Flipping the head of T [SU(N)]: mirror symmetry, spectral duality and monopoles

We consider T [SU(N)] and its mirror, and we argue that there are two more dual frames, which are obtained by adding flipping fields for the moment map on the Higgs and Coulomb branch. Turning on a monopole deformation in T [SU(N)], and following its effect on each dual frame, we obtain four new daughter theories dual to each other. We are then able to construct pairs of 3d spectral dual theories by performing simple operations on the four dual frames of T [SU(N)]. Engineering these 3d spectral pairs as codimension-two defect theories coupled to a trivial 5d theory, via Higgsing, we show that our 3d spectral dual theories descend from spectral duality in 5d, or fiber base duality in topological string. We provide further consistency checks about our web of dualities by matching partition functions on the squashed sphere, and in the case of spectral duality, matching exactly topological string computations with holomorphic blocks.


JHEP04(2019)138
1 Introduction and summary The study of 3d supersymmetric gauge theories and their dualities has received a great deal of attention in the last decade. Thanks to important achievements in the study of supersymmetric localisation (for a review see [1]), it has been possible to calculate exactly JHEP04(2019)138 quantities such as partition functions of N ≥ 2 theories on various 3-manifolds, and test a plethora of old and new IR dualities. Much progress stemmed from the idea of obtaining 3d dualities from 4d Seiberg-like dualities, which was revamped in [2]. 3d dualities obtained from 4d have been in turn observed to generate old and new dual pairs when subject to various types of real mass deformations. Identifying a common ancestor of various apparently unrelated 3d dualities seems a useful organizing principle to attempt charting the vast landscape of 3d dualities. For recent results in this direction see [3,4] and references therein.
In this paper we are interested in a different type of 3d dualities: we will consider mirror and spectral dualities for which we can identify a 5d ancestor, and not a 4d one.
Our starting point is the so-called T [SU(N + 1)] quiver theory of [5], depicted in the figure 1.1. This is a 3d N = 4 supersymmetry theory which can be realised on a set of D3 branes stretched between NS5 and D5 branes. The action of S-duality on Type IIB threeand five-branes can then be used to show that T [SU(N + 1)] has a mirror dual, denoted by T [SU(N + 1)] ‹ , which is described by the same T [SU(N + 1)] quiver with Higgs and Coulomb branch swapped. We will later decorate such a setup by turning on various mass deformations preserving N = 2 * supersymmetry.
Our first observation is that T [SU(N + 1)] has two more dual descriptions, which we denote by Horizontal arrows correspond to mirror dualities, while vertical arrows are new dualities which, as we will explain in more details, can be regarded as a generalization of Aharony duality [6]. We will refer to these new duality as Flip-Flip dualities. In section 2, we discuss the map between operators across the four dual frames. Particularly interesting is the way nilpotent orbits are mapped under Flip-Flip duality.

JHEP04(2019)138
In section 4 we show equality of the partition functions on S 3 b . Contrary to the case of 4d and 3d Seiberg-like dualities, where the equality of localised partition functions reduces to well studied integral identities in the mathematical literature, [7][8][9] and [10], for T [SU(N + 1)] and its dual partition functions there are no analogous results. So we follow the strategy of [11] regarding our partition functions as eigenfunctions of a set of Hamiltonians.
In section 3 we consider the effect of deforming T [SU(N + 1)] by a linear monopole superpotential as in [12,13]. Applying the monopole duality of [15], we show that this deformation has the effect of confining sequentially all the nodes but the last one. The result is a U(N ) theory with (N + 1) flavors and several gauge singlets, which we call theory A. We then follow the monopole deformation across the duality frames and obtain four new dual theories. This is the inner ABCD square shown in the picture below.  Interestingly enough, the horizontal lines in the ABCD diagram also correspond to mirror dualities, while the vertical line connecting A and D is precisely Aharony duality. Mirror symmetry relates A and B in very much the same way of [13], but the connection among theories A, D and C is more involved. In particular, the original monopole deformation on T [SU(N + 1)] translates in F F T [SU(N + 1)] ‹ into a nilpotent vev for the Higgs branch flipping fields. By studying in detail the low energy theory on such nilpotent vev we show that it corresponds to the abelian quiver C in the picture. Then we obtain the same theory by performing piece-wise mirror symmetry to theory D.
In section 5 we move on to the construction of our 3d spectral dualities. Several ingredients goes into it. Firstly, we realise our 3d theories T X as codimension-two defect theories coupled to a (trivial) 5d N = 1 bulk theory. In particular, T X is generated by the so-called Higgsing prescription [17], meaning that T X is generated by turning on appropriate vevs in 5d N = 1 linear quiver theories which is geometrically engineered by (p, q)-web of NS5 and D5' branes. This higgsed configuration can be recognize as a variation of the construction given by [16], and it involves D3-branes stretching between NS5 and D5' branes, on which our defect theories live.

JHEP04(2019)138
Secondly, we also realise our 5d linear quiver theory by the compactification of M-theory on a toric CY three-fold X, with the toric diagram given by the (p, q)-web. This leads to direct interpretation of the 5d instanton partition function as the refined topological string partition function Z X inst,top . In this language, the Higgsing prescription amounts to tuning the values of the Kähler parameters of the CY X to special quantized values in order to obtain the "Higgsed CY" X , and typically it reduces the instanton partition function to an instanton-vortex partition function for the coupled 3d-5d system [18][19][20][21][22][23].
In the cases we are going to consider we have a "complete Higgsing", by which we mean that starting from 5d we are left with a 3d theory coupled to a trivial 5d theory of free hypers. In this way we can unambiguously identify the Higgsed instantonvortex/topological string partition function as the vortex partition function of the defect theory Z α 0 vort,T X = Z X vort,top . More precisely, we can relate the D 2 × S 1 partition function of theory T X evaluated in a certain vacuum α 0 , i.e. the holomorphic block B α 0 T X , to the partition function of the Higgsed topological string: 1 Our 5d linear quiver theories admits a spectral dual description which can be equivalently stated as the invariance of the topological string partition function under a fiber-base transformation: Z X top = Z X top , where we denote by X the fiber-base dual CY. Of course X and X are just two equivalent description of the same toric CY, hence the equality of the partition functions. Therefore, our main idea is to combine Higgsing and fiber-base duality to obtain new 3d spectral pairs which descend from 5d. Summarizing, we first follow the Higgsing process on X down to X , which yields a 3d theory T X . Then we follow this same Higgsing on the fiber-base dual, X down to X , and we obtain another 3d theory T X which we call the 3d spectral dual theory. The fiber-base invariance of the Higgsed topological string partition function implies the equality of the holomorphic blocks B α 0 T X = B α 0 T X of the 3d spectral dual theories T X and T X .
In this paper we propose two examples of spectral dual theories and together with the topological string construction, we support our proposals by providing purely field theory arguments. A third example of spectral duality has been recently discussed in [24].
The first spectral dual pair we construct follows from the duality on the SE-NW di- Notice that F T [SU(N + 1)] has been realised previously in [25] as a defect theory in the square (p, q)-web with (N + 1) D5' and NS5 branes, and there it was also shown that the equality of the holomorphic blocks B ‹ follows indeed from the equality of the topological string partition functions for the fiber-base dual diagrams.
The second spectral dual pair T ↔ T is obtained within the ABCD diagram by flipping the SE-NW diagonal D ↔ B. We discuss the operator map and check the equality JHEP04(2019)138 of the sphere partition functions. We then show how T and T , can be realised as defect theories inside spectral dual 5d theories and obtain their holomorphic blocks B α 0 T , B α 0 T by tuning the Kähler parameters in the fiber-base dual CYs. Again we prove that our 3d spectral pair descends from fiber-base duality in topological strings.
The main novelty of our construction is to provide two completely independent and quantitative tools to check spectral dualities. Indeed, it is quite remarkable that both a field theory computation based on the localised supersymmetric partition function, and the refined vertex of the topological string, exactly agree.
We add an intriguing outlook to our story: both the inner diagram ABCD, and the spectral dual pairs we have constructed, descend, by various flips and deformations, from the self-mirror tail of T [SU(N + 1)], which can be realised with a brane construction. It would be interesting to understand better the interplay between these two types of dualities in string theory. For example, a simple generalization we might consider is to study other Higgsing patterns in the toric diagrams, corresponding to more general nilpotent vevs for T [SU(N +1)], or even more interestingly, corresponding to coupled 3d−5d systems in which the 5d theory is non trivial. As in the examples we have proposed, performing fiber-base duality on a generic Higgsing pattern will produce a new duality for the 3d-5d system. We also point out that our analysis of the nilpotent vev is reminiscent of the discussion about T-branes and 3d N = 2 quiver theories [14].
In this paper we have focused on spectral duality, or fiber-base duality, which is just one element of the S-duality group of the (p, q)-web. It would be interesting to investigate the interplay between Higgsing and the action of the other elements. Some investigations along these lines have been proposed in [24].
Each one of the N round gauge nodes, labelled by its rank k = 1, . . . , N , is associated to a vector multiplet decomposed into an N = 2 vector multiplet and an adjoint chiral field Φ k , represented by a loop. Bifundamental chiral fields Q ab , and antichiral fieldsQãb are represented by lines connecting adjacent nodes and pair up into hypermultiplets 2 . The N + 1 rectangular node is ungauged. In the quiver rapresentation, the flavor node is what we call 'head' of T [SU(N + 1)]. In N = 2 notation the superpotential of the theory is In our conventions the bifundamentals Q (k,k+1) ab transform in the reps ⊗ of U(k) × U(k + 1), and the bifundamentalQ (k,k+1) ab transform in the reps ⊗ of U(k + 1) × U(k)

