATION THEORY 30
invariant, the fields themselves will not be. Thus they must transform in a
certain way. This is where representation theory comes in; they will transform
in some representation of the group of symmetry. This is why group theory
and representation theory is so fundamental in modern physics!
Let me give a slightly more concrete example of this to end this introduc-
tion. In quantum mechanics, we may consider a Hamiltonian that is invariant
under a group of transformations. A very important question then is to find
how the solutions of the Schrodinger equation corresponding to this particular
Hamiltonian behave under these symmetry transformations. This is answered
by representation theory; we can think of the group of symmetries as linear
transformations on the vector space of solutions, which is precisely the no-
tion of a representation of the group. This can then be used to classify the
eigenfunctions of the Hamiltonian according to how they transform under the
symmetry group. Very powerful! So let us study representation theory!
2.1 Representations
Objectives
• Recall the definition of a representation.
• Construct explicitly a few representations of elementary finite groups.

• Recognize when a representation is faithful.
2.1.1 Definition
Let us start by defining a group representation, and then study a bunch of
examples. We will focus here on representations on finite-dimensional complex
vector spaces (but we could easily upgrade vector spaces to Hilbert spaces, and
we could also consider vector spaces over other fields than C). After choosing
a basis, we can think of those as matrix representations.
Definition 2.1.1 Representation of a group. Let G be a group. A
representation of G is a group homomorphism T : G → GL(V ) (see Defini-
tion 1.11.3 for the definition of a group homomorphism), where V is a finite-
dimensional complex vector space of dimension n. We call n the dimension
of the representation.
After choosing a basis on V , we can think of the representation as being
given by n × n complex invertible matrices T (g) for each group element g ∈ G.
♦
In other words, a representation is a mapping that takes group elements into
n×n complex invertible matrices, and this mapping is such that it preserves the
group structure. Just as when we discussed homomorphisms vs isomorphisms,
we want to distinguish between representations that keep the whole group
structure intact, and those that lose some information:
Definition 2.1.2 Faithful representations. If T is one-to-one (that is,
if it is a group isomorphism on its image), then we say that it is a faithful
representation. We say that it is unfaithful otherwise. ♦
Now we can ask many questions about groups. Do every group have a repre-
sentation (aside from the trivial representation)? If so, how many does it have?
What are their dimensions? How do we characterize representations, and how
do we distinguish between them? This is the essence of representation theory.

In other words, we want to characterize and classify all possible ways that a
given group can act. That is, we want to understand its representations.
2.1.2 Examples
But before we study these questions, let us look at a few examples of repre-
sentations.
Example 2.1.3 Identity representation. We start with the most boring
example. For any group G, consider the mapping G → GL(V ), with V = C,
given by sending g 7→ 1 for all g ∈ G. This is certainly a group homomorphism,
but it is a rather boring one. It exists for any group G. It is called the
identity representation (or trivial representation). For any non-trivial G, it
is a dimension one, unfaithful, representation.
Example 2.1.4 The permutation representation of the symmetric
group Sn . Let us look at a more interesting example. Consider the sym-
metric group Sn , which can be understood as the group of permutations of
n objects. Let us focus on S3 forsimplicity.
 Wecan think of the three ob-
1 0 0
jects as vectors v1 = 0, v2 = 1 and v3 = 0. Then a permutation
0 0 1
can be represented as a 3 × 3 matrix acting on the vector space spanned by
these vectors. That is, we can define a representation T : S3 → GL(V ) with
V the three-dimensional vector space spanned by the vectors {v1 , v2 , v3 }. For
instance, the cyclic permutation π = (123) can be represented by the matrix:
 
0 0 1
T (π) = 1 0 0 ,
0 1 0
since by matrix multiplication T (π)v1 = v2 , T (π)v2 = v3 and T (π)v3 = v1 . We

could do the same thing for the other permutations in S3 : this gives a three-
dimensional representation of S3 . In this way, for Sn we would construct an n-
dimensional representation. This is a faithful representation, and is sometimes
called the permutation representation of Sn .
Example 2.1.5 The regular representation of a finite group. Using
the previous example we can easily construct a faithful representation for all
finite groups G. We know from Cayley’s theorem that every finite group G of
order n is isomorphic to a subgroup of Sn . Using the construction of the n-
dimensional representation of Sn above, this gives us for free an n-dimensional
representation of G, by keeping only the subset of matrices corresponding to
the subgroup isomorphic to G. This is called the regular representation of
G; its dimension is the order of G.
For example, one can think of the group of order three Z3 as a subgroup of
S3 . It is easy to see that Z3 is isomorphic to the subgroup of S3 given by the
cyclic permutations {e, (123), (132)}. Thus, if we denote the elements of Z3 by
{e, ω, ω 2 }, we can write a three-dimensional representation as
     
1 0 0 0 0 1 0 1 0
2
T (e) = 0
 1 0 , T (ω) = 1
 0 0 , T (ω ) = 0
 0 1 .
0 0 1 0 1 0 1 0 0
You can check that these matrices satisfy the multiplication rules in Z3 .
Example 2.1.6 Other representations for Sn . Going back to Sn , we can

construct other representations of different dimensions. For instance, there
is a one-dimensional representation T : Sn → GL(V ) with V = C given by
assigning 1 if g ∈ Sn is even, and −1 if it is odd. We have already seen that
this is a group homomorphism. This is an unfaithful representation.
As a further example, one can construct a two-dimensional representation
for S3 . The idea is to think of symmetries of an equilateral triangle. The group
of symmetries is the dihedral group of order 6, that is D3 , which turns out to
be isomorphic to S3 , as you can check. One can then write two-dimensional
matrices that represent the symmetries of the triangle in the plane:
√ √
1 0 1 1 − 3 1 1 3
T (e) = , T (a) = √ , T (b) = √ ,
0 1 2 − 3 −1 2 3 −1
√ √
−1 0 1 −1 − 3 1 −1 3
T (c) = , T (d) = √ , T (f ) = √ .
0 1 2 3 −1 2 − 3 −1

Example 2.1.7 Two-dimensional representation of the quaternion
group. The quaternion group Q is defined by the set Q = {1, −1, i, −i, j, −j, k, −k}
with multiplication rules i2 = j 2 = k 2 = −1, ij = −ji = k, jk = −kj = i, and
ki = −ik = j. One can show that the following 2 × 2 complex-valued matrices
(i here is the imaginary number) form a faithful representation of Q:

1 0 −1 0 i 0
T (1) = , T (−1) = , T (i) = ,
0 1 0 −1 0 −i

−i 0 0 1 0 −1
T (−i) = , T (j) = , T (−j) = ,
0 i −1 0 1 0

0 i 0 −i
T (k) = , T (−k) = ,
i 0 −i 0

Example 2.1.8 One-dimensional representations of Zn . In Exam-
ple 1.1.9, we defined the cyclic group Zn as the set of n’th roots of unity
under multiplication. This is in fact a one-dimensional representation of the
abstract group Zn . Indeed, writing the group elements as ak = e2πik/n for
k ∈ {0, 1, . . . , n − 1} is in fact a faithful one-dimensional representation of
Zn . But there are other one-dimensional representations. One could write
ak = e2πik`/n for some fixed ` ∈ {0, 1, . . . , n − 1}; this gives in total n different
one-dimensional representations for Zn . We recover the original one for ` = 1.
Are these all faithful? Certainly not. Setting ` = 0 gives the trivial rep-
resentation ak = 1 for all k ∈ {0, 1, . . . , n − 1}, which is clearly not faithful.
In general, the representation will be faithful if ` and n are coprime. Can you
check that?
For instance, for Z3 , if we denote the third root of unity ω = e2πi/3 ,
we get three one-dimensional representations, namely {1, 1, 1}, {1, ω, ω 2 } and
{1, ω 2 , ω}. The last two are faithful.
For Z4 , if we denote the fourth root of unity α = e2πi/4 , we get four one-
dimensional representations, namely {1, 1, 1, 1}, {1, α, α2 , α3 } = {1, i, −1, −i}, {1, α2 , 1, α2 } =
{1, −1, 1, −1} and {1, α3 , α2 , α} = {1, −i, −1, i}. The first and third ones are
not faithful, while the second and fourth ones are.
Example 2.1.9 Representations of (R, +). Consider now the group of real
numbers R under addition. To build a representation for this group, we need
to rewrite addition as matrix multiplication. How can we do that?

One simple way is to use the exponential map. For any u ∈ R, we define
the one-dimensional matrix D(u) = eu . Then D(u + v) = D(u)D(v), hence it
is a group homomorphism. In fact, it is an isomorphism, so it gives a faithful
one-dimensional representation.
We can also build a two-dimensional representation easily. Consider the
matrices
1 0
D(u) = .
u 1
Then D(u + v) = D(u)D(v), so it is a group homomorphism. Again, it is
one-to-one, so it gives a two-dimensional faithful representation.
2.2 Properties of representations

Objectives
• Recall the definition of equivalent representations.
• Recall the definition of reducible and irreducible representations.
• Construct the restriction of a representation of a group G to a given
subgroup H ⊂ G.
• Construct the tensor product of two given matrix representations.
2.2.1 Equivalent representations

We defined representations as group homomorphisms between abstract groups
G and operators on a vector space V . After choosing a basis, we can think
of those as invertible matrices. So if we think of representations in terms of
matrices, we want to make sure that we only study “different” representations.
What we mean here is that we do not want to distinguish between matrix
representations that come from the same operators, but in different choices of
bases.
Recall from linear algebra that if you are given a vector equation b = Aa,
and that you want to rewrite this equation in a different basis which is obtained
by applying the transformation B, we get:
Bb = BAa = BAB −1 Ba.
So, in this new basis with b0 = Bb and a0 = Ba, we get the equation
b0 = A0 a0 ,
with A0 = BAB −1 . This is what is called a similarity transformation. So

we want to say that two representations are the same if they are related by a
similarity transformation.
Definition 2.2.1 Equivalent representations. We say that two repre-
sentations T : G → GL(V ) and T 0 : G → GL(V ) are equivalent if, given a
choice of basis for V , all matrices T (g) and T 0 (g) are related by a similarity
transformation:
T 0 (g) = BT (g)B −1 ,
for some invertible matrix B. ♦
In other words, one can think of equivalent representations as mapping the

group elements to the same operators on V , but in a different choice of basis.
Note that any two distinct one-dimensional representations cannot be equiv-
alent, since complex numbers commute.
2.2.2 Irreducible and reducible representations

Our goal is to understand and classify representations of groups in general.
But not all representations are equally interesting. We want to define a notion
of representations that are “minimal”, in the sense that they are somehow the
building block for constructing all representations of a given group. “Minimal”
representations should be representations that have no “sub-structure”. Let us
first define the concept of subrepresentation.
Definition 2.2.2 Subrepresentation. Let T : G → GL(V ) be a represen-
tation of a finite group G. A vector subspace U ⊂ V is said to be T -invariant
if T (g)u ∈ U for all u ∈ U . The restriction of T to the subspace U is called a
subrepresentation. ♦
Thus, if a representation has a subrepresentation, this means that there
exists a subspace U ⊂ V that is closed under the action of the operators in
the image of the representation. So it has some sub-structure. We want our
“minimal” representations, our building block, to have no such sub-structure.
Definition 2.2.3 Reducible and irreducible representations. If a rep-
resentation T : G → GL(V ) has a non-trivial subrepresentation, then it is said
to be reducible. It is otherwise irreducible. ♦
Irreducible representations are the minimal representations that we are
looking for. They are the true building blocks of representation theory. In
fact, a central goal of representation theory is to establish criteria to deter-
mine whether a given representation is irreducible or not, and to classify all
possible irreducible representations of a given group.
By the way in the beginning of this section we said that we would focus
on finite-dimensional representations, but that representations could also be
defined for infinite-dimensional vector spaces. However, there is a theorem
that says that:
Theorem 2.2.4 Irreducible representations of a finite group. All
irreducible representations of a finite group are finite-dimensional.
So as far as irreducible representations of finite groups are concerned, it is
sufficient to consider only finite-dimensional representations.
2.2.3 Semisimple representations

Let us now go back to the study of general representations. There is an easy
way to construct higher-dimensional representations from lower-dimensional
ones. Consider the following example. Suppose that I ask you to give me a
6-dimensional representation of, say, S3 . Well, you could say, sure, I’ll start
with the trivial representation that assigns 1 to all permutations in S3 , and I’ll
assign the 6 × 6 matrix:
 
1 0 0 0 0 0
0 1 0 0 0 0
 
0 0 1 0 0 0
 
0 0 0 1 0 0
 
 
0 0 0 0 1 0
0 0 0 0 0 1
to all permutations in S3 . That is certainly a 6-dimensional representation,

but it’s a rather boring one.
You could be a little fancier, and start instead with the 3-dimensional
representation of S3 that we constructed in Example 2.1.4. If we let T (g) be
the 3×3 matrices associated to the group elements g ∈ S3 in this representation,
you could construct a 6-dimensional representation by constructing the block
diagonal matrices:
T (g) 0
0 T (g)
Those are examples of direct sums of representations.
Definition 2.2.5 Direct sum of representations. Suppose that T : G →
GL(V ) and S : G → GL(W ) are m- and n-dimensional representations of
a group G. We construct the (m + n)-dimensional direct sum representa-
tion T ⊕ S : G → GL(V ⊕ W ) by forming block diagonal matrices, with the
m-dimensional matrices T (g) in the upper left block and the n-dimensional
matrices S(g) in the bottom right block:

T (g) 0
T (g) ⊕ S(g) = .
0 S(g)
♦
Representations that are constructed as direct sums are certainly not min-
imal. More precisely, with our definition of minimality above, we can see that
all direct sum representations are reducible. Indeed, consider a direct sum
representation T ⊕ S : G → GL(V ⊕ W ). Then the subspaces V ⊂ V ⊕ W and
W ⊂ V ⊕ W are certainly T -invariant, as all matrices T (g) are block diagonal,
and hence do not mix elements of V with elements of W . Thus the restrictions
of T ⊕ S to the subspaces V and W are subrepresentations, and hence T ⊕ S
is reducible.
So we have seen that all direct sum representations are reducible. But
this raises an interesting question: is the converse true? Can all reducible
representations be written as direct sums of irreducible representations? In
other words, can all reducible representations be written in block diagonal
form? We will see shortly that the answer is yes for finite groups, but no in
general. Thus it makes sense to distinguish between representations that can
or cannot be written as direct sums of irreducible representations.
Definition 2.2.6 Semisimple representations. A representation is semisim-
ple (also called completely reducible) if it is a direct sum of irreducible repre-
sentations. ♦
With this definition, the question becomes: are all representations of a
given group semisimple? (i.e. either irreducible or direct sums of irreducible
representations.) We will come back to this shortly.
Remark 2.2.7 A consequence of the direct sum construction is that for any
group, there exists an infinite number of reducible representations, which we
can construct as direct sums. This is why, from a classification viewpoint, it is
much more interesting to focus on irreducible representations.
However, in physics one often encounters representations that are reducible
(in fact semisimple). So it is also interesting to develop methods to find the
decomposition of a semisimple representation as a direct sum of irreducible
representations. For instance, in the case of a quantum mechanical Hamiltonian
H with a symmetry group G, we know that solutions will transform according
to some semisimple representation T of G. In other words, we can decompose
the space of solutions as a direct sum of invariant subspaces according to the

decomposition of the semisimple representation as a direct sum of irreducible
representations. As we will see later on, all states in a given invariant subspace
must be eigenstates of the Hamiltonian with the same energy eigenvalue. Thus,
if we are interested in the energy levels of our system, what we first want to do
is decompose the space of states into invariant subspaces, that is, decompose
the representation T into irreducible representations of the symmetry group
G.
2.2.4 Restriction to subgroups

Reducible representations also arise naturally when one considers a subgroup
H of a group G. A representation of G clearly gives a representation of H
if one keeps only the matrices corresponding to the elements of the subgroup
H ⊂ G. However, when restricted to a subgroup H ⊂ G, a representation
that is irreducible for G may become reducible for H. Indeed, the induced
representation for H may have a subrepresentation (suppose for example that
all matrices corresponding to the subgroup are block diagonal), while the orig-
inal representation for G may not (suppose that the matrices corresponding to
elements of G not in H are not block diagonal). A central theme in physics is
to find how a given irreducible representation of G decomposes into a direct
sum of irreducible representations of H upon restriction to a subgroup H ⊂ G.
Example 2.2.8 Restriction to Z2 ⊂ S3 . As an example of this, let us
look back at the two-dimensional representation T : S3 → GL(V ) that we
constructed in Example 2.1.6. We reproduce it here for convenience:
√ √
1 0 1 1 − 3 1 1 3
T (e) = , T (a) = √ , T (b) = √ ,
0 1 2 − 3 −1 2 3 −1
√ √
−1 0 1 −1 − 3 1 −1 3
T (c) = , T (d) = √ , T (f ) = √ .
0 1 2 3 −1 2 − 3 −1
One can show that this representation is irreducible. Let us now consider the
order two subgroup H = {e, c} ⊂ S3 . The restriction of T to the subgroup
H ' Z2 gives the two-dimensional representation:

1 0 −1 0
T (e) = , T (c) = ,
0 1 0 1
which is now reducible (in fact semisimple), as a representation of H, since it

is in block diagonal form. It is the direct sum of the trivial representation of
Z2 and its non-trivial one-dimensional representation.
2.2.5 Tensor product representations

We have already seen one way of constructing higher-dimensional representa-
tions from lower-dimensional ones. Given two representations T : G → GL(V )
and S : G → GL(V 0 ) of the same group, we have already seen how to construct
a (m + n)-dimensional representation on V ⊕ V 0 by adjoining the matrices into
(m + n) × (m + n) block diagonal matrices. We called this construction the
direct sum, and denoted it by T ⊕ S.
There is another way to construct higher-dimensional representations from
lower-dimensional ones. We can also take the “product” of representations to
construct a new mn-dimensional representation: this is called the tensor prod-
uct representation, and we denote it by T ⊗ S. What we are really doing here
is define a new representation T ⊗ S that is an operator on the tensor product

of the underlying vector spaces. If you know what this means, great, otherwise
I will simply be pedestrian here and construct the matrix representation T ⊗ S
explicitly.
Let us start with the simpler case where T and S are two-dimensional
representations, with matrices:

T11 (g) T12 (g) S11 (g) S12 (g)
T (g) = , S(g) = .
T21 (g) T22 (g) S21 (g) S22 (g)
Then the 4 × 4 matrices T ⊗ S(g) are constructed by taking the product of all
matrix entries (this is not a matrix product of course):

T11 (g)S(g) T12 (g)S(g)
T ⊗ S(g) =
T21 (g)S(g) T22 (g)S(g)
 
T11 (g)S11 (g) T11 (g)S12 (g) T12 (g)S11 (g) T12 (g)S12 (g)
T11 (g)S21 (g) T11 (g)S22 (g) T12 (g)S21 (g) T12 (g)S22 (g)
= T21 (g)S11 (g) T21 (g)S12 (g) T22 (g)S11 (g) T22 (g)S12 (g) .

T21 (g)S21 (g) T21 (g)S22 (g) T22 (g)S21 (g) T22 (g)S22 (g)
It is easy to see how this generalizes to arbitrary m and n. The mn × mn

tensor product matrices are constructed as:
 
T11 (g)S(g) . . . T1n (g)S(g)
T ⊗ S(g) =  .. .. ..
,
 
. . .
Tn1 (g)S(g) . . . Tnn (g)S(g)
where of course one could expand out by writing down the matrices S(g)
explicitly as above.
We call the corresponding representation the tensor product represen-
tation. Such tensor product representations occur frequently in physics, for
instance when taking the product of wave-functions in quantum mechanics.
Checkpoint 2.2.9 Check that the tensor product representation constructed
above is indeed a representation, i.e. that T ⊗ S : G → GL(mn, C) is a group
homomorphism.
One important point is that even if T and S are irreducible representations
of G, the tensor product T ⊗ S may be reducible. In fact often in particle
physics one needs to determine how the tensor product of two irreducible rep-
resentations decomposes into a sum of irreducible representations.
2.3 Unitary representations

Objectives
• Determine whether an explicitly given representation is unitary.
• Recall and state the unitarity theorem for finite groups.
One type of representations is particularly important in physics: unitary
representations. Those are widely used in quantum mechanics. So studying
unitary representations in general allows physicists to learn lots of interesting
things about quantum systems. You can look at these slides by Peter Woit if
you want to know more about this, but we will come back to this later on in
this course.
Mathematically, unitary representations are also generally easier to handle
than general representations: they satisfy nice properties not shared by general
representations. It turns out that for finite groups, all representations are
equivalent to unitary representations. This is not so simple for infinite groups
though.
2.3.1 Review of linear algebra

Let us start by reviewing a bit of linear algebra.
Definition 2.3.1 Adjoint of a matrix. Let A ∈ GL(n, C) be an n × n
matrix with complex-valued entries. We write A∗ for its complex conjugate,
obtained by complex conjugating the entries. We write A† = (A∗ )T for the
transpose of the complex conjugated matrix. A† is called the adjoint (or
Hermitian conjugate). ♦
Remark 2.3.2 Note that in the literature, particularly in the mathematics
literature, the notation A∗ is often use to denote the adjoint A† .
Definition 2.3.3 Hermitian matrices. A matrix A ∈ GL(n, C) is Hermi-
tian if A = A† . ♦
Hermitian matrices are also all over the place in quantum mechanics, since
observables are generally understood in terms of Hermitian operators on the
Hilbert space of states. The eigenvalues of these operators correspond to the
possible values of the particular observable variables represented by the oper-
ators. This interpretation makes sense because of the following fundamental
result in linear algebra:
Theorem 2.3.4 Eigenvalues of Hermitian matrices. All eigenvalues of
a Hermitian matrix A ∈ GL(n, C) are real, and A has n linearly independent
eigenvectors that can all be chosen to be orthogonal.
Another type of complex-valued matrices is important:
Definition 2.3.5 Unitary matrices. We say that a matrix A ∈ GL(n, C)
is unitary if AA† = I, that is, A† = A−1 . ♦
Unitary matrices preserve the inner product on a complex vector space,
just like orthogonal matrices preserve the scalar product on real vector spaces.
Given two vectors u, v ∈ Cn , the standard inner product is defined as hu, vi =
u† v. Then a unitary transformation A preserves the inner product, since
hAu, Avi = u† A† Av = u† v = hu, vi.
This is why unitary operators are important in quantum mechanics: they can
be used to do changes of bases while preserving orthogonality between basis
vectors according to the inner product. They behave just like rotations but in
a complex vector space.
Another nice property of unitary matrices concerns their eigenvalues. While
eigenvalues of Hermitian matrices are real, eigenvalues of unitary matrices are
generally complex, but of a very particular form: they must lie on the unit
circle. In other words, they must have modulus 1. This means that they can
be written as eiθ for some angle θ ∈ [0, 2π).
It also turns out that unitary and Hermitian matrices are closely related.
Indeed, the theorem about eigenvalues of Hermitian matrices above relies on
the fundamental statement that any Hermitian matrix can be diagonalized by
an appropriate unitary transformation. That is, for any Hermitian matrix H,

there exists a unitary matrix U such that
D = U † HU
is a diagonal matrix.
Checkpoint 2.3.6 Prove that any Hermitian matrix can be diagonalized by
an appropriate unitary transformation.
2.3.2 The unitarity theorem

Let us now go back to representations.
Definition 2.3.7 Unitary representation. If all matrices T (g) of a repre-
sentation of a group G are unitary, we say that the representation is unitary.
♦
Unitary representations may seem like very special and constrained. Which
raises a question: how common are unitary representations?
Let us consider first one-dimensional representations of a finite group G. To
any element g ∈ G, a one-dimensional representation T associates a complex
number T (g) = reiθ . (This is the polar form of a non-zero complex number,
where r > 0 is the modulus, and θ ∈ [0, 2π).) Since G is finite, we know
that g has finite order, and hence there must exist an integer n such that
g n = e. That is, T (g n ) = rn einθ = T (e) = 1. This is only possible if r = 1.
In other words, what this means is that T (g) has modulus one, and hence is
a one-dimensional unitary matrix, since T † (g)T (g) = r2 = 1. Therefore, all
one-dimensional representations of finite groups are unitary. Note that it was
key in the argument that the group is finite.
Does there exist a similar statement for higher-dimensional representations
of finite groups? This is the content of the important unitarity theorem, the
main result that we prove in this section:
Theorem 2.3.8 Unitarity theorem. Any finite-dimensional representation
of a finite group G is equivalent to a unitary representation. In other words, it
can be brought into unitary form by a similarity transformation.
Proof. Let G be a finite group, and T : G → GL(V ) be an n-dimensional
representation. The proof is constructive: for any such T , we will construct an
equivalent representation that is explicitly unitary.
To this end, let us first introduce the n × n matrix:
X
H= T † (g)T (g).
g∈G
This matrix is interesting. For instance, it satisfies the following invariance

property. For any g 0 ∈ G,
X
T † (g 0 )HT (g 0 ) = T † (g 0 )T † (g)T (g)T (g 0 )
g∈G
X
= (T (g)T (g 0 ))† T (g)T (g 0 )
g∈G
X
= T (gg 0 )† T (gg 0 )
g∈G
=H. (2.3.1)
The last line follows because of the Rearrangement theorem 1.4.2, since mul-
tiplying the group elements g ∈ G by a fixed element g 0 only rearranges the
terms in the sum. This invariance property will be useful later on.
Now notice that H is Hermitian, since H † = H. Thus we know that it has

real eigenvalues, and that is can be diagonalized by a unitary matrix. Hence
we can write D = U † HU , where D is a diagonal matrix with real entries and
U is unitary. Further, we now show that D has real positive entries. We have:
D = U † HU
X
= U † T † (g)T (g)U
g∈G
X
= A† (g)A(g),
g∈G
where we defined A(g) = T (g)U . Now consider the j’th diagonal entry Djj . It
is given by summing over g ∈ G the contributions given by A†j (g)Aj (g), where
Aj (g) denotes the j’th column vector in A(g). Since for each g ∈ G and each
j, Aj (g) is a non-zero vector, then A†j (g)Aj (g) > 0, and hence Djj > 0.
We then define the diagonal matrix D1/2 whose entries are the square roots
of the entries of D, and D−1/2 as the inverse of D1/2 . We now form the matrices
B(g) = D1/2 U † T (g)U D−1/2 , and their adjoints B † (g) = D−1/2 U † T † (g)U D1/2 .
Why are we constructing these matrices? Note that, since U is unitary, and
hence U † = U −1 , the transformation B(g) = D1/2 U † T (g)U D−1/2 is a similar-
ity transformation. Thus the representations furnished by the matrices B(g)
and the T (g) are equivalent. Our goal is to show that the new representation
B(g) is explicitly unitary, which would prove the theorem, namely that any
representation of a finite group is equivalent to a unitary representation.
So let us show that the matrices B(g) are unitary. We have:
B † (g)B(g) =D−1/2 U † T † (g)U D1/2 D1/2 U † T (g)U D−1/2

=D−1/2 U † T † (g)U DU † T (g)U D−1/2
=D−1/2 U † T † (g)HT (g)U D−1/2
=D−1/2 U † HU D−1/2
=D−1/2 DD−1/2
=I,
where we used the invariance property (2.3.1). Thus for any finite-dimensional
representation T of a finite group G, we have constructed a new, equiva-
lent unitary representation B, given by the set of unitary matrices B(g) =
D1/2 U † T (g)U D−1/2 . We have thus proved that all finite-dimensional repre-
sentations of finite groups are equivalent to unitary representations.
Remark 2.3.9 Note here that the requirement of having a finite group G was
crucial in the proof. Otherwise, the expression
X
H= T † (g)T (g)
g∈G
doesn’t even make sense, since the sum would be over an infinite-dimensional
set (or a continuous space if the group is continuous).
Remark 2.3.10 In view of Remark 2.3.9, we may ask: is the unitarity theorem
still true for infinite groups, either discrete or continuous? Consider for example
the infinite continuous group (R, +), and the two-dimensional representation:

1 0
T (u) = , u∈R
u 1
that we encountered in Example 2.1.9. We have:

† 1 u 1 0
T (u)T (u) =
0 1 u 1
2

1+u u
= ,
u 1
which is not the identity for non-zero u ∈ R. So this is not a unitary transforma-
tion, and one can show that it cannot be brought into a unitary transformation
by a similarity transformation. In fact, the same representation restricted to
u ∈ Z is also not unitary for the infinite but discrete group (Z, +).
So for what kind of groups, beyond finite groups, is the unitarity theorem
true? The general result is that the theorem holds if the group is compact.
To define compact groups, we first need to define the concept of topological
groups. Topological groups are groups G that are given the extra structure of
a topology on G, such that the group’s binary operation and function mapping
elements to their inverses are continuous with respect to this topology. Then,
a compact group is a topological group such that its topology is compact.
In the end, they key point is that because of the existence of this compact
topology on G, for compact groups we can “replace” the sum over elements of
the group by an appropriate integral over the continuous group with respect to
some measure (the Haar measure), and the resulting integral then converges.
Using this the proof above goes through with minor modifications, and the
unitarity theorem holds for compact groups. Examples of compact groups
include SO(n) and SU (n), which appear frequently in physics. Note however
that the Lorentz group is not compact.
Note that not only the unitarity theorem holds for compact groups, but
Theorem 2.2.4 also applies to compact groups: all irreducible representations
of compact groups are finite-dimensional.
2.4 Semisimplicity
Objectives
• Recall that every representation of a finite group is semisimple.
In Definition 2.2.3 and Definition 2.2.6 we introduced reducible and semisim-
ple representations. We then asked whether all representations of a given group
are semisimple. That is, whether all representations are either irreducible or
can be written as direct sums of irreducible representations. With our de-
tour study of unitary representations, we are now in a position to answer this
question for finite groups.
We said earlier that unitary representations are quite nice, and satisfy prop-
erties that are not shared by all representations. An example of this is the
following important theorem.
Theorem 2.4.1 Unitary representations are semisimple. All finite-
dimensional unitary representations are semisimple. That is, all finite-dimensional
unitary representations are either irreducible or direct sums of irreducible rep-

resentations.
Proof. Let T : G → GL(V ) be a unitary representation. If T is irreducible,
then we are done. So let us assume that T is reducible. Then there exists a
T -invariant subspace U ⊂ V . Let W be the orthogonal complement of U in
V . Then V = U ⊕ W . We now show that W is also T -invariant, and hence
the representation T must leave both U and its orthogonal complement W
invariant. This means that it must decompose in block diagonal form, i.e. it
is a direct sum of representations.
To prove this, we formulate the T -invariance condition using projection
operators. Recall a few facts from linear algebra. A projection P on a vector
space V is an operator P : V → V such that P 2 = P . Let U ⊂ V be the
range of P , and W ⊂ V be its kernel. Then one can think of P as projecting
vectors onto the subspace U . Further, it follows that V = U ⊕ W , and that
Q = I − P is also a projection, but now onto the subspace W ⊂ V . If we
have a notion of inner product on V , we can define an orthogonal projection
P : a projection for which the range U and the kernel W are orthogonal. A
projection is orthogonal if and only if it is given by a Hermitian matrix.
Let us then introduce an orthogonal (thus Hermitian) projection matrix
onto the subspace U ⊂ V , that is P : V → U , with P 2 = P and P = P † . Then
the statement that U is T -invariant can be written as the condition that
P T (g)P = T (g)P
for all g ∈ G. This is because the projection operator P acts as the identity
operator on the subspace U ⊂ V , and P v ∈ U for all v ∈ V , so the condition
that T (g)u ∈ U for all u ∈ U will be satisfied if and only if P T (g)P v = T (g)P v
for all v ∈ V .
Take the Hermitian conjugate on both sides. We get the condition P T † (g)P =
P T † (g). We know that the matrices T (g) are unitary, and hence T † (g) =
T −1 (g) = T (g −1 ). Thus the condition becomes P T (g −1 )P = P T (g −1 ), but
since this must be true for all g ∈ G, we can write it simply as
P T (g)P = P T (g)
for all g ∈ G. Reorganizing, we get:
P T (g)P − P T (g) − T (g)P + T (g) = −T (g)P + T (g),
that is,
(1 − P )T (g)(1 − P ) = T (g)(1 − P ).
But the projection matrix 1 − P projects on the orthogonal complement W of
the subspace U ⊂ V . Since V = U ⊕ W , this means that T must decompose
in block diagonal form, i.e. it is a direct sum of representations.
We continue this process inductively. Since the representation is finite-
dimensional, the process must stop at some point, and we end up with T being
a direct sum of irreducible representations, that is, a semisimple representation.

Now we can use this key result to study whether representations of finite
groups can be written as direct sums of irreducible representations. Indeed, in
Theorem 2.3.8 we showed that all finite-dimensional representations of finite
groups are equivalent to unitary representations. Therefore Theorem 2.4.1
implies the following corollary:
Corollary 2.4.2 Finite-dimensional representations of finite groups

are semisimple. All finite-dimensional representations of finites groups are
equivalent to semisimple representations. That is, they are either irreducible
or equivalent to direct sums of irreducible representations.
Thus as far as finite groups are concerned, irreducible representations really
are the true building blocks, since all finite-dimensional representations can be
constructed as direct sums of irreducible representations.
Remark 2.4.3 The key result here is Theorem 2.4.1, which applies to finite-
dimensional unitary representations in general, regardless of whether the group
is finite or infinite. Moreover, for the case of compact groups, such as SU (n)
and SO(n), the unitarity theorem Theorem 2.3.8 also holds. Thus one can
generalize Corollary 2.4.2 to compact groups: all finite-dimensional represen-
tations of compact groups are either irreducible or equivalent to direct sums of
irreducible representations.
However, just like the unitarity theorem, this is not true in general for
infinite groups. For instance, consider the two-dimensional representation of
(R, +) given by
1 0
, u∈R
u 1
This representation is not a direct sum, hence it is not semisimple, but it is
0
reducible. Indeed, the subspace spanned by is invariant, hence this two-
1
dimensional representation has a one-dimensional subrepresentation, even if it
is not block diagonal and cannot be brought into block diagonal form by a
similarity transformation.
The outcome of the last two sections is that, as long as finite (or compact)
groups are concerned, we only have to deal with direct sums of irreducible
representations (up to equivalence). Phew!
2.5 Schur’s lemmas

Objectives
• Recall, state and prove Schur’s first and second lemmas.
• Recall that all irreducible representations of finite abelian groups are
one-dimensional.
The two lemmas of Schur are fundamental in representation theory. The
ultimate goal of the next few sections is to develop tools to determine whether
a given representation is irreducible or not. The fundamental result in this
direction is called the “great orthogonality theorem”. But to be able to prove
this theorem, we first need to introduce two lemmas that are interesting in
their own right, known as “Schur’s lemmas”.
Schur’s lemmas are concerned with the type of matrices that commute
with all matrices of irreducible representations of a finite group G. This will
be important to prove the great orthogonality theorem.
2.5.1 Schur’s first lemma

Schur’s first lemma gives us a criterion to determine when two representations
are irreducible.
Lemma 2.5.1 Schur’s first lemma. Let T : G → GL(V ) and S : G →

GL(U ) be two irreducible representations of a finite group G. If a matrix A is
such that
AT (g) = S(g)A ∀g ∈ G,
then either A = 0, or A is an invertible square matrix and hence the two
representations are equivalent.
Proof. Suppose that T is n-dimensional, and S is m-dimensional. First we
note that A is a m × n matrix. Suppose that there is a subspace W ⊂ V such
that Aw = 0 if and only if w ∈ W . Let P be the projection matrix on the
subspace W ⊂ V . Then AP = 0. It follows that
AT (g)P = S(g)AP = 0.
But then, this implies that T (g)P ∈ W , or, in other words, T (g)w ∈ W for any
w ∈ W . This means that W is a T -invariant subspace. Since W is irreducible,
the only possible invariant subspaces are the trivial ones, namely either W = V
or W = {0}. In the first case, this implies that A = 0. In the second case, this
means that Av is never zero for all non-zero v ∈ V .
We can redo the same argument by starting with a subspace W 0 ∈ U such
that uA = 0 if and only if u ∈ W 0 . Following the same steps as above, we
end up with two possibilities; either A = 0 or uA is never zero for all non-zero
u ∈ U.
So we are left with two cases: either A = 0, or Av and uA are never zero
for non-zero v ∈ V and u ∈ U . We now show that the second case implies that
A is an invertible square matrix. First, if the number of rows is less than the
number of columns, than there must exist a v ∈ V such that Av = 0, which is
a contradiction. Similarly, if the number of columns is less than the number of
rows, than there must exist a u ∈ U such that uA = 0, which is a contradiction.
Therefore A must be square, that is, m = n. Finally, we know that a square
matrix has a non-trivial kernel if and only its determinant is zero. Therefore
A must have non-zero determinant, or equivalently, it must be invertible.
Therefore, we conclude that either A = 0 or A is an invertible square matrix.
In the latter case, if then follows that
AT (g)A−1 = S(g),
that is, T and S are equivalent representations.
2.5.2 Schur’s second lemma

The second lemma studies what kind of matrices commute with all matrices
of a given irreducible representation.
Lemma 2.5.2 Schur’s second lemma. Let T : G → GL(V ) be an irre-
ducible representation of a finite group G. If a matrix A commutes with T (g)
for all g ∈ G, that is
AT (g) = T (g)A ∀g ∈ G,
then A = λI for some λ ∈ C. In other words, A is a constant multiple of the

identity matrix.
This is a rather powerful statement. In general, if you were given a bunch of
matrices T1 , . . . , Tn , there would potentially be many matrices that commute
with all of those. However, if the matrices Ti are not arbitrary but actually
form an irreducible representation of a finite group, things are much more
constrained: only multiples of the identity matrix commute with the Ti .

