Patrick Lei

All notes

Informal enumerative geometry seminar

Columbia, Spring 2022.

Informal Enumerative Geometry Seminar
Spring 2022
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

These notes were taken during the seminar using the vimtex package of the editor 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. If you find any errors, please contact me at plei@math.columbia.edu.

1 Melissa (Feb 03): Virtual fundamental classes in algebraic geometry

1.1 Overview

Virtual fundamental classes are well-motivated for us, so we will not motivate them. Let X be a Deligne-Mumford stack, which is virtually smooth (which means it has a perfect obstruction theory). A perfect obstruction theory is a class E=[E−1→E0] in D⁡(X) which is locally isomorphic to a morphism of vector bundles.

In Li-Tian, the authors consider a perfect tangent-obstruction complex

where 𝒯1 is the tangent sheaf and 𝒯2 is the obstruction sheaf. If we consider

we note that the virtual tangent bundle has K-theory class

Thus E has virtual dimension d≔rk⁡E0−rk⁡E−1, and from the data (X,E), we will produce a virtual fundamental class [X,E]vir∈Ad⁢(X,ℚ). If X is a scheme (for example in Donaldson-Thomas theory), then [X,E]vir∈Ad⁢(X,ℤ). If X is smooth, then E=[0→ΩX] and Tvir=TX. Of course, we should recover the usual fundamental class in this case.

1.2 Cones

Recall the Proj construction from Hartshorne, II.7. Let X be a scheme and 𝒮 be a sheaf of graded 𝒪X-algebras generated over 𝒮0=𝒪X by a coherent 𝒮1. Then define C⁡(𝒮)=Spec⁡(𝒮)→X, which is an affine scheme over X. Because Sym∙⁡(𝒮1)↠𝒮, then C⁡(𝒮)↪C⁡(Sym∙⁡(S1)) is a closed embedding.

Example 1.2.1.

Consider X=Spec⁡k and 𝒮=k⁡[x,y,z]/(x⁢z−y2). Of corse Proj⁡𝒮 is a conic in ℙ2, and Spec⁡𝒮 is 𝔸2/(±1), which is a cone. Note that Sym⁡𝒮1=k⁡[x,y,z], so C⁡(Sym⁡(𝒮1))=𝔸3.

Of course, by Hartshorne, Ex. II.5.18, if ℰ is locally free, then Spec⁡Sym⁡(E) is a vector bundle. The analogous notion for a coherent sheaf ℱ is an abelian cone. Note that the Stacks Project calls abelian cones vector bundles for some reason.

1.3 Local theory on a scheme

Recall that if X is a smooth manifold, then it has an atlas Uα of open sets in ℝn. Let X be a scheme. Then X has a cover by affines Uα, and we have open embeddings Uα=Spec⁡Rα→X. Finally, consider a Deligne-Mumford stack X. Then we have an étale atlas Uα=Spec⁡Rα. If X is an Artin stack, then replace étale with smooth.

Now let D⁡(X) denote the derived category of quasicoherent sheaves of 𝒪X-modules. Then let D[−1,0] denote the full subcategory on complexes with vanishing cohomology outside of [−1,0] with h0 and h1 coherent.

Definition 1.3.1.

A perfect obstruction theory on X is a complex E∈D[−1,0]⁢(X) with ϕ:E→LX, where LX is the cotangent complex such that ϕ induces an isomorphism on h0 and a surjection on h−1.

To understand this, assume that U is an affine scheme with an étale morphism to X. Then ϕ:E→LX pulls back to ϕU:EU→LU. If we consider a closed immersion i:U↪W of U into a smooth scheme, we can now do intersection theory a la Fulton. The normal cone CU/W is defined to be

This is a closed subcone of NU/W=Spec⁡Sym⁡ℐ/ℐ2. In this case, we only need to consider the truncated cotangent complex

In fact, if U→W is a local complete intersection, then ℐ/ℐ2 is locally free and LU=[ℐ/ℐ2→i∗⁢ΩW]. Thus LU∨=[i∗⁢TW→𝒩U/W]. If this is surjective, then U is smooth, so we have an exact sequence

and therefore [ℐ/ℐ2→i∗⁢ΩW]=[0→ΩU]=LU. Locally, we have EU=[E−1→E0] a morphism of locally free sheaves, and a morphism ϕ:E→τ≥−1⁢LU.

Proposition 1.3.2.

Up to quasi-isomorphism, we may assume that EU=[E−1→E0] and τ≥−1⁢LU=[L−1→L0=E0] with ϕ0=id. Given these assumptions, E−1→L−1 is surjective.

Proof.

Locally, suppose that U=Spec⁡R. Then let E−1,E0,F−1,F0 be finitely-generated R-modules such that E0,F0 are free. Then consider ϕ:E→F. Next, choose a surjection π:G↠F−1 with G a finitely-generated free module. Then define

which is clearly quasi-isomorphic to E. Then define ϕ~ by ϕ~−1=ϕ−1+π and ϕ~0=ϕ0+dF∘π. Now ϕ~0 is surjective.

Next, we may asume that E=[E−1→E0] such that p⁢h⁢i0 is surjective. We will now replace F by the top row of the cartesian diagram

Because this is a fiber product, we have a map E→F~ that is the identity in the 0-th position. □

Now we return to an étale chart U→X, where X is a Deligne-Mumford stack. Applying Spec⁡Sym⁡(−) to E, we obtain a morphism

Then we define

which is an Artin stack. We also apply this to the intrinsic normal sheaf

Finally, the intrinsic normal cone is defined to be

which is a cone stack over U. We have a closed embedding

of C¯U into an abelian cone stack. Finally, these embed into E¯U=[E1/E0]. Because CU/W has pure dimension dimW, we see that C¯U has dimension 0.

1.4 Virtual fundamental classes

We have intrinsic normal cones locally, and fortunately, they glue! Therefore, we have an intrinsic normal cone C¯X which is a cone stack over X of pure dimension 0, an intrinsic normal sheaf 𝒩¯X which is an abelian cone stack over X, and ℰ¯X, which is a vector bundle stack over X smooth of relative dimension d1−d0=−d.

Recall from Fulton that if E→X is a vector bundle over a scheme of rank r, there is a Gysin map 0E!:Ad(E)→Ad−r(X). We want to do something similar in the case of a vector bundle stack. In Cycles groups for Artin stacks, Kresch gives the following: Let ℰ→X be a vector bundle stack of rank r over a Deligne-Mumford stack. Then there is a Gysin map 0ℰ!:Ad(ℰ¯)→Ad−r. Applying this to our situation, the virtual fundamental class is given by

1.5 Relative version

Let X be a Deligne-Mumford stack and consider πX/M:X→M, where M is a smooth Artin stack of pure dimension m. If πX/M is virtually smooth, then there exists a relative perfect obstruction theory

where LX/M is the relative cotangent complex. We again require that ϕ induces an isomorphism on h0 and a surjection on h−1.

If U is an affine scheme with U→X étale, then U→M is smooth. Also let i:U↪W be a closed embedding of U into a smooth scheme and suppose W→M is smooth of relative domension d0. Then we have the relative intrinsic normal cone

of pure dimension m, the relative intrinsic normal sheaf

and the vector bundle stack ℰ¯U/M. These also glue nicely, and we obtain

where ℰ¯X/M is a vector bundle stack of rank −d=d1−d0. Then [C¯X/M]∈Am⁢(ℰ¯X/M), and then the relative virtual fundamental class is

2 Melissa (Feb 10): Virtual fundamental classes in enumerative geometry

Recall our abstract setup of a Deligne-Mumford stack X and a perfect obstruction theory E. In this setting, Behrend-Fantechi construct a virtual fundamental class [X,E]vir∈Ad⁢(X), where d is the virtual dimension of X. There is an equivariant version of this for the action of a reductive group G on a Deligne-Mumford stack 𝒳. Then we will have L𝒳∈DG⁢(𝒳), so we can consider equivariant perfect obstruction theories E. Here, we will have [𝒳,E]vir∈AdG⁢(𝒳).

For us, we will normally consider T=(ℂ∗)k. Here, there is a fixed substack 𝒳T. If we consider

we know that 𝒯vir=𝒯1−𝒯2=(E0)∨−(E−1)∨. Restricting to the fixed locus, we can decompose the tangent sheaf

into fixed and moving parts, and similarly for the obstruction sheaf 𝒯2. Then we have

and the moving part is

Graber-Pandharipande proved torus localization for virtual fundamental classes, and in particular, we obtain

Because A∗⁢(B⁢ℂ∗)=ℚ⁡[t], we are considering [𝒳,E]vir∈Adℂ∗⁢(𝒳)⊗ℚ⁡[t]ℚ⁡[t,t−1]. Note that if ℂ∗ acts trivially on 𝒴, then Adℂ∗⁢(𝒴)=Ad⁢(𝒴)⊗ℚℚ⁡[t].

2.1 Gromov-Witten theory

Let X be a smooth projective variety and β∈H2⁢(X,ℤ) be an effective curve class. Let 𝒳~=ℳg,npre⁢(X,β) be the moduli space of genus g, n-pointed stable maps to X of degree β. The ℂ-points of this functor are ((C,x1,…,xn),f), where C is a genus g, n-pointed prestable curve and f:C→X has f∗⁢[C]=β.

Taking the locus with finite automorphism groups, we obtain a Deligne-Mumford stack 𝒳=ℳ¯g,n⁢(X,β)⊂𝒳~, which is an Artin stack. 𝒳 is a proper Deligne-Mumford stack with projective coarse moduli space. Recall that there is a map π𝒳~/ℳ:𝒳~→ℳg,npre, which is a smooth Artin stack of dimension 3⁢g−3+n, forgetting the map f. We would like to construct a relative perfect obstruction theory.

