On the size of differential modules (Q1974884)

Let \( G \) be an \( n\times n \) matrix with coefficients in \( {\mathbb Q}(x) \) and consider the differential module \( {\mathcal M }\) associated to the system of linear differential equations \( X' =GX\). For each prime \( p\), one can define its ``\(p\)-adic generic radius of convergence'' \( R_{p}\). Following André's terminology, the differential module \( {\mathcal M }\) is said to be a \( G\)-module if its ``inverse global radius'', namely \( \rho({\mathcal M})=\sum_{p}\max(-\log(R_{p}),0)\), is finite. It is known that all differential modules ``coming from geometry'' are \( G\)-modules and the converse is conjectured. One can also define the size \( \sigma({\mathcal M}) \) of \( {\mathcal M}\). Roughly speaking, it is the mean of the (logarithmic) size of the Taylor coefficients of its generic solutions. The Bombieri-André theorem asserts that \( 0\leq\rho(G)-\sigma(G)\leq n-1 \) and implies that \( G\)-modules are also characterized by the finiteness of their size. The aim of this paper is to improve the knowledge of \( \rho(G)-\sigma(G) \) in connection with the orders of nilpotence \( m(p) \) of the mod \( p \) reduction of \( {\mathcal M }\) (up to finitely many primes \( p \) this is well defined). Especially that difference is computed when \( {\mathcal M }\) is an hypergeometric differential module. This paper is Dwork's last one. The way it gives a deep insight in the \( G\)-functions theory is a fine example of the cleverness of this great mathematician. It has been generalized to the several variables case by \textit{L. Di Vizio} [On the arithmetic size of linear differential equations. J. Algebra 242, No. 1, 31-59 (2001; Zbl 0998.12008)].