Proof. The second lemma is a direct consequence of the first one. A must have
at least one eigenvalue, which we call λ. Then (A − λI) has determinant zero,
and hence is not invertible. But since AT (g) = T (g)A, it follows that:
(A − λI)T (g) = T (g)(A − λI).
From Schur’s first lemma Lemma 2.5.1, it then follows that A − λI = 0, since
A − λI is not invertible. Therefore A = λI for some λ ∈ C.
2.5.3 Irreducible representations of abelian groups

A direct consequence of Schur’s second lemma is the following theorem:
Theorem 2.5.3 Irreducible representations of finite abelian groups.
All irreducible representations of finite abelian groups are one-dimensional.
Proof. Let T : G → GL(V ) be an irreducible representation of a finite abelian
group G. Then for any a ∈ G, T (a) commutes with all T (g), g ∈ G. Thus, by
Schur’s second lemma, T (a) = λI for some λ ∈ C. This is true for all a ∈ G,
and hence all T (a) are scalar multiples of the identity matrix. It then follows
that the only way that T can be irreducible is if it is one-dimensional.
Note that as usual, while we stated the result only for finite groups, it holds
whenever Schur’s second lemma holds. In particular, it is true for compact
groups: all complex irreducible representations of compact abelian groups are
one-dimensional.
2.6 The great orthogonality theorem

Objectives
• Recall, state and prove the great orthogonality theorem.
• Deduce immediate consequences of the great orthogonality theorem.
We are now in a position to prove one of the most fundamental result in
representation theory: the great orthogonality theorem. In essence, what the
theorem says is that if we think of irreducible representations as being given by
“vectors of matrices”, where the “vectors” are indexed by the group elements,
then they satisfy orthogonality relations. This is key to determine the possible
irreducible representations of given group.
Theorem 2.6.1 The great orthogonality theorem. Let T : G → GL(V )
and S : G → GL(V 0 ) be two inequivalent irreducible unitary representations
of a finite group G. Let T (g)ij and S(g)ij denote the matrix elements of the
corresponding matrices, for all g ∈ G. Then
X
T † (g)ij S(g)kl = 0.
g∈G
For the matrices of a single irreducible unitary representation (or two equiv-
alent ones), the relation is:
X |G|
T † (g)ij T (g)kl = δil δjk ,
d
g∈G
where d is the dimension of the representation.

Proof. This theorem follows from Schur’s lemmas. Let T and S be inequiv-
alent irreducible unitary representations of dimensions d and d0 respectively.
Consider an arbitrary d × d0 matrix X, and let
X X
A= T (g)XS(g −1 ) = T (g)XS † (g),
g∈G g∈G
where the second equality follows since S is unitary. Now multiply by T (h) on
the left for any h ∈ G:
X
T (h)A = T (h)T (g)XS(g −1 )
g∈G
X
= T (h)T (g)XS(g −1 )S(h−1 )S(h)
g∈G
X
= T (hg)XS † (g)S † (h)S(h)
g∈G
X
= T (hg)X(S(h)S(g))† S(h)
g∈G
 
X
= T (hg)XS((hg)−1 ) S(h).
g∈G
By the rearrangement theorem Theorem 1.4.2, multiplying all g ∈ G by h ∈ G

does not change the sum over G, it simply rearranges the order of the terms
in the sum. Thus we get:
T (h)A = AS(h).
But then, by Schur’s first lemma Lemma 2.5.1, and the fact that T and S are
assumed to be inequivalent, we conclude that A = 0. Therefore
X
T (g)XS † (g) = 0.
g∈G
Now we can choose X to be any matrix. In particular, we can choose it to be

a matrix with zero entries everywhere except in the j’th row and k’th column
where it is one. Then we get:
X
T (g)ij S † (g)kl = 0,
g∈G
for all i, j, k, l, which is the first orthogonality relation.

For the second one, we start with the matrix
X X
A= T (g)XT (g −1 ) = T (g)XT † (g).
g∈G g∈G
Going through the same steps, we end up with the statement
T (h)A = AT (h).
Schur’s second lemma Lemma 2.5.2 now implies that A = λX I for some com-
plex number λX ∈ C. We wrote λX to remind ourselves that the constant
depends on the choice of matrix X in the definition of A. Thus we have
X
T (g)XT (g −1 ) = λX I.
g∈G
Next, we need to fix the constants λX . We take the trace on both sides of
the equation. On the left-hand-side, we get:
X X
Tr T (g)XT (g −1 ) = Tr XT (g −1 )T (g)

g∈G g∈G
X
= Tr X
g∈G
=|G| Tr X.
On the right-hand-side, we get:
λX Tr I = dλX ,
where d is the dimension of the representation. Therefore we conclude that
|G|
λX = Tr X,
d
and the relation becomes
X |G|
T (g)XT (g −1 ) = I Tr X.
d
g∈G
To conclude the proof, we choose the matrices X as above, with zero entries
everywhere except in position (j, k), wherePit is taken to be 1. Using index
notation, if we write the left-hand-side as g∈G T (g)ij T † (g)kl , we can write
I = δil , and Tr X = δjk , since the trace is non-zero only if the entry 1 in X is
on the diagonal. Therefore, we obtain the second orthogonality relation:
X |G|
T (g)ij T † (g)kl = δil δjk .
d
g∈G

The great orthogonality theorem gives orthogonality relations between the
matrices of the irreducible representations of any finite group G. Those are very
useful, as will become even clearer when we introduce characters. But for the
moment we can already deduce an immediate consequence of the orthogonality
theorem:
Theorem 2.6.2 The number of irreducible representations of a finite
group. A finite group can only have a finite number of inequivalent irreducible
representations.
Proof. This is clear from the orthogonality theorem. We can think of irre-
ducible representations as giving “vectors of matrices” (T (g)ij )g∈G in a vector
space of dimension |G|. The theorem tells us that those vectors must be or-
thogonal. But there are at most |G| orthogonal vectors in a vector space of
dimension |G|, and so the number of irreducible representations must be finite.
In fact we will calculate the number of irreducible representations for any finite
group explicitly when we introduce characters.
2.7 Characters
Objectives
• Compute the character of a representation.
• Recognize that the character is a function of class.
• Calculate the character of direct sum and tensor product representations.

In this section we introduce the notion of the character of a representation.
The idea is simple: we take the traces of the matrices in a given representa-
tion. Characters will give a clean way of studying and classifying irreducible
representations, and determining whether a given representation is reducible
or not.
2.7.1 Definition
Definition 2.7.1 The character of a representation. Let T : G → GL(V )
be a representation of G. The character of this representation is given by
the map χ : G → C, which assigns to every group element the trace of the
corresponding matrix in the representation T :
χ(T ) (g) = Tr T (g).
If T is irreducible, the character is called simple (or irreducible), while it is

called compound otherwise. The degree of a character is the dimension of
the corresponding representation. ♦
2.7.2 A few simple properties

The first thing to note about characters is that they are the same for all
elements that are in the same conjugacy class. This is usually encapsulated in
the statement that “character is a function of class”:
Lemma 2.7.2 Character is a function of class. The character of a repre-
sentation is a class function, meaning that all elements of a group belonging
to the same conjugacy class have the same character.
Proof. Let x, y ∈ G be conjugate. That is, there exists a g ∈ G such that
x = gyg −1 . Since a representation is a group homomorphism, this means that
T (x) = T (g)T (y)T (g −1 ). Taking the trace, we get
Tr T (x) = Tr T (g)T (y)T (g −1 ) = Tr T (y)T (g −1 )T (g) = Tr T (y),

where we used the cyclic property of the trace of a product of matrices.

It is also important to note that equivalent representations have the same
character:
Lemma 2.7.3 Equivalent representations have the same character.
If T : G → GL(V ) and S : G → GL(V ) are equivalent representations, then
χ(T ) (g) = χ(S) (g) for all g ∈ G.
Proof. Two representation T and S are equivalent if their matrices are related
by a similarity transformation T (g) = AS(g)A−1 . Taking the trace, and using
cyclicity again, we get:
Tr T (g) = Tr AS(g)A−1 = Tr S(g)A−1 A = Tr S(g).


What this tells us is that if two representations have different characters,
then they must be inequivalent. We will see that the converse statement (that
two inequivalent representations must have different characters) follows from
the orthogonality theorem.
Another simple property of characters is how they behave for direct sums
and tensor products of representations. We will leave the proof of the following
lemma as an exercise:
Lemma 2.7.4 Character of a direct sum and a tensor product. Let
T : G → GL(V ) and S : G → GL(V ) be two representations of G. Then
χ(T ⊕S) (g) = χ(T ) (g) + χ(S) (g), χ(T ⊗S) (g) = χ(T ) (g)χ(S) (g).
In particular, the character of a semisimple representation T can be written
as
n
X
(T )
χ (g) = ai χ(Ti ) (g),
i=1
where the Ti are the irreducible representations appearing in the direct sum
with multiplicity ai .
2.8 Orthogonality for characters

Objectives
• Recall and prove the orthogonality relations for characters.
• Calculate the number of irreducible representations of a finite group.
• Constrain the dimensions of the irreducible representations of a finite

group.
• Constrain the character table of a group using orthogonality relations for
characters.
Characters are very useful to, indeed, characterize representations. This
becomes clear when we state the orthogonality relations in terms of characters.
We know that equivalent representations have the same character. We will see
that inequivalent representations have different characters, in fact orthogonal.
2.8.1 First orthogonality theorem

Recall that the character of a representation T is a complex-valued function
χ(T ) : G → C. We can think of it as a vector χ(T ) in a |G|-dimensional complex
vector space. The standard inner product is then given by
X ∗
hχ(T ) , χ(S) i = χ(T ) (g)χ(S) (g).
g∈G
In fact, we can do better. We know that the character is a class function,

that is, all group elements in a conjugacy class have the same character. If c
is the number of conjugacy classes in the group G, then we can write
c ∗
(T ) (S)
X
hχ(T ) , χ(S) i = ni χi χi ,
i=1
where ni is the number of elements in the i’th conjugacy class, and we use
(T )
χi to denote the character of these elements. What this means is that we
can think of the characters as vectors in a c-dimensional complex vector space.
Note that we have reduced the dimension of the vector space from |G| to c,
the number of conjugacy classes in G.
What we will now see is that the simple characters (characters of irreducible
representations) are in fact orthogonal vectors. In particular, inequivalent
irreducible representations must have different (in fact orthogonal) characters.
Theorem 2.8.1 Orthogonality of characters. The simple characters of
a finite group G (the characters of its irreducible unitary representations) are
orthogonal:
X ∗ c ∗
(T ) (S)
X
χ(T ) (g)χ(S) (g) = ni χi χi = |G|δT S ,
g∈G i=1
where δT S is zero if T and S are inequivalent, and one if they are equivalent.
Here ni denotes the number of elements in the i’th conjugacy class.
Proof. This is a direct consequence of the great orthogonality theorem Theo-
rem 2.6.1. Using the notation δT S , we can state Theorem 2.6.1 as
X |G|
T † (g)ij S(g)kl = δil δjk δT S .
d
g∈G
We look at the special case with i = j, k = l, and we sum over i and k. We

get:
XX X |G|
T † (g)ii S(g)kk = δik δik δT S .
d
i,k g∈G i,k
(T ) ∗

(g)χ(S) (g). For the right-hand-side,
P
The left-hand-side is simply g∈G χ
we notice that
X X Xd
δik δik = δik = 1 = d.
i,k i,k i=1
Therefore X ∗
χ(T ) (g)χ(S) (g) = |G|δT S .
g∈G

We have shown that the characters are orthogonal vectors in the vector
space of class functions, which has dimensions c, the number of conjugacy
classes. It thus follows immediately that:
Corollary 2.8.2 Upper bound on the number of irreps. The number ρ
of inequivalent irreducible representations of a finite group is less or equal than
the number c of conjugacy classes.
2.8.2 Second orthogonality theorem

So far we looked at the character χ(T ) : G → C of an irreducible representation
T as a vector in the c-dimensional space of class functions. Now we want to
look at a different type of orthogonality. Let ρ be the number of irreducible
representations of a finite group G, which we know is finite and bounded by c,
and label the irreducible representations as T (α) . We denote their characters
(α)
by χ(α) . Now, for each conjugacy class i ∈ {1, . . . , c}, we can think of the χi
as being the components of a vector in the ρ-dimensional vector space of irre-

ducible representations. We can also show that the characters are orthogonal
from this point of view.
(α)
Theorem 2.8.3 Orthogonality of characters II. Let χi be the character
of the irreducible representation T (α) of a finite group G, in the i’th conjugacy
class. Then:
ρ ∗
X (α) (α) |G|
χi χj = δij .
α=1
ni
That is, the characters are orthogonal in the ρ-dimensional vector space of
irreducible representations. Here ni denotes the number of elements in the i’th
conjugacy class.
We will skip the proof of this orthogonality theorem.
2.8.3 Some important consequences

A direct consequence of this theorem is that it completely fixes the number of
irreducible representations of a finite group G:
Theorem 2.8.4 The number of irreps of a finite group. The number
ρ of inequivalent irreducible representations of a finite group is equal to the
number c of conjugacy classes of the group.
Proof. We have already seen that ρ ≤ c. Since the characters are orthogonal
in the vector space of irreducible representations, we also get that c ≤ ρ. Thus
ρ = c.
Remark 2.8.5 It follows from this statement that simple characters form an
orthogonal basis on the space of class functions. Thus any class function can
be written as a linear combination of simple characters.
Using orthogonality of characters we can also determine the possible di-
mensions of irreducible representations. Perhaps surprisingly, not only is the
number of irreducible representations fixed, but their dimensions is also highly
constrained. In other words, for a group of a certain size, its irreducible repre-
sentations cannot be too large.
Theorem 2.8.6 Dimensions of irreducible representations. Let dα be
the dimension of the irreducible representation T (α) of a finite group G. Then
c
X
d2α = |G|,
α=1
where the sum is over all inequivalent irreducible representations of G.

Proof. This follows from the second orthogonality relation Theorem 2.8.3. Set
i = j = 1, where the first conjugacy class is identified with the conjugacy class
which contains only the identity element. We get:
c c
(α) (α)
X X
(χ1 )∗ χ1 = d2α = |G|.
α=1 α=1

Using this result, we can prove the converse of Theorem 2.5.3, which stated
that all irreducible representations of finite abelian groups are one-dimensional.
Theorem 2.8.7 Irreducible representations of abelian groups. A
finite group is abelian if and only if all its irreducible representations are one-
dimensional.
Proof. We already know that the irreducible representations of a finite abelian

group are one-dimensional. We need to show that if a finite group only has
one-dimensional irreducible representations, then it is abelian.
Suppose that all irreducible representations are one-dimensional. Then
from Theorem 2.8.6 we get that
c
X
1 = |G|,
α=1
which means that the number of inequivalent irreducible representations c,

which is equal to the number of conjugacy classes in G, must be equal to the
order of the group |G|. In other words, all conjugacy classes of G must contain
only one element, which implies that the group is abelian.
2.8.4 Character table

The character table of a finite group is a common way of summarizing the
characters of the irreducible representations of a finite group G. It is very useful
in physics, for instance to determine the direct sum decomposition of reducible
representations using the decomposition theorem (see Theorem 2.9.2). In fact
some textbook on group theory for physics compile character tables for various
groups.
The rows of a character table are labeled by the inequivalent irreducible
representations of G, while the columns are labeled by the conjugacy classes of
G. Since the number of inequivalent irreducible representations is equal to the
number of conjugacy classes, the character table is always square. Each entry
in the table gives the character of the given irreducible representation in the
corresponding conjugacy class. Orthogonality relations between characters are
very useful to fill in the character table of a given group, even without knowing
all details of its irreducible representations. We will see how this goes in the
example of S3 shortly (see Section 2.10).
2.9 Reducibility and decomposition

Objectives
• Use the decomposition theorem to calculate the direct sum decomposition
of a reducible representation.
• Use the character of a representation to determine whether it is reducible
or not.
In the previous section we used characters to characterize inequivalent ir-
reducible representations of finite groups. We found that the number of in-
equivalent irreducible representations of G is equal to the number of conjugacy
classes in G, and that the dimensions of the irreducible representations are con-
strained by the order G. We saw that the orthogonality relations for characters
can be neatly encoded in terms of a character table.
In this section we show how characters can also be used to study whether a
given representation is reducible or not: the so-called “reducibility criterion”.
In the case of a reducible (semisimple) representation, we also study how char-
acters can be used to find the decomposition of the representation as a direct
sum of irreducible representations: the so-called “decomposition theorem”.
2.9.1 A criterion for reducibility

Using orthogonality of characters we can find a simple criterion to determine
whether a representation is reducible or not.
Let T : G → GL(V ) be a semisimple representation. It can be decomposed
as a sum of irreducible representations
c
M
T = mα T (α) ,
α=1
where mα is an integer that denotes the number of times that T (α) appears in
the decomposition. Taking the trace, we get a similar relation for the charac-
ters:
Xc
(T )
χ (g) = mα χ(α) (g).
α=1
Therefore, the character of a semisimple representation is a linear combination
of simple characters with non-negative coefficients. (Any class function can be
written as a linear combination of simple characters, but what is special here
is that the coefficients are non-negative.)
A semisimple representation will then be irreducible if and only if only
one of the coefficients mα is non-zero and equal to one. Our goal is to find
a simple criterion in terms of the characters of a representation to determine
when this is the case, without first having to calculate the explicit direct sum
decomposition of the representation.
Theorem 2.9.1 Criterion for reducibility. A representation T of a finite
group G is irreducible if and only if
c
(T )
X
ni |χi |2 = |G|,
i=1
where ni is the number of elements in the i’th conjugacy class. It it is reducible,

then
c
(T )
X
ni |χi |2 > |G|.
i=1
Proof. As explained above, the characters of a semisimple representation are a
linear combination of simple characters with non-negative coefficients:
k
(T ) (α)
X
χi = m α χi .
α=1
The representation is irreducible if and only if only one of the coefficients mα

is non-zero and equal to one.
Taking the complex conjugate we also have
k
(T ) (α)
X
(χi )∗ = mα (χi )∗ .
α=1
We take the product of these two equations, multiply by ni (the number of

elements in the i’th conjugacy class), and sum over i. We get:
c k X k c
!
(T ) (T ) ∗ (α) (β) ∗
X X X
ni χi (χi ) = mβ mα ni χi (χi )
i=1 α=1 β=1 i=1
k
X
=|G| m2α ,
α=1
where we used the first orthogonality theorem Theorem 2.8.1. In particular, T

is irreducible if and only if all mα are zero except one that is equal to one, so
we get that it is irreducible if and only if
c
(T )
X
ni |χi |2 = |G|.
i=1
It thus follows that it is reducible if and only if

c
(T )
X
ni |χi |2 > |G|.
i=1
2.9.2 The decomposition theorem

We now introduce the decomposition theorem, which gives us a way of de-
termining the irreducible representations that appear in the direct sum de-
composition of a semisimple representation. The decomposition theorem uses
orthogonality of characters to determine the coefficients mα in the decompo-
sition.
Theorem 2.9.2 The decomposition theorem. Given a semisimple rep-
(T )
resentation T of a finite group G with characters χi , where i indexes the
conjugacy classes of G, the coefficients mα in the direct sum decomposition
c
M
T = mα T (α)
α=1
are given by
c
1 X (T ) (α)
mα = ni χi (χi )∗ ,
|G| i=1
(α)
where ni is the number of elements in the i’th conjugacy class, and χi are
the characters of the irreducible representation T (α) .
Proof. We know that
c
(T ) (β)
X
χi = m β χi .
β=1
(α)
We multiply by ni (χi )∗ and sum over i. We get:
c c c
!
(T ) (α) (β) (α)
X X X
ni χi (χi )∗ = mβ ni χi (χi )∗
i=1 β=1 i=1
=|G|mα ,
where we used the first orthogonality theorem Theorem 2.8.1. Therefore

c
1 X (T ) (α)
nα = ni χi (χi )∗ .
|G| i=1

The decomposition theorem is useful to determine the direct sum decom-
position of reducible representations. If one knows the characters of all irre-
ducible representations of a given group, then calculating the coefficients mα
of the decomposition of a reducible representation becomes a simple algebraic

calculation.
2.10 An example: S3
Objectives
• Apply the orthogonality theorems to calculate the character table of a
finite group.
• Apply the reducibility criterion to find whether a representation is re-

ducible.
• Apply the decomposition theorem to calculate the direct sum decompo-
sition of a reducible representation.
Let us now study representations of our favourite symmetric group S3 using
all that we have seen about characters. This should make the various theorems
and constructions more explicit.
2.10.1 The character table

First, recall that S3 is the group of permutations of three objects. It has
six elements. They are split into three conjugacy classes, corresponding to the
three possible cycle structures. There is one permutation with three one-cycles
(the identity permutation); 3 permutations with one two-cycle and one one-
cycle (the transpositions); and 2 permutations with a three-cycle (the cyclic
permutations of length 3). Let us denote the corresponding conjugacy classes
as C1 , C2 and C3 , with number of elements respectively n1 = 1, n2 = 3 and
n3 = 2.
Since the number of inequivalent irreducible representations is equal to
the number of conjugacy classes (Theorem 2.8.4), we know that S3 has three
inequivalent irreducible representations. Let us denote those by T (α) with
α = 1, 2, 3. We have already found two of them: the identity representation
(T (1) ) and the parity representation (T (2) ), which are both one-dimensional.
We do not know yet the third one.
From Theorem 2.8.6, we know that the dimensions of the three irreducible
representations must satisfy the equation
3
X
d2α = |S3 | = 6.
α=1
The only possibility is that two of the representations are one-dimensional (we
know them already), and the third one is two-dimensional.
Even if we do not know the third two-dimensional irreducible representation
explicitly, we can still construct the character table of S3 using orthogonality
of characters. The character table should look like:
Table 2.10.1 Constructing the character table for S3
C1 C2 C3
(1)
T 1 1 1
T (2) 1 −1 1
T (3) 2 a b
We have already filled many entries, and left two unknowns (a and b),
which we will determine shortly. The first column is given by the trace of
the matrices corresponding to the identity element of the group, which are just
identity matrices. So these entries are just the dimension of the representations.
This is true in general: the first column is always given by the dimensions of
the representations.
The first row corresponds to the identity representation, which assigns 1 to
all group elements. So the characters are always 1.
The second row corresponds to the parity representation, which assigns 1
to even permutations and -1 to odd permutations. The permutations in the
conjugacy classes C1 and C3 are even, while those in C2 are odd, so we can fill
the second row appropriately.
There are two entries left, a and b. To find them, we use orthogonality of
characters. We will use orthogonality of columns, which is just the statement
of the second orthogonality relation Theorem 2.8.3. This relation says that
3
X (α) (α) |G|
(χi )∗ χj = δij .
α=1
ni
Applying this to the first and second conjugacy classes (the first and second
columns), that is, setting i = 1 and j = 2, the right-hand-side becomes zero,
and we get:
(1)(1) + (1)(−1) + (2)(a) = 0.
Therefore a = 0. Similarly, using the first and third columns (i = 1 and j = 3),
we get:
(1)(1) + (1)(1) + (2)(b) = 0,
that is b = −1. We have thus completed the character table of S3 :
Table 2.10.2 The character table for S3
C1 C2 C3
T (1) 1 1 1
T (2) 1 −1 1
T (3) 2 0 −1
For fun we can check that the other orthogonality relations are also satisfied.
For instance, sticking with the second orthogonality relation Theorem 2.8.3,
but applying it to the third column (i = j = 3), we get:
|G| 6
(1)(1) + (1)(1) + (−1)(−1) = = = 3,
2 2
which is indeed true.
The first orthogonality relation Theorem 2.8.1 correspond to orthogonality
between rows. Recall that it says that:
3
(α) (β)
X
ni (χi )∗ χi = |G|δαβ .
i=1
For instance, using it for the second and third irreducible representations (α =
2 and β = 3), we get:
1(1)(2) + 3(1)(0) + 2(1)(−1) = 0,
which is indeed correct. Applying to the third row (α = β = 3), we get:
1(2)(2) + 3(0)(0) + 2(−1)(−1) = |G| = 6,
which is again correct. Orthogonality relations are indeed satisfied!

We can also check that the reducibility criterion Theorem 2.9.1 is sat-
isfied for the irreducible representations. For instance, looking at the two-
dimensional irreducible representation T (3) , we get:
3
(3)
X
ni |χi |2 = 1(2)2 + 3(0)2 + 2(−1)2 = 6,
i=1
which is indeed equal to |G| = 6, therefore the representation is irreducible,

as we know. In fact, we know this two-dimensional irreducible representation
of S 3 very well: it is the two-dimensional representation constructed from
symmetries of the equilateral triangle in Example 2.1.6. Indeed, one can check
that the characters of this representation match with the third row of the
character table for S 3 .
2.10.2 The decomposition theorem

Let us now study the reducibility criterion and decomposition theorem. Let
us consider the three-dimensional permutation representation T of S3 from
Example 2.1.4. Writing down the 3×3 matrices explicitly, it is easy to calculate
the characters by taking the traces. The trace of the identity matrix is 3; the
trace of the three matrices corresponding to the transpositions is 1; and the
trace of the two matrices corresponding to the cyclic permutations are 0. Thus,
the characters of T are:
(T ) (T ) (T )
χ1 = 3, χ2 = 1, χ3 = 0.
Let us first show that this representation is reducible (we already know that
since S3 has only three irreducible representations of dimensions 1, 1 and 2.)
We use the reducibility criterion Theorem 2.9.1. We calculate:
3
(T )
X
ni |χi |2 = 1(3)2 + 3(1)2 + 2(0)2 = 12,
i=1
which is certainly greater than |G| = 6. Thus the representation is reducible,

as expected.
We now want to find its direct sum decomposition. Recall the decomposi-
tion theorem Theorem 2.9.2. If we write the direct sum decomposition as
c
M
T = mα T (α) ,
α=1
then the coefficients are given by

c
1 X (T ) (α)
mα = ni χi (χi )∗ .
|G| i=1
Let us calculate these coefficients using the character table Table 2.10.2. We
get:
1
m1 = (1(3)(1) + 3(1)(1) + 2(0)(1)) = 1,
6
1
m2 = (1(3)(1) + 3(1)(−1) + 2(0)(1)) = 0,
6
1
m3 = (1(3)(2) + 3(1)(0) + 2(0)(−1)) = 1.
6
We thus obtain the direct sum decomposition:
T = T (1) ⊕ T (3) .
What this means is that the three-dimensional permutation representation of

S3 is equivalent to the direct sum of the trivial one-dimensional representation
T (1) , and the two-dimensional irreducible representation T (3) . This is some-
thing that you could show explicitly by finding a similarity transformation
that brings all six 3 × 3 matrices in block diagonal form. But you see how
much faster this was using characters! No need to find similarity transfor-
mations, only simple algebraic calculations are needed! Characters are really
awesome!
2.11 An example: the regular representation

Objectives
• Find the characters and the direct sum decomposition of the regular
representation of any finite group.
As a last example in this section, let us study the regular representation
for any finite group, which was introduced in Example 2.1.5. Recall that the
regular representation is defined using Cayley’s theorem. We think of a finite
group G of order n as a subgroup of Sn , and keep only the matrices of the n-
dimensional permutation representation of Sn that correspond to the elements
of the subgroup to obtain the regular representation of G. Thus the regular
representation is a n-dimensional representation.
It turns that for any finite group G, the characters of the regular representa-
tion are very easy to compute. First, the matrix corresponding to the identity
element is the identity matrix, thus its trace is n, the dimension of the repre-
sentation. What is the trace of the matrices corresponding to the other group
elements in G? Recall from Cayley’s theorem that one defines the mapping
G → Sn by looking at the permutations of the group elements corresponding
to the rows of the mulitiplication table. But by the rearrangement theorem
Theorem 1.4.2, we know that each row must permute every single element (no
element can be left fixed, since they cannot appear twice in the same column in
the multiplication table). Thus, the corresponding n-dimensional permutation
matrices will necessarily only have zeroes on the diagonal (otherwise it would
leave some elements fixed). Their traces are then necessarily zero!
Therefore we conclude that for any finite group G of order n, its n-dimensional
(R)
regular representation R will have characters χ1 = n for the conjugacy class
(R)
containing the identity element and χi = 0 for all other conjugacy classes
i > 1.
With this we can show that for all finite groups with n ≥ 2 the regular
representation is reducible. Indeed, by the reducibility criterion Theorem 2.9.1,
we get:
c
(R)
X
ni |χi |2 = 1 · n2 + 0 = n2 ,
i=1
which is certainly > |G| = n for all n ≥ 2.

We can also find its decomposition into a direct sum of irreducible repre-
sentations for all finite groups! Recall from the decomposition theorem Theo-
rem 2.9.2 that, if we write the decomposition as

c
M
(S)
T = mα T (α) ,
α=1
the coefficients are given by

c
1 X (S) (α)
mα = ni χi (χi )∗ .
|G| i=1
In the case of the regular representation of a group of order n, since the only
(S)
non-vanishing character is χ1 = n, for any α we calculate:
1 (α)

(α)
mα = n(χ1 )∗ = (χ1 )∗ .
n
But for any irreducible representation, the character of the identity element is
the trace of the identity matrix, which is just the dimension of the representa-
tion:
(α)
χ1 = dα .
Therefore mα = dα , for all irreducible representations!
Therefore for any finite group the regular representation decomposes as the
direct sum of irreducible representations:
c
M
(S)
T = dα T (α) ,
α=1
whose coefficients are precisely the dimensions of the irreducible representa-

tions! In other words, in the decomposition each irreducible representation
appears exactly the same number of times as its dimension. Isn’t that cool?
For instance, the six-dimensional regular representation of S3 has a direct
sum decomposition:
T (S) = T (1) ⊕ T (2) ⊕ T (3) ⊕ T (3) ,
where as before T (1) is the one-dimensional identity representation, T (2) is

the one-dimensional parity representation, and T (3) is the two-dimensional
irreducible representation of S3 (since the latter is two-dimensional it appears
twice in the decomposition of the regular representation).
Remark 2.11.1 The existence of the regular representation for all finite groups
also provide a simple proof of Theorem 2.8.6 about the dimensions of irre-
ducible representations of finite groups. Indeed, the regular representation is
n-dimensional, where n = |G|, and is equivalent to a block diagonal matrix with
blocks of size dα , where dα is the dimension of the irreducible representations,
and such that each block appears exactly dα times. Thus, simply comparing
the dimension of the matrices, we conclude directly that
c
X
d2α = |G|.
α=1
2.12 Real, pseudoreal and complex representa-

tions
Objectives
• Determine whether a given representation is real, pseudoreal or complex.
In this course we focus on group representations on complex vector spaces.
However, sometimes such a representation may be real, just like complex num-
bers may be real. So a natural question is: given a representation on a complex
vector space, can we determine easily when a representation is real or not?
First, how do we know whether a complex number z is actually real? Well,
it is easy, you look at it, and see whether its imaginary part is zero! More
precisely, the statement is that z is real if and only if z = z ∗ , that is, z is equal
to is complex conjugate, which means that its imaginary part vanishes.
We can do the same thing with matrices. We say that a matrix M is real
if and only if M ∗ . Note that this is really the complex conjugate, not the
complex conjugate transpose. So this means that all the matrix entries are
real numbers.
Now what about representations? When is a representation real? This is of
great interest in physics; for instance, once one understands particles in terms
of representations, then anti-particles transform in the complex conjugate rep-
resentations. So understanding the relation between a representation and is
complex conjugate is crucial.
2.12.1 Real, pseudoreal and complex

Let T : G → GL(V ) be a representation of a group G on a complex vector
space V . Let us first define the complex conjugate representation.
Lemma 2.12.1 Complex conjugate representation. Let T : G → GL(V )
be a representation of a group G on a complex vector space V . Then T ∗ , which
is obtained by taking the complex conjugate matrices T (g)∗ , is also a representa-
tion T ∗ : G → GL(V ), and is called the complex conjugate representation.
(T ∗ ) (T )
Note that its characters are χi = (χi )∗ .
Proof. If T is a group homomorphism, then T (gh) = T (g)T (h) for any g, h ∈ G.
Taking the complex conjugate of the matrices, T ∗ (gh) = T (g)∗ T (h)∗ , and
hence T ∗ is also a group homomorphism.
We would like to say that a representation is real if the entries of the
corresponding matrices are real numbers. Or, more precisely, the matrices
should be equivalent (by a similarity transformation) to matrices with only
real numbers. We would also like to translate this condition into the statement
that a real representation is equivalent to its complex conjugate. But this is
subtle; real representations are indeed equivalent to their complex conjugate,
but the converse is not true. Let us be a little more careful.
Definition 2.12.2 Complex representations. We say that T is complex
if it is not equivalent to its complex conjugate representation T ∗ . We say that it
is non-complex otherwise. That is, T is non-complex if there exists a matrix
S such that for all g ∈ G,
T ∗ (g) = ST (g)S −1 .
♦
(T )
Remark 2.12.3 In terms of characters, it follows that if the characters χi
are complex, i.e. not equal to their complex conjugate, then T is complex.
The converse is not true however; if the characters are real, then we cannot
conclude that T and T ∗ are equivalent.

We now show that we can divide non-complex representations further, when
we focus on irreducible representations. We first need a lemma that tells us
there are only two types of possible similarity transformations between an
irreducible representation and its complex conjugate.
Lemma 2.12.4 Symmetric or anti-symmetric. Let T : G → GL(V ) be
a non-complex unitary irreducible representation of a group G on a complex
vector space V . Thus T is equivalent to T ∗ :
T ∗ (g) = ST (g)S −1 ,
for some matrix S. Then S is either a symmetric matrix or an anti-symmetric

matrix.
Proof. We start with T ∗ (g) = ST (g)S −1 . We take the transpose, to get:
T † (g) = (S −1 )T T (g)T S T .
Since T is unitary, the left-hand-side is equal to T (g)−1 = T (g −1 ). Thus we

get:
T (g −1 ) = (S −1 )T T (g)T S T ,
which shows that T (g) and T (g −1 ) are related. Now this equality is true for
any g ∈ G, so we can write it for g −1 instead of g. We get:
T (g) = (S −1 )T T (g −1 )T S T .
But T (g −1 )T = ST (g)S −1 , and hence, substituting back,
T (g) =(S −1 )T ST (g)S −1 S T

−1
= S −1 S T T (g) S −1 S T .

