Seiber Go Logy

Notes on Seibergology
EM-like duality SUSY and applications
Flip Tanedo
Laboratory of Elementary Particle Physics,

Cornell University, Ithaca, NY 14853, USA
E-mail: pt267@cornell.edu
This version: May 7, 2013
Abstract
This is a set of unfinished LATEX’ed notes on Seiberg duality and related topics in phe-
nomenology. It subsumes an older set of notes on metastable supersymmetry breaking.
Corrections are welcome, just don’t expect me to get around to it any time soon.
Contents
1 Introduction 1
1.1 Pre-requisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 A selected non-technical history . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
N = 1 Duality in SQCD 2
2 Nonperturbative SUSY QCD 2
2.1 Effective Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Gauge theory facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Reminder: basic SUSY gauge theory facts . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Moduli space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 The holomorphic gauge coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 The NSVZ β-function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 The Konishi Anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Symmetries of SQCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 F < N : the ADS superpotential 19
3.1 Holomorphic scale as a spurion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 The ADS Superpotential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 ADS: F = N − 1, instantons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Deforming SQCD: Higgsing a squark . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Deforming SQCD: mass perturbations . . . . . . . . . . . . . . . . . . . . . . . . 26
3.6 The coefficient of the ADS superpotential . . . . . . . . . . . . . . . . . . . . . . . 27
3.7 Run, run, runaway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.8 Gaugino condensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.9 Integrating in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.10 Relatied topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4 F = N and F = N + 1: Some special cases 40

4.1 ’t Hooft Anomaly Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Moduli Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 N = F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4 The ADS superpotential from nothing . . . . . . . . . . . . . . . . . . . . . . . . 45
4.5 F = N + 1: s-confinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5 F > N + 1: Seiberg Duality 50

5.1 The Banks-Zaks Fixed point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Some facts about superconformal theories . . . . . . . . . . . . . . . . . . . . . . 53
5.3 32 N < F < 3N : the conformal window . . . . . . . . . . . . . . . . . . . . . . . . 55
5.4 Seiberg Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.5 Checks of Seiberg Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.6 N + 1 < F < 3N/2: compositeness . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.7 RG motivation for Seiberg duality . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.8 The importance of the scale µ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.9 Remarks on the meaning of the duality . . . . . . . . . . . . . . . . . . . . . . . . 63
6 The phase structure of SQCD 64
7 Siblings and Cousins of Seiberg Duality 64
8 Crouching Seiberg, Hidden Gauge Group 64

8.1 Mini review: nonlinear sigma model . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.2 Hidden gauge group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9 Relation to AdS/CFT 65
10 Some current directions of interest 66
11 Unsorted ISS and SUSY breaking notes 66

11.1 Questions originating from Yael’s chiral ISS model . . . . . . . . . . . . . . . . . . 66
2
N = 2 Duality 67
12 Seiberg-Witten 67
13 Gaiotto Dualities 68
Breaking SUSY 68
14 SUSY Breaking: history 68
14.1 SUSY breaking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
14.2 Dynamical supersymmetry breaking . . . . . . . . . . . . . . . . . . . . . . . . . . 68
14.3 What makes SUSY-breaking non-generic? . . . . . . . . . . . . . . . . . . . . . . 69
14.4 Metastable SUSY breaking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
14.5 Problems with gaugino masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
15 Dynamical Supersymmetry Breaking 72

15.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
15.2 Toy example: SUSY QM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
15.3 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
15.4 The 3-2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
15.5 The ITIY Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
15.6 Tools for Noncalculable Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
16 Gauge Mediated SUSY Breaking 75

16.1 SUSY Breaking refresher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
16.2 Set up and features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
16.3 EWSB and the µ-Bµ problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
16.4 Direct, semi-direct, extraordinary, and general . . . . . . . . . . . . . . . . . . . . 84
17 The Nelson-Seiberg R-symmetry theorem 86
18 The Komargodski-Shih Gaugino Mass Theorem 88

18.1 Basics of SUSY breaking in Wess Zumino Models . . . . . . . . . . . . . . . . . . 88
18.2 Tree-level SUSY and R-symmetry breaking . . . . . . . . . . . . . . . . . . . . . . 92
18.3 Application to Gaugino Masses and Model Building . . . . . . . . . . . . . . . . . 93
19 The Landau Pole Problem 95
20 ISS: Metastable SUSY Breaking 95

20.1 Summary in words . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
20.2 Macroscopic Model I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
20.3 Macroscopic Model II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
20.4 Dynamical Realization: the ISS model . . . . . . . . . . . . . . . . . . . . . . . . 106
20.5 Immediate directions beyond the basic ISS model . . . . . . . . . . . . . . . . . . 106
20.6 Extensions and related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3
Appendix 107
A Notation and Conventions 107
A.1 Field labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.2 Spacetime and spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.3 Superfields and superspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A.4 SUSY NLΣM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A.5 2-component plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
A.6 OLD Notation and Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B The Coleman-Weinberg Effective Potential 114

B.1 Quantum Mechanical Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
B.2 Diagrammatic Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
B.3 Functional Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
B.4 Integrating out fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.5 An illustrative example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.6 Cutoff dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
C Phases of Gauge Theories 130
D Review of Anomaly Cancellation 130

D.1 Overview and background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
D.2 The Anomaly Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
D.3 Cancellation of gauge anomalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
D.4 Comments on global anomalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
E Analytic Continuation into Superspace 134

E.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
E.2 Preliminary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
E.3 Gaugino mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
E.4 Analytic Continuation: Wess-Zumino soft terms . . . . . . . . . . . . . . . . . . . 139
E.5 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4
1 Introduction
These are personal notes originally based on the spring 2009 lectures by Csaba Csáki at Cornell
University. They have been augmented with material following the author’s interests in the field
and draw heavily from some of the very excellent ‘Seibergology’ review literature:
• Intriligator and Seiberg’s lectures on SUSY breaking [1] and electromagnetic duality [2]
• Strassler’s lecture notes on SUSY gauge theory [3] and the duality cascade [4].
• Terning’s book on supersymmetry [5] and the accompanying slides which are available online.
Additionally, there are several elements taken from various lectures, review articles, talks, current
literature, and discussions with colleagues. I have tried to give credit and links to further literature
where appropriate. To be clear, this document includes no new results and the pedagogical
approach is an amalgamation from other sources. If unlisted, it is fair to assume that material
and the style of presentation came from one of the aforementioned references. All I did was
interpolate between several sources to weave a narrative which makes sense to me. Comments,
constructive criticism, and corrections are especially welcome.
1.1 Pre-requisites
The reader is expected to already be familiar with supersymmetric gauge theories at the level of
an introductory graduate-level course, e.g. [6]. For more foundational reading on supersymmetric
gauge theories there is now a plethora of review articles and recorded lectures available. As a
rule of thumb any set of lectures after 1996 should review Seiberg duality while any lectures after
2006 should at least mention metastable vacua. Recent reviews which mention the metastable
SUSY-breaking program include Dine’s 2008 Cargese lectures [7] (see also his more recent review
[8]) and Shirman’s 2008 TASI lectures [9]. One may find further references in those papers.
The textbooks by Terning [5] and Dine [10] also mention useful material in modest depth. The
multimedia-inclined are encouraged to view Brian Wecht and Nathan Seiberg’s lectures at the
Isaac Newton Institute’s 2007 “Gauge Fields and Strings” workshop1 , Seiberg’s lectures at PiTP
20102 , or Csaba Csáki’s lectures on supersymmetry3
Further references to appropriate pedagogical or otherwise important literature will be men-
tioned as appropriate in this document.
1.2 A selected non-technical history

[To do: This section to be written.]
1
http://www.newton.ac.uk/programmes/SIS/sisw02p.html
2
http://video.ias.edu/pitp-2010
3
Available to the Cornell LEPP theory group.
1
1.2.1 Dynamical SUSY breaking
1.2.2 N = 1 electromagnetic duality and beyond
1.2.3 Metastable SUSY breaking
N = 1 Duality in SQCD
2 Nonperturbative SUSY QCD
The basic tools we will need come from supersymmetric QCD. We shall now review the key aspects
of supersymmetric SU (N ) gauge theory with F flavors, in particular its nonperturbative for various
values of F and N . Our review will be loosely based on Csáki’s lectures on Beyond the Standard
Model physics4 , which is in turn based on Terning’s textbook [5]. Additional review articles and
lectures include Intriligator and Seiberg’s lectures on SUSY gauge theory and electromagnetic
(i.e. Seiberg) duality [2], Wecht’s lectures at the Newton Institute’s Gauge Fields and Strings
program5 , Argyres’ Cornell lectures on supersymmetry6 , and Peskins TASI lectures on duality in
super Yang-Mills theories [11]. Those unfamiliar with this topic are encouraged to read some of
these references for a proper pedagogical introduction to the subject.
2.1 Effective Actions

We must make an important distinction between two types of effective action. While we don’t
usually care about the exact sense in which an effective action is ‘effective,’ both the 1PI and
Wilsonian effective actions play a key role in SUSY and it is important to be clear which sort of
effective action one is working with. The importance of the distinction between these two was first
elucidated by Shifman and Vainshtein when they explained the apparent contradiction between
the nonrenormalization theorem for the superpotential and their exact-to-all orders (NSVZ) beta
function [12]. Some discussion can also be found in Intriligator and Seiberg’s SUSY electromag-
netic duality lectures [2] and Burgess’ introduction to effective field theory [13].
The 1PI action, Γ, comes from integrating out all quantum effects such that the classical
equations of motion from the 1PI action Γ yield the complete equations of motion including
quantum effects. This object is the generator of 1PI diagrams, i.e. the so-called ‘skeleton diagrams’
that are tree-level but encode multi-loop processes. The momentum-independent part of the 1PI
action is the rather-important-for-this-paper Coleman-Weinberg effective potential. The 1PI
action (and the Coleman-Weinberg potential) by definition take into account all loop effects so
that any actual calculation of these objects must be done as a loop expansion. More formally, the
1PI effective action is the Legendre transform of the classical action with respect to sources and
background fields. The NSVZ ‘exact to all orders’ beta function for the real gauge coupling g is
obtained from the 1PI action.
4
Spring 2009, Cornell University.
5
Notes and recordings: http://www.newton.ac.uk/programmes/SIS/sisw02p.html.
6
http://www.physics.uc.edu/~argyres/661/index.html
2
The Wilsonian action, on the other hand, comes from integrating out heavy fields and the
high-momentum modes of light fields. This is the action that one obtains from the [Wilsonian]
renormalization group flow to lower energies. Unlike the 1PI action, the Wilsonian action must still
be treated quantum mechanically, i.e. one still has to perform the path integral and it is inherently
a ‘theory with a cutoff.’ The one-loop exact (‘Seiberg’) beta function for the holomorphic gauge
coupling τ is obtained with from the Wilsonian action.
Because all quantum excitations are integrated out, 1PI action contains contributions from
massless particles running in loops (∼ log k) and can thus have infrared divergences. A tangible
example of this can be seen in the chiral Lagangian for pions. These “IR ambiguities” can lead
to “holomorphic anomalies.” In other words these divergences tell us that our theory is missing
something important. The Wilsonian action does not have any problems with massless particles
since it only integrates out heavy modes.
2.2 Gauge theory facts

This section is based on Gutowski’s lecture notes, which uses anti-Hermitian genera-
tors. I’ve tried to change these int Hermitian generations, I may have missed a few
places. [Even better: instead I should use Hugh Osborn’s notes... they’re slightly
better on gauge theories.]
The fundamental covariant derivative Dµ is Dµ = ∂µ +ieAµ , so that Dµ ϕ transforms in the same

way as the fundamental field ϕ. This imposes that ieA′µ = gieAµ g −1 − ∂µ gg −1 . More generically,
the covariant derivative associated with a field φ in representation r of the gauge group is
Dµ φ = ∂µ φ + d(ieAµ )φ, (2.1)
where d(ieAµ ) is the associated representation of the Lie algebra on r. In other words, if φ →
D(g)φ where D(g) is the transformation g in the r representation of the gauge group, then Dµ φ →
D(g)Dµ φ. This requires two lemmas to prove,
d(gvg −1 )D(g) = D(g)d(v) (2.2)
( )
dD(g) −1 dg −1
Dg = d g . (2.3)
dt dt
These can be proved by writing g = eth where t is a parameter which we set to t = 1 after taking
appropriate derivatives. See Gutowski or Hugh Osborn’s notes for details. The result for the
adjoint covariant derivative is
Dµ ψ = ∂µ ψ + (ad ieAµ )ψ = ∂µ ψ + [ieAµ , ψ]. (2.4)
For an SU(N ) gauge theory with representation r generators Tra the (quadratic) Casimir Cr (r)
and the Dynkin index T (r) are given by [5]
C2 (r)δnm = (Tra )m a ℓ
ℓ (Tr )n (2.5)
T (r)δ ab = (Tra )m b n
n (Tr )m . (2.6)
These can be understood diagrammaticaly.
3
T (r)δ ab = . C(r)δnm = .
a b n m
In fact, using this diagrammatic interpretation, we can close the external lines of each diagram to
obtain an equivalent two-loop diagram. This leads to the relation d(r)C2 (r) = d(Ad)T (r). We
use the standard normalization that
d(□) = N d(Ad) = N 2 − 1 (2.7)

1
T (□) = T (Ad) = N (2.8)
2
N2 − 1
C2 (□) = C2 (Ad) = N. (2.9)
2N
Comments on gauge redundancy. Gauge (local) symmetries are different from global
symmetries: they are redundancies in the way that a system is described. For example,
for a U(1) gauge theory the photon has only two physical polarizations but is described by
a four-component field. The longitudinal polarization is removed due to the photon being
massless, but the additional degree of freedom that must be removed is precisely the gauge
redundancy—we are free to add to the vector potential (photon) any gauge transformation
since it is projected out in any physical quantity. Note, somewhat subtly, that there is also a
global component of a gauge transformation which gives the current by which the gauge fields
couple to matter fields. Finally, this picture of gauge redundancy is much more subtle when
this ‘symmetry’ is broken in the Higgs phase. It is not technically the case that the gauge
symmetry ‘breaks’ to a smaller subgroup as in the case of a global symmetry—though the
end result is the same. What technically happens is that the parameterization of the order
parameter (Higgs) field introduces an additional redundancy (precisely the subgroup that one
‘breaks’ to) while giving masses to the heavy gauge fields. For a good discussion of this last
point, see chapter 8 of the QFT textbook by Banks [14].
2.3 Reminder: basic SUSY gauge theory facts

Let us quickly remind ourselves of some aspects of a supersymmetric gauge theory. A good
refresher is Section 4 of the latest version (v > 6) of Stephen Martin’s SUSY Primer [15], which
covers superspace techniques.
a a
Recall that under a gauge transformation, a chiral superfield Φ transforms as eΛ T Φ, where
Λ is a chiral superfield parameterizing the gauge transformation. The canonical Kähler potential
Φ† Φ is clearly not gauge invariant and doesn’t include the correct gauge covariant derivative terms
which couple the matter fields to the vector fields. Using Terning’s conventions [5], the Kähler
potential and vector superfield transform as,
K = Φ† egT
aV a
Φ V a → V a + Λa + Λa† + O(V a Λa ). (2.10)
4
More generally, the gauge-invariant Kähler potential takes the form K = K(Φ† , egT V Φ) and the
transformation of the vector superfield obeys
aV a a Λa† aV a a Λa
eT → eT eT eT . (2.11)
In writing the gauge parameter as a chiral superfield we see explicitly that supersymmetry also
enlarges that gauge symmetry. It is conventional to work in the Wess-Zumino gauge in which
the vector superfield is restricted to the vector, gaugino, and auxiliary D term. This super -gauge
choice breaks supersymmetry (by projecting to a subspace of the supermultiplet) but preserves
the usual gauge redundancy of a non-supersymmetric quantum field theory.
Gauge transformations of a vector superfield. It is worth remarking that even through

the entire vector supermultiplet is in the adjoint representation of the gauge group, the defini-
tion of a gauge transformation has also been supersymmetrized. That is to say that the gauge
transformation parameter is itself a superfield with different components. Because of this, the
different components of a vector multiplet transform differently under a gauge transformation.
In Wess-Zumino gauge this can be seen from the different gauge transformation properties of
the vector versus the gaugino. While the gaugino transforms precisely as one would expect for
a matter field in the adjoint representation, the vector plays a special role as the connection
of the gauge redundancy and thus transforms with an additional derivative term. (One could
have also argued this based on Lorentz indices.)
When writing down a supersymmetric Lagrangian we may make use of F terms and D terms
since these are invariant under SUSY transformations modulo total derivatives which vanish in
the action. Of course, the superfields which furnish the F and D terms are generally not the
same as the matter and vector content of the theory. Instead, one must form from these the
appropriate chiral and vector superfields which are gauge invariant and then take the F and D
terms (respectively) of these products as Lagrangian terms.
Note that one may alternately write even the D terms as F terms by using the identity
D Dα (θθ) = −4, where Dα is the SUSY covariant derivative. The conjugate identity also holds,
α
D̄D̄(θ̄θ̄), where we have suppressed the spinor indices of D̄. Thus the D-term of a vector superfield
V can be written as
∫ ∫ ∫
1 1 1( )
V |D = d θ V = −
4
d θ V D̄D̄(θ̄θ̄) = −
4
d4 θ(θ̄θ̄)D̄D̄V = − D̄2 V F , (2.12)
4 4 4
where we dropped total derivatives upon integration by parts. This is the origin of the funny form
of the gauge field strength superfield,
1
Wα = − D̄D̄Dα V, (2.13)
4
from which we can see that this is the same as a Kähler potential term for Dα V , which better
resembles the non-supersymmetric gauge field strength Fµν = ∂[µ Aν] .
5
2.4 Moduli space
We will see that in SYM theories one typically finds flat directions or moduli in the field space.
These are directions in the scalar fields with vanishing potential. When supersymmetry is broken
these tree-level flat directions are often lifted through quantum corrections, i.e. by the Coleman
Weinberg potential. In that case these directions are called pseudomoduli. We can now study
how these flat directions arise in super QCD. At the bare minimum this theory will have a D-term
potential since it is a gauge theory. It needn’t necessarily have any superpotential, so we will
ignore the superpotential contribution for now7 .
Recall that the D terms are the auxiliary fields of the vector superfield. Their value is fixed in
terms of the matter fields by virtue of the on-shell equation of motion which is algebraic. To see
this, recall that the D term pops up both in the canonical gauge-invariant Kähler potential for the
matter fields Φ† egT V Φ and in the field strength superpotential term 41 W2 . The Kähler potential
gives a piece which is linear in D, ∼ ϕ† gT a Da ϕ, while the superpotential term gives a piece which
is quadratic in D. As a reminder for the latter, recall that W ∼ λ + θD + · · · . Thus the equation
of motion sets Da ∼ ϕ† gT a ϕ. This is important for determining the moduli space of the theory.
2.4.1 Example: U (1)

As a toy example, let’s start with U (1) SQED. The squark potential then takes the form
e† Q)2 .
V = (Q† Q − Q (2.14)
e such that V is minimized. Modding out

The moduli space is given by the values of ⟨Q⟩ and ⟨Q⟩
by the gauge redundancy the moduli space is parameterized by a complex parameter a,
e = a.
⟨Q⟩ = ⟨Q⟩ (2.15)
Note that the different values of a parameterize inequivalent vacua of the theory. Compare this to
the the vacuum manifold of the usual Higgs mechanism where each point on the vacuum manifold
is physically equivalent to any other since those points only differ by a transformation by the
unbroken gauge generator.
For a ̸= 0 the gauge group is broken by the super Higgs mechanism. In the usual Higgs
mechanism a massless gauge field obtained a mass by ‘eating’ a scalar field which took the place of
the longitudinal mode. Here we promote the fields to superfields. The gauge superfield acquires a
mass |a| by eating an entire chiral superfield. It is easy to check that the usual Higgs mechanism
is subsumed in the super Higgs mechanism; consider the squark kinetic term,
∫
LQ,kin ∼ Dµ Q† Dµ Q + Dµ Q e
e† Dµ Q, (2.16)
upon giving the squarks vevs as in (2.15), this manifestly gives a mass term |a|2 to the photon. The
U (1) gauge invariance is broken and the photon acquires a mass that depends on the particular
point in the moduli space in which our theory happens to land.
7
In general the superpotential is highly constrained by the global symmetries of the theory.
6
There is an excellent discussion of this theory (and its cousins) in Strassler’s unorthodox review
[3]. We will highlight just one small part of the story. The D-term condition that minimizes (2.14)
tells us that
e 2 = 0,
|Q|2 − |Q| (2.17)
which, in words, says that gauge invariance (the D term conditions) imposes that the vevs of the
two squark fields should have the same magnitudes. In other words, instead of minimizing the
D-term potential we could as well have complexified the gauge redundancy. In other words, (2.17)
modded out by the usual U (1) gauge invariance is completely equivalent to modding out by
Q → αQ (2.18)
e → α∗ Q
Q e (2.19)
for α ∈ C. It is natural to parameterize our moduli space in terms of the vev of a gauge invariant
object, M ≡ QQ, e so that ⟨M ⟩ = a2 tells us everything about where we live on the moduli space.
(We can call M the modulus field.) The classical Kahler potential can also be written in terms of
M,
√
e† e−V Q
Kcl = Q† eV Q + Q e = 2 M †M . (2.20)
√
The ‘meson’ N is the effective gauge-invariant low-energy degree of freedom so that 2 M † M can
be understood to be the effective Kähler potential at low energies. The Kähler potential has a
singularity at X = 0. This is actually a ‘singularity’ and not just a ‘zero’ since the Lagrangian
comes from taking derivatives of the Kähler potential. Singularities are a signal of our theories
trying to tell us something. In this case the theory is telling us that there are new degrees of
freedom that should be in the effective action. We know exactly what these are: at ⟨X⟩ = 0 the
gauge symmetry is unbroken and there are massless gauge fields that cannot be excluded from
any ‘low energy’ action.
Reminder about the Kähler potential. As an aside, remember that the scalar potential
is given by
V (ϕi ) = Wj̄† (K −1 )ij̄ Wi . (2.21)
For a non-canonical Kähler potential this gives us a non-trivial Kähler metric, Kij̄ ̸= δij̄ . When
this is true it is possible that Wi = 0 is no longer sufficient to determine that SUSY is unbroken,
even in a theory of only chiral superfields. Consider, for example, an exercise from Strassler’s
lectures [3]. Given a theory of a single chiral superfield W = yΦ3 , we can define a new chiral
superfield Σ = Φ3 so that dW/dΣ ̸= 0 even when Σ = 0. For an excellent introductory
analysis of supersymmetry with a general Kähler potential, see the lectures by Argyres [16]
and Bilal [17]. Such theories are often called SUSY nonlinear sigma models (NLΣM) because
the Kähler potential forces the scalars to live on a complex manifold for which there is a very
geometric interpretation for the quantities that appear in the Lagrangian. (An analogous and
much simpler thing occurs in spontaneous symmetry breaking phenomena in ‘vanilla’ QFT,
but in those cases the manifold’s geometry is usually trivial.)
7
2.4.2 Case F < N
We assume that we have an SU (N ) theory with F < N flavors of ‘quarks’ ϕim in the fundamental
im
and ‘antiquarks’ ϕ in the anti-fundamental, where i = 1, · · · , F and m = 1, · · · , N . The D-term
for this theory are
∑ † † a
Da = ϕi T a ϕi + ϕi T ϕi
[i ]
∑ ( )in ∑ in ( † )
= ϕ† ϕim − ϕ ϕ (T a )nm .
im
i i
where we understand that the ϕs really mean ⟨ϕ⟩. We can define the N × N matrices Dnm and
n
D m,
( )in
Dnm = ϕ† ϕim
n in
( †)
D m=ϕ ϕ .
im
The condition that our D-term scalar potential vanishes (the ‘D-flatness condition’) then imposes
Da = 0. Since the generators T a are traceless, a solutions is
= α1
n
Dnm − D m
for some overall constant α. We may now use an SU (N ) gauge transformation to diagonalize the
D and D matrices. In the case F < N . Then from their definition we see that the D and D
matrices can have at most F nonzero eigenvalues. Thus they must take the form
D = diag(v12 , v22 , · · · , vF2 , 0, · · · , 0).

| {z }
(N −F )
Imposing D − D = α1 then imposes that D must also be a diagonal matrix. By the structure of
the zero and non-zero entries, we establish that the D-flatness condition can only be satisfied for
α = 0. From this we may write the solutions for our quark fields,
 
v1
 ... 
†  
⟨ϕ⟩ = ⟨ϕ ⟩ =  . (2.22)
 vF 
0 ··· 0
This spontaneously breaks SU (N ) → SU (N − F ). We observe the super Higgs mechanism at

work: we started with (2F ) × N chiral superfields and found a vev where we have a number of
broken generators
( )
(N 2 − 1) − (N − F )2 − 1 = 2N F − F 2 ,
8
each of which ‘eats’ a chiral superfield. The number of D-flat directions is then the number of
chiral superfields minus the number of broken generators,
(2N F ) − (2N F − F 2 ) = F 2 .
In the usual Higgs mechanism a massless vector eats a massless Goldstone boson. The exact same
effect occurs here, but due to supersymmetry the entire superfields must be included. Conceptually
the actual ‘coupling’ of the two superfields occurs between the massless vector component and the
Goldstone scalar, so one can think of the super Higgs mechanism as the joining of two superfields
due to the mixing of one of each of their components due to the regular Higgs mechanism. After
this feast, the remaining F 2 massless degrees of freedom are parameterized by an F × F meson
field,
jn
Mi j = ϕ ϕni . (2.23)
There is actually a more general theorem by Luty and Taylor [18] regarding this:
Theorem 2.1 (Luty-Taylor). The classical moduli space of degenerate vacua can always be pa-
rameterized by independent, holomorphic, gauge-invariant polynomials.
Proof. A heuristic proof is provided in Intriligator and Seiberg’s lecture notes on Seiberg duality
[2]. Setting the [D-term] potential to zero and modding out by the gauge group is equivalent
to modding out by the complexified gauge group. The space of chiral superfields modulo the
complexified gauge group can be parameterized by the gauge invaraint polynomials modulo any
classical relations. Then, Intriligator and Seiberg claim, this theorem follows from geometrical
invariant theory [19]. For a proper proof the reader is directed to the original paper by Luty and
Taylor [18].
2.4.3 Case F ≥ N
Before moving on let’s quickly cover the case F ≥ N . As before the D-flatness condition is still
D − D = ρ1, where ρ is some constant. We can again use the SU (N ) gauge degree of freedom
to diagonalize the D = (ϕ† )i ϕi and D matrices, though now they are of full rank and we may use
the D-flatness condition to write D in terms of the eigenvalues of D and the constant ρ,
   
|v1 |2 |v1 |2 − ρ
 ...   ... 
D=  D= . (2.24)
|vN |2
|vN | − ρ
2
This implies that we may write the ⟨ϕ⟩ and ⟨ϕ⟩ matrices as
 
  v1
v1  .. 
 ...   . 
⟨ϕ⟩ =  0  ⟨ϕ⟩ =  . (2.25)
 vN 
vn
0
9
Now we see that SU (N ) is completely broken at a generic point on the moduli space. This means
that we have (N 2 − 1) broken generators and thus [2N F − (N 2 − 1)] light D-flat directions in field
space. Again we parameterize these degrees of freedom by ‘gauge-invariant polynomials’,
jn
Mi j = ϕ ϕni (2.26)
Bi1 ···iN = ϕn1 i1 · · · ϕnN iN ϵn1 ···nN (2.27)
n1 i 1 nN iN
B i1 ···iN = ϕ ···ϕ ϵn1 ···nN . (2.28)
But wait! We find that we have too many degrees of freedom. That’s okay. We’ve forgotten to
impose the classical constraints to which these fields are subject,
j1 ···jN j
Bi1 ···iN B = M[i1 1 · · · M jNiN ] ∼ detM (2.29)
2.5 The holomorphic gauge coupling

Recall that the action for a vector superfield is conventionally written as
∫
1
L = d2 θ Waα Wαa + h.c. (2.30)
4
In this case, the gauge coupling g shows up in the kinetic term for the chiral superfields
∫
Lkin = d4 θ ϕ† egV T ϕ.
a a
(2.31)
We can redefine W by absorbing the coupling into the vector superfield,
Ve a = gV a , (2.32)
where we are no longer canonically normalized, but we are in some sense using a natural normal-
ization8 . Then the vector Lagrangian takes the form
∫
1
L = 2 d2 θ Waα Wαa + h.c. (2.33)
4g
We know that there are also non-perturbative effects that contribute to this Lagrangian, i.e. the
CP-violating ΘYM term. We can include this effect by defining a holomorphic gauge coupling9 ,
4πi ΘYM
τ≡ + (2.34)
g2 2π
8
This can be understood, for example, by considering the renormalization of the gauge coupling in ordinary
(non-supersymmetric) field theory. The only diagrams that contribute to this renormalization come from loop
contributions to the gauge field propagator. This tells us that g is ‘really’ something associated to the vector field,
not necessarily the coupling of the vector to fermions.
9
As noted in Appendix A, there seem to be many ‘standard’ normalizations for τ which differ by factors of, e.g.,
2π. I audibly groan every time I read a paper with a different normalization.
10
Our vector superfield Lagrangian finally takes the form
∫
1
L = d2 θ τ Waα Wαa + h.c. (2.35)
16πi
The canonically normalized Lagrangian is trivially obtained by multiplying through by g 2 . Since

τ only appears under the d2 θ of the superpotential, it is manifestly a holomorphic parameter.
Note that some people will write this as the imaginary part of the first term, accounting for the
appropriate factors of i; this seemingly odd notation is just the analog of saying 2ReX = X + h.c.
Recall the RG equations for the perturbative coupling,
dg b
µ =− (2.36)
dµ 16π 2
1 b
2
= − 2. (2.37)
g (µ) 8π
By the way, from (2.36) we should already know what the value of b is at any given scale:
b = 3Neff − Feff , (2.38)
where we have been very careful to write that this is the effective number of colors Neff and the
effective number of flavors Feff . This is important since in the following sections we’ll be exploring
the moduli space of SQCD with N colors and F flavors, but as we get away from the origin of the
moduli space the effective number of colors and flavors changes.
Applying (2.36) to τ , we may write
[ ( )]
1 |Λ|
τ1-loop = b log + iΘYM (2.39)
2πi µ
( )
b Λ
= log , (2.40)
2πi µ
where have defined the holomorphic dynamical scale
Λ = |Λ|eiΘYM/b . (2.41)
The real quantity |Λ| plays the role of ΛQCD from non-supersymmetric chromodynamics, but the
holomorphic quantity Λ is what will be very important for us. We can also invert the expression
to write
Λ = µe2πib/τ . (2.42)
Now we claim that this does not receive any further corrections within perturbation theory, i.e.
that (2.40) is the full perturbative expression.
Theorem 2.2. The holomorphic coupling is only perturbatively renormalized at one loop. It does,
however, receive non-perturbative corrections from instanton effects.
11
Proof. We’ve written the one-loop renormalization of g in Eq. (2.40). We now have to show that
this only gets corrections from instantons. The key will be to consider the ΘYM dependence. We
know that ΘYM is a term which multiplies an F Fe in the Lagrangian,
( )
e µνρσ 2
F F = 4ϵ ∂µ Tr Aν ∂ρ Aρ + Aν Aρ Aσ . (2.43)
3
This is a total derivative and has no effect in perturbation theory (as expected from a non-
perturbative instanton effect); in perturbation theory ΘYM is just a constant because it is a total
derivative. However, this term contributes to a topological winding number, n,
∫
ΘYM
d4 x F Fe = nΘYM , (2.44)
32π 2
∫ ∫
c.f. the usual index theorem. In the path integral dA exp (iS) ∼ dA exp (inΘYM ). Thus we
see that the ΘYM must be periodic in 2π, i.e. ΘYM → ΘYM + 2π must be a symmetry of the
theory. Under this transformation the dynamical scale goes as
Λ → e2πi/b Λ. (2.45)
This, in turn, affects the effective superpotential Weff = τ /(16πi)W2 through the dependence of
the holomorphic coupling on Λ,
( )
b Λ
τ= log + f (Λ, µ), (2.46)
2πi µ
where the first term is the one-loop result that we derived and the second term represents an
arbitrary function that includes any higher-loop corrections. Remember from (2.34) that we may
write τ in terms of ΘYM . Under the transformation of Λ in (2.45), we see from (2.41) and (2.34)
that τ → τ + 1. But we can also see that the expected shift is already saturated by the first term
on the right-hand size of (2.46). Since the first term already saturates the correct behavior, the
second term must be invariant under the transformation. We can then write out the second term
as
∑∞ ( )bn
Λ
f (Λ, µ) = an , (2.47)
n=1
µ
where the form is set by demanding weak coupling as Λ → 0 (we want the perturbative result
in this limit). Terms of this form, however, just represent instanton effects. Recall the instanton
action,
( )b
8π 2 Λ
Sinst = 2 ⇒ e Sinst
∼e
2πiτ
= . (2.48)
g µ
Thus instanton effects in SUSY gauge theories will always appear with a prefactor of (Λ/µ)b . Thus
we have the result that τ is only [perturbatively] renormalized at one-loop order.
12
One can also determine the instanton corrections. For example, Seiberg and Witten famously
found exact expressions for the an coefficients in N = 2 SYM. For review see, e.g., [20, 21, 22].
At this point you might want to brush up on your instantons. In a pinch, one can look over the
relevant chapter in Terning [5]. A well-written and more pedagogical treatment of instantons can
be found in the author’s A-exam10 . Comprehensive guides to calculations in supersymmetry can
be found in the notes by Shifman and Vainshtein (the S and V in NSVZ) [23] and a separate set
by Bianchi, Kovacs, and Rossi [24]. More general expositions can be found in Dine [10], Coleman’s
‘The uses of instantons’ in Aspects of Symmetry [25], Vandoren and van Nieuwenhuizen’s lectures
[26], Manton and Sutcliffe [27], Rajaraman [28], and a lecture from Michael Peskin’s 2005 course
on quantum field theory at Stanford University, 11 .
2.6 The NSVZ β-function

The first piece of ‘Seibergology’ that students of supersymmetry learn is the clever use of holo-
morphy (and asymptotics) to prove the non-renormalization of the superpotential. We also know
that the (holomorphic) gauge coupling is only renormalized at one-loop order. We know, however,
that this isn’t the whole story. Even though the holomorphic couplings in superpotential are not
renormalized, there’s nothing protecting the Kähler potential from running. The renormalization
of the Kähler potential then changes the canonical normalization of fields and hence changes the
physical couplings for interactions in the superpotential. With this in mind, now want the β
function for the canonically normalized gauge coupling. NSVZ is an all -loop expression for the β
function for τ . There will be two sources: the canonical normalization of chiral superfields and
that of vector superfields.
Let’s start with one of our favorite toy models of chiral superfields, the Wess Zumino model,
1 1
W = m̂Φ2 + λ̂Φ3 . (2.49)
2 3
As we have written holomorphic quantities with a hat to distinguish them from physical quan-
tities m and λ. At tree level m = m̂ and λ = λ̂, but at loop level we have to account for the
renormalization of the kinetic term K = ZΦ† Φ, leading to
Lkin = Z |∂ϕ|2 + Z ψ̄i∂ψ,

/̄ (2.50)
where Z depends on the parameters of the theory (including the renormalization point µ) via
( )
λ̂λ̂∗ Λ̂2
Z =1+c log . (2.51)
16π 2 µ2
It should be clear that this is true. This represents the wavefunction renormalization from loop
corrections to the two point scalar function. Consider the loop of internal fermions with Yukawa
couplings to the external scalars (the loop with an internal scalar and auxilliary field is related by
SUSY). The prefactor of |λ̂|2 /16π 2 is easy to read off. The logarithm should also be expected: we
10
http://www.lepp.cornell.edu/~pt267/files/documents/A_instanton.pdf
11
http://www.slac.stanford.edu/~mpeskin/Physics332/instantons.pdf
13
know that the contribution to the scalar mass term can be quadratically divergent in a general
QFT. The divergence of the kinetic term is two powers less and is thus only logarithmically
divergent.
To canonically normalize we have to rescale our field
Φ → Φ′ = Z 1/2 Φ (2.52)
and define physical parameters
m̂ λ̂
m= λ = 3/2 . (2.53)
Z Z
The important quantity describing the running of these physical parameters is the familiar anoma-
lous dimension,
∂ log Z
γ=− . (2.54)
∂ log µ
Why is it called an anomalous dimension? In a general (not necessarily supersymmetric)

theory, quantum (loop) effects generate corrections to the kinetic terms that are dimensionful,
L = z|∂ϕ|2 + · · · (2.55)
with [z] ̸= 0. Like any dimensionful parameter, we define a dimensionless parameter in terms
of the natural scale of the theory (the renormalization scale µ),
z
Z ≡ γ, (2.56)
µ
where γ is defined here to be whatever power is required to make Z dimensionless. The
Lagrangian can then be written as
L = Zµγ |∂ϕ|2 = |∂ϕ′ |2 + · · · (2.57)
after re-absorbing a dimensionless factor of Z 1/2 and a dimensionful factor of µγ/2 into the
definition of the field ϕ′ . This means that the field ϕ′ has now picked up an ‘anomalous
dimension’,
1
[ϕ′ ] = 1 + γ, (2.58)
2
where the 1 represents the tree-level dimension. The true scaling dimension of ϕ has changed
due to quantum effects. The formula for γ can be understood by further that noting we could
have written a β function for the dimensionless coupling Z, giving
∂ ln Z
βZ = −γZ ⇒ = −γ. (2.59)
∂ ln µ
These anomalous dimensions turn out to be very important for (super-)conformal theories.
Even in the neighborhood of a conformal theory, one can make use of ‘deep theorems’ in QFT
to give constraints on the anomalous dimension. See, for example, [29]. (See also Strassler’s
lectures for an appropriately unorthodox persepctive [3].) The meaning of the anomalous
dimension, by the way, really comes from the anomaly of dilatations. See Coleman’s Erice
lectures for an excellent discussion [25].
14
In this chiral theory, the anomalous dimensions encode everything there is to know about how
the renormalization of the physical couplings. As explained in Strassler’s lectures [3], the exact β
function for the Yukawa coupling of the Wess-Zumino model is
3
βy = yγ(y), (2.60)
2
where in practice the anomalous dimension must be calculated to some finite order in perturbation
theory. Before moving on to the vector superfields, let us point out that this rescaling of the chiral
superfields may affect the anomaly due to the fermion’s rescaling. The anomaly contributes to
the F Fe term, so we would expect such a modification to feed into the physical gauge coupling
renormalization.
We now perform the analogous analysis for the gauge fields. In the holomorphic basis (anno-
tated by a subscript ‘h’) the gauge kinetic term takes the form
∫
1 1
Lh = d2 θ 2 Waα (Vh )Waα (Vh ) + h.c. (2.61)
4 gh
The factor of 1/gh2 is really shorthand for the holomorphic coupling as τ /4πi. To pass to the
canonical basis one must perform a rescaling of the variables to absorb the coupling into the
gauge field,
Vh = gc Vc , (2.62)
so that the Lagrangian for canonically-normalized fields takes the form

∫ ( )
1 1 ΘYM
Lc = 2
dθ − i 2 Waα (gc Vc )Waα (gc Vg ) + h.c. (2.63)
4 gc2 8π
Again, we are rescaling the whole vector superfield, including the gaugino. For the gaugino picking
up a phase is like a chiral transformation so that this can give an anomaly. Recall that anomalies
come from the change in the path integral measure due to a transformation ψ → eiα ψ. The
formula for a fermion in representation r is
( ∫ ∫ )
i 2 T (r)
D[gc Vc ] = D[Vc ] exp − 4
dx dθ αW (gc Vc )Wα (gc Vc ) + h.c. .
α
(2.64)
4 8π 2
Note that Wα Wα contains our favorite F Fe term. In the present case of interest the transformation
is Vh = gc Vc so that the analog of the eiα transformation is gc , or α = −i log gc . The additional
term in the path integral measure is called the Konishi anomaly. This can be understood as an
IR effect associated with massless particles; while the Wilsonian effective action is holomorphic in
the RG scale µ, the 1PI effective action becomes singular because of the anomaly.
Applying this formula to pure Yang-Mills we obtain
( ∫ ∫ )
i 2 1
D[Vh ] → D[Vh ] exp − d x d θ 2 W (Vh )Wα (Vh ) + h.c.
4 α
(2.65)
4 gh
( ∫ ∫ ( ) )
i 1 2T (Ad)
= D[Vc ] exp − 4
dx dθ 2
− log gc W (gc Vc )Wα (gc Vc ) + h.c. . (2.66)
α
4 gh2 8π
15
In other words, the partition function is
∫
d θd x 2 W(Vh )W(Vh )+h.c.
i
∫ 2 4 1
Z = D[Vh ]e 4 g
h (2.67)
∫ ( )
= D[Vc ]e
i
4
∫
d2 θd4 x 1
g2
h
2T (Ad)
− 8π log gc W(g V )W(g V )+h.c..
c c c c
(2.68)
This tells us that the canonically normalized gauge coupling is

( )
1 1 2T (Ad)
= Re − log gc , (2.69)
gc2 gh2 8π 2
where this expression includes the anomaly. This is the ‘real’ relation between the holomorphic
and canonically normalized gauge coupling.
Now let’s see what happens when we include matter fields. In the pure Yang-Mills case there
was a contribution to the Konishi anomaly coming from the gaugino zero modes. For matter
fields we should also expect a contribution from the matter fermions (quarks). The point is that
integrating out a sliver of momentum space for a species of matter field will generate a non-
holomorphic wavefunction renormalization factor Z. Canonically normalizing with respect to this
wavefunction renormalization shifts the path integral measure so that the contribution to the
anomaly takes the form ln Z WW.
The chiral superfield rescaling is Q′ = Z 1/2 Q. The path integral measure for the Q and Q e
fields in SQCD is
( ∫ )
e = D[Q ]D[Q e ] exp − i T (Ad)
D[Q]D[Q] ′ ′ 2 4
d θd x log Z WW + h.c. .
1/2
(2.70)
4 8π 2
Thus the full expression for the canonically normalized gauge coupling comes from including the
anomalies in the D[Vc ]D[Q′ ]D[Qe′ ] measure,
( )
1 1 2T (Ad) ∑ T (rj )
2
= Re 2
− 2
log gc − log Zj . (2.71)
gc gh 8π j
8π 2
Finally we arrive at an expression for the NSVZ β function. For further discussion see the
original literature [30, 31, 12] or the follow-up works by Arkani-Hamed and Murayama [32, 33].
The β function for the canonically normalized gauge coupling is
( ) ( )
2T (Ad) d log gc ∑ T (rj ) d log Zj
1/2
d 1 −b d 1
= = − − . (2.72)
d log µ gc2 16π 2 d log µ gh2 8π 2 d log µ j
8π 2 d log µ
The first term here is just the flow of the one-loop exact coupling that we wrote in (2.37),
( ) ( )
1 1 −b −1 ∑
= = 3T (Ad) − T (rj ) . (2.73)
d log µ gh2 16π 2 16π 2 j
16
The second term in (2.72) is simply
( )
2T (Ad) d log gc gc2 d 1
− = − , (2.74)
8π 2 d log µ 2 d log µ gc2
which we can move to the left-hand side of the equation. Finally, the last term is simply written
in terms of the anomalous dimensions of the matter fields since
1/2
d log Zj
≡ γj . (2.75)
d log µ
We thus end up with
( )( ) ( )
d 1 T (Ad)gc2 1 ∑ ∑ T (rj )
1− =− 3T (Ad) − T (rj ) − γj , (2.76)
d log µ gc2 8π 2 16π 2 j j
16π 2
from which we derive the NSVZ β function for the running of the canonically normalized gauge
coupling at all-loop order,
( ) ∑
d 1 1 3T (Ad) − j T (rj )(1 − γj )
=− 2 . (2.77)
d log µ gc2 16π 2 1 − T (Ad)g
2
c
8π
2.7 The Konishi Anomaly

This section comes from Jesse Thaler’s TASI 2012 lectures. Suppose you have a gauge theory
with U(1)-charged fermions. The [chiral] anomaly is generated by chiral rotations:
ψ → e+iα ψ (2.78)
ψ̄ → e
c +iα
ψ c
(2.79)
alpha µναβ
L(ψ, ψ c ) → L(e+iα ψ, e+iα ψ c ) + ϵ Fµν Fαβ . (2.80)
64π 2
Now let us promote the chiral rotation parameter α to a chiral superfield. This means we must
have
∫
1
L(ψ, ψ ) → L(e ψ, e ψ ) +
c +iα +iα c
d2 θ αWα Wα + h.c. (2.81)
16π 2
Consider the following Lagrangian:
∫ ( )( X X †
)
4 † 2V
L= d θ Q e Q+Q ee† −2V e
Q 1+ + + ··· (2.82)
Λ Λ
∫ ∫
2
+ d θmQQ e + h.c. + d2 θ 1 Wα Wα + h.c. (2.83)
4g 2
17
If we now integrate out QQ e at the scale m, is there a loop-level coupling X WW? Holomorphy
suggests suggests that this wouldn’t happen since we only get the combination X + X † . However,
if we do a field redefinition
Q → Qe−X/Λ (2.84)
e → Qe
Q e −X/Λ , (2.85)
then we get rid of the linear X + X † terms in the Kähler potential, at the cost of changing the
superpotential by a term
e
∆W = me−2X/Λ QQ. (2.86)
Note, however, that the Konishi anomaly also gives us a term

1 1 1 X
→ − . (2.87)
4g 2 (µ) 4g 2 (µ) 16π 2 Λ
Thus, at the scale µ,

1 1 X 1 b0 − b1 2X 1 X
− = 2 − + O(X 2 ) − . (2.88)
4g 2 (µ, me−2X/Λ ) 2
16π Λ 4g (µ, m) 2
32π Λ 16π 2 Λ
Since b0 − b1 = −1 for integrating out QQ, e this is just g −2 (µ, m)/4 + O(X 2 ). And indeed, the
coupling to X + X † vanishes as expected from holomorphy. Note that if m = 0 then there is no
cancellation, and this falls under the purview of anomaly mediation.
2.8 Symmetries of SQCD

