Group Theory Lecture Notes

Hugh Osborn

latest update: November 23, 2023

Based on part III lectures Symmetries and Groups, Michaelmas Term 2008, revised and
extended at various times subsequently
Books

Books developing group theory by physicists from the perspective of particle physics are

H. F. Jones, Groups, Representations and Physics, 2nd ed., IOP Publishing (1998).
A fairly easy going introduction.
H. Georgi, Lie Algebras in Particle Physics, Perseus Books (1999).
Describes the basics of Lie algebras for classical groups.
J. Fuchs and C. Schweigert, Symmetries, Lie Algebras and Representations, 2nd ed., CUP
(2003).
This is more comprehensive and more mathematically sophisticated, and does not describe
physical applications in any detail.
Z-Q. Ma, Group Theory for Physicists, World Scientific (2007).
Quite comprehensive.
P. Ramond, Group Theory, A Physicists Survey, CUP (2010).
A relatively gentle physics motivated treatment, and includes discussion of finite groups.
A. Zee, Group Theory in a Nutshell for Physicists. Princeton University Press (2016).
Quite lengthy, comprehensive with many physics applications, some nice anecdotal remarks.
P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Lie Groups, Princeton Uni-
versity Press (2009), [Link]
Idiosyncratic, but full of material not found elswhere. Great for doing calculations.

The following books contain useful discussions, in chapter 2 of Weinberg there is a proof
of Wigner’s theorem and a discussion of the Poincaré group and its role in field theory,
and chapter 1 of Buchbinder and Kuzenko has an extensive treatment of spinors in four
dimensions.
S. Weinberg, The Quantum Theory of Fields, (vol. 1), CUP (2005).
J. Buchbinder and S. Kuzenko, Ideas and Methods of Supersymmetry and Supergravity, or
a Walk Through Superspace, 2nd ed., Institute of Physics Publishing (1998).

They are many mathematical books with titles containing references to Groups, Represen-
tations, Lie Groups and Lie Algebras. The motivations and language is often very different,
and hard to follow, for those with a traditional theoretical physics background. Particular
books which may be useful are
B.C. Hall, Lie Groups, Lie Algebras, and Representations, Springer (2004), for an earlier
version see arXiv:math-ph/0005032.
This focuses on matrix groups.
More accessible than most
W. Fulton and J. Harris, Representation Theory, Springer (1991).

Historically the following book, first published in German in 1931, was influential in showing
the relevance of group theory to atomic physics in the early days of quantum mechanics. It
introduces anti-unitary representations. For an English translation
E.P. Wigner, Group Theory and its Application to the Quantum Mechanics of Atomic Spec-
tra, Academic Press (1959).

i
Prologue

The following excerpts are from Strange Beauty, by G. Johnson, a biography of Murray
Gell-Mann1 , the foremost particle physicist of the 1950’s and 1960’s who proposed SU (3)
as a symmetry group for hadrons and later quarks as the fundamental building blocks. It
reflects a time when most theoretical particle physicists were unfamiliar with groups beyond
the rotation group, and perhaps also a propensity for some to invent mathematics as they
went along.

As it happened, SU (2) could also be used to describe the Isopspin symmetry- the group
of abstract ways in which a nucleon can be “rotated” in isospin space to get a neutron or
a proton, or a pion to get negative, positive or neutral versions. These rotations were what
Gell-Mann had been calling currents. The groups were what he had been calling algebras.
He couldn’t believe how much time he had wasted. He had been struggling in the dark
while all these algebras, these groups- these possible classification schemes- had been studied
and tabulated decades ago. All he would have to do was to go to the library and look them
up.
In Paris, as Murray struggled to expand the algebra of the isospin doublet, SU (2), to
embrace all hadrons, he had been playing with a hierarchy of more complex groups, with
four, five, six, seven rotations. He now realized that they had been simply combinations
of the simpler groups U (1) and SU (2). No wonder they hadn’t led to any interesting new
revelations. What he needed was a new, higher symmetry with novel properties. The next
one in Cartan’s catalogue was SU (3), a group that can have eight operators.
Because of the cumbersome way he had been doing the calculations in Paris, Murray
had lost the will to try an algebra so complex and inclusive. He had gone all the way up to
seven and stopped.

1
Murray Gell-Mann, 1929-2019, American, Nobel prize 1969.

ii
Contents
0 Notational Conventions 0

1 Introduction, Definitions and Examples 1

1.1 Basic Definitions and Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Subgroups and Cosets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Normal Subgroups, Quotient Group, Simple Groups and Composition
Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Direct Product of Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Cyclic, Dihedral and Permutation Groups . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Cyclic Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Dihedral Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.3 Symmetric Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.4 Alternating Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Orbit Stabiliser Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Further Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 Automorphisms and Semi-Direct Product . . . . . . . . . . . . . . . . . 10
1.4.2 Wreath Products and Central Products . . . . . . . . . . . . . . . . . . 11
1.4.3 Conjugacy Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.4 Centre, Normaliser, Centraliser and Commutator Subgroups . . . . . 15
1.4.5 Double Coset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.6 Goursat’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Quaternion Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5.1 Tetrahedral, Octahedral and Icosahedral Groups . . . . . . . . . . . . 19
1.6 Matrix Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6.1 Orthogonal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6.2 Unitary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.3 Symplectic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.4 Quaternion Matrix Groups . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6.5 Heisenberg Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.6.6 Compact and Non Compact . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Representations 27
2.1 Schur’s Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Induced Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Unitary Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.1 Infinite Dimensional Unitary Representations . . . . . . . . . . . . . . 34
2.3.2 Real and Pseudo-Real Representations . . . . . . . . . . . . . . . . . . 34
2.4 Orthogonality Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5.1 Further Constraints on Dimensions of Irreducible Representations . . 38
2.6 Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

iii
 2.6.1 Symmetric and Antisymmetric Products . . . . . . . . . . . . . . . . . 40
2.7 Character Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.8 Molien Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.8.1 Anticommuting Molien Series . . . . . . . . . . . . . . . . . . . . . . . . 44
2.8.2 Examples of Molien Series . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.8.3 Molien Series and Wreath Products . . . . . . . . . . . . . . . . . . . . 48
2.9 Symmetries in Quantum Mechanics, Projective and Anti-Unitary Represen-
tations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.9.1 Projective Representations . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.9.2 Anti-Unitary Representations . . . . . . . . . . . . . . . . . . . . . . . . 52

3 Rotations and Angular Momentum, SO(3) and SU (2) 62

3.1 Three Dimensional Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2 Isomorphism of SO(3) and SU (2)/Z2 . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.1 Non Compact Isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Infinitesimal Rotations and Generators . . . . . . . . . . . . . . . . . . . . . . . 67
3.4 Representations of Angular Momentum Commutation Relations . . . . . . . 68
3.5 The ∣j m⟩ basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.5.1 Action of Time Reversal . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5.2 Representation Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5.3 Integration over SO(3) and orthogonality relations . . . . . . . . . . . 74
3.5.4 Characters for SU (2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.6 Tensor Products and Angular Momentum Addition . . . . . . . . . . . . . . . 76
3.7 Examples of the calculation of Clebsch-Gordan coefficients . . . . . . . . . . . 79
3.7.1 Construction of Singlet States . . . . . . . . . . . . . . . . . . . . . . . . 81
3.7.2 Construction of Highest Weight States . . . . . . . . . . . . . . . . . . 81
3.7.3 Special Cases of Clebsch-Gordan Coefficients . . . . . . . . . . . . . . . 82
3.8 3j Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.9 6j Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.9.1 Crossing Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.10 Tensor Products and Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.11 SO(3) Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.11.1 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.12 Molien Series for SU (2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.13 Irreducible Tensor Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.14 Spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.15 Spinor Representation for Angular Momentum . . . . . . . . . . . . . . . . . . 100
3.15.1 Calculation of Action of Derivatives . . . . . . . . . . . . . . . . . . . . 104
3.16 Isospin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.16.1 G-parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

4 Relativistic Symmetries, Lorentz and Poincaré Groups 110

4.1 Lorentz Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

iv
 4.1.1 Proof of Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.1.2 Structure of Lorentz Group . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2 Infinitesimal Lorentz Transformations and Commutation Relations . . . . . . 112
4.3 Lorentz Group and Spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.3.1 Isomorphism SO(3, 1) ≃ Sl(2, C)/Z2 . . . . . . . . . . . . . . . . . . . . 115
4.3.2 Spinors, Dotted and Undotted Indices . . . . . . . . . . . . . . . . . . . 117
4.3.3 Tensorial Representations . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.4 Poincaré Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.5 Irreducible Representations of the Poincaré Group . . . . . . . . . . . . . . . . 121
4.5.1 Massive Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.5.2 Helicity States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.5.3 Massless Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.5.4 Two Particle States and Angular Momentum Decomposition . . . . . 128
4.5.5 Spinorial Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.6 Casimir Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.7 Quantum Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5 Lie Groups and Lie Algebras 136

5.0.1 Vector Fields, Differential Forms and Lie Brackets . . . . . . . . . . . 137
5.1 Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.2 Lie Algebra Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.3 Matrix Lie Algebras and Matrix Lie Groups . . . . . . . . . . . . . . . . . . . . 143
5.3.1 SU (2) Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.3.2 Upper Triangular Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.3.3 Representations and Lie Algebras . . . . . . . . . . . . . . . . . . . . . 144
5.4 Relation of Lie Algebras to Lie Groups . . . . . . . . . . . . . . . . . . . . . . . 146
5.4.1 One-Parameter Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.4.2 Baker Cambell Hausdorff Formula . . . . . . . . . . . . . . . . . . . . . 147
5.5 Simply Connected Lie Groups and Covering Groups . . . . . . . . . . . . . . . 148
5.5.1 Covering Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.5.2 Projective Representations . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.6 Lie Algebra and Projective Representations . . . . . . . . . . . . . . . . . . . . 152
5.6.1 Galilean Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.7 Integration over a Lie Group, Compactness . . . . . . . . . . . . . . . . . . . . 154
5.7.1 SU (2) Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.7.2 Non Compact Sl(2, R) Example . . . . . . . . . . . . . . . . . . . . . . 157
5.8 Adjoint Representation and its Corollaries . . . . . . . . . . . . . . . . . . . . . 158
5.8.1 Killing Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.8.2 Conditions for Non Degenerate Killing Form . . . . . . . . . . . . . . . 160
5.8.3 Decomposition of Semi-simple Lie Algebras . . . . . . . . . . . . . . . . 161
5.8.4 Casimir Operators and Central Extensions . . . . . . . . . . . . . . . . 162

6 Lie Algebras for Matrix Groups 163

v
 6.1 Unitary Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.2 Symplectic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.3 Orthogonal and Spin Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.3.1 Spin Groups and Gamma Matrices . . . . . . . . . . . . . . . . . . . . . 166
6.3.2 Products and Traces of Gamma Matrices . . . . . . . . . . . . . . . . . 168
6.3.3 Construction of Representations of the Clifford Algebra . . . . . . . . 171
6.3.4 Conjugation Matrix for Gamma Matrices . . . . . . . . . . . . . . . . . 172
6.3.5 Special Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.3.6 Fierz Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

7 SU (3) and its Representations 179

7.1 Recap of su(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.2 A su(3) Lie algebra basis and its automorphisms . . . . . . . . . . . . . . . . . 182
7.3 Highest Weight Representations for su(3) . . . . . . . . . . . . . . . . . . . . . 184
7.3.1 Analysis of the Weight Diagram . . . . . . . . . . . . . . . . . . . . . . 188
7.3.2 SU (3) Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.3.3 Casimir operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.3.4 Particular SU (3) Representations . . . . . . . . . . . . . . . . . . . . . 193
7.4 SU (3) Tensor Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.4.1 su(3) Lie algebra again . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.5 SU (3) and Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.5.1 SU (3)F Symmetry Breaking . . . . . . . . . . . . . . . . . . . . . . . . 200
7.5.2 SU (3) and Colour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.6 Tensor Products for SU (3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.6.1 Systematic Discussion of Tensor Products . . . . . . . . . . . . . . . . 203

8 Gauge Groups and Gauge Theories 205

8.1 Abelian Gauge Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
8.2 Non Abelian Gauge Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.2.1 Chern-Simons Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.3 Gauge Invariants and Wilson Loops . . . . . . . . . . . . . . . . . . . . . . . . 216
8.3.1 Wilson Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

9 Integrations over Spaces Modulo Group Transformations 222

9.1 Integrals over Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
9.2 Integrals over Symmetric and Hermitian Matrices . . . . . . . . . . . . . . . . 229
9.2.1 Large n Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
9.3 Integrals over Compact Matrix Groups . . . . . . . . . . . . . . . . . . . . . . . 235
9.4 Integration over a Gauge Field and Gauge Fixing . . . . . . . . . . . . . . . . 241

vi
0 Notational Conventions

Hopefully we use standard conventions. For any Mij , i belonging to an ordered set with
m elements, not necessarily 1, . . . m, and similarly j belonging to an ordered set with n
elements, M = [Mij ] is the corresponding m × n matrix, with of course i labelling the rows,
j the columns. 1 is the unit matrix, on occasion 1n denotes the n × n unit matrix.
For any multi-index Ti1 ...in then T(i1 ...in ) , T[i1 ...in ] denote the symmetric, antisymmetric
parts, obtained by summing over all permutations of the order of the indices in Ti1 ...in ,
with an additional −1 for odd permutations in the antisymmetric case, and then dividing
by n!. Thus for n = 2,

T(ij) = 12 (Tij + Tji ) , T[ij] = 12 (Tij − Tji ) . (0.1)

If some indices are to be omitted from symmetrisation or antisymmetrisation they are

surrounded by ∣ . . . ∣, thus T[i∣k∣j] = 21 (Tikj − Tjki ).
We use µ, ν, σ, ρ as space-time indices, i, j, k are spatial indices while α, β, γ are spinorial
indices.
For a set of elements x then {x ∶ P } denotes the subset satisfying a property P .
A vector space V may be defined in terms of linear combinations of basis vectors {vr },
r = 1, . . . , dim V so that an arbitrary vector can be expressed as ∑r ar vr . For two vector
spaces V1 , V2 with bases {v1r }, {v2s } we may define the tensor product space V1 ⊗ V2 in
terms of the basis of pairs of vectors {(v1r , v2s )} for all r, s and we require the equivalence
relations (v1 + v ′2 , v2 ) ∼ (v1 , v2 ) + (v ′1 , v2 ), (v1 , v2 + v ′2 ) ∼ (v1 , v2 ) + (v1 , v ′2 ), c(v1 , v2 ) ∼
(c v1 , v2 ) ∼ (v1 , c v2 ) so as to extend the vector space properties from V1 , V2 to V1 ⊗ V2 .
An arbitrary vector in V1 ⊗ V2 is then a linear combination v = ∑r,s ars (v1r , v2s ) so that
dim(V1 ⊗V2 ) = dim V1 dim V2 . The tensor product V ⊗V can be decomposed into symmetric
snd antisymmetric subspaces ⋁2 V and ⋀2 V with bases (vr , vs )+(vs , vr ) and (vr , vs )−(vs , vr )
respectively, corresponding to ars = a(rs) and ars = a[rs] . dim ⋁2 V = 12 dim V(dim V + 1) and
dim ⋀2 V = 12 dim V(dim V − 1).
The direct sum V1 ⊕ V2 is defined so that if v ∈ V1 ⊕ V2 then v = v1 + v2 with vi ∈ Vi , and
where (v1 + v ′1 ) + (v2 + v ′2 ) = (v1 + v2 ) + (v ′1 + v ′2 ) and c(v1 + v2 ) = c v1 + c v2 . Equivalently it
has a basis {v1r , v2s } and an arbitrary vector in V1 ⊗ V2 has the form v = ∑r ar v1r + ∑s bs v2s
so that dim(V1 ⊕ V2 ) = dim V1 + dim V2 .

0
1 Introduction, Definitions and Examples

There are nowadays very few papers in theoretical particle physics which do no not mention
groups or Lie algebras and correspondingly make use of the mathematical language and
notation of group theory, and in particular of that for Lie groups. Groups are relevant
whenever there is a symmetry of a physical system, symmetry transformations correspond
to elements of a group and the combination of one symmetry transformation followed by
another corresponds to group multiplication. Associated with any group there are sets
of matrices which are in one to one correspondence with each element of the group and
which obey the same the same multiplication rules. Such a set a of matrices is called a
representation of the group. An important mathematical problem is to find or classify all
groups within certain classes and then to find all possible representations. How this is
achieved for Lie groups will be outlined in these lectures although the emphasis will be
on simple cases. Although group theory can be considered in the abstract, in theoretical
physics finding and using particular matrix representations are very often the critical issue.
In fact large numbers of groups are defined in terms of particular classes of matrices.
Group theoretical notions are relevant in all areas of theoretical physics but they are
particularly important when quantum mechanics is involved. In quantum theory physical
systems are associated with vectors belonging to a vector space and symmetry transforma-
tions of the system are associated with linear transformations of the vector space. With a
choice of basis these correspond to matrices so that directly we may see why group repre-
sentations are so crucial. Historically group theory as an area of mathematics particularly
relevant in theoretical physics first came to the fore in the 1930’s directly because of its ap-
plications in quantum mechanics (or matrix mechanics as the Heisenberg formulation was
then sometimes referred to). At that time the symmetry group of most relevance was that
for rotations in three dimensional space, the associated representations, which are associ-
ated with the quantum mechanical treatment of angular momentum, were used to classify
atomic energy levels. The history of nuclear and particle physics is very much a quest to
find symmetry groups. Initially the aim was to find a way of classifying particles with nearly
the same mass and initially involved isospin symmetry. This was later generalised to the
symmetry group SU (3), the eightfold way, and famously led to the prediction of a new
particle the Ω− . The representations of SU (3) are naturally interpreted in terms of more
fundamental particles the quarks which are now the basis of our understanding of particle
physics.
Apart from symmetries describing observed particles, group theory is of fundamental
importance in gauge theories. All field theories which play a role in high energy physics are
gauge field theories which are each associated with a particular gauge group. Gauge groups
are Lie groups where the group elements depend on the space-time position and the gauge
fields correspond to a particular representation, the adjoint representation. To understand
such gauge field theories it is essential to know at least the basic ideas of Lie group theory,
although active research often requires going considerably further.

1
1.1 Basic Definitions and Terminology

A group G is a set of elements {gi } (here we suppose the elements are labelled by a discrete
index i but the definitions are easily extended to the case where the elements depend on
continuously varying parameters) with a product operation such that

gi , gj ∈ G ⇒ gi gj ∈ G . (1.1)

Further we require that there is an identity e ∈ G such that for any g ∈ G

eg = ge = g , (1.2)

and also g has an inverse g −1 so that

gg −1 = g −1 g = e . (1.3)

Furthermore the product must satisfy associativity

gi (gj gk ) = (gi gj )gk for all gi , gj , gk ∈ G , (1.4)

so that the order in which a product is evaluated is immaterial. A group is abelian if

gi gj = gj gi for all gi , gj ∈ G . (1.5)

For a finite discrete group with n elements then n = ∣G∣ is the order of the group.
For any g ∈ G the smallest integer m such that g m = e is the order of g.
Two groups G = {gi } and G′ = {gj′ } are isomorphic, G ≃ G′ , if there is a one to one
correspondence θ ∶ gi ↔ gj′ between the elements consistent with the group multiplication
rules. Even if G ≃ G′ there is not necessarily a unique choice for θ but of course we must
have θ ∶ e ↔ e′ .
A crucial consequence of the basic group axioms is

{gi g} = {gi } for any g since gj g = gi g ⇒ gj = gi , (1.6)

which implies for a finite group

∑i f (gi ) = ∑i f (gi g) . (1.7)

1.1.1 Subgroups and Cosets

For any group G a subgroup H ⊂ G is naturally defined as a set of elements belonging to G

which is also a group. A proper subgroup H is when H ≠ G and is denoted H < G. For any
subgroup H there is an equivalence relation between gi , g ′i ∈ G,

gi ∼ g ′i ⇔ g ′i = gi h for h ∈ H . (1.8)

2
Each equivalence class {gi } defines a left coset and has ∣H∣ elements. Of course {e} = H.
There can also be right cosets where in (1.8) we take gi = h g ′i instead. The cosets form the
coset space G/H,

G/H ≃ G/∼ = {gH ∶ g ∈ G, g ∼ g ′ if g ′ = gh, h ∈ H} . (1.9)

Since each coset is distinct

dim G/H = ∣G/H∣ = ∣G∣/∣H∣ . (1.10)
The fact that, for any subgroup H ⊂ G, ∣H∣ divides ∣G∣ is Lagrange’s theorem.1 The index of
the subgroup H in G is the number of cosets in G/H, denoted G ∶ H = ∣G∣/∣H∣. The index
is also a divisor of ∣G∣. In general left and right cosets are different.
In general G/H is not a group since gi ∼ g ′i , gj ∼ g ′j does not imply gi gj ∼ g ′i g ′j .

1.1.2 Normal Subgroups, Quotient Group, Simple Groups and Composition

Series

A normal or invariant subgroup is a subgroup N ⊂ G such that

gN g −1 = N for all g ∈ G . (1.11)

This may be denoted by N ◁ G (or G ▷ N ). In this case G/N becomes a group since for
g ′i = gi hi , g ′j = gj hj , with hi , hj ∈ N , then g ′i g ′j = gi gj h for some h ∈ N . Q = G/N is then
the quotient group, or sometimes the factor group. The group G is an extension of Q by N .
The quotient group is expressible in terms of cosets by

G/N ≃ {gN /∼ ∶ g ∈ G, g ∼ gh, h ∈ N } , (1.12)

where elements of the quotient group satisfy

(gN ) (g ′ N ) = (gg ′ N ) , (gN )−1 = (g −1 N ) , e = (N ) , (1.13)

with the group multiplication rule and inverse following from N g = gN , for N a normal
subgroup, and N 2 = N . In general the quotient group Q is not a subgroup of G. For an
abelian group all subgroups are necessarily normal subgroups. A normal subgroup N is
maximal if there is no N ′ ≠ N, G such that N ◁ N ′ ◁ G. As shown later there can be more
than one maximal normal subgroup.
For a normal subgroup the left and right cosets are identical since gh = h′ g for any g ∈ G
and for any h ∈ N there is a corresponding h′ ∈ N .
If H is a subgroup of G and ∣G∣/∣H∣ = 2 then H has to be a normal subgroup since the
right coset other than H has to be equal to the left coset. In this case the quotient group
G/H ≃ Z2 and for g, g ′ ∈ G, g, g ′ ∉ H then gg ′ ∈ H. In this case H is sometimes called the
halving subgroup.
A group G is simple if the only normal subgroups are G and the trivial subgroup formed
by the identity {e} by itself. Simple groups are the building blocks for finite groups. If N is
1
Joseph-Louis Lagrange, born Giuseppe Luigi Lagrangia, 1736-1813, French, after Italian.

3
a maximal normal subgroup of G then the quotient G/N is a simple group. To verify this
if N is not maximal there is a normal subgroup N ′ ◁ G such that N is a subgroup of N ′ ..
Since N is a normal subgroup of G it must also be a normal subgroup of N ′ , N ◁ N ′ . The
corresponding quotient groups can be expressed as

G/N ≃ {gN /∼ ∶ g ∈ G, g ∼ gh, h ∈ N } , N ′ /N ≃ {gN /∼ ∶ h′ ∈ N ′ , h′ ∼ h′ h, h ∈ N } . (1.14)

Directly
(gN ) h′ N (gN )−1 = (gh′ g −1 )N ∈ N ′ /N , (1.15)
′ ′
since N is a normal subgroup. Thus N /N ◁ G/N . Conversely if G/N has a non trivial
normal subgroup then there is a normal subgroup N ′ ◁ G with N a non trivial normal
subgroup of N ′ . Hence if N is maximal G/N is simple.
For any G there is a composition series of successive maximal normal subgroups Ni ◁
Ni−1 , N0 = G where
Nn ◁ ⋅ ⋅ ⋅ ◁ N1 ◁ G , Nn = {e} , (1.16)
and all quotients Ni−1 /Ni are simple groups. The composition series is not necessarily
unique but all composition series for G have the same length n. Of course simple groups
themselves have length one.

1.1.3 Direct Product of Groups

For two groups G1 , G2 we may define a direct product group G1 × G2 formed by pairs of
elements {(g1 , g2 )}, belonging to (G1 , G2 ), which is defined by the rules

(g1 , g2 )(g ′1 , g ′2 ) = (g1 g ′1 , g2 g ′2 ) , (g1 , g2 )−1 = (g1−1 , g2−1 ) , e = (e1 , e2 ) . (1.17)

So long as it is clear which elements belong to G1 and which to G2 we may write the elements
of G1 × G2 as just g1 g2 = g2 g1 and e = e1 e2 . For finite groups ∣G1 × G2 ∣ = ∣G1 ∣ ∣G2 ∣. In any
direct product G1 × G2 then G1 ≃ {(g1 , e2 )}, G2 ≃ {(e1 , g2 )} are both normal subgroups.
For the direct product G × G then G is of course a subgroup but there is also the diagonal
subgroup G formed from elements {(g, g)} which is not a normal subgroup of G × G.

1.2 Cyclic, Dihedral and Permutation Groups

It is worth describing some particular finite discrete groups which appear frequently.

1.2.1 Cyclic Group

The group Zn is defined by integers 0, 1, . . . n−1 with the group operation addition modulo n
and the identity 0. The cyclic group Cn is also defined by the complex numbers e2πir/n , r =
0, . . . , n − 1, of modulus one, under multiplication. Clearly it is abelian and Cn ≃ Zn .
Abstractly Zn , Cn both can be defined by

Zn ≃ Cn = {ar ∶ r = 0, 1, . . . , n, a0 = an = e} . (1.18)

4
Obviously Z1 is the trivial one element group. For p prime Zp has no subgroups, since p
has no divisors, and hence Zp is simple. If n = pq then Zp and Zq are normal subgroups
and Zpq /Zp ≃ Zq . If p, q are coprime (no common factors) Zpq ≃ Zp × Zq and both Zp , Zq are
maximal normal subgroups.
For further illustration we consider Z2 × Z4 where Z2 = {e, b} with b2 = e and Z4 =
{e, a, a2 , a3 } with a4 = e. This has proper subgroups, which by Lagrange’s theorem can only
have order 2 or 4,
Z2 = {e, b}, {e, a2 }, {e, ba2 }, Z4 = {e, a, a2 , a3 }, {e, ba, a2 , ba3 }, K4 = {e, b, a2 , ba2 } . (1.19)
The group K4 = Z2 × Z2 is not cyclic, it is the Klein2 , group, it has 4 elements all but the
identity of order 2. The list in (1.19) is not just a sum of direct products of subgroups of
Z2 and Z4 , as demonstrated by the subgroups {e, ba2 } and {e, ba, a2 , ba3 }.
For Z12 there are three possible composition series Z1 ◁Z2 ◁Z6 ◁Z12 , Z1 ◁Z2 ◁Z4 ◁Z12 or
Z1 ◁Z3 ◁Z6 ◁Z12 , all of length 3. For Zn the sizes of the quotient groups in the composition
series correspond to the prime factors of n and the length of the composition series is the
number of prime factors, allowing for multiplicity.

1.2.2 Dihedral Group

The dihedral group Dn , of order 2n, is the symmetry group for a regular n-sided polygon
and is formed by rotations a through angles 2πr/n together with reflections b. In general
Dn = {ar , ar b ∶ r = 0, 1, . . . , n − 1, a0 = an = e, b2 = e, ab = ban−1 } . (1.20)
For any r (ar b)2 = e. For n > 2 the group is non abelian since ba ≠ ab, note that D2 ≃ Z2 ×Z2 .
In general {e, b} ≃ Z2 is a subgroup of Dn .
The centre of Dn depends on whether n is even or odd, Z(D2n+1 ) = {e} whereas
Z(D2n ) = {e, an } ≃ Z2 .
The normal subgroups of Dn also depend on n. If k divides n, k∣n, then the abelian group
Zk = {an/k r ∶ r = 0, 1, . . . , k−1} is a normal subgroup. This includes Zn . Also D2n has normal
subgroups Dn given by {a2r , a2r b ∶ r = 0, 1, . . . , n − 1} and {a2r , a2r+1 b ∶ r = 0, 1, . . . , n − 1}.
For H an abelian group then a generalised dihedral group can be defined by all elements
{(h, e), (h, b)} for h ∈ H and {e, b} forming the group Z2 so that b2 = e. Group multipli-
cation is defined so that (h1 , e)(h2 , e) = (h1 h2 , e), (h1 , e)(h2 , b) = (h1 h2 , b), (h1 , b)(h2 , e) =
(h1 h2 −1 , b), (h1 , b)(h2 , b) = (h1 h2 −1 , e). Note that (h, e)−1 = (h−1 , e), (h, b)−1 = (h, b).

1.2.3 Symmetric Group

A frequently occurring group is the permutation or symmetric group Sn on n objects. It is

easy to see that the order of Sn is n!. For n = 3 S3 ≃ D3 , as this is the symmetry group of
an equilateral triangle under permutations of the vertices.
2
Felix Klein, 1829-1925, German.

5
 The elements of the permutation group can be decomposed into cycles. Acting on
{1, 2, . . . , n} the 2-cycle (ij), for i ≠ j, takes i ↔ j, the 3-cycle (ijk), with i, j, k all different,
takes i → j → k → i and so on for the arbitrary p-cycle (i1 , . . . , ip ), such that p ≤ n and
ir ≠ is , 1 ≤ ir ≤ n, which generates cyclic permutations of {i1 , . . . , ip }. Trivially the 1-cycle
(i) leaves i invariant. Clearly (i1 , . . . , ip )p = e and for any one of the (np) choices for the p {ir }
there are (p − 1)! choices for the p-cycle involving {ir } since any p-cycle is invariant under
cyclic permutations. For i, j, k, l all distinct (ij)(kl) = (kl)(ij) and also (ij)(jk) = (ijk).
In general a p-cycle can be written as a product of p − 1 2-cycles

(i1 , . . . , ip ) = (i1 i2 )(i2 i3 ) . . . (ip−1 ip ) . (1.21)

To verify the decomposition of the action of some g ∈ Sn into cycles we may consider
for an arbitrary i ∈ {1, 2, . . . , n} all g r i, r = 1, 2, . . . . For some minimal p we must have
g p i = i. The action of g then generates a p-cycle (i1 , . . . , ip ) where i1 = i. Then for j ∈
{1, 2, . . . , n}, j ∉ {i1 , . . . , ip } acting repeatedly with g generates a new q-cycle (j1 , . . . , jq )
for some q and j1 = j. Continuing in this fashion any element of {1, 2, . . . , n} belongs to
some cycle, if gk = k then the decomposition involves the 1-cycle (k). Thus we may write
g = g(i1 ...ip )(j1 ...jq )... . The identity e corresponds to the n 1-cycles (1)(2) . . . (n). Clearly
g and g −1 = g(ip ...i1 )(jq ...j1 )... have the same cycle decomposition. If h corresponds to a
permutation σ where σ{1, 2, . . . , n} = {σ(1), σ(2), . . . , σ(n)} then h g(i1 ...ip )(j1 ...jq )... h−1 =
g(σ(i1 )...σ(ip ))(σ(j1 )...σ(jq ))... .
Elements in Sn which are given by products of non overlapping pi -cycles, i = 1, . . . , r,
with 1 ≤ pr ≤ ⋅ ⋅ ⋅ ≤ p2 ≤ p1 ≤ n and ∑ri=1 pi = n form a subset denoted by [p1 , p2 , . . . pr ].
The identity is obtained for r = n and pi = 1, i = 1, . . . , n. Subsets in which one or
more pi are different are distinct. To count the number of permutations belonging to
each cycle type we first assume the pi are all different so that pi > pi+1 . Then there are
n!/ ∏i pi ! ways of assigning {1, 2, . . . , n} to the different cycles but each cycle is invariant
under pi cyclic permutations so there are (pi − 1)! possibilities for each pi -cycle. This gives
∏ri=1 (pi − 1)! n!/pi ! = n!/ ∏ri=1 pi different permutations. Suppose more generally there are
ji pi -cycles so that ∑ri=1 ji pi = n. Then the previous argument gives a factor pi ji but there
must also be a factor ji ! in the denominator since permutations between the different pi -
cycles give the same permutation. The number of elements in Sn which belong to the
subset corresponding to cycles [p1 (j1 ), p2 (j2 ), . . . , pr (jr )], where pi (ji ) means pi is repeated
ji times, is then
r
n!
N[p1 (j1 ),...,pr (jr )] = r , pi > pi+1 , ji ≥ 1 , 1 ≤ r ≤ n , ∑ ji pi = n . (1.22)
∏i=1 pi ji ji ! i=1

Since the total number of permutations is n! we must have

n
n! = ∑ ∑ ∑ δn,∑ri=1 ji pi N[p1 (j1 ),...,pr (jr )] . (1.23)
r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1

Note that N[1(n)] = 1. Each choice of integer pi > pi+1 > . . . pr > 0, with pi repeated ji times,
r
λ = [p1 (j1 ), p2 (j2 ), . . . , pr (jr )] , ∑ ji pi = n , r = 1, 2, . . . , (1.24)
i=1

6
corresponds to a partition of n. The number of possible partitions increases very rapidly
with n.
All n! elements of Sn can be obtained from products of the n − 1 2-cycles σi = (i i+1)
where the σi may be defined abstractly by requiring

{σi , i = 1, . . . , n − 1 ∶ σi 2 = e, (σi σi+1 )3 = e, i = 1, . . . , n − 2, σi σj = σj σi , ∣i − j∣ > 1} . (1.25)

The condition (σi σi+1 )3 = e can be alternatively expressed as

σi σi+1 σi = σi+1 σi σi+1 . (1.26)

The conditions in (1.25) are sufficient to define the permutation group Sn in terms of all
possible distinct products of σi and can be used to determine the group multiplication. For
each possible product imposing the relations (1.25) ensure the number of σi in the product
is unchanged mod 2. Thus for

σ = σi1 σi2 . . . σip sgn(σ) = εσ = (−1)p , (1.27)

εσ , the sign or signature of σ, does not depend on the particular decomposition of σ.

Even/odd elements of Sn correspond to εσ = ±1. Crucially

εσ εσ′ = εσσ′ , εσ−1 = εσ . (1.28)

The product σ1 σ2 . . . σr−1 corresponds to the r-cycle (1 2 . . . r) and must in consequence

satisfy
(σ1 σ2 . . . σr−1 )r = e , r = 2, 3, . . . , n . (1.29)
This may be verified as a consequence of just the relations (1.25) by induction from

(σ1 . . . σr )s = (σ1 . . . σr−1 )s σr σr−1 . . . σr−s+1 , s = 1, . . . , r , (1.30)

which, subject to (1.29), implies (σ1 . . . σr )r = (σ1 . . . σr )−1 and hence (1.29) for r → r + 1.
In turn (1.30) follows inductively starting from s = 1 by multiplying (1.30) on the right
successively by σ1 . . . σr and using

σr σr−1 . . . σr−s+1 (σ1 . . . σr ) = (σ1 . . . σr−1 ) σr . . . σr−s , s = 1, . . . , r , (1.31)

or, commuting σi , σj for ∣i − j∣ > 1,

σr σr−1 . . . σr−s+1 σr−s σr−s+1 . . . σr = σr−s (σr σr−1 . . . σr−s+2 σr−s+1 σr−s+2 . . . σr )σr−s
= . . . = σr−s σr−s+1 . . . σr−1 σr σr−1 . . . σr−s , (1.32)

repeatedly using (1.26). This argument easily extends to showing

(σi σi+1 . . . σi+p−2 )p = e , (1.33)

so that σi σi+1 . . . σi+p−2 is a p-cycle.

7
 As an illustration of how more general permutations can be generated in terms of prod-
ucts of {σi } then
⎧
⎪σi σi+1 . . . σj σj−1 . . . σi , j > i,
⎪
⎪
⎪
σ̃j = ⎨σi , j = i, (1.34)
⎪
⎪
⎪
⎩σj σj+1 . . . σi−1 σi−2 . . . σj ,
⎪ j < i − 1,
satisfies σ̃j2 = e and corresponds to the 2-cycle (i j+1).
An alternative expression for arbitrary elements belonging to Sn can be obtained in
terms of products of cycles where

ar = σ1 . . . σr−1 , r = 2, . . . , n , a1 = e , arr = e , εar = (−1)r−1 , (1.35)

and then an arbitrary σ ∈ Sn can be written as

σ = anrn an−1rn−1 . . . a2 r2 , ri = 0, 1, . . . , i − 1 . (1.36)

It is easy to see that there are n! possibilities and arbitrary products can be brought to the
form (1.36) by repeatedly using

as at = at2 at−1t−2 as−1 , s = 2, 3, . . . , t − 1 , t = 3, . . . n , (1.37)

as well as arr = e. Of course this rule preserves the signature. For just n = 3 then taking
a2 = b, a3 reproduces the group multiplication rules defining D3 ≃ S3 in (1.20).

1.2.4 Alternating Group

The alternating group An is the normal subgroup of Sn formed by even permutations. It

has n!/2 elements and for n ≥ 5 is simple since, as discussed further later, there are then no
normal subgroups apart from the identity. In general Sn /An ≃ Z2 .
As a consequence of (1.21) the alternating group An can only contain single p-cycles
with p odd or products of distinct p-cycles, p1 , p2 , . . . , with ∑i (pi − 1) even. Every 3-cycle
in Sn is contained in An since for i, j, k distinct

(ijk) = (ik)(ij) . (1.38)

Furthermore any product of 2-cycles is expressible as a product of 3-cycles. If the 2-cycles

are not distinct this has just been shown, otherwise with i, j, k, l all different.

(ij)(kl) = (ij)(ik)(ki)(kl) = (jik)(ikl) . (1.39)

As a consequence An may be generated in terms of 3-cycles just as Sn is generated by

2-cycles.
To verify this for An it is sufficient to consider the 3-cycles si = (12 i+2), i = 1, 2, . . . , n−2,
where for i ≠ j, si sj = (1i)(2j). The group elements si , = 1, . . . , n − 2 obey the abstract
relations
si3 = e , i ≠ j, (si sj )2 = e ⇒ si sj = sj−1 si−1 = sj2 si2 . (1.40)

8
Starting from A3 = {e, s1 , s12 } the groups An may be defined inductively by

An+1 = {An , An sn−1 , An sn−12 , An sn−1 si2 ∶ i = 1, . . . , n − 2} . (1.41)

It is necessary to check An+1 sn−1 = An+1 , which follows from (1.40) since sn−1 si2 sn−1 =
si2 sn−1 si2 and An si = An , and also for j = 1, . . . , n−2, An+1 sj = An+1 , which is a consequence
of sn−1 sj = sj2 sn−12 , sn−12 sj = sj sn−1 sj2 and for j ≠ i, sn−1 si2 sj = si2 sj sn−1 si2 .
All 3-cycles are conjugate in An so long as n ≥ 5. It must be true that there is a σ ∈ Sn
such that
σ(ijk)σ −1 = (i′ j ′ k ′ ) , (1.42)
for arbitrary distinct i, j, k and i′ , j ′ , k ′ . If σ is even then σ ∈ An , otherwise if n ≥ 5 we
can take σ → σ ′ = σ(lm) where l, m are different from i, j, k to achieve the same result and
σ ′ ∈ An . In a similar fashion all n-cycles, for n odd, are conjugate in An+2 . However 5-cycles
in A5 are not necessarily conjugate. There are 6 Z5 subgroups {e, σi , σi2 , σi3 , σi4 } generated
by a 5-cycle σi . Of these two sets of 3 are each conjugate.
As special cases A3 ≃ Z3 and A4 is the symmetry group, without reflections, of a regular
tetrahedron. The group is formed by 2π/3, 4π/3 rotations about axes from each of 4 vertices
to the centre of the opposite face and also rotations of π about the three lines joining the
mid points of opposite edges. Thus, apart from the identity, A4 is composed of 8 different
3-cycles and 3 products of two distinct 2-cycles which, with the identity, form a normal
subgroup.
For arbitrary n any normal subgroup N ◁ An which contains a 3-cycle must, as a
consequence of (1.42), also contain all 3-cycles and so N = An . For n ≥ 5 it is possible to
show that N must contain a 3-cycle and therefore N = An and An is simple so long as n ≥ 5.
To show this let σ ∈ N be a non trivial group element containing p 1-cycles. Then it is
possible to choose τ ∈ An , τ −1 στ ∈ N since σ ∈ N , such that σ ′ = σ −1 τ −1 στ ∈ N contains p′
1-cycles with p′ > p. This process may be continued until p′ = n − 3 and then σ ′ is just a
3-cycle. As an illustration if σ = (12)(34) then taking τ = (12)(35) we have σ ′ = (345) or if
σ = (12345) and τ = (123) then σ ′ = (125).

1.3 Orbit Stabiliser Theorem

An important result which has many applications arises when a group G acts on a space
X = {x} so that for any g ∈ G there is an action x → gx. For any particular x ∈ X the
stabiliser group, or little group, Gx is defined by those elements of G which leave x invariant,
Gx = {h ∶ h ∈ G, hx = x}. It is easy to see that Gx is a subgroup of G. The orbit of x is the
set of points in X obtained by the action of G, Ox = {x′ ∶ x′ = gx}. Ox can be identified
with the coset G/Gx . This is the orbit stabiliser theorem and we have for a finite group G
the dimension of the orbit
dim Ox = ∣G∣/∣Gx ∣ , (1.43)
which is an integer by Lagrange’s theorem. Clearly for x′ ∈ Ox then Gx′ ≃ Gx since hx =
x, x′ = gx implies h′ x′ = x′ for h′ = ghg −1 . In general the space X may be decomposed into
orbits under the action of G.

9
1.4 Further Definitions

Here we give some supplementary definitions connected with groups which play a crucial
role in the theory of groups and introduces important notation.

1.4.1 Automorphisms and Semi-Direct Product

An automorphism of a group G = {gi } is defined as a mapping between elements, gi → ϕ(gi ),

such that the product rule is preserved, i.e.

ϕ(gi )ϕ(gj ) = ϕ(gi gj ) for all gi , gj ∈ G , (1.44)

so that Gϕ = {ϕ(gi )} ≃ G. Clearly we must have ϕ(e) = e and ϕ(g −1 ) = ϕ(g)−1 . In general
for any fixed g ∈ G we may define an inner automorphism by ϕg (gi ) = ggi g −1 , otherwise
the automorphism is outer. It is straightforward to see that the set of all automorphisms
of G itself forms a group Aut G which must include the group of inner autmorphisms
Inn G = G/Z(G) as a normal subgroup, the quotient Out G = Aut G/Inn G defines the outer
automorphism group. For an abelian group there are no non trivial inner automorphisms
but there can be non trivial outer automorphisms, e.g. for Z3 take {e, a, a2 } → {e, a2 , a}.
In this case Aut Z3 = Z2 and Z3 /Z(Z3 ) = {e} the trivial one element group. In a similar
fashion Aut Zn = Zn−1 whenever n is prime.
There is also an antiautomorphism group Anti G which is defined by all maps gi → ϕ(gi )
such that
ϕ(gi )ϕ(gj ) = ϕ(gj gi ) for all gi , gj ∈ G . (1.45)
Clearly Anti G = Aut G if G is abelian.
If H ⊂ Aut G, so that for any h ∈ H and any g ∈ G we have g → ϕh (g) with
h

ϕh (g1 )ϕh (g2 ) = ϕh (g1 g2 ) , (1.46)

and

ϕh1 (ϕh2 (g)) = ϕh1 h2 (g) , ϕh (e) = e , ϕe (g) = g , ϕh−1 (g) = ϕh−1 (g) , (1.47)

we may define a new group called the semi-direct product of H with G, denoted H ⋉ G,
or G ⋊ H. As with the direct product this is defined in terms of pairs of elements (h, g)
belonging to (H, G) but with the rather less trivial product rule

(h, g)(h′ , g ′ ) = (hh′ , g ϕh (g ′ )) , (h, g)−1 = (h−1 , ϕh−1 (g −1 )) . (1.48)

From this it follows that

(h, g)(e, g ′ )(h, g)−1 = (e, gϕh (g ′ )g −1 ) for any g, g ′ ∈ G , h ∈ H . (1.49)

Consequently the subgroup {(e, g)} ≃ G is a normal subgroup of H ⋉ G and hence H ≃

(H ⋉ G)/G.

10
 It is often convenient to write the elements of H ⋉ G as simple products so that we may
abbreviate (h, g) → hg = ϕh (g)h.
As a simple illustration we have Dn ≃ Z2 ⋉ Zn where Z2 = {e, b ∶ b2 = e} and also
Zn = {ar ∶ r = 0, . . . n − 1, an = e}. The semi-direct product is then defined by taking, for any
g = ar ∈ Zn , ϕb (g) = g −1 = b g b−1 . Zn is a normal subgroup of of Dn .
The case of a maximal semi-direct product of G with Aut G

G ⋊ Aut G = Hol G , (1.50)

has a special name, the holomorph of G. From the above example D3 ≃ Hol Z3 .

1.4.2 Wreath Products and Central Products

Another semi-direct product is obtained if we take G → Gn× = G × G × ⋅ ⋅ ⋅ × G, the n-fold

direct product, with H = Sn permuting the elements of each of the factors so that, for any
g = (g1 , g2 , . . . , gn ) ∈ Gn× ,

gσ ≡ ϕσ (g1 , g2 , . . . , gn ) = (gσ(1) , gσ(2) , . . . , gσ(n) ) , σ ∈ Sn . (1.51)

This then defines Sn ⋉ Gn× . This is an example of a wreath product, denoted G ≀ Sn , and has
order n!∣ G∣n . A particular example is Bn = Z2 ≀ Sn , the hyperoctahedral group, which is the
symmetry group of the n-dimensional hypercube. As special cases B2 = D4 , the symmetry
group of the square, and B3 = S4 × Z2 , the symmetry group of the cube.
If the permutations acting on Gn× are restricted to a subgroup of Sn there is a corre-
sponding wreath product, for just cyclic permutations of (g1 , . . . , gn ) this is G ≀ Zn .
Another prescription for combining groups is obtained essentially by taking the direct
product and dividing by some common elements. If G1 , G2 are two groups then for H1 ⊂
Z(G1 ), H2 ⊂ Z(G2 ) then for H1 ≃ H2 with the isomorphism θ such that h1 ↔ h2 then
{(h1 , h2 )} = H, where H ≃ H1 ≃ H2 is a normal subgroup of H1 × H2 . The quotient
(G1 × G2 )/H defines the central product.
A rather trivial example arises if G1 = Z4 , G2 = Z6 . In this case we can take H = Z2 and
(Z4 × Z6 )/Z2 ≃ Z12 .

1.4.3 Conjugacy Classes

If gj = ggi g −1 for some g ∈ G then gj is conjugate to gi , gj ∼ gi . The equivalence relation ∼

divides G into conjugacy classes

Cs = {gi ∶ gi ∼ gi′ = ggi g −1 , g ∈ G} , s = 1, . . . , Nchar . (1.52)

Different conjugacy classes are distinct Cs ∩ Cs′ = ∅, s ≠ s′ , and G = ∪s Cs , ∣G∣ = ∑N char

s=1 ds for
−1
ds = dim Cs . For any g ∈ G, g Cs g = Cs . Clearly the identity is in a conjugacy class C1 by
itself. In general
gi , gj ∈ Cs ⇒ gi−1 , gj−1 ∈ Cs̄ , ∣Cs̄ ∣ = ∣Cs ∣ , (1.53)

11
where Cs̄ = Cs if g, g −1 ∈ Cs . It is sometimes convenient in this case to write Cs̄ = Cs−1 .
For any h ∈ G the associated conjugacy class containing h is given by

C(h) = {ghg −1 ∶ g ∈ G} . (1.54)

G also contains a subgroup defined by

CG (h) = {g ∶ ghg −1 = h, g ∈ G} , (1.55)

and then, by the orbit stabiliser theorem,

dim C(h) = ∣G∣/∣CG (h)∣ . (1.56)

By Lagrange’s theorem ∣CG (h)∣ divides ∣G∣ so that the dimensions of any conjugacy class
must also divide ∣G∣.
Under group multiplication the product of two conjugacy classes must be expressible in
terms of a union of conjugacy classes so that there is a multiplication rule

Cs Ct = ⋃ cst u Cu , (1.57)
u

where cst u takes the values 0, 1, 2, . . . . Since Cs g = g Cs for any g ∈ G then Cs Ct = Ct Cs and
thus cst u = cts u . Furthermore c1t u = δt u , css̄ 1 = ds and ∑u cst u du = ds dt .
For an abelian group all elements have their own conjugacy class necessarily of dimension
1. The elements in a conjugacy class have similar properties such as gi n = e for the same n
for all gi ∈ Cs . Any normal subgroup is composed of conjugacy classes which must include
C1 . For S3 which has elements {e, a, a2 , b, ab, a2 b}, where b = (12), a = (123), there are
three conjugacy classes {e}, {a, a2 }, {b, ab, a2 b}. {e, a, a2 } forms a normal subgroup which
is isomorphic to Z3 .
For the dihedral group Dn , as defined in (1.20), the conjugacy classes are different
according to whether n is even or odd. Labelling them by their size these are

C1 = {e} , C2,r = {ar , an−r } , r = 1, . . . 12 (n − 1) , Cn = {ar b ∶ r = 0, 1, . . . , n − 1} , n odd ,

1
C1,1 = {e} , C1,2 = {a 2 n } , C2,r = {ar , an−r } , r = 1, . . . , 21 (n − 2) ,
C 1 n,1 = {a2r b ∶ r = 0, 1, . . . , 12 n − 1} , C 1 n,2 = {a2r+1 b ∶ r = 0, 1, . . . , 21 n − 1} , n even . (1.58)
2 2

There are then 12 (n + 3) conjugacy classes for n odd, 21 (n + 6) for n even. The conjugacy
classes are all self inverse in that each conjugacy class C contains the inverse for each group
element in C.
Under multiplication for n odd
⎧
⎪
⎪C ∪ C2,r+s or C2,∣r−s∣ ∪ C2,n−r−s r ≠ s
C2,r C2,s = ⎨ 2,∣r−s∣ ,
⎩2 C1 ∪ C2,2r or 2 C1 ∪ C2,n−2r r=s
⎪
⎪
C2,r Cn = Cn C2,r = 2 Cn , Cn Cn = n C1 ∪r C2,r . (1.59)

12
Results for n even are similar.
For n even Dn has an external automorphism which interchanges the two conjugacy
classes involving b. The automorphism group is generated by c ∶ a → an−1 , b → ab. Clearly
c2 is the identity and Out Dn = {e, c ∶ c2 = e} = Z2 . Correspondingly for n even Z2 ⋉Dn ≃ D2n
since for Z2 = {e, c} and Dn as in (1.20) then D2n = {ar , ar c , ar b , ar b c ∶ r = 0, . . . , n − 1}
for c a c = an−1 , c b c = ab. Defining a′ = abc, b′ = ac then a′ 2n = b′ 2 = e and a′ , b′ satisfy the
conditions to generate D2n . Conversely for n odd Z2 × Dn ≃ D2n where the extra element c
commutes with a, b. This follows from Z2 × Zn ≃ Z2n for n odd.
The decomposition of any σ ∈ Sn into non overlapping, or disjoint, cycles is unchanged
under conjugation. For σ ∈ Sn given by a product of non overlapping cycles all other σ ′ ∈ Sn
expressible as a product of the same cycles can be obtained from σ by some permutation
in Sn and in consequence σ ′ can be obtained from σ by conjugation. If σ ′ has a different
expression in terms of cycles then it cannot. Hence for an identical decomposition into
cycles, up to ordering, σ, σ ′ belong to the same conjugacy class. The different conjugacy
classes of Sn are then labelled C[p1 (j1 ),...,pr (jr )] with p1 , p2 , . . . , pr together with j1 , . . . , jr
corresponding to a partition of n such that 1 ≤ pr < ⋅ ⋅ ⋅ < p2 < p1 ≤ n with ∑i ji pi = n. The
dimensions of the conjugacy classes for Sn are given by the general formula (1.22).
As a consequence of (1.28) εσσ′ σ−1 = εσ′ so that the sign of all group elements in Sn
belonging to a particular conjugacy class is the same and is given by

σ ∈ C[p1 (j1 ),...,pr (jr )] , εσ = (−1)∑i ji (pi −1) . (1.60)

Since there equal numbers of σ with εσ = ±1 we must have

n
∑ ∑ ∑ δn,∑ri=1 ji pi N[p1 (j1 ),...,pr (jr )] (−1)∑i ji (pi −1) = 0 . (1.61)
r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1

The elements of the alternating group An correspond to those σ ∈ Sn with a cycle

decomposition in which the number of even n-cycles is also even and εσ = 1. However the
conjugacy classes in An are not just determined by the cycle decomposition as for Sn since
only even permutations are generated by conjugation in An . Conjugacy classes in Sn may
split into two on reduction to An .
The conjugacy classes for Sn can be identified with Young3 diagrams which are formed
by n boxes in rows of decreasing size according to the partition of n. Those for S5 are

[5] , [4, 1] , [3, 2] , [2(2), 1] ,

[3, 1(2)] , [2, 1(3)] , [1(5)] . (1.62)

Such diagrams play an important role in discussions of the permutation group.

3
Alfred Young, 1873-1940, British, 10th wrangler

13
 As a illustration we consider the cases n = 3, 4, 5. For S3 there are just three conjugacy
classes with dimensions

N[3] = 2 , N[2,1] = (32) = 3 , N[1(3)] = 1 . (1.63)

The different conjugacy classes are C[1(3)] = {e}, C[3] = {a, a2 }, C[2,1] = {b, ba, ba2 } where
a3 = b2 = e, ba = a2 b. C[2,1] contains all odd permutations so A3 = {e, a, a2 }. However this is
abelian so each element has its own conjugacy class. Similarly for S4

N[4] = 3! = 6 , N[3,1] = 2(43) = 8 , N[2(2)] = 12 (42) = 3 , N[2,1(2)] = (42) = 6 ,

N[1(4)] = 1 .
(1.64)
The elements of the alternating group A4 belong to the conjugacy classes C[1(4)] , C[3,1]
and C[2(2)] in S4 . The elements of C[2(2)] , a = (12)(34), b = (13)(24), ab = ba = (14)(23),
together e form a normal subgroup of both S4 and A4 , a2 = b2 = e, which is isomorphic to
D2 . However the cycles (123)(4) and (124)(3) belonging to A4 are not conjugate in A4
since the permutation linking them involves one 2-cycle. Hence C[3,1] decomposes into two
equal conjugacy classes of size 4.
For S5 corresponding to (1.62)

N[5] = 4! = 24 , N[4,1] = 3(54) = 30 , N[3,2] = 2(52) = 20 , N[3,1(2)] = 2(53) = 20 ,

N[2(2),1] = 12 (52)(32) = 15 , N[2,1(3)] = (52) = 10 , N[1(5)] = 1 . (1.65)

A5 is given by elements belonging to the conjugacy classes C[3,1(2)] , C[5] , C[2(2),1] as well as
C[1(5)] in S5 . However C[5] splits into two conjugacy classes for A5 each of size 12 and the
dimensions of the five conjugacy classes are then 1 + 20 + 12 + 12 + 15 = 60. In this case none
of the dimensions including 1 for C[1(5)] add up to a divisor of 60 = ∣A5 ∣. Hence A5 has no
non trivial normal subgroups and is simple. It is the smallest non abelian group with this
property.
An alternative labelling of conjugacy classes for Sn instead of [p1 (j1 ), . . . , pr (jr )] for
r = 1, . . . , n is obtained by taking
⎧
⎪ ji ,
⎪ pi = m for some i ,
km = ⎨ m = 1, . . . , n , ∑nm=1 mkm = n , (1.66)
⎩ 0, pi ≠ m for any i ,
⎪
⎪

where N[p1 (j1 ),...,pr (jr )] = n!/ ∏nm=1 mkm m! and εσ = (−1)(m−1)km . For a function on con-
jugacy classes of Sn , f (k1 , . . . , kn ), the corresponding sum over all classes is simply given
by
n!
∑ δn,∑m mkm n f (k1 , . . . kn ) . (1.67)
k1 ,k2 ,...,kn ≥0 ∏m=1 mkm m!

As an example for q = (q1 , q2 , . . . ) there is a generating function obtained by summing

over n
∞ ∞
n
qm k m um qm
F (u, q) = ∑ un ∏ ( ∑ km
) δ n,∑ m mk m = exp ( ∑ ). (1.68)
n=0 m=1 km ≥0 m m! m=1 m

14
If qm = f (xm ) for some function f then
∞
um f (xm )
F (u, q) = PE(u, x; f ) = exp ( ∑ ) = 1 + u f (x) + ∑ um ⋁mf (x) , (1.69)
m=1 m m≥2

is the plethystic exponential formed from the function f and

⋁2f (x) = 12 (f (x)2 + f (x2 )) , ⋁3f (x) = 16 (f (x)3 + 3 f (x2 )f (x) + 2f (x3 )) . (1.70)

For a constant function f (x) = c

∞ ∞
um c 1 1 r
exp ( ∑ )= c
= ∑ r! c(c + 1) . . . (c + r − 1) u . (1.71)
m=1 m (1 − u) r=0

Including the sign factor for odd permutations

∞
um f (xm )
PEA (u, x; f ) = exp ( ∑ (−1)m−1 ) = 1 + u f (x) + ∑ um ⋀mf (x) , (1.72)
m=1 m m≥2

with

⋀2f (x) = 12 (f (x)2 − f (x2 )) , ⋀3f (x) = 16 (f (x)3 − 3 f (x2 )f (x) + 2f (x3 )) . (1.73)

1.4.4 Centre, Normaliser, Centraliser and Commutator Subgroups

The centre of a group G, Z(G), is the set of elements which commute with all elements
of G. This is clearly an abelian normal subgroup. For an abelian group Z(G) ≃ G. The
centre is a normal subgroup and the quotient group G/Z(G) is referred to as the inner
automorphism group.
For a subset of a group H ⊂ G, not necessarily a subgroup, then the elements g ∈ G such
that ghg −1 ∈ H for all h ∈ H, or gHg −1 = H, form a subgroup of G called the normaliser of
H in G, written NG (H). If H is a subgroup then clearly H ◁ NG (H) and if H is a normal
subgroup, NG (H) = G.
The subgroup of G formed by elements {g} such that ghg −1 = h for all h ∈ H forms the
centraliser CG (H). Necessarily CG (H) ⊂ NG (H).
For any two elements of G, g, h, then

[g, h] = g −1 h−1 g h , (1.74)

is the commutator of g, h. If G is abelian then [g, h] = e for all g, h. More generally if [g, h] =
e then g and h commute. In general [g, h]−1 = [h, g] and for any g ′ ∈ G then g ′ [g, h]g ′−1 =
[g ′ gg ′−1 , ghg ′−1 ]. The product of two commutators need not be a commutator (this can
only arise if the order of G is at least 96) but there is a subgroup of G, the commutator
or derived subgroup G′ = [G, G] formed from arbitrary products of commutators. From the
above g[G, G]g −1 = [G, G] for any g ∈ G so [G, G] is a normal subgroup. For any g1 , g2 ∈ G

15
then g1 g2 = g2 g1 [g1 , g2 ] so that the quotient group G/[G, G] is abelian. For any normal
subgroup N ◁ G then the quotient G/N is abelian only if [G, G] < N . Of course if G is
abelian [G, G] = {e}, the trivial group.
A group is perfect if G = [G, G]. A non abelian simple group must be perfect since
[G, G] is a normal subgroup and [G, G] ≠ {e} if G is non abelian. The converse is not
necessary.
As an example we may consider the dihedral group Dn as presented in (1.20). In this
case [Dn , Dn ] = {e, a, a2 , . . . , an−1 } = Zn . The quotient is just Z2 .
The notion of the commutator subgroup may be extended to define the derived series
where G(n) = [G(n−1) , G(n−1) ] where G(0) = G and n = 1, 2, . . . . Evidently ⋅ ⋅ ⋅ ◁ G(n) ◁
G(n−1) ◁ . . . G(1) ◁ G. For a finite group this series must terminate in a perfect group or
G(n) = {e} for some n. In this case G is solvable.

1.4.5 Double Coset

An extension of the notion of a coset is a double coset. If H, K are subgroups H, K ⊂ G

then the equivalence relation (1.8) can be extended to
gi ∼ g ′i ⇔ gi = kg ′i h for h ∈ H , k ∈ K , (1.75)
where under this equivalence relation {gi } defines a double coset and these double cosets
form K/G/H. In general there is a one to one correspondence of K/G/H with H/G/K
since gi−1 = h−1 g ′i−1 k −1 . The potential extension of Lagrange’s theorem is not valid since
in general the cosets belonging to K/G/H have dimensions which need not divide ∣G∣. If
G = D3 and H = {e, b}, K = {e, ba}, which are both Z2 subgroups, then K/G/H is comprised
of {e, b, ba, a2 } and {a, ba2 } whereas H/G/K is formed from {e, b, ba, a} and {a2 , ba2 }.

1.4.6 Goursat’s Lemma

A nice application of the basic definitions of group theory due to Goursat4 shows how
subgroups of direct products are obtained. For G = G1 × G2 and a subgroup H < G then we
may define two subgroups of G1 by
H1 = {h1 } , (h1 , h2 ) ∈ H for any h2 , N1 = {n1 } , (n1 , e2 ) ∈ H . (1.76)
Since (h1 , h2 )(n1 , e2 )(h1 , h2 )−1 = (h1 n1 h1 −1 , e2 ) h1 n1 h1 −1 ∈ N1 for any h1 ∈ H1 so that N1
is a normal subgroup of H1 , N1 ◁ H1 . There is then a quotient group Q1 = H1 /N1 . In a
similar fashion we can define H2 , N2 and Q2 . The aim of the lemma is to show Q1 ≃ Q2 .
To show this assume (h1 , h2 ), (h1 , h2 ′ ) ∈ H. Then (h1 , h2 )−1 (h1 , h2 ′ ) = (e1 , h2 −1 h2 ′ ) so
that h2 −1 h2 ′ ∈ N2 or h2 , h2 ′ both correspond to the same q2 ∈ Q2 . Extending this to h1 , h1′
ensures
(h1 , h2 ), (h1 ′ , h2 ′ ) ∈ H ⇔ (h1 ′ , h2 ′ ) = (h1 n1 , h2 n2 ) for some n1 ∈ N1 , n2 ∈ N2 . (1.77)
4
Édouard Goursat, 1858-1936, French.

16
This implies that any (h1 , h2 ) ∈ H can be expressed as (q1 n1 , q2 n2 ) where q2 determines
q1 and vice versa. In consequence there is a unique mapping ϕ ∶ q1 → q2 with inverse
ϕ−1 ∶ q2 → q1 and this preserves the group properties so that Q1 ≃ Q2 .
The subgroups H < G1 × G2 are therefore determined by subgroups H1 , H2 of G1 , G2
which have normal subgroups N1 , N2 such that H1 /N1 ≃ H2 /N2 ≃ Q. If H1 = N1 and
H2 = N2 , so the quotient groups are trivial, H = H1 × H2 . From this result the order of H,
∣H∣ = ∣N1 ∣ ∣N2 ∣ ∣Q∣ = ∣H1 ∣ ∣H2 ∣/∣Q∣.
As an illustration if we consider the subgroups of Z2 × Z4 in (1.19) which are not simple
direct products we have

H = {e, ba2 } , H1 = {e, b} , N1 = {e} , H2 = {e, a2 } , N2 = {e} ,

H = {e, ba, a2 , ba3 } , H1 = {e, b} , N1 = {e} , H2 = {e, a, a2 , a3 } , N2 = {e, a2 } , (1.78)

In both examples H1 /N1 ≃ H2 /N2 ≃ Z2 . For G × G the diagonal subgroup corresponds to

taking H1 = H2 = G and N1 = N2 = {e} with the map ϕ obtained by taking g1 = g2 .

1.5 Quaternion Groups

Quaternions are defined in terms {i, j, k} which have the properties

i2 = j 2 = k 2 = −1 , ij = −ji = k , jk = −kj = i , ki = −ik = j , (1.79)

were famously discovered, or invented, by Hamilton5 on 16th October 1843 and are an
extension of the usual complex numbers C. They can be defined by requiring i, j to satisfy
just the relations i2 = j 2 = (ij)2 = −1. Then

H = {q ∶ q = x0 1 + x1 i + x2 j + x3 k, xi ∈ R} , (1.80)

where q has a conjugate q̄ = x0 1 − x1 i − x2 j − x3 k and q q̄ = ∣q∣2 1, ∣q∣2 = ∑r xr2 . If q̄ = −q,

so that x0 = 0, then q is imaginary. A unit quaternion has ∣q∣ = 1. A crucial property of
quaternions is that they satisfy ∣q q ′ ∣ = ∣q∣ ∣q ′ ∣ and so form a division algebra.
Any quaternion can be expressed as a product of two imaginary quaternions. To show
this suppose q = q0 1 + Im q and choose an imaginary quaternion r such that

Im q r + r Im q = 0 , r̄ = −r , r2 = −1 ⇒ r q r = −q̄ . (1.81)

For r = r1 i+r2 j+r3 k this just requires (r1 , r2 , r3 ) is a unit 3-vector orthogonal to (x1 , x2 , x3 ).
Then another imaginary quaternion s is given by

s = −r q ⇒ s̄ = −s , s2 = −∣q∣2 1 , q = rs. (1.82)

Furthermore any unit quaternion q can be written as a commutator as in (1.74). For

q = q0 1 + q1 u, ū = −u, u2 = −1, q0 2 + q1 2 = 1 a unit quaternion square root can be defined
5
William Rowan Hamilton, 1805-65, Irish.

17
 √ √
by q = a 1 + b u for a2 = 12 (q0 + ∣q∣), 2 a b = q1 and then taking q = r s for imaginary unit
quaternions r, s implies
q = r s r s = r−1 s−1 r s . (1.83)

Quaternion multiplication is associative and any non zero quaternion q has a unique
inverse q −1 = q̄/∣q∣2 so that H∗ = {q ∶ q ∈ H, q ≠ 0} forms a group. The subgroups of H∗
define several important groups. Since the product of two unit quaternions is also a unit
quaternion there is an infinite continuous non abelian group

Q = {q ∶ q ∈ H, ∣q∣ = 1} , q −1 = q̄ . (1.84)

The centre Z(Q) ≃ {1, −1} and the quotient Q/{1, −1} = {q ∶ ∣q∣ = 1, q ∼ −q}. From (1.83)
Q ≃ [H∗ , H∗ ] and the associated quotient group H∗ /[H∗ , H∗ ] = {∣q∣ ∶ q ∈ H∗ } is just the
group formed by positive real numbers under multiplication.
Q contains abelian subgroups {eθ q̂ = cos θ 1 + sin θ q̂ ∶ q̂ = α i + β j + γ k, ∣q̂∣ = 1, 0 ≤ θ < 2π}.
There are also finite subgroups of Q. It is easy to see that

Cn = {e2πr/n i ∶ r = 0, 1, . . . , n − 1} , n = 1, 2, . . . ,
πr/n i πr/n i
Q4n = {e ,e j ∶ r = 0, 1, . . . , 2n − 1} , n = 1, 2, . . . , (1.85)

form groups of order n, 4n respectively. Clearly Cn ≃ Zn and form the cyclic groups. The
groups Q4n are referred to as dicyclic or binary dihedral or generalised quaternion groups.
They may also be denoted, for later convenience, by 2Dn . For n = 1 the group is not dicyclic
but we may identify 2D1 ≃ Z4 . For n = 2 this group becomes the non abelian quaternion
group
Q8 = {±1, ±i, ±j, ±k} . (1.86)
Q4n contains C2n as a normal subgroup. The conjugacy classes for Q4n are very similar to
those for D2n in (1.58)

C1,1 = {1} , C1,2 = {−1} , C2,r = {eπr/n i , eπ(n−r)r/n i } , r = 1, . . . , n − 1 ,

Cn,1 = {eπ2r/n i j ∶ r = 0, 1, . . . , n − 1} , Cn,2 = {eπ(2r+1)/n i j ∶ r = 0, 1, . . . , n − 1} . (1.87)

Q8 can be decomposed into five conjugacy classes which each have one or two elements,
C1,1 = {1}, C1,2 = {−1}, C2,1 = {±i}, C2,2 = {±j}, C2,3 = {±k}. Q4n contains a normal subgroup
{1, −1} and the quotient Q4n /{1, −1} ≃ Dn which is not a subgroup of Q4n .
The resemblance of Q4n to the dihedral group may be shown by defining it by the
conditions
a2n = e , an = b2 , b a b−1 = a−1 , (1.88)
where in terms of quaternions a = eπ/n i , b = j and of course e = 1 while an = b2 = −1.
There are further interesting finite groups which can be generated using two imaginary
unit quaternions u, v, u2 = v 2 = −1. If we consider arbitrary products of u, v then we must
have for a finite group
(uv)n = 1 for some minimal n . (1.89)

18
The group {±1, ±v, ±(uv)r , ±(uv)r v ∶ r = 1, . . . , n − 1} subject to (1.89), so that (vu)r =
(uv)n−r , is isomorphic to Q4n in (1.85). More general groups are obtained by considering
products of r and s where initially we take
√
r = u , s = eπ/3 v = 21 (1 + 3 v) , r2 = s3 = −1 , (1.90)

and then requiring for finiteness that, for some n = 1, 2, . . . ,

t = rs = ± cos π/n 1 + sin π/n z , z = −z̄, z 2 = −1 ⇒ tn = −(±1)n , (1.91)

where cos π/n ≥ 0. Assuming uv = cos φ 1 + sin φ y, with y an imaginary unit quaternion,
1
√ 1
√ 1
√
2 3 cos φ = ± cos π/n ⇒ − 2 3 ≤ cos π/n ≤ 2 3. (1.92)

This is only possible if n = 2, 3, 4, 5, 6 but for n = 2, cos φ = 0 and the corresponding group
is isomorphic to Q12 and for n = 6 then we may take cos φ = ±1 and hence v = ∓u and the
group isomorphic to C12 . The interesting cases are then6
√
n = 3 , cos 13 π = 12 , n = 4 , cos 41 π = √1 , n = 5 , cos 15 π = 14 ( 5 + 1) , (1.93)
2

and we may take

s = 12 (1 + i + j + k) , n = 3 , r = i , t3 = 1 , n = 4 , r = √1 (i + j) , t4 = −1 ,
2
√ √
n = 5 , r = 2 (i + σ j + τ k) , σ = 2 ( 5 − 1) , τ = 2 ( 5 + 1) , t5 = 1 .
1 1 1
(1.94)

1.5.1 Tetrahedral, Octahedral and Icosahedral Groups

The group corresponding to n = 3 can be obtained by noting that Q8 has an external

automorphism generated by (i, j, k) → (j, k, i) = 14 (1 + i + j + k)(i, j, k)(1 − i − j − k) so that

T = Z3 ⋉ Q8 = Q8 ∪ { 12 (±1 ± i ± j ± k)} , (1.95)

where there are 16 possible choices of ± (each ± is independent) so that T has order 24 and
is referred to as the binary tetrahedral group, also denoted as 2T . For the group elements
not in Q8
( 12 (−1 + q))3 = 1 , ( 21 (1 + q))6 = 1 , q ∈ { ± i ± j ± k} . (1.96)
There are seven conjugacy classes comprising T given by

C1 = {1} , C2 = {−1} , C4 = {±i, ±j, ±k} ,

C3 = 1
{ 2 (−1 + q)} , C ′3 = { 2 (−1 − q)} , C6 = −C ′3 ,
1
C ′6 = −C3 , (1.97)
q ∈ {i + j + k, −i − j + k, −i + j − k, i − j − k} ,

where here we label the conjugacy classes by the order of the group elements comprising
them. These have dimensions 1+1+6+4+4+4+4 = 24. Without any specific choice of r, s other
6
The real part of (eiπ/5 )5 + 1 = 0 leads to a quintic equation for x = cos π/5 which
√ can be factorised as
(x + 1)(4x2 − 2x − 1)2 = 0. The other roots determine cos π = −1 and cos 3π/5 = 14 (− 5 + 1).

19
than satisfying r2 = s3 = −1, (rs)3 = 1 the elements of T are determined by the conjugacy
classes C ′3 = {−s, rs, rsr, srs}, C3 = {s2 , sr, rs2 , s2 r}, C4 = {±r, ±srs2 , ±s2 rs} which along
with C1 , C2 and −C ′3 , −C3 give the whole group. In this case T = T/{1, −1} ≃ A4 where for
q ∈ T, T is obtained by assuming q ∼ −q. Hence T has four conjugacy classes inherited from
those for T which correspond to C1 = {1}, C2 = {i, j, k}, C3 = { 21 (−1 + q)}, C ′3 = { 12 (−1 − q)},
with four possible q as in (1.97), and where in this case Cn is composed of quaternions with
q n = ±1. The dimensions of each class are then 1 + 3 + 4 + 4 = 12.
The binary tetrahedral group T has a Z2 automorphism which is generated by taking
(i, j, k) → (−i, k, j) = − 21 (j + k)(i, j, k)(j + k) linking the conjugacy classes C3 , C ′3 and also
C6 , C ′6 in (1.97). Hence for n = 4 the binary octahedral group, O or 2O, is given by

O = Z2 ⋉ T = T ∪ { √1 (±q ± q ′ ) ∶ q, q ′ = (1, i), (1, j), (1, k), (i, j), (j, k), (k, j)} , (1.98)
2

which is of order 48. There are also seven conjugacy classes for O given, in a similar
notation, by

C1 = {1} , C2 = {−1} , C4 = { ± i, ±j, ±k} ,

C3 = { 21 (−1 ± i ± j ± k)} , C6 = −C3 ,
C ′4 1
= { √ ( ± i ± j), √ ( ± j ± k), √1 ( ± i ± k)} ,
1
2 2 2
C8 = { √1 (±1 ± i), √1 (±1 ± j), √1 (±1 ± k)} , (1.99)
2 2 2

with dimensions 1 + 1 + 6 + 8 + 8 + 12 + 12 = 48. The group O = O/{1, −1} ≃ S4 and

is composed of five conjugacy classes C1 = {1}, C2 = {i, j, k}, C ′2 = { √1 (q ± q ′ ) ∶ q, q ′ =
2
(i, j), (j, k), (k, i)}, C3 = { 12 (−1 + q) ∶ q ∈ { ± i ± j ± k}}, C4 = { √1 (1 ± q) ∶ q = i, j, k} where
2
1 + 3 + 6 + 8 + 6 = 24.
For n = 5 there is a group I, the binary icosahedral group also denoted as 2I, which can
be defined through the conjugacy classes

C1 = {1} , C2 = {−1} ,
C4 = { ± i, ±j, ±k, 12 ( ± i ± σ j ± τ k), 21 ( ± τ i ± j ± σ k), 12 ( ± σ i ± τ j ± k)} ,
C6 = { 21 (1 ± i ± j ± k), 12 (1 ± τ i ± σ j), 21 (1 ± σ i ± τ k), 12 (1 ± τ j ± σ k)} , C3 = −C6 ,
C5 = { 21 (σ ± i ± τ j), 12 (σ ± j ± τ k), 12 (σ ± k ± τ i)} , C10 = −C5 ,
C ′5 = { 21 (−τ ± i ± σ k), 12 (−τ ± j ± σ i), 12 (τ ± k ± σ j)} , C ′10 = −C ′5 , (1.100)

with dimensions 1 + 1 + 30 + 20 + 20 + 12 + 12 + 12 + 12 = 120.7 In this case the quotient

I = I/{1, −1} ≃ A5 with five conjugacy classes, C1 , C3 , C5 , C ′5 , as in (1.100), and also C2 =
{i, j, k, 21 (i ± σj ± τ k), 21 ( ± τ i + j ± σ k), 12 ( ± σ i ± τ j + k)}. Thus 1 + 20 + 12 + 12 + 15 = 60.
The closure of (1.100) under multiplication depends on στ = 1, τ 2 − τ − 1 = 0 implying
τ − σ = 1, τ 2 + σ 2 = 3. There are two solutions, one is given in (1.94), the other is obtained
by taking τ ↔ −σ. The binary icosahedral group contains the subgroup T = 2T , which may
be generated by taking r = 21 (i + σ j + τ k), s = 12 (1 + τ i + σ j), rs = 12 (−1 + τ j + σ k), where
7
Although I has order 120 it is not isomorphic to S5 or Z2 × A5 .

20
r2 = s3 = −1, (rs)3 = 1. The cosets tr T, r = 1, 2, 3, 4, for t = 21 (σ + i + τ j), t5 = 1, give, with
T, the full group I. This verifies ∣I∣ = 5∣T∣ = 120.
More generally, if instead of (1.90), we were to take s = cos π/m 1 + sin π/m v, so that
m
s = −1, then requiring (1.91) leads to − sin π/m ≤ cos π/n ≤ sin π/m. If n = 2 this gives
Q4m . Otherwise for m > 3 the only possibilities are m = 4, n = 3, 4 and m = 5, 6, n = 3. This
just leads to groups already discussed above other than if m = n = 4 when cos φ = ±1 and
the group becomes C8 . In summary the finite quaternion groups which contain the element
−1 are just

2G 2I 2O 2T 2Dn 2Cn
order 120 48 24 4n 2n
normal subgroups Z2 2T, 2D2 , Z2 2D2 , Z2 2D 21 n 2∣n 2Ck k∣n
2Ck k∣n, Ck k∣n k odd Ck k∣n k odd
G = 2G/{1, −1} I ≃ A5 O ≃ S4 T ≃ A4 Dn Cn
(1.101)

where for completeness 2Cn = {eπr/n i ∶ r = 0, 1, . . . , 2n − 1} ≃ C2n and k∣n denotes k divides
n. The only remaining finite group is Cn for n odd, as defined in (1.85). The groups
T, O, I are related to the symmetries of the tetrahedron, cube or equivalently octahedron,
dodecahedron or equivalently icosahedron (in crystallographic literature I is denoted by Y ).
Abstractly they can be generated by elements r, s, t = rs satisfying rl = sm = tn = e. With
labels (n, m, l) then Dn , T, O, I correspond to (n, 2, 2), (3, 3, 2), (4, 3, 2), (5, 3, 2) where for
any such group P
1 1 1 2
+ + =1+ . (1.102)
l m n ∣P ∣
The labels for Dn , T, O, I are the unique integers n, m, l, up to a reordering, such that the
left hand side in (1.102) is > 1.
Under conjugation the real part of any quaternion group element is invariant so the imag-
inary parts of the unit quaternions in each conjugacy class (1.97), (1.99) and (1.100) form
closed sets under conjugation by any element of T, O and I respectively. Under conjugation
the normal Z2 subgroups {1, −1} leave any quaternion invariant so the imaginary parts de-
fine points in three dimensional space which are closed under the action of T, O and I. For
T from (1.97) this gives the four points (1, 1, 1), (−1, −1, 1), (−1, 1, −1), (1, −1, −1) which
form the vertices of a tetrahedron and also the six points (±1, 0, 0), (0, ±1, 0), (0, 0, ±1)
correspond to the midpoints of the tetrahedron edges and form the vertices of an oc-
tahedron. For O using (1.99) the same octahedron reappears and also (±1, ±1, ±1) giv-
ing the eight vertices of a cube. For I the results in (1.100) lead to two sets of 12
points (±1, ±τ, 0), (0, ±1, ±τ ), (±τ, 0, ±1) and (±1, 0, ±σ), (±σ, ±1, 0), (0, ±σ, ±1) which, since
σ = 1/τ , are both the vertices of icosahedron (where 5 triangles meet at each vertex). The 20
points 21 (±1, ±1, ±1), 21 (±τ, ±σ, 0), (0, ±τ, ±σ), 21 (±σ, 0, ±τ ) arising from the conjugacy class
C6 form the vertices of a dodecahedron (where 3 pentagons meet at each vertex). The
remaining cases are, for O from C ′4 , √1 (±1, ±1, 0), √1 (0, ±1, ±1), √1 (±1, 0, ±1) and, for I
2 2 2
from the conjugacy class C4 , (±1, ±1, ±1), 12 (±1, ±σ, ±τ ), 21 (±τ, ±1, ±σ), 12 (±σ, ±τ, ±1) which
correspond to the midpoints of the 12 and 30 edges respectively. These are the vertices
of the cuboctahedron (this has 8 triangle and 6 square faces, two of each meeting at each

21
vertex) and icosidodecahedron (this has 20 triangle and 12 pentagon faces, with two each
meeting at each vertex).

1.6 Matrix Groups

It is easy to see that any set of non singular matrices which are closed under matrix multi-
plication form a group since they satisfy (1.2),(1.3),(1.4) with the identity e corresponding
to the unit matrix and the inverse of any element given by the matrix inverse, requiring
that the matrix is non singular so that the determinant is non zero. Many groups are de-
fined in terms of matrices. Thus Gl(n, R) is the set of all real n × n non singular matrices,
Sl(n, R) are those with unit determinant and Gl(n, C), Sl(n, C) are the obvious exten-
sions to complex numbers. Since det(M1 M2 ) = det M1 det M2 and det M −1 = (det M )−1 the
matrix determinants form an invariant abelian subgroup unless the the conditions defining
the matrix group require unit determinant for all matrices. The commutator subgroup
for Gl(n, R), [Gl(n, R), Gl(n, R)] ≃ Sl(n, R). It is easy to see that A−1 B −1 AB has unit
determinant for any A, B ∈ Gl(n, R) and any element in Sl(n, R) can be obtained as a
commutator. The same of course applies for Gl(n, C) and Sl(n, C). For M ∈ Gl(n, R) there
are n2 real parameters while for M ∈ Sl(n, R), with one condition, there are n2 − 1. The
same applies for Gl(n, C) and Sl(n, C) although the parameters are then complex.
The trivial one dimensional case for Gl(1, R) ≃ R where for x, y ∈ R the group operation
is just addition, the identity is 0 and the inverse of x is −x.
Various matrix groups which are subgroups of Gl(n, R) or Gl(n, C) are obtained by
requiring a bilinear or sesquilinear quadratic form ⟨x, y⟩ for x, y ∈ Rn or x, y ∈ Cn is invariant
under the group action on x, y. For a sesquilinear form ⟨x, y⟩ then ⟨x, y +y ′ ⟩ = ⟨x, y⟩+ ⟨x, y ′ ⟩,
⟨x + x′ , y⟩ = ⟨x, y⟩ + ⟨x′ , y⟩ and ⟨α x, β y⟩ = α∗ β⟨x, y⟩ for α, β ∈ C. The sesquilinear form is
hemitian if ⟨x, y⟩ = ⟨y, x⟩∗ .
Continuous such matrix groups of frequent interest are

1.6.1 Orthogonal

(i) O(n), real orthogonal n × n matrices {M }, so that

MT M = 1 . (1.103)

This set of matrices is closed under multiplication since (M1 M2 )T = M2 T M1 T . For SO(n)
det M = 1. The invariant quadratic form is just ⟨x, x⟩ = xT x. With x ∈ Rn this is positive
definite. A general n × n real matrix has n2 real parameters while a symmetric matrix
has 21 n(n + 1). M T M is symmetric so that (1.103) provides 21 n(n + 1) conditions. Hence
O(n), and also SO(n), have 21 n(n − 1) parameters. For n even ±1 ∈ SO(n) and these
form the centre of the group so long as n ≥ 2. Thus Z(SO(2n)) ≃ Z2 , n = 2, 3, . . . , while
Z(SO(2n + 1)) = {1}, n = 1, 2, . . . is trivial although Z(O(2n + 1)) = {±1} ≃ Z2 , n = 1, 2, . . .
and O(2n + 1) ≃ Z2 × SO(2n + 1). For the trivial one dimensional case O(1) ≃ Z2 .