What this means is that S −1 S T commutes with all matrices T (g) of an irre-
ducible representation. By Schur’s lemma Lemma 2.5.2, we know that S −1 S T =
λI for some λ ∈ C. Thus S T = λS. This means that
S = (S T )T = (λS)T = λS T = λ2 S,
and hence λ2 = 1. It follows that S = ±S, that is, it is either symmetric or

anti-symmetric.
Definition 2.12.5 Real and pseudoreal representations. Let T : G →
GL(V ) be a non-complex unitary irreducible representation of a group G on a
complex vector space V . Thus T is equivalent to T ∗ :
T ∗ (g) = ST (g)S −1 ,
for some matrix S. We say that T is real if S is symmetric, and pseudoreal

if S is anti-symmetric. ♦
Now if a representation is given by matrices T (g) that only have real entries,
then it is certainly real according to our definition, since S can be taken to be
the identity matrix. But is it true that all real representations are equivalent
to representations with matrices that only have real entries? This is what we
show next.
Lemma 2.12.6 Real is real. A real irreducible unitary representation T is
equivalent to a representation whose matrices only have real entries.
Proof. First, we notice that the similarity transformation S can be taken to be

unitary. Indeed, since T ∗ (g) = ST (g)S −1
T
ST (g) = T ∗ (g)S = T (g −1 ) S,
and so S = T (g)T ST (g). We can then calculate:
S † S =T (g)† S † T (g)∗ T (g)T ST (g)

=T (g)† S † ST (g),
since T is unitary. Thus T (g)S † S = S † ST (g). It follows that S † S commutes

with all T (g), and hence by Schur’s lemma Lemma 2.5.2, we know that S † S =
λI for some λ ∈ C.
But the similarity transformation S is defined only up to overall rescaling
(and it is non-zero), so we can choose an appropriate rescaling so that S † S = I,
that is, it is unitary.
Now that we know that S is unitary, we can prove the lemma. We assume
that the representation is real, so that S is a symmetric unitary matrix. One
can show that any symmetric unitary matrix S can be written as S = W 2
for some unitary symmetric matrix W (proof left as exercise). Thus T ∗ (g) =
ST (g)S −1 becomes T ∗ (g) = W 2 T (g)(W −1 )2 . We multiply by W −1 on the left
and W on the right to get:
W −1 T ∗ (g)W = W T (g)W −1 .
Using the fact that W is unitary symmetric, we have W −1 = W † = W ∗ , and

W = (W −1 )∗ . So we can write:
∗
W T (g)W −1 = W −1 T ∗ (g)W = W ∗ T ∗ (g)(W −1 )∗ = W T (g)W −1 .
It thus follows that W T (g)W −1 has only real entries, thus T is equivalent to
a representation with only real entries.
2.12.2 The Frobenius-Schur indicator

We have now defined when an irreducible representation on a complex vector
space is real, pseudoreal, or complex. We understand that a real representation
is equivalent to a representation whose matrices only have real entries. But in
general, the representation may not be given in this form. How do we determine
whether an irreducible representation is real, pseudoreal or complex?
Fortunately there is a neat criterion, known as the Frobenius-Schur indica-
tor, which uses our beloved characters. Here goes!
Theorem 2.12.7 The Frobenius-Schur indicator. Let T be an irreducible
unitary representation of a finite group G, and define the Frobenius-Schur in-
1 2
P
dicator as |G| g∈G χ(g ). Then

1
 if T is real
1 X
2
χ(g ) = −1 if T is pseudoreal,
|G| 
g∈G 
0 if T is complex.
We will leave the proof as an exercise.
Checkpoint 2.12.8 Prove Theorem 2.12.7.
So to determine whether an irreducible representation is real, pseudoreal or
complex, we simply need to evaluate the Frobenius-Schur indicator. We also
note here that if g and h are in the same conjugacy class, than g 2 and h2 also
are; thus the Frobenius-Schur indicator is also a class function, and could be
written as a sum over conjugacy classes.
Remark 2.12.9 The classification between real, pseudoreal and complex irre-
ducible representations also hold for compact groups. Moreover, those can be
distinguished using the Frobenius-Schur indicator as above, but with the sum
replaced by an integral over the compact group.
Example 2.12.10 Irreducible representations of S3 . As an example, let
us look at the irreducible representations of S3 , as studied in Section 2.10. To
evaluate the Frobenius-Schur indicator we will need the square of the group
elements. We note that the square of the identity is still the identity. As for the
transpositions (conjugacy class C2 ), their square is the identity. For the cyclic
permutations (conjugacy class C3 ), we calculate. Take for instance (123). The
square is (123)(123) = (132), and hence it is still a cyclic permutation of length
three. Thus the square of elements in the conjugacy class C3 are still in C3 .
Let us now determine whether the irreducible representations are real, pseu-
doreal or complex. First, the two one-dimensional representations are mani-
festly real, since they map all group elements to real numbers. Let us, for fun,
evaluate the Frobenius-Schur indicator for these two representations.
For the trivial representation, we get:
1 X (1) 2 1
χ (g ) = ((1)(1) + (3)(1) + (2)(1)) = 1,
|S3 | 6
g∈S3
and hence the representation is real, as expected (this is obviously always the
case for the trivial representation).
For the non-trivial one-dimensional representation T (2) (using the notation
in Section 2.10). The sum is:
1 X (2) 2 1
χ (g ) = ((1)(1) + (3)(1) + (2)(1)) = 1,
|S3 | 6
g∈S3
and hence this representation is also real, as expected.

Finally, we look at the two-dimensional representation. We get:
1 X (2) 2 1
χ (g ) = ((1)(2) + (3)(2) + (2)(−1)) = 1,
|S3 | 6
g∈S3
and thus this representation is also real! This is interesting, because depending
on how you think about these 2 × 2 matrices, they may be written as matrices
with complex entries. But in fact in Example 2.1.6 we had already seen one
way to write these matrices with real entries, and thus the representation must
be real indeed.
Example 2.12.11 The two-dimensional representation of the quater-
nion group. Let us now look at the two-dimensional representation of the
quaternion group given in Example 2.1.7. First, we should check that it is ir-
reducible. Using the criterion for irreducibility Theorem 2.9.1, and calculating
the traces of the matrices, we get:
(2)2 + (−2)2 + (0) + (0) + (0) + (0) + (0) + (0) = 8,
which is the order of the quaternion group. Thus it is irreducible. It is the

only irreducible representation of dimension greater than one for the quaternion
group.
Let us determine whether it is real, pseudoreal or complex. First, we note

that for the quaternion group, i2 = (−i)2 = j 2 = (−j)2 = k 2 = (−k)2 = −1,
and of course 12 = (−1)2 = 1. Thus the Frobenius-Schur indicator evaluates
to:
1
((2) + (2) + (−2) + (−2) + (−2) + (−2) + (−2) + (−2)) = −1.
8
We thus conclude that this representation is pseudoreal. That is, it is equiv-
alent to its complex conjugate, but the similarity transformation involves an
anti-symmetric matrix. This is an example where all the characters of the rep-
resentation are real, but the representation is not, it is pseudoreal instead. In
particular, the 2 × 2 matrices cannot be brought into matrices with only real
entries by a similarity transformation.
Example 2.12.12 The cyclic group Zn . As a last example, let us look
at the one-dimensional representation for the cyclic group Z4 given by the
complex numbers {1, −i, −1, i}. We calculate the square of the elements. We
get that 12 = (−1)2 = 1, and (−i)2 = i2 = −1. Thus the Frobenius-Schur
indicator is:
1
(1 + (−1) + 1 + (−1)) = 0,
4
and hence the representation is complex, as expected.
Chapter 3
Applications
In this section we study a few simple physical applications of group and rep-
resentation theory, focusing on finite groups.
3.1 Crystallography
Objectives
• Define what point and space groups are.
• Prove the crystallographic restriction theorem.
Group theory is really all about symmetries. In this section we will see
how group theory can be applied to study objects with symmetries, such as
molecules and crystals. We will see that symmetries are highly constraining,
and that studying symmetries of physical objects such as crystals gives rise to
highly non-trivial, and perhaps unexpected, results.
One should note that there is extensive literature on this subject, in fact
there are whole books dedicated to the study of symmetries of crystals. Physi-
cists and chemists have invented their own notation to denote the symmetry
groups, and reading through this literature can quickly become confusing and
even overwhelming. So what we will do here is just quickly go through some of
the basic results that highlight how group theory is useful to understand the
physics and chemistry of crystals.
3.1.1 Point groups

3.1.1.1 Definition
The starting point of the discussion involves the definition of point groups.
Point groups are groups of geometric symmetries (or isometries) of objects that
fix a point (usually set to be the origin of the coordinate system). The dihedral
groups Dn that we have encountered previously are examples of discrete point
groups in two dimensions.
One can think of points groups in N dimensions as being subgroups of
the orthogonal group O(N ). Roughly speaking, the elements of points groups
correspond to the symmetry operations of rotations, reflections, and “improper
rotations” (i.e. elements with determinant equal to −1). There are infinite
65
CHAPTER 3. APPLICATIONS 66
point groups (as an example, think of the group generated by a rotation in

two dimensions by an irrational number of turns), and finite point groups. In
the following we will mostly focus on finite point groups, which are the most
interesting for crystallography.
Finite point groups can be classified.
• In one dimension, there are two finite points groups: the trivial group,
and the reflection group about the origin, which is isomorphic to Z2 .
• In two dimensions, there are two infinite families of finite point groups:
the cyclic groups of rotations by angle 2π/n, which are isomorphic to
Zn , and the dihedral groups Dn . However, as we will, if we also impose
consistency with the translational symmetries of crystals, only a finite
(and very small) number of those survive, which is the essence of the
crystallographic restriction theorem.
• In three dimensions, there is of course more possibilities. There are 7
infinite familes of finite point groups, and 7 additional point groups.
For more information on those, see for instance the Wikipedia page on
point groups and on point groups in 3D specifically. Again, imposing
consistency with translational symmetries lead to the 32 crystallographic
point groups.
3.1.1.2 Molecular point groups

Three-dimensional point groups are sometimes called molecular point groups,
because they are important in the study of molecules, see for instance the
Wikipedia page on molecular symmetry. For instance, the chemical compound
of Xenon tetrafluoride XeF4 , which consists of four atoms of fluoride placed in
a square planar configuration around an atome of xenon, has point group the
dihedral group D4 of symmetries of the square.
It turns out that representations and characters are crucial in the study
of molecular symmetry. This is because the orbitals of the molecules trans-
form according to irreducible representations of the point group. Thus, to
understand the possible states of the system, it is essential to gain information
about irreducible representations of the point group, which is encapsulated in
its character table. In fact, a lot of the properties of molecules (such as optical
activity, spectroscopy, dipole moments, electronic properties, what not - note
that I have no idea what these terms really mean :-) ) can be analyzed using
irreducible representations of the point group. If you are not convinced, google
orbitals irreducible representations, or something like that. You would be sur-
prised how much group and representation theory appears in quantum chem-
istry! Representations and characters are so important that Wikipedia even
has a list of character tables for chemically important 3D point groups...
3.1.2 Space groups

In the description of crystals, we need more than just point groups. We need
space groups, also called crystallographic groups. Roughly speaking,
space groups are symmetry groups of particular configurations of objects in
space. We do not impose the restriction that a point (the origin) is fixed any-
more. So we allow symmetries such as translational symmetries, which are
crucial in the study of crystals.
In general, one can think of space groups as combinations of point groups
with translational symmetries. More precisely, given a space group, there is an
abelian subgroup of translations. Such a subgroup gives rise to the structure
of a lattice of points that are invariant under these translations. For particular
such lattices, there are sometimes extra symmetries (such as rotations, reflec-
tions, etc.), which, together with translations, form the structure of the space
group.
Just as for point groups, space groups can be classified.
• In one dimension, there are two space groups (also known as “line groups”):
the group of translations, which is isomorphic to the group of integers Z,
and the infinite dihedral group, which includes reflections.
• In two dimensions, there are 17 space groups, also known as wallpaper
groups (have a look at this page, it’s super fun!). Those are obtained by
combining the 5 types of two-dimensional lattices (known as “Bravais lat-
tices”) specifying the translational symmetries with the two-dimensional
point groups that are consistent with translational symmetries.
• In three dimensions, there is a total of 230 different space groups. Those
are obtained by combining the 14 types of three-dimensional Bravais
lattices with the 32 point groups that are consistent with translational
symmetries.
3.1.3 Crystals
We are now ready to discuss symmetry groups of crystals, and the famous
crystallographic restriction theorem, which shows how powerful the study
of symmetries can be in physics and chemistry.
To start with: what is a crystal? In physics and chemistry, a crystal is a
solid material which has a highly ordered microscopic structure: it consists in
a lattice of atoms that extends in all directions and is invariant under transla-
tions. Examples include snowflakes, diamonds, table salt, etc.
Let us focus on crystals in three dimensions. Mathematically, the lattice
of atoms in a crystal is a Bravais lattice: this is an infinite array of discrete
points that are generated by a translation:
T~ = n1 v~1 + n2 v~2 + n3 v~3 ,
where n1 , n2 , n3 ∈ Z and v~1 , v~2 , v~3 are primitive vectors in three dimensions
which lie in different directions (not necessarily perpendicular) and span the
lattice. Thus a crystal is invariant under translations generated by T~ .
Furthermore, a given crystal may be invariant under symmetry operations
that leave a point fixed, such as rotations and reflections. The group of such
symmetry operations is a point group; in this context, it is known as a crys-
tallographic point group. Together with translations, it forms the space group
of the crystal.
We have seen that finite point groups in three dimensions can be classified:
there are 7 infinite familes, and 7 additional point groups. But can all of those
be point groups of crystals? In other words, can all of these point groups arise
as symmetry groups of lattices? The answer to this question is the essence
of the crystallographic restriction theorem, and may appear surprising: only
a very small, finite, number of point groups are symmetry groups of crystals!
For instance, there is no crystal that is invariant under rotations by an angle
2π/5! Isn’t that surprising?
Theorem 3.1.1 Crystallographic restriction theorem. Consider a crys-
tal in three dimensions that is invariant under rotations by an angle 2π/n
around an axis. Then the only allowed possibilities are n = 1, 2, 3, 4, 6, with
the case n = 1 being the trivial case with no rotational symmetry.
Note that the same result holds for a crystal in two dimensions invariant
under rotations.
What? Really? Those are the only possibilities? Imposing translational
symmetry is that constraining? Let’s prove this! There are a number of differ-
ent proofs of this theorem. We will first present a geometric proof, and then
present a representation theoretic, or matrix based, proof, for fun.
Geometric proof. First, we forget about the case n = 1 since it is trivial, as it
involves no rotational symmetry.
Now suppose that a lattice is invariant under rotations by an angle 2π/n
about an axis. Consider two lattice points A and B in the plane perpendicular
to the axis of rotation, and separated by a translation vector ~v of minimal
length (by which I mean that there is no lattice point on the line between A
and B). Let us rotate point A about point B by an angle 2π/n and call the
resulting point A0 . Similarly, let us rotate point B about A by the same angle
2π/n but in the opposite direction, and call the resulting point B 0 .
Let us call v~0 the vector joining A0 and B 0 . We notice that v~0 is parallel to
~v . Thus, if rotation by an angle 2π/n is a symmetry of the lattice, it follows
that the new translation vector v~0 must be an integer multiple of ~v , that is,
v~0 = k~v for some k ∈ Z.
This is because the two points A0 and B 0 must be related by a translation

generated by ~v , and ~v was chosen to have minimal length.
Now we can relate the length of v~0 to the length of ~v and the angle 2π/n.
The four points A, B, A0 , B 0 form a trapezium, with angles 2π/n and π − 2π/n.
By basic geometry, we get:

2π 2π
|v~0 | = |~v | + 2|~v | cos π − = |~v | 1 − 2 cos .
n n
Since we must have v~0 = k~v for some k ∈ Z, and hence |v~0 | = |k||vecv|, we get
the constraint:
2π
|k| = 1 − 2 cos ,
n
that is, 1 − 2 cos 2π
n must be a non-negative integer. Since | cos α| ≥ 1 for any
angle α, the only possible values of |k| are |k| = 0, 1, 2, 3. Solving for n for each
cases, we get, respectively, n = 6, 4, 3, 2. Those are the only possible rotational
symmetries of a crystal, which concludes the proof.
Representation theoretic proof. Let us now give a representation theoretic, or
matrix based, proof. A rotation by an angle 2π/n about an axis can be seen
as a linear operator acting on the two-dimensional vector space perpendicular
to the axis of rotation. After choosing a basis for this vector space, such a
rotation can be written in terms of a 2 × 2 rotation matrix. In the language of
group theory, this gives a two-dimensional representation of the cyclic rotation
group Zn . On general grounds, we know how to write such a rotation matrix
with respect to an orthonormal basis for R2 :
cos 2π sin 2π

n n .
− sin 2π
n cos 2π
n
This matrix expression for the linear operator is only valid with respect to an
orthonormal basis, but its trace (that is, the character of the representation),
which is equal to 2 cos(2π/n), is invariant under changes of basis, as we know
(that is, the characters are the same for equivalent representations related by
similarity transformations). So in a different choice of basis, the 2 × 2 rotation
matrix may look different, but its trace will always be equal to 2 cos(2π/n).
Now, in a lattice, it is appropriate to write down how rotations act with

respect to the basis given by the primitive translation vectors v~i . Those are in
general not orthogonal, nor of length one, so the precise matrix representation
may look ugly. However, if a rotation by an angle 2π/n is a symmetry of the
lattice, then we know that it must map each translation vector into an integer
linear combination of the other translation vectors. In other words, with respect
to this lattice basis, the rotation matrix corresponding to rotation by an angle
2π/n must have integer entries. In particular, its trace must be integer. But
since we know that the trace is invariant under changes of basis, its trace must
also be equal to 2 cos(2π/n). Therefore we conclude that 2 cos(2π/n) = k for
some k ∈ Z. The only possibilities are k = −2, −1, 0, 1, 2, which correspond to,
respectively, n = 2, 3, 4, 6, 1, with the last choice corresponding to the identity
matrix (i.e. no rotational symmetry). These are the only possible rotational
symmetries that are consistent with translational symmetries of a crystal.
This is a highly non-trivial result, and is perhaps unexpected. Why can’t
we build a lattice that is invariant under five-fold rotations? Or seven-fold
rotations? This is related to the fact that we cannot combine objects with a
five-fold or seven-fold apparent symmetry in such a way that they completely
fill the plane. See for instance this webpage for pretty pictures.
Using the crystallographic restriction theorem, we can enumerate the pos-
sible crystallographic point groups in low dimensions.
• In two dimensions, we had two infinite families of point groups: the cyclic
rotations groups (Zn ) and the dihedral groups (Dn ). Since crystals can
only be invariant under rotations by an angle 2π/n with n ∈ {1, 2, 3, 4, 6},
we see that the only crystallographic point groups are the five cyclic
rotation groups Zn and five dihedral groups Dn with n ∈ {1, 2, 3, 4, 6},
for a total of 10 crystallographic point groups in two dimensions.
• In three dimensions, we started with 7 infinite families and 7 additional
point groups. Looking at the rotations that are elements of these point
groups, one can conclude that there are only 32 crystallographic point
groups in three dimensions.
To summarize, the possible symmetries of crystals are highly constrained! This
is neat, and is a nice, direct, application of group theory to physics and chem-
istry. Crystallography is all about group and representation theory!
To end this section, note that the crystallographic restriction theorem is
not the end of the story. The story continues with aperiodic tilings and qua-
sicrystals. It is truly fascinating (and has led to a Nobel prize). Have a look
at these Wikipedia pages and further resources if you are interested!
3.2 Quantum Mechanics

Objectives
• Explain why the states of a quantum mechanical system can be labeled
using irreducible representations of the symmetry group, and how the
degeneracy of states is computed by looking at the dimensions of the
• Use representation theory to determine the form of the states in the case
of a cyclic periodic potential (Bloch’s theorem).
In this section we give a very brief introduction to the power of group and
representation theory in quantum mechanics.
3.2.1 Schrodinger equation

The starting point of quantum mechanics is Schrodinger equation. The state
of a physical system in quantum mechanics is determined by a wave-function
Ψ, which is a solution of a differential equation known as the Schrodinger
equation of the system:
∂
i~ Ψ = HΨ, (3.2.1)
∂t
where H is the Hamiltonian, which is a linear differential operator. It can
be seen as a quantization of the Hamiltonian of classical mechanics, as we will
see later on.
Remark 3.2.1 There is something important here that is not often mentioned:
quantum mechanics is in many ways easier than classical mechanics. That is
because in quantum mechanics, we are solving a linear differential equation to
determine the wave-function Ψ. So the superposition principle applies; given
two solutions Ψ1 and Ψ2 of the Schrodinger equation, any linear combination of
those is also a solution. So we can really think of the vector space of solutions
(the Hilbert space of states), and think of the Hamiltonian as an operator on
that space (if it is finite-dimensional, we can choose a basis and represent it as
a matrix).
On the contrary, in classical mechanics, to solve a problem one needs to
solve the equations of motion for, say, the position function x(t). Newton’s law
can be rewritten as
d2
m 2 ~x(t) = −∇V (~x).
dt
These equations of motion for ~x(t)are generally non-linear, except for the case
of harmonic motion, i.e. masses connected by springs, in which case ∇V (~x)
is linear in ~x. So in general, the superposition principle does not apply in
classical mechanics, and linear combinations of solutions are not solutions of
the equations of motion.
This is why group and representation theory are much more useful and
powerful in quantum mechanics than in classical mechanics.
To solve Schrodinger equation we do a separation of variables. We write
Ψ = ψe−iEt/~ , where ψ does not depend on time. Substituting back in (3.2.1),
we get:
Hψ = Eψ,
which is an eigenvalue problem for the Hamiltonian of the system. E here is
the energy of the system. This is the key equation that we need to solve.
3.2.2 The symmetry group of the Hamiltonian

The power of group and representation theory is that we will be allowed to
characterize the form of the space of solutions for a given quantum mechanical
problem without ever solving the eigenvalue problem. All that we will use are
the symmetry properties of the system.
Suppose that R is another operator acting on the space of states. We know
how an operator transforms: it transforms by a similar transformation. Thus,
the Hamiltonian transforms as
H 0 = R−1 HR.
We say that the Hamiltonian is invariant under R if H 0 = H, that is,
H = R−1 HR,
or, equivalently,
RH = HR.
We can look at all such operators R that commute with H: those form a group
G, which we call the symmetry group of the Hamiltonian.
More precisely, we could start with the symmetry group G, realized in terms
of properties of the Hamiltonian (for instance some explicit symmetry of the
system, such as periodicity). Then the operators R would form a representation
of the group G, acting on the space of states of the system.
3.2.3 Irreducible representations and degeneracy of states

Now, the key is to realize that two eigenfunctions that are mapped into each
other by a symmetry transformation R must have the same energy eigenvalue
E. Indeed, if Hψ = Eψ, then
H(Rψ) = R(Hψ) = R(Eψ) = E(Rψ).
So what this means is that we can group the eigenfunctions of the systems
into subspaces that are invariant under the action of the symmetry group; all
eigenfunctions in each of these subspaces share the same eigenvalue.
This statement follows neatly from representation theory. If R is an irre-
ducible presentation, then, since HR = RH, by Schur’s lemma we must have
H = EI for some number E. What this means is that all states have the
same energy eivenvalue. However, in general R is not irreducible. But one can
L wuch that R decomposes as direct sum
choose a basis for the space of states
of irreducible representations R = α R(α) , and by Schur’s lemma, it follows
that with respect to this choice of basis
L the Hamiltonian also decomposes as a
direct sum of diagonal matrices H = α E (α) I, where I is the identity matrix
of the size of the corresponding irreducible representation R(α) .
To summarize, what this means is that we can choose a basis for the space of
states, composed of eigenfunctions of the Hamiltonian, in which the state space
decomposes as a direct sum of subspaces that transform according to irreducible
representations of the symmetry group. All states in each such subspace are
eigenfunctions of the Hamiltonian with the same energy eigenvalue.
This leads to the very important physical concept of degeneracy.
Definition 3.2.2 Degeneracy of states. We say that two eigenfunctions
of the Hamiltonian are degenerate if they share the same energy eigenvalue.
♦
At first, in the history of physics, when eigenstates that shared the same
energy eigenvalues were discovered, it was very puzzling to physicists. What
should states have the same energy eigenvalue? When you diagonalize a generic
matrix, you expect distinct eigenvalues.
But from the viewpoint of the representation theory, degeneracy is ex-
pected, and in fact can be constrained using representation theory. Indeed,
we know that the “degree of degeneracy” of an eigenfunction (that is, the
number of degenerate states) is at least as large as the dimension of the irre-
ducible representation that it belongs to. It can be larger if two eigenspaces
corresponding to different irreducible representations have the same energy
eigenvalue; we call such degeneracy “accidental”, since it is not determined by
symmetry. However, most degeneracies are determined by symmetry.
This provides a beautiful interplay between mathematics and physics. On

the one hand, if we know the group of symmetry of a system, we can determine
the possible degeneracies, simply by enumerating its irreducible representations
and their dimensions. On the other hand, if we observe experimentally that
say 8 states are degenerate (as was observed in the 50s when eight baryons were
observed with approximately the same mass), then we can use this information
to restrict the possible groups of symmetries: assuming that the degeneracy
is not accidental, the group of symmetries must be such that it has an 8-
dimensional irreducible representation.
The upshot of this discussion could be summarized as follows:
• To each energy eigenvalue we can associate a corresponding irreducible
representation of the symmetry group of the Hamiltonian. The degener-
acy of the eigenvalue is the dimension of the irreducible representation
(it could be larger if there is accidental degeneracy).
• In the basis for the space of states given by eigenfunctions of the Hamilto-
nian, the state space decomposes as a direct sum of subspaces that trans-
form according to irreducible representations of the symmetry group.
All eigenfunctions in a given subspace share the same energy eigenvalue.
Thus, we can label the eigenfunctions of the system using the irreducible
representations of the symmetry group, which we call in physics the
quantum numbers.
3.2.4 Examples
We will look at two simple examples of the construction. But most relevant
examples to quantum mechanics involve continuous groups and their represen-
tations, thus for this we will need to wait until later.
Example 3.2.3 Parity. Let us start with a very simple one-dimensional
example. The Hamiltonian of a one-dimensional system takes the form H =
d2
− 12 dx 2 + V (x) for some potential function V (x). Now suppose that V (−x) =
V (x). Then the mapping x 7→ −x leaves the Hamiltonian H invariant. This

symmetry group is the order two group Z2 = {e, a}.
This means that we can label the eigenfunctions of the system according
to irreducible representations of the symmetry group. Here, Z2 has only two
one-dimensional irreducible representations: the trivial representation (given
by T (e) = T (a) = 1), and the parity representation (given by P (e) = 1,
P (a) = −1). Those act on states of the system (eigenfunctions). States that
transform according to the trivial representation satisfy:
ψ(−x) = T (a)ψ(x) = ψ(x),
that is, they are even. States that transform according to the parity represen-
tation transform as:
ψ(−x) = S(a)ψ(x) = −ψ(x),
that is, they are odd. The upshot of the representation theory is that we can
label the states according to whether they are even or odd.
We also know that there is no degeneracy here (barring accidental degen-
eracy), since all irreducible representations of Z2 are one-dimensional.
This example is of course almost trivial: the goal was just to get familiar
with how to use the language of representation theory to understand states of
a quantum mechanical system.
Example 3.2.4 Bloch’s theorem. The second example is more interesting,

and in fact is of fundamental importance in solid state physics. We stick with
a one-dimensional system, but we consider a potential V (x) that is periodic:
V (x + a) = V (x) for some constant a. We consider the mapping T : x → x + a.
This generates a discrete, infinite, abelian group G.
Since G is abelian, it only has one-dimensional irreducible representations.
So as before, there is no degeneracy of states. We can label states using ir-
reducible representations. Those representations are unitary. So for a one-
dimensional irreducible representation Sk we get the transformation:
ψ(x + a) = Sk ψ(x) = eika ψ(x),
for some real number k. Since eika = eika+i2π , the representations are indexed
by a real number k such that
π π
− ≤k< .
a a
So the states of the system can be labeled by such a real number. This is an
example of a Brillouin zone.
We can look at the special case of a one-dimensional crystal (note that it
can be easily generalized to more than one dimension) with a finite number of
sites, say N . We can impose a periodic boundary condition, meaning that if
we translate from the N th site, we go back to the first one. Thus the group of
symmetry is generated by the translation T , but with the condition T n = e.
This group is the cyclic group ZN .
The abelian group ZN has N one-dimensional irreducible representations
Sk . Those are given by representing the generator T by the roots of unity
Sk (T ) = wk , for k = 0, 1, . . . , N − 1, where w = e2πi/N . So we can label
the states according to how they transform. For instance, a state ψk that
transforms in the kth irreducible representation would transform as
ψk (x + a) = Sk (T )ψk (x) = e2πik/N ψk (x). (3.2.2)
Now let us write the state ψk (x) as
ψk (x) = eiφk (x) uk (x), (3.2.3)
where uk (x + a) = uk (x) and φk (x) is some phase factor. Because of (3.2.2),

we must have:
ψk (x + a) = eiφk (x+a) u(x) = e2πik/N ψk (x) = e2πik/N +iφk (x) u(x).
Thus we must have

2πk
φk (x + a) = + φk (x), (3.2.4)
N
and, iterating, φk (x + ma) = 2πkm
N + φk (x). Thus φk is a linear function of m,
and hence of x + ma. Therefore we can write φk (x) = Ax + B for some A and
B. Substituting back in (3.2.4), we get:
2πk
A= .
Na
Thus φk (x) = 2πk
N a x + B. Subsituting back in (3.2.3), and redefining uk (x) to
include in it the arbitrary constant B, we get:
2πikx
ψk (x) = e Na uk (x).
This is known as a Bloch wave: it is a periodic function uk (x) multiplied by

a plane wave. Representation theory tells us that the quantum mechanical
states of a periodic lattice have this form. This is pretty much the statement
of Bloch’s theorem: that a basis of wavefunctions is given by Bloch waves, and
that these wavefunctions are energy eigenstates of the system.
The amazing thing is that this result follows directly from representation
theory, without ever solving the Schrodinger equation! All that we used is
the periodic symmetry of the lattice. The particular form of the potential
would then be needed to solve for the periodic functions uk (x) and identity the
particular energy eigenvalues corresponding to the eigenstates of the system.

3.3 Coupled harmonic oscillators

Objectives
• Determine the form of the normal modes of a system of masses tied by
springs with particular symmetries using representation theory.
In the previous section we argued that quantum mechanics was somewhat
easier than classical mechanics, because Schrodinger equation is linear, and
hence representation theory naturally plays a role. But there is one particular
context in classical mechanics where the equation of motion is linear: it is the
case of a system of coupled harmonic oscillators. As Michael Peskin, a famous
physicist, said: “Physics is that subset of human experience which can be
reduced to coupled harmonic oscillators.” The dynamics of systems of coupled
harmonic oscillators is crucial in physics, and in fact forms the basis of much
of quantum mechanics and quantum field theory.
In classical mechanics, one can think of a system of coupled harmonic oscil-
lators as a system of masses connected by springs. But in fact what we will do
in this section applies more generally to study oscillations of any physical sys-
tem about a stable equilibrium point, as you may have studied in your classical
mechanics course.
Going back to the spring-mass system, the system is governed by Hooke’s
law, which says that the force on the mass by the spring is proportional to its
displacement from equilibrium. For instance, if we consider the very simple
system of a mass moving in one dimension, and attached by a spring to a fixed
wall, then the equation of motion would be
d2
m x = −kx,
dt2
where m is the mass, k is Hooke’s constant, and x is the displacement of the
mass from its equilibrium position. The goal of classical mechanics is to solve
the equation of motion to find the displacement x as a function of time. In this
case, we see that the equation of motion is linear. We knowp that the general
solution will take the form x(t) = C cos(ωt + δ), with ω = k/m.
For a more general system of N masses moving in D dimensions, each mass
can move in D independent directions, and hence the displacement of the whole
system is given by N D-dimensional displacement vectors. All-in-all, this can
be packaged in a N D-dimensional real vector, which we call ~η (t). The equation
of motions will look like (in vector form):
d2
~η = −H~η , (3.3.1)
dt2
for some N D × N D matrix of coefficients H. The key point here is that the
equation is still linear. But given a particular spring-mass system, the precise
form of the matrix H may be mess, and in fact not so easy to write down. The
goal of this section will be to extract a lot of information about the motion of
the system without ever having to write down H explicitly. As you could guess,
what we will use is simply the symmetries of the system, and representation
theory.
3.3.1 Normal modes

To analyze the behaviour of spring-mass systems in more than one dimension,
one needs to understand the concept of normal modes of the system.
Definition 3.3.1 Normal modes. We say that a system is in a normal
mode if all its parts are oscillating with the same frequency. ♦
Let us go back to (3.3.1). To solve this equation, we will assume that
normal modes exist, and see what they look like. We take the ansatz
~η (t) = ~a cos(ωt + δ),
for some positive constant ω and a vector of coefficients ~a (for the case ω = 0,
we would take the ansatz ~η (t) = ~at + ~b). Plugging back in (3.3.1), we obtain
the equation
H~a = ω 2~a,
which is nothing but an eigenvalue problem for the N D × N D matrix of coef-
ficients H!
In other words, what we have found is that the normal modes are given
by eigenvectors of H, with oscillating frequency given by the square root of
their eigenvalues. Because the equations of motion are linear, we can conclude
that the general motion of the system will be given by a linear combination
(superposition) of these N D normal modes.
A few things to note: for the motion to be oscillatory, all eigenvalues of H
must be non-negative (otherwise we would have an unstable equilibrium point).
For a normal mode with positive ω, all masses are oscillating with the same
frequency ω. For a normal mode with a zero eigenvalue, the corresponding
motion is uniform translation of all masses.
The upshot of this discussion is that to understand the motion of the sys-
tem, we need to find its normal modes, and to find its normal modes, we need to
find the eigenvectors of the N D × N D matrix H. This is where representation
theory comes into play!
3.3.2 Representation theory to the rescue

Our goal is now to find the eigenvectors and eigenvalues of H. As mentioned
previously, for large systems of masses, H may be a mess. Can we get infor-
mation about the normal modes without writing H explicitly?
The idea is to use symmetries of the system. We consider the group G
of symmetries that leave the system invariant. Concretely, this means that
it leaves H invariant. As before, this means that G acts on the space of
solutions as an N D-dimensional representation that commutes with H. This
representation is in general reducible, but one can choose a basis for the space of
states such that it decomposes as a direct sum of irreducible representations.
By Schur’s lemma, then we know that in this basis, the matrix H becomes
diagonal. Thus the basis vectors are eigenvectors of H. Moreover, the space of
states also decomposes as a direct sum of subspaces that transform according

to the irreducible representations in the decomposition, and any two states in
the same subspace share the same eigenvalue.
So given a particular spring-mass system, to understand the different nor-
mal modes, or eigenstates of the system, one proceeds as follows:
• We find the symmetry group G of the system, and how it acts on the
space of states as an N D-dimensional (reducible) representation (in fact
we only need its characters).
• We use Theorem 2.9.2 to calculate how it decomposes as a direct sum of
• We know that in this basis, the matrix H becomes diagonal, with eigen-
values appearing with degeneracy equal to the dimension of the corre-
sponding irreducible representation. The different eigenvalues give us the
possible different normal modes of the system.
• We use physical insight or further calculations to determine what these
normal modes are.
3.3.3 Examples
In this section we consider two examples. The first one is very simple, and is
studied simply to set the stage; the second one is more involved.
Example 3.3.2 Two masses connected by a spring in one dimension.
We start with the simple example of two masses connected by a spring and
restricted to move in one dimension. We want to understand the normal modes
of the system. In this case we can do everything explicitly, so let us start by
doing that. But the goal of this example is to show how representation theory
can be used to gain information on the system without solving explicitly.
Let η1 (t) and η2 (t) be the displacements of the two masses. The equations
of motion are (setting the constants m1 = m2 = k = 1 for simplicity):
d2 d2
η1 = −(η1 − η2 ), η2 = −(η2 − η1 ).
dt2 dt2
In matrix form, those reads:
d2

1 −1
~
η = − ~η .
dt2 −1 1
We could of course solve this equation explicitly here. But let us use represen-
tation theory instead. To get the normal modes we want to find eigenvectors
for the matrix
1 −1
H= . (3.3.2)
−1 1
We want to do so using symmetries of the system.
The symmetry group G here is rather simple. It has two elements: the
identity, and the exchange of the two masses. Thus G ∼ = Z2 . It acts on the
two-dimensional space of states in a two-dimensional representation T . We
can write down the two-dimensional representation explicitly (we will do that
below for completeness), but to find its decomposition in terms of irreducible
representations we only need its characters. The identity element has char-
acter χ(e) = 2, while the non-trivial element does not leave any mass fixed,
hence it must have character χ(a) = 0. For completeness, the two-dimensional
representation is given by the matrices:

1 0 0 1
T (e) = , T (a) = .
0 1 1 0
Now let us decompose T as a direct sum of irreducible representations. Us-

ing the character table for Z2 , and the decomposition theorem Theorem 2.9.2,
we can find the decomposition of T . If we denote by T (1) the trivial represen-
tation, and T (2) the parity representation (the other irreducible representation
of Z2 ), then we can write T = m1 T (1) ⊕ m2 T (2) . From Theorem 2.9.2, we find
the coefficients to be:
1
m1 = ((2)(1) + (0)(1)) = 1,
2
1
m2 = ((2)(1) + (0)(−1)) = 1.
2
Thus T = T (1) ⊕ T (2) . In this basis the representation is given by diagonal
matrices.
By Schur’s lemma, we know that in this basis H is also diagonal, i.e. the
basis vectors are eigenvectors of H, and hence give normal modes of the system.
The eigenvalues here are non-degenerate since the irreducible representations
are one-dimensional. We do not know from symmetry alone however what
the diagonal entries of H should be; those correspond to the eigenvalues (of
frequencies of the normal modes).
But since here we know the specific form of H (see (3.3.2)), and see that
the trace is 2 and the determinant is 0, the diagonal form of H, must be

0 0
H= .
0 2
So there is one normal mode with zero frequency, and one normal mode with a
non-zero frequency. It is easy to identify these normal modes physically. The
first mode corresponds to linear translation of the whole system (the spring is
not stretched at all). The second mode corresponds to the masses oscillating
with the same frequency in opposite directions (the “breathing mode”).
Of course this example was rather trivial. We could have solved the equa-
tions of motion explicitly. But in more complicated systems, H may be difficult
to write down, and studying symmetries provides valuable information on the
normal modes of the system without having to solve any equation of motion.

