Patrick Lei

All notes

Integrability, enumerative geometry, quantization

Lectures by Alexei Oblomkov, Andrei Okounkov, Paolo Rossi, Alexandr Buryak and Vincent Bouchard, Simons Center, Summer 2022.

Integrability, Enumerative Geometry, and Quantization
August-September 2022
Notes by Patrick Lei
Lectures by Various

Simons Center for Geometry and Physics

Disclaimer

These notes were taken during the program 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. 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 Gaëtan Borot, Alexandr Buryak, Melissa Liu, Nikita Nekrasov, Paul Norbury, and Paolo Rossi for organizing this program.

1 Virasoro constraints in enumerative geometry (Alexei Oblomkov)

Here is an equation of a plane curve over formal power series in s, which we will call the Lambert curve:

This curve appears in the work of Oblomkov-Okounkov-Pandharipande [OOP20] which treats the descendent Gromov-Witten/Pandharipande-Thomas correspondence in the stationary, non-fully equivariant case. The main question is to consider deformations of this curve.

We will now state the main formula of interest:

This formula lives on the curve y⁢ey=w⁢ew⁢e−x/θ, where θ−2=−c2⁢(TX) for X a compact threefold. Here, we define

The an satisfy the usual Heisenberg relations [an,am]=m⁢δn+m,0.

Theorem 1.0.1 ([OOP20, Theorems 4, 5]).

There is an equality (after the standard change of variables q=−ei⁢u and up to a monomial in q) of equivariant 2-legged vertices with descendents

modulo (s1+s2)⁢(s2+s3), where

where 𝔽 is the universal stable pair on the PT moduli space. Here, we have

1.1 GW/Hurwitz correspondence for curves

1.1.1 Hurwitz theory

Hurwitz theory counts ramified covers of a curve X with specified ramification data:

Hurwitz proved that these numbers are always finite and that

where Cη=∑g∼(η)g∈ℚ⁢S⁢(d) is actually in the center of ℚ⁢S⁢(d). Then Cη acts on Lλ by the constant function fη⁢(λ)=|Cη|⁢χηλdim(λ).

Because of this, we can write

This can now be computed by a hook length formula, which corresponds to localization in the Hilbert scheme of points on ℂ2. If we fix η, then the function

is a symmetric function on (λi−i), or a so-called shifted symmetric function. If we consider the limit

this is a free algebra on functions f(i) satisfying f(i+1)⁢(λ1,…,λi,0)=f(i)⁢(λ1,…,λn). We now consider the functions

By the work of Vershik-Kerov, the shifted Schur functions can be written as

which after inversion becomes

We can also write the completed conjugacy classes C¯μ=Cμ+∑ρμ,λ⁢Cλ.

Example 1.1.1.

For example, we have (4)¯=(4)+2⁢(2,1)+54⁢(2).

1.1.2 GW/Hurwitz correspondence

Theorem 1.1.2 ([OP06a]).

There is a correspondence

between Gromov-Witten descendents and Hurwitz objects, where ω∈H2⁢(X) is the class of a point. More precisely, we have

This can be related to PT theory as follows: consider Z=ℂ2×ℙ1 with the antidiagonal action of ℂ× on ℂ2 and recall that Hℂ×∗⁢(pt)=ℂ⁡[t]. Then we have the localization formula

The left hand side becomes

where τk⁢(ω)=chk+2⁡(ω).

1.1.3 Pandharipande-Thomas theory

For a threefold Z, Pandharipande-Thomas theory [PT09] considers moduli spaces

of stable pairs, where ℱ is a pure dimension 1 sheaf on X supported on β∈H2⁢(Z) and n=χ⁡(ℱ). This has technical advantages over the older Donaldson-Thomas theory, one of them being that we don’t have to study the Hilbert scheme of points on ℂ3. If we let 𝒪ℙn⁡(Z,β)×Z→𝔽 be the universal stable pair, then we define

for any γ∈H∗⁢(Z).

Now consider Z=ℂ2×ℙ1. Then the first nonempty PT moduli space is

If we let γ∈Hℂ×⁢(ℂ2), we define

where ℐ is the universal ideal sheaf on Hilbd⁢(ℂ2)×ℂ2.

1.1.4 Idea of proof of GW/Hurwitz

We begin with a correspondence between relative Gromov-Witten theory without descendents and Hurwitz theory. Then we can degenerate our target curve with descendents to bubble out the descendents. We now need to show that

which requires us to study the equivariant Gromov-Witten theory of ℙ1.

1.1.5 Fock space

We begin by defining the infinite-dimensional space

Then the semi-infinite exterior power ⋀∞/2V has basis {v→S}, where S={S1>S2>…}⊂ℤ+12 such that

  1. (i)

    The set S+=S∖(ℤ≤0−12) is finite;

  2. (ii)

    The set S−=(ℤ≤0−12)∖S is also finite.

Then we write v→S=s1∧s2∧s3∧⋯.

If we rotate the French way of drawing Young diagrams counterclockwise by π4, we obtain the Russian way of drawing Young diagrams:

Refer to caption
Figure 1.1: Russian convention for Young diagrams.

There are two natural statistics associated to Young diagrams as in Figure 1.1, which are the size and where the vertex touches the bottom. Define

We now define the operators

Then we have

We can check that [ψi,ψj]=[ψi∗,ψj∗]=0 and ψi⁢ψj∗+ψj∗⁢ψi=δi⁢j. If we consider Ei⁢j∈𝔤⁢𝔩⁢(∞), then assigning Ei⁢j↦ψi⁢ψj∗ gives a projective representation of 𝔤⁢𝔩⁢(∞). The Casimir operator is C=∑k∈ℤ+12Ek⁢k, and this acts by

In addition, the operator H=∑k⁢Ek⁢k acts by

Define the operators

where ζ⁡(z)=ez/2−e−z/2. In particular, we have ℰr⁢(0)=∑Ek−r,k=αr for r≠0, where [αk,αr]=k⁢δk+r.

The space (⋀∞/2V)0 has two natural bases given by the vS and by ∏αki⁢v∅. The transition function between them is χμλ. This appears in the Gromov-Witten theory of ℙ1:

where the last equality uses the formula [zk]⁢(ℰ0⁢(z))⁢vλ=ℙk⁡(λ)⁢vλ. The last formula we will write in this section is the relation

1.2 Equivariant Gromov-Witten theory of ℙ1

We are interested in an equivariant version of the GW/Hurwitz correspondence, so we want to compute the Gromov-Witten invariants

where the aki are Getzler descendents, W~ is the operator in Theorem 2.1.1, and μ is ramification data over ∞.

1.2.1 Localization in Gromov-Witten theory

Focusing on the case of ℙ1, we will discuss localization. We begin with the moduli space ℳ¯g,n+m∙⁢(ℙ1,d) of disconnected stable maps, and the fixed points

with respect to the ℂ× action ξ⁡(v1,v2)=(v1,ξ⁢v2) are simply a moduli space of bipartite graphs with extra markings on the vertices (genuses gi and marked points) and the edges (degrees di⁢j) subject to various conditions, for example

Then recall that Hℂ×∗⁢(pt)=ℂ⁡[t] and Hℂ×∗⁢(ℙ1)=ℂ⁡[h,t]/h⁢(h+t).

In particular, we see that

so we need to consider integrals over this space.

One standard technique is to cut the edges in half and consider the vertices and then glue the edges together, and this puts the markings and the degree di⁢j copies of ℙ1 on the same footing. We now need to compute the contribution of a vertex v0 to our integral. We want to compute the contribution of v0 to

which is in fact

1.2.2 Hodge integrals

The main component of this computation is the H∘ terms, which are called Hodge integrals. Let

be the universal curve on the moduli space of curves. Then

is a rank g vector bundle, called the Hodge bundle. Then let λi=ci⁢(𝔼) be the Hodge classes and ψi=c1⁢(Tzi∗⁢C) be the psi classes. The Hodge integral is now

There is a connection between Hodge integrals and Hurwitz numbers in the nonequivariant case, due to Ekedahl-Lando-Schapiro-Vainshtein [ELSV99]. The main contribution of Okounkov-Pandharipande [OP06b] is to replace the Hurwitz numbers in the equivariant case. First, we consider Cg⁢(μ), which counts genus g curves with simple ramification at c1,…,cb and ramification data μ at ∞, where

by the Riemann-Hurwitz formula. By ELSV, this number is computed by the formula

Our next goal is to interpret Cg⁢(μ) in the Fock space analytically. Write

where

is the Pochammer symbol. In the case where m∈ℤ and (a,b)=(m,u⁢m), we obtain

Keeping b as the number of simple branch points, we obtain

Lemma 1.2.1 ([OP06b]).

We have the identity

where we use the formula

We will also discuss some more properties of the 𝒜a,b. First, there is the identity

and second

Example 1.2.2.

We have an explicit example of a Hodge integral

with two insertions.

1.2.3 Application to Gromov-Witten theory

Returning to the vertex contributions, we obtain

If we let Pd be the projection operator on ker⁡(H−d) and write it explicitly as

then we obtain the formula

When m=0, the formula simplifies to

1.2.4 Dressing operator

Recall that we want to compute the Gromov-Witten invariants with Getzler descendents, which are given by

It remains to describe this matrix W and the variant W~=W⁢eα1. Here, the an are defined by

so for example we have

We will work in 𝔤⁢𝔩∞⁢(V). We also write

where H=∑k⁢Ek⁢k and S=α−1=∑Ek,k−1, so S⁢H=(H+1)⁢S. Then define the operator

Both the operators 𝒜,𝒜~ live in 𝔤𝔩(V)[[z±1,u±1,t±1]. Then we will have W−1⁢A⁢W=A~.

Lemma 1.2.3.

Write D=S−1+H, Z=t⁢Su, and D~=D−12⁢(H⁢11−Z+11−Z⁢H). Then there exists a unique solution to the equation

such that

  1. 1.

    W|u=0=1;

  2. 2.

    W−1⁢D⁢W=D~;

  3. 3.

    W is upper triangular.

Sketch of proof.

We will compute the derivative

and this implies that D~|t=0=D|t=0. Some more “basic checking” completes the proof. □

Remark 1.2.4.

The operator B in the lemma was discovered by a computer and it is unclear why it appears.

Theorem 1.2.5 ([OOP20]).

We have the formula

Proof.

Define the operator

and define 𝒜~(m) analogously. In fact, we have 𝒜(m)=(𝒜(1))m and

Thus we obtain W−1⁢𝒜(1)⁢W=𝒜~(1). Differenting both sides by t, we obtain the desired result. □

1.3 GW/PT correspondence

Let X be a threefold and consider Pandharipande-Thomas theory on X, which we defined in Section 1.1.3.

Example 1.3.1.

Let X=ℙ3 and β=ℙ1. Then

and we will call the n points z1,…,zn. Then we see that

and φ has zeroes at z1,…,zn. In particular, we have

This and similar examples were worked out by Pandharipande-Thomas [PT09] and by Moreira [Mor22].

Example 1.3.2.

Consider X=ℙ1×ℂ2. We will consider the moduli space Pχ⁢(ℂ2×ℙ1,n⁢ℙ1). If we consider the torus T0 preserving the symplectic form on ℂ2, then the virtual class satisfies

1.3.1 Descendents in PT theory

Let 𝔽 be the universal stable pair on Pn⁢(X,β)×X. Then if γ∈H∗⁢(X), we define

When X=ℙ1×ℂ2, then

Then there is the formula

Of course, if ℂ× acts on ℂ2 with the antidiagonal action, we see that

where Iλ=(xλ1,xλ2⁢y,…,xλℓ⁢yℓ). Then we have 𝒪/Iλ=ℱ, and so

The S−1⁢(xs) term comes from the presence of the Todd class. Recall the definition

where θ−1=−c2⁢(TX).