22
1.6.2 Unitary

(ii) U (n), complex unitary n × n matrices, so that

M †M = 1 . (1.104)

Closure follows from (M1 M2 )† = M2 † M1 † . For SU (n) det M = 1. The invariant quadratic
form ⟨x, x⟩ = x† x for x ∈ Cn is hermitian. A general n × n complex matrix has 2n2 real
parameters while a hermitian matrix has n2 . M † M is hermitian so that U (n) has n2
parameters. (1.104) requires ∣ det M ∣ = 1 so imposing det M = 1 now provides one additional
condition so that SU (n) has n2 − 1 parameters. The centre of U (n) or SU (n) consists of all
elements proportional to the identity (this follows from Schur’s lemma shown later) so that
Z(SU (n)) = {e2rπi/n 1 ∶ r = 0, . . . n − 1} ≃ Zn , while Z(U (n)) = {eiα 1 ∶ 0 ≤ α < 2π} ≃ U (1).
Note that SO(2) ≃ U (1) since a general SO(2) matrix

cos θ − sin θ
( ), 0 ≤ θ < 2π , (1.105)
sin θ cos θ

is in one to one correspondence with a general element of U (1),

eiθ , 0 ≤ θ < 2π . (1.106)

Topologically U (1) ≃ S 1 , the circle.

1.6.3 Symplectic

(iii) Sp(2n, R) and Sp(2n, C), symplectic 2n × 2n real or complex matrices satisfying

M T J2n M = J2n , (1.107)

where J2n is a 2n × 2n antisymmetric matrix with the standard form

0 1 0
⎛ −1 0 ⎞
0 1
⎜ 0 −1 0 ⎟
J2n = ⎜ ⎟. (1.108)
⎜ ⋱ ⎟
⎝ 0 1 ⎠
−1 0

In this case M T J2n M is antisymmetric so that (1.107) provides n(2n − 1) conditions and
hence Sp(2n, R) has n(2n + 1) parameters. For symplectic transformations there is an anti-
symmetric invariant form ⟨v ′ , v⟩ = −⟨v, v ′ ⟩ = v ′T J2n v so that ⟨x, x⟩ = 0. For an orthonormal
basis {ei } we may define Jij = ⟨ei , ej ⟩.
The condition (1.107) requires det M = 1 so there are no further restrictions as for O(n)
and U (n). To show this we define the Pfaffian8 for 2n × 2n antisymmetric matrices A by
1
Pf(A) = εi ...i Ai i . . . Ai2n−1 i2n , (1.109)
2n n! 1 2n 1 2
8
Johann Friedrich Pfaff, 1765-1825, German.

23
with εi1 ...i2n the 2n-dimensional antisymmetric symbol, ε1...2n = 1. The Pfaffian is essentially
the square root of the usual determinant since

det A = Pf(A)2 , (1.110)

and it is easy to see that

Pf(J2n ) = 1 . (1.111)
The critical property here is that

Pf(M T AM ) = det M Pf(A) since εi1 ...i2n Mi1 j1 . . . Mi2n j2n = det M εj1 ...j2n . (1.112)

Applying (1.112) with A = J2n to the definition of symplectic matrices in (1.107) shows that
we must have det M = 1.
Since both ±1 belong to Sp(2n, R) then the centre Z(Sp(2n, R)) ≃ Z2 .
For M = ( ac db ) ∈ Sp(2, R) the condition (1.107) requires just ad − bc = 1 or det M = 1.
Hence Sp(2, R) ≃ Sl(2, R).

1.6.4 Quaternion Matrix Groups

Matrix groups can also be extended to quaternions where a n × n quaternionic matrix M

has the form
M = a 1 + b i + c j + d k , a, b, c, d real n × n matrices , (1.113)
and the adjoint is
M̄ = aT 1 − bT i − cT j − dT k . (1.114)
Matrix multiplication of non singular, or invertible, n × n quaternionic matrices defines
Gl(n, H) since quaternions obey the crucial associativity property.9 The notion of a deter-
minant with the usual properties is problematic for quaternionic matrices but Sl(n, H) can
be defined as the commutator group [Gl(n, H), Gl(n, H)]. A definition analogous to a de-
terminant due to Dieudonné10 is based on the quotient group Gl(n, H)/[Gl(n, H), Gl(n, H)]
which is a one dimensional abelian group. This can be expressed, for M ∈ Gl(n, H), in terms
of a real ∣M ∣ ≥ 0, ∣M ∣ = ∣M K∣ = ∣KM ∣ for K ∈ [Gl(n, H), Gl(n, H)], satisfying the group
properties ∣M1 M2 ∣ = ∣M1 ∣ ∣M2 ∣, where ∣M̄ ∣ = ∣M ∣ and ∣rM ∣ = ∣r∣n ∣M ∣ for r ∈ R. In general
∣M ∣ = 0 if and only if M is singular, so that there is a quaternionic column vector v such
that M v = 0, and there is no inverse.11 For M ∈ Sl(n, H) then ∣M ∣ = 1 which is preserved
9
The inverse is a little more complicated than the usual matrix inverse since quaternions do not com-
mute. A single non zero quaternion has an inverse. For M ∈ Gl(2, H) then writing M = ( ac db ) =
( ca1−1 01 ) ( a0 d0′ ) ( 10 a 1 b ) with a ≠ 0 and d′ = d − ca−1 b ≠ 0 then M −1 = ( 01 −a1 b ) ( a0 d′0−1 ) ( −ca1 −1 01 ) is both a
−1 −1 −1

right and left inverse for M . Alternatively for d ≠ 0, M = ( 10 bd1 ) ( a0 d0 ) ( d−1 1 0

), a′ = a − bd−1 c leads to an
−1 ′
c 1
−1
equivalent expression for M . These results can be generalised to larger square quaternionic matrices.
10
Jean Dieudonné, 1906-92, French. Founder member of Bourbaki.
11
For M = ( ac db ) ∈ Gl(2, H) then ∣M ∣ = ∣a∣ ∣d − ca−1 b∣ = ∣d∣ ∣a − bd−1 c∣. Note that if a → a + q c, b → b + q d or
b → b + a q, d → d + c q, q ∈ H, which correspond to M → QM or M → M Q for Q = ( 10 1q ), ∣M ∣ is invariant.
For ∣M ∣ = 0 and ∣a∣ ≠ 0, when it is necessary that d − ca−1 b = 0, we can take for the zero eigenvalue vector
v = ( a−1 b ).
−1

24
under matrix multiplication. While M ∈ Gl(n, H) has 4n2 real parameters, M ∈ Sl(n, H)
has 4n2 − 1.
For a single quaternion the group Q defined in (1.84) is isomorphic to SU(2). To show
this we relate the quaternions to 2 × 2 matrices according to

1 → 1 = ( 10 01 ) , i → I = ( 0i −i
0 ), 0 1),
j → J = ( −1 0 k → K = ( 0i 0i ) . (1.115)

Then
x0 + x1 i x3 + x4 i
q = x0 1 + x1 i + x2 j + x3 k, ↔ Q=( ), x0 , x1 , x2 , x3 ∈ R , (1.116)
−x3 + x4 i x0 − x1 i
ensures
q1 q2 ↔ Q1 Q2 , q̄ ↔ Q† , ∣q∣2 ↔ det Q . (1.117)
Furthermore
JQ = Q∗ J . (1.118)
Note that from (1.116) det Q = x0 2 + x1 2 + x2 2 + x3 2 = ∣q∣2 so that SU (2) can be identified
with points on S 3 .
For any quaternion matrix M there is an associated 2n×2n complex matrix M obtained
by replacing quaternions by 2 × 2 matrices as in (1.115)

M → M, M̄ → M† , 1n 1 → 12n , 1n j → J˜2n ⇒ M∗ = −J˜2n MJ˜2n , (1.119)

where M1 M2 → M1 M2 , M −1 → M−1 . With M as in (1.113)

e f 0 1n
M = ( ∗ ∗) , e = a + bi, f = c + di, J˜2n = ( ). (1.120)
−f e −1n 0

With the usual notion of a determinant for complex matrices det M = ∣M ∣2 .

The unitary quaternion matrix group U (n, H) is defined by n×n matrices of quaternions
where for any M ∈ U (n, H) with an associated adjoint M̄ , defined as in (1.113) and (1.114),

M̄ M = 1n 1 , (1.121)

for 1n the unit n × n matrix. From (1.116) U (1, H) ≃ SU (2). A general quaternionic n × n
M then has 4n2 parameters whereas U = M̄ M = Ū is a hermitian quaternion matrix which
has n real diagonal elements and 21 n(n − 1) independent off diagonal quaternionic numbers
giving n(2n − 1) parameters altogether. Hence (1.121) provides n(2n − 1) conditions so that
U (n, H) has n(2n + 1) parameters. The condition (1.121) requires ∣M ∣ = 1. From (1.116)
U (1, H) ≃ SU (2). In this case under the map (1.119) M satisfies, since M∗ = (MT )−1 ,
satisfies (1.107) as well as M† M = 12n and so M ∈ Sp(2n, C) ∩ U (2n). These properties are
preserved under multiplication and define the group U Sp(2n) which is therefore isomorphic
to U (n, H). This group may also be denoted by Sp(n).

25
1.6.5 Heisenberg Group

The Heisenberg12 group H is defined in terms of upper triangular 3 × 3 matrices

⎛1 a c ⎞
A(a, b, c) = ⎜0 1 b ⎟ , a, b, c ∈ R , (1.122)
⎝0 0 1⎠

where under matrix multiplication A(a, b, c)A(a, b′ , c′ ) = A(a + a′ , b + b′ , c + c′ + ab′ ) and

the inverse A(a, b, c)−1 = A(−a, −b, −c + ab). This group is connected with the position
momentum commutation relations in quantum mechanics. The Heisenberg group is non
abelian and since A(a, b, c)A(a, b′ , c′ )A(a, b, c)−1 = A(0, 0, c′ + ab′ − ba′ ) then {A(0, 0, c)} ≃ R
forms its centre Z(H). Clearly {A(a, 0, 0)} ≃ {A(0, b, 0)} ≃ R are abelian subgroups. There
is a discrete infinite subgroup by restricting a, b, c ∈ Z.

1.6.6 Compact and Non Compact

The matrix groups SO(n), SU (n) and U (n, H) are compact, which will be defined precisely
later but for the moment can be taken to mean that the natural paramete rs vary over a
finite range. On the other hand Gl(n, R), Sl(n, R), as well as their complex counterparts,
are non compact. Any one dimensional continuous subgroup of a compact continuous group
must be isomorphic to U (1) while a non compact continuous group will have at least one
dimensional subgroups isomorphic to R. Sp(2n, R) is non compact, which is evident since
matrices of the form
cosh θ sinh θ
( ), −∞ < θ < ∞ . (1.123)
sinh θ cosh θ
belong to Sp(2, R) and form a one dimensional subgroup isomorphic to R. The Heisenberg
group is clearly non compact.
The group U Sp(2n) ≃ U (n, H) is compact.
There are also various extensions of the orthogonal and unitary groups to non compact
groups which arise frequently in physics. Suppose g is the diagonal (n + m) × (n + m) matrix
defined by
1 0
g=( n ), (1.124)
0 −1m
then the pseudo-orthogonal groups O(n, m), and hence SO(n, m), are defined by real ma-
trices M such that
M Tg M = g . (1.125)
The invariant form in this case is ⟨v ′ , v⟩ = v ′T gv is no longer positive. Similarly we may
define U (n, m) and SU (n, m). It is easy to see that O(n, m) ≃ O(m, n) and similarly
for other analogous cases. The parameter count for these groups is the same as for the
corresponding O(n + m) or U (n + m), SU (n + m). Note that matrices belonging to SO(1, 1)
are just those given in (1.123).
12
Werner Karl Heisenberg, 1901-76, German. Nobel prize 1932.

26
2 Representations

In physical applications of groups representations play a crucial role. For any group G a
representation is a set of non singular (i.e. non zero determinant) square matrices {D(g)},
for all g ∈ G, such that

D(g1 )D(g2 ) = D(g1 g2 ) , (2.1)

D(e) = 1 , (2.2)
−1 −1
D(g ) = D(g) , (2.3)

where 1 denotes the unit matrix. If the matrices D(g) are n × n the representation has
dimension n. For each matrix group the definition of course provides a representation
which is termed the fundamental representation.
The representation is faithful if D(g1 ) ≠ D(g2 ) for g1 ≠ g2 . There is always a trivial
representation or singlet representation in which D(g) = 1 for all g. If the representation
is not faithful then there exist group elements h, not equal to e, where D(h) = 1. For all
such h then {h} = H and it is easy to see that H must be a subgroup of G, moreover it is
a normal subgroup.
For complex matrices the conjugate representation is defined by the matrices D̄(g) =
D(g)∗ since complex conjugation preserves matrix multiplication. The matrices (D(g)−1 )T
also define a representation.
Since

det (D(g1 )D(g2 )) = det D(g1 ) det D(g2 ) , det 1 = 1 , det D(g)−1 = ( det D(g))−1 , (2.4)

{det D(g)} form a one-dimensional representation of G which may be trivial and in general
is not faithful.
Two representations of the same dimension D(g) and D′ (g) are equivalent if

D′ (g) = SD(g)S −1 for all g ∈ G , (2.5)

where D(g) → D′ (g) is a similarity transformation.

For any finite group G = {gi } of order n = ∣G∣ we may define the dimension n regular
representation by considering the action of the group on itself

ggi = ∑ gj Dreg,ji (g) , (2.6)

j

where [Dreg,ji (g)] are representation matrices with a 1 in each column and row and with
all other elements zero. In general
⎧
⎪δji ,
⎪ g = e,
Dreg,ji (g) = ⎨ (2.7)
⎩0 , j = i , g ≠ e.
⎪
⎪

27
As an example for D3 = {e, a, a2 , b, ba, ba2 }, where a3 = b2 = e, ab = ba2 , then
001000 000100
⎛1 0 0 0 0 0⎞ ⎛0 0 0 0 1 0⎞
Dreg (a) = ⎜ 00 10 00 00 01 00 ⎟ , Dreg (b) = ⎜ 01 00 00 00 00 10 ⎟ . (2.8)
⎝0 0 0 0 0 1⎠ ⎝0 1 0 0 0 0⎠
000100 001000

A representation of dimension n acts on an associated n-dimensional vector space V,

the representation space. For any vector v ∈ V we may define a group transformation acting
on v by
v → v g = D(g)v . (2.9)
g

Transformations as in (2.5) correspond to a change of basis for V. A representation is

reducible if there is a subspace U ⊂ V, U ≠ V, such that

D(g)u ∈ U for all u ∈ U , (2.10)

otherwise it is an irreducible representation or irrep. For a reducible representation we may

define a representation of lower dimension by restricting to the invariant subspace. More
explicitly with a suitable choice of basis we may write, corresponding to (2.10),

D̂(g) B(g) û
D(g) = ( ) for u = ( ) , (2.11)
0 C(g) 0

where the matrices D̂(g) form a representation of G. If, for any invariant subspace, we may
restrict the representation matrices to the form shown in (2.11) with B(g) = 0 for all g the
representation is completely reducible.
For an abelian group G all irreducible representations are one-dimensional since all
matrices D(g) commute for all g ∈ G and they may be simultaneously diagonalised. For
the n-dimensional translation group Tn , defined by n-dimensional vectors under addition
(with 0 as the unit), then for a representation it necessary, for a ∈ Rn , a → D(a) satisfying
D(a1 )D(a2 ) = D(a1 + a2 ). Irreducible representations are all of the form D(a) = eb⋅a , for
any n-vector b dual to a.
Representations need not be completely reducible, if {R} are n × n matrices forming a
group GR and a is a n-component column vector then we may define a group in terms of
the matrices
R a
D(R, a) = ( ), (2.12)
0 1
with the group multiplication rule

D(R1 , a1 )D(R2 , a2 ) = D(R1 R2 , R1 a2 + a1 ) , (2.13)

which has the abelian subgroup Tn for R = 1. The group defined by (2.13) is then GR ⋉ Tn
with aR = Ra.
In general for a completely reducible representation the representation space V decom-
poses into a direct sum of invariant spaces Ur which are not further reducible, V ≃ ⊕kr=1 Ur ,

28
and hence there is a matrix S such that

⎛D1 (g) 0 ⎞
0 D
SD(g)S −1 = ⎜ 2 (g)
⎜ ⎟
⎟, (2.14)
⎜ ⋱ ⎟
⎝ Dk (g) ⎠

and where Dr (g) form irreducible representations for each r. This can be written as
k
D(g) ≃ ⊕ Dr (g) , (2.15)
r=1

Thus for R the representation given by the matrices D(g) and Rr corresponding to the
irreducible representation matrices Dr (g) then R is decomposed as

R = R1 + ⋅ ⋅ ⋅ + Rk . (2.16)

If there are NG inequivalent irreducible representations they may be labelled Rr , r =

1, . . . nG , and then in general a particular irreducible representation Rr may appear more
than once, with multiplicity mr , in the decomposition (2.16), and (2.15) can be reduced to
just
NG
D(g) ≃ ⊕ mr Dr (g) , (2.17)
r=1

Representations of a finite group are always completely reducible since, as shown later, rep-
resentations can be shown to be equivalent to unitary representations formed from unitary
matrices. This is known as Maschke’s threorem.13
There is always a trivial one dimensional representation R1 which is irreducible and is
given by
D1 (g) = 1 for all g ∈ G . (2.18)

As an example from (2.8)

1 0 0 0 0 0 1 0 0 0 0 0
⎛0 s 0 0 0 0 ⎞ ⎛0 0 1 0 0 0 ⎞
−1 −1
SDreg (a)S = ⎜ 00 0 s̄
0 0
0 0
1 0
0
0 ⎟, S Dreg (b)S = ⎜ 00 1
0
0
0
0
−1
0
0
0 ⎟
0 ,
⎝0 0 0 0 s 0 ⎠ ⎝0 0 0 0 0 −1 ⎠
0 0 0 0 0 s̄ 0 0 0 0 −1 0
1 s̄ s −1 −s̄ −s
1 ⎛1 s s̄ −1 −s −s̄ ⎞ √
−1 1 −1 −1 −1 ⎟
S = √ ⎜ 11 1
s s̄ 1 s s̄ , s = e2πi/3 = − 12 (1 − 3 i) . (2.19)
6⎝1 s̄ s 1 s̄ s⎠
1 1 1 1 1 1

In consequence this representation decomposes into two one-dimensional representations

and two equivalent two-dimensional ones.
13
Heinrich Maschke, 1853-1908, German.

29
2.1 Schur’s Lemmas

Two useful results, which follow almost directly from the definition of irreducibility, char-
acterising irreducible representations are:
Schur’s Lemmas.14 If D1 (g), D2 (g) form two irreducible representations then (i)

SD1 (g) = D2 (g)S , (2.20)

for all g requires that the two representation are equivalent or S = 0. Also (ii)

SD(g) = D(g)S , (2.21)

for all g for an irreducible representation D(g) then S ∝ 1.

To prove (i) suppose V1 , V2 are the representation spaces corresponding to the repre-
sentations given by the matrices D1 (g), D2 (g), so that V1 → V2 . Then the image of S,
S
Im S = {v ∶ v = Su, u ∈ V1 }, is an invariant subspace of V2 , D2 (g) Im S = Im S D2 (g), by
virtue of (2.20). Similarly the kernel of S, Ker S = {u ∶ Su = 0, u ∈ V1 } forms an invariant
subspace of V1 , both sides of (2.20) giving zero. For both representations to be irreducible
we must have Im S = V2 , Ker S = 0, so that S is invertible, det S ≠ 0, (this is only possible
if dim V2 = dim V1 ). Since then D2 (g) = SD1 (g)S −1 for all g the two representations are
equivalent.
To prove (ii) suppose the eigenvectors of S with eigenvalue λ span a space Vλ . Applying
(2.21) to Vλ shows that D(g)Vλ are also eigenvectors of S with eigenvalue λ so that D(g)Vλ ⊂
Vλ and consequently Vλ is an invariant subspace unless Vλ = V and then S = λI.
To obtain (ii) it is necessary in general that Vλ is a complex vector space so that the
reduction to irreducible representations must allow for complex representations. For an
abelian group all matrices D(g) commute and can be simultaneously diagonalised so they
are reducible to one dimensional complex representations.

2.2 Induced Representations

A representation of a group G also gives a representation when restricted to a subgroup H.

If the representation for G is irreducible the restricted representation for H need not be.
Conversely for a subgroup H ⊂ G then it is possible to obtain representations of G in
terms of those for H by constructing the induced representation. Assume

v → D(h)v , h∈H, v∈V, (2.22)

h

with V the representation space for this representation of H. For finite groups the cosets
forming G/H may be labelled by an index i so that for each coset we may choose an element
gi ∈ G such that all elements belonging to the i’th coset can be expressed as gi h for some
14
Issai Schur, 1875-1941, Russian, worked in Germany, forced to leave in 1939 for Palestine.

30
h ∈ H. The choice of gi is arbitrary to the extent that we may let gi → gi hi for some fixed
hi ∈ H. For any g ∈ G then

ggi = gj h for some h∈H, i, j = 1, . . . , N , N = ∣G∣/∣H∣ . (2.23)

Assuming (2.23) determines h the induced representation is defined so that that under the
action of a group transformation g ∈ G,

vi → D(h)vj , vi = (gi , v) , D(h)vj = (gj , D(h)v) . (2.24)

g

In (2.24) h depends on i as well as g and vi ∈ Vi which is isomorphic to V for each i so

that the representation space for the induced representation is the N -fold tensor product
V ⊗N . The representation matrices for the induced representation are then given by N × N
matrices whose elements are D(h) for some h ∈ H,

⎪D(h) , gj−1 g gi = h ∈ H ,
⎧
⎪
Dji (g) = ⎨ (2.25)
⎩0,
⎪
⎪ gj−1 g gi ∉ H .

To show that (2.24) is in accord with the group multiplication rule we consider a subsequent
transformation g ′ so that

vi → D(h)vj →′ D(h′ )D(h)vk = D(h′ h)vk for g ′ gj = gk h′ ⇒ (g ′ g)gi = gk h′ h . (2.26)

g g

The dimension of the induced representation of G is then ∣G∣/∣H∣ dim RH with RH the
representation defined by {D(h)}.
If H = {e}, forming a trivial subgroup of G, and D(h) → 1, the induced representation
is identical with the regular representation for finite groups. This shows that the induced
representation is not in general irreducible.
As a simple example we consider G = Dn generated by elements a, b with an = b2 =
e, ab = ban−1 . H is chosen to be the abelian subgroup Zn = {ar ∶ r = 0, . . . , n − 1}. This has
one-dimensional representations labelled by k = 0, 1, . . . , n − 1 defined by
2πki
v→e n v. (2.27)
a

With this choice for H there are two cosets belonging to Dn /Zn labelled by i = 1, 2 and we
may take g1 = e, g2 = b. Then for v1 = (e, v) transforming as in (2.27) then with v2 = (b, v)
(2.24) requires, using ab = ba−1 ,

v1 , e−
2πki 2πki
(v1 , v2 ) → (e n n v2 ) = (v1 , v2 )Ak , (v1 , v2 ) → (v2 , v1 ) = (v1 , v2 )B , (2.28)
a b

for 2 × 2 complex matrices Ak , B,

2πki
⎛e n 0 ⎞ 0 1
Ak = − 2πki
, B=( ), (2.29)
⎝ 0 e n ⎠ 1 0

which satisfy Akn = 1, B 2 = 1, Ak B = BAkn−1 and so give a two dimensional representation

of Dn for each k. By considering Ak → BAk B it is clear that the representation for k → n−k

31
is equivalent to that in (2.29). For n even we may take k = 1, . . . , (n − 2)/2, for n odd
k = 1, . . . , (n − 1)/2 to give inequivalent two dimensional irreducible representations. If
k = 0 in (2.29) A = 1 then the matrix B is reducible so that there are two one dimensional
representations corresponding to taking b → ±1. For n even then taking k = n/2 and A = −1
there are similarly two more one dimensional representations. The representations are then
given by, for n odd,

R1,1 ∶ (ar , ar b) → (1, 1) , r = 0, . . . , n − 1 ,

r r
R1,2 ∶ (a , a b) → (1, −1) , r = 0, . . . , n − 1 ,
r r
R2,k ∶ (a , a b) → (Akr , Akr B) , r = 0, . . . , n − 1 , k = 1, . . . , 12 (n − 1) , (2.30)

and for n even

R1,1 ∶ (ar , ar b) → (1, 1) , r = 0, . . . , n − 1 ,

r r
R1,2 ∶ (a , a b) → (1, −1) , r = 0, . . . , n − 1 ,
r r
R1,3 ∶ (a , a b) → ((−1)r , (−1)r ) , r = 0, . . . , n − 1 ,
R1,4 ∶ (ar , ar b) → ((−1)r , −(−1)r ) , r = 0, . . . , n − 1 ,
R2,k ∶ (ar , ar b) → (Akr , Akr B) , r = 0, . . . , n − 1 , k = 1, . . . , 12 (n − 1) , (2.31)

The number of representations match the number of conjugacy classes in (1.58). Corre-
sponding to (2.29) there is an equivalent basis

cos 2πk − sin 2πk 1 0

R Ak R−1 = ( n n ) , R B R−1 = ( ), (2.32)
sin 2πk
n cos 2πk
n
0 −1

where R = √1 1 1 ), R−1 = R† .
( −i
2 i

A very similar construction works for the dicyclic group Q4n as defined in (1.85). This
is similar to D2n where we take a2n = e, ab = ba−1 but now an = b2 . Thus
πki
πki
− πki ⎛e n 0 ⎞
(v1 , v2 ) → (e n v1 , e n v2 ) = (v1 , v2 )Ak , Ak = − πki
, k = 0, 1, . . . , n . (2.33)
a ⎝ 0 e n ⎠

Hence an gives (v1 , v2 ) → (−1)k (v1 , v2 ) so that instead of (2.28) for the action of b we may
require

0 1
(v1 , v2 ) → ((−1)k v2 , v1 ) = (v1 , v2 )Bk , Bk = ( k ), k = 0, . . . , n . (2.34)
b (−1) 0

For k = 0, A0 = 1 and B0 ∼ ( 10 −1
0 ), for k = n, A = −1 and B ∼ ( 1 0 ), n even, B ∼ ( i 0 ),
n n 0 −1 n 0 −i
n odd. Hence there are four one dimensional irreducible representations and n − 1 two
dimensional ones. For k = 1 the representation (2.33) and (2.34) is equivalent to using the
quaternion representation in (1.115) in (1.85). The representations follow a similar pattern

32
to the dihedral case, for n odd
R1,1 ∶ (ar , ar b) → (1, 1) , r = 0, . . . , 2n − 1 ,
r r
R1,2 ∶ (a , a b) → (1, −1) , r = 0, . . . , 2n − 1 ,
r r
R1,3 ∶ (a , a b) → ((−1)r , (−1)r i) , r = 0, . . . , 2n − 1 ,
R1,4 ∶ (ar , ar b) → ((−1)r , −(−1)r i) , r = 0, . . . , 2n − 1 ,
R2,k ∶ (ar , ar b) → (Akr , Akr Bk ) , r = 0, . . . , 2n − 1 , k = 1, . . . , n − 1 , (2.35)
and for n even
R1,1 ∶ (ar , ar b) → (1, 1) , r = 0, . . . , 2n − 1 ,
r r
R1,2 ∶ (a , a b) → (1, −1) , r = 0, . . . , 2n − 1 ,
r r
R1,3 ∶ (a , a b) → ((−1)r , (−1)r ) , r = 0, . . . , 2n − 1 ,
R1,4 ∶ (ar , ar b) → ((−1)r , −(−1)r ) , r = 0, . . . , 2n − 1 ,
R2,k ∶ (ar , ar b) → (Akr , Akr Bk ) , r = 0, . . . , 2n − 1 , k = 1, . . . , n − 1 . (2.36)
For k even the representations can be brought to a real form by a similarity transformation
as in (2.32).

2.3 Unitary Representations

For application in quantum mechanics we are almost always interested in unitary represen-
tations where the matrices are require to satisfy
D(g)† = D(g −1 ) = D(g)−1 . (2.37)
For such representation then the usual scalar product on V is invariant, for transformations
as in (2.9) v1 g† v2 g = v2 † v1 . If U is an invariant subspace then the orthogonal subspace U⊥ , as
defined by the scalar product, is also an invariant subspace. Hence unitary representations
are always completely reducible.
Theorem: For a finite group all representations are equivalent to unitary representations.
To show this define
S = ∑ D(gi )† D(gi ) , (2.38)
i
where the sum is over all elements of the group G = {gi }. As a consequence of (1.7)
SD(g)−1 = SD(g −1 ) = ∑ D(gi )† D(gi g −1 )
i
= ∑ D(gi g)† D(gi )
i
= D(g)† ∑ D(gi )† D(gi ) = D(g)† S , (2.39)
i

using that D(g) form a representation and also (AB)† = B † A† . Hence if we define ⟨v1 , v2 ⟩ =
v1 † Sv2 then we have ⟨v1 , D(g −1 )v2 ⟩ = (D(g)v1 , v2 ) or ⟨v1 g , v2 g ⟩ = ⟨v1 , v2 ⟩. With respect to
this scalar product D(g) is unitary (or we may define D′ (g) = S 2 D(g)S − 2 and then show
1 1

D′ (g)† D′ (g) = 1).

33
2.3.1 Infinite Dimensional Unitary Representations

Representations can be infinite dimensional when they are typically expressed in the terms of
the group action on spaces of functions. Such representations arise when considering unitary
representations of non compact groups. As an example we consider unitary representations
for the Heisenberg group as define by products of matrices of the form (1.122). Acting
on complex square integrable functions {f } on R belonging to L2 we define an action
f → T(a,b,c) f , for a, b, c real, by

T(a,b,c) f (x) = ei(bx+c) f (x + a) . (2.40)

It is an exercise to check this satisfies the group multiplication rules and that this forms a
unitary representation.

2.3.2 Real and Pseudo-Real Representations

For any unitary complex representation the conjugate representation may be equivalent so
that
D(g)∗ = C D(g)C −1 for all g ∈ G . (2.41)
Assuming the representation is unitary so that D(g)∗ = D(g)−1T then using (2.41) and its
transpose for D(g)−1

C −1 T C D(g) C −1 C T = D(g) ⇒ [D(g), C −1 C T ] = 0 . (2.42)

For an irreducible representation Schur’s lemma requires

C −1 C T = α1 ⇒ CT = α C ⇒ C = α CT ⇒ α2 = 1 . (2.43)

Hence α = ±1. For α = 1 there is an S such that S T C S = 1 and then S −1 D(g)S = D′ (g) is
a real representation. Otherwise C T = −C and the representation is pseudo-real. For C to
be invertible it must be even dimensional and it is then reducible to the form (1.108).
A prescription for C is obtained by taking

C = ∑ D(gi )T U D(gi ) , (2.44)

i

for arbitrary U since then, using (1.7) again,

C D(g) = ∑ D(gi g −1 )T U D(gi ) = D(g)−1T C . (2.45)

i

Thus for real, pseudo-real representations C T = ±C so that

T T T
∑ D(gi ) U D(gi ) = ± ∑ D(gi ) U D(gi ) ⇒ ∑ Drs (gi )Duv (gi ) = ± ∑ Dus (gi )Drv (gi ) .
i i i i
(2.46)
For a complex representation there can be no matrix C as defined in (2.44) for any U . This
then requires in this case
∑ Drs (gi )Duv (gi ) = 0 . (2.47)
i

34
 For the two dimensional representation of the quaternion groups considered in 1.5 de-
fined by (1.115) then q → Q, q̄ → Q† = Q−1 so that the representation is unitary and also
irreducible since the only 2 × 2 matrix commuting with I, J, K is proportional to 1. Also
as J(I, J, K)J = (I T , J T , K T ) then Q satisfies (2.41) with C → J = −J T . Hence these
representations are pseudo-real.
In general any pseudo-real representation can be written in terms of matrices of quater-
nions, with the quaternions having real coefficients.

2.4 Orthogonality Relations

Schur’s lemmas have an important consequence in that the matrices for irreducible repre-
sentations obey an orthogonality relation. To derive this we define
′
(R ,R) 1 (R′ ) −1 (R)
Srs,uv = ∑ Drv (gi )Dus (gi ) , (2.48)
∣G∣ i
′
where D(R) (g), D(R ) (g) are the matrices corresponding to the irreducible representation
R, R′ . Then

(R′ ,R) 1
(R) (R′ ) −1 (R) (R′ ) (R′ ,R)
Srt,uv Dts (g) = ∑ Drv (ggi )Dus (gi ) = Drt (g) Sts,uv ,
∣G∣ i
(R′ ,R) (R′ ) 1 (R′ ) (R′ ,R)
Srs,uw Dwv (g) = ∑D (gi−1 )Dus
(R) (R)
(ggi ) = Duw (g) Srs,wv , (2.49)
∣G∣ i rv

for any g ∈ G. The proof of (2.49) follows essentially since {gi } = {gi g}. Schur’s lemmas
(R′ ,R) (R′ ,R)
then requires that Srs,uv = 0 unless R′ = R when Srs,uv must be proportional to δrs and
also δuv . Hence we must have
′
(R ,R) 1
Srs,uv = δR′ R δrs δuv , (2.50)
nR

where nR = dim R is the dimension of the representation R. The constant in (2.50) is

(R,R) (R)
determined by considering Sru,us = ∑i Drs (e) = δrs .

2.5 Characters

For any representation R the character is defined by

χR (g) = tr(D(R) (g)) . (2.51)

Since traces are unchanged under cyclic permutations χR (g ′ gg ′−1 ) = χR (g) so that the
character depends only on the conjugacy classes of each element. Hence we may write
χ(gi ) ≡ χ(Cs ) for any gi ∈ Cs where for Nchar different conjugacy classes in G, s = 1, . . . , Nchar .
With previous conventions in (1.53) for gi ∈ Cs , χ(gi −1 ) = χ(Cs̄ ). Similarly the character is
unchanged when calculated for any representations related by an equivalence transformation

35
as in (2.5). Since for a finite group any representation is equivalent to a unitary one we
must also have
χR (g −1 ) = χR (g)∗ or χR (Cs̄ ) = χR (Cs )∗ . (2.52)
For real or pseudo-real representations due to (2.41) characters are real. If the character is
complex then χR ∗ = χR′ for R′ ≠ R the conjugate representation and necessarily there are
group elements such that C(g) ≠ C(g −1 ). As a special case

χR (e) = trR (1) = dim R = nR . (2.53)

For D(R) (g) equivalent to a unitary representation, as is the case for any finite group,
then, since D(R) (g) has nR eigenvalues of modulus one, ∣χR (g)∣ ≤ nR and ∣χR (g)∣ = nR > 1
only for D(R) (g) = ±1nR . For any direct product group G1 × G2 then χG1 ×G2 ((g1 , g2 )) =
χG1 (g1 ) χG2 (g2 ).
For the regular representation defined by (2.6) then by (2.7)

χreg (e) = ∣G∣ , χreg (g) = 0 , g ≠ e . (2.54)

For two representations R, R′ there is a scalar product for the associated characters
defined by

1 ∗ 1 Nchar ∗
⟨χR′ , χR ⟩ = ∑ χR′ (g) χR (g) = ∑ χR′ (Cs ) ds χR (Cs ) , ds = dim Cs . (2.55)
∣G∣ g∈G ∣G∣ s=1

The characters for the irreducible representations play a crucial role. For irreducible
representations {Rr ∶ r = 1, . . . , NG } the corresponding characters χr (g) = χRr (g). For
the singlet representation (2.18) χ1 (g) = 1. As a consequence of the orthogonality relation
for irreducible representations (2.50) then using (2.52) for two irreducible representations
R = Rr , R′ = Rr′ this reduces to

1 ∗ 1 Nchar ∗
⟨χr′ , χr ⟩ = ∑ χr′ (g) χr (g) = ∑ χr′ (Cs ) ds χr (Cs ) = δr′ r . (2.56)
∣G∣ g∈G ∣G∣ s=1

Equivalently
Xrs = χr (Cs ) , Dss′ = ds δss′ ⇒ X D X † = ∣G∣ 1nG . (2.57)
where X is a NG × Nchar matrix and D is a diagonal Nchar × Nchar matrix. (2.57) requires
the number of irreducible representations NG ≤ Nchar . From (2.18) and (2.53)

χ1 (Cs ) = 1 all s , χr (e) = dim Rr = nr . (2.58)

For any f (g) satisfying f (g) = f (g ′ gg ′−1 ) for all g ′ ∈ G, so that f (g) = f (Cs ) for all
∗
g ∈ Cs , we may then define ⟨f, χR ⟩ = ∑N char
s=1 f (Cs ) ds χR (Cs )/∣G∣, as in (2.56). An important
result, demonstrated later, is

⟨f, χr ⟩ = 0 for all r ⇒ f (g) = 0 . (2.59)

36
If NG > Nchar there is a non zero vector such that ∑r vr Xrs = 0. Since this contradicts
(2.59) we must have
NG = Nchar . (2.60)
Hence X is a non singular square matrix and using ∣G∣ X −1 = D X † equivalently
NG
X † X = ∣G∣ D−1 , ∗
∑ χr (Cs )χr (Cs′ ) = ∣G∣/ds δss′ . (2.61)
r=1

For s = s′ = 1 then C1 = {e} is the conjugacy class containing just the identity by itself,
d1 = 1, so that
NG
2
∑ nr = ∣G∣ . (2.62)
r=1
This plays an important role in constraining irreducible representations for finite groups.
(n−2)/2 2
As an illustration for the dihedral group Dn then ∑k=1 2 + 1 + 1 + 1 + 1 = 2n for n even
(n−1)/2 2
and ∑k=1 2 + 1 + 1 = 2n for n odd. For Q4n = 2Dn then (n − 1)22 + 4 × 1 = 4n.
Characters distinguish the different possible representations. From (2.46) and (2.47),
setting s = u, r = v and summing, for an irreducible such representation (note that if g, g ′ ∈ Cs
then g 2 , g ′2 ∈ Cs′ for some Cs′ )
⎧
⎪ ⟨χr , χr ⟩ ⎧
⎪ 1 real
1 ⎪
⎪
⎪ ⎪
⎪
⎪
2
FF S = ∑ χr (g ) = ⎨−⟨χr , χr ⟩ = ⎨−1 pseudo-real . (2.63)
∣G∣ g∈G ⎪
⎪
⎪ ⎪
⎪
⎪
⎩ 0
⎪ ⎩0
⎪ complex

This formula characterising the different possibilities for representations was first obtained
jointly by Frobenius15 and Schur in 1906 and is sometimes referred to as the Frobenius-Schur
indicator.
Using the orthogonality results then for an arbitrary representation R then decomposing
into irreducible representations as in (2.16)
NG
χR (g) = ∑ mr χr (g) , (2.64)
r=1

and the multiplicity of the irreducible representation Rr is then given by

mr = ⟨χr , χR ⟩ . (2.65)

Applying this to the regular representation which is completely reducible

NG NG
Dreg (g) ≃ ⊕ mreg,r DRr (g) , χreg (g) = ∑ mreg,r χr (g) , (2.66)
r=1 r=1

and using (2.52) and (2.54)

1
mreg,r = ⟨χr , χreg ⟩ = χr (e) χreg (e) = nr . (2.67)
∣G∣
15
Ferdinand Georg Frobenius, 1849-1917, German. A pioneer of representation theory.

37
We also then have
NG NG
∣G∣ = χreg (e) = ∑ mreg,r χr (e) = ∑ nr2 , (2.68)
r=1 r=1

which reproduces (2.62).

To prove (2.59) we define for any representation R
1 ∗
TR = ∑ f (gi ) DR (gi ) . (2.69)
∣G∣ i

Using the group property and f (ggi g −1 ) = f (gi ) it is easy to see that

DR (g)−1 TR DR (g) = TR or TR DR (g) = DR (g) TR for all g . (2.70)

Applying this for the irreducible representation Rr and using Schur’s lemma we must have

TRr = c 1nr , (2.71)

with 1nr the identity matrix in this representation. However

1 ∗
⟨f, χR ⟩ = ∑ f (gi ) χR (gi ) = tr(TR ) = 0 , (2.72)
∣G∣ i

by virtue of the assumption (2.59). Hence in (2.71) c = 0 so that TRr = 0 for all irreducible
representations. Since the regular representation can be decomposed into irreducible rep-
resentations as in (2.66)
TRreg = 0 ⇒ ∑ f (gi )∗ gi = 0 , (2.73)
i

from the definition (2.6) since gi = ∑j gj Dreg,jk (gi )gk−1 . Since all gi are independent this is
only possible if f (gi ) = 0 for all i.
For an induced representation as in (2.25) if for the subgroup representation

χ(h) = tr(D(h)) , (2.74)

then
χinduced rep. (g) = ∑ χ(gi−1 g gi ) ∣g −1 g g ∈H . (2.75)
i i
i

If this is applied to the case when H = {e}, giving the regular representation, we get (2.54).

2.5.1 Further Constraints on Dimensions of Irreducible Representations

The dimensions of irreducible representations, nr , r = 1, . . . , NG , for finite groups are con-

strained by (2.62). A further condition is that nr must divide ∣G∣. This is somewhat non
trivial to prove.
For any representation R and conjugacy class C we may define

TC R = ∑ D(R) (g) . (2.76)

g∈C

38
Since g Cg −1 = C then

D(R) (g) TC R D(R) (g)−1 = TC R ⇒ [TC R , D(R) (g)] = 0 for all g ∈ G . (2.77)

By Schur’s lemma if R is irreducible TC R is proportional to the identity so that

dC
TC R = χR (C) 1nR , (2.78)
nR
where the coefficient is determine by taking the trace. Furthermore summing over the
various conjugacy classes Cs and using the orthogonality of characters

∣G∣2
TCs R TCs̄ R =
∣G∣
RR = ∑ 1n . (2.79)
s ds nR2 R

As a consequence of (1.57)

du
TCs Rr TCt Rr = ∑ cst u TCu Rr ⇒ χr (Cs ) χr (Ct ) = nr ∑ cst u χu (Cu ) , (2.80)
u u ds dt

so that
1 ∣G∣
∑ χr (Cs )χr (Ct )χr (Cū ) = cst u . (2.81)
r nr ds dt

Showing that nr divides ∣G∣ depends on applying these results to the regular representa-
tion. From (2.66) and (2.67) this is expressed as a direct sum over irreducible representations
Dreg (g) ≃ ⊕N G
r=1 nr DRr (g). Then

NG dC
Treg,C = ∑ Dreg (g) ≃ ⊕ nr ( χr (C) 1nR ) , (2.82)
g∈C r=1 nr

and
∣G∣ NG ∣G∣2
Rreg = ∑ Treg,Cs Treg,Cs̄ ≃ ⊕ nr ( 2 1nr ) , (2.83)
s ds r=1 nr
using orthogonality of characters again. The eigenvalues of Rreg are then ∣G∣2 /nr 2 , which
has multiplicity nr and are necessarily rational. The eigenvalues are determined by the
roots of det(λ 1∣G∣ − Rreg ) = 0. Since ∣G∣/ds are integers the elements of Rreg are necessarily
integers as Dreg (g) has elements which are only 1 or 0. Hence det(λ 1∣G∣ − Rreg ) can be
expanded as a polynomial in λ with integer coefficients and leading term λ∣G∣ . The rational
root theorem states that in this situation any root which is rational must an integer which is
a factor of det(Rreg ).16 Hence (∣G∣/nr )2 and therefore, since integer square roots are either
integers or irrational, ∣G∣/nr are integers for all r.
16
For a polynomial P (λ) = λn + an−1 λn−1 + ⋅ ⋅ ⋅ + a0 with ai integers then for a rational root P (p/q) = 0
where p, q are integers with no common factors. The equation determining this root can be rewritten as
q(an−1 pn−1 + ⋅ ⋅ ⋅ + a0 q n−1 ) = −pn . Since q then divides pn and does not contain p as a factor this forces q = 1.

39
2.6 Tensor Products

If V1 , V2 are representation spaces for representations R1 , R2 , given by matrices D1 (g),

D2 (g), for a group G then we may define a tensor product representation R1 × R2 in terms
of the direct product of the representation matrices D(g) = D1 (g) ⊗ D2 (g) acting on the
tensor product space V1 ⊗ V2 where D(g)v = ∑r,s ars D1 (g)v1r D2 (g)v2s . Since dim V =
dim V1 dim V2 the tensor product matrices have dimensions which are the products of the
dimensions of the matrices forming the tensor product. If D1 (g), D2 (g) are unitary then so
is D(g).
In general the tensor product representation R1 × R2 for two representations R1 , R2 is
reducible and may be decomposed into irreducible ones. If the irreducible representations
are listed as {Rr } then in general for the product of any two irreducible representations
Rp × Rq = Rq × Rp = ∑ npq,r Rr , (2.84)
r

where npq,r are integers, which may be zero, and npq,r > 1 if the representation Rr occurs
more than once. For non finite groups there are infinitely many irreducible representations
but the sum in (2.84) is finite for finite dimensional representations. The trace of a tensor
product of matrices is the product of the traces of each individual matrix, in consequence
trVp ⊗Vq (D(Rp ) (g) ⊗ D(Rq ) (g)) = trVp (D(Rp ) (g)) trVq (D(Rq ) (g)), so that, in terms of the
characters χr (g) = trVr (D(Rr ) (g)), (2.84) is equivalent to
χp (g)χq (g) = ∑ npq,r χr (g) . (2.85)
r

Using (2.56) the coefficients npq,r can be determined by

1 ∗ 1
npq,r = ∑ χr (gi ) χp (gi )χq (gi ) = ∑ ds χr (Cs̄ )χp (Cs )χq (Cs ) . (2.86)
∣G∣ i ∣G∣ s

The result (2.84) is exactly equivalent to the decomposition of the associated represen-
tation spaces, with the same expansion for Vp ⊗ Vq into a direct sum of irreducible spaces
Vr . If Rp ⊗ Rq contains the trivial or singlet representation then it is possible to con-
struct a scalar product ⟨v, v ′ ⟩ between vectors v ∈ Vp , v ′ ∈ Vq which is invariant under group
transformations, ⟨D(Rp ) (g)v, D(Rq ) (g)v ′ ⟩ = ⟨v, v ′ ⟩.

2.6.1 Symmetric and Antisymmetric Products

The tensor product of vector spaces Vr , r = 1, . . . , n where Vi , Vj ≃ V for all i, j can be

decomposed into representation spaces for the permutation group Sn . The simplest cases
are the one dimensional totaily symmetric or antisymmetric representations of Sn which
can be denoted by
⋁n V = ( V ⊗ ⋅ ⋅ ⋅ ⊗ V )sym , dim ⋁n V = 1
dim V(dim V + 1) . . . (dim V + n − 1) ,
n!
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n
⋀n V = ( V ⊗ ⋅ ⋅ ⋅ ⊗ V )antisym , dim ⋀n V = 1
dim V(dim V − 1) . . . (dim V − n + 1) . (2.87)
n!
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n

40
⋁n V is also commonly denoted as Symn V . The action of permutations commutes with
the group action on V so that ⋁n V and ⋀n V both form representation spaces, in general
reducible, for the group G with a matrix representation D(g) of dimension dim V. The
representation matrices have the form, for the symmetric, antisymmetric cases respectively,
D(r1 ∣s1 (g)Dr2 ∣s2 (g) . . . Drn )sn (g) , D[r1 ∣s1 (g)Dr2 ∣s2 (g) . . . Drn ]sn (g) , (2.88)
involving a sum over n! permutations σ ∈ Sn .
The corresponding characters
χ∨n (g) = D(r1 ∣r1 (g)Dr2 ∣r2 (g) . . . Drn )rn (g) ,
χ∧n (g) = D[r1 ∣r1 (g)Dr2 ∣r2 (g) . . . Drn ]rn (g) , (2.89)
can be reduced to the characters χ(g) = tr D(g) using the decomposition of any permutation
σ into cycles. Thus for a cycle decomposition σ = [p1 , p2 , . . . , pr ], ∑ri=1 pi = n, then one such
permutation of indices r1 , r2 , . . . , rn generates a contribution of the form
Dσ(r1 )r1 (g)Dσ(r2 )r2 (g) . . . Dσ(rn )rn (g)∣σ(r
1 ,...,rn )=(r1 ...rp1 )(rp1 +1 ...rp1 +p2 )...(rn−pr +1 ...rn )
p1 p2 pr
= χ(g ) χ(g ) . . . χ(g ) . (2.90)
Any permutation with the same cycle decomposition generates an identical expression so
that summing over all possible cycles (2.89) becomes
n n
χ⋁n (g) = 1
n! ∑ ∑ ∑ δn,∑ri=1 ji pi N[p1 (j1 ),...,pr (jr )] ∏ χ(g pi )ji ,
r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1 i=1
n n
χ⋀n (g) = 1
n! ∑ ∑ ∑ δn,∑ri=1 ji pi N[p1 (j1 ),...,pr (jr )] (−1)∑i ji (pi −1) ∏ χ(g pi )ji ,
r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1 i=1
(2.91)
with the numbers for ji pi -cycles in each cycle decomposition given by (1.22). Results for
χ⋁2 (g), χ⋁3 (g) are just as in (1.70) and (1.71) is equivalent to the result for dim ⋁n V in
(2.87). Similarly χ⋀2 (g), χ⋀3 (g) are essentially of the form in (1.73) so that
χ 2 n (g)
⋁ V
χ 2 n (g) } = 21 (χV n (g)2 ± χV n (g 2 )) ,
⋀ V

χ 3 n (g)
⋁ V
χ 2 n (g) } = 61 (χV n (g)3 ± 3 χV n (g)χV n (g 2 ) + 2 χV n (g 3 )) . (2.92)
⋀ V

2.7 Character Tables

Knowing the irreducible representations the corresponding characters can readily be com-
puted and can be represented as tables composed of the elements of the square matrix
X.
For abelian groups the irreducible representations are one dimensional as are also the
characters which each correspond to a single group element. For Zn , or equivalently Cn ,
defined by {ar ∶ an = e}, we can take
Xkr = χk (ar ) = e2krπi/n , k, r = 0, 1, . . . , n − 1 , (2.93)

41
where k labels the representation and r the conjugacy class. It is easy to verify that
XX † = N 1N . The character table is just
Zn C1,r FF S
, k, r = 0, . . . , n − 1 . (2.94)
R1,k e2krπi/n 1, k = 0, 21 n(n even), 0 otherwise

For the dihedral group Dn the conjugacy classes are listed in (1.58) and the irreducible
representations in (2.30) and (2.31) with (2.29). The character tables are then, for n odd,
Dn n odd C1 C2,r Cn FF S
R1,1 1 1 1 1
, k, r = 1, . . . , 12 (n − 1) , (2.95)
R1,2 1 1 −1 1
R2,k 2 2 cos 2krπ/n 0 1
where χ2,0 (C) = χ1,1 (C) + χ1,2 (C). For n even,
Dn n even C1,1 C1,2 C2,r C 1 n,1 C 1 n,2 FF S
2 2

R1,1 1 1 1 1 1 1
R1,2 1 1 1 −1 −1 1
, k, r = 1, . . . , 21 (n − 2) .
R1,3 1 (−1)n/2 (−1)r 1 −1 1
R1,4 1 (−1)n/2 (−1)r −1 1 1
k
R2,k 2 2(−1) 2 cos 2krπ/n 0 0 1
(2.96)
In this case χ2,0 (C) = χ1,1 (C)+χ1,2 (C) and χ2,n/2 (C) = χ1,3 (C)+χ1,4 (C). The representations
are all real.
For the dicyclic groups the conjugacy classes are given in (1.87) and irreducible repre-
sentations in (2.35) and (2.36) with (2.33) and (2.34). In this case
Q4n n odd C1,1 C1,2 C2,r Cn,1 Cn,2 FF S
R1,1 1 1 1 1 1 1
R1,2 1 1 1 −1 −1 1
, k, r = 1, . . . , n − 1 , (2.97)
R1,3 1 −1 (−1)r i −i 0
R1,4 1 −1 (−1)r −i i 0
R2,k 2 2(−1)k 2 cos krπ/n 0 0 (−1)k
and
Q4n n even C1,1 C1,2 C2,r Cn,1 Cn,2 FF S
R1,1 1 1 1 1 1 1
R1,2 1 1 1 −1 −1 1
, k, r = 1, . . . , n−1 . (2.98)
R1,3 1 1 (−1)r 1 −1 1
R1,4 1 1 (−1)r −1 1 1
k
R2,k 2 2(−1) 2 cos krπ/n 0 0 (−1)k

42
For both n even and odd χ2,0 (C) = χ1,1 (C) + χ1,2 (C), χ2,n (C) = χ1,3 (C) + χ1,4 (C). The
representations R2,k for n ≥ 2 and k odd are pseudo real. For n = 1 dropping the C2,r
column and R2,k row the character table is identical to that for Z4 .
These character tables satisfy the required orthogonality conditions.

2.8 Molien Series

A nice application of the results for characters is a formula due to Molien.17 Suppose a
group G acts on a representation space V n , of dimension n, so that for x ∈ V n and g ∈ G
there is a linear action x → g x. In terms of coordinates xs , s = 1, . . . , n this becomes a n × n
dimensional representation of G given by
(g x)s = ∑t xt Dts (g) . (2.99)
An important question is then to determine possible G invariant homogeneous polynomials
of degree p, P (λx) = λp P (x), P (g x) = P (x). Let mp be the number of such invariant
polynomials. Then there is a generating function
1
MG (V n , t) = ∑ mp tp = ∑ 1/ det (1 − t D(g)) , (2.100)
p≥0 ∣G∣ g∈G
which is the Molien series.
The determinant in (2.100) can be expanded
tm
1/ det (1 − t D(g)) = exp ( − tr ln(1 − t D(g)) = exp ( ∑ trD(g)m )
m≥1 m
tm
= exp ( ∑ χV n (g m )) = PE(t, g; χV n ) = 1 + χV n (g) + ∑ un χ⋁m V n (g) , (2.101)
m≥1 m m≥2

where the plethystic exponential is defined in (1.69) and χ⋁m V n (g) = ⋁m χV n (g) is the
character for the representation obtained from {D(g)} acting on the m-fold symmetric
tensor product space ⋁m V n as given in (2.91). As a result (2.100) can be expressed as
1
MG (V n , t) = ∑ PE(t, g; χV n ) . (2.102)
∣G∣ g∈G

To verify this result (2.100) we consider all possible homogeneous polynomials of degree
p. There are
⎧
1 Γ(n + p) ⎪ ⎪1 , p=0
Np = (n)p , (n)p = =⎨ , (2.103)
p! Γ(n)
⎩n(n + 1) . . . (n + p − 1) , p ≥ 1
⎪
⎪
such polynomials (here (n)p is the Pochhammer symbol). A basis of degree p polynomials
{Pα (x) ∶ α = 1, . . . , Np } defines a Np -dimensional representation of G
p N
Pα (g x) = ∑β=1 Pβ (x) Dβα (g) . (2.104)
17
Theodor Georg Andreas Molien, 1861-1941, Baltic German, Russian nationality.

43
The representation {D(g)} is formed from a p-fold symmetrised tensor product of the
representation {D(g)} and we must have

Eigenvalues D(g) = {λa (g) ∶ a = 1, . . . , n}

⇒ Eigenvalues D(g) = {∏na=1 λa (g)da ∶ da ≥ 0, ∑na=1 da = p} . (2.105)

There are Np possible choices for {da } so that the number of eigenvalues is equal to the
dimension of the representation. For a particular D(g) then the corresponding character is
obtained by summing over all the eigenvalues

χD (g) = tr(D(g)) = ∏na=1 ∑da ≥0 λa (g)da δ∑a da , p . (2.106)

The representation {D(g)} can be decomposed by using characters

nG
1 ∗
χD (g) = ∑ as,p χs (g) , as,p = ∑ χs (g) χD (g) . (2.107)
i=1 ∣G∣ g∈G

Hence now
n
p 1 ∗ da
∑ as,p t = ∑ χs (g) ∏ ∑ ∑ (t λa (g)) δ∑a da , p
p≥0 ∣G∣ g∈G a=1 p≥0 da ≥0
n n
1 ∗ da 1 ∗ −1
= ∑ χs (g) ∏ ∑ (t λa (g)) = ∑ χs (g) ∏ (1 − t λa (g))
∣G∣ g∈G a=1 da ≥0 ∣G∣ g∈G a=1
1 ∗
= ∑ χs (g) 1/ det (1 − t D(g)) . (2.108)
∣G∣ g∈G

The number of invariant polynomials is equal to the number of singlet representations

contained in the decomposition of {D(g)} so that mp = a1,p . For the trivial singlet repre-
sentation χ1 (g) = 1. In this case (2.108) reduces to (2.100).

2.8.1 Anticommuting Molien Series

From a physics perspective it is also interesting to consider invariants under the group action
of G on a n-dimensional Grassmanian manifold Mn which is defined in terms of anticom-
muting coordinates θ = (θ1 , θ2 , . . . , θn ) with θs θt = −θt θs . The group action is identical to
(2.99). A basis of degree p polynomials in θ, P˜α (θ) satisfying (2.104) with x → θ may also
be constructed but now Np = (np) and we most have p ≤ n, higher degree polynomials vanish.
If va (θ), linear in θ, is an eigenvector of D(g) with eigenvalues λa , a = 1, . . . , n then the
eigenvectors of D(g) are just ∏na=1 va (θ)da with da = 0, 1, since va (θ)2 = 0, and ∑a da = p.
Hence
Eigenvalues D(g) = {∏na=1 λa (g)da ∶ da = 0, 1, ∑na=1 da = p} , (2.109)
and instead of (2.106)

χD (g) = tr(D(g)) = ∏na=1 ∑1da =0 λa (g)da δ∑a da , p . (2.110)

44
The decomposition of χD (g) into irreducible representations remains as in (2.107) and
n 1
p 1 ∗ da
∑ ãs,p t = ∑ χs (g) ∏ ∑ ∑ (t λa (g)) δ∑a da , p
p≥0 ∣G∣ g∈G a=1 p≥0 da =0
n 1 n
1 ∗ da 1 ∗
= ∑ χs (g) ∏ ∑ (t λa (g)) = ∑ χs (g) ∏ (1 + t λa (g))
∣G∣ g∈G a=1 da =0 ∣G∣ g∈G a=1
1 ∗
= ∑ χs (g) det (1 + t D(g)) . (2.111)
∣G∣ g∈G

The number of invariant polynomials of degree p is m̃p = a1,p and the generating function
becomes now
1
M̃G (Mn , t) = ∑ m̃p tp = ∑ det (1 + t D(g)) , (2.112)
p≥0 ∣G∣ g∈G
The first order contribution in and expansion in t, determining the number of linear invari-
ants, is the same as in (2.100) and is given by ∑g χ(g) which will be zero for an irreducible
representation. In general (2.112) is a polynomial of degree n.
In an analogous fashion to (2.101)

tm
det (1 + t D(g)) = exp ( ∑ (−1)m−1 χ n (g m ))
m≥1 m V
= PEf (t, g; χV n ) = 1 + χV n (g) + ∑ un χ⋀m V n (g) , (2.113)
m≥2

with χ⋀m V n (g) = ⋀m χV n (g) the character for the representation acting on the m-fold anti-
symmetric tensor product space (V n ⊗ ⋅ ⋅ ⋅ ⊗ V n )antisym. .

2.8.2 Examples of Molien Series

As a first example we consider Zn which is generated by an = e. Acting on Rn there is a cor-

responding action given by cyclic permutations of the coordinates (x1 , x2 , . . . , xn−1 , xn ) →
(x2 , x3 , . . . , xn , x1 ). This is just an n-cycle in the group of all permutations Sn . The asso-
ciated n × n representation matrix An , where Ann = 1n , is then of the form
0 0 0 ⋯ 0 1
⎛1 0 0 ⋯ 0 0 ⎞
0 1 0 ⋯ 0 0
An = ⎜
⎜0 0 1 ⋯ 0 0
⎟.
⎟ (2.114)
⎝ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮ ⎠
0 0 0 ⋯ 1 0

In general

det (1 − tAnr ) = (1 − tm )n/m , for m, m∣n, the smallest integer such that r m = 0 mod n .
(2.115)
For r = 1, m = n this can be worked out directly. Other cases can be found by generalising
results such as A42 ≃ A2 ⊕ A2 A62 ≃ A3 ⊕ A3 , A63 ≃ A2 ⊕ A2 ⊕ A2 together with det(A ⊕ B) =
det A det B.

45
 A special case of the Molien formula for this case is

1 1 n−1
MZn (Rn , t) = ( + ), n prime . (2.116)
n (1 − t)n 1 − tn

This representation is reducible. Restricting to just the one dimensional irreducible repre-
sentation
1 n−1 1 1
MZn (C, t) = ∑r=0 2rπ i
= . (2.117)
n 1 − exp( n ) t 1 − tn
2qrπ i
The summation can be calculated by expanding in t and using ∑n−1 r=0 exp( n ) = n δq,pn for
rn
p = 0, 1, 2, . . . . For z ∈ C the invariants are just z , n = 1, 2, . . . which are just products of
a single fundamental invariant z n . For n = 2 we may restrict C to R.
Other general results are obtained by considering the two dimensional representation of
n
cos 2π
− sin 2π
n
the cyclic group Cn generated by the real 2×2 rotation matrix A = ( sin 2π cos 2π
) satisfying
n n
An = 12 . Each element has its own conjugacy class and the Molien formula gives18
1 n−1 1 1 + tn
MCn (R2 , t) = ∑ = . (2.118)
n r=0 1 − 2 cos 2rπ
n t+t
2 (1 − t2 )(1 − tn )

For (x, y) ∈ R2 expanding (2.118) in powers of t then gives the number of invariants under
the action of Cn on x, y. These can all be expressed in terms of sums and products of a
finite number of fundamental invariants. There are two primary invariants p1 = x2 +y 2 , p2 =
Re(x + i y)n which correspond to the factors 1 − t2 and 1 − tn in the denominator. There is
also a secondary invariant q = Im(x + i y)n . For n = 2, p2 = x2 − y 2 and q = 2xy. Both p2
and q are invariant under x + i y → e2π/n i (x + i y) but q 2 = p1n − p22 . Of course the role of p2
and q can be interchanged. Any invariant is then obtained by sums and products of p1 , p2
together also with terms linear in q.
For the dihedral group, with the conjugacy classes in (1.58) and the representation given
by (2.29) or (2.32) for k = 1, the Molien formula becomes
⎧ 1
⎪ 1 1 1 n 2
n−1 1
⎪
⎪ 2n ( (1−t)2 +
⎪ (1+t)2
+
1−t2
+ 2 ∑r=1 1−2 cos 2rπ t+t2
), n even ,
2
MDn (R , t) = ⎨ 1
(n−1)
n (2.119)
1
⎪
⎪
⎪ ( 1 2+ n 2
+ 2 ∑r=1 1
), n odd .
⎩ 2n (1−t)
⎪ 1−t2 1−2 cos 2rπ t+t2
n

In either case19
1
MDn (R2 , t) = . (2.120)
(1 − t2 )(1 − tn )
18
Expanding in t the sum becomes n1 ∑n−1
r=0 ∑p,q≥0 t
p+q
exp( 2(p−q)rπ
n
i
) = ∑∞
s=−∞ ∑q≥0,−ns t
2q+ns
which is read-
ily evaluated.
19
For n = 2m the basic sum can be reduced to
m m (p−q)rπ i (p−q)rπ i
= ∑ t ∑r=0 ( exp( m ) + exp(− m ))
p+q
2m ∑r=0 1−2 cos rπ
1 1 1
m
t+t2 4m
p,q≥0 ∞
= ∑ t (δp,q + (−1) ) +
p+q p−q
1
4m
1
2 ∑ ∑ t2q+2ms ,
p,q≥0 s=−∞ q≥0,−2ms

which is then straightforward.

46
In the real basis (2.32) and, for (x, y) ∈ R2 , p1 , p2 as in the Z2 case are still invariants but
q is no longer since it changes sign under y → −y.
A further illustrative example is given by the dicyclic, or binary dihedral, group Q4n .
The conjugacy classes are given in (1.87) and a two dimensional irreducible complex repre-
sentation is given in (2.33) and (2.34) setting k = 1. For this representation det(1 − t Ar ) =
1 − 2 cos rπ 2 r 2
n t + t , det(1 − t A B) = 1 + t , and

1 1 1 2n n−1 1
MQ4n (C2 , t) = ( + + + 2 ∑r=1 )
4n (1 − t) 2 (1 + t) 2 1+t 2 1 − 2 cos rπ
n t+t
2

1 + t2n+2
= . (2.121)
(1 − t4 )(1 − t2n )

There are two primary invariants, for (x, y) ∈ C2 these are p1 = x2 y 2 , p2 = x2n + y 2n , and
also q = xy(x2n − y 2n ) is a secondary invariant since q 2 = p1 p22 − 4 p2n+1 . All invariants are
formed from sums of products of p1 , p2 together with similar expressions linear in q.
There are more invariants when considering the symmetric group Sn acting on the Rn
by permuting the coordinates (x1 , x2 , . . . , xn ) → (xσ(1) , xσ(2) , . . . , xσ(n) ), σ ∈ Sn . The sum
in (2.100) can be reduced to a sum over conjugacy classes which are given by products of
pi -cycles and with the results in subsection 1.4.3 for the numbers of group elements in each
conjugacy class together with using, from (2.115) for n = pi , r = 1, that in any conjugacy
class containing σ ∈ Sn each pi -cycle contributes a factor 1 − tpi to det[1n − t D(g)] so that

det[1n − t D(g)] = ∏ri=1 (1 − tpi ) , g ∈ C[p1 ,...,pr ] , ∑ri=1 pi = n . (2.122)

The general formula (2.100) then becomes, with a sum over conjugacy classes as in (1.23),

1 n 1
MSn (Rn , t) = ∑ ∑ ∑ δn,∑ri=1 ji pi N[p1 (j1 ),...,pr (jr )] r . (2.123)
n! r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1 ∏i=1 (1 − tpi )ji

Applying this for n = 3, 4, 5 with the results given in (1.63), (1.64), (1.65)

MS3 (R3 , t) = 16 ( 1−t

2
3 +
3
(1−t2 )(1−t)
1
+ (1−t)3),

MS4 (R4 , t) = 1
( 6 8 3 6 1
24 1−t4 + (1−t3 )(1−t) + (1−t2 )2 + (1−t2 )(1−t)2 + (1−t)4 ,
)
MS5 (R5 , t) = 1
( 24 30 20 20
120 1−t5 + (1−t4 )(1−t) + (1−t3 )(1−t2 ) + (1−t3 )(1−t)2
+ (1−t215
)2 (1−t)
+ (1−t210
)(1−t)3
1
+ (1−t) 5). (2.124)

The generating function obtained from (2.123) using (1.69) is expressible as a plethystic
exponential
∞
n n 1
∑ u MSn (R , t) = PE(u, t; f ) , f (t) = . (2.125)
n=0 1−t
By expanding in t for this f (t), ∑∞
m=1
um m
m f (t ) = − ∑∞ r
r=0 ln(1 − u t ), and hence

∞ ∞ ∞ ∞
1
∑ t∑r≥1 r ∣
n n n s r
∑ u MSn (R , t) = ∏ r
= ∑ u ∏ . (2.126)
n=0 r=0 1 − u t n=0 r=1 sr ≥0 ∑r≥1 sr ≤n

47
In each case the results are in accord with the general formula
n
1
MSn (Rn , t) = ∏ . (2.127)
r=1 1 − tr
This expression can be interpreted as showing that for every p ≤ n there is a new homo-
geneous symmetric polynomial of degree p, other such polynomials can be represented as
sums, with rational coefficients, of products of symmetric polynomials of lower degree. For
instance we may consider pp (x) = ∑na=1 xap , p = 1, . . . , n as a set of primary polynomial
invariants. For n = 2 (2.123) coincides with (2.116).
For the anticommuting case
n
M̃Sn (Mn , t) = ∑ tr . (2.128)
r=0

so there one invariant for any p ≤ n.

The n-dimensional representation of Sn is reducible since setting all xa equal defines
an invariant subspace. An irreducible (n − 1)-dimensional representation is obtained by
imposing the linear condition ∑a xa = 0, which of course is invariant under Sn . In this case
n
1
MSn (Rn−1 , t) = ∏ . (2.129)
r=2 1 − tr
The lack of the linear 1 − t factor reflects the absence of the ∑a xa invariant. (2.129) is
identical with (2.120) for n = 3.
For the alternating group An conjugacy classes involving odd numbers of 2-cycles are
removed. Following the results in 1.4.3 for n = 3, 4, 5 the general expression is
n
1 1
MAn (Rn , t) = (1 + t 2 n(n−1) )∏ . (2.130)
r=1 1 − tr
With less symmetry there is an extra secondary invariant q(x) = ∏1≤b<a≤n (xa −xb ), of degree
1
2 n(n − 1), as well as the n primary invariants which are present for Sn . qn (x) changes sign
under odd permutations in Sn . However qn (x)2 is invariant under all Sn permutations and
can be expressed in terms of {pp (x)}. This explains the presence of the single factor in the
numerator in (2.130). There is a similar reduction as in going from (2.123) to (2.129) on
restricting to an irreducible representation on Rn−1 .

2.8.3 Molien Series and Wreath Products

The wreath product, whose action is defined in (1.51), plays an important role when groups
act on spaces with symmetry conditions imposed. The simplest illustration is Z2 ≀ Z2 where
Z2 × Z2 acts on (x, y) ∈ R2 by reflecting x and y and there is also a Z2 action obtained
by interchanging x and y. On this two dimensional space the representation of Z2 ≀ Z2 is
obtained by taking
d1 = ( 10 01 ) , d2 = ( 10 −1
0 ) , d3 = ( −1 0
0 1) , d4 = ( −1 0
0 −1 ) ,
d5 = ( 01 10 ) , d6 = ( 01 −1
0 ) ,
0 1) ,
d7 = ( −1 0
0 −1 ) ,
d8 = ( −1 0 (2.131)

48
with d1 , d2 , d3 , d4 representing Z2 × Z2 . Applying (2.100)

1 8 1 1 1 1 4 2
MZ2 ≀Z2 (R2 , t) = ∑ = ( + + + )
8 i=1 det(1 − t di ) 8 (1 − t)2 (1 + t)2 1 − t2 1 + t2
1
= . (2.132)
(1 − t )(1 − t4 )
2

The fundamental invariants are just x2 + y 2 and x4 + y 4 .

More generally for G ≀ Sn the Molien formula allows for significant simplifications. For
a representation d(g) of G acting on Rk then (2.100) gives
1 1
MG≀Sn (Rnk , t) = n ∑ ∑ , (2.133)
∣G∣ n! σ∈Sn g∈Gn× det (1 − t Dσ (gσ ))

where Dσ is the representation of G≀Sn acting on Rnk formed from the n-fold tensor product
of d(g). This representation depends on the decomposition of σ into non overlapping cycles,

σ = σp1 . . . σpr ∈ C[p1 ,...,pr ] ⇒ gσ = g(p1 ) ∪ . . . g(pr ) , g(p) = (g(p)1 , . . . , g(p)p ) (2.134)

with σp , σp p = e, generating cyclic permutations of the group elements in g(p) . In this case
Dσ (gσ ) ≃ Dp1 (g(p1 ) ) ⊕ ⋅ ⋅ ⋅ ⊕ Dpr (g(pr ) ) with
⋯
⎛ d02 0
0
0
0 ⋯
0
0
d1
0
⎞
Dp (g(p) ) = ⎜
⎜ 0⋮ d3 0 ⋯ 0 0 ⎟ , p ≥ 2 , D1 (g(1) ) = d1 ,
⎟ di = d(g(p)i ) . (2.135)
⋮ ⋮ ⋱ ⋮ ⋮
⎝ 0 0 0 ⋯ dp 0 ⎠

Since
1 0 0 ⋯ 0 −t d1 1k 0 0 ⋯ 0 0 1k −tp d1 dp ...d2 0 0 ⋯ 0 −t d1
⎛ −t kd2 1k 0 ⋯ 0 0 ⎞ ⎛ t d2 1k 0 ⋯ 0 0 ⎞ ⎛ 0 1k 0 ⋯ 0 0
⎞
⎜ 0
⎜ −t d3 1k ⋯ 0 0 ⎟ ⎜ t2 d3 d2
⎟⎜ 0 1k ⋯ 0 0 ⎟=⎜
⎟ ⎜ 0 −t d3 1k ⋯ 0 0 ⎟,
⎟
⋮ ⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮
⎝ 0 0 0 ⋯ −t dp 1k ⎠ ⎝ tp−1 d ...d 0 0 ⋯ 0 1k
⎠ ⎝ 0 0 0 ⋯ −t dp 1k ⎠
p 2
(2.136)
evaluating the determinant becomes straightforward giving

det (1pk − t Dp (g(p) )) = det (1k − tp dp . . . d1 ) = det (1k − tp d(g(p)p . . . g(p)1 )) . (2.137)

Using
p−1
∑(g1 ,...,gp )∈Gp× f (gp . . . g1 ) = ∣G∣ ∑g∈G f (g) , (2.138)
the Molien sum in (2.133) can be reduced to a sum over conjugacy classes of Sn

1 n r
1 1
MG≀Sn (Rnk , t) = ∑ ∑ ∑ δn,∑ri=1 ji pi ∏ ∑
n! r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1 pi
i=1 ∣G∣ g∈G det (1k − t d(g))
1 n r
k p
= ∑ ∑ ∑ δn,∑ri=1 ji pi ∏ MG (R , t i ) . (2.139)
n! r=1 j1 ,j2 ,...jr ≥1 p1 >p2 >...pr ≥1 i=1

From (1.69)
n nk k
∑ u MG≀Sn (R , t) = PE(u, t; MG (R )) . (2.140)
n≥0

49
 Various special cases are easily obtained. From (2.116) and (2.123), (2.123)
n
1
MCm ≀Sn (Cn , t) = MSn (Cn , tm ) = ∏ . (2.141)
r=1 1 − tmr

For m = 2, taking Cn → Rn , this is just the Molien series for the symmetry group of the
n-dimensional hypercube. For n = 2 this reduces to (2.132). Using (2.120) with N[2] =
N[1,1] = 1,

1 2 1 + tm+2
MDm ≀Z2 (R4 , t) = (MDm (R2 , t) + MDm (R2 , t2 )) = . (2.142)
2 (1 − t2 )(1 − t4 )(1 − tm )(1 − t2m )
It is an exercise to determine the primary and secondary invariants in this case.

2.9 Symmetries in Quantum Mechanics, Projective and Anti-Unitary

Representations

A symmetry of a physical system is defined as a set of transformation acting on the system

such that the physical observables are invariant. In quantum mechanics the state of a
particular physical system is represented by a vector ∣ψ⟩ belonging to a vector (or Hilbert)
space H. The essential observables are then the probabilities, given that the system is
in a state ∣ψ⟩, of finding, under some appropriate measurement, the system in a state
∣φ⟩. Assuming ∣ψ⟩, ∣φ⟩ are both normalised this probability is ∣⟨φ∣ψ⟩∣2 . For a symmetry
transformation ∣ψ⟩ → ∣ψ ′ ⟩ we must require

∣⟨φ∣ψ⟩∣2 = ∣⟨φ′ ∣ψ ′ ⟩∣2 for all ∣ψ⟩, ∣φ⟩ ∈ H . (2.143)

Any quantum state vector is arbitrary up to a complex phase ∣ψ⟩ ∼ eiα ∣ψ⟩. Making use of
this potential freedom Wigner20 proved that there is an operator U such that

U ∣ψ⟩ = ∣ψ ′ ⟩ , (2.144)

and either

⟨φ′ ∣ψ ′ ⟩ = ⟨φ∣U † U ∣ψ⟩ = ⟨φ∣ψ⟩ , U (a1 ∣ψ1 ⟩ + a2 ∣ψ2 ⟩) = a1 U ∣ψ1 ⟩ + a2 U ∣ψ2 ⟩ , (2.145)

so that U is unitary linear, or

⟨φ′ ∣ψ ′ ⟩ = ⟨φ∣U † U ∣ψ⟩ = ⟨φ∣ψ⟩∗ , U (a1 ∣ψ1 ⟩ + a2 ∣ψ2 ⟩) = a1∗ U ∣ψ1 ⟩ + a2∗ U ∣ψ2 ⟩ , (2.146)

and U is unitary anti-linear. Mostly the anti-linear case is not relevant, if U is continuously
connected to the identity it must be linear.
For the discrete symmetry linked to time reversal t → −t the associated operator T must
∂
be anti-linear, in order for the Schrödinger equation i ∂t ∣ψ⟩ = H∣ψ⟩ to be invariant when
T HT = H (we must exclude the alternative possibility T HT −1 = −H since energies should
−1

be positive or bounded below). This requirement also apparent since if x, p are the position,
20
Eugene Paul Wigner, 1902-1995, Hungarian until 1937, then American. Nobel Prize 1962.

50
momentum operators then the action of time reversal requires T xT −1 = x, T pT −1 = −p and
for this to be compatible with the fundamental commutation relation [x, p] = i 1 it is
necessary that T is anti-linear.
In the simplest case if G is a symmetry for a physical system with a Hamiltonian H we
must require
U [g]HU [g]−1 = H for all g ∈ G . (2.147)
If H has energy levels with degeneracy so that

H∣ψr ⟩ = E∣ψr ⟩ , r = 1, . . . , n , (2.148)

then it is easy to see that

H U [g]∣ψr ⟩ = E U [g]∣ψr ⟩ . (2.149)
For a symmetry group G = {g} there are then unitary operators U [g] where we require
U [e] = 1, U [g −1 ] = U [g]−1 . If the unitary operators satisfy the usual group multiplication
rules U (gi )U (gj ) = U (gi gj ) then
n
U [g]∣ψr ⟩ = ∑ ∣ψs ⟩Dsr (g) , (2.150)
s=1

and furthermore the matrices [Dsr (g)] form a n-dimensional representation of G. If {∣ψr ⟩}
are orthonormal, ⟨ψr ∣ψs ⟩ = δrs , then the matrices are unitary. The representation need not
be irreducible but, unless there are additional symmetries not taken into account or there is
some accidental special choice for the parameters in H, in realistic physical examples only
irreducible representations are relevant.

2.9.1 Projective Representations

However in quantum mechanics, because of the freedom of complex phases, we may relax
the product rule and require only

U [gi ]U [gj ] = eiγ(gi ,gj ) U [gi gj ] . (2.151)

If the phase factor eiγ is present this gives rise to a projective representation. However the
associativity condition (1.4) ensures γ(gi , gj ) must satisfy consistency conditions,

γ(gi , gj gk ) + γ(gj , gk ) = γ(gi gj , gk ) + γ(gi , gj ) . (2.152)

There are always solutions to (2.152) of the form

γ(gi , gj ) = α(gi gj ) − α(gi ) − α(gj ) , (2.153)

for any arbitrary α(g) depending on g ∈ G. However such solutions are trivial since in this
case we may let eiα(g) U [g] → U [g] to remove the phase factor in (2.151). For most groups
there are no non trivial solutions for γ(gi , gj ) so the extra freedom allowed by (2.151) may
be neglected so there is no need to consider projective representations, although there are
some cases when it is essential.

51
 As an extension of the above we may consider wave functions Ψ(X) depending on
variables X ∈ M on which there is a group action X → gX for g ∈ G. There is then an
induced action on Ψ given by
Ψ(X) → Ψg (X) = eiφg (X) Ψ(g −1 X) , (2.154)
where we allow for a phase φg depending on g. Group multiplication requires
−1 X)
eiφg1 g2 (X) = eiφg1 (X) eiφg2 (g1 . (2.155)
A trivial solution is obtained if
−1 X))
eiφg (X) = ei(χ(X)−χ(g , (2.156)
since then we may redefine Ψ(X) → eiχ(X) Ψ(X) and eliminate the phase from the trans-
formation (2.154).
Phases satisfying (2.152) or (2.155) modulo the trivial solutions (2.153) or (2.156) cor-
respond to cohomology classes H 2 (G, U (1)) or H 1 (G, U (1)) of the group G.21
The complex phase in (2.151) may be restricted to a representation of a discrete sub-
group of U (1). Thus we may have U [gi ]U [gj ] = A(gi , gj ) U [gi gj ] with A(gi , gj ) ∈ Cn
and A(gi , gj gk ) A(gj , gk ) = A(gi gj , gk ) A(gi , gj ) with the trivial arbitrariness, correspond-
ing to redefinitions of U (gi ), A(gi , gj ) ∼ A(gi , gj )B(gi gj )B(gi )−1 B(gj )−1 for B(g) ∈ Cn .
The existence of non trivial A(gi , gj ) corresponds to H 2 (G, Zn ). Assuming U (e) = 1 then
A(e, gi ) = A(gi , e) = 1 for all gi .
As an illustration we may consider the group D2 ≃ Z2 × Z2 which has four elements
{e, a, b, ab} with a2 = b2 = e, ab = ba. There is a two dimensional projective representation,
corresponding to non trivial H 2 (D2 , Z2 ), formed by taking D(a) = I, D(b) = J, D(ab) = K
in terms of the quaternion representation matrices in (1.115). For this example A(gi , gj ) is
given by the table
A e a b ab
e 1 1 1 1
a 1 −1 1 −1 . (2.157)
b 1 −1 −1 1
ab 1 1 −1 −1

2.9.2 Anti-Unitary Representations

Anti-unitary representations are possible for a group G = G0 ∪ G1 where G0 is a normal

subgroup and for any h ∈ G0 , a, a′ ∈ G1 then ha, ah ∈ G1 , aa′ ∈ G0 . This requires dim G0 =
21
For real functions of n group elements gi ∈ G, ϕn (g1 , . . . , gn ) ∈ C n , we may define d ∶ C n → C n+1 by
(dϕn )(g1 , . . . , gn+1 ) = ϕn (g2 , . . . , gn+1 ) − ϕn (g1 g2 , g3 , . . . , gn+1 ) + ϕn (g1 , g2 g3 , . . . , gn+1 )
− ⋅ ⋅ ⋅ + (−1)n ϕn (g1 , . . . , gn gn+1 ) + (−1)n+1 ϕn (g1 , . . . , gn ).

Then d2 = 0. Define Z n = ker d∩C n , or {ϕn ∶ dϕn = 0}, and the cohomology class H n (G, R) is defined by H n =
Z n /dC n−1 . The elements of H n (G, U (1)) are given by ei ϕn (g1 ,...,gn ) with ϕn (g1 , . . . , gn ) ∼ ϕn (g1 , . . . , gn )+2π.

