Perfect forms and the moduli space of abelian varieties (Q811851)

Among the toroidal compactifications of the moduli stack \({\mathcal A}_g\) of principally-polarised abelian \(g\)-folds, a few stand out as being of geometric or combinatorial interest. One such is the perfect cone, or first Voronoi, compactification, named after the corresponding decomposition of the cone of real positive definite quadratic forms in \(g\) variables. On the moduli space \(A_g\) and the first Voronoi compactification \(A_g^F\) there are divisor classes \(M\) (of weight~\(1\) modular forms) and \(D=(A_g^F\setminus A_g)_{\text{red}}\). These generate \(\text{NS}(A_g^F)\otimes{\mathbb Q}\). The first of the two main theorems here is that \(aM-D\) is nef if and only if \(a\geq 12\), and ample if and only if \(a>12\). For \(g=1\) this reflects the fact that there is a unique cusp form for \(\text{SL}(2,{\mathbb Z})\) of weight~\(12\) and level~\(1\), with no zeros away from the cusps. The proof of the theorem involves a careful analysis of the base locus of \(| m(12M-D)| \) and how it interacts with the stratification of \(A_g^F\) arising from the different boundary components in the Satake compactification of~\(A_g\). The second main theorem states that, over~\({\mathbb C}\) (which is all that one can hope for in the present state of knowledge of birational geometry), \(A_g^F\) is the relative canonical model of the Satake compactification if \(g\geq 5\); for \(g\geq 12\), it is in fact the canonical model, and for \(g=11\) the canonical model can be precisely described and is not very far from~\(A_{11}^F\). This second main theorem consists of a global part, that certain rings are finitely generated, which follows more or less immediately from the first main theorem, and a local part. The local part asserts that \({\mathcal A}_g^F\) has canonical singularities for~\(g\geq 5\). This was partly proved by \textit{Y. Tai} [Invent. Math. 68, 425--439 (1982; Zbl 0508.14038)]; but Tai, for his purposes, was in principle free to choose a different compactification if need be, and therefore did not need to be as precise as in necessary here. The author therefore tries to give a full proof that the singularities are canonical: in the process he points out, and fills, a gap in Tai's proof of his weaker statement. Unfortunately, Hulek and the reviewer have found a gap also in the proof given here that the singularities of \({\mathcal A}_g^F\) are canonical. The author has since produced a correction, so far only circulated privately but presumably to be made publicly available shortly. This appears to meet our objections; so I believe that the main results here are now all proved, even though the paper as printed is defective. The problem is that the conditions of Proposition~3.2 are not necessarily satisfied. One may only assume that the kernel of the action of the stabiliser~\(G\) on the fibre~\(V\) acts by torus automorphisms (for reasons of space I must refer to the paper for details of the notation): a quotient of \(G\) acts on~\(V\) and one must also analyse this action. Indeed the conclusion of the proof of Proposition~3.2 implies that the singularities at the boundary are products, and in particular non-isolated; but this is definitely false in general. It is therefore impossible to avoid the tedious analysis of the local action that arises in practice: one really cannot hope to predict the singularities at infinity from the singularities of the moduli space.