A construction of hyperkähler metrics through RiemannHilbert problems I
Abstract.
In 2009 Gaiotto, Moore and Neitzke presented a new construction of hyperkähler metrics on the total spaces of certain complex integrable systems, represented as a torus fibration over a base space , except for a divisor in , in which the torus fiber degenerates into a nodal torus. The hyperkähler metric is obtained via solutions of a RiemannHilbert problem. We interpret the KontsevichSoibelman Wall Crossing Formula as an isomonodromic deformation of a family of RH problems, therefore guaranteeing continuity of at the walls of marginal stability. The technical details about solving the different classes of RiemannHilbert problems that arise here are left to a second article. To extend this construction to singular fibers, we use the OoguriVafa case as our model and choose a suitable gauge transformation that allow us to define an integral equation defined at the degenerate fiber, whose solutions are the desired Darboux coordinates . We show that these functions yield a holomorphic symplectic form , which, by Hitchin’s twistor construction, constructs the desired hyperkähler metric.
2010 Mathematics Subject Classification:
Primary1. Introduction
Hyperkähler manifolds first appeared within the framework of differential geometry as Riemannian manifolds with holonomy group of special restricted group. Nowadays, hyperkähler geometry forms a separate research subject fusing traditional areas of mathematics such as differential and algebraic geometry of complex manifolds, holomorphic symplectic geometry, Hodge theory and many others.
One of the latest links can be found in theoretical physics: In 2009, Gaiotto, Moore and Neitzke [6] proposed a new construction of hyperkähler metrics on target spaces of quantum field theories with superysmmetry. Such manifolds were already known to be hyperkähler (see [15]), but no known explicit hyperkähler metrics have been constructed.
The manifold is a total space of a complex integrable system and it can be expressed as follows. There exists a complex manifold , a divisor and a subset such that is a torus fibration over . On the divisor , the torus fibers of degenerate, as Figure 1 shows.
Moduli spaces of Higgs bundles on Riemann surfaces with prescribed singularities at finitely many points are one of the prime examples of this construction. Hyperkähler geometry is useful since we can use Hitchin’s twistor space construction [11] and consider all worth of complex structures at once. In the case of moduli spaces of Higgs bundles, this allows us to consider from three distinct viewpoints:

(Dolbeault) is the moduli space of Higgs bundles, i.e. pairs , a rank degree zero holomorphic vector bundle and a Higgs field.

(De Rham) is the moduli space of flat connections on rank holomorphic vector bundles, consisting of pairs with a holomorphic connection and

(Betti) of conjugacy classes of representations of the fundamental group of .
All these algebraic structures form part of the family of complex structures making into a hyperkähler manifold.
To prove that the manifolds from the integrable systems are indeed hyperkähler, we start with the existence of a simple, explicit hyperkähler metric on . Unfortunately, does not extend to . To construct a complete metric , it is necessary to do “quantum corrections” to . These are obtained by solving a certain explicit integral equation (see (2.12) below). The novelty is that the solutions, acting as Darboux coordinates for the hyperkähler metric , have discontinuities at a specific locus in . Such discontinuities cancel the global monodromy around and is thus feasible to expect that extends to the entire .
We start by defining a RiemannHilbert problem on the slice of the twistor space . That is, we look for functions with prescribed discontinuities and asymptotics. In the language of RiemannHilbert theory, this is known as monodromy data. Rather than a single RiemannHilbert problem, we have a whole family of them parametrized by the manifold. We show that this family constitutes an isomonodromic deformation since by the KontsevichSoibelman WallCrossing Formula, the monodromy data remains invariant.
Although solving RiemannHilbert problems in general is not always possible, in this case it can be reduced to an integral equation solved by standard Banach contraction principles. We will focus on a particular case known as the “Pentagon” (a case of Hitchin systems with gauge group ). The family of RiemannHilbert problems and their methods of solutions is a topic of independent study so we leave this construction to a second article that can be of interest in the study of boundaryvalue problems.
The extension of the manifold is obtained by gluing a circle bundle with an appropriate gauge transformation eliminating any monodromy problems near the divisor . The circle bundle constructs the degenerate tori at the discriminant locus (see Figure 2).
On the extended manifold we prove that the solutions of the RiemannHilbert problem on extend and the resulting holomorphic symplectic form gives the desired hyperkähler metric .
