a

r
X
i
v
:
1
3
0
5
.
1
6
3
3
v
2


[
h
e
p
-
t
h
]


2
0

M
a
y

2
0
1
3
UCSD-PTH-12-17
Aspects of 3d A = 2 Chern-Simons-Matter Theories
Kenneth Intriligator
1
and Nathan Seiberg
2
1
Department of Physics, University of California, San Diego, La Jolla, CA 92093 USA
2
School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540 USA
We comment on various aspects of the the dynamics of 3d A = 2 Chern-Simons gauge
theories and their possible phases. Depending on the matter content, real masses and
FI parameters, there can be non-compact Higgs or Coulomb branches, compact Higgs or
Coulomb branches, and isolated vacua. We compute the Witten index of the theories, and
show that it does not change when the system undergoes a phase transition. We study
aspects of monopole operators and solitons in these theories, and clarify subtleties in the
soliton collective coordinate quantization. We show that solitons are compatible with a
mirror symmetry exchange of Higgs and Coulomb branches, with BPS solitons on one
branch related to the modulus of the other. Among other results, we show how to derive
Aharony duality from Giveon-Kutasov duality.
May 2013
1. Introduction
The use of holomorphy has shed light on the dynamics of A = 1 gauge theories in four
dimensions [1], giving insights into the dynamics and the role of electric magnetic duality.
Following the original papers [2-4], there has been progress in nding a similar picture in
3d A = 2 theories (and interconnections with string theory), see e.g. [5-11], and many
others. The 3d theories exhibit several interesting elements that are not present in 4d:
1. Abelian gauge groups can have non-trivial IR dynamics.
2. There are real parameters: masses and FI terms
1
, which do not reside in (background)
chiral superelds. So the power of holomorphy does not help to control them.
3. Chern-Simons parameters; they are quantized and cannot be continuously varied.
4. Coulomb branches, associated with expectation values of vector superelds; the chiral
operators thus include monopole operators [2,3,5,15,16].
The presence of real parameters has deep consequences: it allows for interesting phase
transitions. This diers from 4d SUSY theories, where all the parameters are background
chiral superelds, so complex. Then, if supersymmetry is unbroken, there are no phase
transitions [17,1]. There are no rst order transitions, since the vacuum energy is zero,
and any second or higher order transitions are associated with the expectation value of
a chiral supereld order parameter. Such regions are subspaces of complex codimension,
e.g. a point in the complex plane that can be in a new phase (e.g. interacting SCFT vs.
IR free). Since it is always possible to move around such regions, there cannot be walls
separating dierent IR phase regions in 4d. But there can be in 3d.
We will here encounter the following phases
2
1. Higgs. This includes non-compact or compact moduli spaces, or isolated points.
2. Coulomb. This includes both non-compact and compact (as in [6]) possibilities.
3. Topological: isolated vacua of theories with Chern-Simons terms k
eff
,= 0.
Most of our examples will be based on U(1) and SU(2) gauge theories, but they can be
easily generalized to more complicated gauge groups.
1
Real Fayet-Iliopoulos terms are also present in 4d. But since they are associated with U(1)
gauge theories, which are not asymptotically free, they cannot be present in the UV Lagrangian.
Moreover, FI terms cannot be generated by the dynamics [12-14] and hence cannot be present in
the IR Lagrangian. Therefore, their impact on the dynamics is less interesting than in 3d.
2
There is also the conning phase, as in [18], for 3d theories with N 1 SUSY; see [19] for
discussion of the conning phase in N = 2 SQED.
1
We will discuss, compare, and make distinctions between, several dierent notions:
1. The supersymmetric vacua and their moduli.
2. Vortices. (Particles, localized at points in space.)
3. Monopole operators. (Localized at points in spacetime.)
These notions are inter-related. The Coulomb branch of say a U(1) gauge theory can be
labeled by expectation values of a chiral supereld X. On the Higgs branch, there can
be vortex particle states, which can be BPS for non-zero FI term. In [3] it was suggested
that the Coulomb branch X can be related to a condensate of massless vortices. This
is motivated by mirror symmetry [20], which exchanges the Higgs and Coulomb branches
of dual theories: the vortices can be the quanta of dual, charged matter. More generally,
we can consider monopole operators, which are UV operators, independent of the vacuum,
which insert a ux unit at a point in spacetime. The above notions will be discussed, and
distinguished, extensively in this work.
1.1. Moduli spaces of vacua
The theories in at spacetime have a rich variety of possible vacua. For generic real
parameters, 3d A = 2 theories have isolated vacua, with a mass gap. In some cases,
supersymmetry is spontaneously and dynamically broken.
In general, upon tuning the real parameters to special values, there can be moduli
spaces of supersymmetric vacua, which can be non-compact or compact. Consider, for
example, a U(1) gauge theory, with some charged matter content. The gauge eld strength
and superpartners can be written in terms of a real, linear multiplet

i
2

V = +. . . +
1
2

2
J
; (1.1)
where is the real scalar in the vector multiplet and . . . includes the photino terms. Here

J
is the current supermultiplet for the conserved, U(1)
J
global symmetry, with charge
q
J
=
_
d
2
xj
0
J
=
_
F
2
= c
1
(F) Z. (1.2)
The linear multiplet (1.1) satises D
2
= D
2
= 0.
When there is a Coulomb branch, the elds in (1.1) are massless, and the real linear
multiplet (1.1) can be dualized to a chiral supereld X, with D

X = 0 or, as an operator
statement, [Q

, X] = 0. The Coulomb modulus X carries q

J
charge 1, so X , = 0
spontaneously breaks U(1)
J
. The eld X can also acquire charges under the other global
symmetries, as can be seen in terms of one-loop induced Chern-Simons terms [3].
2
1.2. Monopole operators
A monopole operator changes the boundary conditions of the gauge eld, inserting
a unit of q
J
charge at a point in spacetime. It is a disorder operator, like the t Hooft
operator in 4d: it is most naturally dened by how it aects the elds in the functional
integral; see e.g. [21,22] for analogous discussion for line and surface operators. As a
UV operator, the denition of the monopole operator is independent of the IR choice of
vacuum. Nevertheless, the monopole operator is not-unrelated to the Coulomb modulus
X, which labels the vacuum, and which also carries q
J
charge. Also, there is a connection
to vortices. When the monopole operator acts on the vacuum, it creates particle states
with q
J
charge, like the vortices discussed above.
It had been anticipated [3] that the Coulomb branch moduli can be given a mirror
dual [20] interpretation as the Higgsing moduli associated with a condensate of vortices.
This notion was made much more precise in the works [5,15,16]. The latter works focus
on the theory on a spatial sphere S
2
, which eliminates some technical challenges. The
physics is quite dierent between the at space theory and the theory on a sphere, and we
will here discuss aspects of the monopole operators in non-compact spacetime, where it is
interesting to understand the connection with the Coulomb branch moduli space of vacua
(which are only superselection sectors on non-compact spacetime).
Inserting a monopole operator of charge q
J
at a spacetime point x
0
introduces a
magnetic source there, with ux q
J
on any (Euclidean) S
2
surrounding x
0
, as in (1.2). For
a chiral monopole operator, this leads to
D
2
= 0, D
2
= q
J
2
(3)
(x x
0
)
2
, (1.3)
(with the roles of D
2
and D
2
reversed for the anti-chiral monopole operator). This
denes the monopole operator in the UV, independent of the choice of vacuum. Indeed,
(1.3) implies that the Coulomb branch coordinate = [ is pushed to an asymptotic
region close to the monopole,

q
J
2r
, for r 0, (1.4)
with r the Euclidean spacetime distance to the insertion point
3
. Here, (1.4) gives the
relation between the microscopic operator, whose denition is independent of the vacuum,
3
This behavior also entered in the analysis of [16] on the theory on S
2
R. The 1/r there ts
with the operator dimension () and one can rescale r 1.
3
and the Coulomb branch moduli space. Moreover, the quantum numbers of the monopole
operator are determined from the various Chern-Simons terms in this asymptotic region.
A Coulomb branch exists precisely when the monopole operator is gauge neutral.
When it is not, one can form a neutral composite of the monopole operator and the charged
matter elds, but they do not have the correct quantum numbers to give a Coulomb branch.
See also [23] for some discussion of monopole operators and the chiral ring.
1.3. Vortices
Though the elementary elds have charge (1.2) q
J
= 0, they can lead to solitonic
particles with charge q
J
,= 0. A classic example is a 3d A = 2 version of the Abelian Higgs
model, which on the Higgs branch has q
J
,= 0 Abrikosov-Nielsen-Olsen (ANO) vortices,
with BPS mass given by the FI parameter, m = [Z[, Z = q
J
. There had been some issues
in literature as to whether there are two or four states in BPS vortex SUSY representations,
whether their spins are half-integral or
1
4
, and issues related to the zero modes, so we
will give a detailed discussion. Some of the issues connect with the distinction between
the monopole operators, which are chiral operators, vs. the vortex BPS states.
We note that, while chiral operators O and BPS states are both annihilated by half the
supercharges, they are annihilated by dierent supercharges (linear combinations). Acting
with a chiral operator on the vacuum gives a tower of energy states, and the BPS state is
found by projecting to the lowest energy state in the tower. The basic BPS representation
is two-dimensional, and it becomes four-dimensional upon including its CPT conjugates.
In theories with non-minimal matter content, there are additional Bose and Fermi
zero modes of the soliton or vortex solutions
4
, which have log-IR-divergent norm when
the theory is on non-compact space. This was rst noted in [24] for solitons in the 3d,
non-SUSY CP
1
sigma model, and it also occurs in the Abelian-Higgs UV completion, and
more generally in theories with more than one charged eld.
We will here give an interpretation
5
of these non-normalizable zero modes, which does
not rely on the presence of vortices: the modes are present also in the q
J
= 0 sector. As we
4
For the monopole operator on S
2
, the = 0 in (1.4) lifts the matter elds, so there are then
no zero modes [16]. Indeed, even the operators in the same supermultiplet have dierent operator
dimension, so they have dierent energies on S
2
.
5
An interpretation discussed in [25] and references cited therein is that these new parameters
are the relative locations of partonic constituents of the solitons or instantons, which are frozen
by strong binding forces, as evident from the mirror dual [20] variables and gauge elds.
4
will further discuss, such zero modes should be interpreted as non-dynamical parameters,
labeling dierent superselection sectors of the quantum eld theory, analogous to vacuum
moduli. For the non-normalizable Fermi zero modes, our prescription is to quantize them,
which leads to a Fock space of states. Because we take space non-compact, this Fock space
gives multiple, disconnected Hilbert spaces. As we discuss, these zero modes aect the
vortex quantum numbers, and we will see how vortex quantum numbers match those of
the monopole operator, thanks in part to the zero mode contributions.
1.4. Generic real masses: a mass gap.
3d A = 2 theories with generic real masses, Chern-Simons terms, and FI parameters
have a mass gap, with isolated vacua. These could be isolated SUSY vacua, or SUSY
could be broken. One of the most basic diagnostics of the dynamics is the Witten index
Tr(1)
F
[26]. The Witten index of a theory in D spacetime dimensions can be computed
by considering the theory on RT
D1
. Let us contrast this torus-index with the sphere-
indices, on R S
D1
or S
1
S
D1
, which have been of recent interest. The torus-index
counts SUSY vacua, whereas the sphere-index (also called the super-conformal index
6
)
index counts the BPS operators. Indeed, the theory on a torus can have multiple vacua,
whereas the theory on a sphere always has a unique vacuum (the identity operator, in the
operator/state correspondence). In theories with non-compact moduli spaces of vacua the
Witten index is ill-dened. This is not an issue for the sphere index.
The index
7
of 3d SUSY G
k
Yang-Mills-Chern-Simons theories, without matter, has
been studied, starting with [28]. But the index of theories with matter has not yet appeared
in the literature
8
(as far as we are aware). We here compute the index of A = 2 theories
with matter, for non-zero real masses. With massless matter, the index is generally ill-
dened. By turning on the real parameters, the space of vacua can be made compact,
which leads to a well-dened, nite index, which we compute
9
.
6
This is a misnomer, as the sphere index is not restricted to superconformal theories, see [27].
7
In what follows, index refers to the Witten (or torus) index. Also, G
k
denotes gauge group
G with Chern-Simons coecient k.
8
It was observed via branes and the s-rule that certain 3d CS theories break supersymmetry
[29,30], and that N = 2 U(N
c
) has SUSY vacua if N
f
N
c
|k| [31].
9
Note that theories related by dimensional reduction generally do not have the same index.
Susy vacua can move in or out from innity, as the radius R of the S
1
R
compactication circle of
dimensional reduction varies between R = 0 and R = . E.g. 4d N = 1 SYM has h = C
2
(G)
SUSY vacua [26,32,33], whereas 3d N = 2 SYM (with k = 0) has a runaway, with no SUSY vacua.
The h 4d SUSY vacua are lost, at innity in eld space.
5
As we vary the parameters there can be phase transitions, where massless moduli arise.
If the moduli are compact, the index remains well-dened and cannot change through the
transition. We will see examples of this, e.g. changing the sign of a FI parameter can
open up a compact CP
N
moduli space. We will also see examples where there are phase
transitions where a non-compact moduli space arises. Since the index is ill-dened at such
a transition, it could in principle dier on either side, with additional vacua coming in
from innity, or disappearing to innity
10
. We show that the index of 3d A = 2 theories
actually does not change through any transitions, even in the non-compact case.
As a general argument as to why the index does not change, consider compactifying
the 3d theory on S
1
R
of radius R (such a compactication is needed anyway to dene the
index). The low-energy theory is a 2d A = (2, 2) theory with massless matter, and a
tower of massive matter elds from the Fourier modes with non-zero momentum on the
S
1
R
. In that theory, the real parameters of the 3d theory become complex parameters,
background expectation values of twisted-chiral superelds
11
. Holomorphy constrains the
2d twisted-chiral supereld parameters, much as in 4d, so any phase transitions are of
complex codimension, and can thus be avoided [36]. This proves that the Witten index
of 3d A = 2 theories cannot change as we vary the real parameters. Even though this
proof is complete, it is nice to verify it explicitly, as we will in several examples. We will
see that the fact that the index does not change looks highly non-trivial, as the vacua are
rearranged from the 3d perspective as we cross the phase transition.
We nd that the index of a general A = 2 G
k
theory, with generic real parameters, is
Tr(1)
F
= J
G
(k

), k

= [k[ h +
1
2

f
T
2
(r
f
), (1.5)
where J
G
(k

) is the number of primary operators of G

k
WZW theory. The

f
in (1.5)
is a contribution from the matter elds, Q
f
, in representations r
f
of G, with T
2
(r
f
) the
quadratic index of the representation (the number of
r
f
Fermion zero modes in a 4d
instanton). When there is no matter, the result (1.5) follows from the argument in [28]:
the h shift in (1.5) comes from integrating out the gauginos (with h =
1
2
T
2
(adj) the dual
Coxeter number, and this contribution is twice that in [28], since here we consider A = 2
10
See e.g. [34,35], and references cited therein, for 4d examples where that happens when SUSY
is broken.
11
Specically, m
R
m
C
and
c
= + i, where is the 2d theta term iF and the
additional parameters in m
C
are twists by global symmetry phases when circling the S
1
.
6
rather than A = 1), and connection between the index of the 3d Chern-Simons theory and
the 2d WZW theory follows from [28,37]. The matter contribution in (1.5), shifting k

by
+
1
2

f
T
2
(r
f
), is a new result, that will be explained here. For example, for U(1)
k
gauge
theory with matter chiral superelds Q
i
of charge n
i
, our result for the index is
Tr(1)
F
= [k[ +
1
2

i
n
2
i
, (1.6)
(with a modication for chiral theories with small [k[). The matter shift of k

could roughly
be anticipated from the shift k k
eff
from integrating out the real-massive matter elds,
though it will be seen that the details are much richer than this rough argument.
If k

< 0 in (1.5), then the theory dynamically breaks supersymmetry for generic real
parameter deformations. For example, that is the case for SU(N
c
) or U(N
c
) with N
f
fundamental avors if k h + N
f
< 0. Though supersymmetry is broken for generic real
parameter deformations, there are generally moduli spaces of supersymmetric vacua for
tuned, non-generic real parameters. This will be illustrated in the examples.
If k

= 0 in (1.5), it follows that there is a unique supersymmetric groundstate

Tr(1)
F
= 1 : G
k
gauge theory with [k[ h +
1
2

f
T
2
(r
f
) = 0 (1.7)
since G
k

=0
WZW theory has only the identity operator. As explained in [28] (for pure
A = 1 SUSY Chern-Simons theory without matter, but the same explanation applies
here), connement of electric ux can occur only if the center of G acts trivially on the
states, and that requires k

= 0. So simply conning theories must have k

= 0.
Dual theories must have matching index. This gives another test of dualities, in
addition to matching the moduli spaces, discrete parity anomaly matching for the global
symmetries in theories with
12
k = 0, and the more detailed checks of matching of the
S
2
S
1
partition function and S
3
sphere indices. For example, it was found in [3] that
SU(N
c
)
0
and U(N
c
)
0
, with N
f
= N
c
, has a simple dual description, to a theory of chiral
superelds, without gauge elds. The matching of Tr(1)
F
gives an immediate check, and
classication of possible generalizations. Such dual theories, without gauge elds, all have
a unique SUSY vacuum upon adding real masses
Tr(1)
F
= 1 : no gauge elds, all matter with real masses. (1.8)
12
When k = 0, P is broken. While the parity anomalies must still match, the check becomes un-
constraining, as global Chern-Simons coecients can then have more general RG running [9,38,39].
7
So we see from (1.8) that only k

= 0 theories can have a simple dual description, without

gauge elds. This matches the observation mentioned above, that the argument of [28]
shows that k

= 0 is a necessary condition for simple connement. The condition (1.8)

is indeed satised e.g. for SU(N
c
)
k
and U(N
c
)
k
when [k[ + N
f
= N
c
, agreeing with the
known cases in [3,31] of theories with simple duals, without gauge elds.
1.5. Aharony Duality and Giveon-Kutasov Duality
We discuss some aspects of Aharony duality [40]. The electric theory is U(N
c
)
0
, with
N
f
avors, and the dual theory is U(N
f
N
c
)
0
with N
f
avors and some added singlets with
superpotential terms. The unusual aspect of this duality is how the electric vs. magnetic
monopole operators are dualized, with the dual monopole operators explicitly appearing in
the dual Lagrangian. This peculiarity does not occur in 4d duals. For example, consider
4d A = 1 electric-magnetic duality [41], e.g. for the case of SO(N) theories [42]: the
magnetic monopoles of the electric theory map to the fundamental matter elds q
i
of the
dual, not to monopoles in the dual. The dierence in 3d can be anticipated from the fact
that monopole operators of U(N
c
) carry a conserved global charge q
J
(1.2).
For theories with non-zero Chern-Simons terms, Giveon and Kutasov [31] proposed
an interesting duality, between U(N
c
)
k
with N
f
avors and U([k[ +N
f
N
c
)
k
with N
f
avors. Here k ,= 0 eliminates the monopole operators. Correspondingly, this duality is
more conventional than Aharony duality, in that the dual theory does not contain monopole
operators in the Lagrangian. It has been shown [31,9] that, starting from Aharony duality
in the UV, there is a deformation that gives an RG ow to Giveon-Kutasov duality in the
IR. So Giveon-Kutasov duality can be derived as a consequence of Aharony duality.
We here discuss a reversed RG ow, from Giveon-Kutasov duality in the UV to
Aharony duality in the IR. In this way, we derive the peculiar monopole operator cou-
pling of Aharony duality as an output, rather than assuming it as an input. There has
been much recent evidence for both Aharony and Giveon-Kutasov duality from consider-
ing the theories on spheres, see e.g. [23,9,43], and our connection here between the two
dualities in at spacetime provides additional evidence. Adding generic real masses to
both sides of either Giveon-Kutasov, or Aharony, duals gives another check: the massive
matter elds can be integrated out, and the low-energy Chern-Simons theories are related
by level-rank duality. In particular, their indices from (1.5), match.
Our methods are also useful for upcoming work on relating 4d and 3d dualities [44].
8
1.6. Outline
The organization of this paper is as follows. In section 2, we give some background
material on U(1) gauge theories and the Higgs, Coulomb, and topological vacua. In section
3, we discuss general aspects of monopole operators and vortices, and the distinction
between them. In section 4, we discuss examples of U(1) gauge theories with Higgs or
Coulomb moduli spaces, which can be either non-compact or compact (in this context,
compact Coulomb branches rst appeared in [6]). In section 5, we discuss some generalities
about the Witten index [26], in general spacetime dimension, and we recall the results of
[28] for the index of 3d SUSY Chern-Simons theories without matter.
In section 6, we discuss the Witten index of 3d A = 2 gauge theory, with gauge group
U(1), with Chern-Simons term k, and general matter content, with generic real mass m
i
and FI term . In section 7, we compare the Tr(1)
F
index of some dual theories. In
section 8, we discuss SU(2)
k
gauge theory with various matter content. In section 9, we
discuss aspects of general SU(N
c
)
k
and U(N
c
)
k
theories. In section 10, we discuss a RG
ow from Giveon-Kutasov duality in the UV to Aharony duality in the IR.
In appendix A, we collect some details about our conventions, the SUSY algebra and
Lagrangians, vacua, and aspects of BPS vortices and zero modes. In appendix B, we recall
how the Coulomb branch is dualized via a Legendre transform between the linear multiplet
and chiral superelds [45,4], and how this can imply non-renormalization of the Kahler form
[46]. In appendix C, we comment about modes with log-divergent norm. In appendix D,
we provide some details about the derivation of Tr(1)
F
for U(1)
k
with matter, explicitly
verifying that the result is independent of variations of the real parameters.
2. 3d, A = 2 SUSY U(1) gauge theories
Consider a U(1) gauge theory with classical Chern-Simons term k, Fayet-Iliopoulos
term , and matter elds Q
i
of charge n
i
, and real mass m
i
. Consistency for compact U(1)
(as opposed to R) requires all n
i
Z and
k +
1
2

i
n
2
i
Z; equivalently, k +
1
2

i
n
i
Z . (2.1)
The unbroken global symmetry for generic and m
i
is U(1)
R
U(1)
J

i
U(1)
i
/U(1),
where U(1)
R
is an R-symmetry, U(1)
J
is associated with the topological current (1.1),
U(1)
i
rotates Q
i
by a phase, and the /U(1) is for the gauged U(1). The parameters m
i
9
and can be thought of as background values of scalars in classical gauge elds for these
global symmetries
13
. Since one linear combination of the U(1)
i
generators is gauged, one
linear combination of m
i
and is redundant, and can be absorbed in shifts of the scalar
= [ in the dynamical gauge supereld (1.1),
+, k, m
i
m
i
n
i
. (2.2)
It is important that the parameters m
i
and are bottom components of linear su-
perelds. Therefore, they are renormalized at most at one loop. Nevertheless, the actual
masses of elds with nonzero real masses are renormalized at all orders, because of Kahler
potential renormalization. We will continue to refer to the parameters m
i
as real masses,
even though they are not precisely masses of particles.
The semi-classical eective potential of the theory, with W
tree
= 0, is
14
V
s.c.
=
e
2
eff
32
2
_

i
2n
i
[Q
i
[
2

eff
k
eff
_
2
+

i
(m
i
+n
i
)
2
[Q
i
[
2
(2.3)
(Q
i
includes the Z
i
renormalization factor). The eective real mass of Q
i
for ,= 0 is
m
i
() = m
i
+n
i
, (2.4)
so Q
i
is massless at =
Q
i
, with m
i
(
Q
i
) = 0:

Q
i
m
i
/n
i
. (2.5)
The quantum-corrected k
eff
and
eff
include renormalization from integrating out matter
Q
i
with real masses m
i
() but, we emphasize, they must be piecewise eld independent
constants, modulo discontinuous jumps at the
Q
i
(2.5). Indeed, we have

eff
= +
1
2

i
n
i
m
i
sign(m
i
())
k
eff
= k +
1
2

i
n
2
i
sign(m
i
())

eff
+k
eff
= +k +
1
2

i
n
i
[m
i
()[ F() .
(2.6)
13
Therefore, if m
i
= 0, any added W
tree
, e.g. complex mass terms, must not break the U(1)
i
symmetry. Doing so would lead to explicit supersymmetry breaking.
14
Our normalization of the FI term has an unconventional factor of 1/2 (and an opposite sign);
this is natural in the context of 3d eld theory, where can be regarded as coming from a mixed
Chern-Simons term k
GJ
= 1 between the U(1) gauge eld and a background eld
J
| = for the
topological U(1)
J
global symmetry. This will eliminate many factors of 2 in later expressions
and is also natural in terms of the mirror symmetry map [20] m .
10
These expressions are similar to the ones in [6,7]. Condition (2.1) ensures that all
k
eff
Z; this is required for consistency. There are no higher order perturbative cor-
rections to (2.6): k
eff
must remain quantized, and
eff
is a component of a background
linear supereld. Higher order perturbative corrections cannot maintain these properties.
The function F() in (2.6) is the combination appearing in the potential (2.3), and is
continuous: the jumps at
Q
i
are in the slopes of F(). Using (2.6), the discontinuities of
k
eff
and
eff
for just above and below
Q
i
for generic m
i
are
k
+
eff,i
k

eff,i
= n
2
i
sign(n
i
) = n
i
[n
i
[, and
+
eff,i

eff,i
= n
i
m
i
sign(n
i
) = [n
i
[m
i
.
(2.7)
The semi-classical vacua of (2.3) can be found via solutions of

i
2n
i
[Q
i
[
2
= F(), and m
i
()Q
i
= 0, for all i . (2.8)
We refer to the SUSY vacua solutions of (2.8) as follows:
1. Higgs: vacua with some Q
i
,= 0, and thus =
Q
i
, as in (2.5). With non-generic
real masses and opposite sign matter, there can be non-compact Higgs moduli spaces
of vacua, similar to those of 4d theories. With generic real masses, on the other hand,
the Higgs vacua are isolated, with a mass gap, and only exist if the sign of F(
Q
i
) in
(2.8) coincides with that of n
i
. For later use, we dene
s
i
(n
i
F(
Q
i
)), where (x) =
_
1 for x > 0
0 for x < 0,
(2.9)
so, for generic real masses, there is a Higgs vacuum at
Q
i
i s
i
= 1.
2. Coulomb branch: vacua with Q
i
= 0, and thus F() = 0 in (2.8). We reserve the
name Coulomb branch for the case where there is a continuous (possibly compact)
moduli space of such vacua, i.e. a range of where
eff
= k
eff
= 0.
3. Topological vacua: there is a low-energy U(1)
k
eff
, with all matter massive and
integrated out, Q
i
= 0. Such vacua are isolated, with a mass gap for k
eff
,= 0,
located at the zeros of F(),

I
=

eff
(
I
)
k
eff
(
I
)
, i.e. F(
I
) +k
I
+
1
2

i
n
i
[m
i
(
I
)[ = 0, (2.10)
11
When there is a Coulomb branch, it is convenient to dualize the linear multiplet
to a chiral superelds X

, with U(1)
J
charge (1.2) q
J
= 1 [3,4] (see appendix B for a
review). Sometimes the X

elds are related to monopole operators, though we will here

make the distinction between the Coulomb vacuum modulus and the monopole operator.
Semi-classically, for a U(1) gauge theory, or for SU(2) U(1), these are given by
15
,
U(1) : X

e
(2/e
2
eff
+ia)
, or SU(2) : Y e
4/g
2
eff
+ia
, (2.11)
with a the 2 periodic scalar dual to the photon, F

= e
2
eff

a/2, and a can also

be regarded as the massless Nambu-Goldstone boson [2] for spontaneously broken U(1)
J
global symmetry for X

,= 0. With real masses, the Coulomb branch can become quite

intricate, with non-compact and/or compact regions, as depicted in Fig. 1. Monopole
operators push to and hence are associated with the non-compact regions (X
1
and X
4+
in Fig. 1) on the ends. The compact Coulomb regions are also parameterized
by chiral superelds (e.g. X
1+
and X
2
= 1/X
1+
in Fig. 1); they do not correspond to
monopole operators in the UV theory. Instead, these compact regions lead to Skyrmions,
which we will describe below.
Fig. 1: Quantum moduli space of U(1) with avors with (CP symmetric,
m
Q
i
= m
Q
i
) real masses. M
i
i
are mesons parameterizing Higgs branches,
X
1
and X
4+
parameterize noncompact Coulomb branches and the remaining
X
i
(with X
1+
= 1/X
2
, etc.) parameterize the compact regions of the
Coulomb branch. For more details, see below.
15
The relation is as follows. The standard normalization for the SU(2) Lagrangian, broken to
U(1) on the Coulomb branch, gives L =
1
4g
2

3
a=1
F
a

F
a

1
g
2
F

1
4e
2
F

, so
g
2
SU(2)
= 4e
2
U(1)
. Here F

1
2
F
a=3

, and A

1
2
A
a=3

, with the
1
2
introduced to couple to the
fundamental with U(1) SU(2) charges 1 rather than
1
2
; likewise,
U(1)
=
1
2

SU(2)
.
12
As usual, it is useful to tabulate the charges of the chiral superelds under the sym-
metries. When k
eff
,= 0, elds with q
J
charge, in particular the elds X

, become charged
under the U(1)
G
gauge group. Likewise, X

get charges under the global U(1)

i
and U(1)
R
symmetries
16
, via their induced mixed Chern-Simons terms with the U(1)
G
gauge eld [3]
k
G,i
=
1
2
n
i
sign(n
i
+m
i
), k
G,R
=
1
2

i
n
i
sign(n
i
+m
i
) (2.12)
from integrating out the Q
i
matter. In particular, for the asymptotic regions ,
which we associate with operators X

, respectively, this gives the charges

U(1)
G
U(1)
j
U(1)
R
U(1)
J
Q
i
n
i

ij
0 0
X

(k k
c
)
1
2
[n
j
[
1
2

i
[n
i
[ 1
(2.13)
where the asymptotic values of k
eff
() for are:
k
eff
( = ) = k k
c
, k
c

1
2

i
n
i
[n
i
[; (2.14)
k
c
vanishes in theories with vector-like matter, where charge conjugation C: n
i
n
i
is a
symmetry. The charges in (2.13) are computed by going to = , which is only possible
when there is a non-compact Coulomb branch, i.e. k = k
c
(or k = +k
c
), so k
eff
= 0 in
the asymptotic region (2.14), and X
+
(or X

) is gauge neutral. The charges (2.13) should

thus be understood as being applicable only when the U(1)
G
charge of X vanishes. For
k
c
= 0, both X
+
and X

can be neutral, while for k

c
,= 0 at most one can be neutral,
corresponding to a non-compact Coulomb branch.
Another condition for the X

Coulomb branch to be unlifted is that the asymptotic

value of the FI parameter obtained from (2.6),

eff
( = ) =
c

c

1
2

i
[n
i
[m
i
(2.15)
must vanish,
c
. This can equivalently be understood from the X

quantum numbers
in (2.13): since m
j
is a background for U(1)
j
and is a background for U(1)
J
, the real
mass of X

(its central charge Z) is determined by the charges in (2.13) to be

m
X

j
m
j
(
1
2
[n
j
[) , (2.16)
16
The U(1)
R
symmetry given here is simply a particular choice; one can obtain other conserved
R-symmetries by mixing with the U(1)
i
global symmetries. In particular, the superconformal R-
symmetry is determined as in [47,48] (including also any accidental global U(1) symmetries).
13
so X

is massless only if k = k
c
and
c
.
The chiral monopole operator is related to an insertion of the operator X

, so it
must have the global quantum numbers of X

in (2.13), in cases where there is a non-

compact Coulomb branch. In [16], the global quantum numbers of the monopole chiral
operator were determined, for the example of U(1) with N
f
vector-like avors, by instead
considering the energy spectrum of the theory on S
2
, with the BPS monopole gauge eld
background, computing the vacuum charges by heat kernel regularization. The charges in
(2.13) reproduce the results of [16] via the simpler (but conceptually equivalent) method
of using the induced Chern-Simons terms (2.12) of the theory on non-compact space.
For a U(1)
k
theory with N
+
matter elds Q
i
of charge +1, and N

elds

Q
i
of charge
1, the global symmetry is SU(N
+
) SU(N

) U(1)
A
U(1)
R
U(1)
J
, with charges
SU(N
+
) SU(N

) U(1)
A
U(1)
R
U(1)
J
Q
i
N
+
1 1 0 0

Q
i
1 N

1 0 0
M
i
i
= Q
i

Q
i
N
+
N

2 0 0
X

1 1
1
2
(N
+
+N

)
1
2
(N
+
+N

) 1
(2.17)
The charges of X

are as in (2.13) and [3,10] and follow from (2.13), via the induced mixed
Chern-Simons terms k
GA
=
1
2
(N
+
+ N

) sign() = k
GR
between the gauge U(1)
G
and
U(1)
A
and U(1)
R
, with sign() = 1 for X

. As in (2.13), the operators X

are only
gauge neutral if k
eff
= 0; if not, X

are removed from the chiral ring of spin zero, gauge

invariant operators and, correspondingly, there is no Coulomb branch. Using (2.6),
k
eff
= k +
1
2
(N
+
N

) sign(), (2.18)
for m
i
= m
i
= 0. So there is a non-compact half-Coulomb branch for k =
1
2
(N
+
N

),
upon taking 0, parameterized by X

.
As a soon-useful aside, we comment on the case when the charges n
i
all have a common
integer factor, i.e. all n
i
= n n
i
, with n and n
i
integers. This motivates the rescaling
V

V nV, e
2
e
2
n
2
e
2
, n
i
n
i
n
i
/n, k

k k/n
2
,

/n.
(2.19)
If the gauge group is non-compact, i.e. R instead of U(1), the original and rescaled theories
would be physically identical. For compact U(1), with

k an integer, the rescaled theory
14
(2.19) is a Z
n
orbifold of the original theory. Indeed, even if all Q
i
, = 0, the original U(1)
gauge group is not fully broken, U(1) Z
n
. If the original theory has q
J
=
_
F/2 Z,
the rescaled theory (2.19) has q
J
=
_

F/2 nZ. The scalar dual to the photon of the
rescaled and original theories are related as a = na/n
2
= a/n (the 1/n
2
is from e
2
), so
while a

= a + 2, the periodicity of a is 2/n, i.e. it is a Z

n
orbifold.
3. Aspects of vortices and monopole operators
As discussed in the introduction, we are here interested in vortices and their connection
with monopole operators in at spacetime. Many aspects of such operators were elucidated
in [15,16], focussing on the theory on S
2
R for technical reasons. Here we focus on at
R
1,2
spacetime. A key physical dierence is that theories on non-compact space can have
multiple vacua, including continuous moduli spaces, which are superselection sectors. We
will discuss some additional superselection parameters. Theories on compact space, e.g.
S
2
R, instead have a unique vacuum, with classical moduli integrated over
17
.
To insert a monopole operator, say at x = 0, one removes a small ball from R
3
around
the origin and imposes boundary conditions on the eld strength, as in (1.2), around the
hole, setting the charged matter elds to zero there, Q
i
= 0. In our supersymmetric
context, we are interested in monopole operators that are chiral, annihilated by half the
supercharges. In superspace, this chiral condition is written as (1.3) for , which requires
that = [ satises

2
E
= 2q
J

(3)
(x) (3.1)
with
2
E
the 3d Euclidean Laplacian. So the solution near the hole is

q
J
2r
, (3.2)
with r the Euclidean radial distance to the hole. Indeed, in order for the conguration to
be chiral, annihilated by half of the supercharges, requires the dimensional reduction of
the 4d condition that the eld strength be self-dual, which gives
r
2

r0
=
1
2
q
J
, (3.3)
17
Moreover, as in [27], to have unbroken supercharges on S
2
R requires a background U(1)
R
gauge eld b

leading to L =
1
2
b

J
FZ

, here with b
0
1/r. For the Euclidean theory S
2
R
E
or S
2
S
1
, these backgrounds are imaginary [27].
15
so the solution near the hole is again given by (3.2).
As the hole becomes small and r 0, the eld at the hole, given by (3.2) has
sign(q
J
). The monopole operator insertion thus forces the eld nearby to be out
in an asymptotic innity region on the Coulomb branch. This is intuitively satisfying: it
explains why this operator is associated with the Coulomb branch. The inserted operator
of charge q
J
is chiral, X
q
J
+
for q
J
> 0, or X
|q
J
|

for q
J
< 0. Anti-chiral operators have
opposite sign in (3.1), (3.2) and (3.3); they are (X

)
q
J
for q
J
> 0 and (X

+
)
|q
J
|
for q
J
< 0.
3.1. Chiral operators vs. BPS states
Monopole operators are chiral operators, whereas vortices are BPS states. We here
emphasize the general distinction between chiral operators vs. BPS states.
The supersymmetry algebra is (see appendix A for details)
Q

, Q

= 2

+ 2i

Z, (3.4)
where the SL(2, R) Lorentz spinor has (Q

) = Q

. The energy P
0
= P
0
is positive. The
central element Z gets contributions only from the global (not gauge, see [3] for discussion),
non-R, U(1) charges and associated real parameters
Z =

i
q
i
m
i
, (3.5)
where q
i
is the charge under the global U(1)
i
symmetry, and the sum includes the U(1)
J
global symmetry, with m
J
= the FI parameter.
It is convenient to dene
Q

=
1
2
(Q
1
iQ
2
), Q

= Q

=
1
2
(Q
1
iQ
2
), (3.6)
such that Q

and Q

have spin
1
2
. In terms of these, the algebra (3.4) is
Q

, Q

= iP
1
+P
2
, Q

, Q

= P
0
Z. (3.7)
A chiral operator O(x

), at spacetime point x

, preserves half the supercharges

18
,
[Q

, O(x)] = 0. (3.8)
18
Note that we cannot pick O to be annihilated by one of the Qs and one of the Qs: that
would be incompatible with Lorentz invariance.
16
The two Qs puts the operator O(x

) in a four dimensional representation

O ;

(O) = [Q

, O] ; F(O) = Q, [Q, O] (3.9)

(If O has spin, it is more complicated.) If O is complex, CPT means that this representation
appears twice. The equation of motion can reduce the representation, and can make it
2d. For example: the bottom component of a free chiral supereld (even with real mass)
satises the Klein-Gordon equation, the Fermions satisfy the Dirac equation, and F = 0.
When Z = 0, the representation is completely analogous to 4d A = 1. On the other
hand, when a chiral operator O is charged under a global non-R U(1)
i
symmetry, with
real parameter m
i
non-zero, then O creates states with Z ,= 0 in (3.5). The representation
is still four-dimensional (3.9), but the algebra acts on this representation dierently.
For a massive particle state, we can go to the rest frame, P
1
= P
2
= 0, P
0
= m. A
BPS state with Z > 0 is annihilated by Q

and its complex conjugate Q

+
in (3.7),
Z > 0 : Q

[BPS = Q
+
[BPS = 0. (3.10)
The remaining two supercharges make a two-dimensional representation
Z > 0 : [BPS =
_
[a
[b
_
, Q

[a = 0, [b = Q
+
[a. (3.11)
The CPT conjugates have Z < 0 and are thus anti-BPS states, with
Z < 0 : Q

[BPS = Q
+
[BPS = 0 (3.12)
and with Q

and Q
+
forming a two-dimensional representation
Z < 0 : [BPS =
_
[a
[b
_
, Q

[a = 0, [b = Q
+
[a. (3.13)
The R-charges and spins of these states are
19
U(1)
R
U(1)
spin
Z
[a r s > 0
[b r 1 s +
1
2
> 0
[a r s < 0
[b r + 1 s +
1
2
< 0
(3.14)
19
The state and its CPT conjugate have the same spin (P and T both ip the sign). For
example, for Chern-Simons-Maxwell theory, the spectrum includes a single kind of particle with
spin sign(k), which is its own CPT conjugate.
17
The simplest example is a free chiral supereld with real mass m. The real mass m
couples to a global U(1) symmetry under which the chiral supereld has charge +1. Z
is the product of this charge and m. For positive m we have two states with charge +1:
a scalar and a Fermion with spin +
1
2
and their CPT conjugate states with charge 1: a
scalar and a Fermion with spin +
1
2
. This corresponds to the states with s = 0 in (3.14).
To clarify the relation between the chiral operator O (3.8) and the BPS states, consider
the state O[0. While O[0 is annihilated by Q

, it is not a BPS state: it is not annihilated

by any Q

. The BPS states (3.11) are obtained by projecting to the lowest energy state
in O[0:
[a = lim

e
(HZ)
O[0 [b = Q
+
[a, (3.15)
where the global charges q
i
of the operator O must be such that (3.5) has Z > 0 for the
limit (3.15) to give a non-zero state. Similarly, upon projection to the lowest energy state,
anti-chiral operators with Z < 0 can create anti-BPS states (the CPT conjugates of (3.15))
[a = lim

e
(H+Z)
O[0 [b = Q
+
[a. (3.16)
3.2. BPS states from solitons
In appendix A.3, we summarize aspects of BPS vortices for general U(1)
k
gauge theory
with matter. There are (anti) BPS eld congurations when the FI parameter ,= 0, which
are particles with m = [Z[, with Z = q
J
. The BPS eld congurations are annihilated by
Q

and Q
+
(3.11), which acting on the Fermions give the BPS dierential equations for
the bosonic elds. The (anti) BPS equations indeed require Z > 0 (Z < 0). The vortex-
soliton with charge q
J
= 1 always has a single normalizable zero mode supermultiplet.
The bosonic component gives the position of the BPS particle in the spatial plane. Its
Fermionic superpartner is quantized and gives the BPS doublet (3.11).
For U(1) with a single matter eld, these are the only zero modes. With additional
matter elds, there are restrictions on the vacuum in order to admit BPS congurations:
essentially, the vacuum must have a single eld having non-zero expectation value. In such
congurations, each matter eld has a zero mode supermultiplet. The matter eld with
expectation value gives a normalizable zero mode with the same interpretation as above,
giving the BPS particle position, and the doublet structure. The additional matter elds
contribute non-normalizable zero modes, whose interpretation is more subtle. These issues
are discussed further in appendix C.
18
A Chern-Simons term k aects the Gauss law constraint, giving
1
e
2
_

E
i
dx
i
q
Gauss
= q
elec
+kq
J
, (3.17)
where q
elec
is the charge of the matter elds. In the Higgs phase, E
i
/e
2
0 in the IR, so
q
Gauss
= 0 and thus vortices have electric charge (which is anyway screened) when k ,= 0,
q
elec
= kq
J
. (3.18)
This aects the spin of the vortices as in [49,50]
s =
1
2
kq
2
J
. (3.19)
We will show in the next section that, when a Coulomb branch exists, the quantum
numbers of a vortex coincide with those of the monopole operator in (2.13). The Coulomb
branch exists precisely when there is a spin zero, neutral vortex state, whose condensation
can be interpreted as giving the Coulomb branch as a mirror Higgs branch.
4. Examples of U(1) gauge theories with moduli, and vortices
We here discuss some examples with compact and non-compact Higgs or Coulomb
branches, and aspects of the connection between the Coulomb branch and moduli opera-
tors. For general U(1)
k
theory, with matter elds Q
i
of charges n
i
, the theory with generic
real parameters m
i
and has only isolated vacua, with a mass gap; that case will be
discussed in section 6, where we compute the Witten index. In the present section, we
discuss tuned values of k, m
i
and , that lead to moduli spaces.
As we noted after (2.14), the asymptotic regions of the Coulomb branch,
corresponding to X

, only exist if k
eff
there vanishes, i.e. if k = k
c
. As seen from (2.8),
a Coulomb branch, where all Q
i
= 0, requires F() = 0, i.e. both k
eff
= 0 and
eff
= 0,
and using (2.6) this requires a particular value of both k and , to get F( ) = 0:
non-compact Coulomb, X = X

: for k = k
c
, and =
1
2

i
[n
i
[m
i
. (4.1)
Semi-classically, the Coulomb branch X
+
in (4.1) has (
max
Q
i
, +), where
max
Q
i
is
the largest of the
Q
i
in (2.5), and likewise X

has (,
min
Q
i
). There will instead
be a compact Coulomb branch if k and are tuned such that the function F() in (2.6)
vanishes in an intermediate region for , in between two
Q
i
values.
There will be a non-compact Higgs branch moduli space if two (or more)
Q
i
, with
opposite sign n
i
, are tuned to coincide. There will be a compact Higgs branch if two
(or more)
Q
i
, with the same sign n
i
, are tuned to coincide and is chosen such that
sign(F(
Q
i
)) = sign(n
i
), to satisfy (2.8).
We now illustrate these moduli spaces in simple, characteristic examples.
19
4.1. U(1)
k
with one eld, Q, of integer charge n > 0
For the n = 1 case, we refer to this theory as N
f
=
1
2
(N
f
= 1 has Q and its charge-
conjugate,

Q). The quantization condition (2.1) is k +
1
2
n Z; so n = 1 has k +
1
2
Z,
hence k ,= 0. As we will discuss, the general n case is a Z
n
orbifold of the n = 1 case.
Vacua
The theory has a single real parameter, which we take to be (using (2.2) to set
m
Q
= 0). The eective CS and FI terms are (2.6)
k
eff
= k +
1
2
n
2
sign(),
eff
= . (4.2)
For generic k and , the theory has a mass gap, with isolated supersymmetric vacua. As
in (4.1), there is a Coulomb branch solution of (2.8) if k
eff
=
eff
= 0, i.e.
non-compact Coulomb, X = X