52
dim G1 . Acting on states {∣ψr ⟩} then U (a) for a ∈ G1 can act anti-linearly so that
n n n n
U [h] ∑ ∣ψr ⟩ar = ∑ ∣ψs ⟩Dsr (h)ar , U [a] ∑ ∣ψr ⟩ar = ∑ ∣ψs ⟩Dsr (a)ar∗ , (2.158)
r=1 r,s=1 r=1 r,s=1

with both D(h), D(a) unitary matrices. Hence

D(hg) = D(h)D(g) , D(ag) = D(a)D(g)∗ for all h ∈ G0 , a ∈ G 1 , g ∈ G , (2.159)

so that D(a2 ) = D(a)D(a)∗ , and

D(a−1 ) = D(a)−1∗ = D(a)T . (2.160)

As was first described by Wigner this defines a co-representation extending the standard
representation of G0 . With these results
∗
D′ (h) = D(a−1 ha) = (D(a)−1 D(h)D(a)) , (2.161)

also forms a representation of G0 . For a co-representation the matrix similarity transform

(2.5) becomes
D(h) → SD(h)S −1 , D(a) → SD(a)S −1∗ . (2.162)
The matrices remain unitary if S is unitary. For S = α 1, ∣α∣ = 1, then D(a) ∼ α2 D(a). The
co-representation is reducible if under the transformation (2.162) the matrices D(h), D(a)
can be made block diagonal for all h, a.
The different possibilities for anti-unitary representations can be obtained by assuming
a decomposition

M (h) 0 0 N (a)
D(h) = ( ), D(a) = ( ), (2.163)
0 M̄ (h) N̄ (a) 0

where M (h), M̄ (h), N (a), N̄ (a) are all unitary matrices with

N̄ (a−1 ) = N (a)T , N (a)N̄ (a)∗ = M (a2 ) , N̄ (a)N (a)∗ = M̄ (a2 ) (2.164)

Necessarily {M (h)} and {M̄ (h)} define representations R0 and R ¯ 0 of G0 which have equal
dimension d0 , and the co-representation R of G defined by (2.163) then has dimension 2d0 .
From (2.161)
∗ ∗
M (a−1 ha) = (N̄ (a)−1 M̄ (h)N̄ (a)) , M̄ (a−1 ha) = (N (a)−1 M (h)N (a)) . (2.165)

Representations with M ↔ M̄ and N ↔ N̄ are equivalent. To obtain an irreducible repre-

¯ 0 are irreducible.
sentation for G we require that the representations R0 , R
For the decomposition (2.163) a special case arises when R0 and R ¯ 0 are equivalent
¯ 0 , and, by a choice of basis, we may take M (h) = M̄ (h) for all h.
representations, R0 ≃ R
Then (2.165) requires

N̄ (a)−1 M (h)N̄ (a) = N (a)−1 M (h)N (a) ⇒ [M (h), N (a)N̄ (a)−1 ] = 0 for all h . (2.166)

53
Assuming R0 is an irreducible representation then by Schur’s lemma N̄ (a)N (a)−1 ∝ 1 and
hence
N̄ (a) = α N (a) ⇒ N̄ (a)−1 = α∗ N (a)−1 ⇒ α∗ α = 1 , (2.167)
since N (a), N̄ (a) are unitary matrices. Substituting in (2.164) with M = M̄

α N̄ (a)N (a)∗ = M (a2 ) , α = α∗ . (2.168)

Hence α2 = 1 and there are two possibilities.

I∶ N̄ (a) = N (a) , D(h) = M (h)12 , D(a) = N (a) ( 01 10 ) ,

II ∶ N̄ (a) = −N (a) , 0 1) .
D(h) = M (h)12 , D(a) = N (a) ( −1 (2.169)
0

The remaining case arises when

¯0 ,
III. R0 ≄ R ¯ 0 ≃ R0∗ ,
(i) R ¯ 0 ≄ R0∗ .
(ii) R (2.170)

The first case is reducible since we may easily construct a real S such that S ( 01 10 ) S −1 =
( 10 −1
0 ) so that

{D(h), D(a)} → {M (h), N (a)} ⊕ {M (h), −N (a)} . (2.171)

The two anti-unitary representations of dimension d0 in this decomposition for case I are
equivalent. Otherwise in cases II, III there is only a single irreducible representation of
dimension 2d0 . If in (2.165) a → a′ then M̄ (h) → M̄ ′ (h) = S M̄ (h)S −1 corresponds, up to
an equivalence, to the same representation R ¯ 0 . If the group G is abelian then only IIIi is
possible in (2.170). The anti-unitary irreducible representations of G are then determined
by the unitary irreducible representations of G0 .
In the trivial case where G = {e, a} ≃ Z2 and M (h), N (a) → 1 then we may identify
U (a) = T , the anti-unitary time reversal operator and cases I or II arise according to whether
T 2 = 1 or −1 and we take N̄ (a) → 1 or −1 respectively. As described later T 2 = (−1)F where
F is the fermion number. For time reversal invariant systems with an odd number of spin- 21
particles there is then a twofold degeneracy. This applies in atomic physics with just an
external electric field, since electric fields are invariant under time reversal, and is termed
Kramers22 degeneracy.
A restriction of general anti-unitary representations arises if the action of a ∈ G1 gener-
ates an inner automorphism on G0 so that

a−1 ha = f hf −1 , f ∈ G0 for all h ∈ G0 , ⇒ [af, h] = 0 , (af )2 ∈ Z(G0 ), . (2.172)

If the centre of G0 , Z(G0 ), is trivial then (af )2 = e. In this case G ≃ G0 × Z2 and we may
identify T = U [a0 f ] as the time reversal operator for some particular a0 ∈ G1 . For such an
anti-unitary representation following (2.159)

DT D(h)∗ = D(h)DT , DT DT∗ = ±1 , DT = D(a0 f ) , (2.173)

allowing (a0 f )2 = e to be projectively represented so that T 2 = ±1. For any a ∈ G1

D(a) = D(aa0−1 )DT . (2.174)

22
Hans Kramer, 1894-1952, Dutch.

54
DT is assumed to be unitary, DT∗ = DT−1T , so that this implies
D(h)∗ = CD(h)C −1 , C = ±C T , C = DT−1 . (2.175)

In case I of the Wigner classification {D(h)} forms an irreducible representation which

is then real or pseudo-real according to whether T 2 = 1 or −1. For case II D(h) decomposes
into two identical irreducible representations M (h) and we can take
0 K
DT = ( ), KM (h)∗ = M (h)K , KK ∗ = ∓1 , (2.176)
−K 0

In this case we can therefore take C = K −1 with C = ∓C T so that the representation defined
by {M (h)} is pseudo-real or real depending on whether T 2 = −1 or 1. Finally for case III
D(h) decomposes into two different irreducible representations M (h), M̄ (h) and we can
take
0 K
DT = ( ), K M̄ (h)∗ = M (h)K , K̄M (h)∗ = M̄ (h)K̄ , K K̄ ∗ = K̄K ∗ = ±1 . (2.177)
K̄ 0
Since M, M̄ are inequivalent the representations are necessarily complex with M̄ equivalent
to the conjugate representation formed from M . The type of representation of G0 , real,
pseudo-real or complex, then determines the associated anti-unitary representation of G to
correspond to cases I, II or III.
Extending results from the standard discussions of representations to the anti-unitary
case is more involved. The complex character
χ(g) = tr(D(g)) , (2.178)
is well defined if g ∈ G1 , as a consequence of (2.162), under changes of basis only if S is
restricted to be real. Restricting to g ∈ G0 the character is independent of the choice of
basis and since, from (2.159) and (2.160),
D(hgh−1 ) = D(h)D(g)D(h)−1 , D(ag −1 a−1 ) = D(a)D(g)TD(a)−1 , h ∈ G0 , a ∈ G 1 ,
(2.179)
the character χ(g)∣g∈G0 is invariant for any g belonging to an extended conjugacy class
defined by C¯G (g) = {hgh−1 , ag −1 a−1 ∶ h ∈ G0 , a ∈ G1 }. In general C¯G (g) ⊃ CG0 (g) but
may also include additional conjugacy classes CG0 (g1 ) if ag −1 a−1 = g1 ∉ CG0 (g) for some a.
If a′ g −1 a′−1 = g2 for some a′ ≠ a then necessarily g2 ∈ CG0 (g1 ) since aa′−1 g2 a′ a−1 = g1
and aa′−1 ∈ G0 . If there is any a ∈ G1 such that ag −1 a−1 = g for some a ∈ G1 then
C¯G (g) = CG0 (g), g ∈ G0 , if there is no such a, so that ag −1 a−1 = g1 ∉ CG0 (g) for any a,
then C¯G (g) = CG0 (g) ∪ CG0 (g1 ).
For the case G = G0 × Z2 then for g ∈ G0 C¯G (g) = CG0 (g) if g −1 ∈ CG0 (g), otherwise
CG (g) = CG0 (g) ∪ CG0 (g −1 ).
¯

From (2.163) and (2.165)

h = a−1 h−1 a for some a ∈ G1 ,

⎧
⎪2 χR0 (h) ,
⎪
χR (h) = ⎨
⎩χR0 (h) + χR̄0 (h) ,
⎪
⎪ χR̄0 (h) = χR0 (a−1 h−1 a) , h ≠ a−1 h−1 a for any a ∈ G1 .
(2.180)

55
If the irreducible representations for G0 are labelled R0,r , with R ¯ 0,r = R0,r̄ , with χr the
corresponding characters and Cs are the conjugation classes for G0 , with r, s = 1, . . . , n0 , then
for G the irreducible anti-unitary representations are then RG,j with associated characters
χG,j and the extended conjugation classes C¯G,k are then determined in terms of these

h ∈ Cs , a−1 h−1 a ∈ Cs̄ , Cs = Cs̄ ,

⎧
¯ ⎪Cs ,
⎪
CG,k = ⎨
⎩Cs ∪ Cs̄ ,
⎪
⎪ h ∈ Cs , a−1 h−1 a ∈ Cs̄ , Cs ≠ Cs̄ ,
⎧
¯ ⎪χr (Cs ) , 2 χr (Cs ) ,
⎪ I, II ,
χG,j (h)∣h∈C¯ ≡ χG,j (CG,k ) = ⎨ (2.181)
⎩χr (Cs ) + χr̄ (Cs ) , χr̄ (Cs ) = χr (Cs̄ ) , III .
G,k ⎪
⎪

The number of irreducible representations for G is equal to the number of extended con-
jugacy classes so that j, k = 1, . . . , n̄. In general n0 − n̄ = n̄III . For IIIi the anti-unitary
characters are real. From the orthogonality relations for characters of G0 (2.56) we may
directly obtain

1 ∗ 1 n̄ ¯ ∗
∑ χG,j ′ (h) χG,j (h) = ∑ dk χG,j ′ (C¯G,k ) χG,j (C¯G,k ) = tj δj ′ j , d¯k = dim C¯G,k ,
∣G0 ∣ h∈G0 ∣G0 ∣ k=1
(2.182)
with tj = 1, 4, 2 according to whether the representation RG,j is of type I, II, III. Since the
number of extended conjugacy classes is equal to the number of anti-unitary representations
n̄
1
χG,j (C¯G,k ) χG,j (C¯G,k′ )∗ = ¯ δk′ k .
∣G0 ∣
∑ (2.183)
j=1 tj dk

As a special case
n̄
1
∑ (dim RG,j )2 = ∣G0 ∣ . (2.184)
j=1 tj

As was described by Dyson23 some problems can be circumvented by adopting a real form
for the representation of G. Starting from a complex co-representation of G of dimension
n satisfying (2.159), (2.160) then corresponding real matrices DR are obtained by taking,
for any h ∈ G0 , a ∈ G1 ,

D(h) = a(h) + i b(h) ∈ U (n) → DR (h) = a(h) 12 + b(h) J ∈ O(2n, R) ,

D(a) = c(a) + i d(a) ∈ U (n) → DR (a) = (c(a) 12 + d(a) J)θ ∈ O(2n, R) , (2.185)

where
0 1) ,
J = ( −1 θ = ( 10 −1
0 ) , θJ = −Jθ . (2.186)
0

{DR (h), DR (a)} then form a standard representation of G and 2 Re χ(h) = tr(DR (h))
for h ∈ G0 . The conjugate representation D(g)∗ → θD(g)θ and so has an equivalent real
representation. The real conjugacy classes for any g ∈ G are then CR (g) = CG (g) if g −1 ∈
CG (g), CG (g) ∪ CG (g −1 ) otherwise.
23
Freeman Dyson, 1923-2020, Britiah then American.

56
 For the three cases described above

I ∶ DR (h) = MR (h) , DR (a) = NR (a) , n = d0 ,

II ∶ DR (h) = MR (h) ⊗ 12 , DR (a) = NR (a) ⊗ J , n = 2d0 ,
III ∶ DR (h) = MR (h) ⊗ P+ + M̄R (h) ⊗ P− ,
DR (a) = NR (a) ⊗ P+ J − N̄R (a) ⊗ P− J , P± = 21 (12 ± θ) , n = 2d0 , (2.187)

where case I has been reduced to a single irreducible component and the real forms are
obtained from (2.163) just as in (2.185). The representations in each case are characterised
by the algebra of matrices commuting with DR (h), DR (a) for all h, a. For {M (h)}, {M̄ (h)}
forming irreducible representations of G0 these have a real basis 1d0 eµ where

I ∶ e 0 = 12 , II ∶ e0 = 12 ⊗ 12 , e1 = J ⊗ θ , e2 = 12 ⊗ J , e3 = J ⊗ θJ ,
III ∶ e0 = 12 ⊗ 12 , e1 = J ⊗ θ , (2.188)

where e02 = e0 and in case II ei ej = −δij e0 + ijk ek for i, j, k = 1, 2, 3, the algebra of quater-
nions, while for case III then e12 = −e0 the algebra corresponding to complex numbers.
Only in case I is the usual form of Schur’s lemma for irreducible representations applica-
ble.24 For case III although {M (h)}, {M̄ (h)} define inequivalent representations of G0 the
corresponding real representations {MR (h)}, {M̄R (h)} may or may not be equivalent under
real conjugation.
Commuting with 1d0 eµ does not uniquely characterise the representation matrices in
(2.187) in each case. It is then necessary to augment the basic group G so that

G → G̃ = {j r G ∶ r = 0, 1, 2, 3, j 4 = e, hj = jh, h ∈ G0 , aj = j 3 a, a ∈ G1 } , ∣G̃∣ = 4∣G∣ . (2.189)

Correspondingly G̃0 ≃ Z4 × G0 . The representation matrices for G are extended to G̃ by

taking DR (j r g) = J˜r DR (g) with J˜ = J or J ⊗ 12 according to case I or cases II, III.
Just as in section 2.4 for two representations R, R′ with dimensions dR , dR′ respectively
we may define in terms of the real representation matrices as in (2.187)

(R′ ,R) 1 (R′ ) −1 (R)

Srs,uv = ∑ DR,rv (g̃ )DR,us (g̃) , (2.190)
∣G̃∣ g̃∈G̃

so that as shown in (2.49) the two sets of real representation matrices for R, R′ are linked
(R′ ,R) (R) (R′ ) (R′ ,R) (R′ ,R) (R′ ) (R) (R′ ,R)
by Srt,uv DR,ts (g̃) = DR,rt (g̃) Sts,uv and Srs,uw DR,wv (g̃) = DR,uw (g̃) Srs,wv . For R, R′
(R′ ,R)
irreducible it is then necessary that Srs,uw = 0 for R ≠ R′ but in this case instead of (2.50)
′
(R ,R)
Srs,uv = δRR′ ∑ cµν
R (1dR eµ )rs (1dR eν )uv . (2.191)
µ,ν

24
The real matrices which commute with DR (h), DR (a) for all h, a form a real vector space which is
closed under matrix multiplication. These may be restricted to form a division algebra where every non
zero element has an inverse. A theorem due to Frobenius says that every such real division algebra has a
basis in terms of real eµ which satisfy the conditions as above for R, C or H. This provides an alternative
characterisation of the three possible anti-unitary representations.

57
It is easy to see that

(R,R) 1
Srs,uv (1dR eω )vu = (1dR eω )rs ⇒ cµν
R ηνω = δ µ ω for tr(eµ eν ) = cR ηµν , (2.192)
tR dR
where, taking cR = 2 or tR = 4 according to whether the representation is of type I or
type II,III, ηµν is diagonal with η00 = 1, ηij = −δij . Hence cµν µν
R = η /cR dR with η
µν
the
(R) (R)
inverse of ηµν . For orthogonal matrices DR,rv (g̃ −1 ) = DR,vr (g̃) and the result (2.191) can
be re-expressed by summing over G̃/G as
1 (R) (R) ˜ (R) (g))vr (JD
˜ (R) (g))us ) = ∑ cµν (1d eµ )rs (1d eν )uv .
∑ (D (g) DR,us (g) + (JD R R R
2∣G∣ g∈G R,vr R R
µ,ν
(2.193)
By contracting indices
1 (R) (R) ˜ (R) (g)) = 1 ∑ D(R) (a2 )
˜ (R) (g)JD
∑ (DR (g)DR (g) + JD R R
2∣G∣ g∈G ∣G∣ a∈G1 R
1
= η µν 1dR eµT eν , (2.194)
cR dR
(R) (R)
since DR (g)J˜ = ±JD
˜ T
R (g) for g = h, a. Using tr(eµ eν ) = cR δµν

⎧
⎪ 1, I,
1 1 ⎪
⎪
¯ (R) 2 2 µν ⎪
FR = ∑ tr(DR (a )) = ∑ Re χR (a ) = η δµν = ⎨−2 , II , (2.195)
∣G∣ a∈G1 ∣G0 ∣ a∈G1 ⎪
⎪
⎪
⎩ 0,
⎪ III .

This result extends the Frobenius-Schur indicator as in (2.63) to anti-unitary representa-

tions. The cases I, III, II may be labelled by R, C, H according to the different forms of
the commutator algebra with the bases given by (2.188).
By taking traces in (2.193)
1 (R) 2 (R) 2 1 ∗ µν
∑ ((tr a (h)) + (tr b (h)) ) = ∑ χR (h) χR (h) = η ηµν , (2.196)
∣G0 ∣ h∈G0 ∣G0 ∣ h∈G0

so that
⎧
⎪ 1, I,
⎪
⎪
⎪
⟨χR , χR ⟩ = tR = ⎨ 4 , II , (2.197)
⎪
⎪
⎪
⎩ 2,
⎪ III .
This is as expected from the usual orthogonality relations for characters of irreducible
representations of G0 since for the three possible cases χR,I = χM , χR,II = 2χM and χR,III =
χM + χM̄ . The result is equivalent to (2.182). For any irreducible representation then

F¯R + ⟨χR , χR ⟩ = F¯R + tR = 2 . (2.198)

Various different possibilities for anti-unitary representations can be illustrated by a

range of different examples which are given in terms of tables where the characters for each

58
extended conjugacy class are given for G0 . For {M (h)} an irreducible complex representa-
tion of G0 of dimension n there is an associated irreducible 2n-dimensional epresentation
of G0 × Z2 given by taking

M (h) 0 0 1
D(h) = ( ), DT = ( ). (2.199)
0 M (h)∗ 1 0

This corresponds to case III. If {±1} ⊂ {M (h)} then there is an anti-unitary representation
for
M (h) 0 0 1
D(h) = ( ), DT = ( ). (2.200)
0 M (h)∗ −1 0
corresponding to the central product G = (G0 × Z4 )/Z2 .
The results described below are all derived from the character tables in 2.7. We also list
for each representation of G0 whether it is real R, pseudo-real H or complex C.
For G = Zn × Z2 , G0 = Zn ,
Zn n odd C¯1 C¯2,r F¯
R1 , R 1 1 1 , k, r = 1, . . . , 12 (n − 1) , (2.201)
R2,k , C 2 2 cos 2krπ/n 0

Zn n even C¯1,1 C¯2,r C¯1,2 F¯

R1,1 , R 1 1 1 1
R2,k , C 2 2 cos 2krπ/n 2(−1) k
0 , k, r = 1, . . . , 12 n − 1 , (2.202)
1
R1,2 , R 1 (−1)r (−1) 2 n 1

For G = Z4n , G0 = Z2n ,

Z2n C¯1,1 C¯2,r C¯1,2 F¯
R1,1 , R 1 1 1 1
k , k, r = 1, . . . , n − 1 . (2.203)
R2,k , C 2 2 cos krπ/n 2(−1) 0
R1,2 , R 2 2(−1)r 2(−1)n −2

For G = Dn , G0 = Zn the irreducible representations are all one dimensional,

Zn n odd C¯1,r F¯
R k=0 , k, r = 0, . . . , n − 1 , (2.204)
R1,k , C k≠0 exp 2krπ i/n 1

Zn n even C¯1,r F¯
R k = 0, 12 n , k, r = 0, . . . , n − 1 . (2.205)
R1,k , C k ≠ 0, 21 n
exp 2krπ i/n 1

59
For G = D4n , G0 = D2n ,
D2n C¯1,1 C¯1,2 C¯2,r C¯2,n F¯
R1,1 , R 1 1 1 1 1
R1,2 , R 1 1 1 −1 1 , k, r = 1, . . . , n − 1 . (2.206)
k
R2,k , R 2 2(−1) 2 cos krπ/n 0 1
R2,n , R 2 2(−1)n 2(−1)r 0 0

For G = Q4n , G0 = Z2n the conjugacy classes for Z2n are not extended and
Z2n C¯1,r F¯
C k ≠ 0, 21 n
R1,k , exp 2krπ i/n 1 r = 0, . . . , 2n − 1 ,
R k = 0, 12 n , (2.207)
k = 0, . . . , n − 1 .
C k≠ 1
(n−1)
R2,k , R k=
2
1
(n−1)
2 exp(2k + 1)rπ i/n −2
2

In this case F¯R = χR (C¯1,n ).

For other examples we may consider assuming quaternionic groups for G0 . For each
quaternion group GQ described in 1.5 there is a corresponding faithful irreducible two
dimensional representation {M (h)} for all h ∈ GQ obtained by using the representation
(1.115) for the quaternions. There is then a four dimensional anti-unitary representation
for G = GQ × Z2 , G0 = GQ obtained by taking

M (h) 0 0 J
D(h) = ( ), DT = ( ), (2.208)
0 M (h) −J 0

which generically corresponds to case II. If {±1} ⊂ GQ then for G = (GQ × Z4 )/Z2 , G0 = GQ ,
there is a two dimensional anti-unitary representation given by

D(h) = M (h) , DT = J . (2.209)

Clearly now DT 2 = −12 .

For G = Q4n × Z2 and G0 = Q4n we may use (2.208)
Q4n n odd C¯1,1 C¯1,2 C¯2,r C¯2n F¯
R1,1 , R 1 1 1 1 1
R1,2 , R 1 1 1 −1 1 r = 1, . . . , n − 1 ,
, (2.210)
R2,n , C 2 −2 2(−1) r
0 0 k = 1, . . . , 12 (n − 1) ,
R2,k , R 2 2 2 cos 2krπ/n 0 1
R4,k , H 4 −4 4 cos(2k − 1)rπ/n 0 −2

60
Q4n n even C¯1,1 C¯1,2 C¯2,r C¯n,1 C¯n,2 F¯
R1,1 , R 1 1 1 1 1 1
R1,2 , R 1 1 1 −1 −1 1
r = 1, . . . , n − 1 ,
R1,3 , R 1 1 (−1)r 1 −1 1 , (2.211)
k = 1, . . . , 12 (n − 2) .
R1,4 , R 1 1 (−1)r −1 1 1
R2,k , R 2 2 2 cos 2krπ/n 0 0 1
R4,k , H 4 −4 4 cos(2k − 1)rπ/n 0 0 −2
(n−1) 1
For the n odd case 4n F¯R = 2 χR (C¯1,1 ) + 2n χR (C¯1,2 ) + 4 ∑r=1 2
χR (C¯2,2r ) while for n even
1
(n−2)
4n F¯R = 2 χR (C¯1,1 ) + 2(n + 1) χR (C¯1,2 ) + 4 ∑ 2 χR (C¯2,2r ). r=1

For G = Q4n × Z4 /Z2 and G0 = Q4n we may use (2.209)

Q4n n odd C¯1,1 C¯1,2 C¯2,r C¯2n F¯
R1,1 , R 1 1 1 1 1
R1,2 , R 1 1 1 −1 1 , k, r = 1, . . . , n − 1 , (2.212)
r
R2,n , C 2 −2 2(−1) 0 0
H k odd
R2,k , R k even 2 2(−1)k 2 cos krπ/n 0 1

Q4n n even C¯1,1 C¯1,2 C¯2,r C¯n,1 C¯n,2 F¯

R1,1 , R 1 1 1 1 1 1
R1,2 , R 1 1 1 −1 −1 1
, k, r = 1, . . . , n − 1 . (2.213)
R1,3 , R 1 1 (−1)r 1 −1 1
R1,4 , R 1 1 (−1)r −1 1 1
H k odd
R2,k , R k even 2 2(−1)k 2 cos krπ/n 0 0 1

(n−1) 1
For the n odd case 4n F¯R = 2n χR (C¯1,1 ) + 2 χR (C¯1,2 ) + 4 ∑r=1 2
χR (C¯2,2r−1 ) while for n even
1
(n−2)
4n F¯R = 2(n + 1) χR (C¯1,1 ) + 2 χR (C¯1,2 ) + 4 ∑ 2 χR (C¯2,2r−1 ). r=1

These various examples illustrate some of the possibilities where the different cases, I, II
and IIIi, IIIii, of anti-unitary representations for G can be combined with real, pseudo-real
or complex representations of G0 . There are ten possibilities altogether (eight are given in
the above tables) since case IIIii for G requires a complex representation for G0 .

61
3 Rotations and Angular Momentum, SO(3) and SU (2)

Symmetry under rotations in three dimensional space is an essential part of general physical
theories which is why they are most naturally expressed in vector notation. The fundamental
property of rotations is that the lengths, and scalar products, of vectors are invariant.
Rotations correspond to orthogonal matrices, since acting on column vectors v, they are
the most general transformations leaving v T v invariant, for real v the length ∣v∣ is given by
∣v∣2 = v T v. For any real orthogonal matrix M then if v is an eigenvector, in general complex,
M v = λv we also have M v ∗ = λ∗ v ∗ , so that if λ is complex both λ, λ∗ are eigenvalues, and
(M v ∗ )T M v = ∣λ∣2 v † v = v † v so that we must have ∣λ∣2 = 1.

3.1 Three Dimensional Rotations

Rotations in three dimensions are then determined by real matrices R ∈ O(3) and hence
satisfying
RT R = 13 ⇒ (det R)2 = 1 . (3.1)
The eigenvalues of R can only be eiθ , e−iθ and 1 or −1 so that a general R can therefore be
reduced, by a real transformation S, to the form

⎛cos θ − sin θ 0 ⎞
SRS −1 = ⎜ sin θ cos θ 0⎟. (3.2)
⎝ 0 0 ±1⎠

For det R = 1, so that R ∈ SO(3), we must have the +1 case when

tr R = 2 cos θ + 1 . (3.3)

Acting on a spatial vector x the matrix R induces a linear transformation

x → x′ = xR , (3.4)
R

where, for i, j, three dimensional indices, we have

x′i = Rij xj , (3.5)

For det R = −1 the transformation involves a reflection. Of course rotations preserve scalar
products and vector products up to a sign

xR ⋅ y R = x ⋅ y , xR × yR = det R (x × y)R . (3.6)

A general R ∈ SO(3) has 3 parameters which may be taken as the rotation angle θ and
the unit vector n, which is also be specified by two angles, and is determined by Rij nj = ni .
n defines the axis of the rotation. The matrix may then be expressed in general as

Rij (θ, n) = cos θ δij + (1 − cos θ)ni nj − sin θ εijk nk , (3.7)

62
where εijk is the three dimensional antisymmetric symbol, ε123 = 1. The parameters (θ, n)
cover all rotations if
n ∈ S2 , 0 ≤ θ ≤ π , (π, n) ≃ (π, −n) , (3.8)
with S 2 the two-dimensional unit sphere. The space corresponding to (3.8) is then the ball
of radius π with opposite points on the surface identified. Equivalently the range of the
angle θ can be extended to 0 to 2π if

R(θ, n) = R(2π − θ, −n) = R(θ + 2π, n) . (3.9)

In (3.4) this requires

xR(θ,n) = cos θ x + (1 − cos θ) n n ⋅ x + sin θ n × x . (3.10)

−1
For an arbitrary R ∈ SO(3) then taking x → xR gives
−1 −1
((xR )R(θ,n) )R = xRR(θ,n)R = cos θ x + (1 − cos θ) nR nR ⋅ x + sin θ nR × x . (3.11)
−1 −1
since nR ⋅ x = n ⋅ xR and nR × x = (n × xR )R . Hence

R R(θ, n)R−1 = R(θ, nR ) , (3.12)

so that all rotations with the same θ belong to a single conjugacy class.
The additional transformation

x′ = x − 2 m m ⋅ x , m2 = 1 ⇒ x′2 = x2 , (3.13)

corresponds to a reflection through the plane perpendicular to m. The associated matrix

given by
Rij = δij − 2 mi mj , Rik Rkj = δij , (3.14)
has eigenvalues 1, 1, −1 and belongs to O(3). Arbitrary elements of O(3) can be obtained
by combining rotations and reflections. For two reflections defined by unit vectors m and
then l

x′ = x − 2 l l ⋅ x − 2 m m ⋅ x + 4 l l ⋅ m m ⋅ x
= (2 (l ⋅ m)2 − 1)x + 2 (l × m) (l × m) ⋅ x + 2 l ⋅ m (l × m) × x , (3.15)

which, by comparing with (3.10), if l ⋅ m = cos φ corresponds to a rotation through an angle

2φ and about an axis n where l × m = sin φ n. Any rotation R(2φ, n) can be expressed
as a product of two reflections by choosing a unit vector m, m ⋅ n = 0, and then defining
l = cos φ m + sin φ (m × n), l2 = 1.
For an infinitesimal rotation R(δθ, n) acting on a vector x and using standard vector
notation we then have
x ÐÐÐÐ→ x′ = x + δθ n × x . (3.16)
R(δθ,n)

It is easy to see that x′2 = x2 + O(δθ2 ).

63
 Although the group SO(3) is infinite dimensional the notion of a sum over group ele-
ments for a finite group can be extended to an integration over the three dimensional unit
ball. The crucial property (1.7) can be extended by requiring

∫ dµ(θ, n) f (R(θ, n)) = ∫ dµ(θ, n) f (R(θ, n)R) for all R ∈ SO(3) . (3.17)

This is satisfied, as shown later, by taking

dµ(θ, n) = dΩn dθ sin2 21 θ, n ∈ S2 , 0 ≤ θ ≤ π . (3.18)

The group volume VSO(3) = ∫SO(3) dµ(θ, n) = 2π 2 .

3.2 Isomorphism of SO(3) and SU (2)/Z2

SO(3) ≃ SU (2)/Z2 , where Z2 is the centre of SU (2) which is formed by the 2 × 2 matrices
{1, −1}, is of crucial importance in understanding the role of spinors under rotations. To
demonstrate this we introduce the standard Pauli 25 matrices, a set of three 2 × 2 matrices
which have the explicit form

0 1 0 −i 1 0
σ1 = ( ), σ2 = ( ), σ3 = ( ). (3.19)
1 0 i 0 0 −1

These matrices satisfy the algebraic relations

σi σj = δij 12 + i ijk σk , (3.20)

and also are traceless and hermitian. The matrices (12 , iσ3 , iσ2 , iσ1 ) provide a two dimen-
sional complex representation of of the quaternion algebra (1.79) identical to (1.115). Adopt-
ing a vector notation σ = (σ1 , σ2 , σ3 ), so that (3.20) is equivalent to a⋅σ b⋅σ = a⋅b 1+i a×b⋅σ,
we have
σ† = σ , tr(σ) = 0 . (3.21)
Using (3.20) then gives
tr(σi σj ) = 2δij , (3.22)
which ensures that any 2 × 2 matrix A can be expressed in the form

A = 21 tr(A) 1 + 12 tr(σA) ⋅ σ , (3.23)

since the Pauli matrices form a complete set of traceless and hermitian 2 × 2 matrices.
The Pauli matrices ensure that there is a one to one correspondence between real three
vectors and hermitian traceless 2 × 2 matrices, given explicitly by

x → x ⋅ σ = (x ⋅ σ)† , x = 12 tr(σ x ⋅ σ) , (3.24)

Furthermore x ⋅ σ satisfies the matrix equation

(x ⋅ σ)2 = x2 1 . (3.25)
25
Wolfgang Ernst Pauli, 1900-58, Austrian. Nobel prize 1945.

64
 √
From (3.25) and (3.21) the eigenvalues of x ⋅ σ must be ± x2 and in consequence we have

det(x ⋅ σ) = −x2 . (3.26)

For any A ∈ SU (2) we can then define a linear transformation x → x′ by

x′ ⋅ σ = A x ⋅ σA† , (3.27)

since we may straightforwardly verify that A x ⋅ σA† is hermitian and is also traceless, using
the invariance of any trace of products of matrices under cyclic permutations and

AA† = 1 . (3.28)

With, x′ defined by (3.27) and using (3.26),

x′2 = − det(x′ ⋅ σ) = − det(A x ⋅ σA† ) = − det(x ⋅ σ) = x2 , (3.29)

using det(XY ) = det X det Y and from (3.28) det A det A† = 1. Hence, since this shows
that ∣x′ ∣ = ∣x∣,
x′i = Rij xj , (3.30)
with [Rij ] an orthogonal matrix. Furthermore since as A → 1, Rij → δij we must have
det[Rij ] = 1. Explicitly from (3.27) and (3.22)

σi Rij = Aσj A† ⇒ Rij = 1

2 tr(σi Aσj A† ) . (3.31)

To show the converse then from (3.31), using (note σj σi σj = −σi ) σj A† σj = 2 tr(A† )1 − A† ,
we obtain
Rjj = ∣tr(A)∣2 − 1 , σi Rij σj = 2 tr(A† )A − 1 . (3.32)
For A ∈ SU (2), tr(A) = tr(A† ) is real (the eigenvalues of A are e±iα giving tr(A) = 2 cos α)
so that (3.32) may be solved for tr(A) and then A,
1 + σi Rij σj
A=± 1 . (3.33)
2(1 + Rjj ) 2

The arbitrary sign, which cancels in (3.31), ensures that in general ±A ↔ Rij . This ensures
SO(3) ≃ SU (2)/Z2 . Any arbitrary 2 × 2 matrix A can be expanded in a basis formed by
the Pauli matrices and the unit matrix and then

A = x0 1 − i x ⋅ σ ∈ SU (2) ⇔ x02 + x2 = 1 ⇔ (x0 , x) ∈ S 3 . (3.34)

The additional transformation

x′ ⋅ σ = −m ⋅ σ x ⋅ σ m ⋅ σ , m2 = 1 ⇒ x′ = x − 2 m m ⋅ x , (3.35)

corresponds to the reflection (3.35).

For a SO(3) rotation R through an infinitesimal angle as in (3.16) then from (3.7)

Rij = δij − δθ εijk nk , (3.36)

65
and it is easy to obtain for the associated SU (2) matrix ,

AR (δθ, n) = 1 − 21 δθ in ⋅ σ . (3.37)

Note that since det(1+X) = 1+tr X, to first order in X, for any matrix then the tracelessness
of the Pauli matrices is necessary for (3.37) to be compatible with det AR = 1. For a finite
rotation angle θ then, with (3.3), (3.32) gives ∣tr(AR )∣ = 2∣ cos 12 θ∣ and the matrix AR can
be found by exponentiation, where corresponding to (3.7),

AR (θ, n) = e− 2 iθ n⋅σ = cos 21 θ 1 − sin 12 θ in ⋅ σ , ,

1
±AR (θ, n) → R(θ, n) (3.38)

The parameters (θ, n) cover all SU (2) matrices for

n ∈ S2 , 0 ≤ θ < 2π , (3.39)

in contrast to (3.8). For the matrices in (3.38) AR (2π, n) = −1 ∈ Z(SU (2)). For the
matrices {AR (θ, n)} integration can be defined just as in (3.18) with the integration range
on θ extended to be from 0 to 2π.

3.2.1 Non Compact Isomorphisms

The relation between the compact groups SU (2) and SO(3) can be extended to related
non compact groups. If we define σ̃ = (σ1 , iσ2 , σ3 ) then the σ̃i matrices are all real and
traceless. Hence for y0 , y real

A = y0 1 + y ⋅ σ̃ ∈ Sl(2, R) ⇔ y02 − y12 + y22 − y32 = 1 . (3.40)

Clearly the parameters have an infinite range, it contains the subgroup corresponding to
(1.123) by taking y0 = cosh θ, y1 = sinh θ, y2 = y3 = 0. Since

det(x ⋅ σ̃) = −x12 + x22 − x32 , (3.41)

then

x′ ⋅ σ̃ = A x ⋅ σ̃A−1 , A ∈ Sl(2, R) ⇔ −x12 + x22 − x32 = −x′12 + x′22 − x′32 . (3.42)

An alternative non compact group is SU (1, 1) which is defined by

B † σ3 B = σ3 , det B = 1 . (3.43)

A basis of generators in terms of the Pauli matrices is σ̂ = (σ1 , σ2 , iσ3 ) where σ3 σ̂ + σ̂ † σ3 = 0.

In consequence

B = y0 1 + y ⋅ σ̂ ∈ SU (1, 1) ⇔ y02 − y12 − y22 + y32 = 1 . (3.44)

In this case
det(x ⋅ σ̂) = −x12 − x22 + x32 , (3.45)

66
and

x′ ⋅ σ̂ = B x ⋅ σ̂B −1 , B ∈ SU (1, 1) ⇔ −x12 − x22 + x32 = −x′12 − x′22 + x′32 . (3.46)

There is an isomorphism of Sl(2, R) and SU (1, 1) by rotating σ2 into σ3 . This may

be achieved for A(y) ∈ Sl(2, R) then ei 4 πσ1 A(y)e−i 4 πσ1 = B(y′ ) ∈ SU (1, 1), with y′ =
1 1

(y1 , y3 , y2 ). In each case, by similar arguments to previously, we then have

Sl(2, R)/Z2 ≃ SU (1, 1)/Z2 ≃ SO(2, 1) . (3.47)

3.3 Infinitesimal Rotations and Generators

To analyse the possible representation spaces for the rotation group it is sufficient to consider
rotations which are close to the identity as in (3.16). If consider two infinitesimal rotations
R1 = R(δθ1 , n1 ) and R2 = R(δθ2 , n2 ) then it is easy to see that the commutator

R = R2−1 R1−1 R2 R1 = 1 + O(δθ1 δθ2 ) . (3.48)

Acting on a vector x and using (3.16) and keeping only terms which are O(δθ1 δθ2 ) we find

x → x′ = x + δθ1 δθ2 (n2 × (n1 × x) − n1 × (n2 × x))

R
= x + δθ1 δθ2 (n2 × n1 ) × x , (3.49)

using standard vector product identities.

Acting on a quantum mechanical vector space the corresponding unitary operators are
assumed to be of the form
U [R(δθ, n)] = 1 − iδθ n ⋅ J , (3.50)
J are the generators of the rotation group. Since U [R(δθ, n)]−1 = 1 + iδθ n ⋅ J + O(δθ2 ) the
condition for U to be a unitary operator becomes

J† = J , (3.51)

or each Ji is hermitian. If we consider the combined rotations as in (3.48) in conjunction

with (3.49) and (3.50) we find

U [R] = 1 − iδθ1 δθ2 (n2 × n1 ) ⋅ J

= U [R2 ]−1 U [R1 ]−1 U [R2 ]U [R1 ]
= 1 − δθ1 δθ2 [n2 ⋅ J , n1 ⋅ J] , (3.52)

where it is only necessary to keep O(δθ1 δθ2 ) contributions as before. Hence we must have

[n2 ⋅ J , n1 ⋅ J] = i (n2 × n1 ) ⋅ J , (3.53)

or equivalently
[Ji , Jj ] = i εijk Jk . (3.54)

67
 Acting on functions of x
J → L = −i x × ∇ , (3.55)
so that, neglecting δθ2 ,

(1 − iδθ n ⋅ L)f (x) = f (x − δθ n × x) , (3.56)

corresponds to an infinitesimal rotation. It is straightforward to verify that Li satisfies the

commutation relations (3.54).
Although (3.50) expresses U in terms of J for infinitesimal rotations it can be extended
to finite rotations since
N
θ
U [R(θ, n)] = exp(−iθ n ⋅ J) = lim (1 − i n ⋅ J) . (3.57)
N →∞ N

Under rotations J is a vector. From (3.12), U [R]U [R(δθ, n)]U [R]−1 = U [R(δθ, nR )]
which in turn from (3.50) implies

U [R]Ji U [R]−1 = (R−1 )ij Jj . (3.58)

For a physical system the vector operator, rotation group generator, J is identified as
that corresponding to the total angular momentum of the system and then (3.54) are the
fundamental angular momentum commutation relations. It is important to recognise that
rotational invariance of the Hamiltonian is equivalent to conservation of angular momentum
since
U [R]HU [R]−1 = H ⇔ [J, H] = 0 . (3.59)
This ensures that the degenerate states for each energy must belong to a representation
space for a representation of the rotation group.

3.4 Representations of Angular Momentum Commutation Relations

We here describe how the commutation relations (3.54) can be directly analysed to deter-
mine possible representation spaces V on which the action of the operators J is determined.
First we define
J± = J1 ± iJ2 , (3.60)
and then (3.54) is equivalent to

[J3 , J± ] = ± J± , (3.61a)
[J+ , J− ] = 2J3 . (3.61b)

The hermeticity conditions (3.51) then become

J+† = J− , J3 † = J3 . (3.62)

68
 A basis for a space on which a representation for the angular momentum commutation
relations is defined in terms of eigenvectors of J3 . Let

J3 ∣m⟩ = m∣m⟩ . (3.63)

Then from (3.61a) it is easy to see that

J± ∣m⟩ ∝ ∣m ± 1⟩ or 0 , (3.64)

so that the possible J3 eigenvalues form a sequence . . . , m − 1, m, m + 1 . . . .

If the states ∣m ± 1⟩ are non zero we define

J− ∣m⟩ = ∣m − 1⟩ , J+ ∣m⟩ = λm ∣m + 1⟩ , (3.65)

and hence
J+ J− ∣m⟩ = λm−1 ∣m⟩ , J− J+ ∣m⟩ = λm ∣m⟩ . (3.66)
By considering [J+ , J− ]∣m⟩ we have from (3.61b), if ∣m ± 1⟩ are non zero,

λm−1 − λm = 2m . (3.67)

This can be solved for any m by

λm = j(j + 1) − m(m + 1) , (3.68)

for some constant written as j(j + 1). For sufficiently large positive or negative m we clearly
have λm < 0. The hermeticity conditions (3.62) require that J+ J− and J− J+ are of the form
O† O and so must have positive eigenvalues with zero possible only if J− or respectively
J+ annihilates the state (⟨ψ∣O† O∣ψ⟩ ≥ 0, if 0 then O∣ψ⟩ = 0). Hence there must be both a
maximum mmax and a minimum mmin for m requiring

J+ ∣mmax ⟩ = 0 ⇒ λmmax = (j − mmax )(j + mmax + 1) = 0 , (3.69a)

J− ∣mmin ⟩ = 0 ⇒ λmmin −1 = (j + mmin )(j − mmin + 1) = 0 , (3.69b)

where also
mmax − mmin = 0, 1, 2, . . . . (3.70)
Taking j ≥ 0 the result (3.68) then requires

mmax = j , mmin = −j . (3.71)

For this to be possible we must have

j ∈ {0, 12 , 1, 32 , . . . } , (3.72)

and then for each value of j

m ∈ {−j, −j + 1, . . . j − 1, j} . (3.73)

The corresponding states ∣m⟩ form a basis for a (2j + 1)-dimensional representation space
Vj .

69
3.5 The ∣j m⟩ basis

It is more convenient to define an orthonormal basis for Vj in terms of states {∣j m⟩}, with
j, m as in (3.72) and (3.73), satisfying

⟨j m∣j m′ ⟩ = δmm′ . (3.74)

These are eigenvectors of J3 as before

J3 ∣j m⟩ = m∣j m⟩ . (3.75)

and j may be defined as the maximum value of m so that

J+ ∣j j⟩ = 0 . (3.76)

A state satisfying both (3.75) and (3.76) is called a highest weight state. In this case the
action of J± gives
±
J± ∣j m⟩ = Njm ∣j m±1⟩ , (3.77)
±
where Njm are determined by requiring (3.74) to be satisfied. From (3.66) and (3.68) we
must then have
+ 2 − 2
∣Njm ∣ = λm = (j − m)(j + m + 1) , ∣Njm ∣ = λm−1 = (j + m)(j − m + 1) . (3.78)
±
By convention Njm are chosen to real and positive so that
±
√
Njm = (j ∓ m)(j ± m + 1) . (3.79)

In general we may then define the the states {∣j m⟩} in terms of the highest weight state by
1
n n! (2j)! 2
(J− ) ∣j j⟩ = ( (2j−n)! ) ∣j j − n⟩ , n = 0, 1, . . . , 2j . (3.80)

An alternative prescription for specifying the states ∣j m⟩ is to consider the operator

J2 = J1 2 + J2 2 + J3 2 . In terms of J± , J3 this can be expressed in two alternative forms
⎧
2 ⎪J− J+ + J32 + J3 ,
⎪
J =⎨ 2
(3.81)
⎩J+ J− + J3 − J3 .
⎪
⎪

With the first form in (3.81) and using (3.76) we then get acting on the highest weight state

J2 ∣j j⟩ = j(j + 1)∣j j⟩ . (3.82)

Moreover J2 is a rotational scalar and satisfies

[J2 , Ji ] = 0 , i = 1, 2, 3 . (3.83)

In particular J− commutes with J2 so that the eigenvalue is the same for all m. Hence the
states ∣j m⟩ satisfy
J2 ∣j m⟩ = j(j + 1)∣j m⟩ , (3.84)
as well as (3.75). Nevertheless we require (3.77), with (3.79), to determine the relative
phases of all states.

70
3.5.1 Action of Time Reversal

The anti-unitary operator T acts on the angular momentum operators according to

T J T −1 = −J , T J± T −1 = −J∓ , T J3 T −1 = −J3 (3.85)

This is compatible with the commutation relations (3.54) or (3.61a), (3.61b) since T is
anti-linear. Hence in terms of the ∣jm⟩ basis if this is invariant under the action of T we
have
T ∣jm⟩ = eiφ (−1)j−m ∣j −m⟩ , (3.86)
for some complex phase eiφ . Hence

T 2 ∣jm⟩ = (−1)2j ∣jm⟩ , (3.87)

which generalises to T 2 = (−1)F with F the fermion number.

3.5.2 Representation Matrices

Using the ∣j m⟩ basis it is straightforward to define corresponding representation matrices

for each j belonging to (3.72). For the angular momentum operator

J(j) m′ m = ⟨j m′ ∣J∣j m⟩ (3.88)

or alternatively
J∣j m⟩ = ∑ ∣j m′ ⟩ J(j) m′ m . (3.89)
m′

The (2j + 1) × (2j + 1) matrices J(j) = [J(j) m′ m ] then satisfy the angular momentum com-
mutation relations (3.54). From (3.75) and (3.77)
(j) (j)
√
J3 m′ m = m δm′ ,m , J± m′ m = (j ∓ m)(j ± m + 1) δm′ ,m±1 . (3.90)

For R a rotation then

(j)
Dm′ m (R) = ⟨j m′ ∣U [R]∣j m⟩ , (3.91)
(j)
defines (2j + 1) × (2j + 1) matrices D(j) (R) = [Dm′ m (R)] forming a representation of the
the rotation group corresponding to the representation space Vj ,
(j)
U [R]∣jm⟩ = ∑ ∣jm′ ⟩Dm′ m (R) . (3.92)
m′

Note that D(0) (R) = 1 is the trivial representation and for an infinitesimal rotation as in
(3.16)
D(j) (R(δθ, n)) = 12j+1 − iδθ n⋅J(j) . (3.93)

To obtain explicit forms for the rotation matrices it is convenient to parameterise a

rotation in terms of Euler angles φ, θ, ψ when

Rφ,θ,ψ = R(φ, e3 )R(θ, e2 )R(ψ, e3 ) , {Rφ,θ,ψ } = SO(3) for 0 ≤ θ ≤ π , 0 ≤ φ, ψ ≤ 2π , (3.94)

71
where e2 , e3 are unit vectors along the 2, 3 directions. For the corresponding two dimensional
matrix

⎛cos 21 θ e− 2 i(φ+ψ) − sin 12 θ e− 2 i(φ−ψ) ⎞

1 1
− 21 iφ σ3 − 12 iθ σ2 − 21 iψ σ3
Aφ,θ,ψ = e e e = 1 1 , (3.95)
⎝ sin 21 θ e 2 i(φ−ψ) cos 21 θ e 2 i(φ+ψ) ⎠

with

{Aφ,θ,ψ } = SU (2) for 0 ≤ θ ≤ π , −2π ≤ φ − ψ ≤ 2π , 0 ≤ φ + ψ ≤ 4π ,

or 0 ≤ θ ≤ π , 0 ≤ φ ≤ 2π , 0 ≤ ψ ≤ 4π . (3.96)

The allowed regions of (φ, ψ) in (3.94), (3.96) are enclosed in blue, red in φ
. Since

Aφ,θ±2π,ψ = Aφ,θ,ψ±2π = −Aφ,θ,ψ , (3.97)

the region in (3.96) maps into (3.94) allowing for a change of sign.
In terms the angular momentum operators J the rotation operators in terms of Euler
angles are then
U [Rφ,θ,ψ ] = e−iφJ3 e−iθJ2 e−iψJ3 , (3.98)
so that in (3.91)
(j) ′ (j) (j)
Dm′ m (Rφ,θ,ψ ) = e−im φ−imψ dm′ m (θ) , dm′ m (θ) = ⟨j m′ ∣e−iθJ2 ∣j m⟩ . (3.99)
(j)
The matrices d(j) (θ) = [dm′ m (θ)] satisfy d(j) (θ)d(j) (θ′ ) = d(j) (θ + θ′ ). For the special cases
of θ = π, 2π,
(j) (j)
dm′ m (π) = (−1)j−m δm′ ,−m , dm′ m (2π) = (−1)2j δm′ ,m . (3.100)
(j)
Since iJ2 = 12 (J+ − J− ) then with the conventions (3.75) and (3.77) dm′ m (θ) are real.

In general D(j) (R(2π, n)) = (−1)2j 12j+1 , which for j a 21 -integer is not the identity. For
representations of SO(3) it would be necessary to take j to be an integer but in quantum
mechanics any j given by (3.72) is allowed since we require representations only up to a
phase factor. From the result for θ = π we have

e−iπJ2 ∣j m⟩ = (−1)j−m ∣j −m⟩ . (3.101)

Using this and e−iπJ3 ∣j m⟩ = e−iπm ∣j m⟩ with e−iπJ3 J2 eiπJ3 = −J2 we must have from the
definition in (3.99)
(j) (j) ′ (j) ′ (j)
dm′ m (θ) = dmm′ (−θ) = (−1)m −m d−m′ −m (θ) = (−1)m −m dm′ m (−θ) . (3.102)
(j) (j) (j) ′ (j)
Furthermore dm′ m′ (π − θ) = ∑m′′ dm′ m′ (−θ)dm′′ m (π) = (−1)j+m dm′ −m (θ).
For the simplest case j = 21 , it is easy to see from (3.90) that

(1) 0 1 (1) 0 0 (1) 1 1 0

J+ 2 = ( ), J− 2 = ( ), J3 2 = ( ), (3.103)
0 0 1 0 2 0 −1

72
and hence we have
J( 2 ) =
1
1
2 σ, (3.104)
where σi , i = 1, 2, 3 are the Pauli matrices as given in (3.19). It is clear that 12 σi must satisfy
the commutation relations (3.54). The required commutation relations are a consequence
of (3.20). For j = 12 we also have

cos 12 θ − sin 21 θ
d( 2 ) (θ) = (
1
). (3.105)
sin 12 θ cos 12 θ

For j = 1
√
(1) 0 2 √0 (1) √0 0 0 (1) 1 0 0
J+ = (0 0 2) , J− =( 2 √0 0 ) , J3 = (0 0 0) , (3.106)
0 0 0 0 2 0 0 0 −1

and √
⎛1√+ cos θ − 2 sin θ 1√− cos θ ⎞
(1) 1
d (θ) = ⎜ 2 sin θ √2 cos θ − 2 sin θ⎟ . (3.107)
2 ⎝
1 − cos θ 2 sin θ 1 + cos θ ⎠
For any integer j
(j)
d00 (θ) = Pj (cos θ) , (3.108)
witth Pj the usual Legendre polynomials.
(j)
For general j there is an expression for dm′ m (θ) in terms of classical Jacobi polynomi-
als. To obtain the associated differential equation we can start from (3.98) to obtain the
differential relations

J± U [Rφ,θ,ψ ] = −J± U [Rφ,θ,ψ ] , J3 U [Rφ,θ,ψ ] = −J3 U [Rφ,θ,ψ ] , (3.109)

where
∂ ∂ ∂ ∂
J± = e±iφ (i cot θ ± − i csc θ ), J3 = −i . (3.110)
∂φ ∂θ ∂ψ ∂φ
This follows using csc θ e−iθJ2 J3 = (cot θJ3 + J1 )e−iθJ2 . It is straightforward to verify that
J± , J3 satisfy the usual commutation relations (3.61a), (3.61b) and
2
∂2 ∂2 ∂ ∂ ∂
J2 = − 2
− 2
− cot θ − ( cot θ − csc θ ) . (3.111)
∂φ ∂θ ∂θ ∂φ ∂ψ

Using this with (3.109) to evaluate ⟨jm′ ∣ J2 U [Rφ,θ,ψ ]∣jm⟩ leads to the differential equation

∂2 ∂ (j) (j)
+ m′ 2 + (m′ cot θ − m csc θ) ) dm′ m (θ) = j(j + 1) dm′ m (θ)
2
(− − cot θ (3.112)
∂θ2 ∂θ

Writing, for m ≥ m′ ,
(j) m+m′ m−m′
dm′ m (θ) = ( cos 12 θ) ( sin 12 θ) P (cos θ) , (3.113)

73
the differential equation translates into a hypergeometric form

(1 − t2 )P ′′ (t) + 2(m′ − (m + 1) t)P ′ (t) + (j − m)(j + m + 1) P (t) = 0 (3.114)

The solutions are Jacobi polynomials giving

1
(j + m)! (j − m)! 2 ′ ′
(j) 1 m+m 1 m−m (m−m′ ,m+m′ )
dm′ m (θ) =( ) ( cos θ) ( sin θ) P (cos θ) . (3.115)
(j + m′ )! (j − m′ )! 2 2 j−m

For m = m′ = 0 this reduces to (3.108). To check the normalisation for θ → 0 and m ≥ m′

(j) 1 ′ ′
dm′ m (θ) ∼ ( 21 θ)m−m ′
⟨jm′ ∣(J− )m−m ∣jm⟩
(m − m )! 1

1 m−m ′ 1 (j + m)! (j − m′ )! 2
= ( 2 θ) ( ) , (3.116)
(m − m′ )! (j + m′ )! (j − m)!
(m−m′ ,m+m′ ) (j−m′ )!
and Pj−m (1) = (j−m)!(m−m′ )! .

Defining
U (j) = D(j) (R π2 , π2 ,− π2 ) , (3.117)
then
(U (j) d(j) (θ) U (j)† )m′ m = eimθ δm′ ,m . (3.118)
As special cases
√
⎛ 1 2 i √−1 ⎞
( 12 ) 1 1 i 1 √
U =√ ( ), U (1) = ⎜ 2 i √0 2 i⎟ . (3.119)
2 i 1 2 ⎝
−1 2i 1 ⎠

3.5.3 Integration over SO(3) and orthogonality relations

Corresponding to the sum over group elements for a finite group there is an integration over
the group parameters, here the Euler angles θ, φ, ψ, for a continuous group. The crucial
requirement is to respect the property (1.7). If dµθ,φ,ψ this requires that dµθ,φ,ψ = dµθ′ ,φ′ ,ψ′
where the change of variables (φ, θ, ψ) → (θ′ , φ′ , ψ ′ ) is obtained by Rθ,φ,ψ R = Rθ′ ,φ′ ,ψ′ for
any rotation R. Infinitesimally with the j = 21 rotation matrices in (3.95) Aθ,φ,ψ Aθ ,φ ,ψ =
Aθ+δθ,φ+δφ,ψ+δψ which gives, to first order in θ , φ , ψ ,
sin ψ
δθ = cos ψ θ , δφ = θ , δψ = φ + ψ − sin ψ cot θ θ . (3.120)
sin θ
In consequence
sin ψ
⎛ 0 − sin ψ cot θ
sin θ sin2 θ
⎞
d(δθ, δφ, δψ) = (dθ, dφ, dψ)J θ , ⎜
J =⎜ 0 0 0 ⎟, (3.121)
⎟
⎝− sin ψ cos ψ
sin θ − cos ψ cot θ⎠

and hence
δ(dθ dφ dψ sin θ) = dθ dφ dψ (sin θ trJ + cos θ cos ψ)θ = 0 , (3.122)

74
so that we may take for the integration measure in terms of Euler angles
1
dµθ,φ,ψ = dθ dφ dψ sin θ , ∫ dµθ,φ,ψ = 1 , (3.123)
8π 2 SO(3)

with the integration ranges as in (3.94).

The orthogonality relation (2.50) now translates for the unitary rotation matrices into

(j ) (j ) 1
dµθ,φ,ψ Dm11′ m1 (Rφ,θ,ψ ) Dm22′ m2 (Rφ,θ,ψ )∗ = δj j δ ′ ′ δ . (3.124)
2j1 + 1 1 2 m1 m2 m1 m2
∫
SO(3)

With the decomposition (3.99) this is reducible to

π 2
1 (j )
2 (j )
∫ dθ dm′ m (θ) dm′ m (θ) = δj j . (3.125)
0 2j1 + 1 1 2

The integration measure in (3.123) in terms of Euler angles can be transformed to other
parameterisations. The overall angle of rotation Θ can be expressed in terms of Euler angles
using, from (3.38) and (3.95),

cos 12 Θ = 1
2 tr(Aφ,θ,ψ ) = cos 12 θ cos 21 (φ + ψ) . (3.126)

For any function of Θ then

1 2π 1 1 1 1
∫ dµθ,φ,ψ f (Θ) = ∫ dΘ sin 2 Θ f (Θ) ∫ dµθ,φ,ψ δ( cos 2 Θ − cos 2 θ cos 2 (φ + ψ)) , (3.127)
2 0
where if cos 12 Θ > 0 and taking φ± = φ ± ψ,

1 1 1
∫ dµθ,φ,ψ δ( cos 2 Θ − cos 2 θ cos 2 φ+ )
1 2π 2π π
= ∫ dφ −∫ dφ +
∫ dθ sin θ δ( cos 21 Θ − cos 12 θ cos 12 φ+ )
16π 2 −2π 0 0
1 Θ 1 2
= cos 12 Θ ∫ dφ+ 1
= sin 21 Θ . (3.128)
π 0 cos2 2 φ+ π

The final result remains if cos 12 Θ < 0 so that the measure then becomes

1
dΘ sin2 12 Θ , 0 ≤ Θ ≤ 2π . (3.129)
π

3.5.4 Characters for SU (2)

With the definition of characters in (2.51) the rotation group characters

χj (θ) = tr(D(j) (R(θ, n))) , (3.130)

depend only on the rotation angle θ. Since

(j)
Dm′ m (R(θ, z)) = δm′ m e−imθ , (3.131)

75
they may be easily calculated
j
−imθ sin(j + 12 )θ
χj (θ) = ∑ e = . (3.132)
m=−j sin 21 θ

The orthogonality properties corresponding to (2.56) follow from using (3.129) with Θ → θ
1 2π 1 2π
2 1
∫ dθ sin 2 θ χj1 (θ)χj2 (θ) = ∫ dθ ( cos(j1 − j2 )θ − cos(j1 + j2 + 1)θ) = δj1 ,j2 ,
π 0 2π 0
2j1 , 2j2 = 0, 1, 2, . . . . (3.133)

Furthermore corresponding to (2.61)

∞
′ 1 ′ ′
(ei 2 n(θ−θ ) − ei 2 n(θ+θ ) )
1 1
∑ χj (θ)χj (θ ) = 2 1 ∑
j=0, 21 ,1,...
4 sin 2 θ n=−∞
∞
π π
= 2 1 ∑ (δ(θ − θ′ − 4nπ) − δ(θ + θ′ − 4nπ)) = δ(θ − θ′ ) , 0 < θ, θ′ < 2π . (3.134)
sin 2 θ n=−∞ sin2 21 θ

For SO(3), when j, j ′ are integral, the integration range may be reduced to [0, π] with the
coefficient on the right hand side of (3.133) is halved. There is a corresponding modification
in (3.134) and it is necessary to restrict 0 < θ, θ′ < π to get a single δ-function.
In addition since

χj (2θ) = ∑jm=−j ei m 2θ = ∑2j

J=0 (−1)
2j−J
χJ (θ) , (3.135)

then
1 2π
2 1 2j
∫ dθ sin 2 θ χj (2θ) = (−1) , (3.136)
π 0
since only χ0 (θ) = 1 survives after integration. By virtue of (2.63) this shows that the
representations are real for j integral, pseudo-real for j half integral.

3.6 Tensor Products and Angular Momentum Addition

The representation space Vj , which has the orthonormal basis {∣j m⟩}, determines an irre-
ducible representation of SU (2) and also the commutation relations (3.54) of the generators
or physically the angular momentum operators. The tensor product Vj1 ⊗ Vj2 of two repre-
sentation spaces Vj1 , Vj2 has a basis

∣j1 m1 ⟩1 ∣j2 m2 ⟩2 . (3.137)

Associated with Vj1 , Vj2 there are two independent angular operators J1 , J2 both satisfying
the commutation relations (3.54)

[J1,i , J1,j ] = iεijk J1,k ,

[J2,i , J2,j ] = iεijk J2,k . (3.138)

76
They may be extended to act on Vj1 ⊗ Vj2 since with the basis (3.137)

J 1 ≡ J 1 ⊗ 12 , J1 (∣j1 m1 ⟩1 ∣j2 m2 ⟩2 ) = J1 ∣j1 m1 ⟩1 ∣j2 m2 ⟩2 ,

J 2 ≡ 11 ⊗ J 2 , J2 (∣j1 m1 ⟩1 ∣j2 m2 ⟩2 ) = ∣j1 m1 ⟩1 J2 ∣j2 m2 ⟩2 . (3.139)

With this definition it is clear that they commute

[J1,i , J2,j ] = 0 . (3.140)

The generator for the tensor product representation, or the total angular momentum oper-
ator, is then defined by
J = J1 + J2 . (3.141)
It is easy to see that this has the standard commutation relations (3.54).
In the space Vj1 ⊗ Vj2 we may construct states which are standard basis states for the
total angular momentum ∣JM ⟩ labelled by the eigenvalues of J2 , J3 ,

J3 ∣JM ⟩ = M ∣JM ⟩ ,
J2 ∣JM ⟩ = J(J + 1)∣JM ⟩ . (3.142)

These states are chosen to be orthonormal so that

⟨J ′ M ′ ∣JM ⟩ = δJ ′ J δM ′ M , (3.143)

and satisfy (3.77). All states in Vj1 ⊗ Vj2 must be linear combinations of the basis states
(3.137) so that we may write

∣JM ⟩ = ∑ ∣j1 m1 ⟩1 ∣j2 m2 ⟩2 ⟨j1 m1 j2 m2 ∣JM ⟩ . (3.144)

m1 ,m2

Here
⟨j1 m1 j2 m2 ∣JM ⟩ , (3.145)
are Clebsch-Gordan coefficients26 .
As J3 = J1,3 + J2,3 Clebsch-Gordan coefficients must vanish unless M = m1 + m2 . To
determine the possible values of J it is sufficient to find all highest weight states ∣JJ⟩ in
Vj1 ⊗ Vj2 such that
J3 ∣JJ⟩ = J∣JJ⟩ , J+ ∣JJ⟩ = 0 . (3.146)
We may then determine the states ∣JM ⟩ by applying J− as in (3.80). There is clearly a
unique highest weight state with J = j1 + j2 given by

∣j1 +j2 j1 +j2 ⟩ = ∣j1 j1 ⟩1 ∣j2 j2 ⟩2 , (3.147)

so that
⟨j1 j1 j2 j2 ∣j1 + j2 j1 + j2 ⟩ = 1 . (3.148)
26
Rudolf Friedrich Alfred Clebsch, 1833-1872, German. Paul Albert Gordan, 1837-1912, German.

77
From ∣j1 +j2 j1 +j2 ⟩ states ∣j1 +j2 M ⟩ for any M are obtained as in (3.80). Using J−n =
n
∑nr=0 ( r ) J1−r J2−n−r for n = 1, . . . j1 + j2 we may then derive
1
(2j1 )!(2j2 )! (J − M )!(J + M ) 2
⟨j1 m1 j2 m2 ∣J M ⟩ = ( ) ,
(2J)! (j1 − m1 )!(j1 + m1 ))!(j2 − m2 )!j2 + m2 ))!
J = j1 + j2 , M = m1 + m2 . (3.149)

Clearly as well as (3.148) ⟨j1 −j1 j2 −j2 ∣j1 +j2 −(j1 +j2 )⟩ = 1. We may then construct the states
∣JM ⟩ for J = j1 + j2 − 1, . . . iteratively. Defining V (M ) ⊂ Vj1 ⊗ Vj2 to be the subspace for
which J3 has eigenvalue M then, since it has a basis as in (3.137) for all m1 + m2 = M , we
have, assuming j1 ≥ j2 , dim V (M ) = j1 + j2 − M + 1 for M ≥ j1 − j2 and dim V (M ) = 2j2 + 1
for M ≤ j1 − j2 . Assume all states ∣J ′ M ⟩ have been found as in (3.144) for j1 + j2 ≥ J ′ > J.
For j1 + j2 > J ≥ j1 − j2 there is a one dimensional subspace in V (J) which is orthogonal to
all states ∣J ′ J⟩ for J < J ′ ≤ j1 + j2 . This subspace must be annihilated by J+ , as otherwise
there would be too many states with M = J + 1, and hence there is a highest weight state
∣JJ⟩. In constructing a normalised ∣JJ⟩ in terms of a real linear combination of the states
∣j1 m1 ⟩1 ∣j2 m2 ⟩2 , J = m1 + m2 there is an overall choice of sign, conventionally the coefficient
for the largest m1 is positive. If M < j1 − j2 it is no longer possible to construct further
highest weight states. Hence we have shown, since the results must be symmetric in j1 , j2 ,
that in Vj1 ⊗ Vj2 there exists exactly one vector subspace VJ , of dimension (2J + 1), for each
J-value in the range

J ∈ {j1 + j2 , j1 + j2 − 1, . . . , ∣j1 − j2 ∣ + 1, ∣j1 − j2 ∣} , (3.150)

or
j1 +j2
Vj1 ⊗ Vj2 = ⊕ VJ . (3.151)
J=∣j1 −j2 ∣

If j1 ≥ j2 we can easily check that

j1 +j2 j1 +j2
∑ (2J + 1) = ∑ ((J + 1)2 − J 2 )
J=j1 −j2 J=j1 −j2

= (j1 + j2 + 1)2 − (j1 − j2 )2 = (2j1 + 1)(2j2 + 1) , (3.152)

so that the basis {∣JM ⟩} has the correct dimension to span the vector space Vj1 ⊗ Vj2 . The
construction of ∣JM ⟩ states described above allows the Clebsch-Gordan coefficients to be
iteratively determined starting from J = j1 + j2 and then progressively for lower J as in
(3.150). By convention they are chosen to be real and for each J there is a standard choice
of the overall sign. With the common conventions

⟨j1 m1 j2 m2 ∣JM ⟩ = (−1)j1 +j2 −J ⟨j2 m2 j1 m1 ∣JM ⟩ ,

= (−1)j1 +j2 −J ⟨j1 −m1 j2 −m2 ∣J −M ⟩ . (3.153)

The first arises since interchanging j1 and j2 changes the overall sign whenever j1 + j2 − J
is odd, the second since construction of Clebsch-Gordan coefficients can equally be given
starting from ∣j1 +j2 −j1 −j2 ⟩ = ∣j1 −j1 ⟩1 ∣j2 −j2 ⟩2 instead of (3.147) but the sign prescription
changes as previously.

78
 For j1 = j2 = j the decomposition of the tensor product Vj ⊗Vj in (3.151) can be separated
into contributions which are symmetric or antisymmetric under interchange
j j−1
j integral, ⋁2 Vj = (Vj ⊗ Vj )sym = ⊕ V2n , ⋀2 Vj = (Vj ⊗ Vj )antisym = ⊕ V2n+1 ,
n=0 n=0
j− 12 j− 21
j half integral, ⋁2 Vj = ⊕ V2n+1 , ⋀2 Vj = ⊕ V2n , (3.154)
n=0 n=0

As a check for j integral the total dimensions of the symmetric and antisymmetric subspaces
in Vj ⊗ Vj are then ∑jn=0 (4n + 1) = (j + 1)(2j + 1), ∑j−1
n=0 (4n + 3) = j(2j + 1).

Since the original basis (3.137) and {∣JM ⟩} are both orthonormal we have the orthog-
onality/completeness conditions
′ ′
∑ ⟨j1 m1 j2 m2 ∣JM ⟩⟨j1 m1 j2 m2 ∣J M ⟩ = δJJ ′ δM M ′ ,
m1 ,m2
′ ′
∑ ⟨j1 m1 j2 m2 ∣JM ⟩⟨j1 m 1 j2 m 2 ∣JM ⟩ = δm1 m′1 δm2 m′2 . (3.155)
JM

Together with, from applying J± to (3.144),

±
NJM ⟨j1 m1 j2 m2 ∣J M ±1⟩ = Nj±1 m1 −1 ⟨j1 m1 ∓1 j2 m2 ∣JM ⟩ + Nj±2 m2 −1 ⟨j1 m1 j2 m2 ∓1∣JM ⟩ ,
M = m1 + m2 ∓ 1 , (3.156)

these determine all Clebsch-Gordan coefficients up to a choice of sign. Using (3.155) (3.144)
can be inverted
∣j1 m1 ⟩1 ∣j2 m2 ⟩2 = ∑ ∣JM ⟩ ⟨j1 m1 j2 m2 ∣JM ⟩ . (3.157)
J,M

For the tensor product representation defined on the tensor product space Vj1 ⊗ Vj2
we may use the Clebsch-Gordan coefficients as in (3.144) to give the decomposition into
irreducible representations for each J allowed by (3.150)
(j ) (j ) ′ ′ ′ ′ (J)
∑ ∑ Dm11′ m1 (R) Dm22′ m2 (R) ⟨j1 m1 j2 m2 ∣J M ⟩⟨j1 m1 j2 m2 ∣JM ⟩ = δJ ′ J DM ′ M (R) .
m1′ ,m1 m2′ ,m2
(3.158)
For rotation matrices expressed in terms of Euler angles as in (3.99) the dependence on
(j) (j)
φ, ψ factorises and the relation holds when Dm′ m (R) → dm′ m (θ). With the aid of the
orthogonality relations (3.155) this can be rewritten as
(j ) (j ) (j)
∑ dm11′ m1 (θ) dm22′ m2 (θ) ⟨j1 m1′ j2 m2′ ∣jm′ ⟩ = dm′ m1 +m2 (θ) ⟨j1 m1 j2 m2 ∣j m1 +m2 ⟩ . (3.159)
m1′ +m2′ =m′

3.7 Examples of the calculation of Clebsch-Gordan coefficients

A simple example where the Clebsch-Gordan coefficients can be quite easily calculated is
when j1 = j and j2 = 21 so that there are 2(2j + 1) states altogether. In this case the
states can be combined using Clebsch-Gordan coefficients to form states with total angular

79
momentum J = j ± 12 and the states ∣JM ⟩ are just linear combinations of ∣j m⟩1 and ∣ 12 ± 12 ⟩2
with m = M ∓ 12 .
In general then
∣j+ 12 m+ 12 ⟩ = am ∣j m⟩1 ∣ 12 12 ⟩2 + bm ∣j m+1⟩1 ∣ 12 − 21 ⟩2 . (3.160)
As a consequence of (3.147) we must have
aj = 1 , bj = 0 . (3.161)
Applying J± to (3.160) we may directly obtain recurrence relations for am−1 in tems of am
and bm+1 in terms of bm where using (3.77) and (3.79)
1 1
j+m 2 j−m−1 2
am−1 = ( ) am , bm+1 = ( ) bm . (3.162)
j+m+1 j−m
These are easily solved
1 1
j+m+1 2 j−m 2
am = ( ) , bm = ( ) , (3.163)
2j + 1 2j + 1
where the normalisation of am is determined from (3.161). To check the normalisation of
bm it is sufficient to note that applying J− to (3.160) also gives
1 1
1 2
j−m 2
bm−1 = ( ) am + ( ) bm , (3.164)
(j + m + 1)(j − m + 1) j−m+1
which is satisfied by (3.163). Clearly a−j−1 = 0, b−j−1 = 1 so that ∣j+ 21 −j− 12 ⟩ = ∣j −j⟩1 ∣ 21 − 12 ⟩2 .
Also am2 + bm2 = 1 which is necessary for ∣j+ 12 m+ 12 ⟩ to be normalised.
1
The corresponding states with J = j − 2 are orthogonal to the states defied in (3.160).
In this case it is sufficient to take
∣j− 12 m+ 12 ⟩ = −bm ∣j m⟩1 ∣ 12 12 ⟩2 + am ∣j m+1⟩1 ∣ 12 − 12 ⟩2 , (3.165)
where here m = j − 1, j − 2, . . . , −j + 1. This result is unique to within an overall phase which
we have taken in accordance with the so-called Condon and Shortley phase convention. We
may directly verify that J+ ∣j− 12 j− 12 ⟩ = 0 and that acting repeatedly with J− respects the
conventions relating ∣j− 21 M −1⟩ to ∣j− 21 M ⟩.
In the end the Clebsch-Gordan coefficients are then
√ √
⟨jm 21 12 ∣j+ 21 m+ 12 ⟩ = j+m+1
2j+1 , ⟨jm 1 1
2 2− ∣j+ 1
2 m− 1
2 ⟩ = j−m+1
2j+1 ,
√ √
j−m j+m
⟨jm 21 12 ∣j− 21 m+ 12 ⟩ = − 2j+1 , ⟨jm 21 − 12 ∣j− 12 m− 12 ⟩ = 2j+1 , (3.166)
√
2j
The Condon and Shortley convention requires taking ⟨jj 12 − 21 ∣j− 12 j− 12 ⟩ = 2j+1 > 0.

By combining the results in (3.166) with (3.159) for j1 = j − 12 , j2 = 1

2 and using (3.105)
we may obtain
(j) (j− 1 ) (j− 1 )
(j ± m) 2 dm′ m (θ) = (j ± m′ ) 2 cos 12 θ dm′ ∓21 m∓ 1 (θ) ± (j ∓ m′ ) 2 sin 12 θ dm′ ±21 m∓ 1 (θ) . (3.167)
1 1 1

2 2 2 2

80
3.7.1 Construction of Singlet States

A special example of decomposition of tensor products is the construction of the singlet

states ∣0 0⟩, which corresponds to the one-dimensional trivial representation and so is in-
variant under rotations. For Vj1 ⊗ Vj2 , as is clear from (3.150) this is only possible for
j1 = j2 = j and the singlet state must have the general form

∣0 0⟩ = ∑ am ∣j m⟩1 ∣j −m⟩2 . (3.168)

m

Requiring J+ ∣0 0⟩ = 0 gives am = −am−1 so that, imposing the normalisation condition,

2j
1 n 1
∣0 0⟩ = √ ∑ (−1) ∣j j−n⟩1 ∣j −j+n⟩2 ⇒ ⟨jm j−m∣00⟩ = √ (−1)j−m . (3.169)
2j + 1 n=0 2j + 1

Note that ∣0 0⟩ is symmetric, antisymmetric under 1 ↔ 2 according to whether 2j is even,

odd.

3.7.2 Construction of Highest Weight States

The construction of ∣0 0⟩ can be generalised to find all highest weight states ∣J J⟩ contained
in Vj1 ⊗ Vj2 . These have the form

∣J J⟩ = ∑ ∣j1 m1 ⟩1 ∣2 j m2 ⟩2 ⟨j1 m1 j2 m2 ∣JJ⟩ . (3.170)

m1 +m2 =J

Requiring J+ ∣J J⟩ = 0 gives

Nj+1 m1 −1 ⟨j1 m1 −1 j2 m2 ∣JJ⟩ + Nj+2 m2 −1 ⟨j1 m1 j2 m2 −1∣JJ⟩ = 0 , (3.171)

which leads to
1

j1 −m1 (j1 + m1 )! (j2 + m2 )! 2

⟨j1 m1 j2 m2 ∣JJ⟩ = (−1) ( ) Aj 1 j 2 J , m1 + m2 = J . (3.172)
(j1 − m1 )! (j2 − m2 )!

For normalisation we require27

j1
(j1 + m1 )! (j2 + J − m1 )!
∑ Aj1 j2 J 2
(j
m1 =J−j2 1 − m1 )! (j 2 − J + m1 )!
(j1 − j2 + J)! (j2 − j1 + J)! (j1 + j2 + J + 1)!
= Aj1 j2 J 2 = 1 . (3.173)
(j1 + j2 − J)! (2J + 1)!

where Aj1 j2 J = Aj2 j2 J is given by the positive square root and ∣j1 − j2 ∣ ≤ J ≤ j1 + j2 .
(K−n)!(L+n)!
27
The required summation is obtained by using, for K ≥ p, L ≥ 0, ∑pn=0 n!(p−n)!
= L!(K−p)!(K+L+1)!
p! (K+L−p+1)!
,
here K − p = j2 − j1 + J, L = j1 − j2 + J, p = j1 + j2 − J.

81
3.7.3 Special Cases of Clebsch-Gordan Coefficients

A very similar discussion to that just given can be applied to obtain an expression for
⟨j1 j1 j2 m2 ∣JM ⟩. Applying J+ to (3.157) gives a two term relation for this Clebsch-Gordan
coefficient which requires
1
(j2 − m2 )! (J + M )! 2
⟨j1 j1 j2 m2 ∣JM ⟩ = ( ) Bj1 j2 J , j1 + m2 = M . (3.174)
(j2 + m2 )! (J − M )!
Bj1 j2 J can be determined in terms of Aj1 j2 J , as given by (3.173), by taking M = J and
comparing with (3.172) for m1 = j1 . This gives
1
1 (2j1 )! (j2 − j1 + J)! 2
Bj1 j2 J = (2J + 1) (
2 ) . (3.175)
(j1 + j2 − J)! (j1 − j2 + J)! (j1 + j2 + J + 1)!

Results for Clebsch-Gordan coefficients may be derived by successively applying J− to

(3.170) with (3.172). The expressions thereby obtained in general can not be reduced to a
However applying J−J to (3.170) and using from (3.80), for integer j and m ≥ 0
single term. √
(j+m)!
, J−m ∣jm⟩ = (j−m)! ∣j 0⟩ then for j1 , j2 , J integers

J
√ J (j1 + m1 )! (j2 + m2 )!
(2J)! ⟨j1 0 j2 0∣J 0⟩ = ∑ ( ) (−1)j1 −m1 Aj j J
m1 =0 m1 (j1 − m1 )! (j2 − m2 )! 1 2
= (−1)j1 −J ∏Jr=1 (j1 − j2 − J − 1 + 2r)(j1 + j2 − J + 2r) Aj1 j2 J
⎧
⎪0 ,
⎪
⎪ j1 + j2 + J odd
=⎨ (j −j +J)! (j −j +J)! ( 1
(j +j +J))!
(−1) 2 (j1 +j2 −J) ( 1 (j −j +J))! ( 1 (j −j +J))!
1 1 2 2 1 2 1 2
⎪
⎪
⎪ A
( 21 (j1 +j2 −J))! j1 j2 J
, j1 + j2 + J even .
⎩ 2 1 2 2 2 1

(3.176)

3.8 3j Symbol

Besides (3.153) there are further symmetry relations for Clebsch-Gordan coefficients. Less
obviously
⟨j1 m1 j2 m2 ∣JM ⟩ = Cj1 j2 J (−1)j1 −m1 ⟨jm1 J −M ∣j2 −m2 ⟩ , m1 + m 2 = M . (3.177)
The dependence on m1 , m2 , M is dictated by the recurrence relations (3.156). Applying
±
these to (3.177), both sides agree as a consequence of Njm = Nj±−m±1 and where (−1)j1 −m1
provides a necessary sign flip in one term when m1 → m1 ± 1. The overall coefficient Cj1 j2 J
is determined by setting m1 = j1 and using (3.174) so that
1
Bj j J 2J + 1 2
Cj1 j2 J = 1 2 =( ) . (3.178)
Bj1 Jj2 2j2 + 1

Defining a 3j symbol by
j j2 j3 (−1)j1 −j2 −m3
( 1 )= √ ⟨j1 m1 j2 m2 ∣j3 −m3 ⟩ , m1 + m2 + m3 = 0 , (3.179)
m1 m2 m3 2j3 + 1

82
then (3.153) and (3.177) ensure

j1 j2 j3 j j1 j3 j j3 j2
( ) = (−1)j1 +j2 +j3 ( 2 ) = (−1)j1 +j2 +j3 ( 1 ). (3.180)
m1 m2 m3 m2 m1 m3 m1 m3 m2

These further entail that the 3j symbols are invariant under cyclic permutations and also
for mi → −mi from (3.153) it follows that

j1 j2 j3 j j2 j3
( ) = (−1)j1 +j2 +j3 ( 1 ). (3.181)
m1 m2 m3 −m1 −m2 −m3

Hence, although equivalent to Clebsch-Gordan coefficients, 3j symbols are much nicer. They
are non zero only if j1 , j2 , j3 satisfy triangular inequalities ∣j1 −j2 ∣ ≤ j3 ≤ j1 +j2 and j1 +j2 +j3
1 1
is an integer. These symmetries are reflected by AJj2 j1 /(2j1 + 1) 2 = Bj1 j2 J /(2J + 1) 2 =
1
Bj1 Jj2 /(2j2 + 1) 2 .
Many results for 3j symbols can be simply expressed diagrammatically

j2 ,m2
j j2 j3
( 1 ) = j3 ,m3 , (3.182)
m1 m2 m3
j1 ,m1

where the diagram is invariant under rotation and for a non zero result ∑i mi = 0. For later
convenience we also define
j2 ,m2
j1 j2 j3
j3 ,m3 = (−1)j3 +m3 ( ), (3.183)
m1 m2 −m3
j1 ,m1

and similarly for other lines. With this convention from (3.181) = .

The orthogonality conditions in (3.155), using (3.181), can then be reexpressed in the
form
(ji +mi ) j j2 j3 j1′ j2 j3
∑m2 ,m3 (−1)∑ ( −m1
1 −m2 −m3
) ( m1′ m2 m3
) = 1
2j1 +1 δj1 j1 δm1 m1
′ ′ ,
j2 ,m2

j1′ ,m1′ 1
∑m2 ,m3 j1 ,m1 = 2j1 +1 j1 ,m1 j1′ ,m1′ ,
j3 ,m3

(ji +mi ) j j2 j3 j j j
∑j3 ,m3 (2j3 + 1)(−1)∑i ( −m1
1 −m2 −m3
) ( m1 ′ m2 ′ m3 )
1 2 3
= δm1 m1′ δm2 m2′ ,
j2 ,m2 j1 ,m1′
j1 ,m1 j1 ,m1′
∑j3 ,m3 (2j3 + 1) = . (3.184)
j3 ,m3 j2 ,m2 j2 ,m2′
j1 ,m1 j2 ,m2′

For each summation ∑m (−1)j+m ∣j−m⟩∣jm⟩ generates an angular momentum singlet.

