The Period map for quantum cohomology of $\mathbb{P}^2$

TL;DR

This paper explicitly inverts the period map associated with the quantum cohomology of , expressing the inverse via Eisenstein series and quasi-modular forms, revealing deep connections between quantum cohomology and modular forms.

Contribution

It provides an explicit inversion of the period map for quantum cohomology of , linking it to Eisenstein series and quasi-modular forms, and extends to the big quantum cohomology case.

Findings

Explicit inverse for small quantum cohomology using Eisenstein series.

Perturbative inverse for big quantum cohomology as a Taylor series.

Connection established between quantum cohomology and modular forms.

Abstract

We invert the period map defined by the second structure connection of quantum cohomology of $\mathbb{P}^2$. For small quantum cohomology the inverse is given explicitly in terms of the Eisenstein series $E_4$ and $E_6$, while for big quantum cohomology the inverse is determined perturbatively as a Taylor series expansion whose coefficients are quasi-modular forms.

Tables

Table 1. Table 1. Transformation rules

	$\tau \mapsto \tau +1$	$\tau \mapsto -1/\tau$
${\theta }_{00}\left(\tau \right)$	${\theta }_{01}\left(\tau \right)$	${\left(-𝐢\tau \right)}^{1/2}{\theta }_{00}\left(\tau \right)$
${\theta }_{01}\left(\tau \right)$	${\theta }_{00}\left(\tau \right)$	${\left(-𝐢\tau \right)}^{1/2}{\theta }_{10}\left(\tau \right)$
${\theta }_{10}\left(\tau \right)$	$e^{2\pi 𝐢/8}{\theta }_{10}\left(\tau \right)$	${\left(-𝐢\tau \right)}^{1/2}{\theta }_{01}\left(\tau \right)$

Equations

$${\mathrm{ad}}_E:{\mathcal{T}}_M\to {\mathcal{T}}_M,v\mapsto [E,v]$$

$$E=\sum _{i=1}^N((1-d_i)t_i+r_i)\partial /\partial t_i,$$

$$0=d_1\le d_2\le \cdots \le d_N=:D$$

$$\theta :{\mathcal{T}}_M\to {\mathcal{T}}_M,v\mapsto [E,v]-\frac{1}{2}(2-D)v$$

$${\nabla }_{\partial /\partial t_i}$$

$${\nabla }_{\partial /\partial z}$$

$${\nabla }_{\partial /\partial t_i}^{(n)}$$

$${\nabla }_{\partial /\partial \lambda }^{(n)}$$

$$(M\times \mathbb{C}{)}^{\prime }\times H\to (M\times \mathbb{C}{)}^{\prime },$$

$$(M\times \mathbb{C}{)}^{\prime }=\{(t,\lambda )|\det (\lambda -E{\bullet }_t)\ne 0\}.$$

$$S(t,z)z^{\theta }{z}^{-\rho },$$

$$S(t,z)S(t,-z{)}^T=1,$$

$$I^{(n)}(t,\lambda )=\sum _{k=0}^{\infty }(-1{)}^k{S}_k(t){\tilde{I}}^{(n+k)}(\lambda ),$$

$$\displaystyle\widetilde{I}^{(m)}(\lambda)=e^{-\rho\partial_{\lambda}\partial_{m}}\Big{(}\frac{\lambda^{\theta-m-\frac{1}{2}}}{\Gamma(\theta-m+\frac{1}{2})}\Big{)}.$$

$$I_a^{(n)}(t,\lambda ):=I^{(n)}(t,\lambda )a,a\in H,n\in \mathbb{Z}.$$

$${\pi }_1((M\times \mathbb{C}{)}^{\prime },(t^{\circ },{\lambda }^{\circ }))\to \mathrm{GL}(H)$$

$${\pi }_1(F^{\circ },{\lambda }^{\circ }){\text{rTo}}^{i_{\ast }}{\pi }_1((M\times \mathbb{C}{)}^{\prime },(t^{\circ },{\lambda }^{\circ })){\text{rTo}}^{p_{\ast }}{\pi }_1(M,t^{\circ })\text{rTo}1,$$

$$(a|b):=(I_a^{(0)}(t,\lambda ),(\lambda -E\bullet )I_b^{(0)}(t,\lambda ))$$

$$w_a(x)=x-(a|x)a.$$

$$Z:((M\times \mathbb{C}{)}^{\prime }{)}^{\sim }\to H^{\ast },(t,\lambda )\mapsto Z(t,\lambda )$$

$$⟨Z(t,\lambda ),\alpha ⟩:=Z_{\alpha }(t,\lambda )=(I_{\alpha }^{(-1)}(t,\lambda ),1).$$

$$\mathbb{C}\times M\to M,(x,t)\mapsto t+x1,$$

$$Z(t,\lambda )=Z(t-\lambda 1,0).$$

$$X:=(B\times \mathbb{C}{)}^{\prime }=\{(t,\lambda )\in B\times \mathbb{C}|\det (\lambda -E\bullet )\ne 0\}.$$

$$\mathcal{M}(\Omega ,W):=\{f\in \Gamma (\Omega ,{\mathcal{O}}_{H^{\ast }}{)}^W|f\circ Z\in \mathcal{O}(B\times \mathbb{C})\},$$

$$Z:(\mathbb{C}\setminus \{0\}{)}^{\sim }\text{rTo}{H}^{\ast }\cong \mathbb{C},Z(\lambda )=\frac{{\lambda }^{-n-\frac{1}{2}}}{\Gamma (-n+\frac{1}{2})}.$$

$$\pi :\mathbb{C}/W\to \mathbb{C},\lambda \mapsto {\lambda }^q,$$

$$\mathbb{C}\setminus \{0\}\text{rTo}\mathbb{C}/W=\mathbb{C},\lambda \mapsto =\frac{{\lambda }^{-p}}{\Gamma (1-\frac{p}{q})}.$$

$${\varphi }_i:=p^{i-1},i=1,2,3,$$

$$M=\{(t_1,Q,t_3)\in \mathbb{C}\times {\mathbb{C}}^{\ast }\times \mathbb{C}||t_3{Q}^{1/3}|<ϵ\},$$

$$({\partial }_i,{\partial }_j)={\delta }_{i+j,4},1\le i,j\le 3,$$

Full text

1. Introduction

The results of this paper are in the settings of quantum cohomology of $\mathbb{P}^{2}$. Nevertheless, the problems that we solve can be given in much more general settings. Let us start by giving the general picture and providing some background and motivation for our results.

1.1. The second structure connection

We assume that the reader is familiar with the definition of a semi-simple Frobenius manifold (see [3] for some background). Let $M$ be a complex semi-simple Frobenius manifold and let $\mathcal{T}_{M}$ be the sheaf of holomorphic vector fields on $M$. By definition the data of Frobenius structure is given by the following list of objects

• (1)

A non-degenerate symmetric bi-linear pairing $(\ ,\ )$ on $\mathcal{T}_{M}$.

• (2)

A commutative associative multiplication $\bullet:\mathcal{T}_{M}\otimes\mathcal{T}_{M}\to\mathcal{T}_{M}$ .

• (3)

A flat vector field ${\mathbf{1}}\in\Gamma(M,\mathcal{T}_{M})$ that is a unity, i.e., $\mathbf{1}\bullet v=v$ for all $v\in\mathcal{T}_{M}.$

• (4)

An Euler vector field $E\in\Gamma(M,\mathcal{T}_{M})$.

We are going to work only with Frobenius manifolds satisfying the following 4 additional conditions:

1. (i)

The tangent bundle $TM$ is trivial and it admits a trivialization given by a frame of global flat vector fields. 2. (ii)

Recall that the operator

[TABLE]

preserves the space of flat vector fields. We require that the restriction of $\operatorname{ad}_{E}$ to the space of flat vector fields is a diagonalizable operator with rational eigenvalues. 3. (iii)

The Frobenius manifold has a calibration (see Section 1.2). 4. (iv)

The Frobenius manifold has a direct product decomposition $M=\mathbb{C}\times B$ such that if we denote by $t_{1}:M\to\mathbb{C}$ the projection along $B$, then $dt_{1}$ is a flat 1-form and $\langle dt_{1},\mathbf{1}\rangle=1.$

Conditions (i)–(iv) are satisfied for all Frobenius manifolds constructed by quantum cohomology or by the primitive forms in singularity theory.

Let us fix a base point $t^{\circ}\in M$ and a basis $\{\phi_{i}\}_{i=1}^{N}$ of the reference tangent space $H:=T_{t^{\circ}}M$. Furthermore, let $(t_{1},\dots,t_{N})$ be a local flat coordinate system on an open neighborhood of $t^{\circ}$ such that $\partial/\partial t_{i}=\phi_{i}$ in $H$. The flat vector fields $\partial/\partial t_{i}$ ($1\leq i\leq N$) extend to global flat vector fields on $M$ and provide a trivialization of the tangent bundle $TM\cong M\times H$. This allows us to identify the Frobenius multiplication $\bullet$ with a family of associative commutative multiplications $\bullet_{t}:H\otimes H\to H$ depending analytically on $t\in M$. Modifying our choice of $\{\phi_{i}\}_{i=1}^{N}$ and $\{t_{i}\}_{i=1}^{N}$ if necessary we may arrange that

[TABLE]

where $\partial/\partial t_{1}$ coincides with the unit vector field $\mathbf{1}$ and the numbers

[TABLE]

are symmetric with respect to the middle of the interval $[0,D]$. The number $D$ is known as the conformal dimension of $M$. The operator

[TABLE]

preserves the subspace of flat vector fields. It induces a linear operator on $H$ which is known to be skew symmetric with respect to the Frobenius pairing $(\ ,\ )$. Following Givental, we refer to $\theta$ as the Hodge grading operator.

There are two flat connections that one can associate with the Frobenius structure. The first one is usually called Dubrovin’s connection. It is a connection on the $H$-trivial bundle on $M\times\mathbb{C}^{*}$ defined by

[TABLE]

where $z$ is the standard coordinate on $\mathbb{C}^{*}=\mathbb{C}-\{0\}$ and for $v\in\Gamma(M,\mathcal{T}_{M})$ we denote by $v\bullet:H\to H$ the linear operator of Frobenius multiplication by $v$.

Our main interest is in the 2nd structure connection

[TABLE]

where $n\in\mathbb{C}$ is a complex parameter. This is a connection on the trivial bundle

[TABLE]

where

[TABLE]

The hypersurface $\det(\lambda-E\bullet_{t})=0$ in $M\times\mathbb{C}$ is called the discriminant.

1.2. Period vectors

The definition of the period map depends on the choice of a calibration $S(t,z)$ of $M$. By definition (see [6]), the calibration is an operator series $S=1+\sum_{k=1}^{\infty}S_{k}(t)z^{-k}$, $S_{k}\in\operatorname{End}(H)$, such that the Dubrovin’s connection has a fundamental solution near $z=\infty$ of the form

[TABLE]

where $\rho\in\operatorname{End}(H)$ is a nilpotent operator, $[\theta,\rho]=-\rho$, and the following symplectic condition holds

[TABLE]

where T denotes transposition with respect to the Frobenius pairing.

Let us fix a reference point $(t^{\circ},\lambda^{\circ})\in(M\times\mathbb{C})^{\prime}$ such that $\lambda^{\circ}$ is a sufficiently large real number. It is easy to check that the following functions provide a fundamental solution to the 2nd structure connection

[TABLE]

where

[TABLE]

The 2nd structure connection has a Fuchsian singularity at infinity, therefore the series $I^{(n)}(t,\lambda)$ is convergent for all $(t,\lambda)$ sufficiently close to $(t^{\circ},\lambda^{\circ})$. Using the differential equations we extend $I^{(n)}$ to a multi-valued analytic function on $(M\times\mathbb{C})^{\prime}$. We define the following multi-valued functions taking values in $H$:

[TABLE]

These functions will be called period vectors. Using analytic continuation we get a representation

[TABLE]

called the monodromy representation of the Frobenius manifold. The image $W$ of the monodromy representation is called the monodromy group.

Under the semi-simplicity assumption, we may choose a generic reference point $t^{\circ}$ on $M$, such that the Frobenius multiplication $\bullet_{t^{\circ}}$ is semi-simple and the operator $E\bullet_{t^{\circ}}$ has $N$ pairwise different eigenvalues $u_{i}^{\circ}$ ($1\leq i\leq N$). The fundamental group $\pi_{1}((M\times\mathbb{C})^{\prime},(t^{\circ},\lambda^{\circ}))$ fits into the following exact sequence

[TABLE]

where $p:(M\times\mathbb{C})^{\prime}\to M$ is the projection on $M$, $F^{\circ}=p^{-1}(t^{\circ})=\mathbb{C}\setminus{\{u_{1}^{\circ},\dots,u_{N}^{\circ}\}}$ is the fiber over $t^{\circ}$, and $i:F^{\circ}\to(M\times\mathbb{C})^{\prime}$ is the natural inclusion. For a proof we refer to [22], Proposition 5.6.4 or [18], Lemma 1.5 C. Using the exact sequence (2) we get that the monodromy group $W$ is generated by the monodromy transformations representing the lifts of the generators of $\pi_{1}(M,t^{\circ})$ in $\pi_{1}((M\times\mathbb{C})^{\prime},(t^{\circ},\lambda^{\circ}))$ and the generators of $\pi_{1}(F^{\circ},\lambda^{\circ})$.

The image of $\pi_{1}(F^{\circ},\lambda^{\circ})$ under the monodromy representation is a reflection group that can be described as follows. Using the differential equations of the 2nd structure connection it is easy to prove that the pairing

[TABLE]

is independent of $t$ and $\lambda$. This pairing is known as the intersection pairing. Suppose now that $\gamma$ is a simple loop in $F^{\circ}$, i.e., a loop that starts at $\lambda^{\circ}$, approaches one of the punctures $u_{i}^{\circ}$ along a path $\gamma^{\prime}$ that ends at a point sufficiently close to $u_{i}^{\circ}$, goes around $u_{i}^{\circ}$, and finally returns back to $\lambda^{\circ}$ along $\gamma^{\prime}$. By analyzing the second structure connection near $\lambda=u_{i}$ it is easy to see that up to a sign there exists a unique $a\in H$ such that $(a|a)=2$ and the monodromy transformation of $a$ along $\gamma$ is $-a$. The monodromy transformation representing $\gamma\in\pi_{1}(F^{\circ},\lambda^{\circ})$ is the reflection defined by the following formula:

[TABLE]

Let us denote by $R$ the set of all $a\in H$ as above determined by all possible choices of simple loops in $F^{\circ}$. We refer to the elements of $R$ as reflection vectors.

1.3. The ring of modular functions

Our main interest is in the period map

[TABLE]

where $((M\times\mathbb{C})^{\prime})^{\sim}$ is the universal cover of $(M\times\mathbb{C})^{\prime}$ and $Z(t,\lambda)\in H^{*}$ is defined by

[TABLE]

Recall that we require that the Frobenius manifold $M$ satisfies condition (iv) from Section 1.1. Under this condition the flow of the unit vector field $\mathbf{1}$ defines a free action of $\mathbb{C}$ on $M=\mathbb{C}\times B$

[TABLE]

where for $t=(t_{1},{}^{\prime}t)\in\mathbb{C}\times B$ we define $t+x\mathbf{1}:=(t_{1}+x,{}^{\prime}t)$. The period map has the following translation symmetry

[TABLE]

Therefore, we will restrict our analysis to the case $t_{1}=0$, i.e., we will assume that $t\in B$ and that the period map is defined on the universal cover of

[TABLE]

Let us denote by $\Omega\subset H^{*}$ the image of the period map $Z$. This is a $W$-invariant subset which will be called the period domain. In general very little is known about such domains. For example it would be interesting to classify semi-simple Frobenius manifolds such that the action of $W$ on $\Omega$ is properly discontinuous and the quotient $[\Omega/W]$ is an orbifold whose coarse moduli space is isomorphic to the Frobenius manifold $M$. Furthermore, we would like to introduce the ring of modular functions

[TABLE]

where $\Gamma(\Omega,\mathcal{O}_{H^{*}})^{W}$ is the ring of $W$-invariant holomorphic functions in $\Omega$. Note that in general if $f\in\Gamma(\Omega,\mathcal{O}_{H^{*}})^{W}$ is an arbitrary function, then the composition $f\circ Z$ defines a holomorphic function on $(B\times\mathbb{C})^{\prime}$. The condition in the above definition requires that $f\circ Z$ extends analytically across the discriminant.

