Published for SISSA by Springer

Received: February 11, 2015

Accepted: May 4, 2015
Published: May 19, 2015

Defects and quantum Seiberg-Witten geometry

JHEP05(2015)095
Mathew Bullimore,a Hee-Cheol Kimb and Peter Koroteevb
a
Institute for Advanced Study,
Einstein Dr., Princeton, NJ 08540, U.S.A.
b
Perimeter Institute for Theoretical Physics,
31 Caroline Street North, Waterloo, Ontario N2L2Y5, Canada
E-mail: [email protected], [email protected],
[email protected]

Abstract: We study the Nekrasov partition function of the five dimensional U(N ) gauge
theory with maximal supersymmetry on R4 ×S 1 in the presence of codimension two defects.
The codimension two defects can be described either as monodromy defects, or by coupling
to a certain class of three dimensional quiver gauge theories on R2 × S 1 . We explain how
these computations are connected with both classical and quantum integrable systems.
We check, as an expansion in the instanton number, that the aforementioned partition
functions are eigenfunctions of an elliptic integrable many-body system, which quantizes
the Seiberg-Witten geometry of the five-dimensional gauge theory.

Keywords: Supersymmetric gauge theory, Duality in Gauge Field Theories, Integrable

Equations in Physics, Differential and Algebraic Geometry

ArXiv ePrint: 1412.6081

Open Access, c The Authors.

doi:10.1007/JHEP05(2015)095
Article funded by SCOAP3 .
Contents

1 Introduction 1

2 Twisted chiral rings 3

2.1 The Nekrasov-Shatashvili correspondence 4
2.2 Quantum/classical duality 6
2.2.1 Electric frame 9

JHEP05(2015)095
2.2.2 Magnetic frame 11
2.3 Line operators and interfaces in N = 4 SYM 12
2.4 Quantum equivariant K-theory 13
2.5 More genetic quiver varieties 14
2.5.1 Higgsing T [U(2)] 15
2.5.2 Higgsing T [U(3)] 15

3 3d partition functions 18
3.1 S 3 partition function 18
3.2 ’t Hooft operators 20
3.3 Interfaces 21
3.4 T [U(N )] 24
3.5 Holomorphic blocks 26
3.5.1 T [U(2)] theory 27
3.5.2 T [U(N )] theory 27
3.5.3 Givental J-functions 30
3.6 Difference equations in electric frame 30
3.6.1 T [U(2)] theory 30
3.6.2 T [U(3)] theory 31

4 5d/3d partition functions 32

4.1 Instanton counting 34
4.2 Ramified instantons 36
4.3 3d decoupling limit 38
4.3.1 U(2) theory 38
4.3.2 U(N ) theory 39
4.4 Wilson loops 39
4.5 5d/3d coupled system 40
4.5.1 U(2) theory 40
4.5.2 U(3) theory 41
4.5.3 Comments on normalizability of wavefunctions 43
4.5.4 Connection to 4d index of class S 43
4.6 5d theory coupled to 3d hypermultiplets 44
4.7 S-transformation 48

–i–
5 Gauge theories with chiral matter and 4d reduction 51
5.1 Complete flags and open Toda chains 51
5.1.1 Chiral limit of T [U(2)] 52
5.1.2 Effective twisted superpotential 53
5.1.3 Open Toda Lax matrix 54
5.2 Pure super Yang Mills theories with defects and closed Toda chains 54
5.2.1 Open Toda from decoupling limit 56
5.2.2 Closed Toda chain 57
5.3 Connections to Kapustin-Willett results 58

JHEP05(2015)095
5.4 4d/2d construction 58
5.4.1 Equivariant quantum cohomology 60
5.4.2 Coupled 4d/2d systems 61

6 Summary and outlook 61

6.1 Integrable systems 61
6.1.1 Calogero-Ruijsenaars-Dell family 61
6.1.2 Toda family 63
6.1.3 Gaudin-spin chain family 63
6.2 Open problems 64

A Conventions of special functions 66

A.1 Q-hypergeometric functions 66
A.2 Double sine functions 66
A.3 Theta functions 66

B Factorization of Tρ partition functions 67

C Perturbative 5d partition functions 69

1 Introduction

Supersymmetric gauge theories provide a rich source of inspiration for various branches of
mathematics. From a practical viewpoint, they can also provide a powerful set of tech-
niques to solve challenging mathematical problems using physics. The interplay between
supersymmetric gauge theories and mathematics is enhanced by introducing defects that
preserve some amount of supersymmetry.
In this work, we study 5d U(N ) N = 2 supersymmetric gauge theories with codimen-
sion two and codimension four defects and how they are connected to the quantization of
the integrable system associated to its Seiberg-Witten geometry [1, 2]. We focus on a class
of codimension two defects preserving N = 4 supersymmetry in three dimensions. They
can be described either as Gukov-Witten monodromy defects [3, 4] or by coupling to a class

–1–
of 3d N = 4 supersymmetric gauge theories. We will concentrate on the surface defect ob-
tained by introducing the most generic monodromy for the gauge field, or alternatively by
coupling to the 3d N = 4 theory T [U(N )] [5]. The codimension four defects are described
by supersymmetric Wilson loops.
As a preliminary step towards understanding and computing with surface defects, we
will first find a reformulation of the twisted chiral ring of a canonical deformation of the
T [U(N )] theory on S 1 × R2 , building on the work of [6]. We will show that the twisted
chiral ring relations are equivalent to the spectral curve of an associated classical N -body
integrable system, known as the complex trigonometric Ruijsenaars-Schneider (RS) sys-
tem. In addition, this provides a reformulation of the equivariant quantum K-theory of

JHEP05(2015)095
the cotangent bundle to a complete flag variety, via the Nekrasov-Shatashvili correspon-
dence [7, 8]. We will also explore the connection with quantum K-theory for more general
linear quiver gauge theories.
This classical integrable system can be quantized by turning on an equivariant param-
eter ǫ for rotations in R2 , otherwise known as the three-dimensional Omega background
S 1 × R2ǫ [9]. As it is technically simpler, we will first consider the squashed S 3 partition
function of T [U(N )] theory using results from supersymmetric localization [10, 11]. The
partition functions on S 1 × R2ǫ can then be obtained by factorization of the S 3 partition
function [12, 13]. We show that these supersymmetric partition functions are eigenfunctions
of the quantized trigonometric Ruijsenaars-Schneider system with the Planck constant pro-
portional to ǫ. The corresponding eigenvalues are given by supersymmetric Wilson loops
for background U(N ) flavor symmetries. The twisted chiral ring relations, or equivalently
the spectral curve of the classical integrable system, are reproduced in the semi-classical
limit ǫ → 0.
In coupling the three-dimensional theory T [U(N )] theory as a surface defect in 5d
U(N ) N = 2 gauge theory, the twisted chiral ring relations are deformed by an additional
complex parameter Q, which is related to the 5d holomorphic gauge coupling. According
to [14] it is expected that this deformation provides a presentation of the Seiberg-Witten
curve of the 5d theory on S 1 × R4 . The Seiberg-Witten curve of the 5d U(N ) N = 2
supersymmetric gauge theory is known to correspond to the spectral curve of the N -body
elliptic Ruijsenaars-Schneider system [15]. This is indeed a deformation of the trigonometric
RS system by an additional complex parameter Q.
In order to test this relationship, we will compute the 5d Nekrasov partition function of
N = 2 U(N ) supersymmetric gauge theory on S 1 × R4ǫ1 ,ǫ2 in the presence of surface defects
wrapping one of the two-planes S 1 × R2ǫ1 . This computation is performed by treating the
surface defect as a monodromy defect and applying the orbifolding procedure introduced
in [16, 17]. In order to check that the Gukov-Witten monodromy defect is reproducing
the same surface defect as coupling to T [U(N )] theory, we check that this computation
reproduces the S 1 × R2ǫ1 partition function of T [U(N )] theory in the limit Q → 0 where
the coupling to the 5d degrees of freedom is turned off. In particular, we note that the
Gukov-Witten monodromy parameters are identified with the Fayet-Iliopoulos parameters
of the 3d gauge theory supported on the defect.

–2–
After performing this preliminary check, we study the full Nekrasov partition function
on S 1 × R4ǫ1 ,ǫ2 as an expansion in the parameter Q. In the the Nekrasov-Shatashvili limit
ǫ2 → 0, we will show that the expectation value of the most generic surface defect is formally
an eigenfunction of the elliptic RS system. Furthermore, we find that the corresponding
eigenvalues are given by the expectation values of supersymmetric Wilson loops wrapping
S 1 in the 5d gauge theory. This computation provides a quantization of the Seiberg-
Witten geometry.
We will also study another type of codimension two defect by coupling directly to 3d
hypermultiplets [18, 19]. We focus on the simplest example where two free hypermultiplets
of U(2) flavor symmetry are coupled to the bulk gauge field of the 5d U(2) N = 2 gauge

JHEP05(2015)095
theory. We will show that the partition function of this coupled system solves an eigenfunc-
tion equation of the so-called two-body dual elliptic Ruijsenaars-Schneider system. Indeed,
the S-transformation of the 3d theory in [20] relates this partition function to the partition
function of the U(2) gauge theory with a monodromy defect.
The paper is organized as follows. In section 2 we will relate the twisted chiral ring of
T [U(N )] to the spectral curve of the classical trigonometric Ruijsenaars-Schneider system
and discuss connections to equivariant quantum K-theory. In section 3 we show that
partition functions on squashed S 3 and S 1 × R2 are eigenfunctions of the corresponding
quantized integrable system. Then in section 4 we explain how to compute the expectation
values of surface defects in N = 2 U(N ) gauge theory on S 1 × R4 and demonstrate that
they are eigenfunctions of the elliptic RS system. Later in section 5 we present a cursory
discussion of various limits and degenerations of the results presented in this paper. Finally,
in section section 6 we summarize the connections of this work to integrable systems and
discuss areas for further research.

2 Twisted chiral rings

In this section, we will study 3d N = 4 linear quiver gauge theories on S 1 ×R2 , deformed by
hypermultiplet masses, FI parameters, and with the canonical N = 2∗ mass deformation.
It was observed by Nekrasov and Shatashvili [7, 8] that the equations determining the
supersymmetric massive vacua on S 1 × R2 , or the twisted chiral ring relations, can be
identified with Bethe ansatz equations for a quantum XXZ integrable spin chain.
We will focus for the most part on the triangular quiver gauge theory: T [U(N )]. We
will reformulate the statement of its twisted chiral ring in terms of the spectral curve of a
classical N -body integrable system known as the complexified trigonometric Ruijsenaars-
Schneider system. Alternatively, it can be viewed as a Lagrangian correspondence that
diagonalizes this classical integrable system. This reformulation will be important when
we come to couple this theory as a codimension two defect in five-dimensions.
We will explain how the corresponding statements for more general linear quivers can
be obtained by a combination of Higgsing and mirror symmetry, and demonstrate this in
a simple example. We will also briefly discuss connections to results in the mathematical
literature on the equivariant quantum K-theory of the cotangent bundles to partial flag
manifolds.

–3–
N1 N2 ..................... NL

M1 M2 ML

Figure 1. 3d N = 4 quiver gauge theory of type AL with gauge group U(N1 ) × · · · × U(NL ) and
Mi fundamental hypermultiplets at i-th gauge node.

JHEP05(2015)095
2.1 The Nekrasov-Shatashvili correspondence

Three-dimensional theories with N = 4 supersymmetry have SU(2)H ×SU(2)C R-symmetry

and flavor symmetries GH × GC acting on the fields parametrizing the Higgs and Coulomb
branches respectively. For the purpose of this paper, it is important to turn on a canonical
deformation preserving only N = 2 supersymmetry. The corresponding U(1)R is the diago-
nal combination of Cartan generators of SU(2)H × SU(2)C . The anti-diagonal combination
becomes an additional flavor symmetry U(1)ǫ with real mass parameter ǫ. Furthermore, we
turn on real deformation parameters by coupling to N = 2 vectormultiplets for GH × GC
and giving a vacuum expectation value to the real scalar. In a UV description, these de-
formation parameters enter as real hypermultiplet masses denoted typically by m and FI
parameters denoted by t. We refer to this setup as N = 2∗ supersymmetry. We refer the
reader to [6] for a more complete description of this setup.
Here, we focus on theories with a UV description as a linear quiver with unitary gauge
groups. Our notation is summarized in figure 1. It is convenient to introduce a U(1)
symmetry acting trivially such that the Higgs branch symmetry is given by U(M1 ) × · · · ×
(j) (j)
U(ML ). The corresponding mass parameters are denoted by {m1 , . . . , mMj }. Similarly, we
introduce an additional topological U(1) so that the Coulomb branch symmetry manifest in
the UV description is U(1)L+1 with corresponding parameters {t1 , . . . , tL+1 }. The physical
FI parameter at the j-th gauge node is tj+1 − tj . This symmetry can be enhanced by
monopole operators up to a maximum of U(L + 1) in the IR.
We will focus on he twisted chiral ring of the effective 2d N = (2, 2) theory obtained
by compactifying on a circle of radius R. In this case, the real deformation parameters are
complexified by background Wilson lines wrapping the circle and behave as twisted masses
in the language of N = (2, 2) supersymmetry.
For generic deformation parameters there is a discrete set of massive supersymmetric
vacua on S 1 ×R2 each with an associated effective twisted superpotential W (i) (m, t, ǫ). This
is a holomorphic function that is independent of superpotential and gauge couplings. In
the UV theory one can integrate out three-dimensional chiral multiplets to find an effective
twisted superpotential W(m, t, ǫ, s) for the dynamical vectormultiplets. The supersymmet-
ric vacua are then solutions to

∂W
exp 2πR = 1, (2.1)
∂si

–4–
which can be identified with the twisted chiral ring relations of the effective two-
dimensional theory.
The effective twisted superpotential of the generic linear quiver shown in figure 1 is
given by
XL Ni
iδj X
W(s, m, t, ǫ) = ti − ti+1 + s(i)
α + W1-loop (s, m, ǫ)
2R
i=1 α=1
L X j Mi
(2.2)
X X
+ ti+1 ma .
j=1 i=1 a=1

JHEP05(2015)095
The first term includes contributions from the FI parameters at each node together with
phase δi = Mi + Ni−1 + Ni + Ni+1 − 1.1 The second term includes the 1-loop contributions
from the KK tower of chiral multiplets. The basic building block of the 1-loop contributions
is the contribution ℓ(m) from a three-dimensional chiral multiplet of mass m, which is a
solution of the differential equation 2πR ∂m ℓ(m) = log(2 sinh πRm). We refer the reader
to [6] for an explicit expression. The final term is included to ensure that mirror symmetry
acts straightforwardly in the presence of the spurious U(1) symmetries.
As the imaginary components of the twisted mass parameters are periodic, it is con-
venient to introduce exponentiated parameters
(i) (i)
µ(i)
a =e
2πRma
, τi = e2πRti , σα(i) = e2πRsα , η = eπRǫ . (2.3)

With this notation, the equations for the supersymmetric vacua are

τi Y ησα(i) −σβ(i−1)
Ni−1 Ni
Y (i) (i)
η −1 σα −ησβ Y ησα(i) −σβ(i+1) Y
Ni+1 Mi (i)
ησα − µa
(i)

(i−1) (i)
· (i) (i)
· (i+1) (i)
· (i) (i)
= (−1)δi ,
τi+1 −1
−σα β6=α η σβ −ησα β=1 ησβ −σα a=1 ηµa −σα
β=1 ησβ
(2.4)
for all i = 1, . . . , L. It was observed by Nekrasov and Shatashvili [7, 8] that these equations
can be identified with the Bethe equations of a quantum integrable XXZ spin chain. A
complete dictionary for linear quivers can be found in [6].
In order to write down the twisted chiral ring it is necessary to introduce a generating
(i)
function for the gauge invariant combinations of the σα ’s. For this purpose, we introduce
an auxiliary parameter u and monic polynomials
Ni
Y Mi
Y
Qi (u) = (u − σα(i) ) , Pi (u) = (u − µ(i)
a ). (2.5)
α=1 a=1

The equations for supersymmetric vacua (2.4) can be expressed in terms of these polyno-
mials as
−− +
τi Pi+ Q+
i−1 Qi Qi+1
η −∆i ++ − = −1 , (2.6)
τi+1 Pi− Q−
i−1 Qi Qi+1
(i)
where the polynomials are understood to be evaluated at u = σα for α = 1, . . . , Ni .
We defined ∆i = Mi + Ni+1 + Ni−1 − 2Ni to be the 1-loop contribution to the scaling
1
Sign conventions are slightly altered compared to [6].

–5–
1 2 N-1 N

Figure 2. A Lagrangian description of the T [U(N )] theory consists of a sequence of gauge groups
U(1) × · · · × U(N − 1) with bifundamental matter and N hypermultiplets at the final node.

dimensions of monopole operators charged under i-th gauge group. The superscripts on
the polynomials are shorthand for multiplicative shifts of the arguments by η, for example
−
Q+

JHEP05(2015)095
−1
i (u) = Qi (ηu), Qi (u) = Qi (η u). The twisted chiral ring relations for gauge invariant
(i)
combinations of σα ’s are given by expanding equations (2.6) in u.
In what follows, it will be useful to introduce another slightly less familiar
Pi−1reformulation
of the twisted chiral ring.2 We first rescale the FI parameters by τei = τi η j=1 ∆j to absorb
the dependence on ∆j . Then we introduce the polynomial equations

τei+1 Q+ e − ei Q− Q
e + = (e
i Qi − τ i i τi+1 − τei )Pi Qi−1 Qi+1 , (2.7)

where Q e i (u) are auxiliary polynomials of rank Mi + Ni−1 − Ni + Ni+1 . To recover equa-
tions (2.6) for the supersymmetric vacua we shift the argument this polynomial equation by
(i)
η ± and evaluate both at roots σα of Qi (u). Dividing one equation by the other, the com-
bination (e e i (u) cancel out and we reproduce (2.4).
τi+1 − τei ) and the auxiliary polynomials Q
The twisted chiral ring relations are given by expanding equations (2.7) in u.
Finally, 3d N = 4 theories have a remarkable duality known as mirror symme-
try [21, 22]. This can be understood from brane constructions in Type IIB String the-
ory [23], where it is realized as the S-duality. Mirror symmetry acts in quite a non-trivial
manner on the data (Ni , Mi ) of the linear quiver, which is spelled out in reference [6].
Mirror symmetry interchanges mass parameters and FI parameters of the quivers and also
acts non-trivially on the N = 2∗ mass deformation. Schematically, we have.3

µ a ↔ τa η ↔ η −1 (2.8)

where the check symbol designates parameters of the dual theory.

2.2 Quantum/classical duality

As mentioned above, the equations for supersymmetric vacua can be identified with the
Bethe equations for a quantum integrable spin chain. Remarkably, the same system of
equations are related to a second, classical integrable system of interacting relativistic par-
ticles in one dimension — the complexified trigonometric Ruijsenaars-Schneider system [6].
This correspondence is most straightforward to understand in the case of the triangular
quiver T [U(N )] (see figure 2) — this will be our main example throughout this paper.
2
We thank Davide Gaiotto for exhibiting us this calculation in a sample example.
3
Here the mirror map for η has a different sign compared to [6], where η was mapped onto −η −1 .

–6–
The effective twisted superpotential (2.2) of this theory is given by

N
X X
j
iδj
W= tj − tj+1 + s(j)
α
2R
j=1 α=1
j X
j+1 ǫ ǫ
N
X −1 X
(j+1) (j+1)
+ ℓ s(j)
α − s α′ + + ℓ −s(j)
α + s α′ + (2.9)
2 2
j=1 α=1 α′ =1
N
X −1 X (j)

+ ℓ s(j)
α − s α′ − ǫ ,
j=1 α6=α′

JHEP05(2015)095
(N )
where δj = j − 1,4 and we define sα = mα and tN +1 = 0 to simplify notation. For
this theory, the Higgs and Coulomb branch symmetries are both U(N ) and we have corre-
sponding exponentiated mass µj and FI parameters τj with j = 1, . . . , N . This theory is
invariant under mirror symmetry with the transformation µj ↔ τj , η ↔ η −1 .
By introducing the conjugate momenta to µj and τj

∂W ∂W
pjµ = exp 2πR , pjτ = exp 2πR , (2.10)
∂mj ∂tj

we provide canonical coordinates on two copies Mµ and Mτ of the cotangent bundle to

(C∗ )N with the following holomorphic symplectic forms

N N
X dµj dpjµ X dτj dpjτ
Ωµ = ∧ , Ωτ = ∧ . (2.11)
j=1
µj pjµ j=1
τj pjτ

This is the phase space of our complex classical integrable system. The defining equations
for the conjugate momenta (2.10) sweep out a complex Lagrangian in the product L ⊂
Mµ × Mτ with holomorphic symplectic form Ωµ − Ωτ and generating function given by
the on-shell twisted effective superpotential W(mi , tj , ǫ).
It is straightforward to find an explicit description of the Lagrangian L for T [U(2)]
theory. The supersymmetric vacua equations read as follows

τ1 (µ1 − ησ1 ) (µ2 − ησ1 )

= 1. (2.12)
τ2 (ηµ1 − σ1 ) (ηµ2 − σ1 )

The conjugate momenta for the FI terms are as follows

µ1 µ2
p1τ = σ1 , p2τ = , (2.13)
σ1

and for the masses

ηµ1 − σ1 ηµ2 − σ1
p1µ = τ2 , p2µ = τ2 . (2.14)
µ1 − ησ1 µ2 − ησ1
4
In the notations of (2.4) Nj = j, so the this definition of δj is consistent with the above conventions
modulo two.

–7–
Given the above definitions of conjugate momenta vacua equation (2.12) can be presented
in two equivalent ways. First, as

τ1 η − τ2 η −1 1 τ2 η − τ1 η −1 2
pτ + pτ = µ 1 + µ 2 , p1τ p2τ = µ1 µ2 , (2.15)
τ1 − τ2 τ2 − τ 1
and, second, as

µ1 η −1 − µ2 η 1 µ2 η −1 − µ1 η 2
pµ + pµ = τ 1 + τ 2 , p1µ p2µ = τ1 τ2 . (2.16)
µ1 − µ2 µ2 − µ1

JHEP05(2015)095
In (2.15) the combinations appearing on the left are the Hamiltonians of the complex
trigonometric RS system for two particles with positions τ1 , τ2 and momenta p1τ , p2τ . The
right hand sides are independent of the momenta p1µ , p2µ so the Lagrangian correspondence
L diagonalizes the system. From this perspective, W(ma , ta , ǫ, sα ) evaluated on super-
symmetric vacua are solutions of the relativistic Hamilton-Jacobi equation. The evident
symmetry under µj ↔ τj and η ↔ η −1 means this Lagrangian also diagonalizes the same
system with the coupling inverted η → η −1 .
To eliminate dynamical vectormultiplet scalars σα from the supersymmetric vacuum
equations in favor of the conjugate momenta pjµ or pjτ in the case N > 2 is rather non-trivial
task. Below we demonstrated that one can do this in two equivalent ways. First as
N
Y
det (u − L(τ, pτ , η)) = (u − µj ) , (2.17)
j=1

and, second, as
N
Y
−1

det u − L(µ, pµ , η ) = (u − τj ) . (2.18)
j=1

In both relations above

N
Q
(αi β − αk β −1 )
k6=j
Lij (α, pα , β) = N
pjα , (2.19)
Q
(αi − αk )
k6=i

is the Lax matrix for the N -body complex trigonometric RS system. One can clearly see
that (2.17) and (2.18) are related to each other by mirror symmetry map (2.8). Therefore,
somewhat artificially, we can refer to the former relation as written in the ‘electric’ frame,
where eigenvalues of L are related to masses µj , and to the latter relation as presented
in the ‘magnetic’ frame, in which, using mirror frame variables, the eigenvalues of L are
identified with FI parameters τj .
As we mentioned above, in order to understand why (2.17) and (2.18) are true it is
convenient to re-formulate the supersymmetric vacuum equations arising from this twisted
superpotential. We introduce monic degree-j polynomials Qj (u), Q e j (u) for each node
j = 1, . . . , N of the quiver. Note that we treat the matter polynomial in a uniform manner,

–8–
(N ) e
that is we define ma = σa and hence P (u) = QN (u). With this definition, we have QQ
equations (2.7)
τj+1 Q+ e− − e+
j Qj − τj Qj Qj = (τj+1 − τj )Qj−1 Qj+1 . (2.20)

Note that in this case ∆j = 0 and there is no need to redefine τj . The original supersym-
metric vacuum equations are obtained by shifting arguments in the above by η ± , evaluating
(j) e j (u). They can be
at the roots σα of Qj (u) and eliminating the auxiliary polynomials Q
expressed uniformly as
+ −− +
τj Qj−1 Qj Qj+1
= −1 , (2.21)
τj+1 Q− ++ −
j−1 Qj Qj+1

JHEP05(2015)095
(j)
evaluated on the roots u = σα .

2.2.1 Electric frame

(j)
Firstly, we explain how to eliminate the σα in favor of the momenta conjugate to the FI
e
parameters (2.17). For this purpose, we will set up an inductive procedure to solve the QQ
equations recursively node by node. We first note that by the definition (2.10)

Qj (0)
pjτ = − (2.22)
Qj−1 (0)

and hence by evaluating the QQ e equation at u = 0 we can immediately solve for the
constant terms in the polynomials Qj (u) and Q̃j (u) as follows

Qj (0) = (−1)j p1τ . . . pjτ

(2.23)
e j (0) = (−1)j p1τ . . . pj−1
Q τ pτ
j+1
.

Now, given the polynomials Qi (u) and Q e i (u) for i ≤ j we can determine Qj+1 (u) from the
QQ̃ equation (2.20). Then, by shifting j → j + 1 in the same equation and evaluating it on
(j)
the j roots σα of Qj (u) we have just enough data to determine the j unknown coefficients
e j+1 (u).
in Q
To illustrate this process, let us perform the first interation explicitly. It is convenient
to introduce the monic degree one polynomials qi = u − piτ . Then from equation (2.23)
we have Q1 (u) = q1 (u) and Q e 1 (u) = q2 (u). Then, from equation from (2.20) with j = 1,
we find
τ2 q1+ q2− − τ1 q1− q2+
Q2 = . (2.24)
τ2 − τ1
Now, evaluating equation (2.20) with j = 2 on the root of the polynomial Q1 (u) it is
e 2 (u) and hence find
straightfoward to compute the coefficient of the linear term in Q
+ − − +
e 2 = τ3 q 1 q 3 − τ1 q 1 q 3 .
Q (2.25)
τ3 − τ1

We can now immediately compute the polynomial Q3 (u) using equation (2.20) with j = 2

–9–
We have implemented this procedure to many orders in j and found experimentally
that the solution can be expressed as follows. We introduce the following j × j matrices
   
qij−1
1
τi1 qij−3
1
· · · τij−1
1
qi1−j
1
1 τi1 · · · τij−11
 . .. ..  . . . 
Mi1 ,...,ij =  .. .
..
. . , M
(0)
=  .. .. . . ...  , (2.26)
  i1 ,...,ij  
j−1 j−3 j−1 1−j j−1
q i j τi j q i j · · · τi j q i j 1 τ i j · · · τi j

where we define qi = u−piτ and we remind the reader that superscripts are not exponentials
but shifts of the argument by η. Then the solution is given by a ratio of Vandermonde-like
determinants

JHEP05(2015)095
det M1,...,j det M1,...,j−1,j+1
Qj (u) = , Qe j (u) = . (2.27)
(0) (0)
det M1,...,j det M1,...,j−1,j+1

Solutions of this form for similar functional equations have appeared in the integrability
literature. We expect that these techniques could be used to prove the solution we have
found (see e.g. [24]).
Since all polynomials qi are monic of degree one the above ratios can be simplified and
(0)
by inverting the matrix M1,...,N , the ratio of determinants can be reexpressed as a single
spectral determinant
QN (u) = det u − L , (2.28)
where
N
Q
τi η − τk η −1
k6=j
Lij = N
pjτ , (2.29)
Q
(τi − τk )
k6=i
is the Lax matrix of the N -body complex trigonometric RS system. At the final stage of
N
Q
the recursion, the polynomial QN (u) becomes the matter polynomial P (u) = (u − µj ),
j=1
providing us with the required relation (2.17). By expanding both sides of (2.17) in u we
find explicitly the Hamiltonians
N
X

det u − L(τi , piτ , η) = Tr (τi , piτ , η)ur , (2.30)
r=0

and their eigenvalues

N
Y N
X
(u − µj ) = χr (µi )ur , (2.31)
j=1 r=0
Thus we can explicitly write the full set of conserved charges for trigonometric RS system
Tr (τi , piτ , η) = χr (µi ), or, more explicitly as
X Y τi η − τj η −1 Y X Y
pkτ = µk , (2.32)
τi − τj
I⊂{1,...,N } i∈I k∈I I⊂{1,...,N } k∈I
|I|=r j ∈I
/ |I|=r

where r = 0, 1, . . . , N .

