Patrick Lei

All notes

Log GLSM and related topics

Sichuan University, Summer 2026.

Abstract.

These are live-TeXed (for all talks besides my own and those of Shuai Guo, some figures were inserted later using Codex or Claude Code) notes from the log GLSM summer school at Sichuan University in July 2026. Note that all talks (besides mine) were given in Chinese. I apologize for any translation errors as well as for any errors in my lectures. In addition, the notes for the two lectures by Shuai Guo are experimental AI-generated transcriptions of his slides. While I did give Claude access to all of my other notes to see what my writing style is, please let me know if there is a mismatch in writing style or if there are any errors. Finally, I would like to thank You-Cheng Chou, Tingting Fang, and Yaoxiong Wen for pointing out mistakes in the notes.

1 Theta-stratifications (Yaoxiong Wen and Bin Wang)

1.1 Our first moduli problems.

Our first moduli space will be ℙN−1, which parameterizes 1-dimensional linear subspaces in ℂN, or equivalently is the geometric quotient (ℂN∖{0})/ℂ×. In this case every ℂ×-orbit in ℂN∖{0} is closed. This construction is our first example of geometric invariant theory. In general, a set-valued moduli problem assigns to every scheme T a collection of isomorphism classes of families over T, and we can think of it as a functor

Example 1.1.

Let C be a smooth projective curve. Then consider the assignment

where Et¯=E|C×{t¯} and ∼ denotes isomorphism of vector bundles on C×T. This parameterizes isomorphism classes of families of vector bundles on C, but it is still not the correct stack-valued moduli problem.

Example 1.2.

By the Yoneda Lemma, every scheme Y defines a moduli problem.

Unfortunately, not every groupoid-valued moduli problem can be represented, as a stack, by a scheme or algebraic space. A scheme or algebraic space has trivial inertia, whereas objects in a moduli problem may have nontrivial automorphisms. For example, if we consider the moduli of line bundles on ℙ1, each line bundle has an entire ℂ× of automorphisms. Therefore, when we consider moduli problems, we need to remember both the objects and the isomorphisms between them, and the target of the functor should be the category of groupoids.

Definition 1.3 (informal).

A stack 𝔛 over S is a pseudofunctor

which satisfies descent for objects and morphisms with respect to the fppf topology.

Example 1.4.

Consider the stack 𝖡𝗎𝗇r,d⁢(C) which sends a scheme T to the groupoid whose objects are vector bundles E on T×C such that every geometric fiber Et¯ has rank r and degree d, and whose morphisms are isomorphisms of vector bundles on T×C.

Unfortunately, there are far too many stacks to consider, so we will need to consider a subclass of stacks which is amenable to the methods of algebraic geometry.

Definition 1.5.

An algebraic stack is a stack with representable (by algebraic spaces) diagonal and with a smooth surjective morphism from a scheme U (which is called an atlas).

Note that the first condition means that for any morphism T→𝔛×𝔛 from an algebraic space, then T×𝔛×𝔛𝔛 is an algebraic space. This is equivalent to all Isom sheaves being algebraic spaces and implies that any morphism from a scheme to 𝔛 is representable.

Remark 1.6.

For sets, if we have morphisms f:A→C and g:B→C, then the fiber product is

For schemes or algebraic spaces, the same formula describes the T-points of the scheme-theoretic fiber product. For stacks, a T-point of the 2-fiber product 𝔄×ℭ𝔅 is a triple (a,b,ϕ), where a∈𝔄⁡(T), b∈𝔅⁡(T), and

is an isomorphism in ℭ⁡(T). Thus equality must be replaced by an isomorphism, and the isomorphism must be recorded as part of the data.

1.2 Quotient stacks.

Let G be an algebraic group and let X be a scheme with a left action of G.

Definition 1.7.

The quotient stack [X/G] sends a scheme T to the groupoid whose objects are diagrams

where P→T is a right principal G-bundle for the fppf topology and φ satisfies φ⁡(p⁢g)=g−1⁢φ⁢(p). A morphism from (P,φ) to (P′,φ′) is an isomorphism of principal G-bundles P→∼P′ compatible with the maps to X.

Example 1.8.

Let 𝔛 be a smooth separated DM stack of finite type over ℂ. If either the generic stabilizer of 𝔛 is trivial or the coarse moduli space of 𝔛 is quasi-projective, then 𝔛 is a global quotient [EHKV01]. Here, a DM stack is an algebraic stack with unramified diagonal, and a global quotient means a stack of the form [U/GLn] for some algebraic space U with an action of GLn.

Exercise 1.9.

Show that the groupoid [X/G]⁢(ℂ) is equivalent to the action groupoid of G⁡(ℂ) acting on X⁡(ℂ).

After choosing one representative xi∈X⁡(ℂ) of each G⁡(ℂ)-orbit, the above exercise implies that, as an abstract groupoid and not as an algebraic stack, we have

where Gxi is the stabilizer of xi.

1.3 Fine vs coarse vs good moduli spaces.

Definition 1.10.

A fine moduli space for a moduli problem is a scheme or algebraic space M which represents the functor.

Unfortunately, it is very hard to find a fine moduli space, so we want to weaken this condition.

Definition 1.11.

A coarse moduli space for an algebraic stack 𝔛 is a morphism π:𝔛→M to an algebraic space such that

  1. (1)

    For every algebraic space Y, composition with π induces a bijection

  2. (2)

    For every algebraically closed field K, the induced map

    is a bijection.

Even these do not always exist. A good moduli space is a different weakening: it replaces the requirement of a bijection on geometric points by a cohomological affineness condition [Alp13].

Definition 1.12.

A quasi-compact morphism q:𝔛→X from an algebraic stack to an algebraic space is a good moduli space if

  1. (1)

    The pushforward functor q∗:𝖰𝖢𝗈𝗁⁡(𝔛)→𝖰𝖢𝗈𝗁⁡(X) is exact;

  2. (2)

    The natural map 𝒪X→q∗⁢𝒪𝔛 is an isomorphism.

Table 1. Types of moduli spaces
Notion Defining property Geometric behavior Feature
Fine moduli Represents the set-valued moduli functor Its points represent the objects recorded by the functor Admits a universal family
Coarse moduli Universal for maps to algebraic spaces and bijective on geometric isomorphism classes One geometric point for each isomorphism class Forgets automorphisms
Good moduli Cohomologically affine and 𝒪X≃q∗⁢𝒪𝔛 May identify distinct nonclosed points Gives S-equivalence in GIT-type examples

The existence of these moduli spaces is not determined only by whether stabilizers are finite or infinite. For example, B⁢𝔾m→Spec⁡ℂ is both a coarse moduli space and a good moduli space, even though its stabilizer is 𝔾m.

Theorem 1.13 (Keel-Mori [KM97]).

Let 𝔛 be an algebraic stack of finite type over ℂ. If its inertia morphism

is finite, then 𝔛 admits a coarse moduli space. The fiber of I𝔛→𝔛 over a geometric point x is the automorphism group Aut⁡(x).

Theorem 1.14 (Alper–Hall–Rydh [AHR20]).

Let

be a good moduli space, where 𝔛 is a Noetherian algebraic stack over ℂ with affine diagonal. Let x∈𝔛⁡(ℂ) be a closed point, and denote its stabilizer group by Gx.

Then there exists an affine scheme Spec⁡A equipped with an action of Gx, together with a Cartesian diagram

where

is an étale neighborhood of q⁡(x).

For r≥2, the stack 𝖡𝗎𝗇r,d⁢(C) is locally of finite type, but it is neither quasi-compact nor separated. To see that it is not quasi-compact, choose line bundles Ln and Md−n of degrees n and d−n, respectively, and consider

The bundle En contains a line subbundle of degree n. In any quasi-compact family of vector bundles of fixed rank and degree, the degrees of line subbundles are bounded above, so the bundles En cannot all lie in a quasi-compact substack as n→∞. To see that the stack is not separated, note that every vector bundle E has scalar automorphisms 𝔾m⊂Aut⁡(E). Stabilizer groups of a separated algebraic stack are proper, whereas 𝔾m is not proper. Passing to semistable bundles does not make the stack separated; rather, it cuts out a bounded open substack of finite type which admits a good moduli space whose closed points are S-equivalence classes.

Definition 1.15 (Slope stability).

For a nonzero vector bundle E, define its slope by

The vector bundle E is semistable if

for every nonzero proper saturated subsheaf F⊂E, and it is stable if the inequality is strict. Since C is a smooth curve, a saturated subsheaf of a vector bundle is a subbundle, so it is equivalent to test nonzero proper subbundles.

Armed with this definition, we may now consider the open substack 𝖡𝗎𝗇r,dss⁢(C). Semistable bundles of fixed rank and degree form a bounded family, so this stack is of finite type, and in particular quasi-compact. It is still not separated because every object has the nonproper scalar stabilizer 𝔾m.

Definition 1.16 (S-equivalence).

Let E be a semistable vector bundle. A Jordan–Hölder filtration of E is a filtration

such that every quotient Ji/Ji−1 is stable of slope μ⁡(E). The filtration is not necessarily unique, but the associated graded vector bundle

is unique up to noncanonical isomorphism. Two semistable vector bundles E and E′ of the same slope are S-equivalent if gr⁡(E)≃gr⁡(E′).

Now we consider the moduli space Mr,dss⁢(C) of S-equivalence classes of semistable vector bundles on C. There is a good moduli space morphism

and the closed points of Mr,dss⁢(C) are in bijection with S-equivalence classes of semistable vector bundles.

Remark 1.17.

The original method of proof for this theorem was to construct the moduli space using geometric invariant theory. The goal of this series of lectures will be to discuss an intrinsic approach using Θ-stratifications together with existence criteria for good moduli spaces, without first choosing a GIT presentation.

1.4 Theta-stratifications.

The framework of Θ-stratifications and numerical invariants used here is developed in [Hal14].

Let 𝔛 be an algebraic stack and consider Θ≔[𝔸1/𝔾m], where the action on points is λ⋅z=λ⁢z; equivalently, the coordinate function t has weight −1. Assume that the mapping stacks below are algebraic. Then define the stack 𝖥𝗂𝗅𝗍⁡(𝔛)≔𝖬𝖺𝗉⁡(Θ,𝔛) of filtered objects and the stack 𝖦𝗋𝖺𝖽⁡(𝔛)≔𝖬𝖺𝗉⁡(B⁢𝔾m,𝔛) of graded objects. There are canonical morphisms

Here ev1 is restriction to 1∈𝔸1, gr is induced by the inclusion B⁢𝔾m={0}/𝔾m↪Θ, and σ is induced by the projection Θ→B⁢𝔾m. In particular, gr∘σ=id. We also write tot=ev1∘σ:𝖦𝗋𝖺𝖽⁡(𝔛)→𝔛.

Definition 1.18.

A Θ-stratum in an algebraic stack 𝔜 is a union of connected components S⊂𝖥𝗂𝗅𝗍⁡(𝔜) such that the restriction

is a closed immersion.

A Θ-stratification of 𝔛 consists of

  1. (1)

    A totally ordered set I with a least element 0;

  2. (2)

    An increasing collection of open substacks 𝔛≤α⊂𝔛, indexed by α∈I, such that

    and such that for every x∈|𝔛|, the set {α∈I|x∈|𝔛≤α|} has a least element;

  3. (3)

    For every α>0, a Θ-stratum Sα⊂𝖥𝗂𝗅𝗍⁡(𝔛≤α) such that

The semistable locus is 𝔛ss≔𝔛≤0, and the stratification gives a disjoint decomposition

Consider the stack 𝖡𝗎𝗇G⁢(C)=𝖬𝖺𝗉⁡(C,B⁢G). We will consider the stack

Any such map f is equivalent to a 𝔾m-equivariant principal G-bundle ℰ over 𝔸1×C. When G=GLn, this is equivalent to a 𝔾m-equivariant vector bundle on 𝔸1×C, or equivalently to a finitely generated graded locally free 𝒪C⁢[t]-module

With the convention above, the coordinate function t has weight −1, so multiplication by t gives maps

Since ℰ is locally free over 𝒪C⁢[t], multiplication by t is injective.

Remark 1.19.

Consider the vector bundle E=ℰ|{1}×C. Since 𝔾m acts freely on 𝔾m×C, equivariant descent gives

where PC:𝔾m×C→C is the projection. Equivalently,

Define

Since t⁢ℰw+1⊂ℰw, we have Fw+1⁢E⊂Fw⁢E. The finite generation and local freeness of ℰ imply that this is a finite, exhaustive, and separated decreasing filtration by subbundles, which may be written as

In the other direction, let E be a vector bundle with a finite decreasing filtration F∙⁢E by subbundles such that the quotients Fw⁢E/Fw+1⁢E are locally free. The Rees construction is the graded 𝒪C⁢[t]-module

It is locally free over 𝒪C⁢[t], and its fibers at t=1 and t=0 satisfy

Therefore, a map from Θ to the stack of vector bundles is equivalent to a finite ℤ-weighted filtration of a vector bundle by subbundles with locally free associated graded pieces.

1.5 Non-abelian localization.

We will begin with the classical setting. Let X be a smooth projective variety with an action of 𝔾m, let F be a 𝔾m-equivariant perfect complex, and write

for the decomposition of the fixed locus into connected components. After localizing the representation ring R⁡(𝔾m) so that the equivariant Euler classes below become invertible, we have

where

We will now consider the stacky version. Let

be a Θ-stratification. Since ev1|Sα is a closed immersion, we may identify Sα with its image in 𝔛. The center of the stratum is

where σ:𝖦𝗋𝖺𝖽⁡(𝔛)→𝖥𝗂𝗅𝗍⁡(𝔛) is induced by the projection Θ→B⁢𝔾m.

Theorem 1.20 (Virtual non-abelian localization [Hal15, Theorem 5.1]).

Let 𝔛 be a quasi-smooth derived stack with a Θ-stratification

where we identify every stratum with its image under ev1, and let Zα be its center. Decompose the restriction of the cotangent complex according to the 𝔾m-weights as

where the homology of 𝕃α−, 𝕃α0, and 𝕃α+ is concentrated in negative, zero, and positive weights, respectively. Define

Here det(𝕃α+) and rk⁡(𝕃α+) denote the determinant and virtual rank of the perfect complex 𝕃α+. We say that a complex of vector spaces is finite-dimensional if the direct sum of all of its cohomology groups is finite-dimensional. Let F∈𝖯𝖾𝗋𝖿⁡(𝔛). Assume that

is finite-dimensional for every α and is acyclic for all but finitely many α. Then R⁢Γ⁢(𝔛,F) is finite-dimensional if and only if R⁢Γ⁢(𝔛ss,F) is finite-dimensional, and

Here quasi-smooth means that 𝕃𝔛 is perfect of Tor-amplitude contained in [−1,1].

1.6 Geometric invariant theory.

For geometric invariant theory, see [MFK94]. The vector-bundle example below uses the Harder–Narasimhan filtration [HN74].

Let G be a reductive group and X be a projective scheme equipped with a G-equivariant line bundle ℒ. The semistable locus can be defined using the Hilbert-Mumford criterion, where it consists of all points where for all cocharacters λ:𝔾m→G, we can define the point

and define the quantity

A point is semistable if μL⁢(x,λ)≤0 for all cocharacters of G. The stable locus consists of all points where the inequality is strict and the stabilizer is finite.

Theorem 1.21 ([MFK94]).

The stable and semistable loci are open in X. Moreover, Xss/G is a good quotient and Xs/G is a geometric quotient.

Example 1.22.

Consider the moduli stack 𝖡𝗎𝗇r,d of vector bundles of rank r and degree d on a curve of genus at least 2. Then recall that the semistable locus was defined to consist of slope semistable vector bundles. In fact, this can be obtained via GIT.

Fix an ample line bundle 𝒪⁡(1) on C. Then there exists some n0 such that for all n≥n0 and E∈𝖡𝗎𝗇r,d, E⁡(n) is globally generated. Therefore, we can consider

If we set V=ℂN, where N=h0⁢(E⁡(n)), then there is a surjection

Therefore, we can think of E as a member of a Quot scheme. On the Quot scheme, we can consider the universal family

Now define the line bundle

for some m≫0. We can also consider the action of SL⁡(V) on the Quot scheme. For some cocharacter λ, we can split

with weights s and N−s. Then consider

and consider the filtration

induced by this. Then we see that

Then we compute the restriction ℒ|grFE and obtain the spaces H0⁢(C,F1⁢E⁢(m)) and H0⁢(C,E⁡(m)/F1⁢E⁢(m)), and therefore the weight of ℒ is given by

which recovers GIT.

We will now consider the unstable locus. Let E be a vector bundle on C. Then there exists a unique filtration

such that

  1. (1)

    All grFi⁡E are semistable;

  2. (2)

    We have μ⁡(grFi⁡E)>μ⁡(grFi−1⁡E) for all i.

Remark 1.23.

From the GIT perspective, we should really take μ to be the reduced Hilbert polynomial.

1.7 Construction of Theta-stratifications.

Let 𝔛 be a reasonable algebraic stack.1 Now consider some family ℰ over 𝔸1×C. If we restrict to the central fiber, we see that 𝔾m acts on grF⁡E, which means we have a map 𝔾m→Aut⁡grF⁡E, and from this we can compute the weight.

Note that the morphism σ:𝖦𝗋𝖺𝖽⁡(𝔛)→𝖥𝗂𝗅𝗍⁡(𝔛) is a Θ-contraction, and therefore

Example 1.24.

Let 𝔛=B⁢G. Then we have

where Pλ is the parabolic subgroup corresponding to the cocharacter λ, and we can always choose representatives in a single Weyl chamber. Moreover, we have

where Lλ is the Levi subgroup corresponding to λ.

Definition 1.25.

A Θ-stratum in 𝔛 is a union of connected components S⊆𝖥𝗂𝗅𝗍⁡(𝔛) such that

is a closed immersion.

Roughly speaking, closed immersions are injective, so an object in S consists of an object in 𝔛 and a “canonical filtration.”

Definition 1.26.

A Θ-stratification is an increasing family of open substacks

where Γ is some totally ordered set, together with Θ-strata

which are required to satisfy

In particular, this implies that

Analogously to the Hilbert-Mumford criterion, we will construct Θ-stratifications using numerical invariants.

Definition 1.27.

A numerical invariant μ with values in a totally ordered set Γ consists of an assignment of a scaling invariant function

for any p∈𝔛⁡(ℂ) and any nondegenerate homomorphism γ:𝔾mq→Aut⁡(p). These are required to satisfy

  1. (1)

    The assignment is invariant under field extensions. This point will not be elaborated on because we are working over ℂ;

  2. (2)

    If we have 𝔾mq↪𝔾mn↪Aut⁡(p), then we have a factorization ℝq∖{0}→ℝn∖{0}→Γ;

  3. (3)

    If we have some ξ:S→𝔛 and γ:𝔾m,Sq→Aut⁡(ξ) with finite kernel, the function μκ⁡(s) is locally constant on S, where s ranges over points of S.

We will give a naive version of a construction starting with a numerical invariant μ. Define x∈𝔛 to be semistable if μ⁡(f)≤0 for any nondegenerate f∈𝖥𝗂𝗅𝗍⁡(𝔛) satisfying f⁡(1)=x. If x is unstable, then we define its instability level

If a filtration f attains this supremum, we call it a “Harder-Narasimhan filtration.”

Definition 1.28.

A numerical invariant μ is standard if

  1. (1)

    μ⁡(x) and μ⁡(−x) are not both positive for x∈ℝq∖{0};

  2. (2)

    Each function μγ is strictly quasi-concave, namely for all δ∈[0,1] and linearly independent x0,x1∈ℝq such that μ⁡(x0),μ⁡(x1)≥0, then

    If either value is strictly positive, the inequality is required to be strict.

The next condition is that we need some kind of rationality. We consider 𝕏∗⁢(T)ℝ, and our goal is to find the rational points. This leads us to condition (R): if μγ attains a positive value for some γ, then it attains a maximal value at a rational point.

Theorem 1.29 ([Hal14]).

Let 𝔛 be a stack and μ be a standard numerical invariant satisfying condition (R). Then μ defines a Θ-stratification if and only if it satisfies the following:

  • •

    For any DVR R with fraction field K and residue field k, then μ⁡(fKHN)≤Mμ⁢(fk) (HN-specialization);

  • •

    For any map B:T→𝔛 from a finite type scheme, there exists a quasi-compact open 𝔛′⊆𝔛 such that for any point of finite type p∈T⁡(k) and a filtration f of B⁡(p) for which μ⁡(f)>0, there exists another filtration f′ of B⁡(p) with μ⁡(f′)≥μ⁡(f) satisfying f′⁢(0)∈𝔛′ (HN-boundedness).

The point here is that the conditions of being standard and (R) combined with HN-boundedness imply existence and HN-specialization implies uniqueness.

2 Intro to log geometry (Tingting Fang and Zijian Han)

2.1 Motivation.

In enumerative geometry, we will consider intersection numbers on compactified moduli spaces. This is going to require our objects to degenerate in some way, for example for domain curves to be nodal, maps to land in boundary divisors, or we degenerate the target. In ordinary geometry, we view these as singular, but in log geometry, there is a way to think of these situations as being log smooth. In log geometry, we are going to replace functions with the data of functions and how they vanish on the boundary (or namely the morphism α:MX→𝒪X).

Example 2.1.

Consider a family of curves (x⁢y=t) over 𝔸1, where the central fiber is nodal. In log geometry, if we place the divisorial log structure on 𝔸1, then this family is log smooth.

2.2 Monoids.

Definition 2.2.

A monoid is a commutative semigroup with a unit.

For a monoid P, we may associate a group Pgp via the standard Grothendieck construction

Definition 2.3.

