Extremal transitions via quantum Serre duality
Introduction
Two varieties Z and Z ~ are said to be related by extremal transition if there exists a degeneration from Z to a singular variety Z ¯ and a crepant resolution Z ~ → Z ¯ . In this paper we compare the genus-zero Gromov–Witten theory of toric hypersurfaces related by extremal transitions arising from toric blow-up. We show that the quantum D-module of Z ~ , after analytic continuation and restriction of a parameter, recovers the quantum D-module of Z. The proof provides a geometric explanation for both the analytic continuation and restriction parameter appearing in the theorem.
Two smooth projective varieties and are said to be related by an extremal transition if there exists a singular variety together with a projective degeneration from to and crepant resolution . Topologically the spaces and are related by a surgery.
Motivation for studying extremal transitions comes from birational geometry, where they provide a bridge between different connected components of moduli. A famous conjecture known as Reid's fantasy [[37]] predicts that any pair of deformation classes of smooth Calabi–Yau 3-folds may be connected via a sequence of extremal transitions. Phrased another way, if we consider two varieties as equivalent if they are related by a crepant resolution, then the moduli space of Calabi–Yau 3-folds is connected. This has been verified for a large class of examples [[4], [11]].
Furthermore, Morrison [[35]] showed that extremal transitions are naturally compatible with mirror symmetry. It was checked in many cases that if is mirror to , then a mirror to may be constructed by applying a "dual transition" to , by first contracting and then smoothing. This relationship was conjectured to hold generally.
The two conjectures together suggest a tantalizing strategy for understanding mirror symmetry generally. If one can determine how both the A-model (Gromov–Witten theory) and the B-model (variation of Hodge structures) vary under extremal transitions, then one could—in principle—prove mirror symmetry for an arbitrary Calabi–Yau 3-fold by connecting it via extremal transitions to an example where the mirror theorem is known. The first steps towards this ambitious goal were initiated by Li–Ruan [[30]], where the behavior of Gromov–Witten theory was determined for small transitions of 3-folds, a type of extremal transition where the the exceptional locus of consists of finitely many rational curves. For the specific case of conifold transitions between Calabi–Yau 3-folds, the A and B models together were systematically studied by Lee et al. [[29]].
Beyond the setting of mirror symmetry and 3-folds, the interplay between Gromov–Witten theory and birational geometry is a rich subject in its own right. One of most prominent and well-studied examples of this is the crepant transformation conjecture (also known as the crepant resolution conjecture) which predicts a deep relationship between the Gromov–Witten invariants of smooth varieties related by crepant birational transformation. While it is not possible to give a complete account of all of the progress on this subject here, examples of results of this type may be found in [[10], [15]–[17], [28]].
In contrast to the crepant transformation conjecture, our understanding of the behavior of Gromov–Witten theory under extremal transitions is much less complete. With the notable exception of small transitions between 3-folds [[29]], there are only a few examples of extremal transitions where the Gromov–Witten theory has been studied.
These examples were studied by Iritani–Xiao in [[27]], and by the first author in [[33]]. Prior to this, it was not clear what kind of statement one should expect for the change of Gromov–Witten theory under general extremal transitions in arbitrary dimensions.
Results
In this paper, we compare the genus-zero Gromov–Witten theory of toric hypersurfaces related by an extremal transition induced from a toric blow-up on the ambient space. This includes a very large family of interesting examples, including non-small extremal transitions in arbitrary dimensions. The setup is roughly as follows.
Let be a smooth toric Deligne–Mumford stack with projective coarse moduli space and let be the blow-up of along a torus-invariant subvariety . Let be a hypersurface in , defined by the vanishing of a general section of a semi-ample line bundle. By degenerating the section appropriately, we obtain a variety which acquires singularities along . Under certain numerical conditions, the proper transform of in , denoted by , will be a crepant resolution of . In this case and are related by extremal transition through the singular variety (see Sect. 4 for details). In the case where and are Calabi–Yau, this was one of the cases considered by Morrison in formulating his conjectural connection to mirror symmetry [[35], Section 3.2].
The above method of constructing extremal transitions reproduces many well-known examples, including, for instance, the following:
Example 1.1
(Conifold transition) Let s be a general degree 5 homogeneous polynomial in the variables , defining a smooth quintic hypersurface in . Degenerate the polynomial s by setting to zero the coefficients of each monomial in s which is not divisible by either or . This new polynomial, , may then be written as , where f and g are general homogeneous polynomials of degree 4. The variety contains and has 16 nodes at
Graph
Let denote the blowup of along , and let denote the proper transform of under the map . Then and are related by a conifold transition.
Our results are formulated in terms of the quantum D-module , a collection of data consisting of:
- the Dubrovin connection , a flat connection lying over the extended Kähler moduli and encoding the genus-zero Gromov–Witten theory of ;
- a pairing , flat with respect to .
More precisely, we consider the ambient quantum D-module of , , defined by restricting the state space to those insertions pulled back from . An additional important piece of data is the -integral structure defined by Iritani. This is a lattice of -flat sections determined by the K-theory .
The main theorem may be paraphrased as follows:
Theorem 1.2
(Theorem 8.2) The Dubrovin connection is analytic along a specified direction , and may be analytically continued to a neighborhood of . The monodromy invariant part around , when restricted to , contains a subquotient which is gauge-equivalent to . This equivalence is compatible with the pairing and integral structures on and .
The form of the correspondence given above is a slight modification of that conjectured by the first author in [[34]]. A similar formulation also appeared in [[27]].
It is interesting to note that the theorem described above consists of two distinct steps: (1) analytic continuation of ; and (2) restriction of flat sections to (restricting to zero). The appearance of analytic continuation is reminiscent of the crepant transformation conjecture, which can be formulated similarly [[15]], but without a restriction of variables. Our proof yields a geometric explanation for both of the above steps.
To our knowledge this is the first result on the Gromov–Witten theory of extremal transitions that includes varieties of arbitrary dimension and includes a large family of non-small extremal transitions. Furthermore, a (somewhat surprising) upshot of our proof is that the integral lattices may also be compared. We expect the techniques developed here to apply more broadly. We plan to explore this in future work, with the hope of better understanding mirror symmetry for a large class of varieties.
Strategy of proof
At the heart of our proof is the use of quantum Serre duality to compare quantum D-modules. Let D and denote the divisors on and which define the hypersurfaces and respectively. Define
Graph
Originally proven by Givental [[23]], quantum Serre duality is the name given to a correspondence between the genus-zero Gromov–Witten theory of and (resp. and ). This was reformulated by the second author in [[38]] as an isomorphism between the ambient quantum D-module of and the compact type quantum D-module of (resp. and ). The compact type quantum D-module is defined by restricting the state space to those insertions which can be represented by classes of compact support.
It therefore suffices to compare the (compact type) quantum D-modules of and . The benefit of this perspective is the existence of an intermediate toric variety, denoted , which relates to both and . The toric variety is obtained from by a flop, and is simultaneously a partial compactification of , compactifying the fibers of .
Figure 1 provides a picture of the fans in the case of , and a point. In this picture the bottom vertex is the origin and each line out of the origin is a primitive ray vector.
Graph: Fig. 1Fans representing the toric varieties (1) T , (2) T¯ , and (3) T~
The proof then proceeds in two steps, each of independent interest. First, using the work of [[15]], we prove a compact type version of the crepant transformation conjecture for non-compact toric varieties. In particular this applies to and .
Theorem 1.3
(Theorem 6.13) Given toric varieties and related by a crepant variation of GIT across a codimension-one wall, the compact type quantum D-modules are gauge equivalent after analytic continuation along a specified path. This equivalence preserves the compact type pairing and identifies the integral lattices.
Second, we use the specific geometry of the partial compactification
Graph
to compare their respective Gromov–Witten invariants.
Theorem 1.4
(Corollary 7.4) The monodromy invariant part of around (where is a coordinate corresponding to the divisor ), when restricted to , contains a submodule which maps surjectively to via . This map is compatible with the flat pairing, and identifies the integral lattice generated by with that generated by .
Theorem 1.2 then proceeds by combining Theorems 1.3 and 1.4 with quantum Serre duality. Schematically we have:
Graph
From this perspective, the variety plays a crucial role in understanding the relationship between the quantum D-modules of and . The appearance of analytic continuation in Theorem 1.2 is explained by the crepant transformation between and , and the restriction of parameters is due to the partial compactification .
Connections to other results
The behavior of Gromov–Witten theory under small transitions among 3-folds was first studied in [[30]]. In [[29]], it was shown that the A and B model of considered together are determined by the A and B models of for conifold transitions between Calabi–Yau 3-folds. Beyond 3-folds, particular cases of small transitions arising as toric degenerations were considered in [[27]], where a similar statement to our main theorem was proven. Examples of non-small extremal transitions were studied by the first author. These include cases of triple point transitions in [[33]] and degree-4 Type II transitions in [[34]]. In the latter paper, a general conjecture was formulated using the language of quantum D-modules.
The types of toric hypersurface transitions we consider have appeared before (in the Calabi–Yau case), for instance in showing that for large classes of known examples, the "web" of Calabi–Yau 3-folds is connected [[4], [11]], an important step towards verifying Reid's fantasy.
The intermediate space has appeared previously in the context of Landau–Ginzburg models. After adding an appropriate superpotential , the corresponding LG model is known as an exoflop, and was studied in [[3]]. In the context of derived categories, the category of matrix factorizations on the LG model was used to prove an equivalence of categories between the derived categories arising from different Berglund–Hübsch–Krawitz mirrors [[21]].
Gromov–Witten theory preliminaries
In this section, we will briefly review the basics of Gromov–Witten theory and set notations for the rest of this paper. Our presentation here mainly follows [[1]].
Quantum D-module
Let be a smooth Deligne–Mumford stack over , whose coarse moduli space is projective. We denote by the inertia stack of , and by the rigidified inertia stack [[1], Section 3.4]. Recall that points of are given by pairs (x, g) where x is a point of and is an element of the isotropy group of x. We denote the twisted sectors of by for ranging over some index set R. Let
Graph
denote the natural involution on components of which maps (x, g) to .
Definition 2.1
Define the Chen–Ruan orbifold cohomology of to be
Graph
where cohomology is always taken with coefficients in unless otherwise specified. Note that when is a smooth variety, this is simply the usual cohomology ring. Define the Chen–Ruan orbifold pairing on to be
Graph
Let denote the cone of effective curve classes. For , , let denote the moduli space of representable degree-d stable maps from genus-g orbi-curves with n markings to the target space [[1], Section 4.3]. For , we have evaluation maps:
Graph
There is a canonical isomorphism , which allows us define for .
Definition 2.2
For an effective curve class and , define the Gromov–Witten invariant
Graph
where is the first Chern class of the th cotangent line bundle over , and is the virtual fundamental class defined as in [[6]] and [[1], Section 4.4].
Notation 2.3
Fix a basis of in such a way that there exists a partition such that is a basis for and forms a basis for Denote the dual basis by . Let . We denote for each . We will make use of the rings and .
We introduce the double bracket notation for generating series of genus-zero Gromov–Witten invariants.
Definition 2.4
For , define
Graph
Here the sum is over all terms in the stable range: .
Definition 2.5
For , the quantum product is defined by the equation
Graph
for all
By the divisor equation,
2.1.1
Graph
where
Graph
Consequently, the quantum product lies in .
The Chen–Ruan orbifold product is defined to be
Graph
Unless otherwise specified, when multiplying classes in we always use the Chen–Ruan product.
Definition 2.6
The Dubrovin connection for is given by the collection of operators defined by
Graph
for .
One can extend the definition of to the z-direction as well. Define the Euler vector field
Graph
where .
Define the grading operator by
Graph
for in . We define
Graph
This meremorphic connection is flat by standard arguments using the WDVV equation and homogeneity [[19]].
Definition 2.7
(Definition 3.1 of [[25]]) The fundamental solution operator
Graph
is defined as
Graph
where the denominator should be interpreted as a power series expansion in 1/z. By the divisor equation, this may be expressed alternatively as
2.1.2
Graph
Definition 2.8
Define a z-sesquilinear pairing on by
Graph
Proposition 2.9
(Proposition 2.4 of [[24]] and Remark 3.2 of [[25]]) The quantum connection is flat. The operator
Graph
is a fundamental solution for the quantum connection, that is:
2.1.3
Graph
for and . The pairing is flat with respect to , that is:
Graph
Definition 2.10
The quantum D-module is defined to be the data
Graph
Compact type quantum D-module
Many of the definitions above can be naturally extended to a non-proper target , provided the evaluation maps are proper. In this scenario, we will introduce a natural sub-D-module of the quantum D-module, which we refer to as the compact type quantum D-module. See [[38]] for details of this construction.
Let be a smooth Deligne–Mumford stack with quasi-projective coarse moduli space. In this section, we assume that all evaluation maps
Graph
are proper.
Definition 2.11
Let be the Chen–Ruan cohomology with compact support. There is natural map
Graph
which "forgets" that a cochain had compact support. Define the compact type (Chen–Ruan) cohomology of , denoted by , to be the image of inside .
For and , define
Graph
There is a non-denegerate pairing on . Given , let denote a lift of , that is an element satisfying . For , we define the pairing
Graph
One checks this definition is independent of the choice of lift [[38], Corollary 2.9].
For , the Gromov–Witten invariant
Graph
is well-defined when at least one of is in . Again using lifts, the invariant is also well-defined if for some .
Using compactly supported cohomology, most of the definitions in Sect. 2.1 go through directly for a non-proper target . For define to be the element satisfying
Graph
for all . Similarly, for and , define to be the element satisfying
Graph
for all .
The Dubrovin connection and fundamental solution are defined exactly as in Sect. 2.1. To avoid confusion we denote these operators by and respectively when they are acting on compactly supported cohomology.
By [[38], Proposition 4.6], the connection and the fundamental solution preserve the compact type cohomology. We denote the restrictions to by and respectively.
Definition 2.12
Define a z-sesquilinear pairing as follows. For and , let
Graph
For , define
Graph
Analogues of Proposition 2.9 hold. In particular we have the following relation between and .
Proposition 2.13
(Proposition 4.4 of [[38]]) The connections and are dual with respect to the pairing , that is:
2.2.1
Graph
Furthermore, for and ,
2.2.2
Graph
The map commutes with the quantum connection and the fundamental solution. It follows that is a flat connection with fundamental solution , and the pairing is flat with respect to . See [[38], Section 4.2] for details. Note that in loc. cit., the compact type cohomology is referred to as narrow cohomology.
Definition 2.14
The compact type quantum D-module for is defined to be the data
Graph
Ambient quantum D-module
In this section, we consider a restricted quantum D-module of a complete intersection.
Let be a smooth Deligne–Mumford stack with projective coarse moduli space.
Definition 2.15
A vector bundle is convex if, for any stable map from a genus-zero orbi-curve to ,
Graph
Let be a convex vector bundle with a transverse section . Let
Graph
be the inclusion of the zero locus .
Definition 2.16
Define the ambient cohomology of to be
Graph
We make the following assumption:
Assumption 2.17
The Chen–Ruan pairing on is non-degenerate.
The above assumption is equivalent to the statement that decomposes as . It holds, for instance, if E is a semi-ample line bundle on a toric variety [[32]], the setting of Sect. 4.
Given , the quantum product preserves , as do the quantum connection and the fundamental solution [[25], Corollary 2.5]. We can therefore make the following definition.
Definition 2.18
The ambient quantum D-module for , is defined to be
Graph
where is defined as in Notation 2.3, but using a basis of .
Integral structures
In [[24]], Iritani defines an integral lattice of flat sections in . We review the construction here.
Given a vector bundle, let denote the restriction of E to the twisted sector . Given a point , the group acts on . The vector bundle splits as
Graph
where is the Eigenbundle for which the stabilizer acts by .
Let denote the Grothendieck group of bounded complexes of vector bundles on . Recall [[40]] the definition of the orbifold Chern character:
Graph
Assumption 2.19
We assume that the stack has the resolution property, and the map is an isomorphism after tensoring with .
Assumption 2.19 holds, for instance, if is a smooth toric stack.
Definition 2.20
Define the Gamma class to be the multiplicative class
Graph
where are the Chern roots of . Denote by the class .
Let
Graph
denote the operator which multiplies a homogeneous class by its (unshifted) degree. In [[24]], Iritani defines a map .
Definition 2.21
For , define to be
Graph
Proposition 2.22
(Section 3.2 of [[25]]) The map identifies the pairing in K-theory with up to a sign:
Graph
Definition 2.23
Define the integral structure of to be
Graph
The integral structure of forms a lattice in the space of flat sections of .
Next we consider the case were is not proper. We can apply Definition 2.21 without change to obtain an integral lattice in . With slight modifications we can also define a lattice in and .
For a closed and proper substack of , let denote the Grothendieck group of bounded complexes of vector bundles which are exact off of . Define
Graph
to be the direct limit over all proper substacks . There exists a compactly supported orbifold Chern character [[38], Definition 8.4]. Let
Graph
denote the natural map. By [[38], Proposition 8.5],
2.4.1
Graph
Assumption 2.24
We assume that the stack has the resolution property and the map is an isomorphism after tensoring with .
Assumption 2.24 holds if is the total space of a vector bundle bundle on a proper toric stack.
Following Definition 2.21, we make the following definition.
Definition 2.25
For , define to be
Graph
Define the integral structure of to be
Graph
We have the following analogue of Proposition 2.22:
Proposition 2.26
For and ,
Graph
By (2.4.1), is supported in the compact type cohomology for all . We can therefore define the following.
Definition 2.27
For , define to be
Graph
Define the integral structure of to be
Graph
Let be the inclusion of a closed substack in a proper Deligne–Mumford stack as in Sect. 2.3. Again with Assumption 2.19, we can define an integral structure on the ambient quantum D-module to be
Graph
Quantum Serre duality
Consider the setting of Sect. 2.3, where is a convex vector bundle, and is the zero locus of a transverse section of E.
Consider the total space of the dual bundle:
Graph
Quantum Serre duality is the name given to a close relationship between the Gromov–Witten theory of and the Gromov–Witten theory of . It was first described mathematically by Givental [[23]] and later generalized by Coates–Givental in [[14]]. The correspondence was formulated as a relationship between quantum D-modules in [[26]]. The formulation below using compact type quantum D-modules was given by the second author in [[38]].
Definition 2.28
Define the map as follows. For , define
Graph
where is a lift of . This is independent of the choice of lift by [[38], Lemma 6.10] Define .
Definition 2.29
Define the map by
Graph
where here and are dual bases of . Define
Graph
Note that lies in .
Theorem 2.30
[[38], Theorem 6.14] The map identifies the quantum D-module with the pullback . Furthermore it is compatible with the integral structures and the functor , i.e., the following diagram commutes:
Graph
where denotes the subgroup of consisting of complexes exact off of and denotes the image of .
Remark 2.31
We note that when is a toric stack, the lattice generated by the left vertical map spans the space of solutions in .
Toric preliminaries
We consider toric varieties constructed as stack quotients via GIT. As this section is primarily to set notation and recall previous results, the exposition is condensed. We refer the reader to [[15]] for further details.
GIT description
The initial data consists of
- A torus ;
- the lattice of co-characters of K;
- a set of characters ;
- a choice of a stability condition .
Given the above, the map
Graph
defines an action of K on . Given , we denote by the complement of I.
Definition 3.1
Define to be the subset
Graph
Define the set of anticones (with respect to ) to be
Graph
For each I, consider the open set
Graph
and let denote the union We define the toric stack to be the GIT quotient
Graph
where brackets denote that we are taking the stack quotient. We will denote the underlying coarse toric variety as
Graph
Define the set to be the collection of such that for all .
Assumption 3.2
As in [[15]], we assume in what follows that
-
;
- for each , the dimension of is maximal (equal to r).
The space of stability conditions has a wall and chamber structure. Let
Graph
Then for any , . We call the extended ample cone.
Fan description
We obtain the more familiar fan description of from the GIT data as follows. Consider the short exact sequence
3.2.1
Graph
where is defined to be the quotient map of by . For , let and let denote the image of in . For , let denote the cone generated by .
Definition 3.3
Define the fan in to be the collection of all cones
Graph
The collection is called an R-extended stacky fan.
The elements for are called extended vectors or ghost rays depending on the source. It can be checked that for , lies in the support of .
Cohomology
Each of the characters for defines a divisor . The class may be defined as the first Chern class of the line bundle given by
3.3.1
Graph
where the action of K on the last factor of is given by the character . Alternatively is Poincaré dual to
Graph
The cohomology ring of is then given by the Stanley–Reisner presentation [[7], Lemma 5.1]
3.3.2
Graph
where
3.3.3
Graph
Note that for , .
It will also be convenient to describe the compact type cohomology of , which follows from the work of [[8]].
Definition 3.4
Let be a fan whose support in is a rational polyhedral cone. We say a cone is an interior cone if the interior of is contained in , the interior of the support of .
Lemma 3.5
The compact type cohomology is generated as an -module by
Graph
Proof
By [[8], Proposition 2.4] and the discussion following [[8], Remark 2.5], the compactly supported cohomology of is given by
Graph
modulo a natural set of relations analogous to those in (3.3.3). Here is just a formal symbol. Furthermore, the map is given by
Graph
The ample cone
Equation (3.3.2) implies, in particular, that . Using (3.2.1), one can find a canonical isomorphism For each , let denote the smallest cone of which contains . Then for . Define by
3.4.1
Graph
Then
3.4.2
Graph
Dualizing we see that
3.4.3
Graph
hence .
Denote by
Graph
the quotient by . Note that this map satisfies .
Under the splitting (3.4.3), the cone
3.4.4
Graph
where is the cone of ample divisors. The Mori cone is then given by
Graph
Chen–Ruan cohomology
For , define
Graph
Define the lattice to be the set
Graph
The set indexes components of the inertia stack . Namely, for , the corresponding component of is given by
Graph
where
3.5.1
Graph
The component may be described concretely as
Graph
In particular, it is also a toric variety constructed as in Sect. 3.1, using K and as before, but restricting to those characters One may construct the analogous exact sequence to (3.2.1):
Graph
Let denote the corresponding fan in .
The Chen–Ruan cohomology (Definition 2.1 below) of is then given as a vector space by
Graph
where may be written explicitly by Sect. 3.3.
Proper evaluation maps
Because we do not assume that our toric stack is proper, we include a preliminary lemma to guarantee that the quantum D-module (as well as the compactly supported and compact type quantum D-module) is well-defined.
Lemma 3.6
Let be a toric stack defined as a GIT quotient as in Sect. 3.1. For fixed g, n, d, and for , the evaluation map is proper.
Proof
The toric stack is defined as a stack GIT quotient , which maps to the coarse underlying toric variety . The latter is projective over the affine space . Given a semi-stable orbifold curve , a morphism must be constant. Any stable map lies in a fiber of . Therefore the space coincides with the moduli space of stable maps to relative to (in the sense of [[2]]). This moduli space is denoted as . It is shown [[2]] that is proper over . The evaluation map fits into the diagram
Graph
The map is proper, therefore is as well, by [[39], Lemma01W6].
Hypersurfaces
Let be a smooth and proper toric Deligne–Mumford stack. Choose a divisor , with . Let
Graph
be the support function for D, a piecewise-linear function which is linear on each cone of and characterized by the condition that for . (See [[20], Chapters 4 and 6] for a discussion of the support function.)
We make the following assumptions on D.
Condition 3.7
The function satisfies the following:
- For each , there exists an element such that
-
Graph
- for ;
- The graph of is convex and for .
See [[20], Chapters 14 and 15] for details on these conditions. Our purpose in making these assumptions is explained in the following lemma.
Lemma 3.8
Under the assumptions above, D is basepoint free and is a convex line bundle on .
Proof
By Part (1) of Condition 3.7, D is pulled back from a Cartier divisor on the coarse space . See [[20], Theorem 15.1.1 and Proposition 15.1.3] for the proof when is a fan rather than a stacky fan. The general case follows from a similar argument. By [[20], Proposition 15.1.3], Part (2) of Condition 3.7 implies that is nef. By [[20], Theorem 15.1.1], is basepoint free and thus so is D.
Let be a stable map from a genus-zero n-marked orbi-curve . Let be the map to the underlying coarse curve. By [[2], Theorem 1.4.1], the map factors through a map . Therefore . We observe that
Graph
where the third equality is the projection formula.
Since is basepoint free, we have a map such that is the pullback of . We conclude that
Graph
As is convex, the above is zero.
Corollary 3.9
For a general section , The zero locus Z(s) is a smooth orbifold.
Proof
By Lemma 3.8D is basepoint free. The result then follows from a general version of the Bertini theorem [[36], 2.8] and [[42], Theorem 2.1].
As another consequence of Lemma 3.8, the results of Sect. 2.5 hold for these hypersurfaces.
For future reference we record the following facts about . We may express a section as
Graph
where each is a torus invariant section of D and . Define
3.7.1
Graph
Then the torus invariant sections are identified with lattice points in . Under this correspondence, the element is identified with the rational function
Graph
Note that given , viewed as a (rational) section of , the order of vanishing of along is then
3.7.2
Graph
Geometric setup for extremal transitions
Definition 4.1
[[35]] Given two smooth projective varieties and , we say they are related by an extremal transition if there exists a singular variety such that is a smoothing of and is a crepant resolution of . More precisely, we require that there exists:
- a projective map where is a smooth variety with distinguished points , such that the fiber over 0 is isomorphic to , the fiber over x is isomorphic to , and the fiber over any point in is a smooth projective variety diffeomorphic to ;
- a projective birational morphism such that ).
The following commutative diagram shows the relationship between the various spaces:
Graph
We may extend this definition to the case where and are smooth Deligne–Mumford stacks with projective coarse moduli space. In this case we require that is an orbifold crepant resolution of , i.e. is smooth as a stack.
Remark 4.2
The relationship between and in Definition 4.1 is more accurately referred to as a geometric transition or simply a transition. A transition is then extremal if the map is an extremal contraction in the sense of Mori theory. Following e.g. [[27]], in this paper we refer to all transitions as extremal transitions to avoid confusion with other uses of the term transition in algebraic geometry.
In this section, we give a general construction which yields extremal transitions between toric hypersurfaces. This is related to a well-known construction [[4], [11]] of transitions between toric Calabi–Yau hypersurfaces. However here we focus on toric blowups, while also generalizing beyond the Calabi–Yau setting.
Toric blow-ups
Let K be a torus and let denote a set of characters. Choose a stability condition to obtain the toric variety as in Sect. 3. Consider a codimension-k toric subvariety defined by the vanishing of a subset of the homogeneous coordinates . By possibly reordering the divisors , we can assume without loss of generality that is defined by the vanishing of the first k coordinates. To avoid a trivial situation we assume is a cone of - in particular .
We may realize the map
Graph
as an instance of toric wall crossing. Let denote the torus and let .
For we define as follows:
4.1.1
Graph
and define
Graph
Let be the map from (3.2.1). We extend this to a map by defining . The following sequence is exact.
Graph
Choose and define
Graph
We will consider the GIT quotients , where the action of K on is given by . Let denote the corresponding sets of anticones. Let denote the respective fans.
Proposition 4.3
The toric stack is equal to .
Proof
This follows immediately from the observation that J is an anticone of if and only if is an anticone of . Therefore .
On the other hand, define
Graph
Proposition 4.4
The toric stack is equal to . More precisely, the fan for contains exactly the following maximal cones:
- For every maximal cone of such that I does not contain , the cone is in .
- For every maximal cone of such that I contains and for each , the cone is in .
Proof
Consider a maximal cone of . Then is a minimal anti-cone of . Assume first that is not contained in I. Then . By assumption, there are constants such that
Graph
Furthermore,
Graph
By shrinking if necessary, we may assume that
Graph
Then
Graph
Therefore is a minimal anticone of .
Next, assume that is contained in I. Then the anticone is disjoint from . As before, there are constants such that
Graph
Therefore,
Graph
for . By Lemma 4.5 below with and , the stability conditions and lie in the same maximal chamber of . Therefore is a minimal anticone of .
We observe that the union of the support of each of the cones described above is equal to . This guarantees that there are no maximal cones of other than those described above.
Lemma 4.5
Let and be as above. Consider the wall and chamber structure on determined by .
Let and in be vectors with last coordinates either both positive or both negative. Consider the stability conditions for . Then for positive and sufficiently small, and lie in the same maximal chamber.
Proof
Let denote the closure of the (maximal) chamber containing . We first claim that with respect to the wall and chamber structure on determined by , the stability condition is not contained in the closure of any wall other than .
To see this, assume the contrary, that is a wall such that . If then for a wall with respect the wall and chamber structure determined by . By Assumption 3.2, must be dimension r and therefore must be . So we may assume that is not a subset of .
By assumption has dimension less than . Therefore there must exist a set with and non-negative constants such that
Graph
Because is a wall not contained in , at least one of has a nonzero last coordinate which then forces to be in J. Without loss of generality we may assume . We conclude that
Graph
Because , this contradicts Assumption 3.2. This proves the claim
We now show that for positive and sufficiently small, lies in a maximal chamber. If we assume the contrary, then the line segment
Graph
must be contained in a wall . By assumption on , is not contained in . But this contradicts the fact that is not contained in the closure of any wall other than .
To show that and lie in the same maximal chamber, we must show there is not a wall through which separates and for positive and sufficiently small. By assumption, the sign of the last coordinate of and is the same, so the wall does not separate them. Then again by the fact that is not contained in the closure of any wall other than , we reach the desired conclusion.
By the above two propositions we see that a toric blow-up is an example of a (discrepant) toric wall crossing. In this case the wall W is generated by the vector .
Extremal transition
Let the setup be as in the previous section. Choose a divisor on satisfying Condition 3.7. Let be a general section. Define the hypersurface
Graph
This is a smooth Deligne–Mumford stack by Corollary 3.9. Recall that we may express the section s as
Graph
where the sum ranges over all of the lattice points in the polytope from (3.7.1) and the are constants.
We may degenerate the section to some by setting some of the to zero. Recall from the previous section that the toric subvariety is defined by the vanishing of the first k homogeneous coordinates. Define
Graph
where is given by (3.7.2). Consider the section . Then
Graph
is a degeneration of . It will contain (and will usually develop singularities on) . If the proper transform of under the blowup map is smooth, then it gives a resolution of . As we will see below, the degeneration of the section s to is chosen precisely to guarantee that this resolution, when it exists, will be crepant.
Let denote the blowup map from Sect. 4.1. The pullback of D to is given by
4.2.1
Graph
Define the divisor to be
4.2.2
Graph
Note that by (3.7.1), is a subset of . Precisely, the element lies in if and only if
Graph
The left hand side of the above equation is equal to (viewing as a section of D), and therefore defines a general section of . To avoid confusion with , we denote this section by , although both and may be expressed as .
We will always assume Condition 3.7 holds for as well as for D. Under this assumption, if we have chosen sufficiently general, then
Graph
will be a smooth variety (or orbifold). As mentioned previously, may also be described as the proper transform of under the map .
Proposition 4.6
The hypersurfaces and are related by an extremal transition through in the sense of Definition 4.1.
Proof
We must check that the resolution is crepant.
We first claim that the singular locus of is codimension at least 2. Note that and are isomorphic outside of , so the claim is immediate if the codimension of in is at least three. We are left with the case that is codimension 2 in (and therefore codimension 1 in ). In this case, the section is defined to be a general section which vanishes on . We may therefore express as
Graph
where f and g are general sections of and respectively. A local coordinate computation shows that the singular locus is given by
Graph
The claim will follow if we can show that at least one of f or g is non-vanishing on . By Condition 3.7, is basepoint free. There must therefore exist at least one torus-invariant section which is non-vanishing on . Representing as a lattice point in , by (4.2.2) and (3.7.2) we have that
Graph
On the other hand, since also lies in ,
Graph
for . Since , it follows that for i equal one of 1 or 2, and for the other Without loss of generality, we assume that . Then corresponds to a torus-invariant section of D which vanishes to order 1 at and order 0 at . In homogeneous coordinates, it may therefore be written as where is nonvanishing on . We conclude that a general section f of is nonvanishing on . This proves the claim that the singular locus of is codimension at least 2.
We may therefore calculate the canonical class on the complement of the singular locus. Let denote the singular locus of . Let denote the complement . Then is a regular section of , whose zero locus is smooth. Applying the adjunction formula to and then extending to all of , we conclude that
Graph
exactly as if were smooth.
Consider the commutative diagram
Graph
We must show that . Again applying the adjunction formula,
Graph
It therefore suffices to show that
4.2.3
Graph
The pullback is equal to
Graph
On the other hand, by (4.2.1) and (4.2.2),
Graph
Equation (4.2.3) follows.
Remark 4.7
The construction of extremal transitions given above can be generalized to the case where the map between the ambient spaces is a weighted blow-up. In this modification, we replace from (4.1.1) with for , where are positive integers, and we replace (4.2.2) with
Graph
The results of the paper hold in this context as well, with the same proofs, although the notation is slightly more cumbersome. We leave the details to the reader.
Examples
This setup includes many well-known examples.
Example 4.8
More generally we can let , choose any m and k with , and let . For any choice of , the divisor
Graph
satisfies Condition 3.7 if is positive. Here H is a hyperplane. The section s defines a degree c hypersurface of dimension in , and is a singular hypersurface which vanishes at . In this case , and
Graph
where H is the pullback of a generic hyperplane in and E is the exceptional divisor. Since is a smooth variety, Part (1) of Condition 3.7 is automatic for . By [[20], Theorem 15.1.1], the divisor will satisfy Part (2) of Condition 3.7 provided is nef. The nef cone for is generated by the rays and . Condition 3.7 is therefore satisfied if lies in the cone generated by (1, 1) and (1, 0), or equivalently, if
Graph
In this way we generate numerous examples of transitions in any dimension for hypersurfaces of any degree. The case of , and gives the conifold transition described in Example 1.1. The case of and gives the cubic transition studied by the first author in [[33]].
Example 4.9
There is no particular reason to restrict to a projective space. Generalizing the above example, we consider a product of projective spaces
Graph
by choosing each from , where is the vector with 1 in the jth coordinate and zeroes in all other coordinates.
By reordering , the set may be given by
Graph
where are homogeneous coordinates on the jth factor of , . After choosing appropriate , the divisor D is conjugate to where is the pullback of the hyperplane class from . Then D will satisfy Condition 3.7 if for all j, and will satisfy Condition 3.7 if, for all j with ,
Graph
Example 4.10
For an orbifold example, let . Choose positive integers , let for , and let . Then is the weighted projective space . The fan for lies in and contains the rays for and . Let and . If d is divisible by each of then D will satisfy Condition 3.7. The divisor will satisfy Condition 3.7 if
Graph
where .
The total spaces
We want to compare the Gromov–Witten theory of and , as defined in Sect. 4. By quantum Serre duality (Theorem 2.30), it suffices to compare the Gromov–Witten theory of the total spaces of the line bundles corresponding to and . Define
Graph
We can represent both and as a toric GIT quotient. We will focus on .
Let denote the torus and recall the definition of from Sect. 4.1. Define
Graph
We will consider the GIT quotients , where the action of on is given by . Let denote the corresponding sets of anticones.
Define by
Graph
where is the map from (3.2.1) for . A simple check shows that
5.0.1
Graph
is exact. Define . We denote the fans corresponding to the GIT quotients by .
Proposition 5.1
The total space may be expressed as the toric GIT quotient . Furthermore if and only if .
Proof
Given a cone of , let . It follows immediately from considering anticones that each cone is a cone of . Note that for ,
Graph
As a consequence, the union
Graph
is the set of all points such that the last coordinate of is greater than or equal to . By the convexity assumption on the support function (Part (1) of Condition 3.7), this set is equal to
Graph
and therefore must be equal to . Therefore must contain every cone of .
We have shown that is a cone of if and only if is a cone of . This implies that and therefore that the map is a vector bundle. To see that it is the total space of , it suffices to recall that
Graph
The second part of the proposition follows as well from the description of the cones of .
Corollary 5.2
The only interior ray of is .
Proof
This follows from the description of in the proof of Proposition 5.1.
Next we investigate the GIT quotient with respect to the stability condition . As the next proposition shows, it is not equal to . Let
Graph
Proposition 5.3
The toric stack is a partial compactification of . The fan for contains precisely the following two sets of maximal cones:
- For every maximal cone of , the cone is in ;
- For every maximal cone of such that I contains , the cone is in .
Furthermore if and only if .
Remark 5.4
The fan formed by the type (1) cones gives the toric stack . The type (2) cones serve to partially compactify . See Corollary 5.6 below.
Proof
By an identical argument as in Proposition 5.1, we see that the type (1) cones lie in .
We now show that the type (2) cones lie in . Convexity of D implies that D lies in the closure of the extended ample cone [[20], Theorem 15.1.1 and Proposition 15.1.3]. Therefore, for all , there exist a set of non-negative constants such that
Graph
Since for , we have
5.0.2
Graph
whenever J is disjoint from . Let be a cone of such that I contains . Let J denote the complement in . We claim that is an anticone of . Choose constants such that
Graph
Using the above two equations we observe that
Graph
Therefore is an anticone of .
The union of the support of all type (1) and (2) cones is easily seen to be equal to the support of , which implies there are no other maximal cones.
The final statement of the proposition follows immediately from the description of maximal cones.
Figure 1 in the introduction provides a picture of the fans in the case of , and a point. In this picture the bottom vertex is the origin and each line out of the origin is a primitive ray vector.
In the table below we list the relevant toric varieties constructed in this section, together with the toric data defining them. Recall that , , .
Graph
Recall that the connected components of are indexed by . These components can be separated into two groups based on whether or not they intersect .
Definition 5.5
Define the following complementary subsets of :
Graph
The following observations are more or less immediate from the above description of . We record them here for future use.
Corollary 5.6
The open locus is equal to . The locus is equal to .
The connected components of are of the following two types:
- The elements index those components of which are partial compactifications of a corresponding component of . In particular the embedding given by induces an isomorphism .
- For , the component of is supported on the locus .
Proof
The first statements follow from the fact that
Graph
where the action of K on the last factor of is given by the restriction of to . The GIT quotient on the right hand side is easily seen to be , by an argument similar to Proposition 5.1.
It follows from the definitions that
Graph
where the isomorphism is induced by the inclusion . The set indexes exactly those twisted sectors which have a nontrivial intersection with .
The elements of correspond to components of supported on .
Comparison of compact type cohomology
In this section, we identify the compact type cohomology of with a subquotient of the compact type cohomology of . We also compare the ample and Mori cones.
Recalling (3.3.2), let and denote the cohomology classes defined by the vanishing of the homogeneous coordinates and respectively, viewed as classes in the untwisted sector .
Lemma 5.7
The compact type cohomology contains the subspace .
Proof
We work with each component of individually. For , we claim that . The toric stack may be represented as the GIT quotient
Graph
where is as in (3.5.1). By assumption on , , so the -th (= st) coordinate of a point in must be zero. It follows that a semistable point of must be nonzero in the nd coordinate or equivalently, that the natural section of is nowhere vanishing. This implies that under the map , the class pulls back to zero. The claim follows by [[38], Remark 2.14 ].
Next consider the untwisted sector . Because and have the same support, it follows from Corollary 5.2 that is an interior ray of . By Lemma 3.5 this implies .
For , the map is a closed embedding. By [[38], Proposition 2.5], maps to . From this we have
Graph
Proposition 5.8
The pullback restricts to a surjective map
Proof
For , the summand is zero by the previous proof. We must check that for , the pullback maps surjectively onto .
For a given twisted sector , consider the Stanley–Reisner presentations of and as described in (3.3.2). The pullback is given simply by setting to zero in the Stanley–Reisner presentation of . It follows that the restriction
Graph
is surjective.
Lemma 5.9
We have the following comparisons between (co-)homology of and .
- The map given by defines a canonical isomorphism .
- The map defines a canonical isomorphism .
- Under the isomorphism above, the extended ample cones are related as:
-
Graph
- The ample cones and Mori cones satisfy the same relationship:
-
Graph
• and
Graph
Proof
By Proposition 5.3, if and only if . By (3.4.2),
Graph
where is as in (3.4.1). Proposition 5.3 also implies that the fan for consists of the Type 1 cones of . Note also that for ,
Graph
The second part of Condition 3.7 for D then implies that for , is contained in one of the cones of . It follows by (3.4.1) that , where is the corresponding element for . Under the splittings
Graph
the summand on the left is identified with on the right.
From the equation above we see that . This implies the second claim by (3.4.3).
To compare the extended ample cones, we use Proposition 5.3, which implies that the anticones are of the form
- if is an anticone of , then is an anticone of ;
- if is an anticone of which is disjoint from , then is an anticone of .
First observe that
Graph
To see that this is equal to it suffices to show that the further intersection with for the anticones of type (2) does not change the set. This follows from the following claim: if is an anticone of which is disjoint from , then is contained in , where and are the corresponding anticones of types (1) and (2) associated to I.
To prove the claim we show that This too follows from the convexity assumption of Condition 3.7 for D. Again using (5.0.2), for I disjoint from ,
Graph
The claim follows and we conclude that
Graph
The comparison of ample cones follows from this and (3.4.4). The comparison of Mori cones is then immediate.
Crepant transformation conjecture
In this section, we recall the crepant transformation conjecture proven in [[15]], and use it to prove analogous statements in compactly supported and compact type cohomology. The results of this section will be applied specifically to and .
Wall crossing
We begin by recalling the general wall crossing setup of [[15]]. To be notationally consistent with the specific setup we will use in Sect. 5, we will consider toric varieties and arising as GIT stack quotients by a torus of dimension . The corresponding rank lattice will be denoted by . Choose stability conditions lying in cones and of maximal dimension which are separated by a codimension-one wall. Denote by and the corresponding toric stacks. Define and denote the extended stacky fans.
Define W denote the hyperplane separating and and define
Graph
Let be a primitive generator of . We assume without loss of generality that . Under these assumptions, and are birational, via a common toric blow-up defined in [[15], Section 6.3.1]
Graph
If we assume further that then and are K-equivalent. This is the setting of [[15]].
Consider the fan consisting of the cones , , and their faces. Let denote the corresponding toric variety. This is an open subset of the secondary toric variety associated to and . Let denote the torus fixed point of associated to the cone , and let denote the torus invariant curve between and . The correspondence given by the crepant transformation conjecture of [[15]] takes place on a formal neighborhood of in . Following [[15]], it is more convenient to work with a smooth cover of .
Choose integral bases and of such that
-
for ;
-
for .
For , let denote the corresponding element of . We have inclusions
Graph
The coordinates are related by the change of variables
6.1.2
Graph
where are determined by the change of basis from to . Note that by assumption . Let denote the toric variety corresponding to the fan whose cones consist of , , and their faces. There is a birational map
Graph
induced by the map of cones.
We will consider a modification of , whose sheaf of functions is analytic in the last coordinate ( ) and formal in the other coordinates. Let denote the analytic complex plane, with coordinates and respectively. These glue to form an analytic via the change of variables Define the sheaves
Graph
which also glue over via (6.1.2). Let denote the corresponding ringed space, and let and be the open sets given by and respectively.
Remark 6.1
The setup above differs slightly from that given in [[15]], where the authors define overlattices and of , and choose to be an integral basis of . These coordinate are used to define a variety [[15], Equation 5.10], which is a smooth cover of . One can check that the bases may be chosen so that the map factors through . There are induced maps between the completions of the analytic spaces.
The overlattices and of correspond to finite covers of and . The deck transformations of these covers corresponds under the mirror theorem to Galois symmetries of the quantum D-module [[24], Sections 2.2 and 2.3].
By [[15], Remarks 5.10 and 6.6], the mirror theorem and crepant transformation conjecture of [[15]] are both compatible with this Galois action, consequently the results in this section are equally valid on and .
Notation 6.2
By Lemma 5.9, for the particular case of and , we may set to be . We fix this choice in later sections.
Mirror theorem and CTC
We recall the main result of [[15]]. In contrast to that paper we work non-equivariantly.
Let
Graph
Let denote .
Definition 6.3
Define the Fourier–Mukai transform
Graph
wherer is as in (6.1.1). Because and are proper, we can also define
Graph
Theorem 6.4
[[15], Theorem 5.14] There exists an open subset such that is a discrete set and . On there exists
- A trivial -bundle
-
Graph
- A flat connection on with logarithmic singularities at for ;
- A mirror map , of the form
- 6.2.1
Graph
• where
Graph
such that is the pullback of the quantum connection of by .
The above is a D-module formulation of the mirror theorem from [[13]], itself a generalization of the work of [[23], [31]]. The connection is generated by the I-function of [[13]] after specializing the Novikov variables to one. An analogous result holds for .
Next we state the crepant transformation conjecture proven in [[15]], which relates the connections and . The comparison may be made on and depends on a path of analytic continuation in from a neighborhood of to a neighborhood of . We choose a path from to such that the real part of is always increasing, the imaginary part of is 0 when or , and contains the point
Graph
Theorem 6.5
(Crepant transformation conjecture [[15]]) There exists a gauge transformation
Graph
such that
- the connections and are gauge equivalent via :
-
Graph
- analytic continuation of flat sections is induced by a Fourier–Mukai transformation:
•
Graph
Remark 6.6
In fact the theorem as stated in [[15]] takes place on an open subset of the universal cover of (as that is where the I-functions are single-valued). However the results are equally valid on , by [[15], Remark 6.6] (see also Remark 6.1).
Compactly supported CTC
We recall from Sect. 2.2 the compactly supported quantum connection and the compactly supported fundamental solution .
Definition 6.7
With and the mirror map given as in Theorem 8.2, define the trivial -bundle over
Graph
Define the connection
Graph
By Proposition 2.13, and are dual with respect to .
Definition 6.8
Define the -valued homomorphism
Graph
via the equation
6.3.1
Graph
for all and .
Theorem 6.9
(Compactly supported CTC) The connections and are gauge equivalent via :
Graph
Furthermore, analytic continuation of flat sections is induced by a Fourier–Mukai transformation:
Graph
for all .
Proof
For all and ,
Graph
The first and third equalities follow from the definition of . The second and fourth follow from Proposition 2.13. The fifth equality follows from Theorem 6.5. Cancelling from the top and bottom expression finishes the proof of the first statement.
To compare flat sections with the Fourier–Mukai transformation, consider and .
Graph
As span the flat sections of and the pairing is nondegenerate, we conclude that
Graph
compact type CTC
In this section, we use the compactly supported crepant transformation conjecture to prove a corresponding statement in compact type cohomology.
Definition 6.10
With and the mirror map given as in Theorem 8.2, define the trivial -bundle over
Graph
Define the connection
Graph
The following proposition was suggested to the second author by Iritani in private communication.
Proposition 6.11
Provided Assumption 2.24 holds for , the following diagram commutes:
Graph
where the vertical arrows are and .
In particular, the map sends to .
Proof
Commutativity of the top square is immediate from the definition of . Commutativity of the front and back squares follows from (2.4.1) and that the fundamental solutions satisfy
Graph
Commutativity of the left and right square are Theorems 6.9 and 6.5 respectively. Commutativity of the bottom square then holds if the maps are surjective. This holds because Assumption 2.24 holds for the total space of a vector bundle on a smooth toric stack.
Definition 6.12
Define to be the restriction of to .
Theorem 6.13
(Compact type CTC) The transformation satisfies the following:
- the connections and are gauge equivalent via :
-
Graph
- analytic continuation of flat sections is induced by a Fourier–Mukai transformation:
•
Graph
- The pairing is preserved:
-
Graph
Proof
The first point and second point are automatic from the fact that , , and are restrictions of , and respectively. The third point is due to the fact that for and ,
Graph
D-module of the partial compactification
In this section, we identify a subquotient of a certain restriction of with . The results rely on the specific geometry of .
Restricting degree
Notation 7.1
Lemma 5.9 allows us to express any , as where and is a multiple of . Following Notation 2.3, by decomposing with respect to twisted sectors and degree, there is a canonical way to write a point as where . By Part 2 of Lemma 5.9, the point may be canonically written as where and s is a scalar. We note that is identified with under (3.4.3). For , let . Let denote . Then for ,
Graph
We will focus on the subspace , which may be alternatively described as
Graph
where is the closed embedding obtained by composing with the inclusion .
Lemma 7.2
After restricting to ( ), the quantum D-module preserves the subspace . Furthermore, the map is compatible with quantum D-modules after this restriction:
7.1.1
Graph
whenever .
Proof
Let be a stable map of degree in . By [[18]] and [[12], Section 2.2], the map f is determined by line bundles (corresponding to the factors of ), and for each character , a section of the corresponding line bundle (here from (3.3.1)). By the condition on d, the line bundle is trivial and the section is constant. Therefore f maps entirely to the locus or This implies in particular that we have the following cartesian diagram:
Graph
(By a slight abuse of notation we will denote both the maps and by i.)
We claim further that f maps entirely to or . From above, we can split the argument into two cases depending on whether f maps to or to . If f maps entirely to then f maps to . If f maps entirely to then f may be described by a map to together with a section of . By assumption D is nef and therefore has non-negative degree. Thus the degree of is non-positive which forces the section to be constant. This shows the claim. We therefore have a second cartesian diagram
Graph
Given , we note that and therefore
Graph
by (7.1.3). From [[22], Proposition 1.7] we have
Graph
therefore
Graph
for . Working term by term on the expression (2.1.2), we conclude that lies in . This shows that preserves the subspace .
Next we prove (7.1.1). It suffices to show
7.1.4
Graph
We make use of (7.1.2). The perfect obstruction theory on is the restriction of the perfect obstruction theory on , and so they are compatible in the sense of [[6], Definition 5.8]. By functoriality [[6], Proposition 5.10],
Graph
From this we observe that
Graph
This gives a term-by-term identification of the left and right sides of (7.1.4).
We consider the monodromy of around .
Lemma 7.3
The monodromy invariant part of the compact type quantum D-module of around is spanned by the flat sections for ranging over
Graph
In particular, it contains the flat sections
Graph
Proof
As described in [[24], Sections 2.2 and 2.3], the monodromy transformation of flat sections is given by the Galois action defined in [[24], Proposition 2.3]. The monodromy around corresponds to the action of : for ,
7.1.5
Graph
where is as in (3.3.1) and is the weight of the generator of the generic isotropy group of on the line bundle .
We prove the second claim first. For , and by Lemma 5.7, is supported on the twisted sectors for which . In this case the section is invariant under the monodromy transformation by (7.1.5).
For the first statement, choose a basis of such that each is homogeneous and supported on a single twisted sector . Then a section is monodromy invariant if and only if
Graph
or equivalently, if
Graph
As multiplication by preserves for each twisted sector, we can work with each summand individually. Let be the subset such that is a basis for . By considering the top degree nonzero homogeneous term of , we see that it must lie in , and that must be zero if is nonzero. Therefore for , for . For , we use descending induction on the top degree of the left hand side to see that
Graph
The previous two lemmas together yield the following.
Corollary 7.4
The monodromy invariant part of , when restricted to , contains a sub-D-module which maps surjectively via onto .
This map of D-modules is compatible with the pairings. It relates the integral structures as follows: for ,
Graph
Proof
By Lemma 7.3, the monodromy-invariant flat sections of contain the span of for . Then by Lemma 7.2,
7.1.6
Graph
This shows that the pullback maps to .
To see that the pairings match, we apply the projection formula [[9], Proposition 6.15] twice:
Graph
We note that the pairing may become degenerate when restricted to .
For , is entirely supported in . The last statement follows from (7.1.6) and the definition of .
Interaction with the mirror theorem
Recall the connection on . By Definition 6.10, this is the pullback of via a mirror map
Graph
We consider the monodromy invariant submodule around .
Proposition 7.5
The monodromy invariant part of , when restricted to , contains a submodule which maps surjectively via onto .
This identification is compatible with the pairings and integral structures as in Corollary 7.4.
Proof
Consider the pullback of the fundamental solution matrix via the mirror map . By Theorem 8.2, the mirror map is given by
Graph
From Notation 6.2 we observe that . Thus we are in the setting of Corollary 7.4, with playing the role of . The result follows.
Main theorem
In this section, we prove the main theorem, relating the compact type quantum D-modules of and . We recall the relations between the various stacks described in Sects. 4 and 5:
Graph
Consider the following diagram, which combines the maps appearing in quantum Serre duality, the compact type crepant transformation conjecture, and the previous section:
Graph
where here and are defined to be the compositions.
Notation 8.1
We recall the following maps:
- Let be the gauge equivalence as given in Definition 6.12 with respect to the wall crossing between and .
- Let denote the quantum Serre duality map for (Definition 2.28).
- Let denote the quantum Serre duality map for .
Our main theorem states, roughly, that the quantum D-module , together with its associated integral structure, can be analytically continued to a neighborhood of ( ), and that the restriction of the monodromy invariant part to has a sub-quotient which is gauge-equivalent to . As a result, the quantum D-module for is determined by the quantum D-module for after pullback by a mirror map. The theorem follows almost immediately from the results of previous sections, namely:
- Theorem 6.13, the (compact type) crepant transformation conjecture, which relates to ;
- Proposition 7.5, which relates to ; and
- Theorem 2.30, quantum Serre duality, which relates to and to .
To formulate a precise statement, we make use of the ambient quantum D-module of . Define the quantum D-module to be the pullback , together with the pulled-back integral structure and pairing.
Theorem 8.2
The quantum D-module on satisfies the following:
- In a neighborhood of , is gauge-equivalent to via the transformation ;
- The monodromy invariant part of around , when restricted to , contains a sub-module which maps surjectively to via the gauge transformation ;
- Both transformations preserves the pairing: for ,
-
Graph
- for
-
Graph
- In particular, for ,
-
Graph
- The integral lattice of is obtained from the restriction of the monodromy invariant sublattice of the integral lattice of to . More precisely, for all ,
•
Graph
Proof
The first bullet follows immediately from Theorems 6.13 and 2.30, as does the statement about the pairing for . These theorems further show that the integral structure on coincides with the integral structure of .
The second bullet follows immediately from Proposition 7.5 and Theorem 2.30, as does the statements about the pairing for . These results also imply that the integral lattice of induced by , when restricted to , maps to the lattice of integral solutions of .
For the reader's convenience, we present the last statement of the theorem in detail:
Graph
Here the first and last equality are Theorem 2.30, the second equality is Theorem 6.13, and the third is Proposition 7.5.
Special cases
In some cases a stronger statement is possible. Consider the following two conditions:
• The map
-
Graph
- is an isomorphism.
- There is an equality
-
Graph
When condition (1) holds, the compact type quantum D-module of may be identified with a submodule of the monodromy invariant part of rather than a subquotient. If condition 2 holds, then by Lemma 7.3 may be identified with a quotient of the monodromy invariant part of rather than a subquotient. If both of the above conditions are satisfied, then the map identifies with (the restriction to of) the monodromy invariant part of . This in turn implies a stronger form of Theorem 8.2.
Given condition (1), the map may be inverted, which in turn defines a map given by composing with the inclusion .
Under conditions (1) and (2), Theorem 8.2 may be rephrased as follows.
Theorem 8.3
When conditions (1) and (2) are satisfied, the quantum D-module on satisfies the following:
- In a neighborhood of , is gauge-equivalent to via the transformation ;
- The monodromy invariant part of around , when restricted to , is guage equivalent to to via the transformation ;
- For ,
-
Graph
- where
- The integral lattice of is equal to the restriction of the monodromy invariant sublattice of the integral lattice of to .
We conclude with examples illustrating when the above conditions hold, and give examples showing they do not always hold.
Example 8.4
Consider Example 4.8, where is a degree-d hypersurface in projective space , and is a hypersurface in . The GIT description of is given by , , , and , with stability condition . We check below whether (1) and (2) are satisfied.
Note that both conditions concern only the cohomology of twisted sectors for . In this case that is only the untwisted sector. Thus we restrict our attention to . By (3.3.2), and the description of the fan given in Proposition 5.3, one computes
8.1.1
Graph
where u and e denote the divisors in corresponding to and respectively. A homogeneous basis is given by
Graph
On the other hand, and
Graph
The map simply sends u to u and e to 0. The vector space is equal to . It follows immediately that condition (1) is satisfied for all choices of m, k, d.
By Lemma 3.5, the compact type cohomology is generated as a module by and . In this case the module is By Condition 3.7 for , . If , then and are scalar multiples of each other, and thus . Condition (2) will therefore be satisfied automatically.
We next consider the case . In this case one checks that lies in the span of and , and therefore lies in . We can express the compact type cohomology more simply as
Graph
From (8.1.1), is equal to This module is rank one in degrees
Graph
and rank two in degree , with generators and . We conclude that
Graph
Condition (2) is therefore satisfied if and only if , which in turn holds if and only if .
In summary, in the setting of Example 4.8, condition (1) is always satisfied. Condition (2) is satisfied if and only if either or if . In particular, the example of the cubic transition of the quintic 3-fold ( and ), both conditions are satisfied and Theorem 8.3. This is consistent with [[33]].
Example 8.5
For an example where condition (1) fails, we consider a particular case of Example 4.10. Let , , , , and . Then . Let and . Then is a degree 8 Calabi–Yau hypersurface in and is a hypersurface in . Note that Condition 3.7 holds for both D and in this case.
The inertia stack has a twisted sector corresponding to the element . The corresponding twisted sectors in and are
Graph
We compute the cohomology rings
Graph
Again the map is given by sending u to u and e to 0. Note that the element is a nonzero element in . Thus condition 1 fails.
Remark 8.6
The general statement of Theorem 8.2, relating to a subquotient of (rather than simply a sub-module) is not surprising. Indeed the examples of extremal correspondences in [[27]] take a similar form.
It would be interesting to better understand what conditions on and would imply conditions (1) and (2) above.
Acknowledgements
The authors are indebted to Y.-P. Lee and Y. Ruan for their mentorship and guidance. In addition to innumerable other helpful conversations, Y. Ruan first explained the significance of extremal transitions in Gromov–Witten theory and mirror symmetry. The role of quantum Serre duality in proving other correspondences was first proposed to M. S. by Y.-P. Lee in the context of the LG/CY correspondence. M. S. also thanks H. Iritani and J. A. Cruz Morales for helpful discussions and correspondences, and thanks H. Iritani for suggesting Proposition 6.11. The authors would also like to thank the anonymous referee for their many helpful comments and suggestions. R. M. was supported by a postdoc fellowship from Center for Mathematical Sciences and Applications at Harvard University. M. S. was partially supported by NSF grant DMS-1708104.
Author Contributions
Both authors contributed equally to this manuscript.
Funding
R. M. was supported by a postdoc fellowship from Center for Mathematical Sciences and Applications at Harvard University. M. S. was partially supported by NSF Grant DMS-1708104
Availability of data and materials
This manuscript has no associated data.
Declarations
Conflict of interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Ethics approval
Not applicable.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Code availability
This manuscript has no associated code.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
[
References
1
Abramovich D, Graber T, Vistoli A. Gromov–Witten theory of Deligne–Mumford stacks. Am. J. Math. 2008; 130; 5: 1337-1398. 2450211. 10.1353/ajm.0.0017. 1193.14070
2
Abramovich D, Vistoli A. Compactifying the space of stable maps. J. Am. Math. Soc. 2002; 15; 1: 27-75. 1862797. 10.1090/S0894-0347-01-00380-0. 0991.14007
3
Aspinwall PS. Exoflops in two dimensions. J. High Energy Phys. 2015; 2015; 7: 104. 3383108. 10.1007/JHEP07(2015)104. 1388.81472
4
Avram AC, Candelas P, Jančić D, Mandelberg M. On the connectedness of the moduli space of Calabi–Yau manifolds. Nuclear Phys. B. 1996; 465; 3: 458-472. 1392861. 10.1016/0550-3213(96)00058-2. 0896.14027
5
Avram AC, Kreuzer M, Mandelberg M, Skarke H. The web of Calabi–Yau hypersurfaces in toric varieties. Nuclear Phys. B. 1997; 505; 3: 625-640. 1490782. 10.1016/S0550-3213(97)00582-8. 0896.14026
6
Behrend K, Fantechi B. The intrinsic normal cone. Invent. Math. 1997; 128; 1: 45-88. 1437495. 10.1007/s002220050136. 0909.14006
7
Borisov LA, Chen L, Smith GG. The orbifold Chow ring of toric Deligne–Mumford stacks. J. Am. Math. Soc. 2005; 18; 1: 193-215. 2114820. 10.1090/S0894-0347-04-00471-0. 1178.14057
8
Borisov LA, Paul Horja R. Applications of homological mirror symmetry to hypergeometric systems: duality conjectures. Adv. Math. 2015; 271: 153-187. 3291860. 10.1016/j.aim.2014.11.020. 1334.14024
9
Bott R, Loring WT. Differential Forms in Algebraic Topology. Graduate Texts in Mathematics. 1982: New York; Springer. 10.1007/978-1-4757-3951-0. 0496.55001
Bryan, J, Graber, T: The crepant resolution conjecture. In: Algebraic Geometry—Seattle 2005. Part 1, Volume 80 of Proceedings of Symposium on Pure Mathematics, pp 23–42. Amer. Math. Soc, Providence, RI, (2009)
Chiang T.-M, Greene B.R, Gross M, Kanter Y: Black hole condensation and the web of Calabi–Yau manifolds, vol 46, pp. 82–95. -duality and mirror symmetry (Trieste, 1995)
Ciocan-Fontanine I, Kim B. Moduli stacks of stable toric quasimaps. Adv. Math. 2010; 225; 6: 3022-3051. 2729000. 10.1016/j.aim.2010.05.023. 1203.14014
Coates T, Corti A, Iritani H, Tseng H-H. A mirror theorem for toric stacks. Compos. Math. 2015; 151; 10: 1878-1912. 3414388. 10.1112/S0010437X15007356. 1330.14093
Coates T, Givental A. Quantum Riemann–Roch. Lefschetz Serre. Ann. Math. (2). 2007; 165; 1: 15-53. 2276766. 10.4007/annals.2007.165.15. 1189.14063
Coates T, Iritani H, Jiang Y. The crepant transformation conjecture for toric complete intersections. Adv. Math. 2018; 329: 1002-1087. 3783433. 10.1016/j.aim.2017.11.017. 1394.14036
Coates T, Iritani H, Tseng H-H. Wall-crossings in toric Gromov–Witten theory. I. Crepant examples. Geom. Topol. 2009; 13; 5: 2675-2744. 2529944. 10.2140/gt.2009.13.2675. 1184.53086
Coates T, Ruan Y. Quantum cohomology and crepant resolutions: a conjecture. Ann. Inst. Fourier (Grenoble). 2013; 63; 2: 431-478. 3112518. 10.5802/aif.2766. 1275.53083
Cox DA. The functor of a smooth toric variety. Tohoku Math. J. (2). 1995; 47; 2: 251-262. 1329523. 10.2748/tmj/1178225594. 0828.14035
Cox DA, Katz S. Mirror Symmetry and Algebraic Geometry. Mathematical Surveys and Monographs. 1999: Providence; American Mathematical Society. 10.1090/surv/068. 0951.14026
Cox DA, Little JB, Schenck HK. Toric Varieties. Graduate Studies in Mathematics. 2011: Providence; American Mathematical Society
Favero D, Kelly TL. Derived categories of BHK mirrors. Adv. Math. 2019; 352: 943-980. 3975030. 10.1016/j.aim.2019.06.013. 1444.14076
Fulton W. Intersection Theory. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)]. 1984: Berlin; Springer
Givental AB. Equivariant Gromov–Witten invariants. Int. Math. Res. Not. 1996; 13: 613-663. 1408320. 10.1155/S1073792896000414. 0881.55006
Iritani H. An integral structure in quantum cohomology and mirror symmetry for toric orbifolds. Adv. Math. 2009; 222; 3: 1016-1079. 2553377. 10.1016/j.aim.2009.05.016. 1190.14054
Iritani H. Quantum cohomology and periods. Ann. Inst. Fourier (Grenoble). 2011; 61; 7: 2909-2958. 3112512. 10.5802/aif.2798. 1300.14055
Iritani H, Mann E, Mignon T. Quantum Serre theorem as a duality between quantum -modules. Int. Math. Res. Not. 2016; 9: 2828-2888. 3519131. 10.1093/imrn/rnv215. 1404.14066
Iritani H, Xiao J. Extremal transition and quantum cohomology: examples of toric degeneration. Kyoto J. Math. 2016; 56; 4: 873-905. 3568645. 10.1215/21562261-3664959. 1360.14132
Lee Y-P, Lin H-W, Wang C-L. Flops, motives, and invariance of quantum rings. Ann. Math. (2). 2010; 172; 1: 243-290. 2680420. 10.4007/annals.2010.172.243. 1272.14040
Lee Y-P, Lin H-W, Wang C-L. Towards theory in conifold transitions for Calabi–Yau threefolds. J. Differ. Geom. 2018; 110; 3: 495-541. 3880232. 10.4310/jdg/1542423628. 1423.14242
Li A-M, Ruan Y. Symplectic surgery and Gromov–Witten invariants of Calabi–Yau 3-folds. Invent. Math. 2001; 145; 1: 151-218. 1839289. 10.1007/s002220100146. 1062.53073
Lian BH, Liu K, Yau S-T. Mirror principle. I. Asian J. Math. 1997; 1; 4: 729-763. 1621573. 10.4310/AJM.1997.v1.n4.a5. 0953.14026
Mavlyutov AR. On the chiral ring of Calabi–Yau hypersurfaces in toric varieties. Compos. Math. 2003; 138; 3: 289-336. 2019444. 10.1023/A:1027367922964. 1117.14052
Mi, R: Cubic extremal transition and Gromov–Witten theory. arXiv:1711.11014
Mi, R: Gromov–Witten theory under degree-4 type II extremal transitions. arXiv:1810.05783
Morrison DR. Through the looking glass. Mirror Symmetry, III (Montreal, PQ, 1995), Volume 10 of AMS/IP Advanced Mathematical Studies. 1999: Providence; American Mathematical Society: 263-277
Ottaviani G. Varieta' Proiettive di Codimensione Piccola. 1995: Aracne, Roma; IDdAM course
Reid M. The moduli space of -folds with may nevertheless be irreducible. Math. Ann. 1987; 278; 1–4: 329-334. 909231. 10.1007/BF01458074. 0649.14021
Shoemaker, M: Narrow quantum D-modules and quantum Serre duality. To appear in Ann. Inst Fourier (Grenoble) (2018)
The Stacks project authors. The stacks project. https://stacks.math.columbia.edu (2020)
Toen B. Théorèmes de Riemann–Roch pour les champs de Deligne–Mumford. K-Theory. 1999; 18; 1: 33-76. 1710187. 10.1023/A:1007791200714. 0946.14004
Tseng H-H. Orbifold quantum Riemann–Roch, Lefschetz and Serre. Geom. Topol. 2010; 14; 1: 1-81. 2578300. 10.2140/gt.2010.14.1. 1178.14058
Wang S-S. On the connectedness of the standard web of Calabi–Yau 3-folds and small transitions. Asian J. Math. 2018; 22; 6: 981-1003. 3919548. 10.4310/AJM.2018.v22.n6.a1. 1441.14134
]
By Rongxiao Mi and Mark Shoemaker
Reported by Author; Author