Now the main formula is that for s=1, we have

On the Gromov-Witten side, we have

This allows us to solve the GW/PT correspondence and match descendents, where

This actually allows us to define negative descendents, but to match the descendents we need to write D~ in terms of the αk.

1.3.2 Matching GW and PT descendents

Our method will be to do the following:

  1. 1.

    Diagonalize D;

  2. 2.

    If A∈𝔤⁢𝔩⁢(V) is diagonal with eigenvalues

    then A acts on ⋀∞/2V by

    There is a boson/fermion correspondence between

The first step is of course to diagonalize D (for simplicity, we will set t=s=1). Clearly we can identify

Then we know

so we need to solve the differential equation D⁢f⁢(z)=λ⁢f⁢(z). The solution is

To make this work, we need λ∈ℤ, and to diagonalize D we take the basis

where g⁡(z)=−i⁢u2⁢z2+1z−12⁢log⁡(1−i⁢uz). The Lambert curve is obtained from fk, which produces the curve

Remark 1.3.3.

Recall the dressing operator W, which we defined using a differential equation. An explicit formula for W is

An upgraded version of the GW/PT correspondence is the following:

Theorem 1.3.4 ([OOP20]).

Subject to t1+t2=0, there is an explicit equality of descendent vertices

Remark 1.3.5.

Using the Pandharipande-Pixton philosophy, we can obtain equality Theorem 1.0.1. This will be enough to prove the stationary GW/PT correspondence for toric varieties.

1.3.3 GW/PT correspondence with descendents

Recall that on the Gromov-Witten side, the insertions of interest are τ0⁢(γ), while in PT theory, the corresponding insertions should be ch2⁡(γ). This correspondence was proved in [MOOP11] for any toric variety and any γ, while Pandharipande-Pixton [PP17] prove this for X a complete intersection in ℙN. There is also a conjectural equality between τk⁢([pt]) and chk+2⁡([pt]).

More generally, [MNOP06b] conjecture that there is a series of “chemical reactions” that take ∏i=1nτki⁢(γi) to an expression involving with Chern classes.

Pandharipande-Pixton [PP14] consider the operators

in the fully equivariant case. This fits into an overall picture

Theorem 1.3.6 ([PP14]).
  • •

    The operator MPT is rational in q;

  • •

    The operators MPT and MGW are upper-triangular.

  • •

    There is an equality

    for any X and any D=∏i=1mchki⁡(γi).

While MGW/PT is unknown, there is an explicit formula for some of the matrix elements, which we call C∙.

Theorem 1.3.7 ([OOP20]).

There is an explicit equality

for any toric X and stationary descendents γi∈H>0⁢(X).

We will consider the formal algebras

Then write

Then we can write

The C∘ are quite complicated, but in a simple case, we have

There is a bumping and splitting that produces things like

where Δ⋅α=∑αiL⊗αiR∈H∗⁢(X×X). We will take the αiL and αiR to have degrees dL,dR, respectively.

1.3.4 Virasoro constraints in PT theory

For a toric variety X, we can apply the GW/PT correspondence to the Virasoro constraints for GW⁡(X). This obtains Virasoro constraints for PT⁡(X), which turns out to be quite simple. The Virasoro constraints in PT theory are:

  • •

    We have

    for any γ∈H2⁢d⁢(X,ℚ). For example, R−1⁢(chi⁡(γ))=chi=1⁡(γ).

  • •

    The constant terms are

  • •

    There are also the operators LkPT=Tk+Rk satisfying

    and operators ℒkPT=LkPT+(k+1)!⁢L−1PT⁢chk+1⁡(p).

Theorem 1.3.8 ([MOOP20]).

For any D∈𝔻PT,st, we have

Conjecture 1.3.9.

The theorem holds for all descendents D∈𝔻PT.

The proof essentially amounts to checking the following formula:

Theorem 1.3.10.

There is the formula

Here, LkGW is the usual Virasoro for Gromov-Witten theory.1 The only contribution to descendents of 1 to LkGW are

Therefore, we can define L~kGW=LkGW−Tk∘ and ℒkGW=ℒ~kGW+(i⁢u)2⁢(k+1)!⁢L~−1GW.

Proposition 1.3.11.
  • •

    The operator ℒkGW preserve stationary descendents, as in

  • •

    We have the vanishing

    for all D∈𝔻GW.

Let S be a surface and X=S×ℙ1. Then

and so the virtual class is

Then if ℐ is the universal ideal on the Hilbert scheme, define

Corollary 1.3.12.

For any toric surface S and γi∈H∗⁢(S), we have

Proof.

This follows from

□

Remark 1.3.13.

A cobordism argument was used by Moreira [Mor22] to extend this to any surface S with H1⁢(S)=0.

2 Counting curves in 1, 3, and 5 dimensions (Andrei Okounkov)

The main object in Gromov-Witten theory is the moduli space ℳGW of stable maps C→X from a curve to X. This is of course highly disconnected, so the genus g component will be weighted by u2⁢g−2 while we will hide the degree component for now. One of the major ideas in this subject is to push forward cohomology classes to the moduli of curves, which forms a cohomological field theory with rich structure coming from maps between different moduli spaces of curves. Instead, we will push things forward to X. Just like we can vary the source curves in families, we can degenerate X. Also, automorphisms of X still act on our moduli space.

Our goal is to give a target space description of integrals over ℳGW. We would like a geometric theory that reproduces the same integrals and an algebraic theory that computes them in some fashion resembling a topological quantum field theory – for example Chern-Simons theory.

X1DX2X
Figure 2.1: Basic degeneration

The most basic degeneration of a variety X is to degenerate

where D is a smooth divisor shared between X1 and X2 as in Figure 2.1. If we write ℙ=ℙ⁡(𝒪D⊕NXi/D), we obtain the relation

in algebraic cobordism. As proved by Levine-Pandharipande [LP09], this relation generates all relations in algebraic cobordism. A similar type of degeneration is the expanded degeneration of Jun Li [Li01], which creates an accordion of ℙ1-bundles as in Figure 2.2.

X1X2D1D2D3Xℙℙ
Figure 2.2: Expanded degeneration

This fits well into the usual TQFT formalism where the boundary divisor D=⨆Di is associated to a vector space ℋ⁡(D)=⨂ℋ⁡(Di) and the manifold X is a vector in this space. Because ℋ⁡(D) is usually the cohomology of some nice space, it has a product ∪, integration, and a form

We can impose conditions at D by either pulling back cohomology classes from D or pushing forward to a cohomology class in ℋ⁡(D).

In reality, we have more general degenerations, where we have a degeneration

where the Xi are glued along a simple normal crossings divisor. This has been extensively studied in geometry, for example in logarithmic Gromov-Witten theory, logarithmic Donaldson-Thomas theory of Maulik-Ranganathan [MR20], and Brett Parker’s “exploded manifolds” [Par12].

2.1 Counting curves in 1 dimension

Now, we will take dimX=1, so we are studying maps f:C→X from a curve. We can take X=ℙ1 or even X=ℂ with the action of ℂ×. Before we begin, we would like to discuss why we prefer dimX to be odd. If we consider X×ℂ, then

where ε∈H2⁢([pt/ℂ×]) is a coordinate on 𝖫𝗂𝖾⁡ℂ×. This can be removed by taking ε→∞, but doing so is very hard and introduces new problems. However, Mumford tells us

and thus changing dimension by 2 is very easy.

In the case of dimX=1, the geometric target space theory is just Hurwitz theory, which as in Section 1.1.1 counts ramified covers of X of degree d with specified ramification data. This ramification data is an instance of a relative condition in Gromov-Witten theory, which is as follows. We can consider the moduli space of relative stable maps ℳGW⁢(X/D), where we impose that f−1⁢(D)=∑μi⁢pi, where μ is a partition and the pi are the marked points of C.

Hurwitz theory has moduli spaces parameterizing maps C→X which are ramified covers with the specified ramification, where (X,x1,…,xn) can vary in moduli. These moduli spaces are finite over the moduli of curves, so the most interesting number is simply the degree of this map, which is the Hurwitz number. Under degenerations X↝X1∪X2, then the ramification at the node x0 should match. These are called admissible covers; see Harris-Morrison [HM98] for a reference. This gives us the relation

where ∙=∑ηζ⁡(η)⁢η⊗η∈ℋ⊗ℋ. This insertion is inverse to the the pairing on ℋ.

The general answer for Hurwitz numbers was known to Frobenius and Burnside as

This contains a blueprint for the modern understanding of all 2d TQFTs as well as handle-gluing operators.

2.1.1 More algebraic way to think about characters of S⁡(d)

There is the action of a central extension of 𝔤⁢𝔩⁢(∞) on ⋀∞/2ℂ∞, known as the Fock space. We need the operators ψk,ψk∗ defined in Section 1.1.5, but we also want the Heisenberg operators

which act on the vacuum v∅ by

We also want co consider the fermionic operators

taken without the normal ordering and ζ-regularized as in

The operator Pm acts with eigenvalue

on |λ⟩. By Vershik-Kerov and Kerov-Olshanski, the central character of μ and λ are polynomials of λ, which are ℂ⁡[p1,p2,…] and in fact is equal to

where the lower order terms are controlled by the Gromov-Witten/Hurwitz correspondence. In particular, there is the remarkable formula

while the case of the elliptic curve is simply the trace of the middle operator without the eα1 terms.

2.1.2 Gromov-Witten/Hurwitz correspondence

There are several features of this correspondence:

  1. 1.

    There are no descendents of 1∈H0⁢(X,ℚ) or H1⁢(X,ℚ) in Hurwitz theory. In Gromov-Witten theory, these can be explicitly reduced to descendents of pt∈H2⁢(X,ℚ).

  2. 2.

    Relative conditions correspond to relative conditions. This will always be true in the things that we study.

  3. 3.

    Descendents of pt∈H2⁢(X,ℚ) are equal to relative conditions glued on a ℙ1 bubble.

This reduces to computing everything in the Gromov-Witten theory of X=(ℙ1,0,∞), where one of the marked points is descendent and the other is relative.

The remarkable formula in Theorem 1.1.2 can be computed explicitly and gives a formula for the completed cycles as conjugacy classes. More generally, if we have two relative points, the invariants can be computed explicitly and are B/C-symmetric in η∪η′. In particular, for d=0, we have

and this was computed by Faber and Pandharipande [FP00].

We now want to consider the equivariant theory for ℙ1.

Theorem 2.1.1 ([OP06b]).

We have the identity

where we take a linear change of times determined by

2.1.3 Role of integrable systems

This last formula was found by Getzler [Get04] and is problematic in the ε→∞ limit. Morally, the Gromov-Witten descendents give a quantum integrable system that looks like a free boson or free fermion. By the Kyoto school of integrable systems, there is a classical integrable system for generating functions. In particular, for ℙ1 with two marked points, the degeneration gives us

which is a tau function for the 2-Toda hierarchy and in particular gives a KP hierarchy for linear Hodge integrals in degree 0.

Remark 2.1.2.

These equations differ from the equations found in 2009 by Kazarian [Kaz09], and this is not well-understood.

Fundamentally, these equations come from Plücker relations for G⁢L⁢(∞)↪End⁡(Fock). This is because in Fock⊗Fock there is an invariant operator ∑ψk⊗ψk∗ commuting with G⁢L⁢(∞). Our goal now is to understand the deformation of the following for dimX=3:

  1. 1.