We will make extensive use of the symmetries of SQCD. Here we summarize the main results.
The charge table is:
SU (N ) SU (F )L SU (F )R U (1)A U (1)B U (1)R′
Q □ □ 1 1 1 0
e
Q □ 1 □ 1 −1 0
λ 1 1 1 1 1 1
ψQ □ □ 1 1 1 −1
ψQe □ 1 □ 1 −1 −1
ΘYM 1 1 1 2F 0 2N − 2F
Here we’ve written the SU (N ) gauge symmetry along with the non-Abelian SU (F )2 flavor sym-
metry. In addition to these, we have three U (1) symmetries. Naı̈vely from QCD we only expect
baryon number U (1)B and the anomalous axial symmetry U (1)A , but we have to remember that
there are other fermions in the theory, namely the gaugino λ. The gaugino is a Weyl spinor and so
may have a U (1) charge, but since the other components in its supermultiplet are real, they can-
not carry this U (1) charge. This means that the gaugino’s U (1) charge must be an R-symmetry,
which we call R′ for now.
18
We’ve also written the charges of the quark and anti-quark fermions: these are the same as
their bosons with the R-charge decremented by one since the superspace coordinate θ soaks up
one unit. As a sanity check, the gaugino having an R-charge (rather than an ordinary global U (1))
is consistent with the usual SQCD matter vertices: gauge-quark2 and gaugino-quark-squark.
The last line of the table is the ΘYM angle in Yang-Mills theory. Below we will be more
sophisticated and package this into the holomorphic scale Λ. For now we can afford to be prosaic.
Under a rotation α of the anomalous axial symmetry, ΘYM is shifted by 2αF coming from the F
quarks and the F anti-quarks running in the triangle diagram. This shift is precisely what is meant
when we say a symmetry is anomalous. We note, however, that ΘYM also shifts under a rotation
of the R′ symmetry: it shifts by 2N coming from the gauginos (in the adjoint representation) and
by −2F from the quarks.
Of course, for a simple gauge group there can only be one anomalous U(1). We may replace R′
take a linear combination of R′ and A such that it is anomaly-free. Looking at the ΘYM charges
we can see that this combination is:
F −N
R = R′ + A. (2.89)
F
With this choice ΘYM is invariant and we have the standard assignment of R-charge to the squarks:
U (1)A U (1)B U (1)R
F −N
Q 1 1 F
e
Q 1 −1 F −N
F
λ 1 1 1
ψQ 1 1 −NF
ψQe 1 −1 −NF
ΘYM 2F 0 0
Here we’ve only written the U(1) charges. We note with foresight that one may choose many
different U (1)R charge assignments, but there is a single unique anomaly-free U(1)R which is
special since it is the symmetry that lives inside the superconformal algebra.
3 F < N : the ADS superpotential

We now review the famous result by Affleck, Dine, and Seiberg in the 1980s that instantons
generate the so-called ADS superpotential [34, 35]. Along the way we’ll learn how to use the
moduli space to go to regions in parameter space where we can make definitive statements that
carry over to the nonperturbative regime. In the following section we’ll make use of the tools that
we’ve developed to go over the F ≥ N case and finally address Seiberg duality.
3.1 Holomorphic scale as a spurion

The trick that we will employ is to promote the instanton power of the holomorphic scale Λb to
a spurion for anomalous symmetries. In particular, anomalies break global symmetries through
19
instanton effects which manifest themselves via the ’t Hooft operator12 ,
∏
O’t Hooft = Λb ψi 2Ti , (3.1)
i
where Ti = T (□) = 1/2 for the fundamental representation. For a one-instanton background and
under a chiral rotation, i.e. a rotation that acts independently on each chiral fermion ψi ,
ψi → eiαqi ψi (3.2)
∑
ΘYM → ΘYM − α nr · 2T (r) (3.3)
r
∑
b −i r nr (2T (r))
Λb → Λ e , (3.4)
where nr is the number of fermions in the representation r. If we recall that Λ = |Λ| exp(iΘYM /b),
we note that we can assign a fake (i.e. spurious) charge to Λ so that the ’t Hooft operator preserves
the chiral symmetry,
∑
qΛ = − 2nr T (r). (3.5)
r
For more on the NSVZ β function and the Konishi anomaly, see the notes by Xi Yin13 .
3.2 The ADS Superpotential

Our goal is to write down the effective superpotential. We know that this is given by gauge-
invariant polynomials. In fact, the symmetries of the theory allow us to further constrain the
superpotential. Let’s explicitly write out the representations of the relevant fields under all of
these symmetries using a funny table of boxes,
SU (N ) SU (F ) SU (F ) U (1)1 U (1)2 U (1)R

Q □ □ 1 1 0 0
Qe □ 1 □ 0 1 0
Λb 1 1 1 F F −2F + 2N
The Λb charge under U (1)1 and U (1)2 are given by the prescription above to absorb anomalous
charges. For example, because of (3.3), we can see that the U (1)1 charge of Λb must be
∑ ( ) ( )
1 e 1
q1 [Λ ] = −
b
2Tr qr = −2 (q1 [Q] + q1 [Q])F = −2 F. (3.6)
r
2 2
For the U (1)R let us remember that the bosons and fermion within a supermultiplet contain
different R-charges,
R[fermion] = R[boson] − 1, (3.7)

12
I thank Leo van Nierop for teaching me how to correctly spell an pronounced Gerard ’t Hooft’s name. Ironically
I still cannot properly pronounce “van Nierop.”
13
http://www.people.fas.harvard.edu/~xiyin/beta.pdf
20
so that R[ψQ ] = R[ψQe ] = −1. Remembering that the Dynkin index for fermions is still 1/2,
the quarks combine to contribute −2F to the R-charge of the spurion, R[Λb ] = −2F . We must
remember, however, that there are other fermions in the theory coming from the gauge supermul-
tiplet. Since R[W ] = 2, we know that R[Wa Wa ] = 2, and so the gaugino has R-charge R[λa ] = 1.
The gaugino Dynkin index is just T (adj) = 1, so this gives a contribution of 2N to R[Λb ]. Thus we
find R[Λb ] = 2(N − F ). Note that all of the U (1) symmetries defined here are anomalous, though
two linear combinations are anomaly-free. In particular, we could have written a non-anomalous
U (1)B and a new U (1)R along with an anomalous U (1)A . We don’t have to worry about this for
now, but for reference the revised table looks like
SU (N ) SU (F ) SU (F ) U (1)A U (1)B U (1)R
F −N
Q □ □ 1 1 1 F
Qe □ 1 □ 1 -1 F −N
F
Λb 1 1 1 2F 0 0
Finally, since the holmorphic scale is the only quantity carrying R-charge, we know that the
superpotential must go as
3N −F
W ∼ Λ N −F . (3.8)
Invariance under the U (1) symmetries forces additional factors,

( ) N −F
1
Λ3N −F
W ∼ F
. (3.9)
QF Q
Further imposing flavor invariance and writing the superpotential in terms of gauge invariant
polynomials (which parameterize the moduli space), we get the ADS superpotential,
( ) N −F
1
Λ3N −F
WADS = CN,F , (3.10)
det M
where we’ve written M to be the gauge-invariant meson field and CN,F is a coefficient that we
have to determine. We’ll now do this for the particular case F = N − 1 and then we’ll show that
there are neat tricks we can do to derive more general combinations (F, N ).
More generally: We could have also written a term
b ln ΛWa Wa (3.11)
which would be invariant under the above symmetries due to the transformation of the path
integral under the anomaly. The gauge-invariant chiral superfields that we have available to us
are for constructing a superpotential are Wα Wα , Λb , and det M . The term above corresponds
to a Wess-Zumino term. More generally, we could have written
W = Λbn (WW)m (det M )p (3.12)
21
from which U (1)A and U (1)R symmetries impose
1−m
2 = 2m + 2p(F − N ) ⇒ n = −p = . (3.13)
N −F
Requiring that our superpotential makes sense in the weak coupling limit Λ → 0 (this boils
down to requiring a Wilsonian effective theory) forces power of Λ in (3.12) to be non-negative.
We know from (2.38) that b = 3N − F > 0, so that we require n ≥ 0. This, in turn, implies
p ≤ 0 andhence m ≤ 1. On the other hand, Wa Wa contains derivatives terms and so locality
requires that it comes with a non-negative as well, m ≥ 0 and m ∈ Z. (The low-energy
Wilsonian effective action must have a sensible derivative expansion.) Thus we are left with
m = 0, 1. The case m = 1 gives (3.11), while m = 0 is precisely (3.10).
Holomorphic? This superpotential might make you a bit unhappy—it’s not holomorphic!
Isn’t one of the mantras of SUSY that our superpotential must be holomorphic? The ADS
superpotential appears to have a pole; certainly having a negative power of a superfield is
not holomorphic, i.e. analytic—infinitely differentiable—over the entire complex plane. The
[pedantic] point is that when physicists refer to the ‘holomorphy’ of the superpotential, what
we really technically mean is meromorphy, i.e. holomorphy up to isolated singularities. (I
won’t bother with a technical definition.) Practically, what we mean is that the superpoten-
tial depends only on the superfield and not its complex conjugate. For most of our favorite
pedagogical toy SUSY superpotentials, the symmetries of the theory require that the super-
fields appear in positive powers. However, this needn’t be true—as evidenced by the ADS
superpotential.
What is the physical meaning of such a pole in the superpotential? Well, the divergence
implies that the potential is very large near that region of field space and the universe will
not want to settle nearby. Further, the divergence is a signal that this is a regime where the
theory breaks down, as we shall see below.
3.3 ADS: F = N − 1, instantons

For F = N − 1, W ∼ Λ3N −F = Λb , so that the ADS superpotential smells like an instanton effect.
This case is rather special since we know that instanton effects requires an integer power of Λ;
this is manifestly guaranteed by setting F = N − 1. In this case the SU (N ) gauge symmetry is
completely Higgsed to SU (N − (N − 1)) = SU (1) = nothing. We can see this since each vev that
we turn on breaks a flavor symmetry; turning on one vev breaks SU (N ) → SU (N − 1), turning
on two vevs breaks SU (N ) → SU (N − 2), and so forth. We have F = N − 1 flavors, so a vev
⟨M ⟩ breaks SU (N ) → SU (1). (We have assumed that we are away from the det M = 0 point of
the moduli space; we do not yet fully understand such a theory [5].)
Does this buy us anything? It sounds bad, this puts us in an asymptotically free (β > 0),
strongly interacting region. However, we can go to a region in moduli space where ⟨M ⟩ is very
large. In particular, we can go to a theory where the gauge group breaks before the theory becomes
22
strongly coupled so that our instanton calculations are reliable in this weakly interacting regime.
Before we jump ahead of ourselves, though, let’s convince ourselves that these really are instanton
effects. The ’t Hooft operator can be drawn as a vertex with an external leg for each zero mode
fermion: the quarks, anti-quarks, and the gauginos.
N −1
Q QN −1
λ2N
This doesn’t quite look like our superpotential. However, we can go along the flat directions to
points in the moduli space where the squarks have very large vevs, v. Now recall that we have the
coupling between squarks and gauginos, λQQ e∗ and λQQ e ∗ . We can use these couplings to connect
the λ and Q, Q legs of the ’t Hooft operator. We have two gaugino legs left over, which we may
convert into quarks as shown in the diagram14 .
v v
e
Q
λ
Q
Q
. v
v Q
λ
v
e
Q
Q
v
This rather complicated diagram gives us a contribution to the ‘quark’ mass (where we’re being
lax about v versus v ∗ )
v 2N QQΛ2N +1 . (3.14)
14
As of the time of this writing, this is the sexiest TikZ diagram that I have ever drawn using purely hand-typed
commands. It is important for two key techniques when drawing Feynman diagrams: (a) using clip and foreach
to draw a shaded blob, and (b) rotating and translating ‘x’s to have them uniform at any angle.
23
To get the right term for the ADS superpotential we need to suppress by the length scale of the
instanton. In the presence of the squark vev, this length scale is
b
ρ2 ∼ ,
16π 2 |v|2
and so we can write our instanton-background Lagrangian as
L ∼ v 2N QQΛ2N +1 (ρ)2N (3.15)

−2N 2N +1
=v QQΛ . (3.16)
This is just the fermion mass term that we get from the ADS superpotential.
Λ2N +1 QQ 2N +1 QQ
WADS = → ∼ Λ2N +1 ( )N ∼ Λ . (3.17)
det M eQ
Q e v 2N
Thus we see that the ADS superpotential for F = N − 1 is really just a one-instanton term.
Grown-ups can do the exact instanton calculation [36]. I don’t know how they do it, and for the
moment I don’t really care. The magical result however, is that the coefficient CN F for F = N − 1
is... drum-roll...
CN,N −1 = 1. (3.18)
Now we understand what we need for the particular case F = N −1. That’s useful for very specific
models, but we are more ambitious. In Section 3.4 and 3.5 we will describe two general tools for
taking a given theory of SQCD with F flavors and N colors and deforming it to a theory with a
different F ′ and N ′ . The principle will be to go out along the moduli space of the original theory
where and either give a squark a vev or otherwise add a mass term to the superpotential so that
the low energy theory below these introduced scales is described by a different pair F ′ and N ′ .
We can determine the ADS coefficient by matching the two theories. The procedure of Higgsing
or adding a mass term to a quark is known as deformations of the original theory and will hold
for general F and N , even when F ≥ N .
3.4 Deforming SQCD: Higgsing a squark

Our first trick will be to assign a large vev to one squark flavor,
⟨qF ⟩ = ⟨q F ⟩ = v. (3.19)
We thus have two scales in the theory that we’d like to relate via the Wilsonian renormalization
group. The original theory has an SU (N ) gauge group with F flavors, while the low-energy
Higgsed theory has SU (N ) → SU (N − 1) and one flavor eaten, i.e. SU (N − 1) with (F − 1)
flavors. Thus this Higgsing has taken us from (N, F ) to (N − 1, F − 1). By matching these two
theories, we can find a way to relate the coefficients CN,F and CN −1,F −1 .
24
We now match of the low energy (with a subscript L) and UV couplings at the scale v. Using
(2.40),
( ) ( ) ( )bL ( )b
8π 8π ΛL Λ ΛL Λ
2
= 2 ⇒ bL log = b log ⇒ = . (3.20)
gL (v) g (v) v µ v v
The value of the β-function coefficients are well known in SUSY QCD,
b = 3N − F ⇒ bL = 3(N − 1) − (F − 1), (3.21)
from which we obtain the so-called scale-matching conditions,
(ΛN,F )3N −F = v 2 (ΛN −1,F −1 )3N −F −2 . (3.22)
Scale matching. This will be one of our most powerful tools to explore the moduli space of
SUSY gauge theories. The ‘big picture’ is that we deform the theory in the UV—in this case
by Higgsing a quark, but in the next section my integrating out a quark—and then check the
effect on the low-energy theory which now has a different number of colors and/or flavors and
that does not care about the particular tweaks we performed at a high scale. This language
should sound very familiar: it is nothing more than the usual story of effective field theory.
We can represent the (F − 1)2 light [scalar] degrees of freedom as an (F − 1) × (F − 1) matrix

ML . This can be related to the analogous F × F matrix in the original (UV) theory via
det M = v 2 det ML . (3.23)
Going back and plugging Eqs. (3.20 - 3.23) into the ADS superpotential in Eq. (3.10), we get
( )1/N −F ( )1/N −F
3N −F −2
Λ3N −F v 2 ΛN −1,F −1
CN,F = CN,F
det M v 2 det ML
( )1/N −F
−F −2
Λ3N
N −1,F −1
≡ CN −1,F −1 , (3.24)
det ML
where in the last line we’ve reminded ourselves of the form of the ADS potential with N − 1 colors
and F − 1 flavors. They take precisely the same form. Coincidence? No, the Higgsed theory is
exactly the same as the (N − 1, F − 1) theory at low energies since in this limit the effects of the
Higgsed flavors decouples. (This is the lesson of Wilsonian renormalization.) Thus what we’ve
discovered is that
CN −1,F −1 = CN,F . (3.25)
In particular, this means that C only depends on (N − F ), i.e. CN,F = CN −F . Thus thanks to
our N = F − 1 solution, we now have a set of solutions for (N − F ) = −1. It turns out there’s
still one more trick we can play.
25
The astute reader will wonder how we came to find such a simple relation in Eq. (3.25). What
ever happened to the usual complications, namely threshold effects? Usually when we integrate out
a field, we get some remnant of the matching in the solutions to the RG equations. The matching
we’ve written without any threshold effects implicitly reflects a choice of the DR subtraction
scheme [37]. In other words, the threshold effects are absorbed into the particular definition of
the cutoff scale.
3.5 Deforming SQCD: mass perturbations

The general principle is clear now: how do we can perturb the UV limit of a super QCD and work
out the consequences for the low energy theory. In that limit the UV perturbations are negligible
effects so that the IR theory characterized by (N ′ , F ′ ) is ‘really’ the (N ′ , F ′ ) super QCD theory.
We can match the C coefficients of the two theories to obtain a relation between CN,F and CN ′ ,F ′ .
The next perturbation we have at our disposal is to give mass m to a flavor without Higgsing
the group,
∆Wmass = mQF QF = mMF F . (3.26)
This allows us to integrate out that flavor in the low energy theory, (N, F ) → (N, F − 1). We can
go ahead and play our scale matching game (really just matching in effective field theory),
( )b ( )bL ( )3N −F ( )3N −(F −1)
Λ ΛL ΛN,F ΛN,F −1
= ⇒ = (3.27)
m m m m
so that we finally obtain

−F +1 −F
Λ3N
N,F −1 = mΛ3N
N,F . (3.28)
Now we’d like to solve the equation of motion in the presence of the mass term. We start with
the mass-perturbed superpotential
( ) N −F
1
Λ3N −F
WADS + ∆Wmass = CN,F + mMF F . (3.29)
det M
The relevant functional derivative of W = WADS + ∆Wmass is

( ) N −F
1
−1 ( )
∂W CN,F Λ3N −F (−Λ3N −F ) ∂detM
j = 0 = + mδiF δFj , (3.30)
∂Mi N −F detM (detM )2 ∂Mi j
where (∂detM/∂Mi j ) = (M −1 )ij detM can be written in terms of cofactors (subdeterminants) by

using the matrix identity
( )i cof(Mj i )
M −1 j = . (3.31)
detM
26
The equations of motion for Mi F (similarly for MF i ) and MF F are
( ) N −F
1
∂W −CN,F Λ3N −F cof(MF i )
=0= (3.32)
∂Mi F
N −F detM detM
( ) N −F
1
∂W −CN,F Λ3N −F cof(MF F )
=0= + m. (3.33)
∂MF F
N −F detM detM
The first of these equations tells us that cof(MF i ) = 0, or (M −1 )F i = 0. This, in turn, tells us
that M must take a block diagonal form,
( )
f
M 0
M= . (3.34)
0 MF F
Combining this with (3.33), we find

( ) N −F
1
CN,F Λ3N −F
= mMF F . (3.35)
N −F detM
Now plugging this back into our superpotential W , we obtain

( )( ) N −F
1
1 Λ3N −F
W = CN,F 1+ . (3.36)
N −F detM
As before, the real trick is to write this in terms of low-energy variables,
detM = MF F detMf (3.37)

( )3N −F +1
mΛ3N −F = ΛN,(F −1) , (3.38)
from which we ultimately obtain a recursion relation,

( ) NN−F−F+1 ( ) N −F1 +1
(N,F −1) CN,F (ΛN,F −1 )3N −F +1
WADS = (N − F + 1) . (3.39)
N −F detM f
Note that this is indeed in the correct form that we proposed in (3.10). In fact, comparing to
(3.10), we can deduce the relation
( ) NN−F−F+1
CN,F
CN,F −1 = (N − F + 1) . (3.40)
N −F
3.6 The coefficient of the ADS superpotential

Now we’re doing quite well. From Higgsing through the moduli space we found a nice relation
(3.25) between the coefficients CN,F and CN −1,F −1 . Further, mass perturbations gave us (3.40),
which relates the coefficients CN,F and CN,F −1 . Using these perturbations to explore the moduli
space, we can relate the ADS superpotential for any N, F with that of any other N ′ , F ′ . All we
27
need to decisively write down the explicit form is the value for CN,F at any particular value. This
is precisely what the we-won’t-derive-it-here instanton calculation in (3.18) gave us.
More explicitly, (3.25) required that CN,F is only a function of (N − F ),
CN,F = f (N − F ). (3.41)
Further, the instanton calculation (3.18) told us the particular value at F = N − 1,
CN,N −1 = f (1) = 1. (3.42)
Finally, the mass perturbation (3.40) gives a recursion relation
( )k/(k+1)
f (k)
f (k + 1) = (k + 1) , (3.43)
k
which can better be written
( )k+1 ( )k
f (k + 1) f (k)
= , (3.44)
k+1 k
which has the solution f (k) = k.
This gives us the explicit form of the ADS coefficient for any N and F ,
CN,F = N − F.. (3.45)
The ADS superpotential (3.10) is thus
( ) N −F
1
Λ3N −F
WADS = (N − F ) . (3.46)
det M
We’ll get back to the ADS superpotential one last time in Section 4.4, where we’ll meet an
even slicker derivation.
3.7 Run, run, runaway

Run run run run runnin’ / Here I go I’m wantin’ you
Run run run run runnin’ / Here I go I’m needing you
Run run run run runnin’ / Here I go I’m loving you
Run run run run runnin’ / Run run run run runaway
–Jefferson Starship
What kind of scalar potential does WADS generate? In other words, given (3.46), what do
we know about the moduli space of our theory? Since det M ∼ M F , the scalar potential looks
heuristically like

∂WADS 2

VADS = ∼ |M | N−2N
−F . (3.47)
∂M
Since we’re presently interested in the regime where F < N , we find a rather unsettling potential
of the form
28
VADS
. ⟨M ⟩
Whoa there. This is like a slide that never ends. It has a minimum infinitely far away. In fact,
you can convince yourself that the minimum at ⟨M ⟩ = ∞ is supersymmetric since VADS (∞) = 0.
This is indeed what we would expect from a Witten index analysis of SQCD.
But effectively the potential has no ground state, we call this a run away potential because
the vacuum just runs, runs, runs away to infinity. Let us make a few remarks about this [16]
• Is it possible for quantum effects to bring the ⟨M ⟩ = ∞ point to some finite value? No; the
regime of very large ⟨M ⟩ is one where we trust perturbation theory.
• Is it possible for quantum effects to generate a minimum for small ⟨M ⟩? For example,
perhaps the inverse Kähler metric g aā has some weird behavior. However, the scalar potential
V ∼ g aā ∂a W ∂ā W only has zeros when ∂a W = 0 and we now know that this oly occurs at
⟨M ⟩ = ∞. Note that modifying the Kähler potential can generate metastable (i.e. only
local) minima, this is a key insight for the industry of metastable SUSY breaking that we
will explore later in this document.
• Finally, one last possibility is that for some finite ⟨M ⟩ the Kähler metric becomes singular so
that the theory tells us that there are new massless degrees of freedom. This is what happens
for the gauge multiplet when det M = 0. One could say that we do not yet understand this
sort of theory without vevs [5].
Let us emphasize that this is what happens when we write down a pure SQCD theory with
no additional tree-level superpotential. In this case the tree-level theory has many classical flat
directions in the moduli space. The ADS superpotential is dynamically generated and produces
a potential which pushes the moduli to infinity where a SUSY-restoring vacuum is waiting. In
this sense the ADS superpotential is the avatar of the SUSY-preserving minima predicted by the
Witten index.
Lift your flat directions! This leads us to a very important lesson for SUSY model-builders.
Usually the goal of a nice SUSY theory is to find a clever way to break supersymmetry, i.e.
to write a model where we live in a nice SUSY-breaking vacuum. One must always make sure
that this nice SUSY-breaking vacuum has no flat directions in the potential, in other words,
29
one should “lift” these flat directions. If you don’t, then the dynamically-generated ADS
superpotential will likely push your SUSY-breaking vacuum to a SUSY-preserving minimum
at infinite vev. This is a surprisingly common pitfall that causes papers to get withdrawn.
Of course, if we add tree-level masses to our theory, then there should be no problems. These
mass terms generate a quadratic potential that, for large field values, will pull back towards the
origin. Thus we would expect the potential to be modified to the following heuristic form,
. ⟨M ⟩
⟨M ⟩min
in which we can see that a minimum is generated for finite ⟨M ⟩. Let us thus see what happens
when we give mass terms to all quark flavors and integrate them out. This will take us to a theory
of SQCD without any matter, i.e. pure super Yang-Mills theory. Fortunately, we already have to
tools to navigate the moduli space, so this should be a piece of cake.
In Section 3.5 We learned how to integrate out flavors one at a time by adding mass pertur-
bations. It’s easy to generalize to the case where we integrate out all flavors:
( ) N −F
1
Λ3N −F
W = (N − F ) + mij M ji . (3.48)
det M
The equation of motion (∂W/∂Mij ) = 0 completely determines the meson matrix,

( ) N −F
1
−1
1 Λ3N −F Λ3N −F
0 = −(N − F ) (M −1 )ij (det M ) + mij , (3.49)
N −F det M (det M )2
where we’ve used ∂M j det M = (M −1 )ij det M . Cleaning this mess up we obtain
i
( ) N −F
1
Λ3N −F
Mi = j
(m−1 )ij . (3.50)
det M
30
Now to simplify this further we’d like to get rid of the det M on the right-hand side. We can do
this by taking determinants of both sides, remembering that the tedious expression on the right
is just an overall number multiplying a matrix element. We find
( 3N −F ) NF−F
1 Λ
det M = (3.51)
det m det M
N 1 3N −F
(det M ) N −F = ΛF N −F . (3.52)
det M
Plugging this back into (3.50) we obtain the meson matrix as promised,
( )1/N
⟨Mi j ⟩min = (m−1 )ij det m Λ3N −F . (3.53)
This formula, while derived for F < N , will be true in the IR limit of theories with F > N massive
flavors since one can always integrate out flavors to get to the F < N and F = 0 limits.
We still have to do scale matching at the mass thresholds. We may do these either step by
step or all at once. The condition ghigh-E = glow-E gives us
( )bHE ( )bLE
ΛHE ΛLE
= . (3.54)
m m
Removing one flavor gives us
e 3N −F +1 ,
Λ3N −F m = Λ (3.55)
so that integrating out all flavors gives

e 3N ,
Λ3N −F det m = Λ (3.56)
where the strong coupling scale Λe on the right-hand side what one obtains for pure SYM with no
flavors (F = 0). This tells us that the ADS superpotential for SYM is
F =0
WADS e3 .
= NΛ (3.57)
F =0
To be a bit pedantic, it is common to write WADS F =0

∼ (Λ3N )1/N , where the N th root is related to
the vacua of the theory, as we will see shortly. Is it weird that W = N Λ e 3 , which looks like a
constant? No, Λ e is not a constant since it is a function of det M . The main results (3.53) and
(3.57) were first derived in [38].
3.8 Gaugino condensation

It is now prudent to wonder what kind of physics might generate the ADS superpotential (3.46).
We saw in Section 3.3 that instantons can generate the F = N − 1 ADS superpotential. What
about the case F < N − 1? (The case F > N − 1 will be the focus of the next section.) From our
previous analysis, we know that this manifestly cannot be an instanton effect. What generates
WADS in this case?
31
For F = N − 1 we saw that the gauge group is completely broken (Higgsed) by the vevs. For
F < N − 1, there simply aren’t enough flavors to break the SU(N ) gauge group, there is always
some unbroken SU(N − F ) subgroup. This SYM theory (with F 2 singlets) is asymptotically free
and becomes strongly coupled. We will now show that it is this strong coupling which generate
WADS . In particular, there will be a leftover coupling between the mesons to the pure Yang-Mills
theory that will cause the gauginos to condense.
Let us start with pure SU(N ) SYM, i.e. F = 0. Here there is no anomaly-free U(1)R symmetry,
λ → eiα λ. This is because the gaugino—the only fermion in the theory—would have R-charge
R[λ] = 1. Triangle diagrams with the R-current are anomalous. A nice way to see this is to think
about the ’t Hooft operator, which goes like OtH ∼ λ2N and hence breaks R-symmetry. Note,
however, that the R-symmetry is not completely broken. In fact it is broken down to a discrete
subgroup, Z2N given by
2πi
λ → e 2N λ. (3.58)
Another way to see this is to note that

∑
ΘYM → ΘYM − α nr 2T (r) = ΘYM − 2N α (3.59)
r
under an R transformation, using T (Ad) = N . When the transformation angle is α = kπ/N this
is just a shift by an integer multiple of 2π.
e matter. Unlike the
Now let’s remember that this SU(N ) theory also has F flavors of Q and Q
case of pure super Yang-Mills, this SQCD theory does have an anomaly-free R-symmetry coming
from the combined rotation of the matter superfields and gauginos. (Of course, this R-symmetry
needn’t be the ‘canonical’ R-symmetry.) This seems to be a contradiction: on the one hand we
know that SU(N ) with F < N − 1 massive flavors goes to a pure SYM theory, but the former
theory has an anomaly-free R-symmetry while the latter theory does not.
We claim that this means that there must be some coupling between the mesons and gauginos
(living in Wα Wα ) that compensates for the anomaly and restores the anomaly-free R-symmetry.
We can do the scale matching for the high and low holomorphic scales (3.20),
( )3N −F ( e )3(N −F )
Λ Λ
= , (3.60)
v v
where v is meant to be the mass scale of the mesons, det M ∼ v 2F . We have written Λ for the
e for the low-scale theory (SU(N − F ) SYM). Thus the
high-scale theory (SU(N ), F flavors) and Λ
holomorphic scale depends on the meson vev and hence induces a dependence of the holomorphic
coupling on the meson vev. This is just the avatar of our purported coupling between M and
Wα Wα . Cleaning up the scale matching above,
e 3(N −F ) = 1 Λ3N −F = 1 Λ3N −F ,
Λ (3.61)
v 2F det M
where in the last step we have restored the det M dependence of the scale matching. We should
read this equation as the dependence of Λe on det M for a fixed UV holomorphic scale Λ.
32
The low-energy gauge coupling can be read off of the kinetic term which, in terms of the
low-energy holmorphic coupling τe, is
∫
1
d2 θ τe Wα Wα + h.c. (3.62)
16πi
Explicitly, τe is given by
eb e
Λ
τe = log , (3.63)
2πi µ
where eb = 3(N − F ) since the unbroken gauge group is SU(N − F ). We should think of τe as a
e τe = τe(det M ). We can read off log Λ
function of det M through its dependence on Λ, e from (3.61),
e = − log det M + · · · ,
3(N − F ) log λ (3.64)
where we’ve neglected terms that are independent of the meson vevs. Plugging this into the kinetic
term we get an expression of the form
∫
1
L⊃ d2 θ(log det M )Wα Wα + h.c. (3.65)
32π 2
Now we can already qualitatively see how this is going to give us an anomaly-free R-symmetry.
The R transformation induces a shift in the F Fe term inside the Wα Wα term. This is the signal
that R-symmetry is anomalous in the pure SYM theory. The R transformation also induces a
phase in M , which becomes a shift when we take log det M . This has the correct form to cancel
the transformation of the ΘYM term. This is indeed what happens. Note that the logarithm
plays an important role in converting the phase to a shift. In fancy parlance, (3.65) is called a
Wess-Zumino term. It is a term which can be understood as being generated in a low energy
theory to protect the anomaly structure of the UV theory. For a delightful exposition, see [39]
(check: I might be confusing the WZ term of Skyrme significance with the WZW term).
We can do this more explicitly in components. The relevant terms in the Lagrangian are
1
L= 2
Tr(FM M −1 )λa λa + Arg det M F Fe + FM
2
+ ··· . (3.66)
32π
The Arg det M term is precisely the phase that restores R symmetry while the other terms are
required by SUSY. Taking the equations of motion we obtain
1
FM = 2
M −1 ⟨λa λa ⟩, (3.67)
32π
where on the right-hand side we’ve restored the angle brackets that we typically leave implicit
when discussing the moduli space. Completely independently of this, however, we can write down
FM coming from the full high-scale ADS superpotential to which we must match,
( ) N −F
1
−1
∂WADS N −F Λ3N −F −1
ADS
FM = = Λ3N −F M −1 det M. (3.68)
∂M N −F det M (det M )2
33
This should look very familiar from our analogous calculation earlier in (3.49). The whole point
ADS
is that FM must match with FM in (3.67). Setting them equal and cleaning things up a little,
we finally obtain
( 3N −F ) N −F
1
1 Λ
⟨λa λa ⟩ = − . (3.69)
32π 2 det M
e 3 in terms of the low energy holomorphic scale defined in (3.61).
The right-hand side is just −Λ
What this means is that the theory has a gaugino condensate,
e 3,
⟨λa λa ⟩ = −32π 2 Λ (3.70)
this condensate generates the ADS superpotential and explains the Λe 3 superpotential in the pure
SYM case (3.57).
What we have not explained is what dynamics actually forms the gaugino condensate. This is
a hard question. It is possible that the condensate is formed through instantons, but unlike the
‘easy’ instanton calculation that we omitted for the F = N − 1 case, this instanton calculation is
in the strong coupling regime and is much more difficult.
The miracle is that we have been able to deduce gaugino condensation, which is a purely
strongly coupled phenomenon. It seems like we’ve gotten away with information that we had no
business deriving. How did we manage this voo-doo? We used holomorphy to connection regions
of strong and weak coupling in the moduli space.
Finally, let’s mention that the gaugino condensate spontaneously breaks the Z2N symmetry
that we originally found in the pure SYM theory:
⟨λλ⟩ : Z2N → Z2 . (3.71)
This is because under an R-transformation, Λ3N → e2iN α Λ3N , so that ⟨λλ⟩ → e2iα ⟨λλ⟩. We end
up with degenerate but distinct vacua in SYM essentially coming from the N th root that we had
to take. This is, of course, no surprise from a Witten index analysis.
3.9 Integrating in
We now turn to a related topic that provides a slightly more general technique to understand
gaugino condensation. This technique is called ‘integrating in’ and, as the name implies, it is
in some sense the opposite of ‘integrating out.’ Recall that we usually integrate out a heavy field
to write down a low-energy theory without that field. Now we would like to ask when is this
process invertible? In other words, we will take a theory and include new operators to account for
additional heavy modes. We can then, in certain cases, interpolate to the case when the additional
modes are massless.
3.9.1 The 1PI effective action

Let’s start by reviewing something completely different that will turn out to be handy: the
Legendre transform in field theory15 . Consider the case of a single scalar field, ϕ(x) with some
15
As a helpful reminder, the geometric meaning of the Legendre transform is explained in chapter 7.3 of Ryder’s
textbook. For more review, see Ramond chapter 3.3 [40] or Peskin chapter 11.3 [41].
34
action S[ϕ]. Recall that the partition function (vacuum-to-vacuum amplitude) for a field theory
is given by
∫
Z[J] = d[ϕ]eiS+Jϕ ≡ exp(iW J), (3.72)
where W [J] is the generating functional of connected diagrams and J is a source for the quantum
field. From this we may calculate Green’s functions such as the expectation value for the field
ϕ(x) itself in the background of a source J(x),
δW [J]
⟨ϕ(x)⟩J = ≡ φ(x). (3.73)
δJ(x)
Here we see that φ(x) is the classical value of the field. The 1PI effective action Γ[φ] written in
terms of the classical field φ is given by the Legendre transform
∫
W [J] = Γ[φ] + dd xJφ, (3.74)
where φ is, by definition, the solution of

δ
Γ[φ] = −J(x) (3.75)
δφ(x)
in the presence of the source J(x). This is our main point: the effective action whose tree-level
Green’s functions represents a resummation of quantum effects is given by a Legendre transform.
In general, the actual analytic form of Γ[φ] must be calculated in a loop expansion in ℏ. Recall
that for φ(x) = φ0 one obtains the effective potential which determines the quantum vacuum
structure (moduli space) of a theory. It is crucial now to remember that Γ is the 1PI effective
action and should absolutely not be confused with the Wilsonian effective action, see Section 2.1.
In no sense have we integrated out any heavy modes. The 1PI effective action is effective in the
sense that if we did not know about quantum effects, Γ[φ] would be the action that we would
write down to describe results from experiments of the theory S[ϕ].
Has this jogged your memory? Good. Let’s get back to the art of integrating in. This was first
presented by Intriligator in [42] and is mentioned in pedagogical contexts in [2, 16, 43]. Before
jumping into the details, the main idea is this: the process of integrating out is ‘invertible’ when
the low-energy superpotential is a Legendre transform of the high-energy superpotential. In this
case one can take a known low-energy superpotential and invert the Legendre transform to obtain
information about the vevs of the high-energy degree of freedom and hence information about the
phase of the gauge theory. Let’s do this systematically. We’ll start by integrating out heavy fields
to go to a low-energy effective theory, pointing out relevant features along the way.
3.9.2 Integrating out

Remembering that we are still working with the 1PI effective action, let us be a bit sly and refer to
the act of ‘integrating out’ to obtain an effective theory. This notion is something that we are most
familiar with in the Wilsonian picture where we very literally integrate out shells of UV momenta.
35
However, now we are working with an action which, in principle, encapsulates all quantum effects
in its tree level couplings. The Legendre transform has replaced the quantum degree of freedom
O by its classical value, ⟨O⟩. The notation here is deliberate: the classical value of a field is, of
course, simply its vacuum expectation value. In other words, the 1PI effective action obtained
from performing a Legendre transform with respect to a particuar operator replaces that operator
by its vev. Conceptually this is precisely what we mean when we integrate out a field, we remove
the dynamical degree of freedom associated with a massive field and leave behind the vev.
First let us note that we can translate all of our results regarding the 1PI effective action to
the superpotential.
∫ 2 This is because adding a source term to a chiral superfield O contributes a
term d θJO to the Lagrangian, which is equivalent to adding JO to the superpotential. If we
are interested in ⟨O⟩, the classical value of O, then we would take functional derivatives with
respect to FJ ,
δL[J] δW [J]
⟨O⟩J = = . (3.76)
δFJ δJ
Now we need to clarify some notation. In Section 3.9.1 we used the standard QFT notation where
S is the classical action, Z is the partition function, W is the generating functional of connected
diagrams, and Γ is the 1PI effecive action. From this point on we will work with superpotentials
so that W will always (unless otherwise stated) refer to a superpotential. We will refer to 1PI the
effective superpotential with explicit subscripts. A useful analogy is thus
S[ϕ] : Γ[φ] :: W [O] : Weff [⟨O⟩]. (3.77)
Explicitly, given a superpotential W , we may write down a 1PI effective superpotential
Weff (⟨O⟩, Λb0 ) = W (J, Λb0 ) + J⟨O⟩. (3.78)
Let’s now follow Intriligator’s presentation in [42] (see also [44] for some context). This requires
a bit of cumbersome notation, so we will write things out as explicitly as possible. Let us start
with two SQCD theories: the upstairs theory with superpotential Wu and holomorphic scale
Λu , and the downstairs theory with superpotential Wd and holomorphic scale Λd . Both theories
describe chiral superfields ϕi and ϕei which form gauge invariant polynomials Xr . The upstairs
ê
theory additionally includes a massive chiral superfield ϕ̂ which forms a meson M = ϕ̂ϕ. There
are also gauge invariant polynomials Za which contain both ϕ and ϕ̂.
ê
We know that as we integrate the heavy quark ϕ̂ (and the corresponding ϕ) out of the upstairs
theory, the low energy look something like the downstairs theory up to the vev of the heavy fields
which manifest themselves as vevs of the Za fields. Given a mass scale m for ϕ̂, we know that the
two holomorphic scales are related by (3.20),
Λbdd = mΛbuu , (3.79)
where bd and bu are the β-function coefficients. Recall b = 3N − F so that the power of m is
correct by dimensional analysis.
36
Let us explicitly write out the field dependence of the superpotential, Wu = Wu (X, Z, Λbuu ),
where we suppress the indices of X, Z, and J. Now let’s set up the Legendre transform. Define
the ‘full’ superpotential to be Wu plus the usual source term,
∑
Wf (X, Z, Λbu , J) = Wu (X, Z, Λbuu ) + JZ. (3.80)
∑
Note that the JZ implicitly includes a ‘source’ for the meson M , which we write as mM , the
notation is meant to be evocative since such a source term is exactly a mass term for ϕ̂. Using
the equation of motion in the background of a source J,

∂Wf
= 0, (3.81)
∂Z J
we may write down the 1PI effective superpotential by taking the Legendre transform with respect
to the Z fields,
W1PI (X, Λbuu , J) =Wf (X, ⟨Z⟩, Λbu , J) (3.82)

bd
=Wd (X, Λ ) + WI (X, Λbuu , J), (3.83)
where we’ve written out the field dependence explicitly. In the second line we’ve recovered our old
friend, the downstairs theory. The leftover interactions coming from the ⟨Z⟩ vev is encapsulated
in WI (X, Λbuu , J), which is RG irrelevant and vanishes for J = 0. We can see the irrelevance since
WI → 0 as m → ∞, as can be seen from the equation of motion (3.81). The final decomposition
of W1PI into Wd and WI depends on the fact that Wf is linear in J which can be taken as an
assumption. (See [44] for some remarks on this.)
Remember that we have not thrown away any information to get to this low energy effective
superpotential. Due to the linearity in the source(s) J, all we have done is performed a Legendre
tranform. The great thing is that such a transform can be inverted.
3.9.3 Integrating back in

Let’s explicitly reconstruct Wu from W1PI . First construct a new superpotential
∑
Wn (X, Y, Λbuu , J) = W1PI (X, Λbuu , J) − JY, (3.84)
where Ya is a set of new gauge invariants which do not affect W1PI . The magic trick will be to
transmogrify Ya back into the Za fields which we integrated out. We make the bold claim that
this can be done simply by integrating out the source(s) J. Playing the same game, we use the
equation of motion (∂W/∂J)Y = 0. Because W is linear in J, the source is just an auxiiary
superfield. Now observe that
∑
Wn (X, Y, Λbuu , J) = Wu (X, ⟨Z⟩, Λbuu ) + J(⟨Z⟩ − Y ), (3.85)
so that J is just a Lagrange multiplier that enforces Y = ⟨Z⟩ which, in turn, sets
Wn (X, Y, Λbuu , J) = Wu (X, Y, Λbuu ). (3.86)
37
So what we’ve found is that we can take Wn (X, Y, Λbuu , J), which only depends on Wd and WI ,
integrate out the source J, and end up with the high scale theory, Wu . How’s that for pulling a
rabbit out of a hat?
To some extent all we’ve been doing is slight of hand using Legendre transforms. The real power
comes from specific examples when the downstairs theory is much simpler than the upstairs theory.
Integrating in fields then allows us a handle on the upstairs theory without having to meddle with
it directly. One remark is worth making: because the upstairs theory includes a heavy field, it
does not necessarily make sense to talk about it as a dynamical degree of freedom. Instead, it gives
us information about the vaccum structure of that theory. The key point is that we’re already
armed with a very powerful tool, the exact ADS superpotential, (3.46), which gives us a natural
handle on the low-energy theory. We’ll put this to good use to rederive a now-familiar result.
3.9.4 Gaugino condensation (from another perspective)

ê
Let’s consider the simple case where the gauge singlets Z are bilinears in ϕ̂ and ϕ. This is
particularly nice because the source terms in the superpotential Wf are simply mass terms, mM ,
and the cumbersome term vanishes WI = 0. You are referred to Intriligator’s paper [42] for the
not-so-simple case.
Fortunately, the simple case is enough to give an alternate demonstration of gaugino conden-
sation. Our upstairs theory and downstairs theories are pure SU (N ) pure Yang-Mills, where the
downstairs theory does not explicitly contain the glueball field. The Z fields are objects like mesons
M and the source terms are simply JZ = Tr(mM ), where the trace is over SU(F )L × SU(N )R
flavor indices. In our case, M is just the glueball superfield
S = −Wα Wα . (3.87)
Let us work with S as the gauge invariant polynomial that we’d like to integrate out, Z = S. The
source for S is the holomorphic coupling τ , which we will simply write as J = b log Λ, where Λ is
the SYM holomorphic scale and we’re ignoring some overall constants. The gaugino condensate
⟨S⟩ ∼ ⟨λa λa ⟩ is simply the classical value of S so that

∂W1PI
⟨S⟩ = . (3.88)
∂ log Λb b
log Λ
Because our gauge invariant polynomial is quadratic, we know that W1PI (X, Λb , log Λb ) = Wd (X, Λb )
since WI vanishes. Note that we’re a little redundant in the superpotential arguments (for exam-
ple, there are no X fields), but we do this to maintain the connection to our general discussion
above where it was very important to keep track of the arguments of each functional. In fact, we
know that the low-energy effective superpotential is exactly the SYM ADS superpotential (3.57),
W ∼ Λ3 . As promised, this rather simple (though it took a bit of tooth-pulling to derive), and we
shall see that it is indeed much simpler than the Wu that we’ll eventually derive, justifying this
entire procedure.
Let’s spell things out now. From the expression above for ⟨S⟩ and remembering that b = 3N ,
we have
N ∂Λ3
⟨S⟩ = = Λ3 . (3.89)
3N ∂ log Λ
38
This is an expression for the vev ⟨S⟩, but the inverse Legendre transform restores this to S. In
the general discussion above, this is just the Y = ⟨Z⟩ step from integrating out J. Thus we may
trade Λ3 → S in Wn to obtain
Wn (X, S, Λb ) = WADS
F =0
− N log Λ3 S (3.90)
= N S − N S log S. (3.91)
We obtain the full ‘upstairs’ theory by also including the gauge kinetic term W = b log Λ S,
[ ( 3 ) ]
ΛN
WVY ≡ Wu = S log +N . (3.92)
S 3N
This theory is the Veneziano-Yankielowicz superpotential, originally derived by anomaly consider-

ations in [45, 46]. It is indeed uglier than the ADS superpotential, justifying the ‘slick’ integrating
in of S. It may sound a bit fishy that S gets pushed out of and into the theory with impunity,
but it is important to remember that S is a massive field whose dynamical degrees of freedom
don’t really belong in either theory. The whole point of reconstructing WVY is that it encodes the
vacuum structure of the glueball field,
∂WVY
=0 ⇒ ⟨S⟩ = e2πik/N Λ3 , (3.93)
∂S
as we saw at the end of Section 3.8.
Just for fun, we can integrate in one more time. This time let’s take the VY theory to be our
downstairs theory Wd = WVY , and 0 < F < N SQCD be our upstairs theory. We would like to
integrate in some matter fields which manifest themselves as gauge-singlet mesons M with mass
terms of the form Tr mM . This time is a straightforward application of the steps outlined above.
We simply write
Wn = WVY − Tr mM, (3.94)
where we assume that in WVY the downstairs holomorphic scale Λd = Λ is written in terms of the
upstairs scale Λu via
3N −F
Λ3N
d = det m Λu . (3.95)
We then integrate out the source m using its equation of motion ∂Wn /∂m = 0, which gives
⟨m⟩ = SM −1 . Finally, we obtain
[ ( ) ]
Λu3N −F
Wu = S log +N −F , (3.96)
e
S 3N −F det QQ
e From here one should properly integrate out
where we’ve written in the heavy squarks Q and Q.
S once again since it is always massive.
39
3.10 Relatied topics
There are some related topics which one may pursue at this point. A detailed discussion is left to
more suitable references and, perhaps, future revisions of this document.
• Generalizing gaugino condensation for order parameters of dimension one. See Dine and
Mason Section 4.3 [8] and the references therein.
• Integrating in for WI ̸= 0. See Intriligator’s original paper, [42]. This is also mentioned in
[43].
• Gaugino condensation can also be derived in a slick way using N = 2 techniques due to
Seiberg and Witten. We postpone a discussion of Seiberg-Witten theory to later in this
document.
• Gaugino condensation is discussed in chapter 8.3 of Terning [5] using methods similar to
those in Section 3.9.4 above, though in somewhat more plain language.
4 F = N and F = N + 1: Some special cases