– 10 –
2.2.2 Magnetic frame
Let us now look at the other presentation of twisted chiral ring (2.18). Now we want to
(j)
eliminate σα in favor of the momentum conjugate to the masses, pµj . In this case, it will
not be possible to provide an argument that lands directly on the Lax matrix formulation
of the complex trigonometric RS model. Instead, we attempt to verify the mirror equations
X Y µi η −1 − µj η Y X Y
pkµ = τk , (2.33)
µi − µj
I⊂{1,...,N } i∈I k∈I I⊂{1,...,N } k∈I
|I|=r j ∈I
/ |I|=r

related to those above by τj ↔ µj and η ↔ η −1 .

JHEP05(2015)095
Let us first consider the first independent Hamiltonian with r = 1. The momentum
conjugate to the masses (2.10) can be expressed in terms of the polynomials Qj (u) as
follows
(N )
α
Q+ (σ )
−1 N −1 N −1 α
pµ = τN (η ) (N )
, (2.34)
Q−
N −1 (σα )
(N )
where we remind the reader that σα = µα . It is now straightforward to see that the first
Hamiltonian can be expressed as a contour integral
N
" # I
X Y µα η −1 − µβ η
α η2 du Q−− Q+ (u)
N (u) N −1
pµ = τN , (2.35)
µα − µβ 1 − η 2 CN u QN (u) Q− N −1 (u)
α=1 β6=α

where the contour CN surrounds the roots of QN (u) i.e. the masses µα . Our proposition
is that this contour integral evaluates to τ1 + · · · + τN .
We prove this proposition by induction. To perform the inductive step, we contract
the contour CN such that it surrounds the roots σαN −1 of Q− N −1 (u), u = 0 and u = ∞,
(N −1)
then eliminate the dependence on QN (σα ) using the supersymmetric vacuum equa-
tions (2.21), and then express the result once again as contour integral. Performing these
steps, we find
I I
η2 du Q−− Q+ (u)
N (u) N −1 η2 du Q− Q++ (u)
N (u) N −1
τN = τ N − τ N
1 − η 2 CN u QN (u) Q− N −1 (u)
1 − η 2 CN −1 u Q+ Q
N (u) N −1
(u)
I −− +
η2 du QN −1 (u) QN −2 (u)
= τN + τN −1
1 − η 2 CN −1 u QN −1 (u) Q− N −2 (u)
= τN + τN −1 + . . . + τ1 , (2.36)
as required. In the second line above contour CN −1 surrounds only roots of QN −1 (u).
The argument for the Hamiltonian appearing at order ur proceeds in a similar manner.
We first express the Hamiltonian as a contour integral
Q
I (um − un )
X Y µi η −1 − µj η Y du1 dur m6=n
k
pµ = ... r
µi − µj CN u 1 ur Q
I⊂{1,...,N } i∈I k∈I (η −2 um − un )
|I|=r j ∈I
/ m,n=1 (2.37)
r
r
Y Q−− Q+ (u )
N (um ) N −1 m
× τN .
QN (um ) Q−
N −1 (um )
m=1

– 11 –
where the contour CN surrounds the poles arising from the denominators QN (um ). Note the
Q
critical role of the numerator factor (um −un ) in ensuring that the non-zero residues are
m6=n
labelled by sets i1 < · · · < ir . This contour integral is the path integral of a supersymmetric
gauged quantum mechanics on the circle, an observation we explain further below. Our
claim is that this contour integral evaluates to
X
χr (τ ) = τi 1 · · · τi r . (2.38)
i1 <...ir

To prove this statement, we again proceed by induction. The inductive step depends on

JHEP05(2015)095
the following contour integral identity
Q
I (um − un ) r
du1 dur m6=n r
Y Q−− Q+ (u )
N (um ) N −1 m
... r τN
CN u 1 ur Q Q (u ) Q− N −1 (um )
(η −2 um − un ) m=1 N m
m,n=1
Q
I (um − un ) r−1
du1 dur−1 m6=n Y Q−− +
N −1 (um ) QN −2 (um )
r−1
= τN ... τ N −1
CN −1 u1 ur−1 r−1Q Q (u ) Q− N −2 (um )
(η −2 um − un ) m=1 N −1 m
m,n=1 (2.39)
Q
I (um − un ) r
du1 dur m6=n r
Y Q−− +
N −1 (um ) QN −2 (um )
+ ... r τN −1
CN −1 u1 ur Q Q (u ) Q− N −2 (um )
(η −2 um − un ) m=1 N −1 m
m,n=1

= τN χr−1 (τ1 , . . . , τN −1 ) + χr (τ1 , . . . , τN −1 )

= χr (τ1 , . . . , τN ) ,

which we have checked in numerous examples.

We expect that this expression as well as (2.35) can be interpreted as partition functions
of quantum mechanics on the 1d supersymmetric defect on S 1 coupled to the 3d gauge
theory on S 1 × R2 . The integral relation (2.39) can be interpreted as an identity between
the partition functions of 1d defects coupled to neighboring nodes of the quiver. This
observation requires further study.

2.3 Line operators and interfaces in N = 4 SYM

Many of the computations presented in the preceding section, in particular the connections
to classical integrable systems, have a useful interpretation in terms of interfaces between
copies of four-dimensional U(N ) N = 2∗ theory.
The starting point for this construction is the moduli space of vacua M of U(N )
N = 2∗ theory on S 1 × R3 . There is always a region at infinity in the moduli space where
the gauge group is broken to the maximal abelian subgroup U(1)N and complex coordinates
(αi , pαi ) valued in (C∗ )2N corresponding to complexified electric and magnetic Wilson lines
in each abelian factor, complexified by vectormultiplet scalars. Classically, we would have
Mcl = (C∗ )2N , which is the phase space of the complex trigonometric RS system [6].

– 12 –
The quantum corrected moduli space M in the appropriate complex structure is given
by the space of GL(N ) flat connections on a torus with puncture. This can be described by
the holonomies A and B around the two cycles of the torus, which must obey ABA−1 B −1 =
E where E has eigenvalues η −2 , . . . , η −2 , η 2N −2 . Remarkably, these equations can be solved
in terms of coordinates (αi , pαi ) such that
X Y
TrΛr (A) = αi
I⊂{1,...,N } i∈I
|I|=r
X Y αi η −1 − αj η Y (2.40)
TrΛr (B) = piα .

JHEP05(2015)095
αi − αj
I⊂{1,...,N } i∈I i∈I
|I|=r j ∈I
/

In particular, there is a choice of gauge where B becomes the Lax matrix of the com-
plexified trigonometric RS system. It can be confirmed by localization computations that
TrΛr (A) and TrΛr (B) correspond respectively to BPS Wilson and ’t Hooft loops in the
anti-fundamental representations of U(N ) wrapping the S 1 .
Interfaces between two theories with moduli spaces ML and MR correspond to La-
grangian submanifolds L ⊂ ML × MR . Let us recall that the three-dimensional T [U(N )]
theory can be identified with the S-duality interface for the U(N ) N = 2∗ theory. In
this context the complex parameters of the three-dimensional theory (µi , piµ ) and (τi , piτ )
are identified with the Darboux coordinates for the moduli space on either side of the
interface. The corresponding Lagrangian submanifold L is then described precisely by the
equations (2.17) and (2.18). These relations are interpreted as Ward identities for line
operators at the interface: a ’t Hooft loop approaching from one side is equivalent to a
Wilson loop approaching from the other. This is expected since Wilson loops and ’t Hooft
loops are interchanged under S-duality (cf. [25]).
Similarly, the contour integral relations in equation (2.39) can be interpreted as bound-
ary Ward identities for ’t Hooft loops at an interface between N = 2∗ theories with gauge
groups U(N ) and U(N − 1). This interpretation are discussed further in section 3.3.

2.4 Quantum equivariant K-theory

From the work of Nekrasov and Shatashvili [7, 8] it is known that the twisted chiral ring
of 3d N = 4 quiver gauge theories on S 1 × R2 represent the equivariant quantum K-theory
ring of the Higgs branch. For generic linear quivers the Higgs branch is a quiver variety
introduced by Nakajima [26–28]. For our main example T [U(N )] this is the cotangent
bundle to the N -dimensional complete flag variety.
It is also known from the work of Givental and collaborators [29–31] that there is a
deep connection between equivariant quantum cohomology/quantum K-theory and classi-
cal many-body integrable systems. The results of this sections can be interpreted in this
light: the equivariant quantum K-theory of the cotangent bundle to a N -dimensional com-
plete flag is determined by the N -body trigonometric RS integrable system. Later in [32]
the results of Givental and Lee were proved using different methods which utilize the action
of quantum groups. This method is also applicable to affine quiver varieties of type-A.

– 13 –
Twisted chiral ring of 3d theory Quantum Equivariant K-theory
FI parameters τi Quantum deformation parameters
N = 2∗ mass η Equivariant parameter for U(1)ǫ
Hypermultiplet masses µi Equivariant parameters for U(N )

Table 1. Identification of mass parameters of 3d quiver theories and equivariant quantum K-rings.

We can therefore formulate the following

JHEP05(2015)095
Proposition 2.1 The T -equivariant quantum K-ring of the cotangent bundle to the com-
plete N-dimensional complex flag variety is given by

QKT• (T ∗ FN ) ≃ C (piτ )±1 , τi±1 , η ±1 , µ±1
i /I , i = 1, . . . , N , (2.41)

where ideal I is given by relations (2.32) and T is the maximal torus of U(N ) × U(1) with
equivariant parameters µ1 , . . . µN for U(N ) and equivariant parameter η for U(1). The
correspondence between physical and geometrical parameters is summarized in table 1.
Note that we have kept equivariant parameters appropriate for U(N ) (rather than SU(N ))
global symmetry. For SU(N ) symmetry one needs to add additionally impose the conditions
p1τ . . . pN
τ = µ1 . . . µN = 1 so that one equivariant parameter and one generator are excluded.
In the limit η → 1 twisted chiral ring relations (2.32) have the simple solutions piτ =
µσ(i) for any permutation σ. In terms of trigonometric RS model, this limit describes N
non-interacting relativistic particles. In terms of 3d gauge theory, this describes the point
in the parameter space where the supersymmetry is enchanted from N = 2∗ to N = 4.
Recently we became aware of reference [33] where quantum K-rings of flag varieties
were discussed. The conjecture the authors make about the quantum K-ring is in agreement
with Proposition 2.1.

2.5 More genetic quiver varieties

So far we have considered only the triangular quiver T [U(N )] which has the property of
being self-dual under three-dimensional mirror symmetry. We now briefly discuss how
aspects of more general linear quivers can be studied through a combination of Higgsing
and mirror symmetry.
For generic hypermultiplet masses and FI parameters, there are only discrete massive
supersymmetric vacua on S 1 × R2 . However, by tuning the mass parameters ma (or FI
parameters ta in the mirror frame) one can open up a Higgs branch. Moving out onto this
branch and flowing to the infrared we can partially Higgs the theory to obtain a new quiver.
Using Type IIB brane constructions one can develop an algorithm for constructing a
generic quiver from T [U(N )] by repeating this procedure together with mirror symmetry. In
terms of the parameter space of hypermultiplet mass deformations, Higgsing corresponds
to restriction to a certain subvariety specified by the Higgs branch locus of the original
quiver theory (see [6] for examples). Below we shall illustrate the idea using couple of
simple examples.

– 14 –
2.5.1 Higgsing T [U(2)]
The twisted chiral ring relation of T [U(2)] theory is equivalent to the spectral curve of the
2-body trigonometric RS model as presented in equation (2.15). These equations reproduce
the Bethe equations upon inserting the on-shell values of the momenta (2.13). In order to
Higgs theory we impose we impose the condition

η −1 µ1 = ηµ2 = σ1 := µ (2.42)

from which we find p1τ = p2τ = µ from the definition of the momenta (2.13). Equivalently,
inserting the condition (2.42) into the spectral curve (2.15) we find

JHEP05(2015)095

−1
µη p1τ µ p1τ η
η+η (τ1 − τ2 ) = τ1 + − τ2 + , (2.43)
p1τ ηµ ηp1τ µ

which determines p1τ = µ. After Higgsing T [U(2)] theory becomes free, hence its chiral ring
relations are trivial.

2.5.2 Higgsing T [U(3)]

The twisted chiral ring of T [U(3)] theory computes the equivariant quantum K-ring of T ∗ F3 .
In what follows we shall study vacua of T [U(3)] theory subject to a Higgsing condition sim-
ilar to (2.42) and compute quantum K-ring for the Nakajima quiver variety corresponding
to the Higgsed theory. From 3d mirror symmetry it can be shown that the quantum K-ring
of the latter variety is canonically isomorphic to the quantum K-ring T ∗ P2 .
The T [U(3)] theory corresponds to the three-particle trigonometric RS model which is
described by the Hamiltonians given in (2.32). As we Higgs the 3d theory we impose (see
figure 3)
(2)
η −1 µ1 = ηµ2 = σ1 := µ , (2.44)
(2)
where σ1 is one of the Coulomb branch parameters for the U(2) gauge group (second node
of the quiver). The theory is therefore reduced to the A2 quiver with labels (1, 1)(1, 1);
its mirror is A1 quiver with labels (1, 3), or in other words (which lead to T ∗ P2 ), the U(1)
theory with three flavors.
Let us first describe the chiral ring relations for the A2 quiver. Constraint (2.44)
affects the r.h.s. of (2.32) by imposing the relationship between the masses and the l.h.s.
by restricting some of the conjugate momenta pτ
(2) (2)
σ σ µ1 µ2 µ3
p1τ =σ (1)
, p2τ = 1 (1)2 , p3τ = (2) (2)
, (2.45)
σ σ1 σ2

as follows
(2)
µσ2 µµ3
p1τ = σ (1) , p2τ = , p3τ = . (2.46)
σ (1) σ2
(2)

which are consistent with the definition of the conjugate momenta for (1, 1)(1, 1) quiver,
where µ = ηµ2 is the mass of the hypermultiplet on the first node of the quiver. Therefore

– 15 –
1
2 2
3
3 1

1 2 1 1

3 1 1

JHEP05(2015)095
Figure 3. Higgsing T [U(3)] theory using Type IIB brane construction. Vertical black lines denote
NS5 branes along directions 012789, horizontal blue lines denote D3 branes along 0123 directions,
and oval red circles stand for D5 branes stretched along 012456 directions. The left side of the
figure describes T [U(3)] theory, whereas the right side shows how to obtain A2 quiver with labels
(1, 1)(1, 1) by Higgsing flavor branes 1 and 2. After applying the S-duality to the right figure D5
and NS5 branes switch roles, and D3 branes are self dual. The newly obtained structure of two
NS5 branes and three D5 branes describes U(1) theory with three flavors.

the only difference between chiral rings on the T [U(3)] (or (3, 2)(1, 0)) theory and the
(1, 1)(1, 1) theory — one simply imposes (2.44) on the mass parameters
T1 = (η −1 + η)µ + µ3 ,
T2 = (η −1 + η)µµ3 + µ2 ,
T3 = µ 2 µ 3 , (2.47)
where Tr are defined in (2.30). One can also explicitly check that the Bethe equations for
(1, 1)(1, 1) theory are equivalent to (2.47).
It is instructive to see how Higgsing applies in the magnetic frame, namely while using
µi and piµ in tRS Hamiltonians (2.33). Thus we want to derive vacua equations for the
Higgsed T [U(3)] together with the conjugate momenta piµ and compare them with the data
of the (1, 1)(1, 1) quiver.
Let us evaluate the effective twisted superpotential (2.9) for T [U(3)] theory at the locus
of the Higgs branch (2.44), which can be written in the original variables as
(2) ǫ
m1 = m2 + ǫ , s1 = m2 + . (2.48)
2
Using the twisted effective superpotential we can compute conjugate momenta to mass
parameters
(ηµ2 − µ3 ) (ηµ2 − σ1 ) (µ2 − ηµ3 ) (ηµ3 − σ2 )
p2µ = τ2 τ3 , p3µ = τ3 , (2.49)
(µ2 − ηµ3 ) (ησ1 − µ2 ) (ηµ2 − µ3 ) (ησ2 − µ3 )
(1) (2)
where we have substituted σ1 = σ1 and σ2 = σ2 . We can see that the above expressions
reproduce the corresponding conjugate momenta for the A2 quiver with labels (1, 1)(1, 1)
ηµ2 − σ1 ηµ3 − σ2
p(1)
µ = τ2 τ3 , p(2)
µ = τ3 , (2.50)
ησ1 − µ2 ησ2 − µ3

– 16 –
up to rational function
ηµ2 − µ3
A= , (2.51)
µ2 − ηµ3
(1) (2)
namely, p2µ = Apµ and p3µ = A−1 pµ . Note that −A provides a contribution of the free
fundamental hypermultiplet of mass m3 − m2 to the effective twisted superpotential. We
can interpret contact term (2.51) as coming from the global symmetry of the Higgs branch
of T [U(3)] theory. After Higgsing the dependence on µ1 (the position of particle 1 on the
right of figure 3) and its momentum disappears.
Let us now explicitly describe the quantum K-ring of T ∗ P2 using vacua equations.

JHEP05(2015)095
Since A2 quiver with labels (1, 1)(1, 1) is mirror dual to U(1) theory with three flavors we
can use vacua equations of the latter theory to derive the desired chiral ring relations. In
the mirror variables one gets
3
Y 3
Y
µ2 (ησ − τj ) + µ3 (ητj − σ) = 0 , (2.52)
j=1 j=1

together with momenta conjugate to the FI parameters

τ1 τ2 τ3
p2 = σ , p3 = , (2.53)
σ
and hence p2 p3 = τ1 τ2 τ3 . We can expand the vacuum equations in powers of σ and then
eliminate σ using p2 := p. By doing so (2.52) becomes

[µ23 ]−1 p3 − χ1 (τ )[µ23 ]0 p2 + χ2 (τ )[µ23 ]1 p − χ3 (τ )[µ23 ]2 = 0 , (2.54)

where we defined
η a µ2 − η −a µ3
[µ23 ]a = , (2.55)
µ2 − µ3
so [µ23 ]0 = 1. Equivalently we could have started with vacua equations for (1, 1)(1, 1)
quiver and using momenta defined in (2.50) derive (2.54).
With more experimentation we can find the general formula for the A1 quiver with N
fundamental hypermultiplets
N
X
(−1)a [µ]a−1 χa (τi ) pa = 0 , (2.56)
a=0

a −a
where [µ]a = η µ−ηµ−1 and χa stand for characters of the a-th antisymmetric tensor rep-
resentation of slN (for the trivial representation s0 = 1). Interestingly, the above formula
provides the classical limit of an N -th order difference equation for the q-hypergeometric
series N FN −1 . Indeed, if we promote p to a shift operator (2.56) is nothing but the func-
tional equation for quantum N FN −1 function. It is also well known that holomorphic blocks
of the U(1) theory with N flavors obey exactly the same equation.5
We can therefore make the following
5
It can presumably be interpreted as Ward identity for line operators.

– 17 –
Corollary 2.2 T -equivariant quantum K-ring of T ∗ PN is isomorphic to

QKT• (T ∗ PN ) ≃ C p±1 , (τi )±1 , η ±1 , µ±1 /I , i = 1, . . . , N , (2.57)

where ideal I is given by relation (2.56).

Note that in the limit when Kähler parameter vanishes µ → 0 we have [µ]a−1 → η a−1 .
Therefore, by redefining pa → η 1−a pa we obtain that the ideal given in (2.56) is reduced to

(p − τ1 ) · · · · · (p − τN ) , (2.58)

which is in the agreement with the known result about classical K-rings and cohomology

JHEP05(2015)095
rings of complex projective spaces. Note also that the dependence on η drops out sug-
gesting that the action of U(1)η is trivial in this case; which confirms the known fact that
equivariantly T ∗ PN retracts to PN .

3 3d partition functions

In section 2, we explained how twisted chiral ring of T [U(N )] theory is encoded in the
spectral curve of a classical integrable system of N interacting relativistic particles called
the trigonometric RS system.
In this section, we will quantize this classical integrable system by studying T [U(N )]
theories on a curved three-dimensional background. For most of this section, we concentrate
on squashed S 3 , where the quantization parameter is identified with squashing parameter
traditionally denoted by b (and, as we explain below, also b−1 ). In this background, the
conjugate momenta to τi and µi are promoted to difference operators piτ and piµ and the
spectral curve becomes an operator equation, which annihilates the partition function.
From the squashed S 3 partition function, we will obtain the partition function of T [U(N )]
on S 1 × R2 by factorization.

3.1 S 3 partition function

Let us begin by summarizing the supersymmetric partition function of 3d N = 4 gauge
theories on squashed S 3 , as computed using supersymmetric localization in [10, 11, 34].
This partition function depends on a squashing parameter b > 0 and real mass parameters
valued in the Cartan subalgebra of the global symmetry group G. The latter are introduced
by coupling to a background N = 2 vectormultiplet for G and giving a vacuum expectation
value to the real scalar. For the most part we consider U(N ) flavor symmetries and use
shorthand notation m ~ = (m1 , . . . , mN ).
As in the previous section, it is convenient to introduce exponentiated mass parameters.
The conjugate momenta now become elementary difference operators. For a U(N ) global
symmetry we define in this section

µj = e2πbmj pjµ = eib∂mj (3.1)

They obey the following relation

piµ µj = q δij µj piµ , (3.2)

– 18 –
2
where q = e2πib . We will also introduce a second set of exponentiated mass parameters
whose variables denoted by µ ej , pjµe and qe. They are defined in exactly the same fashion
except we replace b → b−1 .
The partition function can also depend on real FI parameters for U(N ) gauge groups.
The real FI parameter t associated to a U(N ) gauge group is a real mass parameter for
the topological U(1). The corresponding exponentiated FI parameters τ , τe and momenta
pτ , pτe are defined as above.
As in section 2, it will be important to turn on a real mass parameter m for the
diagonal combination U(1)ε ⊂ U(1)H × U(1)C breaking to N = 2∗ supersymmetry in three
dimensions. It is convenient to introduce the notation

JHEP05(2015)095
ε = b+ + im , ε∗ = b+ − im , (3.3)

together with the exponentiated parameters

η 2 = q e−2πibε = e2πib(b− −im) , (3.4)

where we define b± = (b ± b−1 )/2. We will not need to consider the conjugate momentum
for this symmetry.
Let us now summarize the building blocks of the squashed S 3 partition function of
3d N = 2∗ quiver gauge theories. They are built from bifundamental hypermultiplets,
vectormultiplets and FI parameters,
1. U(N1 ) × U(N2 ) bifundamental hypermultiplet:
N1 Y
Y N2 ∗
(1) (2) ε∗ (1) (2) ε (1) (2)
QN1 ,N2 (m ,m )= S + imi − imj S − imi + imj (3.5)
2 2
i=1 j=1

2. U(N ) vectormultiplet:
N
Y N
Y
−1
νN (m) = S(imi − imj ) S (ε + imi − imj )
i,j=1 i,j=1
i6=j
N
Y
N (N −1)
= (−1) 2 2 sinh πb(mi − mj ) 2 sinh πb−1 (mi − mj ) (3.6)
i<j
N
Y
× S (ε + imi − imj )
i,j=1

3. FI parameter for U(N ) symmetry

e−2πit(m1 +···+mN ) (3.7)

Turning off the N = 2∗ mass deformation ε → b+ , the contributions from the adjoint N = 2
chiral multiplet cancel in pairs and we recover the familiar N = 2 vectormultiplet measure
involving only the hyperbolic sine function. It is important to include these contribution
when m 6= 0.

– 19 –
Here we are using the double sine function S(z) = S2 (z|b, b−1 )−1 . This is meromorphic
in z with simple poles at z = mb + nb−1 for n, m ≤ 0 and simple zeroes for n, m ≥ 1. The
most important properties we will need are

S(z + b± ) = 2 sin(πb± z)S(z) ,

(3.8)
S(x)S(b+ − x) = 1 ,

where b+ ≡ b + b−1 . Further properties are summarized in appendix A.2.

The S 3 partition function is related to a quantization of the Lagrangian submanifold L
of section 2.2. The relevant quantization parameter ~ is related to the squashing parameter

JHEP05(2015)095
by ~ = 2πb2 (for the parameters µ, pµ ,. . . ) and ~ = 2πb−2 (for the parameters µ̃, p̃µ ,. . . ).
In the semi-classical limit b → 0, the S 3 partition function schematically has asymptotic
behavior Z
ZS 3 = dσ e −2πi W(σ) + ... (3.9)

where W(σ) is the effective twisted superpotential with the radius R replaced by the
squashing parameter b. In this limit, the conjugate momenta become pµ = e2πb∂m W and
hence matching their semi-classical counterparts.

3.2 ’t Hooft operators

The building blocks for 3d N = 4 quiver gauge theories have an interesting interplay with
a system of commuting difference operators, which quantize the classical Hamiltonians
studied in section 2. We will consider two sets of operators acting on the mass parameters
m = (m1 , . . . , mN ) of a U(N ) symmetry. The first set is defined by
X Y sin πb (ǫ − imij ) Y ib∂
Tr (m) = e mj
sin πb (−imij )
I⊂{1,...,n} i∈I j∈I
| I |=r j ∈I
/

X Y q 1/2 η −1 µi − ηq −1/2 µj Y (3.10)

= pjµ ,
µi − µj
I⊂{1,...,n} i∈I j∈I
| I |=r j ∈I
/

where r = 1, . . . , N . The second set, which we denote by Ter (m), is obtained by replacing
b → b−1 in the first line or replacing the exponentiated parameters by their tilded counter-
parts in the second line. It can be shown that the two sets of operators commute among
themselves.
Note that Tr (m) is expressed as a sum over the states of the anti-symmetric tensor
representation of U(N ) of rank r where the mass parameters (m1 , . . . , mN ) are shifted by an
amount proportional to the corresponding weight. Similarly, the operators with reversed
masses Tr (−m) are associated to the conjugate anti-symmetric tensor representation of
U(N ) of rank r.
This is no coincidence. The same difference operators appear in the computation of the
expectation value of supersymmetric ’t Hooft loops in N = 4 SYM theory with gauge group

– 20 –
U(N ) on squashed S 4 [35].6 For general squashing, the supersymmetric ’t Hooft loops can
be supported on two Hopf-linked circles corresponding to the operators Tr (m) and Ter (m)
respectively. As shown in references [37, 38], the expectation value of a supersymmetric
’t Hooft loop in the r-th anti-symmetric representation of U(N ) can be massaged into
the form Z
dN m νN (m) G(m, ǫ, τ ) [ Tr (m) · G(m, ǫ, τ ) ] (3.11)

with a similar equation for Ter (m).

