Good lattices of algebraic connections (with an appendix by Claude Sabbah) (Q1740285)

Let \(U\) be a smooth quasi-projective geometrically irreducible variety of dimension \(n\) defined over an algebraically closed field \(k\) of characteristic zero, and consider a good compactification \(j:U\hookrightarrow X\) of it (that is, an embedding to a smooth projective variety \(X\) such that the complement \(D=X\setminus U\) is a normal crossing divisor). Let also \((\mathcal{E}, \nabla)\) be a vector bundle on \(U\) with an integrable \(k\)-linear connection. The first main result of the paper under review involves a comparison between two similar (but a priori different) notions associated with the tuple \((U,(\mathcal{E},\nabla))\). The authors say that a tower \(E_0\subset E_1\subset\ldots\subset E_n\) of \(\mathcal{O}_X\)-coherent subsheaves of \(j_*\mathcal{E}\) is a logarithmic model of \((U,(\mathcal{E},\nabla))\) if the following holds: 1. The \(E_i\) are locally free lattices of \(j_*\mathcal{E}\),\newline 2. \(\nabla(\Omega_X^i(\log D)\otimes_{\mathcal{O}_X}E_i)\subset \Omega_X^{i+1}(\log D)\otimes_{\mathcal{O}_X} E_{i+1}\) for any \(1\leq i\leq n-1\),\newline 3. For every effective divisor \(\Delta\) with support in \(D\), the embeddings \[\Omega_X^\bullet(\log D)\otimes_{\mathcal{O}_X}E_\bullet\subset \Omega_X^\bullet(\log D)\otimes_{\mathcal{O}_X} E_\bullet(\Delta)\subset j_*(\Omega_U^\bullet\otimes_{\mathcal{O}_U}\mathcal{E})\] are quasi-isomorphisms. They also call \(E_\bullet\) a good model of \((U,(\mathcal{E},\nabla))\) if such tower fulfills the following instead: 1. \(j\) resolves the turning points of \((\mathcal{E},\nabla)\),\newline 2. \(E_0\) is the Deligne-Malgrange lattice of \(\mathcal{E}\) and \(E_{i+1}\) is the \(\mathcal{O}_X\)-coherent sheaf spanned by \(E_i\) and \(\Theta_X(\log D)E_i\) for every \(1\leq i\leq n-1\). When \(n=1\), \textit{P. Deligne} introduced in [Equations différentielles à points singuliers réguliers. Berlin-Heidelberg-New York: Springer-Verlag (1970; Zbl 0244.14004)] logarithmic models \(E_0\subseteq E_1\) in order to define and study the irregularity divisor of a connection on a curve. On the other hand, good models are known to exist thanks to the independent work of \textit{K. S. Kedlaya} on the existence of good formal structure for meromorphic connections (see, respectively, [J. Am. Math. Soc. 24, No. 1, 183--229 (2011; Zbl 1282.14037)] and [\textit{T. Mochizuki}, Wild harmonic bundles and wild pure twistor \(D\)-modules. Paris: Société Mathématique de France (SMF) (2011; Zbl 1245.32001)]). In this paper, the authors guarantee the existence of logarithmic models of connections \((\mathcal{E},\nabla)\) by showing that good models are logarithmic too. That is their main result, achieved using certain filtrations on the log-de Rham complexes above. They also provide a formula, predicted by Michael Groechenig, for the characteristic variety of the \(\mathcal{D}_X\)-module \(j_*\mathcal{E}\) in the group \(K_0(\mathcal{O}_X)\), writing it as an alternated sum of the classes \(\left[\omega_X^{-1}\otimes\Omega^i_X(\log D)\otimes E_i\right]\). After an introductory section, the authors explain in the second one the results of Mochizuki and Kedlaya (in the algebraic setting of the latter). They continue with two more sections in which they prove their main result and note some remarks about it. The paper ends with three more sections, the last two as an appendix by the second named author only, in which he tackles the formula for the characteristic variety of \(j_*\mathcal{E}\) and the necessary background to prove it.