Continuity of the radius of convergence of differential equations on \(p\)-adic analytic curves (Q607716)

The author studies meromorphic connections \((\mathcal F,\nabla)\) on compact rig-smooth curves \(X\) within Berkovich's category of non-archimedean analytic spaces over non-archimedean extensions \(k\) of \(\mathbb Q_p\). For any semi-stable model \(\mathfrak X\) of \(X\), he defines the \(\mathfrak X\)-normalized radius of convergence \(R_{\mathfrak X}(\mathcal F,\nabla)\) of \((\mathcal F,\nabla)\) as a \((0,1]\)-valued function on \(X\); he proves that if \(\mathcal F\) has a model on \(\mathfrak X\), then \(R_{\mathfrak X}(\mathcal F,\nabla)\) is continuous, logarithmically concave and logarithmically piecewise linear. Under certain additional assumptions on the singularities of \((\mathcal F,\nabla)\), he proves that \((\mathcal F,\nabla)\) is a Robba connection (which means that \((\mathcal F,\nabla)\) is trivial on every open disc in \(X_{k'}\), for every analytic extension field \(k'\) of \(k\)) if and only if \(R_{\mathfrak X}(\mathcal F,\nabla)\) equals \(1\) on the vertices of the skeleton \(S(\mathfrak X)\subseteq X\) associated to \(\mathfrak X\). As Baldassarri explains in the introduction of his article, the normalized radius of convergence \(R_{\mathfrak X}(\mathcal F,\nabla)\) can be defined in higher dimensions, for any connection \((\mathcal F,\nabla)\) on a rig-smooth analytic space \(X\) over a (possibly trivially valued) non-archimedean field \(k\) with respect to a given nondegenerate polystable formal model \(\mathfrak X\) of \(X\): if \(x\in X\) is a \(k\)-rational point and \(k\) is both non-trivially valued and algebraically closed, then one defines \(R_{\mathfrak X}(\mathcal F,\nabla)(x)\) to be the radius of the maximal open polydisc neighborhood of \(x\) in \(X\setminus S(\mathfrak X)\) where \((\mathcal F,\nabla)\) is trivial, and \(S(\mathfrak X)\subseteq X\) denotes the skeleton of \(X\); the definition is extended to the general case via extension of the base field. One conjectures in general that \(R_{\mathfrak X}(\mathcal F,\nabla)\) is continuous on \(X\) and that, if \(\mathcal F\) has a model on \(\mathfrak X\), the connection \((\mathcal F,\nabla)\) is a Robba connection in the above sense if and only if \(R_{\mathfrak X}(\mathcal F,\nabla)\) is identically \(1\) on \(X\). As we stated above, the present paper settles these conjectures in one-dimensional situations for fields \(k/\mathbb Q_p\), even under the presence of mildly singular connections. The zeroth section of the present article is of an introductory nature; it provides an extensive explanation of the problem and a detailed overview of the paper. For a relatively compact subdomain \(X\) of the analytic affine line \(\mathbb A^1_k\), Baldassarri explains how to relate the normalized radius of convergence function (relative to the analytic skeleton of \(X\)) to the naive non-normalized radius of convergence function \(x\mapsto R(x)\) (which depends on the embedding of \(X\) into \(\mathbb A^1_k\)): the two notions are related via the continuous diameter function. Baldassarri then discusses an example of Christol, for which he explicitly computes \(R(x)\) in all points \(x\) of \(\mathbb A^1_k\). It turns out that in this example, the non-normalized radius of convergence function \(R\) is controlled, via retraction, by a finite graph \(\Gamma\subseteq\mathbb A^1_k\) along whose branches it is constant or linear with respect to a natural parametrization. In the first section, the author reviews the structure of rig-smooth compact connected strictly \(k\)-analytic curves over a non-archimedean field \(k\) with a non-trivial valuation, as well as the parallel theory of semistable reduction of \(k\)-algebraic curves; for simplicity, he assumes \(k\) to be algebraically closed. In the second section, Baldassarri establishes a general criterion, which he attributes to Berkovich, to test whether an \(\mathbb R_{>0}\)-valued function on a smooth \(k\)-affinoid curve over a non-trivially valued non-archimedean field \(k\) is continuous. In Section \(3\), Baldassarri starts assuming that \(k\) has characteristic \(0\). He considers the complement \(X\) of a finite reduced effective divisor \(Z\) in a compact rig-smooth connected \(k\)-analytic curve \(\overline{X}\) as well as connections \((\mathcal F,\nabla)\) on \(X\) with meromorphic singularities along \(Z\). He assumes that \(\mathcal F\) has a model on a semi-stable formal model \(\overline{\mathfrak X}\) of \(\overline{X}\) such that the schematic closure \(\mathfrak Z\) of \(Z\) in \(\overline{\mathfrak X}\) is étale over the valuation ring of \(k\), and he assumes that \(k\) is an extension of \(\mathbb Q_p\). In this situation, he defines the normalized radius of convergence function on \(X\) attached to \((\mathcal F,\nabla)\) and \(\overline{\mathfrak X}\), and he studies its behavior under changes of the model \(\overline{\mathfrak X}\). In Section 4, the author recalls some aspects of the classical theory of differential systems on annuli. In Subsection 5.1, Baldassarri establishes what he calls the main technical tool of his present paper; it is a generalization of the Dwork-Robba theorem [\textit{B. Dwork, G. Gerotto} and \textit{F. J. Sullivan}, An introduction to \(G\)-functions. Annals of Mathematics Studies. 133. Princeton, NJ: Princeton University Press (1994; Zbl 0830.12004), Chapter \ IV, Theorem 3.1] on effective bounds for the growth of local solutions on basic annuli and discs. He then deduces the upper semicontinuity of his normalized radius of convergence function, using methods different from those which had been used in a more special situation by \textit{G. Christol} and \textit{B. Dwork} [Ann. Inst. Fourier 44, No. 3, 663--701 (1994; Zbl 0859.12004)]. In Subsection 5.3, the author shows that the normalized radius of convergence function is continuous on the skeleton of a basic affinoid annulus, and in Subsection 5.4, he discusses how the normalized radius of convergence function for basic annuli (which are not necessarily affinoid subdomains of \(\mathbb A^1_k\)) behaves under Dwork's technique of Frobenius descent. In Section 6, Baldassarri derives the desired continuity property of the normalized radius of convergence function in general, by first reducing to the case of basic affinoid discs or annuli and then using the techniques developed in the previous section. He also obtains a more precise description of the normalized radius of convergence, showing that it is logarithmically concave and logarithmically piecewise linear along the skeleton. Finally, Baldassarri makes use of his previous work on \(p\)-adic Turrittin theory [Adv. Math. 44, 155--179 (1982; Zbl 0493.12030)]: he formulates a condition NL for \((\mathcal F,\nabla)\) at points of \(Z\) by requiring certain formal Fuchs exponents to be \(p\)-adic non-Liouville numbers. Under this assumption, the results from [loc. cit.] relate the asymptotic behavior of the normalized radius of convergence of \((\mathcal F,\nabla)\) with respect to \(\overline{\mathfrak X}\) to the algebraic irregularity of \((\mathcal F,\nabla)\) at \(Z\) at the given point of the divisor. Baldassarri concludes that the normalized radius of convergence is identically \(1\) if and only if it is \(1\) at the vertices of the skeleton and the singularities along \(\mathfrak Z\) are regular. For more details, we refer to the comprehensive introduction of Baldassarri's paper. The article is very well written, and it contains many interesting and instructive examples.