The wavefunction G(m, ǫ, τ ) is expected to be the partition function of U(N ) N = 4
SYM on a hemisphere of squashed S 4 with Dirichlet boundary conditions at the boundary.

JHEP05(2015)095
There is a global U(N ) symmetry at the boundary whose real mass parameter is m. The
boundary is identified with the squashed S 3 geometry we consider in this section. The
supersymmetric partition function on S 4 is computed by including contributions from each
of the hemispheres, reintroducing a 3d N = 4 vectormultiplets with partition function
νN (m) and gauging.
The operators Tr (m) obey some remarkable properties in the way they interact with
the S 3 partition function of 3d N = 4 vectormultiplets and bifundamental hypermultiplets.
Each of these properties can be understood in terms of the action of ’t Hooft loops on
interfaces in 4d N = 4 SYM.

3.3 Interfaces
The first important property of the difference operators is that they are self-adjoint with
respect to the measure νN (m) on the Cartan subalgebra of U(N ) defined by the N = 4
vectormultiplet contribution. As a preliminary step we show that the vectormultiplet
measure obeys the difference equation
" #
Y sin πb(−imij − b) sin πb(ε + imij )
(piµ )−1 νN (m) = νN (m)
sin πb(imij ) sin πb(ε − imij − b)
j6=i
" # (3.12)
Y µi q −1/2 − q 1/2 µj q 1/2 η −1 µj − ηq −1/2 µi
= νN (m)
j6=i
µj − µi η −1 µi − ηq −1/2 µj

together with the same equation for the tilded variables. Let us now pick two functions
f (x) and g(x) and assume that they have no poles in the region −b < Im(x) < b. Then
by deforming the contour integration xi → xi − b for each mass parameter in the set i ∈ I
and using the above result, a short calculation shows that
Z Z
dν(m) f (m) [ Tr (m) · g(m) ] = dν(m) [ Tr (−m) · f (m)] g(m) . (3.13)

This relation can be used to ‘integrate by parts’ inside the contour integral expressions for
partition functions, provided no poles are crossed in deforming the contour.
6
The expectation value of ’t Hooft loops on a round four-sphere were computed in [35]. In the case of a
squashed four-sphere the expectation value can be computed under the assumption of the AGT correspon-
dence from Verlinde loop operators in Toda CFT [36].

– 21 –
This can be interpreted in terms of a trivial interface. Namely, the partition function
of U(N ) N = 4 theory on a four-sphere can be built from two copies of the hemisphere
partition function with Dirichlet boundary conditions by introducing a 3d N = 4 vector-
multiplet on the equator. The partition function is constructed by taking f (m) = G(m, ǫ, τ )
and g(m) = G(m, ǫ, τ ). The equation says that the ’t Hooft loop is unchanged on passing
through the boundary.
The second property states that the partition function of a U(N ) × U(N − M ) bifun-
damental hypermultiplet intertwines a difference operator with its decomposition under
the symmetry breaking pattern U(N ) → U(N − M ). In this case we proceed by example

JHEP05(2015)095
before stating the general result.
Let us begin by considering a bifundamental U(N ) × U(N ) hypermultiplet. The par-
tition function obeys

Tr (m(1) ) − Tr (−m(2) ) QN,N (m(1) , m(2) ) = 0 (3.14)

together with an isomorphic equation involving Ter (m). Firstly, acting on the partition
function of bifundamental hypermultiplets with the momenta piµ(1) and piµ(2) and using the
difference equation obeyed by the double sine function we have

n (1) (2) n (1) (2)

Y µi − η −1 µj Y µj − η −1 µi
piµ(1) Q = (1) (2)
Q, (piµ(2) )−1 Q = (1) (2)
Q.
1/2 η −1 µ −1/2 µ 1/2 η −1 µ −1/2 µ
j=1 q i −q j j=1 q j −q i
(3.15)
Using these results the first line of equation (3.14) is a consequence of the rational function
identity

X Y q 1/2 η −1 µ(1)
i − ηq
−1/2 µ(1)
j
Y (1)
µi − η −1 µj
(2)

(1) (1) (1) (2)

I⊂{1,...,N } i∈I µi − µj i∈I q 1/2 η −1 µi − q −1/2 µj
| I |=r j ∈I
/ j=1,...,N
(3.16)
X Y q 1/2 η −1 µ(2)
j − ηq
−1/2 µ(2)
i
Y (1)
µj − η −1 µi
(2)
= (1) (1) (1) (2)
.
I⊂{1,...,N } i∈I µj − µi i∈I q 1/2 η −1 µj − q −1/2 µi
| I |=r j ∈I
/ j=1,...,N

This has been proven for precisely the same purpose in reference [39]. An isomorphic
argument applies for the equation involving the operators Ter (m). Thus we see that the
partition function of U(N ) × U(N ) hypermultiplets intertwines the quantum hamiltonians
associated with the fundamental and anti-fundamental representations of U(N ).
We now consider the more interesting case of U(N ) × U(N − 1) bifundamental hyper-
multiplets. The partition function obeys
h i
Tr (m(1) ) − Tr (−m(2) ) − Tr−1 (−m(2) ) QN,N −1 (m(1) , m(2) ) = 0 (3.17)

together with an isomorphic equation involving Ter (m). This equation is a consequence of

– 22 –
the identity
(1) (1) (1) (2)
X Y q 1/2 η −1 µi − ηq −1/2 µj Y µi − η −1 µj
(1) (1) (1) (2)
(3.18)
I⊂{1,...,n} i∈I µi − µj i∈I q 1/2 η −1 µi − q −1/2 µj
| I |=r j∈{1,...,n}/I j=1,...,n−1
(2) (2) (1) (2)
X Y q 1/2 η −1 µj − ηq −1/2 µi Y µj − η −1/2 µi
= (1) (1) (1) (2)
I⊂{1,...,n−1} i∈I µj − µi i∈I q 1/2 η −1 µj − q −1/2 µi
| I |=r j∈{1,...,n−1}/I j=1,...,n
(2) (2) (1) (2)
X Y q 1/2 η −1 µj − ηq −1/2 µi Y µj − η −1 µi

JHEP05(2015)095
+ (1) (1) (1) (2)
.
I⊂{1,...,n−1} i∈I µj − µi i∈I q 1/2 η −1 µj − q −1/2 µi
| I |=r−1 j∈{1,...,n−1}/I j=1,...,n

(2) (2)
which can be obtained from equation (3.16) by replacing µN → µN γ and take the limit
γ → ∞+ . This corresponds to making N of the hypermultiplets very heavy and integrating
them out to recover the bifundamental of U(N ) × U(N − 1). The partition function now
intertwines a hamiltonian with the conjugate of its decomposition under U(N ) → U(N −1).
In terms of representations of U(N ) we have
(N −1)
Λr(N ) → Λ̄r(N −1) + Λ̄r−1
(N −1)
(3.19)
Λ̄r(N ) → Λr(N −1) + Λr−1 .

where we have denoted r-th anti-symmetric tensor power of the fundamental representation
(N ) (N )
by Λr and the conjugate representation by Λ̄r .
Although we will not need it for the triangular quiver T (U(N ) it is interesting to study
the partition function of U(N ) × U(N − M ) bifundamental hypermultiplets. The argument
proceeds by induction and the details can be found in [39]. The result of the computation
is that
 
min(r,M )
X
Tr (m(1) ) − Dim(η, Λ(M
s
)
) T(r−s) (−m(2) ) QN,N −M (m(1) , m(2) ) = 0 , (3.20)
s=0

where

Dim(η, Λ(M
s
)
) = χΛ(M ) (q/η 2 )M −1 , (q/η 2 )M −3 , . . . , (q/η 2 )1−M (3.21)
s

(M )
is the quantum dimension of the representation Λs of U(M ) with quantum parameter
η and in the formula we have defined χΛ (x1 , . . . , xM ) to be the Schur polynomial for
representation Λ of U(M ). In the limit η → 1 the quantum dimension becomes the ordinary
dimension. An example is the hamiltonian associated to the fundamental representation
of U(N ),
h i
T1 (m(1) )−T1 (−m(2) )− (q/η)M −1 +(q/η)M −3 +. . .+(q/η)1−M QN,N −M (m(1) , m(2) ) = 0
(3.22)
(2)
where we have used T0 (−m ) = 1.

– 23 –
Summarizing, we have found that the partition function of U(N ) × U(N − M ) bi-
fundamental hypermultiplets intertwines a difference operator associated to a antisym-
metric tensor representation of U(N ) with the conjugate of its decomposition under
U(N ) → U(N − M ). The coefficients in the expansion are the quantum dimensions as-
sociated to the number of times that representation appears in the decomposition with
quantum parameter η.
This equation has an interpretation in terms of an interface between two copies of
the N = 2∗ theory with gauge groups U(N ) and U(N − j) respectively. The hemisphere
partition functions are coupled by adding three-dimensional bifundamental hypermultplets
of U(N )×U(N −M ) and gauging these symmetries on the interface. We can then interpret

JHEP05(2015)095
the intertwining property as the mathematical statement of how a ’t Hooft loop of U(N )
decomposes into ’t Hooft loops of U(N − j) on passing through the interface.
That the ’t Hooft loops should decompose in the same way that characters of the
corresponding representations of U(N ) decompose under U(N − 1) is perhaps clearer in an
S-dual picture where the theories are coupled by a reduction of symmetry due to a Nahm
pole boundary condition. The S-dual of a ’t Hooft loop is a Wilson loop whose contribution
is a character. From this perspective, it is clear that a Wilson loop will decompose according
to the decomposition of the characters of representations under U(N ) → U(N − M ) when
brought to the interface.7

3.4 T [U(N )]
Let us now consider the partition function of T [U(N )], which we denote by ZU(N ) (m, t).
It depends on mass parameters m = (m1 , . . . , mN ) for the Higgs branch symmetry U(N )H
and FI parameters t = (t1 , . . . , tN ) for the Coulomb branch symmetry U(N )C . As before,
the FI parameter associated to the U(j) gauge group is tj − tj+1 .
This theory arises on an interface between two copies of the N = 2∗ theory in four
dimensions with gauge groups U(N ) and holomorphic couplings related by the S-duality
transformation τ → −1/τ . The coupling between the bulk and interface degrees of freedom
is performed by gauging the U(N )H symmetry on one side of the interface and the U(N )C
symmetry on the other. Therefore the mass parameters m and FI parameters t are identified
with the vacuum expectation value of the vectormultiplet scalar of the bulk theory on either
side of the interface.
It is expected that Wilson loops and ’t Hooft loops are interchanged by the S-duality
transformation τ → −1/τ . Therefore, we expect that acting with a ’t Hooft loop one one
side of the interface is equivalent to acting with a Wilson loop on the other. Concentrating
on the minuscule representations, this can be turned into the mathematical statement

(Tr (m) − Wr (t)) ZU(N ) (m, t) = 0 (3.23)

where X
Wr (t) = χr (τ1 , . . . , τN ) = e2πib(ti1 +...+tir ) . (3.24)
i1 <...<ir

7
We thank Davide Gaiotto for suggesting this interpretation.

– 24 –
In words, the partition function should be an eigenfunction of the ’t Hooft loop difference
operators whose eigenvalue if the expectation value of a Wilson loop. This is the quantized
counterpart of the ‘gauge invariant’ formulation of the Bethe equations for supersymmetric
vacua which we have found earlier. As always, in the context of the S 3 partition function
there is an isomorphic statement for the tilde parameters.
We shall prove this conjecture by induction on the rank of U(N ) following an ar-
gument presented in [39]. The initial condition is the three-sphere partition function of
T (U(1)). This is simply an FI parameter t for a background U(1) vectormultiplet with
mass parameter m,
ZU(1) (m, t) = e2πimt . (3.25)

JHEP05(2015)095
It is immediate that this partition function obeys the difference equations pµ = τ and
pµe = τe as required. The partition function of T [U(N )] is now defined recursively by

2πitN
N
P
mj
Z
ZU(N ) (m, t) = e j=1
[ds] νN −1 (s) QN,N −1 (m, s) (3.26)
× ZU(N −1) (s, {t1 − tN , . . . , tN −1 − tN })
where we use shorthand notation m = {m1 , . . . , mN }, t = {t1 , . . . , tN } for mass parameters
of the U(N ) symmetries and s = {s1 , . . . , sN −1 } for the mass parameters of the U(N − 1)
symmetry. The contour is a small deformation away from the imaginary axis.
Let us now act with the difference operator Tr (x). This passes through the exponen-
tial prefactor leaving behind a phase τN r . It now acts on the bifundamental hypermulti-

plet contribution QN,N −1 (m, s) and can be exchanged for the sum of difference operators
Tr (−s) + Tr−1 (−s) according to the intertwining relation (3.20). Finally, we integrate by
parts inside the integral to find the sum of difference operators Tr (s) + Tr−1 (s) acting on
the partition function ZU(N −1) (s, {t1 − tN , . . . , tN −1 − tN }) which is an eigenfunction by
assumption. Combining these factors we find that ZU(N ) (m, t) is an eigenfunction of Tr (m)
with eigenvalue

r τ1 τN −1 τ1 τN −1
τN χr ,..., + χr−1 ,..., = χr (τ1 , . . . , τN ) . (3.27)
τN τN τN τN
This completes the recursive step.
For completeness, let us write down the solution of the recursion relation for the
partition function of T [U(N )] in closed form. For this purpose, it is convenient to introduce
(α)
the triangular array of mass parameters mj where α = 1, . . . , N and j = 1, . . . , α. For
(N )
convenience we denote mj = mj . They are mass parameters for the nodes of a triangular
quiver. We also introduce FI parameters tj with j = 1, . . . , N . The solution is then given
by the formula

ZU(N ) (m, t) = (3.28)

Z NY
−1 (α) (α)
e2πitN (m1 +···mN ) dνα (m(α) ) Qα+1,α m(α+1) , m(α) e2πi(tα −tα+1 )(m1 +···+mα )
α=1

where again the choice of the contour is explained later in section 3.5.

– 25 –
Finally, we note that this mathematical argument has a pleasing interpretation in
terms of interfaces. We can think of T [U(N )] as obtained by colliding a series of symmetry
breaking interfaces between a sequence N = 2∗ theories with gauge groups U(j), j =
1, . . . N − 1 together with boundary FI parameters. Our argument simply brings the ’t
Hooft loop through the sequence of interfaces one-by-one.
T [U(N )] is self-dual under mirror symmetry. A more precise statement is that the S 3
partition function obeys the symmetry property

Z(µ, τ, η) = Z(τ, µ, q 1/2 η −1 ) . (3.29)

JHEP05(2015)095
This symmetry is extremely non-trivial from the integral expression that we have presented
and we will not attempt to prove it directly. However, we note that it is equivalent to saying
that the partition function is simultaneously an eigenfunction of Hamiltonians acting on
the masses and FI parameters,

Tr (τ, pτ , η) · Z(µ, τ, η) = χr (µ) Z(µ, τ, η) ,

(3.30)
Tr µ, pµ , q 1/2 η −1 · Z(µ, τ, η) = χr (τ ) Z(µ, τ, η) .

The classical limit of trigonometric RS relations in the electric (2.17) and magnetic (2.18)
frames is reproduced from the above relations in the q → 1 limit.

3.5 Holomorphic blocks

The squashed S 3 partition function of T [U(N )] is simultaneously an eigenfunction of two

sets of difference operators Tr (m) and Ter (m) related by the transformation b ↔ b−1 . On
the other hand, asking for an eigenfunction of a single set of difference operators, say Tr (m),
leads to a basis of solutions called holomorphic blocks [12, 13].
Let us denote the holomorphic blocks by Bj where the index j = 1, . . . , N ! labels the
supersymmetric vacua of the theory on S 1 × R2 . i.e. the number of solutions of the Bethe
equations. The S 3 partition function can be reconstructed from the holomorphic blocks as
follows
N!
X
ZS 3 (m, t) = Bj (µ, τ ) Bj (e
µ, τe) . (3.31)
b
j=1

The holomorphic blocks can be computed a priori, up to some non-perturbative ambiguities,

by systematically building formal solutions of the relevant difference equations and finding
a basis of contours Cj for the integrals.
Here we follow a less sophisticated approach and derive the holomorphic blocks from
the Coulomb branch integral expression for the S 3 partition function by explicitly evalu-
ating the integral by residues in a certain regime of small FI parameters. This leads to
an alternative Higgs branch representation of the partition function involving the vortex
partition function. Here we will concentrate on T [U(N )] leaving the notationally more
complicated case of T [U(N )]ρ in appendix B.

– 26 –
3.5.1 T [U(2)] theory
Let us start with the simplest example, the partition function of T [U(2)]. The partition
function is
Z Y2 ε∗
[12 ]
ZU(2) = S(ε) dx S ± i(x − mj ) e2πx(t1 −t2 )+2πt2 (m1 +m2 ) . (3.32)
2
j=1

The integrand has two sets of simple poles at x = mj + iε∗ /2 + ikb + ie kb−1 and x =
mj − iε∗ /2 − ikb − ie kb−1 with integers k, e
k ≥ 0. The integral contour can be chosen to
enclose either set of poles, but the result is independent of this choice. We take residues

JHEP05(2015)095
from the first set of poles. The residue sum is
∞
[12 ]
2 X
X 2
Y
ZU(2) = S(ε) Resx=0 S ix − kb − e
kb−1 × S imij − kb − e
kb−1 (3.33)
i=1 k,e
k=0 j6=i
2
Y ε∗ e −1
× S −imij + ε∗ + kb + e
kb−1 e2π(mi +i 2 +ikb+ikb )(t1 −t2 )+2πt2 (m1 +m2 ) .
j=1

Using the identity (B.6), one can evaluate the residue of the double sine function
e e k+1)
e e k(k+1) k( e
kb−1 = i−k−k (−1)k+k+kk q 4 qe 4 (q; q)−1
Resx=0 S ix − kb − e q ; qe)e−1
k (e (3.34)
k
2 2
with q ≡ e2πib , qe ≡ e2πi/b . Plugging this residue and using the identities (B.6) we find
2
X Y
[1,1] ε∗
)(t1 −t2 )S(imij )
ZU(2) = e2πt2 (m1 +m2 )+2π(mi +i 2 (3.35)
S(imij + ε)
i=1 j6=i
   
X∞ 2
k Y (η 2 µ i
; q) X∞ ek Y2 (η 2 µ
ei
; q
e )
τ1 µj k τe1 µ
ej e
k
× qη −2 µi
× η −2
qe µ
e
.
τ2 (q µj ; q)k τe2 (e
q i
; q
e )e
k=0 j=1 e k=0 j=1 µ
ej k

The result takes the form of the Higgs branch representation which is given by sum over
the contributions from two supersymmetric vacua labelled by i = 1, 2. Also, the second
line shows the factorization into the vortex and anti-vortex partition functions.

3.5.2 T [U(N )] theory

Now we turn to the partition function of the T [U(N )] theory. We shall perform the integral
recursively starting with the recursive expression
Z PN −1
ZU(N ) (m, t) = [ds]νN −1 (s)QN,N −1 (m, s)e2π(tN −1 −tN ) i=1 si ZU(N −1) (s, {t1 , · · · , tN −1 })
(3.36) PN
(1)
with ZU(1) = e−2πit1 m .
This partition function differs by a prefactor from e2πtN i=1 mi

the true partition function of the T [U(N )] theory defined in the previous section. We shall
multiply this prefactor to the last expression after integration.
The contour integrand has infinite simple poles and zeros from various contributions
and also from the function ZU(N −1) . We should first ask how the integral contour goes

– 27 –
around those poles. We choose the contour along the real axis assuming that the mass
parameters m as well as the squashing parameter b are real parameters. We then take
residues from the poles above the real axis. One can also pick up the poles in the other
side but the result will be the same.
Although pole structure seems to be complicated as they can be developed by various
terms, it turns out that the relevant poles can arise only from the contributions of the
bifundamental hypermultiplets QN,N −1 . Indeed, we find that poles from ZU(N −1) are
completely cancelled by zeros of the vector multiplet contribution and poles from the vector
multiplet are also cancelled by zeros of ZU(N −1) and QN,N −1 . Namely νN −1 ZU(N −1) has
no relevant pole inside the contour. We will see this after computing ZU(N −1) explicitly.

JHEP05(2015)095
However, before we get ZU(N −1) , we first assume that this is the case and evaluate the
integral by picking up poles only from the hypermultiplet.
The contribution QN,N −1 has simple poles at si = mj + i ε2 + iki b + ie
ki b−1 for ki , e
∗
ki ≥ 0
in the contour. Summing over residues from these poles we obtain
1 X X τN −1 |k| τeN −1 |ek| 2πi(t −t )( (N −1)ε∗ −i PN −1 m )
ZU(N ) = e N −1 N 2 j=1 σ(j)
(N − 1)! τN τeN
σ∈WN k,e
k≥0
QN −1 −1 N
i,j=1 S ε − imσ(i)σ(j) + kij b + e
kij b−1 NY Y
× QN −1 ∗ e
S ε − imσ(i)σ(j) + ki b + ki b −1

i6=j S − imσ(i)σ(j) + kij b + e

kij b−1 i=1 j=1
N
Y −1 Y
N
× Resxi =0 S ixi + imσ(i)σ(j) − ki b − e
ki b−1 ZU(N −1) (s∗ , {t1 , · · · , tN −1 }) .
i=1 j=1
(N −1) PN −1 (N −1)
We denote k = {ki }
and |k| = i=1 ki and similarly for e
k. Here s∗ stands for
the integral variables si evaluated at each pole and they replace the mass parameters in the
partition function ZU(N −1) . In this expression, the (semi-)positive integer numbers ki and
e
ki correspond to the vortex and anti-vortex numbers of the U(N − 1) gauge group, which
will become clearer later when we write the full partition function as a factorized form.
There is the summation over σ in the Weyl group WN and an overall factor 1/(N − 1)!
corresponding the the Weyl factor of the U(N − 1) gauge group. Combining these two,
one notices that the number of supersymmetric vacua added to each recursion step is N .
Therefore the total number of supersymmetric vacua of the T [U(N )] partition function is
N ! which is in a perfect agreement with the quaternionic dimension of the Higgs branch.
We can use the above formula to compute the function ZU(N −1) at s∗ . Using the Weyl
group WN −1 , one can set σ (N −1) = 1 and obtain
X τN −2 |k| τeN −2 |ek| 2πi(t −t )( (N −2)ε∗ −i PN −2 m )
ZU(N−1) (s∗ ) = (N − 1) e N −2 N −1 2 j=1 j
τN −1 τeN −1
k,e
k≥0
QN −2 −2 N −1
i,j=1 S ε−imij +kij b+ e
kij b−1 NY Y (N−1) (N−1) −1

× QN −2 S ε∗ − imij +(ki −kj )b+(e
ki − e
kj )b
i6=j S −imij +kij b+ e
kij b−1 i=1 j=1
N
Y −2 NY
−1
(N−1) e e(N−1) −1
× Resxi =0 S ixi +imij −(ki −kj )b−(ki − kj )b ZU(N−2) (s∗ , {t1 ,· · ·, tN −2 }).
i=1 j=1

– 28 –
(N −2)
Here we have redefined the vortex numbers of the U(N − 2) gauge group as k = {ki +
(N −1) (N −2) (N −1)
ki } and e
k = {eki +e
ki } with i = 1, · · · , N −2. The second line and the residues
in the third line can be further simplified using the equations in (3.34) and (B.6) as follows:

PN −2 PN −1 (N −1) (N −1)
N−2
Y iS(imi,N −1 )
πi (ki −kj )b(b−ε∗ )+(e
ki −e
kj )b−1 (b−1 −ε∗ )
e i=1 j=1
S(ε + imi,N −1 )
i=1
µi
N−2 q ηe−2 µµeeji ; qe)ekij
Y (qη −2 µµji ; q)kij (e N−2 2 η 2 µµeeji ; qe)ek −ek(N−1)
Y (η µj ; q)ki−kj(N−1) (e
Y N−1 i j

i6=j ( µµji ; q)kij ( µµeeji ; qe)ekij i=1 j=1 q µµeeji ; qe)ek −ek(N−1)
(q µµji ; q)k −k(N−1) (e
i j i j

JHEP05(2015)095
We can consecutively apply this procedure and evaluate the partition functions ZU(n) (s∗ )
for lower n < N − 2 in the similar manner. Upon the redefinition of the vortex numbers
(n) P −1 (j) (n) P −1 e(j)
for U(n − 1) gauge group as k (n) = {ki = N j=n ki } and k ki = N
e(n) = {e
j=n ki }
one can see that the partition function takes the same form as ZU(N −1) but just the rank
is reduced to n − 1.
Finally we combine all integrated subpartition functions and find

X N
Y
PN PN S(imσ(i)σ(j) )
e−πiε N−
∗ (N t
ZU(N ) = i=1 ti )+2π i=1 ti mσ(i) ZV (mσ(i) )ZAV (mσ(i) )
S(ε + imσ(i)σ(j) )
σ∈WN i<j
−2 µi
2 µi

X N −1
Y q τn |k | Y qη µj ; q kij
(n) n (n) Y Y η µj ; q ki(n)−kj(n+1)
n n+1
ZV = µi
n=1
η 2 τn+1
i6=j µj ; q k(n) i=1 j=1
q µµji ; q k(n)−k(n+1)
{~k(1) ,··· ,~k(N −1) }≥0 ij i j

ZAV = ZV ((τ, µ, η, q) → (e
τ, µ
e, ηe, qe)) . (3.37)
N P
where |k (N ) | = |e
k (N ) | = 0. Here we have multiplied the prefactor e2πtN i=1 mi which we
omitted in (3.36). This is the Higgs branch representation of the S 3 partition function
of the T [U(N )] theory. We note that the partition function is factorized by the vortex
and anti-vortex partition functions, ZV and ZAV respectively, as well as the perturbative
contributions.
We still need to check the our assumption that the vector multiplet and the function
ZU(N −1) do not develop nontrivial poles in the contour integral. To see this, it is convenient
to first fix σ = 1 in ZU(N −1) using the Weyl group of the U(N − 1) gauge symmetry and
perform the contour integral starting with s1 and so on. We then noticed that the 1-loop
contribution in ZU(N −1) (s) of the form

N
Y −1
S(isij )
(3.38)
S(ε + isij )
i<j

is completely cancelled by the vector multiplet contribution νN −1 (s) of the U(N − 1). Thus
the function ZU(N −1) (s) can have poles only from the vortex and anti-vortex series. Fur-
∗
thermore, we find that possible poles from the vortex series are si = mj + i ε2 + inb + imb−1
with n, m > 0 and they are also cancelled by zeros of the vector multiplet contribution at
si = sj + inb + imb−1 for i > j. The remaining poles are at si = sj + iε + ipb + iqb−1

– 29 –
with p, q ≥ 0 and i > j in νN −1 (s), which are also cancelled by zeros of QN,N −1 (s) at
∗
si = mj − i ε2 + inb + imb−1 . Therefore the integral contour involves no pole from the
vector multiplet contribution and the function ZN −1 . This proves that the relevant poles
can come only from the contributions of the bifundamental hypermultiplets, and therefore
the Higgs branch representation of the partition function (3.37) is correctly derived.

