Polynomial approximation with exponential weights (Q1105797): Difference between revisions

Let \(w_{\lambda}(x)=\exp (-| x|^{\lambda})x\in R\), \(\lambda >0\) and for \(1\leq p\leq \infty\) let f be a measurable function such that \(fw_{\lambda}\in L_ p(R)\). The degree of best weighted polynomial approximation \(E_ n(f,w_{\lambda})_ p=\inf_{p_ n}\| w_{\lambda}(f-p_ n)\|_ p\) was characterized by \textit{G. Freud} and \textit{H. N. Mhaskar} [Ark. Mat. 21, 145-161 (1983; Zbl 0536.41005)] and by \textit{Z. Ditzian} and \textit{V. Totik} [J. Approximation Theory 46, 38-41 (1986; Zbl 0606.41010)]. In particular it is known that for \(1\leq p\leq \infty\), \(\lambda >1\) and \(0<\alpha <r\) (r an integer), \(E_ n(f,w_{\lambda})_ p=O(n^{-\alpha (\lambda -1)/\lambda})\) if and only if \(\| X_ jw_{\lambda}\Delta^ r_ hf\|_ p=O(h^{\alpha})\) \((h\to 0+)\) where \(X_ h\) is the characteristic function of \((-h^{1/1-\lambda},h^{1/1-\lambda}).\) Most of this paper is devoted to characterizing those measurable functions f which satisfy \(\| w_{\lambda}\Delta^ r_ hf\|_ p=O(h^{\alpha})\) \((h\to 0^+)\) by means of the degree of some weighted polynomial approximation. However, there is also an interesting result characterizing the degree of weighted polynomial approximation of f such that \(fw_ 1\in L_ 1(R)\). Theorem: Let \(0<\alpha <r\) (r an integer) and let \(fw_ 1\in L_ 1(R)\). Then \(E_ n(f,w_ 1)_ 1=O((\log n)^{-\alpha})\) if and only if \(\| w_ 1\Delta^ r_ hf\|_ 1=O(h^{\alpha})\) and \(\| \zeta_ hw_ 1f\|_ 1=O(h^{\alpha}))\) \((h\to 0+)\) where \(\zeta_ h\) is the characteristic function of \(T\setminus (-\exp 1/h,\exp 1/h)\).