Patrick Lei

All notes

The B-model

Lectures by Si Li, Bilibili, Feb 2026.

Abstract.

These are notes taken based on the Bilibili minicourse on the B-model given by Si Li in Spring 2025 at the Chinese University of Hong Kong. The minicourse consisted of 12 lectures.

1 Introduction

The B-model has its origins in mirror symmetry, which is a duality of two dimensional superconformal field theories (SCFTs) in physics. These describe the geometry of a mapping space from a surface Σ to a target space X (see Figure 1). The duality gives mathematical predictions which relate symplectic geometry (the A-model) to complex geometry (the B-model).

Σ2fX
Figure 1. Map from a surface Σ to a target space X.

There is a duality

where both sides involve some kind of infinite-dimensional geometry. However, we can reduce this to a finite-dimensional geometry by a form of supersymmetric localization, which we will not explain. The interested reader can refer to the speaker’s lectures at the Perimeter Institute on supersymmetry and TQFTs, which are also on Bilibili.

There is a string description via a worldsheet theory, where we consider maps as in Figure 2. This theory is called string field theory. This describes string theory using the language of quantum field theory. The correspondence goes as in Table 1.

Σ, worldsheetfX
Figure 2. Map in the worldsheet theory X.
Table 1. Idea of string field theory
String Fock space String field theory
|ψ⟩ ψ
Dynamics S⁡[ψ] String field action
|ψ⟩ ψ

We outline some historical developments in both mathematics and physics related to the B-model.

  • •

    In 1992, Zwiebach described a general structure for closed string field theory and discovered that it is described by a string action which contains infinitely many vertices and satisfies the BV master equation, something we will describe later in the language of the B-model. In particular, if we consider the genus 0 closed string field, we will discover a homotopy Lie algebra or an L∞-algebra.

  • •

    In 1997, Zwiebach extended this theory to the open string field, which is described by an A∞-algebra. When coupled with the closed string field, it satisfies the open-closed master equation, which is described via the moduli of bordered Riemann surfaces.

We will consider the B-model from the perspective of string field theory. This is usually referred to as the B-twisted topological field theory, and when coupled to gravity, it becomes the topological string theory. It is localized around constant maps, and near each constant map, we see the “local” string field action. The correspondence goes as in Figure 3.

×××XX∫𝖬𝖺𝗉⁡(Σ,X)(−)∫X(−)
Figure 3. Localization of the B-model.

Here are some developments in the history of describing this new action:

  • •

    Witten described the A-model open string field using Chern-Simons and instanton theories in 1992. As for the B-model open string field, he described it as holomorphic Chern-Simons theory. Finally, he speculated that the theory is actually UV-finite, which we know now is true by the work of M. Wang and J. Yan.

  • •

    The famous work of Bershadsky-Cecotti-Ooguri-Vafa in 1994 described the B-model closed string field theory on a Calabi-Yau threefold as a gauge theory, which they call Kodaira-Spencer gravity. This has some interesting features:

    • –

      At the classical level, the equation of motion describes deformation of complex structures on the Calabi-Yau threefold. By Yau’s proof of the Calabi conjecture, this leads to a family of solutions which are Ricci-flat metrics.

    • –

      The leading cubic vertex of Zwiebach’s string action in the topological B-model admits a full description.

  • •

    There is a construction by Barannikov-Kontsevich which constructs the Frobenius manifold (which encodes classical genus-zero data) of this theory via the geometry of polyvector fields, which will be discussed in the next few lectures.

  • •

    In 2012, the speaker and Kevin Costello gave a mathematical definition of the full string action for the topological B-model on a Calabi-Yau manifold of any dimension and proved that it satisfies the full BV master equation. They refer to this as BCOV theory.

  • •

    In 2015/16, the same authors determined the B-model open-closed quantum coupling in the large N limit and from this perspective they worked out a version of twisted supergravity. Most of the lectures will explain the speaker’s work with Costello.

We will now fix some conventions. We will work with the category of ℤ-graded vector spaces

For any a∈Vm, we will denote its degree by deg⁡(a)=|a|=m. We will denote degree shifts by

or namely, shifting to the left. Finally, we will discuss some conventions on tensors. We will denote the symmetric power by

and the exterior powers by

These are sometimes referred to as the graded symmetric product and the graded skew-symmetric product. Finally, we will consider the symmetric algebra (or polynomial algebra)

and its completion (a power series ring)

For a graded algebra A, we will always use [−,−] to refer to the graded commutator, which is defined by

Finally, if X is a smooth manifold X, we will write

to denote smooth differential forms on X and denote the de Rham differential by d. If X is a complex manifold, we can decompose

and decompose the differential as d=∂¯+∂.

Exercise 1.1.

Show that

2 Calabi-Yau geometry

In this section, we will build some foundations for Calabi-Yau geometry. Let X be a complex manifold of dimension dimℂ=d. We will denote the holomorphic tangent bundle by TX.

We will denote polyvector fields by

and the space of all polyvector fields by

Locally, we can write μ∈𝖯𝖵i,j⁢(X) can be written as

where I={k1,…,ki} and J={s1,…,sj} are multi-indices. Here, the index notation looks like

These look like differential forms, but are in fact subtly different. All μJ¯I are smooth and skew-symmetric for indices in both I and in J.

There is a natural (wedge) product

which makes 𝖯𝖵⁡(X) a bi-graded algebra. We will define the total degree by |μ|=i+j for μ∈𝖯𝖵i,j⁢(X). This makes (𝖯𝖵⁡(X),∧) a graded-commutative algebra, namely that α∧β=(−1)|α|⁢|β|⁢β∧α.

We can also extend the ∂¯-operator to polyvector fields, which has degree 1 and goes as

This is in fact a derivation which satisfies the graded Leibniz rule

There are many definitions for Calabi-Yau manifolds, so we will choose one.

Definition 2.1.

A Calabi-Yau manifold is a pair (X,ΩX) where X is a complex manifold and ΩX is a nowhere-vanishing holomorphic volume form on X.

In local holomorphic coordinates {z1,…,zd} on X, we can write

where φ⁡(z) is a nowhere-vanishing holomorphic function. The Calabi-Yau form induces a linear isomorphism

We now consider the divergence operator

with respect to the volume form ΩX, which is defined using the commutative diagram

Note that this depends on the choice of volume form. Unfortunately, we will abuse notation and write ∂ for ∂Ω and whether we mean the divergence operator or the holomorphic de Rham operator will be clear from context. Unfortunately, the divergence operator is not a derivation. The failure is described by the tensor

This bracket is known in geometry and is related to the Schouten-Nijenhuis bracket [−,−]SN by the sign

In mathematics, this is sometimes called the Bogomolov-Tian-Todorov lemma and in physics it is related to the BV algebra. In fact, the difference between the Schouten-Nijenhuis bracket and the BV bracket is related to shifting the degree by 1. The BV bracket has degree given by

It is in fact intrinsically defined independent of the choice of volume form (or divergence operator).

Remark 2.2.

The lifting of {−,−} to the volume form ΩX or the divergence operator ∂ is related to BV quantization.

Exercise 2.3.

Show that the BV bracket satisfies graded symmetry

the Leibniz relation

and the graded Jacobi identity

Remark 2.4.

