M AT T H E W B W I N G AT E

SYMMETRIES,
PA R T I C L E S ,
AND FIELDS
These notes are for use with the Cambridge University Part III lecture course Symmetries, Particles, and
Fields, given in the Department of Applied Mathematics and Theoretical Physics in Michaelmas Term
2022. They draw on the notes of previous lecturers, in particular Ben Allanach.

This version was compiled 15 November 2022 (8:43).

Contents

Introduction 5

Discrete groups 7

Lie groups and Lie algebras 13

Representations 29

Angular momentum: SO(3) and SU(2) 37

Relativistic symmetries 45

Classification of Lie algebras 47

Bibliography 57
Introduction

Symmetry is an important concept in physics. When we say that

Nature possesses a certain symmetry, we mean that the laws of
physics are unchanged after a corresponding transformation.
We encounter symmetries as soon as we start studying physics.
Special relativity assumes that physics is the same in any iner-
tial reference frame, any two of which are related by translation,
rotation, or constant-velocity boost.
Schrödinger’s equation for a particle in a central, three-dimensional
potential is rotationally symmetric. Angular momentum states.
Even approximate symmetries are useful. Nonrelativistic space-
time. Isospin and flavour symmetries of QCD.
Gauge redundancies.
Notes by the previous lecturer have been published.1 1
B Allanach. Symmetries, Particles and
Fields. 2021. ISBN 978-1-739-94069-0
Discrete groups

In this chapter we introduce a number of concepts in the theory of

discrete groups. Many of these will generalize straightforwardly to
Lie groups.
The texts by Jones2 and Ramond3 have nice introductory chap- 2
H F Jones. Groups, Representations, and
ters on discrete groups, going into more detail than we can. Physics. IOP Publishing, 1990. ISBN
0-85274-029-8
3
P Ramond. Group Theory: A Physi-
cist’s Survey. Cambridge University
2.1 Basic definitions Press, 2010. ISBN 978-0-521-89603-
0. URL https://www.cambridge.
A group G is a set of objects or elements with a binary operation org/core/books/group-theory/
· which we will usually call a product. The group may have a fi- 8BAC137A9F0C43D65448E6420248D841

nite number of elements n, in which case the order | G | of the finite

group is n, or it may have an infinite number of elements, in which
case we say that it is an infinite group, or a group of infinite order.
This introductory chapter reviews a few concepts within the con-
text of discrete groups, those groups with a countable number of
elements

G = { g1 , g2 , . . . , g k , . . . } . (2.1)

The following axioms must hold

1. Closure: The product of any pair of group elements must also

be an element of the group. We use the notation

∀ g1 , g2 ∈ G , g2 · g1 ∈ G (2.2)

which reads “for all elements g1 and g2 in G, their product

g2 · g1 is also an element of G.”

2. Associativity: The group operation must be associative, that

is we must have

g3 · ( g2 · g1 ) = ( g3 · g2 ) · g1 (2.3)

∀ g1 , g2 , g3 ∈ G.

3. Identity: One, and only one, element of G, denoted e, must

satisfy the following for all g ∈ G

e· g = g·e = g. (2.4)

You can prove by contradiction that the identity must be

unique.
8

4. Inverse: Each element g ∈ G must have a unique inverse

g−1 ∈ G, such that

g · g −1 = e = g −1 · g . (2.5)

You can also use proof by contradiction to show that for every
g, g−1 is unique.

Note that we have not insisted that the group operation be com-
mutative. If the following property holds for all g1 , g2 ∈ G

g2 · g1 = g1 · g2 (2.6)

then we say that the group is commutative or Abelian, otherwise

we say the group is noncommutative or nonabelian.

Examples
Some examples of infinite groups

• The integers Z form a group under addition. Let us denote

any three integers by a, b, c. Then closure holds since a + b ∈
Z. Integer addition is associative: ( a + b) + c = a + (b + c).
The identity is 0. The inverse of a is (− a): a + (− a) = 0 =
(− a) + a. The group is commutative since a + b = b + a.

• The rational numbers excluding 0 form a group under mul-

tiplication. Check for yourselves that the group axioms hold
and that the group is commutative. Why must 0 be excluded?

2.2 Groups and symmetries

Let g1 represent a transformation of a physical system which leaves

physical laws invariant, and let g2 represent another such trans-
formation. Then the combination, or composition, of these two
transformations, written

g2 ◦ g1 = g2 g1 (2.7)

and read right-to-left, also leaves the physical system invariant. The
collection of all symmetry transformations { g1 , g2 , . . . , gk , . . .} =
G forms a group under composition. The group axioms are all
satisfied. By construction, closure is satisfied. Composition of these
r
transformations is associative. The identity is the “do nothing”
transformation. And we assume that any transformation can be
undone, so the corresponding group element has its inverse in G.
Let’s focus on the symmetries of a two-sided equilateral triangle.
The triangle is invariant under rotation by an angle 2π/3 and un- m3 m1 m2
der reflection about any of the three lines connecting a vertex to the Figure 2.1: Actions of D3 on an equi-
triangle’s centre (see Fig. 2.1). The symmetry group is the dihedral lateral triangle.
y
group D3 .
Let us denote the anticlockwise rotation by 2π/3 by r, and a
reflection about the vertical axis by m. The key relation for the
 9

←− g1 −→ Table 2.1: Multiplication table for

products g2 g1 in D3 , where m1 = m,
e r r 2 m1 m2 m3
m2 = mr2 , and m3 = mr.
r r 2 e m2 m3 m1
x
r2 e r m3 m1 m2


g2 m1 m3 m2 e r2 r
m2 m1 m3 r e r2

y
m3 m2 m3 r2 r e

dihedral group is that the product mr, an anticlockwise rotation fol-

lowed by a reflection, is equivalent to r −1 m, a reflection followed by
a clockwise rotation. With this we can then determine all the group
products and fill in the group’s multiplication table (Table 2.1). We
can generate all symmetry rotations and reflections using only r
and m; therefore, we call r and m generators of D3 .
Note that each row in the multiplication table is a distinct per-
mutation of the group elements. The same fact holds for the columns.
Convince yourself that if this were not the case, a contradiction
would arise.
There is a very convenient way to specify a group using its
generators. This is called a presentation and takes the form G =
r
hgenerators | rulesi. The rules specify how the generators are re-
lated to the identity and to each other. For example
D  E
D3 = r, m  r3 = m2 = e ; mr = r −1 m . (2.8)


m3
In fact, the dihedral group describing the symmetries ofman
1 n-gon is m2
the generalization

y
D  E
Dn = r, m  r n = m2 = e ; mr = r −1 m . (2.9)


Another concept which will be a major component of this course x

later is that of a group representation. Let’s introduce it here infor-

mally, in the context of D3 . Imagine drawing x- and y-axes on the Figure 2.2: Mapping of the equilateral
triangle, with the origin at its centre (Fig. 2.2). Then vectors in that triangle onto R2 .
2-dimensional vector space transform under r and m according to
left-multiplication by the following matrices
!
2π

cos 2π
3 − sin 2π3
r=R =
3
sin 2π
3 cos 2π
3
!
−1 0
m= . (2.10)
0 1

Matrix representations for the other group elements can be ob-

tained using the multiplication table. Denoting the matrix rep-
resentation of any group element g by D( g), any vector in the
2-dimensional vector space we defined when we drew the axes
transforms as v 7→ D( g)v.
Since all the matrices are distinct, this representation is said to be
faithful.
10

2.3 Useful concepts

In this section, let us recall some concepts which will be useful later

• Conjugacy: Two elements of G, say g1 and g2 , are said to be

conjugate, written g1 ∼ g2 , if there exists any g ∈ G such that

g2 = gg1 g−1 . (2.11)

Conjugacy is an equivalence relation, which partitions G into

conjugacy classes. For example, in D3 the conjugacy classes are

{e}, {r, r2 }, {m1 , m2 , m3 } , (2.12)

that is, the identity, the rotations, and the reflections.

• Subgroup: A subset H ⊂ G is a subgroup if H is a group in its

own right. We write H 6 G. Every group has two improper
subgroups, the trivial group {e} and the group G itself. All other
subgroups are proper subgroups, and the notation H < G is
used. The proper subgroups of D3 are

{e, r, r2 }, {e, m1 }, {e, m2 }, {e, m3 } . (2.13)

Lagrange’s theorem states that if G is finite and H 6 G, then | H |

divides | G |.

• Homomorphism and kernel: A mapping from group ( H, ·) to

group (K, ∗), φ : H → K, is a group homomorphism if it pre-
serves the group structure. That is, we require φ(h1 · h2 ) =
φ(h1 ) ∗ φ(h2 ), for all h1 , h2 ∈ H. The kernel of φ, ker φ, consists
of all elements of H which map to the identity of K.

• Isomorphism: A special case of homomorphism is as follows.

Two groups H and K are said to be isomorphic, denoted H ∼ = K,
if there is a bijective (one-to-one) map (an isomorphism), φ :
H → K which preserves the group structure. It follows that
| H | = |K | and the kernel of φ is solely the identity in H.

• Normal subgroup: A subgroup H < G is normal, written H
G,

if H is a union of conjugacy classes. In our example {e, r, r2 }

D3 , but the order-2 subgroups {e, mi } are not normal. If we wish
to consider normal subgroups which may not be proper, then we
write H E G.

• Coset: Let H < G and g ∈ G. Then the left coset of H with g,

written gH, is the set formed by the left product of g with each
element H:

gH = { gh | h ∈ H } . (2.14)

Similarly, the right coset of H with g, written Hg is

Hg = { hg | h ∈ H } . (2.15)
 11

It is a theorem that, for all g ∈ G, the left and right cosets are
equal if and only if H is normal:

gH = Hg , ∀ g ∈ G ⇐⇒ H
G . (2.16)

(The reader is invited to prove this theorem as an exercise.) For

example, take G = D3 and H = {e, r, r2 }. Then
(
{e, r, r2 } , g∈H
gH = Hg = . (2.17)
{ m1 , m2 , m3 } , g ∈
/H

If we took H to be one of the order-2 subgroups, none of which

are normal, then we would find gH 6= Hg for some g ∈ D3 .

• Coset product: Let us define a product of two cosets to be the

set of the distinct products of all elements. Continuing with our
example

{e, r, r2 } · {m1 , m2 , m3 } = {m1 , m2 , m3 } (2.18)

where we have noted that {em1 , em2 , em3 } is the same set as
{rm1 , rm2 , rm3 } and {r2 m1 , r2 .m2 , r2 m3 }. Let us label the two
cosets

I = {e, r, r2 } and a = {m1 , m2 , m3 } . (2.19)

These two cosets form a group under coset multiplication, in fact

it is the cyclic group of order 2,

{ I, a} ∼
= h a| a2 = ei = C2 (2.20)

where I acts as the identity. More generally the cyclic group of

order n is

Cn = h a | an = ei . (2.21)

We observe that I = {e, r, r2 } is isomorphic to C3 .

• Quotient group: Generalizing the example above, the cosets of

any normal subgroup H E G form a group under the coset
product, called the quotient group (or factor group) G/H, with
H as its identity. The proof of this theorem is nontrivial. The
example above shows that the quotient group D3 /C3 = C2 .

• Direct product: A group G is a direct product of subgroups A

and B, that is G = A × B, if

(i) ab = ba for all a ∈ A and b ∈ B ,

(ii) All g ∈ G can be expressed uniquely as some g = ab .

For example, the dihedral group D2 = hr, m|r2 = m2 = e, mr = rmi

(r ) (m)
is the direct product D2 = C2 × C2 (the superscripts distin-
guish between the cyclic groups with different generators). On
the other hand, despite the fact that D3 /C3 = C2 , D3 is not a
direct product group, D3 6= C3 × C2 .
12

We close this chapter by listing the groups of order 8. In addition

to the cyclic group C8 and the dihedral group D4 , we have the
quaternions, defined succinctly by the presentation
D  E
Q = i, j  i4 = e, i2 = j2 , ij = j−1 i (2.22)


or more familiarly as the set

Q = {±1, ±i, ± j, ±k} (2.23)

with the rules ij = k and i2 = j2 = k2 = ijk = −1. We also have the

( a) (b) (c)
direct product groups C2 × C2 × C2 and C4 × C2 .
Lie groups and Lie algebras

Lie groups and Lie algebras are named

Previously we briefly reviewed basic definitions in the context of after Norwegian mathematician
discrete groups. Discrete groups can be finite or infinite. In either Sophus Lie (1842-1899). His surname
is pronounced /LEE/.
case discrete groups have a countable number of elements. We
could label the group elements of a discrete group G by an integer
index

G = { gi } with i = 1, 2, . . . , | G | . (3.1)

Lie groups, also called continuous groups, are those with an

infinite number of elements, each characterized by one or more
continuous parameters. That is, in addition to satisfying all the
group axioms, a Lie group is also a manifold, one where group
operations are described by smooth maps. Before introducing Lie
groups in full, let us be more specific about manifolds.