Example 3.3.3 The triangular molecule. We now consider the more
interesting example of 3 masses in two dimensions connected by three springs in
the form of an equilateral triangle. Each mass has two displacement coordinates
(horizontal and vertical directions), and hence the displacement vector is 3×2 =
6-dimensional.
Writing down the matrix H for this system would be pretty annoying. So
let us not do that. We will use symmetry instead to gain information on the
normal modes of the system.
The symmetry group here is the dihedral group D3 which is isomorphic
to the symmetric group S3 . It acts on the space of states in the form of a
6-dimensional representation T . We will not write down the representation
explicitly, but simply deduce its characters on the three conjugacy classes of
S3 . We will use the notation in Section 2.10.
In the conjugacy class C1 containing the identity element, the character of
T is χ1 = 6. In the conjugacy class C2 corresponding to the transpositions,

the action should leave one mass fixed, and hence its two displacement coor-
dinates invariant. Therefore, the character must be χ2 = 2. For the class C3
corresponding to the cyclic permutations, nothing is left invariant, and hence
χ3 = 0.
From this we can get the decomposition of T . Using Theorem 2.9.2 and the
character table Table 2.10.2 (with the notation that T (1) is the trivial irrep,
T (2) is the parity one-dimensional irrep, and T (3) is the two-dimensional irrep)
we get the coefficients:
1
m1 = (1(6)(1) + 3(2)(1) + 2(0)(1)) = 2,
6
1
m2 = (1(6)(1) + 3(2)(−1) + 2(0)(1)) = 0,
6
1
m3 = (1(6)(2) + 3(2)(0) + 2(0)(−1)) = 2.
6
Thus T = T (1) ⊕ T (1) ⊕ T (3) ⊕ T (3) . So in this basis the matrix H takes the
diagonal form:
H = diag(ω1 , ω2 , ω3 , ω3 , ω4 , ω4 ),
for some eigenvalues ω1 , ω2 , ω3 , ω4 . Thus there should be four types of normal
modes, two of which would have a two-dimensional eigenspace. At least that’s
what we get from symmetry; there could be additional accidental degeneracies
(i.e. some of the ωi could be identical).
In fact one can identify the zero modes physically. Two of them correspond
to horizontal and vertical translations of the whole system; those have zero
frequency. This corresponds to one of the two-dimensional irreps. There is in
fact an accidental degeneracy here; there is another mode with zero frequency,
corresponding to cyclic rotation of the whole system. This corresponds to one
of the one-dimensional irreps. So there are three zero modes for this system.
We still have one one-dimensonal irrep and one two-dimensional irrep.
There is one “breathing mode”, where all masses oscillate inwards and outwards
simultaneously. This corresponds to the remaining one-dimensional irrep. The
two other normal modes are not so easy to guess, but one could calculate the
eigenvectors from linear algebra, since they should be orthogonal to the four
eigenvectors that we have already determined, and then see how they act on
the three masses. See for instance Section III.2 in Zee’s book.
3.4 The Lagrangian

Objectives
• Use symmetries to specify the form of a Lagrangian, and explain how
symmetries are related to conserved quantities.
Symmetries also play a very important role in determining the physical laws
of a system. In modern physics, we usually formulate classical and quantum
mechanics in terms of either a Lagrangian or Hamiltonian. In this section we
briefly review the Lagrangian formalism, and show how symmetries play an
important role.
3.4.1 The Lagrangian and classical mechanics

In the Lagrangian formulation of classical mechanics, one starts with the La-
grangian L of the system, which is defined as L = T − V , where T is the
kinetic energy of the system and V is its potential energy. We then construct
the action of the system from time t1 to t2 :
Z t2
S= Ldt.
t1
Hamilton’s principle then states that the motion of the system from time
t1 to t2 is such that the action has a stationary value (is extremized). This is
a very profound statement. Out of all possible paths in configuration space,
Nature chooses the path which extremizes the action. The physics is entirely
encoded in its Lagrangian; the equations of motion can be obtained by ex-
tremizing the action. Isn’t that beautiful? All of classical mechanics can be
summarized in this neat variational principle.
Using variational calculus, one finds that the action S has a stationary
value if and only if the Euler-Lagrange equations

d ∂L ∂L
− =0
dt ∂ q̇i ∂qi
are satisfied, for i = 1, . . . , s, where the qi are coordinates for the system, and
q̇i denotes time derivative of the coordinates.
Thus the physics of a system in classical mechanics is encoded in a single
object: its Lagrangian. This makes understanding symmetries of the theory
much easier. In terms of the equations of motion, one would say that a system
has a symmetry if both sides of the equations of motion transform in the same
way, so that the equations are invariant under the symmetry operation. But
here the statement is much simpler. A system if symmetric if its action is
invariant under the symmetry operation. Or, in the language of representation
theory, the action must transform in the trivial representation of the symmetry
group. In fact, in most cases, invariance of the action comes from invariance
of the Lagrangian itself.
In fact, in modern physics one often starts with the desired symmetries
of a given physical system, and use this to determine the form of the La-
grangian, by writing down the most general action which is invariant under
these symmetries. Let me give a very simple example. Consider a free particle
in three-dimensional space. Homogeneity of space and time implies that L can
only depend on |~v |2 : it cannot depend explicitly on ~x or t, nor can it depend
on the direction of ~v . Then, one can show that L must be proportional to |~v |2
by requiring that it is invariant under Galilean transformations of the frame
of reference. This constant of proportionality defines the mass of the particle,
which can be shown to be positive, since otherwise the action would have no
extremum. Thus, simply by requiring invariance under given symmetries, we
are able to recover the Lagrangian of the system! The same can be done for a
system of N particles, see “Mechanics” by Landau and Lifschitz, Sections 1.3
and 1.4.
This approach is what is done for instance in particle physics. Hamilton’s
principle (and its quantum version) is postulated. We then study the symme-
tries of the system, and figure out the most general Lagrangian that is com-
patible with these symmetries. This Lagrangian will involve a certain number
of constants; we then use experiments to fix the value of these constants.
3.4.2 The Lagrangian and quantum physics

Talking about particle physics, I should mention here that quantum mechanics
and quantum field theory can also be obtained from the Lagrangian of the
physics. This beautiful approach is due to Feynman.
While classical mechanics can be summarized by saying that the system
takes the path in configuration space which extremizes the action, quantum
mechanics can be summarized by saying that the systems takes all paths in
configuration space, each of which is weighted by the exponential of the action.
A simple argument for this goes as follows. Consider the standard double-slit
experiment. A particle emitted from a source at time t1 passes through one
or the other of two holes, drilled in a screen, and is detected at time t2 by a
detector located on the other side of the screen. In quantum mechanics, the
amplitude for detection is given by the sum of the amplitudes corresponding
to the two paths that the particle can take. But then Feynman asked the
following. First, what happens if there are three holes in the screen? Well,
then the amplitude is the sum of the three paths. What if there are four holes?
Well, it must be the sum of the four paths. And what if I put another screen
with some holes in it? Then the amplitude becomes the sum of all possible
paths. But then, what if I have an infinity of such screens with an infinity of
holes in them, such that the screens are actually no longer there??? The only
logical conclusion is that the amplitude then must be the sum of all paths from
the source to the detector. It is easy to show (in quantum mechanics) that
each path must be weighted by a factor of
i
e ~ S,
where S is the action of the system. This infinite sum over all paths is what is
called in quantum physics a path integral. When ~ → 0, the classical limit
is recovered, since the path integral “localizes” on the classical configuration
given by the extremum of S. This gives a beautiful conceptual explanation of
the difference between quantum and classical physics.
We have just seen that quantum physics can be understood neatly in terms
of the action of a system. In fact, all of the fundamental laws of physics can
be written in terms of an action principle. This includes electromagnetism,
general relativity, the Standard Model of particle physics, and string theory.
For instance, almost everything we know about Nature can be captured in the
Lagrangian
√ 1
L = g(R + Fµν F µν + ψ̄6Dψ),
2
where the first term is the Einstein term, the second the Yang-Mills (or Maxwell)
term, and the last the Dirac term. Those describe gravity, the forces of Nature
(like electromagnetism and the nuclear forces), and the dynamics of particles
like electrons and quarks. You are welcome to try to understand what these
terms really mean! :-)
3.4.3 Symmetries and conservation laws

Going back to classical mechanics (although the discussion here has a parallel
in quantum physics), from the Lagrangian of a system one can extract the
equations of motion. However, for most systems in Nature, these equations are
not integrable, which means that they can only be solved numerically (with a
big enough computer).
In many cases however a great deal of information can be obtained by
studying conserved quantities, which are quantities whose values remain
constant during motion. These conserved quantities can often be found rela-
tively easily and then used to reduce the differential system to a simpler one
which is easier to solve. In fact, in many cases these conserved quantities are
even more interesting than a full solution to the equations of motion. It is
therefore of interest to learn how to find conserved quantities.
The key result here is that we can associate conserved quantities to sym-
metries of the system! This is the fundamental statement of Noether’s the-
orem: For any continuous symmetry of the action of a system, there is a
corresponding conserved quantity. And this is not just an abstract statement;
we can construct this conserved quantity explicitly. For instance, invariance
under time translations gives rise to conservation of energy. Invariance under
space translations gives rise to conservation of momentum. Invariance under
space rotations gives rise to conservation of angular momentum. Once again,
group theory is fundamental!
Chapter 4
Lie groups and Lie algebras
In this section we introduce Lie groups and Lie algebras.
4.1 Lie groups

Objectives
• Recall the definition of Lie groups and matrix Lie groups.
• Relate the interpretation of Lie groups as groups of matrices, groups of
continuous transformations of a target space, and in terms of the manifold
structure of their parameter spaces.
• Identify the fundamental representation of a matrix Lie group.
In the first few sections we discussed at length finite groups and their rep-
resentations. But in the beginning we also introduced examples of infinite
and continuous groups. In this section we go back to the study of continuous
groups, and focus on those that also satisfy a differentiability criterion, known
as Lie groups. But in fact, all examples of continuous groups that we have seen
previously are also examples of Lie groups.
4.1.1 Continuous (topological) groups and Lie groups: ab-

stract definition
Let us go back to discrete groups for a moment. What is a discrete group? It
is a discrete set G, with a binary operation (that we call group multiplication)
satisfying a bunch of axioms. In particular, the binary operation closes, so we
can think of it as a mapping α : G × G → G, given by α : (a, b) 7→ ab, where
a, b ∈ G and on the right-hand-side we mean group multiplication of a and b.
We can also define another mapping β : G → G, which maps elements to their
inverses: β : g 7→ g −1 for g ∈ G. So we encode the structure of a group in
terms of the mappings α and β.
Now, what is a continuous group? The idea is to replace the discrete set G
by a continuous space M of a certain dimension, which we call the dimension
of the continuous group. More precisely, to define a continuous group we need
to define the notion of continuity. The precise statement is that we add to
the set M the extra structure of a topology (i.e. a notion of open sets, or
82
CHAPTER 4. LIE GROUPS AND LIE ALGEBRAS 83
“proximity”), so that it becomes a topological space. Then, we require that

the mappings α and β be continuous with respect to this topology. This
defines the concept of a topological group, which is what we really mean by
“continuous group”.
The idea of Lie groups is to impose the extra structure of differentiability.
To define what it means to be “differentiable”, we introduce an extra structure
on the set M : we require that M be a smooth (i.e. C ∞ ) manifold. Then, we
require that the mappings α : M × M → M and β : M → M be smooth maps.
Definition 4.1.1 Lie groups. A Lie group is a real smooth (C ∞ ) manifold
M that is also a group, in which the group operations of multiplication α :
M × M → M and inversion β : M → M are smooth maps. ♦
This is a nice, abstract definition, but it requires an understanding of man-
ifolds and differential geometry. Fortunately, as we will see in practice we will
not need to know much about differential geometry, as the Lie groups that we
will work with will all be realized as groups of matrices that depend continu-
ously on a set of parameters. But let us not get ahead of ourselves: let us first
work through a few simple examples.
Example 4.1.2 A one-dimensional Lie group. Consider the set R∗ , with
binary operation given by multiplication. It is easy to see that it satisfies the
axioms of a group. Moreover, R∗ is a smooth manifold, and the group opera-
tions of multiplication and inversion are given by the mappings α : (a, b) 7→ ab
and β : a → a1 , for a, b ∈ R∗ , which are smooth maps. Thus R∗ is a one-
dimensional Lie group.
Example 4.1.3 The circle group. Consider the one-dimensional circle S 1 ,
which can be parameterized by an angle θ ∈ [0, 2π). This is a group under the
operation given by addition of angles. Again, S 1 is a smooth manifold, and
the group operations, given by α : (θ1 , θ2 ) 7→ θ1 + θ2 (mod 2π) and β : θ1 7→
−θ1 (mod 2π) are smooth maps. Thus S 1 is a one-dimensional Lie group.
Example 4.1.4 The affine group of one dimension. Let us now consider
the set of matrices of the form

a b
T = ,
0 1
with a ∈ R>0 and b ∈ R. This set forms a group under matrix multiplication.
Its underlying manifold is given by R>0 ×R, which is a smooth manifold. Using
matrix multiplication, we see that

a1 b1 a2 b2 a1 a2 a1 b2 + b1
T1 T2 = = ,
0 1 0 1 0 1
and hence the group operation on R>0 × R can be written as:
α : ((a1 , b1 ), (a2 , b2 )) 7→ (a1 a2 , a1 b2 + b1 ),
which is smooth. As for inverses, given a matrix

a b
T = ,
0 1
its inverse is
− ab
1
T = a .
0 1
Thus the group inversion on R>0 × R can be written as:

1 b
β : (a, b) 7→ ,− ,
a a
which is also differentiable since a ∈ R>0 . Therefore, this is a two-dimensional

Lie group.
Now that we have endowed our groups with the extra structure of a topol-
ogy, we can define what it means for a group to be compact.
Definition 4.1.5 Compact groups. We say that a topological group is
compact if its topology is compact. ♦
Of the three examples that we saw above, only the circle group S 1 is com-
pact.
4.1.2 Matrix Lie groups

The abstract definition of Lie groups as groups that are also smooth manifolds
is nice, but in practice it is a bit too abstract for our purposes. All the Lie
groups that we will study in this class have very explicit realizations as matrix
groups. To study those, we need to introduce the mother of all Lie groups: the
general linear group, which we have already encountered many times.
Example 4.1.6 GL(n, R) and GL(n, C). Consider the group GL(n, R) of
real n×n invertible matrices. Real invertible matrices have n2 real independent
entries. One can show that this is a Lie group of dimension n2 . The details
of the proof are beyond the scope of this class, but let us briefly sketch the
argument. Let Mn (R) be the set of n × n matrices. Then the subset of real
invertible matrices GL(n, R) is obtained by requiring that the determinant
is non-zero. But since the determinant is a polynomial map, one can show
that the subset GL(n, R) forms an open set in Mn (R) (with respect to the
Zarisky topology). Using this, one can conclude that GL(n, R) is a Lie group
of dimension n2 .
Using a similar line of reasoning, one can show that GL(n, C) is also a
Lie group, with real dimension 2n2 (since complex invertible matrices have n2
independent complex entries, that is, 2n2 independent real parameters).
With this example, we can construct many Lie groups, because of the fol-
lowing crucial theorem, known as the “closed subgroup theorem”:
Theorem 4.1.7 The closed subgroup theorem. If H is a closed subgroup
of a Lie group G, then H is also a Lie group (with the smooth structure agree-
ing with the embedding). In particular, all closed subgroups of GL(n, R) and
GL(n, C) are Lie groups. Those are known as matrix Lie groups.
We will not prove this theorem here, as it is beyond the scope of this class.
But the key is that it implies that all our favourite continuous groups are Lie
groups.
Example 4.1.8 SL(n, R) and SL(n, C). The group SL(n, R) of real matrices
with determinant one is a closed subgroup of GL(n, R), and hence a Lie group.
Imposing that the determinant is equal to one removes one degree of freedom,
and hence the dimension of SL(n, R) is n2 − 1.
Similarly, the group SL(n, C) of complex-valued matrices with determinant
one is a closed subgroup of GL(n, C), and hence a Lie group. Again, imposing
that the determinant is equal to one fixes one constraint, but in this case
it reduces the number of complex degrees of freedom by one, and hence the
number of real degrees of freedom by two. So the real dimension of SL(n, C)
is 2(n2 − 1).
Note that SL(n, R) and SL(n, C) are non-compact.
Example 4.1.9 O(n) and SO(n). The group O(n) of real orthogonal ma-
trices is a closed subgroup of GL(n, R), and hence a Lie group. The number of
independent real parameters in an n × n orthogonal matrix can be calculated
to be 21 n(n − 1), which is the dimension of the Lie group.
The group SO(n) of special orthogonal matrices (with determinant one)
is also a closed subgroup of GL(n, R), and hence a Lie group. In this case,
imposing that the determinant of an orthogonal matrix A is equal to one is a
discrete condition, as any orthogonal matrix must satisfy det A = ±1 to start
with. Thus it does not reduce the number of degrees of freedom, and hence
the dimension of SO(n) is also 12 n(n − 1).
The orthogonal groups O(n) and special orthogonal groups SO(n) are com-
pact Lie groups.
Example 4.1.10 U (n) and SU (n). The group U (n) of unitary matrices is
a closed subgroup of GL(n, C), and hence a Lie group. The number of real
independent parameters in an n × n unitary matrix can be calculated to be n2 ,
which is the dimension of U (n).
The special unitary group SU (n) is also a closed subgroup of GL(n, C), and
hence a Lie group. Imposing the condition that the determinant is equal to
one reduces the number of real degrees of freedom by one. Indeed, any unitary
matrix A has | det A|2 = 1, which means that its determinant can be written
as det A = eiθ for some real number θ. Imposing that det A = 1 fixes this real
parameter θ. As a result, the real dimension of SU (n) is n2 − 1.
Just as for orthogonal groups, both U (n) and SU (n) are compact Lie
groups.
In this course we will focus on matrix Lie groups, such as the orthogonal
and unitary groups, which are realized as closed subgroups of the general linear
groups. This is why we will rarely need to go back to the formal abstract
definition of Lie groups.
4.1.3 Lie groups as groups of transformations of a target

space
We first introduced Lie groups abstractly, as groups that are also smooth man-
ifolds. We then studied an important class of Lie groups, known as matrix Lie
groups, that are realized as closed subgroups of the general linear groups. But
concretely, we often think of Lie groups in a third way, as groups of transfor-
mations of a target space. For instance, we often think of SO(n) as the group
of rotations in Rn . Let us make these statements a little more precise.
Suppose that we are interested in a group of continuous transformations of
some space, which we call the target space T :
L : T → T.
If the target space has dimension d, and we use coordinates x1 , . . . , xd on T ,
we can write these transformations as:
x0i = gi (x1 , . . . , xd ; a1 , . . . , an ), i = 1, . . . , d,
where the a1 , . . . , an are parameters for the continuous transformations. From
this point of view, the abstract continuous group is the parameter space
of the transformations, which is parameterized by the a1 , . . . , an . In other
words, to each choice of real numbers a1 , . . . , an in a certain set M , we assign
a corresponding continuous transformation of the target space as above.
From the point of view of transformations, the group operation consists in

composition of transformations. How does this relate to the group operation on
parameter space? Given two transformations of the target space, correspond-
ing to two points in parameter space, composing the transformations gives a
new transformation of the target space, which is assigned to a new point in
parameter space. Thus, from the point of view of the parameter space, com-
posing transformation of the target space gives a mapping α : M × M → M
on parameter space, which specifies the group operation. If M is a smooth
manifold, we recover the structure of Lie group.
Remark 4.1.11 It is important however to distinguish between the dimension
of the target space (d), and the dimension of the Lie group, which is the
number of parameters n on which the transformations depend. In other words,
the dimension of the Lie group is the dimension of the parameter space M ,
which is the underlying smooth manifold of the Lie group.
All the Lie groups that we have encountered above can be understood as
groups of transformations.
Example 4.1.12 Rescalings in one dimension. Let x ∈ R, and consider
the group of rescalings
x0 = ax
parameterized by a non-zero real number a ∈ R∗ . These transformations form
an abelian group, with the operation being composition. The parameter space
is R∗ . Composing two transformations with parameters a and b, we get:
x00 = ax0 = abx,
and hence the group operation on parameter space is the mapping α : R∗ ×

R∗ → R∗ given by
α : (a, b) 7→ ab.
The underlying parameter space is thus the group R∗ with operation given by
multiplication, which is a one-dimensional Lie group. It is Example 4.1.2.
Example 4.1.13 Affine transformations in one dimension. As a vari-
ation of the previous example, we consider the group of transformations:
x0 = a1 x + b1 ,
with a1 ∈ R>0 and b1 ∈ R. These transformations form a non-abelian group.

Its parameter space is two-dimensional, and given by M = R>0 ×R. Composing
two transformations with parameters (a1 , b1 ) and (a2 , b2 ), we get:
x00 = a1 x0 + b1 = a1 (a2 x + b2 ) + b1 = a1 a2 x + (a1 b2 + b1 ).
So the group operation corresponds to the mapping α : M × M → M on

parameter space given by:
α : ((a1 , b1 ), (a2 , b2 )) 7→ (a1 a2 , a1 b2 + b1 ). (4.1.1)
This is nothing else than the group operation on the two-dimensional Lie group
constructed in Example 4.1.4.
We also introduced matrix Lie groups above as groups of matrices. But
it is also customary to think of them as groups of transformations of target
space.
Example 4.1.14 General linear transformations. Let the target space

be an n-dimensional real vector space V . The linear invertible transformations
L : V → V form the general linear group GL(V ), which we have already
studied. If V = Rn , then we get the general linear group GL(n, R). Note
that, while the dimension of the target space is n, the dimension of the Lie
group (which is the dimension of the parameter space for n × n real invertible
matrices) is n2 .
If we choose the complex vector space V = Cn for target space, then we get
the group GL(n, C). The real dimension of the target space here is 2n, while
the dimension of the Lie group is 2n2 .
Remark 4.1.15 When we think of Lie groups as groups of transformations,
or as groups of matrices, the underlying manifold structure of the parameter
space many not be obvious. We will see that in the following examples.
Example 4.1.16 Orthogonal transformations and rotations in Rn .
One can restrict to general linear transformations in Rn that preserve length
and angles. Those are given by orthogonal transformations, and form the Lie
group O(n), studied in Example 4.1.9. One can also think of orthogonal trans-
formations as symmetries of a real n-dimensional sphere, since they preserve
its defining equation
x21 + x22 + . . . + x2n+1 = R2 .
Note that while the dimension of the target space is n, the dimension of the
Lie group O(n) is 21 n(n − 1), as orthogonal n × n matrices depend on 12 n(n − 1)
real independent parameters.
We can also further restrict to special orthogonal transformations, which
form the group SO(n). These consists in all elements of O(n) that are contin-
uously connected to the identity. Geometrically, it can be understood as the
group of rotations in Rn .
We note that the underlying manifold structure of SO(n) (that is, of its
parameter space) is not obvious to see in general. But for SO(2) it is clear.
Rotations in two dimensions are parameterized in terms of a single angle θ ∈
[0, 2π). Thus the parameter space is the circle S 1 , with the group operation
given by addition of angles. We recover the one-dimensional Lie group of
Example 4.1.3! So instead of thinking of SO(2) as the group of 2 × 2 special
orthogonal matrices, or as the group of rotations in R2 , we can also think of
it as the circle S 1 with group operation given by addition of angles. This also
makes it clear that it is a compact group, since the circle is compact.
For SO(3), things are already more complicated. Rotations in R3 are pa-
rameterized by three angles. So its parameter space is three-dimensional, which
is indeed the dimension of the Lie group. But if you remember how to param-
eterize rotations in three dimensions with angles, it is not so straightforward.
It turns out that the manifold structure of the parameter space here is RP3 ,
the real projective three-dimensional space. This example already higlights the
fact that finding the manifold structure of the underlying parameter space of
a matrix Lie group is not in general straightforward.
Example 4.1.17 Unitary transformations in Cn . The unitary group
U (n), studied in Example 4.1.10 can be realized as the group of transformations
in Cn that preserve the inner product. One could also think U (n) as the group
of symmetries of a “complex n-dimensional sphere”.
As for the the orthogonal group, the underlying manifold structure of U (n)
and SU (n) is not so obvious to see. But for SU (2) we can find the parameter
space directly. Any 2 × 2 special unitary matrix A can be parametrized as:
α −β ∗

A=
β α∗
for α, β ∈ C such that |α|2 + |β|2 = 1. Writing α = a + ib and β = c + id, the

equation |α|2 + |β|2 = 1 becomes
a2 + b2 + c2 + d2 = 1,
which is the equation of a three-sphere S 3 in R4 . Thus, the parameter space

(or underlying manifold structure) of SU (2) is the three-sphere S 3 , which is a
compact manifold.
Remark 4.1.18 When we think of Lie groups as groups of transformations,
or equivalently as subgroups of GL(n, C), formally what we are doing is con-
structing a representation of the underlying abstract group. Indeed, to every
group element (i.e. points in parameter space), we assign a unique matrix (or
a transformation of the target space). So we are building an explicit group
homomorphism between the parameter space and GL(n, C). We call this rep-
resentation the fundamental (or defining) representation of the matrix Lie
group.
For instance, the interpretation of SO(2) as 2×2 special orthogonal matrices
is the fundamental representation of the abstract Lie group, which is isomorphic
to the circle S 1 . Similarly, the interpretation of SU (2) as 2 × 2 special unitary
matrices is the fundamental representation of the abstract Lie group, which is
isomorphic to the three-sphere S 3 .
4.2 Rotations in two and three dimensions

Objectives
• Calculate the generators of two and three dimensional rotations.
• Recover finite rotations by exponentiating the generators.

The very cool thing about Lie groups is that almost all of their structures
is encoded in their linearization. This is the key insight of Lie. In calculus, you
can approximate an analytic function f (x) by looking at its linearization (or
tangent line) at a point, say x = 0. In fact, if the function is analytic, Taylor
tells you that you can completely reconstruct the function locally if you know
all its derivatives at x = 0. But you need to know all derivatives. A similar
statement holds in more than one dimensions; you can approximate a manifold
at a point by looking at its tangent space. But this is only a linearization of
the space.
For Lie groups the situation is different. It turns out that you only need to
know the linearization of a group near the origin to recover the whole group
locally! This is because of the group structure. If you know the first derivative
of the group elements (thought of as matrices) at the origin, you can recover
the group elements through exponentiation. So we can study the structure of
Lie groups by looking at their linearizations at the origin, which are called Lie
algebras. In other words, the Lie algebra of a Lie group can be thought of as
the linearization, or tangent space, of the group manifold at the origin.
To introduce the concept of Lie algebras, we will first study rotations in
two and three dimensions, and then define the formal idea that applies to all
Lie groups.
4.2.1 Rotations in two dimensions

Let us start by looking at the Lie group SO(2) of rotations in two dimensions.
Those are given by 2 × 2 special orthogonal matrices. We know that we can
write any such rotation as a matrix

cos θ sin θ
,
− sin θ cos θ
in terms of an angle of rotation θ ∈ [0, 2π). The deep idea of Lie is that we
can recover all rotations by doing a rotation by a very small angle many many
times. Doesn’t sound very deep, but it is. :-) More precisely, we think of
infinitesimal rotations. If we take the angle θ to be small, we can approximate
cos θ ∼
= 1 + O(θ)2 and sin θ ∼ = θ + O(θ)3 . Keeping only terms of first order in
θ, we can thus approximate the rotation matrix as

cos θ sin θ ∼ 1 θ 0 1
= =I +θ .
− sin θ cos θ −θ 1 −1 0
Conversely, we can do a similar linear approximation for special orthogonal

matrices. If A is a 2 × 2 special orthogonal matrix A which is very close to
the identity matrix, we can write A ∼
= I + M for some “infinitesimal” M . The
condition that AT A = I implies that
(I + M T )(I + M ) ∼
= I + M + M T = I,
where we kept only terms of first order in M . Thus M = −M T , that is, M

is an antisymmetric matrix. But intwo dimensions any real antisymmetric
0 1
matrix is a multiple of X = , thus we can write
−1 0
A∼
= I + θX
for some real parameter θ as before. So we have recovered the same first order
expansion as for rotations. We call X the generator of the Lie group (more
precisely, it is a matrix representation of the generator of the Lie group).
Checkpoint 4.2.1 Starting from orthogonal matrices, we recovered the first
order expansion of rotations. But this should only work for special orthogonal
matrices. It looks like we never imposed the condition that det A = 1. Why?
The key insight of Lie is that we can now recover finite rotations from the
knowledge of the generator X alone. Let R(θ) be a rotation by a finite angle
θ. Pick a large integer N . We can recover R(θ) by doing R(θ/N ) N times. In
the limit as N → ∞, R(θ/N ) becomes an infinitesimal rotation. Thus we get:
N
N θ
R(θ) = lim R(θ/N ) = lim I+ X .
N →∞ N →∞ N
x N
If we naively use the relation ex = limN →∞ (1 + N) for matrices, we would
conclude that
R(θ) = eθX .
In other words, we can recover finite rotations by exponentiating the infinites-
imal generator! That is cool. This is due to the group structure, which says
that we can recover rotations by successively repeating rotations of smaller

angles.
Let us be a little more precise and prove this explicitly, which makes the
relation with the group structure more explicit.
Lemma 4.2.2 The generator of two-dimensional rotations. Let

0 1
X= .
−1 0
We define the exponential of a matrix A through the series:

∞
X 1 n
eA = A .
n=0
n!
Then any two-dimensional rotation R(θ) can be written in terms of the gener-
ator X via exponentiation:
R(θ) = eθX .
Proof. Let us start with a two-dimensional rotation R(θ). Since it is a Lie
group, we can Taylor expand near the identity:
dR 1 d2 R
R(θ) = I + θ+ θ2 + . . . (4.2.1)
dθ θ=0 2 dθ2 θ=0
(By derivative of a matrix here we mean derivative of its entries.) Let us now
identity the derivatives of the rotation matrix. Rotations form a Lie group.
The group structure is given by composition of rotations, which can be written
as the requirement that
R(θ1 + θ2 ) = R(θ1 )R(θ2 ).
Taking the derivative on both sides with respect to θ1 , and then setting θ1 = 0,
we get
dR(θ1 + θ2 ) dR(θ1 )
= R(θ2 ) . (4.2.2)
dθ1 θ1 =0 dθ1 θ1 =0
We can calculate the left-hand-side of (4.2.2) via the chain rule:

dR(θ1 + θ2 ) d(θ1 + θ2 ) dR(θ2 )
= .
d(θ1 + θ2 ) dθ1 θ1 =0 dθ2
For the right-hand-side of (4.2.2), we define the matrix

dR(θ1 ) 0 1
X := = .
dθ1 θ1 =0 −1 0
Thus (4.2.2) becomes

dR(θ)
= XR(θ). (4.2.3)
dθ
In particular,
dR(θ)
= X.
dθ θ=0
In fact, taking repeated derivatives of (4.2.3), we get:
dn R(θ) dn−1 R(θ)

n
=X .
dθ dθn−1
Evaluating at θ = 0, we get, by induction on n,
dn R(θ)
= X n.
dθn θ=0
Thus all derivatives of the rotation matrix at the origin are determined by the
generator X! As is clear from the proof, this follows because of the group
structure of rotations.
Plugging this back into (4.2.1), with X 0 := I, we get
∞
X 1 n n
R(θ) = θ X = eθX ,
n=0
n!
which concludes the proof.

The cool thing here is that all two-dimensional rotations can be recovered
by exponentiating the infinitesimal generator X. So instead of studying the Lie
group SO(2) directly, we could instead study the properties of the generator
X. This is our first example of a Lie algebra. We say that X is an element
of the Lie algebra so(2): we use the weird font to distinguish the Lie algebra
from the Lie group. The algebra is rather trivial however here, since it has
only one generator. Our next example will be less trivial.
4.2.2 Rotations in three dimensions

Let us now consider the Lie group SO(3) consisting of three-dimensional ro-
tations. We think of those as 3 × 3 special orthgonal matrices A. We do an
infinitesimal expansion A ∼
= I + M for an infinitesimal M . Then the orthogo-
nality condition becomes
AT A = (I + M T )(I + M ) ∼
= I + M T + M = I.
Therefore, as for two-dimensional rotations, we conclude that M is a real

antisymmetric 3 × 3 matrix. In fact, it is customary in physics to introduce a
factor of i in our linearization, and define instead A ∼
= I + iL. Then what we
have shown is that (iL)T = −iL. Since (iL)T is a real matrix, then it is equal
to its complex conjugate −iL† . Thus the condition that (iL)T = −iL can be
rewritten as L† = L, that is, L is a purely imaginary Hermitian 3 × 3 matrix.
Any 3 × 3 purely imaginary Hermitian matrix can be written as a linear
combination of three matrices:
     
0 0 0 0 0 −1 0 1 0
L1 = −i 0 0 1 , L2 = −i 0 0 0  , L3 = −i −1 0 0 .
0 −1 0 1 0 0 0 0 0
(4.2.4)
Those are the infinitesimal generators of the Lie group SO(3). We can write
an arbitrary infinitesimal rotation as iθ1 L1 + iθ2 L2 + iθ3 L3 for three real pa-
rameters θ1 , θ2 , θ3 (we thus see that SO(3) is a three-dimensional Lie group).
Following the same argument as for rotations in two dimensions, we can also
conclude that we can write an arbitrary finite rotation in three dimensions (i.e.
an element of the Lie group SO(3)) by exponentiation:
P3
R(θ) = ei i=1 θi Li .
The algebra generated by the generators L1 , L2 , L3 is the Lie algebra so(3).

Now that we have more than one generator, this is more interesting. How do
we define the abstract properties of this algebra? (Instead of writing down
an explicit representation in terms of three-dimensional matrices.) For this,

we need one more element: the notion of a binary operation on the algebra,
which, in matrix form, will be given by the commutator. Let us see how this
comes about.
4.2.3 Commutation and the commutator

What we have seen so far is that for rotations in two and three dimensions,
we can reconstruct the group elements by exponentiating the infinitesimal gen-
erators. This is in fact a general statement for all Lie groups, as we will see
in the next section. In the context of three-dimensional rotations, we found
an explicit representation for the generators in terms of 3 × 3 matrices. But
just as when we defined abstract groups, we would like to obtain an abstract
definition of the algebra of generators of a Lie group. For this, there is one
element missing. In general, rotations do not commute. How can we see that
from the point of view of the infinitesimal generators?
Let R ∼= I + M and R0 ∼ = I + M 0 be infinitesimal rotations. Then
RR0 R−1 ∼
= (I + M )(I + M 0 )(I − M ) = I + M 0 + (M M 0 − M 0 M ),
where we neglected terms of higher order. If we define the commutator

[M, M 0 ] as
[M, M 0 ] := M M 0 − M 0 M,
then non-commutativity of R and R0 is encapsulated in the statement of
whether the commutator of their infinitesimal generators vanishes or not.
Thus, to encode the group structure of a Lie group in terms of the abstract
notion of the algebra of its infinitesimal generators, we need to specify the com-
mutation relations between the generators. This will give an abstract definition
of a Lie algebra, from which a Lie group can be obtained by exponentiation.
For three-dimensional rotations, looking at the representation of the gener-
ators L1 , L2 , L3 in terms of 3 × 3 matrices that we found, it is easy to compute
that
[L1 , L2 ] = iL3 , [L2 , L3 ] = iL1 , [L3 , L1 ] = iL2 .
Checkpoint 4.2.3 Check that the generators of three-dimensional rotations
L1 , L2 , L3 satisfy these commutation relations.
This can be encoded neatly using the Levi-Civita symbol ijk , which is
defined as:

1
 if (i, j, k) is an even permutation of (1, 2, 3);
ijk = −1 if (i, j, k) is an odd permutation of (1, 2, 3);


0 if two indices are repeated.
Then:
3
X
[Li , Lj ] = i ijk Lk . (4.2.5)
k=1
As we can see, the commutator closes, since the right-hand-side is a linear

combination of the generators. Thus it provides a bilinear operation L×L → L
on the vector space L spanned by L1 , L2 , L3 . This is generally true, as we will
see. For any Lie algebra, we are given a bilinear operation, which we write as
a commutator, such that
X
[Li , Lj ] = cijk Lk .
k
We call the cijk the structure constants of the Lie algebra.