The algebra (𝖯𝖵,∂¯,∂,{−,−}) is a fundamental example of a differential graded Gerstenhaber BV algebra and exists in a large class of examples.

We can now define a trace map

This trace is only nontrivial for 𝖯𝖵d,d and will be used later to discuss higher residue pairings.

Exercise 2.5.

Show that the trace map is compatible with the the two differentials via the identities

and

We will now let X be a compact Kähler manifold. The deformation space of complex structures on X is locally given by

Here, the final condition is a gauge-fixing condition and ∂¯μ+12⁢{μ,μ}=0 is known as the Maurer-Cartan equation.

Theorem 2.6 (Bogomolov-Tian-Todorov).

Let X be a compact Kähler Calabi-Yau manifold. Then the deformation space ℳ𝖼𝗑 is smooth and has dimension

In general, we can consider the extended moduli space ℳ by solving

modulo gauge for all types of polyvector fields. We can view the original moduli space as a submanifold of the extended one.

Theorem 2.7 (Barannikov-Kontsevich).

The extended moduli space ℳ is smooth.

Proof.

Consider the divergence-free polyvector fields ker⁢∂⊂𝖯𝖵. Given α,β∈ker⁢∂, we consider their BV bracket

which is ∂-exact. Therefore,

This fact will be crucial.

The complex (𝖯𝖵,∂¯,{−,−}) is actually a (shifted) dg Lie algbera, and we consider the inclusion

This is actually a morphism of dg Lie algebras and is in fact a quasi-isomorphism. If we allow ℍ⊂𝖯𝖵⁡(X) be the harmonic elements, then extending Hodge theory gives us

by the Kähler condition. Now define the projection

By the ∂∂¯-lemma, we see that ∂¯(ker⁢∂)⊂Im⁢∂, so we can consider the map

which is again a morphism of dg Lie algebras. Now we have the zig-zag

where both maps are quasi-isomorphisms. By some abstract nonsense, solving the Maurer-Cartan up to gauge is invariant under quasi-isomorphisms, and solving the Maurer-Cartan equation in (ℍ,0,0) is trivial. Therefore we have ℳ≃ℍ. □

3 Period map

Before we move to the B-model, we need to discuss the period map, which will allow us to write down action functionals. Let X,ΩX be a compact Kähler Calabi-Yau manifold. Given μ∈𝖯𝖵1,1⁢(X) satisfying the Maurer-Cartan equation

we will use [Xμ]∈ℳ𝖼𝗑 to denote the complex manifold specified by the deformation μ. Assume that {zi} are local holomorphic coordinates on X. We can write

Then we can define new holomorphic 1-forms by

and these form a local basis of (1,0)-forms on Xμ.

Let ℋd denote the complex vector bundle over ℳ𝖼𝗑 whose fiber at [Xμ] is the de Rham cohomology Hd⁢(Xμ). This bundle admits a flat connection, called the Gauss-Manin connection, and denoted by ∇𝖦𝖬, which identifies the nearby fibers to the central one as

There is some global issue with monodromy, but locally the identification is canonical. We now define the holomorphic subbundles (the Hodge filtration) ℱp⁢ℋd⊂ℋd with fibers

In particular, there is the line bundle L=ℱd⁢ℋd which picks up the holomorphic volume form on Xμ. In physics, this is known as the vacuum line bundle. Later, we will see that invariants of the B-model are actually sections of powers of this line bundle.

The period map is locally given by the map

Here, we need to use the Gauss-Manin connection to identify the cohomology of Xμ with that of X. The period map is actually a local embedding (the local Torelli theorem). Sometimes it is helpful to remember the actual volume form, and if we consider the ℂ×-bundle

over ℳ𝖼𝗑, then a point here represents a pair (X,ΩX), so we have defined the local moduli of deformations of a pair (X,ΩX). This allows us to lift the period map to an embedding

Our goal is to understand the geometry of the period map, which will allow us to understand the action functional of the B-model. We will begin with the case of Calabi-Yau threefolds, which is classical, and then move to a more modern perspective. Let (X,ΩX) be a Calabi-Yau threefold. Then H3⁢(X,ℂ) has a symplectic pairing given by

Considering the Hodge filtration, the subspace

is a Lagrangian subspace. Over the moduli space ℳ𝖼𝗑 ω defines a fiberwise symplectic pairing on ℋ3 which is flat with respect to the Gauss-Manin connection. Then we can lift the Hodge filtration to a subbundle

which gives a holomorphic family of Lagrangian subspaces.

Proposition 3.1.

The embedding ℳ^𝖼𝗑↪H3 defines a Lagrangian submanifold.

Proof.

Let Ωs be a holomorphic family of Calabi-Yau forms parameterized by s. By Griffiths transversality, we see that

Every fiber is a linear Lagrangian subspace, so we are done. □

From this data, we can cook up B-model invariants as follows. Choose a Lagrangian splitting ℒ⊂H3⁢(X,ℂ) be a transversal linear Lagrangian subspace such that

is an isotropic splitting. This allows us to identify

where we use F2 to refer to F2⁢H3. Then ℳ^𝖼𝗑 is a Lagrangian submanifold as in Figure 4. Therefore, there is some function F^0∈𝒪⁡(F2) such that ℳ^𝖼𝗑 is the graph of d⁢F^0.

F2H3≃T∗⁢F2ℳ^cx
Figure 4. The moduli space ℳ^cx as a Lagrangian submanifold in H3≃T∗⁢F2.
Example 3.2.

Choose ℒ=H1,2⊕H0,3 (the complex conjugate splitting). Physicists would now write down some special geometry and prepotentials. We parametrize ℳ^𝖼𝗑 by

Because ℳ𝖼𝗑=ℳ^𝖼𝗑/ℂ×, we see that F^0 has homogeneous degree 2. The local coordinates on ℳ𝖼𝗑 are given by τi=tit0. Therefore, we can write

Geometrically, F^0 actually defines a local section of L⊗−2, where L is the vacuum line bundle. We can now reparaterize the extended moduli space by

The factor inside the parentheses is called the primitive period map. We now obtain a particular choice of holomorphic volume form on the moduli space given by

We are now finally able to calculate the Yukawa coupling for the moduli space by

In the physics literature, ℱ0⁢(τ) is called the prepotential.

Exercise 3.3.

Prove the formula for the Yukawa coupling in the previous example.

Before we proceed, we need to describe the deformation of the pair (X,ΩX). We know that deformations of X are given by μ∈𝖯𝖵1,1⁢(X) satisfying the Maurer-Cartan equation

In the complex structure Xμ specified by μ, the elements

form a new basis of (1,0)-forms on Xμ.

Lemma 3.4.

We have

for any differential forms α,β.

The lemma now implies that

is of type (d,0) with respect to Xμ. Applying this globally, the form

is of type (d,0) in Xμ. Therefore, there exists a smooth function ρ such that

is a holomorphic volume form on Xμ. Because the type is (d,0), then the form is holomorphic if and only if it is closed, or in other words that d⁡(eρ⁢eμ⁢⌟⁢ΩX)=0. This gives us the two equations

which govern deformations of the pair (X,ΩX). These equations were studied by Barannikov in his work on variations of semi-infinite Hodge structures.

