Supersymmetric Field Theory

University of Cambridge Part III Mathematical Tripos

David Tong
Department of Applied Mathematics and Theoretical Physics,
Centre for Mathematical Sciences,
Wilberforce Road,
Cambridge, CB3 OBA, UK

http://www.damtp.cam.ac.uk/user/tong/susy.html
[email protected]
Recommended Books and Resources

Here is a collection of useful textbooks on supersymmetry.

• Wess and Bagger “Supersymmetry”

This is a strange little book, with chapters that are 2 pages long followed by several
pages of key equations. It’s not particularly good for learning the subject, but makes
a remarkably useful reference guide.

• Bailin and Love “Supersymmetric Gauge Field Theory and String Theory”

Probably the best book covering the basics of supersymmetric Lagrangians.

• Dan Freedman and Toine Van Proeyen “Supergravity”

As the name suggests, this book is mostly focussed on supergravity rather than global
supersymmetry. But it kicks off with a really excellent description of classical field
theory. The section on spinors in various dimensions is particularly useful.

• Steven Weinberg “The Quantum Theory of Fields, Volume III: Supersymmetry”

The third volume of Weinberg’s magnum opus covers supersymmetry. As always, it

contains many important things that are difficult to find elsewhere. As always, these
things are sometimes frustratingly buried in unconventional notation and dressed with
more indices than you can shake a stick at.

• John Terning “Modern Supersymmetry: Dynamics and Duality”

This is one of the few books (possibly the only book) that describes the quantum dy-
namics of supersymmetric field theories, rather than just their classical action. (Wein-
berg has a chapter on the Seiberg-Witten solution, but it feels like his heart isn’t in
it and any mention of Seiberg duality is noticeably absent.) There are, fortunately,
many lecture notes that make up for the deficiency. You can find links on the course
webpage.
Contents
1 Introduction 3
1.1 A First Look at Supersymmetry 7

2 The Supersymmetry Algebra 9

2.1 The Lorentz Group 9
2.1.1 Spinors and SL(2, C) 12
2.1.2 Lagrangians for Spinors 18
2.1.3 The Poincaré Group and its Extensions 20
2.2 The Supersymmetry Algebra 23
2.2.1 R-Symmetry 25
2.2.2 A Consequence: Energy is Positive 25
2.3 Representations on Particle States 27
2.3.1 Representations of the Poincaré Group 28
2.3.2 Massless Representations 30
2.3.3 Massive Representations 33
2.4 Extended Supersymmetry 36
2.4.1 Massless Representations 37
2.4.2 Massive Representations and BPS Bounds 40
2.4.3 Supersymmetry in Other Dimensions 42

3 Chiral Superfields 48
3.1 Superspace 48
3.1.1 The Geometry of Superspace 49
3.1.2 Superfields 52
3.1.3 Constraining Superfields 56
3.1.4 Chiral Superfields 58
3.2 And. . . Action 60
3.2.1 Integrating Over Superspace 60
3.2.2 The Action for Chiral Superfields 62
3.2.3 Supersymmetry of the Wess-Zumino Model Revisited 66
3.2.4 Non-Linear Sigma Models 68
3.3 Non-Renormalisation Theorems 71
3.3.1 R-Symmetry Revisited 72
3.3.2 The Power of Holomorphy 73
3.3.3 Integrating Out Heavy Fields 78

–i–
 3.3.4 A Moduli Space of Vacua 79
3.4 A First Look at Supersymmetry Breaking 82
3.4.1 The Goldstino 83
3.4.2 The Witten Index 85
3.4.3 The O’Raifeartaigh Model 88
3.4.4 R-symmetry and the Nelson-Seiberg Argument 92
3.4.5 More Ways to (Not) Break Supersymmetry 94

4 Supersymmetric Gauge Theories 98

4.1 Abelian Gauge Theories 98
4.1.1 The Field Strength and Action 99
4.1.2 Supersymmetric QED 101
4.2 Non-Abelian Gauge Theories 105
4.2.1 Super Yang-Mills 105
4.2.2 Supersymmetric QCD 107
4.3 The Moduli Space of Vacua 109
4.3.1 The Moduli Space of SQED 112
4.3.2 The Moduli Space of SQCD 116
4.3.3 Briefly, Gauged Linear Sigma Models in 2d 121
4.4 Extended Supersymmetry 124
4.4.1 N = 2 Theories 125
4.4.2 N = 4 Theories 128

5 Boot Camp: Quantum Gauge Dynamics 129

5.1 Strong Coupling 129
5.1.1 The Beta Function 129
5.1.2 Confinement and the Mass Gap 131
5.1.3 Adding Matter 134
5.1.4 Chiral Symmetry Breaking 136
5.1.5 Phases of Massless QCD 137
5.2 Anomalies 140
5.2.1 Gauge Anomalies 142
5.2.2 Chiral (or ABJ) Anomalies 144
5.2.3 ’t Hooft Anomalies 146
5.3 Instantons 149

– ii –
6 Supersymmetric QCD 154
6.1 Super Yang-Mills 157
6.1.1 Confinement and Chiral Symmetry Breaking 157
6.1.2 The Witten Index 159
6.1.3 A Superpotential 162
6.2 A First Look at SQCD 164
6.2.1 Symmetries 165
6.2.2 Runaway for Nf < Nc 167
6.2.3 Adding Masses 169
6.2.4 The Potential at Weak Coupling 173
6.3 A Second Look at SQCD 176
6.3.1 A Deformed Moduli Space for Nf = Nc 177
6.3.2 ’t Hooft Anomaly Matching 179
6.3.3 Confinement Without χSB for Nf = Nc + 1 184
6.4 A Peek in the Conformal Window 188
6.4.1 Facts About Conformal Field Theories 189
6.4.2 Facts About Superconformal Field Theories 192
6.4.3 The Conformal Window for SQCD 194
6.5 Seiberg Duality 196
6.5.1 Matching Symmetries 197
6.5.2 Completing the Phase Diagram for SQCD 200
6.5.3 Deformations of the Theories 203
6.5.4 Why Seiberg Duality is Electromagnetic Duality 207

7 More Supersymmetric Gauge Dynamics 209

7.1 Other Gauge Groups 209
7.1.1 Sp(N ) Quantum Dynamics 209
7.1.2 Seiberg Duality 214
7.1.3 SU (2) Gauge Theory Revisited 215
7.2 A Chiral Gauge Theory 217
7.2.1 SU (N ) with a Symmetric 217
7.2.2 A Chiral Duality 219
7.2.3 Briefly, the Konishi Anomaly 223
7.2.4 Briefly, a-Maximisation 225
7.3 Dynamical Supersymmetry Breaking 226
7.3.1 The SU (3) × SU (2) Model 227
7.3.2 The Quantum Moduli Space Revisited 230

–1–
 Acknowledgements

These lecture notes owe a debt to Graham Shore, from whom I first learned supersym-
metry, and to Philip Argyres’ wonderful notes on the subject. I’m also grateful to Ben
Allanach and Fernando Quevedo, both previous lecturers of the supersymmetry course
in Cambridge.

This is one of the more advanced courses in Part III. It assumes a familiarity with
quantum field theory, in particular the renormalisation group. You will also need to be
comfortable with some group theory.

Spinor Conventions

We work in Minkowski space with signature (+, −, −, −). Spinor indices are raised and
lowered with ψ α = ϵαβ ψβ and ψ̄ α̇ = ϵα̇β̇ ψ̄β̇ where the invariant, anti-symmetric tensor
is
!
0 1
ϵαβ = ϵα̇β̇ = −ϵαβ = −ϵα̇β̇ =
−1 0

Left-handed spinors are contracted as ψχ = ψ α χα and right-handed spinors are con-

tracted as ψ̄ χ̄ = ψ̄α̇ χ̄α̇ . Sigma matrices are defined by

(σ µ )αα̇ = (1, σ i )αα̇ and (σ̄ µ )α̇α = ϵαβ ϵα̇β̇ σβµβ̇ = (1, −σ i )α̇α

and the generators of the Lorentz group in the left-handed and right-handed spinor
representation are, respectively,
i µ ν i µ ν
(σ µν )αβ = (σ σ̄ − σ ν σ̄ µ )αβ and (σ̄ µν )α̇β̇ = (σ̄ σ − σ̄ ν σ µ )α̇β̇
4 4

–2–
1 Introduction
Supersymmetry is the name given to a novel symmetry that relates bosons and fermions.
In many ways it is a surprise that such a symmetry could exist at all. This is because
bosons and fermions are, to put it mildly, different.

Bosons are gregarious. Put many of them in a box and they huddle together to form
a macroscopic quantum object called a Bose-Einstein condensate. In contrast, fermions
are loners, an isolation enforced by the Pauli exclusion principle. Put many fermions
in a box and you get a more familiar, but ultimately even stranger, state of quantum
matter called a Fermi surface.

Within the framework of relativistic quantum field theories, the difference between
fermions and bosons is even more stark. Fermions are matter particles. Bosons are
force carriers. Any symmetry that relates the two must somehow entail a unification
of matter and force.

Of course, we know from our earlier lessons on Quantum Field Theory that the dis-
tinction between bosons and fermion can be traced to something that is, in some sense,
rather minor. They differ only by the simple matter of ℏ/2 in their angular momen-
tum, with the spin-statistics theorem then doing the heavy lifting that ensures the
resulting particles have such different properties. However, this too highlights just how
unusual supersymmetry must be. The angular momentum of a particle is a property
that follows from the symmetries of spacetime. Anything that relates particles with
different angular momentum must involve some kind of extension of the symmetries of
spacetime. And that sounds interesting!

All of this means that it’s not at all obvious that something like supersymmetry can
exist and we should, if nothing else, be curious about how it can come about. But why
else should we care? In the rest of this introduction, I give three reasons why studying
supersymmetric quantum field theories is worthwhile.

Reason 1: Strongly Interacting Quantum Field Theories

Quantum field theory is hard. This is particularly true when coupling constants, which
specify the strength of interactions, are not small. This means that we can no longer
understand the physics using the familiar methods of perturbation theory and Feynman
diagrams. In this case, the word “hard” typically means “no one knows how to solve
it”.

–3–
 Supersymmetric theories are not wildly different from other quantum field theories.
They have a carefully curated collection of fields, with some interactions tuned to take
certain values, but otherwise they exhibit many of the strongly coupled phenomena
expected of any other quantum field theory. The magic of supersymmetry, however,
is that in many cases we are able to make exact statements about the properties of
the theory. This is because supersymmetry places certain restrictions on the kind of
dynamics that can occur. Fortuitously, it turns out that these restrictions are not
strong enough to stop interesting things happening, but are strong enough to allow us
to solve certain aspects of the theory. In this way, supersymmetric field theories provide
an important collection of toy models that allow us to understand what quantum field
theory can do in regimes where we would otherwise have very little control.

Here is an example. The theory of the strong nuclear force, QCD, exhibits a re-
markable property known as confinement. Quarks are always trapped inside hadrons
and we never see isolated quarks on their own. There is no doubt that the theory of
QCD has this property – we can see it clearly in numerical simulations – but we are
a long way from being able to prove confinement from first principles. However, there
are supersymmetric gauge theories, similar to QCD but with slightly different matter
content, where confinement can be proven analytically. (This follows from the famous
Seiberg-Witten solution of N = 2 supersymmetric theories.) While the supersymmet-
ric proof of confinement is not directly applicable to real-world QCD, it nonetheless
gives us good intuition for how confinement might proceed in that context.

These lectures will very much be given in the spirt of using supersymmetry to tell us
interesting things about strongly coupled quantum field theories. We will learn about
topics that exist for real word QCD, such as confinement and chiral symmetry breaking,
and see how these manifest themselves in more tractable supersymmetric theories. We
will also learn about novelties that appear not to be of relevance for QCD but give
us an insight into what strongly interacting quantum field theories can do. Foremost
among these novelties is the concept of duality, the idea that two very different looking
quantum field theories may, in fact, describe the same physics.

Reason 2: Mathematics
As our understanding of supersymmetric field theories grew, increasingly sophisticated
mathematical constructs were found lurking within them. These are primarily, but not
exclusively, ideas from geometry.

This link between supersymmetry and mathematics starts with some simple quantum
mechanical models whose solutions give new perspectives on, among other things, Morse

–4–
theory and index theorems. But the real fun starts when we turn to supersymmetric
field theories. Understanding supersymmetric field theories in d = 1 + 1 dimensions
led to the discovery of mirror symmetry, a relationship between topologically distinct
manifolds. As we move to higher dimensional quantum field theories, we find ever more
elaborate structures, some of which are known to mathematicians and some of which
are novel. It is clear that there is much more to uncover.

We won’t have anything to say about the connection to mathematics in these lectures,
although we will stumble upon the concept of Kähler geometry as we proceed which at
least gives a feel for how interesting geometric concepts arise naturally from supersym-
metry. The companion lectures on Supersymmetric Quantum Mechanics have more
of an eye towards the mathematical aspects of supersymmetry, albeit without getting
very deep into the subject.

Reason 3: Our World

The million dollar question is: does supersymmetry have anything to do with our
world? The rather disappointing answer is: we don’t know.

There is certainly no experimental evidence that supersymmetry is a symmetry of

nature at the fundamental level. Moreover, it’s not for want of trying. To fill in the
details, I’ll first explain what it would mean for our world to be supersymmetric. Then
I’ll explain what reasons we have (or had!) for thinking that this might be the case.

In any supersymmetric theory, particles come in pairs – one a boson, the other a
fermion – and this pair of particles share many of their properties, such as their masses
and the forces that they experience. You don’t need to build an LHC to realise that
our world most certainly does not have this property! There is no bosonic particle
with the same mass and charge as an electron; no massless fermionic particle with the
same properties as the photon. (No, the neutrino doesn’t do it!) There is, in short, no
supersymmetry.

However, not all symmetries are manifest in the world around us. This is because
of the phenomenon of symmetry breaking in which the dynamics of the theory make a
choice which masks the underlying symmetry. There are many examples of symmetry
breaking that we know take place, some mundane and familiar, others more exotic. Here
are two. In a magnet, all the spins align in a given direction, breaking the underlying
rotation symmetry. In the Standard Model, electroweak symmetry is broken by the
Higgs boson ensuring, among other things, that the (left-handed) electron and neutrino
look very different to our low-energy eyes despite the fact that they are indistinguishable
at high energies.

–5–
 It may well be that supersymmetry is a symmetry of our world but is broken and
so hidden at low-energies. If this is the case, the breaking comes with an energy scale
that we will call Msusy . All of the superpartners – the other half of each boson/fermion
pair – would get a mass that sits somewhere around Msusy . So to answer the question
of whether supersymmetry exists in nature we must also address the partner question:
what is the scale of Msusy ?

For many years, supersymmetry was viewed as the most promising candidate for
physics beyond the Standard Model, with Msusy ≈ 1 TeV. At this scale, supersymmetry
provides a compelling solution to the hierarchy problem (the question of why the Higgs
mass is not driven to higher scales by quantum fluctuations). Furthermore, if you adopt
this solution then it comes with a number of happy consequences, from the unification
of coupling constants to enticing candidates for dark matter.

However, with the advent of the LHC we have now explored the TeV scale and there
is no sign of the predicted superpartners. It’s not quite game over: it may well be that
these extra particles are lurking just around the corner, tantalisingly out of reach of
our current accelerator and will be found as we go to higher energies. But it’s certainly
fair to say that the parameter space of allowed theories has shrunk dramatically, as
have our reasons for believing in supersymmetry at the TeV scale. This means that if
supersymmetry is a symmetry of our world, it now appears to be broken at some scale
Msusy ≳ 1 TeV. But where?

There is reason to think that supersymmetry might show up by the time we reach
the Planck scale Mpl ≈ 1015 TeV. This reason is string theory. Of course, we don’t
know that string theory is the right theory of quantum gravity but it is presently the
only viable candidate where a microscopic quantum theory gives the Einstein equations
emerging at large distances. And string theory appears to require supersymmetry. (I
include the word “appears” here because there are some open questions about bosonic
(i.e. non-supersymmetric) string theory that we don’t have a good handle on and it
may be premature to throw this out as a viable theory.)

So if you buy into string theory, then you’ll most likely want supersymmetry to be
manifest by the time you get Mpl . And, as we’ve seen above, it looks like it should be
broken at some scale Msusy ≳ 1 TeV. But there are 15 orders of magnitude between the
TeV scale and the Planck scale. Where in this range should we expect supersymmetry
to be broken if not at the TeV scale, or just above it, to provide a solution to the
hierarchy problem? Sadly, I don’t think that we have any good idea, and there are

–6–
no hints from nature that it is more useful to have Msusy at some large scale ≫ TeV
rather than another.

This leaves us with the current situation, one of no small befuddlement about what
role, if any, supersymmetry has to play in our world. Given this, in these lectures we
won’t make any attempt to describe how supersymmetry may appear in our world.
In particular, we will not devote effort to constructing supersymmetric versions of
the Standard Model (the simplest is known as the MSSM where the first M stands
for “minimal” and you can guess the rest) nor will we describe the many subtleties
that come with how supersymmetry might be broken and how this manifests itself.
Instead we will focus on places where supersymmetry has proved invaluable, viewing
the theories as toy models to guide us in our understanding of quantum field theories.

1.1 A First Look at Supersymmetry

To motivate some of what lies ahead, we’ll jump in with a particularly simple super-
symmetric theory. The theory consists of a single, complex scalar ϕ together with a
2-component Weyl fermion ψα . (If you’re unfamiliar with Weyl fermions, we’ll describe
their properties in detail in Section 2.1.)

The following action has kinetic terms for these two fields, together with some care-
fully tuned interactions
" #
Z 2 2 2 †
∂W 1 ∂ W 1 ∂ W
S = d4 x ∂µ ϕ† ∂ µ ϕ − iψσ µ ∂µ ψ̄ − − ψψ − ψ̄ ψ̄ (1.1)
∂ϕ 2 ∂ϕ2 2 ∂ϕ† 2

Here σ µ = (1, σ i ) with σ i the usual collection of three Pauli matrices. Note that
there is a relation between the scalar potential V (ϕ) = |W ′ (ϕ)|2 and the scalar-fermion
interactions, both of which are dictated by a function W (ϕ) known as the superpotential.
If we want a renormalisable theory, this function should be no more than cubic
1 1
W (ϕ) = mϕ2 + λϕ3
2 3
This ensures that the potential is a quartic polynomical, V (ϕ) = |mϕ + λϕ2 |2 , while the
scalar-fermion interactions take the usual Yukawa form ϕψψ. Crucially, the function
W (ϕ) should be holomorphic: it depends only on ϕ and not on ϕ† . This fact will take
on increasing significance as these lectures progress, but for now we will just take this
as given.

–7–
 Even without doing any detailed calculations, we can see that there’s something
curious about the action (1.1): the boson ϕ and the fermion ψ have the same mass
|m|. Usually in quantum field theory, we shouldn’t ascribe too much meaning to such
an observation since masses receive quantum corrections and there’s no guarantee that
the physical masses of two distinct particles will coincide just because the masses in
the Lagrangian are equal. However, for the particular action (1.1), it turns out that
the equality of bosonic and fermionic masses persists in the full quantum theory. This
arises because the action enjoys a rather surprising symmetry, with the infinitesimal
variation given by
√ √ µ √ ∂W †
δϕ = 2ϵψ and δψ = 2iσ ϵ̄ ∂µ ϕ − 2ϵ † (1.2)
∂ϕ
This is our first example of supersymmetry. It is a symmetry that relates the bosonic
field ϕ with the fermionic field ψ. Because ψ is a Grassmann field, while ϕ is not,
the infinitesimal object ϵ, which parameterises the transformation, must also be a
Grassmann-valued Weyl spinor.

You can’t tell just by staring at the action (1.1) that it is invariant under the super-
symmetry transformation (1.2). Instead, it takes a calculation, one that turns out to
be a little bit of a headache. (Some balm for this headache will be offered in Section
3.2.3.)

The action (1.1) is the simplest supersymmetric theory in d = 3 + 1 dimensions. It

is known as the Wess-Zumino model. The existence of such a symmetry opens up a
number of questions. What, if anything, is the symmetry good for? Are there other
theories that also exhibit such symmetry? What properties might they have? All of
these will be answered as these lectures progress.

There is also another question that might have occurred to you: why is it such a pain
to see that the action (1.1) is invariant under supersymmetry? Usually, the existence
of symmetries in an action jumps out at you. Indeed, one of the main advantages of
working with the Lagrangian approach, rather than the Hamiltonian approach, is that
all symmetries are manifest. Typically you need do little more than ensure that various
indices are contracted in the right way. This suggests that there may be a better way
to write the action (1.1) that makes supersymmetry as obvious as any other symmetry.
And there is. Our first task in these lectures – one that will carry us through much of
Sections 2, 3 and 4 – is to better understand the structure behind supersymmetry and
the corresponding supersymmetric actions.

–8–
2 The Supersymmetry Algebra
The purpose of this section is to describe, in mathematical terms, what supersymmetry
actually is. Usually in physics, we think of symmetries as associated to groups. But,
at least for continuous symmetries, these groups have an underlying algebra and often
that contains all the information that we need. So it is with supersymmetry. We will
describe the algebra that underlies supersymmetry and start to explore some of its
representations.

I should warn you that this section will be a little dry in flavour. There will be few
fields and certainly no dynamics. These will come in later sections. But this section
lays the necessary groundwork for the stories that are to come.

2.1 The Lorentz Group

Minkowski space R1,3 is the stage for relativistic quantum field theory. This space
comes equipped with the Minkowski metric

ηµν = diag(+1, −1, −1, −1)

The set of symmetries of Minkowski space include Lorentz transformations of the form
xµ → Λµν xν where

ΛT ηΛ = η

Embedded among these are a couple of discrete transformations: parity with Λ =

diag(1, −1, −1, −1) and time reversal with Λ = diag(−1, 1, 1, 1). The transformations
that are continuously connected to the identity have detΛ = 1 and Λ00 > 0 and form the
Lorentz group SO(1, 3). (The restriction to Λ00 > 0 is sometimes written as SO+ (1, 3).)

Our main goal in this section is the spell out some properties of the spinor represen-
tations of the Lorentz group. In fact, strictly speaking the group SO(1, 3) doesn’t have
any spinor representations. However, there is a closely related group called Spin(1, 3)
that does admit spinors. This is the double cover, in the sense that

SO(1, 3) ∼
= Spin(1, 3)/Z2

where that Z2 is the famous minus sign that spinors pick up under a 2π rotation, a
minus sign that vectors like xµ are oblivious to. The fact that there are spinors in our
world is the statement that the true symmetry group is Spin(1, 3) rather than SO(1, 3).

–9–
 When we introduced spinors in the Quantum Field Theory course, we did so by first
looking at the algebra so(1, 3) that is shared by both groups Spin(1, 3) and SO(1, 3).
A Lorentz transformation acting on a 4-vector can be written as
 
i µν
Λ = exp − ωµν M (2.1)
2
where ωµν are six numbers that specify what Lorentz transformation we’re doing, while
M µν = −M νµ are a choice of six 4 × 4 anti-symmetric matrices that generate the
different Lorentz transformations. The matrix indices are suppressed in the above
expressions; in their full glory we would write (M µν )ρσ . So, for example
0 1 0 0
! 0 0 0 0
!
(M 01 )ρσ = i 1 0 0 0
0 0 0 0
and (M 12 )ρσ = i 0 0 −1 0
0 1 0 0
(2.2)
0 0 0 0 0 0 0 0

(Note that the generators differ by a factor of i from those defined in the Quantum
Field Theory lectures. This is compensated by an extra factor of i in the exponent
(2.1).) The matrices generate the algebra so(1, 3),

[M µν , M ρσ ] = i (η νρ M µσ − η νσ M µρ + η µσ M νρ − η µρ M νσ ) (2.3)

In the lectures on Quantum Field Theory, we then constructed the spinor representa-
tions by first looking at the Clifford algebra of gamma matrices, {γ µ , γ ν } = 2η µν and,
from these, constructing a new representation of the Lorentz algebra (2.3). Here, we’ll
take a slightly different path. It will be useful to first extract a little more information
from the algebra (2.3).

The six different Lorentz transformations naturally decompose into three rotations
Ji and three boosts Ki , defined by
1
Ji = ϵijk Mjk and Ki = M0i
2
where these j, k = 1, 2, 3 indices are summed over, and ϵ123 = +1. The rotation
matrices are Hermitian, with Ji† = Ji while the boost matrices are anti-Hermitian with
Ki† = −Ki . This ensures that the rotations in (2.1) give rise to a compact group while
the boosts are non-compact. From the Lorentz algebra, we find that these generators
obey

[Ji , Jj ] = iϵijk Jk , [Ji , Kj ] = iϵijk Kk , [Ki , Kj ] = −iϵijk Jk

The rotations form an su(2) sub-algebra. That, of course, is to be expected and is

related to the fact that SO(3) ∼
= SU (2)/Z2 .

– 10 –
 We can, however, find two mutually commuting su(2) algebras sitting inside so(1, 3).
For this we take the linear combinations
1 1
Ai = (Ji + iKi ) and Bi = (Ji − iKi )
2 2
Both of these are Hermitian, with A†i = Ai and Bi† = Bi . They obey
[Ai , Aj ] = iϵijk Ak , [Bi , Bj ] = iϵijk Bk , [Ai , Bj ] = 0 (2.4)
But we know about representations of SU (2): they are labelled by an integer or half-
integer j ∈ 12 Z which, in the context of rotations, we call “spin”. The dimension of the
representation is then 2j + 1. The fact that we can find two su(2) sub-algebras of the
Lorentz algebra tells us that all representations must carry two such labels
1
(j1 , j2 ) with j1 , j2 ∈ Z (2.5)
2
and has dimension (2j1 +1)(2j2 +1). We’ll flesh out the meaning of these representations
more below. But for now, we can identify the simplest such representations just by
counting: we have
(0, 0) : scalar
( 12 , 0) : left-handed Weyl spinor
(0, 12 ) : right-handed Weyl spinor
( 21 , 12 ) : vector
(1, 0) : self-dual 2-form
(0, 1) : anti-self-dual 2-form
We see that the smallest representations of the Lorentz group are the left- and right-
handed Weyl spinors. What we call the physical spin of a particle is the quantum
⃗ this is j = j1 + j2 .
number under rotations J:

There’s something a little odd about the our discovery of two su(2) sub-algebras.
After all, it certainly isn’t true that the Lorentz group is isomorphic to two copies of
SU (2). This is because SU (2) is a compact group: keep doing a rotation and you will
eventually get back to where you started. Indeed, two copies of the group SU (2) give
rotation group of Euclidean space R4 .
Spin(4) ∼
= SU (2) × SU (2) with SO(4) ∼
= Spin(4)/Z2
In contrast, the Lorentz group is non-compact: keep boosting and you get further and
further from where you started. How does this manifest itself in the two su(2) algebras
that we’ve found in (2.4)?

– 11 –
 The answer is a little subtle and is to be found in the reality properties of the
generators Ai and Bi . Recall that all integer, j ∈ Z, representations of SU (2) are real,
while all half-integer spin, j ∈ Z + 12 , are pseudoreal (which means that, while not
actually real, the representation is isomorphic to its complex conjugate). However, the
Ai and Bi in (2.4) do not have these properties. You can see in (2.2) that both Ji and
Ki are pure imaginary. This, in turn, means that the generators Ai and Bi are complex
conjugates of each other

(Ai )⋆ = −Bi

This is where the difference lies that distinguishes SO(4) from SO(1, 3). The Lie algebra
so(1, 3) does not contain two, mutually commuting copies of the real Lie algebra su(2),
but only after a suitable complexification. This means that certain complex linear
combinations of the Lie algebra su(2) × su(2) are isomorphic to so(1, 3). To highight
this, the relationship between the two is sometimes written as

so(1, 3) ∼
= su(2) × su(2)⋆
For our purposes, it means that the complex conjugate of a representation (j1 , j2 )
exchanges the two quantum numbers

(j1 , j2 )⋆ = (j2 , j1 )

Both the scalar representation (0, 0) and the vector representation ( 21 , 12 ) are real, while
the left- and right-handed Weyl spinors ( 12 , 0) and (0, 12 ) are exchanged under complex
conjugation. This last statement will be important as we proceed. In the context
of quantum field theory, if a field appears in a theory then so too does its complex
conjugate. This means that if you have a left-handed spinor, you also have a right-
handed complex conjugated spinor.

2.1.1 Spinors and SL(2, C)

There is another way to discover spinors, this time one that doesn’t involve going
through the algebra. We will use the fact that there is an isomorphism between two
groups

Spin(1, 3) ∼
= SL(2, C) (2.6)

To see this, we first note that we can write a point xµ in Minkowski space as a 2 × 2
Hermitian matrix,
!
x 0 + x 3 x 1 − ix 2
X = xµ σ µ =
x1 + ix2 x0 − x3

– 12 –
where we’ve introduced the 4-vector of 2 × 2 matrices,
! ! !
0 1 0 −i 1 0
σ µ = (1, σ i ) with σ 1 = , σ2 = , σ3 = (2.7)
1 0 i 0 0 −1

The σ i are, of course, the Pauli matrices. The matrix X is Hermitian: X = X † .

Moreover, there is clearly a one-to-one map between 4-vectors xµ and 2 × 2 Hermitian
matrices. The Minkowski inner product is particularly natural in this language: it is

det X = (x0 )2 − (x1 )2 − (x2 )2 − (x3 )2 = xµ xµ

Now consider an SL(2, C) transformation that acts as

X → X ′ = SXS † (2.8)

with S ∈ SL(2, C). We have (X ′ )† = X ′ and det X ′ = det X since det S = 1. This
means that the map (2.8) must be a Lorentz transformation.

In fact, it is not hard to see that we can implement all Lorentz transformations this
way and we’ll give an explicit construction of the generators shortly. For now, we can
just do some simple counting. A general complex 2 × 2 matrix has 4 complex entries.
The requirement that its determinant is 1 reduces this to 3 complex parameters, or 6
real parameters. This agrees with the dimension of the Lorentz group: 6 = 3 rotations
+ 3 boosts. Moreover, the SL(2, C) transformation S = −1 does not act on X, which
is the reason why SL(2, C) coincides with the double cover (2.6).

It is clear that the fundamental representation of SL(2, C) is not a 2 × 2 matrix: it

is a 2-component, complex object ψα = (ψ1 , ψ2 ) that transforms as

ψα → Sαβ ψβ α, β = 1, 2

Clearly it is a complex two-dimensional representation. In terms of our previous clas-

sification (2.5), we take it to correspond to ( 21 , 0): it is what we call a left-handed Weyl
spinor.

Given any complex representation of a Lie group, we can always form another rep-
resentation by taking the conjugate. This is equivalent to the original if we can find a
matrix C for which S ⋆ = CSC −1 . In the present case, no such C exists and the ma-
trix S and its conjugate S ⋆ are inequivalent representations. We denote the complex
conjugate as

(ψα )† = ψ̄α̇

– 13 –
We’ve adopted two notational flourishes to distinguish the two representations. First,
we use different indices α, β = 1, 2 and α̇, β̇ = 1, 2 for the two different representations.
This is useful because the two indices are telling us that the objects transform in
different ways. In addition, we also add a bar over any object, like ψ̄, that transforms
in the conjugate representation. This allows us to identify these objects even when
we suppress the indices. (Note that a bar on a Weyl spinor simply means complex
conjugation while, as we learned in the Quantum Field Theory lectures, a bar on
a Dirac spinor means complex transpose together with multiplication by γ 0 .) The
complex conjugate spinor then transforms as

ψ̄α̇ → (S ⋆ )α̇β̇ ψ̄β̇ α̇, β̇ = 1, 2

In our previous classification (2.5) it is the representation (0, 21 ). It is a right-handed

Weyl spinor.

Some of the index conventions above (and below) differ from what you may have
seen in other contexts and it’s worth quickly explaining why. Suppose that we’ve got
a vector u that transforms in the fundamental of SU (N ). We write the components as
ua with a = 1, . . . , N . The vector u† transforms in the conjugate representation and we
would write these components as (u† )a , with the index raised and no dots in sight. This
reflects the fact that we can contract u† and u to form a singlet: (u† )a ua . However,
the representations of SL(2, C) have a different structure and, as we’ll see shortly, you
can’t contract a spinor and its conjugate to get a singlet. That’s why we introduce
the strange looking dotted indices, rather than raising the index, to distinguish the
conjugate representation

Building Scalars from Spinors

The group SL(2, C) has the following invariant tensors
! !
0 1 0 −1
ϵαβ = ϵα̇β̇ = and ϵαβ = ϵα̇β̇ =
−1 0 1 0

Note that the ϵαβ with indices lowered differs by a minus sign from ϵαβ . This ensures
that one is the inverse of the other: ϵαβ ϵβγ = δγα . This, in turn, means that when we
use epsilon symbols to raise and lower indices (as we will below) then if we choose to
raise an index and subsequently lower it again then we don’t get a minus sign for our
troubles.

– 14 –
 Given, say, two left-handed Weyl fermions ψ and χ, we can use the epsilon tensors
to form invariants. We define
ψχ := ϵαβ ψβ χα = ψ2 χ1 − ψ1 χ2
To see that these are, indeed, invariants under SL(2, C), we just need to perform a
transformation
ψχ → Sαγ Sβδ ϵαβ ψδ χγ = (det S)ϵγδ ψδ χγ = ψχ (2.9)
where, in the first equality we’ve used the fact that Sαγ Sβδ ϵαβ = det S ϵγδ , which you
can confirm simply by checking all the cases γ, δ = 1, 2. In the second equality we’ve
used the fact that det S = 1.

In some ways, the ϵ symbols play a role for spinors that is akin to role played by the
metric η µν for vectors. Of course, one key difference is that ϵαβ is anti-symmetric, but
this tallies nicely with the fact that, in quantum field theory, spinors are anti-commuting
Grassmann variables. We then have
ψχ = ψ2 χ1 − ψ1 χ2 = −χ1 ψ2 + χ2 ψ1 = χψ
In particular, ψψ = 2ψ2 ψ1 is non-vanishing.

We can do something similar for right-handed fermions. However, a fiddly minus

sign rears its head. We define
ψ̄ χ̄ := ϵα̇β̇ ψ̄α̇ χ̄β̇ = ψ̄1 χ̄2 − ψ̄2 χ̄1 (2.10)
With anti-commuting spinors, we again have ψ̄ χ̄ = χ̄ψ̄. Note that the ordering of
the indices in (2.10) differs from (2.9). The reason for choosing this different ordering,
resulting in a minus sign difference in the definitions, is that it ensures that (ψχ)† = ψ̄ χ̄,
since
(ψχ)† = (ψ2 χ1 − ψ1 χ2 )† = χ̄1 ψ̄2 − χ̄2 ψ̄1 = ψ̄ χ̄
We can use the ϵ symbols to raise and lower spinor indices, just as we use the Minkowski
metric to raise and lower vector indices. We have
ψ α = ϵαβ ψβ , ψα = ϵαβ ψ β and ψ̄ α̇ = ϵα̇β̇ ψ̄β̇ , ψ̄α̇ = ϵα̇β̇ ψ̄ β̇
In this notation, the Lorentz scalars (2.10) become
ψχ = ψ α χα and ψ̄ χ̄ = ψ̄α̇ χ̄α̇
Our fiddly minus sign difference between (2.9) and (2.10) has now transmuted into
the following rule: for left-handed spinors we should contract (undotted) indices in the
direction ↘, while for right-handed spinors we should contract (dotted) indices in the
direction ↗.

– 15 –
 We can ask how these new objects ψ α and ψ̄ α̇ fare under Lorentz transformations.
We have

ψ α → ϵαβ Sβγ ψγ = (S −1 T )αβ ψ β

ψ̄ α̇ → ϵα̇β̇ (S ⋆ )β̇γ̇ ψ̄γ̇ = (S −1 † )α̇β̇ ψ̄ β̇ (2.11)

where the equality follows from the following algebra

Sαγ ϵαβ Sβδ = ϵγδ ⇒ (S T )γα ϵαβ Sβδ = ϵγδ ⇒ ϵαβ Sβδ = (S −1 T )αγ ϵγδ

with similar manipulations for the right-handed spinor. The matrices S −1 T don’t form
a new representation of SL(2, C); they are equivalent to the fundamental representa-
tion since, from above, we have ϵSϵ−1 = S −1 T . This means that the covariant and
contravariant left-handed spinors ψα and ψ α transform in equivalent representations.
Similarly, the right- handed spinors ψ̄α̇ and ψ̄ α̇ transform in equivalent representations.

Building Vectors from Spinors

A key take-away from our discussion above is that if you want to form a Lorentz scalar
then you need to a pair of left-handed fermions or a pair of right handed fermions.
Suppose that we instead have one object of each type, say a left-handed spinor ψα and
a right- handed spinor χ̄α̇ . What kind of object can we then build? The answer is clear
from the quantum numbers of these representations:

( 12 , 0) ⊗ (0, 21 ) = ( 12 , 12 )

This is the vector representation of the Poincaré group.

To explicitly construct the vector, we sandwich the Pauli matrices

(σ µ )αα̇ = (1, σ i )αα̇

between two spinors. We write

ψσ µ χ̄ = ψ α (σ µ )αα̇ χ̄α̇

Note that, as shown above, the Pauli matrices σ µ should come with an index of each
type – one undotted, and one dotted – and both subscripts. Taking the conjugate, we
have (ψσ µ χ̄)† = χσ µ ψ̄.

– 16 –
 To see that the object does indeed transform as a 4-vector, we can contract this with
any other 4- vector xµ to give ψX χ̄ with X = xµ σ µ . But we know from (2.8) and (2.11)
how each of these transforms: we then have

ψX χ̄ = ψ α Xαα̇ χ̄α̇ → (ψ β (S −1 )βα )(Sαδ Xδδ̇ S ⋆α̇ δ̇ )(χ̄β̇ (S ⋆ −1 )β̇α̇ ) = ψX χ̄

The fact that ψX χ̄ forms a singlet shows that ψσ µ χ̄ must transform as a vector. In
fancy maths words, we say that the Pauli matrices act as the intertwiner between the
different representations.

We can use the epsilon symbols to raise the spinor indices on the Pauli matrices σαµα̇ .
This gives us a closely related set of matrices that we denote

(σ̄ µ )α̇α = ϵαβ ϵα̇β̇ σβµβ̇

The bar on σ̄ doesn’t denote anything to do with complex conjugation. The σ̄ µ are
simply a different set of 2×2 matrices from σ µ . Note that the indices have not only been
raised, but also switched: σ µ has the undotted index first, while σ̄ µ has the dotted index
first. If we define ϵ = iσ 2 then, viewed as matrix multiplication, we have σ̄ = ϵσ T ϵT .
A quick calculation shows that

(σ̄ µ )α̇α = (1, −σ i )α̇α

We can then similarly construct the vector

χ̄σ̄ µ ψ = χ̄α̇ (σ̄ µ )α̇α ψα

This isn’t a new object: you can check that ψσ µ χ̄ = −χ̄σ̄ µ ψ.

Generators of SL(2, C)
Finally we can give a description of the generators of SL(2, C). We define the anti-
symmetrised product of sigma matrices,
i µ ν
(σ µν )αβ = (σ σ̄ − σ ν σ̄ µ )αβ
4
These are linearly independent and so can be taken as a generators of SL(2, C). Because
of the anti-symmetry in µ and ν, there are six such generators which is the dimension
of the Lorentz group. Indeed, we can see explicitly that these generate the Lorentz
group by computing the commutator

[σ µν , σ ρσ ] = i (η νρ σ µσ − η νσ σ µρ + η µσ σ νρ − η µρ σ νσ )

– 17 –
This reproduces the algebra of the Lorentz group (2.3) as promised. A left-handed
spinor then transforms as
 β
i µν
ψα → exp − ωµν σ ψβ (2.12)
2 α
where ωµν are the same set of six numbers that specify the Lorentz transformation
(2.1).

The conjugate representation is generated by by

i
(σ̄ µν )α̇β̇ = (σ̄ µ σ ν − σ̄ ν σ µ )α̇β̇
4
These too satisfy the algebra of the Lorentz group. Correspondingly, a right-handed
spinor transforms as
 α̇
α̇ i µν
ψ̄ → exp − ωµν σ̄ ψ̄ β̇ (2.13)
2 β̇

Note that, from the positioning of the indices of σ̄ µν , these act naturally as generators
on ψ̄ α̇ , with the index raised.

2.1.2 Lagrangians for Spinors

We can now describe how to construct Lagrangians from a Weyl spinor. Suppose that
we have just a single left-handed Weyl spinor ψ to play with. This necessarily comes
with its conjugate, a right-handed spinor ψ̄ = ψ † . We can then form a kinetic term
Z
SWeyl = − d4 x iψ̄σ̄ µ ∂µ ψ (2.14)

Upon quantisation, this theory gives a single massless, left-handed fermion of helicity
− 21 and massless right-handed anti-particle of helicity of + 12 . The theory has a global
U (1) symmetry under which ψ → eiα ψ; if the left-handed fermion has charge +1 then
the right-handed fermion has charge −1, as befits an anti-particle.

We can add a mass term for a single Weyl fermion. This is known as a Majorana
mass,
m⋆
Z
m
SMaj = d4 x ψψ + ψ̄ ψ̄ (2.15)
2 2
In general, we can take m ∈ C although any complex phase of m can be absorbed into ψ
and, upon quantisation, the resulting particle has mass |m|. Importantly, the Majorana
mass explicitly breaks the global U (1) symmetry, so there is no quantum number to
distinguish particle from anti-particle. Upon quantisation, the theory consists of a
single massive spin 21 particle that is now its own anti-particle.

– 18 –
 Because the Majorana mass term explicitly breaks the U (1) symmetry, it is not
allowed if the U (1) is gauged. Relatedly, it’s not possible to write down such a term
for any fermion ψ that transforms in a complex representation of a gauge group. It is,
however, possible to write down such terms for fermions in real representations.

Recovering Dirac Spinors

All this discussion of spinors and, so far, not a gamma matrix or Clifford algebra in
sight! Yet these played a central role in the discussion of spinors that we met in the
Quantum Field Theory lectures. What’s going on?

The Dirac spinor is not an irreducible representation of the Lorentz group in d = 3+1
dimensions. Instead, it consists of independent left- and right-handed spinors. In our
earlier notation:

( 21 , 0) ⊕ (0, 12 ) : Dirac spinor

We write a Dirac spinor as a 4-component object, consisting of a left-handed Weyl

fermion ψα and a right-handed Weyl fermion χ̄α̇ (note the index up),
!
ψα
Ψ=
χ̄α̇

We also introduce the chiral basis of gamma matrices

!
µ 0 σµ
γ = (2.16)
σ̄ µ 0

These obey the Clifford algebra {γ µ , γ ν } = 2η µν . In the Quantum Field Theory lectures,
we showed that the generators of Lorentz transformations for a Dirac spinor are
!
µν
i σ 0
S µν = [γ µ , γ ν ] =
4 0 σ̄ µν

(As with our earlier definition of M µν , this differs by a factor of i from the conventions in
the Quantum Field Theory lectures.) Under a Lorentz transformation, a Dirac spinor
transforms as Ψ → exp(− 2i ωµν S µν )Ψ. This reproduces the transformations of Weyl
spinors that we saw in (2.12) and (2.13).

– 19 –
 The Dirac action that we met in our Quantum Field Theory lectures is
Z
SDirac = − d4 x iΨ̄γ µ ∂µ Ψ − M Ψ̄Ψ

where, for a Dirac spinor (but not a Weyl spinor!) the bar notation means Ψ̄ = Ψ† γ 0 .
Decomposed in terms of Weyl fermions, it becomes
Z
SDirac = − d4 x iψ̄σ̄ µ ∂µ ψ + iχσ µ ∂µ χ̄ − M (χψ + ψ̄ χ̄) (2.17)

The first term coincides with the kinetic term (2.14) for a left-handed fermion. The
second term is simply a different way of writing this, with the derivative now acting
on a right- handed fermion; if you play around lowering and raising indices then the
second term can be massaged to look like the first.

The mass term in (2.17) is not of the Majorana type (2.15). First, the mass is
necessarily real, M ∈ R, although it can be positive or negative. Second, because the
mass term involves two distinct Weyl fermions it preserves a U (1) symmetry, under
which the phase of ψ and χ rotate oppositely. The result is that, upon quantisation,
the action (2.17) gives a particle of spin + 12 and charge +1, together with a distinct
anti-particle of spin + 12 and charge −1, both with mass |M |.

It is possible to restrict the Dirac fermion Ψ to have the same content as a single
Weyl fermion. In a general basis of gamma matrices, we do this by introducing a
charge conjugation matrix. But in the chiral basis (2.16), it’s particularly simple: we
just restrict χ̄ = ψ̄ ≡ ψ † . A Dirac spinor with such a restriction is called a Majorana
spinor.

Throughout these lectures, we will have no need to resort to 4-component spinors.

We will write everything in terms of 2-component Weyl fermions.

2.1.3 The Poincaré Group and its Extensions

The continuous symmetries of Minkowski space comprise of Lorentz transformations
together with spacetime translations. Combined, these form the Poincaré group. Space-
time translations are generated, as usual, by the momentum 4-vector P µ . Their com-
mutation relations with themselves and with the Lorentz generators M µν are given
by

[P µ , P ν ] = 0 and [M µν , P σ ] = i (P µ η νσ − P ν η µσ ) (2.18)

– 20 –
The latter of these is equivalent to the statement that P µ transforms as a 4-vector
under Lorentz transformations. These commutation relations should be considered in
conjunction with the Lorentz algebra (2.3),

[M µν , M ρσ ] = i (η νρ M µσ − η νσ M µρ + η µσ M νρ − η µρ M νσ ) (2.19)

Together, (2.18) and (2.19) form the algebra of the Poincaré group.

It’s not unusual for quantum field theories to exhibit further continuous symmetries.
Say, a global U (1) symmetry that rotates the phase of a complex field, or perhaps
a non-Abelian SU (N ) symmetry under which a multiplet of fields transforms. The
generators of these symmetries – which we’ll denote collectively as T – correspond to
some conserved charge or isospin and are always Lorentz scalars. This means that they
necessarily commute with the Poincaré generators,

[P µ , T ] = [M µν , T ] = 0

One could ask: is it possible for something less trivial to happen, with the new genera-
tors transforming in some interesting fashion under the Poincaré group? For example,
this would happen if the additional generators T themselves carried some spacetime
index. If this were possilble, the Poincaré group would be subsumed into a larger group.
And that sounds interesting.

A theorem due to Coleman and Mandula greatly restricts this possibility. Roughly
speaking, the theorem states that, in any spacetime dimension greater than d = 1 + 1,
the symmetry group of any interacting quantum field theory must factorise as

Poincaré × Internal (2.20)

We won’t prove the Coleman-Mandula theorem here1 . The gist of the proof is that
Poincaré invariance already greatly restricts what can happen in, say, 2 to 2 scatter-
ing, with only the scattering angle left undetermined. Any internal symmetries that
factorise, as in (2.20), put restrictions on the kinds of interactions that are allowed,
for example enforcing conservation of electric charge. But if the generators T were
to carry a spacetime index then they would put further constraints on the scattering
angle itself and that would be overly restrictive, at best allowing scattering to occur
only at discrete angles. But if one assumes that the scattering amplitudes are analytic
functions of the angle then the amplitude must vanish for all angles and the theory is
free.
1
The original Coleman-Mandula paper is from 1967 and entitled “All Possible Symmetries of the
S-matrix”. Witten’s “Introduction to Supersymmetry” lectures give a clear intuitive explanation of
the theorem. A full proof can be found Weinberg vol III.

– 21 –
 Like all no-go theorems in physics, the Coleman-Mandula theorem comes with a
number of underlying assumptions. Some of these are eminently reasonable, such as
locality and causality. But it may be possible to relax other assumptions to find inter-
esting loopholes to the Coleman-Mandula theorem. Two such loopholes have proven
to be extremely important.

• Conformal Invariance: The Coleman-Mandula theorem assumes that the the-

ory has a mass gap, meaning that all particles are massive. Indeed, it studies
symmetries of the S-matrix which is really only well defined for massive particles
where we don’t have to worry about IR divergences. For theories of massless
particles something interesting can, and often does, happen.

The first interesting thing is that interacting massless theories typically exhibit
scale invariance. This means that physics is unchanged under the symmetry
xµ → λxµ . The associated symmetry generator is called D for “dilatation”. This
can only be a symmetry of a theory that has no dimensionful parameters. In
particular, no masses.

The second interesting thing is more surprising. For reasons that are not en-
tirely understood, theories that exhibit scale invariance also exhibit a further
symmetry known as special conformal transformations of the form
x µ − aµ x 2
xµ →
1 − 2a · x + a2 x2
This transformation depends on a vector parameter aµ and the associated gen-
erator is a 4-vector K µ . The resulting conformal algebra extends the Poincaré
algebra (2.18) and (2.19) with the non-trivial commutators

[D, K µ ] = −iKµ , [D, P µ ] = iP µ

[K µ , P ν ] = 2i(Dη µν − M µν )
[M µν , K σ ] = i (K ν η µσ − K µ η νσ )

Interacting conformal field theories crop up in many places in physics. In their

Euclidean incarnation, they describe critical points, or second order phase transi-
tions, that were the focus of our lectures on Statistical Field Theory. In d = 1 + 1
dimensions the conformal group has rather more structure and a detailed intro-
duction can be found in the lectures on String Theory. We’ll meet examples of
supersymmetric conformal field theories later in Section 6.4 when we discuss the
low-energy physics of certain gauge theories.

– 22 –
 • Supersymmetry: The second loophole to the Coleman-Mandula theorem is
supersymmetry. As you may by now have guessed, exploiting this loophole will
be the topic of the rest of these lectures.

2.2 The Supersymmetry Algebra

Supersymmetry evades the Coleman-Mandula no-go theorem because it is a different
kind of symmetry. In contrast to the symmetries discussed above, it is not characterised
by a Lie algebra. Instead it is characterised by a mathematical structure known as a
Z2 - graded Lie algebra. For our purposes, this simply means that the algebra contains
both commutation and anti-commutation relations.

A generalisation of the Coleman-Mandula theorem to graded Lie algebras was given

by Haag, Lopuszanski and Sohnius. Roughly speaking, it says that the only possibility
is supersymmetry. We will now, finally, explain what this means.

Supersymmetric theories have a new conserved charge that is a left-handed Weyl

spinor Qα , together with its right-handed counterpart Q̄α̇ . This is known as the su-
percharge. It is possible to have multiple supercharges, a situation known as extended
supersymmetry. We will discuss this in Section 2.4 and, for now, stick to just a single
complex supercharge. This is known as N = 1 supersymmetry.

At the heart of the supersymmetry algebra is the anti-commutation relation

{Qα , Q̄α̇ } = 2σαµα̇ Pµ (2.21)

It is no surprise that a spinor should have an anti-commutator. But the structure of

this relation is interesting: it tells us that the supercharges should be viewed as the
square-root of spacetime translations! Our goal in these lectures is to understand what,
exactly, this means.

The full supersymmetry algebra comprises of commutation relations (2.18) and (2.19)
of the Poincaré group, which remain unchanged, together with the (anti)-commutation
relations of the supercharges. The first of these is

[M µν , Qα ] = (σ µν )αβ Qβ and [M µν , Q̄α̇ ] = (σ̄ µν )α̇β̇ Q̄β̇ (2.22)

This is simply the statement that the supercharges transform under a Lorentz trans-
formation in the manner expected of operators that are Weyl fermions. To see this,
first recall from (2.12) that any spinor like Qα transforms as Qα → Uαβ Qβ where
U = exp(− 2i ωµν σ µν ). But Qα is also an operator acting on a Hilbert space and,
viewed through this lens, we get a different expression for how it transforms. Any

– 23 –
state in the Hilbert space transforms as |ϕ⟩ → V |ϕ⟩ with V = exp(− 2i ωµν M µν ). Here,
M µν is the abstract generator of Lorentz transformations and its action on any state
depends on the quantum number of that state. Correspondingly, operators O trans-
form as O → V OV † since this ensures that the matrix elements ⟨ϕ′ |O|ϕ⟩ remains
unchanged. Equating these two ways in which the supercharge transforms, we have
V Qα V † = (U Q)α . The algebra (2.22) is the infinitesimal version of this transformation
law.

The remaining commutation relations are somewhat less interesting, although no less
important

[Qα , P µ ] = {Qα , Qβ } = 0 (2.23)

There are, however, reasons why these commutators take this boring form.

First, why do we necessarily have [Qα , P µ ] = 0? Clearly the right-hand side should
be something with α and µ indices so that the commutator is covariant under Lorentz
transformations. But that leaves the option for [Qα , P µ ] = c(σ µ )αα̇ Q̄α̇ for some c ∈ C.
What forces us to have c = 0?

The answer to this lies in the Jacobi identity

[P µ , [P ν , Qα ]] + [P ν , [Qα , P µ ]] + [Qα , [P µ , P ν ]] = 0

Clearly the last term vanishes, as [P µ , P ν ] = 0. If we choose [Qα , P µ ] = c(σ µ )αα̇ Q̄α̇
and, correspondingly, [Q̄α̇ , P µ ] = c⋆ (σ̄ µ )α̇β Qβ then the Jacobi identity becomes

−cσαν α̇ [P µ , Q̄α̇ ] + cσαµα̇ [P ν , Q̄α̇ ] = |c|2 (σ ν σ̄ µ − σ µ σ̄ ν )αβ Qβ = 0

This requires c = 0.

There is a similar reason for why we must have {Qα , Qβ } = 0. Once again, there is an
alternative since if we just try to pair up indices then we might think that {Qα , Qβ } =
c′ (σ µν )αβ Mµν would be acceptable for any c′ ∈ R. But if we take the commutator
with P ρ then, from the argument above, the left-hand-side must vanish which, because
[P ρ , M µν ] ̸= 0, tells us that c′ = 0.

(An aside: there’s actually a subtlety in this last discussion. While it is true that
{Qα , Qβ } = 0 when sandwiched between any finite energy states, some supersymmet-
ric theories have multiple ground states and it turns out that {Qα , Qβ } can be non-
vanishing when evaluated on the infinite energy domain walls that interpolate between
these ground states. This subtlety is interesting, at least if you care about domain
walls, but somewhat beyond the scope of these lectures.)

– 24 –
2.2.1 R-Symmetry
We started this section by noting that all internal symmetries must commute with the
spacetime symmetries of the Poincaré group. But must they also commute with the
supercharge Qα ? The answer is: almost.

All internal symmetries must commute with Qα with one exception: it may be that
theories admit an internal U (1) symmetry that acts as

Qα → e−iλ Qα and Q̄α̇ → eiλ Q̄α̇ (2.24)

This U (1) symmetry is known as an R-symmetry and is sometimes denoted U (1)R . If

we denote the generator as R then it has commutation relations

[R, Qα ] = −Qα and [R, Q̄α̇ ] = +Q̄α̇ (2.25)

When we turn to theories of extended supersymmetry in Section 2.4, we’ll see different
R-symmetry groups arising. But for theories with N = 1 symmetry we have only
U (1)R . Nonetheless, this will play an important role when we come to analyse the
dynamics of supersymmetric theories in later sections. We’ll see this, for example, in
Section 3.3.

This, then, is the supersymmetry algebra: it comprises of the algebra of the Poincaré
group (2.18) and (2.19), together with the algebra of the supercharges (2.21), (2.22)
and (2.23) and, finally, the R-symmetry (2.25). The next question is: what can we do
with it?