Consider the diagram

with f:C→X. Then we have h0⁢(E𝒳~/ℳ∨)ξ=H0⁢(C,f∗⁢TX) and h1⁢(E𝒳~/ℳ∨)ξ=H1⁢(C,f∗⁢TX). Globally, we obtain

Then we obtain

and therefore we obtain

Our stability condition guarantees that any component contracted by f is already stable as a curve. Now let T be a scheme of finite type with a smooth map to 𝒳. Then we have the following diagram:

with fT:CT→X. We know that πT is flat and projective over T, so define the line bundle

By the stability condition, ℒ is πT-ample because ωπT⁢(x1+⋯+xn) has degree 2⁢gv−2+nv≥−2 on any irreducible component Cv⊂C. Also, fT∗⁢TX is a locally free sheaf on CT, so there exists N≫0 such that fT∗⁢TX⊗ℒN is globally generated and H1⁢(Ct,ft∗⁢TX⊗ℒtN)=0. In addition, π∗⁢(ℒ−N)=0.

Now consider the exact sequence

and use the long exact sequence in cohomology, and thus E∨≃[𝒯1→0𝒯2].

Example 2.1.1.

Note that ℳ¯g,n⁢(X,0)=ℳ¯g,n×X. Here, given ((C,z1,…,zn)⁢f), we must have f⁡(C)=x, for some x∈X. Then Hi⁢(C,f∗⁢TX)=Hi⁢(C,𝒪C)⊗Tx⁢X, and so we obtain

Note that L𝒳/ℳ=[0→pr2∗ΩX. Thus 𝒯1=𝒯𝒳 and 𝒯2=p⁢r2∗⁢TX⊗p⁢r1∗⁢E∨, where E=(πℳ)∗⁢ωπℳ. Finally, we see that

2.2 Remarks on equivariant localization

Let G=ℂ∗ act on a stack 𝒳. Then to do localization, we need to check that

  1. 1.

    There exists a G-equivariant embedding 𝒳↪𝒴 of 𝒳 into a smooth Deligne-Mumford stack.

  2. 2.

    There exists a global locally free resolution E=[E−1→E−2].

Thus doing equivariant enumerative geometry required constructing a G-equivariant embedding of 𝒳 into a smooth stack. Then a result of Chang-Kiem-Li in 2017 allows us to remove the first condition and weaken the second (we only nneed a global resolution on the fixed locus).

3 Song (Feb 17): Crepant transformation conjecture for toric Deligne-Mumford stacks, part 1

The crepant transformation conjecture roughly says:

Conjecture 3.0.1 (Ruan, etc).

For a pair of K-equivalent manifolds/orbifolds/Deligne-Mumford stacks X±, their quantum cohomology/Gromov-Witten theory should be related.

Here, by K-equivalent we mean (following C-L Wang) that there exists a diagram

where φ is birational, f± are projective and birational, and f+∗⁢KX+=f−∗⁢KX−. Coates-Iritani-Jiang formulated and proved the crepant transformation conjecture for toric Deligne-Mumford stacks and complete intersections.

3.1 Toric geometry

Note that toric crepant transformations correspond to wall-crossings of GIT stability conditions. There are three types of this.

Example 3.1.1 (Crepant (partial) resolution).

Here, we have a morphism X+→|X−| that partially resolves singularities of X−. In our example, we will consider X+=Kℙ2⇢[ℂ3/μ3]=X−. Then the toric fan of Kℙ2 is generated by b1=(1,0,1),b2=(0,1,1),b3=(−1,−1,1),b4=(0,0,1). The toric fan of [ℂ3/μ3] is generated by b1,b2,b3, and there is a singularity at b4, so to get Kℙ2 we subdivide the triangle.

The GIT data for Kℙ2 is given by 0→L→ℤ4→(b1,b2,b3,b4)ℤ3→0. Dualizing, we have a surjection (ℤ4)∨→(1,1,1,−3)L∨→0. In Lℝ∨, we see that there is a positive chamber and a negative chamber, and in the positive chamber we obtain Kℙ2 and in the negative chamber we obtain [ℂ3/μ3].

Example 3.1.2 (Flop).

Consider X±=𝒪ℙ1⁢(−1)⊕𝒪ℙ1⁢(−1). These are related by an Atiyah flop. The flop corresponds to exchanging the two diagonals of a square. In this case, we still have a 1-dimensional secondary fan with two chambers.

Example 3.1.3 (Gerbe flop).

Consider X+=ℙ⁡(2,2) and X−=ℙ1×B⁢μ2. Note that both rigidified stacks are isomorphic to ℙ1. The GIT data is given by ℤ3→ℤ×ℤ2→0, where b1=(1,1),b2=(−1,1),b3=(1,0). If we choose b1,b2 we obtain X− and if we choose b2,b3 we obtain X+. The secondary fan has D1=(1,0),D2=(1,2),D3=(0,2), where the bottom chamber corresponds to X+ and the top chamber corresponds to X−.

Returning to the crepant resolution, we can tensor the lattice data to obtain

We will let Q≔(ℂ∗)3 be the dense open torus in X±. Thus T≔(ℂ∗)4 also acts on X±. We will work T-equivariantly.

Now note that I⁢X−=X−⊔B⁢μ3⊔B⁢μ3. The two copies of B⁢μ3 are called the twisted sectors and will be labeled by 13,23. The untwisted sector will be labeled by 0. Then we take Chen1-Ruan cohomology and obtain

where the twisted sectors live in degrees 2,4 respectively. On the other hand, X+ is a scheme, so IX+=X+ and

Note that Lℚ∨≅HC⁢R2⁢(X±). We will call its rank r. We can also define T,Q-equivaraint versions of everything and there exists a T-equivariant orbifold Poincaré pairing

Here, inv:I⁢X→I⁢X takes an automorphism to its inverse.

3.2 Quantum cohomology/Gromov-Witten theory

First, we will define (T-equivariant) Gromov-Witten invariants. Choose g,n∈ℤ≥0, d∈H2⁢(|X|,ℤ), α1,…,αn∈HC⁢R,T∗⁢(X), and k1,…,kn∈ℤ≥0 defining the descendent insertions. Then we define

Note that the moduli spaces are not compact, so we need to use T-equivariant localization.

With the Gromov-Witten invariants, we can now define the quantum product. Let ϕ0=1, ϕ1,…,ϕr be a basis for H2, and ϕr+1,…,ϕN be the completion to a homogeneous basis for H∗. Let τ0,…,τN be the corresponding coordinates. For τ=∑τi⁢ϕi, define

This is a commutative and associative product which has a nonequivariant limit. The reason for this is because the evaluation maps are proper (a consequence of X being semiprojective).

The next thing to define are the J-functions. Again write τ=∑i=1rτi⁢ϕi. Then define the J-function

By a result of Givental and Coates-Corti-Iritani-Tseng, there is a mirror theorem stating that

under the mirror map τ=τ⁡(y). The I-function is an explicit hypergeometric function (here, we need X weak Fano).

We will now discuss I-functions and the extended complex moduli for the example of the crepant resolution. Let λ1,…,λ4∈HT2⁢(pt) be the T-equivariant parameters. Then set

Continuing, let u1±,…,u4±∈HC⁢R,T2⁢(X±) be the T-equivariant Poincaré duals of the “divisors” corresponding to the bi. Also define

Now we may define the I-functions of X± by

where ⟨d⟩ denotes the fractional part of d. The I-functions have some nice properties. For example, I±⁢(y±,z) are locally analytic near the “large radius limit point” P± of X± in the extended complex moduli space, which in our case is just the secondary variety corresponding to the secondary fan, which is simply ℙ⁡(1,3). The point P+ is a smooth point and P− is an orbifold point. Thus we can relate y+=y−−3. Based on this oversimplified picture, we can state the following result:

Theorem 3.2.1 (Crepant transformation conjecture, oversimplified).

There exists a degree-preserving RT-linear isomorphism 𝕌:HC⁢R,T∗⁢(X−,ℂ)→HC⁢R,T∗⁢(X+,ℂ) such that I+⁢(y+,z)=𝕌⁢I−⁢(y−,z) after analytic continuation.

4 Song (Feb 24): Crepant transformation conjecture for toric Deligne-Mumford stacks, part 2

Recall the oversimplified statement of the crepant transformation conjecture discussed previously:

Theorem.

There exists a degree-preserving RT-linear isomorphism 𝕌:HC⁢R,T∗⁢(X−,ℂ)→HC⁢R,T∗⁢(X+,ℂ) such that I+⁢(y+,z)=𝕌⁢I−⁢(y−,z) after analytic continuation.

4.1 Technical discussion of the simplified result

Begin by recalling the I-functions corresponding to X+=Kℙ2 and X−=[ℂ3/μ3]. We should note here that in this case, u4−=0 and u1+⁢u2+⁢u3+=0. We will reproduce the I-functions below:

The main tool here that we will use is the Mellin-Barnes analytic continuation. This is a classical tool for analytic continuation of hypergeometric functions. Replacing z=2⁢π⁢i, consider the functions

This allows us to define the T-equivariant gamma classes of X± by

We can relate I and H by I≈Γ̂⌣inv∗⁡H. Now we will work with the H-functions, and we will consider analytic continuation of H± under the fixed-point basis. At the point p1, we have u1+|p1=0, so we have

because Γ⁡(x) has simple poles at x∈ℤ≤0 and Γ⁡(x)⁢Γ⁡(1−x)=πsin⁡(π⁢x). Call the complicated function in the final expression F⁡(s). This has poles at all integers and at 13⁢u4+|p12⁢π⁢i−13⁢ℤ≥0. But now if C is a contour, we obtain the integral

Therefore H+⁢(y+)|p1 is a sum of residues at poles to the right of C, which is the integral for |y+| small. Taking the analytic continuation from right to left, we obtain the same integral for |y+| large, which is the same as small |y−|. But now H+⁢(y+)|p1 is a sum of residues at poles to the left of C. Omitting the explicit computations, we obtain

After more explicit computations, we obtain the desired result.

For general X±, suppose we have a 3-dimensional cone with chambers ± with a wall. Set ℳ to be approximately the toric variety defined by the two chambers. Then ℳ is the total space of a vector bundle on the curve 𝒞=ℙ⁡(a,b) corresponding to the wall. On 𝒞, there are points p+,p− coresponding to the respective large radius limits. We have analytic problems, so we will consider the formal neighborhhood of 𝒞 in ℳ. Once we do this, we can apply the Mellin-Barnes method.

