WorksMathematics
Gopakumar-Vafa invariants and Mordell-Weil lattices
Abstract
Let be a Weierstrass fibration with integral fibers from a smooth Calabi–Yau threefold to a del Pezzo surface. For a smooth curve with smooth inverse image , we decompose the moduli fiber of stable sheaves with Euler characteristic one and projected cycle . In fiber degree , its components are indexed by all sections of with , where is the Mordell–Weil height pairing. For infinitesimally rigid sections we compute the GV contributions, recovering the theta product over a -curve. We formulate the corresponding conjecture for families of sections.
Introduction
Huang, Katz, Klemm, and Wang (HKKW) compute refined BPS numbers for compact elliptic Calabi–Yau threefolds using Wilson loops [25 §§3.3–3.4 and 5.1]. In their calculation, distinct sections can contribute to the same curve class [25 §§5.1.4 and 5.1.7]. Fix a smooth curve in the base. Which sections over occur in a given fiber degree, and what does each contribute to the Gopakumar–Vafa (GV) invariants after fibers are added? Since the projected cycle is with multiplicity one, each support has exactly one horizontal component, of degree one over . Theorem 3.1 describes the moduli fiber as the section varies, and Theorem 1.1 computes the ambient GV contribution of each infinitesimally rigid section.
In Gromov–Witten (GW) theory, such contributions enter the series of virtual counts of stable maps through the predicted GV multicover formula
where runs over nonzero effective curve classes, is the connected genus- GW invariant, is the genus- GV number in class , is the genus variable, and the formal symbols satisfy [15 Part II, equation (3.2)]. The agreement of these numbers with those defined by vanishing cycles below is conjectural in general [38].
For an elliptic threefold, one can group the terms of this series by base and fiber degrees. Alim–Scheidegger and Klemm–Manschot–Wotschke studied the modular structure and holomorphic anomaly equations that arise in this way [1, 34]. At fixed base degree, Huang, Katz, and Klemm predicted that the all-genus partition function is expressed in meromorphic Jacobi forms with prescribed poles. In their formula the fiber parameter is the modular variable and the string coupling is the elliptic variable [23 equations (1.1)–(1.2)]. Their later work extends the calculation to refined invariants and examines the dependence on complex structure and orientation [24 §§3.5–3.7].
The Jacobi transformation has a monodromy interpretation. Schimmann computed the action of relative Fourier–Mukai monodromy on brane charges [46 §3], and Plioline and Schimmann relate this action to the transformation of the fixed-base partition function [43 §4]. On the enumerative side, Oberdieck and Pixton prove the quasi-Jacobi and holomorphic anomaly equations for GW invariants of rational elliptic surfaces [40]. In Pandharipande–Thomas (PT) theory one takes virtual counts of stable pairs with pure of dimension one and zero-dimensional cokernel [42]. Here Oberdieck and Shen prove the elliptic transformation law for reduced base classes [41 Theorem 1].
HKKW, on the other hand, fix the fiber degree and allow the base degree to vary. Over they propose a finite expansion of the compact partition function in local Wilson loops [25 §3.4.1, equation (3.58)], whose values are computed in [33, 26, 52]. For this expansion, a section and its added fibers count toward the same degree. Our first task is therefore to determine the degree of the section itself. Appendix A formulates the conjecture for families motivated by HKKW’s expansion.
We work over . Let be a smooth projective del Pezzo surface, so that is ample, and let be a Weierstrass fibration with smooth projective total space and zero section . Locally over , the fibration is defined in Weierstrass form by
We assume that the discriminant divisor is reduced with only ordinary cusps at the transverse intersections of and , and that all fibers of are integral. We further assume that has trivial canonical bundle, , is torsion-free, and the Picard and curve groups decompose integrally as
where and is the class of an elliptic fiber. Since the fibers are integral, Wazir’s Shioda–Tate formula gives for the group of rational sections [53 Theorem 3.3]. The first integral hypothesis in (1.2) therefore forces the global Mordell–Weil rank to be zero.
To obtain sections of positive height, we restrict to a smooth connected curve whose preimage is smooth. Set
For , write for the scheme-theoretic fiber. Its class is on and on . By a section of , we mean the image of a morphism right inverse to . Under fiberwise addition on the generic fiber, the sections form the Mordell–Weil group with identity . We write for the height pairing on this group and call the height of [49 Definition 8.5]. Since every fiber of is integral, Shioda’s formula [49 Theorem 8.6] gives
and in particular . By Proposition 2.1, is the fiber degree of in . § 2 compares this degree with the Jacobi index for a global section and distinguishes it from the discrete charge in Schimannek’s GV series.
To record the added fibers and their multiplicities, let denote the Hilbert scheme of length- zero-dimensional subschemes of , let be the -th symmetric product of (with a single reduced point), and let be the Picard variety of degree- line bundles on . Pushing the zero-cycle forward along defines the morphism
We now pass from these cycles to the sheaves they support. For an effective class , let be Simpson’s projective moduli scheme [50 Theorem 1.21] of Gieseker stable pure one-dimensional sheaves on with class and . Purity means that every nonzero subsheaf has one-dimensional support, and we write for the fundamental cycle weighted by generic lengths. Since , stability is independent of the ample polarization and is equivalent to requiring for every proper nonzero subsheaf . In particular, semistability coincides with stability [38 Remark 3.2]. Sending a sheaf to its support cycle defines the Hilbert–Chow morphism
from the underlying reduced scheme to the Chow variety of effective one-cycles of class [45 Corollary 7.15][38 Remark 3.6].