JHEP04(2019)138
where we defined the matrix of bifundamentals Q (L,R) = Q (L,R) ab Q (L,R) ab , labelled by the pair (L, R) attached to the link between a left (L) and a right (R) node. On the first node Q (0,1) = 0. Traces Tr k are taken in the adjoint of U(k).
The global symmetry of T [SU(N + 1)] is SU(N + 1) flavor × SU(N + 1) top . The flavor symmetry SU(N + 1) flavor rotates the fundamental hypers at the head of the tail. The non abelian SU(N + 1) top is an IR symmetry, and the UV Lagrangian only manifests a U(1) N top topological symmetry, coming from the dual photons on the gauge nodes. For each Cartan in the flavor symmetry group and each U(1) top we can turn on a real masses, M p and T p , respectively.
The R-symmetry of a 3d N = 4 theory is SU(2) C × SU(2) H with Cartans U(1) C ⊂ SU(2) C and U(1) H ⊂ SU(2) H . We will work with a family of N = 2 * theories obtained by introducing a real mass parameter for the anti-diagonal combination U(1) A = C − H [26]. We take the UV R-charge equal to the combination R 0 = C + H. In the IR, the Rsymmetry can mix with other abelian symmetries, but since the topological symmetry is non-abelian, R 0 will only mix with U(1) A . Thus we introduce a trial R-charge, defined by For the bifundamental fields we find R = 1−α 2 ≡ r, in agreement with the assignment C = 0, H = 1 2 . For the adjoint fields R[Φ k ] = 2(1 − r) iff the superpotential has R-charge 2. Notice also that R[Φ k ] = 1 + α = 2(1 − r) is consistent with C = 1 and H = 0. The exact value of r can be fixed by F-extremization [27].
We define the gauge invariant (N + 1) × (N + 1) meson matrix: The dynamics might impose additional relations on Q ij , thus restricting the set of generators of the Higgs branch (HB) chiral ring. Classical relations follow from the F-terms, and for T [SU(N + 1)] the F-terms of the fields Φ k imply that Q is nilpotent [5]. The argument goes as follows: Q = Q (N,N +1) Q (N,N +1) has rank at most N by definition. Then, the F-term of Φ N can be used to rewrite which implies Q 2 has at most rank N − 1. Iterating this computation we find that certainly Q N +1 = 0. The Higgs branch is related to the nilpotent cone N for matrices in SL(N +1, C). This space can be organized as the union of all the orbits S ·J λ ·S −1 where S ∈ SL(N +1, C) and J is the Jordan form associated to a partition λ of N + 1, see [28] for a review on related topics. The meson Q comes along with the moment map operator Π Q , which is better suited to describe global symmetries of the theory. Indeed, Π Q is the half-BPS primary in a supermultiplet which contains conserved global currents. In our case, Π Q is defined as Coulomb branch (CB) operators can be obtained from Tr Φ k and monopole operators M f 1 ...f N carrying f i units of flux for the topological U(1) on the i-th node. The R-charge JHEP04(2019)138 of a (BPS) monopole operator is determined by the R-charges of all the fermions ψ of the theory by the formula: where ρ ψ (f 1 , . . . f N ) is the monopole charge of ψ [5,29,30]. 3 We find that monopole operators defined by a string of fluxes of the form [0 n (±1) m 0 p ], where 0 and 1 are repeated with integer multiplicities n, m, and p constrained by n + m + p = N , have the same Rcharge of the adjoint fields, i.e. R[Φ k ] = 2(1 − r). These monopole operators are N (N + 1) and together with the Φ k=1,...N can be arranged into a (N + 1) × (N + 1) matrix, analogous to the meson matrix. For N = 3 this matrix reads where D i are traceless diagonal generators of SU(N + 1) top . The generators of the CB chiral ring can be obtained from such an M ij upon imposing further relations.
In the rest of the paper we will refer to a matrix assembled as in (2.8), as the monopole matrix of the theory under consideration.
The moment map Π Q and the monopole matrix belong to the adjoint of SU(N + 1).

Mirror simmetry
It is well known that T [SU(N +1)] is self-dual under mirror symmetry [5]. The dual theory, hereafter T [SU(N + 1)] ‹ , has quiver diagram 2 3 1 N N + 1 (2.9) and the same field content as T [SU(N + 1)]. In T [SU(N + 1)] ‹ we denote the adjoint chirals by Ω k , the monopoles operators by N f 1 ...f N , and the bifundamental fields by P ab and Pãb. The indexes k and f 1 . . . f N have the same meaning as in T [SU(N + 1)]. We introduce the matrix P (L,R) = P (L,R) abP (L,R) ab for each pair of nodes (L, R). Then the dual the superpotential reads (2.10) Mirror symmetry exchanges the Higgs and Coulomb branch. Therefore the bifundamental fields have now R-charge R[P ab ] = 1 − r. Consequently the monopole operators have R-charge R[N ij ] = 2r. It follows from the superpotential that R[Ω k ] = 2r for any k.  On the Higgs branch we define the meson P and its moment map Π P . The meson is On the Coulomb branch we consider the monopole matrix N ij , which similarly to the previous section, is obtained from TrΩ k and from monopole operators with fluxes valued in [0 n 1 m 0 p ]. For N = 3 we have where again D i are traceless diagonal generators of SU(N + 1). Mirror symmetry exchanges and therefore HB and CB.
The fact that T [SU(N + 1) is self-dual under mirror symmetry can be neatly derived from the IIB brane engineering of the T [SU(N + 1)]. More precisely, T [SU(N + 1)] can be understood as the low energy theory of a system of D3 branes suspended between (N + 1) D5 and NS5 branes [32]. The brane configuration is summarized in table 1 and goes as follows. The NS5 extend along directions 012789, and the D5 branes along directions 012456. The NS5 and D5 branes are separated in the third direction, where (N + 1) D3 branes are stretched in between, so that each NS5 brane is connected to a distinct D5 brane. These D3 branes extend along directions 0123, but since they are bounded in the third direction by D5 and NS5 branes, the low energy dynamics on their wordlvolume is threedimensional. In fact it is precisely the T [SU(N +1)] theory. The R-symmetry group factors SU(2) C and SU(2) H correspond to the rotation symmetry of the NS5 and D5 branes in the directions transverse to the D3 branes, i.e. to SO(3) 456 and SO(3) 789 respectively. The action of mirror symmetry, which exchanges CB and HB, is precisely that of IIB S-duality, which exchanges the NS5 and D5 branes (leaving the system invariant). Equivalently, one can think of this transformation as the exchange of the 456 and 789 directions.

Flip-flip duals
In this section we propose new duals for T [SU(N + 1)] and its mirror. We name them Flip-Flip dualities for reasons that will become soon clear.

JHEP04(2019)138
Let us begin by describing the Flip-Flip dual of T [SU(N + 1)], which we denote by F F T [SU(N + 1)]. This theory has the content of T [SU(N + 1)] plus two extra sets of fields, the flipping fields. We represent F F T [SU(N + 1)] by the quiver where the horizontal loops attached on the flavor node indicate the addition of flipping fields. We have adjoint chiral fields Θ k , bifundamental fields R ab and Rãb, with R[R ab ] = R[ Rãb] = 1 − r, and monopoles operators. Out of them, we define the meson R ij , its moment map Π R ij , and the monopole matrix m ij . The flipping fields are elementary fields, singlet of gauge groups, and transform respectively in the adjoint of SU(N + 1) flavor and SU(N + 1) top. , in particular they are traceless. These are denoted by F R ij and F m ij , since they will couple to R ij and m ij in the superpotential: We then discover that the Flip-Flip duality between T [SU(N + 1)] and F F T [SU(N + 1)] maps: The F-terms of F R ij and F m ij imply Π R ij = 0 and m ij = 0. As a result, the HB and CB will now be described by F R ij and F m ij , respectively. In this sense, F R ij is the flip of the moment map, i.e. the meson R ij , and F m ij that of the monopole matrix. It is interesting to look at the description of the HB. The F-terms of the bifundamentals R (k,k+1) and R (k,k+1) imply the equations where we defined Θ N +1 ≡ F R so to have a uniform notation in (2.17) and (2.18). The F-terms of the diagonal component of the monopole matrix give TrΘ k = 0 for all k ≤ N , in particular Θ 1 = 0. Furthermore, it is always possible to use SU(N ) gauge × SU(N + 1) flavor to put one of the bifundamentals on the last link in a diagonal form. For concreteness we take, with arbitrary v i . The constraint Π R ij = 0 trivializes (2.18). Let us discuss the case N = 1 to start with. From (2.17) we find

JHEP04(2019)138
therefore R (1,2) is in the kernel of Θ 2 . The flipping fields Θ 2 are in the adjoint of SU(2) and we shall take TrΘ 2 = 0. It follows that a traceless matrix in the adjoint with a one dimensional kernel can be put into the form, i.e. Θ 2 is nilpotent. More in general, we can use a recursive argument to show that Θ N +1 can be taken to be nilpotent. So let us assume that Θ N is nilpotent, and consider the matrix R This matrix can be explicitly constructed in the gauge (2.19). Then (2.17) becomes Considering Θ N is in its Jordan Form we introduce a basis { w i } such that: for given coefficients c j i which depend on the partition associated to Θ N as nilpotent matrix. Equations (2.22) now imply the relations Since w i=1,...N is a basis, the span of u i = R (N,N +1) w i is by construction an N -dimensional subspace in N + 1 dimensions. The solution of (2.24) is then where K parametrizes the one dimensional kernel of R (N,N +1) aux , and the coefficients θ i are arbitrary. It is straightforward to plug (2.25) back into (2.24) and check that the equations are satisfied by using R (N,N +1) aux R (N,N +1) = I N ×N . Moreover, this relation implies that K and the set of vectors { u i } are independent, and therefore we can use them to span a basis in N + 1 dimensions. Thus we fix completely Θ N +1 by specifying its action on such an basis, i.e. by adding Θ N +1 K to the list in (2.25). The nilpotent solution is given by where the θ j play a role analogous to θ 2 appearing in (2.21).