: for k =
1
2
n
2
, and = 0. (4.3)
There are conserved U(1)
R
and U(1)
J
global symmetries with charges as in (2.13): the
Coulomb modulus X has R(X) =
1
2
n, and q
J
(X) = sign(k) (we take R(Q) = 0, as in
(2.13), so Q ,= 0 does not break U(1)
R
).
The n = 1 case of (4.3), with [k[ =
1
2
, is dual to a single, free chiral supereld [8,9,51],
which ts with R(X) =
1
2
. The duality requires that the Kahler potential for X be smooth
at the origin (i.e. in the region [X[
2
e
2
)
K([X[)
_
e
2
_
log([X[)
_
2
[X[
[X[
2
[X[ 0
(4.4)
where the U(1)
J
symmetry implies that K only depends on [X[ and we included the large
[X[ dependence thats known from the perturbative U(1) limit. Since the duality maps the
global symmetry U(1)
J
U(1)
X
, it maps m
X
: the FI parameter of the U(1) theory
to the real mass parameter m
X
for the free supereld X of the dual. For = m
X
= 0, the
dual theories have the same, non-compact moduli space. Turning on ,= 0, or m
X
,= 0 in
the dual, gives Tr(1)
F
= 1 supersymmetric vacuum.
The Coulomb branch theory (4.3) for integer n > 1 is, as discussed after (2.19), a Z
n
orbifold of the theory with n = 1. The Coulomb coordinate can be written in terms of

X
which has the quantum numbers of X
1/n
, e.g. it has the U(1)
R
charge of a free eld, but
20
with a Z
n
phase identication. The local theory for n > 1 is free, but there are twisted
sector, codimension two, line-operator elds associated with the orbifold.
Vortices and monopole operators
We take Q to have U(1) charge n = 1. Taking > 0, there is a Higgs vacuum,
[Q
1
[
2
= /2, breaking U(1). This is the A = 2 version of the classic Abelian Higgs
model, which has ANO vortices and has been much discussed (with some debates) in the
literature; see e.g. [52,50,53]. The necessary presence of the Chern-Simons term k ,= 0,
since k +
1
2
Z, has important eects. As in the analogous case of 2d LG theories [54,55]
the BPS solitons are doublets, and the spins are half-integral rather than semionic.
The elds for the static soliton with winding number q
J
have the standard vortex
asymptotics far from the core [56],
lim
z
Q
1
=
_

2
e
iq
J

+. . . A

= q
J
+. . . , (large [z[) (4.5)
where z = x + iy = [z[e
i
is the space coordinate, and the . . . are higher order in 1/[z[.
The deviations from the asymptotic behavior (4.5) is here exponentially suppressed: since
the theory is in the Higgs phase, the magnetic ux is conned by the Meissner eect to a
thin ux tube, with core size set by the photon mass.
For the A = 2 theory, the self-duality equations [56] are the BPS soliton equations
[50]. The vortex with U(1)
J
charge q
J
= 1 has a zero mode which forms one chiral
supereld particle degree of freedom, with complex Bose component associated with the
position of the vortex in the spatial plane. The complex Fermi zero mode superpartner is

+
, and complex conjugate

, associated with the broken supersymmetries Q

+
and Q

.
The
+
zero mode has R-charge +1 and spin
1
2
and its complex conjugate, from

, has
the opposite charges. The quantization of
+
and

leads to the BPS doublet of states

[a and [b in (3.14), with charges (using (3.19))
U(1)
spin
U(1)
J
U(1)
R
[a
1
2
k
1
4
1
1
2
[b
1
2
k +
1
4
1
1
2
(4.6)
Because we consider compact U(1) (as needed for vortices), the k quantization condition
(2.1) is k Z +
1
2
. So the spin in (4.6) is always half-integral or integral, not semionic.
The solitons (4.6) have BPS mass m = [[.
21
We have seen that the theory with k =
1
2
has a gauge-neutral operator X
+
, which
is a chiral supereld with zero spin and R =
1
2
. For = 0, X
+
is massless and there is
an associated non-compact Coulomb branch. For k =
1
2
, we can indeed identify [a with
the projection (3.15) of X
+
[0, since [a has the same quantum numbers (4.6) as X
+
for
k =
1
2
. The state [b in this case is similarly obtained from the Fermionic superpartner
(X
+
)[0. The conjugate states [a and [b are likewise created by the anti-chiral operator
X

+
and its Fermionic partner. For k = +
1
2
, on the other hand, the state [b in (4.6) has
spin 0, with R =
1
2
. In this case, the state [b has the quantum numbers to be identied
with the projection of X

[0, and [a is created by the Fermionic parter of X

.
For [k[ , =
1
2
, the vortex has nonzero charge (charge is anyway screened in the Higgs
phase) (3.18) and spin s ,= 0 (4.6). Therefore, they cannot condense without spoiling
rotational symmetry. This ts with the fact seen from (2.13) that X

are only gauge

neutral operators if [k[ = [k
c
[ =
1
2
, so the Coulomb branch only exists in that case.
4.2. U(1)
k
with N elds Q
i
of charge n
i
= +1
The quantization condition (2.1) here requires k +
1
2
N Z. Using (2.2), we take m
i
in the Cartan subalgebra of the SU(N) global symmetry,

i
m
i
= 0.
Vacua
There are Higgs vacua, for any k, at
Q
i
= m
i
if
eff
> 0. For generic m
i
, these
are isolated SUSY vacua, with a mass gap. Upon tuning two or more m
i
to coincide, they
instead lead to a compact Higgs branch moduli spaces of vacua. Taking all m
i
= 0, the
theory with > 0, and any k has a compact Higgs branch /
H
= CP
N1
.
Now consider Coulomb vacua. Non-compact Coulomb branches (4.1) X

exist for
k =
1
2
N, and = 0. For [k[ <
1
2
N, and m
i
chosen appropriately and appropriately
tuned, there is a compact CP
1
Coulomb branch, for between the neighboring values of

Q
i
= m
i
where k
eff
= 0. Consider e.g. N = 2 elds, with m
1
= m
2
= m > 0. Then
(2.6) gives
F() =
eff
+k
eff
= +k +
1
2
[ +m[ +
1
2
[ m[. (4.7)
For k = 1 (or k = 1) and = 0, there is a non-compact Coulomb branch (4.1) with
< m ( > m). For k = 0, at = m, there is a compact Coulomb branch, [[ < m,
giving /
C

= CP
1
upon including the compact scalar a dual to the photon [6]. The local
theory on this CP
1
Coulomb branch is everywhere smooth. In particular, at = m (or
22
= m), the metric is smooth as in (4.4), as the low-energy theory there is U(1)
k
low
, with
k
low
=
1
2
and one charged eld Q.
For generic m
i
, the non-compact or compact Coulomb branches mentioned above are
all smooth, and free. Upon tuning two or more m
i
to coincide, the low-energy theory
has a singularity associated with additional massless degrees of freedom. To illustrate it,
consider taking all m
i
= 0, so (2.6) gives k
eff
= k +
1
2
N sign() and
eff
= . There is a
non-compact Coulomb branch X
+
(or X

) for k =
1
2
N (k =
1
2
N) and = 0, with 0
( 0). The theory at the origin, X = 0, is singular for N > 1, i.e. there are additional
degrees of freedom there; for general N and k, it is an interacting SCFT.
Solitons (for m
i
= 0)
For > 0 and any k, the vacuum is on the compact Higgs branch /
H
= CP
N1
;
by an SU(N) rotation, we take the vacuum to be Q
i
[ =
_
/2
i,1
. This theory has BPS
vortices, where Q
1
and A

have the same eld conguration solution as in the N = 1 case,

with asymptotic behavior (4.5) far from the core. The solution has the same, associated,
normalizable chiral supereld zero mode, giving the particles spatial position and super-
doublet structure (3.11). The matter superelds, Q
i>1
, provide N 1 additional chiral
supereld zero modes, satisfying D
z
Q
i
= 0, with charge n
i
= 1 coupling to the winding
gauge conguration A

of the vortex.
As discussed in appendix C, these additional zero modes are non-normalizable. The
bosonic components are frozen, superselection sector parameters. The question then arises
whether or not to quantize the non-normalizable Fermion zero modes. As we discuss in
appendix C, the Fermi components are to be quantized, but map between dierent Hilbert
spaces, so they aect the vortices quantum numbers, but not the number of states.
This U(1) linear sigma model gauge theory is the UV completion of the low-energy
CP
N1
non-linear sigma model. The U(1) vortices are, correspondingly, the UV comple-
tion of the Skyrmionic lump solutions [24] of the low-energy CP
N1
sigma model, asso-
ciated with eld congurations that wrap the non-trivial two-cycles in the moduli space,

2
/
H
, with size
_

2
= calibrated by Kahler form , around the two dimensional
space surrounding the soliton. Indeed, there are BPS solutions that interpolate between
the ANO vortices and the CP
N1
Skyrmionic lumps; see e.g. [57-59].
Following the prescription in appendix C, the upshot is that there are BPS vortices as
in (3.14), with U(1)
R
charge r =
1
2
N and U(1)
spin
, s =
1
2
(k
1
2
N). When k =
1
2
N, we
23
nd the spin 0 states, matching the quantum numbers of X
+
and X

, which are neutral,

matching the existence of non-compact Coulomb branches for [k[ =
1
2
N.
When there is a compact Coulomb branch, we can likewise consider BPS Skyrmions
which wrap the compact Coulomb vacua. Consider e.g. the N = 2 case with real masses
m (4.7), where for k = 0 there is a compact Coulomb branch for [[ m, /
C

= CP
1
.
The BPS mass of the Skyrmion is given by the size of the CP
1
, which is calibrated by a
Kahler form which is non-renormalized, see [4,46] and appendix B. So the BPS Skyrmion
mass is
m
BPS
=
_
=
_
1
4
d
i
da
i
= [m[. (4.8)
We will discuss an analogous setup in the next subsection, which will be more interesting,
in that the Coulomb branch Skyrmions map to the quanta of the Higgs branch meson op-
erator. In the present case, this does not happen, as there is no non-compact Higgs branch
and, correspondingly, no gauge-invariant analog of the meson chiral operator, because here
Q
1
and Q
2
both have positive charge. We identify the topological charge of the Skyrmion
in the present example with a U(1) SU(2)
F
; this interpretation is consistent with (4.8).
4.3. U(1)
k
with N
f
= 1 avor of matter, Q and

Q, of charge 1
We take m
Q
= m
Q
= m 0. The function F() is
F() = k
eff
+
eff
= +k +
1
2
[ +m[
1
2
[ m[. (4.9)
Vacua
For m = 0, there is a non-compact Higgs branch moduli space /
H
of SUSY vacua,
labeled by arbitrary M with M = Q

Q, for all k and . The metric on /

H
depends on
, e.g. it can be singular at M = 0 for = 0, where there is an interacting SCFT for all k.
For m ,= 0, /
H
is lifted, and there is instead an isolated supersymmetric Higgs vacuum
at = m (or = +m), for positive (or negative). The low-energy theory there is
U(1)
k
eff
with light eld Q (or

Q) and k
eff
= k
1
2
(or k
eff
= k +
1
2
) from integrating out
the other eld.
Now consider the topological or Coulomb vacua, where (4.9) gives F() = 0. For
k ,= 0, 1, there are no Coulomb vacua, only isolated topological vacua, for any m. For
k = 0, there are non-compact Coulomb branches where , labeled by X

, for
= m. The X
+
non-compact Coulomb branch starts at = m, where Q is massless
24
and

Q can be integrated out, leading to U(1)

1
2
there: this is the theory of subsection 4.1,
which is dual to a smooth, free eld theory X. Likewise, the X

non-compact Coulomb
branch, for < m, comes from the CP conjugate theory, U(1) 1
2
, with massless eld

Q
at = m. Upon tuning m 0, X

, merge together and the theory at the origin is an

interacting SCFT, dual to W = MX
+
X

[3].
For k = 1, there is a compact Coulomb branch for = 0, with F() = 0 for [[ m.
Including the dual photon, this /
C

= CP
1
. The CP
1
is everywhere smooth; in particular,
the local theory at = m is the U(1)
k=
1
2
theory, dual to a free eld.
Solitons
There is a Higgs-branch BPS vortex for non-zero and any k, m. For m ,= 0, where
the Higgs vacua are isolated with a mass gap, the vortex is essentially that of the low-
energy N
f
=
1
2
theory (4.5). For positive (negative), the light eld Q (or

Q) has winding
expectation value (4.5) in the low-energy theory at = m (or = m). The other eld
has real mass, and does not contribute any additional Bose or Fermi zero modes.
For m = 0, for any k, there is a BPS vortex for any non-zero , but only for one, special
vacuum on the noncompact Higgs branch. For > 0 it is the vacuum with [Q[
2
= /2,
with

Q = 0. The BPS soliton has Q = Q
1
and gauge eld as in the N
f
=
1
2
theory,
(4.5). There is the same, normalizable Bose and Fermion zero modes as the N
f
=
1
2
theory, associated with the super-translations of the soliton core. There is a complex, non-
normalizable Bose and Fermion zero mode associated with

Q. Following the prescription
in appendix C, this mode gives superselection sectors, which contributes to the vortex
quantum numbers in the spectrum (3.14), but not its number of states.
There is also a Coulomb branch Skyrmion for k = 1, = 0, where /
C

= CP
1
,
since
2
(/
C
) = Z. As discussed in the previous subsection, for Higgs-branch Skyrmions,
q
J
is given by the
2
(/
H
) winding number. We have here a mirror-dual situation, where
now
2
(/
C
) gives the charge q
M
=
1
2
q
A
under the global U(1) symmetry which assigns
charge 1 to M (so
1
2
to Q and

Q). In particular, the basic Skyrmion wrapping the /
C
Coulomb branch has the same quantum numbers as the operator M, so can potentially
be identied as the BPS state associated with the chiral operator M. The BPS Skyrmion
has mass given as in (4.8). It becomes massless as m 0, and it can there condense and
be interpreted in terms of the non-compact Higgs branch which exists for m = 0.
25
4.4. U(1)
k
with N
+
matter elds Q
i
of charge +1, and N

matter elds of charge 1.

The CS term satises (2.1), k+
1
2
(N
+
+N

) Z. Taking the real masses to vanish, the

global symmetries are as in (2.17). There is a non-compact Coulomb branch, parameterized
by X

, for k =
1
2
(N
+
N

), and = 0. There is a Higgs branch /

H
, of complex
dimension dim
C
/
H
= N
+
+ N

1. For N
+
N

= 0, /
H
is compact and only exists
for of correct sign. For N
+
N

,= 0, the Higgs branch is non-compact, and can be

parameterized by M
i
i
= Q
i

Q
i
with rank 1,

ij...

ij...
M
i
i
M
j
j
= 0 . (4.10)
When
eff
0, the Higgs branch can touch the origin, Q
i
=

Q
i
= 0, where it can connect
with the Coulomb branch, when it exists. For
eff
,= 0, the Higgs branch elds are away
from the origin, and there can be BPS solitons on special points or subspaces of /
H
.
We can turn on real masses m
i
for Q
i
, and m
i
for

Q
i
, in the Cartan subalgebra of
SU(N
+
) SU(N

) U(1)
A
. Let us consider the case m
Q
i
= m
Q
i
= m, i.e. a background
value for U(1)
A
only; by using P, we can take m > 0 without loss of generality. Using
(2.6), we nd
k
eff
= k +
1
2
N
+
sign(m+) +
1
2
N

sign(m),

eff
= +
1
2
mN
+
sign(m+)
1
2
mN

sign(m).
(4.11)
There is a Coulomb branch when k
eff
=
eff
= 0. So there is a non-compact Coulomb
branch X
+
, for > m, if k =
1
2
(N
+
N

) and =
1
2
m(N
+
+ N

), as then k
eff
=

eff
= 0. Likewise, there is a non-compact Coulomb branch X

, for < m, if k =
1
2
(N
+
N

) and =
1
2
m(N
+
+N

). The global charges of X

are as in (2.17).
There is also a compact Coulomb branch, with [[ < m, if k =
1
2
(N
+
+ N

), and
=
1
2
m(N
+
N

). The low-energy theory at = m, after integrating out the massive

Q
i
elds, has N
+
massless elds Q
i
of charge +1, and k
low
=
1
2
N
+
. This low-energy
theory is smooth at = m only for N
+
= 1, otherwise there are additional degrees of
freedom. The low-energy theory at = +m likewise has N

massless

Q
i
elds of charge
1, with k
low
=
1
2
N

, and is smooth there if N

= 1.
26
4.5. U(1) with N
+
= N elds Q
i
of charge +1, and N

= 1 eld
20

Q of charge 1.
These theories are a special case of those of the previous subsection, and the discussion
of the Coulomb branches there directly apply here. The nice aspect of this class is that
the Higgs branch simplies, as it is simply given N unconstrained moduli M
i
= Q
i

Q. The
classical metric on /
H
can be found from
K
cl
= Q

i
e
V
Q
i
+

Q

e
V

QV (4.12)
upon integrating out V , which gives
K
cl
=
_

2
+ 4M

i
M
i
+ log(
_
4M

i
M
i
+
2
). (4.13)
For = 0, K
cl
= 2
_
M

i
M
i
is singular at the origin, while for ,= 0 /
H
is smoothed out.
For < 0, /
H
is topologically trivial, /
H

= C
N
. For > 0, on the other hand,
c
1
(/
H
) ,= 0 and there is a Kahler form that calibrates a two-cycle in /
H
, with volume
, similar to the Fubini-Study Kahler potential of CP
N1
. These two cases can be seen
from the behavior of (4.13) near the origin:
K
cl
(M

i
M
i

2
)
_
1

i
M
i
+ log(M

i
M
i
) > 0
1
||
M

i
M
i
< 0
(4.14)
with both cases smooth at the origin for ,= 0. Far from the origin, (4.12) gives
K
cl
(M

i
M
i

2
) 2
_
M

i
M
i
+
1
2
log(M

i
M
i
). (4.15)
The log(M

i
M
i
) term can be eliminated by a Kahler transformation for N = 1, but leads
to a non-zero contribution to the Kahler metric for N > 1. Quantum eects will of course
locally modify the classical metric, but we expect that the full quantum metric on /
H
has the same topology, with a non-trivial two-cycle, of size for > 0. So the > 0
case has BPS Skyrmion, similar to those of the CP
N1
sigma model.
20
These theories arise via M theory on local fourfolds with G ux [10].
27
4.6. U(1)
0
with N
f
= N
+
= N

= 2, with nonzero real masses

We recall this case [3] to illustrate the possibility of a moduli space with multiple
components, both compact and non-compact. Taking m
Q
1
= m
Q
1
= m
Q
2
= m
Q
2
=
m > 0, and = 0, the moduli space has two singular points, = m, each of which has
a low-energy U(1)
k=0
with N
f
= 1 light avor, i.e. the W = MX
+
X

SCFT. The moduli

space is similar to Fig. 1. The theory is dual to [3]
W = M
1
1
X
1+
X
1
M
2
2
X
2+
X
2
+(X
1+
X
2
1), (4.16)
with a Lagrange multiplier. X
1
and X
2+
parameterize the Coulomb branches that look
like non-compact cones. The compact /
C

= CP
1
region in between has local coordinates
X
1+
and X
2
near the two points where the CP
1
intersects the non-compact Higgs and
Coulomb branches (M
1
1
and X
1
, or M
2
2
and X
2+
, respectively).
Since the low-energy theory at = m and = +m are each a copy of U(1)
0
with
N
f
= 1, each has corresponding Higgs branch BPS vortices for ,= 0. As 0, the
vortices on the M
1
1
Higgs branch can condense to give the X
1
Coulomb branch moduli.
Likewise, vortices on the M
2
2
branch can condense to give the X
2
moduli. On the other
hand, the chiral monopole operators are dened via (1.3) independent of the vacuum,
which leads to (1.4), , so always corresponds to non-compact Coulomb regions;
specically here, X
1
or X
2+
.
We can also consider Skyrmion solitons wrapping the compact CP
1
part of the
Coulomb branch, though the local theory at = m is not smooth, but contains the
additional degrees of freedom of the interacting SCFT there. This soliton has the quan-
tum numbers of M
2
1
or M
1
2
.
5. Generalities about the Witten index
We here recall a few standard aspects about the Witten index [26], I = Tr(1)
F
of SUSY theories in D spacetime dimensions. The index can be computed by reducing
the theory to SQM on T
D1
R, or from the partition function on Euclidean T
D
, with
periodic boundary conditions for the Fermions in every S
1
. The index is ill-dened if the
theory on R
1,D1
has a continuous, non-compact moduli space of vacua. Such vacua are
superselection sectors of the theory in D 3 non-compact spacetime dimensions, but are
integrated over when the theory is put on T
D1
, leading to an ill-dened Tr(1)
F
.
28
5.1. The index for 3d SUSY CS theories with no matter
Consider rst pure, A = 0 Chern-Simons theory, with gauge group G and Chern-
Simons coecient k, without the Maxwell term, gauginos, or matter, on a spatial torus
T
2
. This theory has a moduli space of at connections that are zero energy states and,
upon quantization, are related [37,60] to the conformal blocks of the 2d G
k
WZW theory.
The number of vacua is thus J
G
(k) the number of conformal blocks of G
k
WZW theory.
This is also the number of vacua of the theory with the Maxwell term added, since the
Maxwell term is irrelevant in the IR compared with the Chern-Simons term.
The A = 1 supersymmetric version of this theory, without matter, was analyzed in
[28]. The result is that the vacua counted by J
G
(k) count the supersymmetric groundstates,
once one accounts for a shift of the level from integrating out the gauginos
21
:
Tr(1)
F
I(k) = J
G
([k

[), [k

N=1
[ = [k[
1
2
h, [k

N=2
[ = [k[ h, (5.1)
(h = N for SU(N)). The index of pure A = 2 gauge theory without matter is then
Tr(1)
F
= J
G
([k[ h), (5.2)
with supersymmetry broken
22
for [k[ < h.
For U(1)
k
, the number of conformal blocks is J
U(1)
([k[) = [k[, as seen by bosonizing
the U(1), with k vertex operators V
q
= e
iq/

k
, q = 0, . . . , k 1. So (5.2) gives
U(1)
k
with no matter : Tr(1)
F
=
_
[k[ for k ,= 0
ill dened for k = 0.
(5.3)
For k ,= 0, there is a mass gap and the single classical SUSY vacuum at = 0 acquires a
topological multiplicity of [k[. For k = 0, the index is ill-dened, because of the unlifted
Coulomb branch.
For SU(N
c
)
k
, the number of SU(N
c
)
k
conformal blocks is
J
SU(N
c
)
(k

) =
(N
c
+k

1)!
(N
c
1)!k

!
. (5.4)
21
There is also a +h contribution to the WZW level from integrating out gluons, but it was
stated in [28] that such bosonic contributions are already accounted for, without shifting k

. This
subtlety was debated and analyzed in detail in [61] and references therein. The upshot is that the
original result of [28] is correct: only the Fermionic shifts to k

should be included. See also [62].

