Symmetry and Particle Physics: Michaelmas Term 2007

Symmetry and Particle Physics
Michaelmas Term 2007

Jan B. Gutowski
DAMTP, Centre for Mathematical Sciences
University of Cambridge
Wilberforce Road, Cambridge
CB3 0WA, UK
Email: J.B.Gutowski@damtp.cam.ac.uk
Contents
1. Introduction to Symmetry and Particles 5

1.1 Elementary and Composite Particles 5
1.2 Interactions 7
1.2.1 The Strong Interaction 7
1.2.2 Electromagnetic Interactions 8
1.2.3 The weak interaction 10
1.2.4 Typical Hadron Lifetimes 12
1.3 Conserved Quantum Numbers 12
2. Elementary Theory of Lie Groups and Lie Algebras 14

2.1 Differentiable Manifolds 14
2.2 Lie Groups 14
2.3 Compact and Connected Lie Groups 16
2.4 Tangent Vectors 17
2.5 Vector Fields and Commutators 19
2.6 Push-Forwards of Vector Fields 21
2.7 Left-Invariant Vector Fields 21
2.8 Lie Algebras 23
2.9 Matrix Lie Algebras 24
2.10 One Parameter Subgroups 27
2.11 Exponentiation 29
2.12 Exponentiation on matrix Lie groups 30
2.13 Integration on Lie Groups 31
2.14 Representations of Lie Groups 33
2.15 Representations of Lie Algebras 37
2.16 The Baker-Campbell-Hausdorff (BCH) Formula 38
2.17 The Killing Form and the Casimir Operator 45
3. SU(2) and Isospin 48

3.1 Lie Algebras of SO(3) and SU (2) 48
3.2 Relationship between SO(3) and SU (2) 49
3.3 Irreducible Representations of SU (2) 51
3.3.1 Examples of Low Dimensional Irreducible Representations 54
3.4 Tensor Product Representations 55
3.4.1 Examples of Tensor Product Decompositions 57
3.5 SU (2) weight diagrams 58
3.6 SU(2) in Particle Physics 59
3.6.1 Angular Momentum 59
3.6.2 Isospin Symmetry 59
–1–
3.6.3 Pauli’s Generalized Exclusion Principle and the Deuteron 61
3.6.4 Pion-Nucleon Scattering and Resonances 61
3.7 The semi-simplicity of (complexified) L(SU (n + 1)) 63
4. SU(3) and the Quark Model 65

4.1 Raising and Lowering Operators: The Weight Diagram 66
4.1.1 Triangular Weight Diagrams (I) 69
4.1.2 Triangular Weight Diagrams (II) 71
4.1.3 Hexagonal Weight Diagrams 73
4.1.4 Dimension of Irreducible Representations 77
4.1.5 The Complex Conjugate Representation 77
4.2 Some Low-Dimensional Irreducible Representations of L(SU (3)) 78
4.2.1 The Singlet 78
4.2.2 3-dimensional Representations 78
4.2.3 Eight-Dimensional Representations 81
4.3 Tensor Product Representations 81
4.3.1 3 ⊗ 3 decomposition. 82
4.3.2 3 ⊗ 3̄ decomposition 84
4.3.3 3 ⊗ 3 ⊗ 3 decomposition. 86
4.4 The Quark Model 89
4.4.1 Meson Multiplets 89
4.4.2 Baryon Multiplets 90
4.4.3 Quarks: Flavour and Colour 91
5. Spacetime Symmetry 94
5.1 The Lorentz Group 94
5.2 The Lorentz Group and SL(2, C) 97
5.3 The Lie Algebra L(SO(3, 1)) 99
5.4 Spinors and Invariant Tensors of SL(2, C) 101
5.4.1 Lorentz and SL(2, C) indices 102
5.4.2 The Lie algebra of SL(2, C) 104
5.5 The Poincaré Group 104
5.5.1 The Poincaré Algebra 105
5.5.2 Representations of the Poincaré Algebra 106
5.5.3 Massive Representations of the Poincaré Group: k µ = (m, 0, 0, 0) 111
5.5.4 Massless Representations of the Poincaré Group: k µ = (E, E, 0, 0) 111
6. Gauge Theories 114

6.1 Electromagnetism 114
6.2 Non-Abelian Gauge Theory 115
6.2.1 The Fundamental Covariant Derivative 115
6.2.2 Generic Covariant Derivative 116
6.3 Non-Abelian Yang-Mills Fields 118
–2–
6.3.1 The Yang-Mills Action 119
6.3.2 The Yang-Mills Equations 120
6.3.3 The Bianchi Identity 121
6.4 Yang-Mills Energy-Momentum 122
6.5 The QCD Lagrangian 123
–3–
Recommended Books
General Particle Physics Books
? D. H. Perkins, Introduction to High energy Physics, 4th ed., CUP (2000).
• B. R. Martin and G. Shaw, Particle Physics, 2nd ed., Wiley (1998).
Lie Algebra books written for Physicists
? H. Georgi, Lie Algebras in Particle Physics, Perseus Books (1999).
? J. Fuchs and C. Schweigert, Symmetries, Lie Algebras and Representations, 2nd ed.,
CUP (2003).
• H. F. Jones, Groups, Representations and Physics, 2nd ed., IOP Publishing (1998).
• J. Cornwell, Group Theory in Physics, (Volume 2), Academic Press (1986).
Pure Mathematics Lie Algebra books
• S. Helgason, Differential geometry, Lie groups, and symmetric spaces, 3rd ed., Aca-
demic Press (1978).
? H. Samelson, Notes on Lie Algebras, Springer (1990).
• W. Fulton and J. Harris, Representation Theory, A First Course, 3rd ed. Springer
(1991).
For an introduction to some aspects of Lie group differential geometry not covered in
this course:
• M. Nakahara, Geometry, Topology and Physics, 2nd ed., Institute of Physics Pub-
lishing (2003).
References for Spacetime Symmetry and Gauge Theory Applications
• T-P. Cheng and L-F. Li, Gauge Theory of Elementary Particle Physics, Oxford
(1984).
? S. Pokorski, Gauge Field Theories, 2nd ed., CUP (2000).
• S. Weinberg, The Quantum Theory of Fields, (Book 1), CUP (2005).
? J. Buchbinder and S. Kuzenko, Ideas and Methods of Supersymmetry and Supergrav-

ity, Or a Walk Through Superspace, 2nd ed., Institute of Physics Publishing (1998).
–4–
1. Introduction to Symmetry and Particles
Symmetry simplifies the description of physical phenomena. It plays a particularly impor-

tant role in particle physics, for without it there would be no clear understanding of the
relationships between particles. Historically, there has been an “explosion” in the number
of particles discovered in high energy experiments since the discovery that atoms are not
fundamental particles. Collisions in modern accelerators can produce cascades involving
hundreds of types of different particles: p, n, Π, K, Λ, Σ . . . etc.
The key mathematical framework for symmetry is group theory: symmetry transfor-
mations form groups under composition. Although the symmetries of a physical system are
not sufficient to fully describe its behaviour - for that one requires a complete dynamical
theory - it is possible to use symmetry to find useful constraints. For the physical systems
which we shall consider, these groups are smooth in the sense that their elements depend
smoothly on a finite number of parameters (called co-ordinates). These groups are Lie
groups, whose properties we will investigate in greater detail in the following lectures. We
will see that the important information needed to describe the properties of Lie groups is
encoded in “infinitessimal transformations”, which are close in some sense to the identity
transformation. The properties of these transformations, which are elements of the tangent
space of the Lie group, can be investigated using (relatively) straightforward linear algebra.
This simplifies the analysis considerably. We will make these rather vague statements more
precise in the next chapter.
Examples of symmetries include
i) Spacetime symmetries: these are described by the Poincaré group. This is only an
approximate symmetry, because it is broken in the presence of gravity. Gravity is the
weakest of all the interactions involving particles, and we will not consider it here.
ii) Internal symmetries of particles. These relate processes involving different types
of particles. For example, isospin relates u and d quarks. Conservation laws can be
found for particular types of interaction which constrain the possible outcomes. These
symmetries are also approximate; isospin is not exact because there is a (small) mass
difference between mu and md . Electromagnetic effects also break the symmetry.
iii) Gauge symmetries. These lead to specific types of dynamical theories describing
types of particles, and give rise to conserved charges. Gauge symmetries if present,
appear to be exact.
1.1 Elementary and Composite Particles

The fundamental particles are quarks, leptons and gauge particles.
The quarks are spin 1/2 fermions, and can be arranged into three families
Electric Charge (e)

2
u (0.3 GeV) c (1.6 GeV) t (175 GeV) 3
d (≈ 0.3 GeV) s (0.5 GeV) b (4.5 GeV) − 13
–5–
The quark labels u, d, s, c, t, b stand for up, down, strange, charmed, top and bottom.
The quarks carry a fractional electric charge. Each quark has three colour states. Quarks
are not seen as free particles, so their masses are ill-defined (the masses above are “effective”
masses, deduced from the masses of composite particles containing quarks).
The leptons are also spin 1/2 fermions and can be arranged into three families
Electric Charge (e)

e− (0.5 MeV) µ− (106 MeV) τ− (1.8 GeV) −1
νe (< 10 eV) νµ (< 0.16 MeV) ντ (< 18 MeV) 0
The leptons carry integral electric charge. The muon µ and taon τ are heavy unstable
versions of the electron e. Each flavour of charged lepton is paired with a neutral particle
ν, called a neutrino. The neutrinos are stable, and have a very small mass (which is taken
to vanish in the standard model).
All these particles have antiparticles with the same mass and opposite electric charge
(conventionally, for many particles, the antiparticles carry a bar above the symbol, e.g.
the antiparticle of u is ū). The antiparticles of the charged leptons are often denoted by
a change of − to +, so the positron e+ is the antiparticle of the electron e− etc. The
antineutrinos ν̄ differ from the neutrinos ν by a change in helicity (to be defined later...).
Hadrons are made from bound states of quarks (which are colour neutral singlets).
i) The baryons are formed from bound states of three quarks qqq; antibaryons are formed
from bound states of three antiquarks q̄ q̄ q̄
For example, the nucleons are given by
(
p = uud : 938 M ev
n = udd : 940 M ev
ii) Mesons are formed from bound states of a quark and an antiquark q q̄.
For example, the pions are given by



 π + = ud¯ : 140 M ev
π − = dū : 140 M ev
 π 0 = uū, dd¯ superposition

: 135 M ev
Other particles are made from heavy quarks; such as the strange particles K + = us̄
with mass 494 Mev , Λ = uds with mass 1115 Mev, and Charmonium ψ = cc̄ with mass
3.1 Gev.
The gauge particles mediate forces between the hadrons and leptons. They are bosons,
with integral spin.
–6–
Mass (GeV) Interaction
γ (photon) 0 Electromagnetic
W+ 80 Weak
W− 80 Weak
Z0 91 Weak
g (gluon) 0 Strong
The gluons are responsible for interquark forces which bind quarks together in nucleons.
It is conjectured that a spin 2 gauge boson called the graviton is the mediating particle
for gravitational forces, though detecting this is extremely difficult, due to the weakness of
gravitational forces compared to other interactions.
1.2 Interactions
There are three types of interaction which are of importance in particle physics: the strong,
electromagnetic and weak interactions.
1.2.1 The Strong Interaction

The strong interaction is the strongest interaction.
• Responsible for binding of quarks to form hadrons (electromagnetic effects are much
weaker)
• Dominant in scattering processes involving just hadrons. For example, pp → pp is

an elastic process at low energy; whereas pp −→ ppπ + π − is an inelastic process at
higher energy.
• Responsible for binding forces between nucleons p and n, and hence for all nuclear
structure.
Properties of the Strong Interaction:
i) The strong interaction preserves quark flavours, although q q̄ pairs can be produced
and destroyed provided q, q̄ are the same flavour.
An example of this is:
π
+ u u +
d s K
d s +
p u u Σ
u u
–7–
The Σ+ and K + particles decay, but not via the strong interaction, because of con-
servation of strange quarks.
ii) Basic strong forces are “flavour blind”. For example, the interquark force between
q q̄ bound states in the ψ = cc̄ (charmonium) and Υ = bb̄ (bottomonium) mesons are
well-approximated by the potential
α
V ∼ + βr (1.1)
r
and the differences in energy levels for these mesons is approximately the same.
The binding energy differences can be attributed to the mass difference of the b and
c quarks.
iii) Physics is unchanged if all particles are replaced by antiparticles.
The dynamical theory governing the strong interactions is Quantum Chromodynamics

(QCD), which is a gauge theory of quarks and gluons. This is in good agreement with
experiment, however non-perturbative calculations are difficult.
1.2.2 Electromagnetic Interactions
The electromagnetic interactions are weaker than the strong interactions. They occur in
the interactions between electrically charged particles, such as charged leptons, mediated
by photons.
The simplest electromagnetic process consists of the absorption or emission of a photon
by an electron:
This process cannot occur for a free electron, as it would violate conservation of 4-
momentum, rather it involves electrons in atoms, and the 4-momentum of the entire atom
and photon are conserved.
Other examples of electromagnetic interactions are electron scattering mediated by
photon exchange
–8–
e
and there are also smaller contributions to this process from multi-photon exchanges.
Electron-positron interactions are also mediated by electromagnetic interactions
e+ e
+
e− e−
e
+ e+
γ
+
e− e−
Electron-positron annihilation can also produce particles such as charmonium or bot-

tomonium
+
e H
A
γ
ψ D
R
O
N
e− S
–9–
The dynamic theory governing electromagnetic interactions is Quantum Electrody-
namics (QED), which is very well tested experimentally.
Neutrinos have no electromagnetic or strong interactions.
1.2.3 The weak interaction
The weak interaction is considerably weaker than both the strong and electromagnetic
interactions, they are mediated by the charged and neutral vector bosons W ± and Z 0
which are very massive and produce only short range interactions. Weak interactions
occur between all quarks and leptons, however they are in general negligable when there
are strong or electromagnetic interactions present. Only in the absence of strong and
electromagnetic interactions is the weak interaction noticable.
Unlike the strong and electromagnetic interactions, weak interactions can involve neu-
trinos. Weak interactions, unlike strong interactions, can also produce flavour change in
quarks and neutrinos.
The gauge bosons W ± carry electric charge and they can change the flavour of quarks.
Examples of W -boson mediated weak interactions are n −→ p + e− + ν̄e :
νe
e−
−
W
d u
n d d p
u u
and µ− −→ e− + ν̄e + νµ :
– 10 –
νµ
µ−
−
W
νe
e−
and νµ + n → µ− + p
νµ µ−
+
W
d u
n d d p
u u
The flavour changes within one family are dominant; e.g.
e− ↔ νe , µ− ↔ νµ
u ↔ d, c↔s (1.2)
whereas changes between families, like u ↔ s and u ↔ b are “Cabibbo suppressed”.

The neutral Z 0 , like the photon, does not change quark flavour; though unlike the
photon, it couples to neutrinos. An example of a Z 0 mediated scattering process is ν̄µ e−
scattering:
– 11 –
νµ νµ
0
Z
e− e−
In any process in which a photon is exchanged, it is possible to have a Z 0 boson

exchange. At low energies, the electromagnetic interaction dominates; however at high
energies and momenta, the electromagnetic and weak interactions become comparable.
The unified theory of electromagnetic and weak interactions is Weinberg-Salam theory.
1.2.4 Typical Hadron Lifetimes

Typical hadron lifetimes (valid for most decays) via the three interactions are summarized
below:
Interaction Lifetime (s)

Strong 10−22 − 10−24
Electromagnetic 10−16 − 10−21
Weak 10−7 − 10−13
with the notable exceptional case being weak neutron decay, which has average lifetime
of 103 s.
1.3 Conserved Quantum Numbers

Given a configuration of particles containing particle P , we define N (P ) to denote the
number of P -particles in the configuration. We define various quantum numbers associated
with leptons and hadrons.
Definition 1. There are three lepton numbers. The electron, muon and tauon numbers
are given by
Le = N (e− ) − N (e+ ) + N (νe ) − N (ν̄e )

Lµ = N (µ− ) − N (µ+ ) + N (νµ ) − N (ν̄µ )
Lτ = N (τ − ) − N (τ + ) + N (ντ ) − N (ν̄τ ) (1.3)
In electromagnetic interactions, where there are no neutrinos involved, conservation

of L is equivalent to the statement that leptons and anti-leptons can only be created or
annihilated in pairs. For weak interactions there are more possibilities, so for example, an
– 12 –
election e− and anti-neutrino ν̄e could be created. Lepton numbers are conserved in all
interactions.
There are also various quantum numbers associated with baryons.
Definition 2. The four quark numbers S, C, B̃ and T corresponding to strangeness,

charm, bottom and top are defined by
S = −(N (s) − N (s̄))

C = (N (c) − N (c̄))
B̃ = −(N (b) − N (b̄))
T = (N (t) − N (t̄)) (1.4)
These quark quantum numbers, together with N (u) − N (ū) and N (d) − N (d), ¯ are
conserved in strong and electromagnetic interactions, because in these interactions quarks
and antiquarks are only created or annihilated in pairs. The quark quantum numbers are
not conserved in weak interactions, because it is possible for quark flavours to change.
Definition 3. The baryon number B is defined by
1
B = (N (q) − N (q̄)) (1.5)
3
where N (q) and N (q̄) are the total number of quarks and antiquarks. Baryons therefore
have B = 1 and antibaryons have B = −1; mesons have B = 0. B is conserved in all
interactions.
Note that one can write
1 ¯ + C + T − S − B̃)
B = (N (u) − N (ū) + N (d) − N (d) (1.6)
3
Definition 4. The quantum number Q is the total electric charge. Q is conserved in all
interactions
In the absence of charged leptons, such as in strong interaction processes, one can write
2 1 ¯ − S − B̃)
Q = (N (u) − N (ū) + C + T ) − (N (d) − N (d) (1.7)
3 3
Hence, for strong interactions, the four quark quantum numbers S, C, B̃, T together
¯
with Q and B are sufficient to determine N (u) − N (ū) and N (d) − N (d).
– 13 –
2. Elementary Theory of Lie Groups and Lie Algebras
2.1 Differentiable Manifolds

Definition 5. A n-dimensional real smooth manifold M is a (Hausdorff topological) space
which is equipped with a set of open sets U α such that
1) For each p ∈ M , there is some U α with p ∈ U α
2) For each U α , there is an invertible homeomorphism xα : U α → Rn onto an open
subset of Rn such that if U α ∩ U β 6= ∅ then the map
xβ ◦ xα −1 : xα (U α ∩ U β ) → xβ (U α ∩ U β ) (2.1)
is smooth (infinitely differentiable) as a function on Rn .

The open sets U α together with the maps xα are called charts, the set of all charts
is called an atlas. The maps xα are local co-ordinates on M defined on the U α , and have
components xiα for i = 1, . . . , n. So a smooth manifold looks locally like a portion of Rn .
A n-dimensional complex manifold is defined in an exactly analogous manner to a real
manifold, with Rn replaced by Cn throughout.
Definition 6. Suppose M is a m-dimensional smooth manifold, and N is a n-dimensional

smooth manifold, with charts (U α , xα ), (W A , yA ) respectively. Then the Cartesian product
X = M ×N is a m+n-dimensional smooth manifold, equipped with the standard Cartesian
product topology.
The charts are V α,A = U α × W A with corresponding local co-ordinates
zα,A = xα × yA : U α × W A → Rm+n (2.2)
Definition 7. Suppose M is a m-dimensional smooth manifold, and N is a n-dimensional

smooth manifold, with charts (U α , xα ), (W A , yA ) respectively. Then a function f : M → N
is smooth if for every U α and W A such that f (U α ) ∩ W A 6= ∅, the map
yA ◦ f ◦ x−1 α A
α : xα (U ) → yA (W ) (2.3)
is smooth as a function Rm → Rn .
Definition 8. A smooth curve on a manifold M is a map γ : (a, b) → M where (a, b) is

some open interval in R such that if U is a chart with local co-ordinates x then the map
x ◦ γ : (a, b) → Rn (2.4)
may be differentiated arbitrarily often.
2.2 Lie Groups

Definition 9. A group G is a set equipped with a map • : G × G → G, called group
multiplication, given by (g1 , g2 ) → g1 • g2 ∈ G for g1 , g2 ∈ G. Group multiplication satisfies
– 14 –
i) There exists e ∈ G such that g • e = e • g = g for all g ∈ G. e is called an identity
element.
ii) For every g ∈ G there exists an inverse g −1 ∈ G such that g • g −1 = g −1 • g = e.
iii) For all g1 , g2 , g3 ∈ G; g1 •(g2 •g3 ) = (g1 •g2 )•g3 , so group multiplication is associative.
It is elementary to see that the identity e is unique, and g has a unique inverse g −1 .
Definition 10. A Lie group G is a smooth differentiable manifold which is also a group,
where the group multiplication • has the following properties
i) The map • : G × G → G given by (g1 , g2 ) → g1 • g2 is a smooth map.
ii) The inverse map G → G given by g → g −1 is a smooth map
Henceforth, we shall drop the • for group multiplication and just write g1 • g2 = g1 g2 .
Examples:
Many of the most physically interesting Lie groups are matrix Lie groups in various
dimensions. These are subgroups of GL(n, R) (or GL(n, C)), the n × n real (or complex)
invertible matrices. Group multiplication and inversion are standard matrix multiplication
and inversion.
Suppose that G is a matrix Lie group of dimension k. Let the local co-ordinates
be xi for i = 1, . . . , k. Then g ∈ G is described by its matrix components g AB (xi ) for
A, B = 1, . . . , n. The g AB are smooth functions of the co-ordinates xi . Examples of matrix
Lie groups are (here F = R or F = C):
i) GL(n, F), the invertible n × n matrices over F. The co-ordinates of GL(n, F) are the
n2 real (or complex) components of the matrices.
ii) SL(n, F) = {M ∈ GL(n, F) : det M = 1}
iii) O(n) = {M ∈ GL(n, R) : M M T = In }
iv) U (n) = {M ∈ GL(n, C) : M M † = In }, where † is the hermitian transpose.
v) SO(n) = {M ∈ GL(n, R) : M M T = In and det M = 1}
vi) SU (n) = {M ∈ GL(n, C) : M M † = In and det M = 1}. SU (2) and SU (3) play a
particularly important role in the standard model of particle physics.
vii) SO(1, n − 1) = {M ∈ GL(n, R) : M T ηM = η and det M = 1}

where η = diag (1, −1, −1, · · · − 1) is the n-dimensional Minkowski metric.
There are other examples, some of which we will examine in more detail later. It can
be shown that any closed subgroup H of GL(n, F) (i.e. any subgroup which contains all
its accumulation points) is a Lie group.
– 15 –
Some of these groups are related to each other by group isomorphism; a particularly
simple example is SO(2) ∼= U (1). Elements of U (1) consist of unit-modulus complex
iθ
numbers e for θ ∈ R under multiplication, whereas SO(2) consists of matrices
!
cos θ − sin θ
R(θ) = (2.5)
sin θ cos θ
which satisfy R(θ+φ) = R(θ)R(φ). The map T : U (1) → SO(2) given by T (eiθ ) = R(θ)
is a group isomorphism.
2.3 Compact and Connected Lie Groups
A lie group G is compact if G is compact as a manifold. Recall that a subset of U ⊂ Rn

is compact iff it is closed and bounded, or equivalently iff every sequence un ∈ U has a
subsequence which converges to some u ∈ U .
It is straightforward to see that SU (n) is compact, for if we denote the rows of M ∈
†
SU (n) by RA then RA RB = δAB . Hence the components M AB are all bounded |M AB | ≤ 1.
So it follows that if Mn is a sequence of points in SU (n), then by repeated application of the
Bolzano-Weierstrass theorem, there is a subsequence Mnr which converges to some matrix
N . Moreover as the constraints det Mnr = 1 and Mnr Mn†r = 1 are smooth functions of the
matrix components, one must also have det N = 1 and N N † = 1 in the limit as r → ∞,
i.e. N ∈ SU (n).
[There is a subtlety concerning convergence which we have glossed over, namely how
one actually defines convergence. We assume the existence of some matrix norm (for
example ||M ||sup = max(|M AB |)) with respect to which convergence is defined. As all
(finite-dimensional) matrix norms are equivalent, convergence with respect to one matrix
norm ensures convergence with respect to any norm].
In contrast, the Lorentz group SO(1, n − 1) is not compact. For example, consider for
simplicity SO(1, 1). One can define a sequence of elements Mn ∈ SO(1, 1) by
!
cosh n sinh n
Mn = (2.6)
sinh n cosh n
As the components of Mn are unbounded, it follows that Mn cannot have a convergent

subsequence. Observe that as SO(1, n−1) is a Lie subgroup of both SL(n, R) and GL(n, R)
it must follow that SL(n, R) and GL(n, R) are also non-compact.
A Lie group G is said to be connected if any two points in the group can be linked
together by a continuous curve in G.
O(n) is not connected. To see this, observe that if M ∈ O(n) then M M T = 1 and on
taking the determinant this implies det M = ±1. Now take M ∈ O(n) with det M = −1,
so if O(n) is connected, there is a continuous curve γ : [0, 1] → O(n) with γ(0) = I and
γ(1) = M . We can then compute det γ(t) which must be a continuous real function of t
such that det γ(t) ∈ {−1, 1} for all t ∈ [0, 1] and det γ(0) = 1, det γ(1) = −1. This is not
possible.
– 16 –
We shall say that two points in G are connected if they can be linked with a continuous
curve. This defines an equivalence relation on G, and hence partitions G into equivalence
classes of connected points; the equivalence class of g ∈ G is called the connected component
of g. The equivalence class of points of O(n) connected to I is SO(n), which is connected.
2.4 Tangent Vectors

Suppose that U is an open subset of a manifold M , and that the curve γ passes through
some p ∈ U with γ(t0 ) = p. Then the curve defines a tangent vector at p, denoted by γ̇p ,
which maps smooth real functions f : U → R to R according to
d
γ̇p : f →
(f ◦ γ(t)) t=t0 (2.7)
dt
The components of the tangent vector are
d
γ̇pm = ((x ◦ γ)m ) t=t0 = γ̇p (xm )

(2.8)
dt
Note that one can write (using the chain rule)
d
γ̇p (f ) = (f ◦ γ(t)) t=t0
dt
d
(f ◦ x−1 ◦ x ◦ γ(t)) t=t0

=
dt
n
X ∂ d
= i
(f ◦ x−1 )|x(p) ( (x ◦ γ)i )t=t0
∂x dt
i=1
n
X ∂
= (f ◦ x−1 )|x(p) γ̇pi (2.9)
∂xi
i=1
Proposition 1. The set of all tangent vectors at p forms a n-dimensional vector space
(where n = dim M ), denoted by Tp (M ).
Proof
Suppose that p lies in the chart U with local co-ordinates x. Suppose also that V ,
W ∈ Tp (M ) are tangent vectors at p corresponding to the curves γ, σ, where without
loss of generality we can take γ : (a, b) → M , σ : (a, b) → M with a < t0 < b and
γ(t0 ) = σ(t0 ) = p
Take a, b ∈ R. Consider the curve ρ̂ in Rn defined by
ρ̂(t) = a(x ◦ γ)(t) + b(x ◦ σ)(t) − (a + b − 1)x(p) (2.10)

where scalar multiplication and vector addition are the standard operations in Rn .
Note that ρ̂(t0 ) = x(p).
Then define the curve ρ on U by ρ = x−1 ◦ ρ̂, so that ρ(t0 ) = p.
If f is a smooth function on U then by (2.9) it follows that
n
X ∂ d
ρ̇p (f ) = i
(f ◦ x−1 )|x(p) ( (x ◦ ρ)i )t=t0
∂x dt
i=1
– 17 –
n
X ∂ d
= i
(f ◦ x−1 )|x(p) ( ρ̂i (t))t=t0
∂x dt
i=1
n
X ∂ d
=a (f ◦ x−1 )|x(p) ( (x ◦ γ)i )t=t0
∂xi dt
i=1
n
X ∂ d
+b i
(f ◦ x−1 )|x(p) ( (x ◦ γ)i )t=t0
∂x dt
i=1
= aγ̇p (f ) + bσ̇p (f ) (2.11)
So it follows that aγ̇p + bσ̇p is the tangent vector to ρ at p.

In order to compute the dimension of the vector space it suffices to compute a basis.
To do this, define n curves ρ(i) for i = 1, . . . , n passing through p by
(x ◦ ρ(i))(t)j = (x(p))j + tδij (2.12)
Using (2.9) it is straightforward to compute the tangent vectors to the curves ρ(i) at
p;
∂
ρ̇(i)p (f ) = (f ◦ x−1 )|x(p) (2.13)
∂xi
and hence, if γ is a curve passing through p then (2.9) implies that
n
X
γ̇p (f ) = ρ̇(i)p (f )γ̇pi (2.14)
i=1
Pn i
and hence it follows that γ̇p = i=1 γ̇p ρ̇(i)p .
Hence the tangent vectors to the curves
ρ(i) at p span Tp (M ).
Given the expression (2.13), it is conventional to write the tangent vectors to the curves
ρ(i) at p as
∂
ρ̇(i)p = (2.15)
∂xi p
Lemma 1. Suppose that M1 , M2 are smooth manifolds of dimension n1 , n2 respectively.
Let M = M1 × M2 be the Cartesian product manifold and suppose p = (p1 , p2 ) ∈ M . Then
L
Tp (M ) = Tp1 (M1 ) Tp2 (M2 ).
Proof
Suppose Vp ∈ Tp (M ). Then V is the tangent vector to a smooth curve γ(t), with
γ(t0 ) = p. Write γ(t) = (γ1 (t), γ2 (t)); γi (t) is then a smooth curve in Mi and γi (t0 ) = pi
for i = 1, 2.
Let f be a smooth function f : M → R. Suppose that xa are local co-ordinates on M1
for a = 1, . . . , n1 and y m are local co-ordinates on M2 for m = 1, . . . , n2 corresponding to
charts U1 ⊂ M1 and U2 ⊂ M2 .
Then one has n1 + n2 local co-ordinates z α on M where if q = (q1 , q2 ) ∈ U1 × U2 ,
z(q1 , q2 ) = x1 (q1 ), . . . , xn1 (q1 ), y 1 (q2 ), . . . , y n2 (q2 )

(2.16)
– 18 –
Note that f1 (q1 ) = f (q1 , q2 ) is a smooth function of q1 when q2 is fixed, and f2 (q2 ) =
f (q1 , q2 ) is a smooth function of q2 when q1 is fixed.
Then using the chain rule
nX
1 +n2
∂ d
Vp f = α
(f ◦ z −1 )|z(p) ((z ◦ γ)α (t))|t=t0
∂z dt
α=1
n1
X ∂ d
= (f1 ◦ x−1 )|(x(p1 ),y(q2 )) ((x ◦ γ1 )a (t))|t=t0
∂xa dt
a=1
n2
X ∂ d
+ j
(f2 ◦ y −1 )|(x(p1 ),y(q2 )) ((y ◦ γ2 )j (t))|t=t0
∂y dt
j=1
= (V (1)p + V (2)p )f (2.17)
where V (1)p is the tangent vector to γ1 at p, and V (2)p is the tangent vector to γ2 at
p. Hence Vp = V (1)p + V (2)p . Conversely, given two smooth curves γ1 (t), γ2 (t) in M1 , M2
passing through p1 and p2 at t = t0 , with associated tangent vectors V (1)p and V (2)p , one
can construct the smooth curve γ(t) = (γ1 (t), γ2 (t)) in M passing through p = (p1 , p2 ) at
t = t0 . Then (2.17) shows that V (1)p + V (2)p can be written as Vp ∈ Tp (M ).
2.5 Vector Fields and Commutators

The tangent space of M , T (M ) consists of the union
[
T (M ) = Tp (M ) (2.18)
p∈M
A vector field V on M is a map V : M → T (M ) such that V (p) = Vp ∈ Tp (M ).