83
 The essential normalisation of the 3j symbols is then diagrammatically expressed as
j1 ,m1

∑i (ji +mi ) j j j j j j j2 ,m2

∑mi (−1) ( −m1 −m2 −m3 ) ( m1 m2 m3 ) = ∑m i = 1. (3.185)
1 2 3 1 2 3 j3 ,m3

As a special case from (3.173) and (3.176), for j1 , j2 , j3 integers,

1
j j j (2k1 )! (2k2 )! (2k3 )! 2 J!
( 1 2 3 ) = (−1)J ( ) , j1 +j2 +j3 = 2J , ki = J −ji , (3.186)
0 0 0 (2J + 1)! k1 ! k2 ! k3 !
for J an integer.
If we consider the tensor product space Vj1 ⊗ Vj2 ⊗ Vj3 , then so long as j1 , j2 , j3 obey the
required conditions, we may form a singlet state by

j j2 j3
∣0 0⟩ = ∑ ( 1 ) ∣j1 m1 ⟩1 ∣j2 m2 ⟩2 ∣j3 m3 ⟩3 . (3.187)
m1 ,m2 ,m3 m 1 m 2 m3

This is a singlet since, coupling first ∣j1 m1 ⟩1 ∣j2 m2 ⟩2 to form a state ∣j3 −m3 ⟩, we have
j j2 j3
( 1 ) = (−1)j1 −j2 +j3 ⟨j1 m1 j2 m2 ∣j3 −m3 ⟩⟨j3 −m3 j3 m3 ∣00⟩ using (3.169).
m1 m2 m3

3.9 6j Symbol

3j symbols can be combined to form invariants which can be represented by vacuum graphs
with trivalent vertices. The simplest defines the normalisation as in (3.185). The graph
with 4 vertices defines the 6j symbol as in
j1 j2 j3
} = ∑ (−1)∑i (ji +mi +ki +ni ) ( −m1
j k2 k3 k j k k k j j j j
{ n2 −n3
) ( −n1 −m2 n3 ) ( n1 −n2 −m3 ) ( m1 m2 m3 )
k1 k2 k3 1 1 2 3 1 2 3 1 2 3
mi ,ni

k2 ,n2

j1 ,m1 j3 ,m3
j1 ,m1 j3 ,m3 j2 ,m2
= j2 ,m2
k1 ,n1 or , (3.188)
k2 ,n2
k3 ,n3 k1 ,n1
k3 ,n3

where the sums over ni , mi are constrained by the 4 linear conditions necessary for non
vanishing 3j symbols. Due to these constraints there are just two non trivial summations.
This is non zero so long as ji , ki satisfy the triangle inequalities corresponding to them being
the edge lengths of the dual tetrahedron, obtained in the above picture by tasking ji ↔ ki
for each i, in Euclidean space and the sum of the three ji , ki incident at each vertex is an
integer. Since ∑i mi = 0 in any non zero contribution the associated sign factor can be
dropped. Using the symmetries in (3.180) and (3.181)
j1 j2 j3 j2 j1 j3 j3 j1 j2 k1 k2 j3
{ }={ }={ }={ }. (3.189)
k1 k2 k3 k2 k1 k3 k3 k1 k2 j1 j2 k 3

84
so that the 6j symbol is invariant under permutations of columns and interchanging the
upper and lower elements of any two columns. These correspond to any permutations of
vertices in the figure in (3.188) so they generate the tetrahedral symmetry group S4 .
Directly from (3.188)

j1′ j2 j3
(−1)∑i (ji +mi +ki +ni ) ( −m1
j k2 k3 k j k k k j
∑ −n2 n3
) ( n1 −m2 −n3 ) ( −n1 n2 −m3 ) ( ′ )
1 1 2 3 1 2 3 m1 m2 m3
ni , m1 fixed
j1 j2 j3
={ } 2j11+1 δj1 j1′ δm1 m1′ ,
k1 k2 k3
k2 ,n2 j3 ,m3
j1 j2 j3
∑ j1 ,m1 k1 ,n1 j1′ ,m1′ ={ } 2j11+1 j1 ,m1 j1′ ,m1′ . (3.190)
k1 k2 k3
ni , m1 fixed
k3 ,n3 j2 ,m2

′
Multiplying with (2j1 + 1)(−1)j1 +m1 ( m
j1
′
j2 j3
m2′ m3′
) and summing j1′ , m1′ , using (3.184), gives
1
the vertex relation
j1 j2 j3
(−1)∑i (ki +ni ) ( m1
j k2 k3 k j k k k j j j2 j3
∑ n2 −n3
) ( −n1 m2 n3 ) ( n1 −n2 m3 ) ={ } ( m1 m2 m3
) ,
1 1 2 3 1 2 3 k1 k2 k3 1
ni , mi fixed
j2 ,m2
j2 ,m2
k3 ,n3
j1 j2 j3
∑ (−1)∑i (ki +ni ) j1 ,m1 k1 ,m1 ={ } j1 ,m1 . (3.191)
k2 ,n2 k1 k2 k3
ni , mi fixed j3 ,m3
j3 ,m3

By using the orthogonality relations further this leads to the crossing relations
j1 j4 J j3 j2 J ′ +M ′ j1 j2 J ′ j3 j4 J′
(−1)J+M ( )( ) = ∑ XJJ ′ (−1)J ( )( ),
m1 m4 −M m3 m2 M m1 m2 M′ m3 m4 −M ′
J′
j1 j2 J
M = m1 + m4 , M ′ = m3 + m4 , ∑i mi = 0 , XJJ ′ = (2J ′ + 1)(−1)2j4 { }, (3.192)
j3 j4 J ′

where the summation over J ′ is constrained by ji , J, J ′ forming a tetrahedron, J ′ ≥ ∣M ′ ∣ and

∑4i=1 ji is an integer. Diagrammatically

j3 ,m3 j2 ,m2 j2 ,m2 j3 ,m3

J ′ ,M ′
J,M = ∑ XJJ ′ . (3.193)
J′ j1 ,m1 j4 ,m4
j4 ,m4 j1 ,m1

The 6j symbols play a crucial role in the decomposition of three spins. Tensor products
are associative so that Vj1 ⊗ (Vj2 ⊗ Vj3 ) ≃ (Vj1 ⊗ Vj2 ) ⊗ Vj3 but different bases for given total
angular momentum j4 are formed according to whether we decompose Vj2 ⊗ Vj3 = ⊕J VJ or
Vj1 ⊗ Vj2 = ⊕J ′ VJ ′ . The 6j symbols relate the two bases. Generalisations of 6j symbols can
be obtained by considering further vacuum diagrams with trivalent vertices which are not
reducible using 6j symbol relations.

85
3.9.1 Crossing Relations

The relation (3.193) leads to identities for 6j symbols. We first consider such relations more
generically where we consider couplings between four vectors, labelled by a, b, c, d, which
are expressible diagrammatically, where here the lines do not carry any arrows, as
b c c b a d

J = J = J , (3.194)
a d d a b c

corresponding to a Z2 × Z2 symmetry. The crossing equations then take the form

b c
b c c b
abcd K abcd K
J = ∑ XJK = ∑ YJK . (3.195)
K a d K a d
a d

Consistency with (3.194) requires

abcd badc dcba abcd dcba cdab
XJK = XJK = XJK , YJK = YJK = YJK . (3.196)

Applying (3.195) repeatedly gives the crossing relations

abcd cbad abcd bcad cabd
∑K XJK XKL = δJL , ∑K,L YJK YKL YLM = δJM , (3.197)

with the condition

abcd cbad bacd abcd bcad
∑K,L XJK YKL YLM = ∑K YJK XKM . (3.198)
These ensure X and Y generate S3 corresponding to permutations of a, b, c. For S3 defined
abstractly by elements a, b such that a3 = b2 = e, ba2 = ab it is evident that these relations
correspond to (3.197) and (3.198) with Y ∼ a, X ∼ b.
In terms of 6j symbols these results are applicable by taking
j1 j2 K j1 j3 K
j1 j2 j3 j4
XJK = (2K + 1)(−1)2j4 { j j4 J
}, j1 j2 j3 j4
YJK = (2K + 1)(−1)j2 +j3 +2j4 +J { j j4 J
}.
3 2
(3.199)
(3.196) is modified to
j1 j2 j3 j4 j2 j1 j4 j3 j4 j3 j2 j1
XJK = (−1)2K XJK = (−1)2J XJK ,
j1 j2 j3 j4 j4 j3 j2 j1 j3 j4 j1 j2
YJK = (−1)2J YJK = (−1)2J+2K YJK , (3.200)

and (3.197) reduces to

j1 j2 K j3 j2 L
∑ (2J + 1)(2K + 1) { j j4 J
}{j j4 K
} = δJL ,
3 1
K
J+K+L j1 j3 K j2 j1 L j3 j2 M
∑ (−1) (2J + 1)(2K + 1)(2L + 1) { j j4 J
}{j j4 K
}{j j4 L
} = δJM . (3.201)
2 3 1
K,L

Subject to (3.189) these are equivalent to standard identities for 6j symbols. Furthermore
1 2
∑ YJK
j j j3 j4 j2 j2 j1 j4
XKL = (−1)j2 +j3 +J δJL . (3.202)
K

86
3.10 Tensor Products and Characters

The decomposition of tensor products can equally be determined in terms of the characters
given in (3.132)
j2 j2
1
χj1 (θ) χj2 (θ) = χj1 (θ) ∑ e−imθ = (e(j1 +m+ 2 )θ − e(−j1 +m+ 2 )θ )
1 1

1 ∑
m=−j2 2i sin 2 θ m=−j2
j1 +j2 j1 +j2
= ∑ χj (θ) = ∑ χj (θ) , (3.203)
j=j1 −j2 j=∣j1 −j2 ∣

where if j2 > j1 we use χ−j (θ) = −χj−1 (θ) to show all contributions to the sum for j < j2 − j1
cancel. Comparing with (2.85) the result of this character calculation of course matches the
tensor product decomposition given in (3.152).
For the symmetric and antisymmetric tensor products of the j representation from (2.92)
⌊j⌋
χ⋁2 V (θ) = 21 (χj (θ)2 + χj (2θ)) = ∑ χ2j−2n (θ) ,
j
n=0
⌊j⌋
χ⋀2 V (θ) = 21 (χj (θ)2 − χj (2θ)) = ∑ χ2j−1−2n (θ) , (3.204)
j
n=0

using (3.135). This of course agrees with (3.154).

The results can be extended to three fold tensor products
j1 +j2 +j3
χj1 (θ) χj2 (θ) χj2 (θ) = ∑ NJ χJ (θ) ,
J=j1 −j2 −j3

NJ = min{2j2 + 1, 2j3 + 1, j1 + j2 + j3 − J + 1, j2 + j3 − j1 + J + 1} . (3.205)

Although not manifestly symmetric cancellations arising from terms in the sum for J < 0
ensure it is so. For equal integer j
j 3j
χj (θ)3 = ∑ (2J + 1)χJ + ∑ (3j − J + 1)χJ . (3.206)
J=0 J=j+1

For j half integer a similar result obtains but with the first sum starting at J = 12 . From
(2.92)
χ (θ)
⋁3 Vj
χ (θ) } = 61 (χVj (θ)3 ± 3 χVj (θ)χVj (2θ) + 2 χVj (3θ)) , (3.207)
⋀2 V j

giving, for j an integer,

2j j−1 ⌊ 31 (j−1)⌋ ⌊ 12 (j−1)⌋
χ⋁3 V = ∑ (1 + ⌊ 61 k⌋)χ3j−k + ∑ (1 + ⌊ 31 k⌋)χk − ∑ χ3j−6k−1 − ∑ χj−2k−1 , j ≥ 1 ,
j
k=0 k=0 k=0 k=0
2j−2 j−2 ⌊ 13 (j−2)⌋ ⌊ 21 (j−2)⌋
χ⋀3 V = ∑ (1 + ⌊ 61 k⌋)χ3j−k−3 + ∑ (1 + ⌊ 13 k⌋)χk − ∑ χ3j−6k−4 − ∑ χj−2k−2 , j ≥ 2 ,
j
k=0 k=0 k=0 k=0
(3.208)

87
while χ⋀3 V = χ0 . For half integer j
1

2j j−1 ⌊ 13 j⌋
χ⋁3 V =
j
∑ (1 + ⌊ 61 k⌋)χ3j−k + ∑ (1 + ⌊ 31 (k − 1)⌋)χk − ∑ χ3j−6k−1 , j≥ 5
2 ,
k=0 k= 32 k=0
2j−3 j−1 ⌊ 31 (j−2)⌋
χ⋀3 V = ∑ (1 + ⌊ 61 k⌋)χ3j−k−3 + ∑ (1 + ⌊ 13 (k − 1)⌋)χk − ∑ χ3j−6k−4 , j≥ 5
2 , (3.209)
j
k=0 k= 32 k=0

with, for j = 32 , χ⋁3 Vj = χ 3 + χ 5 + χ 9 , χ⋀3 V = χ 3 and for j = 12 , χ⋁3 Vj = χ 3 , χ⋀3 V = 0.

2 2 2 j 2 2 j

3.11 SO(3) Tensors

In the standard treatment of rotations vectors and tensors play an essential role. For
R = [Rij ] and SO(3) rotation then a vector is required to transform as

Vi → V ′i = Rij Vj . (3.210)
R

Vectors then give a three dimensional representation space V. A rank n tensor Ti1 ...in is
then defined as belonging to the n-fold tensor product V ⊗ ⋅ ⋅ ⋅ ⊗ V and hence satisfy the
transformation rule
Ti1 ...in → T ′i1 ...in = Ri1 j1 . . . Rin jn Tj1 ...jn . (3.211)
R

It is easy to see the dimension of the representation space, V(⊗V)n−1 , formed by rank n
tensors, is 3n . For n = 0 we have a scalar which is invariant and n = 1 corresponds to
a vector. The crucial property of rotational tensors is that they be multiplied to form
tensors of higher rank, for two vectors Ui , Vi then Ui Vj is a rank two tensor, and also that
contraction of indices preserves tensorial properties essential because for any two vectors
Ui Vi is a scalar and invariant under rotations, U ′i V ′i = Ui Vi . The rank n tensor vector space
then has an invariant scalar product T ⋅ S formed by contracting all indices on any pair of
rank n tensors Ti1 ...in , Si1 ...in .
In tensorial analysis invariant tensors, satisfying I ′i1 ...in = Ii1 ...in , are of critical impor-
tance. For rotations we have the Kronecker delta δij

δ ′ij = Rik Rjk = δij , (3.212)

as a consequence of the orthogonality property (3.1), and also the ε-symbol

ε′ijk = Rij Rjm Rkn εlmn = det R εijk = εijk , (3.213)

if R ∈ SO(3). Any higher rank invariant tensor is formed in terms of Kronecker deltas and
ε-symbols, for rank 2n we may use n Kronecker deltas and for rank 2n + 3, n Kronecker
deltas and one ε-symbol, since two ε-symbols can always be reduced to combinations of
Kronecker deltas.
Using δij and εijk we may reduce tensors to ones of lower rank. Thus for a rank two
tensor Tij , Tii = δij Tij , which corresponds to the trace of the associated matrix, is rank zero

88
and thus a scalar, and Vi = 21 εijk Tjk is a vector. Hence the 9 dimensional space formed
by rank two tensors contains invariant, under rotations, subspaces of dimension one and
dimension three formed by these scalars and vectors. In consequence rank 2 tensors do not
form an irreducible representation space for rotations.
To demonstrate the decomposition of rank 2 tensors into irreducible components we
write it as a sum of symmetric and antisymmetric tensors and re-express the latter as a
vector. Separating out the trace of the symmetric tensor then gives
Tij = Sij + εijk Vk + 31 δij Tkk , (3.214)
for
Sij = T(ij) − 31 δij Tkk , Vi = 21 εijk Tjk . (3.215)
Each term in (3.214) transforms independently under rotations, so that for Tij → T ′ij ,
Sij → S ′ij , Vk → V ′k , Tkk → T ′kk = Tkk . The tensors Sij are symmetric and traceless, Skk = 0,
and it is easy to see that they span a space of dimension 5.
These considerations may be generalised to higher rank but it is necessary to identify
for each n those conditions on rank n tensors that ensure they form an irreducible space. If
Si1 ...in is to be irreducible under rotations then all lower rank tensors formed using invariant
tensors must vanish. Hence we require
δir is Si1 ...in = 0 , εjir is Si1 ...in = 0 , for all r, s, 1 ≤ r < s ≤ n . (3.216)
These conditions on the tensor S are easy to solve, it is necessary only that it is symmetric
Si1 ...in = S(i1 ...in ) , (3.217)
and also traceless on any pair of indices. With the symmetry condition (3.217) it is sufficient
to require just
Si1 ...in−2 jj = 0 . (3.218)
Such tensors then span a space Vn which is irreducible.
To count the dimension of Vn we first consider only symmetric tensors satisfying (3.217),
belonging to the symmetrised n-fold tensor product, sym(V ⊗ ⋅ ⋅ ⋅ ⊗ V). Because of the
symmetry not all tensors are independent of course, any tensor with r indices 1, s indices
2 and t indices 3 will be equal to
S 1...1
2...2 3...3 where r, s, t ≥ 0 , r + s + t = n . (3.219)
±±±
r s t

Independent rank n symmetric tensors may then be counted by counting all r, s, t satisfying
the conditions in (3.219), hence this gives
dim (sym(V ⊗ ⋅ ⋅ ⋅ ⊗ V )) = 12 (n + 1)(n + 2) . (3.220)
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n

To take the traceless conditions (3.218) into account it is sufficient, since taking the trace of
rank n symmetric tensors gives rank n − 2 symmetric tensors spanning a space of dimension
1
2 (n − 1)n, to subtract the dimension for rank n − 2 symmetric tensors giving

dim Vn = 12 (n + 1)(n + 2) − 12 (n − 1)n = 2n + 1 . (3.221)

89
Thus this irreducible space Vn may be identified with the representation space j = n, with
n an integer.
For any rank n symmetric tensor Si1 ...in there is a one to one correspondence with
homogeneous polynomials of degree n in x,

Si1 ...in ↔ S (n) (x) = Si1 ...in xi1 . . . xin , (3.222)

and
Si1 ...in−2 jj = 0 ⇔ ∇2 S (n) (x) = 0 , (3.223)
where S (n) is then a harmonic function. As a particular case we have

S (n) (x) = (t ⋅ x)n for t2 = 0 , (3.224)

where t is any complex null vector. Since from (3.55)

L2 = −x2 ∇2 + x ⋅ ∇(x ⋅ ∇ + 1) . (3.225)

and x ⋅ ∇S (n) (x) = n S (n) (x) we have for harmonic polynomials

L2 S (n) (x) = n(n + 1) S (n) (x) , (3.226)

so that symmetric traceless tensors or harmonic polynomials of degree n correspond to

angular momentum n for any integer n. Clearly S (n) (x) = ∣x∣n S (n) (x̂) for x̂ = x/∣x∣ and as
L∣x∣ = 0 (3.226) reduces to

L2 S (n) (x̂) = n(n + 1) S (n) (x̂) , (3.227)

With two symmetric traceless tensors S1,i1 ...in and S2,i1 ...im then their product can be
decomposed into symmetric traceless tensors by using the invariant tensors δij , εijk , gener-
alising (3.214) and (3.215). Assuming n ≥ m, and using only one ε-symbol since two may
be reduced to Kronecker deltas, we may construct the following symmetric tensors

S1,(i1 ...in−r j1 ...jr S2,in−r+1 ...in+m−2r ) j1 ...jr , r = 0, . . . m ,

εjk(i1 S1,i2 ...in−r j1 ...jr j S2,in−r+1 ...in+m−1−2r ) j1 ...jr k , r = 0, . . . m − 1 . (3.228)

For each symmetric tensor there is a corresponding one which is traceless obtained by
subtracting appropriate combinations of lower order tensors in conjunction with Kronecker
deltas, as in (3.215) for the simplest case of rank two. Hence the product of the two
symmetric tensors of rank n, m decomposes into irreducible tensors of rank n + m − r,
r = 0, 1, . . . , m, in accord with general angular momentum product rules.
In quantum mechanics we may extend the notion of a tensor to operators acting on the
quantum mechanical vector space. For a vector operator we require

U [R]Vi U [R]−1 = (R−1 )ij Vj , (3.229)

as in (3.58), while for a rank n tensor operator

U [R]Ti1 ...in U [R]−1 = (R−1 )i1 j1 . . . (R−1 )in jn Tj1 ...jn . (3.230)

90
These may be decomposed into irreducible tensor operators as above. For infinitesimal
rotations as in (3.16), with U [R] correspondingly given by (3.50), then (3.229) gives

[Ji , Vj ] = i εijk Vk , (3.231)

which is an alternative definition of a vector operator. From (3.230) we similarly get

[Ji , Tj1 j2 ...jn ] = i εij1 k Tkj2 ...jn + i εij2 k Tj1 k...jn + ⋅ ⋅ ⋅ + i εijn k Tj1 j2 ...k . (3.232)

The operators x, p are examples of vector operators for the angular momentum operator
given by L = x × p where [xi , pj ] = iδij .

3.11.1 Spherical Harmonics

Rank n symmetric traceless tensors are directly related to spherical harmonics. If we choose
(n,m)
an orthonormal basis for such tensors Si1 ...in , labelled by m taking 2n+1 values and satisfying
′ ′
S (n,m) ⋅ S (n,m ) ∝ δ mm , then the basis may be used to define a corresponding complete set
of orthogonal spherical harmonics on the unit sphere, depending on a unit vector x̂ ∈ S 2 , by
(n,m)
Ynm (x̂) = Si1 ...in x̂i1 . . . x̂in . (3.233)

For a standard basis we require m is an integer with −n ≤ m ≤ n and

±
L± Ynm (x̂) = Nnm Yn m±1 (x̂) , L3 Ynm (x̂) = m Ynm (x̂) , (3.234)
±
where L± , L3 are the angular momentum operators acting on functions of x̂ and Nnm is
defined in (3.79). Defining

x = x̂1 + ix̂2 , x̄ = x̂1 − ix̂2 , z = x̂3 , xx̄ + z 2 = 1 , (3.235)

then
∂ ∂ ∂ ∂ ∂ ∂
L3 = x − x̄ , L+ = −x + 2z L− = x̄ − 2z . (3.236)
∂x ∂ x̄ ∂z ∂ x̄ ∂z ∂x
In terms of usual spherical polar coordinates z = cos θ, x = sin θ eiφ , x̄ = sin θ e−iφ .
Spherical harmonics can be expressed in the form
⎧
⎪(−x)m pnm (z) ,
⎪ m ≥ 0,
Ynm (x̂) = ⎨ ∣m∣ (3.237)
⎩x̄ pn∣m∣ (z) , m < 0,
⎪
⎪

where pnm (z), 0 ≤ m ≤ n, is a polynomial of degree n − m and Ynm (x̂)∗ = (−1)m Yn−m (x̂)
This expression automatically satisfies the L3 equation in (3.234) and from the L± equations
d + d −
pnm (z) = Nnm pnm+1 (z) , −(1 − z 2 ) pnm (z) + 2m z pnm (z) = Nnm pnm−1 (z) . (3.238)
dz dz
− +
Hence, since Nnm Nnm−1 = (n + m)(n − m + 1),

d d d
( (1 − z 2 ) − 2m z + (n − m)(n + m + 1))pnm (z) = 0 . (3.239)
dz dz dz

91
Regular solutions are given in terms of Gegenbauer polynomials which may be defined in
terms of the generating function
−λ
(1 − 2z r + r2 ) = ∑∞ λ n
n=0 Cn (z) r , Cnλ (−z) = (−1)n Cnλ (z) . (3.240)

Since Cnλ ′ (z) = 2λ Cn−1

λ+1
(z), ((2m − 1)!! = (2m)!/(2m m!) = (2m − 1)(2m − 3) . . . 1),
√
(n−m)! m+ 1
pnm (z) = an (2m − 1)!! (n+m)! Cn−m2 (z) . (3.241)

Spherical harmonics are also expressible in terms of a generating function

et⋅x̂ , t ⋅ x̂ = v(z − 12 λ x + 1
2λ x̄) , (3.242)

with x̂ defined in terms of x, x̄, z as in (3.235). With (3.236) we have

∂ t⋅x̂ 1 ∂ ∂ ∂ ∂
L3 et⋅x̂ = λ e , L+ et⋅x̂ = (v + λ ) et⋅x̂ , L− et⋅x̂ = λ(v − λ ) et⋅x̂ , (3.243)
∂λ λ ∂v ∂λ ∂v ∂λ

then as t, as defined by (3.242), is a null vector so that, for arbitrary v, λ, (t ⋅ x̂)n is a

harmonic function we can expand the exponential in the form

et⋅x̂ = ∑∞ n n m
n=0 ∑m=−n bnm v λ Ynm (x̂) . (3.244)

(3.234) is satisfied so long as

+ −
(n + m + 1)bnm+1 = Nnm bnm , (n − m + 1)bnm−1 = Nnm bnm , (3.245)

or
bn
bnm = √ . (3.246)
(n − m)!(n + m)!

The coefficients bn , or an in (3.241), depend on the normalisation of Ynm for differing

n. It is conventional to impose

∗ m
∫ 2dΩ Ynm (x̂) Yn′ m′ (x̂) = δnn′ δmm′ , or ∫ 2dΩ Ynm (x̂) Yn′ m′ (x̂) = (−1) δnn′ δm−m′ .
S S
(3.247)
On integration over S 2 L3† = L3 , L+† = L− .
The required normalisation to ensure (3.247) can be derived by considering the three
dimensional integrals

3 −x 2 +k⋅x 3 1 2 −x 2 3 (2n)! 2 n
∫ d xe = π 2 e4k ⇒ 3 2n
∫ d x e (k ⋅ x) = π 2 (k ) . (3.248)
22n n!
∞
Since d3 x = r2 dr dΩ and using ∫0 dr r2n+2 e−r = 12 Γ(n + 32 ) we obtain
2

2n (2n)! 1 4π
∫ 2dΩ (k ⋅ x̂) = 4π 2n 3
(k2 )n = (k2 )n , ( 32 )n = 3
2 . 52 . . . ( 12 + n) . (3.249)
S 2 n! ( 2 )n 2n + 1

92
If now k = t1 + t2 with t12 = t22 = 0, so that k2 = 2 t1 ⋅ t2 , then

n n ′ n!2
∫ 2dΩ (t1 ⋅ x̂) (t2 ⋅ x̂) = δnn′ 4π (2 t1 ⋅ t2 )n . (3.250)
S (2n + 1)!

For t1 , t2 as defined in (3.242) and expanding ∫S 2 dΩ et1 ⋅x̂+t2 ⋅x̂

′ ′ 4π
∑ bnm bn′ m′ v n λm v ′n λ′m ∫ dΩ Ynm (x̂) Yn′ m′ (x̂) = ∑ (2 t1 ⋅ t2 )n , (3.251)
n,m,n′ ,m′
2 S n (2n + 1)!

where
v1 v2
2 t1 ⋅ t2 = − (λ1 − λ2 )2 , (3.252)
λ1 λ2
so that m
n
(2n)! λ1
(2 t1 ⋅ t2 )n = (v1 v2 )n ∑ (− ) . (3.253)
m=−n (n − m)!(n + m)! λ2
Hence ensuring (3.247) gives
1
4π 1 4π 2
bnm bn−m = ⇒ bn = ( ) . (3.254)
2n + 1 (n − m)!(n + m)! 2n + 1

1
From (3.244) we have bnn Ynn (x̂) = n! (− 12 x)n while from the solution given by (3.237)
√
and (3.241) Ynn (x̂) = an (2n − 1)!!/ (2n)! (−x)n which requires an bn = 1. The expansion
(3.244), with bnm determined by (3.246) and (3.254), is ascribed to Herglotz.28
These results can be extended to the integral of three spherical harmonics. Taking
k = t1 + t2 + t3 with ti three null vectors
n1 n2 n3
∫ 2dΩ (t1 ⋅ x̂) (t2 ⋅ x̂) (t3 ⋅ x̂) n! n1 ! n2 ! n3 !
S = 4π (2 t1 ⋅ t2 )k3 (2 t2 ⋅ t3 )k1 (2 t1 ⋅ t3 )k2 .
(2n + 1)! k1 ! k2 ! k3 !
n1 + n2 + n3 = 2n , k i = n − ni . (3.255)

We then have

bn1 m1 bn2 m2 bn3 m3 ∫ dΩ Yn1 m1 (x̂) Yn2 m2 (x̂) Yn3 m3 (x̂)

S2
4π n! 3
(−1)bi (2ki )!
= ∑ ∏ ∣ . (3.256)
(2n + 1)! k1 ! k2 ! k3 ! b1 ,b2 ,b3 i=1 (ki − bi )! (ki + bi )! b1 −b2 =m3 ,b2 −b3 =m1 ,b3 −b1 =m2

This is non zero only when ∑i mi = 0. For mi = 0 then bi = b and the sum can be evaluated
using

1 (k1 + k2 + k3 )!
∑ (−1)b = . (3.257)
− min ki ≤b≤min ki ∏3i=1 (ki − b)! (ki + b)! (k1 + k2 )! (k2 + k3 )! (k3 + k1 )! k1 ! k2 ! k3 !
28
Gustav Herglotz, 1881-1953, German.

93
 √
4π
Hence, since bn0 = 2n+1 /n!, and with the result (3.186)
1
2
(2n1 + 1)(2n2 + 1)(2n3 + 1) 2 n1 n2 n3
∫ 2dΩ Yn1 0 (x̂) Yn2 0 (x̂) Yn3 0 (x̂) = ( ) (
0 0 0
) . (3.258)
S 4π
For arbitrary mi , ∑i mi = 0, it then follows that

∫ 2dΩ Yn1 m1 (x̂) Yn2 m2 (x̂) Yn3 m3 (x̂)

S
1
(2n1 + 1)(2n2 + 1)(2n3 + 1) 2 n1 n2 n3 n1 n2 n3
=( ) ( )( ), (3.259)
4π 0 0 0 m1 m2 m3
since the dependence on mi on both sides is identical from (3.234) and the recurrence
identities for 3j symbols following from (3.156).

3.12 Molien Series for SU (2)

The Molien series in (2.100) can be readily extended to the continuous group SU (2) by
replacing the finite sum over group elements by the corresponding invariant integegration.
Since the formula only involves a sum over conjugacy classes in this case it reduces, for the
representation j of dimension 2j + 1, to
1 2π 1
MSU (2) (C2j+1 , t) = ∫ dθ (1 − cos θ) , (3.260)
2π 0 det (1 − t D(θ, n))
where the choice of n is irelevant so that from (3.131) we may use
det (1 − t D(θ, z)) = ∏jm=−j (1 − t ei mθ ) . (3.261)

For integer spin the result reduces to the series for SO(3). In this case with z = eiθ
1 1 1
(j−1)(j+2) (1 − z)(1 − z −1 )
MSO(3) (R2j+1 , t) = ∮ dz z 2
j
. (3.262)
1 − t 4πi ∣z∣=1 ∏m=1 (z m − t)(1 − t z m )
For ∣t∣ < 1 this can be evaluated by summing the residues of the poles arising at z m = t and
also for j = 0, 1 at z = 0. This gives for j = 0, 1, 2
1 1 1
MSO(3) (R, t) = , MSO(3) (R3 , t) = , MSO(3) (R5 , t) = . (3.263)
1−t 1 − t2 (1 − t2 )(1 − t3 )
For (x1 , x2 , x3 ) ∈ R3 the fundamental invariant is just x2 . The representation for j = 2 acts
on the five dimensional space formed by traceless symmetric tensors Tij and the fundamental
primary invariants are tr(T 2 ), tr(T 3 ). For larger integer j the results become complicated
and there is no simple general formula.
For half integer spin it is covenient to extend the integration to be from 0 to 4π and the
1
with w = ei 2 θ
1 (j− 1 )(j+ 23 ) (1 − w2 )(1 − w−2 )
MSU (2) (C2j+1 , t) = ∮ dw w 2 j
. (3.264)
4πi ∣w∣=1 ∏m= 1 (w2m − t)(1 − t w2m )
2

94
As special cases for j = 12 , 3
2

1
MSU (2) (C2 , t) = 1 , MSU (2) (C4 , t) = . (3.265)
1 − t4
There are no invariants for the two dimensional spin- 12 but there is one of order 4 for spin- 32 .
Quantum fields with half integer spin are anticommuting so it is more natural to use
(2.112) to count invariants in this case. As above this can be reduced to a contour integral
analogous to (3.263) but only poles at w = 0 are relevant. The lowest cases are

M̃SU (2) (M2 , t) = 1 + t2 , M̃SU (2) (M4 , t) = 1 + t2 + t4 . (3.266)

There is now a quadratic invariant since (V 21 ⊗V 21 )antisym ⊃ V0 . For j = 92 the obvious pattern
does not extend since there are then two quartic and two sextic invariants. Extending to
integer j
M̃SO(3) (M3 , t) = 1 + t3 , M̃SO(3) (M5 , t) = 1 + t5 , (3.267)
where the first result reflects the existence of the invariant tensor ijk .

3.13 Irreducible Tensor Operators

An alternative basis for irreducible tensor operators is achieved by requiring them to trans-
form similarly to the angular momentum states ∣j m⟩. An irreducible tensor operator in the
standard angular momentum basis satisfies
Definition: The set of (2k + 1) operators {Tkq } for

k ∈ {0, 12 , 1, 32 , . . . } , (3.268)

and
q ∈ {−k, −k + 1, . . . , k − 1, k} , (3.269)
for each k in (3.268), constitute a tensor operator of rank k if they satisfy the commutation
relations

[J3 , Tkq ] = q Tkq ,

±
[J± , Tkq ] = Nkq Tkq±1 , (3.270)
±
with Nkq given by (3.79). This definition is of course modelled exactly on that for the ∣j m⟩
states in (3.75) and (3.77) and ensures that we may treat it, from the point of view of its
angular moment properties, just like a state ∣k q⟩.
Examples:
If k = 0 then q = 0 and hence [J , T00 ] = 0, i.e. T00 is just a scalar operator.
If k = 1 then setting
√
V1±1 = ∓ 12 (V1 ± iV2 ) , V10 = V3 , (3.271)

95
ensures that V1q satisfy (3.270) for k = 1 as a consequence of (3.231).
If k = 2 we may form an irreducible tensor operator T2q from two vectors Vi , Ui using
Clebsch-Gordan coefficients
T2q = ∑ V1m U1m′ ⟨1m 1m′ ∣2q⟩ , (3.272)
m,m′

with V1m , U1m′ defined as in (3.271). This gives

√
T22 = V11 U11 , T21 = 12 (V11 U10 + V10 U11 ) ,
√
T20 = 16 (V11 U1−1 + 2V10 U10 + V1−1 U11 ) ,
√
T2−1 = 12 (V10 U1−1 + V1−1 U10 ) , T2−2 = V1−1 U1−1 . (3.273)
The individual T2q may all be expressed in terms of components of the symmetric traceless
tensor Sij = V(i Uj) − 31 δij Vk Uk .
For irreducible tensor operators Tkq their matrix elements with respect to states ∣α, j m⟩,
where α are any extra labels necessary to specify the states in addition to jm, are constrained
by the theorem:
Wigner-Eckart Theorem29
⟨α′ , j ′ m′ ∣Tkq ∣α, j m⟩ = ⟨jm kq∣j ′ m′ ⟩ C , (3.274)
with ⟨jm kq∣j ′ m′ ⟩ a Clebsch-Gordan coefficient. The crucial features of this result are:
(i) The dependence of the matrix element on m, q and m′ is contained in the Clebsch-
Gordan coefficient, and so is known completely. This ensures that the matrix element is
non zero only if j ′ ∈ {j + k, j + k − 1, . . . , ∣j − k∣ + 1, ∣j − k∣}.
(ii) The coefficient C depends only on j, j ′ , k and on the particular operator and states
involved. It may be written as
C = ⟨α′ j ′ ∣∣Tk ∣∣αj⟩ , (3.275)
and is referred to as a reduced matrix element.
The case k = q = 0 is an important special case. If [J , T00 ] = 0, then T00 is scalar
operator and we we have
⟨α′ , j ′ m′ ∣T00 ∣α, j m⟩ = ⟨jm 00∣j ′ m′ ⟩ ⟨α′ j ′ ∣∣T0 ∣∣αj⟩
= δjj ′ δmm′ ⟨α′ j ′ ∣∣T0 ∣∣αj⟩ , (3.276)
with reduced matrix-element independent of m.
To prove the Wigner-Eckart theorem we first note that Tkq ∣α, j m⟩ transforms under the
action of the angular momentum operator J just like the product state ∣k q⟩1 ∣j m⟩2 under
the combined J1 + J2 . Hence

∑ Tkq ∣α, j m⟩⟨kq jm∣JM ⟩ = ∣J M ⟩ (3.277)

q,m
29
Carl Henry Eckart, 1902-1973, American.

96
defines a set of states {∣J M ⟩} satisfying, by virtue of the definition of Clebsch-Gordan
coefficients in (3.144),
±
J3 ∣J M ⟩ = M ∣J M ⟩ , J± ∣J M ⟩ = NJ,M ∣J M ±1⟩ . (3.278)

Although the states ∣J M ⟩ are not normalised, it follows then that

⟨α′ , j ′ m′ ∣J M ⟩ = CJ δj ′ J δm′ M , (3.279)

defines a constant CJ which is independent of m′ , M . To verify this we note

⟨α′ , JM ∣J M ⟩NJM
+ ′
−1 = ⟨α , JM ∣J+ ∣J M −1⟩
= ⟨α′ , JM ∣J−† ∣J M −1⟩ = ⟨α′ , JM −1∣J M −1⟩NJM
−
. (3.280)
+ − ′ ′ ′
Since NJM −1 = NJM we then have ⟨α , JM ∣J M ⟩ = ⟨α , JM −1∣J M −1⟩ so that, for m = M ,
(3.279) is independent of M . Inverting (3.277)

Tkq ∣α, j m⟩ = ∑ ∣J M ⟩⟨kq jm∣JM ⟩ , (3.281)

JM

and then taking the matrix element with ⟨α′ , j ′ m′ ∣ gives the Wigner-Eckart theorem, using
(3.279), with Cj ′ = ⟨α′ j ′ ∣∣Tk ∣∣αj⟩.

3.14 Spinors

For the rotation groups there are spinorial representations as well as those which can be
described in terms of tensors, which are essentially all those which can be formed from
multiple tensor products of vectors. For SO(3), spinorial representations involve j being
half integral and are obtained from the fundamental representation for SU (2).
For the moment we generalise to A = [Aα β ] ∈ SU (r), satisfying (3.28), and consider a
vector η belonging to the r-dimensional representation space for the fundamental represen-
tation and transforming as
ηα → η ′α = Aα β ηβ . (3.282)
A

The extension to a tensor with n indices is straightforward

ψα1 ...αn → ψ ′α1 ...αn = Aα1 β1 . . . Aαn βn ψβ1 ...βn , (3.283)

A

Since A is unitary
(Aα β )∗ = (A−1 )β α . (3.284)
The complex conjugation of (3.282) defines a transformation corresponding to the conjugate
representation. If we define
η̄ α = (ηα )∗ , (3.285)
then using (3.284) allows the conjugate transformation rule to be written as

η̄ α → η̄ ′α = η̄ β (A−1 )β α . (3.286)
A

97
It is clear then that η̄ α ηα is a scalar. A general tensor may have both upper and lower
indices, of course each upper index transforms as (3.282), each lower one as (3.286).
As with the previous discussion of tensors it is critical to identify the invariant tensors.
For the case when A ∈ SU (2) and α, β = 1, 2 we have the two-dimensional ε-symbols,
εαβ = −εβα , ε12 = 1, and εαβ = −εβα , where it is convenient to take ε12 = −1. To verify εαβ
is invariant under the transformation corresponding to A we use

ε′αβ = Aα γ Aβ δ εγδ = det A εαβ = εαβ for A ∈ SU (2) , (3.287)

and similarly for εαβ . The Kronecker delta also forms an invariant tensor if there is one
lower and one upper index since,

δ ′α β = Aα γ δγ δ (A−1 )δ β = δα β . (3.288)

For this two-dimensional case, with the preceding conventions, we have the relations

εαβ εγδ = −δα γ δβ δ + δα δ δβ γ , εαγ εγβ = δα β . (3.289)

Rank n tensors as in (3.283) here span a vector space of dimension 2n . To obtain

an irreducible vector space under SU (2) transformations we require that contractions with
invariant tensors of lower rank give zero. For φα1 ...αn it is sufficient to impose εαr αs φα1 ...αn =
0 for all r < s. The irreducible tensors must then be totally symmetric φα1 ...αn = φ(α1 ...αn ) .
To count these we may restrict to those of the form

φ 1...1
2...2 where r = 0, . . . , n , r + s = n . (3.290)
±±
r s

Hence there are n + 1 independent symmetric tensors φα1 ...αn so that the representation
corresponds to j = 12 n.
The SU (2) vectors ηα and also η̄ α form SO(3) spinors. For this case the two index
invariant tensors εαβ and εαβ may be used to raise and lower indices. Hence we may define

η α = εαβ ηβ , (3.291)

which transforms as in (3.286) and correspondingly

η̄α = εαβ η̄ β , (3.292)

As a consequence of (3.289) raising and then lowering an index leaves the spinors ηα un-
changed, and similarly for η̄ α . In general the freedom to lower indices ensures that only
SU (2) tensors with lower indices, as in (3.283), need be considered.
For an infinitesimal SU (2) transformation, with A as in (3.37), the corresponding change
in a spinor arising from the transformation (3.282) is

δηα = −iδθ 12 (n ⋅ σ)α β ηβ . (3.293)

For a tensor then correspondingly from (3.283)

n
δψα1 ...αn = −iδθ ∑ 21 (n ⋅ σ)αrβ ψα1 ...αr−1 βαr+1 ...αn , (3.294)
r=1

98
where there is a sum over contributions for each separate index.
Making use of (3.289) we have
εαγ εβδ σ γ δ = σ β α , (3.295)
since tr(σ) = 0. From (3.295) we get
εαγ σ γ β = εβγ σ γ α , (3.296)
showing that (εσ)αβ form a set of three symmetric 2 × 2 matrices. Similar considerations
also apply to (σε)αβ . The completeness relations for Pauli matrices can be expressed as
(σε)αβ ⋅ (εσ)γδ = δα γ δβ δ + δα δ δβ γ , (εσ)αβ ⋅ (εσ)γδ = −εαγ εβδ − εαδ εβγ . (3.297)

The Pauli matrices allow symmetric spinorial tensors to be related to equivalent irre-
ducible vectorial tensors. Thus we may define, for an even number of spinor indices, the
tensor
Ti1 ...in = (εσi1 )α1 β1 . . . (εσin )αn βn ψα1 ...αn β1 ...βn , (3.298)
where it is easy to see that Ti1 ...in is symmetric and also zero on contraction of any pair of
indices, as a consequence of (3.297). For an odd number of indices we may further define
Tα i1 ...in = (εσi1 )α1 β1 . . . (εσin )αn βn ψαα1 ...αn β1 ...βn , (3.299)
where Tα i1 ...in is symmetric and traceless on the vectorial indices and satisfies the constraint
(σj )α β Tβ i1 ...in−1 j = 0 . (3.300)

For two symmetric spinorial tensors φ1,α1 ...αn , φ2,β1 ...βm , with j1 = 21 n, j2 = 12 m, their
product can be decomposed into symmetric rank (n + m − 2r)-tensors, for r = 0, . . . m if
n ≥ m, where for each r,
εβ1 γ1 . . . εβr γr φ1,(α1 ...αn−r β1 ...βr φ2,αn−r+1 ...αn+m−2r )γ1 ...γr , r = 0, . . . , m . (3.301)
For two spinors η1α , η2α the resulting decomposition into irreducible representation spaces
is given by
η1α η2β = η1(α η2β) + εαβ 21 η1γ η2γ , (3.302)
where η1(α η2β) may be re-expressed as a vector using (3.298). This result demonstrates
the decomposition of the product of two spin- 12 representations into j = 0, 1, scalar, vector,
irreducible components which are respectively antisymmetric, symmetric under interchange.
For two symmetric spinorial tensors as above with n = m there is a SU (2) invariant
εα1 β1 . . . εαn βn φ1,α1 ...αn φ2,β1 ...βn , (3.303)
which is clearly symmetric, antisymmetric under φ1 ↔ φ2 according to whether n is even,
odd. For four n = 4 symmetric spinorial tensors there is an independent antisymmetric
invariant
εα1 β1 εα2 γ1 εα1 δ1 εβ2 γ2 εβ3 δ2 εγ3 δ3 φ1,α1 α2 α3 φ2,β1 β2 β3 φ3,γ1 γ2 γ3 φ4,δ1 δ2 δ3 , (3.304)
where the indices are contracted so as to correspond to the lines joining the vertices of a
tetrahedron. Interchanging any two vertices then generates a minus sign since ε for the line
joining the two vertices changes sign. The invariant in (3.304) corresponds to the t4 term
appearing in (3.266).

99
3.15 Spinor Representation for Angular Momentum

Finding explicit expressions for 3j and 6j symbols can be a rather involved combinatorial
exercise. This can be made more algebraic by using spinor operators to provide representa-
tions of the angular momentum operators To this end we introduce operators φ̂α satisfying
the commutation relations30
[φ̂α , φ̂β ] = 2 εαβ 1 . (3.305)

The operators φ̂α are arbitrary up to φ̂α → φ̂β Mβ α for [Mβ α ] ∈ Sp(2, C). As a consequence
of (3.305)
J+ = 41 φ̂1 φ̂1 , J− = − 14 φ̂2 φ̂2 , J3 = 18 (φ̂1 φ̂2 + φ̂2 φ̂1 ) , (3.306)
satisfy the standard commutation relations (3.61a), (3.61b) and

∂ x⋅ φ̂ ∂ x⋅ φ̂
[J+ , ex⋅ φ̂ ] = x2 e , [J− , ex⋅ φ̂ ] = x1 e ,
∂x1 ∂x2
1 ∂ ∂
[J3 , ex⋅ φ̂ ] = (x1 − x2 ) ex⋅ φ̂ , x⋅ φ̂ = xα φ̂α (3.307)
2 ∂x1 ∂x2
Hence
j
1 1
ex⋅ φ̂ = ∑ ∑ x1 j+m x2 j−m Oj m , njm = ((j + m)!(j − m)!) 2 , (3.308)
j=0, 12 ,1,... m=−j
njm

ensures that Ojm are irreducible tensor operators satisfying (3.270). The operators Ojm
form a basis for the 2j + 1 symmetric tensor operators φ̂(α1 φ̂α2 . . . φ̂α2j ) . By using (3.305)
any arbitrary product φ̂α1 φ̂α2 . . . φ̂αn can be reduced to a sum over symmetrised products
and hence Ojm for 2j ≤ n.
For the product of two exponentials we have

ex⋅ φ̂ ey⋅ φ̂ = e(x+y)⋅ φ̂ eε xβ yγ

βγ

j
1 1
= ∑ ∑ (x1 + y1 )j+m (x2 + y2 )j−m Oj m ∑ (−1)s (x1 y2 )r (x2 y1 )s . (3.309)
1
j=0, ,1,... m=−j n jm r,s≥0 r!s!
2

Expanding ex⋅ φ̂ in terms of Oj1 m1 and ey⋅ φ̂ in terms of Oj2 m2 as in (3.308) and comparing
coefficients gives

(−1)s
Oj1 m1 Oj2 m2 = nj1 m1 nj2 m2 ∑
r,s≥0 r!s!(j1 + m1 − r)!(j1 − m1 − s)!(j2 + m2 − s)!(j2 − m2 − r)!
× nj1 +j2 −r−s m1 +m2 Oj1 +j2 −r−s m1 +m2

= (−1)j1 −j2 −m3

j j2 j3
∑ F ( m1 m2 m3
) Oj3 −m3 , m3 = −m1 − m2 , (3.310)
1
∣j1 −j2 ∣≤j3 ≤j1 +j2

30
A particular representations for such operators is unimportant for the considerations in this section.
Necessarily such operators act on an infinite dimensional space. A particular choice, acting on functions
f (x), is provided by taking φ̂1 = 2x, φ̂2 = − dx
d
or in terms of creation and annihilation operators φ̂1 =
√ † 2 √
2 a , φ̂ = − 2 a.

100
for
j j2 j3 (−1)s+j1 −j2 −m3
F ( m1 m2 m3
) = nj1 m1 nj2 m2 nj3 m3 ∑ s!(j1 −m1 −s)!(j2 +m2 −s)!(j3 −j2 +m1 +s)!(j3 −j1 −m2 +s)!(j1 +j2 −j3 −s)! .
1
s
(3.311)

In (3.311) the range of the s sum is dictated by the factorials in the denominator. For
j3 = 0, j1 = j2 and m2 = −m1 there is just one term when s = j1 − m1 for a non zero
contribution giving
j j 0
F( ) = (−1)j−m . (3.312)
m −m 0
Another such case arises for j3 = j1 + j2 when we must take s = 0 so that
j1 j2 j1 +j2 nj1 +j2 m1 +m2
F (m m2 −m1 −m2
) = (−1)j1 −j2 +m1 +m2 . (3.313)
1 nj1 m1 nj2 m2
1
For j2 = 2 there remains one term and
1
j j+ 12 1 j 1
j− 12 1
F ( m ±21 −m∓ 21
) = ±(j ± m + 1) 2 (−1)j+m , F ( m ±21 −m∓ 12
) = −(j ∓ m) 2 (−1)j+m . (3.314)
2 2

From (3.311) we may readily derive the symmetry properties by shifting s → j1 +j2 −j3 −s
and s → j2 − j3 − m1 + s

= (−1)j1 +j2 +j3 F ( m2

j j2 j3 j j1 j3 j j1 j3 j j1 j2
F ( m1 m2 m3
) = F ( −m2 −m1 −m3
) m1 m3
) = F ( m3 m1 m2
) . (3.315)
1 2 2 3

Detailed results are based on

j
exp ( − u ∆xy ) ex⋅ φ̂ ez ⋅ φ̂ ey⋅ φ̂ ∣ = ∑ u2j ∑ (−1)j+m Oj −m ez ⋅ φ̂ Oj m
x=y=0
j=0, 21 ,1,... m=−j

1 ′ ∂2 1+u
= 2
eu z ⋅ φ̂ , ∆xy = εαβ , u′ = , (3.316)
(1 − u) ∂xα ∂yβ 1−u

where the calculation of the action of the derivatives is described later. This reduces to
j
1
∑ u2j ∑ (−1)j+m Oj −m Okn Oj m = u′2k Okn . (3.317)
j=0, 12 ,1,... m=−j (1 − u)2

As a special case, taking k = 0, this gives31

j
j+m
∑ (−1) Oj −m Oj m = (2j + 1) 1 , (3.318)
m=−j

31
Alternatively starting from (3.310)

p (p + s)! (2j − s − p)!

j 2j 2j−j ′ j ′
1
∑ (−1)
j+m
Oj −m Oj m = ∑ j ′ ! O2j ′ 0 ∑ ∑ (−1) ,
m=−j j ′ =0 s=0 s! (2j − j − s)! p=0
′ p!2 (j ′ − p)!2

gives the same result.

101
and therefore, as a consequence of (3.318), from (3.310) and using (3.315)
j +j +m +m
∑ (−1) 1 2 1 2 Oj2 −m2 Oj1 −m1 Oj1 m1 Oj2 m2 = (2j1 + 1)(2j2 + 1) 1
m1 ,m2
j1 +j2
j2 j3 ′
(−1)j1 +j2 −m3 Oj3 ′ m3 Oj3 −m3 .
j j j2 j3
= ∑ ∑ F ( m1 m2 m3
) F ( m1 m2 m3
) (3.319)
1 1
m1 ,m2 j3 ′ ,j3 =∣j1 −j2 ∣

For consistency it is necessary that

j j2 j3 ′ j j2 j3 1
∑ F ( m1 ) F ( m1 ) = f (j1 , j2 , j3 ) δj ′ j , (3.320)
m1 ,m2 ,m1 +m2 =−m3
1 m2 m3 1 m2 m3 2j3 + 1 3 3

so that
j j2 j3 j j2 j3
∑ F ( m1 m2 m3
) F ( m1 m2 m3
) = f (j1 , j2 , j3 ) , (3.321)
1 1
mi ,∑ mj =0

where the symmetry conditions (3.315) requires f (j1 , j2 , j3 ) to be totally symmetric.

To obtain an explicit form for f (j1 , j2 , j3 ) we may use a generalisation of (3.316)
′ ⋅ φ̂ ′ ⋅ φ̂
exp ( − r ∆zz ′ − s ∆ww′ ) ez ⋅ φ̂ ew⋅ φ̂ ⊗ ew ez ∣
z=w=z ′ =w′ =0
j1 +j2
= ∑ r2j1 s2j2 ∑
j1 ,j2 =0, 12 ,... j3 ,j3′ =∣j1 −j2 ∣
j1 j3 ′
× ∑ (−1)j1 +m1 +j2 +m2 F ( −m1
j j2 j3 j
−m2 m3
) F ( m2 m1 −m3
) Oj3 −m3 ⊗ Oj3′ m3 . (3.322)
1 2
m1 ,m2

Hence, using (3.319) and (3.318), and also with u′ as in (3.316) and similarly for v ′ in terms
of v,

exp ( − r ∆x2 x7 − s ∆x3 x6 − u ∆x1 x4 − v ∆x5 x8 ) ex1 ⋅ φ̂ ex2 ⋅ φ̂ . . . ex8 ⋅ φ̂ ∣

xi =0
1 2j1 2j2 ′ ′ 2j3 j1 +j2 −j3
= ∑ r s (u v ) (−1) f (j1 , j2 , j3 ) 1
(1 − u) (1 − v)2j
2
1 ,j2 ,j3 =0, 2 ,...
1

1 1
= 1. (3.323)
(1 − u)2 (1 − v)2 (1 + r s − r u′ v ′ − s u′ v ′ )2

Since ∑a,b,c≥0 (−1)c (a + b + c + 1)!/(a! b! c!) xa y b z c = (1 − x − y + z)−2 we then have

(j1 + j2 + j3 + 1)!
f (j1 , j2 , j3 ) = . (3.324)
(j1 + j2 − j3 )! (j2 + j3 − j1 )! (j3 + j1 − j2 )!

From the generating function for f (j1 , j2 , j3 ) given by (3.323) this satisfies
j1 +j2
1
∑ r2j1 s2j2 ∑ (−1)j1 +j2 −j3 f (j1 , j2 , j3 ) =
j1 ,j2 =0, 21 ,... j3 =∣j1 −j2 ∣ (1 − r)2 (1 − s)2
2j 2j
= ∑ (2j1 + 1)(2j2 + 1) r 1 s 2 , (3.325)
j1 ,j2 =0, 12 ,...

102
as is necessary for (3.319) using (3.321).
Manifestly (3.310) is just the usual Clebsch-Gordan angular momentum decomposition
j j j
as in (3.157) so that the 3j symbol ( m1 m2 m3 ) is determined up to a normalisation factor
1 2 3
independent of m1 , m2 , m3 . The normalisation is determined by (3.320) so that

∆(j1 , j2 , j3 ) = f (j1 , j2 , j3 )− 2 .
j j2 j3 j j2 j3 1
( m1 m2 m3
) = ∆(j1 , j2 , j3 ) F ( m1 m2 m3
) , (3.326)
1 1

These techniques can be extended to derive results for 6j symbols. Following a similar
route as before

exp (−u1 ∆x1 x4 −u3 ∆x5 x8 −u2 ∆x9 x12 −v2 ∆x3 x6 −v1 ∆x7 x10 −v3 ∆x2 x1 ) ex1 ⋅ φ̂ ex2 ⋅ φ̂ . . . ex12 ⋅ φ̂ ∣
xi =0
3 ki
= exp (− u1 ∆x1 x4 − u3 ∆x5 x8 − u2 ∆x9 x12 ) ∏ ∑ vi 2ki ∑ (−1)ki +ni
i=1 ki =0, 1 ,... ni =−ki
2
x1 ⋅ φ̂ x4 ⋅ φ̂ x5 ⋅ φ̂ x8 ⋅ φ̂ x9 ⋅ φ̂
×e Ok3 −n3 Ok2 −n2 e e Ok2 n2 Ok1 −n1 e e Ok1 n1 Ok3 n3 ex12 ⋅ φ̂ ∣
xi =0
3
1 j1 j2 j3
ui′2j1 vi 2ji (−1)∑i=1 (2ji +2ki ) F {
3
=∏ 2 ∑ }1
k1 k2 k3
i=1 (1 − ui ) ji ,ki =0, 21 ,...
3
1
=∏ (3.327)
i=1 (1 − ui )2
1
× 2
1,
(1 − u2′ u3′ v1 − u1′ u3′ v2 − u1′ u2′ v3 − v1 v2 v3 + u1′ u2′ v1 v2 + u1′ u3′ v1 v3 + u2′ u3′ v2 v3 )
using that j1 + k2 + k3 , j2 + k1 + k3 are integers to achieve a symmetric form for the sign
factor and where we have required, similarly to (3.190),
j3 j1′
(−1)∑i (ji +mi +ki +ni ) F ( n2
k k3 j1 k k 1 j2 k k2 j3 j
∑ −n3 −m1
)F ( n3 −n1 −m3
)F ( n1 −n2 −m3
)F ( m2 m3 m1
)
2 3 1 2
ni , m1 fixed
j1 j2 j3
=F{ } 2j11+1 δj1 j1′ . (3.328)
k1 k2 k3

Clearly from (3.326)

j1 j2 j3 j1 j2 j3
{ } = ∆(k2 , k3 , j1 ) ∆(k3 , k1 , j2 ) ∆(k1 , k2 , j3 ) ∆(j1 , j2 , j3 ) F { }, (3.329)
k1 k2 k3 k1 k2 k3

and the symmetry properties in (3.189) follow directly from the generating function given
by (3.328). By using
(∑i ai + ∑j bj + 1)! 4 3
1
2
= ∑ (−1)∑j bj ∏ xiai ∏ yjbj (3.330)
(1 − ∑4i=1 xi + ∑3j=1 yj ) ai ,bj ≥0 ∏4i=1 ai ! ∏3j=1 bj ! i=1 j=1

to expand (3.323) with x1 = u2 ′ u3′ v1 , x2 = u1′ u3′ v2 , x3 = u1′ u2′ v3 , x4 = v1 v2 v3 , y1 =

u2′ u3′ v2 v3 , y2 = u1′ u3′ v1 v3 , y3 = u1′ u2′ v1 v2 and matching with ∏3i=1 ui′2j1 vi 2ji requires
taking

a1 = s − j1 − k2 − k3 , a2 = s − j2 − k3 − k1 a3 = s − j2 − k1 − k2 , a4 = s − j1 − j2 − j3 ,
b1 = j2 + j3 + k2 + k3 − s , b2 = j3 + j1 + k3 + k1 − s , b3 = j1 + j2 + k1 + k2 − s . (3.331)

103
with s an integer constrained to a finite range by requiring each ai , bj is a positive integer
or zero and ∑i ai + ∑j bj = s. Hence

j1 j2 j3 (s + 1)!
F{ } = ∑ (−1)s , (3.332)
k1 k2 k3 a1 ! a2 ! a3 ! a4 ! b1 ! b2 ! b3 !
s

with ai , bj as in (3.331) and where the sign (−1)∑j bj = (−1)∑j (2jj +2kj )+s . As a special
case for k1 = 0 there is only one term in the sum with a non zero contribution requiring
k2 = j3 , k3 = j2 and

j1 j2 j3 j1 j2 j3 (−1)∑i ji
F{ 0 j3 j2
} = (−1)∑i ji f (j1 , j2 , j3 ) ⇒ { 0 j3 j2
} = √ . (3.333)
(2j2 + 1)(2j3 + 1)

The generating functions obtained here were first determined by Schwinger.32

3.15.1 Calculation of Action of Derivatives

The calculations to obtain the results in (3.316), (3.322), (3.323) or (3.328) is complicated
by the non commutavity of φ̂α . We consider in general

exp (− ∑ ui,j ∆xi xj ) ex1 ⋅ φ̂ ex2 ⋅ φ̂ . . . exn ⋅ φ̂ ∣ , (3.334)

i<j∈I xi =0,i∈I

for I ⊂ {1, 2, . . . , n} with p = dim I even. For ui,j , ui′ j ′ to be both non zero we require
i ≠ i′ , j ≠ j ′ . With our conventions εαβ = −1 and from (3.316),

⎧
⎪ ui,j i<j
⎪
⎪
⎪
∑ ui,j ∆xi xj = ∑ ∂xi,2 Uij ∂xj,1 , Uij = ⎨−ui,j i>j . (3.335)
i<j∈I i,j∈I
⎪
⎪
⎪
⎩0
⎪ otherwise

U = [Uij ] is then an antisymmetric p × p matrix with one non zero element in each row and
column.
For p = n (3.334) defines an invariant. To evaluate (3.334) we define a normal ordering
whereby all operators φ̂1 are moved to the left of φ̂2 . Thus

ex⋅ φ̂ = N (ex⋅ φ̂ ) e−x1 x2 , N (ex⋅ φ̂ ) = ex1 ⋅ φ̂ ex2 ⋅ φ̂ ,

1 2
(3.336)

and in general
⎧
⎪ 0 i<j
⎪
⎪
x1 ⋅ φ̂ xn ⋅ φ̂ x1 ⋅ φ̂ xn ⋅ φ̂ ⎪
e ...e = N (e ...e ) exp ( − ∑ xi,1 Vij xj,2 ) , Vij = ⎨ 1 i=j ,
i,j ⎪
⎪
⎪
⎩2
⎪ i>j
N (ex1 ⋅ φ̂ . . . exn ⋅ φ̂ ) = exp (∑i xi,1 φ̂1 ) exp (∑i xi,2 φ̂2 ) . (3.337)
32
Juliian Seymour Schwinger, 1918-1994, American. Nobel prize 1965.

104
Defining coherent states such that ⟨φ1 ∣φ̂1 = φ1 ⟨φ1 ∣, φ̂2 ∣φ2 ⟩ = φ2 ∣φ2 ⟩ then
⟨φ1 ∣N (f (φ̂1 , φ̂2 ))∣φ2 ⟩ = f (φ1 , φ2 ) ⟨φ1 ∣φ2 ⟩ . (3.338)
With these results the calculation of (3.334) can be reduced to considering

exp (− ∑ ui,j ∆xi xj ) ⟨φ1 ∣ ex1 ⋅ φ̂ . . . exn ⋅ φ̂ ∣φ2 ⟩ ∣

i<j∈I xi =0,i∈I
n n
= exp (− ∑ ∂xi,2 Uij ∂xj,1 ) exp (∑ (xi,1 φ1 + φ2 xi,2 ) − ∑ xi,1 Vij xj,2 ) ∣ ⟨φ1 ∣φ2 ⟩ ,
i,j∈I i=1 i,j=1 xi =0,i∈I

(3.339)
where, with I ∪ I¯ = {1, 2, . . . , n},
n
∑ xi,1 Vij xj,2
i,j=1

= ∑ xi,1 V (p)ij xj,2 + ∑ xi,1 V (n−p)ij xj,2 + 2 ∑ (xi,1 ∑ xj,2 + ∑ xj,1 xi,2 ) , (3.340)
i,j∈I i,j∈I¯ i∈I ¯
j∈I,j>i ¯
j∈I,j<i

with V (p) the p×p lower triangle matrix with 1’s on the diagonal and 2′ s below the diagonal.
The evaluation of (3.339) follows from

exp ( − ∂x̃ ⋅ U ⋅ ∂x ) exp (x ⋅ ỹ + y ⋅ x̃ − x ⋅ V ⋅ x̃)∣ = exp ( − ∂ỹ ⋅ V ⋅ ∂y ) exp ( − y ⋅ U ⋅ ỹ)

x=x̃=0

= exp ( − ∂ỹ ⋅ V ⋅ ∂y ) det (U −1 ) ∫ dzdz̃ exp (z ⋅ U −1 ⋅ z̃ + z ⋅ ỹ + y ⋅ z̃)

= det (U −1 ) ∫ dzdz̃ exp (z ⋅ (U −1 − V ) ⋅ z̃ + z ⋅ ỹ + y ⋅ z̃)

1
= exp ( − y ⋅ (1 − V ⋅ U )−1 ⋅ U ⋅ ỹ) , (3.341)
det(1 − V ⋅ U )
for x̃, ỹ column vectors, x, y row vectors and U, V non singular square matrices.
For application to (3.316)
x2 φ1 0 u 1 0
x = (x1 y1 ) , x̃ = ( ) , y = (φ2 − 2z1 φ2 ) , ỹ = ( 1 ), U = ( ), V = ( ),
y2 φ − 2z2 −u 0 2 1
(3.342)
so that
exp ( − ∂x̃ ⋅ U ⋅ ∂x ) exp (x ⋅ ỹ + y ⋅ x̃ − x ⋅ V ⋅ x̃)∣ × exp (z1 φ1 + φ2 z2 − z1 z2 )
x=x̃=0
= 1
(1−u)2
exp (u′ (z1 φ1 + φ2 z2 ) − u′2 z1 z2 ) , u′ = 1−u
1+u
. (3.343)
which corresponds to the result in (3.316) after normal ordering. For (3.328) the matrix U
becomes
0 0 0 u1 0 0 0 0 0 0 0 0
⎛ 0 0 0 0 0 0 0 0 0 0 v3 0 ⎞
⎜ 0 0 0 0 0 v2 0 0 0 0 0 0 ⎟
⎜ −u1 0 0 0 0 0 0 0 0 0 0 0 ⎟
⎜ 0 0 0 0 0 0 0 u3 0 0 0 0 ⎟
⎜ 0 0 −v2 0 0 0 0 0 0 0 0 0 ⎟
U =⎜ ⎜ 0 0 0 0 0 0 0 0 0 v1 0 0 ⎟ ,
⎟ (3.344)
⎜ 0 0 0 0 −u3 0 0 0 0 0 0 0 ⎟
⎜ 0 0 0 0 0 0 0 0 0 0 0 u ⎟
⎜ 2⎟
⎜ 0 0 0 0 0 0 −v1 0 0 0 0 0 ⎟
⎝ 0 −v3 0 0 0 0 0 0 0 0 0 0 ⎠
0 0 0 0 0 0 0 0 −u2 0 0 0

105
where 1/ det(112 − V ⋅ U ) gives the final result for the generating function.

3.16 Isospin

The symmetry which played a significant role in the early days of nuclear and particle
physics is isospin, was initially based on the symmetry between neutrons and protons as far
as nuclear forces were concerned. The symmetry group is again SU (2) with of course the
sam1 mathematical properties as discussed in its applications to rotations, but with a very
different physical interpretation. Results for Clebsch-Gordan coefficients are crucial in the
applications of isospin symmetry. In order to distinguish this SU (2) group from various
others which arise in physics it is convenient to denote it as SU (2)I .
From a modern perspective this symmetry arises since the basic QCD lagrangian de-
pends on the Dirac u and d quark fields only in terms of

u ¯ ,
q=( ), q̄ = (ū d) (3.345)
d

in such a way that it is invariant under q → Aq, q̄ → q̄A−1 for A ∈ SU (2). This symmetry
is violated by quark mass terms since mu ≠ md , although they are both tiny in relation to
other mass scales, and also by electromagnetic interactions since u, d have different electric
charges.
Neglecting such small effects there exist conserved charges I± , I3 which obey the SU (2)
commutation relations

[I3 , I± ] = ±I± , [I+ , I− ] = 2I3 or [Ia , Ib ] = i εabc Ic , (3.346)

as in (3.61a),(3.61b) or (3.54), and also commute with the Hamiltonian

[Ia , H] = 0 . (3.347)

The particle states must then form multiplets, with essentially the same mass, which trans-
form according to some SU (2)I representations. Each particle is represented by an isospin
state ∣I I3 ⟩ which form the basis states for a representation of dimension 2I + 1.
The simplest example is the proton and neutron which have I = 12 and I3 = 12 , − 21
respectively. Neglecting other momentum and spin variables, the proton, neutron states
are a doublet (∣p⟩, ∣n⟩) and we must have

I3 ∣p⟩ = 21 ∣p⟩ , I3 ∣n⟩ = − 21 ∣n⟩ , I− ∣p⟩ = ∣n⟩ , I+ ∣n⟩ = ∣p⟩ . (3.348)

Other examples of I = 12 doublets are the kaons (∣K + ⟩, ∣K 0 ⟩) and (∣K̄ 0 ⟩, ∣K − ⟩). The pions
form a I = 1 triplet (∣π + ⟩, ∣π 0 ⟩, ∣π − ⟩) so that
√ √
I3 (∣π + ⟩, ∣π 0 ⟩, ∣π − ⟩) = (∣π + ⟩, 0, −∣π − ⟩) , I− ∣π + ⟩ = 2∣π 0 ⟩ , I− ∣π 0 ⟩ = 2∣π − ⟩ . (3.349)

Another such triplet are the Σ baryons (∣Σ+ ⟩, ∣Σ0 ⟩, ∣Σ− ⟩). Finally we note that the spin- 23
baryons form a I = 32 multiplet (∣∆++ ⟩, ∣∆+ ⟩, ∣∆0 ⟩, ∣∆− ⟩). Low lying nuclei also belong to

106
isospin multiplets, sometimes with quite high values of I. For each multiplet the electric
charge for any particle is given by Q = Q0 + I3 , where Q0 has the same value for all particles
in the multiplet.
Isospin symmetry has implications beyond that of just classification of particle states
since the interactions between particles is also invariant. The fact that the isospin generators
Ia are conserved, (3.347), constrains dynamical processes such as scattering. Consider
a scattering process in which two particles, represented by isospin states ∣I1 m1 ⟩, ∣I2 m2 ⟩,
scatter to produce two potentially different particles, with isospin states ∣I3 m3 ⟩, ∣I4 m4 ⟩. The
scattering amplitude is ⟨I3 m3 , I4 m4 ∣T ∣I1 m1 , I2 m2 ⟩ and to the extent that the dynamics are
invariant under SU (2)I isospin transformations this amplitude must transform covariantly,
i.e.
(I )∗ (I ) ∗ (I ) (I )
′ ′ ′ ′
∑ Dm′33 m3 (R) Dm′44 m4 (R) Dm′11 m1 (R)Dm′22 m2 (R) ⟨I3 m 3 , I4 m 4 ∣T ∣I1 m 1 , I2 m 2 ⟩
m′3 ,m′4 ,m′1 ,m′2

= ⟨I3 m3 , I4 m4 ∣T ∣I1 m1 , I2 m2 ⟩ . (3.350)

This condition is solved by decomposing the initial and final states into states ∣IM ⟩ with
definite total isospin using Clebsch-Gordan coefficients,

∣I1 m1 , I2 m2 ⟩ = ∑ ∣IM ⟩ ⟨I1 m1 , I2 m2 ∣IM ⟩ ,

I,M

⟨I3 m3 , I4 m4 ∣ = ∑ ⟨I3 m3 , I4 m4 ∣IM ⟩⟨IM ∣ , (3.351)

I,M

since then, as in (3.276),

⟨I ′ M ′ ∣T ∣IM ⟩ = AI δI ′ I δM ′ M , (3.352)
as a consequence of T being an isospin singlet operator. Hence we have

⟨I3 m3 , I4 m4 ∣T ∣I1 m1 , I2 m2 ⟩ = ∑ AI ⟨I3 m3 , I4 m4 ∣IM ⟩⟨I1 m1 , I2 m2 ∣IM ⟩ . (3.353)

I

The values of I which appear in this sum are restricted to those which can be formed by
states with isospin I1 , I2 and also I3 , I4 . The observed scattering cross sections depend only
on ∣⟨I3 m3 , I4 m4 ∣T ∣I1 m1 , I2 m2 ⟩∣2 .
As an illustration we consider πN scattering for N = p, n. In this case we can write
√ √
∣π + p⟩ = ∣ 23 23 ⟩ , ∣π 0 p⟩ = 23 ∣ 32 12 ⟩ − 13 ∣ 21 12 ⟩ ,
√ √ √ √
∣π 0 n⟩ = 23 ∣ 32 − 12 ⟩ + 13 ∣ 12 − 21 ⟩ , ∣π − p⟩ = 13 ∣ 32 − 12 ⟩ − 23 ∣ 12 − 12 ⟩ , (3.354)

using the Clebsch-Gordan coefficients which have been calculated in (3.166) for j = 1. Hence
we have the results for the scattering amplitudes

⟨π + p∣T ∣π + p⟩ = A 3 ,
2
− − 1 2
⟨π p∣T ∣π p⟩ = 3 A 32 + 3 A 12 ,
√
⟨π 0 n∣T ∣π − p⟩ = 3
2
(A 3 − A 1 ) , (3.355)
2 2

107
so that three observable processes are reduced to two complex amplitudes A 3 , A 1 . For the
2 2
observable cross sections
2 2 2
σπ+ p→π+ p = k∣A 3 ∣ , σπ− p→π− p = 19 k∣A 3 + 2A 1 ∣ , σπ− p→π0 n = 29 k∣A 3 − A 1 ∣ , (3.356)
2 2 2 2 2

for k some isospin independent constant. There is no immediate algebraic relation between
the cross sections since AI are complex. However at the correct energy A 3 is large due to
2
3 1
the I = 2 ∆ resonance, then the cross sections are in the ratios 1 ∶ 9 ∶ 29 .
An example with more precise predictions arises with N N → πd scattering, where d is
the deuteron, a pn bound state with I = 0. Hence the πd state has only I = 1. Decomposing
N N states into states ∣IM ⟩ with I = 1, 0 we have ∣pp⟩ = ∣11⟩, ∣pn⟩ = √1 (∣10⟩ + ∣00⟩). Using
2
this we obtain σpn→π0 d /σpp→π+ d = 12 .
The examples of isospin symmetry described here involve essentially low energy pro-
cesses. Although it now appears rather fortuitous, depending on the lightness of the u, d
quarks in comparison with the others, it was clearly the first step in the quest for higher
symmetry groups in particle physics.

3.16.1 G-parity

G-parity is a discrete quantum number obtained by combining isospin with charge conjuga-
tion. Charge conjugation is a discrete Z2 symmetry where the unitary charge conjugation
operator C acts on a particle state to give the associated anti-particle state with opposite
charge. If these are different any associated phase factor is unphysical, since it may be
absorbed into a redefinition of the states. In consequence the charge conjugation parity
is well defined only for particle states with all conserved charges zero. For pions we have
without any arbitrariness just
C∣π 0 ⟩ = ∣π 0 ⟩ . (3.357)
The associated
√ ± charged pion states are obtained, with standard isospin conventions, by
0
I± ∣π ⟩ = 2∣π ⟩. Since charge conjugation reverses the sign of all charges we must take
CI3 C −1 = −I3 and we require also CI± C −1 = −I∓ (more generally if CI+ C −1 = −eiα I− , CI− C −1 =
−e−iπα I+ the dependence on α can be absorbed in a redefinition of I± ). By calculating
CI± ∣π 0 ⟩ we then determine unambiguously

C∣π ± ⟩ = −∣π ∓ ⟩ . (3.358)

G-parity is defined by combining C with an isospin rotation,

G = Ce−iπI2 . (3.359)

The action of e−iπI2 on an isospin multiplet is determined for any representation by (3.101).
In this case we have

e−iπI2 ∣π + ⟩ = ∣π − ⟩ , e−iπI2 ∣π 0 ⟩ = −∣π 0 ⟩ , e−iπI2 ∣π − ⟩ = ∣π + ⟩ , (3.360)

and hence on any pion state

G∣π⟩ = −∣π⟩ . (3.361)

108
Conservation of G-parity ensures that in any ππ scattering process only even numbers of
pions are produced. The notion of G-parity can be extended to other particles such as the
spin one meson ω, with I = 0, and ρ± , ρ0 , with I = 1. The neutral states have negative
parity under charge conjugation so the G-parity of ω and the ρ’s is respectively 1 and −1.
This constrains various possible decay processes.

109
4 Relativistic Symmetries, Lorentz and Poincaré Groups

Symmetry under rotations plays a crucial role in atomic physics, isospin is part of nuclear
physics but it is in high energy particle physics that relativistic Lorentz33 transformations,
forming the Lorentz group, have a vital importance. Extending Lorentz transformations
by translations, in space and time, generates the Poincaré34 group. Particle states can be
considered to be defined as belonging to irreducible representations of the Poincaré Group.

4.1 Lorentz Group

For space-time coordinates xµ = (x0 , xi ) ∈ R4 then the Lorentz group is defined to be the
group of transformations xµ → x′µ leaving the relativistic interval

x2 ≡ gµν xµ xν , g00 = 1 , g0i = gi0 = 0 , gij = −δij , (4.1)

invariant. Assuming linearity a Lorentz transformation xµ → x′µ

x′µ = Λµ ν xν , (4.2)

ensures
x′2 = x2 , (4.3)
which requires, for arbitrary x
gσρ Λσ µ Λρ ν = gµν . (4.4)
Alternatively in matrix language
1 0
ΛT gΛ = g , Λ = [Λµ ν ] , g = [gµν ] = ( ). (4.5)
0 −13

Matrices satisfying (4.5) belong to the group O(1, 3) ≃ O(3, 1).

In general we define contravariant and covariant vectors, V µ and Uµ , under Lorentz
transformations by

V µ → V ′µ = Λµ ν V ν , Uµ → U ′µ = Uν (Λ−1 )ν µ . (4.6)
Λ Λ

It is easy to see, using (4.4) or (4.5), V ′T g = V T ΛT g = V T gΛ−1 , that we may use gµν to lower
indices, so that gµν V ν is a covariant vector. Defining the inverse g µν , so that g µλ gλν = δ µν ,
we may also raise indices, g µν Uν is a contravariant vector.

4.1.1 Proof of Linearity

We here demonstrate that the only transformations which satisfy (4.3) are linear. We
rewrite (4.3) in the form
gµν dx′µ dx′ν = gµν dxµ dxν , (4.7)
33
Hendrik Antoon Lorentz, 1853-1928, Dutch. Nobel prize 1902.
34
Jules Henri Poincaré, 1853-1912, French.

110
and consider infinitesimal transformations

x′µ = xµ + f µ (x) , dx′µ = dxµ + ∂σ f µ (x) dxσ . (4.8)

Substituting (4.8) into (4.7) and requiring this to hold for any infinitesimal dxµ gives

gµσ ∂ν f σ + gσν ∂µ f σ = 0 , (4.9)

or, with fµ = gµσ f σ , we have the Killing equation,

∂µ fν + ∂ν fµ = 0 . (4.10)

Then we write

∂ω (∂µ fν + ∂ν fµ ) + ∂µ (∂ν fω + ∂ω fν ) − ∂ν (∂ω fµ + ∂µ fω ) = 2∂ω ∂µ fν = 0 . (4.11)

The solution, defining a Killing vector, is obviously linear in x,

fµ (x) = aµ + ωµν xν , (4.12)

and then substituting back in (4.10) gives

ωµν + ωνµ = 0 . (4.13)

For aµ = 0, (4.12) corresponds to an infinitesimal version of (4.2) with

Λµ ν = δ µν + ω µ ν , ω µ ν = g µσ ωσν . (4.14)

4.1.2 Structure of Lorentz Group

Taking the determinant of (4.5) gives

(det Λ)2 = 1 ⇒ det Λ = ±1 . (4.15)

By considering the 00’th component we also get

(Λ0 0 )2 = 1 + ∑i (Λ0 i )2 ≥ 1 ⇒ Λ0 0 ≥ 1 or Λ0 0 ≤ −1 . (4.16)

The Lorentz group has four components according to the signs of det Λ and Λ0 0 since no
continuous change in Λ can induce a change in these signs. For the component connected
to the identity we have det Λ = 1 and also Λ0 0 ≥ 1. This connected subgroup is denoted
SO(3, 1)↑ .
Rotations form a subgroup of the Lorentz group, which is obtained by imposing ΛT Λ = 1
as well as (4.5). In this case the Lorentz transform matrix has the form,

1 0
ΛR = ( ), R T R = 13 , (4.17)
0 R

where R ∈ O(3), det R = ±1, represents a three dimensional rotation or reflection, obviously
ΛR ΛR′ = ΛRR′ forming a reducible representation of this subgroup.

111
 Another special case is when
Λ = ΛT . (4.18)
To solve the constraint (4.5) we first write

cosh α sinh α nT
Λ=( ), B T = B , nT n = 1 , (4.19)
sinh α n B

where n is a 3-dimensional column vector, and then

1 sinh α(cosh α nT − nT B)
ΛT gΛ = ( ). (4.20)
sinh α(cosh α n − Bn) sinh2 α nnT − B 2

Hence (4.5) requires

Bn = cosh α n , B 2 − sinh2 α nnT = 13 . (4.21)
The solution is just
B = 13 + (cosh α − 1)nnT . (4.22)
The final expression for a general symmetric Lorentz transformation defining a boost is then

cosh α sinh α nT
B(α, n) = ( ), (4.23)
sinh α n 13 + (cosh α − 1)nnT

where the parameter α has an infinite range. Acting on xµ , using vector notation,

x′0 = cosh α x0 + sinh α n ⋅ x ,

x′ = x + (cosh α − 1)n n ⋅ x + sinh α n x0 . (4.24)

This represents a Lorentz boost with velocity v = tanh α n.

Boosts do not form a subgroup since they are not closed under group composition, in
general the product of two symmetric matrices is not symmetric, although there is a one
parameter subgroup for n fixed and α varying which is isomorphic to SO(1, 1) with matrices
as in (1.123). With ΛR as in (4.17) then for B as in (4.23)

ΛR B(α, n)ΛR−1 = B(α, nR ) , (4.25)

gives the rotated Lorentz boost. Any Lorentz transformation can be written as at of a boost
followed by a rotation.
√ To show this we note that ΛT Λ is symmetric and positive so we
may define B = ΛT Λ = B T , corresponding to a boost. Then ΛB −1 defines a rotation since
(ΛB −1 )T ΛB −1 = B −1 ΛT ΛB −1 = 1 and so ΛB −1 = ΛR , or Λ = ΛR B, with ΛR of the form in
(4.17).