2.2.2 A Consequence: Energy is Positive

Even before we write down any field theories, we can derive one feature of supersym-
metric theories from the algebra alone. This follows from the key algebraic relation
(2.21),

{Qα , Q̄α̇ } = 2σαµα̇ Pµ (2.26)

If we compute the expectation of the left-hand side in any state |ϕ⟩ then we find that
it is necessarily positive

⟨ϕ|Qα Q̄α̇ + Q̄α̇ Qα |ϕ⟩ = |(Qα )† |ϕ⟩|2 + |Qα |ϕ⟩|2 ≥ 0 (2.27)

The same must be true of the right-hand side

σαµα̇ ⟨ϕ|Pµ |ϕ⟩ ≥ 0

– 25 –
If we set α = α̇ and sum over α = 1, 2 then we make use of the fact that tr σ 0 = 2
and tr σ i = 0. This then reduces to the statement that the energy of any state in a
supersymmetric theory is necessarily positive

⟨ϕ|P0 |ϕ⟩ ≥ 0

This is curious. Usually in physics, we don’t care about the overall value of the energy:
if you add an overall constant to all energies, then physics remains unchanged. There
are two places where this state of affairs no longer holds. The first is in gravity where
the energy of the vacuum contributes as a cosmological constant. The second is, as
we’ve seen above, in supersymmetric theories where energies are necessarily positive
definite.

Physically, it’s far from clear if there is any deep relation between these two ideas.
In fact, as we will see later in these lectures, the energy of the ground state acts as an
order parameter for the breaking of supersymmetry. This means that the ground state
energy is zero if supersymmetry is exact, otherwise it is non-zero. In our world, it’s
clear that there is no supersymmetry visible at the TeV scale, while the cosmological
constant is many of orders of magnitude smaller, at 10−3 eV. This makes it difficult to
see how supersymmetry can help alleviate the cosmological constant problem.

However, at the formal mathematical level, the relationship between supersymmetry

and gravity has proven rather useful. For example, there exists a greatly simplified
proof of the positive energy theorem in general relativity, due to Witten, that uses
ideas of supersymmetry.

There is one further piece of physics hiding in (2.26). For any other symmetry in field
theory, we can think about gauging it. This means that we try to construct theories
in which the symmetry is realised locally. Supersymmetry is no different. One can
construct theories in which the associated infinitesimal parameter for supersymmetry
transformations depends on xµ . From (2.26), we see that such theories necessarily enjoy
a symmetry in which you do different translations at different points in space. But
such transformations are diffeomorphisms and are the characteristic feature of general
relativity. In other words, theories of local supersymmetry are necessarily theories of
gravity! Such theories are known as supergravity, usually shortened to the ugly acronym
“sugra”. We will mention supergravity only very briefly in this section. In subsequent
sections our interest will be entirely on theories with global supersymmetry.

– 26 –
2.3 Representations on Particle States
Given an algebra, our next task is to explore its representations. There are different
ways that we could approach this. Ultimately, we will be interested in quantum field
theories that enjoy supersymmetry and this means understanding the way supersym-
metry acts on fields. This we will do in later sections. Here, to build some intuition, we
will understand how supersymmetry acts on single particle states in the Hilbert space.

Without doing any work, we can guess that something interesting is going on. The
supercharge Qα is a fermionic operator, both in the sense that it carries spin 21 and in the
sense that it is naturally anti-commuting as in (2.21). This means that, schematically,
we must have

Q|fermion⟩ = |boson⟩ and Q|boson⟩ = |fermion⟩ (2.28)

This is the defining feature of supersymmetry.

In fact, it is straightforward to show that any representation of the supersymmetry

algebra must have an equal number of bosonic and fermionic states. To this end, we
introduce the fermionic number operator (−1)F . This acts on bosonic states as

(−1)F |B⟩ = |B⟩ and (−1)F |F ⟩ = −|F ⟩

Because Qα swaps a bosonic state for a fermionic state, we necessarily have

(−1)F Qα = −Qα (−1)F ⇒ {(−1)F , Qα } = 0

The result that we now want follows straightforwardly from the algebra {Qα , Q̄α̇ } =
2σαµα̇ Pµ . Suppose that we have a finite collection of one-particle states that form a
representation of the supersymmetry algebra. We can take the folllowing trace over
elements of this multiplet

tr (−1)F {Qα , Q̄α̇ } = tr (−1)F Qα Q̄α̇ + (−1)F Q̄α̇ Qα

   

= tr −Qα (−1)F Q̄α̇ + (−1)F Q̄α̇ Qα = 0

 

Here the second equality we’ve uses the fact that {(−1)F , Qα } = 0 while the final
equality uses the cyclicity of the trace. The supersymmetry algebra then tells us that

σαµα̇ tr (−1)F Pµ = 0
 

Note that σαµα̇ sits outside the trace over states: it’s just a bunch of numbers as far as
the trace is concerned. Meanwhile Pµ sits inside the trace because it is an operator

– 27 –
acting on states. We can choose these states to be momentum eigenstates, so that
Pµ |any state⟩ = pµ |any state⟩. We then simply have

σαµα̇ pµ tr (−1)F = 0

But tr(−1)F simply counts the number of bosonic states nB minus the number of
fermionic states nF ,

tr(−1)F = nB − nF = 0

The number of such states must be equal. The quantity tr(−1)F is called the Witten
index.

There’s actually a loophole in the discussion above. It may be that Qα and Q̄α̇
annihilate states in the supersymmetry multiplet. From the supersymmetry algebra
(and the positivity conditions (2.27) that follows from it) this can only happen for
states of zero energy which are necessarily the ground states of the system. This means
that there may be a mismatch between the number of bosonic and fermionic ground
states of a system. It is in studying such ground states that the Witten index really
finds it teeth and we’ll revisit this in Section 3.4.2. More sophisticated examples can
be found in the lectures on Supersymmetric Quantum Mechanics.

We now know that supersymmetry requires an equal number of bosonic and fermionic
states. The next step is to understand exactly what kind of fermion is paired with what
kind of boson.

2.3.1 Representations of the Poincaré Group

To set the scene, let’s first recall how we construct the irreducible representations of the
Poincaré group. In fact, let’s start even more simply: how do we construct irreducible
representations of the rotation group?

We work with the algebra so(3) ∼= su(2) rather than the group. This is, of course,
defined by the familiar commutation relations

[Ji , Jj ] = iϵijk Jk

To construct representations, the first thing we do is look to the Casimirs. These are
operators that commute with all generators of the group. For su(2), there is just a
single Casimir,
3
X
C= Ji2
i=1

– 28 –
Irreducible representation are labelled by their eigenvalue of the Casimir. For su(2),
the eigenvalue of J 2 is j(j + 1) with the spin j taking values in j = 0, 21 , 1, . . .. Each
representation has dimension 2j + 1, with the states within a multiplet identified by
their eigenvalue under, say, J3 whose eigenvalue lies in |j3 | ≤ j. The result is the
familiar one from quantum mechanics: states are labelled by |j, j3 ⟩

Now let’s turn to the Poincaré group. The irreducible representations are what we
call “particles”. Again, they are characterised by the Casimirs. I won’t tell you how
to construct Casimirs, but will instead just present you the result: the Poincaré group
has two Casimirs, given by

C 1 = Pµ P µ and C2 = Wµ W µ

Here W µ = 12 ϵµνρσ Pν Mρσ is the Pauli-Lubański vector. It can be thought of as a

relativistic version of angular momentum.

Representations of the Poincaré group are then labelled by the eigenvalues of C1 and
C2 . The first of these is simply the mass m of a particle: C1 = m2 . What happens
next is a little different depending on whether the particles are massive or massless.

• Massive Particles: In this case, we can always boost to the rest frame of the
particle so that P µ = (m, 0, 0, 0). In this frame, the Pauli-Lubański vector is

W 0 = 0 and W i = −mJ i

with J i the generators of rotations. This means that C2 = −m2 J 2 and so is

specified by the eigenvalue of J 2 . We find the familiar fact that massive particles
are characterised by their mass m and spin j.

• Massless Particles: Now C2 = W 2 = 0, so both Casimirs vanish. Nonetheless,

there is still a way to characterise the representation. To see this, we choose a
frame such that, say, P µ = (E, 0, 0, E). There, we have W µ = M12 P µ , so the
constant of proportionality between W and P is determined by the eigenvalue
of the U (1) rotation in the (x1 , x2 )-plane. The eigenvalue of this rotation is the
helicity, h = 0, 21 , 1, . . .. We learn that massless particles are characterised by
(obviously) m = 0 and their helicity h.

Although the results are different for m = 0 and m ̸= 0, the strategy is the same. In
each case, we boost to a preferred frame of the particle which is then characterised by
how it transforms under the surviving symmetry group. This surviving symmetry —
SU (2) for a massive particle, U (1) for a massless one — is called the little group.

– 29 –
 There is a slight twist to the story when it comes to realising these representations
on the Hilbert space of single particle states. For massive particles, the states take the
form

|pµ ; j, j3 ⟩ (2.29)

where the momentum is restricted to obey pµ pµ = m2 while the azimuthal angular

momentum takes values in j3 ≤ |j|. This fills out the 2j + 1 dimensional set of spin
sets. However, for massless particles, there is just a single state |pµ ; h⟩. This is because
the helicity describes the representation of the Abelian group U (1) generated by M 12
rather than the non-Abelian group SU (2) and irreducible representations of Abelian
groups are one-dimensional.

The problem is that we know that massless particles also have internal degrees of
freedom. For example, the photon necessarily has two polarisation states. Clearly
we’re missing something. What we’re missing is the additional requirement that the
spectrum of states is invariant under CPT. For massive particles, this doesn’t buy us
anything new: the set of states (2.29) is already invariant under CPT. However, for
massless particles CPT flips h 7→ −h and tells us that massless states must come in
pairs

|pµ ; h⟩ and |pµ , −h⟩

This is the origin of the two polarisation states of the photon or graviton, or the two
helicities of a massless Weyl spinor. Note that a massless scalar has helicity h = 0 and
so is CPT self-conjugate. This means that there’s no requirement from CPT to add an
additional degree of freedom in this case.

2.3.2 Massless Representations

We now turn to the representations of the N = 1 supersymmetry algebra. The simple
observation (2.28) tells us that we should expect representations to contain particles of
different spin and this will turn out to be true. Once again we need to treat massless
and massive particles separately.

The supersymmetry algebra also has two Casimirs. The first is familiar:

C1 = P µ P µ

The fact that this is a Casimir tells us that all particles in a supersymmetric multiplet
must have the same mass, C1 = m2 .

– 30 –
 In contrast, the other Casimir of the Poincaré group, Wµ W µ , is not a Casimir of the
supersymmetry algebra. This is because [Wµ , Qα ] ̸= 0 which, in turn, can be traced to
the commutation relation [Mµν , Qα ] ̸= 0. But it was Wµ W µ that told us that repre-
sentations of the Poincaré group are characterised by the spin of a particle. The fact
that Wµ W µ is no longer a Casimir means that representations of the supersymmetry
algebra can contain particles of different spin.

It is possible to construct a new Casimir. First define

1
Yµ = Wµ − Q̄α̇ σ̄µα̇β Qβ
4
Then the second Casimir of the supersymmetry algebra turns out to be
C̃2 = (Yµ Pν − Yν Pµ )(Y µ P ν − Y ν P µ )
However, in what follows we won’t need this result. Instead we will build up a repre-
sentation of the supersymmetry algebra more directly. Our strategy is to start from a
particle (i.e. a representation of the Poincaré group) and then act on it with successive
supersymmetry generators until we build up a representation of the full algebra.

It turns out that things are slightly simpler for massless representations. Consider
a state |pµ , h⟩ of a massless particle of helicity h. We can again boost to a frame in
which pµ = (E, 0, 0, E). Restricted to act on such states, the supersymmetry algebra
becomes
!
µ 1 0
{Qα , Q̄α̇ } = 2σαα̇ Pµ = 2E(1 + σ 3 )αα̇ = 4E
00
From the positivity condition (2.27), we see that Q2 and Q̄2 necessarily annihilate this
state,
⟨pµ , h|{Q2 , Q̄2 }|pµ , h⟩ = 0 ⇒ Q2 |pµ , h⟩ = Q̄2 |pµ , h⟩ = 0
To build a representation of the full supersymmetry algebra, we only need consider
the action of Q1 and Q̄1 . But these act just like fermionic creation and annihilation
operators. Specifically, if we rescale the operators to become
Q1 Q̄1
a= √ and a† = √ ⇒ {a, a† } = 1 and {a, a} = {a† , a† } = 0
4E 4E
The representations of this algebra are straightforward: they consist of two states |0⟩
and |1⟩ such that a|0⟩ = 0 and |1⟩ = a† |0⟩. This ensures that a† |1⟩ = 0. For us, this
means that we can start by taking a state which, by assumption, is annihilated by a,
a|pµ , h⟩ = 0

– 31 –
The full supersymmetry multiplet then consists of |pµ , h⟩ and a† |pµ , h⟩. The question
is: what is the helicity of this second state? This follows from the commutation relation
(2.22)
[M µν , Qα ] = (σ µν )αβ Qβ and [M µν , Q̄α̇ ] = (σ̄ µν )α̇β̇ Q̄β̇ (2.30)
Restricting to rotations in the (x1 , x2 ) plane, which is what we mean by helicity, we
have
1 1
[M 12 , Q1 ] = Q1 and [M 12 , Q2 ] = − Q2
2 2
12 1 1 1 1
[M , Q̄ ] = Q̄ and [M , Q̄ ] = − Q̄2
12 2
2 2
The first equation tells us that Q1 raises the helicity by 12 . This suggests that the
adjoint Q̄1 lowers the helicty by 12 . To see that this is the case, we need to remember
that, after lowering an index, Q̄1 = −Q̄2 so we have
1
[M 12 , Q̄1 ] = − Q̄1
2
So Q̄1 does indeed lower the helicity by 12 as anticipated. We learn that the massless
representations of the supersymmetry algebra consist of just two states:
1 Q̄1
|pµ , h⟩ and |pµ , h − ⟩ = √ |pµ , h⟩
2 4E
As we saw above, for massless states we must also add their CPT conjugates. The
different representations of the supersymmetry algebra then arise by picking different
starting helicities h. There are three representations that are most important:
• If we start with h = 1
2
then we have

h − 12 0 + 12
multiplicity 1 2 1

This is the matter content that we get from quantising a single Weyl spinor
together with a complex scalar. This is known as a chiral multiplet.
The chiral multiplets should be thought of as matter particles. We will devote
Section 3 to studying field theories associated to chiral multiplets. Here we make a
quick comment. The fact that any other internal symmetry generator must com-
mute with Qα means that the fermion and scalar in a given chiral multiplet must
experience the same force. In particular, if one is charged under a gauge group
then so is the other. We’ll see this explicitly when we construct supersymmetry
gauge theories in Section 4.

– 32 –
 • If we start with h = 1 then we have

h −1 − 12 + 12 +1
multiplicity 1 1 1 1

This is the matter content of a photon together with a single Weyl spinor. It is
known as the gauge multiplet or vector multiplet.
We will devote Section 4 to the study of vector multiplets. There we will see
that we can construct supersymmetric versions of Yang-Mills theory with gauge
group G by taking dim G vector multiplets. As usual, the h = 1 gauge bosons
transform in the adjoint of the gauge group. But now, so too, must its fermionic
supersymmetric partner. In this context, the fermion is called a gaugino.

• If we start with h = 2 then we have

h −2 − 23 + 32 +2
multiplicity 1 1 1 1

This is the matter content of a graviton together with a helicity 32 spinor, some-
times known as a Rarita-Schwinger field or, in this context, the gravitino. They
combine to form the supergravity multiplet.

If we keep going, we get massless fields with helicity h > 2. But there are strong
restrictions that prohibit the existence of interacting theories with massless fields of such
high helicity. (This statement is true in Minkowski spacetimes; there are remarkable
”higher spin” theories that include an infinite tower of massless states in de Sitter or
anti de Sitter spacetimes.) We also skipped the h = 23 multiplet for similar reasons; it
turns out that the existence of a massless helicity 32 particle implies the existence of a
local supersymmetry which, in turn, requires that the theory is coupled to gravity.

2.3.3 Massive Representations

We next turn to massive representations of the supersymmetry algebra. In the rest
frame of a particle we have pµ = (m, 0, 0, 0). Acting on such states, the supersymmetry
algebra becomes
!
1 0
{Qα , Q̄α̇ } = 2σαµα̇ Pµ = 2mσα0 α̇ = 2m (2.31)
01

– 33 –
This time, after rescaling, both Q1 and Q2 act as fermionic creation/annihilation op-
erators
Qα Q̄α̇
aα = √ and a†α̇ = √ ⇒ {aα , a†α̇ } = δαα̇
2m 2m

with {aα , aβ } = {a†α̇ , a†β̇ } = 0. We start with a state |Ω⟩ = |pµ ; j, j3 ⟩ that we assume to
be annihilated by aα |Ω⟩ = 0. Then the full supermultiplet consists of four states

|Ω⟩
a†1 |Ω⟩ and a†2 |Ω⟩
a†1 a†2 |Ω⟩

Again, the question is: what is the spin of these other states. We could use the
commutation relations (2.30) to understand how the new states transform under the
SU (2) little group but it’s a little fiddly while the end result is intuitive and straight-
forward. The initial state |Ω⟩ has spin j. The states a†α |Ω⟩ then sit in the tensor
product of representations j ⊗ 21 = (j + 12 ) ⊕ (j − 12 ). The final state can be written
as a†1 a†2 |Ω⟩ = 12 ϵαβ a†α a†β |Ω⟩, where the ϵαβ now contracts the creation operators to be a
spin singlet. This means that the state a†1 a†2 |Ω⟩ once again has spin j.

The upshot is that a massive supermultiplet contains two particles of spin j, a particle
of spin j − 12 and a particle of spin j + 21 . Note that the degeneracy of the two particles
of spin j is precisely equal to the degeneracies of the other two particles:
       
1 1
2 × (2j + 1) = 2 j + +1 + 2 j− +1
2 2

This is simply that statement that we saw previously: a supermultiplet must have an
equal number of bosonic and fermionic degrees of freedom.

There are just two massive supermultiplets that will be of interest

• If we start with j = 0, we have

1
j 0 2
multiplicity 2 1

This is the matter content of a massive complex scalar with a single massive Weyl
fermion. We recognise it as the same matter content as the chiral multiplet that
we met previously, now of course with all particles having a mass.

– 34 –
 • If we start with j = 12 , we have

1
j 0 2
1
multiplicity 1 2 1

In other words, we have a massive spin 1 particle, two massive Weyl fermions,
and a massive spin 0 particle. This is now more states than we found in the mass-
less gauge multiplet. In fact, this collection of states is equivalent to a massless
gauge multiplet and a massless chiral multiplet. But that makes sense. In quan-
tum field theory, a massless gauge boson can become massive only through the
Higgs mechanism, in which the gauge boson “eats” a scalar. The supersymmetric
extension of this is that a massless vector multiplet “eats” a chiral multiplet to
become the massive vector multiplet described above.

There’s one further subtlety that is worth flagging up. This is how parity acts on
the two scalars in the massive chiral multiplet. It turns out that one of them is a scalar
and the other a pseudoscalar. Here, the meaning of a “pseudoscalar” is that it picks up
a minus sign under parity. This statement follows, like everything else in this section,
from the supersymmetry algebra. We denote the parity operator as P̂ to distinguish it
from the momentum operator P µ . By definition, we must have

P̂P µ P̂ −1 = (P 0 , −P i )

Meanwhile, parity also exchanges left-handed and right-handed spinors. This means
that parity must exchange some combination of Qα and Q̄α̇ . One can check that the
supersymmetry algebra remains unchanged if we take

P̂Qα P̂ −1 = (σ 0 )αα̇ Q̄α̇ and P̂ Q̄α̇ P̂ −1 = −(σ 0 )α̇α Qα

(More generally one can include a complex phase in these relations but it will not affect
our discussion here.)

Now our two scalar states in the massive chiral multiplet are |Ω⟩ and |Ω′ ⟩ = a†1 a†2 |Ω⟩ ∼
Q̄1 Q̄2 |Ω⟩. They obey Qα |Ω⟩ = Q̄α̇ |Ω′ ⟩ = 0. Since parity exchanges Qα and Q̄α̇ , it must
also exchange |Ω⟩ and |Ω′ ⟩. This means that the parity eigenstates are

P̂ (|Ω⟩ ± |Ω′ ⟩) = ± (|Ω⟩ ± |Ω′ ⟩)

and we have one scalar (with the + sign) and one pseudoscalar (with the - sign) as
advertised.

– 35 –
2.4 Extended Supersymmetry
It is possible for theories to exhibit more than one supersymmetry. This means that
there is a collection of N supercharges

QIα and Q̄Iα̇ I = 1, . . . , N

Each of these supercharges retains the same commutation relations with the generators
of the Poincaré group,

[M µν QIα ] = (σ µν )αβ QIβ and [P µ , QIα ] = 0

and the key part of the supersymmetry algebra holds for each generator separately

{QIα , Q̄Jα̇ } = 2σαµα̇ Pµ δ IJ

However, there are two novelties. The first is that the anti-commutator of the super-
charges with themselves can be more interesting

{QIα , QJβ } = ϵαβ Z IJ and {Q̄Iα̇ , Q̄Jβ̇ } = ϵα̇β̇ (Z † )IJ (2.32)

Here Z IJ = −Z JI is a central charge, meaning that it commutes with all other elements
of the algebra. The exact nature of these central charges depends on the precise theory
that we consider, but they must be constructed from other conserved quantities that
are at hand. We’ll see the role that these central charges play shortly.

The second novelty is the R-symmetry group. Recall that for N = 1 we had a
U (1)R symmetry (2.24) that rotates the phase of the supercharge. For N > 1, the
R-symmetry rotates the supercharges among themselves. For reasons that will become
clear shortly, our primary interest will be in N = 2 and N = 4 supersymmetry. Here
the R-symmetries are:
• N = 2: The R-symmetry group is U (2)R ∼
= U (1)R × SU (2)R .

• N = 4: A priori, the R-symmetry group is U (4). However, it turns out that

only SU (4) is realised on fields. This is equivalent to SU (4) ∼
= Spin(6). (This is
sometimes written, a little inaccurately, as SO(6) but the supercharges transform
in the spinor representation of Spin(6) which is not a representation of SO(6) =
Spin(6)/Z2 .)
Theories with extended supersymmetry are a subset of those theories with N = 1
supersymmetry. This means that the representations of theories with N > 1 must be
constructed by joining together the N = 1 supermultiplets that we described above.
In the rest of this section, we explain how this works.

– 36 –
2.4.1 Massless Representations
For representations on states |pµ , h⟩ of massless particles, we proceed as before. We
boost to a frame with pµ = (E, 0, 0, E) and restrict attention to the algebra on such
states. We then have
!
1 0
{QIα , Q̄Jα̇ } = 4E δ IJ
00

As previously, we have QI2 |pµ , h⟩ = Q̄I2 |pµ , h⟩ = 0. From (2.32), we then have Z IJ |pµ , h⟩ =
0 which tells us that the central charges play no role for the massless states. We’re left,
as before, just with the QI1 and Q̄I1 operators to deal with. These now form a collection
of N fermionic creation and annihilation operators

QI Q̄I
aI = √ 1 and aI† = √ 1 ⇒ {aI , aJ† } = δ IJ and {aI , aJ } = {aI† , aJ† } = 0
4E 4E

We now start with some fiducial state |Ω⟩ = |pµ , h⟩ satisfying aI |Ω⟩ = 0 and build up
the full representation by acting with successive creation operators. The end result is
a collection of states

|Ω⟩
I†
a |Ω⟩
aI† aJ† |Ω⟩
...
a . . . aN † |Ω⟩
1†

Our initial state |Ω⟩ has helicity h. If we act with p of the a† excitation operators then
there are Np different states, each of which has helicity h − p/2. The full multiplet

consists of 2N different states. If we add the CPT conjugate states then we have 2N +1
states overall. Let’s now look at some specific examples.

N = 2 Supersymmetry
Again, the different multiplets arise by considering initial states |Ω⟩ with different
helicities. We’ll deal with each in turn.

• If we start with h = 21 then there are two states in the first level, aI† |Ω⟩, each
with h = 0, and a single state in the final level, a1† a2† |Ω⟩, with h = − 12 . After
adding the CPT conjugate we end up with

– 37 –
 h − 12 0 + 12
multiplicity 2 4 2

This is called a hypermultiplet. In consists of two chiral multiplets or, equivalently,

two complex scalars and a Dirac fermion (i.e. two Weyl fermions).
You might wonder why we needed to add the CPT conjugate in this case. After
all, starting with h = + 12 gave a single chiral multiplet which is already CPT self-
conjugate. The answer to this is buried in the details of the SU (2)R symmetry
which acts on the scalars aI† |Ω⟩ as a doublet. But this means that each of
these scalars must be complex and that, in turn, requires that we add the CPT
conjugate.

• If we start with h = 0 then we get two additional states with h = − 21 and one
with h = −1. Adding the CPT conjugate gives

h −1 − 12 0 + 12 +1
multiplicity 1 2 2 2 1

This is the N = 2 vector multiplet, comprising of an N = 1 vector multiplet and

N = 1 chiral multiplet.

• If we start with h = 2 then, after adding the CPT conjugate, we end up with

h −2 − 32 −1 +1 + 32 +2
multiplicity 1 2 1 1 2 1

This is the N = 2 supergravity multiplet. It comprises of an N = 1 supergravity

multiplet together with an N = 1 vector multiplet.

There’s one important feature of the spectrum above that is worth highlighting.
The fermions now come in pairs, meaning that they can be viewed as Dirac fermions
rather than Weyl fermions. This puts restrictions on the kind of supersymmetric theo-
ries that we can build. In particular, it’s not possible to construct a chiral gauge theory
with N > 1 supersymmetry. Here a chiral theory is one in which left- and right-handed
fermions experience different forces, like in the Standard Model. Such theories are pos-
sible with N = 1 supersymmetry (or, indeed, N = 0 supersymmetry as in our world!).
But any extended supersymmetry forces the theories to be vector-like.

– 38 –
N = 4 Supersymmetry
We can play the same game with N = 4 supersymmetry.

• If we start with h = 1 then we get the following multiplet

h −1 − 12 0 + 12 +1
multiplicity 1 4 6 4 1

This consists of an N = 2 vector multiplet with an N = 2 hypermultiplet and is

the unique N = 4 multiplet that does not include gravity. Note that there is now
no longer a distinction between forces and matter: once you specify the gauge
group, all matter content is also fixed. Furthermore, all matter fields necessarily
transform in the adjoint representation of the gauge group.
For once, we did not need to add the CPT conjugate to the above multiplet: it’s
already CPT self- conjugate. As we saw above, it was almost possible to achieve
this for the N = 2 matter representation but we fell at the last hurdle when we
considered how the SU (2)R symmetry acts on the scalars. But now we have no
such concern. The scalars are the set of 6 states aI† aJ† |Ω⟩ and transform in the 6
of the SU (4) R-symmetry. But this is a real representation and there is no need
to add the CPT conjugate.

• If we start with h = 2 then, after adding the CPT conjugate multiplet, we have

h −2 − 32 −1 − 12 0 + 12 +1 + 23 +2
multiplicity 1 2 2 2 2 2 2 2 1

This is the N = 4 supergravity multiplet, comprising of an N = 2 supergravity

multiplet and N = 2 vector multiplet.

You may have noticed that we jumped straight from N = 2 to N = 4, missing

out N = 3 in the middle. If you try to build a multiplet of single particle states
with N = 3 supersymmetry starting from, say, h = 12 or h = 1 then you’ll find that
you’re obliged to add the CPT conjugate representation and you just end up with
N = 4 supersymmetry after all. This observation is the key element of a proof that
says any perturbative theory with N = 3 global supersymmetry necessarily has N = 4
supersymmetry.

– 39 –
 The word “perturbative” is important in the above statement. This means that the
theory is weakly coupled and the single particle states that we’re considering here are
a good approximation to the spectrum of the theory. It turns out N = 3 supersym-
metry can be realised in strongly coupled, interacting quantum field theories, with no
perturbative regime.

N = 8 Supersymmetry
If we go beyond N = 4 supersymmetry then we no longer have multiplets with helicities
h ≤ 1. This means that we are now necessarily in the realm of local supersymmetry
and supergravity. Furthermore, by the time we get beyond N = 8 supersymmetry the
multiplets have particles with helicity h > 2. As we mentioned before, such theories
are always free in Minkowski space and therefore of limited interest. In this sense,
N = 8 is the maximum number of supersymmetries possible. The theory has a unique
supergravity multiplet with the following degeneracies

h −2 − 23 −1 − 12 0 + 21 +1 + 32 +2
multiplicity 1 8 28 56 70 56 28 8 1
N = 8 supergravity has some interesting properties and plays a role in string theory.
However, we won’t discuss it further in this course.

2.4.2 Massive Representations and BPS Bounds

Rather than repeating the whole story for massive representations, we will instead
just focus on the novelty. This arises from the central charges Z IJ that appear in the
supersymmetry algebra
{QIα , QJβ } = ϵαβ Z IJ
For reasons that we now explain, this is where much of the power of extended super-
symmetry comes from.

Our goal is to understand representations of this algebra, in conjunction with the

original supersymmetry algebra which, in the rest frame of the particle, reads (2.31)
!
1 0
{QIα , Q̄Jα̇ } = 2m δ IJ
01
We’ll illustrate the story with N = 2 supersymmetry, although the general idea holds
for any theory with extended supersymmetry. With N = 2, the anti-symmetric central
charge is necessarily just a complex number Z
Z IJ = 2ϵIJ Z

– 40 –
For simplicity, we take Z to be real. (Typically it’s not but we’ll dodge this issue
for now and state the full result below.) We then define the following combination of
creation and annihilation operators
! !
1 Q11 + Q̄22 1 Q11 − Q̄22
aα = √ and bα = √
2 Q12 − Q̄21 2 Q12 + Q̄21

Note that we’ve mixed up α and α̇ indices. This is acceptable because we’re working
in the rest frame of the particle and so have already broken Lorentz invariance. The
choice of a and b operators is designed to disentangle the mass and central charge Z,
so their commutation relations read

{aα , a†β } = 2(m + Z)δαβ and {bα , b†β } = 2(m − Z)δαβ

with all other anti-commutators vanishing. The {aα , a†β } and {bα , b†β } are both positive
definite, so the corresponding right-hand sides must be too. But this is only true if the
masses are bounded by the central charges,

m ≥ |Z|

This formula also holds if Z is complex; we just need to redefine the operators a
and b using a phase to derive the same result. This formula is interesting. Although
we haven’t seen yet any specific examples, recall that the central charge Z is some
combination of conserved charges in the quantum field theory. We learn that the masses
of particles is bounded by the charges. This is known as the BPS bound although in
the present context the name Witten-Olive bound would be more appropriate.

What about the representation theory of the algebra? Crucially, this depends on
whether m > |Z| or m = |Z|.

If m > |Z|, then we are in a situation very similar to the massive representation
theory that we saw before. Both a†α and b†α act as creation operators and the result is
that we have a multiplet comprising of 16 states. This is known as a long multiplet. We
can also repeat this story with N supersymmetries to find that long multiplets have
22N states.

More interesting is what happens when m = |Z|. In this case, half of the creation
operators do nothing. For example, when m = Z, the bα operators must just vanish
on all states in the multiplet. Now we’re back to the situation we met when discussing
massless representations, with only a†α acting as creation operators. The result is the
hypermultiplet or vector multiplet that we saw above, each with 8 states, but now with
a mass m = Z. This is known as a short multiplet.

– 41 –
 The existence of short multiplets, whose mass is fixed to be m = |Z|, turns out to
be a wonderfully powerful tool in the study of quantum field theories with extended
supersymmetry. The basic idea is that one can usually solve quantum field theories at
weak coupling. There we can identify the various states and understand the spectrum
of long and short multiplets. As one moves into the strong coupling realm, we typically
lose control over the dynamics. However, the short multiplets are special because their
mass is pinned to be m = |Z|. The mass can’t deviate from |Z| because this would
need there to be extra states in the Hilbert space and these can’t magically appear
from nowhere as some parameter, like a coupling constant, is varied. The only way
that the short multiplets can free themselves from this constraint is if two or more short
multiplets become degenerate and hen combine to become a long multiplet whose mass
is no longer protected. By understanding when this can (or, better yet, can’t) happen
we get a precious handle on the strong coupling dynamics of certain quantum field
theories.

In this way, the study of short BPS multiplets shines a rare light into what happens
at strong coupling. It allows us to effectively solve the dynamics of N = 2 and N = 4
gauge theories. It also allows us to understand the strong coupling limits of string
theory, including the existence of M-theory, and to compute the microscopic entropy
of certain BPS black hole solutions. It is, in short, a very useful tool.

The BPS trick is not available for N = 1 theories and so we won’t be wielding it for
much of these lectures. (Actually, it can be used to compute the tension of domain walls
and vortex strings in certain N = 1 theories, but not the masses of particle states.)

2.4.3 Supersymmetry in Other Dimensions

Throughout these lectures, we will restrict ourselves to supersymmetric theories in
d = 3 + 1 spacetime dimensions. There are, however, many interesting things to say
about supersymmetric theories in other dimensions. Here we merely make a few very
simple comments.

Supersymmetric Gauge Theories in Different Dimensions

We’ve seen that the vector multiplet of N = 1 supersymmetry has a photon paired
with a single massless Weyl spinor. This works because both have two internal degrees
of freedom in d = 3 + 1 dimensions. We can ask: in what other spacetime dimensions
might we be able to pair a photon with a fermion?

The number of polarisation states of a photon is d − 2. So the question really is:

in what dimensions does a spinor have d − 2 degrees of freedom? We will see that we

– 42 –
can have a supersymmetric theory in which a photon pairs with a single fermion in
d = 3, 4, 6 and 10 Lorentzian spacetime dimensions.

The story is simplest in d = 3 + 1 and d = 5 + 1. In even spacetime dimension d, a

Dirac spinor has 2d/2 complex components. But the irreducible representations of the
Lorentz group are Weyl spinors with 2(d−2)/2 complex components. While a complex
scalar has two degrees of freedom, a complex spinor has the same number of degrees
of freedom as the number of components. This is because the Dirac equation (or Weyl
equation) is first order so these components include both “position” and “momentum”.
This means that if we want the number of degrees of freedom of a Weyl spinor to match
those of a photon then we need to solve the equation

2(d−2)/2 = d − 2

The solutions are d = 4 and d = 6 as advertised.

In d = 3 + 1 dimensions we can choose to impose either a Majorana condition or a

chiral projection to a Weyl fermion. However in d = 2 mod 8 spacetime dimensions, it
is possible to impose both a Majorana and Weyl condition. This halves the number of
degrees of freedom of a Weyl fermion. Attempting to match the degrees of freedom of
a Majorana-Weyl fermion to a photon we have

2(d−4)/2 = d − 2 with d = 2 mod 8

The unique solution is d = 10.

Finally we’re left searching solutions in odd spacetime dimensions. It is not hard
to see that there is just one possibility. In d = 2 + 1 dimensions, a photon has just
a single polarisation state. Meanwhile, a Dirac spinor in d = 2 + 1 has two complex
components. However we can impose a Majorana condition to make the spinor real.
(For example, we can take the real Clifford algebra γ 0 = iσ 2 , γ 1 = σ 2 and γ 2 = σ 3 .)
So a Majorana spinor in d = 2 + 1 has two real components and, correspondingly, one
degree of freedom, matching that of the photon.

If we’re not in the magic spacetime dimension d = 3, 4, 6 or 10 then we can still have
supersymmetric theories that relate a photon to a fermion. But now we need to include
extra scalar degrees of freedom as well to make up the numbers.

The fact that the number of fermion degrees of freedom increases exponentially with
d, while the number of bosonic degrees of freedom increases only linearly, suggests that
there may be a maximum spacetime dimension in which supersymmetry is possible.

– 43 –
Indeed this is the case. If we don’t wish to get our hands dirty with supergravity then
d = 9 + 1 dimensions is the highest we can go. If we’re happy to include gravity in
the mix then there is a unique supersymmetry theory in d = 10 + 1 dimensions known,
reasonably enough, as eleven dimensional supergravity. It is extremely interesting and
describes the low-energy behaviour of M-theory.

Extended Supersymmetry and Higher Dimensions

There is a close relationship between supersymmetric theories in higher dimensions and
extended supersymmetry. In particular, theories with N = 2 supersymmetry naturally
descend from d = 5 + 1 dimensions while those with N = 4 supersymmetry come from
d = 9 + 1 dimensions. (This statement, taken at face value, is true only at the classical
level. But there are also a myriad of subtle and wonderful connections at the quantum
level, none of which will be touched upon in these lectures.)

To see this, we will briefly jump ahead of ourselves slightly and use the language of
fields, rather than the language of single particle quantum states that we’ve invoked
until now. The relationship between theories in different dimensions involves a pro-
cess known as dimensional reduction. This means that we take the fields in a higher
dimension and state, by fiat, that they are independent of certain spatial coordinates.
For example, consider a gauge field AM in, say, d = 5 + 1 dimensions. This means
that M = 0, 1, . . . , 5. Upon dimensional reduction, we insist that this gauge field only
depends on xµ with µ = 0, 1, 2, 3. The gauge field itself then decomposes as

AM → (Aµ , ϕ4 , ϕ5 )

That is, we get a d = 3 + 1 dimensional gauge field Aµ together with two real scalars ϕ4
and ϕ5 . But this is precisely the bosonic content of the N = 2 vector multiplet that we
found above. A d = 5 + 1 Weyl fermion decomposes into two d = 3 + 1 Weyl fermions
in a similar fashion (although you have to work a little harder playing around with the
gamma matrices to see this).

Playing the same game with a d = 9 + 1 gauge field, we find a d = 3 + 1 gauge field
together with 10 − 4 = 6 scalars. This is the bosonic content of the N = 4 vector
multiplet that we found above. Decomposing a d = 9 + 1 Majorana-Weyl fermion
completes the story, giving four d = 3 + 1 Weyl fermions.

Finally, if you dimensionally reduce eleven dimensional supergravity you find N = 8

supergravity in d = 3 + 1 dimensions.

– 44 –
Counting Supersymmetries
The way in which we count supersymmetries in different dimensions can be rather
bewildering when you first meet it. In d = 3 + 1 we count supersymmetries by the
number of Weyl spinor supercharges QIα with I = 1, . . . , N . But this is clearly specific
to 4d. In other dimensions the counting depends on what kinds of minimal spinors we
can construct. Moreover, if we dimensionally reduce then what is a minimal supersym-
metry in a higher dimension typically becomes an extended supersymmetry in a lower
dimension.

To avoid this confusion, it can be useful to count the number of components of the su-
percharges. We count these as N (rather than the calligraphic N .) These components
are, sadly, also referred to a supercharges! Because spinors can be real in some dimen-
sions, we count the number of real components or, equivalently, twice the number of
complex components. This means that, in d = 3+1 dimensions, N = 1 supersymmetry
has four supercharges, N = 2 has eight supercharges, and so on.

To orient you, here are a list of some of the most interesting classes of supersymmetric
theories and how they are labelled in various dimensions. The list is by no means
complete but gives some sense of the more compelling supersymmetric stories out there.

The maximum number of supercharges is N = 32. These are all supergravity theories
and can exist in any dimension d = 10 + 1 and below. Upon dimensional reduction,
the number of minimal spinor supercharges N in various dimensions is

Dimension d 11 10 6 4
N=32 supercharges:
Supersymmetry N 1 (1,1) (2,2) 8

This is not an exhaustive list: supersymmetric theories with N = 32 supercharges exist

in all dimension d ≤ 11. But the dimensions listed above are, for various reasons, the
most interesting and well studied.

Note the strange (n, n) notation in d = 5 + 1 and d = 9 + 1. This is because of one

more subtlety of representations of the Clifford algebra. When d = 2 mod 4, the two
types of Weyl spinor are not related by complex conjugation in Lorentzian signature.
This means that you can have a spinor of one chirality without necessarily having the
other. In contrast, when d = 0 mod 4 (including, as we saw in great detail, in d = 3+1)
the complex conjugate of a left-handed spinor is a right-handed spinor, so if you have
one then you always have the other. The notation (n, n) tells us how many left- and
right-handed spinor supercharges we have.

– 45 –
 There is another supergravity theory in d = 9 + 1 dimension which has also 32
supercharges but with N = (2, 0) supersymmetry. This is more commonly known as
Type IIB supergravity, with the N = (1, 1) theory known as Type IIA. They are the
low-energy descriptions of Type IIA and IIB string theories.

Theories with N = 16 supercharges can exist in dimensions d = 9 + 1 and below.

Upon dimensional reduction, the associated supersymmetry is:

Dimension d 10 6 4 3 2
N=16 supercharges:
Supersymmetry N (1,0) (1,1) 4 8 (8,8)

The most famous and well studied of these is the Yang-Mills theory associated to the
N = 4 vector multiplet in d = 3 + 1. It has many remarkable properties, including
electromagnetic duality and the fact that, at strong coupling, it is can be viewed as
a theory of quantum gravity through the AdS/CFT correspondence. There are also
interesting stories to tell about the quantum dynamics of the theories in d = 2 + 1 and
d = 1 + 1 dimensions.

There is one further interesting theory with 16 supercharges. This is a strongly

interacting superconformal quantum field theory in d = 5 + 1 dimensions with N =
(2, 0) supersymmetry. In some ways, it can be viewed as the grandfather of all quantum
field theories. Given its importance, it has a remarkably rubbish name: it is simply
called the (2, 0) theory.

Theories with N = 8 supercharges exist in d = 5 + 1 dimensions and below. Upon

dimensional reduction, the names of the supersymmetries that one finds are

Dimension d 6 4 3 2
N=8 supercharges:
Supersymmetry N (1,0) 2 4 (4,4)

Again, the theories with N = 2 supersymmetry in d = 3 + 1 dimensions are the best

studied and were first solved by Seiberg and Witten.

Theories with N = 4 supercharges exist in d = 3 + 1 dimensions and below. Upon

dimensional reduction, this becomes

Dimension d 4 3 2
N=4 supercharges:
Supersymmetry N 1 2 (2,2)

Much of the focus of these lectures notes will be on understanding the dynamics of
N = 1 theories in d = 3 + 1 dimensions. But there are many beautiful stories in lower

– 46 –
dimensions as well. In particular, the study of superconformal N = (2, 2) theories in
d = 1 + 1 dimensions is where one can first find the mathematical study of mirror
symmetry. There are also interesting 2d theories with N = (0, 4) supersymmetry.

Finally, theories with N = 2 supercharges exist in d = 2 + 1 dimensions and below.

The dimensional reduction to d = 1 + 1 gives

Dimension d 3 2
N=2 supercharges:
Supersymmetry N 1 (1,1)

There are also N = (0, 2) theories that do not descend from d = 2 + 1 dimensions.
Note that these are usually written as (0, 2) rather than (2, 0) to give an extra hint
that we’re talking about 2d theories rather than the 6d theory mentioned above.

I’ve not included d = 0 + 1 theories in the above list, also known as quantum
mechanics, but it’s not for want of things to say. You can read about supersymmetric
quantum mechanics in the companion lecture notes.

– 47 –
3 Chiral Superfields
In the previous section we’ve understood how supersymmetry acts on single particles
states in the Hilbert space. But, ultimately, we want to write down field theories that
are invariant under supersymmetry. Part of this requires understanding how super-
symmetry acts on fields.

We’ve already seen a taster of this in the introduction. The action (1.1) was given
by
" #
Z 2 2 2 †
∂W 1 ∂ W 1 ∂ W
S = d4 x ∂µ ϕ† ∂ µ ϕ − iψ̄σ̄ µ ∂µ ψ − − ψψ − ψ̄ ψ̄ (3.1)
∂ϕ 2 ∂ϕ2 2 ∂ϕ† 2

This involves a complex scalar ϕ and a single Weyl fermion ψα . After our discussion in
the last section, we now recognise this as the fields corresponding to a chiral multiplet.
We claimed in the introduction that this action is invariant under the transformation
√ √ √ ∂W †
δϕ = 2ϵψ and δψα = 2iσαµα̇ ϵ̄α̇ ∂µ ϕ − 2ϵα (3.2)
∂ϕ†
There are a few questions that we’d like to ask. First: how can we construct actions like
(3.1)? After all, it’s not like we can just stare at the action and see that it’s invariant
under the transformations (3.2). It takes a bit of work to show this. Secondly, how are
the transformations (3.2) related to the supercharges and supersymmetry algebra that
we met in the previous section.

The purpose of this section is to answer these questions. In particular, we’ll see how
we can rewrite the action (3.1) in a way that the supersymmetry is manifest. The trick
to doing this is to combine the bosonic field ϕ and the femionic field ψα into a single
object known as a superfield.

3.1 Superspace
Usually, fields are functions of xµ , the coordinates of Minkowski space. But, as we’ve
seen, supersymmetry is an extension of the Poincaré group. Correspondingly, super-
fields live not on Minkowski space, but on an extension of Minkowski space known as
superspace.

The coordinates of superspace are

xµ , θα , θ̄α̇

– 48 –
Here xµ , with µ = 0, 1, 2, 3 are the coordinates of Minkowski space. In superspace these
are augmented with Grassmann-valued spinors θα and θ̄α̇ . In other words, superspace
is not a regular manifold of the kind that we know and love from courses on differential
geometry. Instead it is an example of a supermanifold, with both commuting and
anti-commuting dimensions.

3.1.1 The Geometry of Superspace

In what follows, we’ll explore the idea of fields on superspace and see how they en-
capsulate a collection of fields that transform into each other under supersymmetry.
However, we could reasonably ask: how did we come up with the idea of superspace in
the first place? There is, it turns out, a group theoretic answer to this.

In general, if we’re given a Lie group G, we might want to know what manifolds M
accommodate a natural action of G.

One obvious choice is to take the manifold to be the group itself: M = G. In this
case, each element g ∈ G gives us natural map M 7→ M given by g ′ ∈ M 7→ g · g ′ .

A slightly less obvious choice is to take a coset space. This is the manifold M = G/H
where H ⊂ G is a subgroup of G. A point {g} in the coset G/H is defined by the
equivalence relation among elements of G

g ≡ g · h for all h ∈ H

Again, any element g ∈ G gives us a natural map M = G/H 7→ G/H defined by

{g ′ } ∈ M 7→ {g · g ′ }.

For example, the group G = SU (2) is, as a manifold, G = S3 . We can consider the
subgroup H = U (1) ⊂ SU (2) to get the coset SU (2)/U (1) ∼ = S2 . (Mathematically,
this is known as the Hopf fibration.) Obviously there is a natural action of SO(3) ∼
=
SU (2)/Z2 on S2 .

This, somewhat abstract, way of thinking gives us a new perspective on Minkowski

space itself. It can be viewed as the coset
Poincaré Group
R1,3 = G/H =
Lorentz Group
Here a general element of the Poincaré group G is comprised of Lorentz boosts, gener-
ated by M µν , and translations generated by P µ . We write this as
 
i µν µ
g(ω, a) = exp − ωµν M + iaµ P
2

– 49 –
Meanwhile, the Lorentz group H consists only of Lorentz boosts. This means that coset
space can be parameterised just by aµ which we can equivalently think of as coordinates
xµ = aµ on Minkowski space. The fact that Minkowski space can be viewed a a coset
merely confirms something that we knew already: there is an action of the Poincaré
group on Minkowski space.

Now, however, we would like to construct a space on which the group of supersym-
metry transformations naturally acts. These are given by
 
i µν µ α α̇
g(ω, a, θ, θ̄) = exp − ωµν M + iaµ P + iθ Qα + iθ̄α̇ Q̄ (3.3)
2

with Qα and Q̄α̇ the supersymmetry generators that we met in the previous section.
The spinors θα and θ̄α̇ should be viewed as parameterising the “amount” of super-
symmetry transformation that we’re doing, albeit with the “amount” now somewhat
harder to quantify as it’s a Grassmann valued object. With Grassmann elements of
this kind, g is an element of a super Lie group which, in this case, is known as the
super-Poincaré group. The coset construction continues to work in the same way and
we define superspace to be
Super-Poincaré Group
Superspace = G/H =
Lorentz Group
A point in superspace is now parameterised by xµ = aµ and the Grassmann-valued
spinors θα and θ̄α̇ as advertised above.

Before we go on, a quick comment on nomenclature. The Lorentz group is, of course,
SO(1, 3). (Actually, strictly speaking if we want to include spinor representations it
is SL(2, C) = Spin(1, 3) but we’ll ignore this double cover subtlety.) The Poincaré
group is the semi-direct product ISO(1, 3) = SO(1, 3) ⋉ R4 and Minkowski space is
R1,3 = ISO(1, 3)/SO(1, 3). Meanwhile, the super-Poincaré group is usually written as
ISO(1, 3|1) with the additional “bar 1” or “slash 1” telling us that we have N = 1
supersymmetry. Superspace is then the “4+4” dimensional supermanifold R1,3|4 =
ISO(1, 3|1)/SO(1, 3). We’ll have no need for any of this notation in these lectures.

The Action on Superspace

The whole point of the coset construction of superspace is that it tells us how the
supergroup acts. This will be important in what follows so let’s flesh it out a little.
First, we write the general element of the supergroup (3.3) as

g(ω, x, θ, θ̄) = g̃(x, θ, θ̄)h(ω)

– 50 –
where h(ω) is a Lorentz transformation and g̃(a, θ, θ̄) is the representative of the coset

g̃(x, θ, θ̄) = exp ixµ P µ + iθα Qα + iθ̄α̇ Q̄α̇

This specifies a point (x, θ, θ̄) in superspace,

We now want to see how the momentum operator P and supercharges Q and Q̄
shift the point (x, θ, θ̄) in superspace. Let’s start with the momentum operator. We
introduce the supergroup element

U (a) = exp (iaµ P µ )

Then we have

U (a) g̃(x, θ, θ̄) = eiaP eixP +iθQ+iθ̄Q̄ = ei(x+a)P +iθQ+iθ̄Q̄ = g̃(x + a, θ, θ̄)

This gives us a familiar result: momentum generates translations,

x µ → x µ + aµ

Now we do the same for the supercharges. This time we will find a small twist to the
story. We introduce the supergroup element

V (ϵ, ϵ̄) = exp iϵα Qα + iϵ̄α̇ Q̄α̇

Note that ϵα and ϵ̄α̇ are Grassmann-valued spinors. They shouldn’t be confused with
the anti-symmetric ϵαβ matrices that we met earlier. (Sorry!) Now the action on
superspace is given by

V (ϵ, ϵ̄) g̃(x, θ, θ̄) = eiϵQ+iϵ̄Q̄ eixP +iθQ+iθ̄Q̄ (3.4)

The small twist is that Q and Q̄ do not anti-commute with each other. In fact, now
that we’ve multiplied the supercharges with anti-commuting spinors ϵ and θ, we can
talk about commutation relations rather than anti-commutation relations. We have

Qα Q̄α̇ + Q̄α̇ Qα = 2σαµα̇ Pµ ϵα Qα Q̄α̇ + Q̄α̇ Qα θ̄α̇ = 2(ϵα σαµα̇ θ̄α̇ )Pµ


⇒
⇒ [θ̄α̇ Q̄α̇ , ϵα Qα ] = 2(ϵσ µ θ̄)Pµ (3.5)

where the Grassmann nature of θ̄, ϵ, Q and Q̄ means that we pick up a minus sign in
going from the first line to the second, turning { , } into [ , ].

– 51 –
 We now evaluate (3.4) using the BCH formula
1
eA eB = eA+B+ 2 [A,B]+...

The commutator (3.5), together with the fact that the higher commutator terms . . . in
the BCH formula all vanish in the present case, gives us the result

V (ϵ, ϵ̄) g̃(x, θ, θ̄) = eixP +i(θ+ϵ)Q+i(θ̄+ϵ̄)Q̄+(ϵσθ̄)P −(θσϵ̄)P

= g̃(x + iθσϵ̄ − iϵσ θ̄, θ + ϵ, θ̄ + ϵ̄)

Here we see the twist. The supercharges shift the Grassmann coordinate in superspace
as we might have anticipated. But, at the same time, they also shift the point in
Minkowski space by a Grassmann bilinear

xµ → xµ + iθσ µ ϵ̄ − iϵσ µ θ̄
θ → θ+ϵ (3.6)
θ̄ → θ̄ + ϵ̄

Note that the shift in xµ due to the Grassmann bilinear can’t be thought of as nor-
mal translation by some number. Instead, it’s a more formal expression. Ultimately,
we’ll see how this manifests itself in terms of the superfields and their more familiar
components.

3.1.2 Superfields
A superfield is a function on superspace, Y = Y (x, θ, θ̄). To start, we take this to be a
complex-valued function on superspace.

In principle, the superfield could transform in some non-trivial representation of

the Lorentz group. For example it could carry a vector index µ or a spinor index α.
However, rather remarkably, we will find all the fields that we need – scalar, spinor and
vector – lurking within the simplest scalar superfield. (We will, however, come across
superfields carrying spinor indices in Section 4.)

To see this, we Taylor expand the superfield in θ and θ̄. But this is easy because θ
and θ̄ are Grassmann valued objects obeying, for example,

θα θβ = −θβ θα

This means that the Taylor expansion truncates after some finite length. In particular
we have θα θβ θγ = 0. So the Taylor expansion of Y (x, θ, θ̄) stops after terms quadratic

– 52 –
in θ and θ̄. Expanding the superfield out in this way then reveals a bunch of more
familiar fields lurking within,
Y (x, θ, θ̄) = ϕ(x) + θα ψα (x) + θ̄α̇ χ̄α̇ (x) + θ2 M (x) + θ̄2 N (x)
+ θα θ̄α̇ Vαα̇ (x) + θ2 θ̄α̇ λ̄α̇ (x) + θ̄2 θα ρα (x) + θ2 θ̄2 D(x) (3.7)
Here θ2 = θα θα and θ̄2 = θ̄α̇ θ̄α̇ .

There are a few things to say about this. First, note that the superfield does indeed
contain all the fields that we usually care about: there are four complex scalars ϕ, M ,
N and D, two left-handed spinors ψ and ρ, two right-handed spinors χ̄ and λ̄ and a
vector Vαα̇ = σαµα̇ Vµ .

Second, note that it contains many more fields that we might have thought from our
analysis in the previous section! The representations on single particle states suggested
that there should be a chiral multiplet containing a single complex scalar and a Weyl
fermion and a vector multiplet containing a gauge field and a Weyl fermion. Yet the
superfield Y contains a plethora of such fields. We will shortly see how we can impose
further restrictions on Y that truncate the number of fields lying within to match our
earlier expectation.

Our next task is to understand how superfields transform under supersymmetry

transformations. We’ll again start with translations xµ → xµ + aµ which, as we have
seen, are generated by the unitary operator
U = exp (iaµ Pµ )
Previously, we viewed this as a group element acting on superspace. But in quantum
field theory, it has another avatar as an operator acting on the Hilbert space. The fields
in quantum field theory are, of course, also operators and the superfield is no different.
The action of U on such operators enacts the translation, meaning
U Y (x, θ, θ̄)U † = Y (x + a, θ, θ̄)
For infinitesimal aµ , we expand U = eiaP = 1 + iaµ P µ + O(a)2 . We also Taylor expand
the field, Y (x + a) = Y (x) + aµ ∂µ Y (x) + O(a2 ). Equating the terms linear in a we see
that the translations are captured in the commutation relation on fields
[Pµ , Y ] = −i∂µ Y (3.8)
We can treat the action of the supercharges in a similar fashion. We again have the
unitary operator
V (ϵ, ϵ̄) = exp iϵα Qα + iϵ̄α̇ Q̄α̇

– 53 –
Acting on superfields, this gives
V Y (x, θ, θ̄)V † = Y (x + iθσ µ ϵ̄ − iϵσ µ θ̄, θ + ϵ, θ̄ + ϵ̄)
where we’ve invoked the transformation of the superspace coordinate (3.6). If we
now treat ϵα as an infinitesimal spinor and work to leading order in ϵ, we find the
commutation relations
 
∂ µ α̇
[Qα , Y ] = −i α − σαα̇ θ̄ ∂µ Y (3.9)
∂θ
 
∂ α µ
[Q̄α̇ , Y ] = +i α̇ + θ σαα̇ ∂µ Y (3.10)
∂ θ̄
In this expression, the derivatives with respect to Grassmann coordinates are defined
by
∂
∂α = α with ∂α θβ = δαβ and ∂α θ̄β̇ = 0
∂θ
∂
∂¯α̇ = α̇ with ∂¯α̇ θ̄β̇ = δα̇β̇ and ∂¯α̇ θβ = 0
∂ θ̄
These Grassmann derivatives are themselves Grassmann. This means that they pick
up a minus sign when they pass through other Grassmann variables. So, for example,
if you wish to differentiate χβ θγ , where both χ and θ are Grassmann variables, then
you have
∂ β γ β γ ∂
α
(χ θ ) = δ α θ and α
(χβ θγ ) = −δαγ χβ
∂χ ∂θ
where that extra minus sign in the second expression comes from dragging the ∂/∂θα
through the χβ before it gets to attack its prey.

