On Lagrange interpolatory parabolas to \(|x|^{\alpha}\) at equally spaced nodes (Q1572844)

Let \(\{L_n(f,.)\}\) be the sequence of Lagrange interpolation polynomials which are based on equidistant nodes. It is a classical result that \(\{L_n(|x|,x_0)\}\) diverges for all \(x_0\in(-1,1)\), \(x_0\neq 0\) [\textit{S. Bernstein}, Math. Ann. 79, 1-12 (1918; JFM 46.0417.01)]. This result was generalized by the author showing that the same statement is true for \(\{L_n(|x|^\alpha,x_0)\}\), \(0<\alpha\leq 1\), \(x_0\in(-1,1)\), \(x_0\neq 0\) [\textit{M. Revers}, J. Approximation Theory 103, No. 2, 269-280 (2000; Zbl 0954.41002)]. What happens for \(x_0=0\)? D. L. Berman stated that for \(f(x)=|x|\) the sequence \(\{L_n(|x|,0)\}\) converges to \(0=f(0)\). (A surprising result!) Is this fact true for \(\{L_n(|x|^\alpha,0)\},0<\alpha\leq 1\)? That is the question which is studied in the present paper. The author establishes the estimations from below and from above for \(L_n(|x|^\alpha,0)\), \(n=2m-1\), \(m\in\mathbb{N}\), \(0\leq\alpha\leq 1\). Both the estimation turn out to be best possible.