The supersymmetric index of the 4D theory realized by a brane tiling coincides with the partition function of an integrable 2D lattice model. We argue that a class of half-BPS surface defects in brane tiling models are represented on the lattice model side by transfer matrices constructed from L-operators. For the simplest surface defects in theories with flavor groups, we identify the relevant L-operator as that discovered by Sklyanin in the context of the eight-vertex model. We verify this identification by computing the indices of class- and - theories in the presence of the surface defects.
Remarkable connections have been uncovered in the past several years between supersymmetric field theories and integrable lattice models [1–9]. A prominent example is the correspondence [1–3,7] between 4D quiver gauge theories realized by certain brane configurations in string theory, known as brane tilings [10–12], and the 2D lattice models that Bazhanov and Sergeev [13,14] constructed using elliptic hypergeometric integrals discovered by Spiridonov [15–19]. In this correspondence, the supersymmetric index of a brane tiling model is equated with the partition function of a lattice model, and the integrability of the latter follows from Seiberg duality.
In this paper we study a class of half-BPS surface defects in 4D theories from the perspective of the above correspondence. We argue that a surface defect in this class is represented on the lattice model side by a transfer matrix, an object which we depict as
Since the results obtained in the present work bridge two areas of physics in a way that may be unfamiliar to many readers, in this introduction we provide a somewhat detailed overview.
To begin with, let us briefly review where the correspondence between brane tilings and integrable lattice models comes from, following the discussion in Ref. . We will give a more thorough explanation in Sect. 2.
In fact, the correspondence in question is a combination of two correspondences: one between brane tilings and 2D topological quantum field theories (TQFTs) equipped with line operators, and another one between 2D TQFTs with line operators localized in extra dimensions and integrable lattice models [4,6].
The first correspondence has its origin in six dimensions. A brane tiling model is constructed from a stack of D5-branes on , intersected by NS5-branes that occupy and, roughly speaking, are supported on curves in the Riemann surface . From the view point of the 6D theory living on the D5-branes, the NS5-branes create codimension-1 defects or domain walls. At low energies, the 6D theory compactified on in the presence of these codimension-1 defects is described by a 4D theory. Under nice circumstances, it is a gauge theory characterized by a quiver diagram with nodes drawn on , where is the number of the D5-branes.
The situation is therefore similar to the construction of 4D theories of class [20–22]. A class- theory is defined by compactification of a 6D superconformal field theory on a Riemann surface in the presence of codimension-2 defects. The locations of these defects (“punctures”) in the surface are parameters of the theory, and generally, physical quantities depend on them. If, however, we place the theory on and compute its partition function, the result is the supersymmetric index [23,24] which is a protected quantity independent of continuous parameters and hence determined by the topological data of the punctured surface. It follows that the index of a class- theory is captured by a correlation function of local operators in a TQFT on the associated surface .
By the same logic, the index of a brane tiling model coincides with a topological correlator on , albeit of line operators in another TQFT. We get a different TQFT because of the different 6D origin, namely 6D super-Yang–Mills theory instead of a 6D theory, and line operators rather than local ones since our defects are of codimension-1 and not of codimension-2.
The second correspondence relates TQFTs and integrable lattice models defined on the same surface . Nevertheless, a higher-dimensional point of view is also crucial here.
Given a configuration of line operators in a TQFT on , we can express its correlation function in the form of the partition function of a lattice model. Generically, these operators form a lattice with no three lines meeting at a point. Therefore, we can cut into square pieces in such a way that each of them contains a crossing of two lines,
This fact may not be noteworthy—it is merely a rewriting of the path integral—were it not for the following observation by Costello [4,6]: the lattice model thus obtained would be integrable if there exist extra dimensions in the TQFT.
Thus, a correlation function of line operators in a 2D TQFT with extra dimensions is equal to the partition function of an integrable lattice model. The model is defined on the lattice formed by the line operators, whose coordinates in the extra dimensions provide the spectral parameters.
We have to ask whether the 2D TQFT arising from brane tiling models has desired hidden extra dimensions. It actually has one: the 11th dimension that emerges when the brane system is embedded into M-theory via string dualities. Consequently, the index of a brane tiling model is the partition function of an integrable lattice model.
The main theme of this paper is to incorporate surface defects into the above story of connections between brane tilings, TQFTs and integrable lattice models. We address this question in Sect. 3.
In order to create a surface defect in our 4D theory, we add to the brane system a D3-brane supported on a plane in and ending on the D5-branes along a curve in . The total brane configuration preserves supersymmetry on the plane. Inside the 4D theory, the D3-brane appears as a half-BPS surface defect. This is the most basic example of a class of half-BPS surface defects, all of which admit a similar, if slightly more elaborate, brane construction. Surface defects in this class are specified by a representation of . The one just described corresponds to the fundamental representation.
Since the D3-brane ends on the D5-branes along a curve in , it creates a line operator in the 2D TQFT. In the lattice model, the introduction of the surface defect is therefore translated to insertion of an extra line, which we represent by a dashed line:1).
With lines of a different type in hand, we can write down different versions of the Yang–Baxter equation (4). In particular, we have the relation2) and the R-operator (3). It is called an RLL relation.
It turns out that when the dashed line is labeled with the fundamental representation of , the above RLL relation was studied by Derkachov and Spiridonov in Ref. . According to their work, an L-operator that solves the RLL relation is essentially Sklyanin's L-operator , a matrix whose entries are difference operators acting on meromorphic functions. This observation allows us to infer that in this simplest case, the L-operator (2) for our theory is Sklyanin's L-operator. If this is true, one consequence is that the integrable model realized on a lattice consisting solely of dashed lines should be the eight-vertex model [28,29].
As the Yang–Baxter equations do not determine the L-operator uniquely, it is important to check our proposal by comparing it with independent computations from the gauge theory side. Fortunately, such checks can be performed, as we do in Sects. 4 and 5.
The brane tiling models we mainly consider in this paper are also examples of theories of class [30–33], which are generalizations of class- theories. As such, their indices in the presence of surface defects can be computed by the method developed in Refs. [30,34].
Briefly, the procedure goes as follows. A class- theory of type arises from M5-branes placed on a orbifold singularity and further compactified on a punctured surface. To this surface we introduce an extra puncture carrying a flavor symmetry (known as a “minimal” puncture). The addition of the puncture modifies the 4D theory. The index of the new theory has a series of poles in the fugacity parameter associated with the flavor symmetry of the puncture. These poles are labeled with a pair of nonnegative integers, and the residue at the pole encodes the index of the original theory in the presence of two surface defects, supported on different tori inside and labeled with the th and th symmetric representations, respectively.
String dualities map the class- setup to the brane tiling setup. Under this map, the addition of a minimal puncture corresponds to the introduction of an NS5-brane. In turn, the latter operation inserts a lattice line in the lattice model. Taking the residue converts this line to a dashed one representing the surface defect. Incidentally, the integrability of the lattice model is nothing but the statement that the index is invariant under interchange of the positions of minimal punctures.
We carry out the residue computation for and , first for (i.e., for class- theories) in Sect. 4, then for general in Sect. 5. In each case, the index in the presence of the surface defect is obtained by letting a difference operator act on the bare index, in the absence of the surface defect. We find that this operator precisely matches the corresponding transfer matrix calculated based on our proposal.
A reader familiar with the class- story may ask the following question: The residue method described above produces distinct difference operators for a given pair . How can they all be accommodated in a single transfer matrix? The answer is that although there is only one transfer matrix, it carries a continuous parameter, namely the spectral parameter of the dashed line. The difference operators are unified into this one-parameter family as values of the spectral parameter.
While we focus on a particular class of surface defects in this paper, 4D theories have many other aspects that should be equally illuminated by the integrability structure. In this sense, the present work may be regarded as a first step in the broader program of studying 4D theories through integrability. In Sect. 6 we suggest a couple of possible directions to be taken for next steps. Clearly, though, they are only a tiny fraction of the long list of interesting topics for future research in this ambitious program.