Now we get to the interesting regime. As before, we have to start by developing some tools.
4.1 ’t Hooft Anomaly Matching

The primary tool is the ’t Hooft anomaly matching condition, which ’t Hooft considered
so trivial that the first reference to the technique is in his summer school lecture notes [47].
A more rigorous proof was presented by Banks, Frishman, Schwimmer, and Yankielowicz [48].
Anomaly matching is a way to check the spectrum of a strongly coupled theory in a regime where
perturbative methods have no business being valid.
Suppose we have some asymptotically free gauge group Gg in a theory which also has a flavor
symmetry group GF . By definition the gauge symmetry is just a redundancy in the description
of a physical system, so it must be anomaly free. For global symmetries, however, all bets are off.
There’s nothing inconsistent about a global symmetry being anomalous. Another way of saying
this is that the triangle diagrams associated with three global currents (we shall call this the G3F
anomaly) don’t have any gauge bosons attached. ’t Hooft developed a clever way to use these
anomalies to teach us about the effective degrees of freedom at low energies.
He key step is to pretend that GF is weakly gauged, meaning the coupling constant gF is
arbitrarily weak and perturbative in the range of energies of interest. Eventually we’ll take the
limit gF → 0 and turn off this gauging. Now the GF anomalies are a problem, albeit a pretend
problem. In order to keep the theory consistent, we need to cancel the pretend G3F anomalies.
We do this by adding spectator fields to the theory which are Gg singlets and only cary GF
quantum numbers so that the anomaly coefficients for GF cancel. To be more precise, we only
trust these calculations in the perturbative regime of the Gg gauge theory (we define gF to be
perturbative effectively everywhere). Since Gg is asymptotically free, the precise statement is that
40
the anomaly coefficients for GF cancel in the UV:
AUV (F ) + AUV (spec) = 0. (4.1)
Now that we’ve introduced spectator fields to cancel the G3F anomalies in the UV, the ‘magic’
is that the anomalies must still cancel in the IR theory,
AIR (F ) + AIR (spec) = 0. (4.2)
This is because the IR regime where Gg is strongly coupled looks the same as far as the GF weakly
gauged sector is concerned. In fact, we have AIR (spec) = AUV (spec). Alternatively, this is just the
consistency of the gauge theory. Of course we know that with respect to the ‘real’ gauge group
Gg , the UV and IR theories are very different with totally different degrees of freedom. While the
UV theory describes quarks, the IR theory describes confined states. Thus the calculable triangle
diagrams contributing to AUV (F ) are totally different than those contributing AUV (F ), which may
be much more difficult or impossible to calculate.
Now there are two possibilities as we go to the IR theory.
1. The strong dynamics spontaneously breaks the flavor symmetry so that GF is broken. This is
what happens, for example, in QCD when the chiral condensate breaks SU(3)L ×SU(3)R →
SU(3)D . When this is the case, the anomaly matching condition doesn’t tell us anything
useful.
2. GF is left unbroken by the strong dynamics. In this case we can do something.
When GF is unbroken, we can combine (4.1) and (4.2), remembering that AIR (spec) = AUV (spec),
AIR (F ) = AUV (F ). (4.3)
In other words, the anomalies for the global GF symmetries must match in the UV and IR. At
this point we can forget about the spectator fields and the weak gauging; they’ve served their
purposes valiantly and we now have everything we need.
In fact, there’s a handy corollary to this result. We know that anomalies come from zero-mode
fermions, so if we calculate that the global anomaly is nonzero in the UV, then we can say that
there must be massless fermions in the IR spectrum.
4.2 Moduli Space

We should also discuss the moduli space of the F > N theory. Fortunately, we’ve already done
this in Section 2.4.3. Please refer to that section for a quick refresher.
4.3 N =F
We can now take our first steps beyond the ADS superpotential. We will find an anomaly-matching
spectrum, but some of the symmetries will be spontaneously broken. Our UV and IR fields are:
41
SU(N ) SU(F )L SU(F )R U(1)B U(1)A U(1)R
Q □ □ 1 1 1 0
e
Q □ 1 □ -1 1 0
QQe=M 1 □ □ 0 2 0
QN = B 1 1 1 N N 0
QeN = B
e 1 1 1 −N N 0
m 1 □ □ 0 −2 2
2N
Λ 1 1 1 0 2F 0
In the last two lines we’ve included the mass parameter from W ⊂ mQQ e and the holomorphic scale
which should each be treated as spurions. The holomorphic scale carries the quantum numbers
of the anomalous symmetries. We’ve written out the charges under the theories’ U(1)s: baryon
number, the anomalous abelian symmetry, and the anomaly-free R symmetry. Note that we
cannot use the anomalous symmetry for anomaly matching since the spectators would have to be
charged under Gg .
We have 2N 2 fields subject to (N 2 − 1) D-flatness conditions, this leaves us with N 2 + 1
moduli. Looking at our gauge invariant polynomials, we have N 2 mesons M , and one of each type
of baryon, B and B. e Thus we have N 2 + 2 fields to fit into an N 2 + 1 moduli space. This just
means that there is a classical constraint on the (M, B, B) e space to project it to the moduli
space, as we learned in Section 2.4.3:
e
det M = B B. (4.4)
This constraint describes a complex manifold which is singular at the origin M = B = B e = 0.

e
This is a conical singularity associated with the fact that M , B, and B are complex so that at
the origin their phases are undefined.
We know that physics doesn’t like singularities. In this case, we’ll see that the strong dynamics
will want to smooth out this singularity by modifying the constraint. Let’s look at what happens
at the origin. Let’s recall (3.53), the formula the vev ⟨Mi j ⟩ in the presence of a mass term,
( )1/N
⟨Mi j ⟩min = (m−1 )ij det m Λ3N −F . (4.5)
We originally derived this for F < N , but we remarked that it was general. This is why we pointed
that out. Taking the determinant of both sides,
1 ( )1/N
det M = (det m)N Λ2N = Λ2N . (4.6)
det m
The mass matrix m drops out (a special feature of F = N ), which means we can take the m → 0
limit where there is no mass term in which case we would still have det M = Λ2N . There is one
more remnant of the mass term: for m ̸= 0 we have B = B e = 0 since all fields with baryon number
can be integrated out. When we take the m → 0 limit, we must also address the B, B e ̸= 0 case.
Recall that classically det M = B Be = 0, so (4.6) is telling us that there really must be a quantum
modification to the classical constraint. Is the Λ2N factor the whole correction?
42
We can use the symmetries of the theory to explicitly write out the most general quantum
modification to the classical constraint:
( )
e
det M − B B = Λ 2N 2N a e b c
1 + (Λ ) (B B) (det M ) , (4.7)
e → 0 we
where dimensional analysis requires a + b + c = 0. We also know that in the limit B B
must recover (4.6). This means that b > 0. Further, we don’t want any divergences in the weak
coupling limit (Λ → 0) so that a > 0. We thus have
( )
(Λ 2N a
) (B e b
B)
det M − B Be = Λ2N 1 + Cab . (4.8)
(det M )a+b
e ≫ Λ2N . In this regime the

Now let’s slide along the baryonic branch of moduli space, B B
constraint takes the form
( ) a+b
b−1
det M ∼ B Be . (4.9)
This can be seen by dividing both sides of (4.8) by B B e and dropping terms which are ≪ 1. The
regime B Be ≫ Λ2N corresponds to breaking the gauge group before the theory is strongly coupled
so that this should match the classical result. This tells us that Cab = 0, so that the full quantum
modified constraint on the moduli space is
e = Λ2N .
det M − B B (4.10)
The right-hand side indeed has the correct power for an instanton effect (Λb ). The effect of this
quantum modified constraint is that the origin of moduli space—which previously had a conical
singularity—has cordoned off from the theory. The singularities are smoothed out16 . A direct
consequence of this is that some global symmetries are necessarily broken. M , B, and Be all carry
charges. Eliminating the origin of the moduli space means that at least some of these are broken.
We have at least checked that for F = M the low-energy description of the mesons and
baryons is correct as long as we impose the quantum modified constraint. We can now consider
actual vacua of this theory. We want to preserve as much symmetry as possible. For a somewhat
relevant discussion of why this is the case, see [49]. First consider the branch of moduli space
where only the meson gets a vev, M = Λ2 · 1. Baryon number and R-symmetry are preserved,
but SU(N )L × SU(N )R symmetry is broken down to SU(N )D ; we have chiral symmetry breaking,
just like ordinary QCD. The particle content now has the charges
SU(N ) SU(N )D U(1)B U(1)R
Q □ □ 1 0
e
Q □ □ -1 0
M 1 (Ad + 1) 0 0
B 1 1 N 0
e
B 1 1 −N 0
16
For those who are keeping up with all the hip and cool review literature, these conical singularities should be
familiar from the XY Z model described in Strassler’s unorthodox review [3]. In fact, if you’re not sure what these
cones are all about, I suggest looking over that review.
43
where (Ad+1) refers to the adjoint plus trace decomposition of a bifundamental. Now let’s get to
the matter at hand. Let’s use the ’t Hooft anomaly matching conditions to check that the global
anomalies indeed match. It is instructive to do a few examples
e
• SU(N )3D . In the UV there are N fundamental quarks and N antifundamental antiquarks Q
which sum to give no anomaly. In the IR the adjoint is automatically anomaly free.
• SU(N )2D U(1)B . In the UV we have
T (□)2 (+1) + T (□)2 (−1) = 0 (4.11)
while in the IR we have
T (□)2 (0) + T (□)2 (0) = 0. (4.12)
• SU(F )2D U(1)R . This one is non-trivial. In the UV we have

[ ]2
[N T (□)]2 (−1) + N T (□) = −2N 2 . (4.13)
This just comes from A(r1 ⊗ r2 ) = dim(r1 )A(r2 ) + dim(r2 )A(r1 ). In the IR, the mesons make
no contributions since A(Ad) = 0. The baryons give
N 2 (−1) + (−N )2 (−1) = 2N 2 . (4.14)
• U(1)3B . Both equal zero.

• U(1)B U(1)2R . Both equal zero.
• U(1)2B U(1)R . Both equal −2N .
• U(1)3R . In the UV we sum the quarks, antiquarks, and gauginos:
(−1)N 2 + (−1)N 2 + (N 2 − 1) = −N 2 − 1. (4.15)
On the IR side we sum the meson, baryon, and anti-baryon:
(−1)N 2 − 1 − 1 = −N 2 − 2. (4.16)
It seems like ’t Hooft has failed! However, we must remember that for the IR calculation
we must impost the quantum modified constraint so that one degree of freedom can be
expressed in terms of the others. This means that we really have −N 2 − 1 and the results
match.
And that’s the anomaly matching game. It’s not exactly Starcraft 2, but it’s probably more
interesting than Wheel of Fortune. Let’s see how this works for the baryonic branch where
⟨M ⟩ = 0 and
e = ΛN .
⟨B⟩ = ⟨B⟩ (4.17)
In this region of moduli space the SU(F ) flavor symmetries are left intact, but U(1)B is broken.
Chiral symmetry is preserved, but baryon number is now spontaneously broken. The particle
representations under the remaining symmetries are
44
SU(N ) SU(F )1 SU(F )2 U(1)R
Q □ □ 1 0
e
Q □ 1 □ 0
M 1 □ □ 0
B 1 1 1 0
e
B 1 1 1 0
We’ve labelled by flavor symmetries by 1 and 2 rather than L and R to avoid confusion with the
R-symmetry. Now let’s do a few anomaly matching examples.
• SU(N )31 . The anomaly coefficient is N . In the UV theory this comes from the SU(N )
multiplicity of states, while in the IR this comes from SU(N )2 multiplicity.
• SU(N )2 U(1)R . The anomaly coefficient is −N coming from R-charge. (Recall that the R
charge differs for each element of a supermultiplet so that even though all of the listed R
charges are zero, they include fermion zero modes that carry nonzero R charge.)
• U(1)R and U(1)3R . These have anomaly coefficient −N 2 − 1.
4.4 The ADS superpotential from nothing

Now that we’re comfortable matching anomaly coefficients, it turns out that there is yet another
rabbit we can pull out of our hats. We can derive WADS from this theory. The strategy should be
obvious: integrate out a flavor to go from F = N to F = N − 1 and hopefully recover the ADS
superpotential. In fact, consistency of our entire discussion of SQCD requires that this should be
true. That being said, it does feel like we’re doing a bit of hocus pocus; our F = N theory has no
superpotential, just a quantum modified constraint.
Our first step, then, is to put in a superpotential that we can later mold into WADS . To do this,
we introduce an auxiliary chiral superfield X which will act as a Lagrange multiplier to enforce
the quantum modified constraint.
W = X(det M − B B e − Λ2N ). (4.18)
Thus we’ve swapped our quantum constraint for an actual superpotential. That was a good first
step. Now let’s cut to the chase and add a mass term to the N th flavor,
∆W = mMN N . (4.19)
It helps to write out the meson matrix as
( )
f N
M
M= e MN N . (4.20)
N
We can now take the equation of motion with respect to all fields containing the heavy flavor.
First, the baryons:
∂W e=0
= 0 ⇒ XB (4.21)
∂B
∂W
= 0 ⇒ XB = 0. (4.22)
e
∂B
45
e = 0. Let’s move on to the meson fields.
In other words, we have B = B
∂W f=0
=0 ⇒ m + X det M (4.23)
∂MN N
∂W
=0 ⇒ XMN−1i det M = 0. (4.24)
fN i
∂M
This gives us the constraints
−m
X= (4.25)
f
det M
MN−1i = 0, (4.26)
e = 0. Finally, there’s one more equation of motion to take:

where the last condition sets N = N
that of the auxiliary field X.
∂W f = Λ2N ,
=0 ⇒ MN N det M (4.27)
∂X
f. Substituting these into the superpotential,
which sets MF F = Λ2N / det M
( )
e
W = X det M − B B − Λ 2N
+ mMN N (4.28)
f MF F − XΛ2N + mMN N
= X det M (4.29)
= −XΛ 2N
, (4.30)
f MN N = −mMN N . This boils down to

where we’ve used X det M
mΛ2N
W = . (4.31)
f
det M
Now recall our favorite scale matching condition (3.27), which in this case is
( )2N ( )3N −(N −1)
Λ e
Λ
= . (4.32)
m m
Plugging this into W , we obtain a familiar result,

e 2N +1
Λ
W = WADS = . (4.33)
det Mf
This is the result which more technically minded people derived by an honest instanton calculation.
Even the coefficient is correct. We derived using only the quantum modified constraint and a few
slick moves. Everything really fits together nicely.
46
4.5 F = N + 1: s-confinement
Now we meet a second special case, which also turns out to be the simplest case. In fact, one can
find many theories which exhibit similar behavior as the F = N + 1 scenario. Here the baryons
and mesons perfectly match the anomalies; there is no additional constraint on the moduli space,
quantum or otherwise. The light degrees of freedom (moduli) match the UV degrees of freedom.
There’s a new feature in this theory: s-confinement. This is a really stupid name where
the ‘s’ appears to refer to screening. This is a phase of a gauge theory where there are massless
degrees of freedom. Compare this to QCD where there are no massless fundamental quarks to
screen; there is a linear potential until you hit the lightest quark mass, at which point the QCD
flux tubes break. If there were massless quarks, as in the s-confining case, the flux tubes break
immediately. As a handy summary, s-confinement carries the following implications:
• Confinement does not break chiral symmetry (contrast this with the N = F case)
• All IR degrees of freedom are gauge-invariant composites of the fundamental fields
• It smoothly17 interpolates between the Higgs and confining phases; there is no gauge-
invariant order parameter which distinguishes these phases and no phase transition
• A non-vanishing confining superpotential is dynamically generated
• At the origin of moduli space all global symmetries are unbroken and the global anomalies
in the UV and IR theories match
A very neat feature of this theory is the complementarity between the Higgs and confining phases.
We already saw this in F = N theories; consider the mesons and baryons of the low energy theory.
Near the origin of moduli space the mesons and baryons act like composite states. On the other
hand, in the semiclassical region of large moduli, these fields are Higgsed. We have a smooth
transition between these two phases without a phase transition, and so these phases are identical.
For F = N + 1 no global symmetries are broken (e.g. chiral symmetry) and there are no
quantum modified constraints. B and B e are no longer flavor singlets since there are too many
flavors. The antisymmetrization of the color indices by the ϵ-tensor antisymmetrizes all but one
flavor index. We know that this is equivalent to the antifundamental representation; e.g. the
Young tableaux relation
=.
Alternately, we can see this explicitly from the indices:
Bi1 ···iN = ϵα1 ···αN Qi1 α1 · · · QiN αN (4.34)

ii1 ···iN
=ϵ Bi1 ···iN ≡ B .
i
(4.35)
17
For this reason my adviser proposes that the ‘s’ can be interpreted as ‘smooth,’ though I think he’s the only
one to claim this. On the other hand, he seems to have naming rights [50, 51].
47
The classical constraints are trivially satisfied. Using
M ij Bi = ϵi1 ···iN Qi Qi1 · · · QiN Q̄j = 0, (4.36)
e we find (using ϵ identities)
and the analogous identity for B,
( −1 )i
M det M = Be i Bj . (4.37)
j
We should check whether these classical constraints are quantum modified, as they were in the
F = N case. We will do this by using the same trick of adding a mass term and then taking the
mass term to zero. (4.5) still holds. We end up with
M −1 det M = mΛ2N −1 , (4.38)
where the right-hand side vanishes as m → 0, which is exactly what we expect from the classical
constraint since the mass term sets the baryon moduli to zero. Thus we suspect that the classical
constraints are not quantum modified. This is shown in [52]. We’ll get back to this as soon as we
write down our superpotential and anomaly matching.
There’s one more difference from the F = N case that we should highlight. In the case F = N ,
we were able to assign all superfields (including Λ) to have zero R-charge (where of course we mean
the lowest components). For the F = N + 1 case, we cannot do this for the holomorphic scale.
SU(N ) SU(F )1 SU(F )2 U(1)B U(1)A U(1)R
Q □ □ 1 1 1 0
e
Q □ 1 □ -1 1 0
M 1 □ □ 0 2 0
B 1 □ 1 N N 0
e
B 1 1 □ −N N 0
Λ2N −1 1 1 1 1 2(N + 1) −2
where the U(1)R charge of the Λ2N −1 comes from the expression 2N −2(N +1), which in turn comes
from the contribution of the gauginos, mesons, and baryons to the ’t Hooft instanton operator.
This reflects the anomaly in the canonical U(1)R charge, not to be confused with the anomaly-free
R-charge which may mix the canonical charge with the other U(1)s in the theory. (This depends
on the number of flavors.) This table gives us all of the possible terms for the superpotential. The
most general thing we can write that satisfies these symmetries is
[ ( )]
1 e + β det M + det M f det M
W = 2N −1 αBM B . (4.39)
Λ BM B e
We should be curious about the behavior of the overall 1/Λ2N −1 prefactor in the weak coupling
limit, where we expect W = 0. In fact, we know that the raison d’être of this superpotential is
only to impose the classical constraint as a Lagrange multiplier. If this is the case, then in the
classical limit W = 0 indeed vanishes an there’s no question about the weak limit. We can just
calculate the equations of motion and require that they yield the classical constraints. This fixes
f = 0, β = −α so that we finally have
1 ( )
e .
WF =N +1 = 2N −1 det M − BM B (4.40)
Λ
48
Great. Now we can go and check our ’t Hooft anomaly matching conditions. Now everything
is non-trivial. To do this we should re-write our particle table in terms of the anomaly-free
symmetries. In particular, we should now write the anomaly-free R-symmetry by combining
U(1)s so that Λ2N carries no R-charge.
SU(N ) SU(F )1 SU(F )2 U(1)B U(1)R
Q □ □ 1 1 1/(N + 1)
e
Q □ 1 □ -1 1/(N + 1)
M 1 □ □ 0 2/(N + 1)
B 1 □ 1 N N/(N + 1)
e
B 1 1 □ −N N/(N + 1)
Note that in the F = N case we had a quantum modified constraint that told us that there was
a vev to expand about. In particular,
e = Λ2N .
det M − B B (4.41)
So along the baryonic F = N branch, for example, we could expand B → ⟨B⟩ + δB (and similarly
e so that we end up with linear pieces in the superpotential with respect to the dynamical
for the B)
fields δB. Once you have a linear expression for a field in the superpotential, you can take the
equation of motion to solve for that field. On the baryonic branch we could solve δB in terms
e for example. This is also related to the fact that the origin has been removed from the
of δ B,
moduli space; there has to be a vev that we can expand about and hence we must be able to
eliminate a field using its equation of motion. Back to our present F = N + 1 case, there’s nothing
keeping us away from the origin. And so the origin is part of the anomaly matching. This shows
up non-trivially in the anomaly matching.
Let’s demonstrate a few anomaly coefficients.
• SU(F )3 . Here the UV theory gives N while the IR theory gives (N + 1) − 1.
• SU(F )2 U(1)B . The UV and IR theories both give N .
• SU(N + 1)2 U(1)R . This is a non-trivial one. Recall that we’re really counting the R-charge
of the fermion of each superfield. In the UV we have
( )
1 N2
N −1 =− . (4.42)
N +1 N +1
In the IR we sum the meson and baryon
( ) ( )
2 N −N 2
(N + 1) −1 + −1 = , (4.43)
N +1 N +1 N +1
after a little bit of elbow grease.
You can see that this can get pretty non-trivial.
Let’s get back to checking the quantum modified constraint. As one last plausibility check, let’s
see if we get the correct quantum modified constraint when we flow from F = N + 1 to F = N by
49
adding a mass to the last flavor. We then take the equation of motion to get a quantum modified
constraint. In particular,
∂ 1 ( f − BF B
)
e F + m = 0,
W = − 2N −1 det M (4.44)
∂MF F Λ
where F = N + 1. Note that the BF B e F term is independent of the heavy flavor (which we can
see since it came from the derivative of a BM Be term). We end up with
f − BB
det M e = mΛ2N −1 = Λ
e 2N , (4.45)
where the right-hand side is the dynamical scale of the theory. This is a non-trivial relation among
the remaining degrees of freedom. We have recovered our old quantum modified constraint for
F = N . Looking back at the procedure, we notice that MF F (where F = N + 1) has played the
role of a Lagrange multiplier to enforce the F = N quantum modified constraint.
These types of s-confining theories are quite nice and are the easiest SQCD theories to find.
They don’t suffer from any singularities on the moduli space and are generally well-behaved. In
fact, in the past a young theorist could spend some portion of their life writing a list of all such
theories [50, 51].
5 F > N + 1: Seiberg Duality

Now that we’ve built up some sophistication with our methods, we can get to the good stuff.
No more special cases. On to one of the great theoretical discoveries of the 1990s. While the
main result—Seiberg duality—can be plainly state and used as is, the intuition leading to this
result is firmly rooted in a solid understanding of the renormalization group. There are many
good references for this, but Strassler’s ‘unorthodox introduction’ [3] is perhaps the most intuitive
and pedagogical. As it is highly unlikely that I can improve on these lectures, they are strongly
recommended as background reading. For those who already have a strong background, Strassler’s
follow up lectures on the duality cascade [4] offer a good summary in his style.
Some good lectures. A personal note from the author: seriously, Strassler’s write up on
this subject are among the best lecture notes I have read for any subject.
5.1 The Banks-Zaks Fixed point

For F > N + 1 we need to know about the RG fixed points of the theory. As a starting point,
we know that the one-loop β-function coefficient takes the form b = 3N − F . For F > 3N the
theory flows to a weakly coupled theory at low energies, i.e. to a ‘boring’ trivial fixed point. For
F < 3N , on the other hand, the theory is asymptotically free. Now we are curious about what
happens in the region around F = 3N , where the one-loop beta function vanishes. Could this be
a fixed point?
50
Fixed points in SUSY theories. One thing that we know about SUSY fixed points is
that they typically come from the cancellation with a higher-order term in a loop expansion.
For example, for dimensionful couplings there is a ‘classical’ β-function associated with the
‘engineering’ dimensions of the tree-level interaction. In the rich phase structure of 3D N = 2
models [53], non-trivial fixed points appear when tree-level β function for a relevant coupling
cancels against the loop-level (quantum) β function. (Due to the change in the dimensions
of each field, marginal operators in 4D are relevant in 3D.) In this case the gauge coupling
runs at one-loop order, and we seek to understand a cancellation from the two-loop contribu-
tion. These examples should raise all sorts of concerns since they represent, by definition, a
non-perturbative regime where we don’t have much control. Fortunately, SUSY (essentially
holomorphy) comes to the rescue and preserves this control.
The NSVZ exact β function provides a very powerful tool for checking the existence of a fixed
point. Recall from Section 2.6 that this takes the form
−g 3 3N − F (1 − γ(g 2 ))
β(g) = g2
, (5.1)
16π 2 1 − N 8π 2
where the anomalous dimension is

−g 2 N 2 − 1
γ(g 2 ) = + O(g 4 ). (5.2)
8π 2 N
The γ term in the numerator and the g 2 /8π 2 term in the denominator are higher loop corrections
to the one-loop running. In fact, things are looking pretty optimistic since we can see that the
F γ term in β appears to carry the opposite sign as he one-loop contribution.
Let’s now do this a bit more carefully. We shall take a slightly-weird limit where we take both
F and N to infinity while staying very close to the 3N − F = 0 region where the one-loop β
function vanishes. We’d like to fall into the asymptotically free region, so while taking F, N → ∞,
let us fix
F
=3−ϵ (5.3)
N
for arbitrarily small, but positive, ϵ. We will see shortly that for small ϵ the coupling is at weak
coupling so that perturbative methods are valid. Expanding 16π 2 β in powers of g 2 /8π 2 ,
[ ]−1
g2
16π β(g) = −g (3N − F (1γ)) 1 − N 2
2 4
(5.4)
8π
g2
= −g 3 (3N − F ) + N 2 (−g 3 )(3N − F ) − g 3 F γ (5.5)
8π
g5 g5 N 2 − 1
= −g 3 (3N − F ) − N 2 (3N − F ) − 2 F (5.6)
8π( 8π ) N
g5 F
= −g 3 (3N − F ) − 2 3N 2 − 2N F + , (5.7)
8π N
51
where we’ve dropped terms of O(g 7 ). Here we can plug in our large-N limit with F/N = 3 − ϵ,
3g 5 2
16π 2 β(g) = −g 3 ϵN + (N − 1) + O(g 7 , ϵ2 ). (5.8)
8π 2
We see that there is indeed a fixed point coming from the cancellation of a one loop and a two
loop effect:
8π 2 N
g∗ = ϵ. (5.9)
3 N2 − 1
This is the SQCD version of the celebrated Banks-Zaks fixed point18 of QCD [54]. Though
the two-loop β-function had been known since 1974, Banks and Zaks were the first to seriously
consider the zero of the β-function at F∗ = 16.5. By performing an expansion in (F − F∗ ), where
F is a physically meaningful natural integer. They found that a non-trivial fixed point indeed
exists in the F > F∗ regime, and along with it the found many unexpected features. For example,
chiral symmetry was not broken in the strongly coupled regime as was previously expected.
The phase diagram looks like this:
β(g)
. g
g∗
Recall that the fixed point indicates a scale-invariant theory. For a sensible quantum field
theory of particles with spin less than 2, scale invariance implies a much larger symmetry, conformal
invariance. And when we slap on supersymmetry, we have superconformal invarance.
The ’t Hooft coupling. What is the meaning of the odd N, F → ∞ while F/N = 3 − ϵ
limit which we took? This is just the famous ’t Hooft ‘large N limit’ (see e.g. [25]). We should
have known that this would have shown up. The ’t Hooft limit originally came about when
physicists studied non-supersymmetric O(N ) models where it was found that an expansion
in 1/N can control loop-level effects by allowing the β-function to cancel in a perturbative
18
For a nice set of slides about the history of the Banks-Zaks phase, see http://scipp.ucsc.edu/Symposium/
Peskin_BanksZaks.pdf
52
regime. The key insight is that perturbation theory is not an expansion in g 2 /4π, but rather
in the ’t Hooft coupling,
g2N
λ= . (5.10)
4π 2
(By the way, there are times when the loop factor of 4π turns out to be very important [55].)
The large N expansion corresponds to fixing λ and taking N → ∞. The β-function for λ is
λ2 3 − N
F
(1 − γ)
βλ = − . (5.11)
8π 2
1 − 8πλ2
Perturbation theory breaks down then λ ∼ 1, not necessarily when g 2 ∼ 1. At the SQCD
Banks-Zaks fixed point we find that λ∗ ∼ 1/N , so that the large N limit indeed pushes this
to the perturbative regime.
5.2 Some facts about superconformal theories

We shall invoke a few very useful theorems about superconformal field theories. We’ll be terse,
mostly because doing otherwise would take us a bit far afield.
Fact 5.1. At a superconformal fixed point, there exists a unique anomaly-free R-symmetry which
is part of the superconformal algebra. It is conventional to call this Rsc .
Compare this to the case of the canonical R-symmetry, which is not well-defined since it can
mix with any other U(1) in the theory. In fact, Rsc is precisely the anomaly-free R-symmetry under
which we assigned quantum numbers to our fields when we analyzed the F = N and F = N + 1
theories.
Fact 5.2. The dimension of a chiral operator O and its Rsc charge are related by
3
dim O = |Rsc [O]| . (5.12)
2
Further, D = ±3Rsc /2 where the + (−) sign corresponds to (anti-)chiral superfields. More gen-
erally, in d dimensions, dim O = (d − 1) |Rsc [O]/2|.
This is an extremely powerful result that comes from the fact that Rsc is part of the supercurrent
along with the stress energy tensor, which includes the generator of dilatations. The dimension
of an operator is information about its transformation under scaling, so it is unsurprising that in
a superconformal theory this information should be linked by SUSY to the Rsc charge.
The really neat thing is that the particular dimension that is related to the Rsc charge is
the full quantum dimension of the chiral field: the canonical (‘engineering’) dimension plus the
loop-induced anomalous dimension γ. Thus the real power of this supercurrent relation is that we
can get direct information about γ, even in non-perturbative regions. These observations become
even more powerful when we throw in one more useful fact:
53
Fact 5.3. Near any conformal fixed point all spin-zero, gauge-invariant operators O must have
[scaling] dimension greater than or equal to 1. More generally for d [spacetime] dimensions, O
must have [scaling] dimension (d−2)/2. Further, saturation of this bound implies that the operator
is a free field.
This very bold statement is proved in [29]. These are very ‘deep’ statements about the structure
of field theory and SUSY; we’ll be sure to get good mileage out of them.
Let’s do a couple of quick warm ups from [3]
• Suppose that all gauge couplings are set to zero. Then the chiral superfield Q is gauge in-
variant (gauge redundancy has been turned off) and so it satisfies the conditions for Fact 5.3.
If Q has any residual superpotential interactions, then by (2.58), dim ϕ = 1 + γ/2 > 1, note
the ‘strictly greater than’ symbol. This tells us that γ > 0.
• In 4D, as we assume throughout this document, we know that an operator in the superpo-
tential is relevant if its coupling has dimension greater than 3 and irrelevant if its coupling
is less than 3. We can read this off by checking whether the Rsc charge of the operator is
greater or less than 2.
The chiral ring. This is a phrase which pops up fairly often, so it’s worth addressing its
meaning. The product of chiral operators (‘left chiral superfields’) is also a chiral operator.
Since the Rsc charges of the product of chiral operators is just the sum of the individual Rsc
charges, then the dimension of the product of operators is also just a sum of the individual
dimensions. (Recall that in the usual OPE, a composite operator picks up its own anomalous
dimension.) A set with defined addition and multiplication operations is called a ring.
Now let’s do a little more with all this. In particular, let’s take a look at the role of Rsc -
symmetry in SQCD in F > N . Recall the table of quark charges,
SU(N ) SU(F )1 SU(F )2 U(1)B U(1)Rsc
Q □ □ 1 1 (F − N )/F
Qe □ 1 □ -1 (F − N )/F
The U(1)R charges are found by taking linear combinations of the U(1)s so that the holomorphic
scale (now acting as a spurion for instanton contributions) Λ3N −F is uncharged under it, i.e. so
that it is anomaly-free. We can now calculate the dimension of the meson in the UV theory,
dim(QQ)e = 3 Rsc (QQ)e = 3F − N . (5.13)

2 F
We know that this is equal to the canonical scaling dimension plus an anomalous dimension,
dim(QQ) e = 2 + γ, where the 2 comes from the fact that there is a product of two fields. Thus the
facts laid out in this section tell us that at a superconformal fixed point, γ → γ∗ ,
3N
γ∗ = 1 − . (5.14)
F
As a check, we can plug this into the NSVZ numerator (5.1) to find
β ∝ 3N − F (1 − γ∗ ) = 0, (5.15)
as expected. Thus our technique is consistent.
54
3
5.3 2N < F < 3N : the conformal window
Our first significant step with our new superconformal tools will be to go back to the meson
operator QQ.e Our Rsc analysis above told us that the dimension of this field is 3 − 3N/F . The
unitarity bound then requires
3N
3− ≥ 1, (5.16)
F
i.e. F/N ≥ 3/2. This sets a lower bound on the region where we expect a superconformal fixed
point, the so-called conformal window. Previously we only knew that for F > 3N the theory
goes to a trivial (weakly coupled) fixed point, and that the cases F ≤ N + 1 had been already
discussed. Now it seems like the conformal window is smaller than we might have originally
thought and has a lower bound so that it is constrained to exist in the regime
3
N < F < 3N. (conformal window) (5.17)
2
In fact, there’s something interesting about the conformal window. Because dim(M ) > 1 (strictly),
this theory flows to an interacting superconformal theory. It is in the interacting non-Abelian
Coulomb phase. As we showed it is asymptotically free so that the coupling grows as we flow
into the IR, but unlike ‘real world QCD,’ the coupling doesn’t diverge. Instead, it asymptotically
approaches the fixed value g∗ . The superconformal theory at the IR fixed point is interacting and
not confined.
Seiberg posited that the Banks-Zaks fixed point really exists in the conformal window [56].
This result (to the best of my knowledge) has not been proven rigorously. Our arguments above
are valid for the large N limit with F near the upper limit of the window; for now we shall take
on faith that this fixed point survives even away from these particular limits. The RG flow for a
theory in this window thus looks like
Conformal Window IR Free: F > 3N
.
g=0 g∗ g=0
The lower limit of the conformal window begs the following question: what happens in the
regime N + 1 < F ≤ 3N/2? We’ll get to this shortly. What we already know, however, is that our
Rsc analysis showed and the requirement that β = 0 at a fixed point shows that the Banks-Zaks
point does not occur in this regime. It will turn out that this regime will be dual to conformal
window by a duality, the famous Seiberg duality to which we now turn.
5.4 Seiberg Duality

We shall motivate Seiberg’s remarkable proposal in Section 5.7. First, let’s just dive in and explain
the proposal. Let us call the conformal window SQCD theory by a suggestive name, the electric
theory. Seiberg proposed that there is an equivalent description of this theory, which we shall
call the magnetic theory. The remarkable aspect of this “electric-magnetic” duality is that the
55
magnetic theory looks very different. First, it is a theory with an SU(n) gauge symmetry with
F flavors, were n = F − N . This is weird! Our usual ‘real-world QCD’ intuition from strongly
coupled gauge theories is that quarks confine into mesons which are gauge singlets. Now the low
energy theory contains dual quarks q and qe which are charged under a new gauge group. Further,
while the electric theory is an ‘honest’ SQCD model with no superpotential, the magnetic theory
has an additional superpotential
Wmag ∼ M qe
q, (5.18)
where M is a field independent of q and qe to be associated with the QQ e meson relative to the
electric theory. It is sometimes customary to call this dual theory SQCD+M.
To be more precise, this is an infrared duality in which two different theories flow to the same
fixed point in their IR limits. We will see that while the SQCD electric theory is asymptotically
free, the SQCD+M magnetic theory is IR free and and that the magnetic theory can be understood
to be the low-energy effective theory of the electric theory.
This duality is certainly unexpected. The two theories seem to be very different beasts with
no obvious path connecting them. However, the fact that the two theories have different gauge
groups should not discourage us: recall that the gauge symmetry is just a redundancy of how we
describe the theory. In principle, one is perfectly free to describe the same physics with different
redundancies. On the other hand, while gauge symmetries needn’t match, the global symmetries
of UV and IR theories should match. (Recall our ’t Hooft anomaly matching discussion.) Thus
let us write out the symmetries of our fields in the UV and IR theories:
SU(N ) SU(F ) SU(F ) U(1)B U(1)Rsc
F −N
Q □ □ 1 1 F
e
Q □ 1 □ −1 F −N
F
M 1 □ □ 0 2 F −N
F
q □ □ 1 N
F −N
N
F
−N
qe □ 1 □ F −N
N
F
The meson field M exists as a degree of freedom of the magnetic theory, but we identify it as well
with the QQ e operator of the electric theory. This seems a little odd, since the QQe bound state
has canonical dimension 2 while the ‘fundamental’ M field in the magnetic theory has canonical
dimension 1. Of course, there’s nothing strange about this: the canonical dimensions only hold in
the UV of each theory. The Banks-Zaks superconformal fixed point exists as the IR limit of these
theories19 Thus during the RG flow from the UV to the IR, the QQ e field in the electric theory
and the M field in the magnetic theory pick up anomalous dimensions so that they end up with
dimension
e=3 F −N
dim M = dim QQ . (5.19)
F
To be precise we should define a separate scalar field Mm with dimension 1 in the UV so that
M = QQ e = µMm for some characteristic scale µ. It is conventional to write everything in terms
19
This isn’t quite true, as we’ll discuss in Section 5.7 when we consider the RG flow of the SQCD+M theory.
56
of the meson M and the scale µ so that
1
Wmag = M qe
q. (5.20)
µ
Thus (1/µ) plays the role of a coupling constant.