Abstractly, we can define the Lie algebra so(3) as being the three-dimensional
vector space V of real linear combinations of the generators L1 , L2 , L3 , with a
bilinear operation [·, ·] : V × V → V specified by (4.2.5).
4.2.4 Differential representation

So far we have worked exclusively with matrix representations of the rotation
groups. In fact, we defined the rotation groups in terms of their fundamental,
or defining, representations, as subgroups of GL(n, R).
But there are other types of representations that are very useful. Let us
focus on three-dimensional rotations as an example. We can represent the
infinitesimal generators of rotations as differential operators acting on functions
f (x, y, z) on R3 . To do that, what we need to do is find differential operators
that satisfy the commutation relations (4.2.5). It is not too difficult to check
that the following differential representation works:

∂ ∂ ∂ ∂ ∂ ∂
L1 = −i y −z , L2 = −i z −x , L3 = −i x −y .
∂z ∂y ∂x ∂z ∂y ∂x
(4.2.6)
Checkpoint 4.2.4 Check that the differential operators L1 , L2 , L3 satisfy the
commutation relations (4.2.5).
You may recognize those operators as the angular momentum operators
in quantum mechanics (up to a factor of ~). This is not a coincidence! The
angular momentum operators in quantum mechanics are a representation of the
infinitesimal generators of the group of rotations in three dimensions. In fact,
the possibility of going back and forth between differential representations and
matrix representations of the Lie algebra is the essence of the duality between
the Schrodinger and Heisenberg pictures of quantum mechanics.
4.3 Lie algebras

Objectives
• Recall the definition of Lie algebras.
• Calculate the Lie algebra corresponding to a given Lie group.
• Reconstruct the group elements from the Lie algebra generators by ex-
ponentiation.
In this section we generalize the discussion about two-dimensional and
three-dimensional rotations to more general Lie groups. We will try to come
up with an abstract definition for the algebra of generators of a Lie group.
We will proceed in three steps. First, we will study how to extract the Lie
algebra corresponding to a given matrix Lie group. Second, we will define the
abstract notion of a Lie algebra. Third, we will show how we can recover the
group structure corresponding to a Lie algebra through exponentiation.
4.3.1 The Lie algebra of a Lie group

Let us first describe how, given a matrix Lie group, one can extract the cor-
responding Lie algebra. The discussion here is very similar to what was done
for two-dimensional and three-dimensional rotations in Section 4.2.
Geometrically, the Lie algebra is the linearization of the Lie group at the
origin. In other words, we replace the group manifold locally near the origin
by its tangent space. Since we are working with matrix Lie groups, we can
calculate this linearization explicitly by expanding the matrices of the funda-
mental representation near the identity, and keeping only terms of first order.
Let us be a little more precise.
We start with a matrix Lie group. That is, we start with its defining
representation, so we can think of the group elements as matrices A that depend
on n real parameters. We want to construct the associated Lie algebra. We
proceed in three steps.
1. We expand the group elements near the identity, to first order:
A = I + iL.
We determine the properties of the matrices L by imposing the appro-

priate condition on A. For instance, for SO(n), imposing that A is a
real orthogonal matrix constrains L to be a purely imaginary Hermitian
matrix.
2. We find a basis L1 , . . . , Ln for the n-dimensional vector space of matrices
L satisfying the appropriate constraint. We call the Li the infinitesimal
generators of the Lie group. We write L as a linear combination of the
Li :
Xn
L= θi Li .
i=1
3. For any two group elements A, A0 with first order expansions A = I + iL,
A0 = I + iL0 , the commutator of matrices [L, L0 ] encodes whether the
group elements commute. Since L and L0 are linear combinations of the
generators Li , to know all such commutators we only need to calculate
the commutators of the generators Li . Because of the group structure
we know that the commutator closes, that is, the commutator of two
generators will be itself a linear combination of the generators. So we
can write:
d
X
[Li , Lj ] = cijk Lk (4.3.1)
k=1
for some structure constants cijk . We calculate these structure constants

from the form of the generators of the group.
This determines the Lie algebra associated to a Lie group. It is the vector space
V of real linear combinations of the generators Li , with a binary operation
[·, ·] : V × V → V specified by (4.3.1).
Remark 4.3.1 Note that in this approach, it is crucial that we start with a
matrix representation of the group. This allows us to do an expansion near
the identity matrix, and to determine the bracket [·, ·] by evaluating the com-
mutator of the matrices corresponding to the infinitesimal generators. But in
the end, we end up with an algebra that can be defined abstractly in terms
of the generators and the relation (4.3.1). It does not depend on the matrix
representation anymore. And in fact, had we started with a different matrix
representation of the Lie group, we would have obtained a different matrix
representation of the generators Li , but the resulting algebra would have been
the same, with the same relation (4.3.1).
4.3.2 Abstract definition of a Lie algebra

In the previous section we have seen how we can determine the linearization of
a matrix Lie group at the origin (that is, the associated Lie algebra). We ob-
tained a vector space spanned by the infinitesimal generators of the Lie group,
equipped with a bracket operation. We calculated the Lie algebra associated
to a Lie group using a specific representation of the group (usually the defin-
ing representation). But the end result is an algebraic structure that can be
defined entirely abstractly. Let us now define the abstract concept of a Lie
algebra.
Definition 4.3.2 Lie algebra. A Lie algebra is a vector space g over some
field F (generally taken to be the real numbers in this class), together with a
binary operation [·, ·] : g × g → g, called the Lie bracket, that satisfies the
following axioms:
1. Bilinearity:
[ax + by, z] = a[x, z] + b[y, z],

[z, ax + by] = a[z, x] + b[z, y],
for all a, b ∈ F and all x, y, z ∈ g.

2. Antisymmetry:
[x, y] = −[y, x],
for all x, y ∈ g.
3. The Jacobi identity:
[x, [y, z]] + [z, [x, y]] + [y, [z, x]] = 0,
for all x, y, z ∈ g.
♦
This is the abstract definition of a Lie algebra. Concretely, in this course the
Lie algebras that we will study will be obtained as linearizations of matrix Lie
groups. That is, they will be constructed as in Subsection 4.3.1, starting from
a matrix representation of a Lie group. Then the vector space g is constructed
as real linear combinations of the infinitesimal generators of the group, and the
Lie bracket is realized explicitly as a commutator of the matrix representation
for these generators. We know that the Lie bracket closes, that is, for any two
x, y ∈ g, [x, y] ∈ g, as required for a Lie algebra.
In this context, the three axioms in the definition of Lie algebras are clear.
The commutator of matrices is certainly bilinear and antisymmetric, by defi-
nition. As for the Jacobi identity, one can check that given any three matri-
ces A, B, C, the commutator [A, B] = AB − BA satisfies the Jacobi identity
[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0.
Checkpoint 4.3.3 Check that for any three matrices A, B, C, the commuta-
tor [A, B] = AB − BA satisfies the Jacobi identity [A, [B, C]] + [B, [C, A]] +
[C, [A, B]] = 0.
Thus the procedure of Subsection 4.3.1 always produces a Lie algebra. The
structure of the algebra is encoded in the structure constants as in (4.3.1).
4.3.3 Recovering a Lie group from a Lie algebra

Now suppose that you are given a Lie algebra g. How can you reconstruct the
group structure from the Lie algebra?
The procedure is simple: exponentiation. Start with the vector space g.
We assume that there is a matrix representation for the Lie algebra, so that
we can represent elements of the vector spaces in terms of matrices. Then:
1. We pick a basis Li , i = 1, . . . , n for g. Using the matrix representation
we can think of the elements Li as P matrices. A general element of g can
n
be written as a linear combination i=1 ti Li .
2. We construct a matrix representation for the Lie group by exponentia-
tion: Pn
g(~t) = ei i=1 ti Li .
The matrices Li form a representation of the infinitesimal generators of
the Lie group.
The claim is that the matrices g(~t) thus constructed form a group under
matrix multiplication. Associativity is clear. We need to check that the identity
matrix is in there; that inverses exist; and the matrix multiplication closes.
Exponentiation maps the origin of the vector space ~0 ∈ g to the matrix
g(~0) = e0 = I,
which is the identity matrix.PSo this is good. P

As for inverses, for any i ti Li ∈ g, − i ti Li is also in g, and by expo-
nentiation those are mapped to
Pn Pn
g(~t)g(−~t) = ei i=1 ti Li e−i i=1 ti Li = I.
This follows because for any matrix A, we have eA e−A = I, which can be
checked by expanding the Taylor series for the exponential.
The tricky statement is whether the groupPmultiplication P closes. In other
words, we needP to show that given any two i ti Li and i si Li in g, then
there exists a i ui Li ∈ g such that
g(~t)g(~s) = g(~u).
In other words, for any two A, B ∈ g, we want to find a C ∈ g such that
eA eB = eC .
To show such a thing, we would like to construct C explicitly from A and

B. For C to be in g, this means that we should be able to write it as a linear
combination of A, B and commutators of those.
First, if A and B commute, then it straightforward to show that
eA eB = eA+B .
Thus, in this case, the statement is certainly true since C = A + B, and

exponentiation produces a (abelian) group.
Checkpoint 4.3.4 Prove that eA eB = eA+B for any two commuting matrices
A and B.
However, eA eB is not equal to eA+B anymore when A and B do not com-
mute. If we define C by eC = eA eB , expand the formal series on both sides
of the equality, and determine C order by order in A and B, we obtain the

expansion:
1 1
C = A + B + [A, B] + [A, [A, B]] + . . . ,
2 12
where . . . stands for an infinite formal series of terms that are expressed as
linear combinations of commutators of A and B. This is essentially the Baker-
Campbell-Haussdorff formula. The key point here is that C is a formal series of
terms in g. And if A and B are close enough to the identity (this can be stated
formally), then the formal series converges to a Lie algebra element C ∈ g.
Thus exponentiation constructs a group, at least locally near the identity.
To summarize, from a Lie algebra, one can recover the local behaviour of
a Lie group through exponentiation. Nice!
4.4 SU (2)
Objectives
• Determine the Lie algebra su(2).
• Explain the statement that SU (2) is a double cover of SO(3).
In this course we will mostly focus on the two most important Lie groups
in physics, namely SO(3) and SU (2). We have already calculated the Lie
algebra so(3) associated to the Lie group SO(3) in Section 4.2. We obtained
that so(3) is the three-dimensional vector spanned by elements L1 , L2 , L3 with
the bracket
3
X
[Li , Lj ] = i ijk Lk .
k=1
4.4.1 The Lie algebra su(2)

Let us now study the Lie algebra su(2) associated to the special unitary group
SU (2). The fundamental representation of SU (2) consists in 2 × 2 unitary
matrices U with determinant one:
U † U = I, det U = 1.
To obtain the Lie algebra, we do a first order expansion U = I + iL. To first

order, the unitarity condition becomes
U †U ∼
= (I + iL)† (I + iL) ∼
= I − iL† + iL = I.
Therefore L† = L, which says that L is a Hermitian matrix. We also need to

impose the condition that det U = 1. We can write a general 2 × 2 complex-
valued matrix as
a b
L= ,
c d
for a, b, c, d ∈ C. Then

∼ 1+a b
det U = det(I + iL) = det = (1 + a)(1 + d) − bc.
c 1+d
Since we keep only terms of first order in L, we can ignore any terms that are
order two in the complex numbers a, b, c, d. Thus, to first order, we have
det U ∼
= 1 + a + d.
For this to be equal to one we must have a = −d, that is, the matrix L is
traceless (has vanishing trace).
The conclusion is that L must be a 2 × 2 traceless Hermitian matrix. Any
such matrix can be written as

t1 t2 + it3
L=
t2 − it3 −t1
for some real numbers t1 , t2 , t3 ∈ R. Indeed, since the matrix is Hermitian,

the diagonal terms must be real, and since it is traceless they must add up to
zero. The off-diagonal terms must be complex conjugate since the matrix is
Hermitian.
It thus follows that the Lie algebra su(2) is three-dimensional. A natural
basis for the vector space consists of the matrices:

1 0 1 1 0 −i 1 1 0
T1 = , T2 = , T3 = . (4.4.1)
2 1 0 2 i 0 2 0 −1
Those are known as Pauli matrices. By calculating the commutators, one

finds:
[T1 , T2 ] = iT3 , [T2 , T3 ] = iT1 , [T3 , T1 ] = iT2 .
In other words,
3
X
[Ti , Tj ] = i ijk Tk .
k=1
What? This is the same abstract algebra as so(3)! Indeed, it turns out
that the Lie algebras so(3) and su(2) are isomorphic. Really, they are the
same. This is absolutely fundamental, and, as we will see, is the reason for
the appearance of fermions (particles with half-integer spin) in non-relativistic
physics.
This does not mean however that the Lie groups SO(3) and SU (2) are
isomorphic: in fact they are not. But they are certainly closely related. More
precisely, there exists a two-to-one group homomorphism from SU (2) to SO(3).
We say that SU (2) is a “double covering” of SO(3).
4.4.2 SU (2) vs SO(3)

We have seen in the previous section that the Lie algebras su(2) and so(3) are
isomorphic. What does this mean for the Lie groups SU (2) and SO(3)?
The fact that their Lie algebras are the same means that the two groups are
“locally isomorphic”, which is a rather striking statement. What this means is
that, using exponentiation of the Lie algebra, we would recover the same group
structure locally near the identity. This does not mean however that the two
groups are globally isomorphic: in fact they are not.
But they are certainly closely related. The precise statement is that there
exists a group homomorphism from SU (2) to SO(3) that is two-to-one. This
is what we show next. We say that SU (2) is a “double covering” of SO(3).
Let us now construct the group homomorphism f : SU (2) → SO(3).
Theorem 4.4.1 SU (2) covers SO(3). There exists a two-to-one group
homomorphism f : SU (2) → SO(3), which means that the Lie groups SU (2)
and SO(3), which share the same Lie algebras su(2) ∼ = so(3), are only locally
isomorphic.
We provide a formal proof of this statement using the adjoint representation
of the Lie group SU (2) below. But before we do that, let us write down an
informal argument, which may be more enlightening.
Both SU (2) and SO(3) share the same Lie algebra. For SU (2), we obtained
the Lie algebra by constructing the generators of SU (2), which is given by the
vector space spanned by the Pauli matrices (4.4.1), which we reproduce here
for convenience:

1 0 1 1 0 −i 1 1 0
T1 = , T2 = , T3 = .
2 1 0 2 i 0 2 0 −1
For SO(3), we obtained the algebra by looking at the generators of rotations
in three dimensions (4.2.4), which we also reproduce here:
     
0 0 0 0 0 −1 0 1 0
L1 = −i 0 0 1 , L2 = −i 0 0 0  , L3 = −i −1 0 0 .
0 −1 0 1 0 0 0 0 0
One can show that all special unitary matrices in SU (2) can be obtained by
exponentiation, and similarly for special orthogonal matrices in SO(3) (this is
because SU (2) and SO(3) are compact and connected). Thus we can write an
arbitrary U ∈ SU (2) as
3
!
X
U (θ1 , θ2 , θ3 ) = exp i θk Tk ,
k=1
and an arbitrary R ∈ SO(3) as

3
!
X
R(θ1 , θ2 , θ3 ) = exp i θk Lk .
k=1
The mapping f : SU (2) → SO(3) is then given by U (θ1 , θ2 , θ3 ) 7→ R(θ1 , θ2 , θ3 ).

How do we see that the mapping is two-to-one? Let us argue that it is
two-to-one by looking at the special case with θ1 = θ2 = 0, but the argument
is general (although a bit more tedious) and can be written explicitly for any
choice of parameters.
We know that R(0, 0, θ) = R(0, 0, θ + 2π), since R(0, 0, θ) corrrespond to a
rotation in three dimensions by angle θ about an axis, hence rotating further
by 2π does not do anything. On the other hand, since T3 is diagonal, we can
write !
θ
ei 2 0
U (0, 0, θ) = exp (iθT3 ) = θ .
0 e−i 2
Thus, !
θ
ei 2 +iπ 0
U (0, 0, θ + 2π) = θ = −U (0, 0, θ).
0 e−i 2 −iπ
So shifting θ by 2π does not send the unitary matrix U (0, 0, θ) to itself; rather
it sends it to minus itself. Shifting by 4π would bring back U (0, 0, θ) to itself.
As a result, we see that both U (0, 0, θ) and U (0, 0, θ + 2π) = −U (0, 0, θ)
are mapped to the same rotation matrix R(0, 0, θ), and hence the mapping
is two-to-one. The point here is that rotations have period 2π, while unitary
matrices as written in terms of Pauli matrices have period 4π. This is why
SU (2) covers SO(3) twice.
Let us now write down a formal proof of the theorem.
Proof. We will construct the group homomorphism explicitly. In fact, it will

be given by the adjoint representation of the group SU (2), so let us construct
this explicitly.
The adjoint representation of a Lie group G is given by a representation
of the group elements of G as linear operators acting on the Lie algebra g
of G. Since the Lie algebra g is a n-dimensional vector space, where n is the
dimension of the Lie group, the adjoint representation maps the group elements
to n × nmatrices acting on g. So it gives a representation T : G → GL(g) of the
same dimension as the dimension of the Lie group G. In the case of G = SU (2),
which has dimension 3, the adjoint representation is three-dimensional, and
represents the group of elements of SU (2) as 3 × 3 matrices in GL(su(2)).
As we will show, it turns out that the image of the adjoint representation is
SO(3) ⊂ GL(su(2)), and hence it provides the group homomorphism SU (2) →
SO(3) that we are looking for.
Let us now construct the adjoint representation explicitly. We start with
an element X ∈ su(2), which we think of as a 2 × 2 traceless Hermitian matrix.
We take an element U ∈ SU (2), which is a 2 × 2 special unitary matrix. Then
we define:
AdU (X) = U XU † ,
which is basically a similarity transformation of X by U .
Let us now show that AdU (X) ∈ su(2), i.e. AdU (X) is traceless Hermitian.
The tracelessness property follows since
Tr AdU (X) = Tr(U XU † ) = Tr(XU † U ) = Tr X = 0,
by the cyclic property of traces. As for being Hermitian, we find:
(AdU (X))† = (U XU † )† = U X † U † = U XU † = AdU (X),
since X = X † . Therefore, AdU (X) is Hermitian and traceless. In other words,
we have a mapping AdU : su(2) → su(2), which takes X 7→ AdU (X) = U XU † .
That is, AdU ∈ GL(su(2)).
We can now define the adjoint representation of SU (2). It is given by
the group homomorphism Ad : SU (2) → GL(su(2)), with U 7→ AdU . In
other words, it takes a special unitary matrix U ∈ SU (2), and maps it to
the linear operator AdU on su(2). Since su(2) is a three-dimensional vector
space, this gives a three-dimensional representation of SU (2). Note that it is
straightforward to show that Ad is a group homomorphism, since Ad : U1 U2 7→
AdU1 U2 = AdU1 ◦ AdU2 by definition of AdU .
The image of Ad is in GL(su(2)). We can write any 2×2 Hermitian traceless
matrix X as a real linear combination of the Pauli matrices (4.4.1):
X = x1 T1 + x2 T2 + x3 T3 .
Thus we can identify su(2) ∼= R3 , where we map an Hermitian traceless matrix
X to the vector of coefficients ~x in the basis for su(2) given by Pauli matrices.
With this identification we can see Ad : SU (2) → GL(3, R). But in fact, the
image is in SO(3) ⊂ GL(3, R), as we now show.
We notice that for any Hermitian matrix X = x1 T1 + x2 T2 + x3 T3 , the
determinant is calculated as:
4 det X = (x3 )(−x3 ) − (x1 − ix2 )(x1 + ix2 ) = −|~x|2 .
So we can identify the determinant of X ∈ su(2) with the length of the vector
~x ∈ R3 . Since the determinant is preserved by similarity transformations, it
follows that
det X = det U XU † = det AdU (X).
This means that the vectors in R3 corresponding to X and AdU (X) have the
same length. In other words, AdU is a linear operator on R3 that preserves
the length of vectors: it is an orthogonal transformation in O(3) ⊂ GL(3, R).
Thus Ad : SU (2) → O(3) ⊂ GL(3, R). But in fact, we know even more. From
Example 4.1.10, we know that the underlying manifold structure of SU (2) is S 3 .
In particular, it is connected (in fact, it is also simply connected). Therefore,
the image of the continuous mapping Ad : SU (2) → O(3) must be connected to
the identity in O(3), which means that it must be a subgroup of SO(3) ⊂ O(3).
Thus Ad : SU (2) → SO(3).
What remains to be shown is that the group homomorphism Ad : SU (2) →
SO(3) is surjective, and that it is two-to-one.
We will leave the proof that Ad : SU (2) → SO(3) is surjective as an
exercise. (Note that there are many ways that this can be done.)
To show that Ad : SU (2) → SO(3) is two-to-one, pick any unitary matrix
U1 , and set U2 := −U1 . Then
Ad(U1 ) = AdU1 = AdU2 = Ad(U2 ),
since for any Hermitian matrix X,
AdU1 (X) = U1 XU1† = (−U1 )X(−U1 )† = U2 XU2† = AdU2 (X).
Thus the mapping Ad : SU (2) → SO(3) is two-to-one.

Equivalently, we could show that the kernel is non-trivial. The kernel of
Ad corresponds to 2 × 2 unitary matrices U that are mapped to AdU being the
identity element of SO(3). There are two such matrices,

1 0 −1 0
, ,
0 1 0 −1
since in both cases AdU (X) = X, i.e. AdU is the identity in SO(3). Thus
ker(f ) ∼
= Z2 , where Z2 is realized as ±1 times the identity matrix. Since the
kernel has two elements, this means that Ad : SU (2) → SO(3) is two-to-one.

Remark 4.4.2 We have shown that there is a two-to-one group homomorphism
Ad : SU (2) → SO(3). The kernel is Z2 , realized as ±1 times the identity matrix
in SU (2). Thus, by the first isomorphism theorem, we can write
∼ SU (2)/Z2 .
SO(3) =
The groups however are said to be “locally isomorphic”, since they have iso-
morphic Lie algebras. Equivalently, this can be seen since if U ∈ SU (2) is near
the identity, then −U ∈ SU (2) is not. And hence in a neighborhoud of the
identity the map is a local isomorphism.
4.5 General remarks

The presentation of Lie groups and Lie algebras in this course is necessarily
unsystematic, and does not do justice to this beautiful area of mathematics.
But let us conclude this section with a few general remarks about Lie groups
and Lie algebras.
Remark 4.5.1 To every Lie group is associated a unique Lie algebra. The Lie
algebra is uniquely determined as being the tangent space of the Lie group at
the origin, or, in the language of matrix Lie groups, as being the algebra of the
infinitesimal generators of the group.

Remark 4.5.2 Two Lie groups that have isomorphic Lie algebras are not
necessarily isomorphic. Different Lie groups may share the same Lie algebras,
such as SO(3) and SU (2). However, Lie groups that have the same Lie algebras
are locally isomorphic near the identity. Globally, they are generally related
by surjective group homomorphisms, i.e. covering maps. For instance, SU (2)
covers SO(3) twice, as we have seen. The “largest” of the groups associated to
a Lie algebra is called the universal covering group. It is simply connected
(i.e. it consists of one piece and has no “holes” in it, that is, its fundamental
group is trivial). In the case of SO(3), its universal covering is SU (2), which
is simply connected (since it is diffeomorphic to S 3 , see Example 4.1.10).
Remark 4.5.3 While different Lie groups may share the same Lie algebra,
there is nonetheless a uniqueness result that is crucial. Lie’s third theorem
states that every finite-dimensional real Lie algebra is the Lie algebra of some
simply connected Lie group. Moreover, if two Lie groups are simply connected
and have isomorphic Lie algebras, then the groups themselves are isomorphic.
What this means is that to every finite-dimensional real Lie algebra one can
associate a unique Lie group that is simply connected. SU (2) plays this role
for the Lie algebras so(3) ∼= su(2).
Our goal for the remaining of this class is to study representation theory
of Lie groups such as SO(3) and SU (2). More precisely, from physics we are
particularly interested in representation theory of SO(3), which corresponds
to non-relativistic particles.
Representation theory of Lie algebras is generally easier than representation
theory of Lie groups, because Lie algebras are linear. We have already seen that
given a representation of a Lie algebra, one can reconstruct a representation of
a Lie group using exponentiation. So the goal is to study representation theory
of Lie groups by studying representation theory of Lie algebras. However, one
has to be careful, since different Lie groups may share the same Lie algebra.
If we start with a representation of a Lie algebra, and define a representation
via exponentiation, how do we know what Lie group we are talking about?
The main result here is that every representation of a simply connected
Lie group comes from a representation of its corresponding Lie algebra. So by
starting with representations of the Lie algebra, what we are constructing is
all representations of the unique simply connected Lie group associated to the
algebra (the universal covering). The simply connected property is crucial.
For instance, representation theory of the Lie algebra su(2) ∼ = so(3) con-
structs all representations of the universal covering SU (2). However, some of
those will not be honest representations of the non-simply connected SO(3) ∼ =
SU (2)/Z2 . As we will see, they are so-called “projective”, or “spin”, represen-
tations. Those are mappings that preserve group multiplication but only up
to a constant. So from the point of view of SO(3), representation theory of
the Lie algebra so(3) constructs not only the honest representations of SO(3),
but also some spin representations.
Fortunately, in quantum mechanics it is sufficient to preserve only group
operation up to a constant (since wave-functions are only defined up to phase),
so we actually do care about spin representations, and should include them.
Indeed, as we will see, the odd-dimensional representations (corresponding in
physics to particles with integer spin) coming from so(3) are honest represen-
tations of SO(3), while the even-dimensional ones (corresponding in physics
to particles with half-integer spin) are spin representations of SO(3) (they are
however true representations of the simply connected SU (2)).
Let us now move on to representation theory!
Chapter 5
Representation theory of Lie

groups
5.1 Tensor representations of SO(3)

Objectives
• Recognize tensors as objects that transform according to representations
of SO(3).
• Determine the dimension of the irreducible representations associated to
symmetric traceless tensors.
In this section we study intuitively what representations of SO(3) look like.
This will give us an intuitive understanding of what the representation theory of
SO(3) should be, which we will then construct explicitly using representations
of its Lie algebra in the following sections.
5.1.1 Representations of SO(2)

Let us start by giving a brief and naive construction of the irreducible repre-
sentations of SO(2).
We start with the cyclic group ZN , which we can think of a discrete set
of rotations in two dimensions by angle 2π/N . We know its irreducible rep-
resentations: there are N of them, all one-dimensional, given by sending the
generator of ZN to e2πik/N for some k ∈ {0, 1, . . . , N − 1}.
Well, as N goes to infinity, we can think of the group ZN as turning into the
continuous group SO(2) of rotations in two dimensions. So we could expect
the irreducible representations of SO(2) to be all one-dimensional (which is
true since SO(2) is abelian), and to be given by the complex numbers eikθ for
θ ∈ [0, 2π), and with k a non-negative integer. So there would be an infinite
number of them, indexed by k.
While this construction is of course very naive (since it is not clear what
precisely is meant by the limit N → ∞ here), the intuition is correct. Those
are the irreducible representations of SO(2), and they are indexed by a non-
negative integer k.
103
CHAPTER 5. REPRESENTATION THEORY OF LIE GROUPS 104

Let us now move on to the more interesting case of rotations in three dimen-
sions, that is, SO(3). What are the irreducible representations? After looking
at SO(2), we expect an infinite number of them. But here we do not expect
them to be all one-dimensional, since SO(3) is not abelian.
The simplest representation is of course the trivial one, where we sent every
element R ∈ SO(3) to T (R) = 1. This is as usual a rather boring one. It is
sometimes called the scalar representation in physics, since, if you think
of it as acting on a one-dimensional vector space V , then T (R)v = v for all
v ∈ V , that is, it leaves everything invariant. So a scalar object (which is an
object that is invariant under rotations) transforms according to the trivial
representation of SO(3).
We have already studied another representation of SO(3): its defining rep-
resentation, which associates to every R ∈ SO(3) a three-dimensional rotation
matrix T (R). In physics, we sometimes call this representation the vector
representation. This is because it defines how vectors in R3 transform under
rotations. Indeed, if v ∈ R3 , then under a rotation we get v 0 = Rv, where R is
a 3 × 3 rotation matrix. In index notation, we would write
3
X
va0 = Raj vj .
j=1
Turing this around, one could think of vectors as being objects that transform
according to the three-dimensional defining representation of SO(3).
Remark 5.1.1 This is a good time to introduce a notation that is standard in
physics, which is to denote the representation of a Lie group by its dimension
in bold face. For instance, we would write 3 for the defining representation of
SO(3). The notation is slightly ambiguous though, because there may be more
than one representation of the same dimension. For instance, if a representa-
tion is complex, then its complex conjugate has the same dimension (we would
then write, say, 5 and 5 for the complex conjugate representations). Never-
theless, this is common notation, and it is usually clear from context what
representation is being discussed.
Now you probably see a pattern. How can we get more representations of
SO(3)? A natural way to think about representations is to think about how
they act. Namely, we think of objects that transform in certain ways under
rotations. We have already studied how scalars and vectors transform, which
gave us the one-dimensional and three-dimensional representation of SO(3).
What next?
The next objects to look at are matrices. How do matrices tranform under
rotations? Think of a matrix M as being a linear operator on R3 . Then under
a rotation R (which is a change of basis for R3 ) the matrix M will change
by a similarity transformation M 0 = RM RT , since R is orthogonal. In index
notation, this becomes
3
X 3
X
0 T
Mab = Rai Mij Rjb = Rai Rbj Mij .
i,j=1 i,j=1
This defines a new representation of SO(3). Indeed, we could place the nine
components of the matrix M in a vector, and then the transformation rule
above would give us a nine-dimensional representation of SO(3), where we
associate to every group element of SO(3) the 9 × 9 matrix with components
given by Rai Rbj , where R is the corresponding 3 × 3 rotation matrix. In other
words, if you recall the construction of tensor product representations, what we

are constructing here is the nine-dimensional representation that is the tensor
product of the defining representation with itself:
9 = 3 ⊗ 3.
We know that matrices transform according to this representation of SO(3).

We could keep going and define higher-dimensional representations by look-
ing at how objects with more than two indices transform under rotations.
Those objects are known as tensors, and the corresponding representations
are known as tensor representations. The rank of a tensor is the number
of indices. For instance, a rank 3 tensor Tijk transforms as
3
X
0
Tabc = Rai Rbj Rck Tijk ,
i,j,k=1
which defines a 27-dimensional representation of SO(3), namely 27 = 3⊗3⊗3.

In fact, we can think of tensors as being objects that transform according to
representations of SO(3).
5.1.3 Irreducible representations of SO(3)

In the previous section we constructed an infinite class of representations of
SO(3) by looking at how tensors transform under rotations. But are these
representations irreducible? We will not answer this question rigorously here.
We will instead look at the rank 2 case to get an intuition about what to
expect, and then state the general result without proof.
Let us look at the rank 2 tensor representation (the matrix one) as a starting
point. This 9-dimensional representation maps elements of SO(3) to operators
that act on matrices as
3
X
0
Mab = Rai Rbj Mij .
i,j=1
This representation will be reducible if there exists a proper subspace of the

space of matrices that is invariant under the action of the representation.
The key here is to notice that the transformation preserves the symmetry
property of tensors under permutation of indices. In the case of a matrix, we
can always write a matrix as a sum of its symmetric and antisymmetric parts:
1 1
Mij = (Mij − Mji ) + (Mij + Mji ) := Aij + Sij ,
2 2
where Aij = −Aji and Sij = Sji . The subspaces given by antisymmetric and
symmetric matrices respectively are invariant under the action of the repre-
sentation. Indeed, we get that
1
A0ab = (Mab 0
− Mba0
)
2
3
1 X
= (Rai Rbj Mij − Rbi Raj Mij )
2 i,j=1
3
1 X
= Rai Rbj (Mij − Mji )
2 i,j=1
3
X
= Rai Rbj Aij .
i,j=1
A similar argument holds for Sij . Thus the subspaces of symmetric and anti-
symmetric rank 2 tensors are invariant subspaces, and thus provide subrepre-
sentations. The number of independent components in a rank 2 anti-symmetric
tensor in three dimensions is 3, while the number of independent components
in a rank 2 symmetric tensor in three dimensions is 6. We thus obtain the
decomposition 9 = 6 ⊕ 3 for the nine-dimensional tensor representation.
Now, are the 6 and 3 representations irreducible? It turns out that the
three-dimensional one is, and is in fact dual to the vector representation (since
in three dimensions rank 2 anti-symmetric tensors are dual to vectors, see Zee’s
book, section IV.1, for more on this). However the six-dimensional represen-
tation is still reducible. P3
Indeed, consider the subspace consisting of the trace S = i=1 Sii . Then
one sees that
3
X
S0 = 0
Saa
a=1
d X
X 3
= Rai Raj Sij
a=1 i,j=1
3 d
!
X X
T
= Ria Raj Sij
i,j=1 a=1
3
X
= δij Sij
i,j=1
= S,
where we used the fact that R is an orthogonal matrix, and hence RT R = I.

Therefore the trace is invariant! Thus the subspace spanned by the trace
is a one-dimensional invariant subspace, which transforms according to the
trivial one-dimensional representation. We thus get the further decomposition
6 = 5 ⊕ 1. Overall, the nine-dimensional tensor representation decomposes as
9 = 5 ⊕ 3 ⊕ 1. It turns out that these summands are now all irreducible. The
objects that transform according to the five-dimensional representation are the
symmetric traceless tensors, which can be written as
S
Sij − δij ,
3
and are indeed traceless.
A similar decomposition, albeit slightly more complicated, is possible for
higher rank tensors. For SO(3), it turns out that there is a special duality that
can be used to show that for higher rank tensors, only the symmetric traceless
tensors give rise to new irreducible representations (see Zee’s book, section
IV.1). The end result, which we will not prove here, is that all irreducible rep-
resentations of SO(3) can be constructed by looking at how symmetric traceless
tensors transform.
Our last question for this section is: what is the dimension of these irre-
ducible representations? To find the dimension we need to find the number
of independent components of symmetric traceless tensors. This will give the
dimensions of the irreducible representations of SO(3).
Lemma 5.1.2 Let Sk1 ···kj be a rank j tensor that is fully symmetric under
permutations of indices, and such that it is traceless:
3
X
δk1 k2 Sk1 k2 ...kj = 0.
k1 ,k2 =1
(Note that it does not matter what pair of indices we choose here, since S is
fully symmetric.) Then S has 2j + 1 independent components.
Proof. We need to count the number of independent components of S. First,
let us count the number of components of a symmetric rank j tensor. Suppose
first that each index can only take values 1 and 2. There the possibilities
are 2 · · · 2, 2 · · · 21, 2 · · · 211, and so on. This gives j + 1 possibilities. Then
we add a 3. If we have a 3 for the first index, then we have j possibilities
for 2s and 1s in the remaining j − 1 indices. If we have two 3s, then we
have j − 1 possibilities
P for the remaining indices. And so on. Overall we get:
Pj j 1 1
k=0 (k + 1) = k=0 k + (j + 1) = 2 j(j + 1) + (j + 1) = 2 (j + 1)(j + 2)
possibilities. So a symmetric rank j tensor has 21 (j + 1)(j + 2) independent
components. P3
The tracelessness condition k1 ,k2 =1 δk1 k2 Sk1 k2 ...kj = 0 consists in a num-
ber of independent conditions. The number of conditions here is the number
of values that the indices k3 · · · kn can take. From the previous paragraph, this
is the number of independent components of a rank j − 2 symmetric tensor,
which is 12 j(j − 1). Therefore, the total number of independent components of
a symmetric traceless rank j tensor is
1 1
(j + 1)(j + 2) − j(j − 1) = 2j + 1.
2 2

The end result is that there is an infinite number of irreducible representa-
tions for SO(3), indexed by a non-negative integer j, with dimensions 2j + 1.
The objects that transform according to these representations are symmetric
traceless rank j tensors.
This is the key result of this section. And, if you have done some physics,
you may have encountered the formula 2j + 1 before. It is rather famous in
the history of quantum mechanics and atomic physics. It corresponds to the
degeneracy of states of the hydrogen atom (or spherical harmonics). It also
corresponds to the multiplicity, or quantum states, of particles with integer
spin. This is certainly not a coincidence, as we will see!
5.2 Representations of Lie algebras and the ad-

joint representation
Objectives
• Recall the definition of representations of Lie algebras, and distinguish
with representations of Lie groups.
• Determine the adjoint representation of a Lie algebra.
The irreducible representations of a Lie algebra are in one-to-one correspon-
dence with the irreducible representations of its associated simply connected
Lie group (the universal covering). Concretely, given a representation of a Lie
algebra, one may obtain the corresponding representation of the Lie group by
exponentiation. Since representations of Lie algebras are easier to construct
than for the corresponding Lie groups, we now move on to the construction of
irreducible representations of Lie algebras.
In this section we will define more precisely what representations of Lie
algebras are, and then look at one important representation that exists for all
Lie algebras: the adjoint representation.
5.2.1 Definition
Recall that a Lie algebra g is a vector space with a bracket [·, ·] that satisfies a
bunch of axioms. Suppose that g is n-dimensional, and pick a basis L1 , . . . , Ln
for g. Then we can write
n
X
[Li , Lj ] = cijk Lk , (5.2.1)
k=1
for some choice of structure constant cijk .