4.2 Infinitesimal Lorentz Transformations and Commutation Relations

General infinitesimal Lorentz transformations have already been found in (4.14) with ω µ ν
satisfying the conditions in (4.13). For two infinitesimal Lorentz transformations

Λ1µ ν = δ µν + ω1µ ν , Λ2µ ν = δ µν + ω2µ ν , (4.26)

112
then
Λµ ν = (Λ2−1 Λ1−1 Λ2 Λ1 )µ ν = δ µν + [ω2 , ω1 ]µ ν , (4.27)
where it is clear that Λµ ν = δ µν if either ω1µ ν or ω2µ ν are zero.
For a relativistic quantum theory there must be unitary operators U [Λ] acting on the
associated vector space for each Lorentz transformation Λ which define a representation,

U [Λ2 ]U [Λ1 ] = U [Λ2 Λ1 ] . (4.28)

For an infinitesimal Lorentz transformation as in (4.13) we require

U [Λ] = 1 − i 21 ω µν Mµν , Mµν = −Mνµ . (4.29)

Mµν are the Lorentz group generators. Since we also have U [Λ−1 ] = 1 + i 21 ω µν Mµν (4.27)
requires

U [Λ] = 1 − i [ω2 , ω1 ]µν Mµν

= U [Λ2−1 ]U [Λ1−1 ]U [Λ2 ]U [Λ1 ]
= 1 − [ 21 ω2µν Mµν , 21 ω1σρ Mσρ ] , (4.30)

or

[ 12 ω2µν Mµν , 1
2 ω1σρ Mσρ ] = i [ω2 , ω1 ]µν Mµν , [ω2 , ω1 ]µν = gσρ (ω2µσ ω1ρν − ω1µσ ω2ρν ) . (4.31)

Since this is valid for any ω1 , ω1 we must have the commutation relations

[Mµν , Mσρ ] = i(gνσ Mµρ − gµσ Mνρ − gνρ Mµσ + gµρ Mνσ ) , (4.32)

where the four terms on the right side are essentially dictated by antisymmetry under µ ↔ ν,
σ ↔ ρ. For a unitary representation we must have

Mµν † = Mµν . (4.33)

Just as in (3.229) we may define contravariant and covariant vector operators by requir-
ing
U [Λ]V µ U [Λ]−1 = (Λ−1 )µ ν V ν , U [Λ]Uµ U [Λ]−1 = Uν Λν µ . (4.34)
For an infinitesimal transformation, with Λ as in (4.14) and U [Λ] as in (4.29), this gives

[Mµν , V σ ] = −i(δ σµ Vν − δ σν Vµ ) , [Mµν , Uσ ] = −i(gµσ Uν − gνσ Uµ ) . (4.35)

To understand further the commutation relations (4.32) we decompose it into a purely

spatial part and a part which mixes time and space (like magnetic and electric fields for the
field strength Fµν . For spatial indices (4.32) becomes

[Mij , Mkl ] = −i(δjk Mil − δik Mjl − δjl Mik + δil Mjk ) . (4.36)

Defining
Jm = 21 εmij Mij ⇒ Mij = εijm Jm , (4.37)

113
and similarly Jn = 21 εnkl Mkl we get

[Jm , Jn ] = −i εmij εnkl Mil = 21 i εmnj εilj Mil = i εmnj Jj . (4.38)

The commutation relations are identical with those obtained in (3.54) which is unsurpris-
ing since purely spatial Lorentz transformations reduce to the subgroup of rotations. As
previously, J = (J1 , J2 , J3 ) are identified with the angular momentum operators.
Besides the spatial commutators we consider also

[Mij , M0k ] = −i(δjk M0i − δik M0j ) , (4.39)

and
[M0i , M0j ] = −i Mij . (4.40)
Defining now
Ki = M0i , Ki † = Ki , (4.41)
and, using (4.37), (4.39) and (4.40) become

[Ji , Kj ] = i εijk Kk , (4.42)

and
[Ki , Kj ] = −i εijk Jk . (4.43)
The commutator (4.43) shows that K = (K1 , K2 , K3 ) is a vector operator, as in (3.231).
The − sign in the commutator is (4.43) reflects the non compact structure of the Lorentz
group SO(3, 1), if the group were SO(4) then gµν → δµν and there would be a +.
For δxµ = ω µ ν xν letting ωij = εijk θk and ω 0 i = ω i 0 = vi then we have, for t = x0 and
x = (x1 , x2 , x3 ),
δt = v ⋅ x , δx = θ × x + vt , (4.44)
representing an infinitesimal rotation and Lorentz boost. Using (4.29) with (4.37) and (4.41)
gives correspondingly
U [Λ] = 1 − i θ ⋅ J + i v ⋅ K , (4.45)
which shows that K is associated with boosts in the same way as J is with rotations, as
demonstrated by (3.50).
The commutation relations (4.38), (4.42) and (4.43) can be rewritten more simply by
defining
Ji± = 21 (Ji ± iKi ) , J+† = J− , (4.46)
when they become

[Ji+ , Jj+ ] = i εijk Jk+ , [Ji− , Jj− ] = i εijk Jk− , [Ji+ , Jj− ] = 0 . (4.47)

The commutation relations are then two commuting copies of the standard angular momen-
tum commutation relations although the operators J± are not hermitian.

114
4.3 Lorentz Group and Spinors

For SO(3, 1) there are corresponding spinorial representations just as for SO(3). For SO(3)
a crucial role was played by the three Pauli matrices σ. Here we define a four dimensional
extension by
σµ = (12 , σ) = σµ † , σ̄µ = (12 , −σ) = σ̄µ † . (4.48)
Both σµ and σ̄µ form a complete set of hermitian 2 × 2 matrices. As a consequence of (3.20)
we have
σµ σ̄ν + σν σ̄µ = 2gµν 1 , σ̄µ σν + σ̄ν σµ = 2gµν 1 , (4.49)
and also
tr(σµ σ̄ν ) = 2gµν . (4.50)
1
Hence for a 2 × 2 matrix A we may write A = 2 tr(σ̄ µ A) σµ .

4.3.1 Isomorphism SO(3, 1) ≃ Sl(2, C)/Z2

The relation of SO(3, 1) to the group of 2 × 2 complex matrices with determinant one is an
extension of the isomorphism SO(3) ≃ SU (2)/Z2 . To demonstrate this we first describe the
one to one correspondence between real 4-vectors xµ and hermitian 2 × 2 matrices x where

xµ → x = σµ xµ = x† , xµ = 1
2 tr(σ̄ µ x) . (4.51)

With the standard conventions in (3.19)

x0 + x3 x1 − ix2
x=( 1 ). (4.52)
x + ix2 x0 − x3

Hence
det x = x2 ≡ gµν xµ xν . (4.53)
Defining
x̄ = σ̄µ xµ , (4.54)
then (4.49) are equivalent to
x x̄ = x2 1 , x̄x = x2 1 . (4.55)

For any A ∈ Sl(2, C) we may then define a linear transformation xµ → x′µ by

x → x′ = AxA† = x′† . (4.56)

A

where, using det A = det A† = 1,

det x′ = det x ⇒ x′2 = x2 . (4.57)

Hence this must be a real Lorentz transformation

x′µ = Λµ ν xν . (4.58)

115
From (4.56) this requires

σµ Λµ ν = Aσν A† , Λµ ν = 1
2 tr(σ̄ µ Aσν A† ) . (4.59)

To establish the converse we may use σν A† σ̄ ν = 2 tr(A† ) I to give

Λµ µ = ∣tr(A)∣2 , σµ Λµ ν σ̄ ν = 2 tr(A† ) A , (4.60)

and hence, for trA = eiφ ∣trA∣,

σµ Λµ ν σ̄ ν
A = eiφ √ , (4.61)
2 Λµ µ
where the phase eiα may be determined up to ±1 by imposing det A = 1. Hence for any
A ∈ Sl(2, C), ±A ↔ Λ for any Λ ∈ SO(3, 1). All elements in Sl(2, C) are continuously
connected to the identity so (4.51) does not allow for spatial or time reflections.
As special cases if A† = A−1 , so that A ∈ SU (2), it is easy to see that x′0 = x0 in (4.56)
and this is just a rotation of x as given by (3.27) and (3.30). If A† = A then Λ, given by
(4.59), is symmetric so this is a boost. Taking

AB (α, n) = cosh 12 α 1 + sinh 21 α n ⋅ σ , −∞ < α < ∞ , (4.62)

corresponds to the Lorentz boost in (4.23). Rotations remain in the form in (3.38).
For a general infinitesimal Lorentz transformation as in (4.14) then, using Λµ µ = 4 to
this order and σµ σ̄ µ = 4 1, (4.61) gives

A = 1 + 14 ω µν σµ σ̄ν , (4.63)

setting α = 0, since tr(ω µν σµ σ̄ν ) = 0 as a consequence of ω µν = −ω νµ . From (4.63)

A† = 1 − 14 ω µν σ̄µ σν . (4.64)

Alternatively, with these expressions for A, A† ,

Aσρ A† = σρ + 41 ω µν (σµ σ̄ν σρ − σρ σ̄µ σν ) = σρ + 21 ω µν (gνρ σµ − gρµ σν ) , (4.65)

using, from (4.49),

σµ σ̄ν σρ = gνρ σµ − σµ σ̄ρ σν , σρ σ̄µ σν = 2gρµ σν − σµ σ̄ρ σν , (4.66)

and therefore (4.65) verifies Aσν A† = σµ Λµ ν with Λµ ν given by (4.14).

In general (4.63),(4.64) may be written as

A = 1 − i 12 ω µν sµν , A† = 1 + i 12 ω µν s̄µν , sµν = 12 i σ[µ σ̄ν] , s̄µν = 12 i σ̄[µ σν] , (4.67)

where sµν , s̄µν = sµν † are matrices each obeying the same commutation rules as Mµν in
(4.32). To verify this it is sufficient to check

sµν σρ − σρ s̄µν = i(gνρ σµ − gµρ σν ) , s̄µν σ̄ρ − σ̄ρ sµν = i(gνρ σ̄µ − gµρ σ̄ν ) . (4.68)

116
4.3.2 Spinors, Dotted and Undotted Indices

In a similar fashion to the discussion in section 3.14 spinors are defined to transform under
the action of the Sl(2, C) matrix A. Fundamental spinors ψ, χ are required to transform as

ψα → Aα β ψβ , χα → χβ (A−1 )β α , α, β = 1, 2 . (4.69)
A A

We may also, as hitherto, raise and lower spinor indices with the ε-symbols εαβ , εαβ , where
ε12 = ε21 = 1, so that the representations defined by ψα , χα in (4.69) are equivalent

ψ α = εαβ ψβ , χα = εαβ χβ , (4.70)

as, since det A = 135 ,

(A−1 )β α = εαγ Aγ δ εδβ . (4.71)

The crucial difference between spinors for the Lorentz group SO(3, 1) and those for
SO(3) is that conjugation now defines an inequivalent representation. Hence there are two
inequivalent two-component fundamental spinors. It is convenient to adopt the notational
convention that the conjugate spinors obtained from ψα , χα have dotted indices, α̇ = 1, 2.
In general complex conjugation interchanges dotted and undotted spinor indices. For ψ, χ
conjugation then defines the conjugate representation spinors

ψ̄α̇ = (ψα )∗ , χ̄α̇ = (χα )∗ , (4.72)

which have the transformation rules, following from (4.69),

ψ̄α̇ → ψ̄β̇ (Ā−1 )β̇ α̇ , χ̄α̇ → Āα̇ β̇ χ̄β̇ , (4.73)

A A

for
(Ā−1 )α̇ β̇ = (Aβ α )∗ or Ā−1 = A† . (4.74)
Both A, Ā ∈ Sl(2, C) and obey the same group multiplication rules, since A1 A2 = Ā1 Ā2 .
The corresponding ε-symbols, εα̇β̇ , εα̇β̇ , allow dotted indices to raised and lowered,

ψ̄ α̇ = εα̇β̇ ψ̄β̇ , χ̄α̇ = εα̇β̇ χ̄β̇ , (4.75)

in accord with the conjugation of (4.70).

In terms of these conventions the hermitian 2 × 2 matrices defined in (4.48) are written
in terms of spinor index components as

(σµ )αα̇ , (σ̄µ )α̇α , (4.76)

where
(σ̄µ )α̇α = εα̇β̇ εαβ (σµ )β β̇ , (σµ )αα̇ = εαβ εα̇β̇ (σ̄µ )β̇β . (4.77)
35
Using (3.289), εαγ Aγ δ εδβ = δβα tr(A) − Aβ α = (A−1 )β α , since for any 2 × 2 matrix the characteristic
equation requires A2 − tr(A) A + det A 1 = 0, so that if det A = 1 then A−1 = tr(A) I − A.

117
With the definitions in (4.51) and (4.54) then (4.77) requires tr(xx̄) = 2 det x = 2x2 . Using
the definition of Ā we may rewrite (4.59) in the form

A σν Ā−1 = σµ Λµ ν , Ā σ̄ν A−1 = σ̄µ Λµ ν , (4.78)

showing the essential symmetry under A ↔ Ā.

The independent fundamental spinors ψ, χ and their conjugates ψ̄, χ̄ can be combined
as a single 4-component Dirac36 spinor together with its conjugate in the form

ψ
Ψ = ( α̇α ) , Ψ̄ = (χα ψ̄α̇ ) , (4.79)
χ̄

0 1
where Ψ̄ = Ψ† ( ). Correspondingly there are 4 × 4 Dirac matrices
1 0

0 σµ
γµ = ( ). (4.80)
σ̄µ 0

These satisfy, by virtue of (4.49), the Dirac algebra

γµ γν + γν γµ = 2gµν 14 . (4.81)

For these Dirac matrices

0 1
γ0 γµ γ0 = γµ † since γ0 = ( ), (4.82)
1 0

and from (4.77)

εαβ 0 ε 0
Cγµ C −1 = −γµ T for C = ( ), C −1 = ( αβ α̇β̇ ) . (4.83)
0 εα̇β̇ 0 ε

4.3.3 Tensorial Representations

Both vector and spinor tensors are naturally defined in terms of the tensor products of
vectors satisfying (4.6) and correspondingly spinors satisfying (4.69) or (4.73). Thus for a
purely contragredient rank n tensor

T µ1 ...µn → Λµ1 ν1 . . . Λµn νn T ν1 ...νn . (4.84)

Λ

For a general spinor with 2j lower undotted indices and 2̄ lower dotted indices

Υα1 ...α2j ,α̇1 ...α̇2̄ → Aα1 β1 . . . Aα2j β2j Υβ1 ...β2j ,β̇1 ...β̇2̄ (Ā−1 )β̇1 α̇1 . . . (Ā−1 )β̇2̄ α̇2̄ . (4.85)
A

The invariant tensors are just those already met together with the 4-index ε-symbol,

g µν , εµνσρ , εαβ , εα̇β̇ , (4.86)

36
Paul Adrian Maurice Dirac, 1902-84, English. Nobel prize, 1933.

118
as well as all those derived from these by raising or lowering indices. Here ε0123 = 1 while
ε0123 = −1.
To obtain irreducible tensors it is sufficient to consider spinorial tensors as in (4.85)
which are totally symmetric in each set of indices
Υα1 ...α2j ,α̇1 ...α̇2̄ = Υ(α1 ...α2j ),(α̇1 ...α̇2̄ ) . (4.87)
The resulting irreducible spinorial representation of SO(3, 1) is labelled (j, ̄). Under com-
plex conjugation (j, ̄) → (̄, j). Extending the counting in the SO(3) case, it is easy to
see that the dimension of the space of such tensors is (2j + 1)(2̄ + 1). The fundamental
spinors transform according to the ( 21 , 0) and (0, 12 ) representations while the Dirac spinor
corresponds to ( 12 , 0) ⊕ (0, 21 ). These representations are not unitary since there is no posi-
tive group invariant scalar product, for the simplest cases of a vector or a ( 12 , 0) spinor the
scalar products gµν V µ V ν or εβα ψα ψβ clearly have no definite sign.
The tensors products of irreducible tensors as in (4.87) may be decomposed just as for
SO(3) spinors giving
(j1 , ̄1 ) ⊗ (j2 , ̄2 ) ≃ ⊕ (j, ̄) . (4.88)
∣j1 −j2 ∣≤j≤j1 +j2
∣̄1 −̄2 ∣≤̄≤̄1 +̄2

Rank n vectorial tensors are related to spinorial tensors as in (4.85) for 2j = 2̄ = n by
Tµ1 ...µn = Υα1 ...αn ,α̇1 ...α̇n (σ̄µ1 )α̇1 α1 . . . (σ̄µn )α̇n αn . (4.89)
If Υ is irreducible, as in (4.87), corresponding to the ( 12 n, 12 n) real representation, then
Tµ1 ...µn is symmetric and traceless.
A corollary of εµνσρ being an invariant tensor is, from (4.78),
A εµνσρ σµ σ̄ν σσ σ̄ρ A−1 = εµνσρ σµ σ̄ν σσ σ̄ρ , Ā εµνσρ σ̄µ σν σ̄σ σρ Ā−1 = εµνσρ σ̄µ σν σ̄σ σρ . (4.90)
By virtue of Schur’s lemma these products of σ-matrices must be proportional to the iden-
tity. With (3.20) we get
1 µνσρ 1 µνσρ
24 ε σµ σ̄ν σσ σ̄ρ = σ0 σ̄1 σ2 σ̄3 = i 1 , 24 ε σ̄µ σν σ̄σ σρ = σ̄0 σ1 σ̄2 σ3 = −i 1 , (4.91)
using (σ0 σ̄1 σ2 σ̄3 )2 = σ0 σ̄1 σ2 σ̄3 σ3 σ̄2 σ1 σ̄0 = −1, and similarly (σ̄0 σ1 σ̄2 σ3 )2
= −1, by virtue
of (4.49). The two identities in (4.91) are related by conjugation. In terms of the Dirac
matrices defined in (4.80)

1 µνσρ 12 0
24 ε γµ γν γσ γρ = γ0 γ1 γ2 γ3 = iγ5 , γ5 = (
0 −12
). (4.92)

As a consequence of (4.91) we may further obtain37

1 µνσρ
2ε σσ σ̄ρ = −i σ [µ σ̄ ν] , 1 µνσρ
2ε σ̄σ σρ = i σ̄ [µ σ ν] . (4.93)
37
For a somewhat convoluted demonstration note, that since the indices only take four values,
εµνσρ
σ[µ σ̄ν σσ σ̄ρ σλ] = 51 εµνσρ (σµ σ̄ν σσ σ̄ρ σλ − σµ σ̄ν σσ σ̄λ σρ + σµ σ̄ν σλ σ̄σ σρ − σµ σ̄λ σν σ̄σ σρ + σλ σ̄µ σν σ̄σ σρ ) = 0.
Then using (4.49) move σλ or σ̄λ to the right giving εµνσρ σµ σ̄ν σσ σ̄ρ σλ + 4 ελ νσρ σν σ̄σ σρ = 0. Hence, with
(4.91), iσλ = − 61 ελ νσρ σν σ̄σ σρ . Similarly iσ̄µ = 61 εµ νσρ σ̄ν σσ σ̄ρ . Using these results, i(σλ σ̄µ − σµ σ̄σ ) =
− 61 ελ νσρ (σν σ̄σ σρ σ̄µ + σµ σ̄ν σσ σ̄ρ ). The right hand side may be simplified using (4.49) again and leads to
just (4.93).

119
Since tr(σ[µ σ̄ν] ) = tr(σ̄[µ σν] ) = 0, (εσ[µ σ̄ν] )αβ , (σ̄[µ σν] ε)α̇β̇ are symmetric in α ↔ β, α̇ ↔ β̇
respectively so that for (1, 0) or (0, 1) representations there are associated antisymmetric
tensors
fµν = 12 (εσ[µ σ̄ν] )αβ Υαβ , f¯µν = 21 (σ̄[µ σν] ε)α̇β̇ Ῡα̇β̇ , (4.94)
which satisfy fµν = 12 i εµν σρ fσρ , f¯µν = − 12 i εµν σρ f¯σρ . Only fµν + f¯µν is a real tensor.

4.4 Poincaré Group

The complete space-time symmetry group includes translations as well as Lorentz transfor-
mations. For a Lorentz transformation Λ and a translation a the combined transformation
denoted by (Λ, a) gives
xµ → x′µ = Λµ ν xν + aν . (4.95)
(Λ,a)

These transformations form a group since

(Λ2 , a2 )(Λ1 , a1 ) = (Λ2 Λ1 , Λ2 a1 + a2 ) , (Λ, a)−1 = (Λ−1 , −Λ−1 a) , (4.96)

with identity (I, 0). The corresponding group is the Poincaré group, sometimes denoted as
ISO(3, 1), if det Λ = 1. It contains the translation group T4 , formed by (I, a), as a normal
subgroup and also the Lorentz group, formed by (Λ, 0). A general element may be written as
(Λ, a) = (I, a)(Λ, 0) and the Poincaré Group can be identified with the semi-direct product
O(3, 1) ⋉ T4 .
If we define
(Λ, a) = (Λ2 , a2 )−1 (Λ1 , a1 )−1 (Λ2 , a2 )(Λ1 , a1 ) , (4.97)
then direct calculation gives

Λ = Λ2−1 Λ1−1 Λ2 Λ1 , a = Λ2−1 Λ1−1 (Λ2 a1 − Λ1 a2 − a1 + a2 ) . (4.98)

For infinitesimal transformations as in (4.26) we then have

Λµ ν = δ µν + [ω2 , ω1 ]µ ν , aµ = ω2µ ν a1ν − ω1µ ν a2ν . (4.99)

In a quantum theory there are associated unitary operators U [Λ, a] such that

U [Λ2 , a2 ]U [Λ1 , a1 ] = U [Λ2 Λ1 , Λ2 a1 + a2 ] . (4.100)

For an infinitesimal Lorentz transformation as in (4.14) and also for infinitesimal a we

require
U [Λ, a] = 1 − i 12 ω µν Mµν + i aµ Pµ , P µ † = Pµ , (4.101)
defining the generators Pµ in addition to Mµν = −Mνµ discussed in section 4.2. To derive
the commutation relations we extend (4.30) to give

U [Λ, a] = 1 − i [ω2 , ω1 ]µν Mµν + i (ω2a1 − ω1 a2 )µ Pµ

= U [Λ2 , a2 ]−1 U [Λ1 , a1 ]−1 U [Λ2 , a2 ]U [Λ1 , a1 ]
= 1 − [ 21 ω2µν Mµν − a2µ Pµ , 12 ω1σρ Mσρ − a1σ Pσ ] . (4.102)

120
Hence, in addition to the [M, M ] commutators which are given in (4.31) and (4.32), we
must have
[ 21 ω1σρ Mσρ , a2µ Pµ ] = i (ω1 a2 )µ Pµ , [a2µ Pµ , a1σ Pσ ] = 0 , (4.103)
or
[Mµν , Pσ ] = i(gνσ Pµ − gµσ Pν ) , [Pµ , Pσ ] = 0 . (4.104)
This agrees with general form in (4.35) and shows that Pµ is a covariant 4-vector operator.
Since (Λ, 0)(I, a)(Λ, 0)−1 = (I, Λa) and using (Λa)µ Pµ = aµ (P Λ)µ we have for finite Lorentz
transformations
U [Λ, 0] Pµ U [Λ, 0]−1 = Pν Λν µ . (4.105)

If we decompose
P µ = (H, P) , Pµ = (H, −P) , (4.106)
then using (4.37) and (4.41) the commutation relations become

[Ji , H] = 0 , [Ji , Pj ] = i εijk Pk , (4.107)

and
[Ki , H] = i Pi , [Ki , Pj ] = i δij H . (4.108)

4.5 Irreducible Representations of the Poincaré Group

It is convenient to write

U [Λ, a] = T [a]U [Λ] , U [Λ, 0] = U [Λ] , T [a] = U [1, a] , (4.109)

where T [a] are unitary operators corresponding to the abelian translation group T4 . In
general
µ
T [a] = eia Pµ . (4.110)
As a consequence of (4.100)
U [Λ]T [a] = T [Λa]U [Λ] . (4.111)

The irreducible representations of the the translation subgroup T4 of the Poincaré Group
are one-dimensional and are defined in terms of vector ∣p⟩ such that
µp
Pµ ∣p⟩ = pµ ∣p⟩ , T [a]∣p⟩ = eia µ
∣p⟩ , (4.112)

for any real 4-vector pµ which labels the representation. As a consequence of (4.105)

Pµ U [Λ]∣p⟩ = (pΛ−1 )µ U [Λ]∣p⟩ , (4.113)

so that U [Λ] acting on the states {∣p⟩} generates a vector space V such that ∣p′ ⟩, ∣p⟩ belong
to V if p′ µ = (pΛ−1 )µ for some Lorentz transformation Λ. All such p′ , p satisfy p′2 = p2
and conversely for any p′ , p satisfying this there is a Lorentz transformation linking p′ , p.
The physically relevant cases arise for p2 ≥ 0 and also we require, restricting Λ ∈ SO(3, 1)↑ ,
p0 , p′0 ≥ 0.

121
 The construction of representations of the Poincaré group is essentially identical with
the method of induced representations described in 2.2 for G = SO(3, 1)↑ ⋉ T4 . A subgroup
H is identified by choosing a particular momentum p̊ and then defining

Gp̊ = {Λ ∶ Λp̊ = p̊} , (4.114)

the stability group or little group for p̊, the subgroup of SO(3, 1)↑ leaving p̊ invariant as
discussed in 1.3. For a space Vp̊ formed by states {∣p̊⟩} (additional labels are here suppressed)
where
µ
Pµ ∣p̊⟩ = p̊µ ∣p̊⟩ , T [a]∣p̊⟩ = eia p̊µ ∣p⟩ , (4.115)
then Vp̊ must form a representation space for Gp̊ since U [Λ] ∣p̊⟩ ∈ Vp̊ for any Λ ∈ Gp̊ by virtue
of (4.114). Hence Vp̊ defines a representation for H = Gp̊ ⊗ T4 . The cosets G/H are then
labelled, for all p such that p2 = p̊2 , by any L(p) ∈ SO(3, 1)↑ where

pµ = (p̊L(p)−1 )µ , or equivalently pµ = L(p)µ ν p̊ν , (4.116)

and, following the method of induced representations, a representation space for a repre-
sentation of G is then defined in terms of a basis

∣p⟩ = U [L(p)] ∣p̊⟩ ∈ Vp , for all ∣p̊⟩ ∈ Vp̊ . (4.117)

Finding a representation of the Poincaré group then requires just the determination of
U [Λ] ∣p⟩ for arbitrary Λ. Clearly, by virtue of (4.113), U [Λ] ∣p⟩ must be a linear combination
of all states {∣p′ ⟩} where p′µ = Λµ ν pν . Since p′µ = L(p′ )µ ν p̊ν we have

(L(p′ )−1 ΛL(p))µ ν p̊ν = p̊µ . (4.118)

It follows that
L(Λp)−1 ΛL(p) = Λ̊p ∈ Gp̊ , (4.119)
and hence
U [Λ] ∣p⟩ = U [L(Λp)] U [Λ̊p ] ∣p̊⟩ ∈ VΛp , (4.120)
where U [Λ̊p ]∣p̊⟩ is determined by the representation of Gp̊ on Vp̊ .
For physical interest there are two distinct cases to consider.

4.5.1 Massive Representations

Here we assume p2 = m2 > 0. It is simplest to choose for p̊ the particular momentum

p̊µ = (m, 0) , (4.121)

and, since p̊ has no spatial part, then

Gp̊ ≃ SO(3) , (4.122)

122
since the condition Λp̊ = p̊ restricts Λ to the form given in (4.17). As in (4.116) L(p), for
any p such that p2 = m2 , p0 > 0, is then a Lorentz transformation such that pµ = L(p)µ ν p̊ν .
With (4.17) defining ΛR for any R ∈ SO(3), then (4.119) requires
L(Λp)−1 ΛL(p) = ΛR(p,Λ) , R(p, Λ) ∈ SO(3) . (4.123)
R(p, Λ) is a Wigner rotation. (4.123) ensures that
U [L(Λp)]−1 U [Λ] ∣p⟩ = U [ΛR(p,Λ) ] ∣p̊⟩ . (4.124)
For any R, U [ΛR ]∣p̊⟩ is an eigenvector of P µ with eigenvalue p̊µ and so is a linear combi-
nation of all states {∣p̊⟩}. In this case Vp̊ must form a representation space for SO(3). For
irreducible representations Vp̊ then has a basis, as described in section 3.5, which here we
label by s = 0, 12 , 1 . . . and s3 = −s, −s + 1, . . . , s. Hence, assuming {∣p̊, s s3 ⟩} forms such an
irreducible space,
(s)
U [ΛR ] ∣p̊, s s3 ⟩ = ∑ ∣p̊, s s′3 ⟩ Ds′ 3 s3 (R) , (4.125)
s′3

with D(s) (R) standard SO(3) rotation matrices. Extending the definition (4.117) to define
a corresponding basis for any p
∣p, s s3 ⟩ = U [L(p)] ∣p̊, s s3 ⟩ , (4.126)
then applying (4.125) in (4.124) gives
(s)
U [Λ] ∣p, s s3 ⟩ = ∑ ∣Λp, s s′3 ⟩ Ds′3 s3 (R(p, Λ)) , (4.127)
s′3

with R(Λp, Λ′ )R(p, Λ) = R(p, Λ′ Λ).

The states {∣p, s s3 ⟩ ∶ p2 = m2 , p0 > 0} then provide a basis for an irreducible representa-
tion space Vm,s for SO(3, 1)↑ . The representation extends to the full Poincaré group since
for translations, from (4.112),
µ
T [a] ∣p, s s3 ⟩ = eipµ a ∣p, s s3 ⟩ . (4.128)
The states ∣p, ss3 ⟩ are obviously interpreted as single particle states for a particle with mass
m and spin s.
In terms of these states there is a group invariant scalar product
⟨p′ , s s′3 ∣p, s s3 ⟩ = (2π)3 2p0 δ 3 (p′ − p) δs3 s′3 , (4.129)
which is positive so the representation is unitary. The invariance under Lorentz transfor-
mations follows from
d3 p 1
d4 p δ(p2 + m2 )θ(p0 ) = where p0 = (p2 + m2 ) 2 . (4.130)
2p0

The precise definition of the representation depends on the choice of L(p) satisfying
(4.116). This does not specify L(p) uniquely since if L(p) is one solution so is L(p)Λ for
any Λ ∈ Gp̊ . One definite choice is to take
L(p) = B(α, p̂) for pµ = B(α, p̂)µ ν p̊ν = m(cosh α, sinh α p̂) , (4.131)

123
where B(α, p̂) is the boost Lorentz transformation defined in (4.23). Then in (4.126)

U [L(p)] → U [B(α, p̂)] = eiα p̂⋅K . (4.132)

Using (4.131) if we consider a rotation ΛR then L(ΛR p) = B(α, p̂R ) and, by virtue of
(4.25),
B(α, p̂R )−1 ΛR B(α, p̂) = ΛR , (4.133)
so that
(s)
U [ΛR ] ∣p, s s3 ⟩ = U [L(ΛR p)] U [ΛR ] ∣p̊, s s3 ⟩ = ∑ ∣ΛR p, s s′3 ⟩ Ds′3 s3 (R) . (4.134)
s′3

The Wigner rotation given by (4.123) with this definition of L(p) is then just R(p, ΛR ) = R.

4.5.2 Helicity States

An alternative prescription for L(p) in (4.117), giving a different but equivalent basis for
Vm,s , is to first boost along the 3-direction and then rotate so that

e3 → p̂(θ, φ) = cos θ e3 + sin θ(cos φ e1 + sin φ e2 ) , (4.135)

which is just the radial unit vector in spherical polar coordinates. This rotation corresponds
to Rφ,θ,−φ , in terms of Euler angles, or equivalently R(Θ, n) with n = − sin φ e1 + cos φ e2 .
Hence p = L(p)p̊, with pµ = (p0 , ∣p∣ p̂) can be obtained by taking
cosh α 0 0 sinh α
⎛ cos φ sin θ sinh α sin2 φ+cos2 φ cos θ cos φ sin φ(cos θ−1) cos φ sin θ cosh α ⎞
L(p) = ΛR(Θ,n) B(α, e3 ) = sin φ sin θ sinh α cos φ sin φ(cos θ−1) cos2 φ+sin2 φ cos θ sin φ sin θ cosh α
,
⎝ cos θ sinh α − cos φ sin θ − sin φ sin θ cos θ cosh α
⎠
pµ = m(cosh α , sinh α p̂(θ, φ)) . (4.136)

Helicity states are then defined by

∣p, h⟩ = U [R(Θ, n)] U [B(α, e3 )] ∣p̊, h⟩ , U [R(Θ, n)] = e−iΘn⋅J = e−iφJ3 e−iθJ2 eiφJ3 , (4.137)

suppressing the label s for the particle spin which is fixed. Since J3 commutes with
U [B(α, e3 )] and
e−iΘn⋅J J3 eiΘn⋅J = p̂ ⋅ J , (4.138)
we must have
p̂ ⋅ J ∣p, h⟩ = h ∣p, h⟩ , (4.139)
so that h is the component of spin along the direction of motion, or helicity.
Wigner rotations are defined just as in (4.123). Helicity states transform very simply
under rotations since for any rotation R

R R(Θ, n) = R(Θ′ , n′ ) R(χ, e3 ) = Rφ′ ,θ′ ,−φ′ R(χ, e3 ) , n′ = − sin φ′ e1 + cos φ′ e2 , (4.140)

124
and
ΛR L(p) = L(ΛR p) , (ΛR p)µ = (p0 , ∣p∣ p̂(θ′ , φ′ )) , (4.141)
so that
U [ΛR ] ∣p, h⟩ = ∣ΛR p, h⟩ e−i hχ . (4.142)
For R = R(γ, e3 ) then χ = γ and θ′ = θ, φ′ = φ + γ. For R = R(η, e2 ) and taking φ = 0 then
χ = φ′ = 0 and θ′ = θ + η.
When Lorentz boosts are involved the Wigner rotation mixes different helicity states.
For a boost along the 3-direction

R(p, B(β, e3 )) = Rφ,χ,−φ , cot χ = cot θ cosh α + coth β cosec θ sinh α , (4.143)

so that
(s)
U [B(β, e3 )] ∣p, h⟩ = ∑ ∣p′ , h′ ⟩ Dh′ h (Rφ,χ,−φ ) (4.144)
h′

with p′ = B(β, e3 )p given in terms of α′ , θ′ , φ as in (4.136) with cosh α′ = cosh β cosh α +

sinh β sinh α cos θ, cot θ′ = cosh β cot θ + sinh β coth α cosec θ.

4.5.3 Massless Representations

The construction of representations for the massless case can be carried out in a similar
fashion to that just considered. When p2 = 0 then the method requires choosing a particular
momentum p̊ satisfying this from which all other momenta with p2 = 0 can be obtained by
a Lorentz transformation. There is no rest frame as in (4.121) and we now take

p̊µ = ω̊(1, 0, 0, 1) , ω̊ > 0 , (4.145)

with ω̊ some arbitrary fixed choice. It is then necessary to identify the little group in this
case as defined by (4.114). To achieve this we consider infinitesimal Lorentz transformations
as in (4.14) when the necessary requirement reduces to

ω µ ν p̊ν = 0 , ω µν = −ω νµ . (4.146)

This linear equation is easy to solve giving

ω03 = 0 , ω 1 0 = −ω 1 3 , ω 2 0 = −ω 2 3 , ω30 = 0 . (4.147)

These reduce the six independent ω µν = −ω νµ to three so that

1 µν
2 ω Mµν = ω 12 M12 + ω 01 (M01 + M31 ) + ω 02 (M02 + M32 ) . (4.148)

Identifying the operators

J3 = M12 , E1 = M01 + M31 = K1 + J2 , E2 = M02 + M32 = K2 − J1 , (4.149)

we find the commutators from (4.32), or from (3.54), (4.42) and (4.43),

[J3 , E1 ] = iE2 , [J3 , E2 ] = −iE1 , [E1 , E2 ] = 0 . (4.150)

125
 A unitary operator corresponding to finite group elements of Gp̊ is then

e−i(a1 E1 +a2 E2 ) e−iχJ3 , (4.151)

Noting that

a1χ cos χ − sin χ a1

e−iχJ3 (a1 E1 + a2 E2 )eiχJ3 = a1χ E1 + a2χ E2 , ( )=( )( ) , (4.152)
a2χ sin χ cos χ a2

then if (4.151) corresponds to a group element (χ, a1 , a2 ), with χ an angle with period 2π,
we have the group multiplication rule
′ ′
(χ′ , a′1 , a′2 )(χ, a1 , a2 ) = (χ′ + χ, a1χ + a′1 , a2χ + a′2 ) . (4.153)

The group multiplication rule (4.153) is essentially identical to (4.96). The group is then
isomorphic with the group formed by rotations and translations on two dimensional space,
so that for the massless case we have the little group

Gp̊ ≃ ISO(2) ≃ SO(2) ⋉ T2 . (4.154)

This group is isomorphic to the group of rotations and translations in two dimensions.
The representations of this group can be obtained in a very similar fashion to that of the
Poincaré group. Define vectors ∣b1 , b2 ⟩ such that

(E1 , E2 )∣b1 , b2 ⟩ = (b1 , b2 )∣b1 , b2 ⟩ , (4.155)

and then we assume, consistency with the group multiplication (4.153),

e−iχJ3 ∣b1 , b2 ⟩ = e−ihχ ∣b1χ , b2χ ⟩ , (4.156)

linking all (b1 , b2 ) with constant c = b1 2 + b2 2 . This irreducible representation of ISO(2),

labelled by c, h, is infinite dimensional. However there are one-dimensional representations,
corresponding to taking c = 0, generated from a vector ∣h⟩ such that

E1 ∣h⟩ = E2 ∣h⟩ = 0 , J3 ∣h⟩ = h∣h⟩ , (4.157)

so that the essential group action is

e−iχJ3 ∣h⟩ = e−ihχ ∣h⟩ . (4.158)

For applications to representations of the Poincaré group e−iχJ3 corresponds to a subgroup

of the SO(3) rotation group so it is necessary to require in (4.157) and (4.158)

h = 0, ± 21 , ±1, . . . . (4.159)

For the associated Lorentz transformations then a general element corresponding to the
little group is Λ(a1 ,a2 ) Λχ where

⎛ 1+ 2 (a1 +a2 ) a1 a2 − 12 (a12 +a22 ) ⎞

1 2 2 1 0 0 0
⎛ 0 cos χ − sin χ 0 ⎞
a1 1 0 −a1
Λ(a1 ,a2 ) = ⎜ a2 0 1 −a2 ⎟ , Λχ = ⎜ 0 sin χ cos χ 0 ⎟ . (4.160)
⎝ 21 (a12 +a22 ) a1 a2 1− 12 (a12 +a22 ) ⎠ ⎝0 0 0 1
⎠

126
It is easy to see that Λ(a1 ,a2 ) p̊ = Λχ p̊ = p̊ with p̊ as in (4.145).
The construction of the representation space Vh when p2 = 0 proceeds in a very similar
fashion as in the massive case. Neglecting infinite dimensional representations of the little
group, then starting from a vector ∣p̊, h⟩ satisfying

P µ ∣p̊, h⟩ = p̊µ ∣p̊, h⟩ , J3 ∣p̊, h⟩ = h∣p̊, h⟩ , (4.161)

a basis {∣p, h⟩ ∶ p2 = 0, p0 > 0}, for Vh is formed by

∣p, h⟩ = U [L(p)] ∣p̊, h⟩ , for pµ = L(p)µ ν p̊ν , (4.162)

where L(p) is assumed to be determined uniquely by p. Using

L(Λp)−1 ΛL(p) = Λ(a1 ,a2 ) Λχ ∈ Gp̊ , for a1,2 (p, Λ) , χ(p, Λ) , (4.163)

so that
U [Λ(a1 ,a2 ) ]U [Λχ ] ∣p̊, h⟩ = ∣p̊, h⟩ e−ihχ , (4.164)
then, for any Λ ∈ SO(3, 1)↑ , the action of the corresponding unitary operator on Vh is given
by
U [Λ] ∣p, h⟩ = ∣Λp, h⟩ e−ihχ(p,Λ) . (4.165)
Group multiplication requires χ(p, Λ) + χ(Λp, Λ′ ) = χ(p, Λ′ Λ).
For p̊ as in (4.145), and
pµ = ω(1, p̂) , ω > 0, (4.166)
then L(p), satisfying (4.116), is determined by assuming it is given by the expression (4.136)
with now eα = ω/ω̊. By the same arguments as for massive helicity states

p̂ ⋅ J ∣p, h⟩ = h ∣p, h⟩ , (4.167)

so that the component of the angular momentum along the direction of motion, or helicity,
is again h.
The irreducible representations of the Poincaré group for massless particles require only
a single helicity h, with values as in (4.159). If the symmetry group is extended to include
parity, corresponding to spatial reflections, then it is necessary for there to be particle states
with both helicities ±h. When parity is a symmetry there is an additional unitary operator
P with the action on the Poincaré group generators

PJP −1 = J , PKP −1 = −K , PHP −1 = H , PPP −1 = −P . (4.168)

In consequence PP⋅JP −1 = −P⋅J so that, from (4.167), P∣p, h⟩ must have helicity −h, so we
must have P∣p, h⟩ = η∣p, −h⟩, for some phase η, usually η = ±1. Thus photons have helicity
±1 and gravitons ±2. However neutrinos, if they were exactly massless, which is no longer
compatible with experiment, need only have helicity − 12 since their weak interactions do
not conserve parity and experimentally only involve − 12 helicity.

127
4.5.4 Two Particle States and Angular Momentum Decomposition

Scattering experiments generally involve collisions of two particles. So called in states are
formed from tensor products of free particle states in the distant past. A convenient basis
can be formed starting from two particle states in the centre of mass frame. Initially we
define
∣P̊ , pe3 , h1 h2 ⟩ = U1 [B(α1 , e3 )] ∣p̊1 , h1 ⟩1 U2 [B(α2 , e3 )] ∣p̊2 , −h2 ⟩2 , (4.169)
with α1 , α2 such that

B(α1 , e3 )µ ν p̊1ν = (E1 , p e3 ) , B(α2 , e3 )µ ν p̊2ν = (E2 , −p e3 ) , p > 0,

1 1
µ
P̊ = (E, 0) , E = E1 + E2 E1 = (m12 2
+ p ) , E2 =
2 (m22 +p ) . 2 2 (4.170)

A general centre of mass state is obtained by a rotation taking e3 → p̂ as in (4.135)

∣P̊ , p p̂, h1 h2 ⟩ = U [Rφ,θ,−φ ] ∣P̊ , pe3 , h1 h2 ⟩ ,

∣P̊ , p p̂, h1 h2 ⟩ = ∣p1 , h1 ⟩1 ∣p2 , h2 ⟩2 , p1µ = (E1 , p p̂) , p2µ = (E2 , −p p̂) , (4.171)

where U [R] is a rotation generated by the total angular momentum J = J1 + J2 and p = p p̂

as in (4.135). It is easy to see that

p̂ ⋅ J ∣P̊ , p, h1 h2 ⟩ = h ∣P̊ , p, h1 h2 ⟩ , h = h1 − h2 . (4.172)

A basis for two particle states ∣P, p, h1 h2 ⟩ with arbitrary total 4-momentum P µ can
be obtained by acting on ∣P̊ , p, h1 h2 ⟩ with a boost taking P̊ µ → P µ . For any translation
invariant operator A acting on two particle states the overall momentum conservation δ-
function can factored off to defined a reduced matrix element for centre of mass states
E ′ ′ ′
⟨P ′ , p′ , h1′ h2′ ∣A∣P, p, h1 h2 ⟩ = (2π)4 δ 4 (P ′ − P ) ⟨p̂ , h1 h2 ∣∣A∣∣p̂, h1 h2 ⟩ . (4.173)
p
Since
E 2 ′
(2π)6 4p10 p20 δ 3 (p1′ − p1 ) δ 3 (p2′ − p2 ) = (2π)4 δ 4 (p1′ + p2′ − p1 − p2 ) (4π)2 δ (p̂ , p̂) , (4.174)
p

with E, p, p̂′ , p̂ defined by transforming p1 , p2 to the centre of mass frame as in (4.171), then

⟨p̂′ , h1′ h2′ ∣∣1∣∣p̂, h1 h2 ⟩ = (4π)2 δ 2 (p̂′ , p̂) δh1′ h1 δh2′ h2 . (4.175)

Here δ 2 (p̂′ , p̂) is the delta function on the unit sphere so that ∫ dΩp̂ δ 2 (p̂′ , p̂)f (p̂) = f (p̂′ ).
Using the orthogonality condition (3.124) the centre of mass states in (4.169) can be
projected onto states of definite angular momentum by taking
1 (J)
∣P̊ , JM, h1 h2 ⟩ = NJ 2 ∫ dµθ,φ,ψ U [Rφ,θ,ψ ] ∣P̊ , pe3 , h1 h2 ⟩DM h (Rφ,θ,ψ )∗
8π SO(3)
1 2π 2π
(J) ∗
= NJ ∫ dφ ∫ dθ sin θ ∣P̊ , p p̂, h1 h2 ⟩DM h (Rφ,θ,−φ ) , (4.176)
4π 0 0

128
where for the ψ integration to be non zero requires m = h = h1 − h2 . These states are then
a basis for two particle states of total angular momentum J,
2π 2π
(J ′ ) (J)
⟨J ′ M ′ , h1′ h2′ ∣∣1∣∣JM, h1 h2 ⟩ = NJ 2∫ dφ ∫ dθ sin θ DM ′ h (Rφ,θ,−φ )DM h (Rφ,θ,−φ )∗ δh1′ h1 δh2′ h2
0 0
= δJ ′ J δM ′ M δh1′ h1 δh2′ h2 , for NJ 2 = 2J+1
4π . (4.177)

Using the completeness relation

1 (J) (J) ∗ 2 ′ ′ ′
∑ (2J +1) DM h (Rφ′ ,θ′ ,−φ′ ) DM h (Rφ,θ,−φ ) = δ (p̂ , p̂) = δ(cos θ −cos θ)δ(φ −φ) , (4.178)
4π J,M

(4.176) can be inverted as an expansion over angular momentum states

(J)
∣P̊ , p p̂, h1 h2 ⟩ = ∑ NJ ∣P̊ , JM, h1 h2 ⟩ DM h1 −h2 (Rφ,θ,−φ ) . (4.179)
J,M

For a rotationally scalar operator, which commutes with J,

(J)
⟨J ′ M ′ , h1′ h2′ ∣∣A∣∣JM, h1 h2 ⟩ = Ah1′ h2′ ,h1 h2 δJ ′ J δM ′ M , (4.180)

(J)
with Ah1 h2 independent of M by virtue of the Wigner-Eckart theorem. With the expansion
(4.179)
1 (J) (J) ∗ (J)
⟨p̂′ , h1′ h2′ ∣∣A∣∣p̂, h1 h2 ⟩ = ∑ (2J + 1) Ah1′ h2′ ,h1 h2 DM h′ (Rθ′ ,φ′ ,−φ′ ) DM h (Rφ,θ,−φ ) ,
4π J,M
h′ = h1′ − h2′ , h = h1 − h2 , (4.181)

The two D-functions can be combined using the group property but more simply p̂ can be
chosen to be along the 3-direction giving
1 (J) (J) i(h−h′ )φ
⟨p̂, h1′ h2′ ∣∣A∣∣e3 , h1 h2 ⟩ = ∑ (2J + 1) Ah1′ h2′ ,h1 h2 dhh′ (θ) e , (4.182)
4π J

for p̂ given in terms of θ, φ according (4.135).

This approach to the angular momentum decomposition of relativistic two particle
states was first introduced by Jacob and Wick.38 It avoids the complications of com-
bining spins and then orbital angular momentum using Clebsch-Gordan coefficients. It was
of course designed to apply to scattering amplitudes for spinning particles where taking
A → S, scattering operator, in (4.182) gives the partial wave decomposition. By virtue
(J)
of (4.177) the unitarity condition for elastic scattering amplitudes Sh1′ h2′ ,h1 h2 reduces to
(J) (J)
∑h1′′ ,h2′′ Sh1′′ ,h2′′ ,h1′ h2′ ∗ Sh1′′ h2′′ ,h1 h2 = δh1′ h1 δh2′ h2 . For the spinless case, using (3.108), (4.182)
becomes the standard expansion in terms of Legendre polynomials.
38
Maurice René Michel Jacob, 1933-2007, French. Gian Carlo Wick, 1909-1992, Italian.

129
4.5.5 Spinorial Treatment

Calculations involving Lorentz transformations are almost always much simpler in terms
of Sl(2, C) matrices, making use of the isomorphism described in section 4.3, rather than
working out products of 4 × 4 matrices Λ. As an illustration we re-express some of the above
discussion for massless representations in terms of spinors.
Defining
p0 + p3 p1 − ip2
pαα̇ = pµ (σµ )αα̇ , [pαα̇ ] = ( ), (4.183)
p1 + ip2 p0 − p3
similarly to (4.52), then for p2 = 0

det[pαα̇ ] = 0 ⇒ pαα̇ = λα λ̄α̇ . (4.184)

The spinor λα and its conjugate λ̄α̇ are arbitrary up to the U (1) transformation given by
λα → λα e−iη , λ̄α̇ → λ̄α̇ eiη . To determine λα precisely we choose the phase so that λ1 is real.
For p̊ given by (4.145) then, for simplicity choosing 2ω̊ = 1,

1 0 1
[p̊αα̇ ] = ( ) ⇒ λ̊ = ( ) . (4.185)
0 1 0

If pi2 = pj2 = 0 then

pi,αα̇ = λi,α λ̄i,α̇ , pj,αα̇ = λj,α λ̄j,α̇ , (4.186)
and

2 pi ⋅ pj = pi,αα̇ pjα̇α = ⟨ij⟩[ij] , ⟨ij⟩ = εαβ λi,α λj,β , [ij] = εα̇β̇ λi,β λ̄j,β̇ . (4.187)

As a consequence of the map SO(3, 1) → Sl(2, C) then for the massless case

L(p)p̊ = p ⇒ A(λp )λ̊ = λp , (4.188)

defines p → λp uniquely, at least up to a sign, satisfying (4.184). For any Lorentz transfor-
mation Λ → AΛ then (4.163) becomes equivalently

A(λΛp )−1 AΛ A(λp ) = A(a1 ,a2 ) Aχ , (4.189)

where, from (4.160), we have correspondingly under SO(3, 1) → Sl(2, C)

⎛e− 2 iχ
1
1 a1 − ia2 0 ⎞
Λ(a1 ,a2 ) → A(a1 ,a2 ) = ( ), Λχ → Aχ = 1 . (4.190)
0 1 ⎝ 0 e 2 iχ ⎠

With the definition (4.188) of λp , (4.189) implies

AΛ λp = λΛp e− 2 iχ(p,Λ) .
1
(4.191)

130
 The spinorial formalism a convenient method of calculating χ(p, Λ). For helicity states
with L(p) as in (4.136) the corresponding Sl(2, C) matrix is then

− sin 12 θ e−iφ − 2 α ⎞
1 1
⎛ cos 12 θ e 2 α
AR (Θ, n)AB (α, e3 ) = . (4.192)
⎝sin 21 θ eiφ + 2 α cos 21 θ e− 2 α ⎠
1 1

In (4.188) we can then take

λ1 −λ2∗ /(λ12 + ∣λ2 ∣2 ) 1 p1 + i p 2

A(λ) = ( ), λp,1 = (p0 + p3 ) 2 , λp,2 = , (4.193)
λ2 λ1∗ /(λ12 + ∣λ2 ∣2 ) (p0 + p3 ) 2
1

where the phase ambiguity in λ is resolved by requiring λ1 to be real.

As a illustration we consider three examples

Λ = ΛR(σ,e3 ) ,
λΛp,1 = λp,1 , λΛp,2 = eiσ λp,1 , χ(p, Λ) = σ ,
Λ = ΛR(ρ,e2 ) ,
λ cos 21 ρ − sin 12 ρ λp,1 1 iη 1
1 p + ip
2
( Λp,1 ) = ( ) ( ) e 2 , tan 1
η = tan ρ , χ(p, Λ) = η ,
λΛp,2 sin 12 ρ cos 12 ρ λp,2 2 2 p0 + p3
Λ = B(β, e3 ) ,
(e2β − 1)λp,1 λp,2∗
λΛp,1 = e 2 β λp,1 , λΛp,2 = e− 2 β λp,1 ,
1 1
χ = 0 , a1 − ia2 = .
(λp,12 + ∣λp,2 ∣2 )(λp,12 + e−2β ∣λp,2 ∣2 )
(4.194)

4.6 Casimir Operators

For the rotation group then from the generators J it is possible to construct an invariant
operator J2 which commutes with all generators, as in (3.83), so that all vectors belonging
to any irreducible representation space have the same eigenvalue, for Vj , j(j + 1). Such
operators, which are quadratic or possibly higher order in the generators, are generically
called Casimir39 operators. Of course only algebraically independent Casimir operators are
of interest.
For the Lorentz group, SO(3, 1), there are two basic Casimir operators which can be
formed from Mµν using the invariant tensors
1 µν 1 µνσρ
4 M Mµν , 8ε Mµν Mσρ . (4.195)

In terms of the generators J, K, defined in (4.37),(4.41), and then J± , defined in (4.46),

1 µν
4 M Mµν = 12 (J2 − K2 ) = J+2 + J−2 ,
1 µνσρ
8ε Mµν Mσρ = J ⋅ K = −i(J+2 − J−2 ) . (4.196)
39
Hendrik Brugt Gerhard Casimir, 1909-2000, Dutch.

131
Since J± both obey standard angular momentum commutation relations, as in (4.47), then
for finite dimensional irreducible representations

J+2 → j(j + 1) 1 , J−2 → ̄(̄ + 1) 1 , j, ̄ = 0, 12 , 1, 32 , . . . . (4.197)

For the fundamental spinor representation the generators sµν = 21 i σ[µ σ̄ν] , as in (4.67),
the associated Casimir operators become
1 µν 1 µ ν 3
4 s sµν = − 32 σ σ̄ (σµ σ̄ν − σν σ̄ν ) = 4 1,
1 µνσρ 1 µνσρ
8ε sµν sσρ = 32 ε σµ σ̄ν σσ σ̄ρ = − 43 i 1 , (4.198)

using (4.49) and (4.91). As expected this is in accord with (4.196) and (4.197) for j = 12 ,
̄ = 0. Conversely for s̄µν the role of j and ̄ are interchanged since this is the conjugate
representation.
For the Poincaré group then (4.195) no longer provides Casimir operators because they
fail to commute with Pµ . There is now only a single quadratic Casimir

P 2 = P µ Pµ , (4.199)

whose eigenvalues acting on the irreducible spaces Vm,s , Vs , corresponding to the spaces of
relativistic single particle states, give the invariant m2 in the massive case or zero in the
massless case. However the irreducible representations are also characterised by a spin label
s, helicity in the massless case. To find an invariant characterisation of this we introduce
the Pauli-Lubanski vector,

Wµ = 1 µνσρ
2ε Pν Mσρ = 1 µνσρ
2ε Mσρ Pν . (4.200)

Using εµνσρ Pν Pσ = 0 we have

[W µ , Pν ] = 0 . (4.201)
µνσρ µ
Since ε is an invariant tensor then W should be a contravariant 4-vector, to verify this
we may use

[W µ , 21 ω σρ Mσρ ] = − 12 i εµνσρ (Pλ ω λ ν Mσρ + Pν Mλρ ω λ σ + Pν Mσλ ω λ σ )

= 21 i ω µ λ ελνσρ Pν Mσρ = i ω µ λ W λ , (4.202)

to obtain
[W µ , Mσρ ] = i(δ µσ Wρ − δ µρ Wσ ) . (4.203)
With (4.201) and (4.203) we may then easily derive

[W µ , W ν ] = i εµνσρ Pσ Wρ . (4.204)

It follows from (4.201) and (4.203) that

Wµ W µ , (4.205)

is a scalar commuting with Pν , , Mσρ and so providing an additional Casimir operator.

132
 For the massive representations then, for p̊ as in (4.121),

W 0 ∣p̊, s s3 ⟩ = 0 , W i ∣p̊, s s3 ⟩ = −m εijk Mjk ∣p̊, s s3 ⟩ = −m Ji ∣p̊, s s3 ⟩ , (4.206)

so that
Wµ W µ ∣p̊, s s3 ⟩ = −m2 J2 ∣p̊, s s3 ⟩ = −m2 s(s + 1)∣p̊, s s3 ⟩ . (4.207)
Hence Wµ W µ has the eigenvalue −m2 s(s + 1) for all vectors in the representation space
Vm,s .
For the massless representations then, for p̊ as in (4.145),

W 1 ∣p̊, h⟩ = ω̊ E2 ∣p̊, h⟩ = 0 , W 2 ∣p̊, h⟩ = −ω̊ E1 ∣p̊, h⟩ = 0 ,

W 0 ∣p̊, h⟩ = − ω̊ J3 ∣p̊, h⟩ = −ω̊ h∣p̊, h⟩ , W 3 ∣p̊, h⟩ = −ω̊ J3 ∣p̊, h⟩ = −ω̊ h∣p̊, h⟩ , (4.208)

using (4.157). Since W µ , P µ are both contravariant 4-vectors the result (4.208) requires

(W µ + h P µ )∣p, h⟩ = 0 , (4.209)

for all vectors providing a basis for Vh . This provides an invariant characterisation of the
helicity h on this representation space.

4.7 Quantum Fields

To construct a relativistic quantum mechanics compatible with the general principles of

quantum mechanics it is essentially inevitable to use quantum field theory. The quantum
fields are required to have simple transformation properties under the symmetry trans-
formations belonging to the Poincaré group. For a simple scalar field, depending on the
space-time coordinates xµ , this is achieved by

U [Λ, a]φ(x)U [Λ, a]−1 = φ(Λx + a) , (4.210)

where U [Λ, a] are the unitary operators satisfying (4.100). For an infinitesimal transforma-
tion, with Λ as in (4.14) and U as in (4.101), this gives

− i[ 21 ω µν Mµν − aµ Pµ , φ(x)] = (ω µ ν xν + aµ )∂µ φ(x) , (4.211)

or

[Mµν , φ(x)] = −Lµν φ(x) , Lµν = i(xµ ∂ν − xν ∂µ ) , [Pµ , φ(x)] = −i∂µ φ(x) . (4.212)

Lµν and i∂µ obey the same commutation relations as Mµν and Pµ in (4.32) and (4.104).
Note that, with (4.106), [P, φ] = i ∇φ.
To describe particles with spin the quantum fields are required to transform according
to a finite dimensional representation of the Lorentz group so that (4.210) is extended to

U [Λ, a]φ(x)U [Λ, a]−1 = D(Λ)−1 φ(Λx + a) , (4.213)

133
regarding φ now as a column vector and suppressing matrix indices. For an infinitesimal
Lorentz transformation then assuming

D(Λ) = 1 − i 12 ω µν Sµν , Sµν = −Sνµ , (4.214)

the commutator with Mµν in (4.212) is extended to

[Mµν , φ(x)] = −(Lµν + Sµν )φ(x) . (4.215)

The matrix generators Sµν obey the same commutators as Mµν in (4.32).
The relation of the quantum fields to the particle state representations considered in 4.5
is elucidated by considering, considering first Vm,s ,

⟨0∣φ(x)∣p, s s3 ⟩ = u(p, s3 ) e−ip⋅x , p2 = m2 . (4.216)

Here ∣0⟩ is the vacuum state, which is just a singlet under the Poincaré group, U [Λ, a]∣0⟩ =
∣0⟩. It is easy to check that (4.216) is accord with translation invariance using (4.128).
Using (4.213), for a = 0, Λ → Λ−1 , with (4.127) we get
(s)
D(Λ) u(p, s3 ) = ∑ u(Λp, s′3 ) Ds′3 s3 (R(p, Λ)) , (4.217)
s′3

which is directly analogous to (4.127) but involves the finite dimensional representation
matrix D(Λ). u(p, s3 ) thus allows the complicated Wigner rotation of spin indices given
by R(p, Λ) to be replaced by a Lorentz transformation, in some representation, depending
just on Λ. To determine u(p, s3 ) precisely so as to be in accord with (4.217) it is sufficient
to follow the identical route to that which determined the states ∣p, s s3 ⟩ in 4.5.1. Thus it
is sufficient to require, as in (4.125),
(s)
D(ΛR ) u(p̊, s3 ) = ∑ u(p̊, s′3 ) Ds′3 s3 (R) , (4.218)
s′3

and then define, as in (4.126),

u(p, s3 ) = D(L(p)) u(p̊, s3 ) . (4.219)

For Λ reduced to a rotation ΛR , as in (4.17), the representation given by the matrices

D(ΛR ) decomposes into a direct sum of irreducible SO(3) representations D(j) (R). For
(4.218) to be possible this decomposition must include, by virtue of Schur’s lemmas, the
irreducible representation j = s, with any other D(j) , j ≠ s, annihilating u(p̊, s3 ).
For the zero mass case the discussion is more involved so we focus on a particular case
when the helicity h = 1 and the associated quantum field is a 4-vector Aµ . Replacing (4.216)
we require
⟨0∣Aµ (x)∣p, 1⟩ = µ (p) e−ip⋅x , p2 = 0 . (4.220)
µ (p) is referred to as a polarisation vector. For 4-vectors there is an associated represen-
tation of the Lorentz group which is just given, of course, by the Lorentz transformation
matrices Λ themselves. When p = p̊ as in (4.145) then from the little group transformations
as in (4.163) we require, for h = 1,

Λχ (p̊) = (p̊) e−iχ . (4.221)

134
Using (4.160) this determines (p̊) to be

µ (p̊) = √1 (0, 1, i, 0) , (4.222)

2

with a normalisation ∗ ⋅  = −1. Furthermore from the explicit form for Λ(a1 ,a2 ) also in
(4.160) we then obtain

Λ(a1 ,a2 ) (p̊) = (p̊) + c p̊ , c= √1 (a1 + a2 ) . (4.223)

2

For general momentum p = ω(1, n), p2 = 0, as in (4.166), we may define, for L(p) given by
(4.136),
(p) = L(p) (p̊) = ΛR(n) (p̊) , (4.224)
since B(α, e3 )(p̊) = (p̊), and where the rotation R(n) is determined by n just as in (4.136).
With the definition (4.224)
pµ µ (p) = p̊µ µ (p̊) = 0 . (4.225)

For a general Lorentz transformation Λ then from (4.163) and (4.221),(4.222)

Λ(p) = ((Λp) + c Λp) e−iχ(p,Λ) , (4.226)

for some c depending on p, Λ. This matches (4.165), for h = 1, save for the inhomogeneous
term proportional to c (for h = −1 it is sufficient to take (p) → (p)∗ ). (4.226) shows that
(p) does not transform in a Lorentz covariant fashion. Homogeneous Lorentz transforma-
tions are obtained if, instead of considering just (p), we consider the equivalence classes
polarisation vectors {(p) ∶∼} with the equivalence relation

(p) ∼ (p) + c p , for arbitrary c. (4.227)

This is the same as saying that the polarisation vectors (p) are arbitrary up to the addition
of any multiple of the momentum vector p. It is important to note that, because of (4.225),
that scalar products of polarisation vectors depend only on their equivalence classes so that

′ (p)∗ ⋅ ′ (p) = (p)∗ ⋅ (p) for ′ (p) ∼ (p) . (4.228)

The gauge freedom in (4.227) is a reflection of gauge invariance which is a necessary feature
of field theories when massless particles are described by quantum fields transforming in a
Lorentz covariant fashion.
In general Lorentz covariant fields contain more degrees of freedom than those for the
associated particle which are labelled by the spin or helicity in the massless case. It is then
necessary to impose supplementary conditions to reduce the number of degrees of freedom,
e.g. for a massive 4-vector field φµ , associated with a spin one particle, requiring ∂µ φµ = 0.
For the massless case then there are gauge transformations belonging to a gauge group
which eliminate degrees of freedom so that just two helicities remain. Although this can be
achieved for free particles of arbitrary spin there are inconsistencies when interactions are
introduced for higher spins, beyond spin one in the massive case with spin two also allowed
for massless particles.

135
5 Lie Groups and Lie Algebras

Although many discussions of groups emphasise finite discrete groups the groups of most
widespread relevance in high energy physics are Lie groups which depend continuously on a
finite number of parameters. In many ways the theory of Lie40 groups is more accessible than
that for finite discrete groups, the classification of the former was completed by Cartan41
over 100 years ago while the latter was only finalised in the late 1970’s and early 1980’s.
A Lie Group is of course a group but also has the structure of a differentiable manifold, so
that some of the methods of differential geometry are relevant. It is important to recognise
that abstract group elements cannot be added, unlike matrices, so the notion of derivative
needs some care. For a Lie group G, with an associated n-dimensional differential manifold
MG , then for an arbitrary element

g(a) ∈ G , a = (a1 , . . . , an ) ∈ Rn coordinates on MG . (5.1)

n is the dimension of the Lie group G. For any interesting MG no choice of coordinates is
valid on the whole of MG , it is necessary to choose different coordinates for various subsets of
MG , which collectively cover the whole of MG and form a corresponding set of coordinate
charts, and then require that there are smooth transformations between coordinates on
the overlaps between coordinate charts. Such issues are generally mentioned here only in
passing.
For group multiplication we then require

g(a)g(b) = g(c) ⇒ cr = ϕr (a, b) , r = 1, . . . , n , (5.2)

where ϕr is continuously differentiable. It is generally convenient to choose the origin of

the coordinates to be the identity so that

g(0) = e ⇒ ϕr (0, a) = ϕr (a, 0) = ar , (5.3)

and then for the inverse

g(a)−1 = g(ā) ⇒ ϕr (ā, a) = ϕr (a, ā) = 0 . (5.4)

The crucial associativity condition is then

g(a)(g(b)g(c)) = (g(a)g(b))g(c) ⇒ ϕr (a, ϕ(b, c)) = ϕr (ϕ(a, b), c) . (5.5)

A Lie group may be identified with the associated differentiable manifold MG together with
a map ϕ ∶ MG × MG → MG , where ϕ satisfies (5.3), (5.4) and (5.5).
For an abelian group ϕ(a, b) = ϕ(b, a) and it is possible to choose coordinates such that

ϕr (a, b) = ar + br , (5.6)
40
Marius Sophus Lie, 1842-1899, Norwegian.
41
Élie Joseph Cartan, 1869-1951, French.

136
and in general if we Taylor expand ϕ we must have

ϕr (a, b) = ar + br + crst as bt + O(a2 b, ab2 ) , ār = −ar + crst as at + O(a3 ) . (5.7)

As will become apparent the coefficients crst , or rather f rst = cr[st] , which satisfy conditions
arising from the associativity condition (5.5), essentially determine the various possible Lie
groups.
As an illustration we return again to SU (2). For 2 × 2 matrices A we may express them
in terms of the Pauli matrices by

A = u0 I + i u ⋅ σ , A† = u0 I − i u ⋅ σ . (5.8)

Requiring u0 , u to be real then

A† A = (u02 + u2 ) I , det A = u02 + u2 . (5.9)

Hence
A ∈ SU (2) ⇒ u02 + u2 = 1 . (5.10)
The condition u02 + u2 = 1 defines the three dimensional sphere S 3 embedded in R4 , so that
MSU (2) ≃ S 3 . In terms of differential geometry all points on S 3 are equivalent but here
the pole u0 = 1, u = 0 is special as it corresponds to the identity. For SO(3) then, since
±A correspond to the same element of SO(3), we must identify (u0 , u) and −(u0 , u), i.e.
antipodal points at the ends of any diameter on S 3 . In
√ the hemisphere u0 ≥ 0 we may use
u, ∣u∣ ≤ 1 as coordinates for SU (2), since then u0 = 1 − u2 . Then group multiplication
defines ϕ(u, v) = u + v − u × v + . . . .
For A ∈ Sl(2, C) then if A† A = e2V , for V † = V , R = A e−V satisfies R† R = I. Since then
det R = eiα while det eV = etr(V ) is real, det A = 1 requires both det R = 1 and tr(V ) = 0.
Hence there is a unique decomposition A = R eV with V = Vi σi so that the group manifold
MSl(2,C) = S 3 × R3 .

5.0.1 Vector Fields, Differential Forms and Lie Brackets

For any differentiable n-dimensional manifold M, with coordinates xi , then scalar functions
f ∶ M → R are defined in terms of these coordinates by f (x) such that under a change of
coordinates xi → x′i we have f (x) = f ′ (x′ ). Vector fields are defined in terms of differential
operators acting on scalar functions
∂
X(x) = X i (x) , (5.11)
∂xi
where for the x → x′ change in coordinates we require

∂x′i
X j (x) j
= X ′i (x′ ) . (5.12)
∂x
For each x the vector fields belong to a linear vector space Tx (M) of dimension n, the
tangent space at the point specified by x.

137
 For two vector fields X, Y belonging to Tx (M) the Lie bracket, or commutator, defines
a further vector field
[X, Y ] = −[Y, X] , (5.13)
where
[X, Y ]i (x) = X(x)Y i (x) − Y (x)X i (x) , (5.14)
′
since, for a change x → x and using (5.12),

[X, Y ]′ = [X ′ , Y ′ ] , (5.15)
∂ 2 x′i ∂ 2 x′i
as a consequence of ∂xj ∂xk
= ∂xk ∂xj
. The Lie bracket is clearly linear, so that for any
X, Y, Z ∈ Tx (M)
[αX + βY, Z] = α[X, Y ] + β[Y, Z] , (5.16)
as in necessary for the Lie bracket to be defined on the vector space Tx (M), and it also
satisfies crucially the Jacobi42 identity, which requires

[X, [Y, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0 . (5.17)

This follows directly from the definition of the Lie Bracket as a commutator of differential
operators.
Dual to vector fields are one-forms, belonging to Tx (M)∗ ,

ω(x) = ωi (x) dxi , (5.18)

where ⟨dxi , ∂j ⟩ = δ ij . For x → x′ now

∂xj
ωj (x) ′i
= ω ′i (x′ ) . (5.19)
∂x
For p-forms

ρ(x) = 1
p! ρi1 ...ip (x) dxi1 ∧ ⋅ ⋅ ⋅ ∧ dxip , dxi ∧ dxj = −dxj ∧ dxi , (5.20)

so that ρi1 ...ip = ρ[i1 ...ip ] . The transformations ρ → ρ′ for a change of coordinates x → x′ are
the natural multi-linear extension of (5.19). For an n-dimensional space dx′i1 ∧ ⋅ ⋅ ⋅ ∧ dx′in =
′i
det [ ∂x
∂xj
] dxi1 ∧ ⋅ ⋅ ⋅ ∧ dxin and we may require

dxi1 ∧ ⋅ ⋅ ⋅ ∧ dxin = εi1 ...in dn x (5.21)

with εi1 ...in the n-dimensional antisymmetric symbol and dn x the corresponding volume
element. If ρ is a n-form and Mn a n-dimensional manifold this allows the definition of the
integral
∫ ρ. (5.22)
Mn

The exterior derivative d acts on p-forms to give (p + 1)-forms, dρ = dxi ∧ ∂i ρ. For the
one-form in (5.18) the corresponding two-form is then given by

(dω)ij (x) = ∂i ωj (x) − ∂j ωi (x) . (5.23)

42
Carl Gustav Jacob Jacobi, 1804-1851, German.

138
Of course (dω)′ = d′ ω ′ with d′ = dx′i ∂ ′i . In general d2 = 0. If ρ is a closed p-form then

dρ = 0 . (5.24)

A trivial solution of (5.24) is provided by

ρ = dω , (5.25)

for some (p − 1)-form ω. In this case ρ is exact. If the n-form ρ in (5.22) is exact and if also
if Mn is closed then the integral is zero.

5.1 Lie Algebras

The additional structure associated with a differential manifold MG corresponding to a Lie

group G ensures that the tangent spaces Tg (MG ), for a point on the manifold for which the
group element is g, can be related by group transformations. In particular the tangent space
at the origin Te (MG ) plays a special role and together with the associated Lie bracket [ , ]
defines the Lie algebra g for the Lie group. For all points on MG there is a space of vector
fields which are invariant in a precise fashion under the action of group transformations
and which belong to a Lie algebra isomorphic to g. There are also corresponding invariant
one-forms.
To demonstrate these results we consider how a group element close to the identity
generates a small change in an arbitrary group element g(b) when multiplied on the right,

g(b + db) = g(b)g(θ) , θ infinitesimal ⇒ br + dbr = ϕr (b, θ) , (5.26)

so that
∂ r
dbr = θa µa r (b) , µa r (b) =
ϕ (b, θ))∣ . (5.27)
∂θa θ=0
Here we use a, b, c as indices referring to components for vectors or one-forms belonging to
Te (MG ) or its dual (which must be distinguished from their use as coordinates) and r, s, t
for indices at an arbitrary point. To consider the group action on the tangent spaces we
analyse the infinitesimal variation of (5.2) for fixed g(a),

g(c + dc) = g(a) g(b + db) = g(c) g(θ) , (5.28)

so that, for fixed g(a),

dcr = θa µa r (c) = dbs λs a (b) µa r (c) , (5.29)
using (5.27) and defining λ(b) as the matrix inverse of µ(b),

[λs a (b)] = [µa s (b)]−1 , λs a (b) µa r (b) = δsr . (5.30)

Hence from from (5.29)

∂cr
= λs a (b) µa r (c) . (5.31)
∂bs
If near the identity we assume (5.7) then µa s (0) = δas .

139
 By virtue of (5.31)

∂ ∂cr ∂
Ta (b) = µa s (b) = µ a
s
(b) = Ta (c) , (5.32)
∂bs ∂bs ∂cr
define a basis {Ta ∶ a = 1, . . . , n} of left-invariant vector fields belonging to T (MG ), since
they are unchanged as linear differential operators under transformations corresponding to
g(b) → g(c) = g(a)g(b). Furthermore the corresponding vector space, formed by constant
linear combinations g = {θa Ta }, is closed under taking the Lie bracket for any two vectors
belonging to g and defines the Lie algebra.
To verify closure we consider the second derivative of cr (b) where from (5.31) and (5.32)

∂ 2 cr
µa s (b)µb t (b) = µa s (b)Tb (b)(λs a (b) µa r (c))
∂bs ∂bt
= µa s (b)(Tb (b)λs c (b) µc r (c) + λs c (b) Tb (c)µc r (c)) . (5.33)

For any matrix δX −1 = −X −1 δX X −1 so that from (5.30)

Tb (b)λs c (b) = −λs d (b)(Tb (b)µd u (b))λu c (b) , (5.34)

which allows (5.33) to be written as

∂ 2 cr
µa s (b)µb t (b) = −Tb (b)µa u (b) λu c (b)µc r (c) + Tb (c)µc r (c) , (5.35)
∂bs ∂bt
or, transporting all indices so as to refer to the identity tangent space,

∂ 2 cr
µa s (b)µb t (b) λr c (c) = −(Tb (b)µa r (b)) λr c (b) + (Tb (c)µa r (c))λr c (c) . (5.36)
∂bs ∂bt
Since
∂ 2 cr ∂ 2 cr
= , (5.37)
∂bs ∂bt ∂bt ∂bs
the right hand side of (5.36) must be symmetric in a, b. Imposing that the antisymmetric
part vanishes requires

(Ta (b)µb r (b) − Tb (b)µa r (b))λr c (b) = f cab , (5.38)

where f cab are the structure constants for the Lie algebra. They are constants since (5.36)
requires that (5.38) is invariant under b → c. Clearly

f cab = −f cba . (5.39)

From (5.30), (5.38) can be equally written just as first order differential equations in terms
of µ,
Ta µb r − Tb µa r = f cab µc r , (5.40)
or more simply it determines the Lie brackets of the vector fields in (5.32)

[Ta , Tb ] = f cab Tc , (5.41)

140
ensuring that the Lie algebra is closed.
The Jacobi identity (5.17) requires

[Ta , [Tb , Tc ]] + [Tc , [Ta , Tb ]] + [Tb , [Tc , Ta ]] = 0 , (5.42)

or in terms of the structure constants

f ead f dbc + f ecd f dab + f ebd f dca = 0 . (5.43)

(5.43) is a necessary integrability condition for (5.40) which in turn is necessary for the
integrability of (5.31).
The results (5.31), (5.40) with (5.42) and (5.39) are the contents of Lie’s fundamental
theorems for Lie groups.
Alternatively from (5.33) using

∂ ∂
t
µa r (c) = −µa u (c) t λu c (c) µc r (c) , (5.44)
∂c ∂c
we may obtain

∂ 2 cr ∂ ∂
µa s (b)µb t (b) s t
λr c (c) = µa s (b)µb t (b) t λs c (b) − µb t (c)µa u (c) t λu c (c) . (5.45)
∂b ∂b ∂b ∂c
In a similar fashion as before this leads to
∂ ∂
µa s (b)µb t (b) t
λs c (b) − µb s (b)µa t (b) t λs c (b) = f cab , (5.46)
∂b ∂b
which is equivalent to (5.38), or

∂ ∂
λs c (b) − s λr c (b) = −f cab λr a (b)λsb (b) . (5.47)
∂br ∂b
Defining the left invariant one-forms

ω a (b) = dbr λr a (b) , (5.48)

the result is expressible more succinctly, as consequence of (5.23), by

dω a = − 12 f bbc ω b ∧ ω c . (5.49)

Note that, using d(ω b ∧ ω c ) = dω b ∧ ω c − ω b ∧ dω c , d2 ω a = − 12 f ab[c f bde] ω c ∧ ω d ∧ ω e = 0 by

virtue of the Jacobi identity (5.43).
In general a n-dimensional manifold for which there are n vector fields which are linearly
independent and non zero at each point is parallelisable. Examples are the circle S 1 and
the 3-sphere S 3 . A Lie group defines a parallelisable manifold since a basis for non zero
vector fields is given by the left invariant fields in (5.32), the group U (1) corresponds to S 1
and SU (2) to S 3 .

141
5.2 Lie Algebra Definitions

In general a Lie algebra is a vector space g with a commutator [ , ] ∶ g × g → g satisfying

(5.13), (5.16) and (5.17), or in terms of a basis {Ta }, satisfying (5.41), with (5.39), and
(5.42) or (5.43). Various crucial definitions, which are often linked to associated definitions
for groups, are given below.
Two Lie algebras g, g′ are isomorphic, g ≃ g′ , if there is a mapping between elements of
the Lie algebras X ↔ X ′ such that [X, Y ]′ = [X ′ , Y ′ ]. If g = g′ the map is an automorphism
of the Lie algebra. For any g automorphisms form a group, the automorphism group of g.
The Lie algebra is abelian, corresponding to an abelian Lie group, if all commutators
are zero, [X, Y ] = 0 for all X, Y ∈ g.
A subalgebra h ⊂ g forms a Lie algebra itself and so is closed under commutation. If
H ⊂ G is a Lie group then its Lie algebra h is a subalgebra of g.
An invariant subalgebra or ideal h ⊂ g is such that

[X, Y ] ∈ h for all Y ∈ h, X ∈ g. (5.50)

If H is a normal Lie subgroup then its Lie algebra forms an ideal. Note that

i = [ g, g ] = {[X, Y ] ∶ X, Y ∈ g} , (5.51)

forms an ideal i ⊂ g, since [Z, [X, Y ]] ∈ i for all Z ∈ g. i is called the derived algebra.
The centre of a Lie algebra g, Z(g) = {Y ∶ [X, Y ] = 0 for all X ∈ g}.
A Lie algebra is simple if it does not contain any invariant subalgebra.
A Lie algebra is semi-simple if it does not contain any invariant abelian subalgebra.
Using the notation in (5.51) and we may define in a similar fashion a sequence of
successive invariant derived subalgebras g(n) , n = 1, 2, . . . , forming the derived series by

g(n+1) = [ g(n) , g(n) ] , g(1) = [ g, g ] . (5.52)

A Lie algebra g is solvable if g(n+1) = 0 for some n, and so g(n) is abelian and the derived
series terminates.
Solvable and semi-simple Lie algebras are clearly mutually exclusive. Lie algebras may
be neither solvable nor semi-simple but in general they may be decomposed in terms of such
Lie algebras.
The direct sum of two Lie algebras, g = g1 ⊕ g2 = {X1 + X2 ∶ X1 ∈ g1 , X2 ∈ g2 }, with the
commutator
[X1 + X2 , Y1 + Y2 ] = [X1 , X2 ] + [Y1 , Y2 ] . (5.53)
It is easy to see that the direct sum g contains g1 and g2 as invariant subalgebras so that g
is not simple. The Lie algebra for the direct product of two Lie groups G = G1 ⊗ G2 is the
direct sum g1 ⊕ g2 .

142
 If a Lie algebra g can be defined to act linearly on a Lie algebra h such that
′ ′ ′
Y →YX, (Y X )X − (Y X )X = Y [X ,X] for all Y ∈ h , X, X ′ ∈ g , (5.54)
X

then we may define the semi-direct sum Lie algebra g ⊕s h = {X + Y ∶ X ∈ h, Y ∈ h} with

′
commutators [X +Y, X ′ +Y ′ ] = [X, X ′ ]+Y ′X −Y X +[Y, Y ′ ]. h forms an invariant subalgebra
of g ⊕s h. The semi-direct sum of Lie algebras arises from the semi-direct product of Lie
groups.

5.3 Matrix Lie Algebras and Matrix Lie Groups

The definition of the Lie algebra is more straightforward for matrix Lie groups. For a matrix
group there are matrices D(a), depending on the parameters ar , realising the basic group
multiplication rule (5.2),
D(a)D(b) = D(c) . (5.55)
For group elements close to the identity with infinitesimal parameters θa we can now write

D(θ) = 1 + θa ta , (5.56)

which defins a set of matrices {ta } forming the generators for this matrix group. Writing
∂
D(b + db) = D(b) + dbr D(b) , (5.57)
∂br
then (5.26) becomes
∂
dbrD(b) = θa Ta D(b) = D(b)θa ta , (5.58)
∂br
using (5.27) along with (5.32). Clearly

Ta D(b) = D(b)ta , (5.59)

and it then follows from (5.41) that

[ta , tb ] = f cab tc . (5.60)

The matrix generators {ta } hence obey the same Lie algebra commutation relations as {Ta },
and may be used to directly define the Lie algebra instead of the more abstract treatment
in terms of vector fields.

5.3.1 SU (2) Example

As a particular illustration we revisit SU (2) and following (5.8) and (5.10) write
√
A(u) = u0 1 + i u ⋅ σ u0 = 1 − u2 . (5.61)

This parameterisation is valid for u0 ≥ 0. With, for infinitesimal θ, A(θ) = 1 + i θ ⋅ σ we get,

using the standard results (3.20) to simplify products of Pauli matrices,

A(u + du) = A(u)A(θ) = u0 − u ⋅ θ + i(u + u0 θ − u × θ) ⋅ σ , (5.62)

143
and hence

du = a0 θ − u × θ , or dui = θj µji (u) , µji (u) = u0 δji + uk εjki . (5.63)

The vector fields forming a basis for the Lie algebra su(2) are then

∂
Tj (u) = µji (u) ⇒ T = u0 ∇u + u × ∇u . (5.64)
∂ui
Since
TA(u) = A(u) iσ , (5.65)
and [σi , σj ] = 2iεijk σk , the Lie bracket must be

[Ti , Tj ] = −2 εijk Tk . (5.66)

5.3.2 Upper Triangular Matrices

The upper triangular and the strictly upper triangular matrices

⎧ x . x⎞⎫ ⎧ x . x⎞⎫
⎪
⎪
⎪
⎪⎛x x ⎪
⎪
⎪ ⎪
⎪
⎪⎛0 x ⎪
⎪
⎪
⎪
⎪ 0 x x . x⎟⎪⎪
⎪ ⎪
⎪
⎪ 0 0 x . x⎟⎪⎪
⎪
⎪⎜
⎪⎜ ⎟⎪
⎪ ⎪⎜
⎪⎜ ⎟⎪
⎪
b = ⎨⎜ 0 0 x . x⎟ ⎬, n = ⎨⎜0 0 0 . x⎟ ⎬, (5.67)
⎪
⎪⎜ ⎟⎪ ⎪⎜ ⎟⎪
⎪
⎪
⎪
⎜⋮ ⋮ ⋮ ⋱ ⋮ ⎟⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎜⋮ ⋮ ⋮ ⋱ ⋮ ⎟⎪
⎪
⎪
⎪
⎪⎝ ⎪ ⎪ ⎪
⎩ 0
⎪
⎪ 0 0 . x⎠⎪⎪
⎭
⎪
⎪⎝
⎩ 0 0 0 . 0 ⎠⎪
⎪
⎭
form Lie algebras with the commutator defined by usual matrix multiplication. It is easy
to see that
n = [b, b] , (5.68)
and that the Lie algebras b and hence also n are solvable.