It’s useful to define differential operators associated to the right-hand sides of (3.8),
(3.9) and (3.10). To this end, we write
Pµ = −i∂µ
Qα = −i∂α − σαµα̇ θ̄α̇ ∂µ (3.11)
Q̄α̇ = +i∂¯α̇ + θα σ µ ∂µ
αα̇

Be warned: these differ from the operators Pµ , Qα and Q̄α̇ only by the use of curly
calligraphic script. You can check that anti-commutation relation of these differential
operators is something familiar
{Qα , Q̄α̇ } = 2σαµα̇ Pµ
together with {Qα , Qβ } = {Q̄α̇ , Q̄β̇ } = 0. This is telling us that P, Qα and Q̄α̇
also furnish a representation of the supersymmetry algebra, now acting on fields on
superspace

– 54 –
Supersymmetry Transformation of Fields
We can unpack the supersymmetry transformations (3.9) and (3.10) to see how it acts
on the individual fields sitting with Y . The infinitesimal change of the superfield is
defined to be
δY = i[ϵQ + ϵ̄Q̄, Y ] = i(ϵQ + ϵ̄Q̄)Y (3.12)
Expanding out Y in terms of the components (3.7), the operators Q and Q̄ act on each
term. Q removes a θ (where there is one) and adds a θ̄∂µ (where there aren’t too many
θ̄’s already) Obviously Q̄ is the conjugate. We then compare the various θ and θ̄ and
terms.

For example, the lowest term in Y is the scalar ϕ(x). To compute its variation, we
look for the term in δY with neither θ’s nor θ̄’s. This comes from ∂α acting on the term
θψ and ∂¯α̇ acting on θ̄χ̄. The result is
δϕ = ϵψ + ϵ̄χ̄ (3.13)
Meanwhile, the highest term in Y is the scalar D(x). To compute its variation, we find
the term in δY that comes with the full complement of θ2 θ̄2 . This happens comes from
the θ̄∂µ term in Q and the θ∂µ term in Q̄. The net effect is that the variation of D(x)
is a total derivative
i
δD = ∂µ (ϵσ µ λ̄ − ρσ µ ϵ̄) (3.14)
2
This will prove to be part of the story as we proceed.

It takes a bit of work to get the transformation of all the remaining component fields
in (3.7). You’ll have the pleasure of doing this work in the first examples sheet. The
answer turns out to be
δψ = 2ϵM + (σ µ ϵ̄)(i∂µ ϕ + Vµ )
δ χ̄ = 2ϵ̄N − (ϵσ µ )(i∂µ ϕ − Vµ )
i
δM = ϵ̄λ̄ − ∂µ ψσ µ ϵ̄
2
i µ
δN = ϵρ + ϵσ ∂µ χ̄ (3.15)
2
i
δVµ = ϵσµ λ̄ + ρσµ ϵ̄ + (∂ ν ψσµ σ̄ν ϵ − ϵ̄σ̄ν σµ ∂ ν χ̄)
2
i ν µ
δ λ̄ = 2ϵ̄D + σ̄ σ ϵ̄ ∂µ Vν + iσ̄ µ ϵ ∂µ M
2
i
δρ = 2ϵD − σ ν σ̄ µ ϵ ∂µ Vν + iσ µ ϵ̄ ∂µ N
2
The variation of each has at least two terms, one with a derivative ∂µ and one without.

– 55 –
3.1.3 Constraining Superfields
As we already commented, the superfield Y is too big. It has way more fields than
we expect from the representation theory of Section 2.3. This is because Y is not an
irreducible representation. It can be reduced to something smaller. The question is:
how?

We want to impose constraints on Y such that it remains a superfield. That means

that whatever object we have after the constraint should also transform as (3.9) and
(3.10) under supersymmetry transformations. So our first step to understanding the
possible constraints is to figure out what kind of operations we can perform on super-
fields that keep them as superfields.

There are some obvious operations, albeit ones that won’t help with our constraint.
If we have two superfields Y1 and Y2 then αY1 is a superfield for any α ∈ C, as is Y1 + Y2
and Y1 Y2 . For example, to see that Y1 Y2 is a superfield, we need to note that

[Qα , Y1 Y2 ] = [Qα , Y1 ]Y2 + Y1 [Qα , Y2 ] = (Qα Y1 )Y2 + Y1 (Qα Y2 ) = Qα (Y1 Y2 )

as required.

More pertinent for our purposes, if Y is a superfield then so too is ∂µ Y . However,

crucially, neither ∂α Y nor ∂¯α̇ Y are superfields. Algebraically, this is because

[ϵα Qα , ∂¯α̇ ] = ϵα σαµα̇ ∂µ ̸= 0

To build some intuition for what’s going on, note that ∂¯α̇ Y doesn’t include, for example,
the highest component θ2 θ̄2 D term; there was such a term in Y but one of the θ̄’s is
removed after acting with ∂¯α̇ . However, acting with a supercharge Qα will generate
such a term. In other words, it’s not consistent with supersymmetry to simply state by
fiat that the last term vanishes, D(x) = 0. Act with a supersymmetry transformation
and this will no longer be true. It’s analogous to setting A3 = 0 in a vector field Aµ
and thinking that you’ve found an object with just three components, only to realise
that A3 gets resurrected after a rotation.

However, there is a way forward. We define the covariant derivatives

Dα = ∂α + iσαµα̇ θ̄α̇ ∂µ
D̄α̇ = −∂¯α̇ − iθα σ µ ∂µ
αα̇

These are very similar to the Qα and Q̄α̇ differential operators defined in (3.11), but
with a relative minus sign difference (and an overall factor of i difference). Their key

– 56 –
property is that they anti-commute with Q and Q̄

{Dα , Qβ } = {Dα , Q̄β̇ } = {D̄α̇ , Qβ } = {D̄α̇ , Q̄β̇ } = 0 (3.16)

The covariant derivatives also obey

{Dα , D̄α̇ } = 2σαµα̇ Pµ (3.17)

together with {Dα , Dβ } = {D̄α̇ , D̄β̇ } = 0.

From (3.16), we have

[ϵQ + ϵ̄Q̄ , Dα ] = [ϵQ + ϵ̄Q̄ , D̄α̇ ] = 0

This tells us that both Dα Y and D̄α̇ Y are superfields. For example, under the super-
symmetry transformation (3.12), we have

δY = i(ϵQ + ϵ̄Q̄)Y ⇒ δ(Dα Y ) = i(ϵQ + ϵ̄Q̄)Dα Y

Now we can discuss the various constraints that we can place on a superfield Y . There
are four of interest (of which, only three will play a major role in these lectures).
• A chiral superfield Φ is defined by the constraint

D̄α̇ Φ = 0

• An anti-chiral superfield Ψ is defined by the constraint

Dα Ψ = 0

Note that you can’t impose both chiral and anti-chiral conditions since the anti-
commutator (3.17) would then require that the superfield is actually constant.
Moreover, if Φ is a chiral superfield then Φ̄ = Φ† is an anti-chiral superfield.
(I give a simple way to see this at the end of Section 3.1.4.) The fact that we
can’t impose both conditions simultaneously means that we can’t take Φ to be
real: chiral superfields are necessarily complex. We will see that chiral superfields
correspond to the chiral multiplets that we met in Section 2.3.

• A real superfield V is defined by the simple requirement that

V =V†

We will postpone our discussion of real superfields to Section 4. There we will

see that the real superfields correspond to the vector multiplet that we met in
Section 2.3.

– 57 –
 • Finally, a linear superfield J is defined

J = J† and D2 J = D̄2 J = 0

These play a slightly less prominent role than the (anti)-chiral and real super-
fields. In particular, we won’t build supersymmetry actions out of linear super-
fields. However, it turns out that they are useful homes for certain composite
operators in quantum field theory, most notably Noether currents associated to
global symmetries.

We will spend the rest of this section studying the properties of chiral superfields.

3.1.4 Chiral Superfields

A chiral superfield obeys the constraint

D̄α̇ Φ = 0 (3.18)

We will first solve this equation to understand what it means for the superfield Φ.

There’s a useful trick here. We introduce the coordinate

y µ = xµ + iθσ µ θ̄

The advantage of this coordinate is that we have

 
D̄α̇ y µ = −∂¯α̇ − iθα σαν α̇ ∂ν xµ + iθβ σβµβ̇ θ̄β̇ = −iθα σαµα̇ − i∂¯α̇ (θβ σβµβ̇ θ̄β̇ ) = 0

where to see that the two terms cancel, you have to remember that you pick up an
extra minus sign as the ∂¯α̇ passes through the θβ . In addition, we have

D̄α̇ θβ = 0

This means that if we view a general superfield as a function of Φ = Φ(y, θ, θ̄) then, of
the three arguments, only D̄α̇ θ̄β̇ ̸= 0 and the condition (3.18) tells us

D̄α̇ Φ = 0 ⇒ Φ = Φ(y, θ)

In other words Φ is almost a function only of θ and not of θ̄, the “almost” because
there is in fact a θ̄ buried in the y µ . This means that we can expand in components
√
Φ(y, θ) = ϕ(y) + 2θψ(y) + θ2 F (y)

– 58 –
 √
where the 2 is a convention. We can then further Taylor expand the y µ to get the
expression for a chiral superfield in components
√
Φ(x, θ, θ̄) = ϕ(x) + 2θψ(x) + θ2 F (x)
i 1
+ iθσ µ θ̄∂µ ϕ(x) − √ θ2 ∂µ ψ(x) σ µ θ̄ − θ2 θ̄2 □ϕ(x) (3.19)
2 4
with □ = ∂µ ∂ µ . We see that the chiral superfield contains just three component
fields: a complex scalar ϕ, a Weyl spinor ψ and another complex scalar F . The higher
components of Φ(x) are simply derivatives of the first two fields.

This is much closer to what we expected based on our analysis in Section 2.3. There
we found a chiral multiplet consists of single particle states associated to a complex
scalar ϕ and a Weyl fermion ψ. However, we’ve also got a second complex scalar F .
We will see later that this is an object known as an auxiliary field. For now it’s worth
noticing that, in contrast to ϕ and ψ, there are no terms in the chiral superfield with
∂F . This will be important as we proceed.

The supersymmetry transformations of the chiral multiplet are

√
δϕ = 2ϵψ
√ √
δψ = 2iσ µ ϵ̄ ∂µ ϕ + 2ϵF (3.20)
√
δF = 2iϵ̄σ̄ µ ∂µ ψ

Note that F transforms as a total derivative, just like D in the original unconstrained
superfield (3.14). We’ll see the relevance of this shortly.

There is a very similar story for the anti-chiral superfields. As we mentioned previ-
ously, these can be viewed as the complex conjugate of a chiral superfield. To see this,
note that if a chiral superfield Φ(y, θ) is function of y µ and θ, then its conjugate Φ† (ȳ, θ̄)
is a function of ȳ µ = xµ − iθσ µ θ̄ and θ̄. But it’s simple to check that Dα ȳ µ = Dα θ̄α̇ = 0
and so Φ† is indeed an anti-chiral superfield obeying Dα Φ† = 0. In components, we
have
√
Φ† (ȳ, θ̄) = ϕ† (ȳ) + 2θ̄ψ̄(ȳ) + θ̄2 F † (ȳ)

We can then further expand out ȳ further if we wish to get an expression analogous to
(3.19),
√
Φ† (x, θ, θ̄) = ϕ† (x) + 2θ̄ψ̄(x) + θ̄2 F † (x)
i 1
− iθσ µ θ̄∂µ ϕ† (x) + √ θ̄2 θσ µ ∂µ ψ̄(x) − θ2 θ̄2 □ϕ† (x)
2 4

– 59 –
3.2 And. . . Action
To construct actions that are invariant under Poincaré group, we take suitable La-
grangian densities of fields and integrate them over spacetime. Analogously, to con-
struct actions that are invariant under supersymmetry, we take suitable Lagrangian
densities of superfields and integrate them over superspace.

3.2.1 Integrating Over Superspace

First, let’s remind ourselves how Grassmann integration works. (It is, happily, much
easier than normal integration!) If we have a single Grassmann variable θ then
Z Z
dθ 1 = 0 and dθ θ = 1

This means that if we have a function f (x, θ) = f0 (x) + θf1 (x), then Grassmann
integration picks out the component multiplying θ,
Z
dθ f (x, θ) = f1 (x)

In this manner, integration over Grassmann variables is the same thing as differenti-
R
ation: dθ = ∂/∂θ. In particular, we have a Grassmann version of the fundamental
theorem of calculus
Z Z
∂f
dθ = dθ f0 (x) = 0 (3.21)
∂θ

Here we will need to integrate over superspace, parameterised by θα and θ̄α̇ . We define
Z Z Z Z
2 1 1 2 2 1
dθ= dθ dθ and d θ̄ = − dθ̄1 dθ̄2
2 2
1
Those strange factors of 2
are because θ2 = θα θα = −2θ1 θ2 . We then have
Z Z
d θ θ = − dθ1 dθ2 (θ1 θ2 ) = 1
2 2

where the minus sign disappears when dθ2 moves past θ1 . Note that the measure
d2 θ̄ comes with an extra minus sign but this cancels the corresponding minus sign in
R
θ̄2 = θ̄α̇ θ̄α̇ = +2θ̄1 θ̄2 . Once again, we have d2 θ̄ θ̄2 . Finally, we also use the (not
entirely logical) notation
Z Z
d θ = d2 θ d2 θ̄
4

– 60 –
Now suppose that we build an action out of some function of superfields. That function
will itself be a superfield that we will call K(x, θ, θ̄) but, in contrast to what we’ve
discussed so far, we’ll view K as a composite superfield whose component are functions
of other fields. We the construct the action of the form
Z
S = d4 x d4 θ K(x, θ, θ̄) (3.22)

The action is real if K is a real superfield, obeying K = K † . As we saw above, this is

a valid constraint on a superfield. Under a supersymmetry transformation, we have
Z
δS = d4 x d4 θ δK

where any superfield K must change as (3.12). This means that we have

δK = ϵα (∂α K − iσαµα̇ θ̄α̇ ∂µ K) + (−∂¯α̇ K + iθα σαµα̇ ∂µ K)ϵ̄α̇

But each of these terms involves a derivative. Those terms that are differentiated
with respect to a Grassmann coordinate automatically vanish when integrated over
superspace by virtue of (3.21). Meanwhile, those terms that involve a differential ∂µ
give at most a boundary term which, if fields drop off suitably quickly asymptotically,
also vanishes. We learn that any action of the form (3.22) is necessarily invariant under
supersymmetry:

δS = 0

In fact, we can give an expression for the action. The superfield K has an expansion

K(x, θ, θ̄) = Kfirst (x) + . . . + θ2 θ̄2 Klast (x)

The action (3.22) simply picks up the last of these terms

Z
S = d4 x Klast (x)

We refer to terms in the action that come from integrating over all of superspace as
D-terms. The name isn’t a great one but comes from the fact that the last component
in a real superfield is usually denoted D.

In anticipation of this, in the general expansion of the superfield (3.7) we called the
final term D. We also saw that it transforms as a total derivative under a supersym-
metry transformation (3.14). This gives another way of seeing the result above: any
Lagrangian given by a D-term transforms as a total derivative and so the action is
invariant.

– 61 –
3.2.2 The Action for Chiral Superfields
What does this mean for our chiral superfield Φ? As with any other field, we have a
choice of what action to build. But, typically in quantum field theory, the simplest
possibilities are the most interesting.

Because Φ is complex, we also necessarily have the anti-chiral superfield Φ† to play

with. Multiplying these together gives a real superfield Φ† Φ that we can integrate over
superspace to get the action,
Z
Schiral = d4 x d4 θ Φ† Φ

This means that the action is given by the D-term of Φ† Φ. A short calculation, and
some integration by parts, shows that the action becomes
Z
Schiral = d4 x ∂µ ϕ† ∂ µ ϕ − iψ̄σ̄ µ ∂µ ψ + F † F
 

where we have thrown away some total derivatives. These are just the standard kinetic
terms for a complex scalar ϕ and Weyl fermion ψ. But now we see that there’s some-
thing special about F : it doesn’t have any kinetic terms. Moreover, this will continue
to be true as we write down further supersymmetric interactions. This is what it means
to be an auxiliary field.

Because there are no kinetic terms for F , it has no propagating degrees of freedom
and, when quantised, doesn’t give rise to any particle states. That’s why it didn’t
appear in our representation theory analysis of Section 2.3. Nonetheless, there is a
good reason that F appears in the chiral superfield.

When looking at single particle states, we previously argued that there have to be
equal number of bosonic and fermionic degrees of freedom. And there are. But now
we’re looking at the action, we can ask two variants of this question. First, we can insist
that the number of physical propagating degrees of freedom match. In the context of
field theory, these are said to be “on-shell” degrees of freedom. This means that we
count the degrees of freedom after imposing the equations of motion. The complex
scalar field ϕ has two degrees of freedom, while the non-propagating scalar F has
none. Meanwhile, the Weyl fermion ψα has two complex components but obeys a first
order, rather than second order equation of motion which means that ψα counts both
“position” and “momentum”. So the equation of motion cuts the number of on-shell
degrees of freedom, giving two. This, of course, matches the degrees of freedom of ϕ.

– 62 –
 However, we require the action to be invariant under supersymmetry for all field
configurations, not just those that obey the equations of motion. And this motivates
us to count the “off-shell” degrees of freedom, meaning the number of fields before
equations of motion are imposed. The two complex scalars ϕ and F have two each,
while the Weyl spinor ψα has four off-shell degrees of freedom because it contains two
complex components. The presence of the auxiliary field F is required to match these
off-shell degrees of freedom.

Next we want to write down supersymmetric masses and Yukawa-type interactions

for these fields. These don’t arise from D-terms. Indeed, you could try writing down
a more general function K(Φ, Φ† ) and integrating over d4 θ but you’ll find that it
R

doesn’t generate the kind of interactions we want. (We’ll see what it does generate in
Section 3.2.4.) Instead we have to do something different.

This something different is an option that arises only for chiral superfields. Roughly
speaking, because a chiral superfield depends on only half of superspace, we can get a
supersymmetric action by integrating it over only half of superspace.

More precisely, given a chiral superfield Φ the function W (Φ) is also a chiral super-
field. In components it reads
√ ∂W 1 ∂ 2W
 
2 ∂W
W (Φ) = W (ϕ) + 2 θψ + θ F− ψψ + . . .
∂ϕ ∂ϕ 2 ∂ϕ2

where the + . . . are the extra terms on the second line of (3.19) that include a θ̄ term.
But, as you can see in (3.19), each of these is a total derivative and so will not contribute
to the action. This means that, for the purposes of building an action, we can think
of W (Φ) as a function only of θ and not of θ̄. This means that we can construct a
supersymmetric action by integrating over only half of superspace
Z Z Z 
4 2 2 † †
SW = d x d θ W (Φ) + d θ̄ W (Φ )

where the second term is the Hermitian conjugate of the first and is needed to make
the action real. This action picks out the θ2 term in W (Φ) and is known as an F-term,
so named because the auxiliary field in a chiral multiplet is usually called F .

We see in (3.20) that the F field (and, by extension any F term that multiplies θ2
in a chiral multiplet) transforms as a total derivative under supersymmetry. This gives
us another way to see that the action SW in indeed invariant under supersymmetry.

– 63 –
 Putting together the D-term and F-term contributions, we get our final supersym-
metric action
1 ∂ 2W
Z   
4 † µ µ † ∂W
S = Schiral + SW = d x ∂µ ϕ ∂ ϕ − iψ̄σ̄ ∂µ ψ + F F + F − ψψ + h.c.
∂ϕ 2 ∂ϕ2

This is known as the Wess-Zumino action. The function W (Φ) is called the superpo-
tential.

(An aside: There is a completely different object that is also called the Wess-Zumino
action, or sometimes the Wess-Zumino-Witten or WZW action. This is a topological
term that involves an integral over a higher dimensional space. It has nothing to do
with supersymmetry. You can read about it in the lectures on Gauge Theory.)

As promised, the auxiliary field F appears only algebraically in the action. For such
fields, it is legitimate to eliminate it by the equation of motion which, in this case,
reads simply

∂W † ∂W
F+ †
= 0 and F † + =0
∂ϕ ∂ϕ
Putting this back into the action gives us an action just in terms of those fields that
have propagating degrees of freedom,
" #
Z 2 2 2 †
∂W 1 ∂ W 1 ∂ W
S = d4 x ∂µ ϕ† ∂ µ ϕ − iψ̄σ̄ µ ∂µ ψ − − ψψ − ψ̄ ψ̄
∂ϕ 2 ∂ϕ2 2 ∂ϕ† 2

This is the form of the action that we met back in the introduction in (1.1). We see
that the scalar potential is positive definite and takes the form
2
∂W
V (ϕ, ϕ† ) =
∂ϕ
We still have to specify the form of the superpotential. In general, this can be any
holomorphic function of ϕ. If want to restrict ourselves to theories that are renormalis-
able then we should take a superpotential that is no greater than cubic. For example,
we could take
m 2 λ 3
W (Φ) = Φ + Φ (3.23)
2 3
In general, both m and λ can be complex. This gives the potential
2
V = mϕ + λϕ2

– 64 –
After expanding this out, the mass of the scalar field is |m|. Note that, in addition to
the |ϕ|4 term, there are also cubic terms ϕ2 ϕ† and ϕ† 2 ϕ. These give Feynman diagrams
in which a single ϕ particle splits into two others which means that particle number is
not conserved in the Wess-Zumino model and, relatedly, there is no way to distinguish
particles from anti-particles. This is related to the fact the theory does not have a U (1)
global symmetry in the presence of the general superpotential (3.23) with m, λ ̸= 0.

With a cubic superpotential, the equation of motion for the Weyl fermion is

iσ̄ µ ∂µ ψ + m⋆ ψ̄ = −2λ⋆ ϕ† ψ̄

The fermion also has mass |m|. There is no U (1) symmetry associated to this fermion
and the mass is an example of a Majorana mass. Note also that the Yukawa term
on the right-hand side specifies the interaction between the fermion and scalar and is
characterised by the same coupling λ that determines the self-interaction of the scalar.
This will have important consequences when we turn to the quantum theory.

Multiple Chiral Superfields

There is a straightforward generalisation of the Wess-Zumino action to multiple chiral
superfields Φi . We now take the action
Z X † Z Z 
4 4 4 2
S = d xd θ Φi Φi + d x d θ W (Φ) + h.c. (3.24)
i

where if we wish the theory to be renormalisable we should again restrict to a cubic

superpotential
1 1
W (Φ) = mij Φi Φj + λijk Φi Φj Φk
2 3
The resulting potential is
X ∂W 2
V (ϕ) =
i
∂ϕi

Again, this is positive definite as it must be in a supersymmetric theory since the energy
is necessarily positive.

As we have seen, for a single massive chiral multiplet the Weyl fermion necessarily
has a Majorana mass. With two chiral multiplets, we may have a Dirac mass. Let’s
call the chiral multiplets Φ and Φ̃. Then the simple superpotential

W = mΦ̃Φ

– 65 –
gives rise to two Weyl equations, each of which mixes the spinors ψ and ψ̃,

iσ̄ µ ∂µ ψ + m⋆ ψ̃¯ = 0 and iσ̄ µ ∂µ ψ̃ + m⋆ ψ̄ = 0

This is the Dirac equation, decomposed into two Weyl pieces. (Sorry for the ugliness
of piling a bar on top of a tilde.) Note that it now has a U (1) symmetry, under which
ψ and ψ̃ (or, equivalently the superfields Φ and Φ̃) rotate with opposite charges.

3.2.3 Supersymmetry of the Wess-Zumino Model Revisited

It’s worth pausing for a recap. We’ve derived the Wess-Zumino model which, for a
single chiral superfield, before integrating out F , is given by

1 ∂ 2W
Z   
4 † µ µ † ∂W
S= dx ∂µ ϕ ∂ ϕ − iψ̄σ̄ ∂µ ψ + F F + F − ψψ + h.c.
∂ϕ 2 ∂ϕ2

Our arguments involving superspace have told us that this action is invariant under
the supersymmetry transformations (3.20).
√
δϕ = 2ϵψ
√ √
δψ = 2iσ µ ϵ̄ ∂µ ϕ + 2ϵF
√
δF = 2iϵ̄σ̄ µ ∂µ ψ

together with the hermitian conjugate transformations

√
δϕ† = 2ϵ̄ψ̄
√ √
δ ψ̄ = − 2iϵσ µ ∂µ ϕ† + 2ϵ̄F †
√
δF † = 2iϵσ µ ∂µ ψ̄

But this is something that we can just check. It’s a little tedious but, given the
importance of this result, it’s worth doing. From our discussion above, we know that
the kinetic terms and the superpotential terms should be independently invariant. We
can check each in turn. First the kinetic terms. We have

Z
d4 x ∂ µ ϕ† ∂µ δϕ − iδ ψ̄ σ̄ µ ∂µ ψ + F † δF + h..c
 
δSchiral =

We’ve kept only half the terms, the other half buried in the hermitian conjugate.
(Admittedly, there was some forethought involved in which terms to keep to ensure

– 66 –
that they cancel among themselves.) Using the supersymmetry transformations above,
we have
√ Z 4  µ †
δSchiral = 2 d x ∂ ϕ ϵ∂µ ψ − ∂ν ϕ† ϵσ ν σ̄ µ ∂µ ψ − iF † ϵ̄σ̄ µ ∂µ ψ + iF † ϵ̄σ̄ µ ∂µ ψ + h.c.


We see that the two terms with F † cancel immediately. For the other two terms we have
a little bit of work to do. Note that, by integrating by parts twice, we can symmetrise
over (µν) in the second term. But you can check that σ (ν σ̄ µ) = η µν which then ensures
that the first two terms also cancel and δSchiral = 0.

For the superpotential terms we have

∂ 2W ∂ 2W 1 ∂ 3W
Z  
4 ∂W
δSW = d x δF +F δϕ − ψ δψ − ψψ δϕ + h.c.
∂ϕ ∂ϕ2 ∂ϕ2 2 ∂ϕ3

The final ∂ 3 W/∂ϕ3 term multiplies ψ 3 and so vanishes because ψ is a 2-component

Grassmann field. We’re then left with
√ Z 4 ∂ 2W ∂ 2W ∂ 2W
 
µ ∂W µ
δSW = 2 d x iϵ̄σ̄ ∂µ ψ +F ϵψ − i 2 ψσ ϵ̄∂µ ϕ − F ϵψ + h.c.
∂ϕ ∂ϕ2 ∂ϕ ∂ϕ2

The F ϵψ terms cancel immediately. The other two cancel after an integration by parts,
together with the fact that ψσ µ ϵ̄ = −ϵ̄σ̄ µ ψ. We then have δSW = 0 as promised.

There is also a version of this calculation after we have integrated out the auxiliary
field F , replacing it with its equation of motion F = −∂W † /∂ϕ† . As we’ve seen, the
Wess-Zumino action becomes
" #
Z 2 2 2 †
∂W 1 ∂ W 1 ∂ W
S = d4 x ∂µ ϕ† ∂ µ ϕ − iψ̄σ̄ µ ∂µ ψ − − ψψ − ψ̄ ψ̄
∂ϕ 2 ∂ϕ2 2 ∂ϕ† 2

We can also replace F in the supersymmetry transformations. These become

√ √ µ √ ∂W †
δϕ = 2ϵψ and δψ = 2iσ ϵ̄ ∂µ ϕ − 2ϵ †
∂ϕ

The calculation described above goes through with only minor modifications (although
you can no longer treat the kinetic and superpotential terms independently). This is
the supersymmetry invariance of the Wess-Zumino model that we promised back in the
introduction.

– 67 –
3.2.4 Non-Linear Sigma Models
The restriction to a cubic superpotential above is motivated by the requirement that
the theory be renormalisable. But for theories of scalars, this requirement isn’t always
at the top of our list. The reason is that these theories may arise as the low-energy
description of something more interesting. In this situation, there’s no reason to think
that the low-energy description should be valid at arbitrarily high-energy scales and so
no reason to impose renormalisability.

An illustrative analogy can be found in QCD. At high energies this is a theory of

quarks and gluons but at low energies, after confinement has imposed itself on the
dynamics, it is a theory of light scalar particles called pions. We denote these fields
as π i (x) with i labelling the different pion fields. (For what it’s worth, i = 1, . . . , 8 in
QCD if we include mesons that contain up, down and strange quarks.) The low-energy
dynamics of pions takes the form
Z
SNLSM = d4 x gij (π) ∂µ π i ∂ µ π j (3.25)

Theories of this kind go by the unhelpful name of non-linear sigma models. The fields
π i can be thought of as coordinates on some manifold M that is called the target space.
The interactions are hiding in the derivative terms and are packaged into a collection
of functions gij (π) that can be viewed as a metric on M. The action (3.25) describes
massless scalar fields, although it is always possible to add mass terms if necessary.

Actions of the type (3.25) arise in many places in physics. We first meet them in
General Relativity as the action for particles (rather than fields) moving in a curved
space or spacetime. But they also occur in many places in condensed matter physics
and statistical physics. (The O(N ) models discussed in the lectures on Statistical
Field Theory are an example.) You can learn more about the specific metric gij (π)
that describes pion dynamics in Section 5 of the lectures on Gauge Theory. Here, our
interest is in writing down supersymmetric versions of non-linear sigma models.

We can achieve this simply by introducing more interesting D-terms. We consider

n chiral superfields Φi with i = 1, . . . , n. We’ll denote the anti-chiral superfields as Φ̄ī
with the ī = 1, . . . , n index a useful reminder that these label anti-chiral fields. We
then consider the action
Z
S = d4 x d4 θ K(Φ, Φ̄) (3.26)

with K(Φ, Φ̄) any real function of these superfields. This function is known as the
Kähler potential.

– 68 –
 Previously, we took
X
K= Φ† ī Φi
i

We will refer to this as the canonical Kähler potential. It is the form that we must take
if we want our theory to renormalisable. But if we’re willing to entertain low-energy
effective theories then we can take a general, real function K. To compute the resulting
action, we simply need to compute the D-term of K(Φ, Φ† ). This calculation is a little
laborious but the result is quite beautiful. The supersymmetric non-linear sigma model
takes the form
Z h  i i i µ

4 i µ j̄ i µ j̄ j̄ i j̄
S = d x gij̄ ∂µ ϕ ∂ ϕ̄ + ∂µ ψ σ ψ̄ − ψ σ ∂µ ψ̄ + F F̄
2 2
1 ∂gij̄ 
k i j̄ j̄ µ i k

+ ψ ψ F̄ − i ψ̄ σ ψ ∂ µ ϕ + h.c.
2 ∂ϕk
1 ∂ 2 gij̄ i k j̄ l̄
i
+ (ψ ψ )(ψ̄ ψ̄ ) (3.27)
4 ∂ϕk ∂ ϕ̄l̄
where the metric gij̄ is related to the Kähler potential as

∂ 2K
gij̄ = (3.28)
∂ϕi ϕ̄j̄
Note that this metric only only has components with one holomorphic and one anti-
holomorphic index. We can eliminate the auxiliary field F through its equation of
motion
1 ∂gij̄ k i 1 ∂gij̄ l̄ j̄
gij̄ F i + k
ψ ψ = 0 and gij̄ F̄ j̄ + ψ̄ ψ̄ = 0
2 ∂ϕ 2 ∂ ϕ̄l̄
Substituting this back into the action, we find
Z    
4 i µ j̄ i i µ j̄ i i µ j̄ 1 i k j̄ l̄
S = d x gij̄ ∂µ ϕ ∂ ϕ̄ + Dµ ψ σ ψ̄ − ψ σ Dµ ψ̄ + Rij̄kl̄ (ψ ψ )(ψ̄ ψ̄ )
2 2 4
Rather wonderfully, all the terms now take a nice geometrical form. The kinetic term
for the fermion involves a kind of covariant derivative, defined by

Dµ ψ i = ∂µ ψ i + Γijk ψ j ∂µ ϕk

where, for a metric given by (3.28), the Christoffel symbol is given by

∂gkl̄
Γijk = g il̄
∂ϕj

– 69 –
Meanwhile, the four-fermion interaction terms comes multiplying the Riemann tensor.
For a metric given by (3.28), this too takes a special form
∂Γm ik ∂ 2 gij̄ ∂gin̄ ∂gmj̄
Rij̄kl̄ = gmj̄ = − g mn̄
∂ ϕ̄ l̄ k
∂ϕ ϕ̄ l̄ ∂ϕk ∂ ϕ̄l̄
We have stumbled upon the mathematical framework of Kähler geometry. This is a
particular form of complex geometry that can be placed on manifolds that are even
dimensional and can be endowed with complex coordinates, like the ϕi and above. A
Kähler manifold is a manifold that is endowed with a Kähler two-form
Ω = 2igij̄ dϕi ∧ dϕ̄j̄
such that
dΩ = 0
This requires that the gij̄ satisfies
∂gij̄ ∂gkj̄ ∂gij̄ ∂gil̄
k
= and =
∂ϕ ∂ϕi ∂ ϕ̄l̄ ∂ ϕ̄j̄
This condition is locally equivalent to the existence of a Kähler potential K(ϕ, ϕ̄), with
the metric given by (3.28).

Finally, note that the Kähler potential is not unique. The action (3.26) is invariant
under any shift
K(Φ, Φ̄) + Λ(Φ) + Λ̄(Φ̄)
where Λ(Φ) is any holomorphic function of Φi . This is because Λ(Φ) is a chiral superfield
and necessarily vanishes when integrated over all of superspace. These shifts are called
Kähler transformations.

Supersymmetry has led us to the mathematical framework of Kähler geometry. This

is just one of many close connections between supersymmetry and interesting geo-
metric structures. Some of these connections are explored further in the lectures on
Supersymmetric Quantum Mechanics.

Adding a Superpotential
The supersymmetric non-linear sigma model (3.27) describes massless fields. We can
always add an additional superpotential W (Φ) to the action. We won’t write down the
full action, but simply comment that the scalar potential now takes the form
∂W ∂W †
V (ϕ, ϕ̄) = g ij̄ (3.29)
∂ϕi ∂ ϕ̄j̄
with g ij̄ the inverse metric.

– 70 –
A Comment on Supergravity
Throughout these lectures we will restrict ourselves to theories with global, or rigid,
supersymmetry. As we’ve mentioned previously, if one extends supersymmetry to a
gauge symmetry, making it local, then the resulting theory necessarily includes gravity.
This is supergravity. In this case, the scalar potential for a bunch of chiral multiplets
again has a fixed form, depending only on the Kähler potential K and superpotential
W . It is
!
2
2 |W |
V (ϕ, ϕ̄) = eK/Mpl g ij̄ Di W Dj̄ W † − 3 2 (3.30)
Mpl
where
∂W 1 ∂K
Di W = + W
∂ϕi Mpl2 ∂ϕi
Here Mpl is the Planck mass. In the limit that Mpl → ∞, gravity becomes arbitrarily
weak and the potential (3.30) reduces to our previous potential (3.29).

Perhaps surprisingly, the supergravity potential is not positive definite. This is re-
lated to the fact that supersymmetric theories can exist in anti-de Sitter spacetimes
with a negative cosmological constant.

3.3 Non-Renormalisation Theorems

So far our discussion of supersymmetric theories has been entirely classical. But the
great advantage of supersymmetry is that it allows us to gain control over the quantum
dynamics of the theory.. We can start to understand this already just with chiral mul-
tiplets. In this section we will show that the superpotential does not receive quantum
corrections at any order in perturbation theory. This is known as a non-renormalisation
theorem. In contrast, all bets are off with the Kähler potential: it is no more constrained
than the kinetic terms in any other quantum field theory.

The original proof of the non-renormalisation theorem used Feynman diagrams for
superfields. This means that we write down a diagram in which, say, the propagators
correspond to superfields. These “super-Feynman diagrams” then encode a number of
normal Feynman diagrams, some with bosons running in loops and others with fermions
running in loops. One can then show that the most general super-Feynman diagram
doesn’t contribute to the superpotential.

In these lectures, we’re not going to develop the machinery of superfield Feynman
diagrams. Instead, we will give a much simpler argument that uses only the symmetries
of the problem.

– 71 –
 Before we get going, an important comment. Throughout these lectures, theories of
chiral superfields will typically be viewed as low-energy effective actions. More precisely,
they will be viewed as Wilsonian low-energy effective actions. This means that they
describe physics only on some suitably large length scale, or equivalently at energies less
than some UV cut-off, E ≤ ΛU V . All short distance, or high energy, degrees of freedom
have been integrated out but may, in some cases, leave an imprint on the low-energy
degrees of freedom. We’ll see examples of this as we proceed.

A Wilsonian effective action already takes into account any quantum effects above
the cut-off ΛU V . But not those below. You need to use the action to compute, for
example, loop diagrams to understand the low-energy quantum dynamics. But there
are no UV divergences because the action comes equipped with an explicit cut-off.

There is another, more formal kind of effective action that is common in quantum
field theory. This is the one particle irreducible, better known as 1PI, effective action. It
arises as the Legendre transform of the (log of) the partition function. In contrast to the
Wilsonian effective action, the 1PI effective action is best viewed as a classical action,
with all quantum effects already taken into account. This can be problematic in the
presence of massless particles since the 1PI effective action may have IR singularities.
In contrast, there is no such problem with the Wilsonian effective action.

3.3.1 R-Symmetry Revisited

Given a quantum field theory, one of the first things we should do is understand its
symmetries. The kind of Wess-Zumino models (or, more generally non-linear sigma
models) that we’ve described above could have many different Abelian or non-Abelian
global symmetries acting on the chiral superfields Φi . However, there is one that is of
particular importance. This is the U (1) R-symmetry. It is special because it does not
commute with supersymmetry. Instead, as we saw in (2.25), it obeys

[R, Qα ] = −Qα and [R, Q̄α̇ ] = +Q̄α̇

This means that the R-charge of the scalar ϕ and fermion ψ in a chiral superfield
necessarily differ. If the scalar has charge r, then the other members of the multiplet
have

R[ϕ] = r ⇒ R[ψ] = r − 1 and R[F ] = r − 2 (3.31)

Another way of saying this is to return to the expansion of a chiral superfield (3.19),
√
Φ = ϕ + 2θψ + θ2 F + . . .

– 72 –
We endow the supercoordinate θ with an R-charge

R[θ] = +1

This tallies with our expression (3.11) for the supercharge Q ∼ ∂/∂θ +. . . which tells us
that Q and θ have opposite charges. The upshot is that if the superfield has R-charge
R[Φ] = r, then the other charges in (3.31) follow.

So when do theories enjoy an R-symmetry? Let’s consider the simplest Wess-Zumino

model (3.24) for a single chiral superfield. The D-term, which gives the kinetic terms,
is clearly invariant under any R-symmetry. That leaves the superpotential. This mul-
tiplies d2 θ but Grassmann integration acts in the same way as differentiation which
means that the measure has charge

R[d2 θ] = −2

We see that the action is invariant under R-symmetry only if we can assign charges to
the superfield such that the superpotential has charge

R[W ] = +2 (3.32)

When we have just a single superfield Φ, this is rather limiting. It holds only if the
superpotential is a monomial

W (Φ) = Φn

in which case we can assign R[Φ] = 2/n. For example, if we take W (ϕ) = 21 mϕ2 then
the Lagrangian has an R-symmetry under which ϕ → eiα ϕ and ψ → ψ. This case is a
little boring because there are no interaction terms between ϕ and ψ so obviously we
can rotate them independently. We could, however, take W (ϕ) = 13 λϕ3 in which case
we have the Yukawa term ϕψψ which is invariant under the R-symmetry ϕ → e2iα/3
and ψ → e−iα/3 ψ. However, if we include both mass and Yukawa terms, there is no R-
symmetry. The surprise, as we will now see, is that the lack of an R-symmetry doesn’t
stop it being useful!

3.3.2 The Power of Holomorphy

We will now see what the R-symmetry has to do with the non-renormalisation of the
superpotential. I should warn you that the argument that follows, originally due to
Seiberg, is extremely slick and was developed only after a more nuts and bolts argument
using Feynman diagrams had been found. But the symmetry argument is both easier
and, ultimately, more powerful.

– 73 –
 There are a number of conceptual steps that we need to take before the non-
renormalisation theorem becomes clear. These are all related to the parameters that
appear in the superpotential, things like the mass m and Yukawa coupling λ in (3.23).
Each of these parameters is naturally complex. Moreover, like the chiral superfields
themselves, the superpotential must be a holomorphic function of these parameters.

Of course, as written in (3.23), the superpotential is, by definition, a holomorphic

function of parameters. There’s an m that sits in the first term and a λ in the second
and these are complex. However, the point is that any quantum corrections to the
superpotential must also be holomorphic in parameters. This greatly restrains the
allowed quantum corrections.

There are two ways to argue that the superpotential must be holomorphic in param-
eters. The first is direct, but convoluted, and invokes a kind of supersymmetric Ward
identity. The second way is to say a bunch of words that hopefully makes it obvious.
We’re going to adopt the second way.

In any quantum field theory, we can view parameters as arising from some fixed,
background scalar fields. This means that the parameters may come from some dy-
namical, but very heavy, scalar field with a potential that pins the value of the scalar
to that of the parameter. If this is the case, we wouldn’t notice any difference at low
energies because these new fields are so heavy. We would see the fluctuations of the
parameter only at high energies.

This idea is realised in our world: in the Standard Model the scale of the masses of
all elementary particles is set by the expectation value of the Higgs boson. It’s an idea
that is extended dramatically in string theory where all dimensionless parameters of a
low-energy theory also arise as the expectation value of some scalar. However, it is a
way of thinking that has proven to be useful in many other arenas including, as we will
now see, in supersymmetric theories. The new fields that replace the parameters are
sometimes called spurions.

This change of perspective from parameters to spurions doesn’t change the low-
energy behaviour of the theory. But, remarkably, it does allow us to put constraints
on what this low-energy behaviour can be. These constraints are especially strong
in supersymmetric theories because the spurion must be the lowest component of a
chiral superfield. And, as such, the parameters must appear holomorphically in the
superpotential.

– 74 –
 To understand what this buys us, let’s return to the simple case of a single chiral
superfield with superpotential
1 1
Wtree = mΦ2 + λΦ3 (3.33)
2 3
We refer to this as the tree-level superpotential. Our goal is to understand how it is
changed by quantum corrections.

As we’ve seen above, this theory does not have an R-symmetry. Nonetheless, thinking
of the parameters as spurions suggests that we could think of enlarged symmetries
under which the parameters also transform. In this larger framework, the theory has
two symmetries: one R-symmetry that we call U (1)R and one global symmetry that
commutes with supersymmetry that we call U (1)F . The charges are

U (1)R U (1)F
Φ 1 1
m 0 −2
λ −1 −3
All components of the superfield have the same charge under U (1)F , while the charge
under U (1)R tells us how the lowest scalar component of the superfield transforms,
with other components given by (3.31). Relatedly, the superpotential is invariant under
U (1)F but has charge +2 under U (1)R , as in (3.32).

I stress again that neither U (1)R nor U (1)F are symmetries of our theory since a
true symmetry isn’t allowed to change parameters of the theory. Said another way,
non-vanishing charges for m and λ are telling us that these symmetries are explicitly
broken. Nonetheless, the spurions give a useful book-keeping device to characterise
exactly how the symmetry is broken. Moreover, as we will now see, they also place
strong constraints on the quantum corrections to theory.

Any quantum corrections to the superpotential must be consistent with the two sym-
metries U (1)R and U (1)F . Combined with holomorphy, this becomes a very powerful
constraint on what can appear. We can form a single, dimensionless combination of
superfields that carries no charge at all: this is λΦ/m. (The superfield has the same
dimension as a scalar, namely [Φ] = 1. Meanwhile the mass and Yukawa coupling have
dimensions [m] = 1 and [λ] = 0.) The only kinds of superpotentials that we can write
down consistent with the symmetries are then of the form
 
2 λΦ
Weff = mΦ f
m

– 75 –
Note that holomorphy was key here. In most situations assigning a charge to a complex
parameter isn’t particularly restrictive since, say, |λ|2 carries no charges and so can
appear anywhere. But the fact that only holomorphic quantities can appear in the
superpotential is a game changer.

We still have an arbitrary function f (λΦ/m) that can appear. But this can be
pinned down by studying the theory in different limits. First, for λ ≪ 1, we are in the
weakly coupled limit. This means that for small λ we should reproduce the tree level
superpotential (3.33), perhaps with corrections at order λ2 or higher coming from loop
diagrams. In other words, the expansion of f (x) about x = 0 must take the form
1 1
f (x) = + x + O(x2 )
2 3
However, should also have a well defined superpotential in the limit m → 0 in which we
have massless particles. This tells us that we must have f (x) = 21 + 31 x or, equivalently,

1 1
Weff = mΦ2 + λΦ3 = Wtree
2 3
This is the result we promised: the superpotential receives no quantum corrections to
any order in perturbation theory in λ.

(Looking forward: in Section 6, we will study the quantum dynamics of supersym-

metric gauge theories. There we will find that superpotentials are, in some circum-
stances, dynamically generated. But even there they will not be perturbative effects.
The superpotentials will arise either by some strong coupling effect or by an instanton
effect.)

While the superpotential is immune to quantum corrections, this is not true of the
Kähler potential. There are now no holomorphy restrictions and nothing to prohibit
corrections of order λ2 and higher. This means that the physical masses and Yukawa
couplings do, in fact, receive quantum corrections. To see this, note that typically the
Kähler potential will pick up quantum correction of the form

K(Φ, Φ† ) = Φ† Φ → ZΦ† Φ

where Z = 1 + O(λ2 ) is sometimes, inappropriately, called the wavefunction renormal-

isation. This renormalisation factor will have a characteristic logarithmic form
2
ΛUV
Z = 1 + c|λ|2 log + ... (3.34)
m

– 76 –
Figure 1. The three one-loop diagrams contributing to the mass of the scalar ϕ. As shown
in the last diagram, the dotted line denotes the scalar ϕ and the solid line the fermion ψ.

Here c is a constant whose exact value can be calculated but isn’t of interest for our
purposes and . . . refers to higher loop corrections. This renormalisation changes the
kinetic terms for each of the fields and the action is now
Z Z  
4 4 † 4 2 1 2 1 3
S = d x d θ ZΦ Φ + d x d θ mΦ + λΦ + h.c.
2 3
Importantly, supersymmetry ensures that there is just a single renormalisation Z for
the superfield, meaning that each of the component fields ϕ, ψ and F experiences the
same Z. In such a situation, we should work with the canonically normalised field
Φ̂ = Z 1/2 Φ and the action becomes
Z Z  
4 4 † 4 2 1m 2 1 λ 3
S = d x d θ Φ̂ Φ̂ + d x d θ Φ̂ + Φ̂ + h.c.
2Z 3 Z 3/2
In this way, the non-renormalisation of the superpotential is not enough to protect
the physical mass and Yukawa coupling, which are mphys = m/Z and λphys = λ/Z 3/2
respectively.

This may seem like a disappointing end to our non-renormalisation claim: the super-
potential doesn’t change, but the physical parameters sitting within it do. Nonetheless,
there’s something important going on here. That’s because supersymmetry has ensured
that the mass m2phys picks up only a multiplicative renormalisation.

This contrasts strongly with the mass renormalisation expected of a scalar field in
a typical quantum field theory. Typically, this mass renormalisation is additive. In
particular, any one of the three diagrams shown in Figure 1 would give a contribution
of the form

m2phys ∼ m2 + |λ|2 Λ2UV

This is the statement that quantum fluctuations tend to push the mass of scalar fields
up to the cut-off scale. In the absence of fine tuning (or some other explanation like sym-
metry breaking) scalars in quantum field theory are typically heavy. Yet this doesn’t

– 77 –
happen in supersymmetric theories: miraculously, the additive renormalisation cancels
between each of the diagrams above. This occurs because, as we have seen, the same
coupling λ appears in the Yukawa coupling to the fermions and in the 3-point and
4-point vertices of the scalars. The result is that, in supersymmetric theories, there
is no difficulty with the masses of scalars being small. In particular, if we choose to
set m = 0 in the superpotential so that the chiral multiplet is massless then quantum
corrections do not change this.

This is the key reason that supersymmetry has attracted the interest of phenomenol-
ogists. The mass of the Higgs boson is seemingly much lighter than the cut-off scale of
the Standard Model, an issue referred to as the hierarchy problem. (See the lectures on
Particle Physics for a non-technical account of this.) The existence of supersymmetry
at, say, the TeV scale would provide a natural explanation of this. Sadly, there is no
evidence that this is the explanation favoured by nature.

3.3.3 Integrating Out Heavy Fields

We may sometimes find ourselves in situations in which our theory has two or more
fields with different masses. In this case, we can integrate out the heavier fields, leaving
ourselves with an action just for the lighter ones. This will be an important tool for us
later, so we pause here to see how it works.

Consider the theory of two chiral superfields Φ and Z, both with canonical Kähler
potential K = Φ† Φ + Z † Z, and with superpotential
1 1
W = M Z 2 + λΦ2 Z (3.35)
2 2
In this example, Z is the heavy field with mass M while Φ is massless, but interacts
with Z. If we care only about physics at energies E ≪ M , we can simply integrate out
Z to leave ourselves with a theory for Φ.

Usually in quantum field theory, integrating out fields requires us to evaluate some
complicated functional determinants or Feynman diagrams. But, at the level of the
superpotential, things are straightforward. For a field configuration Φ, the heavy field
will rapidly arrange itself to minimise its energy which it does by adjusting to
∂W λ 2
=0 ⇒ Z=− Φ
∂Z 2M
Substituting this back into the superpotential gives our effective superpotential
1 λ2 4
W =− Φ
8M

– 78 –
This results in a ϕ6 interaction for the scalar, together with the Yukawa-like interaction
for the fermion.

We can also reach the same conclusion by analysing the (spurious) symmetries of the
theory. This time there are two global symmetries, U (1)Φ and U (1)Z in addition to the
R-symmetry. The charges of various fields and parameters are
U (1)R U (1)Φ U (1)Z
Φ 1 1 0
Z 0 0 1
M 2 0 −2
λ 0 −2 −1
The unique superpotential consistent with these symmetries that does not involve Z is
λ2 4
W ∼ Φ (3.36)
M
This symmetry argument doesn’t give the overall constant −1/8 but, as we’ve seen
above, that’s not difficult to get by simply solving the equation of motion.

Note that there’s a different philosophy at play here from when we showed the non-
renormalisation of the superpotetnial (3.33). In the earlier case we insisted that the
superpotential was well behaved as m → 0. However, in the present case the super-
potential clearly diverges as M → 0. But this is to be expected: the theory involving
Φ alone is only supposed to make sense at energies E ≪ M . The fact that the super-
potential diverges as M → 0 is telling us something physical: that we shouldn’t have
discarded the field Z in this limit since it wasn’t heavy. This is a lesson that we will
see several times as these lectures progress: our low-energy theory will break down in
any limit where some field that we have ignored becomes massless.

There’s also a terminological issue here. Physicists refer to the superpotential (3.36)
as “holomorphic” in Φ, λ and M . Strictly speaking it’s not holomorphic in M , but
instead meromorphic because of the pole. As we explained above, the pole certainly
has physical consequence, but we won’t belabour the point and will continue to take
about holomorphy rather than the more accurate meromorphy.

3.3.4 A Moduli Space of Vacua

We can see a twist on this same theme if we study the superpotential (3.35) in the limit
M = 0. We have
1
W = λΦ2 Z (3.37)
2

– 79 –
This theory has a feature that will become increasingly important as these lectures
develop: there is not a unique ground state, or even a finite number of isolated ground
states. Instead the potential energy is given by
2 2
∂W ∂W 1 2
V (ϕ, z) = + = |λϕz|2 + λϕ2
∂ϕ ∂z 4
We’ve now resorted to our earlier notation of referring to the lowest scalar component
of the superfields Φ and Z by the lower case letter ϕ and z respectively. The minima
of the potential are given by

V (ϕ, z) = 0 ⇔ ϕ = 0 and z = anything

This means that the potential has a flat direction. Provided that ϕ = 0, there is no
energy cost to turning on z. We say that there is a moduli space of vacua. In such a
situation, the choice of ground state z is not determined dynamically. Instead, to fully
specify the theory, we must also state the expectation value of the field z. Importantly,
different choices of z give rise to different theories. For example, we can see immediately
from the potential that the mass of ϕ is mϕ = |λz|. In other words, this is moduli space
of inequivalent vacua.

Now the roles of z and ϕ are reversed! Provided that z ̸= 0, the ϕ field is massive
while z is massless. We can again play the kind of game that we saw above: is there
a superpotential W (Z) that we can write down that might arise after Φ is integrated
out? It’s simple to see that the answer is no. Everywhere along the moduli space, we
have

W (Z) = 0

This is important. Had we found W (Z) ̸= 0, it would have meant that there was a
quantum generated potential that lifts the flat direction and that the true quantum
theory has a preferred ground state. But the non-renormalisation theorem tells us that
no such potential is generated. Instead we learn that the moduli space of ground states
survives in the quantum theory.

The existence of a moduli space of inequivalent vacua is commonplace in super-

symmetric theories but never happens in the absence of supersymmetry. In any non-
supersymmetric theory, quantum corrections always generate a potential on the would-
be moduli space. This is known as the Coleman-Weinberg potential and it picks the
true ground state of the system, typically pushing the scalar either to z = 0 or to
z = ∞.

– 80 –
Figure 2. The classical moduli space on the left and the quantum corrected moduli space
on the right, with it’s singularity at z = 0 revealing the massless particle and its negative
signature at large z showing that the quantum theory is ill-defined.

We can get some intuition for the Coleman-Weinberg in a simple quantum mechanics
example. Suppose that we have a quantum particle that can move in the (x, y) plane
but with a potential that we take to be

Vtoy model = x2 y 2

The classical system has two flat directions: x = 0 and y = anything; or y = 0 and
x = anything. Suppose that we sit at some y ̸= 0 but classically set x = 0. We
then look at the quantum system by supposing that y is constant and quantising the
x degree of freedom. But this is just a quantum harmonic oscillator with frequency
given by ω = y. And the ground state energy of the quantum harmonic oscillator is
E ∼ ℏω = ℏy. In this way, the quantisation of x gives rise to an energy that pushes y
back towards the origin. Indeed, this quantum mechanical system has a unique ground
state, localised around the origin.

The Coleman-Weinberg potential is the analogous phenomenon in quantum field

theory. It is generic but is avoided in supersymmetric theories due to a delicate cancel-
lation between bosons and fermions, very similar to those at play in the loop diagrams
in Figure 1. We’ll be meeting many different vacuum moduli spaces as these lectures
progress. Indeed, one of the emerging themes of these lectures is that the geometry of
these moduli spaces contains important clues to the underlying physics.