JHEP04(2019)138
The outcome of our computation is interesting for two reasons: on one hand we obtained nilpotent solutions for the vev of the flipping fields, which in turn supports our duality, i.e. our identification Π Q ↔ F R . On the other hand, this nilpotency condition on the flipping fields shows up in a totally opposite way compared to the case of the meson Q in T [SU(N + 1)]. As reviewed in the Introduction, in order to show that Q is nilpotent in T [SU(N + 1)] we used the F -term constraints starting from the head of the tail back to the first gauge node. For the flipping fields, instead, we used F -term constraints recursively from the first gauge node up to the head of F F T [SU(N + 1)]. We will have more to say about this in section 3.3.
In the case of N = 1, our duality relates T [SU (2)] to a U(1) theory with two flavors and various singlets. This case can be understood as a version of the Aharony duality, as we now argue. Let us recall that Aharony duality maps N = 2 SQED theory with two electric flavors (Q i ,Q j ) and no superpotential, to an abelian theory with two magnetic flavors (q i ,q j ), and extra singlets M ij and S ± . The magnetic superpotential is non trivial: In our language, M ij and S ± are "flipping" fields for the magnetic mesons q iqj , and for the dual monopoles V ± , respectively. Notice that M ij belongs to the adjoint of U(2), so it is not yet our flipping field. In order to get T [SU (2)] out of the electric side of Aharony duality, we introduce an extra singlet φ, and we add a cubic superpotential of the form W el. = φ i Q iQi . This is indeed the tail superpotential for T [SU (2)]. Adding a corresponding singlet field φ also on the magneric side, Aharony duality maps our deformation to the mass term φ Tr(M ). Integrating out these two fields in the full magnetic superpotential, we obtain: At this point the meson q iqj can be replaced by its moment map without changing W magn. . Then, W magn. will be precisely what turns out to be the superpotential of F F T [SU (2)]. The expression of W F F T in this case is, where θF m 3 is the coupling due to the σ 3 generator in (2.8). Similarly for m ± F m ± . Integrating out F m 3 , we recover W magn. upon a trivial field redefinition. The Flip-Flip duality on the mirror side works in a similar fashion: the starting point is T [SU(N + 1)] ‹ and its quiver diagram (2.9). The quiver diagram of F F T [SU(N + 1)] ‹ is essentially (2.9), except for the flavor node on which the new flipping fields are attached. On F F T [SU(N +1)] ‹ we will use the following notation: Ψ k for the adjoint chirals, S ab and Sãb for the bifundamental fields, with R[S ab ] = R[ Sãb] = r, S ij for the mesons and n ij for the monopole matrix. The flipping fields are denoted by F S ij and F n ij . The superpotential is

A commutative diagram
We can represent our four dualities through the following commutative diagram: Horizontal arrows connect mirror dual theories while vertical arrows connect flip-flip dual theories.
We  [27]. Indeed, when we extremize the partition functions we set to zero the fugacities for the non-abelian symmetries, since these can't mix with the R-charge. But, as we will see later, if we turn-off the non-abelian fugacities, the contribution of the two sets of flipping fields cancel-out, hence the extremal R-charges for F F T [SU(N + 1)] and T [SU(N +1)] are the both equal to 1/2, which is the N = 4 value.
In section 5.1 we will argue that the F F T [SU(N + 1)] ‹ theory can be realised on a brane set-up consisting of D3 branes suspended between NS5' and D5' branes preserving N = 4 supersymmetry.

Deformations of the commutative diagram
In this section we consider a certain monopole deformation of T [SU(N + 1)] and follow its RG-flow across the commutative diagram. This computation offers an interesting and novel consistency check about the (mother) T [SU(N + 1)] commutative diagram, and produces JHEP04(2019)138 another set of dual theories, named ABCD, themselves organized as a (daughter) commutative diagram. The final picture is presented in section 3.4 and summarized as follows: The monopole deformation we are interested in turns on the following components of the monopole matrix M ij , The last gauge node is underformed. 4 We denote the deformed superpotential in T [SU(N + 1)] by W T def , namely More generally we will define W T ‹ def and W F F T ‹ def for theories B and C, respectively.

Theory A: monopole deformed T [SU(N + 1)]
The quiver diagram for T [SU(N + 1)] was introduced in section 2, It will be convenient to decompose the adjoint fields on a basis of hermitian generators of U(k), namely Φ k = φ a k T a , and extract from the superpotential (2.2), the abelian components, defined hereafter as, The reason is that abelian and non-abelian components decouple. 5 In the presence of the monopole deformation L where the magnetic fields M ij replace the electric meson. This is the first instance of a family of electric-magnetic dualities introduced in [15]: The map (3.7) does not include adjoint fields, which instead are present on the quiver tail. However, on the first node, Φ 1 is just a singlet, thus it can be taken into account afterwards. Similarly, we add the coupling γ det M on top of W T . Since the magnetic dual of a U(1) gauge theory with two flavors is a Wess-Zumino model, the U(1) dynamics has confined in the IR. The presence of W T allows for a sequence of iterations of this procedure, where at each step the duality (3.7) is used for an increasing value of N c . This is the content of the sequential confinement introduced by [12]. Here we generalize it to the case of T [SU(N + 1)], building on previous work done in [13]. Before presenting results for the final low energy theory, we discuss in detail the confinement of the first two nodes.
Move # 1. Consider the restriction of T [SU(N + 1)] to the first and the second gauge node. Locally, the theory is described by the quiver with superpotential where in (3.9) we have specified the abelian component. Notice the property Tr 1 Tr 2 Q (1,2) = Tr 2 Tr 1 Q (1,2) , i.e. we can commute the two traces. We use the monopole duality (3.6) on the first gauge node. Accordingly, we replace the electric meson, Tr 1 Q (1,2) → M 2 , where M 2 is in the adjoint of U(2), the second gauge node. In the dual theory the superpotential has become: (3.10) where the abelian superpotential reads, The interaction term, γ 2 det M 2 , is part of the duality map. It is convenient to rotate the abelian adjoints to

JHEP04(2019)138
in such a way that An important remark is that M 2 is an elementary field in the dual theory. Then the Fterm of ϕ − 2 and φ a=1,2,3 , determine a vev for M 2 . In particular, M 2 depends on Tr 3 Q (2,3) as follows Equations (3.14) imply that M 2 has the same non abelian components of Tr 3 Q (2,3) but differ by a factor of 1 2 in the abelian component. In matrix form, the solution is Expanding the superpotential around M 2 = M 2 + δM 2 , we find mass terms for ϕ − 2 , δM 2 , and for the non abelian adjoint fields φ a . This is obvious from (3.10) and (3.13). Below a common mass scale, all these fields can be integrated out. As a result, the second node has now only a light U(1) adjoint scalar ϕ + 2 , and the bifundamentals on the (2, 3) link. On the vacuum M 2 there is a novel effective superpotential, which we determine in the next paragraph.
To proceed further, we would like to express det M 2 in terms of traces over matrices in the adjoint of U(3). The reason is that a matrix in the adjoint of U(3) plays the role of the meson matrix for T [SU (3)]. Thinking about iterating the duality (3.7) on node (2), this rewriting is clearly necessary. To achieve the desired result, we first expand Then, we rewrite Finally, two additional manipulations: in the abelian case we interchange the traces in the obvious way, Tr 2 Tr 3 Q (2,3) = Tr 3 Tr 2 Q (2,3) . In the non-abelian case, we notice the property

JHEP04(2019)138
The resulting theory has the following quiver diagram, where the blue loop stands now for ϕ + 2 , instead of the full adjoint, and we remind ourselves of γ 2 by displaying it on the l.h.s. of the diagram. The effective superpotential associated to (3.18) is, Had we chosen N = 2, there would be no monopole deformation in (3.19). Renaming Tr 2 Q (2,3) as the meson matrix Q introduced in section 2, this would be the final result.
Move # 2. We glue (3.18) back to T [SU(N + 1)], and move forward. On nodes (2) and (3), the theory is now described by the modified quiver The superpotential includes the terms The gauged matter content attached at node (2) is again of the form (3.7), plus singlets. We dualize by replacing Tr 2 Q (2,3) → M 3 and add the superpotential term γ 3 det M 3 . As before we study abelian and non abelian contributions separately. In the abelian sector we find, which upon performing the rotation Very much as in move #1, the F-terms of φ a=1,...8 3 and ϕ − 3 imply that M 3 has the same non abelian components of Tr 4 Q (3,4) but differs in the trace. The solution for M 3 is

3
, and fluctuations of δM 3 , we obtain a low energy theory with light ϕ + 3 and bifundamentals Q (3,4) , As in the previous case, we would like to express the superpotential couplings which are linear in γ 2 and γ 3 , in terms of traces over matrices in the adjoint of U(4). When det M 3 is expanded out in Tr 3 , both γ 2 and γ 3 terms can be rearranged by using the following formulas, For the couplings to γ 2 we obtain and for the couplings to γ 3 In both cases, the final results can be expressed in terms of polynomials p 3,3 and p 2,3 in the variable Tr 3 Q (3,4) , which is indeed in the adjoint of U(4). Collecting these contributions, the effective superpotential is determined by where . . . stands for the remaining monopole superpotential L {3,...,N −1} .
Duality moves: from # 1 up to # N − 1. Repeating the reasoning in move #1, and #2, we proceed up to #N − 1. The final gauge theory, which we refer to as theory A, has quiver diagram

JHEP04(2019)138
and superpotential where Tr N Q (N,N +1) = Q is the meson matrix, and p m,n are polynomials generalizing (3.29) and (3.30) at each step. Recall that we did not turn on the monopole superpotential on the last gauge node, therefore (3.33) is the final result. Let us summarize the sequential confinement up to #N − 1. In the order: 1) After each dualization, labelled hereafter by k − 1, we derived an equation for M k which we solved explicitly. In each case, the non abelian components of M k are fixed by the F-terms of φ a=1,...k 2 −1 to be equal to Tr k+1 Q (k,k+1) . As in move #1, and #2, the abelian equation turns out to be always: The solution for M k is 2) Having found the solution (3.35), we integrate the massive fields at node k and we write the superpotential for the light fields. These are Tr k Q (k,k+1) , ϕ (k) + and the collection of {γ m } k m=2 . This step is the most involved, since it requires rearranging the expression of {det m } k m=2 in terms of traces. The final result is packaged into the polynomials p m,k . The structure of traces of such polynomials is fixed, i.e. by construction it coincides with that of det m in its Laplace expansion. In particular, In particular, the first three recursions determine p 2,k and p 3,k for any k.
Final remarks. It is important to emphasize some features of the superpotential W A . A gauge theory U(N ) with N + 1 flavors and no superpotential would have flavor symmetry SU(N + 1) flavor × SU(N + 1) flavor . This is reduced to a single SU(N + 1) flavor because of the superpotential. Even in the absence of γ m contributions, the presence of ϕ + N TrQ guarantees the correct amount of flavor symmetry. In this respect, ϕ + N plays a distinguished role. Since the superpotential has R-charge 2, the R-charges of the singlets γ m acquire a dependence on m, The F-terms of ϕ + N and γ m=2,...N imply sequentially that TrQ 1≤k≤N = 0. Then, from the Cayley-Hamilton theorem it also follows TrQ N +1 = 0. 7 This set of conditions is in fact equivalent to the statement that Q is nilpotent. At this point it is useful to redefine ϕ + N = N !γ 1 and simplify W A by invoking chiral stability arguments [31]. This amounts to drop terms containing TrQ. The final form of the superpotential is then In our discussion there will be no difference between these two versions of W A . However, we should note that this prescription amounts to drop multi-trace contributions to the effective superpotentials, which might affect other details of the theory.