3.1 Manifolds

Throughout this course we will make use of properties of differ-

entiable manifolds. Although some of the jargon may be new at
this stage, the general ideas will be familiar. We will be brief here,
but Chapter 2 of Schutz gives a nice introduction with some more
detail.4 These concepts are also introduced in Part III General Rela- 4
B Schutz. Geometrical Methods of
tivty.5 Mathematical Physics. Cambridge
University Press, 1980. ISBN 0-521-
Recall the definition of an open set. In a space with a metric, i.e. 29887-3. URL https://doi.org/10.
a notion of distance d between two points X and Y, an open set, O , 1017/CBO9781139171540
5
Fore example, see the Part III General
is a collection of points which satisfies the following property. For Relativity notes by H Reall.
any X ∈ O there exists an e > 0 such that every point Y satisfying
d( X, Y ) < e is also in O . In other words, if X ∈ O then there is
a small “neighbourhood” neary X which is also in O . An open
subset of R is an open interval; an example of an open subset of
R2 is an open disc; an open ball is an example of an open set in R3 .
This definition carries over to manifolds even where a metric is not
defined; some notion of nearness will still exist, however, as we will
see.
An n-dimensional manifold M is a set, along with a collection
of open subsets Oα satisfying the following

Oα = M, that is, every point in M is contained in at least

S
1. α
one subset.
14

Figure 3.1: An illustration of subsets

Oα and O β on manifold M, and
their respective charts φα and φβ as
ℳ described in the text.
𝒪α
𝒪β
ϕα
ϕβ

ϕβ ∘ ϕα−1
𝒰α 𝒰β
ϕα(𝒪α ∩ 𝒪β ) ϕβ(𝒪α ∩ 𝒪β )

2. For each α, there is a one-to-one mapping φα : Oα → Uα ,

where Uα is an open subset of Rn . One can say that, locally, a ℳ
𝒪α
𝒪βdimension
manifold “looks like” an open subset of Rn . The
of the manifold is given by n. ϕα
ϕβ
3. If, for any α and β, Oα ∩ O β 6= ∅, then we require a one-to-
one map φβ ◦ φα−1 from the subset φα (Oα ∩ O ⊂ Uα to the
ϕββ∘)ϕα−1
subset φβ (Oα ∩ O β ) ⊂ U β . 𝒰α 𝒰β
(𝒪α ∩ 𝒪φβα) is called
These concepts are illustrated in Fig. 3.1. Aϕαmap ϕβ(𝒪α ∩ a𝒪βcoor-
)
dinate system or a chart, and the collection {φα } is called an atlas.
We will be interested in differentiable manifolds. These are those
for which we can extend notions of differentiability of Rn to the
manifold M. For this, we need all maps φβ ◦ φα−1 to be differen-
tiable. Throughout this course, we can assume that these functions Q θ P
are smooth, i.e. infinitely differentiable.
Take for example S1 , the unit circle. We can represent S1 as a
subset of R2 by the points ( x, y) = (cos θ, sin θ ), with θ ∈ [0, 2π ).
Figure 3.2: The manifold S1.
Since this interval is not open at both ends, we need 2 charts. Let us
consider S1 in two different ways, one removing point P at θ = 0
and the other removing point Q at θ = π (see Fig. 3.2)

φ1 : (S1 − P) → (0, 2π ) , φ1 = θ1
φ2 : (S1 − Q) → (−π, π ) , φ2 = θ2 . (3.2)

These charts intersect everywhere except at P and Q. We can easily

define maps for the overlapping regions

Upper semi-circle: θ2 = φ2 ◦ φ1−1 (θ1 ) = θ1

Lower semi-circle: θ2 = φ2 ◦ φ1−1 (θ1 ) = θ1 − 2π . (3.3)

In both cases θ2 is a smooth function of θ1 , so the transition func-

tions are smooth. This means that the manifold is smooth.

3.2 Lie groups

A Lie group G is a manifold with the addition of a group operation

G × G → G which is a smooth map. Let us consider two simple
examples
 15

1. The set of n-dimensional vectors {~x ∈ Rn } form a Lie group

under vector addition. ~x3 = ~x1 + ~x2 varies smoothly with ~x1
and ~x2 . The dimension of the group is n.

2. The set of points on a circle S1 = { θ | 0 ≤ θ < 2π } form a

Lie group under addition of angles, with the identification
of θ with θ + 2nπ, integer n. In any small neighbourhood,
S1 is isomorphic to R1 , and the dimension of the group is 1.
Equivalently, we could define S1 = { z ∈ C | |z| = 1 } with
the group operation being complex multiplication. The two
definitions are isomorphic, being related by the map z = eiθ .

Just like discrete groups, Lie groups can have subgroups. If the
subgroup is continuous, and the underlying manifold is a submani-
fold of the original, then the subgroup is a Lie subgroup.

Matrix groups
Most of what we will discuss in this course concern matrix Lie
groups, those Lie groups whose elements are square, invertible
(obviously!) matrices, and whose operation is matrix multiplication.
The n-dimensional general linear group, GL(n, F ) is the set of
invertible n × n matrices over a field F. This is a subset of the set
of all n × n matrices over F, Matn ( F ). We shall normally take F to
be the real R or complex C numbers. The dimension of GL(n, R)
is n2 , since the entries of the square matrix are n2 independent real
numbers. The real dimension of GL(n, C) is 2n2 ; this is what we
will usually mean when we write about dimensionality. Sometimes
one might read that the complex dimension of GL(n, C) is n2 , in
which case the author is counting the number complex parameters
needed to specify a group element.
Let us introduce the most important subgroups of GL(n, R) and
GL(n, C):

1. The special linear group:

SL(n, F ) = { M ∈ GL(n, F ) | det M = 1 } . (3.4)

The restriction that the matrices have unit determinant is a

constraint which takes away one degree-of-freedom. Hence
dim SL(n, F ) = dim GL(n, F ) − 1.

2. The orthogonal group:

n  o
O(n) = M ∈ GL(n, R)  MT M = I , (3.5)


where T denotes a matrix transpose and I is the identity

matrix. As an exercise, show that dim O(n) = n(n − 1)/2.
Note that MT M = I implies det M = ±1.
The special orthogonal group is the subgroup of O(n) corre-
sponding to matrices M satisfying det M = +1:

SO(n) = { M ∈ O(n) | det M = 1 } . (3.6)

16

3. The unitary group:

n  o
U (n) = M ∈ GL(n, C)  M† M = I (3.7)


where † is the complex transpose.6 Note that matrices in 6

( M† )ij = ( M ji )∗ .
U (n) have determinant ±1. The special unitary group is the
subgroup of U (n) of matrices of determinant equal to +1

SU (n) = { M ∈ U (n) | det M = 1 } . (3.8)

4. The symplectic group requires us first to define a matrix Ω,

such that Ω is a fixed, antisymetric, 2n × 2n matrix. Usually
this is taken to be
!
0 In
Ω= (3.9)
− In 0

where In is the n × n identity matrix and the 0 should be read

as an n × n matrix of zeros. The symplectic group is formed
of matrices M which satisfy MT ΩM = Ω:
n  o
Sp(2n, R) = M ∈ GL(2n, R)  MT ΩM = Ω . (3.10)


The condition on M, and the fact that det Ω 6= 0, implies that

det M = ±1. In fact, one can show det M = 1. The Pfaffian of
a 2n × 2n, antisymmetric matrix A is defined to be

1
Pf A = e A · · · Ai2n−1 i2n (3.11)
2n n! i1 i2 ...i2n i1 i2
where e is the 2n-dimensional antisymmetric symbol with
e1...2n = 1.7 The crucial step is to show 7
Note that (Pf A)2 = det A.

Pf( MT ΩM ) = det M Pf Ω . (3.12)

Using ( MT ΩM ) = Ω on the left-hand side, then Pf Ω 6= 0

implies det M = 1. (In fact Pf Ω = (−1)n(n−1)/2 .)
The symplectic group can be defined as above, but over the
field of complex numbers
n  o
Sp(2n, C) = M ∈ GL(2n, C)  M† ΩM = Ω . (3.13)


An alternative matrix used in the construction of the group

instead of Ω is
 
0 1
0 ··· 0
 −1 0
 

 
 0 1 
 0 ··· 0 
J=
 − 1 0 
 (3.14)

 .
.. .
.. . .. .
..


 
0 1 
 
···

0 0 .
−1 0

It is straightforward to see that Pf J = 1.

 17

Group elements as transformations

Matrices represent linear operations which act as transformations of
vectors in particular vector spaces. Physics is often concerned with
how vectors or functions of vectors behave under such transforma-
tions.
For example, the group O(n) represents rotations and reflections
of vectors in Rn . The usual inner product is preserved under O(n)
transformations. The usual inner product of two vectors ~v1 and ~v2
in Rn is defined as (~v2 , ~v1 ) := vT
2 v1 , where on the right-hand side
we change font to denote the vectors as column matrices. It follows
then that for R ∈ O(n)

( R~v2 , R~v1 ) := vT T
v2 , ~v1 ) .
2 R Rv1 = (~ (3.15)

Similarly for vectors in ~v1 and ~v2 in Cn , the inner product h~v2 |~v1 i :=
v2† v1 is preserved under U (n) transformations.
Let us consider the Lie group SO(2), the group of rotations of
vectors in R2 :
( ! )
cos θ − sin θ 
SO(2) = R(θ ) =  θ ∈ [0, 2π ) . (3.16)
sin θ cos θ 

The group properties are easily checked. Closure implies that the
product of two rotations is another rotation, and the smooth de-
pendence on the underlying manifold requires that R(θ2 ) R(θ1 ) =
R(θ1 + θ2 ), which can be shown to hold explicitly. The identity cor-
responds to R(0), and R(θ )−1 = R(2π − θ ). Of course associativity
follows from ordinary matrix multiplication.
Rotations of vectors in R3 are described by the Lie group SO(3).
Three-dimensional rotations are specified by a unit vector ~n ∈ S2
corresponding to the axis of rotation and an angle θ; therefore,
dim SO(3) = 3. Note that a rotation of angle θ ∈ [−π, 0] about
~n is equivalent to a rotation of angle −θ about −~n, so it suffices
to confine θ ∈ [0, π ] with ~n ∈ S2 . We can depict the manifold
of SO(3) as a ball of radius π in R3 , each point θ~n corresponding
to an angle and direction. Antipodal points on the surface of the
ball are identified with each other: for any ~n rotations clockwise
and anticlockwise by an angle π are equivalent, so we must have
π~n = −π~n.
Later we will show that elements of SO(3) can be written as

Rij = cos θ δij + (1 − cos θ )ni n j − sin θ eijk nk . (3.17)

Note that the manifolds of SO(2) and SO(3) are compact, that is,
they have finite volume.

Pseudo-orthogonal and pseudo-unitary groups

SO(n) and SU (n) are compact groups. Equally important in physics
are pseudo-orthogonal and pseudo-unitary groups, which are non-
18

compact. We begin by defining an (n + m) × (n + m) metric matrix

!
In 0
η= . (3.18)
0 − Im

The pseudo-orthogonal groups O(n, m) are defined to be the set of

real matrices such that
n  o
O(n, m) = M ∈ GL(n + m, R)  MT η M = η . (3.19)


These matrices act on vectors in Rn+m and preserve the scalar prod-
uct vT
1 ηv2 for vectors ~
v1 and ~v2 in the vector space. The introduc-
tion of the signs in η does not change the dimensionality of the
group manifold compared to the corresponding orthogonal group,
so dim O(n, m) = dim O(n + m).
Examples familiar from the theory of special relativity are
( ! )
cosh ψ sinh ψ 
SO(1, 1) = ψ∈R (3.20)
sinh ψ cosh ψ 

and SO(1, 3) or SO(3, 1), depending on your convention for the

Minkowski metric. Note these groups are noncompact, as their
manifolds have infinite volume.
Pseudo-unitary groups U (n, m) are defined analogously, with
the transpose replaced by the Hermitian conjugate.

