A few remarks on the operator norm of random Toeplitz matrices (Q966501)

The asymptotic behavior of the spectral norm of a random Toeplitz matrix \(T_n\) with independent real random variables \(X_0, X_1, X_2, \dots,\) is analyzed. The result of \textit{M. Meckes} [Electron. Commun. Probab. 12, 315--325, electronic only (2007; Zbl 1130.15018)] concerning the Toeplitz matrices generated by i.i.d. mean zero random variables with positive mean and with finite variance are extended to the case of i.i.d. mean zero random variables with finite variance and to the case of independent nonidentically distributed variables which satisfy some moment bounds. It is proved that if \({\mathbb E}X_0 = 0\), and \(0 < {\mathbb E}{X_0}^2 < \infty \) for i.i.d. random variables \(X_0, X_1, X_2, \dots,\) then \(\lim_{n \to \infty} \|T_n\|/( {\mathbb E}\|T_n\|) = 1\) with probability one (almost sure behavior) and also that \(\lim\sup_{n \to \infty} \|T_n\|/ \sqrt{n \log n}< \infty\) with probability one if and only if \({\mathbb E}X_0 = 0\) and \({\mathbb E}{X_0}^2 < \infty \). Sufficient conditions are given for the case of independent but not necessarily identically distributed variables for the above relation. Tools of proof come from the classical probability in Banach space theory and are based on concentration inequalities for sums of independent Banach space-valued variables according to the modification of recent results by \textit{U. Einmahl} and \textit{D. Li} [Trans. Am. Math. Soc. 360, 6677--6693 (2008; Zbl 1181.60010)].