5.3.3 Representations and Lie Algebras

There is an intimate relation between representations of Lie algebras and Lie Groups. Just
as described for groups in 2, a representation of a Lie algebra g is of course such that for
any X ∈ g there are corresponding matrices D(X) such that D([X, Y ]) = [D(X), D(Y )],
where [D(X), D(Y )] is the matrix commutator. For convenience we may take D(Ta ) = ta
where {ta } form a basis of matrices in the representation satisfying (5.60), following from
(5.41). As for groups an irreducible representation of the Lie algebra is when there are no
invariant subspaces of the corresponding representation space V under the action of all the
Lie algebra generators on V. Just as for groups there is always a trivial representation by
taking D(X) = 0.
The generators may be defined in terms of the representation matrices for group elements
which are close to the identity,

D(g(θ)) = 1 + θa ta + O(θ2 ) , D(g(θ))−1 = 1 − θa ta + O(θ2 ) . (5.69)

144
For unitary representations, as in (2.37), the matrix generators are then anti-hermitian,
ta† = −ta . (5.70)
If the representation matrices have unit determinant, since det(1+X) = 1+ tr(X)+O(2 ),
we must also have
tr(ta ) = 0 . (5.71)

In a physics context it is commonplace to redefine the matrix generators so that ta = −it̂a

so that, instead of (5.70), the generators t̂a are hermitian and satisfy the commutation
relations [t̂a , t̂b ] = if cab t̂c .
Two representations of a Lie algebra {t′a } and {ta } are equivalent if, for some non
singular S,
t′a = S ta S −1 . (5.72)
For both representations to be unitary then S must be unitary. If the representation is
irreducible then, by applying Schur’s lemma,
ta = S ta S −1 or [S, ta ] = 0 ⇒ S ∝I. (5.73)

The complex conjugate of a representation is also a representation, in general it is

inequivalent. If it is equivalent then, for some C,
ta∗ = C ta C −1 , (5.74)
or for a unitary representation, assuming (5.70),
C ta C −1 = −taT . (5.75)
Following the same argument as in 2.3.2, combining (5.75) with its transpose we get
C −1T C ta C −1 C T = ta so that for an irreducible representation
C −1 T C = c I ⇒ C = c CT ⇒ c = ±1 . (5.76)
If C = C T then, by a transformation C → S T CS together with ta → S −1 ta S, we can take
C = I and the representation is real. If C = −C T the representation is pseudo-real. For
det C ≠ 0 the representation must be even dimensional, 2n. By a transformation we may
take C = J, J 2 = −I, where J is defined in (1.108). The representation matrices then satisfy
D(g(θ))† = −JD(g(θ))J, which is just as in (1.79). This is sufficient to ensure that the
pseudo-real representation formed by {D(g(θ))} can be expressed in terms of n×n matrices
of quaternions, and so such representations are also referred to as quaternionic.
The SO(3) spinor representation described in section 3.14 is pseudo-real since
C σ C −1 = −σ T 0 1),
for C = iσ2 = ( −1 0 (5.77)
which is equivalent to (3.296).
A corollary of (5.75) is that,for real or pseudo-real representations,
tr(t(a1 . . . tan ) ) = 0 for n odd . (5.78)
For n = 3 this has important consequences in the discussion of anomalies in quantum field
theories.

145
5.4 Relation of Lie Algebras to Lie Groups

The Lie algebra of a Lie group is determined by those group elements close to the identity.
Nevertheless the Lie group can be reconstructed from the Lie algebra subject to various
topological caveats. Firstly the group must be connected, for elements g ∈ G there is a
continuous path g(s) with g(0) = e and g(1) = g. Thus we must exclude reflections so
that SO(3) and SO(3, 1)↑ are the connected groups corresponding to rotations and Lorentz
transformations. Secondly for a Lie group G having a centre Z(G) which is a discrete
abelian group, then for any subgroup HZ (G) ⊂ Z(G), where HZ (G) = {h} with gh = hg
for all g ∈ G, the group G/HZ (G), defined by g ∼ gh, is also a Lie group with the same Lie
algebra as G. As an example SO(3) and SU (2) have the the same Lie algebra although
SO(3) ≃ SU (2)/Z2 where Z2 = Z(SU (2)).

5.4.1 One-Parameter Subgroups

For any element θa Ta ∈ g there is a one-parameter subgroup of the associated Lie Group G
corresponding to a path in MG whose tangent at the identity is θa Ta . With coordinates ar
the path is defined by ars , with s ∈ R, where

d r d
a = θa µa r (as ) , ar0 = 0 , or g(as ) = θa Ta (as ) g(as ) . (5.79)
ds s ds
To verify that this forms a subgroup consider g(c) = g(at )g(as ) where from (5.2)

cr = ϕr (at , as ) . (5.80)

Using (5.79) with (5.31) we get

∂ r
c = θa µa u (as ) λu b (as ) µb r (c) = θb µb r (c) , cr ∣s=0 = art . (5.81)
∂s
The equation is then identical with (5.79), save for the initial condition at s = 0, and the
solution then becomes
cr = ars+t ⇒ g(at )g(as ) = g(as+t ) . (5.82)
Since
g(as )−1 = g(a−s ) , (5.83)
then {g(as )} forms an abelian subgroup of G depending on the parameter s. We may then
define an exponential map
exp ∶ g → G , (5.84)
by
g(as ) = exp(s θa Ta ) . (5.85)
For any representation we have
at
D(g(as )) = es θ a
, (5.86)
where ta are the matrix generators and the matrix exponential may be defined as an infinite
power series, satisfying of course etX esX = e(s+t)X for any matrix X.

146
5.4.2 Baker Cambell Hausdorff Formula

In order to complete the construction of the Lie group G from the Lie algebra g it is
necessary to show how the group multiplication rules for elements belonging to different
one-parameter groups may be determined, i.e for any X, Y ∈ g we require
exp(tX) exp(tY ) = exp(Z(t)) , Z(t) ∈ g . (5.87)
The Baker Cambell Hausdorff43 formula gives an infinite series for Z(t) in powers of t whose
first terms are of the form
Z(t) = t(X + Y ) + 12 t2 [X, Y ] + 1 3
12 t
([X, [X, Y ]] − [Y, [X, Y ]]) + O(t4 ) , (5.88)
where the higher order terms involve further nested commutators of X and Y and so are
determined by the Lie algebra g. For an abelian group we just have Z(t) = t(X + Y ). The
higher order terms do not have a unique form since they can be rearranged using the Jacobi
identity. Needless to say the general expression is virtually never a practical method of
calculating group products, for once existence is more interesting than the final explicit
formula.
We discuss here the corresponding matrix identity rather than consider the result for
an abstract Lie algebra. It is necessary in the derivation to show how matrix exponentials
can be differentiated so we first consider the matrix expression
f (s) = es(Z+δZ) e−sZ , (5.89)
and then
d
f (s) = es(Z+δZ) δZ e−sZ = esZ δZ e−sZ + O(δZ 2 ) . (5.90)
ds
Solving this equation
1
f (1) = 1 + ∫ ds esZ δZ e−sZ + O(δZ 2 ) , (5.91)
0
so that
1
eZ+δZ − eZ = ∫ ds esZ δZ e(1−s)Z + O(δZ 2 ) . (5.92)
0
Hence for any Z(t) we have the result for the derivative of its exponential
d Z(t) 1 d
e = ∫ ds esZ(t) Z(t) e(1−s)Z(t) . (5.93)
dt 0 dt

If, instead of (5.87), we suppose,

etX etY = eZ(t) , (5.94)
then
d tX tY −tY −tX
(e e ) e e = X + etX Y e−tX
dt
d Z(t) −Z(t) 1 d
= e e = ∫ ds esZ(t) Z(t) e−sZ(t) . (5.95)
dt 0 dt
43
Henry Frederick Baker, 1866-1956, British, senior wrangler 1887. John Edward Cambell, 1862-1924,
Irish. Felix Hausdorff, 1868-1942, German.

147
With the initial condition Z(0) = 0 this equation then allows Z(t) to be determined. To
proceed further, using the formula for the exponential expansion

eA B e−A = B + [A, B] + 12 [A, [A, B]] + . . . , (5.96)

(5.95) can be rewritten as an expansion in multiple commutators

∞
d 1 d
X + etX Y e−tX = Z(t) + ∑ [Z(t), . . . [Z(t), Z(t)] . . . ] , (5.97)
dt n=1 (n + 1)! ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ dt
n

which may be solved iteratively by writing Z(t) = ∑∞ n

n=1 Zn t .

The results may be made somewhat more explicit if we adopt the notation
∞ ∞
f (X ad )Y = ∑ fn [X, . . . [X , Y ] . . . ] for f (x) = ∑ fn xn , (5.98)
n=0 ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ n=0
n

so that (5.96) becomes eA B e−A = eA B. Then, since ∫0 ds esz = (ez − 1)/z, (5.95) can be
ad 1

written as
d ad ad
Z(t) = f (eZ(t) )(X + etX Y ) , (5.99)
dt
for, using the standard series expansion of ln(1 + x),
∞
ln x (−1)n
f (x) = =∑ (x − 1)n . (5.100)
x − 1 n=0 n + 1

Since
eZ(t) U = eZ(t) U e−Z(t) = etX etY U e−tY e−tX = etX etY U ,
ad ad ad
(5.101)
Z(t)ad tX ad ad
we may replace e →e etY on the right hand side of (5.99). With some intricate
combinatorics (5.99) may then be expanded as a power series in t which on integration gives
a series expansion for Z(t) (a formula can be found on Wikipedia).
A simple corollary of these results is

e−tX e−tY etX etY = et

2 [X,Y ]+O(t3 )
, (5.102)

so this combination of group elements isolates the commutator [X, Y ] as t → 0.

5.5 Simply Connected Lie Groups and Covering Groups

For a connected topological manifold M then for any two points x1 , x2 ∈ M there are
continuous paths px1 →x2 linking x1 and x2 defined by functions px1 →x2 (s), 0 ≤ s ≤ 1, where
px1 →x2 (0) = x1 , px1 →x2 (1) = x2 . For three points x1 , x2 , x3 a composition rule for paths
linking x1 , x2 and x2 , x3 is given by
⎧
⎪px →x (2s) ,
⎪ 0 ≤ s ≤ 21 ,
(px1 →x2 ○ px2 →x3 )(s) = ⎨ 1 2 1
(5.103)
⎩px2 →x3 (2s − 1) , 2 ≤ s ≤ 1.
⎪
⎪

148
For any px1 →x2 the corresponding inverse, and also the trivial identity path, are defined by

p−1
x2 →x1 (s) = px1 →x2 (1 − s) , pid
x→x (s) = x . (5.104)

The set of paths give topological information about M by restricting to equivalence, or

homotopy, classes [px1 →x2 ] = {p′x1 →x2 ∶ p′x1 →x2 ∼ px1 →x2 }, where the homotopy equivalence
relation requires that p′x1 →x2 (s) can be continuously transformed to px1 →x2 (s). These ho-
motopy classes inherit the composition rule [px1 →x2 ] ○ [px2 →x3 ] = [px1 →x2 ○ px2 →x3 ]. The
fundamental group for M is defined in terms of homotopy classes of closed paths starting
and ending at an arbitrary point x ∈ M,

π1 (M) = {[px→x ]} . (5.105)

This defines a group using the composition rule for group multiplication and for the identity
−1
e = [pid
x→x ] and for the inverse [px→x ] = [p−1
x→x ]. For M connected π1 (M) is independent of
the point x chosen in (5.105). M is simply connected if π1 (M) is trivial, so that px→x ∼ pid
x→x
for all closed paths. If π1 (M) is non trivial then M is multiply connected, if dim π1 (M) = n
there are n homotopy classes [px1 →x2 ] for any x1 , x2 .
For Lie groups we can then define π1 (G) ≡ π1 (MG ). In many examples this is non trivial.
For the rotation group SO(3), as described earlier, MSO(3) ≃ S 3 /Z2 where antipodal points,
at the end of diameters, are identified. Alternatively, by virtue of (3.8), MSO(3) may be
identified with a ball of radius π in three dimensions with again antipodal points on the
boundary S 2 identified. There are then closed paths, starting and finishing at the same
point, which involve a jump between two antipodal points on S 3 , or the surface of the ball,
and which therefore cannot be contracted to the trivial constant path. For two antipodal
jumps then by smoothly moving the corresponding diameters to coincide the closed path
can be contracted to the trivial path. Hence

π1 (SO(3)) ≃ Z2 . (5.106)

As another example we may consider the group U (1), as in (1.106), where it is clear that
MU (1) ≃ S 1 , the unit circle. For S 1 there are paths which wind round the circle n-times
which are homotopically distinct for different n so that homotopy classes belonging to
π1 (U (1)) are labelled by integers n. Under composition it is straightforward to see that the
winding number is additive so that

π1 (U (1)) ≃ Z , (5.107)

which is an infinite discrete group in this case.

5.5.1 Covering Group

For a non simply connected Lie group G there is an associated simply connected Lie group
G, the covering group, with the same Lie algebra since G and G are identical near the
identity. Assuming π1 (G) has n elements then for any g ∈ G we associate paths pi,e→g
where
pi,e→g (s) = gi (s) , gi (0) = e , gi (1) = g , i = 0, . . . , n − 1 , (5.108)

149
corresponding to the n homotopically distinct paths from the identity e to any g. The
elements of π1 (G) can be identified with [pi,e→e ]. We then define G such that the group
elements are
g i = (g, [pi,e→g ]) ∈ G for all g ∈ G , i = 0, . . . , n − 1 , (5.109)
with a corresponding group product

g1i g2j = gk , for g = g1 g2 , [pk,e→g1 g2 ] = [pi,e→g1 ○ g1 pj,e→g2 ] , (5.110)

using the path composition as in (5.103) and noting that g1 pj,e→g2 defines a path from g1
to g = g1 g2 . For the inverse and identity elements we have, with the definitions in (5.104),

g i −1 = (g −1 , [g −1 p−1
i,g→e ]) , e0 = (e, [p0,e→e ]) , p0,e→e = pid
e→e . (5.111)

These definitions satisfy the group properties although associativity requires some care. G
contains the normal subgroup given by

{ei ∶ i = 0, . . . , n − 1} ≃ π1 (G) , ei = (e, [pi,e→e ]) . (5.112)

Any discrete normal subgroup H of a connected Lie group G must be moreover a

subgroup of the centre Z(G), since if h ∈ H then ghg −1 ∈ H for any g ∈ G, by the definition
of a normal subgroup. Since we may g vary continuously over all G, if G is a connected Lie
group, and since H is discrete we must then have ghg −1 = h for all g, which is sufficient to
ensure that h ∈ Z(G).
The construction described above then ensures that the covering group G is simply
connected and we have therefore demonstrated that

G ≃ G/π1 (G) , π1 (G) ⊂ Z(G) . (5.113)

As an application we consider the examples of SO(3) and U (1). For SO(3) we consider
rotation matrices R(θ, n) as in (3.7) but allow the rotation angle range to be extended to
0 → 2π. Hence, instead of (3.8), we have

n ∈ S2 , 0 ≤ θ ≤ 2π , (θ, n) ≃ (2π − θ, −n) . (5.114)

There are two homotopically inequivalent paths linking the identity to R(θ, n), 0 ≤ θ ≤ π,
which may be defined, with the conventions in (5.114), by

p0,I→R(θ,n) (s) = R(sθ, n) , p1,I→R(θ,n) (s) = R(s(2π − θ), −n) , 0 ≤ s ≤ 1, (5.115)

since p1,I→R(θ,n) involves a jump between antipodal points. The construction of the covering
group then defines group elements R(θ, n)i , for i = 0, 1. For rotations about the same axis
the group product rule then requires

⎪ R(θ + θ′ , n) i+j mod 2 , 0 ≤ θ + θ′ ≤ π ,

⎧
⎪
R(θ, n)i R(θ′ , n)j = ⎨ ′
0 ≤ θ, θ′ ≤ π . (5.116)
⎩ R(θ + θ , n) i+j+1 mod 2 ,
⎪
⎪ π ≤ θ + θ′ ≤ 2π ,

150
It is straightforward to see that this is isomorphic to SU (2), by taking R(θ, n)0 → A(θ, n),
R(θ, n)1 → −A(θ, n), and hence SO(3) ≃ SU (2). For U (1) with group elements as in (1.106)
we may define
pn,1→eiθ (s) = eis(θ+2nπ) , 0 ≤ s ≤ 1, n ∈ Z, (5.117)

which are paths with winding number n. Writing the elements of the covering group U (1)
as gn (eiθ ) we have the product rule
′
⎪ gn+n′ (ei(θ+θ ) ) , 0 ≤ θ + θ′ ≤ 2π ,
⎧
⎪
iθ′
iθ
gn (e ) gn′ (e ) = ⎨ i(θ+θ′ )
0 ≤ θ, θ′ ≤ 2π . (5.118)
⎩gn+n +1 (e
⎪
⎪ ′ ), 2π ≤ θ + θ′ ≤ 4π ,

It is straightforward to see that effectively the group action is extended to all real θ, θ′ so
that U (1) ≃ R.

5.5.2 Projective Representations

For a non simply connected Lie group G then in general representations of the covering group
G generate projective representations of G. Suppose {D(gi )} are representation matrices
for G, where D(g1i ) D(g2j ) = D(gk ) for g1i , g2j , gk ∈ G satisfying the group multiplication
rule in (5.110). To restrict the representation to G it is necessary to restrict to a particular
path, say i, since there is then a one to one correspondence gi → g ∈ G. Then, assuming
g1i g2i = gj for some j,

D(g1i ) D(g2i ) = D(gj ) = D(gj gi−1 ) D(gi ) = D(ek ) D(gi ) , (5.119)

where, by virtue of (5.112) and (5.113),

gj gi−1 = ek ∈ Z(G) for some k . (5.120)

Since ek belongs to the centre, D(ek ) must commute with D(gi ) for any gi ∈ G and so,
for an irreducible representation must, by Schur’s lemma, be proportional to the identity.
Hence, for a unitary representation,

D(ek ) = eiγk 1 , (5.121)

where {eiγk ∶ k = 0, . . . , n − 1} form a one dimensional representation of π1 (G). Combining

(5.119) and (5.121) illustrates that {D(gi )}, for i fixed, provide a projective representation
of G as in (2.151).
For SO(3) we have just eiγk = ±1. For U (1) then there are one-dimensional projective
representations given by eiαθ , for any real α, where we restrict 0 ≤ θ < 2π which corresponds
to a particular choice of path in the covering group. Then the multiplication rules become
′
⎪ eiα(θ+θ ) , 0 ≤ θ + θ′ ≤ 2π ,
⎧
⎪
iαθ iαθ′
e e = ⎨ 2πiα iα(θ+θ′ −2π) (5.122)
⎩e
⎪
⎪ e , 2π ≤ θ + θ′ ≤ 4π .

151
5.6 Lie Algebra and Projective Representations

The possibility of different Lie groups for the same Lie algebra, as has been just be shown,
can lead to projective representations with discrete phase factors. There are also cases when
the phase factors vary continuously which can be discussed directly using the Lie algebra.
We wish to analyse then possible solutions of the consistency conditions (2.152) modulo
trivial solutions of the form (2.153) and show how this may lead to a modified Lie algebra.
For simplicity we write the phase factors γ which may appear in a projective represen-
tation of a Lie group G, as in (2.151), directly as functions on MG × MG so that, in terms
of the group parameters in (5.1), we take γ(g(a), g(b)) ≡ γ(a, b). The consistency condition
(2.152) is then analysed with gi → g(a), gj → g(b), gk → g(θ) with θ infinitesimal and, with
the same notation as in (5.26) and (5.28), this becomes

γ(c, θ) + γ(a, b) = γ(a, b + db) + γ(b, θ) . (5.123)

Defining
∂
γa (b) = γ(b, θ)∣ , (5.124)
∂θa θ=0
and with (5.27) and the definition (5.32) then (5.123) becomes

Ta (b) γ(a, b) = γa (c) − γa (b) . (5.125)

This differential equation for γ(a, b) has integrability conditions obtained by considering

[Ta (b), Tb (b)] γ(a, b) = f cab Tc (b) γ(a, b) (5.126)

which applied to (5.125) and using Ta (b) = Ta (c) from (5.32) leads to a separation of the
dependence on b and c so each part must be constant. This gives

Ta (b) γb (b) − Tb (b) γa (b) − f cab γc (b) = hab = −hba , (5.127)

with hab a constant. Applying Tc (b) and antisymmetrising the indices a, b, c gives, with
(5.41),

0 = Tc hab + Tb hca + Ta hbc = f dab (Td γc − Tc γd ) + f dbc (Td γa − Ta γd ) + f dca (Td γb − Tb γd ) , (5.128)

and hence, with (5.127) and (5.43), there is then a constraint on hab ,

f dab hdc + f dbc hda + f dca hdb = 0 . (5.129)

As was discussed in 2.9 there are trivial solutions of the consistency conditions which are
given by (2.153), and which, in the context of the Lie group considered here, are equivalent
to taking γ(a, b) = α(c) − α(a) − α(b) for α any function on MG . From (5.26) we then have
γ(b, θ) = α(b + db) − α(b) − α(θ) so that (5.124) gives

∂
γa (b) = Ta (b) α(b) − ca , ca = α(θ)∣ , (5.130)
∂θa θ=0

152
and then substituting in (5.127)
hab = f cab cc . (5.131)
It is easy to verify that (5.130) and (5.131) satisfy (5.127) and (5.129)44 .
If there are unitary operators U (a), corresponding to g(a) ∈ G, realising the Lie group
G as a symmetry group in quantum mechanics then (2.151) requires

U (b)U (θ) = eiγa (b)θ U (b + db) ,

a
(5.132)

for infinitesimal θa . Assuming

U (θ) = 1 − iθa T̂a , (5.133)
for hermitian operators T̂a , then, since U (b + db) = U (b) + θa Ta (b)U (b), we have

Ta (b)U (b) = −iU (b)(T̂a + γa (b)) . (5.134)

By considering [Ta , Tb ]U (b) and using (5.127) then this requires that the hermitian opera-
tors {T̂a } satisfy a modified Lie algebra

[T̂a , T̂b ] = if cab T̂c − i hab 1 . (5.135)

The additional term involving hab is a central extension of the Lie algebra, it is the coefficient
of the identity operator which commutes with all elements in the Lie algebra. A central
extension, if present, is allowed by virtue of the freedom up to complex phases in quantum
mechanics and they often play a crucial role. The consistency condition (5.129) is necessary
for {T̂a } to satisfy the Jacobi identity, if (5.131) holds then the central extension may be
removed by the redefinition T̂a → T̂a + ca 1.
As shown subsequently non trivial central extensions are not present for semi-simple Lie
algebras, it necessary for there to be an abelian subalgebra. A simple example arises for
the Lie algebra iso(2), given in (4.150), which has a central extension

[J3 , E1 ] = iE2 , [J3 , E2 ] = −iE1 , [E1 , E2 ] = ic 1 . (5.136)

5.6.1 Galilean Group

As an illustration of the significance of central extensions we consider the Galilean Group.

Acting on space-time coordinated x, t this is defined by the transformations involving rota-
tions, translations and velocity boosts

x′ = Rx + a + vt , t′ = t + b , (5.137)

where R is a rotation belonging to SO(3). If we consider a limit of the Poincaré Lie algebra,
with generators J, K, P, H, by letting K → cK, H → cM + c−1 H and take the limit c → ∞
44
Alternatively, using the left invariant one forms in (5.48) and defining h = 12 hab ω a ∧ ω b , then (5.129)
is equivalent, by virtue of (5.49), to dh = 0, so that h is closed, while the trivial solution (5.131) may be
identified with h = −dc, corresponding to h being exact, for c = ca ω a . Thus projective representations depend
on the cohomology classes of closed, modulo exact, two forms on MG .

153
then the commutation relations from (4.42), (4.43) and (4.107), (4.108) become

[Ji , Jj ] = iεijk Jk , [Ji , Kj ] = iεijk Kk , [Ki , Kj ] = 0 , [Ki , H] = iPi ,

[Ki , Pj ] = iδij M , [J, M ] = [K, M ] = [P, M ] = [H, M ] = 0 . (5.138)

When the Lie algebra is calculated just from the transformations in (5.137) the terms
involving M are absent, the terms involving M are a central extension.
If we consider the just the subgroup formed by boosts and spatial translations then
writing the associated unitary operators as

U [v, a] = e−ia⋅P eiv⋅K , (5.139)

then a straightforward calculation shows that

′
U [v′ , a′ ] U [v, a] = eiM v ⋅a U [v′ + v, a′ + a] . (5.140)

For comparison with the preceding general discussion we should take Ta → (∇v , ∇a ) and
T̂a → (−K, P). From (5.140) then γa → M (0, v) and from (5.127) hab → M ( −I03 I03 ).
For representations of the Galilean group in quantum mechanics the central extension
plays an essential role. Using (5.138)

e−iv⋅K Peiv⋅K = P + M v , eiv⋅K Heiv⋅K = H + P ⋅ v + 21 M v2 . (5.141)

In a similar fashion to the Poincaré group we may define irreducible representations in terms
of a basis for a space VM obtained from a vector ∣0⟩, such that P∣0⟩ = 0, by

∣p⟩ = eiv⋅K ∣0⟩ , p = Mv , (5.142)

so that as a consequence of (5.142)

p2
P∣p⟩ = p ∣p⟩ , H∣p⟩ = (E0 + 2M
)∣p⟩ . (5.143)

Clearly VM corresponds to states of a nonrelativistic particle of mass M . The representa-

tion can easily be extended to include spin by requiring that ∣0⟩ belong to an irreducible
representation of the rotation group.

5.7 Integration over a Lie Group, Compactness

For a discrete finite group G = {gi } then an essential consequence of the group axioms is that,
for any function f on G, the sum ∑i f (gi ) = ∑i f (ggi ) is invariant for any arbitrary g ∈ G.
This result played a vital role in the proof of results about representations such as Schur’s
lemmas and the equivalence of any representation to a unitary representation. Here we
describe how this may be extended to Lie groups where, since the group elements depend
on continuously varying parameters, the discrete sum is replaced by a correspondingly
invariant integration.

154
 If we consider first the simplest case of U (1), with elements as in (1.106) depending on
an angle θ then a general function f on U (1) is just a periodic function of θ, f (θ+2π) = f (θ).
′ ′
Since the product rule for this abelian group is eiθ eiθ = ei(θ +θ) then, for periodic f ,
2π 2π
∫ dθ f (θ) = ∫ dθ f (θ′ + θ) . (5.144)
0 0

provides the required invariant integration over U (1). For the covering group R, formed by
real numbers under addition, the integration has to be extended to the whole real line.
For a general Lie group G then, with notation as in (5.1) and (5.2), we require an
integration measure over the associated n-dimensional manifold MG such that

∫ dρ(b) f (g(b)) = ∫ dρ(b) f (g(c)) for g(c) = g(a)g(b) , (5.145)

G G

where dρ(b) = dn b ρ(b). To determine ρ(b) it suffices just to calculate the Jacobian J
for the change of variables b → c(b), with fixed a, giving for the associated change of the
n-dimensional integration volume elements

∂cr
dn c = ∣J∣ dn b , J = det [ ], (5.146)
∂bs

and then require, to satisfy (5.145),

dρ(b) = dρ(c) ⇒ ρ(b) = ∣J∣ ρ(c) . (5.147)

For a Lie group the fundamental result (5.31), with (5.30), ensures that

det [µ(c)]
J = det [λ(b)] det [µ(c)] = . (5.148)
det [µ(b)]

Comparing (5.146) and (5.148) with (5.147) show that the invariant integration measure
over a general Lie group G is obtained by taking
C
dρ(b) = dn b . (5.149)
∣ det [µ(b)]∣

for some convenient constant C. The normalisation of the measure is dictated by the form
near the identity since for b ≈ 0 then dρ(b) ≈ C dn b.
A Lie group G is compact if the group volume is finite,

∫ dρ(b) = ∣G∣ < ∞ , (5.150)

G

otherwise it is non compact. By rescaling ρ(b) we may take ∣G∣ = 1. For a compact Lie group
many of the essential results for finite groups remain valid, in particular all representations
are equivalent to unitary representations, and correspondingly the matrices representing the
Lie algebra can be chosen as anti-hermitian or hermitian, according to convention. Amongst
matrix groups SU (n), SO(n) are compact while SU (n, m), SO(n, m), for n, m > 0, are non
compact.

155
5.7.1 SU (2) Example

For SU (2) with the parameterisation in (5.61) the corresponding 3 × 3 matrix [µji (u)] was
computed in (5.63). It is not difficult to see that the eigenvalues are u0 , u0 ± i∣u∣ so that in
this case, since u02 + u2 = 1,
det[µji (u)] = u0 . (5.151)
Hence (5.149) requires
1 3
dρ(u) = d u, −1 ≤ u0 ≤ 1 , ∣u∣ ≤ 1 . (5.152)
∣u0 ∣

where range of u0 , u is determined in order to cover SU (2) matrices in (5.8). For the
parameterisation in terms of θ, n, n2 = 1, as given by (3.38)

u0 = cos 21 θ , u = − sin 12 θ n , d3 u = ∣u∣2 d∣u∣ dΩn , (5.153)

so that
dρ(θ, n) = 1
2 sin2 12 θ dθ dΩn , 0 ≤ θ ≤ 2π . (5.154)
Since ∫S 2 dΩn = 4π the group volume is easily found

∫ dρ(θ, n) = 2π 2 . (5.155)
SU (2)

These results verify the integration measure in (3.18) for SO3), where the range of θ is
halved.
For the parameterisation of SU (2) in terms of Euler angles φ, θ, ψ as in (3.96) the

u0 = cos 12 θ cos 12 (φ + ψ) , u3 = − cos 21 θ sin 21 (φ + ψ) ,

u1 = sin 12 θ sin 21 (φ − ψ) , u2 = sin 21 θ cos 12 (φ − ψ) . (5.156)

Using du1 ∧ du2 = − 18 sin θ dθ ∧ d(φ − ψ) and du1 ∧ du2 ∧ du3 = 1

8 sin θ u0 dθ ∧ dφ ∧ dψ then
1
dρ(φ, θ, ψ) = 8 sin θ dθ dφ dψ , (5.157)

reproducing (3.123).
For SO(3), since SU (2) is a double cover, the group volume is halved. In terms of the
parameterisation (θ, n) used in (5.154) we should take 0 ≤ θ ≤ π or in terms of the Euler
angles modify (3.96) to 0 ≤ ψ ≤ 2π.
For compact Lie groups the orthogonality relations for representations (2.50) or charac-
ters (2.56) remain valid if the summation is replaced by invariant integration over the group
and ∣G∣ by the group volume as in (5.150). For SU (2) this corresponds to the results give
in (3.133) and (3.134).

156
5.7.2 Non Compact Sl(2, R) Example

As an illustration of a non compact Lie group, we consider Sl(2, R) consisting of real 2 × 2

matrices with determinant 1. With the Pauli matrices in (3.19) a general real 2 × 2 matrix
may be expressed as in (3.40)

A = v0 + v1 σ1 + v2 iσ2 + v3 σ3 , (5.158)

where, for A ∈ Sl(2, R), v0 , v are real and we must further impose

det A = v02 + v22 − v12 − v32 = 1 . (5.159)

If we choose v = (v1 , v2 , v3 ) as independent parameters, so that we may write A(v), then

for a infinitesimal θ = (θ1 , θ2 , θ3 ) under matrix multiplication

A(v) A(θ) = A(v + dv) , (5.160)

where, using the multiplication rules (3.20),

⎛ v0 v3 v2 ⎞
(dv1 dv2 dv3 ) = (θ1 θ2 θ3 ) ⎜ v3 v0 −v1 ⎟ . (5.161)
⎝−v2 −v1 v0 ⎠

This defines the matrix µ(v), as in (5.27), for Sl(2, R) with the parameter choice in (5.158).
It is easy to calculate, with (5.159),

det µ(v) = v0 , (5.162)

so that the invariant integration measure becomes

1 3
dρ(v) = d v. (5.163)
∣v0 ∣

Unlike the case for SU (2) the parameters v have an infinite range so that the group volume
diverges.
For an alternative parameterisation we may take

v0 = cosh α cos β , v2 = cosh α sin β , v1 = sinh α cos γ , v3 = sinh α sin γ ,

α ≥ 0, 0 ≤ β, γ ≤ 2π . (5.164)

In this case the Sl(2, R) integration measure becomes

1
dρ(α, β, γ) = 2 sinh 2α dα dβ dγ , (5.165)

which clearly demonstrates the diverging form of the α integration. For β, γ = 0 the Sl(2, R)
matrix given by (5.164) reduces to one for SO(1, 1) as in (1.123).

157
5.8 Adjoint Representation and its Corollaries

A Lie algebra g is just a vector space with also a bilinear commutator, [ , ] ∶ g×g → g, subject
only to the requirement that the commutator is antisymmetric and satisfies the Jacobi
identity. The vector space defines the representation space for the adjoint representation
which plays an absolutely fundamental role in the analysis of Lie algebras.
For any X, Y ∈ g then
Y → [X, Y ] = X ad Y , (5.166)
X

defines the linear map X ad ∶ g → g. There is also a corresponding adjoint representation for
the associated Lie group G. For any X ∈ g the associated one parameter group is given by
exp(sX) ∈ G and then the adjoint representation Dad is defined by
∞
ad sn
Y ÐÐÐÐ→ Dad ( exp(sX))Y = esX Y = ∑ [X, . . . [X , Y ] . . . ] , (5.167)
exp(X) n=0 n! ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n

with similar notation to (5.98). To verify that (5.166) provides a representation of the Lie
algebra the Jacobi identity is essential since from

Z ad X ad Y = [Z, [X, Y ]] , (5.168)

we obtain for the adjoint commutator, using (5.17),

[Z ad , X ad ] Y = [Z, [X, Y ]] − [X, [Z, Y ]] = [[Z, X], Y ] = [Z, Y ]ad Y , (5.169)

and hence in general

[Z ad , X ad ] = [Z, Y ]ad . (5.170)

Explicit adjoint representation matrices are obtained by choosing a basis for g, {Ta } so
that for any X ∈ g (5.166) becomes

[X, Ta ] = Tb (X ad )b a . (5.171)

For X → Ta the corresponding adjoint representation matrices are then given by

[Ta , Tb ] = Tc (Taad )c b ⇒ (Taad )c b = f cab , (5.172)

using (5.41). The commutator

[Taad , Tbad ] = f cab Tcad , (5.173)

is directly equivalent to the Jacobi identity (5.42). The group representation matrices
ad
Dad (exp X) = eX , with X ad = Taad X a , are then obtained using the matrix exponential.
Close to the identity, in accord with (5.69),

Dad (exp X) = I + X ad + O(X 2 ) . (5.174)

If the Lie algebra is abelian then clearly X ad = 0 for all X so the adjoint representation
is trivial.

158
 For su(2) with the standard hermitian generators

[Ji , Jj ] = iεijk Jk ⇒ (Jiad )jk = −iεijk , (5.175)

where Jad are three 3 × 3 hermitian matrices. If n is a unit vector (n ⋅ Jad )2 = I − n nT

from which we may deduce that n ⋅ Jad has eigenvalues ±1, 0 so that this is the spin 1
representation. For the the Lie algebra iso(2), as given in (4.150), we have

⎛0 0 0 ⎞ ⎛0 0 1⎞ ⎛0 −1 0⎞
E1ad = i ⎜0 0 −1⎟ , E2ad = i ⎜0 0 0⎟ , J3ad = i ⎜1 0 0⎟ . (5.176)
⎝0 0 0 ⎠ ⎝0 0 0⎠ ⎝0 0 0⎠

5.8.1 Killing Form

The Killing45 form, although apparently due to Cartan, provides a natural symmetric bi-
linear form, analogous to a metric, for the Lie algebra g. It is defined using the trace, over
the vector space g, of the adjoint representation matrices by

κ(X, Y ) = tr(X ad Y ad ) for all X, Y ∈ g , (5.177)

or in terms of a basis as in (5.172)

κab = κ(Ta , Tb ) = f cad f dbc , (5.178)

so that κ(X, Y ) = κab X a Y b . Clearly it is symmetric κab = κba .

The importance of the Killing form arises from the crucial invariance condition

κ([Z, X], Y ) + κ(X, [Z, Y ]) = 0 . (5.179)

The verification of this is simple since, from (5.170),

κ([Z, X], Y ) = tr([Z, X]ad Y ad ) = tr([Z ad , X ad ] Y ad ) , (5.180)

and then (5.179) follows from tr([Z ad , X ad ] Y ad ) + tr(X ad [Z ad , Y ad ]) = 0, using cyclic sym-
metry of the matrix trace. The result (5.179) also shows that the Killing form is invariant
under the action of the corresponding Lie group G since
ad ad
κ(esZ X, esZ Y ) = κ(X, Y ) , (5.181)

which follows from (5.167) and differentiating with respect to s and then using (5.179).
Alternatively (5.179) may be expressed in terms of components using

κ([Tc , Ta ], Tb ) = f dca κ(Td , Tb ) = f dca κdb ≡ fcab , (5.182)

in a form expressing κab as an invariant tensor for the adjoint representation

κdb f dca + κad f dcb = 0 ⇔ fcab + fcba = 0 . (5.183)

45
Wilhelm Karl Joseph Killing, 1847-1923, German.

159
Since, from (5.39), fcab + fcba = 0 this implies

fabc = f[abc] . (5.184)

If the Lie algebra g contains an invariant subalgebra h then in an appropriate basis we

may write
Ta = (Ti , Tr ) , Ti ∈ h [Ti , Tj ] = f kij Tk , [Tr , Ti ] = f jri Tj , (5.185)
so that the Killing form restricted to h is just

κij = f kil f ljk = trh (Tiad Tjad ) . (5.186)

The crucial property of the Killing form is the invariance condition (5.179). If gab also
defines an invariant bilinear form on the Lie algebra, as in (5.179), so that

gab ([Z, X]a Y b + X a [Z, Y ]b ) = 0 , (5.187)

then, for any solution λi of det[κab − λgab ] = 0, hi = {Xi ∶ (κab − λi gab )Xib = 0} forms, by
virtue of the invariance condition (5.187), an invariant subalgebra hi ⊂ g. Restricted to hi
the Killing form κab and gab are proportional. For a simple Lie algebra, when there are no
invariant subalgebras, the Killing form is essentially unique.
For a compact group the adjoint representation Dad may be chosen to be unitary so
that in (5.174) the adjoint Lie algebra generators are anti-hermitian, as in (5.70),

X ad† = −X ad . (5.188)

In this case
κ(X, X) ≤ 0 , κ(X, X) = 0 ⇔ X ad = 0 . (5.189)

For su(2) using the hermitian adjoint generators in (5.175) the Killing form is positive

κij = tr(Jiad Jjad ) = i2 εikl εjlk = 2 δij . (5.190)

However for iso(2) then, if Ta = (E1 , E2 , J3 ), a = 1, 2, 3, it is easy to see from (5.176)

⎛0 0 0⎞
[κab ] = 2 ⎜0 0 0⎟ . (5.191)
⎝0 0 1⎠

5.8.2 Conditions for Non Degenerate Killing Form

For the Killing form to play the role of a metric on the Lie algebra then it should be non-
degenerate, which requires that if κ(Y, X) = 0 for all Y ∈ g then X = 0 or more simply
det[κab ] ≠ 0 so that κab Y b = 0 has no non trivial solution. An essential theorem due to
Cartan gives the necessary and sufficient conditions for this to be true. Using the definition
of a semi-simple Lie algebra given in 5.2 we have;

160
Theorem The Killing form is non-degenerate if and only if the Lie algebra is semi-simple.
To demonstrate that if the Lie algebra is not semi-simple the Killing form is degenerate
is straightforward. Assume there is an invariant abelian subalgebra h with a basis {Ti } so
that
Ta = (Ti , Tr ) ⇒ [Ti , Tj ] = 0 , [Tr , Ti ] = f jri Tj . (5.192)
Then from (5.178)

κai = f cad f dic = f raj f j ir = 0 , since f rsj = f rkj = 0 , (5.193)

which is equivalent to κ(Y, X) = 0 for X ∈ h and all Y ∈ g. The converse is less trivial.
For a Lie algebra g, if det[κab ] = 0 then h = {X ∶ κ(Y, X) = 0, for all Y ∈ g} forms a non
trivial invariant subalgebra, since κ(Y, [Z, X]) = −κ([Z, Y ], X) = 0, for any Z, Y ∈ g, X ∈ h.
Thus g is not simple. The proof that g is not semi-simple then consists in showing that h is
solvable, so that, with the definition in (5.52), h(n) is abelian for some n. The alternative
would require h(n) = h(n+1) , for some n, but this is incompatible with κ(X, Y ) = 0 for all
X, Y ∈ h.
The results (5.190) and (5.191) illustrate that su(2) is semi-simple, whereas iso(2) is
not, it contains an invariant abelian subalgebra.
For a compact Lie group G the result that a degenerate Killing form for a Lie algebra g
implies the presence of an abelian invariant subalgebra follows directly from (5.189) since
if X ad = 0, X commutes with all elements in g. For the compact case the Lie algebra can
be decomposed into a semi-simple part and an abelian part so that the group has the form

G ≃ Gsemi−simple ⊗ U (1) ⊗ ⋅ ⋅ ⋅ ⊗ U (1)/F , (5.194)

with a U (1) factor for each independent Lie algebra element with X ad = 0 and where F is
some finite abelian group belonging to the centre of G.

5.8.3 Decomposition of Semi-simple Lie Algebras

If a semi-simple Lie algebra g contains an invariant subalgebra h then the adjoint represen-
tation is reducible. However it may be decomposed into a direct sum of simple Lie algebras
for each of which the adjoint representation is irreducible. To verify this let

h⊥ = {X ∶ κ(X, Y ) = 0, Y ∈ h} . (5.195)

Then h⊥ is also an invariant subalgebra since, for any X ∈ h⊥ and Z ∈ g, Y ∈ h, κ([Z, X], Y ) =
−κ(X, [Z, Y ]) = 0. Furthermore h⊥ ∩ h = 0 since otherwise, by the definition of h⊥ in (5.195),
there would be a X ∈ h⊥ and also X ∈ h so that κ(X, Z) = 0 for all Z ∈ g which contradicts
the Killing form being non-degenerate. Hence

g = h ⊕ h⊥ . (5.196)

This decomposition may be continued to give until there are no remaining invariant spaces

g = ⊕ gi , gi simple . (5.197)
i

161
For the Lie algebra there is then a basis {Ta(i) }, such that for each individual i this represents
a basis for gi , a = 1, . . . dim gi , and with the generators for gi , gj , i ≠ j commuting as in (5.53)
and κ(Ta(i) , Tb(j) ) = 0, i ≠ j. For any X, Y ∈ g then the Killing form becomes a sum

X = ∑ Xi , Y = ∑ Yi , κ(X, Y ) = ∑ trgi (Xiad Yiad ) , (5.198)

i i i

The corresponding decomposition for the associated Lie group becomes G = ⊗i Gi .

With this decomposition the study of semi-simple Lie algebras is then reduced to just
simple Lie algebras.

5.8.4 Casimir Operators and Central Extensions

For semi-simple Lie algebras we may easily construct a quadratic Casimir operator for any
representation and also show that there are no non trivial central extensions.
The restriction to semi-simple Lie algebras, det[κab ] ≠ 0, ensures that the Killing form
κ = [κab ] has an inverse κ−1 = [κab ], so that κac κcb = δab , and we may then use κab and κab to
raise and lower Lie algebra indices, just as with a metric. The invariance condition (5.183)
becomes κ Taad + TaadT κ = 0 so that from [Taad , κ−1 κ] = 0 we obtain Taad κ−1 + κ−1 TaadT = 0
or
f bad κdc + f cad κbd = 0 , (5.199)
showing that κab is also an invariant tensor. Hence, for any representation of the Lie algebra
in terms of {ta } satisfying (5.60), then

[ta , κbc tb tc ] = κbc (f dab td tc + f dac tb td ) = (κbe f dab + κdc f eac ) td te = 0 . (5.200)

In consequence κab ta tb is a quadratic Casimir operator.

To discuss central extensions we rewrite the fundamental consistency condition (5.129)
in the form
hae f ecd = −hde f eac − hce f eda . (5.201)
Then using (5.199)

hae f ecd f c bg κgd = −hae f gcd f c bg κde = hae κdb κed = hab , (5.202)

and also, with (5.199) again,

(hde f eac + hce f eda )f cbg κgd = (hde f eac f c bg + hec f c bg f ead )κgd
= hde f eac f c bg κgd − hec f c bg f gad κde (5.203)

we may obtain from (5.201), re-expressing (5.203) as a matrix trace,

hab = −tr(h [Taad , Tbad ]κ−1 ) = −tr(h Tcad κ−1 )f cab . (5.204)

Hence hab is of the form given in (5.131) which demonstrates that for sem-simple Lie algebras
there are no non trivial central extensions. Central extensions therefore arise only when are
invariant abelian subalgebras.

162
6 Lie Algebras for Matrix Groups

Here we obtain the Lie algebras g corresponding to the various continuous matrix groups
G described in section 1.6 by considering matrices close to the identity

M = 1 + X + O(X 2 ) , (6.1)

with suitable conditions on X depending on the particular group.

6.1 Unitary Groups

For u(n), X is a complex n × n matrix satisfying X † = −X and for su(n), also tr(X) = 0.
It is convenient to consider first a basis formed by the n2 , n × n, matrices {Ri j }, where Ri j
has 1 in the i’th row and j’th column and is otherwise zero,

j
⎛0 0 . . . 0 ... 0⎞
⎜⋮ ⋮⎟
⎜ ⎟
i⎜0 1 0⎟
i
Rj= ⎜
⎜⋮
⎟, i, j = 1, . . . , n , (Ri j )† = Rj i . (6.2)
⎜ ⋮⎟
⎟
⎜ ⎟
⎜0 ⋱ 0⎟
⎝0 0 . . . 0 ... 0⎠
These matrices satisfy
[Ri j , Rk l ] = δ kj Ri l − δ i l Rk j , (6.3)
and
tr(Ri j Rk l ) = δ kj δ i l . (6.4)

In general X = Ri j X j i ∈ gl(n) for arbitrary X j i so that {Ri j } form a basis for gl(n). If
∑j X j j = 0 then X ∈ sl(n) while if (X j i )∗ = −X i j then X = −X † ∈ u(n). For the associated
adjoint matrices

[X, Ri j ] = X i k Rk j − Ri k X k j ⇒ (X ad )l k, i j = X i k δ l j − X l j δ i k . (6.5)

Hence, for X = Ri j X j i , Y = Ri j Y j i ,

κ(X, Y ) = tr(X ad Y ad ) = 2(n ∑i,j X j i Y i j − ∑i X i i ∑j Y j j ) . (6.6)

Restricting to u(n)

κ(X, X) = −2n ∑i,j ∣X̂ j i ∣2 , X̂ j i = X j i − 1

n δ j i ∑k X k k . (6.7)

Clearly κ(X, X) = 0 for X ∝ I reflecting that u(n) contains an invariant abelian subalgebra.
For su(n), when ∑k X k k = 0 and hence tr(X) = 0, then κ(X, X) = 2n tr(X 2 ) < 0.
A basis of n2 − 1 anti-hermitian generators for su(n), forming the fundamental repre-
sentation of this Lie algrebra, is provided by taking

{ta } = {i(Ri j + Rj i ), (Ri j − Rj i ) ∶ 1 ≤ i < j ≤ n} ∪ {i(Ri i − Ri+1 i+1 ); 1 ≤ i ≤ n − 1} . (6.8)

163
These satisfy

ta ta = − n n−1 1 ,
2
tr(ta ) = 0 , tr(ta tb ) = −2 δab , [ta , tb ] = f c ab tc , (6.9)

and the completeness condition

1
X = ta Xa Xa = 2 tr(ta X) for any X ∈ su(n) . (6.10)

Since
r
∑ tr(ta1 . . . [tc , tai ] . . . tar ) = 0 , (6.11)
i=1

then tr(ta1 ta2 . . . tar ) is an invariant tensor for su(n) in the basis provided by {ta }. Invariant
tensors can be chosen to be totally symmetric since otherwise the commutation relations
(6.9) allow the trace over r generators to be reduced to a sum over traces with r′ < r. A
potential basis for invariant tensors is then provided by

tr(t(a1 ta2 . . . tar ) ) , r = 2, . . . n , (6.12)

since for an n × n matrix A, tr(Ar ) for r > n is reducible to products of traces tr(Am ) with
m ≤ n. The invariant tensors are not unique since any symmetric rank r invariant tensor
dr,a1 ...ar is invariant up symmetrised products of lower rank ri tensors with ∑ ri = r.
Any representation of the su(n) Lie algebra has a basis {Ta } where [Ta , Tb ] = f c ab Tc
and then for invariant tensors dr,a1 ...ar

Cr = dr,a1 ...ar Ta1 . . . Tar satisfies [Ta , Cr ] = 0 , (6.13)

and so Cr is a Casimir operator. For su(n) there are thus n − 1 Casimirs. For an irreducible
representation Cr = cr 1.

6.2 Symplectic Groups

For sp(2n, R) or sp(2n, C) the condition (1.107) translates into

JX = −X T J = (JX)T , (6.14)

where J is the standard antisymmetric matrix given in (1.108). It can be represented by

Jij = −Jji = −(−1)i δij ′ , j ′ = j − (−1)j . (6.15)

It is convenient to define also

J ij = −J ji , J ik Jkj = δ i j , (6.16)

where Jij , Jij can be used to raise and lower indices. A basis for sp(2n, R), or sp(2n, C),
satisfying (6.14) is provided in terms of the 2n × 2n matrices {T i j }

(T i j )k l = δ i l δ k j + J ik Jjl , JT i j = (JT i j )T . (6.17)

164
With the choice (6.15) these have the explicit form

i′ j
⎛0 . . . 0 ... 0 ... 0⎞
⎜⋮ ⋮⎟
⎜ ⎟
i⎜ ⎜0 0 1 0⎟
⎟
i
T j = ⎜ ⎜⋮ ⋮⎟
⎟, 1 ≤ i < j ≤ 2n . (6.18)
j′ ⎜
⎜0 −(−1)i+j 0 0⎟
⎟
⎜ ⎟
⎜⋮ ⋮⎟
⎝0 . . . 0 ... 0 ... 0⎠

A linearly independent basis is given by {T i j , 1 ≤ i < j ≤ 2n; T 2i−1 2i−1 , T 2i 2i−1 , 1 ≤ i ≤ n}.
For n = 1 this prescription gives

1 0 0 2 0 0
T 1 1 = −T 2 2 = ( ), T 12 = ( ), T 21 = ( ). (6.19)
0 −1 0 0 2 0

From (6.17)
T i j = J il Jjk T k l , (6.20)
and the matrices {T i j } satisfy the Lie algebra

[T i j , T k l ] = δ i l T k j − δ k j T i l − J ik Jjm T m l + Jjl J im T k m , (6.21)

and also
tr(T i j T k l ) = 2(δ i l δ k j + J ik Jjl ) . (6.22)
For any X ∈ sp(2n, R), or sp(2n, C) then X = 12 X j i T i j . As a consequence of the symmetry
of T i j X j i should be restricted so that X j i = J jl Jik X k l . Using (6.21)

[X, T i j ] = X k j T i k − X i k T k j , (6.23)
1
so that from (5.171) [X, T i j ] = 2 T l k (X ad )k l ,i j giving

(X ad )k l ,i j = δ i l X k j − δ k j X i l + J ki Jlm X m j − Jlj J km X i m , (6.24)

and hence
κ(X, Y ) = 14 (X ad )kl , i j (Y ad )j i ,l k = (n + 2) X i j Y j i . (6.25)

An alternative basis for the Lie algebra is obtained by taking Tij = Jik T k j = Tji with
tr(Tij Tkl ) = −2(Jil Jjk + Jik Jjl ) and the Lie algebra becoming

[Tij , Tkl ] = Jik Tjl + Jil Tjk + Jjk Til + Jjl Tik , (6.26)

The generators here correspond to symmetric 2n dimensional matrices so the dimension is

clearly 12 2n(2n + 1).
For the corresponding compact group Sp(n) = Sp(2n, C) ∩ SU (2n) we impose, as well
as (6.14),
X † = −X or X j i∗ = −X i j . (6.27)

165
Then (6.25) gives
2
κ(X, X) = −(n + 2) ∑i,j ∣X j i ∣ . (6.28)
From (6.14)
JXJ −1 = −X T , (6.29)
so that, following the discussion in section 5.3.3, the fundamental representation of compact
Sp(n) is pseudo-real.
1
In the alternative basis in (6.26) a general X ∈ sp(2n, C) can be expressed as X = 2 X ij Tij
with X ij symmetric and (6.27) becomes

Xij∗ = J ik J jl Xkl . (6.30)

6.3 Orthogonal and Spin Groups

For o(n) or so(n) then in (6.1) we must require X T = −X so that tr(X) = 0. A basis for
n × n antisymmetric matrices is given by the 12 n(n − 1) matrices {Sij ∶ i < j} where

i j
⎛0 . . . 0 ... 0 ... 0⎞
⎜⋮ ⋮⎟
⎜ ⎟
⎜0
i⎜ 0 1 0⎟
⎟
Sij = −Sji = ⎜
⎜⋮ ⋮⎟
⎟, i ≠ j = 1, . . . , n . (6.31)
j⎜
⎜0 −1 0 0⎟
⎟
⎜ ⎟
⎜⋮ ⋮⎟
⎝0 . . . 0 ... 0 ... 0⎠

These satisfy
[Sij , Skl ] = δjk Sil − δik Sjl − δjl Sik + δil Sjk , (6.32)
and
tr(Sij Skl ) = 2(δil δjk − δik δjl ) . (6.33)
For arbitrary X ∈ so(2n) then X = 21 Xij Sij , where Xij = −Xji is real. From (6.32)
ad
[X, Sij ] = Xki Skj − Xkj Ski ⇒ Xkl,ij = Xki δlj − Xkj δli − Xli δkj + Xlj δki , (6.34)

and hence
ad ad
κ(X, Y ) = 41 Xkl,ij Yij,kl = (n − 2) Xij Yji . (6.35)
The matrices (6.31) are the generators for the vector representation of SO(n) which is of
course real, as described later there are also complex representations involving spinors.

6.3.1 Spin Groups and Gamma Matrices

The relation SO(3) and SU (2), which is described in section 3.2, and also the introduction
of spinorial representations, described in section 3.14, may be extended to higher orthogonal

166
groups. In the discussion for SO(3) and SU (2) an essential role was played by the Pauli
matrices. For SO(n) we introduce similarly gamma matrices, γi , i = 1, . . . , n which form
the basis for a Clifford algebra the Clifford46 algebra,

γi γj + γj γi = 2δij 1 , γi† = γi . (6.36)

The algebra may be extended to pseudo-orthogonal groups such as the Lorentz group, which
involve a metric gij as in (1.124), by taking δij → gij on the right hand side of (6.36). To
obtain explicit gamma matrices for SO(n, m) it is sufficient for each j with gjj = −1 just to
let γj → iγj for the corresponding SO(n + m) gamma matrices. For the non compact group
the gamma matrices are not all hermitian. (For gij as in (1.124) then if A = γ1 . . . γn then
Aγi A−1 = −(−1)n γi† .)
The representations of the Clifford algebra (6.36), acting on a representation space S,
are irreducible if S has no invariant subspaces under the action of arbitrary products of
γi ’s. As will become apparent there is essentially one irreducible representation for even n
and two, related by a change of sign, for odd n. If {γ ′i }, like {γi }, are matrices forming an
irreducible representation of (6.36) then γ ′i = Aγi A−1 , or possibly γ ′i = −Aγi A−1 for n odd,
for some A. As a consequence of (6.36)

(γ ⋅ x)2 = x2 1, x ∈ Rn . (6.37)

This the primary definition of a Clifford algebra where there a product for two vectors
belonging to a vector space V which is proportional to the unit operator on V .
To show the connection with SO(n) we first define

sij = 1
2 γ[i γj] = −sij† . (6.38)

Using just (6.36) it is easy to obtain

[sij , γk ] = δjk γi − δik γj , (6.39)

and hence
[sij , skl ] = δjk sil − δik sjl − δjl sik + δil sjk . (6.40)
This is identical with (6.32), the Lie algebra so(n). Moreover for finite transformations,
which involve the matrix exponential of 12 ωij sij , ωij = −ωji ,

e− 2 ωij sij γ ⋅ x e 2 ωij sij = γ ⋅ x′ , x′ = Rx , R = e− 2 ωij Sij ∈ SO(n) ,

1 1 1
(6.41)

with Sij ∈ so(n) as in (6.31). It is easy to see that x′2 = x2 , as required for rotations, as a
consequence of (6.37). To show the converse we note that γ ′i = γj Rji also satisfies (6.36)
for [Rji ] ∈ O(n) so that γ ′i = A(R)γi A(R)−1 where A(R) = e− 2 ωij sij for R continuously
1

connected to the identity.

The exponentials of the spin matrices form the group

Spin(n) = {e− 2 ωij sij ∶ ωij = −ωji ∈ R} .

1
(6.42)
46
William Kingdon Clifford, 1845-1879, English, second wrangler 1867.

167
Clearly Spin(n) and SO(n) have the same Lie algebra. For n = 3 we may let γi → σi and
sij = 21 iεijk σk so that Spin(3) ≃ SU (2). In general, since ±1 ∈ Spin(n) are mapped to
1 ∈ SO(n), we have SO(n) ≃ Spin(n)/Z2 .
Unlike SO(n), Spin(n) is simply connected and is the covering group for SO(n). For
further analysis we define

Γ = γ1 γ2 . . . γn = (−1) 2 n(n−1) Γ† , Γ† = γn γn−1 . . . γ1 ,

1
(6.43)

so that
1
Γ2 = (−1) 2 n(n−1) 1 . (6.44)
Directly from (6.36)

[Γ, γi ] = 0 , n odd , Γ γi + γi Γ = 0 , n even , i = 1, . . . n . (6.45)

Using, similarly to (3.38),

eα sij = cos 21 α 1 + sin 12 α 2sij , (6.46)
then

e−π ∑i=1 s2i−1 2i = (−1)m Γ ,

m m
eπ ∑i=1 s2i−1 2i = Γ , for n = 2m even . (6.47)

This allows the identification of the centres of the spin groups

⎧
⎪Z2 × Z2 , n = 4m ,
⎪
Z(Spin(n)) = {1, −1, Γ, −Γ} ≃ ⎨
⎩Z4 , n = 4m + 2 ,
⎪
⎪
Z(Spin(n)) = {1, −1} ≃ Z2 , n = 2m + 1 . (6.48)

Spinors for general rotational groups are defined as belonging to the fundamental rep-
resentation space S for Spin(n), so they form projective representations, up to a sign, of
SO(n).

6.3.2 Products and Traces of Gamma Matrices

For products of gamma matrices if the same gamma matrix γi appears twice in the product
then, since it anti-commutes with all other gamma matrices, as a consequence of (6.36),
and also γi2 = 1, it may be removed from the product, leaving the remaining matrices
unchanged apart from a possible change of sign. Linearly independent matrices are obtained
by considering products of different gamma matrices. Accordingly we define, for i1 , . . . ir
all different indices,

Γi1 ...ir = γ[i1 . . . γir ] = (−1) 2 r(r−1) Γi1 ...ir† , Γi1 ...ir† = Γir ...i1 ,
1
r = 1, . . . n , (6.49)
1
where Γi1 ...ir2 = (−1) 2 r(r−1) 1 and with the usual summation convention
1
r! Γi1 ...ir Γir ...i1 = (nr) 1 . (6.50)

168
From the definition (6.43)

Γi1 ...in = εi1 ...in Γ , n = 1, 2, . . . . (6.51)

We also have the relations

1 1
Γi1 ...ir = (−1) 2 n(n−1)+ 2 r(r−1) 1
s! εi1 ...ir j1 ...js Γj1 ...js Γ , r + s = n. (6.52)

A basis for these products of γ-matrices is given by Cr = {Γi1 ...ir ∶ 1 ≤ i1 < i2 < ⋅ ⋅ ⋅ < ir ≤ n},
with dim Cr = (nr), Cn = {Γ}. It is easy to see that C (n) = {±1, ±C1 , . . . , ±Cn−1 , ±Γ} is closed
under multiplication and therefore forms a finite matrix group, with dim C (n) = 2 ∑nr=0 (nr) =
2n+1 . The matrices {1, C1 , . . . , Cn } may also be regarded as the basis vectors for a 2n -
dimensional vector space which is also a group under multiplication, and so this forms a
field.
When n is odd then from (6.45) Γ commutes with all elements in C (n) and so for an
irreducible representation we must have Γ ∝ 1. Taking into account (6.44)
⎧
⎪±1 ,
⎪ n = 4m + 1
Γ=⎨ . (6.53)
⎩±i 1 , n = 4m + 3
⎪
⎪

The ± signs correspond to inequivalent representations, linked by taking γi → −γi . For a lin-
early independent basis then, as a consequence of (6.52), it is necessary to recognise that the
products of gamma matrices are no longer independent if r > 12 n. The matrix groups formed
from the irreducible representations for n odd are then, for the two cases in (6.53), C (4m+1) =
{±1, ±C1 , . . . , ±C2m }, dim C (4m+1) = 24m+1 , and C (4m+3) = {±1, ±i1, ±C1 , ±i C1 , . . . , ±i C2m+1 },
dim C (4m+3) = 24m+4 . Thus for n = 3 there is the Pauli group of order 16 formed from the
2 × 2 matrices {±12 , ±i12 , ±σi , ±iσi }, with σi the usual Pauli matrices or Q8 ∪ i Q8 . This
group corresponds to Q8 × Z4 /Z2 with Z4 = {±1, ±i} and Z2 = {±1}. The Pauli group has a
D4 subgroup formed by {±12 , ±σ1 , ±σ2 , ±iσ3 }.
For n even C (n) does not contain any elements commuting with all γi but

[Γ, sij ] = 0 . (6.54)

Hence we may decompose the representation space S = S+ ⊕ S− , such that ΓS± = S± and,
since γi anti-commutes with Γ, γi S± = S∓ . Hence there is a corresponding decomposition
of the gamma matrices with Γ diagonal and where, using (6.44),
⎧
⎪ ( 1 0 ) , n = 4m ,
⎪ 0 σi s+ij 0
Γ = ⎨ 0 I−10 γi = ( ), sij = ( ). (6.55)
σ̄i 0 0 s−ij
⎩ i ( 0 −I ) , n = 4m + 2 ,
⎪
⎪

Clearly σ̄i = σi† and s+ij = 1

2 σ[i σ̄j] , s−ij = 1
2 σ̄[i σj] and just as in (6.38) we have

s±ij† = −s±ij . (6.56)

With the decomposition in (6.55) the Clifford algebra (6.36) is equivalent to

σi σ̄j + σj σ̄i = 2δij 1 , σ̄i σj + σ̄j σi = 2δij 1 . (6.57)

169
 For traces of gamma matrices and their products we first note that from (6.36)

tr(γj (γi γj + γj γi )) = 2 tr(γj γj γi ) = 2 tr(γi ) = 0 , j ≠ i , no sum on j . (6.58)

We may similarly use γj Γi1 ...ir + Γi1 ...ir γj = 0, when r is odd and for j ≠ i1 , . . . , ir , or
γj Γj i2 ...ir + Γj i2 ...ir γj = 0, when r is even and with no sum on j, to show that

tr(Γi1 ...ir ) = 0 , except when r = n, n odd . (6.59)

Hence in general, for r, s = 0, . . . , n for n even, or with r, s < 12 n for n odd,

tr(Γi1 ...ir Γjs ...j1 ) = δrs dn (Ar )i1 ...ir ,j1 ...jr , (6.60)

where

dn = tr(1) ,
⎧
⎪
⎪
⎪±1 , (j1 , . . . , jr ) an even/odd
⎪ permutation of (i1 , . . . , ir ) ,
(Ar )i1 ...ir ,j1 ...jr = r! δ[i1 ∣j1 ∣ δi2 ∣j2 ∣ . . . δir ]jr =⎨ (6.61)
⎪
⎪
⎩ 0, otherwise
⎪
⎪

so that Ar /r! is a projection operator for antisymmetric rank r tensors.

With dn -dimensional spinorial indices α, β, . . . the Γ-matrices can be represented dia-
grammatically
⎧
⎪ β ⎧
⎪ α
⎪
⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ r=0
⎪
⎪
⎪ r=0 ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ β
⎪
⎪ α ⎪
⎪
(Γi1 ...ir )α β =⎨ β , (Γir ...i1 )α β =⎨ α , (6.62)
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
⎪ r≥1 ⎪
⎪
⎪ r
r≥1
⎪
⎪ r ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪ α ⎪
⎪ β
⎩ ⎩

with = dn , = 0. (6.60) is then equivalent to

, (6.63)
r s = dn δrs

The identity (6.50) corresponds to

r
n .
= ( ) (6.64)
r

These matrices then have a norm

1 n
r! tr(Γi1 ...ir Γir ...i1 ) = = dn ( ) . (6.65)
r

170
 In general these products of gamma matrices form a complete set so that
dn δα δ δγ β = δα β δγ δ + ∑ 1
r! (Γi1 ...ir )α β (Γir ...i1 )γ δ ,
r≥1
β γ β γ β γ

dn = + ∑r≥1 (6.66)
r
α δ α δ α δ

with r = 1, . . . , n, dn = 2 2 n for n even, and r = 1, . . . , 12 (n − 1), dn = 2 2 (n−1) for n odd.

1 1

6.3.3 Construction of Representations of the Clifford Algebra

For n = 2m an easy way to construct the γ-matrices satisfying the Clifford algebra (6.36)
explicitly is to define
ar = 12 (γ2r−1 + iγ2r ) , ar† = 21 (γ2r−1 − iγ2r ) , r = 1, . . . , m . (6.67)
Then (6.36) becomes
ar as + as ar = 0 , ar as† + as ar† = δrs 1 , (6.68)
which is just the algebra for m fermionic creation and annihilation operators, the femionic
analogue of the usual bosonic harmonic oscillator operators. The construction of the es-
sentially unique representation space S for such operators is standard, there is a vacuum
state annihilated by all the ar ’s and all other states in the space are obtained by acting on
the vacuum state with linear combinations of products of ar† ’s. In general, since ar† 2 = 0, a
† sr
basis is formed by restricting to products of the form ∏m r=1 (ar ) with sr = 0, 1 for each r.
There are then 2 independent basis vectors, giving dim S = 2m . For m = 1 then we may
m

take, with the ‘vacuum state’ represented by ( 10 ),

a = σ+ = ( 00 10 ) , a† = σ− = ( 01 00 ) , γ1 γ2 = i σ3 = i( 10 −1
0 ). (6.69)
The general case is obtained using tensor products
ar = 1 ⊗ ⋅ ⋅ ⋅ ⊗ 1 ⊗ σ+ ⊗ σ3 ⊗ ⋅ ⋅ ⋅ ⊗ σ3 , ar† = 1 ⊗ ⋅ ⋅ ⋅ ⊗ 1 ⊗ σ− ⊗ σ3 ⊗ ⋅ ⋅ ⋅ ⊗ σ3 . (6.70)
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
r−1 m−r r−1 m−r

The σ3 ’s appearing in the tensor products follow from the requirement that ar , as , and
ar† , as† , anti-commute for r ≠ s. With (6.70) γ2r−1 γ2r = i 1 ⊗ ⋅ ⋅ ⋅ ⊗ 1 ⊗ σ3 ⊗ 1 ⋅ ⋅ ⋅ ⊗ 1 so that
Γ = i m σ3 ⊗ σ3 ⊗ ⋅ ⋅ ⋅ ⊗ σ3 . (6.71)
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
m

These results are equivalent to defining the gamma matrices for increasing n, where
γ i(n) γ j(n)
+ γ j(n) γ i(n) = 2δij 1(n) , recursively in terms of the Pauli matrices by

γ i(2m+2) = γ i(2m) ⊗ σ3 , i = 1, . . . , 2m ,
(2m+2) (2m+2)
γ2m+1 = 1(2m) ⊗ σ1 , γ2m+2 = 1(2m) ⊗ σ2 ,
Γ(2m+2) = i Γ(2m) ⊗ σ3 . (6.72)

171
Note that we may take γ i(2) = σi , i = 1, 2 with Γ(2) = i σ3 . For odd n the gamma matrices
may be defined in terms of those for n − 1 by
⎧
(2m+1) ⎪±1 , m even ,
⎪
γ i(2m+1) = γ i(2m) , i = 1, . . . , 2m , γ2m+1 = cm Γ(2m) , cm = ⎨ (6.73)
⎩±i , m odd ,
⎪
⎪

where the ± signs correspond to inequivalent representations. Thus γ i(3) = (σ1 , σ2 , ∓σ3 ).

6.3.4 Conjugation Matrix for Gamma Matrices

It is easy to see that γiT also obeys the Clifford algebra in (6.36) so that for an irreducible
representation we must have

Cγi C −1 = − γiT C Γ C −1 = (−1) 2 n(n+1) ΓT

1
⇒
or Cγi C −1 = γiT C ΓC −1 = (−1) 2 n(n−1) ΓT .
1
⇒ (6.74)

When n is even then, by taking C → CΓ, the two cases are equivalent. When n is odd, and
we require (6.53), then for n = 4m + 1, C must satisfy Cγi C −1 = γiT , for n = 4m + 3, then
Cγi C −1 = −γiT . In either case for the spin matrices in (6.38)

C sij C −1 = −sijT , (6.75)

so that for the matrices defining Spin(n)

e− 2 ωij sij C (e− 2 ωij sij ) = C .

1 1 T
(6.76)

With the recursive construction of the gamma matrices γi(n) in (6.72) we may also
construct in a similar fashion C (n) iteratively since, using (5.77),

C (n) γi(n) C (n) −1 = γi(n) T

⇒ C (n+2) = C (n) ⊗ iσ2 ensures C (n+2) γi(n+2) C (n+2) −1 = −γi(n+2) T , (6.77)

and, using σ1 σi σ1 = σiT , i = 1, 2, σ1 σ3 σ1 = −σ3T ,

C (n) γi(n) C (n) −1 = −γi(n) T

⇒ C (n+2) = C (n) ⊗ σ1 ensures C (n+2) γi(n+2) C (n+2) −1 = γi(n+2) T . (6.78)

Starting from n = 0, or n = 2, this construction gives (note that (X ⊗ Y )T = X T ⊗ Y T ),

C γi C −1 = γi T , C Γ C −1 = ΓT , C = CT , n = 8k ,
−1 T −1 T T
C γi C = − γi , C ΓC = −Γ , C = −C , n = 8k + 2 ,
−1 T −1 T T
C γi C = γi , C ΓC =Γ , C = −C , n = 8k + 4 ,
C γi C −1 = − γi T , C Γ C −1 = − ΓT , C = CT , n = 8k + 6 . (6.79)

172
In each case we have Csij C −1 = −sij T . Starting from (6.79) and with the construction in
(6.73) for odd n,

C γi C −1 = γi T , C = CT , n = 8k + 1 ,
−1 T T
C γi C = − γi , C = −C , n = 8k + 3 ,
C γi C −1 = γi T , C = − CT , n = 8k + 5 ,
−1 T T
C γi C = − γi , C=C , n = 8k + 7 . (6.80)

The definition of C for n = 2m + 1 remains the same as in (6.79) for n = 2m since in each n
odd case we have Cγ1 γ2 . . . γn C −1 = (γ1 γ2 . . . γn )T .
If we consider a basis in which Γ is diagonal, as in (6.55), then for n = 8k, 8k+4 [C, Γ] = 0,
so that C is block diagonal, while for n = 8k + 2, 8k + 6 CΓ + ΓC = 0, so that we may take
C to have a block off diagonal form. By considering the freedom under C → S T CS with
SΓS −1 = Γ we may choose with the basis in (6.55),

1 0
C=( ), σ̄i = σiT , s±ij = − s±ijT , n = 8k ,
0 1
0 1
C=( ), σi = σiT , σ̄i = σ̄iT , s±ij = − s∓ijT , n = 8k + 2 ,
−1 0
J 0
C=( ), J = − J T , J σ̄i = −(Jσi )T , Js±ij = (Js±ij )T , n = 8k + 4 ,
0 J
0 1
C=( ), σi = − σiT , σ̄i = −σ̄iT , s±ij = − s∓ijT , n = 8k + 6 . (6.81)
1 0

Here the antisymmetric matrix J can be taken to be of the standard form as in (1.108).
For n = 8k the matrices are real.
Since the generators of the two fundamental spinor representations satisfy (6.56) then as
a consequence of the discussion in section 5.3.3 we have for these representations of Spin(n),
for n even, from (6.81)

Spin(8k) ∶ real , Spin(8k + 4) ∶ pseudo-real ,

Spin(8k + 2) , Spin(8k + 6) ∶ complex . (6.82)

Furthermore for n odd the single spinor representation, from (6.80), satisfies

Spin(8k + 1) , Spin(8k + 7) ∶ real , Spin(8k + 3) , Spin(8k + 5) ∶ pseudo-real . (6.83)

6.3.5 Special Cases

When n = 2 we may take

σi = (1, −i) , σ̄i = (1, i) , (6.84)
while for n = 4 we may express σi , σ̄i in terms of unit quaternions

σi = (1, −i, −j, −k) , σ̄i = (1, i, j, k) . (6.85)

173
 For low n results for γ-matrices may be used to identify Spin(n) with other groups.
Thus

Spin(3) ≃ SU (2) , Spin(4) ≃ SU (2) × SU (2) ,

Spin(5) ≃ Sp(2) , Spin(6) ≃ SU (4) . (6.86)

For n = 3 it is evident directly that e− 2 ωij sij ∈ SU (2). For n = 4 as a consequence of (6.52),
1

with the decomposition in (6.55), we have

s±ij = ± 12 εijkl s±kl , (6.87)

so that e− 2 ωij sij = e− 2 ω+ij s+ij ⊗ e− 2 ω−ij s−ij factorises a 4 × 4 Spin(4) matrix into a product of
1 1 1

two independent SU (2) matrices as ω±ij = 21 ωij ± 14 εijkl ωkl are independent. For n = 5 then
the 4 × 4 matrix e− 2 ωij sij ∈ SU (4) ∩ Sp(4, C), using (6.76) with C T = −C. In this case there
1

are 10 independent sij which matches with the dimension of the compact Sp(2). For n = 6,
e− 2 ω+ij s+ij ∈ SU (4) with the 15 independent 4 × 4 matrices s+ij matching the dimension
1

of SU (4). Note also that, from (6.48), Z(Spin(6)) ≃ Z4 ≃ Z(SU (4)). Using (6.81) with
(6.55), the transformation (6.41) can be rewritten just in terms of the SU (4) matrix

e− 2 ωij s+ij σ ⋅ x (e− 2 ωij s+ij ) = σ ⋅ x′ ,

1 1 T
(6.88)

which is analogous to (3.27). The result that the transformation x → x′ satisfies x2 = x′2
also follows in a similar fashion to (3.29), but in this case using the Pfaffian (1.109) instead
of the determinant since we require Pf(σ ⋅ x) = x2 (from σ ⋅ x σ̄ ⋅ x = x2 I then, with n = 6,
det(σ ⋅ x) = (x2 )2 ).

6.3.6 Fierz Identities

Fierz47 identities which depend on the completeness properties of γ-matrices play a crucial
role in many calculations. To derive Fierz identities it is convenient to map the γ-matrices
to operator fermi fields ψ̂i so that

γi → ψ̂i , {ψ̂i , ψ̂j } = 2 δij 1 , (6.89)

with i a n-dimensional index and where ψ̂i is decomposed into fermionic creation and
annihilation operators

ψ̂i = bi + bi† , {bi , bj } = 0 , {bi , bj† } = δij 1 . (6.90)

Then from (6.49)

Γi1 ...ir → ∶ ψ̂i1 . . . ψ̂ir ∶ , (6.91)
where ∶ ⋅ ⋅ ⋅ ∶ denotes normal ordering, where annihilation operators are moved to the right
of creation operators taking into account anticommutation signs as required but dropping
non zero terms from (6.90), thus ∶ ψ̂i ψ̂i ∶ = 0 while ψ̂i ψ̂i = n 1.
47
Markus Fierz, 1912-2006, Swiss. A student of and assistant to Pauli.

174
 The normal ordered products are conveniently represented in terms of a generating
function

eθi ψ̂i = ∑∞
s=0
1
s! θis . . . θi1 ∶ ψ̂i1 . . . ψ̂is ∶ = ∑∞ 1 s
s=0 s! (−1) ∶ ψ̂is . . . ψ̂i1 ∶ θi1 . . . θis , (6.92)

where {θi } are arbitrary Grassmannian, or anticommuting, variables. These ensure that
only antisymmetrised products of ψ̂i arise in the expansion and therefore the products are
normal ordered. For n an integer the number of terms in the expansion is finite since
∶ ψ̂i1 . . . ψ̂is ∶ = 0 for s > n. As an application we may use

eθi ψ̂i eθ̃j ψ̂j = e(θ+θ̃)i ψ̂i eθ̃j θj , (6.93)

and expanding to O(θr , θ̃s ) gives

θ̃js . . . θ̃j1 θir . . . θi1 ∶ ψ̂i1 . . . ψ̂ir ∶ ∶ ψ̂j1 . . . ψ̂js ∶

min(r,s) r! s!
= ∑t=0 (r−t)! (s−t)! t! θ̃js−t . . . θ̃j1 θ̃kt . . . θ̃k1 θk1 . . . θkt θir−t . . . θi1 ∶ ψ̂i1 . . . ψ̂ir−t ψ̂j1 . . . ψ̂js−t ∶ .
(6.94)

This implies

Γi1 ...ir Γj1 ...js

min(r,s) 1
= ∑t=0 r! s!
(r−t)! (s−t)! t!2
(−1) 2 t(2s−t−1) Γ[i1 ...ir−t ∣[j1 ...js−t (At )js−t+1 ...js ],ir−t+1 ...ir ] , (6.95)

where (As )j1 ...js ,i1 ...is is defined in (6.61) and acts as the δ-function for antisymmetric rank
s tensors so that s!1 (As )j1 ...js ,i1 ...is θ11 . . . θis = θj1 . . . θjs . In (6.95) the i-indices and j-indices
are separately antisymmetrised.

In a similar fashion starting from eφi ψ̂i eθj ψ̂j eφ̃k ψ̂k we may obtain

φ̃kt . . . φ̃k1 θjs . . . θj1 φir . . . φi1 tr(Γi1 ...ir Γj1 ...js Γk1 ...kt ) = dn r! s! t!
a! b! c! (φ̃i θi )a (φ̃j φj )b (θk φk )c ,
a = 12 (s + t − r) , b = 12 (t + r − s) , c = 21 (r + s − t) . (6.96)

This is non zero for ∣r − s∣ ≤ t ≤ r + s, r + s + t even and implies

tr(Γi1 ...ir Γj1 ...js Γk1 ...kt )

= dn r! s! t!
(a! b! c!)2
(−1)sb (Aa )[kt ...kb+1 ,[js ...jc+1 (Ab )kb ...k1 ],[ir ...ic+1 (Ac )jc ...j1 ],ic ...i1 ] . (6.97)

Fierz identities are obtained by considering

2/∂φ
eu ∂ l ∂ φ̃l
eφj ψ̂j eθi ψ̂i eφ̃k ψ̂k ∣φ=φ̃=0 = ∑∞
r=0
1
r! ur ∶ ψ̂jr . . . ψ̂j1 ∶ eθi ψ̂i ∶ ψ̂j1 . . . ψ̂jr ∶
= ∑∞
s=0
1
s! θis . . . θ i1 ∑ ∞
r=0
1
r! ur (−1)rs ∶ ψ̂jr . . . ψ̂j1 ∶ ∶ ψ̂i1 . . . ψ̂is ∶ ∶ ψ̂j1 . . . ψ̂jr ∶ , (6.98)
∂ ∂
where {φj , φ̃j } are further independent Grassmannian variables and ∂φl φj + φj ∂φ l
= δjk .