The role of in the GV count comes from the BPS interpretation in terms of spins [15, 31]. Hosono, Saito, and Takahashi use the relative Lefschetz action for the support morphism to define the invariants [22]. Kiem and Li introduce vanishing cycles into this construction [32]. We use Maulik and Toda’s definition, which extracts the GV numbers from the direct image of vanishing cycles under [38 Definition 3.7].
To specify the vanishing cycles, fix a trivialization of . Then carries a natural -critical structure [5 Corollary 3.19][38 Theorem 3.3] recording local presentations as critical loci of regular functions on smooth varieties [29]. An orientation is a line bundle on whose square is the virtual canonical line bundle [29 Definition 2.31], and it determines a global perverse sheaf of vanishing cycles by Brav, Bussi, Dupont, Joyce, and Szendrői [6 Theorem 6.9].
Following [38 Definition A.4], an orientation on a proper open and closed subscheme is Calabi–Yau with respect to if its line bundle is pulled back from the intermediate variety in the Stein factorization of . For such an orientation, the GV contributions are defined by
where denotes middle perverse cohomology and . When is smooth and endowed with its canonical orientation , one has . The following theorem identifies such a component for each infinitesimally rigid section and each .
Theorem 1.1. Let be a section of whose normal bundle satisfies . For every integer , the locus of stable sheaves with support cycle
forms a smooth projective connected component of , equipped with an isomorphism
Under this identification, the support morphism factors through the projection onto and sends to , where . Moreover, the canonical orientation on is Calabi–Yau with respect to the support morphism. Let be the genus- GV contribution of in (1.5), and put . With recording the number of added fibers, we have
In particular, unless , and
For a fixed pair , the section starts in fiber degree , so its contribution with added fibers occurs in degree . In (8) the Picard variety contributes , while the maps record the added fibers in and give the infinite product. In this sense (8) is the pointwise version of Conjecture A.1, in which varies.
For a rational elliptic surface, the fixed-section product already appears in the relative Lefschetz calculation of Hosono, Saito, and Takahashi [22 Theorem 4.7]. In the compact threefold one must also determine when this surface calculation gives a contribution to the ambient GV invariant. Infinitesimal rigidity makes the fixed-section locus a smooth component of the threefold moduli space, on which we determine the orientation. Over a single line in , Example 5.3 gives six rigid sections in the same ambient class. For very general coefficients in this example, Corollary 5.4 computes the contribution of every nonzero section in the rank-two subgroup they generate.
There is a related local calculation. When all singular fibers of are nodal, Bryan and Kool compute the connected unweighted Donaldson–Thomas series of the total space of [8 Corollary 2]. Setting their variable gives times the product in (8).
Finally, when is a -curve, every section is infinitesimally rigid and is a point. We can therefore sum the fixed-section contributions to obtain the complete GV series in classes . § 6 identifies the resulting product with the one--string elliptic genus.
§ 2 and § 3 relate the heights of sections to their ambient curve classes and describe the moduli space over a fixed divisor. § 4 computes the perverse generating series, § 5 proves the rigidity and orientation statements, and § 6 gives the specialization.
Heights of sections and ambient curve classes
On the smooth surface , we regard curves as divisors. Let denote the Néron–Severi group of line bundles on modulo algebraic equivalence, endowed with the intersection pairing. Relative duality for , adjunction on and , and the zero section yield the surface identities
Since , we have . The Leray spectral sequence for therefore gives , and pullback along induces an isomorphism on identity components of Picard schemes,
with inverse given by restriction to .
Proposition 2.1. For a section , define
Then gives an isomorphism of groups under which the height pairing takes the form . Moreover,
The group is torsion-free and contains only finitely many sections of bounded height, and every nonzero section satisfies .
Proof. Since every fiber of is integral, the local correction terms from reducible fibers in Shioda’s height formula vanish, giving [49 Theorem 8.6]. For , we have and hence , which shows that has no torsion. The Shioda–Tate theorem identifies with [49 Theorem 1.3].
The classes span a free subgroup, and the quotient is torsion-free. Hence is torsion-free as well.
With respect to the basis , the intersection form on has matrix
which has determinant and signature . Orthogonal projection off is therefore defined over , yielding the direct sum decomposition together with the projection and self-intersection . By the Hodge index theorem, the intersection form on is negative definite, so only finitely many elements of can have bounded height.
Finally, for any class , combining the projection formula with (1.2) gives
Indeed, pairing with pullbacks of divisors from determines the coefficient of , while intersecting with and using determines the coefficient of . It follows that , and evaluating (2.4) on yields . □
Proposition 2.1 applies to a section of defined only over . To compare with the height divisor of a global section, let be a smooth elliptic Calabi–Yau threefold with zero section and no reducible fibers in codimension one. Wazir defines a pairing of divisors on with values in by [53 Definition 3.2]. Write also for the divisor of a section. In the convention of Morrison and Park [39 equations (3.4)–(3.8)], the Shioda class of and the height pairing of are
Thus the divisor-valued height pairing is the negative of Wazir’s pairing on the Shioda classes. Put . If is smooth with integral fibers and , then in . The projection formula gives
The number is the Jacobi index of the charge variable in base class [37 equation (3.45)]. Fierro Cota, Klemm, and Schimannek give the corresponding index matrix for several sections [13 equations (4.32)–(4.33)]. Schimannek uses the convention and writes the same index as [47 equations (4.3), (4.8)].
In (1.2), is torsion-free and records the ordinary fiber degree. The discrete charges in Schimannek’s genus-one examples arise from torsion curve classes, weighted by characters of a flat -field [47 equations (3.12)–(3.13)]. The Tate–Shafarevich group parametrizes the locally trivial genus-one torsors of the Jacobian [12]. In Schimannek’s example over , the two sectors in fiber degree one and base degree zero sum to the smooth-model invariant [47 §6.3, Tables 6–7]. Toda computes the local invariants in every pure fiber class of the smooth Weierstrass model [51 Theorem 3.5]. For the torsion grading, Katz, Klemm, Schimannek, and Sharpe propose a definition using stable sheaves on small resolutions of the nodal model [30 §3.5].