For now, let’s go back to our field theory (3.37) and ask: what happens to the moduli
space at z = 0? Here the ϕ field also becomes massless and it should no longer be valid

– 81 –
to ignore it. But how do we see this if we’re focussed on the dynamics of z alone?
The answer to this can be found in the Kähler potential. Classically, this takes the
canonical form K = Z † Z, corresponding to to a flat metric

∂ 2K
ds2 = dz dz̄ = dz̄dz
∂z∂ z̄
However, as we saw above, when we integrate out the massive Φ field the Kähler
potential receives a one-loop quantum correction (3.34) and becomes
!
2
Λ U V
K = Z † Z 1 + c|λ|2 log + ... (3.38)
Z

where |Z| appears in the argument of the logarithm courtesy of the role it plays as the
mass of Φ. This results in a metric on the moduli space given by

∂ 2K
   
2 2 z̄z
ds = dz dz̄ = −c|λ| log + constant + . . . dz̄dz
∂z∂ z̄ Λ2U V

We see that distances diverge as we approach z → 0. The log singularity at z = 0 is

the sign that we have attempted to integrate out a massless particle at that point.

There is also some strange behaviour for large |z|. When |z| ≫ ΛU V , the first
term is negative and, for large enough |z|, will overwhelm the constant term, giving
us a negative metric. This, of course, is nonsensical. It’s telling us that our scalar
theory doesn’t make sense at very high expectation values or, equivalently at very high
energies. In other words, it is capturing the phenomenon of the Landau pole in ϕ4
theory, but now in a novel geometric fashion. A depiction of the classical and quantum
moduli spaces is shown in Figure 2.

3.4 A First Look at Supersymmetry Breaking

A symmetry is said to be spontaneously broken if it acts non-trivially on the ground
state. This means that the Noether charge Q for the symmetry fails to annihilate the
vacuum,

̸ 0
Q|0⟩ =

Broken symmetries have important consequences. If a discrete symmetry is sponta-

neously broken then it implies the existence of multiple, isolated ground states. If a
continuous symmetry is spontaneously broken then it implies the existence of a mass-
less particle called a Goldstone boson. These ideas underlie Landau’s classification of

– 82 –
phases of matter and were discussed in some detail in the lectures on Statistical Field
Theory and the lectures on Gauge Theory. In this section, we will make a first pass
at understanding when supersymmetry may be spontaneously broken and what the
consequences are.

First, some basics. From the supersymmetry algebra {Qα , Q̄α̇ } = 2σαµα̇ Pµ we can
derive an expression for the Hamiltonian
1 1
H = P 0 = {Q†1 , Q1 } + {Q†2 , Q2 }
4 4
We already noted in Section 2.2.2 that this implies that all states in a supersymmetric
theory necessarily have energy E ≥ 0. This means that any state with E = 0 must be
a ground state. These states obey

Eground = ⟨0|H|0⟩ = 0 ⇔ Qα |0⟩ = 0

In this case the supercharges annihilate the ground state which means that supersym-
metry is unbroken. Conversely, supersymmetry is spontaneously broken if and only if
the energy of the ground state is non-vanishing

Eground = ⟨0|H|0⟩ > 0 ⇔ Qα |0⟩ =

̸ 0

In other words, the ground state energy Eground is the order parameter for broken
supersymmetry.

There is another way of looking at this. In theories of chiral multiplets (with a

canonical Kähler potential) the potential energy is given by (3.29)
X X ∂W 2
V (ϕ, ϕ̄) = |Fi |2 =
i i
∂ϕi

The ground state energy is non-zero if and only if the F-term gets an expectation value
in the vacuum
∂W †
Fi = − ̸= 0
∂ ϕ̄i
This is known as F-term supersymmetry breaking. (There is another option that involves
vector multiplets known as D-term supersymmetry breaking.)

3.4.1 The Goldstino

If a normal continuous symmetry is spontaneously broken, it results in a massless
particle known as a Goldstone boson. If supersymmetry is spontaneously broken, it
results in a massless fermion that we call a Goldstino.

– 83 –
 First, some intuition. When a normal, continuous symmetry is spontaneously bro-
ken, the symmetry sweeps out a manifold of equivalent ground states. The canonical
example is the breaking of a U (1) symmetry that gives rise to the S1 rim of the Mexican
hat potential. The massless Goldstone mode then arises from fluctuations along this
flat direction.

Something similar happens for supersymmetry. From the supersymmetry transfor-

mations (3.20), we see that when F i ̸= 0, a supersymmetry transformation acting on
the vacuum turns on a linear combination of fermions
√
δψ i = 2ϵF i

This is the Goldstino.

There is a simple, hands-on way to see the existence of this massless fermion within
the class of theories that we’re discussing here. The ground state of the system, whether
supersymmetric or not, sits at

∂V X ∂ 2 W ∂W † X ∂ 2W
=0 ⇒ = − Fj = 0
∂ϕi j
∂ϕi ∂ϕj ∂ ϕ̄j j
∂ϕi ϕj

If supersymmetry is broken then Fj ̸= 0 for some j and the equation above then tells us
that the matrix ∂ 2 W/∂ϕi ∂ϕj necessarily has an eigenvector with vanishing eigenvector.
But ∂ 2 W/∂ϕi ∂ϕj is the fermion mass matrix in our theory. So we learn that when
supersymmetry is broken there is at least one massless fermion.

There is a more powerful, general approach to show the existence of the Goldstino
that holds for the strongly coupled theories that we will discuss later. This is in close
analogy to the original proof of Goldstone’s theorem and we just give a bare bones
sketch here. The idea is to first construct the supercurrent Sαµ . This is the conserved
current associated to supersymmetry transformations and, like any other conserved
current, obeys ∂µ Sαµ = 0. The supercharge Qα arises from this current in the usual
way:
Z
Qα = d3 x Sα0

The supercurrent obeys the algebra

{Qα , S̄α̇µ } = 2σαν α̇ T µν

– 84 –
with Tµν the energy-momentum tensor. This reproduces the usual supersymmetry
algebra (2.21) when integrated over space. The proof of the existence of a massless
Goldstino then proceeds by computing the two-point function

pµ ⟨Sµα (p) S̄ν α̇ (−p)⟩ = −2σαµα̇ ηµν E0

with E0 the ground state energy. This tells us that whenever E0 ̸= 0 there is a pole in
the ⟨S S̄⟩ 2-point function at p = 0. This pole corresponds to a massless fermion, the
Goldstino.

These lectures are very much focussed on more formal aspects of supersymmetry
rather than any possible application to our world. Nonetheless, the existence of the
Goldstino raises a puzzle. Clearly we don’t see supersymmetry at the energies we
have explored so far, which is roughly speaking E ≲ 100 GeV or so. That, in itself,
is not such a big issue since it may well be that supersymmetry is broken at some
higher energy scale. But, in that case the argument above suggests that we would
expect to see a massless Goldstino in our world and no such particle exists. (You might
wonder if perhaps the neutrino could act as a Goldstino. This isn’t possible because
the Goldstino is created from the vacuum and so should share its quantum numbers,
while the neutrino carries electroweak charge.)

The resolution to this lies in supergravity. Recall that supergravity involves a local,
or gauged, version of supersymmetry. When a normal gauge symmetry is broken, the
would-be massless Goldstone boson is “eaten” by the Higgs mechanism and becomes
massive. The same is true of gauged supersymmetry. In the context of supergravity,
the would-be Goldstino is eaten by the gravitino and both become massive with mass
of order E0 , the supersymmetry breaking scale.

3.4.2 The Witten Index

Not all theories can spontaneously break supersymmetry. There is a topological ob-
struction that they must overcome. This is obstruction is the Witten index.

We met the Witten index briefly back in Section 2.3. It defined as the sum over all
states

Tr(−1)F e−βH (3.39)

The trace is taken over the infinite number of states in the quantum field theory Fock
space. Here F is the fermion number, so that the Witten index counts bosonic states
with a +1 and fermionic states with a −1. In contrast to the discussion in Section

– 85 –
2.3, we’ve now included a factor of e−βH , where H is the Hamiltonian. This acts as
a regulator on the very high energy states. But, as we’ll now show, these high energy
states don’t in fact contribute to the Witten index.

To make the discussion precise, we should really work on a compact space, like T3 .
This ensures that momentum is quantised and, correspondingly, the energy spectrum
is discrete. There are then no subtleties in taking the trace.

The key fact about the Witten index is that any states with energy E > 0 necessarily
come in boson-fermion pairs. This follows from the kind of representation theory that
we did in Section 2.3. More precisely, if we define the combination of supercharges
Q = Q1 + Q†2
then, from the supersymmetry algebra (2.21), it is simple to see that these obey
{Q, Q† } = 4H
Consider the action of this operator on a state with energy H|ϕ⟩ = E|ϕ⟩ with E ̸= 0.
We can then define the fermionic creation and annihilation operators
Q
a= √ ⇒ {a, a† } = 1
2 E
This algebra has a two-dimensional irreducible representation |ϕ⟩ and a† |ϕ⟩, both with
energy E. One of these states is bosonic and the other fermionic, ensuring that they
cancel in their contribution to the Witten index.

Note that the degeneracy of E > 0 states is true whether or not supersymmetry is
broken. If supersymmetry is unbroken, it arises because of mass degeneracy of particles
in a supermultiplet. If supersymmetry is broken then the degeneracy arises simply from
the addition of a zero energy Goldstino mode. (More precisely, on a compact space it
arises from the quantisation of the Goldstino zero mode.) In this case, there is no need
for the masses of bosonic and fermionic particles to be equal.

This argument for the degeneracy of the spectrum breaks down for states of zero
energy. For such supersymmetric ground states there is no obstacle to having just a
single state obeying
Qα |0⟩ = Q†α |0⟩ = 0
More generally, it may well be the case that a theory has multiple ground states. In
this case, each ground state could be bosonic or fermionic. Here a “fermionic” ground
state is nothing exotic: it just means that it sits in the sector of the Hilbert space with
(−1)F |0⟩ = −|0⟩ rather than (−1)F |0⟩ = +|0⟩.

– 86 –
Figure 3. The spectrum on the left has Tr(−1)F e−βH = 2 and cannot break supersymmetry
as parameters are changed. The one in the middle has Tr(−1)F e−βH = 0. It does not break
supersymmetry but as parameters are varied there is nothing to protect it from turning into
the spectrum on the right which does break supersymmetry.

The upshot is that the Witten index (3.39) actually counts the difference in the
number of E = 0 ground states

Tr(−1)F e−βH = nB (E = 0) − nF (E = 0)

In particular, the Witten index is independent of the value of β. Moreover, it is

actually independent of any other parameter in the theory. To see this, consider a
generic spectrum of a supersymmetric theory as shown in Figure 3. All E ̸= 0 states
come in pairs, while E = 0 states may be unpaired. As we vary parameters in the
theory, some of the E = 0 ground states may get lifted and get non-zero energy. But
they can only be lifted in pairs and the Witten index remains unchanged. In this sense,
the Witten index provides a topological classification of theory.

(Actually, this last statement is only true providing that asymptotic nature of the
potential does not change. We’ll see an example below.)

All of this means that supersymmetry can only be spontaneously broken in theories
with Tr(−1)F = 0. In contrast, if Tr(−1)F ̸= 0 for some choice of parameters then the
theory cannot break supersymmetry as the parameters are changed and this remains
true even as the dynamics becomes strongly coupled.

– 87 –
An Example
All of the theories that we will explore in this section are weakly coupled and we can
tell whether supersymmetry is broken simply by looking at the potential. This means
that we don’t really have any need for the Witten index. It starts to show its teeth
only for the strongly interacting theories that we will meet in Section 6. Nonetheless,
it’s useful to get a feeling for how supersymmetric ground states are robust.

Consider a Wess-Zumino model with a single chiral superfield Φ with a superpotential

that is a polynomial of degree p + 1,

W (ϕ) = ap+1 ϕp+1 + ap ϕp + . . . + a1 ϕ

A supersymmetric ground state exists if there are solutions to the equation

∂W
=0 (3.40)
∂ϕ
But there’s always a solution to this equation because we’re solving a polynomial over
the complex numbers. In fact, there are always p such solutions (counted with multi-
plicity). As we vary the coefficients ai the ground states move around, but they are
never lifted. This reflects the fact that this theory has Tr(−1)F e−βH = p. It’s a little
fiddly to show that all ground states contribute the same +1 to the Witten index, rather
than with different signs. You can find the argument in the lectures on Supersymmetric
Quantum Mechanics where the Witten index plays a central role throughout.

There is, however, an important caveat to the statement that the theory always has
p ground states. If we set ap+1 = 0 then the superpotential becomes a polynomial of
degree p and the theory has p − 1 ground states. It’s simple to see what happens here:
as we take the limit ap+1 → 0, one of the ground states starts heading off to infinity
in field space ϕ → ∞. This provides a salutary lesson: the Witten index can change if
we change how the theory behaves in the asymptotic region of field space. We will see
other examples below where, as we vary parameters, a moduli space of ground states
emerges then disappears again. This also provides a scenario where the Witten index
can jump.

3.4.3 The O’Raifeartaigh Model

The Witten index argument, together with some basics facts about roots of polynomi-
als, means that you have to strive to write down theories that break supersymmetry.
Nonetheless, it’s not too difficult to achieve. The first model was constructed in 1975

– 88 –
by O’Raifeartaigh. It contains three chiral superfields that we call Y , Z and Φ with
the superpotential
h
W = Y (Φ2 − µ2 ) + mZΦ (3.41)
2
We take all fields to have a canonical Kähler potential so the theory is renormalisable.
(We will relax this assumption below.) The parameter h is dimensionless, while [µ] =
[m] = 1. It’s useful to note that the potential has an R-symmetry (a real one, not a
spurious one) under which R[Y ] = R[Z] = 2 and R[Φ] = 0.

The fields Y and Z act like Lagrange multipliers in the superpotential, setting
∂W h 2 ∂W
Φ − µ2 = 0 and


= = mΦ = 0
∂Y 2 ∂Z
Clearly there’s no way to set both of these to zero so supersymmetry is spontaneously
broken.

The potential of this model is given by

1 2
V (y, z, ϕ) = hϕ2 − hµ2 + |mϕ|2 + |hyϕ + mz|2
4
Note that y and z are just names of scalar fields here; they are not to be confused with
coordinates on spacetime. The minima of the potential always sits at z = hyϕ/m so
the final term vanishes. What happens next depends on the ratio of parameters

hµ
α=
m

If α < 1 then the minima is at ϕ = z = 0. If α > 1 then this minima splits into two
minima at ϕ = ± something and a saddle. Importantly, in either case y is arbitrary: it
is a flat direction.

It is simple to check that the whole superfield Y is massless. The fermion is the
Goldstino while the phase of y is a Goldstone boson associated to a broken R-symmetry.
The surprise is that |y| is also massless, with no symmetry reason to protect it. As we
now explain, the classical moduli space parameterised by |y| doesn’t survive in the full
quantum theory.

– 89 –
The Quantum Generated Potential
Importantly, the mass spectrum of the O’Raifeartaigh model depends on the the value
of |y|: each point on this moduli space describes different physics. Furthermore, and in
contrast to our earlier supersymmetric models, the masses of the bosons and fermions
are different. This is important because it means that when we integrate out these heavy
fields they will induce a Coleman-Weinberg potential on the moduli space parameterised
by |y|. Here we give some general comments on the form of this potential.

Integrating out heavy fields in a 4d quantum field theory usually give three kinds
of divergences: quartic, quadratic and logarithmic. In each case, bosons give rise to
a positive potential and fermions a negative potential. In a supersymmetric theory,
these exactly cancel which is the reason that moduli space of vacua are not lifted when
supersymmetry is broken. As we now explain, when supersymmetry is spontaneously
broken some, but not all, of this cancellation remains.

First the quartic divergences. These are given by

Veff ∼ Str Λ4U V

where ΛU V is the UV cut-off and Str is the supertrace which means that we sum over all
complex bosonic fields minus the sum over all fermionic fields. (Note that we’re sum-
ming over the different fields of the theory here. This contrasts with the Witten index
where we were performing the much larger sum over all states in the Hilbert space.)
But supersymmetric theories have an equal number of bosonic and fermionic fields so
all quartic divergences disappear regardless of whether supersymmetry is spontaneously
broken or not.

Next up are the quadratic divergences. These take the form

Veff ∼ Λ2U V Str M2 = Λ2U V Tr M2B − Tr M2F

Here M is the tree-level mass matrix, including both bosons and fermions. In the
second equality we’ve written it in terms of a sum over bosonic and fermionic fields
with their appropriate mass matrices MB and MF . Clearly this too vanishes when
there is a degeneracy of masses. But a rather nice result says that it also vanishes when
supersymmetry is spontaneously broken:

Claim: Str M2 = 0 for F-term supersymmetry breaking.

Proof: This holds generally in any theory with N superfields and a canonical Kähler

– 90 –
potential. The proof involves just a little bit of algebra. First, the N × N mass matrix
for a Weyl fermion is

∂ 2W
(MF )ij =
∂ϕi ϕj

We write this in terms of the auxiliary field F̄ī = −∂W/∂ϕi as (MF )ij = −F̄īj . The
mass-squared matrix that appears in the supertrace formula is the Hermitian matrix

(MF )2 = (MF )īj (MF )†j k̄ = F̄īj Fj k̄

Meanwhile, we have to be a little more careful with the bosons because after supersym-
metry breaking the real and complex parts of the scalar will typically have different
mass. (This happens, for example, in the O’Raifeartaigh Model.) This means that we
should break the bosons into real and imaginary pieces and consider the 2N × 2N mass
matrix
2 2
!
∂ V ∂ V
∂ϕi ϕ̄j̄ ∂ϕi ϕl
M2B = ∂2V ∂2V
∂ ϕ̄j̄ ϕ̄k̄ ∂ ϕ̄j̄ ϕl

But V = Fi F̄ī . Plugging this expression into M2B above and taking the trace (remem-
bering that there’s a factor of 12 because we’re now working with real fields rather than
complex) gives the claimed result. □

All of which means that in a theory with spontaneously broken supersymmetry, the
only contribution to the effective potential comes from the logarithmic divergences. It
can be shown that these too take the form a supertrace over the mass matrix
 2
1 4 M
Veff = Str M log
64π 2 ΛU V

Again, this vanishes if supersymmetry is unbroken. But now it does not vanish if
supersymmetry is spontaneously broken. This gives the quantum potential that lifts
flat directions in this case.

The mass matrix M depends on the value of the field y, and hence Veff should be
viewed as a potential that lifts this flat direction. In any theory with a flat direction,
quantum generated potentials typically push the field to one end or another. Computing
the masses shows that here the true ground state of the system sits at y = 0. This is
the unique ground state with spontaneously broken supersymmetry.

– 91 –
3.4.4 R-symmetry and the Nelson-Seiberg Argument
We could continue exploring different models (and we will below!) but it is useful to
first stop and try to understand some general features of supersymmetry breaking. To
this end, let’s first look at a small extension of the O’Raifeartaigh model,
h ν ϵ
W = Y (Φ2 − µ2 ) + mZΦ + Φ2 + Y 2 (3.42)
2 2 2
This differs from the O’Raifeartaigh model by the addition of the last two terms. Note
that these two terms break the R-symmetry and this will be important shortly. For
now, we can simply study the scalar potential arising from this superpotential. It is
1 2
V (y, z, ϕ) = hϕ2 − hµ2 + 2ϵy + |mϕ|2 + |hyϕ + mz + νϕ|2
4
Now the theory does have a supersymmetric ground state, sitting at z = ϕ = 0 and
y = hµ2 /2ϵ.

If, however, we now take ϵ → 0 to remove the last term in (3.42), then the super-
symmetric vacuum moves off to infinity in field space y → ∞ and we once again find
ourselves with a theory that breaks supersymmetry, one that appears to be very simi-
lar to the original O’Raifeartaigh model. However, in one way there is a key difference
between them. To describe this difference we first need to explain what it means for
theories to be “generic”.

All the theories we’re discussing in this section should be viewed as low-energy effec-
tive theories, coming from some unknown UV physics. But there is a mantra that can
be applied to such low-energy theories: anything that is not forbidden is mandatory.
This means that quantum effects will conspire to generate all possible terms in the
potential provided that they are consistent with the symmetries of the theory. A low
energy effective theory that includes all such terms, with no particular fine tuning of
the coefficients, will be said to be “generic”.

In this sense, the O’Raifeartaigh model (3.41) is generic. It has an R-symmetry and
there are no further terms that one can add consistent with this symmetry.

In contrast, the extension of the O’Raifeartaigh model (3.42) is not generic. It no

longer has an R-symmetry, but we have not included Z 2 terms nor Φ3 terms nor many
other terms that we could write down. Despite this, it turns out that the behaviour
we have seen – namely the existence of a supersymmetric ground state – persists if we
add all these extra terms. So it is sufficient for our discussion.

– 92 –
 However, among this large class of theories that do not have an R-symmetry, we
only find one that breaks supersymmetry if we set one of the coefficients to vanish:
ϵ = 0. This is a very particular choice of coefficient. If the theory (3.42) arose as the
low-energy limit of some other theory — one which itself did not have an R-symmetry
— then there would be no reason to expect that ϵ = 0. For this reason, it’s unlikely
that the supersymmetry breaking we’ve found in this model is actually useful.

In fact, one can make these kind of arguments more generally. Consider a theory
with N chiral superfields Φi and a potential W (ϕ). A supersymmetric ground state
obeys
∂W
=0 (3.43)
∂ϕi
Supersymmetry is broken if we can cook up a superpotential for which there are no
solutions to this equation. But these are N equations in N variables and for a generic
W they always have a solution. That means that a supersymmetric ground state can
always be found.

It is, however, appropriate to restrict W by symmetry arguments and we might

wonder if that will help us find a generic W that breaks supersymmetry. For example,
suppose that W is invariant under a U (1) global symmetry under which the superfield
Φi transforms with charge qi ,

Φi → eiαqi Φi

In this case the superpotential can always be written as a function of W = W (Xi ) with
Xi the invariant ratios
Φi
Xi = q /q1
i = 2, . . . , N
Φ1i

But now the conditions for a supersymmetric ground state are just ∂W/∂Xi = 0 for
i = 2, . . . , N which are N − 1 conditions for N − 1 variables. Again, for a generic W
there will be a solution. We see that imposing global symmetries doesn’t help us in
finding supersymmetry breaking potentials.

However, the story is different if there is an R-symmetry. We take the superfields to

transform with charges ri ,

Φi → eiαri Φi

– 93 –
We again form the invariant ratios
Φi
X̃i = r /r1
i = 2, . . . , N
Φ1i
The key difference is that the superpotential must have R-charge +2. This means that
it takes the form
2/r1
W (Φ1 , X̃i ) = Φ1 W̃ (X̃i )
The conditions for a supersymmetric ground state are now ∂ W̃ /∂ X̃i = 0. But, as long
2/r
as Φ1 1 ̸= 0, we must also have W̃ (X̃) = 0. This is now N conditions on N − 1
variables X̃i and generically there will not be a solution.

This is the Nelson-Seiberg argument. It says that models of supersymmetry breaking

with generic superpotentials should have an R-symmetry. This is indeed true of the
O’Raifeartaigh model.

Our main interest in these lectures is not to construct realistic supersymmetric

theories, but rather to explore the strong coupling dynamics of quantum field theo-
ries. Nonetheless, it’s worth mentioning that the argument for the existence of an R-
symmetry causes something of a headache if you’re trying to build realistic models in
which supersymmetry is spontaneously broken. In some models, like the O’Raifeartaigh
model, the non-supersymmetric ground state preserves the R-symmetry (recall that,
ultimately, the quantum potential pushes us to y = 0.). But this causes problems
further down the line because, as we will see in Section 4, an R-symmetry prohibits
masses for the superpartners of gauge fields, known as gauginos. But these must be
heavy in any realistic theory.

Alternatively, we could cook up models in which both supersymmetry and the R-

symmetry are spontaneously broken. But this then leads to a light Goldstone boson
known as the R-axion. Again, we must find a way to give this a mass.

3.4.5 More Ways to (Not) Break Supersymmetry

In the remainder of this section, we briefly discuss a number of other simple models
that illustrate different ways in which supersymmetry can be broken.

Runaway Potentials
Here is a model that looks like it breaks supersymmetry but, on closer inspection, does
something different. It consists of two fields, Z and Φ, with superpotential
h
W = ZΦ2 − λΦ
2

– 94 –
It has an R-symmetry with R[Φ] = 2 and R[Z] = −2 and a scalar potential given by
1
V = |hϕ2 |2 + |hzϕ − λ|2
4
Clearly there is no way to set both terms to zero so we seem to again have a situation
in which supersymmetry is broken. However, instead something slightly different is
happening and the potential slopes to zero asymptotically. To see this, look at the
direction with ϕ = λ/hz for which the potential is given by
2
λ2
V (z) =
2hz 2

Clearly V (z) → 0 as z → ∞. So it is better to say that this theory has no stable

ground state at all: the field is pushed to z → ∞ where supersymmetry is restored.
We will see behaviour like this emerging dynamically in Section 6.

Metastable Supersymmetry Breaking

Let’s now consider a slightly different variant of the model (3.42) that broken R-
symmetry. We take the superpotential
h ϵ
W = Y (Φ2 − µ2 ) + mZΦ + Z 2
2 2
The potential is
1 2
V (y, z, ϕ) = hϕ2 − hµ2 + |mϕ + ϵz|2 + |hyϕ + mz|2
4
This breaks R-symmetry and so, on general grounds, we might expect it to have a
supersymmetric vacuum (provided that we have taken the superpotential to be suitably
generic). This is indeed the case: the supersymmetric ground state is given by ϕ2 = µ
and z = −mϕ/ϵ and y = m2 /hϵ.

For ϵ very small, this ground state sits a long way from the origin of field space.
Moreover, if we look close to the origin, y = 0, then the potential is very similar to the
original O’Raifeartaigh model. In particular, when ϕ = z = 0 there is a flat direction
along y, albeit one that is not a global minimum of the the potential. When we include
quantum corrections, this will be lifted and, for suitable values of the parameters, we
will find a local, supersymmetry breaking vacuum at the origin. A schematic sketch of
this situation is shown in Figure 4.

– 95 –
Figure 4. A schematic sketch of the metastable minima at y = 0 that breaks supersymmetry
and the global, supersymmetric ground state at y ∼ 1/ϵ. (The actual potential should be
plotted in higher dimensions.)

In a quantum field theory, any local minima of a potential that is not the global
minimum is a metastable state, with a finite lifetime. This means that if we initially sit
in the supersymmetry breaking minimum, we will eventually tunnel out into the super-
symmetric ground state. Nonetheless, it is possible to use such metastable minima to
build phenomenologically viable models. You just need to make sure that “eventually”
≫ 100 billion years (or whatever allows you to sleep easy at night).

Playing with the Kähler Potential

So far we haven’t discussed the simplest theory that breaks supersymmetry. This is a
single chiral multiplet with superpotential

W = µ2 Φ

Clearly ∂W/∂ϕ = µ2 ̸= 0. But this feels too cheap. The ground state energy may be
non-zero, but the theory is just a free massless fermion (the Goldstino!) and a free
complex scalar. It’s hard to argue that there’s any deep physics in there.

Things change however if we consider a more general Kähler potential K = K(ϕ† ϕ).
The fermion remains massless but a potential is now generated for the scalar, given by
−1
∂ 2K

4
V (ϕ) = |µ|
∂ϕ∂ϕ†

The price that we pay is that the theory is no longer renormalisable. Of course, as
we’ve stressed above, given that we view these scalar field theory as low energy effective
theories, that is not necessarily a bad thing.

– 96 –
 For example, suppose that, when expanded around the origin, the Kähler potential
takes the form
1
K(ϕ, ϕ† ) = |ϕ|2 − |ϕ|4 + . . .
M2
This kind of behaviour can arise from integrating out heavy particles of mass M . (We
found a log correction to the Kähler potential from integrating out particles in (3.38),
but other interactions can give the power-law above.) We should view M as the UV
cut-off of the theory. Other energy scales in the game should necessarily be much
smaller than the cut-off which, for us, means µ ≪ M .

With such a Kähler potential, the actual potential energy reads

 
† 4 4 2
V (ϕ, ϕ ) = |µ| 1 + 2 |ϕ| + . . .
M

This now has a minima at ϕ = 0. The net result is that the scalar ϕ has a mass
mϕ = 2µ2 /M 2 .

A comment on the scales here. As we’ve mentioned repeatedly, all the theories in
this section should be viewed as low-energy effective theories arising from some high
energy completion. In the present case, our theory is valid at energy scales ∼ µ. We
have integrated out stuff at the much higher scale M ≫ µ and this is what gives rise to
the correction to the Kähler potential. It’s necessary that there is a separation of scales
here. Although the scalar ϕ is not massless, it is light in the sense that 2µ2 /M ≪ µ.

Different Kähler potentials can give the different kinds of behaviour that we saw
above, including runaway potentials and metastable vacua.

– 97 –
4 Supersymmetric Gauge Theories
Finally, we turn to the main subject of these lectures: supersymmetric gauge theory. In
this section we will describe the classical structure of supersymmetric gauge theories.
In Section 6 we turn to their quantum dynamics.

4.1 Abelian Gauge Theories

A gauge field Aµ sits inside a real superfield satisfying V (x, θ, θ̄) = V † (x, θ, θ̄). Expand-
ing out such a superfield in components, we have
V (x, θ, θ̄) = C(x) + θχ(x) + θ̄χ̄(x) + iθ2 M (x) − iθ̄2 M † (x) + θσ µ θ̄ Aµ (x)
   
2 i µ 2 i µ
+ θ θ̄ λ̄(x) + σ̄ ∂µ χ(x) + θ̄ θ λ(x) + σ ∂µ χ̄(x)
2 2
 
1 1
+ θ2 θ̄2 D(x) − □C(x) (4.1)
2 2
The real superfield contains two real scalars, C and D, and a complex scalar M , together
with two Weyl fermions χα and λα . Importantly, it also contains a real vector field
Aµ . This will play the role of the gauge field in what follows. We’ve defined some of
the components to include derivatives of others. This should simply be thought of as
a redefinition of D(x) and λ(x), admittedly one that you wouldn’t write down unless
you had an inkling of what was coming.

If Aµ is to be a gauge field, then it must enjoy a gauge transformation. These too

sit in superfield. We start by taking a chiral superfield Ω
√ i 1
Ω = ω + 2θρ + θ2 G + iθσ µ θ̄∂µ ω − √ θ2 ∂µ ρ σ µ θ̄ − θ2 θ̄2 □ω
2 4
then i(Ω − Ω† ) is a real superfield. Consider the generalised gauge transformation
V → V + i(Ω − Ω† ) (4.2)
The vector component of the real superfield shifts as
Aµ → Aµ − 2∂µ (Re ω) := Aµ + ∂µ α (4.3)
But this is precisely the form of a gauge transformation. But under this generalised
gauge transformation, it’s not just Aµ that shifts. The other fields in V (x, θ, θ̄) also
transform as
C → C − 2 Im ω
√
χ → χ + 2iρ
M → M +G

– 98 –
Importantly, however, λ → λ and D → D remain unchanged. This can be traced to the
extra derivative terms that we included in the superfield expansion (4.1) which were
designed to soak up the shift by a chiral superfield.

We can now use this gauge transformation to simply set C = χ = M = 0. This is

known as Wess-Zumino gauge. Note that it’s not a gauge choice that has done anything
to fix Aµ . It’s more a “super gauge choice” to fix the extraneous components in the
superfield. In Wess-Zumino gauge, the superfield takes the simpler form
1
VW Z = θσ µ θ̄ Aµ + θ2 θ̄λ̄ + θ̄2 θλ + θ2 θ̄2 D (4.4)
2
It contains a gauge field Aµ , a Weyl fermion λα and an extra real scalar D that, as
the top component of a superfield, will prove to be auxiliary. If we quantise Aµ and λ
then we find the single-particle excitations of the gauge multiplet that we anticipated
in Section 2.3.2.

If you act with a supersymmetry transformation on VW Z , then it will take you out
of Wess-Zumino gauge. This isn’t a big headache; it just means that you have to do a
compensating transformation to put yourself back in Wess-Zumino gauge afterwards.
The supersymmetry transformations then act on the fields Aµ , λ and D as

δAµ = ϵσµ λ̄ + λσµ ϵ̄

δλ = ϵD + (σ µν ϵ)Fµν (4.5)
µ µ
δD = iϵσ ∂µ λ̄ − i∂µ λσ̄ ϵ̄

Note that the supersymmetry transformations (3.15) alone give us a term proportional
to ∂µ Aν in δλ. The compensating gauge transformation to take us back into Wess-
Zumino gauge adds another term so this becomes the gauge invariant field strength

Fµν = ∂µ Aν − ∂ν Aµ

Finally, note that

1
VW2 Z = θ2 θ̄2 Aµ Aµ and VW3 Z = 0 (4.6)
2
This will be useful when constructing supersymmetric actions shortly.

4.1.1 The Field Strength and Action

We will build the action out of a field strength superfield, constructed from V by
1
Wα = − D̄2 Dα V
4

– 99 –
This has some nice properties. First, it is a chiral superfield, obeying D̄α̇ Wα = 0. This
follows from the fact that D̄3 = 0. Second, it is invariant under the superfield gauge
symmetry (4.2): the Ω† term is killed immediately by Dα Ω† = 0, while the two D̄’s
contrive to kill the Ω term. (You need one D̄ to get past the Dα and the other D̄ to
kill Ω.) The upshot is that any action formed from Wα will be automatically gauge
invariant.

Next, we compute the components of Wα . This is a straightforward calculation

but the number of terms involved gets rather large. Happily, things are easier if we
appreciate that Wα is a chiral superfield since this means we only need to worry about
the θ terms, with the θ̄ terms following automatically from the expansion (3.19). In
components, the field strength superfield reads

Wα (x, θ) = λα (x) + θα D(x) + (σ µν θα )Fµν (x) − iθ2 σαµα̇ ∂µ λ̄α̇ (x) + . . .

The first component of the chiral superfield Wα is a spinor, rather than a scalar, re-
flecting the fact that Wα is itself a spinor chiral superfield. Importantly, Wα contains
the field strength Fµν .

Since Wα is chiral, we can integrate it over half of superspace to get a supersymmetric

action. We have
Z
1 i
d2 θ W α Wα = − Fµν F µν + Fµν ⋆ F µν − 2iλσ µ ∂µ λ̄ + D2
2 2
where the second term involves the dual field strength

⋆ 1
F µν = ϵµνρσ Fρσ
2
This is like Fµν but with the electric and magnetic fields swapped (one of them with a
minus sign).

The term iFµν ⋆ F µν is imaginary and so, at first glance, it looks like it will cancel
when we add the hermitian conjugate d2 θ̄ Wα̇† W † α̇ . However, it turns out that this
R

term plays an important role (at least this is true in the non-Abelian theories that we
will discuss shortly) and we wish to keep it. This is achieved by introducing the gauge
coupling constant e2 . Because this coupling constant sits in an F -term it is necessarily
complex. We define
ϑ 4πi
τ= + 2
2π e

– 100 –
And then write the Lagrangian
Z Z 
4 2 iτ α
SMaxwell = − d x dθ W Wα + h.c.
16π
Z  
4 1 µν ϑ ⋆ µν i µ 1 2
= d x − 2 Fµν F + Fµν F − 2 λσ ∂µ λ̄ + 2 D (4.7)
4e 32π 2 e 2e
This is the supersymmetric Maxwell action. The propagating degrees of freedom are
the U (1) gauge field and a fermion λ that, in this context, is called the gaugino or,
more specifically, the photino. There is also a real, auxiliary field D.

The parameter e2 is the coupling constant. It doesn’t do anything in Maxwell theory,

which is free, but will come into play when we add matter. Note that we’re working
in a convention where there is a factor of 1/e2 that sits in front of the Maxwell action.
As we’ll see, the gauge coupling doesn’t then sit anywhere else. This differs from the
convention that we first met in Quantum Field Theory where the Maxwell term was
canonically normalised but there was a gauge coupling inside the covariant derivatives.
The two conventions are related by a rescaling Aµ → eAµ . Note that the photino λ
similarly has an unconventionally normalised kinetic term, with a 1/e2 .

Finally, there is the parameter ϑ. This is known as the theta angle. (We’ve used
calligraphic script ϑ to distinguish it from the superspace coordinate θ.) Classically,
the theta angle doesn’t do anything. This is because it multiplies a total derivative
⋆
Fµν F µν = 2∂µ (ϵµνρσ Aν ∂ρ Aσ )
However, things are more interesting in the quantum theory and the addition of such
topological terms in the path integral can affect the dynamics. This is rather subtle
for Maxwell theory, but underlies the story of 3d topological insulators. The effect is
more pronounced in Yang-Mills theory and we’ll discuss it further in Section 6. You
can read (a lot) more about the theta angle in the lectures on Gauge Theory.

4.1.2 Supersymmetric QED

Next we add matter. This comes in the form of chiral multiplets Φi , where i = 1, . . . , N .
We want these to be charged under the U (1) gauge field so that under a gauge trans-
formation
Aµ → Aµ + ∂µ α
The components of the chiral multiplet transform with charges qi ∈ Z. This means
that the lowest components transform as
ϕi → eiαqi ϕi

– 101 –
By necessity, the fermions ψi and auxiliary fields Fi in the chiral multiplet Φi must
have the same charge,

ψi → eiαqi ψi and Fi → eiαqi Fi

From (4.3), this gauge transformation sits within a larger superfield transformation,
under which

Φi → exp (−2iqi Ω) Φi

This, however, means that the canonical Kähler potential that we’ve used so far is not
gauge invariant:
N
X N
X
Φ†i Φi exp −2iqi (Ω − Ω† ) Φi Φ†i


→
i=1 i=1

However, it’s simple to fix up. We simply need to use the new Kähler potential
N
X
K(Φi , Φ†i , V )= Φ†i e2qi V Φi
i=1

with the transformation of V given in (4.2) rendering the whole expression gauge in-
variant. In Wess-Zumino gauge, the formulae (4.6) truncates at e2qV = 1 + 2qV + q 2 V 2 .
Integrating over superspace then gives
Z
4 † 2qV
Z
4
h
2 µ 2
√ †

2
i
d θ Φ e Φ = d x |Dµ ϕ| − iψ̄σ̄ Dµ ψ + |F | − 2q ϕλ̄ψ̄ + ϕ λψ + qD|ϕ|

Here the covariant derivatives are given by

Dµ ϕ = ∂µ ϕ − iqAµ ϕ and Dµ ψ = ∂µ ψ − iqAµ ψ

The full action for an Abelian gauge theory then comes from combining the Maxwell
action (4.7) with the matter fields. It is
N Z
X
S = SMaxwell + d4 x d4 θ Φ†i e2qi V Φi
i=1
Z N
4
h 1 ϑ i X
− 2 Fµν F µν + ⋆ µν µ 2 µ


= dx F µν F − λσ ∂ µ λ̄ + |Dµ ϕi | − i ψ̄σ̄ Dµ ψ i
4e 32π 2 e2 i=1
N
1 X √   i
+ 2 D2 + |Fi |2 − 2qi ϕi λ̄ψ̄i + ϕ†i λψi + qi D|ϕi |2 (4.8)
2e i=1

– 102 –
The first line contains the kinetic terms, the second the interactions. Note that there
is a Yukawa coupling between the gaugino λ and the chiral multiplet fields, with ϕ†
partnering ψ so that the Yukawa term is gauge invariant. In addition, there is a scalar
potential that arises when we integrate out the auxiliary fields. The F terms don’t do
anything unless we also add a superpotential, while integrating out the D term results
in the potential
N
!
1 2 X
V (ϕ) = 2 D with D = e2 qi |ϕi |2 (4.9)
2e i=1

Provided that there are both positive and negative charges qi (and there must be as we
explain below) then the potential has flat directions in which
N
X
qi |ϕi |2 = 0 (4.10)
i=1

The existence of a moduli space of vacua is an important feature of supersymmetric

gauge theories. We will study it more closely in Section 4.3

A First Look at the Anomaly

There’s nothing wrong with (4.8) as a classical theory. But, as a quantum theory, it
has a problem. It turns out that for most choices of the charges qi , the quantum theory
is sick. It has an inconsistency that goes by the name of a gauge anomaly.

We will have a lot to say about anomalies, gauge and otherwise, later in these lectures.
For now we simply mention that the quantum theory only makes sense if the charges
satisfy the following two conditions
N
X N
X
qi = qi3 = 0 (4.11)
i=1 i=1

These conditions are not special to supersymmetric theories. They hold for any theory
that has Weyl fermions coupled to a U (1) gauge group. We’ll say more about where
these conditions come from in Section 5.2. For now, note that they require us to have
fields with both positive and negative charges which, in turn, ensures that there are
solutions to (4.10) with ϕi ̸= 0.

There are non-trivial solutions to the consistency conditions (4.11) but, for the most
part, we will work with trivial solutions in which chiral multiplets come in pairs so
that for each Φ with charge q there is a second chiral multiplet that we call Φ̃ with

– 103 –
charge −q. The conditions (4.11) are then automatically satisfied. Each pair Φ and Φ̃
is sometimes referred to as a flavour. If a flavour is said to have charge q, it means that
Φ has charge q and Φ̃ charge −q.

The simplest example comprises of a U (1) gauge field interacting with N flavours
(which means 2N chiral multiplets) of charge +1. This theory is known as supersym-
metric QED, or SQED for short. The action is
N Z
X  
SSQED = SMaxwell + 4
dxdθ4
Φ†i e2i V Φi + Φ̃†i e−2i V Φ̃i
i=1
Z
4
h 1 ϑ i
= dx − 2
Fµν F µν + 2
Fµν ⋆ F µν − 2 λσ µ ∂µ λ̄
4e 32π e
N 
|Dµ ϕi | + |Dµ ϕ̃i | − iψ̄σ̄ Dµ ψi − iψ̃¯i σ̄ µ Dµ ψ̃i
X 
2 2 µ
+
i=1
N  N
!2
√ X  e2 X i
− 2 ϕ†i λψi − ϕ̃†i λψ̃i + h.c. − |ϕi |2 − |ϕ̃i |2 (4.12)
i=1
2 i=1

where we’ve integrated out both D-term and F -terms so the scalar potential takes the
form (4.9).

When we first met QED in the lectures on Quantum Field Theory, we coupled a
Dirac fermion to a U (1) gauge field. This Dirac fermion contains two chiral fermions,
one left-handed ψ and one right-handed χ̄, both with the same charge. If we conjugate
the right-handed fermion then it becomes a left-handed fermion χ. We now have two
left-handed fermions with equal and opposite charges. That’s precisely the fermionic
matter content in each flavour in (4.12).

Adding Further Terms

There are further terms that we can add to the action (4.12) (or, indeed, to the more
general action (4.8)). We can add any superpotential W (Φ) provided that it is gauge
invariant. For example, we can always add to (4.12) the superpotential
N
X
W (Φ, Φ̃) = mi Φ̃i Φi
i=1

This gives a mass |mi | to each chiral multiplet. In particular, the fermions get a Dirac
mass. Note that such mass terms are only possible if there are pairs of chiral superfields
with opposite charges.

– 104 –
 There is one further, slightly curious term that we can add. This is known as the
Fayet-Iliopoulos term,
Z
LFI = d4 θ 2ζV = ζD (4.13)

It is gauge invariant because D doesn’t shift under the generalised gauge symmetry
(4.2). Here ζ ∈ R is the Fayet-Ilipoulos, or FI, parameter. Since this multiplies the
D-term, it changes only the scalar potential (4.9) which becomes

N
!2
e2 X
V (ϕ) = qi |ϕi |2 − ζ
2 i=1

In particular, supersymmetric vacua with V (ϕ) = 0 now require some scalar field to get
a non-vanishing expectation value which, in turn, breaks the U (1) gauge symmetry.

4.2 Non-Abelian Gauge Theories

We can repeat everything above for non-Abelian gauge fields. We work with a gauge
group G with Lie algebra

[T A , T B ] = if ABC T C

The factor of i in the commutation relations ensures that the generators are Hermi-
tian, so (T A )† = T A . We normalise the generators in the fundamental (i.e. minimal)
representation as
1
Tr T A T B = δ AB (4.14)
2
In what follows, generators T A will always be taken to be in the fundamental represen-
tation. If we need generators in other representations R then we will denote them as
TRA . In these lectures we will mostly work with

G = SU (Nc )

with the subscript on Nc short for the number of “colours”. We’ll also mention results
for other gauge groups as we go and, for now, keep things general.

4.2.1 Super Yang-Mills

Constructing supersymmetric Yang-Mills theory is slightly more fiddly version of what
we did for Maxwell theory. We introduce a real superfield V in the adjoint of the gauge

– 105 –
group. As usual, we can view an object in the adjoint representation as living in the
Lie algebra by writing

V = V AT A A = 1, . . . , dim G

For G = SU (Nc ), if we take T A to be in the fundamental representation then this

means that V is an Nc × Nc matrix. In terms of the components, we have a gauge field,
but this is now accompanied by a fermion λ and auxiliary field D, both of which must
also sit in the adjoint representation. Equivalently, all of them naturally live in the Lie
algebra

Aµ = AA
µT
A
, λα = λA
αT
a
, D = DA T A

Again, for SU (Nc ) this means that each of these should be thought of as an Nc × Nc
matrix (in addition to any vector or spinor index they carry). The fermion is again
called a gaugino or sometimes a gluino.

We again want to generalise the usual non-Abelian gauge symmetry to something

that can act on a superfield. We do this by taking an adjoint valued chiral superfield

Ω = ΩA T A

Since Ω is in the Lie algebra, eiΩ ∈ G and this acts on the real superfield as
†
e2V → e−2iΩ e2V e2iΩ
1
From the Baker-Cambell-Hausdorff formula, eX eY = eX+Y + 2 [X,Y ]+... , we get the trans-
formation law for the superfield itself

V → V + i(Ω − Ω† ) − i[V, Ω + Ω† ] + . . .

We can use the shift that appears in the first term to once again go to Wess-Zumino
gauge where the real superfield takes the form (4.4), now with all fields in the adjoint
of G. You can check that the remaining gauge symmetry acts on Aµ in the usual way,

Aµ → U Aµ U −1 + iU ∂µ U −1

with U ∈ G. The field strength lives in a chiral multiplet, defined as

1
Wα = − D̄2 e−2V Dα e2V


8

– 106 –
Evaluated in Wess-Zumino gauge, we use the fact that V 3 = 0, as in (4.6), to expand
e2V = 1 + 2V + 2V 2 . A short calculation then shows that
1
Wα (y, θ) = − D̄2 (Dα V − [V, Dα V ])
4
= λα (y) + θα D(y) + (σ µν θ)α Fµν (y) − iθ2 σαµβ̇ Dµ λ̄β̇ (y)

with the non-Abelian field strength and covariant derivative defined by

Fµν = ∂µ Aν − ∂ν Aµ − i[Aµ , Aν ] and Dµ λ = ∂µ − i[Aµ , λ]

To construct the action, we again define the complexified gauge coupling

ϑ 4πi
τ= + 2 (4.15)
2π g
The action is then given by
Z Z 
4 2 iτ α
SSYM = − d x Tr dθ W Wα + h.c.
8π
Z
4
h 1 µν ϑ ⋆ µν 2i µ 1 2i
= d x Tr − 2 Fµν F + Fµν F − 2 λσ Dµ λ̄ + 2 D (4.16)
2g 16π 2 g g
This is super Yang-Mills. After all that work, it’s actually a very simple theory: just
Yang-Mills coupled to a single, adjoint Weyl fermion. The factor of 2 differences com-
pared to the Maxwell action (4.7) can be traced to the normalisation convention (4.14).

4.2.2 Supersymmetric QCD

We can add matter transforming in any representation R of the gauge group. The
matter sits, as always, in a chiral superfield Φ that now transforms as

Φ → exp −2iΩA TRA Φ

(4.17)

We construct a gauge invariant, supersymmetric action with the superfield expression

Z
d4 x d4 θ Φ† e2V Φ = Dµ ϕ† Dµ ϕ − iψ̄σ̄ µ Dµ ψ + F † F
√
− 2(ψ̄λA TRA ϕ + ϕ† λA TRA ψ) + ϕ† DA TRA ϕ

Here the covariant derivatives include the gauge field transforming in the appropriate
representation R.

– 107 –
 Again, various anomaly cancellation conditions must be satisfied when coupling Weyl
fermions to non-Abelian gauge groups in complex representations. The simplest way
forward is to work instead with Dirac fermions. This means that we take pairs of
chiral superfields, Φ transforming in some representation R and Φ̃ in the conjugate
representation R̄. (In much of the literature, these superfields are denoted Q and Q̃
but we’ll stick with Φ and Φ̃ to avoid any unnecessary confusion with the supercharges.)

The most common is to take R to be the fundamental representation. We could, for

example, consider G = SU (Nc ) gauge group with Nf flavours of fermions, each in the
fundamental representation. The action is then
Z h 1 ϑ 2i µ i
SSQCD = d4 x Tr − 2 Fµν F µν + F µν
⋆ µν
F − λσ D µ λ̄
2g 16π 2 g2
Nf h
¯
X i
2 2 µ µ
+ |Dµ ϕi | + |Dµ ϕ̃| − iψ̄i σ̄ Dµ ψi − iψ̃i σ Dµ ψ̃i
i=1
Nf h
√ X
ϕ†i λψi − ϕ̃i λ̄ψ̃¯i + h.c. − V (ϕ, ϕ̃)
i
− 2 (4.18)
i=1

Here the covariant derivatives are

Dµ ϕ = ∂µ ϕ − iAµ ϕ and Dµ ψ = ∂µ ψ − iAµ ψ

for the fields in the fundamental representation, and

Dµ ϕ̃ = ∂µ ϕ̃ + iϕ̃Aµ and Dµ ψ̃ = ∂µ ψ̃ + iψ̃Aµ

for those in the anti-fundamental representation. Finally, the scalar potential is again
given by the D-terms
Nf 
1 X 
V (ϕ, ϕ̃) = 2 DA DA with D = −g A 2
ϕ†i T A ϕi − ϕ̃i T A ϕ̃†i (4.19)
2g i=1

with T A the Nc × Nc generators in the fundamental representation. This is the action

of supersymmetric QCD, or SQCD for short. In a nod to the real world, we refer to the
fermions ψ and ψ̃ as quarks. Their supersymmetric scalar partners ϕ and ϕ̃ are called
squarks.

Once again, we can also add masses for the quark multiplets by including the gauge
invariant superpotential
Nf
X
W(Φ, Φ̃) = mi Φ̃i Φi
i=1

– 108 –
This gives an extra term to the scalar potential
Nf  
X
δLmass = − |mi |2 |ϕi |2 + |ϕ̃i |2
i=1

as well as Dirac masses for ψi and ψ̃i .

There is no FI parameter that we can add for non-Abelian theories. The non-Abelian
analog of (4.13) would involve Tr D but the trace of the generators of any non-Abelian
Lie algebra always vanishes. Fayet-Iliopoulos terms can only be introduced for U (1)
gauge theories.

4.3 The Moduli Space of Vacua

In the absence of a superpotential, supersymmetric gauge theories do not have a unique
ground state. Instead, the D-term potential has a flat direction with V (ϕ) = 0. This
is the moduli space of vacua. It will turn out that this moduli space holds the key
to understanding the quantum dynamics of supersymmetric gauge theories. For this
reason, we will spend some time studying its structure.

Consider, for example, U (1) SQED with a single flavour. If we don’t turn on a FI
parameter then the D-term is (4.12)

D = −g 2 (|ϕ|2 − |ϕ̃|2 )

Clearly any solution with

|ϕ|2 = |ϕ̃|2 = v 2

has zero energy. To fully specify the classical theory, we must decide where on this
moduli space we want to sit.

At all points on the moduli space, there are always massless particles. Indeed, the
low-energy physics is dominated by the fluctuations along the moduli space, which
always correspond to massless particles, together with their fermionic superpartners.

Meanwhile, the masses of heavy particles typically depend on where you sit on the
moduli space which, in the current example, means that value of v 2 . Because ϕ is
charged under the U (1) gauge field, when it gets an expectation value, the Higgs mech-
anism kicks in and the photon gets a mass of order

m2γ ∼ e2 v 2

– 109 –
But the Yukawa terms in (4.12) mean that a particular combination of fermions also
gets a mass, given by

mfermion ∼ ev

The fact that this is the same as mγ is, of course, no coincidence: the photon, massive
fermion and an additional massive scalar in the spectrum form a massive vector multi-
plet of the kind discussed in Section 2.3. The origin of the moduli space, at ϕ = ϕ̃ = 0,
is special because here the vector multiplet becomes massless.

The Geometry of Moduli Space

We denote the moduli space of vacua as M. As we now explain, this manifold naturally
comes with a number of interesting geometric structures.

First M is defined by the requirement that V (ϕ) = 0. In the absence of a superpo-

tential, this is equivalent to D(ϕ) = 0. (Note that here ϕ denotes all chiral multiplet
scalars and, for SQED and SQCD, this means both ϕ and ϕ̃.). However, we should
also remember that the gauge group G acts on these scalars. The gauge symmetry is
not really a symmetry of the theory, but rather a redundancy in our description. This
means that any two values of ϕ related by a gauge transformation should be viewed
as physically equivalent. The upshot is that the vacuum moduli space M is defined as
the quotient

M = {ϕ | D(ϕ) = 0}/G (4.20)

We have stumbled upon a construction known to mathematicians as the symplectic

reduction. It’s particularly natural because, as we’ve seen above, the D-term constraint
D(ϕ) = 0 is fully specified by the action of the group G. In this way, the group G gets
to act twice: once as a constraint, and again as a quotient. Mathematicians call the
constraint D(ϕ) = 0 the moment map. If G includes an Abelian factor, the associated
FI parameter is known as the level.

There are two, further ways to describe the moduli space M. We will now describe
these, but won’t prove the equivalence with (4.20). Instead, we will content ourselves
with some heuristic justification, followed by some examples2 .
2
A full proof can be found in the paper by Marcus Luty and Wati Taylor, Varieties of vacua in
classical supersymmetric gauge theories.

– 110 –
 The fact that the group G “acts twice”, is even more apparent if the second way of
writing the moduli space: it is the holomorphic quotient

M = { ϕ }/GC (4.21)

with GC the complexified gauge group. This means that we take the real parameters
α that usually specify a gauge transformation – that is ϕ → eiqα ϕ for Abelian G or
a a
ϕ → eiα TR ϕ for non-Abelian – and quotient by transformations with α ∈ C. You
should think of the D-term constraint in (4.20) as like a gauge-fixing condition for the
non-Hermitian part of the GC transformations.

In fact, looking back at our construction of supersymmetric gauge theories, the gauge
transformations started life in a chiral superfield Ω where everything was complex.
They became real only after moving to Wess-Zumino gauge. From the perspective
of supersymmetric gauge theory, the equivalence of (4.20) and (4.21) is best seen by
looking at the more general gauge transformations before imposing Wess-Zumino gauge.

The final description of the moduli space will, in some circumstances, turn out to be
the most useful. The manifold M can alternatively be viewed as

M = {Gauge invariant, holomorphic monomials} / {Algebraic relations} (4.22)

This is a description of M in terms of what mathematicians call an algebraic variety.

This definition is best elucidated by examples that we will turn to below, but here we
give the basic gist.

There are three key ideas that we need to explain in this definition: gauge invariant,
holomorphic, and the algebraic relations. We cover each in turn:
• Because gauge symmetry is merely a redundancy in our choice of description,
it should be possible to describe the dynamics of massless particles in terms of
some gauge invariant fields. This is the basic idea underlying the characterisation
(4.22)

• It’s always possible to build such gauge invariant fields by taking combinations
like ϕ† ϕ. These are invariant under G, but not invariant under the larger GC
that defines the moduli space according to (4.21). The need to impose invariance
under GC , or equivalently the need to impose the D-term constraint D = 0, means
that we should work with holomorphic gauge invariant combinations, meaning
monomials that involve ϕ alone and not ϕ† . Alternatively, and more physically,
supersymmetry means that we should be able to describe the fields in terms of
chiral multiplets, and these are necessarily holomorphic.