Although for the most basic examples of this construction such as the moduli space of Higgs bundles it was already known that extends to a hyperkähler manifold with degenerate torus fibers, the construction here works for the general case of . Moreover, the functions here are special coordinates arising in moduli spaces of flat connections, Teichmüller theory and Mirror Symmetry. In particular, these functions are used in [4] for the construction of holomorphic discs with boundary on special Lagrangian torus fibers of mirror manifolds.
The organization of the paper is as follows. In Section 2 we introduce the complex integrable systems to be considered in this paper. These systems arose first in the study of moduli spaces of Higgs bundles and they can be written in terms of initial data and studied abstractly. This leads to a formulation of a family of RiemannHilbert problems, whose solutions provide Darboux coordinates for the moduli spaces considered and hence equip the latter with a hyperkähler structure. In Section 3 we fully work the simplest example of these integrable systems: the OoguriVafa case. Although the existence of this hyperkähler metric was already known, this is the first time it is obtained via RiemannHilbert methods. In Section 4, we explicitly show that this metric is a smooth deformation of the wellknown TaubNUT metric near the singular fiber of thus proving its extension to the entire manifold. In Section 5 we introduce our main object of study, the Pentagon case. This is the first nontrivial example of the integrable systems considered and here the Wall Crossing phenomenon is present. We use the KS wallcrossing formula to apply an isomonodromic deformation of the RiemannHilbert problems leading to solutions continuous at the wall of marginal stability. Finally, Section 6 deals with the extension of these solutions to singular fibers of thought as a torus fibration. What we do is to actually complete the manifold from a regular torus fibration by gluing circle bundles near a discriminant locus . This involves a change of the torus coordinates for the fibers of . In terms of the new coordinates, the functions extend to the new patch and parametrize the complete manifold . We finish the paper by showing that, near the singular fibers of , the hyperkähler metric looks like the metric for the OoguriVafa case plus some smooth corrections, thus proving that this metric is complete.
Acknowledgment: The author likes to thank Andrew Neitzke for his guidance, support and incredibly helpful conversations.
2. Integrable Systems Data
We start by presenting the complex integrable systems introduced in [6]. As motivation, consider the moduli space of Higgs bundles on a complex curve with Higgs field having prescribed singularities at finitely many points. In [7], it is shown that the space of quadratic differentials on with fixed poles and residues is a complex affine space and the map is proper with generic fiber , a compact torus obtained from the spectral curve , a doublebranched cover of over the zeroes of the quadratic differential . has an involution that flips . If we take to be the subgroup of odd under this involution, forms a lattice of rank 2 over , the space of quadratic differentials with only simple zeroes. This lattice comes with a nondegenerate antisymmetric pairing from the intersection pairing in . It is also proved in [7] that the fiber can be identified with the set of characters . If denotes the tautological 1form in , then for any ,
defines a holomorphic function in . Let be a local basis of with the dual basis of . Without loss of generality, we also denote by the pairing in . Let be short notation for . Since , .
This type of data arises in the construction of hyperkähler manifolds as in [6] and [13], so we summarize the conditions required:
We start with a complex manifold (later shown to be affine) of dimension and a divisor . Let . Over there is a local system with fiber a rank lattice, equipped with a nondegenerate antisymmetric integer valued pairing .
We will denote by the dual of and, by abuse of notation, we’ll also use for the dual pairing (not necessarily integervalued) in . Let denote a general point of . We want to obtain a torus fibration over , so let be the set of twisted unitary characters of ^{1}^{1}1Although we can also work with the set of unitary characters (no twisting involved) by shifting the coordinates, we choose not to do so, as that results in more complex calculations, i.e. maps satisfying
Topologically, is a torus . Letting vary, the form a torus bundle over . Any local section gives a local angular coordinate of by “evaluation on ”, .
We also assume there exists a homomorphism such that the vector varies holomorphically with . If we pick a patch on which admits a basis of local sections in which is the standard symplectic pairing, then (after possibly shrinking ) the functions
are real local coordinates. The transition functions on overlaps are valued on , as different choices of basis in must fix the symplectic pairing. This gives an affine structure on .
