Sampling scattered data with Bernstein polynomials: stochastic and deterministic error estimates (Q1946479)

In this interesting paper, the authors study the convergence properties of the following generalization of the classical Bernstein polynomials \[ B_{n}^{A}(f)(x)=\sum_{k=0}^{n}p_{n, k}(x)f(x_{n, k}), \] where \(p_{n, k}(x)={n \choose k}x^{k}(1-x)^{n-k}\) and \(A=(x_{n, k})_{0\leq k\leq n, n\in \mathbb{N}}\) is a triangular array in \([0, 1]\) such that, for each \(n\in \mathbb{N}\), the numbers \(x_{n, k}\in [0, 1]\) are arranged in ascending order after \(k=0, 1,\dotsc, n\). One says that \(A=(x_{n, k})_{0\leq k\leq n, n\in \mathbb{N}}\) is uniformly distributed in \([0, 1]\) if, for all \(0\leq \alpha < \beta \leq 1\), we have \(\lim_{n\to \infty}\frac{\operatorname{card}\{x_{n, k}\in [\alpha, \beta]\}}{n}=\beta-\alpha\). The star discrepancy of \(A\) is defined in the paper as \(D_{n}^{*}(A)=\sup_{0\leq \alpha <1}\left |\frac{\operatorname{card}\{x_{n, k}\in [\alpha, \beta]\}}{n} -\alpha \right |\). In Section 2, the probability aspect of the convergence based on Chebyshev's inequality is treated. The main results of the following sections of the paper can be summarized as follows. (i) If \(A\) is uniformly distributed then, for each \(f\in C[0, 1]\), we have the estimate \[ |B_{n}^{A}(f)(x)-f(x)|\leq 2\max\{\omega(f; D_{n}^{*}), 1.1\omega(f; 1/\sqrt{n})\}, \;x\in [0, 1], \;n\in \mathbb{N}. \] Here, \(\omega(f ; h)\) denotes the uniform modulus of continuity and \(C[0, 1]\) is the class of continuous functions on \([0, 1]\). (ii) If \(A\) is uniformly distributed, then \(B_{n}^{A}(f)\) converges uniformly to \(f\in C[0, 1]\). (iii) If \(A=(x_{n, k})_{0\leq k\leq n, n\in \mathbb{N}}\) is a triangular array in \([0, 1]\) with the property that \(\lim_{n\to \infty}\frac{1}{n+1} \sum_{k=0}^{n}|x_{n, k}-\frac{k}{n}|=0\), then, for each \(f\in C[0, 1]\), we have \[ \lim_{n\to \infty}\|B_{n}^{A}(f)-f\|_{L^{1}}=0, \] where \(\|\cdot \|_{L^{1}}\) denotes the \(L^{1}\)-norm. (iv) Let \(1\leq p <\infty\) and \(A=(x_{n, k})_{0\leq k\leq n, n\in {\mathbb N}}\). Assume that \(\lim_{n\to \infty}\sum_{k=0}^{n}W_{n, k}^{(p)} \left |x_{n, k}-\frac{k}{n}\right |=0\), where \(W_{n, k}^{(p)}=\frac{(n+1)^{(p-3)/p}}{(k+1)^{(p-1)/2p}(n-k+1)^{(p-1)/2p}}\). Then, for each \(f\in \operatorname{Lip}^1\), we have \[ \|B_{n}^{A}(f)-f\|_{L^{p}}=0; \] here, \(\operatorname{Lip}^1\) denotes the Lipschitz class of order one on \([0, 1]\).