Now we can match the holomorphic scales of the two theories. We can assign R-charge by
treating Λb as a spurion for the instanton-generated ’t Hooft operator. This can then be used to
determine the scale matching of the (UV) electric theory Λ and the (IR) magnetic theory Λ e
e 3(N −F )−F = (−1)F −N µF ,

Λ3N −F Λ (5.21)
where µ shows up once again to soak up scaling dimensions. This tells us something very im-
portant: since we have the relation Λb Λ eeb = const, as one holomorphic scale increases, the other
decreases. This tells us that this is a strong-weak duality, as one theory become strongly coupled
the other is weakly coupled. This is the origin of the ‘electric-magnetic’ nomenclature, since this is
reminiscent of a g ↔ 1/g duality. (Unlike the N = 4 case, this isn’t quite a ‘true’ electric-magnetic
S-duality.) This is the real power of Seiberg duality: just when one description of the theory is
becoming non-perturbative, the other is returning to perturbativity. (Otherwise there’s no point
to a duality which relates one incalculable regime to an equally incalculable regime.)
That funny ‘duality’ minus sign. The minus sign in the relation between the holomorphic
scales is a familiar sight from more familiar duality transformations. For example, taking the
derivative of the action with respect to log Λ ∼ τ tells us that the electric and magnetic field
strengths are related by
fαW
Wα Wα = −W f α, (5.22)
which should be familiar from the usual electric-magnetic duality in Maxwell theory, wherein
e2 − B
E 2 − B 2 = −(E e 2 ). (5.23)
This sign should be reminiscent of the signs in Fourier and Legendre transforms. The reason
why this minus sign must be here will be seen in Section 5.5.4. It is necessary in order to be
able to identify the quarks of the dual-of-the-dual theory with the original quarks and to have
the dual-of-the-dual superpotential vanish. In other words, this minus sign is there so that the
dual-of-the-dual theory is the same as the original theory.
5.5 Checks of Seiberg Duality

5.5.1 ’t Hooft anomaly matching
The global anomalies match between the magnetic and electric theories. Here’s a summary:
• SU(F )3 : N
57
• SU(F )2 U(1)B : N
• SU(F )2 U(1)R : −N 2 /F
• U(1)2B U(1)R : −2N 2
• U(1)R : −N 2 − 1
• U(1)3B , U(1)B , U(1)B U(1)R : 0
5.5.2 Moduli space

If the two theories indeed describe two regions of the same physics, then their flat directions
(parameterized by gauge invariant polynomials) must match. Writing B and B e for the electric
baryons and b and eb for the magnetic baryons,
SU(F ) SU(F ) U(1)B U(1)R
M □ □ 0 2 F −N
F
B . 1 N N F −N
F
e
B 1 . −N N F −N
F
M □ □ 0 2 F −N
F
. (F −N )N (F −N )N
b 1 F −N F
eb 1 . −(F −N )N (F −N )N
F −N F
where we’ve written .to mean the antisymmetric representation with N indices and .to mean the
conjugate antisymmetric representation with (F − N ) indices. The moduli indeed match because
these two objects both have the same dimensionality, N (1 − N/F ). There is a handy way to relate
the moduli of both theories,
Melec = Mmag (5.24)

B = ∗b (5.25)
Be = ∗eb, (5.26)
where ∗ is the Hodge dual operator.
5.5.3 Navigating the moduli space

Another check of consistency comes from exploring the moduli space. We already learned how to
do this when we studied the ADS superpotential. We have two tools: Higgsing both sides (recall
Section 3.4) and integrating out a flavor from each side (recall Section 3.5).
[To do: write this up. See slides and Seiberg.]
This works out, and we are left with a commutative diagram
58
dual
SU(N )F SU(F − N )F
e FF
⟨QQ⟩ . M qe
q
SU(N − 1)F −1 SU(N )F −1

dual
5.5.4 Dual of the dual

Another check of consistency is that the duality transform takes the magnetic theory back into
the electric theory. The first sign of potential trouble is the relation of the holomorphic scales,
(5.21). Performing the duality transform again and identifying the dual-of-the-dual scale with the
original electric scale, we find
e 3(F −N )−F Λ3N −F = (−)N µ
Λ eF , (5.27)
where 1/e µ is the superpotential coupling for the dual-of-the-dual theory. The left-hand side is
e = −µ. We know
indeed the same as (5.21). Setting the right-hand sides equal give the relation µ
that the dual-of-the-dual theory has gauge group SU(N ) with F flavors. We now want to see
that the superpotential vanishes. Let us quite the dual-of-the-dual quarks as d and de (soon to be
e The independent meson, formed from the
identified with the original electric quarks Q and Q).
magnetic theory’s dual quarks, will be written as N . The duality-induced superpotential is thus
1 e
Wdual2 = N dd. (5.28)
e
µ
This doesn’t look like the original theory, but we must also remember to include the superpotential
that was generated when we first when from the original electric theory to the SU(F −N ) magnetic
theory. This looks like
1
Wdual = M N, (5.29)
µ
where we’ve written the magnetic quarks in terms of the dual-of-the-dual meson N . The complete
dual-of-the-dual superpotential is thus
1 e
W = N (M − dd). (5.30)
µ
This is just a linear (mass) term for N . As we learned in Section 3.9, this means that we can just
use the N equation of motion to integrate it out of our theory by a Legendre transform. In other
words, N is just a Lagrange multiplier. It sets N = 0 and M = dd. e The first relation tells us that
W = 0, while the second relation tells us that dde = QQ, e so that we have indeed recovered the
original electric theory.
59
5.6 N + 1 < F < 3N/2: compositeness
Now let’s get back to the curious case of N + 1 < F < 3N/2. Recall that when we identified the
conformal window, unitarity forced us to impose a lower limit of F > 3N/2. This left a swath
of (F, N ) space unaccounted for: we know that it must exist in a different phase, but how do we
characterize it? This is the real problem. In the case of the conformal window we we were able to
go into the corner of parameter space, F ≈ 3N , where the theory is perturbative. On the other
hand, the regime around F ≈ 3N/2 is badly non-perturbative; what can we do?
Ah! Well, it just so happens that in the conformal window we had this magnetic theory which
was weakly coupled in precisely the same regime where the electric theory is strongly coupled!
Now we have a trick: while we’re in the conformal window, go to the F ≈ 3N/2 limit where
the magnetic theory is weakly coupled. The key is to stay in the magnetic picture where we are
weakly coupled and make the jump across F = 3N/2. Since 3n = 3(F − N ) < F , the theory
is not asymptotically free. In particular, the superpotential is irrelevant and so becomes weakly
coupled in the IR (y flows to zero). This leaves us with an SU(F − N ) gauge theory with massless
magnetically charged quarks and a singlet M . This is the free magnetic phase.
5.7 RG motivation for Seiberg duality

Seiberg duality is often presented somewhat miraculously, as if the details of the duality were
collected by Seiberg on two stone tablets (one describing SQCD in the conformal window and the
other describing the SQCD+M dual!). While this is all that is necessary for a ‘working knowledge’
for most phenomenological applications, one can have a fuller appreciation of the insights involved
by better motivating why Seiberg originally conjectured his duality.
First let us recall the RG flow of an SQCD theory in the conformal window:
.
g∗
There are two fixed points: the trivial fixed point at g = 0 and the Banks-Zaks fixed point at
g = g∗ , where g∗ is perturbative in the large-N limit. We have assumed that g∗ exists more
generally in the conformal window. We’ve argued that the trivial fixed point is unstable and all
theories in the conformal window eventually flow to the Banks-Zaks fixed point.
Note that if the electric theory is in the conformal window, then the gauge group of the
proposed dual theory, SU (n) with n = F − N , is also in the conformal window. (This is trivial
to check.) The dual theory, however, has an additional field M . In the limit where the magnetic
superpotential vanishes, Wmag = 0, then this field is free and is completely decoupled from the
SQCD sector. The magnetic SQCD sector thus still flows to the Banks-Zaks fixed point so that
the IR theory is Banks-Zaks plus a decoupled free scalar field with dimension 1. From Fact 5.2 we
know that Rsc [M ] = 2/3. Further, because M is a free field, we can transform it independently of
the dual quarks with respect to its flavor symmetries. In particular, this Wmag = 0 theory enjoys
a larger flavor symmetry group,
SU (F )L × SU (F )R × U (F )L × U (F )R . (5.31)
Now let’s turn on the superpotential. For simplicity, let’s define the coupling to be y = 1/µ.
We note that Wmag breaks the global symmetry above to SU (F )L × SU (F )R , the same symmetry
60
as the electric theory. In the special case where g = 0 (trivial SQCD fixed point) and y ̸= 0, we
are left a theory with no gauge symmetry and a trilinear superpotential. This is just the XY Z
model investigated for pedagogical reasons in [3]; we know this flows to the trivial fixed point. We
know that this fixed point is unstable since small perturbations in g will send it to the Banks-Zaks
fixed point.
In fact, things are more interesting at the Banks-Zaks fixed point. A good question to ask
is whether the superpotential destabilizes this fixed point. In fact, it does. Let’s look at the
β-function for y in the neighborhood of (g∗ , y = 0). Suppose y takes a very small value. We know
that the dimension of M is very close to 1 since
dim M = 1 + O(y 2 ). (5.32)
Further, we know that the β-function is given by the sum of anomalous dimensions,
βy = y (γM + γq + γqe) . (5.33)
This is simply because the coupling y is protected by holomorphy so that the only renormalization
comes from wavefunction renormalization associated with non-canonical scaling dimensions. (See
Strassler’s notes for an excellent introduction [3].) Thus
( ( ) ( ))
3 3
βy = y (dim M − 1) + dim q − + dim qe − (5.34)
2 2
( )
3n
=y 1− . (5.35)
F
We’ve dropped the higher-order terms in y coming form γM . The key point about βy is that it is
negative for any F < 3n (i.e. in the conformal window). Thus y is a relevant operator! (If you’re
worried about the effect of the higher order terms in y, then unitarity constraints should convince
you that this doesn’t invalidate our argument.) The relevance of y means that the Banks-Zaks
fixed point is an unstable fixed point. Now the RG structure of the theory has become very
interesting, indeed:
y
. g
g∗
Both the trivial and Banks-Zaks fixed points are unstable. Where, oh where, can our little theory
go? It may be that there is no fixed point and that all couplings flow to infinity, this means that
our theory must be defined with a cutoff. That would be boring. Fortunately, Seiberg imagined a
more interesting scenario. The crux of Seiberg duality is that there is a new fixed point at (ĝ, ŷ)
that is stable and is the IR limit of the RG flow:
61
y
(ĝ, ŷ)
. g
g∗
Clearly this is different from the Banks-Zaks fixed point. The key point is that the new fixed point
for the magnetic theory at (ĝ, ŷ) is to be identified with the Banks-Zaks fixed point of the electric
theory. Why should we believe this? consider the dimension of the field M at the IR fixed ∑ point.
From Rsc -symmetry we know that 2γqe = 1 − 3n/F . At a fixed point β = 0, therefore i γi = 0.
This fixes γM = (2F − 3N )/F . Thus in the magnetic theory the meson has dimension
3F − 3N
dim M = 1 + γM = . (5.36)
F
This is indeed exactly what we get from calculating the QQ e dimension at the Bank-Zaks fixed
point of the electric theory, as we saw in (5.13). Thus the field M in the magnetic theory can
plausibly be associated with the meson QQ e of the electric theory. Similarly, one can check the
dimension of the baryon operators in both theories. In fact, we have already shown that they have
the same Rsc charge.
5.8 The importance of the scale µ

Let us return to the scale matching condition (5.21). To avoid (or perhaps generate) confusion,
in this section we will explicitly refer to the dynamical scales as Λel for the electric theory, Λmag
for the magnetic theory, and Λ for the matching scale which we previously called µ. The scale
matching relation can then be written
Λbelel Λbmag
mag
= (−)N Λbel +bmag . (5.37)
This, in turn, can be written as

( ) ( )
1 bel |Λ| −bmag |Λ| −1
2
= 2 log = log = . (5.38)
gel (|Λ|) 8π Λel 8π 2 Λmag 2 (|Λ|)
gmag
This shows us that |Λ| may be interpreted as the scale at which the two dual couplings are equal up
to a sign. As shown in Fig. 1, the size of |Λ| determines the structure of the duality. For Λ > Λel ,
there is an energy range for which there exists no weakly coupled description of the dynamics.
On the other hand, for Λ < Λel , it is possible to have a weakly coupled magnetic description of
the dynamics which matches on to a weakly coupled description of the electric dynamics. Since
Seiberg duality is actually a statement about the far infrared, the ambiguity in the ‘correct’ value
62
of Λ is not usually relevant. The ability to work with two weakly coupled theory, however, can be
used in model building, e.g. [57], from which this section is based.
Naive dimensional analysis suggests that all three scales Λ, Λel , and Λmag have similar mag-
nitude. One way of understanding Seiberg duality is the statement that Λ is a parameter and
that the naive relation of scales needn’t hold. In particular, having a small Λ would allow one to
consider weakly coupled magnetic dynamics.
The condition |Λ| < Λel can also be written as Λy < Λmag where Λy is the Landau pole scale
for the magnetic Yukawa coupling. This is because... [To do: see [57]].
a a
10 10
5 5
m m
0.2 0.4 0.6 0.8 1.0 1.2 1.4 Lel 1 2 3 4 Lel
-5 -5
-10 -10
Figure 1: Values of α = g 2 /4π as functions of the renormalization scale µ for Λ = 1.5Λel on the
2
leftFigure
and Λ1:= Values of the
0.8Λel on the one-loop gauge
right. The couplings
‘electric’ ↵ = g(blue)
coupling /4⇡, is
aspositive
functions forofµthe
>Λ renormaliza-
el whereas the
tion scalecoupling
‘magnetic’ µ, for ⇤(red)
= 1.5is⇤positive
el on the for
leftµand
< Λ⇤mag= .0.8
We ⇤eluse
on Nthe=right.
6 andThe F =“electric”
8. Imagecoupling
from [57].
(shaded in blue for
[To do: redraw in TikZ.] 6 colors and 8 flavors) is positive for µ > ⇤ el , while the “magnetic”
coupling (shaded in red) is positive for µ < ⇤mag .
5.9 We
Remarks on the|⇤|
see that formally is the scale
meaning ofwhere the two dual couplings are equal up to a
the duality
sign, although this means (outside of the conformal window) that we have extrapolated the
[Tocoupling
do: fleshof one thisofout.]
the Seiberg
theories duality is an |⇤|,
to a scale, infrared
that isduality.
beyond its Thisrange
means of that the electric
validity. For
andexample,
magneticif theories
we choose |⇤| > ⇤el for a theory in a free “magnetic” phase, we see that the of
flow to the same IR fixed point. We have taken the additional step
identifying
“magnetic” the theory
magnetic theory as theateffective
is renormalized IR that
a scale |⇤| theory after the
is above electric UV
its Landau pole.theory
In thisbecomes
case
strongly coupled. Indeed, the effective field theory paradigm is a story
there is a gap between ⇤mag and ⇤el and a simple description of the dynamics is not known. of infrared duality. The
perturbative low-energy effective theories which are valid the IR are typically
This situation is shown on the left in Fig. 1. If |⇤| is smaller than ⇤el then the Landau pole non-renormalizable
andofrequire a UV cutoff.
the “magnetic” theoryFor a related
is above ⇤el , discussion
as shown on ofthe
theright
nonlinear
in Fig.sigma
1. The model and the
ambiguity Higgs
in the
20
sector,
valueseeof Nima
|⇤| is Arkani-Hamed’s
irrelevant for Seiberg lectures at PiTP
duality: since2010 .
the duality holds in the extreme infrared,
for a fixed “electric” theory, any value of |⇤| will lead indeed
In their UV limits, the magnetic and electric theories are to the very samedifferent
“magnetic” and have no deep
dynamics
relation. In particular,
sufficiently far in the the duality appears to only exist at the IR fixed point in the continuum
infrared.
limit ofGiven
the RG flow. This is
a particular underlying often accompanied
“electric” theoryby the
thatstatement
is valid abovethat ⇤ suchweancan infrared
run intoduality
at
el
(asleast
is typical in field theory) should be contrasted with the ‘exact’ dualities
two problems when we try to extend Seiberg duality for a free “magnetic” phase beyond which appear in string
theory. This
the infrared. perspective
First there is not
will quite accurate,heavy,
be unknown, since composite
one can indeed states,takethata are
limitnot
where thereinis a
included
finite region in which the two theories actually have the same RG flow rather
the “magnetic” description. However, as long as these states have masses around ⇤el , then as than just approaching
thefar
same fixed
as the point. This
low-energy is discussed
theory goes their in e↵ects
detail can
by Strassler in his two
be “absorbed” into reviews
threshold 4] and forms
[3,corrections
theatbasis for the duality cascade.
the matching scale. The second, more serious, problem is that there should be a specific
value
20 of |⇤| which gives the best description of physics in the weakly coupled “magnetic”
http://video.ias.edu/pitp-2010
theory as the renormalization scale is raised. Currently it is not known how to fix the correct
value of |⇤| for Seiberg duality. NDA suggests that we should choose |⇤| ⇠ ⇤el ⇠ ⇤mag .
If it is possible to select a theory where the correct value of |⇤| is much smaller than
⇤el we would have a “magnetic” theory that63 has a weak gauge coupling at the scale where
it matches on to the underlying strong dynamics. How can this be possible? Recall that
the physics of the “magnetic” phase is parameterized by two coupling constants, the gauge
coupling and the dynamical Yukawa coupling of the “meson” to the dual “quarks”. Both of
We will return to the idea of understanding Seiberg duality from an effective theory perspective
in Section 8, where we describe recent work by Komargodski to relate the dual theories to nonlinear
sigma models from the 60s.
6 The phase structure of SQCD

Here’s a summary of Sections 3 through 5.
Free electric, V ∼ 1
log(ΛR)R
F = 3N
Non-Abelian Coulomb, V ∼ 1
R
3N
F = 2
Free magnetic, V ∼ log(ΛR)

R
F =N +2
F =N +1 s-confining, V = 0
F =N Quantum modified constraint, V = 0
F =N −1 Higgs, V = const.
ADS superpotential
F =1
F = 0 . Confining, V ∼ R
7 Siblings and Cousins of Seiberg Duality

• SO, Sp gauge groups.
• Kutasov Dualities.
8 Crouching Seiberg, Hidden Gauge Group

F*cking magnets, how do they work? – Insane Clown Posse, Miracles (2009)21
21
http://knowyourmeme.com/memes/fcking-magnets-how-do-they-work
64
Despite being rigorously tested, the structure of the Seiberg magnetic dual is surprising. Why
should the infrared effective theory describing confined degrees of freedom have some new emergent
SU (N − F ) gauge symmetry? In 2010 Komargodski presented an alternate motivation for the
duality from the point of view of 1960s meson physics [58].
8.1 Mini review: nonlinear sigma model

From QCD we are familiar with the idea that gauge degrees of freedom can confine and disappear
in an effective theory. For example, in the SU(3)×SU(3)→SU(3) nonlinear sigma model, the
quark degrees of freedom are confined leaving only the (pseudo)-Goldstone excitations in the light
spectrum. Let us quickly review the steps for the nonlinear sigma model for a breakign pattern
G → H. This is a digression, but it will be useful in what follows.
1. Pick a vacuum state, ⟨ϕ⟩ = ϕ0 . In the case of the pion NLΣM, this is ⟨U (x)⟩ = f 13×3 .
2. Transform it by one of the broken generators, g ∈ G/H. We thus have U (x) → gL U (x)gR† .
For the broken axial transformations, we have gR† = gL ≡ g, where we may write g = eiϵ T .
a a
3. Promote the transformation parameter to a field, call it a Goldstone boson. We thus take
ϵa → π a (x)/f . Thus the pion appears as U (x) = exp(2iπ a (x)T a /f )f 1.
The original field U (x) is nonlinear in the low energy degrees of freedom, π a (x). The reason
for this is that we had to constrain our field to live along the non-trivial vacuum manifold from
the G → H breaking pattern. The cost of imposing that our low energy degree of freedom always
points in the Goldstone direction is that it had to come in a rather nasty exponential.
This isn’t the only way to impose the constraint. There are many ways to do this, see for
example the texts by Donoghue [59] or Cheng & Li [60]. The key point, however, is that the low
energy physics is completely independent of how we choose to represent the NLΣM. This elegant
result was first presented by Haag [61] and more completely by Coleman et al. [62, 63].
8.2 Hidden gauge group

Let us return to the original question: how is it that we have an emergent gauge group in the low
energy theory? In our pion Lagrangian, all traces of the QCD gauge group disappeared because
they were confined. From this point of view, it seems ridiculous that the low energy theory should
have any new gauge degrees of freedom. It seems like there is no analog to the magnetic SU(F −N )
gauge symmetry in Seiberg duality. Or is there?
Back in the old days, when QCD was young and the Beatles were still on topping charts,
physicists wondered what to make of the higher resonances in the hadronic spectrum. We
Georgi papers on ρ... [64, 65]
Current algebra, reviewed in [66]... see also refs in flavor notes
9 Relation to AdS/CFT
Mention Steve’s Papers. Mention Csaba’s work. Important: Zohar’s paper (above)
65
From Adam: how to think about this: AdS/CFT takes your strongly coupled theory to a 5D
theory. The 5D theory can be deconstructed. This deconstruction is related to a chiral lagrangian
(breaking of gauge group between links). This chiral lagrangian can be understood as the chiral
lagrangian in Zohar’s ρ meson paper, which relates it to the magnetic gauge field. What’s not
obviously rigorous: relating the 5D theory and the deconstruction... do you require non-local
couplings? Also, is Zohar’s result rigorous in all cases?
10 Some current directions of interest

• In a rather elegant confluence, two of the remarkable dualities of the 1990s, Seiberg du-
ality and the AdS/CFT correspondence, come together in the so-called “warped deformed
conifold” construction. This is stringy construction wherein the supergravitational theory
is dual to a superconformal field theory which, itself, can be understood as a sequence of
Seiberg dualities. The latter has been dubbed the duality cascade, and is elegantly reviewed
in [4]. The gravitational side was described in the author’s A-examination22 .
• In 2006, Seiberg duality was employed to find a way to make generic models of dynamical
SUSY breaking in the ISS model [67]. This had been one of the elusive goals of SUSY
model building for two decades and came at the cost of the SUSY-breaking vacuum being
metastable. This model launched an entire model building industry that continues to thrive.
We present some aspects of this field when we discuss SUSY breaking.
• A very interesting question from a formal as well as model-building perspective is whether

unification survives across a Seiberg duality when the low-energy theory suffers from a
Landau pole before it unifies. This was examined by Abel and Khoze [68, 69].
• The relation between Seiberg duality and AdS/CFT has also recently been applied to model-
building. This is a particularly interesting direction which the author would like to pursue.
See: [70, 71].
• Recently a group of formal theorists have applied methods from algebraic geometry to better
understand the classical moduli space of SQCD [72]. The author’s ignorance in this subject
has left him to pronounce the paper’s title, much less understand it.
11 Unsorted ISS and SUSY breaking notes

11.1 Questions originating from Yael’s chiral ISS model
(Thanks to Yael for really clarifying things that I was confused about. She points out 0705.1074
for a discussion of some of the details of ISS, especially incalculability)
s-confinement: while the phenomenon is interesting physically, this has nothing to do with
why this a nice playground for model building. (cf TY and John’s paper.) The reason why we like
it is that it is easy. Csaba wrote out all s-confining theories and the dualities are easy. In general,
22
http://www.lepp.cornell.edu/~pt267/files/documents/A_ks.pdf
66
it’s difficult to write out the dual theory models that aren’t s-confining. For example, there are
tricks like deconfinement (see Pouliot) that one has to use to get these duals. (Deconfinement:
trick to use s-confinement to engineer dualities.)
The definition of ‘chiral’ that Yael uses is that one cannot write a mass term for all the fields.
In this sense it furnishes a chiral Weyl fermion. For example, the tensor in her SU(6) model cannot
obtain a mass term and so has a chiral fermion. Note that these chiral fermions cannot be used
to furnish the SM matter particles since this is prohibited by tree-level no-go theorems for single
sector susy breaking.
Why are these models interesting, then? The point is that the ISS susy breaking sectors are
vectorlike. One might like to check to see if anything interesting happens if the sector is chiral.
For example, in old-style (global SUSY breaking minimum) models, the chiral models were very
different from the vector models because they weren’t affected by the Witten Index theorem.
Incalculability: Incalculability has to do with higher order corrections to the Kahler potential.
In old-school models with global SUSY-breaking minima, it was otfen the case that one could
construct a model where one knew that a SUSY-breaking global minimum existed, but nothing
more. For example, the ADS potential with its flat directions lifted by a tree-level potential. In
general the region where the minimum occurs has no well controlled expansion, though one can
see that the F-term equations are not satisfied so that SUSY is broken. On the other hand, one
can usually tune the tree-level potential to consider regions where vevs are large and the theory
is weakly coupled. (See the 3-2 model.)
In ISS like models, incalculability is a question of whether one can be sure that all mass-
squareds at the SUSY-breaking local minimum are positive. In other words, the local minimum is
stable. In ISS this is done by controlling the SUSY breaking effects through a parameter µ. All of
the field vevs are proportional to a power of µ so that the higher order corrections to the Kahler
potential are also controlled by µ. One can then tune µ so that the higher order corrections are
under control and the theory is perturbative.
Caveats: Yael warns that this is not necessarily a fruitful business to enter at this time, not
only because one should be data-driven. She points out the gaugino mass problem which is still
has not been solved in a satisfactory way. Further, she notes that while ISS has open up a wide
range of models to play with, as a whole they haven’t yet added anything in terms of fully viable
models. She says that we already have a decent model, MGM. (She did say that it’s still valid to
look for a toy project to learn how to use these tools.)
N = 2 Duality
12 Seiberg-Witten
Reviews: Bilal [22], Alvarez-Gaume [20]
67
13 Gaiotto Dualities
Breaking SUSY
14 SUSY Breaking: history
14.1 SUSY breaking
If you’re reading this then you’re already familiar with the theoretical features of supersymmetry
as an appealing model for physics beyond the TeV scale. At low energies, however, SUSY is
clearly broken. In order to preserve the good features of SUSY, we would like SUSY to be broken
spontaneously (i.e. ‘softly’ in terms of phenomenology) rather than explicitly. Attempts to write
down realistic SUSY-breaking models for the Standard Model are immediately confronted with
problems associated with the supertrace rule, i.e. that the sum of all boson masses minus the sum
of all fermion masses must vanish. Since we haven’t discovered any very light scalar partners to
the leptons or quarks, this imposes the usual modular structure for SUSY breaking.

. messenger

SUSY MSSM
The two questions are (1) how to build a model for the SUSY-breaking sector, and (2) how to
mediate this to the MSSM. We will not say anything about this second question and, for the
remainder of this document, assume gauge mediation which we review in Appendix 16. For the
first question, any student of beyond the Standard Model physics will tell you that breaking su-
persymmetry is not as easy as one might otherwise think. The simplest model of spontaneous
SUSY breaking presented in the literature is the O’Raifeartaigh model which contains three su-
perfields and a very specific superpotential. It turns out that one really has to work hard to kill
supersymmetric vacua! In other words, SUSY-breaking vacua appear to be highly-non-generic.
We will return to this momentarily.
14.2 Dynamical supersymmetry breaking

Dynamical symmetry breaking (DSB) is an elegant idea in which spontaneous symmetry breaking
is realized via the vaccum expectation value of a composite field, e.g. the BCS theory of super-
conductivity. The key point is that the field which obtains a vev is not a ‘fundamental’ field in
the theory and is the result of strong coupling. Dynamical breaking of supersymmetry was first
proposed by Witten in 1981 [73, 74] as a solution to the Hierarchy problem.
“Wait a second,” you ask, “I thought that supersymmetry itself is a solution to the Hierarchy
problem?” Indeed you are correct. The point is that naı̈vely imposing SUSY breaking at the TeV
scale—where it provides a reasonably natural solution to the Higgs mass hierarchy (mH ≪ MPl )—
still leaves one with the question of why the SUSY breaking scale should be so much lower than
68
the fundamental scale, MSUSY ≪ MPl . Obtaining this hierarchy suggests that supersymmetry is
not broken at tree-level, but rather by quantum corrections. In other words, the theory’s vacuum
is manifestly supersymmetric at tree-level but is only rendered non-supersymmetric through the
dynamics of the theory itself. Further, the powerful non-renormalization theorems in supersym-
metry state that if a theory is supersymmetric at tree-level, then it is supersymmetric at all orders
in perturbation theory. Thus the only way for SUSY to be broken dynamically is through nonper-
turbative effects. Since these effects go as e−8π /g , we see that they are strongly suppressed and
2 2
we can hope to explain the hierarchy between MSUSY and MPl .

It was quickly discovered, however, that it is very difficult to build a straightforward and re-
alistic theory of dynamical supersymmetry breaking. Models end up requiring a special structure
on top of a chiral gauge theory. Models based on supersymmetric SU (N ) gauge theories (“super-
QCD,” or SQCD), i.e. what one’s first option for building a DSB sector, run into problems with
strong coupling leading to ‘non-calculability.’ The nail in the coffin for such theories is the obser-
vation that the Witten index for such theories is N , i.e. there exists N supersymmetric vacua23 .
To add insult to injury, we note that gauge mediation, for all of its features, tends to restore super-
symmetry somewhere in field space based on the above Witten index argument. How could one
hope to write down a calculable theory with a non-supersymmetric vacuum? There exist several
famous exceptional cases with silly names like (3-2), (4-1), and ITIY. It was clear, however, that
models realizing Witten’s original hope for dynamical SUSY breaking are very non-generic.
14.3 What makes SUSY-breaking non-generic?

We’ve now twice used the pejorative phrase ‘non-generic.’ Intuitively this means that models of a
given type don’t require any special constructions and are, in a sense, easy to create from simple
building blocks. In other words, models don’t require fine-tuning (in theory space). A particularly
useful notion of genericness that we will make use of is the statement that n equations for n
unknowns generically have a solution. This is of course not always true, but one would expect
this for equations that don’t have special relations (degeneracy).
In 1993 Nelson and Seiberg proved a very powerful theorem that showed that the conditions
for a SUSY breaking vacuum are connected to the existence of an R-symmetry [75]. The Nelson-
Seiberg theorem stated that

SUSY ⇒ ∃ R-symmetry

SSB R ⇒
SUSY,
or, in other words: R-symmetry is a necessary condition for the existence of a SUSY-breaking
vacuum and a spontaneously broken R-symmetry is a sufficient condition for such a vacuum.
This was able to shed light on the difficulty of constructing generic SUSY breaking models.
In order to have a SUSY breaking vacuum, Nelson and Seiberg tell us that the theory must have
an R-symmetry. We know, however, that gaugino masses explicitly break R-symmetry. We must
thus consider the case where the R-symmetry is spontaneously broken, which is fine since this
automatically implies the existence of a SUSY-breaking vacuum. On the other hand, we know
that spontaneously breaking the R-symmetry would give us a Goldstone boson. Thus either
23
This Witten index argument also imposes a chiral structure theories with SUSY breaking global minima.
69
we have a preserved R-symmetry and massless gauginos or a spontaneously broken R-symmetry
and a massless R-Goldstone boson. No matter what we end up with a massless particle which is
unobserved! We could try to be more sophisticated and appeal to gravity: since gravity hates con-
tinuous symmetries24 , we might expect that gravitational effects will give mass to the Goldstone
and save us [76]. Unfortunately, such effects would usually be far too small to give phenomeno-
logically acceptable masses (though see Dine, Nelson, and Shirman [77] for a counterexample).
Alternately, formal theorists might argue with a result by Banks and Dixon that there can be
no such thing as an exact global symmetry in string theory [78]. We are finally forced into the
conclusion that ‘generic’ theories of supersymmetry have supersymmetric vacua. This explains
why it was so damn hard to construct pretty SUSY-breaking models.
14.4 Metastable SUSY breaking

The ‘modern’ (after 2006) approach to this problem is that “when theory gives you lemons, make
models based on lemonade.” We shall squeeze out of the problem of non-generic SUSY-breaking
global vacua by looking instead at models with a local SUSY-breaking vacuum. Until this point
we had assumed that all the vacua we were considering were global, i.e. the ‘true’ vacuum of the
potential. This is sensible since otherwise we’d have to worry about tunneling to the true vacuum25
[79, 80]. However, if we want to hold on to the principle of being able to construct generic models,
then metastable SUSY breaking is inevitable. If you are particularly high-brow—which the author
is not—then you might also appeal to gravitational motivations such as the cosmological constant
(“it’s hard to make sense of deSitter space,” a string landscape).
It is worth noting that models with metastable SUSY-breaking vacua have been around for
some time. Ellis, Llewellyn Smith, and Ross constructed a model with a classically metastable vac-
uum in 1982 [81]. Around fifteen year later, Dimopoulos, Dvali, Rattazzi, and Giudice developed
the first metastable SUSY-breaking models based on pseudomoduli space, i.e. the “quantum-
modified” moduli space [82]. Unencumbered by the burden of finding a SUSY-breaking vacuum,
these models were much simpler to construct: one could use the framework of SQCD straight
out of the box. These theories have an explicit R-symmetry breaking which is realized as an
accidental R-symmetry in the low-energy theory26 . Finally, there is generally a small parameter
ϵ which parameterizes the explicit R-symmetry breaking and the separation of the SUSY and
SUSY-breaking vacua in field space. One can dial in a convenient value for ϵ to evade vacuum
tunneling problems.
These models, however, were faced with the practical problem of calculability. Supersymmetric
QCD in the asymptotically free regime is IR confining. Thus it is not obvious what the low-energy
24
For example, a black holes might eat something charged under a global symmetry and then forget about that
charge. Only if symmetry is gauged would the black hole take on the charge of the object it ate. Compare this to
the popular Kirby video games by Nintendo.
25
There’s a lot of interesting physics involved in such a tunneling. If a bubble of space tunneled to the true
vacuum, there would be a boundary which is stretched over the potential barrier. The volume (∼ R3 ) would live in
a lower-energy state, but its boundary (∼ R2 ) lives in a higher-energy state. Thus for small bubbles the boundary
energy would dominate and would cause the bubble to collapse, while for larger bubbles the volume would dominate
and eventually expand to eat the entire universe.
26
Note that typically instanton effects only break R-symmetry to a discrete subgroup so that more work needs
to be done to allow explicit gaugino masses.
70
degrees of freedom are in this nonperturbative regime and it’s not clear if we can even say that
SUSY is broken since higher order terms in the Kähler potential might ruin one’s construction. It
seems like we’re at a loss for constructing useable models.
In 2006, Intriligator, Seiberg, and Shih (ISS) found a way around calculability by [67] using
the powerful and surprising electromagnetic duality in SQCD discovered by Seiberg [56] (see
Intriligator and Seiberg’s lectures [2] for a review). This so-called Seiberg duality connects an
SU (N ) ‘electric’ theory with F flavors to an SU (n) = SU (F − N ) ‘magnetic’ theory in the same
universality class so that the two theories flow to the same IR fixed point and so describe the
same low-energy physics. These two Seiberg dual theories can be chosen so that one is UV-free
while the other is IR-free, thus allowing us to work perturbatively both in the UV and IR limits.
Intriligator, Seiberg, and Shih were then able to construct a dynamical supersymmetry breaking
ultraviolet theory while maintaining the existence of a long-lived metastable vacuum in the IR.
(Technically the model-building proceeded the other way around, starting with the IR theory and
then making it dynamical in the UV. We will go over this in Section 20.)
The ISS model opened the floodgates for a new wave of model building for metastable SUSY-
breaking vacua. Using Seiberg duality and simple gauge groups, theorists could produce generic,
calculable models and pit them against each other in a beauty pageant of which model is more
elegant than the next. The swimsuit competition of such a beauty pageant, however, is always
the ability of a model to reproduce realistic phenomenology, and it turned out that there were
still a few features of ISS-type models that physicists were having a hard time ironing out.
14.5 Problems with gaugino masses

The main problem was that ISS builders ended up having to use every trick they could muster
to create models with phenomenologically acceptable gaugino masses. Even though R-symmetry
is broken in these metastable models, it turned out that the gaugino masses of any semi-realistic
direct mediation model couldn’t be made to be very heavy without, in turn, pushing the scalar
masses to be dramatically heavier. See Ookuchi for a recent review [83]. This appears to be
true even after the residual discrete R-symmetry (which still protects against gaugino masses) is
broken. This would be nearly automatic disqualification from the SUSY model beauty pageant
since any large scalar masses would contribute to corrections to the Higgs and would thus re-
introduce our old nemesis the Hierarchy problem (defeating the whole point of having SUSY to
begin with).
Now the lesson here is that history tends to repeat itself. Just as it seemed unsettling 15 years
ago that it was so hard to find generic models of SUSY breaking, it was unsettling more recently
that it was so hard to find ISS models of metastable SUSY breaking without ‘anomalously small’
gaugino masses. Once again physicists were faced wondering why such a thing would not be
generic, and once again a clever duo was able to devise a theorem based on deeper feature of the
theory to explain this.
Just as Nelson and Seiberg’s 1993 theorem made clear that the difficulty in constructing SUSY-
breaking vacua was rooted in R-symmetry, Komargodski and Shih wrote down a theorem in 2009
that explained why gaugino masses were so small in ISS-like models [84]. We will thoroughly dissect
their derivation in Section 18. The result, however, is that gaugino masses vanish identically at tree
level if the pseudomoduli space is everywhere non-tachyonic (i.e. stable). It is rather surprising
71
that the pseudomodulus topology is related to the gaugino mass, but the result is able to shed light
on the ‘anomalously small’ gaugino masses encountered by those who tried to construct realistic
ISS models.
15 Dynamical Supersymmetry Breaking

We now remind ourselves of the basics of dynamical supersymmetry breaking (DSB). We shall
roughly follow David Shih’s lectures from TASI 0927 Review articles on dynamical SUSY breaking
models before ISS include those by Poppitz and Trivedi [85], and Shadmi [86]. During the [long]
time that this document was being prepared, Dine and Mason also published an excellent review
that touches on many current topics [8].
15.1 Motivation
The main motivation for dynamical SUSY breaking is understanding why there should be such
a hierarchy between the SUSY breaking scale M
SUSY and the Planck Scale MPl . Naturalness28
suggests that that M ∼ TeV ≪ MPl . As quantum field theorists we can conjecture that
SUSY
maybe this is because SUSY is broken at a higher order in perturbation theory, e.g. radiatively
by multi-loop effects. This does not work in supersymmetry since holomorphy tells us that the
superpotential W is not perturbatively renormalized. Thus if SUSY is unbroken at tree level, it
is left unbroken at all orders in perturbation theory.
The hope, then, is that one might be able to dynamically generate the TeV scale from a much
higher UV scale using nonperturbative effecits in much the same way that the confinement scale
ΛQCD is generated in QCD. In other words, we hope to construct a model where
= MUV e−1/g ≪ MUV .

2
M
SUSY
The primary analogue here is the breaking of chiral symmetry in QCD from the condensation of
quarks to form a QCD vacuum that [spontaneously] breaks the axial SU (3)A symmetry dynami-
cally.
Our ultimate goal is to be able to find a nice model of DSB and use it to build a model of
gauge mediation to write down a viable theory of nature.
15.2 Toy example: SUSY QM

This section is borrowed from the review [8].
Consider a quantum mechanical system with two operators

1
Q1,2 = (σ1,2 P ± σ2,1 W (x)) . (15.1)
2
27
Recordings available at http://www.colorado.edu/physics/Web/tasi09_annc.html.
28
Technically we mean ‘technical naturalness.’
72
By construction these satisfy {Qi , Qj } = δij H where
( )
1 2 2 dW
H= p + W + σ3 . (15.2)
2 dx
This turns out to already exhibit many of the features of 4D SUSY field theories. For example,
if W has a zero then the system has a supersymmetric ground state which is preserved to all
orders in perturbation theory. Ref. [8] gives the example of a harmonic oscillator, W = ωx so
that V = 21 ω 2 x2 . The ground state energy gets a bosonic zero-point contribution 12 ℏω, but also
‘fermionic’ contributions from the ℏσ3 terms which cancel: ∆E = ±ℏω.
Now let’s start to play with what could happen non-perturbatively. The condition for unbroken
SUSY is Qi |ψ⟩ = 0. For Q1 ,
( )
d
i + iσ3 W ψ = 0. (15.3)
dx
We can even solve this:

∫x
dx′ W (x′ )σ 3
ψ = e± −∞ ψ0 . (15.4)
If |W | → ∞ as x → ∞, the state is only normalizable for W odd.
15.3 Basics
Let us start by being very up-front. There are three types of DSB... WHAT?
. ⟨X⟩
73
V
c<0
c>0
. ⟨X⟩
V
x2
. ⟨X⟩
. ⟨X⟩
15.4 The 3-2 Model

Reviewed in Nelson, Dine, ...
74
15.5 The ITIY Model
15.6 Tools for Noncalculable Models
See ADS phenomenology paper: [87] See Hitoshi’s paper using vectors: [88]
16 Gauge Mediated SUSY Breaking

In this appendix we review features of gauge-mediated supersymmetry breaking (GMSB). The
canonical reference for traditional gauge mediation models and their phenomenology is the review
by Giudice and Rattazi [89]. Over the decade since the review was written, however, there has
been significant model-building progress in this direction. After briefly reminding ourselves of the
key features of gauge mediation, we mention relevant pieces of these recent developments. Finally,
a very powerful tool for calculating masses in many gauge mediation models, so-called analytic
continuation into superspace, is reviewed in Appendix E. The material in this section is loosely
based on Patrick Meade’s lectures on gauge mediation from TASI 200929 .
Meade divides the history of gauge mediation into three eras.
1. 1982 – 1993. Gauge mediation originated with the papers by Dine and Fishler [90] and
Nappi and Ovrut [91]. It is now an old and simple idea but built a huge amount of literature
from models of dynamical supersymmetry breaking, e.g. the Affleck-Dine-Seiberg (ADS)
model. The are, however, not many working examples of such models.
2. 1993 – 2006. A watershed moment for gauge mediation was the Dine-Nelson-Shirman
model that built in the now-standard modular structure for supersymmetry breaking [92].
Combined with experimental hints for supersymmetry in the 1990s, this led to a boom in
the number of gauge mediation models.
3. 2006 – present. The Intriligator-Seiberg-Shih (ISS) model demonstrated new ways to
break supersymmetry dynamically with simple, generic models based on metastable vacua
[67]. This led to another boom in the number of gauge-mediated and dynamically broken
models. In 2008 Meade, Seiberg, and Shih generalized the definition of gauge mediation and
described the features that are common to all possible models with this structure [93]. For
the purposes of the present paper we will not explore general gauge mediation (GGM) any
further.
16.1 SUSY Breaking refresher

Recall that the supertrace rule tells us that
∑
STrM 2 ≡ (−)2j (2j + 1)m2j = 0, (16.1)
j
where j labels the spin of a particle and mj is that particle’s mass. This means that in a
(spontaneously-broken) supersymmetric theory, the sum of the masses of the fermion masses and
29
75
the sum of the boson masses, weighted by the number of degrees of freedom for each field, must
vanish. In particular, this implies that naı̈vely supersymmetrizing the Standard Model would
imply the existence of new scalars lighter than, e.g. the up and down quarks. This is a tree-level
relation and could be modified by loop effects (though such effects are small), but the typical
solution is a modular structure in which a hidden sector breaks supersymmetry and mediates such
breaking to the unbroken MSSM.

. messenger

SUSY MSSM
In the SUSY-breaking sector we can parameterize spontaneous supersymmetry breaking by the

vacuum expectation value of a higher component of a superfield30 . For example, we may give a
vev to the F term of a chiral superfield, ⟨X⟩ = θ2 F . This SUSY breaking is communicated to the
MSSM via messenger fields at some mass scale M . At low energies in the visible (MSSM) sector
one would observe that supersymmetry is broken at the scale M . In this regime one can write
down effective operators for the masses of the sfermions and gauginos,
∫
X †X † F2
mQ ∼ d4 θ
2
Q Q ∼ (16.2)
M2 M2
∫
X F
mλ ∼ d2 θ Wa Wa ∼ . (16.3)
M M
This gives us a wide range of scales to play with. At the end of the day we want (F/M ) ∼ 100GeV
for a natural solution to the Hierarchy Problem. The naı̈ve choice is to use M = MPl and assume
gravity mediation. However, this leads to severe problems with flavor since gravity doesn’t respect
global symmetries. Further problems can arise due to incalculability in the regime where gravity
is strongly coupled. We thus want to find a mediator with a lower scale.
A general feature of SUSY breaking is the appearance of a goldstino field. When we promote
‘rigid’ SUSY to local supersymmetry, i.e. supergravity, the goldstino is eaten by the gravitino. In
most theories we don’t have to worry about this since the resulting gravitino mass is
F
m3/2 ∼ , (16.4)
MPl
so that even though this is ∼ 100 GeV in gravity mediation, it is gravitationally (i.e. very weakly)
coupled and would only be relevant for early universe cosmology. As we dial down the messenger
mass√ scale, however, we must correspondingly dial down F to maintain (F/M ) ∼ 100GeV. Thus
for F ∼ 1000 TeV, we are already left with
m3/2 ∼ O(eV). (16.5)
We thus end up with lighter particles and stronger couplings, and we have to consider the decay
of our otherwise-LSP to gravitinos (‘gravintii’).
30
Recall that vevs for the lowest component of a superfield do not break SUSY.
76
16.2 Set up and features
In the gauge mediation scenario SUSY-breaking is transmitted to the MSSM via gauge fields such
that SUSY is restorted in the MSSM sector in the limit when the gauge coupling is taken to zero.
These are naturally flavor-blind so we avoid many of the tight flavor-constraints in the general
MSSM. We ensure calculability by choosing a low messenger scale. Let’s go ahead and build a
gauge mediation model as a concrete example.
16.2.1 Write down a SUSY-breaking sector

Let’s start with the original O’Raifeartaigh model.
W = f X + mΦ1 Φ2 + yXΦ21 .

SUSY (16.6)
We know that at the minimum of the resulting potential there must exist a nonzero F term,
which we shall take to be F = FX ̸= 0. We further take the limit where m is the largest scale and
ϕ1 = ϕ = 2 = 0. We note that
⟨X⟩ = M + θ2 F (16.7)
is undetermined, i.e. it is a flat direction.
16.2.2 Have this sector talk to the MSSM via gauge fields
Now that we’re armed with ⟨X⟩, we would like to couple this SUSY-breaking field to a messenger
sector that is charged under the MSSM. We’ll populate our messenger sector with two left-chiral
superfields, φ and φe which transform as a 5 and 5 of SU (5). These have to be chosen to form
a vectorlike representations since this allows them to have large Dirac masses that can become
heavy31 . We introduce the hidden sector superpotential
e
Whidden = Xφφ. (16.8)
From Eq. (16.7) we know that this gives a mass of mΨ = M to the Dirac spinor Ψ formed from
e The scalar potential is
the Weyl spinors in φ and φ.
V = ⟨X⟩2 φ† φ + ⟨X⟩2 φ
e† φ e = M 2 (φ† φ + φ
e + ⟨X⟩θ2 φφ e† φ)
e + F φφ.
e
This leads to a mass matrix

( 2 )
M F
Mφ2 ∼ ,
F ∗ M2
e† )T . The eigenvalues of this matrix are

which acts on (φ, φ
m2φ± = M 2 ± F. (16.9)
31
Otherwise, purely chiral fermions would be protected by chiral symmetry and would be unacceptably light.
77
We can see that the breaking of SUSY in the hidden sector from an F -term vev is transmitted as
e are charged under
a splitting in the masses of the messenger sector scalars. The fields φ and φ
the MSSM. We want to ensure that these fields don’t obtain a vev that would break color or spoil
electroweak symmetry breaking, so we require
m2φi ≥ 0 ⇒ F ≤ M 2 .
This now completes the information that we need for the messenger sector.
16.2.3 Feed this into the MSSM via gauge interactions.