4.2 Fourier-Mukai transform

Recall that we have the diagram

Because this diagram is a K-equivalence, we have a Fourier-Mukai transform

Theorem 4.2.1.

There exists a commutative diagram

The Ψ± are related to the “Γ~-integral structure” where roughly we have

This result is proved by explicit computation. But now because the Fourier-Mukai transform has a non-equivariant limit, so does 𝕌.

4.3 Related work

First, there is some related work in physics by Hori-Romo. In this, the Mellin-Barnes analytic continuation of the hemisphere partition function (the contour integral) corresponds to transportation of branes (the Fourier-Mukai transform).

Second, there is the work of Borisov-Horja. They replace our diagram by the diagram

where the bottom row consists of solutions spaces to hypergeometric systems. The bottom arrow is also a Mellin-Barnes integral, and the vertical arrows are so-called “mirror symmetry” maps. CIJ is a generalization of this.

Third, there is the work of Gonzalez-Woodward. They have a wall-crossing formula for genus-0 Gromov-Witten theory for general variation of GIT stability condition, which is a generalization of CIJ.

Finally, we can upgrade the simplified version by replacing orbifold cohomology by Givental’s loop spaces. There are other statements of the crepant transformation conjecture in quantum cohomology, but some of them have well-definedness issues. Unfortunately, we do not have the time to discuss these.

5 Che (Mar 03): Quantum groups and quantum cohomology, part 1

5.1 Nakajima quiver varieties

Let Q be a quiver with set of vertices I. Then let Q¯ be the framed double of Q. Choose dimension vectors v,w∈ℤI. Then let Vi=ℂvi,Wi=ℂwi. Then define 𝖱𝖾𝗉Q¯ to be the sum of

and its dual. This has an action of ∏G⁢L⁢(Vi) which preserves the symplectic form, so we have a moment map μ. Then we define

Example 5.1.1.

Let Q be the trivial quiver with one vertex. Then ℳQ⁢(k,n)=T∗⁢G⁢r⁢(k,n).

Example 5.1.2.

If Q is the quiver with one vertex and one loop, then ℳQ⁢(n,1)=Hilb⁡(ℂ2,n).

Nakajima quiver varieties have various group actions. First, define

Then we may define

Here, the ℂℏ× scales the symplectic form with weight ℏ. Now let A⊂GW be the maximal torus.

Example 5.1.3.

Consider T∗⁢G⁢r⁢(k,n). Then A=(ℂ×)n, and we have

Now define X⁡(n)≔⨆kT∗⁢G⁢r⁢(k,n). Clearly we have X⁢(n)A=𝒫⁡(n). Noting that X⁡(1)=pt⊔pt and X⁡(2)=pt⊔T∗⁢ℙ1⊔pt, we in fact have

In general, define ℳQ⁢(w)≔⨆vℳq⁢(v,w). Then we have

Now consider ℂ×↪A, where z↦(z,…,z⏟m,1,…,1⏟n−m). Then in fact (X⁡(n))ℂ×=X⁡(m)×X⁡(n−m).

We also note that (μ−1⁢(0)×Vi)⫽θ∏G⁢L⁢(Vi) is a vector bundle 𝒱i on ℳ!(v,w). The Chern classes of these 𝒱i are called tautological classes.

5.2 Quantum cohomology

Let X be a smooth variety with the action of a reductive group G with XG compact. Define the quantum product by

Here, γi∈HG∗⁢(X) and (γ1,γ2)=∫Xγ1∪γ2. Our goal is to describe this quantum cohomology in terms of the action of a quantum group. This connection seems surprising, but there is some previous literature on this.

  • •

    In 1994, Nakajima constructed an action of 𝒰⁡(𝔤) on H∙⁢(ℳQ⁢(w)), where Q is the quiver with adjacency matrix given by the Cartan matrix of 𝔤.

  • •

    Nekrasov-Shatashvili proved a correspondence between 2D N=2 supersymmetric gauge theories and so-called Heisenberg spin chains. Usually, there is a quantum group action on the Heisenberg spin chains producing operators commuting with the Hamiltonian.

  • •

    Maulik-Okounkov identified (HG∗⁢(X),∗) with a subalgebra of a quantum group called the Yangian.

5.3 Stable envelopes

First, we will construct the stable envelope. This will give us an R-matrix, which will in turn give us the Yangian.

Consider the action of A on X=ℳQ⁢(v,w). Fix a pair of tori A⊂T⊂Gw×Gedge where A preserves the symplectic form. Consider all of the A-weights of 𝒩XA/X. These partition the Lie algebra 𝔞=cochar⁡(A)⊗ℤℝ (the Lie algebra of the maximal compact) into chambers.

Example 5.3.1.

Consider T∗⁢ℙn−1. Then the weights are ai−aj, which are exactly the roots of S⁢Ln.

Definition 5.3.2.

Let C be a chamber of 𝔞 and let Y∈π0⁢(XA). Then the attracting set of Y with chamber C is defined to be

Here, limC(x)=limz→0σ⁡(z) for all σ∈C. Now we may define

where Y′<Y if Y′∩AttrC⁢(Y)¯≠∅.

Example 5.3.3.

In the example of T∗⁢ℙ1, let u=a1−a2. Then at 0, we have weight −u−ℏ in the fiber direction and u in the base direction and at ∞ we have weight −u in the base direction and u−ℏ in the fiber direction. Then choose C={u>0}. Note XA=0⊔∞. We then have AttrC⁢(0)=ℙ1∖∞ and AttrC⁢(∞)=F∞. We also have AttrCf⁢(0)=ℙ1∪F∞ and AttrCf⁢(∞)=F∞.

Definition 5.3.4.

A polarization is a choice of T1/2∈KT⁢(X) such that TX=T1/2⊕ℏ−1⁢(T1/2)∨.

Example 5.3.5.

For T∗⁢ℙ1, let T1/2=π∗⁢(Ωℙ1).

Next, we will define a degree in A. Consider

where 𝔞=𝖫𝗂𝖾⁡(A) and 𝔱=𝖫𝗂𝖾⁡(T). Note that ℂ⁡[𝔱]=ℂ⁡[𝔱/𝔞]⊗ℂ⁡[𝔞]. Thus for all α∈HT∗⁢(XA), we can define degA⁡α to be its degree in ℂ⁡[𝔞].

Theorem 5.3.6.

Fix a chamber C and a polarization T1/2. There exists a unique HT∗⁢(pt)-module map

such that for all Z∈π1⁢(XA) and for all γ∈HT/A∗⁢(Z) and Γ=StabC,T1/2⁢(Γ),

  1. 1.

    supp⁡(Γ)⊂AttrCf⁢(Z);

  2. 2.

    Γ|Z=±e⁡(N−)∪γ, where Ni is the repelling direction of the normal bundle of Z;

  3. 3.

    degA⁡Γ|Z′<12⁢codim⁡Z′ for all Z′<Z.

Example 5.3.7.

Consider T∗⁢ℙ1 again. Consider C={u>0}. Then we have StabC⁢(∞)=−P⁢D⁢([F∞]) and StabC⁢(0)=P⁢D⁢([ℙ1])+P⁢D⁢([F∞]). In particular, StabC⁢(0)|∞=u−ℏ+(−u)=−ℏ. Finally, we see

when restricted to the fixed points.

Remark 5.3.8.

For the example of T∗⁢ℙ1, the conditions in the theorem say that

  1. 1.

    StabC is lower triangular;

  2. 2.

    The diagonal elements are ±e⁡(N−);

  3. 3.

    The off-diagonal elements have A-degree less than 12⁢codim⁡Z′.

6 Che (Mar 10): Quantum groups and quantum cohomology, part 2

We will consider X⁡(n)=⨆kT∗⁢G⁢r⁢(k,n). Then recall that we have A⊂T=A×ℂℏ×, where the ℂℏ× scaled the contangent direction. Also recall that we constructed stable envelopes. Recall that we constructed stable envelopes StabC for a chamber C in the real cocharacters of A.

6.1 R-matrices

Let C,C′ be chambers in 𝔞ℝ. Then define RC′,C=StabC′−1∘StabC. In the example of T∗⁢ℙ1, we have

where

is the swapping operator. Our goal is to eventually construct the Yangian YQ acting on HT∙⁢(ℳQ⁢(v,w)).

Remark 6.1.1.

We can replace A with a subtorus A′. For example, we want X⁢(n)ℂ×=X⁡(n1)×X⁡(n2), where n=n1+n2.

Now let C be a chamber and C′ be a face of C. Then write 𝔞′ for the span of C′ and A′ for the subtorus of A associated to 𝔞′.

Lemma 6.1.2 (Triangle lemma).

The stable envelope factors as

Example 6.1.3.

Consider X⁡(3) with the action of A={(a1,a2,a3)}, C=a1>a2>a3, and C′ to be the face where a1−a2=0. Then define A′={a1=a2}⊂A. Now we have X⁢(3)A′=X⁡(2)×X⁡(1). The lemma tells us we have a diagram

Corollary 6.1.4.

If C1 and C2 are two adjacent chambers separated by a wall C′, then denote by >0,<0 the chambers C1/C′,C2/C′. Then RC2,C1=R<0,>0.

Proof.

Applying the triangle lemma, we can see that the second factors are the same, so we obtain

Theorem 6.1.5 (Yang-Baxter equation for X⁡(3)).

Let R⁡(u)∈End⁡(HT∗⁢(X⁢(2)A)) be the R-matrix we constructed. Then

where Ri⁢j∈HT∗⁢(X⁡(1)×X⁡(1)×X⁡(1)) acts on the i,j-th factor.

Here, in our example, we should have RC1,C6∘⋯∘RC3,C2∘RC2,C1=1. But then we have relations like R12⁢(u)=R21⁢(−u)−1, and this gives us the desired result.

6.2 Construction of the Yangian

This construction is due to Fadeev, Reshetikhin, Takhtajan. Let F−HGedge∗⁢(X⁡(1))=ℂ⁢[ℏ]⊕2. Then define F⁡(u)=F⁡[u], where this notation persists for historical reasons. Now let

where the tensor products are taken over Hℂℏ×∗⁢(pt). If we set W=F⁡(u1)⊗⋯⊗F⁡(un), define RW∈End⁡(F⊗W)⁢(u) by

This corresponds to the below picture where the horizontal line is auxillary space and the vertical lines are physical space.

