Ellipsoid partition function from Seiberg-Witten monopoles

We study Higgs branch localization of N=2 supersymmetric theories placed on compact Euclidean manifolds. We analyze the resulting localization equations in detail on the four-sphere and find that in this case the path integral is dominated by vortex-like configurations as well as singular Seiberg-Witten monopoles located at the north and south pole. The partition function is written accordingly.


Introduction
After it was understood in [1] 1 how to apply localization techniques [3,4] to perform exact computations of partition functions and vacuum expectation values of supersymmetric operators in supersymmetric quantum field theories defined on compact Euclidean manifolds, a wealth of exact computations in theories defined on a variety of geometries in a variety of dimensions has become 1 See also [2] for related earlier work. available, see for example . Such exact, non-perturbative results can be put to excellent use in precision tests of various non-perturbative dualities, but their applications are much richer. Indeed, recently a lot of research has been conducted on interpreting and applying the wide variety of exact results available, resulting in an impressive list of both physical and mathematical developments.
Localization computations are based on the observation that in the path integral of a supersymmetric theory one can add Q-exact 2 deformations to the action without changing the resulting partition function. For a positive semi-definite such deformation, one can then easily argue that a one-loop computation around its zeros is exact. The canonical choice of deformation term has as its zeros certain configurations involving the (bosonic) vector multiplet fields, while all matter multiplet fields are set to zero. Typically, these configurations take the form of arbitrary constant values for the vector multiplet scalars or holonomies around circles. 3 The path integral then collapses to a finite-dimensional matrix integral over this classical Coulomb branch, hence the localization based on the canonical deformation term is called Coulomb branch localization.
Upon choosing a particular additional deformation term (or equivalently, by changing the integration contour of the auxiliary fields in complexified field space) and if certain conditions on the parameters of the theory hold, the localization locus instead consists of a finite number of discrete Higgs vacua, where matter multiplet scalars can acquire a vacuum expectation value solving the D-term equations, accompanied by infinite towers of non-perturbative point-like Higgs branch configurationse.g., vortices or Seiberg-Witten monopoles -located at special points in the geometry.
Such a Higgs branch localization computation was first performed in N = (2, 2) theories on the two-dimensional sphere [5,6], and later applied in and extended to two [7], three [41,42], four [43] and five [44] dimensional theories. 4 In this paper, we apply the Higgs branch localization technique to N = 2 supersymmetric theories placed on compact Euclidean manifolds. We derive the general localization equations and subsequently study their solutions in detail on the four-sphere S 4 . On this geometry, the 2 Here Q denotes a particular supercharge of the theory, which in general is not nilpotent. Then the precise statement is that one add to the supersymmetric action S a deformation term QV, for some fermionic functional V satisfying Q 2 V = 0. 3 In the presence of homological two-cycles in the manifold, also a sum over magnetic fluxes will occur. The integration over holonomies around circles is a particular instance of the more general case which entails integration/summation over the space of flat connections. In higher-dimensional examples, also (point-like) instanton configurations will appear as zeros of the canonically chosen deformation term. 4 These results are closely related to the factorization results initiated in [45], and extended and generalized in [46][47][48][49][50][51][52].
Higgs branch localization locus is given by vortex-like configurations and singular Seiberg-Witten monopoles centered at the north and south pole of S 4 . The appearance of the latter can be understood intuitively as follows: locally around the north and south pole, the theory looks like the (anti-) topologically twisted theory with hypermultiplets. The localization locus of the latter theory is described by the Seiberg-Witten monopole equations (or their non-abelian version, i.e., the generalized monopole equations), which follows immediately by imposing the D-term equations.
In the localization computation on the four-sphere, the effect of the additional deformation term is precisely to impose the D-term equations, while the gauge symmetry is generically broken to the maximal torus by one of the other BPS equations. Finally, we compute the resulting Higgs branch localized ellipsoid partition function. As a byproduct, we formulate a prediction for an interesting relationship between the instanton partition function and the Seiberg-Witten partition function, which capture the equivariant volume of the instanton moduli space and Seiberg-Witten moduli space respectively.
The rest of this paper is organized as follows. In section 2 we derive the general Higgs branch localization equations of N = 2 supersymmetric theories. In section 3 we find various classes of solutions on S 4 . Next, in section 4, we compute the Higgs branch localized partition function.
Finally, in section 5, we match the Coulomb branch and Higgs branch localized results in a simple example. Appendices A and B summarize our conventions and recall the generalized Killing spinor equations and supersymmetry multiplets and variations. Appendix C studies the locally almost complex structure one can define using the Killing spinor solutions. Appendix D finally contains some useful specifics about the ellipsoid.
Note. There has recently appeared a paper on the arXiv claiming to perform Higgs branch localization on S 4 b [53]. Our results are significantly different from theirs. Throughout the paper we will point out the major points of disagreement.