The messenger fields can then transmit SUSY-breaking effects into the visible MSSM sector
through couplings to the gauginos and scalars, inducing SUSY-breaking masses in these fields.
These are, of course, precisely the Standard Model superpartners whose masses we want to lift.
The tree-level gaugino coupling takes the form:
λi .
ψφ
From this we can build loop-level diagrams that contribute to the gaugino mass:
For this diagram explicitly drawn the SUSY-breaking mass insertions. From now on we’ll be
rather lazy and leave that implicit. To get the right insertions one can check that the Weyl arrows
take the right form. One may check that this gives a gaugino mass of
[ ]
αa F 1
m λi = n (1 + x) log(1 + x) + (1 − x) log(1 − x) , (16.10)
4π M x2
where x = F/M 2 ≤ 1 and n is the Dynkin index for the pair φ, φ (for example, n = 1 for the
N + N of SU (N )). We can now take the convenient limit x ≪ 1 for our minimal SU (5) model,
αa F
m λa = n . (16.11)
4π M
78
The scalar masses are much more difficult. The scalars have no direct coupling to the messenger
fields. These couplings are only induced at one-loop, thus the SUSY-breaking masses given to the
scalars only occur at two-loop order. If you’re reading this document, you’re interested in BSM
model-building and probably never intend to calculate a two-loop anything. Fortunately, the
technique of analytic continuation into superspace (reviewed in Appendix E) will allow us
to calculate these masses to leading order in the SUSY-breaking in a slick and elegant way. Just
for the heck of it, here are the diagrams.
φ φ φ φ
. . . .
φ φ φ φ
. . . .
The result of those calculations are scalar masses

F2 ( α )2 F2 ( α )2
i x≪1 i
me 2 = 2 Ca n f (x) −−→ 2n 2 Ca , (16.12)
M 4π M 4π
where Ca are the quadratic Casimirs of the scalar representation32 . We’ve written f to mean a
complicated function of dilogarithms,
[ ( ) ( )]
1+x x 2x
f (x) = log(1 + x) − 2Li2 + Li2 + (x → −x).
x2 1+x 1+x
We won’t care33 about this function or the overall prefactor of 2. What is relevant for us is that
Eq. (16.11) and Eq. (16.12) tell us that our theory has m e ∼ mλ , even the loop factors match.
What this tells us is that the mass splittings come from the factors of Ca and the strength of the
gauge coupling.
This gives us a distinct phenomenology where the heaviest superpartners are those charged
under the largest SU (N ), e.g. colored superpartners are the heaviest, followed by those with
SU (2)L charge, and so forth. The above formulae tell us
Mλ1 : Mλ2 : Mλ3 = α1 : α2 : α3 (16.13)
and
4 3 3
m2qe : m2ℓe : m2Ee = α32 : α22 : α12 . (16.14)
3 4 5
A few general remarks are in order,
32
C1 = 0 for singlets, C2 = 3/4 for weak doublets, C3 = 4/3 for color triplets.
33
Those who would like to show off their calculational prowess can follow the calculation in the appendix of
Martin’s paper on generalized (gauge) messengers [15].
79
1. These relations are independent of the details of the SUSY-breaking sector and even those
of the messenger sector.
2. The gaugino, squark, and slepton masses are all described by the vev of the spurion X.
3. Flavor-changing neutral currents are automatically suppressed and CP violation is conserved
since each of the mass matrices are proportional to the identity and the A terms are highly
suppressed.
4. The µ term (see below) is protected by symmetries so that further model-building is required.
If one is particularly clever, one would object that we appear to have missed something in our
above analysis: one-loop diagrams coming from non-zero (due to SUSY-breaking) hypercharge
D-terms. This is protected, however, by an accidental, approximate symmetry
q↔q ℓ↔ℓ VY ↔ VY .
This symmetry is broken by the MSSM interactions, but the effects of this breaking only occurs
at high-loop order. For details in a more involved model, see Giudice and Dimopoulos [94].
16.2.4 Basic phenomenology

Let us now discuss the features of these [ordinary] gauge mediation models. If we have a set of n
e in the 5, then in the limit
messengers in SU (5), that is to say n fields φ in the 5 and n fields φ
F ≪ M 2 we have the SU (3)c − SU (2)L − U (1)Y hierarchy
( )
αi F
Mλi = n (16.15)
4π M
( )2 ∑ ( )
2 F αi
mϕ = 2n Ci [ϕ]. (16.16)
M i
4π
Note that as n → ∞, Mλ2 /m2ϕ → ∞, so the characteristic scale of the gauginos verses the scalars
can be very different. Within the gauginos and (separately) within the scalars, however, the 3-2-1
hierarchy is preserved. The real danger of n → ∞ are the presence of Landau poles in the MSSM
sector due to a large contribution to the running of the MSSM gauge couplings. A good rule of
thumb for SU (5) models is that n ≲ 5.
For the low SUSY-breaking scales in our gauge mediation models, the gravitino mass m3/2 ∼
F/MPl matters since this is generally the lightest particle in the theory. We then have to recognize
that the field that we would otherwise call the LSP is actually the NLSP and will eventually decay
e Phenomenologically we need to figure out how the gravitino couplings. These
into the gravitino, G.
are predominantly due to the goldstino (which is eaten by the gravitino via the Higgs mechanism)
whose couplings come from the conservation of the supercurrent. The goldstino Lagrangian takes
the form
1 e
L = − JQµ ∂µ G (16.17)
F
1 [ 2 ]
e + ··· .
= (mψ − m2ϕ )ψϕ + mλi λi σ µν Fµν
i
G (16.18)
F
80
Note that the mass terms in the brackets also depend on F so that the expression on the right-
hand side is well defined in the F → 0 limit. This gives us two types of NLSP decay modes,
depending on the type of NSLP.
e
γG
m5χ0
χ0i e
2G Γ∼ F2
e
hG
.
m5τe
τe e
τG Γ∼ F2
For F ≳ 1000 TeV one would expect the NLSP to be collider-stable. For F < 100 TeV one gets a
prompt decay to the gravitino. For intermediate scales one gets a decay inside the detector which
may be measurable as a displaced vertex. The take-home phenomenological lesson is that the
‘smoking gun’ signal for ordinary gauge mediation models are photons plus missing energy34 .
In practice this is enough to go an talk to your favorite experimentalist. It’s important to talk
to an experimentalist who can tell you about the actual assumptions going into what they call
gauge mediation since experimental collaborations typically make assumptions about parameters.
For example, CDF and D0 assume
F
Mmess = 2
M
nmess = 1
tan β = 15
µ > 0.
16.3 EWSB and the µ-Bµ problem

It is worth mentioning the well known µ-Bµ problem of gauge mediation, first identified by Kim
and Nilles before I was born [96]. The µ parameter is the only SUSY-preserving parameter with
dimensions of mass and hence its natural lives at the Planck scale, MPl , while the Bµ parameter is
a soft SUSY-breaking term, Lsoft = Bµ Hu · Hd + h.c.. Note that there are a few different notations
floating around in the literature
Recall the potential for the neutral scalar Higgs in the MSSM,
( ) ( ) ( )
V = µ2 + m2Hu0 |Hµ |2 + µ2 + m2H 0 |Hd |2 − Bµ Hu0 Hd0 + h.c.
d
1( 2 ) ( )2
g + g ′2 Hu0 − Hd0
2 2
+ (16.19)
8
34
Though this shouldn’t be taken too seriously since one can cook up non-supersymmetric models of new physics
that mimic this signature, e.g. [95]
81
This can be found in any self-respecting MSSM phenomenology review or textbook. Nepotism
leads us to suggest the MSSM review written in the mid ’90s by a promising young graduate
student [50]. We note in particular that there is no quartic potential in the direction |Hu0 | = |Hd0 |.
In order to obtain electroweak symmetry breaking, the origin of the neutral Higgs potential
must be destabilized without introducing a run-away direction. In other words, there should be
one direction with a negative (mass)2 , but we cannot have this both directions or else the lack of
a quartic potential in the |Hu0 | = |Hd0 | direction will lead to a run-away. We can ensure this by
taking the determinant of the mass matrix in Eq. (16.19) and imposing that it is negative,
2
|µ| + m2H −Bµ
u
< 0.
−Bµ |µ|2 + m2Hd
This imposes
( )( )
Bµ2 > |µ|2 + m2Hµ |µ|2 + m2Hd . (16.20)
In order to ensure stability, i.e. to avoid the run-away direction, we want to impose that the
(mass)2 is positive along |Hu0 | = |Hd0 |. This gives the constraint
2|µ|2 + m2Hµ + m2Hd − 2Bµ > 0. (16.21)
These two equations relate supersymmetric µ term and the soft SUSY-breaking Bµ term which
naı̈vely have nothing to do with each other. This is a first hint of the µ − Bµ problem. One can
check explicitly that there is no solution to Eqs. (16.20-16.21) for m2Hu = m2Hd . The natural choice
is to have m2Hu < 0 and m2Hu > 0. This can be seen by looking at the running of the soft-breaking
scalar masses, from which we obtain at leading order
( 2)
( 2 ) 6yt2 Λ ( 2 )
mHu = mHu 0 −
2
ln 2
e t − m2t .
m (16.22)
16π2 m
The up-type Higgs couples to the top (s)quark and so the negative renormalization has a large
coefficient. A more detailed discussion along with remarks about fine-tuning can be found in
Section 11.3 of Dine’s
√ textbook [10] or Section 4.5 of Terning’s textbook [5]. Let us assign the
vevs ⟨Hu,d ⟩ = vu,d / 2 with the relations vu = v sin β and vd = v cos β. Minimizing the Higgs
0
potential gives us the famous equations,

2Bµ
sin 2β = (16.23)
2|µ|2 + m2Hu + m2Hd
MZ m2Hd − m2Hu tan2 β
= −µ2 + . (16.24)
2 tan2 β − 1
One can already tell that there’s something strange in neighborhood of Eq. (16.24). In order for
the terms in this equation to avoid fine-tuning, each term must be roughly of the same order.
Thus this equation tells us that naturally,
MZ2 ∼ µ2 ∼ m2Hu,d .
82
Are you unhappy yet? The first term is the physical Z mass which lives at a well-investigated
scale, the second term is a supersymmetric term that appears in the superpotential, and the third
terms are part of the soft SUSY-breaking Lagrangian. Why should these scales all have to be at
roughly the same order? This is a manifestation of the Little Hierarchy problem.
We can play with the µ and Bµ parameters to see what we can do. Since we obtained elec-
troweak symmetry breaking radiatively (i.e. from the running of m2H0 ), one might hope that we
could play the same game and set µ = Bµ = 0 and generate them radiatively such that they exist
at the electroweak scale as EWSB seems to require. One natural symmetry that prohibits both
the µ and Bµ terms is a Peccei-Quinn-type symmetry [97] that sends
Hu,d → eiα Hu,d .
We then assume that the SUSY-breakign sector breaks this symmetry and we cross our fingers
that this produces the necessary values for µ and Bµ at the weak scale. It turns out that this
works out perfectly in gravity meduation and is called the Giudice-Masiero mechanism [98].
The µ term is generated by an effective operator of the form
∫ †
4 X Hu Hd
dθ ,
MPl
where ⟨X⟩ ∼ F θ2 and we get an effective µ term at the scale µ ∼ F/MPl which is at the order of
the soft SUSY-breaking terms. The Bµ term is generated from
∫
X †X
d4 θ Hu Hd
MPl
from which we get Bµ ∼ F 2 /MPl 2

∼ µ2 . That’s great.
Finally we return to gauge mediation, where this µ − Bµ problem is not so easy to solve. Recall
that we now have F ≪ 1011 GeV, from which we get µ and Bµ terms that are much too small.
At the very least, we know from bounds on the Higgsino mass that µ ≥ 100 GeV. If we forbid the
tree level µ and Bµ terms, then then natural value for a radiatively generated µ from the hidden
sector is
1 F
µ∼ . (16.25)
16π 2 M
We will see that this is not a problem to attain. The problem in gauge mediation is to simulta-
neously obtain a value of Bµ of the same order of magnitude. If we have a direct coupling to the
SUSY-breaking sector W = λXHu Hd , where λ generally can encode loop factors, we end up with
µ = λM and Bµ = λF such that Bµ /µ ∼ 10 − 100 TeV. One generically has the problem that µ
and Bµ show up at the same order in λ (e.g. loop order), which means they cannot both end up
at the weak scale35 . We know from our discussion of EWSB above that a large ratio of Bµ to µ
destabilizes the electroweak symmetry-breaking vacuum, or at least introduces fine tuning.
Dvali, Giudice, and Pomarol presented a model in 1996 [99] that highlighted both the nature
of the µ − Bµ problem in gauge mediation and provided a somewhat elaborate strategy to combat
35
If this statement doesn’t make sense then check it with the simple example we just presented.
83
it. The model introduces additional singlets to cook up a scenario where µ is generated by a
more complicated operator which manifestly cannot simultaneously generate a Bµ term, which
must then be generated at a higher-loop order. Such models additionally require a mechanism to
prohibit the operators above that would otherwise generate µ and Bµ simultaneously at a lower
scale.
A second strategy is based on the next-to-minimally supersymmetric Standard Model (NMSSM)
which is reviewed by Maniatis [100]. This involves throwing in a new weak-scale singlet whose
vev produces the µ and Bµ terms, but requires some extra structure to maintain electroweak
symmetry breaking.
A third approach is to use have large renormalization effects suppress Bµ while leaving µ
relatively unaffected. One such model by Roy and Schmaltz used the dynamics of the SUSY-
breaking sector to impose this suppression [101]. The model, however, relies on assumptions
about incalculable anomalous dimensions in the hidden sector.
A final approach is to live with the ‘natural’ µ2 ≪ Bµ hierarchy of gauge mediation to see
if there is another way out. Csáki, Falkowski, Nomura, and Volansky presented this idea by
showing that if µ2 m2Hu ≪ Bµ ≪ m2Hd , then one can still obtain electroweak symmetry breaking,
Eqs. (16.20-16.21) [55]. Such a relation can be engineered if the Higgs fields are directly coupled
to the SUSY-breaking sector.
16.4 Direct, semi-direct, extraordinary, and general

Gauge mediation was born in 1993 with Dine and Nelson’s model [92] which is more or less what
we’ve presented above. We refer to our simple SU(5) 5 ⊕ 5 messenger model as minimal gauge
mediation (MGM), or the general class of such models as ordinary gauge mediation (OGM).
In particular, this is a narrow umbrella where the hidden sector has a single field X obtaining a
SUSY-breaking vev (the mechanism is unimportant) which couples to the vector-like (meaning
they come as conjugate pairs, N ⊕ N) messengers ϕ, ϕe via Yukawa interactions,
WOGM = λij Xϕi ϕej . (16.26)
This model was born roughly at the same time as the World Wide Web (at CERN), and just as
we’ve seen a remarkable growth in the Internet, gauge mediation model-building has come a long
way.
Back in the 90s, along with denim jackets and the TV show Friends, the big question was
whether one could further simplify the modular structure that Dine and Nelson had established.
The messenger sector was valuable to ‘insulate’ the MSSM from the SUSY-breaking sector. One
is able to avoid strict constraints from the supertrace rule and flavor-changing neutral currents.
The cost, however, is a rather arbitrary messenger sector. Theorists were thus driven to try
to construct more elegant models that did away with the messenger sector by completely by
allowing the messenger fields to participate in the SUSY-breaking mechanism, i.e. to incorporate
the messenger sector into the SUSY-breaking (‘hidden’) sector. This is called direct gauge
mediation (DGM).
Note the historical logic here: the original attempts to build dynamical SUSY breaking models
were also ‘single sector’ but were considered unpalatable since it was so hard to find a realistic
84
model. This was because naı̈vely building the ‘simplest’ models invoking only the paradigm of
DSB would never have led one to consider what would (again, naı̈vely) seem like a very arbitrary
set up. Dine and Nelson demonstrated a new paradigm where a messenger sector is introduced to
insulate the MSSM from the ‘dirty laundry’ of the DSB sector. People then took this as a lesson
and went back to the ‘old-style’ DSB models but set them up in such a way that there is still an
effective separation between the MSSM and the DSB fields.
This brought back problems that were already apparent in the original DSB attempts of the 80s.
In particular, having a DSB gauge group which generates the SUSY-breaking scale is effectively a
very large flavor group for the Standard Model gauge fields. The running of the Standard Model
couplings is then enhanced by this flavor factor and they can become nonperturbative before they
unify. This is the so-called Landau pole problem.
The first viable direct mediation model was presented by Poppitz and Trivedi [102] based on
SU (N ) × SU (N − 2) gauge group. The gauge messengers of this model are charged under the
Standard Model, which is embedded in an unbroken flavor symmetry of the SUSY-breaking sector.
The model has a very large SUSY-breaking scale, ∼ 1010 GeV, because of the large N require
to embed the Standard Model36 . At such scales the effects of gravity mediation must be taken
into account, making this a kind of ‘hybrid’ model. One then has to do a lot of work to rule out
flavor-changing neutral currents.
Shortly after Arkani-Hamed, March-Russel, and Murayama developed an alternative model
closer to what we recognize as gauge mediation [103]. Their model utilizes a pseudomodulus X
which is lifted by a non-renormaliziable operator in the superpotential. The field can then get a
very large lowest-component vev while maintaining a small vacuum energy, i.e. ⟨X⟩ = M + F θ2
with M 2 ≪ F which suppresses supergravity contributions. They arranged for the Standard
Model-charged fields to get masses on the order of ⟨X⟩ so that their contribution to gauge coupling
renormalization only appears above the large scale M . This avoids the Landau pole problem and
saves perturbative unification. However, there was a leftover problem that afflicts both this and
the Poppitz-Trivedi model: there are Standard Model-charged fields below 105 GeV whose scalar
components get soft-masses on the order of 104 GeV. This contributes to the renormalization of
the squark and slepton masses at two-loop order and actually drive them to negative values at
low energies.
A third model by Murayama which appeared in short succession was the ‘first phenomenologi-
cally viable’ model of direct mediation and was the gold-standard for direct mediation for about a
decade afterward [104]. The light SM-charged fields in this model do not have large soft masses so
do not make large negative contributions to the squark and slepton mass renormalizations. Fur-
ther, the model is completely chiral and one does not have to forbid mass terms for the messenger
fields by hand, as one had to in the previous models.
The modern era of gauge mediation (post-ISS) has brought more diverse directions, returning
to the modular structure of OGM model (and how this can again teach us about building DGM
models). The first ISS-type models based on vacua whose metastability are established near the
origin via Seiberg duality were developed before Christmas of 2006. Murayama and Nomura high-
lighted the role that metastable vacua play in relieving the Nelson-Seiberg R-symmetry condition
for model-building [105]. Kitano, Ooguri, and Ookouchi presented a direct mediation model with
36
SUSY breaking occurs due to non-renormalizable operators whose dimension grows with N and which are
suppressed by factors of MPl . This leads to the large SUSY-breaking scale.
85
string-inspired deformations [106]. Days afterward, the Three Musketeers developed a low-scale
direct mediation model based on the ISS framework [107].
With metastability making gauge mediation vogue once again, the IAS-Harvard axis started
thinking about jazzing up the framework itself. Seiberg, Volansky, and Wecht developed semi-
direct gauge mediation (sDGM) in which the messenger field exists in the SUSY-breaking sector
but does not itself participate in the breaking of supersymmetry [108]. Cheung, Fitzpatrick, and
Shih explored the consequences of generalizing the messenger sector by allowing its superpoten-
tial to include all renormalizable couplings to any number of hidden sector singlets Xk . They
called their framework (extra)-ordinary gauge mediation (EOGM) since their results can be
understood as a generalization of the ordinary gauge mediation (OGM) formulae. Since one can
perform unitary rotations on the Xk fields so that only one field, X, obtains an F -term vev, the
superpotential coupling SUSY-breaking to the messengers is given by
WEOGM = (λij X + mij )ϕi ϕej , (16.27)
where we’ve written the scalar (lowest-component) vevs of the supersymmetric fields Xk into
mij . The resulting formulae can be cast in terms of quantities identified with effective number of
messengers, by analogy to the OGM formulae Eqs. (16.15-16.16). They classified three types of
models within the EOGM framework:
1. det m ̸= 0
2. det λ ̸= 0
3. det m = det λ = 0.
Theories based on generalized O’Raifeartaigh models, including ISS-type models, fall under the
first class and will be our primary interest.
Next, Meade, Seiberg, and Shih [93] defined a framework for general gauge mediation, i.e.
the ‘essence’ of gauge mediation that is common to all known gauge mediation models (including
DGM). They used current correlators to generate sum rules that characterized the phenomenology
possible gauge mediation models. Under some of the models within gauge mediation one can
actually break the 3-2-1 hierarchy of sparticle masses, leading to very different phenomenology
from ordinary gauge mediation. The technique of using current correlators has since been used to
develop closed formulae for the soft masses of extraordinary gauge mediation [109] and semi-direct
gauge mediation [110].
17 The Nelson-Seiberg R-symmetry theorem

We would like to review the Nelson-Seiberg R-symmetry theorem [75] and provide a sketch of a
proof from Argyres’ well-written 2001 SUSY notes37 The theorem formally states:
Theorem 17.1 (Nelson-Seiberg). If one has a supersymmetric model so that the effective La-
grangian is generic and the theory calculable at low energies, then (1) the existence of an R-
symmetry is a necessary condition for a SUSY-breaking vacuum and (2) a spontaneously broken
R-symmetry is a sufficient condition for a SUSY-breaking vacuum.
37
These are available at http://www.physics.uc.edu/~argyres/661/index.html.
86
The first condition means that the theory is assumed to not have any special relations between
its parameters. We shall use the definition of ‘generic’ provided in Section 14.3, namely that
a system of n equations with n unknowns generically has a solution. The second condition of
calculability is more precisely phrased by saying that the low-energy theory must be described by
a Wess-Zumino model with no gauge fields. Such a theory of only chiral superfields would not
suffer from the problems of nonperturbative dynamics that appear in SU (N ) gauge theories.
In such a theory the scalar potential is given by the square of the F -term, Fi = ∂i W . As we
know the minimum of the scalar potential tells us whether or not SUSY is broken. If min V =
min |F |2 = 0 then SUSY is preserved in the vacuum, otherwise SUSY is broken. If we label
our chiral superfields by i such that our low-energy Wess-Zumino model is composed of fields Φi ,
i = 1, · · · , n then the condition for a SUSY-preserving vacuum is
∂i W (Φ1 , · · · , Φn ) = 0 ∀ i. (17.1)
This is a system of n complex analytic equations for n complex unknowns38 . Thus the system
generically has a solution and hence the theory has a supersymmetric vacuum. Boring. What else
can we do? The only tool that is really at our disposal is to play with global symmetries39 . We
can argue that Eq. (17.1) didn’t take into account the global symmetries that our theory might
have. Under such a global symmetry the superfields each have some charge, Q[Φi ] = qi . Typically
the superpotential must be invariant under this symmetry, imposing a further constraint on the
theory and naı̈vely giving us hope that perhaps we can get generic SUSY-breaking. Suppose for
simplicity that the symmetry is a U (1) and assume without loss of generality that the charge
q1 ̸= 0 (at least one such field must be charged in order for the symmetry to be nontrivial).
If the U (1) is preserved, then the vacuum is given by the state where all of the charged fields
must have vanishing vevs,
⟨Φi ⟩ = 0 if qi ̸= 0. (17.2)
If the first k fields Φi , · · · Φk have nonzero charges q and the rest have vanishing charge, then this
imposes k constraints. Restricted to the remaining subspace of unknown field vevs, superpotential
is still gives (n − k) generically independent equations for (n − k) unknowns. Thus the case for a
preserved global symmetry does not work.
We can consider what happens when the global symmetry is broken spontaneously, in which
case some of the charged fields are allowed to have nonzero vev. The superpotential as a term
in the Lagrangian, must still be neutral. We may incorporate this constraint by writing our
superpotential as a function of only n − 1 superfields,
−qn /q1 −q2 /q1
W (Φ1 , · · · , Φn ) ≡ w(Φ2 Φ1 , · · · , Φn Φ1 ), (17.3)
where all we have done is absorbed the Φ1 dependence into the condition that the superpotential
can be expressed in terms of variables that are uncharged under the global symmetry. Now,
38
Note that we have made use of the standard, but sometimes confusing, notation where we write the vev of a
field using the same notation as the field itself, i.e. Φ = ⟨Φ⟩ when it is clear from context that we are discussing
the vev. This saves a lot of clutter in the notation, but the reader must be a little more careful.
39
Even this is arguable under the banner of genericness, but the point is that we will be interested in R-symmetries
which are generic features in SUSY models.
87
however, we’ve just written everything in terms of a system of (n − 1) equations with (n − 1)
unknowns. SUSY vacua are still generic.
At this point it may look like we’ve exhausted our options, but there is a way out. We assumed
that the superpotential had to be neutral under this symmetry since it is part of the Lagrangian,
∫
L = ··· + d2 θ W (Φi ).
We note that if the superspace coordinate θ were charged under the symmetry, then W must also
be charged. This is precisely what occurs in the R-symmetry which is present in SUSY theories,
the superpotential has R-charge R[W ] = 2. Thus, for the case of an R-symmetry, Eq. (17.3) must
be modified to
2/r1 −r2 /r1 −rn /q1
W (Φ1 , · · · , Φn ) ≡ Φ1 w(Φ2 Φ1 , · · · , Φn Φ1 ), (17.4)
where we’ve written ri as the R-charge of the lowest-component field in Φi . The overall factor of
2/r
Φ1 1 must be included to maintain R[W ] = 2. Now we can see that Eq. (17.1) implies
2 2/r1 −1 −r /r −r /q
Φ1 w(Φ2 Φ1 2 1 , · · · , Φn Φ1 n 1 ) = 0 (17.5)
r1
−r /r −r /q
∂i̸=1 w(Φ2 Φ1 2 1 , · · · , Φn Φ1 n 1 ) = 0. (17.6)
The second equation is just the usual system of (n − 1) equations for (n − 1) unknowns, but the
first equation is an additional constraint imposing w(· · · ) = 0. This gives us a total of n equations
for (n − 1) unknowns and thus the system is overconstrained and generically does not have a
system. We thus conclude that supersymmetry must be broken. This concludes the simple proof
of the Nelson-Seiberg theorem.
18 The Komargodski-Shih Gaugino Mass Theorem

In Komargodski and Shih’s 2009 paper, ‘Notes on SUSY and R-Symmetry Breaking in Wess-
Zumino Models,’ they collect a series of theorems40 about ‘general O’Raifeartaigh’ models that
can appear as the low-energy effective theories of dynamical SUSY breaking models [84]. They
introduces a new technique for ‘generic’ model building by introducing “tree-level R-symmetry
breaking” and, most importantly for our present purposes, they elucidate the nature of the anoma-
lously light gauginos that appear in these models. This latter result is what we shall refer to as
the Komargodski-Shih theorem. In this section we shall review their derivation, following the
structure of their paper very closely.
18.1 Basics of SUSY breaking in Wess Zumino Models

We start with a general, weakly-coupled, renormalizable Wess-Zumino (WZ) model with a canon-
ical superpotential,
1 1
W = fi Φi + mij Φi Φj + λijk Φi Φj Φk , (18.1)
2 6
40
For proper attribution, some of their results were first discussed by S. Ray [111].
88
that breaks supersymmetry through tree-level F -term vevs. We shall call such models generalized
O’Raifeartaigh models. The conditions for a SUSY-breaking minimum are then41 :
1. There exists some i such that Wi ̸= 0. This just says that there is a non-vanishing F term
that causes the vacuum energy not to vanish, ⟨V ⟩ > 0. Note that the fields Φj which preserve
SUSY still have Wj = 0.
{
= 0 if ϕi preserves SUSY
Wi (18.2)
̸= 0 if ϕi breaks SUSY
2. The fields take their values at the minimum of the potential V . In other words,
Wij Wj∗ = 0. (18.3)
Recall that the fermion mass matrix (MF )ij = Wij at tree level, so this is just the familiar
goldstino theorem that the spontaneous breaking of supersymmetry leads to a massless
Goldstone fermion. (Recall that the only fields with Wj ̸= 0 are those that participate in
SUSY breaking.)
3. The boson (mass)2 matrix M2B must be positive definite, i.e. the vacuum is free of tachyons.
Let us recall that form of the boson (mass)2 matrix is
( ∗ )
MF MF F∗
Mb =
2
(18.4)
F MF M∗F
where Fij = Wk∗ Wijk . M2B is manifestly a positive semi-definite Hermitian matrix. For such a
matrix we may always write M2B = A† A for some A. This is obvious if we write
( )
e†i M2B ij ej = ê† U † (M2Diag )U êj . (18.5)
From this we arrive at a handy lemma,
Lemma 18.1. In any SUSY-breaking vacuum of a generalized O’Raifeartaigh model, if there exists
a massless fermion at tree-level, then its scalar superpartner must also be massless at tree-level.
Proof. From the above observation, we see that
w† M2B w = 0 ⇔ MB w = 0. (18.6)
Now suppose that MF has a zero eigenvector, v. This is, of course, a vector in field space. We
( )T
shall construct the bosonic vector v v ∗ . Then we observe that
( )( )
( ∗ ) M∗F MF F∗ v
v v ∗ = v T Fv + c.c. (18.7)
| {z } F M F MF v∗
w†
| {z } | {z }
M2B w
Since M2B is positive semi-definite, this expression must vanish otherwise one may perform a phase
rotation on v to make the right-hand side negative and hence inconsistent. Thus the scalar is also
massless.
41
see Appendix A for our conventions if you are confused about expressions like Wi j.
89
Note that even though we define our generalized O’Raifeartaigh model to be renormalizable,
the proof of this lemma never depended on this property and it turns out to actually hold for
any general polynomial superpotential regardless of renormalizability. From this we can also write
down two corollaries,
Corollary 18.2. If MF v = 0, then Fv = 0.
In other words, MF and F have the same null eigenvector.
Corollary 18.3. For a SUSY-breaking minimum,
Wijk Wi∗ Wj∗ = 0 (18.8)
Proof. For a SUSY-breaking vacuum, we have Eq. (18.3), which can be written as (MF )ij Wj∗ = 0.
To be precise, one can rotate the fields such that the SUSY-breaking linear combination is labelled
ĵ and Eq. (18.3) can be written as (MF )iĵ Wĵ∗ = 0 where there is no sum over ĵ. We thus have a
massless fermion associated with the Wĵ∗ direction. Applying the lemma above we then have
Fiĵ Wĵ∗ = − ⇒ Wij ĵ Wi∗ Wĵ∗ = 0. (18.9)
Rotating back to the original field direction we get precisely Corollary 18.3.
It turns out that not only are the scalar partners of the golstino massless, but it can be extended
to an entire pseudomodulus, i.e. a tree-level flat direction emanating from a SUSY breaking
minimum which obtains a potential from quantum corrections.
Theorem 18.4. The direction ϕi = ϕi + zWi∗ leaves the tree-level potential V [ϕi ] unchanged for
(0)
any z ∈ C, in other words, it is a pseudomodulus.
Proof. An earlier proof was provided by Ray in [111], but the notation is rather cumbersome so
we’ll follow the derivation by Komargodski and Shih. Under this field transformation,
1
δWi = ∂j Wi · δWj + ∂k ∂j Wi · δWk δWj
2
1
= Wij (zWj∗ ) + Wijk (zWj∗ )(zWk∗ ).
2
There are no other terms since W is renormalizable, i.e. Wijkℓ = 0. We know from above that
Wij Wj∗ = 0 (18.3)

Wijk Wi∗ Wj∗ = 0. (18.8)
From this we deduce that δWi = 0 and hence Wi is constant along this direction. This proof is
sufficient for our purposes, though Komargodski and Shih have a more general version of their
theorem in their appendix [84].
90
Now that we’ve proved the existence of a pseudomoduli space, we would next like to show
that we may perform a rotation on our fields such that we may write the superpotential in what
Komargodski and Shih refer to as the canonical form,
1 1 1
W = X(f + λab φa φb ) + mab φa φb + λabc φa φb φc . (18.10)
2 2 6
In this basis the SUSY-preserving fields φ have zero vev, ⟨φa ⟩ = 0 while the SUSY-breaking
pseudomodulus field X ∼ Wi∗ can be arbitrary.
Proof. We rotate our fields according to
ϕi = Uix X + Uia φa
such that
W = fi Uix X + fi Uia φa + · · ·
Thus we may identify f ′ = fi Uix and fa′ = fi Uia . Similarly we may define m′ = Uix Ujx mij and so
forth. Now expanding the ϕs about their vevs ϕa → ⟨ϕa ⟩ + ϕa and reabsorb factors of ⟨ϕa ⟩ into
the coefficients, e.g.
1 ′
λ ⟨ϕc ⟩ ≡ λab .
3 abc
The factors of 1/2 and 1/6 are part of the definition of the new parameters and take care of
permutations of the ϕ fields. We now only have to appeal to the equations above to explain the
form of Eq. (18.10). First of all Eq. (18.2) tells us that Wa = 0 and so there are no terms in W
linear in φ. Next Eq. (18.3) tells us that Wxx X = Waa = 0 and so W cannot have any XX or φX
terms. Finally, Eq. (18.8) tells us Waxx XX = Wxxx XX = 0 so that W cannot have any XXX or
φXX terms. This gives us the canonical form above.
More generally, we will use what Komargodski and Shih refer to as the generic form of the
generalized O’Raifeartaigh superpotential,
W = Xi fi (φa ) + g(φa ). (18.11)
Note the dependence on the genericness assumption: one could easily construct an O’Raifeartaigh
theory that does not take this form, for example one can take a superpotential in the canonical
form and do a rotation of the fields. Such a superpotential, however, would not be generic in
that the couplings would not be independent since they would be related to the couplings of the
original via the unitary transformation and hence would not be generic. For future reference, the
original ISS model is based on the case g = 0.
91
18.2 Tree-level SUSY and R-symmetry breaking
Komargodski and Shih now shift gears a little and introduce the idea of SUSY and R-symmetry
breaking ‘at tree-level.’ The main idea was to identify a set of models where one doesn’t have to
calculate the Coleman-Weinberg effective potential to check vacuum stability of the pseudomoduli
with the hope that this would be a particularly nice place to do realistic model-building. The
main result that we shall take from this, however, will be to identify an incompatibility of the
assumption of vacuum stability with gaugino masses.
Definition 18.5. A model breaks supersymmetry at tree level if
1. The pseudomoduli space is locally stable everywhere.
2. The Coleman-Weinberg potential on the pseudomoduli rises at infinity in every direction.
In other words, tree-level SUSY-breaking models are those where we don’t have to worry about
checking the stability of states along the pseudomoduli. We gan go on to define R-symmetry
breaking ‘at tree level.’
Definition 18.6. Further, a model breaks R-symmetry at tree level if, in addition to the
above conditions,
3. The pseudomoduli space breaks R-symmetry everywhere.
Thus for such models we would not have to calculate the details of the Coleman-Weinberg
potential to be guaranteed that SUSY and R-symmetry are broken in the vacuum. The second
condition requires some knowledge of the full potential, but only at large fields42 .
We now observe from the generic form of the generalized O’Raifeartaigh superpotential that
if g(φ) = 0, then the model cannot break R-symmetry at tree level.
Proof. If g = 0 then the theory has an R symmetry with R[Xi ] = 2 and R[φa ] = 0. Since we’ve
written our variables such that only the Xi fields have non-zero F -terms, Wa = 0. We note,
however, that this means
∂W
= Xi ∂a fi (ϕ)(φ) = 0,
∂φa
in other words, Xi must be a null eigenvector of Mai ≡ ∂a fi . Rescaling Xi then leaves the vacuum
energy unchanged, as one can see explicitly from the form of the potential obtained form the
generic form of the generalized O’Raifeartaigh superpotential,
∑ ∑
V = |fi (φ)|2 + |Xi ∂a fi (φ) + ∂a g(φ)|2 .
i a
This freedom to rescale Xi tells us that the origin {Xi = 0} is a connected element of any
pseudomodulus. Since we’ve shown that the Xi are the only R-charged fields, there is then
always a point on any pseudomodulus where R-symmetry is unbroken (the origin). Hence, by the
definition of “broken R-symmetry at tree level,” we see that for g = 0, R-symmetry cannot be
broken at tree level.
42
In this limit it can generally be computed using the techniques developed by Intriligator, Shih, and Sudano
[112]. For the case of a single pseudomodulus, X, it one show that the potential rises like log X times the [positive]
anomalous dimension of X [113].
92
That is the main result that we’d like to use to start discussing gaugino masses. Before pro-
ceeding let us first make a brief aside since part of the purpose of this document is to collect a
set of tools for metastable model building. Even for a g = 0 generalized O’Raifeartaigh model, we
may engineer it to have tree-level SUSY and R-symmetry breaking. The general idea is to add
new fields φ e and a g(φ, φ)e term to the superpotential that set the R-charges to ‘exotic’ values.
For simplicity, let us assume that the model in question respects an additional U (1) symmetry in
addition to R-symmetry that is spontaneously broken in the vacuum via ⟨φ⟩ ̸= 0. To construct a
‘tree-level R-breaking’ model, we may add ‘by hand’ additional fields φ e and an additional super-
potential term g(φ, φ)e such that both the U (1) and U (1)R are broken explicitly while maintaining
a nontrivial combination U (1)′R ⊂ U (1)R × U (1). As long as the F -terms associated with the new
e fields can be all be set to zero, this doesn’t spoil our tree-level SUSY breaking. The ⟨φ⟩ =
φ ̸ 0 vevs
then breaks R-symmetry ‘at tree level.’ This is illustrated schematically in Figure 2. Komargodski
and Shih give an explicit example of such a construction in their paper [84]
e
g(φ, φ)
U (1)R ×
. U (1) . ′R
U (1)
⟨φ⟩ . ⟨φ⟩
e
g(φ, φ)
U (1)
. R 1.
Figure 2: Schematic representation of how to build a R-symmetry breaking model out of an

R-symmetry preserving g = 0 model.
18.3 Application to Gaugino Masses and Model Building

At the risk of alienating the reader with re-writing the generalized O’Raifeartaigh superpotential
once again, let us write its canonical form Eq. (18.10) with relabeled variables that will make
things easier in the future:
1 e 1
W = f X + (λ e ab )φa φb + + gabc φa φb φc .
ab X + m (18.12)
2 6
This is nothing other than the relabelling λab → λeab , mab → m e ab , and λabc → ga bc. We will rotate
e
the matrices λ and m e such that they can be written as
( ) ( )
e λ 0 m 0
λ= me =
0 0 0 0
so that the ‘reduced determinant’ det(λX + m) ̸= 0 for generic X. This brings us to our very
important result,
Theorem 18.7 (Komargodski-Shih). For models with tree-level SUSY and R-symmetry breaking,
the reduced determinant is a constant function of X,
det(λX + m) = det m. (18.13)
In particular, if this expression is to be violated, then there must exist tachyonic directions at
values of X where the reduced determinant vanishes.
93
The value of this theorem is that the X-derivative expression on the left-hand side of Eq. (18.13)
is precisely what appears in the expression for the gaugino mass in theories of gauge mediation,
as we will show below.
Proof. Suppose Eq. (18.13) does not hold. Then we may write the right-hand side as a polynomial
in X,
∑
det(λX + m) = ci (λ, m)X i . (18.14)
i
Thus there exist values X = X0 ∈ C where det(λX0 + m) = 0. This means that there exists a
direction in field space v such that
(λX0 + m)v = 0. (18.15)
This v is a massless fermion direction. From Lemma 18.1, however, we know that this either
implies the existence of the massless boson in the same direction or else, according to the proof
of that lemma, there must be a tachyonic direction. This massless boson direction tells us that
Fij vj = 0, using the notation from Eq. (18.4) so that (e.g. see Corollary 18.3)
0 = Wk∗ Wijk vj = Wx∗ Wab vb = X0 λab vb ,
where we’ve used Wa∗ = 0 from Eq. (18.2). Combined with Eq. (18.15), this tells us that λv = 0
and hence mv = 0. This contradicts the assumption43 that det(λX + m) ̸= 0. Hence either
det(λX + m) cannot have zeroes at finite points in field space, i.e. it must be a constant function,
or there must be a tachyonic direction at X = X0 .
This theorem has an immediate and important consequence in models of gauge mediation
where the hidden sector is described by a generalized O’Raifeartaigh model. In such models a
subset of the φa fields are charged under the Standard Model gauge group and communicate the
SUSY breaking from the X field to the MSSM. Due to gauge invariance, the mass matrices for
these messengers must factorize at quadratic order and so one can apply the Komargodski-Shih
theorem to these fields independent of the rest of the hidden sector. This results in
det(λX + m)|mess. = constant.
Using the techniques of analytic continuation into superspace reviewed in Appendix E (or tra-
ditional two-loop calculations), one knows that the leading order (in SUSY breaking) gaugino
masses are given by
∂
mλ ∼ F † log det(λX + m)|mess. = 0. (18.16)
∂X
This leads the Komargodski and Shih to proclaim that, “This simple result shows that it is
impossible to build viable theories of gauge mediation with tree-level SUSY breaking, unless one
is prepared to accept an exacerbated little hierarchy problem and the attendant fine tuning coming
from very heavy sfermions.”
43 e and λ to ensure that (Xλ + m) is nondegenerate.
i.e. the entire construction where we defined λ
94
So the point is this: gaugino masses in gauge mediation are zero at leading order (in the
SUSY-breaking parameter) unless the pseudomoduli space is not locally stable everywhere. In
order to construct a realistic gauge mediation model, one requires a tachyonic direction somewhere
on the pseudomoduli space. Of course, this ‘somewhere’ should be away from the vacuum that
we populate, and this will be the game played by of most of this paper.
As a sanity check, we can ask ourselves about the models of gauge mediation that have been
around for 15 years. In minimal gauge mediation (MGM) (see, e.g. Dine and Nelson [92] and
the follow up paper with Shirman [77]), the superpotential takes the form
W ⊃ λX φ
eφ,
e
which is tachyonic at X = 0. A more recent manifestation, the direct gauge mediation (DGM)
models recently studied in the extraordinary gauge mediation (EOGM) scenario by Cheung,
Fitzpatrick, and Shih [114] also have tachyons at X = 0 required for mλ ̸= 0. This result is
broadly applicable for dynamical SUSY breaking with gauge mediation since such models are
often described by renormalizable Wess-Zumino models.
One can also wonder about models whose hidden sectors are not described by generalized
O’Raifeartaigh models at low energies. Such models can be strongly coupled or have non-
renormalizable Kähler- and superpotentials, and tend not to be calculable. Notable exceptions
such as Seiberg, Volansky, and Wecht’s semi-direct gauge mediation (sDGM) model [108] also
still have gaugino masses vanishing at leading order. This leads Komargodski and Shih toopenly
wonder if there is a way to generalize this result to non-canonical Kähler potentials, noting that a
hint may be that the leading-order contribution to gaugino mass is a superpotential term in the
effective action.
19 The Landau Pole Problem

See ADS phenomenology paper: [87]
Good summary in intro of: 0809.4437
20 ISS: Metastable SUSY Breaking

We now review the Intriligator, Seiberg, Shih model of metastable vacua [67]. The original paper
is very readable, though it assumes a core set of background material which is not yet found
in most textbooks. Fortunately, two of the authors provide excellent reviews in their lecture
notes on Seiberg duality [2] and supersymmetry breaking [1]. Their SUSY breaking review should
be taken as supplementary pedagogical notes that were tailor-made for understanding the ISS
model, though the topics are presented in such a way that their significance is only illuminated
in hindsight. Additional thoughts can be found in Dine’s Cargese lectures [7] or Shirman’s TASI
lectures [9].
20.1 Summary in words

The construction of the ISS model proceeds in three steps.
95
1. Construct a theory of chiral superfields were supersymmetry is broken at tree-level. We will
use a theory of matrix fields where supersymmetry is broken by the rank condition. We
will call this the macroscopic model I.
2. Promote this model to one with gauge superfields by gauging a global symmetry. This
generates new supersymmetric vacua, but we will do this in such a way that the vacuum of
the previous theory is preserved as a metastable vacuum. We shall call this the macroscopic
model II.
3. We then use Seiberg duality to realize this metastable supersymmetry breaking dynamically.
This will give us our ISS model.
We will follow the structure of the original paper, including detours to check the consistency of
what we are doing. We will ignore the generalization to SO(N ) and Sp(N ) groups.
20.2 Macroscopic Model I

We shall start by considering a theory where supersymmetry is broken by the rank condition.
This is reviewed pedagogically in Section 2.7 of [1]. The theory will have a global symmetry,
SU(N ) × SU(F ) × SU(F ) × U(1)B × U(1)′ × U(1)R ,
where these are a soon-to-be gauge symmetry, flavor symmetries (for quarks and antiquarks), a
baryon number charge, a U(1) that will be broken by the superpotential, and an R-charge. We
will be particularly interested in the case F > N . The fields and their representation under the
global symmetries of the theory are given by
SU(N ) SU(F )L SU(F )R U(1)B U(1)′ U(1)R

Φ 1 □ □ 0 -2 2
φ □ □ 1 1 1 0
e
φ □ 1 □ -1 1 0
The Kähler potential is taken to be canonical for these matrix-valued fields,
K = Tr φ† φ + Trφ
e† φ
e + Tr Φ† Φ. (20.1)
The most general superpotential compatible with the above symmetries is
e
W0 = hTr φΦφ, (20.2)
where h is some dimensionless coupling constant. We will also add an additional term to this
superpotential that explicitly breaks SU(F )L × SU(F )R × U(1)′ → SU(F ),
∆W = −hµ2 Tr Φ, (20.3)
96
where µ is a parameter with dimensions of mass and our resulting superpotential is W = W0 +∆W .
Our theory’s global symmetry is now broken to
SU(N ) × SU(F ) × U(1)B × U(1)R , (20.4)
where we recall that because the SU(F )L × SU(F )R × U(1)′ symmetry is broken explicitly, there
are no Goldstone bosons associated with it.
We can now check that supersymmetry is broken. The main idea is this: because F > N , the
F -terms cannot all be set to zero by the rank of the relevant matrices. Consider, in particular,
the FΦ term,
e − hµ2 1F ,
−FΦ† = hφφ (20.5)
where this is understood to be an F × F matrix relation. If one is uncomfortable with this, it’s
easy to write out particular components of the F -term by taking derivatives of W with respect to
e is an object of rank N while hµ1F
particular elements of the matrix fields. The first term, hφφ
is manifestly an object of rank F . Since F > N , these two terms cannot sum to zero and so the
scalar potential is manifestly greater than zero,
Vmin = (F − N )|h2 µ4 . (20.6)
We call this supersymmetry breaking by the rank condition.