175
This can be evaluated using48
2/∂φ ∂ φ̃ 2/∂φ ∂ φ̃
eu ∂ l l
eφj ψ̂j eθi ψ̂i eφ̃k ψ̂k ∣φ=φ̃=0 = eu ∂ l l
eφj ψ̂j +φ̃j (ψ̂j +2θj )−φj φ̃j ∣φ=φ̃=0 eθi ψ̂i
1−u
= (1 + u)n e 1+u θi ψ̂i = ∑∞ 1
s=0 s! (1 + u)
n−s
(1 − u)s θis . . . θi1 ∶ ψ̂i1 . . . ψ̂is ∶ . (6.99)
Thus
1
r! ∶ ψ̂jr . . . ψ̂j1 ∶ ∶ ψ̂i1 . . . ψ̂is ∶ ∶ ψ̂j1 . . . ψ̂jr ∶ = Ωrs ∶ ψ̂i1 . . . ψ̂is ∶ , s = 1, 2, . . . ,
1
r! ∶ ψ̂jr . . . ψ̂j1 ∶ ∶ ψ̂j1 . . . ψ̂jr ∶ = Ωr0 1 , (6.100)
for
1 dr
Ωrs = (−1)rs r! dur (1 + u)
n−s
(1 − u)s ∣u=0
min(r,s)
= (−1)rs ∑t=0 (−1)t ( st ) ( n−s
r−t )
n−s
= (−1)rs ( r ) F (−r, −s; n − r − s + 1; −1)
= (−1)rs (nr) F (−r, −s; −n; 2) , (6.101)
where F is a hypergeometric function and the last line follows by the ralation between
hypergeometric functions with argument x and 1 − x. For integer n it is necessary that
r, s ≤ n. Manifestly r!(n − r)! Ωrs = s!(n − s)! Ωsr and, using the properties of hypergeometric
functions, Ωrs = (−1)r(n−1) Ωrn−s = (−1)s(n−1) Ωn−rs . In general Ωr0 = ( nr ), Ω0s = 1, Ω1s =
(−1)s (n − 2s) and also
⎧
⎪(2 + u)n−s (−u)s ∣u=0 ,
⎪ s even
∑r Ωrs = exp ((−1)s du
d
)(1 + u)n−s (1 − u)s ∣u=0 = ⎨ n−s s
⎩u (2 − u) ∣u=0 , s odd
⎪
⎪
⎧
⎪ 2n ,
⎪ s = 0 , s = n , n odd
=⎨ . (6.102)
⎩ 0, otherwise
⎪
⎪

As particular cases for even n

1 1 1 1 1
1 1 1 ⎛ 4 −2 0 2 −4 ⎞
Ω∣n=2 = ( 2 0 −2 ) , Ω∣n=4 = 6 0 −2 0 6 ,
1 −1 1 ⎝ 41 −12 10 −2 −4 ⎠
−1 1
1 1 1 1 1 1 1
⎛ 6 −4 2 0 −2 4 −6 ⎞
15 5 −1 −3 −1 5 15 ⎟
Ω∣n=6 = ⎜
⎜ 20 0 −4 0 4 0 −20 , (6.103)
15 −5 −1 3 −1 −5 15 ⎟
⎝ 6 4 2 0 −2 −4 −6 ⎠
1 −1 1 −1 1 −1 1
48
When the derivatives act on a normal ordered function of ψ̂ then, in a similar fashion to the discussion
in 3.15.1, the operator anticommutation relations are irrelevant and for evaluation we can let ψ̂ → ψ, an
ordinary Grassmann variable. The necessary identity may then be obtained, for φ̃, φ, ψ̃, ψ n dimensional
Grassmann variables and U, V n × n matrices, by using

e−∂/∂ φ̃ U ∂/∂φ e−φ V φ̃ eφψ̃+ψφ̃ ∣φ=φ̃=0 = e∂/∂ ψ̃ V ∂/∂ψ eψ U ψ̃ = e∂/∂ ψ̃ V ∂/∂ψ det U ∫ dn η dn η̃ e−η U
−1
η̃+η ψ̃+ψ η̃

= det U ∫ dn η dn η̃ e−η (U +V ) η̃+η ψ̃+ψ η̃

= det(1 + U V ) eψ (1+U V )
−1 −1
U ψ̃
,

where the crucial step is to express eψ U ψ̃ in terms of an integral over Grassmann variables η, η̃ and with the
conventions here eηψ̃+ψη̃ is an eigenvector for ∂/∂ ψ̃, ∂/∂ψ with eigenvalues −η, η̃. In application to (6.99)
φ → φ + φ̃, ψ̃ → ψ, ψ → −2θ and U → u 1, V → 1 so that 1 − 2(1 + U V )−1 U → 1−u 1+u
.

176
and for n odd (6.101) gives
1 1 1 1 1 1
1 1 1 1 ⎛ 5 −3 1 1 −3 5 ⎞
Ω∣n=3 = ( 33 −1 −1 3) , Ω∣n=5 = ⎜ 10 2 −2 −2 2 10 ⎟ . (6.104)
−1 −1 3 10 2 −2 −2 2 10
1 1 1 1 ⎝ 5 −3 1 1 −3 5 ⎠
1 1 1 1 1 1

By virtue of (6.91), (6.100) is equivalent to the γ-matrix identity

1
r! Γjr ...j1 Γi1 ...is Γj1 ...jr = Ωrs Γi1 ...is ,
(6.105)
r = Ωrs ,
s s

from which it follows that

r
r
s s
= dn Ωrs , s = dn Ωrs (ns) .
(6.106)

Clearly it is necessary that Ωrs (ns) is symmetric under r ↔ s which is evident from (6.101).
With the completeness relation (6.66) this implies a crossing relation as in (3.194)
1
r! (Γjr ...j1 )α δ (Γj1 ...jr )γ β = 1
dn ∑s≥0 Ωrs 1
s! (Γis ...i1 )α β (Γi1 ...is )γ δ ,
β γ
γ
β (6.107)
1 .
r = dn ∑ Ωrs s
s≥0 α δ
α δ

These relations are what is usually referred to as the Fierz identity. For consistency it is
necessary that
2 2 2
∑t≥0 Ωrt Ωts = dn δrs , or Ω = dn 1n+1 , (6.108)
which requires, except when n is an odd integer,
1
dn = 2 2 n . (6.109)

As a check we may use the first line of (6.101) to obtain via a Taylor expansion
t s n−s s
∑t≥0 x Ωts = (1 + (−1) x) (1 − (−1)s x) , (6.110)

and then use this in (6.108) to obtain

1 dr n−s s
∑t≥0 Ωrt Ωts = r! dur
(1 + u + (−1)r+s (1 − u)) (1 + u − (−1)r+s (1 − u)) ∣u=0
⎧ r ⎧
⎪ d r us ∣u=0 ,
⎪ r + s even ⎪δrs ,
⎪ r + s even
= 1
r! 2n ⎨ du
dr n−s
= 2n ⎨ (6.111)
⎩ dur u ∣u=0 , r + s odd ⎩1 , r + s = n odd .
⎪
⎪ ⎪
⎪

177
 For n = 2m + 1 odd {1, Γi1 ...ir ∶ 1 ≤ r ≤ m} form an independent basis and we have, as a
consequence of the symmetry properties of Ω,

Ω̂ Ω̂ C
Ω=( ), Crs = δr+s,m+1 , C 2 = 1 , (6.112)
C Ω̂ C Ω̂ C

where Ω̂, C are (m+1)×(m+1) matrices, C is anti-diagonal. Clearly in this case from (6.104)
det Ω = 0. The result (6.111) for n = 2m + 1 requires Ω2 = 2 ( Ω̂2 C Ω̂2 ) = 22m+1 ( 1C 1C ) and is
2 2

Ω̂ C C Ω̂ C
then satisfied for
Ω̂2 = 22m 1m+1 . (6.113)
Hence in (6.105) and (6.107) we may restrict 0 ≤ r, s ≤ m and take Ω → Ω̂, dn → 2m .
For n = 3, 5 the 2 × 2, 3 × 3 reduced matrices forming Ω̂ in these cases are given by the
corresponding top left hand sub matrices in (6.104). For n = 7, m = 3
1 11 1
7 −5
Ω̂∣n=7 = ( 21 3 −1 ) . (6.114)
9 1 −3
35 −5 −5 3

In general Ωrs is a polynomial in n of degree r and the result (6.101) with (6.109)
may be extended to any complex n. In this case {1, Γi1 ...ir ∶ r = 1, 2, . . . } span an infinite
dimensional space although the norm in (6.65) is no longer positive definite when r > n.
There is an infinite sum in (6.108) which is convergent for Re n > r + s and may then defined
by analytic continuation in n.

178
7 SU (3) and its Representations

SU (3) is an obvious generalisation of SU (2) although that was not the perception in the
1950’s when many physicists were searching for a higher symmetry group, beyond SU (2) and
isospin, to accommodate and classify the increasing numbers of resonances found in particle
accelerators with beams of a few GeV . Although the discovery of the relevance of SU (3)
as a hadronic symmetry group was a crucial breakthrough, leading to the realisation that
quarks are the fundamental constituents of hadrons, it now appears that SU (3) symmetry
is just an almost accidental consequence of the fact that the three lightest quarks have a
mass which is significantly less than the typical hadronic mass scale.
Understanding SU (2) and its representations is an essential first step before discussing
general simple Lie groups. Extending to SU (3) introduces many of the techniques which
are needed for the general case in a situation where the algebra is still basically simple and
undue mathematical sophistication is not required. For general SU (N ) the Lie algebra is
given, for the associated chosen basis, by (6.3) where, since the corresponding matrices in
(6.2) are not anti-hermitian, we are regarding the Lie algebra as a complex vector space.
To set the scene for SU (3) we reconsider first SU (2).

7.1 Recap of su(2)

For the basic generators of su(2) we define in terms of 2 × 2 matrices as in (6.2)

0 1 0 0 1 0
e+ = ( ), e− = ( ), h=( ), (7.1)
0 0 1 0 0 −1

which satisfy the Lie algebra

[e+ , e− ] = h , [h, e± ] = ±2 e± . (7.2)

These matrices satisfy

e+† = e− , h† = h . (7.3)
Under interchange of the rows and columns

0 1
b=( ) ⇒ b{e+ , e− , h}b−1 = {e− , e+ , −h} . (7.4)
1 0

Clearly b2 = 1 and {e+ , e− , h}, {e− , e+ , −h} must satisfy the same commutation relations as
in (7.2) so b generates an automorphism.
For representations of the su(2) Lie algebra then we require operators

l = {E+ , E− , H} , [E+ , E− ] = H , [H, E± ] = ±2 E± . (7.5)

It is easy to see that the commutation relations are identical with (3.61a) and (3.61a),
and also the hermeticity conditions with (3.62), by taking J± → E± , 2J3 → H. Indeed the
representation matrices in (7.1) then correspond exactly with (3.103).

179
 An important role in the general theory of Lie groups is played by the automorphism
symmetries of a privileged basis for the Lie algebra which define the Weyl49 group. For
su(2) the relevant basis is given by (7.5) and then from (7.4) there is just one non trivial
automorphism
l → lR = {E− , E+ , −H} . (7.6)
b
2
Since b = I the Weyl group for su(2), W (su(2)) ≃ Z2 .
For representations we require a finite dimensional representation space on which there
are operators E± , H which obey the commutation relations (7.5) and subsequently require
there is a scalar product so that the operators satisfy the hermeticity conditions in (7.3).
A basis for a representation space for su(2) is given by {∣r⟩} where

H∣r⟩ = r∣r⟩ . (7.7)

The eigenvalue r is termed the weight. It is easy to see from (7.5) that

E± ∣r⟩ ∝ ∣r ± 2⟩ unless E+ ∣r⟩ = 0 or E− ∣r⟩ = 0 . (7.8)

We consider representations where there is a highest weight, rmax = n, and hence a highest
weight vector ∣n⟩hw satisfying
E+ ∣n⟩hw = 0 . (7.9)
The representation space Vn is then spanned by

{E−r ∣n⟩hw ∶ r = 0, 1, . . . } . (7.10)

On this basis
HE−r ∣n⟩hw = (n − 2r) E−r ∣n⟩hw , (7.11)
and using
r−1 r−1
[E+ , E−r ] = ∑ E−r−s−1 [E+ , E− ]E−s = E−r−1 ∑ (H − 2s) = E−r−1 r(H − r + 1) , (7.12)
s=0 s=0

then from (7.9),

E+ E−r ∣n⟩hw = r(n − r + 1) E−r−1 ∣n⟩hw . (7.13)
(7.11) and (7.13) ensure that the commutation relations (7.5) are realised on Vn .
If n ∈ N0 , or n = 0, 1, 2, . . . , then from (7.13)

∣−n − 2⟩hw = E−n+1 ∣n⟩hw ∈ Vn , (7.14)

is also a highest weight vector, satisfying (7.9). From ∣−n − 2⟩hw we may construct, just as
in (7.10), a basis for an associated invariant subspace

V−n−2 ⊂ Vn . (7.15)
49
Hermann Klaus Hugo Weyl, 1885-1955, German.

180
Hence the representation space defined by the basis Vn is therefore reducible under the action
of su(2). An irreducible representation is obtained by restricting to the finite dimensional
quotient space
Vn = Vn /V−n−2 . (7.16)

In general for a vector space V with a subspace U the quotient V /U is defined by

V /U = { ∣v⟩/∼ ∶ ∣v⟩ ∼ ∣v ′ ⟩ if ∣v⟩ − ∣v ′ ⟩ ∈ U } . (7.17)

It is easy to verify that V /U is a vector space and, if V, U are finite-dimensional, dim(V /U ) =

dim V − dim U . If X is a linear operator acting on V then

U →U ⇒ {X∣v⟩/ ∼} = {X∣v ′ ⟩/ ∼} if ∣v⟩ ∼ ∣v ′ ⟩ ⇒ X ∶ V /U → V /U . (7.18)

X

Thus, if U ⊂ V is an invariant subspace under X, then X has a well defined action on V /U .

Furthermore for traces
trV /U (X) = trV (X) − trU (X) . (7.19)

Since V−n−2 is an invariant subspace under the action of the su(2) Lie algebra generators
we may then define E± , H to act linearly on the quotient Vn given by (7.16). On Vn this
ensures
E−n+1 ∣n⟩hw = 0 , (7.20)
so that there is a finite basis {E−r ∣n⟩hw ∶ r = 0, . . . , n}. In terms of the angular momentum
representations constructed in section 3, n = 2j. The space Vn may equally be constructed
from a lowest weight state ∣−n⟩ satisfying H∣−n⟩ = −n ∣−n⟩, E− ∣−n⟩ = 0, in accord with the
automorphism symmetry (7.4) of the su(2) Lie algebra.
If we define a formal trace over all vectors belonging to Vn then
∞
tn+2
Cn (t) = t̃rVn (tH ) = ∑ tn−2r = 2 , (7.21)
r=0 t −1

where convergence of the sum requires ∣t∣ > 1. Then for the irreducible representation defined
on the quotient Vn , by virtue of (7.19), the character is

tn+2 − t−n tn+1 − t−n−1

χn (t) = trVn (tH ) = Cn (t) − C−n−2 (t) = = . (7.22)
t2 − 1 t − t−1
1
This is just the same as (3.132) with t → ei 2 θ and n → 2j. It is easy to see that

χn (1) = dim Vn = n + 1 . (7.23)

Although the irreducible representation of su(2) are labelled by n ∈ N0 the characters may
be extended to any integer n with the property

χn (t) = −χ−n−2 (t) , (7.24)

as follows directly from (7.22). Clearly χ−1 (t) = 0.

181
 The su(2) Casimir operator in this basis

C = E+ E− + E− E+ + 21 H 2 = 2E− E+ + 12 H 2 + H , (7.25)

and it is easy to see that

C∣n⟩hw = cn ∣n⟩hw for cn = 12 n(n + 2) . (7.26)

Note that c−n−2 = cn as required from (7.14) as all vectors belonging to Vn must have the
same eigenvalue for C.

7.2 A su(3) Lie algebra basis and its automorphisms

We consider a basis for the su(3) Lie algebra in terms of 3 × 3 matrices as in (6.2). Thus
we define
⎛0 1 0⎞ ⎛0 0 0⎞ ⎛0 0 1⎞
e1+ = ⎜0 0 0⎟ , e2+ = ⎜0 0 1⎟ , e3+ = ⎜0 0 0⎟ , (7.27)
⎝0 0 0⎠ ⎝0 0 0⎠ ⎝0 0 0⎠
and their conjugates
ei− = ei+† , i = 1, 2, 3 , (7.28)
together with the hermitian traceless diagonal matrices

⎛1 0 0⎞ ⎛0 0 0 ⎞
h1 = ⎜0 −1 0⎟ , h2 = ⎜0 1 0 ⎟ . (7.29)
⎝0 0 0⎠ ⎝0 0 −1⎠

The commutator algebra satisfied by {e1± , e2± , e3± , h1 , h2 } is invariant under simultaneous
permutations of the rows and columns of each matrix. For b corresponding to the permu-
tation (1 2) and a to the cyclic permutation (1 2 3)

⎛0 1 0⎞ ⎛0 0 1⎞
b = ⎜1 0 0⎟ , a = ⎜1 0 0⎟ , (7.30)
⎝0 0 1⎠ ⎝0 1 0⎠

then

b{h1 , h2 }b−1 = {−h1 , h1 + h2 } , b{e1± , e2± , e3± }b−1 = {e1∓ , e3± , e2± } ,
a{h1 , h2 }a−1 = {h2 , −h1 − h2 } , a{e1± , e2± , e3± }a−1 = {e2± , e3∓ , e1∓ } . (7.31)

The matrices in (7.30) satisfy

b2 = 1 , a3 = 1 , ab = ba2 , (7.32)

so that they generate the permutation group S3 = {e, a, a2 , b, ab, a2 b}.

For representations of su(3) it is then sufficient to require operators
1
⎛ 3 (2H1 + H2 ) E1+ E3+ ⎞
i 1
{E1± , E2± , E3± , H1 , H2 } → [R̂ j ] = ⎜ E1− 3 (−H1 + H2 ) E 2+ ⎟, (7.33)
⎝ E3− E2− − 13 (H1 + 2H2 )⎠

182
acting on a vector space, and satisfying the same commutation relations as the corresponding
matrices {e1± , e2± , e3± , h1 , h2 }. The commutation relations may be summarised in terms of
R̂i j by
[R̂i j , R̂k l ] = δ kj R̂i l − δ i l R̂k j , (7.34)
since, for X, Y appropriate matrices, (7.34) requires
[tr(X R̂), tr(Y R̂)] = tr([X, Y ]R̂) , (7.35)
and with the definitions (7.27) and (7.29) we have, from (7.33), tr(ei± R̂) = Ei± , i = 1, 2, 3
and tr(hi R̂) = Hi , i = 1, 2.
Just as with su(2) the possible irreducible representation spaces may be determined
algebraically from the commutation relations of the operators in the privileged basis given
in (7.33). Crucially there are two commuting generators H1 , H2 so that
[H1 , H2 ] = 0 . (7.36)
For Ei+ the commutation relations are
[E1+ , E2+ ] = E3+ , [E1+ , E3+ ] = [E2+ , E3+ ] = 0 . (7.37)
while under commutation with H1 , H2
[H1 , {E1± , E2± , E3± }] = ± {2E1± , −E2± , E3± } ,
[H2 , {E1± , E2± , E3± }] = ± {−E1± , 2E2± , E3± } . (7.38)
The remaining commutators involving Ei± are
[E1+ , E1− ] = H1 , [E1+ , E2− ] = 0 , [E2+ , E2− ] = H2 ,
[E3+ , E1− ] = − E2+ , [E3+ , E2− ] = E1+ , [E3+ , E3− ] = H1 + H2 , (7.39)
together with those obtained by conjugation, [X, Y ]† = −[X † , Y † ], where Ei±† = Ei∓ and
Hi† = Hi .
The su(3) Lie algebra basis in (7.33) can be decomposed into three su(2) Lie algebras,
l1 = {E1+ , E1− , H1 } , l2 = {E2+ , E2− , H2 } , l3 = {E3+ , E3− , H1 + H2 } , (7.40)
where each li satisfies (7.5). From (7.31) the automorphism symmetries of the privileged
basis in (7.33) are generated by
l1 → l1R , l2 → l3 , l3 → l2 , l1 → l2 , l2 → l3R , l3 → l1R , (7.41)
b b b a a a

with the reflected su(2) Lie algebra defined by (7.6). The corresponding Weyl group, defined
in terms of transformations a, b satisfying (7.32), W (su(3)) ≃ S3 .
If we define
H⊥ = √1 (H1 + 2H2 ) , (7.42)
3
then the automorphism symmetries become
√ √
3 3
(H1 , H⊥ ) → (−H1 , H⊥ ) , (H1 , H⊥ ) → (− 21 H1 + 2 H⊥ , − 2 H1 − 12 H⊥ ) . (7.43)
b a

Regarding H1 , H⊥ as corresponding to Cartesian x, y coordinates then b represents a reflec-

tion in the y-axis and a a rotation through 2π/3.

183
7.3 Highest Weight Representations for su(3)

H1 , H2 commute, (7.36), and a standard basis for the representation space for su(3) is given
by their simultaneous eigenvectors ∣r1 , r2 ⟩ where

H1 ∣r1 , r2 ⟩ = r1 ∣r1 , r2 ⟩ , H2 ∣r1 , r2 ⟩ = r2 ∣r1 , r2 ⟩ . (7.44)

As a consequence of (7.38) we must then have

E1± ∣r1 , r2 ⟩ ∝ ∣r1 ± 2, r2 ∓ 1⟩ ,

E2± ∣r1 , r2 ⟩ ∝ ∣r1 ∓ 1, r2 ± 2⟩ ,
E3± ∣r1 , r2 ⟩ ∝ ∣r1 ± 1, r2 ± 1⟩ , (7.45)

unless Ei+ and/or Ei− annihilate ∣r1 , r2 ⟩ for one or more individual i. The set of values
[r1 , r2 ], linked by (7.45), are the weights of the representation. The may be plotted on a
triangular lattice with r1 along the x-axis and √1 (r1 + 2r2 ) along the y-axis.
3

E2+
E3+

E1− E1+
E2−
E3−

For any element σ ∈ W (su(3)) there is an associated action on the weights for su(3),
σ[r1 , r2 ], such that

Hi → H ′i , H ′i ∣r1 , r2 ⟩ = r′i ∣r1 , r2 ⟩ , i = 1, 2 ⇒ [r′1 , r′2 ] = σ[r1 , r2 ] . (7.46)

σ

From (7.41) this is given by

b[r1 , r2 ] = [−r1 , r1 + r2 ] , ab[r1 , r2 ] = [r1 + r2 , −r2 ] , a2 b[r1 , r2 ] = [−r2 , −r1 ] ,

a[r1 , r2 ] = [r2 , −r1 − r2 ] , a2 [r1 , r2 ] = [−r1 − r2 , r1 ] . (7.47)

As will become apparent the set of weights for any representation is invariant under the
action of the Weyl group, thus su(3) weight diagrams are invariant under rotations by 2π/3
and reflections in the y-axis.

184
 For a highest weight representation there is a unique vector ∣n1 , n2 ⟩hw , such that for all
other weights r1 + r2 < n1 + n2 . [n1 , n2 ] is the highest weight and we must then have

E1+ ∣n1 , n2 ⟩hw = E2+ ∣n1 , n2 ⟩hw = 0 ⇒ E3+ ∣n1 , n2 ⟩hw = 0 . (7.48)

The corresponding representation space V[n1 ,n2 ] is formed by the action of arbitrary products
of the lowering operators Ei− , i = 1, 2, 3 on the highest weight vector and may be defined by

V[n1 ,n2 ] = span {E3−t E2−s E1−r ∣n1 , n2 ⟩hw ∶ r, s, t = 0, 1, . . . } . (7.49)

The ordering of E1− , E2− , E3− in the basis assumed in (7.49) reflects an arbitrary choice,
any polynomial in E1− , E2− , E3− acting on ∣n1 , n2 ⟩ may be expressed uniquely in terms of
the chosen basis in (7.49) using the commutation relations given by the conjugate of (7.37).
For these basis vectors

H1 E3−t E2−s E1−r ∣n1 , n2 ⟩hw = (n1 − 2r + s − t) E3−t E2−s E1−r ∣n1 , n2 ⟩hw ,
H2 E3−t E2−s E1−r ∣n1 , n2 ⟩hw = (n2 + r − 2s − t) E3−t E2−s E1−r ∣n1 , n2 ⟩hw , (7.50)

so that the weights of vectors belonging to V[n1 ,n2 ] are those belonging to a 2π/3 segment
in the weight diagram with vertex at [n1 , n2 ], as shown by the shaded region in the figure
below.

[n1 ,n2 ]

The representation of su(3) is determined then in terms of the action of Ei± on the basis
(7.49). For the lowering operators it is easy to see that

E3− E3−t E2−s E1−r ∣n1 , n2 ⟩hw = E3−t+1 E2−s E1−r ∣n1 , n2 ⟩hw ,
E2− E3−t E2−s E1−r ∣n1 , n2 ⟩hw = E3−t E2−s+1 E1−r ∣n1 , n2 ⟩hw ,
E1− E3−t E2−s E1−r ∣n1 , n2 ⟩hw = E3−t E2−s E1−r+1 ∣n1 , n2 ⟩hw
− s E3−t+1 E2−s−1 E1−r ∣n1 , n2 ⟩hw , (7.51)

using [E1− , E2−s ] = −s E3− E2−s−1 .

The action of Ei+ on the basis (7.49) may then be determined by using the basic com-
mutation relations (7.39), with (7.38) and (7.37), and then applying (7.48). Just as in (7.12)
we may obtain

[E1+ , E1−r ] = E1−r−1 r(H1 − r + 1) , [E1+ , E2−s ] = 0 , [E1+ , E3−t ] = −t E3−t−1 E2− , (7.52)

185
so that

E1+ E3−t E2−s E1−r ∣n1 , n2 ⟩hw

= r(n1 − r + 1) E3−t E2−s E1−r−1 ∣n1 , n2 ⟩hw − t E3−t−1 E2−s+1 E1−r ∣n1 , n2 ⟩hw . (7.53)

Similarly

[E2+ , E3−t ] = t E3−t−1 E1− , [E1− , E2−s ] = −s E3− E2−s−1 ,

[E2+ , E2−s ] = E2−s−1 s(H2 − s + 1) , [E2+ , E1−r ] = 0 , (7.54)

which leads to

E2+ E3−t E2−s E1−r ∣n1 , n2 ⟩hw

= s(n2 + r − s − t + 1) E3−t E2−s−1 E1−r ∣n1 , n2 ⟩hw + t E3−t−1 E2−s E1−r+1 ∣n1 , n2 ⟩hw . (7.55)

Furthermore

E3+ E3−t E2−s E1−r ∣n1 , n2 ⟩hw = [E1+ , E2+ ] E3−t E2−s E1−r ∣n1 , n2 ⟩hw
= t(n1 + n2 − r − s − t + 1) E3−t−1 E2−s E1−r ∣n1 , n2 ⟩hw
+ rs(n1 − r + 1) E3−t E2−s−1 E1−r−1 ∣n1 , n2 ⟩hw . (7.56)

The results (7.50), (7.51) with (7.53), (7.55) and (7.56) demonstrate how V[n1 ,n2 ] forms
a representation space for su(3) which is in general infinite dimensional.
The space V[n1 ,n2 ] defines a reducible representation of su(3) when it contains vectors
which satisfy the highest weight condition (7.48) since these generate invariant subspaces.
Highest weight vectors may be constructed in V[n1 ,n2 ] in a similar fashion to the discussion
for su(2). Using, as a special case of (7.53) and (7.55),

E1+ E1−r ∣n1 , n2 ⟩hw = r(n1 − r + 1)E1−r−1 ∣n1 , n2 ⟩hw , E2+ E1−r ∣n1 , n2 ⟩hw = 0 ,
E2+ E2−s ∣n1 , n2 ⟩hw = s(n1 − s + 1)E2−s−1 ∣n1 , n2 ⟩hw , E1+ E2−s ∣n1 , n2 ⟩hw = 0 , (7.57)

then for n1 , n2 positive integers

∣−n1 − 2, n1 + n2 + 1⟩hw = E1−n1 +1 ∣n1 , n2 ⟩hw ,

∣n1 + n2 + 1, −n2 − 2⟩hw = E2−n2 +1 ∣n1 , n2 ⟩hw , (7.58)

satisfy the necessary conditions (7.48). Using the highest weight vectors obtained in (7.58)
we may further obtain, for n1 , n2 positive integers, two more highest weight vectors

∣n2 , −n1 − n2 − 3⟩hw = E2−n2 +n1 +2 ∣−n1 − 2, n1 + n2 + 1⟩hw ,

∣−n1 − n2 − 3, n1 ⟩hw = E1−n2 +n1 +2 ∣n1 + n2 + 1, −n2 − 2⟩hw . (7.59)

This construction may also be applied to the highest weight vectors in (7.59) giving one
further highest weight vector

∣−n2 − 2, −n1 − 2⟩hw = E1−n2 +1 ∣n2 , −n1 − n2 − 3⟩hw = E2−n1 +1 ∣−n1 − n2 − 3, n1 ⟩hw . (7.60)

186
The highest weight vectors determined in (7.58), (7.59), (7.60) are non degenerate, in (7.60)
this depends on the identity50

E1−n2 +1 E2−n1 +n2 +2 E1−n1 +1 = E2−n1 +1 E1−n1 +n2 +2 E2−n2 +1 . (7.61)

Clearly this construction of highest weight vectors terminates with (7.60).

For each of highest weight vectors given in (7.58), (7.59) and (7.60), ∣n′1 , n′2 ⟩hw , there
are associated invariant, under the action of su(3), subspaces V[n′1 ,n′2 ] , constructed as in
(7.49), and contained in V[n1 ,n2 ] . In particular

V[−n1 −2, n1 +n2 +1] , V[n1 +n2 +1,−n2 −2] ⊂ V[n1 ,n2 ] ,
V[n2 ,−n1 −n2 −3] ⊂ V[−n1 −2, n1 +n2 +1] , V[−n1 −n2 −3, n1 ] ⊂ V[n1 +n2 +1,−n2 −2] ,
V[−n2 −2,−n1 −2] ⊂ V[n2 ,−n1 −n2 −3] ∩ V[−n1 −n2 −3, n1 ] . (7.62)

The highest weight vectors which are present are illustrated on the weight diagram below,
with the shaded regions indicating where the associated invariant subspaces are present.

[−n1 −2,n1 +n2 +1] l=0

[−n1 ,n1 +n2 ] [n 1 ,n 2]

k=l+n 1 k=0

[−n1 −n2 ,n1 ] [n 1 +n 2,−n2 ]

[−n1 −n2 −3,n1 ] [n 1 +n 2+1,−n1 −2]

k=n 1 +n 2 k=l+n 2

[−n2 ,−n1 ]
[n 2 ,−n1 −n2 ]
l=n 1 +n 2

[−n2 −2,−n1 −2] [n 2,−n1 −n2 −3]

The reduction of V[n1 ,n2 ] to an irreducible representation space becomes less trivial than
that given by (7.16) for su(2) due to this nested structure of invariant subspaces. Using the
This may be shown using the identity E1−r E2−s = ∑r,s t=0 (−1) ( t ) (s−t)! E3− E2−
50 t r t s−t
s!
E1−r−t , both sides of
(7.61) give rise to the same expansion in E3−t E2−n1 +n2 +2−t E1−n1 +n2 +2−t .

187
same definition of the quotient of a vector space by a subspace as in (7.16) we may define

V (2)
[n1 ,n2 ] = (V[n2 ,−n1 −n2 −3] ⊕ V[−n1 −n2 −3, n1 ] )/ V[−n2 −2,−n1 −2] ,

V (1) (2)
[n1 ,n2 ] = (V[−n1 −2, n1 +n2 +1] ⊕ V[n1 +n2 +1,−n2 −2] )/ V [n1 ,n2 ] ,

V[n1 ,n2 ] = V[n1 ,n2 ] / V (1)

[n1 ,n2 ] . (7.63)

In V[n1 ,n2 ] there then are no highest weight vectors other than ∣n1 , n2 ⟩hw so invariant sub-
spaces are absent and V[n1 ,n2 ] is a representation space for an irreducible representation
of su(3). Although it remains to be demonstrated the representation space is then finite-
dimensional and the corresponding weight diagram has vertices with weights

[n1 , n2 ] , [−n1 , n1 + n2 ] , [n1 + n2 , −n2 ] , [−n1 − n2 , n1 ] , [n2 , −n1 − n2 ] ,

[−n2 , −n1 ] ,
(7.64)
which are related by the transformations of the Weyl group as in (7.47). The sector of the
weight diagram corresponding to highest weight states forming finite dimensional represen-
tations is then
W = {[m, n] ∶ m, n ∈ N0 } , (7.65)
which is illustrated by

7.3.1 Analysis of the Weight Diagram

It is clear that the construction (7.49) for V[n1 ,n2 ] requires that in general the allowed weights
are degenerate, i.e. there are multiple vectors for each allowed weight in the representation
space V[n1 ,n2 ] except on the boundary. For a particular weight [r1 , r2 ], (7.49) there is a
finite dimensional subspace contained in V[n1 ,n2 ] given by

(k,l)
V[n1 ,n2 ] = span {E3−t E2−l−t E1−k−t ∣n1 , n2 ⟩hw ∶ 0 ≤ t ≤ k, l} , (7.66)

where
k = 31 (2n1 + n2 − 2r1 − r2 ) , l = 13 (n1 + 2n2 − r1 − 2r2 ) . (7.67)
Clearly
⎧
(k,l) ⎪k + 1 ,
⎪ k ≤ l,
dim V[n1 ,n2 ] =⎨ (7.68)
⎩l + 1, l ≤ k.
⎪
⎪

188
 To show how (7.63) leads to a finite-dimensional representation we consider how it
applies to for the vectors corresponding to particular individual weights [r1 , r2 ]. In a similar
fashion to (7.66) we may define
(k−n −1,l)
1 (k,l−n −1)
V[−n1 −2,n1 +n2 +1]
, V[n1 +n22+1,−n2 −2] ,
(k−n −1,l−n −n2 −2) (k−n −n −2,l−n2 −1) (k−n −n −2,l−n1 −n2 −2)
V[n2 ,−n1 1 −n2 −3]
1
, 1
V[−n1 −n 2
2 −3,n1 ]
, 1 2
V[−n2 −2,−n 1 −2]
, (7.69)

which form nested subspaces, just as in (7.62), and whose dimensions are given by the
obvious extension of (7.68).
To illustrate how the construction of the representation space V[n1 ,n2 ] in terms of quo-
tient spaces leads to cancellations outside a finite region of the weight diagram we describe
how this is effected in particular regions of the weight diagram by showing that the di-
mensions of the quotient spaces outside the finite region of the weight diagram specified
by vertices in (7.64) are zero and also that on the boundary the dimension is one. For
(k,l)
k ≤ n1 , l ≤ n2 there are no cancellations for V[n1 ,n2 ] . Taking into account the contributions
(k−n −1,l)
1 (k,l−n −1)
from V[−n1 −2,n 1 +n2 +1]
and V[n1 +n22+1,−n2 −2] gives

⎧
(k,l) (k−n1 −1,l) ⎪0
⎪ if k ≥ l + n1 + 1, l ≥ 0 ,
dim V[n1 ,n2 ] − dim V[−n1 −2,n +n +1]
= ⎨ (7.70)
⎩1 if k = l + n1 , l ≥ 0 ,
1 2 ⎪
⎪

and
⎧
(k,l) (k,l−n −1) ⎪0 if l ≥ k + n2 + 1, k ≥ 0 ,
⎪
dim V[n1 ,n2 ] − dim V[n1 +n22+1,−n2 −2] =⎨ (7.71)
⎩1 if l = k + n2 , k ≥ 0 .
⎪
⎪
Furthermore
(k,l) (k−n −1,l)
1 (k,l−n −1)
dim V[n1 ,n2 ] − dim V[−n1 −2,n 1 +n2 +1]
− dim V[n1 +n22+1,−n2 −2]
⎧
⎪l + 1 − n2 − (l − n2 ) = 1 , k = n1 + n2 , n2 ≤ l ≤ n1 + n2 ,
⎪
=⎨ (7.72)
⎩k + 1 − (k − n1 ) − n1 = 1 , l = n1 + n2 , n1 ≤ k ≤ n1 + n2 .
⎪
⎪

The remaining contributions, when present, give rise to a complete cancellation so that the
representation space given by (7.63) is finite dimensional. When l ≥ n2 , k ≥ n1 + n2 + 1,
(k,l) (k−n −1,l)
1 (k,l−n −1) (k−n −n −2,l−n2 −1)
dim V[n1 ,n2 ] − dim V[−n1 −2,n 1 +n2 +1]
− dim V[n1 +n22+1,−n2 −2] + dim V[−n1 −n
1 2
2 −3,n1 ]
⎧
⎪(l + 1) − (l + 1) − (l − n2 ) + (l − n2 ) ,
⎪ k ≥ l + n1 + 1
=⎨
⎩(l + 1) − (k − n1 ) − (l − n2 ) + (k − n1 − n2 − 1) , l ≤ k ≤ l + n1 + 1
⎪
⎪
= 0, (7.73)

and for k, l ≥ n1 + n2 + 1 in an analogous fashion

(k,l) (k−n −1,l)
1 (k,l−n −1) (k−n −1,l−n −n2 −2)
dim V[n1 ,n2 ] − dim V[−n1 −2,n 1 +n2 +1]
− dim V[n1 +n22+1,−n2 −2] + dim V[n2 ,−n1 1 −n2 −3]
1

(k−n −n −2,l−n2 −1)

1 2 (k−n −n −2,l−n1 −n2 −2)
1 2
+ dim V[−n1 −n 2 −3,n1 ]
− dim V[−n2 −2,−n 1 −2]
= 0. (7.74)

189
 For the finite representation space V[n1 ,n2 ] then at each vertex of the weight diagram
as in (7.64) there are associated vectors which satisfy analogous conditions to (7.48), in
particular

(E1− , E3+ ) ∣−n1 , n1 + n2 ⟩ = 0 , (E2− , E3+ ) ∣n1 + n2 , −n2 ⟩ ,

(E2+ , E3− ) ∣−n1 − n2 , n1 ⟩ = 0 , (E1+ , E3− ) ∣n2 , −n1 − n2 ⟩ = 0 ,
(E1− , E2− ) ∣−n2 , −n1 ⟩ = 0 . (7.75)

Each vector may be use to construct the representation space by acting on it with ap-
propriate lowering operators. In this fashion V[n1 ,n2 ] may be shown to be invariant under
W (su(3)).
A generic weight diagram has the structure shown below. The multiplicity for each
weight is the same on each layer. For n1 ≥ n2 there are n2 + 1 six-sided layers and then the
layers become triangular. For the six-sided layers the multiplicity increases by one as one
moves from the outside to the inside, the triangular layers all have multiplicity n2 + 1. In
the diagram different colours have the same multiplicity.

00000000
11111111 00000000
11111111
n1
11111111111111111111111111111
00000000000000000000000000000 00000000
11111111
00000000
11111111 11111111
00000000
00000000
11111111
0000
111100001111
111100001111
111100001111
111100000000
11110000
1111 00000000
11111111
n2
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
0000
1111
0000
1111
0000
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00000000
11111111
00000000
11111111
0000
1111
0000
1111
00001111
1111
111
000
00001111
000
111
00001111
000
111
0000
000
111
0000
1111
000
111
0000
1111
000
111
0000
1111 00000000
11111111
0000
1111
0000
1111
0000
1111 000
111 000
111 000
111 000
111 000
111
0000
1111
000
111
0000
1111
0000
1111 00000000
11111111
00000000
11111111
0000
1111
0000
1111 000
111 000
111
0000
1111 000
111
0000
1111 000
111
0000
1111 000
111
0000
1111 000
111
0000
1111 0000
1111
0000
1111
00000000
11111111
0000
1111
0000
1111 000
1110000
11110000
11110000
11110000
11110000
1111 000
111
0000
1111
0000
1111
00000000
11111111
0000
1111
0000
1111 000
1110000
1111
0001111
111 0000
11110000
11110000
11110000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111 0000
000
111 0000
1111
0000
1111 0000
1111
0000
1111 0000
1111
0000
1111 0000 111
1111
0000
1111 000
0000
1111
0000
1111
0000
1111
0000
1111
000
1110000
11110000
11110000
11110000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
0001111
11100000000
11110000
11110000
1111 000
111
0000
1111
0000
1111
000
111
0000
1111
0000
1111
0000
1111 0000
1111
000
111 0000
1111
0000
1111 0000
1111
0000
1111 0000
1111
0000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
000
1110000
11110000
11110000
1111 0000
1111
000
111
0000
1111
0000
1111 0000
1111
0001111
111 0000
11110000
1111 0000
1111
0000
1111
0000
1111 0000
000
111 0000
1111
0000
1111 0000 111
1111
0000
1111 000
0000
1111
0000
1111
0000
1111
0000
1111
000
1110000
11110000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111 0000
1111
0001111
111 0000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111 00000000
1111
0000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111 000
111
000
1110000
1111 000
111
0000
1111
0000
1111
000
111
0000
1111
0000
1111
0000
1111
0000
1111
0001111
111
0000
1111
0000
1111
0000 111
000
0000
1111
0000
1111
0000
1111
0000
1111 000
111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
000
111 000
111
0000
1111
0000
1111
000 111
111 000
0000
1111
0000
1111
0000
1111
0000
11110000
11110000
1111
00001111
11110000
00001111
11110000
0000
1111
00001111
1111
0000
111100001111
1111
00000000
1111
0000
00001111
111100001111
0000

7.3.2 SU (3) Characters

A much more straightforward procedure for showing how finite dimensional representations
of SU (3) are formed is to construct their characters following the approach described for
SU (2) based on (7.21) and (7.22). For the highest weight representation space V[n1 ,n2 ] we

190
then define in terms of the basis (7.49)

C[n1 ,n2 ] (t1 , t2 ) = t̃rV[n1 ,n2 ] (t1H1 t2H2 ) = ∑ t1n1 −2r+s−t t2n2 +r−2s−t
r,s,t≥0
r s t
= t1n1 t2n2 ∑ (t2 /t12 ) (t1 /t22 ) (1/t1 t2 ) . (7.76)
r,s,t≥0

For a succinct final expression it is more convenient to use the variables

u = (u1 , u2 , u3 ) , u1 = t1 , u3 = 1/t2 , u1 u2 u3 = 1 , (7.77)

so that t2 /t12 = u2 /u1 , t1 /t22 = u3 /u2 , 1/t1 t2 = u3 /u1 and convergence of the sum requires
u1 > u2 > u3 . Then
u1n1 +n2 +2 u2n2 +1
C[n1 ,n2 ] (u) = . (7.78)
(u1 − u2 )(u2 − u3 )(u1 − u3 )
Following (7.63) the character for the irreducible representation of su(3) obtained from the
highest weight vector ∣n1 , n2 ⟩hw is then

χ[n1 ,n2 ] (u) = C[n1 ,n2 ] (u) − C[−n1 −2,n1 +n2 +1] (u) − C[n1 +n2 +1,−n2 −2] (u)
+ C[−n1 −n2 −3,n1 ] (u) + C[n2 ,−n1 −n2 −3] (u) − C[−n2 −2,−n1 −2] (u)
1
=
(u1 − u2 )(u2 − u3 )(u1 − u3 )
× (u1n1 +n2 +2 u2n2 +1 − u2n1 +n2 +2 u1n2 +1 − u1n1 +n2 +2 u3n2 +1
+ u2n1 +n2 +2 u3n2 +1 − u3n1 +n2 +2 u2n2 +1 + u3n1 +n2 +2 u1n2 +1 ) . (7.79)

It is easy to see that both the numerator and the denominator are completely antisymmetric
so that χ[n1 ,n2 ] (u) is a symmetric function of u1 , u2 , u3 , the S3 ≃ W (su(3)).
If we consider a particular restriction we get

1 − q n1 +1 1 − q n2 +1 1 − q −n1 −n2 −2
χ[n1 ,n2 ] (q, 1, q −1 ) = , (7.80)
1−q 1−q 1 − q −2
and hence it is then easy to calculate

dim V[n1 ,n2 ] = χ[n1 ,n2 ] (1, 1, 1) = 21 (n1 + 1)(n2 + 1)(n1 + n2 + 2) . (7.81)

The relation of characters to the Weyl group is made evident by defining, for any element
σ ∈ W (su(3)), a transformation on the weights such that

[r1 , r2 ]σ = σ[r1 + 1, r2 + 1] − [1, 1] . (7.82)

Directly from (7.47) we easily obtain

[r1 , r2 ]b = [−r1 − 2, r1 + r2 + 1] , [r1 , r2 ]ab = [r1 + r2 + 1, −r2 − 2] ,

2
[r1 , r2 ]a = [r2 , −r1 − r2 − 3] , [r1 , r2 ]a = [−r1 − r2 − 3, r1 ] ,
2b
[r1 , r2 ]a = [−r2 − 2, −r1 − 2] . (7.83)

191
Clearly [n1 , n2 ]σ generates the weights for the highest weight vectors contained in V[n1 ,n2 ] ,
as shown in (7.58), (7.59) and (7.60). Thus (7.79) may be written more concisely as

χ[n1 ,n2 ] (u) = ∑ Pσ C[n1 ,n2 ]σ (u) = ∑ C[n1 ,n2 ] (σu) , (7.84)
σ∈S3 σ∈S3

with, for σ ∈ S3 ,
⎧
⎪−1 , σ odd permutation ,
⎪
Pσ = ⎨ (7.85)
⎩1 , σ even permutation ,
⎪
⎪
and where σu denotes the corresponding permutation, so that b(u1 , u2 , u3 ) = (u2 , u1 , u3 ),
a(u1 , u2 , u3 ) = (u2 , u3 , u1 ). The definition of χ[n1 ,n2 ] (u) extends to any [n1 , n2 ] by taking

χ[n1 ,n2 ]σ (u) = Pσ χ[n1 ,n2 ] (u) . (7.86)

2b
Since [−1, r]b = [−1, r], [r, −1]ab = [r, −1] and [r, −r − 2]a = [r, −r − 2] we must then have

χ[−1,r] (u) = χ[r,−1] (u) = χ[r,−r−2] (u) = 0 . (7.87)

This shows the necessity of the three factors in the dimension formula (7.81). It is important
to note that for any [n1 , n2 ]

n1 , n2 ≠ −1, n1 + n2 ≠ −2 , [n1 , n2 ]σ ∈ W for a unique σ ∈ S3 , (7.88)

where W is defined in (7.65).

7.3.3 Casimir operator

For the basis in (7.33) the su(3) quadratic Casimir operator is given by

C = R̂i j R̂j i = ∑3i=1 (Ei+ Ei− + Ei− Ei+ ) + 23 (H12 + H22 + H1 H2 )

= ∑3i=1 Ei− Ei+ + 23 (H12 + H22 + H1 H2 ) + 2(H1 + H2 ) . (7.89)

Acting on a highest weight vector

C∣n1 , n2 ⟩hw = c[n1 ,n2 ] ∣n1 , n2 ⟩hw , (7.90)

where, from the explicit form in (7.89),

c[n1 ,n2 ] = 32 (n12 + n22 + n1 n2 ) + 2(n1 + n2 ) . (7.91)

It is an important check that c[n1 ,n2 ]σ = c[n1 ,n2 ] as required since C has the same eigenvalue
c[n1 ,n2 ] for all vectors belonging to V[n1 ,n2 ] .

192
7.3.4 Particular SU (3) Representations

We describe here how the general results for constructing a finite dimensional su(3) irre-
ducible representation spaces V[n1 ,n2 ] apply in some simple cases which are later of physical
relevance. The general construction in (7.63) ensures that the resulting weight diagram is
finite but in many cases the results can be obtained quite simply by considering the su(2)
subalgebras in (7.40) and then using results for su(2) representations.
The trivial singlet representation of course arises for n1 = n2 = 0 when there is unique
vector ∣0, 0⟩ annihilated by Ei± and Hi .
A particularly simple class of representations arises when n2 = 0. In this case applying
the su(2) representation condition (7.20) the highest weight vector must satisfy
E1−n1 +1 ∣n1 , 0⟩hw = 0 , E2− ∣n1 , 0⟩hw = 0 . (7.92)
Furthermore, using [E3+ , E1−r ] = −rE1−r−1 E2+ ,
E3+ E1−r ∣n1 , 0⟩hw = 0 , (H1 + H2 ) E1−r ∣n1 , 0⟩hw = (n1 − r) E1−r ∣n1 , 0⟩hw , (7.93)
so that E1−r ∣n1 , 0⟩hw is a su(2)i3 highest weight vector so that from (7.20) again
E3−n1 −r+1 E1−r ∣n1 , 0⟩hw = 0 . (7.94)
Hence a finite dimensional basis for V[n1 ,0] is given by

E3−t E1−r ∣n1 , 0⟩hw , t = 0, . . . n1 − r , r = 0, . . . , n1 , (7.95)

where there is a unique vector for each weight [n1 − 2r − t, r − t], which therefore has
multiplicity one. It is easy to check that this is in accord with the dimension of this
representation dim V[n1 ,0] = 21 (n1 + 1)(n1 + 2).
These representations have triangular weight diagrams as shown below.

193
 A corresponding case arises when n1 = 0 and the roles of E1− and E2− are interchanged.
In this case the basis vectors for V[0,n2 ] are just E3−t E2−s ∣0, n2 ⟩hw for t = 0, . . . n2 − s, s =
0, . . . , n2 and the weight diagram is also triangular.

In general the weight diagrams for V[n2 ,n1 ] may be obtained from that for V[n1 ,n2 ] by
rotation by π, these two representations are conjugate to each other.
The next simplest example arises for n1 = n2 = 1. The su(2) conditions (7.20) for the
highest weight state require

E1−2 ∣1, 1⟩hw = E2−2 ∣1, 1⟩hw = E3−3 ∣1, 1⟩hw = 0 . (7.96)

Since E1− ∣1, 1⟩hw is a highest weight vector for su(2)i2 and, together with E2− ∣1, 1⟩hw , is
also a su(2)i3 highest weight vector then the weights and associated vectors obtained from
∣1, 1⟩hw in terms of the basis (7.49) are then restricted to just

[−1, 2] ∶ E1− ∣1, 1⟩hw , [2, −1] ∶ E2− ∣1, 1⟩hw , [0, 0] ∶ E3− ∣1, 1⟩hw , E2− E1− ∣1, 1⟩hw ,
[−2, 1] ∶ E3− E1− ∣1, 1⟩hw , [1, 0] ∶ E3− E2− ∣1, 1⟩hw , E2−2 E1− ∣1, 1⟩hw ,
[−1, −1] ∶ E3−2 ∣1, 1⟩hw , E3− E2− E1− ∣1, 1⟩hw . (7.97)

However (7.96) requires further relations since

E2−2 E1− ∣1, 1⟩hw = (E2− E1− E2− + E3− E2− )∣1, 1⟩hw = 2 E3− E2− ∣1, 1⟩hw , (7.98)

which then entails

E1− E2−2 E1− ∣1, 1⟩hw = −2 E3− E2− E1− ∣1, 1⟩hw
= 2 E1− E3− E2− ∣1, 1⟩hw = 2(E3− E2− E1− − E3−2 )∣1, 1⟩hw , (7.99)

194
so that furthermore
E3−2 ∣1, 1⟩hw = 2 E3− E2− E1− ∣1, 1⟩hw . (7.100)
All weights therefore have multiplicity one except for [0, 0] which has multiplicity two. The
overall dimension is then 8 and V[1,1] corresponds to the SU (3) adjoint representation. The
associated weight diagram is just a regular hexagon, invariant under the dihedral group
D3 ≃ S3 , with the additional symmetry under rotation by π since this representation is
self-conjugate.

7.4 SU (3) Tensor Representations

Just as with the rotational group SO(3), and also with SU (2), representations may be
defined in terms of tensors. The representation space for a rank r tensor is defined by
the direct product of r copies of a fundamental representation space, formed by 3-vectors
for SO(3) and 2-spinors for SU (2), and so belongs to the r-fold direct product of the
fundamental representation. Such tensorial representations are reducible for any r ≥ 2 with
reducibility related to the existence of invariant tensors. Contraction of a tensor with an
invariant tensor may lead to a tensor of lower rank so that these form an invariant subspace
under the action of the group. Tensor representations become irreducible once conditions
have been imposed to ensure all relevant contractions with invariant tensors are zero.
For SU (N ) it is necessary to consider both the N -dimensional fundamental representa-
tion and its conjugate, SU (2) is a special case where these are equivalent. When N = 3 we
then consider a complex 3-vector q i and its conjugate q̄i = (q i )∗ , i = 1, 2, 3, belonging to the
vector space S and its conjugate S, ¯ and which transform as

q i → Ai j q j , q̄i → q̄j (A−1 )i j , [Aj i ] ∈ SU (3) . (7.101)

A (r, s)-tensor Tji11...j

...ir
s
is then one which belongs to S( ⊗ S)r−1 ( ⊗ S) ¯ s and which transforms
as
Tji11...j
...ir
s
→ Ai1 k1 . . . Air kr Tlk11...l
...kr
s
(A−1 )l1 j1 . . . (A−1 )ls js . (7.102)
The conjugate of a (r, s)-tensor is a (s, r)-tensor
∗
T̄ij11...i
...js
r
= (Tji11...j
...ir
s
) . (7.103)

195
 The invariant tensors are a natural extension of those for SU (2), as exhibited in (3.287)
and (3.288). Thus there are the 3-index antisymmetric ε-symbols, forming (3, 0) and (0, 3)-
tensors, and the Kronecker δ, which is a (1, 1)-tensor,

εijk , εijk , δji . (7.104)

That εijk and εijk are invariant tensors is a consequence of the transformation matrix A
satisfying det A = 1. The transformation rules (7.102) guarantee that the contraction of an
upper and lower index maintains the tensorial transformation properties. In consequence
from a tensor Tji11...j
...ir
s
then contracting with εijm jn or εjim in , for some arbitrary pair of
indices, generates a (r + 1, s − 2) or a (r − 2, s + 1)-tensor. Similarly using δ ij we may form
a (r − 1, s − 1)-tensor. Thus the vector space of arbitrary (r, s)-tensors contains invariant
subspaces, except for the fundamental (1, 0) or (0, 1) tensors or the trivial (0, 0) singlet.
Just as for SO(3) or SU (2) we may form an irreducible representation space by requiring all
such contractions give zero, so we restrict to (r, s)-tensors with all upper and lower indices
totally symmetric, and also traceless on contraction of any upper and lower index,
(i ...i )
Sji11 ...j
...ir
s
= S(j11 ...jrs ) , Sji11 ...j
...ir−1 i
s−1 i
= 0. (7.105)

The vector space formed by such symmetrised traceless tensors forms an irreducible
SU (3) representation space V[r,s] . To determine its dimension we may use the result in
(3.220) for the dimension of the space of symmetric tensors, with indices taking three
values, for n = r, s and then take account of the trace conditions by subtracting the results
for n = r − 1, s − 1. This gives

dim V[r,s] = 12 (r + 1)(r + 2) 12 (s + 1)(s + 2) − 12 r(r + 1) 12 s(s + 1)

= 12 (r + 1)(s + 1)(r + s + 2) . (7.106)

This is of course identical to (7.81). The irreducible representation space constructed in

terms of (r, s)-tensors is isomorphic with the finite dimensional irreducible space constructed
previously by analysis of the Lie algebra commutation relations.

7.4.1 su(3) Lie algebra again

For many applications involving SU (3) symmetry it is commonplace in physics papers to

use a basis of hermitian traceless 3 × 3 matrices, forming a basis for the su(3) Lie algebra,
which are a natural generalisation of the Pauli matrices in (3.19), the Gell-Mann λ-matrices
λa , a = 1, . . . , 8,

⎛0 1 0⎞ ⎛0 −i 0⎞ ⎛1 0 0⎞ 1 ⎛
1 0 0⎞
λ1 = ⎜1 0 0⎟ , λ2 = ⎜ i 0 0⎟ , λ3 = ⎜0 −1 0⎟ , λ 8 = √ ⎜0 1 0 ⎟ ,
⎝0 0 0⎠ ⎝0 0 0⎠ ⎝0 0 0⎠ 3 ⎝0 0 −2⎠

⎛0 0 1 ⎞ ⎛0 0 −i⎞ ⎛0 0 0⎞ ⎛0 0 0 ⎞
λ 4 = ⎜0 0 0 ⎟ , λ5 = ⎜0 0 0 ⎟ , λ6 = ⎜0 0 1⎟ , λ7 = ⎜0 0 −i⎟ . (7.107)
⎝1 0 0 ⎠ ⎝i 0 0 ⎠ ⎝0 1 0⎠ ⎝0 i 0 ⎠

196
These satisfy
tr(λa λb ) = 2 δab , (7.108)
and
[λa , λb ] = 2i fabc λc , (7.109)
for totally antisymmetric structure constants, fabc . In terms of the matrices defined in
(7.27) and (7.29) it is easy to see that e1+ = 12 (λ1 + iλ2 ), e2+ = 12 (λ6 + iλ7 ), e3+ = 12 (λ4 + iλ5 )
and also λ3 = h1 , λ8 = √1 (h1 + 2h2 ).
3

The relation between SU (3) matrices and the λ-matrices is in many similar to that for
SU (2) and the Pauli matrices, for an infinitesimal transformation the relation remains just
as in (3.38). (3.23) needs only straightforward modification while instead of (3.20) we now
have
λa λb = 32 1 + dabc λc + ifabc λc , (7.110)
with dabc totally symmetric and satisfying dabb = 0.

7.5 SU (3) and Physics

Besides its virtues in terms of understanding more general Lie groups a major motivation
in studying SU (3) is in terms of its role in physics. Historically SU (3) was introduced, as
a generalisation of the isospin SU (2)I , to be an approximate symmetry group for strong
interactions, in current terminology a flavour symmetry group, and the group in this con-
text is often denoted as SU (3)F . Unlike isospin, which was hypothesised to be an exact
symmetry for strong interactions, neglecting electromagnetic interactions, SU (3)F is intrin-
sically approximate. The main evidence is the classification of particles with the same spin,
parity into multiplets corresponding to SU (3) representations. For the experimentally ob-
served SU (3)F particle multiplets, unlike for isospin multiplets, the masses are significantly
different.
For SU (3)F the two commuting generators are identified with I3 , belonging to SU (2)I ,
and also the hypercharge Y , where [Ii , Y ] = 0 so that Y takes the same value for any
isospin multiplet. Y is related to strangeness S, a quantum number invented to explain
why the newly discovered, in the 1940’s, so-called strange particles were only produced in
pairs, the precise relation is Y = B + S, with B the baryon number. For any multiplet
we must have tr(I3 ) = tr(Y ) = 0. Expressed in terms of the su(3) operators H1 , H2 ,
I3 = H1 , Y = 13 (H1 + 2H2 ). For SU (3)F multiplets the electric charge is determined by
Q = I3 + 21 Y and so must be always conserved, but Y is not conserved by weak interactions
which are responsible for the decay of strange particles into non-strange particles.
For SU (3)F symmetry of strong interactions to be realised there must be 8 operators
satisfying the su(3) Lie algebra. If the same basis as for the λ-matrices in (7.107) is adopted
then these are Fa , a = 1, . . . , 8, where Fa are hermitian, and
[Fa , Fb ] = ifabc Fc , Fi = Ii , i = 1, 2, 3 , F8 = √1 Y . (7.111)
3

From a more modern perspective SU (3)F is understood to be a consequence of the

fact that low mass hadrons are composed of the three light quarks q = (u, d, s) and their

197
 ¯ s̄), corresponding to three quark flavours. These belong respectively
anti-particles q̄ = (ū, d,
to the fundamental [1, 0] and [0, 1] representations, more often denoted by 3 and 3∗ . On
a weight diagram these are the simplest triangular representations. With axes labelled by
I3 , Y these are

Y Y
2/3
s
d u
1/3

I3 I3
−1/2 1/2 −1/2 1/2

−1/3
u d
s
−2/3

The charges of quarks are dictated by the requirement Q = I3 + 12 Y and so for q are fractional,
2 1
3 and − 3 , while for q̄ they are the opposite sign. We may further interpret the quantum
numbers in terms of the numbers of particular quarks minus their anti-quarks, hence I3 =
Nu − Nū − Nd + Nd¯ and S = −Ns + Ns̄ , where each q has baryon number B = 31 and each q̄,
B = − 31 ,
As is well known isolated quarks are not observed, they are present as constituents of
the experimentally observed mesons, which are generally q q̄ composites, or baryons, whose
quantum numbers are consistent with a qqq structure. The associated representations have
zero triality, elements belonging to the centre Z(SU (3)) act trivially, or equivalently the
observed representations correspond to the group SU (3)/Z3 .
For the mesons we have self-conjugate octets belonging to the [1, 1], or 8, SU (3) repre-
sentations. The weight diagram for the lightest spin-0 negative parity mesons is

Mass(Mev) 0
Y +
495 K 1 K

1/2

137 π
−
π0 π+
I3
549 −1 −1/2
η 1/2
1

−1/2

− 0
495 K −1 K

Here the kaons K + , K 0 and K̄ 0 , K − are I = 1

2 strange particles with S = 1 and S = −1. A

198
similar pattern emerges for the next lightest spin one negative parity mesons.

Mass(Mev) *0 Y
892 K 1 K*+

1/2

ρ− ρ0 ρ+
770
I3
783 −1
−1/2 ω 1/2 1

−1/2

892 K* −1 K*0

The lightest multiplet of spin- 12 baryons is also an octet, with a similar weight diagram, the
same set of I3 , Y although of course different particle assignments.

Mass(Mev) Y
939 n 1 p

1/2
0
1193 Σ
− Σ Σ+
I3
Λ0
−1/2 1/2
1116 −1 1

−1/2

− 0
1318 Ξ −1 Ξ

The novelty for baryons is that there are also decuplets, corresponding to the [3, 0] and
[0, 3] representations, or labelled by their dimensionality 10 and 10∗ . The next lightest
spin- 23 baryons and their anti-particles belong to decuplets.

199
 Y
Mass (Mev)
∆− ∆0 ∆+ ∆++
1235 1

− 1/2 +
1385 Σ∗ Σ∗0 Σ∗
I3
−3/2 −1 −1/2 1/2 1 3/2
−1/2
−
Ξ∗ −1 Ξ∗0
1530
−3/2
−
1670 −2 Ω

Except for the Ω− the particles in the decuplet are resonances, found as peaks in the invariant
mass distribution for various cross sections. Since mΞ + mK > mΩ− the Ω− can decay only
via weak interactions and its lifetime is long enough to leave an observable track.

7.5.1 SU (3)F Symmetry Breaking

Assuming quark masses are not equal there are no exact flavour symmetries in strong
interactions, or equivalently QCD, save for a U (1) for each quark. Even isospin symmetry
is not exact since mu ≠ md . Restricting to the three light q = (u, d, s) quarks the relevant
QCD mass term may be written as
¯ − ms s̄s
Lm = − mu ūu − md dd
= − m̄ q̄q − 12 (mu − md ) q̄λ3 q − 1
√ (mu + md − 2ms ) q̄λ8 q , (7.112)
2 3

for m̄ = 13 (mu + md + ms ). If the difference between mu , md is neglected then the strong

interaction Hamiltonian must be of the form
H = H0 + T8 , (7.113)
where H0 is a SU (3) singlet and T8 is part of an octet of operators {Ta } so that, with the
SU (3) operators {Fa } as in (7.111), we have the commutation relations [Fa , H0 ] = 0 and
[Fa , Tb ] = ifabc Tc . The Hamiltonian in (7.113) is invariant under isospin symmetry since
[Ii , T8 ] = 0.
In any SU (3) multiplet the particle states may be labelled ∣II3 , Y ⟩ for various isospins
I and hypercharges Y , depending on the particular representation. For I3 = −I, −I + 1, . . . , I
the vectors ∣II3 , Y ⟩ form a standard basis under SU (2)I . With isospin symmetry the particle
masses are independent of I3 and to first order in SU (3) symmetry breaking
mI,Y = m0 + ⟨II3 , Y ∣T8 ∣II3 , Y ⟩ . (7.114)
It remains to determine a general expression for ⟨II3 , Y ∣T8 ∣II3 , Y ⟩, which is essentially
equivalent to finding the extension of the Wigner-Eckart theorem, described in section 3.13,
to SU (3).

200
 Instead of finding results for SU (3) Clebsch-Gordan coefficients the necessary calculation
may be accomplished, in this particular case, with less effort. It is necessary to recognise that
the crux of the Wigner-Eckart theorem is that, as far as the I, Y dependence is concerned,
⟨II3 , Y ∣T8 ∣II3 , Y ⟩ is determined just by the SU (3) transformation properties of T8 . Hence,
apart from overall undetermined constants, T8 may be replaced by any other operator with
the same transformation properties. For convenience we revert to a tensor basis for the
octet Ta → T i j , T i i = 0, and then with Fa → R̂i j as in (7.33),

[R̂i j , T k l ] = δ kj T i l − δ il T k j , T8 = 31 (T 1 1 + T 2 2 − 2T 3 3 ) . (7.115)

This ensures that T i j is a traceless (1, 1) irreducible tensor operator. Any such tensor
operator constructed in terms of R̂i j has the same SU (3) transformation properties. The
simplest case is if T i j = R̂i j when (7.115) requires

T8 = 31 (H1 + 2H2 ) = Y , (7.116)

with Y the hypercharge operator. An further independent (1, 1) operator is also given by
the quadratic expression T i j = 12 (R̂i k R̂k j + R̂k j R̂i k ) − 31 δ ij R̂k l R̂l k which then leads to

T8 = 41 (R̂1 k R̂k 1 + R̂k 1 R̂1 k + R̂2 k R̂k 2 + R̂k 2 R̂2 k − R̂3 k R̂k 3 − R̂k 3 R̂3 k ) − 61 C , (7.117)

where C is the SU (3) Casimir operator defined in (7.89). Using (7.33) then
2
T8 = 12 (E1+ E1− + E1− E1+ + 12 H12 ) − 1
36
(H1 + 2H2 ) − 16 C
= Ii Ii − 14 Y 2 − 16 C , (7.118)

with Ii the isospin operators and (7.25) has been used for the SU (2)I Casimir operator.
For a 3 × 3 traceless matrix R, R3 − 13 I tr(R3 ) = 12 R tr(R2 ) so that there are no further
independent cubic, or higher order, traceless (1, 1) tensor operators formed from R̂i j .
The results of the Wigner-Eckart theorem imply that, to calculate ⟨II3 , Y ∣T8 ∣II3 , Y ⟩,
it is sufficient to replace T8 by an arbitrary linear combination of (7.116) and (7.118).
Absorbing an I, Y independent constant into m0 and replacing the operators Ii Ii and Y by
their eigenvalues this gives the first order mass formula

mIY = m0 + aY + b(I(I + 1) − 41 Y 2 ) , (7.119)

with a, b undetermined coefficients.

For the baryon octet (7.119) gives 2(mN + mΞ ) = 3mΣ + mΛ , which is quite accurate.
For the decuplet the second term is proportional to the first so that the masses are linear
in Y , again in accord with experimental data. For mesons, for various reasons, the mass
formula is applied to m2 , so that 4mK
2
= 3mπ2 + mη2 .

7.5.2 SU (3) and Colour

The group SU (3) plays a more fundamental role, other than a flavour symmetry group,
as the gauge symmetry group of QCD. Each quark then belongs to the three dimensional

201
fundamental, 3 or [1, 0], representation space for SU (3)colour so that there is an addi-
tional colour index r = 1, 2, 3 and hence, for each of the six different flavours of quarks
q = u, d, s, c, b, t in the standard model, we have q r . The antiquarks belong to the conjugate,
3∗ or [0, 1], representation space, q̄r . The crucial assumption, yet to be fully demonstrated,
is that QCD is a confining theory, the states in the physical quantum mechanical space are
all colour singlets. No isolated quarks are then possible and this matches with the observed
mesons and baryons since the simplest colour singlets are just

q̄1r q2r , εrst q1r q2s q3t . (7.120)

Baryons are therefore totally antisymmetric in the colour indices. Fermi statistics then
requires that they should be symmetric under interchange with respect to all other variables,
spatial, spin and flavour. This provides non trivial constraints on the baryon spectrum which
match with experiment. The additional colour degrees of freedom also play a role in various
dynamical calculations, such as the total cross section for e− e+ scattering or π 0 → γγ decay.

7.6 Tensor Products for SU (3)

Just as for angular momentum it is essential to be able to decompose tensor products

of SU (3) representations into irreducible components in applications of SU (3) symmetry.
Only states belonging to the same irreducible representation will have the same physical
properties, except for dynamical accidents or a hidden addition symmetry.
For small dimensional representations it is simple to use the tensor formalism described
in section 7.4 with irreducible representations characterised by symmetric traceless tensors
as in (7.105). Thus for the product of two fundamental representations it is sufficient to
express it in terms of its symmetric and antisymmetric parts

q1i q2j = S ij + εijk q̄k , S ij = q1(i q2j) , q̄k = 12 εkij q1i q2j . (7.121)

while for the product of the fundamental and its conjugate it is only necessary to separate
out the trace
q̄i q j = Mij + δij S , Mij = q̄i q j − 13 δij q̄k q k , S = 31 q̄i q i . (7.122)
These correspond respectively to

3 ⊗ 3 = 6 ⊕ 3∗ , 3∗ ⊗ 3 = 8 ⊕ 1 . (7.123)

For the product of three fundamental representations then the decomposition may be ex-
pressed in terms of an irreducible (3, 0) tensor, two independent (1, 1) tenors and a singlet

q1i q2j q3k = Dijk + εikl Blj + εjkl Bli + εijl B ′kl + εijk S ,
Dijk = q1(i q2j q3k) , S = 16 εijk q1i q2j q3k ,
Bli = 31 εjkl q1(i q2j) q3k , B ′kl = 21 εijl q1i q2j q3k − δlk S . (7.124)

To verify that this is complete it is necessary to recognise, since the indices take only three
values, that
εijl Blk + εkil Blj + εjkl Bli = εijk Bll = 0 , (7.125)

202
for any Bji belonging to the 8 representation. (7.124) then corresponds to

3 ⊗ 3 ⊗ 3 = 10 ⊕ 8 ⊕ 8 ⊕ 1 . (7.126)

These of course are the baryon representations for SU (3)F .

In general it is only necessary to use the invariant tensors in (7.104) to reduce the tensor
products to irreducible tensors. Thus for the product of two octets the irreducible tensors
are constructed by forming first the symmetric (2, 2), (3, 0), (0, 3) tensors as well as two
(1, 1) tensors and also a singlet by
(i
Bji B ′kl → B(j B ′ l) , εjl(m Bji B ′ l , εik(m Bji B ′kl) , Bji B ′ jl , Bji B ′ ji , Bji B ′ ji .
k) k)
(7.127)

and then subtracting the required terms to cancel all traces formed by contracting upper
and lower indices, as in (7.122). This gives the decomposition

8 ⊗ 8 = 27 ⊕ 10 ⊕ 10∗ ⊕ 8 ⊕ 8 ⊕ 1 . (7.128)

7.6.1 Systematic Discussion of Tensor Products

For tensor products of arbitrary representations there is a general procedure which is quite
simple to apply in practice. The derivation of this is straightforward using characters to find
an algorithm for the expansion of the product of two characters for highest weight irreducible
representations as in (2.85). For su(3), characters are given by (7.79). In general these have
an expansion in terms of a sum over the weights in the associated weight diagram

χΛ (u) = ∑ nΛ, λ u1r1 +r2 +2 u2r2 +1 , Λ = [n1 , n2 ] , λ = [r1 , r2 ] , (7.129)

λ

where nΛ, λ is then the multiplicity in the representation space VΛ for vectors with weight
λ. Due to the symmetry of the weight diagram under the Weyl group we have

nΛ, λ = nΛ, σλ . (7.130)

Using (7.78) it is easy to see that

CΛ (u) χΛ′ (u) = ∑ nΛ′ , λ CΛ+λ (u) , (7.131)

λ

and since, for the weights {λ} corresponding to the representation with highest weight Λ,

{λ} = {σλ} , (Λ + λ)σ = Λσ + σλ , (7.132)

then, with (7.130), we may use (7.84) to obtain

χΛ (u) χΛ′ (u) = ∑ nΛ′ , λ χΛ+λ (u) . (7.133)

λ

However in general Λ + λ ∉ W, as defined in (7.65). In this case (7.86) may be used to

rewrite (7.133) as

χΛ (u) χΛ′ (u) = ∑ nΛ′ , λ Pσ χ(Λ+λ)σ (u) , (Λ + λ)σ ∈ W , (7.134)

λ

203
dropping all terms where Λ+λ satisfies any of the conditions in (7.87) ensuring χΛ+λ (u) = 0,
so that, by virtue of (7.88), σ in (7.134) is then unique. Since in (7.134) some terms may
now contribute with a negative sign there are then cancellations although the final result is
still a positive sum of characters.
The result (7.134) may be re-expressed in terms of the associated representation spaces.
For a highest weight Λ the representation space VΛ has a decomposition into subspaces for
each weight,
(λ) (λ)
VΛ = ⊕ VΛ , dim VΛ = nΛ, λ , (7.135)
λ

and then (7.134) is equivalent to

VΛ ⊗ VΛ′ ≃ ⊕ nΛ′ , λ Pσ V(Λ+λ)σ , (Λ + λ)σ ∈ W . (7.136)

λ

This implies the corresponding decomposition for the associated representations.

As applications we may consider tensor products involving V[1,0] which has the weight
decomposition
V[1,0] → [1, 0] , [−1, 1] , [0, −1] , (7.137)
and then

V[n1 ,n2 ] ⊗ V[1,0] ≃ V[n1 +1,n2 ] ⊕ V[n1 −1,n2 +1] ⊕ V[n1 ,n2 −1]
⎧
⎪ ⊕ V[0,n2 −1] , n1 = 0 ,
= ⎨ [1,n2 ]
⎪V
(7.138)
⎩V[n1 +1,0] ⊕ V[n1 −1,1] , n2 = 0 .
⎪
⎪

It is easy to see that this is in accord with the results in (7.126). For an octet

V[1,1] → [1, 1] , [2, −1] , [−1, 2] , [0, 0]2 , [1, −2] , [−2, 1] , [−1, −1] , (7.139)

so that, for n1 , n2 ≥ 2,

V[n1 ,n2 ] ⊗ V[1,1] ≃ V[n1 +1,n2 +1] ⊕ V[n1 +2,n2 −1] ⊕ V[n1 −1,n2 +2] ⊕ V[n1 ,n2 ]
⊕ V[n1 ,n2 ] ⊕ V[n1 +1,n2 −2] ⊕ V[n1 −2,n2 +1] ⊕ V[n1 −1,n2 −1] , (7.140)

with special cases

V[1,1] ⊗ V[1,1] ≃ V[2,2] ⊕ V[3,0] ⊕ V[0,3] ⊕ V[1,1] ⊕ V[1,1] ⊕ V[0,0] , (7.141)

which is in accord with (7.128), and

V[3,0] ⊗ V[1,1] ≃ V[4,1] ⊕ V[2,2] ⊕ V[3,0] ⊕ V[1,1] , (7.142)

using V[4,−2] ≃ −V[3,0] . Equivalently, labelling the representations by their dimensions

10 ⊗ 8 = 35 ⊕ 27 ⊕ 10 ⊕ 8 . (7.143)

204
8 Gauge Groups and Gauge Theories

Gauge theories are fundamental to our understanding of theoretical physics, many successful
theories such as superconductivity and general relativity are best understood in terms of
an appropriate gauge symmetry and its implementation. High energy particle physics is
based on quantum gauge field theories. A gauge theory is essentially one where there are
redundant degrees of freedom, which cannot in general be eliminated, at least without
violating other symmetries that are present. The presence of such superfluous degrees of
freedom requires a careful treatment when gauge theories are quantised and a quantum
vector space for physical states is constructed. If the basic variables in a gauge theory are
denoted by q then gauge transformations q → q g , for g ∈ G for some group G, are dynamical
symmetries which define an equivalence q ∼ q g . The objects of interest are then functions of
q which are invariant under G, in a physical theory these are the physical observables. For
a solution q(t) of the dynamical equations of motion then a gauge symmetry requires that
q g(t) (t) is also a solution for arbitrary continuously differentiable g(t) ∈ Gt ≃ G. For this
to be feasible G must be a Lie group, group multiplication is defined by g(t)g ′ (t) = gg ′ (t)
and the full group of gauge transformations is then essentially G ≃ ⊗t Gt . A gauge theory in
general requires the introduction of additional dynamical variables which form a connection,
depending on t, on MG and so belongs to the Lie algebra g.
For a relativistic gauge field theory there are vector gauge fields, with a Lorentz index
Aµ (x), belonging to g. Denoting the set of all vector fields, functions of x and taking values
in g, by A, we can then write
Aµ ∈ A . (8.1)
In a formal sense, the gauge group G is defined by

G ≃ ⊗ Gx , (8.2)
x

i.e. an element of G is a map from space-time points to elements of the Lie group G (the
definition of G becomes precise when space-time is approximated by a lattice). Gauge
transformations act on the gauge fields so that

Aµ (x) Ð→ Aµg(x) (x) ∼ Aµ (x) . (8.3)

g(x)

Gauge transformations g(x) are then the redundant variables and the physical space is
determined by the equivalence classes of gauge fields modulo gauge transformations or

A/G . (8.4)

If Aµ (x) is subject to suitable boundary conditions as ∣x∣ → ∞, or we restrict x ∈ M for

some compact M, then this is topologically non trivial.
The most significant examples of quantum gauge field theories are51 ,

Theory: QED Weinberg-Salam model QCD,

Gauge Group: U (1) SU (2) ⊗ U (1) SU (3).
51
Steven Weinberg, (1933-), American. Abdus Salam, (1926-1996), Pakistani. Nobel Prizes 1979.

205
Renormalisable gauge field theories are almost uniquely determined by specifying the gauge
group and then the representation content of any additional fields.

8.1 Abelian Gauge Theories

The simplest example arises for G = U (1), which is the gauge group for Maxwell52 electro-
magnetism, although the relevant gauge symmetry was only appreciated by the 1920’s and
later. For U (1) the group elements are complex numbers of modulus one, so they can be
expressed as eiα , 0 ≤ α < 2π. For a gauge theory the group transformations depend on x so
we can then write eiα(x) . The representations of U (1) are specified by q ∈ R, physically the
charge, so that for a complex field φ(x) the group transformations are

φ Ð→ eiqα φ = φ′ . (8.5)
eiα

If the field φ forms a non projective representation we must have

q ∈ Z = {0, ±1, ±2, . . . } . (8.6)

In quantum mechanics this is not necessary but if the U (1) is embedded in a semi-simple
Lie group then, with a suitable convention, q can be chosen to satisfy (8.6). For U (1) the
multiplication of representations is trivial, the charges just add, and also under complex
conjugation q → −q. It is then easy to construct lagrangians Lφ which are invariant under
(8.5) for global transformations, where α is independent of x. Restricting to first derivatives
this requires
Lφ (φ, ∂µ φ) = Lφ (φ′ , ∂µ φ′ ) , (8.7)
and an obvious solution, which defines a Lorentz invariant theory for complex scalars φ, is
then
Lφ (φ, ∂µ φ) = ∂ µ φ⋆ φµ φ − V (φ∗ φ) . (8.8)

For local transformations, when the elements of the gauge group depend on x, the initial
lagrangian is no longer invariant due to the presence of derivatives since

∂µ φ′ = eiqα (∂µ φ + iq∂µ α φ) , (8.9)

and the ∂µ α terms fail to cancel. This is remedied by introducing a connection, or gauge
field, Aµ and then defining a covariant derivative on φ by

Dµ φ = ∂µ φ − iqAµ φ . (8.10)

If under a local U (1) gauge transformation, as in (8.5), the gauge field transforms as

Aµ Ð→ Aµ + ∂µ α = A′µ , (8.11)
eiα

so that
D′µ φ′ = eiα Dµ φ , (8.12)
52
James Clerk Maxwell, 1831-79, Scottish, second wrangler 1854.

206
and then it is easy to see that, for any globally invariant lagrangian satisfying (8.7),

Lφ (φ, Dµ φ) = Lφ (φ′ , D′µ φ′ ) . (8.13)

It is important to note that for abelian gauge theories Aµ ∼ A′µ , which corresponds precisely
to the freedom of polarisation vectors in (4.227) when Lorentz vector fields are used for
massless particles with helicities ±1.
The initial scalar field theory then includes the gauge field Aµ , as well as the scalar
fields φ, both gauge dependent. For well defined dynamics the scalar lagrangian Lφ must
be extended to include an additional gauge invariant kinetic term for Aµ . In the abelian
case it is easy to see that the curvature

Fµν = ∂µ Aν − ∂ν Aµ = F ′µν , (8.14)

is gauge invariant, since ∂µ ∂ν α = ∂ν ∂µ α. In electromagnetism Fµν decomposes in to the

electric and magnetic fields and is related to the commutator of two covariant derivatives
since
[Dµ , Dν ]φ = −iqFµν φ . (8.15)
The simplest Lorentz invariant, gauge invariant, lagrangian is then
1 µν
L = Lgauge + Lφ (φ, Dµ φ) , Lgauge = − F Fµν , (8.16)
4e2
with e an arbitrary parameter, unimportant classically. It is commonplace to rescale the
fields so that
Aµ → eAµ , Dµ φ = ∂µ φ − ieqAµ φ , (8.17)
so that e disappears from the gauge field term in (8.16). The dynamical equations of motion
which flow from (8.16) are, for the gauge field,

1 µ ∂
2
∂ Fµν = jν = − ν Lφ (φ, Dµ φ) , (8.18)
e ∂A
which are of course Maxwell’s equations for an electric current jν and e becomes the basic
unit of electric charge. A necessary consistency condition is that the current is conserved
∂ ν jν = 0. In addition Fµν satisfies an identity, essentially the Bianchi identity, which follows
directly from its definition in (8.14),

∂ω Fµν + ∂ν Fωµ + ∂µ Fνω = 0 . (8.19)

1
In the language of forms, A = Aµ dxµ , F = 2 Fµν dxµ ∧ dxν = dA, this is equivalent to
dF = d2 A = 0.

8.2 Non Abelian Gauge Theories

In retrospect the generalisation of gauge theories to non abelian Lie groups is a natural
step. A fully consistent non abelian gauge theory was first described in 1954, for the group

207
SU (2), by Yang and Mills53 so they are often referred to, for the particular gauge invariant
lagrangian generalising the abelian lagrangian given in (8.16) and obtained below, as Yang-
Mills theories. Nevertheless the same theory was also developed, but not published, by
R. Shaw54 (it appeared as an appendix in his Cambridge PhD thesis submitted in 1955
although this work was done in early 1954). Such theories were not appreciated at first
since they appeared to contain unphysical massless particles, and also since understanding
their quantisation was not immediate.
Following the same discussion as in the abelian case we first consider fields φ belonging
to the representation space V for a Lie group G. Under a local group transformation then

φ(x) Ð→ g(x)φ(x) = φ′ (x) , (8.20)

g(x)

for g(x) ∈ R for R an appropriate representation, acting on V, of G. Manifestly derivatives

fail to transform in the same simple homogeneous fashion since

∂µ φ(x) Ð→ g(x)(∂µ φ(x) + g(x)−1 ∂µ g(x) φ(x)) = ∂µ φ′ (x) , (8.21)

g(x)

where g −1 ∂µ g belongs to the corresponding representation of the Lie algebra of G, g, which

is assumed to have a basis {ta } satisfying the Lie algebra (5.60). As before to define a
covariantly transforming derivative Dµ it is necessary to introduce a connection belonging
to this Lie algebra representation which may be expanded over the basis matrices ta ,

Aµ (x) = Aaµ (x) ta , (8.22)

and then
Dµ φ = (∂µ + Aµ )φ . (8.23)
Requiring
D′µ φ′ = gDµ φ , (8.24)
or
g −1 A′µ g + g −1 ∂µ g = Aµ , (8.25)
then the gauge field must transform under a gauge transformation as

Aµ Ð→ A′µ = gAµ g −1 − ∂µ g g −1 = gAµ g −1 + g∂µ g −1 . (8.26)

g

Hence if Lφ (φ, ∂µ φ) is invariant under global transformations φ → gφ then Lφ (φ, Dµ φ) is

invariant under the corresponding local transformations, so long as Aµ also transforms as
in (8.26).
It is also useful to note, since G is a Lie group, the associated infinitesimal transforma-
tions when
g = 1 + λ, λ = λa ta . (8.27)
53
Chen-Ning Franklin Yang, 1922-, Chinese then American, Nobel prize 1957. Robert L. Mills, 1927-99,
American.
54
Ron Shaw, 1929-2016, English.

