A non-linear differential equation for the periods of elliptic surfaces

Technical Summary: A Non-Linear Differential Equation for the de Rham Cohomology of Elliptic Surfaces

Problem Statement
The paper addresses the structure of the moduli space of Jacobian elliptic surfaces $f: X \to C$ over $\mathbb{C}$, specifically focusing on the relationship between the geometry of the surface, the ramification of its classifying morphism, and the variation of its Hodge structure. The central problem is to characterize the period map for the primitive $(1,1)$-cohomology, $H^{1,1}_{\text{prim}}(X)$, and to derive a governing differential equation for this map. While the Gauss–Manin connection provides a linear framework for the variation of Hodge structures, the author seeks a non-linear partial differential equation (PDE) system that explicitly constrains the period matrix of the $(1,1)$ classes in terms of the local coordinates of the moduli stack.

Methodology
The approach combines the geometry of moduli stacks, the theory of elliptic surfaces, and the analysis of the Gauss–Manin connection.

1. Moduli and Coordinates: The author considers the stack $\mathcal{J}\mathcal{E}$ of Jacobian elliptic surfaces with irreducible nodal singular fibers. This stack is shown to be smooth of dimension $N = 10h + 8(1-q)$, where $h$ is the geometric genus and $q$ the irregularity. A local coordinate system on the moduli stack is constructed via the branch locus of the classifying morphism $\phi: C \to \mathcal{E}\ell\ell$ (where $\mathcal{E}\ell\ell \cong \mathbb{P}(4,6)$ is the stack of generalized elliptic curves). Specifically, if $Z = \text{Ram}(\phi)$ is the ramification divisor, the branch divisor $B = \phi^*Z$ provides étale local coordinates $(t_i)_{i \in [1,N]}$.
2. Cohomological Basis: For each ramification point $P_i \in Z$, a specific meromorphic 2-form $\tilde{\eta}_i$ of the second kind is constructed in $H^0(X, \Omega^2_X(2E_i))$, where $E_i = f^{-1}(P_i)$. These forms are normalized to vanish on a specific isotropic sublattice $L \subset H^2(X, \mathbb{Z})_{\text{prim}}$. Their projections $\eta_i$ into $H^{1,1}_{\text{prim}}(X)$ form an orthogonal basis.
3. The Period Matrix: The matrix $H = [\eta_1, \dots, \eta_N]$ is treated as an element of the complex orthogonal group $O_N$, representing a multi-valued period map from the moduli stack to the torsor of orthonormal frames.
4. Derivation of the PDE: By analyzing the derivative of the period map along the coordinate curves defined by varying a single branch point, the author relates the variation of the basis forms $\eta_i$ to the Gauss–Manin connection. The key step involves "truncating" the linear system of the Gauss–Manin connection to isolate the non-linear constraints on the $(1,1)$ classes.

Key Contributions and Results

• The "Ecliptic" Equation: The primary result is the derivation of a system of non-linear PDEs satisfied by the period matrix $H$. The system is given by:
  $$\{ \partial_i H = (\partial_i \eta_i \wedge \eta_i) H \}_{i \in [1,N]}$$
  where $\partial_i = \partial/\partial t_i$. The term $\partial_i \eta_i \wedge \eta_i$ is interpreted geometrically as the "ecliptic" of the particle $\eta_i$ (the 2-plane defined by its position and velocity). This equation holds generally, even when the canonical linear system $|K_X|$ has base points, extending previous results that required the absence of base points.
• Orthogonality and Basis Construction: The paper provides a rigorous construction of an orthonormal basis for $H^{1,1}_{\text{prim}}(X)$ using meromorphic 2-forms of the second kind. It proves that these classes are mutually orthogonal, a property previously established only under restrictive assumptions (no base points).
• Infinitesimal Torelli Theorem: A generic infinitesimal Torelli theorem is proven for the map $H$. The derivative of the period map is shown to be generically immersive. The paper analyzes the case where base points exist (specifically when $h=q$ and the base point $Q$ lies in the ramification divisor $Z$), showing that in this specific scenario, the derivative of the period map has corank 1, and infinitesimal Torelli fails.
• Explicit Calculations for Rational Surfaces: For rational elliptic surfaces ($q=0, h=1$), the paper outlines an explicit algorithm to calculate the period matrix. This involves:
  • Constructing the surface as a blow-up of $\mathbb{P}^2$ at nine points.
  • Using the anticanonical model and sextic equations in weighted projective space.
  • Computing intersection numbers $[\eta_i] \cup_{dR} [D]$ with specific curves $D$ (sections of the fibration) using residue formulas derived from the local expansions of the forms.
  • Determining the scalars required to normalize the basis to be orthonormal.

Significance
The paper claims significance in providing a concrete, non-linear differential equation that governs the variation of the $(1,1)$-cohomology of elliptic surfaces, a sector of the cohomology often more difficult to access than the $(2,0)$ or $(0,2)$ parts. By interpreting the period map as a solution to this "ecliptic" equation, the work connects the geometry of the moduli space (via the branch divisor) directly to the Hodge structure. The explicit calculation for rational surfaces demonstrates that these abstract constructions are computable, offering a potential pathway for explicit period calculations in arithmetic geometry. The work also refines the understanding of the infinitesimal Torelli theorem in the presence of base points, clarifying the precise geometric conditions under which the period map degenerates.