Arithmetic properties of power series solutions of algebraic differential equations (Q1167764)

In this paper the authors apply a sophisticated technique to the \(p\)-adic theory of differential equations, refining Newton's iteration procedure that was originally used by J. Moser and V. I. Arnol'd to deal with the problem of the so-called ``small divisors'' appearing in the formal solutions of the nonlinear differential equations of celestial mechanics. The method is based upon the replacement of a system of nonlinear differential equations, brought to one of certain canonical forms, by a sequence of linear systems whose solutions make up a series uniformly converging to a solution of the original system. The main \(p\)-adic ingredients that make work this procedure are a sharp estimate due to Dwork concerning the radii of \(p\)-adic convergence of the solutions of a linear system at an ordinary point and a result of Clark, that gives a sufficient condition for a power series solution of a linear system at a singular point to converge \(p\)-adically. The main result of this paper is the following theorem: Let \(\overline{\mathbb Q}\) be the algebraic closure of the rational numbers. Let \(y= \sum_{n=0}^\infty a_nx^n\), \(a_n\in\overline{\mathbb Q}\), be a formal solution of an algebraic differential equation. (That is, assume that there exists \(F(x_1,x_2,\ldots,x_{n+2}) \in \mathbb Q[x_1,\ldots,x_{n+2}]\) such that \(F(x,y,y',\ldots,y^{(n)})=0\).) Then \(y\) has a positive \(\nu\)-adic radius of convergence for every non-archimedean valuation \(\vert\cdot\vert_\nu\) of \(\mathbb Q\). A function \(y\) not satisfying any algebraic differential equation is called ``transcendentally transcendental''. Therefore the above theorem may be viewed as giving a series of sufficient conditions for a function \(y= \sum_{n=0}^\infty a_nx^n\) to be transcendentally transcendental. This is the case, for example, if for some valuation \(\vert\cdot\vert_\nu\) of \(\overline{\mathbb Q}\), \(y\) does not satisfy an estimate of the type: \(\vert a_m\vert_\nu\le e^{c(\nu)m}\) for all \(m>0\), \(c(\nu) = \) a constant independent of \(m\). Although effective estimates do not appear in the main theorem reported above, most of the intermediate steps of the elaborate proof give explicit estimates for the \(\nu\)-adic size of the coefficients of power series solutions of systems in certain basic standard forms. The authors suggest that these estimates may be of use in refining the archimedean estimates of Popken-Mahler, which imply results of transcendence. The authors also give a generalization of their main result to Pfaffian partial differential systems.