On \(E_1\)-degeneration for the special fiber of a semistable family (Q2192770)

Let \(k\) be a perfect field of characteristic \(p>0\). For a variety \(X\) over \(k\) denote by \(F:X\to X'\) the relative Frobenius morphism. \textit{P. Deligne} and \textit{L. Illusie} proved in [Invent. Math. 89, 247--270 (1987; Zbl 0632.14017)] that the truncated de Rham complex \(\tau^{<p}F_*\Omega^{\bullet}_{X/k}\) of a smooth \(X\) admits a decomposition \[\tau^{< p}F_*\Omega^{\bullet}_{X/k}\simeq\bigoplus\limits_{i=0}^{p-1}\Omega^i_{X'/k}\tag{1}\] in the derived category of quasicoherent \(O_{X'}\)-modules if and only if \(X\) admits a lift to a flat scheme over \(W_2(k)\). In particular, if \(X\) is proper of dimension less than \(p\) and admits such lift then the Hodge-to de Rham spectral sequence for \(X\) degenerates at the first page. The authors investigate how these results extend to the case where \(X\) is a log-scheme arising as the special fiber of a semi-stable family over \(W(k)\). To fix notation, let \(X\) be a semi-stable scheme over \(W(k)\), viewed as a log scheme over the base \((W(k),\mathbb N:1\mapsto p)\) and \(X\) be its special fiber considered as a log scheme over \((k, \mathbb N:1\mapsto 0)\). Building on the previous work of \textit{K. Kato} [in: Algebraic analysis, geometry, and number theory: proceedings of the JAMI inaugural conference, held at Baltimore, MD, USA, May 16--19, 1988. Baltimore: Johns Hopkins University Press. 191--224 (1989; Zbl 0776.14004)], the authors prove that the truncation \(\tau^{<p}F_*\Omega^{\bullet}_{X/k}\) of the logarithmic de Rham complex admits a decomposition (1) if and only if \(X\) lifts to a log smooth log scheme over the base \((W_2(k),\mathbb N:1\mapsto 0)\). Note that by the nature of the setup, the log scheme \(X\) admits a lift to a log smooth log scheme over \((W_2(k),\mathbb N:1\mapsto p)\) given by \(X\times_{W(k)}W_2(k)\) but this lift does not necessarily provide a lift living over \(W_2(k)\) equipped with the log structure of \(\mathbb N:1\mapsto 0\). In fact, the authors prove that if \(X\) has several irreducible components \(X=\bigcup X_i\) then a lift \(X\) of \(X\) over \((W_2(k),\mathbb N:1\mapsto 0)\) has to decompose into irreducible components \(X=\bigcup X_i\) as well, each \(X_i\) being a lift of the corresponding \(X_i\). This observation enables the authors to show that the decomposition (1) might fail to exist in the present setting. As a counterexample, one can take \(X\) to be the blow-up of a projective space \(\mathbb P^n_{W(k)}\) centered in a subvariety \(Z\subset \mathbb P^n_k\) of the special fiber that does not admit a lift over \(W_2(k)\). In this case \(X=X\times_{W(k)}k\) contains an irreducible component isomorphic to a projective bundle over \(Z\), hence \(X\) cannot be lifted to a log smooth log scheme over \((W_2(k),\mathbb N:1\mapsto 0)\). On the positive side, the authors prove that such construction cannot possibly give rise to an example where (in the case of a proper \(X\)) the Hodge-to de Rham spectral sequence for \(X\) does not degenerate at the first page. They prove that both Hodge and de Rham cohomology groups of \(X\) remain unchanged if one replaces \(X\) by a blow-up of a nice enough subvariety of the special fiber, therefore showing that the degeneration property of the Hodge-to de Rham spectral sequence is stable under such blow-ups. This leads the authors to conclude that, assuming an appropriate version of the resolution of singularities, degeneration of the Hodge-to de Rham spectral sequence for the special fiber of \(X\) depends only on the generic fiber of \(X\).