Distribution of eigenvalues of real symmetric palindromic Toeplitz matrices and circulant matrices (Q2385612): Difference between revisions

Real symmetric palindromic Toeplitz matrices whose independent entries are i. i. d. random variables chosen from some distribution \(p\) with mean 0, variance 1 and finite higher moments are considered. Results of the analysis differ from the case when independent entries are taken from a Gaussian density and giving the limiting spectral measure i.e. the density of normalized eigenvalues which converges weakly and almost surely to a distribution which is almost the standard Gaussian. The different behaviour from that of Gaussian is caused by obstructions to solutions of Diophantine equations. It is shown that by imposing an additional symmetry on the matrices by requiring that the first row be a palindrome, the obstructions to the Diophantine equations vanish and the limiting spectral measure converges weakly, in probability and almost surely to the standard Gaussian. In a previous study, \textit{A. Bose} and \textit{J. Mitra} [Stat. Probab. Lett. 60, No.~1, 111--120 (2002; Zbl 1014.60038)] proved weak convergence for an ensemble closely related to palindromic Toeplitz matrices. They combined explicit expressions for the eigenvalues of circulant matrices and probabilistic arguments to construct the empirical spectral distribution and showed that the limiting spectral distribution is the standard Gaussian. In this paper, an alternate proof of Bose and Mitra's result is given. The linear algebra arguments are used for the analysis of the arising Diophantine equations. Then, the eigenvalues of palindromic Toeplitz are interlaced with those of the circulant ensemble. It is shown by Cauchy's interlacing property and the rank inequality that the analysis of the Diophantine equations related to the palindromic Toeplitz ensemble provides an alternate proof of the limiting spectral measure of the circulant ensemble.