Best uniform rational approximation of \(x^ \alpha\) on \([0,1]\). (Q1886148)

The author proves a strong error estimate for the uniform rational approximation of \(x^\alpha\) on \([0,1]\) for arbitrary real \(\alpha>0\)~: If \(E_{nn}(x^\alpha,[0,1])\) denotes the minimal approximation error in the uniform norm for rational functions with both numerator and denominator degree at most \(n\), it is proven that \[ \lim_{n\to\infty} e^{2\pi\sqrt{\alpha n}} E_{nn}(x^\alpha,[0,1]) = 4^{1+\alpha}|\sin\pi\alpha| \] for each \(\alpha>0\). This presentation concludes a series of mathematical contributions initiated by \textit{D. J. Newman} [``Rational approximation to \(|x|\)'', Mich. Math. J. 11, 11--14 (1964; Zbl 0138.04402)], who was the first to show that uniform rational best approximations to \(|x|\) on \([-1,1]\) possess an asymptotically much faster decaying error than uniform polynomial best approximations (note that \(E_{nn}(x^\alpha,[0,1]) = E_{2n,2n}(|x|^{2\alpha},[-1,1])\)). One highlight in this series was the author's proof of a strong asymptotic error estimate for uniform rational best approximation of \(|x|\) on \([-1,1]\), which is a special case of the result discussed here (\(\alpha=1/2\)) [``Best uniform rational approximations of \(|x|\) on \([-1,1]\)'', Russ. Acad. Sci., Sb., Math. 76, No. 2, 461--487 (1993; Zbl 0785.41019); translation from Mat. Sb. 183, No. 8, 85--118 (1992)]. The concrete form of the right hand side of the above strong error estimate has first been conjectured by \textit{R. S. Varga} and \textit{A. J. Carpenter} [``Some numerical results on best uniform approximation of \(x^\alpha\) on \([0,1]\)'', Numer. Algorithms 2, 171--185, (1992; Zbl 0763.41025)] and independently found by the author [``Best uniform rational approximation of \(x^\alpha\) on \([0,1]\)'', Bull. Am. Math. Soc. (New Ser.) 28, 116--122 (1993; Zbl 0768.41015)]. It is quite remarkable that the corresponding limit in the case of uniform polynomial best approximation of \(|x|\) on [-1,1] is still unknown: \textit{R. S. Varga} and \textit{A. J. Carpenter} [``On the Bernstein conjecture in approximation theory'', Constructive Approximation 1, 333--348 (1985; Zbl 0648.41013)] disproved S. Bernstein's conjecture on this limit by high precision numerical computations. The proof of the strong error estimate for uniform rational best approximation impressively combines methods from approximation theory like Chebyshev's theorem on alternation points, explicit integral representations of the approximation error and potential theoretic methods, particularly the representation of certain functions as Green potentials with respect to suitable measures.