F0⁢(u)F⁡(u1)F⁡(u2)F⁡(un)⋯
Figure 6.1: Pictorial description of RW

Now we have Ru∈End⁡(F⊗W)⁢(u), but we wanted elements of End⁡(W). First, we need to take expansions in u−1, and then we view RW as a matrix with values in End⁡(W). Finally, we take YQ to be generated by all limits (as u→∞) of all coefficients of all matrix elements of RW for all possible F0 after dividing by ℏ.

Note that YQ acts on F⁡(u1)⊗⋯⊗F⁡(un). On the other hand, we had

because X⁢(n)A=X⁢(1)n, and therefore YQ acts on HT∗⁢(X⁡(n)) for all n because StabC is an isomorphism after localization. This choice should be independent of the choice of C by the Yang-Baxter equation.

Now let 𝔤Q be the span of the u−1 terms in YQ and 𝔥¯Q be the span of vi¯,wi¯, where for α∈HT∗⁢(ℳQ⁢(v,w)), vi¯⋅α=vi⋅α.

Proposition 6.2.1.
  1. 1.

    𝔤Q is a Lie algebra;

  2. 2.

    𝔥¯Q⊂𝔤Q is a subalgebra;

  3. 3.

    𝔥¯Q is a maximal commutative subalgebra in 𝔤Q.

Example 6.2.2.

Consider X⁡(2)=pt⊔T∗⁢ℙ1⊔pt. We have

Define r=1−S to be the classical r-matrix. This implies

After computation, we will see that 𝔤Q is spanned by

We then see that h1=u¯,h2=w¯−u¯, and that e,f are raising and lowering operators, respectively.

Proposition 6.2.3.

The associated graded algebra of YQ is gr⁡YQ≃𝒰⁡(𝔤Q⁢[u]), where the grading is with respect to degree in u shifted by 1.

Then consider 𝔤Q=𝔥¯Q⊕⨁α>0(𝔤α⊕𝔤−α).

Theorem 6.2.4.

Let λ∈ℂI and c1⁢(λ)=∑i∈Iλi⁢c1⁢(𝒱i). Then

where the quantum product is modified by qβ=(−1)(KX,β)⁢qβ.

Now let X=ℳQ⁢(v,w). This has an action of T, so consider γ∈HT2⁢(X) and γ1,γ2∈HT∗⁢(X). Then we have

But then

7 Kevin (Mar 24): Donaldson-Thomas type invariants via microlocal geometry, part 1

For any finite type Deligne-Mumford stack X/ℂ, Behrend constructs a natural constructible function νX. More precisely, we can use νX to compute enumerative invariants.

Theorem 7.0.1.

Let X be a projective Deligne-Mumford stack (maybe proper with projective coarse moduli space?) with a symmetric obstruction theory. If X is either smooth, a global finite group quotient, or a gerbe over a scheme, then

Example 7.0.2.

Let X be a DT-type moduli space for a Calabi-Yau threefold. Then X has a symmetric obstruction theory, and so we can compute Donaldson-Thomas and Pandharipande-Thomas invariants for Calabi-Yau threefolds using the Behrend function.

We will attempt to explain the following diagram where X↪M is an embedding of X into a smooth stack M:

The left triangle was studied by MacPherson, and is relatively well-known. The right triangle is microlocal because of the presence of the cotagent bundle.

7.1 Obstruction theories

7.1.1 Perfect obstruction theories

Recall that a perfect obstruction theory for a Deligne-Mumford stack X is ϕ:E→τ≥−1⁢LX that is a surjection on h−1 and an isomorphism on h0. We also require that E is a perfect complex with amplitude contained in [−1,0]. Note that τ≥−1⁢LX is locally of the form

where M is a smooth stack containing X.

From the perfect obstruction theory, we can construct the intrinsic normal cone, which is the cone stack ℭX such that if U→X is étale and U→M is an embedding into a smooth scheme, then ℭX|U=[CU/M/TM|U]. Then a perfect obstruction theory gives us a closed immersion ℭX↪𝔈, and the virtual fundamental class is

This is really bad because of the vector bundle stacks, so Behrend gives a different construction that will give us honest vector bundles.

Definition 7.1.1.

A local resolution of E over U→X is a presentation of E as [E1→E0] over U. The obstruction cone is the cone C that is the pullback

We have a surjection E1∨→Ob=H1⁢(E∨).

Proposition 7.1.2.

Let Ω be a vector bundle on X with a surjection Ω↠Ob. Then there exists a unique closed subcone C⊂Ω such that for all local resolutions [E0→E0] on U→X with obstruction cone C′ and lifts

C|U=ϕ−1⁢(C′). This C is called the obstructon cone of Ω↠Ob.

Some nice facts in the case that X is projective are

  • •

    Perfect obstruction theories admit global resolutions;

  • •

    There exists an embedding X↪M into a smooth stack.

Proposition 7.1.3.

Let X be projective with perfect obstruction theory E. Let Ω be a vector bundle with surjection Ω↠Ob and let C⊂Ω be the obstruction cone. Then C has pure dimension rk⁡E+rk⁡Ω and

7.2 Symmetric obstruction theories

Definition 7.2.1.

A symmetric obstruction theory is a perfect obstruction theory E endowed with a nondegenerate symmetric bilinear form of degree 1, which means an isomorphism

such that Θ=Θ∨⁢[1].

Note that if X has a symmetric obstruction theory, which means that [X]vir∈A0⁢(X). Thus we can define the virtual count #vir⁢(X)=deg⁡[X]vir if X is proper.

Example 7.2.2.

Let Y be a projective Calabi-Yau threefold and X be a Hilbert scheme of subschemes of Y of dimension ≤1. Also let ℰ be the universal ideal sheaf on Y×X. Then let ℱ fit into a distinguished triangle

Now E≔R⁢πX⁣∗⁢ℱ⁢[2] is a symmetric obstruction theory. First, it is clear that ℱ is self-dual, and second, by relative Serre duality, we have

Shifting and composing, we obtain E≃E∨⁢[1].

Example 7.2.3.

Let M be a smooth variety and let ω be an almost closed 1-form (which means that dω|Z⁡(ω)=0). Then X has a symmetric obstruction theory

This is clearly self-dual, and the map to the cotangent complex is given by ω:TM|X→I/I2. Then we obtain a closed embedding CX/M↪ΩM|X, and the image of this is the obstruction cone of ΩM|X↠Ob.

Proposition 7.2.4.

Let E be a symmetric obstruction theory on X. Then étale locally, there exists U↪M and an almost closed 1-form ω such that U=Z⁡(ω) and E|U is the symmetric obstruction theory associated to ω.

7.3 The Behrend function

We want to think of νX as an invariant of singularities of X. This means that:

  • •

    If x∈X is a smooth point, then νX⁢(x)=(−1)dimX;

  • •

    If f:X→Y is smooth, then f∗⁢νY⁢(−1)dimX−dimY⁢νX;

  • •

    For any x∈X,y∈Y, we have νX×Y⁢(x,y)=νX⁢(x)⁢νY⁢(y);

  • •

    If X=Z⁡(df)⊂M, then

    Here, Φf is the vanishing cycles and Fx is the Milnor fiber.

Now we want to define νX. The first tool is MacPherson’s local Euler obstruction

which is in the diagram 7.1 and is given by

Here, if we choose x∈V, embed V↪M into a smooth M, and consider the Grassmann bundle G→V of rank p=dimV quotients of ΩM|V. Above the smooth locus Vs, there exists a section Vs→G sending p to ΩV|p. Finally, we let V~ be the closure of this section in G and μ:V~→V be the natural projection, called the Nash blowup. Also, T~ is the universal quotient bundle on G.

Now we can define the Chern-Mather class by

We now need to construct a cycle 𝔠X. Embed X↪M into a smooth stack and consider the normal cone C=CX/M. Also let π:C→X be the projection, and define the cycle

where C′ ranges over the irreducible components of C. When X is smooth, then 𝔠X=(−1)dimX⁢[X].

We may finally define the Behrend function by νX≔Eu⁡(𝔠X). There is also the Aluffi class

When X is smooth, then αX=c⁡(ΩX)∩[X].

Proposition 7.3.1.

In the cases of the main theorem, we have

8 Kevin (Mar 31): Donaldson-Thomas type invariants via microlocal geometry, part 2

Recall the diagram (7.1) from last time. Our first goal is to show that (αX)0=[X]vir.

8.1 Microlocal geometry

Let M be a smooth variety of dimension n.

Definition 8.1.1.

An irreducible closed subset C⊂ΩM is conic if it is invariant under the ℂ∗ action. C is Lagrangian if the symplectic form vanishes generically on C.

Definition 8.1.2.

A closed subscheme of ΩM is conic Lagrangian if all of its irreducible components of conic Lagrangian.

Because these conditions are actually étale local, we can extend the definitions to the case where M is a smooth Deligne-Mumford stack.

Example 8.1.3.

Let X⊂M be irreducible. Then the closure of the conormal bundle to Xsm is a conic Lagrangian cycle.

Definition 8.1.4.

We will define ℒ⁡(ΩM)⊂Zn⁢(ΩM) to be the group of conic Lagrangian cycles and ℒX⁢(ΩM) to be the group of conic Lagrangian cycles supported over M.

Proposition 8.1.5.

There is an isomorphism ℒ⁡(ΩM)≅Z∗⁢(M) given by the maps

and

The point of this microlocal geometry is the following result:

Proposition 8.1.6.

The diagram

commutes.

Lemma 8.1.7 (Fundamental lemma for almost closed 1-forms).

The obstruction cone C (where [X]vir=0ΩM!(C)) is conic Lagrangian.

To prove the lemma, we can simply check locally. Now, if we note that

by definition of 𝔠X, then we obtain

as desired.

