group theory zipped!

world-wide quest to tame group theory

Predrag Cvitanović

Jan 13, 2023

 Overview
I have collected the course notes here, not so much for the notes themselves –they
cannot be understood on their own, without viewing the recorded live lectures– but for
the hyperlinks to various source texts you might find useful later on in your research.

Course outline (8:49 min)

◦ sect. 17 The epilogue: detailed overview of the course

Course policy (2:03 min)
0.1 Navigating the course (4:18 min)
Extras If a video is marked as an ‘extra’, it is not a required viewing, but a supplemen-
tary discussion or a comment or (sometimes) just a rant.
My teaching philosophy : Bologna (2:09 min)
About teaching online (3:15 min)
How does one pronounce ‘Euler’? ‘Cvitanović’? (4:17 min)
If I am allowed to teach group theory... (0:30 sec)

The fastest way to watch any week’s lecture videos is by letting YouTube run the
course playlist
and skipping ‘extras’ manually.
Contents

1 Linear algebra 7
Homework HW1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1 Week’s videos, reading . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Special projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Matrix-valued functions . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 A linear diversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . 14
1.6.1 Yes, but how do you really do it? . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
exercises 23

2 Finite groups - definitions 25

Homework HW2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1 Week’s videos, reading . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1.1 Don’t wonna know group theory . . . . . . . . . . . . . . . . 26
2.1.2 Finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Using group theory without knowing any . . . . . . . . . . . . . . . 28
2.4 Using symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Normal modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Group presentations . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7 Permutations in birdtracks . . . . . . . . . . . . . . . . . . . . . . . 30
2.8 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 32
2.9 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.10 What are cosets good for? (a discussion) . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
exercises 40

3 Group representations 43
Homework HW3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.1 Week’s videos, reading . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.1.1 Matrix representations, Schur’s Lemma . . . . . . . . . . . . 44
3.1.2 Wonderful Orthogonality Theorem . . . . . . . . . . . . . . . 44

3
 CONTENTS

3.2 It’s all about class . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.3 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.1 Hard work builds character . . . . . . . . . . . . . . . . . . . 45
3.3.2 History (optional) . . . . . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
exercises 48

4 Hard work builds character 49

Homework HW4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 51
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
exercises 56

5 It takes class 59
Homework HW5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1 It’s all about class . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.1.1 Dirac characters, Burnside’s method (optional) . . . . . . . . 61
5.1.2 William G. Harter (optional) . . . . . . . . . . . . . . . . . . 62
5.2 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
exercises 65

6 For fundamentalists 67
Homework HW6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 68
6.2 Thoughts (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.3 ChaosBook notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.4 Chaotic 3-spring, integrable 3-vortex systems (optional) . . . . . . . . 73
6.5 Eigenfunctions (optional) . . . . . . . . . . . . . . . . . . . . . . . . 76
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
exercises 80

7 Discrete Fourier representation 83

Homework HW7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.1 Optional reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
exercises 85

8 Space groups 87
Homework HW8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 88
8.2 Thoughts (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.3 Space groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.3.1 Wallpaper groups . . . . . . . . . . . . . . . . . . . . . . . . 91
8.3.2 One-dimensional line groups . . . . . . . . . . . . . . . . . . 93
8.3.3 Time reversal symmetry . . . . . . . . . . . . . . . . . . . . 93

birdtracks.eu/course3 4 2020-12-23
CONTENTS

8.4 Elastodynamic equilibria of 2D solids (optional) . . . . . . . . . . . . 93

8.5 Literature, reflections (optional) . . . . . . . . . . . . . . . . . . . . 95
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
exercises 99

9 Continuous groups 101

Homework HW9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 103
9.2 Discussion (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.3 Continuous symmetries: unitary and orthogonal . . . . . . . . . . . . 106
9.4 Character orthogonality theorem . . . . . . . . . . . . . . . . . . . . 107
9.5 Reps of compact groups are fully reducible . . . . . . . . . . . . . . 107
9.5.1 Projection operators . . . . . . . . . . . . . . . . . . . . . . 108
9.5.2 Spectral decomposition . . . . . . . . . . . . . . . . . . . . . 110
9.6 Clebsch-Gordan coefficients . . . . . . . . . . . . . . . . . . . . . . 111
9.7 Irrelevancy of clebsches . . . . . . . . . . . . . . . . . . . . . . . . . 114
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
exercises 116

10 Lie groups, algebras 119

Homework HW10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
10.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 121
10.2 Discussion (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10.3 SU(2) Pauli matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.4 Two-modes SO(2)-equivariant flow . . . . . . . . . . . . . . . . . . 124
10.5 SO(3) character orthogonality (optional) . . . . . . . . . . . . . . . . 126
10.6 Linear algebra (optional) . . . . . . . . . . . . . . . . . . . . . . . . 127
10.7 Lie groups for pedestrians (optional) . . . . . . . . . . . . . . . . . . 129
10.7.1 Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
10.7.2 Infinitesimal transformations, Lie algebras . . . . . . . . . . . 132
10.8 Nobel Prize in Physics 2020 (optional) . . . . . . . . . . . . . . . . . 135
10.8.1 Quaternionic speculations . . . . . . . . . . . . . . . . . . . 136
10.9 What really happened (optional) . . . . . . . . . . . . . . . . . . . . 137
10.10Birdtracks - updated history . . . . . . . . . . . . . . . . . . . . . . . 138
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
exercises 140

11 SU(2) and SO(3) 143

Homework HW11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
11.1 Discussion (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . 144
11.1.1 Recap of the course, so far (optional) . . . . . . . . . . . . . 145
11.2 SU(2) and SO(3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
exercises 148

2020-12-23 5 birdtracks.eu/course3
 CONTENTS

12 Lorentz group; spin 151

Homework HW12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
12.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 152
12.2 Discussion (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . 153
12.3 Spinors and the Lorentz group . . . . . . . . . . . . . . . . . . . . . 153
12.4 Irreps of SO(n) (optional) . . . . . . . . . . . . . . . . . . . . . . . 155
12.5 SO(4) of the Kepler problem (optional) . . . . . . . . . . . . . . . . 156
12.5.1 Central force problems (optional) . . . . . . . . . . . . . . . 156
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
exercises 157

13 Simple Lie algebras; SU(3) 161

Homework HW13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
13.1 Group theory news (optional) . . . . . . . . . . . . . . . . . . . . . . 162
exercises 163

14 Flavor SU(3) 165

Homework HW14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
14.1 Isotropic quantum harmonic oscillator . . . . . . . . . . . . . . . . . 166
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
exercises 167

15 Many particle systems. Young tableaux 169

Homework HW15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
15.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 170
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
exercises 171

16 Wigner 3- and 6-j coefficients 173

Homework HW16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
16.1 Other sources (optional) . . . . . . . . . . . . . . . . . . . . . . . . 174
16.2 Gruppenpest (optional) . . . . . . . . . . . . . . . . . . . . . . . . . 175
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
exercises 176

17 An overview, and the epilogue 183

17.1 Student suggestions for improvements . . . . . . . . . . . . . . . . . 186

birdtracks.eu/course3 6 2020-12-23
 group theory - week 1

Linear algebra

Homework HW1

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 1.1 Trace-log of a matrix 4 points

Exercise 1.2 Stability, diagonal case 2 points
Exercise 1.3 The matrix square root 4 points

Total of 10 points = 100 % score.

7
 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Diagonalizing the matrix: that’s the key to the whole thing.

— Governor Arnold Schwarzenegger

2021-05-18 Predrag Week 1 Linear algebra

◦ Light purple text in this PDF is a live hyperlink. If you encounter a ChaosBook.org
web login: all copyright-protected references are on a password protected site.
What password? If you are a Georgia Tech student, I can help you with that.

◦ The fastest way to watch any week’s lecture videos is by letting YouTube run the

course playlist

Downside is that the playlist plays also all ‘extra’ videos - you can skip through
those, if you are short on time. Or patience.

1.1 Week’s videos, reading

◦ Sect. 1.4 Matrix-valued functions

Matrices : 2 kinds
Derivative of a matrix function
Exponential, logarithm of a matrix
Determinant is a volume
log det = tr log (updated Aug 18, 2020)
Multi-matrix functions (optional, for the QM inclined)

• Sect. 1.5 A linear diversion

There are two representations of exponential of constant matrix, the Taylor series
(1.7) and the compound interest (Euler product) formulas (1.8).

Linear differential equations

• Sect. 1.6 Eigenvalues and eigenvectors

Hamilton-Cayley equation, projection operators (1.27), any matrix function is
evaluated by spectral decomposition (1.30). Work through example 1.5.

Eigenvalues and eigenvectors

What’s the deal with Hamilton-Cayley?
Spectral decomposition
Spectral decomposition and completeness
Right, left eigenvectors
A projection operators workout

birdtracks.eu/course3 week1 8 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

1.2 Other sources (optional)

Mopping up operations are the activities that engage most sci-
entists throughout their careers.
— Thomas Kuhn, The Structure of Scientific Revolutions

The subject of linear algebra is a vast and very alive research area, generates innumer-
able tomes of its own, and is way beyond what we can exhaustively cover here. Some
linear operators and matrices reading (optional reading for week 1, not required for
this course; whenever the text is colored, you can click on the live hyperlink in the pdf
version of these notes):

Stone and Goldbart [15], Mathematics for Physics: A Guided Tour for Graduate
Students, Appendix A. This is an advanced summary where you will find almost
everything one needs to know.

AWH p. 113 Functions of Matrices. Anything prefixed by AWH, like “Kro-

necker product AWH eq. (2.55)” refers to the more pedestrian and perhaps eas-
ier to read Arfken, Weber & Harris [2] Mathematical Methods for Physicists: A
Comprehensive Guide (Georgia Tech students can get it from GaTech Library).

AWH Section 2.2 Matrices

AWH Example 2.2.6 Exponential of a diagonal matrix

AWH Chapter 2 Determinants and matrices (click here).

AWH Section 6.1 Eigenvalue Equations (click here).

AWH Chapter 6 Eigenvalue problems (click here).

In sect. 1.4 I make matrix functions appear easier than they really are. For an in-
exercise 1.3
depth discussion, consult Golub and Van Loan [8] Matrix Computations, chap. 9
Functions of Matrices (click here).

Much more than you ever wanted to know about linear algebra: Axler [3] Down
with determinants! (click here).

Steve Trettel Linear Algebra and the Periodic Table is a gentle 53 min tour from
vectors to function spaces to quantum mechanics. True, what they teach you
as QM is 95% linear algebra, but Trettel does not mention that QM is 95% one
amazing experimental fact: ~ is a nature-given constant. Mathematicians...

David Austin online PreTeXt textbook Understanding Linear Algebra.

Grant Sanderson Essence of linear algebra ( 3Blue1Brown). Karan Shah likes

the geometrical explanations of linear algebra eigen-values / -vectors, recom-
mends it.

2021-07-29 9 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Question 1.1. Henriette Roux finds course notes confusing

Q Couldn’t you use one single, definitive text for methods taught in the course?
A It’s a grad school, so it is research focused - I myself am (re)learning the topics that we are
going through the course, using various sources. My emphasis in this course is on understanding
and meaning, not on getting all signs and 2π’s right, and I find reading about the topic from
several perspectives helpful. But if you really find one book more comfortable, nearly all topics
are covered in Arfken, Weber & Harris [2].

1.3 Special projects

Several people have been interested in taking a special project, instead of the final in
the course. If you propose to work out in detail some group-theory needed for your
own research (but you have not taken the time to master the theory), that would be
ideal. Random examples of interesting topics (i.e., something that Predrag would like
to learn from you:) –

1. The talk by David Weitz on melting of crystal lattices. Can you do a calculation
on a Wigner lattice or a graphene, or a silicon carbide polytype used as a substrate
in our graphene lab (ask Claire Berger about it), using our group theory methods
as applied to space groups (2- or 3-D lattices)?

2. If you are really wild about string theory, then you can read Giles and Thorn [7]
Lattice approach to string theory, and write up what you have learned as the
project report. The Giles-Thorn (GT) discretization of the worldsheet begins
with a representation of the free closed or open string propagator as a light-
cone worldsheet path integral defined on a lattice. The sequel Papathanasiou and
Thorn [14] Worldsheet propagator on the lightcone worldsheet lattice gives in
Appendix B 2D lattice Neumann open string, Dirichlet open string, and closed
string propagators. Discrete Green’s functions are explained, for example, by
Chung and Yau [4] who give explicitly, in their Theorem 6, a 2-dimensional lat-
tice Green’s function for a rectangular region R[`1 ×`2 ] . The paper is cited over
100 times, maybe there is a better, more up-to-date one to read in that list.

3. 3-springs system of sect. 6.4.

I recommend that you take a final, as these are hard and time-consuming projects,
and the faculty does not want to overburden you with course work. However, if a
project dovetails with your research interests, it might be worth it. Fly it by me.

1.4 Matrix-valued functions

What is a matrix?
—Werner Heisenberg (1925)
What is the matrix?
—-Keanu Reeves (1999)

birdtracks.eu/course3 week1 10 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Why should a working physicist care about linear algebra? Physicists were bliss-
fully ignorant of group theory until 1920’s, but with Heisenberg’s sojourn in Helgoland,
everything changed. Quantum Mechanics was formulated as
i
φ(t) = Û t φ(0) , Û t = e− ~ tĤ , (1.1)

where φ(t) is the quantum wave function t, Û t is the unitary quantum evolution opera-
tor, and Ĥ is the Hamiltonian operator. Fine, but what does this equation mean? In the
first lecture we deconstruct it.
The matrices that have to be evaluated are very high-dimensional, in principle in-
finite dimensional, and the numerical challenges can quickly get out of hand. What
made it possible to solve these equations analytically in 1920’s for a few iconic prob-
lems, such as the hydrogen atom, are the symmetries, or in other words group theory,
which start sketching out in the second lecture (and fill in the details in the next 27
lectures).
Whenever you are confused about an “operator”, think “matrix”. Here we recapit-
ulate a few matrix algebra concepts
Q that we found essential. The punch line is (1.35):
Hamilton-Cayley equation (M − λi 1) = 0 associates with each distinct root λi of a
matrix M a projection onto ith vector subspace
Y M − λj 1
Pi = .
λi − λj
j6=i

What follows - for this week - is a jumble of Predrag’s notes. If you understand the
examples, we are on the roll. If not, ask :)
How are we to think of the quantum operator

Ĥ = T̂ + V̂ , T̂ = p̂2 /2m , V̂ = V (q̂) , (1.2)

corresponding to a classical Hamiltonian H = T + V , where T is kinetic energy, and

V is the potential?
Expressed in terms of basis functions, the quantum evolution operator is an infinite-
dimensional matrix; if we happen to know the eigenbasis of the Hamiltonian, the prob-
lem is solved already. In real life we have to guess that some complete basis set is
good starting point for solving the problem, and go from there. In practice we truncate
such operator representations to finite-dimensional matrices, so it pays to recapitulate
a few relevant facts about matrix algebra and some of the properties of functions of
finite-dimensional matrices.

Matrix derivatives. The derivative of a matrix is a matrix with elements

dA(x) d
A0 (x) = , A0ij (x) = Aij (x) . (1.3)
dx dx
Derivatives of products of matrices are evaluated by the chain rule
d dA dB
(AB) = B+A . (1.4)
dx dx dx

2021-07-29 11 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

A matrix and its derivative matrix in general do not commute

d 2 dA dA
A = A+A . (1.5)
dx dx dx
d −1
The derivative of the inverse of a matrix, if the inverse exists, follows from dx (AA ) =
0:
d −1 1 dA 1
A =− . (1.6)
dx A dx A

Matrix functions. A function of a single variable that can be expressed in terms of

additions and multiplications generalizes to a matrix-valued function by replacing the
variable by a matrix.
In particular, the exponential of a constant matrix can be defined either by its series
expansion, or as a limit of an infinite product:
∞
X 1 k
eA = A , A0 = 1 (1.7)
k!
k=0
 N
1
= lim 1 + A (1.8)
N →∞ N

The first equation follows from the second one by the binomial theorem, so these in-
deed are equivalent definitions. That the terms of order O(N −2 ) or smaller do not
matter for a function of a single variable follows from the bound
 N  N  N
x− x + δxN x+
1+ < 1+ < 1+ ,
N N N

where |δxN | < . If lim δxN → 0 as N → ∞, the extra terms do not contribute. A
proof for matrices would probably require defining the norm of a matrix (and, more
generally, a norm of an operator acting on a Banach space) first. If you know an easy
proof, let us know.

Logarithm of a matrix. The logarithm of a matrix is defined by the power series

∞
X Ak
ln(1 − A) = − . (1.9)
k
k=1

log det = tr log matrix identity. Consider now the determinant

N
det (eA ) = lim (det (1 + A/N )) .
N →∞

To the leading order in 1/N

1
det (1 + A/N ) = 1 + tr A + O(N −2 ) .
N

birdtracks.eu/course3 week1 12 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

hence
 N  N
1 tr A
det e A
= lim −2
1 + tr A + O(N ) = lim 1 + = etr A (1.10)
N →∞ N N →∞ N

Defining M = eA we can write this as

ln det M = tr ln M , (1.11)

a crazy useful identity that you will run into over and over again.

Functions of several matrices. Due to non-commutativity of matrices, generaliza-

tion of a function of several variables to a function of several matrices is not as straight-
forward. Expression involving several matrices depend on their commutation relations.
For example, the Baker-Campbell-Hausdorff commutator expansion

t2 t3
etA Be−tA = B + t[A, B] + [A, [A, B]] + [A, [A, [A, B]]] + · · · (1.12)
2 3!
sometimes used to establish the equivalence of the Heisenberg and Schrödinger pic-
tures of quantum mechanics, follows by recursive evaluation of t derivatives
d
etA Be−tA = etA [A, B]e−tA .


dt
Expanding exp(A + B), exp A, exp B to first few orders using (1.7) yields
1
e(A+B)/N = eA/N eB/N − [A, B] + O(N −3 ) , (1.13)
2N 2
and the Trotter product formula: if B, C and A = B + C are matrices, then
 N
eA = lim eB/N eC/N (1.14)
N →∞

In particular, we can now make sense of the quantum evolution operator (1.1) as a
succession of short free flights (kinetic term) interspersed by small acceleration kicks
(potential term),
 N
e−itĤ = lim e−i∆t T̂ e−i∆t V̂ , ∆t = t/N , (1.15)
N →∞

where we have set ~ = 1.

1.5 A linear diversion

Linear is good, nonlinear is bad.
—Jean Bellissard

(Notes based of ChaosBook.org/chapters/stability.pdf)

2021-07-29 13 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Linear fields are the simplest vector fields, described by linear differential equations
which can be solved explicitly, with solutions that are good for all times. The state
space for linear differential equations is M = Rd , and the equations of motion are
written in terms of a vector x and a constant stability matrix A as
ẋ = v(x) = Ax . (1.16)
Solving this equation means finding the state space trajectory
x(t) = (x1 (t), x2 (t), . . . , xd (t))
passing through a given initial point x0 . If x(t) is a solution with x(0) = x0 and
y(t) another solution with y(0) = y0 , then the linear combination ax(t) + by(t) with
a, b ∈ R is also a solution, but now starting at the point ax0 + by0 . At any instant in
time, the space of solutions is a d-dimensional vector space, spanned by a basis of d
linearly independent solutions.
How do we solve the linear differential equation (1.16)? If instead of a matrix
equation we have a scalar one, ẋ = λx , the solution is x(t) = etλ x0 . In order to solve
the d-dimensional matrix case, it is helpful to rederive this solution by studying what
happens for a short time step ∆t. If time t = 0 coincides with position x(0), then
x(∆t) − x(0)
= λx(0) , (1.17)
∆t
which we iterate m times to obtain Euler’s formula for compounding interest
 m
t
x(t) ≈ 1 + λ x(0) ≈ etλ x(0) . (1.18)
m
The term in parentheses acts on the initial condition x(0) and evolves it to x(t) by
taking m small time steps ∆t = t/m. As m → ∞, the term in parentheses converges
to etλ . Consider now the matrix version of equation (1.17):
x(∆t) − x(0)
= Ax(0) . (1.19)
∆t
A representative point x is now a vector in Rd acted on by the matrix A, as in (1.16).
Denoting by 1 the identity matrix, and repeating the steps (1.17) and (1.18) we obtain
Euler’s formula for the exponential of a matrix:
 m
t
x(t) = J t x(0) , J t = etA = lim 1 + A , (1.20)
m→∞ m
where J t = J(t) is a short hand for exp(tA).

1.6 Eigenvalues and eigenvectors

10. Try to leave out the part that readers tend to skip.
— Elmore Leonard’s Ten Rules of Writing.

birdtracks.eu/course3 week1 14 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Eigenvalues of a [d×d] matrix M are the roots of its characteristic polynomial

Y
det (M − λ1) = (λi − λ) = 0 . (1.21)

Given a nonsingular matrix M, with all λi 6= 0, acting on d-dimensional vectors x, we

would like to determine eigenvectors e(i) of M on which M acts by scalar multiplica-
tion by eigenvalue λi
M e(i) = λi e(i) . (1.22)
If λi 6= λj , e(i) and e(j) are linearly independent. There are at most d distinct eigen-
values which we order by their real parts, Re λi ≥ Re λi+1 .
If all eigenvalues are distinct, e(j) are d linearly independent vectors which can be
used as a (non-orthogonal) basis for any d-dimensional vector x ∈ Rd

x = x1 e(1) + x2 e(2) + · · · + xd e(d) . (1.23)

However, r, the number of distinct eigenvalues, is in general smaller than the dimension
of the matrix, r ≤ d (see example 1.3).
From (1.22) it follows that

(M − λi 1) e(j) = (λj − λi ) e(j) ,

matrix (M − λi 1) annihilates e(i) , thus the product of all such factors annihilates any
vector, and the matrix M satisfies its characteristic equation
d
Y
(M − λi 1) = 0 . (1.24)
i=1

This humble fact has a name: the Hamilton-Cayley theorem. If we delete one term from
this product, we find that the remainder projects x from (1.23) onto the corresponding
eigenspace: Y Y
(M − λj 1)x = (λi − λj ) xi e(i) .
j6=i j6=i

Dividing through by the (λi − λj ) factors yields the projection operators

Y M − λj 1
Pi = , (1.25)
λi − λj
j6=i

which are orthogonal and complete:

r
X
Pi Pj = δij Pj , (no sum on j) , Pi = 1 , (1.26)
i=1

with the dimension of the ith subspace given by di = tr Pi . For each distinct eigen-
value λi of M,
(M − λj 1)Pj = Pj (M − λj 1) = 0 , (1.27)

2021-07-29 15 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

the colums/rows of Pj are the right/left eigenvectors e(j) , e(j) of M which (provided
M is not of Jordan type, see example 1.3) span the corresponding linearized subspace.
The main take-home is that once the distinct non-zero eigenvalues {λi } are com-
puted, projection operators are polynomials in M which need no further diagonaliza-
tions or orthogonalizations.
It follows from the characteristic equation (1.27) that λi is the eigenvalue of M on
Pi subspace:
M Pi = λi Pi (no sum on i) . (1.28)
Using M = M 1 and completeness relation (1.26) we can rewrite M as

M = λ1 P1 + λ2 P2 + · · · + λd Pd . (1.29)

Any matrix function f (M) takes the scalar value f (λi ) on the Pi subspace, f (M) Pi =
f (λi ) Pi , and is thus easily evaluated through its spectral decomposition
X
f (M) = f (λi )Pi . (1.30)
i

This, of course, is the reason why anyone but a fool works with irreducible reps: they
reduce matrix (AKA “operator”) evaluations to manipulations with numbers.
By (1.27) every column of Pi is proportional to a right eigenvector e(i) , and its
every row to a left eigenvector e(i) . In general, neither set is orthogonal, but by the
idempotence condition (1.26), they are mutually orthogonal,

e(i) · e(j) = cj δij . (1.31)

The non-zero constant cj is convention dependent and not worth fixing, unless you feel
nostalgic about Clebsch-Gordan coefficients. We shall set cj = 1. Then it is convenient
to collect all left and right eigenvectors into a single matrix.

Example 1.1. Linear stability of 2-dimensional flows: For a 2-dimensional flow

the eigenvalues λ1 , λ2 of A are either real, leading to a linear motion along their eigen-
vectors, xj (t) = xj (0) exp(tλj ), or form a complex conjugate pair λ1 = µ + iω , λ2 =
µ − iω , leading to a circular or spiral motion in the [x1 , x2 ] plane, see example 1.2.

Figure 1.1: Streamlines for several typical 2-

dimensional flows: saddle (hyperbolic), in node
(attracting), center (elliptic), in spiral.

birdtracks.eu/course3 week1 16 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

saddle out node in node

6 6 6
× ×- ××- ×× -

center out spiral in spiral

×6 6× × 6
- - -
× × ×

Figure 1.2: Qualitatively distinct types of exponents {λ(1) , λ(2) } of a [2×2] Jacobian
matrix. Here the eigenvalues of the Jacobian matrix are multipliers Λ(j) , and the expo-
nents are defined as the deformation rates λ(j) = log(Λ(j) )/t.

These two possibilities are refined further into sub-cases depending on the signs of
the real part. In the case of real λ1 > 0, λ2 < 0, x1 grows exponentially with time, and x2
contracts exponentially. This behavior, called a saddle, is sketched in figure 1.1, as are
the remaining possibilities: in/out nodes, inward/outward spirals, and the center. The
magnitude of out-spiral |x(t)| diverges exponentially when µ > 0, and in-spiral contracts
into (0, 0) when µ < 0; whereas, the phase velocity ω controls its oscillations.
If eigenvalues λ1 = λ2 = λ are degenerate, the matrix might have two linearly
independent eigenvectors, or only one eigenvector, see example 1.3. We distinguish
two cases: (a) A can be brought to diagonal form and (b) A can be brought to Jordan
form, which (in dimension 2 or higher) has zeros everywhere except for the repeating
eigenvalues on the diagonal and some 1’s directly above it. For every such Jordan
[dα ×dα ] block there is only one eigenvector per block.
We sketch the full set of possibilities in figures 1.1 and 1.2.

Example 1.2. Complex eigenvalues: in-out spirals. As M has only real entries, it
will in general have either real eigenvalues, or complex conjugate pairs of eigenvalues.
Also the corresponding eigenvectors can be either real or complex. All coordinates used
in defining a dynamical flow are real numbers, so what is the meaning of a complex
eigenvector?
If λk , λk+1 eigenvalues that lie within a diagonal [2 × 2] sub-block M0 ⊂ M form
a complex conjugate pair, {λk , λk+1 } = {µ + iω, µ − iω}, the corresponding com-
plex eigenvectors can be replaced by their real and imaginary parts, {e(k) , e(k+1) } →
{Re e(k) , Im e(k) }. In this 2-dimensional real representation, M0 → A, the block A is a
sum of the rescaling×identity and the generator of SO(2) rotations in the {Re e(1) , Im e(1) }
plane.      
µ −ω 1 0 0 −1
A= =µ +ω .
ω µ 0 1 1 0
Trajectories of ẋ = A x, given by x(t) = J t x(0), where (omitting e(3) , e(4) , · · · eigen-
directions)  
cos ωt − sin ωt
J t = etA = etµ , (1.32)
sin ωt cos ωt
spiral in/out around (x, y) = (0, 0), see figure 1.1, with the rotation period T and the
radial expansion /contraction multiplier along the e(j) eigen-direction per a turn of the

2021-07-29 17 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

spiral:
exercise 1.4
T = 2π/ω , Λradial = eTµ . (1.33)
We learn that the typical turnover time scale in the neighborhood of the equilibrium
(x, y) = (0, 0) is of order ≈ T (and not, let us say, 1000 T, or 10−2 T).

Example 1.3. Degenerate eigenvalues. While for a matrix with generic real
elements all eigenvalues are distinct with probability 1, that is not true in presence of
symmetries, or spacial parameter values (bifurcation points). What can one say about
situation where dα eigenvalues are degenerate, λα = λi = λi+1 = · · · = λi+dα −1 ?
Hamilton-Cayley (1.24) now takes form
r
Y X
(M − λα 1)dα = 0 , dα = d . (1.34)
α=1 α

We distinguish two cases:

M can be brought to diagonal form. The characteristic equation (1.34) can be re-
placed by the minimal polynomial,
r
Y
(M − λα 1) = 0 , (1.35)
α=1

where the product includes each distinct eigenvalue only once. Matrix M acts multi-
plicatively
M e(α,k) = λi e(α,k) , (1.36)
on a dα -dimensional subspace spanned by a linearly independent set of basis eigen-
vectors {e(α,1) , e(α,2) , · · · , e(α,dα ) }. This is the easy case. Luckily, if the degeneracy is
due to a finite or compact symmetry group, relevant M matrices can always be brought
to such diagonalizable form.

M can only be brought to upper-triangular, Jordan form. This is the messy case,
so we only illustrate the key idea in example 1.4.

Example 1.4. Decomposition of 2-dimensional vector spaces: Enumeration of ev-

ery possible kind of linear algebra eigenvalue / eigenvector combination is beyond what
we can reasonably undertake here. However, enumerating solutions for the simplest
case, a general [2×2] non-singular matrix
 
M11 M12
M= .
M21 M22

takes us a long way toward developing intuition about arbitrary finite-dimensional matri-
ces. The eigenvalues
1 1p
λ1,2 = tr M ± (tr M)2 − 4 det M (1.37)
2 2
are the roots of the characteristic (secular) equation (1.21):

det (M − λ 1) = (λ1 − λ)(λ2 − λ)

= λ2 − tr M λ + det M = 0 .

birdtracks.eu/course3 week1 18 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Distinct eigenvalues case has already been described in full generality. The left/right
eigenvectors are the rows/columns of projection operators (see example 1.5)

M − λ2 1 M − λ1 1
P1 = , P2 = , λ1 6= λ2 . (1.38)
λ1 − λ2 λ2 − λ1

Degenerate eigenvalues. If λ1 = λ2 = λ, we distinguish two cases: (a) M can be

brought to diagonal form. This is the easy case. (b) M can be brought to Jordan form,
with zeros everywhere except for the diagonal, and some 1’s directly above it; for a [2×2]
matrix the Jordan form is
     
λ 1 1 0
M= , e(1) = , v(2) = .
0 λ 0 1

v(2) helps span the 2-dimensional space, (M − λ)2 v(2) = 0, but is not an eigenvector,
as Mv(2) = λv(2) + e(1) . For every such Jordan [dα × dα ] block there is only one
eigenvector per block. Noting that
 m
mλm−1

λ
Mm = ,
0 λm

we see that instead of acting multiplicatively on R2 , Jacobian matrix J t = exp(tM)

! !
tM u tλ u + tv
e =e (1.39)
v v

picks up a power-low correction. That spells trouble (logarithmic term ln t if we bring the
extra term into the exponent).

Example 1.5. Projection operator decomposition in 2 dimensions: Let’s illustrate

how the distinct eigenvalues case works with the [2×2] matrix [11]
 
4 1
M= .
3 2

Its eigenvalues {λ1 , λ2 } = {5, 1} are the roots (1.37):

det (M − λ1) = λ2 − 6 λ + 5 = (5 − λ)(1 − λ) = 0 .

That M satisfies its secular equation (Hamilton-Cayley theorem) can be verified by ex-
plicit calculation:
 2      
4 1 4 1 1 0 0 0
−6 +5 = .
3 2 3 2 0 1 0 0

Associated with each root λi is the projection operator (1.38)

 
1 1 3 1
P1 = (M − 1) = (1.40)
4 4 3 1
 
1 1 1 −1
P2 = − (M − 5 · 1) = . (1.41)
4 4 −3 3

2021-07-29 19 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Matrices Pi are orthonormal and complete. The dimension of the ith subspace is given
by di = tr Pi ; in case at hand both subspaces are 1-dimensional. From the charac-
teristic equation it follows that Pi satisfies the eigenvalue equation M Pi = λi Pi . Two
consequences are immediate. First, we can easily evaluate any function of M by spec-
tral decomposition, for example
 
58591 19531
M7 − 3 · 1 = (57 − 3)P1 + (1 − 3)P2 = .
58593 19529
Second, as Pi satisfies the eigenvalue equation, its every column is a right eigenvector,
and every row a left eigenvector. Picking first row/column we get the eigenvectors:
   
1 1
{e(1) , e(2) } = { , }
1 −3
   
3 1
{e(1) , e(2) } = { , },
1 −1
with overall scale arbitrary. The matrix is not symmetric, so {e(j) } do not form an orthog-
onal basis. The left-right eigenvector dot products e(j) · e(k) , however, are orthogonal
as in (1.31), by inspection. (Continued in example ??.)
Example 1.6. Computing matrix exponentials. If A is diagonal (the system is un-
coupled), then etA is given by
  λ1 t
e
 
λ1 t
λ t
 λ2 t   e 2 
exp  = .
   
 . ..   . .. 
λd t
λd t e
If A is diagonalizable, A = F DF −1 , where D is the diagonal matrix of the eigen-
values of A and F is the matrix of corresponding eigenvectors, the result is simple:
An = (F DF −1 )(F DF −1 ) . . . (F DF −1 ) = F Dn F −1 . Inserting this into the Taylor se-
ries for ex gives eAt = F eDt F −1 .
But A may not have d linearly independant eigenvectors, making F singular and
forcing us to take a different route. To illustrate this, consider [2×2] matrices. For any
linear system in R2 , there is a similarity transformation
B = U −1 AU ,
where the columns of U consist of the generalized eigenvectors of A such that B has
one of the following forms:
     
λ 0 λ 1 µ −ω
B= , B= , B= .
0 µ 0 λ ω µ
These three cases, called normal forms, correspond to A having (1) distinct real eigen-
values, (2) degenerate real eigenvalues, or (3) a complex pair of eigenvalues. It follows
that
 λt     
Bt e 0 Bt λt 1 t Bt at cos bt − sin bt
e = , e =e , e =e ,
0 eµt 0 1 sin bt cos bt
and eAt = U eBt U −1 . What we have done is classify all [2×2] matrices as belonging to
one of three classes of geometrical transformations. The first case is scaling, the second
is a shear, and the third is a combination of rotation and scaling. The generalization of
these normal forms to Rd is called the Jordan normal form. (J. Halcrow)

birdtracks.eu/course3 week1 20 2021-07-29

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

1.6.1 Yes, but how do you really do it?

As M has only real entries, it will in general have either real eigenvalues (over-damped
oscillator, for example), or complex conjugate pairs of eigenvalues (under-damped os-
cillator, for example). That is not surprising, but also the corresponding eigenvectors
can be either real or complex. All coordinates used in defining the flow are real num-
bers, so what is the meaning of a complex eigenvector?
Due to the reality of M, complex eigenvalues form complex conjugate pairs,
{λk , λk+1 } = {µ + iω, µ − iω} ,
and the sum of terms in the spectral decomposition (1.29) of M is real,
M = · · · + (µk + iωk )Pk + (µk − iωk )Pk+1 + · · ·
= · · · + µk Rk + ωk Qk + · · · , (1.42)
where Rk = Pk + Pk+1 and Qk = i(Pk − Pk+1 ) are matrices with real elements.
1
Pk = (R + iQ) , Pk+1 = P∗k ,
2
Any matrix function f (M) takes the scalar value f (λi ) on the Pi subspace, f (M) Pi =
f (λi ) Pi , and is thus easily evaluated through its
they are in a sense degenerate:
is also a projection operator pair, but this time
If two eigenvalues form a complex conjugate pair, {λk , λk+1 } = {µ + iω, µ − iω},
they are in a sense degenerate: while a real λk characterizes a motion along a line, a
complex λk characterizes a spiralling motion in a plane. We determine this plane by
replacing the corresponding complex eigenvectors by their real and imaginary parts,
{e(k) , e(k+1) } → {Re e(k) , Im e(k) }, or, in terms of projection operators: Substitution
    
Pk 1 1 i R
= ,
Pk+1 2 1 −i Q
suggest introduction of a det U = 1, special unitary matrix
eiπ/2 1
 
i
U= √ (1.43)
2 1 −i
which brings the λk Pk + λk+1 Pk+1 complex eigenvalue pair in the spectral decom-
position into the real form,
   
> µ + iω 0 µ −ω
U U= .
0 µ − iω ω µ
     
λ 0 Pk µ −ω R
[Pk , Pk+1 ] = [R, Q] , (1.44)
0 λ∗ Pk+1 ω µ Q
where we have dropped the superscript (k) for notational brevity.
To summarize, spectrally decomposed matrix M acts along lines on subspaces cor-
responding to real eigenvalues, and as a [2×2] rotation in a plane on subspaces corre-
sponding to complex eigenvalue pairs.

2021-07-29 21 birdtracks.eu/course3 week1

 GROUP THEORY - WEEK 1. LINEAR ALGEBRA

Commentary
Remark 1.1. Projection operators. The construction of projection operators given in
sect. 1.6.1 is taken from refs. [5, 6]. Who wrote this down first we do not know, lineage cer-
tainly goes all the way back to Lagrange polynomials [13], but projection operators tend to get
drowned in sea of algebraic details. Arfken and Weber [1] ascribe spectral decomposition (1.30)
to Sylvester. Halmos [9] is a good early reference - but we like Harter’s exposition [10–12] best,
for its multitude of specific examples and physical illustrations. In particular, by the time we
get to (1.27) we have tacitly assumed full diagonalizability of matrix M. That is the case for
the compact groups we will study here (they are all subgroups of U(n)) but not necessarily in
other applications. A bit of what happens then (nilpotent blocks) is touched upon in example 1.4.
Harter in his lecture Harter’s lecture 5 (starts about min. 31 into the lecture) explains this in great
detail - its well worth your time.

References
[1] G. B. Arfken and H. J. Weber, Mathematical Methods for Physicists: A Compre-
hensive Guide, 6th ed. (Academic, New York, 2005).
[2] G. B. Arfken, H. J. Weber, and F. E. Harris, Mathematical Methods for Physi-
cists: A Comprehensive Guide, 7th ed. (Academic, New York, 2013).
[3] S. Axler, “Down with determinants!”, Amer. Math. Monthly 102, 139–154 (1995).
[4] F. Chung and S.-T. Yau, “Discrete Green’s functions”, J. Combin. Theory A 91,
191–214 (2000).
[5] P. Cvitanović, “Group theory for Feynman diagrams in non-Abelian gauge the-
ories”, Phys. Rev. D 14, 1536–1553 (1976).
[6] P. Cvitanović, Classical and exceptional Lie algebras as invariance algebras, Ox-
ford Univ. preprint 40/77, unpublished., 1977.
[7] R. Giles and C. B. Thorn, “Lattice approach to string theory”, Phys. Rev. D 16,
366–386 (1977).
[8] G. H. Golub and C. F. Van Loan, Matrix Computations, 4th ed. (J. Hopkins Univ.
Press, Baltimore, MD, 2013).
[9] P. R. Halmos, Finite-Dimensional Vector Spaces (Princeton Univ. Press, Prince-
ton NJ, 1948).
[10] W. G. Harter, “Algebraic theory of ray representations of finite groups”, J. Math.
Phys. 10, 739–752 (1969).
[11] W. G. Harter, Principles of Symmetry, Dynamics, and Spectroscopy (Wiley, New
York, 1993).
[12] W. G. Harter and N. dos Santos, “Double-group theory on the half-shell and the
two-level system. I. Rotation and half-integral spin states”, Amer. J. Phys. 46,
251–263 (1978).
[13] K. Hoffman and R. Kunze, Linear Algebra, 2nd ed. (Prentice-Hall, Englewood
Cliffs NJ, 1971).

birdtracks.eu/course3 week1 22 2021-07-29

EXERCISES

[14] G. Papathanasiou and C. B. Thorn, “Worldsheet propagator on the lightcone

worldsheet lattice”, Phys. Rev. D 87, 066005 (2013).
[15] M. Stone and P. Goldbart, Mathematics for Physics: A Guided Tour for Graduate
Students (Cambridge Univ. Press, Cambridge UK, 2009).

Exercises
1.1. Trace-log of a matrix. Prove that

det M = etr ln M .

for an arbitrary nonsingular finite dimensional matrix M , det M 6= 0.

1.2. Stability, diagonal case. Verify that for a diagonalizable matrix A the exponential is
also diagonalizable

J t = etA = U−1 etAD U , AD = UAU−1 . (1.45)

1.3. The matrix square root. Consider matrix

 
4 10
A= .
0 9

Generalize the square root function f (x) = x1/2 to a square root f (A) = A1/2 of a
matrix A.
a) Which one(s) of these is/are the square root of A
       
2 2 −2 10 −2 −2 2 −10
, , , ?
0 3 0 3 0 −3 0 −3

b) Assume that the eigenvalues of a [d × d] matrix are all distinct. How many square root
matrices does such matrix have?
c) Given a [2×2] matrix A with a distinct pair of eigenvalues {λ1 , λ2 }, write down a
formula that generates all square root matrices A1/2 . Hint: one can do this using the 2
projection operators associates with the matrix A. 2 points
1.4. Real representation of complex eigenvalues. (Verification of example 1.2.) λk , λk+1
eigenvalues form a complex conjugate pair, {λk , λk+1 } = {µ + iω, µ − iω}. Show that
(a) corresponding projection operators are complex conjugates of each other,

P = Pk , P∗ = Pk+1 ,

where we denote Pk by P for notational brevity.

(b) P can be written as
1
P= (R + iQ) ,
2
where R = Pk + Pk+1 and Q are matrices with real elements.

2021-07-29 23 birdtracks.eu/course3 week1

 EXERCISES

(c)     
Pk 1 1 i R
= .
Pk+1 2 1 −i Q
(d) The · · · + λk Pk + λ∗k Pk+1 + · · · complex eigenvalue pair in the spectral decom-
position (1.29) is now replaced by a real [2×2] matrix
  
µ −ω R
··· + + ···
ω µ Q

or whatever you find the clearest way to write this real representation.

birdtracks.eu/course3 week1 24 2021-07-29

 group theory - week 2

Finite groups - definitions

Homework HW2

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 2.1 Gx ⊂ G 1 point

Exercise 2.2 Transitivity of conjugation 1 point
Exercise 2.3 Isotropy subgroup of gx 1 points
Exercise 2.4 D3 : symmetries of an equilateral triangle 5 points
Exercise 2.5 C4 -invariant potential 7 (+2) points

Bonus points
Exercise 2.6 (a), (b) and (c) Permutation of three objects 2 points
Exercise 2.7 Three masses on a loop 6 points

Total of 15 points = 100 % score.

Extra points accumulate, can help you later if you miss a few problems.

25
 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

2.1 Week’s videos, reading

If I had had more time, I would have written less
— Blaise Pascal, a remark made to a correspondent

Please do not get intimated by the length of this week’s notes - they are here more
for me than for you, as notes on these topics for future reference. If you understand the
main sequence of video clips, and recommended reading, that should suffice to do the
problems. The rest is optional, you can quickly skim over...

2.1.1 Don’t wonna know group theory

The fastest way to watch any week’s lecture videos is by letting YouTube run the
Section playlist
◦ Sect. 2.4 Using symmetries
◦ Sect. 2.5 Normal modes: The free vibrations of systems, for undamped systems
with total energy conserved for which the frequencies of oscillation are real.
Normal modes (9:06 min)
◦ example 2.5 Vibrational spectra of molecules is taken from Gutkin lecture notes
example 5.1 Cn symmetry. The corresponding projection operators (1.25) are
worked out in example 2.6.
◦ Example 2.4 Vibrations of a classical CO2 molecule
A Hamiltonian with a symmetry (4:46 min)
CO2 molecule (4:07 min)
Projection operators (5:33 min)
(Anti)symmetric subspaces (3:04 min)
Zero mode (5:19 min)

2.1.2 Finite groups

Groups, permutations, rearrangement theorem, subgroups, cosets, classes,
all exemplified by the D3 ∼
= C3v ∼
= S3 symmetries of an equilateral trian-
gle.