If the natural morphism P↪Pgp is injective, we say that P is integral.

Example 2.4.

If P has generators a,b,c with relation a+c=b+c, then in Pgp we have a=b, so P is not integral.

Definition 2.5.

A monoid P is

  • •

    fine if P is finitely generated and integral;

  • •

    saturated if for all p∈Pgp and all natural numbers n, if n⁢p∈P then p∈P;

  • •

    torsion-free if Pgp is;

  • •

    sharp if the identity is the only invertible element.

2.3 Log structures.

Let X be a scheme.

Definition 2.6.

A pre-log structure on X is a sheaf of monoids MX on the small étale site of X and a morphism α:MX→𝒪X. It is a log structure if α−1⁢(𝒪X×)≃𝒪X×, where the isomorphism is given by α.

We now consider the exact sequence

and the quotient here is called the characteristic sheaf or the ghost sheaf.

Example 2.7.

Let MX=𝒪X× and α be the identity. Then there is no additional information provided by the log structure.

Example 2.8.

Let X be a regular variety and D be a closed subscheme. Set U=X∖D. Then define

where j:U→X is the open immersion. If D is a divisor, then this is called the divisorial log structure.

For example, if X=𝔸1 and D={0}, then the characteristic sheaf vanishes everywhere except for the origin. Here, we compute

The stalk is 𝒪X× everywhere except at the origin, where it is 𝒪𝔸1,0×⊕ℕ.

Example 2.9.

Now let X=𝔸2 and D be the union of the coordinate axes. Then

Note here that at the general point, we see Mp=𝒪X,p×. At the origin, we see 𝒪X,p×⊕ℕ2. At a general point on the x-axis, we see that xa is invertible, so we see the stalk 𝒪X,p×⊕ℕ⋅b. Similarly, we can compute the stalk at a general point on the y-axis.

Just as we can associate a sheaf to a presheaf, we will associate a log structure to a pre-log structure. First, a pushout in a category 𝒞 will be denoted by B⊔AC and is the colimit of the diagram

In the case where 𝒞 is commutative monoids, then

Example 2.10.

Consider the morphisms 2:ℕ→ℕ and 3:ℕ→ℕ. Then the pushout is the monoid

If we replace 𝒞 by the category of sheaves of commutative monoids, we can form the pushout on sections and then sheafify.

Definition 2.11.

The associated log structure of a pre-log structure α:MX→𝒪X is the pushout

The map to 𝒪X is given by (a,b)→α⁡(a)⁢b.

Definition 2.12.

A morphism of log schemes (X,MX)→(Y,MY) consists of

  1. (1)

    A morphism f¯:X¯→Y¯ of underlying schemes;

  2. (2)

    A map f♭:f∗⁢MY→MX, where f∗⁢MY is the log structure associated to f−1⁢MY.

Definition 2.13.

A chart for a log scheme (X,α:MX→𝒪X) is a morphism from the constant monoid PX→MX such that MX≃PXa≃PX⊕αP−1⁢(𝒪X×)𝒪X×, where αP:PX→MX→𝒪X.

Charts are useful in specifying log structures. For example, if we have PX→𝒪X, then PXa gives a log structure on X.

Example 2.14.

The trivial log structure is given by the chart P={0}.

Example 2.15.

Let X=Spec⁡k and give a log structure by αℕ:ℕ→k which sends 0 to 1 and 1 to 0. This is called the standard log point and can also be formed by restricting the divisorial log structure on 𝔸1 to the origin.

Example 2.16.

Consider the chart ℕ→𝒪𝔸1 which sends n↦tn. Then this recovers the divisorial log structure on 𝔸1 relative to the origin.

Example 2.17.

Consider the chart ℕ2→𝒪𝔸2 which sends (a,b)↦xa⁢yb. This recovers the divisorial log structure on 𝔸2 relative to the toric boundary.

Definition 2.18.

A chart for a log map f:(X,MX)→(Y,MY) is a triple (PX→MX,QY→MY,Q→P), where PX→MX and QY→MY are charts and the diagram

commutes.

Definition 2.19.

A log scheme (X,MX) is finitely generated, fine, integral, saturated if there exists an étale local chart PU→MU for a cover U→X such that P is finitely generated, fine, integral, or saturated, respectively.

2.4 Log smoothness.

Definition 2.20.

Consider a commutative diagram

where j is a strict closed immersion (here, this means that MT0=j∗⁢MT1). A morphism f:X→Y of fine log schemes is log smooth (resp. log étale) if

  1. (1)

    The underlying map f¯:X¯→Y¯ is locally of finite presentation;

  2. (2)

    For any commutative diagram as above, étale locally on T1 there exists a (resp. there exists a unique) morphism g:T1→X such that ϕ=g∘j and ψ=f∘g.

We will now give a chart local criterion for log smoothness.

Theorem 2.21 ([Kat89, Theorem 3.5]).

Let f:X→Y be a morphism of fine log schemes. Assume we have a chart Q→MY where Q is a fine monoid. Then f is log smooth if and only if étale locally on X there exists a chart for f extending Q→MY satisfying

  1. (1)

    The kernel and the torsion part of the cokernel of Qgp→Pgp are finite groups whose orders are invertible on X;

  2. (2)

    The induced morphism X¯→Y¯×Spec⁡ℤ⁡[Q]Spec⁡ℤ⁡[P] is étale in the classical sense.

Example 2.22.

Let X be an affine toric variety given by Spec⁡ℂ⁡[P] for a monoid P. Note that Pa≃MX,X∖T induces the divisorial log structure. Then consider the map f:X→pt, where the point has a trivial log structure. We note that the morphism 0→Pgp has no kernel and the cokernel is torsion-free, and the map

is an isomorphism, so any toric variety is log smooth over a point with trivial log structure.

From this example, we can morally think of log smoothness as being locally toric relative to the base.

Example 2.23.

Recall the family x⁢y=t with log structure relative to the central fiber over 𝔸1 with the divisorial log structure relative to 0. The map of charts we will consider is in fact Δ:ℕ→ℕ2. Taking the groupification, the map becomes Δ:ℤ→ℤ2. The kernel is trivial and the cokernel is torsion-free, so we have verified the first condition. Then we consider the map

which is an isomorphism, so this family is log smooth.

2.5 Stack of log structures.

The stack of log structures and its relation to log smoothness are due to Olsson [Ols03].

Let S be a fine log scheme. We will consider the category 𝖫𝗈𝗀S whose objects are fine log schemes (T,MT)→S and whose morphisms are strict morphisms (T,MT)→(T′,MT′) over S.

  1. (1)

    This category 𝖫𝗈𝗀S is a category fibered in groupoids over 𝖲𝖼𝗁/S classifying all fine log schemes mapping to S;

  2. (2)

    𝖫𝗈𝗀S is an algebraic stack locally of finite type over S;

  3. (3)

    A morphism f:X→S is log smooth if and only if X→𝖫𝗈𝗀S is smooth. The same holds with smooth replaced by étale.

2.6 Tropicalization.

Let X be a fine log scheme. Then the tropicalization of X is defined to be the colimit

where the equivalence relation is given by if a point x specializes to y, then the induced map σx→σy induces an equivalence from σx to its image.

Now suppose there is a stratification of X by Xi such that M¯|Xi is constant. For any Xi, all points x∈Xi satisfy x∈{ηXi}¯, so we only need to consider the generic points of the strata and then glue various faces.

Example 2.24.

Consider X with the trivial log structure. Note that for any point x∈X, we have M¯X,x=0, so σX={0}. Then Σ⁡(X)={0}.

Example 2.25.

Consider 𝔸1 relative to the origin (with divisorial log structure). It is easy to see that σ0=Hom⁡(ℕ,ℝ≥0)=ℝ≥0 and ση={0}, so the tropicalization is ℝ≥0.

Example 2.26.

Consider 𝔸2 with the toric log structure (divisorial log structure relative to x⁢y=0). Then σ(0,0)=ℝ≥02, σ(x,0)=ℝ≥0, σ(0,y)=ℝ≥0, and σ(x,y)={0}, so we obtain the fan of 𝔸2 as a toric variety treated as a cone complex.

Example 2.27.

Let XΣ be a toric variety with the divisorial log structure relative to XΣ∖T. Then

so a possible slogan is that tropicalizing a toric variety recovers the structure of ordinary toric geometry (minus the embedding of the fan in some real vector space).

Example 2.28.

Let X=Spec⁡k⁡[x,y,t]/(x⁢y−tℓ) with divisorial log structure over x⁢y=0. Then there is a chart given by

Note that MX=Qa=Q⊕α−1⁢(𝒪X×)𝒪X× and M¯X=Q. Therefore, we have

at the singular point q, which recovers the cone which gives X as a toric variety. At the generic point ηx of x=0, we see that M¯X,ηx=ℕ⁢et, and therefore σηx=ℝ≥0. Similarly, we see that σηy=ℝ≥0. The remaining problem is gluing.

Note that M¯X,q→M¯X,ηx is given by et↦et, ex↦ℓ⁢et, and ey↦0, and therefore the map σηx→σq sends ρt to (ρt,ρx), where ρx=ℓ⁢ρt.

The morphism X→𝔸t1 giving a family of curves with nodal central fiber has tropicalization given by projecting the cone onto ℝ≥0.

2.7 Log curves.

For now, all monoids and log schemes will be fine (or even fine saturated). There are two reasons for this:

  1. (1)

    If P is a fine monoid, its dual is fs and sharp;

  2. (2)

    If X=(X¯,MX) is a fine log scheme, for any geometric point x¯→X, there exists an étale neighborhood of x¯ which has a chart.

We will now discuss various geometric properties of maps between monoids.

Definition 2.29.

A map θ:P→Q is integral if the map Spec⁡ℂ⁡[Q]→Spec⁡ℂ⁡[P] is flat.

For example, any nonzero map ℕ→ℕ and the diagonal Δ:ℕ→ℕ2 are integral. If a map is integral, then for any P→P′, the monoid P′⊕PQ is fine.

Example 2.30.

Pushouts may not be fine, for example consider

Definition 2.31.

A map f:X→Y between fine log schemes is integral if at every geometric point of X, the map

is integral.

Definition 2.32.

A map between fs monoids P→Q is saturated if for all P→P′, the monoid P′⊕PQ is also fs.

Example 2.33.

The map Δ:ℕ→ℕ2 is saturated, but ℓ:ℕ→ℕ is not.

Definition 2.34.

A log curve over an fs log scheme S is a log smooth and integral map π:𝒞→S between fs log schemes such that the underlying geometric fibers are reduced connected curves.

Proposition 2.35 ([Kat89]).

Let f:X→Y be a log smooth and integral map between fs log schemes. Then f¯ is flat.

Theorem 2.36 ([Kat00]).

The underlying family of a log curve is a family of nodal curves.

Example 2.37.

Let P=ℕ∖{1}. Then Spec⁡(P→k⁡[P]) is log smooth over k, but Spec⁡k⁡[P] is a cusp.

Definition 2.38.

An n-pointed log curve over an fs base S consists of the data (π:𝒞→S),p→={pi}i=1n such that

  1. (1)

    The underlying family is a prestable curve;

  2. (2)

    π is a log smooth map between fs log schemes;

  3. (3)

    On the smooth locus U of π¯, we have

2.8 Canonical log structures on prestable curves.

Let 𝔐g,n be the moduli of n-pointed prestable curves. Then the singular locus is a normal crossings divisor, so there is a canonical log structure

If we consider the universal family ℭg,n, we can form π−1⁢(∂𝔐g,n)⊆ℭg,n. Therefore, we have a map

We can also consider the divisors pi⁢(𝔐g,n)⊂ℭg,n, which also define log structures

Therefore, we can define a log structure

Then if we consider

we can lift π to a map of log stacks.

Theorem 2.39 ([Kat00]).

The map π is a log smooth integral map.

Definition 2.40.

For a family (C¯/S¯,p→) of prestable curves, we can pull back along the diagram

and endow the left edge with a canonical log structure.

Remark 2.41.

Let S¯=Spec⁡k. If Scan=Spec⁡(ℕr→k), then r is the number of nodes on C¯. Working étale locally on a node q, the log structure looks like

Theorem 2.42 ([Kat00]).

Let (𝒞/S,p→) be an n-pointed log curve on an fs base S. Then there exists a unique MScan→MS such that

is a Cartesian diagram in the category of fs log schemes.

A slogan is that a log curve is a prestable curve together with MScan→MS.

Example 2.43.

Let S¯=Spec⁡k⁡[s] and T¯=Spec⁡k⁡[t]. Also choose ℓ∈ℕ. Then consider the diagram

On log structures, consider the case where the bottom left arrow corresponds to 1↦ℓ on ℕ. Then at the top left corner, the monoid becomes ℕ2⊕ℕ,ℓℕ. If we call this P, then the map k⁡[P]→k⁡[ℕ] is not a family of log curves with the canonical log structure, but is a family of log curves.

2.9 Moduli of log curves.

Definition 2.44.

A map between n-pointed log curves (𝒞,S)→(𝒞′,S′) is a pair (ℓ:𝒞→𝒞′,Θ:S→S′) such that Θ is strict and the diagram

is Cartesian. Denote by 𝔐g,nlog the fibered category parameterizing n-pointed log curves.

Proposition 2.45 ([Kat00; Ols03]).

We have 𝔐g,nlog≃𝖫𝗈𝗀(𝔐g,n,M𝔐g,ncan).

In particular, for a scheme T¯, the groupoid 𝔐g,nlog⁢(T¯) parameterizes log structures MT together with log curves 𝒞/T.

Consider a log curve π:𝒞→S,p→ over S=Spec⁡(Q→k). Denote π¯♭:π¯∗⁢M¯S→M¯𝒞. At a generic point or a general smooth point η, the map is

At a marked point pi∈{η}¯, we have M¯𝒞,pi=Q⊕ℕ. In this case we have

the specialization map and the pullback

At a node q, we have M¯𝒞,q=Q⊕ℕℕ2. Let ρq denote the image of 1 in Q, which is the edge length or smoothing parameter at q. Denote by η¯1 and η¯2 the generic points of the two branches of q. Then the generalization maps are

2.10 Tropicalization of log curves.

We will only consider the situation over a log point S=Spec⁡(Q→k). We will consider the category of cones 𝖢𝗈𝗇𝖾, whose objects are pairs (w,Nw) such that w⊆Nw⊗ℤℝ, where w is a full-dimensional strongly convex rational polyhedral cone. The morphisms are maps (w1,Nw1)→(w2,Nw2), which consist of

such that χℝ⁢(w1)⊆w2.

Example 2.46.

Let X be an fs log scheme and consider ((M¯X,x∨)ℝ,(M¯X,x∨)ℤ).

We will now consider the category 𝖦𝖢𝖢 of generalized cone complexes. Its objects are topological spaces |Σ| together with a presentation as the colimit of a diagram in 𝖢𝗈𝗇𝖾 with all arrows being inclusions of faces. Its morphisms are continuous map |Σ|→|Σ′| such that for each (w,Nw)∈Σ, the map w→|Σ′| factors through (w,Nw)→(w′,Nw′).

Example 2.47.

If f:X→Y is a morphism of fs log schemes, the map f¯♭:f∗⁢M¯Y→M¯X induces Σ⁡(X)→Σ⁡(Y) on tropicalizations.

Definition 2.48.

A cone complex is a generalized cone complex Σ such that for each (σ,Nσ)∈Σ, the map σ→|Σ| is injective.

Theorem 2.49 ([ACP15]).

If X is fs and log smooth over ℂ¯ with Zariski log structure, then Σ⁡(X) is a cone complex.

For a monoid Q, we will denote wQ=Qℝ∨=Hom⁡(Q,ℝ≥0) and the lattice Hom⁡(Q,ℤ)=Qℤ∨. At a smooth point, we have

At a marked point, we see that

At a node, we have

Note that the cone is in fact equal to

Now that we understand the cones associated to every point, we now need to discuss the face maps. The generalization maps corresponding to marked points are

and the associated face maps are

At a node, the two face maps are

and

Example 2.50.

Consider a curve which is a union of two rational curves with a marked point p on one branch. Then let S¯=Spec⁡ℂ. Then we see that

Q=ℕ, and Qℝ∨=ℝ≥0. We also see that M¯C,q=ℕ⊕ℕℕ2 and M¯C,p=ℕ⊕ℕ.

wp=ℝ≥0×ℝ≥0wη1wq{(a,b)|b≤a}⊆ℝ≥02wη2

2.11 Tropical curves.

Definition 2.51.

A family of tropical curves (G,g→,ℓ→) over a cone w,Nw consists of

  1. (1)

    A connected graph G=V⁡(G)∪E⁡(G)∪L⁡(G);

  2. (2)

    A leg ordering m→:L⁡(G)→∼{1,…,n};

  3. (3)

    A genus function g→:V⁡(G)→ℕ;

  4. (4)

    An edge length function ℓ:E⁡(G)→Hom⁡(w∩Nw,ℕ)∖{0}.

Now we will construct Σ⁡(G,ℓ). For v∈V⁡(G), we set wv=w. For e∈E⁡(G), we set

For legs ℓ∈L⁡(G), we set wℓ=w×ℝ≥0. We now need to define the face maps. If ℓ is incident to v, then the face map is

If e is incident to two vertices v1,v2, then the face maps are

Remark 2.52.

The choice of v1 and v2 does not change the cone complex up to isomorphism, but it does matter when we want to define contact orders for log maps. However, in general there is no canonical way to make this choice.

2.12 Punctured curves.

Punctured logarithmic curves and maps are developed in [ACGS25].

We consider a nodal curve

with log structures given by ℕ2 with chart (a,b)↦xa⁢yb and ℕ→k being the standard log point. Now consider C¯1=Z⁡(x)=Spec⁡k⁡[y] and mark the node by p:Spec⁡k→Spec⁡k⁡[y]. Note that

Denote MC1∘=MC|C1. Then we see that on log structures we have the diagram

In particular, we can compute the stalk M¯C1∘|p=ℕ2, and therefore we have

However, we should have the map

and therefore C1 is not a log curve. We can then consider the diagram

Therefore, we will consider extending the log structures.

Recall that for a log curve, we have M𝒞=M𝒞′⊕𝒪𝒞×P𝒞, where P𝒞 is the divisorial log structure by the markings.

Definition 2.53.

A punctured curve over an fs log scheme S consists of (𝒞∘→𝑝𝒞→𝜋S,{pi}i=1n), where (𝒞→𝜋S,p→) is a log curve and the map p:𝒞∘→𝒞 is a map of fine log schemes satisfying the following conditions:

  1. (1)

    The underlying map p¯ is an isomorphism;

  2. (2)

    On log structures, the map p♭:M𝒞↪M𝒞∘ is an isomorphism away from the pi and induces

  3. (3)

    For any marking pi and any s∈M𝒞∘,pi∖M𝒞,pi, we have α𝒞∘⁢(s)=0.

Roughly speaking, a punctured curve is a log curve plus the data of a commutative diagram

Example 2.54.

Let (𝒞¯→S¯=Speck,p) be a smooth 1-pointed curve.

  1. (1)

    Let (𝒞/S,p) be endowed with the canonical log structure (so S has no log structure). Let M𝒞=P. Then any puncturing p♭:M𝒞↪M𝒞∘⊂Pgp must be trivial. The reason is that M𝒞∘ cannot have a section s such that α𝒞∘⁢(s)=0. If s∈P=M𝒞, then α𝒞∘⁢(s)=α𝒞⁢(s). Otherwise, if s∉P, then because α𝒞∘⁢(−s) is nonzero, so is α𝒞∘⁢(s).

  2. (2)

    Let M𝒞∘⊆(ℕ⊕𝒪𝒞×)⊕Pgp be a puncturing. Choose a local generator x of 𝒪⁡(−p). Then any s∈M𝒞∘,p has the form s=((n,φ),xm) where n∈ℕ and φ∈𝒪𝒞,p×. If m<0, then we require that α𝒞∘⁢(s)=0. This implies that α𝒞⁢((n,φ),0)=0, and therefore

    In fact, any fine submonoid of the right hand side containing ℕ⊕ℕ can be realized by a puncturing.

  3. (3)

    Let S=Spec⁡(ℕ→k). Consider M𝒞∘ such that M¯𝒞∘,p=ℕ2+ℕ⁡(2,−2). Because (1,−1)∉M¯𝒞∘,p, we see that M𝒞∘ is not saturated.

Lemma 2.55 ([ACGS25]).

Let Q be a sharp fs monoid. Let w≔Hom⁡(Q,ℝ≥0)=Qℝ∨. Let Q∘⊆Q⊕ℤ be finitely generated with Q⊕ℕ⊊Q∘ and Q∘∩{0}×ℤ<0=∅. Then there exists a nonzero, concave, piecewise linear function

with rational slopes such that

Example 2.56.

Let Q=ℕ2 and set

Then

3 Intro to GW theory with a view to higher genus (Patrick Lei)

3.1 Introduction to GW theory.

For the virtual fundamental class construction, see [BF97]; for the axioms and quantum cohomology, see [KM94].

The goal of these lectures is to discuss Gromov-Witten theory in a way that is amenable to doing calculations in higher genus. Gromov-Witten theory concerns itself with integrals on the moduli space ℳ¯g,n⁢(X,β), where X is a smooth projective algebraic variety and β∈H2⁢(X,ℤ) is an effective curve class. There is a diagram

where 𝒞 is the universal curve, σi are the sections of π which give the marked points, f is the universal stable map, and evi=f∘σi are evaluation morphisms. While ℳ¯g,n⁢(X,β) is generally singular, not equidimensional, and reducible, there is a special class

which is known as the virtual fundamental class.

Exercise 3.1.

Use standard arguments in deformation theory to verify that the “expected dimension” of ℳ¯g,n⁢(X,β) is the number given above.

