Generic Newton polygons of Ekedahl-Oort strata: Oort's conjecture (Q609662)

This very technical but very clearly written paper proves Oort's conjecture, elucidating the relation between \(p\)-divisible groups and \(p\)-torsion. The result determines the Newton polygon of a generic point of each Ekedahl-Oort stratum in the moduli space of principally polarised abelian varieties (ppavs) in positive characteristic. The author acknowledges some help from Oort, who first stated his conjecture in print seven years ago. The conjecture is thus not of remarkably long standing; nevertheless, its proof is a notable achievement. The Ekedahl-Oort or EO stratification \(\{{\mathcal S}_w\}\) classifies ppavs according to their \(p\)-kernels \(A[p]\), considered as Barsotti-Tate groups: they are indexed by final elements \(w\in{\mathbb W}_g\), the Weyl group of \(\text{Sp}_{2g}\). The Newton polygon stratification instead classifies ppavs by their \(p\)-divisible group \(A[p^\infty]\): the strata are indexed by Newton polygons \(\xi\). For a Newton polygon \(\xi\) the central stream \({\mathcal Z}_\xi\) is the locus of ppavs whose polarised \(p\)-divisible group agrees, over an algebraically closed field, with a standard polarised \(p\)-divisible group \(H(\xi)\). It is equal to \({\mathcal S}_{w_\xi}\), where \(w_{\xi}\) corresponds to \(H(\xi)[p]\). Oort's conjecture says that as long as the EO stratum \({\mathcal S}_w\) meets the open Newton polygon stratum \({\mathcal W}^0_\xi\), then \({\mathcal Z}_\xi\subseteq \overline{\mathcal S}_w\). What Harashita proves here is that for any final element \(w\in {\mathbb W}_g\), one has \({\mathcal Z}_{\xi(w)}\subseteq \overline{\mathcal S}_w\), where \(\xi(w)\) is the Newton polygon at the generic point of \({\mathcal S}_w\) (one must say this, and much else, slightly differently for the supersingular locus). This, by earlier results of the author, is equivalent to Oort's conjecture. A corollary is that \(\xi(w)\) is the \(\prec\)-maximal Newton polygon among \(\{\xi\mid {\mathcal Z}_\xi\subseteq \overline{\mathcal S}_w\}\). Another corollary is an algorithm to find \(\xi(w)\) given \(w\). The first few sections of the paper give some background on \(p\)-divisible groups, Dieudonné modules, Barsotti-Tate groups and related matters, and the stratifications. After that the author shows that the theorem follows if one can construct a suitable family of \(p\)-divisible groups \(H\) of constant Newton polygon and constant \(H[p]\). To construct the family the steps are to produce first a suitable complex of \(F\)-zips (roughly, Barsotti-Tate groups over a base) and then to lift it to displays (roughly, Dieudonné modules over a base): the latter classify the \(p\)-divisible groups.