    The group G⁢L⁢(∞);

  2. 2.

    The representation G⁢L⁢(∞)→End⁡(Fock);

  3. 3.

    The quantum integrable system;

  4. 4.

    The invariant operator in Fock⊗Fock.

2.2 Counting curves in 3 dimensions

Again, we are looking for a target space description of enumerative theories, but this time for threefolds. Here, we will consider Donaldson-Thomas (sheaf-counting) theories on a threefold X. The connection to Gromov-Witten theory was conjectured by Maulik-Nekrasov-Okounkov-Pandharipande [MNOP06a; MNOP06b] building on the work of Aganagic-Klemm-Marino-Vafa [AKMV05] and the connection to box counting by Okounkov-Reshetikhin-Vafa [ORV06].

2.2.1 Donaldson-Thomas theory

While Gromov-Witten theory thinks of a curve inside X as a map f:C→X, Donaldson-Thomas theory thinks of C as being cut out by the ideal sheaf

which has discrete invariants the degree β∈H2⁢(X,ℤ) and the holomorphic Euler characteristic χ⁡(𝒪C). This has a natural obstruction theory

which is unsuitable for our purposes. Instead, we interpret ℐC as a rank 1 torsion-free sheaf with

and so we obtain the obstruction theory

There are no higher Ext groups (even for orbifolds) by a result of Zhou [Zho18].

In Donaldson-Thomas theory, the descendents are ch⁡(ℑC), where ℑC is the universal sheaf on X×Hilb⁡(X). We would hope that counts in fixed degree (this does work) and fixed genus (this does not work because moduli spaces are empty for g≪0 or χ≪0) may be related, but unfortinately this is too naive. Instead, we expect a GW/DT correspondence

after the standard change of variables

where z is weighted by χ and u is weighted by genus.

All moduli spaces in Donaldson-Thomas theory are schemes, so all DT invariants are actually integers. If we want actual equality of analytic functions with the Gromov-Witten partition function, we must have rationality of the DT partition function (this is related to BPS integrality). For this all to work out, the virtual dimensions

must be the same, so X must be a threefold. Of course, our correspondence requires

  1. 1.

    Identification of descendents;

  2. 2.

    Division by degree 0 contributions in DT theory;

  3. 3.

    Prefactors like z−vir−dim2↔(−i⁢u)vir−dim.

See [Oko19] for precise formulas and additional discussion.

In DT theory, it is possible to remove the degree 0 terms geometrically using Pandharipande-Thomas theory, which was discussed in Section 1.1.3. Some parts of identifying the descendents are easy (for example primary descendents and relative conditions are fine).

2.2.2 Relative conditions in DT theory

In Gromov-Witten theory, a relative condition for a divisor D consists of specifying a condition like

If p2 and p3 are mapped to the same point with p3 tangent to D, then we have an evaluation map to the inertia stack of Sn⁢D∋2⁢f⁢(p1)+3⁢f⁢(p2), where we remember the ramification data (2,1) on f⁡(p2)=f⁡(p3). Therefore, we are really integrating orbifold cohomology classes, where classes in H∗⁢(D) are colored by integers, so we are considering

If we want to resolve the tangency, then we must degenerate X in a family and bubble off of ℙ1-bundle on D (or a whole accordion).

In Donaldson-Thomas theory, we have a map

which is regular as long as multiplying by the equation of D is injective. To ensure this, we must bubble out the accordion as above. Therefore, the relative conditions come from H∗⁢(Hilb⁡(D,n)), which has a Nakajima basis

where the vacuum vector is 1∈H0⁢(Hilb⁡(D,0)). Of course, there is an isomorphism

and so adding a subscheme of length mi along γi∈H∗⁢(D) should correspond to tangency of order mi along γi.

The GW/DT correspondence has been proved for many, but not all, threefolds, and a key step, as in Maulik-Oblomkov-Okounkov-Pandharipande [MOOP11], is to consider a fibration

where Ar is a crepant resolution of ℂ2/μr+1, where μr+1 acts with weights (1,−1). Thus X is a crepant resolution of

When r=0, then X=ℒ1⊕ℒ2→B is a local curve, so for example B×ℂ×ℂ is fine, and we should recover Hurwitz theory for ε1+ε2=0.

Remark 2.2.1.

ADE surfaces are special for many reasons, for example enhanced symmetries of H∗⁢(Hilb) or of Db⁢𝖢𝗈𝗁.

The torus fixed points on Hilb⁡(B×ℂ2) are those curves supported on B, which can be thought of having a partition on B (corresponding to the closure of the generic point) with some extra boxes. The DT analog of the Mumford relations is that if ε1+ε2=0, then the virtual class vanishes for all nonminimal cases. Of course, the fixed points of Hilb⁡(ℂ2,n) correspond to partitions. If |λ|=n, then

where a⁡(□),ℓ⁡(□) are the arm and leg lengths respectively. Then we obtain

and so the virtual class is

When ε1+ε2=0, this becomes

This clearly is related to the representation theory of the symmetric group.

Remark 2.2.2.

Subsequent to [MOOP11], a different perspective was developed, in particular for computations in K-theory. Unfortunately, we cannot say anything about a possible relation to Gromov-Witten theory.

2.2.3 Role of quantum groups

If we return to our Ar-bundle over B, we want some kind of cohomological field theory on the stack of (ℂ×)2-bundles on curves B (which can vary in moduli). This has various building blocks related to B=ℙ1 with a descendent at 0 and relative condition at ∞ with the ℂ× action. The descendent conditions correspond to

while the relative conditions correspond to

By the ideas of Nekrasov-Shatashvili [NS09], this structure should be described by a quantum group, and this idea has been realized (also in K-theory) with the quantum group 𝒰ℏ⁢(𝔤⁢𝔩⁢(r+1)̂̂) in K-theory and the Yangian for cohomology. Here, if t1,t2 are the weights of (ℂ×)2 acting on ℂ2, then ℏ=1t1⁢t2 is the weight of the symplectic form. There is a noticeable difference between 𝔤⁢𝔩 and 𝔰⁢𝔩, and it becomes more dramatic for other quivers.

Remark 2.2.3.

In order to work with these objects, it is important to use a computer, and we would like to advertise the code written by Henry Liu: https://github.com/liu-henry-hl/boxcounting.

Remark 2.2.4.

If we restrict to ε1+ε2=0, then

as a Hopf algebra, so we recover our description of Hurwitz theory.

Example 2.2.5.

Recall that the operators corresponding to ℙ1 with one descendent and two relative conditions commute. These should be Baxter Q-operators, and this is more or less true.

2.2.4 Quantum loop algebra

The quantum loop algebra 𝒰ℏ⁢(ĝ) is a Hopf algebra deformation of 𝒰⁡(𝔤⊗ℂ⁡[t±1]). The original loop algebra has a loop rotation automorphism, where if V is a representation, V⁡(a) is the precomposition with t↦a⁢t. The way we will construct our quantum group is by using the category of modules, which for a Hopf algebra have tensor products and duals. There are particularly important maps. where for ĝ, we take

and for 𝒰ℏ⁢(ĝ) we take

This R12∨⁢(a1/a2) is a rational operator, and it must depend only on a1a2 if the loop rotation is an automorphism. If we have three representations V1,V2,V3, then there are two ways to reverse the order, and the equality between them is called the Yang-Baxter equation, see Figure 2.3.

Refer to caption
Figure 2.3: Yang-Baxter equation sculpture at the Simons Center

Finally, we would like to reconstruct our quantum group given a family of R-matrices. We will do this using

We can always assume that the collection of Vi are closed under the tensor product. If we consider a physical space Vi⁢(a) and auxillary space V0⁢(u), then matrix elements in V0 give operators depending on u in Vi⁢(a). These operators will generate 𝒰ℏ and are closed under product (two auxillary spaces, one physical) and coproduct (one auxillary, two physical). Morphisms in the category of modules are maps commuting with R-matrices, and so we obtain relations in 𝒰ℏ. The R-matrix indeed gives braiding and commutation relations by the Yang-Baxter equations.

Now we will consider a family of subalgebras, called Baxter subalgebras. Let g∈End⁡(Vi⁢(ai)) be such that [g⊗g,R]=0. These come from the Cartan torus, which in geometric cases come from discrete invariants. Then we consider the transfer matrix

These commute for fixed g and all Vi,u, and so we get a family of commutative subalgebras in 𝔘ℏ⁢(ĝ) which are parameterized by a maximal torus in 𝔤.

More generally, we can have several strands on a cylinder and one strand going around as in Figure 2.4 with the same condition on g as before. But then we have commutativity with g⁡(ai)=q⁢ai, and so we get a canonical flat q-difference connection on V1⁢(a1)⊗⋯⊗Vn⁢(an), which is the qKZ equation.

Figure 2.4: Cylindrical braiding condition.

2.2.5 Some more curve counting

We now want to return to curve counting on fibrations

where for now we will specialize B=ℙ1.

We will begin with Gromov-Witten theory. For example, when r=0, then X is the total space of two line bundles on ℙ1, so above 0,∞ there are curves collapsed to a point and above the base there are copies of ℙ1 mapping with degrees μi. Therefore we have something like

The triple Hodge integral terms (Figure 2.5) also involve nonsingular boundary conditions of tangency to π−1⁢(∞) (resp. π−1⁢(0)). Recall that we can break a curve in half while adding relative boundary conditions, and if we have ℙ1 with a ℂ× action, it can be broken in half with nonsingular boundary conditions into edge−1.

μ2μ1x→xμitriple Hodge ∫ψ-classesSimple edgeterms
Figure 2.5: Triple Hodge integral terms.

On the other side, in DT theory, a half edge corresponds to a 3D partition with extra boxes.

The correspondences also include Givental’s J-function for Hilb⁡(ℂ2,n), which is of course a Nakajima quiver variety.

Of course, whenever the target is a GIT quotient (for example in the case of a Nakajima variety), there is a notion of quasimaps [CKM14], which for A0 are actually equal to Pandharipande-Thomas moduli spaces [Oko17]. For r>0, these are not PT moduli spaces, but there is still a correspondence with PT-like theories [Liu21]. The combinatorics of the PT vertex terms looks like Figure 2.6.

HilbPT
Figure 2.6: Combinatorics of one-legged PT vertex.

The half-edge solves differential and difference equations (giving quantum groups) and are a much evolved relative of the hypergeometric function of the form

where the z are the Kähler variables (solving a linear differential equation) and the product term contains the equivariant variables (solving a linear difference equation).

2.2.6 qKZ equation

Donaldson-Thomas and quasimap theories make sense in K-theory, and so we obtain a q-difference equation from quantun groups, for example the qKZ equation. which comes from Figure 2.4. Here, the strands are Vi⁢(ai) and we will call the blue strand Vk⁢(ak). Recall that

and so we have

where z lives in the Cartan torus (also the Kähler variables) and the loop rotation is T−1, where T⁢a=q⁢a.

Consider unrolling the cylinder into a hyperplane arrangement. Consider the simple example of hyperplanes x1=x⁢2+{−1,0,1} with walls like x1=x3 and x2=x3, see Figure 2.7. We can move in three directions, and in the first direction we obtain

In the second direction, we obtain

g−(1)⁢R12⁢g(1)R12⁢(a1/a2)x1=x2+1x1=x2x1=x2−1x2=x3x1=x3
Figure 2.7: Hyperplane arrangement

More generally, from each wall in our hyperplane arrangement ⟨α,x⟩+n=0, we obtain an operator Bw⁢(qn⁢xα). If we consider λ such that ⟨α,λ⟩∈ℤ for all α, then translation by λ is conjugation by 𝒪⁡(λ) of all Bw. This produces commuting q-difference operators, which are transition functions for a vector bundle of the form

For example, for 𝒰ℏ⁢(𝔤̂), we obtain the following:

  1. 1.