4 Extended period map

Introduce a formal parameter t of degree 2, which will play the role of gravitational descendents and consider the combined differential

We can rewrite our two equations as the single equation

where we have defined

Introduce a new (shifted) dg Lie algebra

Remark 4.1.

The data of (𝖯𝖵⁡(X)⁢⟦t⟧,Q) can be viewed as a homological replacement of volume-preserving (divergence-free) vector fields (ker⁢∂,∂¯). We may now write the complex

which behaves badly, so we replace it with (𝖯𝖵⁢⟦t⟧,Q).

Remark 4.2.

The variable t plays the role of gravitational descendants (namely ψ-classes on the A-side) in the B-model.

Remark 4.3.

If (𝖯𝖵,∂¯,{−.−}) gives extended moduli for X, then (𝖯𝖵⁡⟦t⟧,Q,{−,−}) gives extended moduli for the Calabi-Yau geometry.

Remark 4.4.

The variable t can be viewed as an equivariant parameter for rotation on the worldsheet in the worldsheet B-model.

Question 4.5.

Can we express the above equations as the variational critical locus of some action functional?

One reason we would be interested in this question is because this is the usual approach to classical physics, and then we can quantize these using path integrals. This is unfortunately highly nontrivial, and on a Calabi-Yau threefold, BCOV work with some μ∈ker⁢∂. They found that the Maurer-Cartan equation

is equivalent to δ⁢𝖪𝖲δ⁢μ=0 for the Kodaira-Spencer action 𝖪𝖲.

Warning 4.6.

This approach unfortunately violates locality for various reasons:

  • •

    ker⁢∂ is not a module over C∞⁢(X);

  • •

    One of the terms in 𝖪𝖲 is nonlocal.

Consider the Calabi-Yau geometry and let μ∈𝖯𝖵⁡(X)⁢⟦t⟧. We can ask for an action functional whose critical points are solutions of

Unfortunately, the speaker was unable to find this, but there is some magic involved. If we perform an L∞ transformation, this equation becomes

and for this equation there is a BCOV action which was found by Costello and the speaker. The L∞ transformation is produced directly from the period map.

Definition 4.7.

Define

where the shift makes μ+t⁢ρ have degree 0. Then define

This makes 𝖲⁡(X)=𝖲+⁢(X)⊕𝖲−⁢(X), and once we define a symplectic structure it will be a splitting by isotropic subspaces. However, Q does not preserve the splitting, but this will lead to the classical master equation.

We define a t-sesquilinear pairing on 𝖲⁡(X) (inspired by the higher residue pairing of Kyoji Saito)

by the formula

Observe that

If we calculate the degrees, then ω is only nontrivial when

so if we take the original shift by 2 into account, we have proved the following result.

Proposition 4.8.

The space (𝖲⁡(X),ω) has a (6−2⁢d)-shifted symplectic structure.

Remark 4.9.

For a Calabi-Yau threefold X, we have the splitting

and the we will be able to identify this with

via contraction with ΩX, which we will call Γ. The higher residue pairing will be identified with the usual pairing on differential forms. We should note however that this only works for threefolds.

Proposition 4.10.

The operator Q is (graded) skew-symmetric with respect to ω.

Proof.

The proof is left as an exercise. The key point is to study the operators ∂¯ and ∂. □

We now consider the (cochain) period map

Remark 4.11.

This 𝒫 is understood as a map on formal schemes. For any Artinian ring (R,𝔪), we send

This truncates at finite order, so the map is well-defined.

We compute the tangent map at μ=0, and we see

so we see that

is simply the inclusion with a sign of −1. See Figure 5 for a visualization. This implies that 𝒫 is a formal embedding.

𝖲+⁢(X)μ𝒫𝖲+𝖲⁡(X)
Figure 5. The cochain period map

This map is actually an L∞-morphism

One way to see this is to check that the Maurer-Cartan equation is preserved. We can check that

and therefore we see that μ satisfies the Maurer-Cartan equation if and only if 𝒫⁡(μ) does. Therefore, if we view Q as a vector field on 𝖲⁡(X), then Q is tangent to Im⁡(𝒫) and is thus the pushforward of a vector field on 𝖲+⁢(X) which defines the L∞ structure.

The image Im⁡(𝒫)⊂𝖲⁡(X) of the period map is a formal Lagrangian submanifold. If we calculate the variation of the period map, we see that

Proposition 4.12.

For all α,β∈𝖲+⁢(X), we have

Proof.

The proof is left as an exercise. □

5 Classical BCOV action

Some references for this include the surveys Some classical/quantum aspects of Calabi-Yau moduli and Renormalization method and mirror symmetry by the lecturer in addition to his work with Costello.

We will begin with a toy model to describe the geometry of gravitational descendants. Suppose the case when X is a point. In this case, 𝖲=ℂ⁡((t)), 𝖲+=ℂ⁢⟦t⟧, and 𝖲−=t−1⁢ℂ⁢[t−1]. The symplectic pairing is given by

The period map now looks like Figure 6. The image of the period map is a formal Lagrangian submanifold.

𝖲+⁢(X)f⁡(t)𝒫𝖲+𝖲=ℂ⁢⟦t⟧t⁡(1−ef/t)
Figure 6. The cochain period map for X=pt

Under the identification 𝖲=T∗⁢𝖲+, we write the image of the period map as the graph of dI0 for some I0∈𝒪⁡(𝖲+). We will parameterize τ∈𝖲+ as

Then we can write I0=I0⁢(τk), and our goal is to compute this. We see that

Denoting this by φ⁡(τk), we will try to find its geometry. This actually becomes a combinatorial problem, but this actually becomes too complicated. Instead, it can be charcterized by various properties, which correspond to properties of the Lagrangian cone in Gromov-Witten theory.

The first property is the string equation. Let t^:𝖲→𝖲 be the operator which sends tk↦tk+1. Now define

which is essentially multiplying by t−1. We can view L−1 as a vector field on 𝖲, which is actually symplectic, namely satisfying the equation

Writing the symplectic form as

we see that

This implies that

Therefore, we will write the Hamiltonian as

The following result is one of the basic results in the Givental formalism.

Lemma 5.1.

The vector field L−1 is tangent to the object Im⁢𝒫−t, which is known as the dilaton shift.

Proof.

Let α=𝒫⁡(f)−t=−t⁢eft. We calculate the tangent space

We conclude by calculating L−1|α=L−1⁢(α)=−eft⊂Tα⁢(Im⁢𝒫−t). □

We can now write down a Hamilton-Jacobi equation, which says that h−1|Im⁢𝒫−t is constant. In this example, the constant will be zero, we see that

This is equivalent to the PDE

which is exactly the string equation from Gromov-Witten theory.

The second property we will consider is the dilaton equation. Consider the operator

This is also a symplectic operator satisfying

The Hamiltonian function is now given by

Note that all of our operators are linear, so the Hamiltonians are quadratic.

Lemma 5.2.

The vector field L0 is tangent to Im⁢𝒫−t.

Proof.

The proof is similar to the string equation, so it is left as an exercise. □

We can once again write the Hamilton-Jacobi equation, which says that

This implies that

which is exactly the dilaton equation from Gromov-Witten theory. We may continue this and write down the classical Virasoro constraints, but we will stop here.