22
For N = 1, SUSY is broken for |k| <
1
2
h [28].
29
For example, J
SU(2)
(k

) = k

+ 1, from the SU(2)

k
representations with j = 0, . . .
1
2
k

.
Accounting for the shift (5.2), this gives for the index
A = 2 SU(N
c
)
k
: I(k) =
_
_
_
(|k|1)!
(N
c
1)!(|k|N
c
)!
for [k[ N
c
0 (gapped) for 0 < [k[ < N
c
0 (runaway) for k = 0.
(5.5)
For [k[ = N
c
, strictly speaking, the above argument does not work (k

= 0 incorrectly
suggests the absence of a gap), but the result (5.5) is nevertheless applicable, giving the
correct (as we will show) result that Tr(1)
F
= 1 in this case. For 0 < [k[ < N
c
the
vanishing index (along with our results in the following sections) is consistent with super-
symmetry being broken in a stable vacuum, with a massless Goldstino, and a mass gap for
the other elds. For k = 0 there is the non-gapped runaway, with no stable vacuum, as in
the SU(2) case [2].
6. Computing the Witten index for U(1)
k
with matter
We now compute the Witten index for U(1) gauge theory, with CS term k, with matter
elds Q
i
of charge n
i
, and generic real parameters, m
i
and . The vacuum solutions of
(2.8) are then isolated, with a mass gap: the Higgs and topological vacua, which contribute
U(1)
k
eff
topological : Tr(1)
F
[
=
I
= [k
eff
(
I
)[
Q
i
,= 0 Higgsed : Tr(1)
F
[
=
Q
i
= s
i
n
2
i
,
(6.1)
see (2.9). The topological vacua, at =
I
where F(
I
) = 0 (2.10), have [k
eff
[ vacua,
the topological multiplicity (5.3) of the low-energy U(1)
k
eff
theory with no matter. The
Higgsed vacua, at =
Q
i
(2.5) where m
Q
i
() = 0, exist if s
i
= 1 in (2.9), as then
eff
has the right sign for Q
i
,= 0. Now Q
i
breaks U(1) Z
n
i
, with n
i
the U(1) charge of
Q
i
, and the associated Z
n
i
orbifold leads to a Higgs vacua multiplicity
23
of n
2
i
.
The total index is obtained by summing the contributions (6.1) over all the vacua
Tr(1)
F
=

I
[k
eff
(
I
)[ +

Q
i
s
i
n
2
i
. (6.2)
The individual terms (6.1) generally depend on the real parameters, m
i
and , but the sum
(6.2) independent of the real parameters. As we vary m
i
and , the vacua
I
and
Q
i
can
23
In D spacetime dimensions, a Z
n
orbifold gives a n
D1
multiplicity, from the twists on T
D1
.
30
move around and collide, leading to phase transitions. We have already given a general
argument, in the introduction, that Tr(1)
F
must be invariant under deformations of the
real parameters m
i
and , since they become complexied upon compactifying on S
1
R
.
We will explicitly verify the m
i
and independence of (6.2), and evaluate the sum, in
appendix D. The nal result is (in terms of the critical k
c
from (2.14))
Tr(1)
F
= [k[ +
1
2

i
n
2
i
for [k[ [k
c
[, (6.3)
Tr(1)
F
= [k
c
[ +
1
2

i
n
2
i
for [k[ [k
c
[. (6.4)
These results apply whenever the index is well-dened, i.e. whenever there is no non-
compact moduli space of vacua. Non-compact moduli spaces occur on the Higgs branch,
if two
Q
i
, with opposite sign charges n
i
, are tuned to coincide. Such moduli spaces also
occur on the Coulomb branch if k = k
c
and tuned , as in (4.1). It is interesting that the
results (6.3) and (6.4) are independent of m
i
and even in cases where they cross through
Tr(1)
F
ill-dened locations, since vacua could have there moved in or out from innity.
As an illustrative example of (6.2) yielding (6.3) or (6.4), consider U(1)
k
with N elds
Q
i
of charge +1. It suces to set all real masses m
i
= 0, with the FI parameter varying.
If > 0, there is a compact Higgs branch moduli space of SUSY vacua, /
H

= CP
N1
,
with index given by the Euler character, while for < 0 there are no Higgs vacua, so
Tr(1)
F
[
Higgs
=
1
2
(1 + sign())(CP
N1
) =
1
2
N(1 + sign()). (6.5)
Topological vacua exist at
I
= /k
eff
if sign(
I
) ts with the sign of the shift in k
eff
:
U(1)
k
1
2
N
:
I
=

k
1
2
N
, if

k
1
2
N
> 0. (6.6)
For [k[
1
2
N, one or the other of the solutions in (6.6) exist, depending on sign(). For
[k[ < N/2, both exist for < 0, and neither exists for > 0. In sum, this gives
Tr(1)
F
[
topological
=

[k
eff
[ =
_
[k[
1
2
N sign() for [k[
1
2
N
1
2
N(1 sign()) for [k[
1
2
N
. (6.7)
The separate contributions (6.5) and (6.7) depend on sign(). Indeed, at = 0 the
topological and Higgs vacua locations collide,
Q
i
=
I
= 0, so the individual vacua
can there become rearranged. Their total is properly independent of all real parameters,
31
with sign() canceling between (6.5) and (6.7): Tr(1)
F
= [k[ +
1
2
N for [k[
1
2
N, or
Tr(1)
F
= N for [k[
1
2
N, agreeing with (6.3) and (6.4). For k = k
c
=
1
2
N, and
=
crit
= 0, there is an unlifted Coulomb branch, so Tr(1)
F
is ill-dened. Again, we
note that Tr(1)
F
nevertheless takes the same value for >
crit
and <
crit
.
Finally, note that vector-like matter, i.e. pairs Q
i
and

Q
i
with charges n
i
, do not
cancel in the index (6.3) and (6.4). This might seem strange, since one could decouple
such matter with a complex mass term W = m
C
Q
i

Q
i
, and Tr(1)
F
is invariant under
localized, continuous deformations. The point, again, is that real vs. complex masses are
mutually exclusive if SUSY is not explicitly broken. Since (6.3) and (6.4) apply for generic
real masses, they cannot be continuously connected with m
C
,= 0, while keeping Tr(1)
F
well dened. The only option is to go through m
i
= m
C
= 0, where the theory has an
unlifted Higgs branch moduli space of vacua, so Tr(1)
F
is ill-dened, and vacua can
move in or out from innity upon taking m
C
,= 0. The existence of real masses is special
to odd spacetime dimensions, which ts with the index contribution scaling as n
D1
i
, so
vector-like matter contributions do not cancel for D odd.
7. Comparing the index Tr(1)
F
of dual theories
Obviously Tr(1)
F
must agree between dual theories; we here mention a few examples.
As mentioned in the introduction, any gauge theory that is dual to a Wess-Zumino theory,
without gauge elds, must have Tr(1)
F
= 1. Indeed, if the dual theory has elds
i
with
real masses m
i
(suppressing avor indices),
L
eff
K(

i
e
m
i

i
) V
eff
m
2
i
K
[
i
[
2
[
i
[
2
, (7.1)
with a unique SUSY vacuum at
i
= 0 for smooth, non-degenerate K
eff
(

).
On the gauge theory side, Tr(1)
F
= 1 requires (1.8) [k[ h+
1
2

f
T
2
(r
f
) = 0 (with
[k[ replaced with [k
c
[ for [k[ < [k
c
[). For the example of U(1)
k
gauge theory with N

matter elds of charge n

i
= 1, we found Tr(1)
F
= [k[ +
1
2
N
+
+
1
2
N

(as in (6.4) we
replace [k[ with [k
c
[ for [k[ [k
c
[, and here k
c
=
1
2
(N
+
N

), see (2.14)). So the cases

that can be dual to Wess-Zumino theories are the examples that we have mentioned. One
is U(1)
k
, with [k[ =
1
2
and a single matter eld Q of charge [n
i
[ = 1, which is dual to a
free chiral supereld, X. Another is U(1)
0
with N
f
= 1 avor of elds, Q and

Q, of charge
1. This theory is dual to W = MX
+
X

[3]. The dualities for the two cases discussed

above provide a basis for deriving other duality examples, see e.g. [8].
32
Let us explicitly verify that Tr(1)
F
matches for the duals of [3], between A = 2 U(1)
0
with N
f
avors, and the U(1)
N
f
0
/U(1) quiver, analogous to the A = 4 mirror symmetry
dual of [20]. The U(1)
0
theory, for generic FI parameter and real masses has, according
to (6.3) with k = k
c
= 0, Tr(1)
F
= N
f
. To evaluate the index of the U(1)
N
f
0
/U(1)
mirror dual, it is convenient to consider rst U(1)
N
f
, and then mod out by the overall
U(1). Integrating out the bifundamentals, the real masses induce k
ab
mag
=
1
2
n
a
n
b
= C
ab
,
where n
a
is a vector with components n
a
i
the charges of bifundamental matter elds, and
C
ab
is the Cartan matrix of the ane SU(N
f
) extended Dynkin diagram. Note that k
ab
mag
has a zero eigenvalue, corresponding to the overall translation. For example, for N
f
= 2
we get k =
_
2 2
2 2
_
, while for N
f
= 3 we get the Cartan matrix of ane SU(3)
k
mag
=
_
_
2 1 1
1 2 1
1 1 2
_
_
. (7.2)
To get the U(1)
N
f
1
theory we need drop the zero eigenvalue, and rescale the remaining
eigenvalues to have properly normalized U(1) factors. For example, for N
f
= 2 the eigen-
values are 0 and 4, but the relative U(1) has N
f
= 2 (this case is self-dual), so we rescale
to k = 2. In general, for all N
f
, the index of the magnetic dual is given by
Tr(1)
F
mag
=
1
N
f
det

k
mag
= N
f
, (7.3)
where det

means to omit the zero eigenvalue, and the factor of 1/N

f
is to have properly
normalized relative U(1) charges. The result (7.3) indeed agrees with (6.3) in the dual.
8. SU(2)
k
gauge theory
Recall rst the case of A = 2 SU(2)
0
theory, with k = 0. There is a classical Coulomb
branch moduli space, where the real adjoint
adj
diag(, ) breaks SU(2) U(1), and
we impose 0 since the SU(2) Weyl symmetry makes and gauge equivalent. The
moduli space can be labeled by the chiral supereld Y e
/g
2
+ia
(analogous to X for the
low-energy U(1)). For the theory without matter, the Coulomb branch is lifted by a 3d
instanton (4d monopole) [2] and the theory has no static vacuum.
SU(2)
0
, N
f
= 0 : runaway W
dyn

1
Y
. (8.1)
33
For the SU(2)
k
theory with matter elds Q
i
, i = 1 . . . 2N
f
, we can form gauge invari-
ant mesons M
ij
= Q
ic
Q
jd

cd
= M
ji
(with c, d the usually-suppressed SU(2) color in-
dices). The classical Higgs branch, for the theory with zero superpotential and real masses,
is given by M
ij
subject to the classical constraint that rank(M) 2. The SU(2)
0
theo-
ries with 2N
f
fundamental matter elds Q
i
, were analyzed in [3], for vanishing real masses
and integer N
f
.
The result for N
f
= 1 is a quantum moduli space of SUSY vacua, merging the Higgs
(with M M
12
) and Coulomb branches:
SU(2)
0
, N
f
= 1 : smooth SUSY /
quantum
= M, Y [ MY = 1. (8.2)
For N
f
= 2, the theory has a dual with chiral superelds M
ij
and Y , with
SU(2)
0
, N
f
= 2 : simple dual W
dual
= Y PfM . (8.3)
The SU(2)
0
gauge theory and its simple dual (8.3) ow to the same interacting SCFT at
the origin. For N
f
> 2, there is an interacting SCFT at the origin, with additional degrees
of freedom beyond M
ij
and Y [3,40,44]. In this section, we generalize the above to allow
for k ,= 0, real masses, half-integral N
f
, and other matter representations.
For generic k and real masses, there are isolated supersymmetric vacua, with a mass
gap. The index for the theory with 2N
f
matter elds in the fundamental is found to be
Tr(1)
F
= [k[ +N
f
1. (8.4)
More generally, for matter elds Q
i
in representations r
i
of SU(2), the index is
Tr(1)
F
= [k[ +
1
2

i
T
2
(r
i
) 1. (8.5)
Supersymmetry is dynamically broken for [k[ +N
f
1: SUSY is broken with a mass gap
for SU(2)
1
with N
f
= 0 and SU(2)
0
with N
f
= 1 and generic real masses. With non-
generic real masses, and general k and matter content, there can be Higgs and/or Coulomb
branches, and interacting SCFTs. The details will be given in the following subsections.
34
8.1. Some generalities
Consider generally SU(2)
k
with matter elds Q
i
in representations r
i
of SU(2). If
the matter elds are given real masses m
i
, they can be integrated out, and the low-energy
theory is SU(2)
k
eff
with no matter, with (writing k = k
SU(2)
here for claritys sake)
k
SU(2),eff
= k
SU(2)
+
1
2

i
T
2
(r
i
) sign(m
i
), (8.6)
where T
2
(r
i
) = 2(2I
i
+1)I
i
(I
i
+1)/3 is the quadratic index if representation r
i
is labeled by
its SU(2) isospin I
i
(i.e. r
i
is the 2I
i
+ 1 dimensional representation), with T
2
(r
fund
) = 1
for the fundamental. The quantization condition on k, required for gauge invariance, is
k
SU(2)
+
1
2

i
T
2
(r
i
) Z. (8.7)
In particular, for 2N
f
fundamentals (8.7) shows that
k
SU(2)
+N
f
Z. (8.8)
The semi-classical SUSY vacua satisfy
(
a
T
a
r
i
+m
i
1
|r
i
|
)Q
i
= 0, D
a

k
eff
2

a
adj
+

i
Q

i
T
a
Q
i
= 0, (8.9)
where
a
adj
=
a
[, a = 1, 2, 3, are the real scalars in the vector multiplet. When
a
,= 0,
the gauge group is broken, SU(2) U(1), and by an SU(2) gauge rotation we can set

1
=
2
= 0,
3
= 0. The low-energy theory for > 0 is U(1)
k
U(1)
, with k
U(1)
rescaled
to account for the normalization of the generators
k
U(1)
= 2k
SU(2)
2k. (8.10)
More generally, for > 0, breaking SU(2)
k
U(1)
2k
, an SU(2) isospin I
i
rep
matter eld Q
i
decomposes into 2I
i
+ 1 U(1) charged matter elds, Q
n
i
i
, where
1
2
n
i

I
i
, I
i
1, . . . , I
i
, with n
i
Z the U(1) charge. These U(1) charged matter elds
have real masses as in (2.4), m
n
i
i
() = m
i
+ n
i
. So there is a massless component
at
Q
n
i
i
= m
i
/n
i
, if this respects the SU(2) requirement that 0. The charge n
i
component can thus be massless if sign(n
i
) = sign(m
i
). In the low-energy U(1)
2k
theory
we dene, as in (2.6),
F()
eff
+k
U(1),eff
= 2k +
1
2

i
I
i

n
i
2
=I
i
n
i
[m
i
+n
i
[. (8.11)
35
This expression is valid only in the semiclassical region e
2
[k[ , where the signs of
the gauginos real mass are determined by and they do not contribute. In particular, a
fundamental matter eld Q with real mass m contributes
F()
fund
=
1
2
([m+[ [m[) =
_
m > [m[
sign(m) < [m[.
(8.12)
Note that k
U(1),eff
2k for ; comparing with (2.14), here the quantity k
c
= 0,
because SU(2) representations are symmetric under n
i
n
i
. Because k
c
= 0, there will
not be an analog of the case (6.4) for the index of SU(2)
k
theories.
The vacua can be found by a semi-classical analysis in the low-energy U(1)
k
U(1),eff
theory, which is a good approximation for suciently large e
2
, i.e. for [m
i
[ e
2
.
Corrections from instantons in the broken SU(2)/U(1) need to be added to the low-energy
U(1) theory.
The quantum vacua are as follows:
1. Higgs vacua, moduli spaces or isolated points. At the origin of the Coulomb branch,

a
= 0, setting m
i
= 0, the elds Q
i
can have non-zero Q
i
/
H
solving (8.9).
The moduli space /
H
is the same as for the corresponding 4d A = 1 theory, which
can be given as the gauge invariant composite chiral superelds, subject to classical
constraints. The space /
H
is unaected by k and unmodied by quantum eects
(aside from the special case (8.2)), and is generally singular at the origin. The theory
at
a
= Q
i
= 0 is typically an interacting SCFT, which depends on the matter content
and k. For m
i
,= 0, there are Higgs branches where some m
i
+n
i
= 0 in (8.9). These
can be analyzed in terms of the low-energy U(1)
k
U(1)
theory and, depending on the
matter content and m
i
, can be non-compact, compact, or isolated points.
2. Coulomb moduli space, when k
eff
= 0, setting Q
i
= 0 and ,= 0. As we will see, this
can occur in non-compact or compact regions. The semi-classical Coulomb branches
can be lifted by an instanton superpotential, in some cases, similar to [2].
3. Topological vacua. For k
eff
,= 0, when the matter elds have real masses and are
integrated out, the low-energy SU(2)
k
eff
theory with no matter [k
eff
[ 1 topological
vacua at = 0. For ,= 0, there can be [k
eff,U(1)
[ additional topological vacua, at
locations
I
where (8.11) vanishes.
36
The global symmetries for a general theory with matter elds Q
i
and real masses m
i
includes U(1)
i
, which acts only on Q
i
, and a U(1)
R
symmetry, with charges
U(1)
j
U(1)
R
Q
i

ij
0
Y K
j
2 +

j
K
j
(8.13)
K
j
can be computed in the low-energy U(1) theory, where Y X
+
, with charges similar
to (2.13) obtained from the induced, mixed Chern-Simons terms (2.12) between the low-
energy U(1) gauge group and the U(1) global symmetries. This (or, equivalently, the
Callias index theorem [63]) gives
K
j
=
1
2
2I
r
j

n=2I
r
j
nsign(n +m
j
), (8.14)
where the

is a reminder that the sum is only over even or odd n, depending on whether
I
r
j
is integer or half-integer. For example, for matter in the fundamental, I =
1
2
,
K
fund
=
1
2
sign( +m)
1
2
sign( +m) = ( [m[). (8.15)
In general, for [[ > [m
j
[, a Fermion in the 2I + 1 dimensional SU(2) representation has
K
j
= I(I +1) for integer I, or K
j
= (I +
1
2
)
2
for half-integer I. The results (8.1) and (8.2)
and (8.3) are compatible with these charges.
A non-compact Coulomb branch exists semi-classically when k
eff
and
eff
vanish for
. Using (8.11), we nd
k
eff
lim

k
eff
= 2k,
eff
lim

eff
=

i
K
i
m
i
, (8.16)
where K
i
is the index (8.14) in the limit where > [m
j
[ for all m
j
,
K
i
= lim

K
i
=
1
2
2I
r
i

n=2I
r
i
[n[. (8.17)
In particular, if all matter is in the fundamental, then K
i
= 1 and (8.16) gives
eff
=

i
m
i
. There is a non-compact Coulomb branch only if k = 0 and
eff,
= 0. The latter
37
condition ts with the real mass assignment for the eld Y , which follows from the U(1)
i
charge assignment of Y in (8.13)
lim

m
Y
=

i
K
i
m
i
=
eff
, (8.18)
so the non-compact Coulomb branch remains unlifted when m
Y
= 0. When k = m
Y
= 0,
the non-compact region of the Coulomb branch cannot be lifted by an instanton, if there
is matter (unlike pure Yang-Mills [2]), since the U(1)
i
and U(1)
R
charges in (8.13) are
incompatible with a holomorphic superpotential if K
i,
,= 0. We will see examples where
some compact semi-classical Coulomb branches are lifted by W
inst
1/Y , which can
happen in regions where all K
j
= 0, as in (8.15) for < [m
j
[.
8.2. The index for general k and matter elds Q
i
For generic k and real masses, the theory has isolated vacua with a mass gap, so
the index is nite and well-dened. The index is then the sum of contributions from the
SU(2)
k
eff
topological vacua at = 0, and any Higgs or topological vacua at ,= 0:
Tr(1)
F
= Tr(1)
F
[
=0
+

i
Tr(1)
F
[

Q
n
i
i
+

I
Tr(1)
F
[

I
= [k[ 1 +
1
2

i
T
2
(r
i
).
(8.19)
The individual terms on the top line depend on the real parameters, while the sum of
course does not. We here briey illustrate that. The = 0 term is from the topological
vacua of the low-energy SU(2)
k
eff
theory:
Tr(1)
F
[
=0
= [k +
1
2

i
T
2
(r
i
) sign(m
i
)[ 1 (8.20)
The
Q
n
i
i
= m
i
/n
i
and
I
terms in (8.19) are from the possible Higgs or topological
vacua in the low-energy U(1)
k
eff
theory:
Tr(1)
F
[

Q
n
i
i
= n
2
i
(m
i
/n
i
)(n
i
F(
Q
n
i
i
)), (8.21)
Tr(1)
F
[

I
=

dF()
d

, (8.22)
where F() is as in (8.11) and
I
are the topological vacua, where F(
I
) = 0.
38
The fact that (8.19) is independent of the values of m
i
can be proved by compactifying
on an S
1
and using holomorphy in the twisted chiral parameters, as mentioned in the
introduction. Or it can be proved directly, much as in the U(1) case, discussed in appendix
D. We can thus evaluate the index for a convenient choice of m
i
. We use the P symmetry to
take k > 0, and choose all m
i
> 0; in the end, we can replace k [k[ for generality. With
this choice, the index comes entirely from the vacua at the origin (8.20). Indeed, using
(8.11) here yields F() > 0 for all > 0, so there are no vacua
I
,= 0 where F(
I
) = 0,
and there are also no Higgs vacua, since only negative charge n
i
< 0 components Q
n
i
i
become massless, and F() has the wrong sign for Q
n
i
i
,= 0. So the topological vacua at
= 0 yields the result (8.5), which is the complete, general answer.
If some sign(m
i
) ,= sign(k), some Higgs or topological vacua compensate for the dif-
ference, leading again to (8.5). As an illustration consider a theory with 2N
f
fundamental
avors, with k > 0 and all real masses m
i
= m < 0. Then the vacua at the origin contribute
Tr(1)
F
[
=0
= [k N
f
[ 1. (8.23)
To look for the Higgs and topological vacua away from the origin, we consider F() (8.11),
which using (8.12) here gives
F() =
_
2k + 2N
f
m > [m[
2(k N
f
) < [m[
. (8.24)
The low-energy U(1) theory has 2N
f
massless matter elds, with n
i
= +1, at = [m[.
These give a CP
2N
f
1
compact Higgs branch if F[
=|m|
> 0, and no SUSY vacua otherwise:
Tr(1)
F
[
=|m|
= 2N
f
(k N
f
). (8.25)
Finally there are 2k topological vacua at = N
f
[m[/k, if N
f
/k > 1:
Tr(1)
F
[
=N
f
|m|/k
= 2k(N
f
k). (8.26)
Adding (8.23) and (8.24) and (8.25) indeed gives the index (8.4), Tr(1)
F
= [k[ +N
f
1.
If we had taken m > 0, all of these vacua would have instead been at the origin.
This illustrates a point about decoupling matter elds via large [m[. If the sign of m is
the same as the remaining k
eff
, the additional vacuum or vacua associated with the heavy
matter eld remain at the origin, staying in the low-energy theory via increased [k
eff
[. If
the sign of m is opposite to the remaining k
eff
, the additional vacuum or vacua associated
with the heavy matter runs o to innity,
I
heavy
= O([m[), e.g. as in (8.25) or (8.26).
39
8.3. SU(2)
k
with N
f
=
1
2
, a single fundamental matter doublet, Q
The condition (8.8) requires k Z +
1
2
. The global symmetry is U(1)
Q
U(1)
R
. Qs
real mass m is a background eld for U(1)
Q
. By P symmetry (m m and k k),
we could always restrict to m > 0, or k > 0, but we will not do so for the moment. This
theory does not have a Higgs branch at = m = 0, because there is no non-trivial Q ,= 0
solution of the SU(2) D term equations (8.9); this ts with the fact that no gauge invariant
chiral operator can be formed from a single Q eld.
First consider the isolated Higgs and topological vacua for m ,= 0. At = 0, we can
integrate out Q, to obtain a low-energy SU(2)
k+
1
2
sign(m)
theory without matter; so
Tr(1)
F
[
=0
= [k +
1
2
sign(m)[ 1 (8.27)
topological SUSY vacua, with a mass gap. The cases k =
1
2
will be discussed separately.
For ,= 0, breaking SU(2)
k
U(1)
2k
, the low-energy eective U(1) theory has the
Q components Q

