Let $f:X\to S$ be a smooth proper morphism of schemes. If $S$ is of characteristic zero (i.e., $S$ is a $\mathbb{Q}$-scheme), then Deligne has shown:

The Hodge-De Rham spectral sequence $E^{a,b}_1=R^af∗Ω^b(X/S)⇒H^{a+b}_{DR}(X/S)$ degenerates in $E_1$.

This is also true when $S=Spec(k)$ with $k$ a perfect field of characteristic $p$, and $X$ has a smooth $W_2(k)$-lifing. Here $W_2(k)$ is the length 2 truncated Witt ring with values in $k$.

My question is when $S= Spec(W_2(k))$, and $X$ has a smooth $W_3(k)$-lifting, does the statement of $E_1$ degeneration still hold (maybe one needs more conditions)? Has anyone worked on this kind of question before?

  • $\begingroup$ There is something in this direction in the paper by Deligne-Illusie: see eq. (2.2.1) p. 254 in their paper in Invent. Math. 89. $\endgroup$ – Damian Rössler Feb 27 '14 at 8:49

First of all, the statement of Deligne-Illusie as you've given it is slightly wrong; there is only degeneration in degree $\leq \dim(X)$ (and thus full degeneration if $\dim(X)\leq p$).

Yukiyoshi Nakkajima proves an analogous result of the sort you'd like (extending work of Ogus) under the condition that the smooth proper scheme $X/W_n(k), n\geq 2$ admits a Frobenius lift. Here is the statement:

Theorem. Let $X$ be a smooth proper scheme over $W_n(k)$ ($k$ perfect of characteristic $p$, $n\geq 2$). If $X$ admits a lift of the Frobenius of $X\otimes_{W_n(k)} k$ and has a smooth lifting $\tilde{X}$ over $W_{n+1}(k)$, then the Hodge-de Rahm spectral sequence $$E^{i,j}_1=H^j(X, \Omega^i_{X/W_n(k)})\implies H^{i+j}_{dR}(X/W_n(k))$$ degenerates at $E_1$.

Note that unlike the Deligne-Illusie result, there is no condition on the dimension of $X$ or the degree of the Hodge cohomology; rather the whole spectral sequence degenerates. On the other hand, this result requires the (strong) condition of a Frobenius lift, which is vacuous for Deligne-Illusie (where $n=1$).

Ogus proves a similar result (with a degree condition) in 8.2.6 of "F-crystals, Griffiths Transversality, and the Hodge Decomposition."

| cite | improve this answer | |
  • 1
    $\begingroup$ To put a condition on Frobenius lift is too strong. For instance, it forces ordinarity. Also I think that the result you are referring to is basically already in Th. 8.8 of the notes of Berthelot-Ogus on crystalline cohomology. $\endgroup$ – Damian Rössler Mar 4 '14 at 8:52
  • $\begingroup$ I don't see how to deduce this from Theorem 8.8 in Berthelot-Ogus, which doesn't mention a lift to $W_{n+1}$ at all. This theorem in degree $\leq p$ is due to Ogus, as I observe. I'm also confused by your ordinarity comment; Frobenius on a supersingular elliptic curve lifts, by Deuring. Are you sure you're not thinking of Frobenius splitting? $\endgroup$ – Daniel Litt Mar 4 '14 at 9:07
  • 1
    $\begingroup$ That said, I agree that the condition of a Frobenius lift is strong, as I note in my answer. $\endgroup$ – Daniel Litt Mar 4 '14 at 9:08
  • 1
    $\begingroup$ In Berthelot-Ogus, it is supposed that the Frobenius lifts to an entire formal lifting, which is a stronger hypothesis. Nevertheless, the proof of Th. 8.8 in Berthelot-Ogus does not depend on the whole strength of this hypothesis, I think (that is why I wrote 'essentially'). As far as supersingular elliptic curves are concerned, the result of Deuring shows that the curve lifts to a CM curve and therefore a certain power of Frobenius lifts (if $k$ is a finite field). Nevertheless, the lifting will in general be defined over a ramified extension, unlike here. ... $\endgroup$ – Damian Rössler Mar 4 '14 at 17:13
  • $\begingroup$ ... But if you require the Frobenius itself to lift to $W_2(k)$, then the curve has to be ordinary. See for instance Prop. 8.6 in 'Frobenius et dégénérescence de Hodge' by L. Illusie (this is also a result of Nakkajima). $\endgroup$ – Damian Rössler Mar 4 '14 at 17:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.