Algebraic approximation preserving dimension (Q522152)

This paper is the fourth part of a joint work of the three authors. They deal with the approximation of semi-algebraic, analytic and semianalytic sets by algebraic sets in the sense of the Hausdorff distance plus other very interesting conditions. The previous papers are \textit{M. Ferrarotti} et al. [Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 1, No. 1, 1--11 (2002; Zbl 1051.14065); Proc. Am. Math. Soc. 138, No. 5, 1537--1548 (2010; Zbl 1189.14064); Proc. Am. Math. Soc. 143, No. 1, 13--23 (2015; Zbl 1319.14058)] of the bibliography, the reviewer would add here the work, by the same authors, [Pac. J. Math. 194, No. 2, 315--323 (2000; Zbl 1036.14027)], as it probably was an inspiration for the introduction of s-equivalence. The previous papers contain, in the bibliography, more of the history of the problem, to begin with [\textit{L. Bröcker}, Lect. Notes Math. 1524, 145--162 (1992; Zbl 0849.14022)]. The authors mentioned there as well the paper of Marcin Bilski (Kraków), who considered a similar type of approximation in the complex domain (see e.g. [\textit{M. Bilski}, Math. Z. 256, No. 4, 705--716 (2007; Zbl 1115.32005)]). The papers [Zbl 1051.14065; Zbl 1189.14064; Zbl 1319.14058] are also needed because some theorems used in the present work were proved there. They are easy to find, so this is no problem to the reader. The s-equivalence of two closed semialgebraic sets is a very interesting notion, introduced in Definition 2.1. It involves the Hausdorff distance between the intersections of these sets with the sphere \(S_r\) of radius \(r\), centered at a common point \(P\). This distance should tend to zero more rapidly that \(r^s\), where s is a given number not smaller than 1. If it does, the two sets are s-equivalent in \(P\) (one may also use germs here, as all the considerations are local). The main theorem of the paper (Theorem 4.1) states that for any semialgebraic closed set \( A \subset {\mathbb{R}}^n\) that is of non-zero codimension, for a point \(O\) chosen in \(A\) and for any given number \(s \geq 1\) there exists an algebraic set \(S\) in \({\mathbb{R}}^n\) such that \(A\) and \(S\) are s-equivalent in \(O\) and \(\dim_{O} A = \dim_{O} S\). As a corollary, the authors obtain an analogous result for semianalytic closed subsets of \({\mathbb{R}}^n\) (Corollary 4.3). The tools used to prove that are: one of the Łojasiewicz inequalities, some results from the authors' previous papers and a very clever `good presentation' of semialgebraic sets. All the Łojasiewicz inequalities are to be found in his pioneer work on semianalytic sets, whose nice and very useful transcript by Michel Coste is to be found online (see [\textit{S. Łojasiewicz}, Ensembles semi-analytiques. Lecture note I.H.E.S. , Bures-sur-Yvette; réproduit No A 66.765. Ecole Polytechnique, Paris (1965), \url{http://perso.univ-rennes1.fr/michel.coste/Lojasiewicz.pdf}]). An example of a `good presentation' is given in the introduction. The point is that the polynomials of the good presentation are used to construct the algebraic set \(S\). This idea is related to the transversality results of \textit{J. W. Bruce} and \textit{N. P. Kirk} [Bull. Lond. Math. Soc. 32, No. 6, 718--728 (2000; Zbl 1041.58018)]. Even if a good presentation needn't be admitted by any semialgebraic closed set, every such set is a finite reunion of closed semialgebraic sets, \(\Gamma_i\) that have this `good presentation' and whose dimension in \(O\) is the same as \(\dim_{O}A\) except for a small closed semialgebraic set \(\Gamma \subset A\) of dimension strictly lesser than \(\dim_{O}A\). As to the Łojasiewicz inequality used here, the authors adapted for their purpose this one: If \(f\) and \(g\) are two analytic functions defined in a neighbourhood of \(0\) and \(f^{-1}(0) \subset g^{-1}(0)\), then there exist constants \(C, \alpha>0\) and \(\varepsilon>0\) such that: \[ C{|g(x)|}^{\alpha} \leq |f(x)| \] for \(|x|<{\varepsilon}\). It is worth noticing that the adaptation is by no means obvious. The article ends with a nice Fig.1 illustrating such an algebraic approximation of a quadrant. As the authors mention in the introduction (and give a counterexample in [Zbl 1189.14064]) no such result as Corollary 4.3 is true for subanalytic sets. This is not surprising, as subanalytic sets don't even have this nice property of semianalytic sets, whose every germ is contained in an analytic germ of the same dimension. In this context, it seems also worth mentioning that the paper [\textit{Z. Denkowska} and \textit{M. P. Denkowski}, J. Math. Anal. Appl. 387, No. 1, 48--65 (2012; Zbl 1228.32010)] gives a similar dimension-preserving but semi-algebraic approximation for definable sets (with the dimension of the germ preserved and connected components controlled) in a weaker sense of the Kuratowski convergence of closed sets, i.e., of course, without s-equivalence.