Note that T (M ) is a vector space with addition and scalar multiplication defined by
(X + Y )(f ) = X(f ) + Y (f ) (2.19)

where X, Y ∈ T (M ) and f : M → R is smooth, and
(αX)(f ) = αX(f ) (2.20)

for constant α ∈ R.
At a point p ∈ M , one can decompose Vp into its components with respect to a
particular chart as
∂
Vp = Vpi (2.21)
∂xi p
It is conventional to write
∂
V = V i( ) (2.22)
∂xi
where V i = (V ◦ x−1 )(xi ) are functions Rn → R and ( ∂x
∂
i ) is a locally defined vector
field which satisfies
– 19 –
∂
( )xj = δij (2.23)
∂xi
∂
It follows that T (M ) is n-dimensional with a local basis given by the ( ∂x i ). The vector
i n
field is called smooth if the functions V are smooth functions on R .
Suppose now that f is a smooth function on M and that V , W are vector fields on M .
Then note that V f can be regarded as a function M → R defined by
(V f )(p) = Vp f (2.24)
Hence one can act on V f with Wp at some p ∈ M to find
∂
Wp (V f ) = Wpi (V f )
∂xi p
∂ ∂
= Wpi (V j j (f ◦ x−1 ))|x(p)

∂xi p ∂x
∂V j ∂
= Wpi i |x(p) ( j (f ◦ x−1 ))|x(p)
∂x ∂x
∂2
+ Wp Vp ( i j (f ◦ x−1 ))|x(p)
i j
(2.25)
∂x ∂x
The fact that there are second order derivatives acting on f means that we cannot
write Wp (V f ) = Zp f for some vector field Z.
However, these second order derivatives can be removed by taking the difference
∂V j i ∂W
j ∂
Wp (V f ) − Vp (W f ) = Wpi i
|x(p) − Vp i
|x(p) ( j (f ◦ x−1 ))|x(p) (2.26)
∂x ∂x ∂x
which can be written as Zp f where Z is a vector field called the commutator or alter-
natively the Lie bracket of W and V which we denote by [W, V ] with components
∂V j i ∂W
j
[W, V ]j = W i − V (2.27)
∂xi ∂xi
Exercise:
Prove that the Lie bracket satisfies
i) Skew-symmetry: [X, Y ] = −[Y, X] for all smooth vector fields X, Y ∈ T (M ).
ii) Linearity: [αX + βY, Z] = α[X, Z] + β[Y, Z] for α, β constants and X, Y , Z ∈ T (M ).
iii) The Jacobi identity: [[X, Y ], Z] + [[Z, X], Y ] + [[Y, Z], X] = 0 for all X, Y , Z ∈ T (M ).
Definition 11. Let V be a smooth vector field on M . An integral curve σ(t) of V is a

curve whose tangent vector at σ(t) is V |σ(t) , i.e.
d i i
(σ (t)) = Vσ(t) (2.28)
dt
where in a slight abuse of notation, σ i (t) = (x ◦ σ)i (t) for some local co-ordinates x.
Such a curve is guaranteed to exist and to be unique (at least locally, given an initial
condition), by the standard existence and uniqueness theorems for ODE’s.
– 20 –
2.6 Push-Forwards of Vector Fields
Suppose that M , N are two smooth manifolds and f : M → N is a smooth map. Then
there is an induced map
f∗ : T (M ) → T (N ) (2.29)
which maps the tangent vector of a curve γ passing through a point p ∈ M to the
tangent vector of the curve f ◦ γ passing through f (p) ∈ N .
In particular, for each smooth function h on N , and if γ is a curve passing through
p ∈ M with γ(0) = p, and if Vp ∈ Tp (M ) is the tangent vector of γ at p then f∗ Vp ∈ Tf (p) (N )
is given by
d
(f∗ Vp )h = h ◦ (f ◦ γ) t=0
dt
= Vp (h ◦ f ) (2.30)
Hence it is clear that the push-forward map f∗ is linear on the space of tangent vectors.
Note that if M , N and Q are manifolds, and f : M → N , g : N → Q are smooth
functions then if h : Q → R is smooth and p ∈ M ,

(g ◦ f )∗ Vp (h) = Vp (h ◦ (g ◦ f ))
= Vp ((h ◦ g) ◦ f )
= (f∗ Vp )(h ◦ g)

= g∗ (f∗ Vp ) (h) (2.31)
and hence
(g ◦ f )∗ = g∗ ◦ f∗ (2.32)
2.7 Left-Invariant Vector Fields

Suppose that G is a Lie group and a, g ∈ G. Define the operation of left-translation
La : G → G by
La g = ag (2.33)
La defined in this fashion is a differentiable invertible map from G onto G. Hence, one
can construct the push-forward La∗ of vector fields on G with respect to La .
Definition 12. A vector field X ∈ T (G) is said to be left-invariant if
La∗ (X|g ) = X|ag (2.34)
Given v ∈ Te (G) one can construct a unique left-invariant vector field X(v) ∈ T (G)
with the property that X(v)e = v using the push-forward by
– 21 –
X(v)|g = Lg∗ v (2.35)
To see that X(v) is left-invariant, note that
X(v)|ag = L(ag)∗ v (2.36)

but from (2.32) it follows that as Lag = La ◦ Lg we must have
L(ag)∗ v = (La ◦ Lg )∗ v = La∗ (Lg∗ v) = La∗ X(v)g (2.37)

so X(v) is left-invariant. Hence there is a 1-1 correspondence between elements of the
tangent space at e and the set of left-invariant vector fields.
Proposition 2. The set of left-invariant vector fields is closed under the Lie bracket, i.e.
if X, Y ∈ T (G) are left-invariant then so is [X, Y ].
Proof
Suppose that f : G → R is a smooth function. Then

La∗ [X, Y ]g f = [X, Y ]g (f ◦ La )
= Xg (Y (f ◦ La )) − Yg (X(f ◦ La )) (2.38)
But as X is left-invariant, La∗ Xg = Xag so
Xag f = (La∗ Xg )f = Xg (f ◦ La ) (2.39)

so replacing f with Y f in the above we find
Xg ((Y f ) ◦ La ) = Xag (Y f ) (2.40)

Moreover, as Y is left-invariant, it is straightforward to show that

Y (f ◦ La ) g = Yg (f ◦ La )
= (La∗ Yg )f
= Yag (f )
= (Y f )(ag)
= (Y f ) ◦ La )g (2.41)
so Y (f ◦ La ) = (Y f ) ◦ La
Hence
Xg (Y (f ◦ La )) − Yg (X(f ◦ La )) = Xg ((Y f ) ◦ La ) − Yg ((Xf ) ◦ La )

= Xag (Y f ) − Yag (Xf )
= [X, Y ]ag f (2.42)
So La∗ [X, Y ]g = [X, Y ]ag , hence [X, Y ] is left-invariant.
– 22 –
2.8 Lie Algebras
Definition 13. Suppose that G is a Lie group. Then the Lie algebra L(G) associated
with G is Te (G), the tangent space of G at the origin, together with a Lie bracket [ , ] :
L(G) × L(G) → L(G) which is defined by
[v, w] = [L∗ v, L∗ w]e (2.43)
for v, w ∈ Te (G), L∗ v and L∗ w denote the smooth vector fields on G obtained by pushing
forward v and w by left-multiplication (i.e. L∗ v|g = Lg∗ v), and [L∗ v, L∗ w] is the standard
vector field commutator. As the Lie bracket on L(G) is obtained from the commutator of
vector fields, it follows that the Lie bracket is
i) Skew-symmetric: [v, w] = −[w, v] for all v, w ∈ L(G).
ii) Linear: [αv1 + βv2 , w] = α[v1 , w] + β[v2 , w] for α, β constants and v1 , v2 , w ∈ L(G),
iii) and satisfies the Jacobi identity: [[v, w], z] + [[z, v], w] + [[w, z], v] = 0 for all v, w,
z ∈ L(G).
where (ii) follows because the push forward map is linear on the space of vector fields,
and (iii) follows because as a consequence of Proposition 2, Lg∗ [v, w] = [L∗ v, L∗ w]g .
More generically, one can also define a Lie algebra to be a vector space g equipped
with a map [ , ] : g × g → g satisfying (i), (ii), (iii) above.
Definition 14. Suppose that {Ti : i = 1, . . . , n} is a basis for L(G). Then the Ti are called
generators of the Lie algebra. As [Ti , Tj ] ∈ L(G) it follows that there are constants cij k
such that
[Ti , Tj ] = cij k Tk (2.44)
The constants cij k are called the structure constants of the Lie algebra.
The structure constants are constrained by the antisymmetry of the Lie bracket to be
antisymmetric in the first two indices;
cij k = −cji k (2.45)
Also, the Jacobi identity implies
[[Ti , Tj ], Tk ] + [[Tj , Tk ], Ti ] + [[Tk , Ti ], Tj ] = 0 (2.46)
which gives an additional constraint on the structure constants
cij ` c`k m + cjk ` c`i m + cki ` c`j m = 0 (2.47)
– 23 –
2.9 Matrix Lie Algebras
The Lie algebras of matrix Lie groups are of particular interest. Suppose that G is a matrix
Lie group, and V ∈ T (G) is a smooth vector field. Let f be a smooth function of the matrix
components g AB . Then if h ∈ G,
∂f
Vh f = Vhm
∂xm
∂g AB ∂f
= Vhm
∂xm ∂g AB
∂f
= VhAB AB (2.48)
∂g
where
∂g AB
VhAB = Vhm (2.49)
∂xm
defines a tangent matrix associated with the components Vhm of V at h. Each vector
field has a corresponding tangent matrix, and it will often be most convenient to deal with
these matrices instead of more abstract vector fields as differential operators.
In particular, if γ(t) is some curve in G with tangent vector V then
d
Vf = (f ◦ γ(t))
dt
dg AB ∂f
= (2.50)
dt ∂g AB
dg AB
hence the tangent vector to the curve corresponds to the matrix dt . We will fre-
quently denote the identity element of a matrix Lie group by e = I
Examples of matrix Lie algebras are
• a) GL(n, R): the co-ordinates of GL(n, R) are the n2 components of the matrices, so
GL(n, R) is n2 -dimensional. There is no restriction on tangent matrices to curves in
GL(n, R), the space of tangent vectors is Mn×n (R), the set of n × n real matrices.
• b) GL(n, C): the co-ordinates of GL(n, C) are the n2 components of the matrices, so
GL(n, C) is 2n2 -dimensional when viewed as a real manifold. There is no restriction
on tangent matrices to curves in GL(n, C), the space of tangent vectors is Mn×n (C),
the set of n × n complex matrices.
• c) SL(n, R): Suppose that M (t) is a curve in SL(n, R) with M (0) = I. To compute
the restrictions on the tangent vectors to the curve note that
det M (t) = 1 (2.51)
so, on differentiating with respect to t,
– 24 –
dM (t)
Tr M −1 (t) =0 (2.52)
dT
and so if we denote the tangent vector at the identity to be m = dMdt(t) |t=0 then
Tr m = 0. The tangent vectors correspond to traceless matrices. Hence SL(n, R) is
n2 − 1 dimensional.
• d) O(n): suppose that M (t) is a curve in O(n) with M (0) = I. To compute the
restrictions on the tangent vectors to the curve note that
M (t)M (t)T = 1 (2.53)
so, on differentiating with respect to t,
dM (t) dM (t)T
M (t)T + M (t) =0 (2.54)
dt dt
and hence if m = dMdt(t) |t=0 then m+mT = 0. The tangents to the curve at the identity
correspond to antisymmetric matrices. There are 12 n(n − 1) linearly independent
antisymmetric matrices, hence O(n) is 12 n(n − 1)-dimensional.
Note that the Lie algebra of SO(2) is 1-dimensional and is spanned by
!
0 1
T1 = (2.55)
−1 0
As [T1 , T1 ] = 0 it follows trivially that the Lie bracket vanishes
• f) SO(n) the group of n × n real matrices such that if M ∈ SO(n) then M M T = 1

and det M = 1. By the reasoning in (c) and (e) it follows that the tangent matrices
at the identity are skew-symmetric matrices (these are automatically traceless).
As the skew symmetric matrices are automatically traceless, it follows that the Lie
algebra L(SO(n)) of SO(n) is identical to the Lie algebra of O(n). If v, w ∈ L(SO(n))
are skew-symmetric matrices it is straightforward to show that the matrix commu-
tator [v, w] is also skew symmetric, as [v, w]T = (vw − wv)T = wT v T − v T wT =
[w, v] = −[v, w]. Hence [v, w] ∈ L(SO(n)) as expected. We will show that vector
field commutation can be reduced to tangent matrix commutation for matrix Lie
groups.
Exercise
Show that the tangent vectors of U (n) at I consist of antihermitian matrices, and the
tangent vectors of SU (n) at I are traceless antihermitian matrices.
– 25 –
Proposition 3. Suppose that G is a matrix Lie group and V is a smooth vector field on
G and a ∈ G is fixed. If V̂ denotes the tangent matrix associated with V , then the tangent
matrix associated with the push-forward La∗ V is aV̂ .
Proof
Suppose h ∈ G, and f : G → R is a smooth function on G. Consider the tangent
vector La∗ Vh defined at ah
Then
(La∗ Vh )f = Vh (f ◦ La )
= Vh f˜ (2.56)
where f˜(g) = f (ag).

So
∂ f˜
(La∗ Vh )f = VhAB |h
∂g AB
∂f ∂
= VhAB P Q |ah AB ((ag)P Q )
∂g ∂g
∂f
= VhAB P B |ah aP A
∂g
AB ∂f
= (aV̂ ) |ah (2.57)
∂g AB
So it follows that the tangent matrix associated with La∗ Vh is aV̂ .

Using this result, it is possible to re-interpret the commutator of two left-invariant
vector fields in terms of the matrix commutators of their associated matrices.
Proposition 4. Suppose that G is a matrix Lie group and that v, w ∈ Te (G) and V , W
are the left-invariant vector fields defined by Vg = Lg∗ v, Wg = Lg∗ w. Then the matrix
associated with [V, W ]e is the matrix commutator of [v̂, ŵ] where v̂ and ŵ are the matrices
associated with v and w.
Proof
By definition, the matrix associated with [V, W ] is
∂g AB
[V, W ]AB = [V, W ]m
∂xm
∂W m ∂g AB m ∂g AB
p ∂V
= Vp − W
∂xp ∂xm ∂xp ∂xm
∂ ∂g AB AB
p ∂ m ∂g
= V p p Wm − W V
∂x ∂xm ∂xp ∂xm
∂ Ŵ AB ∂ V̂ AB
= Vp p
− Wp (2.58)
∂x ∂xp
where V̂ and Ŵ denote the matrices associated with V and W . But from the previous
AB AB
proposition Vˆg = g AC v̂ CB and Ŵg = g AC ŵCB so
– 26 –
∂g AC P ∂g
AC
[V, W ]AB
e = VP | e ŵ CB
− W |e v̂ CB
∂xp ∂xp
= v̂ AC ŵCB − ŵAC v̂ CB
= [v̂, ŵ]AB (2.59)
as required.
We have therefore shown that if G is a matrix Lie group then the elements L(G) can be
associated with matrices and the Lie bracket is then simply standard matrix commutation
by Proposition 4 (which can be directly checked satisfies all three of the Lie bracket for
Lie algebras). In the literature, it is often conventional to denote the Lie algebra of SO(n)
by so(n), su(n) is the Lie algebra of SU (n), u(n) the Lie algebra for U (n) etc. We will
however continue to use the notation L(G) for the Lie algebra of Lie group G.
Observe that the image [L(G), L(G)] under the Lie bracket need not be the whole of
L(G). This is clear for SO(2), as the Lie bracket vanishes identically in that case. Recall
that the Lie bracket on R viewed as a Lie group under addition vanishes identically as
well. If G is a connected 1-dimensional Lie group then G must either be isomorphic to R
or SO(2) (equivalently U (1)).
2.10 One Parameter Subgroups

Definition 15. A curve σ : R → G is called a one-parameter subgroup if σ(s)σ(t) = σ(s+t)
for all s, t ∈ R.
Note that if σ(t) is a 1-parameter subgroup then σ(0) = e.
We shall show that these subgroups arise naturally as integral curves of left-invariant
vector fields.
Proposition 5. Suppose that V is a left-invariant vector field. Let σ(t) be the integral
curve of V which passes through e when t = 0.
Then σ(t) is a 1-parameter subgroup of G.
Proof
Let x denote some local co-ordinates.
Consider the curves χ1 (t) = σ(s)σ(t) and χ2 (t) = σ(s + t) for fixed s.
These satisfy the same initial conditions χ1 (0) = χ2 (0) = σ(s).
By definition, χ2 satisfies the ODE
d d
((x ◦ χ2 (t))n ) = ((x ◦ σ(s + t))n )
dt d(s + t)
= Vσ(s+t) (xn )
= Vχ2 (t) (xn ) (2.60)
Consider
d d
((x ◦ χ1 (t))n ) = (x ◦ Lσ(s) ◦ σ(t))n

dt dt
– 27 –
d
((x ◦ Lσ(s) ◦ x−1 ) ◦ (x ◦ σ)(t))n

=
dt
∂ d
(x ◦ Lσ(s) ◦ x−1 )n |x◦σ(t) ((x ◦ σ(t))m )

= m
(2.61)
∂x dt
where we have used the chain rule. But by definition of σ(t),
d
((x ◦ σ(t))m ) = Vσ(t)
m
(2.62)
dt
Hence, substituting this into the above:
d ∂
((x ◦ χ1 (t))n ) = m
(x ◦ Lσ(s) ◦ x−1 )n |x◦σ(t)

Vσ(t) m
dt ∂x
= Vσ(t) ((x ◦ Lσ (s))n )
= (Lσ(s)∗ Vσ(t) )(xn ) (by definition of push − forward)
n
= Vχ1 (t) (x ) (as V is left − invariant) (2.63)
So χ1 , χ2 satisfy the same ODE with the same initial conditions.

Hence it follows that σ(s)σ(t) = σ(s + t), i.e. σ defines a 1-parameter subgroup.
The converse is also true: a 1-parameter subgroup σ(t) has left-invariant tangent
vectors
Proposition 6. Suppose σ(t) is a 1-parameter subgroup of G with tangent vector V . Sup-

pose Ve = v. Then Vσ(t) = Lσ(t)∗ v, i.e. the tangent vectors are obtained by pushing forward
the tangent vector at the identity.
Proof
Suppose f : G → R is a smooth function. Then
d
Vσ(t) f = (f ◦ σ)(t)
dt
f (σ(t + h)) − f (σ(t))
= lim
h→0 h
f (σ(t)σ(h)) − f (σ(t))
= lim
h→0 h
d
= (f ◦ Lσ(t) ◦ σ(t0 ))|t0 =0
dt0
= (Lσ(t)∗ v)f (2.64)
so Vσ(t) = Lσ(t)∗ v.
From this we obtain the corollory
Corollory 1. Suppose that σ(t), µ(t) are two 1-parameter subgroups of G with tangent
vectors V , W respectively, with Ve = We = u. Then σ(t) = µ(t) for all t.
Proof
Note that
d
(x ◦ σ(t))n = Vσ(t) xn

dt
– 28 –
= (Lσ(t)∗ u)xn (2.65)
and also
d
(x ◦ µ(t))n = Wσ(t) xn

dt
= (Lµ(t)∗ u)xn (2.66)
So x ◦ σ and x ◦ µ satisfy the same ODE and with the same initial conditions, hence
σ(t) = µ(t).
2.11 Exponentiation
Definition 16. Suppose v ∈ Te (G), Then we define the exponential map exp : Te (G) → G
by
exp(v) = σv (1) (2.67)
where σv (t) denotes the 1-parameter subgroup generated by X(v), and X(v) is the
left-invariant vector field obtained via the push-forward X(v)g = Lg∗ v
Note that exp(0) = e.
Proposition 7. If v ∈ Te (G) and t ∈ R then
exp(tv) = σv (t) (2.68)
and hence exp((t1 + t2 )v) = exp(t1 v)exp(t2 v).

Proof
Take a ∈ R, a 6= 0. Note that σv (at) and σav (t) are both 1-parameter subgroups of G.
The tangent vector to σav (t) at the origin is av.
We also compute the tangent vector to σv (at) at e via
d d
(x ◦ σ(at))n t=0 = a (x ◦ σ(at))n at=0 = av n

(2.69)
dt d(at)
So σv (at) and σav (t) have the same tangent vector av at the origin. Therefore σv (at) =
σav (t).
Hence
exp(tv) = σtv (1) = σv (t) (2.70)
as required.
– 29 –
2.12 Exponentiation on matrix Lie groups
Suppose that G is a matrix Lie group, and v ∈ Te (G) is some tangent matrix. The
d
exponential exp(tv) produces a curve in G with dt (exp(tv))|t=0 = v satisfying exp((t1 +
t2 )v) = exp(t1 v) exp(t2 v)
It is then straightforward to show that
d
(exp(tv))|t=t0 = lim t−1 (exp((t0 + t)v) − exp(t0 v))

dt t→0
= lim t−1 (exp(tv) − I) exp(t0 v)

t→0
= v exp(t0 v) (2.71)
d
Similarly, one also finds dt (exp(tv))|t=t0 = exp(t0 v)v, so v commutes with exp(tv).
d
It is clear that dt exp(tv) = v exp(tv) implies that exp(tv) is infinitely differentiable (as
expected as the integral curve is smooth by construction). Then by elementary analysis,
one can compute the power series expansion for exp(tv) as
∞ n n
X t v
exp(tv) = (2.72)
n!
n=0
with a remainder term which converges to 0 (with respect to the supremum norm on
matrices, for example). Hence, for matrix Lie groups, the Lie group exponential operator
corresponds to the usual operation of matrix exponentiation.
Comment: Suppose that G1 and G2 are Lie groups. Then G = G1 × G2 is a Lie group,
L
and by Lemma 1, L(G) = L(G1 ) L(G2 ).
L
Conversely, suppose Lie groups G, G1 , G2 are such that L(G) = L(G1 ) L(G2 ). Then
by exponentiation, it follows that, at least in a local neighbourhood of e, G has the local
geometric structure of G1 × G2 . However, as it is not in general possible to reconstruct the
whole group in this fashion, one cannot say that G = G1 × G2 globally (typically there will
be some periodic identification somewhere in the Cartesian product group).
In general, one cannot reconstruct the entire Lie group by exponentiating elements of
the Lie algebra. Consider for example, SO(2) and O(2). Both L(O(2)) and L(SO(2)) are
generated by
!
0 1
T1 = (2.73)
−1 0
however it is straightforward to show that

!
θT1 cos θ sin θ
e = (2.74)
− sin θ cos θ
which always has determinant +1. So SO(2) = exp(L(SO(2))) but O(2) 6= exp(L(O(2))).
However, there do exist neighbourhoods B0 of 0 ∈ L(G) and B1 of I ∈ G such that the
map exp : B0 → B1 is invertible. (The inverse is called log by convention).
– 30 –
2.13 Integration on Lie Groups
Suppose that G is a matrix Lie group, and let V be a left-invariant vector field on G, and
suppose that the associated tangent matrix to V at the identity is v̂.
Then if x are some local co-ordinates on G, we know that
∂g(x)
m
g(x)v̂ = Vg(x) (2.75)
∂xm
From this formula, it is clear that if h ∈ G is a constant matrix then
m m
Vg(x) = Vhg(x) (2.76)
If H = {h1 , . . . , hr } is a finite group, and f : H → R is a function, then the integral of

f over H is simply
Xr
f (hi ) (2.77)
i=1
and note that if h ∈ H is fixed then

r
X r
X
f (hi ) = f (hhi ) (2.78)
i=1 i=1
We wish to construct an analogous integral over a matrix Lie group G. Suppose that
x, y are co-ordinates on G and define
dn x = dx1 . . . dxn , dn y = dy 1 . . . dy n (2.79)
Note that dn x and dn y are related by
dn x = J −1 dn y (2.80)
∂y i
where J is the Jacobian J = det ∂x j .
Now suppose that µi for i = 1, . . . , n is a basis of left-invariant vector fields. Then
∂
µi |g(x) = µji,g(x) (2.81)
∂xj
Then we have
∂ ∂y k ∂ ∂
µji,g(x) j
= µ j
i,g(x) j k
= µji,g(y) j (2.82)
∂x ∂x ∂y ∂y
so
∂y j
µji,g(y) = µki,g(x) (2.83)
∂xk
and hence
det µji,g(x) = J −1 det µji,g(y)

(2.84)
Motivated by this, we make the
– 31 –
Definition 17. The Haar measure is defined by
−1
dn x det(µji,g(x) ) (2.85)
Then by the previous reasoning,

−1 −1
dn x det(µji,g(x) ) = dn y det(µji,g(y) ) (2.86)
so the measure is invariant under changes of co-ordinates.

Also, if h is a constant matrix, then as the µi are left-invariant, µji,g(x) = µji,hg(x) , and
so
−1 −1
dn x det(µji,g(x) ) = dn x det(µji,hg(x) ) (2.87)
It follows that if f : G → R, then
Z Z
−1 −1
dn x det(µji,g(x) ) f (g(x)) = dn x det(µji,g(x) ) f (hg(x)) (2.88)
It can be shown that the Haar measure (up to multiplication by a non-zero constant)
is the unique measure with this property.
Example: SL(2, R)
Consider g ∈ SL(2, R),
!
a b
g= (2.89)
c d
for a, b, c, d ∈ R constrained by ad − bc = 1. Note that
!
d −b
g −1 = (2.90)
−c a
We take co-ordinates x1 = b, x2 = c, x3 = d (in some neighbourhood of the identity).

Then
! ! !
∂g c d ∂g 0 0 ∂g −a −b
g −1 1 = 2 , g −1 2 = 1 , g −1 3 = ac
∂x − cd −c ∂x d 0 ∂x d a
(2.91)
Take
! ! !
1 0 0 0 0 1
v1 = , v2 = , v3 = (2.92)
0 −1 1 0 0 0
to be a basis for L(SL(2, R)). Then note that
∂g ∂g ∂g
v1 = −bg −1 1
+ cg −1 2 − dg −1 3
∂x ∂x ∂x
– 32 –
∂g
v2 = dg −1
∂x2
∂g ∂g
v3 = ag −1 1 + cg −1 3 (2.93)
∂x ∂x
It follows that the left-invariant vector fields obtained from pushing-forward the vector
fields associated with v1 , v2 , v3 at the identity with L∗ are
∂ ∂ ∂
µ1 = −b 1 + c 2 − d 3
∂x ∂x ∂x
∂
µ2 = d 2
∂x
∂ ∂
µ3 = a 1 + c 3 (2.94)
∂x ∂x
So the matrix µji is

 
−b c −d
(µji ) =  0 d 0 (2.95)
 
a 0 c
As det(µji ) = d it follows that the Haar measure in these co-ordinates is 1

d db dc dd.
2.14 Representations of Lie Groups

Definition 18. Let V be a finite dimensional vector space (over R or C) and let GL(V )
denote the space of invertible linear transformations V → V . Then a representation of a
Lie group G acting on V is a map D : G → GL(V ) such that
D(g1 g2 ) = D(g1 )D(g2 ) (2.96)
for all g2 , g2 ∈ G. (i.e. D is a homomorphism). The dimension of the representation is

given by dim D = dim V .
Lemma 2. If D is a representation of G then D(e) = 1 where 1 ∈ GL(V ) is the identity

transformation, and if g ∈ G then D(g −1 ) = (D(g))−1 .
Proof
Note that D(e) = D(ee) = D(e)D(e) and so it follows that D(e) = 1 where 1 ∈ GL(V )
is the identity transformation.
If g ∈ G then 1 = D(e) = D(gg −1 ) = D(g)D(g −1 ), so D(g −1 ) = (D(g))−1 .
If M (V ) denotes the set of all linear transformations V → V , and D : G → M (V )
satisfies D(e) = 1 together with the condition (2.96) then it follows from the reasoning
used in the Lemma above that D(g) is invertible for all g ∈ G, with inverse D(g −1 ), and
hence D is a representation.
We next define some useful representations
Definition 19. The trivial representation is defined by D(g) = 1 for all g ∈ G
– 33 –
Definition 20. If G is a matrix Lie group which is a subgroup of GL(n, R) or GL(n, C)
then the group elements themselves act directly on n-component vectors. The fundamental
representation is then defined by D(g) = g.
Definition 21. If G is a matrix Lie group then the adjoint representation Ad : G →
GL(L(G)) is defined by
(Ad(g))X = gXg −1 (2.97)

for g ∈ G and X ∈ L(G) is a tangent matrix.
Lemma 3. Ad(g) as defined above is a representation
Proof
We first verify that if X ∈ L(g) then Ad(g)X ∈ L(G).
Fix g ∈ G. Next, recall that if X ∈ L(G) then there is some smooth curve in G, γ(t)
such that X = dγ(t) −1
dt |t=0 . Define a new smooth curve in G by ρ(t) = gγ(t)g , then the
dρ(t) dγ(t)
tangent matrix to ρ(t) at t = 0 is given by dt |t=0 = g dt |t=0 g −1 = gXg −1 .
Hence Ad(g)X ∈ L(G).
It is clear that Ad(g) is a linear transformation on X ∈ L(G).
Note that Ad(e)X = eXe−1 = X, so Ad(e) = 1. Also, if g1 , g2 ∈ G then
Ad(g1 g2 )X = (g1 g2 )X(g1 g2 )−1

= g1 g2 Xg2−1 g1−1
= g1 (g2 Xg2−1 )g1−1
= g1 (Ad(g2 )X)g1−1
= Ad(g1 )Ad(g2 )X (2.98)
hence Ad(g1 g2 ) = Ad(g1 )Ad(g2 ).

It then follows that Ad(g)Ad(g −1 ) = Ad(gg −1 ) = Ad(e) = 1 so Ad(g) is invertible.
Definition 22. Suppose D is a representation of G acting on V . A subspace W ⊂ V is
called an invariant subspace if D(g)w ∈ W for all g ∈ G and w ∈ W .
Definition 23. Suppose D is a representation of G acting on V . Then D is reducible if
there is an invariant subspace W of V with W 6= 0 and W 6= V . If the only invariant
subspaces of V are 0 and V then D is called irreducible.
Definition 24. A representation D is called totally reducible if there exists a direct sum de-
L L L
composition of V into subspaces Wi , V = W1 W2 ... Wk where the Wi are invariant
subspaces with respect to D and D restricted to Wi is irreducible.
In terms of matrices, if D is totally reducible, then there is some basis of V in which
D has a block diagonal form
 
D1 (g) 0
 0
D(g) =  D2 (g) 
 (2.99)
..
.
– 34 –
Di (g) denotes D(g) restricted to Wi .
Definition 25. A representation D of G acting on V is faithful if D(g) = 1 only if g = e.
Definition 26. Suppose that D is a representation of G acting on V where V is a vector

space over C equipped with an inner product. Then D is a unitary representation if D(g) :
V → V satisfies D(g)D(g)† = 1 for all g ∈ G.
Proposition 8. A finite dimensional unitary representation is totally reducible

Proof
If D is irreducible then we are done. Otherwise, suppose that W is an invariant
subspace. Write V = W ⊕ W⊥ . Suppose v ∈ W⊥ . Then if w ∈ W and g ∈ g,
hD(g)v, wi = hv, D(g)† wi

= hv, D(g)−1 wi
= hv, D(g −1 )wi
= 0 (2.100)
as D(g −1 )w ∈ W because W is an invariant subspace.