BPS Equations
In localization computations, the path integral is localized to the zeros of a positive semi-definite deformation term S def. = QV satisfying Q 2 V = 0. In this section we introduce the relevant deformation terms and derive the resulting BPS equations characterizing their zeros for N = 2 theories placed on manifolds admitting solutions to the generalized Killing spinor equation (B.1) and auxiliary equation (B.2). We restrict our attention to Killing spinors satisfying the orthogonality conditions (B.23) guaranteeing that no scale or U (1) r transformations appear in Q 2 . We also indicate how the equations simplify for the case of the ellipsoid S 4 b .

Vector Multiplet
The canonical deformation Lagrangian for the vector multiplet is given by [1,22] where one considers the reality properties of various fields as in (B.5). We further introduce the With the reality properties (B.5), Qλ I and Qλ I do not satisfy the symplectic-Majorana condition, but can be decomposed in "real" and "imaginary" pieces which do: Explicitly, one finds where the tensor fields S µν ,S µν , T µν , andT µν , are introduced in appendix B (see (B.1) and (B.24)).
The bosonic part of the deformation Lagrangian then becomes L VM def. bos.
To perform Higgs branch localization, we add an additional deformation Lagrangian To derive this result, we made use of the fact that 1 s Θ µν,IJ (κ∧ λ) − µν = 1 sΘ µν,IJ (κ∧ λ) + µν for arbitrary one-form λ thanks to a Fierz identity. Note also that we have effectively taken D IJ away from its purely imaginary integration contour. Substituting back (2.11) in (2.5)+(2.10), we find the following new sum of squares where we used the convenient tensors

Hypermultiplet
The canonical deformation Lagrangian for the hypermultiplet is given by One can split Qψ αA and Qψα A into "real" and "imaginary" pieces, with respect to complex conjugation as in (B.8), using the canonical reality properties for q IA , but anti-canonical ones for F IA (see (B.9), with (B.15)) 6 Re which are set to zero to obtain the BPS equations. Multiplying the BPS equations following from the imaginary pieces with ξ K ,ξ K respectively, and taking their sum and difference using (B.14), one obtains where Ξ KI ′ = (ξ K ζ I ′ ) = (ξ Kζ I ′ ). Similarly, the real equations imply that (2.20)

BPS Equations on Ellipsoid
For the specific case of the ellipsoid S 4 b , one can use the fact that (see appendix D) to simplify the deformation Lagrangian (2.12) to (2.22) 6 The hypermultiplet transformation rules can be found in appendix B. The "·" notation is also explained there.
Here we also used that 1 The arguments of the squares in (2.22) are the ellipsoid BPS equations, which are supplemented by the D IJ equation of motion (2.11), which simplifies to The hypermultiplet equations are given by Equation (2.20) also still holds.

BPS Solutions
In this section, we study the solutions to the ellipsoid BPS equations derived in subsection 2.3.
Depending on the choice of H IJ , we find different classes of solutions. For simplicity, we work on the round four-sphere S 4 ; the generalization to the ellipsoid is expected to be straightforward, but technically somewhat involved.