Parametrization of Lie Groups

Through examples, we have exhibited many properties of Lie
groups. Here we develop some generic notation, suitable for any
Lie group.
Any Lie group G has an associated manifold MG , so that any
element of G we have

g( x ) ∈ G , with x := ( x1 , x2 , . . . , x n ) ∈ Rn , (3.21)

where n is the dimension of Mn , and hence of G. The x are coordi-

nates on MG . Closure under group multiplication states that

g(y) g( x ) = g(z) . (3.22)

The smoothness property mentioned earlier is expressed mathemat-

ically by the statement that for the product above, the components
of z obey

zr = ϕr ( x, y) (3.23)

where ϕr ( x, y) is a continuously differentiable function of x and

y. Usually, one chooses the origin of the manifold such that the
corresponding group element is the identity: g(0) = e. In this case
the property of the group identity implies that

ϕr ( x, 0) = xr and ϕr (0, y) = yr . (3.24)

 19

We can similarly use the group inverse to constrain ϕ. Define x̄

such that g( x̄ ) = g( x )−1 , then it follows that

ϕr ( x̄, x ) = 0 = ϕr ( x, x̄ ) . (3.25)

Associativity,

g(z)[ g(y) g( x )] = [ g(z) g(y)] g( x ) (3.26)

implies

ϕr ( ϕ( x, y), z) = ϕr ( x, ϕ(y, z)) . (3.27)

To summarize, any Lie group G consists of a differentiable mani-

fold MG along with a map ϕ : MG × MG → MG satisfying (3.24),
(3.25), and (3.27).

3.3 Tangent spaces

Here we follow Schutz §2.7.8 Consider a point p on an n-dimensional 8

B Schutz. Geometrical Methods of
manifold M. A curve passing through p can be described in terms Mathematical Physics. Cambridge
University Press, 1980. ISBN 0-521-
of coordinates in the neighbourhood of p as x (λ) = ( x1 (λ), . . . , x n (λ)), 29887-3. URL https://doi.org/10.
where λ is a real parameter such that when λ = λ p , x (λ p ) is the co- 1017/CBO9781139171540

ordinates of p.
Consider a differentiable function f ( x1 , . . . , x n ) (abbreviated f ( x )
from now on) on M, or at least in some neighbourhood around p.
Along the curve there is a differentiable function h(λ) such that

h(λ) = f ( x (λ)) . (3.28)

By the chain rule

dh dxi ∂ f
= , (3.29)
dλ dλ ∂xi
where we sum repeated indices. This should be true for any func-
tions f and h, so we can identify

d dxi ∂
= . (3.30)
dλ dλ ∂xi
We will see that d/dλ is a vector in a tangent space defined in
relation to point p in M, denoted Tp (M). The set {∂/∂xi } form a
basis for the tangent space.
Consider another curve through p, say x̃ (µ), in which case we
derive a differential operator

d d x̃i ∂
= . (3.31)
dµ dµ ∂xi

Generally, this will give a tangent vector different to d/dλ.9 Now 9

Note that while each curve through
if we take a linear combination of two different tangent vectors, we p will have a unique tangent vector,
an infinite number of curves through
find p will have the same tangent vector.
Each tangent vector describes an
dxi d x̃i
 
d d ∂ d equivalence class of curves through p.
a +b = a +b =: . (3.32)
dλ dµ dλ dµ ∂xi dσ
20

In the last step, we assert that we can find a curve x̄ (σ ) which

passes through p having a tangent vector with the components
given by (3.32), which we can surely do. This proves the closure
property of vectors in a vector space; it is easy to prove the other
vector space axioms hold for these differential operators.
We have seen that the space of derivatives of functions along
curves passing through p is in 1-1 correspondence with the vectors
tangent to p. This space is denoted by Tp (M), the tangent space to
manifold M at point p. Tangent vectors at another point in M, say
q, have no relation to those in Tp (M), existing instead in Tq (M).
We now introduce, for later use, the notion of exponentiation of
a differential operator d/dλ along a curve in a manifold. Note that
the following requires the manifold to be analytic, i.e. C ω . Analytic
manifolds are in C ∞ , but with stricter requirements (viz analyticity)
on transformations between coordinates. Coordinates of two nearby
points on a curve xi (λ) can be related by Taylor expansion about a
point p such that x (λ p ) gives the coordinates of p:
dxi  ε2 d2 xi 
 
i i
x (λ p + ε) = x (λ p ) + ε + +...
dλ λ p 2! dλ2 λ p
∞ n
" # 
ε dn 
= ∑ x ( )

n
λ
n! dλ

n =0

λp
  
d
xi (λ) .

=: exp ε (3.33)
dλ λp

The “exponential” serves as a shorthand for the infinite series of

differential operators in the line above. We will use this extensively
in the course, and it will become more concrete when we consider
matrix Lie groups and Lie algebras.

3.4 Lie Algebras

Let us define Lie algebras first and later make the connection with
Lie groups. Recall that a vector space V over a field F (usually F =
R or C for us), along with the following properties. For X, Y ∈ V
and α, β ∈ F:

α( X + Y ) = αX + αY (3.34)
(α + β) X = αX + βX (3.35)
(αβ) X = α( βX ) (3.36)
1X = X (scalar identity) (3.37)
~0 + X = X (vector identity) (3.38)

where 1 is the multiplicative identity of F.

A Lie algebra is a vector space V which additionally has as a
vector product the Lie bracket [·, ·] : V × V → V which must have
the following properties. For X, Y ∈ V:
1. Antisymmetry

[ X, Y ] = −[Y, X ] . (3.39)
 21

2. Jacobi identity

[ X, [Y, Z ]] + [Y, [ Z, X ]] + [ Z, [ X, Y ]] = 0 . (3.40)

3. Linearity. For α, β ∈ F,

[αX + βY, Z ] = α[ X, Z ] + β[Y, Z ] . (3.41)

Note that any vector space which has a product ∗ : V × V → V can

be made into a Lie algebra by defining the Lie bracket to be

[ X, Y ] := X ∗ Y − Y ∗ X , (3.42)

for example, the commutator of matrices.

We can define basis vectors for the Lie algebra V. Denoted by
{ Ta }, with a = 1, . . . , dim V, we will refer to these as generators
of the Lie algebra. We define structure constants f through the Lie
brackets of these basis vectors

[ Ta , Tb ] = f c ab Tc . (3.43)

Antisymmetry of the bracket implies f c ba = − f c ab . The Jacobi

identity (3.40), with X = Ta , Y = Tb , and Z = Tc , implies

f e ad f d bc + f e cd f d ab + f e bd f d ca = 0 . (3.44)

We can write a general elements of the Lie algebra as linear

combinations of the basis vectors, for example X = X a Ta , where the
X a are coefficients in F. Then

[ X, Y ] = X a Y b f c ab Tc . (3.45)

3.5 Lie groups and their Lie algebras

The Lie algebra L( G ) of a Lie group G is the tangent space to G at

the identity e.10 . Below we will see that every Lie group has a cor- 10
It is also common to distinguish
responding Lie algebra. The reverse is not true; several Lie groups a Lie group G from its Lie algebra
g by a change of case and font, e.g.
can have the same Lie algebra. However, for each Lie algebra, there L(SU (n)) = su(n).
is only a single simply-connected Lie group. A Lie group is simply
connected if any closed curve on its manifold can be shrunk to a
point. Before we give a general presentation, let us examine some
simple examples.

Illustrative examples
Recall that SO(2) is 1-dimensional, so we have a single coordi-
nate θ. Define a “curve” θ (λ) in this 1-dimensional space such that
θ (0) = 0. Then the “curve” in SO(2) is
!
cos θ (λ) − sin θ (λ)
g(λ) = , (3.46)
sin θ (λ) cos θ (λ)
22

where g(0) now corresponds to the identity of SO(2). The tangent

vector to the identity for this g is
 ! 
dg  0 −1 dθ 
= . (3.47)
dλ 0 1 0 dλ 0

The tangent space at the origin is spanned by the vector11 11

Note that, despite the notation, eθ is
simply a vector in a vector space. The
! d
notation dθ reminds us that the vector
d 0 −1
= = : eθ . (3.48) space is actually a tangent space,
dθ 1 0 formed from derivatives along curves
on a manifold.
Thus the Lie algebra of SO(2) is
( ! )
0 −a 
L(SO(2)) = a∈R . (3.49)

a 0 

Now for general SO(n). Let us consider a curve in SO(n), i.e. a

single-parameter family of orthogonal matrices, passing through
the identity I: M(λ) ∈ SO(n) such that M(0) = I. By the group
definition, MT (λ) M (λ) = I for all λ. Hence

d h T i
0= M (λ) M(λ)
dλ
dMT dM
= M + MT
dλ dλ
= ṀT M + MT Ṁ , (3.50)

where we introduce the notation that a dot represents directional

differentiation with respect to its parameter. Looking at λ = 0 and
using M (0) = I we infer that

dMT 


T
 dM 
Ṁ  = − Ṁ 0 , or
  =− . (3.51)
0 dλ  dλ  0
0

In other words, the Lie algebra consists of real anti-symmetric n × n

matrices:
n  o
L(SO(n)) ⊂ X ∈ Matn (R)  X T = − X =: Skewn . (3.52)


In fact we can show that L(SO(n)) = Skewn if we can prove that

every antisymmetric matrix is in L(SO(n)). The nicest derivation
will be given in § 3.6. Here we will justify this on dimensional
grounds. To determine the dimensionality of Skewn , we count
the number of parameters required to specify an antisymmetric
n × n matrix. First of all, the diagonal entries must be 0, and then it
suffices to specify the elements in the upper triangular part of the
matrix. Therefore,
1 2 1
dim (Skewn ) = ( n − n ) = n ( n − 1) . (3.53)
2 2
It should be clear, and we will reinforce this below, that a Lie al-
gebra has the same dimension as its underlying group(s), so let
us determine the dimension of SO(n). A matrix M ∈ SO(n) has
 23

n2 entries, but must satisfy MT M = I, or Mki Mkj = δij . These

appear to be independent equations for i 6 j (say), so they im-
pose 21 n(n + 1) constraints. To be fair, we do not know whether all
these constraints are independent, so the most we can say is that
dim SO(n) > 12 n(n − 1). Therefore, we have shown that

n ( n − 1) n ( n − 1)
6 dim L(SO(n)) 6 dim Skewn = . (3.54)
2 2
By the “sandwich theorem” dim L( G ) = dim Skewn , so L( G ) =
Skewn . (However, see § 3.6 for a more satisfying proof.)
Note that L(O(n)) = L(SO(n)), since a matrix R ∈ O(n) which is
in a neighbourhood with the identity also has det R = 1. The matri-
ces in O(n) with determinant −1 correspond to a disconnected part
of the group manifold.
Turning to the unitary groups SU (n), let M (λ) be a curve in
SU (n) with M (0) = I. For small λ let us write

M (λ) = I + λX + O(λ2 ) (3.55)

where X is a matrix of constants. By construction, X = dM (λ)/dλ|0 .

Unitarity of M implies

I = M† M (3.56)
† 2
= I + λ( X + X ) + O(λ ) (3.57)

from which we conclude X † = − X. Writing out

 
1 + λX11 λX12 λX13 ···

 λX21 1 + λX22 λX23 · · ·

M(λ) =  λX31 λX32 1 + λX33 · · ·
 (3.58)
.. .. ..
 
..
. . . .

we have

1 = det M (3.59)
2
= 1 + λTr X + O(λ ) . (3.60)

This shows that X must be traceless.

n  o
L(SU (n)) = X ∈ Matn (C)  X † = − X, Tr X = 0 . (3.61)


For U (n), the determinant can be any phase eiθ , so X need not be
traceless and
n  o
L(U (n)) = X ∈ Matn (C)  X † = − X . (3.62)


Lie algebra of a matrix Lie group

Say we have Lie group G ⊂ GL(n, F ) ⊂ Matn ( F ) with coordinates
x = x1 , . . . , x m , where m = dim G, and the origin coinciding with
the group identity, such that g(0) = e. Then the tangent n space  o
∂g( x ) 
Te ( G ) = L( G ) is the subspace of Matn ( F ) spanned by ∂xi
 .12 12
Note that this implies that the
0 dimension of the Lie algebra is equal
to the dimension of its corresponding
Lie group.
24

We can prove this statement by constructing a linear, one-to-one

map, a bijection, ρ : Te ( G ) → Matn ( F ) such that
  
∂ ∂g 
ρ = . (3.63)
∂xi ∂xi 0

To show this is a bijection consider 2 distinct vectors v = vi ∂∂xi and

w = wi ∂∂xi . We can use linearity to consider the difference between
the images of each of vector
     
i ∂ i ∂ i i ∂
ρ v i − ρ w i = ρ (v − w ) i
∂x ∂x ∂x

i i ∂g 
= (v − w ) i  (3.64)
∂x 0

∂g 
None of the partial derivatives ∂xi  should vanish, unless we have
0
introduced some redundant coordinates, therefore, the
   
i ∂ i ∂
ρ v i − ρ w i = 0 ⇐⇒ vi − wi = 0 ∀i . (3.65)
∂x ∂x

That is, the mapping is one-to-one.