2. Brane tilings and integrable lattice models
In this section we review the correspondence between 4D theories realized by brane tilings and integrable 2D lattice models. A central role is played by a 2D TQFT equipped with line operators that are localized in a hidden extra dimension emerging from M-theory. Our presentation follows Ref. , to which we refer the reader for more details.
2.1. Quiver gauge theories and their supersymmetric indices
Throughout our discussion we will encounter gauge and flavor groups that are either or .1 To each such group, we assign fugacities parameterizing the maximal torus. For instance, an element in the maximal torus of an group takes the form , hence is a set of fugacities for this group, obeying the constraint . Fugacities are also used to label the groups themselves; thus is an gauge or flavor group whose associated set of fugacities is . The quiver diagrams we will deal with involve gauge and flavor nodes. Since is fixed in each quiver, we label the nodes with the fugacities of the corresponding groups, rather than the rank: is a gauge group , while is a flavor group .
Building blocks of 4D supersymmetric quiver gauge theories are vector multiplets and bifundamental chiral multiplets. A vector multiplet is present at a gauge node. A bifundamental chiral multiplet has two flavor groups, say and , and transforms in the fundamental representation under and in the antifundamental representation under . We represent it by an arrow going from to (if and are gauged). In general, a bifundamental chiral multiplet is also charged under an R-symmetry group , which we assume to exist and be anomaly-free, and under additional flavor groups . When we need to indicate the charges under these symmetries, we mark the arrow with
Given a theory with flavor group and another theory with flavor group , we can couple them to obtain a new theory by gauging the diagonal subgroup of . To construct a quiver gauge theory, we take a number of bifundamental chiral multiplets and couple them by gauging all or part of the flavor nodes.
Thanks to supersymmetry, the index (8) receives contributions only from those states whose energies belong to a certain discrete spectrum determined by the R-charge assignment. As a result, it remains invariant under continuous changes of the parameters of the theory. This protected nature of the index will be important for our argument.
The index of a 4D theory with flavor group is a symmetric meromorphic function of the fugacities , , . The symmetricity property reflects the gauge invariance of the index. At the level of the index, gauging of a flavor group is realized by introduction of the corresponding vector multiplet and integration over its fugacities. In particular,
We can readily write down the formula for the index of a general quiver gauge theory. For simplicity, suppose that the theory is described by a quiver that contains no flavor node. Then, the index is computed by
2.2. Supersymmetric index and integrable lattice models
The supersymmetric index (17) of a quiver gauge theory may be interpreted as the partition function of a statistical mechanics model with continuous spins. Indeed, this formula precisely computes the partition function of a spin model in which spins are placed at the gauge nodes. The spin variables at are the fugacities , , , and they interact among themselves as well as with spins at nearest-neighbor nodes, namely those connected by arrows. The Boltzmann weights for the self-interaction and the nearest-neighbor interaction are and , respectively.
This is not particularly surprising in view of the fact that the index is a protected quantity and can be computed in the free theory limit. In this limit, vector and bifundamental chiral multiplets decouple, so their contributions factorize. What is remarkable is that for a certain class of theories, the index is equal to the partition function of an integrable model defined on a 2D lattice.
The connection between the supersymmetric index and the lattice model comes from higher dimensions. Consider a 6D supersymmetric theory equipped with codimension-1 defects. (Here we have in mind the 6D theory living on a stack of D5-branes, though our argument is more general.) We compactify this theory on a two-manifold and place codimension-1 defects along various curves in . Suppose that this kind of configuration preserves four supercharges for any choice of and . Then, at low energies, the system is described effectively by a 4D theory . We can place it on and perform the path integral to compute its index. The index is invariant under continuous changes of the parameters of the theory, and the geometric data of the curves are such parameters. As a consequence, this procedure defines a map from the set of topological configurations of curves to the set of supersymmetric indices, given a choice of codimension-1 defects.
Now start again from the same 6D theory, placed on the space-time , with the same configuration of defects. In the previous paragraph, it was implicitly assumed that the size of is much smaller than and so that the description by the 4D theory is sensible. This time, let us make much larger than ; the index remains invariant under the rescaling of the metric of . In this case, the low-energy physics is described instead by a 2D theory on , and the codimension-1 defects inserted on become line operators supported along in this theory. This consideration leads to the relation
We can compute the above correlation function by dividing into square pieces, each containing segments of two line operators crossing in the middle:34
We will mainly study the case where is either a cylinder or torus and line operators form a square lattice. To be specific, let us take and wrap line operators around 1-cycles , , , making up an lattice. We divide the torus into square pieces as above, and to each side of these squares assign a variable that labels basis vectors for the state space on that side. The situation for is illustrated in Fig. 1. For the computation of the correlation function, first we take the product of the matrix elements of from all squares for each configuration of state variables, then sum over all configurations:
Let us shift our perspective slightly and look at the operator as assigned to the vertices of the lattice, not the square pieces containing them. Also, we think of the state variables as living on the edges of the lattice, not the sides of the squares. If we view the system in this way, the above formula is precisely the partition function of a vertex model in statistical mechanics: spins live on the edges of a lattice, and interact at the vertices with the Boltzmann weight . In the context of vertex models, the S-matrix is known as the R-matrix or R-operator.
So far we have discussed the connection between the supersymmetric index and vertex models. We now explain how integrability comes into the picture.
To talk about the integrability of a vertex model, we should consider the situation that each lattice line carries a continuous parameter, called the spectral parameter assigned to that line. Correspondingly, the R-operator depends on two spectral parameters in general:5
Therefore, for the vertex model constructed from line operators in a 2D TQFT to be integrable, the line operators must carry spectral parameters, and transfer matrices must commute. These two features arise naturally if there are extra dimensions in the theory.
Suppose that our 2D TQFT is really a higher-dimensional theory compactified on some manifold , and the line operators in the correlation function (23) are placed at some points in , which we assume for simplicity to be all different. Suppose also that the correlation function is topological on , but varies nontrivially along . Then, the lattice lines carry continuous parameters, namely their locations in . Furthermore, the commutativity of transfer matrices holds:
Let us recapitulate the logic of our argument. We consider a 4D theory that is constructed from a 6D theory by compactification on a two-manifold , in the presence of codimension-1 defects supported on curves in . Due to its protected nature, the supersymmetric index of the theory is captured by the correlation function of a lattice of line operators in a 2D TQFT on . In turn, by dividing into square pieces, the correlation function can be mapped to the partition function of a vertex model defined on the same lattice. This vertex model is furthermore integrable if the 2D TQFT has hidden extra dimensions along which the correlation function varies nontrivially.
It is clear that the above argument applies to any protected quantities, not just the supersymmetric index on . For example, we can use the index on with the 3-manifold different from . For each protected quantity, there is a corresponding TQFT and hence an integrable lattice model. The case when is a lens space was investigated in Ref. . In this paper we focus on the index since this is a well-understood quantity and, accordingly, there are nice mathematical results available.
2.3. Brane tilings
Now we turn to a specific class of 4D theories that have the desired properties described above. These theories are constructed using branes in string theory.38].
Let us replace the 46-plane with an arbitrary Riemann surface . To preserve supersymmetry, we take the background space-time to be and place the D5-branes on , where is the 4567-space and is embedded in it as the zero section. NS5-branes intersect the D5-branes along curves . More precisely, they are placed on , where are surfaces in such that they restrict to on . Provided that are chosen appropriately, this system preserves 4 supercharges.7
On the D5-branes lives 6D super-Yang–Mills theory with gauge group . The theory is placed on and topologically twisted along , as can be seen by noting that two of its four scalar fields describing fluctuations of the D5-branes are not really scalars, but rather sections of . Thanks to the twisting, 8 of the 16 supercharges are left unbroken by the curvature of . In this 6D theory, the NS5-branes create half-BPS codimension-1 defects, or domain walls, supported on . (The four supercharges preserved by the two types of NS5-branes are compatible with the twisting .) If is compact, the 6D theory is effectively described by a 4D theory. We call 4D theories constructed in this way brane box models .
