The Marcinkiewicz-type discretization theorems (Q1615927)

Let \(\Omega \) be a compact subset of \(\mathbb{R}^d,\, \mu\) a probability measure on \(\Omega\) and \(1\leq q<\infty.\) One says that an \(N\)-dimensional subspace \(X_N\) of \(L_q(\Omega,\mu)\) satisfies a Marcinkiewicz type theorem if there exist a subset \(\{\xi^\nu: 1\leq\nu\leq m\}\) of \(\Omega\) and two constants \(C_1(d,q),C_2(d,q)>0\) such that \[ C_1(d,q)\|f\|^q_q\leq \frac1m\sum_{\nu=1}^m|f(\xi^\nu)|^q\leq C_2(d,q)\|f\|^q_q\,,\leqno(\ast) \] for all \(f\in X_N.\) In the case of the uniform norm one asks that \[ C_1(d,q)\|f\|_\infty\leq\max_{1\leq \nu\leq m}|f(\xi^\nu)| \leq\|f\|_\infty\,, \] for all \(f\in X_N\subset C(\Omega).\) As the author mention, such a result was obtained for the first time by J.~Marcinkiewicz in the case of univariate trigonometric polynomials: \[ C_1(q)\|f\|^q_q\leq \frac1m\sum_{\nu=1}^{2n+1}|f\big(2\pi\nu/(2n+1)\big)|^q\leq C_2(q)\|f\|^q_q\,. \] The author considers also some variants of Marcinkiewicz problem: \(\bullet\)\; the Marcinkiewicz problem with weights (with \(\sum_{\nu=1}^m\lambda_\nu|f(\xi^\nu)|^q \) in \((*)\)); \newline \(\bullet\)\; the Marcinkiewicz problem with \(\varepsilon\) (when \(C_1(d,q)=1-\varepsilon\) and \(C_2(d,q)=1+\varepsilon\)). The aim of the paper is ``to present here a new technique, which works well for discretization of the integral norm. It is a combination of probabilistic technique, based on chaining, and results on the entropy numbers in the uniform norm.'' (quoted from abstract).