On a p-adic invariant cycles theorem

From MaRDI portal
Publication:2634633



Abstract: For a proper semistable curve X over a DVR of mixed characteristics we reprove the "invariant cycles theorem" with trivial coefficients (see Chiarellotto, 1999) i.e. that the group of elements annihilated by the monodromy operator on the first de Rham cohomology group of the generic fiber of X coincides with the first rigid cohomology group of its special fiber, without the hypothesis that the residue field of calV is finite. This is done using the explicit description of the monodromy operator on the de Rham cohomology of the generic fiber of X with coefficients convergent F-isocrystals given in Coleman and Iovita (2010). We apply these ideas to the case where the coefficients are unipotent convergent F-isocrystals defined on the special fiber (without log-structure): we show that the invariant cycles theorem does not hold in general in this setting. Moreover we give a sufficient condition for the non exactness.


Let \(V\) be a discrete valuation ring with perfect residue field \(k\) of characteristic \(p>0\) and with fraction field \(K\) of characteristic \(0\). It contains the fraction field \(K_0\) of \(W(k)\). Let \(X\) be a proper regular scheme over \(V\) with semistable reduction, with special fibre \(X_k\) and generic fibre \(X_K\). The theory of log schemes and log crystalline cohomology allows us to define on the de Rham cohomology groups \(H^i_{\mathrm{dR}}(X_K)\) of \(X_K\) a nilpotent monodromy operator \(N\). On the other hand, we have the rigid cohomology groups \(H^i_{\mathrm{rig}}(X_k/K_0)\) of \(X_k\) (which are finite dimensional \(K_0\)-vector spaces), together with a natural map \(H^i_{\mathrm{rig}}(X_k/K_0)\otimes_{K_0}K{\iota}{\longrightarrow} H^i_{\mathrm{dR}}(X_K)\). The composition \[ H^i_{\mathrm{rig}}(X_k/K_0)\otimes_{K_0}K{\iota}{\longrightarrow} H^i_{\mathrm{dR}}(X_K){N}{\longrightarrow} H^i_{\mathrm{dR}}(X_K) \] is know to be the zero map. A fundamental question now is whether this sequence is actually exact. If \(k\) is finite then this exactness would be implied by the correctness of the monodromy weight conjecture in this context. As this conjecture is known to be true for \(X\) of (relative) dimension \(1\) and \(2\), the exactness of the sequence is known accordingly. For \(i=1\) one even obtains (for such \(X\)) an exact sequence \[ 0\to H^1_{\mathrm{rig}}(X_k/K_0)\otimes_{K_0}K{\iota}{\longrightarrow} H^1_{\mathrm{dR}}(X_K){N}{\longrightarrow} H^1_{\mathrm{dR}}(X_K).\tag{cyc} \] In the case of relative dimension \(1\) (i.e. semistable curves), several alternative definitions of \(N\) acting on \(H^1_{\mathrm{dR}}(X_K)\) have been given in past, as well as versions with nontrivial coefficients (i.e. in log-\(F\)-isocrystals). In the present paper, using one of these alternative definitions, the exactness of (cyc) is proven without assuming that \(k\) be finite. Moreover, the question is studied for which class of nontrivial coefficients (assumed to be induced from convergent isocrystals on the special fibre \(X_k\)) exactness of the corresponding sequence analogous to (cyc) can be expected. For unipotent convergent isocrystals, a criterion for exactness of the sequence is given; it shows in particular that exactness often fails to hold.











This page was built for publication: On a \(p\)-adic invariant cycles theorem

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2634633)