Descent of isocrystals and \({\mathcal D}^\dag\)-modules (Q1284631): Difference between revisions

The existence of a ``weak Frobenius structure'' is a basic tool when studying \(p\)-adic differential equations. This paper explains how to translate it in the framework of \( {\mathcal D}^{\dag}\)-module theory. This leads to an explicit version of a Berthelot construction known as ``descent by Frobenius''. Let \( K \) be a discretely valued field with inequal characteristics and let \( {\mathcal V }\) be its valuation ring. Let \( {\mathcal X }\) be a smooth \( {\mathcal V }\) formal scheme of dimension one (resp. let \( {\mathcal A}(I) \) be the ring of functions analytic on some annulus \( I \) of the projective line over \( K\)) and let \( {\mathcal X}' \) (resp. \( {\mathcal A}(I')\)) be its image by a lifting \( F \) of the Frobenius. Given an isocrystal \( E \) defined over the rigid space \( {\mathcal X}_{K} \) associated to \( {\mathcal X }\) (resp. a free of finite rank \( {\mathcal A}(I)\)-module with connection), the problem is to find an isocrystal over \( {\mathcal X}'_{K} \) (resp. an \( {\mathcal A}(I')\)-module with connection) \( E' \) such that \( E \) is isomorphic to \( F^{*}(E')\). It happens that this is possible if and only if the radius of convergence of \( E \) is large enough. Successively the cases where \( E \) has no singularity on \( {\mathcal X}\), has one logarithmic singularity, has one singularity and is overconvergent are studied. The main idea is to remark that the ``classical'' matrix that enables to construct \( E' \) can be viewed as the action of a global infinite differential operator \( H\). Actually, even if not clear a priori, the problem is basically local. The introduction of a global notion emphasizes functorial properties but every computation needs to use a local coordinate. Then proofs follow exactly the same line as in the differential equation case. This is specially clear in the case of a logarithmic singularity where a sharper insight is given. The last section is devoted to Fuchsian overconvergent isocrystals. Its first part faithfully follows Dwork's computations [\textit{B. M. Dwork}, J. Reine Angew. Math. 484, 85-126 (1997; Zbl 0870.12008)]. Curiously, although Dwork's aim was to get a simplified definition for exponents, that definition is not recalled here. Then, missing this basic notion, the author is forced to state and prove the reduction theorem under an unusable equicontinuity condition : There is no known example where that condition is easiest to check than the statement itself. Why to avoid the ``right hypothesis'', namely non Liouvility of differences of exponents, that is known to be true for all interesting isocrystals (those coming from geometry) and to imply the reduction theorem?