Under the assumption that X is smooth and projective, ℳ¯g,n⁢(X,β) is proper, so we may define GW invariants

for αi∈H∗⁢(X). Here, the ψi are defined by ψi=c1⁢(σi∗⁢(ωπ)).

Unfortunately, calculating individual Gromov-Witten invariants is very difficult, so we will consider the even more difficult problem of studying generating series of GW invariants, for example

when X is a Calabi-Yau threefold. Here, q is a formal variable which keeps track of the curve class and we will denote the ring ℚ⟦qβ|β∈H2(X,ℤ)eff⟧ by Λ. There are some other structures we will consider. First, we may consider the quantum cohomology ring, whose product a⋆𝐭b is given by the formula

Here a,b,c,𝐭∈H∗⁢(X). We will also consider the quantum connection. Let {ei} be a basis for H∗⁢(X). Then the quantum connection is the connection on the trivial bundle with fiber H∗⁢(X) over H∗⁢(X)×ℙz1 given by the formula

Sometimes, it is useful to also consider the connection in the z-direction given by

where μX⁢(α)=deg⁡(α)−dimX2⁢α is the grading operator and

is the Euler vector field. Because we will be mostly interested in Calabi-Yau varieties, we will not consider this.

3.2 Enumerative mirror symmetry in genus zero.

The impetus for mathematicians to develop and study Gromov-Witten theory came in the early 1990s, when Candelas, de la Ossa, Green, and Parkes [COGP91] predicted the number of rational curves of low degree on a general quintic threefold by converting the problem to one of computing period integrals on a different Calabi-Yau threefold called the mirror quintic (equivalently, studying its variation of Hodge structure). In genus zero, all invariants can be recovered from the J-function J𝐭⁢(q,z)∈H∗⁢(X,Λ)⁢[z], which is defined by

for any 𝐭∈H∗⁢(X,Λ)⁢[z] and α∈H∗⁢(X), where (−,−) is the Poincaré pairing. Therefore, the problem of computing all genus-zero Gromov-Witten invariants with descendants (these are the ψ-classes) reduces to the problem of computing the J-function.

Remark 3.2.

We expand 1z−ψ as ∑k=0∞ψk⁢z−k−1.

We should note that the J-function can be obtained via a fundamental solution to the quantum differential equation, which is given by

This has the property that S𝐭∗⁢(−z)⁢S𝐭⁢(z)=Id, which is usually known as the symplectic property. From this, we can recover the J-function by z⁢S𝐭∗⁢(z)⋅1=J𝐭⁢(q,z).

In physics, mirror symmetry is a duality between two different theories called the A-model and B-model, so now we will need to introduce the B-model object. In the interest of time (and notation), we will assume that X is a toric variety and that −KX is nef. Let D1,…,Dr be the complete set of torus-invariant divisors in X. Then define the I-function

Note here that we expand any (Di+m⁢z)−1 which appears as 1m⁢z⁢(1−Dim⁢z+⋯), or namely, as a geometric series in Dim⁢z. In addition, the I-function is annihilated by a differential equation known as the Picard-Fuchs equation, which is some hypergeometric type equation.

Example 3.3.

If 𝐭=0 and X=ℙ1, we obtain

so in fact we see the Bessel function of the first kind as the leading term.

If X is a complete intersection in a toric variety Y cut out by line bundles L1,…,Lk, all of which satisfy H1⁢(f∗⁢Li)=0 for any genus zero stable map to Y (this is called being convex), then the quantum Lefschetz theorem [CG07] allows us to replace [ℳ¯g,n⁢(X,β)]vir with e⁡(⨁π∗⁢f∗⁢Li)∩[ℳ¯g,n⁢(Y,β)]vir and perform all computations on Y. In this case, the I-function is defined to be

where we abuse notation and conflate the Li with their first Chern classes.

Example 3.4.

For the quintic threefold, which is cut out by a section of 𝒪ℙ4⁢(5), the I-function at 𝐭=0 is given by

The content of the genus zero mirror theorem is a correspondence between these two objects. More precisely, in the case where −KX is nef, they are related by a change of variables.

Theorem 3.5 ([Giv98; LLY97]).

Write the I-function with 𝐭=0 as I0⁢(q)⋅z+I1⁢(q)+O⁡(z−1). Then we have

where τ⁡(𝐭)=𝐭+I1⁢(q)I0⁢(q).

Example 3.6.

When X=ℙ1, note that I0⁢(q)=1 and I1⁢(q)=0, so in fact, we have the equality

Example 3.7.

When X is the quintic threefold, we will rewrite the components of the I-function as I0⁢(q)⋅z+I1⁢(q)⁢H+O⁡(z−1) for convenience. We have

so there is a nontrivial mirror map in this case. Specializing to 𝐭=0, we see that τ=I1⁢(q)⁢HI0⁢(q), so we obtain

By the divisor equation, the left hand side also equals J⁡(q⁢eI1/I0,z), and we will also refer to this as the mirror map. Therefore, we can use the mirror theorem to compute ⋆τ. The quantum connection in the direction of H2 becomes the ODE

where D≔q⁢ddq (here, the coordinate is log⁡q) and I11≔1+D⁡(I1⁢(q)I0⁢(q)). An explicit computation using the mirror theorem and the results of Zagier–Zinger [ZZ08] yields

Because Sτ begins with the identity, this computes H⋆τ.

3.2.1 Oscillating integrals.

We will now reinterpret the genus zero mirror theorem, and we will return to this interpretation when we discuss higher genus invariants. In particular, we will interpret the I-function as an integral on some mirror family to X, which for us will be a toric variety. Let

be the fan sequence of X. Applying Hom⁡(−,ℂ×), we obtain the exact sequence

Suppose that the map Λ→ℤr is given by the matrix (mi⁢j)i=1,…,rj=1,…,ρ⁡(X). Then we consider the variety

which is mirror to X and consider the superpotential f=x1+⋯+xr. If we wish to consider the equivariant theory with equivariant parameters λ1,…,λr acting on each homogeneous coordinate of X, we will consider fλ=∑i=1r(xi−λi⁢log⁡xi).

Example 3.8.

When X=ℙn, the mirror to X is the superpotential ∑i=1n+1(xi−λi⁢log⁡xi) on the variety Eq=(x1⁢⋯⁢xn+1=q).

We now choose an appropriate Lefschetz thimble Γα near each critical point (xα) of the (equivariant) superpotential and consider the integral

This corresponds in the I-function to replacing each Di in the expression by Di−λi (turning on the equivariant parameters) and in the J-function to considering the fully equivariant GW theory of X. In genus zero, the limit as λi→0 exists, so we can safely recover the original mirror theorem from the following:

Theorem 3.9 ([Giv96]).

This integral is annihilated by the same Picard-Fuchs equation as the I-function, and up to various prefactors, its asymptotic expansion as z→0− along the negative real axis coincides with the restriction of the I-function corresponding to the T-fixed point on X which corresponds to the critical point (xα).

3.2.2 The Lagrangian cone.

The quantum cohomology of X is an example of a Frobenius manifold, which we will see later can be produced from any cohomological field theory. Write V=H∗⁢(X) and let (eμ) be a basis for V, so we can write 𝐯=∑μ,ntnμ⁢eμ⁢zn∈V⁡⟦z⟧. The structure of a Frobenius manifold comes from the function

which is also called the genus zero descendant potential of X. It satisfies the string equation, dilaton equation, and topological recursion relations:

To state the following result, we consider the infinite-dimensional vector space V⁢⦇z−1⦈ (or a completion of this) with the symplectic form

We also consider the new variable2 𝐪=𝐯−z, so ℱ0 is now a function near 𝐪=−z.

Theorem 3.10 (Givental [Giv04]).

ℱ0 satisfies the above PDEs if and only if the graph ℒ of dℱ0 is a Lagrangian cone with vertex at 𝐪=0 such that its tangent spaces L are tangent to ℒ exactly along z⁢L⊂L.

We may recover the Lagrangian cone ℒ by the following procedure. We find a fundamental solution

to the quantum connection, which satisfies the equation

Then, setting J≔z⁢S∗⁢(z)⁢𝟏, we can then recover the Lagrangian cone by the following procedure:

  • •

    The derivatives ∂∂tμ⁢J form a basis of L∩V⁢⟦z−1⟧;

  • •

    This implies that z⁢∂2∂tμ⁢⁡∂tν⁢J∈L∩V⁡⟦z−1⟧. Writing these in terms of the first derivatives and using the fact that J solves the quantum connection, we recover the Frobenius structure and hence the Lagrangian cone.

Using this, we can interpret the genus zero mirror theorem as:

Theorem 3.11 ([Giv98]).

The I-function I𝐭⁢(q,z) lies on the Lagrangian cone of X.

3.3 Moduli of curves and CohFTs.

The notion of a cohomological field theory was introduced by Kontsevich–Manin [KM94].

We will denote by ℳ¯g,n the moduli space of stable curves of genus g with n marked points. This is nonempty if and only if 2⁢g−2+n>0, in which case it is a Deligne-Mumford stack of dimension 3⁢g−3+n. There is a combinatorial structure to the collection of ℳ¯g,n, which is given by a collection of morphisms.

  • •

    There is the gluing morphism

    which takes two curves and glues them along the last marked point to form a node;

  • •

    There is the self-gluing morphism

    which glues the last two marked points.

There is of course another interesting map, which is the forgetful map

given by deleting the last marked point and then stabilizing. We are now able to define our enumerative theories of interest, which include Gromov-Witten theory.

Definition 3.12.

Given a vector space V with a nondegenerate symmetric bilinear form η and a unit element 𝟏∈V, a Cohomological Field Theory (CohFT) Ω on V is a collection Ωg,n of Sn-equivariant linear maps

which satisfy the basic identity

and the following combinatorial identities (we will let eμ be a basis of V such that e1=𝟏 and eμ be the dual basis):

If in addition the identity

is satisfied, then we say the CohFT has a flat unit (or satisfies the string equation).

All of this can be generalized to super vector spaces, but for simplicity we will not deal with this case.

Example 3.13.

The most important example of a CohFT is the Gromov-Witten theory of a smooth projective variety X. Here, recall that the source of a stable map is a prestable curve, so there is a stabilization morphism

which forgets the map and stabilizes the curve. Then, working over the Novikov ring, we will set V=H∗⁢(X), η to be the Poincaré pairing, and 𝟏 to be the fundamental class of X. The linear maps Ωg,nX are given by the formulae

where evi:ℳ¯g,n⁢(X,β)→X is the i-th evaluation map.

Example 3.14.

Let π:𝒞g,n→ℳ¯g,n be the universal curve. Then the Hodge bundle is the vector bundle

We then consider the vector space V=ℂ with the usual pairing and define the CohFT Ω𝔼 by the formula

The gluing axioms are satisfied by the equation q∗⁢𝔼g+h=𝔼g⊕𝔼h and the exact sequence

Here, because genus zero components do not contribute to global sections of ω and 𝔼 does not depend on the marked points, this CohFT has a flat unit.

Remark 3.15.

We will see later that this is related to the GW CohFT of a point by the quantum Riemann-Roch theorem.

Given a CohFT Ω, we may produce invariants by pairing the classes Ωg,n⁢(𝐯) with various cohomology classes on ℳ¯g,n. The most important ones to consider are the classes

where si:ℳ¯g,n→𝒞g,n is the i-th marked point.

Remark 3.16.

In Gromov-Witten theory, these are called ancestors. There are also descendants, which are given by the same formula except using the moduli space 𝔐g,n of prestable curves instead of ℳ¯g,n, where we factor the morphism st defined above as

The descendant classes are denoted by ψi.

Example 3.17.

We will calculate an invariant which will appear in the GW theory of Calabi-Yau threefolds. For any CohFT Ω, consider the invariant

Using the second gluing equation, the degree zero part of Ω1,1⁢(𝟏) is equal to

which yields the (graded) dimension χ⁡(V) of V (for example, if V=H∗⁢(X) for a Calabi-Yau threefold X, this is the topological Euler characteristic of X). Therefore, we obtain

Example 3.18.

Consider the GW CohFT of a point. This is given by V=ℂ with the usual pairing and the formula

Then we define the invariants

Theorem 3.19 (Kontsevich [Kon92]).

The function

is annihilated by the operators

Remark 3.20.

This result (Virasoro constraints for a point) is equivalent to 𝒟 being a tau-function for the KdV hierarchy and has been generalized by various authors.

3.4 R-matrix action.

In the early 2000s, Givental discovered a remarkable property of axiomatic enumerative theories, namely that one can transform CohFTs by the action of matrices R⁡(z)=R0+R1⁢z+R2⁢z2+⋯∈Hom⁡(V,V′)⁢⟦z⟧ which satisfy the property

Traditionally, the literature requires that V=V′ and R0=Id, but the MSP group has removed this restriction and also allows dimV<dimV′.

In order to define the action of R on a CohFT Ω defined on V, we need to review some of the combinatorial structure of ℳ¯g,n. For a curve C∈ℳ¯g,n⁢(ℂ), we may consider the dual graph of C, which has vertices, edges, and legs, which are defined as follows:

  • •

    The vertices correspond to irreducible components of C and are labelled by a non-negative integer, which is the genus;

  • •

    The edges correspond to nodes of C (in particular, we allow loops);

  • •

    The legs correspond to marked points.

Any graph which appears as the dual graph of a stable curve is called a stable graph.

230αβ312ψψψ2ψ
Figure 1. Example of a stable graph in ℳ¯7,3 and associated tautological class. This stable graph describes the image of a map ℳ¯2,4×ℳ¯3,2×ℳ¯0,5→ℳ¯7,3.

The first type of action on CohFTs is a translation action. Let T∈V⁢⟦z⟧. For a CohFT Ω, we define

whenever the infinite sum makes sense.

Example 3.21.

If ΩX is the GW CohFT of a smooth projective variety X and τ∈H2⁢(X) is a divisor class, then the infinite sum makes sense and we can define the shifted GW CohFT ΩX,τ of X. For example, if X is the quintic threefold and τ=I1I0⁢H, then shifting to τ is the same as the mirror map q↦Q⁡(q)=q⁢eI1I0.

The second type of action is the action of a matrix R as in the beginning of this section. Let Gg,n denote the set of all stable graphs of genus g with n legs. Then we define

where ContΓ is defined by the following construction:

  • •

    At the i-th leg, we place the map

  • •

    At every edge, we place the bivector

  • •

    At every vertex, we place the linear map

  • •

    Finally, we consider the pushforward in cohomology along the gluing morphism ∏vℳ¯gv,nv→ℳ¯g,n.

Definition 3.22.

Let R be as above. Then we define the translation

Whenever it makes sense, we define

Theorem 3.23 (Chang-Guo-Li [CGL26, Theorem 2.7]).

Suppose we work with coefficients in ℂ⁢⟦q⟧. Then if

R.Ω is a well-defined CohFT. Moreover, if dimV=dimV′, 𝟏′ is a unit for R.Ω.

We would like to remark a bit more about the translation action when R0≠Id. For simplicity, we will assume that V=V′ and R0⁢𝟏=c⋅𝟏 for some constant c. Then we set T~R≔z⁡(𝟏−c⁢R⁢(z)−1⁢𝟏) and use the dilaton equation to compute

whenever this makes sense.

3.5 Reconstruction theorem.

Recall that every CohFT defines a Frobenius algebra. We will call a CohFT semisimple if the corresponding Frobenius algebra is semisimple.

Theorem 3.24 (Teleman [Tel12]).

Let Ω be a semisimple CohFT with flat unit and ω be its topological part. Then there exists a unique

such that

Example 3.25.

Recall the Hodge bundle CohFT from before. Recall that it is given by the formula

Taking the degree zero part, we see that ω𝔼 is the GW CohFT of a point. Using Mumford’s computation

where κm=p∗⁢ψ¯n+1m+1, B2⁢k are the Bernoulli numbers, and ι is the inclusion of the boundary up to a 2:1 étale cover, and the formula

for any vector bundle E (or by just using the quantum Riemann-Roch theorem3 directly), we obtain the R-matrix

As a sanity check, we will compute Ω1,1𝔼 using the R-matrix. First, we consider the stable graphs in Figure 2.

10
Figure 2. Stable graphs for g=1,n=1

The first graph Γ1 gives us the contribution

Using the formula B2=16 and the fact that dimℳ¯1,1=1, this becomes

The second graph does not receive tail contributions because dimℳ¯0,3=0. The constant term of the edge contribution is 112, so considering the automorphism and pushing forward to ℳ¯1,1 gives us

where Δ denotes the boundary divisor (with the correct stack structure). Because ψ1=κ1 in this case, we obtain

3.6 Operator formalism and geometric quantization.

We will return to the symplectic formalism of Givental. This is more convenient for certain computations, but it is in fact equal to what we have before, at least when we want to calculate generating functions. Recall that we had a Frobenius manifold structure on V and we considered the vector space 𝒱≔V⁡⦇z−1⦈ with symplectic form

If we consider the polarization given by 𝒱+=V⁡[z] and 𝒱−=z−1⁢V⁢⟦z−1⟧, then letting 𝐪 be as before (with the dilaton shift), let 𝐩 be coordinates on 𝒱− such that (𝐪,𝐩) form a system of Darboux coordinates for 𝒱.

Definition 3.26.

We will call any formal function of the form

an asymptotic element of the Fock space.

Then given such an asymptotic element of the Fock space, we will quantize (infinitesimal) symplectic transformations on 𝒱 by the following formulae:

In particular, this will allow us to understand expressions like R^⁢𝒟. However, we need to be careful because our formulae will involve both the fundamental solution Sτ and the R-matrix, and Sτ is a power series in z−1.4

Theorem 3.27 (Givental [Giv01a]).

An operator of the form S⁡(z−1)=Id+S1⁢z−1+⋯ acts on (asymptotic) elements of the Fock space by the formula

where W=∑η⁡(Wm⁢n⁢qm,qn) is defined by the formula

Operators of the form R⁡(z)=Id+R1⁢z+⋯ act by the formula

where V=∑η⁡(pm,Vm⁢n⁢pn) is defined by

For a semisimple CohFT, there is a system of canonical coordinates uα such that the 1-form du is a homomorphism of algebras Tv⁢V→ℂ. Then near a semisimple point, there is an asymptotic solution to the quantum connection of the form

where Ψ switches from flat coordinates to canonical coordinates and U=diag⁡(uα) is the matrix of canonical coordinates. Finally, define

A corollary of Teleman’s reconstruction theorem is the formula

for the Gromov-Witten theory of any smooth projective variety with semisimple quantum cohomology.

Example 3.28.

As a final example, we will apply Teleman’s theorem to compute F1 of any semisimple CohFT. Let eμ denote an idempotent basis for the quantum product. We will compute

Using the reconstruction theorem, we obtain

Using the identities

where Δμ≔η⁢(eμ,eμ)−1, we obtain the result

which can be placed in the suggestive form

This recovers a result which was already known to Givental.

3.7 How to compute the R-matrix.

We will conclude by explaining how to compute the R-matrix, starting in the setting where the Frobenius manifold is conformal (for example the Gromov-Witten theory of a smooth projective variety, r-spin curves, etc). In this example, we will consider the Frobenius manifold for 3-spin curves, following [PPZ15]. Let V be 2-dimensional with coordinates x and y, where x corresponds to the unit. The metric is given by

and the primary genus zero potential is F⁡(x,y)=12⁢x2⁢y+172⁢y4. There is an Euler vector field

which acts with weights −1 and −23 on ∂x and ∂y, respectively. The conformal dimension is δ=13, and we will define the degree operator

Here, ∂x has degree −16 and ∂y has degree 16.

To make our lives easier, we will set ϕ=y3 and define

and then compute in this frame. In this setting, quantum multiplication is given by

The last thing we need is that quantum multiplication by E is given by

We will now compute the R-matrix R=1+R1⁢z+⋯ by using the recursive equation

Setting Rk=(akbkckdk), we obtain the equation

The unique solution of this is in fact

If you are familiar with tautological relations on the moduli space of curves, you will see that the series ak+bk and ck+dk enter in the formulation of the Faber-Zagier relations and the formulation of Pixton’s relations.

In the case where the Frobenius manifold is not conformal (for example in equivariant Gromov-Witten theory), we will need to use an oscillating integral or some other technique to fix initial values in the R-matrix. We will give an example computation here, which is due to Givental [Giv01b]. In this setting, X=ℙn, but the argument can be made to work for any Fano toric variety. Recall that the equivariant mirror of ℙn is the superpotential

(here, we change the sign of the equivariant parameters for convenience). Therefore, the oscillating integrals will look like

This is annihilated by the operator

which is exactly the Picard-Fuchs equation satisfied by the I-function of projective space.

To expand this integral near the critical points, we first need to compute the critical points. Here, using coordinates t=log⁡q and Ti=log⁡xi, we obtain the equation

where L is the Lagrange multiplier. These satisfy the equation

and therefore for generic values of λ, near q=0, the possible values Li of the Lagrange multiplier are distinguished by which of the λi they begin with. Note that the function W, the values of the Lagrange multipliers, and therefore the critical points, all depend continuously on λ, so they admit non-equivariant limits at λ=0 when q≠0.

When q=0, the matrix Ψ transitions from the flat basis Hi to the basis

which are the leading terms of the canonical coordinates (simply compute the fixed points of the torus action via equivariant localization, and we will obtain this Lagrange interpolation). Therefore, we need to compute Lj⁢(z⁢q⁢ddq)⁢I in the q=0 limit. A direct computation shows that applying Li⁢(z⁢q⁢ddq) to the integral introduces an additional factor proportional to qxi to the integrand of the oscillating integral. Because we can permute the indices, we will compute only the case where i=n+1. Another direct computation shows that the integral vanishes at q=0 unless j=n+1, so in fact we see that R is diagonal in the classical limit. When j=n+1, we have

and the integral becomes

which is simply a product of gamma functions.