We are now in the situation considered before: we have a 6D theory that produces 4D theories by compactification in the presence of codimension-1 defects. By following the same logic, we conclude that the index of a brane box model is given by a correlation function of line operators in a 2D TQFT, and coincides with the partition function of a lattice model.
Furthermore, an extra dimension emerges if the brane system is embedded into M-theory. For the computation of the index, we take the space-time of the 4D theory to be . Thus, we are considering type IIB string theory on . We can apply T-duality along the and lift the resulting type IIA system to M-theory. In this process, the D5-branes are transformed to M5-branes wrapping the 11th dimension, the M-theory circle. On the other hand, the NS5-branes become M5-branes supported at points on the circle. Hence, the M-theory circle provides the extra dimension along which NS5-branes, or line operators in the 2D TQFT, can avoid one another. The existence of the extra dimension implies that the lattice model is integrable.
In a brane box model, the NS5-branes intersect the D5-branes along curves . We can resolve the intersections to trivalent junctions. Each junction connects D5-branes, a single NS5-brane, and their bound state; in the terminology of 5-branes, they are , and 5-branes, respectively. Upon this resolution, the intersection curves spilt into pairs of curves representing the 5-brane junctions. See Fig. 2 for illustration. These curves, which separate into regions supporting different values of the 5-brane charge , are called zigzag paths. We orient them in such a way that increases by 1 as we cross a zigzag path from left to right.
We can consider more general configurations of zigzag paths, not necessarily those obtained by resolving D5–NS5 intersections. Each configuration encodes a 5-brane system: the NS5-branes approach the D5-branes from transverse directions, meet them along the zigzag paths, and together make bound states over some regions. Such a brane configuration is known as a brane tiling. Provided that the NS5-branes wrap appropriate surfaces in , a brane tiling configuration preserves 4 supercharges.
If the string coupling is strong enough, the tension of D5-branes is much larger than that of NS5-branes. In that situation, the shape of the D5-branes is unaffected by the NS5-branes, which simply make 90 degree turns when they hit the D5-branes. (Away from the D5-branes, the NS5-branes wrap the same kinds of supersymmetric cycles as in the case of brane box models.) Hence, the 6D theory on the D5-branes may be regarded as formulated on the fixed space-time , irrespective of the precise configuration of the NS5-branes. From the point of view of the 6D theory, the latter branes create codimension-1 defects supported on the zigzag paths.
For compact , the 6D theory in the presence of these defects is described at low energies by a 4D theory. This construction therefore defines a map from brane tilings on compact Riemann surfaces to 4D theories. Composing it with the supersymmetric index, we get a 2D TQFT equipped with line operators and the associated lattice model.
As before, an extra dimension emerges via embedding into M-theory, implying integrability of the lattice model. Accordingly, each zigzag path naturally carries a circle-valued spectral parameter which is the coordinate of the corresponding M5-brane. If has period , then is identified with a flavor fugacity in the index of the 4D theory. (Flavor fugacities are often analytically continued to complex parameters.) The relevant flavor symmetry comes from the gauge symmetry on the NS5-brane. Via the boundary condition on the brane junction, this symmetry is related to a gauge symmetry on the D5-branes, which gets frozen at low energies and becomes a global symmetry in the field theory.
2.4. Integrable lattice models from quiver gauge theories
In order to actually write down the R-operator of the integrable lattice model arising from brane tilings, we need to know more precisely what 4D theory results from a given configuration of zigzag paths. The answer is known when there is no region supporting 5-brane with on .8 In this case, the 4D theory is a quiver gauge theory.
The rule for reading off the quiver is as follows [11,12,40]. We indicate regions with by dark shading and those with by light shading. Regions with are left unshaded. On an unshaded region there lies an node, produced by open strings attached on this region. This is a flavor node if the region contains part of the boundary of , and otherwise a gauge node. (We allow to have boundary components where the D5-branes end separately on D7-branes.) A crossing of two zigzag paths gives rise to a bifundamental chiral multiplet, produced by open strings that start from one unshaded region and end on another:Sect. 2.3, there is a flavor symmetry for each zigzag path. We choose the convention that the above arrow has charge −1 and +1 under the flavor symmetries and associated with the horizontal and vertical zigzag paths, respectively.
Since every arrow is oppositely charged under two different flavor symmetries coming from zigzag paths, the diagonal combination of all flavor symmetries associated with zigzag paths acts on the theory trivially. Therefore, the zigzag paths provide as many flavor symmetries as their number minus 1, and these generate the nonanomalous flavor symmetries of the theory .
The R-charge is not uniquely determined since it can be shifted by flavor charges. From the point of view of the index, the shift amounts to a redefinition of flavor fugacities by some factors. That said, the R-charge assignment is constrained by two conditions.9 From zigzag paths bounding a shaded region, we get a sequence of arrows making a loop. For example, in our brane tiling we may have a configuration of zigzag paths shown in Fig. 3a. For each such loop, worldsheet instantons induce a superpotential term given by the product of the bifundamental chiral multiplets, with sign determined by the orientation of the arrows. Thus, the R-charges of the arrows must add up to 2. Likewise, from zigzag paths bounding an unshaded region, we get arrows starting from or ending at a gauge node, as in Fig. 3b. For to be free of anomaly, the sum of the R-charges of the arrows must equal the number of the arrows minus 2.
As already said, we would not be able to compute the supersymmetric index without detailed knowledge of the theory. For the purpose of identifying the integrable lattice model, we should therefore study brane tilings in which no regions with appear, at least as a first step of more general analysis. However, even if we do restrict to that case and identify the lattice model, it is not possible to check the integrability directly using the index formula (17). Unfortunately, the Yang–Baxter equation for three zigzag paths always involves regions with .Fig. 4. Note that our R-charge assignment satisfies the two constraints described above.
The R-operator for the corresponding lattice model is given by the supersymmetric index of the quiver (37), and depends on two pairs of spectral parameters. This is in fact the lattice model discovered by Bazhanov and Sergeev in Ref. . The vector space supported on a line is the space of symmetric meromorphic functions of complex variables satisfying the constraint . The variables are to be identified with the fugacities for the node on that line. For example, is the space of symmetric meromorphic functions of the variables or in the above diagram.
Plugging this R-operator into the Yang–Baxter equation (32), we see that the integrability of the lattice model is the statement that the indices of two quivers are equal:42] cyclically four times to the three gauge nodes turns the quiver on the left-hand side to the one on the right-hand side . Fig. 5. The lattice model thus obtained is an interaction-round-a-face (IRF) model, for which spins are assigned on the faces of the lattice. In the open string picture (20), all physical degrees of freedom are localized at the ends of strings as “Chan–Paton factors.” are
The Yang–Baxter equation is simpler for this R-operator. After canceling some factors and using the identity (13), we find that the equation reduces to the following form:16,43] obeyed by the elliptic gamma function, as pointed out in Ref. . Note that we obtained Seiberg duality for flavors from the Yang–Baxter equation, even though the quiver for the lattice model has flavors for each gauge group. This is necessary: the duality transformation would change the rank of the gauge group if the number of flavors were different from .
3. Surface defects as transfer matrices
Having understood how the supersymmetric indices of brane tiling models give rise to integrable lattice models, we now discuss the lattice model realization of a class of half-BPS surface defects in the 4D theories. We will see that these surface defects are mapped to transfer matrices constructed from L-operators. In the simplest case, we will identify the concrete form of the relevant L-operator.