3.5.3 Givental J-functions

Since the Higgs branch of 3d N = 2∗ quiver theories can be identified with the cotan-
gent bundle of the corresponding Nakajima quiver varieties, the vortex partition function

JHEP05(2015)095
thereof, which is the generating function of the BPS states on Higgs branch, gives the
Givental J-function of the corresponding variety [31, 40].
The corresponding quiver variety for T [U(2)] theory is T ∗ P1 . According to (3.35) the
corresponding vortex partition function of the T [U(2)] theory in one of the vacua (when,
say, i = 1) reads
k

∞ η2; q
X η 2 µµ21 ; q qη −2 ττ21
[1,1] k 2 µ1 µ1
ZV = k
· = 2 F1 2
η ,η ; q ; q; qη −2 z , (3.39)
(q µµ12 ; q)k (q; q)k µ2 µ2
k=0

where we used the definition of the Q-hypergeometric function (see appendix A.1).

3.6 Difference equations in electric frame

It is instructive to explicitly demonstrate that holomorphic blocks, which we have computed
above, satisfy difference relations in the electric frame (i.e. second relation in (3.30)). We
shall demonstrate this fact for 3d vortex partition functions (3.37) for T [U(2)] and T [U(3)]
theories.

3.6.1 T [U(2)] theory

We can explicitly check that (3.39) satisfies the following difference relation

ητ1 − η −1 τ2 1 µ2 ητ2 − η −1 τ1 2 [1,1] [1,1]
µ1 η pτ + pτ ZV = (µ1 + µ2 )ZV , (3.40)
τ1 − τ2 η τ2 − τ1

and
[1,1] [1,1]
p1τ p2τ ZV = ZV , (3.41)

where momenta operators act as piτ τj = q δij τj piτ . Note that the operator in the left hand
side of the above equation differs from T1 for trigonometric RS model (2.15) by prefactors
which depend on µ1 , µ2 and η. We can remove those factors by redefining

θ(η −1 τ1 , q)θ(η τ2 , q) [1,1]

ZV = Z , (3.42)
θ(µ1 τ1 , q)θ(µ2 τ2 , q) V

where θ(a, q) = (a, q)∞ (qa−1 , q)∞ is one solution of the difference equation pa θ(a, q) =
−a−1 θ(a, q). Our theta function conventions are given in appendix A.3. Then ZV obeys

– 30 –
the following set of eigenfunction difference equations

ητ1 − η −1 τ2 1 ητ2 − η −1 τ1 2
pτ + pτ ZV = (µ1 + µ2 ) ZV ,
τ1 − τ2 τ2 − τ1 (3.43)
p1τ p2τ Z V = µ1 µ2 Z V ,

which are nothing but conservation conditions for quantum two-body trigonometric RS
Hamiltonians. The eigenvalues on right hand sides are the expectation values of supersym-
metric Wilson loops in the fundamental and skew symmetric tensor representation of U(2).
Thus (3.43) provide the desired quantization of the twisted chiral ring of T [U(2)] theory.

JHEP05(2015)095
Had we chosen another vacuum of the T [U(2)] theory in (3.35) (i = 2) we would
have obtained the second vortex partition function which can be obtained from (3.39)
by replacing µ2 and µ1 . After the redefinition similar to (3.42) (again, with µ2 and µ1
interchanged) the partition function in the second vacuum also satisfies (3.43).
Note that the prefactor in (3.42) is not unique. For example, one may replace the factor
θ(η −1 τ
1 )/θ(µ1 τ1 ) by another one θ(τ1 )/θ(µ1 τ1 η) that obeys the same difference equation.
They correspond to factorizations of mixed Chern-Simons terms (or FI terms) for the 3d
∗
theory living on the surface defect, i.e. eπiε t1 +2πt1 m1 in this case. In the classical limit
b → 0 (or b → ∞), this prefactor reproduces the classical FI terms up to proper rescaling
of the parameters.
At this step let us make the following observation. Vortex partition functions which
were computed in (3.37) are infinite series in FI parameters. Note, however, that for some
values of the coefficients these series truncate and become polynomials. Remarkably, if
we combine them with the corresponding prefactors containing theta-functions, like (3.42),
we can reproduce the so-called Macdonald polynomials, which recently appeared in many
different contexts in mathematical physics. For example, if in (3.42) one puts q −k =
η 2 µ1 /µ2 for k = 0, 1, 2 the series become degree-k symmetric Macdonald polynomials in τ1
and τ2 with parameters q and η. In [41] it was shown that Macdonald polynomials with
proper normalization are eigenfunctions of difference operators of tRS type.

3.6.2 T [U(3)] theory

The vortex partition functions of T [U(3)] theory are slightly cumbersome, so we shall only
specify difference operators which they satisfy. There are 3! = 6 vacua in the T [U(3)] theory,
meaning that there are as many different vortex partition functions which are related to
each other by interchanging mass parameters µ1 , µ2 and µ3 between each other. Thus in
one of the vacua the partition function satisfies the following difference equation

ητ1 − η −1 τ2 ητ1 − η −1 τ3 1 ητ2 − η −1 τ1 ητ2 − η −1 τ3 2

µ1 η 2 pτ + µ 2 pτ
τ1 − τ2 τ1 − τ3 τ2 − τ1 τ2 − τ3
!
µ3 ητ3 − η −1 τ1 ητ3 − η −1 τ2 3 [1,1,1] [1,1,1]
+ 2 pτ Z V = (µ1 + µ2 + µ3 )ZV . (3.44)
η τ3 − τ1 τ3 − τ1

– 31 –
Again, in order to convert the above equation into the momentum conservation of the
three-particle trigonometric RS model we redefine the vortex partition function as

θ(η −2 τ1 , q)θ(τ2 , q)θ(η 2 τ3 , q) [1,1,1]

ZV = Z . (3.45)
θ(µ1 τ1 , q)θ(µ2 τ2 , q)θ(µ3 τ3 , q) V

The prefactors again correspond to factorization of the FI terms.

We have checked in the series expansion in the FI parameters that (3.44) with the
above redefinition is the true eigenfunction of the tRS model. We have also verified that
it is an eigenfunctions of the quadratic tRS operator T2 . Interestingly, one can reproduce

JHEP05(2015)095
the entire vortex partition function using the perturbation theory in FI parameters given
the difference relations of the form (3.44).

4 5d/3d partition functions

The three-dimensional T [U(N )] theory can be coupled to maximally supersymmetric U(N )

Yang-Mills theory in five dimensions to form a surface defect. Following references [14, 42],
if the surface defect is supported on the subspace S 1 × R2 ⊂ S 1 × R4 , we expect that
the twisted chiral ring of T [U(N )] is deformed by an additional parameter that we denote
by Q. This parameter is related to the dimensionless combination of the five-dimensional
gauge coupling g 2 and the radius R of S 1 by the formula
2
− 8π 2R
Q=e g . (4.1)

This deformed twisted chiral ring of the defect theory is expected to provide a representa-
tion of the Seiberg-Witten curve of the five-dimensional theory on S 1 × R4 . The 5d gauge
coupling can be considered as the scalar field in the background vector multiplet coupled
to the topological instanton charge. When the 5d gauge theory is compactified on a circle,
we can turn on a background holonomy in the vector multiplet and it complexifies the 5d
gauge coupling.
The Higgs branch U(N ) symmetry of T [U(N )] is gauged in coupling to five-dimensional
maximal SYM. In particular, the twisted masses µj of the three-dimensional theory are
identified with the vacuum expectation value of the five-dimensional real vectormultiplet
scalar (complexified by the holonomy around S 1 ). Furthermore, the three-dimensional
N = 2∗ deformation η can be identified with the five-dimensional N = 1∗ deformation. The
FI parameters τj are additional parameters associated to the remaining U(N ) symmetry
of the defect. We claim that the twisted chiral ring of T [U(N )] is deformed as follows

X Y θ1 (τi /η 2 τj ; Q) Y
piτ = χr (µ1 , . . . , µN , Q) , (4.2)
θ1 (τi /τj ; Q)
I⊂{1,...,N } i∈I i∈I
|I|=r j ∈I
/

where θ1 (x, q) is the first Jacobi theta function (see appendix A.3). The functions χr (µ) are
the expectation values of supersymmetric Wilson loops wrapping S 1 in the anti-symmetric

– 32 –
tensor representations of U(N ) of rank r = 1, . . . , N , including instanton corrections. It is
straightforward to see that

θ1 (τi /η 2 τj ; Q) η −1 τi − ητj
= + O(Q) ,
θ1 (τi /τj ; Q) τi − τj
X (4.3)
χr (µ1 , . . . , µN ; Q) = µi1 · · · µir + O(Q) ,
i1 <···<ir

and hence that the twisted chiral ring of T [U(N )] is reproduced in the limit Q → 0 where
we turn of the coupling to the five-dimensional degrees of freedom.

JHEP05(2015)095
The combinations appearing on the left are the Hamiltonians of a classical integrable
system: the complexified elliptic RS system. This reduces to the complex trigonometric
RS system in the limit Q → 0. The same integrable system describes the Seiberg-Witten
geometry of five-dimensional maximal SYM. In particular, the expectation values of su-
persymmetric Wilson loops χr (µ1 , . . . , µN , Q) can be taken to parametrize the Coulomb
branch of the five-dimensional theory on S 1 ×R4 , over which equation (4.2) defines a family
of elliptic curves. The Seiberg-Witten differential is identified with the symplectic potential
P
λ = j log pτj ∧ d log τj for the phase space of the elliptic RS system.
The classical integrable system can be quantized by studying this coupled 3d - 5d
system in a curved background preserving some supersymmetry. A natural extension of
our discussion in section 3 would be to study the partition function on S 5 with the surface
defect supported on an S 3 . This provides a formidable technical challenge. However, in
the absence of defects, there is much evidence to suggest that the partition function on
S 5 can be reconstructed from knowledge of the Nekrasov partition function in the Omega
background S 1 × R4ǫ1 ,ǫ2 . Recently, there is some evidence that the same conclusion can be
reached in the presence of a surface defect by enriching the set of building blocks to include
the Nekrasov partition function in the presence of a surface defect supported on S 1 × R2ǫ1 .
We focus solely on this case in what follows.
In principle, one should be able to perform an exact localization computation for
the coupled 5d/3d system. This computation is beyond the scope of the current paper. 8
Instead, we will use an alternative description of the surface defect as a monodromy defect
labelled by the partition [1N ], whose partition function can be computed using ramified
instanton counting [16, 17]. We emphasize that this equivalence has a conjectural status.
Therefore, we first check that the partition function in the presence of a defect holomorphic
blocks of T [U(N )] in the limit Q → 0,

Z (i) (m,
~ ~t, ǫ1 , ǫ2 , Q) = B (i) (m,
~ ~t, ǫ1 ) + O(Q) . (4.4)

The monodromy defect of type [1N ] has Gukov-Witten parameters τ1 , . . . , τN which are
identified with FI parameters of T [U(N )] in the decoupling limit. The leading term is
independent of the equivariant parameter ǫ2 in the plane orthogonal of the defect.
The complete partition function depends on two quantization parameters, ǫ1 and ǫ2 .
At finite Q we will consider the Nekrasov-Shatashvili limit ǫ2 → 0 of the normalized
8
Some work is this direction is now being done in [43].

– 33 –
expectation value of the defect
(i)
Zρ (~a, Q, ǫ1 , ǫ2 , ~t)
Dρ(i) (~a, Q, ǫ1 , ǫ2 , ~t) = lim , (4.5)
ǫ2 →0 Z(a, Q, ǫ1 , ǫ2 )

We will be able to check to a finite order in the instanton expansion in Q that (4.5) form a
basis of solutions to the quantum elliptic RS Hamiltonians. We should emphasize that these
are so far formal solutions: we have neither specified a measure nor checked normalizability
with respect to it. This point is discussed further below.
We shall momentarily discuss the construction of expectation values (4.5), but first

JHEP05(2015)095
we need to build up the necessary ingredients for instanton calculus in the presence of
monodromy defects.

4.1 Instanton counting

Below we review only essential ingredients for our computations further in this section. We
refer the reader to [44–46].
Let us consider the hyper Kähler quotient construction for the instanton moduli space
MN,k with gauge group U(N ) and instanton number k. We introduce vector spaces V = Ck
and W = CN and matrices A, B ∈ Hom(V, V ), P ∈ Hom(W, V ) and Q ∈ Hom(V, W ). An
element g ∈ U(k) acts on these matrices by

(A, B, P, Q) → (gAg −1 , gBg −1 , gP, Qg −1 ) . (4.6)

The hyperkahler moment maps are

µC = [A, B] + P Q ,
(4.7)
µR = [A, A† ] + [B, B † ] + P P † + QQ† ,

which are valued in Hom(V, V ). The instanton moduli space MN,k is then given by the
hyper Kähler quotient construction. As a complex manifold, the instanton moduli space
can be constructed as a Kähler quotient by discarding the real moment map in favor of a
stability condition and dividing by the complex group GL(k, C).
Consider the following action of (C∗ )2 × GL(N, C) on the matrices

(A, B, P, Q) → (t1 A, t2 B, P f −1 , t1 t2 f Q) , (4.8)

where t1 , t2 ∈ C∗ and f ∈ GL(N, C). Note that (C∗ )2 corresponds to rotations in the
two coordinate planes. This commutes with the action of GL(k, C) and preserves the
stability condition, therefore it descends to an action on the instanton moduli space MN,k .
The fixed points of MN,k under this action are labelled by an N -tuple of Young tableaux
~λ = (λ1 , . . . , λN ) where the total number of boxes is k.
Let us introduce the equivariant parameters ǫ1 , ǫ2 and ~a = (a1 , . . . , aN ) that are valued
in the Cartan subalgebra of the symmetry group action, i.e. such that t1 = eǫ1 and t2 = eǫ2
and f = diag(ea1 , . . . , eaN ). Then the equivariant character for the action of the global

– 34 –
symmetry (C∗ )2 × GL(N, C) on the vector spaces W and V at a fixed point ~λ is
N
X
(W )
χ~ = e aI
λ
I=1
N
(4.9)
(V )
X X
aI (1−i)ǫ1 +(1−j)ǫ2
χ~ = e e .
λ
I=1 (i,j)∈λI

The summation (i, j) ∈ λI refers to a sum over the boxes of Young tableaux λI . For
example, the Young tableaux has three boxes with the labels (1, 1), (1, 2) and (1, 3).

JHEP05(2015)095
Next we introduce equivariant Chern character
(E) (W ) (V )
χ~ = χ~ − (1 − e−ǫ1 )(1 − e−ǫ2 )χ~ . (4.10)
λ λ λ

(E)
Using χ~ we can now consider the equivariant character of the tangent space to the
λ
universal bundle over the instanton moduli space at a fixed point λ,
(E) (E ∗ )
χ~ χ~
χ~λN =1 = − λ λ
(1 − e−ǫ1 )(1 − e−ǫ2 )
(W ) (W ∗ )
χ~ χ~ (4.11)
λ λ
=−
(1 − e−ǫ1 )(1 − e−ǫ2 )
(V ) (V ∗ ) (W ) (V ∗ ) (V ) (W ∗ )
− (1 − eǫ1 )(1 − eǫ2 )χ~ χ~ + eǫ1 +ǫ2 χ~ χ~ + χ~ χ~ .
λ λ λ λ λ λ

and the operation ∗ corresponds to reversing the sign of all of the equivariant parameters.
The second line is the perturbative contributions to the Nekrasov partition function. The
third line is the equivariant character of the tangent space to the instanton moduli space,
that is ignoring the universal bundle. They provide the instanton contributions to the
Nekrasov partition function. This is the relevant character for pure N = 1 U(N ) gauge
theory in five dimensions.
Let us now consider the maximally supersymmetric N = 2 U(N ) gauge theory in five
dimensions. This corresponds to an N = 1 U(N ) vectormultiplet together with an adjoint
hypermultiplet. Due to the presence of the adjoint matter we must now consider a vector
bundle on the instanton moduli space. The relevant character is therefore

χ~N
λ
=2
= (1 − em−ǫ1 −ǫ2 ) χ~N
λ
=1
, (4.12)

where m is a mass parameter for the adjoint hypermultiplet. The equivariant character
can always be written in the following form
X
χ~N
λ
=2
= nα e w α , (4.13)
α

where we sum the exponents of the weights at all the fixed points. From this we compute
the five-dimensional instanton partition function
X ~ Y w −nα
α
Zinst = Q|λ| 2 sinh . (4.14)
α
2
λ

– 35 –
In what follows we use exponentiated parameters µj = eaj , eǫ1 = q and em = η −2 q. Since
we will mainly be interested in the limit ǫ2 → 0 we will not need to introduce a parameter
for ǫ2 . The coefficients in the expansion are then rational functions of these exponentiated
parameters.

4.2 Ramified instantons

Let us now consider ramified instanton counting, namely we want to compute instanton
partition function in the presence of a monodromy defect [3]. It was shown in [16, 17]
that the desired answer can be obtained by applying a simple orbifolding procedure to the
above computation.

JHEP05(2015)095
The possible monodromy defects are labelled by a partition ρ = [n1 , n2 , . . . , ns ] where
P
we choose to order n1 ≥ n2 ≥ . . . ≥ ns and si=1 ni = N . This determines the subgroup
L = U(n1 ) × U(n2 ) × · · · × U(ns ) of U(N ) gauge group which is left unbroken by the defect.
The gauge field has a singularity9 in the complex plane orthogonal to the defect which can
be described as follows
I
Aa = 2πma , a = 1, . . . , N , (4.15)
|z2 |=ǫ

where
ma = (m1 , · · · , m1 , m2 , · · · , m2 , · · · , ms · · · , ms ) . (4.16)
| {z } | {z } | {z }
n1 n2 ns

There is an additional label σ which determines how L is embedded into U(N ). Each σ
corresponds to permutation of the monodromy parameters m that are not simply permuta-
tions within each block, that is σ ∈ W/WL where WL is the Weyl group of L. The number
of such permutations is clearly Nρ = N !/(n1 ! . . . ns !).
To compute the ramified instanton partition function, we quotient the standard con-
struction of the instanton moduli space we have reviewed earlier in section 4.1 by a Zs -
action where s is the length of the partition ρ. The Zs - action is embedded inside the
(C∗ )2 × GL(N, C) symmetry of the instanton moduli space. The component in (C∗ )2 acts
on the complex coordinates by (z1 , z2 ) → (z1 , ωz2 ) where ω s = 1. The component in
GL(N, C) acts on the vector space W such that it decomposes

W = ⊕sj=1 Wj , nj = dimC Wj (4.17)

into eigenspaces of the Zs - action. Our convention is that the generator of Zs acts on the
vector space Wj by Wj → ω j Wj . In the sector with instanton number k, we must make
an additional choice of the decomposition of the other vector space
s
X
V = ⊕sj=1 Vj , dimC Vj = kj , kj = k . (4.18)
j=1

Each of these choices corresponds to a distinct topological sector and hence to a distinct
ramified instanton moduli space Mρ,k1 ,...,ks . In summary, the ramified instanton moduli
9
This is a so-called tame ramification, namely when z2 = 0 is a regular singular point, as opposed to a
wild ramification with higher degree singularities.

– 36 –
space Mρ,k1 ,...,ks can be obtained as a Zs quotient of the standard instanton moduli space
P P
MN,k with N = sj=1 nj and k = sj=1 kj .
Let us now explain how to compute the answer. The first statement is that each fixed
point ~λ of the standard instanton moduli space MN,k is also a fixed point of one and only
one ramified instanton moduli space Mρ,k1 ,...,ks . The hardest part of the computation is to
identify which sector {k1 , . . . , ks } a given fixed point ~λ contributes to. It is clear that the
P
total number of boxes in ~λ must add up to k = sj=1 kj . Introduce the following labels for
the Young tableaux
~λ = {λj,α } , j = 1, . . . , s , α = 1, . . . , ns . (4.19)

JHEP05(2015)095
Then the boxes in the i-th column of λj,α contribute to the instanton number ki+j−1 . If
i + j − 1 > N then we count modulo N . We can denote the sector associated to a fixed
point ~λ by kj (~λ).
For example, let us consider U(2) theory in the two-instanton sector k1 + k2 = k =
2, so we need to describe three spaces M2,1,1 , M2,2,0 , M2,0,2 . They are correspondingly
generated by the following tuples of Young tableaux

M2,1,1 : ,∅ , , , ∅, ,

M2,2,0 : ,∅ , M2,0,2 : ∅, . (4.20)

Let us now compute the equivariant character of the tangent space to the ramified
instanton moduli space at a fixed point T~λ Mρ,k1 ,...,ks . Let us denote this character by
χρ,~λ (~a, ǫ1 , ǫ2 ), whilst the character of unramified case is as above χ~λ (~a, ǫ1 , ǫ2 ). According to
the decomposition of space W we have introduced the following notation for the equivariant
parameters
~a = {aj,α } , j = 1, . . . , s , α = 1, . . . , ns . (4.21)
Then we compute the character of the ramified moduli space of instantons via averaging
over the orbifold action, therefore the answer is automatically invariant, which we can
always write using a summation over weights
s
1X ǫ2 + 2πir ǫ2 + 2πir
χρ,~λ (~a, ǫ1 , ǫ2 ) = χλ aj,α − j, ǫ1 ,
s s s
r=1 (4.22)
X
wα
= nα e .
α

Analogously to the unramified case we define

Y
wρ,~λ = (2 sinh(wα /2))−nα . (4.23)
α

In addition, for N = 1∗ theories there is the additional equivariant mass parameter m. We

claim that this mass parameter is invariant under the orbifold action. Finally, the ramified
instanton partition function is given by
X k (~λ) ~
Zρ = Q11 . . . Qsks (λ) wρ,~λ . (4.24)
~λ

– 37 –
Note that the product of instanton parameters for each sector is equal to the unramified
instanton parameter
Q1 · · · · · Qs = Q , (4.25)

which is also implied by the last equality in (4.18).

4.3 3d decoupling limit

4.3.1 U(2) theory
We shall consider N = 1∗ theory with gauge group U(2) in great detail. In this case, there is

JHEP05(2015)095
only one non-trivial monodromy defect labelled by the partition ρ = [1, 1], which is expected
to correspond to the surface defect obtained by coupling to T [U(2)]. The additional label
σ corresponds to permutations ± of the monodromy parameters, corresponding to the two
massive supersymmetric vacua of T [U(2)].
It is convenient to make the following replacements

Q
(+) Q1 = z , Q2 = , (4.26)
z
Q
(−) Q1 = , Q2 = z . (4.27)
z
in order to satisfy (4.25). We shall immediately see physical meaning of both choices. As
we have already mentioned, variable Q can be identified with the holomorphic scale the
five-dimensional bulk theory while z can be understood as a ratio of the FI parameters of
the 3d defect theory coupled to the bulk 5d theory z = ττ21 .
Let us compute the instanton contributions to the Nekrasov partition function (4.24)
and send Q → 0 in order to decouple the five-dimensional degrees of freedom. The first
few terms are

(+) q η 2 − 1 η 2 µ2 − µ1
Z[1,1] = 1 + 2 z
η (q − 1) (q µ2 − µ1 )
(4.28)
q 2 η 2 − 1 η 2 q − 1 η 2 µ2 − µ1 η 2 µ2 q − µ1 2 3
+ z + O(z ) ,
η 4 (q − 1)(q 2 − 1) (q µ2 − µ1 ) (q 2 µ2 − µ1 )

(−) q η 2 − 1 η 2 µ1 − µ2
Z[1,1] =1+ 2 z
η (q − 1) (q µ1 − µ2 )
(4.29)
q 2 η 2 − 1 η 2 q − 1 η 2 µ1 − µ2 η 2 µ1 q − µ2 2
+ z + O(z)3 ,
η 4 (q − 1)(q 2 − 1) (q µ1 − µ2 ) (q 2 µ1 − µ2 )
where we identify q = eǫ1 . Note that the dependence on the equivariant parameter ǫ2 in
the plane orthogonal to the defect has dropped out in this limit. The two expressions are
related by µ1 ↔ µ2 . It is straightforward to recognize them as q-hypergeometric series

(+) 2 2 µ2 µ2 −2
Z[1,1] = 2 F1 η , η , q , q, qη z ,
µ1 µ1
(4.30)
(−) 2 2 µ1 µ1 −2
Z[1,1] = 2 F1 η , η , q , q, qη z .
µ2 µ2

– 38 –
These are indeed the equivariant vortex partition function of U(1) theory with two flavors,
evaluated in each of the supersymmetric vacua. Alternatively they are the non-perturbative
(+)
contributions to the two independent holomorphic blocks. We can now see that Z[1,1]
exactly matches with the vortex partition function of the T [U(2)] theory in one of the
(−)
vacua (3.39), whereas Z[1,1] coincides with the the vortex partition function in the other
vacuum. Recall that both expressions are related merely by interchanging µ1 and µ2 .
After redefinition (3.42) the defect partition function satisfies trigonometric RS differ-
ence relations (3.43), where momenta act as pτj = eǫ1 τj ∂τj .

4.3.2 U(N ) theory

JHEP05(2015)095
When we consider the decoupling limit of the 5d U(N ) N = 1∗ theory with maximal
monodromy defect [1N ] we can eliminate each of Qi in (4.24) from in favor of Q and
[1N ]
other Qi . Thus there are N different partition functions Zi which can be reproduced in
the decoupling limit. Similarly to the T [U(2)] example, they can be mapped onto vortex
partition functions in (3.37).

4.4 Wilson loops

An essential ingredient of our construction is the computation of the vacuum expectation
values of supersymmetric Wilson loops wrapping S 1 . Using the language of equivariant
characters we can easily modify formula (4.14) in order to insert the fundamental Wilson
loop operator inside
P |~λ| (E) Q
wα −nα
~λ q χ~λ α 2 sinh 2
hW(1) i = P ~ Q
wα −nα
. (4.31)
|λ|
~λ q α 2 sinh 2

(E)
The additional character χ~ (4.10) in the numerator represents the contributions from a
λ
heavy charged BPS particle propagating around S 1 . It turns out that the numerator has
the same universal divergence as the denominator in the limit ǫ2 → 0. Thus the expectation
value is finite and we can denote it by

E(1) = lim hW(1) i . (4.32)

ǫ2 → 0

Remember that this is the Wilson loop expectation value of U(N ) gauge group, which
differs from that of the SU(N ) gauge group by the overall U(1) contribution. The overall
U(1) contribution can be interpreted as a contribution from a heavy free BPS particle and
one can compute it using the Wilson loop expectation value of unit charge in the abelian
gauge theory [47]. The U(1) contribution is given by

U(1) (Q/η 2 , Q)∞ (η 2 Q/q, Q)∞

hW(1) i = . (4.33)
(Q, Q)∞ (Q/q, Q)∞
The fundamental Wilson loop expectation value of SU(N ) gauge group is then given by
U(N )
SU(N )
hW(1) i
hW(1) i = U(1)
(4.34)
hW(1) i

– 39 –
Q
with a condition i µi = 1. For later convenience we define
U(1)
N(1) ≡ lim hW(1) i−1 , (4.35)
ǫ2 →0

which will be used in the difference equations below.