Lemma 3.29.

We have the asymptotics

Using this, we see that at q=0, the diagonal matrix element of the R-matrix is given by

By a result of Givental, this implies that this R-matrix recovers the equivariant GW theory of ℙn, and as a corollary, because the non-equivariant limit exists, we obtain Virasoro constraints for ℙn.

4 Decomposition theorem for semi-small maps (Xueqing Wen)

Question 4.1.

Let X be a (manifold, variety, analytic space, etc). How can we compute H∗⁢(X,ℂ)?

Unfortunately, this question is too hard, so we will replace it with a new question.

Question 4.2.

How can we detect H∗⁢(X,ℂ) using smaller spaces?

Theorem 4.3 (Künneth formula).

We have H∗⁢(X×Y,ℂ)=H∗⁢(X,ℂ)⊗H∗⁢(Y,ℂ).

Unfortunately, most spaces are not products, so we consider the following question.

Question 4.4.

Let f:X→Y be a morphism. Can we detect the cohomology of X using the cohomology of Y and the cohomology of the fibers of f?

This question is still very difficult, but at least we can study it. We will begin with the case where f is a fiber bundle between compact manifolds. The answer is of course the Leray or Serre spectral sequences, but before we get there, we need some preliminaries.

Definition 4.5.

A ℂ¯Y-local system on Y is a locally constant sheaf ℒ of vector spaces on Y.

For example, if Y is connected, then local systems correspond to representations of π1⁢(Y). Also, note that if Y is an algebraic variety, we will need to work with its analytification.

Example 4.6.

Let f be a fiber bundle with fiber F as above. Then Ri⁢f∗⁢ℂ¯X is a local system on Y with fiber Hi⁢(F,ℂ).

Proposition 4.7.

We have

Theorem 4.8 (Leray/Serre spectral sequence).

We have a spectral sequence

We will now consider the setting where f:X→Y is a smooth projective map between smooth projective varieties.

Theorem 4.9 (Deligne [Del68]).

We have the non-canonical isomorphism

At the level of cohomology, we have

This is a consequence of the hard Lefschetz theorem for projective varieties and also states that the Serre spectral sequence degenerates at the E2 page. In topology, this fails for the Hopf fibration.

We will now consider the setting where f:X→Y is a proper map between algebraic varieties. Then we have a Whitney stratification5 Y=⨆λYλ such that f−1⁢(Yλ)→Yλ is a C∞ fibration in possibly singular varieties.

Definition 4.10.

A ℂ¯Y-module ℱ on Y is called a constructible sheaf if there exists a Whitney stratification Y=⨆Yλ such that ℱ|Yλ is a local system for all λ.

With this definition, we may consider the derived category 𝒟cb⁢(Y) of constructible sheaves on Y. This assignment admits a six functor formalism, where the functors are f∗,f∗,f!,f!,⊗,ℋom, and satisfies Verdier duality via the functor

Definition 4.11.

A complex ℱ∙∈𝒟cb⁢(Y) is a perverse sheaf if it satisfies the following conditions:

  1. (1)

    (Support condition) We have dimsupp⁡ℋ−i⁢(ℱ∙)≤i for all i;

  2. (2)

    (cosupport condition) We have dimsupp⁡ℋ−i⁢(DY⁢ℱ∙)≤i for all i.

Remarks 4.12.
  1. (1)

    Perverse sheaves are not sheaves, but are complexes in degrees [−dimY,0].

  2. (2)

    They have two origins. One is intersection homology, and the other is the Riemann-Hilbert correspondence.

We will need to use the following fact: the category 𝖯𝖾𝗋𝗏⁡(Y) of perverse sheaves on Y is an abelian category and is moreover the heart of a t-structure on 𝒟cb⁢(Y). In particular, for any 𝒢∈𝒟cb⁢(Y), we can consider the perverse cohomology ℋip⁢(𝒢)∈𝖯𝖾𝗋𝗏⁡(Y).

Theorem 4.13 (Beilinson-Bernstein-Deligne-Gabber [BBD82]).

Let f:X→Y be a proper morphism between algebraic varieties. Assume that X is smooth and connected.6 Then

Moreover, each ℋip⁢(R⁢f∗⁢ℂ¯X) is semi-simple (meaning each subobject is a direct summand).

Definition 4.14.

Let f:X→Y be a proper surjective morphism with X smooth and irreducible. Let Y=⨆Yλ be a stratification such that f−1⁢(Yλ)→Yλ is a C∞-fibration. We say that f is semi-small if for all yλ∈Yλ, we have codimY⁡Yλ≥2⁢dimf−1⁢(yλ). If equality holds, we say that Yλ is relevant. If the only relevant stratum is the dense one, then we say that f is small.

Theorem 4.15 ([CM02]).

If f:X→Y is semi-small, then R⁢f∗⁢ℂ¯X⁢[dimX] is a perverse sheaf on Y.

Proof.

For yλ∈Yλ⊆Y, we see that

Note that this vanishes if i>2⁢dimf−1⁢(yλ)≤codimY⁡(Yλ). If we vary over all strata Yλ, we obtain the support condition. Because f is proper, we have Rf∗=f!, and by using DYRf∗=f!DX, we obtain the cosupport condition. □

Examples 4.16.

We now give some examples of semi-small maps.

  1. (1)

    Any generically finite proper map between surfaces is semi-small.

  2. (2)

    Let Y={x⁢y=tℓ+1}⊆ℂ3. Then there is a crepant resolution f:Y~→Y of singularities, and the fiber over the singular point is a chain of ℙ1s, so it is semi-small (see also the previous example).

Theorem 4.17 (Kaledin [Kal06]).

A projective birational map from a nonsingular holomorphic symplectic variety is semi-small.

Example 4.18.

For example, the Springer resolution and the Grothendieck-Springer resolution are semi-small. In addition, the Hilbert-Chow morphism for the Hilbert scheme of points on a surface is semi-small.

Let f:X→Y be semi-small and consider a stratification Y=⨆Yλ. If Yλ is relevant, consider

where dλ=2⁢dimf−1⁢(yλ). Then ℒλ is semi-simple. Here, note that ℒ|λ=Hdλ(f−1(yλ),ℂ). Now let IC⁡(ℒλ) be the intermediate extension of ℒλ⁢[dimYλ] on Y¯λ, which is a perverse sheaf.

Theorem 4.19 ([CM02]).

Let f:X→Y=⨆Yλ be semi-small. Then there is a canonical decomposition

Example 4.20.

Let G be a reductive group with Lie algebra 𝔤 and choose a Borel B. Then

is small. Therefore, we have

Recall that 𝔤~rs→𝔤rs is a W-torsor. Then for a∈𝔤rs, we have ℒ|a≅ℂ[W], and therefore ℒ=R0⁢π∗⁢ℂ¯𝔤~rs. This implies that W acts on ℒ, and this action can be extended to IC⁡(ℒ) and also on R⁢π∗⁢ℂ𝔤~, and in particular acts on the fiber R⁢π∗⁢ℂ𝔤~|a for any a∈𝔤.

Example 4.21.

Consider the Springer resolution μ:T∗⁢(GL3/B)→𝒩=𝕆(13)⊔𝕆(2,1)⊔𝕆(3). This map is semi-small, but unfortunately all nilpotent orbits are relevant. Note that the fiber above the (2,1) orbit is a union of two ℙ1s glued at a node, the fiber above the (3) orbit is a point, and the fiber above the (13) orbit is GL3/B.

We see that ℒ(3) must be trivial because μ is an isomorphism above this orbit. Then ℒ(2,1)=R2⁢μ∗⁢ℂ¯ is a rank 2 local system on 𝕆(2,1). Finally, ℒ(13)=R6⁢μ∗⁢ℂ¯ is a rank 1 local system on 𝕆(13)=pt. Therefore, we have

Restricting to b∈𝕆(3), we obtain ℒ(3)=H0⁢(pt). Restricting to a∈𝕆(2,1), we obtain

Restricting to 0=𝕆(13), we obtain

As representations of W=S3, these correspond to the trivial representation, two copies of the standard representation, and the sign representation. For the Springer-theoretic construction and these decompositions, see [BM83].

5 Quantum K-theory (Xiaohan Yan and You-Cheng Chou)

5.1 Overview.

Gromov-Witten invariants are some way of “counting” curves in a smooth projective variety X. GW theory produces numbers of the form

for various classes αi∈H∗⁢(X), genus g, number n of marked points, and curve class β∈H2⁢(X,ℤ). This has been used to study mirror symmetry, and outside of mirror symmetry has produced birational invariants. Our goal now is to lift this entire story to K-theory. Quantum K-theory produces numbers

for V1,…,Vn∈K⁡(X). These satisfy various difference equations7 and have found various applications in 3D mirror symmetry and representation theory.

5.2 K-theory.

Definition 5.1.

Let X be a smooth projective variety. Then the K-theory K⁡(X) of X is the Grothendieck group of vector bundles on X with the relation that if

is a short exact sequence, then V2=V1+V3.

Remark 5.2.
  1. (1)

    K⁡(X) is a ring with the operations of direct sum and tensor product;

  2. (2)

    The Chern character gives an isomorphism K⁡(X)→A∗⁢(X), at least working over ℚ.

Example 5.3.

Let X=ℙn and set P=𝒪⁡(−1). Then we have

For a Deligne-Mumford stack which is not assumed to be smooth, then there are two versions

where K0 means coherent sheaves and K0 means perfect complexes. These coincide on a smooth projective variety, so from now on K⁡(X) will mean K0⁢(X). There is a tensor product

and a pushforward

along proper morphisms. If f is a perfect morphism (for example flat, regular embedding, etc), then there is a pullback

5.3 Moduli of curves.

Recall the moduli space

of stable maps to X. This has evaluation maps

There are also the universal cotangent lines Li=σi∗⁢ω𝒞/M¯.

The moduli space is a proper DM stack which has a perfect obstruction theory, which produces a virtual structure sheaf 𝒪g,n,βvir. This comes from the structure of a virtual tangent bundle, where

Roughly, if M¯ is the zero locus of a section of a vector bundle E on a variety Y, then 𝒪vir=𝒪Y⊗Eu⁡(E), where the K-theoretic Euler class is Eu⁡(E)=∑(−1)i⁢Λi⁢E∨.

5.4 KGW invariants.

For quantum K-theory, see [Lee04]; for the permutation-equivariant theory, see [Giv17; Giv15b]. Toën [Toë99] proves the Riemann–Roch formula for Deligne–Mumford stacks used below.

Definition 5.4.

K-theoretic GW invariants are defined by

If V1,…,Vn are actual vector bundles, then the invariant is an integer, but generally we will consider K0 with coefficients in a field of characteristic zero. We may also define gravitational descendents

Using the action of Sn on M¯g,n⁢(X,β) given by permuting the marked points, then the objects ⨂i=1nevi∗⁡V and ⨂i=1nLik are Sn-equivariant.

Definition 5.5.

We define permutation-invariant K-theoretic GW invariants by

Given t⁡(q)=∑n∈ℤtn⁢qn∈K⁡(X)⁢[q,q−1], we can define

More generally, if H=Sa1×⋯×Sam⊂Sn, we can define

Remark 5.6.

We would like to discuss some relation to ordinary GW invariants. Recall that Hirzebruch-Riemann-Roch states that

In the case where M is a smooth DM stack, Töen proved the formula

where I⁢M=M×Δ,M×M,ΔM is the inertia stack of M and the tildes mean taking the orbifold Chern character and Todd class.

Example 5.7.

If M=B⁢μr, then I⁢M=⨆g∈μrB⁢μr. If M=ℙ⁡(1,r) is a weighted ℙ1, then I⁢M=M⊔⨆g≠1∈μrB⁢μr.

Example 5.8.

Let M=B⁢G for a finite abelian group G and V be a vector bundle on B⁢G, which is the same thing as a representation of G. Then χ⁡(M,V)=dimVG. On the other hand, we have

so we have verified Riemann-Roch in this setting.

5.5 Axiomatic properties.

Lemma 5.9 ([Giv15b]).

We have the string equation

Proof.

The proof is the same as in GW theory and comes from computing π∗⁢Li along the map π:M¯g,n+1⁢(X,β)→M¯g,n⁢(X,β). □

Lemma 5.10 ([Giv15b]).

We have the dilaton equation

Proof.

We use the fact that

where π forgets the first marked point and L=ωπ⁢(p1+⋯+pn). This follows from the fact that there is a residue map

The middle term is a direct sum of k trivial representations and one standard representation. □

Notation 5.11.

Fix τ,t∈K⁡(X). We will denote

We will let {ϕα} be a basis of K⁡(X) and {ϕα} be the dual basis. Then define the pairing

This is not quite the right pairing, so we will consider

We will let gα⁢β=(gα⁢β)−1 and Gα⁢β=(Gα⁢β)−1.

Lemma 5.12 ([Giv00]).

Given A,B,C,D∈K⁡(X)⁢[q,q−1], we have the WDVV equation

Proof.

Pull back the equality of the boundary divisors on M¯0,4=ℙ1. In K-theory, we need to apply the inclusion-exclusion principle when relating 𝒪D to 𝒪Ci when D=⋃Ci, which accounts for the appearance of Gα⁢β. □

5.6 Loop space formalism.

Definition 5.13.

The loop space (𝒦,Ω) consists of

with the symplectic form

Remark 5.14.

There is a polarization 𝒦=𝒦+⊕𝒦−, where

and

For example, 11−q∈𝒦−.

Definition 5.15.

The big J-function J:𝒦+→𝒦 is given by

Remark 5.16.

Note that applying Riemann-Roch, we have

Each term becomes of the form ch⁡(trg⁡11−q⁢L), which gives us something of the form ch⁡(11−q⁢ζ⁢L) for some root of unity ζ. Then we can expand

Definition 5.17.

The Givental cone is defined to be ℒ≔Im⁡J⊂𝒦, where we are considering the germ near 1−q in the Q-adic topology.

Example 5.18.

Let X=ℙn. Then we see that

where P=𝒪⁡(−1). As before, we can expand

Proposition 5.19 ([Giv15b]).

In fact, ℒ is actually a cone in 𝒦 and we have

We may also introduce a “huge” J-function

If we fix 𝐭⁡(q), then J⁡(τ) comes from a fundamental solution of a differential equation on the Frobenius manifold in τ. In addition, ℒ𝐭=Im⁡J⁡(τ) is actually a Lagrangian cone.

Example 5.20.

Consider f:ℂ⁡[Λ1±,Λ2±]→ℂ⁡[Λ1±,Λ2±] which is given by the Adams operation

Therefore, this sends Λ1↦Λ1n and Λ2↦Λ2n. However, it is not clear how to take derivatives in t because we would need n⁢v⁢tn−1=Ψn⁢(v), so the J-function does not satisfy a quantum differential equation in 𝐭.

5.7 Equivariant K-theory and localization.

For the fixed-point recursion and the projective-space calculation, see [Giv15a].

Let a torus T=(ℂ×)r act on a smooth projective variety X.

Definition 5.21.

The equivariant K-theory KT⁢(X) is the Grothendieck group of T-equivariant vector bundles on X.

Example 5.22.

For example, we have

where the Λi are the 1-dimensional representations with weight (0,…,1,…,0).

Remark 5.23.

For any X, KT⁢(X) is a KT⁢(pt)-algebra.

Example 5.24.

Consider the fully equivariant K-theory of ℙn. Then we have

When we specialize Λi=1, we recover the non-equivariant K-theory.

Definition 5.25.

We can define the equivariant Euler characteristic as the pushforward

Example 5.26.

For example, we have χT⁢(𝒪ℙ1⁢(1))=Λ0−1+Λ1−1. Similarly, we have

Notation 5.27.

Note that Frac⁡(T)=Frac⁡(Rep⁡(T)). Then we will set

We will need the following fact. Let F be the fixed point set of T on X and consider ι:F↪X. Then there is an isomorphism of Frac⁡(T)-algebras

Remark 5.28.

The inverse is given by

where Eu⁡(V)≔∑j=0rk⁡V(−1)j⁢Λj⁢V∗ is the K-theoretic Euler class.

Example 5.29.

On ℙn, we have n+1 isolated fixed points Fi, and then we obtain

These are idempotents of KTloc⁢(ℙn).

Using equivariant localization, we obtain

Remark 5.30.

In the Λ=1 limit, we can recover the non-equivariant Euler characteristic.

Example 5.31.

Let X=ℙ1. Then we compute

From now on, we will assume that X is a GKM manifold. Our goal now is to use localization on M¯g,n⁢(X,β) to study the big J-function of X. Any T-fixed stable map f:(Σ,p¯)→X has two types of irreducible components. There are those components which map entirely to fixed points and those components who map via ramified covers of 1-dimensional orbits. Unfortunately, we cannot compute the big J-function via virtual localization, but we can still study it.

Let f=f⁡(q)∈𝒦=KTloc⁢(X)⁢(q)⁢⟦Q⟧. Denote the fixed points by Fα for α=1,…,r with inclusions ια:Fα↪X. Then denote

Then f∈ℒ if and only if for all α, the following conditions hold:

  1. (1)

    Given λ∈TFα⁢X and m∈ℕ+, fα⁢(q) has at most simple poles at q=λ1/m, and

    Here, λ∈TFα⁢X means that λ is a 1-dimensional subrepresentation of TFα⁢X. Moreover, β⁡(α,α′) is the degree of the 1-dimensional orbit of T determined by the weight λ and ϕα,α′,m parameterizes m:1 covers of this 1-dimensional orbit.

  2. (2)

    fα⁢(q) has its other poles at 0,∞, and roots of unity, and is regular elsewhere. In addition, there are various minor conditions.

Example 5.32.

Let X=ℙn. Then

This then implies that

so the poles are at q=(ΛiΛj)1/ℓ. Note also that

5.8 Quantum K/GV correspondence.

The quintic correspondence was conjectured in [GS22] and proved in [CL22]. The following Calabi–Yau threefold formula is due to Chou–Lee [CL25].

Our goal here will be to compute JK⁢(0) on a Calabi-Yau threefold. The following result was conjectured by Jockers-Mayr and by Garoufalidis-Scheidegger.

Theorem 5.33 ([CL25]).

We have

where {Φα}α=1N is a basis of K0⁢(X) decomposed such that

such that the collection ch⁡(Φi⁢j) for fixed i form a basis of H2⁢i⁢(X). We then select a dual basis {Φα}α={Φi⁢j}i,j. Moreover, we set the multiple cover formulae

Finally, GV denotes Gopakumar-Vafa invariants, which are defined in genus zero by

By Möbius inversion, we see that

Corollary 5.34 ([CL25]).

We have

Applying Möbius inversion, we obtain8

In addition, we can recover the usual GW invariants by9

Naively, we can think of JK⁢(0)∈ℤ⁡⟦q⟧⁢⟦Q⟧, but to be a bit more precise, we can consider

Here, we expand 11−q⁢L=∑qi⁢Li. We also need to consider the fact that if M is a DM stack and L is a line bundle on M, then the function

converges to a rational function for |q|<ε. Note that there are poles only at roots of unity.

This fact can be proved using the orbifold Riemann-Roch theorem. Decompose the inertia as

where Mζ is the union of twisted sectors such that g acts on L with eigenvalue ζ. Using this, we can compute

and therefore all poles must be at roots of unity. Here, we use the expansion

and take the Chern character.

Example 5.35.

Let L=𝒪⁡(1) on ℙ1. Then we have

Example 5.36.

We will now consider weighted projective space. Then we have

5.9 Adelic characterization.

Recall that there is a relationship between GW and GV invariants using an explicit formula in all genus. On the other hand, we can recover GW invariants from KGW invariants via the q→1 limit. In the other direction, we need the Givental–Tonita adelic characterization [GT14]. The idea is that if we have a rational function f, then we can recover it from the Laurent expansions

at roots of unity ζ.

We will study the inertia stack I⁢M¯0,1⁢(X,β). One can keep track of the components via Figure 3 or via Figure 4.

Refer to caption
Figure 3. A way to keep track of the inertia components due to Givental-Tonita.
fX(a) ζr=1, ζ≠1ℤ2ℤ3fX(b) ζ=1
Figure 4. The components of I⁢M¯0,1⁢(X,β), due to You-Cheng.

We will now explain Figure 4. Let L1 be the cotangent line at the first marked point and let g be the automorphism of a point of I⁢M¯0,1⁢(X,β). Then ζ is defined by the condition that g acts on L1 with eigenvalue ζ. In both pictures, the bullet denotes the first marked point and its image in X, while a thick component is one which is a multiple cover of its image.

First, suppose that ζ=1. Then the black component is the maximal connected component containing the first marked point on which g acts as the identity. The red and blue components are precisely those on which g acts nontrivially, and we refer to them as arms.

Now suppose that ζ is an r-th root of unity for some r≥2. In this case, the nodes of the black component are required to be balanced with respect to the g-action, so we cut off the nonbalanced part, which is colored blue, and refer to it as the tail. On what remains, the black component is the maximal connected component containing the first marked point on which gr acts as the identity. Finally, after taking out the black component and the tail, we refer to the remaining components as legs.

The following definition is due to Givental. We write

where the Xζ term is simply the contribution of Mζ under the Riemann-Roch theorem. Now, we will define arms, legs, and tails. Here, we define the arms

the legs

and the tails

Proposition 5.37 ([GT14]).

The Laurent expansion of the J-function at q=1 is given by

where the fake invariants are computed by applying Riemann-Roch and only considering the contribution of the untwisted sector.

Proposition 5.38 ([GT14]).