8.2 Hilbert schemes of points on threefolds

Let X=Z⁡(ω)⊂M and x∈X. We can choose coordinates x1,…,xn on M and extend to coordinates x1,…,xn,p1,…,pn on ΩM. Then write

Now let η∈ℂ−{0} and define Γη to be the image of 1n⁢ω:M→ΩM. We also define Δ to be the image of the section

Proposition 8.2.1.

For ε>0 small and η small relative to ε, we can express the Behrend function as a linking number via the formula

where Sε is a radius ε copy of S4⁢n−1 around x∈ΩM.

We will not consider torus actions. Let X be a ℂ∗-scheme with an isolated fixed point x and an equivariant symmetric obstruction theory.

Proposition 8.2.2.

There exists an invariant affine neighborhood U∋x where U=Z⁡(ω)⊂M equivariantly such that ω is an invariant almost closed 1-form and dimM=dimTX|x.

Proposition 8.2.3.

Let S1 act on S with finite stabilizers. Let A,A′,B be S1-invariant submanifolds. Suppose we have an S1-equivariant homotopy T×Z→S from A to A′ where T is an open interval containing 0 and 1. Then

The proof of this requires us to talk about orbifolds, so we will not discuss it.

Proposition 8.2.4.

Let M be a smooth n-dimensional ℂ∗ scheme with an isolated fixed point x. Let X=Z⁡(ω), where ω is an invariant almost closed 1-form and x∈X. Then

Proof.

Choose equivariant coordinates x1,…,xn,p1,…,pn on ΩM around x. Then let

be the image of (p1,…,pn)↦(t⁢p¯1,…,t⁢p¯n,p1,…,pn), which is oriented (−1)n times the orientation p1,…,pn. Now let Δt′=Δt∩Sε. There is an S1-equivariant homotopy from Δ0′ to Δ1′ given by the formula

Because x is an isolated fixed points, the action of S1 on Sε has finite stabilizers, so by the previous proposition,

But the left hand side is (−1)n and the right side is νX⁢(x), as desired. □

Finally, we give the following result:

Theorem 8.2.5.

For ℂ3, we have the formula

Theorem 8.2.6.

Recall that the fixed points of the torus action on the Hilbert scheme are actually the monomial ideals I. But then

by MNOP, and therefore

where Mn is the n-th coefficient of the McMahon function.

9 Konstantin (Apr 07): Inductive construction of stable envelopes, part 1

9.1 Elliptic cohomology

There is a general way to construct generalized cohomology theories from formal groups using complex cobordism, but this is totally not useful in computations. There is another more explicit construction using derived moduli stacks of elliptic curves, but we don’t understand infinity-categories, so it is useless to us. Instead, in 1995, Ginzburg, Kapranov, and Vasserot constructed equivariant elliptic cohomology with complex coefficients. This generalizes a construction of Grojnovski, which reduces to ordinary cohomology in the non-equivariant case. Even here, the notation used by Okounkov (which we will use) is very different from that of Ginzburg, Kapranov, and Vasserot.

We will consider the elliptic curve

living over Spec⁡ℂ⁡[q]. This is identified with ℂ/ℤ+τ⁢ℤ by the exponential map ℂ→ℂ× and q=e2⁢π⁢i⁢τ. We will also consider

where ℰz gives the Kähler variables. For an elliptic curve E, we have an exact sequence

Note that any bundle in Pic0⁡(E) has a meromorphic section θ⁡(x⁢z)θ⁡(x), where x∈E,z∈E∨, and

This satisfies the equation θ⁡(q⁢x)=−x−1⁢θ⁢(x), so we have an entire function on ℂ×. Also, we note that θ⁡(qn)=0, so θ⁡(x) is a section of the line bundle corresponding to {0}⊂E.

Let G be an algebraic group, which for us will be a torus. Then we will define

For example, if G=(ℂ×)n, then 𝖤𝗅𝗅G⁢(pt)=En=(ℂ×)n/qℤn. In general, we will have a morphism of schemes

This gives us a sheaf π∗⁢𝒪𝖤𝗅𝗅G⁢(X)=𝖤𝗅𝗅G0⁢(X). Note that elliptic cohomology is periodic, and in fact we have

where ω=πE,∗⁢T1⁢E is a line bundle.

Remark 9.1.1.

If G=G⁢L⁢(r), then 𝖤𝗅𝗅G⁢(pt)=Sr⁢E. If G=μr, then 𝖤𝗅𝗅G⁢(pt)=E⁡[r] is the group of r-torsion points on E.

Remark 9.1.2.

Note that HG∙⁢(X) is a ring over HG∙⁢(pt). Therefore, we can take Spec⁡HG∙⁢(X)=𝖫𝗂𝖾⁡(G) as a scheme. We can do a similar construction for Spec⁡KG⁢(X)→Spec⁡KG⁢(pt)=G. Then there is a map exp:𝖫𝗂𝖾⁡(G)→G, and then of course π:G→G/qcochar⁡G=𝖤𝗅𝗅G⁢(pt).

Now we can construct a sheaf

Example 9.1.3.

We will now compute the elliptic cohomology of X=ℙ1, equivariant with respect to A=(ℂ×)2. First, note that

Note that this is a union of two hyperplanes and projects to u=0. Next, in K-theory, we have

By our choices, we still have a union of hyperplanes, and the projection is onto x=1. We now need to take a quotient with respect to multiplication by q, and this tells us that

where the diagonal is the locus a1=a2. In particular, we have

Unfortunately, there is no easy geometric interpretations of elliptic cohomology classes comparable to cohomology classes being cycles and K-theory classes being sheaves. However, we can consider them as sections of bundles on 𝖤𝗅𝗅G⁢(X). Let V be a vector bundle on X. Then the Thom space of V is defined to be Th⁡(V)=ℙ⁡(V⊕ℂ)/ℙ⁡(V). Its elliptic cohomology is defined to be

which is a module over 𝒪𝖤𝗅𝗅G⁢(X) and thus a pushforward of a sheaf on 𝖤𝗅𝗅G⁢(X). We will call this sheaf Θ⁡(V).

Now denote ℰG≔𝖤𝗅𝗅G⁢(pt) and note that 𝖤𝗅𝗅G⁢(X) is a finite cover of ℰG. In cohomology, we have something like

This has a map to H∙⁢(ℙ⁡(V))=ℂ⁡[u,v1,…,vr,…]/⟨⋅∏i=1r(u−vi)⟩, and this tells us that

Recall from cohomology the Thom isomorphism HG∙⁢(X)→HG⁢(Th⁡(V)) given by α→α⁢∏vi. In elliptic cohomology, if we replace vi by θ⁡(vi), we take

The bundle ΘX⁢(V) satisfies the following properties:

  • •

    ΘX⁢(V⊕W)=ΘX⁢(V)⊗ΘX⁢(W);

  • •

    Θ⁡(V∨)≅Θ⁡(V).

This defines a map ΘX:KG⁢(X)→Pic⁡(𝖤𝗅𝗅G⁢(X)). In fact, this generates all bundles used in Okounkov’s constructions.

Now let f:X→Y. This defines a map 𝖤𝗅𝗅G⁢(X)→𝖤𝗅𝗅G⁢(Y). We also have a map

where Nf=f∗⁢T⁢Y−T⁢X. If f is a closed immersion, then f ⋆ ○ ⁢α=α⁢∏iθ⁡(vi). On the other hand, if f is a flat projection, we have a fiber integration map

9.2 Outline of Okounkov’s construction

Let (X,ω) be a homomorphic symplectic variety with an action of T⊃A, where A preserves the symplectic form. Recall that stable envelopes are related to attracting sets. Recall that if XA⊂X is the fixed locus, consider characters w entering NX/XA. This defines a hyperplane arrangement in 𝔞, and so we choose a chamber. Each weight becomes a positive, negative, or zero weight of a bundle on XA. This gives us attracting and repelling directions to XA.

For each component Fi⊂XA, we can define Attr⁡(Fi)⊂X×Fi to be the set where x∈X→y∈Fi as we approach 0∈ℂ⊃ℂ×. Next time, we will discuss stable envelopes for elliptic cohomology, which will be sections of bundles that satisfy properties analogous to those of attracting sets.

10 Konstantin (Apr 14): Inductive construction of stable envelopes, part 2

10.1 More on elliptic cohomology

Let G=(ℂ×)n. We again take E=ℂ×/qℤ. Set

Recall we have a diagram

We will define the Elliptic cohomology functor X↦π∗X⁢𝒪𝖤𝗅𝗅G⁢(X). Also recall the Thom sheaf, which was produced from the Thom space

This is π∗X⁢Θ⁢(−V), and Θ⁡(V) is a line bundle on 𝖤𝗅𝗅G⁢(X) with a section ∏iθ⁡(vi).

Finally, recall that if f:X→Y is proper, there is a map

To construct this, consider the topological picture X↪V→Y and consider the tubular neighborhood Tub⁡(X). Also define N=TX⁢V=Ni, so we have a map Th⁡(V)→Th⁡(N). We can also assume that Tub⁡(X)⊂V<1 is contained inside the disk bundle on V. Thus we can assume that

This defines a map between the elliptic cohomology of N and the elliptic cohomology of V, so we obtain

and then this becomes

and noting that π∗X=π∗Y⁢f∗, we obtain the desired f ⋆ ○ .

We now discuss correspondences. Consider the diagram

Then we want morphisms π∗X⁢𝒪𝖤𝗅𝗅G⁢(X)→π∗Y⁢𝒪𝖤𝗅𝗅G⁢(Y). For some class α, we will consider the map

To make this work, we require

We can also consider more general line bundles 𝒮X,𝒮Y on 𝖤𝗅𝗅G⁢(X),𝖤𝗅𝗅G⁢(Y). In this case, we want

where 𝒮∇≔𝒮−1⊗Θ⁡(T⁢X).

10.2 Okounkov’s construction

Let X be a holomorphic symplectic variety with symplectic form ω. Let T act on X with A⊂T be the subtorus preserving the symplectic form. Finally, write

as a disjoint union of connected components. From the bundle TXA⁢X=NXA/X, we produce characters wj of A by restricting to each fixed component. Now fix a chamber a chamber 𝒞⊂𝔞≔cochar⁡A⊗ℤℝ where every normal weight is either positive or negative. Then if W is a bundle over XA, we of course have

For each fixed component, we can define the attracting set Attr⁡(Fi)⊂X. Now consider the diagram

Then note that Nȷ=π∗⁢(N<0+T⁢F).