Section playlist
Dresselhaus et al. [4] Chapter 1 Basic Mathematical Background: Introduction.
The MIT course 6.734 online version contains much of the same material.
ChaosBook Chapter 10. Flips, slides and turns

birdtracks.eu/course3 week2 26 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

Discrete symmetry, an example: 3-disk pinball (4:03 min)

What is a group? (10:56 min)

(extra) Discussion: There might be many examples of it, but a ‘group’ itself
is an abstract notion. (3 min)
(extra) Discussion: permutations, symmetric group, simple groups, Italian
renaissance, French revolution, Galois (5:23 min)

by Socratica: (cannot add it to the section YouTube playlist)

a delightful introduction to group multiplication (or Cayley) tables. (7:32)

Active, passive coordinate transformations (3:08 min)

Following Mefisto: symmetry defined three (3) times (7:01 min)

Subgroups, classes, group orbits, reduced state space (7:57 min)

2.2 Other sources (optional)

Group theory and why I love 808,017,· · · ,000 is a great video on group theory
from 3Blue1Brown, writes Andrew Wu. I agree: Well worth of your time, more
motivational than my lectures. What it actually focuses on - the monster group -
is totally useless to us.

AWH Example 6.2.3 Degenerate eigenproblem.

AWH Example 6.5.2 Normal modes.

For a deep dive into this material, here is your rabbit hole.

For deeper insights, read Roger Penrose [7] (click here).

Nathan Carter Visual Group Theory (read it online through your university
library) seems very good. The next two online courses are based on it:

Dana Ernst An inquiry-based approach to abstract algebra.

Matt Macauley Twitter Group actions course and Modern Algebra course.

Tom Judson’s online Abstract Algebra: Theory and Applications.

AWH Chapter 17 Group Theory (click here).

For a typical (but for this course advanced) application see, for example, Stone
and Goldbart [10], Mathematics for Physics: A Guided Tour for Graduate Stu-
dents, Section 14.3.2 Vibrational spectrum of H2 O (click here).

◦ Glance through sect. 2.6 Group presentations and sect. 2.8 Literature, but I do
not expect you to understand this material.

2021-10-30 27 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

Discussion 4 - Homework (3 min)

There is no need to learn all these “Greek” words.

◦ If instead, bedside crocheting is your thing, click here.

2.3 Using group theory without knowing any

It’s a matter of no small pride for a card-carrying dirt physics theorist to claim full and
total ignorance of group theory (read ChaosBook Appendix A.6 Gruppenpest). So what
we will do first is work out a few examples of physical applications of group theory
that you already know without knowing that you have been using “Group Theory.”

2.4 Using symmetries

Tyger Tyger burning bright,
In the forests of the night:
What immortal hand or eye,
Dare frame thy fearful symmetry?
—William Blake, The Tyger

The big idea #1 of this is week is symmetry.

If our physical problem is defined by a (perhaps complicated) Hamiltonian H, another

matrix M (hopefully a very simple matrix) is a symmetry if it commutes with the
Hamiltonian
[M, H] = 0 . (2.1)

Than we can use the spectral decomposition (1.30) of M to block-diagonalize H into

a sum of lower-dimensional sub-matrices,
X
H= Hi , Hi = Pi HPi , (2.2)
i

and thus significantly simplify the computation of eigenvalues and eigenvectors of H,

the matrix of physical interest.

2.5 Normal modes

The big idea #2 of this is week is: many body systems (molecules, neu-
ronal networks, ...) are ruled by collective modes, not individual particles
(atoms, neurons, ...).

birdtracks.eu/course3 week2 28 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

D3 e C C2 σ (1) σ (2) σ (3)

e e C C2 σ (1) σ (2) σ (3)
C C C2 e σ (3) σ (1) σ (2)
C2 C2 e C σ (2) σ (3) σ (1)
σ (1) σ (1) σ (2) σ (3) e C C2
σ (2) σ (2) σ (3) σ (1) C2 e C
σ (3) σ (3) σ (1) σ (2) C C2 e

Table 2.1: The dihedral group D3 group multiplication table. Actually, we prefer
cyclic and dihedral groups notation ‘rotations’ r` and ‘flips’ σm , as in table 4.1.

In the linear, harmonic oscillator approximation, the classical dynamics of a molecule

is governed by the Hamiltonian
N N
X mi 1 X >
H= ẋ2i + x Vij xj ,
i=1
2 2 i,j=1 i

where {xi } are small deviations from the equilibrium, resting points of the molecules
labelled i. Vij is a symmetric matrix, so it can be brought to a diagonal form by an
orthogonal transformation, to a set of N uncoupled harmonic oscillators or normal
modes of frequencies {ωi }.
N
X mi
ẏi2 + ωi2 yi2 .


x → y = U x, H= (2.3)
i=1
2

2.6 Group presentations

Group theory? It is all about class & character.
— Predrag Cvitanović, One minute elevator pitch

Group multiplication (or Cayley) tables, such as Table 2.1, define each distinct
discrete group, but they can be hard to digest. A Cayley graph, with links labeled
by generators, and the vertices corresponding to the group elements, has the same
information as the group multiplication table, but is often a more insightful presentation
of the group.
For example, the Cayley graph figure 2.1 is a clear presentation of the dihedral
group D4 of order 8,

D4 = (e, a, a2 , a3 , b, ba, ba2 , ba3 ) , generators a4 = e , b2 = e . (2.4)

Quaternion group is also of order 8, but with a distinct multiplication table / Cayley
graph, see figure 2.2. For more of such, see, for example, mathoverflow Cayley graph
discussion.

2021-10-30 29 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

Figure 2.1: A Cayley graph presentation of

the dihedral group D4 . The ‘root vertex’ of the
graph, marked e, is here indicated by the letter
F, the links are multiplications by two genera-
tors: a cyclic rotation by left-multiplication by
element a (directed red link), and the flip by b
(undirected blue link). The vertices are the 8
possible orientations of the transformed letter F.

Figure 2.2: A Cayley graph presentation of the

quaternion group Q8 . It is also of order 8, but
distinct from D4 .

2.7 Permutations in birdtracks

The text that follows is a very condensed extract of birdtracks.eu chapter 6 Permuta-
tions, from Group Theory - Birdtracks, Lie’s, and Exceptional Groups [3]. I am usually
reluctant to use birdtrack notations in front of graduate students indoctrinated by their
professors in the 1890’s tensor notation, but I’m emboldened by the very enjoyable ar-
ticle on The new language of mathematics by Dan Silver [9]. Your professor’s notation
is as convenient for actual calculations as -let’s say- long division using roman numer-
als. So leave them wallowing in their early progressive rock of 1968, King Crimsons
of their youth. You chill to beats younger than Windows 98, to grime, to trap, to
hardvapour, to birdtracks.

In 1937 R. Brauer [2] introduced diagrammatic notation for the Kronecker δij op-
eration, in order to represent “Brauer algebra” permutations, index contractions, and
matrix multiplication diagrammatically. His equation (39)

(send index 1 to 2, 2 to 4, contract ingoing (3·4), outgoing (1·3)) is the earliest published
diagrammatic notation I know about. While in kindergarten (disclosure: we were too
poor to afford kindergarten) I sat out to revolutionize modern group theory [3]. But I

birdtracks.eu/course3 week2 30 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

suffered a terrible setback; in early 1970’s Roger Penrose pre-invented my “birdtracks,”

or diagrammatic notation, for symmetrization operators [6], Levi-Civita tensors [8],
and “strand networks” [5]. Here is a little flavor of how one birdtracks:
We can represent the operation of permuting indices (d “billiard ball labels,” tensors
with d indices) by a matrix with indices bunched together:
a a ...aq dp ...d1
σαβ = σb11...b
2
p
,cq ...c2 c1 . (2.5)

To draw this, Brauer style, it is convenient to turn his drawing on a side. For 2-index
tensors, there are two permutations:

identity: 1ab ,cd = δad δbc =

flip: σ(12)ab ,cd = δac δbd = . (2.6)

For 3-index tensors, there are six permutations:

1a1 a2 a3 ,b3 b2 b1 = δab11 δab22 δab33 =

σ(12)a1 a2 a3 ,b3 b2 b1 = δab21 δab12 δab33 =

σ(23) = , σ(13) =

σ(123) = , σ(132) = . (2.7)

Here group element labels refer to the standard permutation cycles notation. There is
really no need to indicate the “time direction" by arrows, so we omit them from now
on.
The symmetric sum of all permutations,
1 n b1 b2 o
Sa1 a2 ...ap ,bp ...b2 b1 = δa1 δa2 . . . δabpp + δab12 δab21 . . . δabpp + . . .
p!
 
1
S = = + + + ... , (2.8)
p!
...

...
...

...

yields the symmetrization operator S. In birdtrack notation, a white bar drawn across
p lines [6] will always denote symmetrization of the lines crossed. A factor of 1/p! has
been introduced in order for S to satisfy the projection operator normalization

S2 = S

= . (2.9)
...

...

You have already seen such “fully-symmetric representation,” in the discussion of

discrete Fourier transforms, ChaosBook Example A24.3 ‘Configuration-momentum’

2021-10-30 31 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

Fourier space duality, but you are not likely to recognize it. There the average was not
over all permutations, but the zero-th Fourier mode φ̃0 was the average over only cyclic
permutations. Every finite discrete group has such fully-symmetric representation, and
in statistical mechanics and quantum mechanics this is often the most important state
(the ‘ground’ state).
A subset of indices a1 , a2 , . . . aq , q < p can be symmetrized by symmetrization
matrix S12...q

(S12...q )a1 a2 ...aq ...ap ,bp ...bq ...b2 b1 =

1 n b1 b2 o
b +1
δ δ . . . δabqq + δab12 δab21 . . . δabqq + . . . δaqq +1 . . . δabpp
q! a1 a2
1
2

...

...
S12...q = q. (2.10)

...
Overall symmetrization also symmetrizes any subset of indices:

SS12...q = S

= . (2.11)
... ...

... ...
... ...

...

... ...

Any permutation has eigenvalue 1 on the symmetric tensor space:

σS = S

= . (2.12)
...
...

Diagrammatically this means that legs can be crossed and uncrossed at will.
One can construct a projection operator onto the fully antisymmetric space in a
similar manner [3]. Other representations are trickier - that’s precisely what the theory
of finite groups is about.

2.8 Other sources (optional)

The exposition (or the corresponding chapter in Tinkham [11]) that we follow here
largely comes from Wigner’s classic Group Theory and Its Application to the Quantum
Mechanics of Atomic Spectra [12], which is a harder going, but the more group theory
you learn the more you’ll appreciate it.
The structure of finite groups was understood by late 19th century. A full list of
finite groups was another matter. The complete proof of the classification of all finite
groups takes about 3 000 pages, a collective 40-years undertaking by over 100 mathe-
maticians, read the wiki. Not all finite groups are as simple or easy to figure out as D3 .
For example, the order of the Ree group 2 F4 (2)0 is 212(26 + 1)(24 − 1)(23 + 1)(2 −
1)/2 = 17 971 200 .

birdtracks.eu/course3 week2 32 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

From Emory Math Department: A pariah is real! The simple finite groups fit into
18 families, except for the 26 sporadic groups. 20 sporadic groups AKA the Happy
Family are parts of the Monster group. The remaining six loners are known as the
pariahs.
Hang in there! And relax. None of this will be on the test. As a matter of fact, there
will be no test.
Question 2.1. Henriette Roux asks
Q What did you do this weekend?
A The same as every other weekend - prepared week’s lecture, with my helpers Avi the Little,
Edvard the Nordman, and Malbec el Argentino, under Master Roger’s watchful eye, see here.

2.9 Examples
Example 2.1. Discrete symmetries in physics:
• Point groups i.e., subgroups of O(3).
• Point groups + discrete translations e.g., symmetry groups of crystals.
• Permutation groups
SΨ(x1 , x2 , . . . xn ) = Ψ(x2 , x1 , . . . xn ).

• Boson wave functions are symmetric while fermion wave functions are anti-symmetric
under exchange of variables.
(B. Gutkin)

Example 2.2. The group multiplication table for D3 : See table 4.1.

Example 2.3. Reflection and discrete rotation symmetries:

(a) Reflection symmetry V (x) = P V (x) = V (−x):
~2 ∂ 2
 
− + V (x) ψ(x) = En ψ(x) (2.13)
2m ∂x2
(see figure 2.3). If ψ(x) is solution then P ψ(x) is also solution. From this and non-
degeneracy of the spectrum follows that either P ψ(x) = ψ(x) or P ψ(x) = −ψ(x).
The first case corresponds to symmetric functions while the second one to anti-
symmetric one. Thus the whole spectrum can be decomposed in accordance to
a symmetry of the Hamiltonian (equations of motion).
(b) Rotation symmetry V (x) = gV (x), G = {e, g, g 2 }: By the same argument we
have three possibilities:
gψ(x) = ψ(x); gψ(x) = ei2π/3 ψ(x); g −1 ψ(x) = e−i2π/3 ψ(x).
In addition, by the time reversal symmetry if ψ(x) is solution then ψ ∗ (x) is solu-
tion with the same eigenvalue as well. From this follows that the spectrum must
be degenerate. The spectrum is split into a real eigenfunction {ψ1 (x)}, and a
degenerate pair of real eigenfunctions
ψ2 (x) = ψ(x) + ψ ∗ (x); ψ3 (x) = i(ψ(x) − ψ ∗ (x)) , where gψ(x) = ei2π/3 ψ(x)
invariant under rotations by 1/3-rd of a circle.

2021-10-30 33 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

L R

Figure 2.3: (left) A reflection-symmetric double-well potential. (right) A 1/3rd-circle

rotation-symmetric plane billiard (infinite wall potential in 2D). (B. Gutkin)

M m M

Figure 2.4: A classical colinear CO2 molecule [1].

(B. Gutkin)

Example 2.4. Vibrations of a classical CO2 molecule: Consider one carbon and
two oxygens constrained to the x-axis [1] and joined by springs of stiffness k, as shown
in figure 2.4. Newton’s second law says

k
ẍ1 = − (x1 − x2 )
M
k k
ẍ2 = − (x2 − x3 ) − (x2 − x1 )
m m
k
ẍ3 = − (x3 − x2 ) . (2.14)
M
The normal modes, with time dependence xj (t) = xj exp(itω) , are the common fre-
quency ω vibrations that satisfy (2.14),
    
A −A 0 x1 x1
2
Hx = −a 2 a −a x2  = ω x2  , (2.15)
0 −A A x3 x3

where a = k/m, A = k/M . Secular determinant det (H − ω 2 1) = 0 now yields a cubic

equation for ω 2 .
You might be tempted to stick this [3×3] matrix into Mathematica or whatever, but
please do that in some other course. What would understood by staring at the output?
In this course we think.

birdtracks.eu/course3 week2 34 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

First thing to always ask yourself is: does the system have a symmetry? Yes! Note
that the CO2 molecule (2.14) of figure 2.4 is invariant under x1 ↔ x3 interchange, i.e.,
coordinate relabeling by matrix σ that commutes with our law of motion H,
   
0 0 1 0 −A A
σ = 0 1 0  , σH = Hσ = −a 2 a −a . (2.16)
1 0 0 A −A 0

We can now use the symmetry operator σ to simplify the calculation. As σ 2 =

1, its eigenvalues are ±1, and the corresponding symmetrization, anti-symmetrization
projection operators (1.38) are
1 1
P+ = (1 + σ) , P− = (1 − σ) . (2.17)
2 2
The dimensions di = tr Pi of the two subspaces are

d+ = 2 , d− = 1 . (2.18)

As σ and H commute, we can now use spectral decomposition (1.30) to block-diagonalize

H to a 1-dimensional and a 2-dimensional matrix.
On the 1-dimensional antisymmetric subspace, the trace of a [1×1] matrix equals
its sole matrix element equals it eigenvalue
1 k
λ− = HP− = (tr H − tr Hσ) = (a + A) − a = ,
2 M
2
so the corresponding eigenfrequency is ω− = k/M . To understand its physical mean-
ing, write out the antisymmetric subspace projection operator (2.18) explicitly. Its non-
vanishing columns are proportional to the sole eigenvector
   
1 0 −1 1
1
P− = 0 0 0  ⇒ e(−) =  0  . (2.19)
2
−1 0 1 −1

In this subspace the outer oxygens are moving in opposite directions, with the carbon
stationary.
On the 2-dimensional symmetric subspace, the trace yields the sum of the remain-
ing two eigenvalues
1 k k
λ+ + λ0 = tr HP+ = (tr H + tr Hσ) = (a + A) + a = +2 .
2 M m
We could disentangle the two eigenfrequencies by evaluating tr H2 P+ , for example, but
thinking helps again.
There is still another, translational symmetry, so obvious that we forgot it; if we
change the origin of the x-axis, the three coordinates xj → xj − δx change, for any
continuous translation δx, but the equations of motion (2.14) do not change their form,

H x = H x + H δx = ω 2 x ⇒ H δx = 0 . (2.20)
(0)
So any translation e = δx = (δx, δx, δx) is a nul, ‘zero mode’ eigenvector of H
in (2.16), with eigenvalue λ0 = ω02 = 0, and thus the remaining eigenfrequency is
2
ω+ = k/M + 2 k/m. As we can add any nul eigenvector e(0) to the corresponding
e(+) eigenvector, there is some freedom in choosing e(+) . One visualization of the

2021-10-30 35 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

corresponding eigenvector is the carbon moving opposite to the two oxygens, with total
momentum set to zero.
(Taken from AWH Example 6.2.3 Degenerate eigenproblem, but done here using
symmetries.)

Example 2.5. Vibrational spectra of molecules: In the linear, harmonic oscillator

approximation the classical dynamics of the molecule is governed by the Hamiltonian
N N
X mi 2 1 X >
H= ẋi + xi Vij xj ,
i=1
2 2 i,j=1

where {xi } are small deviations from the resting the equilibrium, resting points of the
molecules labelled i. Vij is a symmetric matrix, so it can be brought to a diagonal form
by an orthogonal transformation, a set of N uncoupled harmonic oscillators or normal
modes of frequencies {ωi }.
N
X mi 2
ẏi + ωi2 yi2 .


x → y = U x, H= (2.21)
i=1
2

Consider now the ring of pair-wise interactions of two kinds of molecules sketched in
figure 2.5 (a), given by the potential
N  
1X xi
k1 (xi − yi )2 + k2 (xi+1 − yi )2 ,


V (z) = zi = , (2.22)
2 i=1 yi

whose [2N ×2N ] matrix form is (aside to the cognoscenti: this is a Toeplitz matrix):

k1 + k2 −k1 0 0 0 ... 0 0 −k2

 
 −k1 k1 + k2 −k2 0 0 ... 0 0 0 
 
 0 −k2 k1 + k2 −k1 0 ... 0 0 0 
1 0
 
Vij =  0 −k1 k1 + k2 −k2 ... 0 0 0 
2  ..

.. .. .. .. .. .. .. .. 
 .
 . . . . . . . .  
 0 0 0 0 0 ... −k2 k1 + k2 −k1 
−k2 0 0 0 0 ... 0 −k1 k1 + k2

This potential matrix is a holy mess. How do we find an orthogonal transformation (2.21)
that diagonalizes it? Look at figure 2.5 (a). Molecules lie on a circle, so that suggests
we should use a Fourier representation. As the i = 1 labelling of the starting molecule
on a ring is arbitrary, we are free to relabel them, for example use the next molecule
pair as the starting one. This relabelling is accomplished by the [2N ×2N ] permutation
matrix (or ‘one-step shift’, ‘stepping’ or ‘translation’ matrix) M of form

    
0 0 ... 0 I z1 zn
I 0 ... 0 0   z2   z1 
       
1 0 xi

0 I ... 0 0  z3   z2 
  =  , I= , zi = (2.23)
0 1 yi

. .. .. .. ..   ..   .. 
 .. . . . .   .   . 
0 0 ... I 0 zn zn−1
| {z }
M

birdtracks.eu/course3 week2 36 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

Yn X1
ω y1
Xn
x1
Y1
optical m−1

X2 acoustic
x3

Y2
k
y3
(a) (b) −n/2 n/2 (c) x2
y2

Figure 2.5: (a) Chain with circular symmetry. (b) Dependance of frequency on the
representation wavenumber k. (c) Molecule with D3 symmetry. (B. Gutkin)

Projection operators corresponding to M are worked out in example 2.6. They are N
distinct [2N ×2N ] matrices,

I λ̄I λ̄2 I ... λ̄N −2 I λ̄N −1 I

 
 λI I λ̄I ... λ̄N −3 I λ̄N −2 I 
λ2 I λ̄N −4 I λ̄N −3 I 
 
 λI I ... 
2πi

Pk =  ..  , λ = exp k
 
.. .. .. .. ..
 . . . . . .  N
 
λN −2 I λN −3 I λN −4 I ... I λ̄I 
λN −1 I λN −2 I λN −2 I ... λI I
(2.24)
which decompose the 2N -dimensional configuration space of the molecule ring into
a direct sum of N 2-dimensional spaces, one for each discrete Fourier mode k =
0, 1, 2, · · · , N − 1.
The system (2.22) is clearly invariant under the cyclic permutation relabelling M ,
[V, M ] = 0 (though checking this by explicit matrix multiplications might be a bit tedious),
so the Pk decompose the interaction potential V as well, and reduce its action to the kth
2-dimensional subspace. Thus the [2N ×2N ] diagonalization (2.21) is now reduced to
a [2×2] diagonalization which one can do by hand. The resulting kth space is spanned
by two 2N -dimensional vectors, which we guess to be of form:

1 0
   
  0  1 
   
  λ  0 
1  1 
   
η1 = √ ,
 0 η2 = √ λ .

n

n

 .. .. 



 . 
 . 

λn−1   0 
0 λn−1

In order to find eigenfrequences we have to consider action of V on these two vectors:

V η1 = (k1 + k2 )η1 − (k1 + k2 λ)η2 , V η2 = (k1 + k2 )η2 − (k1 + k2 λ̄)η1 .

2021-10-30 37 birdtracks.eu/course3 week2

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

The corresponding eigenfrequencies are determined by the equation:

ω2
  
k1 + k2 −(k1 + k2 λ)
0 = det − I =⇒
−(k1 + k2 λ̄) k1 + k2 2
1 2 k
ω± (k) = k1 + k2 ± |k1 + k2 λ | , (2.25)
2
one acoustic (ω(0) = 0), one optical, see figure 2.5 (b) and the acoustic and optical
phonons wiki. (B. Gutkin)

Example 2.6. Projection operators for cyclic group CN .

Consider a cyclic group CN = {e, g, g 2 , · · · g N −1 }, and let M = D(g) be a [2N ×2N ]
representation of the one-step shift g. In the projection operator formulation (1.25),
the N distinct eigenvalues of M , the N th roots of unity λn = λn , λ = exp(i 2π/N ),
n = 0, . . . N − 1, split the 2N -dimensional space into N 2-dimensional subspaces by
means of projection operators
N −1 −n
Y M − λm I Y λ M − λm I
Pn = = , (2.26)
λn − λm m=1
1 − λm
m6=n

where we have multiplied all denominators and numerators by λ−n . The numerator is
now a matrix polynomial of form (x − λ)(x − λ2 ) · · · (x − λN −1 ) , with the zeroth root
(x − λ0 ) = (x − 1) quotiented out from the defining matrix equation M N − 1 = 0. Using

1 − xN
= 1 + x + · · · + xN −1 = (x − λ)(x − λ2 ) · · · (x − λN −1 )
1−x
we obtain the projection operator in form of a discrete Fourier sum (rather than the
product (1.25)),
N −1
1 X i 2π
Pn = e N nm M m .
N m=0
This form of the projection operator is the simplest example of the key group theory tool,
projection operator expressed as a sum over characters,
1 X
Pn = χ̄(g)D(g) ,
|G| g∈G

upon which stands all that follows in this course. (B. Gutkin and P. Cvitanović)

2.10 What are cosets good for? (a discussion)

Question 2.2. Henriette Roux asks
Q What are cosets good for?
A Apologies for glossing over their meaning in the lecture. I try to minimize group-theory
jargon, but cosets cannot be ignored.
Dresselhaus et al. [4] (click here) Chapter 1 Basic Mathematical Background: Introduction
needs them to show that the dimension of a subgroup is a divisor of the dimension of the group.
For example, C3 of dimension 3 is a subgroup of D3 of dimension 6.

birdtracks.eu/course3 week2 38 2021-10-30

 GROUP THEORY - WEEK 2. FINITE GROUPS - DEFINITIONS

In ChaosBook Chapter 10. Flips, slides and turns cosets are absolutely essential. The
significance of the coset is that if a solution has a symmetry, then the elements in a coset act on
the solution the same way, and generate all equivalent copies of this solution. Example 10.7.
Subgroups, cosets of D3 should help you understand that.

Henriette Roux writes: When talking about the cosets of a subgroup we demonstrated
multiplication between cosets with a specific example, but this wasn’t leading to
something along the lines of that the set of all left cosets of a subgroup (or the
set of all the right cosets of a subgroup) form a group, correct? It didn’t appear
so in the example since the “unit” {E, A} we looked appears to only have the
properties of an identity with multiplication from one direction (the direction
depending on if it is the set of left cosets or the set of right cosets). In the context
of the lecture I think this point was related to Lagrange’s theorem (although we
didn’t call it that) and I vaguely remember cosets being used in the proof of
Lagrange’s theorem but I wasn’t connecting it today. Are we going to cover that
in a future lecture?
Predrag You are right - Lagrange’s theorem (see the wiki) simply says the order of a
subgroup has to be a divisor of the order of the group. We used cosets to partition
elements of G to prove that. But what we really need cosets for is to define (see
Dresselhaus et al. [4] Sect. 1.7) Factor Groups whose elements are cosets of a
self-conjugate subgroup (click here). I will not cover that in a subsequent lecture,
so please read up on it yourself.
Henriette Roux You talked about the period of an element X, and said that that period
is the set
{E, X, · · · , X n−1 } , (2.27)
where n is the order of the element X. I had thought that set was the subgroup
generated by the element X and that the period of the element X was a synonym
for the order of the element X? Is that incorrect?
Predrag To keep things as simple as possible, in Thursday’s lecture I followed Sect. 1.3
Basic Definitions of Dresselhaus et al. textbook [4], to the letter. In Def. 3 the
order of an element X is the smallest n such that X n = E, and they call the set
(2.27) the period of X. I do not like that usage (and do not remember seeing it
anywhere else). As you would do, in ChaosBook.org Chap. Flips, slides and
turns I also define the smallest n to be the period of X and refer to the set (2.27)
as the orbit generated by X. When we get to compact continuous groups, the
orbit will be a (great) circle generated by a given Lie algebra element, and look
more like what we usually think of as an orbit.
I am not using my own ChaosBook.org here, not to confuse things further by
discussing both time evolution and its discrete symmetries. Here we focus on the
discrete group only (typically spatial reflections and finite angle rotations).

2021-10-30 39 birdtracks.eu/course3 week2

 EXERCISES

References
[1] G. B. Arfken, H. J. Weber, and F. E. Harris, Mathematical Methods for Physi-
cists: A Comprehensive Guide, 7th ed. (Academic, New York, 2013).
[2] R. Brauer, “On algebras which are connected with the semisimple continuous
groups”, Ann. Math. 38, 857 (1937).
[3] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[4] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[5] R. Penrose, “Angular momentum: An approach to combinatorical space-time”,
in Quantum Theory and Beyond, edited by T. Bastin (Cambridge Univ. Press,
Cambridge, 1971).
[6] R. Penrose, “Applications of negative dimensional tensors”, in Combinatorial
mathematics and its applications, edited by D. J. A. Welsh (Academic, New
York, 1971), pp. 221–244.
[7] R. Penrose, The Road to Reality - A Complete Guide to the Laws of the Universe
(A. A. Knopf, New York, 2005).
[8] R. Penrose and M. A. H. MacCallum, “Twistor theory: An approach to the quan-
tisation of fields and space-time”, Phys. Rep. 6, 241–315 (1973).
[9] D. S. Silver, “The new language of mathematics”, Amer. Sci. 105, 364 (2017).
[10] M. Stone and P. Goldbart, Mathematics for Physics: A Guided Tour for Graduate
Students (Cambridge Univ. Press, Cambridge UK, 2009).
[11] M. Tinkham, Group Theory and Quantum Mechanics (Dover, New York, 2003).
[12] E. P. Wigner, Group Theory and Its Application to the Quantum Mechanics of
Atomic Spectra (Academic, New York, 1931).

Exercises
2.1. Gx ⊂ G. The maximal set of group actions which maps a state space point x into itself,
Gx = {g ∈ G : gx = x} , (2.28)
is called the isotropy group (or stability subgroup or little group) of x. Prove that the set
Gx as defined in (2.28) is a subgroup of G.
2.2. Transitivity of conjugation. Assume that g1 , g2 , g3 ∈ G and both g1 and g2 are
conjugate to g3 . Prove that g1 is conjugate to g2 .
2.3. Isotropy subgroup of gx. Prove that for g ∈ G, x and gx have conjugate isotropy
subgroups:
Ggx = g Gx g −1

birdtracks.eu/course3 week2 40 2021-10-30

EXERCISES

2.4. D3 : symmetries of an equilateral triangle. Consider group D3 , the symmetry group

of an equilateral triangle:

2 3.
(a) List the group elements and the corresponding geometric operations
(b) Find the subgroups of the group D3 .
(c) Find the classes of D3 and the number of elements in them, guided by the geometric
interpretation of group elements. Verify your answer using the definition of a class.
(d) List the conjugacy classes of subgroups of D3 . (continued as exercise 4.1)

2.5. C4 invariant potential. Consider the Schrödinger equation for a particle moving in a
two-dimensional bounding potential V , such that the spectrum is discrete. Assume that
V is CN -invariant (in some literature, ZN -invariant), i.e., V remains invariant under the
rotation R by the angle 2π/N . For N = 3 case, figure 2.6 (a), the spectrum of the system
can be split into two sectors: {En0 } non-degenerate levels corresponding to symmetric
eigenfunctions φn (Rx) = φn (x) and doubly degenerate levels {En± } corresponding to
non-symmetric eigenfunctions φn (Rx) = e±2πi/3 φn (x).
(a) What is the spectral structure in the case of N = 4, figure 2.6 (b)?
How many sectors appear and what are their degeneracies?
(b) What is the spectral structure for general N ?
(c) A constant magnetic field normal to the 2D plane is added to V .
How will it affect the spectral structure?
(d) (bonus question) Figure out the spectral structure if the symmetry group of potential
is D3 (also includes 3 reflections), figure 2.6 (c).
(Boris Gutkin)

(a) (b) (c)

Figure 2.6: Hard wall potential with (a) symmetry C3 , (b) symmetry C4 , and (c) symmetry
D3 .

2021-10-30 41 birdtracks.eu/course3 week2

 EXERCISES

2.6. Permutation of three objects. Consider S3 , the group of permutations of 3 objects.

(a) Show that S3 is a group.
(b) List the conjugacy classes of S3 .
(c) Give an interpretation of these classes if the group elements are substitution opera-
tions on a set of three objects.
(c) Give a geometrical interpretation in case of group elements being symmetry opera-
tions on equilateral triangle.
2.7. Three masses on a loop. Three identical masses, connected by three identical springs,
are constrained to move on a circle hoop as shown in figure 2.7. Find the normal modes.
Hint: write down coupled harmonic oscillator equations, guess the form of oscillatory
solutions. Then use basic matrix methods, i.e., find zeros of a characteristic determinant,
find the eigenvectors, etc.. (K. Y. Short)

Figure 2.7: Three identical masses are constrained to move on a hoop, connected by
three identical springs such that the system wraps completely around the hoop. Find
the normal modes.

birdtracks.eu/course3 week2 42 2021-10-30

 group theory - week 3

Group representations

Homework HW3

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 3.1 1-dimensional representation of anything 1 point

Exercise 3.2 2-dimensional representation of S3 4 points
Exercise 3.3 3-dimensional representations of D3 5 points

Bonus points
Exercise 3.4 Abelian groups 1 point
Exercise 3.5 Representations of CN 1 point

Total of 10 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

43
 GROUP THEORY - WEEK 3. GROUP REPRESENTATIONS

3.1 Week’s videos, reading

3.1.1 Matrix representations, Schur’s Lemma
2021-05-25 Predrag Lecture 3
Irreps, unitary reps and Schur’s Lemma.

Dresselhaus et al. [1] Sect. 2.4 The Unitarity of Representations.

Dresselhaus et al. [1] Sects. 2.5 and 2.6 Schur’s Lemma.

Lecture 3 (Unedited) Unitarity of reps; Schur’s Lemma (2:04:56 h)

3.1.2 Wonderful Orthogonality Theorem

2021-05-27 Predrag Lecture 4
Section playlist

(extra) Recap of lect. 3 (5:36 min)

◦ Sect. 3.2 It’s all about class

Dresselhaus et al. [1] Sect. 2.7 ‘Wonderful’ Orthogonality Theorem, sect. 2.8
Representations and vector spaces.

Whence "orthogonality"? The ideas: observables are Hermitian; matrix reps

are unitary; average over the group to extract invariants. A matrix rep forms a
complex unit vector, hence "orthogonality". (9:47 min)

Character orthogonality theorem (X:XX min)

Dresselhaus et al. [1] Sects. 3.1 Characters and Class to 3.5 The number of
irreducible representations.

Characters. Character orthogonality. (X:XX min)

Class (17:56 min)

Number of classes equals the number of irreps (6:24 min)

(extra) Discussion: Irrep dimension; Are classes subgroups, cosets? Week’s

homework. Classes and irreps of D3 . N-gon intuition. A LaTeX template. (26:39
min)

Tinkham [3] covers the same material in Chapter 3 Theory of Group Representa-
tions, in a more compact way.

birdtracks.eu/course3 week3 44 2021-07-29

 GROUP THEORY - WEEK 3. GROUP REPRESENTATIONS

3.2 It’s all about class

The essential group theory notions introduced here are the notion of irreducible repre-
sentations (irreps) and their orthogonality

The Group Orthogonality Theorem: Let Dµ , Dµ0 be two irreducible matrix rep-
resentations of a compact group G of dimensions dµ , dµ0 , where the sum is over all
elements of the group, G = {g}, and |G| is their number, or the order of the group:

G
1 X (µ) 0 0 1 0
D (g)a b D(µ ) (g −1 )a b0 = δµµ0 δaa δbb0 .
|G| g dµ

This is a remarkable formula, one relation for each of the d2µ + d2µ0 matrix entries. Still,
the explicit matrix entries reflect largely arbitrary coordinate choices - there should a
more compact statement of irreducibility, and there is: the “character orthogonality
theorem” (3.1).
Consider a reducible representation D(g), i.e., a representation of group element g
that after a suitable similarity transformation takes form
 (a) 
D (g) 0 0 0
 0 (b)
D (g) 0 0
D(g) =  0 ,
 
 0 D(c) (g) 0 
..
0 0 0 .

with character for class C given by

χ(C) = ca χ(a) (C) + cb χ(b) (C) + cc χ(c) (C) + · · · ,

where ca , the multiplicity of the ath irreducible representation (colloquially called “ir-
rep”), is determined by the character orthonormality relations,
class
1 X
ca = χ(a)∗ χ = Nk χ(a) (Ck−1 ) χ(Ck ) . (3.1)
h
k

Knowing characters is all that is needed to figure out what any reducible representation
decomposes into! Work out exercise 4.2 as an example.

3.3 Other sources (optional)

3.3.1 Hard work builds character
Irreps, unitary reps, Schur’s Lemma.

Section playlist

2021-07-29 45 birdtracks.eu/course3 week3

 GROUP THEORY - WEEK 3. GROUP REPRESENTATIONS

Chapter 2 Representation Theory and Basic Theorems

Dresselhaus et al. [1], up to and including
Sect. 2.4 The Unitarity of Representations (click here)

This requires character (1:23 min)

Hard work builds character (15:05 min)

The symmetry group of a propeller (6:13 min)

Irreps of C3 (14:52 min)

Rotation and reflections in a plane: irreps of D3 (13:38 min)

(extra) Discussion: more symmetries, fewer invariant subspaces (2:10 min)

(extra) Discussion: abelian vs. nonabelian (2:11 min)

This week’s Dresselhous exposition (or the corresponding chapter in Tinkham [3])
comes from Wigner’s classic [4] Group Theory and Its Application to the Quantum
Mechanics of Atomic Spectra, which is a harder going, but the more group theory you
learn the more you’ll appreciate it. Eugene Wigner got the 1963 Nobel Prize in Physics,
so by mid 60’s gruppenpest was accepted in finer social circles.
In this course, we learn about full reducibility of finite and compact continuous
groups in two parallel ways. On one hand, I personally find the multiplicative projec-
tion operators (1.25), coupled with the notion of class algebras (Harter [2] (click here)
appendix C) most intuitive - a block-diagonalized subspace for each distinct eigenvalue
of a given all-commuting matrix.
On the other hand, the character weighted sums (here related to the multiplica-
tive projection operators as in ChaosBook Example A24.2 Projection operators for
discrete Fourier transform) offer a deceptively ‘simple’ and elegant formulation of
full-reducibility theorems, preferred by all standard textbook expositions.

3.3.2 History (optional)

The structure of finite groups was understood by late 19th century. A full list of finite
groups was another matter. The complete proof of the classification of all finite groups
takes about 3 000 pages, a collective 40-years undertaking by over 100 mathematicians,
read the wiki.
Alex Kontorovich, Rutgers MAT 640:503 Complex Analysis. A wonderful
lecturer, here he diverges into the story of Cardano
√ and cubics. They are cube-ic for a
reason. Did you know people learned to use −1 before they understood that a number
can be negative, like −1? Listen to his first lecture. Oh no! He just made me solve the
cubic, something I had avoided my entire life. So far. You’ll love it.
According to Kevin Hartnett, The ‘Useless’ Perspective That Transformed Math-
ematics: Representation theory was initially dismissed. Today, it’s central to much of
mathematics.

birdtracks.eu/course3 week3 46 2021-07-29

 GROUP THEORY - WEEK 3. GROUP REPRESENTATIONS

Groups are complicated collections of mathematical objects – like numbers or sym-

metries – that stand in a particular structured relationship with each other. Representa-
tion theory is a way of taking such complicated objects and “representing” them with
simpler objects. It converts the sometimes mysterious world of groups into the well-
trammeled territory of linear algebra, the study of simple transformations performed
on objects called vectors, which are effectively directed line segments. These objects
are defined by coordinates, which can be displayed in the form of a matrix, the core
element of linear algebra, an array of numbers. While groups are abstract and often
difficult to get a handle on, matrices and linear algebra are elementary.
Geordie Williamson, Mathematics in light of representation theory. October
16, 2015 at Urania: Symmetry is all around us. The mathematical study of symme-
try becomes simpler when we linearize, and in doing so we enter the realm of repre-
sentation theory. Representation theory has applications throughout mathematics (the
Fourier transform, monstrous moonshine, the Langlands program, the proof of Fer-
mat’s last theorem, ... ) and science (crystallography and spectroscopy in chemistry,
signal processing in engineering, the standard model in physics, ... ). The lecture is an
introduction to the representation theory of finite groups, both over the complex num-
bers and over fields of positive characteristic (so-called modular representation theory).
Williamson discusses Frobenius’ discovery of the character table in Berlin in 1896,
Brauer’s first steps in modular representation theory in the 1930s, and the role of the
character table in the discovery of the monster simple group in the 1980s. Williamson
finishes with a discussion of recent developments in the modular representations of
symmetric and finite general linear groups.
From Emory Math Department: A pariah is real! The simple finite groups fit into
18 families, except for the 26 sporadic groups. 20 sporadic groups AKA the Happy
Family are parts of the Monster group. The remaining six loners are known as the
pariahs. (Check the notes sect. 5.2 Literature for links to the Ree group and the whole
classification.)

References
[1] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[2] W. G. Harter and N. dos Santos, “Double-group theory on the half-shell and the
two-level system. I. Rotation and half-integral spin states”, Amer. J. Phys. 46,
251–263 (1978).
[3] M. Tinkham, Group Theory and Quantum Mechanics (Dover, New York, 2003).
[4] E. P. Wigner, Group Theory and Its Application to the Quantum Mechanics of
Atomic Spectra (Academic, New York, 1931).

2021-07-29 47 birdtracks.eu/course3 week3

 EXERCISES

Exercises
3.1. 1-dimensional representation of anything. Let D(g) be a representation of a group
G. Show that d(g) = det D(g) is one-dimensional representation of G as well.
(B. Gutkin)
3.2. 2-dimensional representation of S3 .
(i) Show that the group S3 can be generated by two permutations:
   
1 2 3 1 2 3
a= , d= .
1 3 2 3 1 2

(ii) Show that matrices:

     
1 0 0 1 z 0
D(e) = , D(a) = , D(d) = ,
0 1 1 0 0 z2

with z = ei2π/3 , provide proper (faithful) representation for these elements and
find representation for the remaining elements of the group.
(iii) Is this representation irreducible?
(B. Gutkin)
3.3. 3-dimensional representations of D3 . The dihedral group D3 is the symmetry group
of the equilateral triangle. It has 6 elements

D3 = {E, C, C 2 , σ (1) , σ (2) , σ (3) } ,

where C is rotation by 2π/3 and σ (i) is reflection along one of the 3 symmetry axes.
(i) Prove that this group is isomorphic to S3
(ii) Show that matrices
     
1 0 0 z 0 0 0 0 1
(1)
D(E) =  0 1 0  , D(C) =  0 1 0  , D(σ ) =  0 −1 0 ,
0 0 1 0 0 z2 1 0 0
(3.2)
generate a 3-dimensional representation D(g) of D3 . Hint: Calculate products for
representations of group elements and compare with the group table (see lecture).
(iii) Show that this is a reducible representation which can be split into one dimensional
A and two-dimensional representation Γ. In other words find a matrix R such that
 
A(g) 0
RD(g)R−1 =
0 Γ(g)
for all elements g of D3 . (Might help: D3 has only one (non-equivalent) 2-dim
irreducible representation).
(B. Gutkin)
3.4. Abelian groups. Let G be a group with only one-dimensional irreducible representa-
tions. Show that G is Abelian.
(B. Gutkin)
3.5. Representations of CN . Find all irreducible representations of CN .
(B. Gutkin)

birdtracks.eu/course3 week3 48 2021-07-29

 group theory - week 4

Hard work builds character

Homework HW4

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 4.3 All irreducible representations of D4 10 points

Bonus points
Exercise 4.4 Irreducible representations of dihedral group Dn 2 points
Exercise 4.5 Perturbation of Td symmetry 6 points
Exercise 4.7 Two particles in a potential 4 points

Total of 10 points = 100 % score. Bonus points accumulate, can help you later if you
miss a few problems.

49
 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

Table 4.1: The D3 group multiplication table. The same as table 2.1, but written as a
class operator multiplication table.

D3 1 r r2 σ1 σ2 σ3
1 1 r r2 σ1 σ2 σ3 D3 C1 C2 C3
r r r2 1 σ3 σ1 σ2 C1 C1 C2 C3
r2 r2 1 r σ2 σ3 σ1 C2 C2 2C1 +C2 2C3
σ1 σ1 σ2 σ3 1 r r2 C3 C3 2C3 3C1 +3C2
σ2 σ2 σ3 σ1 r2 1 r
σ3 σ3 σ1 σ2 r r2 1

2021-06-01 Lecture 5
Character orthogonality theorem

Character orthogonality relations. (10:53 min)

Character defined. Character of identity = dimension of the representation.
Character orthogonality stated as an average of the group over irrep char-
acters (but not derived). Special cases checked. Completeness verified.
Example: Reflection group in 1 dimension. Characters and their orthogo-
nality checked.
A summary: it is all about class and character (18:50 min)
Presumes knowledge of CN irreps, argues that a reflection (DN ) mixes
them up, thus reducing the number of irreps. 3-disk classes. Character is
labelled by the class and the irrep label. Example: discrete Fourier trans-
form is an [N × N ] unitary matrix. D4 character table.
(extra) Discussion: class and character (7:01 min)

2021-06-01 Predrag Lecture 6

Hard work builds character
Complete Dresselhaus et al. [1] sects. 3.3 “Wonderful Orthogonality Theorem
for Characters” to 3.8 “Setting up Character Tables” (click here). This material
is also covered in Tinkham [7] Chapter 3 Theory of Group Representations.