At any other root of unity ζ≠1, we have

where the stem is some fake invariant of X×B⁢μr defined on the moduli space

It is defined by the formula

where N denotes the normal bundle of M¯0,n+2,βX⁢(ζ)⊂M¯0,n⁢r+2⁢(X,r⁢β).

Remark 5.39.

If we can calculate the fake and stem theories, then we can compute genus zero quantum K invariants.

5.10 Quantum K-theory of a point.

One natural question is to study higher genus quantum K-theory. We will begin with the simplest target, which is a point. One motivation for studying higher-genus would be to try to prove integrality of the Klemm-Pandharipande definition of GV invariants for Calabi-Yau fourfolds or the Pandharipande-Zinger definition of GV invariants for Calabi-Yau fivefolds (in this situation, we only need to consider genus 1). One reason we need to consider the quantum K-theory of a point is because in I⁢M¯1,1⁢(X,β), there are some components where an entire curve of genus 1 is contracted to a point.

Question 5.40.

What is the K-theoretic Witten-Kontsevich theorem?10

We will first recall the cohomological Witten’s conjecture. We will begin with the operator

and deform it to

where we should think of x=T1 and u⁡(x)=u⁡(x,0,0,…). Formally, write

using the property

for all integers n∈ℤ. In particular, we construct the (2-)KdV hierarchy by the equation

for all k≥0. We define a τ-function to be a function τ such that

Now perform the above process using u=2⁢x. Define the function

Note that because the genus is determined entirely by the number of insertions and the degree of the insertions, the notation

uniquely determines a descendant GW invariant of a point.

Theorem 5.41 (Kontsevich [Kon92]).

F is a τ-function of the KdV hierarchy constructed starting with L=∂x2+2⁢x under the change of variables ti=(2⁢i+1)!!⁢T2⁢i+1.

One application of this theorem is to consider virtual localization. For example, suppose we consider M¯g,0⁢(X,d⁡[C]) where C is a unique rational curve in a Calabi-Yau threefold with normal bundle N=𝒪⁡(−1)⊕𝒪⁡(−1). Then in this setting we can reduce to the local model and compute

via virtual localization. If we would like a similar result in K-theory, then we will need to consider terms of the form

so we will need to consider the permutation-equivariant quantum K-theory of a point.

We now proceed with a literature review. First, Lee [Lee97] computed the formula

Later, Lee–Qu [LQ14] gave a way to compute

in principle. Givental [Giv17] computed the formula

Finally, Milanov [Mil25], in a preprint first posted in 2024, computed

The following pluri-Hodge recursions and the new genus-one calculations based on them are due to You-Cheng Chou and were presented in this lecture.

Let s∈ℤ and define the pluri-Hodge sheaf ℋs≔R0⁢π∗⁢(ωπs). Considering the map M¯g,n+1→𝜋M¯g,n, we will denote the upstairs cotangent lines by Li and the downstairs ones by ℓi.

Lemma 5.42.

For all integers s and di, we have

Remark 5.43.

Note that if f⁡(q)∈ℤ⁡[q], then exp⁡∑kΨkk⁢(f⁡(q))∈ℤ⁡⟦q⟧. The reason is that this operation sends f=∑i=1nai⁢qi to ∏i=1n1(1−qi)ai.

Proof.

When all di are zero, we see that

by Serre duality. The next step is to show that

Here, we first consider the exact sequence

and twist by ωπs to obtain

and then compute the pushforward. □

Lemma 5.44.

Let fi be rational functions, for example 11−q⁢L. Then

where fn+1=Ln+1r and (D⁢f)⁢(x)≔f⁡(x)−f⁡(1)x−1.

To prove this, we need to consider the pushforward

Here, we use the fact that Li=(π∗⁢ℓi)⁢(Di) and Ln+1=ωπ⁢(D1+⋯+Dn). Applying the previous lemma, we see that the pushforward becomes

Now, Lee’s formula computes the correlators

via the expansion

Note here that there is no Hodge bundle in genus 0, so in fact the lemma can be used to obtain Lee’s formula.

Lemma 5.45.

We have

The lemmas collectively imply that we can reduce to the case of a single marking at the cost of making the insertions very complicated and introducing pluri-Hodge terms everywhere. In genus 1, life becomes easier because on M¯1,1, we have

In this case, recall the classical formula

Unfortunately, computing

is too hard, so we can try to compute

which is known, but it is unclear how to write a good formula for it. The reason we would consider this is the operation

Unfortunately, You-Cheng does not know how to write

in a nice form. In principle we can write formulae like

but this does not seem to help much.

5.11 Permutation-equivariant theory of a point.

We can compute χ⁡(M¯1,1,11−q⁢L) in another way. Consider the involution on an elliptic curve, which gives a map

Then, Givental [Giv17] computes

In particular, we obtain

and then because the map is a μ2-gerbe (or a 2:1-cover on objects), we obtain the desired formula.

Consider ℋg,1⊂M¯g,1 to be the hyperelliptic locus. Note that the marking lies on a Weierstrass point. Note that the fixed points of the hyperelliptic involution go to 2⁢g+1 points on ℙ1, and therefore one would want to use the calculation

to compute the quantity χ⁡(ℋg,1,11−q⁢L). However, because of the way that the curves upstairs degenerate, we will need to consider the Hassett moduli space [Has03] where the last 2⁢g+1 markings have weight 12.

6 Quasimap wall-crossing (Yang Zhou)

The main wall-crossing theorem and master-space construction in this lecture are from [Zho22]. Earlier genus-zero and higher-genus results appear in [CK14; CK20], and the foundations of stable quasimaps are in [CKM14].

Let X be the quintic threefold. Our goal is to count genus zero curves on X, which is a problem that was solved by Givental and Lian-Liu-Yau. Via the closed immersion X⊂ℙ4, we can write

which is smooth and has a torus action by (ℂ×)5. If a map is fixed by this action, the domain must have some components which map to the fixed points and some components which form an n:1 cover of the 1-dimensional orbits. In particular, the fixed locus is indexed by a set of graphs.

In this setting, we have the quantum Lefschetz principle, which states that

Here, the vector bundle E is defined by

where π is the universal curve and f is the universal stable map. We will give a different perspective on the result that if

then

up to the mirror map Q=q⁢exp⁡(I1I0).

6.1 Naive counting.

We will consider “maps” ℙ1→X⊂ℙ4 from a fixed ℙ1. For simplicity, we will assume X=(∑xi5=0) is the Fermat quintic. This is given by a collection [f0,…,f4] where fi∈ℂ⁡[u,v] such that deg⁡fi=d and ∑fi5=0. To be an actual morphism, we need the fi to have no common zeroes, but this moduli is not proper, so we relax this condition. Then we consider the moduli space

This lives inside the corresponding moduli space

of rational maps to projective space. There are two problems. First, the domain is parameterized, so we need to account for the automorphisms of ℙ1. Second, to obtain stable maps, we need to blow up the domain.

First, let ℂ× act on ℙ1 by scaling u. Consider the special fixed locus F⋆⊂Q⁢G⁢(X,d) defined by

Therefore, F⋆,d={[a0⁢ud,…,a4⁢ud]|[a0,…,a4]∈X}≅X. From this perspective, we can rewrite the I-function as

using standard arguments in deformation theory. This lives in Hℂ×∗⁢(X)⁢⟦q⟧⁢[z−1].

6.2 Wall-crossing without rigor.

Second, consider a nodal curve C with a marking plus a tail with an action of ℂ× and suppose we have a map f:C→X. If we take the limit

we see that the limit of the bubble lands in F⋆,d0, where the common zero is away from the node. Then we can replace this tail by a new marking which carries some terms of the I-function. For the quintic, this is simply I1⁢H−I0⁢ψ=[z⁢I−z]+,z=−ψ. At the end of this process, we are left with C=ℙ1 with various contributions of the I-function.

Note that if we consider ℂ× acting on ℙ4, then f:C→ℙ4 is fixed if for all t∈ℂ×, there exists φt such that

commutes. Therefore, we obtain an action of ℂ× on the domain.

We will define the moduli space Qg,nε⁢(X,d) of ε-stable quasimaps to X, which is analogous to the moduli of stable maps. We want to think of these moduli spaces as being virtually birational to each other, and each of these is a proper DM stack with perfect obstruction theory, and therefore we can define quasimap invariants

Here, ε∈ℚ>0 is a positive rational number. The space of possible ε has a wall and chamber structure, where the walls occur at 1n for positive integers n. When ε>1 we obtain Gromov-Witten theory (namely stable maps), and as we move to the left, we allow fewer rational tails but more basepoints. Eventually, in the ε→0+ limit, no rational tails are allowed. In the end, we obtain the following wall-crossing formula around the wall at 1d0:

Here, μ=[z⁢I−z]+,qd0,z=−ψ. In the special case, where g=0, n=1, and ε=0+, we obtain

Crossing all walls from the GW chamber, we obtain

which produces the mirror map via the dilaton and divisor equations.

6.3 Quasimaps.

Definition 6.1.

A prestable quasimap to the quintic threefold X consists of

where (C,x1,…,xn) is a prestable curve, ℒ∈Pic⁡C, and φ0,…,φ4∈H0⁢(C,ℒ), such that

is finite and away from the special points. In addition, we require that ∑φi5=0.

Importantly, we will refer to φ−1⁢(0) as the set of base points. Away from φ−1⁢(0), we have an actual morphism C∖φ−1⁢(0)→X. Geometrically, suppose we have a family over 𝔸1∖0 where in the limit as t→0, the zero loci φi−1⁢(0) will end up colliding. We can either preserve this quasimap in the central fiber, or we can blow up the central fiber some number of times to obtain an actual morphism at the cost of bubbling out a rational tail. Therefore, any moduli space of prestable quasimaps cannot be separated, so we must consider some kind of stability condition.

Definition 6.2.

A prestable quasimap (C,𝐱,ℒ,φ) is ε-stable if

  1. (1)

    For all x∈φ−1⁢(0), we have len⁡(x)≤1ε, where len⁡(x)=min⁡{ordx⁡(φi)};

  2. (2)

    The ℚ-line bundle ℒε⊗ωClog has positive degree on every irreducible component of C.

Note that for all irreducible components C′⊆C, we have deg⁡ℒ|𝒞′≥0. Therefore, the things which change are curves of genus 0 with ≤2 special points or curves of genus 1 with no marked points. Note that in the (0,0) and (1,0) cases, we have C=C′, and in the (0,2) case, stability is equivalent to deg⁡ℒ>0, so in the (0,1) case, the second condition is equivalent to

Example 6.3.

In the ε>1 chamber, note that the first condition implies that there are no basepoints, and the second condition implies that all unstable components have positive degree, and therefore we have recovered stable maps.

If we move into the chamber where 12<ε<1, then we must contract rational tails of degree 1. For example if we have a chain of two degree 1 rational curves ending in a tail, then we contract the outermost tail and are left with a single degree 2 rational tail with a basepoint of length 1.

If we move into the 13<ε<12 chamber, then we must contract all rational tails of degree 2 and replace them with basepoints of length 2.

Remark 6.4.

In the case of projective space, there is an actual morphism

For a general target, this is not possible, but this idea gives intuition for the phenomenon of quasimap wall-crossing.

We will now discuss how to generalize this to more general target spaces. Let

where V is affine, G is reductive, and θ is a character of G. We need to require that

  1. (1)

    Vss⁢(θ)=Vs⁢(θ)≠∅;

  2. (2)

    V has at worst lci singularities and Vs⁢(θ) is smooth.

Note that we require the stable locus to have finite stabilizers. For convenience, we will also require

  1. (1)

    V⫽G is projective;

  2. (2)

    G acts freely on the stable locus.

Example 6.5.

In the case of the quintic, V is the affine cone over the quintic. The unstable locus consists of the origin, which is the unique singular point.

Example 6.6.

Another example is that we can consider ℂN⫽(ℂ×)r with generic stability condition and obtain a toric variety.

Example 6.7.

We can consider the Grassmannian

where the unstable locus consists of matrices with rank strictly less than 2.

Remark 6.8.

The condition on the stable locus is always satisfied when G is abelian for generic stability conditions, but for non-abelian G this is not always possible.

Remark 6.9.

For a general smooth projective variety, the origin is not necessarily lci, so it is unclear how to define quasimaps in a way where we can still define a virtual cycle.

Definition 6.10.

A prestable quasimap to V⫽G is a map u:C→[V/G] such that the base locus

is finite and away from the special points. By definition, u consists of a principal G-bundle P→C together with an equivariant map P→V, which is equivalent to a section of the associated bundle P×GV.

Example 6.11.

For the quintic, this exactly recovers a line bundle ℒ∈Pic⁡C (which is equivalent to a ℂ×-torsor) combined with a section of Tot⁡(ℒ⊕5) which satisfies the defining equation.

In order to define degrees, we can consider the line bundle ℒθ on [V/G]. Then we can define the degree of C′⊂C under u to be

Example 6.12.

In the G⁡(2,4) example, note that a principal GL2-bundle is equivalent to rank 2 vector bundle ℰ on C. Therefore, P×GV is equivalent to the data of ℰ⊕4, and therefore quasimaps to the Grassmannian are equivalent to stable quotients. In this case, we are looking for generically surjective

and because θ=det, we see that u∗⁢ℒθ=detℰ.

Example 6.13.

Consider V=ℂN+1 and note that ℙN=V⫽ℂ×. If we consider θ1:z↦z, then we have

However, if we consider θ:z↦z2, then we have

Here, we see that the scheme structure is sensitive to the choice of character, so we cannot simply naively define the length of a basepoint. Note that

In particular, we must define

Example 6.14.

Let X=pt=ℂ⫽θℂ×. Then a prestable quasimap is simply

where φ−1⁢(0) is finite and away from the special points. In addition, if deg⁡ℒ=1, we have precisely recovered the data of a new marked point in Csm.

Example 6.15.

Now consider X=pt=ℂn⫽(ℂ×)n. Also suppose that the degree is (1,…,1). Now we will set

Therefore, note that

Because these all live in degree 1, we see that for a smooth point x∈Csm, we have

If we set {pi}=φi−1⁢(0), then the condition that ε⋅len⁡(x)≤1 is equivalent to requiring that

The degree condition now states that

Equivalently, this states that

so we recover a weighted pointed curve with weights (ε⁢a1,…,ε⁢an).

6.4 Master space.

The proof uses virtual localization [GP99]; Hassett's weighted moduli spaces [Has03] provide one of the models for the construction.

Our method will be to perform virtual localization on a master space which interpolates the two moduli. Consider the ℂ×-action on ℙn given by

The fixed loci consists of a ℙn−1 where x0=0 and a point [1,0,…,0]. Then we can compute

by transfering the computation to the other fixed locus. In particular, we lift to H∈Hℂ×∗⁢(ℙn)=ℚ⁡[H,z]/(H−z)⁢Hn and compute

Taking residues, we see that

which recovers the desired integral. The key fact we used here is that projective space is proper.

Example 6.16 (Second toy example).

We will consider a second toy example. Recall Hassett’s moduli of weighted pointed curves

where C is nodal and the xi∈Csm can collide. The collisions are regulated by the fact that each marked point is given a weight δi∈ℚ∩(0,1] and are required to satisfy the following conditions:

  1. (1)

    For all x∈Csm, then

  2. (2)

    We have the stability condition ωC⁢(δ1⁢x1+⋯+δn⁢xn)>0. For example, if we have a rational tail, then ∑xi∈C′δi>1.

Example 6.17 (Another toy example).

Our goal will be to once again compute

Our first goal will be to change the stability condition. If we set

we see that the stability condition does not change. However, if we set

the moduli space does not change, but the universal curve in fact becomes ℙ1×ℙ1 with the section x4 being the diagonal. This tells us that

and we will use this to compute in the δ+ chamber. We can try to form a master space by both allowing rational tails with xi,x4 and allowing x1=x4, but if we do this, we end up with a semistable point with a rational tail where x1=x4. We want to end up with this by taking a quotient from a master space, so we will add the data of a tangent vector to the moduli. Therefore, define the δ± master space to consist of objects of the form

where C is nodal, xi are smooth points, x1,x2,x3 are disjoint,

for all rational tails C′, if v=0, then x4 is disjoint from the other xi, and when v=∞, there are no rational tails where x4≠xi for i=1,2,3.

We now give this master space an action of ℂ× by scaling v. There are various fixed loci:

  1. (1)

    When v=0, we obtain M¯0,4δ+;

  2. (2)

    When v=∞, we obtain M¯0,4δ−;

  3. (3)

    There are fixed loci where v≠0,∞ and we have a rational tail with xi=x4.

If we consider a general point (with all four marked points disjoint), it will travel directly to itself along the ℂ×-action. However, if we start with a rational tail containing x1 and x4, then we first travel to a rational tail with general v but x1=x4, and if we smooth out the node, we obtain a single curve with x1=x4, but we allow v→∞, which is another fixed point. Now we can integrate

Taking residues at z=0, we obtain

where in the final term the ±z in the denominator are node smoothing parameters, and therefore we recover the desired integral.

With all of our toy examples, we are finally ready to discuss the quasimap master space. Our first step at the 1d0 wall is to allow both degree d0 rational tails and length d0 base points. However, we can consider curves with a rational tail and a quasimap of the form

which have infinitely many automorphisms. However, we cannot attach tangent vectors to the base points (because deformations will remove them), and there are also situations where there exists more than one degree d0 with a base point of length d0 (and then the automorphism group contains ℂ×).

Remark 6.18.

There is a key observation. Suppose that we have two rational tails of degree d0 and call the two nodes p1 and p2. Now define the space of smoothing parameters

where Ei is the rational tail and Cg is the rest of the curve. We will consider the moduli

of weighted semistable curves labelled by degrees. Note that this morphism is highly non-separated. We will consider the open substack 𝔐g,n,dwt,ε0 where the curves have no tail of degree strictly less than d0. Then we can consider the Cartier divisor 𝒵 where there exists at least one tail of degree d0. In particular, we see that

for the two branches 𝒵1 and 𝒵2 at the point of interest. We also see that Θ1⊗Θ2=𝒪⁡(𝒵)|ξ, so we can add the data of

and the data of

The actual construction proceeds as follows. Consider 𝒵⊂𝔐=𝔐g,dwt,ε0. Define 𝔐~→⋯→𝔐 to be an iterated blowup along strata of 𝒵. Then an object of the master space (a quasimap with calibrated tails) over a scheme S is given by the data of a quasimap

inducing a morphism p:S→𝔐g,dwt,ε0, a choice of a lift

a line bundle 𝒩∈Pic⁡S, and v=(v1,v2) where v1∈H0⁢(w∗⁢𝒪⁢(−𝒵)) and v2∈H0⁢(𝒩), where (v1,v2) is nonvanishing. We also disallow basepoints of length strictly larger than d0 and require stability such that

  1. (1)

    When v=0, we obtain ε+-stability;

  2. (2)

    When v=∞, we obtain ε−-stability;

  3. (3)

    Any constant tail (degree d0 with base point of length d0) is an entangled tail. Here, we consider the last blowup such that w maps to an exceptional locus ⋂i∈I𝒵~i, and then the Ei are entangled tails for i∈I;

  4. (4)

    If there exists a degree d0 tail, then all length d0 base points are on degree d0 tails.

7 Log GLSM (Qile Chen)

Let X be a smooth projective variety (or even a DM stack). Let E be a vector bundle on X with a section s∈H0⁢(E) which defines a smooth zero section Z of the expected codimension. We will refer to X and E as ambient and s and Z as the complete intersection.

Question 7.1.

What is the relationship between the GW theory of Z and the GW theory of (X,E)?

When g=0, Givental and Lian-Liu-Yau used this relationship to compute the GW invariants of complete intersections, and when g=1, this was done by Zinger. In general, there are two approaches: (N)MSP and log GLSM. In this series of lectures, we will discuss log GLSM.11

7.1 R-maps.

We will consider the R-torus ℂω×≔ℂ×. This has a weight 1 representation, which we will call ℂω. It descends to a line bundle ℒω≔[ℂω/ℂω×] on B⁢ℂω×. The targets of R-maps will be stacks

which are DM-type, separated, and (log) smooth. Now, an R-map consists of a (2-)commutative diagram

Here, the curve could be prestable, twisted, log, or even punctured.

Example 7.2.

Our first example is that of stable maps. Let 𝔓=B⁢ℂω×. Because in the diagram

the vertical arrow is the second projection, we see that this reduces to a map to X.

Example 7.3.

Consider maps

If we assume that C is unmarked, we see that the data of this map is simply a section of ωC, and therefore we recover holomorphic differentials.

Example 7.4.

Consider a morphism ℂR×→ℂω× given by z↦zrin. Then we see that

Now a diagram

with C unmarked is equivalent to the data of a line bundle ℒR|C and an isomorphism ℒR⊗rin≃ωC. Alternatively, we can consider R-maps of the form

which parameterize curves with an R-spin line bundle together with a section.

Example 7.5.

Let X be a smooth projective variety. Then consider R-maps of the form

This is equivalent to a map C→X together with a section ρ∈H0⁢(E∨|C⊗ωC).

We will consider targets in the GW case. Consider the data of an ambient (X,E). Then we will set 𝔓∘≔E∨⊠ℒω. The morphism 𝔓∘→B⁢ℂω× is very nice, but it is not proper. To fix this, we will consider the log stack

Because we have a smooth divisor in a smooth stack, the map 𝔓→B⁢ℂω× is proper and log smooth. Note that we have the zero section

We will now consider targets in the r-spin case. Consider the log stack

This morphism is proper and log smooth.

7.2 Log and punctured R-maps.

We will consider diagrams

where C→S is a log or a punctured (possibly twisted) curve and S is an fs log scheme. The stability condition in the GW case is that f is representable and

for all 0<δ≪1 and k≫1, where H is an ample line bundle on the coarse moduli space of X. For punctured R-maps, we further require pre-stability.