Moduli of stable sheaves over a fixed curve
We use the relative Fourier–Mukai equivalence of Bridgeland and Maciocia [7], with the normalization and moduli comparison of Diaconescu [11 §3.1 and Theorem 1.1]. Let be the two projections and the relative diagonal with ideal sheaf . Define the normalized kernel and integral transform by
Here is identified with the relative compactified degree-zero Picard scheme over , which parametrizes rank-one torsion-free sheaves of degree zero on the fibers, via , where is the ideal of in its fiber. On a smooth fiber over , this is . See [11 §3.1] for this normalization and the resulting derived equivalence.
For , put . Diaconescu’s comparison theorem [11 Theorem 1.1 and equations (3.8)–(3.9)] identifies with the moduli space of adiabatically semistable pure two-dimensional sheaves on with Chern character
where we identify and with homology classes via Poincaré duality and view as a rational number by taking degree. Concretely, if we fix polarizations and with and chosen relative to the numerical class, adiabatic semistability means Gieseker semistability with respect to for all . This comparison is an isomorphism of moduli stacks compatible with flat families, while itself does not depend on . The pushforward of any rank-one torsion-free sheaf on the smooth integral surface is Gieseker stable with respect to every ample divisor on , and hence belongs to the adiabatically stable locus whenever its Chern character is (3.2).
Next, we construct the projection to the linear system ,
where denotes the scheme-theoretic fiber over . For a family of stable sheaves on flat over , the relative transform with kernel (3.1) gives a -flat family of pure codimension-one sheaves on . The divisorial support of each fiber is the sum of its support divisors weighted by the lengths at their generic points. In families these divisors form a relative effective Cartier divisor on whose formation commutes with base change [45 §11, Corollary 11.8][36, Lemmas 7.29 and 7.32]. Since , the line bundle associated with this relative divisor agrees with up to twist by a line bundle from . Because , the divisor is pulled back from a unique relative effective divisor on in . The construction descends to because it commutes with base change and twisting a family by a line bundle from . To identify its value at , localize at the generic point of each component of . Filter the localized finite-length sheaf by powers of a uniformizer of . Over an algebraic closure of the residue field, the graded pieces have a total of point factors. Each point factor transforms to a rank-one sheaf on the elliptic fiber [11 Lemma 3.1]. Hence the multiplicity of in the divisorial support is the same. Components over closed points of contribute no divisor, so .
On an elliptic surface with integral fibers, Yoshioka uses the shift and a subscheme of length in his Fourier–Mukai correspondence [54 §3.2, equations (3.13)–(3.15) and Theorem 3.15]. For a rational elliptic surface, Hosono, Saito, and Takahashi give the support morphism in a fixed section class and attribute the description to Yoshioka [22 Proposition 4.8]. We identify the scheme-theoretic fiber in the compact threefold moduli space.
Theorem 3.1. The scheme-theoretic fiber in (3.3) is smooth and decomposes as
Under this isomorphism, a point in the component indexed by corresponds to a stable sheaf with Fourier–Mukai transform
where is the ideal sheaf of . The multiplicities in the support cycle of (1.6) are determined by .
Proof. Let be a geometric point of and set . Because the divisorial support of is with multiplicity one, has length one at the generic point of and is annihilated there by any local defining equation of . Multiplication on by a local equation of therefore has image supported in dimension at most one. Purity forces this image to vanish, so for a rank-one torsion-free sheaf on . Since is a smooth surface, is a line bundle and for a unique zero-dimensional subscheme .
Put , , and . Applying Grothendieck–Riemann–Roch to yields
Combining (15), (10), and (13), we find
By Proposition 2.1, there exists a unique section such that
which by (11) is equivalent to for a unique line bundle . Furthermore,
which establishes the classification of closed points together with the numerical relations in (18). Conversely, every sheaf of the form (18) has Chern character (15) and is Gieseker stable for every polarization, so its inverse transform defines a point of .
Now choose a Poincaré line bundle on and the universal ideal sheaf on . The resulting family gives a morphism to . Since the source is reduced and maps every geometric point to , this morphism factors through . The map from the disjoint union of the products is proper and bijective on geometric points. To check the scheme structure, consider the change of rings sequence
where identifies the normal term. Since and ,
On , the sheaf is invertible. In a first-order deformation, a local equation of acts on a lift of a local generator as multiplication by , where . The normal map sends this deformation to , and its divisorial support has equation . Hence and agree on . They agree globally as sections of , giving .
Since , Serre duality gives . Riemann–Roch gives . With , we obtain
By Fogarty’s theorem [14 Theorem 2.4], each product is smooth of dimension . The family map is finite by properness and quasi-finiteness. It is a homeomorphism, so the local dimension at each closed point equals the tangent dimension. Thus is regular. In characteristic zero the map is birational on each component, hence an isomorphism by normality.
It remains to determine the support cycle. Write . Restricting the kernel (14) along gives up to a pullback from . Thus is a line bundle on , extended by zero to . For , let be its skyscraper sheaf. Then is rank one and torsion-free on [11 Lemma 3.1 and equation (3.5)]. A composition series of shows that is a sheaf filtered by the , each with multiplicity . Applying to therefore gives a short exact sequence
Additivity of fundamental cycles gives , as claimed.