Theory B: monopole deformations on the mirror
In this section we follow the monopole deformation in the mirror frame T [SU(N + 1)] ‹ , which is represented by the quiver diagram below, Our notation in section 2.1 used bifundamentals P andP on each link, and adjoints Ω on each gauge node. The monopole deformation L T {1,...,N −1} we considered in (3.2) can

JHEP04(2019)138
actually be written, more suggestively, in terms of the Jordan matrix where J k is a single Jordan block of size k and zero eigenvalue. It follows that L T is mirror to a nilpotent mass deformation for the meson P ij . By introducing the vectors and the total superpotential is thus The discussion next will closely follow [13]. The F-term equations ofp a=2,...,N and p b=1,...,N −1 are non trivial due to the mass deformation. Let us begin from the F-terms of the fieldsp a=2,...,N , which read The solution is expressed in terms of p N as follows: Equivalently, the F-terms of the fields p b=1,...,N −1 are solved by On the vacuum p b → p b + δp b , andp a → p a + δp a , the pair of field (p 1 , p N ), and (p N +1 ,p N +1 ) do not get a mass terms from the deformation (3.43), thus they remain in the low energy spectrum. 8 The effective superpotential for these light fields is

JHEP04(2019)138
The low energy theory is then described by the quiver, where we have isolated the fields (d,d) ≡ (p 1 ,p N ) on the bottom of the diagram. Compared to the tail, these fields are the ones with a special superpotential interaction. The matrix N ) is now truncated to a one flavor component p N +1pN +1 , and the total superpotential of theory B is We then have the following map between the singlets of theory A and dressed mesons of theory B: The duality between theories A and B is a particular case of the SQCD mirror dual discussed in [13] with a minor difference, i.e. we have kept the fields γ i on the side of theory A.

Theory C: nilpotent Higgsing from monopoles
The theory F F T [SU(N + 1)] ‹ introduced section 2.2 is described by the quiver where we denoted the bifundamentals on each link by S and S, the adjoint chiral on each gauge node by Ψ, and finally the flipping fields by F n ij and F S ij . The deformation L The total superpotential is then It will convenient to momentarily modify our notation, and denote F S ij by Ψ N +1 . Then S ij F S ij fits with the pattern of the N = 4 superpotential, and will allow us to display some recursions in a neat way.
We first consider the F-term of Ψ N +1 , which set This equation shows that the bifundamentals on the last link of the tail acquire a non trivial vev. 9 By definition, our bifundamentals are rectangular matrices. However, it is convenient to describe the vev in terms of square matrices where we specify which column/row has to be dropped. In this way, the solution of (3.57) is Up to gauge and flavor rotations, the natural strategy to solve an equation of the form (3.57) is to take S equal to the nilpotent vev on the r.h.s, and S such that the equation is satisfied. In particular, S has an identity block of rank N − 1 [34]. The solution when N = 1 reduces to zero, since there is no monopole potential in this case.
The F -terms of the fields Ψ k≤N have two types of contributions. One is coming from the superpontential of the tail, i.e. W T [Ψ, S], and a second one originates from the coupling n ij F n ij : indeed, recall from the definition (2.8) that the monopole matrix has traceless diagonal components of the form TrΨ k D k . We will study a vacuum for which F n ij = 0. Therefore the F -term of Ψ k will be Reading (3.59) from right to left, we conclude that the nilpotent vev (3.58) propagates along the quiver, from the last node towards the left. The solution of this recursion is drop the first row (3.60) Note that for k > 2 the vev S (k−1,k) is always next to a maximal Jordan matrix J k . At the terminating value k = 2 both vevs vanish. These correspond to the bifundamentals on the first link (1, 2).

JHEP04(2019)138
Nilpotent vevs. A supersymmetric nilpotent vev should satisfy the F-terms of the bifundamentals, and finally D-terms. In matrix notation, the F -terms of the bifundamentals are These equations put constraints on the fields Ψ k , and before proceeding, let us recall that an additional constraint comes from the F-terms equations for the diagonal generators of F n ij , which imply the condition TrΨ k = 0 for any k ≤ N . A trivial solution to (3.62) is Ψ k = 0 for any k. However, this solution will not be consistent with vanishing of D-terms, as we now explain.
Notice that equations (3.61) and (3.62) don't fix a solution, rather they impose a constraint on Ψ k+1 which depends on S (k,k+1) , S (k,k+1) and Ψ k . In this new recursion, the starting point is the beginning of the tail, i.e. the U(1) gauge node, and the first link (1, 2). Consistency of this recursion requires that the solution in the case of T [SU(k)] uplifits to T [SU(N + 1)] for any k ≤ N . The study of the first few cases will be enough to understand the nilpotent vev in the adjoint sector.
Consider T [SU (2)]. The D-term on the U(1) gauge node is A short computation shows that the terms labelled by 'R' cancel each other. Since S (1,2) = S (1,2) = 0, the terms labelled by 'L' do not contribute. The case of T [SU (3)] is again special and the solution Ψ 2 = 0, ξ 2 = 0 is consistent. In particular T [SU (2)] uplifits to T [SU (3)]. Finally equations (3.61) and (3.62) imply the relations These equations do not fix all the components of Ψ 3 . The trivial solution is possible, but we claim that the actual solution, compatible with T [SU(4)], is a nilpotent vev for Ψ 3 , i.e.
Again, we are back to the nilpotent vev for the flipping fields we started with.

JHEP04(2019)138
The lesson from the previous case is the following: by moving forward to the right of a longer quiver tail we will have to deal with D-term equations of the form for any R = (k, k + 1) and L = (k − 1, k). We show in appendix A that terms labelled by 'R' always cancel each other. On the other hand, terms labelled by 'L' do not, and give a non zero commutator [Ψ k , Ψ † k ]. The solution of F-and D-terms induced by the next-to-maximal nilpotent vev (3.57) is: It is important to point out that D-terms equations are automatically solved by the SU (2) relation, which follows from the construction of the nilpotent vev. In our case the embedding ρ k of the σ 3 element is In this solution, the FI parameters ξ k are zero for any k.
The list of scalar vevs includes the real scalars in the vector multiplets, which do not play any role, i.e. σ k = 0.
The low energy theory. Given the nilpotent vevs found in the previous section, we can explicitly study the Higgs mechanism and obtain the massless field content.
Let us begin from the gauge sector. It is useful to recall that a generic gauge transformation on the quiver acts on the matter fields in the following fashion: where G k is the action restricted to a single gauge node U(k). Taking the connection we will find where we listed all the different matter representations. A broken generator does not leave the vev invariant, therefore T a R z = 0. Here z stands (at least) for one among all the fields of the tail and the various representation have been indicated by R. Unbroken generators annhilate the vevs. The determination of unbroken generators is equivalent to the study of the kernel of the mass matrix obtained from the expansion of the covariant derivatives. More details on such a matrix are collected in the appendix A. In conclusion, fixing a basis of T a we find a solution for the coefficients JHEP04(2019)138 g a k , which corresponds to a single unbroken generator A k for gauge group U(k). Its explicit form is very simple, As far as the gauge groups are concerned, the quiver (3.54) is Higgsed to The next task will be to deduce the massless matter content by the studying the kernel of the mass matrix for all the scalar fields. We focus on the chiral multiplets. 10 The mass matrix is hermitian and admits an eigenvector decomposition, which we split into ker ⊕ ker ⊥ , where the latter describes massive fields. The massless sector will contain bifundamentals charged under (L, R) gauge groups, fundamentals and anti-fundamentals charged under a single gauge group, other massless neutral fields, and finally Goldstone bosons. The 'physical' massless fields of the IR theory correspond to those vectors in ker which cannot be written only as linear combination of Goldstone bosons. On the other hand, a physical configuration might still have components along the directions parametrized by the Goldstone bosons.
The deformation L F F T ‹ breaks explicitly the non abelian flavor symmetry, therefore all the Goldstone bosons we will find correspond only to the action of broken gauge generators on the nilpotent vev. In the field variables S, S, and Ψ, these Goldstone bosons are described by independent field configurations, of the form, for parameters {g a=1,...k 2 −1 } N k=1 corresponding to broken gauge generators. We computed the mass matrix generated from the superpotential. Holomophy implies the existence of N k=2 (k 2 − 1) = 1 6 (N − 1)N (2N + 5) complex Goldstone bosons. The resulting ker can be quite cumbersome at first, but the physical massless fields can be brought to a simple form by taking linear combinations with Goldstone bosons, i.e. setting to zero unwanted components. We checked all our computations with computer algebra up to N = 6. After all this work is done, we find that most of the final answer can presented in a more intuitive way. This is the case for charged fields, as we now argue. IR flipping fields will have instead a more complicated description.
Let us begin from bifundamental fields in the abelian quiver (3.74), i.e. fields simultaneously charged under a left and right gauge node. Considering the UV S (k,k+1) , we want 10 Real scalar fields {σ 1 . . . σ N } in the vector multiples have the same mass matrix as the gauge fields.
This follows from unbroken susy and it is obvious from a 4d perspective. In the 3d Lagrangian is manifests in matter couplings of the type z † σ 2 R z. In particular, on the vacuum σ k = 0 there are no non trivial off-diagonal mass terms with chiral multiplets.