Coulomb Branch
Let us start by recalling the standard Coulomb branch localization locus, obtained by solving the BPS equations for H IJ = 0. It was argued in [1,22] that the solution sets all hypermultiplet fields to zero, while the smooth vector multiplet solution reads for a a constant, which can be chosen to lie in the Cartan subalgebra. Additionally, since s = sin 2 ρ 2 ,s = cos 2 ρ 2 vanish at the north pole (ρ = 0) and the south pole (ρ = π) respectively, we see from (2.22) (with T =T = 0 on the round four-sphere) that at the north pole the equations on the field strength relax to F + = 0 and at the south pole to F − = 0, allowing for point-like (anti-) instantons.
Before studying the solutions that become available upon turning on H IJ we introduce some notation. The A = 1 (A = 2) components of the hypermultiplet transform in representations R (R) of the combined gauge and flavor group (see also appendix B). We introduce a vector multiplet for this combined symmetry group, whose gauge group components are dynamical while its flavor group components are background, and denote its scalars as Φ 1 , Φ 2 . To preserve supersymmetry, the background components need to satisfy the vector multiplet BPS equations of subsection 2.3 (for H IJ = 0). In particular, from (3.1), it is clear that one can give a vev to the background piece of Φ 1 (and the background auxiliary field) which corresponds to turning on a (real) mass for the hypermultiplet. 7 Decomposing R into irreducible representations of the gauge group as where m j is a mass for the U (1) flavor symmetry carried by the hypermultiplet transforming in gauge representation R j (R j ).
We choose where the sum runs over the irreducible gauge symmetry representations R j and its generators T a R j . Furthermore, ζ is a dimensionless adjoint-valued parameter defined as ζ ≡ h a :u(1) ζ a h a , where the sum runs over the generators h a of u(1) factors of the Lie algebra of the gauge group, and ζ a are real parameters. It will turn out to be useful to split the ζ a parameter in two pieces as ζ a = ζ a vac. + ζ a SW , with ζ a vac. , ζ a SW of the same sign, and define H SW IJ = H IJ + ζvac. ℓ w IJ .

Deformed Coulomb branch
The deformed Coulomb branch is characterized by vanishing hypermultiplet scalars. Then, using which are a direct consequence of the generalized Killing spinor equations on the four-sphere S 4 , one can write the vector multiplet equations as Notice that on S 4 one can explicitly verify that With this fact, it is easy to check that 8 for constant a is a solution. Again, a can be diagonalized. On top of this smooth Abelian solution, we again can have point-like (anti-) instantons located at the poles of the sphere.