For example, if X=T∗⁢ℙ1, let ∞=F1 and 0=F2. Then Attr⁡(F1) is the fiber above ∞ and Attr⁡(F2) is the base ℙ1∖∞. Now define Fi<Fj if Fi⊂Attr⁡(Fj)¯, and we can upgrade this to a total order, so write F1<F2<⋯<FN. Finally, define the total attracting locus

In the example of X=T∗⁢ℙ1, we see that Attrf⁢(F2) is the union of the zero section and the fiber above ∞. Finally, define

Now define 𝒮A≔i∗⁢𝒮⊗Θ⁡(−N<0).

Proposition 10.2.1.

We have

for any 𝒮. Thus ȷ ⋆ ○ provides a section of Θ⁡(Nj) on Xi.

Proof.

We have the identity

Then the right hand side becomes

Now consider the diagram

This defines us a morphism

Lemma 10.2.2.

The map ȷ¯ ⋆ ○ ⁢π∗ is injective, so the long exact sequence for the pair Xi⊃Xi+1 becomes

This result gives us an interpolation from Xi+1 to Xi. Tensoring the previous exact sequence with 𝒮⊠𝒮A∇ on Xi×Fi, we obtain

We want to make the support Δ of π∗Xi⁢(𝒮A|Fi⊠𝒮A∇|Fj) closed on ℰT. If we succeed, we obtain a unique section of 𝒮⊠𝒮A∇⁢(∞⁢Δ) (section with poles of arbitrary order at Δ) starting from ȷ ⋆ ○ on Xi×Fi. By the inductive construction, we obtain a section on X×Fi.

To make the support closed, we do the following. Let ℒ→E be a line bundle. If ℒ has degree 0, then H∗⁢(ℒ)≠0 if and only if ℒ≅𝒪E. Thus we want to make deg⁡𝒮A=0 over ℰA, which is most simply achieved by polarizations.

Lemma 10.2.3.

If T1/2∈KA⁢(X) is such that T⁢X=T1/2+(T1/2)∨, then we can take 𝒮=Θ⁡(T1/2).

This automatically gives us deg⁡𝒮A=0, and to make 𝒮A nontrivial, we can extend the base and take

and then 𝒮A will be nontrivial for generic z.

11 Patrick (Apr 21): DT/PT correspondence using motivic Hall algebras

11.1 Introduction

Let 𝒳 be a proper smooth Deligne-Mumford stack with ω𝒳=𝒪𝒳, H1⁢(𝒳,𝒪𝒳)=0, and projective coarse moduli space X. Note here that X is Gorenstein, Calabi-Yau, and has at worst quotient singularities. We will also impose that 𝒳 satisfies the hard Lefschetz condition, which says that age is invariant under the map I:I⁢𝒳→I⁢𝒳 taking (x,g)↦(x,g−1).

Now let N⁡(𝒳) denote the numerical K-theory of 𝒳, that is K⁡(D⁡(𝒳)) modulo the Euler pairing

Then denote N≤1⁢(𝒳) denote the classes with support in dimension at most 1 and N0⁢(𝒳) be the classes with 0-dimensional support. Then choose1 a splitting

Finally, we call a class β∈N1⁢(𝒳) effective if it is represented by an actual sheaf F∈𝙲𝚘𝚑≤1⁢(𝒳).

Now let Y=Quot⁡(𝒪𝒳,[𝒪x]) for a non-stacky point x∈𝒳. Then Y is a smooth projective Calabi-Yau threefold and there is a map f:Y→X, which is a crepant resolution. Then, if we consider the universal sheaf on Y×𝒳 by 𝒪𝒵, the Fourier-Mukai transform

is a derived equivalence, called the McKay correspondence. Then there is a map N≤1⁢(Y)→N≤1⁢(𝒳), and thus we obtain a splitting N≤1⁢(Y)=Nn-exc⁢(Y)⊕Nexc⁢(Y). Finally, write Nmr⁢(𝒳)=Φ⁡(N≤1⁢(Y)) and N1,mr⁢(𝒳)=Φ⁡(Nn-exc⁢(Y)). Also, write Ψ≔Φ−1.

Then define following DT and PT generating series:

Theorem 11.1.1 (DT/PT correspondence).

Let β∈N1,mr⁢(𝒳). Then there is an equality of generating series

11.2 Categorical preliminaries

Definition 11.2.1.

Let 𝙱 be an abelian category. Then a torsion pair is a pair of full subcategories (𝚃,𝙵) such that Hom⁡(T,F)=0 for all T∈𝚃,F∈𝙵 and every E∈𝙱 fits into a (unique) short exact sequence

with TE∈𝚃,FE∈𝙵.

This can be naturally generalized to the notion of a torsion n-tuple (𝙱1,…,𝙱n).

Example 11.2.2.

Let 𝚃=𝙲𝚘𝚑≤d⁢(𝒳) and 𝙵=𝙲𝚘𝚑≥d+1⁢(𝒳) be the category of sheaves with no subsheaves of dimension ≤d. Then (𝚃,𝙵) is a torsion pair on 𝙲𝚘𝚑⁡(𝒳).

For a torsion pair (𝚃,𝙵), we can define a new abelian category

This is called tilting at the torsion pair (𝚃,𝙵).

Example 11.2.3.

Define the category 𝙲𝚘𝚑♭⁢(𝒳) by tilting at the torsion pair of the previous example for d=1.

Unfortunately, this category is too large for our purposes, so we will define a new category. First, if 𝒞1,…,𝒞n⊂D⁡(𝒳) are full subcategories, let ⟨𝒞1,…,𝒞n⟩ex⊂D⁡(𝒳) be the smallest full subcategory of D⁡(𝒳) containing all of the 𝒞i and closed under extensions. Now we may define

This is a Noetherian abelian category with the inclusions 𝙲𝚘𝚑≤1⁢(𝒳)⊂𝙰⊂D⁡(𝒳) being exact. In addition, 𝙰 contains all ideal sheaves IC⁢[1], PT stable pairs, and Bryan-Steinberg pairs.

Definition 11.2.4.

Let 𝚃,𝙵 be full subcategories of 𝙲𝚘𝚑≤1⁢(𝒳). Write 𝙿𝚊𝚒𝚛⁡(𝚃,𝙵)⊂𝙰 for the full subcategory on objects E such that rk⁡E=−1, Hom⁡(T,E)=0 for T∈𝚃, and Hom⁡(E,F)=0 for F∈𝙵.

Example 11.2.5.

For example, if 𝚃PT=𝙲𝚘𝚑0⁢(𝒳) and 𝙵PT=𝙲𝚘𝚑1⁢(𝚇), then a (𝚃PT,𝙵PT)-pair is a PT stable pair.

For a pair of full subcategories as above, we may also define

If (𝚃,𝙵) and (𝚃~,𝙵~) are torsion pairs, then (𝚃,𝚅⁡(𝚃,𝙵~),𝙵~) is a torsion triple on 𝙰.

11.3 Moduli stack

There is a stack 𝔐⁢𝔲⁢𝔪Y parameterizing objects E∈D⁡(Y) such that Ext<0⁡(E,E)=0 which is an Artin stack locally of finite type, which was constructed by Lieblich in 2006.2 Because Fourier-Mukai transforms behave well in families, by the McKay correspondence, we have a corresponding stack 𝔐⁢𝔲⁢𝔪𝒳. In addition, there is a decomposition

into open and closed substacks by the numerical K-theory class. If 𝙲⊂D⁡(𝒳) is a full subcategory of objects with vanishing self-Ext<0 and defines an open substack of 𝔐⁢𝔲⁢𝔪𝒳, then we will call the corresponding open substack 𝙲¯⊂𝔐⁢𝔲⁢𝔪𝒳.

Definition 11.3.1.

A torsion pair (𝚃,𝙵) on 𝙲𝚘𝚑≤1⁢(𝒳) is open if 𝚃,𝙵 are open subcategories.

Lemma 11.3.2.

The following conditions define open substacks of 𝔐⁢𝔲⁢𝔪𝒳 for full, open subcategories 𝚃,𝙵 of 𝙲𝚘𝚑⁡(𝒳):

  1. 1.

    H0⁢(E)∈𝚃 and E∈D[−1,0]⁢(𝒳);

  2. 2.

    H−1⁢(E)∈𝙵 and E∈D[−1,0]⁢(𝒳).

In particular, if (𝚃,𝙵) is an open torsion pair, then the objects of the tilt of 𝙲𝚘𝚑⁡(𝒳) at (𝚃,𝙵) form an open substack of 𝔐⁢𝔲⁢𝔪𝒳

Proposition 11.3.3.

Let (𝚃,𝙵) be open torsion pairs on 𝙲𝚘𝚑≤1⁢(𝒳). Suppose that 𝙲𝚘𝚑0⁢(𝒳)⊂𝚃. Then the substack 𝙿𝚊𝚒𝚛¯⁢(𝚃,𝙵~)⊂𝔐⁢𝔲⁢𝔪𝚇 is open.

In particular, 𝙲𝚘𝚑♭⁢(𝒳) is an open subcategory. In addition, the category 𝙰rk≥−1⊂𝙲𝚘𝚑♭⁢(𝒳) is open. This means that

are inclusions of open substacks, and in particular, the first two stacks are Artin stacks locally of finite type.

11.4 Motivic Hall algebras

Recall the Grothendieck ring of varieties K⁡(𝚅𝚊𝚛ℂ) (taken with ℚ-coefficients), which is spanned by isomorphism classes of varieties with the relations [X]=[Z]+[X∖Z] for a closed subvariety Z⊂X and [X]⋅[Y]=[X×Y]. This will not be helpful when we consider stacks, so instead we will consider the following two relations.

Definition 11.4.1.

Let Y be connected. Then a morphism f:X→Y is a Zariski fibration if there is an open cover Y=⋃Ui such that f−1⁢(Ui)=Ui×Fi.3

Definition 11.4.2.

