Necessary conditions for mean convergence of Lagrange interpolation on an arbitrary system of nodes (Q2563798): Difference between revisions

Given a triangular matrix \(X\) of nodes \(x_{1n}>x_{2n}>\cdots>x_n\) \((n=1,2,\dots)\) let the Lagrange interpolant to \(f\in C[-1,1]\) be denoted by \(L_n(X,f)= \sum^n_{k=1} f(x_{kn})\ell_{kn}(x)\), \(n=1,2,\dots\), where \(\ell_{kn}(x)\) are the fundamental polynomials, with \(\omega_n(x):=\omega_n(X,x):=\prod^n_{k=1}(x-x_{kn})\). If \(X\) denotes the zeros of the orthogonal polynomial \(P_n(d\alpha, x)\) \((=\gamma_nx^n+\cdots,\gamma_n>0)\) with respect to a nondecreasing function \(\alpha(x)\) on \((-\infty,\infty)\) with infinitely many points of increase, we write \(X(d\alpha)\) and \(x_{kn}(d\alpha)\). Also \(d\alpha\) is called a regular measure if \(\lim_{n\to\infty} (\gamma_n(d\alpha))^{1/n}=2\). In 1980, \textit{P. Turán} [J. Approximation Theory 29, 23-89 (1980; Zbl 0454.41001)] proposed the following problem (Problem XI) among others: Given \(p>1\), what is the necessary and sufficient condition that \[ \lim_{n\to\alpha} \int^1_{-1}(f(x)-L_n(X,f,x))^pdx=0 \] for every \(f\in C[-1,1]\)? Here, the author proposes a slightly more general problem: Let \(d\alpha\) be a measure supported in \([-1,1]\) and \(X\subset[-1,1]\). Given \(p>0\) what is a necessary and sufficient condition that for every \(f\in C[-1,1]\), \[ \lim_{n\to\infty} \int^1_{-1}|f(x)-L_n(X,f,x)|^pd\alpha(x)=0?\tag{\(*\)} \] The following notation is used: Let \(I:=[-1,1]\). \[ |f|_{d\alpha,p}:=\Biggl\{\int_I|f(x)|^pd\alpha(x)\Biggr\}^{1/p},\quad |f|:=\max_{x\in I} |f(x)|, \] \[ |L_n(X)|_{d\alpha,p}:=\sup_{|f|\leq1} |L_n(x,f)|_{d\alpha,p},\quad |L_n(X)|:=\Biggl|\sum^n_{k=1}|\ell_{kn}(x)|\Biggr|, \] \[ S_n(X,x):=\sum^n_{k=1}|(x-x_{kn})\ell_{kn}(x)|,\quad \gamma_n(X):=\sum^n_{k=1} {1\over |\omega_n'(x_{kn})|}, \] \[ \phi(z):=z+\sqrt{z^2-1},\quad z\in\mathbb{C}, \] \[ N_n(X,a,b)=\text{cardinality of the set }\{k:x_{kn}\in[a,b]\subset I\}. \] Using the above notation, the author proves the following theorem as a partial answer to the above problem. Theorem. Let \(d\alpha\) be a regular measure supported in \([-1,1]\), \(X\subset[-1,1]\) and \(0<p<\infty\). If \((*)\) holds, then for every \(f\in C[-1,1]\), the following hold: (a) \(\lim_{n\to\infty} (\gamma_n(X))^{1/n}=2\), (b) \(\lim_{n\to\infty} |S_n(X)|^{1/n}=1\), (c) \(\lim_{n\to\infty} |L_n(X)|^{1/n}=1\), (d) \(\lim_{n\to\infty} |\omega_n(X,x)|^{1/n}={1\over 2} |\phi(z)|\), \(z\in\mathbb{C}\backslash I\), where the branch of the square root is taken so that \(\sqrt{z^2-1}\) behaves like \(z\) near infinity, (e) \(\lim_{n\to\infty} |\omega_n(X)|^{1/n}=1/2\), (f) \(\lim_{n\to\infty} {N_n(X,a,b)\over n}={1\over\pi}\int^b_a{dx\over\sqrt{1-x^2}}\). Under the assumptions of the theorem it appears that the asymptotic behaviour of \(\omega_n(X,x)\) is very closely related to the regular (\(n\)th root) asymptotic behaviour of \(\omega_n(d\alpha,x)\).