, with mass m

= m . When [m[ e
2
, we can analyze the theory
semi-classically in the low-energy U(1) theory, with (8.11) given by
F()
eff
+k
eff
= 2k +
1
2
[m+[
1
2
[m[ . (8.28)
It follows from (8.21) that there is a Higgs vacuum at
Q
= [m[ if F([m[) < 0 for m > 0,
or if F([m[) > 0 for m < 0. There are no topological vacua
I
,= 0 to contribute to (8.22),
since F() has no non-trivial zeros. The total index is then given by the result (8.4) for
N
f
=
1
2
,
Tr(1)
F
= [k[
1
2
. (8.29)
The cases k =
1
2
dynamically break SUSY, tting with Tr(1)
F
= 0 (8.29). The
case k = +
1
2
sign(m) leads to a mass gap, k
eff
= k +
1
2
sign(m) = 1, and none of the
topological nor Higgs vacua exist in this case, since [k
eff
[ = 1 and F([m[) has the wrong
sign for Q ,= 0. The case k =
1
2
sign(m) is more involved, since k
eff
= 0, and thus
a semi-classical Coulomb branch, for [0, [m[], which gives /
S.C.
Coulomb

= CP
1
upon
including the dual photon. The region > m is lifted by a potential, since the low-energy
U(1)
eff
theory has k
eff
= 2k
SU(2)
,= 0 (and
eff
= m).
The semi-classical /
S.C.
Coulomb

= CP
1
for k =
1
2
sign(m) is lifted by an SU(2) in-
stanton superpotential, W
dyn
= 1/Y , where Y is a local coordinate on the CP
1
. This
W
dyn
is generated as in the N
f
= 0 case [2] and is compatible with the symmetries (8.13),
40
because the instanton does not have a Q Fermion zero mode for < [m[ (see (8.15)).
Unlike the N
f
= 0 case, here there cannot be a runaway to innite distance on the moduli
space, because CP
1
is compact: W
dyn
pushes Y to the end where semiclassically [m[
(1/Y 0) where there is a stable vacuum, with SUSY broken by the small non-zero
vacuum energy from W
dyn
(and the global symmetries unbroken).
The case k = 3/2 is also interesting, since (8.29) then gives Tr(1)
F
= 1. For
k = 3/2 and m > 0, the vacuum is at = 0, while for k = 3/2 and m < 0 the vacuum is
at = O([m[).
8.4. N
f
= 1: matter elds Q
1
and Q
2
.
The global symmetry group is SU(2)
F
U(1)
Q
U(1)
R
. Consider rst the theory
with all masses set to zero. For all k, there is a Higgs branch, with M arbitrary, where
M = Q
1
Q
2
is an SU(2)
F
singlet, with U(1)
Q
charge equal to 2. For k = 0, the quantum
moduli space is the smooth space MY = 1 (8.2), while for k ,= 0 there is no Coulomb
branch Y modulus. For [k[ = 1 and zero mass, as we will see via added matter and owing
down, the theory at the origin is an IR-free eld theory, consisting of the chiral supereld
M, with W = 0. For [k[ > 1, the theory at M = 0 is an SCFT labeled by k.
We now turn on real masses, m
1,2
for Q
1
and Q
2
; m
1
+ m
2
is a background gauge
supereld of the U(1)
Q
symmetry, and m
1
m
2
is a background for U(1) SU(2)
F
.
Without loss of generality we can take m
1
[m
2
[. The modulus M has real mass m
M
=
m
1
+ m
2
. So if m
1
+ m
2
,= 0, the Higgs branch is lifted by m
M
,= 0. For m
1
= m
2
,
m
M
= 0 and the Higgs branch is unlifted, for all k. For generic m
1
and m
2
, there is a mass
gap and (8.4) gives Tr(1)
F
= [k[. In particular, for k = 0, SUSY is broken for generic
m
i
. Let us illustrate some aspects in more detail.
We consider rst the semiclassical theory with
e
2
[m
1
[, [m
2
[ . (8.30)
The interesting points on the Coulomb branch are = 0, [m
1
[, [m
2
[. Integrating out the
massive matter gives an SU(2)
k
eff
theory at = 0, with, see (8.6),
k
eff
= k +
1
2
(1 + sign(m
2
)). (8.31)
There can be semi-classical Coulomb vacua when k
eff
= 0, i.e. k = 0 with m
2
< 0, or
k = 1 with m
2
> 0. If k
eff
,= 0, the IR theory at = 0 has a mass gap, with [k
eff
[ 1
41
SUSY vacua, and no SUSY vacua for [k
eff
[ 1. For ,= 0, we can use the low-energy
U(1) eective theory (with instanton corrections), with F() given by (8.11) and (8.12),
F() = k
eff
+
eff
=
_
(2k + (1 + sign(m
2
))) 0 < < [m
2
[
(2k + 1) +m
2
[m
2
[ < < m
1
2k +m
1
+m
2
m
1
<
(8.32)
Consider the k = 0 case. The low energy theory around = 0 is SU(2) 1
2
(1+sign(m
2
))
with N
f
= 0, which does not have a SUSY vacuum. It is clear from (8.32) that for
k = 0 there are no
I
,= 0 topological vacua, where F(
I
) = 0. There are also no Higgs
vacua: the massless eld at = m
1
has charge 1, and F(m
1
) = m
1
+ m
2
> 0 has the
wrong sign for it to have an expectation value. At = [m
2
[, the massless eld has charge
sign(m
2
) and the low-energy theory is U(1)
1+
1
2
sign(m
2
)
, and shifting to vanish there
leads to = m
2
+[m
2
[. There is no Higgs vacuum there since either vanishes or has the
wrong sign for the massless charged matter to get a Higgs vev. In sum, the k = 0 theory
with generic real masses breaks SUSY.
Now consider k = 0 for m
1
= m
2
. For m
1
= m
2
, the low energy theory around
= m
1
is U(1)
1
with F() = 2m
1
and two massless chiral superelds with charge 1.
This theory breaks supersymmetry, since F() has the wrong sign for satisfying the D-
term equations with nonzero charged matter vev. For m
1
= m
2
, on the other hand, the
low energy theory around = m
1
is U(1)
0
with = 0 and N
f
= 1 massless avor of chiral
superelds of charge 1. This low-energy theory has a moduli space of SUSY vacua with
three non-compact branches, two Coulomb and one Higgs, meeting at = m
1
, i.e. it is
the W
low,U(1)
= MX
+
X

theory.
We now consider nonperturbative corrections from SU(2). For m
1
+ m
2
,= 0 the
potential is everywhere nonzero, except perhaps at = 0, where we know that SU(2)
0
has
a runaway to large and in SU(2)
1
supersymmetry is broken. So the m
1
+m
2
,= 0 theory
breaks supersymmetry; we will momentarily give an independent derivation.
Now consider the SU(2) nonperturbative corrections for m
1
= m
2
, where we have
seen that the low-energy U(1) theory near = m
1
is the W
low,U(1)
= MX
+
X

theory.
The low-energy SU(2)
0
theory = 0 gives a W
inst
correction, coinciding with that of [2]
since the low-energy SU(2)
0
theory has no light matter: W
inst
= 1/Y . Now the Coulomb
branch modulus Y at = 0 is related to the modulus X

at = [m
2
[ by Y X

= 1;
this is exactly as in the identication in [3] in the context of Fig 2 there, written there as
42
V
i+
V
i+1
= 1. The instanton of [2] thus generates a term W
inst
= X

in the low-energy
theory at = m
1
, leading to the dual theory there (as also briey discussed in [3]):
W = MX
+
X

. (8.33)
The F-terms of (8.33) set X

= 0 and MX
+
= 1. So the low energy theory has a smooth
moduli space of vacua, MX
+
= 1, and therefore Tr(1)
F
is ill-dened.
Let us see how we can derive the same results in the limit opposite to (8.30)
e
2
[m
1
[, [m
2
[ . (8.34)
We start with the smooth moduli space of N
f
= 1 with m
i
= 0, MY = 1 [3], and turn on
the real masses m
i
as a small perturbation. Since the global SU(2) symmetry does not act
on the low energy theory, only m
1
+ m
2
aects it. In particular, if m
1
+ m
2
= 0 the real
mass has no eect and we end up with the same smooth one dimensional moduli space of
vacua MY = 1. This result is in accord with our conclusion in the opposite limit (8.30).
For m
1
+m
2
,= 0, there is a background expectation value for U(1)
Q
, which leads to
a real mass on the MY = 1 moduli space of vacua, m
M
= m
Y
m
1
+ m
2
, since the
U(1)
Q
charge of M is 2, and that of Y is 2. This breaks supersymmetry, reminiscent of
the quantum moduli space DSB models in 4d [34,64], with the classical supersymmetric
vacuum at the origin eliminated by the quantum constraint. Using the asymptotic form
of the Kahler potential
K
_
[M[ =
1
|Y |
[Y [ 0
(log(Y ) + log(Y ))
2
[Y [
(8.35)
we learn that the potential is V = K(Y e
m
Y

Y )[

2 = m
2
Y
K
Y Y
[Y [
2
:
V
_
(m
1
+m
2
)
2
[M[ =
(m
1
+m
2
)
2
|Y |
[Y [ 0
(m
1
+m
2
)
2
[Y [
(8.36)
and it is never zero. Since, before turning on the real masses, the moduli space was
smooth, and the symmetry associated with m
1
+m
2
is everywhere broken, supersymmetry
is necessarily broken when m
1
+ m
2
,= 0. This breaking can be associated with one or
several metastable states at nite Y . The Fermion component of the MY = 1 modulus
remains massless, playing the role of the Goldstino. For Y there is a classically
massless pseudomodulus superpartner, seen from the classical at direction in (8.36). To
43
determine whether the potential has a stable minimum, or a Y runaway, we now
determine the leading quantum correction to K in (8.35) for Y .
The one-loop correction to the Coulomb branch metric can be written as in [65,4]
ds
2
=
1
4
_
1
e
2
+
s

_
d
2
+
_
1
e
2
+
s

_
1
da
2
(8.37)
where s = N
f
3 for A = 2 SUSY SU(2) with 2N
f
doublets, see e.g. [66,67] and references
therein. The chiral supereld = log Y and the linear multiplet are related by a
Legendre transform [45,4,46], as reviewed in appendix B, which here implies that
K
Y Y
[Y [
2
= K

( + ) =
_
( + )

_
1
=
_
1
e
2
+
s

_
1
. (8.38)
The potential V = m
2
Y
K
Y Y
[Y [
2
thus has slope with sign(dV/d) = sign(s), pushing
toward the origin for s > 0, or away from the origin for s < 0. In the present context,
SU(2) with N
f
= 1, s = 2 < 0, so the potential is a runaway, [Y [ . For small m,
we can regard this as a small correction to the MY = 1 quantum moduli space. Then
Y means the runaway vacuum has M 0.
The result that the m
1
+m
2
,= 0 theory breaks supersymmetry is in accord with our
conclusion in the opposite limit (8.30), and it shows that SU(2)
1
without matter breaks
supersymmetry, consistent with Tr(1)
F
= [k[ 1, vanishing for k = 1. This is also
compatible with our result that SU(2)

1
2
, with N
f
=
1
2
breaks supersymmetry, since we
can ow to that case by decoupling say Q
2
, by taking [m
2
[ [m
1
[.
In conclusion, SU(2)
0
with N
f
= 1 and m
1
+ m
2
,= 0 breaks supersymmetry. For
m
1
+m
2
= 0, there is moduli space of SUSY vacua, MY = 1, with Kahler potential (8.35).
8.5. N
f
=
3
2
: matter elds Q
i=1,2,3
, with k Z +
1
2
For zero mass, there is a Higgs branch labeled by M
i
=
ijk
Q
j
Q
k
, in the 3 of the global
SU(3)
F
. The theory at the origin of this Higgs branch is an interacting SCFT, labeled by
k, for [k[ >
1
2
(it is perturbative for large k). For k =
1
2
, on the other hand, the theory
at the origin is a free eld theory of the unconstrained chiral superelds M
i
. This will be
justied in the following subsection, upon giving a real mass to RG ow from N
f
= 2, with
k = 0, to N
f
=
3
2
, with [k[ =
1
2
. A check is that, for generic real masses, we get for the
index Tr(1)
F
= [k[ +
1
2
SUSY vacua, so the k =
1
2
theory has Tr(1)
F
= 1, matching
that of a dual theory of free chiral superelds (1.8). The SU(3)
F
global symmetry has,
44
in terms of the original Q elds, background CS term k
SU(3)
Z. The theory of the M
i
elds contributes k
SU(3)
F
Z+
1
2
, so parity anomaly matching implies that the low-energy
theory requires an added contribution k
SU(3)
=
1
2
, which can be induced by the RG ow,
as in [39], because the original SU(2) theory anyway has parity violating k
SU(2)
,= 0.
We now consider the SU(2) 1
2
theory, turning on various real masses m
i
for the matter
elds. The location of the vacua depend on the sign of the m
i
. For example, if all m
i
> 0,
the SUSY vacuum contributing to Tr(1)
F
= 1 is at the origin, from the low-energy
SU(2)
2
theory. If, on the other hand, all m
i
= m < 0, the low-energy theory at the
origin is SU(2)
1
, which does not have a SUSY vacuum; the SUSY vacuum is at = 3[m[
and runs o to innity as [m[ . The location of the vacuum is not evident in the
dual description of the free M
i
elds, which gives a supersymmetric vacuum at M
i
= 0
regardless of the sign of the m
i
.
Consider the SU(2) 1
2
theory with a real mass m
3
for Q
3
, with m
1
= m
2
= 0. This gives
real mass to M
1
and M
2
, leaving M = M
12
= M
3
massless, so there is a Higgs branch
labeled by M, and the low-energy theory near the origin is SU(2) 1
2
(1+sign(m
3
))
with
N
f
= 1 massless avor. This low-energy theory is thus qualitatively dierent depending
on the sign(m
3
). For m
3
> 0, the low energy SU(2)
1
has no Coulomb branch, and there
are no other vacua for ,= 0. For m
3
< 0, the low-energy theory at the origin is SU(2)
0
with N
f
= 1, so there is a classical Coulomb branch, with 0 [[ [m
3
[. The low-energy
U(1) theory for ,= 0 has
F() = +
1
2
([m
3
+[ [m
3
[) =
eff
+k
eff
(8.39)
so there is a compact Coulomb branch, where F() = 0, for [[ < [m
3
[; the Coulomb
branch is lifted for [[ m
3
, where F() = + m
3
. Classically, there is thus a compact
Coulomb branch which is topologically a CP
1
. The end near [m
3
[ is smooth, since the
low-energy U(1) theory near = m
3
is U(1) 1
2
with a single charge +1 matter eld, dual
to a free-eld theory. The Coulomb modulus Y has R(Y ) = 0 (Q
1,2
have a Fermion zero
mode (8.15) K
1,2
= 1, while Q
3
has K
3
= 0), so W
dyn
(Y ) = 0. The low-energy N
f
= 1
theory has the smooth, quantum-deformed moduli space (8.2), MY = 1, with Y
corresponding to [m
3
[.
In conclusion, the theory with nite m
3
has a smooth moduli space parameterized
by M = M
12
. As m
3
the point M = 0 is pushed to innite distance. In the dual
description, this detailed structure is simply replaced with a free eld M. The Kahler
45
potential information about the distance to, or location of M = 0 is not immediately
apparent in the dual description.
Now consider SU(2) 1
2
for m
2,3
,= 0, with m
1
= 0. If m
2
+m
3
= 0, there is an unlifted
Higgs branch moduli space of supersymmetric vacua, labeled by the massless eld M
1
. For
m
2
+ m
3
,= 0, the low energy theory has a mass gap. For m
2
and m
3
both positive, the
low-energy theory near the origin is SU(2) 3
2
with N
f
=
1
2
, which has Tr(1)
F
= 1 SUSY
vacuum (8.29). For m
2
and m
3
both negative, the low-energy theory near the origin is
SU(2)

1
2
, which does not have a SUSY vacuum. The location of the SUSY vacuum can
be seen from the low-energy theory on a hypothetical Coulomb branch, using (8.11)
k
eff
+
eff
F() = +
1
2
([m
2
+[ [m
2
[) +
1
2
([m
3
+[ [m
3
[). (8.40)
For m
2
and m
3
of opposite signs, the low-energy theory near the origin is SU(2) 1
2
, which
again does not have a SUSY vacuum. Taking the decoupling limit of large m
2,3
, the SUSY
vacuum is near the origin only if m
2
and m
3
are both positive, but in every other case the
SUSY vacuum should decouple with large m
2,3
.
Now consider SU(2) 1
2
for m
1,2,3
all non-zero and generic. The theory has a mass gap
and Tr(1)
F
= 1 SUSY vacuum. In terms of the low-energy theory of the M
i
elds, the
real masses are m
M
i

j=i
m
j
, e.g. M
3
is massless if m
1
+m
2
= 0, and there is a mass
gap if all m
M
i


j=i
m
j
,= 0. Note that the SUSY vacuum must run o to innite
distance for [m
1
[ [m
2
[ [m
3
[ , since we can there decouple Q
3
and connect to the
low-energy SU(2) 1
2
(1+sign(m
3
))
eective theory, with N
f
= 1, which does not have a SUSY
vacuum for m
1
+ m
2
,= 0. Again, these Kahler metric details are less evident in the dual
description in terms of the M
i
elds.
8.6. N
f
= 2 avors: Q
i=1...4
, with k Z.
For m
i
= 0 and any k, there is a Higgs branch labeled by M
ij
= Q
i
Q
j
, subject to the
classical constraint PfM = 0; the theory at M
ij
= 0, for all k, is an interacting SCFT. The
k = 0 theory has the dual (8.3), which matches the moduli space and the parity anomalies
of the original SU(2)
0
theory [3]. For generic real masses, the theory with general k has
index (8.4), Tr(1)
F
= [k[ + 1. So the k = 0 theory has Tr(1)
F
= 1, which is an
additional check of the duality of (8.3).
In the rest of this subsection, we consider the k = 0 theory, with various non-zero real
masses m
i
for the Q
i
, owing down to the theories considered in the previous subsections.
46
Without loss of generality, we can take m
4
> 0 and m
4
[m
3
[ [m
2
[ [m
1
[. In terms of
the dual (8.3), the masses are
m
M
ij = m
i
+m
j
, m
Y
= (m
1
+m
2
+m
3
+m
4
), (8.41)
as seen from matching the global symmetries.
Using (8.12), the low-energy theory on a hypothetical Coulomb branch has
F()
eff
+k
eff
=
_

_
m
1
+m
2
+m
3
+m
4
> m
4
m
1
+m
2
+m
3
+ m
4
[m
3
[
m
1
+m
2
+(1 + sign(m
3
)) [m
3
[ [m
2
[
m
1
+(1 + sign m
3
+ sign m
2
) [m
2
[ [m
1
[
(1 + sign(m
3
) + sign(m
2
) + sign(m
1
)) [m
1
[ 0.
(8.42)
The asymptotic > m
4
region of the Coulomb branch has
eff
=

