Error estimates for least squares approximation by polynomials (Q793250)

Let \(q_ 0,q_ 1,...\), be the normalized orthogonal polynomials associated with the distribution \(d\alpha\) on the fundamental interval [- 1,1]. The weighted least squares approximation to f is given by \(H[f]=\sum^{n}_{\nu =0}q_{\nu}\int^{1}_{- 1}f(t)q_{\nu}(t)d\alpha(t).\) Let \(\| \cdot \|\) mean the sup-norm on [-1,1]. The main result is the following: Theorem. Let \(d\alpha\) be a distribution with the following properties: (i) if f is any continuous function, then \(\int^{1}_{-1}f(x)d\alpha(x)=\int^{1}_{-1}f(- x)d\alpha(x),\) (ii) \(\| q_{\nu}\| =q_{\nu}(1)\), \(\nu =0,1,...,n+1\). If the derivative \(f^{(n+1)}\) exists, then \[ \| f- H[f]\|<(\| q_{n+1}\| /\| q_{n+1}^{(n+1)}\|)\| f^{(n+1)}\|. \]