We can use the global symmetries to parameterize our classical moduli space by
( ) ( ) ( )
0 0 φ0 e0
φ
Φ= φ= e =
φ T
, (20.7)
0 Φ0 0 0
where these are understood to be up to SU(N )×SU(F )×SU(F ) rotations. Note that we’ve written
the upper (left) blocks of these matrices to be N × N , so that Φ0 is (F − N ) × (F − N ) while φ0
and φ0 are N × N . We can now choose the vacuum that preserves as much of the global symmetry
Eq. (20.4) as possible,
Φ0 = 0 e0 = µ1N .
φ0 = φ (20.8)
This gives us a spontaneous breaking
SU(N ) × SU(F )2 × U(1)B × U(1)R → SU(N )D × SU(F − N ) × U(1)B ′ × U(1)R . (20.9)
The next thing that we’d like to do is to determine the Coleman-Weinberg effective potential,
which we introduce in some detail in Appendix B. The main question we want to answer is whether
or not our SUSY-breaking vacuum is stable on the moduli space under quantum corrections. In
order to do this, we know that we need the mass spectrum of of the fields. To figure this out, we
expand about the vacuum Eq. (20.8):
( ) ( ) ( )
δY δZ T µ + √12 (δχ+ + δχ− ) µ + √12 (δχ+ − δχ− )
Φ= φ= φ= ,
δ Ze δ Φ̂ √1 (δρ+ + δρ− )
2
√1 (δρ+ − δρ− )
2
(20.10)
97
where the division into N and (F − N ) blocks are as before. Our choice of parameterization will
simplify (though not by much) some of the expressions for the mass eigenstate fields. We label
the ‘dynamical’ fields with a δ prefix, which is meant to distinguish the field from the background
value; i.e the δχ± fields are perturbations about the ϕ0 = ϕe0 = µ background value. Follow-up
papers have dropped this cumbersome notation, but for the sake of bop-you-over-the-head clarity,
we’ll follow the original ISS conventions here.
Before working out some details about the spectrum, let’s stop to discuss what we expect.
Most fields should get tree-level masses ∼ |hµ|, since this is the only mass term in the superpo-
tential. We also expect to find some tree-level massless scalars which come in two flavors: (1) the
Goldstone bosons associated with the breaking in Eq. (20.9) and (2) the fluctuations about the
pseudomoduli space. The Goldstones are protected against quantum mass terms, but the pseu-
domoduli generically get mass terms from the Coleman-Weinberg potential. Alright? Allons-y!
Let’s get our hands a little dirty because it’s good for us. Let’s start by writing out the
superpotential in all its indexed glory. This way we can convince ourselves that the derivatives
we take to get the scalar potential actually work in the ‘intuitive’ way. (Then we can stare at it
a little and the slap our foreheads because it was obvious to begin with.)
ejc − hµ2 Φij δ ij .
W = hφci Φij φ (20.11)
We now take the appropriate derivatives,
∂W
= hΦij φjc
∂φci
∂W
= hφci Φij
e
∂φ jc
∂W ( )
ejc − µ2 δ ij .
= h φci φ
∂Φij
The scalar potential is
V = |Wφ |2 + |Wφe|2 + |WΦ |2 ,
where we mean

∑ ( ∂W ) ( ∂W )† ∂W
|Wϕ | =
2
= Tr .
ij
∂ϕij ∂ϕij ∂ϕ
The factor of |h|2 end up everywhere, so for simplicity we’ll just set h = 1. Given the form of the
superpotential, it’s easy to put them back at the end.
20.2.1 Tree-level massive fields

Let’s now work out the spectrum of massive fields at tree-level. We will explicitly derive the
simple case of the spectrum at the origin of the pseudomoduli space. We will see later from the
calculation of the Coleman-Weinerg potential that this is where the theory prefers to live. The
scalar potential can be written as
V = |Φφ| e − µ2 1|2 ,
e 2 + |φΦ|2 + |φφ (20.12)
98
where we should clarify what we mean, e.g. in the first term
Φφe = Φij φ
ejc = (Φφ)
e ic
[ ]i
|Φφ|
e = (Φφ) e † = Φij φ
e ic (Φφ) e†kc Φ†ki = Tr Φφ
ej c φ eφe† Φ† .
c
In terms of our matrix fields, this gives us

 ( ) 
δY µ + 2 (δχ+ − δχ− ) + δZ 2 (δρ+ − δρ− )
√1 T √1
Φφ e= ( ) . (20.13)
δ Ze µ + √12 (δχ+ − δχ− ) + δ Φ̂ √12 (δρ+ − δρ− )
We only care about the mass term inside |Φφ| e 2 , i.e. terms that are bilinear in the fields. Thus we
e which are linear in the fields, i.e. the µ term. It is easy to see that
want terms in Φφ

|Φφ| e 2.
e 2 mass = |µ δY |2 + |µ δ Z| (20.14)
Analogously,

|φΦ|2 mass = |µ δY |2 + |µ δZ|2 . (20.15)
This covers the first two terms in Eq. (20.12). Let’s sketch out the last term.
 ( ) 
1
(δχ + + δχ − )(δχ + − δχ − ) µ + √1
(δχ + + δχ − ) √1
(δρ + − δρ − )
e − µ2 1 =  ( ) .
2 2 2
φφ
√ (δρ+ + δρ− ) µ + √ (δχ+ − δχ− )
1 1 1
(δρ+ + δρ− )(δρ+ − δρ− ) − µ 2
2 2 2
(20.16)
Boy, that’s ugly looking. However, we know that we only care about the diagonal terms in the
trace, so let’s remind ourselves that
( )( † ) ( † )
A B A C† AA + BB †
= .
C D B † D† CC † + DD†
Thus
1 1
e − µ2 |2mass = |µ(δρ+ − δρ− )|2 + |µ(δρ+ + δρ− )|2
Tr|φφ
2 2
1 † 2 1
− (µ ) (δρ+ + δρ− )(δρ+ − δρ− ) − µ2 (δρ+ + δρ− )† (δρ+ − δρ− )† .
2 2
Ack! It still looks really ugly, especially since µ comes in as µ2 , (µ† )2 , and |µ|2 . However, upon
further inspection, this is easy to fix. We just have to absorb the µ into our δρ fields:
µ∗
δρ± → δρ′± = δρ± , (20.17)
|µ|
99
where the |µ|−1 is there to preserve canonical normalization. We can now drop the ′ to clean up
our notation. We end up with
1 1
e − µ2 |2mass = |(δρ+ − δρ− )|2 + |(δρ+ + δρ− )|2
Tr|φφ
2 2
1 1
− (δρ+ + δρ− )(δρ+ − δρ− ) − (δρ+ + δρ− )† (δρ+ − δρ− )† . (20.18)
2 2
Putting in some more elbow grease, we get
† †
e − µ2 |2mass = |δρ+ |2 −
2Tr|φφ + δρ− −
δρ − δρ+ + |δρ− |
δρ 2
+ |δρ+ |2 + † †

δρ − δρ+ + |δρ− |
δρ 2
+ δρ− +
(
(((
− δρ2+(
−δρ ((
− δρ
(( 2
+ + δρ+ δρ− + δρ−
((
† (( †
(
(((
− (δρ†+ )2(
+δρ
((
†
((
− δρ
† † 2
+ − δρ+ δρ− + (δρ− ) . (20.19)
This still requires some massage work. Let’s split the δρ± fields into its real and imaginary parts
(as matrices),
δρ± = a± + ib± . (20.20)
What does this buy us? recall that
(a + ib)(c + id) = ac + iad + ibc − bd

(a − ib)(c − id) = ac − iad − ibc − bd.
In terms of these fields we get
e − µ2 |2mass = 2(a2+ + b2+ ) + 2(a2− + b2− )

2Tr|φφ
− 2(a2+ − b2+ ) + 2(a2− − b2− ) (20.21)
= 4b2+ + 4a2−
= 4Im(δρ+ )2 + 4Re(δρ− )2 . (20.22)
Good. Thus we’ve found that the fields δY , δ Z,e δZ, Im(δρ+ ), and Re(δρ− ) have all acquired
tree level masses on the order of |hµ|. Still with us? Good, because that was the easy part.
Let us remark that later on we will calculate the Coleman-Weinberg effective potential to
determine how the pseudomoduli are lifted. We will find that the vacuum of the theory lives at
the origin of pseudomoduli space so that the tree-level spectrum above turns out to be accurate.
Note, however, that this is not the spectrum that we plug into the Coleman-Weinberg formula.
In order to calculate the effective potential for the pseudomoduli, we will have to determine the
spectrum about an arbitrary point on the pseudomoduli space (we’ll see that it is sufficient to
restrict to a submanifold). In this case, the spectrum will become a function of the pseudomoduli
and, in particular, one will obtain mass terms which mix the above fields.
Now let’s work out the linear combinations that appear as pseudomoduli and Goldstones.
100
20.2.2 Pseudomoduli fields
The pseudomoduli are actually trivial. These are the directions that are associated Eq. (20.7), up
to rotations by our global symmetries. Thus pseudomoduli are precisely
δ Φ̂ and δ χ̂ ≡ δχ− + h.c., (20.23)
where we’ve also rescaled the χ fields to absorb µ:

µ∗
δχ± → δχ′± = δχ± , (20.24)
|µ|
and then we again drop the ′ for simplicity. Note that it is important that δ χ̂ take the precise form
above. The excitation has to be the δχ− part of φ and φ e because this is the antisymmetric part:
this is the part that will cancel in the last term of Eq. (20.12). Note that in the vacuum of the
theory these excitations manifestly do not contribute to the first two terms. This cancellation only
occurs for the real part of this field, which we isolate by summing with the Hermitian conjugate.
20.2.3 Goldstone bosons

Finally, let’s identify the Goldstone bosons coming from the spontaneous breaking of global sym-
metries. Let’s write down our symmetry table once again:
SU(N ) SU(F )L SU(F )R U(1)B U(1)′ U(1)R

Φ 1 □ □ 0 -2 2
φ □ □ 1 1 1 0
e
φ □ 1 □ -1 1 0
Recall that there is an explicit breaking SU(F )L × SU(F )R × U(1)′ → SU(F ) by the ∆W term in
Eq. (20.3). This means, in particular, that Φ transforms as an Ad ⊕ 1, i.e. an adjoint plus the
trace. Because one typically doesn’t work with spontaneous symmetry breaking of with multiple
fields getting related vevs, let’s work through this section somewhat carefully.
Let’s review how fields transform under the fundamental and anti-fundamental representations
of a Lie group44 .
φi → φ′i = φi + iϵa (T a )ij φj

φi → φ′i = φi + iϵa (−T ∗a )ij φj ,
44
This should be a very basic review, but it recently came to my attention that many students are unaware of
handy references for the representation theory of Lie groups for particle physicists. I am not particularly enthused by
the canonical text by Georgi [115]. Instead, for a quick introduction Cheng & Li do a good job [116] while the lecture
notes for the Cambridge Part III course ‘Symmetries and Particle Physics’ are usually very good. Two recent sets of
lecture notes can be found at http://www.mth.kcl.ac.uk/~jbg34/Site/Dr._Jan_Bernard_Gutowski.html and
http://www.damtp.cam.ac.uk/user/ho/GNotes.pdf.
101
where φi transforms as a fundamental □ and φi transforms as an anti-fundamental □. We may
use anti-Hermiticity to relate the generators fundamental and anti-fundamental representations
T□a = −(T□a )∗ = −(T□a )T ,
from which we obtain
φ′i = φi − iϵa (T a )j i φj = φi − iϵa φj (T a )j i .
The U(1)s are all generated by identity matrices, 1 with respect to the matrix Lie groups. This
just means that they are the traces of the multi-dimensional matrices that generate our global
symmetry. For now let’s not worry about them because they’re easy. The generators of our
[broken] SU(N )× SU(F )L ×SU(F )R symmetry are
T A = TSU(N ) ⊗ TSU(F )L ⊗ TSU(F )R .

A A A
(20.25)
Let’s see how this acts on a bifundamental F × F field like Φij .
iϵA T A Φ = iϵaL (TLa )iL kL ΦkL jR + iϵbR (TRb )jR kR ΦiL kR (20.26)
= iϵaL (T a )iL kL ΦkL jR − iϵbR (T b )kR jR ΦiL kR , (20.27)
where for clarity we’ve labelled the SU(F )L and SU(F )R indices separately and used the above
observation that since Φ is a fundamental under SU(F )L and an anti-fundamental under an iden-
tical SU(F )R , we can write everything with respect to the fundamental generators of SU(F ). Now
the main point is that the explicit breaking SU(F )L × SU(F )R → SU(F ) enforces
ϵL = ϵR . (20.28)
This is just the analog of chiral symmetry breaking in QCD (only this is done explicitly).
Now let’s get to the good stuff. We know that the Goldstone bosons are constructed by acting
on the vev by the broken generators since this determines the flat directions in field space. The
somewhat novel feature here relative to what is found in introductory field theory texts is that two
e obtain vevs. The procedure is the same, but one must remember to act on both
fields (φ and φ)
vevs simultaneously with each broken generator. The Goldstone directions in field space will then
be a linear combination of both fields. This is obvious in retrospect, though a helpful mnemonic
might be to imagine a single multi-component field φ ⊕ φ e which is transformed by generators
A e
T ⊕ T and which obtains a vev ⟨φ⟩ ⊕ ⟨φ⟩. In this case it is clear that the correct procedure is
A
to act on both φ and φ e simultaneously by the broken generators.

We already know what the vevs are from Eq. (20.8). We’ll consider the spontaneous breaking in
two steps. First we consider the breaking SU(F ) → SU(N )F ×SU(F −N )×U(1)B ′ . The first factor
is the upper-left N × N part of the SU(F ) generators, the second factor is the lower-right (F −
N ) × (F − N ) part, and the U(1)B ′ corresponds to the diagonal generator diag(a, · · · , a, b, · · · , b).
Note that we are not yet considering that the vevs in Eq. (20.8) take the form µ1 and so break
SU (N ) × SU (N )F → SU (N )D . All we’re considering for now is that we’ve broken SU (F ) into
two disconnected blocks.
102
Let’s remind ourselves what our set of generators look like:
         
1 i a
 T        
 ... 
,
0 1 0 -i 0 t b
where clearly generators of the second and third type are broken since they mix the N × N block
which obtains a vev µ1N with the lower N × (F − N ) block in φ and φ eT . This makes it clear
that the Goldstone directions are precisely these lower N × (F − N ) blocks. We emphasize, once
again, that the directions obtained by doing this for each field are not independent. Fortunately,
we’ve already intelligently split up our fields in Eq. (20.10).
Acting with the real broken generators, we see that our Goldstone directions are (writing
T ⟨φ⟩ + T A ⟨φ⟩)
A
e
1 1
√ Re(ρ+ + ρ− ) + √ Re(ρ+ − ρ− ) ∝ Reρ+ . (20.29)
2 2
Similarly, the imaginary broken generators give us the Goldstones
1 1
√ Im(ρ+ + ρ− ) − √ Im(ρ+ − ρ− ) ∝ Imρ− , (20.30)
2 2
where the minus sign comes from the fundamental versus the anti-fundamental representation.
Now let’s move on. The particular form of the vevs in Eq. (20.8) break SU(N )×SU(F )N ×U(1)′B
to SU(N )D . Note that this is a different spontaneous breaking from the SU(F ) → SU(N )F ×SU(F −
N )×U(1)B ′ we considered above: that had to do with breaking SU(F ) into blocks. Now we’re
dealing with the actual form of the vev in the nontrivial block.
Let us write the upper N × N blocks of φ and φ e as φN and φeN . Then recalling how φ and φe
transform under SU(N ) and SU(F )N , we see
φN → UN φN UF†
eN UN† .
eN → UF φ
φ
This is preserved by the vevs if the transformation parameters are such that ϵN = ϵF , i.e. we
break to the diagonal subgroup. This breaking is precisely analogous to the pattern of chiral
symmetry breaking in QCD. The broken generators then have ϵ = ϵN = −ϵF , i.e. they are the
axial generators. Let’s work out the change in the φ field after an axial transformation:
iϵA T A ⟨φN ⟩ = iϵA (TNA )iN kN ⟨φN ⟩kN jF − iϵA
F ⟨φN ⟩ kF (TF ) jF .
iN A kF
(20.31)
Recalling that ⟨φN ⟩iN jF = µδ iN jF , we have
iϵA T A φN = 2iϵA (T A )iN jF . (20.32)
This is a basis of traceless anti-Hermitian matrices. The analysis for φe gives the same result
(there’s an overall minus sign). There’s one missing piece: the U(1)B ′ generator which is also
broken. This gives a trace part to the Goldstone fields. Thus our Goldstones are the trace-
included anti-Hermitian matrices,
χ− − χ†− . (20.33)
That wraps up our summary of the spectrum of fields.
103
20.2.4 The Coleman-Weinberg potential
At one-loop order the pseudomoduli are lifted by the Coleman-Weinberg potential. One must
check that the potential has positive curvature rather than negative curvature, or else our stable
vacuum will be spoiled. Using the global symmetries (e.g. the unbroken U(1)), the fact that only
single traces appear in the Coleman-Weinberg potential, and some dimensional analysis for the
overall factor, the relevant piece of the effective potential is
( )
1 †
VCW = |h µ |
4 2
aTr δ χ̂ + bTr δ Φ̂ δ Φ̂ + · · · ,
2
(20.34)
2
for some coefficients a and b which we’d like to establish are greater than zero. Because h is
marginally irrelevant in the IR, this one-loop contribution to the effective potential dominates
over higher-order corrections.
[Check Why is h marginally irrelevant?] Note that the way we’ve defined h is precisely
analogous to the use of ℏ to count loops in Appendix B.3. If we take h → 0 with f, X, q ∼ h−1
then the classical Lagrangian goes as h−2 , the one-loop corrections go as h0 , and higher loop
contributions go as h2n for n > 0.
Now recall that it is not correct to simply plug in the tree-level spectrum that we’ve derived
above. These masses are all dependent on the point on the pseudomoduli space in which we live.
With some foresight, we calculated that spectrum at the origin of the pseudomoduli space. In order
to determine the effective potential of the pseudomoduli, however, it is necessary to determine the
spectrum for an arbitrary point on the pseudomoduli space so that the potential can be written
as a function of the pseudomoduli. Thus, generally calculating the effective potential requires
some work since there are so many pseudomoduli (counting each component of the matrix fields).
Fortunately, we can simplify the analysis significantly since we don’t care about the full effective
potential: we only need the quadratic part which tells us about the local stability of a point. Thus
we can be clever and only choose to move along specific directions along the pseudomoduli space.
We will pick directions labelled by X0 and θ:
( ) ( θ ) ( −θ )
δY δZ T µe 1N + δχ µe 1N + δ χ e
Φ= φ= e =
φ T
,
δ Ze X0 1(F −N ) + δ Φ̂ δρ δe
ρ
(20.35)
where we can assume X0 and θ are small. (If we determine a and b anywhere on the pseudomoduli
space then we’ve determined it everywhere.) Plugging this into the formula for VCW (see Appendix
B), we get
( )
1 ∗ 2
VCW = const + h µ4 2
aN µ (θ + θ ) + b(F − N )|X0 | + · · · .
2 2
(20.36)
2
Our task is to determine a and b. We need to find the tree-level masses associated with a point
(X0 , θ) on the pseudomoduli submanifold, so we plug in our parameterization into the superpo-
tential,
[( ) ( ) ( )
W = hTr µeθ δχ δY µe−θ + δ χ e + δρδ Ze µe−θ + δ χ
e
( ) ( ) ] [ ]
e ρ + δρ X0 + δ Φ̂ δe
+ µeθ + δχ δ Zδe ρ − hµ2 Tr δY + X0 1(F −N ) + δ Φ̂ (20.37)
104
Keep in mind what’s going on: fields prefixed with a δ are dynamical excitations, while fields with-
out a δ are background fields (i.e. pseudomoduli). This is why the δ notation, while cumbersome,
is handy.
We can now invoke a bit of a trick: we know that the contributions to the Coleman-Weinberg
effective potential come from SUSY-breaking, since the manifestly supersymmetric parts cancel
in the supertrace. We know how supersymmetry is broken in this model, so we can be clever
and identify which fields have masses which actually couple to the SUSY-breaking F -terms. So,
pop-quiz: which fields obtain F -term vevs?
We know that the fields φ and φ eT obtain vevs along their upper N × N blocks. These vevs
come from using the available global symmetries to cancel as much of the the µ2 term in FΦ , c.f.
Eq. (20.5). The remaining N × (F − N ) matrix of fields are those which could not cancel the
remaining terms in the µ2 diagonal matrix and hence it is these fields which break supersymmetry.
These are just the δρ and δeρ fields. It is easy to see in the superpotential that when the lower-right
(F − N ) × (F − N ) block ⟨FΦ ⟩ is nonzero, the δρ and δe ρ fields obtain obtain SUSY-breaking scalar
masses.
The SUSY-breaking δρ and δe ρ scalars mix with other fields at tree level45 . Thus the fields
which make a nontrivial contribution to the Coleman-Weinberg potential are those which mix
with the δρ or δeρ fields. Writing out the quadratic part of the superpotential, we get
[ ( ) ( ) ( ) ]
W = hTr µeθ δZ T δe e + µe−θ δ ZeT δρ + δY T δχ + δe
ρ + δY δ χ ρδρT − µ2 (X0 + δ Φ̂) + · · · .
For reasons that will become clear shortly, we’ve written out (X0 + δ Φ̂) even though this term
includes non-quadratic terms. At this point the ISS paper (see their Appendix B) makes a cryptic
remark that the off-diagonal components of δ Φ̂ do not contribute to the mass matrix. This is not
quite an accurate or relevant observation since we have chosen a submanifold of the pseudomoduli
space where we are only expanding about the diagonal background (parameterized by X0 ) of the
δ Φ̂ field. The δ Φ̂ itself is not a background value but a physical excitation. At any rate, this is
neither here nor there so we may move on. We can make the more important observation that the
e, and δY do not mix with the SUSY-breaking fields at tree-level. They certainly have
fields δχ, δ χ
higher-power couplings with δρ and δe ρ, but those could only contribute mixing at the one-loop
level. Thus, these superfields have manifestly supersymmetric spectra and do not contribute at
all to the one-loop effective potential. The remaining terms which are of interest may be written
(F −N )
∑ ( ) ( ) ( ) ( )
Wmass = h (X0 + δ Φ̂ii ) δρδe ρδZ T ii + µe−θ δρδ ZeT − µ2 X0 + δ Φ̂ii .
ρT ii + µeθ δe
ii
i=1
Now we make a very handy observation that also justifies our choice of writing out (X0 + δ Φ̂)
explicitly. This looks precisely like (F − N ) decoupled copies of an O’Raifeartaigh-like model with
superpotential
( )
W = h Xϕ1 · ϕ2 + µe−θ ϕ1 · ϕ3 + µeθ ϕ2 · ϕ4 − µ2 X .
45
Again, this is true for some generic point on the pseudomoduli space, but we calculated above that for the case
of the origin of the pseudomoduli the imaginary and real parts of the δρ and δ ρe fields are independent physical
degrees of freedom.
105
The Coleman-Weinberg potential for this model can be worked out straightforwardly for home-
work. Contrary to my usual practice I won’t work it out here46 , but helpful points for the derivation
of the effective potential for simple O’Raifeartaigh model can be found in Intriligator and Seiberg’s
SUSY-breaking notes [1]. We may invoke these results to determine that the Coleman-Weinberg
potential for this pseudomoduli submanifold is
( )
(1) h4 µ2 (log 4 − 1)N (F − N ) 1 2 ∗
VCW = constant + µ (θ + θ) + |X| + · · · ,
2
(20.38)
8π 2 2
from which we determine the coefficients

log 4 − 1 log 4 − 1
a= (F − N ) b= N. (20.39)
8π 2 8π
We don’t actually care what these precise values are, only that they are positive definite in the
regime of interest and therefore our pseudomoduli are indeed stabilized about the origin as we
assumed above.
20.3 Macroscopic Model II

20.4 Dynamical Realization: the ISS model
20.5 Immediate directions beyond the basic ISS model
20.6 Extensions and related work
Acknowledgements
I am especially grateful to my the teachers, collaborators, mentors, and friends who were patient
in explaining many of these topics to me. Foremost are Csaba Csáki, Patrick Meade, David Shih,
Zohar Komargodski, Yael Shadmi, John Terning, Yuhsin Tsai, and David Curtin. This work is
supported in part by the NSF grant number PHY-0355005, an NSF graduate research fellowship,
and a Paul & Daisy Soros Fellowship For New Americans. Part of this work was completed at the
Kavli Institute for Theoretical Physics (as part of the ‘First Year of the LHC’ program) which is
supported in part by the National Science Foundation under Grant No. NSF PHY05-51164. Part
of this work was completed at CERN as part of the SUSY Breaking 2011 workshop. The contents
of this article do not necessarily represent the views of any of the above institutions.
46
I did a simpler case on scratch paper but decided that the calculation was so non-illuminating that it wasn’t
worth typing up.
106
Appendix
A Notation and Conventions
Here we present a set of self-consistent notation and conventions that we (try to) use in this
document.
A.1 Field labels

Chiral superfields are typically written with capital Roman letters, e.g., S, N , X. Complex
conjugation is denoted by a star, (a + ib)∗ = (a − ib). A bar, on the other hand, is used to
distinguish pairs of vector-like chiral superfields, e.g, N and N̄ have opposite charges under a
particular symmetry. Do not confuse this bar with complex conjugate. To avoid confusion, it is
typical to use a tilde to denote the vector-like pair, e.g., N and Ne . We denote the axino by χ
rather than the usual e
a to avoid cumbersome notation and to reinforce its identity as dark matter.
The dual gauge field strength ∗F is defined in component notation relative to the field strength
via
1
Feµν = F αβ ϵαβµν . (A.1)
2
A.2 Spacetime and spinors

There is no completely standard set of spacetime and spinor conventions in the SUSY literature,
but the choices that make the most sense to us are those by Dreiner et al. [117]; see their appendix
for a thorough discussion of how to passing between metric conventions47 . See also Problem 1
of Appendix C in Binetruy’s supersymmetry textbook [118] which identifies all possible sources
of sign ambiguities and writes relevant formulae with all choices made explicit. Pedagogical
introductions to Weyl and Majorana spinors can be found in Aitchison [119] and the article by
Pal [120].
4D Minkowski indices are written with lower-case Greek letters from the middle of the alphabet,
µ, ν, · · · . We use the particle physics (‘West Coast,’ mostly-minus) metric for Minkowski space,
(+, −, −, −). Our convention for σ 0 and the three Pauli matrices ⃗σ is
( ) ( ) ( ) ( )
0 1 0 1 0 1 2 0 −i 3 1 0
σ = σ = σ = σ = . (A.2)
0 1 1 0 i 0 0 −1
The un-barred Pauli matrices have indices σαµβ̇ while the barred Pauli matrices, σ̄ µ = (σ 0 , −⃗σ ),
have indices σ̄ µα̇β . The two types of Pauli matrices are related by
σ̄ µα̇α = ϵα̇β̇ ϵαβ σ µβ β̇ , (A.3)

47
To see this in action, see their source file at http://zippy.physics.niu.edu/spinors.html, which includes a
macro to allow one to change metric conventions. The implementation is an excellent example of where the metric
choice is (and isn’t) relevant.
107
where our convention for the sign of ϵ is given below. The Weyl representation for the Dirac γ
matrices is
( ) ( )
µ σµ 5 0 1 2 3 −1
γ = γ = iγ γ γ γ = . (A.4)
σ̄ µ 1
Note that the definition of γ 5 is the usual 4D Weyl basis convention, whereas the sensible 5D
convention is Γ5 = diag(i, −i) so that the 5D Clifford algebra is satisfied. The antisymmetric
products of Pauli matrices are
i i
σ µν = σ [µ σ̄ ν] σ̄ µν = σ̄ [µ σ ν] . (A.5)
4 4
I don’t like the factor of i, but this is the price of sticking with the conventions in [117].
The totally antisymmetric tensor [densities] are chosen to have
ϵ12 = ϵ21 = 1 ϵ0123 = −ϵ0123 = 1. (A.6)
This convention agrees with Wess & Bagger [121], Terning [5], and Dreiner et al. [117] but has a
relative sign from Bailin and Love [122]. The significance of this choice is described in footnotes
4–6 of Dreiner et al. [117], but the point is that Weyl spinor indices are raised and lowered via
matrix multiplication from the left,
ψα = ϵαβ ψ β ψ α = ϵαβ ψβ ψ̄α̇ = ϵα̇β̇ ψ̄ β̇ ψ̄ α̇ = ϵα̇β̇ ψ̄β̇ , (A.7)
where we’ve introduced the notation ψ̄α̇ = (ψα )∗ and χα = (χ̄α̇ )∗ . Note the use of ∗ here rather
than †, though the distinction is mostly poetic. If one is perturbed by this, an excellent reference
is the relevant chapter in Aitchison’s elementary text [119]. The relative sign between ϵ12 and
ϵ12 sets ϵαρ ϵρβ = δαβ so that no signs appear when an index is raised and then lowered again.
Alternately, this relative sign appears when relating the ϵ tensor to charge conjugation as we
will see below. With this convention, special care is required to keep track of minus signs when
raising and lowering indices of ϵ tensors (see [117]), but this is usually a silly thing to do to begin
with. Using Lorentz invariance, one can write relations like θα θβ ∝ ϵαβ θθ. The overall constant
of proportionality can be found by contracting the indices of both sides. One finds
1 1
θα θβ = − ϵαβ θθ θα θβ = + ϵαβ θθ (A.8)
2 2
1 1
θ̄α̇ θ̄β̇ = + ϵα̇β̇ θ̄θ̄ θ̄α̇ θ̄β̇ = − ϵα̇β̇ θ̄θ̄. (A.9)
2 2
Similarly,
1
θσ µ θ̄ θσ ν θ̄ = + θ2 θ̄2 η µν (A.10)
2
1
(θψ)(θχ) = − (ψχ)(θθ) (A.11)
2
1
(θ̄ψ̄)(θ̄χ̄) = − (ψ̄ χ̄)(θ̄θ̄). (A.12)
2
108
The placement of Weyl spinors (with their natural index placement) within a Dirac spinor is
( )
ψα
ΨD = . (A.13)
χ̄α̇
Spinor contractions are descending for undotted indices and ascending for dotted indices:
ψχ ≡ ψ α χα ψ̄ χ̄ ≡ ψ̄α̇ χ̄α̇ . (A.14)
With this convention, contractions are independent of the order of the spinors: ψχ = χψ and
similarly for the barred spinors ψ̄ χ̄ = χ̄ψ̄. The Dirac conjugate spinor is given by
( ) ( )
† 0
( †α † ) σ 0αβ̇ ( †α † ) 1αβ̇ ( α )
Ψ̄D = Ψ γ = ψ χ̄α̇ = ψ χ̄ ≡ χ ψ̄ . (A.15)
σ̄ 0α̇β α̇ 1 α̇β β̇
One may take this as a definition of χ and ψ̄ in terms of ψ and χ̄ in ΨD . It shows how γ 0 is used
to convert the dotted index of χ̄† into the undotted index of χ (and vice versa for ψ † and ψ̄).
The charge conjugate of a Dirac fermion Ψc is given by
( 2 ) ( )
c T iσ̄ ϵαβ
Ψ = C Ψ̄ C= = , (A.16)
iσ 2 ϵα̇β̇
This comes from taking the Hermitian conjugate of the Dirac equation
←−
i(∂/ − ieA)Ψ
/ = 0 ⇒ −iΨ̄γ 0 γ µ† ( ∂ µ + ieAµ ) = 0 ⇒ −iγ µT (∂µ + ieA)Ψ̄T = 0, (A.17)
where we’ve made use of the identities γ 0 γ µ† γ 0 = γ µ and (γ 0 ) = 1. Because −γ µT satisfies the
2
4D Clifford algebra, there exists a charge conjugation matrix C such that C −1 γ µ C = −γ µT . In

particular, C Ψ̄T is a solution to the Dirac equation with opposite charge,
iγ µ (∂µ + ieAµ )C Ψ̄T = 0. (A.18)
The above property of C implies that C ∼ γ 0 γ 2 . The constant of proportionality must be a pure
phase so that (Ψc )c = Ψ. We choose this proportionality so that
C = iγ 0 γ 2 , (A.19)
which matches (A.16). This can be understood as the reason why the ϵ tensor density appears
with a different overall sign when written with upper versus lower indices; the sign comes from σ 2
versus σ̄ 2 . Writing out indices slightly more carefully,
( 2 ) ( α) ( ) ( α) ( )
c T iσ̄ χ ϵαβ χ χα
Ψ = C Ψ̄ = 2 = = . (A.20)
iσ ψ̄α̇ ϵ α̇ β̇ ψ̄α̇ ψ̄ α̇
A Majorana fermion obeys ΨM = ΨcM so that
( ) ( )
ψα χα
= , (A.21)
χ̄α̇ ψ̄ α̇
that is ψα = χα and χ̄α̇ = ψ̄ α̇ . In other words,
( )
ψα
ΨM = . (A.22)
ψ̄ α̇
Sometimes the right-hand side is written somewhat impressionistically as (ψ, iσ̄ 2 ψ ∗ )T ; the intended
meaning is identical to the above expression.
109
A.3 Superfields and superspace
The superspace measure is
1 1 1 1
d2 θ = − dθα dθβ ϵαβ = − dθα dθα d2 θ̄ = − dθ̄α̇ dθβ̇ ϵα̇β̇ = − dθ̄α̇ dθ̄α̇ . (A.23)
4 4 4 4
A [left] chiral superfield is given by
√
Φ(x) = ϕ(y) + 2θψ(y) + (θθ)F (y), (A.24)
where the shifted coordinate is y µ = x − iθσ µ θ̄. The minus sign here is important for, among
other things, obtaining the correct sign on the fermion kinetic term. Expanding in terms of fields
evaluated at x, we have
1 √ √
Φ(x) = ϕ − i(θσ µ θ̄)∂µ ϕ − (θσ µ θ̄)(θσ ν θ̄)∂µ ∂ν ϕ + 2θψ − i 2θ(θσ µ θ̄)∂µ ψ + (θθ)F. (A.25)
2
Using the relations (A.10) and (A.11) we may simplify this to
√ i 1
Φ(x) = ϕ + 2θψ + (θθ)F − i(θσ µ θ̄)∂µ ϕ + √ (θθ)(∂µ ψσ µ θ̄) − (θθ)(θ̄θ̄)∂ 2 ϕ (A.26)
2 4
√ i 1
Φ† (x) = ϕ∗ + 2θ̄ψ̄ + (θ̄θ̄)F ∗ + i(θσ µ θ̄)∂µ ϕ∗ − √ (θ̄θ̄)(θσ µ ∂µ ψ̄) − (θθ)(θ̄θ̄)∂ 2 ϕ∗ . (A.27)
2 4
The field strength superfield is
W = −iλ + [D − σ µν Fµν ] θ − θθσ∂ λ̄, (A.28)

∫ 2 1
so that the SYM Lagrangian is L = d θ 4 WW + h.c.; occasionally I may write W instead of W.
I’ve chosen the definition σ µν = 4i σ [µ σ̄ ν] , c.f. (A.1).
A.4 SUSY NLΣM

The SUSY nonlinear sigma model (NLΣM) provides a nice application of these conventions. We
provide a somewhat detailed derivation as a sanity check, but the reader may skip to the final result
below. We expand Φ and Φ† about the Grassmann directions so that the expansion parameters
are
√ i 1
∆ = 2θψ + θθF + iθσ µ θ̄ ∂µ ϕ − √ (θθ)(∂µ ψ)σ µ θ̄ − (θθ)(θ̄θ̄)∂ 2 ϕ (A.29)
2 4
√
∆¯ = 2θ̄ψ̄ + θ̄θ̄F ∗ − iθσ µ θ̄ ∂µ ϕ∗ + √i (θ̄θ̄)θσ µ (∂µ ψ̄) − 1 (θθ)(θ̄θ̄)∂ 2 ϕ∗ . (A.30)
2 4
We can thus write the Kähler potential as
K →K + Ka ∆a + Kā ∆ ¯ ā + 1 Kab ∆a ∆b + 1 Kāb̄ ∆ ¯ b̄ + Kab̄ ∆a ∆

¯ ā ∆ ¯ b̄ (A.31)
2 2
1 ¯ ā ∆b ∆c + 1 Kab̄c̄ ∆a ∆ ¯ c̄ + 1 Kabc̄d¯∆a ∆b ∆
¯ b̄ ∆ ¯ d¯.
¯ c̄ ∆
+ Kābc ∆ (A.32)
2 2 4
110
where we’ve written barred indices to denote derivatives with respect to conjugate fields. The K
term on the right-hand side carries no Grassmann directions and can so can be dropped under
the d4 θ. We may now write out the relevant products of ∆ and ∆s. ¯
√
∆∆ =2(θψ)(θψ) − 2 2i(θψ)(θσ µ θ̄)∂µ ϕ − (θσ µ θ̄)(θσ ν θ̄)(∂µ ϕ)(∂ν ϕ)
∆∆ ¯ =i(θ̄ψ̄)(θθ)(∂µ ψσ µ θ̄) − i(θψ)(θ̄θ̄)(θσ µ ∂µ ψ̄) + (θσ µ θ̄)(θσ ν θ̄)(∂µ ϕ)(∂ν ϕ∗ ) + (θθ)(θ̄θ̄)|F |2 + · · ·
√
∆∆∆ ¯ =2 2(θψ)(θψ)(θ̄ψ̄) − 4i(θψ)(θ̄ψ̄)(θσ µ θ̄) ∂µ ϕ + 2(θψ)(θψ)(θ̄θ̄)F ∗ + · · ·
¯ 2 =4(θψ)(θψ)(θ̄ψ̄)(θ̄ψ̄).
∆2 ∆
The combinations with more ∆s ¯ than ∆s are given by the replacement θ ↔ θ̄, ψ → ψ ∗ , and
F → F ∗ . We’ve dropped indices for simplicity, i.e. a term like θψ a ϕb + θψ b ϕa is written as 2θψϕ.
Since all of the indices are summed over in the expansion for K, this is a reasonable simplification.
(We’ll restore indices as necessary below.) To simplify we use some of the handy expressions in
the Appendix.
1
∆∆|θ4 = − (∂ϕ)2 (A.33)
2

∆∆¯ 4 = ψσ µ ∂µ ψ̄ − i ∂µ ψσ µ ψ̄ + 1 |∂ϕ|2 + |F |2
i
(A.34)
θ 2 2 2

∆∆∆¯ 4 = −i(ψσ µ ψ̄)(∂µ ϕ) + (ψψ)F ∗ (A.35)
θ
¯∆
∆∆∆ ¯ 4 = (ψψ)(ψ̄ ψ̄). (A.36)
θ
Plugging this in to the expression for K (and ignoring the constant with no support over d4 θ),
1 1 1 1
K d4 θ = − Ka ∂ 2 ϕa − Kā ∂ 2 ϕ∗a − Kab (∂ϕa )(∂ϕb ) − Kāb̄ (∂ϕ∗ā )(∂ϕ∗b̄ )
4 [ 4 4 4 ]
i a µ i 1 µ ∗b̄ a ∗b̄
+ Kab̄ ψ σ ∂µ ψ̄ − ∂µ ψ σ ψ̄ + (∂µ ϕ )(∂ ϕ ) + F F
ā a µ ā a
2 2 2
1 [ ] 1 [ ]
∗ā ∗c̄
+ Kābc −i(ψ σ ψ̄ )(∂µ ϕ ) + (ψ ψ )F
b µ ā c b c a µ b̄ b̄ c̄
+ Kab̄c̄ i(ψ σ ψ̄ )(∂µ ϕ ) + (ψ̄ ψ̄ )F a
2 2
1 ¯
+ Kabc̄d¯(ψ a ψ b )(ψ̄ c̄ ψ̄ d ). (A.37)
4
We now invoke a bit of a trick. (There are other, equivalent, ways to do this, for example
by taking supercovariant derivatives or by writing the Lagrangian in terms of Kähler geometric
quantities.) Consider the total [spacetime] derivative term,
∂ 2 K =∂ (Ka ∂ϕa + Kā ∂ϕ∗ā ) (A.38)

= Kab ∂ϕa ∂ϕb + Ka ∂ 2 ϕa + Kab̄ ∂ϕa ∂ϕ∗b̄
+ Kāb̄ ∂ϕ∗ā ∂ϕ∗b̄ + Kā ∂ 2 ϕ∗ā + Kab̄ ∂ϕa ∂ϕ∗b̄ . (A.39)
We can thus add to the Kahler potential a total derivative ∂ 2 K/4. This cancels the first line and
111
adds a factor of 2 to the usual complex scalar kinetic term in the second line so that
[ ]
i a µ i µ ∗b̄ a ∗b̄
K d θ = Kab̄ ψ σ ∂µ ψ̄ − ∂µ ψ σ ψ̄ + (∂µ ϕ )(∂ ϕ ) + F F
4 ā a µ ā a
2 2
1 [ ]
+ Kābc −i(ψ b σ µ ψ̄ ā )(∂µ ϕc ) + (ψ b ψ c )F ∗ā
2
1 [ ]
a µ b̄ ∗c̄ b̄ c̄ a
+ Kab̄c̄ i(ψ σ ψ̄ )(∂µ ϕ ) + (ψ̄ ψ̄ )F
2
1 ¯
+ Kabc̄d¯(ψ a ψ b )(ψ̄ c̄ ψ̄ d ). (A.40)
4
We can solve the equation of motion for the auxiliary fields,
δL 1
∗ā
= Kbā F b + Kābc ψ b ψ c = 0. (A.41)
δF 2
We thus find
1 1
F a = − K aā Kābc ψ b ψ c ≡ − Γabc ψ b ψ c (A.42)
2 2
1 1
F ∗ā = − K aā Kab̄c̄ ψ̄ b̄ ψ̄ c̄ ≡ − Γāb̄c̄ ψ̄ b̄ ψ̄ c̄ , (A.43)
2 2
where we use upper indices to denote the inverse Kähler metric and have defined the Christoffel
symbols. Plugging this back into the Kähler potential,
[ ]
i a µ i µ ∗b̄
K d θ = Kab̄ ψ σ ∂µ ψ̄ − ∂µ ψ σ ψ̄ + (∂µ ϕ )(∂ ϕ )
4 ā a µ ā a
2 2
1 [ ] 1 [ ]
+ Kābc −i(ψ b σ µ ψ̄ ā )(∂µ ϕc ) + Kab̄c̄ i(ψ a σ µ ψ̄ b̄ )(∂µ ϕ∗c̄ )
2 2
1 ¯
+ Rab̄cd¯(ψ a ψ c )(ψ̄ b̄ ψ̄ d ), (A.44)
4
where we’ve written the Riemann tensor
¯
Rab̄cd¯ = Kab̄cd¯ − Kdac
¯ Γ b̄d¯ = Kab̄cd¯ − Kdb̄d¯Γ ac .
d d
(A.45)
We can further simplify by defining the Kähler covariant derivatives
( )
Dµ ψ a = ∂µ δca + Γabc ∂µ ϕb ψ c (A.46)
( )
Dµ ψ̄ ā = ∂µ δc̄ā + Γāb̄c̄ ∂µ ϕ∗b̄ ψ̄ c̄ . (A.47)
This allows us to group the K ( 3) terms with the K ( 2) terms to obtain,

[ ] 1
a µ←→ ā µ ∗b̄ ¯
4 a
K d θ = Kab̄ iψ σ D µ ψ̄ + (∂µ ϕ )(∂ ϕ ) + Rab̄cd¯(ψ a ψ c )(ψ̄ b̄ ψ̄ d ), (A.48)
4
where we use the notation
−→ ←−
←
→ Dµ − Dµ
Dµ = . (A.49)
2
112
A.5 2-component plane waves
See [117] for details.
x x†
y† y
A.6 OLD Notation and Conventions

[Flip: please check these for consistency and include above.]
4D Minkowski indices are written with lower-case Greek letters from the middle of the alphabet,
µ, ν, · · · . We use the particle physics (‘West Coast,’ mostly-minus) metric for Minkowski space,
ds2 = (+, −, −, −). Typically (but not always) we will write superfields using capital Greek or
Roman characters, e.g. Φ or Z. If we stray from this notation we will use the lowest component of
the superfield to also refer to the entire superfield, e.g. φ. The components of a chiral superfield
will be written as Φ = ϕ + θψ + θθF . It is a terrible practice, but we will follow the standard
convention that we will also refer to the vacuum expectation value of a field by the same symbol
when there is no ambiguity, i.e.
⟨Φ⟩ = Φ,
when it is clear from context that the object being considered is the vacuum value, not the dynam-
ical field itself. The Kähler potential and superpotential are denoted by K and W respectively.
We will frequently take derivatives of these potentials in field space. For simplicity of notation we
will frequently write these field derivatives (technically variations of functionals) compactly as
δ
∂i W (Φi ) ≡ W (Φi ).
δΦi
We will further truncate this by writing
Wi ≡ ∂i W (Φi )|Φ=⟨Φ⟩ , (A.50)
this is the standard notation in the literature.