4
i=1
m
i
and is lifted
unless
eff
= 0; this agrees with the dual theory (8.41), since m
Y
=
eff
.
Consider rst the case m
4
> 0, with m
1
= m
2
= m
3
= 0. Using (8.42), the Coulomb
branch is always lifted, either by
eff
= m
4
for > m
4
, or by k
eff
= 1 for < m
4
. This
ts with m
Y
= m
4
,= 0 in (8.41). The Higgs branch moduli M
i4
are also lifted by m
4
,= 0.
The remaining massless moduli in the low energy theory are M
k
=
ijk
M
[ij]
=
ijk
Q
[i
Q
j]
(with i, j, k = 1, 2, 3) in the 3 of the unbroken SU(3)
F
. In terms of the dual (8.3), the elds
M
i4
and Y are massive and can be integrated out, and the low-energy theory consists of
the M
i
elds, with W = 0. This derives the dual of the N
f
=
3
2
theory, discussed in the
previous subsection, from the N
f
= 2 dual (8.3). Note that integrating out M
i4
from the
N
f
= 2 dual (8.3) induces the background Chern-Simons term for the unbroken SU(3)
F
global symmetry, k
FF
=
1
2
, which we saw in the previous subsection was needed as an
additional contribution in the low-energy theory of the M
k
elds, to match the SU(3)
F
parity anomaly k
FF
Z of the original Q elds. Note also that the SU(2)
0
theory with
N
f
= 2 and the SU(2) 1
2
theory with N
f
=
3
2
have the same index, Tr(1)
F
= 1, so no
vacua need to move out or in from innity in the m
4
decoupling limit.
We now consider the SU(2)
0
theory with m
3,4
,= 0, with m
1,2
= 0. For general
m
3
+ m
4
,= 0, only M
12
remains massless in (8.41); the remaining M
ij
, and Y become
massive. In particular, (8.42) shows that the region of the Coulomb branch is
lifted by = m
3
+ m
4
, if non-zero. For m
3
+ m
4
= 0, on the other hand, Y and M
34
become massless. More generally, M
12
, M
34
and Y are all massless when m
3
+ m
4
=
47
m
1
+ m
2
= 0. Setting the other, massive elds to zero in the superpotential (8.3), the
low-energy superpotential for the light elds in the dual theory in these cases is
W
low
= Y M
12
M
34
, (8.43)
which looks similar to the W = MX
+
X

dual of N
f
= 1 SQED it ows to a nontrivial
xed point, which is often referred to as the XY Z-theory.
The dual (8.43) in this case seems at odds with the semi-classical picture of the moduli
space, from the electric variables and (8.42). The semi-classical picture on the electric
side suggests that there are two distinguished points on the Coulomb branch, = [m
2
[
and = m
4
, where the two Higgs branches have their roots. These points separate the
Coulomb branch to three distinct branches: 0 < [m
2
[, [m
2
[ m
4
and > m
4
. If
we consider [m
i
[ e
2
, the low-energy theory near both = [m
2
[ and = m
4
is a copy of
N
f
= 1 SQED, so the situation looks similar to (4.16).
In fact, instanton eects modify this semi-classical picture, and the result is perfectly
compatible with (8.43). First, an instanton leads to W
dyn
,= 0 in the region < [m
3
[, as
in [2], since there the instanton does not have any matter Fermion zero modes (see (8.15)).
Next, the Coulomb branch region m
4
> > [m
3
[ gets quantum-merged, as in (8.2), with
the M
12
Higgs branch; this is because the instanton there has precisely two two matter
Fermion zero modes (for the elds Q
1
and Q
2
). The upshot is perfectly compatible with
the dual (8.43): the M
12
and M
34
branches do intersect each other, despite their semi-
classical separation by distance = [m
4
[ [m
3
[, because the separation becomes part
of the M
12
Higgs branch. The non-compact Y Coulomb branch also intersects this point,
and there is an interacting SCFT there.
Now consider the case where m
3
and m
4
are non-zero, with m
3
+ m
4
,= 0, for m
1
=
m
2
= 0. There is the unlifted M
12
Higgs branch at = 0. For sign(m
3
) = sign(m
4
) =
1, the low-energy theory is SU(2)
k=0
, and there is the associated unlifted Coulomb
branch [m
3
[ 0, where F() = 0 in (8.42); this region gets quantum-merged together
with the M
12
Higgs branch, as in (8.2). The other regions of the Coulomb branches are
lifted by F() ,= 0 in (8.42). As seen from the dual (8.3), the low-energy theory consists
of the IR-free eld modulus M
12
. In terms of (8.43), the elds Y and M
34
are set to zero
by their real masses m
M
34 = m
Y
= m
3
+ m
4
. For m
3
= m
4
, the low-energy theory
at = m
4
is SQED with two massless elds of charge 1, but we see from (8.42) that
F( = m
4
) > 0, which is the wrong sign for obtaining a compact CP
1
Higgs branch there.
48
For m
3
and m
4
large, the low-energy theory is SU(2) 1
2
(1+sign(m
3
))
with N
f
= 1 light
avor. According to (8.4), this low-energy theory has Tr(1)
F
= 0 for m
3
< 0, or
Tr(1)
F
= 1 for m
3
> 0. In sum, upon turning on small m
1
and m
2
to lift all mod-
uli, the supersymmetric vacuum of the N
f
= 2 theory should either run o to innity and
decouple, if m
3
< 0, or remain near the origin for m
3
> 0. Again, these Kahler metric
details are less evident in the W = Y PfM dual description.
8.7. A = 2 SU(2)
k
with matter Q in a triplet
One can consider matter in general representations, with no analog of the 4d asymp-
totic freedom restriction on the matter representation. Here we just briey mention one
example. Consider the theory of a single matter eld Q in the adjoint representation.
The index (8.5) gives
24
Tr(1)
F
= [k[ + 1 when Q is given a real mass. The theory
with massless Q has a moduli space, where M = Q
2
has an expectation value, break-
ing SU(2)
k
U(1)
2k
, with no matter. The non-compact M moduli space means that
Tr(1)
F
is well-dened only if the adjoint is given a real mass m
Q
,= 0.
The index gives a simple check of the duality in [68] between SU(2)
1
, with an adjoint
and a free eld theory with eld M = Q
2
, plus a topological sector. If not for the topological
sector, there would be a mismatch. Upon giving Q a real mass, the index is Tr(1)
F
= 2,
whereas an IR-free eld M contributes Tr(1)
F
= 1 for m
M
,= 0. The index matching
works upon including the tensor product with the topological sector U(1)
2
, which is the
low-energy theory left unbroken by M for m
Q
= 0, and which indeed has Tr(1)
F
= 2.
9. Preliminary aspects of SU(N
c
) and U(N
c
)
We here briey discuss, in parallel, gauge groups G = SU(N
c
) and G = U(N
c
), with
N
+
matter elds Q
f
N
c
and N

matter elds

Q
f
N
c
. We can turn on real masses m
g
f
and m
g
f
for the matter elds, and Chern-Simons term k. For the case of U(N
c
), we can also
24
This theory diers from N = 3 SYM, which has a complex mass superpotential term, with
m
C
= k, see e.g. [5]. Since m
C
= 0, the adjoint of N = 3 SYM cannot be given a real mass, and so
the theories need not have the same index. Indeed, the N = 3 SYM theory has |k

N=3
| = |k

|
3
2
h
in (5.1), so it has Tr(1)
F
= |k| 2. The N = 4 case has k = 0, so the index is ill-dened.
49
add an FI term L
_
d
4

2
TrV . (In this case we can also have dierent Chern-Simons
coecients for SU(N
c
) and U(1), but we will not do it here.) The classical vacua satisfy
_

c

f

f
+
c

c
m
f

f
_
Q
cf
= 0
_

c

f

f
+
c

c
m
f

f
_

Q
c

f
= 0
D
c

c

k
2

c


2

c
+Q

cf
Q
c

Q
c

f

Q
c f
= 0 .
(9.1)
For G = SU(N
c
), we set = 0 and relax the RHS of the last line in (9.1) to be proportional
to
c

c
. We can always choose the real Coulomb moduli
c

c
=
c

c
to be diagonal by a
gauge rotation; the o-diagonal components are eaten by the massive gauge elds in the
breaking to the Cartan subgroup, U(N
c
) U(1)
N
c
or SU(N
c
) U(1)
N
c
1
. We can
restrict the
c
to the Weyl chamber
1

2
. . .
N
c
.
The semi-classical vacua are obtained via (9.1) with k and replaced with the one-
loop exact quantum corrected expressions, much as in (2.6). One then adds in the eects
of instanton corrections. Consider, for example, U(N
c
)
k
, with massless matter, expanded
around
c

c

c

c
, with large. Then U(N
c
) is approximately unbroken, and the matter
elds are massive and can be integrated out leading to
k
eff
( ) = k k
c
, k
c

1
2
(N
+
N

) (9.2)
as in the U(1) case (2.14). For k ,= k
c
, this region of the Coulomb branch is lifted by
k
eff
,= 0. For k = k
c
, there is a non-compact Coulomb branch, which is partially lifted
by instantons.
When the
c
all dier, we are out along the Coulomb branch, G U(1)
r
, and the
theory can be approximately analyzed in terms of the low-energy U(1)
r
gauge theory, with
the matter charges given by its G-rep weight vector. We then need to add instanton eects
in the broken SU(2) subgroups of G. In particular, each low-energy U(1) gauge factor yields
an approximate U(1)
J
global topological symmetry, so there is an approximate

r
i=1
U(1)
J
i
global symmetry. Instantons explicitly break N
c
1 of these global U(1) factors, so U(N
c
)
has a single U(1)
J
global symmetry, while SU(N
c
) has none. The Coulomb branch can be
described either in terms of real, linear multiplets
c
, in the U(1)
r
Cartan subalgebra, or
dualized to chiral superelds. In U(N
c
) we can have X
i
e
(
i
/e
2
eff,i
+ia
i
)
, with charges
1 under the erstwhile global U(1)
J
i
charge of the low-energy U(1)
r
Abelian theory. In
SU(N
c
) it is better to use Y
i
X
i+1,
X
i,+
.
50
As discussed in [3,4], instantons indeed generate superpotential terms which lift most
of the Coulomb branch moduli. For SU(N
c
), a single Coulomb branch chiral supereld, Y ,
remains unlifted. For U(N
c
), two Coulomb branch chiral superelds, X

, remain unlifted.
It will be soon useful to recall here how this works. Semi-classically, X
i+
1/X
i
and the
SU(2) subgroups of G have Y
i
X
i+1,
X
i,+
1/X
i+1,+
X
i,
, i = 1 . . . N
c
1, with global
U(1) charges which are similar to (8.13), but with the Callias index giving Fermion zero
modes, K
j
,= 0, only for the Y
i
0
instanton factor which has
i
0
+1
> 0 >
i
0
. Instantons in
these SU(2) subgroups of G generate the dynamical superpotential [2,44]
W
dyn
=

i
1
Y
i
. (9.3)
The eect of this superpotential, is to set
2
= =
N
c
1
= 0. For SU(N
c
), there is a
single, unlifted Coulomb modulus, Y , given by
Y =
N
c
1

i=1
Y
i
. (9.4)
For U(N
c
), there are two unlifted Coulomb moduli, X
+
X
1+
and X

X
N
c

, and we
can replace Y X
+
X

.
We recall, for example, that SU(N
c
)
0
and U(N
c
)
0
, with N
f
= N
c
avors, has a nice
dual description in terms of the unlifted Coulomb moduli, together with the Higgs branch
moduli, with superpotential [3]
SU(N
c
)
0
: with N
f
= N
c
: W = Y (detM B

B), (9.5)
U(N
c
)
0
: with N
f
= N
c
: W = X
+
X

detM. (9.6)
The quantum description for N
f
< N
c
follows from (9.5) or (9.6), upon adding complex
mass terms and integrating out avors. For example, the theories with N
f
= N
c
1 have
quantum moduli space of supersymmetric vacua, with a constraint that follows from (9.5)
or (9.6) upon adding W = m
N
c
M
N
c
N
c
and integrating out the massive avors:
SU(N
c
)
0
with N
f
= N
c
1 : Y detM = 1. (9.7)
U(N
c
)
0
with N
f
= N
c
1: X
+
X

detM = 1. (9.8)
For N
f
< N
c
1, this yields a runaway superpotential for Y or X
+
X

, so the theory with

zero masses has no stable vacuum; upon adding real masses, they break supersymmetry.
51
The quantum description for N
f
> N
c
requires additional degrees of freedom, e.g. those
of the Aharony dual [40].
We now consider these theories with k ,= 0, and more general matter content. Consider
the case of N
+
matter elds Q in the fundamental and N

matter elds

Q in the anti-
fundamental. For k = 0, the quantum numbers are similar to those in [40], with N
f
there
replaced with N
f

1
2
(N
+
+N

) for the U(1)

A
and U(1)
R
charges. For SU(N
c
)
0
,
SU(N
c
)
0
SU(N
+
) SU(N

) U(1)
A
U(1)
R
Q
i
N
+
1 1 0

Q
i
1 N

1 0
M
i
i
= Q
i

Q
i
N
+
N

2 0
Y 1 1 2N
f
2(N
f
N
c
+ 1)
(9.9)
For U(N
c
)
0
, there is the extra U(1)
J
symmetry, and we replace Y X
+
X

U(N
c
)
0
SU(N
+
) SU(N

) U(1)
A
U(1)
R
U(1)
J
Q
i
N
+
1 1 0 0

Q
i
1 N

1 0 0
M
i
i
= Q
i

Q
i
N
+
N

2 0 0
X

1 1 N
f
N
f
N
c
+ 1 1
(9.10)
There is no analog of superpotentials (9.5) or (9.6) if N
+
,= N

, since the holomorphic

SU(N

) singlet object detM can only be formed from M

i
i
if N
+
= N

.
Consider the general N

case, with all matter elds massless, m

f

f
= m
f

f
= 0 in (9.1).
There is a non-compact, classical Higgs branch /
cl
H
moduli space of solutions of (9.1), at

c
= 0, as long as N
+
N

,= 0. Taking N
+
N

> 0, on /
cl
H
the gauge group is broken
as SU(N
c
) SU(N
c
N

), or U(N
c
) U(N
c
N

). So the SU(N
c
) gauge group is
fully broken if N

N
c
1, and the U(N
c
) gauge group is fully broken if N

N
c
. In the
cases where the gauge group is fully broken, there is a quantum Higgs branch moduli space
of SUSY vacua, /
H

= /
cl
H
. When the gauge group is not fully broken, the unbroken
gauge group will break supersymmetry if [k[ < N
c
N

, with a runaway Coulomb branch

in the k = 0 case.
52
10. From Giveon-Kutasov to Aharony
Aharony duality [40] is unusual in that the dual Lagrangian contains terms explicitly
involving the dual theorys monopole operators. One might reasonably question whether
or not this is a legitimate theory: is one allowed to include monopole operators in the
dening Lagrangian of a theory? We here show how to UV-complete this theory to one
without the monopole operators in the Lagrangian. We do this by showing that Aharony
duality can be derived via an RG-ow, as an IR-consequence of Given-Kutasov duality in
the UV. The monopole operators in the dual Lagrangian are in this way UV-completed to
a more-conventional theory without them.
In the following subsections, we briey review these dualities in greater detail, and
then show how to derive Aharony duality as a consequence of Given-Kutasov duality.
10.1. Review of Aharony duality
The electric theory is a U(N
c
)
0
gauge theory, with N
f
avors in the fundamental
and anti-fundamental, Q
cf
and

Q
c f
. The gauge and global charges, including those of
the composite meson M = Q

Q and the monopole operators X

, are as in (9.10) with

N
+
= N

= N
f
. The dual theory [40] is a U(n
c
= N
f
N
c
)
0
gauge theory with quarks
and anti-quarks q and q, and the elementary elds M, X

, with quantum numbers

U(n
c
)
0
SU(N
f
) SU(N
f
) U(1)
A
U(1)
J
U(1)
R
q
c
f
n
c
N
f
1 1 0 1
q
f
c
n
c
1 N
f
1 0 1
M
f
f
1 N
f
N
f
2 0 0
X

1 1 1 N
f
1 N
f
N
c
+ 1
(10.1)
The dual theory has the superpotential [40]
W = Mq q +X
+

+X

X
+
. (10.2)
where

X

are the monopole operators of the magnetic theory. It should be stressed

that, due to the explicit appearance of the magnetic operators

X

in the Lagrangian, the

magnetic theory is not a standard weakly coupled renormalizable Lagrangian. At weak
coupling the last two terms in (10.2) are high dimension operators.
53
10.2. Giveon-Kutasov Duality
The electric theory is U(N
c
)
k
theory with N
f
quarks. The quantum numbers are as in
(9.10) with N
+
= N

= N
f
except that, for k ,= 0, we should omit the operators X

which
are not gauge invariant (as in (2.13), they get charge k from the overall U(1) U(N
c
)).
U(N
c
)
k
SU(N
f
) SU(N
f
) U(1)
A
U(1)
J
U(1)
R
Q
cf
N
c
N
f
1 1 0 0

Q
c f
N
c
1 N
f
1 0 0
M
f
f
1 N
f
N
f
2 0 0
(10.3)
The dual theory [31] has gauge group U(n
c
)
k
, with n
c
= N
f
+[k[ N
c
and matter
U(n
c
)
k
SU(N
f
) SU(N
f
) U(1)
A
U(1)
J
U(1)
R
q
c
f
n
c
N
f
1 1 0 1
q
c f
n
c
1 N
f
1 0 1
M
f
f
1 N
f
N
f
2 0 0
(10.4)
The dual has the superpotential [31]
W = Mq q . (10.5)
As a warmup, we outline how Giveon-Kutasov duality can be derived from Aharony
duality. Consider starting with an electric U(N
c
)
0
with N
f
+ [k[ avors, and turn on
equal mass m for [k[ avors, with sign(k) = sign(m). In the U(N
f
+ [k[ N
c
)
0
Aharony
dual of the original electric theory, the real masses map to real masses m for [k[ avors,
real masses for X

, and real masses for the (N

f
+ [k[)
2
N
2
f
components of the singlets
M with avor indices for the massive electric avors. The masses follow from the U(1)
A
assignments in (10.1). Integrating out the massive avors on both sides, the electric theory
ows in the IR to U(N
c
)
k
with N
f
avors, and the Aharony dual ows in the IR to the
U(N
f
+[k[ N
c
)
k
Giveon-Kutasov dual. The peculiar

X

of the UV Aharony dual (10.2)

decouple in the IR limit, together with the elementary elds X

, which got real masses.

54
10.3. Turning on a real mass for one avor in U(N
c
)
k
We start with the electric side (10.3) of Giveon-Kutasov duality, U(N
c
)
k
theory with
N
f
quark avors, Q
f
and

Q
f
, and consider turning on real masses. The classical vacua
satisfy (9.1), where we set the FI term = 0. For simplicity, consider the case where we
give a non-zero real mass to only one avor, giving the same real mass to Q
N
f
and

Q
N
f
,
m
f

f
= m
fN
f

N
f
; m
f

f
= m
fN
f

N
f
(10.6)
and we take m > 0; the results for m < 0 are similar. The solutions of (9.1) are
25
:
1. = 0, with Q
cN
f
=

Q
cN
f
= 0. The theory here has a Higgs branch for the remaining
N
f
1 massless avors. Integrating out the massive avor, the low-energy theory at
the root of the Higgs branch is a U(N
c
)
k+1
gauge theory with N
f
1 avors. The
case k = 1, where there is a Coulomb branch, will be discussed separately below.
2.
N
c
N
c
m, with the other entries of vanish, breaking the gauge symmetry as
U(N
c
) U(N
c
1) U(1). The unbroken U(N
c
1) factor is similar to the previous
case: it has N
f
1 massless avors and one massive avor, which leads to U(N
c
1)
k+1
with N
f
1 massless avors. The unbroken U(1) gauge factor has N
f
elds of charge
1, from the c = N
c
color component of the original Q
f
and

Q
f
. Of these N
f
charged
avors, N
f
1 have real masses
N
c
N
c
, so they can be integrated out and their one-loop
corrections to
eff
and k
eff
in (2.6) cancel. The U(1) charge +1 matter eld Q
N
c
,N
f
has real mass
N
c
N
c
+m 0, so it is light and kept in the low-energy theory. The U(1)
charge 1 matter eld

Q
N
c
,N
f
has real mass m
N
c
N
c
2m, so it is integrated out
and then (2.6) gives + m, k k +
1
2
. It is natural to shift
N
c
N
c
by = m
to vanish in the vacuum; as in (2.2), this additionally shifts + km. In sum, in
addition to U(N
c
1)
k+1
with N
f
avors, there is a low-energy, decoupled U(1)
k+
1
2
factor, with a single light eld Q = Q
N
c
,N
f
, of charge +1 and
= m(k + 1) . (10.7)
We analyze this low-energy U(1)
k+
1
2
theory as in the discussion following eqn. (2.8),
with k
low
= k +
1
2
and given by (10.7). According to (6.3), this U(1)
k+
1
2
theory
generically has Tr(1)
F
= [k +
1
2
[ +
1
2
SUSY vacua, but the topological vacua do
not count, since they are far from
N
c
N
c
m, so outside of our low-energy eective
25
A closely related discussion appeared in [69].
55
theory; only the Higgs vacua are in the low-energy theory. For generic k < 1, the
eective FI term F(
Q
) < 0 and there is no supersymmetric vacuum. For k > 0, there
is an isolated supersymmetric vacuum, with a mass gap, on the Higgs branch, with
Q

. The U(1) gauge eld and charged matter Q are massive, so the remaining
low energy theory is U(N
c
1)
k+1
with N
f
1 avors. Again, k = 1 is a special
case, giving
low
= 0, k
low
= +
1
2
, which we discuss separately below.
3.
N
c
N
c
m, with the other entries of vanish. The analysis here is similar to the
previous case with Q

Q. The low-energy theory is U(N
c
1)
k+1
U(1)
k+
1
2
, where
the U(1) has a single light eld, of charge 1, with = m(k + 1). For generic k,
there is a supersymmetric vacuum on the Higgs branch if k > 0, with

Q
N
c
,N
f

,
and no supersymmetric vacuum if k < 1.
4. The only nonzero entries of are
N
c
N
c
=
N
c
1
N
c
1
= m, breaking U(N
c
)
k
U(N
c

2)
k+1
U(1)
k+
1
2
U(1)
k+
1
2
. This is similar to the above two cases: the U(N
c
2) factor
has N
f
1 massless avors, and the U(1)
k+
1
2
each have a single charged eld, N
f
=
1
2
.
Again, for k < 1 these U(1) factors do not give supersymmetric vacua within the
low-energy theory, while for k > 0 they give a single supersymmetric vacua, with only
U(N
c
2)
k+1
with N
f
1 avors remaining light.
Next, consider the special case k = 1 and follow the four cases mentioned above:
1. The low energy theory is U(N
c
)
0
with N
f
1 avors and a Coulomb branch.
2. The U(1) factor is U(1)

1
2
with a single charge +1 chiral supereld and no FI-term
(see (10.7)). This theory is dual at low energies to a single chiral supereld U
+
(the
reason for the subscript will be clear below), representing half of the Coulomb branch
of this chiral Abelian theory. The U
+
chiral supereld must couple to the U(N
c
1)
0
elds and in particular, to the operator X

associated with the U(1) U(N

c
1).
The original U(N
c
) theory had a single topological symmetry under which U
+
and
X

have opposite charges. Therefore, eects in the broken high energy U(N
c
) theory
must break the separate topological symmetries of the two factors U(N
c
1) U(1).
In particular, monopole congurations that acts as 3d instantons must do the job and
produce a superpotential as in (9.3)
W U
+
X

. (10.8)
In conclusion, the low energy theory is U(N
c
1)
0
with N
f
1 avors and a chiral
supereld U
+
with the superpotential (10.8).
56
3. This case is similar to the previous one. The low energy theory is U(N
c
1)
0
with
N
f
1 avors and a chiral supereld U

. In addition, it has a superpotential

W U

X
+
, (10.9)
where X
+
is constructed out of the U(N
c
1) elds.
4. Repeating the discussion in the two previous cases, we nd a U(N
c
2)
0
theory with
N
f
1 avors and two chiral superelds U

with the superpotential

W U
+
X

+U

X
+
. (10.10)
10.4. Flowing by real masses from a pair of dual Giveon-Kutasov theories.
The previous discussion can be interpreted as owing from the electric Giveon-Kutasov
theory. We see that the answers depend on the value of k. It is straightforward to repeat
this analysis in the magnetic theory U(n
c
= N
f
+ [k[ N
c
)
k
with N
f
avors, chiral
superelds M and the superpotential (10.5). The electric real masses (10.6) map, according
to the global symmetries, to real masses in the dual for q
N
f
, q
N
f
, M
f
N
f
, and M
N
f
f
. The
discussion is similar to the above so we will not repeat it in detail. Again, there are four
kinds of vacua to consider and the answers depend on if k > 0, k < 1, or k = 1. The
main point is that the four kinds of vacua in the electric theory and the analogous four
kinds of vacua in the magnetic theory are mapped as
1 4
2 3
(10.11)
In particular, starting with U(N
c
)
1
with N
f
avors on the electric side, vacuum 1
gives a low-energy U(N
c
)
0
theory with N
f
1 avors. In the magnetic dual, we start with
U(N
f
+1 N
c
)
1
and ow in vacuum 4 to U(N
f
1 N
c
)
0
theory with the superpotential
(10.2), where the X
+

+X

X
+
terms come from the analog of (10.10) in the magnetic
dual, so U

and X

.
We learn several lessons from this analysis. First, owing from a Giveon-Kutasov
dual pair we nd a pair of Aharony duals. This gives additional evidence for the two
kinds of dualities. Furthermore, we see where the mysterious elds X

in Aharony duality
come from, and how to interpret the superpotential (10.10), which couples to complicated
operators U

. We also derived a new duality (which can be derived trivially from Aharony
duality) from the 2 3 vacua duality exchange: The electric theory is a U(N
c
)
0
theory
57
with N
f
avors, a singlet U
+
and a superpotential (10.8) U
+
X

. The magnetic theory

is a U(N
f
N
c
)
0
theory with N
f
avors, singlets M, a singlet

X
+
, and a superpotential
qM q + U

X
+
where U

is constructed out of the magnetic gauge elds. Finally, we can

extend this analysis and ow in the entire chain of dualities by turning on positive and
negative real masses. The main subtlety is how to ow to k = 0, where additional elds
show up. Our discussion shows how to do it.
Acknowledgments:
We would like to thank Antonio Amariti, Cyril Closset, Thomas Dumitrescu, Guido
Festuccia, Davide Gaiotto, Daniel Jaeris, Hans Jockers, Zohar Komargodski, John Mc-
Greevy, Shlomo Razamat, Brian Willett, and Edward Witten for discussions. We especially
thank Ofer Aharony and David Tong for helpful discussions, correspondence, participation
in some of this work, and thoughtful comments about the manuscript. We would also
like to thank Jihye Seo and McGill University for the N=2 Jeometry And ApplicationZ
workshop, where this collaboration was initiated. The research of NS is supported in part
by DOE grant DE-FG02-90ER40542. The research of KI was supported in part by DOE-
FG03-97ER40546. KI also thanks the IHES, CdF, Jussieu U., and the ENS for hospitality
and support in Paris in August 2012, while some of this work was done.
Appendix A. Notation and additional details, review
We here set our notation. There are some crucial relative signs and normalizations,
between the Chern-Simons term, the Fayet-Iliopoulos term, and the sign of their shifts
when matter is integrated out, which must be properly xed to obtain consistent results.
A.1. Susy algebra and conventions
We use the conventions of [70], reduced to 3d, e.g. the metric is
26

= diag(++).
The supersymmetry algebra is
Q

, Q

= 2

+ 2i

Z. (A.1)
26
In Euclidean space, iS
M
S
E
, e.g. introducing an i factor in the Chern-Simons action.
58
The 3d

matrices satisfy (

. We choose the spinor

basis
27
that diagonalizes
0

=0,1,2

= 1,
1
,
3
, (A.2)
which are the (

of [70], reduced along the x

=2
direction, and the sign of
0
ts with
positive energy, P
0
= P
0
0.
Indices are raised, lowered, and contracted with

or

(as in appendix B of [71]),

e.g.

