Canonical integral structures on the de Rham cohomology of curves (Q848122)

Let \(R\) be a complete discrete valuation ring and \(K\) its fraction field. Let \(X_K\) be a smooth and proper curve over \(K\). Under some mild hypothesis, the author explains how to give the de Rham cohomology \(H^1_{\mathrm{dR}}(X_K/K)\) a canonical integral structure. The problem is complicated by the fact that, when \(X_K\) does not have a good reduction, the (relative) de Rham cohomology of a proper and flat (even regular) model can in general be rather pathological (for functoriality and under duality). One of the motivations of studying this integral structure is to understand the completion of \(\lim_{n \to \infty} H^1_{\mathrm{dR}}(X(p^n)/\mathbb Q_p)\), where \(X(p^n)\) is the modular curve of level \(p^n\). This completion is related to \(p\)-adic local Langlands correspondence. More specifically, the author assumes that \(X_K\) has an admissible model that is a proper relative curve \(X\) over \(R\), which is normal with generic fiber isomorphic to \(X_K\), satisfying the following two conditions: (1) \(H^1(X, \mathcal O_X)\) is torsion-free, and (2) \(X\) has rational singularities, i.e. there exists a resolution of singularities \(\rho: X' \to X\) such that \(R^1\rho_* \mathcal O_{X'} = 0\). This hypothesis can be verified in many cases, for example modular curves. The key point of this paper is to use the relative dualizing sheaf \(\omega_{X/R}\) instead of the differential sheaf \(\Omega^1_{X/R}\). The author defines the integral structure of \(H^1_{\mathrm{dR}}(X_K/K)\) to be the hypercohomology of the complex \(\mathcal O_X \to \omega_{X/R}\); it is a finite and free \(R\)-module independent of the choice of the admissible model \(X\). Moreover, the author checks that the integral structure (together with the induced filtration) is canonical in the sense that it is functorial in finite (generically étale) \(K\)-morphisms of \(X_K\) and is preserved by cup-product auto-duality on \(H^1_{\mathrm{dR}}(X_K/K)\). The author also shows that the canonical lattice naturally contains the lattice furnished by the (truncated) de Rham complex of a regular proper \(R\)-model of \(X_K\) and that the length of the cokernel for this inclusion of lattices is a numerical invariant of \(X_K\) (which he calls the de Rham conductor \(\mathrm{dRcond}(X_K/K)\)). In relation with prior works, the author proves that \(\mathrm{dRcond}(X_K/K)\) is bounded above by \(\mathrm{Art}(X_K/K)\), defined by \textit{S. Bloch} [Duke Math. J. 54, 295--308 (1987; Zbl 0632.14018)], and is bounded below by \(\mathrm{Effcond}(X_K/K)\), defined by \textit{Q. Liu} and \textit{T. Saito} [J. Algebr. Geom. 9, No. 3, 409--424 (2000; Zbl 0992.14008)]. The author also studies how the position of canonical lattice inside the de Rham cohomology of \(X_K\) is affected by finite extension of scalars. This paper is very interesting and well written.