– 111 –
 • Finally, it will turn out that, for some examples, not all of the gauge invariant
combinations are independent. This is why there is the need to quotient by certain
relations between them. This is best illustrated when we turn to examples below.

Mathematically, the equivalence between the quotient constructions (4.20) and (4.21)
and the algebraic description (4.22) goes by the name of geometric invariant theory.

4.3.1 The Moduli Space of SQED

We’ll start by looking at the simpler case of SQED. This is a U (1) gauge theory coupled
to N flavours. If we set the FI parameter to zero for now, then the D-term condition
is (4.12)
N
X
|ϕi |2 − |ϕ̃i |2 = 0 (4.23)
i=1

In addition, we should quotient by the U (1) gauge action

ϕi → eiβ ϕi and ϕ̃i → e−iβ ϕ̃i (4.24)

We started with 2N fields ϕ and ϕ̃. There is one real constraint (4.23) which, together
with the quotient (4.24) reduces the complex dimension of the vacuum moduli space
by one. We then have

dim M = 2N − 1 (4.25)

Let’s see how to reproduce this counting when thinking of M as an algebraic variety
defined by (4.22). The gauge invariant monomial are the bilinears

Mj i = ϕ̃j ϕi (4.26)

We will refer to these, not entirely accurately, as “mesons”. There are N 2 such fields
and, at first glance, it looks like we have way too many. However, they are not all
independent and this is where the algebraic relations in (4.22) come into play.

The meson matrix M is built from vectors ϕ and ϕ̃ and so has, at most, rank 1. This
means that there are N − 1 eigenvalues that are guaranteed to vanish. In general, the
determinant of an N × N matrix A can be written as

ϵi1 ...iN Ai1j1 . . . AiNjN = detA ϵj1 ...jN

The rank 1 matrix M therefore obeys

ϵi1 ...iN (Mj1i1 − λδj1i1 ) . . . (MjNiN − λδjNiN ) = det(M − λ) ϵj1 ...jN = λN −1 (λ − λ0 )ϵj1 ...jN

– 112 –
This tells us that if we expand out the left-hand side, all terms of order λN −2 and lower
must vanish for a rank 1 matrix. In other words, we have the constraints

ϵi1 ...iN Mj1i1 Mj2i2 = 0 (4.27)

with all other constraints following by contracting with further Mj i . Our next task is
to count how many independent constraints we have here. The i3 , . . . , iN indices are
left hanging so by picking these we can restrict i1 and i2 to run over any pair. But
the resulting constraints aren’t all independent. For example, there is a constraint
that arises from (i1 , i2 ) = (1, 2) and another that arises from (i1 , i2 ) = (1, 3). But
dividing the first constraint by the second, and rearranging, gives the constraint that
arise from (i1 , i2 ) = (2, 3). In fact, it’s not hard to convince yourself that the constraints
that come from (i1 , i2 ) = (1, anything but 1) are independent and sufficient to give all
others. Clearly there are N − 1 of these.

For each of these constraints, we still have the (j1 , j2 ) indices hanging. These too
are anti-symmetrised and the same argument that we gave above for (i1 , i2 ) also holds
for (j1 , j2 ). This means that the total number of constraints from (4.27) is (N − 1)2 .
The algebraic variety M, defined by all mesons (4.26) subject to the constraints (4.27)
then has complex dimension

dim M = N 2 − (N − 1)2 = 2N − 1

in agreement with our earlier counting (4.25).

The Metric on the Vacuum Moduli Space

The vacuum moduli space inherits a natural metric. Indeed, if we restrict to very low
energies the dynamics is that of the massless fields, corresponding to fluctuations along
the moduli space. This is the realm of the non-linear sigma model that we discussed
in Section 3.2.4. On general grounds, we know that not only is there a metric on M
but this metric must be Kähler.

It is straightforward to compute this metric. Here we do it in two different ways for

the simplest case of N = 1 flavour. The easiest way to proceed is to start with the
Kähler potential

K = ϕ† ϕ + ϕ̃† ϕ̃

Note that the Kähler potential for a gauge theory involves terms like e2qV , with V the
real superfield, to ensure gauge invariance. We simply set the gauge fields to zero in the

– 113 –
following calculation, so the Kähler potential is the canonical one above. Restricting to
the moduli space (4.23), we have |ϕ|2 = |ϕ̃|2 . Furthermore, if we work with the meson
field M = ϕ̃ϕ, the Kähler potential becomes
√
K = 2|ϕ|2 = 2 M † M (4.28)

The associated metric is just

|dM |2
ds2 = (4.29)
2|M |

We see immediately that the metric is singular at the origin M = 0. This singularity
is telling us something important: when ϕ = ϕ̃ = 0, there are new massless degrees of
freedom. This is simply the photon and its superpartner which become massless at the
origin because the Higgs mechanism turns off.

This is a lesson that we’ve seen before. When we integrated out heavy fields in Section
3.3, we found that the low-energy effective theory had singularities at points where the
heavy fields became light. This is a general feature of low-energy effective theories,
and one that will be important in Section 6 when we come to discuss the quantum
dynamics of these theories. For now, the lesson is worth repeating one more time:
singularities in the low-energy effective action signal the emergence of new, massless
degrees of freedom.

There is a more prosaic way to do this same calculation that highlights our original
quotient description of the vacuum moduli space (4.20). The general solution to the
constraint (4.23) is

ϕ = veiα eiβ and ϕ̃ = veiα e−iβ

with v > 0. The e±iβ has been taken to coincide with the gauge action (4.24), so that
v and α provide the coordinates on the moduli space M.

At this point, there’s an important factor of 2 that we have to take care of. The
parameter β corresponding to the U (1) gauge transformation has range β ∈ [0, 2π).
In contrast, we have α ∈ [0, π). This follows because we can always implement a
gauge transformation with β = π which flips the sign of ϕ and ϕ̃ or, equivalently, takes
α → α + π.

– 114 –
 The metric on M is inherited from the kinetic terms for the scalar fields. To this
end, we promote v, α and β to fields that vary slowly over spacetime. The covariant
derivatives are

Dµ ϕ = (∂µ v + iv(∂µ α + ∂µ β − Aµ )) ei(α+β)

Dµ ϕ̃ = (∂µ v + iv(∂µ α − ∂µ β + Aµ )) ei(α−β)

We now choose Aµ = ∂µ β to absorb the variation of β. This how the quotient in (4.20)
manifests itself in this calculation. The kinetic terms for the scalar fields, restricted to
the vacuum moduli space, then become
h i
Leff = |Dϕ|2 + |Dϕ̃|2 = 2 ∂v 2 + v 2 ∂α2 (4.30)

which we interpret as a metric like the non-linear sigma models (3.25) we discussed
earlier. It’s straightforward to check that this coincides with the metric (4.29) written
in terms of the meson field.

At first glance, (4.30) looks like a flat metric. And, indeed, it is. But it’s not the flat
metric on C because the angular coordinate α doesn’t have periodicity 2π. Instead, it’s
the flat metric on C/Z2 and has a conical singularity at the origin v = 0. This how we
see the emergence of the massless photon at this point.

Turning on the FI Parameter

A small variation on this calculation provides yet another perspective on the importance
of singularities in the low-energy effective action. We again consider SQED with N = 1
flavour, but this time turn on a FI parameter. The D-term constraint now reads

|ϕ|2 − |ϕ̃|2 = ζ (4.31)

We assume that ζ ≥ 0. In the ground state, we necessarily have |ϕ|2 ̸= 0 meaning that
the photon now gets a mass on all points of the moduli space.

We can see how this manifests itself in the moduli space metric. The condition (4.31)
is solved by
p
ϕ = v 2 + ζ eiα eiβ and ϕ̃ = veiα e−iβ

Our previous calculation to compute the metric on M is now a little more involved.
The subtlety lies in figuring out what expression we should take for the gauge field Aµ .
The answer can be found in its equation of motion. Or, more precisely, the equation of

– 115 –
Figure 5. The moduli space of SQED. When ζ = 0, the moduli space is the singular cone
C/Z2 shown on the left. The singularity at the origin reflects the existence of the massless
photon. When ζ ̸= 0 the singularity is resolved and the moduli space is the smooth cone
shown on the right. Now the photon is Higgsed everywhere on the moduli space.

motion in the limit e2 → ∞ where we neglect the Maxwell term. This is the appropriate
limit when the gauge field responds immediately to fluctuations in the scalar and gives
ζ
Aµ = ∂µ α + ∂µ β
2v 2+ζ
It reduces to our previous, pure gauge, choice when ζ = 0. Inserting this expression
into the kinetic terms for ϕ and ϕ̃, we compute the metric on the vacuum moduli space
2v 2 + ζ 4v 2 (v 2 + ζ)2 2
 
2 2 2
Leff = |Dϕ| + |Dϕ̃| = 2 ∂v + ∂α (4.32)
v +ζ (2v 2 + ζ)2
Importantly, as we approach the origin, v 2 → 0, the metric is well approximated by

ds2 ≈ dv 2 + 4v 2 dα2 = dv 2 + v 2 d(2α)2

That extra factor of 2 makes all the difference! We now get the flat metric with the
angular coordinate 2α ∈ [0, 2π) which means that close to v = 0 the metric really does
look like flat space. The resulting moduli space is sketched in Figure 5.

4.3.2 The Moduli Space of SQCD

We now play the same game for SQCD. We will take gauge group

G = SU (Nc )

coupled to Nf fundamental flavours, ϕia in the fundamental representation and ϕ̃ai in the
anti-fundamental. Here a = 1, . . . , Nc labels is the gauge group index while i = 1, . . . Nf
is the flavour index.

– 116 –
 The generators (T A )ab in the fundamental representation are the set of Hermitian,
traceless, complex Nc ×Nc matrices. Meanwhile, the generators in the anti-fundamental
representation are simply T̄ A = −T A . The Nc2 − 1 D-term conditions (4.19) are then
ϕ†i T A ϕi − ϕ̃i T A ϕ̃† i = 0 A = 1, . . . Nc2 − 1
where there is an implicit sum over i = 1, . . . , Nf . To get a better sense of these
constraints, let us first relax the requirement that T A is traceless. (This is what we
would get if the gauge group was U (Nc ) rather than SU (Nc ).) In this case, the T A
provide a basis for all Hermitian matrices and the D-term condition is Nc2 constraints
ϕ†i a ϕib − ϕ̃ai ϕ̃†b i = 0 a, b = 1, . . . Nc for U (Nc )
But the fact that we’re working with SU (Nc ) rather than U (Nc ) means that there’s no
reason to set the trace to zero. So our true D-term constraint is
1  †c i 
ϕ†i a ϕib − ϕ̃ai ϕ̃†b i = ϕi ϕc − ϕ̃ci ϕ̃†c i δ ab (4.33)
Nc
At first glance, this looks like it’s still Nc2 conditions. But if you take the trace then
you find that both sides are trivially equal. This means that, in fact, it’s only Nc2 − 1
conditions, with no condition on the trace. This is what we wanted.

To understand the vacuum moduli space, we must first solve the equations (4.33).
As we will now see, the nature of the solutions is different for Nf < Nc and Nf ≥ Nc .
We deal with each in turn.

Nf < Nc
We’d like to count the dimension of the moduli space M, defined by (4.33) modulo
gauge transformations. It’s tempting to think that there are just Nc2 − 1 constraints
in (4.33) but how do we know that they are all independent? In fact, it’s simple to
see that these constraints cannot all be independent when Nf < Nc because then we
would have more constraints than degrees of freedom. Yet solutions to (4.33) certainly
exist! To proceed, we use the fact that the D-terms and gauge symmetry are closely
entwined. The D-terms only bite when the gauge symmetry does.

When Nf < Nc , we can always use an SU (Nc ) gauge transformations and SU (Nf )
flavour rotations to put the matrix ϕ in the block-diagonal form
 
v1 . . . 0
 . 
 . . 
i
ϕa =  (4.34)
 

 0 . . . vN 
 f 
0 ... 0

– 117 –
Here the columns have length Nc and the rows length Nf . We can then use the other
SU (Nf ) to rotate ϕ̃ to be in upper-diagonal form. (We can’t make it fully diagonal
because we’ve already used up the SU (Nc ) to diagonalise ϕ). However, now we invoke
the D-term conditions (4.33). The only solutions to these conditions require that the
off-diagonal terms in ϕ̃ vanish. (You could check this for a simple case, say Nc = 3 and
Nf = 2 to get a feel for why this is the case.) We’re left with

ϕ̃†a i = ϕia

As before, points on the moduli space related by a gauge transformation are to be

physically identified. On a generic point on the moduli space (with vi ̸= vj ̸= 0 when
i ̸= j) the gauge group is broken to

SU (Nc ) → SU (Nc − Nf )

The number of broken gauge generators is then

# broken generators = (Nc2 − 1) − ((Nc − Nf )2 − 1)

Each of these is eaten by one of the original 2Nc Nf bosons ϕ and ϕ̃. This means that
the resulting vacuum moduli space has complex dimension

dim M = 2Nc Nf − [# broken generators] = Nf2

Note that we only divide out by the points on the moduli space related by the SU (Nc )
gauge symmetry. There will still be points on the moduli space related by the flavour
symmetry SU (Nf ) but these are physically distinct vacua.

We can also view the moduli space as an algebraic variety. Once again, the holomor-
phic monomials are the meson fields

Mj i = ϕ̃aj ϕia (4.35)

This time the name “meson” is more appropriate: we have contracted the gauge indices
of ϕ and ϕ̃ to form a gauge invariant composite. The mesons form Nf2 fields but, in
contrast to SQED, there is no constraint on M . The contracted gauge indices in (4.35)
run over a = 1, . . . , Nc > Nf so there is no obstacle to M being maximal rank. We see
immediately that dim M = Nf2 , in agreement with our result above.

– 118 –
 We can compute the metric on M along the same lines as we saw for SQED. The
Kähler potential is

K = ϕ†i a ϕia + ϕ̃ai ϕ̃†a i

We want to write this in terms of the meson field (4.35). To do this, first note that
for Nf < Nc the trace term on the right-hand side of the D-term (4.33) vanishes when
restricted to the moduli space and we have

ϕ†i a ϕib = ϕ̃ai ϕ̃†b i (4.36)

From this, we have

(M † M )j i = ϕ̃†a i ϕ†k a ϕkb ϕ̃bj = (ϕ̃†a i ϕ̃ak )(ϕ̃†b k ϕ̃bj )

where, in the last equality, we’ve√used (4.36). Taking the square root of this matrix
equation tells us that (ϕ̃† ϕ̃)ij = ( M † M )ij , and so the Kähler potential is
√
K = 2 Tr M † M (4.37)

Just like the Kähler potential for SQED (4.28), the resulting metric will have singulari-
ties whenever M −1 ceases to exist. Again, these singularities correspond to new degrees
of freedom becoming massless. At a generic point on the moduli space, there will be
massless gauge bosons associated to the unbroken SU (Nc − Nf ) gauge symmetry. But
along the loci on which M is not invertible we have an enhancement of the gauge group
and new massless gauge bosons.

Nf ≥ Nc
For Nf ≥ Nc , the story is different. First, we can now use SU (Nc ) and SU (Nf ) trans-
formations to find solutions to the D-term equations (4.33), again in block-diagonal
form
   
v1 . . . 0 0 ṽ1 . . . 0 0
 .
.. .. 
 and ϕ̃†a i =  . . .
 .. 
ϕia = 
 .   .
0 . . . vNc 0 0 . . . ṽNc 0

with

|va |2 = |ṽa |2 + ρ a = 1, . . . , Nc

where ρ must be independent of a. This reflects the fact that the trace term on the
right-hand side of (4.33) can now be non-zero.

– 119 –
 At a generic point on M, the SU (Nc ) gauge symmetry is completely broken. The
complex dimension of the moduli space is therefore
dim M = 2Nc Nf − (Nc2 − 1) (4.38)
How can we describe this moduli space as an algebraic variety? The meson fields (4.35)
provide Nf2 degrees of freedom, but now there are constraints of the kind we met for
SQED since M is at most rank Nc . In addition, there are also new gauge invariant
fields. These are baryons, built from the totally anti-symmetric invariant tensor of
SU (Nc ),
B i1 ...iNc = ϕia11 . . . ϕiaNNcc ϵa1 ...aNc
a
B̃i1 ...iNc = ϕ̃ai11 . . . ϕ̃iNNcc ϵa1 ...aNc
Each of these is anti-symmetric in the Nc different flavour indices i1 , . . . , iNc . There are
then a bunch of further constraints between these baryons and mesons. Rather than
doing this in full generality, we’ll instead just describe how this works for the two cases
that will prove most interesting in Section 6.
• Nf = Nc : In this case, anti-symmetry properties mean that there is just a single
baryon of each type
a
B = ϕ1a1 . . . ϕN
aNc ϵ
c a1 ...aNc
and B̃ = ϕ̃a11 . . . ϕ̃NNcc ϵa1 ...aNc
The meson M can have rank Nf , so there are no constraints there. But there is
a single relation between the mesons and baryons, given by
B̃B = det M (4.39)
This means that there are Nf2 + 2 degrees of freedom in M , B and B̃ and a single
relation, giving a moduli space of dimension dim M = Nf2 + 1 in agreement with
(4.38). The relation (4.39) will play a starring role when we come to consider the
quantum theory in Section 6.3.

• Nf = Nc + 1: Now there are Nf baryons of each type,

Bj = ϵji1 ...iNc B i1 ...iNc and B̃ j = ϵji1 ...iNc B̃i1 ...iNc
This time the constraints are less obvious, but they turn out to be
Adj(M )i j = B i B̃j and Mj i B j = Mj i B̃i = 0 (4.40)
where Adj(M ) is the adjugate matrix, which is the transpose of the matrix of
cofactors. The adjugate matrix is most familiar when M is invertible, in which
case Adj(M ) = (det M )M −1 . However, the conditions BM = M B̃ = 0 tell us
that M has a zero eigenvalue and so is not invertible.

– 120 –
 At this point, things start to get a little messy! It turns out that not all the
relations (4.40) are independent, but there’s no way to write them as a smaller set.
Mathematicians say that the resulting variety is not a complete intersection. We’ll
simply duck the issue which, it turns out, will not hinder us from understanding
the physics.

There is one sense in which the use of the words “mesons” and “baryons” might be
misleading. In QCD, mesons and baryons are bound states of quarks, stuck together
because of confinement. But confinement is a surprising and poorly understood prop-
erty of the quantum theory. Here we are not invoking anything so dramatic. Indeed,
we haven’t yet discussed any quantum effects and what we’ve call SQCD might better
be called SCCD for our current purposes. Instead, we’re using meson and baryon fields
simply because they are gauge invariant and so free of any gauge redundancy. We’ll
turn on the Q in SQCD in Section 6 where we’ll see how this tallies with ideas of
confinement.

4.3.3 Briefly, Gauged Linear Sigma Models in 2d

We’ve learned that we can construct interesting geometric spaces as the moduli spaces
of vacua of supersymmetric gauge theories. This kind of construction goes by the name
of gauged linear sigma models. It turns out that it’s a particularly useful method when
wielded in quantum field theories in d = 1 + 1 dimensions.

To see why, first consider the action for a non-linear sigma model in general d-
dimensional spacetime
Z
S = dd x gij (π) ∂µ π i ∂ µ π j (4.41)

Here π i are coordinates on a manifold M with metric gij .

When d = 0 + 1, we’re dealing with the quantum mechanics of particle moving on

M. But we know what happens in this case: the wavefunction will spread over M and
there will typically be a unique ground state.

This is conceptually very different from what happens in d = 3 + 1 dimensions.

There, each point on M defines a different ground state of the system. There is no
spread of the wavefunction.

The reason for this different behaviour can be traced to the long-distance property
of the propagator. The propagator grows in d = 0 + 1 and d = 1 + 1 dimensions
(logarithmically in the latter case) while it decays in d = 2 + 1 and higher. This fact

– 121 –
is closely related to the Mermin-Wagner theorem which says that global symmetries
cannot be spontaneously broken in d = 0 + 1 and d = 1 + 1 dimensions. (We met this
theorem in the lectures on Statistical Field Theory and Gauge Theory.)

In the context of non-linear sigma models of the type (4.41), this long-distance be-
haviour of the propagator is telling us that d = 0+1 and d = 1+1 dimensions are special
because the wavefunction spreads over the manifold M. This means that the ground
state of the system has a chance of knowing something about the global structure of
the manifold M, like its topology. Indeed, studying the dynamics of low-dimensional
quantum systems on M has been a very fruitful source of developments in mathemat-
ics. This beginnings of this story are told in the lectures on Supersymmetric Quantum
Mechanics.

The story is particularly rich for theories in d = 1 + 1 dimensions where, in addition

to the wavefunction spreading over M, the UV divergences of the quantum field theory
mean that the metric on M is renormalised. At one-loop, the running is captured by
the beautifully geometric RG equation
∂gij
µ = Rij (4.42)
∂µ
where µ is the RG scale and Rij the Ricci tensor. This formula is known as Ricci
flow. It plays an important role in String Theory and has a number of applications in
pure mathematics. Note that the flow stops only if the metric becomes Ricci flat, with
Rij = 0. At this point we have a 2d conformal field theory. However, not all manifolds
admit such a Ricci flat metric.

Things become even more interesting when we throw supersymmetry into the mix.
This is what we called N = (2, 2) supersymmetry in Section 2.4.3. It not only gives
us an important level of control over the dynamics but, as we’ve seen already in these
lectures, dovetails nicely with some interesting mathematical structures. It turns out
that the gauge theory approach to realising non-linear sigma models as the vacuum
moduli space is particularly powerful in this context. Here we just give a hint of how
this works

First, the anomaly cancellation conditions (4.11) are for 4d quantum field theories
and are not needed in two dimensions. (A 4d Weyl fermion reduces to a 2d Dirac
fermion and so the theories we construct are not chiral in 2d.) This means that there
is nothing to stop us considering U (1) coupled to N chiral multiplets of charge +1 in

– 122 –
d = 1 + 1 dimensions. The D-term condition is
N
X
|ϕi |2 = ζ
i=1

where we turn on a FI parameter ζ > 0. Taken on its own, this condition defines a
sphere S2N −1 . But we still have to quotient by the U (1) action to get the vacuum
moduli space and this gives

M = S2N −1 /U (1) = CPN −1

Here CPN −1 is complex projective space, defined as the space of complex lines in CN .
This can also be seen in the definition (4.21) of the moduli space.

Things get more interesting if we add, in addition, a chiral superfield P with charge
−q. The D-term condition is now
N
X
D= |ϕi |2 − q|p|2 − ζ = 0
i=1

After quotienting by the U (1) action, the vacuum moduli space is a non-compact man-
ifold. But we now have the option of introducing a gauge invariant superpotential

W (P, Φ) = P G(Φ1 , . . . , ΦN )

with G a homogeneous polynomial of degree q. The potential energy now also includes
contributions from the F-terms
N 2
X ∂G
VF = |p|2 + |G|2
i=1
∂ϕi

If we choose G to be transverse, meaning

∂G
= 0 ∀i ⇔ ϕi = 0
∂ϕi

then VF = 0 only if p = 0 which means that we’re back onto the CPN −1 vacuum
manifold. But now, in addition, we must satisfy G(ϕ) = 0. The resulting vacuum
moduli space is now a compact manifold given by a degree q hypersurface, M ⊂ CPN −1 .

– 123 –
 To give a sense of why the gauge theory description is useful in understanding the
geometric properties of the vacuum manifold, here’s a short anecdote. It turns out that
the gauge theory flows to a conformal field theory only when q = N . (Only then does
the FI parameter not run.) In this case, the vacuum moduli space X is a degree N
hypersurface CPN −1 . But it is known that such spaces defines what mathematicians
call a Calabi-Yau manifold. One of the key properties of these spaces (conjectured by
Calab and proven by Yau) is that they admit a Ricci flat metric. This ties in nicely
with the gauge theory expectation because, as we have seen in (4.42), such a Ricci flat
metric is necessary for conformal symmetry.

There are many more geometrical properties that can be extracted from a study of
gauge theories in 2d dimensions, including mirror symmetry of Calabi-Yau manifolds3 .

4.4 Extended Supersymmetry

We discussed the representations of extended supersymmetry algebras in Section 2.4.
For theories with N = 2 supersymmetry (or eight supercharges) there are two different
multiplets:
N = 2 vector multiplet = N = 1 vector multiplet (Aµ , λα , D)
+ N = 1 chiral multiplet (ϕ, χα , F )
Here the chiral multiplet necessarily sits in the adjoint representation of the gauge
group. There is also the N = 2 matter multiplet
N = 2 hypermultiplet = N = 1 chiral multiplet (q, ψα , F )
+ N = 1 chiral multiplet (q̃, ϕ̃α , F̃ )
If the first of these transforms in the representation R of the gauge group then the
second transforms in the conjugate representation R̄. We can tune the matter content
and interactions of N = 1 theories to give theories with extended supersymmetry.

With N = 4 there is just a single multiplet (at least restricting to non-gravitational

theories) with content
N = 4 vector multiplet = N = 1 vector multiplet (Aµ , λ1α , D)
+ 3 × N = 1 chiral multiplets (ϕi , λi+1 i
α ,F ) i =, 1, 2, 3
In addition to the gauge field, we have three complex scalars and four Weyl fermions,
all sitting in the adjoint representation of the gauge group.
3
The use of gauge theories as a method to understand geometry was pioneered by Edward Witten
in the paper Phases of N = 2 Theories. You can read more in Kentaro Hori’s lecture notes which
comprise Part 2 and Part 3 of the book Mirror Symmetry.

– 124 –
 To construct theories with N = 2 and N = 4 supersymmetry, we could try to build
an extended superspace. It turns out that there is a superspace for N = 2 theories,
known as harmonic superspace, but it’s rather cumbersome to work with. In contrast,
there is no superspace for N = 4 theories. Instead, we will build Lagrangians for both
by tuning the interactions of N = 1 theories. The key is to get Lagrangians that exhibit
larger R-symmetries.

4.4.1 N = 2 Theories
N = 2 super Yang-Mills comprises of a vector multiplet V and an adjoint chiral multi-
plet Φ. The N = 2 Lagrangian is constructed by simply turning off any superpotential
for Φ. It is
Z  Z
iτ α 1
L = −Tr dθ2
W Wα + h.c. + 2 d4 θ Φ† e2V Φ
8π g
 
2 1 µν µ µ † µ ϑ
= 2 Tr − Fµν F − iλσ Dµ λ̄ − iχσ Dµ χ + Dµ ϕ D ϕ + Tr Fµν ⋆ F µν
g 4 16π 2
2 h√ √ 1 i
+ 2 Tr 2iλ[ϕ† , χ] + 2iλ̄[ϕ, χ̄] − [ϕ† , ϕ]2 (4.43)
g 2
The potential term comes from integrating out the D-term from the N = 1 vector
multiplet: we’ll look more closely at the moduli space of vacua below.

Of more immediate importance are the fermion terms: the two Weyl fermions λ
and χ sit on the same footing in the final Lagrangian, despite their origins in different
N = 1 multiplets. This means that there is an SU (2) symmetry that rotates them,
under which they sit in a doublet 2. The bosonic field ϕ does not transform under
this symmetry, which tells us that this must be an SU (2)R R-symmetry. This is the
smoking gun for N = 2 supersymmetry. There is also a U (1)R symmetry, under which
R[ϕ] = 2 and R[λ] = R[χ] = 1.

There is another way to derive the N = 2 Lagrangian. You can write down a minimal
super Yang-Mills theory in d = 5 + 1 dimensions, consisting of a gauge field coupled to
a Weyl fermion. Upon dimensional reduction, this gives the Lagrangian (4.43).

We can couple matter to (4.43) in the form of hypermultiplets. These comprise of

two chiral multiplet, Q and Q̃. (Note: until now the letter Q has always meant a
supercharge, but it’s not unusual to also use it to denote a chiral multiplet, with Q
standing for “quark”.) As we mentioned above, if Q sits in the representation R then
Q̃ necessarily sits in the conjugate representation R̄. This suffices to determine the

– 125 –
interaction with the vector multiplet V ,
Z h i
Lvector = d4 θ Q† e2V Q + Q̃† e−2V Q̃

But in addition we should couple Q and Q̃ to the N = 2 vector multiplet field Φ in

such a way that the SU (2)R symmetry between λ and χ remains. This is achieved by
the superpotential term
√ Z 2
Lchiral = 2 d θ Q̃ΦQ + h.c.

The interactions between Q̃ and Q themselves are greatly limited by the extended
supersymmetry: we can add only mass terms
√
W = 2mQ̃Q

A general N = 2 theory is specified by the gauge group G and the representations Ri of

any matter multiplets, together with their masses. (If G contains Abelian factors, we
can also add FI terms. We will not include these in the following.) The scalar potential
comes, as always, from integrating out D and F-terms. After some rearranging, the
potential can be expressed as the sum of positive definite terms. For SU (Nc ), it is
dim G
!2 dim 2
1 g2 X X † A XG X
† 2 A † 2 A
V (ϕ, q, q̃) = 2 Tr[ϕ , ϕ] + qi TR qi − q̃i TR q̃i +g q̃i TR qi
g 2 A=1 i A=1 i
X †
+ qi {ϕ† − m†i , ϕ − mi }qi + q̃i {ϕ† − m†i , ϕ − mi }q̃i† (4.44)
i

(Initially, the D-term contains both ϕ and the q’s and q̃’s. The first two terms on the
first line both arise from this D-term, but the cross-term has sneaked into the third
line, where it turns ϕ† ϕ into the anti-commutator {ϕ† , ϕ}.)

The hypermultiplet scalars q and q̃ † transform as a doublet 2 under the SU (2)R sym-
metry. Conversely, their fermionic superpartners ψ and ψ̃ are singlets under SU (2)R .
The second and third terms in the potential (4.44) can be rewritten in way that makes
the SU (2)R symmetry manifest. We introduce the doublet
!
qi
ωi =
q̃i†

– 126 –
The second term in (4.44) is a real D-term while the third is a complex F-term. But,
with N = 2 supersymmetry they are better viewed as a potential V = g12 D ⃗ 2 arising
from triplet of D-terms
X
⃗ A = g2
D ωi† TRA⃗σ ωi
i

⃗ transforms in the 3 of SU (2)R .

where ⃗σ are the Pauli matrices. The triplet D

The potential (4.44) has some interesting properties. Let’s take the masses to vanish:
mi = 0. In this case, the second line takes the schematic form |ϕ|2 (|q|2 + |q̃|2 ). That
means that if we’re looking for vacuum states with V (ϕ, q, q̃) = 0 then there are two
possibilities: either ϕ = 0 and the hypermultiplet scalars q, q̃ are turned on; or q̃ = q = 0
and the vector multiplet scalar ϕ is turned on. Geometrically, this means that the
vacuum moduli space factorises as

M = MC × M H

There are defined as follows:

• MC is called the Coulomb branch. It is defined as the space q̃ = q = 0 with ϕ

restricted to obey

[ϕ† , ϕ] = 0

This is solved by ϕ sitting in the Cartan sub-algebra. For G = SU (Nc ). this

P
means that ϕ = diag(ϕ1 , . . . , ϕNc ) with a ϕa = 0. At a typical point, the gauge
group is broken to the Cartan subalgebra with a bunch of surviving, massless
photons. For example, for G = SU (Nc ), this means G → U (1)Nc −1 . At some
special points, the surviving gauge group will be enhanced further.

When the gauge group is broken to U (1)’s, all charged matter experiences a
Coulomb force, hence the name of this branch of vacua.

• MH is called the Higgs branch. It is defined as the space ϕ = 0 with q̃ and q

constrained to obey the conditions

⃗A = 0
D

In addition, we should quotient by the action of G. At a general point, the gauge

group is completely Higgsed, hence the name of this branch of vacua.

– 127 –
 The Higgs branch has real dimension that is a multiple of four and is a special
case of a Kähler manifold, known as a hyperKähler manifold. (For what it’s
worth, a hyperKähler manifold has three independent complex structures while a
Kähler manifold has just one.) The definition of the Higgs branch is an extension
of the idea of symplectic reduction that gives a hyperKähler metric and is known
as the hyperKähler quotient construction.

4.4.2 N = 4 Theories
The more supersymmetry we have, the more restrictive the theory.

With N = 1 supersymmetry, we are free to specify the gauge group and (chiral) mat-
ter content. In addition to the gauge coupling and masses, both suitably complexified,
we can also introduce any superpotential interactions that we wish.

With N = 2 supersymmetry, we are again free to specify the gauge group and (now
non-chiral) matter content. But we have no freedom in the choice of interactions: the
only arbitrary parameters are the gauge coupling and masses.

With N = 4 supersymmetry, we get to specify only the gauge group and gauge
coupling. All other terms in the Lagrangian are then dictated by supersymmetry.

There are a number of different ways to construct N = 4 super Yang-Mills. It can

be viewed as minimal super Yang-Mills in d = 9 + 1 dimensions, dimensionally reduced
to d = 3 + 1. Alternatively, it can be viewed as an N = 2 theory with a single adjoint
hypermultiplet. The theory contains four adjoint Weyl fermions, transforming in the
4 of SU (4)R R-symmetry and six real scalars φi with i = 1, . . . , 6, transforming in the
6. The scalar potential is
X
V (φ) = −g 2 [φi , φj ]2
i<j

There is now just a Coulomb branch, with G broken to the Cartan subalgebra at a
generic point.

– 128 –
5 Boot Camp: Quantum Gauge Dynamics
Our ultimate aim in these lectures is to understand the quantum dynamics of supersym-
metric gauge theories. But before we can appreciate this, we really need to understand
something about the quantum dynamics of ordinary gauge theories. The purpose of
this section is to provide the necessary background.

I should warn you that, in contrast to the rest of these lecture notes, we won’t
attempt to prove any of the statements made in this section. Indeed, some of them –
like the phenomenon of confinement – can’t currently be proven, although we do have
overwhelming evidence that it takes place, both from numerics and from toy models,
not least supersymmetric theories. (Not to mention experimental results like the fact
that you are literally stuck together by confinement.) Other phenomena – like the
one-loop beta function and the anomaly – have some technical calculations underlying
them. Here we omit the technicalities and just state the relevant facts, meaning that
you can relax and enjoy this section as something akin to the middle eight in a song. If
you want to see the gory details that underlie these results then they can all be found
in the lectures on Gauge Theory.

5.1 Strong Coupling

Our interest throughout this section will be on non-Abelian gauge theories. We start
with Yang-Mills. The Lagrangian is
Z
1
LY M = d4 x − 2 Tr Fµν F µν (5.1)
2g
Here the field strength is given by Fµν = ∂µ Aν − ∂ν Aµ − i[Aµ , Aν ]. As you can see,
we work with the convention in which the coupling constant sits in front of the kinetic
term.

5.1.1 The Beta Function

The key feature of Yang-Mills which makes it both subtle and hard is that the coupling
g 2 runs under RG. At a scale µ the coupling is given by
1 1 b0 Λ2U V
= − log (5.2)
g 2 (µ) g02 (4π)2 µ2
where g02 is the coupling constant evaluated at the cut-off scale ΛU V . Here b0 is the
coefficient of the 1-loop beta function and, for pure Yang-Mills, is given by
11
b0 = I(adj)
6

– 129 –
 G SU (N ) Sp(N ) SO(N ) E6 E7 E8 F4 G2
I(adj) 2N 2(N + 1) N −2 4 3 1 3 4

Table 1. The quadratic Casimir I(adj) for all compact Lie groups.

This depends on a group theoretic factor I(adj), known as the quadratic Casimir. It
has another avatar as the Dynkin index in the adjoint representation. (Note that we’ve
defined I(R) with a factor of 2 difference from the Gauge Theory lecture notes.) The
quadratic Casimirs for the various compact Lie groups are shown in Table 1. In these
lectures, we will focus almost exclusively on gauge group G = SU (N ).

The running of the coupling constant is often summarised in terms of the one-loop
beta function
dg b0 3
β(g) ≡ µ =− g (5.3)
dµ (4π)2

whose solution gives the logarithmic behaviour (5.2).

The all-important feature of the beta function is the overall minus sign. This means
that the theory is weakly coupled at high energies, a phenomenon known as asymptotic
freedom. Conversely, it means that the theory is strongly coupled at low energies. It is
this low-energy physics that we would like to understand.

What do we mean by low and high energy here? Where’s the dividing line? The
answer to this can be found within the formula (5.2). This is because we can construct
a strong coupling scale

8π 2
 
Λ = µ exp − 2 (5.4)
b0 g (µ)

This has the property that dΛ/dµ = 0. In other words, it is an RG invariant. This is
the scale at which the Yang-Mills theory becomes strong.

There’s already something remarkable about the existence of the scale Λ. Classically,
the Yang-Mills theory (5.1) has no dimensionful parameter. That means that there is
nothing to set a scale. Instead, there is just a dimensionless coupling constant g 2 . But
the logarithmic running succeeds in turning this into a dimensionful parameter Λ! One
way to see this is to note that to define the quantum theory, we necessarily had a

– 130 –
dimensionful parameter lurking all along. This is the UV cut-off of the theory, ΛU V .
The strong coupling scale (5.4) is related to the UV cut-off by
2 /b g 2
Λ = ΛU V e−8π 0 0

This means that if the bare coupling is small, g0 ≪ 1, as it should be then the physical
scale Λ is exponentially suppressed relative to the UV cut-off: Λ ≪ ΛU V .

5.1.2 Confinement and the Mass Gap

When the coupling is small, quantum field theories look similar to their classical coun-
terparts. For example, classical Maxwell theory provides a decent guide to what you
might expect from QED. In contrast, when the coupling is large, all bets are off. The
quantum theory and classical theory may be completely different. Yang-Mills provides
the archetypal example.

If you solve the classical Yang-Mills equations, you will find waves that propagate at
the speed of light. This suggests that the quantum theory will give rise to a massless
particle called a gluon, similar to the photon. Indeed, if you stare at the action there
is no A2µ term that might suggest a mass.

Nonetheless, we now know that quantum Yang-Mills contains no massless particles.

We say that the theory is gapped which means that the first excited state has a finite
energy above the ground state. This additional energy is, of course, just E = mc2
where m is the mass of the lightest particle in the theory. The gap is of order the
strong coupling scale, m ∼ Λ.

We don’t currently have the technology to prove the Yang-Mills mass gap. Indeed,
it is generally considered one of the most important and challenging open problems in
mathematical physics. We do, however, have very compelling numerical evidence that
this occurs, together with some intuition built from various toy models and heuristic
explanations for why it occurs. You can read about some of these in the lectures on
Gauge Theory. We’ll meet others later in these lectures.

In our world, the strong force is governed by an SU (3) gauge theory known as QCD.
The associated strong coupling scale is Λ ≈ 300 MeV and is usually referred to as
ΛQCD . No massless gluons are seen in Nature, but there is good evidence for states
known as glueballs with masses around the scale Λ.

– 131 –
 The existence of a mass gap goes hand in hand with another phenomenon: this
is confinement. To explain this, consider placing two charged test particles in the
Yang-Mills field. To be specific, we’ll consider G = SU (N ) and take a quark in the
fundamental representation N and an anti-quark in N̄. We simply ask: what force do
they feel?

It’s best to compute the potential energy between the two particles. You can first
do this in the classical theory. There’s a little bit of group theoretic fiddliness but the
final result is very intuitive: the potential energy scales with the separation r between
particles as

g2
V (r) ∼ (5.5)
r
This, of course, is the same scaling that we see in the Coulomb force of electromag-
netism.

What about the quantum theory? If the separation between particles is small, mean-
ing r ≪ 1/Λ, you don’t notice much difference. At these short distances the theory is
weakly coupled and we again see the Coulomb-like potential (5.5) between test parti-
cles. We should replace the coupling constant in (5.5) with g 2 (µ) = g 2 (1/r) so it’s more
accurate to say that the potential scales as V (r) ∼ log r/r but this is a mild correction
to the physics.

In contrast, at large separation things are radically different. For distances r ≫ 1/Λ,
the potential between test particles takes the form

V (r) ∼ σr (5.6)

The coefficient σ necessarily has dimension [σ] = 2 and this scale, like everything else
in Yang-Mills, is set by σ ∼ Λ2 . For reasons that we will explain shortly, σ is called
the string tension. The force law (5.6) is, to put it mildly, a dramatic departure from
what we’re used to. The potential energy now increases with separation. Indeed, it
costs an infinite amount of energy to pull the quark anti-quark pair to infinity. This
kind of potential energy is said to be confining.

The phenomenon of confinement is, like the mass gap, something that we can’t prove
from first principles. Once again, however, there is clear numerical evidence together
with a plethora of heuristic explanations.

– 132 –
Figure 6. A rough sketch of the non-Abelian field lines in the Coulomb phase, on the left,
and in the confining phase, on the right.

To get some very rough intuition for what’s going on, we can repeat Faraday’s old
experiment (now in thought only!) and try to understand what the field lines look
like. At short separation, in the Coulomb-like phase (5.5), the field lines form the
familiar pattern, first spreading out radially before they bend over to combine with
those emitted by the anti-particle. This is shown on the left-hand side of Figure 6.
However, as the particles are separated to larger distances, the fact that the gauge field
is massive makes itself known. The field lines no longer spread out, but instead lie
closely together to form a collimated flux tube. This flux tube acts very much like a
string, connecting the two quarks. If its tension, or energy per unit length, is σ then it
gives rise to a confining force law like (5.6).

The above description of confinement should be taken with something of a pinch

of salt. After all, we are in a strongly interacting quantum field theory and there is
no single field configuration that governs the physics. Instead, there are many fields
configurations that we should sum over that contribute to the path integral. The
discussion above should be understood to mean that those field configurations that
resemble the flux tube dominate.

The story above was told in terms of test particles. When we introduce dynamical
matter fields into the theory, one would naively expect the associated particles to bind
together like the test particles above. And, roughly speaking, this is indeed what
happens, at least if the number of light species is small enough. (We’ll flesh out this
statement shortly.) For example, in QCD the quarks bind together into mesons and
baryons. Mesons contain a quark anti-quark pair while baryons contain three quarks
and are a colour singlet by dint of the ϵabc invariant tensor. For G = SU (N ) we would
get mesons which again contain a quark anti-quark pair and baryons containing N
quarks.

– 133 –
 There is much more to say about confinement. In particular, the correct, mathemat-
ical description of the confining phase lies involves a non-local operator known as the
Wilson loop
 I 
W [C] = Tr P exp i A
C

Here C is a closed curve in spacetime, while P stands for “path ordering”. In a

Coulomb-like phase, the expectation value scales as ⟨W [C]⟩ ∼ exp(−L[C]) where L[C]
is the length of the perimeter of C. Meanwhile, in the confining phase the expectation
value scales as ⟨W [C]⟩ ∼ exp(−A[C]) where A[C] is the area spanned by the curve C.
An explanation of why this is the right diagnostic, together with its significance, can
be found in the lectures on Gauge Theory.

5.1.3 Adding Matter

Until now, we’ve considered pure Yang-Mills and its response to test particles. Now
we wish to add dynamical matter. The first thing that this does is change the beta
function.

Suppose that we have a bunch of Weyl fermions transforming in some representations

Rf and a bunch of scalars transforming in some representation Rs . Then the one-loop
beta function (5.2) becomes
11 2 X 1 X
b0 = I(adj) − I(Rf ) − I(Rs ) (5.7)
6 6 fermions 6 scalars

Here the group theoretical factors are Dynkin indices. For the representation R, the
Dynkin index I(R) is defined by the normalisation of the trace
1
Tr TRA TRB = I(R) δ AB (5.8)
2
Our previous normalisation (4.14) means that we’re taking the fundamental represen-
tation to have I(fund) = 1. Some examples of I(R) for SU (N ) representations are
collected in Table 2.

Strictly speaking, the beta function takes the form (5.7) only if the matter is massless.
If the matter has some mass m, then the beta function runs like (5.7) for energies µ > m,
but as we drop below the mass scale m the matter decouples and its contribution to
the one-loop beta function is removed.

– 134 –
 Irrep □ adj
1 1
dim N N2 − 1 2
N (N + 1) 2
N (N − 1)
I(R) 1 2N N +2 N −2
A(R) 1 0 N +4 N −4

Table 2. Some group theoretic properties of SU (N ) representations. Here is the symmet-

ric representation and the anti-symmetric. Conjugate representations have I(R̄) = I(R)
and A(R̄) = −A(R).

Again, the first thing to notice is the signs. Both fermions and scalars give a con-
tribution to the beta function that has the opposite sign to the gauge bosons. This
means that if we have too much matter then we will have b0 < 0 and, correspondingly,
β(g) > 0 and the theory will be weakly coupled in the infra-red. In this case, the quan-
tum theory looks very much like classical Yang-Mills at low energies, with massless
gauge bosons. Here we would like to understand what happens when b0 > 0 and the
theory is strongly coupled.

To illustrate this, we will consider a specific set of matter particles. We take

G = SU (Nc )

with Nf flavours of quarks in the fundamental representation. This means that we have
a i
a collection of left-handed Weyl spinors ψαi and ψ̃αa . Here a = 1, . . . , Nc is the gauge
index and i = 1, . . . , Nf the flavour index. We take ψ to transform in the fundamental
Nc representation and ψ̃ in anti-fundamental representation N̄c representation. (If we
take the complex conjugate of ψ̃, we get a Dirac spinor in the Nc representation.) The
action is
Nf
1
iψ̄i σ̄ µ Dµ ψi + iψ̃¯i σ̄ µ Dµ ψ̃ i
Xh i
LQCD = − 2 Tr Fµν F µν + (5.9)
2g i=1

with

Dµ ψ = ∂µ ψ − iAµ ψ and Dµ ψ̃ = ∂µ ψ̃ + iψ̃Aµ

We could add a mass for the quarks by introducing terms like

Nf
X
Lmass = mi ψ̃ i ψi
i=1

However, our interest will be on the case with massless quarks, with mi = 0.

– 135 –
 You might wonder why this is interesting. After all, the quarks in our world aren’t
massless. But they are almost massless! The up and down quarks have masses of a
few MeV, much less than the relevant scale ΛQCD ≈ 300 MeV. Meanwhile, the strange
quark has a mass mstrange ≈ 95 MeV, still smaller than ΛQCD although not by much.
This means that understanding the behaviour of massless QCD is not a bad starting
point for understanding the full theory.

5.1.4 Chiral Symmetry Breaking

The important observation is that massless QCD (5.9) has an extra symmetry that the
massive theory doesn’t have, under which the ψ and ψ̃ fermions rotate independently.
The global symmetry includes

GF = SU (Nf )L × SU (Nf )R (5.10)

Here SU (Nf )L acts on the ψ while SU (Nf )R acts on the ψ̃,

ψi → (L† )i j ψj and ψ̃ i → Ri j ψ̃ j (5.11)

with L ∈ SU (Nf )L and R ∈ SU (Nf )R . (In fact, the full symmetry of the classical
theory is U (Nf )L × U (Nf )R ; we’ll discuss these additional U (1) factors in Section 5.2.)
The group GF is known as the chiral symmetry, chiral because it acts on Weyl spinors
rather than Dirac spinors. This kind of symmetry only exists when the masses mi = 0.

The question that we want to ask is: what becomes of this chiral symmetry? The
answer to this depends on the number of flavours Nf in a way that is not fully un-
derstand. However, for suitably small Nf the theory develops a vacuum expectation
value

⟨ψ̃ i ψj ⟩ ∼ Λ3 δ i j

The formation of this condensate is a strong coupling effect and, like confinement,
poorly understood. In contrast, the consequence of the condensate is both well un-
derstood and dramatic. First, note that the condensate does not preserve the chiral
symmetry (5.11). Indeed, it transforms as

⟨ψ̃ i ψj ⟩ → Λ3 Ri k (L† )kj (5.12)

This is the phenomenon of chiral symmetry breaking, sometimes shortened to χSB. The
surviving subgroup requires us to set L = R in (5.11), meaning

SU (Nf )L × SU (Nf )R → SU (Nf )diag

– 136 –
The spontaneous breaking of chiral symmetry means that massless QCD actually has
a moduli space of vacua, since each choice of L ̸= R in (5.12) gives a different, equally
valid, ground state, albeit one that is entirely equivalent to the original because they
are related by a global symmetry. The vacuum moduli space is the coset

M = [SU (Nf )L × SU (Nf )R ] /SU (Nf )diag (5.13)

with dimension

dim M = Nf2 − 1

There is an important difference between this vacuum moduli space and those that
arise in supersymmetric theories. All points on M in QCD are equivalent because any
point is related to any other by the action of a symmetry. This is not the case for the
supersymmetric moduli space.

Nonetheless, there is one important feature that is common whenever we have flat
directions and this is the importance of massless particles, corresponding to fluctuations
along M. When the flat directions arise from broken symmetries, as in the present
case, these massless particles are Goldstone bosons.

We learn something interesting. Yang-Mills theory has a mass gap. But massless
QCD, at least for Nf > 1, does not. Even if the theory confines, giving massive
baryons and glueballs, chiral symmetry breaking means that there are massless Gold-
stone bosons. These can be identified with certain meson states called pions.

Of course, in our world the pions are not massless. But this is because the constituent
quarks are not exactly massless so the chiral symmetry is not exact. Nonetheless, the
chiral symmetry is an approximate symmetry which, in turn, means that the would-be
Goldstone bosons are light, but not exactly massless. Indeed, the pions are notably
lighter than all other hadrons in QCD.

5.1.5 Phases of Massless QCD

We’re now in a position to describe the different phases of massless QCD as we vary
Nc and Nf . There is much that we don’t yet understand (here “we” means everyone,
not just those following these lectures!) and there are a few subtleties that I will sweep
under the carpet. But, with broad brush, we we can sketch the different phases of the
theory.

– 137 –
 We start with low Nf :

• When Nf = 0, we have pure Yang-Mills. The theory sits in the confining phase,
with a mass gap.

• When Nf = 1, there is no chiral symmetry group (5.10) and so no chiral symmetry

breaking. The theory is again thought to have a mass gap, with quarks bound in
mesons and baryons.

• When 2 ≤ Nf ≤ N ⋆ the theory confines and exhibits chiral symmetry breaking.

This means that the low energy theory consists of freely interacting Goldstone
bosons, parameterising the moduli space (5.13).

The big question here is: what is the maximum value N ⋆ for which chiral sym-
metry breaking occurs? We don’t know the answer to this. Various approaches,
including numerics, suggest that it is somewhere around

N ⋆ ≈ 4Nc

Our lack of knowledge of this simple question highlights just how poorly we
understand strongly interacting field theories.

Now let’s jump to high values of Nf and we’ll then try to fill in the details in the
middle.

• When Nf ≥ 11 2
Nc , the beta function is positive. You can see this from the general
expression (5.7) which, for massless QCD, becomes
11 2
b0 = Nc − Nf (5.14)
3 3
This means that theory is weakly coupled in the infra-red: the low-energy physics
consists of massless gluons, weakly interacting with massless quarks. As we go to
smaller and smaller energies, the interactions become weaker and weaker. Strictly
speaking, in the far IR, the physics is free.
On the flip side, these become arbitrarily strongly coupled in the UV, with the
gauge coupling diverging at some very high scale. This doesn’t mean that we
should discard them, but they don’t make sense at arbitrarily high energies scales.
Said another way, we can’t take the UV cut-off ΛU V to infinity while keeping any
low-energy interactions. Nonetheless, it’s quite possible that these theories may
arise as the low-energy limit of some other theory. We will see examples in Section
6 when we discuss supersymmetric extensions of QCD.

– 138 –
Figure 7. The beta function for Nf slightly below the asymptotic freedom bound has a zero
which indicates the existence of an interacting conformal field theory.

That leaves us with the physics in the middle region. We’ll keep working down
from the asymptotic freedom bound 11Nc /2.

• When N ⋆⋆ < Nf < 11 2

Nc , things are more interesting. To see what happens, we
need the two-loop beta function
b0 3 b1 5
β(g) = − 2
g − g + ...
(4π) (4π)4

with the one-loop coefficient b0 given in (5.14) and the two-loop coefficient

34Nc2 Nf (Nc2 − 1) 10Nf Nc

b1 = − −
3 Nc 3
In the window of interest, b0 > 0 and b1 < 0, so we can play the one-loop
contribution against the two-loop contribution to find a zero of the beta function
b0
g⋆2 = −(4π)2
b1
with β(g⋆ ) = 0. The beta function is shown in Figure 7. The existence of such a
fixed point is telling us that we have an interacting conformal field theory: there
are massless modes, but they are no longer free in the infra-red. This is known
as the Banks-Zaks fixed point.
N
Importantly, when Nf lies just below the asymptotic freedom bound, so Nfc =
11
2
− ϵ, this fixed point lies at g⋆ ≪ 1 which means that we can trust the analysis
without having to worry about higher order corrections. Moreover, because g⋆ is
small we can use perturbation theory to calculate anything that we want.

– 139 –
Figure 8. The expected phases of massless QCD. The asymptotic freedom bound is Nf =
11
2 Nc . The lower edge of the conformal window is not known but is expected to be somewhere
around Nf ≈ 4Nc .

However, as Nf decreases, the value of the fixed point g⋆ increases until we can
no longer trust the analysis above. The expectation is that we get a conformal
field theory only for some range of Nf , lying within N ⋆⋆ < Nf < 11 2
Nc . This is
known as the conformal window. We don’t currently know the value of N ⋆⋆ .

That leaves us with understanding what happens in the middle when N ⋆ < Nf ≤
N ⋆⋆ . Our best guess is that there is no such regime, and the upper edge of the chiral
symmetry breaking phase coincides with the lower edge of the conformal window,

N ⋆⋆ = N ⋆

This guess is motivated partly by numerics and partly by a lack of any compelling
alternative. For us, the lesson to take away is that strongly interacting quantum field
theories are hard and even the most basic questions are beyond our current abilities.
A summary of the expected behaviour of massless QCD is shown in Figure 8.

5.2 Anomalies
The next topic that we need to cover is anomalies. This is a beautiful subject and, in
many ways, the place in which quantum field theory intersects most cleanly with topics
in mathematics. Here we won’t describe any of these mathematical underpinnings, but
instead just cover the minimum material necessary for our later applications.

The main idea is to understand how certain symmetries manifest themselves in quan-
tum field theory. To this, end consider a single left-handed Weyl fermion in d = 3 + 1
dimensions. The action is
Z
S = d4 x iψ̄σ̄ µ ∂µ ψ

– 140 –
This action is clearly invariant under the U (1) global symmetry ψ → eiαψ , with the
corresponding current j µ = ψ̄σ µ ψ. To illustrate the anomaly, we will couple this current
to a gauge field Aµ with charge q ∈ Z. The action is now
Z
S = d4 x iψ̄σ̄ µ Dµ ψ

where the covariant derivative contains the new coupling Dµ ψ = ∂µ ψ − iqAµ ψ. This is
action is now invariant under the gauge symmetry

ψ → eiqα(x) ψ and Aµ → Aµ + ∂µ α (5.15)

Before we proceed, I should mention that there are two distinct ways to think about
the gauge field Aµ and this distinction will be important when we come to look at the
various implications of anomalies. They are:

• Aµ could be a dynamical gauge field. In the classical theory, this means that we
treat it as a dynamical variable, with its own equation of motion, typically after
adding a Maxwell term to the action. In the quantum theory, it means that we
integrate over Aµ in the path integral.

• Aµ could be a background gauge field. This means that it is something fixed,

under our control, and should be viewed as a parameter of the theory. Turning
it on typically breaks Lorentz symmetry, but could be useful to explore how our
system responds to the presence of an electric or magnetic field. In the quantum
theory, Aµ appears as a source on which the partition function depends.

We will consider gauge fields of both types in what follows. However, for now, we will
consider Aµ to be a background gauge field, something that is under our control.

