On the fine and rough theory of Lagrange type interpolation of higher order (Q1971923)

Given an odd number \(m\), let \(X=\{x_{nq}\}\) be a triangular matrix of nodes and let \(A_{jk}\) be the fundamental Hermite polynomials of degree \(nm-1\), i.e., \(A^{(p)}_{jk}(x_{nq})=\delta_{jp}\delta_{kq}\), \(j,p=0,\dots,m-1\), \(k,q=1,\dots,n\). The Hermite interpolation operator for \(f\in C[-1,1]\) is defined by \(H_{nm}(X,f,x):=\sum_{j=0}^{m-1}\sum_{k=1}^nf(x_{nk})A_{jk} (x)\) and for odd \(m\) it is called of Lagrange-type. \textit{J. Szabados} [Acta Math. Hungar. 61, No.~3-4, 357--368 (1993; Zbl 0801.41001)] has proved that the Lebesgue constants \(\mu_n(X)\) for Lagrange-type interpolation operators are unbounded, in fact he has shown that \(\mu_n(X)\geq c\log n\), no matter what the matrix \(X\) is. Hence, in particular, one cannot expect uniform convergence for all continuous functions. The author proves that if \(\mu_n(X)\sim n^\delta\), \(0<\delta<1\) and if \(\delta<\gamma\leq 1\), then \(\lim_{n\to\infty}\|H_{nm}(X,f,\cdot)-f\|=0\), for all \(f\in\text{Lip}\gamma\). He also shows that if \(0<\gamma<m\delta /(\delta+2m)\), then there exists an \(f\in\text{Lip}\gamma\) for which the above is not valid. In between, convergence may hold for some matrices of nodes and not for others. See the paper for details.