By differentiating and evaluating in , we get 1forms on which are linear on . For a local basis as in the previous paragraph, let denote its dual basis on . We write as short notation for
(2.1) 
where we sum over repeated indices. Observe that the antisymmetric pairing and the antisymmetric wedge product of 1forms makes (2.1) symmetric. We require that:
(2.2) 
By (2.2), near , can be locally identified with a complex Lagrangian submanifold of .
In the example of moduli spaces of Higgs bundles, as approaches a quadratic differential with nonsimple zeros, one homology cycles vanishes (see Figure 1). This cycle is primitive in and its monodromy around the critical quadratic differential is governed by the PicardLefschetz formula. In the general case, let be a component of the divisor . We also assume the following:

as for some .

is primitive (i.e. there exists some with ).

The monodromy of around is of “PicardLefschetz type”, i.e.
(2.3)
We assign a complex structure and a holomorphic symplectic form on as follows (see [13] and the references therein for proofs). Take a local basis of . If and is its dual, let
(2.4) 
By linearity on of the 1forms, is independent of the choice of basis. There is a unique complex structure on for which is of type (2,0). The 2form gives a holomorphic symplectic structure on . With respect to this structure, the projection is holomorphic, and the torus fibers are compact complex Lagrangian submanifolds.
Recall that a positive 2form on a complex manifold is a real 2form for which for all real tangent vectors . From now on, we assume that is a positive 2form on . Now fix . Then we can define a 2form on by
This is a positive form of type (1,1) in the complex structure. Thus, the triple determines a Kähler metric on . This metric is in fact hyperkähler (see [5]), so we have a whole worth of complex structures for , parametrized by . The above complex structure represents , the complex structure at in . The superscript stands for “semiflat”. This is because is flat on the torus fibers .
Alternatively, it is shown in [6] that if
(2.5) 
Then the 2form
(where the DeRham operator is applied to the part only) can be expressed as
for , that is, in the twistor space of [11], is a holomorphic section of (the twisting by is due to the poles at and in ). This is the key step in Hitchin’s twistor space construction. By [6, §3], is hyperkähler.
We want to reproduce the same construction of a hyperkähler metric now with corrected Darboux coordinates . For that, we need another piece of data. Namely, a function such that . Furthermore, we impose a condition on the nonzero . Introduce a positive definite norm on . Then we require the existence of such that
(2.6) 
for those such that . This is called the Support Property, as in [6].
For a component of the singular locus and for the primitive element in for which as , we also require
To see where these invariants arise from, consider the example of moduli spaces of Higgs bundles again. A quadratic differential determines a metric on . Namely, if , . Let be the curve obtained after removing the poles and zeroes of . Consider the finite length inextensible geodesics on in the metric . These come in two types:
On the branched cover , each geodesic can be lifted to a union of closed curves in , representing some homology class . See Figure 5.
In this case, counts these finite length geodesics: every saddle connection with lift contributes and every closed geodesic with lift contributes .
Back to the general case, we’re ready to formulate a RiemannHilbert problem on the slice of the twistor space . Recall that in a RH problem we have a contour dividing a complex plane (or its compactification) and one tries to obtain functions which are analytic in the regions defined by the contour, with continuous extensions along the boundary and with prescribed discontinuities along and fixed asymptotics at the points where is nonsmooth. In our case, the contour is a collection of rays at the origin and the discontinuities can be expressed as symplectomorphisms of a complex torus:
Define a ray associated to each as:
We also define a transformation of the functions given by each :
(2.7) 
Let denote the space of twisted complex characters of , i.e. maps satisfying
(2.8) 
has a canonical Poisson structure given by
The glue together into a bundle over with fiber a complex Poisson torus. Let be the pullback of this system to . We can interpret the transformations as birational automorphisms of . To each ray going from 0 to in the plane, we can define a transformation
(2.9) 
Note that all the ’s involved in this product are multiples of each other, so the commute and it is not necessary to specify an order for the product.
To obtain the corrected , we can formulate a RiemannHilbert problem for which the former functions are solutions to it. We seek a map with the following properties:

depends piecewise holomorphically on , with discontinuities only at the rays for which .

The limits as approaches any ray from both sides exist and are related by
(2.10) 
obeys the reality condition
(2.11) 
For any , exists and is real.