Hence it follows that if v ∈ W⊥ then D(g)v ∈ W⊥ and so W⊥ is also an invariant
subspace. Repeating this process by considering D restricted to the invariant subspaces W
and W⊥ one obtains a direct sum decomposition of V into invariant (orthogonal) subspaces
Wi such that D restricted to Wi is irreducible.
Proposition 9. Let V1 , V2 be finite dimensional vector spaces. Suppose D is a represen-

tation of G acting on V1 , and A : V1 → V2 is an invertible linear transformation. Define
D̃(g) = AD(g)A−1 . Then D̃ is a representation of G on V2 .
Proof
As D̃ is a composition of invertible linear transformations, D̃ is also an invertible linear
transformation on V2 .
Also, if g1 , g2 ∈ G
D̃(g1 g2 ) = AD(g1 g2 )A−1

= AD(g1 )D(g2 )A−1
= AD(g1 )A−1 AD(g2 )A−1
= D̃(g1 )D̃(g2 ) (2.101)
and hence D̃ is also a representation.
Definition 27. Suppose D is a representation of G acting on V1 , and A : V1 → V2 is an

invertible linear transformation. Define D̃(g) = AD(g)A−1 . Then D and D̃ are said to be
equivalent representations.
– 35 –
Proposition 10. Schur’s First Lemma: Suppose that D1 and D2 are two irreducible rep-
resentations of G acting on V1 and V2 respectively and there exists a linear transformation
A : V1 → V2 such that
AD1 (g) = D2 (g)A (2.102)
for all g ∈ G. Then either D1 and D2 are equivalent representations, or A = 0.
Proof First note that
Ker A = {ψ ∈ V1 : Aψ = 0} (2.103)
is an invariant subspace of D1 , because if ψ ∈ Ker A then
AD1 (g)ψ = D2 (g)Aψ = 0 (2.104)

so D1 (g)ψ ∈ Ker A for all g ∈ G. But D1 is irreducible on V1 , so one must have
Ker A = 0 or Ker A = V1 , so A is 1-1 or A = 0.
Similarly,
Im A = {φ ∈ V2 : φ = Aψ for some ψ ∈ V1 } (2.105)

is an invariant subspace of D2 , because if φ ∈ Im A then there is some ψ ∈ V1 such
that φ = Aψ and hence
D2 (g)φ = D2 (g)Aψ = AD1 (g)ψ (2.106)

and hence D2 (g)φ ∈ Im A for all g ∈ G. But D2 is irreducible on V2 , so one must have
Im A = 0 or Im A = V2 , i.e. A = 0 or A is onto.
Hence either A = 0 or A is both 1-1 and onto i.e. A is invertible. If A is invertible
then D1 and D2 are equivalent.
Proposition 11. Schur’s Second Lemma: Suppose that D is an irreducible representation
of G on V , where V is a vector space over C, and A : V → V is a linear transformation
such that
AD(g) = D(g)A (2.107)

for all g ∈ G. Then A = λ1 for some λ ∈ C.
Proof
As V is over C, A has at least one eigenvalue. Let λ ∈ C be an eigenvalue of A, with
corresponding eigenspace U (U 6= 0). Then U is an invariant subspace of V with respect
to D, for if ψ ∈ U then
Aψ = λψ (2.108)
and if g ∈ G, then
AD(g)ψ = D(g)Aψ = D(g)(λψ) = λD(g)ψ (2.109)
so D(g)ψ ∈ U .
But D is irreducible on V , so this implies U = V (as U 6= 0).
Hence it follows that A = λ1.
– 36 –
Definition 28. Suppose that D1 and D2 are representations of the Lie group G over vector
N
spaces V1 and V2 respectively. Let V = V1 V2 be the standard tensor product vector space
of V1 and V2 consisting of elements v1 ⊗ v2 (v1 ∈ V1 and v2 ∈ V2 ) in the vector space dual
to the space of bilinear forms on V1 × V2 . If v1 ⊗ v2 ∈ V then v1 ⊗ v2 acts linearly on
bilinear forms Ω via v1 ⊗ v2 Ω = Ω(v1 , v2 ). V is equipped with pointwise addition and scalar
multiplication which satisfy (v1 + w1 ) ⊗ (v2 + w2 ) = v1 ⊗ v2 + v1 ⊗ w2 + w1 ⊗ v2 + w1 ⊗ w2
and α(v1 ⊗ v2 ) = (αv1 ) ⊗ v2 = v1 ⊗ (αv2 ).
Then the tensor product representation D is defined as a linear map on V satisfying
D(g)v1 ⊗ v2 = D1 (g)v1 ⊗ D2 (g)v2 (2.110)
for g ∈ G and v1 ∈ V1 and v2 ∈ V2
Proposition 12. The tensor product representation defined above is a representation.

Proof
The map D(g) is linear by construction, also
D(e)v1 ⊗ v2 = D1 (e)v1 ⊗ D2 (e)v2 = v1 ⊗ v2 (2.111)
because D1 (e) = 1 and D2 (e) = 1. Hence D(e) = 1. And if g1 , g2 ∈ G then
D(g1 g2 )v1 ⊗ v2 = D1 (g1 g2 )v1 ⊗ D2 (g1 g2 )v2

= D1 (g1 )D1 (g2 )v1 ⊗ D2 (g1 )D2 (g2 )v2
= D(g1 )(D1 (g2 )v1 ⊗ D2 (g2 )v2 )
= D(g1 )D(g2 )v1 ⊗ v2 (2.112)
so D(g1 g2 ) = D(g1 )D(g2 ). Hence, this together with D(e) = 1 implies that D(g) is
invertible. So D(g) is a representation.
N
Note that if D1 is irreducible on V1 and D2 is irreducible on V2 then D = D1 D2 is
N
not generally irreducible on V = V1 V2 . Indeed, we shall be particularly interested in
decomposing D into irreducible components in several explicit examples.
2.15 Representations of Lie Algebras

Definition 29. Let V be a finite dimensional vector space (over R or C) and let M (V )
denote the space of linear transformations V → V . Suppose that L(G) is the Lie algebra of a
Lie group G. Then a representation of L(G) acting on V is a linear map d : L(G) → M (V )
satisfying
d([X, Y ]) = d(X)d(Y ) − d(Y )d(X) (2.113)
for all X, Y ∈ L(G). The dimension of the representation is the dimension of V .
Definition 30. The trivial representation of L(G) on V is given by d(X) = 0 for all
X ∈ L(G)
– 37 –
Definition 31. If G is a matrix Lie group and hence L(G) is a matrix Lie algebra, then
the tangent vectors can be regarded as matrices acting directly on n-component vectors.
Then we define the fundamental representation of L(G) on V by d(X) = X
There is a particularly natural representation associated with any Lie algebra.
Definition 32. Let L(G) be a Lie algebra. Then the adjoint representation is a represen-
tation of L(G) over the vector space L(G), ad : L(G) → M (L(G)) defined by
(ad v)w = [v, w] (2.114)

for v, w ∈ L(G).
It is clear from the above that (ad v)w is linear in w , hence ad v ∈ M (L(G)), and
ad v is also linear in v.
Moreover, if v1 , v2 , w ∈ L(G) then
(ad [v1 , v2 ])w = [[v1 , v2 ], w]

= [v1 , [v2 , w]] − [v2 , [v1 , w]] (using the Jacobi identity)
= (ad v1 )[v2 , w] − (ad v2 )[v1 , w]
= (ad v1 )(ad v2 )w − (ad v2 )(ad v1 )w (2.115)
so ad is indeed a representation.
2.16 The Baker-Campbell-Hausdorff (BCH) Formula

The BCH formula states that the product of two exponentials can be written as an expo-
nential:
1 1 1
exp(v) exp(w) = exp(v + w + [v, w] + [v, [v, w]] + [[v, w], w] + ...) (2.116)
2 12 12
where ... indicates terms of higher order in v and w. For simplicity we shall consider
only matrix Lie groups, in which case the Lie algebra elements are square matrices.
To obtain the first few terms in this formula, consider etv etw as a function of t and set
eZ(t) = etv etw (2.117)

At t = 0 we must have eZ(0) = I, which is solved by taking Z(0) = 0 (this solution is
unique if we limit ourselves to the neighbourhood of the identity on which exp is invertible).
Hence we can write Z(t) as a power series
1
Z(t) = tP + t2 Q + O(t3 ) (2.118)
2
where we determine the matrices P and Q by expanding out (2.117) in powers of t:
1 1
I + t(v + w) + t2 (w2 + v 2 + 2vw) + O(t3 ) = I + tP + t2 (Q + P 2 ) + O(t3 ) (2.119)
2 2
from which we find P = v + w and Q = [v, w].
– 38 –
Proposition 13. All higher order terms in the power series expansion of Z(t) in the BCH
formula depend only on sums of compositions of commutators on v and w.
Proof
Suppose that Z(y) is an arbitrary square matrix. Consider
∂ xZ(y)
f1 (x, y) = (e ) (2.120)
∂y
and
Z 1
∂Z(y) (1−t)Z(y)
f2 (x, y) = e(x−1+t)Z(y) e dt (2.121)
1−x ∂y
These both satisfy
∂fi ∂Z(y) xZ(y)

= e + Z(y)fi (x, y) (2.122)
∂x ∂y
and fi (0, y) = 0 for i = 1, 2. Hence f1 (x, y) = f2 (x, y).
Now suppose that Z(t) is the matrix appearing in the BCH formula, i.e.
eZ(t) = etv etw (2.123)

Then consider the identity f1 (1, t) = f2 (1, t). This implies that
Z 1
tv −tv ∂Z(t) −yZ(t)
v + e we = eyZ(t) e dy (2.124)
0 ∂t
Now consider the function
g(t) = etv we−tv (2.125)

this satisfies g(0) = Y and
dn g
= etv (ad v)n we−tv (2.126)
dtn
hence the power series expansion of g(t) is given by
∞ n
tv −tv
X t
e we =w+ (ad v)n w (2.127)
n!
n=1
Applying this expression to both sides of (2.124) and performing the y-integral, one
finds
∞ n ∞
X t dZ X 1 dZ
v+w+ (ad v)n w = + (ad Z(t))n (2.128)
n! dt (n + 1)! dt
n=1 n=1
P∞
tn
Then by expanding out Z(t) = n=1 Zn (as we know that Z(0) = 0), it follows by
induction using the above equation that the Zn can be written as sums of compositions of
commutators on v and w.
Exercise: Suppose that [v, w] = 0. Show that ev ew = ev+w .
– 39 –
Proposition 14. Suppose that D is a representation of the matrix Lie group G acting on
V . Then there is a representation d of L(G) also on V defined via
d
d(v) = D(exp(tv)) |t=0 (2.129)
dt
for v ∈ L(G).
Proof
It is convenient to expand out up to O(t3 ) by
D(etv ) = 1 + td(v) + t2 h(v) + O(t3 ) (2.130)

Note that as D is a representation we must have D(e(t1 +t2 )v ) = D(et1 v )D(et2 v )
Hence
1 + (t1 + t2 )d(v) + (t1 + t2 )2 h(v) + O(t3i ) = (1 + t1 d(v) + t21 h(v))(1 + t2 d(v) + t22 h(v)) + O(t3i )
(2.131)
1 2
and so on equating the t1 t2 coefficient we find h(v) = 2 d(v) .
Next consider for v, w ∈ L(G)
1 1
D(e−tv e−tw etv etw ) = (1 − td(v) + t2 d(v)2 )(1 − td(w) + t2 d(w)2 )
2 2
1 2 1
× (1 + td(v) + t d(v) )(1 + td(w) + t2 d(w)2 ) + O(t3 )
2
2 2
= 1 + t2 (d(v)d(w) − d(w)d(v)) + O(t3 ) (2.132)
But using the BCH formula
1 2 3 1 2
[v,w]+O(t3 )
e−tv e−tw etv etw = e−t(v+w)+ 2 t [v,w]+O(t ) et(v+w)+ 2 t
2 3
= et [v,w]+O(t ) (2.133)
and so
2 [v,w]+O(t3 )
D(e−tv e−tw etv etw ) = D(et ) = 1 + t2 d([v, w]) + O(t3 ) (2.134)
Comparing (2.132) with (2.134) we find that
d([v, w]) = d(v)d(w) − d(w)d(v) (2.135)

as required.
To show that d is linear, suppose v, w ∈ L(G) and α, β are constants. Then
D(etαv etβw ) = D(etαv )D(etβw )

= (1 + tαd(v) + O(t2 ))(1 + tβd(w) + O(t2 ))
= 1 + t(αd(v) + βd(w)) + O(t2 ) (2.136)
– 40 –
But by the BCH formula
2
D(etαv etβw ) = D(et(αv+βw)+O(t ) ) = 1 + td(αv + βw) + O(t2 ) (2.137)
Hence, comparing the O(t) terms in (2.136) and (2.137) it follows that d(αv + βw) =
αd(v) + βd(w).
Proposition 15. If G is a matrix Lie group, then the representation Ad : G → GL(L(G))

induces the representation ad : L(G) → M (L(G)).
Proof
If v, w ∈ L(G) then
Ad (etv )w = etv we−tv

= (1 + tv + O(t2 ))w(1 − tv + O(t2 ))
= w + t[v, w] + O(t2 )
= (I + tad v + O(t2 ))w (2.138)
and hence it follows that
d
(Ad(etv ))|t=0 = ad v (2.139)
dt
as required.
We have seen that a representation D of the matrix Lie group G acting on V gives rise
to a representation d of the Lie algebra L(G) on V . A partial converse is true.
Definition 33. Suppose that G is a matrix Lie group. Let d denote a representation of
L(G) on V . Then a representation D is induced locally on G via
D(g) = ed(v) (2.140)
for those g ∈ G such that g = ev .

Here we assume that the representation d(v) is realized as a matrix linear transforma-
tion on V , so that the standard matrix exponentiation ed(v) may be taken. The represen-
tation D induced by d is generally not globally well-defined, but it is locally well-defined
on the neighbourhood of the identity on which exp is invertible.
Proposition 16. The map D given in (2.140) which is locally induced by the representation
d of L(G) on V defines a representation.
Proof
Clearly, D(g) defines a linear transformation on V .
As I = e0 it follows that D(e) = ed(0) = e0 = 1 where d(0) = 0 follows from the
linearity of d.
Also, suppose that g1 , g2 have g1 = ev1 , g2 = ev2 . Then by the BCH formula we have
1
g1 g2 = ev1 +v2 + 2 [v1 ,v2 ]+... (2.141)
– 41 –
where . . . denotes a sum of higher order nested commutators by Proposition 13. Hence
1
D(g1 g2 ) = ed(v1 +v2 + 2 [v1 ,v2 ]+... )
1
= ed(v1 )+d(v2 )+ 2 d([v1 ,v2 ])+d(... ) (by the linearity of d)
1
= ed(v1 )+d(v2 )+ 2 [d(v1 ),d(v2 )]+... (using (2.113))
= ed(v1 ) ed(v2 )
= D(g1 )D(g2 ) (2.142)
So D is at least locally a representation. Note that we have made use of the fact that all
higher order terms in the BCH expansion can be written as sums of commutators, together
with the property (2.113) of representations of L(G) in proceeding from the second to the
third line of the above equation.
Proposition 17. Suppose that G is a matrix Lie group. If D is a unitary representation

of G on V then the induced representation d of L(G) on V is antihermitian.
Conversely, suppose d is a antihermitian representation of L(G) on V , then the (lo-
cally) induced representation D of G on V is unitary.
Proof
First suppose that D is a unitary representation.
Recall that d satisfies D(etX ) = I + td(X) + O(t2 ) for t ∈ R and X ∈ L(G).
As D is unitary it follows that D(etX )D(etX )† = I.
Hence (I + td(X) + O(t2 ))(I + td(X) + O(t2 ))† = I, so expanding out, the O(t) terms
imply d(X) + d(X)† = 0, i.e. d is antihermitian.
Conversely, suppose that d is an antihermitian representation of L(G) on V . Let D
denote the (locally) induced representation of G on V .
Suppose that g ∈ G is given by g = eX for X ∈ L(G). Then
D(g) = ed(X) (2.143)
Then
D(g)D(g)† = ed(X) (ed(X) )†

†
= ed(X) ed(X)
= ed(X) e−d(X)
= 1 (2.144)
Hence D(g) is unitary .

There are directly analogous definitions for irreducibility of representations of Lie al-
gebras
Definition 34. Suppose d is a representation of L(G) acting on V . A subspace W ⊂ V is

called an invariant subspace if d(X)ω ∈ W for all X ∈ L(G) and ω ∈ W .
Definition 35. Suppose d is a representation of L(G) acting on V . Then d is reducible

if there is an invariant subspace W of V with W 6= 0 and W 6= V . If the only invariant
subspaces of V are 0 and V then d is called irreducible.
– 42 –
Definition 36. A representation d of L(G) is called totally reducible if there exists a direct
L L L
sum decomposition of V into subspaces Wi , V = W1 W2 ... Wk where the Wi are
invariant subspaces with respect to d and d restricted to Wi is irreducible.
Proposition 18. Suppose that G is a matrix Lie group. If D is a representation of G on V
with invariant subspace W , then W is an invariant subspace of the induced representation
d of L(G) on V
Conversely, suppose d is a representation of L(G) on V with invariant subspace W ;
then W is an invariant subspace of the (locally) induced representation D of G on V .
Proof
Suppose that D is a representation of G on V with invariant subspace W with respect
to D. Let d be the induced representation of L(G) on V . If w ∈ W and X ∈ L(G) then
d d
d(X)w = (D(etX ))t=0 w = (D(etX )w)t=0 (2.145)
dt dt
As D(etX )w ∈ W for all t ∈ R it follows that d(X)w ∈ W .
Conversely, suppose that d is a representation of L(G) on V , and W is an invariant
subspace of V with respect to d. Let D be the locally defined representation of G induced
by d. Then if g ∈ G is given by g = eX for some X ∈ L(G) then if w ∈ W ,
∞
X 1 n
D(g)w = ed(X) w = d (X)w (2.146)
n!
n=0
However, as W is an invariant subspace of V with respect to d, it follows that dn (X)w ∈ W

for all n ∈ N. Hence D(g)w ∈ W .
Note that in this proof we made implicit use of the closure of W .
There is also a natural concept of equivalent representations of Lie algebras.
Definition 37. Suppose d is a representation of L(G) acting on V1 , and B : V1 → V2 is
˜
an invertible linear transformation. Define d(X) = Bd(X)B −1 for X ∈ L(G). Then d and
d˜ are said to be equivalent representations.
Exercise: Show that d˜ defined above is a representation of L(G). Also show that if D1 ,
D2 are equivalent representations of G on vector spaces V1 and V2 then the corresponding
induced representations d1 and d2 on V1 and V2 are equivalent; and conversely, if d1 and
d2 are equivalent representations of L(G) on V1 and V2 then the locally defined induced
representations D1 and D2 of G are equivalent.
Note that Schur’s lemmas may be applied to representations of Lie algbras in exactly
the same way as to representations of Lie groups.
Hence we have shown that there is (at least locally) a 1-1 correspondence between
irreducible representations of the Lie group G and the Lie algebra L(G). This is useful,
because it enables us to map the analysis of representations of G to those of L(G), and
thus the problem reduces to one of linear algebra.
Proposition 19. Suppose that d1 and d2 are representations of L(G) acting on V1 and
N
V2 and let V = V1 V2 . Define d = d1 ⊗ 1 + 1 ⊗ d2 as a linear map on V . Then d is a
representation of L(G) acting on V .
– 43 –
Proof
If w1 , w2 ∈ L(V ) and α, β are scalars and v1 ⊗ v2 ∈ V then
d(αw1 + βw2 )v1 ⊗ v2 = d1 (αw1 + βw2 )v1 ⊗ v2 + v1 ⊗ d2 (αw1 + βw2 )v2

= (αd1 (w1 ) + βd1 (w2 ))v1 ⊗ v2 + v1 ⊗ (αd2 (w1 ) + βd2 (w2 ))v2
= αd1 (w1 )v1 ⊗ v2 + βd1 (w2 )v1 ⊗ v2
+ αv1 ⊗ d2 (w1 )v2 + βv1 ⊗ d2 (w2 )v2
= α(d1 (w1 )v1 ⊗ v2 + v1 ⊗ d2 (w1 )v2 )
+ β(βd1 (w2 )v1 ⊗ v2 + v1 ⊗ d2 (w2 )v2 )
= αd(w1 )v1 ⊗ v2 + βd(w2 )v2 ⊗ v2 (2.147)
so d is linear on L(G). Also
d([w1 , w2 ])v1 ⊗ v2 = d1 ([w1 , w2 ])v1 ⊗ v2 + v1 d2 ([w1 , w2 ])v2

= (d1 (w1 )d1 (w2 ) − d1 (w2 )d1 (w1 ))v1 ⊗ v2
+ v1 ⊗ (d2 (w1 )d2 (w2 ) − d2 (w2 )d2 (w1 ))v2
= d1 (w1 )d1 (w2 )v1 ⊗ v2 − d1 (w2 )d1 (w1 )v1 ⊗ v2
+ v1 ⊗ d2 (w1 )d2 (w2 )v2 − v1 ⊗ d2 (w2 )d2 (w1 )v2 (2.148)
Also note that

d(w1 )d(w2 )v1 ⊗ v2 = d(w1 ) d1 (w2 )v1 ⊗ v2 + v2 ⊗ d2 (w2 )v2
= d1 (w1 )d1 (w2 )v1 ⊗ v2 + d1 (w2 )v1 ⊗ d2 (w1 )v2
+ d1 (w1 )v1 ⊗ d2 (w2 )v2 + v1 ⊗ d2 (w1 )d2 (w2 )v2 (2.149)
where the sum of the second and third terms in this expression is symmetric in w1 and w2 .
Hence
d(w1 )d(w2 )v1 ⊗ v2 − d(w2 )d(w1 )v1 ⊗ v2 = d1 (w1 )d1 (w2 )v1 ⊗ v2 − d1 (w2 )d1 (w1 )v1 ⊗ v2
+ v1 ⊗ d2 (w1 )d2 (w2 )v2 − v1 ⊗ d2 (w2 )d2 (w1 )v2
= d([w1 , w2 ])v1 ⊗ v2 (2.150)
as required .
Proposition 20. Suppose that D1 and D2 are representations of matrix Lie group G on
V1 and V2 with induced representations of L(G) on V1 and V2 denoted by d1 and d2 . Let
N N
D = D1 D2 denote the representation of G on the tensor product V1 V2 . Then the
N
corresponding induced representation of L(G) on V1 V2 is d = d1 ⊗ 1 + 1 ⊗ d2 .
Proof Suppose w ∈ L(G), then expanding out in powers of t;
D(etw )v1 ⊗ v2 = D1 (etw )v1 ⊗ D2 (etw )v2
– 44 –
= (v1 + td1 (w)v1 + O(t2 )) ⊗ (v2 + td2 (w)v2 + O(t2 ))
= v1 ⊗ v2 + t d1 (w)v1 ⊗ v2 + v1 ⊗ d2 (w)v2 + O(t2 )

= v1 ⊗ v2 + t(d1 ⊗ 1 + 1 ⊗ d2 )v1 ⊗ v2 + O(t2 ) (2.151)
and hence from the O(t) term we find the induced representation d = d1 ⊗ 1 + 1 ⊗ d2
as required .
2.17 The Killing Form and the Casimir Operator

Definition 38. Suppose that G is a matrix Lie group with Lie algebra L(G). Then for
X ∈ L(G), ad X can be realized as a matrix linear transformation on L(G). The Killing
form κ is defined by
κ(X, Y ) = Tr (ad Xad Y ) (2.152)
Suppose that Ta is a basis for L(G). Then from the Killing form one obtains a sym-
metric matrix
κab = κ(Ta , Tb ) (2.153)
Denote the matrix elements of ad Ta by (ad Ta )b c ; then note that
(ad Ta )Tb = [Ta , Tb ] = cab c Tc

= (ad Ta )b c Tc (2.154)
Hence (ad Ta )b c = cab c . So it follows that
κab = cad e cbe d (2.155)
Lemma 4. The Killing form is associative: κ(X, [Y, Z]) = κ([X, Y ], Z)

Proof
κ(X, [Y, Z]) = Tr (ad Xad [Y, Z])

= Tr (ad X(ad Y ad Z − ad Zad Y ))
= Tr (ad Xad Y ad Z) − Tr (ad Xad Zad Y )
= Tr (ad Xad Y ad Z) − Tr (ad Y ad Xad Z)
= Tr ((ad Xad Y − ad Y ad X)ad Z)
= Tr (ad [X, Y ]ad Z)
= κ([X, Y ], Z) (2.156)
as required. .
As κab is symmetric, one can always choose an adapted basis for L(G) in which κab is
a diagonal matrix, and by rescaling the Lie algebra generators, the diagonal entries can be
set to +1, 0 or −1.
Definition 39. The Killing form is non-degenerate if κab has no zero diagonal entries in
the adapted basis. The Lie algebra L(G) is then called semi-simple. If all the diagonal
entries are −1 then L(G) is said to be a compact Lie algebra.
– 45 –
Lemma 5. Suppose that L(G) is semi-simple. Define cabc = cab d κdc (i.e. lower the last
index of the structure constants with the Killing form). Then cabc is totally antisymmetric
in a, b, c.
Proof
Note that
κ(Ta , [Tb , Tc ]) = cbc d κ(Ta , Td ) = cbc d κad = cbca (2.157)

and
κ([Ta , Tb ], Tc ) = cab d κ(Td , Tc ) = cab d κdc = cabc = −cbac (2.158)

But by associativity of κ, (2.157) and (2.158) are equal. Hence cbca = −cbac . As cabc is
skew symmetric in both the first two and the last two indices, it follows that cabc is totally
antisymmetric.
Definition 40. Suppose that L(G) is a Lie algebra with non-degenerate Killing form, and
d is a representation of L(G) on V . The Casimir operator C of L(G) is defined by
X
C=− (κ−1 )ab d(Ta )d(Tb ) (2.159)
a,b
where κ−1 denotes the inverse of the Killing form.
Proposition 21. The Casimir operator commutes with d(X) for all X ∈ L(G)
Proof
It suffices to show that [C, d(Ta )] = 0 for all Ta .
Note that
X
(κ−1 )bc [d(Ta ), d(Tb )d(Tc )]

[d(Ta ), C] = −
b,c
X
(κ−1 )bc [d(Ta ), d(Tb )]d(Tc ) + d(Tb )[d(Ta ), d(Tc )]

=−
b,c
X
(κ−1 )bc d[Ta , Tb ]d(Tc ) + d(Tb )d[Ta , Tc ]

=−
b,c
X
(κ−1 )bc cab ` d(T` )d(Tc ) + cac ` d(Tb )d(T` )

=−
b,c
= −ca c` d(T` )d(Tc ) − ca c` d(Tc )d(T` )
=0 (2.160)
where we have defined X

ca bc = (κ−1 )b` ca` c (2.161)
`
and we note that ca bc is antisymmetric in b, c.

Note that if d is irreducible then by Schur’s second lemma, C must be a scalar multiple
of the identity.
– 46 –
If L(G) is compact, then working in the adapted basis, C takes a particularly simple
form X
C= d(Ta )d(Ta ) (2.162)
a
– 47 –
3. SU(2) and Isospin
3.1 Lie Algebras of SO(3) and SU (2)

We have shown that the Lie algebra of SO(3) consists of the 3 × 3 antisymmetric real
matrices.
A basis for L(SO(3)) is given by
     
0 0 0 0 0 1 0 −1 0
T1 =  0 0 −1 T2 =  0 0 0 T3 =  1 0 0 (3.1)
     
0 1 0 −1 0 0 0 0 0
Exercise: Show that

[Ta , Tb ] = abc Tc (3.2)
where abc is totally antisymmetric and 123 = 1.

Next consider SU (n). Let M (t) be a smooth curve in SU (n) with M (0) = I . Then
M (t) must satisfy
det M (t) = 1 M (t)M (t)† = I (3.3)
Differentiating these constraints we obtain
dM (t)†

dM (t) dM (t)
Tr M (t)−1 =0 M (t)† + M (t) =0 (3.4)
dt dt dt
Setting t = 0 we find
Tr m = 0 m + m† = 0 (3.5)
where m = dMdt(t) |t=0 . Hence L(SU (n)) consists of the traceless antihermitian matrices.
Exercise: Verify that if X and Y are traceless antihermitian square matrices then so is
[X, Y ].
It is convenient to make use of the Pauli matrices σa defined by
! ! !
0 1 0 −i 1 0
σ1 = σ2 = σ3 = (3.6)
1 0 i 0 0 −1
which satisfy σa σb = δab I + iabc σc .
Then a basis of traceless antihermitian 2 × 2 matrices is given by taking Ta = − 2i σa .
It follows that
[Ta , Tb ] = abc Tc (3.7)
Comparing (3.2) and (3.7) it is clear that SO(3) and SU (2) have the same Lie algebra.
We might therefore expect SO(3) and SU (2) to be similar, at least near to the identity.
We shall see that this is true.
Exercise: Using this basis, show that the Lie algebra L(SU (2)) is of compact type.
– 48 –
3.2 Relationship between SO(3) and SU (2)
Proposition 22. The manifold SU (2) can be identified with S 3 .
Proof
Consider
!
α µ
U= ∈ SU (2) (3.8)
β ν
for α, β, µ, ν ∈ C. ! !
α µ
Then U U † = I implies that be orthogonal to with respect to the standard
β ν
!
α
inner product on C2 . As the orthogonal complement to in C2 is a 1-dimensional
β
!
−β̄
complex vector space spanned by it follows that µ = −σ β̄, ν = σ ᾱ for σ ∈ C.
ᾱ
! !
α µ
We also require that and have unit norm, which fixes
β ν
|σ|2 = |α|2 + |β|2 = 1 (3.9)

Finally, the constraint det U = 1 fixes σ = 1.
Hence we have shown that !
α −β̄
U= (3.10)
β ᾱ
where αᾱ + β β̄ = 1. Such U is automatically an element of SU (2). Hence this is the form
of the most generic element of SU (2).
Set α = y0 + iy3 , β = −y2 + iy1 for y0 , y1 , y2 , y3 ∈ R. Then it is straightforward to see
that
U = y0 I + iyn σn (3.11)
and αᾱ + β β̄ = 1 is equivalent to y02 + y12 + y22 + y32 = 1, i.e. y ∈ S 3 .
This establishes a smooth invertible map between SU (2) and S 3 .
Note that this map provides an explicit way of seeing that SU (2) is connected, because
S 3 is connected.
Proposition 23. There is a 2 − 1 correspondence R : SU (2) → SO(3) between SU (2) and

SO(3). The map R is a group homomorphism.
Proof Suppose U ∈ SU (2). Define a 3 × 3 matrix R(U ) via
1
R(U )mn = Tr σm U σn U †

(3.12)
2
By writing U = y0 I +iym σm for y0 , ym ∈ R satisfying y02 +yp yp = 1, it is straightforward
to show that
Rmn = (y02 − yp yp )δmn + 2mnq y0 yq + 2ym yn (3.13)
– 49 –
(here we have written Rmn = R(U )mn ).
It is clear that if yp = 0 for p = 1, 2, 3 so that U = ±I, then R = I, so R ∈ SO(3).
More generally, suppose that yp yp 6= 0. Then one can set y0 = cos α, yp = sin αzp for
0 < α < 2π and α 6= π. Then the constraint y02 + yp yp = 1 implies that zp zp = 1, i.e z is a
unit vector in R3 . The expression (3.13) can be rewritten as
Rmn = cos 2αδmn + sin 2αmnq zq + (1 − cos 2α)zm zn (3.14)
It is then apparent that
Rmn zn = zm (3.15)
and if x is orthogonal to z then
Rmn xn = cos 2αxm + sin 2αmnq xn zq (3.16)
The transformation R therefore corresponds to a rotation of angle 2α in the plane with

unit normal vector z.
It is clear that any non-trivial rotation in SO(3) can be written in this fashion. How-
ever, the correspondence is not 1 − 1. To see this explicitly, suppose that two rotations
corresponding to y ∈ S 3 and u ∈ S 3 are equal. Then
(y02 − yp yp )δmn + 2mnq y0 yq + 2ym yn = (u20 − up up )δmn + 2mnq u0 uq + 2um un (3.17)
where y02 + yp yp = u20 + up up = 1.