    First, of course the qKZ equation, which depends on z in the Cartan torus of 𝔤.

  2. 2.

    Universally, there exists a q-difference connection in z that commutes with the qKZ operator. Operators in 𝒰ℏ⁢(𝔤̂) act on V1⁢(a1)⊗⋯⊗Vn⁢(an). This was studied first by Etingof-Varchenko [EV02] for finite-dimensional Lie algebras 𝔤, where the α are simply the roots of 𝔤. They write elements

    and these are called quantum dynamical Weyl groups. More generally, Okounkov-Smirnov [OS22] consider more complicated arrangements, for example for 𝒰ℏ⁢(𝔤⁢𝔩⁢(1)̂̂), where we need all roots of of 𝔤⁢𝔩⁢(1)̂, and so the hyperplane arrangement is ℚ. This is clearly preserved by translation by ℤ, so the root subalgebras are

The way to prove the above is to

  1. 1.

    Find qKZ among the q-difference equations in the equivariant variables;

  2. 2.

    Use commutation with qKZ and other known results to conclude.

We know that K⁡(Hilb⁡(ℂ2,pts)) is a module over a quantum group. Then

where ideal⊕ideal lives inside the space M⁡(2) of rank 2 framed sheaves on ℂ2. Here, a framed sheaf is a sheaf ℱ on ℙ2 such that

Clearly, there is an action of G⁢L⁢(2) on framed sheaves, so the sum of two ideals is the fixed locus of the maximal torus. Normalizing one of the weights of the torus, the locus of two ideals is the fixed locus of matrices of the form

and so shifts in a give the qKZ equation.

Remark 2.2.6.

The Hilbert scheme Hilb⁡(ℂ2,n) is the Nakajima quiver variety M⁡(n,1) for the quiver with one vertex and one loop.

Then the qKZ equation is

and comes from a twist by 𝒪⁡(1) in the framing. In the case of Ar for r>0, there are more fixed points of a and there are “constant sections” whose degrees are encoded in the Kähler variables.

2.2.7 Stable envelopes

We would now like a map

such that if we insert the relative conditions in the Ar-fibration, all contributions cancel except those of the constant curves. Once this stable envelope exists, we will have

Vanishing is easier in cohomology [MO19], but is more difficult in K-theory. The reason for this is that vanishing in K-theory is usually proved using rigidity (properness, which implies the result is a polynomial in a, and a degree bound, which implies the result has negative degree).

The stable envelope is supported on the full attracting locus inside Y×Ya (or repelling for Unstab), which gives properness. For degree bounds, the Newton polygon

is contained a fixed polygon.

Remark 2.2.7.

Rigidity in K-theory should come from a rigidity argument in elliptic cohomology, where K-theory is thought of as coming from a nodal degeneration of an elliptic curve.

More abstractly, let A be a torus acting on a variety Y. Then we choose a chamber in 𝖫𝗂𝖾⁡A, where the walls are weights on NY/YA. This is a choice of attracting and repelling directions and corresponds to ξ∈𝖫𝗂𝖾⁡A such that Yξ≠YA. Now for all adjacent strata in our chamber structure, there should be an attaching map, and the attaching maps should commute. These can be constructed with some assumptions on Y, and now the R-matrix is simply the composition of one attaching operator and the inverse of the other. By the original commutation relation, we obtain the Yang-Baxter equation.

Example 2.2.8.

If we consider the rank r sheaves M⁡(r) and write r=r1+r2+r3, then there is an action of (ℂ×)3. Therefore,

More generally, for Nakajima varieties we can consider w=∑w(i).

2.2.8 From enumerative geometry to quantum integrable systems

A quantum integrable system is controlled by a commutative subalgebra of

One interesting problem is to find the eigenvectors and eigenvalues of this algebra. The Hilbert scheme of Ar can be replaced by some Nakajima quiver variety

where the framing w is fixed.

The quantum integrable system is actually the q→1 limit of some flat q-difference connection. In this setting, the problem of finding eigenthings is the problem of finding an integral solution to the q-difference connection.

In the classical setting, consider a hyperplane arrangement and take hyperplanes

where the β1 is fixed and the yi vary, then we obtain integrals that look like

where the ωi⁢(x) are rational and the γj are homology classes on the complement of the hyperplane arrangement. This is like an Euler integral, which has the form

Our integral solves a differential equation (the Gauss-Manin connection) as well as a difference equation.

We now need to consider a q-analog. Here, the γk will live on the complement of translates of subtori inside a tori, and the exponentials become gamma functions. Thus we are now integrating

which is a q-analog of a Mellin-Barnes integral. These are fundamental solutions Ψi⁢j to the q-difference equation.

The q-difference equation has the form

In the q→1 limit, our solutions become

where ψj⁢(a) is the eigenvector with eigenvalue λj⁢(a). The eigenvectors and eigenvalues correspond to critical points of the ∫log⁡λj⁢daa term, where the eigenvalues are e∂S∂a and the eigenvectors are given by evaluating ωi⁢(x) at the critical points x of S. The ωi⁢(x) are called the off-shell Bethe eigenfunctions and also depend on some auxillary variables. Note that we need to substitute in solutions of the Bethe equations

Remark 2.2.9.

There is an important detail in which it is better to do the replacement

Geometrically, we will work with quasimaps to Nakajima quiver varieties. The fundamental solution will correspond to the ℙ1 with one nonsingular point and one relative point. On the other hand, the vertex (see [Oko17, Section 7.4]) is the Mellin-Barnes integral.

Example 2.2.10.

Consider ℒ=𝒪⁡(d) on ℙ1. This has weight w above 0, q along the ℙ1, and q−d⁢w above ∞. Therefore,

if d>0, and so the Euler class is

which is a ratio of two q-gamma functions.

The ideas in this example work more generally, and now it remains to find the elliptic term. This will come from some consideration of elliptic cohomology. The moduli space

of quasimaps (or for a general Nakajima variety) has evaluation maps

This gives us a push-pull morphism

This gives us an elliptic function ℓj⁢(x) which goes into the missing argument of the bottom-left 𝖤𝗅𝗅. Also, the ratios of gamma functions were known to Iritani in the classical setting.

Remark 2.2.11.

The consideration of elliptic cohomology does not affect the gamma function terms of the fundamental solution. The gamma function terms control the spectrum and can be observed already in the case of just one nonsingular point. This was observed already in [NS09].

2.2.9 Elliptic cohomology

Recall that

where x=c1⁢(𝒪⁡(1)). If we consider the action of (ℂ×)n, we can consider the equivariant cohomology

This comes from the observation that ℙn has n ways to lift x to an equivariant class (consider each of the toric divisors). If we consider

this is the union of hyperplanes x=−ai, which are the cohomologies of the fixed points. Lifting to k-theory, we have the same picture after replacing x=−ai with x=ai−1 and taking a∈A instead of x∈𝖫𝗂𝖾⁡(A).

Returning to the case of Nakajima varieties, the ring

is generated by the tautological bundle and its exterior powers. Then its spectrum

is again a union of planes with data coming from plane partitions (which are the fixed points).

Finally, it makes sense to replace ℂ× by E=ℂ×/qℤ, where all of the exact same equations hold (subject to writing the group operation on E as multiplication). The elliptic cohomology of Hilb⁡(ℂ2,n) is drawn in Figure 2.8.

λSn⁢Ea1,a2∈E2
Figure 2.8: Elliptic cohomology of Hilb⁡(ℂ2,n)
Remark 2.2.12.

It is important to note that 𝖤𝗅𝗅G⁢(X) is a scheme and not an algebra, so it is covariant in both G and X.

Note that if G is connected, then

is the moduli space of semistable degree 0 G-bundles on the dual elliptic curve E∨. Also note that 𝖤𝗅𝗅(ℂ×)n⁢(pt)=En.

2.2.10 Computations in elliptic cohomology of Hilb⁡(ℂ2,n)

In order to compute things, we need Euler classes. Suppose p is a fixed point of A acting on X, V is a vector bundles, and A acts with weights wi on Vp. Then

Note that these theta functions are not functions but are sections of line bundles on elliptic curves. Recall that V is a pullback

This gives us a map

and on Sn⁢E there is the theta divisor Θ, which is the locus of bundles with a section. Pulling back to 𝖤𝗅𝗅⁡(X), we obtain a line bundle Θ⁡(V), which is where Euler⁡(V) lives.

Pushforwards in the elliptc theory involve Euler classes and thus twists by line bundles. If we use localization, we obtain the factors in Table 2.1.

Table 2.1: Localization in different cohomology theories
In H∗ In K∗ In 𝖤𝗅𝗅
1∏wi 1∏(1−wi−1) 1ϑ⁡(wi)

Then the pushforward X→pt gives us a map Θ⁡(T⁢X)→𝖤𝗅𝗅⁡(pt) and the inclusion of a fixed point in X gives us a map 𝖤𝗅𝗅⁡(pt)→Θ⁡(T⁢X).

Remark 2.2.13.

There is an inclusion

The Kähler variables live on the left hand side, and quantum computations in K-theory embed inside classsical computations in elliptic cohomology.

Remark 2.2.14.

In principle, one could study enumerative counts in elliptic cohomology or even in another complex oriented cohomology theory like complex cobordism (see [She16]). However, we cannot form generating functions in these situations (for example in elliptic cohomology everything lives on a different line bundle), so we want to study the difference equations directly without generating functions.

We want to actually compute with the line bundles Θ⁡(V) for vector bundles on Hilb⁡(ℂ2,n). For the tautological bundle, we have

For the tangent bundle, we obtain

Note that if we restrict t1⁢t2=1, we have a square root. Thus, we can find a polarization

Thus we will need sections of a line bundle of the form

We will call the determinant det⁡Taut≕𝒪⁡(1). Recall that taking the determinant corresponds to taking

Now recall that Pic0⁡(E)=E∨=E, so for z∈Pic0⁡(E) (the Kähler variable), we can consider the meromorphic section

of some degree 0 line bundle. More generally, we can take z∈Pic⁡(X)⊗E.

Recall that we have a map

We want to take a Lagrangian class on Hilb and extend it to the ambient stack. This is like taking a section of a line bundle 𝒮 on 𝖤𝗅𝗅⁡(Hilb) and extending it to the ambient stack. The ambient stack has a Harder-Narasimhan stratification in which the stable locus is the first piece. This allows us to take an inductive approach to the interpolation, see [Oko21]. There is a nonabelian analogue of this, see [Oko20].

One step of the induction is the following. Consider the inclusion U=X∖Attr⁡(F)↪X. Then in ordinary cohomology, we have the long exact sequence

By Atiyah-Bott, the normal bundle NX/Attr⁡(F) has a nontrivial action, so all of the connecting maps vanish. By excision, we can replace (X,U) by an embedding of F into a vector bundle V over F, which produces

In elliptic cohomology, we obtain an exact sequence of line bundles. Consider the diagram

This produces us an inclusion 𝖤𝗅𝗅⁡(V∖0)↪𝖤𝗅𝗅⁡(V), which gives us the divisor Θ⁡(V) on 𝖤𝗅𝗅⁡(V). From a section of ℒ|Θ⁡(V), we have an exact sequence

In studying sheaves on 𝖤𝗅𝗅⁡(F), we want global sections. Consider the long exact cohomology sequence

Note that because 𝒮∈Θ⁡(T12)⊗Pic0, ℒ≔𝒮⊗Θ⁡(−V) has degree 0. It is also nontrivial (sections have poles related to enumerative geometry). However, if 𝒮 is a nontrivial degree 0 line bundle on an abelian variety ℰ, it has no cohomology whatsoever.