There is a third equation, called the grading equation. Consider the degree operator

Here, we are remembering that there is actually a shift by 2 that we have been suppressing and that t has degree 2.

Lemma 5.3.

G is tangent to the Lagrangian Im⁡𝒫 (with no dilaton shift).

Unfortunately, this is not a Hamiltonian vector field (in particular the symplectic form has a nonzero degree). The lemma is equivalent to the statement that Im⁢𝒫 is preserved by the ℂ×-action taking tk↦λ2⁢k−2⁢tk. In particular, we see that I0⁢(τk) is homogeneous of degree 6 if we assign deg⁡τk=2−2⁢k. Combining this with the string and dilaton equations, it is easy to show the following.

Proposition 5.4.

For each k1,…,kn, we have

The general case follows easily from the point case, and in particular we can determine the classical BCOV action. Denote the image of the cochain period map (see Figure 5) by ℒX, which is a formal Lagrangian submanifold of 𝖲⁡(X)=𝖲+⁢(X)⊕𝖲−⁢(X)=T∗⁢𝖲+⁢(X). Writing ℒX as the graph of dI0X, we can determine I0X as a function on 𝖲+⁢(X)=𝖯𝖵⁡(X)⁢⟦t⟧⁢[2]. Denote

This extends 𝖯𝖵⁡(X)-linearly to a map

Theorem 5.5 (Costello-Li).

We have the formula

This is known as the classical BCOV action.

If we write μ=μ0+μ1⁢t+μ2⁢t2+⋯ for μk∈𝖯𝖵⁡(X), we see that

In the Kodaira-Spencer gravity of BCOV, the fields are μ0∈ker⁢∂⊂𝖯𝖵⁡(X), and their action is given by

This looks a bit unusual, because we can find the Maurer-Cartan equation, but the original theory is nonlocal, which poses problems for quantization (which is why we replace ker⁢∂ with the complex [𝖯𝖵→t⁢∂t⁢𝖯𝖵→t⁢∂t2⁢𝖯𝖵→⋯]).

6 Hodge structure and primitive forms

We can pass our picture from the cochan level to cohomology (or really the Maurer-Cartan functor). We will let

The extended period map

Because Q is compatible with ω, there is an induced symplectic form (shifted by 6−2⁢d) on H∙⁢(𝖲⁡(X),Q), which we also call ω. If we consider the subcomplex

then we see an isotropic subspace

However, 𝖲− is not preserved by Q, so we do not have a preferred choice of splitting at this level.

Definition 6.1.

An opposite filtration of H∙⁢(𝖲⁡(X),Q) is a choice of linear isotropic subspace ℒ⊂H∙⁢(𝖲⁡(X),Q) such that

  1. (1)

    There is an isotropic splitting H∙⁢(𝖲⁡(X),Q)=H∙⁢(𝖲+⁢(X),Q)⊕ℒ;

  2. (2)

    ℒ is preserved by t−1.

A choice of opposite filtration splits the map

by the natural isomorphism

If we write ℋ≔H∙⁢(𝖲⁡(X),Q) and ℋ+=H∙⁢(𝖲+⁢(X),Q), then ℋ=Bℒ⁢((t)) and ℋ+=Bℒ⁢⟦t⟧, where we choose ℒ=t−1⁢Bℒ⁢[t−1]. This allows us to write

Using this, the period map looks like Figure 7. We then see that 𝒫⁡(ℳX) is the graph of dℱ0X,ℒ, where ℱ0X,ℒ is a function on ℋ+=H∙⁢(𝖯𝖵,∂¯)⁢⟦t⟧, which is called the B-model state space (and ℱ0X,ℒ is the genus-zero B-model potential).

ℳX𝒫ℋ+ℋ≃T∗⁢ℋ+
Figure 7. The period map at the cohomology level

We now consider the geometry of the opposite filtration. Define the map

for μ∈𝖯𝖵i,j. This intertwines the operator Q with the standard de Rham differential d=∂¯+∂. Observe that

where ℱ𝓅⁢Ω⁢(X)=Ω≥p,∗⁢(X) is the Hodge filtration. Here, we can see that

where we identify p=d−i−k. In particular, we see that

gives us a sum of graded pieces of the Hodge filtration. In particular, we see that ℒ is equivalent to a splitting of the Hodge filtration. Some examples include the following:

  1. (1)

    If we let ℒ=ℱ¯p is the complex conjugate splitting (the harmonic splitting), we obtain the usual Hodge decomposition

    However, this splitting does not give the correct B-model invariants.

  2. (2)

    We can let ℒ be the monodromy splitting, which arises near the large complex structure limit. This is in fact the correct splitting in the B-model and gives the mirror to Gromov-Witten theory. If we move to a different point (maybe the Gepner point), we get a different monodromy splitting, which is mirror to FJRW theory.

Remark 6.2.

For a Calabi-Yau threefold, we have a symplectic subspace

and ΓΩ identifies this with the space

We will now introduce primitive forms (due to Kyoji Saito), or equivalently, J-functions (due to Givental). Fix a splitting ℒ and a subspace

with basis {φα}.

Definition 6.3 (Higher residue pairing).

Define the pairing

When we come to the Landau-Ginzburg B-model, this is literally Saito’s higher residue pairing. In addition, note that

If we reduce modulo t, this simply becomes the trace pairing on H∙⁢(X,Λ∗⁢T⁢X)=H∙⁢(𝖯𝖵,∂¯).

There are some peculiar properties here. First, the isotropic splitting

implies that

Together, these tell us that

sends Bℒ to constants. In our basis, we will write

Saito calls this choice a good basis. With our choice of splitting, write

for the corresponding projection. Under the choice of good basis Bℒ, we see that

simply eliminates the terms of negative degree in t.

Now consider the fiber product

Let’s write coordinates {τkα}k≥0 for ℋ and coordinates {τ0α,τ1α=0,…} for Bℒ. This identifies a family

solving the Maurer-Cartan equation

This will play the role of the primitive form in Saito’s theory. The solution also satisfies

Here, the notation [−]Q means taking the cohomology class with respect to Q. This can be rewritten as the equation

using the fact that the period map has image the graph of dℱ0ℒ. We can see that the right-hand side of this equation is the J-function J⁡(τ0α;t), which is the analogue of the J-function defined by Givental in Gromov-Witten theory. We can again rewrite this as

and the left-hand side here is the primitive form defined by Saito.

Proposition 6.4.

The quantity

is independent of t.

Proof.

By the expression of J, we see that

On the other hand, we can calculate

where we use the fact that this last expression lies in the tangent space to the Lagrangian cone. The conclusion follows by comparing our two calculations. □

We see now that {∂τ0αJ} defines a family of good bases parameterized by τ0α. Geometrically, we see that

At τ0α=0, we see that T0⁢ℳ∩t⁢ℒ=ℋ+∩t⁢ℒ=Bℒ is our original choice of good basis. Because we have the splitting

we see that the splitting

holds for nearby τ0α. The idea (due to Givental) is that the set {∂τ0αJ} forms a basis of Tτ0α⁢ℳ∩t⁢ℒ.