JHEP04(2019)138
to select those components which transform non trivially under A k and A k+1 , where the gauge field is given explicitly in (3.73). It is simple to see that the first row of S (k,k+1) transforms non trivially under A k , while the first column transforms non trivially under A k+1 . For k = 1, . . . N − 1 the low energy bifundamentals, s (k,k+1) ands (k,k+1) , embedded into S (k,k+1) and S (k,k+1) , are indeed in the (1, 1) entry, The nilpotent vev ( The IR quiver theory until now is described by the diagram We move on to the study of neutral fields. The simplest case is represented by singlet fields on each gauge node, which we denote by ψ k . These are given by the following embedding: Ψ k = diag(ψ k , 0 k−1 ), k = 1, . . . N , as it could have been anticipated. The analysis of IR massless fields originating from the flipping fields Ψ N +1 is less straighforward, and there is a novelty: it is not possible to localize such fields on components of Ψ N +1 , but the corresponding vectors in ker will have components on both Ψ N +1 and Ψ i≤N . Furthermore, the latter cannot be eaten up by taking linear combinations with Goldstone bosons.
We begin by assuming Ψ N +1 ∈ U(N + 1) for simplicity, and we will obtain the case Ψ N +1 ∈ SU(N + 1), which is of interest for our duality web, by a minor modification. This procedure is instructive since it will have a counterpart in the next section.
Let us introduce first the IR fields Γ i=2,...N . In terms of components of Ψ N +1 , we find

JHEP04(2019)138
where the rewriting on the r.h.s. makes manifest that these fields parametrize nilpotent directions inside Ψ N +1 . The configuration (3.78) extends on the UV adjoint fields Ψ i≤N , as follows Matrices are multiplied a number of times defined by the upper index, i.e. J # N −k = J N −k · · · J N −k # times. For example, notice that Γ 2 extends backwards to Ψ 3 , Γ 3 extends backwards to Ψ 4 , and so on. (Ψ 1 and Ψ 2 vanish in this case.) On a similar footing we find the field ∆, which has the following UV embedding, Finally, there are three other fields, Σ ± and δ. These ones are localized on specific compo- The IR flipping fields presented so far have been obtained by refining the output of ker. We now explain how to see explicitly that these fields do not get a mass term from the superpotential. For each link (L, R) of the quiver, consider the mass terms coming from fluctuations δΨ L , δΨ R and δS (L,R) , on top of the nilpotent vev. We find We have to show that when looking at the components of δΨ k≤N +1 parametrized by Γ i=2,... or ∆, each term in (3.82) vanishes for any (L, R). Indeed, because of the form of the nilpotent vev (3.60), the equation are solved precisely by (3.78)-(3.79) and (3.80). Moreover, on the link (N, N + 1), this same computation shows that the directions parametrized by Σ ± and δ are also massless, since when we multiply by the vev, these matrix elements are shifted either to the right, or to the bottom, by two units, i.e. they disappear from the equations. The case of interest, Ψ N +1 ∈ SU(N + 1), is quite simple to deduce. Indeed only the IR fields ∆ and δ parametrize directions which overlap with the identity. Therefore, out of JHEP04(2019)138 these two, we should consider the traceless combination and drop the other. We associate to such a combination the IR field Γ 1 , whose UV embedding is Collecting all the fields, the final low energy theory, which we denote by Theory C, is described by the abelian quiver A simple counting shows that we have determined 4(N + 1) chiral fields.

The ABCD of monopole deformed T [SU(N + 1)]
In this section we define theory D as the Aharony dual of theory A, and we show that the mirror of theory D is precisely theory C.
Let us remind that theory A is U(N ) SQCD with N + 1 flavors coupled to additional singlets γ m through the superpotential We apply Aharony duality [6] to Theory A and obtain Theory D, which is a U(1) gauge theory with N + 1 flavors U i andŨ j , flipping fields F U ij for the meson U ij = U iŨj , and flipping

JHEP04(2019)138
fields σ ± for the U(1) monopoles m ± . (For simplicity we borrow from the F F T [SU(N + 1)] the notation for the monopoles). In addition we denote by θ m the dual of the singlets γ m . The quiver diagram is The flipping fields F U ij of theory D are dual to the electric meson of Theory A, so the superpotential W D becomes Both U ij and F U belong to adjoint of U(N + 1), since they originate from Aharony duality. The mirror of theory D, which we will identify with theory C, is now obtained by applying piecewise mirror symmetry [35]. This procedure amounts to replace each flavor U i , U i with an SQED theory coupled to a singlet χ i , and do the functional integration on the U(1) gauge node of (3.88). For each SQED theory, there is a cubic superpotential is of the form χs s, where we use schematically s ands to denote the flavors. We redefine the set of χ i as follows, . . . (3.90) Then the cubic superpotentials can be presented in the form

JHEP04(2019)138
The mirror of theory D is completed once we map the superpotential W D . In order to do so we should refine the map of the operators. Mirror symmetry would relate the meson U ij to the monopole matrix n ij . But since the meson U ij is not traceless, there is a mismatch of representations we have to take care of. More precisely, we claim that the SU(N + 1) degrees of freedom of U ij are mirror to the monopole matrix n ij , which is traceless, while the trace Tr U is in correspondence with δ. The rest of the dictionary is standard: the monopole fields m ± of theory D are mirror to the long meson L + = N i=0 s i and L − = N i=0 S i , and the flipping fields F U ij , σ ± and θ m are mapped to an equivalent number of singlets, F n ij , Σ ± , and Γ m .
The terms Γ 1 TrF n ij and δ TrF n ij combine into the mass term (Γ 1 + δ)TrF n ij . Then, both (Γ 1 + δ) and TrF n ij can be integrated out, while the field Γ 1 ≡ (Γ 1 − δ) remains massless. After trivial redefinitions, The notation ψ i , s (i,i+1) , F n ij and n ij should be familiar from the study of theory C. We have found F n ij ∈ SU(N +1), n ij ∈ SU(N +1), and other 4(N +1) fields. These corresponds to bifundamentals, and fundamentals on the right and and left of (3.92), in addition to the singlets ψ i=1,.N , the fields Σ ± , and Γ i=1,...N . Remarkably, this number is precisely the same number we determined in section 3.3 from the nilpotent Higgsing.
We have not discussed how the deformation L F F T brings F F T [SU(N + 1)] down to Theory D. This would require a study of the Higgsing process on monopole fields, a challenge which is the behind the immediate scope of this paper.
Operator map. We conclude this section by recording the chiral ring generators which we are able to map across the four dual frames ABCD:

Partition functions
In this section we study partition functions of our theories on the squashed three-sphere S 3 b , and we show that they are all equal as we move in the commutative diagram: We follow the notation of [37]. We introduce the vectors M = (M 1 , . . . M N +1 ) and T = (T 1 , . . . T N +1 ) of real mass parameters for the flavor and topological symmetries and the real mass m A associated to the U(1) A symmetry. We also define Q ≡ b + b −1 , where b is the squashing of the three-sphere. Then, the partition function of T [SU(N + 1)] can be obtained by the following set of rules: • Each one of the N gauge nodes, labelled by (k) with k = 1, . . . N , carries a measure i } represents the Coulomb Branch coordinates on the localizing locus.

JHEP04(2019)138
• The contribution from vector multiplets and adjoint chirals attached to a node (k) is • The contribution of bifundamentals on a link (k, k + 1) is As pointed out in [27], the partition function depends holomorphically on the combination of the real mass parameter m A , and the coefficient determining the IR R-symmetry. Then, we will take Im(m A ) = − Q 2 α with α parametrizing the mixing R = C + H + α(C − H). In this conventions, a chiral multiplet of R-charge r contributes with s b ( iQ 2 (1 − r) − . . .) to the partition function [37], and from Z bif and Z adj we read off This is indeed the same assignment we discussed in section 2. Putting all together the partition function of the tail (2.1) is: where ξ (k) = T k − T k+1 , and x (N +1) ≡ M a constant vector, i.e. not integrated over. The exponential factors in (4.6) correspond in the field theory to mixed Chern-Simons terms. Those inside the integral couple the topological and the gauge symmetry, while those in front of the integral are mixed background Chern-Simons terms coupling the topological and the flavor symmetry. These terms are related to contact terms in the twopoint function of the associated conserved currents and their importance in the context of dualities has been pointed out in [38,39] (see also for a review [40]).
Finally, the partition function of T [SU(N + 1)] is a specification of Z T to the case i=1 M i = 0, consistent with the non-abelian global symmetry SU(N + 1) flavor × SU(N + 1) top . Notice that when this condition is imposed the contribution of Chern-Simons contact terms in (4.6) vanishes. However it is important to keep track of all contact terms when gauging a flavor symmetry, because in such a case the background Chern-Simon terms become dynamical, and play a crucial role.

Difference operators and dual partition functions
In this section we consider the outer diagram In particular Z T ‹ is given by the same matrix integral as T [SU(N + 1)], where masses M a and FI parameters T a are swapped, and the sign of the axial mass m A inverted. This is consistent with the fact that mirror symmetry exchanges HB and CB.
Our prescription for the partition function of F F T [SU(N + 1)] is Our proof is based on [11] where, building on the results of [33], it was shown that Z T is eigenfunctions of two sets of trigonometric Ruijsenaars-Schneider (RS) Hamiltonians. We introduce a first set of RS Hamiltonians, where r = 1, · · · , N , and χ r ( T ) are eigenvalues. Due to a peculiar property of the RS system, the same eigenfunction Z T satisfies also the so-called p-q dual equation: The same steps can be repeated for the RS Hamiltonian in which b → 1 b , thus mirror symmetry is proven [11]. Quite interestingly, by considering the action of K[ M , m A ] on the Hamiltonians 11 : we can also show from (4.14) that on the partition function of T [SU(N +1)]. First, let us observe that on each node where the monopole potential is turned on, the symmetry U(1) A × U(1) top is broken to the diagonal and consequently the FI parameters take special values related to the axial mass: The last node is underformed, so there is no constraint on ξ (N ) . Following the logic of the sequential confinement, spelled out in section 3.1, we dualize the first gauge node, and sequentially all the nodes, by using the duality [15] between U(k) with k + 1 flavors and W = M + ↔ W Z model with W = γ det M. (4.22) 11 To see this we use the following property of the double sine function . (4.16)

JHEP04(2019)138
At the level of the partition functions this duality is obtained from the following evaluation formula: with the definition and the constraint from the monopole superpotential We will actually need the identity (4.23) specialised to the case in which in the electric theory the fundamental chirals couple to the adjoint breaking the SU(k + 1) × SU(k + 1) global symmetry to the diagonal and consequently the parameters µ a are specialised to µ a = iQ 4 + m A 2 for a = 1, · · · , k + 1. The constraint now reads, and we find adj ) −1 . At this point, we can apply this identity to Z T [N, m A ; M , T ], with T specialised as in eq. (4.21), starting from the first node, where the adjoint is a (gauge) singlet, and sequentially by promoting each time the real mass parameters to dynamical variables, i.e. M i → x i . Consider the first few dualizations as a warm-up, we will highlight some crucial simplifications. Focusing on the integrands, the partition function reads adj Z (3,4) bif . . .