Since L( G ) ⊂ Matn ( F ), we have a natural definition for the Lie
bracket as the matrix commutator: [ X1 , X2 ] = X1 X2 − X2 X1 .
Let X1 and X2 be elements of the Lie algebra L( G ). We express
them as elements of the tangent space of G at g = e:
 
dg1 (λ)  dg2 (µ) 
X1 = = ġ1 |0 X2 = = ġ2 |0 (3.66)
dλ 0 dµ 0

for curves g1 (λ), g2 (µ) ∈ G with g1 (0) = g2 (0) = e. Let us vary

λ and µ linearly with respect to a third parameter, t. That is, let
λ = λ̃t and µ = µ̃t, with λ̃ and µ̃ held constant. Taking a group
product g3 (t) = g2 (µ̃t) g1 (λ̃t) the result

d  
ġ3 |0 = g2 (µ̃t) g1 (λ̃t) 0
dt


= λ̃g2 ġ1 + µ̃ ġ2 g1 0
= λ̃X1 + µ̃X2 (3.67)

is another vector in the tangent space, i.e. an element of L( G ).

We can use the group structure to show that the Lie bracket here
is a commutator. Now consider two curves on the manifold, both
parametrized by t (setting λ̄ = µ̄ = 1 and such that g1 (0) = g2 (0) =
e. Taylor expanding about the origin, we have

g1 (t) = e + tX1 + t2 W1 + O(t3 )

g2 (t) = e + tX2 + t2 W2 + O(t3 ) . (3.68)

Then respective products are given by

g2 (t) g1 (t) = e + t( X1 + X2 ) + t2 ( X2 X1 + W1 + W2 ) + O(t3 )

g1 (t) g2 (t) = e + t( X1 + X2 ) + t2 ( X1 X2 + W1 + W2 ) + O(t3 ) (3.69)
 25

Taking the inverse of one of these (which is also an element of G)

and then the product with the other yields

h(t) = [ g2 (t) g1 (t)]−1 g1 (t) g2 (t)

= e + t 2 [ X1 , X2 ] + O ( t 3 ) . (3.70)

Note that the steps above could equally well have been done taking
µ̄ = −1, which would result in a minus sign in front of the t2 term.
Thus we can reparametrize this curve as

h̃(s) = e + s[ X1 , X2 ] + . . . . (3.71)

Thus h̃ is a curve in the manifold, passing through the identity,

which has with tangent vector at the origin [ X1 , X2 ] ∈ L( G ). This
demonstrates that L( G ) is closed under the commutator, which
satisfies the defining properties of a Lie bracket (3.39), (3.40), and
(3.41).
Note that the commutator is only nonzero if and only if the
group G is nonabelian. If G is abelian, then h(t) = e for all t, so the
commutator must vanish.

Tangent space to G at general element g

Let us consider a matrix Lie group G < GL(n, F ) and examine the
tangent space at a group element p = g(t0 ) which sits on a curve
C = g(t). The tangent space is denoted Tp ( G ). The curve need not
pass through the group identity e. Looking at a nearby point along
the curve, g(t0 + ε), let us expand

g(t0 + ε) = g(t0 ) + ε ġ(t0 ) + O(ε2 ) . (3.72)

Since g(t0 + ε) is a group element and the group operation is

smooth, there must be some element of G, near the identity, h p (ε)
which satisfies

g ( t0 + ε ) = g ( t0 ) h p ( ε ) . (3.73)

The p subscript reminds us that h will be different if we choose a

different point along C . Expanding about small ε we have

h p (ε) = e + εX p + O(ε2 ) (3.74)

for some X p ∈ L( G ), again dependent on which point p we are ex-

amining. h p (ε) is the group element which generates the translation
t0 7→ t0 + ε.
Working through O(ε)

1 + εX p = h p (ε) = g(t0 )−1 g(t0 + ε)

= g(t0 )−1 [ g(t0 ) + ε ġ(t0 )]
= e + εg(t0 )−1 ġ(t0 ) . (3.75)

We can identify X p with g(t0 )−1 ġ(t0 ). In other words, g(t0 )−1 ġ(t0 ) ∈
L( G ) for any t0 . By a similar argument ġ(t0 ) g(t0 )−1 ∈ L( G ). This
26

shows that, although vectors ġ(t0 ) ∈ Tp ( G ) are not in the Lie

algebra, one can map them to the Lie algebra by left or right multi-
plication by g(t0 )−1 .13 13
When we come to study gauge the-
Conversely, consider any X ∈ L( G ). Then there exists a curve ories, we encounter the following.
Say g( x1 , . . . , x n ) is a G-valued func-
g(t) satisfying tion on Rn . Then ∂x∂ i ≡ ∂i g is in the
tangent space Tg ( G ). Consequently
g−1 (t) ġ(t) = X . (3.76) (∂i g) g−1 ∈ L( G ) and g−1 (∂i g) ∈ L( G ).

If we specify an “initial condition”, i.e. that the curve go through

a point such that g(t0 ) = g0 , then the curve satisfying (3.76) will
be unique. Existence and uniqueness of this curve follow from the
usual existence and uniqueness theorem for differential equations,
extended to more general manifolds. The solution is given by the
exponential map

g(t) = (exp tX ) g0 . (3.77)

For matrix Lie algebras, this is just the matrix exponential

∞
1
exp tX = ∑ k!
(tX )k . (3.78)
k =0

One parameter subgroups

Given an element of the Lie algebra X ∈ L( G ), the curve

gX (t) = (exp tX )e (3.79)

is an Abelian subgroup of G, said to be generated by X. The group

properties are easily checked. We have the identity gX (0) = e
and unique inverses gX (t)−1 = gX (−t). Associativity is manifest.
Finally we have closure and commutativity since

g X ( t2 ) g X ( t1 ) = g X ( t2 + t1 ) = g X ( t1 + t2 ) = g X ( t1 ) g X ( t2 ) . (3.80)

The curve is isomorphic to (R, +) if only gX (0) = e and to S1 if

there is an additional gX (t0 ) = e where t0 6= 0. (Of course there is
the trivial case where gX (t) = e for all t.)

3.6 Lie groups from Lie algebras

Let us apply the exponential map to elements of the Lie algebra

L( G ). We will see that we can recover some components of group
G. We have already seen that several groups can have the same
algebra, so we cannot expect to generate any Lie group solely from
its algebra.
We use the exponential map exp : L( G ) → G such that, for all
X ∈ L( G )

X 7→ exp X . (3.81)

Locally, the map is bijective (one-to-one). However, globally the

map is generally not one-to-one. For example the map f : R → S1
 27

given by f (θ ) = exp iθ is not one-to-one since exp(2πin) = 1 for all

integer n.
In general, the image of the exponential map is not the whole of
G, but a subset of G connected to the identity. That is, the exponen-
tial map is not necessarily surjective (onto). If the group is compact,
then the image of exp X is the whole of the connected part of G.
However, if G is noncompact, then the exponential map may not be
surjective. This is the case for SL(2, R). The interested reader may
consult Hall for further details.14 14
B C Hall. Lie Groups, Lie Algebras,
Having a map from the algebra to the group, it remains to check and Representations: An Elementary
Introduction. Springer, 2015. ISBN
that, given the Lie bracket, can we determine the group product? 978-3319134666
The answer is yes, via the Baker–Campbell–Hausdorff relation:

exp tX exp tY = exp tZ (3.82)

where

t t2  
Z = X + Y + [ X, Y ] + [ X, [ X, Y ]] + [Y, [Y, X ]] + O(t3 ) . (3.83)
2 12
The proof of this proceeds order-by-order in t and quickly becomes
tedious. This universal formula (3.82)–(3.83) shows that the group
structure of G near the identity e can be determined by the algebra
L ( G ).
For example consider the Lie group O(n). Let X ∈ L(O(n)) and
M = exp tX. We want to show that M ∈ O(n).

MT = (exp X )T
= exp X T
= exp(− X ) = M−1 . (3.84)

Therefore, M is orthogonal. Let’s check the determinant. Since X is

antisymmetric,

Tr X + Tr X T = Tr X − Tr X = 0 (3.85)

while also

Tr X + Tr X T = Tr( X + X T ) = 2 Tr X . (3.86)

Therefore Tr X = 0. Let X have eigenvalues λ1 , λ2 , . . . , λn . Then

det M = det exp X = eλ1 eλ2 · · · eλn

= exp Tr X = 1 . (3.87)

A ha! We see that exp X is necessarily in SO(n). We cannot gen-

erate an element of O(n) whose determinent is −1 from the Lie
algebra. The orthogonal group consists of disconnected manifolds,
one containing the proper SO(n) rotations and the other containing
improper rotations (those with a reflection).
We are now in a position to return to our statement that L(SO(n))
contains every element of Skewn . Let A be any matrix in Skewn .
28

Let us use the exponential map to define a curve of matrices γ(t)

on some manifold

γ(t) := exp tA . (3.88)

In fact, by the above derivation,

(γ(t))T γ(t) = I (3.89)

and det γ(t) = 1. Therefore, γ(t) ∈ SO(n). By construction,

γ̇(0) = A (3.90)

so matrix A is tangent to a curve in SO(n), and is consequently an

element of L(SO(n)).
Representations

As interesting as Lie groups and their algebras are, what we need

in theoretical physics are the actions of group elements on vec-
tors in vector spaces. We want to understand what types of vector
spaces admit Lie group (and algebra) representations.

4.1 Lie group representations

A representation D of a group G is a smooth group homomor-

phism, mapping group elements to transformations on a vector
space V; D : G → GL(V ). The general linear group on a vector
space V is the group of all automorphisms of V, i.e. bijective, linear
maps V → V. V is the representation space of D. If V is finite di-
mensional and a basis is chosen, then GL(V ) is isomorphic to the
general linear group of matrices GL( N, F ), where N = dim V. That
is, for all g ∈ G, D ( g) : V → V is a linear, invertible map such that

v → D ( g)v for v ∈ V . (4.1)

Linearity of the map implies

D ( g)(αv1 + βv2 ) = αD ( g)v1 + βD ( g)v2 (4.2)

for all α, β ∈ F and v1 , v2 ∈ V. In order to be a group homomor-

phism the mapping must preserve the group operation, i.e. we
must have

D ( g2 g1 ) = D ( g2 ) D ( g1 ) . (4.3)

This property (4.3) implies that the identity of the group maps to
the identity map on V, IN if we have a matrix representation, and
that the inverses map to inverses:

D (e) = IN (4.4)
−1 −1
D( g ) = D ( g) . (4.5)

The dimension of a representation is just the dimension of its rep-

resentation space.
A representation is faithful if D ( g) = IN only for g = e, the
group identity. That is, the representation is faithful if ker D = e.15 The kernel of a map D : G →
15

This is enough to imply that D is injective, i.e. D ( g1 ) = D ( g2 ) =⇒ GL( N, F ) consists of the elements of G
which map to IN .
g1 = g2 .
30

Each group has a trivial representation, where

D0 ( g) = 1 , ∀g ∈ G , (4.6)

the 1 × 1 identity, for all g ∈ G. The trivial representation is not

faithful: its dimension is 1 and the kernel of the homomorphism
D0 is all of G. (In fact, we can form trivial representations of any
dimension M by mapping D0 ( g) = I M for all g ∈ G.)
If G is a Lie group of matrices of dimension n, i.e. if G 6 GL(n, F ),
then the representation

D f ( g) = g , ∀g ∈ G , (4.7)

is called the fundamental representation of G. The dimension of

the fundamental representation is clearly also n. The kernel of the
homomorphism D f is trivial, so the fundamental representation is
faithful. The dimension dim D f = n.
Let G be a matrix Lie group and consider the case where the
vector space V is the corresponding matrix Lie algebra L( G ). The
adjoint representation of G is the representation of G on L( G ):
Dadj : G → GL( L( G )). The adjoint representation plays such a
special role that it has a special denotation: Adg := Dadj ( g). The
mapping Adg : L( G ) → L( G ) is given, for any g ∈ G and any
X ∈ L( G ), by

Adg X := gXg−1 .

