The method is really the same as what Deligne-Mumford do to handle the moduli space of curves (creating a smooth cover from a part of a Hilbert scheme), except facts about curves (e.g., Riemann-Roch and cohomological vanishing results) are replaced with analogues for abelian varieties. Below is a guide to relevant literature for why mathcalAg,d is a separated DM stack of finite type over mathbfZ. (Question had d=1.)
Everything needed about abelian varieties is in Mumford's book on abelian varieties and the (self-contained!) Chapter 6 on abelian schemes in his GIT book. The key ingredients are (i) the "Riemann-Roch and Vanishing" theorems from section 16 of Mumford's book on abelian varieties, (ii) Proposition 6.13 in Mumford's GIT book (a relativization of results proved over an algebraically closed field in his book on abelian varieties), (iii) Proposition 6.11 in GIT, and (iv) Proposition 6.14 in GIT.
The references (i) and (ii) ensure that if you consider an abelian scheme f:ArightarrowS of relative dimension g>0 (with dual Armt and Poincar'e bundle mathcalP) and a polarization phi:ArightarrowArmt on A of degree d2 then for the resulting
S-ample line bundle mathcalL=(1,phi)ast(mathcalP) on A the pushforward mathcalE=fast(mathcalLotimes3) is a vector bundle on S whose formation commutes with any base change, has rank determined entirely by g and d, and defines a closed immersion AhookrightarrowmathbfP(mathcalE). (In these arguments with relative ampleness, EGA IV3, 9.6.4 is very useful for bootstrapping from fields to a general base.) Thus, upon making a universal (smooth) base change to equip mathcalE with a global frame, one gets a map from the new base to a certain Hilbert scheme for a projective space (over mathbfZ) and Hilbert polynomial determined by g and d. Since an abelian scheme has a section, this map naturally factors through the universal object X over that Hilbert scheme (which represents the functor of also specifying a section).
Now (iv) comes in: it says that over X, the condition on the universal marked object that it admit a (necessarily unique) structure of abelian scheme with the marked section as the identity is represented by an open subscheme. Restricting to this locus (and again using the Hilbert polynomial to track the rank of the pushforward of its mathcalO(1)) gives a kind of first approximation to a "universal" abelian scheme mathcalA, but we need to universally equip it with a polarization whose cube recovers the mathcalO(1) from the projective embedding as its "associated" ample line bundle (via the construction with the Poincar'e bundle). That's where (iii) comes in: it says that such a polarization is unique if it exists and that its existence is represented by a closed subscheme of the base. Over that subscheme we get a degree-d2 polarization on an abelian scheme of relative dimension g, and the way we arrived at it showed that the base of the family (of finite type over mathbfZ) is a smooth cover of the moduli stack (the "smooth cover" coming about because of the arrangement to get a global frame for mathcalE).
Since one checks before any of the above that the moduli stack does satisfy effective descent (thanks to descent theory with the ample line bundle associated to the polarization; see 6.1/7 in the book "Neron Models"), we indeed have an Artin stack of finite type over mathbfZ. The finite 'etale property of automorphism groups of geometric points implies it is actually a DM-stack (using the "unramified diagonal" criterion for an Artin stack to be a DM stack: Theorem 8.1 in the Laumon/Moret-Bailly book). Finally, to check separatedness one uses the valuative criterion, which is the N'eronian property for abelian schemes over a discrete valuation ring (1.2/8 in "Neron Models"). QED
No comments:
Post a Comment