We have now produced a stable envelope

which interpolates sections of line bundles. We also have the traditional elliptic stable envelope [AO21]

As we change attracting and repelling directions, we obtain elliptic R-matrices. The K-theoretic and cohomological versions can be obtained using degeneration.

2.2.11 Quantum q-difference equations

We would like to solve these equations using integrals. Recall that in K-theory, we have

Recall that the solutions are of the form

As q→1, the gamma functions give the Bethe equation, while the ωi⁢(x) are the q=0 limit of the ℓi⁢(x) in [AO17]. This is called the off-shell Bethe eigenfunction. For quotients by groups of the form ∏G⁢L⁢(Vi), the nonabelian stable envelope reduces to the abelian one, so there is a formula in terms of R-matrices.

Consider the X⁢X⁢Z spin chain, where we have states like ↑↓↓↑↓↓↓. Take the vacuum to be ↓⋯↓. Then the physical diagram will look like Figure 2.9.

↓↓↓↓↓↓↓↓↓↓↑↓↓↑↑↓↓↓↓↓↓↑(u1)↑(u2)↑(u3)T∗⁢Gr⁢(0,9)T∗⁢Gr⁢(0,3)T∗⁢Gr⁢(3,3)T∗⁢Gr⁢(3,9)
Figure 2.9: Example of XXZ spin chain

If the spectral variables ui solve the Bethe equation, then the bottom state is an eigenvector. This is the origin of the algebraic Bethe ansatz.

In our setting, consider the picture of Figure 2.10.

ℳ⁡(0,v)ℳ⁡(v,v)ℳ⁡(0,w)ℳ⁡(v,w)
Figure 2.10: Simple model for Nakajima varieties

The analog of all up arrows here is when v=w and the maps W→V are isomorphisms while the maps V→W are zero. This happens because the Bethe roots sit inside a maximal torus of ∏G⁢L⁢(Wi)=∏G⁢L⁢(Vi) here.

2.2.12 An open problem

In dimension 1, we computed the equivariant ℙ1 in terms of the action of G⁢L⁢(∞) on the Fock space and concluded that the ℙ1 with two marked points is a τ-function for the 2-Toda hierarchy. This comes from the action of G⁢L⁢(∞) on Fock⊗Fock, which has the action of ∑ψi⊗ψi∗

In dimension 3, we know both the equivariant ℙ1 with one marked and one nonsingular point and with one marked point and one relative point in terms of 𝒰ℏ⁢(𝔤⁢𝔩⁢(1)̂̂). In the ℏ=1 limit, we recover 𝒰⁡(𝔤⁢𝔩⁢(∞)), which corresponds to going down to 1 dimension. If we take 𝒰ℏ acting on Fock⊗Fock⁡(shift) along with a screening operator, where

what is the deformation of 2-Toda?

The case of one marked point and one relative point is rational in z [Smi16]. Three key facts are

  1. 1.

    The equivariant ℙ1 with one descendent and one relative condition vanishes for ℳ⁡(v,w) for w≫0 (the so-called large framing vanishing) [Oko17].

  2. 2.

    Splitting the equivariant ℙ1 in half and adding two nonsingular points, consider w=w0+w1, where w0 is fixed and w1 is large. Then a∈ℂ× acts on W1 by scaling. We now have

    so the descendent is a Laurent polynomial in the Chern roots of the tautological bundle V. Twisting by det⁡V, we obtain an ordinary polynomial. Localizing on ℙ1 and a and taking a→0, we reduce to computing the same descendent for w=w0.

  3. 3.

    In the a→0 limit, the fundamental solution for the q-difference equation becomes the product of an explicit fusion operator and two fundamental solutions.

3 Double ramification cycles and integrable hierarches (Paolo Rossi and Alexandr Buryak)

Note: see [Ros17] for a complete discussion.

3.1 Integrable systems

3.1.1 Notation and formalism

We will precisely define things like the KdV equation, tau functions, and higher symmetries. This comes from a formal algebraic discussion of things like dynamical systems. For a detailed reference, see [DZ01]. Define the phase space

We consider the coordinates uα:S1→ℂ as formal variables. Also, we will let x be the coordinate on S1.

Definition 3.1.1.

A differential polynomial is an element of the ring

where ukα=∂xkuα. This has a grading where uk∗ has degree k and ε has degree −1.

This is not quite a ring of functions on 𝒫, so we will correct this. Define the operator

This gives us a space ℱ̂=𝒜̂/(Im⁢∂x⊕ℂ) of local functionals and should be thought of as integrating over the S1. There is a way to reintroduce a ring structure on local functionals, but it will not be necessary for our purposes.

As changes of coordinates on the phase space, consider the Miura transformations

For some amount of invertibility, we require

Remark 3.1.2.

Miura transformations are invertible in the sense that they can be solved for u∗ as a system of ODEs order by order in ε.

We may define Fourier coordinates on 𝒫 by taking the substitution

Here, the pbα are the Fourier coordinates. We force the coefficients to have polynomial dependence on b and to sum to 0 in analogy with the Fourier transform.

3.1.2 Poisson structure

We will now place a Poisson structure

on 𝒫 or equivalently on ℱ̂. Here, we take

where f¯=∫f⁢dx and the variational derivatives are defined by

and the Poisson operator Kα⁢β is defined by

This is not quite a Poisson structure in general as we still need to impose skew-symmetry and the Jacobi identity.

3.1.3 System of evolutionary PDEs

A system of evolutionary PDEs (or a vector field on 𝒫) is a differential equation

where Qα is a differential polynomial. Analogously, recall that a vector field X on a finite-dimensional manifold can be written as either

or as

Define the operator

and its adjoint

This is related to coordinate transformations on tensors by the formula

Theorem 3.1.3 ([Get02]).

There exists a Miura transformation bringing any Poisson structure of the form

with gα⁢β a nondegenerate flat metric on ℂn to the form

where ηα⁢β nondegenerate constant symmetric. Note that the Jacobi identity imposes that the bγα⁢β are the Christoffel symbols for the Levi-Cevita connection of gα⁢β, see [DN83].

Example 3.1.4 (KdV equation).

Let 𝒫={u:S1→ℂ}. Then the ring of differential polynomials is

and we will consider the vector field

This is a Hamiltonian operator for the Poisson structure, where K=∂x. The Hamiltonian that produces this system is

Now we can compute

which is exactly the differential polynomial we want.

We now want to lift Kα⁢β to a map

There is a standard way to do such a thing by taking

In Example 3.1.4, we computed the vector field K⁢(∂h¯∂u)={u,h¯}.

Example 3.1.5 (2-KdV equation).

Consider 𝒫={u:S1→ℂ2} and the Hamiltonian

Also, consider the Poisson operator

The resulting PDE is called the 2-KdV equation.

3.1.4 Integrable systems and tau-structures

Consider a collection of Hamiltonians

that commute with respect to the Poisson structure, i.e.

for all α1,α2=1,…,N and d1,d2≥0. We obtain the differential equation

or symmetry of the mixed partials. Then our differential equations can be written as

Commutativity guarantees the existence of a formal solution

with the initial condition

We will now discuss τ-structures. Assume that

Then a τ-structure is a choice of hβ,d∈𝒜̂[0] for β=1,…,N and d≥−1 such that

  1. 1.

    Kα⁢μ⁢𝛿h¯β,−1𝛿uμ=0 for all β,α=1,…,N;

  2. 2.

    The integrals of the hβ,d satisfy

  3. 3.

    There is an equality of Poisson brackets

    for all β1,β2⁢d1,d2.

Assuming that a tau-structure exists, we see that

The third condition implies that

Remark 3.1.6.

For all solutions

of the integrable system, there exists a unique F∈ℂ⁡[[t∗∗,ε]] such that

For example, the definition of the tau-structure tells us that

In particular, we obtain

Assume that Kuα⁢β|ε=0=ηα⁢β⁢∂x. Then

so in particular

Definition 3.1.7.

Given a tau-structure, its system of normal coordinates is

Note that

3.1.5 KdV hierarchy revisited

Recall the Hamiltonian

and consider K=∂x. Then we see that

We will now give a new construction of the KdV hierarchy.

Theorem 3.1.8 ([BR16]).

Define gd∈𝒜̂[0] for d≥−1 via g−1=u and

where we choose

Then

Note that

Lemma 3.1.9.

Let Kα⁢β=ηα⁢β⁢∂x be in the full Getzler normal form and

Then

is a tau-structure.

Proof.

We will suppress the Greek indices. We now compute

By antisymmetry, we obtain a tau-structure. □

Using this tau-structure, we can write the following well-known result:

Theorem 3.1.10 ([Kon92]).

The τ-function F⁡(t∗,ε) for the solution of the KdV hierarchy with initial condition

is

We would like to consider changes of coordinates that preserve tau-structures. These are called normal Miura transformations. Let

be an integrable system with tau-structure. Recall that we have normal coordinates

Fix a differential polynomial P∈𝒜̂[−2]. The normal Miura transformation generated by P is

Then

is a tau-structure for

Remark 3.1.11.

Note that for all solutions, the tau-function changes under normal Miura transformations as

3.2 Double ramification hierarchy

Recall that if Kα⁢β=ηα⁢β⁢∂x+O⁡(ε), then

Because the integrand is now a total derivative in x, commutativity of the Hamiltonians (in the dispersionless limit) requires that

is invariant under exchanging β,γ. This encodes relations on ℳ¯0,4, where points 2,3 have insertions ψd1,ψd2, which should be invariant under exchange of 1,4. Therefore, we should think of commutativity under exchange of indices as the WDVV equation. However, this approach only works in genus 0.

We will now define

and for any cohomological field theory,

3.2.1 Double ramification cycle

Define the double ramification cycle DRg⁢(a1,…,an)∈ℛg⊂H2⁢g⁢(ℳ¯g,n,ℂ) as follows. First, define

This has a morphism

and so we may define

This is a tautological cycle, and DRg⁢(A)=0 unless ∑ai=0. Recently, an explicit formula was found by Janda-Pandharipande-Pixton-Zvonkine [JPPZ17].

Now define a compact type version

where aJ=∑i∈Jai and δhJ is the cycle corresponding to the stable graph with genus h with J marked points and genus g−h with Jc marked points. Note δ0{i}=−ψi. Also note that

Instead of working with ℳ0,n¯, we need to work with the Losev-Manin compactification

of the moduli space of curves. Now we have a diagram

The Losev-Manin compactifications still have the WDVV equation, so we have the formula

Here, x,y are internal marked points on the two components, and our expression must be invariant under exchanging x,y.

3.2.2 Formation of integrable hierarchy

We are now ready to define

for any cohomological field theory. We now attach the Poisson structure

We can see that

It is unclear that this vanishes, and in fact we will need to quantize our integrable hierarchy or correct our formulas.

To make corrections in the classical case, we will add λ-classes to the WDVV equation as follows:

Then our Hamiltonians become

Our system is related to the formalism of the previous section by taking the Fourier expansion

Now our Poisson bracket takes the form

We may now rewrite our Hamiltonians as differential polynomials by

Note here that λg⁢DRg⁢(a1,…,an)∈ℛg⁢[a1,…,an]2⁢g.