While the classical theory is clearly invariant under the gauge transformation (5.15),
the question that we really want to ask is: what about the quantum theory? For this,
we should turn to the path integral, with the partition function in Euclidean space
defined as
Z  Z 
4 µ
Z[A] = DψDψ̄ exp − d x iψ̄σ̄ Dµ ψ

Clearly the action in the exponent remains invariant under gauge transformations. But
now we must also worry about the measure in the path integral, and this takes some
care to define. The statement of the anomaly is that the measure is not invariant under

– 141 –
gauge transformations. Instead, it turns out that the measure, and hence the partition
function, changes by a phase
 3 Z 
iq 4 ⋆ µν
Z[A] → exp d x αFµν F Z[A] (5.16)
32π 2
with ⋆ F µν = 12 ϵµνρσ Fρσ . The purpose of this section is to understand the implication of
this calculation and a number of variants. As we now explain, there are three different
avatars of the anomaly. We deal with them each in turn.

5.2.1 Gauge Anomalies

The first implication of the anomaly (5.16) is that it is an obstruction to gauging.
Although the action is invariant under the gauge symmetry, the measure is not and
neither is the partition function. That means that we cannot promote the gauge field
Aµ to a dynamical field, where we integrate over it in the path integral. If we attempted
to do this, we would get a sick theory.

There are a number of ways to see why the theory is sick but here is a simple one.
Recall that when we first attempted to quantise the gauge field Aµ in the lectures on
Quantum Field Theory we had some work to do to decouple the negative norm states
that arise from quantising A0 . That work ultimately boiled down to using the gauge
invariance to remove these states. But in an anomalous theory, we no longer have
that gauge invariance at our disposal and the Hilbert space will involve negative norm
states. That’s bad.

The upshot is that a U (1) gauge theory, coupled to a single Weyl fermion, is not
consistent. To proceed, we must have multiple, left-handed Weyl fermions ψi , each
with some charge qi . (If we have right-handed fermions, simply conjugate them to
make them left-handed.) The phase in (5.16) is then proportional to the sum of qi3 .
The gauge theory is consistent only if
X
qi3 = 0 (5.17)
i

This was one of the conditions that we met previously in (4.11). This condition is
sometimes written in a different way. One, very simple way to solve this constraint is
to take pairs of Weyl fermions with charges ±q. If we conjugate one of them to become
a right-handed Weyl fermion, we then have a single Dirac fermion with charge q. These
are called vector-like theories and QED is the most familiar example.

There are, however, more interesting solutions to (5.17) that do involve ± pairs.
These are known as chiral gauge theories.

– 142 –
 The discussion above holds for an Abelian gauge symmetry. There is a similar story
for a non-Abelian gauge symmetry G. For a single Weyl fermion, transforming in the
representation R of G, the anomaly is proportional to the group theoretic factor A(R).
For the fundamental representation, A(R) = 1. For other representations, it is given
by

Tr TRA {TRB , TRC } = A(R) Tr T A {T B , T C }

Some examples of A(R) for SU (N ) representations are collected in Table 2. To be

consistent, a non-Abelian gauge theory coupled to a bunch of left-handed Weyl fermions
must obey
X
A(Ri ) = 0 (5.18)
i

which is the non-Abelian version of (5.17). If R is a complex representation, then it’s

simple to show that A(R̄) = −A(R). This means that we can again always satisfy (5.18)
by taking Dirac fermions, rather than Weyl fermions, since these have a left-handed
fermion in a representation R and another in R̄.

One consequence of the relation A(R̄) = −A(R) is that A(R) = 0 for any real
representation. This means that there is no obstacle to coupling a single Weyl fermion
in a real representation to a non-Abelian gauge group. Indeed, we’ve seen this already
in these lectures: pure super-Yang-Mills has a single adjoint Weyl fermion, but the
adjoint representation is real so there is no problem.

Relatedly, here’s a comment that will prove useful shortly: only massless fermions
contribute to the anomaly. If you have a Weyl fermion ψ in a complex representation
R of a group G, then to give it a mass preserving G you need a second Weyl fermion ψ̃
in representation R̄. You can then write down a Dirac mass term mψ̃ψ. But the two
Weyl fermions ψ̃ and ψ cancel in their contribution to the anomaly. Alternatively, you
can write down as Majorana mass mψψ for any fermion in a real representation of G
but, as we have seen, there is no contribution to the anomaly from fermions in a real
representation. This means that only fermions that cannot get a mass preserving G
contribute to the anomaly for G.

When we previously discussed the requirements of anomaly cancellation in (4.11),

we gave a further condition on U (1) gauge theories. We asked that they also satisfy
X
qi = 0 (5.19)
i

– 143 –
This, it turns out, is a little more subtle and it follows from the requirement that the
theory can be consistently coupled to gravity. There is no corresponding requirement
for non-Abelian gauge theories (essentially because Tr T A = 0 for any generator of a
simply connected Lie algebra).

The upshot is that if you want to have a theory with a dynamical gauge field, them
you better make sure that the anomaly (5.17) or (5.18) cancels. Furthermore, if you
want your theory to be compatible with gravity, then you have one further hoop (5.19)
to jump through.

5.2.2 Chiral (or ABJ) Anomalies

Here is a slight variant on the same calculation that leads to a physically very different
conclusion. Again, consider a single Weyl fermion, now coupled to a background non-
Abelian gauge field A in some representation R of the global symmetry G. It’s useful
to think of G = SU (N ), and R either the fundamental or adjoint representation. We
can construct the partition function
Z  Z 
4 µ
Z[A] = DψDψ̄ exp − d x iψ̄σ̄ Dµ ψ

now with Dµ ψ = ∂µ ψ − iAA A

µ TR ψ. We know that the partition function isn’t invariant
under gauge transformations of G. But here we instead ask a different question: is it
invariant under U (1) rotations of the fermion?
ψ → eiqα ψ (5.20)
The answer is again no, with the partition function transforming as
 Z 
iqI(R) 4 ⋆ µν
Z[A] → exp d x αTr Fµν F Z[A] (5.21)
16π 2
with I(R) the Dynkin index defined previously in (5.8). This looks very similar to our
previous result, but it should now be thought of a mixed anomaly between the U (1)
symmetry (5.20) and the non-Abelian symmetry G. This can be seen in the coefficient
qI(R) which is still cubic but now a mix of Abelian and non-Abelian generators.

An interesting consequence of this is that, in the presence of background gauge fields

for G, the U (1) symmetry is no longer conserved. If we repeat Noether’s theorem,
including the anomaly (5.21), we find that the U (1) current associated to the symmetry
(5.20) now obeys
qI(R)
∂µ j µ = Tr Fµν ⋆ F µν (5.22)
32π 2
When the right-hand side is non-zero, the current is no longer conserved.

– 144 –
 An important example of this occurs in the theory of massless QCD that we intro-
duced in the last section. The gauge group is G = SU (Nc ) and the Lagrangian is
(5.9),
Nf
1
iψ̄i σ̄ µ Dµ ψi + iψ̃¯i σ̄ µ Dµ ψ̃ i
Xh i
LQCD = − 2 Tr Fµν F µν + (5.23)
2g i=1

We have added extra fermions to cancel the gauge anomaly in G, as we should. But,
as we will see, a mixed anomaly of the type (5.21) remains.

Classically, the theory (5.23) has a U (Nf )L × U (Nf )R global symmetry, with each
factor rotating ψ and ψ̃ independently. We studied the SU (Nf )L × SU (Nf )R subgroup
in some detail in the previous section, but didn’t mention the two U (1) factors. These
are usually written as
U (1)B : ψi → eiβ ψi and ψ̃ i → e−iβ ψ̃ i
U (1)A : ψi → eiα ψi and ψ̃ i → eiα ψ̃ i (5.24)
The subscript B stands for “baryon” since this is the vector-like symmetry under which
baryons are charged. Since ψ and ψ̃ have opposite charges under U (1)B , there is no
obstacle to gauging it should we wish. Moreover, the ± charges also cancel on the
right-hand side of (5.22), and the U (1)B current is conserved in the quantum theory.

In contrast, the axial symmetry U (1)A has the same charges for ψ and ψ̃. This means
that the associated current is, following (5.22), no longer conserved. Instead, it obeys
Nf
∂µ jAµ = Tr Fµν ⋆ F µν (5.25)
16π 2
Note that the gauge fields on the right-hand side are now dynamical SU (Nc ) gauge
fields that fluctuate. There is now no way to set them to zero. There is no axial U (1)A
symmetry in the quantum theory.

This also explains why we didn’t include U (1)A when discussing chiral symmetry
breaking in the previous section. Since it is not a symmetry, there is no corresponding
Goldstone boson. (In the real world, the meson associated to U (1)A is called the η ′ and
is significantly heavier than the pion Goldstone bosons.)

This, then, is the second avatar of the anomaly. It manifests itself as a symmetry of
the classical theory that does not survive the quantisation procedure. In fact, this is
how the anomaly was first discovered. In this context, it usually goes by the name of the
chiral anomaly, or the ABJ anomaly after Adler, Bell and Jackiw who first uncovered
this subtle effect of quantum field theory. (Yes, that Bell.)

– 145 –
 There is one further way to think about the chiral anomaly. Non-Abelian gauge
theories have an additional, topological term
Z
ϑ
Sϑ = d4 x Tr Fµν ⋆ F µν
16π 2
This is the theta term. We already met it when constructing super Yang-Mills theory
in (4.16). Comparing with the form of the mixed anomaly (5.21), we see that axial
transformation (5.24) can be thought of as shifting the theta angle

U (1)A : ϑ → ϑ + 2α (5.26)

We’ve met this kind of idea previously in Section 3.3, where we found it useful to think
of parameters – supurions – transforming under symmetries (which, of course, means
that the symmetries aren’t actually symmetries). In Section 6, we’ll learn how we can
combine the shift of the ϑ angle with holomorphy in supersymmetric theories.

5.2.3 ’t Hooft Anomalies

So far we have discussed two manifestations of the anomaly:
• For a gauge symmetry, the anomaly better cancel. Or else.

• A mixed anomaly between a global symmetry and gauge symmetry means that
the global symmetry isn’t.
But what if we have an anomaly just for a global symmetry? What are the conse-
quences? From what we’ve discussed above, we know that the symmetry isn’t conserved
if we couple it to background gauge fields. But nothing compels us to do so. So what
else can we learn from this?

The answer is both subtle and powerful. An anomaly for a purely global symmetry
puts strong constraints on the low-energy dynamics of the theory. The anomaly should
be thought of as a robust way of characterising the theory, and this characterisation
cannot change under RG flow, now under any other deformation of the theory, provid-
ing that the symmetry remains unchanged. Such anomalies in global symmetries are
referred to as ’t Hooft anomaly.

We will first explain the basic idea and then give a concrete example. Suppose that
we have some quantum field theory – typically a non-Abelian gauge theory – that is
weakly coupled in the UV, but flows to strong coupling in the IR. We will abstractly
call the UV theory TU V . We assume that it has some global symmetry GF . This should
be a true symmetry of the quantum theory, meaning that it has no mixed anomalies
with the gauge symmetry.

– 146 –
 This UV theory may have an anomaly for GF . If GF is Abelian, anomaly is simply
P 3 P
q as in (5.17); if it is non-Abelian the anomaly is A(R) as in (5.18). Either way,
we will denote this anomaly as AU V and assume AU V ̸= 0,

The theory now flows under RG to a theory TIR in the IR which, as we’ve seen, will
typically be very different. We have the following result:

Claim: Either the symmetry GF is spontaneously broken, or the anomalies match

meaning
AU V = AIR
This is a wonderfully powerful result. If GF is spontaneously broken then we necessarily
have massless Goldstone bosons. But if GF is unbroken then we must have massless
fermions that reproduce the anomaly. This is known as ’t Hooft anomaly matching.

Proof: The argument for ’t Hooft anomaly matching is very slick. Suppose that
AU V ̸= 0 then we know from the discussion above that we’re not allowed to couple GF
to dynamical gauge fields. That would lead to a sick theory.

To proceed, we introduce a bunch of extra massless Weyl fermions transforming

under GF . We call these spectator fermions. These won’t interact directly with our
original fields in TU V , but they are designed so that the total anomaly of the original
fields and these new fermions vanishes:
AU V + Aspectator = 0
Now there’s nothing to stop us introducing dynamical gauge fields for GF . We do so,
but with a very very small coupling constant. We’ll see the importance of this shortly.

Now let’s go back to our original theory TU V . It will flow to strong coupling at some
scale Λ and we’d like to understand the physics TIR below this scale. If the gauge
coupling for GF is small enough, then this RG flow takes place entirely unaffected by
the presence of the GF gauge fields. This means that one of two things could have
happened. It may be that the strong coupling dynamics of TU V spontaneously breaks
the symmetry GF . (For example, as we’ve seen, this is expected to happen if we take
GF to be the chiral symmetry of QCD.) This was the first possibility of our claim.
Alternatively, GF may be unbroken at low-energies. In this case, we’re left with TIR ,
together with the spectator fermions, all coupled to the GF gauge fields. But this can
only be consistent if
AIR + Aspectator = 0

– 147 –
Clearly, this is only consistent if AIR = AU V . □

Triangle Diagrams
Until now, we’ve explained the anomaly as a transformation of the fermion measure in
the path integral. However, the anomalies also show up in perturbation theory when
computing corrections to Ward identities like (5.25). In this way of looking at things,
one has to compute so called triangle diagrams. Schematically, these take the form

X
Anomaly =
fermions

where you sum over all Weyl fermions running in loops. The outer legs are currents,
either gauge or global. The fact that there are three legs reflects the fact that the anoma-
lies are always proportional to the cube of generators. Our three kinds of anomalies
are related to the different types of currents on the legs
• Gauge3 : This is a gauge anomaly.

• Global × Gauge2 : This is the chiral anomaly.

• Global3 : This is the ’t Hooft anomaly.

An Application: Confinement Implies Chiral Symmetry Breaking

We saw in the last section that massless QCD exhibits two, distinct strong coupling
phenomena: confinement and chiral symmetry breaking. We will now show that they’re
not quite as unrelated as they first appear.

As we’ve seen, the U (1)A symmetry of massless QCD is anomalous. The true sym-
metry group is therefore

GF = U (1)B × SU (Nf )L × SU (Nf )R

Let’s first compute the ’t Hooft anomalies in the ultra-violet, where the quarks con-
tribute. There is no ’t Hooft anomaly for U (1)3B because this is a vector-like symmetry.
In contrast, there is a ’t Hooft anomaly associated to the chiral, SU (Nf ) factors. In
fact, there are two. The first is the purely non-Abelian anomaly
X
[SU (Nf )L ]3 : A = A(□) = −Nc

Here the anomaly A arises because each quark ψ carries a colour index a = 1, . . . , Nc .
The ψ fermions transform in the □ of SU (Nf )L and A(□) = −1. But there are Nc such

– 148 –
fermions. Hence the result Nc A(□) = −1. There is a similar anomaly for SU (Nf )R .
In addition, there is a mixed ’t Hooft anomaly between U (1)B and SU (Nf ). This is
X
[SU (Nf )L ]2 × U (1)B : A′ = qI(□) = Nc

which again simply counts the number of quarks.

Now the question is: what happens in the infra-red? For suitably low Nf , we’ve
already explained the chiral symmetry GF is expected to be broken down to U (1)B ×
SU (Nf )diag , but we didn’t give any justification for this. The idea of ’t Hooft anomaly
matching goes some way to help.

Here is the idea. We will assume that the theory confines and, moreover, that in the
infra-red, the physics is described by weakly interacting mesons and baryons. (This is
in contrast to the conformal field theories that we see at larger Nf .) In such a situation,
’t Hooft anomaly matching shows that the chiral symmetry must be broken.

Here is the argument. Suppose that GF is unbroken in the infra-red. Then they must
be massless fermions around that can reproduce the anomalies A and A′ . Moreover,
by assumption, these massless fermions must be bound states of quarks, either mesons
or baryons.

Mesons certainly can’t do the job because these are bosons. Baryons, meanwhile,
contain Nc quarks so these too are bosons when Nc is even. This is telling us that when
Nc is even, a confining theory contains no fermions at low-energies and so certainly can’t
reproduce the anomalies. We learn that chiral symmetry breaking must occur when
Nc is even.

What about Nc odd? Now baryons are fermions. Is it possible that some of these
baryons could be massless and reproduce the ’t Hooft anomalies? This time we have
something of a calculation to do. First, you have to figure out what representations
of GF the baryons sit in. Then you have to figure out what combination of massless
baryons could match the anomalies A and A′ . It takes some work, but the answer is
that the baryons can never reproduce the anomalies. (You can find the calculation in
Section 5.6 of the lectures on Gauge Theory.) This means that if QCD confines into
weakly interacting colour singlets, then chiral symmetry is necessarily broken.

5.3 Instantons
One of the new ingredients in these lectures is the Yang-Mills theta angle
Z
ϑ
Sϑ = d4 x Tr Fµν ⋆ F µν
16π 2

– 149 –
This deserves some explanation.

First, the theta term is a total derivative,

Z  
θ 4 µ µ µνρσ 2i
Sϑ = 2 d x ∂µ K with K = ϵ Tr Aν ∂ρ Aσ − Aν Aρ Aσ
8π 3

This means that it does not affect the classical equations of motion. Nonetheless,
it can affect the quantum dynamics of gauge theories. This arises because the path
integral receives contributions from field configurations that have something interesting
going on at infinity so that the boundary term Sϑ is non-vanishing. This something
interesting can be found in the topology of the gauge group.

To explain this, we first Wick rotate so that we work in Euclidean spacetime R4 .

Configurations that have a finite action from the Yang-Mills term must asymptote to
pure gauge,

Aµ → iΩ∂µ Ω−1 as x → ∞ (5.27)

with Ω ∈ G. This means that finite action, Euclidean field configurations involve a
map

Ω(x) : S3∞ 7→ G

with S3∞ = ∂R4 . Maps of this kind fall into disjoint classes. This arises because
the gauge transformations can “wind” around the spatial S3 in such a way that one
gauge transformation cannot be continuously transformed into another. Such winding
is characterised by homotopy theory. In the present case, the maps are labelled by an
element of the homotopy group which is

Π3 (G) = Z

for all simple, compact Lie groups G. In words, this means that the winding of gauge
transformations (5.27) at infinity is classified by an integer n.

This statement is most intuitive for G = SU (2) since SU (2) ∼

= S3 and the homotopy
group counts the winding from one S3 to another. For higher dimensional G, it turns
out that it’s sufficient to pick an SU (2) subgroup of G and consider maps which wind
within that. You then need to check that these maps cannot be unwound within the
larger G.

– 150 –
 It can be shown that, in general, the winding n ∈ Z is computed by
Z
1
n(Ω) = d3 S ϵijk Tr (Ω∂i Ω−1 )(Ω∂j Ω−1 )(Ω∂k Ω−1 ) (5.28)
24π 2 S3∞

Evaluated on any configuration, the theta term becomes (5.27)

Sϑ = ϑn (5.29)

It is the contribution from configurations with n ̸= 0 in the path integral that means
that observables in quantum gauge theories can depend on ϑ.

We can say more if we work in a regime in which the theory is weakly coupled.
Here the path integral is dominated by the saddle points, which are solutions to the
classical equations of motion. This means that any ϑ dependence should come from
field equations that wind at infinity, so n ̸= 0, and solve the classical equations of
motion,

Dµ F µν = 0 (5.30)

There is a cute way of finding solutions to this equation. The Yang-Mills action is
Z
1
SY M = 2 d4 x tr Fµν F µν
2g
Note that in Euclidean space, the action comes with a + sign. This is to be contrasted
with the Minkowski space action (5.1) which comes with a minus sign. We can write
this as
8π 2
Z Z
1 4 ⋆ 2 1 4 ⋆ µν
SY M = 2 d x tr (Fµν ∓ Fµν ) ± 2 d x tr Fµν F ≥ 2 |n|
4g 2g g
where, in the last line, we’ve used the result (5.29). We learn that in the sector with
winding n, the Yang-Mills action is bounded by 8π 2 n/g 2 . The action is minimised when
the bound is saturated. This occurs when

Fµν = ±⋆ Fµν (5.31)

These are the (anti) self-dual Yang-Mills equations. The argument above shows that
solutions to these first order equations necessarily minimise the action is a given topo-
logical sector and so must solve the equations of motion (5.30). In fact, it’s straightfor-
ward to see that this is the case since it follows immediately from the Bianchi identity
Dµ ⋆ F µν = 0.

– 151 –
 Solutions to the (anti) self-dual Yang-Mills equations (5.31) have finite action, which
means that any deviation from the vacuum must occur localised in Euclidean spacetime.
In other words, they are point-like objects in R4 . Because they occur for just an “instant
of time” they are known as instantons.

There is much to say about instantons. You can read about the role they play in
quantum Yang-Mills in the lectures on Gauge Theory and more about the structure
of the solutions to (5.31) in the lectures on Solitons. For our purposes, it will suf-
fice to point out that the contributions of instantons to any quantity comes with the
characteristic factor
2 |n|/g 2
e−Sinstanton = e−8π eiϑn (5.32)
2 2
Famously, the function e−8π /g has vanishing Taylor expansion about the origin g 2 = 0.
This is telling us that effects due to instantons are smaller than any perturbative contri-
bution, which takes the form g 2n . Nonetheless, that doesn’t mean that instantons are
useless since they can contribute to quantities that apparently vanish in perturbation
theory.

The theta dependence eiϑn associated to an instanton is also interesting. It is a

complex phase. The fact that it is complex can be traced to the ϵµνρσ tensor in Sϑ .
This means that Sθ contains a single time derivative and so, upon Wick rotation, still
sits in the path integral with a factor of i. The fact that n ∈ Z means that ϑ is a
periodic variable, with

ϑ ∈ [0, 2π)

Instantons are usually referred to as non-perturbative effects. This is a little bit of a

misnomer. The use of instantons requires weak coupling g 2 ≪ 1, so in this sense they
are just as perturbative as usual perturbation theory. The name non-perturbative really
means “not perturbative around the vacuum”. Instead, the perturbation theory occurs
around the instanton solution.

This also means that the theta dependence (5.32) is only expected at weak coupling
2
g ≪ 1. As we’ve seen, in the far infra-red non-Abelian gauge theories are typically
strongly coupled and the theta dependence of quantities can take a different form. We’ll
see examples in what follows.

– 152 –
An Example: An Instanton in SU (2)
It is fairly straightforward to write down the instanton solutions with winding n = 1.
For SU (2), such a configuration is given by
1 a
Aµ = 2 2
ηµν xν σ a (5.33)
x +ρ
a
Here ρ is a parameter whose role we will describe shortly. The ηµν are usually referred
to as ’t Hooft matrices. They are three 4 × 4 matrices which provide an irreducible
representation of the su(2) Lie algebra. They are given by
     
0 1 0 0 0 0 1 0 0 0 0 1
1  −1 0 0 0 2  0 0 0 −1  3 0 0 1 0
ηµν = 0  , ηµν =  −1  , ηµν =

0 0 1 0 0 0 0 −1 0 0

0 0 −1 0 0 1 0 0 −1 0 0 0

These matrices are self-dual: they obey 21 ϵµνρσ ηρσ

i i
= ηµν . (Note that we’re not being
careful about indices up vs down as we are in Euclidean space with no troublesome
minus signs.) In the solution (5.33), the ’t Hooft matrices intertwine the su(2) group
index a = 1, 2, 3 with the spacetime index µ and this implements the asymptotic
winding of the gauge fields.

The associated field strength is given by

2ρ2
Fµν = − ηa σa
(x2 + ρ2 )2 µν

This inherits its self-duality from the ’t Hooft matrices: Fµν = ⋆ Fµν and therefore solves
the Yang-Mills equations of motion, Dµ Fµν = 0.

We can get some sense of the form of this solution. First, the non-zero field strength
is localised around the origin x = 0. (By translational invariance, we can shift xµ →
xµ − X µ to construct a solution localised at any other point X µ .) The solution depends
on a parameter ρ which can be thought of as the size of the instanton lump. The fact
that the instanton has an arbitrary size follows from the classical conformal invariance
of the Yang-Mills action.

– 153 –
6 Supersymmetric QCD
We now turn our attention to the quantum dynamics of supersymmetric gauge theories.
Our focus will be on understanding the physics of super Yang-Mills and super QCD.
There is, as we shall see, a wonderfully rich array of behaviour on display.

First, some basics. There are a number of facts that we’ve seen already in these
lectures that we can combine to great effect in supersymmetric theories. First, we
know that the gauge coupling runs

1 1 b0 Λ2U V
= − log
g 2 (µ) g02 (4π)2 µ2

where g02 is the coupling constant evaluated at the cut-off scale ΛU V . The general
expression for the 1-loop beta function in non-supersymmetric theories is (5.7)
11 2 X 1 X
b0 = I(adj) − I(Rf ) − I(Rs )
6 6 fermions 6 scalars

In supersymmetric theories this simplifies. Gauge bosons are necessarily accompanied

by an adjoint Weyl fermion and chiral multiplets come in fermion/boson pairs. The
upshot is that
3 1 X
b0 = I(adj) − I(R) (6.1)
2 2 chirals

In the quantum theory, the running gauge coupling is replaced by the dynamical scale
Λ, below which the non-Abelian gauge theory is strongly coupled. For reasons that will
become clear shortly, we will refer to this as |Λ|. (It was always a real, positive energy
scale so there’s nothing lost in doing this.) This was defined in (5.4) as

8π 2
 
|Λ| = µ exp − 2
b0 g (µ)

It is RG invariant, meaning that Λ is independent of the scale µ.

Importantly, something novel happens in supersymmetric theories. This is because,

as we have seen, the gauge coupling constant sits as the imaginary part of a complex
coupling (4.15)

ϑ 4πi
τ (µ) = + 2 (6.2)
2π g (µ)

– 154 –
The theta angle does not run, essentially because it is a periodic variable ϑ ∈ [0, 2π)
and so has nowhere to go. This motivates us to define the complexified strong coupling
scale
 
2πiτ (µ)
Λ = µ exp = |Λ|eiϑ/b0 (6.3)
b0

Recall from Section 3.3 that superpotentials are holomorphic in both fields and pa-
rameters. The complexified scale Λ is therefore crying out to sit in the superpotential.
We’ll see many examples of this as we proceed.

The complexified scale also ties together two other ideas that we’ve encountered
previously. First, when discussing what kinds of superpotentials can arise in a quantum
theory in Section 3.3, we found it useful to think of a larger class of symmetries under
which parameters also transform as so-called “spurions”. Of course, if a symmetry
changes a parameter then it’s not a true symmetry of the theory but nonetheless we
saw that these spurious symmetries can prove useful in restricting the kind of behaviour
that can occur in supersymmetric theories.

Second, when discussing chiral anomalies in Section 5.2, we saw that a symmetry
of the classical theory can fail to be a symmetry of the quantum theory by shifting
the theta angle (5.26). In the supersymmetric context, a transformation of theta angle
manifests itself as a complex rotation of Λ. This means that Λ acts as a spurion for
anomalous U (1) symmetries. It also means that we can use anomalous symmetries
to restrict the form of quantum corrections to a theory, just as we used other broken
symmetries in Section 3.3. Again, we’ll see many examples of this as we proceed.

A Comment on Exact Beta Functions

There is an interesting, and somewhat subtle, story about higher order corrections to
the beta function. We can write the one-loop correction in a more revealing way by
inverting (6.3),
 
b0 Λ
τ (Λ; µ) = log (6.4)
2πi µ

Importantly, the periodicity of ϑ ∈ [0, 2π) is manifest on both sides of this equation
through

ϑ → ϑ + 2π ⇔ τ →τ +1 ⇔ Λ → Λe2πi/b0

– 155 –
Any corrections to (6.4) should retain this property. But that’s tricky to achieve while
retaining the holomorphy implied by supersymmetry. The most general form of holo-
morphic corrections, consistent with the periodicity of ϑ, is
  X ∞  b0 n
b0 Λ Λ
τ (Λ; µ) = log + an (6.5)
2πi µ n=1
µ

for some unknown coefficients an . (The restriction to n > 0 comes from requiring that
this is a weak coupling expansion and should not diverge as Λ → 0.) But these addi-
2 2
tional terms are proportional to e−8π n/g and are identified as instanton effects (5.32).
We see that all higher perturbative contributions vanish and, as far as perturbation
theory is concerned, the beta function is one-loop exact.

The fact that the beta function is one-loop exact in supersymmetric theories is a
striking statement. It appears to be even more striking when you actually compute the
two-loop contribution and find that it doesn’t vanish! What’s going on?

The resolution is that one should be careful about what quantity is actually being
computed. The holomorphic gauge coupling τ originates in a superpotential term
R 2
d θ τ W α Wα such that 1/g 2 sits in front of the Yang-Mills action. The story that we
told above assumes a renormalisation scheme in which this holomorphy is protected.

Meanwhile, the physical gauge coupling is computed after a rescaling Aµ → gAµ , so

that the coupling now appears in vertices. But absorbing the gauge coupling into the
gauge field in this way is not an entirely innocent thing to do and there is a price to pay
in the form a Jacobian in the path integral. This means that while the holomorphic
gauge coupling is one-loop exaxt, the physical gauge coupling can, and does, receive
contributions at all loops4 . (It’s not dissimilar to our discussion in Section 3.3 where
we saw that the physical parameters are renormalised even though the superpotential
is not.)

Nonetheless, it turns out that the one-loop exactness of the holomorphic gauge cou-
pling puts strong constraints on the beta function for the physical gauge coupling
which is known as the NSVZ beta function (after Novikov, Shifman, Vainshtein, and
Zakharov).
4
You can read more about these issues in the paper by Nima Arkani-Hamed and Hitoshi Muryama.

– 156 –
6.1 Super Yang-Mills
We will start our study of quantum dynamics with pure super Yang-Mills. The theory
consists of a non-Abelian gauge field coupled to a single, adjoint Weyl fermion,
Z    
4 1 1 µν µ ϑ ⋆ µν
SSYM = d x Tr 2 − Fµν F − 2iλσ Dµ λ̄ + Fµν F
g 2 16π 2
We will work with gauge group G = SU (Nc ).

The one-loop beta function (6.1) is b0 = 3Nc and the theory flows to strong coupling
at the scale |Λ|. The question that we want to answer is: what happens?

6.1.1 Confinement and Chiral Symmetry Breaking

Our first port of call is to understand the global symmetries of the theory. Classically
the theory has a U (1)R symmetry, under which

U (1)R : λ → eiα λ

This symmetry does not survive quantisation: it suffers an anomaly which can be
viewed as a transformation of the theta angle

U (1)R : ϑ → ϑ + I(adj) α = ϑ + 2Nc α (6.6)

Equivalently, we can think of the strong coupling scale (6.3) transforming as

U (1)R : Λ → e2iα/3 Λ

We say that Λ has R-charge R[Λ] = 32 . As we’ve stressed repeatedly, the shift of ϑ
means that U (1)R is not a symmetry of the quantum theory.

However, all is not lost. We can see from (6.6) that a shift by α = 2π/2Nc transforms
ϑ → ϑ + 2π. This means that a discrete Z2Nc subgroup of the R-symmetry survives,
rotating the fermion as

λ → ωλ with ω 2Nc = 1

We learn that SU (Nc ) super Yang-Mills has a discrete Z2Nc R-symmetry.

Next we should start to understand the quantum dynamics. We don’t have enough
control over the strong coupling physics of N = 1 supersymmetric theories to show
from first principles that theory confines. (It turns out that we do have such control in
theories with N = 2 supersymmetry.) We assume that, as with pure Yang-Mills, the
theory confines with a mass gap. There is little doubt that this is correct.

– 157 –
 Furthermore, as in non-supersymmetric QCD, a fermion bilinear forms

⟨ Trλλ ⟩ ∼ Λ3 (6.7)

This time supersymmetry does help us get a handle on this. We’ll see how as we proceed
through this section and, in particular, will be able to pin down the dimensionless
coefficient that sits in front of the right-hand side. But first let us understand the
consequences of the condensate.

As in non-supersymmetric QCD, this condensate spontaneously breaks a symmetry.

The difference is that it in super Yang-Mills the condensate breaks our discrete R-
symmetry,

⟨ Trλλ ⟩ → ω 2 ⟨ Trλλ ⟩

This, however, is a spontaneous breaking rather than an explicit breaking: the theory
is invariant under Z2Nc but the ground state is not. The discrete R-symmetry is broken
to

Z2Nc → Z2

where the surviving Z2 acts as fermion parity λ → −λ. This is subgroup of the
Spin(1, 3) Lorentz group and, as such, cannot be spontaneously broken.

When a continuous symmetry is spontaneously broken, we get massless Goldstone

modes. When a discrete symmetry is spontaneously broken, we get multiple ground
states. These ground states are characterised by the phase of the gluino condensate
(6.7) which, in general, can take the form

⟨ Trλλ ⟩ = aω 2k Λ3 k = 0, 1, . . . , Nc − 1 (6.8)

with ω = eπi/Nc and a ∈ R an undetermined coefficient. The upshot is that SU (Nc )

super Yang-Mills has Nc distinct ground states that differ by the phase of the condensate
(6.8)

Before we go on, it’s worth pointing out that the condensate takes the form
2 /g 2 N
Λ3 ∼ e−8π c
eiθ/Nc

This isn’t of the form (5.32) expected from an instanton contribution. Roughly, it looks
like the contribution from 1/Nc of an instanton! But we should acknowledge that the
condensate arises in the strongly coupled regime of the theory and instantons are not
a good guide to what’s going on.

– 158 –
 So far we haven’t managed to figure out the overall constant a in front of the conden-
sate. In non-supersymmetric theories, the equivalent calculation is not possible. But
in supersymmetric theories it can be done, albeit with a fairly technical computation.
Conceptually the idea is to deform the theory so that it is weakly coupled. We then
compute the gluino condensate in that regime and argue, using holomorphy, that it
remains unchanged as we move back. The end result is

a = 16π 2 (6.9)

There are (at least) two methods to get this result. One is to study the theory on
R3 × S1 rather than R4 . It turns out that the theory can be made weakly coupled
when the S1 has radius R ≪ 1/|Λ|. Moreover, rather wonderfully, when placed on a
circle instantons actually do fractionalise into Nc smaller objects and can be shown to
generate the gluino condensate5 . We’ll see another method to determine a = 16π 2 later
in these lectures.

6.1.2 The Witten Index

There is another way to see the existence of Nc supersymmetric ground states. This is
to compute the Witten index, defined in Section 3.4.2 as

Tr (−1)F e−βH

This counts the number of supersymmetric ground states of the theory, weighted with
a sign.

The beauty of the Witten index is that it stays the same no matter what you do to
the theory as long as you preserve supersymmetry. This means that if we can deform
super Yang-Mills in some way so that the theory becomes weakly coupled, then we
can just compute the Witten index using standard perturbative quantum field theory,
safe in the knowledge that it can’t then change as we deform back to the strongly
coupled regime that we care about. So the question becomes: how can we make super
Yang-Mills weakly coupled?

The way to do this is fairly dramatic. We consider the theory on a spatial torus T3
and take the radius of each circle to be R, so that the volume is V = (2πR)3 . We know
5
This calculation can be found in the paper by Davies, Hollowood, Khoze and Mattis . Be warned:
the computation of background determinants in this paper is incorrect, although the final answer is
right.

– 159 –
that super Yang-Mills is weakly coupled in the UV, but flows to strong coupling at a
scale |Λ|. If we take the spatial torus to be very small, so that
1
R≪
|Λ|
then the RG flow never reaches strong coupling. Of course, the physics of the theory
on such a tiny spatial torus is very different from the physics that we might care about.
In particular, the size of space is now much smaller than the Compton wavelength of
any massive particle so this is not going to be any good to compute, say, the S-matrix.
But there’s one thing that we can compute and that’s the Witten index.

When we compactify space in this way, nearly all states will have an energy set by
E ∼ 1/R. We can ignore these if we want to compute the number of ground states and
focus only on those modes that, classically, have zero energy. These degrees of freedom
come from both the gauge field and the fermions and we deal with each in turn.

On a torus T3 , there are gauge configurations Ai that have vanishing field strength
Fij = 0, but are nonetheless not gauge equivalent to the vacuum. These are parame-
terised by mutually commuting holonomies around each of the three different cycles
 I 
Ui = Tr P exp i Ai i = 1, 2, 3

where P is path ordering. We can use an SU (Nc ) gauge transformation to diagonalise

each of these, so that they read
i i
Ui = diag(eiθ1 , . . . , eiθNc )
The zero energy modes are the coordinates θai , with i = 1, 2, 3 labelling the spatial
directions and a = 1, . . . , Nc the gauge indices. Because Ui ∈ SU (Nc ), these coordinates
are not all independent but are constrained to obey
Nc
X
θia = 0 mod 2π (6.10)
a=1

We should quantise each of these periodic rotors θia , subject to this constraint. But
this is essentially the same as the quantisation of a particle on a circle and we know
that there is a unique ground state in which the wavefunction is independent of the θ’s.
Physically, this can be understood because a non-zero momentum for θ corresponds to
non-Abelian electric field F0i ̸= 0. This means that there’s no subtlety in quantising
the gauge field and we get a unique ground state6 .
6
A different way to count ground states can be found in Witten’s original paper “Constraints on
Supersymmetry Breaking”.

– 160 –
 We’re left with the adjoint fermion. We impose periodic boundary conditions and
the zero modes are simply the constant modes over the torus. We can again diagonalise
the fermions by an SU (Nc ) gauge transformation and write

λα = diag(λ1α , . . . , λN
α )
c

with α = 1, 2 the spinor index. Each of these is a complex Grassmann mode. Because
λ sits in the algebra su(Nc ), these are constrained to obey
Nc
X
λaα = 0 (6.11)
a=1

Let’s first recall what usually happens with such modes in quantum mechanics. A
single Grassmann mode ψ has anti-commutation relations {ψ, ψ † } = 1 and gives rise
to a qubit. This arises by first defining a fiducial state |0⟩ that obeys ψ|0⟩ = 0. The
Hilbert space then consists of two states |0⟩ and ψ † |0⟩.

We can quantise the zero modes λaα in the same way, except we have to make sure
that the end result is gauge invariant. Diagonalising λ has already exhausted much of
the gauge symmetry, but we’re still left with the Weyl group which permutes the λaα .
This means that any wavefunction must be invariant such permutations.

We begin by again introducing a fiducial state that obeys λaα |0⟩ = 0 for all α = 1, 2
and a = 1, . . . , Nc . We can build zero energy excited states by acting with (λaα )† ,
subject to the requirement of gauge invariance and (6.11). It’s straightforward to see
that there is no such state where we excite just a single (λaα )† : the requirement that
it is invariant under permutations means that it has to take the form a (λaα )† |0⟩ but
P

this vanishes by virtue of (6.10).

There is a single state with two (λaα )† excited. We first construct the gauge invariant
combination
Nc
X
S = Tr λλ = ϵαβ λaα λaβ
a=1

and then build a ground state S † |0⟩. All gauge invariant states with more λ† excitations
then arise by acting with further copies of S † . The end result is that there are Nc ground
states, given by

|k⟩ = (S † )k |0⟩ k = 0, . . . , Nc − 1

The series ends at |Nc − 1⟩ because the Grassmann nature of λaα , together with the
constraint (6.10), means that (S † )Nc = 0.

– 161 –
 G SU (N ) Sp(N ) Spin(2N + 1) Spin(4N ) Spin(4N + 2) E6 E7 E8 F4 G2
h N N +1 2N − 1 4N − 2 4N 12 18 30 9 4

Table 3. The dual Coxeter number h for all simply connected gauge groups.

Each of the states |k⟩ contains an even number of Grassmann operators and so
contributes to the Witten index with the same sign. We learn that in the regime
R ≪ 1/|Λ|, where the theory is weakly coupled, the Witten index of SU (Nc ) super
Yang Mills is given by
Tr (−1)F e−βH = Nc
But now we are at liberty to take R as large as we like, safe in the knowledge that the
Witten index does not change. Indeed, the counting above agrees with the expectations
from discrete chiral symmetry breaking (6.8), although the physics underlying these Nc
states looks very different in the two regimes.

Other Gauge Groups

There is a similar story for other gauge groups G. The R-symmetry group of super
Yang-Mills Z2h where h is a group theoretic quantity known as the dual Coxeter number.
The value of h is shown for various groups G in Table 3. The fermionic condensate
(6.7) then spontaneously breaks
Z2h → Z2
giving h distinct vacua. Similarly, one can compute the Witten index on T3 to find the
same result7
Tr (−1)F e−βH = h
In fact, there is a further subtlety in the computation on T3 . It turns out that the
Witten index depends on the global structure of the gauge group meaning that, for
example, the number of supersymmetric ground states for G = Spin(N ) and G =
SO(N ) are different. You can read more about this in Yuji Tachikawa’s lecture notes.

6.1.3 A Superpotential
Later in this section we will derive Wilsonian effective actions for light degrees of
freedom. But for super Yang-Mills there are no light degrees of freedom. The theory
has mass gap, with the lightest states having mass around ∼ |Λ|.
7
The original Witten index paper contains a subtle mistake for Spin(N ) gauge groups that was
corrected by Witten in a subsequent appendix, with further elaborations in this paper.

– 162 –
 Nonetheless, there is an interesting effective action that we can write down. It doesn’t
involve any dynamical degrees of freedom and instead depends only the parameter
Λ. We’ve already seen that the R-charge of this parameter is R[Λ] = 2/3 and the
superpotential must have R-charge 2, which means that the only thing we can write
down is

Weff = cΛ3 (6.12)

for some, as yet, undetermined constant c.

What’s the meaning of such an effective action when it doesn’t contain any dynamical
fields? In fact, it’s just another way of capturing the gluino condensate (6.7). Here we
explain why.

First, recall how we compute expectation values in the path integral. We add a
source J(x) for the operator of interest. We then compute the path integral in the
presence of the source
Z  Z 
iSSY M 4
Z[J] = D(fields) e exp i d x J Trλλ + h.c. (6.13)

The expectation value is then given by

∂ log Z
⟨ Tr λλ ⟩ =
∂J J=0

Now let’s go back to the original action for super Yang-Mills, written in terms of
superfields (4.16)
Z Z 
4 2 iτ α
SSYM = − d x dθ Tr W Wα + h.c.
8π

The lowest component of the chiral superfields is W α Wα = λα λα + . . .. But this means

that a source for the gluino bilinear naturally arises if we promote the parameter τ to
be a chiral superfield with its full complement of components
√
τ = τ + 2θψτ + θ2 Fτ

The source appears as the F-term: J = Fτ /8π.

– 163 –
 The low-energy effective action is what we get when we do the path integral, so
Z[J] = eiSeff
To write the effective action we again promote τ to a chiral superfield. There can be
a complicated Kähler potential for τ but this doesn’t concern us. (It will give terms
proportional to Fτ Fτ† but these will vanish when we set J = 0 in (6.13).) All we need
for our purposes is the contribution to Seff from an effective superpotential
Z Z
∂Weff
Seff ⊃ d x d θ Weff + h.c. = d4 x
4 2
Fτ + h.c.
∂τ
The goal is to write down a Weff that captures the right physics. Repeating the steps
above, we have
∂Seff ∂Weff
⟨ Tr λλ ⟩ = 8πi = 8πi
∂Fτ ∂τ
In this way, the effective superpotential is simply a device to encode the value of the
gluino condensate.

With these path integral gymnastics under our belt, let’s now turn to the superpo-
tential (6.12). As we’ve seen, it’s the only thing that we can write down consistent
with the (anomalous) R-symmetry. In terms of τ is is
16π 2 c 3
Weff = cµ3 e2πiτ /Nc ⇒ ⟨ Tr λλ ⟩ = Λ
Nc
in agreement with our previous result (6.8). To match the normalisation (6.9), the
coefficient c should be
c = Nc (6.14)
Note that Weff hasn’t taught us anything new about the theory. In particular, there’s
nothing to fix the coefficient c and we will have some work to do to make sure that
it’s non-vanishing. However, it will turn out that Weff will be useful in making contact
with the results that we will derive from SQCD.

6.2 A First Look at SQCD

Now we add matter. We will consider supersymmetric QCD: SU (Nc ) gauge theory
coupled to Nf massless flavours. In superspace, the Lagrangian is
Z  Z Nf h
iτ α X i
LSQCD = Tr 2
dθ 4
W Wα + h.c. + d θ Φ†i e2V Φi + Φ̃i † e−2V Φ̃i
8π i=1

The action written in component fields can be found in (4.18) .

– 164 –
 Each flavour consists of two chiral multiplets, Φ in the fundamental representation
Nc and Φ̃ in the conjugate representation N̄c . The one-loop beta function (6.1) is

b0 = 3Nc − Nf

For Nf ≥ 3Nc , the theory is non-renormalisable and infra-red free. Here the low-energy
physics is easy. We want to understand what happens when Nf < 3Nc .

6.2.1 Symmetries
The first step in understanding any quantum field theory is to get the symmetries nailed
down. Let’s start with the classical symmetries. These are:

SU (Nc ) SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R′

ϕ □ □ 1 1 1 0
ϕ̃ □ 1 □ −1 1 0
ψ □ □ 1 1 1 −1
ψ̃ □ 1 □ −1 1 −1
λ adj 1 1 0 0 1

Some obvious comments to make sure that we’re all on the same page. The first column
denotes the SU (Nc ) gauge symmetry; all others are flavour symmetries. For the non-
Abelian symmetries, □ denotes the fundamental, □ denotes the anti-fundamental, and
1 means that it is a singlet.

(As an aside: the symmetries above are actually incomplete for Nc = 2 because
the fundamental 2 is pseudoreal and so equivalent to the 2̄. This gives an enhanced
SU (2Nf ) symmetry. We won’t need this subtlety in what follows.)

Both U (1)B and U (1)A are flavour symmetries, as evidenced by the fact that the
scalars and fermions in the same multiplet transform the same way. Meanwhile, U (1)R′
is an R-symmetry, meaning that the component fields in a chiral multiplet transform
as

R[fermion] = R[boson] − 1 (6.15)

We’ve called this symmetry U (1)R′ rather than U (1)R for a reason that will become
clear shortly. The choice of R[ϕ] = 0 is somewhat arbitrary since we could always
define a new R-symmetry by combing it with any amount of the global A-symmetry.
The important point is that the R-charge of the scalars ϕ and fermions ψ differ by 1.
Note that the gluino λ always has charge +1 under the R-symmetry.

– 165 –
 Not all the classical symmetries survive quantisation. U (1)B is left unscathed as it
is vector-like, but both U (1)A and U (1)R′ suffer chiral anomalies. As we saw in (5.22),
the current conservation equation becomes
A X
∂µ j µ = Tr Fµν ⋆ F µν with A = qI(R)
32π 2 fermions

where q is the charge and R the representation under SU (Nc ). For the two symmetries
U (1)A and U (1)R′ , we have

AA = Nf × 1 + Nf × 1 = 2Nf (6.16)

and

AR′ = Nf × (−1) + Nf × (−1) + 2Nc × 1 = 2(Nc − Nf )

However, we can form a linear combination of these currents that remains conserved.
This is given by
Nf − Nc
R = R′ + A
Nf

This is an R-symmetry, rather than a flavour symmetry, because the chiral multiplet
components still obey (6.15) and R[λ] = 1. (The convention of fixing the normalisation
by insisting that R[λ] = 1 comes with the unhappy side effect that other charges are
fractional.) We can now draw up a table of the true quantum symmetries of the theory:

SU (Nc ) SU (Nf )L SU (Nf )R U (1)B U (1)R

Nf −Nc
ϕ □ □ 1 1 Nf
Nf −Nc
ϕ̃ □ 1 □ −1 Nf
Nc
ψ □ □ 1 1 −N f
Nc
ψ̃ □ 1 □ −1 −N f

λ adj 1 1 0 1

However, this misses some crucial information. This is because, as we’ve seen previ-
ously, it’s useful to keep the anomalous symmetry as a spurious symmetry. The full
symmetry structure of the theory should be thought of as reinstating the anomalous
U (1)A , but with a transformation on Λ showing that it’s not a true symmetry of the
theory:

– 166 –
 SU (Nc ) SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R
Nf −Nc
Φ □ □ 1 1 1 Nf
Nf −Nc
Φ̃ □ 1 □ −1 1 Nf

Λb 0 1 1 1 0 2Nf 0

Some of the previous information is hidden in this table. In particular, the R-symmetry
charge is that of the scalar component of the chiral multiplet and you have to remember
that R[fermion] = R[boson] − 1, together with the fact that R[λ] = 1. The final row
shows how the anomalous symmetries act on Λb0 ∼ eiϑ . We see that Λ transforms
only under the anomalous U (1)A , with the charge given by (6.16). We’ll have cause to
return to this table a number of times in what follows.

6.2.2 Runaway for Nf < Nc

The dynamics of SQCD will depend crucially on the ratio Nf /Nc . We start with small
number of colours

Nf < Nc

We already discussed the classical theory back in Section 4.3. The theory has a moduli
space of vacua M parameterised by the Nf2 gauge invariant, massless meson fields

Mj i = Φ̃j Φi

At a generic point on the moduli space M, the gauge group is spontaneously broken
to

SU (Nc ) → SU (Nc − Nf ) (6.17)

The mesons are neutral under SU (Nc − Nf ) (otherwise they would break it further) so,
at the classical level, we have massless SU (Nc − Nf ) gauge bosons essentially decoupled
from the massless mesons. We want to know what happens in the quantum theory.

We already know what will happen to the SU (Nf − Nc ) gauge bosons: they will
confine and get a mass. That leaves us with the mesons. It’s useful to start by asking:
what could possibly happen? At the crudest level, the massless fields could remain
massless, or they too could get a mass. If the latter happens, it would manifest itself in
terms of a potential generated on the moduli space. And this potential would appear
in the form of a superpotential. So we should check if it’s possible that quantum
corrections generate a superpotential that lifts the moduli space.

– 167 –
 Such a superpotential should be written in the terms of the low-energy meson fields
and must respect the various symmetries of the problem. The meson field itself trans-
forms in the (□, □) of SU (Nf )L × SU (Nf )R , so to get something invariant we should
consider det M . Under the remaining U (1) symmetries, the relevant charges are then

U (1)B U (1)A U (1)R

det M 0 2Nf 2(Nf − Nc )
3Nc −Nf
Λ 0 2Nf 0

Recall that the superpotential should have R-charge R[W ] = 2 and must be neutral
under U (1)A and U (1)B . There is a unique combination that is allowed by symmetries
1
Λ3Nc −Nf
 N
c −Nf
Weff = C (6.18)
det M
with some coefficient underdetermined coefficient C = C(Nc , Nf ).

We’ve learned that symmetries allow for a superpotential only of the specific form
(6.18). But is it actually generated? In other words, is C ̸= 0? There is a general
rule of thumb in quantum field theory that anything that isn’t prohibited by some
symmetry or other principle always occurs. The superpotential (6.18) is constructed to
be invariant under all symmetries. It is also physically sensible, with a positive power
of Λ reflecting the fact that it could be generated by strong coupling effects. Indeed, it
turns out that it is generated with the coefficient C(Nc , Nf ) given by

C(Nc , Nf ) = Nc − Nf

The result (6.18) is known as the Affleck-Dine-Seiberg, or ADS, superpotential. We’ll

give an incomplete explanation of how to determine C(Nc , Nf ) in Section 6.2.4.

Note that if we set Nf = 0, then the ADS superpotential agrees with our previ-
ous result (6.12) that captures the gluino condensate. However, when Nf ≥ 1, the
superpotential Weff is a function of dynamical fields M and tells us the fate of those
fields.

First, let’s understand the physics of the superpotential Weff . The moduli space of
vacua is a large dimensional space but we can get a sense for what happens if we think
of det M ∼ M Nf . The superpotential is then Weff ∼ M −Nf /(Nf −Nc ) . If we ignore the
Kähler potential, then the scalar potential takes the form
2
†∂Weff
V (M, M ) ∼ → 0 as |M | → ∞
∂M

– 168 –
 Figure 9. The runaway potential on the moduli space for Nf < Nc massless flavours.

This is rather striking behaviour. Classically we had an infinite number of vacua,

forming the moduli space M. Quantum mechanically we have none! The potential is
non-zero everywhere, asymptoting to V → 0 only as M → ∞ as shown schematically
in Figure 9. This is known as a runaway potential. We have a quantum theory with no
ground state. This is not something that we saw in non-supersymmetric QCD. Indeed,
it should be clear that it arises in SQCD only because of the existence of massless
scalars and their moduli space.

There are a number of caveats regarding the form of the potential, all deriving from
the fact that we don’t have good control over the Kähler potential which, as we know
from (3.29), affects the actual potential V (M ). In some circumstances, it may well be
possible that V (M ) does not increase monotonically towards the interior of the moduli
space but has some local, non-supersymmetric, minima at V (M ) ̸= 0. If so, these
would be metastable ground states, with some finite lifetime before tunnelling out and
rolling down to infinity.

6.2.3 Adding Masses

The runaway behaviour arises for massless matter. What happens if we add a mass
term? This arises from the addition of a superpotential to the our original theory,

Wmass = mj i Q̃j Qi

with mij the mass matrix. (Sorry for the proliferation of “M ” variables. To remind
you, M is the meson, m is the mass, and M is the moduli space!) We can always use
the SU (Nf ) symmetries to diagonalise the mass matrix

m = diag(m1 , . . . , mNf )

– 169 –
However, in what follows we won’t lose anything by considering a general m.

We care about the low-energy physics. We can again play the same game to determine
the superpotential using symmetries and holomorphy. In addition to M and Λ, we
now also have the mass matrix m. The transformation properties of the fields and
parameters are

SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R

2(Nf −Nc )
M □ □ 0 2 Nf

Λ3Nc −Nf 1 1 0 2Nf 0

2Nc
m □ □ 0 −2 Nf

Again, we can ask: what possible superpotentials are consistent with the symmetry?
The answer is that we can have any function
1
Λ3Nc −Nf
 N
c −Nf
Weff = f (x)
det M

where f (x) is any holomorphic function of the unique holomorphic variable x that is
invariant under all symmetries
 N 1
det M c −Nf
x = Tr(mM )
Λ3Nc −Nf

We can pin down the function f (x) by taking various limits. In the limit m → 0 and
Λ → 0, we must have f (x) = C + x so the superpotential is just the sum of the mass
term and the dynamically generated superpotential (6.18),
1
Λ3Nc −Nf
 N
c −Nf
Weff = (Nc − Nf ) + Tr(mM ) (6.19)
det M
But this limit encompasses all possible values of x, meaning that this is the exact
superpotential.

What is the physics now? We can start by looking at the case Nf = 1 where there is
a just a single complex meson M = Φ̃Φ. The superpotential now has a critical point,

∂Weff Λ3Nc −1
=0 ⇒ M Nc = (6.20)
∂M mNc −1
This is an interesting result. First, there is now a supersymmetric minimum, with the
potential sketched in Figure 10. Moreover, there are actually Nc such minima coming

– 170 –
 Figure 10. The rescued runaway, with a supersymmetric minimum when mass is added.

from taking the Ncth root in (6.20). This is to be expected since it coincides with the
Witten index for super-Yang Mills. As the mass m → 0, the minima move off to infinity
in field space. In the opposite regime, |m| ≫ |Λ|, the flavour decouples and the theory
reduces to super Yang-Mills.

Decoupling
We can look more closely at what happens in the limit |m| ≫ |Λ|. For simplicity, we’ll
take m real in what follows. Clearly this theory should reduce to super Yang-Mills but,
to make this precise, we need to be more careful about the strong coupling scales. In
particular, when we try to decouple some heavy degrees of freedom like this, there are
two strong coupling scales at play. This is because the running of the gauge coupling
happens in two steps:
• E > m: Here the gauge coupling runs with the beta function b0 = 3Nc − 1 that
is appropriate for Nf = 1 flavours. We have
 2 
1 1 b0 ΛU V
2
= 2− 2
log
g (µ) g0 (4π) µ2
If we continued this running to energies lower than m then we would hit strong
coupling at a scale that we will call
2 /b 2 2 /b 2 (m)
Λold = ΛU V e−8π 0 g0
= me−8π 0g

where, in the second equality, we’ve used the fact that Λ is an RG invariant. This
Λold is the scale Λ that appears in the formulae (6.19) and (6.20) above. However,
when the chiral multiplets have a mass, it is better thought of as something of
a counterfactual scale. The RG running never gets as low as Λold < m because
something changes along the way . . .