A morphism f:X→Y is a geometric bijection if f induces a bijection on ℂ-points.

Lemma 11.4.3.

K⁡(𝚅𝚊𝚛ℂ) is spanned by isomorphism classes of varieties with the following relations:

  1. 1.

    If f:X→Y is a geometric bijection, then [X]=[Y];

  2. 2.

    [X1]+[X2]=[X1⊔X2].

Now we say that a morphism f:X→Y of stacks is a Zariski fibration if for any scheme T, the base change fT:X×YT→T is a Zariski fibration of schemes. Of course, the definition of geometric bijection should be changed to that of inducing an equivalence on groupoids of ℂ-points.

Definition 11.4.4.

The Grothendieck ring of stacks K⁡(𝚂𝚝ℂ) is the ℚ-vector space spanned by symbols [X] for finite type Artin stacks with affine stabilizers X with the following relations:

  1. 1.

    [X⊔Y]=[X]+[Y];

  2. 2.

    If f:X→Y is a geometric bijection, then [X]=[Y];

  3. 3.

    If X1,X2→Y are Zariski fibrations with the same fibers, then [X]=[Y].

For any Artin stack S locally of finite type with affine stabilizers, then there is a relative version K⁡(𝚂𝚝S), which is a module over K⁡(𝚂𝚝ℂ). For our purposes, we will take 𝙲=𝙲𝚘𝚑♭⁢(𝒳) and let S=𝙲¯.

Proposition 11.4.5.

There exists an Artin stack 𝙲¯(2) locally of finite type parameterizing short exact sequences in 𝙲. This stack is equipped with maps

for i=1,2,3. Then the morphism (π1,π3):𝙲¯(2)→𝙲¯ is of finite type.

Now, for [X1→𝙲¯],[X2→𝙲¯]∈K⁡(𝚂𝚝𝙲¯), define the stack X1∗X2 as the fiber product

This defines an associative product on K⁡(𝚂𝚝𝙲¯), and we define the motivic Hall algebra H⁡(𝙲)≔(K⁡(𝚂𝚝𝙲¯),∗).

There is an inclusion K⁡(𝚅𝚊𝚛ℂ)⁢[𝕃−1,(1+⋯+Ln)−1∣n≥1]→K⁡(𝚂𝚝ℂ). Then we define the subalgebra of regular elements Hreg⁢(𝙲) to be the K⁡(𝚅𝚊𝚛ℂ)⁢[𝕃−1,[ℙn]−1∣n≥1]-submodule generated by schemes [Z→𝙲¯].

Proposition 11.4.6.

Hreg⁢(𝙲) is closed under ∗ and the quotiennt Hreg⁢(𝙲)/(𝕃−1)⁢Hreg⁢(𝙲) is commutative with Poisson bracket given by {f,g}=f∗g−g∗f𝕃−1. This is called the semi-classical Hall algebra.

Let ℚ⁡[N⁡(𝒳)] with product tα1⋆tα2=(−1)χ⁡(α1,α2)⁢tα1+α2 be the Poisson torus. The Poisson bracket is defined by {tα,tβ}=(−1)χ⁡(α,β)⁢χ⁢(α,β)⁢tα+β. Then there is a Poisson algebra homomorphism

where ν:𝙲→ℤ is the Behrend function.

We will not be considering the motivic Hall algebra but rather a completed version Hgr⁢(𝙲) which is analogous to the completion from Laurent polynomials to Laurent series. There is an algebra Hgr,reg⁢(𝙲) of regular elements and a semi-classical algebra Hgr,sc⁢(𝙲), which comes with an integration morphism

11.5 Wall-crossing and the DT/PT correspondence

Definition 11.5.1.

A torsion pair (𝚃,𝙵) on 𝙲𝚘𝚑≤1⁢(𝒳) is called numerical if whenever [T∈𝚃]=[F∈𝙵] in N⁡(𝒳), then T=F=0.

All of the torsion pairs we will consider are numerical.

Proposition 11.5.2.

Let (𝚃,𝙵) be an open numerical torsion pair and suppose that ℳ⊂𝙿𝚊𝚒𝚛¯⁢(𝚃,𝙵) is an open and finite type substack. Then (𝕃−1)⁢[ℳ]∈Hreg⁢(𝙲).

Lemma 11.5.3.

Assume that 𝙿𝚊𝚒𝚛⁡(𝚃−,𝙵−),𝙿𝚊𝚒𝚛⁡(𝚃+,𝙵+),𝚆 define elements of Hgr⁢(𝙲) and let α∈N⁡(𝙰). The following are equivalent:

  1. 1.

    The stack 𝙿𝚊𝚒𝚛¯⁢(𝚃+,𝙵−)α is of finite type.

  2. 2.

    The stack (𝙿𝚊𝚒𝚛¯⁢(𝚃+,𝙵+)∗𝚆¯)α is of finite type.

  3. 3.

    The stack (𝚆¯∗𝙿𝚊𝚒𝚛¯⁢(𝚃−,𝙵−))α is of finite type.

Definition 11.5.4.

A full subcategory 𝚆⊂𝙲𝚘𝚑≤1⁢(𝒳) is log-able if

  1. 1.

    𝚆 is closed under direct sums and summands;

  2. 2.

    𝚆 defines an element of Hgr⁢(𝙲);

  3. 3.

    If α∈N⁡(𝒳), there are only finitely many ways to write α=α1+⋯+αn, where each αi represents a nonzero element in 𝚆.

Theorem 11.5.5 (No-poles).

If 𝚆 is log-able, then

We are finally ready to do wall-crossing, but we need one more definition.

Definition 11.5.6.

Let (𝚃±,𝙵±) be open torsion pairs on 𝙲𝚘𝚑≤1⁢(𝒳) such that 𝚃+⊂𝚃−. Let 𝚆=𝚃−∩𝙵+. These torsion pairs are wall-crossing material if 𝚆 is log-able and the categories 𝙿𝚊𝚒𝚛⁡(𝚃+,𝙵−),𝙿𝚊𝚒𝚛⁡(𝚃−,𝙵+),𝙿𝚊𝚒𝚛⁡(𝚃+,𝙵−) define elements of Hgr⁢(𝙲).

Theorem 11.5.7.

Suppose (𝚃±,𝙵±) are open torsion pairs with 𝚃+⊂𝚃− which are wall-crossing material. Then w≔I⁡((𝕃−1)⁢log⁡[𝚆¯]) is well-defined and

Proof of DT/PT correspondence.

Let 𝚃DT=0,𝙵DT=𝙲𝚘𝚑≤1⁢(𝒳),𝚆=𝙲𝚘𝚑0⁢(𝚇). Then the torsion pairs (𝚃DT,𝙵DT),(𝚃PT,𝙵PT) are wall-crossing material, so we can apply the previous theorem. Isolating the terms with class β and noting that (𝚃DT,𝙵DT)-pairs are ideal sheaves I⁡[1], we obtain

Now let c∈N0⁢(𝒳). Applying the McKay equivalence, we have Ψ⁡(c)∈N≤1⁢(Y). Because β is multi-regular, Ψ⁡(β,c′)∈N≤1⁢(Y) for all c′∈N0⁢(𝒳). But then the Euler pairing is trivial on N≤1⁢(Y), so

Next, write w=∑c∈N0⁢(𝒳)wc⁢qc. We then have {w,zβ⁢qc′}=0, so

Combining this with the first DT-PT identity, we have

Finally, noting that PT⁢(𝒳)0=1 because 𝒪𝒳⁢[1] is the only stable pair with β=0, we obtain

12 Patrick (Apr 28): The DT crepant resolution conjecture

12.1 Brief review

Recall that 𝒳 is a projective Calabi-Yau 3-orbifold (where we require H1⁢(𝒪𝒳=0)), that X is the coarse moduli space (which is a projective, Gorenstein, Calabi-Yau variety with at worst quotient singularities), and Y was a distinguished crepant resolution of X. Also recall that we have derived equivalences Φ:D⁡(Y)↔D⁡(𝒳):Ψ.

Let 𝙲 be the category 𝙲𝚘𝚑⁡(𝒳) tilted at the torsion pair (𝙲𝚘𝚑≤1⁢(𝒳),𝙲𝚘𝚑≥2⁢(𝒳)). We consider the graded motivic Hall algebra Hgr⁢(𝙲), which is a module over K⁡(𝚂𝚝ℂ). Also, recall the category

Finally, recall the integration map I:Hgr,sc⁢(𝙲)→{∑α∈N⁡(𝒳)nα⁢qα}, where Hsc⁢(𝙲) is a quotient of the algebra

of regular elements.

12.2 Stability conditions

Fix an ample class ω∈N1⁢(Y) and an ample line bundle A on X.

Definition 12.2.1.

A stability condition on 𝙲𝚘𝚑≤1⁢(𝒳) consists of a slope function μ:N≤1⁢(𝒳)→S to a totally ordered set (S,<) such that if

then either μ⁡(A)>μ⁡(B)>μ⁡(C) or μ⁡(A)<μ⁡(B)<μ⁡(C) or μ⁡(A)=μ⁡(B)=μ⁡(C) and every F∈𝙲𝚘𝚑≤1⁢(𝒳) has a Harder-Narasimhan filtration 0=F0⊂⋯⊂Fn=F.

We will now define a number of stability conditions. First, we fix a “generating” vector bundle V (where every coherent sheaf on 𝒳) is locally a quotient of V⊕n for some n. We can assume that V=V∨ (by taking V⊕V∨). Now we define a modified Hilbert polynomial for a sheaf F by

Definition 12.2.2.

Define the Nironi slope of F to be

if F∉𝙲𝚘𝚑0⁢(𝒳) and ν⁡(F)=∞ otherwise. Also write ν+⁢(F),ν−⁢(F) for the slopes of the Harder-Narasimhan factor of F with largest (resp. smallest) slope.

Definition 12.2.3.

Define the stability condition ζ on N1eff⁢(𝒳)∖0 by

for β=0 and ζ⁡(0,c)=(∞,∞). Here, we use the lexicographical ordering on (−∞,∞].

