No label defined (Q4820347)

Forty years after the seminal work by \textit{A. Andreotti} and \textit{A. L. Mayer} [Ann. Scuola Norm. Sup. Pisa (3) 21, 189--238 (1967; Zbl 0222.14024)], subloci of the moduli space of abelian varieties with specific restrictions on the theta divisor are still little known. The authors make a new contribution, which they prove in a very original way by revisiting a seemingly forgotten result of Corrado Segre on the geometry of quadrics. The first part of the paper gives a very useful overview of what is known in the context of interest. We mention the chief dimensional results. Denoting by \({\mathcal A}_g\) the moduli space of principally polarized abelian varieties (ppav's) of dimension \(g\), for an integer \(k\) with \(0\leq k\leq g-2,\) the authors denote by \(N_{g,k}\) the subloci of \({\mathcal A}_g\) defined as the Zariski closure of the sets: \[ \left\{[(X,\Theta_X)]\in {\mathcal A}_g({\mathbb C})| \dim\text{ Sing}(\Theta)=k\right\} \] where \((X,\Theta_X)\) is a p.p.a.v. of dimension \(g\) with theta divisor \(\Theta_X\). Andreotti and Mayer proved in the paper cited above (where the letter \(k\) is used for codimension instead of dimension) that the Jacobian locus \(J_g\) is an irreducible component of \(N_{g,g-4}\) and the hyperelliptic locus \({\mathcal H}_g\) is an irreducible component of \(N_{g,g-3}\). \textit{O. Debarre} proved that the Prym locus \(P_g\) (defined by unramified double-covers of a curve, with \(P_g\) the kernel of the norm-map) in \({\mathcal A}_g\) is an irreducible component of \(N_{g,g-6}\) for \(g\geq 7\) [Duke Math. J. 60, No.3, 599--630 (1990; Zbl 0716.14029)]. The authors' result gives the following lower bound for the codimension of \(N_{g,k}\), which refines Mumford's estimate: (i) if \(g\geq 4\) and \(1\leq k\leq g-3,\) then \(\text{ codim}\,N_{g,k}\geq k+2\); (ii) if \(g\geq 5\) and \(g/3 < k\leq g-3\), then \(\text{ codim}\,N_{g,k}\geq k+3.\) They also prove that \(N_{g,0}\) is a divisor properly containing \(\bigcup_{k\geq 1}N_{g,k}\). However, it is not known if \(N_{g,k}\) properly contains \(\bigcup_{i\geq k+1}N_{g,i}\). Finally, they conjecture the following ``much stronger bound'': Conjecture: If \(1\leq k\leq g-3\) and if \(M\) is an irreducible component of \(N_{g,k}\) whose generic point corresponds to a simple abelian variety, then \(\text{ codim}\,M\geq {{k+2\choose 2}},\) and moreover equality holds if and only if \(g=k+3\) and \(M={\mathcal H}_g\) (resp., \(g=k+4\) and \(M=J_g\)). The method of proof, different from an unpublished, not equally strong, estimate of Debarre, is based on the heat equation satisfied by the theta function. Thanks to a modern, cohomological, interpretation of the heat equation in terms of deformation theory [\textit{G. E. Welters}, Compos. Math. 49, 173--194 (1983; Zbl 0576.14042)], criteria on the singularities of the theta divisor can be recast in terms of quadrics in projective spaces (whose underlying linear spaces are cohomology groups). At this point the authors conclude by a beautiful geometric criterion: Let \(L\) be a pencil of singular quadrics of rank \(\leq n+1-r\) in \({\mathbb P}^n\) with \(n\geq 2\) whose generic member has rank \(n+1-r\). Assume the vertex is not constant in the pencil. Then the Zariski closure of the generic vertex (namely the union of the vertices of the generic quadrics) is a variety of dimension \(r\) and degree \(m-r+1\) in a projective linear subspace \({\mathbb P}^m\subset{\mathbb P}^n\) with \(m\leq (n+r-1)/2\) and \(r\leq (n+1)/3\) [\textit{C. Segre}, Torino Atti XIX. 878--897 (1884; JFM 16.0604.02)]. In the last section the authors provide evidence for the conjecture when \(k=1\). For the entire collection see [Zbl 1044.53002].