A representation of a Lie algebra is simply a mapping that represents the
elements of g as matrices, with the bracket [·, ·] realized as commutator of
matrices. Concretely, one maps the basis L1 , . . . , Ln to n matrices that satisfy
the commutation relations (5.2.1). As usual, if the matrices are d × d, we call
d the dimension of the representation.
More formally:
Definition 5.2.1 Representation of a Lie algebra. Let g be a Lie al-
gebra, and let V be a vector space. Let gl(V ) be the Lie algebra of all linear
endomorphisms (i.e. linear maps V → V ) with bracket given by commutator
of matrices [X, Y ] = XY − Y X. A representation Γ of g is a mapping
Γ : g → gl(V )
that is compatible with the brackets (this is called a Lie algebra homomor-
phism):
Γ([x, y]) = [Γ(x), Γ(y)] := Γ(x)Γ(y) − Γ(y)Γ(x),
for all x, y ∈ g. The dimension of the representation is the dimension of V .
♦
As for groups, we can talk about irreducible representations, which
are representations such that V has no proper invariant subspace, that is, no
proper subrepresentation.
Remark 5.2.2 When we define representations of groups, we map group ele-
ments to linear operators in GL(V ), that is, invertible matrices. Invertibility is
required because of the group properties, since we map the group operation to
matrix multiplication. For Lie algebras, we map elements of the vector space
to linear operators in gl(V ), i.e. matrices that are not necessarily invertible.
This is an important distinction.
We have already seen examples of Lie algebra representations.
Example 5.2.3 Trivial representation. As for groups, the trivial repre-
sentation Γ always exists for Lie algebra. It is defined by sending all elements
of g to the zero endomorphism 0 of a one-dimensional vector space V , which
maps all elements of V to the origin. Indeed, if Γ(Li ) = 0 for i = 1, . . . , n,
where L1 , . . . , Ln is a basis for g, then it certainly preserves the bracket, since

n
X
[Γ(Li ), Γ(Lj )] = cijk Γ(Lk ),
k=1
for any choice of structure constants.

Example 5.2.4 The defining representation for SU (2) and SO(3). We
constructed the Lie algebras su(2) ∼= so(3) by looking at the infinitesimal gen-
erators of SU (2) and SO(3) in their fundamental, or defining, representations.
This gives rise to a corresponding representation of the Lie algebras.
For SU (2), we started with the 2-dimensional representation of SU (2) con-
sisting of 2×2 special unitary matrices. We found in (4.4.1) that the generators
are the Pauli matrices:

1 0 1 1 0 −i 1 1 0
T1 = , T2 = , T3 = . (5.2.2)
2 1 0 2 i 0 2 0 −1
Those satisfy the commutation relations

3
X
[Ti , Tj ] = i ijk Tk ,
k=1
which defines the three-dimensional Lie algebra su(2). More precisely, the Pauli
matrices (5.2.2) furnish a two-dimensional representation of the Lie algebra
su(2), since they map the generators to 2×2 matrices that obey the appropriate
commutation relations.
In the case of SO(3), we started with the 3-dimensional representation of
SO(3) consisting of 3 × 3 special orthogonal matrices. We found in (4.2.4) that
the infinitesimal generators are:
     
0 0 0 0 0 −1 0 1 0
L1 = −i 0 0 1 , L2 = −i 0 0 0  , L3 = −i −1 0 0 .
0 −1 0 1 0 0 0 0 0
(5.2.3)
Those also satisfy the commutation relations:
3
X
k=1
Thus (5.2.3) defines another representation of the same Lie algebra su(2) ∼
=
so(3), this time a three-dimensional one.
5.2.2 The adjoint representation

Before we move on to the study of irreducible representations of the Lie algebra
su(2), let us define an important representation that exists for all Lie algebras
g, which is called the adjoint representation.
Lemma 5.2.5 The adjoint representation. Let g be a n-dimensional Lie
algebra, and pick a basis L1 , . . . , Ln , with bracket
n
X
[Li , Lj ] = cijk Lk .
k=1
The structure constants define a n-dimensional representation of g, called the

adjoint representation, as follows. One defines n matrices Γ1 , . . . , Γn by
(Γi )jk := −cijk .
What this equation means is that the jk’th component of the matrix Γi is given
by the structure constant −cijk . Then these Γ1 , . . . , Γn form a n-dimensional
representation of g.
Proof. To show that it is a representation, we need to check that the mapping
respects the bracket, that is, that the matrices Γi satisfy the commutation
relations
n
X
[Γi , Γj ] = cijk Γk .
k=1
Well, it turns out that this is a direct consequence of the Jacobi identity. Recall
that for a Lie algebra g, the bracket [·, ·] satisfies the Jacobi identity
[x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0,
for all x, y, z ∈ g. In particular, if we pick x = Li , y = Lj , z = Lk , we get
[Li , [Lj , Lk ]] + [Lj , [Lk , Li ]] + [Lk , [Li , Lj ]] = 0. (5.2.4)

Pn
But [Li , Lj ] = k=1 cijk Lk . Thus we can rewrite (5.2.4) as
n
X
(cjka [Li , La ] + ckia [Lj , La ] + cija [Lk , La ]) = 0.
a=1
Doing the same thing one more time we get:

n X
X n
(cjka ciab + ckia cjab + cija ckab ) Lb = 0.
a=1 b=1
Since the Lb are linearly independent, this means that the coefficients of the lin-
ear combination must be identically zero, so we obtain the following condition
on the structure constants:
n
X
(cjka ciab + ckia cjab + cija ckab ) = 0.
a=1
We note that cijk = −cjik , so we can write:

n
X n
X
(cika cjab − cjka ciab ) = − cija cakb .
a=1 a=1
Now if we use the definition of the matrices of the adjoint representation

(Γi )jk = −cijk , we can rewrite this equation as
n
X n
X
((Γi )ka (Γj )ab − (Γj )ka (Γi )ab ) = cija (Γa )kb .
a=1 a=1
This is an equation for the kb’th component of a matrix, where the sum over
a comes from matrix multiplication. Thus it can be rewritten in matrix form
as the equation
n
X
Γi Γj − Γj Γi = cija Γa ,
a=1
which is precisely the commutation relation of the Lie algebra.

We can define the adjoint representation a little more abstractly, for those
who like this stuff. The adjoint representation is a representation that maps
elements of the vector space g to linear operators on the vector space g itself.
That is, we think of elements of the Lie algebra as operators acting on the
algebra itself. The formal definition is the following:
Definition 5.2.6 The adjoint representation (bis). Let g be a Lie algebra,
and let gl(g) be the Lie algebra of linear endomorphisms of g. The adjoint
representation is defined as the mapping
ad :g → gl(g)
x 7→ adx := [x, ·],
where adx = [x, ·] is the operator on g defined by
adx :g → g
y 7→ [x, y].
♦
In this more abstract formulation, we send an element x ∈ g to the operator
that acts on g by evaluating the bracket of any y ∈ g with the fixed x ∈ g. It
may not be obvious at first that this is equivalent to the definition in terms of
structure constants, but it is. We will leave the comparison as an exercise.
Checkpoint 5.2.7 Show that the abstract definition of the adjoint represen-
tation is equivalent to the definition in terms of structure constants, by cal-
culating the components of the matrices corresponding to the linear operators
adLi .
Example 5.2.8 The adjoint representation of su(2). Let us now construct
the adjoint representation of su(2). We recall the commutation relations
3
X
[Ti , Tj ] = i ijk Tk .
k=1
The structure constants are thus cijk = iijk . We construct the adjoint repre-
sentation by defining the three matrices Γi , i = 1, 2, 3 with components given
by
(Γi )jk = −iijk .
Recalling the definition of the Levi-Civita symbol, we see that these matrices
take the form
 
0 0 0
Λ1 = − i 0 0 1 ,
0 −1 0
 
0 0 −1
Λ2 = − i 0 0 0  ,
1 0 0
 
0 1 0
Λ3 = − i −1 0 0 .
0 0 0
Those are precisely the generators of SO(3), see (5.2.3)! Therefore, the ad-
joint representation of su(2) ∼
= so(3) constructs the infinitesimal generators of
SO(3), and by exponentiation the defining representation of SO(3). Note that
this is particular to SO(3) though; in general, the adjoint representation of a

Lie algebra does not construct the defining representation of its associated Lie
group. For instance, for SO(n), the defining representation is n-dimensional
(consisting of n × n rotation matrices), while the adjoint representation of the
Lie algebra so(n) is 21 n(n − 1)-dimensional (the dimension of the Lie algebra).
These dimensions match only when n = 3.
5.3 The highest weight construction for su(2)

Objectives
• Construct irreducible representations of su(2) using the highest weight
construction.
• Determine the matrices of an irreducible representation in the orthonor-

mal basis of states given by the highest weight construction.
• Relate these representations to spherical harmonics.
Let us now do a systematic construction of irreducible representations of the
Lie algebra su(2) using the highest weight construction. This approach is rather
general and can be suitably generalized to construct irreducible representations
of Lie algebras in general.
The idea is to construct representations by looking at how they act on
vector spaces. In other words, we will construct first the space on which a
representation acts, and from there extract the matrices of the representation
in an appropriate basis of states.
5.3.1 The highest weight construction

Let L1 , L2 , L3 be a basis for the three-dimensional Lie algebra su(2), with
bracket
3
X
k=1
We want to construct representations of this algebra. To construct a represen-
tation, we are looking for three matrices Γ1 , Γ2 , Γ3 with commutator
3
X
[Γi , Γj ] = i ijk Γk . (5.3.1)
k=1
Since we are constructing representations up to equivalence, we can always

use similarity transformations to make the representation as simple as possible.
Our goal is to use a similarity transformation to simultaneouly diagonalize as
many of the matrices Γi s as possible. But since two matrices that can be
diagonalized by the same similarity transformation must commute, we can
only diagonalize one of the three Γi s. Its is conventional to choose Γ3 as being
diagonal. In other words, we will be constructing representations of su(2) such
that Γ3 is a diagonal matrix.
The idea is to construct the vector space on which the representation acts.
We will call a state an element of the vector space on which the representation
acts and that is an eigenvector of Γ3 . We use the standard Dirac “bra-ket”
notation. We write the eigenstate of Γ3 with eigenvalue m as |mi:
Γ3 |mi = m|mi.
At this stage, all that we know is that m is a real number, since representations
of su(2) are composed of Hermitian matrices (and hence with real eigenvalues).
Remark 5.3.1 The Dirac bra-ket notation. The Dirac bra-ket notation is
conventional in the treatment of Lie algebras in both physics and mathematics.
In this notation, we represent a complex vector v as |vi. We represent its
complex conjugate transpose (v ∗ )T as hv|. Thus the norm square of v can be
written as
n
X
hv|vi = (v ∗ )T v = |vi |2 .
i=1
In general, for two vectors u, v,

n
X
hu|vi = (u∗ )T v = u∗i vi .
i=1
We are constructing finite-dimensional representations. So we know that
Γ3 only has a finite number of eigenvectors. We pick the state with the largest
eigenvalue, and call this state the highest weight state. We will denote it
by |ji, with the letter j reserved to the largest eigenvalue of Γ3 in a given
representation. We normalize this state so that
hj|ji = 1.
We now define a lowering operator Γ− and a raising operator Γ+ by

1
Γ± = √ (Γ1 ± iΓ2 ).
2
It will be convenient, for reasons that will become clear shortly, to use Γ− , Γ+ , Γ3
as basis for our representation, instead of Γ1 , Γ2 , Γ3 . So we should transform
the commutation relations (5.3.1) into commutations relations for the new ba-
sis. A straightforward calculation shows that
[Γ3 , Γ± ] = ±Γ± , [Γ+ , Γ− ] = Γ3 . (5.3.2)
So instead of constructing matrices Γ1 , Γ2 , Γ3 satisfying (5.3.1), we are now

trying to construct a diagonal matrix Γ3 and two matrices Γ± satisfying (5.3.2).
Γ± are called lowering and raising operator for the following reason. Pick
a state |mi, and act with Γ3 Γ± . We get:
Γ3 Γ± |mi =(Γ± Γ3 ± Γ± )|mi

=(m ± 1)Γ± |mi.
In other words, Γ± |mi is also an eigenstate of Γ3 , with eigenvalue (m ± 1).

So acting with Γ− on an eigenstate of Γ3 produces a new eigenstate with
eigenvalue lowered by 1 (hence the name lowering operator), while acting with
Γ+ produces a new eigenstate with eigenvalue raised by 1 (hence the name
raising operator).
We will fix the normalization of the eigenstate of Γ3 using Γ− :
|m − 1i := Γ− |mi.
Then, acting on the highest weight state |ji with Γ− , we produce a tower of
states:
|ji, |j − 1i, ..., |qi.
We know that the tower must end at some point, since we are constructing a
finite-dimensional representation. Thus there must be an eigenstate |qi of Γ3
such that
Γ− |qi = 0.
The next step is to figure out what q is for a given j. This will tell us the
dimension of the corresponding representation. To this end, we notice that we
have not fully determined how Γ+ acts on an eigenstate of Γ3 . All that we
know is that
Γ+ |mi = Nm |m + 1i
for some normalization constant Nm .
Lemma 5.3.2 Let |mi be eigenstates of Γ3 such that Γ− |mi = |m − 1i, with
highest weight state |ji (that is, Γ+ |ji = 0). Then
Γ+ |mi = Nm |m + 1i
with
1 1
Nm = j(j + 1) − m(m + 1).
2 2
Proof. We can compute Nm as follows:
Nm |m + 1i =Γ+ |mi
=Γ+ Γ− |m + 1i
=(Γ− Γ+ + Γ3 )|m + 1i
=(Nm+1 + (m + 1))|m + 1i.
Therefore, we get a recursion
Nm = Nm+1 + (m + 1).
We can solve this recursion as follows. First, since |ji is the highest weight
state, we know that Nj = 0. Thus
Nj−1 = j.
Then
Nj−2 = Nj−1 + (j − 1) = j + (j − 1).
By induction, we get
s−1
X 1 s
Nj−s = (j − k) = sj − s(s − 1) = (2j − s + 1),
2 2
k=0
for s = 0, . . . , 2j. We redefine the index as m = j − s to get:

j−m 1 1
Nm = (j + m + 1) = j(j + 1) − m(m + 1).
2 2 2

With this result we can determine what the lowest weight state |qi is. By
definition, it is the end of the tower, so we must have Γ− |qi = 0. Then:
0 =Γ+ Γ− |qi
=(Γ− Γ+ + Γ3 )|qi
=(Nq + q)|qi,
1
with Nq = 2 j(j + 1) − 21 q(q + 1). Thus we are looking for a solution to the
equation
1 1 1 1
j(j + 1) − q(q + 1) + q = j(j + 1) − q(q − 1) = 0.
2 2 2 2
There are two solutions: q = j + 1 and q = −j. The first one does not make
sense, since |ji is a highest weight state. Therefore q = −j.
What this means is that we have constructed the following tower of eigen-
states for Γ3 :
|ji, |j − 1i, ..., | − j + 1i, | − ji.
There are 2j + 1 states. Since these states span the vector space on which our
representation acts, 2j + 1 must be a positive integer. That is, j must be a
non-negative half-integer.
The outcome of this construction is the following important result:
Theorem 5.3.3 Irreducible representations of su(2). There is an infi-
nite tower of irreducible representations for su(2), indexed by a non-negative
half-integer j, with dimensions 2j + 1. We call the (2j + 1)-dimensional rep-
resentation the spin-j representation, and denote its states by |j, mi (those
are eigenstates of Γ3 with eigenvalues m = −j, −j + 1, . . . , j − 1, j ) .
We have not proved here that these representations are irreducible, and
that we have obtained all irreducible representations: this is beyond the scope
of this class.
5.3.2 Normalization of states

In the construction above we define the states |j, mi as:
Γ3 |j, mi = m|j, mi, Γ− |j, mi = |j, m − 1i, Γ+ |j, mi = Nm |j, m + 1i,
with Nm = 12 j(j + 1) − 12 m(m + 1). One can show that these states are orthog-
onal, that is,
hj, m0 |j, mi = 0 if m0 6= m.
Checkpoint 5.3.4 Show that hj, m0 |j, mi = 0 if m0 6= m.
However, the states |j, mi are not normalized. We normalized the highest
weight state by requiring that hj, j|j, ji = 1, but hj, m|j, mi =
6 1 for m ≤
2
j − 1. Let Cm = hj, m|j, mi. We can construct a new basis of states that are
orthonormal as
^ 1
|j, mi = |j, mi.
Cm
Lemma 5.3.5 Let |j, ^ ^
mi be a tower of eigenstates Γ3 |j, ^
mi = m|j, mi, for
m = −j, −j + 1, . . . , j − 1, j that are orthonormal:
^
hj, m0 |j,
] mi = δm0 ,m .
Then
r
^ 1
Γ+ |j, mi = (j + m + 1)(j − m)|j,^
m + 1i,
2
r
^ 1
Γ− |j, mi = (j + m)(j − m + 1)|j,^
m − 1i.
2
Proof. We can construct a basis of orthonormal states from our previous basis
as
^ 1
|j, mi = |j, mi.
Cm
Clearly, this does not change how Γ3 acts, since
^
Γ3 |j, ^
mi = m|j, mi.
We want to find how Γ+ and Γ− act on this new basis.

Since Γ± = √12 (Γ1 ± iΓ2 ), and Γ1 , Γ2 are Hermitian, we find Γ†+ = Γ− and
Γ†− = Γ+ . Thus we can write
hj, m − 1|j, m − 1i = hj, m|Γ+ Γ− |j, mi = Nm−1 hj, m|j, mi,
where for the second equality we applied Γ+ Γ− on the ket |j, mi. Thus, if we
2
write Cm = hj, m|j, mi, we get the relation
2 2 1 2
Cm−1 = Nm−1 Cm = (j + m)(j − m + 1)Cm . (5.3.3)
2
Now we know that
Γ− |j, mi = |j, m − 1i.
In terms of the normalized basis, this becomes
^ Cm−1 ^
Γ− |j, mi = |j, m − 1i
Cm
Substituting (5.3.3), we get:
r
^ 1
Γ− |j, mi = (j + m)(j − m + 1)|j,^
m − 1i.
2
As for Γ+ , we know that
j−m
Γ+ |j, mi = Nm |j, m + 1i = (j + m + 1)|j, m + 1i.
2
In terms of the normalized basis, we get:
^ Cm+1 j − m
Γ+ |j, mi = (j + m + 1)|j,^
m + 1i.
Cm 2
Substituting (5.3.3):
s
^ 2 j−m
Γ+ |j, mi = (j + m + 1)|j,^
m + 1i
(j + m + 1)(j − m) 2
r
1
= (j + m + 1)(j − m)|j,^
m + 1i.
2

Remark 5.3.6 For clarity, we will now drop the tilde symbol on the normalized
states. From now on all states will be assumed to be normalized.
5.3.3 Matrix representations

We have constructed the irreducible representations of su(2) using the highest
weight construction. We can now write down explicit matrices for these repre-
sentations, using the orthonormal basis for the vector space on which they act.
We will do explicitly the most common representations in physics, the spin 1/2
and spin 1 representations.
Example 5.3.7 The spin 1/2 representation. We consider the spin 1/2
representation, with j = 1/2. There are two states: |1/2, 1/2i and |1/2, −1/2i.
Those are normalized eigenstates of Γ3 with eigenvalues 1/2 and −1/2.
Using Lemma 5.3.5, we get:
1
Γ− |1/2, 1/2i = √ |1/2, −1/2i, Γ− |1/2, −1/2i = 0,
2
and
1
Γ+ |1/2, 1/2i = 0, Γ+ |1/2, −1/2i = √ |1/2, 1/2i.
2
To write down a matrix representation, we think of the states as basis vector
of a two-dimensional vector space. We define

1 0
= |1/2, 1/2i, = |1/2, −1/2i.
0 1
Then the action of the operators Γ3 and Γ± can be rewritten in matrix form
as

1 1 0 1 0 1 1 0 0
Γ3 = , Γ+ = √ , Γ− = √ .
2 0 −1 2 0 0 2 1 0
We are usually interested in going back to the Γ1 , Γ2 , Γ3 basis. Since Γ± =

√1 (Γ1 ± iΓ2 ), we have
2
1 i
Γ1 = √ (Γ+ + Γ− ), Γ2 = − √ (Γ+ − Γ−). (5.3.4)
2 2
Thus

1 0 1 i 0 1 1 1 0
Γ1 = , Γ2 = − , Γ3 = .
2 1 0 2 −1 0 2 0 −1
Those are none other than the Pauli matrices giving the defining representation
of SU (2)! See Example 5.2.4. Thus we have recovered the two-dimensional
representation of su(2) that is obtained from the infinitesimal generators of
SU (2).
Example 5.3.8 The spin 1 representation. Consider now the spin 1
representation with j = 1, with the three normalized states |1, 1i, |1, 0i and
|1, −1i. The action of Γ± can be calculated to be:
Γ− |1, 1i = |1, 0i, Γ− |1, 0i = |1, −1i, Γ− |1, −1i = 0,
and
Γ+ |1, 1i = 0, Γ+ |1, 0i = |1, 1i, Γ+ |1, −1i = |1, 0i.
Using the basis vectors

     
1 0 0
0 = |1, 1i, 1 = |1, 0i, 0 = |1, −1i,
0 0 1
we get the matrices

     
0 1 0 0 0 0 1 0 0
Γ+ = 0 0 1 , Γ− = 1 0 0 , Γ3 = 0 0 0 .
0 0 0 0 1 0 0 0 −1
Using (5.3.4), we get:

     
0 1 0 0 1 0 1 0 0
1  i 
Γ1 = √ 1 0 1 , Γ2 = − √ −1 0 1 , Γ3 = 0 0 0 .
2 2
0 1 0 0 −1 0 0 0 −1
One can check that these matrices indeed satisfy the commutation relations of
the su(2) Lie algebra. In fact, they are equivalent to the defining representation
of SO(3) as presented in Example 5.2.4. Indeed, one can find a similarity
transformation that brings the matrices of the defining representation of SO(3)
into the matrices above (this is the similarity trasnformation that diagonalizes
Γ3 ).
5.3.4 Casimir invariant

I will not develop the general theory of Casimir elements here, but let me just
say a few words in the context of su(2). Roughly speaking, a Casimir element
is a combination of the elements of the Lie algebra that commutes with all
elements of the algebra. However, to construct a Casimir element we may take
products of matrices in a given representation, and hence we actually leave the
algebra (more precisely, a Casimir element lives in the “universal enveloping
algebra” of a Lie algebra).
An important Casimir element is the quadratic Casimir invariant. In the
context of su(2), this is the operator
Γ2 := Γ21 + Γ22 + Γ23 = Γ+ Γ− + Γ− Γ+ + Γ23 .
Using simple properties of commutators, it is easy to see that
[Γ2 , Γ1 ] = [Γ2 , Γ2 ] = [Γ2 , Γ3 ] = 0,
and hence Γ2 commutes with all elements of the Lie algebra su(2).
Since Γ2 commutes with all elements of the Lie algebra, it follows that
2
Γ |j, mi can only depend on j; it cannot depend on m. (That’s because it
must be a multiple of the identity matrix, as an operator acting on the vector
space spanned by the states |j, mi.) Thus we can calculate Γ2 |j, mi by acting
on the highest weight vector:
Γ2 |j, ji = Γ+ Γ− + Γ− Γ+ + Γ23 |j, ji

=(j + j 2 )|j, ji.

Therefore
Γ2 |j, mi = j(j + 1)|j, mi.
This Casimir invariant in physics corresponds to the total angular momentum,
which you may have seen in your quantum mechanics class. It is the same for
all states in a given irreducible representation of su(2).
5.3.5 Differential representation and spherical harmonics

In most of this class we work with matrix representations of Lie groups and
Lie algebras. But we have already seen in Subsection 4.2.4 that we can also
represent operators as differential operators acting on the space of functions of
some variables. For instance, in (4.2.6) we wrote down a representation for the
basis vectors L1 , L2 , L3 of the Lie algebra su(2) as differential operators acting
on functions f (x, y, z), which we reproduce here for convenience:

∂ ∂ ∂ ∂ ∂ ∂
L1 = −i y −z , L2 = −i z −x , L3 = −i x −y .
∂z ∂y ∂x ∂z ∂y ∂x
As mentioned there, these operators correspond to the angular momentum

operators in quantum mechanics.
Now comes the power of abstract mathematics. We can apply everyting
that we have found about representation theory of su(2) ∼ = so(3) to this par-
ticular representation of the Lie algebra. We know that we can decompose the
space of functions of (x, y, z) in terms of how they transform under the irre-
ducible representations of SO(3). It is easier to work in spherical coordinates
(r, θ, φ), since the group of symmetries here is rotation, and so functions that
are invariant under rotations will depend only on (θ, φ). Then we can write a
basis for spherically symmetric functions f (θ, φ) given by functions Yjm (θ, φ)
that are eigenfunctions of L3 :
L3 Yjm (θ, φ) = mYjm (θ, φ).
Moreover, by the construction of the Casimir invariant we know that these

functions satisfy:
L2 Yjm (θ, φ) = j(j + 1)Yjm (θ, φ).
In other words, we decompose the space of spherically symmetric functions as
a direct sum of subspaces that transform according to the irreducible repre-
sentations of su(2) ∼ = so(3). Here we want true representations of SO(3), the
group of rotations, so we keep only the representations with j a non-negative
integer. As a result, any spherically symmetric function can be written as a
linear combination of the functions Yjm (θ, φ) satisfying the properties above,
with j = 0, 1, 2, 3, . . . and m = −j, −j + 1, . . . , j − 1, j. The functions Yjm (θ, φ)
are known as spherical harmonics, and will certainly show up when you
study for instance the hydrogen atom in quantum mechanics. In our context,
this is just a different point of view on representation theory of the group of
rotations SO(3).
5.4 Projective and spin representations

Objectives
• Define projective and spin representations of a group.
• Recognize the spin representations of SO(3) from the representations of
its Lie algebra so(3).
Now that we understand the irreducible representations of the Lie algebra

∼ so(3), let us now go back to the representations of the Lie group SO(3).
su(2) =
We know that all irreducible representations of su(2) correspond to irreducible
representations of the universal cover SU (2) by exponentiation. But what do

they correspond to in terms of the Lie group SO(3)?
To answer this question we need to define projective and spin representa-
tions, to which we now turn to.
5.4.1 Projective representations

One motivation for the definition of projective representations comes from
quantum mechanics. In quantum mechanics, the state of a system is specified
by a vector in some vector space (or Hilbert space) of possible states. This
state is called the “wave-function” of the system.
But this is not quite precise. The physical probability of finding the system
in a state specified by a wave-function ψ is equal to the norm square |ψ|2 =
ψ ∗ ψ. In particular, two wave-functions that only differ by a phase give the
same physical probability. So the wave-function is only really defined up to a
phase factor: mathematically, the statement is that the state of a system is
a vector in a projective vector space (or projective Hilbert space), where two
vectors that differ by overall rescaling are identified.
Now suppose that the system has a group of symmetries. As we have
already seen, the group of symmetries act on the space of states of the system.
In other words, we get a representation of the group as acting on the space
of states. But in fact, since states are only defined up to overall rescaling, we
do not really need an honest representation of the group; it is sufficient for
the symmetry operations to transform the wavefunctions only up to rescaling.
This is the idea behind projective representations.
Definition 5.4.1 Projective representations. A projective representa-
tion of a group G is a collection of matrices T (g), g ∈ G, such that it preserves
the group structure up to a constant:
T (g)T (h) = c(g, h)T (gh)
for some constants c(g, h).

More mathematically, a projective representation of a group G on a vector
space V over a field F (such as C or R) is a group homomorphism from G to
the projectivization of GL(V ):
T : G → GL(V )/F ∗ ,
where elements of GL(V )/F ∗ are equivalence classes of invertible linear trans-
formations of V that differ by overall rescaling. ♦
This is precisely what we want in quantum mechanics, since overall rescaling
of the wave-function by a phase factor does not change the physics. In other
words, given a group of symmetry for a quantum mechanical system, we know
that it acts on the space of states of the system as a projective representation.
So our goal to understand the system is to classify projective representations
of the group of symmetries.
But how do we construct projective representations of a group G? We have
already seen that the irreducible representations of a Lie algebra are in one-to-
one correspondence with the irreducible ordinary representations of the asso-
ciated simply-connected Lie group. As for projective representations, the key
result is Bargmann’s theorem, which tells us that for a large class of groups
(which includes SO(3), the Lorentz group and the Poincare group), every pro-
jective unitary representation of G comes from an ordinary representation of
the universal cover. Great! In fact, for finite-dimensional representation this
is always true regardless of the group. But in quantum mechanics the Hilbert
space is often infinite-dimensional, so one has to be careful, but Bargmann’s
theorem holds for the groups of interest such as rotations, translations and
Lorentz transformations.
In the context of SO(3), we know that its universal cover is SU (2). So what
this means is that all irreducible projective representations of SO(3) come from
ordinary irreducible representations of SU (2), which in turn correspond to irre-
ducible representations of the Lie algebra su(2) =∼ so(3), which we have already
constructed. So we already know all projective representations of SO(3)!
Before we look at those representations more closely, let us define a partic-
ular type of projective representations that is very important in physics.
5.4.2 Spin representations

There is a particular type of projective representations that is fundamental
in physics. Those are called spin representations. They are defined for
the special orthogonal groups SO(n) (and more generally for SO(p, q) - the
discussion below can be easily adapted to this more general setup).
Roughly speaking, spin representations are representations of the elements
of SO(n) (which you can think of as rotations) that pick a sign when you
rotate by an angle of 2π about an axis. Those are particular types of projective
representations, since they preserve the group law only up to a sign. We call
spinors the objects that transform according to spin representations of SO(n),
which should be contrasted with tensors, which are objects that transform
according to ordinary representations of SO(n).
The more precise definition of spin representations is in terms of the double
cover of SO(n). We have already seen that SO(3) has a double cover, which is
given by the group SU (2). It turns out that in general, the group SO(n) has a
unique connected double cover. We call this double cover the spin group, and
denote it by Spin(n). Then there is a group homomorphism Spin(n) → SO(n)
whose kernel is Z2 . Spin representations of SO(n) are defined as being the
ordinary representations of the spin group Spin(n) that do not come from
ordinary representations SO(n).
In the case of SO(3), its unique connected double cover is SU (2), and
hence Spin(3) ∼= SU (2). Therefore, we know that all ordinary representations
of SU (2) that do not come from ordinary representations of SO(3) are spin
representations.
5.4.3 Spin representations for SO(3)

With this under our belt, we can now go back to representation theory of
SO(3). In the previous section we constructed all irreducible representations
of the Lie algebra su(2) ∼= so(3) using the highest weight construction. We
obtained an infinite tower of irreducible representations, indexed by a non-
negative half-integer j, with dimensions 2j + 1.
We know that these representations are in one-to-one correspondence with
the irreducible representations of the simply connected Lie group SU (2). But
what we are often interested in in physics is the group of rotations SO(3),
not SU (2). What do the representations of the Lie algebra su(2) = ∼ so(3)
correspond to in terms of the non-simply connected Lie group SO(3)?
From the previous discussion, from Bargmann’s theorem we know that
exponentiating the representations of su(2) will generally give projective rep-
resentations of SO(3). Moreover, since SU (2) is the unique connected double
cover of SO(3), we know that they will all be either ordinary or spin represen-
tations of SO(3). How do we distinguish between the two?
Let us first work out an example. Consider the spin-1/2 representation of
su(2), with j = 1/2. The representation is given by the Pauli matrices (4.4.1),
reproduced here for convenience:

1 0 1 1 0 −i 1 1 0
T1 = , T2 = , T3 = .
2 1 0 2 i 0 2 0 −1
Now consider two elements of the Lie algebra given by θ1 T3 and θ2 T3 for some
θ1 , θ2 ∈ R. By exponentiation, these give rise to matrices
R(θ1 ) = eiθ1 T3 , R(θ2 ) = eiθ2 T3 .
Since exponentiating a diagonal matrix means exponentiating the diagonal

components, we get:
iθ /2 iθ /2
e 1 0 e 2 0
R(θ1 ) = , R(θ 2 ) = .
0 e−iθ1 /2 0 e−iθ2 /2
Now suppose that θ1 + θ2 = 2π. Then we get:

i(θ +θ )/2
e 1 2 0
R(θ1 )R(θ2 ) =
0 e−i(θ1 +θ2 )/2

1 0
=−
0 1
= − I,
where I is the 2×2 identity matrix. But in SO(3), rotating about an axis by 2π
should be the identity transformation, not minus the identity transformation!
Thus we conclude that this representation does not quite preserve the group
structure: it preserves it only up to sign. That is, it is a spin representation of
SO(3)!
This was somehow expected, because when we constructed representa-
tions of SO(3) in terms of how tensors transform, we only obtained (2j + 1)-
dimensional representations with j a non-negative integer: we never saw the
representations with j a half-integer.
But what about the other representations of SO(3) with half-integer j?
Are they ordinary or spin representations?
Consider the highest weight construction in the previous section. In this
construction, Γ3 was always a diagonal matrix, whose diagonal entries corre-
sponded to the values of m between −j and j. If j is a half-integer, then those
diagonal entries are all half-integers. Then if we construct matrices R(θ) by
exponentiating the matrices θΓ3 , the same argument above will go through,
and we will always end up with the statement that a rotation by 2π gives minus
the identity matrix. (Note that this does not happen if the diagonal values of
Γ3 are integers, which is the case when j is an integer.)
Therefore we obtain the very important result that finite-dimensional irre-
ducible projective representations of SO(3) come in two classes:
• An infinite class of ordinary representations, indexed by a non-negative
integer j, with dimensions 2j + 1. Those are tensor representations,
and we call objects that transform according to these representations
tensors (for j = 0, we call them scalars, and for j = 1, we call them
vectors).
• An infinite class of spin representations, indexed by a positive half-integer

j (which is not an integer), with dimensions 2j + 1. Those are spin
representations, and we call objects that transform according to these
representations spinors.
Isn’t that cool? Now you see how spinors come about in physics. As you
may already know, in physics we call particles that transform according to
ordinary representations bosons (they have integer spin), and particles that
transform according to spin representations fermions (they have half-integer
spins). Those have very different physical properties, due to the difference be-
tween Bose-Einstein and Fermi-Dirac statistics (the Pauli exclusion principle).
For instance, particles that exchange forces, such as photons, gluons, the W
and Z bosons, and the Higgs, are bosons. On the other hand, particles that
make up matter, such as electrons, protons, muons, quarks, etc., are fermions.
Cool, hey?
5.5 Tensor representations of SU (N )

Objectives
• Recognize tensors as objects that transform according to representations
of SU (N ).
• Determine the dimension of the irreducible representations of SU (N ) by
looking at the corresponding tensors.
• Show that the defining representation of SU (2) is pseudo-real.
• List the irreducible representations of SU (3) as tensor representations.
In Section 5.1 we studied representations of SO(3) by looking at how tensors

transform under rotations. In this section we use a similar approach to study
tensor representations of SU (N ).
5.5.1 Tensor representations of SO(N )

Let us start with a brief recap of the construction of tensor representations of
SO(3) in Section 5.1, which applies just as well to tensor representations of
SO(N ). The idea was to construct representations of SO(N ) by constructing
objects that transform according to these representations. These objects were
called “tensors”, and denoted by a letter with many indices, for instance: Tijk .
The number of indices (called the “rank” of the tensor) tells us how these
objects transform under a rotation R ∈ SO(3):
3
X
0
Tabc = Rai Rbj Rck Tijk . (5.5.1)
i,j,k=1
If we were to put all independent components of Tijk in a column vector,

the transformation rule above would define a matrix representation of SO(3),
which in this case would be the 27-dimensional tensor product representation
27 = 3 ⊗ 3 ⊗ 3.
We also noticed that symmetry properties of the objects Tijk under per-
mutations of indices are preserved by the transformation rules (for instance,
symmetric tensors are mapped to symmetric tensors). Thus restricting to ten-

sors with specific symmetry properties defines proper invariant subspaces, or
subrepresentations. In this way we can build all kinds of tensor represen-
tations of SO(3). We claimed in Section 5.1 that for SO(3), all irreducible
representations correspond to symmetric traceless tensors, which are indexed
by a non-negative integer j (the rank of the tensor, or the number of indices),
and have dimension 2j + 1.
The case of SO(N ) is very similar. Tensors are still defined by how they
transform under n-dimensional rotations, as in (5.5.1), but with R ∈ SO(N ).
For instance, a tensor Tijk , as in (5.5.1), transforms in the N 3 -dimensional
tensor product representation N3 = N ⊗ N ⊗ N. As for SO(3), tensors with
particular symmetry properties define subrepresentations of these tensor prod-
uct representations. However, for SO(N ) irreducible representations are not
simply given by symmetric traceless tensors anymore. The study of irreducible
representations of SO(N ) in terms of symmetry properties of tensors uses
Young tableaux, which we will not discuss in this class (but feel free to re-
search them if you are interested!). The idea is to look at the general tensor
product representation N ⊗ · · · ⊗ N, which is generally reducible, and look at
how it decomposes as a direct sum of irreducible representations. This involves
the study of Clebsch-Gordan coefficients, which is another interesting topic
that we will not study in this class.
Let us also remark that we can identify the special representation known
as the adjoint representation (which is the exponentiated version of the adjoint
representation for the Lie algebra) in terms of tensors for SO(N ) in general.
Rank two anti-symmetric tensors Aij = −Aji transform under the adjoint
representation. Indeed, a rank two anti-symmetric tensor has 21 N (N − 1)
independent components, which is the dimension of the adjoint representation
of SO(N ) (also the dimension of the Lie group).
5.5.2 Tensor representations of SU (N )

Let us now move on to the study of tensor representations of SU (N ). We
now want to contruct tensors that transform in certain ways under special uni-
tary transformations U ∈ SU (N ). The key difference with SO(N ) is that the
matrices U are unitary, and, in particular, complex-valued. So the transfor-
mations U and U † are different transformations. So we should really construct
objects that transform according to products of U s and U † s. More precisely,
what we are saying here is that the defining N representation of SU (N ) is
complex-valued, so we can construct tensor products of N with itself, but also
with its complex conjugate representation N.
There is a neat way of keeping track of objects that transform according
to U and U † . We will use upper and lower indices. An object with upper
indices will transform according to products of U s, while an object with lower
indices will transform according to products of U † s. We can of course also
have objects with both lower and upper indices, which will transform accord-
ingly. Concretely, what we are doing is identifying the operation of complex
conjugation as raising or lowering an index:
ψi := (ψ ∗ )i .
This is of course just a notational convention, but it is very useful to keep track
of the transformation properties of tensors.
We then denote the components of a unitary matrix U by U i a . Its transpose
matrix U † then have components (U † )a i . With this convention the upshot is
that to construct objects with appropriate transformation properties under
unitary transformations, we should always sum only over one index that is
upper and one that is lower. This follows because of the unitary property that
U † U = I, or, in index notation,
N
X
(U † )b j U j a = δab .
j=1
As an example, a tensor Tkij transforms as

N
X
(T 0 )ij
k = U i a U j b (U † )c k Tcab .
a,b,c=1
It thus defines the tensor product representation N ⊗ N ⊗ N.