Let us consider some examples. For U(2) N = 1∗ theory with a fundamental Wilson
loop around S 1 we find the following expectation value
" #
µ µ η 2 +q η 4 +η 2 +q −(µ +µ )2 η 2 q
U(2) 1 2 1 2
E(1) = (µ1 + µ2 ) 1−(1−η 2 )(q−η 2 ) Q + O(Q2 ) .
η 4 q (µ1 q − µ2 ) (µ2 q − µ1 )

JHEP05(2015)095
(4.36)
Similarly, the Wilson loop expectation value in the fundamental representation of U(3) is
given by

U(3) (1−η 2 )(q−η 2 )(µ1 −η 2 µ2 )(µ1 −η 2 µ3 )(η 2 µ1 −qµ2 )(η 2 µ1 −qµ3 ) 2
E(1) = µ1 1− Q + O(Q )
η 6 q(µ1 − µ2 )(µ1 − µ3 )(µ1 − qµ2 )(µ1 − qµ3 )
+ (µ1 , µ2 , µ3 cyclic permutations) . (4.37)

In order to insert a Wilson loop in the r-th skew symmetry power of the fundamental
representation of U(N ) we must modify the computation. It will be discussed elsewhere.

4.5 5d/3d coupled system

4.5.1 U(2) theory
Turning on the parameter Q, the partition function in the presence of the defect is no longer
finite in the limit ǫ2 → 0. The divergence exponentiates and the anomalous dimension is
universal, that is, it is the same divergence without the defect. The saddle point equations
arising from the anomalous dimension will fix the mass parameters µ1 and µ2 to be some
discrete solutions. For now we will ignore this issue and concentrate on the normalized
expectation value of the surface defect with unconstrained µ1 and µ2 .
As the divergence is universal, the normalized expectation value of the monodromy
defect is well defined in the limit ǫ2 → 0. We introduce the notation
(±)
(±)
Z[1,1]
D[1,1] = lim . (4.38)
ǫ2 →0 Z
The first few terms of the expansion are

(+) η 2 − 1 q η 2 µ2 − µ1 η 2 − 1 q η 2 µ1 − µ2 Q
D[1,1] = 1 + 2 z+ + ··· ,
η (q − 1) (µ2 q − µ1 ) η 2 (q − 1) (µ1 q − µ2 ) z
(4.39)
(−) η 2 − 1 q η 2 µ1 − µ2 η 2 − 1 q η 2 µ2 − µ1 Q
D[1,1] = 1 + 2 z+ + ··· .
η (q − 1) (µ1 q − µ2 ) η 2 (q − 1) (µ2 q − µ1 ) z
One may check that the above expressions reduce to the vortex partition functions in the
limit Q → 0. It is very important that we have a regular expansion in Q1 and Q2 and hence
at higher orders in the Q expansion there are negative powers of z = τ2 /τ1 . It appears that
D(+) and D(−) are related by interchanging the mass and FI parameters (µ1 , τ1 ) ↔ (µ2 , τ2 ).

– 40 –
As a regular expansion in Q1 and Q2 we have checked up to order O(Qn1 1 Qn2 2 ) with
n1 + n2 = 5 that the above normalized expectation values obey the following difference
equations
!
µ1 θ1 τ1 /η 2 τ2 , Q 1 θ1 τ2 /η 2 τ1 , Q 2 (+)
p + ηµ2 p D(+) = N(1) E(1) D[1,1] ,
η θ1 (τ1 /τ2 , Q) τ θ1 (τ2 /τ1 , Q) τ
! (4.40)
µ2 θ1 τ1 /η 2 τ2 , Q 1 θ1 τ2 /η 2 τ1 , Q 2 (−)
p + ηµ1 p D(−) = N(1) E(1) D[1,1] ,
η θ1 (τ1 /τ2 , Q) τ θ1 (τ2 /τ1 , Q) τ

JHEP05(2015)095
where E(1) is the normalized expectation value of a Wilson loop in the fundamental rep-
resentation of U(2) in the limit ǫ2 → 0 which we have computed earlier in (4.36) and
N(1) is the inverse of the overall U(1) contribution defined in (4.35). Therefore if we im-
Q
pose the traceless condition i µi = 1, the eigenvalues of the difference equations are the
fundamental Wilson loop expectation value of SU(2) gauge group.
As before, we can define new partition functions

θ(η τ1 , q)θ(η −1 τ2 , q) (+)

D(+) = D ,
θ(µ1 τ1 , q)θ(µ2 τ2 , q) [1,1]
(4.41)
(−) θ(η −1 τ1 , q)θ(η τ2 , q) (−)
D = D ,
θ(µ2 τ1 , q)θ(µ1 τ2 , q) [1,1]

which obey the same difference equations

!
θ1 τ1 /η 2 τ2 , Q 1 θ1 τ2 /η 2 τ1 , Q 2
p + p D(±) = N(1) E(1) D(±) ,
θ1 (τ1 /τ2 , Q) τ θ1 (τ2 /τ1 , Q) τ (4.42)
p1τ p2τ D(±) = µ1 µ2 D(±) .

The same non-uniqueness caveats as before holds also here. One should also check that
the perturbative contributions obtained from orbifolding factorize nicely into contributions
from the defect and the bulk. At this point we claim to have found formal eigenfunctions
and eigenvalues of the two-body elliptic (complexified) Ruijsenaars-Schneider integrable
system, at least as a series expansion.

4.5.2 U(3) theory

The generalization to many-body system is straightforward — one needs to compute

ramified instanton partition function of U(N ) N = 1∗ theory in the presence of full
ρ = [1, . . . , 1] := [1N ] monodromy defect. In this section we will present the results for U(3)
theory. There are 3! different embeddings of the Levi subgroup L = U(1)3 into U(3) labelled
by a permutation σ. We shall focus on a particular embedding σ = 1 in what follows.
We are interested in the normalized partition function in the presence of a monodromy
defect in the limit ǫ2 → 0. As discussed in the previous sections, this limit is well defined

– 41 –
and we end up with a finite expression. The partition function is expanded as

Z[13 ] q(1 − η 2 ) (µ1 − η 2 µ2 ) (1 − η 2 )(η 2 µ1 − µ3 )(η 2 µ2 − qµ3 )
lim =1+ 2 z1 + z2 z3
ǫ2 →0 Z η (1 − q) (µ1 − qµ2 ) η 2 (1 − q)(qµ1 − µ3 )(µ2 − µ3 )
(η 2 µ1 −µ2 )(η 2 µ2 −qµ3 )(µ2 −η 2 µ3 ) (1−qη 2 )(µ1 −η 2 µ2 )(µ1 −qη 2 µ2 ) 2
+ 2 z2 z3 + 2 z
η (qµ1 − µ2 )(µ2 − µ3 )(µ2 − qµ3 ) η (1 − q 2 )(µ1 − qµ2 )(µ1 − q 2 µ2 ) 1

+ (µ1 , µ2 , µ3 and z1 , z2 , z3 cyclic permutations) + · · · , (4.43)

where z1 = ττ12 , z2 = ττ32 , z3 = ττ31 Q. We have also assumed that |z1 |, |z2 |, |z3 | < 1 in the

JHEP05(2015)095
expansion. In the decoupling limit Q → 0, this partition function precisely reproduces
the vortex partition function of the T [U(3)] theory in (3.37) in one of the vacua, after a
relabelling (µi , τi ) → (µ−1 −1
i , τi ).
Let us define a new function as

θ(η 2 τ1 , q)θ(τ2 , q)θ(η −2 τ3 , q) Z[13 ]

D[13 ] = lim . (4.44)
θ(µ1 τ1 , q)θ(µ2 τ2 , q)θ(µ3 τ3 , q) ǫ2 →0 Z

It follows that this function satisfies three difference equations:

θ1 (τ2 /η 2 τ1 , Q) θ1 (τ3 /η 2 τ1 , Q) 1 θ1 (τ1 /η 2 τ2 , Q) θ1 (τ3 /η 2 τ2 , Q) 2
pτ + p (4.45)
θ1 (τ2 /τ1 , Q) θ1 (τ3 /τ1 , Q) θ1 (τ1 /τ2 , Q) θ1 (τ3 /τ2 , Q) τ

θ1 (τ1 /η 2 τ3 , Q) θ1 (τ2 /η 2 τ3 , Q) 3
+ p D[13 ] = N(1) E(1) D[13 ] ,
θ1 (τ1 /τ3 , Q) θ1 (τ2 /τ3 , Q) τ

θ1 (τ1 /η 2 τ2 , Q) θ1 (τ1 /η 2 τ3 , Q) 2 3 θ1 (τ2 /η 2 τ3 , Q) θ1 (τ2 /η 2 τ1 , Q) 3 1
pτ pτ + p p
θ1 (τ1 /τ2 , Q) θ1 (τ1 /τ3 , Q) θ1 (τ2 /τ3 , Q) θ1 (τ2 /τ1 , Q) τ τ

θ1 (τ3 /η 2 τ1 , Q) θ1 (τ3 /η 2 τ2 , Q) 1 2
+ pτ pτ D[13 ] = N(1) E(1,1) D[13 ]
θ1 (τ3 /τ1 , Q) θ1 (τ3 /τ2 , Q)

and p1τ p2τ p3τ D[13 ] = µ1 µ2 µ3 D[13 ] . Here N(1) is the inverse of the overall U(1) Wilson loop
factor and E(1) is the expectation value of a Wilson loop in the fundamental representation
of U(3). E(1,1) is the Wilson loop expectation value in the rank-two antisymmetric tensor
representation. We can compute E(1,1) using the fundamental Wilson loop expectation
value by a simple replacement of the gauge fugacities such as

E(1,1) (µ1 , µ2 , µ3 ) = E(1) (µ1 ,µ2 ,µ3 )→(µ2 µ3 ,µ3 µ1 ,µ1 µ2 )

, (4.46)

which holds only for U(3). The difference equations have been checked in a series expansion
in zi ’s up to O(z1n1 z2n2 z3n3 ) with n1 + n2 + n3 = 5.
The difference equations are the eigenvalue equations of the three-particle eRS inte-
grable system. Therefore we found that the monodromy defect partition functions are
eigenfunctions of the integrable Hamiltonians and the Wilson loops in rank 1 and 2 anti-
symmetric representations are their eigenvalues. In the decoupling limit Q → 0 one can
also notice that these equations reduce to the equations for three-body tRS Hamiltoni-
ans in (3.44).

– 42 –
4.5.3 Comments on normalizability of wavefunctions
The complexification plays a crucial role here. Indeed, normally one may expect that elliptic
system has a discrete spectrum, whereas our eigenfunctions are parametrized by parameters
variables µ1 and µ2 . The point is that we have found only formal eigenfunctions in that
they may not be normalizable wavefunctions. In solving the additional Bethe equations
arising from the saddle point analysis of the effective twisted superpotential W of the
defect theory coupled to the 5d theory in the divergence log Z → ǫ12 W + O(1) we will fix
the parameters µ1 and µ2 to discrete set of values. It is possibly the case that for these
values we obtain normalizable wavefunctions and hence obtain a discrete spectrum.

JHEP05(2015)095
The quantum elliptic Ruijsenaars-Schneider model in connection with supersymmet-
ric gauge theories has been discussed in the the literature before. For instance, in [48],
some approximate solutions to the elliptic RS eigenvalue problem are labeled by a discrete
parameter. We therefore expect that extremization of eigenfunctions (4.38) with respect
to the mass parameters µi (putting the solution on shell) will resolve the normalizability
issues. We plan to discuss it elsewhere.

4.5.4 Connection to 4d index of class S

The elliptic RS system also arises in the study of the superconformal index of 4d theories
of class S [48]. The hamiltonians act on the superconformal index by introducing a surface
defect and the eigenfunctions are the wavefunctions associated to a so-called ‘maximal’
puncture in the TQFT that computes the superconformal index. In fact the wavefunctions
appropriate for the superconformal index are eigenfunctions of a pair of elliptic RS sytems
and are labelled by a dominant integral weight λ.
In reference [47] two of the author’s have conjectured that the eigenfunctions appro-
priate for the 4d superconformal index are equivalent to the superconformal index of the
(2, 0) theory on Sβ1 × S 5 in the presence of a codimension 2 defect of ‘maximal’ type wrap-
ping Sβ1 × S 3 and a codimension 4 defect labelled by the dominant weight λ wrapping
Sβ1 × S 1 where the S 3 and the S 1 are Hopf-linked inside S 5 . The 6d superconformal index
is computed by the partition function of 5d SU(N ) maximal SYM on S 5 together with a
monodromy defect on S 3 and a Wilson loop in the representation λ on S 1 . The parameters
of the 4d and 6d superconformal index are identified according to equation (5.2) of refer-
ence [47]. This conjecture was checked explicitly in the ‘Schur’ limit where only a single
parameter of the superconformal index is turned on.
Now, in the localization computation of the S 5 partition function, the integrand fac-
torizes into contributions from 3 fixed points in the base of the fibration S 5 → CP2 . In the
presence of a monodromy defect wrapping S 3 , the contributions from 2 of the fixed points
are two copies of the Nekrasov partition function on S 1 × R4 with a monodromy defect
wrapping S 1 × R2 . The identification of parameters given in equation (5.2) of [47] realizes
the NS-limit for these Nekrasov partition functions in the planes transverse to the defect.
Thus we arrive at the partition functions studied in this paper.
The results of this paper imply that the S 5 partition function with a monodromy defect
on S 3 is an eigenfunction of a pair of RS systems, one associated to each fixed point on CP2

– 43 –
intersected by the S 3 . The additional parameter λ is incorporated by adding a Wilson loop
at the remaining fixed point. This provides some further evidence of the conjecture made
in [47]. However, it should be emphasized that evaluating the Coulomb branch integral of
the S 5 partition function and checking the result takes the form of a superconformal index
remains a formidable challenge.

4.6 5d theory coupled to 3d hypermultiplets

We now consider a different type of co-dimension two defect in five dimensions. We shall
couple 3d hypermultiplets to the 5d maximal supersymmetric SU(N ) gauge theory. The 3d

JHEP05(2015)095
hypermultiplets have U(1) × SU(N ) flavor symmetry and we couple this SU(N ) symmetry
to the bulk gauge symmetry. Our motivation is that the U(2) gauge theory with this type
of defect turns out to be related to the U(2) gauge theory with a monodromy defect of
type ρ = [12 ] by bispectral duality.
In this section, however, we carry out the computation using U(N ) and we set
QN
i=1 µi = 1 to getQ the results for SU(N ). Naively U(N ) and SU(N ) are the same un-
der the constraint N i=1 µi = 1 as the diagonal U(1) part decouples in the field theory
limit. However, this is no longer true in the instanton computation. The subtle issues
related to the U(1) part in the 5d partition functions are discussed in [49–51]. Here we
assume that the U(1) part does not affect the gauge theory dynamics and its contribution
to the partition function can be subtracted by hand.
In order to compute the partition function in the presence of this 3d defect we shall
follow the prescription given in [19]. The defect introduces an additional three dimensional
vector bundle on the instanton moduli space of the 5d gauge theory. The 3d fields of the
defect theory are in the fundamental and anti-fundamental representations of the SU(2)
bulk gauge group and thus the corresponding bundles are the universal bundle and its
conjugation, respectively. The equivariant characters of the 3d hypermultiplets can be
computed using the equivariant character χ(E) of the universal bundle at k instantons.
We get

m−ǫ1 (E)
"P k
#
−x 2 X
e 2 χk m−ǫ1
−x e aI −ǫ
χ3d
chiral = =e 2 I=1
− (1 − e ) 2
e φ i
,
(1 − e−ǫ1 ) (1 − e−ǫ1 )
i=1
m−ǫ1 (E ∗ )
"P k
#
2 X
e 2 +x χk m−ǫ1 e −aI
χ3d
anti = = e 2 +x I=1
− (1 − e−ǫ2 )eǫ1 +ǫ2 e−φi , (4.47)
(1 − e−ǫ1 ) (1 − e−ǫ1 )
i=1

where χchiral and χanti are the equivariant characters of the 3d chiral and anti-chiral multi-
plets respectively, and x is the equivariant parameter for the U(1) flavor symmetry. These
characters are expressed in term of the equivariant parameter eφi for the auxiliary gauge
group U(k) of the k instanton moduli space. Due to the 3d contribution the saddle point
value of φi is not fully classified by the N -tuple of Young tableaux. We note that the path
integral on the instanton moduli space has extra saddle points from the 3d factors other
than the previous saddle points labeled by Young tableaux.

– 44 –
We find from the above character formulae the extra contribution of the 3d fields to
the k instanton partition function in the integral expression. It is given by

k
Y (1 − η −1 τ −1 ρi )(1 − pqη −1 τ ρ−1
i )
Zk3d = , (4.48)
i=1
p(1 − p−1 η −1 τ −1 ρi )(1 − qη −1 τ ρ−1
i )

where η 2 = e−m+ǫ1 , ρi = eφi , τ = ex , p = eǫ2 .

There are subtleties in the perturbative contribution related to the boundary condition

JHEP05(2015)095
on ∂(S 1 × R2ǫ1 ) ∼
= T 2 due to the presence of the Omega background. The superpotential
and the Chern-Simons term in the 3d theory are in general not invariant under the super-
symmetry in the presence of a boundary [52, 53]. We will discuss the latter in the next
subsection when we gauge the 3d flavor symmetry.
Let us focus on the standard N = 4 superpotential. For being supersymmetric with
boundary we should impose the relevant boundary conditions for the fields. For the ad-
joint chiral multiplet in the N = 4 vector multiplet, we impose the Dirichlet boundary
condition. The boundary conditions for the hypermultiplets, we impose the Neumann
boundary condition on the chiral multiplets and the Dirichlet boundary condition on the
anti-chiral multiplets, or vice versa. This choice guarantees the supersymmetry invariance
of the superpotential in the presence of the boundary.
The partition function of 3d theories on solid torus S 1 × D2 with specified boundary
conditions was recently computed in [54] using localization. We will compute the 1-loop
determinant for the 3d hypermultiplets using character formula (4.47) and the appropriate
boundary conditions. The terms independent of the instanton number k give the per-
turbative contributions. See appendix C for detailed 1-loop computations. The 1-loop
determinant of the 3d hypermultiplets is given by

2
3d − ǫm (πi−x)
Y (qη −1 τ µ−1
I ; q)∞
Z1−loop =e 1
−1 . (4.49)
I=1
(ητ µ I ; q) ∞

The prefactor comes from regularization. Note that this 1-loop contribution includes the
mx
prefactor e ǫ1 which amounts to the mixed Chern-Simons term between the background
U(1) flavor gauge field and R-symmetry current.
The full partition function of the N = 1∗ theory in the presence of the 3d hypermul-
tiplets is given by

∞
X I Y
k
dρi 5d 3d
Z 3d/5d = Z1−loop
3d 5d
Z1−loop Qk Zk , Zk = Z Z . (4.50)
2πiρi k k
k=0 i=1

5d
Z1−loop and Zk5d are the 1-loop and k instanton partition function of the original 5d theory

– 45 –
without the 3d fields,
2
Y 1/2
5d (p; p, q)∞ (q; p, q)∞ (µI /µJ ; p, q)∞ (pqµI /µJ ; p, q)∞
Z1−loop ∼
(pη −2 ; p, q)∞ (qη −2 ; p, q)∞ (pη −2 µI /µJ ; p, q)∞ (qη −2 µI /µJ ; p, q)∞
I6=J

Y k
2 Y k
(1 − p−1 η −2 ρi /µI )(1 − qη −2 µI /ρi ) Y
Zk5d = (1 − ρi /ρj ) ,
(1 − ρi /µI )(1 − pqµI /ρi )
I=1 i=1 i6=j
k
Y (1 − pqρi /ρj ) (1 − η −2 ρi /ρj )(1 − qp−1 η −2 ρi /ρj )
× . (4.51)
(1 − pρi /ρj )(1 − qρi /ρj ) (1 − p−1 η −2 ρi /ρj )(1 − qη −2 ρi /ρj )
i=j

JHEP05(2015)095
One can obtain the instanton part using the quantum mechanics on the instanton
moduli space. For the N = 1∗ theory with U(N ) gauge group, the Witten index of the
N = (4, 4) ADHM gauged quantum mechanics gives the instanton partition function, which
was computed in [55] using localization. When we couple the 3d fields, the supersymmetry
reduces to N = (2, 2) and one can deduce from (4.48) that there would be extra degrees
of freedom, a chiral and an anti-chiral multiplets, added to the quantum mechanics.
The contour integral (4.50) over ρi can be evaluated using the Jeffrey-Kirwan (JK)
residue prescription introduced in [56, 57]. The JK prescription has also been applied
to the localization of the index in quantum mechanics, which we will briefly review now.
See [51, 58, 59] for more detailed explanations.
We can first define a hyperplane in the φ plane for each charge vector Qi ∈ Rk of the
multiplets where the integrand in Zk (4.50) becomes singular

Hi = {φ ∈ Ck Qi (φ) + z = 0} . (4.52)

Here z denotes other chemical potentials (or log of fugacities). Let us then consider when
n ≥ k hyperplanes meet at a point φ = φ∗ and denote by Q(φ∗ ) ≡ {Qi φ ∈ Hi } a set of n
charge vectors at the point. The residue at the singular point can be computed using the
JK prescription.
The integrand of Zk is Laurent expanded around the singular point φ∗ in negative
powers of Qi (φ − φ∗ ). Among others, the JK residue yields nonzero result only at simple
poles of the form
1
, (4.53)
Qi1 (φ − φ∗ ) · · · Qik (φ − φ∗ )
where Qi1 , · · · , Qik are in Q(φ∗ ).
We will choose a reference vector η arbitrarily in Rk . This vector η should not be
confused with N = 1∗ mass parameter. The JK residue defined in [56] is:
(
dQi1 (φ) ∧ · · · ∧ dQik (φ) |det(Qi1 · · · Qik )|−1 if η ∈ Cone(Qi1 , · · · , Qik )
JK-Res(Q∗ , η) =
Qi1 (φ) · · · Qik (φ) 0 otherwise
(4.54)
where ’Cone’ denotes the cone formed by the k independent charge vectors. The result
turns out to be independent of the choice of η. The JK prescription can be applied when

– 46 –
the projective condition is satisfied [56]. This turns out to be the case with our Zk . The
partition function now takes the form of

1 X
Zk = JK-Res(Q∗ , η) Zk5d (φ, z)Zk3d (φ, z) (4.55)
|W |
φ∗

where W is the Weyl group of the U(k) gauge group.

For the 5d U(N ) gauge theory without coupling the 3d fields, the JK prescription
reproduces the Young tableaux sum rule of the instanton partition function. To check this,
we need to choose the reference vector like η = (1, 1, · · · , 1). However, after coupling the

JHEP05(2015)095
3d fields, the residue prescription implies that there would be nontrivial contributions at
the extra poles developed by the 3d factors as well as the Young tableaux summation.
For instance, at the single instanton sector, we should pick up all the poles from the
1
fields of positive charge which give the factors of the form Qi φ1 with Qi > 0. From the
sinh 2
formula for Zk=1 given above one can see that the poles at {ρ1 = µ1 , ρ1 = µ2 , ρ1 = pητ }
are inside the contour. Note that the first two poles are from the 5d factors which were
labeled by the N -tuple of Young tableaux, whereas the last pole is the new pole arising
from the 3d factors. The JK residue at the last pole is nontrivial and should be involved
in the Zk=1 computation.
At two instanton sector, we have nontrivial poles due to the 3d factors at

(ρ1 = pητ, ρ2 = µ1 ), (ρ1 = pητ, ρ2 = µ2 ), (ρ1 = pητ, ρ2 = q −1 ρ1 ), (ρ1 = pητ, ρ2 = pη 2 ρ1 )

(4.56)
apart from the poles in the Young tableaux classification. Summing over all JK residues at
the these poles, one can compute the two instanton partition functions Zk=2 in the presence
of the 3d matter fields. We expect that the Jeffrey-Kirwan method also works for higher
instanton sectors Zk>2 .
Let us now define the normalized partition function of the 3d-5d coupled system and
take the limit ǫ2 → 0
− mx Z 3d/5d
D3d/5d = e ǫ1 lim . (4.57)
ǫ2 →0 Z
− mx
Here we turned on the classical mixed Chern-Simons term e ǫ1 to cancel the induced
Chern-Simons term. This normalized partition function obeys the difference equation

ητ θ(q 2 η −4 pτ , Q) + η −1 τ −1 θ(pτ , Q) D3d/5d = θ(qη −2 pτ , Q)N(1) E(1) D3d/5d , (4.58)

where, as before, we denoted by pτ the difference operator satisfying pτ τ = qτ pτ so that

pτ becomes the conjugate momentum of τ . The factor N(1) is given in (4.35) and E(1) is
the Wilson loop expectation value in (4.36). We have checked this relation by expanding
both sides up to two instanton order.
One can also recast the above difference equation as an eigenfunction equation

θ1 (q 2 η −4 pτ , Q) 1 θ1 (p−1 τ , Q)
τ + D3d/5d = N(1) E(1) D3d/5d , (4.59)
θ1 (q 2 η −2 pτ , Q) τ θ1 (η 2 p−1
τ , Q)

– 47 –
To verify this equation we should expand both sides first in Q and later in η1 . We have
checked for some lowest orders in these expansions. Therefore we have computed the
spectrum of the two-body dual elliptic Ruijsenaars-Schneider Hamiltonian (l.h.s. of (4.59)),
whose eigenvalue is the expectation value of the Wilson loop (4.36) and the eigenfunction,
up to a normalization, is the partition function of the 5d N = 1∗ theory coupled to two
free 3d hypermultiplets (4.57).