1.4. Example

In general, one might try to investigate a more general period map defined by $(I^{(n)}(t,\lambda),\phi_{i})$ for any $n\in\mathbb{C}$ and $1\leq i\leq N$. Since $(I^{(n)}(t,\lambda),\phi_{i})=-\partial_{t_{i}}(I^{n-1}(t,\lambda),\mathbf{1})$ the choice $i=1$, which yields $\phi_{1}=\mathbf{1}$, is quite natural. Let us discuss the possibility of choosing different values of $n\in\mathbb{C}$ in the case of $A_{1}$-singularity. Since the periods defined by the second structure connection locally near a generic point on the discriminat have the same leading order terms as the periods of $A_{1}$-singularity, one can get a good intuition of what values of $n$ could be interesting to investigate.

The period map takes the form

[TABLE]

The monodromy group is a cyclic group $W=(w)$ and the action of $w$ on $\mathbb{C}$ is multiplication by $e^{-\pi\sqrt{-1}(2n+1)}$. If $n$ is a complex non-rational number then the quotient $\mathbb{C}/W$ might even fail to be a Hausdorf space. Let us assume that $n\in\mathbb{Q}$, so that the quotient is an orbifold. The quotient space $\mathbb{C}/W$ has the structure of a smooth complex manifold isomorphic to $\mathbb{C}$. The isomorphism is induced from

[TABLE]

where we write $2n+1=\tfrac{2p}{q}$ with $q>0$ and $p$ relatively prime integers. The period map induces a holomorphic map

[TABLE]

The above map extends holomorphically across the discriminant $\lambda=0$ if and only if $p\leq 0$. Moreover it is an isomorphism if and only if $p=-1$, i.e., $n=-\tfrac{1}{2}-\tfrac{1}{q}$.

The conclusion is that in general, there might be other period maps which would allow us to identify the Frobenius manifold with the orbit space of the corresponding monodromy group. Our choice $n=-1$ is motivated by the applications of semi-simple Frobenius manifold to integrable hierarchies and representations of lattice vertex algebras (see the Appendix for more details).

1.5. Riemann–Hilbert problem for Gromov–Witten invariants

The notion of a Frobenius manifold was invented by Dubrovin [3] in order to give a geometric interpretation of the properties of quantum cohomology of a smooth projective variety $X$. It was conjectured by Givental in [5] and proved by Teleman in [23] that if the quantum cohomology is semi-simple as a Frobenius manifold then the higher genus Gromov–Witten invariants are uniquely determined by genus-0, i.e., by the underlying semi-simple Frobenius structure. On the other hand, the entire semi-simple Frobenius structure is uniquely determined by the second structure connection. The latter is a Fuchsian connection and hence can be recovered uniquely as a solution to a classical Riemann–Hilbert problem (see [4, 13]). The problem that we are interested in is how to express the higher genus Gromov–Witten invariants of a manifold with semi-simple quantum cohomology in terms of the monodromy data of the second structure connection. We refer to such a problem as the Riemann–Hilbert problem for Gromov–Witten invariants. By definition such a problem has a solution. Namely, by solving a classical Riemann Hilbert problem we can recover the second structure connection from its monodromy data, then we can recover the semi-simple Frobenius structure, and finally it remains to recall Givental’s higher genus reconstruction. The solutions to the Riemann–Hilbert problems are usually highly transcendental. Therefore, it seems that the dependence of the Gromov–Witten invariants on the monodromy data should also be quite complicated. However, in a series of examples (see [1, 9, 16, 15]) the monodromy data leads to a highest weight representation of a vertex algebra or to the Hirota bilinear equations of an integrable hierarchy which allow us to uniquely determine the invariants. In other words, we are looking for a representation of a Lie algebra or more generally a vertex algebra which will allow us to express all invariants via the monodromy data in a simple combinatorial way. The main point is not that we will find a way to compute Gromov–Witten invariants, but rather that we can understand a conceptual question. Namely is there a strong relation between semi-simple Frobenius manifolds and Lie algebras. Note that the set of reflection vectors can be used to generalize the notion of root systems and to propose various constructions of Lie algebras. This was actually done in the settings of singularity theory by several authors (e.g. see [20, 21]). The problem is whether such Lie algebras have interesting applications. The Riemann–Hilbert problem for Gromov–Witten invariants can be viewed as a motivation to develop Lie algebra theory for semi-simple Frobenius manifolds.

All examples in which some interesting relation to Lie theory was established have conformal dimension $D\leq 1$ ($D$ is the complex dimension of the manifold $X$ in the case of quantum cohomology). The case of quantum cohomology of $\mathbb{P}^{2}$ is a very good candidate to make progress in conformal dimension $>1$ because the geometry of $\mathbb{P}^{2}$ relevant for studying quantum cohomology and mirror symmetry is very well understood. The problem in the current paper comes from our attempt to generalize the work in [1]. Namely, we would like to find differential operator constraints for the total descendant potential of $\mathbb{P}^{2}$. This is still a very difficult problem. We will argue in the appendix that the genus-0 reduction of a differential operator constraints yields a Hamilton–Jacobi equation given by a holomorphic function in $\mathcal{M}(\Omega,W)$. This motivates to some extent our interest in the ring of modular functions.

Definition 1.1**.**

The period map $Z$ is said to be invertible if there exists a set of modular functions $f_{i}\in\mathcal{M}(\Omega,W)$ $(1\leq i\leq N)$ such that the set of holomorphic functions $f_{i}\circ Z$ $(1\leq i\leq N)$ is a coordinate system on $B\times\mathbb{C}$. A set of such modular functions $\{f_{i}\}_{i=1}^{N}$ is called the inverse of the period map.

There are two reasons why we are interested in finding the inverse of the period map. The first one is related to the discussion above. We expect that if the period map is invertible then the corresponding modular functions $f_{i}$ will give a complete set of recursion relations, which would allow us to determine the genus-0 total descendant potential in terms of the monodromy data of the Frobenius manifold via an explicit recursion (see Appendix for more details).

The second reason is related to the problem of uniformizing a semi-simple Frobenius manifold. We expect that semi-simple Frobenius manifolds relevant in the study of mirror symmetry are quotients of a simply connected domain by a discrete group. At this point we can only speculate, but we believe that the problem of uniformizing the Frobenius manifold corresponding to the quantum cohomology of some smooth projective variety $X$ is related to the problem of constructing the manifold of stability conditions of the bounded derived category $D^{b}(\operatorname{Coh}\,X)$.

Acknowledgements. I would like to thank M. Kapranov for pointing out to me that if the solutions to some differential equations satisfy a quadratic relation, then this is an indication that the differential equation itself might be a symmetric square. I would like to thank the anonymous referee for the useful comments and remarks that helped me to improve the exposition. This work is partially supported by JSPS Grant-In-Aid (Kiban C) 17K05193 and by the World Premier International Research Center Initiative (WPI Initiative), MEXT, Japan.

2. Quantum cohomology of $\mathbb{P}^{2}$

From now on we will work only in the settings of quantum cohomology of $\mathbb{P}^{2}$. The goal of this section is to introduce the necessary notation and to state our results.

2.1. Frobenius manifold structure

Let $H=H^{*}(\mathbb{P}^{2};\mathbb{C})$ and $t_{i}$ $(1\leq i\leq 3)$ be the linear coordinates on $H$ corresponding to the basis

[TABLE]

where $p=c_{1}(\mathcal{O}(1))$ is the hyperplane class. Quantum cohomology defines a Frobenius manifold structure on the space

[TABLE]

where $\epsilon$ is a sufficiently small positive real number and the coordinate $Q:=e^{t_{2}}$ is identified with the Novikov variable. By definition, the linear coordinates $t_{i}$ are flat, the Frobenius pairing is given by the Poincare pairing

[TABLE]

where $\partial_{i}=\partial/\partial t_{i}$, while the multiplication is given by the quantum cup product. The latter is defined by

[TABLE]

where $F(t)$ is the genus-0 potential

[TABLE]

where $t:=t_{1}+t_{2}p+t_{3}p^{2}$. Following Kontsevich and Ruan–Tian we can derive an explicit recursive formula for $F$ as follows. Using the string equation, the divisor equation, and the dimension formula of the virtual fundamental cycle we get that $F$ has the form

[TABLE]

where the coefficient $N_{d}$ can be interpreted as the number of rational curves in $\mathbb{P}^{2}$ of degree $d$ passing through $3d-1$ points in general position. The system of WDVV equations contains a single non-trivial equation

[TABLE]

where the index $i$, $i=2,3$, denotes partial derivative with respect to $t_{i}$. Comparing the coefficients in front of $Q^{d}$ yields

[TABLE]

which together with $N_{1}=1$ determines $N_{d}$ for all $d>1$. The first few values are

[TABLE]

Let us point out that the number $\epsilon$ in the definition of the domain $M$ is chosen in such a way that the radius of convergence of the series (5) is $\epsilon|Q|^{-1/3}$.

Furthermore, the Euler vector field has the form

[TABLE]

and the Hodge grading operator is

[TABLE]

2.2. The $\Gamma$-integral structure of Iritani

Let us recall the notation of Section 1.2. Following Givental (see [6]), we equip the quantum cohomology with calibration

[TABLE]

defined by

[TABLE]

The fundamental solution corresponding to such calibration is $S(t,z)z^{\theta}z^{-\rho}$, where the nilpotent operator $\rho$ is given by classical cup product multiplication by $c_{1}(T\mathbb{P}^{2})=3p$.

There is a very elegant way to describe the reflection lattice, i.e., the $\mathbb{Z}$-submodule of $H$ spanned by all reflection vectors $a\in R.$ Namely, using Iritani’s $\Gamma$-class modification of the Chern character map, we will obtain an explicit description of all reflection vectors in terms of $K^{0}(\mathbb{P}^{2})$ – the $K$-ring of topological vector bundles on $\mathbb{P}^{2}$. Following Iritani’s construction in [11], let us introduce the map

[TABLE]

defined by

[TABLE]

where ${\rm deg}:H\to H$ is the degree operator ${\rm deg}(p^{i})=ip^{i}$ and the $\Gamma$ function should be expanded as a Taylor series at $p=0$, i.e.,

[TABLE]

Recall that $K^{0}(\mathbb{P}^{2})=\mathbb{Z}[L]/(L-1)^{3}$, where $L=\mathcal{O}(1)$. The above formula gives

[TABLE]

Slightly abusing the notation we identify $L^{m}$ with its image $\Psi(L^{m})$.

Let us choose a reference point $(t^{\circ},\lambda^{\circ})\in(M\times\mathbb{C})^{\prime}$ such that $t_{1}^{\circ}=t_{2}^{\circ}=t_{3}^{\circ}=0$ and $\lambda^{\circ}$ is a sufficiently large real number. Recall that for $t_{3}=0$ the quantum cup product $\bullet_{t}$ turns $H$ into the following algebra

[TABLE]

We get that the eigenvalues of $E\bullet_{t^{\circ}}$ are $u_{i}^{\circ}=3\zeta^{-i+1}$ $(1\leq i\leq 3)$, where $\zeta=e^{2\pi\sqrt{-1}/3}.$ Let us denote by $[a,b]$ $(a,b\in\mathbb{C})$ the straight segment in $\mathbb{C}$ from $a$ to $b$.

Let $\gamma_{i}$ $(1\leq i\leq 3)$ be the composition of the arc

[TABLE]

and the straight segment $[\zeta^{-i+1}\lambda^{\circ},u_{i}^{\circ}]$ (see Figure 1).

Proposition 2.1**.**

Let $\gamma_{i}$ $(1\leq i\leq 3)$ be the paths constructed above. Then $L^{i-1}$ is the reflection vector corresponding to the path $\gamma_{i}$.

Corollary 2.2**.**

The set of all reflection vectors is given by

[TABLE]

The proof of Proposition 2.1 and Corollary 2.2 will be given in Section 3.

Since the intersection pairing (3) is independent of $t$, by setting $t_{1}=t_{3}=0$ and passing to the limit $\operatorname{Re}(t_{2})\to-\infty$ (i.e. $Q\to 0$) we get the following formula for the intersection pairing

[TABLE]

Recalling Corollary 2.2 we get that the lattice

[TABLE]

coincides with the reflection lattice. Furthermore, recalling Proposition 2.1 we get Dubrovin’s result (see [4]) that the monodromy group $W\cong\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$. The construction of this isomorphism amounts to choosing an appropriate $\mathbb{Q}$-basis $(E_{1},E_{2},E_{3})$ of the reflection lattice. In our notation this basis is given by

[TABLE]

2.3. The period map for quantum cohomology of $\mathbb{P}^{2}$

Now we are in position to state our results. We identify $H^{*}\cong\mathbb{C}^{3}$ via the linear functions on $H^{*}$ corresponding to $E_{i}$, $i=1,2,3$. The period map takes the form

[TABLE]

where $Z_{i}(t,\lambda):=Z_{E_{i}}(t,\lambda)$. As we have explained in the introduction, we may restrict our analysis to parameters $t\in B:=\{t\in M\ |\ t_{1}=0\}.$ Put $B_{\rm small}=\{t\in B\ |\ t_{3}=0\}=\mathbb{C}^{*}$ and

[TABLE]

Let us introduce the domain

[TABLE]

Let us point out that there is a natural isomorphism

[TABLE]

under which the action of the monodromy group takes a very simple form (see Lemma 4.4). We will prove later on (see Lemma 5.2) that $X_{\rm small}$ is a deformation retract of $X$. Therefore, the universal cover $\widetilde{X}_{\rm small}$ is an analytic submanifold of $\widetilde{X}$ and we can introduce the restriction of the period map $Z_{\rm small}:=Z|_{\widetilde{X}_{\rm small}}$. Recall the Eisenstein series

[TABLE]

where $q=e^{2\pi\sqrt{-1}\tau}$. Our first result can be stated as follows.

Theorem 2.3**.**

a) The image of $Z_{\rm small}$ is $\Omega_{\rm small}\setminus{\{E_{6}(-z_{2}/(2z_{3}))=0\}}$.

b) Let $\widetilde{\pi}_{\rm small}:\widetilde{X}_{\rm small}\to X_{\rm small}$ be the universal cover and

[TABLE]

be the map defined by

[TABLE]

where $(\tau,y)$ is the coordinate system on $\Omega_{\rm small}$ introduced above. Then $\pi_{\rm small}\circ Z_{\rm small}=\widetilde{\pi}_{\rm small}$.

c) The fibers of the map $\pi_{\rm small}$ are the $W$-orbits in $\Omega_{\rm small}$, i.e., the small quantum cohomology $B_{\rm small}\times\mathbb{C}$ is the coarse moduli space for the orbifold $[\Omega_{\rm small}/W]$.

Generalizing the results of Theorem 2.3 to big quantum cohomology is a very challenging problem. We expect that $M_{\rm small}:=\mathbb{C}\times B_{\rm small}$ is an analytic subvariety in a larger Frobenius manifold $N$ and that $M$ is just a tubular neighborhood of $M_{\rm small}$ in $N$. We were able to prove an interesting result about the holomorphic thickening of $Z_{\rm small}$, which might be viewed as the first step towards constructing a global Frobenius manifold.

The period map $Z$ maps a small open neighborhood of $\widetilde{X}_{\rm small}$ in $\widetilde{X}$ into a small open neighborhood of $\Omega_{\rm small}$ in $\mathbb{C}^{3}$. Therefore we have an induced map of ringed spaces

[TABLE]

where $\widetilde{\iota}:\widetilde{X}_{\rm small}\to\widetilde{X}$ and $\iota:\Omega_{\rm small}\to\mathbb{C}^{3}$ are the natural inclusion maps. The ring of regular functions on $\Omega$ is by definition $\Gamma(\Omega_{\rm small},\mathcal{O}_{\mathbb{C}^{3}})$, i.e., functions defined and holomorphic in an open neighborhood of $\Omega_{\rm small}$ in $\mathbb{C}^{3}$. This ring is equipped with the action of the monodromy group $W$. The pullback via the period map $Z$ defines a ring homomorphism

[TABLE]

where $\Gamma(\Omega_{\rm small},\mathcal{O}_{\mathbb{C}^{3}})^{W}$ is the subring of $W$-invariant functions. The coordinate functions $Q=e^{t_{2}}$, $t=t_{3}$, and $\lambda$ of $B\times\mathbb{C}$ are elements of $\Gamma(X_{\rm small},\mathcal{O}_{X})$. We will prove that $Q,t$, and $\lambda$ are pullbacks via the period map of $W$-invariant functions. Moreover, the latter have some interesting property, which can be stated as follows. Let us construct an open neighborhood of $\Omega_{\rm small}$ as the image of the map

[TABLE]

where $\mathbb{H}=\{\tau\in\mathbb{C}\ |\ \operatorname{Im}(\tau)>0\}$ is the upper half-plane. The image of $\Phi$ is the coarse moduli space for the orbifold quotient $[\mathbb{H}^{2}\times\mathbb{C}^{*}/\mu_{2}]$, where $\mu_{2}$ is the cyclic group of order 2 whose generator acts on $\mathbb{H}^{2}$ by permutation $(\tau_{1},\tau_{2})\mapsto(\tau_{2},\tau_{1})$.

Theorem 2.4**.**

a) There are $W$-invariant functions in $\Gamma(\Omega_{\rm small},\mathcal{O}_{\mathbb{C}^{3}})^{W}$ of the form

[TABLE]

where $\tau_{12}:=(\tau_{1}+\tau_{2})/2$, such that their pullbacks via the period map coincide with the coordinate functions $Q,\lambda$, and $t$.

b) The coefficients $Q_{n}(\tau),\lambda_{n}(\tau)$ $(n\geq 0)$ are quasi-modular forms, i.e., they are polynomials in the Eisenstein series $E_{i}(\tau)$, $i=2,4,6$.

Note that if $\tau_{1}=\tau_{2}$ then we recover the formulas from Theorem 2.3. In particular

[TABLE]

We have computed the quasi-modular forms $\lambda_{n}$ and $Q_{n}$ for $n=1,2,3$. The answer is the following

[TABLE]

and

[TABLE]

where $\Delta:=E_{4}^{3}-E_{6}^{2}$.

3. Reflection vectors in quantum cohomology of $\mathbb{P}^{2}$

The main goal in this section is to prove Proposition 2.1.

3.1. Mirror symmetry for the calibration

In this section we recall an identity expressing the operator series $S(t,z)$ in terms of an oscillatory integral. Let us denote by $S(Q,z)$ the restriction of $S(t,z)$ to $t_{1}=t_{3}=0$. Recall that a Givental’s mirror model of $\mathbb{P}^{2}$ is given by the family of functions

[TABLE]

depending on the parameter $Q\in\mathbb{C}^{*}$ and the holomorphic form on $(\mathbb{C}^{*})^{2}$ defined by $\omega=\tfrac{dx_{1}}{x_{1}}\wedge\tfrac{dx_{2}}{x_{2}}$.

Let us fix $Q=e^{2\pi\sqrt{-1}s}q$, where $q>0$ and $s$ are real numbers and define the semi-infinite cycle $\alpha_{Q}:=e^{2\pi\sqrt{-1}s/3}(\mathbb{R}_{>0})^{2}$, i.e., $\alpha_{Q}$ consists of all $x\in(\mathbb{C}^{*})^{2}$ such that $x_{i}=e^{2\pi\sqrt{-1}s/3}|x_{i}|$ ($i=1,2$). The following result is due to Iritani (see [11], Theorem 4.14).

Lemma 3.1**.**

Suppose that $z=e^{2\pi\sqrt{-1}s/3}w$ where $w<0$ is a real number. Then the following formula holds

[TABLE]

Proof.

Note that the LHS is independent of $s$, because if we substitute $x_{i}=e^{2\pi\sqrt{-1}s/3}y_{i}$, then the form $\omega$ is invariant, the cycle $\alpha_{Q}$ is transformed into $\alpha_{q}$, and the function

[TABLE]

Therefore, we may assume that $s=0$ and that $Q=q>0$ and $z=w<0$ are real numbers.

Let us simplify the RHS of (6). Recalling the definition of $S$ we get

[TABLE]

Using the divisor and the string equations the above formula can be transformed into

[TABLE]

where recall that $t=t_{2}p$ and $q=e^{t_{2}}$. The above formula is by definition Givental’s J-function of $\mathbb{P}^{2}$. The J-functions of all projective spaces and certain classes of complete intersections are computed explicitly in [8], Theorem 9.1. We get

[TABLE]

Using the commutation relation $(-w)^{-\theta}p=(-w)p(-w)^{-\theta}$ and $\theta(\mathbf{1})=\mathbf{1}$ we get

[TABLE]

Put $\tau:=t_{2}-3\log(-w)$. Then the RHS of (6) takes the form

[TABLE]

where $[\mathbb{P}^{2}]$ is the fundamental class of $\mathbb{P}^{2}$ and we used the following formulas $\Psi(\mathcal{O})=\tfrac{1}{\sqrt{2\pi}}\Gamma(1+p)^{3}$, $\theta^{T}=-\theta$, $\rho^{T}=\rho$. The above integral can be written as a residue. Namely if $f(p)\in\mathbb{C}[p]$ is a polynomial, then we have

[TABLE]

Using this fact and the identity $\Gamma(1+p)=p(p-1)\cdots(p-d)\Gamma(p-d)$ we get that the RHS of formula (6) has the form

[TABLE]

Let us transform the LHS of (6). Note that the oscillatory integral in (6) can be written as

[TABLE]

where $\tau=t_{2}-3\log(-w)$ and we used the substitution $x_{i}=-we^{y_{i}}$ $(1\leq i\leq 2)$. If $\epsilon>0$ is a fixed real number then $I(\epsilon,\tau):=e^{\epsilon\tau}I(\tau)\in L^{1}(\mathbb{R})$. Recalling the Fourier inversion formula we get

[TABLE]

Let us substitute in the above formula $I(\epsilon,\tau)=e^{\epsilon\tau}I(\tau)$, $I(\epsilon,s)=e^{\epsilon s}I(s)$, and make the substitution $p=\epsilon-\sqrt{-1}\xi$. We get

[TABLE]

where the orientation of the contour $\epsilon+\sqrt{-1}\mathbb{R}$ is from $\epsilon-\sqrt{-1}\infty$ to $\epsilon+\sqrt{-1}\infty$. The integral with respect to $s$ can be computed explicitly as follows:

[TABLE]

where the first equality is justified as follows. The function

[TABLE]

with respect to $(y_{1},y_{2},s)\in\mathbb{R}^{3}$ is of class $L^{1}$ for all $p$ satisfying $\operatorname{Re}(p)>0$. Therefore we can use the Fubini’s theorem to transform the iterated integral $\int_{\mathbb{R}}ds\int_{\mathbb{R}^{2}}dy_{1}dy_{2}$ as a multiple integral. It remains only to change the integration variables $(y_{1},y_{2},s)$ into $(y_{1},y_{2},y_{3})$ via the substitution $s=y_{1}+y_{2}+y_{3}$.

The oscillatory integral takes the form

[TABLE]

The integrand has poles at $p=-d$ for $d=0,1,2,\dots$. Using the Cauchy residue theorem and some standard estimates based on the Stirling formula for the $\Gamma$-function we get

[TABLE]

This completes the proof of (6). ∎

Remark 3.2**.**

The proof of Lemma 3.1 is inspired by the work of K. Hori and M. Romo (see [10]). The main idea is that the Laplace transform with respect to the Novikov’s variables of the oscillatory integral is the integrand of a Mellin–Barnes integral. Therefore, using the inverse Laplace transform we can identify the oscillatory integral with a Mellin–Barnes integral. In fact, this method seems to be quite general and although there are technical difficulties it will be interesting to investigate more complicated targets.

3.2. Twisted thimble integrals

Let us continue to work with $Q=e^{2\pi\sqrt{-1}s}q$ and $z=e^{2\pi\sqrt{-1}s/3}w$ such that $q>0$, $w<0$, and $s$ are real numbers. The key object that would allow us to prove Proposition 2.1 is the following integral

[TABLE]

where $m\geq 1$ is an integer, the parameter $\lambda=e^{2\pi\sqrt{-1}s/3}r$ with $r\geq 3q^{1/3}$ a real number, and $\alpha_{Q,\lambda}\subset(\mathbb{C}^{*})^{2}$ is the submanifold with boundary defined by

[TABLE]

The integrand is a multivalued analytic function. We fix an analytic branch as follows. Note that

[TABLE]

and that the expression in the brackets is non-negative due to the mean arithmetic and mean geometric inequality and our assumption that $r\geq 3q^{1/3}$. We define

[TABLE]

Let us point out that $\alpha_{Q,\lambda}$ is a Lefschetz thimble, i.e., it is swept out by vanishing cycles in the following sense. The function $f(x,Q)$ induces a map

[TABLE]

where $u_{Q}=3Q^{1/3}=3e^{2\pi\sqrt{-1}s/3}q^{1/3}$ is the critical value of $f(x,Q)$ corresponding to the critical point $\xi_{Q}=(Q^{1/3},Q^{1/3})$. The fiber over $\mu\in(u_{Q},\lambda]$ is diffeomorphic to a circle, while the fiber over $u_{Q}$ is the critical point $\xi_{Q}$.

Lemma 3.3**.**

Suppose that $Q=e^{2\pi\sqrt{-1}s/3}q$ and $z=e^{2\pi\sqrt{-1}s/3}w$, where $q>0$ and $w<0$ are real numbers. Then

[TABLE]

where the integral on the LHS is along the ray $\lambda=e^{2\pi\sqrt{-1}s/3}r$, $r\in[3q^{1/3},+\infty)$.

Proof.

Using Fubini’s theorem we have

[TABLE]

Therefore, the LHS of (7) can be written as

[TABLE]

Using $e^{\lambda/z}=e^{(\lambda-\mu)/z}e^{\mu/z}$ and changing the order of integration we get

[TABLE]

Let us change the integration variable $\lambda$ into $t$ via the following substitution $\lambda-\mu=-tz$. Note that the integration range for $t$ is $[0,+\infty)$ and that the integral with respect to $\lambda$ turns into $(-z)^{m+\tfrac{1}{2}}$. Therefore, the LHS of (7) is

[TABLE]

Recalling again the Fubini’s theorem we get that the above integral coincides with the RHS of (7). ∎

3.3. Mirror symmetry for the second structure connection

We will prove that the thimble integrals can be used to construct a solution to the second structure connection.

Lemma 3.4**.**

Suppose that $Q=e^{2\pi\sqrt{-1}s}q$ and $\lambda=e^{2\pi\sqrt{-1}s/3}r$, where $q>0$ and $r\geq 3q^{1/3}$ are real numbers. Then there exists a constant vector $E\in H$ independent of $Q$ and $\lambda$ such that

[TABLE]

for all integers $m\geq 2$.

Proof.

Let us denote by $L(Q,\lambda)$ the $H$-valued function on $Q$ and $\lambda$ defined uniquely in such a way that $(L(Q,\lambda),\phi_{i})$ coincides with the RHS of (9) for all $i=1,2,3$. Let us denote by $A(Q):=p\bullet_{t}\in\operatorname{End}(H)$ the linear operator of quantum multiplication by $t$, where $t=t_{2}p$ and $t_{2}=\log Q=\tfrac{2\pi\sqrt{-1}s}{3}+\log q$. Since $I^{(-m-1)}(t,\lambda)$ is a fundamental solution to the second structure connection, the restriction of $I^{(-m-1)}(t,\lambda)$ to $t_{1}=t_{3}=0$, $t_{2}=\log Q$ is a fundamental solution to the following system of equations

[TABLE]

In order to prove formula (9), it is sufficient to prove that the vector valued function $L(Q,\lambda)$ is a solution to the above system of equations.

Recalling the definition of $\mathcal{I}^{(-m)}(Q,\lambda)$ we get that it has the following scaling symmetry

[TABLE]

for every real number $c>0$. Differentiating in $c$ and setting $c=1$ yields the following differential equation

[TABLE]

This implies that

[TABLE]

On the other hand, since $\theta(\phi_{i})=(2-i)\phi_{i}$ we have

[TABLE]

It follows that $L(Q,\lambda)$ satisfies (11). Note that $\partial_{\lambda}\mathcal{I}^{(-m-1)}(Q,\lambda)=\mathcal{I}^{(-m)}(Q,\lambda)$ (see formula (8)). Recalling the definition of $L(Q,\lambda)$ we get

[TABLE]

for $i=1,2$, where we used that $A(Q)$ is self-adjoint with respect to the Poincare pairing. Since $A(Q)\phi_{3}=Q\phi_{1}$ in order to prove that $L(Q,\lambda)$ is a solution to (10) it remains only to prove that

[TABLE]

Note that $-Q\partial_{Q}\mathcal{I}^{(-m)}(Q,\lambda)$ is equal to

[TABLE]

where we used the identity

[TABLE]

and integration by parts. Similarly,

[TABLE]

Differentiating the above formula with $-Q\partial_{Q}$ yields (12). ∎

3.4. Proof of Proposition 2.1

Let us fix an analytic branch of $S(Q,z)$ for all $Q$ sufficiently close to $Q^{\circ}:=1$. Recall that $\lambda^{\circ}$ is a sufficiently large real number. Let us fix also $m$ to be a sufficiently large positive integer number (e.g. $m\geq 2$ would work). If $\lambda$ is sufficiently close to $\lambda^{\circ}$, then we fix a branch of $I^{(-m-1)}(Q^{\circ},\lambda)$ – for example pick the principal branch of $\log\lambda$, then this would determine $\widetilde{I}^{(n)}(\lambda)$ for all $\lambda$ close to $\lambda^{\circ}$ and hence $I^{(-m-1)}(Q^{\circ},\lambda)$ is also uniquely determined.

Recalling Lemma 3.4 we get that there exist a vector $E$ such that formula (9) holds with $Q=Q^{\circ}$. Since the RHS of (9) is an integral over a Lefschetz thimble, we get that it vanishes when $\lambda$ approaches $u_{Q^{\circ}}$ along the line segment $(u_{Q^{\circ}},\lambda^{\circ}]$. Therefore, since $u_{Q^{\circ}}=3=u_{1}^{\circ}$ and the line segment $(u_{1}^{\circ},\lambda^{\circ}]$ is precisely the path $\gamma_{1}$, we get that $E$ is proportional to the reflection vector corresponding to the path $\gamma_{1}$. We claim that $E=\sqrt{2\pi}\Psi(\mathcal{O})$. Since the intersection pairing $(\Psi(\mathcal{O})|\Psi(\mathcal{O}))=2$ this would prove that $\mathcal{O}$ is the reflection vector corresponding to the path $\gamma_{1}$ as claimed.

Let us prove that $E=\sqrt{2\pi}\Psi(\mathcal{O})$. Let us apply to formula (9) the Laplace transform $\int_{u_{Q^{\circ}}}^{\infty}d\lambda e^{\lambda/z}$. Recalling (7) we get

[TABLE]

where we allow $Q$ to be a real deformation of $Q^{\circ}$. Recalling formula (6) with $s=0$ and using the quantum differential equation $-Q\partial_{Q}S(Q,z)=(-z)^{-1}p{\bullet}S(Q,z)$ we get that the RHS of the above identity is

[TABLE]

In other words we proved that

[TABLE]

We would like to take the limit $Q\to 0$. Although this limit does not exists, it is not hard to characterize the singularities of both sides at $Q=0$. Using the divisor equation we have $S(Q,z)=T(Q,z)e^{P\log Q/z}$, where $P:=p\cup$ denotes the operator of classical cup product multiplication by $p$, $T(Q,z)$ is analytic at $Q=0$ and $T(0,z)=1$. Let us write $T(Q,z)=\sum_{k=0}^{\infty}T_{k}(Q)z^{-k}$. We claim that

[TABLE]

Indeed, by definition

[TABLE]

On the other hand, using that $P\theta=(\theta+1)P$ we get

[TABLE]

Substituting this formula in (15) and summing over all $l$ we get exactly (14). Furthermore,

[TABLE]

where we used that $-\operatorname{ad_{\theta}}(P)=P$. Now it is clear that both sides are polynomials in $\log Q$ of degree 2, whose coefficients take values in the ring of $H$-valued convergent power series in $Q$. Comparing the coefficients in front of $(\log Q)^{0}$ and passing to the limt $Q\to 0$ we get

[TABLE]

The LHS of (16) is by definition

[TABLE]

Using the substitution $\lambda=-tz$, we get that the integral in the above formula is $(-z)^{\theta+m+\tfrac{3}{2}}$ and hence the LHS of (16) takes the form

[TABLE]

Comparing with the RHS of (16) we get $E=\sqrt{2\pi}\Psi(\mathcal{O})$.