1. theory of finite groups are a natural generalization of discrete Fourier represen-

tations

2. it is all about class and character. “Character", in particular, I find very surprising
- one complex number suffices to characterize a matrix!

birdtracks.eu/course3 week4 50 2021-07-29

 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

4.1 Other sources (optional)

Group theory? It is all about class & character.
— Predrag Cvitanović, One minute elevator pitch

For a continuous group version of the character orthogonality theorem, see sect. 9.4. In
particular, the replacement of an irrep matrix representation D(µ) (g)a b by its character
χ(µ) (g) (a single scalar quantity) leads to no loss of any of the matrix indices structure.
I enjoyed reading Mathews and Walker [6] Chap. 16 Introduction to groups. You
can download it from here. Goldbart writes that the book is “based on lectures by
Richard Feynman at Cornell University.” Very clever. Try working through the exam-
ple of fig. 16.2: deadly cute, you get explicit eigenmodes from group theory alone. The
main message is that if you think things through first, you never have to construct the
representation matrices in explicit form - recasting the calculation in terms of invari-
ants, such characters, will get you there much faster.
You might find Gutkin notes useful:
Lect. 4 Representation Theory II, up to Sect. 4.5 Three types of representations:
Character tables. Dual character orthogonality. Regular representation. Indicators for
real, pseudo-real and complex representations. See example 4.3 “Irreps for quaternion
multiplication table.”
Oliver Pierson ChaosBook.org chapter Discrete factorization - Character tables
(10:05 min)
Oliver Pierson ChaosBook.org chapter Discrete factorization - Projection into
invariant subspaces (5:31 min)
Lect. 5 Applications I. Vibration modes go through Wigner’s theorem, Cn symme-
try and D3 symmetry. Study Example 5.1. Cn symmetry. More quantum mechanics
applications follow in
sect. 6.2 Applications II. Quantum Mechanics, Sect. 2. Perturbation theory.
Does the proof in the Lect. 4 Representation Theory II Appendix that the number
of irreps equals the number of classes make sense to you? For an easy argument, see
Vedensky Theorem 5.2 The number of irreducible representations of a group is equal
to the number of conjugacy classes of that group. For a proof, work though Murnaghan
Theorem 7. If you prefer a proof that your professor cannot understand, click here.
For the record (I retract the heady claim I made in class):
Mathworld.Wolfram.com: “A character table often contains enough information to
identify a given abstract group and distinguish it from others. However, there exist
nonisomorphic groups which nevertheless have the same character table, for example
D4 (the symmetry group of the square) and Q8 (the quaternion group).”
exercise 4.3
Fun read along these lines: Hart and Segerman [2] discuss the distinction between
abstract groups and symmetry groups of objects. They exhibit two very different ob-
jects with
D4 = hr, σ | σrσ = r−1 , r4 = σ 2 = ei (4.1)
symmetry (describing the group this way is called a presentation of D4 ), and explain
the Cayley graph for D4 (its edges with arrows correspond to rotations, the other edges

2021-07-29 51 birdtracks.eu/course3 week4

 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

correspond to reflections). For quaternions they discuss a 1-dimensional space group

built of “monkey blocks” (but do not identify its crystallographic name). Q8 is a
subgroup of the symmetries of the 3-dimensional sphere S 3 , the unit sphere in R4 .
They offer a visualisation of the action of Q8 on a hypercube and construct a sculpture
whose symmetry group is Q8 , using stereographic projection from the unit sphere in
4-dimensional space. Q8 is discussed here in example 4.3.

Simon Berman You would think that the analysis of three masses connected by har-
monic strings, see figure 4.1, is a simple exercise finding irreps of D3 symmetry,
but no, it merits a 2019 Phys. Rev. Lett., see Katz and Efrati [3] Self-driven frac-
tional rotational diffusion of the harmonic three-mass system. The article even
starts with our figure 4.1. We continue the discussion in sect. 6.4.

Example 4.1. D3 symmetry: Reflections and rotations of a triangle, figure 2.5 (c)
   
0 0 0 0 1 0 −1 0 0 0 0 0
0 0 0 0 0 1 0 1 0 0 0 0
   
1 0 0 0 0 0 0 0 0 0 −1 0
D(T ) = 
0
, D(σ1 ) =   (4.2)
 1 0 0 0 0

0
 0 0 0 0 1
0 0 1 0 0 0 0 0 −1 0 0 0
0 0 0 1 0 0 0 0 0 1 0 0

   
0 0 0 0 −1 0 0 0 −1 0 0 0
0 0 0 0 0 1 0 0 0 1 0 0
   
0 0 −1 0 0 0 −1 0 0 0 0 0
D(σ2 ) = 
0
, D(σ3 ) =  
 0 0 1 0 0
0
 1 0 0 0 0
−1 0 0 0 0 0 0 0 0 0 −1 0
0 1 0 0 0 0 0 0 0 0 0 1
(4.3)

G = {[e]; [g, g 2 ]; [σ1 , σ2 , σ3 ]}, χ(1) = {1, 1, 1}, χ(2) = {1, 1, −1}, χ(3) = {2, −1, 0}

ri = χ(e)χ(i) (e)/6; ri = {1, 1, 2} =⇒ D = 2E ⊕ A1 ⊕ A2 .

1 X (i)
Pi = χ (g)D(g)
3 g∈G
   
0 0 0 0 0 0 1 0 1 0 1 0
0 1 0 1 0 1 0 0 0 0 0 0
   
1 0 0 0 0 0 0 1 1 0 1 0 1 0
P1 =  , P2 =   (4.4)
30 1 0 1 0 1 30 0 0 0 0 0

0 0 0 0 0 0 1 0 1 0 1 0
0 1 0 1 0 1 0 0 0 0 0 0
The 3 equal masses connected by harmonic springs system of figure 4.1 is a text-
book example of such system, see for example problems 6.37 and 9.16 in Kotkin and
Serbo [4] Collection of Problems in Classical Mechanics.

birdtracks.eu/course3 week4 52 2021-07-29

 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

ν1 ν2
(ν1 3 −ν)2 2

Figure 4.1: Modes of a molecule with D3 symmetry. (B. Gutkin)

The vibrational modes associated with the two 1-dimensional representations are
given by
   
0 1
1 0
   
0 1
P1V = α 
1
 and P 2V = β  ,
0
   
0 1
1 0
respectively. Here P1V represents symmetric mode shown in figure 4.1 (red). The sec-
ond mode P2V corresponds to the rotations of the whole system. The projection opera-
tor for the two-dimensional representation is
 
2 0 −1 0 −1 0
0
 2 0 −1 0 −1
2 2 1 −1 0 2 0 −1 0
P3 = (2D(I) − D(T ) − D(T )) =   (4.5)
6 30 −1 0 2 0 −1
−1 0 −1 0 2 0
0 −1 0 −1 0 2

From this we have to separate two vectors corresponding to shift in x and y directions.
   
1 0
 0   1 
   √ 
 −1/2  − 3/2
ηx =  √
 3/2  , ηy = 
 
 −1/2 
  √ 
 −1/2   3/2 
√
− 3/2 −1/2
 

 
      
  



 2 0 0 0 




 0 0 2  0 
         
1  −1 1 1 1 0 1  0 
 
P3V = α √   +β √   +γ √   +δ √   ,
       

 6 0  2 0  6 −1 2  1 


 −1 −1 0  0 

 
0 0 −1 −1

 


 


 | {z } | {z } | {z } | {z }

ξ1 ξ2 ξ3 ξ4
p p
where ηx = 3/2(ξ4 + ξ1 ), ηy = 3/2(ξ3 − ξ2 ). Vectors ξi are columns of P3 and their

2021-07-29 53 birdtracks.eu/course3 week4

 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

linear combinations. The orthogonal vectors are given by

   
1 0
 0   1 
  √ 
 −1/2   3/2 
√
p p
ν1 = 3/2(ξ1 − ξ4 ) = 
,
 ν2 = 3/2(ξ2 + ξ3 ) = 
 .
− 3/2  −1/2
√


 −1/2  − 3/2
√
3/2 −1/2

(B. Gutkin)

Example 4.2. (Pseudo)real and complex representations. There are three types
of representation: real, pseudo-real and complex (see Montaldi for details). For real
representations matrices D(g) can be brought into real form such that Dij (g) = D̄ij (g).
This implies in particular that all the characters are real. For pseudo-real representation
the characters are also real but matrices D(g) cannot be brought into real form. Finally,
for complex representations the characters are complex. In the last case D(g) and
the conjugate D̄(g) constitute two different representation (since they characters are
different), while in the real and pseudo-real case both representations are equivalent,
i.e., D̄(g) = U D(g)U † .
Indicator. To distinguish between three types of representations one looks at the indi-
cator:
1 X (α) 2
Ind(α) = χ (g ) ∈ {1, 0, −1} , (4.6)
|G| g∈G
where 1, −1, 0 are obtained for real, complex and pseudo-real representations, respec-
tively.
Proof: For a general irreducible representation we have

D(α) (g) = U D̄(β) (g)U † , (4.7)

where α 6= β for a complex representation (since χ(α) (g) 6= χ̄(α) (g)) and α = β for real
and pseudo-real representations. From D(α) (g 2 ) = D(α) (g)D(α) (g) follows
mα X mα X
X 1 X (α) (β) †
Ind(α) = Uk,j Di,k (g)D̄j,n (g)Uni ,
i,j=1 g∈G
|G| g∈G
k,n=1

with mα being dimension of α. By the orthogonality theorem this expression is zero for
α 6= β which is the case of complex α. For real and pseudo-real representations we
have
1

Ind(α) = tr U Ū .
mα
Now note, that for α = β eq. (4.7) yields

D(α) (g)U Ū = U Ū D(α) (g).

By the first Schur’s lemma it follows then that U Ū = γI, or U = γU > which also implies
γ 2 = 1. This leaves only two possibilities γ = 1 for real and γ = −1 for pseudo-real
representations. In the first case we have U U > = I and Ind(α) =
1, while in the second
one U U > = −I and Ind(α) = −1. Note finally, that 1 = det U Ū = γ mα . So γ = −1
might appear only if mα is even. In other words, a pseudo-real irreducible representation
must be of even dimension.

birdtracks.eu/course3 week4 54 2021-07-29

 GROUP THEORY - WEEK 4. HARD WORK BUILDS CHARACTER

Example 4.3. Quaternions: Quaternion multiplication table is

{±1, ±i, ±j, ±k} i2 = j 2 = k 2 ; ij = k.

This group has five conjugate classes:

{1}, {−1}, {±i}, {±j}, {±k}.

The only possible solution for the equation 5i=1 m2i = 8 is mi = 1, i = 1, . . . 4, m5 = 2.

P
In addition to fully symmetric representation, the other three one-dimensional represen-
tations are easy to find: χ(1) = 1, χ(−1) = 1, while χ(i) = −1, χ(j) = −1, χ(k) = 1;
χ(i) = −1, χ(k) = −1, χ(j) = 1 or χ(k) = −1, χ(j) = −1, χ(i) = 1. The two-
dimensional representation can be find by the orthogonality relation:

2 + χ(−1) ± χ(k) ± χ(i) ± χ(j) = 0, =⇒ χ(−1) = −2, χ(k) = χ(i) = χ(j) = 0 .

Since the indicator equals

Ind = (2χ(1) + 6χ(−1))/8 = −1,

the last representation is pseudo-real. Note that this representation can be realized
using Pauli matrices:
{±I, ±σx , ±σy , ±σz }.
(B. Gutkin)

References
[1] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[2] V. Hart and H. Segerman, The quaternion group as a symmetry group, in Proc.
Bridges 2014: Mathematics, Music, Art, Architecture, Culture, edited by G. H.
G. Greenfield and R. Sarhangi (2014), pp. 143–150.
[3] O. Katz-Saporta and E. Efrati, “Self-driven fractional rotational diffusion of the
harmonic three-mass system”, Phys. Rev. Lett. 122, 024102 (2019).
[4] G. L. Kotkin and V. G. Serbo, Collection of Problems in Classical Mechanics
(Elsevier, 2013).
[5] L. D. Landau and E. M. Lifshitz, Quantum Mechanics: Non-Relativistic Theory
(Pergamon Press, Oxford, 1959).
[6] J. Mathews and R. L. Walker, Mathematical Methods of Physics (W. A. Ben-
jamin, Reading, MA, 1970).
[7] M. Tinkham, Group Theory and Quantum Mechanics (Dover, New York, 2003).

2021-07-29 55 birdtracks.eu/course3 week4

 EXERCISES

Exercises
4.1. Characters of D3 . (continued from exercise 2.4) D3 , the group of symmetries of an
equilateral triangle: has three irreducible representations, two one-dimensional and the
other one of multiplicity 2.

(a) All finite discrete groups are isomorphic to a permutation group or one of its sub-
groups, and elements of the permutation group can be expressed as cycles. Express
the elements of the group D3 as cycles. For example, one of the rotations is (123),
meaning that vertex 1 maps to 2, 2 → 3, and 3 → 1.
(b) Use your representation from exercise 2.4 to compute the D3 character table.
(c) Use a more elegant method from the group-theory literature to verify your D3 char-
acter table.
(d) Two D3 irreducible representations are one dimensional and the third one of multi-
plicity 2 is formed by [2×2] matrices. Find the matrices for all six group elements
in this representation.

4.2. Decompose a representation of S3 . As an illustration of the utility of the character

orthonormality relations (3.1), let’s work out the reduction of the matrix representation of
S3 permutations. The identity element acting on three objects [a b c] is a 3 × 3 identity
matrix,  
1 0 0
D(E) = 0 1 0
0 0 1
Transposing the first and second object yields [b a c], represented by the matrix
 
0 1 0
D(A) = 1 0 0
0 0 1

since     
0 1 0 a b
1 0 0  b  = a
0 0 1 c c

1. Find all six matrices for this representation.

2. Split this representation into its conjugacy classes.
3. Evaluate the characters χ(Cj ) for this representation.
4. Determine multiplicities ca of irreps contained in this representation.
5. (bonus) Construct explicitly all irreps.
6. (bonus) Explain whether any irreps are missing in this decomposition, and why.

4.3. All irreducible representations of D4 . Dihedral group D4 , the symmetry group of

a square, consists of 8 elements: identity, rotations by π/2, π, 3π/2, and 4 reflections
across symmetry axes: D4 = hg, σ|g 4 = σ 2 = e, gσ = σg 3 i

(a) Find all conjugacy classes.

birdtracks.eu/course3 week4 56 2021-07-29

EXERCISES

(b) Determine the dimensions of irreducible representations using the relationship

X 2
di = |G|, (4.8)
i

where di is the dimension of ith irreducible representation.

(c) Determine the remaining items of the character table.
(d) Compare with the character table of quaternions, example 4.3. Are they the same
or different?
(e) Determine the indicators for all irreps of D4 . Are they the same as for the irreps of
the quaternion group?
If you are at loss how to proceed, take a look at Landau and Lifschitz [5] Vol.3: Quantum
Mechanics
(Boris Gutkin)
4.4. Irreducible representations of dihedral group Dn .
(a) Determine the dimensions of all irreps of dihedral group Dn , n odd.
(b) Determine the dimensions of all irreps of dihedral group Dn , n even.
This exercise is meant to be easy - guess the answer from the irreps dimension sum rule
(4.8), and what you already know about D1 , D3 and D4 . Working out also D2 case
(cut a disk into two equal halves) might be helpful. A more serious attempt would require
counting conjugacy classes first. This exercise might help you later, when you are looking
at irreps of the orthogonal groups O(n); turns out they are different for n odd or even
n, and that has physical consequences: what you learn by working out a problem in 2
dimensions might be misleading for working it out in 3 dimensions.
4.5. Perturbation of Td symmetry.
A non-relativistic charged particle moves in an infinite bound potential V (x) with Td
symmetry. Consult exercise 5.1 Vibration Modes of CH4 for the character table and other
Td details.
(a) What are the degeneracies of the quantum energy levels? How often do they appear
relative to each other (i.e., what is the level density)?
A weak constant electric field is now added now along one of the 2π/3 rotation axes,
splitting energy levels into multiplets.
(b) What is the symmetry group of the system now?
(c) How are the levels of the original system split? What are the new degeneracies?
(Boris Gutkin)
4.6. Selection rules for Td symmetry.
The setup is the same as in exercise 4.5, but now assume that instead of a constant field, a
time dependent electric field E 0 cos(ωt) is added to the system, with E 0 not necessarily
directed along any of the symmetry axes. In general, when |En − Em | = ~ω, such
time-dependent perturbation induces transitions between energy levels En and Em .

(a) What are the selection rules? Between which energy levels of the system are tran-
sitions possible?

2021-07-29 57 birdtracks.eu/course3 week4

 EXERCISES

(b) Would the answer be different if a magnetic field B 0 cos(ωt) is added instead?
Explain how and why.

4.7. Two particles in a potential.

Two distinguishable particles of the same mass move in a 2-dimensional potential V (r)
having D4 symmetry. In addition they interact with each other with the term λW (|r 1 −
r 2 |).
(a) What is the symmetry group of the Hamiltonian if λ = 0? If λ 6= 0?.
(b) What are the degeneracies of the energy levels if λ = 0?
(c) Assuming that λ  1 (weak interaction), describe the energy level structure, i.e.,
degeneracies and quasi-degeneracies of the energy levels. What will be the answer
if the interaction is strong?
Hint: when interaction is weak we can think about it as perturbation. (Boris Gutkin)

birdtracks.eu/course3 week4 58 2021-07-29

 group theory - week 5

It takes class

Homework HW5

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort

Exercise 5.1 Vibration modes of CH4 , parts (a) (b) (c) i 8 points
Exercise 5.2 Keep it classy (a) 2 points

Bonus points
Exercise 5.1 Vibration modes of CH4 , part (c) ii 2 points
Exercise 5.2 Keep it classy (b) 2 points
Exercise 5.2 Keep it classy (c) 4 points

Total of 10 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

59
 GROUP THEORY - WEEK 5. IT TAKES CLASS

Show class, have pride, and display character. If you do, win-
ning takes care of itself.
— Paul Bryant

2021-06-03 Predrag Lecture 9 It takes class

Section playlist
◦ Sect. 5.1 It’s all about class
Dresselhaus et al. [6] Sect. 3.6 Second Orthogonality Relation for Characters.
6.1 A character table is a unitary matrix from classes to irreps (9:12 min)
6.2 Projection operator perspective (11:35 min)
Harter’s Sect. 3.2 First stage of non-Abelian symmetry analysis
group multiplication table (3.1.1); class operators; class multiplication table (3.2.1b);
all-commuting or central operators;
6.3 Example: the projection operator reduction of D3 (23:00 min)
Harter’s Sect. 3.3 Second stage of non-Abelian symmetry analysis
projection operators (3.2.15); 1-dimensional irreps (3.3.6); 2-dimensional irrep
(3.3.7); Lagrange irreps dimensionality relation (3.3.17)
6.4 Example: A reduction by two commuting operators (Harter problem 1.2.6).
What comes next: a nonlinear symmetry reduction; translations; Fourier series.
(35:14 min)

2021-06-08 Predrag Lecture 10 It takes grit

Gutkin notes, Lect. 5 Applications I. Vibration modes: Example 5.1. Cn sym-
metry completed.

5.1 It’s all about class

In week 1 we introduced projection operators (1.27). How are they related to the char-
acter projection operators constructed in the previous lecture? While the character or-
thogonality might be wonderful, it is not very intuitive - it’s a set of solutions to a set of
symmetry-consistent orthogonality relations. You can learn a set of rules that enables
you to construct a character table, but it does not tell you what it means. Similar thing
will happen again when we turn to the study of continuous groups: all semisimple Lie
groups will be classified by Killing and Cartan by a more complex set of orthogonal-
ity and integer-dimensionality (Diophantine) constraints. You obtain all possible Lie
algebras, but have no idea what their geometrical significance is.
In my own Group Theory book [4] I (almost) get all simple Lie algebras using
projection operators constructed from invariant tensors. What that means is easier to
understand for finite groups, and here I like the Harter’s exposition [8] best. Harter

birdtracks.eu/course3 week5 60 2021-07-29

 GROUP THEORY - WEEK 5. IT TAKES CLASS

constructs ‘class operators’, shows that they form a basis for the algebra of ‘central’
or ‘all-commuting’ operators, and uses their characteristic equations to construct the
projection operators (1.27) from the ‘structure constants’ of the finite group, i.e., its
class multiplication tables. Expanded, these projection operators are indeed the same
as the ones obtained from character orthogonality.
I find Harter’s Sect. 3.3 Second stage of non-Abelian symmetry analysis particu-
larly illuminating. It shows how physically different (but mathematically isomorphic)
higher-dimensional irreps are constructed corresponding to different subgroup embed-
dings. One chooses the irrep that corresponds to a particular sequence of physical
symmetry breakings.
You might want to have a look at Harter [9] Double group theory on the half-shell
(click here). Read appendices B and C on spectral decomposition and class algebras.
Article works out some interesting examples.
See also remark 1.1 Projection operators and perhaps watch Harter’s online lecture
from Harter’s online course.
There is more detail than what we have time to cover here, but I find Harter’s
Sect. 3.3 Second stage of non-Abelian symmetry analysis particularly illuminating. It
shows how physically different (but mathematically isomorphic) higher-dimensional
irreps are constructed corresponding to different subgroup embeddings. One chooses
the irrep that corresponds to a particular sequence of physical symmetry breakings.

5.1.1 Dirac characters, Burnside’s method (optional)

I told you that everybody who understands anything about group theory, writes a book.
This weeks winner is Daniel Arovas, who is writing up his Lecture Notes on Group
Theory in Physics. Check them out - they are cute, and even contain !jokes!.
For example, here I learn for the first time that Harter’s central operators ( Harter’s
Sect. 3.2 First stage of non-Abelian symmetry analysis) are in condensed matter physics
known as ‘Dirac characters’.
Dirac characters were introduced by Dirac [5] in The Principles of Quantum Me-
chanics (1930) (click here). He refers to them as “[...] what is called in group theory
a character of the group of permutations.” Corson [3] Note on the Dirac character
operators (1948) writes:

[...] the evaluation of Dirac and similar character operators is all that is
required for the solution of the standard molecular problems in the spirit
of Dirac’s original program which avoids appeal to formal group theory.

Dirac characters use not only the abstract group information, but also account for
the symmetry information contained in the basis set used. The diagonalization of Dirac
characters has three main advantages:
1. It can be realized by means of a quite simple and general algorithm.
2. The projective irreps obtained are just the ones that are needed to reduce the
starting basis set into irreducible sets.
3. No tabulated quantities are required to construct the projective irreps.

2021-07-29 61 birdtracks.eu/course3 week5

 GROUP THEORY - WEEK 5. IT TAKES CLASS

The scheme is completely general, in the sense that it applies to all space groups.

Ananda Dasgupta had 1.68K followers on YouTube, now he has one more:

playlist for his Symmetries in Physics course:

Lecture 15 (start at about 35 min into the lecture) has a nice discussion of Dirac
characters, their relation to characters, and motivates the algorithmic Burnside’s
method for computing characters via class multiplication tables (Hi )jk .

PH4213 Discussion Class 8 applies Burnside’s method to D4 .

More generally, the whole course is of interest, if covers most topics of our course
in greater depth:

PH4213 Discussion Class 9 gets projection operators out of characters.

Lecture 16 Projection operators attempts to give you an appreciation of the

power of the Wigner Eckart theorem (what in my book is described as all calcu-
lations being ‘vacuum bubbles’, maybe not precisely in these words).

A few textbooks that use Dirac characters:

Cini [2] Topics and Methods in Condensed Matter Theory (2007) (click here)

Jacobs [10] Group Theory with Applications in Chemical Physics, (click here)
(2005)

El-Batanouny and Wooten [1] Symmetry and Condensed Matter Physics: A Com-
putational Approach (2008) (click here). In sect. 4.3 they describe the Burnside’s
method. They give an example of Mathematica code that constructs the charac-
ter table. If needed, on might use Dixon’s method, which is more clever for
numerical computations.

Big Chemical Encyclopedia: Dirac character.

The CRYSTAL package performs ab initio calculations of the ground state energy,
energy gradient, electronic wave function and properties of periodic systems.
Uses Dirac characters.

For a bit of history, see J. E. Humphreys review of Pioneers of representation

theory.

5.1.2 William G. Harter (optional)

Who is Bill Harter? He is a prodigy who at age 16 taught himself group theory by
reading Hamermesh [7]. He was a graduate student at Caltech (1964-65), together with
Ron Fox. They hated the atmosphere there and the teaching was terrible (Feynman did
not teach that year but Harter and Feynman were good friends). Harter and Fox shared
an interest in group theory and discovered that most of the group theory books in the

birdtracks.eu/course3 week5 62 2021-07-29

 GROUP THEORY - WEEK 5. IT TAKES CLASS

physics library had been checked out in 1960-62 by Gell-Mann, Zweig and Glashow.
That only half of the entering students were meant to complete their PhD’s there led to
lots of ugly competition. Harter transferred to UC Irvine, and, upon graduation, got a
job at USC in LA. After a few years he suggested in a faculty meeting that the way they
could improve their quality as a department was “to get rid of all the old farts." These
same “old farts" soon voted to deny him tenure. He ended up in Campinas, Brazil. Fox
rescued him from there by bringing him for an interview at Georgia Tech, where he
was hired in late 1970’s. He was brilliant, an asset for teaching, making all sorts of
demonstration devices. He built a giant rotating table upon which he placed billiard
balls, a wonderful demonstration of mechanical analogues for charged particle motion
in crossed E and B fields. Everyone (except for one nefarious character) liked him, his
work, and especially his devices. The faculty unanimously supported his promotion
to tenure. He did not, however, think much of the Director of School of Physics, and
made that clear. After an argument with the Director, he stormed out, offended. So, he
was denied tenure and moved in 1985 to University of Arkansas where he is a professor
today.
In 1987 Harter and Weeks used Harter’s theory of the rotational dynamics of mole-
cules to calculate the rotational-vibrational spectra of the soccer ball-shaped molecule
Buckminsterfullerene, C60, or “buckyball." C60 had been proposed in 1985 by chemists,
who had seen a mass-spectra peak of atomic mass 720. By 1989 the Harter theory cal-
culations led to a realization that chemists had been making C60 since the early 1970s.
In 1992 Science named C60 “Molecule of the Year," and in 1996 Curl, Kroto and
Smalley were awarded the Nobel Prize in Chemistry for their discovery of fullerenes.
You can find here many Soft Elegant Educational tools developed by Harter, and
follow his lectures on line. He is a great teacher. Georgia Tech’s loss.

5.2 Other sources (optional)

Continuing reading Mathews and Walker [11], now Chap. 14. Porter works out nicely
the normal modes of the D3 springs and masses (again!).
Not all finite groups are as simple or easy to figure out as D3 . For example, the order
of the Ree group 2 F4 (2)0 is 212(26 + 1)(24 − 1)(23 + 1)(2 − 1)/2 = 17 971 200 .

5.3 Discussion
Henriette Roux I have a few questions about the exercise 5.1 part (d) Vibration modes
of CH4 : Find all modes of the methane molecule.

1. When we use the angle of improper rotation, is it true that reflection equals
to the π improper rotation?
2. I assume it is π and it gives me other characters are zero. In the case of all
symmetry, this will give the , which we usually get non-negative integer.
As a result, I’m not perfectly sure that the character formulas you give are
correct.

2021-07-29 63 birdtracks.eu/course3 week5

 GROUP THEORY - WEEK 5. IT TAKES CLASS

3. Moreover seems it’s in the representation of [12 × 12] matrices instead of

[24 × 24] matrices.
Predrag The solution set is very detailed, so how about waiting Tuesday afternoon,
when it gets posted on T-square? Then –if it is still unclear– we continue the
discussion.

1. If g ∈ SO(3) is a rotation, and D(i)r = −r is the inversion transformation,

then rotation combined with the inversion g i is an improper rotation g i ∈
O(3). If g ∈ T (a discrete tetrahedron rotation) then g i is an improper
element of Td .
2. ? (check the solution set).
3. The proper rotations group T of order 12 is a normal subgroup. However, I
do not think you can have an improper rotations subgroup of Td , as gi igj i
is a proper rotation.

References
[1] M. El-Batanouny and F. Wooten, Symmetry and Condensed Matter Physics: A
Computational Approach (Cambridge Univ. Press, Cambridge UK, 2008).
[2] M. Cini, Topics and Methods in Condensed Matter Theory - From Basic Quan-
tum Mechanics to the Frontiers of Research (Springer, Berlin, 2007).
[3] E. M. Corson, “Note on the Dirac character operators”, Phys. Rev. 73, 57–60
(1948).
[4] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[5] P. A. M. Dirac, The Principles of Quantum Mechanics (Oxford Univ. Press,
1930).
[6] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[7] M. Hamermesh, Group Theory and Its Application to Physical Problems (Dover,
New York, 1962).
[8] W. G. Harter, Principles of Symmetry, Dynamics, and Spectroscopy (Wiley, New
York, 1993).
[9] W. G. Harter and N. dos Santos, “Double-group theory on the half-shell and the
two-level system. I. Rotation and half-integral spin states”, Amer. J. Phys. 46,
251–263 (1978).
[10] P. Jacobs, Group Theory with Applications in Chemical Physics (Cambridge
Univ. Press, 2005).
[11] J. Mathews and R. L. Walker, Mathematical Methods of Physics (W. A. Ben-
jamin, Reading, MA, 1970).

birdtracks.eu/course3 week5 64 2021-07-29

EXERCISES

a) b)

Figure 5.1: a) Two classes of rotational symmetries, and a class of reflection symme-
tries of a tetrahedron. (left) Hold the Tetra Pak by a tip, turn it by a third. (middle)
Hold the Tetra Pak by the midpoints of a pair of opposing edges, make a half-turn.
(right) Exchange the vertices outside the reflection plane. b) Methane molecule with
the symmetry Td .

Exercises
5.1. Vibration modes of CH4 .
Tetrahedral group T describes rotational symmetries of a tetrahedron. The order of the
group is |T| = 12, and its conjugacy classes are:
• The identity mapping.
• Four rotations by ϕ = 2π/3, with each of the four rotation axes going through a
vertex, and piercing the midpoint of the triangle opposite.
• Four inverse rotations by ϕ = −2π/3.
• Three rotations by ϕ = π, one for each of the three rotation axes going through
midpoints of opposing edges.
The full group of tetrahedron symmetries Td includes also reflections. This is the sym-
metry group of molecules such as methane CH4 , see figure 5.1).
(a) What is the order of the group Td ? Show that the group is isomorphic to i) the
group of permutations S4 ; ii) to the group O of rotational symmetries of the cube.
iii) Show that T is normal subgroup of Td .
(b) Find all conjugacy classes of the group. Which of these classes correspond to proper
(det R(ϕ) = +1), improper (det R(ϕ) = −1) rotations ?
Information on T might help. Note that ϕ might be also 0.
(c) i) Find all irreducible representations of the group & build the character table.
A shortcut: find all one-dimensional representations, assume that characters are
integers, then use the orthogonality relationship between characters.
ii) Really compute the character table, without assuming that characters are integers
(2 bonus points).
One-dimensional representations + orthogonality of characters is not enough to
build the whole character table for Td . One needs more black magic, such as rep-
resentation of permutation group by matrices.
(d) Find all modes of the methane molecule. Which of them correspond to vibrations,
translations and rotations? What are the degeneracies?

2021-07-29 65 birdtracks.eu/course3 week5

 EXERCISES

Figure 5.2: Three identical masses are constrained to move on a hoop, connected by
three identical springs such that the system wraps completely around the hoop. Find
all symmetries of the equations of motion.

Path: Find characters of the full (reducible) representation by using formulas from
the lecture:
(
ng (1 + 2 cos(ϕ)) rotation,
χ(g) =
ng (−1 + 2 cos(ϕ)) improper rotation .

Here ng is the number of atoms staying at the same place under the action of g, ϕ is
the rotation angle corresponding to g = R(ϕ). Then decompose this representation
into irreducible representations. Identify the rotational and translational parts.
(e) To what representation corresponds the most symmetric ”breathing“ mode and
why? Is it infrared active, i.e., can this mode can be excited by electromagnetic
field?
(B. Gutkin)
5.2. Keep it classy. Check out Harter’s PowerPoint presentation :)
(a) Go through the derivation of the three projection operators for D3 .
(3) (3) (3)
(b) Decompose P(3) = P1 + P2 . Construct Pij . Verify that they are idempotent.
(d) Compute the [2 × 2] irreducible matrix representation D(3) (g)ij for every group
element g, in the spirit of Harter’s slides 13-8 and 13-9.
5.3. Three masses on a loop. (Exercise 2.7 revisited.) Three identical masses, connected
by three identical springs, are constrained to move on a circle hoop as shown in figure 5.2.

(a) Find all symmetries of the equations of motion.

(b) Find the normal modes using group-theoretic decompositions to irreps and charac-
ter orthonormality.
(c) How many eigenvalues are there in all?
(d) Interpret the eigenvalues and eigenvectors from a group-theoretic, symmetry point
of view.

birdtracks.eu/course3 week5 66 2021-07-29

 group theory - week 6

For fundamentalists

Homework HW6

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 6.1 3-disk symbolic dynamics 2 bonus points

Exercise 6.2 Reduction of 3-disk symbolic dynamics to binary 3 bonus points
Exercise 6.3 3-disk fundamental domain cycles 2 bonus points
Exercise 6.4 C2 -equivariance of Lorenz system 3 points
Exercise 6.5 Proto-Lorenz system parts 1.-5. 7 points
Exercise 6.5 Proto-Lorenz system parts 6.-8. 6 bonus points

Total of 10 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

67
 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

2021-06-08 Predrag Lecture 11 True grit

Lecture 7 (Unedited)
◦ We work out the symmetry reduction and a breaking of the D3 symmetry in
the [3 × 3] permutation matrices representation
◦ A preview; reduction of a cyclic CN symmetric molecule chain to a [2 × 2]
frequency matrix calculation. It will take a couple of weeks of discrete
Fourier lectures to fill in the details. (2:07:51 h)
So far we have covered what any QM fixated Group Theory textbook since
1930’s and on covers. Today to turn to what we actually use group theory for
today, here, in Howey, and for that there is no book but ChaosBook.org.
Many fundamental problems of fluid dynamics and more generally non-linear
field theories are studied in experimental settings equipped with symmetries.
That is the subject of dynamical systems theory (of which classical, quantum
and stochastic mechanics and field theories are but specialized branches). We
start gently, with the famed Lorenz butterfly.
2021-06-10 Predrag Lecture 12 Nonlinear symmetry reduction
Lecture 8 (Unedited) Symmetries and nonlinear systems. First a mini-course
on nonlinear dynamics and chaos. I use the symmetry reduction on the
Lorenz, and turn it into Van Gogh. Conclusion: can use either a funda-
mental domain, or invariant polynomial bases to reduce symmetries of a
nonlinear system. (2:28:00 h)
◦ Lorenz flow example. Read ChaosBook Chapter 11. World in a mirror Chaos-
Book.org Chapter 11 World in a mirror. Maybe start with ChaosBook ex-
ample 10.6 Equivariance of the Lorenz flow, example 11.8 Desymmetriza-
tion of Lorenz flow, and then work your way back if needed.
◦ example 6.1 Equivariance of the Lorenz flow
◦ example 6.2 Desymmetrization of Lorenz flow
The reading and the homework for this week, is augmented by - if you find that
helpful - by ‘live’ online blackboard lectures: click here.

6.1 Other sources (optional)

An example: a 1-dimensional system with a symmetry
Fundamental domain
Tiling of state space by a finite group
Make the “fundamental tile" your hood
Symmetry-reduced dynamics
Regular representation of permuting tiles

birdtracks.eu/course3 week6 68 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

6.2 Thoughts (optional)

How I think of the fundamental domain is explained in my online lectures, Week 14,
in particular the snippet Regular representation of permuting tiles.
Unfortunately - if I had more time, that would have been shorter, this goes on and
on, Week 15, lecture 29. Discrete symmetry factorization, and by the time the dust
settles, I do not have a gut feeling for the boundary conditions when it comes to higher-
dimensional irreps (see also last week’s sect. 6.2 Discussion).
The basic insight is that if the symmetry and dynamics commute, one can imple-
ment the stratification of the state space by the symmetry first, paying no heed to the
dynamics. In arbitrary coordinates, the state space is a stratified by a jumble of group
orbits. It is an ‘orbitfold’, in the sense that in general it contains subspaces on which
group orbits are of the dimension of a symmetry subgroup, with the group action on
invariant subspaces trivial, and on which group orbits are points.
On the linear level, the natural stratification is implemented by decomposing the
state space into irreps of the symmetry group. This is a linear reshuffling of coor-
dinates that makes the action of the symmetry operators as simple as possible. You
can think of the new basis vectors as eigenvectors of the symmetry operators (Fourier
modes, spherical eigenfunctions, etc.). The nonlinear terms in dynamical equations
jumble everything up. They are re-expressed in this basis using Kronecker-product
decompositions into sums over products of irreps.
Unfortunately –if I had more time, that would have been shorter– this goes on and
on, ChaosBook course 2, Week 15, lecture 29. Discrete symmetry factorization.

Henriette Roux What do the parameters σ, ρ and b stand for in the Lorenz equations
(6.4)?

Predrag The short answer is the truncation of the Navier-Stokes that leads to Lorenz
equations is so drastic that they have no longer any physical meaning; in his
1963 paper [17] Lorenz played with the parameters until he empirically found
an interesting example of deterministic chaos. Since then, applied mathemati-
cians have reverse-engineered various physical systems to find situations where
parameters σ, ρ and b mean something, see remark 6.1 (copied to here from
ChaosBook.org). The discrete symmetry of the original Navier-Stokes system
(‘left’ is as good as ‘right’) happened to survive the drastic truncation from 105
Fourier modes (for physically accurate simulations) to 3. I prefer to teach non-
linear dynamics using the Rössler system, precisely because it has no discrete
symmetry, just chaos.

6.3 ChaosBook notes

Copied here are a few snippets from this week’s lecture notes, needed here just because
exercises refer to them - read the full lecture notes instead.

2021-07-29 69 birdtracks.eu/course3 week6

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

(a) (b)

Figure 6.1: (a) The pair of full-space 9-cycles, the counter-clockwise 121232313 and
the clockwise 131323212 correspond to (b) one fundamental domain 3-cycle 001.

Definition: Flow invariant subspace. A typical point in fixed-point subspace MH

moves with time, but, due to equivariance
f (gx) = gf (x) , (6.1)
its trajectory x(t) = f t (x) remains within f (MH ) ⊆ MH for all times,
hf t (x) = f t (hx) = f t (x) , h∈H, (6.2)
i.e., it belongs to a flow invariant subspace. This suggests a systematic approach to
seeking compact invariant solutions. The larger the symmetry subgroup, the smaller
MH , easing the numerical searches, so start with the largest subgroups H first.
We can often decompose the state space into smaller subspaces, with group acting
within each ‘chunk’ separately:

Definition: Invariant subspace. Mα ⊂ M is an invariant subspace if

{Mα | gx ∈ Mα for all g ∈ G and x ∈ Mα } . (6.3)
{0} and M are always invariant subspaces. So is any Fix (H) which is point-wise
invariant under action of G.