4.7 S-transformation
We will now show that S-transformation which is an extension of the S-duality in three
dimensions can relate two different types of defects in five dimensions: a monodromy

JHEP05(2015)095
defect and a defect of 3d hypermultiplets. More precisely, we will act the S operation on
the partition function Z 3d/5d of (4.50) and show that the result agrees with the partition
function of the U(2) gauge theory in the presence of a monodromy defect.
Let us first review the S transformation in three dimension. There is a natural SL(2, Z)
action on 3d CFTs with U(1) symmetry [20]. This action maps a theory to other theo-
ries which could be inequivalent to the original theory. It is sufficient to understand two
generators T and S in the SL(2, Z) action. The T transformation simply shifts by one
unit the Chern-Simons level of the background gauge field A for the U(1) symmetry. The
S transformation is more complicated. Firstly we gauge the U(1) global symmetry and
make the background gauge field A as a dynamical gauge field. The theory then has a new
global symmetry whose conserved current is the magnetic flux of A. Secondly we introduce
a mixed Chern-Simons term AdB at unit level with background gauge field B for the new
global symmetry.
When the theory is on the manifold with boundary, the S transformation should be
carefully taken. If there exists the effective Chern-Simons term of the background gauge
field A, the naive gauging procedure fails to work due to the gauge anomaly. The Chern-
Simons term makes the gauge invariance anomalous on the boundary. To keep the gauge
invariance one need to couple relevant boundary degrees of freedom such that the flavor
anomaly of the 2d theory compensates the gauge anomaly on the boundary. In addition,
the mixed Chern-Simons term introduced by the S transformation also breaks the gauge
invariance. The 2d theory should be chosen to cancel this anomaly as well.
In the partition function, the gauge invariance is associated to the periodicity of the
holonomy parameter, a ∼ a + 2πi. The Chern-Simons coupling induces the terms violating
this periodicity. To cancel these gauge non-invariant terms, we should multiply the elliptic
genus of the proper 2d theory. We will now see this with our example.
Let us first consider the S transformation on the 3d partition function of the defect
after decoupling the 5d theory. The 3d theory consists of free hypermultiplets of the
SU(2) doublet. The S transformation acts on the U(1) flavor symmetry with parameter
x. We turn off all the classical background Chern-Simons terms. However, the 1-loop
effect generates the nontrivial effective mixed Chern-Simons term between the U(1) flavor
mx
symmetry and the U(1) R-symmetry. In the 3d partition function (4.49), the prefactor e ǫ1
induced from the 1-loop determinant amounts to the dynamically generated Chern-Simons
coupling. This prefactor violates the periodicity of the parameter x. Moreover, the mixed

– 48 –
− yx
Chern-Simons term AdB in the S transformation adds a term like e ǫ1 where y is the
holonomy parameter for the new U(1) flavor symmetry, which also violates the periodicity
of x. These gauge non-invariant terms must be canceled. In the previous section, we cancel
the former anomalous factor by turning on the classical Chern-Simons term corresponding
− mx
to e ǫ1 . In this section we couple instead boundary N = (0, 2) multiplets and cancel the
anomalous boundary terms.
The N = (0, 2) theory on the boundary T 2 consists of two fermi and one chiral multi-
plets. We refer the reader to [52, 60] for details of elliptic genera of the N = (0, 2) theories.
The regularized elliptic genus is summarized in appendix C. We introduce the boundary

JHEP05(2015)095
theory whose the elliptic genus is given by

2d m+x−y ǫ1 x ǫ1 m−y ǫ1
Z = exp −2πiζ2 0,− |1, +2πiζ2 0,− |1, +2πiζ2 0,− |1,
2πi 2πi 2πi 2πi 2πi 2πi
−1 −1
θ(τ ; q)θ(q η u; q)2
· (4.60)
θ(q −1 η 2 uτ −1 ; q)

with u = ey . One can check that the prefactor precisely cancels the gauge non-invariant
terms of the 3d theory on the boundary.
We now gauge the U(1) symmetry parametrized by τ . Then the S transformation on
the 3d partition function can be written as [13]
Z
dτ yx
ZS3d (u) = e ǫ1 Z 2d (u, τ ) Z1−loop
3d
(τ )
τ
Z 2
dτ −F θ(τ −1 ; q)θ(q −1 η 2 u; q) Y (qη −1 τ µ−1
I ; q)∞
= e −1 2 −1 −1 , (4.61)
τ θ(q η uτ ; q) (ητ µI ; q)∞
I=1

where the prefactor is F = πim − πiǫ1 /2 + π 2 /3 − ǫ2 /12 /ǫ1 and τ = ex .
The integral can be evaluated using the residue theorem with an appropriate pole
prescription. We propose that the integral contour encloses the poles at τ = q −k η −1 µI , k ≥
0 which come from the 1-loop determinant of the 3d fundamental chiral multiplet. It may
be possible to justify this pole prescription by the JK-like residue prescription, but we will
not attempt to do it. Let us focus on the residues at poles of τ = q −k η −1 µ1 . Applying the
residue theorem, we obtain

3d,(+) (qη −2 ; q)∞ (qη −2 µ1 /µ2 ; q)∞ 2 2 µ2 µ2 −2
ZS (u) = Z0 × × 2 F1 η , η , q ; q, qη u , (4.62)
(q; q)∞ (µ1 /µ2 ; q)∞ µ1 µ1

where
θ(ηµ−1 −1 2
1 ; q)θ(q η u; q)
Z0 = −e−F . (4.63)
θ(q −1 η 3 uµ−1
1 ; q)
Apart from Z0 , this S transformed partition function reproduces the holomorphic block
of the 3d U(1) gauge theory with 2 flavors in one of the supersymmetric vacua. The q-
(+)
hypergeometric series 2 F1 is precisely the vortex partition function Z[12 ] of the monodromy
defect ρ = [12 ] in (4.30) upon the identification u = τ2 /τ1 . This result shows that the
3d theories on the defects of two different types are related by the S transformation. The

– 49 –
residues at the poles τ = q k η −1 µ2 yields the same result with µ1 and µ2 exchanged which
corresponds to the holomorphic block in the second vacuum.
The S transformation can be promoted to five dimensions. It acts on the U(1) flavor
symmetry of the 3d theory on the defect as we did above in the decoupling limit. The S
transformation on the surface defect partition function is defined as in [19]
Z
3d/5d dτ yx
ZS (u, Q) = e ǫ1 Z 2d (u, τ ) Z 3d/5d (τ, Q) . (4.64)
τ

The contour is chosen to enclose the poles at τ = q −k η −1 µI , k ∈ Z. We note that the

instanton contribution also has poles at τ = q k η −1 µI for negative integer k < 0 and the

JHEP05(2015)095
contributions from the residues at these poles are crucial to compare with the partition
function of the dual theory.
We sum over all residues of the poles at τ = q −k η −1 µ1 and find

3d/5d,(+) (qη −2 ; q)∞ (qη −2 µ1 /µ2 ; q)∞ inst,(+)

ZS (u, Q) = Z0 × × ZS (u, Q) . (4.65)
(q; q)∞ (µ1 /µ2 ; q)∞

The instanton partition function after S transformation is expanded in u and Q/u and the
first few terms are given by

inst,(+) (η 2 − 1)q(η 2 µ2 − µ1 ) (η 2 − 1)q(η 2 µ1 p − µ2 )

ZS (u, Q) = 1 + u + Q/u + · · · (4.66)
η 2 (q − 1)(µ2 q − µ1 ) η 2 (q − 1)(µ1 pq − µ2 )

Remarkably, having identified the parameter as u = τ2 /τ1 , this instanton partition function
(+)
reproduces the ramified instanton partition function Z[12 ] in section 4

inst,(+) τ2 e Z (+)
ZS (u = , Q) = N [12 ]
(4.67)
τ1
up to the factor
Q (η 2 − 1)(η 4 p − q)
e = PE
N , (4.68)
1 − Q (q − 1)(η 2 p − 1)
which is independent of the mass and FI parameters.10 We have confirmed this relation up
to u2 (Q/u)2 order. This result provides a strong evidence that the two types of defects, the
monodromy defect and the defect of 3d hypermultiplets, are dual to each other under the
S transformation: i.e. bispectral dual. One can also obtain the second ramified partition
(−)
function Z[12 ] from the sum over residues at the second set of poles τ = q −k η −1 µ2 .
The S transformation exchanges the mass parameter and the corresponding momentum
as follows:
τ → p−1
u , pτ → q −1 η 2 u . (4.69)
Therefore the normalized partition function defined as
3d/5d,(+)
(+) ZS
DS = lim (4.70)
ǫ2 →0 Z
10
f (xn ) .
P∞
PE denotes the Plethystic exponential defined as PE[f (x)] = exp n=1

– 50 –
1 2 N-1 N

Figure 4. A Lagrangian description of the U(1) × · · · × U(N − 1) theory with fundamental matter
and N chiral multiplets at the final node.

obeys the difference equation

θ1 (η −2 u, Q) −1 θ1 (η −2 u−1 , Q) (+) (+)
p + pu DS = N(1) E(1) DS , (4.71)
θ1 (u, Q) u θ1 (u−1 , Q)

JHEP05(2015)095
which is in perfect agreement with the eigenfunction equation of the 2-body elliptic Rui-
jsenaars system in (4.42).

5 Gauge theories with chiral matter and 4d reduction

In this section we discuss chiral limits of our 5d/3d construction. Gauge theories in three
dimensions which we have considered so far included both chiral and antichiral matter fields
whose masses were split via the twisted mass of the axial U(1) subgroup of the N = 4 R-
symmetry. Since the supersymmetry is already broken down to N = 2 by this axial mass,
nothing prevents us from examining the possibility of more extreme mass splitting when
some of the mass parameters become large. This is the goal of the current section where we
shall study chiral limits of three-dimensional theories and their 5d/3d completions along
the lines of the previous sections.

5.1 Complete flags and open Toda chains

The chiral version of the quiver theory from figure 2 is formulated using complete N-
dimensional complex flag variety FN : 0 ⊂ C ⊂ C2 ⊂ · · · ⊂ CN . We start with the
connection of equivariant quantum K-theory of complete flags and twisted chiral rings of
the corresponding 3d gauge theories along the lines of section 2.4. In physics literature
complete N-flags arise as target spaces of supersymmetric sigma models with chiral matter
(see figure 4).
In order to see how the vacua equations for chiral quivers arise form vacua equations
of T [U(N )] theories let us rewrite equation (2.4) as follows

Nj−1 (j) (j−1) Y Nj (j) (j) Nj+1 (j) (j+1) Y Mj (j) (j)
τj Y ηεσn −σn′ η −1 σn −ησn′ Y ησn −εσn′ ηεσn −µa
· · · = (−1)δj ,
τj+1 ′ ησ (j−1) −εσ
(j)
n n′ 6=n η −1 σ
(j)
−ησ
(j)
n ηεσ
(j+1)
−σ
(j)
n a=1 ηµ
(j)
a −εσ
(j)
n
n =1 n′ n′ n′ =1 n′
(5.1)
where we scaled
(j) (j)
σi → εj σ i , µ(j)
a →ε
j−1 (j)
µa , (5.2)

and then take the limit

ε → 0, η → ∞, (5.3)

– 51 –
such that εη = 1 to get the following
Nj−1 Nj Nj+1 Mj
τj Y σn(j) − σ (j−1) Y (j)
σn′ Y ησn
(j) Y
n′
(j−1)
· (j)
· (j+1) (j)
· σn(j) − µ(j)
a = (−1)δj , (5.4)
τj+1 ησn′ σn σn′ − σn
n′ =1 n′ 6=n n′ =1 a=1

For flag manifold we have Nj = j, so, after additional scaling

τj → ǫ2j τj , (5.5)

we get

JHEP05(2015)095
j+1 (j) (j−1) j j−1 Mj
τj Y σn − σn′ Y (j)
Y 1 Y
(j−1)
· σn ′ · (j+1) (j)
· σn(j) − µ(j)
a = (−1)δj . (5.6)
τj+1 σn′ σn ′ − σn
n′ =1 n′ 6=n n′ =1 a=1

We have just described how to obtain a chiral theory which lives on a complete flag from
the theory supported on T ∗ FN . In order to derive other chiral theories corresponding to
non complete flags one needs start from theories of type T [U(N )]ρ (see appendix B, where
we complete the corresponding vortex partition functions) and take a limit similar to (5.3).

5.1.1 Chiral limit of T [U(2)]

The simplest example of a complete flag variety is P1 . Let us see how to describe its
quantum cohomology starting from T ∗ P1 . From [6] we get
τ1 ησ − µ1 ησ − µ2 τ1 τ2
= 1, p1τ = σ , p2τ = , (5.7)
τ2 ηµ1 − σ ηµ2 − σ σ
The set of the above three equations can be rewritten as
ητ1 − η −1 τ2 1 ητ2 − η −1 τ1 2
pτ + pτ = µ 1 + µ 2 , p1τ p2τ = τ1 τ2 . (5.8)
τ1 − τ2 τ2 − τ1
Performing scaling limits in (5.2) and (5.5) we get for (5.7)
τ1 −2 ηεσ − µ1 ηεσ − µ2
ε = 1. (5.9)
τ2 ηµ1 − εσ ηµ2 − εσ
or
τ2
= e2πRΛ = ℓ ,
(σ − µ1 )(σ − µ2 ) = µ1 µ2 (5.10)
τ1
where Λ is the dynamically generated scale. By expanding the quadratic expression in σ
in the l.h.s. of the above expression and using the definition of the momenta we arrive at
µ1 µ2 − ℓ 1
pτ + p2τ = µ1 + µ2 , (5.11)
τ1 τ2
which can be identified with the trace of two-body open Toda Lax matrix. From (5.7) we
can derive
p2 (µ1 µ2 − ℓ) − p(µ1 + µ2 )τ1 τ2 + τ12 τ22 = 0 , (5.12)
where we put p1τ = p. One can recognize in the above equation equivariant quantum K-ring
relation for P1 (cf. (2.56) for N = 2).

– 52 –
5.1.2 Effective twisted superpotential
Equivalently we can derive the above equations from the first principles by considering
N = 2 supersymmetric gauge theories in three dimensions built around a triangular quiver.
(i)
Let us fix and integer N ≥ 2. Similarly to (2.9) we introduce real mass parameters sj with
(i) (N ) (j) (j)
i, j = 1, . . . , N such that sj = 0 if i < j and sj = mj . The parameters {s1 , . . . , sj }
are gauge parameters for the j-th node and {m1 , . . . , mN } are mass parameters for U(N )
flavor symmetry. We also introduce FI parameters {t1 , . . . , tN }.
The effective twisted superpotential for theory in figure 4 reads
N X
X I
I−1 X

JHEP05(2015)095
(I−1) (I)
W= ℓ si − sj
I=2 i=1 j=1
N
X −1 X
I
(I) (I)
+ ℓ si − sj
I=1 i6=j
N j (5.13)
X iδj X (j)
+ tj − tj+1 + si
2R
j=1 i=1
 ! 2

N −1 I N
1 X  X (I) X (I+1) (I) 
+ si − si sj .
2
I=1 i=1 i,j=1

Each line of the above formula corresponds to chiral multiplets, vectormultiplets, FI terms
and Chern-Simons terms respectively, which become manifest in this form of the twisted
superpotential.11
The vacuum equations can be readily derived from (5.13) and read as
j
Y j+1
Y
τj (j) (j) (j) (j+1)
Qj+1 (σi ) σk = (−1)δj Qj−1 (σi ) σk , i = 1...,j , (5.14)
τj+1
k=1 k=1
k6=i

j
Q N
Q
(j)
where Qj (u) = (u − σi ). In what follows we denote QN (u) = M (u) = (u − µi ).
i=1 i=1
These equations describe (5.4) for the complete N -flag. As in the T [U(N )] case they may
appear to be slightly complicated, but it is favorable to express them the in functional
(1)
form similarly to (2.20). Using the recursive nature of conjugate momenta p1τ = σ1 ,
(2) (2) (1)
p2τ = σ1 σ2 /σ1 , and so on it is easy to see that they are recovered from the following
functional equations
τj+1 e j (u) ,
Qj+1 (u) − (−1)δj Qj−1 (u) σ pj+1
τ = Qj (u) Q (5.15)
τj

where Qe j (u) is monic monomial of degree one. As before, these equations serve as a
powerful toll for deriving the Lax matrix of Toda systems from the vacuum equations.
11
Recall that each function ℓ(s) contains a Chen-Simons term s2 /4 which is generated in one loop, and
it is the same for chiral and anti-chiral fields. After we take the chiral limit from (2.9) and anti-chiral fields
become infinitely massive, quadratic contributions remain in the last line of (5.13).

– 53 –
Again, let us consider N = 2 for simplicity. There is a single equation
τ2 e
M (u) − (−1)δ1 σ p2τ = Q1 (u) Q(u) . (5.16)
τ1

By evaluating the equation at u = 0 it is straightforward to see that Q1 (u) = u − p1τ and

e 1 (u) = u − p2τ and therefore
Q
τ2
M (u) = (u − p1τ )(u − p2τ ) + (−1)δ1 u p2τ
τ1
(5.17)
2 1 2 δ1 τ 2
= u − u pτ + pτ − (−1) p + p1τ p2τ .
2
τ1 τ

JHEP05(2015)095
Thus we have recovered the Q-Toda Hamiltonians.
Now we can graduate on to N = 3 case. We have two equations
τ2 e 1 (u)
Q2 (u) − (−1)δ1 u p2τ = Q1 (σ) Q
τ1
τ3 (5.18)
M (u) − (−1)δ2 Q1 (u) u p3τ = Q2 (u) Q e 2 (u) .
τ2

We can recycle the information from the N = 2 case so that Q1 (u) = u − p1τ and Q e 1 (u) =
u−p2τ . Furthermore, by evaluating the second equation at u = 0 we find that Q e 2 (u) = u−p3 .
τ
Recall that Q2 (u) is the matter polynomial of the N = 2 example. Thus we have

δ2 1 τ3 3 3 1 2 δ1 τ2 2
M (u) = (−1) u(u − pτ ) pτ + (u − pτ ) (u − pτ )(u − pτ ) + (−1) u pτ
τ2 τ1

τ2 τ3
= u3 − u2 p1τ + p2τ + p3τ − (−1)δ1 p2τ − (−1)δ2 p3τ (5.19)
τ1 τ2

1 2 2 3 3 1 δ1 τ 2 2 3 δ2 τ 3 3 1
+ u pτ pτ + pτ pτ + pτ pτ − (−1) p p − (−1) p p − p1τ p2τ p3τ ,
τ1 τ τ τ2 τ τ

which are Hamiltonians of the three-body relativistic Toda system.

5.1.3 Open Toda Lax matrix

Now we systematize the above computation in the general case N > 2 to find the Lax
matrix of the open relativistic Toda chain. We should find the matrix L(u) with non-zero
components
τi+1
Li+1,i = 1 , Li,i = u − pi , Li,i+1 = −(−1)δi u pτ , (5.20)
τi i+1
and then M (u) = det(L(u)) analogously to (2.17). Thus we have constructed the limit
when trigonometric Ruijsenaars-Schneider model is reduced to relativistic Toda lattice
using gauge theories.

5.2 Pure super Yang Mills theories with defects and closed Toda chains
Thus far we have identified parameter space of supersymmetric vacua of quiver theories with
chiral matter with global symmetry of rank N and the phase space of the (complexified)

– 54 –
open Toda chain with N particles. In other words, we have described the ‘chiral’ version of
the XXZ/tRS duality from section 2. Now, along the lines of section 4, we shall investigate
how the above construction is modified when the same theory is coupled as a surface defect
in pure U(N ) N = 1 SYM in five dimensions. We expect a deformation of the twisted chiral
ring depending on the dynamical scale of the five dimensional theory. In what follows, we
use a dimensionless quantity Λ obtained by multiplying by the dynamical scale by the
radius of the circle.
We therefore claim that the twisted chiral ring of quiver theory in figure 4 is described
by the functional equations
τj+1

JHEP05(2015)095
Qj+1 (u) − (−1)δj e j (u)
Qj−1 (u) u pτj+1 = Qj (u) Q (5.21)
τj
for j = 1, . . . , N − 2 and with Q0 (u) = 1, whereas at the final node there is a modification
of the equation to
τN τ1
M (u) − (−1)δN −1 QN −2 (u) u pτN − (−1)δN QN −1 (u)u p1τ
τN −1 τN (5.22)
e N −1 (u) .
= QN −1 (u) Q
For example, for N = 3 we would find

3 2 1 2 3 δ1 τ 2 2 δ2 τ 3 3 δ3 τ 1 1
M (u) = u − u pτ + pτ + pτ − (−1) p − (−1) p − (−1) p
τ1 τ τ2 τ τ3 τ

1 2 2 3 3 1 δ1 τ 2 2 3 δ2 τ 3 3 1 δ3 τ 1 1 2
+ u pτ pτ + pτ pτ + pτ pτ − (−1) p p − (−1) p p − (−1) p p
τ1 τ τ τ2 τ τ τ3 τ τ
− p1τ p2τ p3τ , (5.23)
which are nothing but closed Toda spectral relations. Thus we claim that coupling the
theory to five dimensions transforms open relativistic Toda chain to closed relativistic Toda
chain — a natural generalization of the similar statement found in [14] for nonrelativistic
Toda systems.
We now discuss the computation of ramified instanton partition functions of N = 1
SYM theories in five dimensions. For simplicity we start with gauge group U(2). In order
to compute the corresponding instanton partition function we use character for N = 1
theory (4.11). Then we introduce the monodromy defect according to the prescription of
section 4.2. For U(2) theory the instanton partition function will have the form (4.24)
labelled by the partition ρ = [1, 1]. We make the replacement
1 1 Q
(+) Q1 = √ z, Q2 = √ , (5.24)
µ1 µ2 µ1 µ2 z
1 Q 1
(−) Q1 = √ , Q2 = √ z. (5.25)
µ1 µ2 z µ1 µ2
√
Again, later we shall replace z = ττ12 . The normalization factors of µ1 µ2 are included to
remove fractional powers of µ1 and µ2 in the formulae for the Nekrasov partition function
in the presence of the defect, which we shall address later in this section. The same
normalization will be used for the difference equations. Note that they would cancel if we
move to the center of mass frame i.e. consider gauge group SU(2) instead of U(2).

– 55 –
5.2.1 Open Toda from decoupling limit
Analogously to section 4.3 we discuss the decoupling limit first, so we send Q → 0 and
decouple the five-dimensional degrees of freedom.12 In this limit we expect to find con-
tributions that can be accounted for by degrees of freedom supported on the defect. The
answer should be consistent with chiral limits (5.3), (5.5) of holomorphic blocks and vortex
partition functions which we have computed earlier in the paper. Indeed, this turns out to
be the case.
For example, in order to find the vortex partition function for P1 sigma model we can
start with Givental J-function (3.39) for T ∗ P1 and perform the chiral limit in order to

JHEP05(2015)095
see that X ∞
µ1 (qz)k
ZV → 2 F1 0, 0; q ; q; z = Qk , (5.26)
µ2 l l
l=1 (µ2 − µ1 q ) (1 − q )
k=0
which leads us to the formula for equivariant K-theoretic J-function for P1 from [31]. At
the last step we rescaled z → z/µ2 . The above result can be understood as hypergeometric
series of type 0 F1 .
Equivalently, from the computation of the ramified instanton partition functions in the
Q → 0 limit using notations (5.25) we get
∞
X
(+) q n(n+1)/2
Z[1,1] = Qn j )(µ − q j µ )
zn ,
j=1 (1 − q 1 2
n=1
∞ (5.27)
(−)
X q n(n+1)/2
Z[1,1] = Qn j )(µ − q j µ )
zn ,
j=1 (1 − q 2 1
n=1
which are related by interchanging µ1 ↔ µ2 . The above solutions represent the non-
perturbative contributions to the two independent holomorphic blocks of U(1) theory with
two chiral multiplets.
It is straightforward to prove that expansions (5.27) obey the following difference
equations

1 2 τ2 (+) (+) (+) (+)
µ 1 pτ + µ 2 pτ − Z[1,1] = (µ1 + µ2 ) Z[1,1] , p1τ p2τ Z[1,1] = Z[1,1] ,
τ1
(5.28)
1 2 τ2 (−) (+) 1 2 (−) (−)
µ 2 pτ + µ 1 pτ − Z[1,1] = (µ1 + µ2 ) Z[1,1] , pτ pτ Z[1,1] = Z[1,1] .
τ1
The second equations in each line impose that the answer depends only on the ratio τ2 /τ1 .
In order to remove the dependence on µ1 and µ2 on the left we must include contributions
to the classical action from FI parameters. These are exactly the same ambiguities we have
addressed in (3.42). We could choose, for example
θ(τ1 , q)θ(µ1 , q) θ(τ2 , q)θ(µ2 , q) (+)
Z (+) = Z[1,1]
θ(τ1 µ1 , q) θ(τ2 µ2 , q)
(5.29)
θ(τ1 , q)θ(µ2 , q) θ(τ2 , q)θ(µ1 , q) (−)
Z (−) = Z[1,1] ,
θ(τ1 µ2 , q) θ(τ2 µ1 , q)
12
One can also study ‘chiral’ version of section 3, namely study 3d partition functions of chiral quivers
and prove that they satisfy Q-Toda difference equations. This approach (albeit without reference to gauge
theories in three dimensions) was pursued in [61].

– 56 –
where, we recall, θ(a, q) obeys pa θ(a, q) = −a−1 θ(a, q). Now both of the functions Z (±)
obey the same difference equations

1 2 τ2
p τ + pτ − Z (±) = (µ1 + µ1 )Z (±) , p1τ p2τ Z (±) = µ1 µ2 Z (±) , (5.30)
τ1
which are the Hamiltonians of the two-body open Q-Toda integrable system. The holo-
morphic blocks in this example were discussed in detail in [13].
In [62] the K-theoretic J-function of the flag variety was shown to be the (universal)
eigenfunction of the relativistic Toda system for any simply laced group and in [63] the
analysis was extended to non-simply laced groups.

JHEP05(2015)095
5.2.2 Closed Toda chain
Turning on the parameter Q, the partition function the presence of the defect diverges in
the limit ǫ2 → 0 since we have degrees of freedom propagating in this plane. However,
the divergence exponentiates and is universal i.e. it is independent of the presence of the
defect. Thus the expectation value of the defect is finite in the limit
(±)
Zρ=[12 ]
(±)
R = lim . (5.31)
ǫ2 →0 Z
The first few terms in the expansion are
q q Q
R(+) = 1 + z+ + ···
(1 − q) (µ1 − µ2 q) (1 − q) (µ2 − µ1 q) z
(5.32)
q q Q
R(−) =1+ z+ + ···
(q − 1) (µ2 − µ1 q) (1 − q) (µ1 − µ2 q) z
We can see that, unlike the elliptic RS eigenfunctions, these two functions are no longer
related by µ1 ↔ µ2 at higher order in the Q expansion. This fact is not surprising since one
needs to scale Q as we obtain the above expressions via taking the limit from the elliptic
RS eigenfunctions. Yet, the expressions are symmetric if one interchanges z = τ1 /τ2 with
Qτ2 /τ1 . Starting from (4.39) and taking the limit described after (5.1) we arrive to the
above expressions provided that Q is also rescaled as follows
τ1 τ1
→ ǫ2 , Q → Qǫ4 , ηǫ = 1 , (5.33)
τ2 τ2
as ǫ → 0. In addition to the above scaling one can redefine the FI parameters as
τ1 1 τ1 Q
→ , Q→ , (5.34)
τ2 µ 2 τ2 µ1 µ2
then the elliptic RS eigenfunctions take the form of (5.32).
We have checked to order O(Qn1 1 Qn2 2 ) with n1 + n2 = 5 that (5.32) obey the difference
equations

1 2 τ2 τ 1
µ 1 pτ + µ 2 pτ − − Q R(+) = E(1) R(+) , p1τ p2τ R(+) = R(+) ,
τ1 τ2
(5.35)
1 2 τ2 τ1 (−) (−) 1 2 (−) (−)
µ 2 p τ + µ 1 pτ − − Q R = E(1) R , p τ pτ R =R ,
τ1 τ2

– 57 –
where
µ1 µ2 qQ
E(1) = (µ1 + µ2 ) 1 + + O(Q2 ) (5.36)
(µ1 q − µ2 ) (µ1 − µ2 q)
is the vacuum expectation value of the Wilson loop. We now multiply R(±) by the same
factors as in (5.29). From a three-dimensional perspective they were contributions to
classical action from FI parameters. The final result is that both functions obey

1 2 τ2 τ1
pτ + p τ − − Q R(±) = E(1) R(±) , p1τ p2τ R(±) = µ1 µ2 R(±) . (5.37)
τ1 τ2

Thus we can see how turning on the parameter Q in the difference equation controls the

JHEP05(2015)095
five-dimensional instanton corrections. We have found formal eigenfunctions of the two-
body closed Q-Toda Hamiltonians. Note that the spectrum should be discrete, whereas
we have continuous parameters µ1 and µ2 . Again, as in the non chiral case, they should
be fixed by Bethe equations coming from the divergent prefactor. It could be that this is
related to normalizability of the wavefunctions.