Lemma 3.2.1.

If Cg,n is a cohomological field theory with unit e1, then

Corollary 3.2.2.

The DR hierarchy is τ-symmetric with

This leads naturally to the following conjecture:

Conjecture 3.2.3 ([BDGR18]).

The DR hierarchy is normal Miura equivalent to the Dubrovin-Zhang hierarchy.

3.2.3 Quantization

Define

The Poisson bracket is defined by

To compute the commutator of two local functionals, we need to take a deformation of the Moyal product for the old Poisson structure as follows:

and then take

Theorem 3.2.4.

The Hamiltonians G¯α,d commute:

3.2.4 DR hierarchy for DR cycles

Recall that the double ramification cycles satisfy the following properties:

  • •

    DR0⁢(a1,…,an)=1∈H0⁢(ℳ¯g,n). Therefore, DR0⁢(a,b,0)=δa+b,0∈H0⁢(ℳ¯0,3);

  • •

    DRg⁢(0,…,0)=(−1)g⁢λg;

  • •

    If π:ℳ¯g,n+1→ℳ¯g,n is the forgetful map, then

  • •

    If gl:ℳ¯g1,n1+1×ℳ¯g2,n2+1→ℳ¯g,n is the gluing map, then

We would now like to compute the flows corresponding to this partial CohFT. Here, set

Substituting our double ramification cycles for the Cg,n terms, this integral vanishes unless n=3.

Theorem 3.2.5 ([BR21]).

There is the explicit formula

for all g≥0.

Proof.

We will begin with Hain’s formula

where

Moreover, we have

Then the Hodge classes vanish outside of the compact type locus, so we now need to prove that

Our main tool is an explicit formula for as⁢ψs⁢DRg⁢(a1,…,an) analogous to our formula for the WDVV equation:

where

If we add Hodge classes, this formula simplifies to become

We are now ready to compute

The first term becomes

The integral splits as

The first integral in this product vanishes unless h1=g1−1,h1=0, in which case it is is

and therefore the overall integral becomes

We can use the same ideas for the other terms, and we obtain

We can use the same ideas to compute

Finishing the computation, we obtain

For any a∈ℤ, we now obtain

Using the Fourier expansion, we now obtain

Here, we see the Moyal star-product

and therefore we obtain

More generally, we have

3.2.5 Noncommutative integrable hierarchy

We now want to consider

Then the flows ∂∂td0 of the DR hierarchy begin with

which is a kind of noncommutative KdV equation. The higher flows are the higher flows of the ncKdV hierarchy, which can be described using the Lax representation. This leads naturally to the next problem:

Problem 3.2.6.

How can we describe the flows ∂∂tdα with α≠0?

There are of course other questions, including the relationship with the Dubrovin-Zhang hierarchy as in Conjecture 3.2.3, which implies the following conjecture.

Conjecture 3.2.7 ([BR22]).

Intersection numbers with Pixton’s class give a solution of the noncommutative KdV hierarchy.

4 Airy ideals and topological recursion: an investigative tool for enumerative geometry, VOAs, and gauge theories (Vincent Bouchard)

Topological recursion was invented by Eynard-Orantin in their work [EO07] on matrix models in 2007 and then reformulated in terms of Airy structures (or Airy ideals) by Kontsevich-Soibelman [KS18] in 2017. This is a correspondence between geometry, algebra, and differential equations.

4.1 Witten’s conjecture

This story began with theories of 2d quantum gravity. Define the partition function

where we define

To simplify our notation, we will write

Theorem 4.1.1 ([Kon92]).

The function

is the unique solution to the KdV hierarchy with initial condition u⁡(t0,0,…)=t0.

4.1.1 Reformulation as differential constraints for 𝒵

For k≥−1, define the operator

where

Now we can reformulate Theorem 4.1.1 as

Theorem.

𝒵 is the unique solution to the constraints Lk⁢𝒵=0 for all k≥−1.

Note that the Lk satisfy the relation

Thus we have a representation of a subalgebra of the Virasoro algebra, which is defined by

Note that Lk has the expansion

which makes it easy to see uniqueness once we have existence.

4.1.2 Potential generalizations

The first possible generalization is the Virasoro conjecture [EHX97], where we remove uniqueness of the solution and preserve the geometry (using Gromov-Witten theory of projective varieties) and the Virasoro algebra. This has been proved in several cases, including toric varieties [Giv01; Iri07], flag varieties [JK20], Grassmannians [BCK05], varieties with semisimple quantum cohomology [Tel12], Calabi-Yau varieties [Get99], and toric bundles over varieties satisfying the Virasoro conjecture [CGT24]. There is also progress on Virasoro constraints for sheaf-counting theories, see [MOOP20].

The second possible generalization is to keep existence and uniqueness of the solution 𝒵 to the constraints Hi⁢𝒵=0. More precisely, we want to find constraints on Hi such that Hi⁢𝒵=0 always has a unique solution of the form

with the initial condition 𝒵⁢_⁢xA=0=1. This is the generalization that we will take.

4.2 Airy ideals

4.2.1 Weyl algebra and modules

Let A be a (possibly infinite) indexing set. The Weyl algebra 𝒟A is defined as

This is only a filtered algebra, so it is difficult to form power series. We will use the Rees construction to turn it into a graded algebra.

Here, we will use the filtration

where

For example, F2⁢𝒟A contains terms like x1 and x1⁢∂2 but not terms like x12⁢∂2.

Now we may define the Rees Weyl algebra

where deg⁡ℏ=1, and the completion

This is in fact a good way to handle power series, and we will consider Hi∈𝒟̂Aℏ.

We will now consider the polynomial module ℳA, which has the following properties:

  • •

    ℳA is generated by 1∈ℳA;

  • •

    The annihilator Ann𝒟A⁡(1) is the left ideal generated by ∂i;

  • •

    We recover

We can now introduce ℏ directly into our module, but we need a filtration on ℳ such that

A natural choice is to let Fi⁢ℳA be polynomials of degree at most i. We can now repeat the Rees construction by considering

The partition function 𝒵 will live in a twist of ℳ̂Aℏ rather than in the module itself. Some properties of ℏ are as follows:

  • •

    ℳ̂Aℏ is a cyclic left 𝒟̂Aℏ-module generated by 1;

  • •

    Ann𝒟̂Aℏ⁡(1) is the left ideal ℐcan generated by ℏ⁢∂i;

  • •

    We can recover the module by

We can now reformulate our problem as

Problem 4.2.1.

What conditions should a left ideal ℐ⊂𝒟̂Aℏ satisfy such that ℐ⋅𝒵 has a unique solution of the form

4.2.2 Characterization of ideals solving our problem

From our expression for 𝒵, it is clear that the operators

satisfy H¯i⁢𝒵=0. Denote the ideal generated by the H¯i by ℐ¯, which is in fact Ann𝒟̂Aℏ⁡(𝒵).

Definition 4.2.2.

Define an automorphism Φ:𝒟̂Aℏ→𝒟̂Aℏ by

Note that this is a valid construction because [H¯i,ℏ⁢xj]=ℏ2⁢δi⁢j and [H¯i,H¯j]=0.

Given the twist Φ, define the twisted module ℳ̂ℏAΦ by the twisted product

This satisfies the same properties as the untwisted version. Because Φ⁡(P)=𝒵⋅P⋅𝒵−1, we can define a module of exponential type by 𝒵⁢ℳAℏ̂, which is cyclic and generated by 𝒵. Clearly we have

Definition 4.2.3 (Airy ideal).

Let ℐ¯⊆𝒟̂Aℏ be a left ideal. It is Airy if it is generated by (H¯i)i∈A.

Theorem 4.2.4.

The function 𝒵 defined in Equation 4.1 is the unique solution to ℐ¯⋅𝒵=0 with 𝒵|xA=0=1.

Of course, any ideal can have many different generating sets. How can we recognize Airy ideals in a more intrinsic manner?

Theorem 4.2.5.

[KS18] Let ℐ⊆𝒟̂Aℏ be a left ideal. Suppose that:

  1. 1.

    ℐ is generated by (Hi)i∈A such that Hi=ℏ⁢∂i+O⁡(ℏ2);

  2. 2.

    [ℐ,ℐ]⊂ℏ2⁢ℐ.

Then ℐ is Airy.

Remark 4.2.6.

Clearly [ℐ,ℐ]⊆ℐ for any left ideal ℐ. In addition, in our situation we always have [ℐ,ℐ]⊆ℏ2⁢𝒟̂Aℏ. However, in general we do not have [ℐ,ℐ]⊂ℏ2⁢ℐ.

Example 4.2.7.

Let ℐ⊂𝒟̂Aℏ be the ideal generated by the Kontsevich-Witten differential operators

For the second condition, note the (shifted) Virasoro relations

Computing the full commutator of any A,B∈ℐ, we have

and clearly this lands in ℏ2⁢ℐ.

Sketch of proof of Theorem 4.2.5.

We need to prove that there exists a generating set (H¯)i∈A taking the desired form.

  1. 1.

    First, we find H¯i∈ℐ of the form

  2. 2.

    Next, if ℐ¯ is the ideal generated by the H¯i, we need to show that ℐ⊆ℐ¯. Therefore, ℐ=ℐ¯.

  3. 3.

    Finally, we prove that [H¯i,H¯j]=0 for all i,j.

To complete the first step, start with the Hi=ℏ⁢∂i+O⁡(ℏ2). We can then replace

for the right-most derivative, and now we have

where Q∈ℐ. Simply set H¯i to be the first two terms of Hi.

To complete the second step, consider the original generators Hi=ℏi+O⁡(ℏ2). Then replace

and so Hi is the sum of a polynomial and Q¯, where Q¯∈I¯. The following lemma completes this step:

Lemma 4.2.8.

There are no nonzero polynomials in ℐ.

To complete the third step, we use brute force to obtain

By the lemma, we are finished. □

Trivially, we see that

  • •

    ℐ=Φ⁡(ℐcan) for some transvection Φ;

  • •

    𝒟̂Aℏ/ℐ≃ℳℏAΦ≃ℳAℏ⁢𝒵;

  • •

    𝒵 is uniquely determined with 𝒵|xA=0=1.

4.2.3 Are Airy ideals interesting?

Returning to the algebra-geometry-integrable systems correspondence, we want either that 𝒵 is the τ-function for some integrable system or that Fg,n are some kind of interesting enumerative invariants.

Definition 4.2.9.

We say that an Airy ideal ℐ⊆𝒟̂Aℏ is ℏ-polynomial if it is generated by (Hi)i∈A that are ℏ-polynomial. In addition, we say that ℐ is ℏ-finite if there exists N∈ℕ such that the ℏ-degree of all Hi is at most N.

Airy ideals are integrable when they are ℏ-finite. If

then Hi⁢𝒵=0 has the solution

Example 4.2.10.

Suppose that ℐ has degree 2. Then we can write

Because [ℐ,ℐ]⊆ℏ2⁢ℐ, we see that

and therefire the Hi are a representation of a Lie algebra. We can write an explicit recursive formula for 𝒵, which is of the form

The Fg,n satisfy a topological recursion, with the formula

Here, we impose the initial conditions

4.2.4 Where we can find Airy ideals

Let A={1,…,n}. We are interested in a classification of quadratic Airy ideals.

Example 4.2.11.

Consider the Lie algebra 𝔰⁢𝔩2⁢(ℂ) which is generated by E,H,F satisfying the relations

This does have a representation that is an Airy ideal, which is of the form

See [ABCO17] for details.

Unfortunately, 𝔰⁢𝔩2 is one of the rare simple Lie algebras admitting a representation that generates an Airy ideal in finitely many variables.