The dimension of the adjoint representation is equal to the dimen-

sion of the Lie algebra L( G ).
We can check that this is a well-defined representation:

• Closure: For a given X, there is some curve in G, g(t) =

I + tX + . . ., with tangent X at t = 0. For any h ∈ G, we
have another curve in G given by g̃(t) = hg(t)h−1 . Near the
identity

g̃(t) = I + thXh−1 + . . . . (4.8)

Since hXh−1 is tangent to the curve at t = 0, it is a vector in

the tangent space, thus Adh X = hXh−1 ∈ L( G ).

• Group operation:

Adg2 g1 X = ( g2 g1 ) X ( g2 g1 )−1
= g2 g1 Xg1−1 g2−1
= Adg2 Adg1 X . (4.9)

Let’s consider a simple example, with G = (R, +). Any represen-

tation D must satisfy

D (α + β) = D (α) D ( β) with α, β ∈ R . (4.10)

Each of these are representations of this group:

 31

(a) D (α) = ekα for some k ∈ R. This a faithful representation as

long as k 6= 0.

(b) D (α) = eikα for some k ∈ R. This is not a faithful representa-

tion since ker D = 2π k n ∀n ∈ Z .
 


(c) We can form a 2-dimensional representation

!
cos α − sin α
D (α) = . (4.11)
sin α cos α

This representation is not faithful because of the periodicity

of the trigonometric functions.

(d) We can form an infinite dimensional representation by

considering V to be the vector space of all real functions
f : R → R and letting

( D (α) f ) ( x ) = f ( x − α) . (4.12)

The representation is faithful since D (α) f ( x ) = f ( x ) for all

f only if α = 0, i.e. ker D = {0}.

4.2 Lie algebra representations

The representation of a Lie algebra L( G ) is a map from L( G ) to

the set of automorphisms V → V, not necessarily invertible. That
is, d : L( G ) → gl(V ), where gl(V ) = L( GL(V )) is the algebra of
GL(V ). If we have a basis for V, then d : L( G ) → gl( N, F ), but this
is just the set of N × N matrices over F, Mat N ( F ). For v ∈ V and
any X ∈ L( G ), we have d( X ) : V → V such that

v 7→ d( X )v . (4.13)

The representation must preserve the Lie bracket, i.e.

d ([ X, Y ]) = [d( X ), d(Y )] . (4.14)

The dimension of the representation d is the dimension of its corre-

sponding vector space V.
The trivial representation simply maps each X ∈ L( G ) to the
zero vector of V

d( X ) = 0 , ∀ X ∈ L( G ) . (4.15)

Considering matrix Lie group G 6 GL(n, F ), the fundamental

representation of L( G ) is d f : L( G ) → Matn ( F ) with

d f (X) = X , ∀ X ∈ L( G ) . (4.16)

dim d f = n.
The adjoint representation of (any) Lie algebra, adX : L( G ) →
L( G ), is given by

adX Y = [ X, Y ] . (4.17)

dim(ad) = dim L( G ). We will discuss the adjoint representation of

the Lie algebra more in the next subsection.
32

Algebra representations from group representations

Consider tangents to curves in the group manifold which pass
through the identity

g(t) = e + tX + . . . ∈ G, (4.18)

where X ∈ L( G ). Let D be a representation of G and V be its

representation space. Along the curve in the group manifold, we
have

D ( g(t)) = I + td( X ) + . . . . (4.19)

where I is the identity map on V and (4.19) implicitly defines d( X )

in relation to D ( g). We can check that these d( X ) form a Lie algebra
by checking their Lie bracket. Consider two curves

g1 (t) = e + tX1 + t2 W1 + . . .
g2 (t) = e + tX2 + t2 W2 + . . . (4.20)

and two ways of writing the representation of the product g1−1 g2−1 g1 g2 :

D ( g1−1 g2−1 g1 g2 ) = D ( g1 )−1 D ( g2 )−1 D ( g1 ) D ( g2 ) . (4.21)

On the lefthand side we have

D ( g1−1 g2−1 g1 g2 ) = D (e + t2 [ X1 , X2 ] + . . .)
= I + t2 d([ X1 , X2 ]) + . . . (4.22)

and on the righthand side

D ( g1 )−1 D ( g2 )−1 D ( g1 ) D ( g2 ) = I + t2 [d( X1 ), d( X2 )] + . . . . (4.23)

Therefore, we conclude that

d([ X1 , X2 ]) = [d( X1 ), d( X2 )] , (4.24)

as required.
Let us see this for the case of the adjoint representations of ma-
trix Lie groups and algebras. For g ∈ G and X, Y ∈ L( G ),

Adg Y = gYg−1
= ( I + tX )Y ( I − tX ) + . . .
= Y + t[ X, Y ] + . . .
= ( I + tadX + . . .)Y (4.25)

where in the last line we identify adX Y = [ X, Y ] as before.

Group representations from algebra representations

Given that d is a representation of L( G ) and X ∈ L( G ), let g =
exp X ∈ G and D ( g) = exp d( X ). We can use the Baker–Campbell–
Hausdorff formula to confirm that, for g1 , g2 ∈ G, D obeys

D ( g2 g1 ) = D ( g2 ) D ( g1 ) . (4.26)
 33

However, D may not be a representation of G since the exponential

map is not guaranteed to be surjective (onto). Generally, it will not
be. We need two conditions in order for exp d to be a representa-
tion of G.

1. Every element in G can be written as exp X.

2. G must be simply connected. That is, all closed curves in G

can be shrunk to a point.

4.3 Useful concepts

In this section we introduce a number of concepts which are useful

in discussing group and algebra representations. Keep in mind that
whatever we say about a Lie group representation has implications
for its Lie algebra representation, and whatever we say about a Lie
algebra’s representation informs us about the representations of the
groups which share the Lie algebra.

• Equivalent representations: Representations D1 and D2 of G

or d1 and d2 of L( G ) are equivalent if there exists an invertible
matrix R or S such that,

D2 ( g) = R D1 ( g) R−1 ∀ g ∈ G
or d2 ( X ) = S d1 ( X ) S−1 ∀ X ∈ L( G ) . (4.27)

• A representation d of L( G ) with corresponding vector space V

has an invariant subspace W ⊂ V if

d( X )w ∈ W (4.28)

for all w ∈ W and all X ∈ L( G ). All representations have two

trivial invariant subspaces, {0} and V. The above definitions
are the same for group representations.

• An irreducible representation, or irrep, is a representation

with no nontrivial invariant subspaces. A representation
which has a nontrivial subspace is reducible.

• The direct sum of two vector spaces U and W

U ⊕ W = { u ⊕ w | u ∈ W, w ∈ W } (4.29)

where the direct sum of vectors obeys

( u 1 ⊕ w1 ) + ( u 2 ⊕ w2 ) = ( u 1 + u 2 ) ⊕ ( w1 + w2 ) (4.30)
λ(u ⊕ w) = (λu) ⊕ (λw) . (4.31)

Note that dim (U ⊕ W ) = dim U + dim W.

• A totally reducible representation d of L( G ) (or D of G) can

be decomposed into irreducible pieces. That is, V can be writ-
ten as a direct sum of invariant subspaces V = W1 ⊕ W2 ⊕ . . .,
34

where d( X )wi ∈ Wi for any wi ∈ Wi and all X ∈ L( G ). A basis

for V exists in which each d( X ) is block-diagonal
 
d1 ( X ) 0 0 ···
 0

d2 ( X ) 0 ··· 
d( X ) =  0

0 d3 ( X ) · · · 
 (4.32)
.. .. ..
 
..
. . . ..

• An N-dimensional group representation D is unitary if

D ( g) ∈ U ( N ) for all g ∈ G. This implies the corresponding Lie
algebra representation d( X ) is antihermitian for all X ∈ L( G ).
If all D ( g) are also real, then D is said to be orthogonal.

Let us investigate an important theorem combining many of the

concepts introduced above.

Theorem (Maschke): A finite-dimensional, unitary representation

is either irreducible or totally reducible. We sketch the proof here,
leaving more detail for an Examples Sheet question. For each invari-
ant subspace W, one can show that the orthogonal compement W⊥ ,
constructed using the usual inner product and unitarity of the repre-
sentation, is also an invariant subspace. This implies V = W ⊕ W⊥ . If
W and W⊥ have any nontrivial invariant subspaces, the we repeat the
process. Since the representation is finite-dimensional, this process
must terminate.

In the case of discrete groups and compact Lie groups, Maschke’s

theorem can be extended to finite representations which are not el-
ements of U ( N ). One defines a new group-invariant inner product
with respect to which D ( g) is unitary. The proof is not too compli-
cated for discrete groups, but the derivation for compact Lie groups
goes beyond the scope of this course. Nevertheless, we will rely on
the applicability of Maschke’s theorem to compact Lie groups in
later chapters.
Let us consider a simple example, again with the group (R, +).
Let V be the space of all 2π-periodic functions f : R → R, with
f ( x + 2π ) = f ( x ). We take D as a representation of f such that

( D (α) f )( x ) = f ( x − α) . (4.33)

Unlike the example given in (4.12), the kernel of this representation

is nontrivial; for α = 2πk and integer k, we have

( D (2πk) f )( x ) = f ( x ) , ∀f . (4.34)

Therefore, D must be reducible. The invariant (one-dimensional)

subspaces are given by
n  o
Wn = f ( x ) = cn einx  n ∈ Z . (4.35)


Closure is evident since

cn ein( x−α) = einα cn einx ∈ Wn , ∀α . (4.36)

 35

The familiar Fourier decomposition is an illustration of this total

reducibility. Any 2π-periodic function can be decomposed into the
Fourier sum
∞
f (x) = ∑ cn einx . (4.37)
n=−∞

Thus the vector space V can be written as the infinite direct sum
∞
M
V = . . . ⊕ W−1 ⊕ W0 ⊕ W1 ⊕ W2 ⊕ . . . =: Wn (4.38)
n=−∞

where we introduce the symbol for a direct sum in the last step.
Note that each invariant subspace occurs exactly once in V.

Tensor product
Given vector spaces V and W, the tensor product space V ⊗ W is
spanned by vectors of the form v ⊗ w ∈ V ⊗ W, where v and w are
basis elements of V and W, respectively. The tensor product of two
vectors v ∈ V and w ∈ W obeys the following distributive properies

v ⊗ ( λ 1 w1 + λ 2 w2 ) = λ 1 v ⊗ w1 + λ 2 v ⊗ w2
( λ1 v1 + λ2 v2 ) ⊗ w = λ1 v1 ⊗ w + λ2 v2 ⊗ w . (4.39)

The dimension of V ⊗ W is equal to (dim V )(dim W ). A vector

Φ ∈ V ⊗ W which is equal to the direct product of two vectors
v ∈ V and w ∈ W can be written as

Φ A := Φαa := vα wa (if Φ = v ⊗ w) (4.40)

where α = 1, 2, . . . , dim V, a = 1, 2, . . . , dim W, and A = 1, 2, . . . , dim (V ⊗

W ). A simple map between these indices is A = α(dim) W + a.
Note that not all elements of V ⊗ W can be written as a direct
product of a vector in V with a vector in W. For example, the linear
combination

λ 1 v 1 ⊗ w1 + λ 2 v 2 ⊗ w2 ∈ V ⊗ W (4.41)

may generally not be written as λ3 v3 ⊗ w3 for some v3 ∈ V, w3 ∈ W.

Tensor products allow one to combine representations of Lie
groups (and consequently their Lie algebras). Let D (1) and D (2) be
representations of G with respective representation spaces V and
W. That is, for all g ∈ G

D (1) ( g ) : vα 7→ D (1) ( g)αβ v β , v∈V

D (2) ( g ) : wa 7→ D (2) ( g) ab wb , w ∈W. (4.42)

The tensor product representation D (1) ⊗ D (2) acts on vector space

V ⊗ W such that

( D (1) ⊗ D (2) )( g)αa,βb = D (1) ( g)αβ D (2) ( g) ab . (4.43)

36

This acts on vectors in the tensor product space, Φ ∈ V ⊗ W as

Φ A = Φαa 7→ D (1) ( g)αβ D (2) ( g) ab Φ βb = Φ B . (4.44)

Let d(1) and d(2) be the algebra representations corresponding to

D (1) and D (2) , respectively. We define their tensor product as, for
X ∈ L( G )

(d(1) ⊗ d(2) )( X ) = d(1) ( X ) ⊗ IW + IV ⊗ d(2) ( X ) , (4.45)

where IV and IW are the identity maps on V and W. An examples

sheet question will explore the relation between D (1) ⊗ D (2) and
d(1) ⊗ d(2) in more detail.
An important corollary to Maschke’s theorem states that finite
representations of d(1) ⊗ d(2) can be written as the direct sum of
irreps of L( G ):

d(1) ⊗ d(2) = d˜1 ⊕ d˜2 ⊕ . . . ⊕ d˜k = d˜i .