Definition: Irreducible subspace. A space Mα whose only invariant subspaces un-

der the action of G are {0} and Mα is called irreducible.
Example 6.1. Equivariance of the Lorenz flow. The velocity field in Lorenz equa-
tions [17]
        
ẋ σ(y − x) −σ σ 0 x 0
 ẏ  =  ρx − y − xz  =  ρ −1 0   y  +  −xz  (6.4)
ż xy − bz 0 0 −b z xy

is equivariant under the action of cyclic group C2 = {e, r} acting on R3 by a π rotation

about the z axis,
r(x, y, z) = (−x, −y, z) . (6.5)

birdtracks.eu/course3 week6 70 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

Table 6.1: D3 correspondence between the binary labeled fundamental domain prime cycles
p̃ and the full 3-disk ternary labeled cycles p, together with the D3 transformation that maps
the end point of the p̃ cycle into the irreducible segment of the p cycle. White spaces in the
above ternary sequences mark repeats of the irreducible segment; for example, the full space
12-cycle 1212 3131 2323 consists of 1212 and its symmetry related segments 3131, 2323. The
multiplicity of p cycle is mp = 6np̃ /np . The shortest pair of fundamental domain cycles related
by time reversal (but no spatial symmetry) are the 6-cycles 001011 and 001101.
p̃ p gp̃ p̃ p gp̃
0 12 σ12 000001 121212 131313 σ23
1 123 C 000011 121212 313131 232323 C2
01 12 13 σ23 000101 121213 e
001 121 232 313 C 000111 121213 212123 σ12
011 121 323 σ13 001011 121232 131323 σ23
0001 1212 1313 σ23 001101 121231 323213 σ13
0011 1212 3131 2323 C2 001111 121231 232312 313123 C
0111 1213 2123 σ12 010111 121312 313231 232123 C2
00001 12121 23232 31313 C 011111 121321 323123 σ13
00011 12121 32323 σ13 0000001 1212121 2323232 3131313 C
00101 12123 21213 σ12 0000011 1212121 3232323 σ13
00111 12123 e 0000101 1212123 2121213 σ12
01011 12131 23212 31323 C 0000111 1212123 e
01111 12132 13123 σ23 ··· ··· ···

Example 6.2. Desymmetrization of Lorenz flow: (continuation of example 6.1) Lorenz

equation (6.4) is equivariant under (6.5), the action of order-2 group C2 = {e, r}, where
r is [x, y]-plane, half-cycle rotation by π about the z-axis:

(x, y, z) → r(x, y, z) = (−x, −y, z) . (6.6)

(r)2 = 1 condition decomposes the state space into two linearly irreducible subspaces
M = M+ ⊕ M− , the z-axis M+ and the [x, y] plane M− , with projection operators
onto the two subspaces given by
   
0 0 0 1 0 0
+ 1 − 1
P = (1 + r) =  0 0 0  , P = (1 − r) =  0 1 0  . (6.7)
2 2
0 0 1 0 0 0

so
      
ẋ− −σ σ x− 0
= +
ẏ− ρ −1 y− −z x−
1
ż+ = −b z+ + (x+ + x− )(y+ + y− ) , (6.8)
4
where z+ = z. As (ẋ+ , ẏ+ ) = (0, 0), values of (x+ , y+ ) are conserved parts of the initial
condition. We define the fundamental domain by the (arbitrary) condition x̂− ≥ 0, and
whenever exits the domain,we replace the function dependence by the corresponding
fundamental domain coordinates,

(x− , y− ) = r(x̂− , ŷ− ) = (−x̂− , −ŷ− ) if x− < 0 ,

2021-07-29 71 birdtracks.eu/course3 week6

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

(a) (b)

Figure 6.2: (a) Lorenz flow cut by y = x Poincaré section plane P through the z axis
and both E1,2 equilibria. Points where flow pierces into section are marked by dots.
To aid visualization of the flow near the E0 equilibrium, the flow is cut by the second
Poincaré section, P 0 , through y = −x and the z axis. (b) Poincaré sections P and P 0
laid side-by-side. (E. Siminos)

and record that we have applied r (that is the ‘reconstruction equation’ in the case of a
discrete symmetry). When we integrate (6.8), the trajectory coordinates (x̂− (t), ŷ− (t))
are discontinuous whenever the trajectory crosses the fundamental domain border.
That, however, we do not care about - the only thing we need are the Poincaré sec-
tion points and the Poincaré return map in the fundamental domain.
Poincaré section hypersurface can be specified implicitly by a single condition, through
a function U (x) that is zero whenever a point x is on the Poincaré section,

x̂ ∈ P iff U (x̂) = 0 . (6.9)

In order that there is only one copy of the section in the fundamental domain, this con-
dition has to be invariant, U (gx̂) = U (x̂) for g ∈ G, or, equivalently, the normal to it has
to be equivariant
∂j U (gx̂) = g∂j U (x̂) for g ∈ G . (6.10)
There are two kinds of compact (finite-time) orbits. Periodic orbits x(Tp ) = x(Tp )
are either self dual under rotation r, or appear in pairs related by r; in the fundamental
domain there is only one copy x̂(Tp ) = x̂(Tp ) of each. Relative periodic orbits (or ‘pre-
periodic orbits’) x̂(Tp ) = rx(Tp ) they are periodic orbits.
As the flow is C2 -invariant, so is its linearization ẋ = Ax. Evaluated at E0 , A com-
mutes with r, and the E0 stability matrix A decomposes into [x, y] and z blocks.
The 1-dimensional M+ subspace is the fixed-point subspace, with the z-axis points
left point-wise invariant under the group action

M+ = Fix (C2 ) = {x ∈ M | g x = x for g ∈ {e, r}} (6.11)

(here x = (x, y, z) is a 3-dimensional vector, not the coordinate x). A C2 -fixed point x(t)
in Fix (C2 ) moves with time, but according to (6.2) remains within x(t) ∈ Fix (C2 ) for all
times; the subspace M+ = Fix (C2 ) is flow invariant. In case at hand this jargon is a bit

birdtracks.eu/course3 week6 72 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

of an overkill: clearly for (x, y, z) = (0, 0, z) the full state space Lorenz equation (6.4) is
reduced to the exponential contraction to the E0 equilibrium,

ż = −b z . (6.12)

However, for higher-dimensional flows the flow-invariant subspaces can be high-dim-

ensional, with interesting dynamics of their own. Even in this simple case this subspace
plays an important role as a topological obstruction: the orbits can neither enter it nor
exit it, so the number of windings of a trajectory around it provides a natural, topological
symbolic dynamics.
The M− subspace is, however, not flow-invariant, as the nonlinear terms ż = xy−bz
in the Lorenz equation (6.4) send all initial conditions within M− = (x(0), y(0), 0) into
the full, z(t) 6= 0 state space M/M+ .
By taking as a Poincaré section any r-equivariant, non-self-intersecting surface that
contains the z axis, the state space is divided into a half-space fundamental domain
M̃ = M/C2 and its 180o rotation rM̃. An example is afforded by the P plane section
of the Lorenz flow in figure 6.3. Take the fundamental domain M̃ to be the half-space
between the viewer and P. Then the full Lorenz flow is captured by re-injecting back
into M̃ any trajectory that exits it, by a rotation of π around the z axis.
As any such r-invariant section does the job, a choice of a ‘fundamental domain’
is here largely mater of taste. For purposes of visualization it is convenient to make
the double-cover nature of the full state space by M̃ explicit, through any state space
redefinition that maps a pair of points related by symmetry into a single point. In case at
hand, this can be easily accomplished by expressing (x, y) in polar coordinates (x, y) =
(r cos θ, r sin θ), and then plotting the flow in the ‘doubled-polar angle representation:’

(x̂, ŷ, z) = (r cos 2θ, r sin 2θ, z) = ((x2 − y 2 )/r, 2xy/r, z) , (6.13)

as in figure 6.4 (a). In contrast to the original G-equivariant coordinates [x, y, z], the
Lorenz flow expressed in the new coordinates [x̂, ŷ, z] is G-invariant. In this representa-
tion the M̃ = M/C2 fundamental domain flow is a smooth, continuous flow, with (any
choice of) the fundamental domain stretched out to seamlessly cover the entire [x̂, ŷ]
plane.
(E. Siminos and J. Halcrow)

6.4 Chaotic 3-spring, integrable 3-vortex systems (op-

tional)
Continued from sect. 4.1.

Simon Berman According to the 2019 Phys. Rev. Lett., of Katz-Saporta and Efrati [15],
Self-driven fractional rotational diffusion of the harmonic three-mass system, a
system of three masses connected by harmonic springs might be the simplest
mechanical system (homonuclear triatomic molecule, such as ozone, except the
three couplings are not the same) that exhibits a geometric phase. Away from
its resting configuration the system is nonlinear, and once its rotational SO(2)
symmetry is reduced, and as its energy is increased, it exhibits all kinds of shape-
dependent chaotic geometric phases. Katz and Efrati [15] mostly do numerical

2021-07-29 73 birdtracks.eu/course3 week6

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

simulations and plot displacement vs. time diffusion plots in its 6D phase space,
like this is still early 1960’s. The earlier arXiv:1706.09868 version has more
information than the PRL (no doubt thanks to impatience of the referees, plus
space constraints of a PRL). One suspects that a bit of thinking along periodic
orbit theory lines could yield some insight into the diffusive properties of its
shape-changing dynamics.
In the symmetry-reduced or the ‘shape’ state space there is a D3 symmetry. One
sees it in their [15] Hamiltonian (2): the bij vectors can be viewed as the three
coordinates of an equilateral triangle in the w1 −w2 plane. Since the Hamiltonian
only depends on |w| and in a symmetric way on w · bij , it has a D3 symmetry for
(w1 , w2 ) components of the w vector, and a reflection symmetry for w3 . So the
total symmetry group is D3 × C 1/2 .

Predrag As the system is D3 symmetric, the symmetry should be quotiented as in

(this week’s lectures) and ChaosBook.org. The students from Weizmann (as
well as all our local plumber apprentices) believe they have been born knowing
everything, and thus they do not need to take ChaosBook.org/course1, so they
would have no idea that

• they are supposed to quotient the symmetry

• probability densities (eigenfunctions of the evolution operator; Perron--
Frobenius and its generalizations) block diagonalize as irreps of D3 , and
• that makes all calculations, numerical and periodic orbit-type more trans-
parent and more convergent.

By going to relative w’s coordinates, one has quotiented only the 2D Euclidean
translations and SO(2) rotations, no discrete symmetries, so D3 still remains.
Now, anyone who has taken ChaosBook.org/course1 knows that the next step
is to quotient D3 , and do the calculation in the 1/6th of the phase space, i.e., the
fundamental domain.
I’m curious whether I’m right, because soon we’ll look at space groups (infinite
lattices with discrete symmetries) and there I have confused understanding of
how to quotient the space group, but that is related to diffusion in space, rather
than the angular diffusion, as in this 3-springs system.
We can make this a course project for a student in this course (a project instead
of taking the final). To be especially pedagogical, we’ll ask them to do it in Julia
(there is one potential candidate on Piazza).

Predrag proposal: 2-body, 3-spring system We need the simplest illustration of a

geometric phase, and its diffusion along the continuous symmetry direction in-
duced by chaotic (“turbulent”) shape-changing dynamics. So let’s take one of
the masses infinite. Still 3 springs, but only 2 bodies moving in a plane. We still
have SO(2) continuous symmetry to reduce. What remains is the D2 = {e, σ}
symmetry of exchanging the two particles, with two irreps, the symmetric and
the antisymmetric normal modes. There is shape-changing dynamics, with the

birdtracks.eu/course3 week6 74 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

potential a nonlinear function of wj ’s, so for larger energies we expect angu-

lar geometric phase diffusion, but in a lower-dimensional phase space than that
of the free 3-springs system. Easier to work out and look at Poincaré sections,
search for relative equilibria and relative periodic orbits, compute the angular
diffusion constant from its cycle expansion formulation.

Predrag: N vortex system Went to hear Tomoki Ohsawa ( Google Scholar), talk
about Symplectic reduction and the Lie–Poisson dynamics of point vortices on
the plane, arXiv:1808.01769 .
I had previously written to Tomoki’s friend Molei about how much I had already
suffered through Weinstein, Marsden, etc. moment maps, for decades. We all
have to do symmetry reductions, but with Marsden it is always the moment map,
and then the climax is the rigid 3D body example which is the end-all of every
article and book. Perhaps due to my pleas, Ohsawa gave us a gentle, sensible
seminar, Weinstein-Marsden for humans, where he explained why moment map
is called ‘moment,’ etc.. As nice a birthday present one could hope for, see the
slides here.
Ohsawa develops a Hamiltonian formulation of the dynamics of the “shape" of
N point vortices on the plane and the sphere. If N = 3, it is the dynamics of the
shape of the triangle formed by three point vortices, regardless of the position
and orientation of the triangle on the plane/sphere.
For the planar case, reducing the basic equations of point vortex dynamics by the
special Euclidean group SE(2) yields a Lie-Poisson equation for relative config-
urations of the vortices. The shape dynamics is periodic in certain cases. The
approach can be extended to the spherical case by first lifting the dynamics from
the two-sphere to C 2 and then performing reductions by symmetries.

Figure 6.3: (green) Level set of quadratic Casimir C2 ellipsoid. (orange) Level set of
Hamiltonian h. The intersection is the unique periodic orbit of the symmetry-reduced
N = 3 vortex system. See also the corresponding figure for 3 vortices on the sphere
on p. 46 of the slide presentation. T. Ohsawa

2021-07-29 75 birdtracks.eu/course3 week6

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

I think Ohsawa discovery is that the system has a previously un-noted quartic
Casimir, whose invariance reduces the dimension of the symmetry reduced phase
space by one degree of freedom (dof). The implies that the symmetry-reduced
dynamics in the N = 3 case is 1 dof, i.e., integrable, see figure 6.3. In addition,
if the sum of the vortex circulations is 0, the N = 4 case is integrable. This fact
is not yet explained - my intuition is that the zero total circulation implies extra
rotational symmetry. For more vortices I expect the usual Hamiltonian mixed
phase space.
Parenthetically, statement that there is quartic order Casimir that is invariant un-
der the symmetry group can probably be written as a syzygy constraint on the
invariant polynomial basis in (Hilbert’s) theory of invariant polynomial bases.
In the Katz-Saporta and Efrati [15] example there is no quartic Casimir, so one
ends with a generic chaotic system. Ohsawa’s geometric technique works be-
cause of the simple symplectic structure on the point-vortex problem (there is no
‘momentum’), whereas Katz-Saporta and Efrati problem is a standard classical-
mechanical one on the cotangent bundle of a configuration space, with momen-
tum there. Ohsawa believes that one can apply the techniques developed by
Richard Montgomery to this setting as well. (Montgomery’s paper motivated
him to work on the point-vortex problem).
There are also examples in cardiac (!) dynamics where one must reduce 2D
Euclidean symmetry first, with similar outcome to yours, but no moment maps,
as such PDEs have no variational formulation (that I am aware of). Googling
“Barkley model" might do the trick. I do not think there is a variational (La-
grangina) formulation.
But that is the whole point - any flow with a symmetry has to have the symmetry
quotiented out. It’s easier to understand this for flows which are not symplectic
- in that case, every continuous symmetry parameter reduces the dimension of
the symmetry-reduced state space by one. The Hamiltonin case is a pain (or
bliss, if you love moment maps) because every continuous symmetry reduces
the dimension of the phase space by one degree of freedom (ie, by 2). Also
variational problems obey Noether’s theorem, our (dissipative) problems usually
do not. If I understand this right...

Kimberly Short Foulkes PhD thesis [7] Drift and Meander of Spiral Waves (2009)
might be an user friendly introduction for students that need to understand Eu-
clidean symmetry? Covers refs. [2, 8]. Page 14 solves a differential equation
with SE(2) symmetry. Appendix 8.9 discusses symmetries, and gives the con-
dition for equivariance.

6.5 Eigenfunctions (optional)

What follows is an inconclusive discussion of eigenfunctions over fundamental do-
mains - feel free to ignore...

birdtracks.eu/course3 week6 76 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

(a) (b)

Figure 6.4: (a) Lorenz attractor plotted in [x̂, ŷ, z], the doubled-polar angle coordinates
(6.13), with points related by π-rotation in the [x, y] plane identified. Stable eigen-
vectors of E0 : e(3) and e(2) , along the z axis (6.12). Unstable manifold orbit W u (E0 )
(green) is a continuation of the unstable e(1) of E0 . (b) Blow-up of the region near E1 :
The unstable eigenplane of E1 defined by Re e(2) and Im e(2) , the stable eigenvector
e(3) . The descent of the E0 unstable manifold (green) defines the innermost edge of
the strange attractor. As it is clear from (a), it also defines its outermost edge. (E.
Siminos)

Predrag Heilman and Strichartz [13] Homotopies of Eigenfunctions and the Spectrum
of the Laplacian on the Sierpinski Carpet, arXiv:0908.2942, is not an obvious
read for us, but they compute a spectrum on a square domain, and we might
have to be mindful of it: “ Since all of our domains are invariant under the
D4 symmetry group, we can simplify the eigenfunction computations by reduc-
ing to a fundamental domain. On this domain we impose appropriate boundary
conditions according to the rep-resentation type. For the 1-dimensional repre-
sentation, we restrict to the sector 0 ≤ θ ≤ π/4 . Recall that the functions will
extend evenly when reflected about θ = 0 in the 1++ and 1– cases, and oddly
in the 1-+ and 1+- cases. Note that performing an even extension across a ray is
equivalent to imposing Neumann boundary conditions on that ray. Similarly, the
odd extension is equivalent to Dirichlet conditions. For the 2-dimensional rep-
resentation our fundamental domain is the sector 0 ≤ θ ≤ π/2 , and we impose
Neumann boundary conditions on the ray θ = 0 and Dirichlet conditions on the
ray θ = π/2. Note that our fundamental domains are simply connected. ”
This seems to be saying that one gets the 2-dimensional representation by dou-
bling the fundamental domain and mixing boundary conditions. Do you under-
stand that?

Boris Here is my present understanding of the fundamental domains issue: If you want
simple boundary conditions like Dirichlet or Neumann you stick to 1d represen-
tations only. They connect eigenfunction to itself at the fundamental domain
boundaries – otherwise you would need to connect pair of functions (would be

2021-07-29 77 birdtracks.eu/course3 week6

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

something like boundary conditions for spinor in case of 2d representations.) So

what you do is the following: take the largest abelian subgroup Z2 × Z2 (for
D4 ) and split its spectrum with respect to its fundamental domain defined as
1/4 of the square (twice the fundamental domain of the full group). Then your
see that Dirichlet-Dirichlet and Neumann-Neumann Hamiltonians still have Z2
symmetry so your split them further into the Hamiltonians of the 1/8 fundamen-
tal domain. But Dirichlet-Neumann remains 1/4th of the square.

Predrag Your argument is in the spirit of Harter’s class operators construction (see
week 5) of higher-dimensional representations by using particular chains of sub-
groups, but I am not able to visualize how that larger fundamental domain (of
the lower-order subgroup) folds back into the small fundamental domain of the
whole group. By the time the dust settles, I have the symmetry factorization of
the determinants that we need, but I do not have a gut feeling for the boundary
conditions that you do, when it comes to higher-dimensional irreps.

Commentary
Remark 6.1. Lorenz equation. The Lorenz equation (6.4) is the most celebrated early illus-
tration of “deterministic chaos” [17] (but not the first - that honor goes to Dame Cartwright [3]
in 1945. Amusingly, Denisov and Ponomarev [6] argue that Ben F. Laposky might have been
the first to observe chaotic attractors as early as 1953, which, strictly speaking falls after 1945,
even in Russia). Lorenz’s 1963 paper, which can be found in reprint collections refs. [5, 12],
is a pleasure to read, and it is still one of the best introductions to the physics motivating such
models (read more about Lorenz here). The equations, a set of ODEs in R3 , exhibit strange
attractors. W. Tucker [24–26] has proven rigorously (via interval arithmetic) that the Lorenz
attractor is strange for the original parameters (no stable orbits) and that it has a long stable
periodic orbit for slightly different parameters. In contrast to the hyperbolic strange attractors
such as the weakly perturbed cat map [4], the Lorenz attractor is structurally unstable. Frøy-
land [9] has a nice brief discussion of Lorenz flow. Frøyland and Alfsen [10] plot many periodic
and heteroclinic orbits of the Lorenz flow; some of the symmetric ones are included in ref. [9].
Guckenheimer-Williams [11] and Afraimovich-Bykov-Shilnikov [1] offer an in-depth discus-
sion of the Lorenz equation. The most detailed study of the Lorenz equation was undertaken by
Sparrow [22]. For a geophysics derivation, see Rothman course notes [20]. For a physical inter-
pretation of ρ as “Rayleigh number,” see Jackson [14] and Seydel [21]. The Lorenz truncation
to 3 modes, however, is so drastic that the model bears no relation to the geophysical hydro-
dynamics problem that motivated it. Just for fun, as Lorentz was such a lovable weatherman,
in 1972 Willem Malkus constructed [18], by a feat of reverse engineering, a physical system,
a “water wheel”, popularized by Strogatz [23], that is described by Lorentz equations. You
can see it simulated on wolfram.com, and tested experimentally at http://www.ace.gatech.edu.
There is no deep physics in this lovely game, it is but a cute distraction. For detailed pictures of
Lorenz invariant manifolds consult Vol II of Jackson [14] and “Realtime visualization of invari-
ant manifolds” by Ronzan. The Lorenz attractor is a very thin fractal – as we shall see, stable
manifold thickness is of the order 10−4 – whose fractal structure has been accurately resolved
by D. Viswanath [27, 28]. If you wonder what analytic function theory has to say about Lorenz,

birdtracks.eu/course3 week6 78 2021-07-29

 GROUP THEORY - WEEK 6. FOR FUNDAMENTALISTS

check ref. [29]. Modular flows are your thing? E. Ghys and J. Leys have a beautiful tale for you.
Refs. [16, 19] might also be of interest.

References
[1] V. S. Afraimovich, B. B. Bykov, and L. P. Shilnikov, “On the origin and structure
of the Lorenz attractor”, Dokl. Akad. Nauk SSSR 234, In Russian, 336–339
(1977).
[2] I. V. Biktasheva, D. Barkley, V. N. Biktashev, G. V. Bordyugov, and A. J. Foulkes,
“Computation of the response functions of spiral waves in active media”, Phys.
Rev. E 79, 056702 (2009).
[3] M. L. Cartwright and J. E. Littlewood, “On non-linear differential equations of
the second order”, J. London Math. Soc. 20, 180–189 (1945).
[4] S. C. Creagh, “Quantum zeta function for perturbed cat maps”, Chaos 5, 477–
493 (1995).
[5] P. Cvitanović, Universality in Chaos, 2nd ed. (Adam Hilger, Bristol, 1989).
[6] S. Denisov and A. V. Ponomarev, “Oscillons: An encounter with dynamical
chaos in 1953?”, Chaos 21, 023123 (2011).
[7] A. J. Foulkes, Drift and Meander of Spiral Waves, PhD thesis (Univ. of Liver-
pool, Liverpool, UK, 2020).
[8] A. J. Foulkes and V. N. Biktashev, “Riding a spiral wave: numerical simulation
of spiral waves in a comoving frame of reference”, Phys. Rev. E 81, 046702
(2010).
[9] J. Frøyland, Introduction to Chaos and Coherence (Taylor & Francis, Bristol,
1992).
[10] J. Frøyland and K. H. Alfsen, “Lyapunov-exponent spectra for the Lorenz model”,
Phys. Rev. A 29, 2928 (1984).
[11] J. Guckenheimer and R. Williams, “Structural stability of the Lorenz attractor”,
Publ. Math. IHES 50, 55–72 (1979).
[12] B.-L. Hao, Chaos II (World Scientific, Singapore, 1990).
[13] S. M. Heilman and R. S. Strichartz, “Homotopies of eigenfunctions and the spec-
trum of the Laplacian on the Sierpinski carpet”, Fractals 18, 1–34 (2010).
[14] E. A. Jackson, Perspectives of Nonlinear Dynamics, Vol. 1 (Cambridge Univ.
Press, Cambridge UK, 1989).
[15] O. Katz-Saporta and E. Efrati, “Self-driven fractional rotational diffusion of the
harmonic three-mass system”, Phys. Rev. Lett. 122, 024102 (2019).
[16] J. B. Laughlin and P. C. Martin, “Transition to turbulence of a statically stressed
fluid”, Phys. Rev. Lett. 33, 1189 (1974).
[17] E. N. Lorenz, “Deterministic nonperiodic flow”, J. Atmos. Sci. 20, 130–141
(1963).

2021-07-29 79 birdtracks.eu/course3 week6

 EXERCISES

[18] W. V. R. Malkus, “Non-periodic convection at high and low Prandtl number”,

Mem. Societe Royale des Sciences de Liege, 125–128 (1972).
[19] P. Manneville and Y. Pomeau, “Different ways to turbulence in dissipative dy-
namical systems”, Physica D 1, 219–226 (1980).
[20] D. Rothman, Nonlinear Dynamics I: Chaos, 2006.
[21] R. Seydel, From Equilibrium to Chaos: Practical Bifurcation and Stability Anal-
ysis (Elsevier, New York, 1988).
[22] C. Sparrow, The Lorenz Equations: Bifurcations, Chaos and Strange Attractors
(Springer, New York, 1982).
[23] S. H. Strogatz, Nonlinear Dynamics and Chaos (Westview Press, Boulder, CO,
2014).
[24] W. Tucker, “The Lorenz attractor exists”, C. R. Acad. Sci. Paris Sér. I Math. 328,
1197–1202 (1999).
[25] W. Tucker, “A rigorous ODE solver and Smale’s 14th problem”, Found. Comp.
Math. 2, 53–117 (2002).
[26] M. Viana, “What’s new on Lorenz strange attractors?”, Math. Intelligencer 22,
6–19 (2000).
[27] D. Viswanath, “Symbolic dynamics and periodic orbits of the Lorenz attractor”,
Nonlinearity 16, 1035–1056 (2003).
[28] D. Viswanath, “The fractal property of the Lorenz attractor”, Physica D 190,
115–128 (2004).
[29] D. Viswanath and S. Şahutoğlu, “Complex singularities and the Lorenz attrac-
tor”, SIAM Rev. 52, 294–314 (2010).

Exercises
6.1. 3-disk symbolic dynamics. As periodic trajectories will turn out to be our main tool to
breach deep into the realm of chaos, it pays to start familiarizing oneself with them now
by sketching and counting the few shortest prime cycles. Show that the 3-disk pinball has
3 · 2n−1 itineraries of length n. List periodic orbits of lengths 2, 3, 4, 5, · · · . Verify that
the shortest 3-disk prime cycles are 12, 13, 23, 123, 132, 1213, 1232, 1323, 12123, · · · .
Try to sketch them. (continued in exercise 6.3)
A comment about exercise 6.1, exercise 6.2, and exercise 6.3: If parts of these problems
seem trivial - they are. The intention is just to check that you understand what these
symbolic dynamics codings are - the main message is that the really smart coding (fun-
damental domain) is 1-to-1 given by the group theory operations that map a point in the
fundamental domain to where it is in the full state space. This observation you might not
find deep, but it is - instead of absolute labels, in presence of symmetries one only needs to
keep track of relative motions, from domain to domain, does not matter which domain in
absolute coordinates - that is what group actions do. And thus the word ‘relative’ creeps
into this exposition.

birdtracks.eu/course3 week6 80 2021-07-29

EXERCISES

6.2. Reduction of 3-disk symbolic dynamics to binary. (continued from exercise 6.1)
(a) Verify that the 3-disk cycles
{1 2, 1 3, 2 3}, {1 2 3, 1 3 2}, {12 13 + 2 perms.},
{121 232 313 + 5 perms.}, {121 323+ 2 perms.}, · · · ,
correspond to the fundamental domain cycles 0, 1, 01, 001, 011, · · · respectively.
(b) Check the reduction for short cycles in table 6.1 by drawing them both in the full
3-disk system and in the fundamental domain, as in figure 6.3.
(c) Optional: Can you see how the group elements listed in table 6.1 relate irreducible
segments to the fundamental domain periodic orbits?
(continued in exercise 6.3)
6.3. 3-disk fundamental domain cycles. Try to sketch 0, 1, 01, 001, 011, · · · . in the
fundamental domain, and interpret the symbols {0, 1} by relating them to topologically
distinct types of collisions. Compare with table 6.1. Then try to sketch the location of
periodic points in the Poincaré section of the billiard flow. The point of this exercise is
that while in the configuration space longer cycles look like a hopeless jumble, in the
Poincaré section they are clearly and logically ordered. The Poincaré section is always to
be preferred to projections of a flow onto the configuration space coordinates, or any other
subset of state space coordinates which does not respect the topological organization of
the flow.
6.4. C2 -equivariance of Lorenz system. Verify that the vector field in Lorenz equations
(6.4)    
ẋ σ(y − x)
ẋ = v(x) =  ẏ  =  ρx − y − xz  (6.14)
ż xy − bz
is equivariant under the action of cyclic group C2 = {e, r} acting on R3 by a π rotation
about the z axis,
r(x, y, z) = (−x, −y, z) ,
as claimed in example 6.1.
6.5. Proto-Lorenz system. Here we quotient out the C2 symmetry by constructing an
explicit “intensity” representation of the desymmetrized Lorenz flow.

1. Rewrite the Lorenz equation (6.4) in terms of variables

(u, v, z) = (x2 − y 2 , 2xy, z) , (6.15)

show that it takes form

   
u̇ −(σ + 1)u + (σ − r)v + (1 − σ)N + vz
 v̇  =  (r − σ)u − (σ + 1)v + (r + σ)N − uz − N z 
ż v/2 − bz
p
N = u2 + v 2 . (6.16)

2. Show that this is the (Lorenz)/C2 quotient map for the Lorenz flow, i.e., that it
identifies points related by the π rotation (6.6).
3. Show that (6.15) is invertible. Where does the inverse not exist?
4. Compute the equilibria of proto-Lorenz and their stabilities. Compare with the
equilibria of the Lorenz flow.

2021-07-29 81 birdtracks.eu/course3 week6

 EXERCISES

5. Plot the strange attractor both in the original form (6.4) and in the proto-Lorenz
form (6.16)

45
40

35
30

W
25
20
15
10

5
0 200 400 600 800
V

for the Lorenz parameter values σ = 10, b = 8/3, ρ = 28. Topologically, does it
resemble more the Lorenz, or the Rössler attractor, or neither? (plot by J. Halcrow)
6. Show that a periodic orbit of the proto-Lorenz is either a periodic orbit or a relative
periodic orbit of the Lorenz flow.
7. Show that if a periodic orbit of the proto-Lorenz is also periodic orbit of the Lorenz
flow, their Floquet multipliers are the same. How do the Floquet multipliers of
relative periodic orbits of the Lorenz flow relate to the Floquet multipliers of the
proto-Lorenz?
8. Show that the coordinate change (6.15) is the same as rewriting
r
ṙ = (−σ − 1 + (σ + ρ − z) sin 2θ
2
+(1 − σ) cos 2θ)
1
θ̇ = (−σ + ρ − z + (σ − 1) sin 2θ
2
+(σ + ρ − z) cos 2θ)
r2
ż = −bz + sin 2θ . (6.17)
2
in variables
(u, v) = (r2 cos 2θ, r2 sin 2θ) ,
i.e., squaring a complex number z = x + iy, z 2 = u + iv.

birdtracks.eu/course3 week6 82 2021-07-29

 group theory - week 7

Discrete Fourier representation

Homework HW7

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 7.1 Am I a group? 2 points

Exercise 7.2 Product of two groups 2 points
Exercise 7.3 Laplacian is a non-local operator 4 points
Exercise 7.4 Lattice Laplacian diagonalized 8 points
Exercise 7.5 Work through ChaosBook Example A24.2
− Projection operators for discrete Fourier transform. 6 bonus points

Total of 16 points = 100 % score. Bonus points accumulate, can help you later if you
miss a few problems.

83
 GROUP THEORY - WEEK 7. DISCRETE FOURIER REPRESENTATION

2021-06-15 Predrag Lecture 13 Fundamentalist vision

How I think of the fundamental domain is explained in my online lectures, Week
14, in particular the snippet Regular representation of permuting tiles.
2021-06-15 Predrag Lecture 14 Diffusion confusion
You also might find my online lectures, Week 13 helpful.

Discretization of continuum, lattices, discrete derivatives, discrete Fourier representa-

tions.
The fastest way to watch any week’s lecture videos is by letting YouTube run
the course playlist (2h 50 min + 45 min for extras).

Symmetry is your friend - overview. The power of thinking. (9 min)

Applied math version: how to discretize derivatives:
ChaosBook Appendix A24 Deterministic diffusion
Sects. A24.1 to A24.1.1 Lattice Laplacian.

Lattice discretization, lattice state (7 min)

Lattice derivative (6 min)
Shift operator: the generator of discrete translations (15 min)
(extra) Discussion: Shift matrix must have the periodic b.c.; Derivative
being nonlocal is easiest to grasp on discrete lattice. It’s so easy to make
errors in the continuum formulation. (14 min)
Derivative is a linear operator (15 min)
Lattice Laplacian (5 min)
Derivative is a non-local operator (6 min)
(extra) Discussion: Lattice discretization; What if geometry is not flat in all
directions, but spherical? What about General Relativity? Life’s persistent
questions, skated around. (14 min)
(extra) Discussion: What is a derivative? Hypercubic lattice is a graph,
with nodes connected by links. Every graph has a notion of derivative
associated with it; in particular a Laplacian. I was not allowed to say
"Laplacian" here, as I have not gotten to defining it in my lecture at that
point... (2 min)
A periodic lattice as the simplest example of the theory of finite groups:
ChaosBook Sects. A24.1.2 to A24.3.1.
ChaosBook Example A24.2 Projection operators for discrete Fourier represen-
tation.
ChaosBook Example A24.3 ‘Configuration-momentum’ Fourier space duality.

birdtracks.eu/course3 week7 84 2021-07-29

EXERCISES

Have symmetry? Use it! (14 min)

(extra) Rant: Symmetrize you must. Karl Schwarzschild found his exact so-
lution in 1915, a month after the publication of Einstein’s theory of general
relativity, while serving on a World War I front. (3 min)

Have symmetry? Go to "eigen"subspace! Fourier decomposition of a 2-sites

periodic lattice. (7 min)

Periodic lattices (5 min)

Fourier eigenvalues (9 min)

Discrete Fourier representation (6 min)

Laplacian in Fourier representation (9 min)

Propagator in Fourier representation (6 min)

A meta truth; We live in The Matrix; Fourier transformation is just a matrix (10
min)

7.1 Optional reading

A theoretical physicist’s version of the above notes: Quantum Field Theory - a
cyclist tour, Chapter 1 Lattice field theory motivates discrete Fourier represen-
tations by computing a free propagator on a lattice.

(extra) Quantum Mechanics in a box: Sometimes it is simplest to impose the

periodic b.c. on a localized solution, than relax it towards the correct (infinite
extent) continuum solution. (5 min)

(extra) Rocket science needs complex numbers; Why Fourier? Digital image
processing! (8 min)

Exercises
7.1. Am I a group? Show that multiplication table

e a b c d f
e e a b c d f
a a e d b f c
b b d e f c a
c c b f e a d
d d f c a e b
f f c a d b e

2021-07-29 85 birdtracks.eu/course3 week7

 EXERCISES

describes a group. Or does it? (Hint: check whether this table satisfies the group axioms.)
7.2. Product of two groups. Let G1 and G2 be two finite groups. The elements of the
product set G = G1 × G2 are defined as pairs (g1 , g2 ), g1 ∈ G1 g2 ∈ G2 .
(a) Show that G is a group with the multiplication operation (g1 , g2 ) · (g10 , g20 ) =
(g1 g10 , g2 g20 ).
Let D1 be an irreducible representation of G1 and let D2 be an irreducible representation
of G2 . For each g = (g1 , g2 ) ∈ G define D(g) = D1 (g1 ) × D2 (g2 )
(b) Show that D = D1 × D2 is an irreducible representation of G. What are the
characters of D?
7.3. Laplacian is a non-local operator.
While the Laplacian is a simple tri-diagonal difference operator, its inverse (the “free”
propagator of statistical mechanics and quantum field theory) is a messier object. A way
to compute is to start expanding propagator as a power series in the Laplacian
∞
1 1 X 1
= 2 ∆n . (7.1)
m2 1−∆ m n=0 m2n

As ∆ is a finite matrix, the expansion is convergent for sufficiently large m2 . To get a

feeling for what is involved in evaluating such series, show that ∆2 is:
6 −4 1 1 −4
 
 −4 6 −4 1 
 
 1 −4 6 −4 1
1 

∆2 = 4  . (7.2)

a  . .
 1 −4 . 

 6 −4 
−4 1 1 −4 6
What ∆3 , ∆4 , · · · contributions look like is now clear; as we include higher and higher
powers of the Laplacian, the propagator matrix fills up; while the inverse propagator
is differential operator connecting only the nearest neighbors, the propagator is integral
operator, connecting every lattice site to any other lattice site.
This matrix can be evaluated as is, on the lattice, and sometime it is evaluated this way,
but in case at hand a wonderful simplification follows from the observation that the lattice
action is translationally invariant, exercise 7.4.
P (k)
7.4. Lattice Laplacian diagonalized. Insert the identity P = 1 wherever you
profitably can, and use the shift matrix eigenvalue equation to convert shift σ matrices
into scalars. If M commutes with σ, then (ϕ†k · M · ϕk0 ) = M̃ (k) δkk0 , and the matrix
M acts as a multiplication by the scalar M̃ (k) on the kth subspace. Show that for the
1-dimensional lattice, the projection on the kth subspace is
   
2 2π
(ϕ†k · ∆ · ϕk0 ) = 2 cos k − 1 δkk0 . (7.3)
a N
In the kth subspace the propagator is simply a number, and, in contrast to the mess gen-
erated by (7.1), there is nothing to evaluating it:
1 δkk0
ϕ†k · · ϕk0 = 2 2 , (7.4)
m2 1 − ∆ m − a2
(cos 2πk/N − 1)
where k is a site in the N -dimensional dual lattice, and a = L/N is the lattice spacing.

birdtracks.eu/course3 week7 86 2021-07-29

 group theory - week 8

Space groups

Homework HW8

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 8.1 Space group 2 points

Exercise 8.2 Band structure of a square lattice 8 points

Bonus points
Exercise 8.3 Tight binding model 8 points

Total of 10 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

87
 GROUP THEORY - WEEK 8. SPACE GROUPS

2021-06-17 Predrag Lecture 15 Space groups 2019-02-21 Claire Berger

has no time to teach this lecture. But if she did, she would: (i) Start with 2D
square lattice. (ii) Define Bravais lattice unit cell. (iii) Show that rotation sym-
metries compatible with a 2D lattice are (none), 2-, 3-, 4-, or 6-fold. (iv) Sketch
the resulting 17 wallpaper groups, sect. 8.3.1.

2021-06-17 Predrag Lecture 16 Reciprocal lattice

2019-02-21 Claire Berger has no time to teach this lecture. But if she did, she
would: (i) Start with Bragg diffraction off 2-layer square lattice to motivate the
reciprocal lattice. (ii) Show her group’s graphene diffraction measurements that
identify and distinguish the one- and the two-layer graphene. Reciprocal lattice
is not a mathematical construct - it is what experimentalists see. (iii) Construct
the reciprocal lattice and the first Brillouin zone. (iv) Show the Brillouin zone
for graphene, explain what is seen in experiments.

Lecture 10 (Unedited) Space groups. Bravais lattice. Reciprocal lattice. Brillouin

cell. Fundamental domain. (2:29:20 h)

9.1 Space groups (24:49 min; included in the above)

Gutkin lecture notes Lect. 7 Applications III. Energy Band Structure, Sect. 7.2
Lattice symmetries.

8.1 Other sources (optional)

Gutkin lecture notes Lect. 7 Applications III. Energy Band Structure, Sect. 7.2
Band structure.
If you are curious about graphene, work out Gutkin lecture notes sect. 7.3 Ap-
plications III. Energy Band Structure

Liang and Cvitanović [23] A chaotic lattice field theory in one dimension (2022).

Also good reads: Dresselhaus et al. [11] chapter 9. Space Groups in Real Space
(click here), and Cornwell [9] chapter 7. Crystallographic Space Groups (click here).
Walt De Heer learned this stuff from Herzberg [15] Molecular Spectra and Molec-
ular Structure. Condensed matter people like Kittel [21] Introduction to Solid State
Physics, but I am not a fan, because simple group theoretical facts are there presented
as condensed matter phenomena.
Quinn and Yi [25] Solid State Physics: Principles and Modern Applications intro-
duction to space groups looks compact and sensible. Band structure of graphene.
Martin Mourigal found the Presqu’île Giens, May 2009 Contribution of Symmetries
in Condensed Matter Summer School very useful. Villain [30] Symmetry and group
theory throughout physics gives a readable overview. The overheads are here, many of
them are of potential interest (Mourigal recommended).
Canals and Schober [8] Introduction to group theory. It is very concise and precise,
a bastard child of Bourbaki and Hamermesh [13]. Space groups show up only once, on

birdtracks.eu/course3 week8 88 2023-01-13

 GROUP THEORY - WEEK 8. SPACE GROUPS

p. 24: “By working with the cosets we have effectively factored out the translational
part of the problem.”
Ballou [1] An introduction to the linear representations of finite groups appears
rather formal (and very erudite).
Grenier, B. and Ballou [12] Crystallography: Symmetry groups and group repre-
sentations.
The word crystal stems from Greek ‘krustallas’ and means “solidified by the cold.”
Schober [28] Symmetry characterization of electrons and lattice excitations gives
an eminently readable discussion of space groups.
Rodríguez-Carvajal and Bourée [26] Symmetry and magnetic structures
Schweizer [29] Conjugation and co-representation analysis of magnetic structures
deals with black, white and gray groups that Martin tries not to deal with, so all Mouri-
gal groups are gray.
Villain discusses graphene in the Appendix A of Symmetry and group theory through-
out physics [30].

8.2 Thoughts (optional)

This week’s notes are long, because I’m fascinated why –of all fields of physics where
problems are formulated on lattices– only condensed matter utilizes the theory of irreps
of space groups. For the course itself, read sect. 8.3 Space groups and sect. 8.3.1 Wall-
paper groups - the rest is speculations, mostly.
Why do I care? In this course we are learning theory of space groups as applied to
quantum mechanics of crystals - rather than diagonalizing the Hamiltonian and com-
puting energy levels, one works on the reciprocal lattice, and computes energy bands
(continuum limit of finely spaced discrete eigenvalues of finite, periodic lattices). If
fluctuations from strict periodicity are small, one can often identify the crystal by mea-
suring the intensities of Bragg peaks.
Then there are other kinds of lattices. In computational field theory (classical
and quantum) one discretizes the space-time, often on a cubic lattice; one example is
worked out here in sect. 8.4 Elastodynamic equilibria of 2D solids. The there are Ising
models in one, two, three dimensions, problems like deterministic diffusion on periodic
lattices of scatterers, coupled maps lattices. None of that literature ever (to best of my
knowledge) reduces the computations to the reciprocal space Brilluion zone. Why?
The funny thing is - I know the answer since 1976, but the siren song of classi-
cal crystallography is so enchanting that it has blinded me with science. I think that
is due to a deep and under-appreciated “chaos / turbulence” physics underlying these
problems. If deviations from the strict periodic structure are small (the basic “long
wavelength” assumption of sect. 8.4), the “integrable” thinking in terms of normal
modes applies, and you should use the crystallography described here. If the symmetry
of the law you are studying is a space group, but the deviations of typical solutions
are large (our deterministic diffusion, Ising models, ...), we have to think again. One
fundamental thing we learned in studies of transitions to chaos is that the traditional
Fourier analysis is useless - it just yields broad, shapeless continuous spectra. The

2023-01-13 89 birdtracks.eu/course3 week8

 GROUP THEORY - WEEK 8. SPACE GROUPS

powerful way to think about these problems is Poincaré’s qualitative theory of solu-
tions of differential equations : analyse the geometry of their flows in their state space.
I know for a fact (from a study of cat maps and spatiotemporal cat maps - see links to
talks in ChaosBook.org/overheads/spatiotemporal; the papers are slowly being written
up) that in that case the translational eigenfunctions are hyperbolic sinhes and coshes,
rather than the sines and cosines we are used to as Cn eigenfunctions. For finite dis-
crete symmetries you saw that irreps were fine for linear problems, like coupled arrays
of springs, but symmetry reduction for a nonlinear problem like Lorenz equations re-
quired quite different techniques. For space group symmetries the analogous nonlinear
problems seem still quite unexplored.

8.3 Space groups

Kepler noted that there are only three regular tessellations (tilings) of the plane, by
triangles, by squares, and by hexagons. In 1900 David Hilbert posed his 23 problems,
including “Is there in n-dimensional Euclidean space only a finite number of essentially
different kinds of groups of motions with a fundamental region?” In 1910 Bieberbach
solved this problem and proved that in dimension n there were only finitely many
Bieberbach groups, extensions of the translation group, which is isomorphic to Zn , by
a finite subgroup of GL(n, Z). In 1948 Zassenhaus gave an algorithm to determine a
complete set of representatives of the types of n-dimensional space groups. In the mid
1970’s computers helped to determine that there are 4783 four-dimensional groups.
For the above history and references, see David E. Joyce.
A space group, a subgroup of the group of rotations and translations in three di-
mensions, is the set of transformations that leave a crystal invariant. A space group
operator is commonly denoted as
{R|t} , (8.1)
where t belongs to the infinite set of discrete translations, and R is one of the finite
number of discrete orientation (point group) symmetries. Translation symmetry, i.e.,
the periodicity of a crystal, manifests itself physically through phonons, magnons, and
other smooth, long-wavelength deformations. Discrete orientation symmetry manifests
itself through macroscopic anisotropies of crystals, and its natural faces. The experi-
mental challenge is to determine the crystal structure, typically by diffraction (study of
the reciprocal lattice). It is a challenge, as one measures only the intensities of Bragg
peaks, not their phases, but the answer should be one of the 230 space groups listed in
the International Tables for Crystallography, the “Bible” of crystallographers.
Unless you have run into a quasicrystal :). In that case Claire has a story to tell,
but it will have to remain private.
Understanding the Bible requires much more detail than what we can cover in a
week or two (it could take a lifetime), and has been written up many places. I found
Dresselhaus et al. [11] Chapter 9. Space Groups in Real Space quite clear on matrix
representation of space groups (click here). (The MIT course 6.734 online version
contains much of the same material.) I also found Béatrice Grenier’s overview over
crystallography helpful. Many online tools are available to ease the task, for example
the FullProf suite of crystallographic programs. The Bible was completed in 19th

birdtracks.eu/course3 week8 90 2023-01-13

 GROUP THEORY - WEEK 8. SPACE GROUPS

century, but the field is undergoing a revival, as the study of topological insulators
requires diving deeper into crystallography than simply looking up the tables.
The translation group T , the set of translations t that put the crystallographic struc-
ture in coincidence with itself, constitutes the lattice. T is a normal subgroup of G. It
defines the Bravais lattice. Translations are of the form

t = tn = n1 a1 + n2 a2 + n3 a3 , nj ∈ Z .