JHEP04(2019)138
The effect of the confinement of the U(1) node has been to cancel the adjoint on the U(2) node and shift the FI. Both these modifications are such that we can apply (4.27) to the U(2) node. This procedure goes on sequentially.
After confining all nodes but the last one we obtain: On the first line we recognize the contribution of the fields γ l . Indeed, as explained around eq. (4.5), we can read out the R-charges by looking at the arguments of the s b functions and we find: (1 − R γ l ) = −1 + (1 − α)(l + 1), from which follows the solution R γ l = 2 − 2(l + 1)r. This is the same assignment of R-charges we read off from the superpotential , if we identify the indexes as l + 1 = m. The partition function of theory A is finally obtained by making explicit the values of T 1 , . . . T N , using the constraint N i=1 T i = 0. Then 4.31) and The result is where we highlighted the cancellations in the last line. The partition function of theory B is then where we used N i=1 T i = 0 in the prefactor, and defined The contributions Z d corresponds to the fundamentals fields d andd. Looking at the coefficient of iQ 2 we see that 1 − R d = N 2 (1 − α) so R d = 1 − N r as expected. The contribution Z (N,1) originates from two fundamentals chirals with R-charge (1 − r), which are still part of the tail in the quiver diagram.

Partition functions on the A-to-D side
We obtain theory D from theory A by applying Aharony duality. As reviewed in appendix B, Aharony duality is implemented by the following integral identity, with Z (Nc,N f ) defined in (4.24). Notice that here we have a background Chern-Simon term (the exponential prefactor in the second line) coupling the topological and the flavor symmetry. For N c = N , N f = N + 1 and λ = 2T N +1 , the l.h.s. of (4.38) coincides with the integrand of Z A . However, since theory A has a non trivial superpotential, which breaks for a = 1, · · · , N f . Finally, we obtain Z D upon including the prefactor associated to the dual of the fields γ l≥0 of theory A. We find: a,b=1 Some comments on Z D are in order. There is a cancellation of contact terms when using (4.38) on the integrand of theory A. This is so because the FI of theory A, compared to T [SU(N + 1)], has been reduced to T N +1 during the sequential confinement.
In the notation of section 3.4, we recognize in the first line of Z D the contribution of the two singlets, σ ± , and that of (N + 1) 2 singlets F U D . The fields σ ± are flipping fields for the monopoles, and the fields F U D flip the meson. From the arguments of the s b we read out Notice also that the contribution of θ 1 (the coefficient with l = 0 in the product) cancels the diagonal contributions of the singlets F U D , effectively enforcing the tracelessness of F U D .

T [SU(2)]
The case of T [SU (2)] is simple enough to compute the partition function explicitly. In this case, our monopole deformation is empty, thus the partition function of theory A is directly that of T [SU (2)], with the specification T 1 = 0. The presence of flipping fields, and the non trivial mapping of parameters across the commutative diagram, makes the equalities of partition functions a nice exercise to go through. In these computations we The integrand of the T [SU (2)] partition function can be quickly evaluated by residue integration [41]. The full partition function will also include a factor of Z (1) adj = s b (m A ), and an exponential prefactor. In our conventions, ξ = T 1 − T 2 ,

JHEP04(2019)138
We pick poles of the two s b functions at the numerator, and assign to the set of poles labelled by M i , the series S i defined as, Recall the definitions q = e 2πibQ andq = e 2πiQ/b . The hypergeometric function 2 ϕ (q) 1 admits the series representation If |q| < 1 and |q| > 1 we use the relation where the r.h.s. can be expanded out as in (4.42). The partition function is invariant under b ↔ 1/b and can be written as the sum 2 i=1 e 2πiT 2 (M 1 +M 2 ) s b (m A ) S i . We will work with its factorized expression [45], namely In (4.44) we isolated the following exponential prefactor and defined the holomorphic blocks B i=1,2 associated to the series S i=1,2 : 12 The holomorphic blocks of theory A, B A i=1,2 , can be defined from B i=1,2 by setting T 1 = 0. Whenever needed we will understand b → b −1 in the conjugate blocks B 12 Compared to [45] we do not introduce Θ-functions to factorize exponential terms, but we use the 'factorization' matrix (4.44).

JHEP04(2019)138
Aharony duality. The partition function of theory D is given by (4.39), The FI parameter ξ should be fixed to be −T 2 , but we are keeping it generic for comparison with Z F F T in the next section. The integrand (4.48) can be evaluated by residue integration, as in (4.41). The modifications are minors so we will not repeat them. Instead, after writing the partition function in the factorized form, we show how the blocks map into each other when going from theory A to theory D. The factorization of the partition function is where P is the same prefactor (4.45) and the holomorphic blocks are Flip-Flip duality. In order to compute Z F F T we follow our prescription (4.8). Considering Z T [1, −m A ; . . .], with Z T given in (4.40), we obtain Very similarly to the T SU(2) computation, we evaluate the integrand in (4.56) and factorize the result into where, by using again (4.52) on the holomorphic blocks B F F i=1,2 , we find Thus we have shown that Notice that our definition of Z F F T in (4.8), which was strongly motivated by the use of difference operators, has correctly captured possible field theory contact terms. In section 2.2 we used a field theory argument to show that Aharony duality applied to T [SU (2)] is related to Flip-Flip duality. This is consistent with the observation that (4.58) and (4.54) follow from the same Heine's identity (4.52), i.e. they are not independent. However, we did not obtain Flip-Flip duality directly, and we insisted on some additional manipulations. These manipulations will also be visible at the level of the partition function: consider the action of Aharony duality on T [SU(2)] by implementing (4.38) on the integrand of Z T . We denote this by A • Z T . Then, we find the relation The contribution of the SU(2) flavor × SU(2) top flipping fields in F F T [SU (2)] comes out as follows: on the r.h.s, of (4.59) we find where the terms in parenthesis [. . .] are introduced by the Aharony duality. Then, one of the diagonal contributions, i.e. a = b = 1 or a = b = 2, simplifies with the original s b (m A ) of T [SU (2)], and we recover the same prefactors as in (4.56). Furthermore, when the constraint M 1 + M 2 = 0 is imposed, the relation (4.59) implies the equalities as expected from the field theory argument.

Mirror symmetry. Explicit computations about the partition functions of T [SU(2)] and T [SU(2)]
‹ have been done in [45]. For sake of completeness, we repeat them in our notation to show consistency. We shall refer directly to theory A and theory B, since the monopole deformation is empty.
We write ζ = M 1 − M 2 and (4.34). In its factorized form we can then extract the same prefactor P, given in (4.45), and obtain, The map between blocks under mirror symmetry is more involved than (4.53). It can be derived from the use of the first Heine's identity and the analitic continuation formulas [45]. The result is For the conjugate blocks we get The connection matrix in (4.70) is essentially the inverse of (4.67), transposed.

JHEP04(2019)138 5 Spectral dualities
In this section we connect the dualities discussed in the first part of the paper to a class of 3d dualities which we call spectral dualities since they have their origin in 5d spectral dualities, or fiber-base duality in topological string.
In the introduction we claimed that 3d spectral dual pairs can be regarded as 3d N = 2 theories living on a codimension-two defect which is coupled to a (trivial) 5d N = 1 theory. The starting point of this construction is a toric CY three-fold X which engineers a 5d N = 1 linear quiver theory. A 3d − 5d coupled system can be obtained via Higgsing, by tuning the Kähler parameters of the CY X in a specific way. The resulting CY will be denoted by X 13 and since we will be considering a complete Higgsing it will correspond to the 3d theory T X coupled to 5d free hypermultiplets. In particular the topological string partition function we started with reduces to the partition function of our 3d theory T X . From the original fiber-base duality of the CY, we can then infer the existence of 3d dualities, which we will discuss in the next section.
More precisely, we have found the following relation between the holomorphic block (D 2 × S 1 partition function) B α 0 T X evaluated on a contour α 0 and the partition function of the Higgsed topological string on X : We have separated the topological string partition function on the r.h.s. of eq. (5.1) into two pieces, Z X 1−loop,top and Z X vort,top which coincide with the vortex part of the 3d partition function. Z X 1−loop,top is independent of the 3d FI parameters, and hence of the corresponding Kähler parameters of the CY, while Z X vort,top does depend on them. Finally G denotes a possible fiber-base invariant prefactor and E T a contact term.
The choice of contour α 0 on which the holomorphic block is evaluated corresponds to a particular way of tuning the Kähler parameters to implement the Higgsing. More concretely, the different contours correspond to Higgsed toric CY's, in which the spectral parameters in the external legs of the toric diagram are fixed, while the internal ones can vary. For example, two Higgsed CY's corresponding to two contours (or vacua) of the F T [SU (2)] theory are shown in figure 1 (see section 5.3 for notations).
In the following we present our two main spectral dual pairs: 2) T ↔ T , which is obtained from the D ↔ B duality in the ABCD framework.
After discussing the field theory evidence of these dualities we will see how the holomorphic blocks of each theory can be obtained via Higgsing from a topological string partition function and we will then explicitly see how the spectral duality descends from the fiber-base duality. Figure 1. Two toric CY diagrams corresponding to two vacua of the 3d F T [SU (2)] theory. Notice that the spectral parameters of external legs are the same in both cases.