Proposition 4.2.12 ([ABCO17, Proposition 6.9]).

The following Lie algebras do not admit a representation in finitely many variables generating an Airy ideal:

  • •

    An for n∉{1,5};

  • •

    Cn for n≥6;

  • •

    Dn for n≥4;

  • •

    E6,E7,E8.

4.3 Applications

We will give some examples of Airy ideals that arise as representations of infinite-dimensional algebras and attempt to give them an interpretation in terms of enumerative geomtry. We will consider 𝒲-algebras or vertex operator algebras as our algebras and then consider the algebra of modes. Then we want to realize our vertex operator algebras as sub-VOAs of the Heisenberg VOA. Roughly, a 𝒲-algebra is a generalization of the Virasoro algebra, but we will not give a precise definition.

Example 4.3.1.

𝒲⁡(𝔰⁢𝔩2) is the Virasoro algebra and is generated by a single field

where Wn2=Ln.

4.3.1 Kontsevich-Witten partition function

Example 4.3.2.

The algebra 𝒲⁡(𝔤⁢𝔩2) is generated by two fields

We would like to find expressions for our 𝒲-algebra fields as differential operators, which will create Airy ideals for us. We would like to realize 𝒲⁡(𝔤⁢𝔩2) as a subalgebra of the rank 2 Heisenberg algebra, which is defined by

One basic representation is

More precisely, we can start with any module for the Heisenberg algebra.

We begin with the Z2-twisted representation for the rank 2 Heisenberg algebra. This gives us the following formulae for W⁡(𝔤⁢𝔩2):

We now want to fix a subset of modes such that

and then consider a dilaton shift, which breaks ℏ-homogeneity.

We will first choose modes Wk1 for k≥1 and Wk2 for k≥−1. Recall that the Wk2 are Virasoro modes, so we have no problems there. To break homogeneity, we will shift

and this gives us

If we compute the partition function 𝒵, we obtain the Kontsevich-Witten partition function, which is defined by

Instead of choosing Wk2 for k≥−1, we can take k≥0 (so we delete W−12). The old dilaton shift does not work for indexing reasons, so we take

We then obtain

The resulting 𝒵 is the BGW τ-function for the KdV hierarchy, where our new initial condition is

where

This 𝒵 is defined by

where Θg,n is the Norbury/Chiodo class [Nor17].

4.3.2 The example of W⁡(𝔤⁢𝔩r)

The algebra 𝒲⁡(𝔤⁢𝔩r) is generated by W1⁢(z),…,Wr⁢(z). There is a natural embedding

into the rank r Heisenberg algebra. We of course consider the ℤr twisted module for ℋr.

We want to classify subsets of modes such that there is a valid dilaton shift subject to including all non-negative modes. These are indexed by

and were computed in [Bor+18]. For example, if s=r+1, we choose Wki for all k≥−i+1+δi,1 and if s=1, we choose all Wki for k≥0+δi,1. Then our dilaton shift must be

When s=r+1, then Fg,n are the r-spin intersection numbers and 𝒵 is the τ-function for the r-KdV hierarchy. This recovers the W-constraints for 𝒵. It is unclear what happens for other s. There is a natural candidate, which is called the Chiodo class. This is only known to work for s=r−1 by work of Chidambaram–Garcia-Falide–Giachetto [CGG22].

We may instead consider the permutation σ=(1⁢⋯⁢r−1)⁢(r) and consider the σ-twisted module for ℋr. We then obtain a choice of some s∈{1,…,r}. When r=3 and s=r, we recover the W3 constraints of Alexandrov [Ale15] for open intersection numbers. Conjecturally, when r>3 and s=r, we should recover the open r-spin intersection numbers. For other choices of s, it is unclear what we obtain.

4.3.3 Other examples coming from Lie algebras

For any 𝔤, we can construct the 𝒲-algebra 𝒲⁡(𝔤). For some choices of embeddings into the Heisenberg algebra and twists, Airy structures have been constructed for 𝒲⁡(𝔰⁢𝔬2⁢n),𝒲⁡(𝔢k),𝒲⁡(𝔰⁢𝔭2⁢n), see [Bor+18; BCJ22].

4.3.4 Whittaker vectors

We want to find a state such that

for all k≥1. These states are called Whittaker vectors. These can be computed in our formalism (see [BBCC21]).

4.4 Spectral curve topological recursion

4.4.1 Spectral curves

Definition 4.4.1.

A spectral curve is a quadruple S=(Σ,x,y,B) such that

  • •

    Σ is a Riemann surface;

  • •

    x,y are functions on Σ that are holomorphic except potentially at a finite number of points;

  • •

    B is a symmetric differential Σ×Σ with a double pole on the diagonal and biresidue 1.

For example, we can take Σ=ℙ1 and

Definition 4.4.2.

A spectral curve S is meromorphic if x,y are meromorphic on Σ and is compact if Σ is compact.

Note that if x is meromorphic and Σ is compact, then x is a finite branched covering of ℙ1. In general, we can consider infinite-degree covers. In any case, consider the ramification locus R⊂Σ. For any a∈R, we can write, we can write equations like

Definition 4.4.3.

A spectral curve S is admissible if at all a∈R, ra=±1(modsa), where roughly we have the behavior

near each ramification point a. Here, sa∈{1,…,ra+1}.

In general, suppose we have some equation P⁡(x,y)=0. Of course, if S is compact and meromorphic, then P is polynomial and S is a compactification of P⁡(x,y)=0.

Example 4.4.4.

Let Σ=ℙ1 and

where s∈{1,…,r+1} such that r=±1(mods). Then

These (r,s)-curves recover the (r,s)-partition functions discussed previously.

Example 4.4.5.

Let Σ=ℂ∖cut and

Then

and this spectral curve recovers the stationary Gromov-Witten theory of ℙ1.

Example 4.4.6.

Let Σ=ℂ and

These satisfy the equation

This example reproduces the r-spin Hurwitz number and if Σ=ℂ∞, then we recover the Atlantes Hurwitz numbers.

Example 4.4.7.

Let f∈ℤ and consider the curve

This produces the open Gromov-Witten theorey of ℂ3. More generally, if we take the toric web diagram of a toric Calabi-Yau threefold X, we can produce the mirror spectral curve of X, which recovers the open Gromov-Witten theory of X.

4.4.2 Topological recursion

Topological recursion is a machine that produces symmetric differentials

on Σn from the initial data

This is done by performing local analysis around the ramification points.

For simplicity, we will assume that all a∈R are have simple ramification, so x=x⁡(a)+φ2, and that Σ has genus 0. Then the formula becomes

This is related to our Airy structures as follows. Consider all a∈R and attach a copy of 𝒲⁡(𝔤⁢𝔩ra). We then consider the (ra,sa)-representation constructed previously, and the partition function 𝒵 of this Airy ideal is the same as the expresssion computed by topological recursion. Choose a∈R and write

Choose the vector space

There is a positive subspace V+⊆V with basis dξk=φk−1⁢dφ. The choice of V− is dictated by B. There is a symplectic pairing

Thus we can choose V−⊂V with basis dξ−k such that

The simplest choice is the standard polarization

This can be deformed into

We can now write

Theorem 4.4.8.

The partition function 𝒵 defined by the Fg,n is a solution to the Airy ideal defined by the representations of 𝒲⁡(𝔤⁢𝔩ri).

4.4.3 Enumerative applications

Note that the Airy ideals only see the local structure of the spectral curve topological recursion. However, the differential forms are globally defined, and this gives interesting structure.

Example 4.4.9.

Consider the example of the Gromov-Witten theory of ℙ1. Then there is a mixing operator P̂ such that

The way to obtain the Gromov-Witten invariants of ℙ1 is to consider expansions of ωg,n at the punctures of Σ:

Then the Gromov-Witten theory of ℙ1 will come from the punctures.

In general, expansions at the ramification points will give us integrals over ℳ¯g,n (or r-spin versions), while expansions at the punctures will give enumerative invariants like the Gromov-Witten theory of ℙ1, Gromov-Witten theory of of a toric Calabi-Yau threefold, or Hurwitz numbers.

4.4.4 Some open questions

  1. 1.

    Can we do spectral curve topological recursion for other Airy ideals? We only considered fully twisted representations of 𝒲⁡(𝔤⁢𝔩r). The global structure contains more information, so we would like to find a global structure in some cases.

  2. 2.

    Is there some way to globalize the notion of Airy structures? We would like to consider so-called “x-twisted modules” for the Heisenberg algebra, which will gives information about 𝒲⁡(𝔤⁢𝔩r).

  3. 3.

    If we begin with a spectral curve, can we construct a quantum version?

  4. 4.

    There is a conjectural relationship between knot theory and topological recursion, where the Jones and HOMFLY polynomials should be constructed using topological recursion.

  5. 5.

    If I have a family of spectral curves, what happens in the limit? For example, consider

