Polynomial approximation in several variables (Q2297378)

In approximation theory one of the most interesting questions is to characterize the rate of approximation of functions by a system of polynomials in terms of some modulus of continuity or smoothness. For instance, if \(f\) is a continuous function on \([-1,1]\) and \(r\) is a positive integer number in [\textit{Z. Ditzian} and \textit{V. Totik}, Moduli of smoothness. New York: Springer Verlag (1987), Zbl 0666.41001], the \(\varphi\)-modulus of smoothness of order \(r\) given by \(\omega_{\varphi}^{r} (f,\delta)= \sup \{ |\Delta^{r}_{h \varphi(x)} f(x)|, x\in[-1,1], 0< h \leq \delta\}\) is introduced. There \(\varphi(x)= \sqrt{1-x^2}\) and \(\Delta^{r}_{h} f(x)= \sum_{k=0}^{r}(-1)^{k} \binom{r}{k} f(x+(\frac{r}{2}- k) h)\) is the \(r\)th symmetric difference at \(x.\) Thus, the error \(E_{n}(f)= \inf\{ ||f - P_{n}||\}\) of the best polynomial approximation in the sup-norm on \([-1,1]\) of a continuous function \(f\) by polynomials \(P_{n}\) of degree at most \(n\) satisfies the Jackson-type estimate \(E_{n}(f)\leq C_{r} \omega_{\varphi}^{r} (f, 1/n).\) The weak converse \(\omega^{r}_{\varphi} (f, 1/n) \leq \frac{C_{r}}{n^r} \sum_{k=0}^{n} \frac{E_{k}(f)}{(k+1)^{r-1}}\) also holds. The aim of the paper under review is the analysis of the analogues of the above results when you deal with continuous functions in some domains of \(\mathbb{R}^d\) and you search estimates of the error of best approximation on linear subspaces of polynomials of total degree \(n\). A closed set \(K\in \mathbb{R}^d\) is a convex polytope if it is the convex hull of finitely many points. The analogue of the \(\varphi\)-modulus of smoothness on \(K\) is introduced in Chapter 12 of the monograph by Ditzian and Totik and, according to a geometric approach, you must use lines in different directions in order to introduce \(r\)th symmetric differences at a point. The main result of the first part is presented as Theorem 1.1 where an upper bound of the error of the best approximation (a Jackson type estimate) is obtained in terms of the \(r\)th modulus of smoothness in the set of directions assuming \(n\geq rd.\) For simple polytopes and for cubes/parallelepipes in all dimensions a weaker result has been obtained when all directions are used in the modulus of smoothness. In the proof of such a theorem the author deals with polytopes satisfying the property that if you cut off a small pyramid around every vertex, then the remaining set is a simple polytope. The proof is done by induction on the dimension but you need the approximations to be given by linear operators and the modulus of smoothness should involve only edge-directions. Pyramidal convering and approximation on general polytopes constitute the basic background in order to analyze approximation on pyramides in order to get the statement of Theorem 1.1 for large degrees. In a next step, the proof of the Jackson type estimate on convex polytopes for small degree polynomial approximation is presented as Theorem 5.1. In the second part, best polynomial approximation in domains with smooth (algebraic) boundary is studied. Notice that this problem has been analyzed in the literature for balls. In such a case the rate of approximation has been characterized by using appropriate moduli of smoothness (see [\textit{F. Dai} and \textit{Y. Xu}, Approximation theory and harmonic analysis on spheres and balls. New York, NY: Springer (2013; Zbl 1275.42001)]). For algebraic domains, the Jackson type theorem and its converse are stated as Theorems 6.5 and 6.6, respectively. Their proofs are based on the reduction to an approximation problem on suitable sets attached to the boundary, by taking into account the modulus of smoothness on the whole domain. Finally, the case when the domain is a disk in \(\mathbb{R}^2\) is analyzed. The motivation and geometric approach to polynomial best approximation on domains in the multivariate case presented in this interesting paper constitute a remarkable and stimulating starting point for future developments on these topics.