Example 7.6.

Consider the case of holomorphic differentials. Then consider f∈H0⁢(ωC) such that all zeroes are concentrated in a single Weierstrass point, where C has genus 2. If we consider limλ→∞λ⋅f, then the limit consists of a genus 2 component mapping entirely to ∞ and a rational component which remembers the zero.

We will now discuss the tropicalization of log or punctured R-maps. Recall that a graph

consists of a set 𝕍 of vertices, a set ℍ of half edges, an involution ı on the half edges, and a root map from half edges to vertices. Then we will let the set 𝕃 of legs be

The edges are pairs of distinct half-edges which are sent to each other by the involution:

Now a decorated graph is a graph combined with a genus function

an assignment

and an ordering

Example 7.7.
12120

For a log or punctured R-map

we will tropicalize it by taking the associated generalized cone complex of the entire diagram. Because B⁢ℂω× has no log structure, what we see is

Now suppose that S=Spec⁡(Q→ℂ) is a log point with Q an fs sharp monoid. In particular, this means that Σ⁡(S)=Qℝ∨. Because everything is (piecewise) linear, it suffices to consider an interior point t∈Σ⁡(S) and consider Σ⁢(C)t.

From each half edge, we treat it as pointing outward from the attached vertex, and in particular, this means that a linear map to ℝ≥0 corresponds to a slope. We will call these contact orders. In particular, we have c⁡(h)=−c⁡(h^) if h≠h^ are related by ı. For all v∈𝕍, we also define the degeneracy

In particular, ev∈Q is an integral point.

The next thing we need to introduce are punctured legs. They will be denoted by

We will define prestability at h∈𝕃∘ such that 0∈Σ⁢(f)t⁢(h)¯, which is a technically necessary assumption. Under the assumption that all punctured legs are prestable, we can define the edge length function

Note that ℓ⁡(E)∈Q is also an integral point for edges E. For punctured legs, this is true up to subdivision.

7.3 Log GLSM invariants.

Definition 7.8.

We will define the decorated tropical type of a log map f as

Definition 7.9.

A log or punctured map f is of compact type if

  1. (1)

    Im⁡Σ⁢(f)t is bounded for all t;

  2. (2)

    f⁡(ph)∈0𝔓 for all half-edges h with contact order 0.

We are now able to define the category ℛ⁡(𝔓,τ) of stable punctured R-maps fibered over the category 𝖫𝗈𝗀𝖲𝖼𝗁 of fs log schemes and marked by τ.

Theorem 7.10 ([CJR21; CJR22]).

The stack ℛ⁡(𝔓,𝛕) is represented by a proper DM stack with basic/minimal log structure.

Remark 7.11.

If 𝕍={v} is a singleton, then we will write

Consider the diagram

where we set ℛ≔ℛ⁡(𝔓,𝝉). Then the perfect obstruction theory is defined by

Here, both tangent bundles refer to the log tangent complex. From the perfect obstruction theory, this defines the canonical virtual cycle

The main drawback of this virtual cycle is that it does not reproduce Gromov-Witten theory. However, we can consider how it differs from GW theory, but for this we need a birational modification. Let Q be a sharp monoid. It has a partial ordering by a<b if there exists a nonzero c∈Q such that a+c=b.

Definition 7.12.

We say that either f or Σ⁡(f) has uniform maximal degeneracy if there exists emax∈{ev|v∈𝕍} such that emax≥ev for all v∈𝕍.

Remark 7.13.

If f is compact type and has uniform maximal degeneracy, then Σ⁡(f) is uniformly bounded by emax.

Definition 7.14.

Define 𝒰⁡(𝔓,𝝉) to be the stack of stable punctured or log R-maps marked by 𝝉 with uniform maximal degeneracy.

Remark 7.15.

If 𝕍={v} is a singleton, then we will write

If one of the contact orders is negative, then we will write

Note that there are two types of vertices: if a vertex is mapped to 0, then either there is at least one half-edge which points upwards or all contact orders are 0, in which case all are contracted. These are called log vertices. The other type of vertex is one that does not go to 0. Either all half-edges point downwards, or there are some half-edges which point upwards. These are called ∞-vertices. Moreover, the vertices which are compact types are the ones without upwards-pointing half-edges.

If we forget the condition of having uniform maximal degeneracy, then there is a map

which is a subcategory over 𝖫𝗈𝗀𝖲𝖼𝗁. However, as a morphism of algebraic stacks, it is log étale, proper, and surjective. In addition, 𝒰 has a canonical perfect obstruction theory which is pulled back from ℛ. In the case when the set of vertices is a singleton, we have

In the case of a log vertex, we in fact have inclusions from the moduli of stable maps with p-fields as

If we denote the moduli of p-fields by ℛ∘, then the boundary

is a Cartier divisor. By the condition of having uniform maximal degeneracy, we have a section

This is equivalent to the data of a morphism

which pulls 0 back to Δ.

We will define a superpotential to be a function

Its critical locus is denoted by

We say that W has proper critical locus if Crit⁡(W)→B⁢ℂω× is proper.

For example, suppose we are interested in GW theory. Let X be a smooth projective variety, V be a vector bundle on X, and s∈H0⁢(V). Set Z=(s=0). Then we obtain

Lemma 7.16.

W has proper critical locus if and only if Z is smooth of expected codimension.

Example 7.17.

Let X be a point, V=ℂ, and s≠0. Then dW=s≠0, so the critical locus is empty.

Using the Kiem–Li method of cosection localization [KL13], we can obtain a virtual cycle

which is supported on the critical locus. In GW theory, this equals the GW virtual cycle,

where the sign can be explicitly calculated.

Theorem 7.18 ([CJR21; CJW21]).

Let 𝛕 be of compact type. In the compact type setting,

  1. (1)

    𝒰⁡(𝔓,𝝉) has a reduced perfect obstruction theory, which yields a reduced virtual cycle [𝒰⁡(𝔓,𝝉)]vir;

  2. (2)

    Δ has a reduced perfect obstruction theory, which yields a reduced virtual cycle [Δ]vir;

  3. (3)

    We have [𝒰g,c=n⁢(𝔓,β)]red=±[M¯g,n⁢(Crit⁡(W),β)]vir in the case of a single log vertex. Moreover, there is the formula

    where r~ is the pole order of W along ∞.

7.4 Tropical decomposition and finite generation.

The tropical decomposition and its applications below are from the working draft [CJR25]. The effective invariants and their axioms are developed in [CJR22].

From now on, we will consider a simpler setup where X is a smooth projective variety and V is a line bundle on X. In this case, we will assume that Z=(s=0) is smooth of codimension 1. Moreover, we will assume that Z is a Calabi-Yau threefold. We will consider the invariants

and the generating series

Conjecture 7.19 (Yamaguchi-Yau, Alim-Länge [YY04; AL07]).

Suppose g=0 mirror symmetry holds for Z. Then there exists a finitely-generated ℚ-algebra R⊂Λ such that for all g≥2, Fg∈(I0⁢(q))2⁢g−2⁢R.

Our main goal is the following.

Theorem 7.20 ([CJR25]).

Let Z be a Calabi-Yau threefold hypersurface in a Fano fourfold X with no odd cohomology. Let {ϕj} be a basis of H∗⁢(X) and {ϕj} be the dual basis. Then for all g≥0, we have

Example 7.21.

Let Y be a toric variety with its toric log structure. Then Σ⁡(Y) is the toric fan treated as a cone complex, and we can consider its boundary ΔY=∑ρ∈Σ⁢(Y)1Δρ.

If we consider [Δ]red, we should obtain the similar formula

The question now is what the ρ represent. In fact, they parameterize rigid tropical curves. In the tropicalization

we see that fρ is rigid if and only if the source tropical curve is a bipartite graph with one set of vertices going to emax and the other set of vertices going to 0.

Example 7.22.

Let X be a point and Z=∅. Also consider g=2,n=0. Then the possible graphs are given in Figure 5.

g=2g=1g=0g=0g=0g=0g=0g=1g=0g=0g=1g=2g=2ABCDEF11111223
Figure 5. Graphs of tropical types for X=pt and Z=∅ when g=2.
Theorem 7.23 (Tropical quantum Lefschetz [CJR25]).

Let X and Z be as before. For any β∈H2⁢(X), any α1,…,αn∈H∗⁢(X), and any k1,…,kn∈ℕ, we have

Remark 7.24.

There is also a virtual localization formula, announced in [CJR], which can be applied to either the reduced theory (with only the R-torus action) or the canonical theory (which can be applied to the full automorphisms of the ambient space).

Now consider an ∞-vertex. Set τv=(g,c,β) and set 𝒰v=𝒰g,c⁢(∞,β). Then the first computation is that

where the tropical dimension is 3⁢g−4+n. Note that all contact orders are negative, so the second term is at most zero. In particular, we have proved that

Corollary 7.25 ([CJR22]).

If vir⁢dimM¯g,n⁢(Z,β)<0, then [𝒰v]red=0.

We also have a balancing condition, which states that

This can be rewritten as

Proposition 7.26 ([CJR22]).

We see that ∞-vertices do not exist in the following cases:

  1. (1)

    We have g=0 and V is nef;

  2. (2)

    We have g=1 and V is ample, unless β=0 and all c⁡(h)=−1.

Definition 7.27.

A marking is a unit if its contact order is −1.

Notation 7.28.

We will denote the unit sector by 𝟏 and denote τv+𝟏=(g,c+𝟏,β), 𝒰v+𝟏=𝒰g,c+𝟏⁢(∞,β), and ℛv+𝟏=ℛg,c+𝟏⁢(∞,β).

Theorem 7.29 ([CJR22, Theorems 1.16–1.17]).
  1. (1)

    There is a map F𝟏:𝒰v+𝟏→𝑠𝒞∘→𝜋𝒰v, where s is birational;

  2. (2)

    We have the equality s∗⁢[𝒰v+𝟏]red=π∗⁢[𝒰v]red of reduced virtual cycles;

  3. (3)

    If D is a divisor on X, then F𝟏⁣∗⁢(ev𝟏∗⁡D∩[𝒰v+𝟏]red)=(∫βD)⁢[𝒰v]red;

  4. (4)

    Consider the diagram

    Then F∗⁢[𝒰v+𝟏]red=p∗⁢F∗⁢[𝒰v]red.

Example 7.30.

Consider the setting where g=1 and Zd⊂X=ℙN is a hypersurface. We only need to consider β=0 and c⁡(h)=−1 for all h∈ℍ. Then we see that

Proposition 7.31 ([CJR22]).

We in fact have

and the equality

Example 7.32.

Now consider the quintic threefold Z5⊂ℙ4. Then we see that red⁡dim𝒰v=∑h∈ℍ(c⁡(h)+2). The unit axiom reduces us to the case when c⁡(h)≤−2. We must now determine the number of new invariants we need to consider if all c⁡(h)=−2, which is equivalent to finding the number of possible β. The balancing condition states that

which implies that 0≥∑(c⁡(h)+1)=5⁢β−(2⁢g−2) and therefore that β≤2⁢g−25.12

Conjecture 7.33 (Janda [CJR25, Conjecture 1.6]).
  1. (1)

    We have K⁡(Q)=log⁡I0⁢(q)⋅c1⁢(KX∨)−I1I0

  2. (2)

    The canonical theory satisfies Ω∈I0⁢(q)2⁢g−2+2⁢n1+3⁢n2⋅R.

Corollary 7.34 ([CJR25, Theorem 1.7]).

The above conjecture for (X,V) implies Yamaguchi-Yau finite generation for Z.

8 Structure of higher-genus GW invariants of the quintic (Shuai Guo)

8.1 The problem.

We will consider the quintic threefold X=X5⊂ℙ4 and ask how many curves of genus g and degree d it contains. In genus zero, the answers begin

Because X is Calabi-Yau, the moduli space ℳ¯g,n⁢(X,β) has virtual dimension zero, and the string equation gives ⟨𝟏,γ2,…,γn⟩g,n,β=0 unless (g,n,β)=(0,3,0). We therefore define the primary invariants by

and we package these numbers in the generating series

The invariants Ng,β are rational numbers, while the integers we actually want are the BPS invariants ng,β. These are defined from the positive-degree part Fg+ by the multicover transformation

Note that only the ng,β are expected to be integral, and that the numbers above are the n0,d.

Table 2. Leading BPS invariants ng,d of the quintic threefold
g d=1 d=2 d=3 d=4 d=5
0 2875 609250 317206375 242467530000 229305888887625
1 0 0 609250 3721431625 12129909700200
2 0 0 0 534750 75478987900
3 0 0 0 8625 −15663750
4 0 0 0 0 49250
5 0 0 0 0 1100
6 0 0 0 0 10
7 0 0 0 0 0

We see that every column of Table 2 eventually vanishes. This is the Castelnuovo bound, which for the quintic states that ng,d=0 once g>d2+5⁢d+1010 and was proved by Liu–Ruan [LR22]. Integrality of the entries is the Gopakumar-Vafa conjecture, which was proved by Ionel–Parker [IP18].

In genus zero, Candelas-de la Ossa-Green-Parkes computed F0 using period integrals on the mirror, which we discussed in Section 3.2. In higher genus, Bershadsky-Cecotti-Ooguri-Vafa gave a recursion which produces Fg from the Fh with h<g up to finitely many unknown constants. Our goal is to explain how log GLSM determines this ambiguity.

8.2 Genus zero and the polynomial structure of the B-model.

Recall from Section 3.2 that for the quintic we consider the I-function

whose components are annihilated by the Picard-Fuchs operator

If we write Jk=Ik/I0 and I1I0=log⁡q+τ⁡(q), so that Q=eI1/I0=q⁢eτ⁡(q)=q+O⁡(q2), then the genus zero mirror theorem states that

For our purposes, it is useful to place the quintic formula in the general mirror-symmetry picture. Let 𝒳={(X,ω)} be a family of Calabi-Yau threefolds with Kähler moduli ℳ∋Q=e2⁢π⁢i⁢ω and let 𝒳∨={(X∨,J)} be a mirror family with complex moduli ℳ∨∋q. The solutions of the Picard-Fuchs equation are periods on X∨ and assemble into

which lies in the span of z⁢φ0,φα,φα⁢z−1,φ0⁢z−2. Here, I1=∑αI1α⁢φα, I2=∑αI2,α⁢φα, and I3=I3,0⁢φ0, and the mirror map is again Q=eI1/I0=q⁢eτ⁡(q).

We will use three special points on the complex moduli line of the quintic: the large complex structure point q=0, the conifold point q=5−5 (where a period vanishes), and the orbifold (or Landau-Ginzburg) point q=∞. The mirror map is defined at q=0 and the expansion there produces Gromov-Witten invariants, while in higher genus we will impose boundary conditions at the other two points.

We now restate the mirror theorem in terms of three-point functions. Pick local coordinates qi on ℳ∨, let {ϕi} be the corresponding basis on the A-side, and define

Theorem 8.1 (Mirror theorem [Giv98; LLY97]).

We have Ci⁢j⁢kA⁢(Q⁡(q))=Ci⁢j⁢kB⁢(q). Moreover Ci⁢j⁢kB∈Δ−1⁢ℚ⁢[q], so it is a globally defined meromorphic function on ℳ∨ with poles only along the discriminant.

In particular, after multiplying by the correct power of I0 and applying the mirror map, a genus-zero generating function becomes a rational function of q with poles only along the discriminant. Therefore, it is determined by finitely many coefficients.

Example 8.2.

For the quintic in the hyperplane basis,

Example 8.3.

For X=X3,3⊂ℙ2×ℙ2, the Yukawa couplings are

where

In higher genus, we expect normalized potentials fg,iA satisfying

where the right hand side is computed by a sum over graphs.

8.3 The BCOV Feynman rule.

The B-model Feynman rule and holomorphic anomaly equation originate in [BCOV94]. For the quintic, Chang–Guo–Li prove polynomiality in [CGL21] and the Feynman rule in [CGL26].

We will now discuss the BCOV recursion as a Feynman rule. It comes from a finite-dimensional model for the quantization of a symplectic transformation as in Section 3.6. Let ℋ=ℋ−⊕ℋ+⊂ℋ∞ be symplectic with d+1=dimℋ+=dimℋ− and set

Given a total potential F⁡(ℏ,𝐱), we formally define the transformed potential by the Gaussian integral

Expanding by stationary phase gives a sum over graphs. Here, D−1⁢C supplies the propagator on each edge and F supplies the vertices.

For a Calabi-Yau threefold the propagators are written Sφ⁢φ, Sφ, and S according to the number of indices, and the vertices are the normalized A-model correlators

Note that these have nothing to do with the operator Sτ⁢(z) solving the quantum differential equation, which will appear later.

Example 8.4.

For the quintic in genus two with no markings, the Feynman rule reads

where χ=χ⁡(X5)=−200. This is a cubic polynomial in the propagators.

We can interpret this as a sum over stable graphs, as in Section 3.4. A vertex v contributes the lower-genus correlator Pgv,val⁡(v); an internal edge contracts two of the insertions against a propagator; the legs record the remaining insertions. Each graph is weighted by 1|Aut⁡Γ|. As usual

and fgB is the sum of the contributions of all stable graphs of genus g.

8.4 The holomorphic anomaly equation.

We will now rewrite the recursion as a differential equation. The B-model amplitudes ℱg⁢(z,z¯) are not holomorphic, and for g≥2 their failure to be holomorphic is given by

Here, Ci⁢j⁢k is the Yukawa coupling, C¯i¯j⁢k is obtained from C¯i¯⁢j¯⁢k¯ by raising two indices with the Weil-Petersson metric, and Dj is the covariant derivative on the appropriate power of the Hodge line bundle. The two terms on the right are the two boundary strata of ℳ¯g, namely a non-separating and a separating node. The 12 forgets the ordering of the two branches.

If we choose local ∂¯-primitives

so that, modulo a holomorphic term, the anomaly equation becomes the polynomial PDE

These propagators are only defined up to holomorphic tensors, and this freedom is the source of the holomorphic ambiguity.

Remark 8.5.

In the one-parameter normalized variables we will use for the quintic, this reads

To see why the graph sum satisfies this equation, note that in an unmarked potential the non-holomorphic generators occur only in the edge factors. A derivative in a propagator direction therefore picks out a single edge and cuts it. The two half-edges become markings on the resulting lower-genus vertices, and the covariant derivatives correspond to those markings by the divisor equation. We obtain Pg−1,2 by cutting a loop and Pg1,1⁢Pg2,1 by cutting a separating edge.

Remark 8.6.

For clarity, the implication only goes one way. The BCOV graph rule implies the anomaly equation, but the anomaly equation by itself does not determine the graph rule. To recover the same solution, we also need the propagator convention, the lower-genus vertex data, and the boundary conditions.

If all lower-genus vertices are known, then the right hand side is known, so we can integrate with respect to the non-holomorphic generators to obtain

where finite generation gives the holomorphic ambiguity

We see that the anomaly equation leaves 3⁢g−2 unknown coefficients in each genus.

Remark 8.7.

For the quintic, the degree-zero data together with orbifold regularity remove ⌈3⁢g−35⌉+1 of these freedoms and the conifold gap supplies 2⁢g−2 further conditions, so that

parameters remain and are genuinely new effective invariants. This is exactly the number of ∞-vertex unknowns predicted by Equation 1, namely the number of β with β≤2⁢g−25.

We therefore obtain the following conjectural algorithm proposed by physicists: the BCOV Feynman rule, orbifold regularity at the Landau-Ginzburg point (a form of the LG/CY correspondence, which for the quintic is work in progress of Guo-Ruan-Janda), and the conifold gap condition. To these we may add the Gopakumar-Vafa integrality and the Castelnuovo bound.

8.5 Four theories.

Recall from Section 7 that the log GLSM carries two virtual cycles. The reduced cycle recovers quintic Gromov-Witten theory, while the canonical equivariant cycle admits a root of unity specialization which is semisimple and therefore computable by Givental-Teleman. The bipartite graph sum

expresses the former in terms of the latter, at the cost of introducing the effective invariants cg,βeff at the ∞-vertices. Replacing these by formal parameters 𝐂={cg,β} gives the extended 𝐂-family, whose vertex data is supplied by the meromorphic and rubber log GLSM theories.

We will consider the following four theories.

  • •

    The t-twisted geometric theory sets λi=0 and keeps the auxiliary weight t. It has both stable-map and stable-quotient versions, and the quotient version gives the genus two localization formula below.

  • •

    The formal quintic, or formal λ-twisted theory, sets t=0 and λi=ζ5i⁢λ. It is a semisimple CohFT, so it has a universal and computable R-matrix.

  • •

    The extended 𝐂-family interpolates between these, and is where bipartite graphs with both formal and meromorphic vertices live.

  • •

    Quintic Gromov-Witten theory is obtained by setting 𝐂=𝐂eff and then letting λ→0, and is the theory we want to compute.

Warning 8.8.

In higher genus the t0 coefficient of the t-twisted stable quotient theory is not the λ0 coefficient of the formal λ-twisted theory.

8.6 The formal quintic.

The formal quintic and its holomorphic anomaly equations are studied in [LP19]; the comparison with the extended theory is developed in [GJR18].

Let E=𝒪ℙ4⁢(5) and let π:𝒞→ℳ¯g,n⁢(ℙ4,d) be the universal curve. The t-twisted cycle is

where [1] has equivariant first Chern class t. Starting instead from the fully (ℂ×)5-equivariant cycle, the formal quintic is the root of unity specialization

where ζ5=e2⁢π⁢i/5. The relation H5=λ5 makes the quantum product generically semisimple while retaining λ as a parameter.

