Computational approach to the Schottky problem

We present a computational approach to the classical Schottky problem based on Fay’s trisecant identity for genus $g \ge 4$. For a given Riemann matrix $B \in Hg$, the Fay identity establishes linear dependence of secants in the Kummer variety if and only if the Riemann matrix corresponds to a Jacobian variety as shown by Krichever. The theta functions in terms of which these secants are expressed depend on the Abel maps of four arbitrary points on a Riemann surface. However, there is no concept of an Abel map for general $B \in Hg$. To establish linear dependence of the secants, four components of the vectors entering the theta functions can be chosen freely. The remaining components are determined by a Newton iteration to minimize the residual of the Fay identity. Krichever’s theorem assures that if this residual vanishes within the finite numerical precision for a generic choice of input data, then the Riemann matrix is with this numerical precision the period matrix of a Riemann surface. The algorithm is compared in genus $4$ for some examples to the Schottky-Igusa modular form, known to give the Jacobi locus in this case. It is shown that the same residuals are achieved by the Schottky-Igusa form and the approach based on the Fay identity in this case. In genera $5$, $6$ and $7$, we discuss known examples of Riemann matrices and perturbations thereof for which the Fay identity is not satisfied. This is work with E. Brandon de Leon and J. Frauendiener.