F T [SU(N + 1)] and its spectral dual
We now add another set of (N + 1) 2 singlets F T ij and deform the F F T [SU(N + 1)] theory by the superpotential δW = F T ij F m ij . We are basically flipping twice the Coulomb branch of T [SU(N + 1)] and since f lip 2 = 1, as it is easy to see by using the equations of motion, we find a new theory, which we call F T [SU(N + 1)], where only the Higgs branch moment map is flipped: On the dual side F T [SU(N + 1)] ‹ , we proceed similarly. We add new (N + 1) 2 , singlets which we call F P ij , and the superpotential deformation δW = F P ij Π P ij . This deformation is dual to that for ‹ we again assign R-charge r to the quarks so that again we will find a monopole matrix with R[N ij ] = 2 − 2r, and R[F P ij ] = 2 − 2r. The operator map will be: The first evidence of this duality was obtained in [25] using difference operators acting on the holomorphic blocks. The argument is similar to our discussion in section 4.1.
The partition function of the F T [SU(N + 1)] theory is simply obtained by multiplying the partition function of T [SU(N + 1)] by the contribution of the flipping singlets which transform in the adjoint of the SU(N + 1) flavor symmetry: where the prime indicates that we removed the trace part from the singlet contribution.

JHEP04(2019)138
Considering the map of operators in (5.5) we see that flavor and topological fugacities will be swapped in the partition function of the dual theory, but the sign of m A will not change, consistently with our R-charge assignment. We have: (5.7) Proving our spectral duality at the level of partition functions requires to prove the following identity:  The additional flipping, which lead us to the spectral dual pair, is trivially implemented by moving the contribution of the singlets from the left to the right: which up to m A → −m A is the identity we were looking for. It is interesting to observe that F T [SU(N + 1)] and its spectral dual, similarly to T [SU(N + 1)] and its mirror dual, describe the low energy theory on a stack of D3 branes suspended between NS5 and D5 branes. Crucially, however, the IIB brane setup for F T [SU(N + 1)] involves D5 branes spanning the 012478 directions as shown in table 2 (see also figure 2), so we call them D5' to distinguish them from the ones relevant for T [SU(N + 1)] in figure 3. The difference between the two set-up is a "brane flip" -the D5' and D5 branes are transformed into each other under the exchange of directions 56 ↔ 78. The set-up in table 2 and figure 2, which preserves N = 2 supersymmetry, is also invariant under the action of Type IIB S-duality which turns the NS5 branes into D5' branes leaving the D3 branes invariant and explains the spectral self-duality of F T [SU(N + 1)]. Notice also that in the brane-realisation the (N + 1) 2 singlets fields which flip the mesons correspond to the degrees of freedom of the D3 branes moving in directions 78 between two D5' branes (one hyper for each D3 segment) and between a D5' and an NS5 [46].
At this point it is tempting to speculate that performing also the flip of the CB moment map to obtain F F T [SU(N +1)] corresponds to rotating also the NS5 into NS5'. This would give a new N = 4 set-up with NS5' and D5' equivalent to the one in table 1 NS5 NS5 x 3 x 8 that all the five-branes sit at the same point in x 3 direction. The NS5 and D5' branes will form a (p, q) web in the 49 plane, as shown in figure 4(a) for the simplest example of N = 1. The worldvolume theory on the five-branes is the 5d N = 1 gauge theory living in the 01278 space. The positions of the five-branes in the 49 plane correspond to Coulomb moduli, couplings and masses of the gauge theory. In particular for the "square" (p, q)-web formed by (N + 1) NS5 and (N + 1) D5' the worldvolume theory is the U(N + 1) N 5d linear quiver theory with (N + 1) fundamental hypermultiplets at each end.
If we now go to the Higgs branch of this 5d theory where the NS5 and D5' branes are separated in the x 3 direction we can stretch D3 branes between them as in figure 4(b), arriving precisely at the setup of table 2. Hence we explicitly realize the F T [SU(N + 1)] theory as a defect theory appearing in the Higgs branch of the 5d theory. This realisation of F T [SU(N + 1)] as a defect theory has been discussed extensively in section 3 of [25], here we summarize the salient points.

JHEP04(2019)138
x 9 NS5 NS5 x 9 x 4 x 3 where G is a fiber-base invariant factor. The parameters µ i , τ j are identified with Kähler parameters while the exponentiated axial mass t is identified with one of the equivariant Ω-background parameters on R 4 q,t × S 1 . Then in [25] it was observed that the topological string partition function is invariant under fiber-base duality (which in the case of the square diagram S is self -duality) even after Higgsing: which implies the 3d spectral self-duality of the 3d blocks: Fiber-base duality exchanges the Kähler parameters of the base with that of the fiber, thus it exchanges µ i and τ j , but t is left untouched since it is the parameter of the Ω-background (or, equivalently, the refinement parameter of refined topological string).

A new spectral dual pair
The reasoning that led us to state the spectral duality between F T [SU(N + 1)] and F T [SU(N + 1)] ‹ can be used on theory D and theory B to obtain a daughter spectral duality. Recall that theory D is SQED with (N + 1) flavors, U i and U i , mesonic and monopole flipping fields, and superpotential (5.14) The flipping fields σ ± and F U ij originated from Aharony duality on theory A. To arrive at theory T we flip the singlets σ ± and θ m≥2 , since f lip 2 = 1 we arrive at: where we redefined θ 1 = φ, for simplicity. We then can use the F-terms of φ and consider traceless flipping fields. Theory T is obtained from theory B upon repeating the same two operations that define theory T . From the operator map given in section 3.4, we see that the fields θ m≥2 correspond to dressed mesons of theory B: while the monopoles σ ± are mapped to the two mesonsdp and dp. So we have: The equality of the partition functions Z T = Z T follows from the equality Z D = Z B simply by reshuffling the flipping fields and we obtain: 15 In the first line we can notice the cancellation of the trace-part of the flipping fields with the singlet φ.

JHEP04(2019)138
Holomorphic blocks. In this section realise theory T and T as defect theories via Higgsing. First of all we need the holomorphic blocks, i.e. D 2 × S 1 partition functions evaluated on a reference contour. The block integrands Υ T and Υ T can be easily obtained by taking the "square-root" of the S 3 b integrand as observed in [42], and reviewed in [43]. Their explicit expression can be found in eqs. (C.1) and (C.3) in the appendix.
We then evaluate the block integrands on a basis of contours Γ α with α = 1, · · · N + 1 which are in one-to-one correspondence withe the (N + 1) SUSY vacua of the theory. Similarly we will evaluate the block integrand for the spectral dual theory on a basis of contour to obtain the blocks B β T : Testing the the spectral duality at the level of the blocks requires to establish a map between each element of the basis of theory T and T . In terms of field theory objects, the matrix elements of this map are partition functions of 2d theories living on the interface between theory T in vacuum α and T in vacuum β. Geometrically the interface is a torus ∂(D 2 × S 1 ) = T 2 , the equivariant parameter q of the D 2 × S 1 background plays the role of the complex structure of the boundary torus and the 2d partition function is a version of elliptic index, hence expressed in terms of Jacobi theta-functions θ q . However we will not be concerned with evaluating fully the matrix of transition coefficients. We limit ourselves to the evaluation of the blocks of T on a reference contour Γ α 0 . On the dual side we are able to identify the corresponding contour which we also call Γ α 0 . The details of the calculations can be found in the appendix, here we give the final result: The explicit forms of Z α 0 cl,T Z α 0 1−loop,T and Z α 0 vort,T are given in in eqs. (C.11), (C.10). On the dual side we have: The explicit forms of Z α 0 cl,T Z α 0 1−loop,T and Z α 0 vort,T are given in eqs. (C.17), (C.15).

Spectral duality from fiber-base
In this section we explain how the 3d spectral duality between theories T and T follows from fiber-base duality of refined topological string. First of all we need to establish the Higgsing prescription which allows us to obtain B α T and B α T from refined topological string partition functions with tuned Kähler parameters.
Refined topological strings provide a deformation of the topological A-model partition function on toric CY threefolds. Apart from the exponentiated string coupling q = e −gs the deformation depends on an additional parameter t, so that for t = q the conventional JHEP04(2019)138 partition function is recovered. The rules for computing partition were introduced in [47].
Here we briefly recall that the main ingredient is the trivalent refined vertex It will be more convenient for us to use spectral parameters, assigned to all the legs of the diagram, instead of Kähler parameters associated only with the intermediate edges. Figure 5 explains the identification for the basic example we will need in our setup, the resolved conifold geometry. The piece of the partition function corresponding to the resolved conifold from figure 5 is given by In what follows we normalize Z conifold so that it is an identity when all the external legs are empty, i.e. we divide by Z

JHEP04(2019)138
diagrams). Namely for the situation pictured in figure 6, the lengths of the diagrams on the vertical leg before and after the crossing are constrained as follows: These constraints are valid irrespective of the diagrams propagating on the horizontal leg. For example if the resolved conifold fragment (5.24) sits in the lowest part of the diagram, then Y = ∅ since it corresponds to an external leg, and therefore W is constrained to be empty. The block (5.25) in the same situation would constraint the diagram W to have just one column, i.e. W = [k], k ∈ Z ≥0 . The integers k in this construction will correspond to the summation variables in the 3d vortex series. We will denote the "Higgsed" CY manifold (i.e. the CY with discrete choice of Kähler parameters) corresponding to the 3d theory T by Y and that corresponding to T by Y . Of course, Y is the fiber-base dual (the mirror image along the diagonal) of Y. Below we give some details of the topological string computations for Y and Y .
CY Y. The toric diagram for the CY Y in the case N = 4 looks as follows: Here we have explicitly indicated the Young diagrams propagating on the intermediate vertical legs. These diagrams are constrained by the rules, (5.24), (5.25), so that [k] is the single column Young diagram. It is this variable over which the summation in the vortex series is performed. We normalize our partition function so that Z Y vortex ( µ, ξ, q, t) is a series in e 2πbξ which starts with identity. The partition function can then be computed explicitly e.g. using the resolved conifold formula from eq. (5.23) 16 and the result coincides with the series vortex series (C.10).
The relative prefactor Z Y top,1−loop ( µ, ξ, q, t) is easy to calculate -it is what remains of the partition function when τ 1 τ 2 goes to zero. This limit corresponds to an infinitely large 16 There is, however, a more compact and convenient operator product technique [48][49][50], which we don't present here not to overcomplicate the presentation with technicalities.