Higgs Branch and Seiberg-Witten Monopoles
Higgs-like solutions They are characterized by the requirement that H SW IJ vanishes. The vector multiplet equations then reduce to the deformed Coulomb branch equations of the previous subsection with deformation parameter ζ vac. , and have the solutions (3.6). The value of ζ vac. will be fixed momentarily. In particular, the field φ 1 is a diagonal, constant matrix. Its values are further constrained by the hypermultiplet equation (2.26), i.e. Φ 1 · q IA = 0. The combined set of equations are in fact the standard vacuum equations of an N = 2 supersymmetric theory in the presence of an FI-parameter. Their solutions strongly depend on the choice of gauge and matter representations of the hypermultiplets. We will be interested in cases where generic masses and generic Fayet-Iliopoulos parameters ζ a completely break the gauge group. More precisely, we restrict ourselves to cases where the first vacuum equation in (3.7) uniquely determines the components of φ 1 in terms of the m j , and moreover where all components take distinct values, thus breaking the gauge group G to its maximal torus U (1) rank G . The hypermultiplet scalars taking on a vacuum expectation value further break these U (1)s via the Higgs mechanism. One arrives at a discrete set of Higgs vacua. It is clear then that after the gauge group is broken to U (1) rank G , it is sufficient to analyze a U (1) gauge group with a single flavor, which, up to rescaling of U (1) charges, we can take to be fundamental. We will do so henceforth.
Let us consider the particular example of a U (N c ) gauge group with N f ≥ N c fundamental hypermultiplets. Then it is well-known that the vacuum equations have N f Nc solutions, essentially differing by the choice of N c out of the N f hypermultiplets to acquire a vev. For positive value of ζ SW , and choosing the first N c hypermultiplets, one solution is given concretely as φ 1 = −diag(m 1 , . . . , m Nc ) and q ja = ℓ −1 √ ζ SW δ ja , for j = 1, . . . , N c , q ja = 0 for j = N c + 1, . . . , N f , and q ja = 0, where a denotes the gauge index and we introduced the standard notations q ≡ q I=1,A=1 , andq ≡ q I=1,A=2 (see also (B.10)), while for negative values of ζ theq get a vev. The U (1) vacua can be obtained as a special case.
To complete the Higgs-like solution, we should still make sure that (2.24) and (2.25) are satisfied.
On the round sphere, their combination (2.20) simplifies to which can be decomposed in terms of the scalars q,q † as and their complex conjugates. It is clear that in the vacuum where (only) q gets a vev, these equations are solved for ζ vac. = +6, while ifq gets a vev, one finds ζ vac. = −6. It is straightforward to verify that then also (2.24) and (2.25) are solved.
Smooth "(m, n)-vortex" solutions Let us now relax the constraint H SW IJ = 0 and study more general smooth solutions with non-zero H SW IJ 9 . We will focus on generalizing the vacuum solutions where ζ is positive and thus q acquires a vev, knowing that the case where ζ is negative andq gets a vev can be treated completely similarly. Let us further denote the deformed Coulomb branch configuration for ζ vac. as A vac. = 1 3ℓ ζ vac. κ, and (φ 2 ) vac. = 1 6ℓ ζ vac. cos ρ. The smooth solutions we are about to uncover carry winding around the circles parametrized by ϕ and χ, and thus have their combined core at the north pole and the south pole. In fact, they resemble the standard two-dimensional vortex solutions, and as in that case, we do not have an analytic expression for the solution, but study its behavior in the far away region, and near the core. It is trivial to verify that away from the north and south pole one has the solutionq = 0 and while φ 1 still takes its vacuum value determined in terms of the masses. This solution is valid in the region ρ 2 ≫ m+n ζ SW +2(m+n) , as we will derive momentarily. To analyze the behavior around the north pole and south pole, we study the vector and hypermultiplet equations to linear order in ρ or π − ρ respectively. The geometry is approximated by flat C 2 . Indeed, around the north pole ρ = 0 and with r 1 = ℓρ cos θ and r 2 = ℓρ sin θ the metric simply becomes The hypermultiplet equations read forq = 0 while the vector multiplet equations can be written as where we omitted terms involving |q| 2 and constant times which assuming smoothness can only contribute at order ρ 2 or higher. Furthermore, we wrote the equations in such a way as to highlight the vortex-like behavior in the planes (r 1 , ϕ) and (r 2 , χ) evident in equations (3.12),(3.18) and (3.12), (3.19) respectively if F χr 1 = F ϕr 2 = 0.
One finds the solution to the full set of equations (3.12)-(3.19) for ρ 2 = r 2 1 +r 2 2 ℓ 2 ≪ m+n ζ SW +2(m+n) to beq = 0 and for some constant B. Note that m and n are necessarily positive and that F χr 1 = F ϕr 2 = 0 indeed.
One can easily estimate the size of these smooth solutions. From (3.10), one can find via Stokes' theorem the vorticity of A−A vac. carried in the (r 1 , ϕ) and (r 2 , χ) planes to be given by −m and −n respectively. Then approximating 1 2 + m + n on disks of radii ǫ 1 and ǫ 2 respectively, one easily estimates ǫ 1 ≈ ℓ m ζ SW +2(m+n) and ǫ 2 ≈ ℓ n ζ SW +2(m+n) . For sufficiently large values of ζ SW the smooth solutions squeeze to zero size, justifying the first order approximations we made.