The basis vectors aj span the unit cell. There are 6 simple (or primitive) unit cells
that contain a single point, specified by the lengths of the unit translations a, b, c and
pairwise angles α, β, γ between them. The most symmetric among them is the cubic
cell, with a = b = c and α = β = γ = 90o .
The lattice unit cell (primitive cell) is always a generating region (a tile that tiles the
entire space). The smallest generating region –the fundamental domain– is a minimal
region that generates the whole pattern through its images under all symmetries. At
each lattice point the identical group of “atoms” constitutes the motif. The lattice and
the motif completely characterize the crystal.
The cosets by translation subgroup T (the set all translations) form the factor (AKA
quotient) group G/T , isomorphic to the point group g (rotations). All irreducible rep-
resentations of a space group G can be constructed from irreducible representations of
g and T . This step, however, is tricky, as, due to the non-commutativity of translations
and rotations, the quotient group G/T is not a normal subgroup of the space group G.
The quantum-mechanical calculations are executed by approximating the infinite
crystal by a triply-periodic one, and going go to the reciprocal space by deploying CNj
discrete Fourier transforms. This implements the G/T quotienting by translations and
reduces the calculation to a finite Brilluoin zone. That is the content of the ‘Bloch
theorem’ of condensed matter physics. Further work is then required to reduce the
calculations to the point group irreps.
Point symmetry operations leave at least one point fixed. They are (a) inversion
through a point, (b) rotation around an axis, (c) roto-inversion around an axis and
through a point and (d) reflection through a mirror plane. The rotations have to be
compatible with the translation symmetry: in 3 spatial dimensions they can only be of
orders 1, 2, 3, 4, or 6. They can be proper (det = +1) or improper (det = −1).
The spectroscopists’ Schoenflies notation labels point groups as: cyclic Cn , dihe-
dral Cn0 , tetrahedral T and octahedral O rotation point groups, of order n = 1, 2, 3, 4, 6,
0
respectively. The superscript refers to either v (parallel mirror plane) or h (perpen-
dicular mirror plane). The crystallographer’s preferred classification is, however, the
international crystallographic (Hermann-Mauguin) notation.

8.3.1 Wallpaper groups

Pedagogically, it pays to start with a discussion of two-dimensional space groups. In
1924 George Pólya and Paul Niggli proved that there are exactly 17 different sym-
metry types of‘wallpaper’ pattern (says Twitter). The 17 wallpaper groups, classify
the distinct systems of symmetries that can occur in a periodic tiling of the plane. In
a wallpaper pattern, there are translational symmetries in two independent directions,

2023-01-13 91 birdtracks.eu/course3 week8

 GROUP THEORY - WEEK 8. SPACE GROUPS

(a) (b)

Figure 8.1: The shaded (or yellow) area indicates a fundamental domain, i.e., the
smallest part of the pattern whose repeats tile the entire plane. (a) For the most sym-
metric 2D square lattice, with point group p4mm, the fundamental domain is indicated
by the shaded triangle ΓΛRSX∆Γ which constitutes 1/8 of the Brillouin zone, and
contains the basic wave vectors and the high symmetry points (Fig. 10.2 of Dressel-
haus et al. [11]). (b) For the 2D square lattice with the glide and reflect point group
p4g the fundamental domain is indicated by the yellow triangle (Figure drawn by M.
von Gagern).

forming a normal subgroup. Rotations can only have order 1,2,3,4 or 6 (crystallo-
graphic restriction).
For wallpaper groups the Hermann-Mauguin notation begins with either p or c, for
a primitive cell or a face-centred cell. This is followed by a digit, n, indicating the
highest order of rotational symmetry: 1-fold (none), 2-fold, 3-fold, 4-fold, or 6-fold.
The first, resp. second of the next two symbols indicates the symmetry relative to one
translation axis of the pattern, referred to as the main, resp. second one. The symbols
are either m, g, or 1, for mirror, glide reflection, or none.
Section 9.3 Two-Dimensional Space Groups of Dresselhaus et al. [11] discusses
the most symmetric of the wallpaper groups, the tiling of a plane by squares, which in
the international crystallographic notation is denoted by #11, with point group p4mm.
We work out this space group in exercise 8.2. The largest invariant subgroup of C4v is
C4 . In that case, the space group is p4, or #10. Prefix p indicates that the unit cell is
primitive (not centered). This is a ‘simple’, or symmorphic group, which makes calcu-
lations easier. There is, however, the third, non-symmorphic two-dimensional square
space group p4g or #12 (p4gm), see Table B.10 of ref. [11]. If someone can explain
its ‘Biblical’ diagram to me, I would be grateful. The wiki explanation, reproduced
here as figure 8.1 (b), is the best one that I have found so far, but I’m still scratching
my head:) The Bravais lattice ‘unit cell’ is a square in all three cases. In the crystallo-
graphic literature the ChaosBook’s ‘fundamental domain’ makes an appearance only in
the reciprocal lattice, as the Brilloun zone depicted for p4mm in figure 8.1 (a). How-
ever, the ‘wallpaper groups’ wiki does call ‘fundamental domain’ the smallest part of
the configuration pattern that, when repeated, tiles the entire plane.
The quantum-mechanical calculations are carried out in the reciprocal space, in our
case with the full Γ point, k = 0, wave vector symmetry (see Table 10.1 of ref. [11]),

birdtracks.eu/course3 week8 92 2023-01-13

 GROUP THEORY - WEEK 8. SPACE GROUPS

and ‘Large Representations’.

Sect. 10.5 Characters for the Equivalence Representation look like those for the
point group, sort of.

8.3.2 One-dimensional line groups

One would think that the one-dimensional line groups, which describe systems exhibit-
ing translational periodicity along a line, such as carbon nanotubes, would be simpler
still. But even they are not trivial – there are 13 of them.
The normal subgroup of a line group L is its translational subgroup T , with its fac-
tor group L/T isomorphic to the isogonal point group P of discrete symmetries of its
1-dimensional unit cell x ∈ (−a/2, a/2]. In the reciprocal lattice k takes on the values
in the first Brillouin zone interval (−π/a, π/a]e. In Irreducible representations of the
symmetry groups of polymer molecules. I, Božović, Vujičić and Herbut [7] construct
all the reps of the line groups whose isogonal point groups are Cn , Cnv , Cnh , S2n , and
Dn . For some of these line groups the irreps are obtained as products of the reps of the
translational subgroup and the irreps of the isogonal point group.
According to W. De Heer, the Mintmire, Dunlap and White [24] paper Are Fullerene
tubules metallic? which took care of chiral rotations for nanotubes by a tight-binding
calculation, played a key role in physicists’ understanding of line groups.

8.3.3 Time reversal symmetry

Consequences of time-reversal symmetry on line groups are discussed by Božović [6];
In the case when the Hamiltonian is invariant under time reversal [14], the symmetry
group is enlarged: L + θL. It is interesting to learn if the degeneracy of the levels is
doubled or not.
Johnston [19] Group theory in solid state physics is one of the many reviews that
discusses Wigner’s time-reversal theorems for a many-electron system, including the
character tests for time-reversal degeneracy, the double space groups, and the time-
reversal theorems (first discussed by Herring [14] in Effect of time-reversal symmetry
on energy bands of crystals).
For a very different take on reflection symmetries on spatiotemporal lattices, of
which time reversal on a temporal lattice is the simplest example, see Liang and Cvi-
tanović [23] A chaotic lattice field theory in one dimension (2022).

8.4 Elastodynamic equilibria of 2D solids (optional)

Artificial lattices are often introduced to formulate classical field theories (described
by partial differential equations) and quantum field theories (described by path inte-
grals) as finite-dimensional problems, either for theoretical reasons (QM in a periodic
box), or in order to port them to computers. For example, lattice QCD approximates
Quantum Chromodynamics by a 4-dimensional cubic crystal. What follows is a simple
example of such formulation of a classical field theory, taken from Mehran Kardar’s
MIT course, Lect. 23.

2023-01-13 93 birdtracks.eu/course3 week8

 GROUP THEORY - WEEK 8. SPACE GROUPS

Consider a perfect two-dimensional solid at T = 0. The equilibrium configuration

of atoms forms a lattice,
r 0 (m, n) = me1 + ne2 ,
where e1 and e2 are basis vectors, a = |ej | is the lattice spacing, and {m, n} are inte-
gers. At finite temperatures, the atoms fluctuate away from their equilibrium position,
moving to
r(m, n) = r 0 (m, n) + u(m, n) ,
As the low temperature distortions do not vary substantially over nearby atoms, one can
define a coarse-grained distortion field u(x), where x = (x1 , x2 ) is treated as continu-
ous, with an implicit short distance cutoff of the lattice spacing a. Due to translational
symmetry, the elastic energy depends only on the strain matrix,

uij (x) = 1
2 (∂i uj + ∂j ui ) .

Kardar picks the triangular lattice, as its elastic energy is isotropic (i.e., invariant under
lattice rotations, see Landau and Lifshitz [22]). In terms of the Lamé coefficients λ and
µ,
Z
1
βH = d2 x (2µ uij uij + λ uii ujj )
2
Z
1
= − d2 x ui [2µ  δij + (µ + λ) ∂i ∂j ] uj . (8.2)
2
(here we have assumed either infinite or doubly periodic lattice, so no boundary terms
from integration by parts), with the equations of motion something like (FIX!)

∂t2 ui = [2µ  δij + (µ + λ) ∂i ∂j ] uj . (8.3)

(Note that Kardar keeps time continuous, but discretizes space. In numerical compu-
tations time is discretized as well.) The symmetry of a square lattice permits an addi-
tional term proportional to ∂x2 u2x + ∂y2 u2y . In general, the number of independent elastic
constants depends on the dimensionality and rotational symmetry of the lattice in ques-
tion. In two dimensions, square lattices have three independent elastic constants, and
triangular lattices are “elastically isotropic” (i.e., elastic properties are independent of
direction and thus have only two [22]).
The Goldstone modes associated with the broken (PC: why “broken”?) transla-
tional symmetry are phonons, the normal modes of vibrations. Eq. (8.3) supports two
types of lattice normal modes, transverse and longitudinal.
The order parameter describing broken translational symmetry is

ρG (x) = eiG·r (x) = eiG·u(x) ,

where G is any reciprocal lattice vector. Since, by definition, G · r0 is an integer

multiple of 2π, ρG = 1 at zero temperature. Due to the fluctuations,

hρG (x)i = heiG·u(x) i

birdtracks.eu/course3 week8 94 2023-01-13

 GROUP THEORY - WEEK 8. SPACE GROUPS

decreases at finite temperatures, and its correlations decay as hρG (x)ρ∗G (0)i . This is
the order parameter ChaosBook and Gaspard use in deriving formulas for deterministic
diffusion. Kardar computes this in Fourier space by approximating G · q with its
angular average G2 q 2 /2, ignoring the rotationally symmetry-breaking term cos q · x,
and getting only the asymptotics of the correlations right (the decay is algebraic).
The translational correlations are measured in diffraction experiments. The scat-
tering amplitude is the Fourier transform of ρG , and the scattered intensity at a wave-
vector q is proportional to the structure factor. At zero temperature, the structure factor
is a set of delta-functions (Bragg peaks) at the reciprocal lattice vectors.
The orientational order parameter that characterizes the broken rotational symmetry
of the crystal can be defined as

Ψ(x) = e6iθ(x) ,

where θ(x) is the angle between local lattice bonds and a reference axis. The factor of
6 accounts for the equivalence of the 6 possible D3 orientations of the triangular lattice.
(Kardar says the appropriate choice for a square lattice is exp(4iθ(x)) - shouldn’t the
factor be 8, the order of C4v ?) The order parameter has unit magnitude at T = 0, and
is expected to decrease due to fluctuations at finite temperature. The distortion u(x)
leads to a change in bond angle given by

θ(x) = − 12 (∂x uy − ∂y ux ) .

(This seems to be dimensionally wrong? For detailed calculations, see the above Kar-
dar lecture notes.)

8.5 Literature, reflections (optional)

Predrag The story of quantum scattering off crystals, I believe, starts with the Bouck-
aert, Smoluchowski and Wigner (1936) paper [5].
To understand the order of the full group Oh of symmetries of the cube, exer-
cise 5.1 a.ii, it is instructive to look at figure 8.2 (figs. 8.8 and 8.12 in Joshi [20]).
When a cube is a building block that tiles a 3D cubic lattice, it is referred to as
the ‘elementary’ or ‘Wigner-Seitz’ cell, and its Fourier transform is called ‘the
first Brillouin zone’ in ‘the reciprocal space’. The special points and the lines
of symmetry in the Brillouin zone are shown in figure 8.2 (a). The tetrahedron
ΓXM R, an 1/48th part of the Brillouin zone, is the fundamental domain, as the
action of the 48 elements of the point group Oh on it tiles the Brillouin zone
without any gaps or overlaps.
Predrag OK, I’ll confess. The reason why it is lovely to teach graduate level physics
is that one is allowed to learn new things while doing it. I’ll now sketch one,
perhaps wild, direction that you are completely free to ignore.
Here is the problem of space groups in the nutshell. The Euclidean invari-
ance on Newtonian space-time (including its subgroups, such as the discrete
space groups), and the Poincaré invariance of special-relativistic space-time is a

2023-01-13 95 birdtracks.eu/course3 week8

 GROUP THEORY - WEEK 8. SPACE GROUPS

(a) (b)

Figure 8.2: (a) The special points and the lines of symmetry in the first Brillouin zone of a
simple cubic lattice define its fundamental domain, the tetrahedron ΓXM R. (b) Just not to get
any ideas that this is easy: the fundamental domain for the first Brillouin zone of a bcc lattice.
(From Joshi [20].)

strange brew: the space is non-compact (homogeneity), while rotations are com-
pact (isotropy). That leads to the conceptually awkward situation of mixing a
group of additions (translations) with a group of multiplications (rotations). To
work with such group we first translate objects to the origin and then rotate them
with the respect to the origin. That’s not nice, because by translation invariance
any point is as good as any other, there is no preferred origin. There is no reason
why one should translate first, rotate second. What one needs is a formalism that
implements translations and rotations on the same footing.
If I understand Hestenes [16] right (also David Finkelstein and perhaps Holger
Beck Nielsen have told me things in this spirit) a way to accomplish that is to
replace the flat translational directions by a compact manifold where translations
and rotations are non-commuting multiplicative group operations.
A part of the Hestenes program is redoing crystallography. I have read Hestenes [17]
paper (but not the Hestenes and Holt [18] follow up). It looks very interesting,
but I will spare you from my comments here, as I do not know how to make this
formalism work for our purposes (character; explicit computations), so I should
not waste your time on that. If you do have a look at his, or at Coxeter [10]
discussion of planar tilings, please do report back to me.

Predrag Graphene is a two-dimensional sheet of carbon in which the carbon atoms

are arranged in a honeycomb lattice: each carbon atom is connected to three
neighbors. It was exfoliated by Schafhaeutl [4, 27] in 1840 (more recently, a con
man got a Nobel Prize for that), and formally defined for chemists by Boehm [3]
in 1986. In 1947 Wallace [31] calculated the electronic structure of graphene, as
a preliminary exercise to calculating electronic structure of graphite, and noted
that the velocity of the electrons was independent of their energies: they all travel
at the same speed (about 100 km per second, about 1/3000 of the speed of light):
plot of the energy of the electrons in graphene as a function of its momentum
(which is inversely proportional to its wavelength) is V shaped since the energy
of the electron is linearly proportional to its momentum (Wallace [31] Eq. 3.1).

birdtracks.eu/course3 week8 96 2023-01-13

 GROUP THEORY - WEEK 8. SPACE GROUPS

The energy of a free electron is proportional to the square of its momentum, but
not so in a crystal. As this is reminiscent of massless elementary particles like
photons and neutrino’s, it has been renamed since ‘Dirac cones’, but Dirac has
nothing whatsoever to do with that. To learn more, talk to people from the Claire
Berger and Walt De Heer’s group [2] - I have extracted above history of graphene
from De Heer’s notes (the “con man” is my own angle on what went down with
this particular Nobel prize).

References
[1] R. Ballou, “An introduction to the linear representations of finite groups”, EPJ
Web Conf. 22, 00005 (2012).
[2] C. Berger, Z. Song, T. Li, X. Li, A. Y. Ogbazghi, R. Feng, Z. Dai, A. N. Marchenkov,
E. H. Conrad, P. N. First, and W. A. De Heer, “Ultrathin epitaxial graphite: 2D
electron gas properties and a route toward graphene-based nanoelectronics”, J.
Phys. Chem. B 108, 19912–19916 (2004).
[3] H. Boehm, R. Setton, and E. Stumpp, “Nomenclature and terminology of graphite
intercalation compounds”, Carbon 24, 241–245 (1986).
[4] H. P. Boehm., A. Clauss, G. O. Fischer, and U. Hofmann, “Dünnste Kohlenstoff-
Folien”, Z. Naturf. B 17, 150–153 (1962).
[5] L. P. Bouckaert, R. Smoluchowski, and E. P. Wigner, “Theory of Brillouin zones
and symmetry properties of wave functions in crystals”, Phys. Rev. 50, 58–67
(1936).
[6] I. B. Božović, “Irreducible representations of the symmetry groups of polymer
molecules. III. Consequences of time-reversal symmetry”, J. Phys. A 14, 1825
(1981).
[7] I. B. Božović, M. Vujičić, and F. Herbut, “Irreducible representations of the
symmetry groups of polymer molecules. I”, J. Phys. A 11, 2133 (1978).
[8] B. Canals and H. Schober, “Introduction to group theory”, EPJ Web Conf. 22,
00004 (2012).
[9] J. F. Cornwell, Group Theory in Physics: An Introduction (Academic, New York,
1997).
[10] H. S. M. Coxeter, Introduction to Geometry, 2nd ed. (Wiley, New York, 1989).
[11] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[12] B. Grenier and R. Ballou, “Crystallography: Symmetry groups and group repre-
sentations”, EPJ Web Conf. 22, 00006 (2012).
[13] M. Hamermesh, Group Theory and Its Application to Physical Problems (Dover,
New York, 1962).
[14] C. Herring, “Effect of time-reversal symmetry on energy bands of crystals”,
Phys. Rev. 52, 361–365 (1937).

2023-01-13 97 birdtracks.eu/course3 week8

 GROUP THEORY - WEEK 8. SPACE GROUPS

[15] G. Herzberg, Molecular Spectra and Molecular Structure (Van Nostrand, Prin-
centon NJ, 1950).
[16] D. Hestenes, Space-time Algebra, 2nd ed. (Springer, 1966).
[17] D. Hestenes, “Point groups and space groups in geometric algebra”, in Appli-
cations of Geometric Algebra in Computer Science and Engineering, edited by
L. Dorst, C. Doran, and J. Lasenby (Birkhäuser, Boston, MA, 2002), pp. 3–34.
[18] D. Hestenes and J. W. Holt, “Crystallographic space groups in geometric alge-
bra”, J. Math. Phys. 48, 023514 (2007).
[19] D. F. Johnston, “Group theory in solid state physics”, Rep. Prog. Phys. 23, 66
(1960).
[20] A. W. Joshi, Elements of Group Theory for Physicists (New Age International,
New Delhi, India, 1997).
[21] C. Kittel, Introduction to Solid State Physics, 8th ed. (Wiley, 2004).
[22] L. D. Landau and E. M. Lifshitz, Theory of Elasticity, 3rd ed. (Pergamon Press,
Oxford, 1970).
[23] H. Liang and P. Cvitanović, “A chaotic lattice field theory in one dimension”, J.
Phys. A 55, 304002 (2022).
[24] J. W. Mintmire, B. I. Dunlap, and C. T. White, “Are Fullerene tubules metallic?”,
Phys. Rev. Lett. 68, 631–634 (1992).
[25] J. J. Quinn and K. . Yi, Solid State Physics: Principles and Modern Applications
(Springer, Berlin, 2009).
[26] J. Rodríguez-Carvajal and F. Bourée, “Symmetry and magnetic structures”, EPJ
Web Conf. 22, 00010 (2012).
[27] C. Schafhaeutl, “LXXXVI. On the combinations of carbon with silicon and iron,
and other metals, forming the different species of cast iron, steel, and malleable
iron”, Philos. Mag. Ser. 3 16, 570–590 (1840).
[28] H. Schober, “Symmetry characterization of electrons and lattice excitations”,
EPJ Web Conf. 22, 00012 (2012).
[29] J. Schweizer, “Conjugation and co-representation analysis of magnetic struc-
tures”, EPJ Web Conf. 22, 00011 (2012).
[30] J. Villain, “Symmetry and group theory throughout physics”, EPJ Web Conf. 22,
00002 (2012).
[31] P. R. Wallace, “The band theory of graphite”, Phys. Rev. 71, 622–634 (1947).

birdtracks.eu/course3 week8 98 2023-01-13

EXERCISES

Figure 8.3: Square lattice of atoms

Exercises
8.1. Space group.
(a) Show that for any space group, the translations by vectors from Bravais lattice form
a normal subgroup.
(b) Can rotations of the lattice at a fixed point constitute a normal subgroup of a space
group?
(B. Gutkin)
8.2. Band structure of a square lattice. A charged particle (without spin) moves in a po-
tential created by an infinite square lattice of atoms, see figure 8.3.
(a) What are the symmetry groups of the Bravais and reciprocal lattices?
(b) Plot the 1st Brillouin zone. What is its symmetry? What is the corresponding
fundamental domain?
Let k be quasi-momentum and En (k) the energy of the nth band.
(c) At which points of the Brillouin zone is the group G(k) (the group which leaves
vector k invariant) nontrivial? What is it?
(d) What is the symmetry of En (k) as a function of k? At which points of the Brillouin
zone is the group velocity ∇En (k) equal 0?
(e) At which points of the Brillouin zone neighboring bands (generically) stick to each
other? How many bands can stick? Explain from the group theory prospective.
(f) Assume now that the lattice is slightly squeezed along one of the axis. What will
be the new symmetry of the system and its 1st Brillouin zone? Will the sticking
between bands be lifted or persiss?
(B. Gutkin)
8.3. Tight binding model. Verify your solution of exercise 8.2 within the 2-state tight bind-
ing model. Assume that particle can hop either from corner to corner of the square lattice
with coefficient t1 or from corner to the middle of the square with coefficient t2 (and vice
versa).
(a) Show the obtained energy bands Ei (k) as both contour- and 3-dimensional plots.
(b) Compare with the results from exercise 8.2.

2023-01-13 99 birdtracks.eu/course3 week8

 group theory - week 9

Continuous groups

Homework HW9

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 9.1 Irreps of SO(2) 2 points

Exercise 9.2 Reduction of product of two SO(2) irreps 1 point
Exercise 9.3 Irreps of O(2) 2 points
Exercise 9.4 Reduction of product of two O(2) irreps 1 point

Bonus points
Exercise 9.5 A fluttering flame front 4 points
Exercise 9.6 O(2) fundamental domain for a PDE (difficult) 10 points

Total of 6 points = 100 % score.

101
 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

2022-11-15 Predrag Lecture 17 Continuous groups

2022-11-15 Predrag Lecture 18 Lie groups
The fastest way to watch any week’s lecture videos is by letting YouTube run the

lecture playlist

These lectures are about the basic ideas of how one goes from finite groups to the
continuous ones. We have worked one example out in week 2, the discrete Fourier
transform of example 2.6 Projection operators for cyclic group CN . The cyclic group
CN is generated by the powers of the rotation by 2π/N , and in the N → ∞ limit one
only needs to understand
P the algebra of T` , generators of infinitesimal transformations,
D(θ) = 1 + i ` θ` T` . Applied to functions, they turn out to be partial derivatives.

Continuous symmetries - an introduction (2 min)

They still do not get it! (6 min)

◦ Lie groups, sect. 9.3: Definition of a Lie group; Cyclic group CN → contin-
uous SO(2) plane rotations; Infinitesimal transformations; SO(2) generator of
rotations.

What is a symmetry? (8 min)

Group element; transformation generator (8 min)

What is a symmetry group? (7 min)

What is a group orbit? (3 min)

What is dynamics? (2 min)

Group SO(2) (3 min)

Unitary groups are mothers of all finite / compact symmetries.

(1 h 4 min)

• The N → ∞ limit of CN gets you to the continuous Fourier transform as a

representation of SO(2), but from then on this way of thinking about continuous
symmetries gets to be increasingly awkward. A fresh restart is afforded by ma-
trix groups, and in particular the unitary group U(n) = U(1) ⊗ SU(n), which
contains all other compact groups, finite or continuous, as subgroups.

Special orthogonal group SO(n) (9 min)

Symplectic group Sp(n) (9 min)

birdtracks.eu/course3 week9 102 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

9.1 Other sources (optional)

Do not get intimidated by this week’s lectures notes.

• What’s the payback? While for you the geometrically intuitive representation is
the set of rotation [2×2] matrices, group theory says no! They split into pairs of 1-
dimensional irreps, and the basic building blocks of our 2-dimensional rotations
on our kitchen table (forget quantum mechanics!) are the U(1) [1×1] complex
unit vector phase rotations.
Reading: C. K. Wong Group Theory notes, Chap 6 1D continuous groups,
Sects. 6.1-6.3 Irreps of SO(2). In particular, note that while geometrically
intuitive representation is the set of rotation [2×2] matrices, they split into
pairs of 1-dimensional irreps.
Reading: C. K. Wong Group Theory notes, Chap 6 1D continuous groups,
Sect. 6.6 completes discussion of Fourier analysis as continuum limit of
cyclic groups Cn , compares SO(2), O(2), discrete translations group, and
continuous translations group.
Chen, Ping and Wang [2] Group Representation Theory for Physicists, Sect
5.2 Definition of a Lie group, with examples (click here).
AWH Chapter 17 Group Theory, Sect. 17.7 Continuous groups (click here).
◦ Sect. 9.4 Character orthogonality theorem
Infinitesimal symmetries: Lie derivative (8 min)
Tell no Lie to plumbers (39 sec)
It’s a matter of no small pride for a card-carrying dirt physics theorist to claim
full and total ignorance of group theory (XX min)

9.2 Discussion (optional)

Calligraphic M denotes the state space manifold as well as any subspace, such
as a group orbit (3:38 min)
Why are continuous transformation group elements represented by exponentials?
(5:39 min)
How did we get the Lie algebra? Why is (almost) every symmetry we care about
a subgroup of an unitary group? (9 min)
How did we get the SO(2) generator? (2 min)
Orthogonal and unitary transformations (7 min)
Fourier modes are so simple, that no one calls them irreps. But add more sym-
metries, and there have to be fewer irreps. (11 min)

2022-11-15 103 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

Why did we move from orthogonal group O(n) to special orthogonal group
SO(n)? (3:32 min)
Why SU(n) rather than U(n)? (6:30 min)
Why is SU(n) dimension n2 − 1? (56 sec)
What are these “characters”? And why is there a Journal of Linear Algebra,
today? Inconclusive blah blah. (12 min)
Rant 1 - Is beauty symmetry? The first piece of art found in China is a perfect
disk carved out of jade. All of Bach is symmetries. (9 min)
Rant 2 - students find letter A beautifully symmetric, but Predrag finds zero ‘O’
the most beautiful grade. (1 min)
Rant 3 - SO(3) & SU(2) preview and a long rant - listen to it at your own risk.
Roger Penrose thoughts on quantum spacetime and quantum brain. Are laws of
physics time invariant? Waiting for dark energy to go away. Arrow of time. (17
min)
Rant 4 - SO(3) & SU(2) preview and a long rant - listen to it at your own risk.
Get this: math uses 2d complex vectors (spinors) to build our real 3d space. And
all we see - starlight, graphene, greenhouse effect, helioseismography, gravita-
tional wave detectors - it is all irreps! (12 min)
Rant 5 - Help me, I’m bullied by a mathematician. (3 min)
Rant 6 - you can always count on Prof. Z. (1/2 min)

Question 9.1. Henriette Roux, pondering exercise 4.2, writes

Q I want to make sure I understand the concept of irreducible representations.
1. If a representation (which can be thought of as a sort of basis) is reducible, all group
element matrices can be simultaneously diagonalized. I want to be able to see how this
definition of reducibility matches with the notion of block diagonalizability of an overall
representation D(g).
2. AWH p. 822-823 has a discussion of this, but I’m wondering if there’s an intuitive way to
connect these two definitions or if it’s just linear algebra.
3. We have familiarized ourselves with the concept of (conjugacy) classes. Here, we now
add in the concept of character, which is just the trace of any matrix in a given class
(and every matrix of the same class will have the same trace b/c of the properties of
classes/traces).
4. So to find the characters for a given representation, we just need to find the classes and
then take the trace of a matrix representation in each class?
5. My next and related question then concerns what character means conceptually. Does it
relate classes to other classes within a given representation, or different representations
(whether reducible or not), or both? AWH says that “the set of characters for all elements
and irreducible representations of a finite group defines an orthogonal finite-dimensional
vector space."

birdtracks.eu/course3 week9 104 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

6. How does a vector space come about from a set of traces, each of which I normally think
of as just a number, like the determinant? And finally,
7. How can we use our knowledge of classes/character to find irreducible representations,
since that seems to be an important goal in examining a group.
8. Exercise 4.2 (c) says to find the characters for this representation, which seems to imply
that character depends on representation. But I would’ve thought that character, which is
a trace of a matrix, is invariant under any similarity transform, which is how you get from
a reducible representation to an irreducible representation.
9. Do the multiplicities of irreducible representations correspond to the multiplicity of char-
acters (i.e. the number of elements in each class)? If so, why? (Or if not, why not?)
10. The same thing for classes, correct? Classes shouldn’t depend on representation b/c they
can be thought of as corresponding to a physical operation (e.g. transposition or cyclic
permutation), something which is independent of basis.
.
A Great framing for a discussion, thanks! I’ll probably reedit this post several times, every-
body’s input is very welcome. Items numbered as in above:
(2) My favorite step-by-step, pedagogical exposition are the chapters 2 Representation The-
ory and Basic Theorems and 3 Character of a Representation of Dresselhaus et al. [6].
There is too much material for our course, but if you want to understand it once for all
times, it’s worth your time.
(3) Correct.
(4) Correct. Note, however, that while every matrix representation has a trace, and thus a
character, you want to decompose this character into the sum of irrep characters, as it is
obvious after the block diagonalization has been attained.
(5) The unitary diagonalization matrix, whose entries are characters, takes character-weighted
sums of classes in order to project them onto irreps, just like what the Fourier represen-
tation does. The result (as we know from projection operators analysis), are mutually
orthogonal sub-spaces.
(6) Whenever you do not understand something about finite groups, ask yourself - how does
it work for finite lattice Fourier representation?
There the vector space comes via a unitary transformation from the configuration coor-
dinates (where each group element is represented by a full matrix) to the diagonalized,
irreducible subspaces coordinates (Fourier modes).
The unitary F matrix is full of ω ij , ie, characters of the cyclic group Cn . That’s where
the characters come from.
Now mess up C3 by adding a reflection. Dihedral group D3 , the group of rotations and re-
flections, has more symmetry constrains, it cannot have 6 irreps, as reflection invariance
mixes together the two senses of rotation. Now there are 3 classes, ie, kinds of things
the group does: nothing, flip, rotate. The unitary transformation that diagonalizes group
element matrices is now morally a smaller unitary [3 × 3] matrix from ‘classes’ in config-
uration space to ‘irreps’ in the diagonalized representation, where some sub-spaces must
have dimension higher than one.
The surprise, for me, is that the entries in the unitary diagonalization matrix can still be
written as traces of irreps, ie, characters. For me it is a calculation, a beautiful example
of mathematics leading us somewhere where our intuition falls short. If you find a good
intuitive explanation somewhere, please let us all know.

2022-11-15 105 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

(7) That’s automatic, now. Each irrep has the projection operator associated with it; we
construct it as a sub-product of factors in Hamilton-Cayley formula. Now we know we
can write it -just as we did with the Fourier representation- as sum over all class group
actions, each weighted by a the irrep’s character.
(8) Characters are elements of the unitary matrix with one index running over classes, the
other over irreps. So you expect character to differ from representation to representation;
very clear from D3 character table. As always, you already know that from the Fourier
representation example.
(9) They do not. Dresselhaus et al. [6] has the answer - enter it here once you understand it.
(10) Correct.

Question 9.2. Henriette Roux, digesting sect. 10.7.1, asks

Q Please explain when one keeps track of the order of tensorial indices?
A In a tensor, upper, lower indices are separately ordered - and that order matters. The simplest
example: if some indices form an antisymmetric pair, writing them in wrong order gives you a
wrong sign. In a matrix representation of a group action, one has to distinguish between the “in”
set of indices – the ones that get contracted with the initial tensor, and the “out” set of indices
that label the tensor after the transformation. If you understand Eq. (3.22) in birdtracks.eu, you
get it. Does that answer your question?

Question 9.3. Henriette Roux asks

Q Please explain the Mµν,δρ generators of SO(n).
A Let me know if you understand the derivation of Eqs. (4.51) and (4.52) in birdtracks.eu.
Does that answer your question?

9.3 Continuous symmetries: unitary and orthogonal

This week’s lectures are not taken from any particular book, they are about basic
ideas of how one goes from finite groups to the continuous ones that any physicist
should know. We have worked one example out earlier, in week 9 and ChaosBook
Sect. A24.4. It gets you to the continuous Fourier transform as a representation of
U(1) ' SO(2), but from then on this way of thinking about continuous symmetries
gets to be increasingly awkward. So we need a fresh restart; that is afforded by matrix
groups, and in particular the unitary group U(n) = U(1) ⊗ SU(n), which contains all
other compact groups, finite or continuous, as subgroups.
The main idea in a way comes from discrete groups: the cyclic group CN is gen-
erated by the powers of the smallest rotation by ∆θ = 2π/N , and in the N → ∞
limit one only needs to understand the commutation relations among T` , generators of
infinitesimal transformations,
X
D(∆θ) = 1 + i ∆θ` T` + O(∆θ2 ) . (9.1)
`

These thoughts are spread over chapters of my book Group Theory - Birdtracks,
Lie’s, and Exceptional Groups [5] that you can steal from my website, but the book
itself is too sophisticated for this course.

birdtracks.eu/course3 week9 106 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

9.4 Character orthogonality theorem

You might like my intuitive derivation [5] of the character orthogonality theorem for
continuous compact lie groups, birdtracks.eu sect. 8.2 Characters.
Note that the replacement of an irrep matrix representation D(µ) (g)a b by its char-
acter χ(µ) (g) (a single scalar quantity) does not mean that any of the matrix indices
structure is lost; the full D(µ) (g)a b can be recovered by differentiation, as in bird-
tracks.eu eq. (8.27).

9.5 Reps of compact groups are fully reducible

(copied from Group Theory - Birdtracks, Lie’s, and Exceptional Groups [5])

The objective of physicists’ group-theoretic calculations is a description of the spec-

troscopy of a given theory. This entails identifying the levels (irreducible multiplets),
the degeneracy of a given level (dimension of the multiplet) and the level splittings
(eigenvalues of various casimirs). The basic idea that enables us to carry this program
through is extremely simple: a hermitian matrix can be diagonalized. This fact has
many names: Schur’s lemma, Wigner-Eckart theorem, full reducibility of unitary reps,
and so on (see sect. 9.5.1). We exploit it by constructing invariant hermitian matrices
M from the primitive invariant tensors. The M ’s have collective indices and act on
tensors. Being hermitian, they can be diagonalized
 
λ1 0 0 ...
 0 λ1 0 
 
† 0 0 λ
CM C =  1 ,

 λ 2

..
 
..
. .
and their eigenvalues can be used to construct projection operators that reduce multi-
particle states into direct sums of lower-dimensional reps (see sect. 9.5.1):

.. . 
. .. ... 0
 
 ... 0 
 

 1 0 . . . 0 

Y M − λj 1  . 0 1 . 
Pi = = C† 
 .. ..  C . (9.2)

.. . . .
.
λi − λj  . . . 
j6=i  

 0 ... 1 

 0 ... 
0 ...
 
.. . .
. .
An explicit expression for the diagonalizing matrix C (Clebsch-Gordan coefficients or
clebsches, sect. 9.6) is unnecessary — it is in fact often more of an impediment than
an aid, as it obscures the combinatorial nature of group-theoretic computations (see
sect. 9.7).

2022-11-15 107 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

All that is needed in practice is knowledge of the characteristic equation for the
invariant matrix M (see sect. 9.5.1). The characteristic equation is usually a simple
consequence of the algebraic relations satisfied by the primitive invariants, and the
eigenvalues λi are easily determined. The λi ’ s determine the projection operators
Pi , which in turn contain all relevant spectroscopic information: the rep dimension is
given by tr Pi , and the casimirs, 6-j’s, crossing matrices, and recoupling coefficients)
are traces of various combinations of Pi ’s. All these numbers are combinatoric; they
can often be interpreted as the number of different colorings of a graph, the number of
singlets, and so on.

The key results are the construction of projection operators from invariant matri-
ces, the Clebsch-Gordan coefficients rep of projection operators (9.26), the invariance
conditions (10.41) and the Lie algebra relations (10.45).
The basic idea is simple: a hermitian matrix can be diagonalized. If this matrix
is an invariant matrix, it decomposes the reps of the group into direct sums of lower-
dimensional reps. Most of computations to follow implement the spectral decomposi-
tion
M = λ1 P1 + λ2 P2 + · · · + λr Pr ,
which associates with each distinct root λi of invariant matrix M a projection operator
(9.5):
Y M − λj 1
Pi = .
λi − λj
j6=i

The exposition given here in sect. 9.5.1 is taken from refs. [3, 4]. Who wrote this down
first I do not know, but I like Harter’s exposition [9–11] best.

9.5.1 Projection operators

For M, a hermitian matrix, there exists a diagonalizing unitary matrix C such that

λ1 ... 0
 
.. 0 0
.
 
 
 
 0 . . . λ1 
 

 λ2 0 ... 0 

†
CMC = 
 0 λ2 .

(9.3)
 0 .. .. .. 0 

 . . . 


 0 ... λ2 

 λ3 ... 
0 0
 
.. ..
. .

Here λi 6= λj are the r distinct roots of the minimal characteristic polynomial

r
Y
(M − λi 1) = 0 (9.4)
i=1

birdtracks.eu/course3 week9 108 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

(the characteristic equations are discussed in sect. 2.10).

In the matrix C(M − λ2 1)C † the eigenvalues corresponding to λ2 are replaced by
zeroes:

λ1 − λ2
 
 λ1 − λ2 
λ1 − λ2
 
 
 

 0 


 . .. 
,
 

 0 


 λ3 − λ2 


 λ3 − λ2 

..
.

and so on, so the product over all factors (M − λ2 1)(M − λ3 1) . . . , with exception of
the (M − λ1 1) factor, has nonzero entries only in the subspace associated with λ1 :
 
1 0 0

 0 1 0 0 


 0 0 1 

0
Y Y
†
C (M − λj 1)C = (λ1 − λj )  .
 
j6=1 j6=1
 0 
0
 

 0 

..
.

In this way, we can associate with each distinct root λi a projection operator Pi ,
Y M − λj 1
Pi = , (9.5)
λi − λj
j6=i

which acts as identity on the ith subspace, and zero elsewhere. For example, the pro-
jection operator onto the λ1 subspace is
 
1
 .. 

 . 

 1 
†
 
P1 = C  0 C .
 (9.6)

 0 

 .. 
 . 
0

The matrices Pi are orthogonal 

Pi Pj = δij Pj , (no sum on j) , (9.7)

2022-11-15 109 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

and satisfy the completeness relation

r
X
Pi = 1 . (9.8)
i=1

As tr (CPi C † ) = tr Pi , the dimension of the ith subspace is given by

di = tr Pi . (9.9)

It follows from the characteristic equation (9.4) and the form of the projection operator
(9.5) that λi is the eigenvalue of M on Pi subspace:

MPi = λi Pi , (no sum on i) . (9.10)

Hence, any matrix polynomial f (M) takes the scalar value f (λi ) on the Pi subspace

f (M)Pi = f (λi )Pi . (9.11)

This, of course, is the reason why one wants to work with irreducible reps: they reduce
matrices and “operators” to pure numbers.

9.5.2 Spectral decomposition

Suppose there exist several linearly independent invariant [d × d] hermitian matrices
M1 , M2 , . . ., and that we have used M1 to decompose the d-dimensional vector space
Ṽ = Σ ⊕ Vi . Can M2 , M3 , . . . be used to further decompose Vi ? This is a standard
problem of quantum mechanics (simultaneous observables), and the answer is that fur-
ther decomposition is possible if, and only if, the invariant matrices commute:

[M1 , M2 ] = 0 , (9.12)

or, equivalently, if projection operators Pj constructed from M2 commute with pro-

jection operators Pi constructed from M1 ,

Pi Pj = Pj Pi . (9.13)

Usually the simplest choices of independent invariant matrices do not commute.

In that case, the projection operators Pi constructed from M1 can be used to project
commuting pieces of M2 :
(i)
M2 = Pi M2 Pi , (no sum on i) .
(i)
That M2 commutes with M1 follows from the orthogonality of Pi :

(i) (i)
X
[M2 , M1 ] = λj [M2 , Pj ] = 0 . (9.14)
j

(i)
Now the characteristic equation for M2 (if nontrivial) can be used to decompose Vi
subspace.

birdtracks.eu/course3 week9 110 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

An invariant matrix M induces a decomposition only if its diagonalized form (9.3)

has more than one distinct eigenvalue; otherwise it is proportional to the unit matrix
and commutes trivially with all group elements. A rep is said to be irreducible if all
invariant matrices that can be constructed are proportional to the unit matrix.
An invariant matrix M commutes with group transformations [G, M] = 0, see
(10.30). Projection operators (9.5) constructed from M are polynomials in M, so they
also commute with all g ∈ G:
[G, Pi ] = 0 (9.15)
(remember that Pi are also invariant [d×d] matrices). Hence, a [d×d] matrix rep can
be written as a direct sum of [di ×di ] matrix reps:
X X X
G = 1G1 = Pi GPj = Pi GPi = Gi . (9.16)
i,j i i

In the diagonalized rep (9.6), the matrix G has a block diagonal form:
 
G1 0 0
CGC † =  0 G2 0  ,
X
G= C i Gi Ci . (9.17)
 
.. i
0 0 .

The rep Gi acts only on the di -dimensional subspace Vi consisting of vectors Pi q,

q ∈ Ṽ . In this way an invariant [d×d] hermitian matrix M with r distinct eigenvalues
induces a decomposition of a d-dimensional vector space Ṽ into a direct sum of di -
dimensional vector subspaces Vi :
M
Ṽ → V1 ⊕ V2 ⊕ . . . ⊕ Vr . (9.18)
The theory of class algebras [9–11] offers a more elegant and systematic way of
constructing the maximal set of commuting invariant matrices Mi than the sketch of-
fered in this section.

9.6 Clebsch-Gordan coefficients

(copied from Group Theory - Birdtracks, Lie’s, and Exceptional Groups [5])

Consider the product

 
0

 0 


 1 


 1 


 1  C

(9.19)

 0 


 0 


 0 

..
.

2022-11-15 111 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

of the two terms in the diagonal representation of a projection operator (9.6). This
matrix has nonzero entries only in the dλ rows of subspace Vλ . We collect them in a
[dλ × d] rectangular matrix (Cλ )α
σ , α = 1, 2, . . . d, σ = 1, 2, . . . dλ :

(Cλ )11 . . . (Cλ )d1 

 

Cλ =  ... ..  d . (9.20)

.  λ
(Cλ )ddλ

| {z }
d

The index α in (Cλ )ασ stands for all tensor indices associated with the d = n
p+q
-
p q
dimensional tensor space V ⊗ V̄ . In the birdtrack notation these indices are explicit:

b1
λ
(Cλ )σ , abpq...b1
= . (9.21)

...

...
...a2 a1
aq

Such rectangular arrays are called Clebsch-Gordan coefficients (hereafter referred to

as clebsches for short). They are explicit mappings V → Vλ . The conjugate mapping
Vλ → V̄ is provided by the product
 
0

 0 


 1 


 1 

†
C  1 ,

(9.22)

 0 


 0 


 0 

..
.

which defines the [d × dλ ] rectangular matrix (C λ )σα , α = 1, 2, . . . d, σ = 1, 2, . . . dλ :

(C λ )11 (C λ )d1λ 
 
... 
 .. ..  d
Cλ =  . . 

(C λ )ddλ

| {z }
dλ
b1
b2
a a ...aq σ λ σ
...

(C λ )b11...b
..

2
p
, = . (9.23)
..

aq

The two rectangular Clebsch-Gordan matrices C λ and Cλ are related by hermitian

conjugation.

birdtracks.eu/course3 week9 112 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

The tensors transform as tensor products of the defining rep. In general, tensors
transform as tensor products of various reps, with indices running over the correspond-
ing rep dimensions:

a1 = 1, 2, . . . , d1
a2 = 1, 2, . . . , d2
..
xaap+1 ...ap+q
1 a2 ...ap
where . (9.24)
ap+q = 1, 2, . . . , dp+q .

The action of the transformation g on the index ak is given by the [dk × dk ] matrix rep
Gk .
Clebsches are notoriously index overpopulated, as they require a rep label and a
tensor index for each rep in the tensor product. Diagrammatic notation alleviates this
index plague in either of two ways:

1. One can indicate a rep label on each line:

λ 11
00
aλ 00
11
μ 11
00 σ
Caaλµ aν , aσ = aμ 111
000 0011 aσ . (9.25)
ν 111
000 1100
aν 000
111
000
111
(An index, if written, is written at the end of a line; a rep label is written above
the line.)
2. One can draw the propagators (Kronecker deltas) for different reps with different
kinds of lines. For example, we shall usually draw the adjoint rep with a thin line.

By the definition of clebsches (9.6), the λ rep projection operator can be written out in
terms of Clebsch-Gordan matrices C λ Cλ :

C λ Cλ = Pλ , (no sum on i)
a a2 ...ap α a a2 ...dp dq ...d1
(C λ )b11...bq
, (Cλ )α , cpd...c
q ...d1
2 c1
= (Pλ )b11...bq
, cp ...c2 c1 (9.26)

λ
= Pλ .
...
...

...
...

A specific choice of clebsches is quite arbitrary. All relevant properties of pro-

jection operators (orthogonality, completeness, dimensionality) are independent of the
explicit form of the diagonalization transformation C. Any set of Cλ is acceptable as
long as it satisfies the orthogonality and completeness conditions. From (9.19) and
(9.22) it follows that Cλ are orthonormal:

Cλ C µ = δλµ 1 ,
a a ...ap
(Cλ )β , b11...b
2
q
(C µ )apbq...a
...b1
2 a1
,α = δβα δλµ

λ μ λ μ
= . (9.27)
...

2022-11-15 113 birdtracks.eu/course3 week9

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

Here 1 is the [dλ × dλ ] unit matrix, and Cλ ’s are multiplied as [dλ × d] rectangular
matrices.
The completeness relation (9.8)
X
C λ Cλ = 1, ([d × d] unit matrix) ,
λ
a a2 ...ap α d
X
(C λ )b11...bq
, (Cλ )α , dcpq ...c
...d1
2 c1
= δca11 δca22 . . . δbqq
λ

X λ
= (9.28)

...
...

...
λ

C λ Pµ = δλµ C λ ,
Pλ C µ = δλµ C µ , (no sum on λ, µ) , (9.29)

follows immediately from (9.7) and (9.27).

9.7 Irrelevancy of clebsches

(copied from Group Theory - Birdtracks, Lie’s, and Exceptional Groups [5])

As was emphasized in sect. 9.6, an explicit choice of clebsches is highly arbitrary; it

corresponds to a particular coordinatization of the dλ -dimensional subspace Vλ . For
computational purposes clebsches are largely irrelevant. Nothing that a physicist wants
to compute depends on an explicit coordinatization. For example, in QCD the physi-
cally interesting objects are color singlets, and all color indices are summed over: one
needs only an expression for the projection operators, not for the Cλ ’s separately.
Again, a nice example is the Lie algebra generators Ti . Explicit matrices are often
constructed (Gell-Mann λi matrices, Cartan’s canonical weights); however, in any sin-
glet they always appear summed over the adjoint rep indices. The summed combination
of clebsches is just the adjoint rep projection operator, a very simple object compared
with explicit Ti matrices (PA is typically a combination of a few Kronecker deltas),
and much simpler to use in explicit evaluations. As we shall show by many examples,
all rep dimensions, casimirs, etc.. are computable once the projection operators for the
reps involved are known. Explicit clebsches are superfluous from the computational
point of view; we use them chiefly to state general theorems without recourse to any
explicit realizations.
However, if one has to compute noninvariant quantities, such as subgroup embed-
dings, explicit clebsches might be very useful. Gell-Mann [8] invented λi matrices in
order to embed SU (2) of isospin into SU (3) of the eightfold way. Cartan’s canonical
form for generators, summarized by Dynkin labels of a rep is a very powerful tool in
the study of symmetry-breaking chains [7, 12]. The same can be achieved with de-
composition by invariant matrices (a nonvanishing expectation value for a direction

birdtracks.eu/course3 week9 114 2022-11-15

 GROUP THEORY - WEEK 9. CONTINUOUS GROUPS

in the defining space defines the little group of transformations in the remaining di-
rections), but the tensorial technology in this context is underdeveloped compared to
the canonical methods. And, as Stedman [13] rightly points out, if you need to check
your calculations against the existing literature, keeping track of phase conventions is
a necessity.

References
[1] N. B. Budanur and P. Cvitanović, “Unstable manifolds of relative periodic orbits
in the symmetry-reduced state space of the Kuramoto-Sivashinsky system”, J.
Stat. Phys. 167, 636–655 (2015).
[2] J.-Q. Chen, J. Ping, and F. Wang, Group Representation Theory for Physicists
(World Scientific, Singapore, 1989).
[3] P. Cvitanović, “Group theory for Feynman diagrams in non-Abelian gauge the-
ories”, Phys. Rev. D 14, 1536–1553 (1976).
[4] P. Cvitanović, Classical and exceptional Lie algebras as invariance algebras, Ox-
ford Univ. preprint 40/77, unpublished., 1977.
[5] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[6] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, New York, 2007).
[7] L. Frappat, P. Sorba, and A. Sciarrino, Dictionary on Lie algebras and superal-
gebras (Academic Press, 2000).
[8] M. Gell-Mann, The Eightfold Way: A Theory of Strong Interaction Symmetry
(CalTech, 1961).
[9] W. G. Harter, “Algebraic theory of ray representations of finite groups”, J. Math.
Phys. 10, 739–752 (1969).
[10] W. G. Harter, Principles of Symmetry, Dynamics, and Spectroscopy (Wiley, New
York, 1993).
[11] W. G. Harter and N. dos Santos, “Double-group theory on the half-shell and the
two-level system. I. Rotation and half-integral spin states”, Amer. J. Phys. 46,
251–263 (1978).
[12] R. Slansky, “Group Theory for Unified Model Building”, Phys. Rep. 79, 1–128
(1981).
[13] G. E. Stedman, Diagram Techniques in Group Theory (Cambridge U. Press,
Cambridge, 1990).

2022-11-15 115 birdtracks.eu/course3 week9

 EXERCISES

Exercises
9.1. Irreps of SO(2). Matrix  
0 −i
T = (9.30)
i 0
is the generator of rotations in a plane.
(a) Use the method of projection operators to show that for rotations in the kth Fourier
mode plane, the irreducible 1D subspaces orthonormal basis vectors are
1  (k) (k)

e(±k) = √ ±e1 − i e2 .
2

How does T act on e(±k) ?

(b) What is the action of the [2×2] rotation matrix
 
cos kθ − sin kθ
D(k) (θ) = , k = 1, 2, · · ·
sin kθ cos kθ

on the (±k)th subspace e(±k) ?

(c) What are the irreducible representations characters of SO(2)?
9.2. Reduction of a product of two SO(2) irreps. Determine the Clebsch-Gordan series for
SO(2). Hint: Abelian group has 1-dimensional characters. Or, you are just multiplying
terms in Fourier series.
9.3. Irreps of O(2). O(2) is a group, but not a Lie group, as in addition to continuous
transformations generated by (9.30) it has, as a group element, a parity operation
 
1 0
σ=
0 −1

which cannot be reached by continuous transformations.

(a) Is this group Abelian, i.e., does T commute with R(kθ)? Hint: evaluate first the
[T, σ] commutator and/or show that σD(k) (θ)σ −1 = D(k) (−θ) .
(b) What are the equivalence (i.e., conjugacy) classes of this group?
(c) What are irreps of O(2)? What are their dimensions?
Hint: O(2) is the n → ∞ limit of Dn , worked out in exercise 4.4 Irreducible repre-
sentations of dihedral group Dn . Parity σ maps an SO(2) eigenvector into another
eigenvector, rendering eigenvalues of any O(2) commuting operator degenerate.
Or, if you really want to do it right, apply Schur’s first lemma to improper rotations
   
0 cos kθ − sin kθ cos kθ sin kθ
R (θ) = σ=
sin kθ cos kθ sin kθ − cos kθ

to prove irreducibility for k 6= 0.

(d) What are irreducible characters of O(2)?
(e) Sketch a fundamental domain for O(2).
9.4. Reduction of a product of two O(2) irreps. N Determine the Clebsch-Gordan series for
O(2), i.e., reduce the Kronecker product D(k) D(`) .

birdtracks.eu/course3 week9 116 2022-11-15

EXERCISES

9.5. A fluttering flame front.

(a) Consider a linear partial differential equation for a real-valued field u = u(x, t)
defined on a periodic domain u(x, t) = u(x + L, t):

ut + uxx + νuxxxx = 0 , x ∈ [0, L] . (9.31)

In this equation t ≥ 0 is the time and x is the spatial coordinate. The subscripts x
and t denote partial derivatives with respect to x and t: ut = ∂u/d∂, uxxxx stands
for the 4th spatial derivative of u = u(x, t) at position x and time t. Consider the
form of equations under coordinate shifts x → x + ` and reflection x → −x. What
is the symmetry group of (9.31)?
(b) Expand u(x, t) in terms of its SO(2) irreducible components (hint: Fourier expan-
sion) and rewrite (9.31) as a set of linear ODEs for the expansion coefficients. What
are the eigenvalues of the time evolution operator? What is their degeneracy?
(c) Expand u(x, t) in terms of its O(2) irreducible components (hint: Fourier expan-
sion) and rewrite (9.31) as a set of linear ODEs. What are the eigenvalues of the
time evolution operator? What is their degeneracy?
(d) Interpret u = u(x, t) as a ‘flame front velocity’ and add a quadratic nonlinearity to
(9.31),
ut + 12 (u2 )x + uxx + νuxxxx = 0 , x ∈ [0, L] . (9.32)
This nonlinear equation is known as the Kuramoto-Sivashinsky equation, a baby
cousin of Navier-Stokes. What is the symmetry group of (9.32)?
(e) Expand u(x, t) in terms of its O(2) irreducible components (see exercise 9.3) and
rewrite (9.32) as an infinite tower of coupled nonlinear ODEs.
(f) What are the degeneracies of the spectrum of the eigenvalues of the time evolution
operator?
9.6. O(2) fundamental domain for Kuramoto-Sivashinsky equation. You have C2
discrete symmetry generated by flip σ, which tiles the space by two tiles.
• Is there a subspace invariant under this C2 ? What form does the tower of ODEs
take in this subspace?
• How would you restrict the flow (the integration of the tower of coupled ODEs) to
a fundamental domain?
This problem is indeed hard, a research level problem, at least for me and the grad students
in our group. Unlike the beautiful full-reducibility, character-orthogonality representation
theory of linear problems, in nonlinear problems symmetry reduction currently seems to
require lots of clever steps and choices of particular coordinates, and we am not at all sure
that our solution is the optimal one. Somebody looking at the problem with a fresh eye
might hit upon a solution much simpler than ours. Has happened before :)
Burak Budanur’s solution is written up in Budanur and Cvitanović [1] Unstable mani-
folds of relative periodic orbits in the symmetry-reduced state space of the Kuramoto-
Sivashinsky system sect. 3.2 O(2) symmetry reduction, eq. (17) (get it here).
9.7. Lie algebra from invariance. Derive the Lie algebra commutator and the Jacobi
identity as particular examples of the invariance condition, using both index and birdtracks
notations. The invariant tensors in question are “the laws of motion,” i.e., the generators
of infinitesimal group transformations in the defining and the adjoint representations.

2022-11-15 117 birdtracks.eu/course3 week9

 group theory - week 10

Lie groups, algebras

Homework HW10

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 10.1 Conjugacy classes of SO(3) 2 points (+ 2 bonus points, if complete)

Exercise 10.2 The character of SO(3) 3-dimensional representation 1 point
Exercise 10.3 The orthonormality of SO(3) characters 2 point
Exercise 10.4 U(1) equivariance of two-modes system for finite angles 3 points
Exercise 10.6 SO(2) or harmonic oscillator slice 2 points

Bonus points
Exercise 10.5 Integrate the two-modes system 4 point
Exercise 10.7 Invariant subspace of the two-modes system 1 point
Exercise 10.8 Slicing the two-modes system 1 point
Exercise 10.9 The symmetry reduced two-modes flow (difficult) 6 points

Total of 10 points = 100 % score.

119
 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

2021-06-24 Predrag Lecture 19 Lie groups, algebras

The fastest way to watch any week’s lecture videos is by letting YouTube run the
lecture playlist

There is way too much material in this week’s notes. Watch the main sequence of video
clips, that and recommended reading should suffice. The rest is optional.

• Bridging the step from discrete to continuous compact groups: invariant integra-
tion measures, characters, character orthonormality and completeness relations:
Rotations in 3 dimensions (30 min)
Lie algebra (21 min)

Birdtracks (6 min)
◦ Sect. 10.7 Lie groups for pedestrians is advanced material, safely ignored,
here only to whet your appetite for things not done in 19th century. It
is a very condensed extract of chapters 3 Invariants and reducibility and
4 Diagrammatic notation from Group Theory - Birdtracks, Lie’s, and Ex-
ceptional Groups [11]. I am usually reluctant to use birdtrack notations in
front of graduate students indoctrinated by their professors in the 1890’s
tensor notation, but now I’m emboldened by the very enjoyable article on
The new language of mathematics by Dan Silver [22].
◦ Ditto for sect. 10.10 Birdtracks - updated history. Your professor’s notation
is as convenient for actual calculations as -let’s say- long division using
roman numerals. So leave them wallowing in their early progressive rock
of 1968, King Crimsons of their youth. You chill to beats younger than
Windows 98, to grime, to trap, to hardvapour, to birdtracks.
◦ Go to week 16 to learn more.

• OK, I see that formally SU(2) ' SO(3), but who ordered ”spin?”
Rotations in 2 complex dimensions (42 min)
◦ Read sect. 10.3 SU(2) Pauli matrices
For overall clarity and pleasure of reading, I like Schwichtenberg [21]
(click here) discussion best. If you read anything for this week’s lectures,
read Schwichtenberg sects. 3.4 to 3.6.

2021-06-29 Predrag Lecture 20 O(2) symmetry sliced

◦ sect. 10.4 Two-modes SO(2)-equivariant flow
lecture playlist

birdtracks.eu/course3 week10 120 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

In two real dimensions dynamics is boring (4:28 min)

Symmetries of solutions (18 min)

Symmetry reduction (1:44 min)

Moving frames: with freedom comes responsibility (8:35 min)

Phase of a relative periodic orbit, choice of moving frame (9:18 min)

Comoving frames (23:09 min)

Slice (4:00 min)

How to slice a continuous symmetry (14:16 min)

optional: Low dimensional slices; 2D flat heart (1:40 min)

Slices are not sections! (17 sec)

optional: Cross-sections, orbitfolds (1:18 min)

Symmetry reduced equations of motion (6:12 min)

Sections and slices are local, good up to a border (1:18 min)

A spatial Fourier expansion (5:11 min)

First Fourier mode slice (3:00 min)

In-slice time (1:54 min)

For the two-modes SO(2)-equivariant flow long version, see

ChaosBook example Two-modes flow.

ChaosBook chapt. Slice & dice, sect. 13.1 Only dead fish go with the flow
to sect. 13.5 First Fourier mode slice.

This is difficult material, so it is OK if you do not get it this time around. None
of this will be on the final - the main point is that once you face a nonlinear
problem, nothing is easy - not even rotations on a circle.

10.1 Other sources (optional)

• You can glance through

◦ sect. 10.5 SO(3) character orthogonality

◦ sect. 10.6 Linear algebra

but I do not expect you to master this material.

2021-07-29 121 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

C. K. Wong Group Theory notes, Chap 6 1D continuous groups, works out in

full detail the representations and Haar measures for 1-dimensional Lie groups,
and explains the difference between rotations and translations.
Chen, Ping and Wang [7] Group Representation Theory for Physicists, Sect 5.3
Lie algebras and Sect 5.4 Finite transformations work out several SU(2) and
O(3) examples (click here). Sects 5.5, 5.6 and 5.7 also merit a quick read.
In his group theory notes D. Vvedensky, chapter 8, sect. 8.3 Axis–angle repre-
sentation of proper rotations in three dimensions, has a very nice discussion of
the (10.7) parametrization of the SO(3) 3-dimensional group manifold: the pa-
rameter space corresponds to the interior of a sphere of radius π, and the over the
classes of SO(3) is given by integral over spherical shells. In sect. 8.4 he derives
the Haar measure (without calling it so).
In sect. 8.5 Vvedensky says: “For SO(2), we were able to determine the charac-
ters of the irreducible representations directly, i.e., without having to determine
the basis functions of these representations. The structure of SO(3), however,
does not allow for such a simple procedure, so we must determine the basis
functions from the outset.” That I disagree with; in birdtracks.eu sect. 15.1 Reps
of SU(2) I construct the irreps and label them by their Young tableaus with no
recourse to spherical harmonics.
Reading: Chen, Ping and Wang [7] Group Representation Theory for Physicists,
Sect 5.2 Definition of a Lie group, with examples (click here).
• Dirac belt trick applet
If still anxious, maybe this helps: Mark Staley, Understanding quaternions and
the Dirac belt trick arXiv:1001.1778.
ChaosBook Sect 26.1 Compact groups
I have enjoyed reading Mathews and Walker [16] Chap. 16 Introduction to groups
(click here). Goldbart writes that the book is ”based on lectures by Richard Feyn-
man at Cornell University.” Very clever. In particular, work through the example
of fig. 16.2: it is very cute, you get explicit eigenmodes from group theory alone.
The main message is that if you think things through first, you never have to
go through using explicit form of representation matrices - thinking in terms of
invariants, like characters, will get you there much faster.
Any book, of 100s available, like Cornwell [8] Group Theory in Physics: An
introduction that covers group theory might be more to your taste.
Hamilton’s quaternions (3:18 min)
Stone and Goldbart [24] (click here) Chapter 17 Sect 17.6 Analytic functions and
topology (wherein stereographic projection is revealed to be the geometric origin
of the spinor representations of the rotation group)
This week’s lectures are related to AWH Chapter 3 Vector Analysis (click here)
and Chapter 16 Angular Momentum (click here).

birdtracks.eu/course3 week10 122 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

10.2 Discussion (optional)

Question 10.1. Henriette Roux asks
Q Why is this complex 2-dimensional vector called a ’spinor’?
A Historical, as Arfken, Weber & Harris [4] explain: ”It turns out that half-integral angular mo-
mentum states are needed to describe the intrinsic angular momentum of the electron and many
other particles. Since these particles also have magnetic moments, an intuitive interpretation is
that their charge distributions are spinning about some axis; hence the term spin. It is now un-
derstood that the spin phenomena cannot be explained consistently by describing these particles
as ordinary charge distributions undergoing rotational motion, [...] ”
Schwichtenberg [21]: ”[...] spinors have properties that usual vectors do not have. For
instance, the factor 1/2 in the exponent. This factor shows us that a spinor1 is after a rotation by
2π not the same, but gets a minus sign. This is a pretty crazy property, because all objects we
deal with in everyday life are exactly the same after a rotation by 360o = 2π.

Question 10.2. Henriette Roux asks

Q What’ relation of Pauli exclusion principle to the spinor 2π rotation amounting to overall
minus sign?
A I think of fermion/Grassmann statistics as Archimedes principle + linearity, see my Field
Theory [10] chap. 4 Fermions. Basically, usually a constraint is imposed by eliminating a
variable, for example, givenpthe constraint is x2 + y 2 + z 2 = 1, one gets rid of z by replacing
it everywhere with z → 1 − x2 − y 2 . This makes a fully symmetric theory asymmetric
and ugly. In linear setting, another option is to keep all the variables and the symmetry, but
add a new variable which by construction subtracts a degree of freedom, what I call [12] a
”negative dimension”. In quantum field theory such variable is called a ’ghost’; it needs to be
anti-commuting or Grassmann.

Question 10.3. Henriette Roux asks

Q This course is all about eigenfunctions of symmetry operators. Why are you not teaching us
Bessel functions?
A Blame Feynman: On May 2, 1985 my stay at Cornell was to end, and Vinnie of college town
Italian Kitchen made a special dinner for three of us regulars. Das Wunderkind noticed Feynman
ambling down Eddy Avenue, kidnapped him, and here we were, two wunderkinds, two humans.
Feynman was a very smart, forever driven wunderkind. He naturally bonded with our very
smart, forever driven wunderkind, who suddenly lurched out of control, and got very competive
about at what age who summed which kind of Bessel function series. Something like age twelve,
do not remember which one did the Bessels first. At that age I read Palle Alone in the World,
while my nonwunderkind friend, being from California, watched television 12 hours a day.
When Das Wunderkind taught graduate E&M, he spent hours crafting lectures about sym-
metry groups and their representations as various eigenfunctions. Students were not pleased.
So, fuggedaboutit! if you have not done your Bessels yet, they are eigenfunctions, just like
your Fourier modes, but for a spherical symmetry rather than for a translation symmetry; wiggle
like a cosine, but decay radially.
When you need them you’ll figure them out. Or sue me.

Question 10.4. Predrag asks

Q You are the best of students now. Are you ready for The Talk?
A Henriette Roux: I’m ready!

2021-07-29 123 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

10.3 SU(2) Pauli matrices

A lightning, bullet points review.
• U(n): unitary transformation U = eiH
• Unitarity: U † U = 1 ⇒ H † = H, the generator is hermitian.
• SU(n): special unitary transformation det U = 1
• Must know: ln det = tr ln for any matrix, so the generator is traceless
ln det U = tr ln U= tr H = 0
 
a c
• SU(2) : H = , a, b, c, e ∈ C , eight real numbers in all.
eb
 
a c + id
• H is hermitian: H = , a, b, c, d ∈ R ,
c − id b
• H is traceless: 0 = tr H ⇒ a + b = 0 , three real rotation parameters in all, so

H = cσx + dσy + aσz

     
0 1 0 −i 1 0
= c +d +a (10.1)
1 0 i 0 0 −1

where σj are known as Pauli matrices.

10.4 Two-modes SO(2)-equivariant flow

Consider the pair of U(1)-equivariant complex ODEs

ż1 = (µ1 − i e1 ) z1 + a1 z1 |z1 |2 + b1 z1 |z2 |2 + c1 z 1 z2

ż2 = (µ2 − i e2 ) z2 + a2 z2 |z1 |2 + b2 z2 |z2 |2 + c2 z12 , (10.2)

with z1 , z2 complex, and all parameters real valued.

This system is a generic example of a few-modes truncation of a Fourier represen-
tation of some physical flow, such as fluid dynamics convection flow, truncated in such
a way that the model exhibits the same symmetries as the full original problem, while
being drastically simpler to study. It is a merely a toy model with no physical interpre-
tation, just like the iconic Lorenz flow. We use it to illustrate the effects of continuous
symmetry on chaotic dynamics.
We refer to this toy model as the two-modes system. It belongs to the family of
simplest ODE systems that we know that (a) have a continuous U(1) ' SO(2), but no
discrete symmetry (if at least one of ej 6= 0). (b) models ‘weather’, in the same sense
that Lorenz equation models ‘weather’, (c) exhibits chaotic dynamics, (d) can be easily
visualized, in the dimensionally lowest possible setting required for chaotic dynamics,
with the full state space of dimension d = 4, and the SO(2)-reduced dynamics taking
place in 3 dimensions, and (e) for which the method of slices reduces the symmetry by
a single global slice hyperplane.

birdtracks.eu/course3 week10 124 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

ŷ2
y2

1.0 0.2
0.0
0.0 −0.2
−1.0
−1.5 −1.5 −1.5
0
0.75
x2 0
1.5 1.5
0
x1 x̂2 0.75 0 1.5 x̂ 1
(a) (b)

Figure 10.1: Two-modes flow before (a) and after (b) symmetry reduction by first
Fourier mode slice. Here a long trajectory (red and blue) starting on the unstable man-
ifold of the TW1 (red), until it falls on to the strange attractor (blue) and the shortest
relative periodic orbit 1 (magenta). Note that the relative equilibrium becomes an equi-
librium, and the relative periodic orbit becomes a periodic orbit after the symmetry
reduction.

The model has an unreasonably high number of parameters. After some experi-
mentation we fix or set to zero various parameters, and in the numerical examples that
follow, we settle for parameters set to
µ1 = −2.8 , µ2 = 1 , e1 = 0 , e2 = 1 ,
a1 = −1 , a2 = −2.66 , b1 = 0 , b2 = 0 , c1 = −7.75 , c2 = 1 , (10.3)
unless explicitly stated otherwise. For these parameter values the system exhibits
chaotic behavior. Experiment! If you find a more interesting behavior for some other
parameter values, please let us know. The simplified system of equations can now be
written as a 3-parameter {µ1 , c1 , a2 } two-modes system,
ż1 = µ1 z1 − z1 |z1 |2 + c1 z 1 z2
ż2 = (1 − i) z2 + a2 z2 |z1 |2 + z12 . (10.4)
In order to numerically integrate and visualize the flow, we recast the equations in real
variables by substitution z1 = x1 + i y1 , z2 = x2 + i y2 . The two-modes system (10.2)
is now a set of four coupled ODEs
exercise 10.5
ẋ1 = (µ1 − r2 ) x1 + c1 (x1 x2 + y1 y2 ) , r2 = x21 + y12
ẏ1 = (µ1 − r2 ) y1 + c1 (x1 y2 − x2 y1 )
ẋ2 = x2 + y2 + x21 − y12 + a2 x2 r2
ẏ2 = −x2 + y2 + 2 x1 y1 + a2 y2 r2 . (10.5)
Try integrating (10.5) with random initial conditions, for long times, times much
beyond which the initial transients have died out. What is wrong with this picture?
Figure 10.4 (a) is a mess. As we show here, the attractor is built up by a nice ‘stretch
& fold’ action, hidden from the view by the continuous symmetry induced drifts. That
exercise 10.6
is fixed by ‘quotienting’ model’s SO(2) symmetry, and reducing the dynamics to a
exercise 10.7
3-dimensional symmetry-reduced state space, figure 10.4 (b).
exercise 10.8

2021-07-29 125 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

10.5 SO(3) character orthogonality (optional)

In 3 Euclidean dimensions, a rotation around z axis is given by the SO(2) matrix
   
cos ϕ − sin ϕ 0 0 −1 0
R3 (ϕ) =  sin ϕ cos ϕ 0 = exp ϕ 1 0 0 . (10.6)
0 0 1 0 0 0

An arbitrary rotation in R3 can be represented by

Rn (ϕ) = e−iϕ n·L L = (L1 , L2 , L3 ) , (10.7)

where the unit vector n determines the plane and the direction of the rotation by angle
ϕ. Here L1 , L2 , L3 are the generators of rotations along x, y, z axes respectively,
     
0 0 0 0 0 1 0 −1 0
L1 = i 0 0 1 , L2 = i  0 0 0 , L3 = i 1 0 0 , (10.8)
0 −1 0 −1 0 0 0 0 0

with Lie algebra relations

[Li , Lj ] = iεijk Lk . (10.9)
All SO(3) rotations by the same angle θ around different rotation axis n are conjugate
to each other,
eiφn2 ·L eiθn1 ·L e−iφn2 ·L = eiθn3 ·L , (10.10)

with eiφn2 ·L and e−iθn2 ·L mapping the vector n1 to n3 and back, so that the rotation
around axis n1 by angle θ is mapped to a rotation around axis n3 by the same θ. The
conjugacy classes of SO(3) thus consist of rotations by the same angle about all distinct
rotation axes, and are thus labelled the angle θ. As the conjugacy class depends only on
exercise 10.2
θ, the characters can only be a function of θ. For the 3-dimensional special orthogonal
representation, the character is

χ = 2 cos(θ) + 1 . (10.11)

For an irrep labeled by j, the character of a conjugacy class labeled by θ is

sin(j + 1/2)θ
χ(j) (θ) = (10.12)
sin(θ/2)

To check that these characters are orthogonal to each other, one needs to define
the group integration over a parametrization of the SO(3) group manifold. A group
element is parametrized by the rotation axis n and the rotation angle θ ∈ (−π, π] ,
with n a unit vector which ranges over all points on the surface of a unit ball. Note
however, that a π rotation is the same as a −π rotation (n and −n point along the
same direction), and the n parametrization of SO(3) is thus a 2-dimensional surface of
a unit-radius ball with the opposite points identified.

birdtracks.eu/course3 week10 126 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

The Haar measure for SO(3) requires a bit of work, here we just note that after the
integration over the solid angle (characters do not depend on it), the Haar measure is

dθ dθ
dg = dµ(θ) = (1 − cos(θ)) = sin2 (θ/2) . (10.13)
2π π
With this measure the characters are orthogonal, and the character orthogonality the-
exercise 10.3
orems follow, of the same form as for the finite groups, but with the group averages
replaced by the continuous, parameter dependant group integrals
Z
1 X
→ dg .
|G| G
g∈G

The good news is that, as explained in ChaosBook.org Chap. Relativity for cyclists
(and in Group Theory - Birdtracks, Lie’s, and Exceptional Groups [11]), one never
needs to actually explicitly construct a group manifold parametrizations and the corre-
sponding Haar measure.

10.6 Linear algebra (optional)

In this section we collect a few basic definitions. A sophisticated reader might prefer
skipping straight to the definition of the Lie product (10.21), the big difference between
the group elements product used so far in discussions of finite groups, and what is
needed to describe continuous groups.

Vector space. A set V of elements x, y, z, . . . is called a vector (or linear) space

over a field F if

(a) vector addition ”+” is defined in V such that V is an abelian group under addi-
tion, with identity element 0;

(b) the set is closed with respect to scalar multiplication and vector addition

a(x + y) = ax + ay , a, b ∈ F , x, y ∈ V
(a + b)x = ax + bx
a(bx) = (ab)x
1x = x, 0x = 0. (10.14)

Here the field F is either R, the field of reals numbers, or C, the field of complex
numbers. Given a subset V0 ⊂ V , the set of all linear combinations of elements of V0 ,
or the span of V0 , is also a vector space.

A basis. {e(1) , · · · , e(d) } is any linearly independent subset of V whose span is V.

The number of basis elements d is the dimension of the vector space V.

2021-07-29 127 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

Dual space, dual basis. Under a general linear transformationPg ∈ GL(n, F), the
row of basis vectors transforms by right multiplication as e(j) = k (g−1 )j k e(k) , and
the column of xa ’s transforms by left multiplication as x0 = gx. Under left multiplica-
tion the column (row transposed) of basis vectors e(k) transforms as e(j) = (g† )j k e(k) ,
where the dual rep g† = (g−1 )> is the transpose of the inverse of g. This observation
motivates introduction of a dual representation space V̄ , the space on which GL(n, F)
acts via the dual rep g† .
Definition. If V is a vector representation space, then the dual space V̄ is the set of all
linear forms on V over the field F.
If {e(1) , · · · , e(d) } is a basis of V , then V̄ is spanned by the dual basis {e(1) , · · · , e(d) },
the set of d linear forms e(k) such that

e(j) · e(k) = δjk ,

where δjk is the Kronecker symbol, δjk = 1 if j = k, and zero otherwise. The compo-
nents of dual representation space vectors ȳ ∈ V̄ will here be distinguished by upper
indices
(y 1 , y 2 , . . . , y n ) . (10.15)
They transform under GL(n, F) as

y 0a = (g† )a b y b . (10.16)

For GL(n, F) no complex conjugation is implied by the † notation; that interpretation

applies only to unitary subgroups U (n) ⊂ GL(n, C). In the index notation, g can be
distinguished from g† by keeping track of the relative ordering of the indices,

(g)ba → g a b , (g† )ba → g b a . (10.17)

Algebra. A set of r elements tα of a vector space T forms an algebra if, in addition

to the vector addition and scalar multiplication,

(a) the set is closed with respect to multiplication T · T → T , so that for any two
elements tα , tβ ∈ T , the product tα · tβ also belongs to T :
r−1
X
tα · tβ = ταβ γ tγ , ταβ γ ∈ C ; (10.18)
γ=0

(b) the multiplication operation is distributive:

(tα + tβ ) · tγ = tα · tγ + tβ · tγ
tα · (tβ + tγ ) = tα · tβ + tα · tγ .

The set of numbers ταβ γ are called the structure constants. They form a matrix rep of
the algebra,
(tα )β γ ≡ ταβ γ , (10.19)

birdtracks.eu/course3 week10 128 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

whose dimension is the dimension r of the algebra itself.

Depending on what further assumptions one makes on the multiplication, one ob-
tains different types of algebras. For example, if the multiplication is associative

(tα · tβ ) · tγ = tα · (tβ · tγ ) ,

the algebra is associative. Typical examples of products are the matrix product

(tα · tβ )ca = (tα )ba (tβ )cb , tα ∈ V ⊗ V̄ , (10.20)

and the Lie product

(tα · tβ )ca = (tα )ba (tβ )cb − (tα )bc (tβ )ab , tα ∈ V ⊗ V̄ (10.21)

which defines a Lie algebra.

10.7 Lie groups for pedestrians (optional)

[...] which is an expression of consecration of angular momen-
tum.
— Mason A. Porter’s student

Definition: A Lie group is a topological group G such that (i) G has the structure of
a smooth differential manifold, and (ii) the composition map G × G → G : (g, h) →
gh−1 is smooth, i.e., C∞ differentiable.
Do not be mystified by this definition. Mathematicians also have to make a living.
The compact Lie groups that we will deploy here are a generalization of the theory of
SO(2) ' U(1) rotations, i.e., Fourier analysis. By a ‘smooth differential manifold’
one means objects like the circle of angles that parameterize continuous rotations in a
plane, figure 10.2, or the manifold swept by the three Euler angles that parameterize
SO(3) rotations.
By ‘compact’ one means that these parameters run over finite ranges, as opposed
to parameters in hyperbolic geometries, such as Minkowsky SO(3, 1). The groups we
focus on here are compact by default, as their representations are linear, finite-dimen-
sional matrix subgroups of the unitary matrix group U(d).

Example 1. Circle group. A circle with a direction, figure 10.2, is invariant under rota-
tion by any angle θ ∈ [0, 2π), and the group multiplication corresponds to composition
of two rotations θ1 + θ2 mod 2π. The natural representation of the group action
is by a complex numbers of absolute value 1, i.e., the exponential eiθ . The composi-
tion rule is then the complex multiplication eiθ2 eiθ1 = ei(θ1 +θ2 ) . The circle group is
a continuous group, with infinite number of elements, parametrized by the continuous
parameter θ ∈ [0, 2π). It can be thought of as the n → ∞ limit of the cyclic group Cn .
Note that the circle divided into n segments is compact, in distinction to the infinite
lattice of integers Z, whose limit is a line (noncompact, of infinite length).

2021-07-29 129 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

Figure 10.2: Circle group S 1 = SO(2), the symmetry group of a circle with directed
rotations, is a compact group, as its natural parametrization is either the angle φ ∈
[0, 2π), or the perimeter x ∈ [0, L).

An element of a [d×d] -dimensional matrix representation of a Lie group continu-

ously connected to identity can be written as
N
g(φ) = eiφ·T ,
X
φ·T = φa Ta , (10.22)
a=1

where φ · T is a Lie algebra element, Ta are matrices called ‘generators’, and φ =

(φ1 , φ2 , · · · , φN ) are the parameters of the transformation. Repeated indices are summed
throughout, and the dot product refers to a sum over Lie algebra generators. Sometimes
it is convenient to use the Dirac bra-ket notation for the Euclidean product of two real
vectors x, y ∈ Rd , or the product of two complex vectors x, y ∈ Cd , i.e., indicate
complex x-transpose times y by
d
X
hx|yi = x† y = x∗i yi . (10.23)
i

Finite unitary transformations exp(iφ · T ) are generated by sequences of infinitesimal

steps of form

g(δφ) ' 1 + iδφ · T , δφ ∈ RN , |δφ|  1 , (10.24)

where Ta , the generators of infinitesimal transformations, are a set of linearly indepen-

dent [d×d] hermitian matrices (see figure 10.3 (b)).
The reason why one can piece a global transformation from infinitesimal steps is
that the choice of the “origin” in coordinatization of the group manifold sketched in
figure 10.3 (a) is arbitrary. The coordinatization of the tangent space at one point on
the group manifold suffices to have it everywhere, by a coordinate transformation g,
i.e., the new origin y is related to the old origin x by conjugation y = g −1 xg, so all
tangent spaces belong the same class, they are geometrically equivalent.
Unitary and orthogonal groups are defined as groups that preserve ‘length’ norms,
hgx|gxi = hx|xi, and infinitesimally their generators (10.24) induce no change in the

birdtracks.eu/course3 week10 130 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

g
x’ x
Manifold traced out by action of G
for all possible group elements g

(a) (b)

Figure 10.3: (a) Lie algebra fields {t1 , · · · , tN } span the tangent space of the group
orbit Mx at state space point x, see (10.26) (figure from WikiMedia.org). (b) A global
group transformation g : x → x0 can be pieced together from a series of infinitesimal
steps along a continuous trajectory connecting the two points. The group orbit of state
space point x ∈ Rd is the N -dimensional manifold of all actions of the elements of
group G on x.

norm, hTa x|xi + hx|Ta xi = 0 , hence the Lie algebra generators Ta are hermitian for,

Ta† = Ta . (10.25)

The flow field at the state space point x induced by the action of the group is given by
the set of N tangent fields
ta (x)i = (Ta )ij xj , (10.26)
which span the d-dimensional group tangent space at state space point x, parametrized
by δφ.
For continuous groups the Lie algebra, i.e., the algebra spanned by the set of N gen-
erators Ta of infinitesimal transformations, takes the role that the |G| group elements
play in the theory of discrete groups (see figure 10.3).

10.7.1 Invariants
One constructs the irreps of finite groups by identifying matrices that commute with
all group elements, and using their eigenvalues to decompose arbitrary representation
of the group into a unique sum of irreps. The same strategy works for the compact Lie
groups, (10.30), and is indeed the key idea that distinguishes the invariance groups clas-
sification developed in Group Theory - Birdtracks, Lie’s, and Exceptional Groups [11]
from the 19th century Cartan-Killing classification of Lie algebras.

Definition. The vector q ∈ V is an invariant vector if for any transformation g ∈ G

q = Gq . (10.27)

Definition. A tensor x ∈ V p ⊗ V̄ q is an invariant tensor if for any g ∈ G

a a ...ap c c ...c
xb11...b
2
q
= Ga1 c1 Ga2 c2 . . . Gb1 d1 . . . Gbq dq xd11 ...d
2
q
p
. (10.28)

If a bilinear form m(x̄, y) = xa Ma b yb is invariant for all g ∈ G, the matrix

Ma b = Ga c Gb d Mc d (10.29)

2021-07-29 131 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

is an invariant matrix. Multiplying with Gb e and using the unitary, we find that the
invariant matrices commute with all transformations g ∈ G:

[G, M] = 0 . (10.30)

Definition. An invariance group G is the set of all linear transformations (10.28) that
preserve the primitive invariant relations (and, by extension, all invariant relations)

p1 (x, ȳ) = p1 (Gx, ȳG† )

p2 (x, y, z, . . .) = p2 (Gx, Gy, Gz . . .) , ... . (10.31)

Unitarity guarantees that all contractions of primitive invariant tensors, and hence all
composed tensors h ∈ H, are also invariant under action of G. As we assume unitary
G, it follows that the list of primitives must always include the Kronecker delta.

Example 2. If pa qa is the only invariant of G

a
p0 qa0 = pb (G† G)b c qc = pa qa , (10.32)

then G is the full unitary group U (n) (invariance group of the complex norm |x|2 =
xb xa δba ), whose elements satisfy
G† G = 1 . (10.33)
Example 3. If we wish the z-direction to be invariant in our 3-dimensional space,
q = (0, 0, 1) is an invariant vector (10.27), and the invariance group is O(2), the group
of all rotations in the x-y plane.

10.7.2 Infinitesimal transformations, Lie algebras

A unitary transformation G infinitesimally close to unity can be written as

Ga b = δab + iDab , (10.34)

where D is a hermitian matrix with small elements, |Dab |  1. The action of g ∈ G on

the conjugate space is given by

(G† )b a = Ga b = δba − iDba . (10.35)

D can be parametrized by N ≤ n2 real parameters. N , the maximal number of inde-

pendent parameters, is called the dimension of the group (also the dimension of the Lie
algebra, or the dimension of the adjoint rep).
Here we shall consider only infinitesimal transformations of form G = 1 + iD,
|Dba |  1. We do not study the entire group of invariant transformation, but only the
transformations connected to the identity. For example, we shall not consider invari-
ances under coordinate reflections.
The generators of infinitesimal transformations (10.34) are hermitian matrices and
belong to the Dba ∈ V ⊗ V̄ space. However, not any element of V ⊗ V̄ generates an

birdtracks.eu/course3 week10 132 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

allowed transformation; indeed, one of the main objectives of group theory is to define
the class of allowed transformations.
This subspace is called the adjoint space, and its special role warrants introduction
of special notation. We shall refer to this vector space by letter A, in distinction to
the defining space V . We shall denote its dimension by N , label its tensor indices by
i, j, k . . ., denote the corresponding Kronecker delta by a thin, straight line,

δij = , i, j = 1, 2, . . . , N , (10.36)

and the corresponding transformation generators by

1
(CA )i , ab = √ (Ti )ab = a, b = 1, 2, . . . , n
a
i = 1, 2, . . . , N .

