Cut-by-curves criterion for the log extendability of overconvergent isocrystals (Q641885)

The paper, together with its companion paper [\textit{A. Shiho}, Math. Ann. 348, No. 2, 467--512 (2010; Zbl 1268.12005)] provides a \(p\)-adic analogue of the following classical result of \textit{P. Deligne} [Equations différentielles à points singuliers réguliers. Berlin-Heidelberg-New York: Springer-Verlag (1970; Zbl 0244.14004)]: Let \(\bar X\) be a smooth complex algebraic variety, \(Z\subseteq \bar X\) a divisor with normal crossings, \(E\) a vector bundle with an integrable connection on \(X:=\bar X - Z\). Then (1) \(E\) extends to a vector bundle with a connection with logarithmic poles on \(\bar X\) if and only if \(E\) has regular singularities, (2) \(E\) has regular singularities if and only if for any locally closed smooth curve \(C\) in \(X\), \(E|_C\) has regular singularities. In the setting of the two papers, the field of complex numbers is replaced by a (here assumed uncountable) characteristic \(p\) field \(k\), vector bundles with an integrable connection -- by overconvergent isocrystals, regular singularities -- by ``\(\Sigma\)-unipotent monodromy'', connection with logarithmic poles -- by isocrystals on the log convergent site of \((\bar X, X)\) with ``exponents in \(\Sigma\)''. Under this dictionary, the author's paper [loc. cit., Theorem 3.16] provided an analogue of (1) and the paper under review gives an analogue of the ``cut-by-curves criterion'' (2) (Theorem 0.1). Here \(\Sigma\) is a collection of sets of \(p\)-adic integers assigned to the irreducible components of \(Z\) with non-integer, non-\(p\)-adically Liouville differences. The proof makes heavy use of the theory of \(p\)-adic differential equations, namely the study of the \(\nabla\)-modules on \(p\)-adic annuli around the generic points of the components of \(Z\), based on methods and notions (highest ramification break and exponent of a \(\nabla\)-module) due to Christol, Mebkhout, Dwork, and Kedlaya. The author provides a useful review of their results as well as the results and definitions of the companion [loc. cit.] in the first section of the paper. In the course of the proof the author finds ``cut-by curves criteria'' for solvability, highest ramification break and exponent of a \(\nabla\)-module. Finally, he shows (Corollary 2.12) that one can remove the uncountability assumption of \(k\) at the expense of considering curves \(C\) defined over all extensions of \(k\).