Growth of log-analytic functions (Q6156373)

This article studies the lack of polynomially bounded growth of log-analytic functions at infinity. Log-analytic functions were introduced by \textit{J.-M. Lion} and \textit{J.-P. Rolin} [Ann. Inst. Fourier 47, No. 3, 859--884 (1997; Zbl 0873.32004)], as the smallest class that contains both global subanalytic functions and logarithm and that is closed by composition. This class was mostly studied in the late 1990s and early 2000s as an intermediate class between subanalytic functions and functions definable in \(\mathbb R_{\text{an},\exp}\). Log-analytic functions admit a preparation theorem [loc. cit.], which implies directly that a log-analytic function of one variable is bounded by some power function at infinity. In higher dimension, since finding the critical locus of a log-analytic function involves solving log-analytic equations, exponentials might intervene; and with exponentials involved, there might be no power of the radius \(r\) that bounds the maximal value of a log-analytic function \(f\) on the sphere of radius \(r\) (\(f\) continuous). The article gives an explicit example of this fact. The main theorem presented here shows that the locus where this happens is however ``small'' at infinity. Given a continuous log-analytic function \(f\), a set \(U\) can be constructed, that is definable in \(\mathbb R_{\text{an},\exp}\) and contains all germs of lines at infinity, in such a way the restriction of \(f\) to \(U\) is bounded by a power of the radius. The proof mostly relies on the preparation theorem for log-analytic functions together with elementary o-minimal geometry.