5.3 Connections to Kapustin-Willett results

Kapustin and Willett [64] computed quantum K-ring of a family of line bundles of Grass-
manninas T (M, N, k), which naturally appear in the study of twisted chiral rings of 3d
N = 2 U(N ) theories with M fundamental twisted chirals and Chern-Simons level k. In
other words, the authors computed the quantum K-ring for M copies of the tautological
bundle O(−1)M → Gr(M − k, N ) over the Grassmannian.13
Let us write Bethe equations for 3d N = 2∗ U(N ) theory with M flavors
M N
τ2 Y ησi − µa Y η −1 σi − ησj
= 1. (5.38)
τ1 ηµa − σi η −1 σj − ησi
a=1 j6=i

As we know these are quantum ring relations for T ∗ (Gr(M, N )). For simplicity let us
consider N = 1 and M = 3 first. The variables in the above equation then can be rescaled
as follows
η → ǫ−1 η , σ → ǫσ , µ3 → ǫ2 µ1 µ2 → ǫµ2 , (5.39)
to get
η − µ1 σ −1 τ2
qσ 2 = 1, q= , (5.40)
ηµ1 σ −1 − 1 τ1 µ 1 µ 2
which is the chiral ring relation for T (3, 2, 1).

5.4 4d/2d construction

For completeness we exhibit 2d analogues of some properties of 3d theories we have used in
the main text. Quantum cohomology of complete flag varieties and their cotangent bundles
in connection with monodromy defects were discussed in [65]. Here we discuss chiral limit
of the twisted chiral ring of (2, 2)∗ theory in two dimensions.
13
We changed k to −k for convenience.

– 58 –
Supersymmetric vacua equations can be obtained from (2.4) and read as follows

Nj−1 (j) (j−1) Nj (j) (j) Nj+1 (j) (j+1) Mj (j) (j)
τj Y sn −sn′ + 2ǫ Y sn −sn′ −ǫ Y sn −sn′ + 2ǫ Y sn −ma + 2ǫ
· · · = (−1)δj ,
τj+1 ′ s(j−1) −s
(j) ǫ
n + s
(j)
−s
(j)
n −ǫ s
(j+1)
−s
(j)
n + ǫ
m
(j)
a − s
(j) ǫ
n +
n =1 n′ 2 n′ 6=n n′ n′ =1 n′ 2 a=1 2
(5.41)
(j) (j) ǫ
Now we make some shifts sn → sn − j 2

Nj−1 (j) (j−1) Nj (j) (j) Mj (j) Nj+1

(j)
τj Y sn −sn′ Y Y sn(j) −s(j+1)
sn −sn′ −ǫ +ǫ Y sn −ma +(1−j) 2ǫ
n′
(j−1) (j)
· (j) (j)
· (j+1) (j)
· (j) (j)
τj+1 sn′ −sn +ǫ n′ 6=n sn′ −sn −ǫ n′ =1 sn′ − sn ǫ
n′ =1 a=1 ma −sn +(1+j) 2

JHEP05(2015)095
= (−1)δj , (5.42)

(j) (j) (j)

We now take the limit ǫ → ∞ combined with ma → ∞ such that −ma +(1−j) 2ǫ = −Ma
is kept fixed. In order to keep the whole expression finite the FI couplings have to run
thereby generating a scale Λj for each gauge group, so we have the following chiral ring
relations
Nj−1 Mj Nj+1
Y (j−1)
Y M +N −N
Y (j+1)
(s(j)
n − s n′ ) · (s(j)
n − Ma(j) ) = Λj j j−1 j+1 (s(j)
n − s n′ ), (5.43)
n′ =1 a=1 n′ =1

which is the quiver generalization of a CPN sigma model. We can also write the above
equation in Baxter form
Qi−1 Mi = ΛD
i Qi+1 ,
i
(5.44)

where
Di = Mj + Nj−1 − Nj+1 . (5.45)

For a complete flag case only M1 = Q0 is nonzero, the rest Mi vanish and vacua equa-
tions (5.44) can be written as

ei ,
Qi−1 − Λ2i Qi+1 = Qi Q (5.46)

e i . It is assumed that QN = 1. Following [14] we rewrite the

using auxiliary polynomial Q
above equation using Toda Lax matrix L

Li,j Qj = δi,1 M , (5.47)

where
e i δij + Λ2i δi,j+1 − δi,j−1 ,
Li,j = Q (5.48)

and it also follows that

M = detL . (5.49)
e i and the coordinates
Interestingly the momenta in Toda are the auxiliary polynomials Q
are dynamical scales Λi at each gauge group.

– 59 –
5.4.1 Equivariant quantum cohomology

One of the motivating results for this note is the famous theorem by Givental and Kim,
which we list below for completeness.

Theorem 5.1 (Givental, Kim 1993) The equivariant quantum cohomology ring of the
complete manifold of flags inside CN is isomorphic to

QHT• (FN ) ≃ C[p1 , . . . , pN , τ1 , . . . τN −1 , µ1 , . . . µN ]/IToda (5.50)

JHEP05(2015)095
where ideal IToda is generated by the coefficients of the following polynomial

N
Y
det (u − LToda (p, τ )) = (u − µi ) , (5.51)
j=1

where matrix LToda (Lax matrix of open Toda chain)

 
p1 τ1 0 . . . 0
−1 p2 τ 2 . . . 0 
 
 
LToda = 0 −1 p3 . . . 0  . (5.52)
 
· · · 
0 0 0 −1 pN

In the Lax matrix above we have identified momenta pi and (exponential of) relative
coordinates between Toda particles τi = eti −ti+1 with dynamical scales Λi and polynomials
Qe i respectively in (5.48). This theorem can now be viewed as an important limiting case
of our construction. In order to obtain the above result from Proposition 2.1 we need to
first, shrink the compactification radius R → 0, and, second, take the chiral limit in mass
parameters (5.3).
A detailed analysis of twisted chiral rings of (2, 2) and (2, 2)∗ quiver theories in two
dimensions in the context of ramified instanton counting and equivariant quantum coho-
mology was done in [65]. In particular, two-dimensional vortex partition functions for the
theories on two-dimensional defects inside four-dimensional N = 2 and N = 2∗ gauge the-
ories were computed, and the identification with Givental J-functions was presented. One
can also obtain these expressions by taking the 2d limit of 3d holomorphic blocks (3.37).
Note that currently in mathematical literature all necessary tools for computing quan-
tum cohomology of vector bundles over complex projective spaces and their generalizations
is already available. Thus using the results of [66] one can derive quantum J-function for
the cotangent bundle to the flag variety in the framework of [67] (see [65] for details).
Further connections to integrability in N = (2, 2)∗ theories were studied in [68], in
e
particular, QQ-type relations, similar to (2.7) were examined using cluster algebra methods.

– 60 –
5.4.2 Coupled 4d/2d systems
Using the results of the previous section we can compute twisted chiral rings of two di-
mensional quiver theories coupled to four dimensional N = 2∗ gauge theory by taking
the radius of the compact circle in the 5d computation to zero. Classical analysis has
been conducted in [14], in particular, the 4d analogue of the deformed twisted chiral ring
relation (4.2) was constructed in the cited paper for the SU(N ) N = 2∗ theory.

6 Summary and outlook

JHEP05(2015)095
6.1 Integrable systems
In this section we outline how the results of this paper fit into the wider context of integrable
many-body systems.
There exists a correspondence between the spectral curves of a family of many-body
integrable systems (Calogero-Moser-Sutherland [69–71], Ruijsenaars-Schneider [72], and
Double-elliptic [73–76]) and Seiberg-Witten curves of supersymmetric gauge theories with
adjoint matter in four, five and six dimensions (see [74] and references therein). These are
the examples that we have focussed on in this paper.
Recently, in [77] this correspondence was extended and systematized to a large class
four dimensional quiver gauge theories of finite and affine ADE type. Later in [46] the five
and six-dimensional versions of the Seiberg-Witten geometry together with its quantum
deformation are also discussed.

6.1.1 Calogero-Ruijsenaars-Dell family

Figure 5 exhibits a family on many-body classical integrable systems and our proposal
for how their eigenfunctions are realized by the partition functions of supersymmetric
gauge theories in various dimensions. The integrable models are characterized by how the
periodicity properties of the coordinates and momenta: each may be rational, trigonometric
and elliptic. The most general system studied in this paper is the elliptic RS system
(corresponding to 5d N = 2 gauge theory with a 3d surface defect), which has elliptic
dependence on positions and trigonometric dependence on momenta. We have indicated
how some of the remaining members of the family can be obtained by various limits.
In figure 5 the first row contains Calogero-Moser-Sutherland (CMS) family, the second
row describes Ruijsenaars-Schneider (RS) family, whereas the last row contains models
that are bispectrally dual to eCMS, eRS, and, finally, the double elliptic (Dell) system.
Blue double arrows describe bispectral dualities between the models in the table. Because
the properties of the Dell system are largely unexplored, the anticipated duality between
Dell and itself is designated by the dotted blue double arrow. Various limits are shown by
arrows. Thus the right column describes elliptic models, whose ellipticity parameter plays
a role of the bulk (4,5, or 6-dimensional) instanton fugacity Q. When we send Q → 0 bulk
degrees of freedom decouple and we are left with the corresponding defect theories which
are located in the middle column. These models are trigonometric in coordinates. Next,
one may take a limit when the adjoint mass ǫ vanishes. Provided that an appropriate

– 61 –
JHEP05(2015)095
Figure 5. Classification of integrable many-body systems according to their periodicity properties
in coordinates q (columns) and momenta p (rows).

scaling is chosen (see [6]), one obtains a family of rational models from the ǫ → 0 limit.
Equivalently, one can take the limits row-wise. First, starting from the Dell model and its
descendants in the last row take a limit Re → 0 thereby shrinking one of the extra compact
direction of the 6d theory. This limit moves us to the RS family. Further on, shrinkage of
the other circle (again, with a properly chosen scaling of parameters) R → 0 descends us
further to the CMS family. One can use the arrows we described above to navigate through
the diagram starting from Dell.
Recall that the classical tRS model was derived from the twisted chiral ring of T [U(N )]
theory (2.32). After quantization these relations become operator equations which we
solved in (3.30), where the eigenfunctions are written in (3.28) as a partition function
on the Coulomb branch of T [U(N )] theory, then in terms of vortex partition functions
in (3.37), and finally as a decoupling limit of the 5d/3d partition function in section 4.3.
The eigenvalues are characters of antisymmetric tensor representation of U(N ).
Next, we have found the eigenfunctions of elliptic RS system for two and three particles
as partition functions of 5d N = 1∗ U(2) (4.42) and U(3) (4.45) gauge theories in the
presence of monodromy defects of maximal type section 4.5. The eigenvalues in this case
can be computed in (4.31) and correspond to VEVs of Wilson loops in the skew powers of
the fundamental representation of U(N ).
Finally, we were able to find the spectrum of the two-body dual eRS model using
the U(2) 5d gauge theory coupled to 3d free hypermultiplets (4.59). By applying the S-
transformation (4.64) we have shown that the eigenfunctions of the elliptic RS model (4.71)
can be reproduced from the eigenfunctions of the dual model.

– 62 –
p finite affine

open Toda closed Toda

r 2d N=(2,2) 4d N=2 SYM
chiral quiver theory 2d defect

open relativistic Toda closed relativistic Toda

t 3d N=2 5d N=1 SYM
chiral quiver theory 3d defect

JHEP05(2015)095
closed elliptic Toda
open elliptic Toda
(??)
e 4d N=1
6d (1,0)
chiral quiver theory
4d defect

Figure 6. Classification of rational, trigonometric (relativistic) and elliptic Toda integrable sys-
tems. Closed Toda chains exhibit affine Lie algebra symmetry, whereas open chain are symmetric
under the action of the Lie algebra itself.

6.1.2 Toda family

In section 5 we have discussed 3d quiver gauge theories with chiral matter and pure 5d
SYM theories with defects which support chiral quiver theories. We have found quantum
spectra of relativistic open (5.30) and closed (5.37) Toda chains. In the limit when the
compactification radius of the 3d theory becomes small we recover nonrelativistic Toda
chains (see section 5.4). The classification is presented in figure 6. In the last row of the
table one can find elliptic Toda systems which were studied in [78].

6.1.3 Gaudin-spin chain family

For completeness we also discuss the quantum integrable models whose Bethe equations
reproduce the twisted chiral rings of the supersymmetric gauge theories appearing (after
decoupling Q → 0) on codimension two surface defects in the four, five and six-dimensional
theories we have discussed above. This is illustrated in figure 7 and can be compared to
figure 5.
The spin chain family (XXX, XXZ, and XYZ) in the middle column of figure 7 can
be described using the Nekrasov-Shatashvili correspondence using quiver gauge theories in
dimensions two, three, and four respectively. The Gaudin family in the first column can
be obtained from the spin chain family by turning off the ‘Planck constant’, which in our
construction is represented by the adjoint mass ǫ (see [6] for the details about XXX and
XXZ models). We refer the reader to [79] and references therein for the discussion about
XYZ spin chain and XYZ (elliptic) Gaudin model. The limit ǫ → 0 should be supplemented
by scaling of anisotropy parameters such that ǫmi remain constant.

– 63 –
JHEP05(2015)095
Figure 7. Classification of integrable models of Gaudin and spin chain type according to the
periodicity properties in twists t (columns) and anisotropies m (rows). Blue double arrows describe
bispectral dualities between the models in the table.

To the best of our knowledge there are no known quantum models with elliptic de-
pendence on twists which could be consistently placed in the right column of figure 7.
Moreover, our analysis involving defects in higher dimensional gauge theories with adjoint
matter fields does not immediately suggest any candidates for such models. Their existence,
is certainly an intriguing question on its own and should be pursued independently. 14

6.2 Open problems

Let us now articulate what we believe to be some important questions unanswered by the
present work. We list them below in the random order.

• In this paper we have conjectured the structure of formal eigenfunctions of the elliptic
Ruijsenaars-Schneider model and (in the 2-body case) its bispectral dual. We have
checked our conjectures by expanding the eigenfunctions (instanton partition func-
tions of 5d theories with codimension two defects) to first several orders. It should
be possible to prove these conjectures. We understand that work in this direction is
being pursued in [83].

• The Hamiltonians of trigonometric Ruijsenaars-Schneider model form a commutative

subalgebra inside double affine Hecke algebras (DAHA) and Cherednik algebras [84].
Further understanding of the role of Cherednik algebras in gauge theories is due.
14
Some work understanding these integrable models is being done in [80–82].

– 64 –
• We have analyzed in this paper five-dimensional theories with defects. An immediate
generalization would be to study six dimensional (1, 1) theory with a four dimensional
defect and use it to find formal eigenfunctions of the double-elliptic (Dell) model. It
would be interesting to understand the action of bispectral duality in this case.

• In the presence of monodromy defects, it is known that the Alday-Gaiotto-

Tachikawa [85] correspondence is modified to conformal blocks of affine algebras.
Understanding of five-dimensional gauge theories with surface defects might lead to
new ideas in quantum affine algebras (see, e.g. [86–89] for a related study of q-Toda
degenerate conformal blocks).

JHEP05(2015)095
• There is a notion of regular and irregular monodromy defects in gauge theories. In
this paper we have only addressed the regular case and computed instanton partition
functions in the presence of such monodromy defects. One may wonder what happens
in the case of irregular singularities.

• The Calogero-Moser-Sutherland family of integrable models is connected to the Ko-

rteweg — de Vries/Benjamin-Ono/Intermediate Long Wave hierarchy [80, 81, 90–93]
in the limit of large number of particles. One may study generalizations of the cor-
respondence to the Ruijsenaars-Schneider family. Some work in this direction has
already been done in [94–96].

• In this paper we discussed the relationship between the many-body trigonometric

Ruijsenaars-Schneider system and Bethe ansatz of quantum XXZ spin chains. It is
a legitimate question to ask how the correspondence is generalized to the related
elliptic many-body system.

Acknowledgments

We would like to thank Davide Gaiotto, Jaume Gomis, Nikita Nekrasov, Vasily Pestun,
Mina Aganagic, Shamil Shakirov, Nathan Haouzi, Satoshi Nawata, Amihay Hanany, An-
drey Zayakin, Mikhail Bershtein, Aleksey Morozov, Antonio Sciarappa, Alessandro Tanzini,
Giulio Bonelli, Heeyeon Kim for fruitful discussions.
MB gratefully acknowledges support from the Perimeter Institute for Theoretical
Physics and IAS Princeton through the Martin A. and Helen Choolijan Membership. PK
thanks W. Fine Institute for Theoretical Physics at University of Minnesota, Kavli Institute
for Theoretical Physics at University of California Santa Barbara, University of California
at Berkeley, and Simons Center for Geometry and Physics where part of his work was done,
for kind hospitality. HC gratefully acknowledges support from the Perimeter Institute for
Theoretical Physics, the organizers of “Exact Results in SUSY Gauge Theories in Various
Dimensions” at CERN and also CERN-Korea Theory Collaboration funded by National
Research Foundation (Korea) for the hospitality and support. This work was presented at a
number of conferences and seminars including “IV SISSA Workshop on Dualities and Geo-
metric Correspondences” and KITP program “New Methods in Nonperturbative Quantum
Field Theory”. We are grateful to the organizes for invitations and kind hospitality.

– 65 –
Our research was partly supported by the Perimeter Institute for Theoretical Physics.
Research at Perimeter Institute is supported by the Government of Canada through Indus-
try Canada and by the Province of Ontario through the Ministry of Economic Development
and Innovation.

A Conventions of special functions

We summarize our conventions on various functions we have used in the main text.

A.1 Q-hypergeometric functions

JHEP05(2015)095
We use the following definition for Q-hypergeometric functions
∞
X (a; q)k (b; q)k zk
2 F1 (a, b; c; q, z) = . (A.1)
(c; q)k (q; q)k
k=1

A.2 Double sine functions

Here we use the double sine function S2 (z|~
ω ) where ω
~ = (ω1 , ω2 ) with ω1 , ω2 > 0 defined for
example in the appendix of. Let us summarize the properties written there. This function
has an integral representation
Z
iπ ezt dt
log S2 (z|~
ω ) = B2,2 (z|~
ω) + ω t ω t
(A.2)
2 (e − 1)(e − 1) t
1 1
R+i0

where B2,2 (z|~

ω ) is the multiple Bernoulli polynomial. The most important property for
this paper is the functional relations
S2 (z + ω1 |~
ω)
. (A.3)
S2 (z|~
ω)
A.3 Theta functions
We have used basic theta-function in the computations in the main text
Y
−1 m q m+1
θ(a, q) = (a, q)∞ (qa , q)∞ = (1 − aq ) 1 − , (A.4)
a
m≥0

where, we recall, the q-Pochhammer symbol is defined as follows

 Q
 n−1 i
 i=0 (1 − xq ) for n > 0 ,
(x; q)n ≡ 1 for n = 0 , (A.5)
 Q−n (1 − xq −i )−1 for n < 0 .

i=1

It is related to Kac-Weyl character for Â1 as

Y
θKW (a, q) = θ(a, q) · (1 − q m+1 ) , (A.6)
m≥0

which in turn is related to conventional elliptic theta functions associated to the elliptic
curve constructed on complex vector q. For instance
1 1
θ1 (a, q) = −iq − 8 a 2 θKW (a, q) . (A.7)

– 66 –
N 1 N 2
N L N

Figure 8. Quiver diagram for T [U(N )]ρ theory where ρTi = Ni+1 − Ni .

B Factorization of Tρ partition functions

The T [U(N )]ρ theory is the 3d N = 4 quiver gauge theory with gauge group U(N1 ) × · · · ×

JHEP05(2015)095
U(NL ) and N fundamental hypermultiplets for the last gauge group (see figure 8).
In this appendix we shall compute the Higgs branch representation of the ellipsoid
partition function for T [U(N )]ρ theory. The partition function is given by

Z Y
L
ρ dx(n) 2π PLn=1 PNn x(n) (tn −tn+1 )
ZU(N ) = e i=1 i (B.1)
Nn !
n=1
Nn
Y Nn
Y Nn NY
Y n+1
(n) (n) ε∗ (n) (n+1)
× 2 sinh(πb± xij ) Sb ε + ixij Sb ± i(xi − xj ) ,
2
i<j i,j=1 i=1 j=1

(L+1)
with NL+1 = N and xi = mi , the N mass parameters for the flavor symmetry.
The integrand has infinite number of poles and zeros. We suppose that the physi-
cally relevant poles are only from the bifundamental hypermultiplets. 15 Then the contour
integral can be done by taking residues from the relevant poles. The poles take the form of

(n) (L − n + 1)ε∗ (n) (n)

xi = mj + i + iki b + ie
ki b−1 (B.2)
2

(n) (n)
with j = 1, · · · , N and non-negative integers ki , e
ki ≥ 0. It is convenient to define new
(n)
integral variables si such as

(n) (n) (L − n + 1)ε∗ (n) (n)

xi = si + mj + i + iki b + ie
ki b−1 (B.3)
2

(n) (n) (n)

and perform the integral over si instead of xi . This effectively shifts all poles to si = 0.
With new variables the integral can be rewritten as

ρ 1 X Z Y Nn
L Y
ds
(n)
i
ZU(N ) (mi , tn ) = QL Z(s, mσ(j) , t) (B.4)
i=1 ρi ! σ∈SN
2πi
n=1 i=1

15
This is proven for T [U(N )] theory in section 3.5.

– 67 –
where
PL P Nn (n) (L−n+1)ε∗ (n) (n)
i=1 (tn −tn+1 )(si +mi +i +iki b+ie
ki b−1 )
Z(s, m, t) = e2π n=1 2 (B.5)
X Nn
L Y
Y
(n) (n) (n)
× 2 sinh πb± (sij +mij +ikij b+ie
kij b−1 )
~k,~e n=1 i<j
k≥0
Nn
Y
(n) (n) (n)
× S ε−isij −imij +kij b+ e
kij b−1
i,j=1

(n) (n+1) (n) (n+1) (n) (n+1) −1
Nn NY S −isi + isj − imij + (ki − kj )b + (e
ki − e kj )b + ε∗

JHEP05(2015)095
Y n+1

×
(n) (n+1) (n) (n+1) (n) (n+1)
i=1 j=1 S −isi + isj − imij + (ki − kj + 1)b + (eki − e kj + 1)b−1

(L+1) (L+1) (L+1)

with si = e
= 0 and ki
ki = 0. SN is the Weyl group of the U(N ) flavor
~
symmetry and we denote collectively by ~k and e
k all the (semi-)positive integer numbers
(n) e(n)
ki and ki respectively. The integrand can be further simplified using the following
identity

in+m+2mn e−πix(nb+mb
−1 )

Sb (x + nb + mb−1 ) =
−1 x
(e2πibx ; q)n (e2πib ; qe)m Sb (x) (B.6)
q n(n−1)/4 qem(m−1)/4
2 2
with q ≡ e2πib , qe ≡ e2πi/b . The function Z reduces to

L (n) |ek(n) |
PL P Nn (n) (L−n+1)ε∗ X Y τn |k | τen
Z(s, m, t) = e 2π n=1 i=1 (tn −tn+1 )(si +mi +i 2
)
τn+1 τen+1
~k,~e n=1
k≥0
(n) (n)
σi µi σ
ei µ
ei
Nn
Y Nn
Y qη −2 (n) µj ; q k(n) η −2
q̃e (n) ej ; q
µ e (n)
σj σ
ej e
kij
(n) (n)
× 2 sinh πb± (sij +mij ) S(ε+isij +imij ) · (n)
ij
· (n)
σi µi σ
ei µ
ei
µj ; q k(n) ej ; q
e (n)
i<j i,j=1 (n)
σj σ
ej
(n) µ e
kij
ij
(n) (n)
σi µi σ
ei µ
ei
Nn NY (n) (n+1) η2 µj ; q k(n)−k(n+1) ηe2 ej ; q
e (n) (n+1)
Y n+1
S(ε∗ −isi +isj −imij ) (n+1)
σi i j
σ
(n+1)
ei µ e
ki −e
kj
× (n) (n+1)
· (n) · (n)
S(2b+ −isi +isj −imij ) σi µi σ
ei µ
ei
i=1 j=1 q (n+1) µj ; q k(n) −k(n+1) qe (n+1) ej ; q
µ e (n) (n+1)
σi σ
ei e
ki −e
kj
i j
1 PL (n) | 1 PL e(n) |
×(qη −2 ) 2 n=1 (Nn+1 −Nn−1 )|k q ηe−2 ) 2
(e n=1 (Nn+1 −Nn−1 )|k , (B.7)

where

Nn
X Nn
X
(n) (n)
(n) (n) (n) (n)
σi =e 2πbsi
, |k (n)
|= ki , σ
ei =e 2πb−1 si
, |e
k (n) | = e
ki . (B.8)
i=1 i=1

(n)
All relevant poles are simple poles located at the origin si = 0, so the residue computation
is straightforward. Finally, the residue sum provides the Higgs branch representation of

– 68 –
the partition function of the T [U(N )]ρ theory

ρ 1 X
ZU(N ) = QL Zclass (mσ(i) , t)Z1−loop (mσ(i) )ZV (mσ(i) )ZAV (mσ(i) ) ,
i=1 ρi ! σ∈SN
P P Nn (L−n+1)ε∗
2π L i=1 (tn −tn+1 )(mi +i )
Zclass = e n=1 2 ,
Nn
L Y Nn Nn QNn+1
Y Y Y j6=i Sb (imij )
±
Z1−loop = 2 sinh πb (mij ) S(imij + ε) QNn+1 ,
n=1 i<j i,j=1 i=1 j=1 Sb (imij + ε)
X L
Y |k(n) | PL
τn 1 n=1 (Nn+1−Nn−1 )|k
(n) |
ZV = qη −2 2

JHEP05(2015)095
τn+1
~k≥0 n=1

qη −2 µi ; q η 2 µi ; q
L N N N
YY n µj k
(n) Y n Y n+1 µj (n) (n+1)
k −kj
× ij i ,
µi µi
n=1 i6=j µj ; q (n) i=1 j=1 q µj ; q (n) (n+1)
kij ki −kj

ZAV = ZV ((z, µ, η, q) → (e
z, µ
e, ηe, qe)) . (B.9)

The result consists of the classical and 1-loop determinant and non-perturbative parts and
the non-perturbative contribution factorizes into the vortex and anti-vortex series.

