A purity theorem for abelian schemes (Q1885565)

Given a discrete valuation ring \(O\) and a regular \(O\)-scheme \(Y\), suppose that \(U\subset Y\) is an open subscheme whose complement has codimension \(\geq 2\). What kinds of object can, in this situation, always be extended from \(U\) to \(Y\)? More precisely, given a stack \({\mathcal S}\) over the category of \(O\)-schemes (with the Zariski topology), is the pullback \({\mathcal S}_Y\to {\mathcal S}_U\) surjective on objects? For many stacks this is true, but an example due to Raynaud and others shows that it fails for the stack of abelian schemes, if the field of fractions \(K\) of \(O\) has characteristic zero. This contradicts some assertions made in the book of \textit{G. Faltings} and \textit{C.-L. Chai} [``Degeneration of abelian varieties'' (1990; Zbl 0744.14031)]: Chapter~V, \(6.4\), \(6.4'\) and \(6.8\) (this list is misprinted in the paper under review). It also raises the question of what conditions are needed on \(U\) and \(Y\) to ensure that every abelian scheme (or, respectively, every \(p\)-divisible group) can be extended from \(U\) to \(Y\). The scheme \(Y\) is said to be healthy (respectively, \(p\)-healthy) if such extensions exist as soon as \(U\supset Y_K\). Here these problems are addressed in the case that \(O\) is of mixed characteristic \((0,p)\), the other cases being already understood. If the index of ramification of \(O\) is \(e\leq p-2\) then \(Y\) is healthy and \(p\)-healthy (this is due to Faltings). The main theorem here is that if \(e=1\) then any regular formally smooth \(O\)-scheme is healthy and \(p\)-healthy. (In fact in the present context \(p\)-healthy implies healthy, as is also shown here.) The main theorem is only strictly new in the important case \(p=2\), but the proof given here works for all~\(p\). The proof itself is highly technical and although the exposition is clear it is impossible to give a useful short summary.