From the antisymmetric portion of this matrix we find y0 yp = u0 up .
From the diagonal elements with n = m, we see that
y02 − yp yp + 2(ym )2 = u20 − up up + 2(um )2 (3.18)
where p is summed over but m is fixed. Summing over m we find
3y02 − yp yp = 3u20 − up up (3.19)
where p is summed over; which together with y02 + yp yp = u20 + up up = 1 implies that
y02 = u20 and yp yp = up up (sum over p). Substituting back into (3.18) we find ym2 = u2 for
m
each m = 1, 2, 3.
Suppose first that y0 6= 0, then u0 6= 0, and it follows that y0 = ±u0 and yp = ±up for
each p = 1, 2, 3 (with the same sign throughout).
Next, suppose y0 = 0. Then u0 = 0 also, and ym yn = um un for each m, n = 1, 2, 3.
Contracting with yn we get (yn yn )ym = (yn un )um for m = 1, 2, 3. As yn yn = 1, this implies
that ym = λum for m = 1, 2, 3 where λ is constant. Hence, (1 − λ2 )um un = 0. Contracting
over m and n then gives 1 − λ2 = 0, so λ = ±1. Therefore yp = ±up for p = 1, 2, 3 (with
the same sign throughout).
– 50 –
Hence, we have shown that each R ∈ SO(3) corresponds to two elements U and
−U ∈ SU (2). These two elements correspond to antipodal points ±y ∈ S 3 . This establishes
the correspondence.
It remains to check that R(U1 U2 ) = R(U1 )R(U2 ) for U1 , U2 ∈ SU (2). Note that on
writing
U1 = y0 I2 + iyn σn , U2 = w0 + iwn σn (3.20)
for y0 , yp , w0 , wp ∈ R satisfying y02 + yp yp = w02 + wp wp = 1 then
U1 U2 = u0 I2 + iun σn (3.21)
where u0 = y0 w0 − yp wp and um = y0 wm + w0 ym − mpq yp wq satisfy u20 + up up = 1. It then

suffices to evaluate directly
R(U1 U2 )mn = (u20 − up up )δmn + 2mnq u0 uq + 2um un (3.22)
and compare this with
R(U1 )mp R(U2 )pn = (y02 −y` y` )δmp +2mpq y0 yq +2ym yp (w02 −wr wr )δpn +2pnr w0 wr +2wp wn

(3.23)
On expanding out these two expressions in terms of y and w it becomes an exercise in
algebra to show that R(U1 U2 )mn = R(U1 )mp R(U2 )pn as required.
Exercise: Verify the identity R(U1 U2 )mn = R(U1 )mp R(U2 )pn .
It is conventional to write SU (2) = S 3 and SO(3) = S 3 /Z2 , where S 3 /Z2 is the 3-
sphere with antipodal points identified. SU (2) is called the double cover of SO(3); and
there is an isomorphism SO(3) ∼ = SU (2)/Z2 .
It can be shown (using topological methods outside the scope of this course) that
SU (2) and SO(3) are not homeomorphic. This is because SU (2) and SO(3) have different
fundamental groups π1 . In particular, as SU (2) ∼ = S 3 , and S 3 is simply connected, it
follows that π1 (SU (2)) is trivial. However, it can be shown that π1 (SO(3)) = Z2 .
3.3 Irreducible Representations of SU (2)

The standard basis of L(SU (2)) which we have been using is Ta = − 2i σa . Suppose that
d is a finite-dimensional irreducible representation of L(SU (2)) on V , where V is a vector
space over C.
Define
i
J3 = id(T3 ), J± = √ (d(T1 ) ± id(T2 )) (3.24)
2
Then
[J3 , J± ] = ±J± , [J+ , J− ] = J3 (3.25)
As V is a complex vector space, there exists an eigenstate |φi of J3 with some eigenvalue
λ, and using (3.25) it follows that
– 51 –
J3 J± |φi = (λ ± 1)J± |φi (3.26)
and so by simple induction
J3 (J± )n |φi = (λ ± n)(J± )n |φi (3.27)

for non-negative integer n; i.e. J± are the standard raising and lowering operators.
The eigenvalues of J3 are called weights. We have therefore shown that (J± )n |φi either
vanishes or is a J3 eigenstate with eigenvalue λ ± n.
Consider the states (J+ )n |φi. If non-vanishing, these states are linearly independent
(as they have different J3 eigenvalues). Hence, as V is finite dimensional, there exists a J3
eigenstate (J+ )n |φi which we will refer to as |j, ji with eigenvalue j such that J+ |j, ji = 0.
Definition 41. j is called the highest weight of the representation. In the context of particle
physics, it is called the spin.
Note that by acting on |j, ji with J− other distinct eigenstates of J3 are obtained. As
we are interested in finite dimensional representations, it follows that (J− )N |j, ji = 0 for
some positive integer N (otherwise one could just keep going and the representation would
be infinite dimensional). Let k + 1 be the smallest positive integer for which this happens,
and set |ψk i = (J− )k |j, ji, so, by construction, J− |ψk i = 0.
Define for ` = 0, . . . , k
|ψ` i = (J− )` |j, ji (3.28)

Then |ψ` i are (non-vanishing) j − ` eigenvectors of J3 .
Note that
J+ |ψ` i = J+ (J− )` |j, ji

= (J+ J− )(J− )`−1 |j, ji
[J+ , J− ] + J− J+ (J− )`−1 |j, ji

=
J3 + J− J+ (J− )`−1 |j, ji

=
= (j − (` − 1))(J− )`−1 |j, ji + J− J+ (J− )`−1 |j, ji (3.29)
Repeating this process, one finds by induction

J+ |ψ` i = j − (` − 1) + j − (` − 2) + · · · + j − 1 + j |ψ`−1 i
1
= `(j − (` − 1)) |ψ`−1 i (3.30)
2
In order to constrain j recall that J− |ψk i = 0, so
0 = J+ J− |ψk i

= [J+ , J− ] + J− J+ |ψk i
– 52 –

= J3 + J− J+ |ψk i
1
= (j − k) |ψk i + J− k(j − (k − 1)) |ψk−1 i
2
1
= j − k + k(j − (k − 1)) |ψk i
2
1
= (k + 1)(2j − k) |ψk i (3.31)
2
As k is non-negative, it follows that k = 2j

Using this it is straightforward to prove the following
Proposition 24. V = span{J− k |j, ji , J k−1 |j, ji , . . . , |j, ji} and the highest weight state is
−
unique.
Proof
Consider the vector space V 0 spanned by |ψi i for i = 0, . . . , k. This is an invariant
subspace of V with respect to the representation d. As the representation is irreducible on
V it follows that V = V 0 . In particular, J3 is diagonalizable on V and each eigenspace is
1-dimensional.
To prove uniqueness suppose that |φi ∈ V satsfies J+ |φi = 0. As |φi ∈ V it follows
that we can write
X2j
|φi = ai (J− )i |j, ji (3.32)
i=0
for constants ai . Applying (J+ )2j to both sides of this equation implies a2j = 0. Then,
applying (J+ )2j−1 implies a2j−1 = 0. Continuing in this way, we obtain a1 = a2 = · · · =
a2j−1 = a2j = 0, and so |φi = a0 |j, ji. So the highest weight state in an irreducible
representation of L(SU (2)) is unique (up to scaling). .
Hence, we have shown that j is half (non-negative) integer, and the representation is
2j + 1-dimensional. The irreducible representations are therefore completely characterized
by the value of the weight j.
It is possible to go further, and prove the following theorem (the proof given is that
presented in [Samelson]):
Theorem 1. **NON-EXAMINABLE** Suppose that d is a representation of L(SU (2))

on a complex vector space V . Then V can be decomposed as V = V1 ⊕ · · · ⊕ Vp where Vi
are invariant subspaces of V such that d restricted to Vi is irreducible.
Proof Given in Appendix A.
Note that we have not as yet assumed that the representation originates from a unitary
representation of the Lie group. However, in order to compute normalized states, it will
be convenient to assume this; so in particular, the d(Ta ) are antihermitian and hence
J3† = J3 , †
J± = J∓ (3.33)
We will assume that the highest weight state |j, ji is normalized to 1.

Exercise:
– 53 –
Show using (3.30) that hψ` | ψ` i = 2` (2j − ` + 1) hψ`−1 | ψ`−1 i, and hence that
(2j)!`!
N` ≡ hψ` | ψ` i = (3.34)
2` (2j
− `)!
It is conventional to define normalized states |j, mi for m = −j, . . . , j via

1
|j, mi = p |ψj−m i (3.35)
Nj−m
the first label denotes the highest weight value j, the m label denotes the eigenstate
of J3 , J3 |j, mi = m |j, mi. These satisfy (check!)
1 p
J− |j, mi = √ (j + m)(j − m + 1) |j, m − 1i
2
1 p
J+ |j, m − 1i = √ (j + m)(j − m + 1) |j, mi (3.36)
2
Exercise: Show that the Casimir operator C is given by C = − 12 (J+ J− + J− J+ + (J3 )2 )

and satisfies
1
C |j, mi = − j(j + 1) |j, mi (3.37)
2
3.3.1 Examples of Low Dimensional Irreducible Representations
It is useful to consider several low-dimensional representations.
i) j = 12 . A normalized basis is given by | 12 , 12 i and | 21 , − 12 i, with
1 1
J+ | , i = 0
2 2
1 1 1 1 1
J+ | , − i = √ | , i (3.38)
2 2 2 2 2
and
1 1 1 1 1
J− | , i = √ | , − i
2 2 2 2 2
1 1
J− | , − i = 0 (3.39)
2 2
ii) j = 1. A normalized basis of states is |1, 1i, |1, 0i and |1, −1i with
J+ |1, 1i = 0
J+ |1, 0i = |1, 1i
J+ |1, −1i = |1, 0i (3.40)
– 54 –
and
J− |1, 1i = |1, 0i
J− |1, 0i = |1, −1i
J− |1, −1i = 0 (3.41)
iii) j = 23 . A normalized basis of states is | 23 , 32 i, | 32 , 12 i, | 32 , − 21 i and | 32 , − 23 i. with
3 3
J+ | , i =0
2 2 r
3 1 3 3 3
J+ | , i = | , i
2 2 2 2 2
3 1 √ 3 1
J+ | , − i = 2| , i
2 2 r 2 2
3 3 3 3 1
J+ | , − i = | ,− i (3.42)
2 2 2 2 2
and
r
3 3 3 3 1
J− | , i = | , i
2 2 2 2 2
3 1 √ 3 1
J− | , i = 2| ,− i
2 2 r 2 2
3 1 3 3 3
J− | , − i = | ,− i
2 2 2 2 2
3 3
J− | , − i =0 (3.43)
2 2
3.4 Tensor Product Representations

Suppose that d1 and d2 are two irreducible representations of L(SU (2)) over vector spaces
N
V (1), V (2). Let V = V (1) V (2) be the tensor product, and d = d1 ⊗ 1 + 1 ⊗ d2 be the
representation on V .
We wish to decompose V into irreducible representations of d (i.e. invariant subspaces
of V on which the restriction of d is irreducible).
Denote the states of V (1) by |j1 , mi for m = −j1 , . . . , j1 and those of V (2) by |j2 , ni
for n = −j2 , . . . , j2 , where j1 and j2 are the highest weights of V (1) and V (2) respectively.
Note that |j1 , mi ⊗ |j2 , ni for m = −j1 , . . . , j1 and n = −j2 , . . . , j2 provides a basis for V .
Set
i
J3 (1) = id1 (T3 ), J± (1) = √ (d1 (T1 ) ± id1 (T2 )) (3.44)
2
and
i
J3 (2) = id2 (T3 ), J± (2) = √ (d2 (T1 ) ± id2 (T2 )) (3.45)
2
– 55 –
with
J3 = J3 (1) ⊗ 1 + 1 ⊗ J3 (2), J± = J± (1) ⊗ 1 + 1 ⊗ J± (2) (3.46)
Exercise: Check that J± and J3 satisfy J3† = †
J3 , J± = J∓ and
[J3 , J± ] = ±J± , [J+ , J− ] = J3 (3.47)
In order to construct the decomposition of V into irreducible representations, first note
that if |ψ(1)i ∈ V (1) is a state of weight m1 with respect to J3 (1), and |ψ(2)i ∈ V (2) is
a state of weight m2 with respect to J3 (2) then |ψ(1)i ⊗ |ψ(2)i ∈ V is a state of weight
m1 + m2 with respect to J3 , i.e. weights add in the tensor product representation.
Using this, it is possible to compute the degeneracy of certain weight states in the
tensor product representation. In particular, the maximum possible weight must be j1 + j2
which corresponds to |j1 , j1 i ⊗ |j1 , j2 i.
Consider the weight j1 + j2 − k for k > 0. In general, j1 + j2 − k can be written as a
sum of two integers m1 + m2 , for m1 ∈ {−j1 , ..., j1 } and m2 ∈ {−j2 , ..., j2 } in k + 1 ways:
j1 + j2 − k = (j1 − k) + j2
= (j1 − k + 1) + (j2 − 1)
...
= (j1 − 1) + (j2 − k + 1)
= j1 + (j2 − k) (3.48)
provided that j1 − k ≥ −j1 and j2 − k ≥ −j2 , or equivalently
k ≤ 2min(j1 , j2 ) = j1 + j2 − |j1 − j2 | (3.49)

Consider the state of weight j1 + j2 . There is only one such state, and there is no state
of higher weight. Hence it must be a highest weight state of a subspace of V on which
the tensor product representation is irreducible. This irreducible subspace has dimension
2(j1 + j2 ) + 1, and is denoted by Vj1 +j2 Write V = V 0 ⊕ Vj1 +j2 .
Next consider the states in V 0 . The highest possible weight is j1 + j2 − 1. In V there
were two linearly independent states of this weight, however one of these lies in Vj1 +j2 , and
does not lie in V 0 . The remaining state of weight j1 + j2 − 1 is of highest weight in V 0 and
hence must be a highest weight state of a subspace (of dimension 2(j1 + j2 ) − 1) of V 0 , on
which the tensor product representation is irreducible. Denote this subspace by Vj1 +j2 −1 .
Note that by construction Vj1 +j2 ∩ Vj1 +j2 −1 = {0}.
One can continue inductively in this fashion: for each j = |j1 −j2 |, |j1 −j2 |+1, . . . j1 +j2
there is a subspace Vj (of dimension 2j +1) on which the tensor product representation is ir-
reducible, with highest weight j, and Vj ∩Vk = {0} if j 6= k. In fact these subspaces account
L L
for all of the elements in V . To see this, compute the dimension of V|j1 −j2 | · · · Vj1 +j2 :
We find
M M jX
1 +j2
dim V|j1 −j2 | ··· Vj1 +j2 = (2j + 1)

j=|j1 −j2 |
– 56 –
j1 +j2 −|j1 −j2 |
X
= (2(|j1 − j2 | + n) + 1)
n=0
= (1 + 2j1 )(1 + 2j2 )
= dim V (3.50)
L L
Hence we have decomposed V = V|j1 −j2 | · · · Vj1 +j2 into irreducible subspaces Vj
where Vj has highest weight j and is of dimension 2j + 1.
3.4.1 Examples of Tensor Product Decompositions

We shall consider two simple spin combinations which are of physical interest.
Firstly, take the spin 1/2 ⊗ spin 1/2 composite system. As j1 = j2 = 21 there are two
possible values for the composite spin, j = 1 or j = 0, and the tensor product space is
4-dimensional.
For the j = 1 states, the highest weight state can be written as
1 1 1 1
|1, 1i = | , i ⊗ | , i (3.51)
2 2 2 2
Applying J− to both sides we find
1 1 1 1 1 1 1 1 1
|1, 0i = √ | , − i ⊗ | , i + | , i ⊗ | , − i (3.52)
2 2 2 2 2 2 2 2 2
and applying J− once more
1 1 1 1
|1, −1i = | , − i ⊗ | , − i (3.53)
2 2 2 2
The remaining possible spin is j = 0. This has only one state, which must have the
form
1 1 1 1 1 1 1 1
|0, 0i = c0 | , i ⊗ | , − i + c1 | , − i ⊗ | , i (3.54)
2 2 2 2 2 2 2 2
for constants c0 , c1 to be fixed. Applying J+ to both sides we get
1 1 1 1
(c0 + c1 ) | , i ⊗ | , i = 0 (3.55)
2 2 2 2
so c1 = −c0 . Then on making the appropriate normalization we find
1 1 1 1 1 1 1 1 1
|0, 0i = √ | , − i ⊗ | , i − | , i ⊗ | , − i (3.56)
2 2 2 2 2 2 2 2 2
Next, take the spin 1 ⊗ spin 1/2 composite system. As j1 = 1, j2 = 21 there are two
possible values for the composite spin, j = 12 or j = 32 , and the tensor product space is
6-dimensional.
For the j = 32 states, the state with greatest weight is
3 3 1 1
| , i = |1, 1i ⊗ | , i (3.57)
2 2 2 2
– 57 –
Applying J− to both sides gives
r
3 1 2 1 1 1 1 1
| , i= |1, 0i ⊗ | , i + √ |1, 1i ⊗ | , − i (3.58)
2 2 3 2 2 3 2 2
Further applications of J− give

r
3 1 1 1 1 2 1 1
| , − i = √ |1, −1i ⊗ | , i + |1, 0i ⊗ | , − i (3.59)
2 2 3 2 2 3 2 2
and
3 3 1 1
| , − i = |1, −1i ⊗ | , − i (3.60)
2 2 2 2
1
For the j = 2 states, the state with maximal weight can be written as a linear combi-
nation
1 1 1 1 1 1
| , i = c0 |1, 1i ⊗ | , − i + c1 |1, 0i ⊗ | , i (3.61)
2 2 2 2 2 2
for some constants c0 , c1 to be determined. Then
1 1 1 1 1
0 = J+ | , i = (c1 + √ c0 ) |1, 1i ⊗ | , i (3.62)
2 2 2 2 2
so c1 = − √12 c0 . On making a unit normalization, we also fix c0 and find

r
1 1 2 1 1 1 1 1
| , i=− |1, 1i ⊗ | , − i + √ |1, 0i ⊗ | , i (3.63)
2 2 3 2 2 3 2 2
and applying J− to both sides this gives

r
1 1 1 1 1 2 1 1
| , − i = − √ |1, 0i ⊗ | , − i + |1, −1i ⊗ | , i (3.64)
2 2 3 2 2 3 2 2
3.5 SU (2) weight diagrams

One can plot the weights of an irreducible L(SU (2)) representation on the real line; for
example:
The spin j=4 SU(2) irreducible representation
−6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6
The weight diagrams of irreducible representations have the following properties
i) The diagram is reflection symmetric about the origin.
ii) The weights are all half integer; and are distributed with unit distance between each
weight. The highest weight is j for 2j ∈ N, and the lowest weight is −j, and there
are no “holes” in the weight diagram- −j, −j + 1, . . . , j − 1, j are all weights.
– 58 –
iii) Each weight has multiplicity 1.
One can also plot the weight diagram of a generic (not necessarily irreducible) represen-
tation. For example, the weight diagram of the tensor product (j1 = 12 )⊗(j2 = 21 )⊗(j3 = 12 )
eight dimensional tensor product representation is
−3 −2 −1 0 1 2 3
This has a highest weight j = 32 and a lowest weight − 32 both with multiplicity 1, and
weights ± 12 each with multiplicity 3. In general, a non-irreducible representation will have
a highest weight, but it need not be of multiplicity 1. For a generic weight diagram
i) The diagram (together with weight multiplicities) is reflection symmetric about the
origin.
ii) The weights are all half-integer.
iii) Each weight need not be of multiplicity 1. However, as one proceeds from a particular
weight towards the origin (in unit steps from either the left or the right), the weight
multiplicities do not decrease.
3.6 SU(2) in Particle Physics

3.6.1 Angular Momentum
The orbital angular momentum operators La acting on wavefunctions are given by
∂
La = −iabc xb (3.65)
∂xc
These operators satisfy
[La , Lb ] = iabc Lc (3.66)
and hence correspond to a (complexified) representation of SU (2). Particles also carry a
spin angular momentum S, which commutes with the orbital angular momentum [L, S] = 0.
The total angular momentum is defined by J = L + S.
3.6.2 Isospin Symmetry

It is observed that the proton and neutron have similar mass, and also that the strong
nuclear forces between nucleons are similar. Heisenberg introduced the concept of a SU (2)
isospin symmetry to systematize this. Particles are grouped into multiplets of isospin
value I (previously called j) and labelled by the weights, which are the eigenvalues of I3 .
Originally, this was formulated for nucleons, but later extended to describe all mesons and
baryons.
Particles in the same isospin multiplet have the same baryon number, the same content
of non-light quarks, the same spin and parity and almost the same mass. Isospin is a
conserved quantum number in all known processes involving only strong interactions: it is
related to the quark content by
– 59 –
1 ¯
I3 = (N (u) − N (ū) − (N (d) − N (d))) (3.67)
2
Isospin symmetry arises in the quark model because of the very similar properties of
the u and d quarks.
Examples:
i) Nucleons have isospin I = 12 ; the proton has I3 = 12 , and the neutron has I3 = − 12 :
1 1
n = | ,− i
2 2
1 1
p=| , i (3.68)
2 2
ii) The pions have I = 1 with
π − = |1, −1i
π 0 = |1, 0i
π + = |1, 1i (3.69)
iii) The strange baryons have I = 0 and I = 1
Σ− = |1, −1i
Σ0 = |1, 0i
Σ+ = |1, 1i (3.70)
and
Λ0 = |0, 0i (3.71)
1
iv) The strange mesons lie in two multiplets of I = 2
1 1
K0 = | , − i
2 2
1 1
K+ = | , i (3.72)
2 2
and
1 1
K− = | , − i
2 2
0 1 1
K̄ = | , i (3.73)
2 2
K ± are antiparticles with the same mass, but are in different isospin multiplets
because of their differing quark content.
– 60 –
1
v) The light quarks have I = 2
1 1
d = | ,− i
2 2
1 1
u=| , i (3.74)
2 2
and all other quarks are isospin singlets I = 0.
3.6.3 Pauli’s Generalized Exclusion Principle and the Deuteron

Consider first NN nucleon-nucleon bound states. From the previously obtained decompo-
sition of the (I1 = 1/2) ⊗ (I2 = 1/2) tensor product we find the following isospin states
1
|1, 1i = pp, |1, 0i = √ (np + pn), |1, −1i = nn (3.75)
2
which are are symmetric under exchange of isospin degrees of freedom, and the remaining
state is
1
|0, 0i = √ (np − pn) (3.76)
2
which is anti-symmetric under exchange of isospin degrees of freedom.
The deuteron d is a pn bound state, which has no pp or nn partners. There is therefore
only one possibility; d = |0, 0i.
In general, the total wavefunction for a NN state can be written as a product of space,
spin, and isospin functions
ψ = ψ(space)ψ(spin)ψ(isospin) (3.77)
The generalized Pauli exclusion principle requires allowed N N wave-functions be anti-

symmetric under exchange of all degrees of freedom. As ψ(isospin) is anti-symmetric,
ψ(space)ψ(spin) must be symmetric- in fact ψ(space) is symmetric (the nucleons in the d
are in a ` = 0 angular momentum state), and ψ(spin) as also symmetric, as d is a spin 1
particle.
3.6.4 Pion-Nucleon Scattering and Resonances

Isospin can be used to investigate scattering processes; consider for example pion-nucleon
scattering processes. From the decomposition of the (I1 = 1) ⊗ (I2 = 1/2) tensor product
we find
3 3
| , i = π+p
2 2 r
3 1 2 0 1
| , i= π p + √ π+n
2 2 3 r3
3 1 1 − 2 0
| ,− i = √ π p + π n
2 2 3 3
– 61 –
3 3
| , − i = π−n (3.78)
2 2
and
r
1 1 1 0 2 +
| , i = √ π p− π n
2 2 r3 3
1 1 2 − 1
| ,− i = π p − √ π0n (3.79)
2 2 3 3
These equations can be inverted to give
3 3
π+p = | , i
r 2 2
0 2 3 1 1 1 1
π p= | , i+ √ | , i
3 2 2 r3 2 2
1 3 1 2 1 1
π+n = √ | , i − | , i
3 2 2 r3 2 2
1 3 1 2 1 1
π−p = √ | , − i + | ,− i
r3 2 2 3 2 2
2 3 1 1 1 1
π0n = | ,− i − √ | ,− i
3 2 2 3 2 2
3 3
π−n = | , − i (3.80)
2 2
Consider πN scattering. The scattering is described by a S-matrix, which, in processes
dominated by strong interactions is taken to be isospin invariant: [Ij , S] = 0 and so by
Schur’s lemma
hI 0 , m0 | S |I, mi = φ(I)δII 0 δmm0 (3.81)
The cross sections of the scattering processes are given by
σ(in → out) = K| hin| S |outi |2 (3.82)
for constant K. Hence
3
σ(π + p → π + p) = K|φ( )|2 (3.83)
2
2 3 1
σ(π 0 n → π − p) = K|φ( ) − φ( )|2 (3.84)
9 2 2
1 3 1
σ(π − p → π − p) = K|φ( ) + 2φ( )|2 (3.85)
9 2 2
For all three of these processes, a marked resonance is measured at approximately 1236
Mev. The ratio of the cross-sections of these resonances is
2 1
σ(π + p → π + p) : σ(π 0 n → π − p) : σ(π − p → π − p) = 1 : : (3.86)
9 9
which is consistent with the supposition that the resonance corresponds to a particle of
isospin 23 (and so |φ( 32 )| |φ( 12 )|). This particle is the ∆ particle, which lies in an
isospin I = 32 multiplet with states ∆− , ∆0 , ∆+ , ∆++ having weights I3 = − 32 , − 12 , 12 , 32
respectively.
– 62 –
3.7 The semi-simplicity of (complexified) L(SU (n + 1))
To proceed, define the n + 1 × n + 1 square matrices Eij by (Ei,j )pq = δip δjq . Recall that
L(SU (n+1)) consists of the traceless antihermitian matrices, and is therefore spanned over
R by i(Ei,i − En+1,n+1 ) for i = 1, . . . , n and Ei,j − Ej,i , i(Ei,j + Ej,i ) for i < j. Hence, on
complexification, the Lie algebra is spanned by traceless diagonal matrices together with
the Ei,j for i 6= j.
P
Suppose that h = diag (a1 , a2 , . . . , an+1 ), ai = 0 is a traceless diagonal matrix.
Then if i 6= j, observe that [h, Ei,j ] = (ai − aj )Ei,j .
Proposition 25. Complexified L(SU (n + 1)) is semi-simple.

Proof
Note by direct computation that
[Ei,j , Er,s ] = δjr Ei,s − δis Er,j (3.87)
so
(ad Ep,q ad Ei,j )Er,s = δjr δiq Ep,s − δjr δps Ei,q − δis δqr Ep,j + δis δpj Er,q (3.88)
and hence the component of (ad Ep,q ad Ei,j )Er,s in the E`,t direction is
δjr δiq δp` δst − δjr δps δi` δqt − δis δqr δp` δjt + δis δpj δr` δqt (3.89)
There are various cases to consider
i) κ(Ep,q , Ei,j ) for p 6= q, i 6= j. We must compute Tr (ad Ep,q ad Ei,j ). If r 6= s then

the component of (ad Ep,q ad Ei,j )Er,s in the Er,s direction is
δjr δiq δpr − δjr δps δir δqs − δis δqr δpr δjs + δis δpj δqs = δjr δiq δpr + δis δpj δqs (3.90)
(as i 6= j and p 6= q). So the contribution to the trace from these terms is
n+1
X
δjr δiq δpr + δis δpj δqs = 2nδiq δpj (3.91)
r=1,s=1,r6=s
We also compute the component of (ad Ep,q ad Ei,j )Er,s in the direction Ek,k −
En+1,n+1 for k = 1, . . . , n. Observe that the component of δiq Ep,s − δps Eiq in this
direction is δiq δps (δpk − δik ) (if i 6= q or p 6= s then the diagonal components of
δiq Ep,s −δps Eiq vanish). Hence the component of (ad Ep,q ad Ei,j )Er,s in the direction
Ek,k − En+1,n+1 for k = 1, . . . , n is
δjr δiq δps (δpk − δik ) + δis δpj δrq (δrk − δpk ) (3.92)
It follows that the component of (ad Ep,q ad Ei,j )(Ek,k − En+1,n+1 ) along (Ek,k −
En+1,n+1 ) is
δjk δiq δpk (δpk − δik ) + δik δpj δqk (1 − δpk )
– 63 –

− δj,n+1 δiq δp,n+1 (δpk − δik ) − δi,n+1 δpj δq,n+1 δpk (3.93)
which can be simplified using i 6= j and p 6= q to
δjk δiq δpk + δik δpj δqk + δj,n+1 δiq δp,n+1 δik + δi,n+1 δpj δq,n+1 δpk (3.94)
On taking the sum from k = 1, . . . , n this gives a contribution to the trace of
2δiq δpj (3.95)
and so
κ(Ep,q , Ei,j ) = 2(n + 1)δiq δpj (3.96)

P
ii) κ(Ep,q , h) for h = diag (a1 , a2 , . . . , an+1 ) with i ai = 0 and p 6= q. The only
contribution to the trace Tr (ad Ep,q ad h) comes from terms
(ad Ep,q ad h)Ei,j = ad Ep,q (ai − aj )Ei,j = (ai − aj )(δqi Ep,j − δp,j Ei,q ) (3.97)
for i 6= j. The component of this matrix along the Ei,j direction is
(ai − aj )(δqi δpi − δp,j δq,j ) = 0 (3.98)
as p 6= q. Hence
κ(Ep,q , h) = 0 (3.99)
P
iii) κ(h, g) where h = diag (a1 , a2 , . . . , an+1 ) and g = diag (b1 , b2 , . . . , bn+1 ) and i ai =
P
i bi = 0,
The only contribution to the trace Tr (ad h ad g) is from the terms
(ad h ad g)Ei,j (3.100)
for i 6= j. But
(ad h ad g)Ei,j = (ai − aj )(bi − bj )Ei,j (3.101)
so taking the sum over i and j (i 6= j) we obtain
X
κ(h, g) = 2(n + 1) ai bi (3.102)
i
Hence κ is negative definite over the span over R of i(Er,r − En+1,n+1 ) for r = 1, . . . , n;
and this span is orthogonal to the span of the Ei,j − Ej,i and i(Ei,j + Ej,i ) (i 6= j).
Furthermore, κ is diagonal and negative definite over the span over R of the Ei,j − Ej,i and
i(Ei,j + Ej,i ). κ is therefore non-degenerate.
We also have the immediate corollory:
Corollory 2. The Lie algebra of SU (n + 1) (as a real Lie algebra) is compact.
– 64 –
4. SU(3) and the Quark Model
The Lie algebra of SU (3) consists of the traceless antihermitian 3 × 3 complex matrices.
It is convenient to define the following matrices
   1 
1
0 0 √
2 3
0 0
2
1
h1 =  0 − 21 0 h2 =  0 √ 0 
  
2 3 
0 0 0 1
0 0 − √3
 1
  
0 √
2
0 0 0 0
1 1  √1 0 0
e+ =  0 e− =  2
 
0 0 
0 0 0 0 0 0
   
0 0 0 0 0 0
e2+ √1  2
 0 0  0 0 0
= e− = 

2 
0 0 0 0 √1 0
   2 
1
0 0 √
2
0 0 0
e3+ 3  0 0 0
= 0 e− =  (4.1)
 
0 0 
0 0 0 √1 0 0
2
Then ih1 , ih2 and i(em m m m

+ +e− ), e+ −e− for m = 1, 2, 3 form a basis for the antihermitian
traceless 3 × 3 matrices (over R), and hence are a basis for L(SU (3)). Suppose that d is
the irreducible representation of L(SU (3)) acting on a complex vector space V which is
induced from an irreducible representation of SU (3) acting on V .
It is convenient to set
m
H1 = d(h1 ), H2 = d(h2 ), E± = d(em
± ) for m = 1, 2, 3 (4.2)
Then we find the following commutators:
[H1 , H2 ] = 0
1 1 2 1 2 3 1 3
[H1 , E± ] = ±E± , [H1 , E± ] = ∓ E± , [H1 , E± ] = ± E±
2√ 2
√
1 2 3 2 3 3 3
[H2 , E± ] = 0, [H2 , E± ]=± E , [H2 , E± ] = ± E (4.3)
2 ± 2 ±
and
1 1
[E+ , E− ]=H√1
2 2 3 1
[E+ , E− ]= H2 − H1
√2 2
3 3 3 1
[E+ , E− ]= H2 + H1 (4.4)
2 2
The remaining commutation relations are
1 2 1 3 1 2 1 3
[E+ , E+ ] = √ E+ , [E− , E− ] = − √ E−
2 2
– 65 –
1 3 1 2 1 3 1 2
[E+ , E− ] = − √ E− , [E− , E+ ] = √ E+
2 2
2 3 1 1 2 3 1 1
[E+ , E− ] = √ E− , [E− , E+ ] = − √ E+ (4.5)
2 2
and
1 2 1 2 1 3 1 3 2 3 2 3
[E+ , E− ] = [E− , E+ ] = [E+ , E+ ] = [E− , E− ] = [E+ , E+ ] = [E− , E− ]=0 (4.6)
Note in particular that H1 , H2 commute. The subalgebra of L(SU (3)) spanned by

ih1 and ih2 is called the Cartan subalgebra. It is the maximal commuting subalgebra of
L(SU (3)).
4.1 Raising and Lowering Operators: The Weight Diagram

