Distribution of eigenvalues for the ensemble of real symmetric Toeplitz matrices (Q2576791): Difference between revisions

The authors consider random matrix ensembles with less degrees of freedom than independently identically distributed symmetric matrices. In the latter case, a well-known result from \textit{E. P. Wigner} [Ann. Math. (2) 62, 548-564 (1955; Zbl 0067.08403)] implies that the limiting distribution of the density of eigenvalues is the semi-circle law. In the paper under review, random real symmetric Toeplitz matrices are considered, that is matrices \((b_{j-i})_{i,j}\) where the entries are independent identically distributed random variables with a common probability distribution \(p\) with mean 0, variance 1, and finite moments of all orders. For such a matrix \(A\), the following normalized moments are considered: \[ M_k(A,N)=\frac1{N^{(k/2)+1}}\,\sum_{i=1}^N\lambda_i(A). \] The limit when \(N\rightarrow\infty\) of the average of \(M_k(A,N)\) over the ensemble, with each \(A\) weighted by its distribution, is shown to converge to the moments of a distribution with unbounded support and such that the \(0,2\) and \(k\)-th moments agree with the Gaussian moments, while the even moments for \(k\geq4\) are less than the Gaussian moments. The convergence is shown to be almost sure when \(p\) is an even function. The bulk of the paper consists in the proof of the aforementioned result. In calculating the moments, combinatorial methods are used. Starting with the 4th moment and going up with the even moments, the combinatorics used show that there are diophantine obstructions that prevent the moments to be Gaussian. At the end of the paper, cases with less degrees of freedom are discussed, together with some numerical evidence regarding the spacings between adjacent normalized eigenvalues.