We can now compute the second derivatives ∂τ0α∂τ0βJ in two ways as we did with the higher residue pairing. On one hand, the expression of J tells us that

On the other hand, we calculate

In particular, the two calculations imply that

This now implies that there exists Aα⁢βγ⁢(τ0∙) such that

Passing to the primitive form, this is now equivalent to the statement that

Remark 6.5.

There is a perturbative algorithm for the Landau-Ginzburg B-model due to Li-Li-Saito, which allows us to compute the primitive form order by order using a method similar to Birkhoff factorization.

7 Classical master equation

Recall the period map as in Figure 5. Using the string and dilaton equations, we determined that Im⁢𝒫 was the graph of dI0X, where

Our goal now will be to study some structure relating to Q, which will allow us to pass to a finite-dimensional model. Q is odd (whereas the Virasoro constraints are even), so we will write down an odd version of the Hamilton-Jacobi equation, which will be the classical master equation.

Because Q preserves ω, we will write down the quadratic Hamiltonian hQ, which will satisfy hQ|Im⁢𝒫=0. The differential Q=∂¯+t⁢∂ satisfies Q2=0, and is therefore a cohomological vector field. Here are some properties of the operator Q:

  • •

    It preserves 𝖲+⁢(X);

  • •

    It does not preserve 𝖲−⁢(X).

To give an idea of the general picture and how to quantize the theory, we will consider a finite-dimensional toy model. Let

be a dg (even) symplectic vector space (and for simplicity, we will assume the degree is 0), so in particular we can consider one of

By definition, the differential Q is skew-symmetric with respect to ω. Assume that we have a splitting

where both V+ and V− are Lagrangian subspaces. We will require that Q preserves V+ but not necessarily V−. The splitting will allow us to make the identification V=T∗⁢V+ and consider the projection

We know that ω−1∈Λ2⁢V⊂(V+⊗V−)⊕(V−⊗V+). To write down the Hamiltonian function, consider

Denote the component in V+⊗V+ by

which plays the role of the BV kernel and measures the failure of Q to preserve V−.

Proposition 7.1.

P defines a degree 1 element of Sym2⁡(V+) and is Q-compatible, namely that

Proof.

Using the fact that Q is skew-symmetric with respect to ω, we see that

This implies that

is symmetric, and projecting to V+⊗V+ concludes the proof. □

This is actually a dual description of the Hamiltonian function hQ. Geometrically, we can consider

which exactly gives the quadratic Hamiltonian associated to Q. Using the fact that

we see that

We see that the Hamilton-Jacobi equation will have a linear term coming from the first two components and a quadratic term coming from the last component.

Definition 7.2.

Denote the differential operator

by contracting with P∈Sym2⁡(V+). More explicitly, we define

Also define the BV bracket

Proposition 7.3.

The data (𝒪⁡(V+)=Sym∙⁡(V+∗),Q,∂P,{−,−}P) defines a dg Gerstenhaber-Batalin-Vilkovisky (dGBV) algebra.

Proof.

Because (Q⊗1+1⊗Q)⁢P=0, we see that [Q,∂P]=0. The rest of the properties are straightforward to check. □

Now assume that we have a Lagrangian submanifold

which is the graph of dI0 for some function I0∈𝒪⁡(V+).

Proposition 7.4.

Q is tangent to ℒ if and only if

This is called the classical master equation.

Proof.

The odd Hamilton-Jacobi equation says that hQ|ℒ=0. The mixed part of hQ corresponds to Q and the purely momentum part corresponds to the BV bracket. □

Now denote by

the corresponding cohomological vector field. The classical master equation can now be rewritten as δ2=0. In fact, with the picture as in Figure 8, we see that

Because Q2=0, we see that δ2=0 as well.

V+V≃T∗⁢V+Qπ+
Figure 8. Q tangent to ℒ and the map π+.

We will now return to our original infinite-dimensional setting of the B-model on a Calabi-Yau manifold X. The structure is a bit more complicated but is still similar. Recall that

Also recall that

The problem comes from the trace pairing, which causes the Poisson tensor ω−1 to involve distributions as

where δΔ is a distributional element of 𝖯𝖵⁡(X)⊗𝖯𝖵⁡(X) and is the integral kernel for the identity operator

Note that this is fine classically but will presernt problems for quantizing the theory. We can still write down the BV kernel

because Q only fails to preserve 𝖲−⁢(X) on t−1⁢𝖯𝖵 via the t⁢∂ term. Given a local functional I on 𝖲+⁢(X)=𝖯𝖵⁡(X)⁢⟦t⟧ of the form I=I⁡(μk)=∫⋯, its variation can be computed as

We define the BV bracket for two local functionals I1 and I2 by

which produces another local functional. Copying the finite-dimensional case, we have the following theorem.

Theorem 7.5 (Costello-Li).

The BCOV action I0X=Tr⁡⟨eμ⟩0 satisfies the classical master equation

Remark 7.6.

The theorem is proved directly in the lecturer’s PhD thesis, where he finds that the classical master equation is equivalent to the genus-zero topological recursion relations.

8 Quantum master equation

Note that the map π+∘𝒫:𝖲+⁢(X)→𝖲+⁢(X) is a diffeomorphism. However, we gain the classical interaction I0X and the Maurer-Cartan equation with respect to δ becomes

where ℓ2,ℓ3,… emerge from the {I0X,−} term.

We will now return to our toy model (V,Q,ω) of a dg symplectic vector space. We also continue to assume the Poisson kernel ω−1 and the isotropic splitting V=V+⊕V−. Q preserves V+ but not V−, and we wrote down the tensor

This produced a dGBV algebra

Definition 8.1.

We define the (formal) Weyl algebra

where ℏ is the formal (quantum) parameter and the relation ∼ is generated by

Note that the classical limit

is the formal functions on V. We denote the annihilator Ann⁡(V+)⊂V∗ to be

Because the differential Q preserves V+, the induced differential Q:V∗→V∗ preserves Ann⁡(V+) as well.

Proposition 8.2.

The annihilator Ann⁡(V+)⊂V∗ is a subcomplex.

Definition 8.3.

The (formal) Fock space is defined by

Remark 8.4.

We treat Ann⁡(V+) as annihilation operators.

Because Q is compatible with ω, we see that (𝒲⁡(V),Q) defines a cochain complex with subcomplex 𝒲⁡(V)⁢Ann⁡(V+)⊂𝒲⁡(V). This induces a differential

The splitting V=V+⊕V−≃T∗⁢V+ induces a natural embedding V+∗⊂V∗ given by extension by zero on V−. Let

be the formal functions on V+. This induces

and realizes 𝒪^⁢(V+)⁢⟦ℏ⟧ as creation operators (which for physicists look like expressions of the form ⋯⁢a†⁢|0⟩). Under this isomorphism, there is a differential Q^ on 𝒪^⁢(V+)⁢⟦ℏ⟧, namely the unique one making the diagram

commute. However, this is not the same as the one induced from Q:V+→V+ (which extends to Q:O^⁢(V+)→𝒪^⁢(V+)).

Proposition 8.5.

The difference between Q^ and Q is precisely the failure of Q to preserve the splitting, namely

Proof.

This is standard quantization of quadratic Hamiltonians and is left as an exercise. □

Definition 8.6.

