Cohomology of the second Voronoi compactification of \(\mathcal A_4\) (Q455487)

Compared to how much we know about the cohomology of moduli spaces of curves, our knowledge of the cohomology of moduli spaces of abelian varieties is very meager. The cohomology of the spaces \(A_g\) of principally polarized abelian varieties and/or their compactifications have been determined completely only when \(g \leq 3\), and for these cases the methods of proof have exploited the fact that every abelian variety of dimension at most three is a product of Jacobians, and our knowledge of the cohomology of the spaces \(M_g\). The paper under review determines all cohomology groups of the space \(\overline A_4\), the bar denoting the 2nd Voronoi compactification, except the middle one. Their method is very indirect. The idea is to apply the spectral sequence associated to the filtration of \(\overline A_4\) by torus rank, except the open stratum is taken to be the complement of the closed locus of Jacobians in \(A_4\). The strata defined by positive torus rank are fibered over the spaces \(A_g\), \(g < 4\), which implies that their cohomology can be computed in terms of cohomology of these spaces (possibly with local coefficients). The cohomology of the locus of Jacobians in \(A_4\) can be determined from knowledge of the cohomology of \(M_4\). Hence all these strata have cohomology which can be computed, together with their mixed Hodge structure, by reducing to moduli spaces of curves. For the open stratum they apply a theorem of Igusa, stating that the complement of the Jacobian locus in \(A_4\) is affine. By the affine vanishing theorem, we know that `half' its cohomology groups are zero. Since \(\overline A_4\) is smooth and compact, we also know Poincaré duality and purity of the corresponding Hodge structures. Putting the ingredients together one can determine all cohomology groups of \(\overline A_4\) except the middle one. They find that all cohomology is algebraic. They also determine all cohomology groups below the middle for the perfect cone compactification, where again all cohomology is algebraic (below the middle). They remark that merely knowing the Euler characteristic (i.e. the alternating sum of Betti numbers) of \(A_4\) would be enough to determine also the last cohomology group; however, not even this is known, a fact which was slightly shocking to the referee. Clearly there is much more work to be done here.