JHEP04(2019)138
Kähler parameter between the two horizontal legs in (5.26), so that the diagram splits into a product of two horizontal strip partition functions. Indeed, in this limit only k = 0 contributes and we have: Here we have written out the spectral parameter of the vertical legs explicitly. The wellknown formula for the refined strip partition function gives [51]: CY Y . The toric diagram for the spectral dual CY Y is simply the mirror image along the diagonal of that of Y (5.26), so that: Here we have again used the rules (5.24), (5.25) to constraint the Young diagrams on the vertical legs. The integers k (a) i are precisely the integers in the 3d vortex sum and Z Y vortex can be checked to reproduce (C.15).
The computation of the one-loop factor is similar to section 5.3: the toric diagram in the limit τa τ a+1 → 0 splits into (N + 1) horizontal strips. Using the result for the strips we obtain The fiber-base duality of the topological string partition function (after Higgsing) yields the following equality: A simple brute force check of eq. (5.31) to lower orders in the Kähler parameters is given in appendix D.

JHEP04(2019)138
Match of field theory with Z top . Finally we relate our gauge theory results for the holomorphic blocks with the results of the Higgsing prescription. We find that: and where G T , G T denote fiber-base invariant prefactors.
Since we checked the fiber-base duality of the refined string eq. (5.31))we are left to check that: In fact eqs. (5.35), (5.34) can be relaxed slightly: the r.h.s. can be a q-periodic function, e.g. a combination of θ q -functions which also becomes an identity when glued into the S 3 b partition function. Notice that the topological string partition function lacks the classical (i.e. power function) part, so the relation (5.35) is essentially the requirement that the classical part of the field theory holomorphic block be fiber-base duality invariant on its own.

JHEP04(2019)138
Dividing eq. (5.36) by eq. (5.37) we get Thus, to get the invariance we need to have And indeed we obtained the ratio of the contact terms as a determined from the gauge theory partition function calculation in eq. (C.6) in the appendix C.
We have thus established the spectral duality for theories T and T using topological string computation. It is remarkable that the field theory computation matches the topological string not only qualitatively but with such a quantitative finesse.

JHEP04(2019)138
The two objects D µ S (k,k+1) and D µ S (k,k+1) , are themselves bifundamentals. The covariant derivative for the adjoint scalars on a node U(k) is Nilpotent vev and D-terms. We discussed in the main text the role of D-terms in the solution of our nilpotent vev. Our notation for a D-term there was the following: for a gauge node U(k), with bifundamentals on the left, L = (k − 1, k), and on the right, R = (k, k + 1), we have Then, it is straightforward to compute on the nilpotent vev (3.60) the following matrix products Gauge multiplets mass matrix. Given the Lagrangian of the theory, the mass matrix for spin-1 fields can be obtained from the covariant derivatives of the charged fields. We expect that the bifundamentals S and S, whose kinetic term is will be responsible for mass terms between different gauge nodes. We quote the form of the mass matrix coming from the bifundamentals S (k,k+1) , since it is instructive: In this formula A N +1 = 0 since the last node is a flavor node, i.e. it is ungauged. Matrix elements are understood on the nilpotent vev. The contribution of S-type bifundamentals is similar to (A.10). Then, if we split the total mass matrix into the contributions of S, S, and Φ, adjoint fields will not couple different gauge nodes. The total mass matrix has the following block structure, with non zero crossed blocks. This structure resemble indeed that of the quiver: all but the first gauge node get two contributions, one from bifundamentals on the left, one from the right.

JHEP04(2019)138
After careful evaluation of (A.11) we were able to double-check the solution quoted in (3.73). This same solution can then be understood in a simpler way by thinking about the action of broken gauge generators, along the lines of what we stated in section 3.3.
A basis for massive chiral fields. When discussing Theory C we described, within the set of UV fields, an explicit basis for the massless fields on the nilpotent vev. This basis contained two subspaces: physical IR massless fields and goldstone bosons. In the physical sector we then had a further splitting: bifundamentals, and adjoints. This splitting is orthogonal by default. However, physical massless fields are not orthogonal to goldstone bosons. (This is OK, since both are in the kernel of the matrix, and it might happen that is just convenient, but not needed, that physical massless are taken to be orthogonal among themselves).
In order to obtain a basis for massive chiral fields we can adopt the following strategy.
• We split the set of UV fields, call them B, into the set of physical IR fields and its orthogonal, hereafter denoted by K ⊥ . (This is not ker ⊥ ). The only non trivial construction in K ⊥ regards the adjoints, since as we mentioned, bifundamentals and adjoints are orthogonal by default. In practise we construct where the first set contains only physical massless fields in the IR. We check that . v #uv }, then we know that {goldostone bosons} ⊂ K ⊥ . We do not impose orthogonality among the vectors in {v # ir+1 , . . . v #uv }.
• For each goldstone boson, call it G k , we impose the orthogonality condition These linear equations fix a number of parameters equal to the number of goldstone bosons. The resulting free parameters provide a span for the massive fields, i.e. the actual ker ⊥ . Vectors in this basis are not orthogonal among themselves, but they are automatically orthogonal to physical massless fields which is what we were looking for. Concluding we have splitted B in the form B = K ⊕ G ker ⊕ ker ⊥ (A.14) Let us come back on the first part of this construction, i.e. a convenient basis for adjoint fields. Note indeed that massless IR fields in the adjoint are not directly aligned with a basis of hermitian matrices for U(N ), so it is better to use an alternative basis. Consider the map ι k : R d×d → R k with k ≤ d defined as Then, for each node U(n) we construct a basis Φ n of the adjoint rep recursively. Define Φ n−1 to be the basis of U(n − 1) built out of ι k for k ≤ n − 1. We can embed Φ n−1 in Φ n in two ways The embedding on the right of (A.17) will be needed for the U(N + 1) flavor node. The other one is used on the gauge nodes of the tail. In order to find a complete orthogonal basis we only need elements parametrizing the remaining row and a column of Φ n . Finally we normalize. The basis ⊕ N +1 n=1 Φ n parametrize the 2N +3 fields {Γ i , ψ k , Σ ± , δ}, in a natural way. A basis orthogonal to these 2N +3 fields is also simple to construct.
More general nilpotent deformations. The nilpotent vev we studied, together with the Higgsing, can be generalized outside next-to-extremality. For example, let us label the F-term deformation generated by the monopoles using a partition, i.e. the following set of integers: I = {n 1 , . . . , n N } with n i ≥ 0 and N l=1 ln l = N . In (3.57) we considered I = (0 N −1 , 1), which naturally generalize to,

B Bookkeeping integrals
In this appendix we collect some useful results about hyperbolic integrals.
Double-sine function. The double-sine function, s b , appeared in the very first computation of [37], as a building block for the localized partition function of 3d N = 2 theories on the squashed sphere S 3 b . It can be introduced with an infinite product representation, which is perhaps familiar to the physics literature, It satisfies the following non trivial properties where z = e iπλ and u i are arbitrary. For example, if we take u i = − iQ 4 − m A 2 , we find Z bif as defined in the main text with a minus sign for the masses.

JHEP04(2019)138
with the constraint r µ r = (m + 1)(ω 1 + ω 2 ) . (B.17) Notice that (B.16) provides an evaluation formula when m = 0. Gauge theory parameters N c and N f enter with the following dictionary: N c = n and N f = (m + n + 2), thus m = N f −N c −2. The background parameter b, which measures the squashing of the threesphere, enters through ω 1 = ib and ω 2 = i/b, thus ω 1 +ω 2 = iQ. Finally Γ h (x) = s b ( iQ 2 −x). Summary. In the notation of [15], the equality Z T M = Z T M is obtained from (B.16) by taking the limit Then [15] find other two results: Derive the monopole duality [section 8 of [15]], which we used in this paper, U(N c ) ⊕ N flav. and W = M + ↔ N 2 f ⊕ 1 singlets M ij ⊕ γ and W = γ det M • Recover Aharony duality, The corresponding integral identities can be deduced from (B.16) as follows, constraint becomes • t → ∞, (B. 21) constraint becomes In the next paragraphs we present some important details on contact terms involved in these computation.
Details on (B.18). Starting from (B.16), it is trivial to substitute the m, n dependence with gauge theory parameters N c and N f . The strategy of [15] is to rewrite the integrals as

JHEP04(2019)138
In terms of N f , the vector µ has 2N f component, and s enter with different signs, specified in (B.18). Since the integration variables are also shifted by +s, it will happen that out of the combinations, µ r ± x i and ±x i ± x j , which appear in (B.15), some are invariant and the others are shifted twice. For example Taking into account these shifts, and the asymptotics expansion (B.24), we find two (different) prefactor in (B.16), one for the l.h.s. and one for r.h.s. These two prefactors depend on m i ,m j , the integration variables, and s. In particular there is a divergent part. However, upon imposing the constraint, the dependence on the integration variables drops, and the simplified prefactors cancels each other from r.h.s. to l.h.s. It follows that and Details on (B.19). On the electric side, i.e. Z T M with N f + 1 flavors, we consider the following manipulations on the integrand: split the product over j = 1, . . . N f + 1 into 1 ≤ j ≤ N f and the last one, and take the limit (B.19), The total prefactor will be Nc i=1 e iπt(iQ−ζ) e iπ(ζ−iQ)x i . On the magnetic side, i.e. Z T M , the same kind of manipulations lead to

JHEP04(2019)138
Additionally, on the magnetic side the prefactors produce extra terms, Details on (B.21). On the electric side, i.e. Z T M with N f + 2 flavors, we single out the two extra flavors and take the limit, thus producing Then, the total prefactor is Nc i=1 e iπt(2iQ−ζ) e iπλx i . Similarly on the magnetic side, We can finally re-introduce the notation The result is

C Holomorphic blocks calculations for T and T
In this section we evaluate the holomorphic blocks for T and T over the reference contours. We first list the integrals obtained via factorisation of the S 3 b partion function, which is a consequence of the factorisation property of the double sine (B.2).
For theory T we have is the contribution of the flipping fields and E T is a contact term. We have also introduced the exponentiated variables µ i = e 2πbM i and q = e 2πib 2 , t = q β . On the dual side, for theory T we have: