Patrick Lei

All notes

Blowup formula in GW theory

Columbia, Spring 2024.

Enumerative invariants and birational geometry
Spring 2024
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

These notes were taken during the lectures using neovim. Any errors are mine and not the speakers’. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the lecturers. Also, notation may differe between lecturers. If you find any errors, please contact me at plei@math.columbia.edu.

Acknowledgements

I would like to thank Shaoyun Bai for co-organizing the seminar with me. I would also like to thank Song Yu for bringing the works [PSW24; LSW24] about wall-crossing for Grassmannian flops to our attention.

References used

We used the following references in the seminar:

  1. 1.

    The main references for Section 1.1 are [Giv04; Dub96; Coa03].

  2. 2.

    The reference for Section 1.2 is [CG07].

  3. 3.

    The reference for Section 1.3 is [Iri17].

  4. 4.

    The main references for Section 1.4 are [CIJ18; AGV08; Tse10]

  5. 5.

    The reference for Section 1.5 is [CIJ18, Section 3].

  6. 6.

    The reference for Section 2.1 and Section 2.2 is [CIJ18].

  7. 7.

    The reference for Section 2.3 is [PSW24]. Also see the very similar [LSW24].

  8. 8.

    The reference for Chapter 3 is [IK24].

  9. 9.

    The reference for Chapter 4 is [Iri23].

1 Preliminaries

1.1 Givental formalism (Patrick, Feb 01)

1.1.1 Introduction

Let X be a smooth projective variety. Then for any g,n∈ℤ≥0,β∈H2⁢(X,ℤ), there exists a moduli space ℳ¯g,n⁢(X,β) (Givental’s notation is Xg,n,β) of stable maps f:C→X from genus-g, n-marked prestable curves to X with f∗⁢[C]=β. It is well-known that ℳ¯g,n⁢(X,β) has a virtual fundamental class

In addition, there is a universal curve and sections

In this setup, there are tautological classes

This allows us to define individual Gromov-Witten invariants by

These invariants satisfy various relations. The first is the string equation:

The next is the dilaton equation:

Finally, we have the divisor equation when one insertion is a divisor D∈H2⁢(X):

It is often useful to package Gromov-Witten invariants into various generating series.

Definition 1.1.1.

The quantum cohomology Q⁢H∗⁢(X) of X is defined by the formula

for any t∈H∗⁢(X). This is a commutative and associative product.

The small quantum cohomology is obtained by setting t=0 and the ordinary cohomology is obtained by further setting Q=0.

Remark 1.1.2.

Convergence of the formula does not hold in general, so quantum cohomology needs to be treated as a formal object.

Definition 1.1.3.

Let ϕi be a basis of H∗⁢(X) and ϕi be the dual basis. Then the J-function of X is the cohomology-valued function

Definition 1.1.4.

The genus-0 GW potential of X is the (formal) function

The associativity of the quantum product is equivalent to the PDE

for any a,b,c,d, which are known as the WDVV equations. Here, we choose coordinates on H∗⁢(X) and set z=0 (only consider primary insertions). In addition, set ηe⁢f to be the components of the Poincaré pairing and let ηe⁢f be the inverse matrix.

1.1.2 Frobenius manifolds

A Frobenius manifold can be thought of as a formalization of the WDVV equations.

Definition 1.1.5.

A Frobenius manifold is a complex manifold M with a flat symmetric bilinear form ⟨−,−⟩ (meaning that the Levi-Civita connection has zero curvature) on T⁢M and a holomorphic system of (commutative, associative) products ⋆t on Tt⁢M satisfying:

  1. 1.

    The unit vector field 𝟏 is flat: ∇𝟏=0;

  2. 2.

    For any t and a,b,c∈Tt⁢M, ⟨a⋆tb,c⟩=⟨a,b⋆tc⟩;

  3. 3.

    If c⁡(u,v,w)≔⟨u⋆tv,w⟩, then the tensor (∇zc)⁢(u,v,w) is symmetric in u,v,w,z∈Tt⁢M.

If there exists a vector field E such that ∇∇⁡E=0 and complex number d such that:

  1. 1.

    ∇∇⁡E=0;

  2. 2.

    ℒE⁢(u⋆v)−ℒE⁢u⋆v−u⋆ℒE⁢v=u⋆v for all vector fields u,v;

  3. 3.

    ℒE⁢⟨u,v⟩−⟨ℒE⁢u,v⟩−⟨u,ℒE⁢v⟩=(2−d)⁢⟨u,v⟩ for all vector fields u,v,

then E is called an Euler vector field and the Frobenius manifold M is called conformal.

Example 1.1.6.

Let X be a smooth projective variety. Then we can give H∗⁢(X) the structure of a Frobenius algebra with the Poincaré pairing and the quantum product. Note that the quantum product does not converge in general, so we must treat this as a formal object. The Euler vector field is given by

where a general element of H∗⁢(X) is given by t=∑iti⁢ϕi. We will also impose that ϕ1=𝟏. There is another very important structure, the quantum connection, which is given by the formula

Here, μX is the grading operator, defined for pure degree classes ϕ∈H∗⁢(X) by

Finally, in the direction of the Novikov variables, we have

Remark 1.1.7.

For a general conformal Frobenius manifold (H,(−,−),⋆,E), there is still a deformed flat connection or Dubrovin connection given by

Definition 1.1.8.

The quantum D-module of X is the module H∗(X)[z]⟦Q,t⟧ with the quantum connection defined above.

Remark 1.1.9.

It is important to note that the quantum connection has a fundamental solution matrix SX⁢(t,z) given by

It satisfies the important equation

Using this formalism, the J-function is given by SX∗⁢(t,z)⁢𝟏=z−1⁢JX⁢(t,z).

1.1.3 Givental formalism

The Givental formalism is a geometric way to package enumerative (CohFT) invariants cleanly. We begin by defining the symplectic space

with the symplectic form

This has a polarization by Lagrangian subspaces

giving ℋ≅T∗⁢ℋ+ as symplectic vector spaces. Choose Darboux coordinates p¯,q¯ on ℋ. For example, there is a choice in Coates’s thesis which gives a general element of ℋ as

Taking the dilaton shift

we can now think of ℱX has a formal function on ℋ+ near q=−z. This convention is called the dilaton shift.

Before we continue, we need to recast the string and dilaton equations in terms of ℱX. Write tx=∑tki⁢ϕi. Then the string equation becomes

and the dilaton equation becomes

There are also an infinite series of topological recursion relations

We can make sense of these three relations for any (formal) function ℱ on ℋ+.

Now let

be the graph of dℱ. This is a formal germ at q=−z of a Lagrangian section of the cotangent bundle T∗⁢ℋ+ and is therefore a formal germ of a Lagrangian submanifold in ℋ.

Theorem 1.1.10.

The function ℱ satisfies the string equation, dilaton equation, and topological recursion relations if and only if ℒ is a Lagrangian cone with vertex at the origin q=0 such that its tangent spaces L are tangent to ℒ exactly along z⁢L.

Because of this theorem, ℒ is known as the Lagrangian cone. It can be recovered from the J-function by the following procedure. First consider ℒ∩(−z+z⁢ℋ−). Via the projection to −z+H along ℋ−, this can be considered as the graph of the J-function. Next, we consider the derivatives ∂J∂ti, which form a basis of L∩z⁢ℋ−, which is a complement to z⁢L in L. Then we know that

so

Writing these in terms of the first derivatives ∂J∂ti and using the fact that J is a solution of the quantum connection, so we recover the Frobenius structure of quantum cohomology.

We will now express some classical results in this formalism. Let X be a toric variety with toric divisors D1,…,DN such that D1,…,Dk form a basis of H2⁢(X) and Picard rank k. Then define the I-function

Theorem 1.1.11 ([Giv98]).

The formal functions IX and JX coincide up to some change of variables, which if c1⁢(X) is semi-positive is given by components of the I-function.

Theorem 1.1.12 (Mirror theorem in this formalism).

For any t, we have

Another direction in Gromov-Witten theory is the Virasoro constraints. In the original formulation, these involved very complicated explicit differential operators, but in the Givental formalism, there is a very compact formulation.

Define ℓ−1=z−1 and

Then define

Theorem 1.1.13 (Genus-0 Virasoro constraints).

Suppose the vector field on ℋ defined by ℓ0 is tangent to ℒ. Then the same is true for the vector fields defined by ℓn for any n≥1.

Proof.

Let L be a tangent space to ℒ. Then if f∈z⁢L⊂ℒ, the assumption gives us ℓ0⁢f∈L. But then z⁢ℓ0⁢f∈z⁢L, so ℓ0⁢z⁢ℓ0⁢f=ℓ1⁢f∈L. Continuing, we obtain ℓn⁢f∈L for all n. □

Later, we will learn that the Quantum Riemann-Roch theorem can be stated in this formalism. Let ℒtw be the twisted Lagrangian cone (where the twisted theory will be defined next week).

Theorem 1.1.14 (Quantum Riemann-Roch).

For some explicit linear symplectic transformaiton Δ, we have ℒtw=Δ⁢ℒ.

1.1.4 Quantization

In the last part of the talk, we will briefly discuss the quantization formalism, which encodes the higher-genus theory. In Darboux coordinates pa,qb, we will quantize symplectic transformations by the standard rules

This determines a differential operator acting on functions on ℋ+.

We also need the genus-g potential

and the total descendent potential

In this formalism, the Virasoro conjecture can be expressed as follows. Let Ln=ℓ̂n+cn, where cn is a carefully chosen constant.

Conjecture 1.1.15 (Virasoro conjecture).

If L−1⁢𝒟=L0⁢𝒟=0, then Ln⁢𝒟=0 for all n≥1.

In this formalism, the higher-genus version of the Quantum Riemann-Roch theorem takes the very simple form

Theorem 1.1.16 (Quantum Riemann-Roch).

Let 𝒟tw be the twisted descendent potential. Then

1.2 Quantum Riemann-Roch (Shaoyun, Feb 08)

We will state and prove the Quantum Riemann-Roch theorem in genus 0, following Coates-Givental.

1.2.1 Twisted Gromov-Witten invariants

Again, let X be a smooth projective variety. Let E be a vector bundle on X. We should note that

is the universal curve, and the universal morphism is simply evn+1. We will consider the sheaf

We need to check that this is a well-defined K-theory class. Choose an ample line bundle L→X. By definition, for N≫1, the cohomology

whenever i≥1. This gives us an exact sequence

For any stable map f:Σ→X of positive degree, we obtain a long exact sequence

so we obtain

This expresses E0,n,β as a difference of vector bundles.

We will now introduce a universal characteristic class

where s0,s1,s2,… are formal variables and chk is the k-th Chern character

where xi are the Chern roots.

Example 1.2.1.

Let E→X be a vector bundle and equip it with the fiberwise ℂ∗-action by scaling. Let λ be the equivariant parameter and ρi be the Chern roots. Then

We then rewrite

so for the (equivariant Euler class), we obtain

We are now ready to define the (E,𝐜)-twisted Gromov-Witten invariants.

Definition 1.2.2.

Define the twisted Gromov-Witten invariants by

for αi∈H∗⁢(X) and ki∈ℤ≥0.

We will now construct the Lagrangian cone for the twisted theory. Let R be the coefficient ring containing s0,s1,… and define

We also introduce the twisted Poincaré pairing

The symplectic structure is defined by

There is a polarization

with

Finally, we have the twisted genus-0 descendent potential

Identifying ℋXtw with T∗⁢ℋ+tw, we obtain the twisted Lagrangian cone ℒXtw as the graph of dℱX,tw0. Denote the untwisted Lagrangian cone as ℒX.

Theorem 1.2.3.

We have

where

Here, the Bernoulli numbers B2⁢m are defined by

1.2.2 Proof of Theorem 1.2.3

The idea is to use the Grothendieck-Riemann-Roch theorem.

Proposition 1.2.4.

We can write

where

Here, Z0,n+1,β is formed by the nodes of π, Z~0,n+1,β is a double cover of Z0,n+1,β formed by a choice of branch of the nodes, ψ+ and ψ− are the ψ-classes at the two branches of the nodes, and

is the “inclusion.”

Proof.

We will first assume that M¯0,n+1⁢(X,β), ℳ¯0,n⁢(X,β), and Z0,n+1,β are all smooth and that π⁡(Z0,n+1,β) is a normal crossings divisor. In general, we need a Cartesian diagram

Continuing in the ideal situation, we apply Grothendieck-Riemann-Roch1 to obtain

where td∨ is the dual Todd class, defined by −x1−et⁢x, and Ωπ is the sheaf of relative differentials.

We then have two short exact sequences

and

where Di is the divisor where the marked points i,n+1 collide and their component has exactly three special points. Now we obtain

in K-theory. Using the facts that c1⁢(Ln+1)=ψn+1, Di∩Dj=∅ for i≠j, and Di∩Z0,n+1,β=∅, we see that Ln+1 is trivial when restricted to Di and Z0,n+1,β. Now we apply the dual Todd class.

Lemma 1.2.5.

If x1∪x2=0, then

Using the lemma, we obtain

The first term in the statement comes from the dual Todd class of Ln+1, the second comes from

and the relation between 𝒪⁡(−Di) and Li, and the last term can be found in Appendix A of Coates-Givental. □

To obtain the Quantum Riemann-Roch theorem, we use the previous proposition and manipulate the generating function. If E is convex and Y⊂X is a complete intersection defined by E, then ℒXtw is closely related to ℒY, so we are able to study the Gromov-Witten theory of Y using this.

1.3 Shift operators (Melissa, Feb 15)

Let X be a semiprojective smooth variety. This means that X is projective over its affinization. Also assume that X has an action by T=(ℂ×)m such that all T-weights in H0⁢(X,𝒪) are contained in a strictly convex cone in Hom⁡(T,ℂ×)ℝ and H0⁢(X,𝒪)T=ℂ. All such X imply that

  1. (a)

    The fixed locus XT is projective;

  2. (b)

    The T-variety X is equivariantly formal. This means that HT∗⁢(X) is a free module over HT∗⁢(pt)=ℚ⁡[λ]≔ℚ⁡[λ1,…,λm] and there is a non-canonical isomorphism

    as HT∗⁢(pt)-modules.

  3. (c)

    The evaluation maps evi:X0,n,d→X are proper.

Using (b), we may choose a basis {ϕi}i=0N of HT∗⁢(X) over HT∗⁢(pt). Let τi be the dual coordinates.

1.3.1 Equivariant big quantum cohomology

Let (−,−) be the T-equivariant Poincaré pairing, which in general takes values in ℚ⁡(λ). Then the T-equivariant big quantum product is defined by

This can also be defined using the evaluation maps

as

1.3.2 Quantum connection

We will define

by setting

We can view z as the loop variable by setting T̂=T×ℂ×. If the extra copy of ℂ× acts trivially on X, then

This has a fundamental solution

where

This satisfies the differential equation

which is equivalent to

The solution has the form

1.3.3 Shift operators

Let k:ℂ×→T be a cocharacter of T. Then define a T̂-action ρk on X by

for t∈T,u∈ℂ×,x∈X. Under the group automorphism

the identity map (X,ρ0)→(X,ρk) is T̂-equivariant, so we obtain isomorphisms

Now define the bundle

where ℂ× acts by

This is an X-bundle over ℙ1 with an action on T̂ by

Setting 0=[1,0] and ∞=[0,1], we see that T̂ acts on X0 by ρ0 and X∞ by ρk.

Definition 1.3.1.

A cocharacter k:ℂ×→T is seminegative if all weights of H0⁢(X,𝒪) are nonpositive with respect to k and is negative if all nonzero weights of H0⁢(X,𝒪) are negative.

Lemma 1.3.2.

If k is seminegative, then Ek is semiprojective.

Now let π:Ek→ℙ1 be the projection. We now consider section classes, which are those effective classes in H2⁢(Ek,ℤ) satisfying π∗⁢d=[ℙ1]. For the ℂ×-action on X given by k, there is a unique fixed component Fmin whose normal weights are all positive (one way to see this is to consider the moment map of the corresponding circle action). Therefore, there is a minimal section class σmin corresponding to Fmin.

Lemma 1.3.3.

Given τ∈HT∗⁢(X), there exists τ̂∈HT̂∗⁢(Ek) such that τ̂|X0=τ and τ̂|X∞=Φk⁢(τ).

Lemma 1.3.4.

If k is seminegative, then

Definition 1.3.5.

Let k:ℂ×→T be seminegative. Given τ∈HT∗⁢(X), we define the shift operator

by the formula

where α∈Ht̂,ρ0∗⁢(X) and β∈HT̂,ρk∗⁢(X). We also define

Theorem 1.3.6.

We have the formula

where 𝒮k is defined via the commutative diagram

Here, we define

where

is the normal bundle of Fi in X and ρi,α,j are its Chern roots.

The idea of the proof is to decompose

Using the exact sequence

we obtain the explicit formulae

This gives the virtual normal bundle, and using virtual localization, we obtain

where

Using the unitarity property of M, we obtain the desired result.

1.4 Orbifold stuff (Patrick, Apr 04)

1.4.1 Orbifold Gromov-Witten theory

Let X be a smooth and separated Deligne-Mumford stack of finite type over ℂ.

Definition 1.4.1.

The inertia stack of X is the fiber product in the diagram

More concretely, we may think about I⁢X as parameterizing pairs (x,g), where x∈X and g∈Aut⁡(x). There is another description of I⁢X if X lives over ℂ. In general, I⁢X is disconnected. We will write

It also has an important morphism inv:I⁢X→I⁢X given by (x,g)↦(x,g−1).

Definition 1.4.2.

A morphism X→Y of algebraic stacks is representable if for all schemes S and morphisms S→Y, the fiber product X×SY is an algebraic space.

Theorem 1.4.3.

Let

denote the stack of representable morphisms from classifying stacks of roots of unity to X (the cyclotomic inertia stack). Then Iμ⁢X≃I⁢X.

We need to make one more definition, which will appear as a degree shift on cohomology. Let (x,g)∈Xi. Because ⟨g⟩⊂Aut⁡(x) is cyclic, there is a decomposition

where Vℓ is the eigenspace with eigenvalue e2⁢π⁢−1⁢ℓri and ri is the order of g. Then the function

is constant on Xi, so we denote its value by age⁡(Xi).

Recall that by the Keel-Mori theorem, X (which has finite inertia) has a coarse moduli space |X|, which is an algebraic space satisfying two properties:

  • •

    The morphism π:X→|X| is bijective on k-points whenever k is an algebraically closed field;

  • •

    |X| is initial for morphisms from X to any algebraic space.

From now on, we will assume that |X| is quasiprojective, and in particular that it is a scheme.

Moduli of stable maps
Definition 1.4.4.

The moduli space of stable maps ℳ¯g,n⁢(X,β) parameterizes objects

where

  1. 1.

    C is a prestable balanced twisted curve of genus g. This means that C has stacky structure only at nodes and marked points, and the nodes are formally locally [(Spec⁡ℂ⁡[x,y]/x⁢y)/μr], where μr acts by ζ⁡(x,y)=(ζ⁢x,ζ−1⁢y);

  2. 2.

    Σi⊂C is an étale cyclotomic gerbe over T with a trivialization for all i;

  3. 3.

    f:C→X is representable and the induced morphism between coarse moduli spaces is a stable map of degree β with n marked points.

We see that ℳ¯g,n⁢(X,β) has evaluation maps evi:ℳ¯g,n⁢(X,β)→I⁢X. It is also disconnected, with the connected components being indexed by components of I⁢X. Let

Then

Each component has a virtual fundamental class

of virtual dimension

given by the relative perfect obstruction theory (R⁢π∗⁢f∗⁢T⁢X)∨, where π:C→ℳ¯g,n⁢(X,β) is the universal curve, over the moduli stack 𝔐g,ntw of prestable twisted curves. Because we chose to work with trivialized gerbe markings, we need to multiply the virtual fundamental class as follows. Note that the j-th marked point is

Here, if x=[B⁢μr→X]∈Xij⊂I⁢X, then rij=r. Then set

Now consider the morphism p:ℳ¯g,n⁢(X,β)→ℳ¯g,n⁢(|X|,β) given by taking the coarse moduli space. Let C|X|→ℳ¯g,n⁢(|X|,β) be the universal curve and σi,|X| be the marked points. Then the descendant classes2 are defined to be

Quantum cohomology

We are now able to define Gromov-Witten invariants. Let αj∈Hpj⁢(Xij,ℂ). Then define

We are still able to form generating series ℱg,JX,… as before, and the invariants satisfy the string, dilaton, and divisor equations (although we have to be careful that the marked point we delete is a scheme point), so the orbifold Gromov-Witten theory has a Lagrangian cone ℒX⊂ℋ.

The orbifold Poincaré pairing is defined by the formula

where ∪ denotes the usual cup product. This is well-defined because of the formula

when X is proper. When X is not proper, we will assume we are working equivariantly. Now we may define the quantum product by the formula

for a,b,c,τ∈H∗⁢(I⁢X,ℂ). Restricting to the degree 0 part and setting τ=0, we obtain the orbifold cup product, which is given by

Denote HCR∗⁢(X)≔(H∗⁢(I⁢X,ℂ),⋆). The orbifold cup product is graded for the grading deg⁡(a)=p+2⁢age⁡(Xi) for a∈Hp⁢(Xi). Using the quantum product, we may define the quantum connection and its fundamental solution.

1.4.2 Toric Deligne-Mumford stacks

We will assume the reader is familiar with the fan presentation of a toric variety. If you are not, there are many references.

Definition 1.4.5.

An extended stacky fan is a quadruple 𝚺=(N,Σ,β,S) of

  1. 1.

    A finitely generated abelian group N of rank n;

  2. 2.

    A rational simplicial fan Σ in Nℝ=N⊗ℝ;

  3. 3.

    A homomorphism β:ℤm→N. We will write bi=β⁡(ei)∈N for the image of the standard basis vector ei∈ℤm and b¯i for its image in Nℝ;

  4. 4.

    A subset S⊂{1,…,m}

satisfying the following conditions:

  1. 1.

    The set Σ⁡(1) of 1-dimensional cones is exactly the set {ℝ≥0⋅b¯i|i∉S};

  2. 2.

    For all i∈S, b¯i∈|Σ|.

We will now assume that |Σ| is convex and full-dimensional and, that there is a strictly convex piecewise linear function f:|Σ|→ℝ which is linear on each cone, and that β is surjective. From this data, we will now obtain a GIT presentation. Define 𝕃 by the exact sequence

Then define K≔𝕃⊗ℂ×. Then define Di∈𝕃∨ to be the image of the i-th standard basis vector in (ℤm)∨ under the last arrow in the exact sequence

Finally, set

Choose a stability condition

Then we define

The ample cone is Cω′⊂𝕃ℝ∨/∑i∈Sℝ⁢Di≅H2⁢(X𝚺,ℝ), which is defined in the same way as Cω after deleting S from the extended stacky fan, and the cone of effective curve classes is its dual.

Orbifold cohomology

First, we will describe the equivariant cohomology of X𝚺. Let 𝒬=(ℂ×)m/K. Then if ui is Poincaré dual to (xi=0⊂(ℂm)s)/K, we have

where

There is a combinatorial description of the components of the inertia stack I⁢X𝚺. Because X𝚺 is a global quotient, the components of the inertia stack correspond to elements g∈K such that ((ℂm)s)g is nonempty. Equivalently, if we define

then the components of I⁢X𝚺 are in bijection with 𝕂/𝕃. To give a description in terms of the fan, for any σ∈Σ⁡(n), define

and then

Then there is a natural bijection 𝕂/𝕃≅Box⁡(𝚺). For any f∈𝕂/𝕃, Xf is a toric DM stack with K,𝕃,ω the same as for Xω and characters Di for i such that Di⋅f∈ℤ. At the level of fans, this corresponds to killing the minimal cone of Σ containing the corresponding v¯.

We will now give the orbifold cohomology of X𝚺. Define the deformed group ring ℂ⁢[N]𝚺 as the vector space ℂ⁡[N] with product given by

Then there is an isomorphism of rings [BCS05]

Remark 1.4.6.

This result also works in families over a base B [Jia08], where ℂm is replaced by a direct sum of m line bundles on B. Then we need to add a c1⁢(Lχ) to the relations and obtain

1.5 Gamma-integral structure (Patrick, Apr 04)

Let I⁢X=⨆v∈𝖡Xv and qv:Xv→X be the restriction of I⁢X→X. Let E be a T-equivariant vector bundle on X. Recall that v corresponds to some gv∈K, so we obtain an eigenbundle decomposition

where Ev,f is the subbundle where gv acts by e2⁢π⁢i⁢f. We now define the orbifold Chern character to be

Now let δv,f,j be the Chern roots of Ev,f. We define the orbifold Todd class to be

The Γ̂-class should be a square root of this and is defined by

where we expand Γ around 1−f. The reflection formula for the Γ-function implies that the Xv-component of Γ̂⁢(E∨)∪Γ̂⁢(E) is given by

Here, deg0 is the grading operator given by the degree without age shifting.

Definition 1.5.1.

Define the K-group framing 𝔰:KT(X)→HCR,T∗(X)⊗RTRT[logz]⦇z−1k⦈⟦Q,τ⟧ by the formula

where L⁡(τ,z) is the fundamental solution to the quantum connection, μ is the usual grading operator given by 12⁢(deg−dimX) on homogeneous elements, and ρ=c1⁢(T⁢X)∈H2⁢(X).

Proposition 1.5.2.

Define the equivariant Euler pairing by

and the modified version χz⁢(E,F) by replacing the equivariant parameters λj by 2⁢π⁢i⁢λjz. Then

Remark 1.5.3.

Everything we have discussed so far makes sense for toric DM stacks after specializing Q=1.

2 Crepant transformation conjecture

2.1 Toric wall-crossings in GW theory (Davis, Apr 11)

Our goal is to study GIT wall-crossing in toric Gromov-Witten theory. For toric varieties X± related across a wall, we want to define ℳ⊃U± such that X±∈U±, as well as quantum connections on U± that restrict to X± appropriately.

2.1.1 GIT wall-crossing

Recall that the data of an (extended) stacky fan is equivalent to GIT data consisting of

  1. 1.

    A vector space V of dimension m;

  2. 2.

    A torus K;

  3. 3.

    Characters Di∈char⁡(K) for i=1,…,m;

  4. 4.

    A stability condition ω∈charℝ⁡(K);

  5. 5.

    A set of anticones Aω={I⊂{1,…,m}|ω∈∑i∈Iai⁢Di,ai∈ℝ>0}.

Remark 2.1.1.

The DM stack Xω is defined to be [Uω/K], where

is the semistable locus. Therefore, if ω1,…,ω2 are stability conditions, there is a birational map Xω1⇢Xω2 induced by identifying the dense open tori.

Suppose that ω1∈C1,ω2∈C2 are separated by a wall W in the space of stability conditions. Choose ω0∈W∩C1=W∩C2 and let

which may not be Deligne-Mumford. Choose a resolution X~ of Xω1,Xω2 completing the diagram

We will assume that ∑i=1MDi∈W. Under this assumption, the wall-crossing is crepant:

Example 2.1.2.

If e is a vector perpendicular to W, the 𝔾m it generates may not be fixed. For example, let K=𝔾m2 and V=ℂ3. Let D1=(1,0), D2=(1,2), and D3=(0,2). Therefore,

Let ω≔(ωx,ωy). If ωx,ωy>0, then Aω⊃{{1,2,3},{1,3}}. Then D2 appears if either

Therefore, there is a wall givven by 2⁢ωx=ωy. In the chamber ωy>2⁢ωx>0, we have

and therefore Uω=(ℂ2∖0)×ℂ×. Taking the quotient, we see that

In the other chamber, we have

which implies that Uω=ℂ××(ℂ2∖0), so

For a stability condition ω0 on the wall, we will obtain

The middle ℂ× is fixed by the cocharacter (2,−1), so X0 is non-DM.

2.1.2 The secondary variety

Definition 2.1.3.

The wall and chamber structure on charℝ⁡(K) defines a fan on charℝ⁡(K) called the secondary fan. The associated toric variety is called the secondary toric variety. For us, we will consider the subfan consisting of cones Cω1,Cω2, and their faces. Call the corresponding toric variety ℳ. This is a moduli space of Landau-Ginzburg models mirror to the Xi.

Unfortunately, ℳ is generally singular, so we will consider a smooth finite cover ℳreg. Explicitly, we see that

Call the charts U+,U−. Define

and let 𝕃~i be the free ℤ-submodule of cocharℚ⁡(K) generated by 𝕂i.

Example 2.1.4.

Continung with the previous example, recall that

If we want (fx,fy)⋅(0,2)∈ℤ, then we see that fy∈12⁢ℤ, so 𝕃~1=ℤ×12⁢ℤ.

Remark 2.1.5.

Recall that

where

is the set of indices contained in every anticone. Therefore, we can split

Claim 2.1.6.

There exist ξj± such that

  1. 1.

    ξj+|W=ξj−|W for all j∈S+∩S−;

  2. 2.

    For all j∈S+⁢Δ⁢S−, ξj±|W=0;

  3. 3.

    All ξj±∈𝕂±.

Claim 2.1.7.

We have a decomposition

Furthermore, W∩𝕃~+∨=W∩L~−∨.

Proof.

If j∈S+, then j∈I for all I∈Aω+, so Dj⁢f∈ℤ. Thus Dj∈L~i+. In the other direction, if v∈L~+∨, v⁢ξi∈ℤ because ξj∈𝕂+. Therefore,

Corollary 2.1.8.

There exists an integral basis for L~i∨ of the form

such that pk±∈C¯±′ live in the closure of the ample cones of Xi and pk+=pk− as classes of CW′.

Now we have maps

given by the formulae

for cocharacters d∈cochar⁡(K).

Remark 2.1.9.

We may reorder the basis pi as {q1,…,qr} such that qr is the unique vector not on the wall W. Dually, we may write {yi,xj} as {z1,…,zr} such that zr is dual to qr. In these coordinates, we can write

Let ℳ̂reg be the analytification of ℳreg=U+∪U−.

2.1.3 Extending the I-functions

Let

be the decomposition of some d∈cochar⁡(K). Then the I-functions of X± are given by

Precisely, we have

In the above formula, we will make the substitutions Q=1 and

Here, λj=θ+⁢(Dj), where θ+ is the restriction to H2, and c0⁢λ=∑λi.

A more explicit formula is given by

where [−d]∈𝕂/𝕃 is the twisted sector.

Claim 2.1.10.

The I-function I+ defines a convergent power series.

Idea of proof.

Choose f∈𝕂/𝕃 and d such that [−d]=f. Then X+f⊂I⁢X+ is a connected component and a closed substack of X+. The restriction

is injective. Then

where βj=ujz, x=yrpr+⋅e, and

For k≪0, because Rj=⟨Dj⋅d⟩=0 for j∈δ⊂If and Dj⋅e>0, the xk-term vanishes. Convergence follows by the ratio test. □

Because Φ solves an explicit differential equation1 with singularities only at 0,∞,∏(Dj⋅e)Dj⋅e, it can be analytically continued to

For just X+, there are differential operators Pi such that

where L−1=1+O⁡(z−1) and Γ=const+O⁡(z). Therefore, we may determine L−1 by Birkhoff factorization. This gives us quantum connections on ℳ~±.

2.2 Crepant transformation conjecture for toric complete intersections (Davis, Apr 18)

Our goal is now to relate I± to each other. More precisely, they will differ by a gauge transformation, which is a Fourier-Mukai transform. The main tool will be to compute in localized equivariant cohomology.

2.2.1 Gauge transformation

First, we will rewrite the I-functions as ratios of Γ-functions:

where ι[−d] is the age. We will not work directly with these. Instead, we will consider

Using the Γ̂ class, it is related to the I-function by the formula

Recall that T-fixed points on IX+ correspond to pairs (δ,f), where δ is a minimal anticone and f∈𝕂+/𝕃. Then we compute

Note that for j∈δ, uj⁢(δ)=0. In addition, Dj⋅d∈ℤ, so if it is negative, then the corresponding Γ-function vanishes. Therefore, we define

and the I-function becomes

If δ∈𝒜ω+∩𝒜ω−, then ι∗⁢(δ,f)⁢H+=ι(δ,f)∗⁢H−. Otherwise, δ has the form {j1,…,jr−1,j+}, where all Dji∈W for i<r and Dj+⋅e>0 for the normal vector to W pointing towards ω+.

Definition 2.2.1.

A pair (δ±∈𝒜ω±) are next to each other if both are of the form

where ±Dj±⋅e>0. Likewise, (δ±,f±) are next to each other if (δ±) are and f−=f++α⁢e∈𝕃ℚ/𝕃.

Lemma 2.2.2.

If δ+ is next to δ−, for all j and for any j−∈δ−∩δ+c,

Proof.

Expand

Then Dj⋅e=c−⁢Dj−⋅e. Because

and uj−⁢(δ−)=0, we see that

This implies that

Claim 2.2.3.

We have the formula

where c is some explicit matrix2 made of products of things like

Proof.

Idea of proof If d∈δ+, then Dj⋅d≥0 for all j∈δ+, so we can write d=D++k⁢e, where d+∈δ+v⁢e⁢e and D+−e∉δ+∨. We now obtain

Using the reflection formula

we may write

We obtain poles at the following locations:

  1. (1)

    Γ⁡(1−s) gives poles at s=1,2,3,…;

  2. (2)

    Γ⁡(s) gives poles at s=0,−1,−2,…;

  3. (3)

    The last term gives poles at s=(−Dj−⋅e)−1⁢((2⁢π⁢i)−1⁢uj⁢(δ+)+Dj−⋅d+−n).

We can then rewrite this as a contour integral where the contour is such that all poles of type (1) are on the right and all poles of types (2) and (3) except the one at 0 are on the left. For small y, we have

and for large y, we have

Choose some j− for which Dj−⁢e<0, we obtain

Setting δ−={j1,…,jr−1,j−} and

we use the lemma to rewrite some Γ⁡(…⁢uj⁢(δ+)) in terms of uj⁢(δ−) and conclude. □

We have therefore obtained the formula

Now define

We have a commutative diagram

2.2.2 The Fourier-Mukai transform

We will now prove that 𝕌 is given by the Fourier-Mukai transform. Here, let X~ be a common toric blowup of X± as in

and 𝔽⁢M≔(f+)∗⁢(f−)∗≔KT0⁢(X−)→KT0⁢(X+). We see that KT0⁢(X±) is generated by [L±(ρ)], where ρ∈char⁡(K) and KT0⁢(X~) is generated by [L±⁢(ρ,n)]. We will now move to the fixed point basis and see that

Remark 2.2.4.

Let xδ be a fixed point of X± and ρ be in its isotropy. Then eδ,ρ=(ιδ)∗⁢ρ.

We now compute 𝔽⁢M⁡(eδ,ρ).

  1. 1.

    If δ∈𝒜−∩𝒜+, then 𝔽⁢M⁡(eδ,ρ)=eδ,ρ;

  2. 2.

    Otherwise, we have a complicated formula.

The computation involves doing localization and relating anticones of X± to those of X~. We then obtain a commutative diagram

by direct computation.

2.2.3 Crepant transformation conjecture

We are now ready to state the crepant transformation conjecture.

Theorem 2.2.5.

Let ℋ⁡(X±)≔HCR,T∗⁢(X±)⁢⦇z−1⦈. Then there exists a degree-preserving symplectic transformation

such that

  1. 1.

    I+=𝕌⁢I− after analytic continuation;

  2. 2.

    If g±:X±⁢X¯= is the map to the toric blowdown, then

    for all v∈HT2⁢(X¯0);

  3. 3.

    𝕌 fits into the commutative diagram Equation 2.1.

Theorem 2.2.6 (Crepant transformation conjecture for toric DM stacks).

Let (F±,E±,∇±) be the quantum connections on X±. Then there exists a gauge transformation

such that

  1. 1.

    ∇+Θ=Θ⁢∇−;

  2. 2.

    Θ is homogeneous of degree 0;

  3. 3.

    Θ preserves the orbifold Poincaré pairings.

The proof starts by applying differential operators to 𝕌⁢I−=I+ to obtain

We then set Θ=γ+⁢γ−−1 and obtain

2.3 Wall-crossing for Grassmannian flops (Kostya, Apr 25)

We will now consider a wall-crossing in a nonabelian situation. This relies on reducing to the abelian case and citing CIJ.

2.3.1 Geometric setup

Let B be a smooth projective variety and let E,F be vector bundles on B which split as direct sums of line bundles

Now let

and G=G⁢Lr. G acts on the total space of V by the formula

Let X±=V⫽±detG be the corresponding GIT quotients of V. We obtain

The birational transformation here is known as a Grassmannian flop.

Example 2.3.1.

Consider the quiver given by

The corresponding moduli of semistable representations is

Then

There is an action of 𝕋=(ℂ×)2⁢n on the total space of E⊕F, which induces an action of 𝕋 on V. Denote

Let R=ℂr, viewed as a right G-representation. Then R×V→V is a G-equivariant vector bundle on V which descends to X± and are the pullbacks of S+∨ and S−, repsectively. Now denote by

the Chern roots of R∨. Then

where I+ is the ideal given by

and I− is defined similarly. Let R𝕋=H𝕋∗⁢(pt) and S𝕋=R𝕋,loc. Finally, let Ŝ𝕋 be a completion of S𝕋 containing exponentials and Γ-functions. The 𝕋-fixed loci in X± correspond to r×r minors of the corresponding Grassmannians. We encode them by

which are strictly order-preserving, as in δi≔δ⁡(i)<δ⁡(j)≕δj if i<j. In other words,

Lemma 2.3.2.

We have

This implies that

2.3.2 Abelianization

Let (C×)r⊂G=GLr be the maximal torus which acts diagonally on ℂr. Then the abelianized quotients are

Note that T has many more stability conditions than T. Denote

and the GIT quotient

Therefore, XT,0=XT,− and XT,r=XT,+.

The cohomology of XT,j is given by

where the ideal Ijab is given by

The 𝕋-fixed loci are indexed by points in ℙn−1. These are denoted by arbitrary functions

Note that f is not required to be injective. Thus we have a decomposition

The first abelianization in cohomology is

Next, note the diagram

gives us a rational map between the abelian and nonabelian quotients. This gives us a surjection

The kernel of this morphism are those f which are not injective.

2.3.3 Fourier-Mukai transform

Consider the mutual blowup

and let

Here,

Then the morphisms are

The fixed loci in X~ are indexed by δ+,δ−.

Lemma 2.3.3.

The projections f± map B(δ+,δ−) isomorphically to Bδ+ or Bδ−, respectively. Moreover,

Lemma 2.3.4.

The localized 𝕋-equivariant K-theory of X− is spanned by classes of the form

for A∈K⁡(B) and similarly for X+.

Lemma 2.3.5.

We have

Proposition 2.3.6.

The following diagram commutes:

Here, 𝕌H is an explicit Ŝ𝕋-linear operator given by the formula

where

The strategy to prove this result is first to replace cohomology by the symplectic space and replace the Chern character by the K-group framing. Then we use abelianization for I-functions [Web23] and repeatedly apply CIJ to each wall-crossing from the Xj. We then compose the symplectic transformations from CIJ and check that it equals the one given above. Finally, a statement about deforming the contour into the Weyl-invariant part gives the desired result.

3 Quantum cohomology of projective bundles

3.1 Mirror theorem (Che, Feb 22)

3.1.1 Setup

Let X be a smooth projective variety, {ϕi}i=0s be a basis of H∗⁢(X), {ϕi}i=0s be the dual basis, and

We will let

which is the J-function in Definition 1.1.3 multiplied by z−1.1. Also, recall the inverse of the fundamental solution of the quantum D-module

which is defined by

Remark 3.1.1.

By the string equation, we have

3.1.2 The vector bundle case

Now let V→B be a vector bundle with rk⁡V≥2. This has an action of ℂ× scaling the fibers. Then we have

Now we may take τ0,…,τs to be ℂ⁡[λ]-valued coordinates.

Remark 3.1.2.

Equivariant localization is required to define the Gromov-Witten invariants of V, which lie in ℂ⁡[λ,λ−1].

In order to avoid this issue, we will assume that V∨ is globally generated. This implies that V is semiprojective, meaning that the evaluation maps ev:V0,n,d→V are proper. As before, we may define the fundamental solution

and the J-function

Because the evaluation maps are proper, they can be defined without localization.

3.1.3 Statement and discussion of the mirror theorem

Theorem 3.1.3.

Define the H∗⁢(ℙ⁡(V))-valued function

where δ are the Chern roots of V, q is the Novikov variable, and p=c1⁢(𝒪ℙ⁡(V)⁢(1)). Then z⁢Iℙ⁡(V)⁢(τ,t) lies on the Lagrangian cone of ℙ⁡(V).

Let ℒXorig be the Lagrangian cone for X, which has the explicit form

Definition 3.1.4.

For a set of variables x=(x1,x2,…), we say that f∈ℋX⁢⟦x⟧ is a ℂ⟦Q,x⟧-valued point on ℒXorig if f is of the form 3.1 for some t⁡(z)∈ℋ+⁢⟦x⟧ with t⁡(z)|Q=x=0=0.

Example 3.1.5.

The point z⁢JX⁢(τ)|z↦−z is a ℂ⟦Q,τ⟧-valued point on ℒXorig.

Given this, define ℒX≔ℒXorig|z↦−z. By Theorem 1.1.10, we obtain

which means that any ℂ⟦Q,x⟧-valued point on ℒX can be written as z⁢MX⁢(τ)⁢f for some τ∈H∗(X)⟦Q,x⟧ and f∈ℋ+⁢⟦x⟧ such that τ|Q=x=0=0 and f|Q=x=0=1. This property will be used to construct the Fourier transform later.

3.1.4 Proof of Theorem 3.1.3

We will now sketch a proof of Theorem 3.1.3. First, we will need Quantum-Riemann-Roch for a vector bundle W→X in two cases:

  1. (a)

    When the vector bundle W is convex, which means that H1⁢(C,f∗⁢W)=0 for all stable maps f:C→X of genus 0, and 𝐜=e⁡(λ) is the equivariant Euler class, which corresponds to setting

  2. (b)

    When W is globally generated and 𝐜=eλ−1.

In the first case, we obtain the Gromov-Witten invariants of the zeroes of a regular section Z⊂X of W via

In the second case, we obtain the Gromov-Witten invariants of W via

We are now ready to begin the proof. Because V∨ is globally generated, there is a surjection

This gives an exact sequence

embedding ℙ⁡(V)↪B×ℙN−1. By a result of Brown-Elezi, we have

Now define

on B×ℙ1. This has a section s given by

which satisfies s−1⁢(0)=ℙ⁡(V). Because Q⁡(1) is convex, we use Quantum-Riemann-Roch in case (a) to relate the Gromov-Witten theory of ℙ⁡(V) to the (Q⁡(1),eλ)-twisted Gromov-Witten theory. We now require two more techinical ingredients.

Moving points on the Lagrangian cone via differential operators
Lemma 3.1.6.

Let x=(x1,x2,…) and y=(y1,y2,…) be formal variables. Let

be a differential operator. Then exp⁡(F/z) preserves ℂ⟦Q,x,y⟧-valued points on ℒX.

Definition 3.1.7.

A ℂ⟦Q,τ,y⟧-valued point f on ℒX is called a miniversal slice if

For example, the J-function is a miniversal slice.

Lemma 3.1.8.

Any miniversal slice on ℒX can be obtained from z⁢JX⁢(τ) be applying exp⁡(F/z) for some differential operator F as in the previous lemma satisfying F|Q=y=0=0.

The rest of the proof (ignoring convergence issues)

First, we introduce

Because log⁡ΔWλ and log⁡Γ⁡(x) have similar asymptotic expansions, we have

Using the exact sequence

we see that

which preserves the Lagrangian cone ℒB. We see that

Applying Quantum-Riemann-Roch in case (b), we see that

and thus

By Lemma 3.1.8, there exists F such that

By Lemma 3.1.6, we obtain

Now we compute

where ε runs over the Chern roots of Q. Taking the non-equivariant limit λ→0, we obtain the I⁡(τ,t) in the statement of Theorem 3.1.3.

3.2 Fourier transform (Kostya, Feb 29)

Technically, there are two different Fourier transforms:

  1. 1.

    The discrete Fourier transform QDMS1⁡(V)→FTQDM⁡(ℙ⁡(V));

  2. 2.

    The continuous Fourier transform QDM⁡(ℙ⁡(V))loc→⨁i=0r−1QDM⁡(B).

3.2.1 Quantum D-modules and symplectic spaces

Let V→B be a rank r vector bundle. The quantum D-module of V will be

where λ is the equivariant variable, Q is the Novikov variable, and 𝝉={τi,k} for i counting a basis of H∗⁢(V) and k records the degree of λ. It is equipped with the Dubrovin connection

given by

Note the last line is only ℂ-linear. Recall the fundamental solution MV−1⁢(𝝉,z) is a fundamental solution in the cohomology directions (but not the conformal direction) in the sense that

and intertwines the shift operator by

where

Now, the symplectic space for V is

with its Lagrangian cone ℒV. It has the important property that f⁡(𝝉) is in ℒV means that there exists τ̂⁢(𝝉) and f~∈QDMS1⁡(V) such that

which can be seen as a Birkhoff factorization.

We now turn to ℙ⁡(V). There is a decomposition

where δ runs over the Chern roots of V. This receives the Kirwan map

Thus the quantum D-module for ℙ⁡(V) is

where 𝝉̂={τ̂i} is a basis for H∗⁢(ℙ⁡(V)) and q is the Novikov variable of the fiber curve class. It is equipped with the connections ∇τ̂i,∇ξ⁢Q⁢∂Q,∇q⁢∂q,∇z⁢∂z.

Remark 3.2.1.

Note there are no shift operators, but there is an additional q-direction in the quantum D-module for ℙ⁡(V)

We also have the symplectic space ℋℙ⁡(V), the Lagrangian cone ℒℙ⁡(V), and the fundamental solution Mℙ⁡(V)⁢(𝝉̂,z). Finally, we will recall the mirror theorem in the form that

lies on ℒℙ⁡(V).

3.2.2 Discrete Fourier transform

Definition 3.2.2.

The discrete Fourier transform ℋV→ℋℙ⁡(V) is the transform

In this framing, the mirror theorem states that the discrete Fourier transform of the J-function of V lies on the Lagrangian cone of ℙ⁡(V).

Theorem 3.2.3.

There exists a “mirror map”

and an isomorphism

of ℂ[z]⟦Q,𝛕⟧-modules intertwining the connections in the natural ways.

Remark 3.2.4.

One has to be careful with the Novikov variables and think about approximately eight other points of the theorem, but we will ignore these for now.

Because the Fourier transform intertwines the connections, we have the commutative diagram

Idea of proof.

The idea of the proof is to start from the mirror theorem (the bottom row) and apply Birkhoff factorization. The mirror theorem states that

for some mirror map 𝝉̂⁢(𝝉) and Υ∈QDM⁡(ℙ⁡(V)). Using the intertwining properties of M, we see that

Defining

and 𝝉̂ to be the mirror map appearing in the Birkhoff factorization, we are done. □

Remark 3.2.5.

The mirror map satisfies

and the Fourier transform satisfies

Remark 3.2.6.

The Fourier transform intertwines the natural pairings on the quantum D-modules.

3.2.3 Continuous Fourier transform

Definition 3.2.7.

Define

where r′=r⁢ or ⁢2⁢r depending on parity.

Theorem 3.2.8.

For j=0,…,r−1, there exist maps H∗⁢(ℙ⁡(V))→H∗⁢(B) given by

and an isomorphism

intertwining the pairings and quantum connections in a natural way, namely that

Writing Φ=(Φ0,…,Φj), we have

Idea of proof.

We use another realization of the Fourier transform on QDMS1⁡(V) and

If we consider ΔVλ arising from Quantum Riemann-Roch, it is given as

Shifting by −z, we see that

We now consider the transformation

for s∈QDMS1⁡(V). Because this integral intertwines 𝕊 with q and λ with z⁢∇q⁢∂q, it formally gives a solution to QDM⁡(ℙ⁡(V)).

To make sense of this terrible expression, we use the stationary phase expansion of the integral. Setting

where φ⁡(λ)z is the Stirling asymptotics of ΔVλ, given by

The critical points of φ are given by

which tells us that λr=q. Thus, we obtain r solutions

We now consider the formal expansions around λj. These produce a “continuous Fourier transform”

such that

These intertwine the quantum connection and multiplication by λ, as in

so z⁢ℱj⁢(JV⁢(𝝉)) is on the Lagrangian cone of B. We then use the following result:

Proposition 3.2.9.

We have

for some σj,νj.

Unfortunately, ℱj does not intertwine ∇z⁢∂z correctly. To fix this, define

and Φj by a shift of νj. □

3.2.4 Discrete equals continuous

Warning 3.2.10.

Everything in this subsection may be false.

Consider the Fourier transform

This can be computed either using residues or using stationary phase asymptotics. Using residues, we obtain

which is precisely

Using stationary phase asymptotics, we obtain the I⁡(s) defined previously.

4 Quantum cohomology of blowups

4.1 Setup (Kostya, Mar 07)

Let X be a smooth projective variety, Z⊂X be a smooth closed subvariety of codimension r, and

be the blowup of X with center Z. Denote the exceptional divisor by

As a vector space, there is an isomorphicm

Our goal is to upgrade this to the level of quantum cohomology. However, this is very tricky because of convergence issues, so it will be a corollary of a decomposition theorem for quantum D-modules.

Recall that the quantum D-module of X is given by

with the Dubrovin connection

and pairing

4.1.1 Statement of the theorem

The first problem we need to fix is that the cohomological and Novikov variables of X,Z,X~ are different.

Definition 4.1.1.

The extended Novikov ring is defined as

where d∈NEℕ⁡(X) and d~∈NEℕ⁡(X~).

Introducing another formal variable q, we can embed the Novikov rings of Z,X,X~ into ℂ⁡⦇q−1s⦈⁢⟦𝒬⟧, where s is either r−1 or 2⁢(r−1) depending on the parity of r (we want s to be even):

  • •

    ℂ⁢⟦Q⟧ embeds as Qd↦Qd;

  • •

    ℂ⁢⟦Q~⟧ embeds as Q~d~↦Qφ∗⁢d~⁢q−[D]⋅d~;

  • •

    ℂ⁢⟦QZ⟧ embeds as QZd↦Qι∗⁢d⁢q−ρz⋅dr−1, where ρZ=c1⁢(NZ/X).

Later, we will see that q=y⁢S−1, where S−1 will be an equivariant variable for a ℂ∗ action. Denote

We can now state the main result.

Theorem 4.1.2.

There exists a formal invertible change of variables

denoted by

and an isomorphism

such that Ψ intertwines the quantum connections and the pairings.

4.1.2 The master space

Definition 4.1.3.

Define the master space

to be the degeneration to the normal cone of Z⊂X. This is endowed with a T=ℂ×-action extending the action

for (x,u)∈X×ℙ1. See Figure 4.1 for a picture of W.

X~X=X×{∞}DD̂ZZ×ℙ1tϵ0μ
Figure 4.1: W=BlZ×{0}⁡(X×ℙ1) and a moment map μ:W→ℝ

The fixed locus of the ℂ∗-action is given by

We can also see that

Note here that

so a general class in N1T⁢(W) can be written as

The dual notion is the group of 1-cycles, which is given by

whose Novikov variable is

Note that

The effective curve classes are those which are equivariant, so they either lie in the fixed loci or are 1-dimensional orbits. There are classes C1 lying entirely inside X, classes C2, which are x×ℙ1 for x∉Z, classes C3⊂Z×ℙ1, and classes C4⊂D̂. The corresponding Novikov variables are Qd,x,x⁢y−1, and y, respectively.

Lemma 4.1.4.

The monoid of effective curve classes is generated by C1,C2,C3,C4,S.

Example 4.1.5.

Let X=ℙ1×ℙ1 and Z=(0,0). Then a toric diagram for W with C1,C2,C3,C4 is given in Figure 4.2.

X~XZD⊂D̂C3⊂Z×ℙ1C1C2C4
Figure 4.2: Toric diagram of W for X=ℙ1×ℙ1 and Z=(0,0) with curve classes C1,C2,C3,C4.

The T-ample cone CT⁢(W) of W is the dual to the cone of effective curve classes. Then ω̂∈CT⁢(W) if the set of ω̂-stable points under the ℂ∗-action is nonempty. Then there is a decomposition

into pieces where the GIT quotient

respectively. The stable points are

  • •

    For ω̂∈CX, the stable points are X×ℂ∗;

  • •

    For ω̂∈CX~, the stable points are 𝒪X~⁢(−D̂)∖X̂.

The dual cones CX∨ and CX~∨ correspond to embeddings of the effective cones of X and X~, respectively, into the effective cone of W as in Figure 4.3.

NE⁡(W)y⁢S−1y−1⁢Sx⁢S−1SX~X
Figure 4.3: A schematic picture of the cones CX∨ and CX~∨ in N1T⁢(W).

Recall the Kirwan map

where Y is either X or X~.

Now consider the equivariant quantum D-module

endowed with the quantum connection and the action of the shift operators

which are defined by

where β is identified with (d,k,ℓ,m) under the factorization N1⁢(X)⊕ℤ3.

We can obtain the quantum D-modules for X and X~ from the equivariant quantum D-module of W by taking affine charts and completing. Here, we think of QDMT⁡(W) as a global Kähler moduli space.

qx⁢y−1y⁢q−1q−1x⁢y−1⁢qy𝔐X~𝔐X
Figure 4.4: Global Kähler moduli space associated with QDMT⁡(W). See also Figure 4.3.

4.1.3 Fourier transforms

To construct the map in Theorem 4.1.2, Iritani uses both the discrete and continuous Fourier transforms. For f∈ℋW, the discrete Fourier transform is given by

The continuous Fourier transform for Z∈π0⁢(WT) is given by sending f∈ℋW to the element of ℋZ given by the formal asymptotic expansion of

Remark 4.1.6.

We can think of X and X~ either as fixed components X,X~∈π0⁢(WT) (corresponding to the continuous Fourier transform) or as GIT quotients W⫽ω̂T, corresponding to the discrete Fourier transform. These are in fact equal up to a factor.

Note that Z is not a GIT quotient of W, but is a fixed component, so there is a continuous Fourier transform to W. In total, we have three morphisms

4.2 Fourier analysis for blowups (Sam, Mar 21)

In this section, denote the Chern roots of 𝒩Z/X by ε1,…,εr.

4.2.1 Extended Kirwan maps

We will begin by describing the Kirwan maps in coordinates. On HT2⁢(W), we have

for Y=X

for Y=X~. The dual Kirwan maps on N1⁢(Y) are given by

when Y=X and Y=X~, respectively. It may appear that we don’t see the equivariant parameters in the dual Kirwan map, but we will fix this.

Definition 4.2.1.

The extended Givental space is

which is a base change of ℋY.

The shift operators on ℋWrat=Hloc∗⁢(W)⁢[Z±]⁢⟦Q⟧ are now given by

where

4.2.2 Discrete Fourier transform

We are now able to make the following definition.

Definition 4.2.2.

The discrete Fourier transform is

defined by

Because we invert the equivariant parameters, this may not necessarily be well-defined. However, we do have the following result.

Proposition 4.2.3.

The discrete Fourier transform FY is well-defined on tangent spaces to ℒW.

To prove this result, we need various ingredients:

  1. 1.

    Some regularity at λ=0, which follows from properness;

  2. 2.

    Not needing arbitrarily high powers of Q−1 (landing in the target).

As in the case of projective bundles, the shift operators satisfy various nice properties. For example, because

we obtain the formulae

for any ξ∈H2T⁢(W).

4.2.3 Continuous Fourier transform

Let F∈{X,X~,Z} be a fixed component in WT. Denote the Chern roots of 𝒩F/W by ρ1,…,ρn. Now define

More specifically, for each individual fixed component, we have

This GF actually intertwines all 𝒮̂β, which are defined for all

Given F, we have

which are obtained by section classes on Ek as in Figure 4.5. Now GF satisfies the equation

which follows from the same argument as in the projective bundle case.

σk⁢(Z)σk⁢(X~)σk⁢(X)
Figure 4.5: Toric diagram of Ek when X,Z are as in Figure 4.2.

We are now able to describe the continuous Fourier transform formally as an integral which has the same properties as the discrete one. Define

where SF=σ1⁢(F). Formally, we see that

which suggests that this is the right object to study the asymptotics of. Recall the Stirling approximation

where we have the decomposition

ρα=c1⁢(𝒩α) is the first Chern class, rα=rk⁡𝒩α, and Δα is the quantum Riemann-Roch operator. We now obtain

We have a critical point

where

Also, for j=0,…,|cF|−1, we have critical points

Definition 4.2.4.

The continuous Fourier transform is given by the asymptotic expansion

where we take the asymptotics as z→0 and substitute

and expand in u-powers as u→0.

Via taking a formal asymptotic analysis of what we did before, we have

Proposition 4.2.5.
  1. 1.

    The continuous Fourier transform intertwines the shift operators as

  2. 2.

    The continuous Fourier transform intertwines the equivariant parameter via

  3. 3.

    There is a similar property for the Euler vector field.

Now let

Proposition 4.2.6.

There exists

and

such that

In other words, we can recover QDMF from ℱF,j after a change of basis and an étale cover of the Kähler moduli space.

Proposition 4.2.7.

The discrete and continuous Fourier transforms agree for Y∈{X,X~}. In other words, we have

4.3 Decomposition (Sam, Mar 28)

4.3.1 A general picture

We begin by stating some general conjectures that the blowup result fits into.

Conjecture 4.3.1 (Dubrovin).

Let X be smooth and projective. If there exists a semiorthogonal decomposition

then there is a decomposition

More specifically, if X admits a full exceptional collection, QDM⁡(X) can be extended to 𝒪an⁢⟦z⟧. In addition the Gram matrix of the pairing χ⁡(−,−) on K⁡(X) is recovered as the Stokes matrix

of flat sections of the irregular connection

On the level of derived categories, we have a decomposition

so we expect a decomposition

In the limit Q⁢τ→0, we obtain QDM⁡(X) at μ=0 and r−1 copies of QDM⁡(Z) at roots of unity. The shift away from μ=0 will correspond to the shift of saddle points in the Fourier transform.

In the equivariant setting, we have semiorthogonal decompositions

We then expect

Conjecture 4.3.2.

Setting W⫽θT≕Y, then

lies on the Lagrangian cone of Y.

4.3.2 Decomposition of the quantum D-module

First, we would like to give a more precise formulae for the quantities appearing in Proposition 4.2.6. First, τ is given by

where

Then we have

where

for λj=e1⁢π⁢i⁢jcF⁢λ0.

For X,X~, the continuous Fourier transform ℱF,j intertwines Sβ and λ with ξ⁢S⁢∂∂S and EWT⋆τ+μW with EX~⋆τ+μX+12. On the other hand, for F=Z, the critical points are different, so there is a shift in λ. For example, we have

because the integral satisfies the formula

A similar argument yields

We next need a space on which to compare QDM⁡(X), QDM⁡(X~), and QDMT⁡(W) via the dual Kirwan maps. Define

and extend ∇ trivially on CY,ℕ∨. As in Figure 4.4, we have 𝔐X~ and 𝔐X, but we also need

The three charts can be compared on the chart

The comparison of the effective cones of curves is given in Figure 4.3.

The comparison also requires a completion and localization of QDMT⁡(W). In order to compare with QDM⁡(X~) we need the action of extended shift operators 𝕊β. Define

The completion is given by

Now let

such that

By taking derivatives ∂θi⁢k, the Fourier transform is lifted to a map of quantum D-modules.

Theorem 4.3.3.

There is an isomorphism

and a projection

They intertwine the quantum connection and the pairing up to a 12 shift in μ.

These extend to the completions, but we will not prove this here.

For F=Z and any j, there are coordinates

such that

Theorem 4.3.4.

There are projections

which intertwine λ with z⁢∇S⁢∂S+λj and z⁢∇ξ⁢Q⁢∂Z with z⁢∇Ŝ⁢∂Ŝ+ιpt∗⁢ξ|λ=λi. Here, QDM⁡(Z)ext,loc is defined by the extension

We will now shift our variables in order to make our comparison. Let

Then

intertwines the quantum connections up to a 12 shift in μZ. Combining the Fourier transforms for the different fixed loci, we obtain

Theorem 4.3.5 (Main theorem).

The diagram

induces an isomorphism

over 𝔘 that intertwines the quantum connections and Poincaré pairings.

Bibliography

  • [AGV08]
    Dan Abramovich, Tom Graber, and Angelo Vistoli. “Gromov-Witten theory of Deligne-Mumford stacks”. In: Amer. J. Math. 130.5 (2008), pp. 1337–1398. issn: 0002-9327,1080-6377. doi: 10.1353/ajm.0.0017.
  • [BCS05]
    Lev A. Borisov, Linda Chen, and Gregory G. Smith. “The orbifold Chow ring of toric Deligne-Mumford stacks”. In: J. Amer. Math. Soc. 18.1 (2005), pp. 193–215. issn: 0894-0347,1088-6834. doi: 10.1090/S0894-0347-04-00471-0.
  • [CG07]
    Tom Coates and Alexander Givental. “Quantum Riemann-Roch, Lefschetz and Serre”. In: Ann. of Math. (2) 165.1 (2007), pp. 15–53. issn: 0003-486X,1939-8980. doi: 10.4007/annals.2007.165.15.
  • [CIJ18]
    Tom Coates, Hiroshi Iritani, and Yunfeng Jiang. “The crepant transformation conjecture for toric complete intersections”. In: Adv. Math. 329 (2018), pp. 1002–1087. issn: 0001-8708,1090-2082. doi: 10.1016/j.aim.2017.11.017.
  • [Coa03]
    Tom Coates. “Riemann-Roch theorem in Gromov-Witten theory”. PhD thesis. University of California at Berkeley, 2003.
  • [Dub96]
    Boris Dubrovin. “Geometry of 2D topological field theories”. In: Integrable systems and quantum groups (Montecatini Terme, 1993). Vol. 1620. Lecture Notes in Math. Springer, Berlin, 1996, pp. 120–348. isbn: 3-540-60542-8. doi: 10.1007/BFb0094793.
  • [Giv04]
    Alexander B. Givental. “Symplectic geometry of Frobenius structures”. In: Frobenius manifolds. Vol. E36. Aspects Math. Friedr. Vieweg, Wiesbaden, 2004, pp. 91–112. isbn: 3-528-03206-5.
  • [Giv98]
    Alexander Givental. “A mirror theorem for toric complete intersections”. In: Topological field theory, primitive forms and related topics (Kyoto, 1996). Vol. 160. Progr. Math. Birkhäuser Boston, Boston, MA, 1998, pp. 141–175. isbn: 0-8176-3975-6.
  • [IK24]
    Hiroshi Iritani and Yuki Koto. Quantum cohomology of projective bundles. 2024. arXiv: 2307.03696 [math.AG].
  • [Iri17]
    Hiroshi Iritani. “Shift operators and toric mirror theorem”. In: Geom. Topol. 21.1 (2017), pp. 315–343. issn: 1465-3060,1364-0380. doi: 10.2140/gt.2017.21.315.
  • [Iri23]
    Hiroshi Iritani. Quantum cohomology of blowups. 2023. arXiv: 2307.13555 [math.AG].
  • [Jia08]
    Yunfeng Jiang. “The orbifold cohomology ring of simplicial toric stack bundles”. In: Illinois J. Math. 52.2 (2008), pp. 493–514. issn: 0019-2082,1945-6581.
  • [LSW24]
    Wendelin Lutz, Qaasim Shafi, and Rachel Webb. Crepant Transformation Conjecture For the Grassmannian Flop. 2024. arXiv: 2404.12302 [math.AG].
  • [PSW24]
    Nathan Priddis, Mark Shoemaker, and Yaoxiong Wen. Wall crossing and the Fourier-Mukai transform for Grassmann flops. 2024. arXiv: 2404.12303 [math.AG].
  • [Tse10]
    Hsian-Hua Tseng. “Orbifold quantum Riemann-Roch, Lefschetz and Serre”. In: Geom. Topol. 14.1 (2010), pp. 1–81. issn: 1465-3060,1364-0380. doi: 10.2140/gt.2010.14.1.
  • [Web23]
    Rachel Webb. “The abelian-nonabelian correspondence for I-functions”. In: Int. Math. Res. Not. IMRN 3 (2023), pp. 2592–2648. issn: 1073-7928,1687-0247. doi: 10.1093/imrn/rnab305.