Perverse direct images and generating series
Recall that sends a length- subscheme to the divisor under the elliptic fibration . Define
Thus records the Euler characteristics on of the perverse cohomology sheaves of . Hosono, Saito, and Takahashi discuss the extension of their rational elliptic surface calculation to other elliptic surfaces in the remark following the proof of [22, Theorem 4.7].
Lemma 4.1. With and , one has
Proof. Let denote the open locus over which is smooth, and set with intersection complex . The decomposition theorem of Beilinson, Bernstein, Deligne, and Gabber [4 Theorem 6.2.5], applied to , gives
At , the multiplicity of a skyscraper summand is one less than the number of irreducible components of , hence zero. This also follows by comparing degree-two stalks, since , and applying Verdier duality.
By the canonical bundle formula for elliptic surfaces and Noether’s formula, the topological Euler characteristic of is . Since , we obtain . Taking hypercohomology in (22) shows that vanishes for and has dimension for . The nonzero bigraded hypercohomology spaces are therefore recorded by
For a partition of , consider the finite map
Pushing forward the Hilbert–Chow decomposition [17 Theorem 3 and Proposition 1] (in the form of [10 Theorem 4.1.1]) to gives an isomorphism of rational constructible complexes
where denotes the invariant summand of the finite pushforward of to with the usual Koszul signs. We omit Tate twists for rational complexes. For the induced perverse filtration on the Hilbert scheme, see [55 Proposition 4.12] and [48 Proposition 2.2].
Since finite pushforwards are perverse exact [4 §2.2.6], external products and the Künneth formula reduce the calculation to the bigraded vector space in (23), with permutation parity and Euler characteristic sign . For each part size , the three rows in (23) contribute , , and , respectively, and multiplying over proves (21). $\square`
Setting in Lemma 4.1 recovers the Euler characteristic specialization of Göttsche’s formula [16],
For a fixed section and added fibers, the total fiber degree is . The morphism
is injective on geometric points since the one-cycle recovers its fiber multiplicities, and it is finite by properness and quasi-finiteness. On the component of Theorem 3.1, the support map is , where is projection to . The Künneth formula gives
Here is the constant sheaf on the product shifted by . Since finite pushforward is perverse exact and
the contribution of the intersection complex on is .
Proposition 4.2. Fix . On the fiber over the projected curve , let be the sum of the constant sheaves on the components of {(xref)}(3.4), each shifted by the dimension of that component. Define
Then
Proof. By Proposition 2.1, the height bound ensures that each coefficient of is finite. Summing (25) over the components of {(xref)}(3.4) gives
Multiplying by and summing over gives {(xref)}(4.6) by Lemma 4.1. □
Formula {(xref)}(4.6) concerns the intrinsic intersection complexes on the fixed scheme-theoretic fiber . When or deforms in , this fiber need not be open in , and the next section determines when the component calculation computes the ambient GV invariants {(xref)}(1.5).
Normal deformations and vanishing cycles
Proposition 5.1. Let be the sheaf on from {(xref)}(3.5), with . Then
Proof. The differential of splits along , so is smooth there. By adjunction, . The normal bundle sequence for is therefore
Let be the connecting map. As , its kernel is . Moreover, the trace map is an isomorphism, since its Serre dual is the isomorphism given by the identity on .
For , let be the corresponding deformation over . Choose a local equation for and a lift of , so that has equation . For a flat deformation of , multiplication by identifies with . Under this identification, multiplication by induces the endomorphism of given by , where is the normal map in {(xref)}(3.7). Thus $\eta(E)=v precisely when on each such chart. Equivalently, is the pushforward of a flat lift of to . Exactness of (20) gives
Write for the Kodaira–Spencer class of and for the Atiyah class of . Since is smooth, is perfect for every . Its lifting obstruction is , whose trace is the obstruction to lifting [27 Corollary 3.4 and §4.2]. The trace is an isomorphism, so lifts if and only if lifts. Now
The factor extends from , and extends to the induced deformation of since . Hence precisely when lifts to . We show that this is equivalent to lifting the divisor . Let be its canonical section. On every integral genus-one fiber has degree one, so spans and . Cohomology and base change give
Indeed, by (10).
If lifts to over , choose a lift . The obstruction to lifting lies in . Twisting by a line bundle with transition functions changes that obstruction by the class of . The displayed isomorphism allows us to cancel it. The lifted section cuts out a relative Cartier divisor lifting . Conversely, a lifted Cartier divisor gives a lift of .
The normal sequence (28) says that lifts inside precisely when . Hence
Substitution into (20) proves (27).
Proof of Theorem 1.1. By Theorem 3.1, the product in (7) is a smooth projective connected component of the moduli fiber . Since the Fourier–Mukai equivalence preserves , Proposition 5.1 and the vanishing imply that the ambient tangent dimension at each closed point of is . At such a point the local dimension lies between and the tangent dimension. The ambient local ring is therefore regular. The ideal of in each such regular local ring is zero, since a nonzero ideal in a local domain lowers the dimension of the quotient. Thus the closed immersion is also open and identifies with a connected component of .
Let be the second projection. The line bundle on descends to a line bundle on because point stabilizers act trivially on its fibers. Since the Picard variety has trivial canonical bundle, , and the Hilbert–Chow morphism is crepant, the local construction of the canonical bundle of in [3 §6] shows that is the pullback of the symmetric descent of on , and hence
The map sends to . Over with distinct and smooth fibers , its fiber is . Thus the proper surjective map has geometrically connected generic fiber, also for . Since its source is smooth and connected and its target is normal, Stein factorization gives
The map from to the Chow variety adds to the fiber cycle. It is finite, so this is the Stein factorization of the support morphism. On a smooth -critical scheme, the virtual canonical line bundle is [29 Theorem 2.28], and (30) shows that its square root descends to the Stein factor and therefore defines a Calabi–Yau orientation in the sense of [38 Definition A.4]. With this orientation, the smooth critical chart with zero potential gives [6 Theorem 6.9]
Consequently, the left-hand side of (5) is , and Lemma 4.1 proves (8).
Finally, each is a symmetric Laurent polynomial with exponents between and , and hence is an integral polynomial in of degree at most . Multiplying by gives the genus bound, while extracting the coefficient of in (21) gives , proving (9). ∎
Remark 5.2. For and a smooth plane curve of degree , we have and . In this case, the vanishing condition in Theorem 1.1 is equivalent to the injectivity of
from (28), where both source and target have dimension . By Proposition 2.1, every nonzero section satisfies , which for matches the threshold for additional sections in HKKW [25 §5.1.7].
Example 5.3. Let be homogeneous coordinates on and let . Choose Weierstrass coefficients
where and are general homogeneous polynomials of the indicated degrees. Along , the discriminant restricts to , which has 36 simple zeros, and hence meets the discriminant transversely. Off , varying and prescribes the first jets of independently. For general coefficients the pair is transverse to the smooth part of and to its cusp point. Thus is smooth with the required discriminant, and is smooth.
The smooth Weierstrass models form an open subset of the affine space of coefficients. Its general member is the crepant resolution of a degree-18 hypersurface in [24 §2]. The five rays corresponding to the vertices generate the toric lattice . The formulas of Batyrev and Kreuzer [2 Theorems 1.6 and 3.8] therefore give trivial fundamental and Brauer groups. In particular and are torsion-free, using the exponential and universal coefficient sequences. The same torsion vanishing holds throughout the smooth family. Diaconescu’s integral bases modulo torsion [11 Lemma 2.1] now give (2).
In the chart , let be the section over given by
Indeed, for these three values of . Each section is disjoint from , so . Any two meet only at , where and are local coordinates on since . Their local intersection has length six. By Proposition 2.1 the height matrix of and is
of determinant 27. Thus and are independent. The three displayed points are collinear on the generic cubic, so their Mordell–Weil sum is .
Differentiating the Weierstrass equation along gives
where . At , the three summands on global sections span the monomials with , respectively. Hence is an isomorphism for all three sections for general . Each normal bundle has degree and no global sections, so . Inversion gives three further rigid sections with . Thus in the single ambient class , Theorem 1.1 gives at least six isolated points, one for each of the six sections.
More generally, write for the Mordell–Weil multiple by a nonzero integer . On the smooth locus of , the Weierstrass group law is regular [28 Corollary 12]. Multiplication by is étale there since its relative differential is multiplication by after translation. It maps isomorphically to over , identifying their normal bundles. Since , every such section gives a component isomorphic to in class for each . For , its GV contributions are
Let . At , (8) gives . This is the factor for adding fibers to an isolated section over a line in Greer’s genus-zero GW calculation [18 p. 3, before Conjecture 1.6]. Greer uses Noether–Lefschetz theory to count rational curves of base degree one on very general Weierstrass fibrations [19].
Corollary 5.4. In Example 5.3, put and . For very general , every nonzero section in is infinitesimally rigid in . With recording total fiber degree, the combined GV contribution of these sections is
@page 10/14 starts_mid=0 ends_mid=1 Remark 5.2. For and a smooth plane curve of degree , we have and . In this case, the vanishing condition in Theorem 1.1 is equivalent to the injectivity of
from (28), where both source and target have dimension . By Proposition 2.1, every nonzero section satisfies , which for matches the threshold for additional sections in HKKW [25 §5.1.7].
Example 5.3. Let be homogeneous coordinates on and let . Choose Weierstrass coefficients
where and are general homogeneous polynomials of the indicated degrees. Along , the discriminant restricts to , which has 36 simple zeros, and hence meets the discriminant transversely. Off , varying and prescribes the first jets of independently. For general coefficients the pair is transverse to the smooth part of and to its cusp point. Thus is smooth with the required discriminant, and is smooth.
The smooth Weierstrass models form an open subset of the affine space of coefficients. Its general member is the crepant resolution of a degree-18 hypersurface in [24 §2]. The five rays corresponding to the vertices generate the toric lattice . The formulas of Batyrev and Kreuzer [2 Theorems 1.6 and 3.8] therefore give trivial fundamental and Brauer groups. In particular and are torsion-free, using the exponential and universal coefficient sequences. The same torsion vanishing holds throughout the smooth family. Diaconescu’s integral bases modulo torsion [11 Lemma 2.1] now give (2).
In the chart , let be the section over given by
Indeed, for these three values of . Each section is disjoint from , so . Any two meet only at , where and are local coordinates on since . Their local intersection has length six. By Proposition 2.1 the height matrix of and is
of determinant 27. Thus and are independent. The three displayed points are collinear on the generic cubic, so their Mordell–Weil sum is .
Differentiating the Weierstrass equation along gives
where . At , the three summands on global sections span the monomials with , respectively. Hence is an isomorphism for all three sections for general . Each normal bundle has degree and no global sections, so . Inversion gives three further rigid sections with . Thus in the single ambient class , Theorem 1.1 gives at least six isolated points, one for each of the six sections.
More generally, write for the Mordell–Weil multiple by a nonzero integer . On the smooth locus of , the Weierstrass group law is regular [28 Corollary 12]. Multiplication by is étale there since its relative differential is multiplication by after translation. It maps isomorphically to over , identifying their normal bundles. Since , every such section gives a component isomorphic to in class for each . For , its GV contributions are
Let . At , (8) gives . This is the factor for adding fibers to an isolated section over a line in Greer’s genus-zero GW calculation [18 p. 3, before Conjecture 1.6]. Greer uses Noether–Lefschetz theory to count rational curves of base degree one on very general Weierstrass fibrations [19].
Corollary 5.4. In Example 5.3, put and . For very general , every nonzero section in is infinitesimally rigid in . With recording total fiber degree, the combined GV contribution of these sections is
where is the contribution from added fibers in (21).
Proof. Keep fixed and specialize . The coordinates then satisfy the Weierstrass equation modulo , so extends to a section over the doubled line . The normal map for is independent of , hence is rigid for by Example 5.3. Translation by the extended section on the smooth group locus over identifies the first infinitesimal neighborhoods of and in . Indeed, the square of the ideal of pulls back into the square of the ideal of either section. Their normal bundles are therefore isomorphic.
For each fixed the curve is independent of the coefficients. By upper semicontinuity of of its normal bundle, rigidity holds on a nonempty open set when . For and the same conclusion follows from Example 5.3. Outside the countable union of the complementary closed sets all these sections are rigid. Their half-heights are by the height matrix above. Theorem 1.1, with , gives the stated sum. □
The complete formula over a -curve
Suppose now that with and smooth, so that and is a rational elliptic surface. Let
be the lattice with its standard Euclidean inner product, and let
be its theta series, where is the normalized Eisenstein series of weight four.
Corollary 6.1. For , the full moduli scheme and its canonically oriented sheaf of vanishing cycles are
Here on the component indexed by . With , the GV invariants satisfy
In particular,
and the genus-zero series is
Proof. Since , any effective divisor on linearly equivalent to must contain with trivial residual divisor, so is a point and . Moreover, for every section , the sequence (28) becomes
which splits and gives , so Theorem 1.1 applies to every component.
Because is a rational elliptic surface with integral fibers, the orthogonal complement of in its unimodular Néron–Severi lattice is negative definite, even (since ), and unimodular of rank eight, so its Mordell–Weil lattice is isomorphic to [49, 22]. Proposition 2.1 therefore gives , and combining Theorem 3.1 with (8) yields (33) and (34).
At degree in (34), the maximal exponent of is , which arises solely from the factor paired with the constant terms of all other factors and has coefficient . Since has leading term , this gives (35), while setting (that is, ) yields (36). □`
At , the 240 roots of give the decomposition
which accounts for the 240 additional section contributions identified by HKKW for the exceptional curve of the Hirzebruch surface [25 §5.1.4]. The first coefficients of (34) are
The factor multiplying in (34) is the fixed-section product of Hosono, Saito, and Takahashi [22 Theorems 4.7 and 4.10]. Summing over sections supplies the theta series, as in their earlier formula [21 equation (5.18)]. At genus zero, Bryan and Leung compute the contribution of a fixed section class on the rational elliptic surface [9 Theorem 6.2]. Rose and Yui sum over the lattice in their compact example [44 Theorems 7.3–7.4]. Dividing (34) by also recovers Oberdieck and Shen’s PT series in base class , normalized by the pure-fiber PT series [41 Theorem 2].
Klemm, Mayr, and Vafa obtained (36) for a singly wound string [35 equation (4.1) and Table 1]. At all genera, let be the elliptic genus of one E-string with all mass parameters zero and rotation parameters , as in [20 equation (6.30)]. With and , the Jacobi triple product gives
The compact moduli spaces realizing these coefficients are given in (33).
When moves, Proposition 5.1 identifies the extra tangent directions at the fixed-divisor locus with .
Appendix A. The direct image for a family of sections
Let parametrize smooth connected curves for which is smooth. Let be the universal curve and . Write for the relative symmetric product and for the relative Hilbert scheme of points. The maps give the proper morphism
For each integer , let be the reduced locus in the relative Hilbert scheme of parametrizing pairs with a section satisfying . For , adding fibers gives a cycle in class . Define
At , . Over , write for the direct image of vanishing cycles under the support map.
Conjecture A.1. For a choice of orientations that restrict to the canonical ones in Theorem 1.1, there are isomorphisms of constructible complexes on
where and are the projections to and , respectively.
To compare with Theorem 1.1, fix . On , proper base change identifies the right-hand side of (A.1) with
If , Theorem 1.1 at identifies with . For every , the sheaves over these cycles form the open and closed component . On this slice is the finite injective map of Section 4. Pulling back (25) gives (A.1). Taking the perverse Euler polynomial on gives , the product of the Picard and Hilbert scheme contributions. Summing over gives (8). For a moving or obstructed section, (25) computes the intrinsic contribution over , while (A.1) retains the ambient stalk .
For and of degree , a section with added fibers has total fiber degree . Its cycle meets in degree
For the section over a line in Example 5.3, . At the curve is disjoint from , as in HKKW’s charge-zero contribution [25 §5.1.7]. HKKW’s Wilson-loop sum ranges over residual plane degrees [25 §3.4.1, equation (3.58)]. Conjecture A.1 is the author’s proposed description of the direct image over the smooth-curve locus. Deriving the coefficients of HKKW’s sum from that direct image remains open.
Use of generative AI
The author used ChatGPT (OpenAI) to help draft and revise the manuscript and to check steps in the proofs. The author takes responsibility for the final text and its mathematical content.
Acknowledgements
I thank Sheldon Katz for introducing me to this problem. I also thank Min-xin Huang, Kaiwen Sun, Xin Wang, and Yukinobu Toda for useful discussions. This work was supported by JSPS KAKENHI Grant Number JP25K17226. Kavli IPMU is supported by the World Premier International Research Center Initiative (WPI), MEXT, Japan.
References
Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo, Japan
Email address: [email protected]
References
- [1]M. Alim and E. Scheidegger, Topological strings on elliptic fibrations, Commun. Number Theory Phys. 8 (2014), 729–800. [arXiv:1205.1784].arxiv.org/abs/1205.1784
- [2]V. Batyrev and M. Kreuzer, Integral cohomology and mirror symmetry for Calabi–Yau 3-folds, in Mirror Symmetry V, AMS/IP Stud. Adv. Math. 38, Amer. Math. Soc., 2006, 255–270. [arXiv:math/0505432].DOI
- [3]A. Beauville, Variétés kählériennes dont la première classe de Chern est nulle, J. Differential Geom. 18 (1983), 755–782. [doi:10.4310/jdg/1214438181].
- [4]A. A. Beilinson, J. Bernstein and P. Deligne, Faisceaux pervers, Astérisque 100 (1982), 5–171.
- [5]O. Ben-Bassat, C. Brav, V. Bussi and D. Joyce, A ’Darboux theorem’ for shifted symplectic structures on derived Artin stacks, with applications, Geom. Topol. 19 (2015), 1287–1359. [arXiv:1213.0090].
- [6]C. Brav, V. Bussi, D. Dupont, D. Joyce and B. Szendrői, Symmetries and stabilization for sheaves of vanishing cycles, with an appendix by J. Schürmann, J. Singul. 11 (2015), 85–151. [arXiv:1211.3259].DOI
- [7]T. Bridgeland and A. Maciocia, Fourier–Mukai transforms for K3 and elliptic fibrations, J. Algebraic Geom. 11 (2002), 629–657. [arXiv:math/9908022].arxiv.org/abs/math/9908022
- [8]J. Bryan and M. Kool, Donaldson–Thomas invariants of local elliptic surfaces via the topological vertex, Forum Math. Sigma 7 (2019), e7. [arXiv:1608.07369].DOI
- [9]J. Bryan and N. C. Leung, The enumerative geometry of K3 surfaces and modular forms, J. Amer. Math. Soc. 13 (2000), 371–410. [arXiv:alg-geom/9711031].DOI
- [10]M. A. A. de Cataldo and L. Migliorini, The Douady space of a complex surface, Adv. Math. 151 (2000), 283–312. [arXiv:math/9811159].
- [11]D.-E. Diaconescu, Vertical sheaves and Fourier–Mukai transform on elliptic Calabi–Yau threefolds, Commun. Number Theory Phys. 10 (2016), no. 3, 373–431. [arXiv:1509.07749].
- [12]I. Dolgachev and M. Gross, Elliptic threefolds I. Ogg–Shafarevich theory, J. Algebraic Geom. 3 (1994), 39–80. [arXiv:alg-geom/9210009].
- [13]C. Fierro Cota, A. Klemm and T. Schimannek, Topological strings on genus one fibered Calabi–Yau 3-folds and string dualities, J. High Energy Phys. 11 (2019), 170. [arXiv:1910.01988].
- [14]J. Fogarty, Algebraic families on an algebraic surface, Amer. J. Math. 90 (1968), 511–521.DOI
- [15]R. Gopakumar and C. Vafa, M-theory and topological strings—I, II, 1998. [arXiv:hep-th/9809187], [arXiv:hep-th/9812127].
- [16]L. Göttsche, The Betti numbers of the Hilbert scheme of points on a smooth projective surface, Math. Ann. 286 (1990), 193–207. [doi:10.1007/BF01453572].
- [17]L. Göttsche and W. Soergel, Perverse sheaves and the cohomology of Hilbert schemes of smooth algebraic surfaces, Math. Ann. 296 (1993), 235–245. [doi:10.1007/BF01445104].DOI
- [18]F. Greer, Counting curves on a Weierstrass model, 2017. [arXiv:1701.06596].arxiv.org/abs/1701.06596
- [19]F. Greer, Modular forms from Noether–Lefschetz theory, Algebra Number Theory 14 (2020), no. 9, 2335–2368. [arXiv:1801.00375].DOI
- [20]B. Haghighat, G. Lockhart and C. Vafa, Fusing E-strings to heterotic strings: E + E → H, Phys. Rev. D 90 (2014), 126012. [arXiv:1406.0850].
- [21]S. Hosono, M.-H. Saito and A. Takahashi, Holomorphic anomaly equation and BPS state counting of rational elliptic surfaces, Adv. Theor. Math. Phys. 3 (1999), 177–208. [arXiv:hep-th/9901151].arxiv.org/abs/hep-th/9901151
- [22]S. Hosono, M.-H. Saito and A. Takahashi, Relative Jefschetz action and BPS state counting, Internat. Math. Res. Notices 2001, no. 15, 783–816. [arXiv:math/0105148].DOI
- [23]M.-x. Huang, S. Katz and A. Klemm, Topological string on elliptic CY 3-folds and the ring of Jacobi forms, J. High Energy Phys. 10 (2015), 125. [arXiv:1501.04891].