M
(4.46)
i

We will refer to this as decomposition of a direct product represen-

tation into its constituent irreducible representations.
Angular momentum: SO(3) and SU(2)

SO(3) is the group of rotations in three dimensions. Naturally, rep-

resentations of SO(3) correspond to physical states with orbital
angular momentum, as we will see. In fact, not all angular momen-
tum states can be described by SO(3) representations. Spin angular
momentum states of particles with half-integer spin require SU (2)
representations. This chapter investigates the relation between these
groups and their representations.

5.1 Relationship between SO(3) and SU(2)

The Lie algebra of SU (2) consists of the set of traceless, antihermi-

tian matrices
n  o
su(2) = L(SU (2)) = X ∈ Mat2 (C)  X † = − X, Tr X = 0 (5.1)


i
Ta = i σa (5.2)
2
where the σa are the Pauli matrices.16 16
In most physics literature, the con-
Recall the identity σa σb = Iδab + ieabc σc . vention is to use Hermitian generators
t a = −iTa so that exp( X a Ta ) becomes
1 exp(iX a t a ).
[ Ta , Tb ] = − (σa σb − σb σa )
4
1
= − (ieabc − iebac )σc
4
= eabc Tc . (5.3)
c =e .
We can see that the structure constants for su(2) are f ab abc
The elements of the Lie algebra of SO(3) are the 3 × 3 antisym-
metric, real matrices

so(3) = L(SO(3)) = Skew3 . (5.4)

Basis
     
0 0 0 0 0 1 0 −1 0
T̃1 = 0 0 −1 , T̃2 =  0 0 0 , T̃3 = 1 0 0 ,
     
0 1 0 −1 0 0 0 0 0
(5.5)

or, more briefly ( T̃a )bc = −eabc . After using eacd ebde = −δab δce +
δae δbc in one direction and then the other, we see that [ T̃a , T̃b ] =
38

c = e , as for the basis vectors of su(2). The

eabc T̃c , and thus f ab abc
fact that the two algebras have the same structure constants, and
therefore are isomorphic to each other, suggests that there will be
similarities between the groups SU (2) and SO(3).
Let us consider the group manifolds. We discussed SO(3) as
an example in § 3.2. The manifold is a 3-ball of radius π, with an-
tipodes identified. An element of U ∈ SU (2) can be written as

U = a0 I + i~a ·~σ (5.6)

with ( a0 ,~a) real and a20 + |~a|2 = 1. This manifold is then the unit
sphere in R4 , namely S3 .
Recall that the centre of a group is the set of all x ∈ G such that

xg = gx ∀ g ∈ G . (5.7)

The centre Z ( G ) E G is a normal subgroup of G (since gxg−1 ∈

Z ( G ), ∀ g ∈ G). The group SU (2) has centre Z (SU (2)) = { I, − I } ∼
=
Z2 . If we look at the coset formed for any U ∈ SU (2) with the
centre, we have

U Z (SU (2)) = { U, −U } . (5.8)

The set of all such cosets form a quotient group SU (2)/Z2 (under
coset multiplication) whose manifold is S3 , now with antipodes
identified.
We can draw the manifold as the upper half of S3 (i.e. a0 > 0)
with opposite points on the equator identified. This is just a curved
version of the SO(3) manifold. Therefore,

SO(3) ∼
= SU (2)/Z2 . (5.9)

We can write down an explicit map ρ : SU (2) → SO(3) such that,

for A ∈ SU (2)

1
ρ( A) = R where R has components Rij = Tr(σi Aσj A† ) . (5.10)
2
The map is 2-to-1, since ρ(− A) = ρ( A), and is called a double
covering of SO(3). That is, SU (2) is the double cover of SO(3).
There is a theorem which states that every Lie algebra is the
Lie algebra of exactly one simply-connected Lie group. Any other
Lie group with the same Lie algebra is covered by the simply con-
nected group.

5.2 Representations of L(SU (2))

It will be convenient to work with complex vector spaces rather

than real ones. Let V be a real vector space and let { Ta } be a basis
of V:

V = { λ a Ta | λ a ∈ R } . (5.11)
 39

The complexification of V is the complex span of the same basis

set:

VC = { λ a Ta | λ a ∈ C } . (5.12)

Note that when we talk a vector space being real or complex, we

are referring to the field in which the coefficients λ a live; it has
nothing to do with the { Ta }.
Let g = L( G ) be a real Lie algebra, and denote its complexifica-
tion by gC = L( G )C . A representation d of g can be extended to gC
by imposing

d( X + iY ) = d( X ) + id(Y ) (5.13)

where X, Y ∈ g. Conversely, if we have a representation dC of gC ,

we can restrict it to d by writing

d ( X ) = dC ( X ) (5.14)

where X ∈ g ⊂ gC .
A real form of a complex Lie algebra h is a real Lie algebra g
whose complexification is h, i.e. such that

gC = h . (5.15)

A complex Lie algebra can have multiple nonisomorphic real forms,

as we will see later.
Now to our case, su(2). Its complexification is

su(2)C = { λ a σa | λ a ∈ C } . (5.16)

Note that, while the elements of su(2) are the traceless, antihermi-
tian 2 × 2 matrices, the complexification breaks the antihermiticity
property, extending the algebra to all traceless matrices. This is just
the Lie algebra of SL(2, C). In fact this is true su(n)C , so we have

su(n)C ∼
= sl(n, C) . (5.17)

Within the complex vector space of su(2)C , we can employ a

more useful basis (Cartan–Weyl)
!
1 0
H = σ3 =
0 −1
!
1 0 1
E+ = (σ1 + iσ2 ) =
2 0 0
!
1 0 0
E− = (σ1 − iσ2 ) = . (5.18)
2 1 0

These satisfy

[ H, E± ] = ±2E±
[ E+ , E− ] = H . (5.19)
40

Recall that adX Y = [ X, Y ], then the first relation is equivalent to

ad H E± = ±2E± , (5.20)

and we also have

ad H H = [ H, H ] = 0 . (5.21)

We see that E− , H, E+ are eigenvectors of ad H , with respective

eigenvalues −2, 0, 2. These eigenvalues of H are called the roots of
su(2).
Let d be a finite-dimensional irreducible representation of su(2)
with representation space V. Write an eigenvector of d( H ) as vλ :

d( H )vλ = λvλ (5.22)

The eigenvalues λ of d( H ) are the weights of the representation

d. The operators E± are called ladder or step operators. Let’s see
why. Apply d( H ) to the vectors resulting from applying the ladder
operators to some vλ :

d( H )d( E± )vλ = {d( E± )d( H ) + [d( H ), d( E± )]} vλ

= {d( E± )λ + d ([ H, E± ])} vλ
= (λ ± 2)d( E± )vλ . (5.23)

Therefore, as long as d( E± )vλ 6= 0, we see that d( E± )vλ are eigen-

vectors of d( H ) with eigenvalues λ ± 2.
If we have a finite-dimensional representation, then there are
only a finite number of eigenvalues. Let Λ be the highest weight.
Then we have both

d( H )vΛ = ΛvΛ and d( E+ )vΛ = 0 . (5.24)

The latter relation must be true, otherwise Λ + 2 would be an eigen-

value, contradicting our assumption. Now apply d( E− ) n times to
define

vΛ−2n = (d( E− ))n vΛ . (5.25)

The process must terminate for some n = N, again since the repre-
sentation is finite. This implies a basis for the irrep

{vΛ , vΛ−2 , . . . , vΛ−2N } , (5.26)

where we have

d( H )d( E+ )vΛ−2n = (Λ − 2n + 2)d( E+ )vΛ−2n . (5.27)

However, is the vector we create with the raising operator (5.27)

the same one we generated by repeatedly applying the lowering
operator (5.25)? If not, then we would have degenerate eigenvalues.
Let’s check.

d( E+ )vΛ−2n = d( E+ )d( E− )vΛ−2n+2

= {d( E− )d( E+ ) + [d( E+ ), d( E− )]} vΛ−2n+2
= d( E− )d( E+ )vΛ−2n+2 + (Λ − 2n + 2)vΛ−2n+2 (5.28)
 41

where we have used [d( E+ ), d( E− )] = d( H ) between the second

and third lines.
Eq. 5.28 is a recursion relation we can solve as follows. First
consider n = 1

d( E+ )vΛ−2 = 0 + ΛvΛ . (5.29)

Applying the raising operator to vΛ−2 just gives us a vector parallel

to the vΛ we started with. Let’s look at n = 2

d( E+ )vΛ−4 = d( E+ )d( E− )vΛ−2 + (Λ − 2)vΛ2

= (2Λ − 2)vΛ−2 , (5.30)

using (5.29) to obtain the last line. Again, the raising operator just
undoes the lowering operator up to a multiplicative factor. In gen-
eral we will have the form

d( E+ )vΛ−2n = rn vΛ−2n+2 . (5.31)

Substituting this into (5.28) we find

rn = rn−1 + Λ − 2n + 2 (5.32)

with r1 = Λ from (5.29). The solution is

rn = (Λ + 1 − n)n . (5.33)

Recalling that we can only have a finite number of eigenvalues

and that we denoted the smallest one Λ − 2N, we must have
d( E− )vΛ−2N = 0, which implies

r N +1 = 0 = (Λ − N )( N + 1) (5.34)

Since N + 1 is a positive integer, we have Λ = N, an integer.

We conclude that the finite-dimensional irreducible representa-
tions of su(2) are labelled by Λ ∈ Z>0 as dΛ with corresponding
weights

SΛ = { −Λ, −Λ + 2, . . . , Λ − 2, Λ } . (5.35)

The work above showed that these weights are nondegenerate;

consequently dim dΛ = Λ + 1. Some special cases:

• d0 is the trivial representation.

• d1 is the fundamental representation.

• d2 is the adjoint representation.

Of course we also have higher dimension irreps.

The discussion of this section parallels treatment of spin angular
momentum in quantum mehanics. There one introduces a Hermi-
tian angular momentum operator ~J = ( J1 , J2 , J3 ) which commutes
with the Hamiltonian, and choose one component, say J3 , to com-
plete the set of commuting operators. The eigenvectors are labelled
42

by j and m, where 2j ∈ Z>0 and m ∈ { − j, − j + 1, . . . , j − 1, j }, with

eigenvalues
~J 2 | jmi = j( j + 1) | jmi
J3 | jmi = m | jmi . (5.36)

We identify

d( H ) = 2J3
d( E± ) = J1 ± iJ2 (5.37)

along with Λ = 2j and n = 2m.

5.3 Representations of SU (2) and SO(3)

In this section, we use what we learned about representations of the

Lie algebra su(2) to construct representations of the groups SU (2)
and SO(3).
SU (2) is simply connected, so an su(2) irrep dΛ yields a repre-
sentation DΛ of SU (2) through the exponential map (§ 4.2). Given
g = exp X, with X ∈ su(2) and g ∈ SU (2),

DΛ ( g) = exp dΛ ( X ) . (5.38)

Turning to SO(3), we recall that SO(3) ∼ = SU (2)/Z2 ; any ele-

ment in SO(3), say A, corresponds to a pair of elements {− A, A}
in SU (2). The SU (2) representation DΛ given above is also a repre-
sentation of SO(3), D̃Λ , if and only if it respects the identification of
pairs A and − A

D̃Λ (A) = DΛ ( A) = DΛ (− A) . (5.39)

It is sufficient to check whether DΛ ( I ) = DΛ (− I ), since (5.39) can

then be obtained by multiplying both sides by DΛ ( A).
First we note that, after multiplication by iπ, the basis element
H ∈ su(2) (5.18) maps to

− I = exp iπH ∈ SU (2) . (5.40)

Therefore, the SU (2) and su(2) representations satisfy

DΛ (− I ) = exp iπ dΛ ( H ) . (5.41)

We found that the representation dΛ has weights (i.e. dΛ ( H ) has

eigenvalues) in λ ∈ {−Λ, −Λ + 2, . . . , Λ}. Note that if Λ is odd
(even), then all the λ are odd (even). The eigenvalues of DΛ (− I )
are then given by

exp iπλ = (−1)λ = (−1)Λ . (5.42)

Therefore, we conclude that DΛ (− I ) = DΛ ( I ) if and only if Λ is

even. Only the Λ even representations of su(2) (and SU (2)) can
work as representations of SO(3). These correspond to integer val-
ues of spin angular momentum j. Representations of su(2) with
odd Λ correspond to spinor representations, where the spin angu-
lar momentum j is a half-integer.
 43

5.4 Tensor products of L(SU (2)) representations

Let us investigate the tensor product representations of L(SU (2)).

