Arithmetic differential equations and \(E\)-functions (Q819225)

Let \(K\) be a number field, and let \(\phi \in K[X,d/dX]\) be a differential operator. The Fourier transform of \(\phi\) is defined by replacing \(X\) with \(d/dX\) and \(d/dX\) with \(-X\). Recall that \(\phi\) is called a \(G\)-operator if it is the minimal differential operator annihilating a \(G\)-function (see for instance 0830.12004 Dwork, Bernard; Gerotto, Giovanni; Sullivan, Francis J. An introduction to \(G\)-functions) and recall also that \(\psi \in K[X,d/dX]\) is called an \(E\)-operator if it is the Fourier transform of a \(G\)-operator. These operators are important because they annihilate certain arithmetic Gevrey series, by a result of \textit{Y. André} [Ann. Math. (2) 151, No. 2, 705--740 (2000; Zbl 1037.11049)]. The main result of the paper (theorem 3.1) is the following characterization of \(E\)-operators (here \(A_\psi\) is the companion matrix of \(\psi\)). A differential operator \(\psi \in K[X,d/dX]\) of rank \(\mu\) is an \(E\)-operator if and only if (1) The coefficients of \(\psi\) are not all in \(K\) (2) The slopes of the Newton-Ramis polygon of \(\psi\) lie in \(\{-1,0\}\) (3) The differential system \(dZ/dX = A_\psi Z\) has a solution of the form \(Y(1/X) \cdot (1/X)^\Gamma \cdot \exp(-\Delta X)\) where \(Y(X) \in \mathrm{GL}_\mu(K((X)))\) satisfies \(\prod_{v} \min(R_v(Y)\pi_v,1) \neq 0\) (the product being taken over all finite places \(v\) of \(K\) and \(\pi_v\) being a uniformizer), where \(\Gamma \in \text{M}_\mu(K)\) has eigenvalues in \(\mathbb{Q}\), and where \(\Delta \in \mathrm{M}_\mu(K)\) is diagonal and commutes with \(\Gamma\). Section 3 of the paper is devoted to the proof that an \(E\)-operator satisfies (1), (2) and (3) while the rest of the paper is devoted to the proof of the converse statement. An important ingredient is the formal Laplace transform.