One can similarly analyze the asymptotic behavior near the south pole. One finds in terms of . We have constructed solutions to the BPS equations in a small neighborhood around the north pole ρ = 0 and the south pole ρ = π. We claim however that for m = 0 the core of the solution (defined as the zeros of the complex scalar field q) in fact wraps the two-sphere S 2 θ=0 defined by θ = 0, and similarly for n = 0 the core wraps the θ = π/2 two-sphere S 2 θ=π/2 . While heuristically clear, such behavior can be argued for rigorously from the full hypermultiplet equations by starting in the core at either north or south pole and verifying that any motion along the relevant twosphere keeps q zero, but we won't do so here. In these cases we thus found a vortex-like object in R 2 wrapping an S 2 . For generic m, n = 0 the BPS configurations are a non-trivial superposition of the m-vortex near S 2 θ=0 and the n-vortex near S 2 θ=π/2 , with the core lying again on these two spheres. The cores overlap at the two intersection points of the two two-spheres, i.e. the north pole and the south pole of S 4 . As we will see momentarily, there are additional point-like solutions to the BPS equations supported at these points.
The approximations made above are valid only in the limit where ζ SW → ∞ and the solutions squeeze to zero size (in the planes carrying winding). For finite values of ζ SW the solutions will require both finite size and curvature corrections. Nonetheless, using the BPS equations, we can deduce important properties of the solutions valid for any value of ζ SW . Namely, since q only vanishes in the core of the solution, we find from (3.9) the exact relation, Moreover, due to the compact nature of S 4 , the winding numbers m, n are not without restrictions; instead, given ζ SW , they are required to satisfy a certain bound, which we now derive. Consider the integral On the one hand, one can substitute the F ± using the BPS equations, and obtain On the other hand, observing that d * dκ = −3ℓ −2 sin ρ − sin θe 1 + cos θe 2 ∧ e 3 ∧ e 4 , it is easy to show that (using the fact that any complex line bundle on S 4 is trivial, and therefore globally Combining the equations (3.26) and (3.27), we have where we used that on the solution ζ ℓ 2 − |q| 2 + |q| 2 ≤ ζ ℓ 2 . Finally, using the exact relation (3.24) which implies that the integrand on the left hand side is constant on S 4 , we obtain a bound on the winding numbers 10 we can introduce ordinary differential forms α ∈ Ω 0,0 J and β ∈ Ω 0,2 J defined as To extract the equations describing the singular configurations, we further split off the vacuum deformed Coulomb branch solution for ζ = ζ vac. = 6, i.e. A = A vac. + a, φ 2 = (φ 2 ) vac. + ∆φ 2 . Then, 10 When instead analyzing the case of negative ζ and thus non-trivialq solutions, and introducing the positive winding numbersm,ñ as inq † ∼ e imϕ+iñχ , one finds the exact relation 2ιRA + φ2 (s − s) = −ℓ −1 (m +ñ + 1) and at the north pole (ρ = 0), the equations become: where the superscriptJ denotes the component proportional to the (1, 1)J formJ. By standard arguments, by combining (3.31) and (3.33), it is easy to show that either α = 0 or β = 0. As for the smooth solutions, we consider solutions with non-trivial α, which trivially implies that ∆φ 2 = 0 (and β = 0). 11 The equations then reduce to the standard Seiberg-Witten equations on C 2 [54], see [55] for a nice introduction. Moreover, we demand that the singular solutions share the same winding numbers with the smooth solutions found above.
In the patch containing the south pole, the anti-self-dual complex structure J is well-defined, see (D.8). It is then straightforward to derive another set of Seiberg-Witten equations at the south pole with respect to this complex structure.
Solutions (α = α(z, w), β = 0) to the Seiberg-Witten equations on C 2 z,w can be constructed from complex algebraic curves, as discussed in [56]. 12 More precisely, given a polynomial p(z, w), there exists a solution to the Seiberg-Witten equations such that α(z, w) vanishes on the zeros of p and such that near its zeros, α looks like the polynomial to leading order. In particular, given a polynomial p (z, w) = We are looking for single-centered solutions with winding numbers matching those of the smooth solutions on top of which the singular Seiberg-Witten solution is defined. It is clear then that the relevant polynomial is given by p(z, w) = z m w n . It will be important later to note that in particular when m = 0 or n = 0 we are dealing with a vortex solution in the w or z-plane respectively. 11 If we had split off the vacuum deformed Coulomb branch solution for ζ = ζvac. = −6, i.e. had considered smooth solutions with non-trivial β, then only (3.32) changes. It becomes (∆φ2 − 2ℓ −1 )α = ∆φ2β = 0, thus again setting ∆φ2 = 0. 12 We would like to thank Clifford Taubes for communication on this point.