3.1. Surface defects and L-operators
For the sake of clarity, let us go back to the brane box configuration (34) and explain the construction of these surface defects in this situation; adapting the construction to brane tilings is straightforward. To this configuration we add D3-branes:Fig. 6). Out of the 4 supercharges preserved by the other branes, they preserve the half that generate supersymmetry on the 01-plane. From the point of view of the 6D theory on the D5-branes, they create a codimension-3 defect. In the 4D theory obtained by compactifying the 46-plane, this is a codimension-2 or surface defect.
In the absence of NS5-branes extending along the 012367 directions, this brane configuration is related, via T-duality along the -direction, to the familiar configuration of D2-branes creating a surface defect  in a 4D gauge theory realized by a D4–NS5 system . If those NS5-branes are present, the T-duality converts them to an orbifold which breaks the supersymmetry to . Still, our setup is locally identical to the case, and we can rely on various results that have been obtained in that context.
Instead of letting the D3-branes extend indefinitely along the -axis, we can make them end on an NS5-brane that spans the 012345 directions and is located at some (Fig. 7a). The number of D3-branes can be any integer, so this configuration may be thought of as corresponding to a symmetric representation of . Hence, we label the surface defect created by this configuration of D3-branes with the th symmetric representation .
Rather than the above NS5-brane, we may also introduce an NS5-brane extending along the 014589 directions and have the D3-branes end on it (Fig. 7b). In this case, due to the fermionic nature of D-branes, the other ends of the D3-branes must attach to separate D5-branes (“s-rule”). Thus, cannot exceed . Furthermore, if we pass the NS5-brane to the other side of the D5-branes, by the Hanany–Witten transition we obtain a similar configuration with D3-branes. This means that we can label the corresponding surface defect with the th antisymmetric representation of . This configuration is dual to that for a Wilson line in the antisymmetric representation in 4D super Yang–Mills theory [47–49].
We can generate many more surface defects by taking products of these basic ones. It is known that they form a class of surface defects classified by irreducible representations of . The brane configuration for a surface defect labeled with a general representation , after a slight deformation, looks as in Fig. 8 [48–50].11 Dual configurations realizing line operators in three dimensions were considered in Ref. .
In the above construction, we may replace the 46-plane with any Riemann surface and let the D3-branes end on the D5-branes along curves in . The NS5-branes can also take more general configurations representing a brane tiling.
Now that we have a 4D theory with a half-BPS surface defect, we can place it on and compute its supersymmetric index. Assuming that the system flows to a conformal fixed point, we can do this by conformally mapping the Euclidean space-time (minus the origin) to , and then compactifying the radial direction . After the mapping, the surface defect wraps . The first factor may be taken to be either or in the parameterization of . These are the only circles in that are left invariant under the action of the isometry group , for the orbit of a point outside these circles is two-dimensional.
The index in the presence of the surface defect is again given by a correlation function of line operators in a 2D TQFT on . The difference is that this time, the correlator contains a new line operator created by the D3-branes ending on the D5-branes. We represent it by a dashed arrow:
In any case, the correlation function equals the partition function of a lattice model whose lattice is made of two kinds of lines, zigzag paths coming from NS5-branes and the dashed line coming from the D3-branes. An extra dimension emerges as the M-theory circle if the brane system is embedded in the M-theory via T-duality along the second factor. Under this embedding, the D3-branes are mapped to M2-branes supported at points on the M-theory circle. Thus, the inclusion of the dashed line does not spoil the integrability of the lattice model.
The position of the M2-branes on the M-theory circle (which is also the position of the M5-brane on which they have one end) provides a spectral parameter for the dashed line. From the viewpoint of the theory on the D3-branes, this is the holonomy around the second of the dual gauge field for the diagonal subgroup of the gauge group. For the theory on the D2-branes obtained by T-duality along the -direction, it is the holonomy of the gauge field dual to a periodic scalar.
We denote by the vector space for a dashed line labeled . At least when one of the representations is trivial, , it is natural to expect that this space is isomorphic to the representation space of for the following reason. Under the M-theory embedding, the D5-branes become M5-branes and support the 6D superconformal theory of type , placed on . It is known that a BPS sector of the 6D theory compactified on is equivalent to Chern–Simons theory with gauge group [54–59]. The D3-branes, on the other hand, become M2-branes and create a half-BPS codimension-4 defect supported on , where is the curve along which the D3-branes are attached on the D5-branes, and is a point on the M-theory circle. In the Chern–Simons theory, this defect reduces to a line operator labeled . This is a Wilson line operator in the representation , which may be thought of as the worldline of a heavy charged particle whose Hilbert space is . Thus we expect
Let us ask how the introduction of the surface defect is represented on the lattice model side. Consider a general brane tiling configuration (which may or may not have a quiver description), and suppose that the D3-branes end on the D5-branes along a loop in . Due to the periodic boundary condition, the dashed line crosses zigzag paths coming from the right as many times as those coming from the left. By deforming these zigzag paths near the dashed line, we can always make the two cases occur alternately. Then, the neighborhood of the dashed line looks like49) created by the surface defect is the transfer matrix
We emphasize that the surface defect is represented by an object that is defined locally near the dashed line. Hence, the same dashed line acts on the indices of any brane tiling models in the same way, as long as the neighborhoods of the dashed line in the respective models are topologically equivalent and the spectral parameters match. This locality holds even when we couple a brane tiling model to an arbitrary 4D theory by gauging appropriate flavor groups: to compute the index of the combined theory in the presence of a surface defect, we can first let the corresponding transfer matrix act on the index of the brane tiling model, and then couple the result to the index of the other theory by formula (9). This property is a consequence of “associativity” of the gauging operation. The surface defect considered here may be thought of as a 2D theory coupled to a 4D theory. To insert it in the combined 4D theory, we may first couple the 2D theory to a brane tiling model and then couple the resulting 2D–4D system to the other 4D theory.
3.2. Fundamental representation of
Let us consider the simplest interesting setup where we have two D5-branes and a single D3-brane, and identify the concrete form of the transfer matrix (49) in this case. For , and 5-branes are related by an transformation of type IIB string theory. Therefore, we can go to a duality frame in which the transfer matrix only involves either and regions, or and regions. The two cases are on an equal footing, and in fact related in a simple way, as we will see. We first consider the transfer matrix in the background.
We denote the L-operator in this case by since it is the operator that arises when a dashed line is inserted in a brane tiling model described by the diamond quiver constructed from the R-operator (37). In the situation under consideration, the gauge group of a brane tiling model is a product of groups, and the surface defect is labeled ; the D3-brane wraps the circle in . Thus, is the space of meromorphic functions such that , and . Accordingly, we can represent as a matrix whose entries are operators acting on functions in . The R-operator is a matrix. These operators, together with the R-operator , satisfy the Yang–Baxter equations (32), (52), (53), and (54).
Sklyanin constructed  an L-operator that solves the RLL relation
The RLL relations actually admit more degrees of freedom than just the overall normalization. For example, we can multiply by a function of its spectral parameters, and the result still solves the RLL relations. In Sects. 4 and 5 we will check our proposal by comparing it with independent computations from gauge theory.
So far we have treated the surface defect labeled . Of course, we may also consider the case with in the same manner, by letting surface defects wrap around the other inside . Hence, there are two sets of L-operators related by the symmetry exchanging and . The underlying algebraic structure is the product of two copies of the Sklyanin algebra, known as the elliptic modular double .
3.3. Relation to the Bazhanov–Sergeev model
The reason that we introduced the thick line (36) by pairing up two zigzag paths was that brane tiling diagrams constructed using this line do not contain regions supporting 5-branes with . If those regions are present, in general we do not have a description of the 4D theory in terms of a quiver and hence cannot use the formula (17) for the supersymmetric index. Since the Yang–Baxter equation for three zigzag paths always involves undesirable regions, simply restricting ourselves to the quiver case is not sufficient for checking the integrability of the model explicitly.