The task now is to identify proper invariant subspaces, or subrepresen-
tations, of these tensor representations. Using the exact same argument as
for SO(N ), it is straightforward to show that symmetry properties of tensor
according to permutations of upper indices and lower indices separately are
preserved by the transformation rule. For instance, a tensor that is fully sym-
metric under permutations of its upper indices is mapped to a tensor that
is also fully symmetric under permutations of its upper indices. However, we
cannot exchange upper and lower indices; symmetries under such permutations
would not be preserved by the transformation rule.
Is there an analog of the trace that was preserved by special orthogonal
transformations? Yes, but it must now involve summing over an upper and
lower index. Indeed, one sees that, for instance,
N
X N
X N
X
(T 0 )ji
i = U j a U i b (U † )c i Tcab
i=1 i=1 a,b,c=1
N N
!
X X
j i † c
= U a U b (U ) i Tcab
a,b,c=1 i=1
N
X
= U j a (δbc ) Tcab
a,b,c=1
N N
!
X X
j
= U a Tbab .
a=1 b=1
PN
Therefore, the object i=1 Tiji transforms like a tensor with one single upper
index (i.e. in the N defining representation of SU (N )). So summing over one
upper and one lower index for any tensor creates a subrepresentation. We call
this operation taking the trace of a tensor.
Since taking the trace always defines a subrepresentation, to understand
irreducible representations of SU (N ) we know that we should look for traceless
tensors that have appropriate symmetry properties upstairs and downstairs.
In general, the study of irreducible representations from this point of view
involves Young tableaux, as already mentioned for SO(N ). The idea is similar,
where we now look at general tensor products N ⊗ · · · N ⊗ N ⊗ · · · ⊗ N, and
study their decompositions as direct sums of irreducible representations using
Clebsch-Gordan coefficients.
We will not discuss Young tableaux in this course, but let us at least enu-
merate the first few non-trivial tensor representations of SU (N ):
Table 5.5.1 The first few tensor representations of SU (N )
Tensor Symmetry property Dimension of the representation

Ti - N
1
T ij = −T ji Anti-symmetric 2 N (N − 1)
1
T ij = T ji Symmetric 2 N (N + 1)
PN
Tji Traceless ( i=1 Tii = 0) N2 − 1
We note that the N -dimensional, 12 N (N − 1)-dimensional and 21 N (N +
1)-dimensional representations also have complex conjugate representations,
corresponding to similar tensors but with lower indices. We also note that the
representation given by the traceless tensor Tji is the adjoint representation of
SU (N ), which is equivalent to its complex conjugate.
5.5.3 Tensor representations of SU (2)

We already know the irreducible representations of SU (2), since they are in
one-to-one correspondence, through exponentiation, with the irreducible rep-
resentations of its Lie algebra su(2). We already constructed those using the
highest weight construction, and obtained an infinite family indexed by a non-
negative half-integer j, with dimensions 2j + 1. Let us see how those arise from
the point of view of tensor representations.
Just as for SO(3), it turns that SU (2) is very special. In the case of SO(3),
what was particular about it is that irreducible representations were all given
by symmetric traceless tensors. In the case of SU (2), it turns out that all
irreducible representations are constructed from symmetric tensors with only
upper indices. In other words, for SU (2), we do not need tensors with lower
indices. We will see in a second why.
So all irreducible representations of SU (2) correspond to fully symmetric
tensors T i1 ···in . So we obtain an infinite family of irreducible representations,
indexed by a non-negative integer n. What is the dimension of such a repre-
sentation? Recall that the indices i1 to in can only take values 1 or 2, since
we are considering SU (2), and its defining representation is two-dimensional.
Thus the independent components of a fully symmetric tensor T i1 ···in are:
T 11···1 , T 21···1 , T 221···1 , ... T 22···2 .
There are exactly n + 1 of them. So we obtain an infinite family of irreducible

representations indexed by a non-negative integer n with dimensions n + 1.
This is exactly what we obtained before with the highest weight construction,
if we let n = 2j with j a half-integer. Great!
But why is it that we only have to consider tensors with upper indices for
SU (2)? Concretely, this means that we are only considering tensor products
of the defining representation 2 with itself: we do not consider tensor products
with its complex conjugate 2. Why?
Well, let us look at the defining representation 2 more closely. Recall that
it is given by the Pauli matrices (4.4.1):

1 0 1 1 0 −i 1 1 0
T1 = , T2 = , T3 = .
2 1 0 2 i 0 2 0 −1
The question is: is the group representation obtained by exponentiating the

Pauli matrices complex, real, or pseudo-real (see Section 2.12)? If it was real
or pseudo-real, this would mean that it is equivalent to its complex conjugate
representation, which would justify only looking at tensor products of 2 with
itself (since 2 would be equivalent to 2).
Lemma 5.5.2 The defining representation of SU (2) is pseudo-real.

The defining, two-dimensional, representation of SU (2) is pseudo-real.
Proof. Recall that a representation is real or pseudo-real if it is equivalent to
its complex conjugate representation. That is, there exists an invertible matrix
S such that
T (g)∗ = ST (g)S −1
for all g ∈ SU (2). It is real if S is symmetric, and pseudo-real if S is anti-
symmetric.
The defining representation of SU (2) is obtained by exponentiating a gen-
eral real linear combination of Pauli matrices:
3
X
T (g) = exp(i θk Tk ).
k=1
Then its complex conjugate is:

3
X
T (g)∗ = exp(−i θk Tk∗ ).
k=1
Thus,
T (g)∗ = ST (g)S −1
for some S if and only if
Tk∗ = −STk S −1
for the Pauli matrices Tk , k = 1, 2, 3.
One can check using matrix multiplication that Pauli matrices satisfy the
property that
Ti Tj = −Tj Ti for i 6= j, Ti2 = I.
Since T1 and T3 are real-valued, it thus follows that
T2 T1∗ T2 = T2 T1 T2 = −T1 , T2 T3∗ T2 = T2 T3 T2 = −T3 .
As for T2 , we see that T2∗ = −T2 , and thus
T2 T2∗ T2 = −T2 T2 T2 = −T2 .
Thus, if we set S = T2 , and hence S −1 = T2 , we get that Tk∗ = −STk S −1 ,

as required. Finally, since T2 is anti-symmetric, we conclude that the defining
representation of SU (2) is pseudo-real.
This is the key property that makes SU (2) special. Since its defining rep-
resentation is pseudo-real, that is, 2 ∼= 2, we do not need to consider tensor
products involving both 2 and 2, since
2 ⊗ ··· ⊗ 2 ⊗ 2 ⊗ ··· ⊗ 2 ∼
= 2 ⊗ · · · ⊗ 2.
Concretely, what this means is that we only need to consider tensors with
upper indices for SU (2).
Note that this is certainly not true for SU (N ) in general though. The
defining representation is pseudo-real only for SU (2). For N > 2, the defining
representation is complex.
5.5.4 Tensor representations of SU (3)

Let us now look at SU (3). In this case, the defining representation 3 is complex,
and we need to consider general tensor products 3 ⊗ · · · 3 ⊗ 3 ⊗ · · · ⊗ 3.
For SU (3), the statement is that all irreducible representations are given by
traceless tensors that are fully symmetric upstairs an downstairs individually.
i ···i
That is, tensors of the form Tk11 ···kpq that are traceless and fully symmetric
under permutations of the ia and the kb separately.
Thus we conclude that irreducible representations of SU (3) are indexed by
two non-negative integers (p, q), corresponding to the number of upper and
lower indices respectively. What is the dimension of such a representation?
Lemma 5.5.3 Dimension of irreducible representations of SU (3). A
i ···i
traceless tensor Tk11 ···kpq that is fully symmetric under permutations of the ia
and the kb separately has
1
(p + 1)(q + 1)(p + q + 2)
2
independent components. Thus, the dimension of the (p, q) irreducible repre-
sentation of SU (3) is 12 (p + 1)(q + 1)(p + q + 2).
i ···i
Proof. A tensor Tk11 ···kpq that is fully symmetric under permutations of the ia
and the kb separately has

1 1
(p + 1)(p + 2) (q + 1)(q + 2) (5.5.2)
2 2
independent components. We then need to impose the tracelessness condi-

tion, which will impose a number of constraints. Let us count the number of
constraints, which are independent, and then substract this number from the
number of above to get the number of independent components of a traceless
tensor that is fully symmetric upstairs and downstairs.
To impose that a tensor is traceless, we need to sum over one upper and one
lower index. We can choose any of those, since the tensor is fully symmetric
upstairs and downstairs. Thus we need to impose the conditions
3
ai ···i
X
Tak22 ···kpp = 0.
a=1
How many constraints does this give? Well, the indices i2 , . . . , ip and k2 , . . . , kp
can take any values here. So the number of constraints is the number of
independent components of a tensor with p − 1 upper indices and q − 1 lower
indices that is fully symmetric upstairs and downstairs. By (5.5.2), this is

1 1
(p)(p + 1) (q)(q + 1) . (5.5.3)
2 2
Therefore, the number of independent components of a traceless tensor

i ···i
Tk11 ···kpq that is fully symmetric under permutations of the ia and the kb sepa-
rately is (5.5.2) minus (5.5.3), that is,
1 1 1
(p + 1)(p + 2)(q + 1)(q + 2) − p(p + 1)q(q + 1) = (p + 1)(q + 1)(p + q + 2).
4 4 2

To end this section, let us list the first few irreducible representations of
SU (3) in terms of the integers (p, q). We write the representation in terms of
dimension, and give its name when appropriate.
Table 5.5.4 The first few irreducible representations of SU (3)
(p, q) Representation (name)

(0, 0) 1 (trivial)
(1, 0) 3 (fundamental)
(0, 1) 3 (anti-fundamental)
(2, 0) 6
(0, 2) 6
(1, 1) 8 (adjoint)
(3, 0) 10
(0, 3) 10
In fact, there is an interesting story here. In the early 1960s, nine short-lived
baryonic particles had been observed. They all shared similar properties and
masses. So it was believes that those should live some irreducible representation
of the symmetry group of the theory. Using the fact that 8 spin 0 mesons and 8
spin 1/2 baryons, Gell-Mann proposed that the symmetry group of the strong
force should be SU (3), and that those came in the adjoint representation of
SU (3). Then, since there is no 9-dimensional irreducible representation for
SU (3), he made the striking prediction that there should be a 10th short-lived
baryonic particle with similar properties and masses as the nine other ones,
so that they transformed in the 10-dimensional irreducible representations of
SU (3). The missing particle was soon found. Gell-Mann’s prediction is a
remarkable achivement of representation theory in particle physics!
5.6 The Standard Model of particle physics and

GUTs
Objectives
• Recognize that particles of the Standard Model can be labelled as repre-
sentations of the gauge group SU (3) × SU (2) × U (1).
• State the general idea, from the perspective of representation theory, of

Grand Unified Theories (GUTs).
We are now ready to do some cool physics and see how particles in the
Standard Model can be labeled by representations, and how the idea of Grand
Unified Theories (GUTs) then naturally arises.
5.6.1 The Standard Model of particle physics

In this section we summarize some representation-theoretic aspects of the Stan-
dard Model in particle physics. For more on the mathematical formulation of
the Standard Model, see for instance this wikipedia page.
It has been emphasized a number of times in this course that states of a
quantum mechanical systems can be labeled using irreducible representations
of the group of symmetries. In quantum field theory, these states correspond to
particles. So in the particle physics, we can label particles in terms of how they
transform under the group of symmetries, i.e., in terms of the corresponding
representations.
There are two sources of symmetries in the Standard Model of particle

physics. One is the Lorentz group (or Poincare group) corresponding to sym-
metries of spacetime. So we can label particles according to how they trans-
form under the Lorentz group. We have not seen yet the representations of
the Lorentz group, but, roughly speaking, they are similar to the representa-
tions of SO(3). In particular, this source of symmetry gives rise to the spin
of a particle, which tells us in which representation of the Lorentz group it
transforms.
Another source of symmetries is the gauge group of the Standard Model,
which is a group of transformations of the theory that leaves the physics in-
variant (such as the U (1) transformations of the electromagnetic potential in
electromagnetism). It turns out that the gauge group of the Standard Model
is SU (3) × SU (2) × U (1). The SU (2) × U (1) factor comes from the electroweak
force, while the SU (3) factor corresponds to the strong force. Therefore, parti-
cles are labeled by how they transform under this gauge group: in other words,
they are labeled by representations of SU (3), SU (2) and U (1).
Recall the irreducible representations of U (1) are all one-dimensional (since
U (1) is abelian), and take the form eikθ with θ ∈ [0, 2π) and k a non-negative
integer. So they are indexed by an integer, which we call the charge. It
is customary in the Standard Model to specify this charge as being the so-
called “weak hypercharge”, which is not quite an integer, but rather a rational
number.
So we usually specify a representation of SU (3) × SU (2) × U (1) as (p, q)n ,
where p is the p-dimensional representation of SU (3), q is the q-dimensional
representation of SU (2), and n is the weak hypercharge. With this notation,
we can list the particles of the Standard Model, with their spin (representation
of the Lorentz group) and their representation under SU (3) × SU (2) × U (1).
The following standard table is pretty much taken from wikipedia.
Table 5.6.1 The Standard Model of particle physics
Particle Name Representation
Spin 1
B Z boson (1, 1)0
W W boson (1, 3)0
G gluon (8, 1)0
Spin 1/2
qL left-handed quark (3, 2)1/3
ucL left-handed antiquark (up) (3, 1)−4/3
dcL left-handed antiquark (down) (3, 1)2/3
`L left-handed lepton (1, 2)1
`cL left-handed antilepton (1, 1)2
Spin 0
H Higgs boson (1, 2)1
What is pretty cool is that you should now understand what the notation
stands for, and be able to read this table!
A few things to note:
• The spin 1 particles, which correspond to the gauge bosons (that carry
forces), all come in the adjoint representations of SU (3), SU (2) and U (1).
This is a general feature, gauge bosons generally transform according to
the adjoint representation.
• The matter particles, i.e. those that compose matter, such as electrons,
muons, quarks, etc. all correspond to spin 1/2 particles (i.e. they are
spinors).
• The Higgs boson is the only spin 0 particle.
• However, with the recent discovery of neutrino masses and neutrino os-
cillation, it is generally believed that one more particle should be added
to the Standard Model, which would responsible for giving masses to
the neutrinos. This particle is the “right-handed neutrino”, which would
transform according to the trivial representation (1, 1)0 . The existence
of the right-handed neutrino has not been confirmed experimentally how-
ever.
This is all very cool. But, from a theoretical physics viewpoint, the represen-
tations that appear in the Standard Model appear rather random. Why are
there particles transforming in the (3, 2)1/3 representation, but not in, say, the
(3, 2)1/3 representation? Who chose this seemingly random list of representa-
tions?
One answer to this question could be that there is no answer, and that
this is just what Nature says. That is a fine answer. But often, when there
is something that seems rather random or mysterious, this means that we are
missing something. And going a little deeper into representation theory, one
sees that indeed, these representations are not random. This is the fundamental
idea behind Grand Unified Theories (GUTs).
5.6.2 SU (5) GUT

The idea of GUTs is that perhaps the gauge group of the Standard Model is
in fact larger than SU (3) × SU (2) × U (1), but is somehow broken to SU (3) ×
SU (2) × U (1) at some higher energy scale. Why would that be a good idea?
You will see!
Perhaps the simplest embedding of SU (3) × SU (2) × U (1) as a subgroup
of a larger group is
SU (3) × SU (2) × U (1) ⊂ SU (5).
This embedding can be seen in terms of the defining representations. Roughly

speaking, one constructs a subgroup of 5×5 special unitary matrices by looking
at those that are block diagonal, with a 3 × 3 block with determinant one
(SU (3)) and a 2 × 2 block with determinant one (SU (2)). You also include
diagonal matrices of the form diag(a, a, b, b, b) with a2 b3 = 1, which generates
a copy of U (1).
Now given irreducible representations of SU (5), one can restrict to the sub-
group SU (3) × SU (2) × U (1) ⊂ SU (5). Those will generally now be reducible
representations of the subgroup. One can work out how those decompose as
direct sums of irreducible representations of the subgroups. Such restriction
and decomposition are called branching rules in physics, and some textbooks
include tons of table of such branching rules (see for instance Slansky’s book
Group Theory for Unified Model Building.)
Let us then look at the simplest irreducible representations of SU (5). We
know what those are: they are the tensor representations that we constructed
earlier. Here are the first few, with how they decompose as direct sums of
irreducible representations of the subgroup SU (3) × SU (2) × U (1) ⊂ SU (5):
1 →(1, 1)0 ,
5 →(3, 1)−1/3 ⊕ (1, 2)1/2 ,
5 →(3, 1)1/3 ⊕ (1, 2)−1/2 ,

10 →(3, 2)1/6 ⊕ (3, 1)−2/3 ⊕ (1, 1)1 ,
24 →(8, 1)0 ⊕ (1, 3)0 ⊕ (1, 10 ⊕ (3, 2)−5/6 ⊕ (3, 2)5/6 .
The last representation is the adjoint representation, whose decomposition

always includes the adjoint representation of the subgroup, and extra stuff.
Now the really cool thing here is that all the seemingly random repre-
sentations corresponding to the matter content of the Standard Model (see
Table 5.6.1) are in fact the irreducible representations that appear in the de-
composition of the representation 5 ⊕ 10 of SU (5)! (Note that the weak hyper-
charge in the decomposition above has been rescaled by 2 in comparison with
Table 5.6.1, which is just a choice of convention.) Thus the representations
do not appear so random anymore. This is so nice that it cannot simply be a
coincidence.
As for the other particles of the Standard Model, the gauge bosons arise
from the decomposition of the adjoint 24 of SU (5), and the Higgs arises from
the 5 and 5. The right-handed neutrino, if observed, would need to be added
as coming from the trivial representation 1 of SU (5).
This is what motivated Georgi and Glashow to try to come up with a gauge
theory based on SU (5) that would underly the Standard Model. However, the
simplest such models run into trouble; for instance, they predict proton decay
with a lifetime that is too short to be consistent with experiments (proton decay
has never been observed). But this just means that the simple Georgi-Glashow
model cannot be correct. The beauty of the representation-theoretic unification
may appear in a different context. It is just too nice to be coincidental.
5.6.3 SO(10) GUT

In fact, one can keep going and look at further embeddings into larger gauge
groups. The next level of embedding consists in looking at the subgroup
SU (5) × U (1) ⊂ SO(10).
As you will see, things become even nicer!

As before, we look at irreducible representations of SO(10), and see how
their restriction to the subgroup decomposes as a direct sum of irreducible
representations of the subgroup. The first few non-trivial representations (we
include the spin representation 16) decompose as follows.
10 →5−2 ⊕ 52 ,
16 →101 ⊕ 5−3 ⊕ 15 ,
45 →240 ⊕ 10−4 ⊕ 104 ⊕ 10 .
Amazingly, all the matter content, including the right-handed neutrino,

arises from the decomposition of a single representation, the spin representation
16 of SO(10)! Isn’t that amazing? This cannot be a coincidence.
As for the rest of the particles, the gauge bosons as always come from the
decomposition of the adjoint 45 of SO(10), while the Higgs comes from the 10
of SO(10).
The SO(10) GUTs do not suffer from the proton decay problem of SU (5).
However, they have other issues, such as the so-called “doublet-triplet splitting
problem”.
Nevertheless, the point here is that the representation-theoretic “unifica-
tion” of the particles of the Standard Model is really quite elegant. What it
points to is that the larger groups SU (5) and/or SO(10) may perhaps play a
role in whatever theory of physics goes beyond the Standard Model. If this
is the case, it would be a tremendous achievement of representation theory in
particle physics.
5.6.4 Beyond SO(10)

Why stop at SO(10)? In fact, there is a natural chain of embeddings:
SU (5) ⊂ SO(10) ⊂ E6 ⊂ E7 ⊂ E8 ,
where E6 , E7 , E8 are exceptional Lie groups. The fundamental representation
of E6 is 27-dimensional, and it turns out that for the subgroup SO(10)×U (1) ⊂
E6 it decomposes as
27 → 161 ⊕ 10−2 ⊕ 14 .
In particular, it now gives rise to not only the matter content of the Standard
Model (161 ), but also the Higgs (10−2 ). All those unify into a single repre-
sentation of E6 ! Let me state that one more time: all the matter content of
the Standard Model, including the Higgs and the right-handed neutrino, all
combine into the fundamental representation of E6 . Wow!
As for the gauge bosons, as always, they come from the decomposition of
the adjoint of E6 .
Why then stop at E6 ? Well, now there is a good reason to stop. E6 is the
only exceptional group that has complex representations, and those are needed
to give rise to chiral fermions as in the Standard Model through standard
symmetry breaking by a mechanism of Higgs-type. So standard GUT theories
must somehow stop at E6 . It is no good to try to construct standard GUT
theories with gauge group larger than E6 . In any case, all the matter content
has already unified into a single representation, so there is no reason really to
look at larger gauge groups.
However, in string theory we can go further, and in fact we often must.
But in string theory we can break symmetries using other mechanisms, such
as string mechanisms, so we are allowed to go further. In fact, some flavours
of string theory (such as heterotic string theory) naturally come with an E8
gauge group. So it seems like this representation-theoretic unification does
take place in string theory, at least in some string models. Of course, we do
not know yet whether string theory is a valid description of Nature. But it is
quite nice that it naturally gives rise to a representation-theoretic unification
of the particles of the Standard Model.
In any case, I hope that I have convinced you in this section that represen-
tation theory is essential to understand the Standard Model of particle physics.
And that it points towards some sort of unification of the matter content into
representations of a larger gauge group, which, if physically correct, would be
a striking prediction of representation theory.
5.7 Representations of the Lorentz group

Objectives
• Construct the ordinary and spin representations of SO(4) by looking at
its spin group.
• Recognize the representations of the Lorentz group, and determine which
ones are spin representations.

So far we studied representations of SU (N ), and ordinary and spin repre-
sentations for SO(3). But the symmetry group of spacetime in special relativity
is the Lorentz group, which is composed of Lorentz transformations. Thus in
quantum field theory, which is the unique formalism that naturally combines
quantum mechanics and special relativity, we want to think of our fields in
terms of how they transform under the Lorentz group. That is, we want to
think of them in terms of representations of the Lorentz group. This is what
we study in this section.

Before we look at the Lorentz group it is instructive to look at the group
of rotations in four dimensions, SO(4). This is because the Lorentz group
SO(3, 1) is also a group of rotations in four dimensions, but with respect to
the Minkowski metric instead of the Euclidean metric. As a result, many of
the properties of representations of SO(4) carry through to the Lorentz group
SO(3, 1).
Remark 5.7.1 More precisely, in this section we focus on the component of
SO(3, 1) that is connected to the identity, which is often denoted by SO(3, 1)+ .
This is the group that is physically represented in terms of Lorentz transforma-
tions of spacetime. To avoid cluttering notation in the following we will denote
this component by SO(3, 1), but keep in mind that we always only consider its
component connected to the identity.
5.7.1.1 The spin group Spin(4)

The double cover of SO(4) is the spin group Spin(4). We know that the
representations of Spin(4) all descend to either ordinary or spin representations
of SO(4), depending on whether they assign the same matrix to the elements
of the kernel of the map Spin(4) → SO(4) (if they do, they descend to ordinary
representations; if they do not, they descend to spin representations). Thus,
to understand the representations of SO(4), we want to study representations
of Spin(4).
In the case of SO(3), we found that Spin(3) ∼ = SU (2), which was nice and
easy. It turns out that Spin(4) is also isomorphic to nice and easy group:
Spin(4) ∼= SU (2) × SU (2).
Remark 5.7.2 Note that the spin groups Spin(N ) in general are not necessar-
ily isomorphic to such nice and easy groups. But there are such isomorphisms
for small enough N . For instance, Spin(6) ∼ = SU (4).
∼
How can we see that Spin(4) = SU (2) × SU (2)? Intuitively, one can think
of it as follows. The group SU (2) is in fact the unit three-sphere S 3 as a
manifold, as we saw in Example 4.1.10. In other words, an element of SU (2)
corresponds to a point on a S 3 , that is, a point in R4 that satisfies the equation
x21 + x22 + x23 + x24 = 1.
Thus, specifying an element of the product group SU (2) × SU (2) is equivalent

to specifying two elements of SU (2), or, in other words, two points on the
three-sphere S 3 . But the group of symmetries of a three-sphere is the group
of rotations SO(4) in four dimensions. Thus, these two points will be related
by a rotation. In summary, to an element of SU (2) × SU (2), we can assign
an element of SO(4), corresponding to the rotation that brings the first point
on S 3 to the second. Finally, one sees that this mapping is not one-to-one,
since the two points U, V ∈ S 3 and the opposite points −U, −V ∈ S 3 will be
mapped to the same rotation in SO(4).
From the point of view of Lie algebras, the statement that SU (2) × SU (2)
is a double cover of SO(4) implies that they must share the same Lie algebra
∼ su(2) ⊕ su(2) (the Lie algebra of a product group is the direct sum of
so(4) =
the Lie algebras of the individual factors). Let us see this explicitly.
Lemma 5.7.3 The isomorphism between so(4) and su(2) ⊕ su(2). The
Lie algebra so(4) is isomorphic to the Lie algebra su(2) ⊕ su(2).
Proof. Following along the same lines as we did for so(3) and su(2), we can
find an explicit description of the Lie algebra so(4). It is six-dimensional, and
there is a natural choice of basis, which we denote by Ji , Ki with i = 1, 2, 3,
with the bracket:
3
X
[Ji , Jj ] =i ijk Jk , (5.7.1)
k=1
3
X
[Ji , Kj ] =i ijk Kk , (5.7.2)
k=1
3
X
[Ki , Kj ] =i ijk Jk . (5.7.3)
k=1
To see the isomorphism with the Lie algebra su(2) ⊕ su(2), we do a change of
basis. We define a new basis
1
J±,i = (Ji ± Ki ), i = 1, 2, 3.
2
It is straightforward to check that the bracket becomes, in this new basis,
[J+,i , J−,j ] =0, for all i, j = 1, 2, 3,

3
X
[J+,i , J+,j =i ijk J+,k ,
k=1
3
X
[J−,i , J−,j =i ijk J−,k .
k=1
We recognize that the J+,i and the J−,i generate two copies of the su(2) Lie
algebra. Moreover, those commute, and hence the Lie algebra is a direct sum
su(2) ⊕ su(2). We conclude that so(4) ∼
= su(2) ⊕ su(2).
5.7.1.2 Representations of SO(4)

Now that we have established that Spin(4) ∼ = SU (2) × SU (2), we can find
the irreducible representations of Spin(4). The irreducible representations of
SU (2)×SU (2) are given by tensor products T ⊗S of irreducible representations
S and T of SU (2). But from Section 5.3, we already know the irreducible repre-
sentations of SU (2). There is an infinite tower of finite-dimensional irreducible
representations, indexed by a non-negative half-integer j, of dimension 2j + 1.
Therefore, the irreducible representations of SU (2)×SU (2) are indexed by two
non-negative half-integers j1 and j2 , and have dimensions (2j1 + 1)(2j2 + 1).
We know that all these representations descend to either ordinary irre-

ducible representations of SO(4), or spin representations. We also know that
all irreducible ordinary and spin representations of SO(4) arise from irreducible
representations of Spin(4). So to conclude our study of representation theory
of SO(4), all we need to do is determine which of the representations above
descend to ordinary representations of SO(4), and which descend to spin rep-
resentations.
Theorem 5.7.4 Representations of SO(4). The finite-dimensional ir-
reducible projective representations of SO(4) are indexed by two non-negative
half-integers j1 and j2 and have dimensions (2j1 + 1)(2j2 + 1). They are ordi-
nary representations if j1 + j2 ∈ Z, and spin representations otherwise.
Proof. We will simply give a rough argument here. To determine whether a
representation of SU (2) × SU (2) descends to ordinary representation of SO(4),
we look at the kernel of the double covering SU (2) × SU (2) → SO(4). If the
given representation assigns the same matrix to all elements of the kernel, then
it descends to an ordinary representation of SO(4). If it doesn’t, it descends
to a spin representation of SO(4).
The kernel of the double covering SU (2) × SU (2) → SO(4) is Z2 . Its non-
trivial element is given by the tensor product of the two non-trivial elements
in the kernel of the mapping SU (2) → SO(3) that we studied before. In terms
of the fundamental representation of SU (2) given by 2 × 2 special unitary
matrices, the kernel of SU (2) → SO(3) was given by the two matrices {I, −I},
where I is the 2×2 identity matrix. For the mapping SU (2)×SU (2) → SO(4),
the kernel is then given by the tensor products {U ⊗ I, (−I) ⊗ (−I)}.
Let us continue with this particular representation. It corresponds to j1 =
j2 = 1/2. It is a four-dimensional representation of SU (2) × SU (2). Does it
descend to an ordinary representation of SO(4)? To the identity element it
assigns the matrix I ⊗ I, which is of course the 4 × 4 identity matrix. To the
non-trivial element in the kernel, it assigns the 4 × 4 matrix given by the tensor
product
(−I) ⊗ (−I) = I ⊗ I,
which is again the 4 × 4 identity matrix. Therefore, it descends to an ordinary
representation of SO(4): it is in fact the fundamental representation of SO(4)
(which is interestingly built by taking the tensor product of two copies of the
spin-1/2 representation of SU (2)).
In general, the non-trivial element of the kernel is given by the matrix
(1) (2)
obtained as the tensor product e2πiT3 ⊗ e2πiT3 , where we use the notation of
Section 5.4. Recall that for a given irreducible representation of SU (2) indexed
by a half-integer j, the matrix T3 is a diagonal matrix with entries that are half-
integers (but not integers). Thus e2πiT3 becomes minus the identity matrix. If
j is an integer, then T3 is a diagonal matrix with integer entries, and hence
e2πiT3 is the identity matrix. Therefore, since minus signs cancel in the tensor
product, we conclude that a representation of SU (2) × SU (2) descends to an
ordinary representation if and only if j1 + j2 is an integer. It is spin otherwise.