Computation of the partition function
To complete the localization computation of the partition function on S 4 , we need to evaluate the classical action on and the one-loop determinant of quadratic fluctuations around the BPS configurations and subsequently sum/integrate over the space of BPS solutions. Since H IJ is introduced through a Q-exact deformation (2.10), all (appropriate) choices of H IJ should leave the partition function invariant. We will see in detail how this expectation works out.
Since the classical action for the hypermultiplet is Q-exact, we only need to evaluate the Yang-Mills action (B.4) and the Fayet-Iliopoulos action (B.7). Furthermore, through an index theorem, the one-loop determinants on S 4 b can be computed straightforwardly [1,22]: where α ∈ g are the roots of the gauge Lie algebra g, w ∈ R are the weights of representation R in which the hypermultiplet transforms, and Q = b + b −1 , with b = ℓ/l. Here we assumed that the (rescaled) equivariant gauge transformation parameter (see (B.16)) evaluated at north and south pole are equal: Since we are working on S 4 , we have ℓ =l, b = 1 and Q = 2. In this section we will keep using ℓ andl indicating how our results are naturally generalized to the squashed sphere at the level of the partition function.

Coulomb branch
The classical actions evaluated on the Coulomb branch solution (3.1) give 3) The gauge equivariant parameter is easily evaluated to beâ = ℓla, which can be plugged into (4.1). Taking into account the point (anti-) instantons at north and south pole, the total partition function can then be written as where Z inst. (x, M, ǫ 1 , ǫ 2 , q) denotes the instanton partition function [2,57], we introduced x = ℓla, and we also included the (rescaled) hypermultiplet masses M . The integrations are along the real line. It is furthermore relevant to mention that the poles of the instanton partition function cancel against the zeros of the vector multiplet one-loop determinant and thus the integrand only has poles originating from the hypermultiplet one-loop determinant. We would like to write the partition function in terms of only the contributions of Higgs branch configurations, i.e. we would like to find a regime of parameters for which the deformed Coulomb branch contribution vanishes in the limit ζ a → ±∞. Using the asymptotic behavior of Υ b (z) for large z, which can be derived from the asymptotics of the double gamma function [58],

Deformed Coulomb branch
where γ is the Euler-Mascheroni constant, and introducing the U (1) charges of the gauge representation R j as q  This divergence arises since we are considering a point on the Coulomb branch where these hyper-multiplets are effectively massless. As explained in [6], the computation of the combined vector and hypermultiplet system is performed by removing the corresponding zero modes via a residue prescription. The total one-loop determinant can thus be written as Reŝ a→â (m,n) HV  then (4.13) precisely contributes the residues of the crossed poles, as anticipated in the previous subsection. Unfortunately, we are not aware of an independent computation of Z HV,z m w n SW,non-pert. (M, b, b −1 , q), so instead we put forward (4.14) as a prediction. As a particular case of (4.14), we find for n = 0 that Z inst. (x HV +i(mb+Q/2), M, b, b −1 , q) equals the non-perturbative piece of the four-dimensional N = 2 supersymmetric m-vortex partition function in the Ω-background.

Higgs branch and Seiberg-Witten Monopoles
The ζ → ∞ limit For ζ → +∞, we found around (4.8) that in favorable cases the deformed Coulomb branch contribution vanishes. Moreover, in this limit the smooth "(m, n)-vortex" solutions were found to squeeze to zero size and their winding numbers are unbounded. The path integral is thus dominated by the squeezed vortex solutions as well as the singular Seiberg-Witten solutions.
We will denote their total resummed contribution in the Higgs vacuum HV rather unimaginatively as Z resummed and an overall classical factor. The latter is given by is the S 2 partition function of the m-vortex worldvolume theory. 17 It has indeed the structure of a Higgs branch localized partition function on the two-sphere, see [5,6,62]. We will see an explicit example in the next section. Similar considerations are valid for m = 0.