- [24]M.-x. Huang, S. Katz and A. Klemm, Towards refining the topological strings on compact Calabi–Yau 3-folds, J. High Energy Phys. 03 (2021), 266. [arXiv:2010.02910].DOI
- [25]M.-x. Huang, S. Katz, A. Klemm and X. Wang, Refined BPS numbers on compact Calabi–Yau threefolds from Wilson loops, J. High Energy Phys. 08 (2025), 178. [arXiv:2503.16270].DOI
- [26]M.-x. Huang, K. Lee and X. Wang, Topological strings and Wilson loops, J. High Energy Phys. 08 (2022), 207. [arXiv:2205.02366].DOI
- [27]D. Huybrechts and R. P. Thomas, Deformation-obstruction theory for complexes via Atiyah and Kodaira–Spencer classes, Math. Ann. 346 (2010), 545–569. [arXiv:0805.3527].arxiv.org/abs/0805.3527
- [28]J. Jin, Homogeneous division polynomials for Weierstrass elliptic curves. [arXiv:1303.4327].DOI
- [29]D. Joyce, A classical model for derived critical loci, J. Differential Geom. 101 (2015), no. 2, 289–367. [arXiv:1304.4508].arxiv.org/abs/1304.4508
- [30]S. Katz, A. Klemm, T. Schimannek and E. Sharpe, Topological strings on non-commutative resolutions, Commun. Math. Phys. 405 (2024), 62. [arXiv:2212.08655].DOI
- [31]S. Katz, A. Klemm and C. Vafa, M-theory, topological strings and spinning black holes, Adv. Theor. Math. Phys. 3 (1999), 1445–1537. [arXiv:hep-th/9910181].arxiv.org/abs/hep-th/9910181
- [32]Y.-H. Kiem and J. Li, Categorification of Donaldson–Thomas invariants via perverse sheaves, 2012. [arXiv:1212.6444].arxiv.org/abs/1212.6444
- [33]H.-C. Kim, M. Kim and S.-S. Kim, 5d/6d Wilson loops from blowups, J. High Energy Phys. 08 (2021), 131. [arXiv:2106.04731].DOI
- [34]A. Klemm, J. Manschot and T. Wotschke, Quantum geometry of elliptic Calabi–Yau manifolds, Commun. Number Theory Phys. 6 (2012), 849–917. [arXiv:1205.1795].arxiv.org/abs/1205.1795
- [35]A. Klemm, P. Mayr and C. Vafa, BPS states of exceptional non-critical strings, Nucl. Phys. B Proc. Suppl. 58 (1997), 177–194. [arXiv:hep-th/9607139].DOI
- [37]S.-J. Lee, W. Lerche and T. Weigand, Tensionless strings and the weak gravity conjecture, J. High Energy Phys. 10 (2018), 164. [arXiv:1808.05958].DOI
- [38]D. Maulik and Y. Toda, Gopakumar–Vafa invariants via vanishing cycles, Invent. Math. 213 (2018), 1017–1097. [arXiv:1610.07303].arxiv.org/abs/1610.07303
- [39]D. R. Morrison and D. S. Park, F-theory and the Mordell–Weil group of elliptically-fibered Calabi–Yau threefolds, J. High Energy Phys. 10 (2012), 128. [arXiv:1208.2695].DOI
- [40]G. Oberdieck and A. Pixton, Gromov–Witten theory of elliptic fibrations: Jacobi forms and holomorphic anomaly equations, Geom. Topol. 23 (2019), 1415–1489. [arXiv:1709.01481].DOI
- [41]G. Oberdieck and J. Shen, Curve counting on elliptic Calabi–Yau threefolds via derived categories, J. Eur. Math. Soc. 22 (2020), no. 3, 967–1002. [arXiv:1608.07073].
- [42]R. Pandharipande and R. P. Thomas, Curve counting via stable pairs in the derived category, Invent. Math. 178 (2009), 407–447. [arXiv:0707.2348].DOI
- [43]B. Pioline and T. Schimannek, Revisiting the quantum geometry of torus-fibered Calabi–Yau threefolds, SciPost Phys. 21 (2026), 037. [arXiv:2510.23722].DOI
- [44]S. Rose and N. Yui, Elliptic Calabi–Yau threefolds over a del Pezzo surface, in Arbeitstagung Bonn 2013, Progr. Math. 319, Birkhäuser, 2016, 337–354. [arXiv:1312.0911].arxiv.org/abs/1312.0911
- [45]D. Rydh, Families of cycles, Paper IV in Families of cycles and the Chow scheme, doctoral thesis, KTH, Stockholm, 2008.
- [46]T. Schimannek, Modularity from monodromy, J. High Energy Phys. 05 (2019), 024. [arXiv:1902.08215].DOI
- [47]T. Schimannek, Modular curves, the Tate–Shafarevich group and Gopakumar–Vafa invariants with discrete charges, J. High Energy Phys. 02 (2022), 007. [arXiv:2108.09311].DOI
- [48]J. Shen and Z. Zhang, Perverse filtrations, Hilbert schemes, and the P = W conjecture for parabolic Higgs bundles, Alg. Geom. 8 (2021), no. 4, 465–489. [doi:10.14231/AG-2021-014].DOI
- [49]T. Shioda, On the Mordell–Weil lattices, Comment. Math. Univ. St. Paul. 39 (1990), no. 2, 211–240.DOI
- [50]C. T. Simpson, Moduli of representations of the fundamental group of a smooth projective variety I, Publ. Math. Inst. Hautes Études Sci. 79 (1994), 47–129. [doi:10.1007/BF02698877].
- [51]Y. Toda, Gopakumar–Vafa invariants and wall-crossing, J. Differential Geom. 123 (2023), 141–193. [arXiv:1710.01843].
- [52]X. Wang, Wilson loops, holomorphic anomaly equations and blowup equations, 2023. [arXiv:2305.09171].arxiv.org/abs/2305.09171
- [53]R. Wazir, Arithmetic on elliptic threefolds, Compos. Math. 140 (2004), no. 3, 567–580. [arXiv:math/0112259].DOI
- [54]K. Yoshioka, Moduli spaces of stable sheaves on abelian surfaces, Math. Ann. 321 (2001), no. 4, 817–884. [arXiv:math/0009001].
- [55]Z. Zhang, Multiplicativity of perverse filtration for Hilbert schemes of fibered surfaces, Adv. Math. 312 (2017), 636–679. [arXiv:1512.04643].arxiv.org/abs/1512.04643