In the case of SO(3), we showed in Lemma 5.5.2 that the 2-dimensional spin
representation of SO(3), which descends from the fundamental representation
of Spin(3) ∼ = SU (2), is pseudo-real. For SO(4), we have found two lowest-
dimensional spin representations, the (1/2, 0) and (0, 1/2), which also have
dimension 2. One can ask: are those complex conjugate of each other, or are
they both independently real or pseudo-real?
We will not prove this here, but it turns out that for SO(4), the two 2-
dimensional spin representations are pseudo-real. In fact, there is a nice clas-
sification result about spin representations of SO(N ) in general. One needs to

distinguish between SO(2N ) and SO(2N + 1). For SO(2N ), there are always
two lowest-dimensional spin representations, and they have dimension 2N −1 .
For SO(2N + 1), there is only one lowest-dimensional spin representation, and
it has dimension 2N −1 . The type of these spin representations (complex, real
or pseudo-real) is summarized in the table below. It turns out that the pattern
is periodic, with periodicity eight. So, if we write SO(8k +m) for some integers
k ∈ Z and 0 ≤ m ≤ 7, then the type of spin representations depends only on
m.
Table 5.7.5 Spin representations for SO(8k + m)
m 0 1 2 3 4 5 6 7
Type Real Real Complex Pseudo-real Pseudo-real Pseudo-real Complex Real
To end this section, let us check that the table is consistent with what we
have found so far. SO(3) corresponds to k = 0 and m = 3, which says that
the one 2-dimensional spin reprepresentation is pseudo-real, which is what we
have found. For SO(4), we have k = 0 and m = 4, and hence from the table
we conclude that the two 2-dimensional spin representations are pseudo-real,
which is indeed what we stated above. Great!
5.7.2 Representation theory of the Lorentz group

This is all very nice, but what does it have to do with the Lorentz group
SO(3, 1)? The Lorentz group is certainly more complicated. For instance,
it is non-compact, which implies that it has no non-trivial finite-dimensional
unitary representations. Fortunately, it is however still semi-simple, so all
representations are equivalent to direct sums of irreducible representations. So
we only need to care about irreducible representations. In any case, at the
level of Lie algebras, the irreducible representations of SO(3, 1)are indexed in
a way very similar to those of SO(4). Let us see why.
The Lie algebra so(4) was presented in (5.7.3). It turns out that the Lie
algebra so(3, 1) of the Lorentz group is very similar. It is also six-dimensional,
and can be written in terms of generators Ji , Ki with i = 1, 2, 3 as
3
X
[Ji , Jj ] =i ijk Jk , (5.7.4)
k=1
3
X
[Ji , Kj ] =i ijk Kk , (5.7.5)
k=1
3
X
[Ki , Kj ] = − i ijk Jk . (5.7.6)
k=1
The only difference with (5.7.3) is the sign in the bracket [Ki , Kj ]. This sign
is however quite important.
To determine the representations of SO(3, 1), we want to construct its
double cover, as we did for SO(4). Let us first do it at the level of Lie algebras.
As in (5.7.3), we want to find a change of basis to construct an isomorphism
of Lie algebras.
The key is to realize that if we do a change of basis Kj 7→ iKj , the brackets
in (5.7.6) become exactly equal to those in (5.7.3). But we are only allowed to
do such a change of basis if we consider the “complexified Lie algebra”, that
is, we consider the complex vector space with basis given by the Ji , Ki . We
denote this complex Lie algebra as so(3, 1)C . Thus what we have found is that
the complexified Lie algebra so(3, 1)C and so(4)C are isomorphic. In particular,
doing the change of basis as for (5.7.3), we conclude that
so(3, 1)C ∼
= su(2)C ⊕ su(2)C .
But in the end we want to work with the real Lie algebra so(3, 1). So we want
to turn the right-hand-side into the complexification of a single Lie algebra
(instead of direct sum of complex Lie algebras).
To do so we proceed through a sequence of isomorphisms of Lie algebra.
The key is the isomorphism su(2)C = ∼ sl2 (C), which can be checked from the
commutation relations of those. Then we get:
so(3, 1)C ∼
=su(2)C ⊕ su(2)C (5.7.7)
∼
=sl2 (C) ⊕ sl2 (C) (5.7.8)
∼
=sl2 (C) ⊕ isl2 (C) (5.7.9)
∼
=sl2 (C)C , (5.7.10)
where in the last line we mean the complexification of the real Lie algebra
sl2 (C) associated to the Lie group SL(2, C).
As a result, we get so(3, 1)C ∼
= sl2 (C)C , which is an isomorphism between
the complexifications of two Lie algebras. Upon restriction to the real Lie
algebras this becomes the isomorphism
so(3, 1) ∼
= sl2 (C).
Thus the Lie groups SO(3, 1) and SL(2, C) share the same Lie algebra.
It turns out that SL(2, C) is simply connected, and is in fact the universal
cover of SO(3, 1). Moreover, the map SL(2, C) → SO(3, 1) is a double cover-
ing, and hence Spin(3, 1) ∼= SL(2, C). We have found the spin group Spin(3, 1)!
Thus all finite-dimensional irreducible projective representations of SO(3, 1)
descend from irreducible finite-dimensional representations of SL(2, C).
What are the irreducible representations of SL(2, C)? Well, it is easier to
work at the level of Lie algebras. I will be rather sketchy here for brevity. The
isomorphisms (5.7.10) are very useful. First, it turns out that the complex rep-
resentations of the complexification sl2 (C)C are in one-to-one correspondence
with the real representations of sl2 (C). Thus, the latter are in one-to-one cor-
respondence with the complex representations of su(2)C ⊕ su(2)C . But those
are equivalent to the representations we constructed earlier when we studied
SO(4). They are indexed by two half-integers j1 and j2 and have dimensions
(2j1 + 1)(2j2 + 1). Therefore, the same is true of the real irreducible represen-
tations of SL(2, C).
It turns out that as for SO(4), the representations with j1 + j2 ∈ Z descend
to ordinary irreducible representations of SO(3, 1), while the other ones are
spin. The (1/2, 0) and (0, 1/2) are the two 2-dimensional spin representations;
the (1/2, 1/2) is the four-dimensional representation (which corresponds to how
a four-vector in spacetime transforms).
As for SO(4), we can ask whether the spin representations are real, pseudo-
real, or complex. It turns out that for SO(3, 1), the two 2-dimensional spin
representations are in fact complex conjugate of each other. This is one key
difference with SO(4), in which case they were pseudo-real.
In fact, the table Table 5.7.5 can be generalized for arbitrary special or-
thogonal groups SO(p, q). The type of spin representations is again periodic
with periodic eight, but it now depends on p − q mod 8. We get the following
generalized table, which reduces to Table 5.7.5 for q = 0.
Table 5.7.6 Spin representations for SO(p, q)
p − q mod 8 0 1 2 3 4 5 6 7
Type Real Real Complex Pseudo-real Pseudo-real Pseudo-real Complex Real
5.8 Unitary representations of the Poincare group

Objectives
• Sketch how unitary irreducible representations of the Poincare group are
constructed using little groups.
• State Wigner’s classification of non-negative energy unitary irreducible
representations of the Poincare group.
In the previous section we studied finite-dimensional representations of the
Lorentz group. As we saw, because the Lorentz group is non-compact, all
those representations are not unitary, since there is not non-trivial unitary
finite-dimensional representation. But in quantum mechanics and in particle
physics in general, we are interested in unitary representations. The reason
is that the representations act on the Hilbert space of states, i.e. they tell
us how wave-functions transform. Since the norm square of the wavefunction
gives the probability that the system is observed in a given state, we want the
representations to preserve the norm square, or, in other words, to preserve
the inner product. That is, we want unitary representations.
Moreover, the symmetry group of spacetime is larger than the Lorentz
group: it also includes translations. This is the so-called Poincare group (also
called “inhomogeneous Lorentz group” in older references). Putting this to-
gether, in particle physics we can in fact identify particles by how they trans-
form under the group of symmetries. So we think of particles as irreducible
unitary representations of the Poincare group. More precisely, since wave-
functions are only defined up to scale, we allow irreducible unitary projective
representations. This is what we study in this section.
5.8.1 The Poincare group

Let us start by recalling the main features of the Poincare group. The Poincare
group acts on a four-dimensional spacetime with coordinates xµ as
4
X
xµ 7→ (x0 )µ = Lµν xν + aµ ,
ν=1
where Lµν are the components of a Lorentz transformation. Thus the Poincare
group includes both Lorentz transformations and translations.
We can extract the Lie algebra of the Poincare group as usual. It is 10-
dimensional. A convenient choice of basis uses the generators Jµν = −Jνµ for
the Lorentz transformations, and four generators Pµ for the translations. The
commutation relations can be written as
[Pµ , Pν ] =0,
[Jµν , Pρ ] = − i(ηµρ Pν − ηνρ Pµ ),
[Jµν , Jρσ ] = − i(ηµρ Mνσ − ηµσ Mνρ − ηνρ Mµσ + ηνσ Mµρ ),
where ηµν are the components of the Minkowski metric (i.e. the diagonal
matrix diag(−1, 1, 1, 1)).
Remark 5.8.1 The generators Jµν for Lorentz transformations are connected
to the generators of rotation Ji and boosts Ki introduced in (5.7.6) by
3
1 X
Ji = imn J mn , Ki = Ji0 ,
2 m,n=1
where we raised indices using the Minkowksi metric as usual.
5.8.2 Wigner’s classification

The goal of this section is to classify unitary representations of the Poincare
group. First, since the group is non-compact, we know that the only finite-
dimensional unitary representation is the trivial representation. So all non-
trivial unitary representations are infinite-dimensional.
The classification of unitary representations of the Poincare group is known
as Wigner’s classification, since it was originally achieved in Wigner’s fa-
mous paper “On unitary representations of the inhomogeneous Lorentz group”.
This is a very nice paper in fact, have a look!
More precisely, Wigner was interested in representations that are physical,
and may correspond to physical particles. Thus, what he classified are irre-
ducible unitary representations that have non-negative, real, mass (m ≥ 0) -
negative mass representations are not expected to be physical. We will see
in a second how mass arises from the point of view of representations of the
Poincare group.
The method that he used is the method of induced representations, also
known as the method of little groups in physics. In this approach, one
constructs the representations of a group starting from the representations of
a subgroup. We will not explain how the method of induced representations
works in these notes, but will simply go through the main steps in the context
of the Poincare group. We will be very hand-wavy, but I suppose that’s ok.
For more information one could for instance refer to these lecture notes.
The starting point is to consider the subgroup of the Poincare group con-
sisting of translations. From the point of view of Lie algebras, we consider the
subspace generated by the translation generators Pµ . Those generators com-
mute, and hence can be simultaneously diagonalized. We pick an eigenvector
|pi, with (real) eigenvalues pµ :
Pµ |pi = pµ |pi.
Remark 5.8.2 We note in passing P4that the Poincare algebra has two Casimir
invariants. One of them is P 2 = µ=1 Pµ P µ , and the other is constructed as
P4
W 2 = µ=1 Wµ W µ with
4
1 X
Wσ = − µνρσ J µν P σ .
2 µ,ν,σ=1
Wσ is known as the Pauli-Lubanski pseudovector. In any case, since those

are Casimir invariants, we can label the irreducible representations by the
eigenvalues of these
P4 operators. In particular, the eigenvalue of P 2 on the state
µ
|pi is −m := µ=1 pµ p . What this means is that the “mass squared” m2 is
2
one of the label for representations of the Poincare group. Thus mass appears
naturally in this context; it is one of the label that specifies the representations
of the Poincare group.
In any case, going back to the classification, we fix an eigenvector |pi, with
eigenvalues pµ . The idea of the method of little groups is to now consider
the subgroup of the Poincare group that leaves pµ invariant: this is called the
little group (in mathematics, it is called the “stabilizer subgroup”). The idea
of the method is to induce unitary irreducible representations of the whole
group from the unitary irreducible representations of the little group. In other
words, we classify the states with a fixed momentum pµ in terms of the unitary
irreducible representations of the little group.
Concretely, we need to separate the classification into three cases, depend-
ing on the choice of starting eigenvector |pi:
1. Positive mass m > 0, in which case we choose |pi to have eigenvalues pµ =
(m, 0, 0, 0) (that is, we use Lorentz transformations to bring ourselves to
the rest frame of the particle);
2. Zero mass m = 0 but with non-zero pµ , in which case we choose |pi to
have eigenvalues pµ = (p, 0, 0, p), p > 0;
3. Zero mass m = 0 and zero eigenvalues pµ = (0, 0, 0, 0).

The next step is to determine the little group for each of these cases, and
classify the irreducible representations of the little groups.
5.8.2.1 Zero mass and zero momentum

Let us start with the case m = 0 and pµ = (0, 0, 0, 0). In this case, the little
group that leaves pµ fixed is the whole Lorentz group. But we know that
its only finite-dimensional unitary representation is the trivial representation.
Thus what this case corresponds to is the trivial representation for the Poincare
group. We call a state that transforms according to that representation the
vacuum, since it is invariant under all symmetries of the Poincare group.
5.8.2.2 Positive mass

Next, let us consider the case where m > 0 and we are in the rest frame of the
particle, that is, pµ = (m, 0, 0, 0). In this case, the little group is the subgroup
of the Poincare group that leaves (m, 0, 0, 0) invariant. It turns out that this
is given by the group of rotations in three-dimensional space, that is, SO(3).
Nice! Because we already know the irreducible unitary projective represen-
tations of SO(3). We know that there is an infinite family of them, indexed by
a non-negative half-integer j called the spin, and with dimensions 2j + 1. We
know that when j ∈ Z we get an ordinary representation, while when j ∈ / Z
we get a spin representation.
Thus, massive particles are classified by an irreducible representation of
SO(3), identified by its spin j, which is half-integer.
5.8.2.3 Zero mass with non-zero pµ

Let us now do the case of massless particles, but with non-zero pµ . We choose
a frame such that pµ = (p, 0, 0, p).
The little group here that fixes (p, 0, 0, p) is not so obvious to see. But it
turns out to be given by the special Euclidean group SE(2), which consists
of rotations and translations in two dimensions (with Euclidean signature).
Thus, to complete Wigner’s classification, we need to understand the unitary
representations of SE(2).
The Lie algebra se(2) associated to the special Euclidean group SE(2) is
three-dimensional. One generator J is the generator of infinitesimal rotations,
with the two other generators P1 and P2 being generators of infinitesimal trans-
lations. The commutation relations are:
[J, P1 ] = iP2 , [J, P2 ] = −iP1 , [P1 , P2 ] = 0.
We are interested in irreducible unitary representations of SE(2).

Since SE(2) is non-compact, all its non-trivial unitary representations are
infinite-dimensional. But they come in two types: so-called “finite spin repre-
sentations” and “continuous spin representations”.
The easiest way to construct representations of SE(2) is to in fact use
the method of induced representations again. In this case, we consider the
subspace of the Lie algebra generated by translations, and pick an eigenstate
|ki with eigenvalues ki under translations:
Pi |ki = ki |ki, i = 1, 2.
We then consider the two cases with either ki = 0 or ki 6= 0.

In the ki = 0 case, the little group is the group of rotations SO(2). Thus
irreducible unitary representations of SE(2) in this case are induced from the
unitary irreducible representations of SO(2), which are all one-dimensional,
and indexed by an integer h which we call the helicity. In fact, since we also
allow spin representations (more precisely, the little group should have been
the double cover of SE(2)), we allow the helicity h to be a half-integer.
In the case where ki 6= 0, we choose a frame such that ki = (k, 0). The
little group is SO(1), which is trivial. In fact, this case gives rise to so-called
“continuous spin” (or “infinite spin”) representations. There are two of them
here. We will not comment further on this, since those do not appear to be
physical.
5.8.2.4 The end result of Wigner’s classification

The end result of the classification is rather remarkable. What it says is that
unitary irreducible representations of the Poincare group, which we call “par-
ticles” in physics, are indexed by two parameters: a continuous parameter m,
that we call the “mass of the particle”, and a discrete parameter j ∈ 12 Z or
h ∈ 12 Z, which we call the “spin” or “helicity” of a particle. More precisely,
focusing on the case of non-negative mass, particles come in two kinds (we
omit the vacuum case here and the continuous spin representations):
• Massive particles, which are indexed by a positive real number m > 0

(the mass), and a half-integer j (the spin).
• Massless particles, which are indexed by a zero mass (m = 0), and a
half-integer h (the helicity).
Isn’t that great? This is why it makes sense to talk about the mass and spin
(or helicity) of particles: that’s because those index the unitary irreducible
representations of the Poincare group! Moreover, the distinction between mas-
sive and massless particles which is crucial in physics appear naturally here in
terms of representations. Beautiful!
5.9 Classification of simple Lie algebras

Objectives
• Sketch the main ideas behind the classification of simple Lie algebras.
• State the result of the classification of simple Lie algebras.
In this section we will sketch the main ideas behind one of the most funda-
mental result in Lie theory, the classification of simple Lie algebras.
5.9.1 Simple Lie algebras

Recall the definition of simple groups (see Definition 1.9.5): a group is simple
if it has no non-trivial (or proper) normal subgroups. It turns out that there
is a similar concept of a simple Lie algebra, which we now define.
Definition 5.9.1 Lie subalgebra. Let g be a Lie algebra. A Lie subalgebra
is a subspace h ⊂ g that is closed under the bracket. That is, for all X, Y ∈ h,
[X, Y ] ∈ h. ♦
Definition 5.9.2 Invariant Lie subalgebra. A Lie subalgebra h ⊂ g
is invariant if for all X ∈ h and all Y ∈ g, [X, Y ] ∈ h. In mathematical
terminology, this is also called an ideal in the Lie algebra g. ♦
The notion of invariant subalgebras is closely connected to the notion of
normal subgroups. Indeed, the definition is pretty much the same, with the
operation of conjugation for groups replaced by the bracket operation for al-
gebras. Concretely, one can show that invariant Lie subalgebras give rise to
normal subgroups of Lie groups by exponentiation.
Definition 5.9.3 Simple Lie algebra. A Lie algebra is simple if it has no
non-trivial invariant Lie subalgebras. It is semisimple if it is a direct sum of
simple Lie algebras. ♦
This is the algebra counterpart to the notion of simple groups. It turns out
that simple Lie algebras are very nice, and in fact can be fully classified. This
is what we now turn to.
5.9.2 Construction of the classification

To classify simple Lie algebras one needs to introduce a number of concepts
that we have not discussed yet. We will be very brief here and only sketch
the construction. The starting point is a generalization of the highest weight
construction of irreducible representations of su(2) to general Lie algebras. It
turns out that the adjoint representation plays a special role in this construc-
tion, and in fact knowing the adjoint representation is sufficient to recover
the whole Lie algebra (as expected, since the adjoint is determined by the
structure constants). In the end, we use an explicit description of the adjoint
representation in terms of weights to classify all simple Lie algebras.
Let us start by introducing a few concepts. Recall that the starting point
of the highest weight construction for su(2) was to diagonalize one of the gen-
erators of the Lie algebra. In general, we want to simultaneously diagonalize
as many generators as possible. For this to be possible, these generators must
commute. We are led to the following definition:
Definition 5.9.4 Cartan subalgebra. Let g be a Lie algebra. The Cartan

subalgebra h ⊂ g is the Abelian subalgebra spanned by the largest set of
commuting Hermitian generators Hi , i = 1, . . . , m. We call m the rank of the
Lie algebra g. ♦
Note that by definition, the generators Hi of the Cartan subalgebra satisfy
Hi = Hi† , [Hi , Hj ] = 0.
Remark 5.9.5 One should not confuse the dimension of a Lie algebra g with
its rank. The dimension of g is the dimension of the vector space, while the
rank of g is the dimension of its Cartan subalgebra. For instance, su(2) has
dimension 3, while its rank is 1, since it has only one Cartan generator (see
Section 5.3).
Example 5.9.6 Cartan subalgebra of su(3). Let us first extract the Lie
algebra su(3). In general, it is straightforward to show that the Lie algebra
su(N ) is the algebra of N × N traceless Hermitian matrices, with bracket
given by the standard matrix commutator (more precisely, this defines the
fundamental representation of su(N )).
In the case of su(3), the dimension of the space of traceless 3 × 3 Hermitian
matrices is 32 − 1 = 8. We can write down an explicit basis for 3 × 3 traceless
Hermitian matrices. This is given by the so-called Gell-Mann matrices:
     
0 1 0 0 −i 0 1 0 0
λ1 = 1 0 0 , λ2 =  i 0 0 , λ3 = 0 −1 0 , (5.9.1)
0 0 0 0 0 0 0 0 0
   
0 0 1 0 0 −i
λ4 = 0 0 0 , λ5 = 0 0 0  , (5.9.2)
1 0 0 i 0 0
     
0 0 0 0 0 0 1 0 0
1
λ6 = 0 0 1 , λ7 = 0 0 −i , λ8 = √ 0 1 0  . (5.9.3)
3
0 1 0 0 i 0 0 0 −2
It is conventional to define the generators of su(3) in terms of the Gell-Mann

matrices as Li = 21 λi . The commutation relations then give
8
X
[Li , Lj ] = i cijk Lk ,
k=1
with the non-vanishing structure constants given by (with other non-vanishing

ones related by symmetry):
c123 =1,
1
c147 = − c156 = c246 = c257 = c345 = −c367 = ,
√ 2
3
c458 =c678 = .
2
Looking at the Gell-Mann matrices, we see that there are two diagonal
generators with real eigenvalues: L3 and L8 . We conclude that these generate
the Cartan subalgebra of su(3). Thus, while the dimension of su(3) is 8, its
rank is 2. In general, the dimension of su(N ) is N 2 − 1 while its rank is N − 1.

We want to continue mimicking the highest weight construction for su(2).
We construct representations of g by constructing the vector space on which
they act. We take a basis for that vector space consisting of eigenvectors of the
simultaneously diagonalizable Cartan generators. To each such basis vector
|αi is associated a vector α = (α1 , . . . , αm ) ∈ Rm of eigenvalues of this basis
vector under the action of the Cartan generators H1 , . . . , Hm . That is,
Hi |αi = αi |αi.
Definition 5.9.7 Weight vector. The weight vectors of a representation
Γ of a Lie algebra g are the vectors of eigenvalues of the eigenvectors of the
Cartan generators. These vectors live in Rm , where m is the rank of g. We
call Rm the weight space of g. ♦
Remark 5.9.8 Note that the weight space is the same for all representations
of a Lie algebra g, but the weight vectors change. In fact, one can think of a
representation as being determined by a set of weights in weight space.
This is indeed the natural generalization of the highest weight construction
for su(2). In this case, there was only one Cartan generator, so the weight space
was one-dimensional, i.e. it was just R. The weight vectors of a representation
were given by the half-integers −j, −j+1, . . . , j−1, j ∈ R, for some non-negative
half-integer j.
Example 5.9.9 The weights of the fundamental representation of
su(3). Going back to su(3), with generators given by the Gell-Mann matrices
(see (5.9.3), we can extract the weights of the representation by calculating
 
1
the eigenvalues of the Cartan generators for the basis vectors e1 = 0,
0
   
0 0
e2 = 1, e3 = 0 on which the representation acts. We get
0 1
1 1
T3 e1 = e1 , T8 e1 = √ e1 ,
2 2 3
1 1
T3 e2 = − e2 , T8 e2 = √ e2 ,
2 2 3
1
T3 e3 =0, T8 e3 = − √ .
3
1
So the weight vectors of the fundamental representation are ( 12 , 2√ 3
), (− 12 , 2√
1
3
)
1 2
and (0, − 3 ), which live in the weight space R .
√
It turns out that the adjoint representation plays a very special role.
Definition 5.9.10 Root. The roots of a Lie algebra g are the non-zero
weights of its adjoint representation. ♦
The roots play an important role in the highest weight construction of
representations. First, one can show that if α is a root, then −α is also a root.
So roots come in pairs ±α. It turns out that the generators of the Lie algebra
that are not generators of the Cartan subalgebra can also be grouped in pairs
E±α , which play the role of raising and lowering operators. Indeed, one can
show that if |µi is an eigenvector of the Cartan generators Hi , i = 1, . . . , m
with eigenvalue µ = (µ1 , . . . , µm ), then
Hi E±α |mui = (µi ± αi )E±α |µi.
In other words, E±α |µi is also an eigenvector of the Cartan generators, but
with eigenvalues raised or lowered by the value of the corresponding root α.
In this way, we can define the notion of highest weight vector, and construct
representations using the raising and lowering operators associated to the roots
of the Lie algebra.
In fact we can also use the result above to determine the roots. If we
can rearrange the remaining generators of the Lie algebra into pairs of raising
and lowering operators E±α , when we can determine the roots by evaluating
[Hi , E±α ] = ±αi E±α . The αi give the components of the root vector associated
to this pair of lowering and raising operators. Note that to calculate these
commutators, we can use whatever representation we want. So in this way,
if we can identity the pairs of raising and lowering operators, we can obtain
the roots without ever writing down the adjoint representation, for instance
by calculating commutators using the simpler fundamental representation.
Example 5.9.11 Roots of su(3). One way to find the roots of su(3) is to
write down the 8 × 8 matrices corresponding to the Cartan generators T3 and
T8 in the adjoint representation, and extract their eigenvalues. Or, we can use
our knowledge of the Gell-Mann matrices (5.9.3) and deduce the three pairs
of raising and lowering operators. Indeed, (T1 , T2 ), (T4 , T5 ) and (T6 , T7 ) form
pair of Pauli matrices embedded in 3 × 3 matrices. So the pairs of raising and
lowering operators for su(3) should be:
1
E±α(1) = √ (T1 ± iT2 ),
2
1
E±α(2) = √ (T4 ± iT5 ),
2
1
E±α(3) = √ (T6 ± iT7 ).
2
Then we can calculate the commutators to determine the roots α(1) , α(2) , α(3) .
First, for α(1) , the commutator with T3 is just the same as for su(2), so we
get:
[T3 , E±α(1) ] = ±E±α(1) ,
(1)
which means that α1 = 1. Furthermore, T8 commutes with T1 and T2 , and
hence
[T8 , E±α(1) ] = 0,
(1)
that is α2 = 0. Thus the first two roots are ±(1, 0).
For α(2) and α(3) we can calculate the commutator explicitly using the
fundamental representation (Gell-Mann matrices). We get:
√
1 3
[T3 , E±α(2) ] = ± E±α(2) , [T8 , E±α(2) ] = ± E±α(2) ,
2 √2
1 3
[T3 , E±α(3) ] = ± E±α(3) , [T8 , E±α(3) ] = ∓ E (3) .
2 2 ±α
√ √
Thus the remaining four roots are ±( 12 , 3
2 ) and ±( 12 , − 3
2 ). The resulting
root system is shown in Figure 5.9.12.
Figure 5.9.12 The root system for su(3) (taken from wikipedia).

But our goal here is not to construct representations, but to classify simple
Lie algebras. It turns out that the roots are key for this as well. Indeed,
the roots tell us not only something about the adjoint representation, but in
fact they can be used to reconstruct the whole algebra itself. This is perhaps
expected, since the adjoint representation is defined in terms of the structure
constants, and hence should package the information content of the Lie algebra
itself.
To see how this goes we need a few more definitions.
Definition 5.9.13 Positive and simple root. A positive weight is a
weight of a representation such that the first non-zero component is positive.
A positive root is a positive weight for the adjoint representation. A simple
root is a positive root that cannot be written as a linear combination of other
positive roots with all coefficients being positive. ♦
One can show that knowing the simple roots is in fact sufficient to determine
all roots of a Lie algebra, because of the symmetries of a root system. The
simple roots in fact form a basis for Rm , so, in particular, there are m simple
roots, where m is the rank of the Lie algebra.
Example 5.9.14 Simple roots for su(3).√ In Example 5.9.11 √
we determined
that the roots of su(3) are ±(1, 0), ±( 21 , 23 ) and ±( 21 , − 23 ). The positive
√ √
roots are then (1, 0), ( 12 , 23 ) and ( 12 , − 23 ). Since
√ √
1 3 1 3
(1, 0) = ( , ) + ( ,− ),
2 2 2 2
√
we see that (1, 0) is not a simple root. The other two positive roots, ( 12 , ± 3
2 ),
are the simple roots of su(3).
5.9.3 Root systems

The discussion of the preceding section shows that to a Lie algebra g, we
can associate a bunch of roots living in weight space Rm by looking at the
eigenvalues of the adjoint representation. These roots form what is called a
root system. It turns out that root systems are very constrained. Random
bunch of vectors in Rm certainly do not form root systems for some Lie algebra.
It turns out that root systems can be defined abstractly, independently of Lie
algebras, which we now do. In fact, we note here that root systems appear in
many different contexts in mathematics, not just in the context of Lie theory.
Definition 5.9.15 Root system. A root system Φ in Rm is a finite number

of vectors (roots) in Rm such that:
• The roots span Rm .
• For any root α ∈ Φ, the only scalar multiple of α that is also in Φ is −α.
• For every α ∈ Φ, Φ is closed under reflections through the hyperplane
perpendicular to α. This can be rephrased as the condition that for any
two α, β ∈ Φ, we must have that
α·β
σα (β) = β − 2 ∈ Φ.
α·α
• For any two α, β ∈ Φ,

α·β
2 ∈ Z.
α·α
♦
Note that one can construct root systems by “putting together” smaller root
systems. Two roots systems can be combined into a bigger one by considering
the spaces Rm1 and Rm2 as mutually orthogonal subspaces of a bigger space
Rm1 +m2 . Then the two roots systems form a bigger root system in Rm1 +m2 .
We say that root systems that cannot be obtained in this way are irreducible.
In other words, a root system Φ is irreducible if it cannot be partitioned into
two roots systems Φ = Φ1 ∪ Φ2 such that α · β = 0 for all α ∈ Φ1 and β ∈ Φ2 .
The crucial result in our context is that there is a one-to-one correspondence
between simple Lie algebras and irreducible roots systems (up to isomorphisms).
This is absolutely key. We have already seen how, given a Lie algebra g, we can
calculate its roots, which are the weights of its adjoint representation. Then
one can show that for any simple Lie algebra, these roots form an irreducible
root system, according to the definition above. So we know that we can assign
a unique irreducible root system to a simple Lie algebra.
What needs to be shown is the reverse direction. First, we must show that
starting with an irreducible root system, we can assign at most one associated
simple Lie algebra. In other words, two non-isomorphic simple Lie algebras
cannot share the same irreducible root system. This is not so trivial, but
intuitively, it is expected, since two non-isomorphic simple Lie algebras cannot
have the same structure constants, and hence cannot have the same roots.
Second, we must show that to any irreducible root system one can assign at
least one simple Lie algebra. In other words, that any irreducible root system
arises as the root system of a simple Lie algebra. As we will see, this is clear
for the four infinite families of irreducible root systems, since those arise as
root systems of basic matrix Lie groups. But it is not so obvious to prove for
the five exceptional root systems.
5.9.4 Classification of simple Lie algebras

The fact that simple Lie algebras are in one-to-one correspondence with irre-
ducible root systems (up to isomorphisms) imply that to classify simple Lie
algebras, we only need to classify irreducible root systems, which is a nice com-
binatorial problem. Root systems are in fact nicely encapsulated into so-called
Dynkin diagrams.
We construct the Dynkin diagram associated to a root system Φ as follows.
We assign a node to each of the simple roots. Then we draw edges between
nodes according to the angle between the roots. The definition of a root system
implies that the only possible angles between root vectors are π/2, 2π/3, 3π/4
and 5π/6. We draw and edge between nodes as follows:
• No edge if the root vectors are perpendicular (π/2);
• A single edge if the angle between the root vectors is 2π/3;
• A double edge if the angle between the root vectors is 3π/4;

• A triple edge if the angle between the root vectors is 5π/6.
Moreover, in the last two cases, one can show that the two root vectors cannot
have equal length. Thus we draw an arrow pointing towards the shorter vector.
Example 5.9.16 Dynkin diagram for su(3). √
Going back to su(3), we found
that the two simple roots are given by ( 12 , ± 23 ). So the Dynkin diagram will
have two nodes. The angle between the two simple roots is 2π/3, and hence
the two nodes are connected by a single edge. The Dynkin diagram of su(3) is
shown in Figure 5.9.17.
Figure 5.9.17 The Dynkin diagram for su(3).

Using Dynkin diagrams, one can classify all possible irreducible root sys-
tems, and hence all simple Lie algebra (up to isomorphisms). The classification
is shown in Figure 5.9.18. It turns out that there are four infinite families, usu-
ally denoted by An , Bn , Cn and Dn , and five exceptional cases, denoted by G2 ,
F4 , E6 , E7 and E8 .
Figure 5.9.18 Classification of irreducible root systems in terms of Dynkin

diagrams (taken from wikipedia).
The infinite families correspond to the root systems of well-known simple
Lie algebras. We have that An ∼ = su(n + 1), Bn ∼= so(2n + 1), Cn ∼= sp(2n),
∼
and Dn = so(2n). The only family that we have not studied in this class
is Cn ∼
= sp(2n), which corresponds to the Lie algebra associated to 2n × 2n
symplectic matrices.
The five exceptional root systems also correspond to simple Lie algebras,
but those are not realized in terms of matrices with simple properties (such
as orthogonal, unitary, symplectic, etc.) But these exceptional Lie algebras
are common in physics: for instance, in one flavour of string theory (so-called
“heterotic string theory”), the starting gauge group of the theory is E8 × E8 ,
where E8 is the Lie group whose Lie algebra has root system given by the E8
exceptional case.
Remark 5.9.19 Looking at Figure 5.9.18, we see that the infinite families An
and Dn , as well as the exceptional cases E6 , E7 , E8 are special because they
only have single edges. We call such diagrams simply laced. Those Dynkin
diagrams arise often in many different contexts in mathematics. Whenever
these diagrams arise, we refer to their classification in mathematics as an ADE
classification.

Notes Group Theory in Physics

Uploaded by

Notes Group Theory in Physics

Uploaded by

MA PH 464 - Group Theory in

March 30, 2020

Symmetry is the language of modern physics. When confronted with a phys-

• Anthony Zee, Group Theory in a Nutshell for Physicists, Princeton Uni-

• Eugene A. Lim, lecture notes on Symmetry in Physics.

1 Basic concepts of group theory 1

3.4 The Lagrangian . . . . . . . . . . . . . . . . . . 78

4 Lie groups and Lie algebras 82

5 Representation theory of Lie groups 103

Basic concepts of group the-

• Show in particular examples that the group axioms are satisfied.

• Identity: There exists a group element e ∈ G, known as the identity, such

It is easy to show a few elementary properties of groups:

• Determine the cyclic subgroup generated by an element of a group.

Example 1.2.2 Rotations in 3D around a fixed axis. Consider only

subgroup of invertible linear transformations of Rn that preserve the Euclidean

(u∗ )T v = u∗1 v1 + u∗2 v2 + . . . + u∗n vn .

Thus, just as for orthogonal transformations, it is easy to see that unitary

u · v = −u1 v1 − . . . − up vp + up+1 vp+1 + . . . + un vn ,

1.2.3 A few universal subgroups

[a, b] = a−1 · b−1 · a · b.

If a and b commute, then [a, b] = e, thus we can think of the commutator as

1.3 Direct product

3. Every g ∈ G can be written as g = h1 h2 for some h! ∈ H1 and h2 ∈ H2 .

be given a group structure by defining the operation

(g, h) ? (g 0 , h0 ) = (gg 0 , hh0 )

1.4 Multiplication table

1.4.1 The idea

1.4.2 The rearrangement theorem

Theorem 1.4.2 Rearrangement theorem. Let G = {g1 , . . . , gn } be a finite

contains each element of the group once and only once.

1.4.3 Groups of order 2 and 3

Similarly, the cyclic group with n elements Zn would be written as:

The direct product Z2 × Z2 would be written as:

ha, b|a2 = b2 = e, ab = bai.

2×2 matrices can be obtained by repeatedly multiplying together two matrices:

In terms of fractional linear transformations, what that means is that the

hS, T |S 2 = I, (ST )3 = Ii.

• Use Lagrange’s theorem to rule out when a subset cannot be a subgroup

is a right coset of H. Similarly, the set

gH = {ge, gh1 , . . . , ghr }

is a left coset of H. Cosets are generally not subgroups of G; they will be

1.6.2 Cosets and Lagrange’s theorem

1.7 Conjugacy classes

There is a particularly interesting equivalence relation, known as conjuga-

1.7.2 Properties of conjugacy classes

bn = (gag −1 )n = (gag −1 )(gag −1 ) · · · (gag −1 ) = gan g −1 = geg −1 = e.

It follows that bn = e, and hence the order of b is ≤ n.

1.8 The symmetric group Sn

1.8.1 Cayley’s theorem

Generically, a permutation π that takes 1 to π(1), 2 to π(2), and so on, would

Example 1.8.2 The symmetric group S3 . As an example, consider S3 .

To construct the multiplication table for S3 , we need to multiply elements in

Example 1.8.4 The cycle structure of permutations in S3 . Consider

Example 1.8.9 Multiplication of cycles vs composition of permu-

We claim that this is equivalent to the product of cycles (123)(45)(1)(2)(345).

(i1 i2 . . . ir ) = (i1 ir )(i1 ir−1 ) · · · (i1 i3 )(i1 i2 ).

Example 1.8.13 Decomposition of a permutation in S6 . As an exam-

Of course, this decomposition is not unique (for instance (15)(14)(24)(24)(23)(6)

1.8.4 Conjugacy classes

then it follows that

The conjugate permutation is then

which has cycle structure σ ◦ π ◦ σ −1 = (214)(3). It is indeed the same cycle

(ν1 + ν2 + . . . + νn ) + (ν2 + ν3 + . . . + νn ) + (ν3 + . . . + νn ) + . . . + (νn ) = n.

It is also clear that λ1 ≥ λ2 ≥ . . . ≥ λn ≥ 0, since λ1 − λ2 = ν1 , λ2 − λ3 = ν2 ,

Consider the partition 2 + 2. We get ν1 = 0, ν2 = 2, ν3 = ν4 = 0. The

1.9 Normal subgroups

G, we define the product of S1 and S2 as being the subset

That is, we multiply each element of S1 with each element of S2 . ♦

Example 1.9.7 Subgroups of abelian groups are always normal. All

1.10 Quotient groups

group operation (aH)(bH) = abH.

((aH)(bH))(cH) = (abH)(cH) = abcH = (aH)(bcH) = (aH)((bH)(cH)).

(k1 + mZ) · (k2 + Z) = k1 + k2 + mZ = k + mZ,

where k = k1 + k2 (mod m).

1.11 Homomorphisms and isomorphisms