Matching with the Coulomb branch integral
In this section we briefly show that Coulomb branch and Higgs branch localization indeed produce the same partition function for the simplest case of a U (1) theory with N f fundamental hypermultiplets. The computation amounts to closing the integration contour in the Coulomb branch integral and computing the residues of the encircled poles. 18 We start with the Coulomb branch integral (4.4) specialized to the U (1) case Here the masses satisfy the relation N f k=1 M k = 0 since the flavor symmetry group is SU (N f ). Furthermore, the instanton partition function is given by [2,57] as a sum over Young diagrams Y . Each Young diagram Y encodes a non-increasing sequence of is the width of the Young diagram. The total number of boxes in the diagram is denoted by |Y |. The instanton counting parameter is given by q = e 2πiτ , with τ = θ 2π + 4πi g YM 2 . The first factor captures the contribution from the hypermultiplets, while the second factor those of the vectormultiplet. We denote them by z HM and z VM respectively. The latter can be simplified by splitting the infinite products over r, s into the four regions r, s ≤ W, and r ≤ W, s > W, and r > W, s ≤ W, and r > W, s > W, where W ≥ W (Y ) is any integer larger than or equal to the width of the Young 17 Considering a four-dimensional N = 2 gauge theory with gauge group U (Nc) and N f ≥ Nc fundamental hypermultiplets, the m-vortex worldvolume theory is given by a two dimensional N = (2, 2) gauge theory with gauge group U (m) and Nc fundamental chiral multiplets, N f − Nc anti-fundamental chiral multiplets and an adjoint chiral multiplet [60,61]. Note that the vortex theory contains vortices itself. 18 The U (N ) generalization of this computation has been considered in [53], which however performs an incorrect truncation of the sum over Young diagrams in the instanton partition function upon plugging in the position of the pole.
diagram. The last region does not contribute, while various manipulations in the other products result in where we use the Pochhammer symbol (y) n = n−1 i=0 (y + i) and already specified the Ω-deformation parameters as ǫ 1 = b, ǫ 2 = b −1 .
As discussed around equation (4.8), the contour in (5.1) can be closed in either upper or lowerhalf plane. We choose to close the contour in the upper-half plane and pick up the poles at 4) which are located at the zeros of the Υ b functions in the denominator. Using the shift formula In total we then find where we should insert (5.3) and (5.6) in the last line. As discussed below (5.6), the sum over Young diagrams Y is effectively truncated to those diagrams whose shape is such that they do not contain a box with coordinates (column, row) = (m + 1, n + 1).
The Special case of n = 0. In the previous section we have argued that for n = 0, Z = after some straightforward manipulations and using in the second line that the masses sum to zero.
Before writing down Z j,z m SW,non-pert. we should first remark that for n = 0 only Young diagrams Y not containing a box with coordinates (column, row) = (m + 1, 1), i.e., satisfying W (Y ) ≤ m, have non-vanishing contributions. We can thus use m as W in (5.3) and (5.6) and obtain we find almost 19 perfect agreement upon identifying 4π and where we denoted the masses in the two-dimensional theory as m for the fundamental chiral multiplet,m for the anti-fundamental chiral multiplet, and m X for the adjoint chiral multiplet. In particular we find that Z j,z m SW,non-pert. precisely equals the vortex partition function of the two-dimensional theory.

Acknowledgements
We would like to thank Armen Sergeev and Clifford Taubes for communication on Seiberg-Witten solutions. We are grateful to Jaume Gomis, Leonardo Rastelli and Martin Rocek for comments on the manuscript. The work of Y.P. and W.P. is supported in part by NSF Grant PHY-1316617.