208
Then from (8.20) and (8.24), for arbitrary λa (x),

δφ = λφ , δDµ φ = λDµ φ , (8.28)

and from (8.26)

δAµ = [λ, Aµ ] − ∂µ λ ⇒ δAaµ = −f abc Abµ λc − ∂µ λa . (8.29)

The associated curvature is obtained from the commutator of two covariant derivatives,
as in the abelian case in (8.15), which gives

[Dµ , Dν ]φ = Fµν φ , Fµν = F aµν ta , (8.30)

so that
Fµν = ∂µ Aν − ∂ν Aµ + [Aµ , Aν ] , (8.31)
or
F aµν = ∂µ Aaν − ∂ν Aaµ + f abc Abµ Acν . (8.32)
Unlike the abelian case, but more akin to general relativity, the curvature is no longer linear.
The same result is expressible more elegantly using differential form notation by

F = dA + A ∧ A , A = Aµ dxµ , A ∧ A = 21 [Aµ , Aν ] dxµ ∧ dxν . (8.33)

For a gauge transformation as in (8.26)

Fµν Ð→ F ′µν = gFµν g −1 , (8.34)

g

or, infinitesimally,

δFµν = [λ, Fµν ] ⇒ δF aµν = −f abc F bµν λc , (8.35)

which are homogeneous.

As a consistency check we verify the result (8.35) for δF aµν from the expression (8.32)
using (8.29) for δAaµ . First

δ(∂µ Aaν − ∂ν Aaµ ) = −f abc (∂µ Abν − ∂ν Abµ )λc − f abc (Abν ∂µ λc − Abµ ∂ν λc ) . (8.36)

Then
δ(f abc Abµ Acν )∣∂λ = −f abc (∂µ λb Acν + Abµ ∂ν λc ) , (8.37)
which cancels, using (5.39), the ∂λ terms in (8.36). Furthermore

δ(f abc Abµ Acν )∣λ = − f abc (f bde Adµ λe Acν + Abµ f cde Adν λe )
= − (f af d f fbe + f acf f fbe )Abµ Adν λe = −f af e f fbd Abµ Adν λe , (8.38)

by virtue of the Jacobi identity in the form (5.43). Combining (8.36), (8.37) and (8.38)
demonstrates (8.35) once more.

209
 The gauge fields Aaµ are associated with the adjoint representation of the gauge group
G. For any adjoint field Φa ta then the corresponding covariant derivative is given by

Dµ Φ = ∂µ Φ + [Aµ , Φ] ⇒ (Dµ Φ)a = ∂µ Φa + f abc Abµ Φc . (8.39)

This is in accord with the general form given by (8.23), with (8.22), using (5.172) for the
adjoint representation generators. Note that (8.29) can be written as δAaµ = −(Dµ λ)a and
for an arbitrary variation δAaµ from (8.32),

δF aµν = (Dµ δAν )a − (Dν δAµ )a . (8.40)

From the identity

([Dω , [Dµ , Dν ]] + [Dν , [Dω , Dµ ]] + [Dµ , [Dν , Dω ]])φ = 0 , (8.41)

for any representation, we have the non abelian Bianchi identity, generalising (8.19),

Dω Fµν + Dν Fωµ + Dµ Fνω = 0 , (8.42)

where the adjoint covariant derivatives are as defined in (8.39). Alternatively with the
notation in (8.33)
dF + A ∧ F − F ∧ A = 0 . (8.43)

To construct a lagrangian leading to dynamical equations of motion which are covariant

under gauge transformations it is necessary to introduce a group invariant metric gab = gba ,
satisfying (5.187) or equivalently

gdb f dca + gad f dcb = 0 , (8.44)

which also implies, for finite group transformations g and with X, Y belonging to the asso-
ciated Lie algebra,
gab (gXg −1 )a (gY g −1 )b = gab X a Y b . (8.45)
If X, Y are then adjoint representation fields the definition of the adjoint covariant derivative
in (8.39) gives
∂µ (gab X a Y b ) = gab ((Dµ X)a Y b + X a (Dµ Y )b ) , (8.46)
in a similar fashion to covariant derivatives in general relativity.
The simplest gauge invariant lagrangian, extending the abelian result in (8.16), is then,
as a result of the transformation properties (8.34) or (8.35), just the obvious extension of
that proposed by Yang and Mills for SU (2)
1
LYM = − gab F aµν F bµν . (8.47)
4
It is essential that the metric be non degenerate det[gab ] ≠ 0, and then using (8.40) requiring
the action to be stationary gives the gauge covariant dynamical equations

(Dµ Fµν )a = 0 . (8.48)

210
These equations, as well as (8.42) and unlike the abelian case, are non linear. As described
before a necessary consequence of gauge invariance is that if Aµ is a solution then so is
any gauge transform as in (8.26) and hence the time evolution of Aµ is arbitrary up to this
extent, only gauge equivalence classes, belonging to (8.4), have a well defined dynamics. If
the associated quantum field theory is to have a space of quantum states with positive norm
then it is also necessary that the metric gab should be positive definite. This requires that
the gauge group G should be compact and restricted to the form exhibited in (5.194). Each
U (1) factor corresponds to a simple abelian gauge theory as described in 8.1. If there are
no U (1) factors G is semi-simple and gab is determined by the Killing form for each simple
group factor. For G simple then by a choice of basis we may take
1
gab = δab , (8.49)
g2
with g the gauge coupling. For G a product of simple groups then there is a separate
coupling for each simple factor, unless additional symmetries are imposed.
If the condition that the metric gab be positive definite is relaxed then the gauge group
G may be non compact, but there are also examples of non semi-simple Lie algebras with
a non-degenerate invariant metric. The simplest example is given by the Lie algebra iso(2)
with a central extension, which is given in (5.136). Choosing Ta = (E1 , E2 , J3 , 1) then it is
straightforward to verify that

⎛1 0 0 0⎞
⎜0 1 0 0⎟
[gab ] = ⎜ ⎟, β arbitrary , (8.50)
⎜0 0 β c⎟
⎝0 0 c 0⎠

is invariant. The Killing form only involves the matrix with the element proportional to β
non zero. Since it is necessary that c ≠ 0 for the metric to be non-degenerate the presence of
the central charge in the Lie algebra is essential. For any β it is easy, since det[gab ] = −c2 ,
to see that [gab ] has one negative eigenvalue.
An illustration of the application of identities such as (8.46) is given by the conservation
of the gauge invariant energy momentum tensor defined by

T µν = gab (F aµσ F bν σ − 14 g µν F aσρ F bσρ ) . (8.51)

Then

∂µ T µ ν = gab ((Dµ F µσ )a F bνσ + F aµσ (Dµ Fσν )b − 12 F aσρ (Dν Fσρ )b )

= gab (Dµ F µσ )a − 12 gab F aσρ ((Dρ Fνσ )b − (Dσ Fνρ )b + (Dν Fσρ )b )
= gab (Dµ F µσ )a , (8.52)

using the Bianchi identity (8.42). Clearly this is conserved subject to the dynamical equation
(8.48).

211
8.2.1 Chern-Simons Theory

The standard gauge invariant lagrangian is provided by (8.47). However in order to obtain
a gauge invariant action, given by the integral over space-time of the lagrangian, it is only
necessary that the lagrangian is invariant up to a total derivative. This allows for additional
possibility for gauge field theories, with gauge group G a general Lie group, in three space-
time dimensions, termed Chern-Simons55 theories.
First we note that in four dimensions the Bianchi identity (8.42) may be alternatively
be written using the four dimensional antisymmetric symbol as

εµνσρ Dν Fσρ = 0 . (8.53)

Apart from (8.47) there is then another similar gauge invariant and Lorentz invariant
1 µνσρ
ε gab F aµν F bσρ , (8.54)
4
which may be used as an additional term in the lagrangian. However the corresponding
contribution to the action is odd under x → −x or t → −t. Such a term does not alter the
dynamical equations since its variation is a total derivative and thus the variation of the
corresponding term in the action vanishes. To show this under arbitrary variations of the
gauge field we use (8.40) and (8.53) to give
1
δ εµνσρ gab F aµν F bσρ = εµνσρ gab (Dµ δAν )a F bσρ = ∂µ (εµνσρ gab δAaν F bσρ ) . (8.55)
4
This allows us to write
1 µνσρ
ε gab F aµν F bσρ = ∂µ ω µ , (8.56)
4
where
ω µ = εµνσρ gab (Aaν ∂σ Abρ + 13 f bcd Aaν Acσ Adρ ) , (8.57)
since this has the variation

δω µ = εµνσρ gab (δAaν ∂σ Abρ + Aaν ∂σ δAbρ + f bcd δAaν Acσ Adρ )
= εµνσρ gab ∂σ (Aaν δAbρ ) + εµνσρ gab δAaν (2∂σ Abρ + f bcd Acσ Adρ )
= εµνσρ gab ∂ν (δAaσ Abρ ) + εµνσρ gab δAaν F bσρ , (8.58)

using that gab f bcd is totally antisymmetric as a consequence of (8.44). The result is then in
agreement with (8.55).
If the variation is a gauge transformation so that

Aaµ → A′aµ ⇒ ω µ → ω ′µ , (8.59)

g g

then since (8.56) is gauge invariant we must require

∂µ ω µ = ∂µ ω ′µ . (8.60)
55
Shiing-Shen Chern, 1911-2004, Chinese, American after 1960. James Harris Simons, 1938-, American.

212
This necessary condition may be verified for an infinitesimal gauge transformation by setting
δAaν = −(Dν λ)a in (8.58) which then gives, using the Bianchi identity (8.53) again,

δω µ = − εµνσρ gab ∂ν ((Dσ λ)a Abρ ) − εµνσρ gab (Dν λ)a F bσρ
= εµνσρ gab ∂ν (λa (Dσ A)bρ − λa F bσρ )
= − εµνσρ gab ∂ν (λa ∂σ Abρ ) . (8.61)

Hence it is evident that the result in (8.61) satisfies

∂µ δω µ = 0 . (8.62)

In three dimensions the identities for ω µ may be applied to

LCS = ενσρ gab (Aaν ∂σ Abρ + 13 f bcd Aaν Acσ Adρ ) , (8.63)

which defines the Chern-Simons lagrangian for gauge fields. For an infinitesimal gauge
transformation, by virtue of (8.61), LCS becomes a total derivative since

δAaν = −(Dν λ)a ⇒ δLCS = −ενσρ gab ∂ν (λa ∂σ Abρ ) , (8.64)

so that the corresponding action is invariant. Under a general variation

δ ∫ d3 x LCS = ∫ d3 x ενσρ gab δAaν F bσρ , (8.65)

so that the dynamical equations are

F aµν = 0 , (8.66)
so the connection Aµ is ‘flat’ since the associated curvature is zero (Cherns-Simons theory is
thus similar to three dimensional pure gravity where the Einstein equations require that the
Riemann curvature tensor vanishes). In a Chern-Simons theory there are no perturbative
degrees of freedom, as in the case of Yang-Mills theory, but topological considerations play
a crucial role.
Topology also becomes relevant as the Chern-Simons action is not necessarily invariant
under all gauge transformations if they belong to topological classes which cannot be con-
tinuously connected to the identity. To discuss this further it is much more natural again
to use the language of forms, expressing all results in terms of A(x) = Aaµ (x)ta dxµ a Lie
algebra matrix valued connection one-form, [ta , tb ] = f cab tc as in (5.60), and replacing the
group invariant scalar product by the matrix trace. For any set of such Lie algebra matrices
{X1 , . . . , Xn } the trace tr(X1 . . . Xn ) is invariant under the action of adjoint group trans-
formations Xr → gXr g −1 for all r. Since the wedge product is associative and the trace is
invariant under cyclic permutations we have

tr( A ∧ ⋅ ⋅ ⋅ ∧ A ) = tr((A ∧ ⋅ ⋅ ⋅ ∧ A) ∧ A) = (−)n−1 tr(A ∧ (A ∧ ⋅ ⋅ ⋅ ∧ A))

´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶
n n−1 n−1
=0 for n even . (8.67)

213
 The Chern-Simons theory is then defined in terms of the three-form

ω = tr(A ∧ dA + 23 A ∧ A ∧ A) = tr(A ∧ F − 13 A ∧ A ∧ A) , (8.68)

with the two-form curvature F as in (8.33). It is easy to see that

dω = tr(dA ∧ dA + 2 dA ∧ A ∧ A) = tr(F ∧ F ) , (8.69)

which is equivalent to (8.56) and (8.57). For a finite gauge transformation

A′ = gAg −1 + gdg −1 , F ′ = gF g −1 , (8.70)

so that, from (8.68),

ω ′ = ω + tr(dg −1 g ∧ (F − A ∧ A)) − tr(dg −1 g ∧ dg −1 g ∧ A)

− 31 tr(dg −1 g ∧ dg −1 g ∧ dg −1 g) . (8.71)

Using
dg −1 g = −g −1 dg , d(g −1 dg) = −g −1 dg ∧ g −1 dg , (8.72)
we get
ω ′ = ω + d tr(g −1 dg ∧ A) + 31 tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) . (8.73)
In this discussion g −1 dg is unchanged under g → g0 g, for any fixed g0 , and so defines a
left invariant one-form. If br are coordinates on the associated group manifold MG then
g −1 (b)dg(b) = ω a (b)ta where ω a (b) are the one forms defined in the general analysis of Lie
groups in (5.48).
Since, using (8.72),

d tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) = −tr(g −1 dg ∧ g −1 dg ∧ g −1 dg ∧ g −1 dg) = 0 , (8.74)

by virtue (8.67), we have

dω ′ = dω , (8.75)
which is equivalent to (8.60). However although tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) is therefore a
closed three-form it need not be exact so that its integration over a three manifold M3 may
not vanish, in which case we would have

∫ ω′ ≠ ∫ ω, (8.76)
M3 M3

for some g(x). The Cherns-Simons action is not then gauge invariant for such gauge trans-
formations g.
To discuss
I =∫ 1
tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) , (8.77)
M3 3

we note that for a variation of g, since

δ(g −1 dg) = g −1 d(δg g −1 ) g , (8.78)

214
then
δ 31 tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) = tr(d(δg g −1 ) ∧ dg g −1 ∧ dg g −1 )
= d tr(δg g −1 ∧ dg g −1 ∧ dg g −1 ) , (8.79)
since d(dg g −1 ∧ dg g −1 ) = −d2 (dg g −1 ) = 0. Hence, for arbitrary smooth variations δg,
δI = 0 , (8.80)
so that I is a topological invariant, only when g(x) can be continuously transformed to the
identity must I = 0.
If we consider g(θ) ∈ SU (2) with coordinates θr , r = 1, 2, 3 then
1
3 tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) = ρ(θ) d3 θ , (8.81)
The integration measure in (8.81) is defined in terms of the left invariant Lie algebra one
forms so that for g(θ′ ) = g0 g(θ) we have
ρ(θ′ ) d3 θ′ = ρ(θ) d3 θ . (8.82)
Up to a sign, depending just on the sign of det[∂ ′r /∂θs ], this is identical with the re-
quirements for an invariant integration measure described in section 5.7. To check the
normalisation we assume that near the origin, θ ≈ 0, then g(θ) ≈ I + iσ ⋅ θ and hence
1
3 tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) ≈ 13 i3 tr(σ ⋅ dθ ∧ σ ⋅ dθ ∧ σ ⋅ dθ)
i
= 2
3 εijk dθ ∧ dθj ∧ dθk = 4 d3 θ , (8.83)
assuming (5.21) and standard formulae for the Pauli matrices in (3.20) with (3.22). Thus
ρ(0) = 4 and the results for the group integration volume for SU (2) in (5.155) then imply,
integrating over MSU (2) ≃ S 3 ,

∫
1
tr(g −1 dg ∧ g −1 dg ∧ g −1 dg) = 8π 2 . (8.84)
MSU (2) 3

In general the topological invariant defined by (8.77), for a compact 3-manifold M3 ,

corresponds to the index of the map defined by g(x) from M3 to a subgroup SU (2) ⊂ G,
i.e. the number of times the map covers the SU (2) subgroup for x ∈ M3 . The result (8.84)
then requires that in general
I = 8π 2 n for n ∈ Z. (8.85)
In the functional integral approach to quantum field theories the action only appears in
the form eiS . In consequence S need only be defined up to integer multiples of 2π. Hence
despite the fact that the action is not invariant under all gauge transformations a well
defined quantum gauge Chern-Simons theory is obtained, on a compact 3-manifold M3 , by
employing as the action
k 2
SCS = ∫ tr(A ∧ dA + 3 A ∧ A ∧ A) , k ∈ Z, (8.86)
4π M3
so that, unlike Yang-Mills theory, the coupling is quantised. There is no requirement for k
to be positive, the cubic terms become effectively small, and the theory is weakly coupled,
when k is large.

215
8.3 Gauge Invariants and Wilson Loops

Only gauge invariant quantities have any significance in gauge field theories. Although
it is necessary in non abelian gauge theories to solve the dynamical equations for gauge
dependent fields, or in a quantum theory, to integrate over the gauge fields, only for gauge
invariants is a well defined calculational result obtained. For abelian gauge theories this is
a much less significant issue. The classical dynamical equations only involve Fµν which is
itself gauge invariant, (8.14). However even in this case the associated quantum field theory,
QED, requires a much more careful treatment of gauge issues.
For a non abelian gauge theory Fµν = F aµν ta is a matrix belonging to a Lie algebra
representation for the gauge group which transforms homogeneously under gauge transfor-
mations as in (8.34). The same transformation properties further apply to products of F ’s,
at the same space-time point, and also to the gauge covariant derivatives Dα1 . . . Dαr Fµν .
Since [Dα , Dβ ]Fµν = [Fαβ , Fµν ] the indices α1 , . . . , αn may be symmetrised to avoid linear
dependencies. A natural set of gauge invariants, for pure gauge theories, is then provided by
the matrix traces of products of F ’s, with arbitrarily many symmetrised covariant deriva-
tives, at the same point,

tr(Dα11 . . . Dα1r1 Fµ1 ν1 Dα21 . . . Dα2r2 Fµ2 ν2 . . . Dαs1 . . . Dαsrs Fµs νs ) . (8.87)

Such matrix traces may also be further restricted to a trace over a symmetrised product
of the Lie algebra matrices, since any commutator may be simplified by applying the Lie
algebra commutation relations, and also to just one of the s invariants, in the above example,
related by cyclic permutation as the traces satisfy tr(X1 . . . Xs ) = tr(Xs X1 . . . Xs−1 ). If the
gauge group G has no U (1) factors then tr(ta ) = 0. The simplest example of such an
invariant then involves just two F ’s, which include the energy momentum tensor as shown
in (8.51). In general there are also derivative relations since
s
∂µ tr(X1 . . . Xs ) = ∑ tr(X1 . . . Dµ Xi . . . Xs ) . (8.88)
i=1

However, depending on the gauge group, the traces in (8.87) are not independent for
arbitrary products of F ’s, even when no derivatives are involved. To show this we may
consider the identity
det(1 − X) = etr ln(1−X) , (8.89)
which is easy to demonstrate, for arbitrary diagonaliseable matrices X, since both sides
depend only on the eigenvalues of X and the exponential converts the sum over eigenvalues
provided by the trace into a product which gives the determinant. Expanding the right
hand side gives

det(1 − X) = e− ∑r≥1 tr(X

r )/r

= 1 − tr(X) + 21 (tr(X)2 − tr(X 2 )) − 61 (tr(X)3 − 3 tr(X)tr(X 2 ) + 2 tr(X 3 )) + . . . . (8.90)

If X is a N × N matrix then det(I − X) is at most O(X N ) so that terms which are of higher

216
order on the right hand side must vanish identically56 . If N = 2 this gives the relation

tr(X 3 ) = 3
2 tr(X) tr(X 2 ) − 12 tr(X)3 , (8.91)

and if N = 3, and also we require tr(X) = 0, the relevant identity becomes

tr(X 4 ) = 1
2 tr(X 2 )2 . (8.92)

In general tr(X n ) when n > N is expressible in terms of products of tr(X s ) for s ≤ N .

For G = SU (N ) and taking ta to belong to the fundamental representation these results
are directly applicable to simplifying symmetrised traces appearing in (8.87) since the results
for tr(X n ) are equivalent to relations for tr(t(a1 . . . taN ) ).

8.3.1 Wilson Loops

The gauge field Aµ is a connection introduced to ensure that derivatives of gauge dependent
fields transform covariantly under gauge transformations. It may be used, as with connec-
tions in differential geometry, to define ‘parallel transport’ of gauge dependent fields along a
path in space-time between two points, infinitesimally for x → x + dx this gives dxµ Dµ φ(x),
where φ is a field belonging to a representation space for the gauge group G and Dµ is the
gauge covariant derivative for this representation. Any continuous path Γx,y linking the
point y to x may be parameterised by xµ (t) where xµ (0) = y µ , xµ (1) = xµ . For all such
paths there is an associated element of the gauge group G, as in (8.2), which is obtained by
integrating along the path Γx,y . For the particular matrix representation R of G determined
by φ this group element corresponds to P (Γx,y ) ∈ R where P (Γx,y )φ(y) transforms under
local gauge transformations g(x) ∈ R belonging to Gx while φ(y) transforms as in (8.5) for
g(y) belonging to Gy .
For simplicity we consider an abelian gauge theory first. In this case P (Γx,y ) ∈ U (1) and
under gauge transformations transforms as a local field at x and its conjugate at y. For a
representation specified by a charge q as in (8.5), this is defined in terms of the differential
equation
d dxµ
( − iq ẋµ (t)Aµ (x(t)))P (t, t′ ) = 0 , P (t, t) = 1 , ẋµ = , (8.93)
dt dt
which has a solution,
t
P (t, t′ ) = eiq ∫t′ dτ ẋ
µ (τ )A
µ (x(τ ))
. (8.94)
We then require
iq ∫Γx,ydxµ Aµ (x)
P (Γx,y ) = P (1, 0) = e ∈ U (1) , (8.95)
56
Equivalently if F (z) = det(1 − zX) = 1 + ∑N
r=1 ar (X) z then
r

F ′ (z) ∞
− = tr(X(1 − zX)−1 ) = ∑ z r tr(X r+1 ) ,
F (z) r=0

and expanding the left hand side determines tr(X n ) for all n solely in terms of ar , r = 1, . . . N which are
also expressible in terms of tr(X n ) for n ≤ N .

217
which is independent of the particular parameterisation of the path Γx,y . Under the abelian
gauge transformation in (8.11)
iq ∫Γx,ydxµ ∂µ α(x)
P (Γx,y ) Ð→ P (Γx,y ) e = eiq α(x) P (Γx,y ) e−iq α(y) , (8.96)
eiα

demonstrating that, for φ transforming under gauge transformations as in (8.5),

P (Γx,y )φ(y) Ð→ eiqα(x) P (Γx,y )φ(y) . (8.97)

eiα

If Γ is a closed path, with a parameterisation xµ (t) such that xµ (1) = xµ (0) = xµ ∈ Γ,

then Γ = Γx,x for any x on Γ. It is evident from (8.96) that P (Γ) is gauge invariant. In
this abelian case P (Γ) may be expressed just in terms of the gauge invariant curvature in
(8.14) using Stokes’ theorem
µ (x) µν (x)
µA 1 µν F
P (Γ) = eiq ∮Γ dx = e 2 iq ∫S dS , (8.98)

for S any surface with boundary Γ and dS µν = −dS νµ the orientated surface area element
(in three dimensions the identity is ∮Γ dx ⋅ A = ∫S dS ⋅ B, B = ∇ × A with dSi = 12 εijk dS jk ).
For the non abelian case (8.93) generalises to a matrix equation

d
(1 + A(t))P (t, t′ ) = 0 , A(t) = ẋµ Aµ (x(t)) , P (t, t) = 1 , (8.99)
dt
where A(t) is a matrix belonging to the Lie algebra for a representation R of G. (8.99) may
also be expressed in an equivalent integral form
t
P (t, t′ ) = 1 − ∫ dτ A(τ )P (τ, t′ ) . (8.100)
t′

Solving this iteratively gives

t t1 tn−1
P (t, t′ ) = 1 + ∑ (−1)n ∫ dt1 ∫ dt2⋯ ∫ dtn A(t1 )A(t2 ) . . . A(tn )
′ t ′ ′ t t
n≥1
1 n t
= 1 + ∑ (−1)n ∏ ∫ ′ dtr T {A(t1 )A(t2 ) . . . A(tn )} . (8.101)
n≥1 n! r=1 t

where T denotes that the non commuting, for differing t, A(t) are t-ordered so that

⎪A(t)A(t′ ) , t ≥ t′ ,
⎧
⎪ ′
T {A(t)A(t )} = ⎨ ′
(8.102)
⎩A(t )A(t) ,
⎪
⎪ t < t′ .

The final expression can be simply written as a T -ordered exponential

t
P (t, t′ ) = T {e− ∫t′ dτ A(τ ) } . (8.103)

The corresponding non abelian generalisation of (8.95) is then

− ∫Γx,ydxµ Aµ (x)
P (Γx,y ) = P (1, 0) = P{e } ∈ R, (8.104)

218
with P denoting path-ordering along the path Γ (this is equivalent to t-ordering with the
particular parameterisation xµ (t)). These satisfy the group properties

P (Γx,y ) P (Γy,z ) = P (Γx,y ○ Γy,z ) , (8.105)

where Γx,y ○ Γy,z denotes path composition, and, if R is a unitary representation

P (Γx,y )−1 = P (Γ−1 †

y,x ) = P (Γx,y ) , (8.106)

with Γ−1
y,x the inverse path to Γx,y .

For a gauge transformation as in (8.26), g(x) ∈ R, then in (8.99)

A(t) → g(t)A(t)g(t)−1 − ġ(t)g(t)−1 , g(t) = g(x(t)) ⇒ P (t) → g(t)P (t, t′ )g(t′ )−1 ,
g g
(8.107)
and hence
P (Γx,y ) Ð→ g(x)P (Γx,y ) g(y)−1 . (8.108)
g

For Γ = Γx,x a closed path then we may obtain a gauge invariant by taking the trace

W (Γ) = tr(P (Γx,x )) . (8.109)

W (Γ) is a Wilson57 loop. It depends on the path Γ and also on the particular representation
R of the gauge group. Wilson loops form a natural, but over complete, set of non local
gauge invariants for any non abelian gauge theory. They satisfy rather non trivial identities
reflecting the particular representation and gauge group. Subject to these the gauge field can
be reconstructed from Wilson loops for arbitrary closed paths up to a gauge transformation.
The associated gauge groups elements for paths connecting two points, as given in (8.104),
may also be used to construct gauge invariants involving local gauge dependent fields at
different points. For the field φ, transforming as in (8.5), φ(x)† P (Γx,y )φ(y) is such a
gauge invariant, assuming the gauge transformation g is unitary so that (8.5) also implies
φ(x)† → φ(x)† g(x)−1 .
If a closed loop Γ is shrunk to a point then the Wilson loop W (Γ) can be expanded in
terms of local gauge invariants, of the form shown in (8.87), at this point. As an illustration
we consider a rectangular closed path with the associated Wilson loop

W (◻) = tr(P (Γx,x+bej ) P (Γx+bej ,x+aei +bej ) P (Γx+aei +bej ,x+aei ) P (Γx+aei ,x )) , (8.110)

where here Γ are all straight line paths and ei , ej are two orthogonal unit vectors. To
evaluate W (◻) as a, b → 0 it is convenient to use operators x̂ν , ∂ˆµ with the commutation
relations
[x̂µ , x̂ν ] = 0 , [∂ˆµ , ∂ˆν ] = 0 , [∂ˆµ , x̂ν ] = δµν , (8.111)
which have a representation, acting on vectors ∣x⟩, x ∈ R4 , where

x̂µ ∣x⟩ , ∂ˆµ ∣x⟩ = −∂µ ∣x⟩ . (8.112)

57
Kenneth Geddes Wilson, 1936-, American. Nobel prize 1982.

219
In terms of these operators, since x̂ν e−te = e−te
µ D̂ µ D̂
µ µ
(x̂ν + teν ),

e−te
µ D̂
µ
∣x⟩ = ∣x(t)⟩ P (Γx(t),x ) , D̂µ = ∂ˆµ + Aµ (x̂) , xν (t) = xν + teν , (8.113)

which defines P (Γx(t),x ) for the straight line path Γx(t),x from x to x(t), with P (Γx,x ) = I.
To verify that P (Γx(t),x ) agrees with (8.103) we note that

∂ −teµ D̂µ ∂
∣x⟩ = −eµ D̂µ e−te D̂µ ∣x⟩ = ( ∣x(t)⟩ − ∣x(t)⟩ eµ Aµ (x(t)))P (Γx(t),x ) ,
µ
e (8.114)
∂t ∂t

using (8.112) as well as (8.113). It is then evident that (8.114) reduces to

∂
P (Γx(t),x ) = −eµ Aµ (x(t)) P (Γx(t),x ) , (8.115)
∂t
which is identical to (8.93). For the rectangular closed path in (8.110)

∣x⟩ P (Γx,x+bej ) P (Γx+bej ,x+aei +bej ) P (Γx+aei +bej ,x+aei ) P (Γx+aei ,x )

= ebD̂j eαD̂i e−bD̂j e−αD̂i ∣x⟩
= eab [D̂j ,D̂i ]− 2 a b [[D̂j ,D̂i ],D̂i ]∣+ 21 ab2 [D̂j ,[D̂j ,D̂i ]]+...
1 2
∣x⟩
−ab Fij (x)− 21 a2 b Di Fij (x)− 12 ab2 Dj Fij (x)+...
= ∣x⟩ e , (8.116)

using the Baker Cambell Hausdorff formula described in 5.4.2 and [D̂i , D̂j ] = Fij (x̂). Hence,
for a N -dimensional representation with tr(ta ) = 0, the leading approximation to (8.110) is
just

W (◻) = N + 12 a2 b2 (1 + 21 a∂i + 12 b∂j + 61 a2 ∂i2 + 16 b2 ∂j2 + 14 ab∂i ∂j ) tr(Fij Fij )

1 4 2
− ) 1 2 4
24 a b tr(Di Fij Di Fij − 24 a b tr(Dj Fij Dj Fij
)
1 3 3
6 a b tr(Fij Fij Fij + . . . , no sums on i, j . (8.117)
− )

For completeness we also consider how P (Γx,y ) changes under variations in the path
Γx,y . For this purpose the path Γ is now specified by xµ (t, s), depending continuously on
the additional variable s, which includes possible variations in the end points at t = 0, 1. If
we define t, s covariant derivatives on these paths by
∂ ∂ ∂xµ ∂xµ
Dt = 1 + At (t) , Ds = 1 + As (t) , At (t) = Aµ (x) , As (t) = Aµ (x) , (8.118)
∂t ∂s ∂t ∂s
leaving the dependence on s implicit, then
∂xµ ∂xν
[Dt , Ds ] = F (t) = Fµν (x) . (8.119)
∂t ∂s
With the definitions in (8.118), (8.99) becomes Dt P (t) = 0. Acting with Ds gives

Dt Ds P (t, t′ ) = F (t)P (t, t′ ) , Ds P (t, t) = As (t) , (8.120)

220
which has a straightforward solution giving
d 1
P (1, 0) + As (1)P (1, 0) − P (1, 0)As (0) = ∫ dt P (1, t)F (t)P (t, 0) . (8.121)
ds 0

The result (8.121) may be recast as

δΓ P (Γx,y ) + δxν Aν (x) P (Γx,y ) − P (Γx,y ) δy ν Aν (y)

=∫ dz µ P (Γx,z ) Fµν (z)δxν (z) P (Γz,y ) , (8.122)
Γx,y

where
Γx,y = Γx,z ○ Γz,y for z ∈ Γx,y . (8.123)
For a Wilson loop
δΓ W (Γ) = ∮ dxµ tr(Fµν (x)δxν (x) P (Γx,x )) . (8.124)
Γ

For a pure Chern-Simons theory then, as a consequence of the dynamical equation (8.66),
there are no local gauge invariants and also Wilson loops are invariant under smooth changes
of the loop path. The Wilson loop W (Γ) ≠ N only if it is not contractable to a point.

221
9 Integrations over Spaces Modulo Group Transformations

In a functional integration approach to quantum gauge field theories it is necessary to

integrate over the non trivial space of gauge fields modulo gauge transformations, as in (8.4)
with the definitions (8.1) and (8.2). This often becomes rather involved with somewhat
formal manipulations of functional integrals but the essential ideas can be illustrated in
terms of well defined finite dimensional integrals.
To this end we consider n-dimensional integrals of the form
n
∫ nd x f (x) , (9.1)
R

for classes of functions f which are invariant under group transformations belonging to a
group G,
f (x) = f (xg ) , for x → xg for all g ∈ G . (9.2)
g

Necessarily we require
−1
(xg1 )g2 = xg1 g2 , (xg )g = x, (9.3)
and also we assume, under the change of variable x → xg ,

dn x = dn xg . (9.4)

The condition (9.4) is an essential condition on the integration measure in (9.1), which is
here assumed for simplicity to be the standard translation invariant measure on Rn . If the
group transformation g acts linearly on x then it is necessary that G ⊂ Sl(n, R) ⋉ Tn , which
contains the n-dimensional translation group Tn .
For any x the action of the group G generates the orbit Orb(x) and those group elements
which leave x invariant define the stability group Hx ,

Orb(x) = {xg } , Hx = {h ∶ xh = x} . (9.5)

Clearly two points on the same orbit have isomorphic stability groups since

Hxg = g −1 Hx g ≃ Hx ⊂ G . (9.6)

We further require that for arbitrary x, except perhaps for a lower dimension subspace, the
stability groups are isomorphic so that Hx ≃ H. Defining the manifold M to be formed by
the equivalence classes [x] = {x/ ∼}, where xg ∼ x, or equivalently by the orbits Orb(x),
then M ≃ Rn /(G/H). We here assume that G, and also in general H, are Lie groups, and
further that H is compact. In this case M has a dimension which is less than n. Although
Rn is topologically trivial, M may well have a non trivial topology.
In the integral (9.1), with a G-invariant function f , the integration may then be reduced
to a lower dimensional integration over M, by factoring off the invariant integration over
G. To achieve this we introduce ‘gauge-fixing functions’ P (x) on Rn such that,

for all x ∈ Rn then P (xg ) = 0 for some g ∈ G ,

if P (x0 ) = 0 then P (x0g ) = 0 ⇒ g = h ∈ H , x0h = x0 . (9.7)

222
In consequence the independent functions P (x) ∈ Rn̂ where n̂ = dim G−dim H. The solutions
of the gauge fixing condition may be parameterised in terms of coordinates θr , r = 1, . . . , n−n̂,
so that
P (x0 (θ)) = 0 ⇒ θr coordinates on M , dim M = n − n̂ . (9.8)
For any P (x) an associated function ∆(x) is defined by integrating over the G-invariant
measure, as discussed in 5.7, according to
n̂ g
∫ dρ(g) δ (P (x )) ∆(x) = 1 . (9.9)
G

Since by construction dρ(g) = dρ(g ′ g) then it is easy to see that

∆(xg ) = ∆(x) for all g ∈ G. (9.10)

Using (9.9) in (9.1), and interchanging orders of integration, gives

n n n̂ g
∫ nd x f (x) = ∫ dρ(g) ∫ nd x δ (P (x )) ∆(x) f (x)
R G R

= ∫ dρ(g) ∫ dn xg δ n̂ (P (xg )) ∆(xg ) f (xg )

G n R

= ∫ dρ(g) ∫ dn x δ n̂ (P (x)) ∆(x) f (x) . (9.11)

G n R

using the invariance conditions (9.2), (9.4) and (9.10), and in the last line just changing the
integration variable from xg to x. For integration over M we then have a measure, which
is expressible in terms of the coordinates θr , given by

dµ(θ) = dn x δ n̂ (P (x)) ∆(x) . (9.12)

To determine ∆(x) in (9.9) then, assuming (9.7), if

g(α, h) = exp(α) h , α ∈ g/h , (9.13)

we define a linear operator D, which may depend on x0 , such that

x0g(α,h) = x0 + D(x0 )α , for α ≈ 0 , D(x0 ) ∶ g/h → Rn . (9.14)

If {Tâ } is a basis for g/h (if g has a non degenerate Killing form κ then κ(h, Tâ ) = 0 for all
â and we may write g = h ⊕ g/h) then

α = αâ Tâ , (9.15)

and, with the decomposition in (9.13),

n̂
dρ(g) ≈ ∏ dαâ dρH (h) for αâ ≈ 0 , (9.16)
â=1

for dρH (h) the invariant integration measure on H. For x near x0 we define the linear
operator P ′ by

P (x0 + y) = P ′ (x0 )y for y ≈ 0 , P ′ (x0 ) ∶ Rn → Rn̂ . (9.17)

223
Then in (9.9), with (9.16),

n̂ g n̂ g n̂ n̂ ′
∫ dρ(g) δ (P (x )) = ∫ dρ(g) δ (P (x0 )) = VH ∫ d α δ (P (x0 )D(x0 )α)
G G
1
= VH , VH = ∫ dρH (h) . (9.18)
∣ det P ′ (x 0 )D(x0 )∣ H

Hence in (9.9)
1
∆(x) = ∣ det P ′ (x0 )D(x0 )∣ for x = x0g . (9.19)
VH
In a quantum gauge field theory context det P ′ (x0 )D(x0 ) is the Faddeev-Popov58 deter-
minant. The determinant is non vanishing except at points x0 such that P (x0g ) = 0 has
solutions for g ≈ e and g ∉ H and the gauge fixing condition P (x) = 0 does not sufficiently
restrict g. The resulting measure, since

P (x) = 0 ⇒ x = x(θ, α) = x0 (θ)g(α,h) , (9.20)

from (9.12) becomes, with a change of variables x → θ, α,

1 n n̂
dµ(θ) = d x δ (P (x)) ∣ det M (θ)∣ , M (θ) = P ′ (x0 (θ))D′ (x0 (θ)) . (9.21)
VH
Note that
δ n̂ (P (x(θ, α))) ∣ det M (θ)∣ = δ n̂ (α) , (9.22)
and therefore the measure over M may also be expressed in terms of the Jacobian from θ, α
to x since
∂x ∂x
dµ(θ) = dn−n̂ θ ∣ det [ , ]∣ . (9.23)
∂θ ∂α α=0

With these results, for G compact, (9.11) gives

n n n̂
∫ nd x f (x) = VG ∫ nd x δ (P (x)) ∆(x) f (x) = VG ∫ dµ(θ) f (x0 (θ)) . (9.24)
R R M

As an extension we consider the situation when there is a discrete group W , formed by

transformations θ → θgi , such that

W = {gi ∶ x0 (θgi ) = x0 (θ)g(gi ) , g(gi ) ∈ G} . (9.25)

It follows that M (θgi ) = M (θ) and dµ(θgi ) = dµ(θ). Since the stability group H leaves
x0 invariant g(gi ) is not unique, hence in general it is sufficient that g(gi )g(gj ) = g(gi gj )h
for h ∈ H. In many cases it is possible to restrict the coordinates {θr } so that W becomes
trivial but it is also often natural not to impose such constraints on the θr ’s and to divide
(9.21) by ∣W ∣ to remove multiple counting so that
1
dµ(θ) = dn x δ n̂ (P (x)) ∣ det M (θ)∣ , (9.26)
∣W ∣ VH
58
Ludvig Dmitrievich Faddeev, 1934-2017, Russian. Viktor Nikolaevich Popov, 1937-1994, Russian.

224
9.1 Integrals over Spheres

As a first illustration of these methods we consider examples where the group G is one of
the compact matrix groups SO(n), U (n) or Sp(n) and the orbits under the action of group
transformations are spheres.
For the basic integral over x ∈ Rn in (9.1), where x = (x1 , . . . , xn ), we then consider

f (x) = F (x2 ) , (9.27)

where x2 = xi xi is the usual flat Euclidean metric. In this case we take G = SO(n) which
acts as usual x → x′ = Rx, regarding x here as an n-component column vector, for any
R
R ∈ SO(n). Since det R = 1 of course dn x′ = dn x. The orbits under the action of SO(n) are
all x with x2 = r2 fixed and so are spheres S n−1 for radii r. A representative point on any
such sphere may be chosen by restricting to the intersection with the positive 1-axis or

x0 = r(1, 0, . . . , 0, 0) , r > 0. (9.28)

In this case the stability group, for all r > 0, H ≃ SO(n − 1) since matrices leaving x0 in
(9.28) invariant have the form

1 0
R(R̂) = ( ), R̂ ∈ SO(n − 1) . (9.29)
0 R̂

Note that dim SO(n) = 21 n(n − 1) so that in this example n̂ = dim SO(n) − dim SO(n − 1) =
n − 1, and therefore n − n̂ = 1 corresponding to the single parameter r.
Corresponding to the choice (9.28) the corresponding gauge fixing condition, correspond-
ing to δ n̂ (P (x)), is
n
F(x) = θ(x1 ) ∏ δ(xi ) . (9.30)
i=2

The condition x1 > 0 may be omitted but then there is a residual group W ≃ Z2 correspond-
ing to reflections x1 → −x1 . For the generators of SO(n) given by (6.31) we have

Ss1 x0 = r(0, . . . , 1 , . . . , 0) , s = 2, . . . , n , (9.31)

®
s′ th place

so that in (9.13) we may take

n
α = ∑ αs Ss1 , (9.32)
s=2

so that
exp(α)x0 = r(1, α2 , . . . , αn ) for α ≈ 0 . (9.33)
For the measure we assume a normalisation such that

dρSO(n) (R) ≈ dn−1 α dρSO(n−1) (R̂) for R = exp(α)R(R̂) , α ≈ 0, (9.34)

225
where R(R̂) is given in (9.29). With the gauge fixing function in (9.30)
n
1
∫ dρSO(n) (R) F(Rx) = VSO(n−1) ∫ dn−1α ∏ δ(αs ∣x1 ∣) = VSO(n−1) . (9.35)
SO(n) s=2 ∣x1 ∣n−1

Hence
1
∆(x) = rn−1 , x2 = r 2 , r > 0 . (9.36)
VSO(n−1)
With this (9.24) becomes

VSO(n) ∞
n 2 n 2
∫ nd x F (x ) = VSO(n) ∫ nd x F(x) ∆(x) F (x ) = ∫ dr rn−1 F (r2 ) . (9.37)
R R VSO(n−1) 0

Of course this is just the same result as obtained by the usual separation of angular variables
for functions depending on the radial coordinate r.
For a special case
VSO(n) ∞ VSO(n)
−x
dr rn−1 e−r =
2 1 2
n
∫ nd x e = π 2n = ∫
1
2 Γ( 12 n) , (9.38)
R VSO(n−1) 0 VSO(n−1)

giving
1
VSO(n) 2π 2 n
= Sn = , (9.39)
VSO(n−1) Γ( 12 n)
where Sn is the volume of S n−1 . Since VSO(2) = 2π, or VSO(1) = 1, in general
1
n−1 π 4 n(n+1)
VSO(n) = 2 . (9.40)
∏ni=1 Γ( 12 i)

For the corresponding extension to the complex case we consider integrals over Cn ≃ R2n ,
of real dimension 2n, with coordinates Z = (z1 , . . . , zn ), zi ∈ C. The analogous integrals are
then
n
2n
∫ nd Z F (Z̄Z) , Z̄Z = ∑ ∣zi ∣2 , (9.41)
C i=1

and where
n
d2n Z = ∏ d2 zi , d2 z = dxdy for z = x + iy . (9.42)
i=1

In this case we may take G = U (n) ⊂ O(2n) where the transformations act Z → U Z for
U
U ∈ U (n) so that Z̄Z is invariant, as is also d2n Z. As in the discussion for SO(n) we may
take on each orbit
Z0 = r(1, 0, . . . , 0, 0) , r > 0. (9.43)
The stability group H ≃ U (n − 1) corresponding to matrices

1 0
U (Û ) = ( ), Û ∈ U (n − 1) . (9.44)
0 Û

226
In this case dim U (n) = n2 so that n̂ = dim U (n) − dim U (n − 1) = 2n − 1. The orbits are just
specified again by the single variable r.
Corresponding to (9.43) the gauge fixing condition becomes
n
F(Z) = θ(Re z1 )δ(Im z1 ) ∏ δ 2 (zi ) , δ 2 (z) = δ(x)δ(y) , z = x + iy . (9.45)
i=2

In terms of the generators defined in (6.2) we let

n
α = iα1 R1 1 + ∑ (αs Rs 1 − αs∗ R1 s ) , α1 ∈ R , αs ∈ C , s ≥ 2 . (9.46)
s=2

Hence
exp(α)Z0 = r(1 + iα1 , α2 , . . . , αn ) , α ≈ 0, (9.47)
and we take

dρU (n) (U ) ≈ dα1 ∏ns=2 d2 αs dρU (n−1) (Û ) for U = exp(α)U (Û ) , α ≈ 0. (9.48)

With these results

1
∫ dρU (n) (U ) F(U Z) = VU (n−1) , (9.49)
∣z1 ∣2n−1
which implies
1
∆(Z) = r2n−1 , Z̄Z = r2 , r > 0 . (9.50)
VU (n−1)
Finally
VU (n) ∞
2n 2n
∫ nd Z F (Z̄Z) = VU (n) ∫ nd Z F(Z) ∆(Z) F (Z̄Z) = ∫ dr r2n−1 F (r2 ) .
C C VU (n−1) 0
(9.51)
Corresponding to (9.39), (9.51) requires
VU (n)
= S2n . (9.52)
VU (n−1)
Taking VU (1) = 2π we have, with our normalisation,
1
π 2 n(n+1)
VU (n) = 2n . (9.53)
∏ni=1 Γ(i)

Since U (n) ≃ SU (n) × U (1)/Zn

2π
VU (n) = VSU (n) . (9.54)
n

A very similar discussion applies in terms of quaternionic numbers which are relevant
for Sp(n). For Q = (q1 , . . . , qn ) ∈ Hn the relevant integrals are
n
4n
∫ nd Q F (Q̄Q) , Q̄Q = ∑ ∣qi ∣2 , (9.55)
H i=1

227
and where
n
d4n Q = ∏ d4 qi , d4 q = dx dy du dv for z = x + iy + ju + kv . (9.56)
i=1

Q̄Q is invariant under Q → M Q for M ∈ Sp(n) ⊂ SO(4n), regarded as n × n quaternionic

M
unitary matrices M satisfying (1.121). As before we choose
Q0 = r(1, 0, . . . , 0, 0) , r > 0. (9.57)
The stability group H ≃ Sp(n − 1) corresponding to quaternionic matrices where M is
expressible in terms of M̂ ∈ Sp(n − 1) in an identical fashion to (9.44). We now have
dim Sp(n) = n(2n + 1) so that n̂ = dim Sp(n) − dim Sp(n − 1) = 4n − 1.
The associated gauge fixing condition becomes
n
F(Q) = θ(Re q1 )δ 3 (Im q1 ) ∏ δ 4 (qi ) , δ 4 (q) = δ(x)δ(y)δ(u)δ(v) , q = x + iy + iu + iv .
i=2
(9.58)
In terms of the generators defined in (6.2) we let
n
α = α1 R1 1 + ∑ (αs Rs 1 − αs R1 s ) , αs ∈ H , Re α1 = 0 . (9.59)
s=2
and
dρSp(n) (M ) ≈ d3 α1 ∏ns=2 d4 αs dρSp(n−1) (M̂ ) α ≈ 0. (9.60)
Hence we find
1
∆(Q) = r4n−1 , Q̄Q = r2 1 , r > 0. (9.61)
VSp(n−1)

The integral in (9.55) becomes

VSp(n) ∞
4n 4n
∫ nd Q F (Q̄Q) = VSp(n) ∫ nd Q F(Q) ∆(Q) F (Q̄Q) = ∫ dr r4n−1 F (r2 ) ,
H H VSp(n−1) 0
(9.62)
and corresponding to (9.39), (9.62) requires
VSp(n)
= S4n . (9.63)
VSp(n−1)
Since Sp(1) = {q ∶ ∣q∣2 = 1}, with the group property depending on ∣q1 q2 ∣ = ∣q1 ∣ q2 ∣, the group
manifold is just S 3 and

VSp(1) = ∫ d4 q δ(∣q∣ − 1) = S4 = 2π 2 , (9.64)

just as in (5.155). Hence

π n(n+1)
VSp(n) = 2n . (9.65)
∏ni=1 Γ(2i)

The results for the group volumes in (9.40), (9.53) and (9.65) depend on the conven-
tions adopted in the normalisation of the group invariant integration measure which are
here determined by (9.34), (9.48) and (9.60) in conjunction with (9.32), (9.46) and (9.59)
respectively.

228
9.2 Integrals over Symmetric and Hermitian Matrices

A class of finite dimensional group invariant integrals which are rather more similar to
gauge theories are those which involve integrals over real symmetric or complex hermitian
matrices.
For the real case for n × n symmetric matrices X the relevant integrals are of the form

1 1
n
n(n+1)
∫ d2 X f (X) , X = XT , d 2 n(n+1) X = ∏ dXii ∏ dXij , (9.66)
i=1 1≤i<j≤n

and we assume the invariance

f (X) = f (RXR−1 ) , R ∈ SO(n) . (9.67)

1
The measure d 2 n(n+1) X is invariant under X → RXR−1 . A standard result in the discussion
of matrices is that any symmetric matrix such as X may be diagonalised so that

⎛λ1 . . . 0⎞
RXR−1 = Λ = ⎜ ⋮ ⋱ ⋮ ⎟, (9.68)
⎝ 0 ... λn ⎠

where λi are the eigenvalues of X. If {λi } are all different there is no continuous Lie
subgroup of SO(n) such that RΛR−1 = Λ since

dim{X ∶ X = X T } − dim SO(n) = 21 n(n + 1) − 12 n(n − 1) = n , (9.69)

corresponding to the number of independent λi . The orbits of X under the action of SO(n)
are then determined by the eigenvalues {λi }. For any SO(n) invariant function as in (9.67)
we may write
f (X) = fˆ(λ) , λ = (λ1 , . . . λn ) . (9.70)

However there is a discrete stability group for Λ. The diagonal matrices corresponding
to reflections in the i-direction
i (9.71)
⎛ 10 01 ............ 0 ⎞
⎜ ⋮ ⋱ ⋮ ⎟
⎜ 1
⎟
⎜ ⎟
Ri = i ⎜ ⋮ −1 ⋮ ⎟ ∈ O(n) , i = 1, . . . , n , Ri2 = In . (9.72)
⎜ 1 ⎟
⎜ ⋮ ⋱ ⋮ ⎟
⎜ ⎟
⎝ 0 ............ 1 0⎠
01

generate the discrete group

{Ra1 ...an = R1a1 . . . Rn−1an ∶ ai = 0, 1} ≃ Z2×n , (9.73)

such that for any element

Ra1 ...an ΛRa1 ...an−1 = Λ . (9.74)

229
Furthermore for any permutation σ ∈ Sn there are corresponding matrices Rσ ∈ O(n), such
that (Rσ )ij xj = xσ(i) . The matrices {Rσ } form a faithful representation of Sn and

⎛λσ(1) . . . 0 ⎞
Rσ ΛRσ−1 = Λσ = ⎜ ⋮ ⋱ ⋮ ⎟. (9.75)
⎝ 0 ... λσ(n) ⎠

For the permutation (i i+1) the associated matrix is

i (9.76)
⎛ 10 01 ........... 0 ⎞
⎜ ⋮ ⋱ ⋮ ⎟
⎜ 01 ⎟
R(i i+1) = i ⎜ ⋮ 1 0 ⋮ ⎟, i = 1, . . . , n − 1 , (9.77)
⎜ ⋮ ⋱ ⋮ ⎟
⎜ ⎟
⎝ 0 ........... 1 0⎠
01

which may be used to determine Rσ for any σ by group multiplication. Since Rσ Ri Rσ−1 =
Rσ(i) the groups generated by permutations and reflections may be identified as the semi-
direct product Sn ⋉ Z2×n . However det Rσ = Pσ = ±1, with Pσ defined in (7.85), and also
det Ra1 ...an = (−1)∑i ai . Restricting to the subgroup belonging to SO(n), having determinant
one, we then take for the group W , as defined in (9.25),

W = (Sn ⋉ Z2×n )/Z2 , ∣W ∣ = 2n−1 n! . (9.78)

It is possible to restrict W to Z2 ×n−1 , formed by Ra1 ...an with ∑i ai even, by requiring that
the eigenvalues in (9.68) are ordered so that λi ≤ λi+1 . However the choice of W in (9.78)
is generally more convenient.
Taking X0 = Λ, as in (9.68), the corresponding gauge fixing condition is

F(X) = ∏ δ(Xij ) . (9.79)

1≤i<j≤n

For a rotation
R(α) = exp(α) , α = −αT , (9.80)
and the group invariant integration is then assumed to be normalised such that, for R as
in (9.80),
dρSO(n) (R(α)) ≈ ∏ dαij , α ≈ 0. (9.81)
1≤i<j≤n

With these assumptions, applying (9.9),

1
=∫ dρSO(n) (R) F(RXR−1 ) = ∏ ∫ dαij δ(αij (λj − λi )) , (9.82)
∆(X) SO(n) 1≤i<j≤n

so that
ˆ
∆(X) = ∣∆(λ)∣ for ˆ
∆(λ) = ∏ (λi − λj ) . (9.83)
1≤i<j≤n

230
 The resulting SO(n) invariant integration over symmetric matrices becomes

1
n(n+1)
VSO(n) 1
n(n+1)
∫ d2 X f (X) = ∫ d2 X F(X)∆(X) f (X)
∣W ∣
VSO(n)
= n ˆ
∫ d λ ∣∆(λ)∣ fˆ(λ) . (9.84)
2n−1 n!
Since the normalisations chosen in (9.80) and (9.81) are compatible with those assumed
previously we may use (9.40) for VSO(n) .
For the particular example
n n
f (X) = e− 2 κ tr(X ) ,
1 2
tr(X 2 ) = ∑ Xii2 + 2 ∑ Xij2 = ∑ λi2 , (9.85)
i=1 1≤i<j≤n i=1

then 1
π 4 n(n+1)
X e− 2 κ tr(X
1
n(n+1) 1 1 2)
∫ d2 = 22n ( ) . (9.86)
κ
Using (9.40) this defines a normalised probability measure for the eigenvalues for a Gaussian
ensemble of symmetric real matrices
1
κ 4 n(n+1) n
e− 2 κ ∑i λi .
1 2
dµ(λ)symmetric matrices =
ˆ
∏ dλi ∣∆(λ)∣ (9.87)
3
2 2
n
∏ni=1 Γ(1 + 12 i) i=1

There is a corresponding discussion for complex hermitian n × n matrices when the

integrals are of the form
n
X = X† ,
n 2 2
∫ d X f (X) , dn X = ∏ dXii ∏ d2 Xij , (9.88)
i=1 1≤i<j≤n

where f satisfies
f (X) = f (U XU −1 ) , U ∈ U (n) . (9.89)
Just as before hermitian matrices may be diagonalised

U XU −1 = Λ , (9.90)

where the diagonal elements of Λ are the eigenvalues of X as in (9.68). In this case there
is a non trivial continuous subgroup of U (n) leaving Λ invariant formed by the diagonal
matrices
⎛eiβ1 0 ... 0 ⎞
⎜ 0 eiβ2 ⋮ ⎟
U0 (β) = ⎜
⎜ ⋮
⎟, (9.91)
⎜ ⋱ ⋮ ⎟⎟
⎝ 0 . . . . . . . . . eiβn ⎠
and hence we may identify
H ≃ U (1)×n . (9.92)
In addition we may identify W = Sn formed by {Rσ } ⊂ U (n) which permute the eigenvalues
in Λ.

231
 The gauge fixing condition restricting X to diagonal form is now
2
F(X) = ∏ δ (Xij ) . (9.93)
1≤i<j≤n

In this case we may write for arbitrary U ∈ U (n),

U (α, β) = exp(α)U0 (β) , α = −α† , αii = 0 all i , (9.94)

and the group invariant integration is then assumed to be normalised such that, for U as
in (9.94),
n
2
dρU (n) (U (α, β)) ≈ ∏ d αij ∏ dβi , α ≈ 0 , 0 ≤ βi < 2π . (9.95)
1≤i<j≤n i=1

With these assumptions

∫ dρU (n) (U ) F(U XU −1 ) = (2π)n ∏ ∫ d2 αij δ 2 (αij (λj − λi )) . (9.96)

U (n) 1≤i<j≤n

Since
1 2
δ 2 (λz) = δ (z) , (9.97)
∣λ∣2
this gives
1 2 1 ˆ
∆(X) = ∏ (λi − λj ) = ∆(λ)2 . (9.98)
(2π)n 1≤i<j≤n (2π)n

The result for U (n) invariant integration over hermitian matrices becomes

n 2 VU (n) n 2
∫ d X f (X) = ∫ d X F(X)∆(X) f (X)
n!
VU (n) n ˆ 2 ˆ
= ∫ d λ ∆(λ) f (λ) , (9.99)
n! (2π)n

where we may use (9.53) for VU (n) .

For a Gaussian function
1 n n
n − κ tr(X
2 1 2) 1 π 2 n2
∫ d Xe 2 = 2 2 n( ) , tr(X 2 ) = ∑ Xii2 + 2 ∑ ∣Xij ∣2 = ∑ λi2 . (9.100)
κ i=1 1≤i<j≤n i=1

Using (9.53) this defines a normalised probability measure for the eigenvalues for a Gaussian
ensemble of hermitian matrices
1 2
κ2n n
ˆ 2 − 2 κ ∑ i λi
1 2
dµ(λ)hermitian matrices = 1 ∏ dλi ∆(λ) e . (9.101)
(2π) 2
n
∏ni=1 i! i=1

Extending this to quaternionic hermitian n × n matrices the relevant integrals are

n
n(2n−1)
∫ d X f (X) , X = X̄ , dn(2n−1) X = ∏ dXii ∏ d4 Xij , (9.102)
i=1 1≤i<j≤n

232
where X̄ is defined by (1.114) and integration over quaternions is given by (9.56). f is now
assumed to satisfy
f (X) = f (M XM −1 ) , M ∈ U (n, H) . (9.103)
Such quaternionic matrices may be diagonalised so that, for a suitable M ∈ U (n, H),

M XM −1 = Λ , λi = λ̄i . (9.104)

Using the correspondence with 2n × 2n complex matrices provided by (1.115) and (1.79),
when M → M ∈ U Sp(2n, C) and X → X where

X = X† , X = −JX T J . (9.105)

The eigenvalues of X must then be ±λi , i = 1, . . . n, and (9.104) is equivalent to the matrix
theorem that the 2n × 2n antisymmetric matrix X J may be reduced to a canonical form in
terms of {λi },
0 λ1
⎛ −λ1 0 0 ⎞
⎜ 0 0 λ2 ⎟
M X J MT = ⎜ −λ2 0 ⎟ for M ∈ U (2n) . (9.106)
⎜ ⋱ ⎟
⎜ ⎟
⎝ 0 λn ⎠
−λn 0

In (9.104) the subgroup of U (n, H) leaving Λ invariant is formed by quaternionic matrices

⎛q1 0 . . . 0 ⎞
⎜ 0 q2 ⋮⎟
M0 (q) = ⎜ ⋮ ⋱ ⋮⎟ , ∣qi ∣ = 1 , (9.107)
⎜ ⎟
⎝ 0 . . . . . . . qn ⎠

giving
H ≃ U (1, H)×n . (9.108)
As before W = Sn formed by {Rσ } ⊂ U (n, H) which permute the diagonal elements in Λ.
The gauge fixing condition restricting X to diagonal form is now
4
F(X) = ∏ δ (Xij ) . (9.109)
1≤i<j≤n

In this case we may write for arbitrary M ∈ U (n, H),

M (α, q) = exp(α)M0 (q) , α = −ᾱ , αii = 0 all i , (9.110)

and the group invariant integration is then assumed to be normalised such that, for M as
in (9.110),
n
4 4
dρU (n,H) (M (α, q)) ≈ ∏ d αij ∏ d qi δ(∣qi ∣ − 1) , α ≈ 0. (9.111)
1≤i<j≤n i=1

With these assumptions and using (9.64)

∫ dρU (n,H) (M ) F(M XM −1 ) = (2π 2 )n ∏ ∫ d4 αij δ 4 (αij (λj − λi )) . (9.112)

U (n,H) 1≤i<j≤n

233
In this case
1 4 1 ˆ 4
∆(X) = ∏ (λi − λj ) = ∆(λ) . (9.113)
(2π 2 )n 1≤i<j≤n (2π 2 )n

The result for U (n, H) invariant integration over quaternion hermitian matrices becomes

n(2n−1)
VSp(n) n 2
∫ d X f (X) = ∫ d X F(X)∆(X) f (X)
n!
VSp(n) n ˆ 4 ˆ
= ∫ d λ ∆(λ) f (λ) , (9.114)
n! (2π 2 )n

where we may use (9.65) for VSp(n) .

For the Gaussian integral
1 n n
π 2 n(2n−1)
X e− 2 κ tr(X
n(2n−1) 1 2) 1
∫ d = 2 2 n( ) , tr(X 2 ) = ∑ Xii2 + 2 ∑ ∣Xij ∣2 = ∑ λi2 .
κ i=1 1≤i<j≤n i=1
(9.115)
Using (9.65) we therefore obtain a normalised probability measure for the eigenvalues for a
Gaussian ensemble of hermitian quaternionic matrices
1
2 2 n κ 2 n(2n−1) n
1
ˆ 4 − 12 κ ∑i λi2
dµ(λ)hermitian quaternionic matrices = ( ) n ∏ dλi ∆(λ) e . (9.116)
π ∏i=1 (2i)! i=1

9.2.1 Large n Limits

The results for the eigenvalue measure dµ(λ), given by (9.87), (9.101) and (9.116) for a
Gaussian distribution of real symmetric and hermitian complex and quaternion matrices,
can be simplified significantly in a limit when n is large. In each case the distribution has
the form
1
dµ(λ) = Nn dn λ e−W (λ) , W (λ) = κ ∑ λi2 − 12 β ∑ ln ∣λi − λj ∣ , (9.117)
2 i i,j,i≠j

where β = 1, 2, 4 and we may order the the eigenvalues so that

λ1 < λ2 < ⋅ ⋅ ⋅ < λn . (9.118)

For a minimum W (λ) is stationary when

1
κ λi = β ∑ . (9.119)
j≠i i λj
λ −

In the large n limit we may approximate λi by a smooth function,

n 1
i dx
λi → λ(x) , x= , ∑ = n ∫ dx = n ∫ dλ ρ(λ) , ρ(λ) = > 0, (9.120)
n i=1 0 dλ

234
where ρ(λ) determines the eigenvalue distribution and is normalised since
1
∫ dλ ρ(λ) = ∫ dx = 1 . (9.121)
0

As n → ∞ the distribution is dominated by λ(x) such that W (λ) is close to its minimum.
The minimum is determined by (9.119) or, taking the large n limit,
κ 1
λ = P ∫ dµ ρ(µ) , (9.122)
nβ λ−µ
where P denotes that the principal part prescription is used for the singularity in the integral
at µ = λ.
(9.122) is an integral equation for ρ. To solve this we define the function
R 1 1
F (z) = ∫ dµ ρ(µ) ∼ as z → ∞, (9.123)
−R z−µ z

using (9.121) and assuming

ρ(µ) > 0 , ∣µ∣ < R , ρ(µ) = 0 , ∣µ∣ > R . (9.124)

F (z) is analytic in z save for a cut along the real axis from −R to R. The integral equation
requires
κ
F (µ ± i) = µ ∓ iπ ρ(µ) , ∣µ∣ < R . (9.125)
nβ
Requiring F (z) = O(z −1 ) for large z this has the unique solution
κ √
F (z) = (z − z 2 − R2 ) . (9.126)
nβ

The large z condition in (9.123) requires

2nβ
R2 = . (9.127)
κ
This then gives
2 √ 2
ρ(λ) = R − λ2 . (9.128)
πR2
This is Wigner’s semi-circle distribution and is relevant for nuclear energy levels.

9.3 Integrals over Compact Matrix Groups

Related to the discussion of integrals over group invariant functions of symmetric or her-
mitian matrices there is a corresponding treatment for integrals over functions of matrices
belonging to the fundamental representation for SO(n), U (n) or Sp(n). For simplicity we
consider the unitary case first.

235
 For matrices U ∈ U (n) the essential integral to be considered is then defined in terms of
the n2 -dimensional group invariant measure by

∫ dρU (n) (U ) f (U ) , (9.129)

U (n)

where
f (U ) = f (V U V −1 ) for all V ∈ U (n) . (9.130)
Just as for hermitian matrices U can be diagonalised so that

V U V −1 = U0 (θ) , θ = (θ1 , . . . , θn ) , (9.131)

where U0 is defined in (9.91). For θi all different V is arbitrary up to V ∼ V U0 (β), for any
β = (β1 , . . . , βn ) so that the associated stability group H = U (1)×n . The remaining discrete
symmetry group in this case is then

WU (n) ≃ Sn , (9.132)

since, for any permutation σ ∈ Sn , there is a Rσ ∈ O(n) such that

Rσ U0 (θ)Rσ−1 = U0 (θσ ) , θσ = (θσ(1) , . . . , θσ(n) ) . (9.133)

Thus we use the gauge fixing condition

2
F(U ) = ∏ δ (Uij ) . (9.134)
1≤i<j≤n

Using the same results as given in (9.94) and (9.95) we then get

∫ dρU (n) (V ) F(V U V −1 ) = (2π)n ∏ ∫ d2 αij δ 2 (αij (eiθj − eiθi )) , (9.135)

U (n) 1≤i<j≤n

so that, using (9.97),

1 iθ iθ 2 1 2
∆(U ) = ∏ ∣e j − e i ∣ = ∏ (2 sin 12 (θi − θj ))
(2π)n 1≤i<j≤n (2π)n 1≤i<j≤n
1 ˆ iθ ˆ −iθ
= ∆(e )∆(e ) , (9.136)
(2π)n

with the definition (9.83). The basic formula (9.26) then gives an integration measure over
the θi ’s
n
1 2
dµU (n) (θ) = n ∏ dθi ∏ (2 sin 21 (θi − θj )) , 0 ≤ θi ≤ 2π . (9.137)
n! (2π) i=1 1≤i<j≤n

By restricting f (U ) = 1 in (9.129) it is clear that this integration measure is normalised,

∫ dµU (n) (θ) = 1, since VU (n) may be factored from both sides.
To reduce to SU (n) we let θi = θ + θ̂i , i = 1, . . . , n − 1, θn = θ − ∑n−1
i=1 θ̂i , where now
n n−1
0 ≤ θ̂i ≤ 2π and 0 ≤ θ ≤ 2π/n and also ∏i=1 dθi = ndθ ∏i=1 dθ̂i . The θ integral may

236
then be factored off, corresponding to the decomposition U (n) ≃ SU (n) × U (1)/Zn , or
equivalently θn is no longer an independent variable but determined by ∑i θi = 0. For any
Rσ if det Rσ = −1 we may define R̂σ = eπi/n Rσ and otherwise R̂σ = Rσ so that {R̂σ } ⊂ SU (n)
and also R̂σ U0 (θ)Rσ−1 = U0 (θσ ). Hence, as in (9.132), we still have

WSU (n) ≃ Sn . (9.138)

Restricting (9.137) to SU (n) we then obtain

n−1 n−1
1 2
dµSU (n) (θ) = ∏ dθi ∏ (2 sin 21 (θi − θj )) , θ n = − ∑ θi . (9.139)
n! (2π)n−1 i=1 1≤i<j≤n i=1

For real orthogonal matrices in a similar fashion

∫ dρSO(n) (R) f (R) , f (R) = f (SRS −1 ) for all S ∈ SO(n) . (9.140)

SO(n)

In this case it is necessary to distinguish between even and odd n. For any R ∈ SO(2n) it
can be transformed to

⎛r(θ1 ) 0 ... 0 ⎞
0 r(θ
SRS −1 = R0 (θ) = ⎜ ⋮
⎜ 2 ) ⋮ ⎟
⋱ ⋮ ⎟ , S ∈ SO(2n) , (9.141)
⎜ ⎟
⎝ 0 . . . . . . . . . . r(θn )⎠

where R0 (θ) is written as a n × n matrix of 2 × 2 blocks with

r(θ) = ( −cos θ sin θ

sin θ cos θ ) . (9.142)

In (9.141) S ∼ SR0 (β), for arbitrary β = (β1 , . . . , βn ), so that the stability group for R0 (θ)
is then SO(2)×n . The discrete group defined by 2n × 2n matrices {S} ∈ O(2n) such that
SR0 (θ)S −1 = R0 (θ′ ) is Sn ⋉ Z2×n , with the permutation group Sn formed by {Rσ × 12 } and
Z2×n generated by
i
1
⎛ 2 . . . . . . . . ........ 0 ⎞
⎜ ⋮ ⋱ ⋮ ⎟
⎜ ⋮ 12 ⋮ ⎟
Ri = i ⎜ ⋮
⎜ ⋮ σ 3 ⋮ ⎟ ∈ O(2n) , i = 1, . . . , n , Ri2 = 12n , (9.143)
⎜ 12 ⋮ ⎟
⎟
⎜ ⋮ ⋱ ⋮ ⎟
⎝ 0 . . . . . . . . . . . . . . . . 12 ⎠

since σ3 r(θ) σ3 = r(2π −θ), for σ3 = ( 10 −1

0 ). Restricting to the subgroup formed my matrices

with determinant one

WSO(2n) ≃ (Sn ⋉ Z2×n )/Z2 . (9.144)

Writing R ∈ SO(2n) in terms of 2 × 2 blocks Rij , i, j = 1, . . . n, the gauge fixing condition

is then taken as
F(R) = ∏ δ 4 (Rij ) , (9.145)
1≤i<j≤n

237
with the definitions

δ 4 (A) = δ(a)δ(b)δ(c)δ(d) , d4 A = da db dc dd for A = ( ac db ) . (9.146)

For a general rotation S ∈ SO(2n) we may write

S = eA S0 (β) , AT = −A , Aii = 0 all i , (9.147)

and then
n
4
dρSO(2n) (S) ≈ ∏ d Aij ∏ dβi for A ≈ 0. (9.148)
1≤i<j≤n i=1

Using (9.148) is then sufficient to obtain

∫ dρSO(2n) (S) F(SR0 (θ)S −1 ) = (2π)n ∏ ∫ d4 Aij δ 4 (Aij r(θj )−r(θi )Aij ) . (9.149)
SO(2n) 1≤i<j≤n

With
1
δ 4 (Ar(θ) − r(θ′ )A) = δ 4 (A) , (9.150)
4(cos θ − cos θ′ )2
we then get for SO(2n)
1 2 1 ˆ cos θ))2 ,
∆(R) = ∏ (2(cos θi − cos θj )) = (∆(2 (9.151)
(2π)n 1≤i<j≤n (2π)n

ˆ is defined by (9.83).
where ∆
Combining the ingredients the measure for integration reduces in the SO(2n) case to
an integral over the n θi ’s given by
n
1 ˆ cos θ))2 .
dµS0(2n) (θ) = ∏ dθi (∆(2 (9.152)
2n−1 n! (2π)n i=1

For SO(2n+1) (9.141) may be modified, by introducing one additional row and column,
to
⎛r(θ1 ) 0 ... 0 0⎞
⎜ 0 r(θ 2 ) ⋮⎟
SRS −1 = R0 (θ) = ⎜ ⋮
⎜ ⋮⎟
⋱ ⎟, S ∈ SO(2n + 1) , (9.153)
⎜ ⎟
⎜ 0 r(θn ) 0 ⎟
⎝ 0 ................. 1⎠
with r(θ) just as in (9.142). Instead of (9.143) we may now take
i
⎛12 . . . . . . . . . . . . . . . . 0 0 ⎞
⎜ ⋮ ⋱ ⋮ ⎟
⎜ ⋮
⎜ 12 ⋮ ⎟
⎟
⎜ ⋮ σ3 ⋮ ⎟
Ri = i ⎜
⎜ ⋮ 12 ⋮
⎟ ∈ SO(2n + 1) ,
⎟ i = 1, . . . , n , (9.154)
⎜ ⋮ ⋱ ⋮ ⎟
⎜ ⎟
⎜0 12 ⎟
0 ⎟
⎜
⎝ 0 ..................... −1 ⎠

238
and in a similar fashion, for any permutation σ ∈ Sn , there is a Rσ ∈ SO(2n + 1), with the
matrix Rσ having 1, −1 in the bottom right hand corner according to whether σ is even,odd,
such that Rσ R0 (θ)Rσ−1 = R0 (θσ ). Hence

WSO(2n+1) ≃ Sn ⋉ Z2×n . (9.155)

In this case R ∈ SO(2n + 1) is expressible in terms of 2 × 2 blocks Rij , i, j = 1, . . . n, 2 × 1

blocks Ri n+1 and also 1 × 2 blocks Rn+1 i for i = 1, . . . n. The gauge fixing condition is now
n
4 2
F(R) = ∏ δ (Rij ) ∏ δ (Ri n+1 ) , (9.156)
1≤i<j≤n i=1

with δ 2 ( ab ) = δ(a) δ(b), similarly to (9.146). Expressing S ∈ SO(2n + 1) in the same form as
(9.147) we now have
n n
4 2
dρSO(2n+1) (S) ≈ ∏ d Aij ∏ d Ai n+1 ∏ dβi for A ≈ 0, (9.157)
1≤i<j≤n i=1 i=1

so that

∫ dρSO(2n+1) (S) F(SR0 (θ)S −1 )

SO(2n+1)
n
= (2π)n ∏ ∫ d4 Aij δ 4 (Aij r(θj ) − r(θi )Aij ) ∏ ∫ d2 Ai n+1 δ 2 ((I2 − r(θi ))Ai n+1 ) .
1≤i<j≤n i=1
(9.158)

In the SO(2n + 1) case this implies

n
1 ˆ cos θ))2 ∏ (2 sin 1 θi )2 ,
∆(R) = (∆(2 2 (9.159)
(2π)n i=1

and in consequence
n
1 2
ˆ cos θ))2 .
dµS0(2n+1) (θ) = n n ∏ dθi (2 sin 12 θi ) (∆(2 (9.160)
2 n! (2π) i=1

The remaining case to consider is for integrals over M ∈ Sp(n) ≃ U (n, H) of the form

∫ dρSp(n) (M ) f (M ) , f (M ) = f (N M N −1 ) for all N ∈ Sp(n) . (9.161)

Sp(n)

By a suitable transformation the quaternion matrix M can be reduced to the diagonal form
iθ
⎛e 1 0 ... ⎞ 0
⎜ 0 eiθ2
N M N −1 = M0 (θ) = ⎜ ⋮
⎟ ⋮
⋱ ⎟,⋮ N ∈ Sp(n) , (9.162)
⎜ ⎟
⎝ 0 iθn ⎠
......... e

239
As before N ∼ N M0 (β) so the stability group is U (1)×n . The remaining discrete group
generated by Rσ 1 ∈ Sp(n), for σ ∈ Sn and 1 the unit quaternion, and also by
i
⎛1 . . . . . . . . . . . . . 0⎞
⎜⋮ ⋱ ⋮⎟
⎜⋮ 1 ⋮⎟
Ni = i ⎜ ⋮ j ⋮ ⎟ ∈ Sp(n) , i = 1, . . . , n . (9.163)
⎜⋮ 1 ⋮⎟
⎜⋮ ⋱ ⋮⎟
⎝0 . . . . . . . . . . . . . 1⎠

In this case Ni2 = N0 (β), with eiβi = −1, eiβj = 1, j ≠ i, so that Ni corresponds to a Z2
symmetry. Hence for Sp(n) we have

WSp(n) ≃ Sn ⋉ Z2×n . (9.164)

For the Sp(n) case we take

n
4 2
F(M ) = ∏ δ (Mij ) ∏ δ (Mii ) , (9.165)
1≤i<j≤n i=1

where, for any quaternion q, δ 4 (q) is defined as in (9.58) and also here

δ 2 (q) = δ(u)δ(v) for q = x + iy + ju + kv . (9.166)

Writing then, for any N ∈ Sp(n),

N = eα M0 (β) , αij = −ᾱji ∈ H , i ≠ j , αii = jui + kvi , (9.167)

we have
n n
4 2
dρSp(n) (N ) ≈ ∏ d αij ∏ d αii ∏ dβi for α ≈ 0, (9.168)
1≤i<j≤n i=1 i=1

so that

∫ dρSp(n) (N ) F(N M0 (θ)N −1 )

Sp(n)
n
= (2π)n ∏ ∫ d4 αij δ 4 (αij eiθj − eiθi αij ) ∏ ∫ d2 αii δ 2 (αii eiθi − eiθi αii ) . (9.169)
1≤i<j≤n i=1

For this case we may use

′ 1
δ 4 (α eiθ − eiθ α) = δ 4 (α) ,
4(cos θ − cos θ′ )2
1
δ 2 (α eiθ − eiθ α) = δ 2 (α) for α = ju + kv , (9.170)
4 sin2 θ
to obtain
n
1 ˆ cos θ))2 ∏ (2 sin θi )2 .
∆(M ) = (∆(2 (9.171)
(2π)n i=1

240
Hence
n
1 2
ˆ cos θ))2 .
dµSp(n) (θ) = n n ∏ dθi (2 sin θi ) (∆(2 (9.172)
2 n! (2π) i=1

As special cases we have dµSp(1) (θ) = dµSU (2) (θ), dµSO(3) (θ) = 2 dµSp(1) ( 12 θ) and also,
from SO(4) ≃ (Sp(1) × Sp(1))/Z2 , dµSO(4) (θ1 − θ2 , θ1 + θ2 ) = 2 dµSp(1) (θ1 ) dµSp(1) (θ2 ) with,
from SO(5) ≃ Sp(2)/Z2 , dµSO(5) (θ1 − θ2 , θ1 + θ2 ) = 2 dµSp(2) (θ1 , θ2 ), and, from SO(6) ≃
SU (4)/Z2 , dµSO(6) (θ2 + θ3 , θ3 + θ1 , θ1 + θ2 ) = 2 dµSU (4) (θ1 , θ2 , θ3 ).

9.4 Integration over a Gauge Field and Gauge Fixing

An example where the reduction of a functional integral over a gauge field A ∈ A can be
reduced to A/G, where G is a the gauge group, in an explicit fashion arises in just one
dimension. We then consider a gauge field A(t) with the gauge transformation, following
(8.26),
A(t) Ð→ A(t)g(t) = g(t)A(t)g(t)−1 − ∂t g(t) g(t)−1 , (9.173)
g

where here we take

A(t) = −A(t)† ∈ u(n) , g(t) ∈ U (n) . (9.174)
The essential functional integral has the form

∫ d[A] f (A) , f (Ag ) = f (A) , (9.175)

where we restrict to t ∈ S 1 by requiring the fields to satisfy the periodicity conditions

A(t) = A(t + β) , g(t) = g(t + β) . (9.176)

In one dimension there are no local gauge the discussion in 8.3 and the periodicity
requirement (9.176), the gauge invariant function f in (9.175) should have the form

f (A) = fˆ(U ) where fˆ(U ) = fˆ(g U g −1 ) for all g ∈ U (n) . (9.177)

In particular
Pβ (U ) = tr(U ) , (9.178)
is gauge invariant, being just the Wilson loop for the circle S 1 arising from imposing peri-
odicity in t. Pβ (U ) is a Polyakov59 loop.
The general discussion for finite group invariant integrals can be directly applied to the
functional integral (9.175). It is necessary to choose a convenient gauge fixing condition.
For any A(t) there is a gauge transformation g(t) such that

A(t)g(t) = iX , X† = X . (9.179)

In consequence we may choose a gauge condition ∂t A(t) = 0 or equivalently take

F[A] = δ ′ [A] , (9.180)

59
Alexander Markovich Polyakov, 1945-, Russian.

241
where δ ′ [A] is a functional δ-function, δ ′ denoting the exclusion of constant modes. For a
general Fourier expansion on S 1

A(t) = iX + ∑ An e2πint/β , X† = X , An† = −A−n , (9.181)

n≠0

where X is a hermitian and An are complex n × n matrices, then

δ ′ [A] = ∏ Nn δ 2n (An ) .
2
(9.182)
n>0

Nn is a normalisation factor which is chosen later. With the expansion (9.181) the functional
integral can also be defined by taking
2 1 2n2
d[A] = dn X ∏ d An . (9.183)
n>0 Nn

The integral (9.9) defining the Faddeev Popov determinant then becomes

′ g
∫ dµ(g) δ [A ] where A(t) = (iX)g(t) for some g(t) , (9.184)
G

and where dµ(g) is the invariant measure for the gauge group G. From (9.173) for an
infinitesimal gauge transformation

(iX)g(t) = iX + i[λ(t), X] − ∂t λ(t) for g(t) ≈ 1 + λ(t) , λ(t)† = −λ(t) . (9.185)

If
g(t) = g0 (1 + λ(t)) for λ(t) ≈ 0 , λ(t) = ∑ λn e2πint/β , λn† = −λ−n , (9.186)
n≠0

then we may take

2
dµ(g) ≈ dρU (n) (g0 ) d[λ] , d[λ] = ∏ d2n λn . (9.187)
n>0

Hence from (9.184) we define

′ g
VU (n)
∫ dµ(g) δ [(iX) ] = , (9.188)
G ∆(X)

where
1 2 2πn
= ∫ d[λ] δ ′ [i[λ, X] − ∂t λ] = ∏ Nn ∫ d2n λn δ 2n (
2
λn − i[X, λn ])
∆(X) n>0 iβ
2n2
2n2 2n2 β β
= ∏∫ d λn δ (λn + [X, λn ]) for Nn = ( ) , (9.189)
n>0 2πn 2πn

which gives
2
β
∆(X) = ∏ ( det (1n2 + X ad )) . (9.190)
n>0 2πn

242
The essential functional integral in (9.175) then reduces to just an integral over hermitian
matrices X,
1 n2
∫ d[A] f (A) = ∫ d X ∆(X) f (iX) . (9.191)
VU (n)

There is a remaining invariance under X → gXg −1 for constant g ∈ U (n). This may be
used to diagonalise X so that gXg −1 = Λ where Λ is the diagonal matrix in terms of the
eigenvalues λ1 , . . . , λn , as in (9.68). In terms of these

eigenvalues{X ad } = λi − λj , i, j = 1, . . . , n . (9.192)

Hence
β (λi − λj )2 β 2
det (1n2 + X ad ) = ∏ (1 − ). (9.193)
2πn 1≤i<j≤n 4π 2 n2
Using
θ2 sin θ
∏ (1 − 2 2
)= , (9.194)
n>0 π n θ
we get
2
sin 12 (λi − λj )β
∆(X) = ∏ ( 1
) . (9.195)
1≤i<j≤n 2 (λi − λj )β

As a consequence of (9.99) we further express (9.191) in terms of an integral over the

eigenvalues {λi } using
1 n 2 1 n 2
∫ d X→ ∫ d λ ∏ (λi − λj ) . (9.196)
VU (n) n! (2π)n 1≤i<j≤n

Using this in conjunction (9.195) in (9.191) gives finally

1
∫ d[A] f (A) = ∫ dµU (n) (βλ) f (iΛ) , (9.197)
β n2

with the measure for integration over U (n) determined by (9.137).

Although the freedom of constant gauge transformations has been used in transforming
X → Λ there is also a residual gauge freedom given by
2πr
g(t) = e2πirt/β 1 , r = 0, ±1, ±2, . . . ⇒ Λg(t) = Λ − 1. (9.198)
β

For this to be a symmetry for f (iX) = f (iΛ) we must have

f (iX) = fˆ(e−iβX ) , (9.199)

where fˆ is defined in terms of the line integral over t in (9.177). The final result (9.197)
shows that the functional integral over A(t) reduces after gauge fixing just to invariant
integration over the unitary matrix U .

243