A formal function

satisfies the quantum BV master equation (QME) if

Equivalently, we can rewrite this as

Note that in the classical limit ℏ→0, we recover the classical master equation

We now discuss what is going on in a physical sense. Consider the state

Then the quantum master equation is equivalent to the statement that Q⁢|F⟩=0, which in physics is called the quantum gauge consistency. Passing to Q-cohomology (what physicists call physical states), we pass to

which usually an element of a finite-dimensional space. If we let ℋ=H∙⁢(V,Q) and ℋ+=H∙⁢(V+,Q), then we have a symplectic vector space (ℋ,ω) with ℋ+⊂ℋ as an isotropic subspace. If we choose a polarization ℋ=ℋ+⊕ℒ, the surjection 𝒲⁡(V)↠ℱ⁢ock⁡(V+) reduces to a surjection

and the target is once again isomorphic to 𝒪^⁢(ℋ+)⁢⟦ℏ⟧ because the polarization gives an embedding ℋ+∗⊂ℋ∗. Because Q⁢|F⟩=0, we obtain a state

The choice of splitting gives us a formal function

Unfortunately, this is extremely hard to compute.

Warning 8.7.

If we try to redo everything in the actual B-model, the “quantum master equation”

is ill-defined because P is singular. This will require choosing a Kähler metric, doing renormalization, and defining everything in a homotopical way.

9 Homotopic renormalization

In quantum theory, we will replace (V+,Q) by ℰ,Q, where ℰ is the space of fields. Usually, we will take

to be the global (smooth) sections of some graded vector bundle and

to be an elliptic operator (for example ∂¯ or ∂). This data (ℰ,Q) will be called an elliptic complex. The (shifted) symplectic space will be replaced by a (shifted) Poisson kernel P, which is a distributional section of Sym2⁡(ℰ) which is usually supported on the diagonal of X×X (here, note that P comes from the inverse of integration, which involves delta functions).

We will now follow the toy model until we reach an obstruction. The first modification we need to make is to force ℰ∗=Hom𝖼𝗈𝗇𝗍⁡(ℰ,ℂ) to be the continuous dual (these are essentially distributions on X). The next modification we need is to consider that ℰ∗ is a nuclear space, so when we write ℰ∗⊗ℰ∗, we mean the completed tensor product, which we think of as distributions on X×X. This allows us to define

We may now discuss functions on ℰ as

Our next goal is to define the BV operator, which in the toy model was

We would like to define

but this is not well-defined because we cannot pair two distributions, which in QFT is known as the ultraviolet problem and arises in any infinite-dimensional model.

We will resolve this problem by using Costello’s homotopic renormalization theory. We will change notation and replace P by K0∈𝖣𝗂𝗌𝗍⁡(Sym2⁡(ℰ)), which is the (shifted) Poisson kernel. This is Q-closed, namely that

The key observations are

  1. (1)

    K0 is a Q-closed distribution;

  2. (2)

    Elliptic regularity means that

    and therefore we can write

    where Kr is a smooth kernel and Pr is a distribution known as the parametrix.

We will call Kr the renormalized BV kernel with respect to the parametrix Pr. Now, we can define

by contracting with the smooth kernel Kr. This is now well-defined and gives us a renormalized BV operator

(Note that working with formal power series means we are working perturbatively, and the non-perturbative theory is much more complicated). Because Kr is Q-closed, we see that

Definition 9.1.

The data (𝒪⁡(ℰ),Q,∂Kr) defines a dGBV algebra called the renormalized dGBV with respect to the parametrix Pr.

Question 9.2.

How do we resolve the possible dependence on the choice of parametrix Pr?

Let Pr1 and Pr2 be two parametrices such that

Then we see that Pr1r2≔Pr2−Pr1 is smooth inside Sym2⁡(ℰ).

Definition 9.3.

Define the operator

by contracting with the smooth kernel Pr1r2.

Proposition 9.4.

The following diagram commutes:

Proof.

Note that Kr1−Kr2=Q⁡(Pr1r2). This implies that

This implies that

Definition 9.5.

Define

Note that in physics this is called interaction theory. We define the homotopic renormalization group flow

by the formula

The real content is that

This tells us that

is a sum over connected Feynman diagrams, where contributions are as in Figure 9. The physicists would call I the vertex and Pr1r2 the propagator.

IIPr1r2
Figure 9. Feynman diagram representation of the homotopic renormalization group flow.
Definition 9.6.

A solution of the effective quantum master equation is an assignment I⁡[r]∈𝒪+⁢(ℰ)⁢⟦ℏ⟧ for each parametrix Pr satisfying

  1. (1)

    The renormalized quantum master equation (RQME)

  2. (2)

    For any two parametrices Pr1 and Pr2, we have

Because of the chain homotopy

we see that the two conditions are compatible. The picture is as in Figure 10, where the two points r1 and r2 represent two different choices of parametrix and the arrow represents the homotopic renormalization group flow. The central dot represents the original singular kernel K0, which is homotopic to both Kr1 and Kr2 via the parametrices Pr1 and Pr2.

××r1r2HRG0
Figure 10. Homotopic renormalization group flow.

10 Quantum BCOV theory

We are finally ready to discuss the quantum B-model. Recall that our space of fields is

with Q=∂¯+t⁢∂, which had the (shifted) Poisson kernel

Because this kernel is singular, we will need to run the homotopic renormalization group flow. First, choose a Kähler metric g on X. This gives us

which is the adjoint of ∂¯. We can then write down the Laplacian and then the heat kernel

for any r>0, which represents

for any α∈𝖯𝖵⁡(X). Now let

(this limit holds because the heat kernel converges to the delta function). Also, note that because of the grading,

is the usual Laplacian. Now the parametrix is given by

Proposition 10.1.

We have K0=Kr+Q⁡(Pr).

Proof.

This is the integral kernel representation of the operator equation

Because Q=∂¯+t⁢∂ and the Kähler condition implies that

we see that

Definition 10.2.

Given 0<ε<L, define the effective propagator

This is the homotopy from Kε to KL.

Definition 10.3 (Costello-Li).

A perturbative quantization of BCOV theory on X is a family of functionals (or distributions)

for each L>0 such that

  1. (1)

    They respect the homotopic renormalization group flow, namely

  2. (2)

    The renormalized quantum master equation

    where ΔL=∂KL, holds;

  3. (3)

    The locality axiom that F⁡[L] has a small L asymptotic expansion by local functionals as L→0;

  4. (4)

    The classical limit

    recovers the classical BCOV action;

  5. (5)

    The degree axiom and Hodge weight axiom hold (this is analogous to Gromov-Witten theory) such that FgL has Hodge weight (3−dimX)⁢(g−1), where tm⁢𝖯𝖵k,∙ has Hodge weight k+m−1.

Remark 10.4.

This quantization is not necessarily unique.

Given a quantization {F⁡[L]}L, we consider the IR limit K→+∞. In this limit, we have

and therefore the quantum master equation becomes

This implies that Q⁢F⁢[∞]=0, so it represents some cohomology class

A choice of Kähler metric g gives a splitting of the Hodge filtration, which gives an isomorphism

This now gives quantum B-model invariants Fg,n,Xℒ, which by the mirror conjecture should be related to the Gromov-Witten invariants ⟨−⟩g,n,Xˇ𝖦𝖶 on the mirror Xˇ.

Calculations are very hard, but we can do them using UV finiteness and locality. In practice, constructing a quantization is very complicated, but the program usually goes as follows.

  1. (1)

    Start with a local functional I∈𝒪𝗅𝗈𝖼†⁢(ℰ)⁢⟦ℏ⟧ (the classical interaction).

  2. (2)

    Find an ε-dependent local functional

    such that the limit

    exists. If we can find the counterterms, the family {I⁡[L]} automatically satisfies the homotopic RG flow. However, it does not necessarily satisfy the quantum master equation.

  3. (3)

    Find further corrections to solve the quantum master equation. This step may be obstructed, and the obstruction is known in physics as the gauge anomaly.

Definition 10.5.

We say the theory is UV finite if the limit

exists for the class of local functionals I in question without the need for counterterms.

Now assume that the theory is UV finite (all chiral theories like BCOV theory, holomorphic Chern-Simons, etc are in fact UV finite). Unfortunately, the equation

is ill-defined at L=0, so we regularize to an equation of the form

If the theory is UV finite, we can take the L→0 limit to obtain something like

where the ℓk can be viewed as higher brackets for local functionals and come from resolving the ill-defined term ℏ⁢Δ⁢I. We will refer to this as the renormalized local quantum master equation. Unfortunately, what the ℓk are is not known in general.

Exercise 10.6.

Do this for holomorphic Chern-Simons theory.

There is one case where the higher brackets are known explicitly.

Theorem 10.7 (Li).

2D chiral theories are UV finite and the renormalized local quantum master equation is equivalent to the Maurer-Cartan equation in chiral vertex operator algebras.

The renormalization behaves as in Figure 11, and a reference is the paper Vertex algebras and quantum master equation by the lecturer. From this, we can obtain an explicit canonical solution for the quantum B-model on elliptic curves and establish the quantum mirror conjecture.

{I,I}renormalized
Figure 11. Renormalization of 2D chiral theories
Theorem 10.8 (Wang-Yan).

Chiral theories of any dimension are UV finite (which implies that BCOV theory and holomorphic Chern-Simons theory are UV finite).

The result of Wang-Yan implies that quantum BCOV theory has a local expression.

Remark 10.9.

Gui calculated the higher chiral operator product expansions.

11 Landau-Ginzburg B-model

There are many ways to describe the Landau-Ginzburg B-model (for example via the category of matrix factorizations), but we will take the approach of string field theory. We will consider the Landau-Ginzburg B-model associated to the data (X,Ω,f), where

  1. (1)

    X is a non-compact complex manifold;

  2. (2)

    Ω is a holomorphic volume form on X;

  3. (3)

    f is a holomorphic function1 (called the superpotential) on X with compact critical locus (which we call Crit⁡(f)).

We will consider polyvector fields 𝖯𝖵⁡(X)=Ω0,∙⁢(X,Λ∙⁢TX) and the compactly supported polyvector fields 𝖯𝖵c⁢(X)⊂𝖯𝖵⁡(X). We will twist the Dolbeault operator and write

We will still refer to the divergence operator with respect to Ω as ∂. Then we can consider the dGBV algebras

Lemma 11.1.

The inclusion (𝖯𝖵c⁢(X),∂¯f)⊂(𝖯𝖵⁡(X),∂¯f) is a quasi-isomorphism.

Proof.

The operation {f,−} becomes the Koszul complex, which is exact outside of (a small neighborhood of) the critical locus Crit⁡(f). □

To apply the Barannikov-Kontsevich construction to obtain a Frobenius manifold on the cohomology H∙⁢(𝖯𝖵,∂¯f) , a number of things are required:

  1. (1)

    Hodge-to-de Rham degeneration at the E1 page for the t-adic filtration on 𝖯𝖵⁡(X)⁢⟦t⟧,∂¯f+t⁢∂ (or the compactly supported version), where unfortunately both are bad. If this exists, then this is smooth formal in the sense of Barannikov-Kontsevich and therefore has a univeral period map (which would give us a J-function allow us to write down the Frobenius structure);

  2. (2)

    A trace pairing Tr⁡(−,−) on polyvector fields which is compatible with ∂¯f and ∂, which is required to write down a formula like

    which of course works better for compactly supported polyvector fields;

  3. (3)

    A compatible splitting

    which is compatible with the trace pairing (and the higher residue pairing). This is usually referred to as a good basis.

Remark 11.2.

When f has isolated critical points, the cohomology

and the Hodge-to-de Rham degeneration holds automatically.

To define the trace pairing, it is of course better to use 𝖯𝖵c⁢(X).

Definition 11.3.

Define the trace pairing

Proposition 11.4.

We have the formulae

Corollary 11.5.

The trace pairing defines a pairing on cohomology given by

and the higher residue pairing

given by the formula

Remark 11.6.

Assume that f:ℂn→ℂ has only an isolated critical point at the origin. Let Ω=dnz be the standard volume form. Then we have

Then the trace pairing

is precisely the residue pairing, where we see that

A reference is the paper by Changzheng Li, the lecturer, and Kyoji Saito. In particular, if we write

then K0 is the residue as just described, and all of the K>0 are called higher residues. Here, we should consider holomorphic functions as elements of H∙⁢(𝖯𝖵⁡(X)⁢⟦t⟧,Qf).

Regarding the Hodge-to-de Rham degeneration issue, we can look at the various models:

  • •

    𝖯𝖵⁡(X) behaves well here, but is bad for the trace pairing;

  • •

    𝖯𝖵c⁢(X) behaves well for the trace pairing, but is too small for the Hodge decomposition (for example harmonic functions are outside of this);

  • •

    If we consider L2 polyvector fields, this works well for both the trace pairing and the Hodge decomposition, but is “bad” for non-linear structures (like products).

To solve these problems, we will introduce yet another model, which was introduced by the lecturer and Wen and are called adapted polyvector fields. It will sit as

and will have integration, the Hodge decomposition, and non-linear structures (products) all be well-behaved. It is given by

Definition 11.7.

The data (X,g,Ω) (here g is a complete Kähler metric) is called a bounded Calabi-Yau geometry if (X,g) is a bounded geometry (namely a complete Riemannian manifold, all derivatives of curvature are bounded, and positive injectivity radius), and all covariant derivatives of Ω and Ω−1 are bounded.

Definition 11.8.

Let (X,g) be a bounded geometry. A holomorphic function f:X→ℂ is strongly elliptic if for all ε>0 and k≥2, then ε⁢|∇f|k−|∇kf|→+∞ as z→∞ approaches the boundary.

Example 11.9.

Here are some examples which fall into this setup:

  1. (1)

    Let (X,ΩX)=(ℂn⁢dnz) and g be the standard flat metric. Next, let f be a non-degenerate quasi-homogeneous polynomial. This is mirror to FJRW theory and the full mirror theorem holds by the work of Li-Li-Saito-Shen and He-Li-Shen-Webb.

  2. (2)

    Let (X,ΩX)=((ℂ×)n,dz1z1∧⋯∧dznzn), g be the standard metric, and f be a convenient non-degenerate Laurent polynomial (here the origin lies in the interior of the Newton polytope). This is mirror to toric varieties;

  3. (3)

    Let X→ℂn/G be a crepant resolution for G⊂SU⁡(n) is a finite subgroup and g be an ALE metric. The superpotential is given by f:ℂn→ℂ which is G-invariant. This is the orbifold Landau-Ginzburg B-model. Note here, that the superpotential lifts as

Theorem 11.10 (Li-Wen).

Let (X,g,Ω) be a bounded Calabi-Yau geometry and f:X→ℂ be holomorphic and strongly elliptic with compact critical locus.

  1. (1)

    The data (𝖯𝖵f,∞⁢(X),∂¯f,∂) forms a dGBV algebra with the trace pairing and Hodge-to-de Rham degeneration holds;

  2. (2)

    The inclusions (𝖯𝖵c⁢(X),∂¯f)⊂(𝖯𝖵f,∞⁢(X),∂¯f)⊂(𝖯𝖵⁡(X),∂¯f) are all quasi-isomorphisms. In particular, Hodge-to-de Rham degeneration holds for (𝖯𝖵⁡(X),∂¯f,∂) as well.

  3. (3)

    (Poincaré duality) The trace map induces

    this is non-degenerate, and Kf generalizes Saito’s higher residue pairing.

The upshot is that a Frobenius manifold structure can be constructed on the cohomology H∙⁢(𝖯𝖵⁡(X),∂¯f).

Remark 11.11.

There is an algebraic approach based on positive characteristic methods, including the work of Sabbah, Ogus-Vologodsky, Esnault-Sabbah-Yu, and Katzarkov-Kontsevich-Pantev.

12 Open-closed B-model

So far, we have discussed the B-model closed string field. This leads to a generalized version of BCOV theory, which was developed by Costello and the lecturer. We will now discuss the open-closed B-model. As a reminder, Witten in 1992 gave two models of the open string field theory.

  • •

    The A-model open string field is Chern-Simons theory with instanton corrections;

  • •

    The B-model open string field is holomorphic Chern-Simons theory.

Let X be a Calabi-Yau threefold and E be a holomorphic vector bundle on X. The space of fields will be

and the action (in the BV formalism) is given by

If we decompose

where Ai∈Ω0,i⁢(X,End⁡(E)), then A0 is the ghost, A1 is the gauge (physical) field, A2 is the anti-field, and A3 is the anti-ghost. For the physical field A1, the equation of motion is

which says that if we set ∇¯=∂¯+A1, then ∇¯2=0, meaning ∇¯ defines a new holomorphic vector bundle structure on E. This is what is meant when it is said that holomorphic Chern-Simons theory describes the moduli of holomorphic vector bundles.

The space ℰ is (−1)-shifted symplectic with the symplectic form

The shifted Poisson kernel is given by

and defines a BV bracket {I1,I2}o for local functionals I1,I2. Holomorphic Chern-Simons theory satisfies the classical master equation

If we write δ𝖧𝖢𝖲={𝖧𝖢𝖲,−}o (the BRST transformation), then we obtain

and this implies that δ𝖧𝖢𝖲2=0. Equivalently, this says that Ω0,∙⁢(X,End⁡(E)) is a dg Lie algebra.

A bit more universally, embed E↪L⊕N as a subbundle for some sufficiently ample line bundle L. This allows us to represent E by some idempotent matrix

by choosing a smooth splitting. Using this, we write E=P⁡(L⊕N), and then we can embed

Here, N must be very large, so we will consider the behavior as N→∞. We will focus on

and look for functionals on ℰN which have a stable expression when N→∞. We will call these admissible functionals.

Example 12.1.

The functional

is admissible.

We will now study the large N limit from the perspective of large N duality. For A∈ℰN, we can write

Unlike ordinary Chern-Simons theory (which only allows us to tune the coupling constant), holomorphic Chern-Simons has many degrees of freedom to deform. One deformation is to deform the Calabi-Yau structure, and in fact this gives all deformations. Consider a first order deformation

where J is an admissible local functional. The first-order classical master equation says that

J is the local dual of Lie algebra homology

where the differential is ∂¯ and the Lie bracket is given by [−,−]. This looks hard to compute, but in fact there is a way to compute it.

Theorem 12.2 (Loday-Quillen-Tsygan).

The large N limit of the Lie algebra homology is given by

Here, 𝖧𝖢 is called cyclic homology.

Applying this to our model, we obtain

by the Hochschild-Kostant-Rosenberg theorem (use 𝖧𝖢∙⁢(R)∼(C∙⁢(R)⁢[t−1],b+t⁢B)). By the trace pairing, this is local dual to

Therefore, the BCOV field (topological gravity) is the single trace operator at N→+∞ (this is also related to the AdS/CFT correspondence) in the gauge/gravity duality.

Example 12.3.

Suppose μ∈𝖯𝖵k,∙⁢(X). This gives a first-order deformation

which is in fact another way to express the HKR theorem.

Example 12.4.

If we include descendents to make μ∈tm⁢𝖯𝖵k,∙⁢(X), then (see Figure 12)

A×μ
Figure 12. μ at the bulk and A at the boundary.

A general formula for higher order deformations at the disk level is given by Calaque-Willwacher’s cyclic extension of Kontsevich formality (see Figure 13). This is called a Poisson σ-model with formulae given by

AAA×μ×μ××
Figure 13. Higher order deformation

We will now discuss the open-closed master equation. We are interested in distributions

where g is the genus and h is the number of boundary components.

××××AA
Figure 14. An example of Σ2,2.

Zwiebach’s open-closed master equation requires {Ig,h} to satisfy equations based on δ⁡(Σg,h), which are various degenerations.

Example 12.5.

At the disk level with g=0 and h=1, the boundary is in Figure 15 and the corresponding equation is

This is in fact Kontsevich formality.

××××∂∼××××+××××
Figure 15. Boundary degeneration at the disk level in the open-closed master equation.
Example 12.6.

Now suppose g=2 and h=0. Then the boundary is given in Figure 16 and the corresponding equation is

××××××++×⁣××
Figure 16. Boundary degeneration of a genus-two closed surface.
Example 12.7.

In the case of an annulus with g=0 and h=2, the boundary is given in Figure 17 and the corresponding equation is

The Δo⁢I0,1 term is the one-loop gauge anomaly, and is cancelled out by the tree-level closed string sector via the Green-Schwarz mechanism.

+++
Figure 17. Four boundary strata in the degeneration of the annulus.
Theorem 12.8 (Costello-Li).

On X=ℂn, start with:

  • •

    Holomorphic Chern-Simons theory on Ω0,∙⁢(X)⊗𝔤⁢𝔩N|N. This is necessary because Tr𝔤⁢𝔩N⁡(○)=N≠0, but Tr𝔤⁢𝔩N|N⁡(○)=0.

  • •

    A first order deformation.

There exists a canonical quantization {Ig,h} for all g,h which satisfies the open-closed master equation.

Remark 12.9.

These quantizations glue to any local Calabi-Yau with a non-trivial ℂ×-action.

Remark 12.10.

The open-closed B-model which is BCOV coupled to holomorphic Chern-Simons theory is UV finite on any Calabi-Yau manifold (assumed to be Kähler).