– 171 –
 • E < m: Now the massive chiral multiplets decouple and no longer contribute to
the beta function which becomes that of pure super Yang-Mills, with b′0 = 3Nc .
We can continue the running of the gauge coupling with this new beta function,
now starting at the scale m
b′0
 2
1 1 m
2
= 2 − 2
log
g (µ) g (m) (4π) µ2
Now it hits strong coupling at a scale that we will call
2 /b′ g 2 (m)
Λnew = me−8π 0

This is the actual scale at which the gauge coupling becomes strong.

Comparing the two results above, we have the matching condition

 b  b′
Λold 0 Λnew 0
= (6.21)
m m
In principle there can be additional multiplicative factors that arise from the matching
at scale m at higher loops. These go by the name of threshold effects. One can always
choose a regularisation scheme in which they vanish.

The result (6.21) can be used generally. For our specific purposes, we decouple from
the theory with Nf = 1 to pure super Yang-Mills, and this equation reads
c −1
Λ3N
old
3Nc
m = Λnew

In this case, Λnew > Λold . This is because the presence of matter slows the running of
the coupling. When that matter is removed, the running speeds up and so raises the
strong coupling scale.

We can now evaluate the formulae (6.19) and (6.20) in terms of the true, low-energy
scale Λnew . First we determine the expectation value M in the vacuum (6.20). Then
we substitute this into the superpotential (6.19) at the vacuum. A short calculation
shows that

Weff = Nc Λ3new

This, of course, we’ve seen before. It is precisely the superpotential (6.12) for super
Yang-Mills, now with the strong coupling scale Λnew . Even the coefficient (6.14) comes
out correctly. In this way, the Affleck-Dine-Seiberg superpotential correctly predicts
the value of the gluino condensate in super Yang-Mills.

– 172 –
A General Mass Matrix
We can repeat the calculation above for Nf flavours and a general mass matrix mij .
We just need to find the critical point
∂Weff
=0
∂M ij
of the superpotential (6.19). To do so, we should Jacobi’s formula

δ(det M ) = tr(Adj(M ) δM ) (6.22)

with Adj(M ) the adjugate matrix. If M is invertible then this coincides with the more
familiar δ(det M ) = (det M ) tr(M −1 δM ). Assuming that M is indeed invertible, we
find that the critical point obeys
1
 3Nc −Nf  N −N
Λ c f
Mj i = (m−1 )j i (6.23)
det M
We take the determinant of both sides to find
 3Nc −Nf  N N−N
f
1 Λ c f
1/Nc
det M = ⇒ Mj i = (m−1 )j i det m Λ3Nc −Nf
det m det M
Again, we see that the vacua sit at a position inversely proportional to the mass,
ensuring that they move off to infinity as m → 0. The Ncth root on the right-hand side
provides the phase ambiguity that gives rise to the Nc ground states expected from the
Witten index.

6.2.4 The Potential at Weak Coupling

There is something special that happens when Nf = Nc − 1. This is because, with
this number of flavours, at a generic point on the moduli space M the gauge group is
generically completely broken.

This is important. For any Nf < Nc − 1, there is always a residual unbroken

SU (Nc − Nf ) non-Abelian gauge group which means that the theory is necessarily
strongly coupled. However, for Nf = Nc − 1 the theory can be weakly coupled.

However, weak coupling isn’t guaranteed. For simplicity, let’s consider the point on
the moduli space where all scalars have the same expectation value (4.34),
 
v ... 0 0
 . .. 
ϕia = ϕ̃†a i = 

.. . (6.24)
0 ... v 0

– 173 –
The Higgs mechanism halts the running of the gauge coupling at the scale v of breaking,
so in the infra-red g 2 = g 2 (v). This is small provided that

v≫Λ

In other words, we can trust our weakly coupled intuition when we are far out on the
Nf = Nc − 1 moduli space, with |M | ∼ v 2 ≫ Λ. This means that, in this regime,
we should be able to compute the Affeck-Dine-Seiberg superpotenial in some more
traditional manner.

The form of the superpotential itself tells us where to look. When Nf = Nc − 1,

(6.18) becomes

Λ2Nc +1
Weff = C⋆ (6.25)
det M
2 /g 2 +iϑ
with C⋆ = C(Nc , Nc − 1). This is proportional to Λb0 ∼ e−8π , which, as we saw
in (5.32), is the characteristic signature of an instanton .

This gives a window of opportunity. Until now, our results for the quantum dynam-
ics have relied on symmetries and, crucially, holomorphy. Supersymmetry, of course,
bought us the latter. But this approach can only get us so far and, as we have stressed,
there is nothing to fix the overall constant C. In particular, we need to check that it
doesn’t vanish. This requires us to roll up our sleeves and do a weak coupling, instanton
computation. And the theory with Nf = Nc − 1 is the place to do it. The calculation
is rather technical and we won’t describe it here8 . But the result is

C⋆ = 1

Decoupling: From Weak to Strong Coupling

The single coefficient C⋆ = 1 for Nf = Nc − 1 is sufficient for us to derive the coefficient
C(Nc , Nf ) for all other values of Nf < Nc . We do this by decoupling arguments.

Let’s start with the theory with Nf = Nc − 1 flavours. We will give a large mass m
to k of these flavours. We then expect to flow down to the theory with

Nf′ = Nc − (k + 1) (6.26)

We want to derive the effective superpotential for this new theory.

8
The instanton calculation was first done by Affleck, Dine and Seiberg who showed that C⋆ ̸= 0.
The exact result C⋆ = 1 was first derived by Finnell and Pouliot.

– 174 –
 Our starting point is the superpotential (6.19) for Nf = Nc − 1

Λ2N
old
c +1

W = + Tr(mM ) (6.27)
det M
where now the coefficient C⋆ = 1 in front of the first term should be viewed as fixed by
the weak-coupling instanton calculation. Note that we’ve added the subscript “old” to
the strong coupling scale in anticipation of the fact that we will integrate out matter
to flow to a new theory with Nf′ flavours. We give a mass matrix of the form
!
0 0
m=m
0 1k

The critical point ∂W/∂Mji = 0 solves, from (6.23),

Λ2N
old
c +1

mM = 1Nf (6.28)
det M
We should pause to understand what this is telling us. The meson matrix M takes the
form
!
M̃ 0
M=
0 Z

where Z is a k × k matrix and M̃ is a (Nf − k) × (Nf − k) matrix. Note that the

off-diagonal terms in M must vanish by the equation of motion (6.28).

At first glance, it looks tricky to solve the matrix equation (6.28) because of all those
zeroes in the upper left corner of m make it difficult for the left-hand side to be equal
to the identity matrix 1Nf . But the physics is actually clear. The massive k flavours
in the matrix Z have an expectation value that’s stabilised as Z ∼ 1/m. Meanwhile,
the remaining massless flavours in the matrix M̃ have a runaway behaviour M̃ → ∞
as we’ve seen before.

Here our interest is subtly different. We will integrate out the heavy degree of
freedom Z. This means that we solve (6.28) only for Z and substitute it back in to
get an effective action for M̃ . This effective action will then tell us that M̃ suffers a
runaway, which we knew anyway. But our goal is only to find the overall coefficient
C(Nc , Nf ) in front of this runaway superpotential.

– 175 –
 Focussing on the k × k part of (6.28) gives the matrix equation
2Nc +1
Λold
mZ = 1k
det M̃ det Z
Taking traces and determinants gives
 2Nc +1 k
kΛ2N
old
c +1
k+1 Λold
mTr Z = and (det Z) =
det M̃ det Z m det M̃
If we substitute this back into the original superpotential (6.27), then we get a super-
potential purely for the M̃ mesons. It is
1
 2Nc +1 k  k+1
Λold m
W = (k + 1)
det M̃
From (6.26), we know that k + 1 = Nc − Nf′ . Meanwhile, the kind of RG matching
arguments that led us to (6.21) reveal that the numerator is the strong coupling scale
associated to SU (Nc ) with Nf′ massless flavours
3Nc −Nf′ 2Nc +1 k
Λnew = Λold m
The upshot is that we reproduce the Affleck-Dine-Seiberg superpotential for the light
meson fields as expected,
1
3Nc −Nf′ ′
! N −N
c
′ Λ new
f
W = (Nc − Nf )
det M̃
But with the added bonus that we’ve derived the long-promised coefficient C(Nc , Nf ) =
Nc − Nf .

6.3 A Second Look at SQCD

We’ve seen that the moduli space of vacua is lifted for Nf < Nc . Now we look at what
happens for higher Nf .

Our first observation is that the superpotential (6.18)

1
 3Nc −Nf  N −N
Λ c f
Weff = C
det M
is the only one allowed by the symmetries, regardless of Nf . But it makes no sense for
Nf ≥ Nc . First, it clearly diverges when Nf = Nc . Moreover, for Nf < Nc < 3Nc it
2
has negative powers of Λ, which means that the superpotential scales as e+1/g (with
some coefficient). But this diverges as g 2 → 0 and so isn’t compatible with the weak
coupling limit. In particular, we know that if we set g 2 = 0 then the theory is simply
free and nothing can be going on. This rules out the possibility of a superpotential.

– 176 –
 When Nf < 3Nc , the superpotential does have a positive power of Λ. But this
corresponds to the situation where b0 < 0 and the theory is infra-red free and no
superpotential can be generated. (Another way of saying this is that the putative
strong coupling scale scale Λ is actually bigger than the UV cut-off and shouldn’t be
trusted.) We’ll look at this theory in more detail below.

All of this means that for Nf ≥ Nc there is no possible superpotential that can arise.
The moduli space of vacua survives and, correspondingly, there are necessarily massless
degrees of freedom. Our goal is to understand them.

We will start in this section by looking at two special cases: Nf = Nc and Nf = Nc +1.
Both exhibit interesting phenomena9 . In later sections we’ll then look at higher Nf .

6.3.1 A Deformed Moduli Space for Nf = Nc

Recall that for Nf = Nc , the moduli space is parameterised by mesons Mj i = Φ̃j Φi and
baryons
a
B = ϕ1a1 . . . ϕN
aNc ϵ
c a1 ...aNc
and B̃ = ϕ̃a11 . . . ϕ̃NNcc ϵa1 ...aNc

These fields, gauge invariant composites, and parameters transform under the following
symmetries:

SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R

Φ □ 1 1 1 0
Φ̃ 1 □ −1 1 0
M □ □ 0 2 0
B 1 1 Nc Nc 0
B̃ 1 1 −Nc Nc 0
2Nc
Λ 1 1 0 2Nc 0

The classical moduli space is defined as an algebraic variety, with a single constraint
(4.39) between the fields

det M − B̃B = 0 (6.29)

9
The original paper is by Nati Seiberg, “Exact Results on the Space of Vacua of Four Dimensional
SUSY Gauge Theories”.

– 177 –
Figure 11. The singular space xy = 0 on the left and the smooth space xy = ϵ2 on the right.
This is a cartoon for the moduli space of SQCD when Nf = Nc . On the left the classical,
singular modular space; on the right, the smooth quantum moduli space.

We know that this can’t be lifted by a superpotential. But it turns out that the space
is deformed. The quantum moduli space satisfies the constraint

det M − B̃B = Λ2Nc (6.30)

There are a number of questions that spring to mind. First, what is the meaning of
this deformation? And second, how do we know that it happens?

Let’s start by answering the first of these. The mathematics is all about of the
singularities of the space, the physics all about their meaning. We can start by looking
at a much simpler example. Consider the algebraic variety defined by

xy = 0

with x, y ∈ C. This is obviously the intersection of two complex lines. (The complex
line, or often just “line” is the name given by algebraic geometers to what you used to
think of as a plane.) The space is obviously singular at the origin x = y = 0. The way
to see this mathematically is to look a the tangent vectors, δx and δy. These obey

δx y + xδy = 0 (6.31)

For any point other than the origin, there is a unique complex tangent vector. For
example, if x ̸= 0 then the tangent vector is δx since we necessarily have δy = 0. But
at the origin there is no constraint on δx and δy which is telling us that tangent vector
is ill-defined and, correspondingly, the space is singular.

– 178 –
 We can compare this to the deformed variety
xy = ϵ2
Again, this is a space with one complex dimension and, far from the origin, looks much
like xy = 0. But the origin x = y = 0 is no longer part of this space and this means
that the singularity has now been removed. Tangent vectors must still obey (6.31) but
now there is a unique tangent vector for each point obeying xy = ϵ2 . The singular and
deformed spaces are shown in Figure 11.

This simple example captures the key features of the moduli space M. The classical
moduli space (6.29) is singular. This is obviously true at the origin M = B̃ = B = 0,
but more generally it is singular on any submanifold where B̃ = B = 0 and the meson
matrix has rank(M ) ≤ Nc −2. In contrast, the quantum moduli space (6.30) is smooth.
All singularities have been removed. What is this telling us?

As we’ve seen in numerous examples in Section 4.3, singularities in the moduli space
signify the existence of new massless degrees of freedom. In the present case, there is no
mystery to this: the new massless degrees of freedom are gauge bosons. In particular,
when rank(M ) = k ≤ Nc − 2, an SU (k) gauge group is unbroken.

But these singularities are removed in the quantum theory. This tells us that the
additional particles at the origin of moduli space that were classically massless have
now gained a mass. This is the famous mass gap problem! Here we see that the a
complicated quantum effect – namely the fact that gauge bosons get a mass through
strong coupling – arises in a surprising geometric manner.

Now for the second question: how do we know that the quantum deformation of
the moduli space takes place? The first thing to note is that it’s consistent with the
symmetries and, as we’ve noted before, anything that isn’t prohibited typically occurs.
Of course, you might be forgiven for not being aware that deforming the constraint
through quantum effects was even something that could happen, but the discussion
above about the meaning of removing singularities will hopefully serve to allay such
doubts. However, we should strive to find more convincing evidence than this. And,
indeed, there are two very compelling reasons to believe that the deformation happens.

6.3.2 ’t Hooft Anomaly Matching

Our picture of physics described by the quantum modified constraint assumes that the
only massless degrees of freedom are the mesons and baryons. There are a number
of interesting constraints that this picture must satisfy. These come from ’t Hooft
anomalies.

– 179 –
 The original global symmetry of the theory is

GF = SU (Nf )L × SU (Nf )R × U (1)B × U (1)R

The ’t Hooft anomalies must be matched at each point on the quantum moduli space.
At different points, the global symmetry is broken to some subgroup, GF → HF and
this surviving subgroup changes as we move around M. But importantly, the point
M = B = B̃ = 0 where the full global symmetry GF would be completely unbroken
has been removed by the quantum deformation (6.30). There are, however, two points
where the surviving symmetry HF is maximal and anomaly matching is most stringent.
These are

• B = B̃ = 0 with M = Λ2 1Nc . At this point, the surviving global symmetry group

is

HF = SU (Nf )diag × U (1)B × U (1)R (6.32)

This is not dissimilar to the chiral symmetry breaking pattern in non-supersymmetric

QCD

• M = 0 with B̃ = B = ΛNc . At this point, the surviving global symmetry group

is

HF = SU (Nf ) × SU (Nf ) × U (1)R (6.33)

This is a symmetry breaking pattern that doesn’t (we think!) occur in non-
supersymmetric QCD. The non-Abelian chiral symmetry is unbroken but, in
contrast, baryon number is broken.

We do anomaly matching at each of these points in turn. For what follows, we will need
to frequently turn to the table of symmetries that we constructed at the beginning of
this subsection.

The Point with B̃ = B = 0

We need to match anomalies for symmetries, and any mixed anomalies between sym-
metries, for HF given in (6.32). We’ll do each in turn, starting with the non-Abelian
SU (Nf )diag symmetry.

SU (Nf )3diag : In the UV, we have the quarks ψ and ψ̃. But these cancel in their con-
tribution to the anomaly, giving AU V = 0. In the infra-red, only the meson carries
non-Abelian charge. Under the diagonal SU (Nf )diag it transforms in □ ⊗ □ = adj ⊕ 1.

– 180 –
But the adjoint is a real representation and doesn’t contribute to the anomaly, so we
have AIR = 0.

SU (Nf )2diag · U (1)B : In the UV, the quarks ψ and ψ̃ carry opposite U (1)B charge and
so cancel in their contribution, giving AU V = 0. In the IR, the mesonic fermions are
uncharged under U (1)B so also give AIR = 0.

SU (Nf )2diag · U (1)R : This is more interesting. We need to remember that the charges
listed in the table are for bosons in the chiral multiplet, with R[fermion] = R[boson]−1.
In the UV, we have

AU V = Nc × I(□) × (−1) + Nc × I(□) × (−1) = −2Nc

where the factors of Nc are because each quark has Nc colours. Meanwhile, in the IR,
the contribution from the fermionic mesons is

AIR = I(adj) × (−1) = −2Nf

Now there is no contribution from colour degrees of freedom because the mesons are
confined. Instead there is only the SU (Nf )diag group theory factor I(adj). Nonetheless,
we have AU V = AIR because we are working in the theory with Nf = Nc .

U (1)2B · U (1)R : In the UV, the quarks contribute

AU V = Nc Nf × (+1)2 × (−1) + Nc Nf × (−1)2 × (−1) = −2Nc Nf

In the IR, only the fermionic baryons contribute. These give

AIR = (Nc )2 × (−1) + (−Nc )2 × (−1) = −2Nc2

Again, AU V = AIR .

U (1)3R : This time we have to remember that there are Nc2 − 1 gluinos with charge
R[λ] = +1 in the UV. These didn’t contribute to any of the anomalies above, but they
do now. Including both gluinos and quarks, we have

AU V = (Nc2 − 1) × (+1)3 + Nc Nf × (−1)3 + Nc Nf × (−1)3 = Nc2 − 2Nf Nc − 1

In the IR, both mesons and baryons contribute to the anomaly, all with R-charge −1.
This is the first time that all the IR fields contributed and this means that it’s the first
time we need to take into account the constraint (6.30). This is a constraint not just

– 181 –
on the expectation values, but also on the fluctuations of the fields. This means that
the number of massless IR fields is dim M = Nf2 + 2 − 1 with the +2 the baryons B
and B̃ and the −1 coming from the constraint. The upshot is that the IR anomaly is

AIR = dim M × (−1)3 = −Nf2 − 1

Again, we see the anomaly matches with the UV.

There are two remaining anomalies, U (1)3B and U (1)2R · U (1)B . You can check that
both have AU V = AIR = 0 because U (1)B is vector-like.

In addition, we can match mixed U (1)-gravitational anomalies. This simply means

that the sum of U (1) charges must be the same in the UV and IR. However, in the
P
present case these don’t really give anything new. For U (1)B , we have qB = 0 in both
P P 3
UV and IR. For U (1)R all charges are qR = ±1 so qR = qR and this reduces the
3
U (1)R calculation that we did above. When we consider other theories the matching
of mixed gauge-gravitational anomalies will give more compelling results.

The Point with M = 0

We now need to match anomalies for HF given in (6.33). The only real difference from
the calculation above lies in the SU (Nf )3L anomaly. In the UV. In the UV, just the
quarks ψ contribute and give

AU V = Nc × A(□) = Nc

In the IR, the Nf2 mesons contribute. We have

AIR = Nf × A(□) = Nf

Again, AU V = AIR because we’re working in the theory with Nf = Nc . The anomaly
matching for SU (Nf )2L · U (1)R works in much the same way, giving AU V = AIR = −Nc .
The anomaly matching for U (1)3R works in the same way as we saw above.

The calculations of anomaly matching are straightforward. But the agreement is

not entirely trivial. In particular, it’s clear that it works only when Nf = Nc . As we
proceed, we’ll see anomaly matching working in more intricate ways.

Decoupling
There is a second way to see the need for the quantum deformation of the moduli space.
This uses a trick that we’ve seen before: we look at the fate of the theory when we give
one flavour a mass and decouple it.

– 182 –
 It’s not immediately obvious how to do this since, as we saw above, we don’t have
a superpotential to start with! The trick is to view the constraint (6.30) itself as a
superpotential
 
W = X det M − B̃B − Λ2Nc

where we’ve introduced a new chiral superfield X whose sole role is to act as a Lagrange
multiplier, imposing the constraint. We now add a mass for just one flavour. The
superpotential is
 
2Nc
W = X det M − B̃B − Λold + Tr(mM ) (6.34)

We’ve added the superscript “old” appears because we’re playing an integrating out
game. We’re going to look at what happens when |m| ≫ |Λold | so that we have one
massive flavour and Nf = Nc − 1 massless flavours. In this case, we should be able to
re-derive the appropriate Affleck-Dine-Seiberg superpotential. Let’s see how it works.

The rest of the calculation is very similar to the decoupling that we saw in previous
sections. The critical point for the mesons sits at ∂W/∂Mj i = 0, or

mM = −X det M 1Nf (6.35)

If we turn on a mass term for just the final Nfth flavour, with m = diag(0, . . . , 0, m).
The meson fields take the form
!
M̃ 0
M=
0 Z
N
with Z = MNf f the final flavour and the off-diagonal terms set to zero at the critical
point (6.35). The equation arising from ∂W/∂Z in (6.35) tells us that
m
X=−
det M̃
Meanwhile, the critical points for B and B̃ are
∂W ∂W
= −X B̃ = 0 and = −BX = 0
∂B ∂ B̃
which, since X ̸= 0, means that we must have B̃ = B = 0. So far Z is undetermined,
but this is fixed by the equation of motion for X which, of course, is simply the
constraint itself. It now reads

Z det M̃ = Λ2N
old
c

– 183 –
We now substitute this back into the superpotential (6.34). Only the final Tr(mM ) =
mZ term contributes and gives
2Nc
Λold m Λ2Nc +1
W = = new
det M̃ det M̃
2Nc +1 2Nc
with the now familiar RG matching giving Λnew = Λold m. This we recognise as
the Affleck-Dine-Seiberg superpotential (6.25) in the case Nf = Nc − 1 (with even
the coefficient correct). Notice that the quantum deformation of the constraint was
necessary for us to reproduce the known physics when we integrate out massive flavours.
This is our first piece of evidence (beyond the symmetries) that the deformation actually
occurs.

6.3.3 Confinement Without χSB for Nf = Nc + 1

The case of Nf = Nc + 1 also exhibits some rather startling behaviour and is worth
exploring in some detail. Recall from Section 4.3 that, in addition to the mesons Mj i ,
we now have Nc baryons of each type

Bj = ϵji1 ...iNc B i1 ...iNc and B̃ j = ϵji1 ...iNc B̃i1 ...iNc

This satisfy the constraints (4.40)

Adj(M )i j = B i B̃j and Mj i B j = Mj i B̃i = 0 (6.36)

Recall that if the adjugate matrix Adj(M ) is invertible then it is given by Adj(M ) =
(det M )M −1 . We can gather the various gauge fields together to list their symmetries
in a now-familiar table
SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R
1
Φ □ 1 1 1 Nf
1
Φ̃ 1 □ −1 1 Nf
2
M □ □ 0 2 Nf
Nc
B □ 1 Nc Nc Nf
Nc
B̃ 1 □ −Nc Nc Nf

Λ2Nc −1 1 1 0 2Nf 0

As we’ve already seen, there can be no superpotential generated on the moduli space.
But, this time, there can be no quantum deformation of the constraints either! There
is no possibility consistent with the symmetries and various weakly coupled limits. Our
quantum moduli space has singularities.

– 184 –
 What are we to make of this? As we’ve seen in several examples, the singularities
signify new massless degrees of freedom. Classically, these degrees of freedom are gauge
bosons. It’s tempting to conclude that the singularities in the quantum theory are
telling us that the gauge bosons are free at the origin of the moduli space. However,
it turns out that this is not the case. Instead, the quantum interpretation of the
singularities is rather different.

In fact an obvious quantum interpretation suggests itself if we assume that the theory
confines. This means that the low-energy fields are necessarily mesons and baryons
which, in general, are constrained by (6.36). Geometrically, the singularities of M
arise when the fluctuations of M , B and B̃ are no longer restricted to lie on M.
Physically, this translates into the suggestion that the singularities of M might be due
to unconstrained mesons and baryons. In particular, it would suggest that at the origin
of moduli space M = B = B̃, we should think of the physics as described by free,
massless mesons and baryons.

This interpretation of the singularity is rather remarkable, not least because we would
have confinement without the accompanying chiral symmetry breaking. At the origin
of moduli space, the full chiral symmetry

GF = SU (Nf )L × SU (Nf )R × U (1)B × U (1)R

is unbroken. Famously, confinement without chiral symmetry breaking is not possible

in QCD. (We sketched the argument in Section 5.2.3.) The suggestion is that this does
happen in SQCD with Nf = Nc + 1.

The phenomenon of confinement without chiral symmetry breaking in SQCD some-

times goes by the name of s-confinement. It’s a rubbish name. Here “s” can stand for
“smooth” or perhaps “screening” depending on taste.

More ’t Hooft Anomaly Matching

There is a fairly stringent test that any proposal for confinement without chiral sym-
metry breaking must pass. This is ’t Hooft anomaly matching. Let’s see how we do.

SU (Nf )3L : In the UV, we have the quarks contributing to give AU V = Nc . In the
IR, we have both mesons M , which contribute Nf and the baryons B which contribute
−1 as they sit □. Together they give AIR = Nf − 1 = Nc .

SU (Nf )2L · U (1)B : The quarks give AU V = Nc . In the infra-red, the mesons don’t
contribute while the baryon B gives AIR = Nc .

– 185 –
SU (Nf )2L · U (1)R : Now things get more fiddly, largely because of the fractional R-
charges. In the UV, the quarks give
Nc2
 
1
AU V = Nc −1 =−
Nf Nc + 1
In the IR, both the meson and baryon contribute:
   
2 Nc
AIR = Nf −1 + −1
Nf Nf
A little algebra reassuringly shows that AU V = AIR .

The remaining anomaly matching involving U (1)R gets a little messy. For example,
we have

U (1)R : The mixed U (1)R gravitational anomaly simply requires that we add up the
R-charges. Including the gluinos, we have
 
2 1
AU V = (Nc − 1) + 2Nc Nf − 1 = −Nf2 + 2Nf − 2
Nf
Meanwhile,
   
2 Nc
AIR = Nf2 − 1 + 2Nf − 1 = AIR
Nf Nf

U (1)3R : The calculation is the same as above, but with R3 instead of R. We have
3
Nf4 − 6Nf3 + 12Nf2 − 8Nf + 2

1
AU V = (Nc2 − 1) + 2Nc Nf −1 =−
Nf Nf2
Meanwhile,
 3  3
2 Nc
AIR = Nf2 − 1 + 2Nf −1
Nf Nf
Again, we find AU V = AIR .

By now, you won’t be surprised to hear that all other ’t Hooft anomalies also match.
The messier the computation, the more compelling the evidence. It certainly feels like
there is something deep going on when these complicated algebraic expressions are
found to agree.

– 186 –
Decoupling
For Nf < Nc , we built up an impressive pattern of consistency, understanding how our
new results can be used to imply our earlier ones. We can do this again here. But
there’s a curious lesson awaiting us.

You might think that we should impose the constraints (6.36) by introducing a bunch
of Lagrange multipliers. This, it turns out, doesn’t work. Instead the constraints arise
in a slightly different way. To see this, note that the symmetries allow us to introduce
the superpotential
1  
W = − 2Nc −1 det M − BM B̃ (6.37)
Λ
Using Jacobi’s formula (6.22), equations of motion from this superpotential are (ignor-
ing the overall factor of Λ2Nc −1 for now)

∂W ∂W ∂W
= M B̃ = 0 , = BM = 0 , = −Adj(M )i j + B i B̃j = 0
∂B ∂ B̃ ∂Mj i

The upshot is that the superpotential (6.37) gives the constraints (6.36) as the equations
of motion, rather than through a Lagrange multiplier. This, it turns out, is the way
the constraints should be imposed when Nf = Nc + 1.

This is a much softer way to implement constraints. A Lagrange multiplier imposes

a constraint absolutely in the path integral. In contrast, the classical equations of
motion are merely a gentle suggestion that, at weak coupling, certain configurations
carry more weight in the path integral. Presumably this is related to the fact that the
unconstrained mesons and baryons manifest themselves at the origin.

There is one further unusual aspect of (6.37) and that’s the negative power of Λ.
In previous sections, we discarded some possible superpotentials on the grounds that
2
they scale as e+1/g (with some appropriate exponent) and so didn’t reproduce our
weak coupling needs. But in this case the constraints are classical constraints and the
classical limit g → 0 simply imposes them more strenuously. So there’s nothing to be
concerned about.

We know the deal by now. We introduce a mass for the last flavour, so the superpo-
tential reads
1  
W = − 2Nc −1 det M − BM B̃ + Tr(mM )
Λold

– 187 –
with m = diag(0, . . . , 0, m). The critical point of the meson now sits at
c −1
det M − BM B̃ = Λ2N
old mM (6.38)

The meson and baryon fields can be shown to take the form,
! ! !
M̃ 0 0 0
M= , Bi = , B̃j =
0 Z B B̃

N
with Z = MNf f the final flavour. The constraints BM = M B̃ = 0 tell us that Z = 0
if B, B̃ ̸= 0. But we should still impose the equation of motion. And, indeed, Z drops
out of the equation (6.38) which becomes
2Nc −1
det M̃ − B̃B = mΛold = Λ2Nc
new

This, of course, is the quantum modified constraint (6.30) of the theory with Nf = Nc .

6.4 A Peek in the Conformal Window

At this point, we will jump to the other end of the flavour spectrum. We know that
SQCD is no longer asymptotically free when Nf ≥ 3Nc . In this situation, the low-energy
physics is easy: it is just weakly interacting gluons, gluinos and massless (s)quarks.

What if we now lower Nf slightly below the asymptotic freedom bound. Here, too,
the physics is well understood. This is for the same reason that we saw in non-
supersymmetric QCD: there is a zero of the beta function at weak coupling where
we trust the calculation. This is the Banks-Zaks fixed point. The argument holds for
SQCD just as it does for normal QCD.

Now let’s lower Nf still further. The expectation is that we will continue to flow to
an interacting conformal field theory for some range of Nf , presumably with a different
CFT for each Nc and Nf . The question is: how low can Nf go?

We don’t know the answer in the non-supersymmetric case. But it turns out, we do
know the answer for SQCD. We flow to an interacting conformal field theory in the
regime
3Nc
< Nf < 3Nc (6.39)
2
This is the conformal window.

– 188 –
 Obviously we should ask how we know the lower bound of the conformal window.
This, it turns out, follows from certain properties of supersymmetric conformal field
theories. In the rest of this section we will state these properties, although we won’t
derive them. Then, in Section 6.5, we’ll turn to the outstanding question of what
happens in the gap between Nf = Nc + 1 and the conformal window at Nf > 3Nc /2.

6.4.1 Facts About Conformal Field Theories

A conformal field theory (or CFT) describes the dynamics of interacting massless parti-
cles. Its defining feature is that it is invariant under scale transformations, also known
as dilatations,

xµ → λxµ

Such a scaling would be broken by any dimensionful parameter, such as a mass, which
is one way of seeing that conformal field theories can only describe massless excitations.

Any relativistic, scale invariant theory appears to also enjoy a more dramatic addi-
tional symmetry known as special conformal transformations. This acts as
x µ − aµ x 2
xµ →
1 − 2a · x + a2 x2
In d = 1 + 1 dimensions, there is a proof that scale invariance implies conformal
invariance. In higher dimensions, the proofs are not complete but, nonetheless, it is
thought to be true in any interacting conformal field theory.

The generators of dilatations D and of special conformal transformations Kµ take

the form

D = −ixµ ∂µ , Kµ = −i(2xµ xν ∂ν − x2 ∂µ )

They combine with the usual generators of the Poincaré algebra to form the conformal
algebra, which has the additional commutation relations

[D, K µ ] = −iKµ , [D, P µ ] = iP µ

[K µ , P ν ] = 2i(Dη µν − M µν )
[M µν , K σ ] = i (K ν η µσ − K µ η νσ )

The kinds of questions that we want to ask about conformal field theories are somewhat
different from what we’re used to. We no longer care about the masses of particles
because they’re all zero. Nor do we usually care about the S-matrix which is challenging
to define in a theory of massless particles where there can be arbitrarily low energy
excitations of increasingly long wavelengths.

– 189 –
 Instead, in a CFT we care about correlation functions. In particular, we care about
scaling dimensions. This means that we want to find operators O(x) that have the nice
property
O(λx) = λ−∆ O(x)
with ∆ the scaling dimension. If we then look at the two-point function of these
operators, we necessarily have
1
⟨ O† (x)O(0) ⟩ ∼ 2∆
|x|
These scaling dimensions are closely related to the critical exponents that were the
focus in the lectures on Statistical Field Theory.

It’s useful to look to a free, massless scalar field as an example of a trivial CFT. Here
the theory is described by the action
Z
1
S = dd x ∂µ ϕ∂ µ ϕ
2
The scaling dimension of ϕ coincides with what we often call the “engineering dimen-
sion”, or sometimes just “dimension”. It is
d−2
∆[ϕ] =
2
We don’t have Lagrangian descriptions for interacting CFTs. The closest we can get
is to write down the Lagrangian for a field theory in the UV that flows, in the IR, to
an interacting CFT. This, for example, is what happens in massless (S)QCD with a
suitable number of flavours. It may be that the resulting CFT is weakly coupled, such
as for a Banks-Zaks fixed point, in which case we can compute the scaling dimensions
∆ perturbatively. Or it may be that resulting CFT is strongly coupled, in which case
we need to turn to some other method. Other methods on the table include numerics,
the ϵ expansion that we met in Statistical Field Theory, an approach known as the
bootstrap and, as we will see, supersymmetry.

There is one important result that we will need. The interactions always serve to
increase the scaling dimension. Or, said more precisely, the dimension of any scalar
operator in a unitary, interacting CFT is bounded by
d−2
∆[O] ≥
2
10
This is known as the unitarity bound . In the language of perturbative quantum field
theory, this is telling us that the anomalous dimensions of operators are always positive.
10
It is not too difficult to derive this bound. They key step is to quantise the theory on S3 × R where
we get to use the so-called state-operator map that relates local operators to states in the Hilbert space.

– 190 –
 In addition, any operator that saturates the bound corresponds to a free field. This
means that it must decouple from everything else that’s going on in the theory.

Conformal field theories are of interest in many dimensions d. But our interests lie
strictly in d = 3 + 1. The unitarity bound reads

∆[O] ≥ 1 (6.40)

Any operator with ∆[O] = 1 is free.

Perturbing Conformal Field Theories

Suppose that you sit at a conformal fixed point. As we mentioned above, typically
there’s no action that can describe these dynamics directly but, for the sake of discus-
sion, it will be useful to pretend. So lets call it SCFT . (If you’re worried about this, it’s
better to think in terms of a partition function in the presence of sources.)

Now we perturb the CFT. We do this by adding an extra term to the action. This
extra term is some operator O(x) which, if you’re in the setting of Lagrangian field
theory, would be some combination of fields. The new action is
Z
S = SCFT + λ dd x O(x)

with λ the coefficient that governs the perturbation. The question is: what happens
next?

The answer to this depends on the dimension ∆[O]. Roughly speaking, there are
three possibilities
• ∆ < d: Such perturbations are called relevant. They change the dynamics in
the infra-red and should be thought of as initiating an RG flow from our original
CFT to somewhere else. An example is a mass term for a free, massless scalar
field. In this case, the end point is a gapped theory. However, it’s not true that a
relevant deformation always pushes us to a gapped phase. We may, instead, flow
to a different CFT.

• ∆ > d: These perturbations are irrelevant. They don’t change the low-energy
dynamics of the CFT. An example is a ϕ6 interaction in d = 3 + 1 dimensions: it
is important at high energies, but is insignificant at low energies.
Then you simply require the positivity of an arbitrary state |Pµ P µ |ϕ⟩|2 > 0 and the unitary bound
follows after a few commutation relations using the conformal algebra. What is more challenging is
to show that there is not a more stringent bound coming from some other requirement. You can find
details in the excellent Lectures on Conformal Field Theory by Joshua Qualls.

– 191 –
 • ∆ = d: These perturbations are called marginal. This arises when the parameter
λ is dimensionless.
Now things are a little more subtle. Typically, once you deform the theory by an
arbitrarily small, marginal perturbation then the dimension of λ changes and runs
under RG. It may become smaller as you flow to the IR and such perturbations are
said to be marginally irrelevant. This happens, for example, for a ϕ4 deformation
or Yukawa terms in d = 3 + 1. Alternatively, the perturbation may grow stronger
as you flow towards the IR as is the case for the coupling constant of Yang-Mills.
Such perturbations are said to be marginally irrelevant.
Alternatively, it may be that λ doesn’t run at all under RG. In this case it is said
to be exactly marginal and it means that we have a line of different conformal field
theories, parameterised by λ. This situation is rare, but does occur for certain
supersymmetric conformal field theories.

6.4.2 Facts About Superconformal Field Theories

When a theory with N = 1 supersymmetry flows to an interacting conformal fixed
point, it gives rise to a superconformal field theory (or SCFT). In addition to the
supercharges Qα and Q̄α̇ there are now superconformal charges Sα and S̄α̇ .

Importantly, SCFTs necessarily have a U (1)R symmetry. Recall that this was some-
what optional in ordinary quantum field theories. For example, U (1)R is anomalous in
super Yang-Mills and this is reflected in the transformation of the strong coupling scale
Λ. But in an SCFT U (1)R is not an option. These theories always have an R-symmetry.

The N = 1 superconformal algebra augments the conformal algebra with the Grass-
mann generators. There are commutators
1 1
[D, Qα ] = Qα , [D, Sα ] = − Sα
2 2
[R, Qα ] = Qα , [R, Sα ] = −Sα
[K µ , Qα ] = iσαµα̇ S̄ α̇ , [P µ , Sα ] = iσαµα̇ Q̄α̇

and anti-commutators

{Qα , Q̄α̇ } = σαµα̇ P µ {Sα , S̄α̇ } = 2σαµα̇ K µ

,
{Qα , Sβ } = Mµν σαµα̇ (σ ν )α̇β − i D − 23 R ϵαβ

Now there is a slight twist to the unitarity bound. The fact that the R-symmetry and
dilatation operator sit within the same algebra means that there is a rather remarkable

– 192 –
relation between them. It can be shown that the dimension of any operator is bounded
by its R-charge
3
∆[O] ≥ R[O]
2
Furthermore, chiral operators necessarily saturate this bound. Any chiral superfield Φ
has
3
∆[Φ] = R[Φ] (6.41)
2
while any anti-chiral superfield Φ̄ has
3
∆[Φ̄] = − R[Φ̄]
2
This is an extraordinarily powerful result. Usually in conformal field theories (at least
in dimension d > 2) the scaling dimensions are extremely difficult to compute. And
this remains true for most operators in a superconformal field theory. But there are
a special class of operators – those described by chiral superfields – where the scaling
dimension is trivial to compute. We just need to know its R-charge.

There is a way to get a feel for the factor of 3/2 in (6.41). Consider the Wess-Zumino
model with W (Φ) = λΦ3 , which leads to a V (ϕ) ∼ |ϕ|4 potential. This potential is
classically marginal but one can show that it is marginally irrelevant at one-loop. This
is the statement that λ → 0 in the infra-red, so that the theory becomes free at low
energies. Nonetheless, the classical potential fixes the R-charge to be R[Φ] = 2/3 so
that R[W ] = 2 as it should. Correspondingly, ∆[Φ] = 1 in the infra-red which is indeed
the right result for a free chiral multiplet.

The powerful result (6.41) also makes life easier in another way. If we have two chiral
superfields Φ1 and Φ2 then Φ1 Φ2 is also a chiral superfield. Their R-charges simply add:
R[Φ1 Φ2 ] = R[Φ1 ] + R[Φ2 ]. But so too do their dimensions: ∆[Φ1 Φ2 ] = ∆[Φ1 ] + ∆[Φ2 ].
This is unusual in a conformal field theory. Typically if you multiply operators together
then you get divergences as their positions come close and regulating these divergences
changes the dimension of the composite. But for chiral superfields, things are much
easier. We say that the chiral operators form the chiral ring.

There is, however, a small fly in the ointment. You’ve got to be able to identify the
correct R-symmetry that appears in the superconformal algebra. For example, suppose
that your theory has an R-symmetry R and a global symmetry F . Then there’s nothing
to stop us from saying that R + αF is also a valid R-symmetry for any α ∈ R. How
do we know that this isn’t the thing that we should use when computing dimensions?!
This loophole threatens to make the wondrous relation (6.41) completely toothless.

– 193 –
 Happily, there is a procedure for figuring out what combination of symmetries forms
the correct R-symmetry. This procedure is known as a-maximization. This is important
for understanding many theories and we will describe the procedure in Section 7.2.4.
However, as we’ll now see, it is not needed for SQCD.

6.4.3 The Conformal Window for SQCD

We determined the symmetries of SQCD back in Section 6.2. The charges of the chiral
superfields under the non-anomalous R-symmetry are
Nf − Nc
R[Φ] = R[Φ̃] =
Nf

This means the R-charge of the meson M = Φ̃Φ is

2(Nf − Nc )
R[M ] = (6.42)
Nf

Given the discussion above, one might wonder if we should worry about mixing of
U (1)R with U (1)B . Happily, the meson M is neutral under U (1)B so it’s not something
that we have to worry about. We can say immediately that the dimension of the meson
operator is

3(Nf − Nc )
∆[M ] = (6.43)
Nf

Let’s first test drive this formula by looking at what happens when Nf ≥ 3Nc where
SQCD is infra-red free. At the edge, we have

Nf = 3Nc ⇒ ∆[M ] = 2 (6.44)

But this is precisely what we expect. The theory is effectively free in the infra-red, so
the fields ϕ and ϕ̃ both have their canonical dimension ∆[ϕ] = ∆[ϕ̃] = 1 which agrees
with the result (6.44). The result (6.44) is telling us that the scalar fields ϕ and ϕ̃
(together with their fermionic partners) are free at Nf = 3Nc .

Note that there’s already something a little surprising here. We knew that the theory
was infra-red free at Nf = 3Nc , but only by computing the beta function. In contrast,
the result above uses only the non-anomalous R-charge! Yet the two coincide. It’s a
sign that all these things are interconnected in SQCD in a way that doesn’t happen in
the absence of supersymmetry.

– 194 –
 What happens if we now change Nf ? We can start by looking at Nf > 3Nc where,
at first glance it appears that we become a little unstuck. Here the theory remains
free and so we should still have ∆[M ] = 2. But that’s not what the formula (6.41)
seems to be telling us. However, since the theory is free in the IR, the anomalous U (1)A
symmetry is reincarnated and can now mix with the R-symmetry, changing the answer.
This is a salutary warning: there can be subtleties in blindly following (6.41).

Now let’s look at what happens as we decrease Nf below the asymptotic freedom
bound of Nf = 3Nc . We know that when Nf = 3Nc − ϵ, for some small ϵ, we’re sitting
in a weakly coupled Banks-Zaksesque superconformal field theory. The formula (6.43)
tells us that the meson has dimension
1 ϵ
∆[M ] = 2 − + ...
3 Nc
In other words, it’s slightly less than two. You should think of the meson as describing
a loosely bound state of ϕ and ϕ̃. But as Nf decreases, so too does the dimension
∆[M ]. This is telling us that the state is becoming more and more tightly bound. At
some point, the Banks-Zaks superconformal field theory becomes strongly coupled and
we lose control over its dynamics. But, by the magic of supersymmetry, we remarkably
keep control over the dimension of the chiral meson field! Eventually, the dimension of
the meson his the bound (6.40). This occurs when
3
Nf = Nc ⇒ ∆[M ] = 1
2
But, as we mentioned above, any scalar operator that has dimension 1 is necessarily
a free scalar field. This equation is telling us that the binding between ϕ and ϕ̃ has
become so strong that the composite meson operator M is actually no longer composite!
It is acting just like a fundamental scalar field. Moreover, it is now decoupled and is
free.

How should we think of this? The proposal is that the meson becoming free signifies
the end of the conformal window (6.39). In fact, we will argue shortly that the theory
at Nf = 3Nc /2 is a completely free theory in the IR with a whole bunch of other fields
joining M in the sense that they become non-interacting at low energies.

To argue this, we will turn to a new description of the physics that holds throughout
the conformal window and, also, for Nf < 3Nc /2. This is known as the dual description.

– 195 –
6.5 Seiberg Duality
Throughout this section, our interest has been in massless SQCD, defined as

SU (Nc ) gauge theory to coupled to Nf flavours Φ and Φ̃

We’ve found a plethora of interesting physics as Nf is varied. But we haven’t yet

understood what happens when Nf + 2 ≤ Nc ≤ 3Nc /2. Moreover, at the lower end of
conformal window, where we might expect a strongly interacting CFT, we’ve seen that
the meson becomes free. It would certainly be good to understand this better.

Some light comes from a rather remarkable direction. Consider the following theory

SU (Ñc ) gauge theory to coupled to Nf flavours q and q̃ and Nf2 singlets M

In the absence of the singlets, this clearly coincides with our earlier theory just with the
number of colours renamed as Ñc . However, we arrange the singlets as a matrix Mj i
with i, j = 1, . . . , Nf which is subsequently coupled to the squark superfields through
the superpotential

W = λq̃M q (6.45)

with λ a dimensionless coupling. This is now a slight twist on our original SQCD and
its dynamics may differ. We’ll see how below. Note that we’ve given the singlets the
name M . You may recall that this is the also the name that we gave to the meson in
our original theory. This is what writers call foreshadowing.

For our purposes, it’s particularly interesting to consider the case where the number
of colours in the two theories are related by

Ñc = Nf − Nc (6.46)

This second theory is known as magnetic SQCD (or mSQCD). We’ll also at time refer to
the original SU (Nc ) SQCD as the electric theory and we’ll elucidate the reasons behind
these names as we go along. We now make the following, somewhat astonishing, claim:

SU (Nc ) SQCD and SU (Nf − Nc ) mSQCD have the same low-energy physics

This relationship is known as Seiberg duality11 . The purpose of this section is to give
evidence for the claim and to understand its consequences.
11
This was first proposed by Seiberg in the paper “Electric-Magnetic Duality in Supersymmetric
Non-Abelian Gauge Theories”.

– 196 –
6.5.1 Matching Symmetries
First let’s look at some evidence. Given that the one of the two theories is always
strongly coupled, it is challenging to do any direct calculations. The simplest thing
that we can check is agreement of the symmetries.

Gauge Symmetries are Redundancies

First, the elephant in the room. The gauge symmetries are not the same! Should we
care? The answer is no. Gauge symmetries are not true symmetries of a theory: they
are merely a redundancy in the way we choose to describe the theory.

These are easy words to wheel out, but they also grate with other things we know
about physics. The theory of electromagnetism is synonymous with U (1) gauge theory.
The Standard Model of particle physics is defined as having gauge group SU (3) ×
SU (2) × U (1). If the gauge symmetry is something that isn’t actually inherent to a
theory, but just a redundancy in our choice of description, why do we hang so much on
it elsewhere?

The reason is that gauge symmetry is an extraordinarily useful redundancy when

theories are weakly coupled. In that situation, attempting to describe the physics
in terms of anything other than the gauge field, with particular gauge group, is so
ridiculously complicated that it borders on the absurd. You could, for example, choose
to describe quantum Maxwell theory in terms of the field strengths Fµν and all possible
H

Wilson line operators exp i A which carry the gauge invariant information. But
that’s certainly not easier than our usual gauge dependent description in terms of Aµ .

This means that when gauge theories are weakly coupled, the description in terms
of the gauge symmetry G is indispensable. But when things become strongly coupled,
the story is very different. In this case, the gauge symmetry reveals itself for what it
is: a redundancy. Seiberg duality makes this stark. You can describe the same physics
using two very different gauge theories. Sometimes one formulation is best suited to the
problem at hand because the physics is weakly coupled in those variables. Sometimes
the other formulation is easiest. But neither formulation is ever wrong and the fact
that the gauge symmetries don’t match in the two dual theories is a feature, not a bug.

Global Symmetries
The story is different for global symmetries. These must match. Moreover, as both
theories are claimed to flow to the same infra-red physics, their UV ’t Hooft anomalies
must match as well. Let’s see how we do.

– 197 –
 It’s useful to list, one last time, how the various fields transform. In the electric
theory, we have

SU (Nc ) SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R

Nf −Nc
Φ □ □ 1 1 1 Nf
Nf −Nc
Φ̃ □ 1 □ −1 1 Nf

Λb 0 1 1 1 0 2Nf 0

with b0 = 3Nc − Nf . For the magnetic theory, we have

SU (Nf − Nc ) SU (Nf )L SU (Nf )R U (1)B U (1)A U (1)R

Nc Nc
q □ □ 1 Nf −Nc
−1 Nf
q̃ □ 1 □ − NfN−N
c
c
−1 Nc
Nf
2(Nf −Nc )
M 1 □ □ 0 2 Nf

Λ̃b̃0 1 1 1 0 −2Nf 0

Here Λ̃ is the strong coupling scale of the magnetic theory with b̃0 = 3(Nf − Nc ) − Nf =
2Nf − 3Nc the 1-loop beta function.

The normalisation of the non-anomalous U (1)R charge is fixed, as usual, by the

requirement that the (magnetic) gluinos have charge +1. (This, in turn, follows from
the fact that the superspace coordinate has R[θ] = −1.) This, in turn, fixes the
R-charge for the dual squarks which came be written as R[q] = R[q̃] = Nc /Nf =
(Nf − Ñc )/Nf , where we see that it mimics the form in the original theory. The
requirement that the superpotential has R[W ] = 2 then fixes the R-charge of the
singlet M .
2(Nf − Nc )
R[M ] =
Nf

But this is the same as the R-charge as the meson Φ̃Φ in the original electric theory.
Moreover, because these are chiral fields, if their R-charges match then so too do
their dimensions. This provides our first, and most important, entry in the dictionary
relating the electric and magnetic theories: the singlet fields M in the magnetic theory
correspond to the meson in the electric theory.

M ∼ Φ̃Φ

This matching provides an opportunity to reiterate a lesson from above. We have

not attempted to match individual quarks and gluons on the two sides of the duality.

– 198 –
This is because these are not gauge invariant objects and so have no physical meaning
on their own. However, gauge invariant observables or fields should match across the
duality.

Next the U (1)B charges. We want to identify U (1)B in the two theories but there’s
an ambiguity in the normalisation. We’ve fixed this ambiguity in the table above by
ensuring that the dual baryons b ∼ q Nf −Nc and b̃ ∼ q̃ Nf −Nc have the same U (1)B charges
as their electric counterparts B and B̃. Crucially, their R-charges also match. This
then provides the second entry in our dictionary between the two theories: B ∼ b and
B̃ ∼ b̃. We will look a little closer at the identification of these operators shortly.

’t Hooft Anomaly Matching

Now we can play the increasingly familiar ’t Hooft anomaly game. We denote the ’t
Hooft anomalies in the original theory as Ael and those in the dual as Amag . We have

SU (Nf )3L : The quarks contribute Ael = Nc while the dual quarks and mesons give
Amag = −(Nf − Nc ) + Nf . Note that it was important that the dual quarks sit in the
□ of SU (Nf )L while the quarks sit in the □. This was also need to ensure that the
meson fields M have the same quantum numbers.

SU (Nf )2L · U (1)B : We have Ael = Amag = Nc .

N −N N −2N
SU (Nf )2L · U (1)R : We have Ael = −Nc2 /Nf and Amag = (Nf −Nc )× cNf f +Nf × f Nf c
which agree. This same counting essentially ensures that the mixed U (1)R -gravitational
anomaly also matches.

The ’t Hooft anomalies for U (1)B and ‘U (1)3B trivially vanish in both the electric and
magnetic theories because U (1)B is a vector-like symmetry. However, we do have the
mixed anomaly
 
Nc
U (1)2B · U (1)R : Ael = 2Nf Nc × − N f
= −2Nc2 . The magnetic theory has Amag =
 2  
Nc −Nf
2(Nf − Nc )Nf NfN−N c
c
× Nf
= −2Nc2

For the final matchings involving just U (1)R , we need to remember the existence of
the gluinos.

– 199 –
U (1)R : We have Ael = (Nc2 − 1) + 2Nc Nf (−Nc /Nf ) = −(Nc2 + 1). And
   
2 Nc − Nf Nf − 2Nc
Amag = ((Nf − Nc ) − 1) + 2(Nf − Nc )Nf + Nf2 = −(Nc2 + 1)
Nf Nf

U (1)3R : Now
 3
Nc
Ael = (Nc2 − 1) + 2Nc Nf −
Nf

and
 3  3
2 Nc − Nf Nf − 2Nc
Amag = ((Nf − Nc ) − 1) + 2(Nf − Nc )Nf + Nf2
Nf Nf

Both are equal. We see that all the anomalies do indeed match and seemingly in a
non-trivial fashion.

6.5.2 Completing the Phase Diagram for SQCD

Next, let’s look at some of the more immediate consequences of the duality. Clearly
magnetic SQCD, as defined in (6.46), only makes sense when Nf ≥ Nc + 2 so the claim
of Seiberg duality is that it has something to tell us about the original theory in this
regime. Moreover, we know that mSQCD is no longer asymptotically free when
3
Nf ≥ 3Ñc ⇒ Nf ≤ Nc
2
But this is precisely the regime Nc + 2 ≤ Nf ≤ 3Nc /2 that was left unresolved by our
previous methods.

If Seiberg duality is correct (and we have every reason to believe that it is!) then it
gives a very surprising answer for what happens in this regime: the original SU (Nc )
gauge theory becomes strongly coupled and flows, in the infra-red, to an entirely differ-
ent SU (Nf − Nc ) gauge theory, coupled to the the matter q, q̃ and M . This is known
as the free magnetic phase.

Note that there is no suggestion that SU (Nf − Nc ) is a subgroup of SU (Nc ), one

that perhaps arises through a Higgs mechanism. The gluons of SU (Nf − Nc ) are not
the gluons of SU (Nc )! Instead they are new, composite spin 1 particles that emerge
at strong coupling, presumably some complicated bound states of all the degrees of
freedom of the original electric theory. . We will have more to say about how the two
gauge groups are related in Section 6.5.4.

– 200 –
 Let’s now increase Nf for fixed Nc . When the electric theory sits in the conformal
window, so too does the magnetic dual
3 3
Nc < Nf < 3Nc ⇔ 3Ñc > Nf > Ñc
2 2
However, crucially, when one theory is weakly coupled, the other is necessarily strongly
coupled. For example, at the far end of the conformal window, Nf = 3Nc − ϵ, the
original electric theory is at a Banks-Zaks fixed point and under control, while the
magnetic theory is something strongly coupled. In contrast, at the lower end of the
conformal window, Nf = 23 Nc + ϵ, it is the other way around: the dual magnetic theory
sits at (something like) a Banks-Zaks fixed point, while the electric theory is strongly
coupled.

To understand the fate of the magnetic theory, we also need to take into account the
effect of the superpotential

W ∼ q̃M q

Viewed from the perspective of the UV, this superpotential gives Yukawa terms between
various fermions and scalars in the magnetic theory. The parameter λ is dimensionless,
so this appears to be a marginal operator. But, a one-loop calculation shows that λ
initially decreases as we flow towards the infra-red. The superpotential is a marginally
irrelevant operator of the free, UV fixed point.

However, this story is different when viewed from the infra-red. Suppose that we
first flow to the fixed point within the conformal window of mSQCD and then add the
superpotential (6.45). What now happens? To understand this, we need to compute
the dimension of the superpotential W at the IR fixed point.