In [6], this RH problem is formulated as an integral equation:
(2.12) 
One can define recursively, setting :
(2.13) 
More precisely, we have a family of RH problems, parametrized by , as this defines the rays , the complex torus where the symplectomorphisms are defined and the invariants involved in the definition of the problem.
We still need one more piece of the puzzle, since the latter function may not be continuous. In fact, jumps along a real codimension1 loci in called the “wall of marginal stability”. This is the locus where 2 or more functions coincide in phase, so two or more rays become one. More precisely:
The jumps of are not arbitrary; they are governed by the KontsevichSoibelman wallcrossing formula.
To describe this, let be a strictly convex cone in the plane with apex at the origin. Then for any define
(2.14) 
The arrow indicates the order of the rational maps . is a birational Poisson automorphism of . Define a good path to be a path in along which there is no point with and . (So as we travel along a good path, no rays enter or exit V.) If are the endpoints of a good path , the wallcrossing formula is the condition that are related by parallel transport in along . See Figure 6.
2.1. Statement of Results
We will restrict in this paper to the case , so . We want to extend the torus fibration to a manifold with degenerate torus fibers. To give an example, in the case of Hitchin systems, the torus bundle is not the moduli space of Higgs bundles yet, as we have to consider quadratic differentials with nonsimple zeroes too. The main results of this paper center on the extension of the manifold to a manifold with an extended fibration such that the torus fibers degenerate to nodal torus (i.e. “singular” or “bad” fibers) for .
We start by fully working out the simplest example known as OoguriVafa [3]. Here we have a fibration over the open unit disk . At the discriminant locus , the fibers degenerate into a nodal torus. The local rank2 lattice has a basis and the skewsymmetric pairing is defined by . The monodromy of around is . We also have functions , for holomorphic and admitting an extension to . Finally, the integervalued function in is here: and for any other . There is no wall of marginal stability in this case. The integral equation (2.12) can be solved after just 1 iteration.
For all other nontrivial cases, in order to give a satisfactory extension of the coordinates, it was necessary to develop the theory of RiemannHilbertBirkhoff problems to suit these infinitedimensional systems (as the transformations defining the problem can be thought as operators on , rather than matrices). It is not clear that such coordinates can be extended, since we may approach the bad fiber from two different sides of the wall of marginal stability and obtain two different extensions. To overcome this first obstacle, we have to use the theory of isomonodromic deformations as in [2] to reformulate the RiemannHilbert problem in [6] independent of the regions determined by the wall.
Having redefined the problem, we want our to be smooth on the parameters and , away from where the prescribed jumps are. Even at , there was no mathematical proof that such condition must be true. In the companion paper [8], we combine classical Banach contraction methods and ArzelaAscoli results on uniform convergence in compact sets to obtain:
Theorem 2.1.
If the collection of nonzero satisfies the support property (2.6) and if the parameter of (2.5) is large enough (determined by the values ), there exists a unique collection of functions with the prescribed asymptotics and jumps as in [6]. These functions are smooth on and the torus coordinates (even for at the wall of marginal stability), and piecewise holomorphic on .
Since we’re considering only the case , is a rank1 lattice over the Riemann surface and the discriminant locus where the torus fibers degenerate is a discrete subset of .
From this point on, we restrict our attention to the next nontrivial system, known as the Pentagon case [13]. Here with 2 bad fibers which we can assume are at and is the twicepunctured plane. There is a wall of marginal stability where all are contained in the same line. This separates in two domains and a simplyconnected . See Figure 7.
On we can trivialize and choose a basis with pairing . This basis does not extend to a global basis for since it is not invariant under monodromy. However, the set is indeed invariant so the following definition of makes global sense:
The Pentagon case appears in the study of Hitchin systems with gauge group . The extension of was previously obtained by hyperkähler quotient methods in [1], but no explicit hyperkähler metric was constructed.
Once the are obtained by Theorem 2.1, it is necessary to do an analytic continuation along for the particular for which as . Without loss of generality, we can assume there is a local basis of such that in . After that, an analysis of the possible divergence of as shows the necessity of performing a gauge transformation on the torus coordinates of the fibers that allows us to define an integral equation even at . This series of transformations are defined in (6.9), (6.10), (6.11) and (6.27), and constitute a new result that was not expected in [6]. We basically deal with a family of boundary value problems for which the jump function vanishes at certain points and singularities of certain kind appear as . As this is of independent interest, we leave the relevant results to [8] and we show that our solutions contain at worst branch singularities at 0 or in the plane. As in the case of normal fibers, we can run a contraction argument to obtain Darboux coordinates even at the singular fibers and conclude:
Theorem 2.2.