For a stability condition μ and s∈S, define a torsion pair by

Also, call the category of (𝚃μ,s,𝙵μ,s)-pairs 𝙿μ,s. Finally, define the category of semistable sheaves of slope s by ℳμss⁢(s).

In order to control DT-like invariants coming from stability conditions, we need our categories of semistable objects and of pairs to satisfy openness and boundedness properties.

Proposition 12.2.4.
  1. 1.

    For any δ∈ℝ, the torsion pair (𝚃ν,δ,𝙵ν,δ) is open.

  2. 2.

    For any (γ,η)∈ℝ>0×ℝ, the torsion pair (𝚃ζ,(γ,η),𝙵ζ,(γ,η)) is open. In addition, the moduli stack ℳ¯ζss⁢(a,b)⊂𝙲𝚘𝚑¯≤1⁢(𝒳) is open for any (a,b)∈ℝ2.

We will now discuss boundedness. For any real number γ>0, define the function

This will control the series expansion of the rational function fβ⁢(q), where PT⁢(𝒳)β is the expansion of fβ⁢(q) in ℚ⁢[N0⁢(𝒳)]deg (this means roughly that degree is bounded below).

Definition 12.2.5.

Let S⊂N0⁢(𝒳) and L:N0⁢(𝒳)→ℝ be a homomorphism. Then S is L-bounded if the set

is finite for every M∈ℝ. We also say that S is weakly L-bounded if the set

is finite for every M∈ℝ.

The main results about semistable sheaves and pairs are the following. Recall that a category 𝚆 is log-able if (𝕃−1)⁢log⁡[𝚆¯]∈Hgr,reg⁢(𝙲).

Proposition 12.2.6.

Let (a,b)∈ℝ2. The set

is Lγ-bounded. Moreover, the category ℳγss⁢(a,b) is log-able.

Proposition 12.2.7.

For any (γ,η)∈ℝ>0×ℝ, the set

is Lγ-bounded. Moreover, the stack 𝙿¯ζ,(γ,η)⁢(β,c) is of finite type.

Corollary 12.2.8.

The category 𝙿ζ,(γ,η) defines an element of Hgr⁢(𝙲).

Finally, we will locate regions in which the notion of a (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pair is constant.

Lemma 12.2.9.

Let β∈N1⁢(𝒳). 𝚃γ,(γ,η)∩𝙲𝚘𝚑≤1⁢(𝒳)≤β and 𝙵ζ,(γ,η)∩𝙲𝚘𝚑≤1⁢(𝒳)≤β are constant on the components of (ℝ>0×ℝ)∖(Vβ×ℝ), where

For γ∈Vβ, the categories are locally constant on {γ}×ℝ∖Wβ, where Wβ=1ℓ⁡(β)!⁢ℤ.

12.3 DT-like invariants

We define DT-like invariants counting objects in the categories that we have defined. Once we do this, we will cross our infinitely many walls.

Recall that ℳζss⁢(a,b) is log-able for any (a,b)∈ℝ2. Therefore, we have an element

Therefore, we can define DT-type (Joyce-Song) invariants by

Now let (γ,η)∈ℝ>0×ℝ be away from the walls. By a result of Abramovich-Corti-Vistoli, there is an element (𝕃−1)⁢[𝙿¯ζ,(γ,η)⁢(β,c)]∈Hreg⁢(𝙲). Then we can define DT-type invariants by

Finally, we can form generating series

Lemma 12.3.1.

Let β∈N1⁢(𝒳) and let γ>γ′ for all γ′∈Vβ. An object E∈𝙰 of class (−1,β,c) is a (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pair if and only if it is a PT stable pair.

The following is a technical lemma whose proof requires the Hard Lefschetz condition.

Lemma 12.3.2.

For every γ∈Vβ, there is a unique class βγ∈N1⁢(𝒳) with 0<βγ≤β such that Lγ⁢(A⋅βγ)=0. Class the class cγ≔A⋅βγ∈N0⁢(𝒳).

Now we will discuss wall-crossing. Once we cross all of the walls in Vβ, we will have Bryan-Steinberg invariants, which we will define later. First, we need to understand what happens when we reach a wall γ∈Vβ. Note that the set

is finite for any x∈ℝ, so we can define

Lemma 12.3.3.

Let γ∈Vβ, E∈𝙰 be of class (−1,β,c), and let

If η>ηγ,(β,c)+, then E is a (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pair iff E is a (𝚃ζ,(γ+ε,η),𝙵ζ,(γ+ε,η))-pair. If η<ηγ,(β,c)−, then E is a (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pair iff E is a (𝚃ζ,(γ−ε,η),𝙵ζ,(γ−ε,η))-pair.

This tells us that on each wall γ∈Vβ, we can enter {γ}×ℝ at ∞ from the right and then leave the wall to the left at −∞. Now we need to understand what happens at the walls Wβ as we move from ∞ to −∞.

Proposition 12.3.4.

Let β∈N1⁢(𝒳), γ∈Vβ, and η∈Wβ. Then

Now define the series

Lemma 12.3.5.

Let β∈N1⁢(𝒳), c0∈N0⁢(𝒳), γ∈Vβ, and η0≤−ℓ⁡(β). Then

is a rational function of degree less than deg⁡(β,0)+Mβ,γ,Lγ⁢(c0)++n0⁢ℓ⁢(β)+ℓ⁢(β)2.

Taking n0→−∞, we obtain

Corollary 12.3.6.

Let β,c0,γ be as above. Then DT(β,c0+ℤ⁢cγ)ζ,(γ,∞) and DT(β,c0+ℤ⁢cγ)ζ,(γ,−∞) are equal as rational functions.

Theorem 12.3.7.

Let β∈N1⁢(𝒳), γ∈R>0∖Vβ, η∈ℝ. Then DTβζ,(γ,η) is the expansion of fβ⁢(q) in ℤ⁢[N0⁢(𝒳)]Lγ.

12.4 Bryan-Steinberg invariants

In order to have a crepant resolution conjecture, we need some kind of enumerative invariants on Y. Define

Then define 𝙵f=𝚃f⟂. We can define a Bryan-Steinberg pair (F,s) as a (𝚃f,𝙵f)-pair in 𝙰Y. Equivalently, we have

Definition 12.4.1.

A Bryan-Steinberg pair (F,s) consists of F∈𝙲𝚘𝚑≤1⁢(Y) and s∈H0⁢(Y,F) such that R⁢f∗⁢coker⁡(s)∈𝙲𝚘𝚑0⁢(X) and F admits no maps from elements of 𝚃f.

For a class (β,n)∈N1⁢(Y)⊕ℤ, let 𝙿¯BS⁢(β,n) be the moduli stack of Bryan-Steinberg pairs of class (−1,β,n)∈ℤ⊕N≤1⁢(Y). Then we can define the BS invariant BS⁢(Y/X)(β,n) via the Behrend function.

Before we continue, we will say a little but more about the McKay correspondence. Define the category 𝙿𝚎𝚛⁡(Y/X) to be the category of complexes E∈D⁡(Y) such that R⁢f∗⁢(E)∈𝙲𝚘𝚑⁡(X) and such that for any F∈𝙲𝚘𝚑⁡(Y) with R⁢f∗⁢F=0, Hom⁡(F⁡[1],E)=0 (called perverse coherent sheaves).

Proposition 12.4.2.

The equivalence Φ:D⁡(Y)→D⁡(𝒳) restricts to an equivalence of abelian categories 𝙿𝚎𝚛⁡(Y/X)≃𝙲𝚘𝚑⁡(𝒳).

We now need to relate Bryan-Steinberg pairs to objects living on 𝒳. We first define a new torsion pair.

Definition 12.4.3.

Let 𝚃ζ,0⊂𝙲𝚘𝚑≤1⁢(𝒳) denote the subcategory of sheaves T such that if T↠Q, then either Q∈𝙲𝚘𝚑0⁢(𝒳) or

Let 𝙵ζ,0⊂𝙲𝚘𝚑≤1⁢(𝒳) be the full subcategory on sheaves F such that if S↪F, then S has pure dimension 1 and degY⁡(ch2⁡(Ψ⁡(S⋅A))⋅ω)≥0.

The following result justifies the inclusion of ζ,0 in the subscript.

Lemma 12.4.4.

Let β∈N1⁢(𝒳). If 0<γ<minγ′∈Vβ⁡γ′, then for any η∈ℝ an object E∈𝙰 of class (−1,β,c) is a (𝚃ζ,0,𝙵ζ,0)-pair if and only if it is a (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pair.

We should think of (𝚃ζ,0,𝙵ζ,0) as being the limit of (𝚃ζ,(γ,η),𝙵ζ,(γ,η))-pairs as γ→0. Finally, we relate (𝚃ζ,0,𝙵ζ,0)-pairs to (𝚃f,𝙵f)-pairs, and as a corollary, we can relate enumerative invariants on 𝒳 to enumerative invariants on Y.

Lemma 12.4.5.

We have the following:

These give us the following:

Lemma 12.4.6.

If E is a (𝚃ζ,0,𝙵ζ,0)-pair with βE∈N1,mr⁢(𝒳), then Ψ⁡(E) is an f-stable pair. On the other hand, if E=(𝒪Y→F) is an f-stable pair, then Φ⁢E is a (𝚃ζ,0,𝙵ζ,0)-pair.

This implies that 𝙿¯BS⁢(β,n)≅𝙿¯ζ,(γ,η)⁢(Φ⁡(β,n)) for 0<γ<minγ′∈Vβ⁡γ′, so we have proven

Theorem 12.4.7 (Crepant resolution conjecture).

There exists a unique rational function fβ∈ℚ⁡(N0⁢(𝒳)) such that

  1. 1.

    The Laurent expansion of fβ with respect to deg is the series PT⁢(𝒳)β;

  2. 2.

    The Laurent expansion of fβ with respect to Lγ for 0<γ<minγ′∈Vβ⁡γ′ is the series BS⁢(Y/X)β.

Using results of Bryan-Steinberg and of the previous lecture, we have

Corollary 12.4.8 (Crepant resolution conjecture, original formulation).

There is an equality of rational functions