Toric reduction and a conjecture of Batyrev and Materov (Q706137)

The paper deals with the mirror symmetry statement that, on the one hand, the generating function of intersection numbers on moduli spaces of curves in some variety (``A-side'') and, on the other, the function arising from the variation of Hodge structures on the ``B-side'' coincide. While, in the original version of \textit{V. V. Batyrev} and \textit{E. N. Materov} [Mosc. Math. J. 2, No. 3, 435--475 (2002; Zbl 1026.14016)], the ``Toric Residue Mirror Conjecture'' deals with the mutually dual families of CY-varieties arising from a pair of mutually dual reflexive polyhedra, the present paper proves this conjecture in a more general context -- reflexivity is not assumed anymore. Let \(\alpha:\mathbb Z^n\to M\cong\mathbb Z^r\), and denote by \(\beta:(\mathbb Z^n)^\ast\to{\widetilde N}\cong\mathbb Z^d\) the ``Gale-dual'', i.e.\ the dual map of \([{\widetilde M}:=\ker\alpha\hookrightarrow\mathbb Z^n]\). Under some technical conditions (\(\alpha\) is ``projective'' and ``spanning''), every \(u\in \alpha(\mathbb Q^n_{\geq 0})\) gives rise to a compact polytope \(\Delta_u:=\alpha^{-1}(u)\cap \mathbb Q^n_{\geq 0}\) sitting in a translate of \({\widetilde M}_\mathbb Q\). This translates into a \(d\)-dimensional projective toric variety \(V_A(c)\) (the ``A-side'') that depends only on the chamber \(c\) containing \(u\). The participating torus is \({\widetilde N}\otimes_\mathbb Z\mathbb C^\ast\), and the vector space \(M_\mathbb Q\) may be identified with \(H^2(V_A(c),\mathbb Q)\). In particular, polynomials \(P(z_1,\ldots,z_n)\) give rise to cohomology classes \(P(\alpha_1,\ldots,\alpha_n)\in S^{{\scriptscriptstyle \bullet}} M_\mathbb Q= H^{2{\scriptscriptstyle \bullet}}(V_A(c),\mathbb Q)\), and one may consider \(\int_{V_A(c)} P(\alpha_1,\ldots,\alpha_n)\). Using a result of \textit{M. Brion} and \textit{M. Vergne} [Ann. Sci. Éc. Norm. Supér., IV. Sér. 32, 715--741 (1999; Zbl 0945.32003)], this integral is representable as the so-called Jeffrey-Kirwan residue of \(P(\alpha)/\prod_i\alpha_i\), which in turn equals the integral of \(P(\alpha)/\prod_i\alpha_i\), now understood as a polynomial function on \(N_\mathbb R\), over some real \(r\)-cycle. Fixing a class \(\lambda\in c^\vee\subseteq N_\mathbb Q=H_2(V_A(c),\mathbb Q)\), one defines the so-called Morrison-Plesser moduli space \(\text{MP}_\lambda\) of \(\lambda\)-curves in \(V_A(c)\) in a similar manner as \(V_A(c)\) itself. For a polynomial \(P(z_1,\ldots,z_n)\), one defines \(\langle P\rangle_{\lambda,A,c}\) as an integral similar to that over \(V_A(c)\) -- yielding the generating function \(\langle P\rangle_{A,c}:=\sum_\lambda \langle P\rangle_{\lambda,A,c}\cdot z^\lambda\). It can be presented as the integral of a real form \(\Lambda\) over the \(r\)-cycle on \(N_\mathbb R\) mentioned above. The ``B-side'' is the \(d\)-dimensional, polarized toric variety \(V_B\) defined by the convex hull of the points \(\beta_1,\ldots,\beta_n\). Now, every polynomial \(P(z_1,\ldots,z_n)\) gives rise to a global section \(P(z_1\beta_1,\ldots,z_n\beta_n)\) of a tensor power of the ample line bundle, i.e.\ to an element of the homogeneous coordinate ring. Applying Cox's toric residue (also depending on \(z\) [ cf. \textit{D. Cox}, Ark. Mat. 34, 73--96 (1996; Zbl 0904.14029)]), one defines \(\langle P\rangle_B(z):= \text{TorRes}_z P(z_1\beta_1,\ldots,z_n\beta_n)\). This function may be obtained via cumulating values of a certain function on a finite subset of the embedded torus \({\widetilde M}\otimes_\mathbb Z\mathbb C^\ast\). Now, the miracle is that this finite set can naturally be seen as a subset of \(N_\mathbb C\) -- hence, applying the functions on it involves restricting them to \(N_\mathbb C\), i.e.\ applying \(\alpha\). Writing the result as an integral via using ordinary residues, one is surprised once more: The integral involves the same \(r\)-form \(\Lambda\) as we met on the A-side. Eventually, comparing several real \(r\)-cycles in \(N_\mathbb C\), the conjecture reduces to a new residue formula for the intersection pairings of toric manifolds. The method of the proof uses tropical geometry. The authors reach the result that \(\langle P\rangle_{A,c}(z)=\langle P\rangle_B(z)\). The right hand side does not depend on \(c\). However, on the left hand side, different \(c\)'s lead to different regions of convergence.