The situation is different when the number of D5-branes, . In this case, any 5-brane falls into one of two equivalence classes under the duality of type IIB string theory: either or 5-brane, which we may visualize as an unshaded or shaded region. Every unshaded region generates an gauge or flavor group. This fact raises the hope that a general brane tiling with leads to a quiver gauge theory. If so, there should be the corresponding integrable lattice model whose R-matrix is made out of the bifundamental factor .
For brane tilings on flat surfaces, such an integrable lattice model was indeed discovered by Bazhanov and Sergeev in Ref. . Given a brane tiling, we can map it to the Bazhanov–Sergeev model as follows. First of all, we assume that we can deform the zigzag paths so that each of them heads either upward or downward and its slope is never zero (taking the -direction as horizontal and the -direction as vertical, say). With this assumption, the orientations of zigzag paths are actually irrelevant for the lattice model, so we omit them from the brane tiling diagram. Then the diagram consists of two building blocks, and we assign quivers to them:40) for bifundamental chiral multiplets with R-charge . Note that is symmetric under exchange of and , as is consistent with the fact that arrows are unoriented. A23) and expresses the RG flow from SQCD with three flavors to the infrared theory: 37) and (42).
Similarly, the RLL relation (52) follows from two Yang–Baxter equations involving a dashed line, namely
Following Ref. , let us define an operator by
As mentioned already, this R-operator satisfies the Yang–Baxter equation thanks to the star-triangle relation (82). In the operator form used here, the last relation arises naturally from the Bailey lemma proved in Ref. . Its higher-rank generalization  leads to a web of dualities connecting 4D quiver gauge theories .
4. Surface defects in theories of class
Now we aim to check our proposal on surface defects and transfer matrices by comparing it with independent computations. In this section we perform the simplest such check for surface defects in theories of class [21,22], which arise from compactification of the 6D theory of type on punctured Riemann surfaces. The action of surface defects on the supersymmetric indices of class- theories have been studied before [34,46,51,52]. Here we review the computation for the surface defect labeled with the fundamental representation of based on the method developed in Ref. , and show that the result agrees with the prediction from the transfer matrix (74).
4.1. linear and circular quiver theories
Prototypical examples of class- theories are gauge theories characterized by linear and circular quivers with nodes. They are actually also examples of brane tiling models discussed in the previous sections. As such, they allow us to translate key notions in class- theories to the language of brane tilings, and vice versa. Our first task is to describe these theories as class- theories as well as brane tiling models, and understand the relation between the two descriptions. Although we are mainly interested in the case with , for now we keep general.
Let us consider the standard type IIA brane configuration for an linear quiver theory with nodes. It consists of D4-branes spanning the 01236 directions, intersected by NS5-branes extending along the 012345 directions:Fig. 9 illustrates the correspondence between the quiver and the sphere.
The R-symmetry of the theory is , where originates from the rotational symmetry of the 789-space, and from the rotational symmetry of the 45-plane. The flavor node from each end of the quiver is associated to the maximal puncture on the corresponding side of the sphere. The th gauge node is associated to the region between the th and th minimal punctures. To the th minimal puncture is associated a flavor symmetry which acts on the hypermultiplet charged under the th and th gauge nodes.
Following the philosophy of class- theories, we decompose this theory into basic building blocks by decoupling gauge fields. Roughly speaking, the gauge coupling of the th gauge node is inversely proportional to the length between the th and th minimal punctures. To make the gauge couplings small, we take the minimal punctures far apart from one another. Then the geometry looks like a string of spheres, each containing a single minimal puncture, connected by long tubes. The smaller the gauge couplings get, the longer the tubes become, and eventually these spheres spilt up as the couplings go to zero. Each of the spheres represents a bifundamental hypermultiplet, which is a linear quiver with , so it has one minimal and two maximal punctures. The quiver thus breaks into a collection of three-punctured spheres, or trinions.
Conversely, a sphere with 2 maximal and minimal punctures is obtained by gluing trinions together, i.e., by replacing pairs of maximal punctures with tubes. In general, we can connect two Riemann surfaces with a tube at maximal punctures. From the point of view of gauge theory, gluing corresponds to gauging the diagonal combination of the flavor symmetries associated to the maximal punctures involved. Using trinions with one minimal and two maximal punctures, we can obtain any linear quiver in this way, and for that matter, also a circular quiver by further gluing the two ends of a linear quiver together. In this sense, these trinions are building blocks for linear and circular quivers. As these two kinds of quivers can be treated essentially in the same manner, we will focus on linear quivers.
To make contact with brane tilings, we need to describe the linear quiver theory as an quiver gauge theory. In terms of supermultiplets, the vector multiplet for the th gauge node decomposes into a vector multiplet and a chiral multiplet in the adjoint representation with , while the th hypermultiplet consists of two bifundamental chiral multiplets , with . Here is a Cartan generator of . The pair transforms in the doublet of and have charge . From the point of view of supersymmetry, the symmetry generated by the combination
It is helpful for us to prepare two copies for each node of the quiver and impose identification between them. We draw the arrows in such a way that connects the two copies of the th node and makes a triangle with and , as in Fig. 10. Drawn in this form, it is clear that the linear quiver is a special case of the triangle quiver described in Sect. 2, except that the vertical arrow is missing between the flavor nodes at the right end. The corresponding brane tiling diagram is therefore essentially the same, as shown in Fig. 10. Note that the cubic superpotentials, generated around the triangles by worldsheet instantons, are precisely what we need for the theory to have supersymmetry.
As we can split the -punctured sphere into a collection of trinions, we can also break the brane tiling diagram into basic pieces. Each piece represents a single trinion and is made of three zigzag paths; see Fig. 11. Gluing two trinions corresponds to concatenating two such diagrams side by side. In the course of this operation, we must interchange the positions of the zigzag paths labeled and near the glued side of one of the diagrams. This results in an additional vertical arrow in the combined quiver, which is the adjoint chiral multiplet in the vector multiplet used in the gauging.
Before proceeding, we should mention a peculiarity in the case. When , the flavor symmetry of a bifundamental hypermultiplet is enhanced to due to the fact that the fundamental representation of is pseudoreal. For this reason, there is no distinction between minimal and maximal punctures, and each trinion can be regarded as a half-hypermultiplet in the trifundamental representation of . This is reflected in the index of a trinion,
4.2. Surface defects in class- theories
In Ref. , it was explained how to construct a surface defect labeled with a pair of integers , and how to determine its action on the supersymmetric index. Although the method applies to general theories with flavor symmetry, here we review it in the language of class- theories.
Suppose we have a class- theory associated to a Riemann surface that contains a maximal puncture, whose flavor group we call . To this surface we introduce an extra minimal puncture. Concretely, we can do this as follows. First, we rename the flavor group to . Then, we take a trinion representing a hypermultiplet with flavor symmetry , and glue it to by gauging the diagonal subgroup of . The resulting theory has one more flavor symmetry, , than . Correspondingly, the surface associated to has one more minimal puncture than the original surface.
The theory is related to via the RG flow induced by a diagonal constant vev given to the quark , or equivalently, to the baryon . (We may instead give a vev to the antibaryon , but this does not lead to anything different because of the symmetry.) The vev higgses the gauge group and breaks down to the diagonal subgroup. Moreover, it turns the cubic superpotential into a quadratic one that makes and massive, where is the adjoint chiral multiplet introduced in the gluing. Up to Nambu–Goldstone multiplets that survive the higgsing, in the infrared the multiplets we added are gone and we recover , with replaced with . In effect, the minimal puncture introduced by gluing the trinion is “closed.” The R-charge is broken by the vev, but the combination is preserved and identified with a Cartan generator of the infrared R-symmetry.
To create a surface defect in , we instead give the baryon a position-dependent vev . Here, as before, and are complex coordinates of the two orthogonal planes rotated by and , respectively. Away from the origin, the effect of the position-dependent vev is the same as that of the constant vev, so we get in the infrared. If , however, the infrared theory is modified on the plane since the vev vanishes there. By the same token, the theory is modified on the plane if . Hence, in general we obtain with the insertion of a surface defect labeled with the pair of integers , supported on the planes and . This surface defect is to be identified with the surface defect labeled with the pair ) of symmetric representations of discussed in the previous section .