The state space, unit, and pairing are given by

and the CohFT classes are

where ρ forgets the map and stabilizes the domain. If we shift by τ⁡(q)⁢H with τ⁡(q)=I1λ⁢(q)/I0λ⁢(q), then the shifted theory Ωλ,τ is semisimple. After extending scalars to ℚ⁡(ζ5,λ), the idempotent basis satisfies

Then Theorem 3.24 gives

with the decorations of Section 3.4. A leg carrying γ gets R−1⁢(ψ¯)⁢γ, an edge gets

and a vertex gets the topological field theory ω, with the missing unit-preserving part inserted at forgotten markings via T⁡(z)=z⁢(𝟏−R−1⁢(z)⁢𝟏).

Remark 8.9.

We see that the edge factor has no pole. Setting z2=−z1 and using the symplectic condition in the form R−1⁢(z)=R∗⁢(−z), the numerator vanishes. Therefore, it is divisible by z1+z2, and VR is a power series in the two cotangent line variables. The coefficient of z1p⁢z2q is the class placed at a node with ψ¯1p⁢ψ¯2q, and contracting it against the two adjacent topological field theory tensors glues the branches.

8.7 How to compute the R-matrix.

To compute the R-matrix, we begin with the formal equivariant I-function

The coefficients through H are independent of λ, so I1λ=I1−I0⁢log⁡q and the mirror shift τ⁡(q)=I1λ⁢(q)/I0⁢(q) is free of logarithms. Set

and introduce the bases

where the diagonal Birkhoff factors are defined recursively by

so that in particular I1,1=1+D⁢τ. The idempotents and canonical coordinates are given by

We will use the quantum differential equation to determine the q-dependence of the R-matrix. The fundamental solution satisfies

and in the normalized canonical direction α the stationary phase asymptotics give

so that

Here, Cα⁢(z) is the q=0 quantum Riemann-Roch (equivalently, Gamma function) factor, and together with the symplectic condition it fixes the constants that the differential equation cannot see. As at the end of Section 3.7, the differential equation gives a recursion and an oscillating integral gives the initial conditions.

Writing R⁡(z)=Id+R1⁢z+R2⁢z2+⋯, we can use the equation above to determine each Rk+1 from the preceding ones. There are two features which make the quintic tractable. First, the root of unity basis has five sectors, and a mod 5 charge rule permits only one residue pattern in each row. Then most matrix entries vanish before any computation starts. Second, every coefficient lies in the finitely generated ring

so the stable graph sum reduces to a finite algebraic computation.

Remark 8.10.

In particular, a fixed genus only needs finitely much of R for dimensional reasons. On ℳ¯g,n only classes of total degree at most 3⁢g−3+n contribute and everything else integrates to zero, so for an unmarked potential we need only R0,…,R3⁢g−3. Genus 4 stops at z9 and genus 5 at z12.

8.8 Genus two.

The genus-two mirror formula discussed here is due to Guo–Janda–Ruan [GJR17].

We will now carry out both routes in genus two. The geometric route uses the t-twisted stable quotient theory, applies virtual localization to get five contributions, and then applies stable quotient wall crossing to land on quintic Gromov-Witten theory. The formal route uses the semisimple λ-twisted theory and the R-matrix to compute a stable graph sum. This gives only the formal leading part. Then the effective and meromorphic rest graphs must still be added.

Notation 8.11.

Write

where ℒ is the universal stable quotient line bundle and c1ℂ×⁢([−1])=−t (a sign convention we return to below). This is the stable quotient sign convention, and is not the same as the [1] appearing in the stable map cycle above.

0∞Σ2⁢(d)Γ2Σ1⁢(d)Σ1⁢(0)Γ1Σ0⁢(d)Σ1⁢(0)Σ1⁢(0)Γb0Σ0⁢(d)Σ1⁢(0)Γa0
Figure 6. The four non-constant localization graphs in genus two, with each vertex drawn as its Riemann surface component.

The four non-constant graphs are drawn in Figure 6, and their contributions are

and, writing H1=H⊗1 and H2=1⊗H for the double edge graph,

Note that these correlators live on ℙ4, so H4 is present and nonzero.

There is a fifth contribution from the degree zero graph, which is the constant

and adding the five contributions, we obtain the localization identity

where the two factors of 12 are the automorphisms of Γa0 and Γb0.

Set Q=q⁢eτ⁡(q) and L=(1−55⁢q)−1/5, where τ⁡(q)=I1⁢(q)/I0⁢(q)−log⁡q is the ordinary mirror shift. We now apply stable quotient wall crossing to obtain the exact identity

Introducing Du=L−1⁢q⁢ddq and the generators

the individual twisted graph contributions involve extra generators. The important point for us is that these cancel in the above linear combination and leave precisely the finite generation variables. Writing X=𝒳1 and Y1=𝒴1, we obtain

If we expand at large radius, then we obtain

and these coefficients are the Gromov-Witten invariants N2,d, not the BPS integers of Table 2.

We now turn to the formal route. The Givental-Teleman formula for Ω2,0λ is a sum over the seven stable graphs of genus two with no legs, which are drawn in Figure 7. Only R−1⁢(z) through z3 is needed, and the translations expand the seven topologies into 17 decorated strata.

2111121121012018001800112
Figure 7. The seven stable graphs of genus two with no legs, labelled by the genera of the vertices and weighted by 1|Aut⁡Γ|. On the slides each vertex was drawn as its Riemann surface component.
Remark 8.12.

These are dual graphs of stable curves and are not the graphs of Figure 5, which are the rigid tropical types obtained by applying the tropical decomposition formula to the reduced theory of the log GLSM. The two lists of topologies happen to coincide here, but a vertex means something different in each. Here it is a component of the domain curve carrying a topological field theory tensor. There it is a level of a bipartite tropical curve carrying either a log or an ∞-vertex moduli problem.

For the formal route, we define the normalized potential intrinsically by

The complete graph sum lies in ℚ⁡[L,𝒳1,𝒳2,𝒳3,𝒴1] and equals

For clarity, this is the exact formal λ-twisted answer and is not F2GW. Note also that Y=L5 is the discriminant generator while 𝒴1 is a logarithmic derivative generator.

8.9 Finite generation and the extended family.

Recall the finitely generated ring ℛ=ℚ⁡[A,B1,B2,B3,Y] from above, which we will grade by

For ϕk=I1,1⁢⋯⁢Ik,k⁢Hk as before, set

Theorem 8.13 (Graded finite generation [GJR18]).

We have

Consequently, after assembling the bipartite rest graphs, the non-equivariant component of the extended family satisfies

In particular, at each fixed (g,n) only finitely many graded monomials can appear, and only R-matrix orders up to 3⁢g−3+n can contribute.

We organize the extended theory using the bipartite graph sum

where C∞ is an effective or meromorphic vertex factor, Cλ is a formal zero-vertex correlator, the μv and δ→v record incident degrees, and Ve is the edge factor. The limit λ→0 is taken only after the complete graph sum. Setting 𝐂=0 leaves the formal λ-twisted leading sector. Setting 𝐂=𝐂eff and λ→0 recovers quintic Gromov-Witten theory. Differentiating a universal edge cuts it and produces the holomorphic anomaly equation, while leaving the effective vertices untouched.

To derive the holomorphic anomaly equation, write Pg𝐂≔Pg,0𝐂 and introduce

together with the two non-holomorphic directions

Theorem 8.14 (Holomorphic anomaly equation for the extended family [GJR18]).

We have

We prove this by the cutting argument of Section 8.4, which works because the derivative of the R-matrix propagator is simply

Derivatives of R−1 and of every edge remain inside the same graded ring, so cutting a differentiated edge closes up on lower-genus correlators in ℛ. If we set 𝐂=0, then we recover the same equations for the formal λ-twisted theory.

8.10 Boundary conditions.

We will use two sources of conditions on the holomorphic ambiguity.

First, we use regularity at the orbifold or Landau-Ginzburg point q=∞, equivalently L→0. For either the formal or an assembled extended potential Fg⋆, set

Theorem 8.13 bounds the possible Laurent powers of L, and the formal LG/CY identification says that after analytic continuation and identification of the normalized CY and LG bases, the same formal R-matrix polynomial describes both phases. Therefore

and for fg⁢(Y)=∑i=03⁢g−3ai,g⁢Yi with Y=L5 the quintic bound is

Note that this is a statement about the assembled normalized graph contribution, and not about individual bare meromorphic or rubber vertex correlators.

Second, we study the conifold. Set

where I0co,I1co form a local conifold period pair normalized so that the vanishing flat coordinate satisfies Tc=0 at the conifold. Given a rational expression f in the finite generation data, we obtain its conifold value fco by substituting the Laurent expansions of all generators as series in Tc. Theorem 8.13 gives the pole bound

but what we need is the stronger gap condition

namely that every intermediate negative power Tc−(2⁢g−3),…,Tc−1 vanishes. This is 2⁢g−2 conditions and is not implied by finite generation.

Conjecture 8.15 (Formal conifold gap).

The formal λ-twisted potential has the universal leading term

where B2⁢g is a Bernoulli number.

Example 8.16.

In genus two the computed expansion begins

This is the genus two gap, not merely the pole order bound.

We must also consider the formal correlators with rational descendant insertions which occur in the rest graphs, namely

with δi≥1 and ai∈{0,1,2,3,4}.

Conjecture 8.17 (Regularity with special insertions).

We have (Cg,δ→λ⁢(a→))co=O⁡(1).

For fixed a→ and δ→, individual terms in the ψ-expansion may have poles. Note that the assertion concerns the complete rational descendant insertion, whose negative powers cancel before the rest graph sum over a→ is taken. It is a statement about these specific formal correlators rather than about every unassembled meromorphic correlator.

We can use Conjectures 8.15 and 8.17 to transfer the gap to quintic Gromov-Witten theory. A rest graph contribution has the shape

and the C∞ factors are finite polynomials in q, so they are conifold regular. Then the formal leading gap together with regularity of the special correlators implies the gap for the extended family, and therefore for the quintic.

Remark 8.18.

Conjectures 8.15 and 8.17 are open in general, but have been verified for the quintic through g≤5 by Guo-Janda-Ruan. The pole bound for the extended family is a theorem, and the gap follows from the two conjectures and is proved for g≤5 by explicit computation. At the time of these lectures, the all-genus gap for quintic Gromov-Witten theory was therefore conditional on them and unconditional for g≤5.13

Putting all of this together, we begin with the Picard-Fuchs data and compute the Givental R-matrix and the operator Sτ⁢(z). We then obtain the formal stable graphs and meromorphic rest graphs, use the holomorphic anomaly equation and finite generation, and finally impose the orbifold and conifold boundary conditions to determine FgGW.

9 Structure of higher-genus GW theory of local ℙ1 (Shuai Guo)

This lecture is based on joint work with Jingyi Xu and Qingsheng Zhang.

9.1 The geometry of Xp.

We will consider the total space

which is a non-compact toric Calabi-Yau threefold. For p=0 this is the resolved conifold.

Let N≃ℤ3 be a lattice of rank 3 and let M≔Hom⁡(N,ℤ) be the dual lattice, which we identify with the characters of 𝕋≃(ℂ×)3. The one-dimensional cones of the fan of Xp are generated by

and Xp contains 𝕋 as an open dense subset. Using the dual vector e3∨∈M we obtain the Calabi-Yau subtorus

whose action preserves the Calabi-Yau form. The toric graph consists of two triangles, which correspond to the two fixed points of the torus action. The weights of the 𝕋′-action at the first fixed point are

and at the second fixed point they are

We can therefore express the invariants in terms of equivariant cohomology by localizing at these two fixed points. Here, H𝕋′∗⁢(pt)=ℂ⁡[λ1,λ2].

Unfortunately, for p≥2 the toric graph of Xp is not convex. Then Xp is not semiprojective. This is the reason the case at hand is not covered by the existing literature, as we will explain in Section 9.5.

9.2 The conjecture of CGMPS.

In 2007, Caporaso, Griguolo, Mariño, Pasquetti, and Seminara studied the Gromov-Witten theory of Xp. Writing

for the total free energy, they computed the genus-one part exactly and proposed the ansatz

for g≥2. Here, the variable w is related to the Kähler parameter q by w=1−q and wc=1−1p2. Our convention will differ by FgXp=(−1)g−1⁢FˇgXp for g≥2.

Analyzing the asymptotics of the genus-zero invariants, CGMPS defined the scaled string coupling by

and introduced the double-scaling limit Fds⁢(z), which is obtained from FˇXp⁢(w,ℏ) by letting w→wc and ℏ→0 with z fixed. Combining their genus-zero and genus-one results with the ansatz gives

where bg=(p84⁢wc5)g−1⁢𝒫~g⁢(wc,p).

Conjecture 9.1 (CGMPS [Cap+07]).

The double-scaling limit Fds⁢(z) coincides with the free energy F(3,2)⁢(z) of the (3,2) minimal model.

The free energy of the (3,2) minimal model has the form

where

so we see that Conjecture 9.1 is equivalent to the statement that bg=ag for all g≥2.

9.3 The main results.

The global mirror symmetry picture of Section 8 applies here as well. Our goal is to describe the behavior of FgXp at the three boundary points of the one-dimensional B-model moduli space. The limit q→0 is the large complex structure limit and gives the Gromov-Witten expansion in the A-model variable Q. At q→1, we will use the framed R-matrix to prove that Fg remains regular. The interesting point is q→1p2, which is a higher-order critical point. Unlike an ordinary conifold singularity, whose genus-g leading pole has order 2⁢g−2, this degeneration has pole order at most 5⁢g−5. Its leading coefficient is governed by the (3,2) minimal model.

Theorem 9.2.

For any integer p≥2 and any g≥2, the Gromov-Witten potential of Xp is computed by the Eynard-Orantin topological recursion of a spectral curve 𝒞, namely

In particular, this gives a non-semiprojective toric Calabi-Yau example of the remodeling conjecture.

Theorem 9.3.

For g≥2, the potential FgXp is the expansion of a rational function

where 𝒫g∈1(p2−1)2⁢g−2⁢ℚ⁢[p2]⁢[Δ]≤5⁢g−5.

In particular, this proves the structural ansatz Equation 2.

To identify the double-scaling limit, it remains to determine the constant term of 𝒫g at Δ=0.

Theorem 9.4.

The constant term of 𝒫g is

Because 𝒫~g⁢(wc,p)=𝒫g⁢(0,p2)⁢(1p2−1p4)5⁢g−5, Theorem 9.4 gives

which proves Conjecture 9.1.

9.4 Gromov-Witten theory of Xp.

For the equivariant theory of local curves and its localization formulas, see [BP08].

Because Xp is noncompact, the moduli space ℳ¯g,0⁢(Xp,d) is not proper and the degree of the virtual class is not defined a priori. To define the invariants, we will introduce a torus action and pass to a twisted theory on ℙ1, where we can apply virtual localization.

We choose a framing f∈ℤ and let 𝕋f≔ker⁡(λ2−f⁢λ1). The two localized weight rows, each of which includes one zero weight, are

If f≠0, then rescaling the character gives an equivalent symmetric weight vector

and we denote the adjusted torus by 𝕋𝔣. If H𝕋𝔣∗⁢(pt)=ℂ⁡[λ], this normalization is

Let π:ℳ¯g,1⁢(ℙ1,d)→ℳ¯g,0⁢(ℙ1,d) be the forgetful map and let ev be the evaluation at the marking. Then we define

where e𝕋𝔣 is the equivariant Euler class. For p=0 this specializes to the usual resolved conifold formula.

We obtain a CohFT Ωλ,𝔣 with state space the equivariant cohomology ring

and pairing

In the flat basis {𝟏,H} the Gram matrix is given by

We define the correlators by

where st is the forgetful map to ℳ¯g,n. The twisted virtual class is given by

Set Q=(−1)p+1⁢e−t. Then the genus-g potential is given by

Remark 9.5.

The degree-positive invariants Ng,d≥1Xp are independent of 𝔣 by the topological vertex [LLLZ09], and their λ-independence also follows from Theorem 9.8. We will use the 𝔣-independence in the proof of Theorem 9.3.

In genus zero, we have

which allows us to compute the Yukawa coupling exactly:

Here, q and Q are related by the mirror map Q=q⁢(1−q)p2−1. Evaluating the quantum product using (v1⋆v2,v3)=Ω0,3λ,𝔣⁢(v1,v2,v3), we obtain

or in matrix form

The eigenvalues Lα of this matrix are generically distinct. Then Ωλ,𝔣 is semisimple. We write {eα} for the idempotent basis, Δα−1=(eα,eα) for the lengths, and e¯α≔Δα1/2⁢eα for the normalized canonical basis.

Because the theory is semisimple, Theorem 3.24 gives

where ΩβKW is a one-dimensional trivial CohFT and TR⁢(z)=z⁢(𝟏~−R−1⁢(z)⁢𝟏) with 𝟏~≔∑βe¯β. We determine the R-matrix from the quantum differential equation

Equivariant localization and quantum Riemann-Roch relate the fundamental solution to the R-matrix by

where the initial condition Cα⁢(z) is an exponential involving Bernoulli numbers. Writing Ri⁢α¯⁢(z)≔(R∗⁢(z)⁢Hi,e¯α), we obtain from the quantum differential equation the second-order equation

where D=−∂t. We can then compute the R-matrix order by order in z.

9.5 Topological recursion and the remodeling conjecture.

We now pass to the B-model side. Our goal is to identify the reconstruction supplied by topological recursion with the Givental-Teleman reconstruction above. The Eynard–Orantin topological recursion [EO07] takes as input a spectral curve (Σ,x,y,B), where Σ is a Riemann surface; x is locally defined with meromorphic differential dx having finitely many simple zeros; y is holomorphic near those zeros, with dy≠0 there; and B is a fundamental bidifferential on Σ×Σ. Near a zero zβ of dx, let z¯≠z be the local conjugate point with x⁡(z¯)=x⁡(z) and define the recursion kernel

If we set ω0,0=ω1,0=0, ω0,1⁢(z)=y⁡(z)⁢dx⁡(z), and ω0,2⁢(z1,z2)=B⁡(z1,z2), then the multi-differentials are defined recursively for 2⁢g−2+n+1>0 by

where the primed sum excludes ω0,1. For g≥2 the free energy is given by

The remodeling conjecture [BKMP09] of Bouchard, Klemm, Mariño, and Pasquetti asserts that the all-genus open and closed Gromov-Witten invariants of a toric Calabi-Yau threefold X coincide with the invariants ωg,n produced by topological recursion on its mirror curve. It was proved for ℂ3, the resolved conifold, and local curves by Eynard-Orantin, Zhou, and Chen-Li-Liu. Then Fang–Liu–Zong [FLZ20] proved it for all semiprojective toric Calabi-Yau threefolds. Unfortunately, Xp is not semiprojective for p≥2, so it falls outside the scope of that proof.

To compare the two sides we use the reconstruction of Dunin-Barkowski, Orantin, Shadrin, and Spitz [DOSS14], which identifies the stable topological recursion correlators with the Givental reconstruction of a semisimple CohFT. Around each critical point zβ, define the local Airy coordinate ηβ⁢(z) by

and the globally defined meromorphic reference form

The state space V of the associated CohFT is spanned by the basis {e¯β} corresponding to the critical points, and the pairing is diagonal with η⁡(e¯β,e¯γ)=δβ,γ. The curve R-matrix Rˇ is defined by the asymptotic Laplace transform

and the analogous transform of dy defines T⁡(u), and if the resulting CohFT has a flat unit, then T⁡(u)=u⁢(𝟏~−Rˇ−1⁢(u)⁢𝟏). Then Rˇ is symplectic, the curve CohFT is Ωˇ=Rˇ.T.(⨁βΩβKW), and writing dζkβ¯≔(−d∘1d⁢x)k⁢dζβ¯, the correspondence reads

For Xp, we will consider the spectral curve

where

Solving dx𝔣=0, we obtain the critical points

and the inner products are given by η⁡(eα,eβ)=δα,β⁢y′⁢(zα)2x𝔣′′⁢(zα). Then the resulting lengths satisfy Δˇα=Δα. Therefore, the Frobenius structure of the spectral curve is the quantum product of Xp.

We have now matched the Frobenius structures, so it remains to match the two R-matrices. Integration by parts on the Lefschetz thimbles gives

where

are exactly the entries of the quantum product computed above, so these agree with the quantum differential equation for the Gromov-Witten R-matrix. To match the initial conditions, we take the limit q→0 of the Laplace transforms along the thimbles, which produces Beta functions and therefore products of Gamma functions. Then Stirling’s expansion matches the Bernoulli-number initial condition Cα⁢(z).

Proposition 9.6.

The two R-matrices coincide, namely R⁡(z)=Rˇ⁢(u)|u=−z.

We have now shown that the canonical lengths and the R-matrices agree. After the substitution u=−z, the genus-g stable graph sums differ by the overall parity (−1)3⁢g−3=(−1)g−1, which proves Theorem 9.2.

9.6 Rationality.

Because the higher-genus potential is independent of 𝔣, we may, for simplicity, set 𝔣=0. This simplifies both the canonical basis and the quantum differential equation, and the latter becomes

and we denote the entries of the R-matrix by (Rk)iα. Set

We now use induction on the coefficients of the quantum differential equation to obtain a polynomial structure.

Proposition 9.7.

For k∈ℤ≥0 and i=0,1, we have

Analyzing the degree contributions of the stable graph sum in the Givental-Teleman reconstruction, and tracking the powers of λ, Δ, and p2−1, we obtain the following.

Theorem 9.8.

