Modular forms from Noether-Lefschetz theory (Q2212124)

Let \(Y\subset\mathbb{P}^{n+1}\) be a smooth hypersurface of degree \(d\) and dimension \(m\geq 2\). For an ample line bundle \({\mathcal{L}}={\mathcal{O}}_Y(k)\), a Weierstrass fibration over \(Y\) is a hypersurface \(X\subset{\mathbb{P}}({\mathcal{L}}^{\otimes -2}\oplus {\mathcal{L}}^{\otimes -3}\oplus{\mathcal{O}}_Y)\) cut out by a global Weierstrass equation. \(X\) is smooth of dimension \(m+1\), and the morphism \(\pi:X\to Y\) is flat with generic fiber of genus one, in fact, an elliptic curve ( as \(\pi\) admits a section \(i:Y\to X\)). The second homology group of \(X\) is given by \(H_2(X,{\mathbb{Z}})={\mathbb{Z}}\ell+{\mathbb{Z}}f\) where \(\ell\) is the line class of \(Y\) pushed forward via \(i\), and \(f\) is the class of a fiber. The main result of this paper gives a finite emuneration to the following question: How many smooth rational curves are there on \(X\) in the homology class \(\ell+nf\)? Theorem. Suppose that \(\dim(X)=3\), i.e., \(X\) is a Calabi-Yau threefold and this holds whenever \(k=2n-d(>0)\). Let \(r_X(n)\) denote the number of such curves. Then for \(k\leq 4\), the generating series is given by \[\sum_{n\geq 1} r_X(n)q^n=\varphi(q)-\Theta(q)\] where \(\varphi(q)\in\text{Mod}(6k-2, \mathrm{SL}_2(\mathbb{Z}))\), and \(\Theta(q)\in\mathbb{Q}[\theta_{A_1}, \theta_{A_2},\theta_{A_3}]\), a polynomial of weighted degree less than \(k\). Here for a lattice \(A\), \(\Theta_A(q)=\sum_{v\in A} q^{(u.v)/2}\). Note that both \({\text{Mod}}(6k-2,\mathrm{SL}_2(\mathbb{Z}))\) and \(\mathbb{Q}[\theta_{A_1}, \theta_{A_2},\cdots]\) are finite-dimensional \(\mathbb{Q}\)-vector spaces. The quantity \(r_X(n)\) is computed using Neother-Lefschetz theory. The method of modularity proof uses topological intersection products on a period stack and cohomological theta correspondence of \textit{S. S. Kudla} and \textit{J. J. Millson} [Publ. Math., Inst. Hautes Étud. Sci. 71, 121--172 (1990; Zbl 0722.11026)]. The result here apply only in base degree \(1\), but heuristics for higher base degree match predictions from the topological string partition functions.