The rest of the proposition is easy to complete. Let us look at formula (9) for $Q=e^{2\pi\sqrt{-1}s}Q^{\circ}$, $\lambda=e^{2\pi\sqrt{-1}s/3}\lambda^{\circ}$ and decrease $s$ continuously from $s=0$ to $s=-1$. The vector $E=\sqrt{2\pi}\Psi(\mathcal{O})$ does not change, so according to formula (14) the LHS will be transformed into $(I^{(-m-1)}(Q^{\circ},\lambda^{\circ}\zeta^{-1})e^{2\pi\sqrt{-1}p}E,\phi_{i})$, which is the same as the analytic continuation of $(I^{(-m-1)}(Q^{\circ},\lambda^{\circ})e^{2\pi\sqrt{-1}p}E,\phi_{i})$ along the arc $\lambda=e^{2\pi\sqrt{-1}s/3}\lambda^{\circ}$, $-1\leq s\leq 0$. On the other hand, on the RHS of (9) the only change will be that in the corresponding thimble integrals the cycle $\alpha_{Q^{\circ},\lambda^{\circ}}$ will be transformed to the Lefschetz thimble $\alpha_{Q^{\circ}e^{-2\pi\sqrt{-1}},\lambda^{\circ}\zeta^{-1}}$. The conclusion is that $(I^{(-m-1)}(Q^{\circ},\lambda)e^{2\pi\sqrt{-1}p}E,\phi_{i})$ vanishes as $\lambda$ approaches $u_{Q^{\circ}e^{-2\pi\sqrt{-1}}}=u_{Q^{\circ}}\zeta^{-1}=u^{\circ}_{2}$ first along the arc $\lambda=e^{2\pi\sqrt{-1}s/3}\lambda^{\circ}$ ($-1\leq s\leq 0$) and then along the line segment $[\lambda^{\circ}\zeta^{-1},u^{\circ}_{2}]$, i.e., as $\lambda$ travels along the path $\gamma_{2}$. Therefore, $e^{2\pi\sqrt{-1}p}E=\sqrt{2\pi}e^{2\pi\sqrt{-1}p}\Psi(\mathcal{O})=\sqrt{2\pi}\Psi(\mathcal{O}(1))$ is proportional to the reflection vector corresponding to the path $\gamma_{2}$. The intersection pairing $(\Psi(\mathcal{O}(1))|\Psi(\mathcal{O}(1)))=2$, so $\Psi(\mathcal{O}(1))$ is a reflection vector. The argument that $\Psi(\mathcal{O}(2))$ is the reflection vector corresponding to the path $\gamma_{3}$ is similar – one just has to decrease further $s$ from $-1$ to $-2$. ∎

Let us sketch the proof of Corollary 2.2. First, let us point out that our proof of Proposition 2.1 implies that $L^{n}=\mathcal{O}(n)$ is a reflection vector for all $n\in\mathbb{Z}$. Let $\ell_{i}\in\pi_{1}(F^{\circ},\lambda^{\circ})$ be a simple loop around $u_{i}^{\circ}$ corresponding to the path $\gamma_{i}$ and $M_{i}\in W$ be the monodromy transformation representing $\ell_{i}$. Let $c\in\pi_{1}((M\times\mathbb{C})^{\prime},(t^{\circ},\lambda^{\circ}))$ be the loop that under the projection $(M\times\mathbb{C})^{\prime}\to M$ maps to the loop in $\pi_{1}(M,t^{\circ})\cong\pi_{1}(\mathbb{C}^{*},Q^{\circ})\cong\mathbb{Z}$ that goes once around $Q=0$ in clockwise direction. Formula (14) implies that the monodromy transformation $K\in W$ representing the loop $c$ is multiplication by $e^{2\pi\sqrt{-1}p}$ which via Iritani’s map $\Psi$ corresponds to K-theoretic multiplication by $\mathcal{O}(1)$. By definition $W_{R}=\langle M_{1},M_{2},M_{3}\rangle$ and $W=\langle M_{1},M_{2},M_{3},K\rangle.$

Every path $\gamma\subset F^{\circ}$ from $\lambda^{\circ}$ to one of the punctures $u_{j}^{\circ}$ $(1\leq j\leq 3)$ determines uniquely up to a sign a reflection vector $E\in R$. Since $M_{i}^{2}=1$ we get that $M_{i}(E)$ is the reflection vector corresponding to the path $\gamma\circ\ell_{i}$. We may assume that the paths $\gamma_{j}$ and $\gamma$ coincide in a sufficiently small neighborhood of the puncture $u_{j}^{\circ}$. The composition $\gamma_{j}^{-1}\circ\gamma=\ell_{i_{1}}\circ\cdots\circ\ell_{i_{r}}$. Therefore, the reflection vector corresponding to the path $\gamma=\gamma_{j}\circ\ell_{i_{1}}\circ\cdots\circ\ell_{i_{r}}$ is $M_{i_{r}}\cdots M_{i_{1}}(L^{j-1})$. Since the path $\gamma$ is arbitrary we get that $R=W_{R}\cdot 1\cup W_{R}\cdot L\cup W_{R}\cdot L^{2}$. Finally, in order to complete the proof of Corollary 2.2 we need only to use that $W_{R}$ is a normal subgroup of $W$, the monodromy transformation $K$ and $W_{R}$ generate $W$, and $K(L^{j-1})=L^{j}\in R$.

Remark 3.5**.**

One can prove that $R=\{E\in K^{0}(\mathbb{P}^{2})\ |\ \chi(E)=1\}$. In particular, the set of reflection vectors is bigger than the set formed by the K-theoretic classes of the exceptional objects in the bounded derived category $\mathcal{D}^{b}\operatorname{Coh}(\mathbb{P}^{2})$.

4. The period map for small quantum cohomology

The goal of this section is to prove Theorem 2.3. Let us assume that $t_{3}=0$ and denote by $Z(Q,\lambda)$ the value of the period map at the point $(Q,\lambda)\in X_{\rm small}$. We do not use an explicit notation, but we will always keep in mind that $Z(Q,\lambda)$ depends on the choice of a reference path.

4.1. The monodromy group of the second structure connection

Let us sketch the main steps in computing the monodromy group $W$. The matrix of the intersection form in the basis (over $\mathbb{Q}$) $E_{1},E_{2},E_{3}$ takes the form

[TABLE]

In other words the only non-vanishing pairings are $(E_{1}|E_{3})=4$ and $(E_{2}|E_{2})=-8$. Let us denote by $R_{i}$ the monodromy transformation of the basis $E=(E_{1},E_{2},E_{3})$ corresponding to analytic continuation along the path $\gamma_{i}$ (i.e. the path that turns $L^{i-1}$ into a reflection vector). We represent $R_{i}$ by a matrix such that the monodromy transformation of the row $E$ is $E\,R_{i}$. A direct computation yields

[TABLE]

The matrices $R_{i}$ ($1\leq i\leq 3)$ generate reflection group $W_{R}$ that can be embedded as a finite index subgroup of the modular groups as follows. Let

[TABLE]

be the isomorphism identifying $\mathbb{C}^{3}$ with the space of symmetric quadratic forms on $\mathbb{C}^{2}$. More precisely

[TABLE]

The modular group $\Gamma:=\operatorname{PSL}_{2}(\mathbb{Z})$ acts naturally on the space of quadratic forms

[TABLE]

Note that the above action is a right action: $q\cdot(g_{1}g_{2})=(q\cdot g_{1})\cdot g_{2}$ . Let us define a group homomorphism

[TABLE]

such that $\phi(z)\cdot g=\phi(z\rho(g))$, where $z=(z_{1},z_{2},z_{3})$ is a row vector and the matrix $\rho(g)$ acts on $z$ via matrix multiplication from the right. Explicitly

[TABLE]

Note that $R_{i}=-\rho(g_{i})$, where

[TABLE]

The monodromy group is generated by $R_{1},R_{2},R_{3}$, and $K$, where $K$ is the monodromy transformation corresponding to the analytic continuation in the $Q$-plane along a loop around $Q=0$ in clockwise direction. Recall that $K$ coincides with the operator of K-theoretic multiplication by $\mathcal{O}(1)$ (see the discussion after the proof of Proposition 2.1 in Section 3.3). Therefore $E=(E_{1},E_{2},E_{3})$ transforms into $E\,K$ where

[TABLE]

The relation between $W_{R}$, $W$, and the modular group can be described as follows. The matrices $g_{1}\kappa$ and $g_{1}$ have orders respectively $3$ and $2$ and we have $\operatorname{PSL}_{2}(\mathbb{Z})=\langle g_{1}\kappa\rangle*\langle g_{1}\rangle.$ Using this presentation of the modular group we define the characters

[TABLE]

such that

[TABLE]

Let us define a group homomorphism

[TABLE]

Using that

[TABLE]

we get that the image of the map (20) is the monodromy group $W$. It is not very difficult to check that the map is also injective, so it gives a group isomorphism $W\cong\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$. Finally, using that the reflection group $W_{R}$ is generated by $R_{1},R_{2},R_{3}$ we get that the map

[TABLE]

is a group isomorphism.

4.2. Quadratic relation

The second structure connection for small quantum cohomology takes the form

[TABLE]

and

[TABLE]

Using these equations, we get that the period map satisfies the following differential equations

[TABLE]

and

[TABLE]

where we used that in small quantum cohomology $P\bullet P\bullet P=Q$ and that $\theta(1)=1$.

The second equation implies that the period map has the form

[TABLE]

while the first equation implies that the vector valued function $z(x)$ is a solution to the hypergeometric equation of type $(3,2)$ defined by the differential operator

[TABLE]

where $D=x\partial_{x}$, $\rho_{1}=\frac{1}{3}$, $\rho_{2}=\frac{2}{3}$, and $\alpha_{1}=\alpha_{2}=\alpha_{3}=\frac{1}{6}.$

Lemma 4.1**.**

The image of the period map $Z_{\rm small}:\widetilde{X}_{\rm small}\to\mathbb{C}^{3}$ is contained in the quadratic cone $Z_{2}^{2}=4Z_{1}Z_{3}$.

Proof.

The equation of the quadratic cone coincides with

[TABLE]

where $\eta^{ij}$ are the entries of the matrix inverse to the matrix $\eta$ whose entries are the intersection numbers $\eta_{ij}:=(E_{i}|E_{j})$ (see formula (17)). Let us denote by $I^{(n)}$ the matrix whose $(i,j)$-entry is given by $(I^{(n)}_{E_{i}}(t,\lambda),\phi_{j})$. Using the equations of the second structure connection we get

[TABLE]

where

[TABLE]

On the other hand, the entries of the intersection pairing are

[TABLE]

where $g^{k\ell}$ are the entries of the matrix inverse to the matrix of the Poincare pairing. Therefore

[TABLE]

Using that $Z=(Z_{1},Z_{2},Z_{3})=(I^{(-1)}e_{1})^{T}$ we get

[TABLE]

Recalling the formula for $\eta$ from above we get

[TABLE]

On the other hand

[TABLE]

Therefore $Z\eta^{-1}Z^{T}$ is the (1,1)-entry of the matrix

[TABLE]

4.3. Connection Formula

The differential equation (21) has the following basis of solutions near $x=\infty$.

[TABLE]

where the coefficients $a_{n}$ are defined by

[TABLE]

Note that $z^{\infty}_{1}(x)$ coincides with the generalized hypergeometric function

[TABLE]

The second solution is

[TABLE]

where the constants $b_{n}$ $(n\geq 0)$ are defined by $b_{0}=0$ and

[TABLE]

Finally, the third solution is given by

[TABLE]

where the constants $c_{n}$ $(n\geq 0)$ are defined by $c_{0}=0$ and

[TABLE]

Let us find the transition matrix $C^{\infty}$ defined by

[TABLE]

where

[TABLE]

To begin with, note that the analytic continuation of $Z(Q,\lambda)$ along a clock-wise loop around [math] in the $Q$-plane corresponds to analytic continuation of $z(x)$ along an anticlock-wise loop around [math] and $1$ in the $x$-plane. Since the analytic continuation transforms $Z(Q,\lambda)$ into $Z(Q,\lambda)\,K$ where $K$ is the matrix (19) we get that $z(x)=Q^{1/6}Z(Q,\lambda)$ transforms into

[TABLE]

On the other hand, the analytic continuation of $z^{\infty}(x)$ is

[TABLE]

Therefore we have the following relation

[TABLE]

Let us denote by $C^{\infty}_{i}$ the $i$-th column of $C^{\infty}$. Then comparing the columns in the above relation we get

[TABLE]

and

[TABLE]

Therefore, we need just to find the first column of $C^{\infty}_{1}$, i.e., the coefficients in the relation

[TABLE]

The leading order term in the expansion of the RHS of (23) at $x=\infty$ is precisely

[TABLE]

By definition $z_{1}(x)=Q^{1/6}(I^{(-1)}_{E_{1}}(Q,\lambda),\mathbf{1})$, where $E_{1}=1+2L-L^{2}$. Recalling formula (14) we get that the leading order term of the expansion of the LHS of (23) coincides with the leading order term of

[TABLE]

Recalling the definitions of the period $\widetilde{I}^{(-1)}(\lambda)$ and the Iritani’s map $\Psi$ we get that the above expression is precisely

[TABLE]

Therefore,

[TABLE]

and we get

[TABLE]

4.4. Symmetric square of a hypergeometric equation

We are going to prove that the solution space of the generalized hypergeometric equation (21) has a basis of the form

[TABLE]

where $u$, and $v$ is a basis of solutions for the differential equation defined by the differential operator

[TABLE]

Note that the above operator defines a differential equation equivalent to the classical hypergeometric equation defined by the differential operator

[TABLE]

with $a=b=\frac{1}{12}$ and $c=\frac{2}{3}$.

In order to compute the symmetric square of (25) we have to find 3 functions $p_{i}(x)$ $(1\leq i\leq 3)$ such that

[TABLE]

where $z_{i}$ are given by (24) with $\{u,v\}$ a basis of solutions to (25). Using that $u$ and $v$ are solutions to (25) we can express the equations of the above linear system for $p_{0},p_{1}$, and $p_{2}$ as differential polynomials in $u$ and $v$. After a direct computation we get

[TABLE]

where

[TABLE]

and

[TABLE]

From here we get that $c_{1}=c_{2}=c_{3}=0$, so

[TABLE]

[TABLE]

and

[TABLE]

Finally, it remains only to verify that the differential operator $(1-x)^{-1}(D^{3}-p_{2}D^{2}-p_{1}D-p_{0})$ coincides with (21).

The classical hypergeometric equation (25) has a basis of solutions near $x=\infty$ of the following form

[TABLE]

and

[TABLE]

where the constants $u_{n},v_{n}$ $(n\geq 0)$ are defined by

[TABLE]

and

[TABLE]

Comparing the leading coefficients in the Laurent series expansion near $x=\infty$ we get

[TABLE]

Recalling the quadratic relation $Z_{2}^{2}=4Z_{1}Z_{3}$, we get that the period map $Z_{\rm small}$ can be expressed in terms of the hypergeometric functions $u^{\infty}$ and $v^{\infty}$ as follows:

[TABLE]

where $x=\tfrac{\lambda^{3}}{27Q}$,

[TABLE]

and

[TABLE]

where (27) and the second equality in (28) are obtained by formula (22).

4.5. The Schwarz map

The map (28) is known as the Schwarz map for the hypergeometric equation (25) (with $a=b=\tfrac{1}{12}$, $c=\tfrac{2}{3}$). The exponents of (25) at $x=0,1$, and $\infty$ are respectively $|c-a-b|=\tfrac{1}{2}$, $|1-c|=\tfrac{1}{3}$, and $|a-b|=0=\tfrac{1}{\infty}$. Therefore, the holomorphic branches of $\tau(x)$ define maps whose images are hyperbolic triangles that define a triangulation of the upper-half plane. Let us work out the precise structure of the triangulation.

Let us fix $x^{\circ}=\tfrac{(\lambda^{\circ})^{3}}{27}$ as a reference point and define the holomorphic branch of $\tau(x)$ near $x=x^{\circ}$ to be the one for which $\log x^{\circ}$ is a real number. If $x$ is in a neighborhood of [math], then the hypergeometric equation admits the basis of solutions

[TABLE]

If $x$ is in a neighborhood of $x=1$, then a basis of solution is given by

[TABLE]

The values of the above functions are defined to be real for all real $x\in(0,1)$ while for other $x\in\mathbb{C}\setminus{\{0,1\}}$ the values are specified only after we choose a reference path from $x$ to the interval $(0,1)$.

The general theory of the Gauss hypergeometric equations provides explicit formulas for the two linear transformations that relate the bases $(u^{\infty},v^{\infty})$ and $(u^{0},v^{0})$ and the bases $(u^{0},v^{0})$ and $(u^{1},v^{1})$ (see [2]). The first linear transformation takes the form

[TABLE]

where $u^{0}(x)$ and $v^{0}(x)$ are analytically extended to a neighborhood of $x=x^{\circ}$ along a path consisting of the arc $x=x^{\circ}e^{\pi\sqrt{-1}s}$ ($0\leq s\leq 1$), an interval $[-x^{\circ},-\epsilon]$ such that $0<\epsilon<1$, and the arc $x=\epsilon e^{\pi\sqrt{-1}s}$ ($-1\leq s\leq 0$). The derivation of the above formulas uses formulas (18) and (23) from Section 2.1.4 in [2] and the following identities

[TABLE]

[TABLE]

[TABLE]

[TABLE]

where $\psi(z)=\tfrac{d}{dz}\log\Gamma(z)$ is the logarithmic derivative of the Gamma function. For the last two formulas we used a formula due to Gauss that expresses the rational values $\psi(p/q)$ in terms of elementary transcendental functions (see formula (29) from Section 1.7.4 in [2]).

Similarly, using formula (1) from Section 2.10 and formulas (5) and (6) from Section 2.9 in [2] we get

[TABLE]

for all $x$ in a contractible open neighborhood of the open interval $(0,1)$.

Put $\tau^{i}(x)=\tfrac{u^{i}(x)}{v^{i}(x)}$ for $i=0,1,\infty$. By definition $\tau^{\infty}(x)+\log(1728)=-2\pi\sqrt{-1}\tau(x)$. Therefore, using the linear relation between $(u^{\infty},v^{\infty})$ and $(u^{0},v^{0})$ we get

[TABLE]

where $\rho:=e^{\tfrac{\pi\sqrt{-1}}{3}}=\tfrac{1}{2}+\tfrac{\sqrt{-3}}{2}$. This formula allows us to determine the image of the ray $(-\infty,0)$ under the Schwarz map (28). Indeed, since

[TABLE]

we get that $\operatorname{Re}(\tau(x))=-\tfrac{1}{2}$ and $\lim_{x\to-\infty}\tau(x)=-\tfrac{1}{2}+\infty\,\sqrt{-1}$. Using the relation (29), we get $\lim_{x\to 0}\tau(x)=-\rho^{-1}$. In other words, the image of $(-\infty,0)$ is the geodesic line in $\mathbb{H}$ from $-\tfrac{1}{2}+\infty\,\sqrt{-1}$ to $-\rho^{-1}$.

Furthermore, let us determine the image of the interval $(1,+\infty)$ under the Schwarz map (28). The linear relation between the bases $(u^{0},v^{0})$ and $(u^{1},v^{1})$ yields

[TABLE]

where $\alpha=\tfrac{\Gamma(5/12)}{\Gamma(11/12)}$ and $\beta=\tfrac{\Gamma(1/12)}{\Gamma(7/12)}$. If $x\to 1$ with real values, then $\tau^{1}(x)\to 0$, so using relation (31) we find that the limit $\lim_{x\to 1}\tau(x)=\sqrt{-1}$. Using (30) we get that if $x\in(1,+\infty)$, then $\operatorname{Re}(\tau(x))=0$ and $\lim_{x\to+\infty}\tau(x)=+\infty\sqrt{-1}$. The conclusion is that the image of the interval $(1,\infty)$ is the geodesic line in $\mathbb{H}$ from $\sqrt{-1}$ to $+\infty\,\sqrt{-1}$.

Finally, the image of the interval $(0,1)$ under the Schwarz map (28) must be a geodesic in $\mathbb{H}$ connecting $-\rho^{-1}$ and $\sqrt{-1}$. Therefore it must be the arc on the unit circle connecting $-\rho^{-1}$ and $\sqrt{-1}$. Using again formula (30) we fined that the real part of $\tau(\sqrt{-1}x^{\circ})$ is $-\tfrac{1}{4}$. Therefore, the Schwarz map (28) maps the upper-half plane $\operatorname{Im}(x)>0$ into the interior of the hyperbolic triangle $\Delta_{+}$ in $\mathbb{H}$ with vertices $\infty$, $-\rho^{-1}$ and $\sqrt{-1}$.

The image of the lower half-plane $\operatorname{Im}(x)<0$ can be determined by monodromy reasons. Namely, the values of the Schwarz map above and below the interval $(1,+\infty)$ vary continuously as we cross the interval, while the values above and below the intervals $(0,1)$ and $(-\infty,0)$ make a jump that can be computed if we new the monodromy of the Schwarz map around respectively the loop around $x=1$ and the loop around both $x=1$ and $x=0$. Let us compute the monodromy of the Schwarz map around these two loops. Using formula (30) we get that the analytic continuation of $\tau(x)$ along the circle $x=x^{\circ}e^{-2\pi\sqrt{-1}s}$ ($0\leq s\leq 1$) transforms $\tau(x^{\circ})$ into $\tau(x^{\circ})+1$, i.e., changing the reference path by pre-composing it with a big loop $\ell_{\infty}$ around $x=0$ and $x=1$ (with clock-wise orientation) transforms the holomorphic branch $\tau\mapsto\tau+1$. Similarly, the analytic continuation along the loop $\ell_{1}$ that starts at $x^{\circ}$ approaches $x=1$ along the real axes, goes around $x=1$ anti-clockwise along a small circle, and returns back to $x^{\circ}$ can be determined by formula (31). Indeed, such analytic continuation transforms $\tau^{1}\mapsto-\tau^{1}$. Using formula (31) after a direct computation we find that $\tau\mapsto-\tau^{-1}$. Let us summarize the conclusions of our computations.

Proposition 4.2**.**

a) The analytic continuation of the Schwarz map (28) defines a representation

[TABLE]

such that the generators $\ell_{1}$ and $\ell_{\infty}$ are mapped respectively to the transformations $\tau\mapsto-\tau^{-1}$ and $\tau\mapsto\tau+1$.

b) The image of the upper (resp. lower) half-plane $\operatorname{Im}(x)\geq 0$ (resp. $\operatorname{Im}(x)\leq 0$) under the Schwarz map is the hyperbolic triangle $\Delta_{+}$ (resp. $\Delta_{-}$) with vertices $\infty$, $-\rho^{-1}$, $\sqrt{-1}$ (resp. $\infty,\sqrt{-1},\rho$).

Note that the image of the monodromy representation (32) is the modular group $\operatorname{PSL}_{2}(\mathbb{Z})$ and that $\Delta_{+}\cup\Delta_{-}$ is its fundamental domain. Therefore all Schwarz triangles are obtained from $\Delta_{+}$ and $\Delta_{-}$ by the action of $\operatorname{PSL}_{2}(\mathbb{Z})$. The vertices of the triangulation that are inside $\mathbb{H}$ are the two orbits $\operatorname{PSL}_{2}(\mathbb{Z})\cdot\sqrt{-1}=\{\tau\in\mathbb{H}\ |\ E_{4}(\tau)=0\}$ and $\operatorname{PSL}_{2}(\mathbb{Z})\cdot\rho=\{\tau\in\mathbb{H}\ |\ E_{6}(\tau)=0\}$. The Schwarz map allows us to factorize the universal covering map $(\mathbb{C}\setminus{\{0,1\}})^{\sim}\to\mathbb{C}\setminus{\{0,1\}}$ as follows {diagram} where $\tau$ is the Schwarz map and $J$ is the regular covering corresponding to the kernel of the monodromy representation (32).

Lemma 4.3**.**

a) The function $J$ extends to a holomorphic function on $\mathbb{H}$ and it coincides with the $J$-invariant, i.e., the unique $\operatorname{PSL}_{2}(\mathbb{Z})$-invariant holomorphic function on $\mathbb{H}$ such that $J(\sqrt{-1})=1$ and the Fourier series of $J(\tau)$ has a pole of order 1 at $q=0$, where $q=e^{2\pi\sqrt{-1}\tau}$.

b) The pullback of any holomorphic branch of $(v^{\infty}(x))^{12}$ to $\mathbb{H}$ via the map $J$ extends to an analytic function on $\mathbb{H}$ and it coincides with the modular form $E_{4}(\tau)^{3}-E_{6}(\tau)^{2}$.

Proof.

a) Suppose that $\tau\in\mathbb{H}$ is not a vertex of a Schwarz triangle and let us put $x=J(\tau)$. Recall that the Schwarz triangles are biholomorphic to the upper or the lower half-plane and define a triangulation of $\mathbb{H}$. Therefore, there exists a reference path $C$ in $\mathbb{C}\setminus{\{0,1\}}$ that defines a branch $\tau_{C}$ of the Schwarz map in a neighborhood of $x$ such that $\tau_{C}(x):=\tau$. The choice of such a path is not unique, but the branch $\tau_{C}$ is independent of the choice of $C$. In other words, locally the map $J$ is defined by inverting the Schwarz map.

Let us prove that $J$ is $\operatorname{PSL}_{2}(\mathbb{Z})$-invariant. Suppose that $\tau^{\prime\prime}=g(\tau^{\prime})$ for some $g\in\operatorname{PSL}_{2}(\mathbb{Z})$. Pick a point $x^{\prime}$ and a reference path $C^{\prime}$ such that $\tau_{C^{\prime}}(x^{\prime})=\tau^{\prime}$. Let $C$ be a loop based at $x^{\circ}$ such that the monodromy representation (32) maps the homotopy class of $C$ to $g$. Then $\tau_{C\cdot C^{\prime}}(x^{\prime})=g(\tau_{C^{\prime}}(x^{\prime}))=\tau^{\prime\prime}$. In particular, $x^{\prime}=J(\tau^{\prime\prime})$. On the other hand, the identity $\tau_{C^{\prime}}(x^{\prime})=\tau^{\prime}$ implies $J(\tau^{\prime})=x^{\prime}$, so $J(\tau^{\prime})=J(\tau^{\prime\prime})=J(g(\tau^{\prime}))$.

Since $J$ is $\operatorname{PSL}_{2}(\mathbb{Z})$-invariant, in order to prove that $J$ extends to a holomorphic function on $\mathbb{H}$, it is enough to prove that it extends holomorphically in a neighborhood of $\tau=\sqrt{-1}$ and $\tau=-\rho^{-1}$. Let us give the argument for $\tau=\sqrt{-1}$. The other case is similar. If $x$ is sufficiently close to $x=1$ then substituting $\tau^{1}(x)=(1-x)^{1/2}(1+O(1-x))$ in (31) and solving for $1-x$ in terms of $\tau$, we get that $1-x=f(\tau(x))$ for some function $f(\tau)=c_{2}(\tau-\sqrt{-1})^{2}+c_{3}(\tau-\sqrt{-1})^{3}+\cdots$ holomorphic near $\tau=\sqrt{-1}$. Since $x=J(\tau(x))=1-f(\tau(x))$ for all $x$ in a neighborhood of $x=1$, we conclude that the holomorphic functions $1-f(\tau)$ and $J(\tau)$ agree on a dense subset of a neighborhood of $\tau=\sqrt{-1}$, so they must coincide identically. This proves that $J(\tau)$ is holomorphic at $\tau=\sqrt{-1}$ and moreover that $J(\sqrt{-1})=1$.

It remains only to prove that the Fourier series of $J(\tau)$ has a pole of order 1. If $x$ is sufficiently close to $\infty$ in $\mathbb{C}\setminus{\{0,1\}}$, then let us exponentiate relation (30). We get the following formula:

[TABLE]

Formula (33) can be solved for $x^{-1}$ in terms of $q$. We get that $x=1728q^{-1}+O(1)$. This completes the proof of part a).

b) Let us first check that the pullback of $(v^{\infty})^{12}$ is holomorphic on the complement in $\mathbb{H}$ of the vertices of the Schwarz triangles. The pullback of $v^{\infty}$ to the universal cover is an analytic function and the complement of the vertices of the Schwarz triangles is a quotient of the universal cover by the kernel of the monodromy representation (32). Therefore, we have to verify that $(v^{\infty}(x))^{12}$ is invariant under the analytic continuation of every loop $C$ whose homotopy class is in the kernel of the monodromy representation. On the other hand, according to Proposition 4.2, a), the monodromy representation (32) maps the loops $\ell_{1}$ and $\ell_{\infty}$ to the standard generators of the modular group $\operatorname{PSL}_{2}(\mathbb{Z})$. The relations between the standard generators are well known, namely the kernel of the monodromy representation is the normal subgroup of $\pi_{1}(\mathbb{C}\setminus{\{0,1\}},x^{\circ})$ generated by $\ell_{1}^{2}$ and $(\ell_{1}\ell_{\infty})^{3}$.

Let us work out the analytic transformation of $v^{\infty}(x)$ along the loops $\ell_{1}$ and $\ell_{\infty}$. The analytic continuation along $\ell_{\infty}$ acts on the vector column with entries $u^{\infty}$ and $v^{\infty}$ as multiplication by the matrix

[TABLE]

In particular, $\tau(x)\mapsto\tau(x)+1$ and $v^{\infty}(x)\mapsto e^{2\pi\sqrt{-1}/12}\,v^{\infty}(x).$ The analytic continuation along $\ell_{1}$ requires a long but straightforward computation. Let us point out the main steps leaving the details as an exercise. Using the linear transformation formulas between the bases $(u^{\infty},v^{\infty})$, $(u^{0},v^{0})$, and $(u^{1},v^{1})$ we find

[TABLE]

where $A=-\pi\sqrt{-1}+\log(1728)-\pi\sqrt{3}$ and $B=-\pi\sqrt{-1}+\log(1728)+\pi\sqrt{3}.$ From this formula, we find that the analytic continuation along $\ell_{1}$ acts on the column vector with entries $u^{\infty}$ and $v^{\infty}$ by multiplication by the matrix

[TABLE]

Note that

[TABLE]

Therefore, the analytic continuation along $\ell_{1}$ transforms

[TABLE]

The above formula implies that $v^{\infty}(x)$ is invariant under the analytic continuation along $\ell_{1}^{2}$. The analytic continuation along the path $\ell_{1}\ell_{\infty}$ can be represented by the following diagrams {diagram} and {diagram} Therefore, the analytic continuation along $(\ell_{1}\ell_{\infty})^{3}$ transforms $v^{\infty}(x)$ into $-v^{\infty}(x)$. We get that $(v^{\infty}(x))^{2k}$ for all integers $k\geq 0$ is invariant under the analytic continuation along $\ell_{1}^{2}$ and $(\ell_{1}\ell_{\infty})^{3}$.

Let us denote by $V(\tau)$ the pullback of $(v^{\infty}(x))^{12}$, i.e., $V$ is a holomorphic function on $\mathbb{H}\setminus{\{E_{4}(\tau)=0\}\cup\{E_{6}(\tau)=0\}}$ such that $V(\tau(x))=(v^{\infty}(x))^{12}$, where the reference path defining the branch of the Schwarz map $\tau(x)$ is chosen to be the same as the reference path specifying the branch of $v^{\infty}(x)$. The transformation formulas of $\tau(x)$ and $v^{\infty}(x)$ under the analytic continuation along $\ell_{1}$ and $\ell_{\infty}$ yield the following symmetries

[TABLE]

This implies that $V$ transforms as a modular form of weight 12. Let us check that $V$ is analytic at the cusp $\tau=\infty$. Suppose that $\tau$ belongs to a Schwarz triangle $\Delta$ with vertex at infinity. According to part a), the $J$-invariant defines a map that is inverse to the analytic branch of the Schwarz map whose image is the Schwarz triangle $\Delta$. Then $V(\tau)=v^{\infty}(J(\tau))^{12}$, so by substituting the Fourier series of the $J$-invariant $x=J(\tau)=\tfrac{1}{1728}(q^{-1}+744+O(q))$ in $v^{\infty}(x)^{12}$ we get that $V(\tau)$ has a Fourier expansion in terms of $q$ that has only positive powers. The leading order term is $V(\tau)=1728q+O(q^{2})$. Similar argument shows that $V(\tau)$ extends analytically across the vertices of the Schwarz triangles. Therefore, $V(\tau)$ is a modular form of weight 12. The space of modular forms of weight 12 has dimension 2 and it is spanned by $E_{4}(\tau)^{3}$ and $E_{6}(\tau)^{2}$. We get $V(\tau)=c_{1}E_{4}(\tau)^{3}+c_{2}E_{6}(\tau)^{2}$. Comparing the coefficients in the corresponding Fourier series in front of $q^{0}$ and $q^{1}$ we get $c_{1}=1$ and $c_{2}=-1$. ∎

4.6. Proof of Theorem 2.3, a)

Let us recall formula (26). If $\lambda\neq 0$, then $-Z_{2}(Q,\lambda)/(2Z_{3}(Q,\lambda))=\tau(x)$ is a point in the uper-half plane satisfying the condition $J(\tau(x))=x=\tfrac{\lambda^{2}}{27Q}\neq 1$, i.e., $Z_{\rm small}(Q,\lambda)\in\Omega_{\rm small}$. Since the period map is locally analytic near $\lambda=0$, its value $Z_{\rm small}(Q,0)=\lim_{n\to\infty}Z_{\rm small}(Q,\lambda_{n})$ can be computed by choosing any convergent sequence $\lambda_{n}\to 0$. For example, let us take $\lambda_{n}$ to be real. The Schwarz map $\tau(\lambda_{n})$ has a limit $\tau_{0}$ whose value depends on the reference path, but in any case $\tau_{0}$ is a vertex of a Schwarz triangle and $J(\tau_{0})=\lim_{n}J(\tau(\lambda_{n}))=\lim\lambda_{n}=0$, i.e., $\tau_{0}$ belongs to the orbit $\operatorname{PSL}_{2}(\mathbb{Z})\cdot\rho$. This completes the proof that $Z_{\rm small}:\widetilde{X}_{\rm small}\to\Omega_{\rm small}\setminus{\{E_{6}(-z_{2}/(2z_{3}))=0\}}$. We have to check that this map is surjective.

Let us define the analytic isomorphism

[TABLE]

Then we have to prove that the following map is surjective

[TABLE]

Given a point $(\tau,y)\in\mathbb{H}\times\mathbb{C}^{*}$ we first pick $(Q,\lambda)$ such that $\lambda^{3}/(27Q)=J(\tau)$ and then we fix $Q$ in such a way that $Q^{1/6}z_{3}(\lambda^{3}/27Q)=y$. The surjectivity follows. ∎

4.7. Proof of Theorem 2.3, b)

Recall that the period map $Z_{\rm small}(Q,\lambda)$ has the form (26). According to Proposition 32 if we put $\tau=-\tfrac{Z_{2}(Q,\lambda)}{2Z_{3}(Q,\lambda)}$, then we have

[TABLE]

where $x:=\tfrac{\lambda^{3}}{27Q}$. Therefore,

[TABLE]

The formula for $\lambda$ follows from the relation

[TABLE]

which implies that

[TABLE]

where $\xi^{3}=1$. On the other hand, $Z_{3}(Q,\lambda)^{2}\lambda=8\pi^{2}+O(\lambda^{-1})$, so $\xi=1$.

4.8. Proof of Theorem 2.3, c)

Let us first identify the period domain $\Omega_{\rm small}$ with $\mathbb{H}\times\mathbb{C}^{*}$ via the isomorphism (34). This identification will induce an action of the monodromy group $W$ of quantum cohomology on $\mathbb{H}\times\mathbb{C}^{*}$. Let us work out this action explicitly. Let us define a left action of $\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$ on $\mathbb{H}\times\mathbb{C}^{*}$ by

[TABLE]

Lemma 4.4**.**

Let $w\in W\subset\operatorname{GL}(\mathbb{C}^{3})$ be the monodromy transformation corresponding to an element $(g,\sigma)\in\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$ via the group homomorphism (20). Then

[TABLE]

where $s=\begin{bmatrix}0&1\\ 1&0\end{bmatrix}$ and $(\tau,y)\in\mathbb{H}\times\mathbb{C}^{*}.$

Proof.

Put $z:=\Phi_{\rm small}(\tau,y)$, then by definition $\tau=\tau(z):=-z_{2}/(2z_{3})$ and

[TABLE]

After a straightforward computation we get

[TABLE]

where we used that $z_{1}/z_{3}=z_{1}z_{3}/z_{3}^{2}=\tau^{2}$. It remains only to use that

[TABLE]

and $\chi_{2}(sg^{-1}s^{-1})=\chi_{2}(g)$. ∎

The proof of part c) can be completed as follows. Suppose that $\pi_{\rm small}(\tau,y)=\pi_{\rm small}(\tau^{\prime},y^{\prime})$. Since $J(\tau)=J(\tau^{\prime})$, there exists $g\in\operatorname{PSL}_{2}(\mathbb{Z})$ such that $\tau^{\prime}=g(\tau)$. There are 2 cases. First, if $J(\tau)\neq 0$, then $E_{4}(\tau)\neq 0$ and we get

[TABLE]

so

[TABLE]

for some $\sigma^{\prime}\in\{\pm 1\}$. Defining $\sigma=\sigma^{\prime}\chi_{2}(g)$ we get $(\tau^{\prime},y^{\prime})=(g,\sigma)\cdot(\tau,y)$.

The second case is the case when $J(\tau)=0$. Such a $\tau$ is a vertex in the triangulation by Schwarz triangles and it is in the same $\operatorname{PSL}_{2}(\mathbb{Z})$-orbit as $\zeta=e^{\tfrac{2\pi\sqrt{-1}}{3}}$. We may assume that $\tau=\tau^{\prime}=\zeta$, because the point $(\tau,y)$ (resp. $(\tau^{\prime},y^{\prime})$) is in the $W$-orbit of a point of the form $(\zeta,\widetilde{y})$ (resp. $(\zeta,\widetilde{y^{\prime}})$). Using that

[TABLE]

we get that $y^{\prime}=\xi y$, for some $\xi$ such that $\xi^{6}=1$. Note that the stabilizer of $\zeta\in\mathbb{H}$ in the modular group is a cyclic group of order 3 generated by $g_{1}\kappa$, where the matrices $g_{1},\kappa\in\operatorname{PSL}_{2}(\mathbb{Z})$ are defined in Section 4.1. We have

[TABLE]

where $i$ is an integer. Since we can always choose $\sigma\in\{\pm 1\}$ and $i$ such that $\sigma^{i}\zeta^{2i}=\xi$ we get that $(\tau,y)$ and $(\tau,y^{\prime})$ are in the same $W$-orbit. ∎

5. Holomorphic thickening

Let us return to the general case of the period map for the big quantum cohomology. Recall that

[TABLE]

and

[TABLE]

Let us introduce coordinates $(Q,t,\lambda)$ on $X$ such that $Q:=e^{t_{2}}$, $t:=t_{3}$. We fix a base point $y^{\circ}=(Q^{\circ},\lambda^{\circ})$ in $X_{\rm small}$, which will be used as a base point of $X$ as well. Given a point $(Q,\lambda)\in X_{\rm small}$ we will be interested in the Taylor’s series expansion

[TABLE]

We are going to construct a covering of $X$ such that the pullback of $Z^{(n)}(Q,\lambda)$ is a quasi-modular form. This would allow us to find the inverse of the period map and generalize to some extend the statement of Theorem 2.3.

5.1. Auxiliary covering

Let $\mathcal{U}\subset\mathbb{H}\times\mathbb{C}^{*}\times\mathbb{C}$ be an open subset and let

[TABLE]

be the holomorphic map defined by

[TABLE]

Remark 5.1**.**

So far the variable $x$ was used to denote the coordinate on the domain $\mathbb{C}\setminus{\{0,1\}}$ of the Schwarz map. We will no longer deal with the Schwarz map. From now on we will use $x$ to denote the coordinate on the complex circle $\mathbb{C}^{*}\subset\mathcal{U}$, except for some local proofs where the demand for letters is hard to meet without using $x$.

We choose $\mathcal{U}$ to be the trivial disk bundle

[TABLE]

where

[TABLE]

is a smooth function defined as follows. We choose $\delta$ in such a way that the preimage under $\pi^{\rm aux}$ of the discriminant is the analytic hypersurface $E_{6}(\tau)=0$. More precisely, the equation of the discriminant has the form

[TABLE]

where $g\in\mathcal{O}(B)$ is some holomorphic function. If $(t,Q,\lambda)=\pi^{\rm aux}(\tau,x,s)$, then the above equation becomes

[TABLE]

For fixed $(\tau,x)\in\mathbb{H}\times\mathbb{C}^{*}$ we choose $\delta(\tau,x)$ such that $\pi^{\rm aux}(\tau,x,s)\in B$ and $|s\,g\circ\pi^{\rm aux}(\tau,x,s)|<8(2\pi)^{6}$ for all $|s|<\delta(\tau,x)$.

Lemma 5.2**.**

If we choose the constant $\epsilon$ in the definition of the domain $B$ sufficiently small, then the subvariety $X_{\rm small}$ is a deformation retract of $X$.

Proof.

A deformation retraction

[TABLE]

can be taken in the form

[TABLE]

where $u_{i}(Q,t)$ ($1\leq i\leq 3$) are the eigenvalues of the quantum multiplication by $E\bullet_{Q,t}$, i.e., the canonical coordinates. Note that $u_{i}(Q,t)=Q^{1/3}u_{i}(1,tQ^{1/3})$ and $u_{i}(Q,0)=3\zeta^{i}Q^{1/3}$, where $\zeta=e^{2\pi\sqrt{-1}/3}$. Given a real number $\delta>0$ we can always choose $\epsilon$ sufficiently small so that $|u_{i}(1,tQ^{1/3})-u_{i}(1,0)|<\delta$ for all $(Q,t)$, s.t., $|tQ^{1/3}|<\epsilon$. We claim that if we choose $\delta<\frac{3}{16}|1-\zeta|$ and $\rho_{i}$ ($1\leq i\leq 3$) to be smooth functions such that $\rho_{i}(x)=1$ for all $|x-3\zeta^{i}|<4\delta$ and $\rho_{i}(x)=0$ for all $|x-3\zeta^{i}|>8\delta$, then formula (35) defines a deformation retract, i.e., a homotopy between the identity map and a retraction $X\to X_{\rm small}$.

Clearly we have $\Psi(x,0)=x$ for all $x\in X$ and $\Psi(y,s)=y$ for all $y\in X_{\rm small}$ and for all $s$. We have to verify that $\Psi(Q,t,\lambda,s)$ is not a point on the discriminant. There are two cases. First, if $|\lambda Q^{-1/3}-3\zeta^{i}|>8\delta$ for all $i$, then $\Psi(Q,t,\lambda,s)=(Q,\lambda,(1-s)t)$. We have

[TABLE]

so $\lambda\neq u_{j}(Q,(1-s)t)$ for all $j$, i.e., $(Q,\lambda,(1-s)t)\in X$. The second case is if $|\lambda Q^{-1/3}-3\zeta^{i}|\leq 8\delta$ for some $i$. Note that if $j\neq i$, then

[TABLE]

Therefore the second component of $\Psi(Q,t,\lambda,s)$ is

[TABLE]

We have to prove that the above number does not coincide with $u_{j}(Q,(1-s)t)$ for all $j$. Let us assume that this is not the case, i.e., the number coincides with $u_{j}(Q,(1-s)t)$ for some $j$. Using the estimate

[TABLE]

we get that we must have $j=i$ and $\rho_{i}(\lambda Q^{-1/3})=1$. Therefore our assumption implies that

[TABLE]

This however contradicts the fact that $(Q,t,\lambda)\in X$. ∎

Proposition 5.3**.**

a) Let $\pi^{\prime}:\mathcal{U}^{\prime}=(\pi^{\rm aux})^{-1}(X)\to X$ be the map induced from $\pi^{\rm aux}$. Then the period map admits a holomorphic lift $Z^{\rm aux}:\mathcal{U}^{\prime}\to\mathbb{C}^{3}$.

b) The map $Z^{\rm aux}$ extends holomorphically on the entire domain $\mathcal{U}$.

Proof.

a) Note that our definition of $\mathcal{U}$ implies that $\mathcal{U}^{\prime}=\mathcal{U}-\{E_{6}(\tau)=0\}$. Let us define an action of the monodromy group $W=\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$ on $\mathcal{U}$

[TABLE]

Note that the points with non-trivial stabilizers are given by the analytic hypersurfaces $\{E_{4}(\tau)=0\}$ and $\{E_{6}(\tau)=0\}$. Let $u^{\circ}=(\tau^{\circ},x^{\circ},0)\in\mathcal{U}^{\prime}$ be a reference point, such that $E_{4}(\tau^{\circ})\neq 0$ and $\pi^{\rm aux}(u^{\circ})=y^{\circ}$.

Let us construct a lift $Z^{\prime}$ of the period map on $\mathcal{U}^{\prime}$. If $u=(\tau,x,t)\in\mathcal{U}^{\prime}$, then we pick a reference path $\gamma\subset\mathcal{U}^{\prime}$ and define $Z^{\prime}(u)=Z(\pi^{\rm aux}(u))$ where the value of $Z(\pi^{\rm aux}(u))$ is defined via the reference path $\pi^{\rm aux}(\gamma)$. We claim that choosing a different reference path $\gamma^{\prime}\subset\mathcal{U}^{\prime}$ does not change the value of $Z^{\prime}(u)$. In other words we claim that if $L\in\pi_{1}(\mathcal{U}^{\prime},u^{\circ})$ is a loop based at $u^{\circ}$, then the image $\pi^{\rm aux}(L)$ is in the kernel of the monodromy representation ${\rm per}:\pi_{1}(X)\to W$. By making a small perturbation (without changing the homotopy class) we can arrange that $L$ is a loop in $\mathcal{U}^{\prime}-\{E_{4}=0\}$. Note that the projections $r^{\prime}:(\tau,x,s)\mapsto(\tau,x)$ and $r^{\prime\prime}:(Q,t,\lambda)\mapsto(Q,\lambda)$ give rise to a commutative diagram

[TABLE]

where $\pi^{\prime}$ and $\pi^{\prime\prime}$ are the maps induced from $\pi^{\rm aux}$,

[TABLE]

and the horizontal arrows are deformation retractions. The map $\pi^{\prime\prime}$ induces a covering

[TABLE]

According to Theorem 2.3, locally the inverse of the covering map (36) is given explicitly by the following formulas

[TABLE]

Therefore we have a commutative diagram in which the horizontal arrow is induced from the natural inclusion $X_{\rm small}^{\prime}\subset X_{\rm small}$ and the two diagonal arrows are given by the monodromy representations respectively of the covering and the period maps. On the other hand the lift of the loop $r^{\prime\prime}\circ\pi^{\prime}(L)$ is $r^{\prime}(L)$, which is a loop, so the corresponding monodromy transformation

[TABLE]

fixes the reference point $(\tau^{\circ},x^{\circ})\in\mathcal{U}^{\prime}$. Therefore $w=1$, because the stabilizer of $(\tau^{\circ},x^{\circ})$ is trivial. We get that the homotopy class of $r^{\prime\prime}\circ\pi^{\prime}(L)$ in $\pi_{1}(X_{\rm small},y^{\circ})$ is in the kernel of the monodromy representation of the period map. Using that $r^{\prime\prime}$ is a deformation retract, we get that $\pi^{\prime}(L)$ is homotopic to $r^{\prime\prime}\circ\pi^{\prime}(L)$ in $X^{\prime}$. Finally, since $X_{\rm small}$ is a deformation retract of $X$ (see Lemma 5.2) we get that the homotopy class of $\pi^{\prime}(L)$ in $\pi_{1}(X,y^{\circ})$ must be in the kernel of the monodromy representation of the period map.

b) It remains only to prove that $Z^{\prime}$ extends analytically to the entire domain $\mathcal{U}$. The complement of $\mathcal{U}^{\prime}$ in $\mathcal{U}$ is an analytic hypersurface. Recalling the Riemann extension theorem, we get that it is sufficient to prove that the values of $Z^{\prime}(u)$ are bounded in a neighborhood of an arbitrary point $u_{0}\in\mathcal{U}-\mathcal{U}^{\prime}$. Note that $\pi^{\rm aux}(u_{0})=:(Q_{0},t_{0},\lambda_{0})$ is a point on the discriminant. Then by definition the periods $I^{(-1)}(Q,t,\lambda)\sim(\lambda-u)^{1/2}$ where $\lambda=u$ is the local equation of the discriminant near the point $(Q_{0},t_{0},\lambda_{0})$. Therefore, the map $Z^{\prime}$ is bounded. ∎

5.2. The Taylor’s coefficients $Z^{(n)}$

Recall the notation in the proof of Lemma 4.1. Let us denote by $I^{(n)}$ the matrix whose $(i,j)$ entry is $(I^{(n)}_{E_{i}},p^{j-1})$. We claim that the matrix $I^{(-1)}$ can be expressed in terms of the Wronskian matrix

[TABLE]

Indeed, put

[TABLE]

The differential equation of the second structure connection can be written us

[TABLE]

where $\Omega_{i}=P^{i-1}\bullet$ is the matrix of quantum multiplication by $P^{i-1}$. We get

[TABLE]

Therefore, $\operatorname{Wr}=I^{(-1)}T$, where $T$ is the matrix with columns

[TABLE]

The proves of the next two Lemmas involve some long computations. Although they could be done by hand within acceptable amount of time, we recommend the use of a computer software such as Mathematica or Maple.

Lemma 5.4**.**

The matrix $T$ can be expressed in terms of the genus 0 potential $F$ as follows

[TABLE]

where $\Delta=\operatorname{det}(\lambda-E\bullet)$,

[TABLE]

[TABLE]

and

[TABLE]

The homogeneous degree with respect to $Q,t,\lambda$ of the $i$th row of $T$ is $1-i$.

Proof.

Using that

[TABLE]

and that $F$ is homogeneous of degree 1 we get

[TABLE]

We can express $A$ in terms of the partial derivatives of $F$ and after some long but straightforward computation we get the formulas stated in the Lemma. ∎

Another long but straightforward computation yields that

[TABLE]

where

[TABLE]

Lemma 5.5**.**

a) There exists an operator

[TABLE]

whose coefficients are rational functions in $\lambda$ depending analytically on $(Q,t)\in B$ such that

[TABLE]

b) The coefficients $L_{i}(Q,t,\lambda)$ have the form

[TABLE]

where $\ell_{i}$ is a polynomial in $\lambda$ of degree $2+i$ whose coefficients are polynomials in the partial derivatives of $F$. Moreover, the weigh of $\ell_{i}$ with respect to the variables $Q,t,\lambda$ is 4.

Proof.

Let $Z$ be the column with entries $Z_{1},Z_{2},Z_{3}$ and $\{e_{i}\}_{i=1}^{3}\subset\mathbb{C}^{3}$ be the standard basis. We have

[TABLE]

The 3rd column $Ae_{3}$ of the matrix $A$ can be expressed in terms of the partial derivatives of $F$ as explained above. Therefore the coefficients of the differential operator are given by

[TABLE]

The rest of the proof is a straightforward computation. ∎

Let us point out that at $t=0$ we have

[TABLE]

and

[TABLE]

The differential operator takes the form

[TABLE]

Lemma 5.6**.**

At $t=0$ the period map satisfies the following differential equation

[TABLE]

Proof.

We just need to check that the substitution $Z_{i}(Q,\lambda)=Q^{-1/6}z_{i}(x),$ with $x=\lambda^{3}/(27Q)$ transforms the above differential equation into the generalized hypergeometric equation (21). This however is a straightforward computation. ∎

We would like to change the coordinates $(Q,\lambda)\in\mathbb{C}^{*}\times\mathbb{C}$ using the covering map $\pi$ in Theorem 2.3, i.e.,

[TABLE]

Recall the Ramanujan’s differential equations for the Eisenstein series

[TABLE]

Lemma 5.7**.**

Under the change of coordinates (39) we have

[TABLE]

Proof.

After a short computation we get

[TABLE]

The formula for $Q\partial_{Q}$ is easy to derive from here. ∎

Proposition 5.8**.**

Under the change of coordinates (39) we have

a) The Taylor coefficient $Z_{3}^{(0)}(Q,\lambda)=x$ and

[TABLE]

b) We have $Z_{2}^{(0)}(Q,\lambda)=-2\tau x$ and

[TABLE]

Proof.

Using Lemma 5.5 and 5.6 we get that

[TABLE]

where $M^{(n)}$ is a second order differential operator of the form

[TABLE]

whose coefficients are rational functions in $Q$ and $\lambda$ with poles only at $\lambda=0$ and $\lambda^{3}-27Q=0$. According to Theorem 2.3, under the change of coordinates (39) we have

[TABLE]

Using Lemma 5.7 we also have

[TABLE]

where

[TABLE]

Note that in the above formula for $M^{(n)}$ we have replaced $x\partial_{x}$ with 1, because $x\partial_{x}$ commutes with $L$ and it acts on $Z(Q,\lambda)$ by multiplication by 1. Hence

[TABLE]

and

[TABLE]

The statements of part a) and b), modulo the order of the poles at $E_{4}=0$ and $E_{6}=0$, follows from the fact that $Z^{(n)}$ is homogeneous of degree $n-\frac{1}{2}$, $\tau$ has degree 0 and $x$ has degree $-\frac{1}{2}$. The statement that $Z^{(n)}_{3}$ and $Z^{(n)}_{2}+2\tau Z^{(n)}_{3}$ do not have a pole at $E_{4}=0$ and have a pole of order at most $2n$ at $E_{6}=0$ follows from Proposition 5.3. Indeed, according to the Proposition the series

[TABLE]

is convergent for all $(\tau,x,s)\in\mathcal{U}$, so in particular the coefficient in front of $s^{n}$ must be holomorphic for all $(\tau,x)\in\mathbb{H}\times\mathbb{C}^{*}$ and for all $n\geq 0$. ∎

The first component $Z_{1}(Q,t,\lambda)$ of the period map is determined from the remaining two via the following relation.

Lemma 5.9**.**

We have

[TABLE]

Proof.

The argument is the same as in the proof of Lemma 4.1. Namely, we have

[TABLE]

and the same argument as in Lemma 4.1 proves that the LHS is the $(1,1)$-entry of the matrix

[TABLE]

The entries of $\lambda-E\bullet$ can be expressed in terms of the partial derivatives of the genus zero potential $F$

[TABLE]

Note that the $(1,1)$-entry of $g(\lambda-E\bullet)$ is $t$, so the $(1,1)$-entry of (40) is $4t$. ∎

5.3. Extension of the period domain

Recall that we have identified $\mathbb{C}^{3}$ with the space of quadratic forms in two variables (see (18)). Let us define an open neighborhood of $\Omega_{\rm small}$ in $\mathbb{C}^{3}$ as the image of the following map:

[TABLE]

Recalling the definition of $\phi$ we get that

[TABLE]

Let us equip $\mathbb{H}^{2}\times\mathbb{C}^{*}$ with a left $W$-action. If $w=(g,\sigma)\in W=\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$, then we define

[TABLE]

Lemma 5.10**.**

Let $w\in\operatorname{GL}(\mathbb{C}^{3})$ be the monodromy transformation corresponding to an element $(g,\sigma)\in\operatorname{PSL}_{2}(\mathbb{Z})\times\{\pm 1\}$ via the map (20). Then

[TABLE]

where $s=\begin{bmatrix}0&1\\ 1&0\end{bmatrix}$ and $(\tau_{1},\tau_{2},y)\in\mathbb{H}^{2}\times\mathbb{C}^{*}.$

Proof.

We prove that the quadratic forms corresponding to the LHS and the RHS (of the identity that we have to prove) coincide. The quadratic form corresponding to the LHS is

[TABLE]

where $g=\begin{bmatrix}a&b\\ c&d\end{bmatrix}$. Note that

[TABLE]

and $\chi_{2}(sg^{-1}s^{-1})=\chi_{2}(g)$. Therefore, the quadratic form corresponding to the RHS is

[TABLE]

It remains only to verify that the above formula coincides with (42). ∎

5.4. Proof of Theorem 2.4

We would like to invert the period map

[TABLE]

i.e., express $(Q,t,\lambda)$ in terms of $(z_{1},z_{2},z_{3})$. Recall that

[TABLE]

Since the inverse of $\Phi$ is straightforward to find, it is sufficient to find the relation between the coordinate systems $(\tau,x,t)$ and $(\tau_{1},\tau_{2},y)$.

To begin with note that according to Lemma 5.9 we have

[TABLE]

According to Proposition 5.8 we have

[TABLE]

and

[TABLE]

where $y_{n},\widetilde{\tau}_{12,n}\in\mathbb{C}[E_{2},E_{4},E_{6}]E_{6}^{-2n}$. Using the formula for $y$ we can express $x$ in terms of $y,t,$ and $E_{i}$ ($i=2,4,6$):

[TABLE]

where $x_{n}\in\mathbb{C}[E_{2},E_{4},E_{6}]E_{6}^{-2n}$. Substituting this into the formula for $\tau_{12}$ we get

[TABLE]

where $\tau_{12,n}\in\mathbb{C}[E_{2},E_{4},E_{6}]E_{6}^{-2n}.$ Using Taylor series expansion at $\tau_{12}=\tau$ and the Ramanujan’s differential equations. We get

[TABLE]

Therefore, we can express $E_{i}(\tau)$ ($i=2,4,6$) in terms of $E_{i}(\tau_{12})$ ($i=2,4,6$)

[TABLE]

Since $ty^{-2}=-(\tau_{1}-\tau_{2})^{2}/32$ we get inversion formulas of the following type

[TABLE]

where $Q_{n}$ and $\lambda_{n}$ are polynomial expression in $E_{2}(\tau_{12}),E_{4}(\tau_{12})$, and $E_{6}(\tau_{12})^{\pm 1}$. We have to prove that $Q_{n}$ and $\lambda_{n}$ depend polynomially on $E_{i}$, i.e., there is no negative powers of $E_{6}$. This follows from the fact that the period map for the second structure connection is locally invertible.

Lemma 5.11**.**

a) The value of the Jacobian determinant

[TABLE]

at $t=0$ up to a non-zero constant coincides with $(\lambda^{3}-27Q)^{-1/2}$.

b) The value of the Jacobian determinant

[TABLE]

at $s=0$ up to a non-zero constant coincides with $E_{6}(\tau)^{2}x^{-6}Q(\lambda^{3}-27Q)^{1/2}$.

Proof.

We will use the notation from section 5.2. Using the differential equations of the second structure connection we get

[TABLE]

Therefore

[TABLE]

The above expression should be evaluated at $t=0$. We get

[TABLE]

By definition $\operatorname{Wr}=I^{(-1)}T$ and the $T(Q,0,\lambda)$ is given by (38). Therefore,

[TABLE]

The Jacobian determinant takes the form

[TABLE]

Recall that $Z_{i}(Q,\lambda)=Q^{-1/6}z_{i}(x),$ with $x=\lambda^{3}/(27Q)$. Therefore, the Wronskian determinant takes the form

[TABLE]

i.e.,

[TABLE]

On the other hand, $\{z_{i}\}_{i=1}^{3}$ form a basis of solutions of the hypergeometric equation (21). Therefore, the above determinant can be expressed easily in terms of the Wronskian of the differential equation. After a short computation we get

[TABLE]

where $C$ is a non-zero constant. The precise value of $C$ is irrelevant, but for the sake of completeness, let us compute it. Using the connection formulas in Section 4.3, we can compute the leading order term of the Wronskian near $x=\infty$

[TABLE]

We get $C=-\frac{16}{3\sqrt{6}}$. Finally, for the Jacobian determinant we get

[TABLE]

b) This is an elementary consequence of Ramanujan’s differential equations. ∎

Proposition 5.12**.**

The coefficients

[TABLE]

i.e., they are quasi-modular forms with respect to $\tau_{12}\in\mathbb{H}$.

Proof.

Using Proposition 5.3 and Lemma 5.11, b) we get that the map $Z^{\rm aux}:\mathcal{U}\to\mathbb{C}^{3}$ induces an isomorphism between an open neighborhood in $\mathbb{H}\times\mathbb{C}^{*}\times\mathbb{C}$ of

[TABLE]

and an open neighborhood in $\mathbb{C}^{3}$ of

[TABLE]

In particular, the coordinates $(\tau,x,s)$ and $(z_{1},z_{2},z_{3})$ of the two neighborhoods are biholomorphic. Therefore

[TABLE]

define functions that are holomorphic in an open neighborhood in $\mathbb{C}^{3}$ of $\Omega_{\rm small}^{\prime}$. Note that by definition the functions $(Q,t,\lambda)$ give an inversion of the period map. More precisely, if $(Q^{*},t^{*},\lambda^{*})\in X$ with $t^{*}=0$ and $z^{*}=Z(Q^{*},t^{*},\lambda^{*})\in\Omega_{\rm small}$ is a value of the period map (depending on the choice of a reference path), then in a neighborhood of $z^{*}\in\mathbb{C}^{3}$ the functions $(Q,t,\lambda)$ coincide with the unique solution to the equations

[TABLE]

where the branch of $Z_{i}$ is fixed by $Z_{i}(Q^{*},t^{*},\lambda^{*})=z_{i}^{*}$.

Clearly $t$ is a holomorphic function on $\mathbb{C}^{3}$. We claim that $\lambda$ and $Q$ extend to holomorphic functions defined in a neighborhood of $\Omega_{\rm small}$ in $\mathbb{C}^{3}$. The statement is local, so let $(Q^{*},t^{*},\lambda^{*})$ with $t^{*}=0$ be a point on the discriminant and let $z^{*}=Z(Q^{*},t^{*},\lambda^{*})\in\Omega_{\rm small}$ be a value of the period map. Let $\lambda=u(Q,t)$ be the local equation of the discriminant at the point $(Q^{*},t^{*},\lambda^{*})$. Locally, the components of the period map can be written as

[TABLE]

where $\alpha\in H$ is a vector whose local monodromy around the discriminant is given by $\alpha\mapsto-\alpha$, $Z_{\alpha}:=\langle Z,\alpha\rangle$, and $Z_{i}^{\rm inv}$ corresponds to the invariant part of $E_{i}$, i.e.,

[TABLE]

On the other hand

[TABLE]

where $\widetilde{Z}_{\alpha}(Q,t,\lambda)$ is holomorphic in a neighborhood of $(Q^{*},t^{*},\lambda^{*})$ and $\widetilde{Z}_{\alpha}(Q^{*},t^{*},\lambda^{*})\neq 0$. Let us choose $i$ such that $(E_{i}|\alpha)\neq 0$. Then we get

[TABLE]

The above equation defines a branched double covering of a neighborhood of $(Q^{*},t^{*},\lambda^{*})$ and $(Q,t,\mu)$ is a holomorphic coordinate system on the double cover. We claim that the local lift of the period map is an isomorphism. Indeed, the local lift is a single valued analytic map, because

[TABLE]

We have to check that the corresponding Jacobian determinant does not vanish at the point $(Q^{*},t^{*},\mu^{*})=(Q^{*},0,0).$ Using

[TABLE]

the chain rule, and Lemma 5.11, a) we get

[TABLE]

where $C$ is a non-zero constant, $\zeta=e^{2\pi\sqrt{-1}/3}$, and we used that

[TABLE]

Therefore $Q,t,$ and $\mu$ are holomorphic in a neighborhood of $z^{*}\in\mathbb{C}^{3}$, which implies that $Q,t,$ and $\lambda$ are also holomorphic.

To complete the proof of the proposition we note that

[TABLE]

Therefore, $Q,t,$ and $\lambda$ must be holomorphic functions in $y,\tau_{12},(\tau_{1}-\tau_{2})^{2}$. In particular $Q_{n}(\tau_{12})$ and $\lambda_{n}(\tau_{12})$ must be holomorphic in $\tau_{12}$, so the corresponding polynomial expressions in $E_{2}(\tau_{12})$, $E_{4}(\tau_{12})$, and $E_{6}(\tau_{12})^{\pm 1}$ could not have negative powers of $E_{6}(\tau_{12})$. ∎

5.5. The ring of modular functions

Note that the ring $\Gamma(\Omega_{\rm small},\mathcal{O}_{\mathbb{C}^{3}})$ is equipped with the action of the monodromy group $W$. Let $\Gamma(\Omega_{\rm small},\mathcal{O}_{\mathbb{C}^{3}})^{W}$ be the ring of $W$-invariant functions. We introduce a subring of $W$-invariant functions as follows. Let $\mathbb{C}[Q,\lambda]\{t\}$ be the ring of power series in $t$ whose coefficients depend polynomially on $Q$ and $\lambda$ and such that for every $(Q,\lambda)\in\mathbb{C}^{*}\times\mathbb{C}$ the radius of convergence is non-zero. Then we define

[TABLE]

Proposition 5.12 implies that the tautological map

[TABLE]

is an isomorphism. Although the invariant functions corresponding to $Q$ and $\lambda$ can be find recursively, it will be nice to have a more intrinsic characterization. Unfortunately we could not achieve this goal. On the other hand we have managed to find explicitly invariant functions $E_{4}^{(2)},\Delta^{(2)}$ that generate $\mathcal{M}(\Omega,W)$ in the following sense. If $f\in\mathcal{M}(\Omega,W)$ then

[TABLE]

where $c_{n}\in\mathbb{C}[E_{4}^{(2)},\Delta^{(2)}]$ are some polynomials. The above equality should be interpreted as equality between formal power series in $t$.

In order to find such functions $E_{4}^{(2)}$ and $\Delta^{(2)}$ it is enough to construct two $W\times\mu_{2}$-invariant holomorphic functions in $\mathbb{H}^{2}\times\mathbb{C}^{*}$ whose restrictions to $\mathbb{H}\times\mathbb{C}^{*}$ coincide with

[TABLE]

This could be done easily using the Jacobi theta constants

[TABLE]

For the reader’s convenience we have recorded in Table 1 the transformation rules for the theta constants under the two modular transformations $\tau\mapsto\tau+1$ and $\tau\mapsto-1/\tau$. For more details we refer to [17]. It is easy to check that

[TABLE]

Let us define

[TABLE]

and

[TABLE]

It is straightforward to check that $E^{(2)}_{4}$ and $\Delta^{(2)}$ are $W\times\mu_{2}$-invariant holomorphic functions on $\mathbb{H}^{2}\times\mathbb{C}^{*}$, so they define $W$-invariant analytic functions on the domain $\mathbb{H}^{2}\times\mathbb{C}^{*}/\mu_{2}$. In particular $E^{(2)}_{4},\Delta^{(2)}\in\mathcal{M}(\Omega,W)$.

Appendix A Genus-0 constraints

Suppose that we are in the settings of Sections 1.1 and 1.2. Following [7] we will assume in addition that the Frobenius structure arises from a set of gravitational descendants. The latter are organized into a generating function $\mathcal{F}^{(0)}(\mathbf{t})$, where $\mathbf{t}=(t_{k,i})_{k=0,1,2,\dots}^{i=1,2,\dots,N}$ is a set of formal variables, satisfying the following 3 axioms.

1. (DE)

Dilaton Equation:

[TABLE] 2. (SE)

String Equation:

[TABLE]

where $t_{0}=\sum_{i=1}^{N}t_{0,i}\phi_{i}$. 3. (TRR)

Topological Recursion Relations:

[TABLE]

where $g_{ij}=(\phi_{i},\phi_{j})$ is the matrix of the Frobenius pairing and $g^{ij}$ are the entries of the inverse matrix.

Finally, we are going to consider only the case when the intersection pairing is non-degenerate.

A.1. The Heisenberg vertex operator algebra

In this section we recall the main construction from [1]. Although the work in [1] is in the settings of simple singularities the generalisation to an arbitrary semi-simple Frobenius manifold is straightforward (see [14], Section 5). Let us equip the vector space $H[s,s^{-1}]\oplus\mathbb{C}$ with the structure of a Heisenberg Lie algebra such that

[TABLE]

Let $\mathfrak{F}:=\operatorname{Sym}(H[s^{-1}]s^{-1})$ be the corresponding Fock space, i.e., the unique irreducible highest weight representation with highest weight vector $1$ defined by

[TABLE]

Following [1] we introduce the W-algebra $\mathcal{W}\subset\mathfrak{F}$ as the kernel of all screening operators

[TABLE]

where the linear operators $e^{\alpha}_{(n)}$ are defined as the Fourier coefficients of the following vertex operator:

[TABLE]

where $|\alpha|^{2}=(\alpha|\alpha)$. The vector space $\mathfrak{F}$ is equipped with the structure of a Vertex Operator Algebra (VOA) such that the state field correspondence

[TABLE]

is defined by

[TABLE]

and the following operator product expansion formula holds:

[TABLE]

Let us fix a basis $\{\phi_{i}\}_{i=1}^{N}$ of $H$ and define the following space of formal power series

[TABLE]

where $\mathbb{C}_{\hbar}:=\mathbb{C}(\!(\hbar^{1/2})\!)$ and $q_{k}=(q_{k,1},\dots,q_{k,N})$ is a sequence of formal vector variables. The vector space $\mathfrak{M}$ is equipped with the structure of a twisted VOA module structure over the VOA $\mathfrak{F}$ in the following way. Following Givental we equip the vector space $H[z,z^{-1}]$ with a symplectic form

[TABLE]

Let $H[z,z^{-1}]\oplus\mathbb{C}$ be the Heisenberg Lie algebra with bracket

[TABLE]

We define a representation of this Heisenberg Lie algebra on $\mathfrak{M}$ as follows

[TABLE]

where $\{\phi^{i}\}_{i=1}^{N}$ is a basis of $H$ dual to $\{\phi_{i}\}_{i=1}^{N}$ with respect to the Frobenius pairing. Put

[TABLE]

where

[TABLE]

We define $Y^{\mathfrak{M}}(a,\lambda)$ for all $a\in\mathfrak{F}$ in such a way that the following operator product expansion formula holds

[TABLE]

A.2. Genus-0 reduction

The total descendant potential of a semi-simple Frobenius manifold $M$ is a formal power series of the type

[TABLE]

where $\mathbf{q}=(q_{0},q_{1},\dots)$ is the sequence of formal variables from above and $\mathcal{F}^{(g)}(\mathbf{q})$ ($g\geq 0$) is the so called genus-g descendant potential. By definition

[TABLE]

where $\mathcal{K}\subset M$ is the analytic hypersurface of all non-semisimple points and $q_{0}=(q_{0,1},\dots,q_{0,N})$ should be identified with a flat coordinate system on $M$. For precise definitions and more details we refer to [5, 6]. Let us fix a flat coordinate system defined in a neighborhood of a semi-simple point $t^{\circ}\in M$, s.t., all coordinates $t^{\circ}_{i}=0$. Then by taking the Taylor series expansion at $q_{0}=0$ we identify $\mathcal{D}(\hbar,\mathbf{q})$ with an element in the Fock space $\mathfrak{M}$

First of all note that if $w\in\mathfrak{F}$ is monodromy invariant then $Y^{\mathfrak{M}}(w,\lambda)\mathcal{D}$ is a formal power series in $\mathbf{q}$ whose coefficients are formal Laurent series in $\hbar$ whose coefficients are Laurent series in $\lambda^{-1}$. The proof of Theorem 1.1 in [1], Section 8 is straightforward to generalize to the current settings. We get that if $w\in\mathcal{W}$, then $Y^{\mathfrak{M}}(w,\lambda)\mathcal{D}$ is a formal power series in $\mathbf{q}$, formal Laurent series in $\hbar$ whose coefficients are polynomials in $\lambda$. For brevity in the latter case we say that $Y^{\mathfrak{M}}(w,\lambda)\mathcal{D}$ is regular in $\lambda$. Note that the regularity is equivalent to a sequence of differential operator constraints for $\mathcal{D}$. Indeed expanding $Y^{\mathfrak{M}}(w,\lambda)$ into a Laurent series in $\lambda^{-1}$ yields a sequence of differential operators acting on $\mathfrak{M}$. The regularity means that the differential operators in front of negative powers of $\lambda$ annihilate $\mathcal{D}$.

Let us compute the leading order term in the Laurent series expansion in $\hbar$ of $Y^{\mathfrak{M}}(w,\lambda)\mathcal{D}/\mathcal{D}$. It is convenient to embed $H\to\mathfrak{F}$ via $\alpha\mapsto\alpha s^{-1}$. Let us fix a basis $\alpha_{1},\dots,\alpha_{N}$ of $H$ and denote by

[TABLE]

Then the Fock space

[TABLE]

We will refer to $\partial^{k}\alpha_{i}$ for $k>0$ as jet variables. For a given monomial

[TABLE]

we define its weight to be $l_{1}(k_{1}+1)+\cdots+l_{N}(k_{N}+1)$ and its degree to be $l_{1}+\cdots+l_{N}$. Let us introduce a grading of $\mathfrak{F}$ such that the homogeneous piece $\mathfrak{F}_{n}$ is spanned by monomials of weight $n$. The W-algebra $\mathcal{W}$ is a graded subspace of $\mathfrak{F}$, i.e., if $w\in\mathcal{W}$ and $w=\sum_{n=0}^{\infty}w_{n}$ is a decomposition into homogeneous components $w_{n}\in\mathfrak{F}_{n}$ then $w_{n}\in\mathcal{W}$.

Suppose now that $w\in\mathcal{W}$ is homogeneous, i.e., $w\in\mathfrak{F}_{n}$ for some $n$. Note that $w=w^{(n)}+w^{(n-1)}+\cdots+w^{(1)}$, where $w^{(i)}$ is a linear combination of monomials of degree $i$. Moreover the term $w^{(n)}=f(\alpha_{1},\dots,\alpha_{N})\in\mathbb{C}[\alpha_{1},\dots,\alpha_{N}]$. Recalling the operator product expansion formula we get

[TABLE]

Therefore

[TABLE]

where the dots stand for terms that involve higher powers of $\hbar$, $\mathbf{p}=(p_{0},p_{1},\dots)$ is a sequence of vector variables $p_{k}=(p_{k,1},\dots,p_{k,N})$, the substitution $\mathbf{p}=\partial_{\mathbf{q}}\mathcal{F}^{(0)}$ means $p_{k,i}=\frac{\partial\mathcal{F}^{(0)}}{\partial q_{k,i}}$, and

[TABLE]

Note that $Y^{\mathfrak{M}}(w,\lambda)\mathcal{D}/\mathcal{D}$ is a Laurent series in $\hbar^{1/2}$ whose leading order term is

[TABLE]

We get that if $w\in\mathcal{W}$ is homogeneous of weight $n$ then the expression (46) is regular in $\lambda$, where $f=w^{(n)}$ is the degree $n$ part of $w$. By definition the regularity condition means that in the Laurent series expansion in $\lambda^{-1}$ all coefficients in front of negative powers of $\lambda$ must vanish. On the other hand, every $\alpha_{i}\in H$ is a linear function on $H^{*}$. Therefore the polynomial $f$ defines a holomorphic function on $H^{*}$. The main result in this Appendix can be stated as follows.

Proposition A.1**.**

If $f\in\mathbb{C}[\alpha_{1},\dots,\alpha_{N}]$ is a $W$-invariant polynomial, then $f\in\mathcal{M}(\Omega,W)$ and the expression (46) is regular.

Remark A.2**.**

We expect that Proposition A.1 can be generalized in the following way: if $f\in\mathcal{O}_{H^{*}}(\Omega)$ is a $W$-invariant holomorphic function then the expression (46) makes sense and it is regular in $\lambda$ if and only if $f\in\mathcal{M}(\Omega,W)$.

Remark A.3**.**

If the monodromy group $W$ is not finite (which is almost always the case), then the ring of modular functions $\mathcal{M}(\Omega,W)$ contains very few polynomials. Therefore the W-algebra defined in [1] is not big enough to characterize the total descendant potential. Nevertheless there is still some hope that the construction from [1] can be generalized, because we can replace the Fock space $\mathfrak{F}$ with a larger one, such as

[TABLE]

It will be interesting to find out if the modular functions for $\mathbb{P}^{2}$ found in this paper can be extended to W-constraints for the total descendant potential of $\mathbb{P}^{2}$.

A.3. Givental’s symplectic space formalism

The proof of Proposition A.1 relies on the properties of a certain Lagrangian cone (see [7]). Let us recall the necessary background.

Put $\mathcal{H}=H(\!(z^{-1})\!)$ and let us define a symplectic structure on $\mathcal{H}$ via formula (44). We have $\mathcal{H}=\mathcal{H}_{+}\oplus\mathcal{H}_{-}$, where $\mathcal{H}_{+}:=H[z]$ and $\mathcal{H}_{-}:=H[\![z^{-1}]\!]z^{-1}$ are Lagrangian subspaces. The formal variables $\mathbf{q}$ are identified with coordinates on $\mathcal{H}_{+}$ via

[TABLE]

The linear structure on $\mathcal{H}_{+}$ gives a natural identification of each tangent space

[TABLE]

Using the symplectic form $\Omega$ we identify

[TABLE]

Therefore the cotangent bundle $T^{*}\mathcal{H}_{+}\cong\mathcal{H}$. Note that under the identification $T_{\mathbf{q}}\mathcal{H}_{+}\cong\mathcal{H}_{+}$ the 1-form $dq_{k,i}=\Omega(\phi^{i}(-z)^{-k-1},\cdot)$. The axioms of gravitational descendants can be reformulated in the following geometric way. Let us identify $\mathcal{F}^{(0)}(\mathbf{t})$ with a function on $\mathcal{H}_{+}$ via the substitution $\mathbf{t}(z)=\mathbf{q}(z)+z$. This change of variables is known as the dilaton shift. Let

[TABLE]

Intuitively as $\mathbf{q}$ varies in $\mathcal{H}_{+}$ the values of $J(\mathbf{q},z)$ defines a subset $\mathcal{L}\subset\mathcal{H}$, which under the isomorphism $\mathcal{H}\cong T^{*}\mathcal{H}_{+}$ coincides with the graph of the differential $d_{\mathbf{q}}\mathcal{F}^{(0)}$. The 3 axioms DE, SE, and TRR are equivalent to the following properties of $\mathcal{L}$:

1. (1)

$\mathcal{L}$ is a quadratic cone 2. (2)

If $\mathbf{f}\in\mathcal{L}$ is a smooth point then the tangent space $L=T_{\mathbf{f}}\mathcal{L}$ is Lagrangian, $zL=L\cap\mathcal{L}$, and if $\mathbf{g}\in zL$, then $T_{\mathbf{g}}\mathcal{L}=L$.

Formally the properties of $J(\mathbf{q},z)$ can be stated as follows. Let us recall that if the Frobenius structure comes from gravitational descendants then there is a natural choice of a calibration

[TABLE]

There exists

[TABLE]

such that

[TABLE]

Note that $J(\mathbf{q},z)=-z+J(\mathbf{t},z)$ and that $S_{1}(t)1=t$. Multiplying both sides of the above identify by $S(\tau(\mathbf{t}),z)$ and comparing the coefficients in front of $z^{0}$ yields

[TABLE]

This equation allows us to solve for $\tau$ in terms of $\mathbf{t}$. Finally $v(\mathbf{t},z)=[S(\tau(\mathbf{t}),z)(\mathbf{t}(z)-z)]_{+}$, where $[\cdot]_{+}:\mathcal{H}\to\mathcal{H}_{+}$ is the projection. In other words, the formal series $J(\mathbf{q},z)$ is uniquely determined from the calibration $S(t,z)$.

A.4. Proof of Proposition A.1

Recalling the definition (45) we get that

[TABLE]

There exists $\tau(\mathbf{q})$ and $v(\mathbf{q},z)$ such that $J(\mathbf{q},z)=S(\tau(\mathbf{q}),z)^{-1}v(\mathbf{q},z)$ (see Section A.3). The calibration $S$ is a symplectic transformation. Therefore

[TABLE]

where we used that

[TABLE]

Let $\mathfrak{m}_{k}$ ($k\geq 1$) be the ideal in $\mathcal{O}_{M}(M)[\![q_{1}+\mathbf{1},q_{2},\dots,]\!]$ generated by $(q_{k}+\mathbf{1}\delta_{k,1},q_{k+1},\dots)$. Note that $\tau(\mathbf{q})\in q_{0}+\mathfrak{m}_{1}$ and

[TABLE]

We have

[TABLE]

Suppose now that $f\in\mathbb{C}[\alpha_{1},\dots,\alpha_{N}]$ is a homogeneous polynomial of degree $n$. Note that the expression (46) is obtained from the polynomial $f(\alpha_{1},\dots,\alpha_{N})$ via two substitutions: first

[TABLE]

and second $t\mapsto\tau(q)$. We claim that the first substitution yields an expression regular in $\lambda$. This would complete the proof of the proposition, because the second substitution preserves the regularity property. Suppose that we make the substitution (47) in $f(\alpha_{1},\dots,\alpha_{N})$. Let us fix $t\in M$. Then the resulting expression is a formal power series in $q_{1}+\mathbf{1},q_{2},\dots$ whose coefficients are polynomial expressions in $(I^{(-k)}_{\alpha_{i}}(t,\lambda),\phi_{j})$ with $k>0$. Since $f$ is a $W$-invariant polynomial, each coefficient is a single-valued holomorphic function on $\mathbb{C}\setminus{\{u_{1},\dots,u_{N}\}}$, where $u_{i}$ are the eigenvalues of the operator $E\bullet_{t}$. Since the periods have finite order pole at $\lambda=\infty$ the regularity condition will be established if we manage to prove that the coefficients are holomorphic at $\lambda=u_{i}$. This however follows from the Riemann’s extension theorem, because the period vectors $(I^{(-k)}_{\alpha_{i}}(t,\lambda),\phi_{j})$ for $k>0$ are bounded in a neighborhood of $\lambda=u_{i}$. Note that if $v_{k}=0$ for $k\geq 0$ then this argument proves that $f\in\mathcal{M}(\Omega,W)$. ∎