Writing [⋅]d for the codimension-d component, we have

Taking n=0 and d=3⁢g−3=dimℳ¯g in Theorem 9.8 gives

so Theorem 9.3 will follow if the degree of the numerator in Δ is bounded by 5⁢g−5. We can see this bound at q=1. Because Δ has a simple pole there, regularity of FgXp forces the numerator to cancel the highest powers of Δ. We have therefore reduced rationality to regularity at q=1. This regularity follows from the 𝔣-independence of FgXp together with the existence of the limit q→1 of the R-matrix for 𝔣≠0.

9.7 The limit Δ=0.

To determine the remaining constant term, we now study the limit Δ=0, which is precisely the double-scaling limit. We rescale the canonical lengths by

and define the modified R-matrix entries by isolating the leading terms at Δ=0 in Proposition 9.7, namely

Then the graph sum for FgXp becomes a formula for 𝒫g⁢(0,p2) after replacing Δα by δα and (Rk)iα by (R¯k)iα.

Lemma 9.9.

Solving the leading-order quantum differential equation at Δ=0 gives the closed forms

After factoring out the p-dependent terms in Lemma 9.9, we obtain purely rational coefficients. We now construct the topological field theory Ωtop on V=span⁡{ϕ0,ϕ1} with unit ϕ0, where

Applying reconstruction by R⁡(z) together with its flat-unit translation, we obtain a constant Cg with

and it remains to identify the reconstructed CohFT.

The (2,3) minimal model has spectral curve

with flat coordinate τ1=ε4/2. If we set τ0=0 and ε=1 and identify ϕ0=∂τ0 and ϕ1=∂τ1, then the R-matrix obtained from Xp at Δ=0 coincides with the R-matrix of this curve. Therefore, Cg is exactly the free energy of the (2,3) minimal model.

Finally, we apply the symplectic swap (x,y)↦(y,−x), under which the curve becomes

The swapped curve has one-dimensional state space, trivial R-matrix, and translation T⁡(u)=u2⁢e¯, so that

which is Theorem 9.4. Note that this swap is the correspondence between the (2,3) and the (3,2) minimal models, which is why the intersection numbers appearing here are those of Conjecture 9.1.

References

  • [ACGS25]
    Dan Abramovich, Qile Chen, Mark Gross, and Bernd Siebert. Punctured logarithmic maps. Vol. 15. Memoirs of the European Mathematical Society. EMS Press, Berlin, 2025. viii+156. doi: 10.4171/mems/15.
  • [ACP15]
    Dan Abramovich, Lucia Caporaso, and Sam Payne. “The tropicalization of the moduli space of curves”. In: Ann. Sci. Éc. Norm. Supér. (4) 48.4 (2015), pp. 765–809. doi: 10.24033/asens.2258.
  • [AHR20]
    Jarod Alper, Jack Hall, and David Rydh. “A Luna étale slice theorem for algebraic stacks”. In: Ann. of Math. (2) 191.3 (2020), pp. 675–738. doi: 10.4007/annals.2020.191.3.1.
  • [AL07]
    Murad Alim and Jean Dominique Länge. “Polynomial structure of the (open) topological string partition function”. In: J. High Energy Phys. 10, 045 (2007). doi: 10.1088/1126-6708/2007/10/045.
  • [Alp13]
    Jarod Alper. “Good moduli spaces for Artin stacks”. In: Ann. Inst. Fourier (Grenoble) 63.6 (2013), pp. 2349–2402. doi: 10.5802/aif.2833.
  • [BBD82]
    A. A. Beĭlinson, J. Bernstein, and P. Deligne. “Faisceaux pervers”. In: Analysis and topology on singular spaces, I (Luminy, 1981). Vol. 100. Astérisque. Soc. Math. France, Paris, 1982, pp. 5–171. url: https://www.numdam.org/item/AST_1982__100__1_0/.
  • [BCOV94]
    M. Bershadsky, S. Cecotti, H. Ooguri, and C. Vafa. “Kodaira-Spencer theory of gravity and exact results for quantum string amplitudes”. In: Comm. Math. Phys. 165.2 (1994), pp. 311–427. doi: 10.1007/BF02099774.
  • [BF97]
    K. Behrend and B. Fantechi. “The intrinsic normal cone”. In: Invent. Math. 128.1 (1997), pp. 45–88. doi: 10.1007/s002220050136.
  • [BKMP09]
    Vincent Bouchard, Albrecht Klemm, Marcos Mariño, and Sara Pasquetti. “Remodeling the B-model”. In: Comm. Math. Phys. 287.1 (2009), pp. 117–178. doi: 10.1007/s00220-008-0620-4.
  • [BM83]
    Walter Borho and Robert MacPherson. “Partial resolutions of nilpotent varieties”. In: Analysis and topology on singular spaces, II, III (Luminy, 1981). Astérisque 101–102. Soc. Math. France, Paris, 1983, pp. 23–74. url: https://numdam.org/item/AST_1983__101-102__23_0/.
  • [BP08]
    Jim Bryan and Rahul Pandharipande. “The local Gromov-Witten theory of curves”. In: J. Amer. Math. Soc. 21.1 (2008). With an appendix by Bryan, C. Faber, A. Okounkov and Pandharipande, pp. 101–136. doi: 10.1090/S0894-0347-06-00545-5.
  • [Cap+07]
    Nicola Caporaso, Luca Griguolo, Marcos Mariño, Sara Pasquetti, and Domenico Seminara. “Phase transitions, double-scaling limit, and topological strings”. In: Phys. Rev. D 75.4, 046004 (2007). doi: 10.1103/PhysRevD.75.046004.
  • [CG07]
    Tom Coates and Alexander Givental. “Quantum Riemann-Roch, Lefschetz and Serre”. In: Ann. of Math. (2) 165.1 (2007), pp. 15–53. doi: 10.4007/annals.2007.165.15.
  • [CGL21]
    Huai-Liang Chang, Shuai Guo, and Jun Li. “Polynomial structure of Gromov-Witten potential of quintic 3-folds”. In: Ann. of Math. (2) 194.3 (2021), pp. 585–645. doi: 10.4007/annals.2021.194.3.1.
  • [CGL26]
    Huai-Liang Chang, Shuai Guo, and Jun Li. “BCOV Feynman rule of the quintic threefold”. In: Duke Math. J. 175.4 (2026), pp. 607–675. doi: 10.1215/00127094-2025-0038.
  • [CGYZ26]
    Huai-Liang Chang, Shuai Guo, Lei You, and Huaigong Zhang. Conifold Gap for the Quintic Threefold. Preprint. 2026. arXiv: 2609.13921 [math.AG].
  • [CJR]
    Qile Chen, Felix Janda, and Yongbin Ruan. “Structural formulae in logarithmic Gauged Linear Sigma Models III: The localization formula”. In preparation. url: https://fjanda.web.illinois.edu/research.html (visited on 09/16/2026).
  • [CJR21]
    Qile Chen, Felix Janda, and Yongbin Ruan. “The logarithmic gauged linear sigma model”. In: Invent. Math. 225.3 (2021), pp. 1077–1154. doi: 10.1007/s00222-021-01044-2.
  • [CJR22]
    Qile Chen, Felix Janda, and Yongbin Ruan. Punctured logarithmic R-maps. Preprint. 2022. arXiv: 2208.04519 [math.AG].
  • [CJR25]
    Qile Chen, Felix Janda, and Yongbin Ruan. “Structural formulae in logarithmic Gauged Linear Sigma Models I: The tropical decomposition formula”. Working draft. Jan. 6, 2025. url: https://fjanda.web.illinois.edu/Tropical.pdf.
  • [CJW21]
    Qile Chen, Felix Janda, and Rachel Webb. “Virtual cycles of stable (quasi-)maps with fields”. In: Adv. Math. 385, 107781 (2021). doi: 10.1016/j.aim.2021.107781.
  • [CK14]
    Ionuţ Ciocan-Fontanine and Bumsig Kim. “Wall-crossing in genus zero quasimap theory and mirror maps”. In: Algebr. Geom. 1.4 (2014), pp. 400–448. doi: 10.14231/AG-2014-019.
  • [CK20]
    Ionuţ Ciocan-Fontanine and Bumsig Kim. “Quasimap wall-crossings and mirror symmetry”. In: Publ. Math. Inst. Hautes Études Sci. 131 (2020), pp. 201–260. doi: 10.1007/s10240-020-00114-0.
  • [CKM14]
    Ionuţ Ciocan-Fontanine, Bumsig Kim, and Davesh Maulik. “Stable quasimaps to GIT quotients”. In: J. Geom. Phys. 75 (2014), pp. 17–47. doi: 10.1016/j.geomphys.2013.08.019.
  • [CL22]
    You-Cheng Chou and Y. -P. Lee. Quantum K-invariants and Gopakumar–Vafa invariants I. The quintic threefold. Preprint. 2022. arXiv: 2211.00788 [math.AG].
  • [CL25]
    You-Cheng Chou and Yuan-Pin Lee. “Quantum K-invariants and Gopakumar-Vafa invariants, II: Calabi-Yau threefolds at genus zero”. In: Geom. Topol. 29.9 (2025), pp. 4665–4693. doi: 10.2140/gt.2025.29.4665.
  • [CM02]
    Mark Andrea A. de Cataldo and Luca Migliorini. “The hard Lefschetz theorem and the topology of semismall maps”. In: Ann. Sci. École Norm. Sup. (4) 35.5 (2002), pp. 759–772. doi: 10.1016/S0012-9593(02)01108-4.
  • [COGP91]
    Philip Candelas, Xenia C. de la Ossa, Paul S. Green, and Linda Parkes. “A pair of Calabi-Yau manifolds as an exactly soluble superconformal theory”. In: Nuclear Phys. B 359.1 (1991), pp. 21–74. doi: 10.1016/0550-3213(91)90292-6.
  • [Del68]
    P. Deligne. “Théorème de Lefschetz et critères de dégénérescence de suites spectrales”. In: Inst. Hautes Études Sci. Publ. Math. 35 (1968), pp. 107–126. doi: 10.1007/BF02698925.
  • [DOSS14]
    P. Dunin-Barkowski, N. Orantin, S. Shadrin, and L. Spitz. “Identification of the Givental formula with the spectral curve topological recursion procedure”. In: Comm. Math. Phys. 328.2 (2014), pp. 669–700. doi: 10.1007/s00220-014-1887-2.
  • [EHKV01]
    Dan Edidin, Brendan Hassett, Andrew Kresch, and Angelo Vistoli. “Brauer groups and quotient stacks”. In: Amer. J. Math. 123.4 (2001), pp. 761–777. doi: 10.1353/ajm.2001.0024.
  • [EO07]
    B. Eynard and N. Orantin. “Invariants of algebraic curves and topological expansion”. In: Commun. Number Theory Phys. 1.2 (2007), pp. 347–452. doi: 10.4310/CNTP.2007.v1.n2.a4.
  • [FLZ20]
    Bohan Fang, Chiu-Chu Melissa Liu, and Zhengyu Zong. “On the remodeling conjecture for toric Calabi-Yau 3-orbifolds”. In: J. Amer. Math. Soc. 33.1 (2020), pp. 135–222. doi: 10.1090/jams/934.
  • [Giv00]
    Alexander Givental. “On the WDVV equation in quantum K-theory”. In: Michigan Math. J. 48 (2000). Dedicated to William Fulton on the occasion of his 60th birthday, pp. 295–304. doi: 10.1307/mmj/1030132720.
  • [Giv01a]
    Alexander B. Givental. “Gromov-Witten invariants and quantization of quadratic Hamiltonians”. In: Mosc. Math. J. 1.4 (2001), pp. 551–568, 645. doi: 10.17323/1609-4514-2001-1-4-551-568.
  • [Giv01b]
    Alexander B. Givental. “Semisimple Frobenius structures at higher genus”. In: Internat. Math. Res. Notices 23 (2001), pp. 1265–1286. doi: 10.1155/S1073792801000605.
  • [Giv04]
    Alexander B. Givental. “Symplectic geometry of Frobenius structures”. In: Frobenius manifolds. Quantum Cohomology and Singularities. Ed. by Claus Hertling and Matilde Marcolli. Aspects Math. E36. Friedr. Vieweg, Wiesbaden, 2004, pp. 91–112. doi: 10.1007/978-3-322-80236-1_4.
  • [Giv15a]
    Alexander Givental. Permutation-equivariant quantum K-theory II. Fixed point localization. Preprint. 2015. arXiv: 1508.04374 [math.AG].
  • [Giv15b]
    Alexander Givental. Permutation-equivariant quantum K-theory VII. General theory. Preprint. 2015. arXiv: 1510.03076 [math.AG].
  • [Giv17]
    Alexander Givental. “Permutation-equivariant quantum K-theory I. Definitions. Elementary K-theory of ℳ¯0,n/Sn”. In: Mosc. Math. J. 17.4 (2017), pp. 691–698. doi: 10.17323/1609-4514-2017-17-4-691-698.
  • [Giv96]
    Alexander B. Givental. “Equivariant Gromov-Witten invariants”. In: Internat. Math. Res. Notices 13 (1996), pp. 613–663. doi: 10.1155/S1073792896000414.
  • [Giv98]
    Alexander Givental. “A mirror theorem for toric complete intersections”. In: Topological field theory, primitive forms and related topics (Kyoto, 1996). Ed. by Masaki Kashiwara, Atsushi Matsuo, Kyoji Saito, and Ikuo Satake. Vol. 160. Progr. Math. Birkhäuser Boston, Boston, MA, 1998, pp. 141–175. doi: 10.1007/978-1-4612-0705-4_5.
  • [GJR17]
    Shuai Guo, Felix Janda, and Yongbin Ruan. A mirror theorem for genus two Gromov-Witten invariants of quintic threefolds. Preprint. 2017. arXiv: 1709.07392 [math.AG].
  • [GJR18]
    Shuai Guo, Felix Janda, and Yongbin Ruan. Structure of Higher Genus Gromov-Witten Invariants of Quintic 3-folds. Preprint. 2018. arXiv: 1812.11908 [math.AG].
  • [GP99]
    T. Graber and R. Pandharipande. “Localization of virtual classes”. In: Invent. Math. 135.2 (1999), pp. 487–518. doi: 10.1007/s002220050293.
  • [GS22]
    Stavros Garoufalidis and Emanuel Scheidegger. “On the quantum K-theory of the quintic”. In: SIGMA Symmetry Integrability Geom. Methods Appl. 18, 021 (2022). doi: 10.3842/SIGMA.2022.021.
  • [GT14]
    Alexander Givental and Valentin Tonita. “The Hirzebruch-Riemann-Roch theorem in true genus-0 quantum K-theory”. In: Symplectic, Poisson, and noncommutative geometry. Ed. by Tohru Eguchi, Yakov Eliashberg, and Yoshiaki Maeda. Vol. 62. Math. Sci. Res. Inst. Publ. Cambridge Univ. Press, New York, 2014, pp. 43–91. doi: 10.1017/9781107297258.003.
  • [Hal14]
    Daniel Halpern-Leistner. On the structure of instability in moduli theory. Preprint. 2014. arXiv: 1411.0627 [math.AG].
  • [Hal15]
    Daniel Halpern-Leistner. Remarks on Theta-stratifications and derived categories. Preprint. 2015. arXiv: 1502.03083 [math.AG].
  • [Has03]
    Brendan Hassett. “Moduli spaces of weighted pointed stable curves”. In: Adv. Math. 173.2 (2003), pp. 316–352. doi: 10.1016/S0001-8708(02)00058-0.
  • [HN74]
    G. Harder and M. S. Narasimhan. “On the cohomology groups of moduli spaces of vector bundles on curves”. In: Math. Ann. 212 (1974–1975), pp. 215–248. doi: 10.1007/BF01357141.
  • [IP18]
    Eleny-Nicoleta Ionel and Thomas H. Parker. “The Gopakumar-Vafa formula for symplectic manifolds”. In: Ann. of Math. (2) 187.1 (2018), pp. 1–64. doi: 10.4007/annals.2018.187.1.1.
  • [Kal06]
    D. Kaledin. “Symplectic singularities from the Poisson point of view”. In: J. Reine Angew. Math. 600 (2006), pp. 135–156. doi: 10.1515/CRELLE.2006.089.
  • [Kat00]
    Fumiharu Kato. “Log smooth deformation and moduli of log smooth curves”. In: Internat. J. Math. 11.2 (2000), pp. 215–232. doi: 10.1142/S0129167X0000012X.
  • [Kat89]
    Kazuya Kato. “Logarithmic structures of Fontaine-Illusie”. In: Algebraic analysis, geometry, and number theory (Baltimore, MD, 1988). Ed. by Jun-ichi Igusa. Johns Hopkins Univ. Press, Baltimore, MD, 1989, pp. 191–224. url: https://math.uchicago.edu/~drinfeld/p-adic_periods/Kato_log-structures.pdf.
  • [KL13]
    Young-Hoon Kiem and Jun Li. “Localizing virtual cycles by cosections”. In: J. Amer. Math. Soc. 26.4 (2013), pp. 1025–1050. doi: 10.1090/S0894-0347-2013-00768-7.
  • [KM94]
    M. Kontsevich and Yu. Manin. “Gromov-Witten classes, quantum cohomology, and enumerative geometry”. In: Comm. Math. Phys. 164.3 (1994), pp. 525–562. doi: 10.1007/BF02101490.
  • [KM97]
    Seán Keel and Shigefumi Mori. “Quotients by groupoids”. In: Ann. of Math. (2) 145.1 (1997), pp. 193–213. doi: 10.2307/2951828.
  • [Kon92]
    Maxim Kontsevich. “Intersection theory on the moduli space of curves and the matrix Airy function”. In: Comm. Math. Phys. 147.1 (1992), pp. 1–23. doi: 10.1007/BF02099526.
  • [Lee04]
    Y.-P. Lee. “Quantum K-theory. I. Foundations”. In: Duke Math. J. 121.3 (2004), pp. 389–424. doi: 10.1215/S0012-7094-04-12131-1.
  • [Lee97]
    Yuan-Pin Lee. “A formula for Euler characteristics of tautological line bundles on the Deligne-Mumford moduli spaces”. In: Internat. Math. Res. Notices 8 (1997), pp. 393–400. doi: 10.1155/S1073792897000263.
  • [LLLZ09]
    Jun Li, Chiu-Chu Melissa Liu, Kefeng Liu, and Jian Zhou. “A mathematical theory of the topological vertex”. In: Geom. Topol. 13.1 (2009), pp. 527–621. doi: 10.2140/gt.2009.13.527.
  • [LLY97]
    Bong H. Lian, Kefeng Liu, and Shing-Tung Yau. “Mirror principle. I”. In: Asian J. Math. 1.4 (1997), pp. 729–763. doi: 10.4310/AJM.1997.v1.n4.a5.
  • [LP19]
    Hyenho Lho and Rahul Pandharipande. “Holomorphic anomaly equations for the formal quintic”. In: Peking Math. J. 2.1 (2019), pp. 1–40. doi: 10.1007/s42543-018-0008-0.
  • [LQ14]
    Y.-P. Lee and F. Qu. “Euler characteristics of universal cotangent line bundles on ℳ¯1,n”. In: Proc. Amer. Math. Soc. 142.2 (2014), pp. 429–440. doi: 10.1090/S0002-9939-2013-11800-9.
  • [LR22]
    Zhiyu Liu and Yongbin Ruan. Castelnuovo bound and higher genus Gromov-Witten invariants of quintic 3-folds. Preprint. 2022. arXiv: 2210.13411 [math.AG].
  • [MFK94]
    D. Mumford, J. Fogarty, and F. Kirwan. Geometric invariant theory. 3rd ed. Vol. 34. Ergebnisse der Mathematik und ihrer Grenzgebiete (2) [Results in Mathematics and Related Areas (2)]. Springer-Verlag, Berlin, 1994. xiv+292. doi: 10.1007/978-3-642-57916-5.
  • [Mil25]
    Todor Milanov. “K-theoretic Heisenberg algebras and permutation-equivariant Gromov-Witten theory”. In: Int. Math. Res. Not. IMRN 11, rnaf139 (2025). doi: 10.1093/imrn/rnaf139.
  • [Ols03]
    Martin C. Olsson. “Logarithmic geometry and algebraic stacks”. In: Ann. Sci. École Norm. Sup. (4) 36.5 (2003), pp. 747–791. doi: 10.1016/j.ansens.2002.11.001.
  • [PPZ15]
    Rahul Pandharipande, Aaron Pixton, and Dimitri Zvonkine. “Relations on ℳ¯g,n via 3-spin structures”. In: J. Amer. Math. Soc. 28.1 (2015), pp. 279–309. doi: 10.1090/S0894-0347-2014-00808-0.
  • [Tel12]
    Constantin Teleman. “The structure of 2D semi-simple field theories”. In: Invent. Math. 188.3 (2012), pp. 525–588. doi: 10.1007/s00222-011-0352-5.
  • [Toë99]
    B. Toën. “Théorèmes de Riemann-Roch pour les champs de Deligne-Mumford”. In: K-Theory 18.1 (1999), pp. 33–76. doi: 10.1023/A:1007791200714.
  • [YY04]
    Satoshi Yamaguchi and Shing-Tung Yau. “Topological string partition functions as polynomials”. In: J. High Energy Phys. 7, 047 (2004). doi: 10.1088/1126-6708/2004/07/047.
  • [Zho22]
    Yang Zhou. “Quasimap wall-crossing for GIT quotients”. In: Invent. Math. 227.2 (2022), pp. 581–660. doi: 10.1007/s00222-021-01071-z.
  • [ZZ08]
    Don Zagier and Aleksey Zinger. “Some properties of hypergeometric series associated with mirror symmetry”. In: Modular forms and string duality. Ed. by Noriko Yui, Helena Verrill, and Charles F. Doran. Vol. 54. Fields Inst. Commun. Amer. Math. Soc., Providence, RI, 2008, pp. 163–177. doi: 10.1090/fic/054/07.