    Is there some kind of limit

Bibliography

  • [ABCO17]
    Jorgen Ellegaard Andersen, Gaëtan Borot, Leonid O. Chekhov, and Nicolas Orantin. The ABCD of topological recursion. 2017. doi: 10.48550/ARXIV.1703.03307.
  • [AKMV05]
    Mina Aganagic, Albrecht Klemm, Marcos Mariño, and Cumrun Vafa. “The topological vertex”. In: Commun. Math. Phys. 254 (2005), pp. 425–478.
  • [Ale15]
    Alexander Alexandrov. “Open intersection numbers, Kontsevich-Penner model and cut-and-join operators”. In: J. High Energ. Phys. 28 (2015).
  • [AO17]
    Mina Aganagic and Andrei Okounkov. “Quasimap counts and Bethe eigenfunctions”. In: Mosc. Math. J. 17.4 (2017), pp. 565–600.
  • [AO21]
    Mina Aganagic and Andrei Okounkov. “Elliptic stable envelopes”. In: J. Amer. Math. Soc. 34 (2021), pp. 79–133.
  • [BBCC21]
    Gaëtan Borot, Vincent Bouchard, Nitin Kumar Chidambaram, and Thomas Creutzig. Whittaker vectors for 𝒲-algebras from topological recursion. 2021. doi: 10.48550/ARXIV.2104.04516.
  • [BCJ22]
    Vincent Bouchard, Thomas Creutzig, and Aniket Joshi. Airy ideals, transvections and 𝒲⁡(𝔰⁢𝔭2⁢N)-algebras. 2022. doi: 10.48550/ARXIV.2207.04336.
  • [BCK05]
    Aaron Bertram, Ionuţ Ciocan-Fontaine, and Bumsig Kim. “Two proofs of a conjecture of Hori and Vafa”. In: Duke Math. J. 126.1 (2005), pp. 101–136.
  • [BDGR18]
    Alexandr Buryak, Boris Dubrovin, Jér ́emy Guéré, and Paolo Rossi. “Tau-structure for the double ramification hierarchies”. In: Comm. Math. Phys. 363.1 (2018), pp. 191–260.
  • [Bor+18]
    Gaëtan Borot, Vincent Bouchard, Nitin K. Chidambaram, Thomas Creutzig, and Dmitry Noshchenko. Higher Airy structures, W algebras and topological recursion. 2018. doi: 10.48550/ARXIV.1812.08738.
  • [BR16]
    Alexandr Buryak and Paolo Rossi. “Recursion relations for double ramification hierarchies”. In: Comm. Math. Phys. 342.2 (2016), pp. 533–568.
  • [BR21]
    Alexandr Buryak and Paolo Rossi. “Quadratic double ramification integrals and the noncommutative Kd V hierarchy”. In: Bull. Lond. Math. Soc. 53.3 (2021), pp. 843–854.
  • [BR22]
    Alexandr Buryak and Paolo Rossi. “A generalisation of Witten’s conjecture for the Pixton class and the noncommutative KdV hierarchy”. In: J. Inst. Math. Jussieu (2022), pp. 1–23.
  • [CGG22]
    Nitin Kumar Chidambaram, Elba Garcia-Failde, and Alessandro Giacchetto. Relations on ℳ¯g,n and the negative r-spin Witten conjecture. 2022. doi: 10.48550/ARXIV.2205.15621.
  • [CGT24]
    Tom Coates, Alexander Givental, and Hsian-Hua Tseng. “Virasoro constraints for toric bundles”. In: Forum Math. Pi 12 (2024), Paper No. e4, 28. issn: 2050-5086. doi: 10.1017/fmp.2024.2.
  • [CKM14]
    Ionuţ Ciocan-Fontaine, Bumsig Kim, and Davesh Maulik. “Stable quasimaps to GIT quotients”. In: J. Geom. Phys. 75 (2014), pp. 17–47.
  • [DN83]
    Boris Dubrovin and Sergei Novikov. “The Hamiltonian formalism of one-dimensional systems of hydrodynamic type and the Bogolyubov-Whitlam averaging method”. In: Dokl. Akad. Nauk SSSR 240.4 (1983), pp. 781–785.
  • [DZ01]
    Boris Dubrovin and Youjin Zhang. Normal forms of hierarchies of integrable PDEs, Frobenius manifolds and Gromov - Witten invariants. 2001. doi: 10.48550/ARXIV.MATH/0108160.
  • [EHX97]
    Tohru Eguchi, Kentaro Hori, and Chuan-Sheng Xiong. “Quantum cohomology and Virasoro algebra”. In: Phys. Lett. B 402.1 (1997), pp. 71–80.
  • [ELSV99]
    Torsten Ekedahl, Sergei Lando, Michael Shapiro, and Alek Vainshtein. “On Hurwitz number and Hodge integrals”. In: C. R. Acad. sci. Ser. I Math. 328.12 (1999), pp. 1175–1180.
  • [EO07]
    Bertrand Eynard and Nicolas Orantin. “Invariants of algebraic curves and topological expansion”. In: Commun. Number Theory Phys. 1.2 (2007), pp. 347–452.
  • [EV02]
    Pavel Etingov and Alexander Varchenko. “Dynamical Weyl groups and applications”. In: Adv. Math. 167.1 (2002), pp. 74–127.
  • [FP00]
    Carel Faber and Rahul Pandharipande. “Hodge integrals and Gromov-Witten theory”. In: Invent. Math. 139 (2000), pp. 173–199.
  • [Get02]
    Ezra Getzler. “A Darboux theorem for Hamiltonian operators in the formal calculus of variations”. In: Duke Math. J. 113.3 (2002), pp. 535–560.
  • [Get04]
    Ezra Getzler. “The equivariant Toda conjecture”. In: Publ. Res. Inst. Math. Sci. 40 (2004), pp. 507–536.
  • [Get99]
    Ezra Getzler. “The Virasoro conjecture for Gromov-Witten invariants”. In: Algebraic geometry: Hirzebruch 70 (Warsaw, 1998). Vol. 241. Contemp. Math. Providence: Amer. Math. Soc., 1999, pp. 147–176.
  • [Giv01]
    Alexander B. Givental. “Gromov-Witten invariants and quantization of quadratic Hamiltonians”. In: vol. 1. 4. Dedicated to the memory of I. G. Petrovskii on the occasion of his 100th anniversary. 2001, pp. 551–568, 645. doi: 10.17323/1609-4514-2001-1-4-551-568.
  • [HM98]
    Joe Harris and Ian Morrison. Moduli of curves. Graduate Texts in Mathematics. New York: Springer, 1998.
  • [Iri07]
    Hiroshi Iritani. “Convergence of quantum cohomology by quantum Lefschetz”. In: J. reine angew. Math. 610 (2007), pp. 29–69.
  • [JK20]
    Dosang Joe and Bumsig Kim. “Equivariant mirrors and the Virasoro conjecture for flag manifolds”. In: Int. Math. Res. Not. 2003.15 (2020), pp. 859–882.
  • [JPPZ17]
    F. Janda, R. Pandharipande, A. Pixton, and D. Zvonkine. “Double ramification cycles on the moduli spaces of curves”. In: Publ. Math. Inst. Hautes Études Sci. 125 (2017), pp. 221–266.
  • [Kaz09]
    Maxim Kazarian. “KP hierarchy for Hodge integrals”. In: Adv. in Math. 221.NN (2009), pp. 1–21.
  • [Kon92]
    Maxim Kontsevich. “Intersection theory on the moduli spaces of curves and the matrix Airy function”. In: Commun. Math. Phys. 147.1 (1992), pp. 1–23.
  • [KS18]
    Maxim Kontsevich and Yan Soibelman. “Airy structures and symplectic geometry of topological recursion”. In: Topological recursion and its influence in analysis, geometry, and topology. Ed. by Chiu-Chu Melissa Liu and Motohico Mulase. Vol. 100. Proceedings of symposia in pure mathematics. Providence: American Mathematical Society, 2018, pp. 433–490.
  • [Li01]
    Jun Li. “Stable morphisms to singular schemes and relative stable morphisms”. In: J. Diff. Geom. 57 (2001), pp. 509–578.
  • [Liu21]
    Henry Liu. “Quasimaps and stable pairs”. In: Forum Math. Sigma 9 (2021), e32.
  • [LP09]
    Marc Levine and Rahul Pandharipande. “Algebraic cobordism revisited”. In: Invent. Math. 176 (2009), pp. 63–130.
  • [MNOP06a]
    Davesh Maulik, Nikita Nekrasov, Andrei Okounkov, and Rahul Pandharipande. “Gromov-Witten theory and Donaldson-Thomas theory, I”. In: Compos. Math. 142.5 (2006), pp. 1263–1285.
  • [MNOP06b]
    Davesh Maulik, Nikita Nekrasov, Andrei Okounkov, and Rahul Pandharipande. “Gromov-Witten theory and Donaldson-Thomas theory, II”. In: Compos. Math. 142.5 (2006), pp. 1286–1304.
  • [MO19]
    Davesh Maulik and Andrei Okounkov. Quantum groups and quantum cohomology. Astérisque. Paris: Société Mathématique de France, 2019.
  • [MOOP11]
    Davesh Maulik, Alexei Oblomkov, Andrei Okounkov, and Rahul Pandharipande. “Gromov-Witten/Donaldson-Thomas correspondence for toric 3-folds”. In: Invent. Math. 186 (2011), pp. 435–479.
  • [MOOP20]
    Miguel Moreira, Alexei Oblomkov, Andrei Okounkov, and Rahul Pandharipande. Virasoro constraints for stable pairs on toric 3-folds. 2020. doi: 10.48550/ARXIV.2008.12514.
  • [Mor22]
    Miguel Moreira. “Virasoro conjecture for the stable pairs descendent theory of simply connected 3-folds (with applications to the Hilbert scheme of points of a surface)”. In: J. London Math. Soc. 106 (2022), pp. 154–191.
  • [MR20]
    Davesh Maulik and Dhruv Ranganathan. Logarithmic Donaldson-Thomas theory. 2020. doi: 10.48550/ARXIV.2006.06603.
  • [Nor17]
    Paul Norbury. A new cohomology class on the moduli space of curves. 2017. doi: 10.48550/ARXIV.1712.03662.
  • [NS09]
    Nikita Nekrasov and Samson Shatashvili. “Supersymmetric vacua and Bethe ansatz”. In: Nucl. Phys. Proc. Suppl. 192–193 (2009), pp. 91–112.
  • [Oko17]
    Andrei Okounkov. “Lectures on K-theoretic computations in enumerative geometry”. In: Geometry of moduli spaces and representation theory. Ed. by Roman Bezrukavnikov, Alexander Braverman, and Zhiwei Yun. Vol. 24. IAS/PCMI Mathematics Series. Providence: American Mathematical Society, 2017, pp. 251–380.
  • [Oko19]
    Andrei Okounkov. “Takagi lectures on Donaldson-Thomas theory”. In: Jpn. J. Math. 14 (2019), pp. 67–133.
  • [Oko20]
    Andrei Okounkov. Nonabelian stable envelopes, vertex functions with descendents, and integral solutions of q-difference equations. 2020. doi: 10.48550/ARXIV.2010.13217.
  • [Oko21]
    Andrei Okounkov. “Inductive construction of stable envelopes”. In: Lett. Math. Phys. 111.141 (2021).
  • [OOP20]
    Alexei Oblomkov, Andrei Okounkov, and Rahul Pandharipande. “GW/PT descendent correspondence via vertex operators”. In: Commun. Math. Phys. 374 (2020), pp. 1321–1359.
  • [OP06a]
    Andrei Okoumkov and Rahul Pandharipande. “Gromov-Witten theory, Hurwitz theory, and completed cycles”. In: Ann. of Math. 163 (2006), pp. 517–560.
  • [OP06b]
    Andrei Okoumkov and Rahul Pandharipande. “The equivariant Gromov-Witten theory of 𝐏1”. In: Ann. of Math. 163 (2006), pp. 561–605.
  • [ORV06]
    Andrei Okounkov, Nikolai Reshetikhin, and Cumrun Vafa. “Quantum Calabi-Yau and classical crystals”. In: The unity of mathematics: in honor of the ninetieth birthday of I.M. Gelfand. Ed. by Pavel Etingof, Vladimir Retakh, and I.M. Singer. Boston: Birkhäuser Boston, 2006, pp. 597–618.
  • [OS22]
    Andrei Okounkov and Andrey Smirnov. “Quantum difference equations for Nakajima varieties”. In: Invent. Math. 229 (2022), pp. 1203–1299.
  • [Par12]
    Brett Parker. “Exploded manifolds”. In: Adv. in Math. 229.6 (2012), pp. 3256–3319.
  • [PP14]
    Rahul Pandharipande and Aaron Pixton. “Gromov-Witten/pairs descendent correspondence for toric 3-folds”. In: Geom. Topol. 18 (2014), pp. 2747–2821.
  • [PP17]
    Rahul Pandharipande and Aaron Pixton. “Gromov-Witten/pairs correspondence for the quintic 3-fold”. In: J. Amer. Math. Soc. 30 (2017), pp. 389–449.
  • [PT09]
    Rahul Pandharipande and Richard Thomas. “Curve counting via stable pairs in the derived category”. In: Invent. Math. 178 (2009), pp. 407–447.
  • [Ros17]
    Paolo Rossi. “Integrability, quantization and moduli spaces of curves”. In: SIGMA Symmetry Integrability Geom. Methods Appl. 13 (2017), Paper No. 060, 29.
  • [She16]
    Junliang Shen. “Cobordism invariants of the moduli space of stable pairs”. In: J. London Math. Soc. 94 (2016), pp. 427–446.
  • [Smi16]
    Andrey Smirnov. Rationality of capped descendent vertex in K-theory. 2016. doi: 10.48550/ARXIV.1612.01048.
  • [Tel12]
    Constantin Teleman. “The structure of 2D semi-simple field theories”. In: Invent. Math. 188 (2012), pp. 525–588.
  • [Zho18]
    Zijun Zhou. “Relative orbifold Donaldson-Thomas theory and the degeneration formula”. In: Algebraic Geom. 5.4 (2018), pp. 464–522.