A Sigma matrices and Spinors
In this appendix we review our conventions, following [22], for spinors and σ-matrices.
Finally, given symplectic-Majorana spinors ξ I orξ I , one can define several useful bilinears, where we used the σ-matrices introduced in the next subsection.
B N = 2 Vector Multiplet and Hypermultiplet N = 2 Killing spinors on Eculidean four-manifolds. As discussed in [22] (see also [23,63] Vector Multiplet. An off-shell N = 2 vector multiplet contains the gauge field A µ , complex scalars φ,φ, (anti-) chiral symplectic-Majorana spinors λ I andλ I , and an SU (2) R triplet of auxiliary fields D (IJ) . Their transformation rules are given by [22] 3) The supersymmetric Yang-Mills action is given by [22] which is positive definite upon imposing the reality properties on the bosonic fields, while one maintains the symplectic-Majorana nature of λ I ,λ I . If the gauge group contains a U (1) factor, one can also introduce a Fayet-Iliopoulos term. Introducing an SU (2) R triplet background field w IJ satisfying one can write the invariant action [22] where Tr FI denotes a trace that weighs each U (1) factor in the gauge group with its own Fayet-Iliopoulos parameter ξ FI .
Hypermultiplet. An off-shell N = 2 hypermultiplet 20 consists of scalar fields q IA , the fermions ψ A ,ψ A , and the auxiliary fields F I ′ A . Here A, B, C, . . . = 1, 2 denote U Sp(2) indices (which is broken to the Cartan upon gauging), and I ′ , J ′ , ... = 1, 2 are SU (2) R ′ indices. Furthermore ψ A and ψ A are Ω-symplectic-Majorana spinors, while q and F (canonically 21 ) have reality properties where Ω 12 = −Ω 12 = −1 is the symplectic form of U Sp (2). Note that the reality property of q implies it can be written as (B.10) The A = 1 and A = 2 components reside in complex conjugate representations R,R of the gauge and/or flavor group G. A hermitian generator T ∈ g in representation R acts on any field K A as and thus an adjoint field Ξ = Ξ a T a acts as The supersymmetry transformation rules are [22] where the extra symplectic-Majorana spinors ζ andζ satisfy 14) The supersymmetric action for the hypermultiplet is Q-exact [22] on the ellipsoid and thus does not play a role in the localization computations of this paper. One should remark though that it is only positive definite upon choosing the alternative reality properties for the auxiliary fields Supersymmetry algebra. The supersymmetry algebra takes the form where L A+V +V R denotes a gauge, SU (2) R and SU (2) R ′ -covariant Lie derivative along the vector field R, and SU (2) R ′ rotates the hypermultiplet auxiliary fields. The parameters are given by

C Killing Spinors And Complex Structures
In this subsection we introduce almost complex structures whose existence is guaranteed by having a solution to the generalized Killing spinor equations (B.1) and the auxiliary equations (B.2). They will turn out to be useful when analyzing the singular solutions to the BPS equations.
The interplay between supersymmetry and geometry is quite rich, as for example observed for four-dimensional theories with four or fewer supercharges in [64][65][66][67][68]. It is clear that we are only scratching the surface here, and a more in depth analysis would be very interesting.

C.1 Locally Almost Complex Structures
Let ξ I andξ I to be the solutions to the generalized Killing spinor equations (B.1) and the auxiliary equations (B.2). Then given any symplectic-Majorana spinor χ I andχ I such that (ξ I χ I ) = 0, (ξ I χ I ) = 0, one can define two almost complex structures away from the zeros of ξ andξ where s χ ≡ χ I χ I , sχ ≡ (χ Iχ I ) are both positive semi-definite. Using Fierz identities, it is easy to check that Where ξ is non-zero, one can write χ I = m I J ξ J or χ I = m µν σ µν ξ I and similarly forχ, where m IJ is a triplet of functions satisfying m IJ = ε II ′ ε JJ ′ m I ′ J ′ , and m µν is a real anti-self-dual 2-form.
The two representations are interchangeable, for instance, m IJ and m µν are related by In the following, when we need to, we pick the representation of χ I using m IJ , and similarly for On open sets where the locally almost complex structures are defined, one can introduce the decomposition of tangent vectors with respect to J andJ respectively: A (p, q)-type vector can be characterized using spinorial equations. First of all, taking JX = iX as an example, Multiplying X µ to the two equations on the right, and subsequently taking their sum, it is easy to verify that one obtains a semi-positive product I,α ∆α I ∆α I = X µ s −1 χ χ I σ µν χ I + 2is −1/2 s −1/2 χ χ I σ µν ξ I + s −1 ξ I σ µν ξ I X ν 0 , (C. 6) where ∆α I = X µ (σ µ )α γ s −1/2 χ χ Iγ + is −1/2 ξ Iγ . Therefore, Similarly, one can derive the spinorial condition for X to be (p, q)-vector of J orJ:

C.2 Integrability
It is possible that the almost complex structures induced by Killing spinors are integrable. In the following, we consider χ I = m I J ξ J and study the conditions for J to be integrable.
In [22], a solution to the generalized Killing spinor equations (B.1) and the auxiliary equations The corresponding explicit expressions for the auxiliary fields T µν ,T µν , S µν ,S µν , V µ and M can be found in [22].
(D. 8) where J is defined away from the north pole (ρ = 0), andJ is defined away from the south pole. One can verify that the conditions (C.17) are satisfied and thus that they are integrable.