The Lie algebra of L(SU (3)) can be used to obtain three sets of L(SU (2)) algebras. In
particular, we find that
1 1 1 1
[H1 , E± ] = ±E± , [E+ , E− ] = H1 (4.7)
and √ √
3 1 2 2 2 2 3 1
[ H2 − H1 , E± ] = ±E± , [E+ , E− ] = H2 − H1 (4.8)
2 2 2 2
and √ √
3 1 3 3 3 3 3 1
[ H2 + H1 , E± ] = ±E± , [E+ , E− ] = H2 + H1 (4.9)
2 2 2 2
In particular, there are three pairs of raising and lowering operators E± m.
For simplicity, consider a representation d of L(SU (3)) obtained from a unitary repre-
sentation D of SU (3) such that d is an anti-hermitian representation- so that√H1 and H2
are hermitian, and hence diagonalizable with real eigenvalues. Hence, H1 and 23 H2 ± 12 H1 ,
can be simultaneously diagonalized, and the eigenvalues are real. (In fact the same can be
shown without assuming unitarity!)
Suppose then that |φi is an eigenstate of H1 with eigenvalue p and also an eigenstate
of H2 with eigenvalue q. It is convenient to order the eigenvalues as points in R2 with
position vectors (p, q) where p is the eigenvalue of H1 and q of H2 . (p, q) is then referred
to as a weight.
From the commutation relations we have the following properties
1 |φi = 0 or E 1 |φi is an eigenstate of H and H with eigenvalue (p, q)±(1, 0)

i) Either E± ± 1 2
√
2 |φi = 0 or E 2 |φi is an eigenstate with eigenvalue (p, q) ± (− 1 , 3
ii) Either E± ± 2 2 )
√
3 |φi = 0 or E 3 |φi is an eigenstate with eigenvalue (p, q) ± ( 1 , 3
iii) Either E± ± 2 2 )
Moreover, from the properties of L(SU (2)) representations we know that

√ √
2p = m1 , 3q − p = m2 , 3q + p = m3 (4.10)
– 66 –
√
for m1 , m2 , m3 ∈ Z. It follows that 2 3q ∈ Z. It is particularly useful to plot the
sets of eigenvalues (p, q) as points in the plane. The resulting plot is known as the weight
diagram. As the representation is assumed to be irreducible, there can only be finitely
many points on the weight diagram, though it is possible that a particular weight may
√
correspond to more than one state. Moreover, as 2p ∈ Z, 2 3q ∈ Z, the weights are
constrained to lie on the points of a lattice. From the effect of the raising and lowering
operators on the eigenvalues, it is straightforward to see that this lattice is formed by the
tessalation of the plane by equilateral triangles of side 1. This is illustrated in Figure 1,
where the effect of the raising and lowering operators is given (in this diagram (0, 0) is a
weight, though this need not be the case generically).
3
2 E+
E+
E+1
E−1
2
E−
3
E−
The weight diagram has three axes of symmetry. To see this, recall that if m is a weight
of a state in an irreducible representation of L(SU (2)) then so is −m. In the context of
the three L(SU (2)) algebras contained in L(SU (3)) this means that from the properties
of the algebra in (4.7), if (p, q) is a weight then so is (−p, q), i.e. the diagram is reflection
symmetric about the line θ = π2 passing through the origin. Also, due to the symmetry of
the L(SU (2)) algebra in (4.8), the weight diagram is reflection symmetric about the line
√ √
θ = π6 passing through the origin: so if (p, q) is a weight then so is ( 21 (p+ 3q), 21 ( 3p−q)).
And due to the symmetry of the L(SU (2)) algebra in ((4.9) the weight diagram is reflection
symmetric about the line θ = 5π 6 passing through the origin: so if (p, q) is a weight then so
1
√ 1
√
is ( 2 (p − 3q), 2 (− 3p − q)).
Using this symmetry, it suffices to know the structure of the weight diagram in the
sector of the plane π6 ≤ θ ≤ π2 . The remainder is fixed by the reflection symmetry.
Motivated by the treatment of SU (2) we make the definition:
Definition 42. |ψi is called a highest weight state if |ψi is an eigenstate of both H1 and
H2 , and E+m |ψi = 0 for m = 1, 2, 3.
Note that there must be a highest weight state, for otherwise one could construct
infinitely many eigenstates by repeated application of the raising operators E+ m . Given
– 67 –
a highest weight state, let V 0 be the vector space spanned by |ψi and states obtained by
acting with all possible products of lowering operators E−m on |ψi. As there are only finitely
many points on the weight diagram, there can only be finitely many such terms. Then, by
making use of the commutation relations, it is clear that V 0 is an invariant subspace of V .
As the representation is irreducible on V , this implies that V 0 = V , i.e. V is spanned by
|ψi and a finite set of states obtained by acting with lowering operators on |ψi. Suppose
that (p, q) is the weight of |ψi. Then V is spanned by a basis of eigenstates of H1 and H2
with weights confined to the sector given by π ≤ θ ≤ 5π 3 relative to (p, q)- all points on the
weight diagram must therefore lie in this sector.
Lemma 6. The highest weight state is unique.

Proof Suppose that |ψi and |ψ 0 i are two highest weight states with weights (p, q), (p0 , q 0 )
respectively. Then (p0 , q 0 ) must make an angle π ≤ θ ≤ 5π
3 relative to (p, q) and (p, q) must
make an angle π ≤ θ ≤ 3 relative to (p , q ). This implies that p = p0 , q = q 0 .
5π 0 0
Next suppose that |ψ1 i and |ψ2 i are two linearly independent highest weight states
(both with weight (p, q)). Let V1 and V2 be the vector spaces spanned by the states
obtained by acting with all possible products of lowering operators E− m on |ψ i and |ψ i
1 2
respectively; one therefore obtains bases for V1 and V2 consisting of eigenstates of H1 and
H2 . By the reasoning given previously, as V1 and V2 are invariant subspaces of V and the
representation is irreducible on V , it must be the case that V1 = V2 = V . In particular, we
find that |ψ2 i ∈ V1 . However, the only basis element of V1 which has weight (p, q) is |ψ1 i,
hence we must have |ψ2 i = c |ψ1 i for some constant c, in contradiction to the assumption
that |ψ1 i and |ψ2 i are linearly independent.
Having established the existence of a unique highest weight state |ψi, we can proceed
to obtain the generic form for the weight diagram. Recall that the highest weight j of an
irreducible representation of L(SU (2)) is always non-negative. By acting on |ψi with the
lowering operators E− m , one obtains three irreducible representations of L(SU (2)). Non-
negativity of the highest weight corresponding to the L(SU (2)) irreducible representation
generated by E− 1 implies that the highest weight must lie in the half-plane to the right of
the line θ = π2 , or on the line θ = π2 :
Non-negativity of the highest weight corresponding to the L(SU (2)) irreducible represen-
tation generated by E− 2 implies that the highest weight must lie in the half-plane above
the line θ = π6 , or on the line θ = π6 :
– 68 –
Finally, non-negativity of the highest weight corresponding to the L(SU (2)) irreducible
representation generated by E− 3 implies that the highest weight must lie in the half-plane
above the line θ = 5π 5π

6 , or on the line θ = 6 :
As the highest weight must lie in all three of these regions, it must lie in the sector
π
6 ≤ θ ≤ π2 relative to (0, 0), or at the origin:
Lemma 7. If the highest weight is (0, 0), then there is only one state in the representation,
which is called the singlet.
Proof
Let |ψi be the highest weight state with weight (0, 0). Suppose that E− m |ψi 6= 0 for
some m. Then by the reflection symmetry of the weight diagram, it follows that E+ m |ψi =
6 0,
i
in contradiction to the fact that E+ |ψi = 0 for i = 1, 2, 3, as |ψi is the highest weight
state. Hence E± m |ψi = 0 for m = 1, 2, 3. Also H |ψi = H |ψi = 0. It follows that the
1 2
0
1-dimensional subspace V spanned by |ψi is an invariant subspace of V , and therefore
V = V 0 as the representation is irreducible.
There are then three possible locations for the highest weight state |ψi.
4.1.1 Triangular Weight Diagrams (I)

Suppose that the highest weight lies on the line θ = π2 . In this case, by applying powers of
E−2 the states of the L(SU (2)) representation given in (4.8) are generated. These form a
line orthogonal to the axis of reflection θ = π6 , about which they are symmetric, and there
– 69 –
are no states outside this line, as these points cannot be reached by applying lowering
operators. Then, by using the reflection symmetry, it follows that the outermost states
from an equilateral triangle with horizontal base. Each lattice point inside the triangle
corresponds to (at least) one state which has this weight, because each lattice point in the
triangle lies at some possible weight within the L(SU (2)) representation given in (4.7),
and from the properties of L(SU (2)) representations, we know that this has a state with
this weight (i.e. as the L(SU (2)) weight diagram has no “holes” in it, neither does the
L(SU (3)) weight diagram).
This case is illustrated by
Proposition 26. Each weight in this triangle corresponds to a unique state.

Proof
Note that all of the states on the right edge of the triangle correspond to unique states,
because these weights correspond to states which can only be obtained by acting on |ψi
with powers of E−2 . It therefore follows by the reflection symmetry that all of the states on
the edges of the triangle have multiplicity one.

Now note the commutation relation
1 2 1 3
[E− , E− ] = − √ E− (4.11)
2
This implies that products of lowering operators involving E−3 can be rewritten as
linear combinations of products of operators involving only E−1 and E 2 (in some order).
−
In particular, we find
1 2 n 1 2 2 n−1 2 1 2 n−1
(E− )(E− ) |ψi = [E− , E− ](E− ) |ψi + E− E− (E− ) |ψi
1 3 2 n−1 2 1 2 n−1
= − √ E− (E− ) |ψi + E− E− (E− ) |ψi
2
...
– 70 –
n 3 2 n−1
= − √ E− (E− ) |ψi (4.12)
2
1 |ψi = 0 and [E 2 , E 3 ] = 0.
by simple induction, where we have used the fact that E− − −
A generic state of some fixed weight in the representation can be written as a linear
2 and E 1 lowering operators acting on |ψi of the form
combination of products of E− −
1 2
Π(E− , E− ) |ψi (4.13)
where Π(E−1 , E 2 ) contains m powers of E 2 and ` powers of E 1 where m, ` are uniquely

− − −
determined by the weight of the state- only the order of the operators is unfixed.
Using (4.12), commute the E− 1 states in this product to the right as far as they will
1 acting directly on |ψi), or
go. Then either one finds that the state vanishes (due to an E−
1
one can eliminate all of the E− terms and is left with a term proportional to
2 m−` 3 `
(E− ) (E− ) |ψi (4.14)
2 , E 3 ] = [E 1 , E 3 ] = 0.
where we have used the commutation relations [E− − − −
Hence, it follows that all weights in the diagram can have at most multiplicity 1.
However, from the property of the L(SU (2)) representations, as the weights in the outer
layers have multiplicity 1, it follows that all weights in the interior have multiplicity at
least 1.
Hence, all the weights must be multiplicity 1.
4.1.2 Triangular Weight Diagrams (II)
Suppose that the highest weight lies on the line θ = π6 . In this case, by applying powers of
E−1 the states of the L(SU (2)) representation given in (4.7) are generated. These form a
horizontal line orthogonal to the axis of reflection θ = π2 , about which they are symmetric,
and there are no states outside this line, as these points cannot be reached by applying
lowering operators. Then, by using the reflection symmetry, it follows that the outermost
states from an inverted equilateral triangle with horizontal upper edge. Each lattice point
inside the triangle corresponds to (at least) one state which has this weight, because each
lattice point in the triangle lies at some possible weight within the L(SU (2)) representation
given in (4.7), and from the properties of L(SU (2)) representations, we know that this has
a state with this weight (i.e. as the L(SU (2)) weight diagram has no “holes” in it, neither
does the L(SU (3)) weight diagram).
– 71 –
Proposition 27. Each weight in this triangle corresponds to a unique state.
Proof
Note that all of the states on the horizontal top edge of the triangle correspond to
unique states, because these weights correspond to states which can only be obtained by
acting on |ψi with powers of E− 1 . It therefore follows by the reflection symmetry that all
of the states on the edges of the triangle have multiplicity one.

Now, using (4.11) it is straightforward to show that
2 1 n n 3 1 n−1
E− (E− ) |ψi = √ E− (E− ) |ψi (4.15)
2
for n ≥ 1, where we have used E− 2 |ψi = 0.
Now consider a state of some fixed weight in the representation; this can be written as
a linear combination of terms of the form
1 2
Π(E− , E− ) |ψi (4.16)
where Π(E− 1 , E 2 ) contains m powers of E 1 and ` powers of E 2 in an appropriate order,

− − −
where m and ` are determined uniquely by the weight of the state in question.
Using (4.15), commute the E− 2 states in this product to the right as far as they will
go. Then either one finds that the state vanishes (due to an E−2 acting directly on |ψi), or
one can eliminate all of the E−1 terms and is left with a term proportional to
1 m−` 3 `
(E− ) (E− ) |ψi (4.17)
where we have used the commutation relations [E− 2 , E 3 ] = [E 1 , E 3 ] = 0.

− − −
Hence, it follows that all weights in the diagram can have at most multiplicity 1.
However, from the property of the L(SU (2)) representations, as the weights in the outer
layers have multiplicity 1, it follows that all weights in the interior have multiplicity at
least 1.
Hence, all the weights must be multiplicity 1.
– 72 –
4.1.3 Hexagonal Weight Diagrams
Suppose that the highest weight lies in the sector π6 < θ < π2 . In this case, by applying
powers of E−1 the states of the L(SU (2)) representation given in (4.7) are generated. These
form a horizontal line extending to the left of the maximal weight which is orthogonal
to the line θ = π2 , about which they are symmetric, There are no states above, as these
points cannot be reached by applying lowering operators. Also, by applying powers of E− 2
the states of the L(SU (2)) representation given in (4.8) are generated. These form a line
extending to the right of the maximal weight which is orthogonal to the axis of reflection
θ = π6 , about which they are symmetric, and there are no states to the right of this line,
as these points cannot be reached by applying lowering operators.
Then, by using the reflection symmetry, it follows that the outermost states form a
hexagon. Each lattice point inside the hexagon corresponds to (at least) one state which
has this weight, because each lattice point in the hexagon lies at some possible weight
within the L(SU (2)) representation given in (4.7), and from the properties of L(SU (2))
representations, we know that this has a state with this weight (i.e. as the L(SU (2)) weight
diagram has no “holes” in it, neither does the L(SU (3)) weight diagram).
The multiplicities of the states for these weight diagrams are more complicated than
for the triangular diagrams. In particular, the weights on the two edges of the hexagon
leading off from the highest weight have multiplicity 1, because these states can only be
constructed as (E− 1 )n |ψi or (E 2 )m |ψi. So by symmetry, all of the states on the outer layer
−
of the hexagon have multiplicity 1. However, if one proceeds to the next layer, then the
multiplicity of all the states increases by 1. This happens until the first triangular layer is
reached, at which point all following layers have the same multiplicity as the first triangular
layer.
– 73 –
Suppose that the top horizontal edge leading off the maximal weight is of length m,
and that the other outer edge is of length n, with m ≥ n. This situation is illustrated
below
11111111
00000000
00000000
11111111
11111111111111111111111111111
00000000000000000000000000000
m 00000000
11111111
00000000
11111111
00000000
11111111
00000000
11111111
0000
1111
0000
1111
0000
1111
00001111
1111
0000
1111
00001111
00001111
11110000
00001111
00001111
111100000000
00001111
11110000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00000000
11111111
n
00000000
11111111
0000
1111
0000
11110000
11110000
1111
00001111
1111 0000
1111
000011110000
1111
00001111
00000000
1111
0000
1111
0000
1111
0000
1111 00000000
11111111
0000 11111111
0000 00000000
0000
111100001111
111100001111
00001111
00000000
11110000
1111
0000 111
000 000
111 000
111 000
111 000
111 000
111
0000
1111 1111
1111
0000
1111
0000
1111 000
111 000
111 000
111 000
111 000
111 000
111
1111
0000
1111
0000
1111 00000000
11111111
00000000
11111111
0000 111
0000
1111
000 000
111
0000
1111 000
111
0000
1111 000
111
0000
1111 000
111
0000
1111 0001111
111
0000
1111 0000
1111
0000
1111
000
1110000
11110000
11110000
11110000
11110000
1111
0000
1111
000
111 00000000
11111111
0000
1111
0000
1111
0000 111
0000
1111
1111 0001111
00000000
11110000
11110000
11110000
1111 00000000
11111111
0000
1111
0001111
1110000
1111
0000
1111
0000
1111
0000
1111 000
1110000
11110000
1111
0000
1111 0000
1111
0000
1111 0000
1111
0000
1111 0000
1111
0000
1111 000
1110000
0000
1111
0000 111
000
0000
1111
0000
1111 000
1110000
1111
00001111
0000
00001111
0000
00001111
0000
0000
1111 0000
1111
1111
0000
1111 000
111
0000 111
0000
1111
0001111111111110000 111
1111 0001111
0000
0000
1111
0000 1111
1111
0000
0000 111
000 0000
1111
0000
1111 0000
1111
0000
1111 0000
1111
0000
1111 000
1110000
1111
0000
1111
1111
1111
0000
1111
000
1110000
11110000
11110000
1111 000
111
000
111
0000
1111
0000
1111
0000
1111
0001111
00000000
11110000
1111
0000
1111 0000
1111
0000 111
0000
1111
0000
11110000
1111
0000
1111 0000 111
1111 0000000
1111
1111
0000
1111
000
1110000
11110000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
000
1110000
11110000
1111 000
111
0000
1111
0000
1111
0000
1111
0000
1111
0001111
11100000000
1111
0000
1111 000
111
0000
1111
0000
1111
0000
1111 0000
1111 0000
1111
0000 111
000 000
111
0000
1111
0000
1111
1111
000
1110000
1111 000
111
0000
1111
0000
1111
0001111
0000
0000
1111 0000
1111
0000
1111
0000 1110000 111
000
0000
1111
1111 1111
0000 111
0000
1111
0000
1111
0000
1111
1111 000 000
1110000
1111
0000
1111
0000
1111
0000
1111 000
111
0000 111
0000
1111
1111 000 000
111
000
111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00001111
0000
00001111
0000
111111110000
1111
1111
0000
0000
11110000
1111
0000
11110000
1111
0000
1111
0000
1111
00001111
0000
00001111
0000
111111110000
1111
The highest weight is then at ( m 1

2 , 2 3 (m + 2n)). The outer n layers are hexagonal,
√
whereas the n + 1-th layer is triangular, and all following layers are also triangular. As
one goes inwards through the outer n + 1 layers the multiplicity of the states in the layers
increases from 1 in the first outer layer to n + 1 in the n + 1-th layer. Then all the states
in the following triangular layers have multiplicity n + 1 as well.
We will prove this in several steps.
Proposition 28. States with weights on the k-th hexagonal layer for k = 1, . . . , n or the
k = n + 1-th layer (the first triangular layer) have multiplicity not exceeding k.
Proof
In order to prove this, consider first a state on the upper horizontal edge of the k-th
layer for k ≤ n + 1. The length of this edge is m − k + 1. A general state on this edge is
obtained via
2 1
Π(E− , E− ) |ψi (4.18)
where Π(E− 2 , E 1 ) contains (in some order) k − 1 powers of E 2 and ` powers of E 1 for
− − −
` = k − 1, . . . , m.
Now use the commutation relation (4.11) to commute the powers of E− 2 to the right as
far as they will go. Then the state can be written as a linear combination of the k vectors
3 i−1 1 `−i+1 2 k−i
|vi i = (E− ) (E− ) (E− ) |ψi (4.19)
for i = 1, . . . , k.
It follows that this state has multiplicity ≤ k.
– 74 –
Next consider a state again on the k-th level, but now on the edge leading off to the
right of the horizontal edge which we considered above; this edge is parallel to the outer
edge of length n. Take k ≤ n + 1, so the edge has length n − k + 1. A state on this edge is
obtained via
1 2
Π̂(E− , E− ) |ψi (4.20)
where Π̂(E− 1 , E 2 ) contains (in some order) k − 1 powers of E 1 and ` powers of E 2 where
− − −
` = k − 1, . . . , n. Now use the commutation relation (4.11) to commute the powers of E− 1
to the right as far as they will go. Then the state can be written as a linear combination
of the k vectors
3 i−1 2 `−i+1 1 k−i
|wi i = (E− ) (E− ) (E− ) |ψi (4.21)
for i = 1, . . . , k.
So these states also have multiplicity ≤ k. By using the reflection symmetry, it follows
that the all the states on the k-th hexagonal layer have multiplicity k.
We also have the
Proposition 29. The states with weights in the triangular layers have multiplicity not
exceeding n + 1.
Proof
Consider a state on the k-th row of the weight diagram for m + 1 ≥ k ≥ n + 1 which
lies inside the triangular layers. Such a state can also be written as
2 1
Π(E− , E− ) |ψi (4.22)
where Π(E−2 , E 1 ) contains (in some order) k − 1 powers of E 2 and ` powers of E 1 for
− − −
` = k − 1, . . . , m. and hence by the reasoning above, it can be rewritten as a linear
2 )k−i |ψi = 0.
combination of the k vectors |vi i in (4.19), however for i < k−n, |vi i = 0 as (E−
The only possible non-vanishing vectors are the n + 1 vectors |vk−n i , |vk−n+1 i , . . . , |vk i.
Hence these states have multiplicity ≤ n + 1.
Next note the lemma
Lemma 8. Define |wi,k i = (E− 3 )i−1 (E 1 )k−i (E 2 )k−i |ψi for i = 1, . . . , k, k = 1, . . . , n + 1.

− −
Then the sets Sk = {|w1,k i , . . . , |wk,k i} are linearly independent for k = 1, . . . , n + 1.
Proof
By using the commutation relations, it is straightforward to prove the identities
√
3 3 1 i
E+ |wi,k i = (i − 1) q + p + + 1 − k |wi−1,k−1 i
2 √2 2
1 3 1 i 1 k
− √ (k − i)2 q − p + + − |wi,k−1 i
2 2 2 2 2 2
2 1 1
E+ |wi,k i = E− √ (i − 1) |wi−1,k−1 i
2√
3 1 1
+ (k − i) q − p − (k − i − 1) |wi,k−1 i (4.23)
2 2 2
(with obvious simplifications in the cases when i = 1 or i = k)
– 75 –
Note that S1 = {|ψi} is linearly independent. Suppose that Sk−1 is linearly indepen-
dent (k ≥ 2). Consider Sk . Suppose
k
X
ci |wi,k i = 0 (4.24)
i=1
3 to (4.24) and using the linear independence of

for some constants ci . Applying E+
Sk−1 we find the relation
√ √
3 1 i 3 1 2 3 1 i 1 1
i( + p + + − k)ci+1 − √ (k − i) ( q − p + + − k)ci = 0 (4.25)
2 2 2 2 2 2 2 2 2 2
2 to (4.24) another recursion relation is obtained

for i = 1, . . . , k − 1. Applying E+
√
1 3 1 i 1 1
√ ici+1 + (k − i)( q − p + + − k)ci = 0 (4.26)
2 2 2 2 2 2
√
Combining these relations we find ci+1 = 0 for i = 1, . . . , k − 1. If 23 q − 12 p + 2i + 12 − 12 k 6= 0
when i = 1 then one also has c1 = 0. This holds if k ≤ n + 1, however if k = n + 2 then c1
is not fixed by these equations. The induction stops at this point.
These results are sufficient to fix the multiplicity of all the states. This is because √ the
vectors in Sk for 1 ≤ k ≤ k + 1 correspond to states with weight (p, q) − (k − 1)( 2 , 23 ) 1
which are at the top right hand corner of the k-th hexagonal (or outermost triangular for
k = n + 1) layer. We have shown therefore that these weights have multiplicity both less
than or equal to, and greater than or equal to k. Hence these weights have multiplicity
k. Next consider the states on the level k edges which are obtained by acting with the
L(SU (2)) lowering operators E− 1 and E 2 on the “corner weight” states. Observe the
−
following:
Lemma 9. Let d be a representation of L(SU (2)) on V be such that a particular L(SU (2))
weight m > 0 has multiplicity p. Then all weights m0 such that |m0 | ≤ m have multiplicity
≥p
whose proof is left as an exercise.
By this lemma, all the states on the k-th layer obtained in this fashion have multi-
plicity k also. Then the reflection symmetry implies that all states on the k-th layer have
multiplicity k.
In particular, the states on the outer triangular layer have multiplicity n + 1. We
have shown that the states on the triangular layers must have multiplicity not greater than
n + 1, but by the lemma above together with the reflection symmetry, they must also have
multiplicity ≥ n + 1. Hence the triangular layer weights have multiplicity n + 1, and the
proof is complete.
This was rather long-winded. There exist general formulae constraining multiplicities
of weights in more general Lie group representations, but we will not discuss these here.
– 76 –
4.1.4 Dimension of Irreducible Representations
Using the multiplicity properties of the weight diagram, it is possible to compute the
dimension of the representation. We consider first the hexagonal weight diagram for m ≥ n.
Then there are 1 + · · · + (m − n) + (m − n + 1) = 12 (m − n + 1)(m − n + 2) weights in the
interior triangle. Each of these weights has multiplicity (n + 1) which gives 12 (n + 1)(m −
n + 1)(m − n + 2) linearly independent states corresponding to weights in the triangle.
Consider next the k-th hexagonal layer for k = 1, . . . , n. This has 3((m + 1 − (k − 1)) +
(n + 1 − (k − 1)) − 2) = 3(m + n + 2 − 2k) weights in it, and each weight has multiplicity
k, which gives 3k(m + n + 2 − 2k) linearly independent states in the k-th hexagonal layer.
The total number of linearly independent states is then given by
n
1 X 1
(n+1)(m−n+1)(m−n+2)+ 3k(m+n+2−2k) = (m+1)(n+1)(m+n+2) (4.27)
2 2
k=1
This formula also applies in the case for m ≤ n and also for the triangular weight diagrams
by taking m = 0 or n = 0. The lowest dimensional representations are therefore 1,3,6,8,10...
4.1.5 The Complex Conjugate Representation

Definition 43. Let d be a representation of a Lie algebra L(G) acting on V . If v ∈ L(G),
then viewing d(v) as a matrix acting on V , the complex representation d¯ is defined by
¯
d(v)u = (d(v))∗ u (4.28)
for u ∈ V , where ∗ denotes matrix complex conjugation.

Note that as d(v) is linear in v over R, it follows that (d(v))∗ is also linear in v over
R. Also, as
d([v, w]) = d(v)d(w) − d(w)d(v) (4.29)
for v, w ∈ L(G), so taking the complex conjugate of both sides we find
¯ w]) = d(v)
d([v, ¯ d(w)
¯ ¯ d(v)
− d(w) ¯ (4.30)
i.e. d¯ is indeed a Lie algebra representation.

Suppose that Ta are the generators of L(G) with structure constants cab c . Then as d
is a representation,
[d(Ta ), d(Tb )] = cab c d(Tc ) (4.31)
Taking the complex conjugate, and recalling that cab c are real, we find
¯ a ), d(T
[d(T ¯ c)
¯ b )] = cab c d(T (4.32)
¯ a ) satisfy the same commutation relations.

i.e. the d(Ta ) and d(T
In the context of representations of L(SU (3)), the conjugate operators to iH1 , iH2 ,
m
i(E+ +E− m ) and E m −E m are denoted by iH̄ , iH̄ , i(Ē m + Ē m ), and Ē m − Ē m respectively
+ − 1 2 − + + −
and are given by
iH̄1 = (iH1 )∗
– 77 –
iH̄2 = (iH2 )∗
m m m m ∗
i(Ē− + Ē+ ) = (i(E+ + E− ))
m m m m ∗
Ē+ − Ē− = (E+ − E− ) (4.33)
which implies
H̄1 = −(H1 )∗ , H̄2 = −(H2 )∗ , m

Ē± m ∗
= −(E∓ ) (4.34)
Then H̄1 , H̄2 and Ē±m satisfy the same commutation relations as the unbarred opera-
tors, and also behave in the same way under the hermitian conjugate. One can therefore
plot the weight diagram associated with the conjugate representation d, ¯ the weights being
the (real) eigenvalues of H̄1 and H̄2 . But as H̄1 = −(H1 ) and H̄2 = −(H2 )∗ it follows
∗
that if (p, q) is a weight of the representation d, then (−p, −q) is a weight of the represen-
¯ So the weight diagram of d¯ is obtained from that of d by inverting all the points
tation d.
(p, q) → −(p, q). Note that this means that the equilateral triangular weight diagrams N
and H of equal length sides are conjugate to each other.
4.2 Some Low-Dimensional Irreducible Representations of L(SU (3))
4.2.1 The Singlet

The simplest representation has only one state, which is the highest weight state with
weight (0, 0). This representation is denoted 1.
4.2.2 3-dimensional Representations

Take the fundamental representation. Then as h1 and h2 are already diagonalized, it is
straightforward to compute the eigenstates and weights.
– 78 –
State
  Weight
1
0 ( 12 , 2√
1
)
 
3
0
 
0
1 (− 12 , 2√
1
)
 
3
0
 
0
0 (0, − √13 )
 
1
 
1
The state of highest weight is 0 which has weight ( 12 , 2√
1
). The weight diagram is
 
3
0
d u
This representation is denoted 3. It will be convenient to define the following states

in the 3 representation.
     
1 0 0
u = 0 , d = 1 , s = 0 (4.35)
     
0 0 1
so that u has weight ( 21 , 2√

1
3
), d has weight (− 12 , 2√
1
3
) and s has weight (0, − √13 ).
√ √
The lowering operators have the following effect: d = 2e1− u, s = 2e3− u and s =
√ 2
2e− d.
The complex conjugate of this representation is called 3̄ and the weights are obtained
by multiplying the weights of the 3 representation by −1.
– 79 –
State
  Weight
1
0 (− 21 , − 2√
1
)
 
3
0
 
0
1 ( 12 , − 2√
1
)
 
3
0
 
0
0 (0, √13 )
 
1
 
0
The state of highest weight is 0 which has weight (0, √13 ). The weight diagram is
 
1
u d
It will be convenient to define the following states in the 3̄ representation.

     
1 0 0
ū = 0 ,
  ¯
d = 1 ,
 
s̄ = 0
 
(4.36)
0 0 1
so that ū has weight (− 12 , − 2√

1
3
), d¯ has weight ( 12 , − 2√
1
3
) and s̄ has weight (0, √13 ).
√ √
The lowering operators have the following effect: ū = − 2ē3− s̄, d¯ = − 2ē2− s̄ and
√
¯ where ēm = −(em )∗ .
ū = − 2ē1− d; ± ∓
Exercise: Verify that all other lowering operators ēm − (except those given above) an-
¯
nihilate ū, d, s̄. Also compute the effect of the raising operators ēm +.
– 80 –
4.2.3 Eight-Dimensional Representations
Consider the adjoint representation defined on the complexified Lie algebra L(SU (3)), i.e.
ad(v)w = [v, w]. Then the weights of the states can be computed by evaluating the
commutators with h1 and h2 :
State v [h1 , v] [h2 , v] Weight

h1 0 0 (0, 0)
h2 0 0 (0, 0)
e1+ e1+ 0 (1, 0)
e1− −e1− 0 (−1, 0)
√ √
3 2 3
e2+ − 21 e2+ 2√ e+ (− 21 , )
√2
e2− 1 2
2 e− − 23 e2− 1
(2, − 2 ) 3
√ √
3 3
e3+ 1 3
2 e+ 2√ e+ ( 12 , 23 )
√
e3− − 21 e3− − 23 e3− (− 12 , − 23 )
√
The highest weight state is e3+ with weight ( 12 , 23 ). All weights have multiplicity 1
except for (0, 0) which has multiplicity 2. The weight diagram is a regular hexagon:
4.3 Tensor Product Representations

Suppose that d1 , d2 are irreducible representations of L(SU (3)) acting on V1 , V2 respec-
N
tively. Then let V = V1 V2 and d = d1 ⊗ 1 + 1 ⊗ d2 be the tensor product representation
of L(SU (3)) on V . In general d is not irreducible on V , so we want to decompose V into
a direct sum of invariant subspaces on which the restriction of d is irreducible.
To do this, recall that one can choose a basis of V1 which consists entirely of eigenstates
of both d1 (h1 ) and d1 (h2 ). Similarly, one can also choose a basis of V2 which consists entirely
of eigenstates of both d2 (h1 ) and d2 (h2 ). Then the tensor product of the basis eigenstates
N
produces a basis of V1 V2 which consists of eigenstates of d(h1 ) and d(h2 ).
Explicitly, suppose that |φi i ∈ Vi is an eigenstate of di (h1 ) and di (h2 ) with weight
(pi , qi ) (i.e. di (h1 ) |φi i = pi |φi i and di (h2 ) |φi i = qi |φi i) for i = 1, 2. Define |φi =
– 81 –
|φ1 i ⊗ |φ2 i. Then
d(h1 ) |φi = (d1 (h1 ) |φ1 i) ⊗ |φ2 i + |φ1 i ⊗ (d2 (h1 ) |φ2 i)
= (p1 |φ1 i) ⊗ |φ2 i + |φ1 i ⊗ (p2 |φ2 i)
= (p1 + p2 ) |φi (4.37)
and similarly
d(h2 ) |φi = (q1 + q2 ) |φi (4.38)
So the weight of |φi is (p1 + p2 , q1 + q2 ); the weights add in the tensor product repre-
sentation.
Using this, one can plot a weight diagram consisting of the weights of all the eigenstates
in the tensor product basis of V , the points in the weight diagram are obtained by adding
the pairs of weights from the weight diagrams of d1 and d2 respectively, keeping careful
track of the multiplicities (as the same point in the tensor product weight diagram may be
N
obtained from adding weights from different states in V1 V2 .)
Once the tensor product weight diagram is constructed, pick a highest weight, which
corresponds to a state which is annihilated by the tensor product operators E+ m for m =
1, 2, 3. (Note that as the representation is finite-dimensional such a state is guaranteed

to exist, though as the representation is no longer irreducible, it need not be unique). If
there are multiple highest weight states corresponding to the same highest weight, one
can without loss of generality take them to be mutually orthogonal. Picking one of these,
generate further states by acting on a highest weight state with all possible combinations
of lowering operators. The span of these (finite number) of states produces an invariant
subspace W1 of V on which d is irreducible. Remove these weights from the tensor product
weight diagram. If the multiplicity of one of the weights in the original tensor product
diagram is k, and the multiplicity of the same weight in the W1 weight diagram is k 0 then
on removing the W1 weights, the multiplicity of that weight must be reduced from k to
k − k0 .
Repeat this process until there are no more weights left. This produces a decomposition
L L
V = W1 ... Wk of V into invariant subspaces Wj on which d is irreducible.
Note that one could also perform this process on triple (and higher order) tensor
N N
products e.g. V1 V2 V3 . In this case, one would construct the tensor product weight
diagram by adding triplets of weights from the weight diagrams of d1 on V1 , d2 on V2 and
d3 on V3 respectively.
This process can be done entirely using the weight diagrams, because we have shown
that for irreducible representations, the location of the highest weight fixes uniquely the
shape of the weight diagram and the multiplicities of its states.
We will see how this works for some simple examples:
4.3.1 3 ⊗ 3 decomposition.
Consider the 3 ⊗ 3 tensor product. Adding the weights together one obtains the following
table of quark content and associated weights
– 82 –
Quark content and weights for 3 ⊗ 3
Quark Content Weight
u⊗u (1, √13 )
d⊗d (−1, √13 )
s⊗s (0, − √23 )
u ⊗ d, d ⊗ u (0, √13 )
u ⊗ s, s ⊗ u ( 12 , − 2√
1
3
)
d ⊗ s, s ⊗ d (− 12 , − 2√
1
3
)
Plotting the corresponding weight diagram gives
The raising and lowering operators are E± m = em ⊗ 1 + 1 ⊗ em . The highest weight state
± ±
is u ⊗ u with weight (1, √13 ). Applying lowering operators to u ⊗ u it is clear that a
six-dimensional irreducible representation is obtained. The (unit-normalized) states and
weights are given by
States and weights for the 6 in 3 ⊗ 3

State Weight
u⊗u (1, √13 )
d⊗d (−1, √13 )
s⊗s (0, − √23 )
√1 (d ⊗ u + u ⊗ d) (0, √13 )
2
√1 (u ⊗ s + s ⊗ u) ( 12 , − 2√
1
)
2 3
√1 (d ⊗ s + s ⊗ d) (− 12 , − 2√
1
)
2 3
– 83 –
which has the following weight diagram
This representation is called 6. Removing the (non-vanishing) span of these states

from the tensor product space, one is left with a 3-dimensional vector space. The new
highest weight is at (0, √13 ) with corresponding state √12 (d ⊗ u − u ⊗ d) (this is the unique
linear combination- up to overall scale- of d ⊗ u and u ⊗ d which is annihilated by all the
raising operators). This generates a 3̄. The states and their weights are
States and weights for the 3̄ in 3 ⊗ 3

State Weight
√1 (d ⊗ u− u ⊗ d) (0, √13 )
2
√1 (d ⊗ s − s ⊗ d) (− 12 , − 2√
1
)
2 3
√1 (s ⊗ u − u ⊗ s) ( 12 , − 2√
1
)
2 3
Hence 3 ⊗ 3 = 6 ⊕ 3̄. The states in the 6 are symmetric, whereas those in the 3̄ are
antisymmetric.
4.3.2 3 ⊗ 3̄ decomposition
For this tensor product the quark content/weight table is as follows:
– 84 –
Quark content and weights for 3 ⊗ 3̄
√
3
u ⊗ s̄ ( 12 , 2 )
u ⊗ d¯ (1, 0)
√
3
d ⊗ s̄ (− 12 , 2 )
¯ s ⊗ s̄
u ⊗ ū, d ⊗ d, (0, 0)
d ⊗ ū (−1, 0)
√
3
s ⊗ ū (− 21 , − )
√2
s ⊗ d¯ ( 12 , − 23 )
with weight diagram
ds us
du dd,uu,ss ud
su sd
The raising and lowering operators are E± m = em ⊗ 1 + 1 ⊗ ēm All weights have
± ±
multiplicity 1, except
√ for (0, 0) which has multiplicity 3. The highest weight state is u ⊗ s̄
1 3
with weight ( 2 , 2 ). Acting on this state with all possible lowering operators one obtains
an 8 with the following states and weights
States and weights for the 8 in 3 ⊗ 3̄

State Weight
√
3
u ⊗ s̄ ( 12 , 2 )
u ⊗ d¯ (1, 0)
√
3
d ⊗ s̄ (− 12 , 2 )
√1 (d
2
⊗ d¯ − u ⊗ ū), √16 (d ⊗ d¯ + u ⊗ ū − 2s ⊗ s̄) (0, 0)
d ⊗ ū (−1, 0)
√
3
s ⊗ ū (− 21 , − )
√2
s ⊗ d¯ 1 3
(2, − 2 )
– 85 –
Removing these weights from the weight diagram, one is left with a singlet 1 with
weight (0, 0), corresponding to the state
1 ¯
√ (u ⊗ ū + s ⊗ s̄ + d ⊗ d) (4.39)
3
which is the unique linear combination- up to an overall scale- of u⊗ ū, s⊗ s̄ and d⊗ d¯ which
m . Hence we have the decomposition 3⊗ 3̄ = 8⊕1.
is annihilated by the raising operators E+
4.3.3 3 ⊗ 3 ⊗ 3 decomposition.
For this tensor product the quark content/weight table is as follows:
Quark content and weights for 3 ⊗ 3 ⊗ 3

√
3
u⊗u⊗u ( 32 ,
2 )
√
s⊗s⊗s (0, − 3)
√
3
d⊗d⊗d (− 23 , 2 )
u ⊗ u ⊗ s, u ⊗ s ⊗ u, s ⊗ u ⊗ u (1, 0)
√
3
u ⊗ u ⊗ d, u ⊗ d ⊗ u, d ⊗ u ⊗ u ( 12 , 2√ )
s ⊗ s ⊗ u, s ⊗ u ⊗ s, u ⊗ s ⊗ s ( 21 , − 23 )
√
s ⊗ s ⊗ d, s ⊗ d ⊗ s, d ⊗ s ⊗ s (− 12 , − 23 )
d ⊗ d ⊗ s, d ⊗ s ⊗ d, s ⊗ d ⊗ d (−1, 0)
√
3
d ⊗ d ⊗ u, d ⊗ u ⊗ d, u ⊗ d ⊗ d (− 21 , 2 )
u ⊗ d ⊗ s, u ⊗ s ⊗ d, d ⊗ u ⊗ s,
d ⊗ s ⊗ u, s ⊗ u ⊗ d, s ⊗ d ⊗ u (0, 0)
with weight diagram
ddd ddu duu uuu
dus
dds uus
dss uss
sss
The raising and lowering operators are E± m = em ⊗ 1 ⊗ 1 + 1 ⊗ em ⊗ 1 + 1 ⊗ 1 ⊗ em .

± ± ±
There are six weights of multiplicity 3, and the weight (0, 0) has multiplicity 6. The highest
– 86 –
√
weight is u ⊗ u ⊗ u with weight ( 23 , 23 ). By applying lowering operators to this state, one
obtains a triangular 10-dimensional irreducible representation denoted by 10, which has
normalized states and weights:
States and weights for 10 in 3 ⊗ 3 ⊗ 3

State Weight
√
3
u⊗u⊗u ( 32 ,
2 )
√
s⊗s⊗s (0, − 3)
√
3
d⊗d⊗d (− 32 , 2 )
√1 (u ⊗ u ⊗ s + u ⊗ s ⊗ u + s ⊗ u ⊗ u) (1, 0)
3 √
√1 (u ⊗ u ⊗ d + u ⊗ d ⊗ u + d ⊗ u ⊗ u) 3
3
( 12 , 2 )
√
√1 (s ⊗ s ⊗ u + s ⊗ u ⊗ s + u ⊗ s ⊗ s)
3
( 12 , − 23 )
√
√1 (s ⊗ s ⊗ d + s ⊗ d ⊗ s + d ⊗ s ⊗ s)
3
(− 12 , − 23 )
√1 (d ⊗ d ⊗ s + d ⊗ s ⊗ d + s ⊗ d ⊗ d) (−1, 0)
3 √
√1 (d ⊗ d ⊗ u + d ⊗ u ⊗ d + u ⊗ d ⊗ d) 3
3
(− 12 , 2 )
√1 (u ⊗ d ⊗ s + u ⊗ s ⊗ d + d ⊗ u ⊗ s+
6
d ⊗ s ⊗ u + s ⊗ u ⊗ d + s ⊗ d ⊗ u) (0, 0)
The 10 weight diagram is
ddd ddu duu uuu
dds dus uus
dss uss
sss
Removing the (non-vanishing) span of these states from the tensor product space, one
is left with a 17-dimensional vector space. The new weight diagram is
– 87 –
ddu duu
dus
dds uus
dss uss
√
Note that the highest weight is now ( 12 , 23 ). This weight has has multiplicity 2. It
should be noted that the subspace consisting of linear combinations of d ⊗ u ⊗ u, u ⊗ d ⊗ u
and u ⊗ u ⊗ d which is annihilated by all raising operators E+ m is two-dimensional and
1
is spanned by the two orthogonal states √6 (d ⊗ u ⊗ u + u ⊗ d ⊗ u − 2u ⊗ u ⊗ d) and
√1 (d⊗ u ⊗ u − u ⊗ d ⊗ u). By acting on these two states with all possible lowering
2
operators, one obtains two 8 representations whose states are mutually orthogonal.
The states and weights of these two 8 representations are summarized below:
States and weights for an 8 in 3 ⊗ 3 ⊗ 3

State Weight
√
√1 (d ⊗ u ⊗ u + u ⊗ d ⊗ u − 2u ⊗ u ⊗ d) 3
6
( 12 , 2 )
√1 (s ⊗ u ⊗ u + u ⊗ s ⊗ u − 2u ⊗ u ⊗ s) (1, 0)
6 √
√1 (2d ⊗ d ⊗ u − d ⊗ u ⊗ d − u ⊗ d ⊗ d) 3
6
(− 12 , 2 )
1
√
2 3
(s ⊗ d ⊗ u + s ⊗ u ⊗ d + d ⊗ s ⊗ u
+u ⊗ s ⊗ d − 2d ⊗ u ⊗ s − 2u ⊗ d ⊗ s),
1
√
2 3
(2s ⊗ d ⊗ u + 2d ⊗ s ⊗ u − s ⊗ u ⊗ d
−d ⊗ u ⊗ s − u ⊗ s ⊗ d − u ⊗ d ⊗ s) (0, 0)
√1 (s ⊗ d ⊗ d + d ⊗ s ⊗ d − 2d ⊗ d ⊗ s) (−1, 0)
6 √
√1 (2s ⊗ s ⊗ u − s ⊗ u ⊗ s − u ⊗ s ⊗ s) 3
6
( 12 , −2 )
√
√1 (2s ⊗ s ⊗ d − s ⊗ d ⊗ s − d ⊗ s ⊗ s)
6
(− 12 , − 23 )
– 88 –
States and weights for another 8 in 3 ⊗ 3 ⊗ 3
State Weight
√
√1 (d ⊗ u ⊗ u 3
2
− u ⊗ d ⊗ u) ( 12 , 2 )
√1 (s ⊗ u ⊗ u − u ⊗ s ⊗ u) (1, 0)
2 √
√1 (d ⊗ u ⊗ d 3
2
− u ⊗ d ⊗ d) (− 12 , 2 )
1
2 (s ⊗ d ⊗ u + s ⊗ u ⊗ d − d ⊗ s ⊗ u − u ⊗ s ⊗ d),
1
2 (s ⊗ u ⊗ d + d ⊗ u ⊗ s − u ⊗ s ⊗ d − u ⊗ d ⊗ s) (0, 0)
√1 (s ⊗ d ⊗ d − d ⊗ s ⊗ d) (−1, 0)
2 √
√1 (s ⊗ u ⊗ s − u ⊗ s ⊗ s) 3
2
( 12 , −2 )
√
√1 (s ⊗ d ⊗ s − d ⊗ s ⊗ s)
2
(− 12 , − 23 )
Removing these weights from the weight diagram, we are left with a singlet 1 with
weight (0, 0). The state corresponding to this singlet is
1
√ (s ⊗ d ⊗ u − s ⊗ u ⊗ d + d ⊗ u ⊗ s − d ⊗ s ⊗ u + u ⊗ s ⊗ d − u ⊗ d ⊗ s) (4.40)
6
which is the only linear combination-up to overall scale- of u ⊗ d ⊗ s, u ⊗ s ⊗ d, d ⊗ u ⊗ s,

d ⊗ s ⊗ u, s ⊗ u ⊗ d and s ⊗ d ⊗ u which is annihilated by all the raising operators.
Hence we have the decomposition 3 ⊗ 3 ⊗ 3 = 10 ⊕ 8 ⊕ 8 ⊕ 1 where the states in 10
are symmetric, but the state in 1 is antisymmetric. The 8 states have mixed symmetry.
4.4 The Quark Model
It is possible to arrange the baryons and the mesons into SU (3) multiplets; i.e. the states
lie in Hilbert spaces which are tensor products of vector spaces equipped with irreducible
representations of L(SU (3)). To see examples of this, it is convenient to group hadrons into
multiplets with the same baryon number and spin. We plot the hypercharge Y = S + B
where S is the strangeness and B is the baryon number against the isospin eigenvalue I3
for these particles.
4.4.1 Meson Multiplets
The pseudoscalar meson octet has B = 0 and J = 0. The (I3 , Y ) diagram is
– 89 –
Mass(Mev) 0
Y +
495 K 1
K
1/2
− +
137 π π0 π
I3
549 −1 −1/2 η 1/2 1
−1/2
− −1
0
K K
495
There is also a J = 0 meson singlet η 0 .

The vector meson octet has B = 0 and J = 1. The (I3 , Y ) diagram is
Mass(Mev) 0
∗ Y +∗
892 K 1
K
1/2
770 ρ− ρ0 ρ+
I3
−1 −1/2 ω 1/2 1
783
−1/2
−∗ −1
0∗
K K
892
There is also a J = 1 meson singlet, φ.
4.4.2 Baryon Multiplets
3
The baryon decuplet has B = 1 and J = 2 with (I3 , Y ) diagram
– 90 –
Y
Mass (Mev)
+
∆− ∆0 1 ∆ ∆++
1235
1/2
−∗ ∗
1385 Σ Σ0 Σ+ ∗
−3/2 −1 −1/2 1/2 1 3/2
I3
−1/2
−∗ 0∗
Ξ −1 Ξ
1530
−3/2
−
1670
−2
Ω
There is also an antibaryon decuplet with (I3 , Y ) → −(I3 , Y ).

The baryon octet has B = 1, J = 21 with (I3 , Y ) diagram
Mass(Mev) Y
1
939 n p
1/2
− 0 +
1193 Σ Σ Σ
0 I3
1116 −1 −1/2 Λ 1/2 1
−1/2
− +
Ξ −1
Ξ
1318
1
and there is also a J = 2 baryon singlet Λ0∗ .
4.4.3 Quarks: Flavour and Colour

√
On making the identification (p, q) = (I3 , 23 Y ) the points on the meson and baryon octets
and the baryon decuplet can be matched to points on the weight diagrams of the 8 and 10
of L(SU (3)).
Motivated by this, it is consistent to consider the (light) meson states as lying within
a 3 ⊗ 3̄; as 3 ⊗ 3̄ = 8 ⊕ 1, the meson octets are taken to correspond to the 8 states, and
the meson singlets correspond to the singlet 1 states. The light baryon states lie within a
3 ⊗ 3 ⊗ 3; the baryon decuplet corresponds to the 10 in 3 ⊗ 3 ⊗ 3 = 10 ⊕ 8 ⊕ 8 ⊕ 1; the
baryon octet corresponds to appropriate linear combinations of elements in the 8 irreps,
and the baryon singlet corresponds to the 1.
– 91 –
In this model, the fundamental states in the 3 are quarks, with basis states u (up),
d (down) and s (strange). The basis labels u, d, s are referred to as the flavours of the
¯ s̄. Baryons are composed of
quarks. The 3̄ states are called antiquarks with basis ū, d,
bound states of three quarks qqq, mesons are composed of bound states of pairs of quarks
and antiquarks q q̄. The quarks have J = 12 and B = 13 whereas the antiquarks have J = 21
and B = − 13 which is consistent with the values of B and J for the baryons and mesons.
The quark and antiquark flavours can be plotted on the (I3 , Y ) plane:
Y Y
2/3
s
d 1/3 u
−1/2 1/2
I3 −1/2 1/2
I3
−1/3
u d
−2/3
s
We have shown that mesons and baryons can be constructed from q q̄ and qqq states
respectively. But why do qq particles not exist? This problem is resolved using the notion
of colour. Consider the ∆++ particle in the baryon decuplet. This is a u ⊗ u ⊗ u state with
J = 32 . The members of the decuplet are the spin 32 baryons of lowest mass, so we assume
that the quarks have vanishing orbital angular momentum. Then the spin J = 32 is obtained
by having all the quarks in the spin up state, i.e. u ↑ ⊗u ↑ ⊗u ↑. However, this violates the
Pauli exclusion principle. To get round this problem, it is conjectured that quarks possess
additional labels other than flavour. In particular, quarks have additional charges called
colour charges- there are three colour basis states associated with quarks called r (red), g
(green) and b (blue). The quark state wave-functions contain colour factors which lie in
a 3 representation of SU (3) which describes their colour; the colour of antiquark states
corresponds to a 3̄ representation of SU (3) (colour). This colour SU (3) is independent of
the flavour SU (3).
These colour charges are also required to remove certain discrepancies (of powers of
3) between experimentally observed processes such as the decay π 0 → 2γ and the cross
section ratio between the processes e+ e− → hadrons and e+ e− → µ+ µ− and theoretical
predictions. However, although colour plays an important role in these processes, it seems
that one cannot measure colour directly experimentally- all known mesons and baryons
are SU (3) colour singlets (so colour is confined). This principle excludes the possibility
of having qq particles, as there is no singlet state in the SU (3) (colour) tensor product
decomposition 3 ⊗ 3, though there is in 3 ⊗ 3 ⊗ 3 and 3 ⊗ 3̄. Other products of 3 and 3̄
can also be ruled out in this fashion.
Nevertheless, the decomposition of 3 ⊗ 3 is useful because it is known that in addition
to the u, d and s quark states, there are also c (charmed), t (top) and b (bottom) quark
flavours. However, the c, t and b quarks are heavier than the u, d and s quarks, and are
– 92 –
unstable- they decay into the lighter quarks. The SU (3) symmetry cannot be meaningfully
extended to a naive SU (6) symmetry because of the large mass differences which break the
symmetry. In this context, meson states formed from a heavy antiquark and a light quark
can only be reliably put into 3 multiplets, whereas baryons made from one heavy and two
light quarks lie in 3 ⊗ 3 = 6 ⊕ 3̄ multiplets.
– 93 –
5. Spacetime Symmetry
In this section we examine spacetime symmetry in the absence of gravity. Spacetime

is taken to be 4-dimensional Minkowski space, M 4 , with real co-ordinates xµ for µ =
0, 1, 2, 3, equipped with the Minkowski metric which has the non-vanishing components
η00 = −η11 = −η22 = −η33 = 1, or as a matrix
 
1 0 0 0
 0 −1 0 0
(η)µν = (η)µν =  (5.1)
 
−1

 0 0 0
0 0 0 −1
5.1 The Lorentz Group

A Lorentz transformation is a linear transformation Λ : M 4 → M 4 which transforms
co-ordinates
x0µ = Λµ ν xν (5.2)
for Λµ ν ∈ R, but leaves the length invariant
ηµν x0µ x0ν = ηµν xµ xν (5.3)
for all x.
This condition can be rewritten in matrix notation as
ΛT ηΛ = η (5.4)
Suppose that Λ1 , Λ2 are two 4 × 4 matrices satisfying (5.4) then it is straightforward to

see that Λ1 Λ2 satisfies (5.4). Also, if Λ satisfies (5.4) then det Λ = ±1, hence Λ is invertible,
with inverse Λ−1 = η −1 ΛT η, and
(Λ−1 )T ηΛ−1 = ηΛη −1 .η.η −1 ΛT η

= ηΛη −1 ΛT η
= ηΛΛ−1
= η (5.5)
so Λ−1 is also a Lorentz transformation. Hence, the set of Lorentz transformations forms
a group, under matrix multiplication.
Write a generic Lorentz transformation as
!
λ βT
Λ= (5.6)
α R
where λ ∈ R, α, β ∈ R3 and R is a 3 × 3 real matrix

Then the constraint (5.4) is equivalent to
λ2 = 1 + α.α (5.7)
– 94 –
and
λβ = RT α (5.8)
and
RT R − ββ T = I3 (5.9)
Note that (5.7) implies that

p
λ = ± 1 + α.α (5.10)
so in particular λ ≤ −1 or λ ≥ +1. Then (5.8) fixes β in terms of α and R by
1
β = ±√ RT α (5.11)
1 + α.α
which can be used to rewrite (5.9) as
1
RT R − RT ααT R = I3 (5.12)
1 + α.α
Define
1
ααT R

R̂ = 1 − √ √ (5.13)
1 + α.α(1 + 1 + α.α)
or equivalently
1
ααT R̂

R= 1+ √ (5.14)
1+ 1 + α.α
Then (5.12) implies
R̂T R̂ = I3 (5.15)
i.e. R̂ ∈ O(3). Moreover, it is straightforward to check directly that
1
det R̂ = √ det R (5.16)
1 + α.α
where we have used the formula det(I3 + KααT ) = 1 + Kα.α for any K. Also, one can
write ! ! !
λ 0 1 λ−2 αT 1 0
Λ= (5.17)
α I3 0 I3 − λ−2 ααT 0 R
and hence
λ λ
det Λ = det R = √ det R̂ (5.18)
1 + α.α 1 + α.α
O(3) has two connected components, the connected component of I3 (which is SO(3))
whose elements have determinant det R̂ = +1, and the connected component of −I3 , whose
elements have determinant det R̂ = −1.
There are therefore four connected components of the Lorentz group, according as
Λ 0 ≤ −1 or Λ0 0 ≥ 1 and det Λ = +1 or det Λ = −1. It is not possible to construct a
0
smooth curve in the Lorentz group passing from one of these components to the other.
The set of Lorentz transformations with det Λ = +1 forms a subgroup of the Lorentz
group.
– 95 –
Note that (5.13) implies that
1
R̂T α = √ RT α (5.19)
1 + α.α
and hence
1
β.β = αT RRT α = α.α (5.20)
1 + α.α
So, if Λ and Λ0 are two Lorentz transformations
! !
λ βT λ0 β 0T
Λ= Λ0 = (5.21)
α R α0 R0
with λ ≥ 1 and λ0 ≥ 1, then

√
(ΛΛ0 )0 0 = λλ0 + β.α0 ≥
p p p
1 + α.α 1 + α0 .α0 − α.α α0 .α0 ≥ 1 (5.22)
Hence the set of Lorentz transformations with Λ0 0 ≥ 1 also forms a subgroup of the
Lorentz group.
The subgroup of Lorentz transformations with det Λ = +1 and Λ0 0 ≥ 1 is called the
proper orthochronous Lorentz group, which we denote by SO(3, 1)↑ .
We note the useful lemma
Lemma 10. Suppose that Λ ∈ SO(3, 1)↑ . Then there exist S1 , S2 ∈ SO(3) and z ∈ R such
that
 
! cosh z sinh z 0 0 !
1 0T  sinh z cosh z 0 0 1 0T
Λ= (5.23)
 
 
0 S1  0 0 1 0 0 S2
0 0 0 1
Proof
From the analysis of the Lorentz group so far, we have shown that if Λ ∈ SO(3, 1)↑
then there exists α ∈ R3 and R̂ ∈ SO(3) such that
√ !
1 + α.α αT R̂
Λ= 1
ααT R̂

α 1 + 1+√1+α.α
√ ! !
1 + α.α αT 1 0T
= 1 (5.24)
α 1 + 1+√1+α.α ααT 0 R̂
There also exists S1 ∈ SO(3) and z ∈ R such that

 
sinh z
α = S1  0  (5.25)
 
0
The result follows on substituting this into Λ and setting S2 = (S1 )T R̂.
– 96 –
5.2 The Lorentz Group and SL(2, C)
Consider the spacetime co-ordinates xµ . Define the matrices
! ! ! !
1 0 0 1 0 −i 1 0
σ0 = σ1 = σ2 = σ3 = (5.26)
0 1 1 0 i 0 0 −1
Then given real spacetime co-ordinates xµ , define the 2 × 2 complex hermitian matrix
!
µ x0 − x3 −x1 + ix2
x̃ = xµ σ = (5.27)
−x1 − ix2 x0 + x3
Observe that any hermitian 2 × 2 matrix can be written as x̃ for some real xµ .
Note that
det x̃ = (x0 )2 − (x1 )2 − (x2 )2 − (x3 )2 = ηµν xµ xν (5.28)
ηµν xµ xν is invariant under the action of SO(3, 1). det x̃ is invariant under the action of
SL(2, C), the complex 2 × 2 matrices with unit determinant.
Proposition 30. There exists an isomorphism π : SL(2, C)/Z2 → SO(3, 1)↑ where SL(2, C)/Z2
consists of elements ±N ∈ SL(2, C) with +N identified with −N .
Proof
Given N ∈ SL(2, C) consider the 2 × 2 complex matrix N x̃N † . The components of
this matrix are linear in the spacetime co-ordinates xµ . As x̃ is hermitian, it follows that
N x̃N † is also hermitian. Hence there exist Λµ ν ∈ R (independent of x) for µ, ν = 0, . . . , 3
such that
N x̃N † = (Λx)
] (5.29)
] for all x, and therefore Λ is
Taking the determinant of both sides we find det x̃ = det (Λx)
a Lorentz transformation.
Set
Λ = π(N ) (5.30)
Note that
] = 2Λ0 µ xµ = Tr (N † N x̃)
Tr (Λx) (5.31)
Setting x0 = 1, x1 = x2 = x3 = 0 we find Λ0 0 = 21 Tr (N † N ) > 0.
If N1 , N2 ∈ SL(2, C) then
(N1 N2 )x̃(N1 N2 )† = (π(N

^1 N2 )x) (5.32)
But
(N1 N2 )x̃(N1 N2 )† = N1 (N2 x̃N2† )N1†

^ †
= N1 (π(N 2 )x)N 1
= π(N^
1 )π(N2 )x (5.33)
– 97 –
^
Hence (π(N ^
1 N2 )x) = π(N1 )π(N2 )x for all x, which implies π(N1 N2 ) = π(N1 )π(N2 ).
Next we will establish that π is onto SO(3, 1)↑ . First recall that any R ∈ SO(3, 1)↑ of
the form !
1 0T
R= (5.34)
0 R̂
can be written as a product of rotations around the spatial co-ordinate axes
 
1 0 0 0
 0 1 0 0 
R1 (φ1 ) =  (5.35)
 

 0 0 cos φ1 sin φ1 
0 0 − sin φ1 cos φ1
 
1 0 0 0
 0 cos φ2 0 − sin φ2 
R2 (φ2 ) =  (5.36)
 

 0 0 1 0 
0 sin φ2 0 cos φ2
and
 
1 0 0 0
 0 cos φ3 sin φ3 0
R3 (φ3 ) =  (5.37)
 
− sin φ3

 0 cos φ3 0
0 0 0 1
iφj
σj
By a direct computation we find π(e 2 ) = Rj for j = 1, 2, 3; and
 
cosh z sinh z 0 0
z 1
 sinh z cosh z 0 0
π(e− 2 σ ) =  (5.38)
 

 0 0 1 0
0 0 0 1
iφj j z 1
where e 2 σ ∈ SL(2, C) for j = 1, 2, 3 and e− 2 σ ∈ SL(2, C).
Hence, if Λ ∈ SO(3, 1)↑ , it follows that one can write Λ = Λ1 .Λ2 . . . Λk where Λi are
elementary rotation or boost transformations in SO(3, 1)↑ , and from the above reasoning,
Λi = π(Ni ) for some Ni ∈ SL(2, C). Therefore, Λ = π(N1 .N2 . . . Nk ), so π is onto SO(3, 1)↑ .
Next, suppose that π(N ) = π(M ) for N, M ∈ SL(2, C). Then
N x̃N † = M x̃M † (5.39)
Set Q = M −1 N , so that
Qx̃Q† = x̃ (5.40)
Setting x0 = 1, x1 = x2 = x3 = 0, we obtain QQ† = I2 , so Q ∈ SU (2). Hence
Qσ i = σ i Q (5.41)
– 98 –
for i = 1, 2, 3. The only Q ∈ SU (2) satisfying this is Q = ±I2 , so M = ±N . Hence π is a
2 − 1 map.
Lastly, we must prove that if N ∈ SL(2, C) then π(N ) ∈ SO(3, 1)↑ . We have already
shown that π(N ) is orthochronous. Suppose that det(π(N )) = −1. Consider
 
1 0 0 0
 0 1 0 0
Λ̂ =   π(N ) (5.42)
 
 0 0 1 0
0 0 0 −1
The det Λ̂ = +1, so Λ̂ ∈ SO(3, 1)↑ . Hence, there exists some N 0 ∈ SL(2, C) such that
Λ̂ = π(N 0 ), so
 
1 0 0 0
 0 1 0 0
 π(N ) = π(N 0 ) (5.43)
 

 0 0 1 0
0 0 0 −1
Setting Y = N 0 N −1 , we obtain
 
1 0 0 0
 0 1 0 0
 = π(Y ) (5.44)
 

 0 0 1 0
0 0 0 −1
for some Y ∈ SL(2, C). This implies that
Y x µ σ µ Y † = x 0 I2 + x 1 σ 1 + x 2 σ 2 − x 3 σ 3 (5.45)
for all xµ . In particular, for x0 = 1, x1 = x2 = x3 = 0 we find Y Y † = I2 , so Y ∈ SU (2).

