Purity for overconvergence (Q657837)

The basic Riemann-Hilbert correspondence relates vector bundles with integrable connection on a smooth manifold to representations of its fundamental group. In characteristic \(p\) geometry, modules with integrable connection are not well-behaved (for example, they can have nonzero Chern classes) and there is a complex hierarchy of notions of ``locally constant objects'': crystals and isocrystals (Grothendieck and Berthelot), convergent isocrystals (Berthelot and Ogus), overconvergent isocrystals (Berthelot) etc. Naturally, one would like to establish for them analogues of the well-known results for the fundamental group. Zariski-Nagata purity says that removing a codimension at least two closed subset from a smooth variety does not change its (algebraic) fundamental group. The main result of the paper under review is the following analogue of Zariski-Nagata purity for overconvergent F-isocrystals (Theorem 3.1): Let \((X, \bar X)\) be a smooth pair over a field of positive characteristic and let \(\bar Z\subseteq Z\subseteq \bar X\) be closed subsets of codimension at least two. Then the restriction functor between the categories of overconvergent F-isocrystals on \((X, \bar X)\) and \((X\setminus Z, \bar X\setminus \bar Z)\) is an equivalence. This theorem is almost a generalization of a result of Kedlaya, which deals with the case of \(\bar Z =\emptyset\) (though with no regularity assumptions on \(\bar X\)). Its proof is a rather long and complex sequence of reductions, consisting of 11 steps, eventually reducing the proof to the case of an affine space and coordinate hyperplanes. The author gives the following Riemann-Hilbert type result as an application of his theorem. It is known (Crew) that unit-root convergent F-isocrystals correspond to representations of the étale fundamental group with coefficients in a certain \(p\)-adic field \(K^\sigma\). This has been generalized to the overconvergent case by Tsuzuki (for X a curve) and Kedlaya (arbitrary dimension), overconvergent F-isocrystals being identified under Crew's correspondence with representations having ``finite local monodromy'' (or being ``potentially unramified''). The author finds a new, simpler definition of ``finite local monodromy'', depending in the smooth pair case only on the finitely many valuations given by the boundary divisors, and proves that it coincides with Kedlaya's definition. The proof uses de Jong's alterations to reduce to the smooth pair case and use Theorem 3.1.