Happily, supersymmetry gives us a handle on this because W is a chiral and so its

dimension is related to its R-charge. As we’ve seen above, the R-charges of the dual
squarks are R[q] = R[q̃] = Nc /Nf . That leaves us with the meson field M . And here
there’s something of a subtlety.

We already listed the R-charge of M in the table above but we need to revisit this.
That R-charge was determined by assuming that R[W ] = 2 which is pre-judging the
answer! This is not what we want for the present calculation. Instead, we need to
remember that before we add the superpotential, M is just a free field, decoupled from
everything else. This means that it has dimension ∆[M ] = 1 and, correspondingly,

– 201 –
Figure 12. The RG flow in mSQCD The free fixed point and the fixed point in the conformal
window are shown as black points. The superpotential induces a further flow to the red point.
This is conjectured to coincide with the fixed point of SQCD.

R[M ] = 2/3. This means that, from the perspective of the IR, the superpotential
W = q̃M q has dimension
 
3 3 2 2Nc 3Nc
∆[W ] = R[W ] = + =1+
2 2 3 Nf Nf
When we first enter the lower bound of the conformal window, we have
3
Nf > Nc ⇒ ∆[W ] < 3
2
But this means that the superpotential is always a relevant deformation in the conformal
R
window! (The measure in the action is d4 x d2 θ and [d4 x] = −4 while [d2 θ] = +1 which
is the why the the bound for a relevant superpotential is ∆[W ] < 3.)

The RG flows are shown in Figure 12. There are three fixed points in the magnetic
theory: the free theory at g = λ = 0 that can be thought of as the starting point in
the UV; the fixed point without a superpotential in the conformal window with λ = 0
and g ̸= 0; and the final fixed point with g, λ ̸= 0. The claim of Seiberg duality is that
this final fixed point of the dual theory, shown as the red dot, coincides with the fixed
point in the conformal window of the electric theory.

By the time we reach our final fixed point, shown by the red dot in the figure, we
should now take R[W ] = 2. This gives us the R-charge R[M ] that we listed in the table
with the corresponding dimension
2(Nf − Nc ) 3(Nf − Nc )
R[M ] = ⇒ ∆[M ] =
Nf Nf
It’s only when we reach this fixed point that the R-charge and dimension of M in the
magnetic theory coincides with those of the meson in the original theory.

– 202 –
Figure 13. Seiberg duality is a statement about RG flows, although the precise statement
changes as we vary Nf /Nc .

As we increase Nf ≥ 3Nc , there is no mystery about our electric theory: it is free

in the infra-red. In contrast, the magnetic theory flows to strong coupling but now
becomes the weakly interacting SU (Nc ) theory in the infra-red. We see again that
Seiberg duality is an example of a strong-weak coupling duality. When one theory is
strongly coupled, the other may be weakly coupled and vice versa. This makes it useful.

Of course there are also regimes – notably in the middle of the conformal window –
when both theories are strongly coupled. So the duality isn’t a magic bullet, solving
all our woes. But it is a dramatic and unexpected step forward.

All of this means that the exact interpretation of Seiberg duality depends on the
value of Nf /Nc . For small Nf , the electric theory flows to the weakly coupled magnetic
theory. For large Nf , the opposite happens: the magnetic theory flows to a weakly
coupled electric theory. While for Nf in the conformal window, both theories flow
to the same infra-red fixed point. This is summarised in Figure 13. However, in all
cases Seiberg duality is a statement about RG flows. This should be distinguished
from other “exact dualities” of quantum field theories or many body systems, where
there are two very different descriptions that hold at any energy scale. Examples
of exact dualities includes the high/low temperature duality of the Ising model, or
electromagnetic dualities of N = 2 and N = 4 supersymmetric theories.

6.5.3 Deformations of the Theories

So far we’ve focussed on the fixed point. But both theories also have a moduli space of
vacua, and this too should match. However, showing this isn’t straightforward because,

– 203 –
Figure 14. The phases of massless SQCD. For the values of Nf shown in red, we have a dual
description in terms of mSQCD. This dual description is weakly coupled from Nf = Nc + 2
to Nf = 3Nc /2 + ϵ.

as we saw in Section 4.3, there are some non-trivial constraints between the mesons
and baryons.

Nonetheless, we can see roughly how things work. We’ve already seen that the
singlets M are dual to the mesons in the electric theory

Φ̃Φ ∼ M (6.47)

The symmetries also allow us to match the baryon degrees of freedom

B i1 ...iNc ∼ ϵi1 ...iNc j1 ...jÑc bj1 ...jÑc

B̃i1 ...iNc ∼ ϵi1 ...iNc j1 ...jÑc bj1 ...jÑc

Nc


Each transforms in the Nf
-antisymmetric representation of SU (Nf ) which, of course,
Nf −Nc


is equivalent to the Nf
-antisymmetric representation.

The magnetic theory also has its own meson fields m̃ = q̃q and you might wonder
what becomes of these. But the equation of motion for the singlets M is simply m̃ = 0
so these dual mesons don’t give us any further light degrees of freedom.

Masses and Expectation Values

We can now perform some simple tests of the duality. Suppose that we turn on the
electric meson fields to move out on the moduli space. To start we just turn on a single

– 204 –
entry
 
v
 
 0 
ϕ̃ϕ = 
 
.. 

 . 

0
This breaks the gauge symmetry SU (Nc ) → SU (Nc − 1), now with Nf − 1 flavours.
We would like to see this behaviour in the dual theory. In fact, this is straightforward.
Giving the singlet M the same expectation value, we have
Wmag ∼ q̃M q = v q̃1 q1
This is just a mass term for the dual squark and we can integrate it out, giving us
SU (Ñc ) with Nf − 1 flavours. This is the expected dual.

Alternatively, we could give a mass to one of the quarks in the electric theory by
adding the superpotential
Wel = mΦ̃1 Φ1
After integrating out this massive flavour, we’re left with SU (Nc ) with Nf − 1 flavours.

In the magnetic theory, this same mass deformation gives

Wmag = q̃M q + mM11
The equation of motion for the singlet M then induces an expectation value for the
dual squark
q̃1 q1 = −m
This, in turn, breaks the dual gauge group SU (Ñc ) → SU (Ñc − 1). The upshot is that
we’re left with the dual theory of an SU (Nf − Nc − 1) gauge group coupled to Nf − 1
flavours. This is the expected result.

We see that these simple deformations respect the duality, with a mass term on one
side mimicked by a Higgs effect on the other.

Matching RG Scales
There’s a slight subtlety that we’ve brushed under the carpet so far. The key element
in our dictionary relating mesons Φ̃Φ ∼ M can’t quite be right. This is because the
quarks on the left-hand side are defined in the UV of SQCD and each have dimension
1 so Φ̃Φ has dimension 2. Meanwhile the singlet M is a free field in the dual theory so
has dimension 1. So our dimensional analysis is amiss.

– 205 –
 This should be straightforward to patch up: we just need some invariant RG scale
to take up the slack. But this scale should be holomorphic and, moreover, we don’t
want it to mess up the symmetries on the two sides. Either the electric RG scale Λ or
magnetic scale Λ̃ change the (admittedly spurious) U (1)A charge. But we can introduce
a new scale µ which is some geometric mean of the two
Λ3Nc −Nf Λ̃3(Nf −Nc )−Nf = (−1)Nf −Nc µNf (6.48)
The scale µ is, by construction, invariant under all symmetries, spurious or otherwise.
A better characterisation of the dictionary is then
Φ̃Φ
=M
µ
The strange looking minus sign in (6.48) is largely a convention, but it can be shown
to ensure that the dual of the dual theory brings us back to the original.

The Theory Nf = Nc + 1 Again

We’ve advertised Seiberg duality as holding for Nf ≥ Nc + 2. But it also gives the
right answer for Nf = Nc + 1, at least if we include the additional term det M in the
superpotential so that (6.45) becomes
W ∼ det M + q̃M q
This is the expected superpotential (6.37) for the Nf = Nc + 1 theory, with the dual
quarks q and q̃ identified with the baryons B and B̃.

A Glimpse of the Superconformal Index

Until now, we’ve given no more than plausible evidence for Seiberg duality. The sym-
metries and ’t Hooft anomalies match and it passes some simple tests as we deform
the theory. It turns out that there is a much more quantitative test that the duality
passes. This comes from computing an object known as the superconformal index.

The superconformal index is an extension of the Witten index. While the Witten
index receives contributions only from the ground states, the superconformal index
receives contributions from a much larger, but still restricted class of states. Moreover,
it can be reliably computed for theories even at weak coupling.

The superconformal index is defined for superconformal theories on S3 × R. It is a

function of two variables, p and q, by tracing over all states
1 1
I(p, q) = Tr (−1)F pj1 +j2 − 2 R q j1 −j2 − 2 R
Here R is the R-charge of the state while j1 and j2 are the two angular momenta
associated to the rotation group SO(4) ∼
= SU (2) × SU (2).

– 206 –
 The formulae for the superconformal indices are fairly complicated and, at first
glance, look very different for SQCD and mSQCD. It is a highly non-trivial mathe-
matical fact that these formulae do, in fact, coincide12 .

6.5.4 Why Seiberg Duality is Electromagnetic Duality

There is one feature of Seiberg duality that perhaps remains mysterious: why have we
called the dual theory “magnetic” and the original theory “electric”? The answer to this
gets to the heart of how to think about Seiberg duality and other related phenomena.

The basic idea goes back to Maxwell theory. The equations of motion are usually
written as
∂µ F µν = J µ and ∂µ ⋆ F µν = 0
with J µ the electric current. If there are no charged particles in the theory then
J µ = 0 and the Maxwell equations exhibit a surprising symmetry in which we exchange
F µν → ⋆ F µν . In terms of the underlying electric and magnetic fields, this means
E → B and B → −E
This is electromagnetic duality. It is broken in electromagnetism because our world has
electric sources, but no magnetic sources.

However, one could imagine a theory in which there are particles carrying both
electric and magnetic charges. The latter are called magnetic monopoles. In this case,
Maxwell’s equations should be replaced by
∂µ F µν = Jeµ and ∂µ ⋆ F µν = Jm
µ

with Jeµ and Jm µ

the electric and magnetic currents respectively. In such a theory,
electromagnetic duality may be restored, now with the electric and magnetic particles
interchanged. However, there is a consistency condition between electric charges qel and
magnetic charges qmag : they can be shown to obey the Dirac quantisation condition
qel qmag
∈Z
2π
A derivation of this can be found in the lectures on Gauge Theory. This has an interest-
ing consequence. The electric charge is a measure of the strength of the electromagnetic
force. (For example, the fine structure constant is α = qel2 /4πϵ0 ℏc.) The Dirac quanti-
sation conditions tells us that if the electric charges are weakly coupled, then magnetic
charges will necessarily be strongly coupled.
12
For more information about the superconformal index, see the lectures by Yuji Tachikawa or by
Abhijit Gadde.

– 207 –
 It’s not so easy to write down versions of QED that include both electric and magnetic
charges. This is because we must work with the gauge field Aµ , and the resulting
Bianchi identity ∂µ ⋆ F µν = 0 immediately implies that there are no magnetic monopoles.
However, the story becomes richer in certain non-Abelian gauge theories. It turns out
that some non-Abelian gauge theories necessarily have magnetic monopoles arising as
solitons. This means that although we start by writing a theory purely of electric
charges, the actual theory includes both electric and magnetic charges. Examples of
theories with solitonic magnetic monopoles include N = 2 and N = 4 super Yang-Mills.

However, the N = 1 SQCD theories that we’ve been considering in this Section do not
obviously contain magnetic monopoles. There are certainly no classical soliton solutions
that one can construct that have magnetic charge. On the other hand, the theories are
strongly coupled and it’s not at all clear what properties their excitations have. Part
of the claim of Seiberg duality is that the dual description should really be thought of
as a kind of electromagnetic duality, with the SU (Nf − Nc ) gauge group related to the
original SU (Nc ) gauge group by something morally equivalent to swapping electric and
magnetic fields. Correspondingly, the dual baryons b and b̃ should be viewed as some
kind of magnetic excitation from the perspective of the original theory.

You may have noticed that I’m saying a lot of words here and not writing down any
formulae! That’s because it’s difficult to make the above claims precise. There are,
however, some hints that this is the right way to think about things. For example, the
relationship (6.48) between the scales

Λ3Nc −Nf Λ̃2Nf −3Nc ∼ constant

This formalises something that we’ve already seen: Seiberg duality is a strong-weak
duality. As the gauge coupling in one theory gets smaller, the coupling in the other
gets larger. This is reminiscent of the behaviour in electromagnetic duality.

However, the best evidence that Seiberg duality should be viewed as electromagnetic
duality comes from exploring other theories. In particular, N = 2 and N = 4 theories
both exhibit a form of electromagnetic duality where both electric and magnetic degrees
of freedom can be made manifest. The existence of a duality means that there are two
formulations of the theory, one in which the electric objects are viewed as fundamental
particles and the other in which magnetic objects are fundamental particles. In either of
these descriptions, the other particles arise as solitons. Its only when Seiberg duality is
viewed within this larger context as one of many dualities among quantum field theories,
that it becomes clearer that it is, indeed, a version of electromagnetic duality.

– 208 –
7 More Supersymmetric Gauge Dynamics
There are many more interesting properties of N = 1 gauge theories. In this section,
we describe a few of them.

7.1 Other Gauge Groups

One obvious generalisation of the previous results comes from looking at other gauge
groups. There is a similar story for both Sp(N ) and SO(N ) gauge groups, with a
runaway potential for a small number of flavours and a dual description available in
the conformal window. It tuns out that SO(N ) is significantly more complicated, with
a number of twists and turns along the way13 . Here we give the details only for the
much simpler case Sp(N ).

The classical Lie group Sp(N ) is subgroup of SU (2N ) that leaves invariant the anti-
symmetric tensor

J = 1N ⊗ iσ 2

The group Sp(N ) has dimension N (2N +1), rank N and the fundamental representation
has dimension 2N . For the lowest rank we have

Sp(1) = SU (2)

Be warned: you will find different naming conventions for this group in the literature.
Some authors prefer U Sp(2N ) to Sp(N ), where the argument now describes the di-
mension of the smallest representation rather than the rank. More confusingly, other
authors write Sp(2N ) for Sp(N )!

7.1.1 Sp(N ) Quantum Dynamics

In this section, we consider Sp(Nc ) gauge theory coupled to 2Nf chiral multiplets Qi
in the fundamental representation14 .

The representations of Sp(Nc ) are pseudoreal which means that there’s no sense
in which the matter comes in conjugate pairs. Nonetheless, there’s a subtle effect in
Sp(Nc ) gauge theories called the Witten anomaly that means that Sp(Nc ) gauge theories
only make sense when coupled to an even number of fundamental Weyl fermions. Hence
the 2Nf above.
13
A question on the examples sheet covers the key duality. You can find the full details in the
original paper by Ken Intriligator and Nati Seiberg.
14
This theory was first discussed by Ken Intriligator and Philippe Pouliot.

– 209 –
 To understand this theory, we can largely follow the path laid down in the previous
section. The 1-loop beta function is given by

b0 = 3(Nc + 1) − Nf

Next, the symmetries. In the case of Nf = 0, the U (1)R symmetry is anomalous with a
surviving Z2(Nc +1) . This, in turn, is spontaneously broken to Z2 by a gluino condensate
⟨ Tr λλ ⟩ =
̸ 0, giving Nc + 1 ground states. Indeed, this coincides with the Witten index

Tr (−1)F e−βH = Nc + 1

When Nf > 0, there is a surviving R-symmetry. Taking into account the anomaly, the
symmetries of the theory are

Sp(Nc ) SU (2Nf ) U (1)A U (1)R

Nc +1
Q □ □ 1 1− Nf

Λb 0 1 1 2Nf 0

This is largely sufficient for us to understand what becomes of the quantum dynamics
of this theory.

First, we should understand the classical dynamics. For Sp(Nc ) gauge theories there
are no baryons and the classical moduli space is parameterised solely by mesons,

Mij = Qia Qjb J ab (7.1)

with a, b = 1, . . . , 2Nc the group index and i, j = 1, . . . 2Nf the flavour index. Impor-
tantly, these mesons are anti-symmetric in the flavour indices: Mij = −Mji .

When Nf ≤ Nc , there are no further constraints on these mesons. The classical

moduli space has dimension dim M = Nf (2Nf − 1). At a generic point, the gauge
group is broken from Sp(Nc ) to Sp(Nf − Nc ).

For Nf > Nc , there is a constraint arising from the fact that the mesons M have
rank(M ) ≤ 2Nc . This classical constraint can be written as

ϵi1 ...i2Nf Mi1 i2 Mi3 i4 . . . Mi2Nc +1 i2Nc +2 = 0 (7.2)

At a generic point, the Sp(Nc ) gauge group is broken completely. As with the SU (Nc )
theories, this moduli space has singularities whenever the rank drops below the maxi-
mal. These signify the emergence of massless, unbroken gauge bosons.

– 210 –
 So much for the classical theory. What about the quantum? Given our earlier
results about SQCD, we might expect that a superpotential is generated, lifting the
moduli space for some low Nf . We can use the symmetries above to determine what
superpotential is possible. First, we need to form an object that is invariant under
the SU (2Nf ) flavour symmetry. For SU (Nc ) SQCD, this was the determinant of the
meson matrix. But for Sp(Nc ), we have something a little different. This is because
the meson (7.1) is necessarily anti-symmetric in the i, j flavour indices which means
that it’s natural to consider the Pfaffian, defined by

(Pf M )2 = det M

This has U(1) charges R[Pf M ] = 2(Nf − Nc − 1) and A[Pf M ] = 2Nf .

Runaway for Nf ≤ Nc
The symmetries allow a unique dynamically generated superpotential
 3(Nc +1)−Nf 1/Nc +1−Nf
Λ
W =C (7.3)
Pf M
for some coefficient C. This superpotential only makes sense for Nf ≤ Nc where it
gives rise to a runaway potential, lifting all ground states. For the case Nf = Nc , the
gauge group is completely broken and here the superpotential arises from an instanton
with the characteristic signature Λb0 . An explicit weak coupling calculations shows that
C ̸= 0 and the superpotential is indeed generated.

As for SQCD, giving the flavours a mass stabilises the vacua at a finite distance and
reveals the Nc + 1 ground states expected by the Witten index. If we crank up the
mass and integrate out the massive flavours, we can derive the runaway superpotential,
together with the coefficient C, for all smaller values of Nf .

Deformed Moduli Space for Nf = Nc + 1

For Nf = Nc + 1, the classical constraint (7.2) reads

Pf M = 0

For this choice of Nf , we have R[M ] = 0 and there is an opportunity for the classical
constraint to pick up a quantum deformation to

Pf M ∼ Λ2(Nc +1) (7.4)

The classical moduli space had singularities arising from massless gauge bosons. These
are removed in the quantum moduli space, signalling confinement.

– 211 –
 To see this the quantum deformation does indeed occur, we can repeat the analysis
of SQCD and integrate out the last flavour. The only real difference comes from the
fact that Mij is now anti-symmetric. We start with a superpotential imposing the
constraint, together with a mass term for the final flavour which we call Z
2(N +1)
W = X(Pf M − Λold c ) + mZ with Z = M2Nc +1,2Nc +2 (7.5)

where we’re not being too careful about the overall coefficient in front of the quantum
deformation. (There are some annoying factors of 2 that appear in the Sp(Nc ) analysis
that aren’t there for SU (Nc ).) We write the meson matrix as
 
M̃
 
M = 0 Z 

−Z 0

The equation of motion for Z and X give

2(N +1)
m Λold c
X=− and Z =
Pf M̃ Pf M̃
Substituting this back into the constrained superpotential (7.5) reproduces the expected
2(N +1)
runaway behaviour (7.3) with the matched RG scales Λ2N new
c +1
= Λold c m.

We can also do some ’t Hooft anomaly matching. When M satisfies the quantum
modified constraint (7.4), the global symmetry is broken to

SU (2Nf ) × U (1)R → Sp(Nf ) × U (1)R

There is no need to match the Sp(Nf ) anomalies because the relevant group theoretic
cubic invariant simply vanishes for Sp(Nc ). But we still have others

Sp(Nf )2 · U (1)R : In the UV we have just the quarks with R[ψ] = −1. The ’t Hooft
anomaly is

AUV = −2Nc

In the IR, we have only mesons. The chiral superfields have R-charge R[M ] = 0, so
the fermions have charge −1. They transform in the anti-symmetric representation of
Sp(Nf ). This has dimension dim( ) = Nf (2Nf − 1) − 1 and Dynkin index I( ) =
2Nf − 2. The ’t Hooft anomaly is then

AIR = −(2Nf − 2) = −2Nc

– 212 –
U (1)3R : In the UV we have both gluinos and quarks, contributing

AUV = Nc (2Nc + 1) × (+1)3 + 4Nc Nf × (−1)3 = −Nc (2Nc + 3)

In the IR, we have just the mesons, giving

AIR = −Nf (2Nf − 1) − 1

which agrees with AU V . A similar counting also shows that the mixed U (1)R -gravitational
anomaly matches.

Confinement Without χSB for Nf = Nc + 2

Now there can be neither a superpotential generated on the moduli space, nor a quan-
tum deformation of the constraints. We are left with the classical moduli space, subject
to the classical constraint (7.2). This space has a singularity at the origin.

As with SQCD, the constraints are not imposed by a Lagrange multiplier, but instead
arise as the equations of motion from the superpotential
Pf M
W =
Λ2Nc +1
Once again, we propose that the quantum interpretation of this singularity is differ-
ent from the classical interpretation. The gauge gauge bosons, which are classically
massless, are thought to confine with the singularity at M = 0 arising because all
1
2
× (2Nf ) × (2Nf − 1) elements of the anti-symmetric meson matrix M are massless.

Once again, this proposal must pass the stringent tests of ’t Hooft anomaly matching.
We have

SU (2Nf )3 : In the UV, the quarks give AU V = 2Nc . In the infra-red, the mesons
sit in the anti-symmetric representation and AIR = A( ). This is given by A( ) =
2Nf − 4 = AU V .

SU (2Nf )2 · U (1)R : The quarks now have R-charge R[ψ] = −(Nc + 1)/(Nc + 2) and
so contribute to the UV ’t Hooft anomaly as AU V = −2Nc (Nc + 1)/(Nc + 2). In the IR,
the mesons have R-charge R[M ] = 2/Nf and, of course, the fermions in this chiral mul-
tiplet have R-charge R[M ] − 1. For SU (2Nf ), the Dynkin index of the anti-symmetric
representation is I( ) = 2Nf − 2, so we have AIR = 2(Nf − 1) × (2/Nf − 1) = AU V .

– 213 –
U (1)3R : The gluinos and quarks give
3
(2Nf − 1)(Nf − 2)3

3 1
AU V = Nc (2Nc + 1) × (+1) + 4Nc Nf × −1 =
Nf Nf2

Meanwhile, the mesons give

 3
2
AIR = Nf (2Nf − 1) × − 1 = AIR
Nf

U (1)R : This time the mixed U (1)R -gravitational anomaly gives a different counting.
We have
 
1
AU V = Nc (2Nc + 1) × (+1) + 4Nc Nf × − 1 = −2Nf2 + 5Nf − 2
Nf
Meanwhile, the mesons give
 
2
AIR = Nf (2Nf − 1) × − 1 = AIR
Nf
Again, we see that all ’t Hooft anomalies match as they should.

7.1.2 Seiberg Duality

For Nf ≥ Nc + 3, we turn to a dual description. The claim is that Sp(Nc ) with 2Nf
chiral multiplets is dual to

Sp(Ñc ) with 2Nf chiral multiplets q in the fundamental and singlets Mij

Here Mij sits in the anti-symmetric representation of the SU (2Nf ) flavour symmetry
and is coupled to the other fields through the superpotential

W = Mij qai qbj J ab

with a, b = 1, . . . , Ñc and i, j = 1, . . . , Nf . The rank of the dual gauge group should be
taken to be

Ñc = Nf − Nc − 2

One can perform all the same tests of Seiberg duality that we saw for SU (Nc ) SQCD.
The proposal passes them all.

– 214 –
Figure 15. The phases of Sp(Nc ) gauge theory with 2Nf massless, fundamental chiral
multiplets.

For now, we can use the duality to put together the phase diagram for Sp(Nc ) with
2Nf fundamental chiral multiplets. It looks very similar to the SU (Nc ) case, with just
the numbers changing.

Jumping first to large Nf , the original electric theory is infra-red free when Nf ≥
3(Nc + 1). For Nc + 3 ≤ Nf ≤ 32 (Nc + 1), the magnetic theory is infra-red free. For
3
2
(Nc + 1) < Nf < 3(Nc + 1), both theories flow to the same conformal fixed point.
The upshot is that the phase diagram for Sp(Nc ) theories looks very similar to that of
SU (Nc ) SQCD. It is shown in Figure 15.

7.1.3 SU (2) Gauge Theory Revisited

As we mentioned at the beginning of this section, Sp(1) = SU (2). That means that
we now have two different stories for SU (2) gauge theory, one presented here and the
other in Section 6. We should check to make sure that they are consistent.

Things start out looking fine. For Nf = 0, the Witten index tells us that there
are two ground states. For Nf = 1, there is just a single meson field M and in both
descriptions we have the superpotential

Λ5
W =
M
For Nf = 2, our two descriptions are the same, but with slightly different names
for various objects. In the SU (Nc ) language, we introduced four mesons Mij , with
i, j = 1, 2 and two baryons B and B̃, making 6 in total. In the Sp(1) language, we only

– 215 –
have mesons that, to avoid confusion, we’ll call M̂ij . These have i, j = 1, . . . , 4 with
the requirement that M̂ij = −M̂ji again making 6 in total. One can show that

det M − B̃B = Pf M̂

This means that both the classical constraint, and the quantum deformed constraint,
coincide in the two descriptions.

There is a similar story when Nf = 3. Now in the SU (2) description there are 9
mesons M and 6 baryons B and B̃, while in the Sp(1) description there are 12 × 6 × 5
mesons M̂ .

Things start to get more interesting when we move into the realm Nf ≥ 4 where the
dual description is available to us. The gauge invariant operators M , B and B̃ still
match the mesons M̂ . But the dual descriptions are very different.

To see this, let’s look at SU (2) with Nf = 4 flavours. The two dual descriptions
are based on SU (Nf − Nc ) and Sp(Nf − Nc − 2) gauge theories respectively, which
happily coincide for Nc = 2 and Nf = 4. But the singlet fields which couple through a
superpotential are different. The SU (Nf − Nc ) dual gives
P4
SU (2) with Nf = 4 flavours and W = i,j=1 q̃i Mij qi

The global symmetry of this theory is SU (4)2 ×U (1), acting on the q̃ and q individually.
Meanwhile the Sp(Nf − Nc − 2) dual gives
P8
SU (2) with Nf = 8 chiral multiplets and W = i,j=1 qi M̂ij qi

Now we haven’t split the matter into two sets, q and q̃. Correspondingly, the theory
has a much larger SU (8) global symmetry. From our discussion above, both of these
theories must flow to the same IR fixed point. This means that the first theory must
develop the full SU (8) flavour symmetry in the infra-red. In fact, it turns out that
there are a number of other ways to split the matter multiplets, giving different duals.
You can read more about this in the lectures by Yuji Tachikawa.

For Nf ≥ 5, things start to look even more different. For example, when Nf = 5 one
dual is an SU (3) gauge theory while the other is an Sp(2) = Spin(5) gauge theory. We
see that dual theories can come in different forms: there is nothing that tells us that
there is a unique dual (or, indeed, any dual) for a given gauge theory.

– 216 –
7.2 A Chiral Gauge Theory
A chiral gauge theory is defined to be one in which left and right handed fermions
transform differently under the gauge group. In the supersymmetric context, this
means that chiral multiplets do not come in conjugate pairs.

It’s not completely straightforward to write down consistent chiral gauge theories
because we have to make sure that there are no gauge anomalies. Furthermore, in the
absence of supersymmetry, chiral theories are those that we understand least, in large
part because the Nielsen-Ninomiya theorem provides an obstacle to simulating these
theories on a computer. Notably, the Standard Model is an example of a chiral gauge
theory, albeit one where the chiral interactions are weakly coupled and so we can use
perturbation theory to understand what’s going on.

The purpose of this section is to describe the dynamics of some simple supersym-
metric chiral theories.

7.2.1 SU (N ) with a Symmetric

Consider a G = SU (N ) gauge theory, with a single chiral multiplet S in the sym-
metric representation and N + 4 chiral multiplets Q̃i in the anti-fundamental. This
is a consistent chiral theory because the the symmetric representation contributes
A( ) = N + 4 to the SU (N ) anomaly, which is subsequently cancelled by the Q̃i with
i = 1, . . . , N + 4, each of which contributes A(□) = −1.

The symmetry structure of the theory is

SU (N ) SU (N + 4) U (1)F U (1)R
−2
S 1 N +4 −N
N +2
Q̃ □ □ −(N + 2) 1

There is a large classical moduli space, parameterised as always by gauge invariant,

holomorphic monomials of the matter fields. These are:
mesons : M ij = Q̃i S Q̃j
flavour singlet : U = det S
baryons : B = Q̃N (7.6)
′ N
more baryons : B = (Q̃S)
where the baryons are contracted with an SU (N ) epsilon symbol; there are NN+4 of

them. As always, there are some constraints among these operators, including M N =
U B 2 and B ′ = U B.

– 217 –
 There is no superpotential that we can write down consistent with the symmetries,
so this moduli space survives in the quantum theory. (The flavour singlet U has charge
under U (1)B , while other flavour singlets that you might think you could construct,
such as det M or M 4 B 2 vanish identically.)

We can move out along the moduli space in various directions, breaking the gauge
and global symmetries in some manner. The physics far out along the moduli space
can be understood using weakly coupled analysis (possibly with some strong coupling
physics of the unbroken part of the gauge group still to deal with). Here we would like
to understand what happens at the origin of the moduli space.

First note that there’s no issue with asymptotic freedom in these theories. As the
number of flavours increases, so too does the number of colours and the theories are
asymptotically free for all N . However, there is an issue with the unitarity bound
(6.40). This tells us that any chiral operator in an interacting superconformal theory
must have R-charge
2
RIR [O] > (7.7)
3
where, crucially, RIR is the R-charge at the superconformal point. In general, this
may not coincide with the R-symmetry that we identify in the UV. Indeed, there’s an
ambiguity in our choice of R-symmetry in the table above: we made a specific choice,
but we could equally as well have chosen a new R-symmetry which involved the old
one, together with a mix of U (1)F . In general, the IR R-symmetry could be a mix

RIR = R + αF (7.8)

for some α ∈ R. We don’t yet have any way to determine which combination should
be identified with the R-symmetry of the conformal field theory.

We will, in fact, explain how we can identify RIR in Section 7.2.4. But for now,
let’s take the most general case (7.8) and look at the R-charges of two of our chiral
operators, M and U . They are
 
N +6 N −2
RIR [M ] = − αN and RIR [U ] = N − + α(N + 4)
N +2 N +2
You can see immediately that, for large N , there is going to be a problem satisfying the
unitarity bound (7.7). The first term for RIR [U ] is negative, so we must take α > 0.
But then, for large enough N , we will necessarily have RIR [M ] < 0. A short calculation
shows that there is no choice of R-symmetry for which RIR [M ] > 32 and RIR [U ] > 23
whenever N ≥ 13.

– 218 –
 This suggests that the chiral theory flows to a free infra-red theory when N ≥ 13 and
to an interacting SCFT when N < 13. In fact, for the intermediate case of N = 13,
there is a choice for which RIR [M ] = RIR [U ] = 23 , suggesting again that these fields
may be free.

7.2.2 A Chiral Duality

To better understand the infra-red physics, we can try to find a dual description. It
turns out that the chiral gauge theory described above has a rather startling dual15 . It
has gauge group

G̃ = Spin(8)

This group, which is the double cover of SO(8), is rather special as it has three,
inequivalent represents all of dimension 8. These are the vector 8v , the spinor 8s
and the conjugate spinor 8c . The dual theory has a single chiral multiplet p in the
spinor representation and N + 4 chiral multiplets qi in the vector representation. In
addition, there are Spin(8) singlet fields M ij and U and a superpotential

W = M ij qi qj + U pp (7.9)

The symmetry structure of the theory is

Spin(8) SU (N + 4) U (1)′F U (1)′R

q 8v □ −1 1
p 8s 1 N +4 −5
M 1 2 0
U 1 1 −2(N + 4) 12

Let’s first see why these two theories might be dual to each other. First, each have the
same global symmetry SU (N + 4) × U (1)2 . Note, however, that we haven’t yet made
any attempt to match the two Abelian symmetries across the duality. We’ll do this
shortly.
15
This was first found by Philippe Pouliot and Matt Strassler. A closely related duality was previ-
ously found by Pouliot, and his name is sometimes attached to these dualities. They are also referred
to simply as Seiberg Dualities as a catch-all for this kind of behaviour.

– 219 –
 In addition, the gauge invariant chiral superfields match. For our Spin(8) theory, the
obvious qq and pp mesons are killed by the equations of motion of the superpotential.
(Indeed, this is largely the purpose of the superpotential.) We do, however, have the
singlets M ij and U whose names already suggest how they might map to the original
theory,

Q̃i S Q̃j ←→ M ij
det S ←→ U

Moreover, we can use these to understand how the Abelian symmetries map across
both sides of the duality. The symmetries match if we rescale the global symmetry a

2F = −N F ′

We can’t rescale the R-symmetry because it’s fixed by the requirement that R[gluino] =
1. However, the two R-symmetries on either side of the duality can differ by a flavour
symmetry. You can check that the R-symmetries match if we take
1 N +6
R′ = R + F
N N +2
With these redefinitions, our group of symmetries read

Spin(8) SU (N + 4) U (1)F U (1)R

1 N −2
q 8v □ 2
N 2(N +2)
N 2 +4
p 8s 1 − 12 N (N + 4) 2(N +2)
N +6
M 1 −N N +2
N (N −2)
U 1 1 N (N + 4) − N +2

These most likely aren’t the R-symmetries that you would have chosen. But they’re
the R-symmetries we’ve got!

We haven’t yet discussed the baryons of either theory. It turns out that these too
agree, as do the moduli spaces, but there’s a subtlety awaiting us so we will postpone
that discussion to Section 7.2.3. Instead, with the symmetries in hand we can turn
to the next check: ’t Hooft anomaly matching. For example, those involving the non-
Abelian global symmetry are

SU (N + 4)3 : In the electric theory, we have Ael = N . In the magnetic theory, the
q contribute Amag = −8 while the mesons M contribute Amag = (N + 4) + 4, so

– 220 –
Ael = Amag as it should.

SU (N + 4)2 · U (1)F : In the electric theory, we have Ael = −N × (N + 2). In the

magnetic theory we have Amag = 8 × ( 12 N ) + (N + 4 + 2) × (−N ) = Ael .

SU (N + 4)2 · U (1)R : Since R[Q̃] = 1 the corresponding fermions are uncharged and we
N −2
have Ael = 0. In the magnetic theory, Amag = 8×( 2(N +2)
−1)+(N +4+2)×( N +6
N +2
−1) =
0.

We won’t check all of the others, but here are a couple to give you a sense. For the
mixed U (1)R -gravitational anomaly we have

N −2
U (1)R : This has Ael = (N 2 − 1) + 12 N (N + 1) × (− N +2
− 1) = (N −2)(N
N +2
+1)
where the
contributions are from the gluino and the S field respectively. In the Spin(8) magnetic
theory, we have
   2 
N −2 N +4
Amag = 28 + 8(N + 4) −1 +8 −1
2(N + 2) 2(N + 2)
   
1 N +6 N (N − 2) (N − 2)(N + 1)
+ (N + 4)(N + 5) −1 − −1 =
2 N +2 N +2 N +2
while for the U (1)3R anomaly we have

N −2

3
U (1)3R : This has Ael = − 12 N (N + 1) N +2
+ 1 . Meanwhile,
 3  2 3
N −2 N +4
Amag = 28 + 8(N + 4) −1 +8 −1
2(N + 2) 2(N + 2)
 3  3
1 N +6 N (N − 2)
+ (N + 4)(N + 5) −1 − −1
2 N +2 N +2
A little algebra (or Mathematica) shows you that Ael = Amag . Needless to say, the other
’t Hooft anomalies involving U (1)F and mixed U (1)R , U (1)F also coincide. As alway,
the agreement of these fairly complicated algebraic expressions gives some confidence
that the two theories are indeed related in some way.

Consequences for the Infra-Red Dynamics

Let’s now run with the conjecture that these two theories are dual. The magnetic
Spin(8) theory has the one-loop beta function given by
3 1 1
b0 = × (8 − 2) − (N + 5) = (13 − N )
2 2 2

– 221 –
We see that the theory is asymptotically free only when N < 13. But this agrees
perfectly with our previous analysis of the conformal window of the electric theory!
The duality tells us that the chiral theory is indeed infra-red free when N ≥ 13, but
the free theory is a Spin(8) gauge theory, with the matter described above. Needless
to say, it’s unlikely that we would have guessed this starting the SU (N ) gauge theory.

Meanwhile, for 2 ≤ N ≤ 12, both theories are expected to flow to an interacting

SCFT. The statement of Pouliot duality here is that, once we include the superpotential
(7.9), the two theories flow to the same SCFT.

A Deformation of the Duality

As always, given a duality we can deform it in different ways to derive new (or perhaps
old) dualities. Indeed, understanding how connections in the web of different dualities
is an important consistency check on any new proposal.

There are many ways to deform our chiral duality. Here we just mention two partic-
ularly straightforward ones. First, suppose that we add
W = det S (7.10)
to the electric side. We have the same gauge theory, just with this additional superpo-
tential.

It’s obvious what happens on the magnetic side: the superpotential (7.9) becomes
W = M qq + U (pp + 1)
where we’re not being careful about including coefficients, dimensionful or otherwise,
for these various terms. The equation of motion for U now means that p ̸= 0 in
the ground state. This induces a Higgs mechanism and breaks the magnetic gauge
symmetry Spin(8) → Spin(7) in such a way that the other chiral superfields q, that
previously transformed in 8v , now transform in the 8 spinor representation of Spin(7).

This gives us a new duality: the electric chiral theory with superpotential (7.10) is
dual to Spin(7) gauge theory with N +4 chiral multiplets in the spinor representation 8,
coupled to singlets through W = M qq. (This is actually the original “Pouliot duality”.)
The magnetic theory is now infra-red free for any N ≥ 11.

This version of Pouliot duality has a surprising feature. Our original SU (N ) theory
was a chiral gauge theory. But its Spin(7) dual is non-chiral! In particular, for N ≥ 11,
the chiral SU (N ) theory flows in the infra-red to the non-chiral Spin(7) theory. There
is a lesson in this: the question of whether or not a theory is chiral depends on the
energy scale at which you look. It is not a property that is preserved under RG.

– 222 –
Another Deformation
Alternatively, we could give an expectation value to U = det S. On the electric side,
this gives a mass to the spinor p, allowing us to integrate it out. We’re left just with
SO(8) gauge theory coupled to N +4 chiral multiplets in the 8v , still, of course, coupled
to the superpotenial W = M qq. (I’m ignoring global issues of the gauge group here.)

What happens on the original electric side? We give an expectation value to the
symmetric S ̸= 0. This breaks SU (N ) → SO(N ), so we’re left with an SO(N ) gauge
theory coupled to N + 4 fundamental chiral multiplets. The claim is that this is dual
to the SO(8) theory above.

In fact, this is part of the SO(N ) Seiberg dualities which, in general, relate an SO(Nc )
theory to an SO(Nf − Nc + 4) theory.

7.2.3 Briefly, the Konishi Anomaly

There’s a loose thread hanging from our discussion of Pouliot duality. The electric
theory includes two baryon operators

B = Q̃N and B ′ = (Q̃S)N

We haven’t yet seen what they map to on the magnetic side. Happily, the Spin(8)
theory also contains two baryon operator which, schematically, take the form

b = q 4 p2 and b′′ = q 8

Here the q 8 in b′′ are contracted with an epsilon tensor. We need a little group theory to
explain how b is put together. The vectors q 4 combine in an anti-symmetric fashion into
35s +35c and the latter is contracted with the two spinors which combine symmetrically
into 35s so that the whole thing is a singlet of Spin(8).

It seems reasonable to think that these operators might map into each other under
duality. To see this, we can check the flavour and R-symmetry charges. We have

F [B] = −N (N + 2) and R[B] = N

4N
F [B ′ ] = 2N and R[B ′ ] =
N +2
and

F [b] = −N (N + 2) and R[b] = N

4(N − 2)
F [b′′ ] = 4N and R[b′′ ] =
N +2

– 223 –
It’s close but, sadly, no cigar! First, it’s clear that under the duality we should match
b ←→ B
But while the flavour charge of B ′ and b′′ agree, their R-charge does not! What’s going
on?

In fact, there is a subtlety in this duality that didn’t rear its head in our previous
examples. To fully understand the structure of chiral superfields, we should include one
further field from each theory, each of which involves the chiral superfield that houses
the field strength. We call this Wα for the electric theory and W̃α for the magnetic
theory. Then consider
B ′′ = (Q̃N −4 S N −2 )Wα W α and b′ = q 4 W̃α W̃ α
If we use the fact that R[W 2 ] = R[W̃ 2 ] = 2, we find F [B ′′ ] = F [b′′ ] and R[B ′′ ] = R[b′′ ]
and F [b′ ] = F [B ′ ] and R[b′ ] = R[B ′ ]. So this solves our matching problem: the baryons
on one side are paired chiral fields that include the field strength of the other
b′ ←→ B′
b′′ ←→ B ′′
But this also opens up a whole can of worms! Why are we suddenly including the field
strength in the story? Or, said differently, why didn’t we include the field strength in
Section 6 when discussing SU (N ) SQCD?

The answer to this is a little subtle. Here I don’t give all the details, but sketch the
basic idea. It turns out that one can derive an equation in SQCD that, for each chiral
multiplet, reads
∂W 1
D̄2 (Q† Q) = Q + 2 Tr Wα W α
∂Q 8π
This equation is known as the Konishi anomaly and is the supersymmetric version of
the chiral anomaly which says that a rotation Q → eiαQ results in a shift of the theta
angle. It tells us that, at least as far as the chiral ring is concerned, the operator
Tr Wα Wα can be replaced by Q∂W/∂Q, so we’re not missing anything if we neglect it.

However, in other theories there are a number of these additional chiral multiplets,
dressed with Wα , that you need to include. This first rears its head in the duality for
SO(Nc ) theories (which we didn’t describe in these lectures notes, in part to duck this
particular issue). For the chiral duality that we’ve described above, it turns out that
you need to include the extra B ′′ and b′ (and, in fact, one further operator from each
theory that depends linearly on Wα or W̃α respectively).

– 224 –
7.2.4 Briefly, a-Maximisation
We’ve seen a few times in these lectures that many theories don’t have a unique R-
symmetry. Instead, we can always add any linear combination of other Abelian flavour
symmetries and this also provides a good candidate R-symmetry. This becomes an
issue only when we flow to an interacting SCFT, where the R-symmetry dictates the
dimension of chiral operators
3
∆[O] = RIR [O]
2
But for this to be useful, we need to know exactly what R-symmetry we’re dealing with
in the infra-red.

Happily, there is a simple prescription to determine this. This prescription, known

as a-maximisation16 , is straightforward to state but somewhat harder to prove. Here
we just give the statement, dressed with a little context.

First, in any conformal field theory the trace of the stress tensor necessarily vanishes:
⟨T µµ ⟩ = 0. At least, this is true in flat space. But if the theory is placed on a curved
manifold, there is a so-called trace anomaly and we get
c a ∗
⟨T µµ ⟩ = 2
Cµνρσ C µνρσ − Rµνρσ ⋆ Rµνρσ
16π 16π 2
where Cµνρσ is the Weyl tensor and ∗ R is the dual of the Riemann tensor. (We proved
the analogous statement for 2d CFTs in the lectures on String Theory.) The two
coefficients a and c are known as central charges and provide a way to characterise the
CFT.

Of the two, a is the more interesting. First, it can be proven that a always de-
creases under RG flow. Second, in superconformal field theories it turns out that a is
determined by the R-charge
3 X 3
a= 3RIR − RIR
32 fermions

where the sum should be taken over left-handed Weyl fermions.

16
It was first derived by Ken Intriligator and Brian Wecht. After landing a faculty job in London,
Brian moved on to an infinitely more interesting career as a ninja, fighting dinosaurs with lasers and,
ultimately, finding a much larger audience for his derivation of a-maximisation.

– 225 –
 Once again, it’s important that we use the right R-symmetry RIR when computing
the central charge a. However, the beauty of this calculation is that it gives us a way
to figure out what the right central charge is. Suppose that we have a collection of
candidate central charges in the UV, parameterised by some coefficients α as in (7.8).
For each of these we can compute the would-be central charge
3 X
a(α) = 3R(α)3 − R(α)
32 fermions

The R-symmetry that appears in the superconformal algebra turns out to the one that
maximises the value of a. This gives a simple way to compute RIR and, therefore, the
dimensions of chiral operators in the SCFT.

7.3 Dynamical Supersymmetry Breaking

All the gauge theories that we’ve discussed so far have supersymmetric vacua with
vanishing energy. In some cases these vacua are pushed off to infinity by a runway
potential, but we can always rescue them by giving masses to the matter multiplets,
bringing them in to finite distance. One might wonder: do all supersymmetric gauge
theories have supersymmetric ground states? Or is it possible that some gauge theories
spontaneously break supersymmetry, with a ground state that has energy E > 0?

We already met some models that break supersymmetry back in Section 3.4. There,
we worked only with chiral multiplets and the game was to cook up a superpotential
which for which no critical points exist. In searching for gauge theories that break
supersymmetry, the game is similar. The difference is that now there is the option for
the superpotential to be generated by quantum effects. Such theories are said to break
supersymmetry dynamically.

Where should we look for dynamical supersymmetry breaking? An obvious obstacle

is the Witten index. This is non-vanishing for super Yang-Mills theory with any gauge
group. (It is given by the dual Coxeter number and is listed for all gauge groups in
Table 3.) If we add matter in any vector-like representation, we can always give it
a mass and reduce to super Yang-Mills with its non-vanishing Witten index. This
suggests two places to look for supersymmetry breaking.
• We could consider chiral gauge theories in which it’s not possible to give the
matter mass.

• Alternatively, we could consider gauge theories with a quantum moduli space of

vacua for which the Witten index is ill-defined. It may then be possible to deform
these theories in some other way that doesn’t involve giving masses.

– 226 –
In this section, we give two examples of dynamical supersymmetry breaking, one of
each kind.

7.3.1 The SU (3) × SU (2) Model

One of the simplest chiral gauge theories we can write down is based on the gauge
group
G = SU (3) × SU (2)
We introduce a collection of four chiral multiplets, with quantum numbers given by

SU (3) SU (2) U (1)Y U (1)R U (1)A U (1)′A

Q 3 2 1 −1 1 1
Ũ 3 1 −4 0 1 0
D̃ 3 1 2 0 1 0
L 1 2 −3 3 0 1
Λ73 1 1 1 0 −4 0
Λ42 1 1 1 0 0 −4

We’ve also included both non-anomalous and anomalous U (1) symmetries in this table.
Classically there is a U (1)4 symmetry, but quantum mechanically only a U (1)2 survives.
The anomalous U (1) symmetries are U (1)A and U (1)′A , as shown by the transformation
of the strong coupling scales. The exponents in these strong coupling scales can be
traced to the one-loop beta functions, which are
SU (3) : b0 = 9 − 2 = 7 and SU (2) : b0 = 6 − 2 = 4
If you know the smallest amount of particle physics, these quantum numbers should look
very familiar! They are the representations of the quarks and leptons of the Standard
Model. (The right-handed electron is missing.) The symmetry U (1)Y coincides (up to
a normalisation) with the hypercharge symmetry of the Standard Model, here a global
rather than gauge symmetry.

It’s curious that, as we shall see, this theory dynamically breaks supersymmetry al-
though it doesn’t seem particularly useful for real-world purposes: the MSSM must
include the Higgs fields (which, of course, also sit in chiral multiplets). Various phe-
nomenological constraints means that supersymmetry breaking is thought to take place
in an entirely different sector before being communicated to the Standard Model by
so-called “messenger” fields. Here we study the theory simply to get a feeling for what
chiral gauge theories do.

– 227 –
 First, the classical moduli space. As we’ve seen, this is parameterised by the gauge
invariant holomorphic monomials. For our current theory, there are three:

Y1 = Ũ QL , Y2 = D̃QL , Z = Ũ Q D̃Q

where the SU (2) gauge indices are contracted with an ϵab symbol in each. These have
R-charge R[Y1 ] = R[Y2 ] = 2 and R[Z] = −2. This means that we can add a tree level
superpotential that preserves the R-symmetry,

Wtree = λD̃QL = λY2

with λ a (classically) dimensionless constant. This superpotential is renormalisable and

also preserves U (1)Y .

The superpotential Wtree lifts the vacuum moduli space. To see this, note that the
critical point requires
∂Wtree
=0 ⇒ D̃Q = 0 ⇒ Y2 = Z = 0
∂L
and
∂Wtree
=0 ⇒ QL̃ = 0 ⇒ Y1 = X 2 = 0
∂ D̃
This means that if there is supersymmetric ground state then it necessarily sits at the
origin of moduli space where the theory is strongly coupled.

Now let’s turn to the quantum dynamics. For λ suitably small, we can ignore the
tree-level superpotential and import our results from Section 6. Things are easiest if
we assume that |Λ3 | ≫ |Λ2 | so that the SU (3) dynamics becomes strong first. In this
case we have SU (3) with Nf = 2 flavours which, we know, is the situation where a non-
perturbative superpotential is generated by instantons. Adding this to our tree-level
superpotential gives

Λ73
W = λY1 + (7.11)
Z
The quantum generated superpotential gives a runaway that pushes the ground state to-
wards infinity. Meanwhile, we’ve already seen that the tree level superpotential pushes
the ground state towards the origin. The net result is shown in Figure 16, with a
ground state that sits at energy E > 0 and hence breaks supersymmetry.

– 228 –
Figure 16. The tree level superpotetnial, shown in green competes with the dynamically
induced superpotential, shown in red. The sum of the two, shown in blue, has a minimum at
E > 0 and so breaks supersymmetry.

The above analysis was very quick. You might wonder if perhaps one can play off
the two contributions to find a minimum at zero energy after all. In fact there’s a cute
argument that say this can’t happen. Here’s why. First note that each of Y1 , Y2 and
Z carry non-zero R-charge. Wherever the minimum of (7.11) sits, one of these must
get an expectation value and so R-symmetry is broken with a corresponding Goldstone
mode called an R-axion. This is a compact scalar. If supersymmetry is unbroken,
then there must be another non-compact, massless scalar that joins with the R-axion
to form the lowest component of a chiral multiplet. Usually such non-compact scalars
take us out along the moduli space. But we’ve seen that the moduli space is lifted by
the tree-level superpotential, so no such massless scalar exists and supersymmetry is
necessarily broken.

We could be more precise, finding the minima of the potential in terms of the funda-
mental fields but this is a little fiddly. However, there’s one feature that is important.
From the form of the superpotential (7.11), we would expect the expectation value v
of the fundamental fields to scale as
Λ3
v∼
λ1/7
This means that for λ ≪ 1, we have v ≫ |Λ3 | ≫ |Λ2 |. As long as the expectation values
break the gauge group completely the theory is weakly coupled and we can compute
everything reliably. In particular, we are free to use the canonical Kähler potential in
this regime.

– 229 –
7.3.2 The Quantum Moduli Space Revisited
As a second example of supersymmetry breaking, we take a theory that has a moduli
space of vacua, and hence an ill defined Witten index. We then deform it in such a
way that supersymmetry is broken.

To this end, consider SU (2) gauge theory coupled to four chiral multiplets Φi , i =
1, . . . , , 4, each in the fundamental representation. The gauge invariant operators consist
of six mesons

M ij = Φia Φjb ϵab

(This is the Sp(1) language of Section 7.1. In the SU (2) language of Section 6, both
mesons and baryons are housed in the 4 × 4 matrix M ij = −M ji .)

Classically, the mesons obey the constraint Pf M = 0 where the Pfaffian is defined
by

Pf M = ϵijkl M ij M kl

We now add six singlet fields Sij = −Sji to our original theory. These couple to the
original fields through the tree-level superpotential

Wtree = λSij Φi Φj

This lifts the moduli space parameterised by M which must take value M = 0, but the
theory retains a classical moduli space, parameterised by the expectation values of Sij .

Now we turn to the quantum theory. We know from our discussion in Section 6.3
(or from Section 7.1) that, before adding the singlets, the quantum moduli space is
deformed in the quantum theory and becomes Pf M = Λ4 . The superpotential of our
theory with the singlets is now

W = λSij M ij + X Pf M − Λ4

with X a Lagrange multiplier field. But it’s clear that the equations of motion of
X and of S cannot be simultaneously satisfied: therefore this simple model breaks
supersymmetry17 .
17
This model was first proposed by Izawa and Yanagida and Intriligator and Thomas.

– 230 –
 In fact, we should be a little more careful. This theory has a flat direction, albeit one
with energy E > 0. To see suppose that we place ourselves far out along the classical
direction S ̸= 0. This gives the original quarks Φ a large mass and so they can be
integrated out. The low-energy superpotential is

1/2
Weff ∼ λ2 Λ4 Sij S ij (7.12)

The behaviour on S follows on symmetry grounds, including the fact that R[S] = 2.
The behaviour on the couplings can be deduced from matching scales after integrating
out the quarks, with Λ6new = Λ4old m2 = Λ4old λ2 S 2 and the superpotential is simply
Weff = Λ3new as in (6.12).

If we assume a canonical Kähler potential for S, then the superpotential (7.12) results
in the potential

Sij S † ij
V ∼ |λΛ2 |2
|Sij S ij |

As we vary the phases of different Sij components, this potential diverges in some
directions, but also has flat directions in which V ∼ |λΛ2 |2 .

Because we’ve broken supersymmetry, these flat directions will surely be lifted by
quantum effects. (They are sometimes called pseudo-flat directions for this reason).
The concern is that these quantum effects might lead to a runaway behaviour, so that
rather than breaking supersymmetry we instead have a theory with no good ground
state. Integrating out the quarks gives a logarithmic correction to the Kähler potential
for S, along the lines of (3.38). You need to be careful about the signs, but it turns out
that this causes the potential to grow as we move out along the flat directions. The
ground state is pushed towards smaller values of S and breaks supersymmetry.

Because this model is vector like, we could add masses for the quark fields. What
then happens? To see this, it’s actually useful to add to mass terms: one for the
quarks and another for S. After the quantum modification of the moduli space, the
superpotential becomes

Weff = λSij M ij + mij M ij + m̃Pf S + X(Pf M − Λ4 )

Now there are supersymmetric ground states! They sit at

1/2 1/2
Λ4 Λ4
 
ij ijkl mij
M ∼ϵ mjk and Sij ∼
Pf m m̃ Pf m

– 231 –
The square roots allow for two different signs, and these are the two expected super-
symmetric ground states since Tr(−1)F e−βH = 2 for SU (2) super Yang-Mills. But we
can also see what happens as the masses are removed. As mij → 0, we get a smooth
limit for M ij (because Pf m ∼ m2 ). But as m̃ → 0, the supersymmetric ground state
decouples as S → ∞. Naively, one might think that this leads to runaway behaviour
(as it does, for example, for SU (Nc ) with Nf < Nc flavours). The novelty in the cur-
rent case is that there is an infinite barrier between the supersymmetric ground state
at S → ∞ and the supersymmetry breaking ground state at finite S. If you like, the
maximum of this barrier must also have moved off to infinity as m̃ → 0.

It is straightforward to construct generalisations of this model using other theories

that exhibit a quantum deformed moduli space, including SU (Nc ) with Nf = Nc and
Sp(Nc ) with Nf = Nc + 1.

– 232 –