On a Ditzian-Totik theorem (Q1329028)

The so-called Bernstein interpolation process is considered: \(B_ n (f,x) = \sum^ n_{k=1} f(x_ k) m_ k(x)\), where \(f(x) \in C[-1,1]\), \(m_ 1(x) = {1 \over 4} [3 \ell_ 1(x) + \ell_ 2 (x)]\), \(m_ n(x) = {1 \over 4} [\ell_{n-1} (x) + 3 \ell_ n(x)]\), \(m_ k(x) = {1 \over 4} [\ell_{k-1} (x) + 2 \ell_ k (x) + \ell_{k+1} (x)]\), \(k=2, \dots, n-1\); and \(\ell_ k(x) = (-1)^{k+1} (1-x^ 2_ k) T_ n (x)/n(x- x_ k)\) \((k=1, \dots, n)\) are the fundamental polynomials of Lagrange interpolation at zeros of Chebyshev polynomials \(T_ n(x)\). The main result is the following Theorem. If \(f \in C[-1,1]\), \(0 \leq \lambda \leq 1\), then we have \[ \bigl | B_ n (f,x) - f(x) \bigr | \leq C(\lambda) \omega_{\varphi^ \lambda} \bigl( f_ j n^{-1} \delta_ n(x)^{1 - \lambda} \bigr), \] where \(\varphi (x) - \sqrt{1-x^ 2}\), \(\delta_ n(x) = n^{-1} + \sqrt {1-x^ 2}\), \(c(\lambda)\) is a constant depending only on \(\lambda\), and \(\omega_{\varphi^ \lambda}\) is the modulus of Ditzian-Totik. This theorem improves a weaker version of the above result due to Ditzian and Jiang which only asserted the existence of a sequence of polynomials of degree \(n\) satisfying the above inequalities.