C Perturbative 5d partition functions

In this appendix, we briefly discuss perturbative partition functions and regularization

issue. We will compute the 1-loop determinants of 5d theory on S 1 × C2 and 3d theory
on S 1 × C where S 1 denotes the time circle. We will also consider the elliptic genus of 2d
theory which can arise as a boundary theory of the 3d theory.
Let us start with the 1-loop determinants of 5d theories. We can use the equivariant
indices for the vector and hypermultiplets

1 + e−2ǫ+ X
χvec = − eα(a) ,
2(1 − e−ǫ1 )(1 − e−ǫ2 ) α
em−2ǫ+ X
χhyper = eρ(a) . (C.1)
(1 − e−ǫ1 )(1 − e−ǫ2 ) ρ

where α are the roots and ρ are the weights of the gauge group and m is the equivariant
parameter for the flavor symmetry. The denominator factors is understood as a power
series expansion in terms of eǫ1 and eǫ2 . When we compute the 1-loop determinants using
these equivariant indices, there is an issue related to the boundary condition on ∂C2 , which
has not been clarified yet. We will not attempt to clarify this issue in this paper, but we
will adopt the following prescription. For the multiplets in the complex representations
in the gauge group, we will simply use the above equivariant indices and compute the
corresponding 1-loop determinants. However, for the real multiplets, we will first take
average of the equivariant index with its charge conjugation and then compute the 1-loop
determinant. This prescription respects the invariance of the real multiplets under the

– 69 –
charge conjugation. We also remind the reader that the equivariant indices implicitly have
P 2πi
t
the factor t∈Z e β for the Kaluza-Klein momenta along the temporal circle. In what
follows we set β = 1 for convenience, which can be restored by scaling other chemical
potentials.
We find the following 1-loop determinants
YY Y ∞
5d,vec 2πi
Z1−loop = t + pǫ1 + qǫ2 + α(a)
α t∈Z p,q≥0
β
1/2
2πi
× t + (p + 1)ǫ1 + (q + 1)ǫ2 + α(a) ,

JHEP05(2015)095
β
YY Y ∞ −1
5d,hyper 2πi
Z1−loop = t + pǫ1 + qǫ2 + α(a) + m . (C.2)
ρ
β
t∈Z p,q≥0

where we assumed that ρ is in a complex representation.

These partition functions are divergent infinite products which need to be properly
regularized. We will regularize them using Barnes’ multiple gamma functions. Barnes’
gamma functions are defined as regularized infinite products [97, 98]
Y
ΓN (z|w1 , · · · , wN ) = exp (∂s ζN (s, z; w1 , · · · , wN )|s=0 ) ∼ (z + n · w)−1 , (C.3)
n1 ,··· ,nN ≥0

where Barnes’ zeta function is defined by the series

X
ζN (s, z|w1 , · · · , wN ) = (z + n · w)−s . (C.4)
n1 ,··· ,nM ≥0

When Im(w) > 0, the following identity holds

Y
ΓN +1 (z|1, w)ΓN +1 (1 − z|1, −w) = e−πiζN +1 (0,z|1,w) (1 − e2πi(z+n·w) )−1 . (C.5)
n1 ,··· ,nN ≥0

The 1-loop determinants regularized using Barnes’ gamma functions can be written as

5d,vec
Y α(a) ǫ1 ǫ2 α(a) ǫ1 ǫ2
Z1−loop = Γ3 1, , Γ3 1 − 1, − ,−
α
2πi 2πi 2πi 2πi 2πi 2πi
α(a)+2ǫ
+ ǫ1 ǫ2 α(a)+2ǫ+ ǫ1 ǫ2 −1/2
× Γ3 1, , Γ3 1− 1, − ,− ,
2πi 2πi 2πi 2πi 2πi 2πi
5d,hyper
Y ρ(a)+m ǫ1 ǫ2 ρ(a)+m ǫ1 ǫ2
Z1−loop = Γ3 1, , Γ3 1− 1, − ,− . (C.6)
ρ
2πi 2πi 2πi 2πi 2πi 2πi

Assuming q = eǫ1 , p = eǫ2 < 1, they can be further simplified to

5d,vec 5d
Y Y h i1/2
Z1−loop = e−Fvec (a;ǫ1 ,ǫ2 ) (1 − eα(a) pn1 q n2 )(1 − eα(a) pqpn1 q n2 ) ,
α n1 ,n2 ≥0
5d Y Y
5d,hyper
Z1−loop = e−Fhyper (a,m;ǫ1 ,ǫ2 ) (1 − eρ(a)+m pn1 q n2 )−1 , (C.7)
ρ n1 ,n2 ≥0

– 70 –
where the prefactors induced after the regularization are written as

5d πi X α(a) ǫ1 ǫ2 α(a) + 2ǫ
+ ǫ1 ǫ2
Fvec = − ζ3 0, 1, , + ζ3 0, 1, , ,
2 α 2πi 2πi 2πi 2πi 2πi 2πi
X ρ(a) + m ǫ1 ǫ2
5d
Fhyper = πi ζ3 0, 1, , . (C.8)
ρ
2πi 2πi 2πi

These prefactors encode the 1-loop corrections to the effective action of the 5d gauge theory.
We now turn to the perturbative partition functions of N = 2 field theories in three
dimensions. The 3d partition function on S 1 × D2 has been recently computed in [13] and

JHEP05(2015)095
in [54] using localization.
The equivariant index of the chiral multiplet with R-charge ∆ is
∆
em e− 2 ǫ1 X ρ(a)
χ3d
chiral = e . (C.9)
1 − e−ǫ1 ρ

In the 1-loop computation, we need to take into account a proper boundary condition
in supersymmetric and gauge invariant fashion. For the N = 2 chiral multiplet, we can
choose either the Neumann or the Dirichlet boundary condition. It appears that the
boundary condition determines how to expand the denominator of the equivariant index.
The computation in [54] implies that we have to expand it in power series of e−ǫ1 for the
chiral multiplet with the Neumann boundary condition, while expand it in power series
of eǫ1 for the chiral multiplet with the Dirichlet boundary condition. If we impose the
Neumann boundary condition, the 1-loop determinant of the chiral multiplet becomes
YY Y
3d
Zchiral,N = (2πit − (∆/2 + n)ǫ1 + ρ(a) + m)−1
ρ t∈Z n≥0
Y −ρ(a) − m + ∆
ǫ1 −ρ(a) − m + ∆
ǫ1
2 ǫ1 2 ǫ1
= Γ2 1, Γ2 1 − 1, −
ρ
2πi 2πi 2πi 2πi
Y −ρ(a)−m+ ∆
2 ǫ1 |1,− ǫ2 )
Y
= e−πiζ2 (0, 2πi 2πi (1 − e−ρ(a)−m q ∆/2 q n )−1 . (C.10)
ρ n≥0

We regularized the infinite products using the Barnes’ gamma functions as we did in 5d
theories. On the other hand, the chiral multiplet with the Dirichlet boundary condition
has the following 1-loop determinant
YY Y
3d
Zchiral,D = (2πit + (−∆/2 + n + 1)ǫ1 + ρ(a) + m)
ρ t∈Z n≥0
" #−1
Y ρ(a)+m+(1− ∆ )ǫ ǫ1 ρ(a)+m+(1 − ∆
ǫ1
2 1 2 )ǫ1
= Γ2 1, Γ2 1− 1, −
ρ
2πi 2πi 2πi 2πi
Y ρ(a)+m+(1− ∆
2 )ǫ1 |1, ǫ1 )
Y
= e−πiζ2 (0, 2πi 2πi (1 − eρ(a)+m q 1−∆/2 q n ) . (C.11)
ρ n≥0

The prefactors involve the 1-loop corrections to the (mixed-) Chern-Simons terms by the
matter fields.

– 71 –
Lastly, let us compute the elliptic genera of 2d (0, 2) multiplets. See [57, 60, 99] for
details. The chiral multiplet with R-charge ∆ contributes to the elliptic genus as
Y Y
2d
Zchiral = (2πin1 − (n2 + ∆/2)ǫ1 + ρ(a) + m)−1
ρ n1 ,n2 ∈Z
Y −ρ(a) − m + ∆
ǫ1 −ρ(a) − m + ∆
ǫ1
2 ǫ1 2 ǫ1
= Γ2 1, Γ2 1 − 1, −
ρ
2πi 2πi 2πi 2πi
ρ(a) + m + (1 − ∆
ǫ1 ρ(a) + m + (1 − ∆
ǫ1
2 )ǫ1 2 )ǫ1 )ǫ1
× Γ2 1, Γ2 1 − 1, −
2πi 2πi 2πi 2πi
Y −ρ(a)−m+ ∆

JHEP05(2015)095
−2πiζ2 (0, 2 ǫ1 |1, ǫ1 )
= e 2πi 2πi θ(e−ρ(a)−m q ∆/2 ; q)−1 . (C.12)
ρ

On the other hand the elliptic genus of the fermi multiplet with R-charge ∆ is given by
Y Y
2d
Zfermi = (2πin1 − (n2 + ∆/2)ǫ1 + ρ(a) + m)
ρ n1 ,n2 ∈Z
Y −ρ(a)−m+ ∆
2 ǫ1 |1, ǫ1 )
= e2πiζ2 (0, 2πi 2πi θ(e−ρ(a)−m q ∆/2 ; q) . (C.13)
ρ

Open Access. This article is distributed under the terms of the Creative Commons
Attribution License (CC-BY 4.0), which permits any use, distribution and reproduction in
any medium, provided the original author(s) and source are credited.

References
[1] N. Seiberg and E. Witten, Electric-magnetic duality, monopole condensation and
confinement in N = 2 supersymmetric Yang-Mills theory, Nucl. Phys. B 426 (1994) 19
[hep-th/9407087] [INSPIRE].
[2] N. Seiberg and E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2
supersymmetric QCD, Nucl. Phys. B 431 (1994) 484 [hep-th/9408099] [INSPIRE].
[3] S. Gukov and E. Witten, Gauge Theory, Ramification, And The Geometric Langlands
Program, hep-th/0612073 [INSPIRE].
[4] S. Gukov and E. Witten, Rigid Surface Operators, Adv. Theor. Math. Phys. 14 (2010) 87
[arXiv:0804.1561] [INSPIRE].
[5] D. Gaiotto and E. Witten, S-duality of Boundary Conditions In N = 4 Super Yang-Mills
Theory, Adv. Theor. Math. Phys. 13 (2009) 721 [arXiv:0807.3720] [INSPIRE].
[6] D. Gaiotto and P. Koroteev, On Three Dimensional Quiver Gauge Theories and
Integrability, JHEP 05 (2013) 126 [arXiv:1304.0779] [INSPIRE].
[7] N.A. Nekrasov and S.L. Shatashvili, Quantum integrability and supersymmetric vacua, Prog.
Theor. Phys. Suppl. 177 (2009) 105 [arXiv:0901.4748] [INSPIRE].
[8] N.A. Nekrasov and S.L. Shatashvili, Supersymmetric vacua and Bethe ansatz, Nucl. Phys.
Proc. Suppl. 192-193 (2009) 91 [arXiv:0901.4744] [INSPIRE].
[9] N.A. Nekrasov and S.L. Shatashvili, Quantization of Integrable Systems and Four
Dimensional Gauge Theories, arXiv:0908.4052 [INSPIRE].

– 72 –
[10] A. Kapustin, B. Willett and I. Yaakov, Exact Results for Wilson Loops in Superconformal
Chern-Simons Theories with Matter, JHEP 03 (2010) 089 [arXiv:0909.4559] [INSPIRE].
[11] N. Hama, K. Hosomichi and S. Lee, SUSY Gauge Theories on Squashed Three-Spheres,
JHEP 05 (2011) 014 [arXiv:1102.4716] [INSPIRE].
[12] S. Pasquetti, Factorisation of N = 2 Theories on the Squashed 3-Sphere, JHEP 04 (2012)
120 [arXiv:1111.6905] [INSPIRE].
[13] C. Beem, T. Dimofte and S. Pasquetti, Holomorphic Blocks in Three Dimensions, JHEP 12
(2014) 177 [arXiv:1211.1986] [INSPIRE].
[14] D. Gaiotto, S. Gukov and N. Seiberg, Surface Defects and Resolvents, JHEP 09 (2013) 070

JHEP05(2015)095
[arXiv:1307.2578] [INSPIRE].
[15] N. Nekrasov, Five dimensional gauge theories and relativistic integrable systems, Nucl. Phys.
B 531 (1998) 323 [hep-th/9609219] [INSPIRE].
[16] L.F. Alday and Y. Tachikawa, Affine SL(2) conformal blocks from 4d gauge theories, Lett.
Math. Phys. 94 (2010) 87 [arXiv:1005.4469] [INSPIRE].
[17] H. Kanno and Y. Tachikawa, Instanton counting with a surface operator and the chain-saw
quiver, JHEP 06 (2011) 119 [arXiv:1105.0357] [INSPIRE].
[18] J. Lamy-Poirier, Localization of a supersymmetric gauge theory in the presence of a surface
defect, arXiv:1412.0530 [INSPIRE].
[19] D. Gaiotto and H.-C. Kim, Surface defects and instanton partition functions,
arXiv:1412.2781 [INSPIRE].
[20] E. Witten, SL(2, Z) action on three-dimensional conformal field theories with Abelian
symmetry, hep-th/0307041 [INSPIRE].
[21] K.A. Intriligator and N. Seiberg, Mirror symmetry in three-dimensional gauge theories, Phys.
Lett. B 387 (1996) 513 [hep-th/9607207] [INSPIRE].
[22] J. de Boer, K. Hori, H. Ooguri, Y. Oz and Z. Yin, Mirror symmetry in three-dimensional
gauge theories, sl(2, z) and d-brane moduli spaces, Nucl. Phys. B 493 (1997) 148
[hep-th/9612131].
[23] A. Hanany and E. Witten, Type IIB superstrings, BPS monopoles and three-dimensional
gauge dynamics, Nucl. Phys. B 492 (1997) 152 [hep-th/9611230] [INSPIRE].
[24] V. Kazakov and P. Vieira, From characters to quantum (super)spin chains via fusion, JHEP
10 (2008) 050 [arXiv:0711.2470] [INSPIRE].
[25] N. Nekrasov, A. Rosly and S. Shatashvili, Darboux coordinates, Yang-Yang functional and
gauge theory, Nucl. Phys. Proc. Suppl. 216 (2011) 69 [arXiv:1103.3919] [INSPIRE].
[26] H. Nakajima, Instantons on ale spaces, quiver varieties, and kac-moody algebras, Duke Math.
J. 76 (1994) 365.
[27] H. Nakajima, Homology of moduli spaces of instantons on ALE spaces. 1, J. Diff. Geom. 40
(1994) 105 [INSPIRE].
[28] H. Nakajima, Quiver varieties and kac-moody algebras, Duke Math. J. 91 (1998) 515.
[29] A. Givental and B.-s. Kim, Quantum cohomology of flag manifolds and Toda lattices,
Commun. Math. Phys. 168 (1995) 609 [hep-th/9312096] [INSPIRE].

– 73 –
[30] B. Kim, Quantum cohomology of flag manifolds G/B and quantum Toda lattices,
alg-geom/9607001.
[31] A. Givental and Y.P. Lee, Quantum k-theory on flag manifolds, finite-difference toda lattices
and quantum groups, math/0108105.
[32] A. Braverman and M. Finkelberg, Finite difference quantum Toda lattice via equivariant
k-theory, Transform. Groups 10 (2005) 363 [math/0503456] [INSPIRE].
[33] R. Rimanyi, V. Tarasov and A. Varchenko, Trigonometric weight functions as K-theoretic
stable envelope maps for the cotangent bundle of a flag variety, J. Geom. Phys. 94 (2015) 81
[arXiv:1411.0478].

JHEP05(2015)095
[34] N. Hama, K. Hosomichi and S. Lee, Notes on SUSY Gauge Theories on Three-Sphere, JHEP
03 (2011) 127 [arXiv:1012.3512] [INSPIRE].
[35] J. Gomis, T. Okuda and V. Pestun, Exact Results for ’t Hooft Loops in Gauge Theories on
S 4 , JHEP 05 (2012) 141 [arXiv:1105.2568] [INSPIRE].
[36] J. Gomis and B. Le Floch, ’t Hooft Operators in Gauge Theory from Toda CFT, JHEP 11
(2011) 114 [arXiv:1008.4139] [INSPIRE].
[37] M. Bullimore, M. Fluder, L. Hollands and P. Richmond, The superconformal index and an
elliptic algebra of surface defects, JHEP 1410 (2014) 62 [arXiv:1401.3379] [INSPIRE].
[38] M. Bullimore, Defect Networks and Supersymmetric Loop Operators, JHEP 02 (2015) 066
[arXiv:1312.5001] [INSPIRE].
[39] M. Hallnäs and S. Ruijsenaars, Kernel functions and Bäcklund transformations for relativistic
Calogero-Moser and Toda systems, J. Math. Phys. 53 (2012) 123512 [arXiv:1206.3786].
[40] K. Taipale, K-theoretic J-functions of type A flag varieties, arXiv:1110.3117 [INSPIRE].
[41] A. Braverman, M. Finkelberg and J. Shiraishi, Macdonald polynomials, Laumon spaces and
perverse coherent sheaves, arXiv:1206.3131.
[42] D. Gaiotto, Surface Operators in N = 2 4d Gauge Theories, JHEP 11 (2012) 090
[arXiv:0911.1316] [INSPIRE].
[43] J. Lamy-Poirier, Localization of a supersymmetric gauge theory in the presence of a surface
defect, arXiv:1412.0530 [INSPIRE].
[44] A. Losev, N. Nekrasov and S.L. Shatashvili, Issues in topological gauge theory, Nucl. Phys. B
534 (1998) 549 [hep-th/9711108] [INSPIRE].
[45] H. Nakajima, More lectures on Hilbert schemes of points on surfaces, arXiv:1401.6782.
[46] N. Nekrasov, V. Pestun and S. Shatashvili, Quantum geometry and quiver gauge theories,
arXiv:1312.6689 [INSPIRE].
[47] M. Bullimore and H.-C. Kim, The Superconformal Index of the (2,0) Theory with Defects,
arXiv:1412.3872 [INSPIRE].
[48] D. Gaiotto, L. Rastelli and S.S. Razamat, Bootstrapping the superconformal index with
surface defects, JHEP 01 (2013) 022 [arXiv:1207.3577] [INSPIRE].
[49] L. Bao, V. Mitev, E. Pomoni, M. Taki and F. Yagi, Non-Lagrangian Theories from Brane
Junctions, JHEP 01 (2014) 175 [arXiv:1310.3841] [INSPIRE].
[50] H. Hayashi, H.-C. Kim and T. Nishinaka, Topological strings and 5d TN partition functions,
JHEP 06 (2014) 014 [arXiv:1310.3854] [INSPIRE].

– 74 –
[51] C. Hwang, J. Kim, S. Kim and J. Park, General instanton counting and 5d SCFT,
arXiv:1406.6793 [INSPIRE].
[52] A. Gadde, S. Gukov and P. Putrov, Fivebranes and 4-manifolds, arXiv:1306.4320
[INSPIRE].
[53] Y. Yoshida, Localization of Vortex Partition Functions in N = (2, 2) Super Yang-Mills
theory, arXiv:1101.0872 [INSPIRE].
[54] Y. Yoshida and K. Sugiyama, Localization of 3d N = 2 Supersymmetric Theories on
S 1 × D2 , arXiv:1409.6713 [INSPIRE].
[55] H.-C. Kim, S. Kim, E. Koh, K. Lee and S. Lee, On instantons as Kaluza-Klein modes of

JHEP05(2015)095
M5-branes, JHEP 12 (2011) 031 [arXiv:1110.2175] [INSPIRE].
[56] L.C. Jeffrey and F.C. Kirwan, Localization for nonabelian group actions, alg-geom/9307001.
[57] F. Benini, R. Eager, K. Hori and Y. Tachikawa, Elliptic Genera of 2d N = 2 Gauge
Theories, Commun. Math. Phys. 333 (2015) 1241 [arXiv:1308.4896] [INSPIRE].
[58] C. Cordova and S.-H. Shao, An Index Formula for Supersymmetric Quantum Mechanics,
arXiv:1406.7853 [INSPIRE].
[59] K. Hori, H. Kim and P. Yi, Witten Index and Wall Crossing, JHEP 01 (2015) 124
[arXiv:1407.2567] [INSPIRE].
[60] A. Gadde and S. Gukov, 2d Index and Surface operators, JHEP 03 (2014) 080
[arXiv:1305.0266] [INSPIRE].
[61] S. Kharchev, D. Lebedev and M. Semenov-Tian-Shansky, Unitary representations of U (q)
(sl(2, R)), the modular double and the multiparticle q deformed Toda chains, Commun. Math.
Phys. 225 (2002) 573 [hep-th/0102180] [INSPIRE].
[62] A. Braverman and M. Finkelberg, Semi-infinite Schubert varieties and quantum k-theory of
flag manifolds, arXiv:1111.2266 [INSPIRE].
[63] A. Braverman and M. Finkelberg, Twisted zastava and $q$-Whittaker functions,
arXiv:1410.2365.
[64] A. Kapustin and B. Willett, Wilson loops in supersymmetric Chern-Simons-matter theories
and duality, arXiv:1302.2164 [INSPIRE].
[65] S. Nawata, Givental J-functions, Quantum integrable systems, AGT relation with surface
operator, arXiv:1408.4132 [INSPIRE].
[66] T. Coates and A. Givental, Quantum Riemann-Roch, Lefschetz and Serre, math/0110142.
[67] I. Ciocan-Fontanine, B. Kim and C. Sabbah, The Abelian/Nonabelian correspondence and
Frobenius manifolds, Invent. Math. 171 (2007) 301 [math/0610265].
[68] F. Benini, D.S. Park and P. Zhao, Cluster algebras from dualities of 2d N = (2, 2) quiver
gauge theories, arXiv:1406.2699 [INSPIRE].
[69] R. Donagi and E. Witten, Supersymmetric Yang-Mills theory and integrable systems, Nucl.
Phys. B 460 (1996) 299 [hep-th/9510101] [INSPIRE].
[70] H. Itoyama and A. Morozov, Integrability and Seiberg-Witten theory: curves and periods,
Nucl. Phys. B 477 (1996) 855 [hep-th/9511126] [INSPIRE].
[71] H. Itoyama and A. Morozov, Prepotential and the Seiberg-Witten theory, Nucl. Phys. B 491
(1997) 529 [hep-th/9512161] [INSPIRE].

– 75 –
[72] H.W. Braden, A. Marshakov, A. Mironov and A. Morozov, The Ruijsenaars-Schneider model
in the context of Seiberg-Witten theory, Nucl. Phys. B 558 (1999) 371 [hep-th/9902205]
[INSPIRE].
[73] H.W. Braden, A. Marshakov, A. Mironov and A. Morozov, On double elliptic integrable
systems. 1. A Duality argument for the case of SU(2), Nucl. Phys. B 573 (2000) 553
[hep-th/9906240] [INSPIRE].
[74] A. Gorsky and A. Mironov, Integrable many body systems and gauge theories,
hep-th/0011197 [INSPIRE].
[75] H.W. Braden, A. Gorsky, A. Odessky and V. Rubtsov, Double elliptic dynamical systems

JHEP05(2015)095
from generalized Mukai-Sklyanin algebras, Nucl. Phys. B 633 (2002) 414 [hep-th/0111066]
[INSPIRE].
[76] H.W. Braden and T.J. Hollowood, The Curve of compactified 6 − D gauge theories and
integrable systems, JHEP 12 (2003) 023 [hep-th/0311024] [INSPIRE].
[77] N. Nekrasov and V. Pestun, Seiberg-Witten geometry of four dimensional N = 2 quiver
gauge theories, arXiv:1211.2240 [INSPIRE].
[78] I.M. Krichever, Elliptic analog of the Toda lattice, hep-th/9909224 [INSPIRE].
[79] E.K. Sklyanin and T. Takebe, Algebraic Bethe ansatz for XYZ Gaudin model, Phys. Lett. A
219 (1996) 217 [q-alg/9601028] [INSPIRE].
[80] A.V. Litvinov, On spectrum of ILW hierarchy in conformal field theory, JHEP 11 (2013) 155
[arXiv:1307.8094] [INSPIRE].
[81] M.N. Alfimov and A.V. Litvinov, On spectrum of ILW hierarchy in conformal field theory II:
coset CFT’s, JHEP 02 (2015) 150 [arXiv:1411.3313] [INSPIRE].
[82] N. Nekrasov and A. Okounkov, in progress.
[83] N. Nekrasov, Non-perturbative Dyson-Schwinger Equations and BPS/CFT Correspondence,
to appear.
[84] I. Cherednik, A unification of knizhnik-zamolodchikov and dunkl operators via affine hecke
algebras, Invent. Math. 106 (1991) 411.
[85] L.F. Alday, D. Gaiotto and Y. Tachikawa, Liouville Correlation Functions from
Four-dimensional Gauge Theories, Lett. Math. Phys. 91 (2010) 167 [arXiv:0906.3219]
[INSPIRE].
[86] F. Nieri, S. Pasquetti and F. Passerini, 3d and 5d Gauge Theory Partition Functions as
q-deformed CFT Correlators, Lett. Math. Phys. 105 (2015) 109 [arXiv:1303.2626]
[INSPIRE].
[87] F. Nieri, S. Pasquetti, F. Passerini and A. Torrielli, 5D partition functions, q-Virasoro
systems and integrable spin-chains, JHEP 12 (2014) 040 [arXiv:1312.1294] [INSPIRE].
[88] M. Aganagic, N. Haouzi and S. Shakirov, An -Triality, arXiv:1403.3657 [INSPIRE].
[89] M. Aganagic, N. Haouzi, C. Kozcaz and S. Shakirov, Gauge/Liouville Triality,
arXiv:1309.1687 [INSPIRE].
[90] D.R. Lebedev and A.O. Radul, Generalized internal long waves equations: construction,
hamiltonian structure and conservation laws, Commun. Math. Phys. 91 (1983) 543 [INSPIRE].

– 76 –
[91] P. Dorey, C. Dunning and R. Tateo, The ODE/IM Correspondence, J. Phys. A 40 (2007)
R205 [hep-th/0703066] [INSPIRE].
[92] A.G. Abanov, E. Bettelheim and P. Wiegmann, Integrable hydrodynamics of
Calogero-Sutherland model: Bidirectional Benjamin-Ono equation, J. Phys. A 42 (2009)
135201 [arXiv:0810.5327] [INSPIRE].
[93] G. Bonelli, A. Sciarappa, A. Tanzini and P. Vasko, Six-dimensional supersymmetric gauge
theories, quantum cohomology of instanton moduli spaces and gl(N) Quantum Intermediate
Long Wave Hydrodynamics, JHEP 07 (2014) 141 [arXiv:1403.6454] [INSPIRE].
[94] Y. Tutiya and J. Shiraishi, On some special solutions to periodic Benjamin-Ono equation
with discrete Laplacian, arXiv:0911.5005.

JHEP05(2015)095
[95] J. Shiraishi and Y. Tutiya, Periodic ILW equation with discrete Laplacian, J. Phys. A 42
(2009) 4018 [arXiv:0904.2644].
[96] J. Shiraishi and Y. Tutiya, Periodic Benjamin-Ono Equation with Discrete Laplacian and
2D-TODA Hierarchy, in New Trends in Quantum Integrable Systems, B. Feigin, M. Jimbo
and M. Okado, eds., World Scientific, Singapore (2011), pg. 357 [arXiv:1004.1455].
[97] S. Ruijsenaars, On barnes’ multiple zeta and gamma functions, Adv. Math. 156 (2000) 107.
[98] E. Friedman and S. Ruijsenaars, Shintani-Barnes zeta and gamma functions, Adv. Math.
187 (2004) 362.
[99] F. Benini, R. Eager, K. Hori and Y. Tachikawa, Elliptic genera of two-dimensional N = 2
gauge theories with rank-one gauge groups, Lett. Math. Phys. 104 (2014) 465
[arXiv:1305.0533] [INSPIRE].

– 77 –