We will make use of a notation originally inspired by Fernando Quevedo and write the field
strength superfield as a ‘sophisticated W,’ Wa , in order to avoid confusion with the superpotential
W or its derivatives Wa = ∂a W . For whatever reason there are many silly normalizations for SUSY
113
gauge theories. We shall employ what we feel is the least silliest (and what is used by [5]),
∫
1
LSYM = d2 θ τ Wαa Waα + h.c. (A.51)
16πi
1 ΘYM e i 1
= − 2F2 − 2
F F + 2 λ † σ µ Dµ λ + 2 D 2 (A.52)
4g 32π g 2g
Wαa = −iλaα + θα Da (y) − (σ µν θ)α Fµνa
(y) − (θθ)σ µ Dµ λa† (y) (A.53)
4πi ΘYM
τ= 2 + (A.54)
g 2π
1
Feµν
a
= ϵµναβ Fαβ
a
, (A.55)
2
where y µ = xµ + iθσ µ θ.
For SUSY QCD we denote the number of colors by N and the number of flavors by F . This is
a slight deviation from the canonical review literature which refers to these quantities as Nc and
NF respectively.
We make use of several abbreviations: ISS (Intriligator, Seiberg, Shih; metastable vacua
according to [67]), GKK (Giveon, Katz, Komargodski; uplifted metastable vacua according to

[123]), SUSY (supersymmetry), SUSY (supersymmetry breaking), MSSM (minimal supersymmet-
ric Standard Model), FCNC (flavor-changing neutral current), WZ (Wess Zumino), LSP (lightest
supersymmetric particle), χSF (chiral superfield), SYM (super Yang-Mills), SQCD (super QCD),
DSB (dynamical SUSY breaking), MGM (minimal gauge mediation), OGM (ordinary gauge me-
diation), EOGM (extraordinary gauge mediation), DGM (direct gauge mediation), sDGM (semi-
direct gauge mediation)...
B The Coleman-Weinberg Effective Potential

The Coleman-Weinberg potential, also known as the [quantum] effective potential, is the potential
term in the effective action after taking into account quantum corrections (say, to a given loop
order). In other words, it is the potential that determines the vacuum expectation value of fields.
Typically in field theory the Coleman-Weinberg potential is a small correction on top of the tree-
level potential and it’s not usually worth the trouble to calculate. In the case where the tree-level
potential is flat, however, the Coleman-Weinberg potential determines the vacuum of the theory
and whether or not spontaneous symmetry breaking occurs [124]. This is precisely what we have
shown to occur in generalized O’Raifeartaigh models, c.f Theorem 18.4.
There are three roads to deriving the Coleman-Weinberg potential. We shall review each them
for pedagogical value.
1. Quantum Mechanically: vacuum energy as a harmonic oscillator problem
2. Diagrammatically: calculate the vacuum bubble diagrams with vev insertions
3. Functionally: identify the momentum-independent part of the quantum effective action
114
B.1 Quantum Mechanical Derivation
We begin with the most straightforward procedure48 . This is the method presented in Dine’s
Cargese lectures [7]. We want to determine the vacuum energy of the theory. If you remember
quantum mechanics from back when you were in kindergarden, you’ll remember that it’s really
easy to calculate vac– I mean, zero-point energies. At least it’s easy in the case of a harmonic
oscillator. Fortunately, quantum fields are nothing but harmonic oscillators. The zero-point energy
is
1 ω
V0 = ℏω = (B.1)
2 2
This is precisely the object that we want to promote to the Coleman Weinberg potential, VCW = V0 .
In particular,
∫ √
1 1
VCW = ω = d¯3 k k 2 + m2
2 2
∫ √
1 1
= · 4π · dk k 2
k 2 + m2
2 (2π)3
∫ ( )
1 1 m2 1 m4
= 3
dk k 1 + − + ··· ,
(2π)2 2 k2 8 k4
where we’ve written k to mean 3-momentum. In the SUSY gauge theories that we’ll be interested
in, the mass is generally a function of the pseudomoduli, m = m(X). Some comments are in
order. First of all, it’s not necessarily obvious why there’s an integral over k, especially if you’re
trying to connect to formulae from quantum mechanics. It can sometimes be subtle going form
QM → QFT. Recall that ω is the frequency (energy) of a single quantum mechanical oscillator.
In quantum field theory these oscillators are tied together to form fields (see, e.g. Zee chapter 1
[125]). The ω that appears in the quantum mechanical expression is the term which appears in
the potential for that oscillator. The potential for a given oscillator depends on the wave mode
of the quantum field 49 . Thus the integral over the quantum field’s momentum k is interpreted by
the single quantum mechanical oscillator as a sum over a continuum of different potentials (i.e.
different oscillator systems). Next you might be concerned that we are only considering one such
quantum oscilator. Indeed, the vacuum energy is given by the contribution from each quantum
oscillator. This would just multiply the above result by the (infinite) volume of space. We must
recall, however, that the Coleman-Weinberg potential is given by peeling this factor off of the
vacuum energy, so our expression above is correct.
There is an additional source of infinities: the dk integral over the first two terms in the sum.
These infinities give a dependence on the UV cutoff Λ that diverges as Λ → ∞. Fortunately,
in a theory of supersymmetry these contributions cancel within supermultiplets: for each boson
48
We thank Felix Yu for sharing this derivation based on notes from Yuri Shirman’s lectures.
49
Here’s a handy example if you’re confused: consider a string (i.e. a field) with some stationary sinusoidal
oscillation. Consider a single point on that string an quantize it. That is, imagine that it oscillates in some other
sense (e.g. perpendicular to the string’s plane) and that the classical Hooke’s√ coefficient kH for this oscillation
depends on the potential energy relative to the string oscillation. The ω (∝ kH ) for the quantum mechanical
system depends on the displacement of the point relative to the string and hence really depends on the wave
number/momentum, k, of the string oscillation.
115
contribution there is a corresponding fermion contribution with opposite sign. This may seem
strange, but one must recall that all of these terms are really an expansion in vacuum bubble
diagrams with (pseudomoduli-dependent) mass insertions so that fermions really do pick up a
minus sign. The cancellation of these terms (particularly the m2 term) still carries over when
SUSY is spontaneously broken due to the supertrace rule. Thus for supersymmetric theories the
leading contribution comes from the m4 piece,
∑ 1 ∫ Λ 4
Fi +1 mi 1
VCW = 2
(−) (B.2)
i
(2π) m i
8 k
∑ 1 m2i
= (−)Fi +1 m
2 i
4
ln 2
(B.3)
i
64π Λ
1 4 M2
= STrM ln 2 , (B.4)
64π 2 Λ
where Fi is the fermion number of the state, STr is the supertrace, and M represents the mass
matrix for the supermultiplet with eigenvalues mi .
B.2 Diagrammatic Derivation

While the quantum mechanical derivation is simple, it’s perhaps unpalatable to those who are field
theorists at heart. When particle physicists calculate things, they want to see an expansion in
Feynman Diagrams. For this we turn to Coleman50 , whose pedagogy on this subject are highlighted
in his Erice lectures in Aspects of Symmetry [25], chapter 5.3 and its appendix. Coleman also does
the functional derivation in his section 5.3.4, but we’ll get to this in the next section. Coleman
also wrote the original paper on this subject [124].
Our strategy is to calculate the vacuum energy via bubble diagrams.
. + + +···
Where we’ve written the black dot to mean the two-point function including all tree-level
n-point vertices with (n − 2) vev insertions.
⟨ϕ⟩
⟨ϕ⟩ ⟨ϕ⟩
. = + + + ···
50
We are also grateful to Johannes Heinonen and Jay Hubisz for their insights on this derivation. We made use
of Jay’s solution set for Csaba Csáki’s Physics 662: Quantum Field Theory II course at Cornell University.
116
We have to sum over all such diagrams, where the solid line can be a scalar, vector, or fermion.
We just need to write down the appropriate two-point function for each case. Let’s start with the
scalar. The two-point function is
. = −iU ′′ (ϕcl ),
where we’ve written ϕcl = ⟨ϕ⟩ for readibility. For different flavors, we make the replacement U ′′ →
∂i ∂j U . The loop diagrams are just a momentum integral with alternating two-point insertions
and propagators. Thus the nth two-point insertion diagram is
∫ ( ′′ )n
1 U (ϕcl )
Mn = Tr d¯ k4
, (B.5)
2n k2
where we’ve written the symmetry factor 1/2n coming from rotations and reflections. We can
explicitly do the sum over each diagram,
∑∞ ∫ ∑ ∞ ( )n ∫ ( )
1 1 U ′′ (ϕcl ) 1 U ′′ (ϕcl )
Mn(scalar) 4
= Tr d¯ k 2
= − Tr d¯ k ln 1 −
4
2
. (B.6)
n=1
2 n=1
n k 2 k
∑ n
where we’ve used a /n = − ln(1 − a). We’ll do the integrals in a moment.
Next we can do the same summation for the fermions. The two-point functions are the usual
Dirac masses,
. = im = i(H + Aγ 5 ),
We’ve explicitly written the mass as a Hermitian plus and anti-Hermitian part as necessary for
the Lagrangian to be real. If there are multiple flavors which mix under the mass term, we can
add indices as appropriate. Now the nth two-point contribution is
∫ ( )n ∫ ( )n
1 1 1 † 1 mm†
Mn(fermion)
= − Tr d¯ k 4
m m =− 4
d¯ k . (B.7)
2n k/ k/ 2n k2
Once again, we sum this using the same identity,

∑ ∫ ( )
1 mm†
Mn (fermion)
= Tr d¯ k ln 1 − 2
4
. (B.8)
i
2 k
We’d like to move on to the gauge contribution. There are two things that we worry about.
First, we have to choose a gauge. Not a big deal. Next, we worry about diagrams with both gauge
bosons and scalars running in the loop, like
117
.
which come from the coupling of the scalar to a gauge boson. Fortunately, this problem goes away
in Landau gauge these graphs vanish. This is because the gauge boson propagator goes like
( )
i kµ kν
Landau
∆µν = 2 gµν − 2 . (B.9)
k k
The scalar coupling to the gauge boson is proportional to the scalar momentum, as one can check
by dimensional arguments. Thus since the mixed scalar-gauge boson diagrams contain terms like
k µ ∆µν k ν = 0, we don’t have to worry about them. Instead, all we need to include are the vector
mass insertions,
. = iM .
As before we can include indices if a gauge group is broken so that gauge bosons mix (e.g. Bµ
and Wµ3 mixing). The nth diagram is
∫ [ ( ) ]n
1 1 k µ kν
Mn (gauge)
= 4
Tr d¯ k δ ν− 2
µ
M 2
(B.10)
2n k2 k
∫ ( ) ( 2 )n
1 k µ kµ M
= Tr d¯ k δ µ − 2
4 µ
(B.11)
2n k k2
∫ ( 2 )n
3 4 M
= Tr d¯ k , (B.12)
2n k2
where we were a little sloppy with indices in the first line, but made it clear that the indices all
contract. It is not hard to see that
( )n ( )
k µ kν k µ kµ
δ ν− 2
µ
= δ µ− 2
µ
, (B.13)
k k
where on the left-hand side we assume that the indices are contracted. You can check the case
n = 2 and prove inductively. The sum gives
∑ ∫ ( )
3 M
Mn (gauge)
= − Tr d¯ k ln 1 − 2 .
4
(B.14)
i
2 k
Great! All of the integrals take the same form, so we can just do them all in one fell stroke.
∫ ( ∫ Λ2 ( )
a) i a
d¯ k ln 1 − 2 =
4 2
dkE ln 1 + 2 . (B.15)
k 16π 2 0 kE
118
One can explicitly do the integral on the right-hand side using the usual tricks, e.g. invoking the
appendix in Peskin and Schroeder [41]. Or, more practically, one can plug it into Mathematica,
i [ 2 ( a) ( a) ]
= Λ a + Λ ln 1 + 2 − a ln(Λ + a) − a 1 + 2 + a ln a
4 2 2 2 2
(B.16)
32π 2 [ Λ ] Λ
i 1
= 2
2Λ2 a − a2 − a2 ln Λ2 + a2 ln a + O(Λ−2 ) (B.17)
32π 2
Great. Plugging this in we get a nasty general formula
Λ2 [ ′′ †
]
VCW = Tr U (ϕ cl ) − mm + 3M 2
32π 2
1 [ ( ) ]
′′ 2 † 2
+ Tr (U (ϕcl )) − mm + 3(M ) 2 2
128π 2 [ ]
1 ′′ 2 U ′′ (ϕcl ) ( )
† 2 mm† M2
+ Tr (U (ϕcl )) ln − mm ln 2 + 3(M ) ln 2 .
2 2
(B.18)
64π 2 Λ2 Λ Λ
What a mess! This is the formula that you’d want to scribble down on your ‘handy general
formulae’ page. For this current document, all of our Lagrangians are supersymmetric, so several
cancellations occur. Let us ignore the gauge bosons (practically set M = 0), then the sums
between U ′′ (ϕcl ) and mm† are really supertraces. Thus the first two lines of the above formula all
cancel, and we’re left with the usual formula, Eq. (B.4).
Before we move on, let’s address a point about the loop expansion. One might wonder in
which sense the loop expansion is valid, i.e. how do we explain the loop expansion in terms of
some expansion parameter? Coleman (see also Srednicki chapter 21 [126]) shows us how to do
this by parameterizing the loop expansion by a dimensionless parameter that we will suggestively
call ℏ [127]. We will set ℏ = 1 after we’ve proved what we wanted. Let us write the Lagrangian
in terms of ℏ as
1
L (ℏ) = L. (B.19)
ℏ
For a given Feynman diagram, we now define P to be the power of ℏ appearing in the expression
for that graph. Each propagator carries a power of a since it is the inverse of the kinetic term.
Each interaction gives a power of a−1 . Thus, if we write I be the number of internal lines and V
be the number of vertices, we have
P = I − V. (B.20)
From the usual graph-ology, we know that the number of loops L is given by
L = I − V + 1. (B.21)
You can prove this by counting δ functions over momentum, appealing to fancy-schmancy graph
theory, or just drawing a few diagrams and convincing yourself. Combining these equations, we
get
P = L − 1, (B.22)
119
so that indeed, ℏ counts the number of loops. Great. Now what? Alright, so a loop expansion
corresponds to an expansion in ℏ. We still want to understand why this expansion is meaningful.
When we draw Feynman diagrams, we are expanding in small couplings. But we certainly aren’t
claiming that ℏ is a small parameter: we set it to one. Instead, (quoting Coleman)
The point is, rather, since the loop expansion corresponds to expansion in a parameter
that multiplies the total Lagrange density, it is unaffected by shifts of fields, and by the
redefinition of the division of the Lagrangian into free and interacting parts associated
with such shifts.
B.3 Functional Derivation

We’ve now given two derivations for the Coleman-Weinberg potential. The quantum mechanical
derivation was quick and easy. The diagrammatic derivation was intuitive. Now we review a third
derivation which has the benefit of field theoretic elegance. It is based on the functional integral.
This method is a little “old school” and is what you would find in books that refer to things
like “skeleton diagrams.” Modern textbooks with useful presentations include those by Greiner
[128] (whose notation is a bit odd in that W [J] ↔ Z[J]), Srednicki [126], and Banks [14]. A nice
treatment can be found in the lecture by Hugh Osborn [129]. An explicit calculation can be found
in Peskin (who writes W [J] as −E[J]) chapter 11.4 [41], which in turn follows the surprisingly
readable paper by Jackiw from 1974 [130]. We will roughly follow these last two references.
The general strategy is to calculate (practically to some low order in a loop expansion) the full
1PI quantum effective action Γ and then read off the momentum-independent term, i.e. the term
which survives when we specialize to a constant background field ϕcl . We discuss some nuances
about the effective action in Section 2.1, but for now it is sufficient to identify it as the action
whose tree level matrix elements represent a summation over all loop diagrams contributing to
the process. In other words, the vertices are actually ‘blob’ vertices which include a sum over 1PI
contributions.
Let us remind ourselves about the usual objects in the path integral formalism. The generating
functional of Green’s functions is a function of the source J(x) via
∫
Z[J] = dϕ eiS[ϕ]+iJ(x)ϕ(x) , (B.23)
which allows us to calculate n-point Greeen’s functions by taking n functional derivatives of the
source at the point in function space J(x) = 0. This generically is a sum over connected and
disconnected diagrams. It turns out that W [J], the generator of only connected diagrams, has a
simple relation to Z[J],
Z[J] = eiW [J] . (B.24)
Proof. For posterity, let’s discuss why this is true. This is easiest to see diagrammatically.
120
. = c + c +···+ c
(B.25)
The black blobs represent the Green’s function (connected and disconnected contributions) while
the white blobs are connected Green’s functions. Each external line represents a functional deriva-
tive with respect to J(xi ), where xi is the endpoint of the external line. Each black blob on the
right-hand side also has an expansion in products of lower order blobs. Each term in the sum,
we’ll call it Ta for transition matrix element51 ,
∑
. = Ta(6) .
a
Each Ta is a diagram which is generically disconnected. Let us write connected diagrams as Mi .

As an explicit diagrammatic example,
c c
. = + +···
c
Each term on the right-hand side is one Ta , while each connected diagram contributing to a given
Ta is a Mi . This is of course just a heuristic rewriting of Eq. (B.25) where we’ve fully expanded
each Green’s function (black blobs) in terms of connected Green’s functions. The first term on the
right-hand side is just (M3 )2 . This contribution implicitly contains a symmetry factor between
each identical connected piece52 . We can write this out explicitly as
1 ∏
Ta = (Mi )ni . (B.26)
Sa i
The symmetry factor only counts the interchange of identical connected diagrams so is given by
∏
Sa = ni !. (B.27)
i
51
Recall that the matrix element, M, is a component of the scattering matrix, S = T − 1.
52
This is not the same as the symmetry factor for a given connected diagram, which we will keep implicit.
121
Let us now write out the generating function of Green’s functions, Z[J]. By definition, this is just
the sum of all diagrams (up to a normalization which we ignore)
∑∑
Z[J] = Ta (n)
n a
∑∏ 1
= (Mi )ni
n i !
{ni } i
∏ ∑
= eMi = e i Mi , (B.28)
i
∑
where i Mi ≡ W [J] by definition. This then gives the desired result.
Let us prove this in a slightly more rigorous way. We can write out Eq. (B.25) more technically
as
( )n ∑ n ∑ ( )r ( )(n−r)
δ δ δ
Z[J]J=0 = i W [J]J=0 · Z[J]J=0
δJ r=1 comb.
δJ δJ
∑n−1 ∑ ( )r ( )(n−r−1)
δ δ δ
=i W [J]J=0 · Z[J]J=0 , (B.29)
r=0 comb.
δJ(x1 ) δJ δJ
where we’ve explicitly written a sum over combinations of the external points {xi } but for simplic-
ity of notation suppressed the position of each functional derivative. In the second line we pulled
out an explicit factor of δ/δJ(x1 ) for future convenience. We can write the sum over combinations
more explicitly as
∑ ∑ ( )
1 ∑ n
= = , (B.30)
n! r
comb. {i1 ,··· ,ir }⊂{1,···n} perm.
where ‘perm.’ means a sum over permutations of {1, · · · , n}. Then we may invoke the generalized
Leibniz rule,
∑n ( )
dn n dr dn−r
(f (x)g(x)) = f (x) · g(x). (B.31)
dxn r dxr dxn−r
r=0
Plugging this into Eq. (B.29), we get

( )
δ δ δ δ δW [J]
··· Z[J]J=0 = i ··· Z[J] . (B.32)
δJ(x1 ) δJ(xn ) δJ(x2 ) δJ(xn ) δJ(x1 ) J=0
This result is manifestly symmetric in the xi and so the sum over permutations gives a factor of
n! which just cancels the 1/n! above. Since this equation holds for each value of n, we can reduce
it to a simple [functional] differential equation, (i.e. the differential equation holds at each order
in the Taylor expansion)
δ δW [J]J=0
Z[J]J=0 = i Z[J]J=0 , (B.33)
δJ(x1 ) δJ(x1 )
whose solution is simply Eq. (B.24).
122
Ok, that was a bit of a long aside. Let’s move on to the 1PI quantum effective action, Γ[ϕcl ].
We define Γ to be the generator of 1PI diagrams. This means that if we treated Γ to be the
action of the theory, the tree-level diagrams would be exact (quantum mechanically) and there
would be no loop corrections to those diagrams. In other words Γ generates diagrams that already
include loop effects. In practice, of course, this can only be calculated to a given order in a loop
expansion. Let’s see how we can formalize this. Let’s define a generating functional ZΓ and a
generating functional of connected graphs WΓ associated with this effective action,
∫ ∫ d
ZΓ [J] = dϕ eiΓ[ϕ]+i d x J(x)ϕ(x) = eiWΓ [J] . (B.34)
WΓ is a sum of connected diagrams whose internal lines are exact propagators and whose vertices
are 1PI. By definition the restriction of WΓ to tree-level diagrams is equivalent to the usual
unrestricted W . We can use this to get a handle on Γ, but first we need to figure out how to
restrict WΓ to one-loop diagrams.
Fortunately we already discussed how to do this at length in the previous section when we
did the diagrammatic derivation of the Coleman-Weinberg potential. We found that the natural
parameter that counted the powers of loops is ℏ. Restoring this dependence, we have the
∫ ∫ 4
i
ZΓ,ℏ [J] = dϕ e ℏ (Γ[ϕ]+ d x J(x)ϕ(x)) = eiWΓ,ℏ [J] , (B.35)
where we know how to write the expansion in ℏ in terms of number of loops,

∑
∞
WΓ,ℏ [J] = ℏL−1 WΓ,L [J]. (B.36)
L
So our first step in connecting Γ to our usual objects, Z[J] and W [J] is the relation
W [J] = WΓ,L=0 [J]. (B.37)
We can go on and bring Γ into the mix by evaluating ZΓ,ℏ in Eq. (B.35) using the stationary
phase approximation,
δΓ[ϕ]
= −J(x). (B.38)
δϕ(x)
This is sometimes called the quantum equation of motion. Define the classical field to be the field
configuration ϕ(x) = ϕcl (x) that satisfies this equation. Then the generating functional associated
with Γ can be written as
[ ( ∫ ) ]
i
ZΓ,ℏ = exp Γ[ϕcl ] + d x J(x)ϕcl (x) + O(ℏ ) .
4 0
(B.39)
ℏ
Putting this together with our loop expansion in ℏ we get the important relation
∫
Γ[ϕcl ] = −W [J] + d4 x J(x)ϕcl (x). (B.40)
123
In other words, Γ[ϕcl ] is the Legendre transform of W [J]. Some treatments take this as the
definition of the effective action and from there derive the more intuitive definition above, though
we find it is more instructive to do things in this order.
We can better motivate the name ‘classical field’ by remembering that in the background |Ω⟩
of some general source J(x), the ‘background’ value of the field is
δW [J]
⟨Ω|ϕ(x)|Ω⟩J =
δJ(x)
∫
δΓ[ϕcl ] δϕcl (y)
= + ϕcl (x) + d4 y J(y)
δJ(x) δJ(x)
∫ ∫
4 δΓ[ϕcl ] δϕcl (y) δϕcl (y)
= dy + ϕcl (x) + d4 y J(y)
δϕcl (y) δJ(x) δJ(x)
∫ ( )
δϕcl (y) δΓ[ϕcl ]
= d4 y + J(y) + ϕcl (x)
δJ(x) δϕcl (y)
= ϕcl (x). (B.41)
Good. Now that we’ve thoroughly reviewed the basics, let’s calculate the Coleman Weinberg
potential. Our method will not be direct, but I promise it will be elegant. First let’s expand
about the classical field,
ϕ(x) = ϕcl (x) + φ(x). (B.42)
Now consider the generator of connected diagrams,

∫
W [J] = d4 x (L [ϕcl ] + J(x)ϕcl (x))
∫ ( )
4 δL
+ d x φ(x) + J(x)
δϕ
∫
1 δ2L
+ d4 x d4 y φ(x)φ(y)
2 δϕ(x)δϕ(y)
∫
1 δ3L
+ 4 4 4
d x d y d z φ(x)φ(y)φ(z) + ··· . (B.43)
3! δϕ(x)δϕ(y)δϕ(z)
Let’s drop the terms of O(φ3 ) and perform the quadratic integral. The Gaussian integral is
our bread-and-butter tool for path integrals, so you knew this was coming. Using the usual
manipulations, we can write the generating functional Z[J] as
∫ [ ∫ ∫ ]
4 i 4 4 δ2L
dφ exp i d x (L [ϕcl ] + J(x)ϕcl (x)) + d x d y φ(x) φ(y)
2 δϕ(x)δϕ(y)
[ ∫ ]( [ 2 ])−1/2
δ L
= exp i d x (L [ϕcl ] + J(x)ϕcl (x)) det −
4
, (B.44)
δϕδϕ
We can see explicitly the classical contribution and the first order contribution from quantum
corrections. If we included higher order terms in φ we would get a Feynman diagram expansion
124
with respect to the classical background field. We can take the logarithm of Z to get
∫ [ 2 ]
1 δ L
iW [J] = i d x (L [ϕcl ] + J(x)ϕcl ) − det −
4
+ ··· , (B.45)
2 δϕδϕ
where the “· · · ” represents connected diagrams and counter terms. We’re not going to worry too
much about the counter terms since we know from that in the supersymmetric limit there are no
UV divergences that we have to regulate. But if you wanted to be precise, you would need to
replace L → L + Lc.t. , where the first term (what we’ve written explicitly in our derivation here)
is the renormalized Lagrangian and the second term contains counter terms. We would then need
to identify J(x) as the renormalized source which satisfies
δL [ϕcl ]
+ J(x) = 0, (B.46)
δϕ(x)
and a counter term source δJ(x) which acts to enforce ⟨ϕ(x)⟩J = ϕcl . Upon expanding about
ϕcl , the counter term Lagrangian just provides the usual counter term vertices and an overall
constant that can be used to satisfy renormalization conditions for any divergences in the functional
determinant. Those who really want to be careful with counter terms can follow the exposition in
Peskin’s chapter 11.4 [41].
We learned above that to get the effective action (finally!) we just take a Legendre transform
of this object. We obtain
∫ [ 2 ]
i δ L
Γ[ϕcl ] = d x L [ϕcl ] + ln det −
4
− i(· · · ). (B.47)
2 δϕδϕ
As a sanity check, note that there is no J(x) dependence. Γ is only a function of ϕcl . Alright.
Now we’re getting somewhere. The effective potential is the momentum-independent part of the
effective action, i.e. the part that isn’t kinetic. It’s easy to identify this: we just have to specialize
to the case of a constant background classical field. Then VCW = −Γ[ϕcl ]/(vol), with ϕcl = const
and the volume of spacetime being factored out,
[ 2 ]
i δ L
VCW = V (ϕcl ) − ln det − + ··· (B.48)
2(volume) δϕδϕ
To calculate these functional determinants we use the handy relation
ln det ∆ = Tr ln ∆, (B.49)
where the trace is over eigenvalues of the operator ∆. For our purposes,
δ2L
= ∂ 2 − U ′′ (ϕcl ) = ∂ 2 + m2 + · · · . (B.50)
δϕδϕ
Let’s assume that U ′′ = −m2 . Since U ′′ is constant (because ϕcl is constant), the eigenfunctions
are plane waves whose eigenvalues are −k 2 +m2 . The trace over the logarithm of these eigenvalues
can be defined rigorously by taking the continuum limit of a discrete system (e.g. a large box),
∑ ∫
ln(−k + m ) → (volume) d¯4 k ln(−k 2 + m2 ).
2 2
(B.51)
k
125
Plugging this in and doing an implicit Wick rotation, we get
∫
1
VCW = V (ϕcl ) + d¯4 k ln(k 2 + m2 ). (B.52)
2
This now is now of the same form as the integrals of logarithms that we did in the previous section.
Just to show off a little, we’ll pull out a few more tricks to do these integrals explicitly. First we’ll
use a handy representation of the natural logarithm,
∫ ∞
a dz ( −az )
− ln = e − e−bz . (B.53)
b 0 z
We can use this with a = k 2 + m2 and b = 1 to let us write the quantum correction as
∫ ∫ ∫ ∞
1 1 dz ( −(k2 +m2 )z −z
)
− 4 2 2
d¯ k ln(k + m ) = 4
d¯ k e −e . (B.54)
2 2 0 z
The second term on the right-hand side is divergent and will ultimately be eaten by counter terms,
so we’ll just drop it like it’s hot. The next trick that we’ll do is to perform the d¯4 k integral, which
is now Gaussian.
∫ ∫ ∫
1 1 ∞ dz −m2 z
d¯4 k e−k z
2
d¯ k ln(k + m ) = −
4 2 2
e
2 2 0 z
∫
1 ∞ dz −m2 z 1
=− e
2 0 z (4πz)d/2
∫ ∞
1
dz z −1−d/2 e−m z
2
=− d/2
2(4π)
∫0 ( )
1 1 ∞ d −d/2 −m2 z
= dz z e
(4π)d/2 d 0 dz
∫
m2 1 ∞
dz z −d/2 e−m z
2
= d/2
(4π) d 0
( ) ( )
md d md d
= d/2
Γ 1− =− d/2
Γ −
(4π) 2 2(4π) 2
Using the MS renormalization prescription with a renormalization scale µ to remove divergences

as d = 4 − ϵ → 4, this gives us
( ) ( ( 2) )
md d m4 m 3
− d/2
Γ − = 2
ln − . (B.55)
2(4π) 2 64π µ2 2
This gives the leading quantum correction to the classical potential. One will note that the term
with the logarithm matches that of Eq. (B.18), where we use m2 → −U ′′ (ϕcl ). However, the term
proportional to (U ′′ )2 does not appear to match. Osborn notes that this is due to the arbitrariness
of our renormalization prescription, in particular the freedom to add a finite counter term that is
a quartic polynomial in ϕ [129]. This would affect the coefficient of (U ′′ )2 so that this coefficient
cannot be physical. One will note, on the other hand, that the coefficient of the (U ′′ )2 ln U ′′ term is
physical. These ambiguities can be removed by specifying the derivatives of the effective potential.
126
B.4 Integrating out fields
As we know from our study of supersymmetric QCD, of the knobs that we have to play with53 is
to integrate out massive fields. Intriligator and Seiberg make some important pedagogical notes
about what this means for our cherished arguments of holomorphy and the effective potential
[2, 1]. Consider a superpotential which (say, at some point on the pseudomoduli space) takes the
form
1
W = Φa Mab Φb + · · · . (B.56)
2
Integrating out Φ will give us an effective Kähler potential of the form
[ ( )]
1 † MM†
Keff = − Tr M M log . (B.57)
32π 2 Λ2
[Check: Check this formula, e.g. see hep-th/9605149 or Kuzenko. I probably need to do some
supergraph calculations.] Now, in the limit of small SUSY-breaking, we can use this effecive
Kähler potential as a trick to approximate the Coleman-Weinberg potential. Suppose that the
mass matrix M depends on the pseudomodulus X. Then the approximate CW potential, which
Intriligator and Seiberg call the ‘truncated’ potential, is
Vtrunc = (Keff X,X )−1 |∂X W |2 . (B.58)
This is just the tree-level scalar potential that one would get with a non-trivial Kähler potential.
Vtrunc approximates VCW to leading order in
FX = −(Keff X,X )−1 ∂X W . (B.59)
This is verified in the ISS paper [67].
One ought to be careful at the origin of a theory where fields have been integrated out. At
the points on the pseudomoduli space (usually the origin) where the integrated-out fields become
massless, the effective theory becomes singular. This non-analyticity is the way the theory is
telling us that another degree of freedom becomes operative, i.e. our effective theory is breaking
down. This is of course what we would expect since it makes no sense to integrate out a massless
(or very light relative to the scale) field.
B.5 An illustrative example

In their SUSY-breaking lecture notes, Intriligator and Seiberg make some important notes about
the Coleman-Weinberg potential [1]. We will follow their analysis of a simple metastable SUSY-
breaking model (from Section 2.3 of their lectures) to do a sample calculation and explore some
further technical details.
The simple model is composed of two chiral superfields, X and q with a canonical Kähler
potential and a superpotential
1
W = hXq 2 + f X. (B.60)
2
53
Giggle if you’re British.
127
This already has a form that is similar to the ISS model. We note that if the first term were absent
this would simply be the Polonyi model W = f X, which is a simple (the simplest?) SUSY-beaking
model. A simple analysis of the potential for this model, however, yields
2
1 2
V = |hXq| + hq + f
2
2
√
so that supersymmetric vacua exist at ⟨X⟩ = 0 and ⟨q⟩ = ± −2f /h. Now some SUSY intuition
should kick in: when restricted to the submanifold ⟨q⟩ = 0, there is a pseudoflat direction param-
eterized by ⟨X⟩. We can then move to a region of large ⟨X⟩, where the qs thus obtain large mass
terms and can be integrated out. In this regime we return to the SUSY-breaking Polonyi model.
Thus we can start thinking about constructing a metastable vacuum along this pseudomoduli.
(One can pause to briefly reflect on how this is a simple case of the ISS ‘macroscopic model I’ in
Section 20.2.)
Let’s start by considering the spectrum along this pseudoflat direction. We can be optimistic
and hope that the Coleman-Weinberg potential stabilizes a SUSY-breaking vacuum. (It will not.)
The quarks obtain masses
m20 = |hX|2 ± |hf | m1/2 = hX,
where we’ve been lazy and have written X = ⟨X⟩. We note immediately that the squarks are
tachyonic if |X|2 < |f /h|. This means the potential slopes downward along the ⟨q⟩ direction down
to the supersymmetric vacuum described above. That’s fine, we should have expected this to
happen at tree-level since we already knew the lower-energy SUSY vacuum existed. Let’s work in
the non-tachyonic regime |X|2 > |f /h| so that we may expand in the parameter

f
z ≡ 2 .

X h
Let’s work out how the Coleman-Weinberg potential lifts the pseudomodulus. Recall that
( )
1 M2
VCW = STr M log 2 ,
4
(B.61)
64π 2 Λ
where M is the classical (i.e. tree-level) mass matrix. Thus the Coleman-Weinberg potential for
the pseudomodulus X is
(1) 1 |hX|2 ∑ 1 ( )2 |hX|2 ± |hf |

VCW = − 2|hX| 4
log + |hX|2
± |hf | log (B.62)
64π 2 Λ2 ±
64π 2 Λ2
[ ]
|hf |2 hX 2 3 z 2
= log + − + O(z 4 ) , (B.63)
32π 2 Λ 2 12
where we’ve expanded in our ‘small parameter’ z. This potential lifts the degeneracy of the
pseudomoduli ⟨X⟩ (recall that for brevity we’ve been habitually droping the angle brackets) in
such a way that the potential increases with |X|. Thus we see that the Coleman-Weinberg potential
is indeed pushing us back into the tachyonic region that we were hoping to avoid.
128
So we’ve now worked through a very simple example of not-quite metastale SUSY-breaking.
It’s nice to see an example where the effective potential does not stabilize our pseudomoduli where
we want, since most papers only present successful cases. In practice when dealing with larger
global symmetries (e.g. super QCD with some number of flavors) it can become very tedious to
calculate pseudomoduli by hand. One can usually get away with tricks to determine the stability of
the pseudomoduli space (e.g. in the ISS macro model discussed in Section 20.2.4), but to compute
the entire one-loop Coleman-Weinberg effective potential one generally has to diagonalize mass
matrices via some computer algebra system like Mathematica.
One interesting development on this front is a computational tool by Korneel van den Broek
called Vscape [131]. It is a software package that calculates the effective potential for the pseu-
domoduli space of an ungauged theory of chiral superfields, such as the ISS macroscopic model
I.
B.6 Cutoff dependence

Now it’s somewhat important to discuss the cutoff dependence of the Coleman-Weinberg formula
that we’ve derived. Let’s focus on the case of a supersymmetric theory, where we do not have
any divergent terms in the CW effective potential, but we still have the explicit appearance of Λ.
One might be somewhat perturbed by the Λ in the formula for the effective potential: what does
it mean and how do we pick it to get meaningful results?
It turns out that the cutoff dependence can be removed explicitly if we work with running cou-
plings [1]. Now things are starting to sound familiar from the standard theory of renormalization.
Consider the simple illustrative example in the previous section, where we are again living on the
pseudomoduli ⟨q⟩ = 0. One can see that Eq. (B.63) has explicit terms with |Λ| appearing. Let us
define the running coupling
[ ( ) ]
|h|2 3 µ2
f (µ) = f0 1 + + log 2 + O(h ) .
4
(B.64)
64π 2 2 Λ
We will motivate this in a moment. Let us first behold a ‘miracle’: with respect to this running
coupling, the Coleman-Weinberg potential is independent of the cutoff Λ:
[ ( 2 ) ]
|h|2 z
VCW = |f (|hX|)| 1 −
2
− + O(z ) + O(h ) ,
4 4
(B.65)
32π 2 12
where we’ve evaluated f (µ) at the scale of the massive fields q: µ = |hX|.
We review super QCD below, but you might wonder why we’re talking about a running coupling
when we know from Seiberg-ology that the holomorphic couplings in the superpotential do not
run, i.e. they are not renormalized. On the other hand, we do know that there is still wavefunction
renormalization and indeed, we can understand the above running in terms of the wavefunction
renormalization ZX of the field X.
The tree-level potential above comes from FX , so that at leading order only ZX can affect V .
−1 −1
Veff = ZX |WX |2 + finite = ZX |f |2 + finite. (B.66)
129
This gives us
∂Veff 1
− = γX |f |2
= STr M4 + O(h2 ), (B.67)
∂ log Λ2 64π 2
where γX is the anomalous dimension of X.
[Work: Flesh this out a little bit, it’s kind of important.]
C Phases of Gauge Theories

See: Srednicki chapter 82, Preskill notes. Larsen/Terning notes. Banks. Fradkin54 .
Phase V (r)
Coulomb ∼ 1r
Free electric ∼ r log(Λr)
1
Free magnetic ∼ log(Λr)

r
Higgs ∼ constant
Confining ∼ r.
Under electromagnetic duality,

Free electric ↔ Free magnetic
Coulomb ↔ Coulomb
Higgs ↔ Confining
Can check for confinement using Wilson loops. In s-confined theories the Higgs and Confined
phases are identical. (Why?)
*** Oblique confined phase. http://www.springerlink.com/content/l563v000661j125r/
D Review of Anomaly Cancellation

Here we collect the basic technical machinery for calculating the cancellation of gauge anomalies.
This is not meant to be a comprehensive review, please refer to other resources.
D.1 Overview and background

An anomaly is a classical symmetry which is broken quantum mechanically. For example, the
chiral (ABJ, abelian) anomaly is often the first example of an anomaly where a global symmetry
is broken in a gauge background. Classically, the axial current is conserved ∂µ JAµ = 0, where as
quantum mechanically it is proportional to a non-perturbative term, ∂µ JAµ ∼ F µν Feµν . Since this
can be calculated from the triangle diagram with an axial current and two photons, we call this
the U(1)A U(1)2EM anomaly. A few other remarks about the chiral anomaly:
54
http://webusers.physics.illinois.edu/~efradkin/phys583/physics583.html
130
• The anomaly is generated by chiral ‘zero mode’ (with respect to the Dirac operator) fermions
and are independent of the fermion mass.
• The anomaly is one-loop exact; higher order corrections are lower superficial degree of di-
vergence.
• In dimensional regularization, the anomaly appears in the definition of γ 5 in d ̸= 4.
• Fujikawa showed that the anomaly comes from the non-invariance of the path integral mea-
sure.
• In non-abelian theories, Green’s functions with odd numbers of axial couplings up to 5-point
functions contribute anomalous terms. However, if the triangle diagram vanishes then so do
all other anomalous diagrams.
Other examples of anomalies include gravitational anomalies and the conformal anomaly; the
latter famously manifested through the renormalization group.
The main anomaly we’ll consider here are non-Abelian gauge symmetries55 . Since gauge sym-
metries are really redundancies of how we describe a theory, an anomaly in this symmetry would
be manifestly non-sensical. We thus require theories to be gauge anomaly-free. Non-Abelian
anomalies are intimately related to instantons56 .
Anomalies can be calculated perturbatively through triangle diagrams with chiral fermions, or
alternately non-perturbatively using the path integral methods pioneered by Fujikawa. As this is
standard fare in quantum field theory, we will not dwell on the technical calculation. Fore more
details about anomalies in the spirit of this document, see Preskill’s review57 [132] or one’s favorite
textbooks (Terning, Banks, Nakahara, and Weinberg are especially good).
D.2 The Anomaly Coefficient

The anomaly is quantified by the non-vanishing divergence of the Noether current associated with
the anomalous symmetry. In the divergence is non-zero, but is proportional to a total derivative.
More importantly, for gauge anomalies, it is proportional to the anomaly coefficient,
[ ]
Aabc = Tr T a {T b , T c } , (D.1)
where the T s are the generators of the appropriate symmetry and Aabc depends on the fermion
representation. The trace here refers to a sum over all fermions running in the loop, as Burgess and
Moore say, “every color of every flavor of quark and every lepton in each generation with T denoting
the action of the symmetry on that particular particle type.” You can see that this is precisely the
group theoretic factor that appears wen you draw a triangle diagram with gauge currents at each
55
In fact, it is worth pointing out the very elegant differential geometry that rigorously unifies many of the
heuristic manifestations of anomalies in quantum field theory, e.g. the relation of the Abelian and non-Abelian
anomalies in different dimensions via the Stora descent equations. For more on this see http://www.lepp.cornell.
edu/~pt267/files/BSMclub/Flip_11April11_notes.pdf.
56
57
http://www.theory.caltech.edu/~preskill/pubs/preskill-1991-anomalies.pdf
131
corner. To simplify gauge anomaly calculations, it is convenient to define an anomaly coefficient
A(r) for fermions in representation r relative to the fundamental representation,
[ ]
Aabc (r) = A(r)Tr TFa {TFb , TFc } . (D.2)
Now some useful properties of the anomaly coefficient:
• If r is a complex representation, then A(r̄) = −A(r).
• A(r1 ⊕ r2 ) = A(r1 ) + A(r2 )
• A(r1 ⊗ r2 ) = A(r1 ) dim(r2 ) + A(r2 ) dim(r1 ).
This tells us, for example, that chiral fermions in vector-like (left-right symmetric) representations
(e.g. Dirac fermions) also do not contribute to anomalies. Further, chiral fermions in real (r̄ = r)
or pseudo-real (r̄ = U † rU ) representations do not contribute to anomalies. The value of A(r) for
various representations of SU(N ) is given below, copied from [5].
Irrep dim(r) 2T (r) A(r)

N 1 1
Ad N2 − 1 2N 0
N (N −1)
2
N −2 N −4
N (N +1)
2
N +2 N +4
N (N −1)(N −2) (N −3)(N −2) (N −3)(N −6)
6 2 2
N (N +1)(N +2) (N +2)(N +3) (N +3)(N +6)
6 2 2
N (N −1)(N +1)
3
N −3
2
N −9
2
N 2 (N +1)(N −1) N (N −2)(N +2) N (N −4)(N +4)
12 3 3
N (N +1)(N +2)(N +3) (N +2)(N +3)(N +4) (N +3)(N +4)(N +8)
24 6 6
N (N +1)(N −1)(N −2) (N −2)(N 2 −N −4) (N −4)(N 2 −N −8)
8 2 2
It is conventional to work with only left handed fields, e.g. L and ēR .
D.3 Cancellation of gauge anomalies

The key point is that for a sensible gauge symmetry, the anomaly must vanish. In other words,
∑
A(ri ) = 0. (D.3)
i
Before getting to the nitty-gritty of checking such an expression, let’s remark that the easy way for
anomalies to cancel it to work within theories where there is no anomaly. This condition condition
of having only (pseudo-)real gives representations us a list of groups which, in four dimensions,
have vanishing anomaly coefficients: SU(2), SO(2n + 1), SO(4n), SO(4n + 2), Sp(2N ), G2 , F4 ,
E6 , E7 and E8 . Note that SO(10) and E6 are potential GUT candidates because they can fit
the Standard Model as a subgroup. Alternately, for an arbitrary gauge group, one can enforce
132
anomaly-freedom by only including fermions in vector-like representations. A useful fact is that if
anomalies cancel in a group, then anomalies will cancel in any subgroup. Thus if you construct a
unified theory without anomalies, e.g. SU(5) with the 10 ⊕ 5̄ representation, then you know that
the anomalies of the Standard Model must also cancel.
Next let’s note that the importance of anomaly cancellation only holds for gauge symmetries.
There is nothing ‘wrong’ with a theory whose global symmetries are anomalous.
Let us now confirm that the Standard Model is anomaly-free. This list is from Burgess and
Moore [13].
• A(3, 3, 3): Since the quarks are left-right symmetric (vector-like) with respect to SU(3)c , the
anomalies cancel.
• A(3, 3, 2): This one is also easy. We can ignore the SU(3) parts and just focus on the SU(2)L
piece. The trace will include a trace over the Pauli matrices for each doublet. Since the
pauli matrices are traceless, this anomaly vanishes.
• A(3, 3, 1): Now we consider the case when there are U(1) generators. To do this it is
useful to note that λa , λb = 43 δab + 2fabc λc , where the λs are Gell-Mann matrices. The
color trace gives a factor of three in the first term and causes the second term to vanish.
Thus the anomaly is given by the sum of the hypercharges of each quark: A(3, 3, 1) =
3(2Y [QL ] + Y [ŪR ] + Y [D̄R ]) = 2[2(1/6) + (−2/3) + (1/3)] = 0.
• A(3, X, Y ): For X, Y ̸= 3 this will be proportional to the trace of a Gell-Mann matrix and
so vanishes (just like A(3, 3, 2)).
• A(2, 2, 2): Unlike color, the the electroweak group is not left-right symmetric. However, we
noted above that SU(2) is anomaly free. This is because it is pseudo-real: σ̄ i = −σ 2 τ i σ 2 .
• A(2, 2, 1): Here we have {σ i , σ j } = 2δ ij . Counting the generation and color multiplicities,
we thus have a sum over the hypercharge of each doublet, A(2, 2, 1) = 3(Y [L] + 3Y [Q]) =
3[(−1/2) + 3(1/6)] = 0.
• A(2, 2, 1): This is proportional to the trace of a single generator and vanishes.
• A(1, 1, 1): This is the sum over all fermions with respect to the cube of their hypercharges,
A(1, 1, 1) = 3(2Y [L]3 + Y [ĒR ]3 + 6Y [Q]3 + 3Y [ŪR ]3 + 3Y [D̄R ]3 ) (D.4)

[ ( ) ( )3 ( )3 ]
3
1 1 2
=3 2 − + (1)3 + 6 +3 − + 3 (13)3 = 0. (D.5)
2 6 3
Note that anomaly cancellation sets a rigorous, non-trivial condition on the hypercharges and
cubes of hypercharges of particles. This prevents giving an arbitrarily small, but finite, charge to
the neutrino by shifting its hypercharge by an small amount. Finally, Witten an Alvarez-Gaumé
showed that gravitational anomalies impose an additional constraint on U(1) gauge group factors:
in
∑ order for consistent gravitational coupling, the U(1) generators must be traceless over fermions,
Y = 0.
133
D.4 Comments on global anomalies
These are mainly form Burgess and Moore.
• One can also calculate the anomalies for global symmetries. We already met the chiral
anomaly, A(A, 1, 1). We can also consider baryon number, for which A(3, 3, B) = 0 but
A(2, 2, B) = 3. Similarly, lepton number, e.g. A(2, 2, L) = 1. Note that by ‘lepton number’
here we mean a particular flavor of lepton.
• Note that A(B, B, B) = 0 while A(L, L, L) − 2. Further, A(G, G, B) = 0 while A(G, G, L) =
1, where G represents gravity.
• Global anomalies needn’t vanish. The effect of the anomalies on low-energy physics can be
interpreted as topological objects, instantons and sphaelerons. These effects are proportional
to e−8π/g so that anomalous global symmetries with respect to weakly coupled gauge groups
2
are good approximate symmetries, whereas anomalous global symmetries with respect to
strongly coupled gauge groups are strongly broken.
• The anomaly-free global symmetries of the Standard Model are given by linear combinations
of the anomalous symmetries above. Including gravitational anomalies, these are Le − Lµ
and Le − Lτ , where Lµ − Lτ is linearly dependent on the other two.
• Notice that all of the SM anomalies are the same for baryon number as they are for total
lepton number (3L). The gravitational, B 3 , and L3 anomalies agree if we include right-
handed neutrinos. Thus the combination B − L is anomaly free in the theory with right-
handed neutrinos.
• The η ′ problem: see my A-exam for more details58 . The chiral U(3) symmetries of QCD
are generally anomalous. The anomalies with SU(3)c are all proportional to the trace of the
generator’s 3 × 3 representation. Thus the traceless symmetries are non-anomalous in the
limit where the electroweak interactions are negligible. Since, as an equation of Lie algebras,
U (3) = SU (3) × U (1), only the U (1) generator carries a trace and is thus strongly violated
by SU(3)c . Thus QCD anomalies break U (3)L × U (3)R → SU (3)L × SU (3)R × U (1)B where
U (1)B is the non-anomalous U (1) that is vectorlike with respect to the quarks. The ‘broken’
U (1)A symmetry is the reason why there is no ninth pseudo-Goldstone pion, i.e. why the η ′
is so heavy.
E Analytic Continuation into Superspace

As we saw in Section 16.2.3, unless you are the Rambo of loop integrals, the gauge-mediated
SUSY-breaking masses in the visible sector (i.e. the soft terms in the MSSM) can be very tedious
to calculate since these terms appear at one- and two-loop order. The exact formulae were rather
involved, even though the limit of small (F/M 2 ) simplified the expressions dramatically to forms
that one could have guessed from pure ‘dimensional analysis59 .’. Fortunately we can do better.
58
59
By ‘dimensional analysis’ we mean: α carries units of ‘gauge-ness,’ (4π) carries units of inverse loop, F carries
units of SUSY-breaking, and M −1 carries units of mediation.
134
We now present a very handy trick for easily calculating the soft SUSY-breaking terms in that
limit based on holomorphy. This so-called analytic continuation into superspace was first
developed by Giudice and Rattazzi [133] and was later expanded to include higher-loop correc-
tions in collaboration with Arkani-Hamed and Luty [134]. Further references are Patrick Meade’s
TASI09 lectures60 and John Terning’s textbook [5].
E.1 Overview
In gauge mediated supersymmetry breaking, a chiral superfield (or set of superfields) X in the
hidden sector spontaneously breaks SUSY by obtaining a vacuum expectation value
⟨X⟩ = M + θ2 F.
In minimal gauge mediation, the lowest-component (SUSY-preserving) vev M gives a mass to the
messenger fields ϕ, φ which transmit SUSY breaking to the MSSM. The higher component vev θ2 F
is the actual SUSY-breaking term and is transmitted to the MSSM only through the messengers.
A sensible thing to consider is to use the power of effective field theory by integrating out the
messenger fields and considering effective operators with MSSM fields coupled to the vevs of the
SUSY-breaking hidden-sector fields X. In such a formalism we treat the X as a SUSY-breaking
spurion in the visible sector61 . In such a set-up the effective operators would heuristically take
the form
∫ ∫ †
2 X 4 X X †
Leff = c1 d θ Wa W + c2 d θ 2 Q Q.
a
M M
The problem with this approach is now staring us in the face: in order to go through the EFT
procedure straightforwardly62 , one still has to compute one- and two-loop diagrams and do a
matching to determine the c1 and c2 coefficients. Our usual approach has failed us63 .
Now we can be clever. Giudice and Rattazzi reminded us that the lowest-order vevs for these
effective operators, i.e. the non-SUSY-breaking vevs, are just terms that are contributions to
the usual kinetic Lagrangian in supersymmmetry. The coefficients of these terms are just the
(holomorphic) gauge coupling τ ∼ g −2 and the wavefunction renormalization Z of the chiral
superfields. Further, we already know the RG behavior of the gauge coupling and wavefunction
renormalization from well-known one-loop calculations. It would be great if we could insert these
physical quantities could serve as the lowest component of the spurion coefficients in Leff and then
60
61
We can proceed as if the X field is nothing more than a ‘trick’ in the visible sector to parameterize the a priori
unknown physics of SUSY-breaking. In such a framework we would never have to consider whether or not X is in
any sense a physical field. In this case, however, by the assumption of gauge mediation we know that X is actually
a physical field that is just hidden from the visible sector through couplings via heavy messengers. In this sense
we can interpret our spurion analysis ‘literally.’
62
The implementation of EFT in particle physics is an under-appreciated skill. Good introductions can be found
in, e.g., Witek Skiba’s lectures at TASI09 at http://www.colorado.edu/physics/Web/tasi09_annc.html or the
lectures by Cliff Burgess [13] or James Wells [135]. The most immediate application of the effective field theory
framework are electroweak precision observables; the main papers for phenomenologists are Barbieri, Pomarol,
Rattazzi, and Strumia [136], Han and Skiba [137], and Cacciapaglia, Csáki, Marandella, and Strumia [138].
63
http://xkcd.com/55/
135
‘promote’ their well-known RG dependence to a form for the SUSY-breaking higher-component
spurion vevs. We can, in fact, do this. The running values of τ (µ, Λ) and Z(µ, Λ) at some scale µ
and for some UV cutoff Λ, are given by the solution of the RG equations. These solutions include
terms that come from integrating out the messenger fields at the scale M . If we promote the M -
dependence of these expressions to a dependence on the spurion superfield X, then we convert τ
and Z into superfields whose higher-component (SUSY-breaking) vevs are given straightforwardly
in terms of the X vevs. This is called analytic continuation into superspace. We will see
that the miracle is that in the F ≪ M limit which is usually sufficient in most gauge mediation
models, these higher-component vevs have precisely the coefficients that we would obtain via
explicit calculation of two-loop results.
This result is at first magical and then, after some thought, tautological: such a result had to
be true due to holomorphy and the constraints of supersymmetry. In a broader sense, this is an
example of the use of supersymmetry to constrain the behavior of a quantum field theory that
would otherwise be much more difficult to ascertain.
E.2 Preliminary results