The remaining constraints are
Y σ 1 = σ 1 Y, Y σ 2 = σ 2 Y, Y σ 3 = −σ 3 Y (5.46)
This is not possible, because [Y, σ 1 ] = [Y, σ 2 ] = 0 implies that Y = αI2 for some α ∈ C.
As det Y = 1 this implies Y = ±I2 , but then Y σ 3 6= −σ 3 Y . Hence if N ∈ SL(2, C) then
π(N ) ∈ SO(3, 1)↑ .
Although π : SL(2, C) → SO(3, 1)↑ is not 1-1, we have shown that the restriction of π
to SL(2, C)/Z2 , in which N is identified with −N is 1-1.
5.3 The Lie Algebra L(SO(3, 1))

To compute the constraints on the tangent matrices, consider a curve in the Lorentz group
Λ(t) with Λ(0) = I4 . This is constrained by
Λ(t)T ηΛ(t) = η (5.47)
– 99 –
Differentiate this constraint and set t = 0 to obtain
mT η + ηm = 0 (5.48)
where m = ( dΛ(t)
dt )|t=0 . The generic solution to this constraint is
!
0 χ
mµ ν = (5.49)
χT S
for any χ ∈ R3 and S is a real 3 × 3 antisymmetric matrix; S = −S T . There are three real
degrees of freedom in χ and three real degrees of freedom in the antisymmetric matrix S.
Hence the Lie algebra is six-dimensional.
Define the 4 × 4 matrices M µν for µ, ν = 0, 1, 2, 3 by
(M µν )α β = i(η µα δ ν β − η να δ µ β ) (5.50)
note that M µν = −M νµ , so there are only six linearly independent matrices defined here.
By direct computation, we find
   
0 i 0 0 0 0 i 0
 i 0 0 0  0 0 0 0
 
M 01 =   M 02 = 
 

 0 0 0 0   i 0 0 0
0 0 0 0 0 0 0 0
   
0 0 0 i 0 0 0 0
 0 0 0 0
 0
 0 −i 0
M 03 =   M 12 = 
 

 0 0 0 0  0 i 0 0
i 0 0 0 0 0 0 0
   
0 0 0 0 0 0 0 0
 0 0 0 −i
 0
 0 0 0
M 13 =   M 23 =  (5.51)
 
−i

 0 0 0 0  0 0 0
0 i 0 0 0 0 i 0
and
[M µν , M ρσ ] = i M µσ η νρ + M νρ η µσ − M µρ η νσ − M νσ η µρ

(5.52)
which defines the complexified Lie algebra of the Lorentz group.
Define
1
Ji = ijk Mjk
2
Ki = M0i (5.53)
for i, j, k = 1, 2, 3. Then it follows that
[Ji , Jj ] = iijk Jk
[Ki , Kj ] = −iijk Jk
[Ji , Kj ] = iijk Kk (5.54)
– 100 –
So, setting
1 1
Ai = (Ji − iKi ), Bi = (Ji + iKi ) (5.55)
2 2
we obtain the commutation relations
[Ai , Aj ] = iijk Ak
[Bi , Bj ] = iijk Bk
[Ai , Bj ] = 0 (5.56)
Hence the complexified Lorentz algebra L(SO(3, 1)) can be written as the direct sum of
two commuting complexified L(SU (2)) algebras. It follows that one can classify irreducible
representations of the Lorentz algebra by spins (A, B) for 2A, 2B ∈ N.
5.4 Spinors and Invariant Tensors of SL(2, C)

Definition 44. The left handed Weyl spinors are elements of a 2-dimensional complex
vector space V on which the fundamental representation of SL(2, C) acts via D(N )ψ = N ψ
where N ∈ SL(2, C). In terms of components, if ψ ∈ V has components ψα for α = 1, 2
with respect to some basis of V , then under the action of SL(2, C), ψ transforms as
ψα → ψα0 = Nα β ψβ (5.57)
where N ∈ SL(2, C).
Definition 45. The right handed Weyl spinors are elements of a 2-dimensional com-
plex vector space V̄ on which the complex conjugate of the fundamental representation of
SL(2, C) acts as D∗ (N )χ̄ = N ? χ̄ where N ∈ SL(2, C) and N ? is the complex conjugate
of N . In terms of components, if χ̄ ∈ V̄ has components χ̄α̇ for α̇ = 1, 2, then under the
action of SL(2, C), χ̄ transforms as
χ̄α̇ → χ̄0α̇ = N ∗ α̇ β̇ χ̄β̇ (5.58)
where N ∈ SL(2, C).

Note: One should regard α and α̇ as being entirely independent! The components of
these spinors anticommute.
We also define αβ and αβ to be totally skew-symmetric with
! !
αβ 0 1 0 −1
= αβ = (5.59)
−1 0 1 0
and observe that αβ βγ = δ α γ . One defines α̇β̇ and α̇β̇ similarly.
Note that αβ is invariant under SL(2, C), as
αβ → 0αβ = µν Nα µ Nβ ν = det(N )αβ = αβ (5.60)
or in matrix notation N N T = .
– 101 –
If we define the contravariant representation DCV on V via
DCV (N )ψ → (N T )−1 ψ (5.61)
and the complex conjugate contravariant representation by

∗
DCV (N )χ̄ → (N ∗T )−1 χ̄ (5.62)
then N N T = implies that (N T )−1 = −1 N , so DCV is equivalent to the fundamental

representation. The complex conjugate representations are similarly equivalent.
The tensors αβ and αβ are called invariant tensors as they transform into themselves
under the action of SL(2, C). For SO(3, 1), the invariant tensors are ηµν and η µν , which
can be used to raise and lower indices. We will raise and lower SL(2, C) indices using αβ ,
αβ , so if ψα , χ̄α̇ are in the fundamental and conjugate representations respectively, then
we define
ψ α = αβ ψβ , χ̄α̇ = α̇β̇ χ̄β̇ (5.63)
One can construct a tensor product representation of the fundamental representation
acting on n products of V , V ⊗ V ⊗ V · · · ⊗ V . In terms of components, elements of the
tensor product vector space have components ψα1 ...αn which transform under the action of
SL(2, C) as
ψα1 ,...,αn → ψα0 1 ,...,αn = Nα1 β1 . . . Nαn βn ψβ1 ...βn (5.64)
for N ∈ SL(2, C). Similarly, tensor product representations of the complex conjugate
representation correspond to complex tensors with components χ̄α˙1 ...α˙m which transform
as
χ̄α̇1 ...α̇m → χ̄0α̇1 ...α̇m = N ∗ α̇1 β̇1 . . . N ∗ α̇m β̇m χ̄β̇1 ...β̇m (5.65)
By taking the tensor product n tensor products of V acted on by the fundamental
representations, with m tensor products of V̄ acted on by the conjugate representation, one
obtains a vector space which has elements with components ψα1 ...αn β̇1 ...β̇m which transform
as
ψα1 ...αn β̇1 ...β̇m → ψα0 = Nα1 µ1 . . . Nαn µn N ∗ β̇1 ν̇1 . . . N ∗ β̇m ν̇m ψβ1 ...βn ν̇1 ...ν̇m (5.66)
1 ...αn β̇1 ...β̇m
This representation is in general not irreducible.
5.4.1 Lorentz and SL(2, C) indices

It is straightforward to map between Lorentz invariant tensors and SL(2, C) invariant
tensors. In particular, recall that the relationship between N ∈ SL(2, C) and the corre-
sponding Lorentz transformation Λ = Λ(N ) is given by
N xµ σ µ N † = ηνρ Λρ γ xγ σ ν (5.67)
which implies that
N σ µ N † = ηνρ Λρ γ η γµ σ ν = (η −1 ΛT η)µ ν σ ν = (Λ−1 )µ ν σ ν (5.68)
– 102 –
So, denoting the components of σ µ by σαµβ̇ , one finds
σαν β̇ = Nα λ Nβ̇∗ γ̇ Λν µ σλµγ̇ (5.69)
which implies that σαµβ̇ is invariant. One can also define
(σ̄ µ )α̇α = αβ α̇β̇ σβµβ̇ (5.70)
so that σ̄ 0 = σ 0 , σ̄ i = −σ i for i = 1, 2, 3.
Exercise: Prove the following useful identities
i) σ µ σ̄ ν + σ ν σ̄ µ = 2η µν I2
ii) Tr σ µ σ̄ ν = 2η µν
iii) σαµα̇ σ̄µβ̇β = 2δαβ δα̇β̇ .

Definition 46. Define the 4 × 4 matrices γ µ by
!
0 σµ
γµ = (5.71)
σ̄ µ 0
Then these matrices satisfy
{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 2η µν I4 (5.72)
An algebra satisfying this property is called a Clifford algebra.
Definition 47. A Dirac spinor ΨD is a 4-component spinor constructed from left and right
handed Weyl spinors ψα , χ̄α̇ via
!
ψα
ΨD = (5.73)
χ̄α̇
The gamma matrices act on Dirac spinors.

The σ-matrix identities are useful. For example; recall that the correspondence be-
tween Λ ∈ SO(3, 1)↑ and N ∈ SL(2, C) is given by N σ µ N † = ηνρ Λρ γ η γµ σ ν . Then using
(ii) above the components of Λ are given by
1
Λµ ν = Tr σ̄ µ N σν N †

(5.74)
2
Also, it is straightforward to relate tensors with SL(2, C) indices to tensors with
Lorentz indices. Given a 4-vector with Lorentz indices V µ one can define a tensor with
SL(2, C) indices via
Vαα̇ = V µ (σµ )αα̇ (5.75)
The invariance of (σ µ )αα̇ ensures that if V µ transforms as V µ → Λµ ν V ν under the action
of the Lorentz group, then Vαα̇ transforms as Vαα̇ → Nα β Nα̇∗ β̇ Vβ β̇ under SL(2, C). This
expression can be inverted using (ii) of the above exercise to give
1
V µ = (σ̄ µ )α̇α Vαα̇ (5.76)
2
Similar maps between higher order tensors can also be constructed.
– 103 –
5.4.2 The Lie algebra of SL(2, C)
The Lie algebra of SL(2, C) consists of traceless complex 2 × 2 matrices; which has six real
dimensions. This is to be expected, as L(SO(3, 1)) is six-dimensional. It is convenient to
define the matrices
i µ ν
(σ µν )α β = (σ σ̄ − σ ν σ̄ µ )α β
4
i
(σ̄ µν )α̇ β̇ = (σ̄ µ σ ν − σ̄ ν σ µ )α̇ β̇ (5.77)
4
so that σ 0i = − 2i σ i , σ jk = 12 jk` σ ` , σ̄ 0i = 2i σ i , σ̄ jk = 21 jk` σ ` . It is clear that the σ µν span
the 2 × 2 traceless matrices over R (as do the σ̄ µν ), hence they are generators of the Lie
algebra of SL(2, C). By a direct computation we obtain the commutation relations:
[σ µν , σ ρσ ] = i η νρ σ µσ + η µσ σ νρ − η νσ σ µρ − η µρ σ νσ

(5.78)
which is the same commutation relation as for the Lie algebra L(SO(3, 1)). Similarly, we
find
[σ̄ µν , σ̄ ρσ ] = i η νρ σ̄ µσ + η µσ σ̄ νρ − η νσ σ̄ µρ − η µρ σ̄ νσ

(5.79)
Hence the σ µν and σ̄ µν correspond to representations of L(SO(3, 1)).
The action of SL(2, C) on left-handed Weyl spinors is given by
µν
ψα → eωµν σ α β ψβ (5.80)
Just as for the Lorentz algebra, one can define Ji = 12 ijk σjk = 12 σ i and Ki = σ0i = 2i σ i .
Hence Ai = 12 σ i , Bi = 0. Therefore the fundamental representation corresponds to a spin- 21
L(SU (2)) representation generated by A, and a L(SU (2)) B-singlet. This representation
is denoted by ( 21 , 0).
The action of SL(2, C) on right-handed Weyl spinors is given by
µν β̇
χ̄α̇ → eωµν σ̄ α̇ χ̄β̇ (5.81)
Again, define Ji = 12 ijk σ̄jk = 12 σ i and Ki = σ̄0i = − 2i σ i . Hence Ai = 0, Bi = 21 σ i .

Therefore this representation corresponds to a spin- 12 L(SU (2)) representation generated
by B, and a L(SU (2)) A-singlet. This representation is denoted by (0, 21 ).
5.5 The Poincaré Group

The Poincaré group consists of Lorentz transformations combined with translations; which
act on the spacetime co-ordinates by
xµ → Λµ ν xν + bµ (5.82)
where Λ is a Lorentz transformation and b ∈ R4 is an arbitrary 4-vector. One can denote

the generic Poincaré group element by a pair (Λ, b) which act in this way. Note that under
the action of (Λ, b) followed by (Λ0 , b0 ); x → Λ0 Λx + Λ0 b + b0 , hence one defines the group
product to be
(Λ0 , b0 )(Λ, b) = (Λ0 Λ, Λ0 b + b0 ) (5.83)
– 104 –
so the Poincareé group is closed under this multiplication. The identity is (I4 , 0) and the
inverse of (Λ, b) is (Λ−1 , −Λ−1 b).
One can construct a group isomorphism between the Poincaré group and the subgroup
of GL(5, R) of matrices of the form
!
Λ b
(5.84)
0 1
where Λ is a Lorentz transformation and b is an arbitrary 4-vector, as under matrix mul-

tiplication
! ! !
Λ0 b0 Λ b Λ0 Λ Λ0 b + b0
= (5.85)
0 1 0 1 0 1
5.5.1 The Poincaré Algebra

Consider a curve
!
Λ(t) b(t)
(5.86)
0 1
in the Poincaré group passing through the identity when t = 0, so Λ(0) = I4 , b(0) = 0.
Differentiating with respect to t and setting t = 0 we note that the generic element of the
Poincaré Lie algebra is of the form
!
m v
(5.87)
0 0
where m ∈ L(SO(3, 1)) and v ∈ R4 is unconstrained. Hence a basis for the Lie algebra is
given by the 5 × 5 matrices M µν and P ν for µ, ν = 0, 1, 2, 3 where
(M ρσ )µ ν = i(η ρµ δ σ ν − η σµ δ ρ ν )
(M ρσ )4 ν = (M ρσ )µ 4 = (M ρσ )4 4 = 0 (5.88)
and
(P ν )µ 4 = iη µν
(P ν )µ λ = (P ν )4 λ = (P ν )4 4 = 0 (5.89)
(labeling the matrix indices by µ, ν = 0, 1, 2, 3 and the additional index is “4”). The M ρσ
generate the Lorentz sub-algebra
!
m 0
(5.90)
0 0
for m ∈ L(SO(3, 1)); they satisfy the usual Lorentz algebra commutation relations

(5.91)
– 105 –
The P ν generate the translations
!
0 iv
(5.92)
0 0
for v ∈ R4 . The P µ satisfy

[P µ , P ν ] = 0 (5.93)
and
[P µ , M ρσ ] = iη ρµ P σ − iη σµ P ρ (5.94)
The commutation relations

[P µ , M ρσ ] = iη ρµ P σ − iη σµ P ρ
[P µ , P ν ] = 0 (5.95)
define the Poincaré algebra.
5.5.2 Representations of the Poincaré Algebra

Definition 48. Suppose that d is a representation of the Poincaré algebra. Let µνρσ be
the totally antisymmetric tensor with 0123 = 1. Then the Pauli-Lubanski vector is defined
by
1
Wµ = µρσν d(M ρσ )d(P ν ) (5.96)
2
Proposition 31. The Pauli-Lubanski vector satisfies the following commutation relations:
1) [Wµ , d(Pν )] = 0
2) [Wµ , d(Mρσ )] = iηµρ Wσ − iηµσ Wρ
3) [Wµ , Wν ] = −iµνρσ W ρ d(P σ )
Proof
We will use the identities
µαβγ µρστ = −6δ ρ [α δ σ β δ τ γ] = −6δ [ρ α δ σ β δ τ ] γ (5.97)

and
µναβ µνρσ = −4δ ρ [α δ σ β] = −4δ [ρ α δ σ] β (5.98)

To prove (1) is straightforward:
1
[Wµ , d(Pν )] = µρσθ [d(M ρσ )d(P θ ), d(Pν )]
2
– 106 –
1
µρσθ d(M ρσ )[d(P θ ), d(Pν )] + [d(M ρσ ), d(Pν )]d(P θ )

=
2
1
= µρσθ d(M ρσ )d([P θ , Pν ]) + d([M ρσ , Pν ])d(P θ )

2
1
= µρσθ − iδ ρ ν d(P σ ) + iδ σ ν d(P ρ ) d(P θ )

2
=0 (5.99)
To prove (ii) is an unpleasant exercise in algebra:
1
µλχθ [d(M λχ )d(P θ ), d(Mρσ )

[Wµ , d(Mρσ )] =
2
1
µλχθ d(M λχ )[d(P θ ), d(Mρσ )] + [d(M λχ ), d(Mρσ )]d(P θ )

=
2
1
µλχθ d(M λχ )d([P θ , Mρσ ]) + d([M λχ , Mρσ ])d(P θ )

=
2
1
µλχθ d(M λχ ) iδ θ ρ d(Pσ ) − iδ θ σ d(Pρ )

=
2
i d(M λ σ )δ χ ρ − d(M χ σ )δ λ ρ − d(M λ ρ )δ χ σ + d(M χ ρ )δ λ σ d(P θ )

+
i
= µλχθ d(M λχ )δ θ ρ d(Pσ ) − d(M λχ )δ θ σ d(Pρ )
2
2d(M λ σ )δ χ ρ d(P θ ) − 2d(M λ ρ )δ χ σ d(P θ )

+
i
(ηστ δ θ ρ − ηρτ δ θ σ )µλχθ d(M λχ )d(P τ ) − 2d(M λτ )d(P χ )

=
2
3i
= (ηστ δ θ ρ − ηρτ δ θ σ )µλχθ d(M [λχ )d(P τ ] ) (5.100)
2
But
1 λχτ γ
λχτ γ W γ = γ ν1 ν2 ν3 d(M ν1 ν2 )d(P ν3 )
2
= 3d(M [λχ )d(P τ ] ) (5.101)
Hence
i
[Wµ , d(Mρσ )] = (ηστ δ θ ρ − ηρτ δ θ σ )µλχθ λχτ γ W γ
2
i
= (ηστ δ θ ρ − ηρτ δ θ σ )(−2)(δ τ µ ηθγ − δ τ θ ηµγ )W γ
2
= iηµρ Wσ − iηµσ Wρ (5.102)
as required.
(3) follows straightforwardly from (2):
1
[Wµ , Wν ] = νρσθ [Wµ , d(M ρσ )d(P θ )]
2
1
= νρσθ [Wµ , d(M ρσ )]d(P θ ) + d(M ρσ )[Wµ , d(P θ )]

2
1
= νρσθ [Wµ , d(M ρσ )]d(P θ )
2
– 107 –
1
= νρσθ (iδ ρ µ W σ − iδ σ µ W ρ )d(P θ )
2
= −iµνρσ W ρ d(P σ ) (5.103)
as required. .
From this we find the
Corollory 3. The following commutation relations hold
1) [Wµ W µ , d(Pν )] = 0
2) [Wµ W µ , d(Mρσ )] = 0
Proof
(1) follows because [Wµ W µ , d(Pν )] = Wµ [W µ , d(Pν )] + [Wµ , d(Pν )]W µ = 0
(2) holds because
[Wµ W µ , d(Mρσ )] = Wµ [W µ , d(Mρσ )] + [Wµ , d(Mρσ )]W µ

= Wµ (iδ µ ρ Wσ − iδ µ σ Wρ ) + (iηµρ Wσ − iηµσ Wρ )W µ
=0 (5.104)
as required.
Hence we have shown that Wµ W µ is a Casimir operator. d(Pµ )d(P µ ) is another Casimir
operator:
Proposition 32. The following commutation relations hold
1) [d(Pµ )d(P µ ), d(Pν )] = 0
2) [d(Pµ )d(P µ ), d(Mρσ )] = 0
Proof
(1) follows because
[d(Pµ )d(P µ ), d(Pν )] = d(Pµ )[d(P µ ), d(Pν )] + [d(Pµ ), d(Pν )]d(P µ ) = 0 (5.105)
(2) holds because
[d(Pµ )d(P µ ), d(Mρσ )] = d(Pµ )[d(P µ ), d(Mρσ )] + [d(Pµ ), d(Mρσ )]d(P µ )

= d(Pµ )(iδ µ ρ d(Pσ ) − iδ µ σ d(Pρ )) + (iηµρ d(Pσ ) − iηµσ d(Pρ ))d(P µ )
=0 (5.106)
as required.
We shall show that irreducible representations are classified by the values of the two
Casimir operators Wµ W µ and d(Pµ )d(P µ ).
– 108 –
In particular, suppose that D is a unitary representation of the Poincaré group acting
on V . Such representations arise naturally in the context of quantum field theory when
V is taken to be a Hilbert space, and it is assumed that Poincaré transformations do not
affect transition probabilities. We will assume that this is the case.
Note that iMµν and iP µ form a basis for the (real) Poincaré algebra. Hence one can
locally write the Poincaré transformation as
i µ +ω µν )
e− 2 (bµ P µν M
(5.107)
for real bµ and skew-symmetric real ωµν , and

i µ +ω µν ) i µ )+ω µν ))
D(e− 2 (bµ P µν M
) = e− 2 (bµ d(P µν d(M
(5.108)
where d is a representation of the Poincaré algebra acting on V . As D is unitary, d(Mρσ )

and d(P µ ) are hermitian.
As the d(P µ ) commute with each other and are hermitian, they can be simultaneously
diagonalized, with real eigenvalues. For a 4-vector q µ define the subspace Vq of V to be
the simultaneous eigenspace
Vq = {|ψi ∈ V : d(P µ ) |ψi = q µ |ψi , µ = 0, 1, 2, 3} (5.109)
and M
V = Vq (5.110)
q
Then on Vq , d(Pµ )d(P µ ) = q µ qµ = q 2 . We will assume that for configurations of

physical interest, such as when q is the 4-momentum of a massive particle or of a photon,
that q 2 ≥ 0 and q 0 > 0. We will only consider these cases.
Consider first the operators
it µ )+ω µν )) it µ )+ω µν ))
hλ (t) = e 2 (bµ d(P µν d(M
d(P λ )e− 2 (bµ d(P µν d(M
(5.111)
Differentiating with respect to t we find
dhλ it µ µν i it µ µν
= e 2 (bµ d(P )+ωµν d(M )) [d(P λ ), − (bµ d(P µ ) + ωµν d(M µν ))]e− 2 (bµ d(P )+ωµν d(M ))
dt it µ µν
2 it µ µν
= ω λ χ e 2 (bµ d(P )+ωµν d(M )) d(P χ )e− 2 (bµ d(P )+ωµν d(M ))
= ω λ χ hχ (5.112)
with the initial condition hλ (0) = d(P λ ). Therefore

it µν
hλ (t) = (etω )λ ρ d(P ρ ) = (e− 2 ωµν M )λ ρ d(P ρ ) (5.113)
Hence, setting t = 1 we find

i µ )+ω µν )) i µν i µ )+ω µν ))
d(P λ )e− 2 (bµ d(P µν d(M
= (e− 2 ωµν M )λ ρ e− 2 (bµ d(P µν d(M
d(P ρ ) (5.114)
i µ )+ω µν )) i µν
So, if |ψi ∈ Vq then e− 2 (bµ d(P µν d(M
|ψi ∈ Vq0 where q 0 = e− 2 ωµν M q.
– 109 –
Definition 49. The stability subgroup Hq (or “little group”) which is associated with Vq
is the subgroup of the Poincaré group defined by
i µ +ω µν ) i µ )+ω µν ))
Hq = {e− 2 (bµ P µν M
: e− 2 (bµ d(P µν d(M
|ψi ∈ Vq for |ψi ∈ Vq } (5.115)
It can be shown that Hq is a Lie subgroup of the Poincaré group. Suppose then that
it µν
− 2i (bµ P µ + ωµν M µν ) ∈ L(Hq ). It follows that e− 2 ωµν M q = q for t ∈ R. Expanding out
in powers of t we see that this constraint is equivalent to
ωµν q ν = 0 (5.116)
which has a general solution

ωµν = µνρσ nρ q σ (5.117)
− 2i (bµ P µ +ωµν M µν )
where nρ is an arbitrary constant 4-vector. Hence, if |ψi ∈ Vq and e ∈ Hq
then
i µ µν i µ µ
e− 2 (bµ d(P )+ωµν d(M )) |ψi = e− 2 bµ q e−in Wµ |ψi (5.118)
so we have reduced the action of Hq on Vq to the action of Wµ on Vq .
The action of a generic Poincaré transformation (Λ, b) does not leave Vq invariant,
because
Vq → Vq0 = VΛq (5.119)
as q 0 = Λq. However, q 2 = q 02 is invariant. Hence we can split V into invariant subspaces Vq
corresponding to vectors q µ which have the same value of q 2 . We will therefore henceforth
work with such an invariant subspace, and consider q 2 = m2 to be fixed.
If m2 > 0 then there is a Lorentz transformation Λ0 (q) such that q = Λ0 (q)k where
k µ = (m, 0, 0, 0). Alternatively, if m = 0 then there is a Lorentz transformation Λ(q) such
that q = Λ0 (q)k where k µ = (E, E, 0, 0). These Lorentz transformations can be taken to
be fixed functions of the q.
The key step is to show that the action of the entire Poincaré group on the Vq (with
q = m2 fixed) is fixed by the action of Hk on Vk (which is in turn determined by the
2
action of the Pauli-Lubanski vector on Vk ).

To show this, first note that if |ψk i ∈ Vk , then one can write
|ψq i = D(Λ0 (q), 0) |ψk i (5.120)
where |ψq i ∈ Vq , and these transformations can be used to obtain all elements of Vq from
those in Vk .
It is then straightforward to show that the action of the representation of the whole
Poincaré group on {Vq : q 2 = m2 } is determined by the action of D(Hk ) acting on Vk .
To see this explicitly, suppose that |ψk,M i for M = 1, . . . , `k is a basis of Vk . Then one
can define
|ψq,M i = D(Λ0 (q), 0) |ψk,M i (5.121)
– 110 –
and the |ψq,M i then form a basis for Vq . The representation of Hk on Vk is determined by
the coefficients D(h)M N , where h ∈ Hk , in the expansion
X
D(h) |ψk,M i = D(h)M N |ψk,N i (5.122)
N
Suppose that (Λ, b) is a generic Poincaré transformation; then one can write
D(Λ, b) |ψq,M i = D(Λ, b)D(Λ0 (q), 0) |ψk,M i

= D(Λ0 (Λq), 0)D(Λ0 (Λq)−1 ΛΛ0 (q), Λ0 (Λq)−1 b) |ψk,M i (5.123)
However, ((Λ0 (Λq))−1 ΛΛ0 (q), Λ0 (Λq)−1 b) ∈ Hk , so it is possible to expand
X
D(Λ, b) |ψq,M i = D(Λ0 (Λq), 0)D(Λ0 (Λq)−1 ΛΛ0 (q), Λ0 (Λq)−1 b)M N |ψk,N i
N
X
= D(Λ0 (Λq)−1 ΛΛ0 (q), Λ0 (Λq)−1 b)M N |ψΛq,N i (5.124)
N
We will examine the action of Hk on Vk in the timelike and null cases separately.
Although
M
V = Vq (5.125)
q
is not in general finite-dimensional, we shall assume that Vk (and hence the Vq ) are
finite dimensional.
5.5.3 Massive Representations of the Poincaré Group: k µ = (m, 0, 0, 0)

We compute the action of Wµ on Vk . If |ψi ∈ Vk then
1
W0 |ψi = ij` d(M ij )d(P ` ) |ψi = 0
2
1
Wi |ψi = − ij` d(M j` )d(P 0 ) |ψi = −md(Ji ) |ψi (5.126)
2
We have already shown that d(Ji ) generates a L(SU (2)) algebra, hence the little group for
massive representations is SO(3). For irreducible representations, the spin is fixed by the
value taken by the Casimir on Vk ; Wµ W µ |ψi = −m2 d(Ji )d(Ji ) |ψi.
5.5.4 Massless Representations of the Poincaré Group: k µ = (E, E, 0, 0)

Again, we compute the action of Wµ on Vk . If |ψi ∈ Vk then
1
W0 |ψi = ij` d(M ij )d(P ` ) |ψi = d(M 23 )d(P 1 ) |ψi = Ed(J1 ) |ψi
2
W1 |ψi = −Ed(M 23 ) |ψi = −Ed(J1 ) |ψi
1
W2 |ψi = 2µνλ d(M µν )d(P λ ) |ψi
2
– 111 –
= E(d(M 13 ) − d(M 03 )) |ψi
= E(−d(J2 ) + d(K3 )) |ψi
1
W3 |ψi = 3µνλ d(M µν )d(P λ ) |ψi
2
= E(−d(M 12 ) + d(M 02 )) |ψi
= E(−d(J3 ) − d(K2 )) |ψi (5.127)
Observe that the following commutation relations hold:
[−d(J2 ) + d(K3 ), −d(J3 ) − d(K2 )] = 0

[d(J1 ), −d(J3 ) − d(K2 )] = −i(−d(J2 ) + d(K3 ))
[d(J1 ), −d(J2 ) + d(K3 )] = i(−d(J3 ) − d(K2 )) (5.128)
These expressions may be simplified slightly by setting R1 = −d(J3 ) − d(K2 ), R2 =

−d(J2 ) + d(K3 ), J = d(J1 ); so that
[R1 , R2 ] = 0, [J, R1 ] = −iR2 , [J, R2 ] = iR1 (5.129)
R1 and R2 are commuting hermitian operators on Vk , and hence can be simultaneously

diagonalized over R. Consider a state |ψi ∈ Vk with R1 |ψi = r1 |ψi, R2 |ψi = r2 |ψi for
r1 , r2 ∈ R. Define
f (θ) = e−iθJ R1 eiθJ |ψi

g(θ) = e−iθJ R2 eiθJ |ψi (5.130)
for θ ∈ R. Differentiating with respect to θ and using the commutation relations we

df
find dθ = −g, dg
dθ = f . Solving these equations with the initial condition f (0) = r1 |ψi,
g(0) = r2 |ψi we find
f (θ) = (r1 cos θ − r2 sin θ) |ψi

g(θ) = (r1 sin θ + r2 cos θ) |ψi (5.131)
which implies
R1 eiθJ |ψi = (r1 cos θ − r2 sin θ)eiθJ |ψi

R2 eiθJ |ψi = (r1 sin θ + r2 cos θ)eiθJ |ψi (5.132)
Hence, unless r1 = r2 = 0, there is a continuum of R1 , R2 eigenstates which implies Vk

cannot be finite-dimensional. We must therefore have R1 = R2 = 0 on Vk . J is also a
hermitian operator on Vk , and can also be diagonalized. For irreducible representations, J
can have only one eigenvalue, σ ∈ R. σ is called the helicity of the particle. It follows that
Wµ = σkµ , so it is clear that σ is a Lorentz invariant quantity.
There is no algebraic constraint fixing the value of the helicity σ in the massless case,
as there is to fix the spin in the massive case. However, for physically realistic systems, one
can make a topological argument to fix 2σ ∈ Z. This is because eiθJ describes a rotation of
angle θ in the spatial plane in the 2, 3 directions. So in particular, setting θ = 2π we find
e2πiJ |ψi = e2πσi |ψi (5.133)
– 112 –
We require that e2πσi = ±1 (for a projective representation) and so 2σ ∈ Z. Neutrinos
have helicity ± 12 , photons have helicity ±1 and gravitons have helicity ±2.
– 113 –
6. Gauge Theories
Lie groups and Lie algebras play an important role in the various dynamical theories which
govern the behaviour of particles - the gauge theories. Though we will not examine the
quantization of these theories, we shall present the relationship between Lie algebras and
gauge theories.
Before examining non-Abelian gauge theories, we briefly recap some properties of the
simplest gauge theory, which is the U (1) gauge theory of electromagnetism.
6.1 Electromagnetism
The gauge theory of electromagnetism contains a field strength
fµν = ∂µ aν − ∂ν aµ (6.1)
where µ, ν = 0, 1, 2, 3 and xµ are co-ordinates on Minkowski space (indices raised/lowered

with the Minkowski metric η), and ∂µ ≡ ∂x∂ µ . aµ ∈ R is the 4-vector potential.
Under a gauge transformation aµ −→ a0µ = aµ − ∂µ λ where λ is a real function,
0
fµν → fµν = ∂µ (aν − ∂ν λ) − ∂ν (aµ − ∂µ λ) = ∂µ aν − ∂ν aµ = fµν (6.2)
since ∂µ ∂ν λ = ∂ν ∂µ λ. Hence fµν is invariant under gauge transformations.