Let be a local basis for in a small sector centered at such that as . For the Pentagon integrable system, the local function admits an analytic continuation to a punctured disk centered at in . There exists a gauge transformation that extends the torus fibration to a manifold that is locally, for each point in , a (trivial) fibration over with fiber coordinatized by and with one fiber collapsed into a point. For big enough, it is possible to extend and to , still preserving the smooth properties as in Theorem 2.1.
After we have the smooth extension of the by Theorem 2.2, we can extend the holomorphic symplectic form labeled by as in [11] for all points except possibly one at the singular fiber. From we can obtain the hyperkähler metric and, in the case of the Pentagon, after a change of coordinates, we realize locally as the TaubNUT metric plus smooth corrections, finishing the construction of and its hyperkähler metric. The following is the main theorem of the paper.
Theorem 2.3.
For the Pentagon case, the extension of the manifold constructed in Theorem 2.2 admits, for large enough, a hyperkähler metric obtained by extending the hyperkähler metric on determined by the Darboux coordinates .
3. The OoguriVafa Case
3.1. Classical Case
We start with one of the simplest cases, known as the OoguriVafa case, first treated in [3]. To see where this case comes from, recall that by the SYZ picture of K3 surfaces [9], any K3 surface is a hyperkähler manifold. In one of its complex structures (say ) is elliptically fibered, with base manifold and generic fiber a compact complex torus. There are a total of 24 singular fibers, although the total space is smooth. See Figure 8.
Gross and Wilson [10] constructed a hyperkähler metric on a K3 surface by gluing in the OoguriVafa metric constructed in [14] with a standard metric away from the degenerate fiber. Thus, this simple case can be regarded as a local model for K3 surfaces.
We have a fibration over the open unit disk . At the locus (in the literature this is also called the discriminant locus), the fibers degenerate into a nodal torus. Define as , the punctured unit disk. On there exists a local system of rank2 lattices with basis and skewsymmetric pairing defined by . The monodromy of around is . We also have functions . On we have local coordinates for the torus fibers with monodromy . Finally, the integervalued function in is here: and for any other . There is no wall of marginal stability in this case.
We call this the “classical OoguriVafa” case as it is the one appearing in [14] already mentioned at the beginning of this section. In the next section, we’ll generalize this case by adding a function to the definition of .
Let
(3.1) 
These functions receive corrections defined as in [6]. We are only interested in the pair which will constitute our desired Darboux coordinates for the holomorphic symplectic form . The fact that gives that . As , and approach 0. Thus . Since the actual is obtained after only 1 iteration of (2.13). For each , let be the ray in the plane defined by . Similarly, .
Let
(3.2) 
For convenience, from this point on we assume is of the form , where is a positive number, is fixed and . Moreover, in , , for , and a similar parametrization holds in .
Lemma 3.1.
For fixed , as in (3.2) has a limit as .
Proof.
Writing , we want to find the limit as of
(3.3) 
For simplicity, we’ll focus in the first integral only, the second one can be handled similarly. Rewrite:
(3.4) 
Observe that
and after a change of variables , we get  
Thus, (3.4) reduces to
(3.5) 
If , (3.3) diverges to , in which case . Otherwise, is bounded away from 0. Consequently, in . As , the integrals are dominated by
if . Hence we can interchange the limit and the integral in (3.5) and obtain that, as , this reduces to
(3.6) 
where
are the (unique) holomorphic solutions in the simply connected domain to the ODEs
This forces us to rewrite (3.6) uniquely as
(3.7) 
Here denotes the principal branch of the log in both cases, and the equation makes sense for (recall that by construction, we have the additional datum ). We want to conclude that
(3.8) 
still using the principal branch of the log. To see this, define as . This is an analytic function on and clearly . Thus is constant in . It is easy to show that the identity holds for a suitable choice of (for example, if is not real, choose ) and by the above, it holds on all of ; in particular, for .