, and

. In this notation
28
, () =

= ,
and i and

are real, with e.g. ()

2
=
1
2

2
. Supercharges Q

are represented in
superspace by dierential operators Q

which act on superelds O similar to [70]

O = i[QQ, O] = (QQ)O, (A.3)

where the supercharges can be written as
Q

, Q

+i

. (A.4)
The superspace derivatives, anti-commuting with (A.4) are
D

+i

, D

. (A.5)
A.2. Gauge elds and Lagrangians
A U(1) gauge theory has vector multiplet (in Wess-Zumino gauge [70]) V = i

+i
2
i
2
+
1
2

2
D. The gauge eld strength is in the real linear multiplet

i
2

V 2
J
= + + +
1
2

+
+iD +
i
2

i
2

+
1
4

2
,
(A.6)
with D
2
= D
2
= 0. Then (A.3) gives, for example, the photino variation
iQ

(i

+
1
2
F

) +i

D. (A.7)
27
Another sometimes-convenient basis is one that diagonalizes (
0
)

, and the
1
2
[
1
,
2
] angular
momentum contribution, (
=0,1,2
)

= (i
3
,
2
,
1
).
28
The notation of [72], where indices are contracted with C

= i

nicely eliminates the

minus sign and i above, but we will use the above convention to remain closer to [70].
59

J
is the current supereld for the U(1)
J
global symmetry, with topological current,
j

J

1
2

, so j
0
J
=
F
12
2

B
2
. The gauge kinetic, CS, and FI terms are
29
L
gauge
=
_
d
4

1
e
2

k
4
V

2
V
_
(A.8)
where e is the gauge coupling constant. The terms involving the gauge eld include
L
gauge

1
4e
2
F

+
k
4

+A

matter
, (A.9)
so the A

equations of motion are

1
e
2

= j

matter
+kj

J
j

Gauss
. (A.10)
The CS term k thus imparts electric charge to states carrying U(1)
J
topological charge,
as in (3.17). For states with q
Gauss
= 0, as in (3.18), q
matter
= kq
J
. Vortex states with
q
J
,= 0 acquire electric charge k
eff
q
J
, from (3.18) in the eective theory. The consistency
condition (2.1), k
eff
Z, then follows from the Dirac quantization condition.
For U(1)
k
with chiral superelds Q
i
of charges n
i
, the matter has classical terms
L
kin
=
N

i=1
_
d
4
Q

i
e
n
i
V +im
i

Q
i
. (A.11)
The auxiliary D component of the U(1) gauge multiplet appears in the Lagrangian as
L
cl

1
2e
2
D
2
+
D
2
_

i
n
i
[Q
i
[
2

k
2


2
_
. (A.12)
Solving for D,
D =
e
2
2
(

i
n
i
[Q
i
[
2

2

k
2
), (A.13)
the classical potential of the system is
V
cl
=
e
2
32
2
_

i
2n
i
[Q
i
[
2
k
_
2
+

i
(m
i
+n
i
)
2
[Q
i
[
2
. (A.14)
29
Our sign convention is that integrating out a Fermion of real mass m > 0 induces k > 0.
The following are odd under parity and time reversal: k, m
i
, . Under charge conjugation C,
V V , or equivalently we can leave V alone and take n
i
n
i
; other quantities are C even.
The following quantities all have dimensions of mass: e
2
, , ; while k and V are dimensionless.
60
Quantum corrections modify and k in (A.14), as in (2.3). More generally, including
the gauge kinetic term, the quantum corrections to (A.8) are of the form
L
gauge,e
=
_
d
4

_
f() +
1
4
(k
eff
V + 2
eff
V )
_
, (A.15)
where k
eff
and
eff
are one loop exact, as in (2.6). The function f() in (A.15) is given
at one loop by
f() =
1
e
2

2
+ 2slog(/e
2
) + , (A.16)
with s =
1
2

i
n
2
i
for the U(1) theory. The non-Abelian case is similar, see [65,4], with
negative contributions to s in (A.16) from integrating out the massive, non-Cartan gauge
elds on the Coulomb branch, e.g. s = N
f
3 for SU(2) with N
f
avors.
For a non-Abelian gauge theory, the A = 2 SUSY Chern-Simons term is
S
N=2
CS
=
k
4
_
Tr(AdA+
2
3
A
2
+ 2D) (A.17)
(see e.g. [73-76] for S
N=2
CS
written in superspace). If the gauge group is U(N
c
) we can also
add a FI term L
_
d
4

2
TrV . For the A = 2 theory with matter chiral superelds in
representations r
f
of the gauge group G, the quantization condition is k +
1
2

f
T
2
(r
r
)
Z, with T
2
(r
f
) = 1 e.g. for a fundamental or anti-fundamental of SU(N
c
) or U(N
c
).
Integrating out massive matter shifts k
eff
= k +
1
2

f
T
2
(r
f
) sign(m
f
()). For k
eff
= 0,
there is a Coulomb branch with G U(1)
r
; when k
eff
,= 0, the Coulomb branch is lifted.
A.3. Vortices and their zero modes
We here review and collect some formulae for vortices in 3d, A = 2 SUSY, U(1)
k
Chern-Simons-Maxwell theory with matter. See e.g. [56,50] for more details.
BPS eld congurations are annihilated by Q

and Q
+
, and anti-BPS congurations
are annihilated by Q
+
and Q

, with Q

=
1
2
(Q
1
iQ
2
) and Q

=
1
2
(Q
1
iQ
2
) (3.6). In
Wess-Zumino gauge, the variations of the photino, with

=
1
i
2
, are
Q

= F
zt
+
z

,
+
= F
zz
+
t
iD
Q
+
,
+
= F
zt

Q
+
,

= F
zz
+
t
+iD
(A.18)
61
with F
zz
= iF
xy
since z = x+iy. The complex conjugate equations are obtained via Q

, z z. The variation of a Fermions in chiral multiplets Q

i
=
i
+
i
+
2
F
i
,
with charges n
i
, for mass m
i
= 0, are (with D

in
i
A

and D
z

1
2
(D
x
iD
y
))
Q
+
,
i+
= D
z

i
,
Q
+
,
i
= iD
t

i
+n
i

i
,
Q

,
i+
= F
i
,
Q

,
i
= D
z

i
,
Q

,
i+
= iD
t

i
+n
i

i
,
Q
+
,
i
= F
i
.
(A.19)
with similar complex conjugate equations for the Fermions in the anti-chiral multiplets.
The BPS equations obtained from (A.18) and (A.19) are thus

t
= 0, F
zt
+
z
= 0, i
t

i
+n
i
( +A
0
)
i
= 0, (A.20)
F
zz
iD = 0, (A.21)
D
z

i
= F
i
= 0. (A.22)
Equations (A.20) can be solved in a static conguration, setting all
t
0 and A
0
= .
Integrating (A.21) over the spatial plane yields
2q
J
= i
_
d
2
zF
zz
=
_
d
2
zD
e
2
4
R
2
core
, (A.23)
where R
core
is the length scale of the vortex core, where the
I
0 and 0 in (A.13).
The signs in (A.23) show that (A.21) is only consistent if Z = q
J
> 0. Likewise, anti-BPS
states have z z in (A.21) and (A.22), and the analog of (A.23) requires Z = q
J
< 0.
Appendix B. Coulomb branch and Kahler form nonrenormalization
Consider rst an Abelian gauge theory. The massless modes along the Coulomb
branch are the microscopic gauge elds V
r
, r = 1, ..., n. They can be described by the
linear superelds
r
=
i
2
DDV
r
, and the eective Lagrangian is given by a real function
f:
L
eff
=
_
d
4
f(
r
). (B.1)
The kinetic terms for
r
=
r
[
==0
is determined from the metric
ds
2
= f
rs
d
r
d
s
, with f
rs

r

s
f(
c
). (B.2)
62
In a sensible theory, the Hessian f
rs
> 0 is a positive denite matrix.
It is convenient to dualize the linear multiplets
r
[45,4]. Locally, this is achieved by
adding Lagrange multiplier chiral superelds U
r
and studying the Lagrangian
30
_
d
4

_
f(
r
) + (U
r
+U
r
)

r
2
_
. (B.3)
Where the bottom component of U
r
[ has
(U
r
+U
r
)[ = 2
r
, (U
r
U
r
)[ = 2ia
r
, (B.4)
so e.g. U
r
+U
r
2

a
r
. The normalization of the added term in (B.3) is such that
a
r
have 2 periodicity, given that
r
=
r
/2, the conserved current for a U(1)
J
r
global
symmetry, has integral charges q
J
r
=
_
d
2
x
1
2

0
F

/2 (1.2). The shift symmetries

U
r
U
r
+i
r
, (B.5)
with real
r
, shifting a
r
a
r
+
r
in (B.4), reect the conserved U(1)
J
r
currents
r
/2.
Since the low energy theory is IR free, we can use the semiclassical approximation and
integrate out
r
by a Legendre transform
U
r
+U
r
= 2

r
f(
t
)[

t
=
U
t
, (B.6)
where
U
t
is the solution of (B.6), and we nd the action
L
eff
=
_
d
4
K
K = K(U
r
+U
r
) = f(
U
r
) +
1
2
(U
r
+U
r
)
U
r
.
(B.7)
The Legendre transform gives

U
r K =
U
r
K =
1
2

U
r
(B.8)
and the Hessian of K,
K
rs
=
U
r
U
s K =
1
2

U
s
U
r
, (B.9)
gives the inverse matrix of the Hessian of f in (B.2)
K
rs
= (2)
2
f
rs
. (B.10)
30
We use Lorentzian signature; in Euclidean space the second term has an imaginary coecient.
63
This determines the Kahler metric,
ds
2
= 2K
rs
dU
r
dU
s
= 2K
rs
(d
r
d
s
+da
r
da
s
). (B.11)
The Legendre transform breaks down at the vanishing eigenvalues of f
rs
or K
rs
, where
there are singularities on the Coulomb branch.
To illustrate the above procedure, consider a single Coulomb variable with f()

2
/e
2
for large . Then (B.6) and (B.7) give K
e
2
16
2
(U +U)
2
, with
2/e
2
,

a
2
e
2
F

, X = e
+ia
e
2/e
2
+ia
. (B.12)
The coordinates a
r
are natural because they are associated with the U(1)
n
isometry
(B.5) of the space, and
r
are related to them by the complex structure. Similarly, the
coordinates
r
are natural because they are related by supersymmetry to the eld strengths
and conserved currents. For large , (B.12) gives 2/e
2
. But it is important that
r
and a
r
reside in dierent superelds.
Using the general Kahler metric (B.11), for any f(
r
) in (B.7), the Kahler form is
=
i
2
K
rr
dU
r
dU
r
= K
rs
d
r
da
s
=
1
4
d
r
da
r
, (B.13)
where we used the above properties of the Legendre transform. In this context, this result
was derived in [46], where it was emphasized that the f independence of (B.13) means
that the Kahler form on the Coulomb branch is given exactly by its classical value.
This conclusion has several consequences and it can be extended in several directions.
First, consider a Coulomb branch of a non-Abelian theory. Because of magnetic monopoles,
the semiclassical
r
cannot be used beyond perturbation theory. In fact, as in [2], one
must rst dualize to the variables U
r
and then the shift symmetry (B.5) is violated. Hence,
the Kahler form can be renormalized non-perturbatively.
It turns out that in many situations we can still prove such a non-renormalization
result. Consider for example an SU(2) gauge theory. The theory has a one dimensional
Coulomb branch and it is invariant under a global U(1) symmetry that semiclassically acts
on it as in (B.5). Depending on the number of avors this symmetry is an R-symmetry
or an ordinary symmetry. Therefore, the Kahler potential must be of the form K(U +U).
Note that this is true even in the N
f
= 0 theory, where a superpotential for U is generated
non-perturbatively. Now, we can run the discussion above in reverse, dene using the
Legendre transform and show that the Kahler form is given exactly by d da.
64
More generally, whenever the theory has an n dimensional branch with n shift symme-
tries (B.5) (which could be accompanied with an R-transformation), the Kahler potential
has the form K(U
r
+U
r
). Then we can dene
r
as in (B.8), and conclude that the Kahler
form has the simple form (B.13).
This discussion breaks down at the singularities of the Coulomb branch. They occur
when the Hessian K
rs
has vanishing or divergent eigenvalues. Comparing with known
examples we can nd one of the following situations:
1. At some nite point,
r
=
(0)
r
the Kahler metric K
rs
has a vanishing eigenvalue. This
means that in terms of U
r
the space ends. Nevertheless, the moduli space of vacua
can continue beyond this point. Then, a new set of chiral superelds

U
r
are needed
on the other side. An example of this situation arises in a U(1) gauge theory with
several avors with real masses m
i
. Here, the Coulomb branch of the moduli space of
vacua is parameterized by < < +. It includes various regions m
i
< < m
i+1
parameterized by dierent chiral superelds U
i
(see, e.g. Fig. 1).
2. The classical Coulomb branch ends at some nite
r
=
(0)
r
, but in the quantum
theory
r
is extended beyond this point. This is the case in an SU(2) gauge theory
with N
f
= 1. The moduli space is parameterized by a meson M, which is related to
U through Me
U
= 1.
In the discussion above, we ignored the fact that the functions f and K could depend
on various parameters. In particular, locations of singularities on the Coulomb branch can
depend on real masses m
i
, or Fayet-Iliopoulos parameters. Such parameters can aect the
Coulomb branch because they can be regarded as background linear multiplet superelds.
It is then easy to argue that whenever the discussion above applies (i.e. when the theory
has a U(1)
n
isometry so that the Kahler potential depends only on U
r
+U
r
) the locations
of the singularities is not corrected from the semiclassical values. The singularities are at
=
i
, where
i
is a background linear supereld, which accounts for the real masses and
FI parameters. The quantum corrections to the linear superelds
i
are very restrictive,
the only ambiguity is in shifts by some linear superelds, and the ambiguity by such shifts
is easily determined semiclassically.
This conclusion has important consequences. It is often the case that the theory has
such a U(1)
n
global symmetry so that we can apply the result above, and the Coulomb
branch has singularities. There can then be BPS states that wrap the regions in the
Coulomb branch between these singularities. The BPS masses of such Skyrmions are
65
determined by the Kahler form. Since the Kahler form is not renormalized, and its singu-
larities are easily determined in the classical approximation, we can easily determine the
masses of the BPS states. This result was found in Abelian theories in [46], and here we
extend it also to many non-Abelian theories.
Appendix C. Comments about modes with logarithmically divergent norm
The purpose of this appendix is to clarify a confusing issue in the study of vortices
in theories with non-minimal matter, where it was found, starting with [24], that there
are non-normalizable zero modes. A typical example is the semi-local vortex of the
Abelian Higgs model with several elds [57]; it occurs both for vortex particles in 3d and
for vortex strings in 4d; see e.g. [77-82]. Various prescriptions have been given for handling
the non-normalizable modes either to regulate the divergence, e.g. by putting the theory
on compact space, or to remove them, e.g. by giving them mass. We here interpret these
modes and explain how to treat them, without regulators or removing them.
For concreteness, consider a 3d U(1) gauge theory with two complex scalar elds
i
of charge +1, and with FI term > 0. This system describes the bosonic sector of our
SQED with two Qs with charge +1 (section 4.2). The low energy dynamics of this system
is the CP
1
model. Therefore, if our space is R
2
, the global SU(2) symmetry of the system
is spontaneously broken to U(1). If our space is compact, e.g. a sphere or torus, then the
symmetry is not broken.
More concretely, in nite volume V = L
2
the system has a zero mode associated with
a global SU(2) rotation on the doublet
i
. The norm of this constant mode is L
2
and,
since it is nite, this mode is dynamical. As the volume of the compact space becomes
large, the norm of this mode diverges like L
2
, and it is frozen in the innite volume
limit. Correspondingly, the Hilbert space breaks into superselection sectors labeled by the
boundary conditions on CP
1
. Up to a global SU(2) rotation, we will take the boundary
conditions at innity to be

2
0 ; [
1
[
_

2
. (C.1)
The congurations fall into sectors labeled by the vortex number (1.2)
q
J
=
i
2
_
d
2
zF
zz
, (C.2)
66
where z and z parameterize space. The BPS static congurations with q
J
> 0 are deter-
mined by a meromorphic function of z [24,59]
w(z) =

2

1
=
q
J

I=1

I
z z
I
, (C.3)
where for simplicity we assume that all the poles are simple. When [
I
[ are much smaller
than the separation between z
I
this conguration can be interpreted as q
J
vortices each
centered around z
I
. [
I
[ corresponds to the scale size of the Ith vortex and arg(
I
) is the
orientation in the target space. The moduli space of these vortices and the metric on that
space are analyzed in detail in [59]. It was shown in [57,58,59] that these vortices smoothly
interpolate between the ANO vortices and the CP
1
Skyrmions.
The large z behavior of (C.3) is
w(z)

I

I
z
+O(1/z
2
) . (C.4)
The mode =

I

I
is interesting. Unlike all other modes in the system, which have
nite norm, the norm of is logarithmically divergent. (In the nite volume system this
mode has nite norm proportional to log L.) Therefore, it is not dynamical in the innite
volume system and it is frozen, as noted in [24].
We have seen two modes that should be frozen. The constant mode associated with
SU(2) rotations has quadratically divergent norm and , whose norm is logarithmically
divergent. The constant mode is visible already in the sector of the theory with q
J
= 0
and its physics leads to the superselection sectors there and the corresponding spontaneous
symmetry breaking. But the physical interpretation of appears to be mysterious. In
particular, if the mode of the vortices is not treated as a collective coordinate, then
the spectrum of the vortices appears to be a continuum labeled by . This looks quite
unphysical, and indeed we will now argue that this conclusion is wrong.
We claim that, as with the constant mode, is a superselection parameter already
for the q
J
= 0 sector. Consider e.g. a single, real, massless scalar eld , with boundary
conditions lim
|z|
= 0. Naively, we can set up the quantization of this system in a
standard way. However, consider a smooth conguration of at some xed time t = t
0
(z, z, t = t
0
) = Re

z
+O(1/[z[
2
) (C.5)
with some complex . Setting the initial conditions (C.5) with
t
(z, z, t = t
0
) = 0
the conguration has nite energy and it spreads out. But since Re

z
satises the static
67
equations of motion, no matter how much time we wait, the large [z[ behavior of is
always as in (C.5). The value of can be measured at innity, and it does not change.
We see that the congurations of the classical system are labeled by the complex
parameter and its value cannot change with time. This means that in the quantum
theory labels dierent superselection sectors in the Hilbert space. Indeed, starting with
a state with some value of and acting on it with any local operators, and letting the
system evolve for any nite time, we cannot nd states with any dierent value of .
Normally, we focus only on the states with = 0. These states include the standard
vacuum and we can take its energy to be zero. The states with nonzero are unusual.
Their energy is bounded below by zero. But there is no state with zero energy among
them. These states are necessarily time dependent.
Now we can move to the states with nonzero q
J
. Here we nd static states with any
. The states with nonzero q
J
can be created out of states with q
J
= 0 by acting on them
with monopole operators. Such local operators cannot change the value of .
We conclude that the Hilbert space breaks into superselection sectors labeled by the
continuous complex parameter . If we want to focus on the standard vacuum with = 0
we can limit the discussion of vortices to vortices with = 0. Vortices with nonzero are
excitations of states with q
J
= 0 with nonzero .
The discussion above of the Abelian Higgs model is easily generalized to N+1 complex
charged elds
i
. Here the low energy theory is the CP
N
model. It has N modes with
quadratically divergent norm associated with the boundary values lim
|z|

i
(with xed

i
[
i
[
2
). They are frozen when the system is in R
3
signaling the spontaneous breaking of
SU(N+1) SU(N). In addition, there are N logarithmically divergent modes associated
with the
1
z
fall-o of the scalar elds at innity. As with our single in the example above,
they are superselection sectors, and are present both with and without vortices.
We now consider the Fermionic zero modes. Again, some of them can have logarith-
mically divergent norms. On a compact space, such modes have nite norm and they
should be quantized. But when the system is on R
3
they are frozen. A simple way to
handle them, which is clearly motivated by our discussion above of the bosonic modes, is
the following. Start with the system in a compact space, where the quantization leads to
a big Hilbert space. Next, take the innite volume limit and then this big Hilbert space
breaks to distinct superselection sectors. Doing that with the Fermions, these dierent
superselection sectors have dierent charges. Clearly, there are no transitions between
states in distinct superselection sectors and we can focus only on one of them.
68
The new point is the identication of all these superselection sectors, associated both
with the bosonic and with the Fermionic modes, also when q
J
= 0 and there are no solitons.
Therefore, solitons with dierent values of , and dierent non-normalizable Fermi zero
modes, are not merely in dierent sectors of the one body problem: they are in dierent
sectors of the whole quantum eld theory. In each superselection sector of the whole
second-quantized system, we have solitons only with a xed value of , and with some
charges associated with the Fermion zero modes.
We end this appendix by pointing out that a similar phenomenon can take place in
four dimensions. In addition to its zero mode, a massless real scalar eld can have
asymptotic behavior at spatial innity analogous to (C.5), specied by a real parameter
(x, t = t
0
) =