The index of has a pole in the -plane at , and the residue there gives the index of in the presence of the surface defect of type . The reason is the following. The position-dependent vev breaks , , and . At this value of , however, the only combinations of charges that enter the trace defining the index are those that are preserved by the vev. Thus, we can still define the index in this background. As explained above, flows to plus Nambu–Goldstone multiplets in the infrared. The latter contains massless degrees of freedom, and they contribute to the index by a diverging factor, in fact a simple pole in the -plane. Therefore, the residue at this pole gives the index of , together with some factor associated with the Nambu–Goldstone multiplets.
We wish to compute this residue and determine the action of the surface defect on the index in the simplest nontrivial case, namely when and . But first, let us look at the trivial case to gain a better understanding of the computation.
In the construction of a surface defect described above, and actually play no role. The essential point is that the vev given to the baryon built from replaces with in the infrared. So we couple just to for the moment. The index of the combined theory is given by
In order to express this result in a concise form, we introduce the notation of “striking out an arrow” in a quiver diagram to indicate that a constant vev is given to the baryonic operator built from the bifundamental chiral multiplet represented by that arrow, and the contributions from the accompanying Nambu–Goldstone multiplets are discarded. In this notation, what we just found is the identity15). This identity holds when the index of any theory with flavor symmetry (or more generally, any meromorphic function such that ) is coupled to the right node.
With the help of this identity, we can readily show that when a constant vev is turned on for (and the Nambu–Goldstone multiplets are thrown away), the index of reduces to that of . All we have to do is to look at the part of describing the coupling to the trinion, and compute the relevant residue:107) and set , and in the second we canceled the pair of arrows making a loop. Thus, the vev transforms the trinion into the original flavor node of .
Let us calculate the residue (109) for . At , the index of has four sets of colliding poles in the -plane. Two of them are
The difference operator acts on the fugacity for the maximal puncture on which the surface defect was constructed. This fact has a natural interpretation. To construct the surface defect, we first introduced an extra minimal puncture, and then took the residue of a pole in the fugacity of the associated flavor symmetry. The latter step can be thought of as transforming the minimal puncture to another kind of puncture which represents the surface defect. By construction, this puncture is located in the neighborhood of a maximal puncture contained in a trinion. We can take the surface defect puncture and collide it to the maximal puncture. The collision produces a new puncture, and defines the action of the surface defect on the maximal puncture.
4.3. Comparison with the transfer matrix
Let us compare the result with our proposal. For clarity of presentation, take a minimal puncture in and move it close to the maximal puncture on which the surface defect acts. Then the neighborhood of these punctures looks like a trinion glued to another maximal puncture, and is represented by zigzag paths as in Fig. 12.101), we see that if we set
As noted in Ref. , the above transfer matrix is essentially the Hamiltonian of the elliptic Ruijsenaars–Schneider model [64,65] of type . This fact follows from a general result obtained in Ref. .
Here we have considered only the surface defect of type , but the general story is similar. The surface defect of type acts on the index by a difference operator . This operator is expected to coincide with the transfer matrix for an appropriate L-operator. If so, by the RLL relation (53), the operators for all should commute with one another. This is indeed true . From the class- point of view, the mutual commutativity is guaranteed by the fact that the index is independent of the positions of punctures representing surface defects. Therefore, the order in which they act on a maximal puncture is irrelevant. Note that this argument also exploits the existence of an extra dimension, which is the M-theory circle that emerges as the type IIA brane configuration is lifted to M-theory.
For the same reason, a surface defect puncture can be placed between any two punctures, whether minimal or maximal, and still yield the same result. From the point of view of the type IIA system, this property appears to be quite nontrivial and is known as the “hopping invariance” of the index . From the lattice model viewpoint, this is guaranteed by the other RLL relation (52).
4.4. theories of class
There are generalizations of class- theories that preserve only supersymmetry. For these theories, we can compute the index in the presence of a surface defect either by the residue method or using the transfer matrix, and compare the results.
Suppose we have an theory of class , obtained by compactification of M5-branes on a punctured Riemann surface . In the ordinary class- case, is embedded in the cotangent bundle . If we modify this setup in such a way that becomes a holomorphic curve in a generic Calabi–Yau threefold, the supersymmetry gets broken to . A situation commonly studied in the literature is when is the zero section of the total space of the direct sum of two line bundles over , satisfying an appropriate topological condition [67–73].
For our purpose, it is sufficient to consider theories that are realized by simple modifications of the type IIA brane configuration (96) for linear quiver theories. In order to break supersymmetry by half, we rotate some of the NS5-branes so that they span the 012389 directions. We refer to these rotated NS5-branes as , while calling the unrotated ones . Lifted to M-theory, the two types of NS5-branes, and , both become M5-branes supported at points on a cylinder. Correspondingly, there are now two types of minimal punctures labeled with a sign . We denote positive and negative minimal punctures by and , respectively.
Recall that in the case, there is no distinction between minimal and maximal punctures. This fact suggests that maximal punctures also come in two types, positive and negative, denoted by and . To incorporate maximal punctures of different signs, we have to modify the brane setup slightly. In the type IIA picture, we terminate the D4-branes on a D6-brane on each side of the brane system, rather than letting them continue to . Then, a D6-brane extending along the 0123789 directions represents a maximal puncture with , and along the 0123457 directions represents one with . The total brane configuration is summarized as follows:
D4-branes suspended between two NS5-branes of the same sign give rise to an vector multiplet as before. From those suspended between NS5-branes of different signs, we get an vector multiplet. D4-branes suspended between an NS5-brane and a D6-brane of different signs produce an extra chiral multiplet in the adjoint representation of the flavor symmetry of the maximal puncture.
Flipping the sign of a puncture can be understood geometrically in terms of an operation on zigzag paths. The trinion in Fig. 11 has three punctures with . Changing the sign of a maximal puncture to amounts to interchanging the positions of the corresponding pair of zigzag paths so that the adjoint chiral multiplet arises from the crossing; see Fig. 13. Reversing the sign of a minimal puncture entails flipping of the orientation of the corresponding vertical zigzag path. A trinion with all punctures having is shown in Fig. 14.
Let us compute the action of the surface defect of type on the index of an theory that contains a negative maximal puncture. To this end, we couple the trinion in Fig. 13 to the theory by connecting the positive maximal puncture of the former and the negative maximal puncture of the latter, and take the residue of the index of the resulting theory at the pole . The computation is the same as in the case when the surface defect acts on a positive maximal puncture, except that the factor (114) is replaced with
In the brane tiling picture, the surface defect is represented by a dashed line traversing zigzag paths sandwiching an 5-brane region, as illustrated in Fig. 15. Plugging the relations (101) and (117) into formula (77), we see that the transfer matrix reproduces the above difference operator.
5. Surface defects in theories of class
Lastly, we study surface defects in theories of class , which are 4D superconformal theories obtained by compactification of the 6D superconformal theory of type , or two M5-branes probing a orbifold singularity [74,75]. After reviewing basic elements of class- theories with emphasis on their relation to brane tilings, we compute their supersymmetric indices in the presence of simple surface defects, extending calculations in Ref. . The results agree with our proposal based on the lattice model approach: these surface defects are represented by transfer matrices constructed from copies of the relevant L-operator.
5.1. Class- theories
As in our discussion on class- theories in the previous section, we first treat theories for general . We will later set when we actually carry out the index computation.
Let us consider a brane tiling model with gauge groups described by the quiver shown in Fig. 16. The quiver consists of columns, each containing nodes. The vertical direction is periodic, whereas the horizontal direction is a finite interval. When , the theory reduces to the linear quiver theory considered in the previous section. Like that case, we could make the horizontal direction also periodic by gluing the leftmost and rightmost columns in a consistent manner, but we will leave the quiver as it is in the following discussion.