Recalling the general way in which tensor products work (4.45), we
have, for X ∈ L(SU (2)),

(dΛ ⊗ dΛ0 )( X )(v ⊗ v0 ) = (dΛ ( X )v) ⊗ v0 + v ⊗ dΛ0 ( X )v0

(5.43)

Note that dim(dΛ ⊗ dΛ0 ) = (dim dΛ )(dim dΛ0 ) = (Λ + 1)(Λ0 + 1).

We seek a decomposition
M
Λ 00
dΛ ⊗ dΛ0 = LΛ,Λ 0 d Λ00 (5.44)
Λ00 ∈Z>0

00
where LΛΛ,Λ0 are nonnegative integers, viz multiplicities, (Littlewood–
Richardson coefficients), counting the number of times irrep dΛ00
appears in the decomposition of the tensor product representation.
Given bases for the representation spaces VΛ and VΛ0 respec-
tively

{ v λ } : λ ∈ SΛ such that dΛ ( H )vλ = λvλ

 0 0
v λ 0 : λ ∈ SΛ 0 such that dΛ0 ( H )v0λ0 = λ0 v0λ (5.45)

a basis for tensor product space can be formed

vλ ⊗ v0λ0  λ ∈ SΛ and λ0 ∈ SΛ0 .

 
(5.46)

Combining (5.43) and (5.45)

(dΛ ⊗ dΛ0 )( X )(vλ ⊗ v0λ0 ) = (λ + λ0 )(vλ ⊗ v0λ0 ) (5.47)

Therefore, weights add in the tensor product to give the weight set
of the tensor product representation as

SΛ,Λ0 = λ + λ0  λ ∈ SΛ and λ0 ∈ SΛ0

 
(5.48)

where we count the weights with multiplicity – the same weight

can appear more than once.
The highest weight has multiplicity 1 since there is only one way
+Λ0
to obtain Λ + Λ0 , so LΛ
Λ,Λ0
= 1 and we can partial decompose the
tensor product representation into the irrep dΛ+Λ0 and a remainder
d˜Λ,Λ0

dΛ ⊗ dΛ0 = dΛ+Λ0 ⊕ d˜Λ,Λ0 . (5.49)

The remainder representation has weight set S̃Λ,Λ0 such that

SΛ,Λ0 = SΛ+Λ0 ∪ S̃Λ,Λ0 . (5.50)

The highest weight in S̃Λ,Λ0 is Λ + Λ0 − 2. This weight had multi-

plicity 2 in SΛ,Λ0 , but one instance of it is included in SΛ+Λ0 , so it
occurs only once in S̃Λ,Λ0 . Now the argument repeats itself so that
eventually we find the tensor product representation fully decom-
posed into irreps

d Λ ⊗ d Λ 0 = d Λ + Λ 0 ⊕ d Λ + Λ 0 −2 ⊕ . . . ⊕ d | Λ − Λ 0 | . (5.51)
44

Take, for example, the case Λ = Λ0 = 1, corresponding to the

tensor product of j = j0 = 12 spinor representations. The weight sets

S1 = { −1, 1 }
S1,1 = { −2, 0, 0, 2 } . (5.52)

The highest weight is 2 with multiplicity 1, so (5.50) becomes

S1,1 = { −2, 0, 2 } ∪ { 0 } = S2 ∪ S0 . (5.53)

Thus we find

d1 ⊗ d1 = d2 ⊕ d0 (5.54)

or the combination of two spin- 21 states can yields spin-1 and spin-0
states. Sometimes equations like (5.54) are written using the repre-
sentations’ dimensions, e.g.

2⊗2 = 3⊕1. (5.55)

Relativistic symmetries

In the previous chapter, we investigated the implications of rota-

tional symmetry. In this chapter, we wish to combine rotations with
relativistic boosts and discrete symmetries. At the end, we will also
consider translational invariance.

6.1 Lorentz group

4-vector x µ = ( x0 , { xi })
Lorentz group are the set of transformations x µ 7→ x 0 µ which
leave scalar products such as xµ ηµν x ν invariant, where we take the
Minkowski metric to be
 
1 0 0 0
0 −1 0 0 
ηµν :=  . (6.1)
 
0 0 −1 0 
0 0 0 −1

Let Λ be a linear operator associated with such a transformation

x 0 = Λµ ν x ν .
µ
(6.2)

Invariance of the scalar product implies

x µ ηµν x ν = x σ Λµ σ ηµν Λν ρ x ρ
=⇒ ηρσ = Λµ σ ηµν Λν ρ
η = ΛT ηΛ . (6.3)

The condition (6.3) is exactly that we gave for elements of the

pseudo-orthogonal group O(1, 3) in (3.19). Since the metric is sym-
metric, (6.3) consists of 10 constraints. This implies there are 6 in-
dependent parameters determining the 4 × 4 matrices Λ ∈ O(1, 3).
(We will also see that the dimension of L(O(1, 3)) is equal to 6.)
The Lorentz group is disconnected; it consists of 4 disjoint sets
depending on the signs of det Λ and Λ0 0 .

det(ΛT ηΛ) = det η

det ΛT det Λ = 1
(det Λ)2 = 1
det Λ = ±1 . (6.4)
46

Also, set ρ = σ = 0 in (6.3)

Λµ 0 ηµν Λν 0 = η00 = 1
( Λ0 0 )2 − ∑ ( Λ i 0 )2 = 1
i
( Λ0 0 )2 > 1 (6.5)

so either Λ0 0 > 1 or Λ0 0 6 −1.

The case with det Λ = 1 and Λ0 0 > 1 contains the identity, and
forms the subgroup SO(1, 3)↑ , the proper, orthochronous Lorentz
group. The other parts of O(1, 3) can be obtained from elements of
SO(1, 3)↑ by using the time reversal and parity reversal operators
   
−1 0 0 0 1 0 0 0
 0 1 0 0 0 −1 0 0 
[ T µ ν ] :=   and [ Pµ ν ] := .
   
−1

 0 0 1 0 0 0 0 
0 0 0 1 0 0 0 −1
(6.6)

Special cases of elements of SO(1, 3)↑

1. Rotations
!
1 0
Λ R := with R ∈ SO(3) . (6.7)
0 R

Note that 3 parameters are required to specify and element of

SO(3).

2. Lorentz boosts
!
cosh ψ nT sinh ψ
Λ B := (6.8)
−n sinh ψ I − nnT (cosh ψ − 1)

where n is the column matrix of a unit 3-vector ~n. The boost

velocity is ~v = ~n tanh ψ (in c = 1 units). The boosts also
require 3 parameters: the rapidity ψ and two components of
~n.

While the rotations form a subgroup, SO(3) < SO(1, 3)↑ , the
boosts do not. The boosts do not close under composition (at least
in any spacetime with more than 1 spatial dimension).

6.2 Lie algebra of the Lorentz group

6.3 Poincaré group and algebra

Classification of Lie algebras

In this chapter we extend what we did for SU(2) to a broader class

of Lie algebras. Along the way, we will introduce the notion of
roots as geometric objects in a space isomorphic to Rn and derive
conditions which restrict and characterize Lie algebras.

7.1 Definitions

Here we define a few key terms required in the rest of the chapter.

• A subalgebra of a Lie algebra g is a vector subspace which is

also a Lie algebra under the Lie bracket.

• An ideal (or invariant subalgebra) of a Lie algebra g is a

subalgebra h such that [ X, Y ] ∈ h for all X ∈ g and all Y ∈ h.
An ideal is to an algebra what a normal subgroup is to a
group. Every algebra g has two trivial ideals, {0} and g.

• The derived algebra of a Lie algebra g is

i( g) = [g, g] := spanF { [ X, Y ] | X, Y ∈ g } . (7.1)

This is an ideal of g.

• The centre of g is

J (g) := { X ∈ g | [ X, Y ] = 0, ∀Y ∈ g } . (7.2)

This is also an ideal of g, which can be seen by applying the

Jacobi identity.

• A Lie algebra is Abelian if [ X, Y ] = 0 for all X, Y ∈ g. In this

case J (g) = g and i(g) = {0}.

• A Lie algebra is simple if it is nonabelian and has no non-

trivial ideals. This is equivalent to having i(g) = g and
J (g) = {0}.

• A Lie algebra is semisimple if it has no Abelian, nontrivial

ideals. It can be shown that a semisimple Lie algebra can be
decomposed into the direct sum of simple Lie algebras.
48

7.2 Killing form

i : V×V → F (7.3)

i (v, w) 6= 0 (7.4)

The Killing form of a Lie algebra g is the inner product κ :

g × g → F such that

κ ( X, Y ) = Tr (adX ◦ adY ) . (7.5)

(adX ◦ adY ) Z = [ X, [Y, Z ]]

= X a Y b Z c [ Ta , [ Tb , Tc ]]
= X a Y b Z c f d bc [ Ta , Td ]
= X a Y b Z c f d bc f e ad Te
=: M( X, Y )ec Z c Te (7.6)

defining M( X, Y )ec := X a Y b f d bc f e ad .

κ ab := f d bc f e ad ∈ F . (7.7)

Note that κba = κ ab . The Killing form κ ( X, Y ) meets all the criteria
for being an inner product on the Lie algebra.
Adjoint representation is how elements Z ∈ g act on g. For
X, Y ∈ g

X 7→ X + adZ X
Y 7→ Y + adZ Y . (7.8)

The effect on the Killing form, for an infinitesimal transformation

(Z close to the zero vector of g)

κ ( X, Y ) 7→ κ ( X + adZ X, Y + adZ Y )
≈ κ ( X, Y ) + κ (adZ X, Y ) + κ ( X, adZ Y ) . (7.9)

Invariance of the Killing form under adZ implies

κ (adZ X, Y ) + κ ( X, adZ Y ) = 0
i.e. κ ([ Z, X ], Y ) + κ ( X, [ Z, Y ]) = 0 . (7.10)

We will show this invariance property of the Killing form explicitly

next.
First, observe for Z, W, U ∈ g, by applying the Jacobi identity and
antisymmetry, we have

ad[ Z,W ] U = [[ Z, W ], U ]
= −[[W, U ], Z ] − [[U, Z ], W ]
= [ Z, [W, U ]] − [W, [ Z, U ]]
= (adZ adW − adW adZ )U . (7.11)
 49

Thus,

κ ([ Z, X ], Y ) = Tr(ad[ Z,X ] adY )

= Tr(adZ adX adY ) − Tr(adX adZ adY ) (7.12)
κ ( X, [ Z, Y ]) = Tr(adX adZ adY ) − Tr(adX adY adZ ) . (7.13)

The sum of the two equations above gives 0, in accordance with

(7.10).
Theorem (Cartan): The Killing form of a Lie algebra g is non-
degenerate if and only if g is semisimple. We will only prove this
in one direction, namely that κ nondegenerate implies that g is
semisimple.
Suppose g is not semisimple. There there exists a nontrivial,
Abelian ideal a ⊂ g. That is, for all A ∈ a and all X ∈ g

[ X, A] ∈ a . (7.14)

We will show that this implies the Killing form is degenerate. Let
us choose a basis { Ti } for a and extend it to g:

{ TB } = { Ti | i = 1, . . . , dim a } ∪ { Tα | α = 1, . . . , dim g − dim a }

(7.15)

B is an index running 1, . . . , dim g. The fact that a is Abelian, or

[ Ti , Tj ] = 0 =⇒ f ijB = 0 , (7.16)

and that it is an ideal, or

β
[ Ti , Tα ] = 0 =⇒ f iα = 0 . (7.17)

If we look at the structure constant in (7.16) restricting the super-

script to the basis vectors not in a, B = β, then the combination
with (7.17) implies
β β
f iB = 0 = f Bi . (7.18)

Now we look at the Killing form for A ∈ a and X ∈ g: κ ( X, A) =

κ Bi X B Ai with
C D
κ Bi = f BD f iC
D j
α
= f BD f iα + f BD f ijD
α β α j
= f Bβ f iα + f Bj f iα = 0 . (7.19)

In going from the first line to the second, we split index {C } =

{α} ∪ { j}. In the second line f ijD vanishes by (7.16). We split index
β
{ D } = { β} ∪ { j} to arrive at the third line, where f iα = 0 by (7.17)
α = 0 by (7.18). Therefore, we see that if g is not semisimple,
and f Bj
then κ ( X, A) = 0 for A ∈ a and all X ∈ g, so the Killing form must
be degenerate.
Looking at su(2), for example, the structure constants f ab c = e .
abc
Therefore the coefficients of the Killing form are
c d
κ ab = f ad f bc = eadc ebcd = −2δab , (7.20)
50

so κ ( X, Y ) = −2X a Y b δab which is not generally zero. Since the

Killing form is nondegenerate, we’ve proved that su(2) is semisim-
ple. (In fact, it is simple.)
TODO: Adapted basis

7.3 Casimir elements

TODO: This section

7.4 Cartan–Weyl basis

We begin with some definitions.