The field equations of electromagnetism are
∂ µ fµν = jν (6.3)
and
λµνρ ∂µ fνρ = 0 (6.4)
Equation (6.4) holds automatically due to the existence of the vector potential. Conversely,
if fµν satisfies (6.4) then it can be shown that a vector potential aµ exists (though only
locally) such that fµν = ∂µ aν − ∂ν aµ .
Using the vector potential, one defines a covariant derivative Dµ by
Dµ ψ = ∂µ ψ + iaµ ψ (6.5)
where ψ = ψ(x). Under a gauge transformation
ψ −→ ψ 0 = eiλ ψ, aµ −→ a0µ = aµ − ∂µ λ (6.6)
where λ = λ(x), it is straightforward to see that
Dµ ψ −→ (Dµ ψ)0 = ∂µ (eiλ ψ) + i(aµ − ∂µ λ)eiλ ψ

= eiλ Dµ ψ (6.7)
so that Dµ ψ transforms like ψ. This means that the Dirac equation
iγ µ Dµ Ψ − mΨ = 0 (6.8)
– 114 –
is gauge invariant. Ψ is a 4-component Dirac spinor constructed from left and right handed
Weyl spinors ψα , χ̄α̇ via
!
ψα
Ψ= (6.9)
χ̄α̇
and the 4 × 4 matrices γ µ are given by

!
µ 0 σµ
γ = (6.10)
σ̄ µ 0
These matrices satisfy the Clifford algebra
{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 2η µν I4 . (6.11)
The standard Lagrangian governing the interaction of electrodynamics with scalar

fields
1 1
L = − fµν f µν + (Dµ φ)∗ Dµ φ + V (φ∗ φ) (6.12)
4 2
where φ is a complex scalar field, and V is a real function of φ∗ φ is also gauge covariant.
It is possible to obtain the gauge field strength in a natural way from the commutator
of covariant derivatives. If φ is a scalar field then
Dµ Dµ φ = (∂µ + iaµ )(∂ν φ + iaν φ)

= ∂µ ∂ν φ + i(aν ∂µ φ + aµ ∂ν φ) − aµ aν φ + i∂µ aν φ (6.13)
and hence
[Dµ , Dν ]φ = i(∂µ aν − ∂ν aµ )φ = ifµν φ (6.14)
6.2 Non-Abelian Gauge Theory

6.2.1 The Fundamental Covariant Derivative
Suppose that G is a compact matrix Lie group acting on a vector space V via the funda-
mental representation. Consider a scalar field Φ(x) which is an xµ dependent element of
V (which can be thought of as a column vector of ordinary scalar fields). Suppose that Φ
transforms under the fundamental representation as
Φ(x) −→ Φ0 (x) = g(x)Φ(x) (6.15)
where g(x) ∈ G.
Definition 50. The fundamental covariant derivative Dµ is defined by
Dµ Φ = ∂µ Φ + Aµ Φ (6.16)
where Aµ ∈ L(G) is an element of the Lie algebra of G acting on V .

We require that Dµ Φ should transform under local gauge transformations in the same
way as Φ. Suppose that Aµ → A0µ under local gauge transformations. Then we need
∂µ Φ0 + A0µ Φ0 = g(∂µ Φ + Aµ Φ) (6.17)
– 115 –
which implies
∂µ (gΦ) + A0µ gΦ = g(∂µ Φ + Aµ Φ) (6.18)
and hence
∂µ gΦ + A0µ gφ = gAµ Φ (6.19)
As this must hold for all Φ, we find the transformation rule
A0µ = gAµ g −1 − ∂µ gg −1 (6.20)
Before proceeding further, there is a question of consistency: namely if Aµ ∈ L(G) then

we must verify that A0µ given above is also an element of L(G). This is proved using the
Lemma 11. If g(t) is a curve in the matrix Lie group G then dg dt g(t)
−1 ∈ L(G).
Proof
Suppose that g(t) = g0 when t = t0 . Set h(t) = g(t + t0 )g0−1 . Then h(t) is a smooth
curve in G with h(0) = I, and
dh dg dg −1
|t=0 = |t=t0 g0−1 = g |t=t0 (6.21)
dt dt dt
dg −1
But dh
dt |t=0 ∈ L(G) by definition, and hence dt g |t=t0 ∈ L(G) for all t0 .
Hence we have shown that ∂µ gg −1 ∈ L(G), and from our previous analysis of the
adjoint representation, we know that gAµ g −1 ∈ L(G); so A0µ ∈ L(G) as required.
6.2.2 Generic Covariant Derivative

Definition 51. Suppose that G is a matrix Lie group with representation D acting on V ,
and let d denote the associated representation of the Lie algebra acting on V . Let elements
θ ∈ V transform as θ → θ0 = D(g(x))θ under local gauge transformations.
Then the covariant derivative Dµ associated with D acting on V is defined by
Dµ θ = ∂µ θ + d(Aµ )θ (6.22)
where Aµ ∈ L(G) transforms as Aµ → A0µ = gAµ g −1 − ∂µ gg −1 .

In order to show that this covariant derivative transforms as Dµ θ → D(g)Dµ θ we must
prove the
Lemma 12. Suppose that D is a representation of G acting on V , with associated repre-

sentation d of L(G) acting on V . Then
i) If v ∈ L(G) and g ∈ G, then d(gvg −1 )D(g) = D(g)d(v)
ii) If g(t) is a curve in G then dD(g) −1 dg −1

dt D(g ) = d dt g
Proof
(Caveat: in this proof t is simply a parameter along a curve in G, not the spacetime
co-ordinate x0 !)
– 116 –
To prove (i), set g = eh for h ∈ L(G), so that D(g) = ed(h) . Then (i) is equivalent to
e−d(h) d(eh ve−h )ed(h) = d(v) (6.23)
Set
f (t) = e−td(h) d(eth ve−th )etd(h) (6.24)
for t ∈ R. Then
df
e−td(h) [d(eth ve−th ), d(h)] + d(eth [h, v]e−th ) etd(h)

=
dt
e−td(h) d([eth ve−th , h]) + d(eth [h, v]e−th ) etd(h)

=
e−td(h) d(eth [v, h]e−th ) + d(eth [h, v]e−th ) etd(h)

=
= 0 (6.25)
and f (0) = d(v). Hence, f (1) = e−d(h) d(eh ve−h )ed(h) = f (0) = d(v) as required.
To prove (ii), suppose g(t) = g0 at t = t0 . Set h(t) = g(t + t0 )g0−1 , so that h(t) is a
smooth curve in G with h(0) = I. Then
dD(g) dD(h(t)g0 ) dD(h(t))

D(g −1 )|t=t0 = |t=0 D(g0−1 ) = |t=0 (6.26)
dt dt dt
and
dg −1 dh(t)
d( g |t=t0 ) = d( |t=0 ) (6.27)
dt dt
2)
As h(0) = I we can set h(t) = eth1 +O(t for some constant matrix h1 . Then
dh(t)
|t=0 = h1 ∈ L(G) (6.28)
dt
so
dh(t)
d( |t=0 ) = d(h1 ) (6.29)
dt
But by definition
d d
D(eth1 ) |t=0 =

d(h1 ) = D(h(t)) |t=0 (6.30)
dt dt
Therefore
dD(g) dg
D(g −1 )|t=t0 = d(h1 ) = d( g −1 |t=t0 ) (6.31)
dt dt
are required.
Proposition 33. The covariant derivative Dµ associated with D transforms as
Dµ θ → (Dµ θ)0 = D(g)Dµ θ (6.32)
under local gauge transformations.

Proof
Note that
∂µ θ0 + d(A0µ )θ0 = ∂µ (D(g)θ) + d(gAµ g −1 − ∂µ gg −1 )D(g)θ
– 117 –
= D(g)(∂µ θ + d(Aµ )θ) + d(gAµ g −1 )D(g) − D(g)d(Aµ )
∂µ (D(g)) − d(∂µ gg −1 )D(g) θ

+
D(g)Dµ θ + d(gAµ g −1 )D(g) − D(g)d(Aµ ) θ

=
∂µ (D(g)) − d(∂µ gg −1 )D(g) θ

+ (6.33)
However, by the previous lemma, we have proved that d(gAµ g −1 )D(g)−D(g)d(Aµ ) = 0

and ∂µ (D(g)) − d(∂µ gg −1 )D(g) = 0, hence
∂µ θ0 + d(A0µ )θ0 = D(g)Dµ θ (6.34)
as required.
Given this property of transformations of covariant derivatives, one can define the
adjoint covariant derivative
Definition 52. Suppose that θ ∈ L(G) transforms under the adjoint representation Ad of
G. The the covariant derivative associated with the adjoint representation is
Dµ θ = ∂µ θ + (ad Aµ )θ = ∂µ θ + [Aµ , θ] (6.35)
To summarize, we have shown that if Φ transforms under the action of the fundamental
representation as Φ → Φ0 = gΦ, then in order for the fundamental covariant derivative to
transform in the same way, one must impose the transformation
Aµ → A0µ = gAµ g −1 − ∂µ gg −1 (6.36)
on the gauge potential. We then have shown that if Φ transforms under the action of a
generic representation D, Φ → Φ0 = D(g)Φ, then the same transformation rule Aµ →
A0µ = gAµ g −1 − ∂µ gg −1 is sufficient to ensure that the generic covariant derivative Dµ Φ =
∂µ Φ + d(Aµ )Φ also transforms in the same way as Φ. Caveat: a covariant derivative is
always defined with respect to a particular representation
6.3 Non-Abelian Yang-Mills Fields

Following from the relationship of the U (1) electromagnetic field strength with the com-
mutator of the U (1) covariant derivatives acting on scalars, we consider the commutator
of the fundamental covariant derivative Dµ acting on Φ(x) ∈ V , which transforms under
the fundamental representation as Φ → Φ0 = gΦ:

[Dµ , Dν ]Φ = (∂µ + Aµ ) ∂ν Φ + Aν Φ − (∂ν + Aν ) ∂µ Φ + Aµ Φ

= ∂µ Aν − ∂ν Aµ + [Aµ , Aν ] Φ (6.37)
Definition 53. The non-abelian Yang-Mills field strength is
Fµν = ∂µ Aν − ∂ν Aµ + [Aµ , Aν ] (6.38)
– 118 –
Note that as Aµ ∈ L(g) it follows that Fµν ∈ L(G). Note also that by construction
0 is the transformed
[Dµ , Dν ]Φ transforms like Φ under a gauge transformation. Hence if Fµν
gauge field strength, then Fµν0 Φ0 = gF Φ. As this must hold for all Φ, we find
µν
0
Fµν = gFµν g −1 (6.39)
so that F transforms like the homogeneous part of Aµ .

Exercise: Verify this transformation rule for Fµν directly from the definition Fµν =
∂µ Aν − ∂ν Aµ + [Aµ , Aν ] together with the transformation rule of Aµ .
Lemma 13. The non-abelian field strength satisfies
Dµ Dν F µν = 0 (6.40)
where here Dµ is the adjoint covariant derivative.

Proof
Dµ Dν F µν = Dµ (∂ν F µν + [Aν , F µν ])
= ∂µ ∂ν F µν + [Aµ , ∂ν F µν ] + ∂µ [Aν , F µν ] + [Aµ , [Aν , F µν ]]
= [Aµ , Dν F µν ] + [∂µ Aν , F µν ] + [Aν , ∂µ F µν ]
1
= [Aµ , Dν F µν ] + [Fµν − [Aµ , Aν ], F µν ] + [Aν , Dµ F µν − [Aµ , F µν ]]
2
1
= [Aµ , Dν F ] + [Aν , Dµ F µν ] − [[Aµ , Aν ], F µν ] − [Aν , [Aµ , F µν ]]
µν
2
1
= − [[Aµ , Aν ], F ] − [Aν , [Aµ , F µν ]]
µν
2
=0 ( using the Jacobi identity) (6.41)
as required.
6.3.1 The Yang-Mills Action

Definition 54. The non-abelian Yang-Mills Lagrangian is
1
L= κ(Fµν , F µν ) (6.42)
4e2
where κ is the Killing form of the compact matrix Lie group G.
Proposition 34. The non-Abelian Yang-Mills Lagrangian is gauge invariant

Proof
Under a gauge transformation
κ(Fµν , F µν ) −→ κ(gFµν g −1 , gF µν g −1 ) (6.43)

Suppose that X, Y, Z ∈ L(G). Then for t ∈ R, compute
d
κ(etZ Xe−tZ , etZ Y e−tZ ) = κ(−etZ [X, Z]e−tZ , etZ Y e−tZ ) + κ(etZ Xe−tZ , −etZ [Y, Z]e−tZ )
dt
– 119 –
= κ(−[etZ Xe−tZ , Z], etZ Y e−tZ ) + κ(etZ Xe−tZ , −[etZ Y e−tZ , Z])
= −κ(etZ Xe−tZ , [Z, etZ Y e−tZ ]) + κ(etZ Xe−tZ , −[etZ Y e−tZ , Z])
=0 (6.44)
where we have used the associativity of the Killing form. Therefore
κ(eZ Xe−Z , eZ Y e−Z ) = κ(X, Y ) (6.45)

and hence it follows that
κ(gFµν g −1 , gF µν g −1 ) = κ(Fµν , F µν ) (6.46)
as required.
Note that the coupling constant e plays an important role in the dynamics. If one
attempts to rescale A so that Aµ = eÂµ , it is possible to eliminate the explicit factor of e
from the Yang-Mills Lagrangian, and write
1 1
κ(Fµν , F µν ) = κ(F̂µν , F̂ µν ) (6.47)
4e2 4
where here F̂µν = ∂µ Âν −∂ν Âµ +e[Âµ , Âν ]. Although the explicit e-dependence of the Yang-
Mills Lagrangian appears to have been removed, observe that the gauge field strength now
has an e-dependent term, which arises from the commutator which is quadratic in A. So the
dependence on e in the non-abelian theory cannot be removed by rescaling. If, nevertheless,
one performs this rescaling (and then drops theˆon all terms), then the generic covariant
derivative is modified via Dµ Φ = ∂µ Φ + ed(Aµ )Φ, and the gauge potential transformation
rule is also modified: Aµ → A0µ = gAµ g −1 − e−1 ∂µ gg −1 . Whether e appears as an overall
factor in the Yang-Mills Lagrangian, or within the covariant derivative and gauge field
strength, depends on convention. Until stated otherwise, we shall however retain the 4e12
outside the Lagrangian, and work with the un-rescaled gauge fields.
If the representation D of G is unitary, then one can couple addtional gauge invariant
scalar terms to the Yang-Mills Lagrangian:
1 1
L= 2
κ(Fµν , F µν ) + (Dµ Φ)† Dµ Φ − V (Φ† Φ) (6.48)
4e 2
where Φ → D(g)Φ under a gauge transformation.
6.3.2 The Yang-Mills Equations

Consider a first order variation to the gauge potential Aµ → Aµ +δAµ . Under this variation
Fµν → Fµν + (∂µ δAν + [Aµ , δAν ]) − (∂ν δAµ + [Aν , δAµ ]) (6.49)
1
κ(Fµν , F µν )d4 x is given
R
Hence, the first order variation of the Yang-Mills action S = 4e2
by
Z
1
δS = 2 κ(δFµν , F µν )d4 x
2e
– 120 –
Z
1
= κ(∂µ δAν + [Aµ , δAν ], F µν )d4 x
e2Z
1
= − 2 κ(δAν , ∂µ F µν + [Aµ , F µν ])d4 x + surface terms (6.50)
e
where we have made use of the associativity of the Killing form. Neglecting the surface
terms (assuming that the solutions are sufficiently smooth and fall off sufficiently fast at
infinity), and requiring that δS = 0 for all variations δAν we obtain the Yang-Mills field
equations
∂µ F µν + [Aµ , F µν ] = 0 (6.51)
or equivalently
Dµ F µν = 0 (6.52)
where here Dµ is the adjoint covariant derivative.
6.3.3 The Bianchi Identity

There is a non-abelian generalization of the identity λµνρ ∂µ fνρ = 0 in electromagnetism.
Let Dµ be the adjoint covariant derivative. Consider
Dµ Fνρ = ∂µ (∂ν Aρ − ∂ρ Aν + [Aν , Aρ ]) + [Aµ , ∂ν Aρ − ∂ρ Aν + [Aν , Aρ ]]

= ∂µ (∂ν Aρ − ∂ρ Aν ) − ([Aµ , ∂ρ Aν ] + [Aρ , ∂µ Aν ])
+ [Aµ , ∂ν Aρ ] + [Aν , ∂µ Aρ ] + [Aµ , [Aν , Aρ ]] (6.53)
Consider the contraction λµνρ Dµ Fνρ . As the terms ([Aµ , ∂ρ Aν ] + [Aρ , ∂µ Aν ]) are sym-
metric in µ, ρ and the terms [Aµ , ∂ν Aρ ] + [Aν , ∂µ Aρ ] are symmetric in µ, ν, these terms give
no contribution to λµνρ Dµ Fνρ . Also,
λµνρ [Aµ , [Aν , Aρ ]] = 0 (6.54)
from the Jacobi identity, and just as for electromagnetism,
λµνρ ∂µ (∂ν Aρ − ∂ρ Aν ) = 0 (6.55)
because the partial derivatives commute with each other. Hence we find λµνρ Dµ Fνρ = 0,
or equivalently
Dµ Fνρ + Dν Fρµ + Dρ Fµν = 0 (6.56)
This equation is called the Bianchi identity.
This can be used to prove a Jacobi identity for all covariant derivatives. Suppose that
Dµ is the covariant derivative associated with the representation D, and Φ ∈ V transforms
under the representation D as Φ → Φ0 = D(g)Φ. Let d be the induced representation of
L(G).
Then note that
Dν Dλ Φ = Dν (∂λ + d(Aλ )Φ)

= ∂ν (∂λ Φ + d(Aλ )Φ) + d(Aν )∂λ Φ + d(Aν )d(Aλ )Φ
– 121 –

= ∂ν ∂λ Φ + d(Aλ )∂ν Φ + d(Aν )∂λ Φ + d(∂ν Aλ ) + d(Aν )d(Aλ ) Φ (6.57)
and hence
[Dν , Dλ ]Φ = d(Fνλ )Φ (6.58)

Then
[Dµ , [Dν , Dλ ]]Φ = Dµ (d(Fνλ )Φ) − d(Fνλ )Dµ Φ

= ∂µ (d(Fνλ )Φ) + d(Aµ )d(Fνλ )Φ − d(Fνλ )(∂µ Φ + d(Aµ )Φ)

= ∂µ d(Fνλ ) + [d(Aµ ), d(Fνλ )] Φ
= d(Dµadj Fνλ )Φ (6.59)
where Dµadj denotes the adjoint covariant derivative.

Hence

[Dµ , [Dν , Dλ ]] + [Dν , [Dλ , Dµ ]] + [Dλ , [Dµ , Dν ]] Φ = d(Dµadj Fνλ + Dνadj Fλµ + Dλadj Fµν )Φ
=0 (6.60)
using the Bianchi identity on F . As this must hold for all Φ we obtain the Jacobi
identity for covariant derivatives:
[Dµ , [Dν , Dλ ]] + [Dν , [Dλ , Dµ ]] + [Dλ , [Dµ , Dν ]] = 0 (6.61)
6.4 Yang-Mills Energy-Momentum

The energy momentum tensor of the Yang-Mills field strength is
1
Tµν = κ(Fµλ , Fν λ ) − ηµν κ(Fρσ , F ρσ ) (6.62)
4
This differs from the canonical energy-momentum tensor by a total derivative. Observe
that Tµν is gauge invariant by construction.
Proposition 35. If the Yang-Mills field strength F satisfies the Yang-Mills field equations,
the energy-momentum tensor satisfies
∂ µ Tµν = 0 (6.63)
Proof
1
∂ µ Tµν = ∂ µ (κ(Fµλ , Fν λ )) − ∂ν (κ(Fρσ , F ρσ ))
4
1
= κ(∂ Fµλ , Fν ) + κ(Fµλ , ∂ µ Fν λ ) − κ(∂ν Fρσ , F ρσ )
µ λ
2
= κ(Dµ Fµλ − [Aµ , Fµλ ], Fν λ ) + κ(Fµλ , Dµ Fν λ − [Aµ , Fν λ ])
– 122 –
1
− κ(Dν Fρσ − [Aν , Fρσ ], F ρσ ) (6.64)
2
where here Dµ is the adjoint covariant derivative. Note that by the Bianchi identity
κ(Fµλ , Dµ Fν λ ) = κ(Fµλ , −Dν F λµ − Dλ F µ ν )

= κ(Fµλ , −Dν F λµ ) − κ(Fµλ , Dµ Fν λ ) (6.65)
and so
1
κ(Fµλ , Dµ Fν λ ) = κ(Fµλ , Dν F µλ ) (6.66)
2
Hence
∂ µ Tµν = κ(Dµ Fµλ , Fν λ ) − κ([Aµ , Fµλ ], Fν λ ) − κ(Fµλ , [Aµ , Fν λ ])

1
+ κ([Aν , Fρσ ], F ρσ ) (6.67)
2
However, using the associativity of the Killing form
κ([Aν , Fρσ ], F ρσ ) = κ(Aν , [Fρσ , F ρσ ]) = 0 (6.68)
and
κ([Aµ , Fµλ ], Fν λ ) + κ(Fµλ , [Aµ , Fν λ ]) = κ([Aµ , Fµλ ], Fν λ ) + κ([Fµλ , Aµ ], Fν λ )

=0 (6.69)
hence
∂ µ Tµν = κ(Dµ Fµλ , Fν λ ) = 0 (6.70)
from the Yang-Mills field equations, as required.
6.5 The QCD Lagrangian

The QCD Lagrangian consists of a non-abelian Yang-Mills gauge theory coupled to fermions.
The gauge group is G = SU (3)colour . Take generators Ta of G; as G is compact, these
may be chosen so that the Killing form has components
κab = −δab (6.71)
where a = 1, . . . , 8 (SU (3) is an 8-dimensional Lie group).

The Lagrangian is given by
8 6 6
1 X a aµν X X
LQCD =− Fµν F +i Ψ̄A µ B a B
f γ (δA ∂µ + eA µ (Ta )A )ΨBf − mf Ψ̄A
f ΨAf (6.72)
4
a=1 f =1 f =1
– 123 –
Here we have decomposed the non-abelian gluon field strength F and potential A into
components
a
Fµν = Fµν Ta , Aµ = Aaµ Ta (6.73)
where
a
Fµν = ∂µ Aaν − ∂ν Aaµ + e[Aµ , Aν ]a = ∂µ Aaν − ∂ν Aaµ + ecbc a Abµ Acν (6.74)
The indices f = 1, . . . , 6 are flavour indices; whereas the indices A, B = 1, 2, 3 are
SU (3)colour indices. The SU (3)colour structure constants are cab c . mf is the mass of the f -
flavour quark. The SU (3) gauge coupling is e; observe that the gauge potentials have been
re-scaled as mentioned previously in order to remove the gauge coupling factor from the
Yang-Mills action; although this means that e now appears in the fundamental covariant
derivative via ∂µ + eAµ , and also in the gauge transformations via
Aµ → gAµ g −1 − e−1 ∂µ gg −1 (6.75)
where g ∈ G.
The ΨAf are fermionic fields associated to the quarks- they are 4-component Dirac
spinors. We set ΨA AB Ψ A A † 0
f =δ Bf , and Ψ̄f ≡ (Ψf ) γ .
Gauge Invariance
The Yang-Mills term − 41 8a=1 Fµν a F aµν = 1 κ(F , F µν ) in the QCD action is auto-
P
4 µν
matically gauge invariant by construction.
It remains to consider the terms involving fermions: the fermions transform under the
fundamental representation of SU (3)colour via
ΨAf → gA B (x)ΨBf (6.76)

where g(x) ∈ SU (3). It is then straightforward to see that the covariant derivative
(δA B∂ B
µ + e(Aµ )A )ΨBf then transforms as
(δA B ∂µ + e(Aµ )A B )ΨBf → gA C (δC B ∂µ + e(Aµ )C B )ΨBf (6.77)

Also, the Ψ̄A
f transform as
AP ∗ Q
Ψ̄A
f → Ψ̄Qf δ (g )P (6.78)
Hence we see that under these gauge transformations
AP ∗ Q
Ψ̄A
f ΨAf → Ψ̄Qf δ (g )P gA C ΨCf
QC
= Ψ̄Qf δ ΨCf
= Ψ̄A
f ΨAf (6.79)
where g ∈ SU (3) implies δ AP (g ∗ )P Q gA C = δ QC .

Similarly, under this transformation
AP ∗ Q
Ψ̄A B B
f (δA ∂µ + e(Aµ )A )ΨBf → Ψ̄Qf δ (g )P gA C (δC B ∂µ + e(Aµ )C B )ΨBf
– 124 –
= Ψ̄A B B
f (δA ∂µ + e(Aµ )A )ΨBf (6.80)
It follows that all terms in the Lagrangian are gauge invariant.

Equations of Motion
There are two sets of equations of motion. Varying the gauge potentials Aaν we obtain
the variation of the action
Z 6
X
d4 x δAaν ∂µ F aµν + e[Aµ , F µν ]a + ie Ψ̄A ν B

δS = f γ (Ta )A ΨBf (6.81)
f =1
from which we obtain
6
X
aµν
∂µ F + ecbc Abµ F cµν
a
+ ie Ψ̄A ν B
f γ (Ta )A ΨBf = 0 (6.82)
f =1
Defining
8 X
X 6
J ν = −ie Ψ̄A ν B
f γ (Ta )A ΨBf Ta (6.83)
a=1 f =1
this can be rewritten as
Dµ F µν = J ν (6.84)
Observe that Dµ J µ = 0 as a consequence of Lemma 14.
There are also fermionic equations of motion; these may be obtained by varying Ψ̄A f
and ΨBf in the action. These can be varied independently; from the variation of Ψ̄A
f (not
varying ΨBf ) we obtain the equation:
iγ µ δA B ∂µ + eAaµ (Ta )A B ΨBf − mf ΨAf = 0

(6.85)
Next vary ΨBf (not varying Ψ̄A
f )- then the action variation is
Z
d4 x
iΨ̄A µ B a B B

δS = f γ δA ∂µ + eA (T
µ a A) − mf f δΨBf
Ψ̄
Z
4 A µ B A µ a B B
= d x − i∂µ Ψ̄f γ δA + ieΨ̄f γ Aµ (Ta )A − mf Ψ̄f δΨBf (6.86)
Hence we obtain
−i∂µ Ψ̄A µ B A µ a B B
f γ δA + ieΨ̄f γ Aµ (Ta )A − mf Ψ̄f = 0 (6.87)
Take the hermitian transpose of the above equation we obtain
∗ B A
iγ 0 γ µ ∂µ ΨB 0 µ a 0 B
f − ieγ γ Aµ (Ta )A Ψf − mf γ Ψf = 0 (6.88)
where we have made use of the identity γ µ† = γ 0 γ µ γ 0 . As Ta is anti-hermitian, we
therefore recover (6.85).
– 125 –

Symmetry and Particle Physics: Michaelmas Term 2007

Uploaded by

Copyright:

Available Formats

Symmetry and Particle Physics: Michaelmas Term 2007

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Symmetry and Particle Physics: Michaelmas Term 2007

Uploaded by

Copyright:

Available Formats

Symmetry and Particle Physics

Michaelmas Term 2007

1. Introduction to Symmetry and Particles 5

2. Elementary Theory of Lie Groups and Lie Algebras 14

3. SU(2) and Isospin 48

4. SU(3) and the Quark Model 65

6. Gauge Theories 114

General Particle Physics Books

? D. H. Perkins, Introduction to High energy Physics, 4th ed., CUP (2000).

• B. R. Martin and G. Shaw, Particle Physics, 2nd ed., Wiley (1998).

Lie Algebra books written for Physicists

? H. Georgi, Lie Algebras in Particle Physics, Perseus Books (1999).

• J. Cornwell, Group Theory in Physics, (Volume 2), Academic Press (1986).

Pure Mathematics Lie Algebra books

? H. Samelson, Notes on Lie Algebras, Springer (1990).

References for Spacetime Symmetry and Gauge Theory Applications

? S. Pokorski, Gauge Field Theories, 2nd ed., CUP (2000).

• S. Weinberg, The Quantum Theory of Fields, (Book 1), CUP (2005).

? J. Buchbinder and S. Kuzenko, Ideas and Methods of Supersymmetry and Supergrav-

Symmetry simplifies the description of physical phenomena. It plays a particularly impor-

1.1 Elementary and Composite Particles

Electric Charge (e)

Electric Charge (e)

For example, the pions are given by

1.2.1 The Strong Interaction

• Dominant in scattering processes involving just hadrons. For example, pp → pp is

Properties of the Strong Interaction:

iii) Physics is unchanged if all particles are replaced by antiparticles.

The dynamical theory governing the strong interactions is Quantum Chromodynamics

1.2.2 Electromagnetic Interactions

Electron-positron annihilation can also produce particles such as charmonium or bot-

Neutrinos have no electromagnetic or strong interactions.

1.2.3 The weak interaction

The flavour changes within one family are dominant; e.g.

whereas changes between families, like u ↔ s and u ↔ b are “Cabibbo suppressed”.

In any process in which a photon is exchanged, it is possible to have a Z 0 boson

1.2.4 Typical Hadron Lifetimes

Interaction Lifetime (s)

1.3 Conserved Quantum Numbers

Le = N (e− ) − N (e+ ) + N (νe ) − N (ν̄e )

In electromagnetic interactions, where there are no neutrinos involved, conservation

Definition 2. The four quark numbers S, C, B̃ and T corresponding to strangeness,

S = −(N (s) − N (s̄))

Definition 3. The baryon number B is defined by

2.1 Differentiable Manifolds

is smooth (infinitely differentiable) as a function on Rn .

Definition 6. Suppose M is a m-dimensional smooth manifold, and N is a n-dimensional

zα,A = xα × yA : U α × W A → Rm+n (2.2)

Definition 7. Suppose M is a m-dimensional smooth manifold, and N is a n-dimensional

Definition 8. A smooth curve on a manifold M is a map γ : (a, b) → M where (a, b) is

may be differentiated arbitrarily often.

2.2 Lie Groups

ii) For every g ∈ G there exists an inverse g −1 ∈ G such that g • g −1 = g −1 • g = e.

i) The map • : G × G → G given by (g1 , g2 ) → g1 • g2 is a smooth map.

ii) The inverse map G → G given by g → g −1 is a smooth map

ii) SL(n, F) = {M ∈ GL(n, F) : det M = 1}

iii) O(n) = {M ∈ GL(n, R) : M M T = In }

iv) U (n) = {M ∈ GL(n, C) : M M † = In }, where † is the hermitian transpose.