If we apply T-duality to the vertical direction (which we take to be the -direction), we arrive at the type IIA brane configuration (96) for an linear quiver theory, superposed on a orbifold singularity, with acting on and by
Various symmetries of the 4D theory arise from six dimensions as follows. The global symmetry of the 6D theory is . The theory is topologically twisted along the punctured sphere by a subgroup of the R-symmetry (i.e., the structure group of the sphere is replaced with the diagonal subgroup of ). Due to the twisting, only the part of commutes with the rotation group, and it descends to an R-symmetry of the 4D theory.14 Also, we turn on Wilson lines for so that the 4D theory has a nice Lagrangian description. The Wilson lines break the flavor symmetry to its abelian part . Additionally, the 4D theory inherits flavor symmetries , , , from the minimal punctures. The symmetries associated to the zigzag paths come from these flavor symmetries. Finally, each maximal puncture gives rise to a set of flavor symmetries, represented in the quiver by a column of flavor nodes.
A building block of quivers of this kind is a strip of bifundamental chiral multiplets depicted in Fig. 17. It is associated to a trinion with one minimal and two maximal punctures. (Anticipating introduction of punctures of different types, we have drawn the minimal and maximal punctures with a plus sign as and , respectively.) A puncture is labeled with its flavor symmetry, or . In addition, a maximal puncture carries labels called “color” and “orientation” . The orientation simply distinguishes the two maximal punctures in the trinion. In our pictures, we will always place the maximal puncture with positive orientation on the left and the one with negative orientation on the right. The color is defined by the relation between fugacities: arrows with fugacities and start from or end at the same node in the column of nodes corresponding to a maximal puncture labeled . The color of the positively oriented puncture is greater than that of the negatively oriented puncture by 1.
The brane tiling diagram for a trinion is also shown in Fig. 17. The minimal puncture corresponds to the vertical zigzag path in the middle, while the maximal punctures correspond to the unshaded regions on the sides of the diagram. The two sets of fugacities and are related by
To reconstruct the quiver with columns that we started with, we glue together copies of trinions to get the -punctured sphere. Gluing can be done only between two maximal punctures with opposite orientation and the same color. This operation gauges the diagonal combination of the flavor symmetries of the maximal punctures, and at the same time, adds in bifundamental chiral multiplets corresponding to arrows going upward between the gauged nodes. The restriction on the color and orientation ensures that the mixed anomalies for and cancel.
The rule for gluing is transparent in the brane tiling picture. When we concatenate two brane tiling diagrams, we must connect the zigzag paths in a way consistent with their labels, or the associated flavor symmetries would be lost. Therefore, the colors of the maximal punctures glued together are required to match. Furthermore, each pair of horizontal paths near the glued sides is forced to cross once, resulting in the additional vertical arrows in the combined quiver.
From the brane tiling perspective, it is also clear that the color of the positively oriented maximal puncture increases by 1 as we glue a trinion to it, since the zigzag paths with fugacities shift upward when they cross a vertical path. In particular, the color comes back to the original value after trinions are glued.
5.2. Turning on flux
By definition, class- theories arise from compactification of the 6D theory on punctured Riemann surfaces. In order to completely specify a class- theory, however, we need more data than just a punctured Riemann surface. When we compactify the 6D theory, we can turn on flux for the abelian part of its flavor symmetry, i.e., we have a choice of the associated line bundles. Consequently, there are different theories associated to the same punctured Riemann surface, corresponding to different flux backgrounds for , , and . In fact, for we have already analyzed the case with flux in Sect. 4. This is the class- counterpart of the case with flux, which we will treat in Sect. 5.5. Here we discuss flux for and .
A procedure for turning on flux in class- theories was proposed in Ref. . Suppose we want to create flux for in a class- theory, where is a fixed index. To do that, we glue a trinion (of the sort depicted in Fig. 17) to a maximal puncture of the associated Riemann surface. The new surface thus obtained has one more minimal puncture than the original surface does, hence one more flavor symmetry . Then we “close” this puncture: we give a constant vev to the baryon made of the bifundamental chiral multiplet with fugacity , and “flip” the other baryons in the same column, whose fugacities are with . By “flipping a chiral operator ,” we mean coupling to an external chiral multiplet through a superpotential . After closing the puncture, we obtain a theory associated to the same surface as the original one, but with the color of the maximal puncture shifted by 1. The result is interpreted as a theory with one unit of flux turned on.
Similarly, we can turn on (minus) one unit of flux by giving a vev to the antibaryon with fugacity . More generally, we can repeat the above procedure to add any amount of flux for and . If we turn on flux for more than one flavor symmetries, there are different orders of doing this. However, they all lead to the same result due to the S-duality permuting minimal punctures.
An important point is that adding one unit of flux for every (or one unit of flux for every ) is equivalent to doing nothing. This is because the symmetries come from the symmetry of the 6D theory, hence the sum of their charges is zero. We can see this property more explicitly as follows.
To create one unit of flux for each and every in a given theory, we attach to the theory a sphere with 2 maximal and minimal punctures, and give vevs to baryons charged under distinct . Let us suppose that the maximal puncture to which we attach the sphere has . We align the minimal punctures horizontally and number them 1, , from left to right. Denoting by the fugacity for the flavor symmetry associated to the th puncture, we choose to give vevs to the baryons with fugacities , , , . Fig. 18a illustrates the situation.
The vevs identify pairs of nodes as we explained before, and also turn the cubic superpotentials involving the stuck-out arrows into quadratic ones that give masses to the other arrows participating in these superpotentials. After the massive arrows are integrated out, we have the situation in Fig. 18b. The quiver now contains a number of gauge nodes with two arrows attached. These nodes exhibit confinement and chiral symmetry breaking, and as a result, further pairs of nodes are identified, as in Fig. 18c. The identification of nodes results in more gauge nodes with two arrows attached, which again equate pairs of nodes. This process continues until all flavor nodes are identified with the gauge nodes coming from the maximal puncture to which the sphere was attached. There are baryons left over from the confinement, but they couple to the scalars introduced in the flip operation and together become massive. In the end, all minimal punctures are gone, and we recover the original theory.
5.3. Surface defects in theories
Surface defects in class- theories can be realized via RG flows in much the same way as in their counterparts in class- theories. Given a theory , we construct another theory by gluing a trinion to it. Then, we give a position-dependent vev to a baryon charged under the flavor symmetry of the new minimal puncture, while flipping the other baryons in the same column. The vev triggers to flow to the original theory , but in the presence of a surface defect labeled with the pair of integers , or the pair ) of symmetric representations of .
A novelty in class- theories is that maximal punctures have colors. For the color of the maximal puncture of to remain unchanged by the above operation, the trinion we attach to it must have two maximal punctures of the same color. This is not the case for the trinion in Fig. 17, as it has maximal punctures differing in their colors by 1.
Luckily, we know how to make the colors of the maximal punctures match: we glue to it more trinions of the same type, and close all minimal punctures introduced in the process. More precisely, we prepare a sphere with 2 maximal and minimal punctures, pick , and for all , give constant vevs to the baryons with fugacities and flip those with fugacities , . Then, the -punctured sphere flows to a trinion with one minimal puncture and two maximal punctures of the same color, with minus one unit of flux turned on for .
In summary, a surface defect in is realized as follows. First, we construct by coupling to a sphere with 2 maximal and minimal punctures. Next, we give constant vevs to the baryons with fugacities , . Finally, we give the position-dependent vev to the baryon with fugacity , and flip all other baryons. There are choices for the index , and each choice leads to a different surface defect. Of course, we could follow the same procedure with the roles of baryons and antibaryons exchanged, so in total there are inequivalent surface defects that may be constructed in this way for a given pair of integers .