• Let g be a Lie algebra. Element X ∈ g is ad-diagonalizable if

map adX : g → g is diagonalizable. In the Cartan–Weyl basis
(5.18) of su(2)C for example, H is ad-diagonalizable, while E±
are not.

• A maximal Abelian subalgebra h is not contained in any

larger subalgebra.

• If g is a complex, semisimple Lie algebra, then a Cartan subal-

gebra (CSA) h of g is a complex space such that

1. h is Abelian: for all H1 and H2 ∈ h, their Lie bracket van-

ishes: [ H1 , H2 ] = 0.
2. h is a maximal Abelian subalgebra: For any X ∈ g, if
[ H, X ] = 0 for all H ∈ h, then X ∈ h.
3. All H ∈ h are ad-diagonalizable.

It can be proved that semisimple Lie algebras have Cartan

subalgebras, but we will not do this here.

• The dimension of an algebra’s CSA is called the rank of the

Lie algebra: rank g = dim h.

TODO: Examples su(2)C and su(n)C .

Eigenvectors of ad H
As we did for su(2) in (5.21), we wish to study the eigenvectors of
the adjoint representation of h. For any H, H 0 ∈ h

ad H H 0 = [ H, H 0 ] = 0 (7.21)

which implies

[ad H , ad H 0 ] = 0 . (7.22)

All the adjoint maps communte and are therefore simultaneously

diagonalizable. This means that the Lie algebra g is spanned by the
set of simultaneous eigenvectors of the {ad H }. The elements of the
 51

CSA h are the zero-eigenvectors. Within this subspace, let us choose

a basis

{ Hi | i = 1, . . . , r } (7.23)

where r = dim h = rank g. Since the CSA is maximal, no other

eigenvectors of {ad H } can have eigenvalue zero. Denote the rest of
the eigenvectors Eα , where the subscript α labels each eigenvector
and is called a root. The corresponding eigenvalue αi will generally
depend on our choice of basis (7.23)

ad Hi Eα = [ Hi , Eα ] = αi Eα (7.24)

where αi ∈ C. It is a fact, which we will not prove here, that the

nonzero simultaneous eigenvectors of {ad H } are nondegenerate,
unique up to a normalization. We will call these eigenvectors lad-
der or step operators. The collection of possible roots α is called the
root set and denoted Φ.
We wish to show that the roots are vectors in the dual vector
space h∗ . Let us recall the definition. Given a vector space V over
a field F, the dual vector space V ∗ is the vector space of linear
functions f : V → F. As long as V is finite-dimensional, dim V ∗ =
dim V. Given a basis {vi } for V, we can find a dual basis in V ∗ ,
{vi∗ } such that vi∗ (v j ) = δij .
Any element H ∈ h can be written as a linear combination of
basis vectors (7.23) as17 17
I have switched from coefficients ei
to ρi since e is used in other contexts in
H = ρi Hi , with ρi ∈ C . (7.25) this chapter.

Then the action of this general ad H gives

[ H, Eα ] = ρi [ Hi , Eα ] = ρi αi Eα =: α( H ) Eα (7.26)

defining α( H ) := ρi αi ∈ C. We see that α( H ) : h → C, but is it

linear? Yes. Here is the proof. Take any H and H 0 ∈ h. Then

α( H + H 0 ) Eα = [ H + H 0 , Eα ]
= [ H, Eα ] + [ H 0 , Eα ]
= α( H ) + α( H 0 ) Eα .


(7.27)

Therefore the roots α ∈ Φ are vectors in the dual vector space h∗ .

We will have a lot more to say about this in the next section.
The Cartan–Weyl basis for g is given by

{ Hi | i = 1, . . . , r } ∪ { Eα | α ∈ Φ } . (7.28)

TODO: 5 lemmas regarding the Killing form

[ Hi , Hj ] = 0 (7.29)
[ Hi , Eα ] = αi Eα (7.30)

 Nα,β Eα+ β
 α+β ∈ Φ
[ Eα , Eβ ] = κ ( Eα , Eβ ) Hα α+β = 0 (7.31)

 0 otherwise
52

[ Hα , Eβ ] = (κ −1 )ij α j [ Hi , Eβ ]
= (κ −1 )ij α j β i Eβ
= (α, β) Eβ . (7.32)

We see the inner product on h∗ is induced by the inner product κ.

It is convenient to choose another normalization such that
2
hα := Hα
(α, α)
s
2
eα : = Eα . (7.33)
(α, α) κ ( Eα , E−α )

The Lie brackets (7.31) in this basis are then

[ hα , h β ] = 0 (7.34)
2(α, β)
[ hα , e β ] = e (7.35)
(α, α) β

 nα,β eα+ β
 α+β ∈ Φ
[ eα , e β ] = hα α+β = 0 . (7.36)

 0 otherwise

TODO: subalgebras su(2)α

7.5 Geometry of roots

7.6 Simple roots

We have more roots than are needed to span hR ∗ . We can pick a

hyperplane of dimension r − 1 to divide the roots into two halves,

one we arbitrarily call the positive half, the other negative. The
hyperplane should go through 0 and not contain any roots in it.
This is possible since there are a finite number of roots. If a root α
∗ , we say α is a positive root and − α is a
is in the positive half of hR
negative root. All roots are then divided into equal size root sets

Φ = Φ+ ∪ Φ− . (7.37)

We have two properties (one stated above)

α ∈ Φ+ =⇒ −α ∈ Φ− (7.38)
α, β ∈ Φ+ =⇒ α + β ∈ Φ+ . (7.39)

A simple root is a positive root which cannot be written as the

sum of two positive roots. The set of simple roots is denoted ΦS .
We will derive a few properties of the simple roots.

(i) If α and β are simple roots, then α − β is not a root. The

proof is by contradiction. Suppose α − β ∈ Φ+ . Then

α = (α − β) + β (7.40)
 53

which implies α is not a simple root, as it is the sum of

two positive roots. Otherwise suppose α − β ∈ Φ− . Then
β − α ∈ Φ+ and

β = ( β − α) + α (7.41)

which contradicts β ∈ ΦS . Therefore α − β cannot be a

root. This property will be useful when we write down the
Lie brackets using simple roots.

(ii) For α, β ∈ ΦS , the α-root string through β has length

2(α, β)
`α,β = 1 − ∈ Z+ . (7.42)
(α, α)
Recall the root string is the set

Sα,β = { β + nα | n ∈ Z, n− 6 n 6 n+ } (7.43)

where n− 6 0 and n+ > 0 and

2(α, β)
n+ + n− = − ∈ Z. (7.44)
(α, α)
From (i) we know β − α is not a root, so n− = 0. Then

2(α, β)
n+ = − ∈ Z>0 . (7.45)
(α, α)
We add 1 to obtain the length

`α,β = n+ − n− + 1 ∈ Z+ (7.46)

and the expression (7.42). The length property will be

useful in the next section, and it has a useful corollary as
the next property (iii).

(iii) The fact that (α, β) 6 0 for α, β ∈ ΦS and β 6= α follows

immediately from (ii).

(iv) Any positive root, β ∈ Φ+ , can be written as a linear

combination of simple roots with positive integer coeffi-
cients. Proof: If β ∈ ΦS , we are done. If, on the other hand,
β = β 1 + β 2 ; if both β 1 and β 2 are simple, we are done,
otherwise we iterate until all the summands are simple.

(v) The next three properties show that the simple roots form
∗ . Anticipating this, let us denote the set of
a basis for hR
simple roots with as
n  o
ΦS = α(i)  i = 1, . . . , |ΦS | . (7.47)


First we show that all roots α ∈ Φ can be written as a

linear combination

α= ∑ a i α (i ) (7.48)
i
54

with ai ∈ Z. This follows straightforwardly from . (Note

that, by (i), the nonzero coefficients ai are either all posi-
∗ is spanned by the roots, we
tive, or all negative.) Since hR
now see that the simple roots are sufficient to span the
space.

(vi) The simple roots are linearly independent. By (vi), all dual
∗ can be written as
vectors λ ∈ hR

λ= ∑ c i α (i ) , with ci ∈ R . (7.49)
i

We now prove the simple roots are linear independent by

showing

λ = 0 ⇐⇒ ci = 0 ∀i . (7.50)

Assume that one or more ci 6= 0, and separate the positive

and negative coefficients into two sets

J± := { i | c ≷ 0 } (7.51)

and write

λ+ := ∑ ci α(i) and
i ∈ J+

λ− := − ∑ c i α (i ) = ∑ bi α ( i ) , (7.52)
i ∈ J− i ∈ J−

where we define positive coefficients bi = −ci for i ∈ J− .

Then

λ = λ+ − λ− = ∑ c i α (i ) − ∑ bi α ( i ) . (7.53)
i ∈ J+ i ∈ J−

∗ we have
Using the inner product on hR

(λ, λ) = (λ+ , λ+ ) + (λ− , λ− ) − 2(λ+ , λ− )

> −2( λ + , λ − )
= −2 ∑ ∑ c i b j ( α (i ) , α ( j ) ) > 0 (7.54)
i ∈ J+ j∈ J−

since ci b j > 0 by construction and (α(i) , α( j) ) 6 0 by

(iii). Since at least one of (λ+ , λ+ ) or (λ− , λ− ) is positive
definite, we have shown (λ, λ) > 0. Therefore, the only
way λ = 0 is if all coefficients ci vanish; consequently, the
simple roots are linearly independent.

∗
(vii) A corollary of (v) and (vi), is that there are r = dim hR
simple roots, i.e.

|ΦS | = r , (7.55)

∗.
and that they form a basis for hR
 55

7.7 Classification

In the previous section we showed that the simple roots form a

∗ , so the next step is to write the Lie algebra in this basis,
basis for hR
known as the Chevalley basis.
We define the r × r Cartan matrix A (r = dim hR ∗ ) which has

matrix elements
2( α ( j ) , α (i ) )
A ji = . (7.56)
( α (i ) , α (i ) )

Note that A is not generally symmetric. We know from earlier that

A ji ∈ Z.
Now using (7.36) and focussing on the simple roots, we have

[ h α (i ) , h α ( j ) ] = 0
[hα(i) , e±α( j) ] = ± A ji e±α( j)
[eα(i) , e−α( j) ] = δij hi , (7.57)

with no sums over indices above. The first and second relations are
straightforward transcriptions of (7.36). The third one is clear in the
/ Φ.
case where i = j. For i 6= j we use the fact that α(i ) − α( j) ∈
From (7.36) we have

[eα(i) , eα( j) ] = adeα(i) (eα( j) ) ∝ eα(i)+α( j) (7.58)

as long as α(i ) + α( j) ∈ Φ. In this case, then α(i ) + α( j) is part

of a root string which we can write without loss of generality as
nα(i ) + α( j) whose length we know (7.42), and corresponds to
n = 0, . . . , − A ji . For ` = 1 − A ji applications of adeα(i) , then the
string ends, so we obtain the Serre relation

(adeα(i) )1− A ji (eα( j) ) = 0 . (7.59)

The relations (7.57) and (7.59) completely characterize the Lie al-
gebra. It can further be proven that any finite-dimensional, simple,
complex Lie algebra is uniquely determined by its Cartan matrix.
We proceed by classifying the possible Cartan matrices, and
showing how to reconstruct the corresponding Lie algebras.
Bibliography

B Allanach. Symmetries, Particles and Fields. 2021. ISBN 978-1-739-

94069-0.

B C Hall. Lie Groups, Lie Algebras, and Representations: An Elemen-

tary Introduction. Springer, 2015. ISBN 978-3319134666.

H F Jones. Groups, Representations, and Physics. IOP Publishing,

1990. ISBN 0-85274-029-8.

P Ramond. Group Theory: A Physicist’s Survey. Cambridge

University Press, 2010. ISBN 978-0-521-89603-0. URL
https://www.cambridge.org/core/books/group-theory/
8BAC137A9F0C43D65448E6420248D841.

B Schutz. Geometrical Methods of Mathematical Physics. Cambridge

University Press, 1980. ISBN 0-521-29887-3. URL https://doi.
org/10.1017/CBO9781139171540.