Matrices Ti are called the generators of infinitesimal transformations. Here a is an

(uninteresting) overall normalization fixed by the orthogonality condition

(Ti )ab (Tj )ba = tr (Ti Tj ) = a δij

= a . (10.37)

For every invariant tensor q, the infinitesimal transformations G = 1 + iD must sat-

isfy the invariance condition (10.27). Parametrizing D as a projection of an arbitrary
hermitian matrix H ∈ V ⊗ V̄ into the adjoint space, D = PA H ∈ V ⊗ V̄ ,
1
Dba = (Ti )ab i , (10.38)
a
we obtain the invariance condition which the generators must satisfy: they annihilate
invariant tensors:
Ti q = 0 . (10.39)
To state the invariance condition for an arbitrary invariant tensor, we need to define
the action of generators on the tensor reps. By substituting G = 1 + i · T + O(2 )
and keeping only the terms linear in , we find that the generators of infinitesimal
transformations for tensor reps act by touching one index at a time:
a a ...ap dq ...d1 d
(Ti )b11...b
2
q
, cp ...c2 c1 = (Ti )ac11 δca22 . . . δcapp δbd11 . . . δbqq
d d
+δca11 (Ti )ac22 . . . δcapp δbd11 . . . δbqq + . . . + δca11 δca22 . . . (Ti )acpp δbd11 . . . δbqq
d d
− δca11 δca22 . . . δcapp (Ti )db11 . . . δbqq − . . . − δca11 δca22 . . . δcapp δbd11 . . . (Ti )bqq . (10.40)
This forest of indices vanishes in the birdtrack notation, enabling us to visualize the
formula for the generators of infinitesimal transformations for any tensor representa-
tion:

11
00 11
00
00
11 0011 00111100 11
00
00
11 1100 00
11
00
11 T 1100 = 11001100 + 00111100 − , (10.41)
00
11 1100
00
11 1100 00
11 00
11
00
11 1100 00
11 00
11

2021-07-29 133 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

with a relative minus sign between lines flowing in opposite directions. The reader will
recognize this as the Leibnitz rule.
The invariance conditions take a particularly suggestive form in the birdtrack nota-
tion. Equation (10.39) amounts to the insertion of a generator into all external legs of
the diagram corresponding to the invariant tensor q:

0= + −

+ − . (10.42)

The insertions on the lines going into the diagram carry a minus sign relative to the
insertions on the outgoing lines.
As the simplest example of computation of the generators of infinitesimal transfor-
mations acting on spaces other than the defining space, consider the adjoint rep. Where
does the ugly word “adjoint” come from in this context is not obvious, but remember
it this way: this is the one distinguished representation, which is intrinsic to the Lie
algebra, with the explicit matrix elements (Ti )jk of the adjoint rep given by the fully
antisymmetric structure constants iCijk of the algebra (i.e., its multiplication table un-
der the commutator product). It’s the continuous groups analogue of the multiplication
table, or the regular representation for the finite groups. The factor i ensures their re-
ality (in the case of hermitian generators Ti ), and we keep track of the overall signs by
always reading indices counterclockwise around a vertex:

− iCijk = (10.43)

0110
1010
=− 11001010 . (10.44)

As all other invariant tensors, the generators must satisfy the invariance conditions
(10.42):

0=− + − .
Redrawing this a little and replacing the adjoint rep generators (10.43) by the structure
constants, we find that the generators obey the Lie algebra commutation relation
i j

− = (10.45)

birdtracks.eu/course3 week10 134 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

In other words, the Lie algebra commutator

Ti Tj − Tj Ti = iCijk Tk . (10.46)

is simply a statement that Ti , the generators of invariance transformations, are them-

selves invariant tensors. Now, honestly, do you prefer the three-birdtracks equation (10.45),
or the mathematician’s page-long definition of the adjoint rep? It’s a classic example
of bad notation getting into way of understanding a relation of beautiful simplicity. The
invariance condition for structure constants Cijk is likewise

0= + + .

Rewriting this with the dot-vertex (10.43), we obtain

− = . (10.47)

This is the Lie algebra commutator for the adjoint rep generators, known as the Jacobi
relation for the structure constants

Cijm Cmkl − Cljm Cmki = Ciml Cjkm . (10.48)

Hence, the Jacobi relation is also an invariance statement, this time the statement that
the structure constants are invariant tensors.

10.8 Nobel Prize in Physics 2020 (optional)

Students –really, anybody who has learned some physics– often ask me: is space con-
tinuous or discrete?
We do not know, but this week’s SO(3) ≈ SU(2) correspondence is one of the
gateway drugs to speculations about quantum underpinnings of the observed spacetime.
It start’s with Hamilton’s quaternions - the discovery that the building blocks of our
apparent 3 Euclidian dimensions are 2-dimensional complex spin 1/2 ’spinors’, and it
leads different people to different theories of quantum spacetime - one direction is the
one taken by David Ritz Finkelstein, another one leads to Roger Penrose’s description
of Minkowski spacetime in terms of twistors.
In what follows, Erin Wells Bonning from Emory University and Predrag Cvi-
tanović from the Georgia Tech explain the 2020 Nobel Prize in physics in terms acces-
sible to all.
A half of the 2020 Nobel Prize in Physics was awarded to Roger Penrose, for
the discovery that black hole formation is a robust prediction of the general theory of
relativity. In 1957 Penrose, then a graduate student, met Georgia Tech’s late David
Ritz Finkelstein in a fateful meeting that changed both men’s lives forever after. It was
Finkelstein’s extension of the Schwarzschild metric which provided Penrose with an

2021-07-29 135 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

opening into general relativity and set him on the path to his 1965 discovery celebrated
by this year’s prize.
A half of the 2020 Nobel Prize in Physics was awarded jointly to Reinhard Genzel
and Andrea Ghez for the discovery of –in Ghez’s words- ”The Monster at the heart of
the Milky Way," a black hole whose existence had been hypothesized since the early
1970s. In order to visually observe an object that famously does not emit any light,
precise measurements of stars moving in the black hole’s gravitational field had to be
carried out. The independent work of Genzel and Ghez mapping the positions of these
stars over many years has led to the clearest evidence yet that the center of our Milky
Way galaxy contains “The Monster”, that possibly every galaxy contains a black hole,
and that the environment near it looks nothing like what was expected.

Nobel Lecture: Roger Penrose, Nobel Prize in Physics 2020 (34 min)
Nobel Prize in Physics 2020 (56 min)
Abstract
Penrose slides for Predrag’s 1/2 of the presentation
2020 Nobel Prizes in Chemistry and Physics, Explained
Roger Penrose gets Nobel Prize. How David Ritz Finkelstein and Roger Penrose
met, and exchanged their lives’ paths.
Negative dimensions (6 min)
Andrea Ghez: ”The Monster at the Heart of our Galaxy"
Veritasium: ”The Infinite Pattern That Never Repeats"

10.8.1 Quaternionic speculations

Predrag: putting this here for a further re-examination - safely ignored:)
Marek Danielewski (AGH), December 29, 2020, and L. Sapa: Foundations of the
Quaternion Quantum Mechanics Foundations of the Quaternion Quantum Mechanics,
Entropy, 2020, 22, 1424:
”We show that quaternion quantum mechanics has well-founded mathematical roots
and can be derived from the model of the elastic continuum by Cauchy, i.e., it can be
regarded as representing the physical reality of elastic continuum. Starting from the
Cauchy theory (classical balance equations for isotropic Cauchy-elastic material) and
using the Hamilton quaternion algebra, we present a rigorous derivation of the quater-
nion form of the non- and relativistic wave equations. The family of the wave equations
and the Poisson equation are a straightforward consequence of the quaternion represen-
tation of the Cauchy model of the elastic continuum. This is the most general kind of
quantum mechanics possessing the same kind of calculus of assertions as conventional
quantum mechanics. The problem of the Schrödinger equation, where imaginary ’i’
should emerge, is solved. This interpretation is an attempt to describe the ontology of
quantum mechanics, and demonstrates that, besides Bohmian mechanics, the complete
ontological interpretations of quantum theory exists.”

birdtracks.eu/course3 week10 136 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

It has a quack feel to it, but should be easy to work through...

For a different approach, straightforward, no quackery, see Pavel A. Bolokhov
Quaternionic wave function arXiv:1712.04795: “ quaternions form a natural lan-
guage for the description of quantum-mechanical wave functions with spin. We use
the quaternionic spinor formalism which is in one-to-one correspondence with the
usual spinor language. No unphysical degrees of freedom are admitted, in contrast
to the majority of literature on quaternions. We build a Dirac Lagrangian in the quater-
nionic form, derive the Dirac equation and take the nonrelativistic limit to find the
Schrödinger’s equation. We show that the quaternionic formalism is a natural choice to
start with, while in the transition to the noninteracting nonrelativistic limit, the quater-
nionic description effectively reduces to the regular complex wave function language.
We provide an easy-to-use grammar for switching between the ordinary spinor lan-
guage and the description in terms of quaternions. As an illustration of the broader
range of the formalism, we also derive the Maxwell’s equation from the quaternionic
Lagrangian of Quantum Electrodynamics. In order to derive the equations of motion,
we develop the variational calculus appropriate for this formalism. ”
Commentary:
Manfried Faber, Richard Gill Quaternions were invented by Benjamin Olinde Ro-
drigues, before Hamilton. (He is also known for Rodrigues formula for Legendre
polynomials.) In 1840 he published a result on transformation groups which ap-
plied Leonhard Euler’s four squares formula, a precursor to the quaternions of
William Rowan Hamilton, to the problem of representing rotations in space. In
1846 Arthur Cayley acknowledged Euler’s and Rodrigues’ priority describing
orthogonal transformations.
Manfried Faber MathsHistory.st-andrews: In 1840 Rodrigues published a mathemat-
ical paper which contains the second result for which he is known today, namely
his work on transformation groups where he derived the formula for the compo-
sition of successive finite rotations by an entirely geometric method. Rodrigues’
composition of rotations is basically the composition of unit quaternions.
Predrag I teach my students that SU(2) is double cover of SO(3) and do not do more
with quaternions. Octonions is another story...
Richard Gill According to Stigler’s law of eponomy, everything worth remembering
is associated with the name of someone we want to remember, who did some-
thing else.

10.9 What really happened (optional)

They do not make Norwegians as they used to. In his brief biographical sketch of So-
phus Lie, Burkman writes: ”I feel that I would be remiss in my duties if I failed to
mention avery interesting event that took place in Lie’s life. Klein (a German) and Lie
had moved to Paris in the spring of 1870 (they had earlier been working in Berlin).
However, in July 1870, the Franco-Prussian war broke out. Being a German alien in

2021-07-29 137 birdtracks.eu/course3 week10

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

France, Klein decided that it would be safer to return to Germany; Lie also decided to
go home to Norway. However (in a move that I think questions his geometric abilities),
Lie decided that to go from Paris to Norway, he would walk to Italy (and then presum-
ably take a ship to Norway). The trip did not go as Lie had planned. On the way, Lie
ran into some trouble–first some rain, and he had a habit of taking off his clothes and
putting them in his backpack when he walked in the rain (so he was walking to Italy
in the nude). Second, he ran into the French military (quite possibly while walking
in the nude) and they discovered in his sack (in addition to his hopefully dry clothing)
letters written to Klein in German containing the words ’lines’ and ’spheres’ (which the
French interpreted as meaning ’infantry’ and ’artillery’). Lie was arrested as a (insane)
German spy. However, due to intervention by Gaston Darboux, he was released four
weeks later and returned to Norway to finish his doctoral dissertation.”

10.10 Birdtracks - updated history

Predrag Cvitanović November 7, 2018

Young tableaux and (non-Hermitian) Young projection operators were introduced by

Young [26] in 1933 (Tung monograph [25] is a standard exposition). In 1937 R. Bra-
uer [5] introduced diagrammatic notation for δij in order to represent “Brauer algebra”
permutations, index contractions, and matrix multiplication diagrammatically. R. Pen-
rose’s papers were the first to cast the Young projection operators into a diagrammatic
form. In 1971 monograph [18] Penrose introduced diagrammatic notation for sym-
metrization operators, Levi-Civita tensors [20], and “strand networks” [17]. Penrose
credits Aitken [2] with introducing this notation in 1939, but inspection of Aitken’s
book reveals a few Brauer diagrams for permutations, and no (anti)symmetrizers. Pen-
rose’s [19] 1952 initial ways of drawing symmetrizers and antisymmetrizers are very
aesthetical, but the subsequent developments gave them a distinctly ostrich flavor [19].
In 1974 G. ’t Hooft introduced a double-line notation for U(n) gluon group-theory
weights [1]. In 1976 Cvitanović [9] introduced analogous notation for SU(N ), SO(n)
and Sp(n). For several specific, few-index tensor examples, diagrammatic Young pro-
jection operators were constructed by Canning [6], Mandula [15], and Stedman [23].
The 1975–2008 Cvitanović diagrammatic formulation of the theory of all semi-
simple Lie groups [11] as a way to compute group theoretic wights without any re-
course to symbols goes conceptually and profoundly beyond the Penrose notation
(indeed, Cvitanović “birdtracks” bear no resemblance to Penrose’s “fornicating os-
triches” [19]).
A chapter in Cvitanović 2008 monograph [11] sketches how birdtrack (diagram-
matic) Young projection operators for arbitrary irreducible representation of SU(N )
could be constructed (this text is augmented by a 2005 appendix by Elvang, Cvi-
tanović and Kennedy [13] which, however, contains a significant error). Keppeler and
Sjödahl [14] systematized the construction by offering a simple method to construct
Hermitian Young projection operators in the birdtrack formalism. Their iteration is
easy to understand, and the proofs of Hermiticity are simple. However, in practice, the

birdtracks.eu/course3 week10 138 2021-07-29

 GROUP THEORY - WEEK 10. LIE GROUPS, ALGEBRAS

algorithm is inefficient - the expression balloon quickly, the Young projection operators
soon become unwieldy and impractical, if not impossible to implement.
The Alcock-Zeilinger algorithm, based on the simplification rules of ref. [3], leads
to explicitly Hermitian and drastically more compact expressions for the projection op-
erators than the Keppeler-Sjödahl algorithm [14]. Alcock-Zeilinger fully supersedes
Cvitanović’s formulation, and any future full exposition of reduction of SU(N ) ten-
sor products into irreducible representations should be based on the Alcock-Zeilinger
algorithm.

References
[1] G. ’t Hooft, “A planar diagram theory for strong interactions”, Nucl. Phys. B 72,
461–473 (1974).
[2] A. C. Aitken, Determinants & Matrices (Oliver & Boyd, Edinburgh, 1939).
[3] J. Alcock-Zeilinger and H. Weigert, “Compact Hermitian Young projection op-
erators”, J. Math. Phys. 58, 051702 (2017).
[4] G. B. Arfken, H. J. Weber, and F. E. Harris, Mathematical Methods for Physi-
cists: A Comprehensive Guide, 7th ed. (Academic, New York, 2013).
[5] R. Brauer, “On algebras which are connected with the semisimple continuous
groups”, Ann. Math. 38, 857 (1937).
[6] G. P. Canning, “Diagrammatic group theory in quark models”, Phys. Rev. D 18,
395–410 (1978).
[7] J.-Q. Chen, J. Ping, and F. Wang, Group Representation Theory for Physicists
(World Scientific, Singapore, 1989).
[8] J. F. Cornwell, Group Theory in Physics: An Introduction (Academic, New York,
1997).
[9] P. Cvitanović, “Group theory for Feynman diagrams in non-Abelian gauge the-
ories”, Phys. Rev. D 14, 1536–1553 (1976).
[10] P. Cvitanović, Field Theory, Notes prepared by E. Gyldenkerne (Nordita, Copen-
hagen, 1983).
[11] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[12] P. Cvitanović and A. D. Kennedy, “Spinors in negative dimensions”, Phys. Scr.
26, 5–14 (1982).
[13] H. Elvang, P. Cvitanović, and A. Kennedy, “Diagrammatic Young projection
operators for U(n)”, J. Math. Phys. 46, 043501 (2005).
[14] S. Keppeler and M. Sjödahl, “Hermitian Young operators”, J. Math. Phys. 55,
021702 (2014).
[15] J. E. Mandula, Diagrammatic techniques in group theory, Univ. of Southampton,
Notes taken by S. N. Coulson and A. J. G. Hey, 1981.

2021-07-29 139 birdtracks.eu/course3 week10

 EXERCISES

[16] J. Mathews and R. L. Walker, Mathematical Methods of Physics (W. A. Ben-

jamin, Reading, MA, 1970).
[17] R. Penrose, “Angular momentum: An approach to combinatorical space-time”,
in Quantum Theory and Beyond, edited by T. Bastin (Cambridge Univ. Press,
Cambridge, 1971).
[18] R. Penrose, “Applications of negative dimensional tensors”, in Combinatorial
mathematics and its applications, edited by D. J. A. Welsh (Academic, New
York, 1971), pp. 221–244.
[19] R. Penrose, The Road to Reality - A Complete Guide to the Laws of the Universe
(A. A. Knopf, New York, 2005).
[20] R. Penrose and M. A. H. MacCallum, “Twistor theory: An approach to the quan-
tisation of fields and space-time”, Phys. Rep. 6, 241–315 (1973).
[21] J. Schwichtenberg, Physics from Symmetry (Springer, Berlin, 2015).
[22] D. S. Silver, “The new language of mathematics”, Amer. Sci. 105, 364 (2017).
[23] G. E. Stedman, Diagram Techniques in Group Theory (Cambridge U. Press,
Cambridge, 1990).
[24] M. Stone and P. Goldbart, Mathematics for Physics: A Guided Tour for Graduate
Students (Cambridge Univ. Press, Cambridge UK, 2009).
[25] W.-K. Tung, Group Theory in Physics (World Scientific, 1985).
[26] A. Young, “The application of substitutional analysis to invariants - III”, Phil.
Trans. Roy. Soc. Lond. A 234, 79–114 (1935).

Exercises
10.1. Conjugacy classes of SO(3): Show that all SO(3) rotations (10.7) by the same angle
θ around any rotation axis n are conjugate to each other:

eiφn2 ·L eiθn1 ·L e−iφn2 ·L = eiθn3 ·L (10.49)

Check this for infinitesimal φ, and argue that from that it follows that it is also true for
finite φ. Hint: use the Lie algebra commutators (10.9).
10.2. The character of SO(3) 3-dimensional representation: Show that for the 3-dimen-
sional special orthogonal representation (10.7), the character is

χ = 2 cos(θ) + 1 . (10.50)

Hint: evaluate the character explicitly for Rx (θ), Ry (θ) and Rz (θ), then explain what is
the intuitive meaning of ‘class’ for rotations.

birdtracks.eu/course3 week10 140 2021-07-29

EXERCISES

10.3. The orthonormality of SO(3) characters: Verify that given the Haar measure (10.13),
the characters (10.12) are orthogonal:
Z
0
hχ(j)|χ(j )i = dg χ(j) (g −1 ) χ(j ) (g) = δjj 0 .
0
(10.51)
G

10.4. U(1) equivariance of two-modes system for finite angles: Show that the vector field
in two-modes system (10.2) is equivariant under (10.22), the unitary group U(1) acting
on R4 ∼
= C2 as the k = 1 and 2 modes:

g(θ)(z1 , z2 ) = (eiθ z1 , ei2θ z2 ) , θ ∈ [0, 2π) . (10.52)

10.5. Integrate the two-modes system: Integrate (10.5) and plot a long trajectory of two-
modes in the 4d state space, (x1 , y1 , y2 ) projection, as in figure 10.4 (a). To save you time
(typing in (10.5) is tedious), we have prepared for you python code, and online graded
problem set here. If you do this exercise, please get started early, in order to make sure
that the autograder is working, and forward to us the grades that you receive from the
autograder.
10.6. SO(2) or harmonic oscillator slice: Construct a moving frame slice for action of
SO(2) on R2
(x, y) 7→ (x cos θ − y sin θ, x sin θ + y cos θ)
by, for instance, the positive y axis: x = 0, y > 0. Write out explicitly the group
transformation that brings any point back to the slice. What invariant is preserved by this
construction?
10.7. Invariant subspace of the two-modes system: Show that (0, 0, x2 , y2 ) is a flow invari-
ant subspace of the two-modes system (10.5), i.e., show that a trajectory with the initial
point within this subspace remains within it forever.
10.8. Slicing the two-modes system: Choose the simplest slice template point that fixes the
1. Fourier mode,
x̂0 = (1, 0, 0, 0) . (10.53)
(a) Show for the two-modes system (10.5), that the velocity within the slice, and the
phase velocity along the group orbit are

v̂(x̂) = v(x̂) − φ̇(x̂)t(x̂) (10.54)

φ̇(x̂) = −v2 (x̂)/x̂1 (10.55)

(b) Determine the chart border (the locus of point where the group tangent is either not
transverse to the slice or vanishes).
(c) What is its dimension?
(d) What is its relation to the invariant subspace of exercise 10.7?
(e) Can a symmetry-reduced trajectory cross the chart border?
10.9. The symmetry reduced two-modes flow: Pick an initial point x̂(0) that satisfies
the slice condition for the template choice (10.53) and integrate (10.54) & (10.55). Plot
the three dimensional slice hyperplane spanned by (x1 , x2 , y2 ) to visualize the symmetry
reduced dynamics. Does it look like figure 10.4 (b)?

2021-07-29 141 birdtracks.eu/course3 week10

 group theory - week 11

SU(2) and SO(3)

Homework HW11

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 11.1 The characters of SO(3) representations 4 points

Bonus points
Exercise 11.2 Real and pseudo-real representations of SO(3) 4 points
Exercise 11.3 Total spin of N particles 5 points

Total of 4 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

143
 GROUP THEORY - WEEK 11. SU(2) AND SO(3)

2021-07-01 Predrag Lecture 21 SU(2) and SO(3)

The fastest way to watch any week’s lecture videos is by letting YouTube run the

lecture playlist

Gutkin notes, Lect. 9 SU(2), SO(3) and their representations, Sects. 1-3.2.

14.1 Recap: Irreps of SO(2) are not what you would have expected (24:37 min)

14.2 Defining reps of SO(3) and SU(2) (5:18 min)

14.3 Cartan root lattices; irreps of SO(3) (22:50 min)

(optional) Anthony Zee [3] Group Theory in a Nutshell for Physicists: Cartan
classification of Lie algebras and Dynkin diagrams (5 lecture course)

◦ Read sect. 11.2 SU(2) ' SO(3)

For overall clarity and pleasure of reading, I like Schwichtenberg [2] (click
here) discussion best. If you read anything for this week’s lectures, read
Schwichtenberg sects. 3.4 to 3.6

2021-07-01 Predrag Lecture 22 SO(3) in QM

Andrew Scherbakov: Eigenvalues of J 2 and Jz operators (10:56 min)

Andrew Scherbakov: Raising, lowering operators J+ , J− (10:40 min)

(optional) 14.3A Who ordered J+, J-? (7:37 min)

Andrew Scherbakov: Jx , Jy , and Jz operators for spin-1 particle (3:58 min)

11.1 Discussion (optional)

(Still to be uploaded) 14.4 Discussion: How is SU(2) a double cover of SO(3)
and what are the physical consequences? For dimensions higher than four, SO(n)
and SU(n) get a divorce, and so far quaternions, octonion ideas have not panned
out; instead, particle physics has followed ideas of internal SU(2), SU(3) etc
symmetries, that currently culminate in the Standard Model. Negative dimen-
sions. (X:XX min)

birdtracks.eu/course3 week11 144 2022-07-16

 GROUP THEORY - WEEK 11. SU(2) AND SO(3)

11.1.1 Recap of the course, so far (optional)

Predrag This course is all about class (physically distinct symmetry operations) and
character (mining numbers from symmetries).
Here are some question by the dream student Henriette Roux (pseudonym) that
I have answered in part in class discussions, but still have to write up:

Henriette Roux Why is it that the Fourier transformation works? The presence of a
discrete but infinite translational symmetry in a system calls for its use of it to
diagonalize the matrix and thus make calculations easier, but exactly why is the
Fourier transform able to do this?

Henriette Roux How is this Fourier transform as we have studied in the space/point
groups section related to that which we have derived from the projection opera-
tors?

Henriette Roux As an extension of the Fourier transform, are there any equivalent
of Fourier transforms for rotations or other infinite but discrete symmetries as
well? So for example, if there is a system with a discrete but infinite rotational
symmetry, is there a “rotational” transform where the representing matrix is di-
agonalized? Are there whole classes of such transformations?

Henriette Roux You say that position and momentum are “dual” to each other, and
so is the real space and reciprocal space (I guess it’s the same thing as position
and momentum but just for argument sake). The commonality between these are
the fact that they can be Fourier transformed from one space to another. Does
this mean that unitary operations, eiHt , suggest a Fourier transform from the
“energy” or “frequency” space to “time” space as well?

Henriette Roux This seems very closely related to Noether’s theorem as well, is there
a way to explain this similarity?

Henriette Roux The special thing about Lie groups is that there exist analytic func-
tions which link g(a) and c = f (b, a) for g(c) = g(a)g(b). Does this need
for analytic functions come from the fact that to construct a group manifold, the
maps relating different “local” Euclidean spaces need to be C ∞ , or smooth? If
so, is there a reference we can refer to which explains how the Lie groups sat-
isfy all the other conditions of a manifold (establishing an open ball, building an
atlas and so on) as well? Just as an extension, how do you even study groups
which do not fall under the realm of a manifold? Don’t common functions like
differentials and integrations not apply in spaces outside a manifold?

Henriette Roux Why is that we Taylor expand the group in the first place? How is
this connected to the shift to left/right group operators?
The next few questions are about General Relativity, and how is what is covered
in this course applicable to GR:

2022-07-16 145 birdtracks.eu/course3 week11

 GROUP THEORY - WEEK 11. SU(2) AND SO(3)

Henriette Roux We keep to the first order in the expansion for g(θ) as we are con-
sidering the tangent space to the manifold. In the context of the GR, the tangent
space was defined as the space of directional derivatives at a point. In our case,
we are studying groups, which are not, in general, vectors (well I guess they can
be [1×1] vectors/matrices but that’s only specific irreps, so how do we understand
the concept of tangent space as you have define it?
- Or does it work out since Lie groups are always Abelian and thus have an
infinite number of 1D irreps?
- What happens if we keep the expansion to the 2nd order? Does the mathematics
change in any way? Is there a good reason to ignore the 2nd and higher order
expansions, not just in the physics sense (keeping to largest order of significance)
but in the mathematical way of understanding things?
Question 11.1. Henriette Roux asks
Q Why is this complex 2-dimensional vector called a ‘spinor’?
A Historical, as Arfken, Weber & Harris [1] explain: “It turns out that half-integral angular mo-
mentum states are needed to describe the intrinsic angular momentum of the electron and many
other particles. Since these particles also have magnetic moments, an intuitive interpretation is
that their charge distributions are spinning about some axis; hence the term spin. It is now un-
derstood that the spin phenomena cannot be explained consistently by describing these particles
as ordinary charge distributions undergoing rotational motion, [...] ”
Schwichtenberg [2]: “[...] spinors have properties that usual vectors do not have. For in-
stance, the factor 1/2 in the exponent. This factor shows us that a spinor1 is after a rotation by
2π not the same, but gets a minus sign. This is a pretty crazy property, because all objects we
deal with in everyday life are exactly the same after a rotation by 360o = 2π.

11.2 SU(2) and SO(3)

K. Y. Short
An element of SU(2) can be written as
Un (φ) = eiφ σ·n̂/2 (11.1)
where σj is a Pauli matrix and φ is a real number. What is the significance of the 1/2
factor in the argument of the exponential?
Consider a generic position vector x = (x, y, z) and construct a Hermitian matrix
of the form
σ·x = σx x + σy y + σz z
     
0 x 0 −iy z 0
= + +
x 0 iy 0 0 −z
 
z x − iy
= (11.2)
x + iy −z
Its determinant
 
z x − iy
det = −(x2 + y 2 + z 2 ) = −x2 (11.3)
x + iy −z

birdtracks.eu/course3 week11 146 2022-07-16

 GROUP THEORY - WEEK 11. SU(2) AND SO(3)

gives the length of a vector. Consider a SU(2) transformation (11.1) of this matrix,
U † (σ · x)U . Taking the determinant, we find the same expression as before:
det U (σ · x)U † = det U det (σ · x) det U † = det (σ · x) . (11.4)
Just as SO(3), SU(2) preserves the lengths of vectors.
To make the correspondence between SO(3) and SU(2) more explicit, consider a
SU(2) transformation on a complex two-component spinor
 
α
ψ= (11.5)
β
related to x by
1 2 i
x= (β − α2 ), y = − (α2 + β 2 ), z = αβ (11.6)
2 2
Check that a SU(2) transformation of ψ is equivalent to a SO(3) transformation on x.
From this equivalence, one sees that a SU(2) transformation has three real parameters
that correspond to the three rotation angles of SO(3). If we label the ”angles” for the
SU(2) transformation by α, β, and γ, we observe, for a ”rotation” about x̂
 
cos α/2 i sin α/2
Ux (α) = , (11.7)
i sin α/2 cos α/2
for a ”rotation” about ŷ,
 
cos β/2 sin β/2
Uy (β) = , (11.8)
− sin β/2 cos β/2
and for ”rotation” about ẑ,
eiγ/2
 
0
Uz (γ) = . (11.9)
0 e−iγ/2
Compare these three matrices to the corresponding SO(3) rotation matrices:
   
1 0 0 cos φ 0 sin φ
Rx (ζ) = 0 cos ζ sin ζ  , Ry (φ) =  0 1 0 
0 − sin ζ cos ζ − sin φ 0 cos φ
 
cos θ sin θ 0
Rz (θ) = − sin θ cos θ 0 (11.10)
0 0 1
They’re equivalent! Result: Half the rotation angle generated by SU(2) corresponds
to a rotation generated by SO(3).
What does this mean? At this point, probably best to switch to Schwichtenberg [2]
(click here) who explains clearly that SU(2) is a simply-connected group, and thus the
”mother" or covering group, or the double cover of SO(3). This means there is a two-
to-one map from SU(2) to SO(3); an SU(2) turn by 4π corresponds to an SO(3) turn
by 2π. So, the building blocks of your 3-dimensional world are not 3-dimensional real
vectors, but the 2-dimensional complex spinors! Quantum mechanics chose electrons
to be spin 1/2, and there is nothing Fox Channel can do about it.

2022-07-16 147 birdtracks.eu/course3 week11

 EXERCISES

References
[1] G. B. Arfken, H. J. Weber, and F. E. Harris, Mathematical Methods for Physi-
cists: A Comprehensive Guide, 7th ed. (Academic, New York, 2013).
[2] J. Schwichtenberg, Physics from Symmetry (Springer, Berlin, 2015).
[3] A. Zee, Group Theory in a Nutshell for Physicists (Princeton Univ. Press, Prince-
ton NJ, 2016).

Exercises
11.1. The characters of SO(3) representations: Show that for an irrep labeled by j, the
character of a conjugacy class labeled by θ

sin(j + 1/2)θ
χ(j) (θ) = (11.11)
sin(θ/2)

can be obtained by taking the trace of Rzj (θ). Verify that for j = 1 this character is the
three dimensional special orthogonal representation character (10.11).
11.2. Real and pseudo-real representations of SO(3). Recall (Gutkin notes, sect. 4.5
Representation Theory II, Sect. 5 5. Three types of representations) that there are exist
three types of representation which can be distinguished by the indicator (4.6):

Z  +1 real
dµ(g)χ(`) (g 2 ) = 0 complex . (11.12)
G 
−1 pseudo-real

Determine for which values of ` = 0, 1/2, 1, 3/2, 2 . . . the representation D` of SO(3)

is real or pseudo-real.

Hint: The characters and Haar measure (10.13) of SO(3) are given by

sin ` + 12 θ
 

dθ
χ(`) (g) = , dµ(g) = sin2 (θ/2) (11.13)
sin 21 θ


π

where θ is rotation angle for the group element g.

(B. Gutkin)
11.3. Total spin of N particles. Consider a system of four particles with spin 1/2. Assuming
that all (except spin) degrees of freedom are frozen the Hilbert space of the system is given
by V = V1/2 ⊗ V1/2 ⊗ V1/2 ⊗ V1/2 , with V1/2 being two-dimensional space for each
spin. V = ⊕Vs can be decomposed then into different Psectors Vs having the total spin s
i.e., Ŝ 2 v = s(s + 1)v, for any v ∈ Vs . Here Ŝ 2 = ( 4i=1 ŝi )2 and ŝi = (ŝxi , ŝyi , ŝzi ) is
spin operator for i-th particle.

(a) What are possible values s for the total spin of the system?

birdtracks.eu/course3 week11 148 2022-07-16

EXERCISES

(b) Determine dimension of the subspace of V0 with 0 total spin. In other words: how
many times trivial representation enters into product:

D = D1/2 ⊗ D1/2 ⊗ D1/2 ⊗ D1/2 ? (11.14)

(c) What is the answer to the above questions for N spins?

Hint: it is convenient to use (11.13) to decompose D into irreps.
(B. Gutkin)
11.4. Splitting of degeneracies in a central potential. Hamiltonian H0 has rotational
symmetry of SO(3).
(a) What are the possible energy level degeneracies of H0 ?
A weak perturbation V with a symmetry Td of full tetrahedron group is added (e.g., V is
a potential created by lattice of atoms with a symmetry of Td ).

(b) What will be the degeneracies of new Hamiltonian H0 + V ?

(c) Assuming that the total angular momentum of the system before the perturbation is
l = 2. How the degeneracies of the corresponding energy level will be split after
the perturbation is applied?

(B. Gutkin)
11.5. Quadrupole transitions.
a) Write Q1 = xy, Q2 = zy, Q3 = x2 − y 2 and Q4 = 2z 2 − x2 − y 2 as components of
spherical tensor of rank 2. Hint: use spherical harmonics Ylm (θ, ϕ).
b) The last quantity Q4 is known as quadrupole moment. What are the selection rules for
transitions induced by Q4 in a system with SO(3) symmetry? In other words, for which
m, l and k, j the transition rates:

Pm,l→k,j ∼ |hm l|Q4 |j ki|2

are non-zero?
c) By using Wigner-Eckart theorem write down the relationship between |hm l|Q4 |j ki|2
and |hm l|Q1 |j ki|2 in terms of Clebsch-Gordan coefficients.
(B. Gutkin)

2022-07-16 149 birdtracks.eu/course3 week11

 group theory - week 12

Lorentz group; spin

Homework HW12

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 12.1 Lie algebra of SO(4) and SU(2) ⊗ SU(2) 6 points

Exercise 12.2 SO(n) Clebsch-Gordan series for V ⊗V . 3 points
Exercise 12.3 Lorentz spinology 5 points
Exercise 12.4 Lorentz spin transformations 5 points

Bonus points
Exercise 12.5 The unbearable lightness of SO(4) Lie algebra 15 points

Total of 19 points = 100 % score.

151
 GROUP THEORY - WEEK 12. LORENTZ GROUP; SPIN

Lecture 23
2021-07-08 Predrag
SO(4) = SU(2) ⊗ SU(2); Lorentz group
Lecture 15 (Unedited) SU(2) irreps. SO(4) = SU(2) × SU(2). More importantly:
Minkowski metric, Lorentz group SO(1, 3) irreps are also labeled by pairs of
SU(2) × SU(2) irrep labels. (2:29:20 h)

Gutkin notes sect. 9.2 Representations of SU(2) and SO(3).

Gutkin notes sects. 9.5-9.8 Product representations of SO(3)

◦ sect. 12.3 Spinors and the Lorentz group

For Lorentz group, read Schwichtenberg [3] Sect. 3.7 (click here).

2021-07-13 Predrag Lecture 24 SO(1, 3); Spin

12.1 Other sources (optional)

◦ sect. 12.4 Irreps of SO(n)

14.3A Who ordered J+, J-? (7:37 min)

For SO(n) see also birdtracks.eu Chapt. 10 Orthogonal groups, pp. 121-123.

For SO(4) = SU(2) ⊗ SU(2) see also birdtracks.eu sect. 20.3.1 SO(4) or Cartan
A1 + A1 algebra.

◦ sect. 12.5 SO(4) of the Kepler problem

◦ sect. 12.5.1 Central force problems

Peter Voit [5] Quantum Theory, Groups and Representations (2017) has a nice
calculation of spherical harmonics as SO(3) eigenvectors in polar coordinates.
Used as problem set #6 for Quantum Mechanics I Georgia Tech PHYS-6105,
November 17, 2022, should make it a section in these notes. S

John Wood’s (click here) notes and exercise 12.5 The unbearable lightness of
SO(4) Lie algebra. The challenge: achieve some elegance in deriving the SO(4)
commutator relations.

Ilya Kuprov [2] ‘What exactly is spin?’ (40 min). Starts out with very enter-
taining bits of physics history. Wlad Sobol writes: “Kuprov derives the Dirac
equation from Wigner symmetry theorem, from a product of two Casimir opera-
tors of the Poincaré group." Revisit sect. 10.8.1 Quaternionic speculations for a
different point of view.

birdtracks.eu/course3 week12 152 2021-07-29

 GROUP THEORY - WEEK 12. LORENTZ GROUP; SPIN

12.2 Discussion (optional)

Henriette Roux In this course the Levi-Civita tensor appears to be the unique con-
nection for SO(4); but in GR, I learnt that the choice of connection is actually
arbitrary and there are theories of gravity which need not use the Levi-Civita
tensor. Are these two different concepts which are not necessarily linked?
Predrag Sean Carroll answers your question in arXiv:9712019. He does not un-
derstand that the invariant tensors are good, as they are what defines a given
symmetry group:
It is a remarkable property of the above tensors – the metric, the
inverse metric, the Kronecker delta, and the Levi-Civita tensor – that,
even though they all transform according to the tensor transformation
law, their components remain unchanged in any Cartesian coordinate
system in flat spacetime. In some sense this makes them bad examples
of tensors, since most tensors do not have this property.
However, he then goes on to explain that while in curved spacetime lengths and
volumes are measured in the spacetime dependent way, we still need a notion of
a volume of a hypercube as a skew product of its edges, ie, the determinant:
The Kronecker tensor can be thought of as the identity map from vec-
tors to vectors (or from dual vectors to dual vectors), which clearly
must have the same components regardless of coordinate system. The
other tensors (the metric, its inverse, and the Levi-Civita tensor) char-
acterize the structure of spacetime, and all depend on the metric. We
shall therefore have to treat them more carefully when we drop our
assumption of flat spacetime.
What he then does in his eq. (2.39) is to promote Levi-Civita from ‘tensor’ to
‘symbol’ in order to be able to compute determinants, just like we do in flat
space SO(n).
See also MathWorld discussion.
Are you happy now?
(A side, nomenclature remark: Levi-Civita is not a ‘connection’ in the sense the
word ‘connection’ is used in GR.)

12.3 Spinors and the Lorentz group

A Lorentz transformation is any invertible real [4 × 4] matrix transformation Λ,

x0µ = Λµν xν (12.1)

which preserves the Lorentz-invariant Minkowski bilinear form ΛT ηΛ = η,

xµ yµ = xµ ηµν y ν = x0 y 0 − x1 y 1 − x2 y 2 − x3 y 3

2021-07-29 153 birdtracks.eu/course3 week12

 GROUP THEORY - WEEK 12. LORENTZ GROUP; SPIN

with the metric tensor η = diag(1, −1, −1, −1).

A contravariant four-vector xµ = (x0 , x1 , x2 , x3 ) can be arranged [4] into a Her-
mitian [2×2] matrix in Herm(2, C) as
 0
x + x3 x1 − ix2

µ
x = σµ x = (12.2)
x1 + ix2 x0 − x3

in the hermitian matrix basis

σµ = σ̄ µ = (12 , σ) = (σ0 , σ1 , σ2 , σ3 ) , σ̄µ = σ µ = (12 , −σ) , (12.3)

with σ given by the usual Pauli matrices

     
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = . (12.4)
1 0 i 0 0 −1

With the trace formula for the metric

1
tr (σµ σ̄ν ) = ηµν , (12.5)
2
the covariant vector xµ can be recovered by
1 1
tr (xσ̄ µ ) = tr (xν σν σ̄ µ ) = xν ηνµ = xµ (12.6)
2 2
The Minkowski norm squared is given by

det x = (x0 )2 − (x1 )2 − (x2 )2 − (x3 )2 = xµ xµ , (12.7)

and with (12.3)

x0 − x3 −x1 + ix2
   
x0 + x3 x1 − ix2
x= = , (12.8)
−x1 − ix2 x0 + x3 x1 + ix2 x0 − x3

the Minkowski scalar product is given by

1
xµ yµ = tr(x y) . (12.9)
2
The special linear group SL(2, C) in two complex dimensions is given by the set
of all matrices Λ such that

SL(2, C) = {Λ ∈ GL(2, C) | det Λ = +1}. (12.10)

Let a matrix Λ ∈ SL(2, C) act on x ∈ Herm(2, C) as

x 7→ x0 = ΛxΛ† (12.11)

where † denotes Hermitian conjugation. The Minkowski scalar product is preserved,

det x0 = det x . Thus x0 can also be represented by a real linear combination of gener-
alized Pauli matrices
x0 = σµ x0µ with x0µ x0µ = xµ xµ (12.12)

birdtracks.eu/course3 week12 154 2021-07-29

 GROUP THEORY - WEEK 12. LORENTZ GROUP; SPIN

and Λ explicitly acts as a Lorentz transformation (12.1), with Λµν = 12 tr (σ̄ µ Λσν Λ† ) .
The mapping is two-to-one, as two matrices ±Λ ∈ SL(2, C) generate the same Lorentz
transformation ΛxΛ† = (−Λ)x(−Λ)† . This Λ belong to the proper orthochronous
Lorentz group SO+ (1, 3), and it can be shown that SL(2, C) is simply connected and
is the double universal cover of the SO+ (1, 3).
Consider the fully antisymmetric Levi-Civita tensor ε = −ε−1 = −εT in two
dimensions  
0 1
ε = iσ2 = . (12.13)
−1 0
This defines a symplectic (i.e., skew-symmetric) bilinear form hu, vi = −hv, ui on two
spinors u and v, elements of the two-dimensional complex vector (or spinor) space C2
 1   1 
u v
u= , v= , (12.14)
u2 v2
equipped with the symplectic form
hu, vi = u1 v 2 − u2 v 1 = uT εv . (12.15)
This symplectic form is SL(2, C)-invariant
hu, vi = uT εv = hΛu, Λvi = uT ΛT εΛv , (12.16)
so one can interpret the group acting on spinors as SL(2, C) ∼
= Sp(2, C) , the complex
symplectic group in two dimensions
Sp(2, C) = {Λ ∈ GL(2, C) | ΛT εΛ = ε} . (12.17)
Summary. The group of Lorentz transformations of spinors is the group SL(2,C)
of [2×2] complex matrices with determinant 1, i.e., the invariant tensor is the 2-index
Levi-Civita εAB . The SL(2,C) matrices are parametrized by three complex dimensions
and therefore six real ones (the matrices have four complex numbers and one complex
constraint on the determinant). This matches the 6 dimensions of the group manifold
associated with the Lorentz group SO(1, 3).
Andrew M. Steane writes “A spinor is the most basic mathematical object that can
be Lorentz-transformed.” His An introduction to spinors, arXiv:1312.3824, might help
you develop intuition about spinors.
Andrzej Trautman tracks the origin of spinors to Euclid, and General Relativity to
Clifford. He includes a letter from Hades saying, inter alia, “Unfortunately, it appears
that there is now in your world a race of vampires, called referees, who clamp down
mercilessly upon mathematicians unless they know the right passwords.”

12.4 Irreps of SO(n) (optional)

The dimension of the defining representation of is SO(n) is given by the trace of the
adjoint projection operator:

n(n − 1)
N = tr PA = = . (12.18)
2

2021-07-29 155 birdtracks.eu/course3 week12

 GROUP THEORY - WEEK 12. LORENTZ GROUP; SPIN

Young tableaux × = • + +
n(n−1) (n+2)(n−1)
Dimensions n2 = 1 + 2
+ 2
111
000  
1 000
111
000
111 1
Projectors = n
+ 000
111 + −
000
111
000
111
n

Table 12.1: SO(n) Clebsch-Gordan series for V ⊗V , worked out in detail in Group The-
ory – Birdtracks, Lie’s, and Exceptional Groups, birdtracks.eu Chapt. 10 Orthogonal
groups.

Dimensions of the other reps are listed in table 12.1.

12.5 SO(4) of the Kepler problem (optional)

One of “hidden” symmetries of quantum mechanics is the SO(4) of the Kepler prob-
lem.
John Baez discusses it in a fun read here: “if we take the angular momentum
together with the Runge–Lenz vector, we get 6 conserved quantities—and these turn
out to come from the group of rotations in 4 dimensions, SO(4), which is itself 6-
dimensional. The obvious symmetries in this group just rotate a planet’s elliptical
orbit, while the unobvious ones can also squash or stretch it, changing the eccentricity
of the orbit. [...] wavefunctions for bound states of hydrogen can be reinterpreted as
functions on the 3-sphere, S 3 . The sneaky SO(4) symmetry then becomes obvious: it
just rotates this sphere! And the Hamiltonian of the hydrogen atom is closely connected
to the Laplacian on the 3-sphere. The Laplacian has eigenspaces of dimensions n2
where n = 1, 2, 3, . . . , and these correspond to the eigenspaces of the hydrogen atom
Hamiltonian. ”
When the energy is fixed, the symmetry becomes Lie algebra SO(3, 1) for positive-
energy, scattering states, or SO(4) for negative-energy, bound states.
Michele Cini’s lecture notes, p. 18 gives hydrogen as an example of why we don’t
believe in miracles such as “accidental” eigenvalue degeneracies, but assume that we
must have missed a “hidden” symmetry. Cini writes: “Wolfgang Pauli in 1926 first
solved [...] the H atom using the SO(4) symmetry.” I didn’t know that it was Pauli...
To dig deeper, skim through Baez Mysteries of the gravitational 2-body problem.
Bander and Itzykson [1] Group theory and the hydrogen atom (I) might be OK, but
I have not read it.