Before we do any heavy-lifting, let us remind ourselves of a few results and notation. Recall that
the soft SUSY-breaking terms for a SUSY gauge theory (in particular, a Wess-Zumino model
coupled to a gauge field) take the form
1 1 1 1
Lsoft = − mλ λλ − Aϕ3 − Bϕ2 − ϕ∗ ϕ + h.c. (E.1)
2 6 2 2
These are terms which are manifestly non-supersymmetric but that do not spoil the cancellation
of quadratic divergences that solves the Hierarchy problem. This is identical to saying that these
are the terms that can appear when supersymmetry is broken spontaneously. Heuristically this
is sensible since at energies well above the scale of the vacuum energy (the SUSY-breaking order
parameter) the theory should appear supersymmetric with all nasty divergences canceling. We
can write this out more formally,
Theorem E.1. The soft supersymmetry-breaking terms that otherwise respect all of the sym-
metries and constraints of the theory (e.g. gauge symmetries, renormalizability) are identical to
terms which can be obtained by promoting the couplings of the manifestly supersymmetric theory
to spurion superfields which obtain higher-component vacuum expectation values.
Proof. If one sits down and thinks about this for a moment then it follows tautologically. More
formally, one can construct an isomorphism between any soft breaking term to an appropriate
SUSY-preserving term which would contribute to the soft breaking term if its coupling obtained
a higher-component vev. Let us assume renormalizability so that we only have to look at the su-
perpotential. The generalization to non-renormalizable theories is straightforward. First, consider
any soft SUSY-breaking term and promote each of the fields to superfields. By the assumption
that the soft-terms respect all of the symmetries and constraints of the theory, this term must
exist in the superpotential. Next we want to map any term in the superpotential to a soft term
when the the superpotential term’s coupling is given an F -term vev. This is also straightforward
since the θ2 in the higher-component coupling creates a term in the Lagrangian in which all of
136
the non-spurion superfields are constrained to their lowest components. This term is manifestly a
part of the soft breaking Lagrangian.
Now we are reassured that we can really describe all soft-SUSY breaking terms by discussing
the higher-component vevs of the couplings, i.e. by promoting the couplings to spurion superfields.
As mentioned above, these spurions will be defined via the usual running (non-superfield) couplings
by promoting the dependence on the messenger threshold M to a superfield X. It is now useful
to discuss the usual notation employed to describe the higher component vevs of these objects. If
f (M ) is a non-superfield analytic function of the scale M , we may promote M → X = M + θ2 F
so that f obtains a higher component vev given by Taylor expansion in θ2 F/M :
( ( )) ∂f (M )
f (⟨X⟩) = f M 1 + θ2 F/M = f (M ) + θ2 F,
∂X
where we note that the expression on the right-hand side is exact since θ4 = 0. We can also write
the F -term using partial derivatives with respect to logarithms of superfields,
∂f (M ) ∂ ln f (M ) ∂ ln f (M ) F
f (⟨X⟩)|θ2 = F = f (M )F = f (M ) . (E.2)
∂X ∂X ∂ ln X M
Finally, it is worth noting that the meaning of a logarithm of a superfield is given by its Taylor
expansion,
F
ln X = ln(M + θ2 F ) = ln M + θ2 ,
M
which again terminates and is thus exact.
Now we are ready to derive our main results from analytic continuation into superspace. The
discussion in this section should prepare you to compare all of our derivations to the results in
the original literature. Let us emphasize that the following results depend on the assumption of
gauge mediation as the only source of SUSY-breaking. They are invalidated if there are other
contributions to the soft terms of non-negligible strength. Further, the results that we obtain will
assume the F ≪ M 2 limit.
E.3 Gaugino mass

Let’s start by determining the gauge mediation prediction for the gaugino mass soft-term. As
mentioned above, the real ‘trick’ is to use a result that we already know: the renormalization
group equations for the physical gauge coupling. The running coupling depends on the messenger
sector via the threshold at µ = M where the messenger fields are integrated out. The RG flow is
shown in Fig. 3. In the limit F/M 2 ≪ 1, we may neglect the threshold effects of supersymmetry
breaking in our renormalization. In other words, when the SUSY-breaking scale is low (e.g. just
above the√TeV scale), we can (1) neglect the non-supersymmetric renormalization group flow
between F and the electroweak √ scale and consider only the manifestly supersymmetric flow
from the cutoff scale down to F and (2) we can neglect the non-supersymmetric effects when
integrating over thresholds (when we integrate out the gauginos).
137
Λ, UV scale
N messengers SUSY RGE, b′
M , messenger scale
no messengers SUSY RGE, b
√
F,
SUSY scale
.
Figure 3: The renormalization group evolution of a gauge-mediated model with F ≪ M 2 .
The renormalization group equation for the coupling g at a scale below M can be integrated
to yield
1 1 b′ M b µ
2
= 2
+ 2
ln + 2 ln . (E.3)
g (µ) g (Λ) 8π Λ 8π M
In terms of the holomorphic coupling, this is written as
b′ M b µ
τ (µ) = τ (Λ) + i
ln + i ln . (E.4)
2π Λ 2π M
One of the nice results of SU (Nc ) super-Yang-Mills theories is that the beta function is written
simply in terms of the number of superfields transforming in the fundamental64 ,
b0 = 3Nc − Nf .
Thus we know that the difference in the beta functions is precisely the number of messenger fields
n at the scale M ,
b − b′ = n.
The expression for τ (µ) depends on the messenger scale M . We can ask ourselves where the scale
M comes from. In minimal gauge mediation, for example, we know that it is the vev of lowest
component of the SUSY-breaking field X, e.g. Eq. (16.7),
⟨X⟩ = M + θ2 F.
The trick behind analytic continuation into superspace is to promote M back to the superfield
from whence it originated. This, in turn, promotes τ into a chiral superfield,
b′ X b µ
τ (µ) = τ (Λ) + i ln + i ln (E.5)
2π Λ 2π X
b′ − b
=i ln X + · · · .
2π
∑
64
The fancy way of writing this for a group SU (Nc ) and representation r is b0 = 3C2 (Nc ) − i C(ri ), where
C2 1 = (ta ta )r and C(r)δ ab = Trr (ta tb ). Am I the only one who forgets these things?
138
We know from the form of LSYM in Eq. (A.51) that the soft term corresponding to the gaugino
mass can be written as
τ
m̂λ = −2 . (E.6)
16πi θ2
where the factor of 2 comes from the 1/2 in front of the gaugino mass in the soft breaking
Lagrangian, Eq. (E.1). [Comment: I’m not sure where the minus sign comes from since W ∼ iλ,
thus WW ∼= −λ2 already.] We’ve labelled m̂λ with a hat to indicate that it is not yet canonically
normalized. Recall that we’ve written our gauge Lagrangian with the ‘natural’ normalization in
which the kinetic term has an overall factor of g −2 = τ /4πi. Upon canonical normalization the
gaugino mass takes the form
g2 1
mλ = −2 τ (X)|θ2 = − τ (X)|θ2 .
16πi 2τ
Note that canonically normalizing cancels any arbitrariness in how we defined τ relative to g −2
so that this equation is correct no matter what prefactor multiplies τ ∼ g −2 . Let’s now use the
grown-up notation Eq. (E.2) and the expression Eq. (E.5) to write this more elegantly,

1 ∂ ln τ F i b′ − b F ng 2 F α F
mλ = − = − = 2
=n . (E.7)
2 ∂ ln X X=M M 2τ 2π M 16π M 4π M
Lo and behold we get exactly Eq. (16.11), the leading order contribution in the SUSY-breaking
parameter F/M 2 . Take a moment and bask in the glory of what we’ve done: we’ve reproduced
the leading order contribution to what would otherwise have been a two-loop calculation. Armed
with the one-loop exact beta function for the gauge coupling, we didn’t even have to calculate any
loops.
Before moving on to the other soft terms, let us make the following emphatic caveat: this trick
is only valid in the limit F ≪ M 2 . We relied on the assumption that SUSY-breaking effects were
small as we went through renormalization group thresholds. For example, we did not pick up the
logarithms in the full loop calculation for mλ in Eq. (16.10) nor would we pick up the dilogarithms
in the full two-loop calculation for the scalar masses.
E.4 Analytic Continuation: Wess-Zumino soft terms

Let’s proceed the formulae for the Wess-Zumino (i.e. superpotential) soft terms. Thanks to
Seiberg know that the parameters of the superpotential don’t renormalize [139], but we do know
that we have wavefunction renormalization and that this affects the physical couplings. We would
like to identify the M dependence of Z(M ) and then promote it to X dependence. There is an
immediate subtlety: Z is not a holomorphic quantity, but √ a real function. Thus it has to be a
†
function of both X and X , in particular, Z = Z(M ) = Z( XX † ). The one-loop expansion for
Z coming from the Wess-Zumino model takes the form

Λ
Z(µ) = 1 + |λ| ln .
2
(E.8)
µ
139
However, this is not the quantity that we want to calculate since to this order it doesn’t involve the
messengers which only couple via gauge interactions, and hence it is manifestly supersymmetric.
What we want is the wavefunction renormalization from gauge interactions, which is succinctly
written in the RGE
d ln Z C2 (r)
= α(µ).
d ln µ π
We’ve already calculated α(µ) = iτ −1 in Eq. (E.5), so that
b0 M2
α−1 (µ) = α−1 (Λ) + ln 2 .
4π Λ
We can then integrated the RGE taking into account the threshold at M ,
[ ]2c/b′ [ ]2c/b
α(Λ) α(M )
Z(Λ, M, µ) = Z(Λ) . (E.9)
α(M ) α(µ)
√
So we seem to be well on our way to performing analytic continuation, we just have to plop X † X
everywhere we see M . Not so fast. We should not forget to canonically normalize our fields with
respect to Z. We can go ahead and write
∫ ( )
2 2
Lkin = d4 θ Z + FZ θ2 + FZ∗ θ + DZ θ2 θ Φ† Φ
∫ ( )
∂Z ∂Z ∗ 2 ∂ 2Z ∗ 2 2

4
= d θ Z+ 2
Fθ + F θ + F F θ θ Φ† Φ.
∂X ∂X † ∂X∂X † X=M
In the second line we just wrote FZ in terms of the F -terms of the spurion X. We can canonically
2
normalize up to order O(θ2 , θ ) by redefining our fields
( )
′ ∂ ln Z 2
Φ→Φ =Z 1/2
1+ Fθ Φ.
∂X X=M
From now on we drop the prime on the field, Φ′ → Φ. When we need to we’ll refer to the original,
non-canonically normalized superfield as Φ0 . I know, we’re being excessively pedantic, but I’m
easily confused. Our normalization doesn’t get rid of the D-term, so that the kinetic term now
looks like
∫ [ ( ) ]
∂ ln Z ∂ ln Z 1 ∂ 2Z
∗ 2 2
Lkin = d θ 1 −
4
†
− †
FF θ θ Φ† Φ. (E.10)
∂X ∂X Z ∂X∂X X=M
2
e 2,
The θ2 θ is precisely a scalar mass term, m

∂ 2 ln Z FF∗
e =−
m 2 . (E.11)
∂ ln X∂ ln X † X=M M 2
But wait, there’s more! If we go back to the superpotential and plug in our rescaled field Φ, we
get A and B terms ‘for free.’ Of course, we expected this since we know that the only running of
140
the terms in the physical superpotential terms comes from wavefunction renormalization. Let’s
see how this works. The superpotential was written in terms of the non-canonically normalized
field,
( ( ) )
−1/2 ∂ ln Z F 2
W (Φ0 ) = W Z 1− θ Φ .
∂ ln X M
We want to isolate the soft terms that appear when one of the non-canonically normalized fields
picks up the F θ2 . We will write this down by taking a derivative of W with respect to the
non-canonically normalized field and multiply by the F θ2 term,
( )
∂W −1/2 ∂ ln Z F
∆Lsoft = Z − .
∂Φ0 Φ0 =ϕ0 ∂ ln X M
Thus to leading order (Z = 1) we obtain
∂ ln Z F
A = 3λ . (E.12)
∂ ln X M
Thus we have the useful result that the A terms will be suppressed by the Yukawa coupling times
powers of F/M and will thus be small.
We could proceed to plug in our simple Wess-Zumino superpotential to extract the exact form
of the B terms, but we’ll stop here since we now that B terms are a sensitive subject in gauge
mediation since it needn’t be generated by loops of the messenger fields. In other words, B (or
‘Bµ ’ in the Standard Model), is a hard parameter.
E.5 Remarks
Now that we’ve established our main results and demonstrated our method, let’s make a few
important remarks.
First of all, we might ask what we can do to incorporate higher orders in the messenger loops?
Before analytic continuation into superspace, one would have to calculate two loop diagrams for
the gaugino masses and three loop diagrams for the scalar masses. Patrick Meade remarks, “Now
I’ve never calculated a three loop diagram; maybe some of you have, but it sounds hard.” Just as
we were able to capture the one and two loop effects using well known RG equations at one loop
order, we may calculate the two and three loop effects by using the RG equations at two loop
order. There are subtleties when we go to higher loops due to the higher-loop evolution of τ . In
any practical renormalization scheme (e.g. DRED), τ loses its holomorphicity at two-loop order.
This is precisely due to the dilogarithms (and n-logarithms at higher orders) that we saw in the
full two-loop formula for the soft masses in Section 16.2.3. We thus can no longer simply promote
M → X in our analytic continuation. Giudice and Rattazzi teamed up with Arkani-Hamed in a
follow-up paper that shows how to tip-toe through these subtleties for to analytically continue in
superspace to all orders in perturbation theory [134].
Note that we are always stuck in the F ≪ M 2 limit, no matter how many messenger loops
we include. The threshold effects that we throw away in this limit are functions of logarithms
and dilogarithms presented in Section 16.2.3; we will never obtain such functions using analytic
141
continuation. Just how small does F have to be relative to M 2 ? Giudice and Rattazzi found that
this approximation is still very good for F/M 2 ∼ 0.3 [133]. They note that this is true because
the actual expansion parameter is F 2 /M 4 .
Next let use make some general remarks about the wavefunction renormalization spurion su-
perfield Z(X, X † , µ) following the discussion by Giudice and Rattazzi around their equation (16).
They remark that this superfield is a power series in logarithms of the form
( ) ( )
LΛ = ln µ2 /Λ2 LX = ln µ2 /XX † .
Thus to ℓ-loop order we may write
Z(X, X † , µ) = αℓ−1 (Λ) Pℓ (α(Λ)LX , α(Λ)LΛ ), (E.13)
where Pℓ is a function that comes from integrating the ℓ-loop RG equation. This means, for
example, that the scalar mass in Eq. (E.11) takes the form
e 2 = αℓ+1 (µ)Peℓ (α(µ)LX ) ,

m (E.14)
for Pe related to the second derivative of P . Thus we can see explicitly that it is sufficient to
consider the ℓ = 1 loop result to obtain O(α2 ) contributions to the soft scalar masses.
Moving on, we expressed our wavefunction renormalization in terms of α, which we related to
the renormalization of τ from Eq. (E.5). It is important to recognize, however, that the Z spurion
is a real superfield while the τ spurion is a chiral superfield. Thus the proper identification is
b′ XX †
α−1 (X) = Im (τ ) = α−1 (Λ) + ln 2 . (E.15)
2π Λ
[Check: There should also be a b’ term.] With this we can write out more explicit forms of our
scalar mass and A term by plugging into Eq. (E.9),
α2 (µ) [ 2 n ] ( F )2
e (µ) = 2C2
m 2
n ξ + (1 − ξ )
2
(E.16)
16π 2 b M
Ci α(µ) F
Ai (µ) = 2 n(ξ − 1) , (E.17)
b 4π M
where
( )−1
α(M ) b M
ξ≡ = 1+ α(µ) ln . (E.18)
α(µ) 2π µ
If the superfield Φ is charged under multiple gauge groups, then the appropriate generalization is
to sum over the contributions from the different gauge couplings. Note that we’ve even explicitly
included the leading log effect from fthe renormalization from M down to µ: Ai = 0 at µ = M ,
but at low energies acquires a renormalization proportional to the gaugino mass.
Now let us close by reminding ourselves of something to be happy about: we have been able
to determine the leading-order (in F/M 2 ) effect of supersymmetry breaking in the hidden sector
without having to calculate any loops and in a way that is by and large insensitive to the details
of how supersymmetry is broken in the hidden sector. We should be very proud of ourselves.
142
References
[1] K. A. Intriligator and N. Seiberg, “Lectures on Supersymmetry Breaking,” Class. Quant.
Grav. 24 (2007) S741–S772, arXiv:hep-ph/0702069.
[2] K. A. Intriligator and N. Seiberg, “Lectures on supersymmetric gauge theories and electric-
magnetic duality,” Nucl. Phys. Proc. Suppl. 45BC (1996) 1–28, arXiv:hep-th/9509066.
[3] M. J. Strassler, “An unorthodox introduction to supersymmetric gauge theory,”

arXiv:hep-th/0309149.
[4] M. J. Strassler, “The duality cascade,” arXiv:hep-th/0505153.
[5] J. Terning, Modern Supersymmetry: Dynamics and Duality. Oxford University Press,
USA, 2006.
[6] S. Krippendorf, F. Quevedo, and O. Schlotterer, “Cambridge Lectures on Supersymmetry

and Extra Dimensions,” arXiv:1011.1491 [hep-th].
[7] M. Dine, “Supersymmetry Breaking at Low Energies,” Nucl. Phys. Proc. Suppl. 192-193
(2009) 40–60, arXiv:0901.1713 [hep-ph].
[8] M. Dine and J. D. Mason, “Supersymmetry and Its Dynamical Breaking,”

arXiv:1012.2836 [hep-th].
[9] Y. Shirman, “TASI 2008 Lectures: Introduction to Supersymmetry and Supersymmetry

Breaking,” arXiv:0907.0039 [hep-ph].
[10] M. Dine, “Supersymmetry and string theory: Beyond the standard model,”. Cambridge,
UK: Cambridge Univ. Pr. (2007) 515 p.
[11] M. E. Peskin, “Duality in supersymmetric Yang-Mills theory,” arXiv:hep-th/9702094.
[12] M. A. Shifman and A. I. Vainshtein, “Solution of the Anomaly Puzzle in SUSY Gauge
Theories and the Wilson Operator Expansion,” Nucl. Phys. B277 (1986) 456.
[13] C. P. Burgess, “Introduction to effective field theory,” Ann. Rev. Nucl. Part. Sci. 57
(2007) 329–362, arXiv:hep-th/0701053.
[14] T. Banks, Modern Quantum Field Theory: A Concise Introduction. Cambridge University
Press, 2008.
[15] S. P. Martin, “Generalized messengers of supersymmetry breaking and the sparticle mass
spectrum,” Phys. Rev. D55 (1997) 3177–3187, arXiv:hep-ph/9608224.
[16] P. Argyres, “N=1 d=4 global supersymmetry.” Available on the author’s website., 1996,
2001.
[17] A. Bilal, “Introduction to supersymmetry,” arXiv:hep-th/0101055.
143
[18] M. A. Luty and W. Taylor, “Varieties of vacua in classical supersymmetric gauge
theories,” Phys. Rev. D53 (1996) 3399–3405, arXiv:hep-th/9506098.
[19] D. Mumford, Geometric invariant theory. Springer, Berlin [u.a.], 3. enlarged ed. ; 3.
print. ed., 2002.
[20] L. Alvarez-Gaume and S. F. Hassan, “Introduction to S-duality in N = 2 supersymmetric
gauge theories: A pedagogical review of the work of Seiberg and Witten,” Fortsch. Phys.
45 (1997) 159–236, arXiv:hep-th/9701069.
[21] W. Lerche, “Introduction to Seiberg-Witten theory and its stringy origin,” Nucl. Phys.
Proc. Suppl. 55B (1997) 83–117, arXiv:hep-th/9611190.
[22] A. Bilal, “Duality in N=2 SUSY SU(2) Yang-Mills Theory: A pedagogical introduction to
the work of Seiberg and Witten,” arXiv:hep-th/9601007.
[23] M. A. Shifman and A. I. Vainshtein, “Instantons versus supersymmetry: Fifteen years
later,” arXiv:hep-th/9902018.
[24] M. Bianchi, S. Kovacs, and G. Rossi, “Instantons and supersymmetry,” Lect. Notes Phys.
737 (2008) 303–470, arXiv:hep-th/0703142.
[25] S. Coleman, Aspects of Symmetry. Feb., 1988.
[26] S. Vandoren and P. van Nieuwenhuizen, “Lectures on instantons,” arXiv:0802.1862
[hep-th].
[27] N. Manton and P. Sutcliffe, Topological Solitons. Cambridge University Press, 2004.
[28] R. Rajaraman, SOLITONS AND INSTANTONS. AN INTRODUCTION TO SOLITONS
AND INSTANTONS IN QUANTUM FIELD THEORY. Amsterdam, Netherlands:
North-holland ( 1982) 409p.
[29] S. Minwalla, “Restrictions imposed by superconformal invariance on quantum field
theories,” Adv. Theor. Math. Phys. 2 (1998) 781–846, arXiv:hep-th/9712074.
[30] V. A. Novikov, M. A. Shifman, A. I. Vainshtein, and V. I. Zakharov, “Supersymmetric
instanton calculus: Gauge theories with matter,” Nucl. Phys. B260 (1985) 157–181.
[31] V. A. Novikov, M. A. Shifman, A. I. Vainshtein, and V. I. Zakharov, “Beta Function in
Supersymmetric Gauge Theories: Instantons Versus Traditional Approach,” Phys. Lett.
B166 (1986) 329–333.
[32] N. Arkani-Hamed and H. Murayama, “Holomorphy, rescaling anomalies and exact beta
functions in supersymmetric gauge theories,” JHEP 06 (2000) 030,
[33] N. Arkani-Hamed and H. Murayama, “Renormalization group invariance of exact results in
supersymmetric gauge theories,” Phys. Rev. D57 (1998) 6638–6648,
144
[34] I. Affleck, M. Dine, and N. Seiberg, “Dynamical Supersymmetry Breaking in
Supersymmetric QCD,” Nucl. Phys. B241 (1984) 493–534.
[35] I. Affleck, M. Dine, and N. Seiberg, “Supersymmetry Breaking by Instantons,” Phys. Rev.
Lett. 51 (1983) 1026.
[36] D. Finnell and P. Pouliot, “Instanton calculations versus exact results in four- dimensional
SUSY gauge theories,” Nucl. Phys. B453 (1995) 225–239, arXiv:hep-th/9503115.
[37] W. Siegel, “Supersymmetric Dimensional Regularization via Dimensional Reduction,”
Phys. Lett. B84 (1979) 193.
[38] A. C. Davis, M. Dine, and N. Seiberg, “THE MASSLESS LIMIT OF
SUPERSYMMETRIC QCD,” Phys. Lett. B125 (1983) 487.
[39] E. Witten, “Global Aspects of Current Algebra,” Nucl. Phys. B223 (1983) 422–432.
[40] P. Ramond, Field Theory : A Modern Primer. Frontiers in Physics Series, Vol 74.
Westview Press, 2nd edition ed., 2001.
[41] M. E. Peskin and D. V. Schroeder, “An Introduction to quantum field theory,”. Reading,
USA: Addison-Wesley (1995) 842 p.
[42] K. A. Intriligator, “’Integrating in’ and exact superpotentials in 4-d,” Phys. Lett. B336
(1994) 409–414, arXiv:hep-th/9407106.
[43] D. S. Berman and E. Rabinovici, “Supersymmetric gauge theories,”
[44] K. A. Intriligator, R. G. Leigh, and N. Seiberg, “Exact superpotentials in
four-dimensions,” Phys. Rev. D50 (1994) 1092–1104, arXiv:hep-th/9403198.
[45] G. Veneziano and S. Yankielowicz, “An Effective Lagrangian for the Pure N=1
Supersymmetric Yang-Mills Theory,” Phys. Lett. B113 (1982) 231.
[46] T. R. Taylor, G. Veneziano, and S. Yankielowicz, “Supersymmetric QCD and Its Massless
Limit: An Effective Lagrangian Analysis,” Nucl. Phys. B218 (1983) 493.
[47] G. ’t Hooft, (ed. ) et al., “Recent Developments in Gauge Theories. Proceedings, Nato
Advanced Study Institute, Cargese, France, August 26 - September 8, 1979,” NATO Adv.
Study Inst. Ser. B Phys. 59 (1980) 1–438.
[48] Y. Frishman, A. Schwimmer, T. Banks, and S. Yankielowicz, “The Axial Anomaly and the
Bound State Spectrum in Confining Theories,” Nucl. Phys. B177 (1981) 157.
[49] M. E. Peskin, “The Alignment of the Vacuum in Theories of Technicolor,” Nucl. Phys.
B175 (1980) 197–233.
[50] C. Csaki, “The minimal supersymmetric standard model (MSSM),” Mod. Phys. Lett. A11
(1996) 599, arXiv:hep-ph/9606414.
145
[51] C. Csaki, M. Schmaltz, and W. Skiba, “Confinement in N = 1 SUSY gauge theories and
model building tools,” Phys. Rev. D55 (1997) 7840–7858, arXiv:hep-th/9612207.
[52] N. Seiberg, “Exact results on the space of vacua of four-dimensional SUSY gauge
theories,” Phys. Rev. D49 (1994) 6857–6863, arXiv:hep-th/9402044.
[53] O. Aharony, A. Hanany, K. A. Intriligator, N. Seiberg, and M. J. Strassler, “Aspects of N

= 2 supersymmetric gauge theories in three dimensions,” Nucl. Phys. B499 (1997) 67–99,
[54] T. Banks and A. Zaks, “On the Phase Structure of Vector-Like Gauge Theories with
Massless Fermions,” Nucl. Phys. B196 (1982) 189.
[55] C. Csaki, A. Falkowski, Y. Nomura, and T. Volansky, “New Approach to the mu-Bmu
Problem of Gauge-Mediated Supersymmetry Breaking,” Phys. Rev. Lett. 102 (2009)
111801, arXiv:0809.4492 [hep-ph].
[56] N. Seiberg, “Electric - magnetic duality in supersymmetric nonAbelian gauge theories,”

Nucl. Phys. B435 (1995) 129–146, arXiv:hep-th/9411149.
[57] C. Csaki, Y. Shirman, and J. Terning, “A Seiberg Dual for the MSSM: Partially
Composite W and Z,” Phys.Rev. D84 (2011) 095011, arXiv:1106.3074 [hep-ph].
[58] Z. Komargodski, “Vector Mesons and an Interpretation of Seiberg Duality,” JHEP 02

(2011) 019, arXiv:1010.4105 [hep-th].
[59] J. Donoghue, E. Golowich, and B. Holstein, Dynamics of the Standard Model. Cambridge
University Press, 1994.
[60] L.-F. L. Ta-Pei Cheng, Gauge Theory of Elementary Particle Physics: Problems and
Solutions. Oxford University Press, 2000.
[61] R. Haag, “Quantum field theories with composite particles and asymptotic conditions,”
Phys. Rev. 112 (1958) 669–673.
[62] C. G. Callan, Jr., S. R. Coleman, J. Wess, and B. Zumino, “Structure of phenomenological

Lagrangians. 2,” Phys. Rev. 177 (1969) 2247–2250.
[63] S. R. Coleman, J. Wess, and B. Zumino, “Structure of phenomenological Lagrangians. 1,”

Phys. Rev. 177 (1969) 2239–2247.
[64] H. Georgi, “NEW REALIZATION OF CHIRAL SYMMETRY,” Phys. Rev. Lett. 63

(1989) 1917–1919.
[65] H. Georgi, “Vector Realization of Chiral Symmetry,” Nucl. Phys. B331 (1990) 311–330.
[66] M. C. Birse, “Effective chiral Lagrangians for spin-1 mesons,” Z. Phys. A355 (1996)
231–246, arXiv:hep-ph/9603251.
146
[67] K. A. Intriligator, N. Seiberg, and D. Shih, “Dynamical SUSY breaking in meta-stable
vacua,” JHEP 04 (2006) 021, arXiv:hep-th/0602239.
[68] S. Abel and V. V. Khoze, “Direct Mediation, Duality and Unification,” JHEP 11 (2008)
024, arXiv:0809.5262 [hep-ph].
[69] S. Abel and V. V. Khoze, “Dual unified SU(5),” JHEP 01 (2010) 006, arXiv:0909.4105
[hep-ph].
[70] S. Abel and F. Brummer, “Holographic metastability,” JHEP 05 (2010) 070,

[71] S. Abel and T. Gherghetta, “A slice of AdS5 as the large N limit of Seiberg duality,”
[72] J. Gray, A. Hanany, Y.-H. He, V. Jejjala, and N. Mekareeya, “SQCD: A Geometric
Apercu,” JHEP 05 (2008) 099, arXiv:0803.4257 [hep-th].
[73] E. Witten, “Dynamical Breaking of Supersymmetry,” Nucl. Phys. B188 (1981) 513.
[74] E. Witten, “Constraints on Supersymmetry Breaking,” Nucl. Phys. B202 (1982) 253.
[75] A. E. Nelson and N. Seiberg, “R symmetry breaking versus supersymmetry breaking,”

Nucl. Phys. B416 (1994) 46–62, arXiv:hep-ph/9309299.
[76] J. Bagger, E. Poppitz, and L. Randall, “The R axion from dynamical supersymmetry
breaking,” Nucl. Phys. B426 (1994) 3–18, arXiv:hep-ph/9405345.
[77] M. Dine, A. E. Nelson, and Y. Shirman, “Low-energy dynamical supersymmetry breaking

simplified,” Phys. Rev. D51 (1995) 1362–1370, arXiv:hep-ph/9408384.
[78] T. Banks and L. J. Dixon, “Constraints on String Vacua with Space-Time

Supersymmetry,” Nucl. Phys. B307 (1988) 93–108.
[79] C. G. Callan, Jr. and S. R. Coleman, “The Fate of the False Vacuum. 2. First Quantum
Corrections,” Phys. Rev. D16 (1977) 1762–1768.
[80] S. R. Coleman, “The Fate of the False Vacuum. 1. Semiclassical Theory,” Phys. Rev. D15
(1977) 2929–2936.
[81] J. R. Ellis, C. H. Llewellyn Smith, and G. G. Ross, “WILL THE UNIVERSE BECOME
SUPERSYMMETRIC?,” Phys. Lett. B114 (1982) 227.
[82] S. Dimopoulos, G. R. Dvali, R. Rattazzi, and G. F. Giudice, “Dynamical soft terms with
unbroken supersymmetry,” Nucl. Phys. B510 (1998) 12–38, arXiv:hep-ph/9705307.
[83] Y. Ookouchi, “Light Gaugino Problem in Direct Gauge Mediation,” arXiv:1107.2622

[hep-th].
147
[84] Z. Komargodski and D. Shih, “Notes on SUSY and R-Symmetry Breaking in Wess-Zumino
Models,” JHEP 04 (2009) 093, arXiv:0902.0030 [hep-th].
[85] E. Poppitz and S. P. Trivedi, “Dynamical supersymmetry breaking,” Ann. Rev. Nucl.
Part. Sci. 48 (1998) 307–350, arXiv:hep-th/9803107.
[86] Y. Shadmi and Y. Shirman, “Dynamical supersymmetry breaking,” Rev. Mod. Phys. 72
(2000) 25–64, arXiv:hep-th/9907225.
[87] I. Affleck, M. Dine, and N. Seiberg, “Dynamical Supersymmetry Breaking in

Four-Dimensions and Its Phenomenological Implications,” Nucl. Phys. B256 (1985) 557.
[88] H. Murayama, “Studying Non-calculable Models of Dynamical Supersymmetry Breaking,”

Phys. Lett. B355 (1995) 187–192, arXiv:hep-th/9505082.
[89] G. F. Giudice and R. Rattazzi, “Theories with gauge-mediated supersymmetry breaking,”

Phys. Rept. 322 (1999) 419–499, arXiv:hep-ph/9801271.
[90] M. Dine and W. Fischler, “A Phenomenological Model of Particle Physics Based on

Supersymmetry,” Phys. Lett. B110 (1982) 227.
[91] C. R. Nappi and B. A. Ovrut, “Supersymmetric Extension of the SU(3) x SU(2) x U(1)
Model,” Phys. Lett. B113 (1982) 175.
[92] M. Dine and A. E. Nelson, “Dynamical supersymmetry breaking at low-energies,” Phys.

Rev. D48 (1993) 1277–1287, arXiv:hep-ph/9303230.
[93] P. Meade, N. Seiberg, and D. Shih, “General Gauge Mediation,” Prog. Theor. Phys. Suppl.
177 (2009) 143–158, arXiv:0801.3278 [hep-ph].
[94] S. Dimopoulos and G. F. Giudice, “Multi-messenger theories of gauge-mediated

supersymmetry breaking,” Phys. Lett. B393 (1997) 72–78, arXiv:hep-ph/9609344.
[95] C. Csaki, J. Heinonen, J. Hubisz, and Y. Shirman, “Odd Decays from Even Anomalies:
Gauge Mediation Signatures Without SUSY,” Phys. Rev. D79 (2009) 105016,
arXiv:0901.2933 [hep-ph].
[96] J. E. Kim and H. P. Nilles, “The mu Problem and the Strong CP Problem,” Phys. Lett.
B138 (1984) 150.
[97] R. D. Peccei and H. R. Quinn, “CP Conservation in the Presence of Instantons,” Phys.
Rev. Lett. 38 (1977) 1440–1443.
[98] G. F. Giudice and A. Masiero, “A Natural Solution to the mu Problem in Supergravity

Theories,” Phys. Lett. B206 (1988) 480–484.
[99] G. R. Dvali, G. F. Giudice, and A. Pomarol, “The µ-Problem in Theories with

Gauge-Mediated Supersymmetry Breaking,” Nucl. Phys. B478 (1996) 31–45,
arXiv:hep-ph/9603238.
148
[100] M. Maniatis, “The NMSSM reviewed,” arXiv:0906.0777 [hep-ph].
[101] T. S. Roy and M. Schmaltz, “A hidden solution to the mu/Bmu problem in gauge
mediation,” Phys. Rev. D77 (2008) 095008, arXiv:0708.3593 [hep-ph].
[102] E. Poppitz and S. P. Trivedi, “New models of gauge and gravity mediated supersymmetry
breaking,” Phys. Rev. D55 (1997) 5508–5519, arXiv:hep-ph/9609529.
[103] N. Arkani-Hamed, J. March-Russell, and H. Murayama, “Building models of

gauge-mediated supersymmetry breaking without a messenger sector,” Nucl. Phys. B509
(1998) 3–32, arXiv:hep-ph/9701286.
[104] H. Murayama, “A Model of direct gauge mediation,” Phys. Rev. Lett. 79 (1997) 18–21,
[105] H. Murayama and Y. Nomura, “Gauge mediation simplified,” Phys. Rev. Lett. 98 (2007)
151803, arXiv:hep-ph/0612186.
[106] R. Kitano, H. Ooguri, and Y. Ookouchi, “Direct mediation of meta-stable supersymmetry

breaking,” Phys. Rev. D75 (2007) 045022, arXiv:hep-ph/0612139.
[107] C. Csaki, Y. Shirman, and J. Terning, “A simple model of low-scale direct gauge
mediation,” JHEP 05 (2007) 099, arXiv:hep-ph/0612241.
[108] N. Seiberg, T. Volansky, and B. Wecht, “Semi-direct Gauge Mediation,” JHEP 11 (2008)
004, arXiv:0809.4437 [hep-ph].
[109] D. Marques, “Generalized messenger sector for gauge mediation of supersymmetry

breaking and the soft spectrum,” JHEP 03 (2009) 038, arXiv:0901.1326 [hep-ph].
[110] R. Argurio, M. Bertolini, G. Ferretti, and A. Mariotti, “Patterns of Soft Masses from
General Semi-Direct Gauge Mediation,” arXiv:0912.0743 [hep-ph].
[111] S. Ray, “Some properties of meta-stable supersymmetry-breaking vacua in Wess-Zumino

models,” Phys. Lett. B642 (2006) 137–141, arXiv:hep-th/0607172.
[112] K. Intriligator, D. Shih, and M. Sudano, “Surveying Pseudomoduli: the Good, the Bad
and the Incalculable,” JHEP 03 (2009) 106, arXiv:0809.3981 [hep-th].
[113] E. Witten, “Mass Hierarchies in Supersymmetric Theories,” Phys. Lett. B105 (1981) 267.
[114] C. Cheung, A. L. Fitzpatrick, and D. Shih, “(Extra)Ordinary Gauge Mediation,” JHEP 07

(2008) 054, arXiv:0710.3585 [hep-ph].
[115] H. Georgi, “LIE ALGEBRAS IN PARTICLE PHYSICS. FROM ISOSPIN TO UNIFIED

THEORIES,” Front. Phys. 54 (1982) 1–255.
[116] T. P. Cheng and L. F. Li, “Gauge Theory of Elementary Particle Physics,”. Oxford, Uk:
Clarendon ( 1984) 536 P. ( Oxford Science Publications).
149
[117] H. K. Dreiner, H. E. Haber, and S. P. Martin, “Two-component spinor techniques and
Feynman rules for quantum field theory and supersymmetry,” Phys. Rept. 494 (2010)
1–196, arXiv:0812.1594 [hep-ph].
[118] P. Binetruy, Supersymmetry: Theory, Experiment, and Cosmology. Oxford Graduate

Texts. Oxfor University Press, 2007.
[119] I. Aitchison, Supersymmetry in Particle Physics: An Elementary Introduction. Cambridge

University Press, 2007.
[120] P. B. Pal, “Dirac, Majorana and Weyl fermions,” arXiv:1006.1718 [hep-ph].
[121] J. Wess and J. Bagger, Supersymmetry and Supergravity. Princeton University Press, 2
revised ed., 1992.
[122] D. Bailin and A. Love, Supersymmetric Gauge Field Theory and String Theory. Taylor &
Francis, 1 ed., 1994.
[123] A. Giveon, A. Katz, and Z. Komargodski, “Uplifted Metastable Vacua and Gauge
Mediation in SQCD,” JHEP 07 (2009) 099, arXiv:0905.3387 [hep-th].
[124] S. R. Coleman and E. J. Weinberg, “Radiative Corrections as the Origin of Spontaneous

Symmetry Breaking,” Phys. Rev. D7 (1973) 1888–1910.
[125] A. Zee, “Quantum field theory in a nutshell,”. Princeton, UK: Princeton Univ. Pr. (2003)
518 p.
[126] M. Srednicki, “Quantum field theory,”. Cambridge, UK: Univ. Pr. (2007) 641 p.
[127] Y. Nambu, “S Matrix in semiclassical approximation,” Phys. Lett. B26 (1968) 626–629.
[128] W. Greiner and J. Reinhardt, “Field quantization,”. Berlin, Germany: Springer (1996) 440
p.
[129] H. Obsorn and S. Gielen, “Advanced quantum field theory, lecture notes..”
http://www.damtp.cam.ac.uk/user/ho/Notes.pdf.
[130] R. Jackiw, “Functional evaluation of the effective potential,” Phys. Rev. D9 (1974) 1686.
[131] K. van den Broek, “Vscape V1.1.0: An interactive tool for metastable vacua,” Comput.
Phys. Commun. 178 (2008) 52–72, arXiv:0705.2019 [hep-ph].
[132] J. Preskill, “Gauge anomalies in an effective field theory,” Annals Phys. 210 (1991)
323–379.
[133] G. F. Giudice and R. Rattazzi, “Extracting Supersymmetry-Breaking Effects from Wave-

Function Renormalization,” Nucl. Phys. B511 (1998) 25–44, arXiv:hep-ph/9706540.
150
[134] N. Arkani-Hamed, G. F. Giudice, M. A. Luty, and R. Rattazzi, “Supersymmetry-breaking
loops from analytic continuation into superspace,” Phys. Rev. D58 (1998) 115005,
[135] J. D. Wells, “Lectures on higgs boson physics in the standard model and beyond,”
0909.4541v1. http://arxiv.org/abs/0909.4541v1.
[136] R. Barbieri, A. Pomarol, R. Rattazzi, and A. Strumia, “Electroweak symmetry breaking

after LEP-1 and LEP-2,” Nucl. Phys. B703 (2004) 127–146, arXiv:hep-ph/0405040.
[137] Z. Han and W. Skiba, “Effective theory analysis of precision electroweak data,” Phys. Rev.
D71 (2005) 075009, arXiv:hep-ph/0412166.
[138] G. Cacciapaglia, C. Csaki, G. Marandella, and A. Strumia, “The minimal set of

electroweak precision parameters,” Phys. Rev. D74 (2006) 033011,
[139] N. Seiberg, “Naturalness Versus Supersymmetric Non-renormalization Theorems,” Phys.

Lett. B318 (1993) 469–475, arXiv:hep-ph/9309335.
151

Seiber Go Logy

Uploaded by

Copyright:

Available Formats

Seiber Go Logy

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Seiber Go Logy

Uploaded by

Copyright:

Available Formats

Notes on Seibergology

EM-like duality SUSY and applications

Laboratory of Elementary Particle Physics,

This version: May 7, 2013

4 F = N and F = N + 1: Some special cases 40

5 F > N + 1: Seiberg Duality 50

6 The phase structure of SQCD 64

7 Siblings and Cousins of Seiberg Duality 64

8 Crouching Seiberg, Hidden Gauge Group 64

10 Some current directions of interest 66

11 Unsorted ISS and SUSY breaking notes 66

15 Dynamical Supersymmetry Breaking 72

16 Gauge Mediated SUSY Breaking 75

17 The Nelson-Seiberg R-symmetry theorem 86

18 The Komargodski-Shih Gaugino Mass Theorem 88

19 The Landau Pole Problem 95

20 ISS: Metastable SUSY Breaking 95

B The Coleman-Weinberg Effective Potential 114

C Phases of Gauge Theories 130

D Review of Anomaly Cancellation 130

E Analytic Continuation into Superspace 134

1.2 A selected non-technical history

2.1 Effective Actions

2.2 Gauge theory facts

The fundamental covariant derivative Dµ is Dµ = ∂µ +ieAµ , so that Dµ ϕ transforms in the same

d(□) = N d(Ad) = N 2 − 1 (2.7)

2.3 Reminder: basic SUSY gauge theory facts

Gauge transformations of a vector superfield. It is worth remarking that even through

2.4.1 Example: U (1)

e such that V is minimized. Modding out

V (ϕi ) = Wj̄† (K −1 )ij̄ Wi . (2.21)

D = diag(v12 , v22 , · · · , vF2 , 0, · · · , 0).

This spontaneously breaks SU (N ) → SU (N − F ). We observe the super Higgs mechanism at

2.5 The holomorphic gauge coupling

We can redefine W by absorbing the coupling into the vector superfield,

The canonically normalized Lagrangian is trivially obtained by multiplying through by g 2 . Since

b = 3Neff − Feff , (2.38)

where have defined the holomorphic dynamical scale

2.6 The NSVZ β-function

Lkin = Z |∂ϕ|2 + Z ψ̄i∂ψ,

Why is it called an anomalous dimension? In a general (not necessarily supersymmetric)

so that the Lagrangian for canonically-normalized fields takes the form

This tells us that the canonically normalized gauge coupling is

2.7 The Konishi Anomaly

Note, however, that the Konishi anomaly also gives us a term

Thus, at the scale µ,

2.8 Symmetries of SQCD

3 F < N : the ADS superpotential

3.1 Holomorphic scale as a spurion

3.2 The ADS Superpotential

SU (N ) SU (F ) SU (F ) U (1)1 U (1)2 U (1)R

R[fermion] = R[boson] − 1, (3.7)

Invariance under the U (1) symmetries forces additional factors,

More generally: We could have also written a term

W = Λbn (WW)m (det M )p (3.12)

3.3 ADS: F = N − 1, instantons

and so we can write our instanton-background Lagrangian as

L ∼ v 2N QQΛ2N +1 (ρ)2N (3.15)

3.4 Deforming SQCD: Higgsing a squark

b = 3N − F ⇒ bL = 3(N − 1) − (F − 1), (3.21)