What we have to do for the computation of the supersymmetric index with a surface defect is clear now. We couple the quiver shown in Fig. 18a to by gauging the diagonal combination of the flavor symmetry in the leftmost column of the quiver and the flavor symmetry of a maximal puncture of . Then we replace one of the arrows with , and compute the residues accordingly. The result is the index of in the presence of a surface defect of type .
Let us carry out this computation in the simplest case of theories and , when the surface defect is labeled with the fundamental representation of .
Without loss of generality, we assume that the maximal puncture of to which we glue the -punctured sphere has color . The baryon that is given the position-dependent vev is made of the bifundamental chiral multiplet with fugacity , where . Since the index is independent of the position of punctures, we can rearrange the minimal punctures so that the *th puncture comes to the rightmost position and the rest follow in descending order in their labels. After the rearrangement, the neighborhood of the baryon looks as in Fig. 19a, where we have introduced the symbols and for convenience.
The identity (107) obtained in the previous section states that the constant vevs given to arrows identify pairs of nodes and make the neighboring arrows massive, producing the quiver shown in Fig. 19b.
For the arrow with the position-dependent vev, we use relation (113). This relation says that is set to or after taking the residue. To be specific, let us consider the former case. In this case, at the -integral involves 12 factors of elliptic gamma functions of the form :
To evaluate the - and -integrals properly, we can multiply the fugacities of the first and last factors in the product (123) by and , respectively, and later take the limit . If we apply identity (A23) after this shift of fugacities, we find that the double poles are resolved into four simple poles located at . It is straightforward to compute the residues. The end result of the calculation is that from the - and -integration, we get
Proceeding to the -integral, we encounter the same calculation after the -integral is performed, and the index receives contributions analogous to those found above for the -integral. The same is true for every other -integral. Apart from these contributions, the structure of the calculation is essentially identical to the constant vev case illustrated in Fig. 18. All in all, the effect of the surface defect of type is represented in the index by the action of the difference operator
5.4. Comparison with transfer matrices
In the brane tiling picture, the above surface defects should be represented by a dashed line that crosses pairs of horizontal zigzag paths as in Fig. 20. We see that it inserts the transfer matrix constructed from the product of . Using formula (74) and relation (122), we can check
Interestingly, the transfer matrix unifies the difference operators , , corresponding to the choices of the position-dependent vev given to the trinion attached to , into a single one-parameter family of difference operators. These operators differ simply in the value of the spectral parameter for the dashed line.
5.5. The case with flux
Finally, we consider surface defects in theories with flux [30–32]. This is the -orbifold version of the class- theories discussed in Sect. 4.4. The type IIA brane construction for theories associated to a cylinder or torus is the same as in the class- case, except that the branes are placed at the orbifold singularity. Therefore, each puncture has a sign specifying which type of NS5- or D6-brane it comes from.
The trinion in Fig. 17 has for all punctures. If we flip the sign of a maximal puncture, there arise additional bifundamental chiral multiplets between the flavor nodes coming from that puncture; see Fig. 22. On the other hand, flipping the signs of all punctures gives the quiver in Fig. 22.
Looking at the brane tiling diagrams for these quivers, we notice that the sign of a puncture is correlated with the 5-brane charge distribution in the relevant area: a minimal puncture with sign corresponds to a column of 5-brane regions in the middle, while a maximal puncture of orientation and sign corresponds to a column of 5-brane regions on a side. (This point of view makes it clear that each is really a -tuple of signs, as pointed out in Ref. .) The change in the color between the left and right maximal punctures is equal to of the minimal puncture. When we glue two trinions together, the maximal punctures connected by a tube must have opposite signs in order for the 5-brane charges to be conserved. Otherwise, we have to flip the sign of one of the punctures, as we did in the case of gluing two trinions with positive punctures.
Let us determine how surface defects act on a negative maximal puncture. We take a trinion with three positive punctures and glue it to the trinion in Fig. 22 from the left. Then, we close the minimal puncture contained in the latter trinion by giving a vev to its baryon charged under for some , to obtain a trinion with one minimal and two maximal punctures having the same color and different signs. This trinion has minus one unit of flux for . We attach it to a negative maximal puncture (which we assume to have color ) in some theory from the right, and give the position-dependent vev to the baryon charged under in the trinion. This gives a surface defect of type acting on the negative maximal puncture.
In the brane tiling diagram, the surface defect creates a dashed line as in Fig. 23. From formula (77) and relations (122) and (130), we can check that the above difference operators are reproduced from the transfer matrix.
The key element underlying various connections between 4D supersymmetric field theories and integrable lattice models is the emergence of the structure of a 2D TQFT equipped with line operators that are localized in extra dimensions [4,6,7]. Branes in string theory, combined with protected quantities such as supersymmetric indices, provide a natural framework in which such structures may be found.
In this paper we have utilized this framework to elucidate the integrability aspect of surface defects in 4D theories. As we have seen, under the correspondence between brane tilings and integrable lattice models, a class of half-BPS surface defects labeled with representations of are mapped to transfer matrices constructed from L-operators. In the case of the fundamental representation of , the relevant L-operator is that of Sklyanin, which satisfies an RLL relation with Baxter's R-operator for the eight-vertex model. We have shown that the corresponding transfer matrix unifies the difference operators obtained by the residue method for theories of class .
Our analysis is far from complete, however. Obviously lacking is the answer to the following question: What is the L-operator for a general representation of ?
We may approach this important question either from the lattice model side or from the field theory side. The strategy on the lattice model side would be to search for an L-operator that solves the appropriate Yang–Baxter equations, as we have done for the fundamental representation of .
From the field theory side, the strategy is to somehow compute the indices of brane tiling models in the presence of general surface defects, and read off the L-operator from them. For example, we may combine the residue method for class- theories, which can handle the symmetric representations, and analysis of the algebra generated by the resulting difference operators. This program had some success in the class- case [51,52]. A different method is to realize a surface defect as a 2D theory, and compute the index of the coupled 4D–2D system by localization of the path integral. This was done in Ref.  for and symmetric representations. A related computation appeared in Ref. .
Either strategy is not without shortcomings. The Yang–Baxter equations do not uniquely determine the L-operator. The supersymmetric indices, on the other hand, encode transfer matrices but not the L-operator directly. We would therefore need to combine approaches from both sides.
Another direction we have left unexplored is the study of the 2D TQFT itself, which in our discussion just served as an intermediate step whereby brane tilings and integrable lattice models were connected. String theory predicts that this TQFT is related to the 2D TQFT arising from the indices of class- theories through a duality exchanging line operators in the former and local operators in the latter. It may be possible to determine these TQFTs by localization computations along the lines of Refs. [59,80–84].
There are many more interesting questions to be asked in relation to 4D supersymmetric field theories and integrable lattice models. We wish to answer some of them in future work.
We are grateful to Giulio Bonelli and Alessandro Tanzini for their invitation to the workshop “V Workshop on Geometric Correspondences of Gauge Theories” at SISSA, during which this project was initiated. We also thank Hironori Mori, Jaewon Song, and Shigeki Sugimoto for useful discussions, and Michio Jimbo, Saburo Kekei, Satoshi Nawata, Shlomo Razamat, Vyacheslav Spiridonov, Piotr Sułkowski, and Yuji Yamada for helpful comments. K.M. would like to thank Piotr Sułkowski for his hospitality at the University of Warsaw, where part of this work was carried out. The work of K.M. is supported by the EPSRC Programme Grant EP/K034456/1 “New Geometric Structures from String Theory.” The work of J.Y. is supported by the ERC Starting Grant no. 335739 “Quantum fields and knot homologies” funded by the European Research Council under the European Union's Seventh Framework Programme.
Open Access funding: SCOAP3.
Appendix. Definitions and useful formulas
In this appendix we collect definitions and useful formulas concerning special functions we encounter in this paper.
A.1. Theta functions
A.2. Elliptic gamma function
Let , , , 6 be six complex parameters such that and . Then, we have the following identity proved in Ref. :