12.5.1 Central force problems (optional)

For another way of looking at the H atom (and all solvable central force problems)
download John Wood’s chapter (click here) from Quantum Mechanics for Nuclear
Structure: I. A Primer, IOP science series.
exercise 12.5

birdtracks.eu/course3 week12 156 2021-07-29

EXERCISES

The SO(2, 1) method can be extended to solve relativistic central force problems
(one of my students did his Ph.D. thesis on this 20 years ago).
Q: Is the geometry associated with these algebraic structures, as applied to central
force problems, explored?

References
[1] M. Bander and C. Itzykson, “Group theory and the hydrogen atom (I)”, Rev.
Mod. Phys. 38, 330–345 (1966).
[2] I. Kuprov, Spin: From Basic Symmetries to Quantum Optimal Control (Springer,
New York, 2022).
[3] J. Schwichtenberg, Physics from Symmetry (Springer, Berlin, 2015).
[4] E. Wigner, “On unitary representations of the inhomogeneous Lorentz group”,
Ann. Math. 40, 149–204 (1939).
[5] P. Woit, Quantum Theory, Groups and Representations (Springer, New York,
2017).

Exercises
12.1. Lie algebra of SO(4) and SU(2) ⊗ SU(2). One particle Hamiltonian with a central
potential has in general SO(3) symmetry group. It turns out, however, that for Coulomb
potential the symmetry group is actually larger - SO(4), rather than SO(3). This explains
why the energy level degeneracies in the hydrogen atom are anomalously large. So SO(4)
and its representations are of a special importance in atomic physics.

(a) Show that the Lie algebra so(4) of the group SO(4) is generated by real antisym-
metric 4 × 4 matrices.
(b) What is the dimension of so(4)?

A natural basis in so(4) is provided by antisymmetric matrices Mµν , µ, ν ∈ 1, 2, 3, 4,

µ 6= ν, generators of SO(4) rotations which leave invariant the µν-plane. The elements
of these matrices are given by

(Mµν )ij = δiµ δjν − δjµ δiν (12.19)

(c) Check that these matrices satisfy the commutation relationship

[Mab , Mcd ] = Mad δbc + Mbc δad − Mac δbd − Mbd δac . (12.20)

(d) Show that Lie algebras of the groups SO(4) and SU(2) × SU(2) are isomorphic.
Path:

2021-07-29 157 birdtracks.eu/course3 week12

 EXERCISES

(d.i) Define matrices

1
Jk = εkij Mi,j , Kk = Mk4 , k = 1, 2, 3
2

and

1 1
Ak = (Jk + Kk ) and Bk = (Jk − Kk ) .
2 2
(d.ii) Show that A and B satisfy the same commutation relations as two copies of
su(2).
(e) How does one construct irreps of so(4) out of irreps of su(2)?
(f) Are groups SO(4) and SU(2) ⊗ SU(2) isomorphic to each other?

(B. Gutkin)
12.2. SO(n) Clebsch-Gordan series for V ⊗V .
(a) Show that the product of two n-dimensional reps of SO(n) decomposes into three
irreps:
000
111  
1 000
111 1
= + 000
111 + − . (12.21)
000
111
n 000
111 n
000
111

(b) Compute the dimensions of the three irreps.

(c) Which one is the adjoint one, and why? Hint: check the invariance condition. (This
is worked out in detail in Group Theory – Birdtracks, Lie’s, and Exceptional Groups,
birdtracks.eu Chapt. 10 Orthogonal groups.)
12.3. Lorentz spinology.
Show that
(a)
x2 = xµ xµ = det x (12.22)
(b)
1
xµ y µ = (det (x + y) − det (x) − det (y)) (12.23)
2
(c)
1
xµ y µ = tr (x y) , (12.24)
2
µ
where y = σ̄µ y
12.4. Lorentz spin transformations.
Let a matrix Λ ∈ SL(2, C) act on hermitian matrix x as

x 7→ x0 = ΛxΛ† . (12.25)

(a) Check that x0 is Hermitian, and the Minkowski scalar product (12.23) is preserved.
(b) Show that Λ explicitly acts as a Lorentz transformation x0µ = Λµν xν .
(c) Show that the mapping from a Λ ∈ SL(2, C) to the Lorentz transformation in
SO(1, 3) is two-to-one.

birdtracks.eu/course3 week12 158 2021-07-29

EXERCISES

(d) Consider the Levi-Civita tensor  = −−1 = −T in two dimensions,

 
0 1
= , (12.26)
−1 0

and the associated symplectic form

hu, vi = uT εv = u1 v 2 − u2 v 1 . (12.27)

Show that this symplectic form is SL(2, C)-invariant

hu, vi = uT εv = hΛu, Λvi = uT ΛT εΛv . (12.28)

12.5. The unbearable lightness of SO(4) Lie algebra. Download John Wood’s (click here)
notes. The challenge: achieve some elegance in deriving the SO(4) commutator bracket
relations, for example reduce the number of steps in the calculation by 30% or 50%.
The prize: a case of beer, details to be negotiated with John.
The challenges start on p. 9-8, following eq. (9.21), i.e., “(i)”, “(iv)”, and “(v)”. For
instance, on p. 9-11 John indicates all of the cancellations. These suggest that his solution
is “calculating zero” unnecessarily. One could take linear combinations of the operators
that possess these commutator bracket relations; but the combinations do not seem a priori
warranted on the basis of the dynamics of the problem.
(J. Wood)

2021-07-29 159 birdtracks.eu/course3 week12

 group theory - week 13

Simple Lie algebras; SU(3)

Homework HW13

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Exercise 13.1 Root systems of simple Lie algebras 5 points

Exercise 13.2 Meson octet 5 points

Bonus points
Exercise 13.3 SU(3) symmetry in 3D harmonic oscillator 5 points

Total of 10 points = 100 % score. Extra points accumulate, can help you later if you
miss a few problems.

161
 GROUP THEORY - WEEK 13. SIMPLE LIE ALGEBRAS; SU(3)

2021-07-13 Predrag Lecture 25 Reps of simple Lie algebras

In week 4 we learned that for finite groups there is one very special matrix rep, the
regular representation constructed from the group multiplication table, that is intrinsic
to the abstract group itself, and whose reduction yields all irreps of a given group.
In week 9 we saw that for continuous groups we need to study the Lie algebra
of the finite number of generators Tj , rather than the infinity of group elements g =
exp(iφ · T ). Here the finite group multiplication table is replaced by the ‘Lie product’,
i.e., the table of Lie commutators’ fully antisymmetric structure constants iCijk .
So far we have chosen the hermitian basis Tj . But non-hermitian bases are also
OK, as we know from raising / lowering operators of SU(2) of quantum mechanics
irrep constructions.

(Unedited 34:45 min) 16.1 Adjoint representation and Killing form.

Gutkin notes, sect. 10.1 Adjoint representation and Killing form.
(Unedited 15:13 min) 16.2 Cartan sub-algebra and roots.
Gutkin notes, Sect. 10.2 Cartan sub-algebra and roots.
(Unedited 19:02 min) 16.3 Root systems.
Gutkin notes, Sect. 10.3 Main properties of root systems.
(Unedited 21:30 min) 16.4 Construction of representations. SU(2) example; anal-
ogy to presentations of finite groups, such as Dn (4.1); ‘grand circles’ on Cartan
lattices.
Gutkin notes, Sect. 10.4 Building up representations of g.

2021-07-15 Predrag Lecture 26 Cartan’s SU(3) irreps

(Unedited 2:03:14 h) Representations of SU(3).
Gutkin notes, Sect. 10.5 Representations of SU(3).

13.1 Group theory news (optional)

Turns out, applications of group theory go way beyond what is covered in this course:

Mathematicians map E8 , and it is bigger than the human genome.

Group theory of defamation: The officers argued Sawant’s statements impugned
them individually even though she only spoke about the police department as
a whole. The court says suing as individuals and advancing a group theory of
defamation takes far more than the officers showed in their complaint.

birdtracks.eu/course3 week13 162 2021-07-29

EXERCISES

(a) (b)

Figure 13.1: (a) The meson (pseudoscalars) octet. (b) The quark triplet, the anti-quark
triplet and the gluon octet. (Wikipedia).

[W]hether proceeding under an individual or group theory, Plain-

tiffs must plead that the statements “specifically” identified or singled
them out, or was understood as “referring to [them] in particular.”
Sims, 20 Wn. App. at 236.

Exercises
13.1. Root system of simple Lie algebras.
a) Determine dimensions of Lie algebras so(N ), su(N ) and dimensions of their Cartan
subalgebras. What is the number of the positive roots for these Lie algebras?
b) Show that N × N diagonal matrices Hi with zero traces and uper/lower corner N × N
(a,b)
matrices E (a,b) with the elements Ei,j = δia δib provide Cartan-Weyl basis of su(N ).
To put it differently, show that E (a,b) are eigenstates for adjoint representation of Hi ’s.
(B. Gutkin)
13.2. Meson octet. In Gutkin lecture notes, Lect. 11 Strong interactions: flavor SU(3), the
meson octet, figure 13.1 (a)
 π0 η
π+ K+

√ + √
2 6
Φ =  π− π0
−√ + √η6 K0 
 
2
K− K0 −√2η
6
 0  
π +
π 0 K+
  
√ 0 0 1 0 0
 −2 0 0 η
= π π + 0 0 K  + √ 0 1 0  (13.1)
,

−√ 0
2
K− K0 0 6 0 0 −2
0 0 0

is interpreted as arising from the adjoint representation of SU(3), i.e., the traceless part of
the quark-antiquark 3 ⊗ 3 = 1 ⊕ 8 outer product (see figure 13.1 (b)),
 
uu ud us
du dd ds . (13.2)
su sd ss

2021-07-29 163 birdtracks.eu/course3 week13

 EXERCISES

where we have replaced in (13.1) the constituent q ⊗ q combinations by the names of the
elementary particles they build.
Given the quark quantum numbers

Q I I3 Y B
u 2/3 1/2 1/2 1/3 1/3
d -1/3 1/2 -1/2 1/3 1/3
s -1/3 0 0 -2/3 1/3

verify the strangeness and charge assignments of figure 13.1 (a).

13.3. SU(3) symmetry in 3D harmonic oscillator. The Hamiltonian of 3D isotropic har-
monic oscillator is given by
3 3
X p2i mω 2 2 X
H= + xi = ~ω (a†i ai + 1/2),
i=1
2m 2 i=1
q
is creation ( a†i resp. annihilation) operator satisfying
p mω 1
where ai = 2~
xi + i p
2mω~ i
[ai , a†j ]
= δij , [ai , aj ] = 0.
a) Show that ai → Ui,j aj , with U ∈ U (3) is a symmetry of the Hamiltonian. In other
words isotropic 3D harmonic oscillator has U (3) rather than O(3) symmetry!
b) Calculate degeneracy of the n-th level En = ω~(n + 3/2) of the oscillator.
c) By comparison of dimensions find out which representations of SU(3) appear in the
spectrum of harmonic oscillator.
(B. Gutkin)

birdtracks.eu/course3 week13 164 2021-07-29

 group theory - week 14

Flavor SU(3)

Homework HW14

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Bonus points
Exercise 14.1 Gell-Mann–Okubo mass formula 8 points
Exercise 15.3 Young tableaux for SU(3) 3 points
Exercise 15.4 Irrep projection operators for unitary groups 5 points

All points are bonus points. Extra points accumulate, can help you if you had missed a
few problems.

165
 GROUP THEORY - WEEK 14. FLAVOR SU(3)

Figure 14.1: A lattice gauge theory calculation of the light QCD spectrum. Horizontal
lines and bands are the experimental values with their decay widths. The π, K and and
Ξ have no error bars because they are used to set the light and strange quark masses
and the overall scale respectively. From Scholarpedia.

2021-07-22 Predrag Lecture 27 Flavor SU(3)

(Unedited 1:26:11 h) Gell-Mann–Okubo mass formula. Pions and kaons fit into an
SU(3) octet, but the strange mesons masses are much larger, breaking the sym-
metry SU(3) → SU(2) × U (1). That leads to a Gell-Mann – Okubo parameter-
free constraints on various masses, which were verified to high accuracy, and
lead to predictions of masses for yet undiscovered particles.
Gutkin notes, Lect. 11 Strong interactions: flavor SU(3). Heisenberg isospin
SU(2). Gell-Mann flavor SU(3). Gell-Mann-Okubo mass formula.

The Gell-Mann-Okubo mass sum rules [1, 2, 4] are an easy consequence of the
approximate SU(3) flavor symmetry. Determination of quark masses is much harder -
they are parameters of the standard model, determined by optimizing the spectrum of
particle masses obtained by lattice QCD calculations as compared to the experimental
baryon and meson masses. The best determination of the mass spectrum as of 2012
is given in figure 14.1. Up, down quarks are about 3 and 6 MeV, respectively, with
strange quark mass about 100 MeV, all with large error brackets. As of 2021, I have
not found an update to figure 14.1, but the latest on the subject can probably be traced
in Georg von Hippel’s latticeqcd.blogspot.com.

14.1 Isotropic quantum harmonic oscillator

One of the “hidden" quantum symmetries is the SU(3) of the 3D isotropic quantum har-
monic oscillator. Murgan and Zender [3] Energy eigenvalues of the three-dimensional
quantum harmonic oscillator from SU(3) cubic Casimir operator is a nice pedagog-
ical intro to this SU(3). Would prefer no explicit irreps (see www.birdtracks.eu) but
working this out is a good (but long) exercise.

birdtracks.eu/course3 week14 166 2021-07-29

EXERCISES

References
[1] M. Gell-Mann, The Eightfold Way: A Theory of Strong Interaction Symmetry
(CalTech, 1961).
[2] M. Gell-Mann, “Symmetries of baryons and mesons”, Phys. Rev. 125, 1067–
1084 (1962).
[3] R. Murgan and A. Zender, “Energy eigenvalues of the three-dimensional quan-
tum harmonic oscillator from SU(3) cubic Casimir operator”, Eur. J. Phys. 40,
015405 (2018).
[4] S. Okubo, “Note on unitary symmetry in strong interactions”, Progr. Theor.
Phys. 27, 949–966 (1962).

Exercises
14.1. Gell-Mann–Okubo mass formula. The mass symmetry-breaking interaction for
an isospin multiplet is proportional to the 3rd component of the isospin operator, I3 .
Similarly, the symmetry-breaking interaction of SU(3) for the meson octet is given by the
8th component of the octet operator Y = λ8 . Derive the GMO mass formula for mesons

4 m2K − m2π
m2η = . (14.1)
3
by eliminating the parameter for the strength of this interaction, as in Gutkin lecture notes,
Lect. 11 Strong interactions: flavor SU(3) .

2021-07-29 167 birdtracks.eu/course3 week14

 group theory - week 15

Many particle systems. Young

tableaux

Homework HW15

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Bonus points
Exercise 15.1 Representations of SU(3) 5 points
Exercise 15.2 Young tableaux for S5 3 points
Exercise 15.3 Young tableaux for SU(3) 3 points
Exercise 15.4 Irrep projection operators for unitary groups 5 points

All points are bonus.

169
 GROUP THEORY - WEEK 15. MANY PARTICLE SYSTEMS. YOUNG
TABLEAUX

(no videos) Predrag Lecture 28 Many particle systems. Young

tableaux
Gutkin notes, Lect. 12 Many particle systems.
Excerpt from Predrag’s monograph [5], fetch it here: Sect. 9.3 Young tableaux.
(no videos) Predrag Lecture 29 Young tableaux
Excerpts from Predrag’s monograph [5], fetch them here:
Sect. 2.2 First example: SU(n) (skim over casimirs and beyond: this example
gives you a flavor of birdtracks computations, you do not need to work it out in
detail),
Sect. 6.1 Symmetrization,
Sect. 6.2 Antisymmetrization,
Sect. 9.1 Two-index tensors,
Sect. 9.2 Three-index tensors, and Table 9.1.
Reading for this week: Sect. 9.3 Young tableaux.

Young tableaux for SU(3) and SU(n) have not yet been covered in the lectures, but
you can easily learn them yourself, from, for example, Gutkin notes, Lect. 12 Young
tableaux. Boris Gutkin is a professor, beyond learning new stuff, so he follows old
fashioned references such as Fulton and Harris [6]. The resulting simple recipe with 0
explanation can be found, for example, here: Young diagrams by C.G. Wohl.
A modern exposition is given in Group Theory – Birdtracks, Lie’s, and Exceptional
Groups, birdtracks.eu Chapt. 9 Unitary groups. Currently I am a fan of the Alcock-
Zeilinger algorithm [1–3], based on the simplification rules of ref. [2], which leads to
explicitly Hermitian and compact expressions for the projection operators.
Probably best to read Alcock-Zeilinger course The Special Unitary Group, Bird-
tracks, and Applications in QCD notes [4]. Alcock-Zeilinger fully supersedes Cvi-
tanović’s formulation, and any future full exposition of birdtracks reduction of SU(N )
tensor products into irreducible representations should be based on the Alcock-Zeilin-
ger algorithm.

15.1 Other sources (optional)

The clearest current exposition and the most powerful irrep reduction of SU(n) is given
in the triptych of papers by Judith Alcock-Zeilinger and her thesis adviser H. Weigart,
University of Cape Town:

Simplification rules for birdtrack operators [3],

Compact Hermitian Young projection operators [2], and
Transition operators [1].

Probably best to read Alcock-Zeilinger course The Special Unitary Group, Bird-
tracks, and Applications in QCD notes [4]. You want to study these in detail if your
research leads you to study of multiparticle states.

birdtracks.eu/course3 week15 170 2021-07-29

EXERCISES

References
[1] J. Alcock-Zeilinger and H. Weigert, “Transition operators”, J. Math. Phys. 58,
051702 (2016).
[2] J. Alcock-Zeilinger and H. Weigert, “Compact Hermitian Young projection op-
erators”, J. Math. Phys. 58, 051702 (2017).
[3] J. Alcock-Zeilinger and H. Weigert, “Simplification rules for birdtrack opera-
tors”, J. Math. Phys. 58, 051701 (2017).
[4] J. M. Alcock-Zeilinger, The Special Unitary Group, Birdtracks, and Applica-
tions in QCD, tech. rep. (Univ. Tübingen, 2018).
[5] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[6] W. Fulton and J. Harris, Representation Theory (Springer, New York, 1991).

Exercises
15.1. Representations of SU(3). Any irrep of SU(3) can be labeled D(p, q) by its highest
weight λ = pλ1 + qλ2 , where λ1,2 are the two fundamental weights.
(a) Find all irreps D(p, q) of SU(3) with the dimensions less then 20 (see lecture notes
for the dimensions of D(p, q)).
(b) Draw the lattice Λ generated by λ1,2 and mark there all the weights v (i.e., lattice
nodes) which belong to irrep. D(3, 0). Is D(3, 0) a real irrep?
(c) Consider product (reducible) representation 3 ⊗ 3, where 3 = D(1, 0) is the fun-
damental irrep. Mark all the weights v on Λ which belong to 3 ⊗ 3. Using this find
out decomposition of 3 ⊗ 3 into irreps:

3 ⊗ 3 =  ⊕ 4,  =?, 4 =?

Hint: see lecture notes for similar exercise on 3 ⊗ 3̄.

(d) Using previous results find decomposition of 3 ⊗ 3 ⊗ 3 into irreps.
(B. Gutkin)
15.2. Young tableaux for S5 .
(a) Draw all Young diagrams for the symmetric group S5 . How many irreducible rep-
resentations has it? Which of the diagrams correspond to one-dimensional irreps?
(b) Find Young diagram corresponding to the irrep of S5 with the largest dimension?
Draw Young tableaux corresponding to this irrep/Young diagram. What is the di-
mension of this irrep?
(c) What are the dimensions of the remaining irreps?
(B. Gutkin)

2021-07-29 171 birdtracks.eu/course3 week15

 EXERCISES

15.3. Young tableaux for SU(3). Solve exercise 15.1 (c,d) by using Young tableaux.
Remark: If Young tableaux for SU(3) are not covered in the lectures, learn them your-
self from, for example, birdtracks.eu Group Theory Birdtracks, Lie’s, and Exceptional
Groups. The resulting simple recipe with 0 explanation can be found, for example, here
C.G. Wohl.
(B. Gutkin)
15.4. Irrep projection operators for unitary groups. Derive projection operators and
dimensions for irreps of the Kronecker product of the defining and the adjoint reps of
SU(n) listed in Group Theory Birdtracks, Lie’s, and Exceptional Groups, birdtracks.eu
table 9.3. (Ignore “indices," we have not defined them here.)

birdtracks.eu/course3 week15 172 2021-07-29

 group theory - week 16

Wigner 3- and 6-j coefficients

Homework HW16

== show all your work for maximum credit,

== put labels, title, legends on any graphs
== acknowledge study group member, if collective effort
== if you are LaTeXing, here is the source code

Bonus points
Exercise 16.1 Gravity tensors, part (a) 2 points
Exercise 16.1 Gravity tensors, part (b) 4 points
Exercise 16.1 Gravity tensors, part (c) 1 point
Exercise 16.1 Gravity tensors, part (d) 2 points
Exercise 16.1 Gravity tensors, part (e) 3 points
Exercise 16.1 Gravity tensors, part (f) 4 points
Exercise 16.1 Gravity tensors, part (g) 3 points
Exercise 16.1 Gravity tensors, part (h) 6 points
Exercise 16.1 Gravity tensors, part (i) 4 points
Exercise 16.1 Gravity tensors, part (j) 10 points

All points are bonus.

173
 GROUP THEORY - WEEK 16. WIGNER 3- AND 6-J COEFFICIENTS

2021-07-27 Predrag Lecture 30 Wigner 3- and 6-j coefficients

(Unedited 2:04:59 h) Sex, Lies and Videotape.
What are we really trying to compute? Wigner 3n-j coefficients, birdtracks. How
simple calculations lead to all Lie groups, including the exceptional ones.

The webbook for cyclists (medium level): Tracks, Lie’s, and Exceptional Magic.
Most of the webbook at a cyclist pace, in 50 overheads.

Birdtracks: Excerpts from Predrag’s monograph [4], fetch them here:

Background reading on groups, vector spaces, tensors, invariant tensors, invari-
ance groups (my advice is to start with Sect. 5.1 Couplings and recouplings, then
backtrack to these introductory sections as needed):
Sect. 3.2 Defining space, tensors, reps,
Sect. 3.3 Invariants,
Sect. 4.1 Birdtracks,
Sect. 4.2 Clebsch-Gordan coefficients, and
Sect. 4.3 Zero- and one-dimensional subspaces.

The final result is invariant and highly elegant: any group-theoretical invariant
quantity can be expressed in terms of Wigner 3- and 6-j coefficients:
Sect. 5.1 Couplings and recouplings,
Sect. 5.2 Wigner 3n-j coefficients, and
Sect. 5.3 Wigner-Eckart theorem.
The rest is just bedside reading, nothing technical:
Sect. 4.8 Irrelevancy of clebsches and
Sect. 4.9 A brief history of birdtracks.

Course finale: Indiana Jones vs. Fancy Footwork. (1:30 min)

It’s a matter of no small pride for a card-carrying dirt physics theorist to claim
full and total ignorance of group theory.

16.1 Other sources (optional)

Predrag’s monograph [4], Group Theory Birdtracks, Lie’s, and Exceptional Groups,
Sect. 2.1 Basic concepts; Sect. 2.1 First example: SU(n); Chap. 1 explains pretty
well what the monograph is about.

◦ A birdtracks refresher: sect. 2.7 Permutations in birdtracks

◦ Sect. 10.7 Lie groups for pedestrians

◦ sect. 10.10 Birdtracks - updated history

birdtracks.eu/course3 week16 174 2021-07-29

 GROUP THEORY - WEEK 16. WIGNER 3- AND 6-J COEFFICIENTS

16.2 Gruppenpest (optional)

A practically-minded physicist always has been, and continues to be resistant to grup-
penpest. Apparently already in 1910 James Jeans wrote, while discussing what should
a physics syllabus contain: “We may as well cut out the group theory. That is a subject
that will never be of any use in physics.” In 1963 Eugene Wigner got the Nobel Prize
in Physics, so by mid 60’s gruppenpest was accepted in finer social circles.

ChaosBook Appendix A.6 Gruppenpest

Woit writes here about the “The Stormy Onset of Group Theory in the New Quan-
tum Mechanics,” citing Bonolis [2] From the rise of the group concept to the
stormy onset of group theory in the New Quantum Mechanics. A saga of the
invariant characterization of physical objects, events and theories.

Chayut [3] From the periphery: the genesis of Eugene P. Wigner’s application of
group theory to quantum mechanics traces the origins of Wigner’s application
of group theory to quantum physics to his early work as a chemical engineer,
in chemistry and crystallography. “In the early 1920s, crystallography was the
only discipline in which symmetry groups were routinely used. Wigner’s early
training in chemistry exposed him to conceptual tools which were absent from
the pedagogy available to physicists for many years to come. This both enabled
and pushed him to apply the group theoretic approach to quantum physics. It
took many years for the approach first introduced by Wigner in the 1920s – and
whose reception by the physicists was initially problematical – to assume the
pivotal place it now holds.” Another historical exposition is given by Scholz [6]
Introducing groups into quantum theory (1926–1930).

So what is group theory good for? By identifying the symmetries, one can apply
group theory to determine good quantum numbers which describe a physical state (i.e.,
the irreps). Group theory then says that many matrix elements vanish, or shows how
are they related to others. While group theory does not determine the actual value of a
matrix element of interest, it vastly simplifies its calculation.
The old fashioned atomic physics, fixated on SO(3) / SU(2), is too explicit, with too
many bras and kets, too many square roots, too many deliriously complicated Clebsch-
Gordan coefficients that you do not need, and way too many labels, way too explicit for
you to notice that all of these are eventually summed over, resulting in a final answer
much simpler than any of the intermediate steps.
I wrote my book [4] Group Theory - Birdtracks, Lie’s, and Exceptional Groups to
teach you how to compute everything you need to compute, without ever writing down
a single explicit matrix element, or a single Clebsch-Gordan coefficient. There are two
versions. There is a particle-physics / Feynman diagrams version that is index free,
graphical and easy to use (at least for the low-dimensional irreps). The key insights
are already in Wigner’s book [8]: the content of symmetry is a set of invariant numbers
that he calls 3n-j’s. Then there are various mathematical flavors (Weyl group on Cartan
lattice, etc.), elegant, but perhaps too elegant to be computationally practical.

2021-07-29 175 birdtracks.eu/course3 week16

 EXERCISES

But it is nearly impossible to deprogram people from years of indoctrination in

QM and EM classes. The professors have no time to learn new stuff, and students love
manipulating their mu’s and nu’s.

Nothing to be done... (2:18:01 h)

Bonus: While waiting, exhale. (2:18:01 h)

References
[1] S. L. Adler, J. Lieberman, and Y. J. Ng, “Regularization of the stress-energy ten-
sor for vector and scalar particles propagating in a general background metric”,
Ann. Phys. 106, 279–321 (1977).
[2] L. Bonolis, “From the Rise of the Group Concept to the Stormy Onset of Group
Theory in the New Quantum Mechanics. A saga of the invariant characterization
of physical objects, events and theories”, Rivista Nuovo Cim. 27, 1–110 (2005).
[3] M. Chayut, “From the periphery: the genesis of Eugene P. Wigner’s application
of group theory to quantum mechanics”, Found. Chem. 3, 55–78 (2001).
[4] P. Cvitanović, Group Theory: Birdtracks, Lie’s and Exceptional Groups (Prince-
ton Univ. Press, Princeton NJ, 2008).
[5] R. Penrose, “Applications of negative dimensional tensors”, in Combinatorial
mathematics and its applications, edited by D. J. A. Welsh (Academic, New
York, 1971), pp. 221–244.
[6] E. Scholz, “Introducing groups into quantum theory (1926-1930)”, Hist. Math.
33, 440–490 (2006).
[7] S. Weinberg, Gravitation and Cosmology: Principles and Applications of the
General Theory of Relativity (Wiley, New York, 1972).
[8] E. P. Wigner, Group Theory and Its Application to the Quantum Mechanics of
Atomic Spectra (Academic, New York, 1931).

Exercises
16.1. Gravity tensors. In this problem we will apply diagrammatic methods (“birdtracks”)
to construct and count the numbers of independent components of the “irreducible rank-
four gravity curvature tensors.” However, any notation that works for you is OK, as long
as you obtain the same irreps and their dimensions. The goal of this exercise (longish, as
much of it is the recapitulation of the material covered in the book) is to give you basic
understanding for how Young tableaux work for groups other than U(n). We start with

Part 1 : U(n) Young tableaux decomposition.

birdtracks.eu/course3 week16 176 2021-07-29

EXERCISES

(a) The Riemann-Christoffel curvature tensor of general relativity has the following
symmetries (see, for example, Weinberg [7] or the Riemann curvature tensor wiki):
Rαβγδ = −Rβαγδ (16.1)
Rαβγδ = Rγδαβ (16.2)
Rαβγδ + Rβγαδ + Rγαβδ = 0. (16.3)
Introducing a birdtrack notation for the Riemann tensor
α
β
Rαβγδ =
γ R , (16.4)
δ
check that we can state the above symmetries as
Rαβγδ = −Rβαγδ

R = R , (16.5)

Rαβγδ = Rγδαβ

R = R , (16.6)

Rαβγδ + Rβγαδ + Rγαβδ = 0

R + R + R =0. (16.7)

The first condition says that R lies in the ⊗ subspace.

(b) The second condition says that R lies in the ↔ interchange-symmetric sub-
space.

Use the characteristic equation for

to split into the and irreps:

!
1 4
+ = + . (16.8)
2 3

(c) Show that the third condition (16.7) says that R has no components in
the irrep:

R + R + R =3 R = 0 . (16.9)

Hence, the symmetries of the Riemann tensor are summarized by the irrep
projection operator [5]:
α α´
0 0 0 0 4β β´
(PR )αβγδ ,δ γ β α = γ´ (16.10)
3γ
δ δ´

2021-07-29 177 birdtracks.eu/course3 week16

 EXERCISES

(d) Verify that the Riemann tensor is in the subspace

0 0 0
γ β α0
(PR R)αβγδ = (PR )αβγδ ,δ Rα0 β 0 γ 0 δ0 = Rαβγδ
4
3
R = R . (16.11)

(e) Compute the number of independent components of the Riemann tensor Rαβγδ by
taking the trace of the irrep projection operator:

n2 (n2 − 1)
dR = tr PR = . (16.12)
12

Part 2 : SO(n) Young tableaux decomposition

The Riemann tensor has the symmetries of the irrep of U(n). However, gravity
is also characterized by the symmetric tensor gαβ , that reduces the symmetry to a local
SO(n) invariance (more precisely SO(1, n − 1), but compactness is not important here).
The extra invariants built from gαβ ’s decompose U(n) reps into sums of SO(n) reps.
Orthogonal group SO(n) is the group of transformations that leaves invariant a symmetric
quadratic form (q, q) = gµν q µ q ν , with a primitive invariant rank-2 tensor:
gµν = gνµ = µ ν µ, ν = 1, 2, . . . , n . (16.13)
∗ µν
If (q, q) is an invariant, so is its complex conjugate (q, q) = g qµ qν , and
g µν = g νµ = µ ν (16.14)
is also an invariant tensor. The matrix Aνµ σν
= gµσ g must be proportional to unity, as
otherwise its characteristic equation would decompose the defining n-dimensional rep. A
convenient normalization is
gµσ g σν = δµν
= . (16.15)
As the indices can be raised and lowered at will, nothing is gained by keeping the arrows.
Our convention will be to perform all contractions with metric tensors with upper indices
and omit the arrows and the open dots:
g µν ≡ µ ν. (16.16)

The U(n) 2-index tensors can be decomposed into a sum of their symmetric and antisym-
metric parts. Specializing to the subgroup SO(n), the rule is to lower all indices on all
tensors, and the symmetrization projection operator is written as
0 0
Sµν,ρσ = gρρ0 gσσ0 Sµν ,ρ σ
1
= (gµσ gνρ + gµρ gνσ )
2
From now on, we drop all arrows and g µν ’s and write the decomposition into symmetric
and antisymmetric parts as

= +

1 1
gµσ gνρ = (gµσ gνρ + gµρ gνσ ) + (gµσ gνρ − gµρ gνσ ) . (16.17)
2 2

birdtracks.eu/course3 week16 178 2021-07-29

EXERCISES

The new invariant tensor, specific to SO(n), is the index contraction:

Tµν,ρσ = gµν gρσ , T= . (16.18)

Its characteristic equation

T2 = = nT (16.19)

yields the trace and the traceless part projection operators. As T is symmetric, ST = T,
only the symmetric subspace is reduced by this invariant.

(f) Show that SO(n) 2-index tensors decompose into three irreps:

traceless symmetric:

1 1 1
(P2 )µν,ρσ = (gµσ gνρ + gµρ gνσ ) − gµν gρσ = − ,
2 n n
(16.20)
1 1
singlet: (P1 )µν,ρσ = gµν gρσ = , (16.21)
n n
1
antisymmetric: (P3 )µν,ρσ = (gµσ gνρ − gµρ gνσ ) = .
(16.22)
2
What are the dimensions of the three irreps?
(g) In the same spirit, the U(n) irrep is decomposed by the SO(n) intermediate
2-index state invariant matrix

Q= . (16.23)

Show that the intermediate 2-index subspace splits into three irreducible reps by
(16.20) – (16.22):
( )
1 1
Q = + − +
n n
= Q0 + QS + QA . (16.24)

Show that the antisymmetric 2-index state does not contribute

P R QA = 0 . (16.25)

(Hint: The Riemann tensor is symmetric under the interchange of index pairs.)
(h) Fix the normalization of the remaining two projection operators by computing
Q2S , Q20 :

2
P0 = , (16.26)
n(n − 1)
( )
4 1
PS = − (16.27)
n−2 n

2021-07-29 179 birdtracks.eu/course3 week16

 EXERCISES

and compute their dimensions.

This completes the SO(n) reduction of the U(n) irrep (16.11):

U(n) → SO(n)
→ + + ◦
PR = PW + PS + P0
n2 (n2 −1) (n+2)(n+1)n(n−3) (n+2)(n−1)
12
= 12
+ 2
+ 1

(16.28)
The projection operator for the SO(n) traceless irrep is:

PW = PR − PS − P0
4 4 2
PW = − + (16.29).
3 n−2 (n − 1)(n − 2)

(i) The above three projection operators project out the standard, SO(n)-irreducible
general relativity tensors:

Curvature scalar:
R=− R = Rµνµν (16.30)

Traceless Ricci tensor:

1 1
Rµν − gµν R = − R +n R (16.31)
n

Weyl tensor:

Cλµνκ = (PW R)λµνκ

4 2
= R − n−2 R + (n − 1)(n − 2) R
1
= Rλµνκ + (gµν Rλκ − gλν Rµκ − gµκ Rλν + gλκ Rµν )
n−2
1
− (gλκ gµν − gλν gµκ )R . (16.32)
(n − 1)(n − 2)

The numbers of independent components of these tensors are given by the dimen-
sions of corresponding irreducible subspaces in (16.28).
What is the lowest dimension in which the Ricci tensor contributes? the Weyl tensor
contributes? Show that in 2, respectively 3 dimensions, we have

n=2: Rλµνκ = (P0 R)λµνκ = 21 (gλν gµκ − gλκ gµν )R

n=3: = gλν Rµκ − gµν Rλκ + gµκ Rλν − gλκ Rµν (16.33)
− 12 (gλν gµκ − gλκ gµν )R .

birdtracks.eu/course3 week16 180 2021-07-29

EXERCISES

(j) The last example of this exercise is an application of birdtracks to general relativ-
ity index manipulations. The object is to find the characteristic equation for the
Riemann tensor in four dimensions.
The antisymmetrization tensor Aa1 a2 ... ,bp ...b2 b1 has nonvanishing components, only
if all lower (or upper) indices differ from each other. If the defining dimension is
smaller than the number of indices, the tensor A has no nonvanishing components:
1
2
=0 if p > n . (16.34)

...
...
p

This identity implies that for p > n, not all combinations of p Kronecker deltas are
linearly independent. A typical relation is the p = n + 1 case

... ... ... ...

0= = − + − ... . (16.35)
1 2 ... n+1

Contract (16.34) with two Riemann tensors:

0= R , (16.36)

and obtain the characteristic equation by expanding with (16.35):

R R R R
0 = 2 −4

−4
R R + 2R
R (16.37)

R +1 R R
( )
R2 R
− −2 .
2 2

This identity has been used by Adler et al., eq. (E2) in ref. [1].

2021-07-29 181 birdtracks.eu/course3 week16

 group theory - week 17

An overview, and the epilogue

If I had had more time, I would have written less

— Blaise Pascal, a remark made to a correspondent

This whole course has only one message:

If you have a symmetry, use it!

These notes in isolation do not make much sense - the essence of teaching is unveiling
of concepts on a black/white board at human pace, interacting in live time. But never-
theless the notes might be useful to you, as they are hyperlinked to the literature that
develops a given topic into depth. Here is a brief summary of the course, the ideas you
want to take with you:

week 1 Linear algebra

The key idea: Projection operators (1.27) use eigenvalues of a matrix to split
(reduce) a vector space into subspaces.

week 2 Finite groups

Groups, permutations, group multiplication tables, rearrangement theorem, sub-
groups, cosets, classes.

week 3 Group representations

Irreps, regular representation. So far, everything was intuitive: a representation
of a group was bunch of 0’s and 1’s indicating how a group operation permutes
physical objects. But now the first surprise:
Any representation of any finite group can be put into unitary form, and so
complex-valued vector spaces and unitary representation matrices make their
entrance.

183
 GROUP THEORY - WEEK 17. AN OVERVIEW, AND THE EPILOGUE

week 4 Characters
Schur’s Lemma. Unitary matrices can be diagonalized, and from that follows
the Wonderful Orthogonality Theorem for Characters (coordinate independent,
intrinsic numbers), and the full reducibility of any representation of any finite
group.
week 5 Classes
The algebra of central or ‘all-commuting’ class operators, connects the reduction
in terms of characters to the projection operators of week 1. The key idea:
Define a group by what objects (primitive invariant tensors) it leaves invariant.
week 6 Fundamental domain
Dynamical systems application: the Lorenz flow, its C2 symmetry and its desym-
metrization: if the system is nonlinear, its symmetry reduction is not easy.
week 7 Discrete Fourier representation
So far, everything was finite and compact. Next: two distinct ways of going
infinite: (a) discrete translations, exemplified by deterministic diffusion and spa-
ce groups of week 8, and (b) continuous Lie groups, exemplified by rotations of
week 9.
week 8 Space groups
Translation group, Bravais lattice, wallpaper groups, reciprocal lattice, Brilluoin
zone.
week 9 Continuous groups
Lie groups. Matrix representations. Invariant tensors. Lie algebra. Adjoint
representation, Jacobi relation. Birdtracks.
Irreps of SO(2) and O(2) Clebsch-Gordan series (i.e., reduction of their prod-
ucts).
week 10 Lie groups, algebras; O(2) symmetry sliced
(a) Group integrals. SO(3) character orthogonality.
(b) Continuous symmetry reduction for a nonlinear system is much harder than
discrete symmetry reduction of week 7. “Slicing” is a research level topic.
week 11 SU(2) and SO(3)
SU(2) ' SO(3) correspondence leads to the next rude awakening; our 3-dim-
ensional Euclidean space is not fundamental! All irreps of SO(3) are built from
2-dimensional complex vectors, or 1/2 spins.
week 12 Lorentz group; spin
(a) We now loose compactness: even though the SO(1, 3) Lorentz invariance
group of the Minkowski space symmetries is not compact, its Lie algebra still
closes, as for the compact SO(4).

birdtracks.eu/course3 weekEpilogue 184 2021-04-06

 GROUP THEORY - WEEK 17. AN OVERVIEW, AND THE EPILOGUE

(b) SO(4) ' SU(2) ⊗ SU(2) correspondence leads to the Minkowski 4-dimen-
sional space not being fundamental either - all irreps of the Lorentz group are
built from combinations of 2-dimensional complex vectors, or spinors.
(c) together with general relativity, this leads to replacement of the Minkowski
continuum by a 4-dimensional spacetime (or quantum) foam, a candidate theory
of quantum gravity.
week 13 Simple Lie algebras; SU(3)
The next profound shift:
So far all our group notions were based on tangible, spatial intuition: permuta-
tions, reflections, rotations. But now Lie groups take on a life of their own.
(a) The SO(3) theory of angular momenta generalizes to Killing-Cartan lattices,
and a fully abstract enumeration of all possible semi-simple compact Lie groups.
(b) SU(2) is promoted to an internal isospin symmetry, decoupled from our Eu-
clidean spatial intuition. Modern particle physics is born, with larger and larger
internal symmetry groups, tacked onto higher and higher dimensional continuum
spacetimes.
week 14 Flavor SU(3)
Gell-Mann–Okubo formula. The next triumph of particle physics is yet another
departure; observed baryons and mesons are built up from quarks, particles by
assumption unobservable in isolation.
week 15 Young tableaux
We have come full circle now: as a much simpler alternative to the Cartan-
Killing construction, irreps of the finite symmetric group Sn classify the irreps
of the continuous SU(n) symmetry multi-particle states.
week 16 Wigner 3- and 6-j coefficients
The goal of group theory is to predict measurable numbers, numbers independent
of any particular choice of coordinate. The full reducibility says that any such
number is built from 3- and 6-j coefficients: they are the total content of group
theory.

If you are reading this in preparation for a final exam, think of it as an opportunity to
rethink the key ideas of this branch of mathematics, take with you the few essential
insights that may serve you well in your career later on.

2021-04-06 185 birdtracks.eu/course3 weekEpilogue

 GROUP THEORY - WEEK 17. AN OVERVIEW, AND THE EPILOGUE

17.1 Student suggestions for improvements

Extracted from studentEvals17.pdf, studentEvals19.pdf.

improvements For this specific course, there are so many resources. It covers the best
parts of many books most probably, but from the student’s perspective every time
they open the middle of a new book they can find themselves looking at the very
first page at that book by following previous equations. Also the mathematical
aspect of the course can screen the physics side of the picture. As an experimen-
talist I got lost in the second half of the course. I found it easier to follow when
Dresselhaus or Tinkham book were being followed. Tinkham’s book was hard
for me at some point as well but these two books were easier to read for me than
other resources.
The course could be dramatically improved with a stronger emphasis on funda-
mentals of group structure and constructive examples throughout, accompanied
by a much more deliberate focus on a smaller set of topics. As an example of
a fundamental topic which I think has been inadequately covered: I am still un-
sure what distinguishes irreducible representations from any representation of a
group on an intuitive level. It is, as far as I can tell, something I should simply
know by now, but has not been adequately explained to me.
comments This course has been bewildering for the latter half of the semester. On
several occasions I have co-opted class time to have concepts from homework
and previous lectures explained carefully, which has been consistently unhelp-
ful. Overall, this course is by far the most stressful, confusing, and frustrating
course I have taken at Georgia Tech and on more than one occasion has given
me nightmares.
Instructor improvements The most needed improvement in the lectures is a stronger
focus on examples (especially those reinforcing fundamental concepts) which
reflect the Cvitanović expectations for the rest of the course.
to be CONTINUED

birdtracks.eu/course3 weekEpilogue 186 2021-04-06