The loci of abelian varieties with points of high multiplicity on the theta divisor (Q1018070)

Ever since the famous paper of \textit{A. Andreotti} and \textit{A. L. Mayer} [Ann. Sc. Norm. Super. Pisa, Sci. Fis. Mat., III. Ser. 21, 189--238 (1967; Zbl 0222.14024)] it has been known that valuable information is concealed in the stratification of \({\mathcal A}_g\) (the moduli space of principally polarised abelian varieties of dimension~\(g\)) according to the dimension of the singular locus of the theta divisor. It is well concealed, though: even after another forty years we still do not know even the dimensions of the Andreotti-Mayer loci \(N_k=\{(X_\tau,\Theta_\tau)\mid \dim\text{Sing}\Theta_\tau\geq k\}\subset{\mathcal A}_g\), which in any case are reducible. This paper works with a different, but related, stratification: by the multiplicity of points of \(\Theta\) rather than the dimension. As usual we let \(\Theta\) denote the universal theta divisor and \(\Theta_{\text{sing}}\) its singular locus. We put \[ G_k=\{(X_\tau,\Theta_\tau)\mid \dim(X_\tau \cap \Theta_{\text{sing}})\geq k\} \] which, by the heat equation, is the same as \[ \{(X_\tau,\Theta_\tau)\mid \dim\{z\in X_\tau \mid \text{mult}_z\Theta_\tau\geq 2\}\geq k\}. \] It is quite hard even to get much information about \(G_0\). Also define \[ (\partial\theta)_{\text{null}}=\{(X_\tau,\Theta_\tau)\mid X_\tau[2]^{\text{odd}} \cap \Theta_{\text{sing}}\neq\emptyset\} \] and \[ (\partial^k\theta)_{\text{null}}=\{(X_\tau,\Theta_\tau)\mid \exists x\in X_\tau[2]\;\text{mult}_x\Theta\in k+2{\mathbb N}\}, \] analogously to the locus \(\theta_{\text{null}}=\{(X_\tau,\Theta_\tau)\mid X_\tau[2]^{\text{even}} \cap \text{Sing}\Theta\neq\emptyset\}\), which is a component of \(N_0\). Working on \({\mathcal A}(4,8)\) to avoid orbifold (or stack) difficulties, the authors write down modular forms cutting out some components of \((\partial^2\theta)_{\text{null}}\). They show that for \(k\leq g-4\) the locus \((\partial^k\theta)_{\text{null}}\) is irreducible: one component is \(\theta_{\text{null}}^{(g-k)}\times{\mathcal A}_1\times\cdots\times{\mathcal A}_1\), which has codimension \(gk+1-{{1}\over{2}}(k^2+k)\) in \({\mathcal A}_g\). More generally they say when \(X\times{\mathcal A}_1\) is an irreducible component of \((\partial^k\theta)_{\text{null}}\) in \({\mathcal A}_{g+1}\) if \(X\) is an irreducible component of \((\partial^k\theta)_{\text{null}}\) in \({\mathcal A}_g\). Similarly they exhibit two decomposible irreducible components of~\(G_0\). The last part of the paper includes several conjectures on the codimensions of these loci and some evidence for them. According to these, \((\partial\theta)_{\text{null}}\) and \(G_0\) should have codimension~\(g\) and \((\partial^2\theta)_{\text{null}}\) should have codimension~\(2g-2\), while the codimension of \(\Theta_{\text{sing}}\) in the universal family \({\mathcal X}_g\) should be \(2g\) and \(\text{codim}G_k\) should be at least \(g+k\) (all for \(g\geq 4\) at any rate). These conjectures are related to one another, and by examining rank~\(1\) degenerations the authors are able to prove a small part of them, and show equivalences among other parts, on the assumption that every component of the relevant loci does have rank~\(1\) degenerations. For \(\Theta_{\text{sing}}\) the issue is instead whether it is pure-dimensional or not, and it is easy to show that the conjectures on \(\text{codim}\Theta_{\text{sing}}\) and on \(\text{codim}G_k\), for given~\(g\), are equivalent.