[x[
+O(1/[x[
2
). (C.6)
With such asymptotics, the energy is nite, the -term satises the local equations of
motion, and the norm of the -mode is (linearly) divergent. Hence, leads to a new
superselection parameter.
Appendix D. Derivation of Tr(1)
F
for U(1)
k
with matter
We here derive the index (6.3) and (6.4), by explicitly verifying that the sum (6.2)
is independent of the real parameters m
i
and , and then evaluating it for a convenient
choice. Phase transitions can occur when, upon varying m
i
and , two (or more) vacua
locations collide. The Higgs vacua can exist at
Q
i
, the locations where the slope, k
eff
(),
of F() can change. Topological vacua are at
I
, where F() = 0. Varying the real
parameters, two
Q
i
vacua can cross each other, or a
Q
i
can cross a
I
; two
I
cannot
cross each other, since the equation for
I
is piecewise linear. We here verify that, although
the individual terms can vary, the sum (6.2) is unchanged by all such vacuum crossings.
As we vary the real parameters, the number of Higgs vacua at some
Q
i
can only
change if its s
i
in (2.9) changes, which only happens if
Q
i
crosses a
I
where F(
Q
i
)
passes through a zero. Likewise, the number of topological vacua at some
I
can only
change if [k
eff
(
I
)[ changes, which also only happens if
I
crosses a
Q
i
. We verify that
these changes always cancel in the total Tr(1)
F
. The dependence of (6.5) and (6.7) on
sign() illustrates this general result.
69
Consider rst the case of two crossing
Q
i
and
Q

i
, at the point, where
Q
i
=
Q

i
.
If sign(n
i
) = sign(n

i
), then s
i
= s
i
. For s
i
= s
i
= 1, there is a compact space of Higgs-
vacua /
H
at
Q
i
=
Q
i

, and continuity of Tr(1)

F
requires that these vacua contribute
Tr(1)
F
[
M
H
= n
2
i
+n
2
i
, to equal those of the Higgs vacua at
Q
i
and
Q

i
when they are
separated. For n
i
= n
i
= 1, the compact space of vacua is /
H

= CP
1
, which contributes
to the index according to its Euler character, Tr(1)
F
[
M
H
= (/
H
) = 2. More generally,
for n
i
= n
i
, /
H
is a Z
n
i
orbifold of CP
1
, which gives Tr(1)
F
[
M
H
= 2n
2
i
. These cases
indeed agree with n
2
i
+n
2
i
. For n
i
,= n
i
, the space /
H
is a singular weighted projective
space and the contribution to Tr(1)
F
is better computed via the UV U(1) linear sigma
model, which must give Tr(1)
F
[
M
H
= n
2
i
+ n
2
i
. If sign(n
i
) = sign(n

i
), then s
i
,= s
i
:
only one contributes, both before and after the crossing. At the crossing point, there is
a non-compact Higgs branch, so Tr(1)
F
is ill-dened; elsewhere, the index (6.2) is well
dened, and unchanged by the crossing.
Now consider the transition where a
Q
i
hits a
I
, i.e. where F(
Q
I
) changes sign.
The eect depends on the relative sign of the slopes (2.7) k

eff,i
of F() on either side of

Q
i
. If sign(k

eff,i
) = sign(F(
Q
i
)), there is a
I
topological vacuum at
I
<
Q
i
, and
if sign(k
+
eff,i
) = sign(F(
Q
i
)), there is a
I
vacuum at
I
>
Q
i
. So if k

eff,i
have the
same sign, the transition is that a single
I
crosses over to the other side of
Q
i
. If k

eff,i
are of opposite sign, the transition is instead that two topological vacua
I
and
I
pair
create or annihilate when they meet at
Q
i
.
Consider rst the same sign case, sign(k
+
eff,i
) = sign(k

eff,i
) sign(k
eff,i
), and sup-
pose that initially
I
<
Q
i
, and after the crossing that

I
>

Q
i
; the reversed situation is
analogous. So initially k
eff
(
I
) = k

eff,i
, with sign(F(
Q
i
)) = sign(k
eff,i
), and afterwards
k
eff
(

I
) = k
+
eff,i
, with sign(F(

Q
i
)) = sign(k
eff,i
). Using (2.7),
k
eff
(

I
) = k
+
eff,i
= k

eff,i
+n
2
i
sign(n
i
), (D.1)
so the number of topological vacua counted in (6.2) changes after the crossing, by
Tr(1)
F
topo
= [k

eff,i
+n
2
i
sign(n
i
)[ [k

eff,i
[ = n
2
i
sign(n
i
k
eff,i
). (D.2)
The number of Higgs vacua counted in (6.2) also changes after the crossing, by
Tr(1)
F
Higgs
= n
2
i
(s
i
s
i
) = n
2
i
((n
i
k
eff,i
) (n
i
k
eff,i
)) . (D.3)
70
The changes in (D.2) and (D.3) indeed cancel: the total number of vacua remains constant,
though some of the Higgs vacua can become topological vacua, or visa versa.
Now consider the case of sign(k

eff,i
) = sign(k
+
eff,i
), where topological vacua can
pair create or annihilate at
Q
i
. Consider the annihilation process; the reverse process is
analogous. Initially, sign(k

eff,i
) = sign(F(
Q
i
)) = sign(k
+
eff,i
), so there are
I
and
I

vacua on either side of

Q
i
, which then annihilate when F(
Q
i
) changes sign. The number
of topological vacua counted in (6.2) thus changes by
Tr(1)
F
topo
= [k

eff,i
+n
2
i
sign(n
i
)[ [k

eff,i
[ = n
2
i
sign(n
i
k

eff,i
). (D.4)
The number of Higgs vacua also changes, and using (2.9) here gives
Tr(1)
F
Higgs
= n
2
i
(s
i
s
i
) = n
2
i
_
(n
i
k

eff,i
) (n
i
k

eff,i
)
_
. (D.5)
Again, the changes (D.4) and (D.5) cancel.
We have veried that (6.2) is invariant under all m
i
and deformations, and can thus
evaluate (6.2) by a convenient m
i
, choice. Note from (2.14) that F() has asymptotic
slopes of the same (opposite) sign if [k[ > [k
c
[ ([k[ < [k
c
[), and hence an even (odd) number
of topological vacua. By the
I
annihilation process, we can always adjust such that
F() has only one zero for [k[ > [k
c
[, or no zeros for [k[ < [k
c
[.
We rst consider [k[ > [k
c
[. We take sign m
i
= sign n
i
, so that
Q
i
= m
i
/n
i
< 0
for all elds, and choose such that there is a single topological vacuum, at
I
= 0,
F( = 0) =
eff
= +
1
2

i
n
i
[m
i
[ = 0. The topological contribution to (6.2) is then
Tr(1)
F
topo
= [k
eff
( = 0)[ = [k[ +
1
2

i
n
2
i
sign(n
i
k). (D.6)
For the Higgs contributions, note that signF(
Q
i
) = sign k, so
Tr(1)
F
Higgs
=

i
s
i
n
2
i
=
1
2

i
n
2
i
(1 sign(n
i
k)). (D.7)
Adding (D.6) and (D.7) gives the result (6.3) for the total index.
We now consider the [k[ < [k
c
[ case, and take such that there are no topological
vacua, F() ,= 0, by taking [[ suciently large with sign = signk
c
, so sign F() = sign k
c
everywhere. Then (6.2) yields the result (6.4):
Tr(1)
F
= Tr(1)
F
Higgs
=
1
2

i
n
2
i
(1 + signn
i
k
c
) = [k
c
[ +
1
2

i
n
2
i
. (D.8)
71
References
[1] K. A. Intriligator and N. Seiberg, Lectures on supersymmetric gauge theories
and electric - magnetic duality, Nucl. Phys. Proc. Suppl. 45BC, 1 (1996). [hep-
th/9509066].
[2] I. Aeck, J. A. Harvey and E. Witten, Instantons and (Super)Symmetry Breaking
in (2+1)-Dimensions, Nucl. Phys. B 206, 413 (1982).
[3] 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. B 499, 67
(1997) [arXiv:hep-th/9703110].
[4] J. de Boer, K. Hori and Y. Oz, Dynamics of N=2 supersymmetric gauge theories in
three-dimensions, Nucl. Phys. B 500, 163 (1997). [hep-th/9703100].
[5] A. Kapustin and M. J. Strassler, On mirror symmetry in three-dimensional Abelian
gauge theories, JHEP 9904, 021 (1999). [hep-th/9902033].
[6] N. Dorey and D. Tong, Mirror symmetry and toric geometry in three-dimensional
gauge theories, JHEP 0005, 018 (2000). [hep-th/9911094].
[7] D. Tong, Dynamics of N=2 supersymmetric Chern-Simons theories, JHEP 0007,
019 (2000). [hep-th/0005186].
[8] T. Dimofte, D. Gaiotto and S. Gukov, Gauge Theories Labelled by Three-Manifolds,
[arXiv:1108.4389 [hep-th]].
[9] F. Benini, C. Closset and S. Cremonesi, Comments on 3d Seiberg-like dualities,
JHEP 1110, 075 (2011) [arXiv:1108.5373 [hep-th]].
[10] K. Intriligator, H. Jockers, P. Mayr, D. R. Morrison and M. R. Plesser, Coni-
fold Transitions in M-theory on Calabi-Yau Fourfolds with Background Fluxes,
[arXiv:1203.6662 [hep-th]].
[11] M. Cvetic, T. W. Grimm and D. Klevers, Anomaly Cancellation And Abelian Gauge
Symmetries In F-theory, JHEP 1302, 101 (2013). [arXiv:1210.6034 [hep-th]].
[12] M. Dine, Fields, Strings, and Duality: TASI 96, eds. C. Efthimiou and B. Greene,
(World Scientic, Singapore, 1997).
[13] S. Weinberg, Nonrenormalization theorems in nonrenormalizable theories, Phys.
Rev. Lett. 80, 3702 (1998). [hep-th/9803099].
[14] Z. Komargodski and N. Seiberg, Comments on the Fayet-Iliopoulos Term in Field
Theory and Supergravity, JHEP 0906, 007 (2009). [arXiv:0904.1159 [hep-th]].
[15] V. Borokhov, A. Kapustin and X. -k. Wu, Topological disorder operators in three-
dimensional conformal eld theory, JHEP 0211, 049 (2002). [hep-th/0206054].
[16] V. Borokhov, A. Kapustin and X. -k. Wu, Monopole operators and mirror symmetry
in three-dimensions, JHEP 0212, 044 (2002). [hep-th/0207074].
[17] N. Seiberg and E. Witten, Monopoles, duality and chiral symmetry breaking in N=2
supersymmetric QCD, Nucl. Phys. B 431, 484 (1994). [hep-th/9408099].
72
[18] A. M. Polyakov, Quark Connement and Topology of Gauge Groups, Nucl. Phys.
B 120, 429 (1977).
[19] M. J. Strassler, Conning phase of three-dimensional supersymmetric quantum elec-
trodynamics, In *Shifman, M.A. (ed.): The many faces of the superworld* 262-279.
[hep-th/9912142].
[20] K. A. Intriligator and N. Seiberg, Mirror symmetry in three-dimensional gauge the-
ories, Phys. Lett. B 387, 513 (1996). [hep-th/9607207].
[21] A. Kapustin, E. Witten and , Electric-Magnetic Duality And The Geometric Lang-
lands Program, Commun. Num. Theor. Phys. 1, 1 (2007). [hep-th/0604151].
[22] S. Gukov, E. Witten and , Rigid Surface Operators, Adv. Theor. Math. Phys. 14
(2010). [arXiv:0804.1561 [hep-th]].
[23] D. Bashkirov, Aharony duality and monopole operators in three dimensions,
[arXiv:1106.4110 [hep-th]].
[24] R. S. Ward, Slowly Moving Lumps In The Cp**1 Model In (2+1)-dimensions, Phys.
Lett. B 158, 424 (1985).
[25] B. Collie and D. Tong, The Partonic Nature of Instantons, JHEP 0908, 006 (2009).
[arXiv:0905.2267 [hep-th]].
[26] E. Witten, Constraints on Supersymmetry Breaking, Nucl. Phys. B 202, 253 (1982).
[27] G. Festuccia and N. Seiberg, Rigid Supersymmetric Theories in Curved Superspace,
JHEP 1106, 114 (2011). [arXiv:1105.0689 [hep-th]].
[28] E. Witten, Supersymmetric index of three-dimensional gauge theory, In *Shifman,
M.A. (ed.): The many faces of the superworld* 156-184. [hep-th/9903005].
[29] O. Bergman, A. Hanany, A. Karch and B. Kol, Branes and supersymmetry breaking
in three-dimensional gauge theories, JHEP 9910, 036 (1999). [hep-th/9908075].
[30] K. Ohta, Supersymmetric index and s rule for type IIB branes, JHEP 9910, 006
(1999). [hep-th/9908120].
[31] A. Giveon and D. Kutasov, Seiberg Duality in Chern-Simons Theory, Nucl. Phys.
B 812, 1 (2009). [arXiv:0808.0360 [hep-th]].
[32] E. Witten, Toroidal compactication without vector structure, JHEP 9802, 006
(1998). [hep-th/9712028].
[33] V. G. Kac and A. V. Smilga, Vacuum structure in supersymmetric Yang-Mills the-
ories with any gauge group, In *Shifman, M.A. (ed.): The many faces of the super-
world* 185-234. [hep-th/9902029].
[34] K. A. Intriligator and S. D. Thomas, Dynamical supersymmetry breaking on quan-
tum moduli spaces, Nucl. Phys. B 473, 121 (1996). [hep-th/9603158].
[35] K. A. Intriligator and N. Seiberg, Lectures on Supersymmetry Breaking, Class.
Quant. Grav. 24, S741 (2007). [hep-ph/0702069].
[36] E. Witten, Phases of N=2 theories in two-dimensions, Nucl. Phys. B 403, 159
(1993). [hep-th/9301042].
73
[37] E. Witten, Quantum Field Theory and the Jones Polynomial, Commun. Math.
Phys. 121, 351 (1989).
[38] C. Closset, T. T. Dumitrescu, G. Festuccia and Z. Komargodski, Supersymmetric
Field Theories on Three-Manifolds, [arXiv:1212.3388 [hep-th]].
[39] C. Closset, T. T. Dumitrescu, G. Festuccia, Z. Komargodski and N. Seiberg, Com-
ments on Chern-Simons Contact Terms in Three Dimensions, [arXiv:1206.5218 [hep-
th]].
[40] O. Aharony, IR duality in d = 3 N=2 supersymmetric USp(2N(c)) and U(N(c)) gauge
theories, Phys. Lett. B 404, 71 (1997). [hep-th/9703215].
[41] N. Seiberg, Electric - magnetic duality in supersymmetric nonAbelian gauge theo-
ries, Nucl. Phys. B 435, 129 (1995). [hep-th/9411149].
[42] K. A. Intriligator and N. Seiberg, Duality, monopoles, dyons, connement and oblique
connement in supersymmetric SO(N(c)) gauge theories, Nucl. Phys. B 444, 125
(1995). [hep-th/9503179].
[43] C. Hwang, H. -C. Kim and J. Park, Factorization of the 3d superconformal index,
[arXiv:1211.6023 [hep-th]].
[44] O. Aharony, S. S. Razamat, N. Seiberg and B. Willett, 3d dualities from 4d dualities,
[arXiv:1305.3924 [hep-th]].
[45] N. J. Hitchin, A. Karlhede, U. Lindstrom and M. Rocek, Hyperkahler Metrics and
Supersymmetry, Commun. Math. Phys. 108, 535 (1987).
[46] M. Aganagic, K. Hori, A. Karch and D. Tong, Mirror symmetry in (2+1)-dimensions
and (1+1)-dimensions, JHEP 0107, 022 (2001). [hep-th/0105075].
[47] D. L. Jaeris, The Exact Superconformal R-Symmetry Extremizes Z, JHEP 1205,
159 (2012). [arXiv:1012.3210 [hep-th]].
[48] C. Closset, T. T. Dumitrescu, G. Festuccia, Z. Komargodski and N. Seiberg, Contact
Terms, Unitarity, and F-Maximization in Three-Dimensional Superconformal Theo-
ries, JHEP 1210, 053 (2012). [arXiv:1205.4142 [hep-th]].
[49] F. Wilczek, Magnetic Flux, Angular Momentum, and Statistics, Phys. Rev. Lett.
48, 1144 (1982).
[50] B. -H. Lee and H. Min, Quantum aspects of supersymmetric Maxwell Chern-Simons
solitons, Phys. Rev. D 51, 4458 (1995). [hep-th/9409006].
[51] C. Beem, T. Dimofte and S. Pasquetti, Holomorphic Blocks in Three Dimensions,
[arXiv:1211.1986 [hep-th]].
[52] A. S. Goldhaber, A. Rebhan, P. van Nieuwenhuizen and R. Wimmer, Quantum cor-
rections to mass and central charge of supersymmetric solitons, Phys. Rept. 398, 179
(2004). [hep-th/0401152].
[53] L. Mezincescu and P. K. Townsend, Semionic Supersymmetric Solitons, J. Phys. A
A 43, 465401 (2010). [arXiv:1008.2775 [hep-th]].
74
[54] P. Fendley and K. A. Intriligator, Scattering and thermodynamics of fractionally
charged supersymmetric solitons, Nucl. Phys. B 372, 533 (1992). [hep-th/9111014].
[55] P. Fendley and K. A. Intriligator, Scattering and thermodynamics in integrable N=2
theories, Nucl. Phys. B 380, 265 (1992). [hep-th/9202011].
[56] C. -k. Lee, K. -M. Lee and H. Min, Selfdual Maxwell Chern-Simons solitons, Phys.
Lett. B 252, 79 (1990).
[57] T. Vachaspati and A. Achucarro, Semilocal cosmic strings, Phys. Rev. D 44, 3067
(1991).
[58] M. Hindmarsh, Semilocal topological defects, Nucl. Phys. B 392, 461 (1993). [hep-
ph/9206229].
[59] R. A. Leese and T. M. Samols, Interaction of semilocal vortices, Nucl. Phys. B 396,
639 (1993).
[60] S. Elitzur, G. W. Moore, A. Schwimmer and N. Seiberg, Remarks on the Canonical
Quantization of the Chern-Simons-Witten Theory, Nucl. Phys. B 326, 108 (1989).
[61] A. V. Smilga, Once more on the Witten index of 3d supersymmetric YM-CS theory,
JHEP 1205, 103 (2012). [arXiv:1202.6566 [hep-th]].
[62] M. Henningson, Ground states of supersymmetric Yang-Mills-Chern-Simons theory,
JHEP 1211, 013 (2012). [arXiv:1209.1798 [hep-th]].
[63] C. Callias, Index Theorems on Open Spaces, Commun. Math. Phys. 62, 213 (1978).
[64] K. -I. Izawa and T. Yanagida, Dynamical supersymmetry breaking in vector - like
gauge theories, Prog. Theor. Phys. 95, 829 (1996). [hep-th/9602180].
[65] N. Seiberg and E. Witten, Gauge dynamics and compactication to three-dimensions,
In *Saclay 1996, The mathematical beauty of physics* 333-366. [hep-th/9607163].
[66] N. Dorey, D. Tong and S. Vandoren, Instanton eects in three-dimensional super-
symmetric gauge theories with matter, JHEP 9804, 005 (1998). [hep-th/9803065].
[67] I. L. Buchbinder, N. G. Pletnev and I. B. Samsonov, Eective action of three-
dimensional extended supersymmetric matter on gauge supereld background, JHEP
1004, 124 (2010). [arXiv:1003.4806 [hep-th]].
[68] D. Jaeris and X. Yin, A Duality Appetizer, [arXiv:1103.5700 [hep-th]].
[69] I. Shamir, Aspects of three dimensional Seiberg duality, M. Sc. thesis submitted to
the Weizmann Institute of Science, April 2010.
[70] J. Wess and J. Bagger, Supersymmetry and supergravity, Princeton, USA: Univ.
Pr. (1992) 259 p.
[71] T. T. Dumitrescu and N. Seiberg, Supercurrents and Brane Currents in Diverse
Dimensions, JHEP 1107, 095 (2011). [arXiv:1106.0031 [hep-th]].
[72] S. J. Gates, M. T. Grisaru, M. Rocek and W. Siegel, Superspace Or One Thousand
and One Lessons in Supersymmetry, Front. Phys. 58, 1 (1983). [hep-th/0108200].
[73] B. M. Zupnik and D. G. Pak, Topologically Massive Gauge Theories In Superspace,
Sov. Phys. J. 31, 962 (1988).
75
[74] E. A. Ivanov, Chern-Simons matter systems with manifest N=2 supersymmetry,
Phys. Lett. B 268, 203 (1991).
[75] S. J. Gates, Jr. and H. Nishino, Remarks on the N=2 supersymmetric Chern-Simons
theories, Phys. Lett. B 281, 72 (1992).
[76] D. Gaiotto and X. Yin, Notes on superconformal Chern-Simons-Matter theories,
JHEP 0708, 056 (2007). [arXiv:0704.3740 [hep-th]].
[77] A. Hanany and D. Tong, Vortices, instantons and branes, JHEP 0307, 037 (2003).
[hep-th/0306150].
[78] R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi and A. Yung, Nucl. Phys. B 673, 187
(2003). [hep-th/0307287].
[79] M. Shifman and A. Yung, NonAbelian string junctions as conned monopoles, Phys.
Rev. D 70, 045004 (2004). [hep-th/0403149].
[80] A. Hanany and D. Tong, Vortex strings and four-dimensional gauge dynamics, JHEP
0404, 066 (2004). [hep-th/0403158].
[81] S. Olmez and M. Shifman, Revisiting Critical Vortices in Three-Dimensional SQED,
Phys. Rev. D 78, 125021 (2008). [arXiv:0808.1859 [hep-th]].
[82] P. Koroteev, M. Shifman, W. Vinci and A. Yung, Quantum Dynamics of Low-
Energy Theory on Semilocal Non-Abelian Strings, Phys. Rev. D 84, 065018 (2011).
[arXiv:1107.3779 [hep-th]].
76