Distribution of the eigenvalues of random block-matrices. (Q1418980): Difference between revisions

The asymptotic behaviour of the eigenvalues of an \(n \times n\) random symmetric block-matrix \({W}\) is studied as \(n \rightarrow \infty\). In this study, the symmetric matrix \({W}\) is decomposed as the sum of a deterministic matrix \({B}\) and a stochastic matrix \({P}\). \({B}\) stands for a block-diagonal matrix with diagonal blocks \({B}_i\), \(i=1, \ldots, k\). The \(n_i \times n_i\) symmetric matrix \({B}_i\) has the positive real numbers \(\mu_i\)'s as entries, except in its main diagonal where it has entries \(\nu_i\)'s. Here, \(k\) is a fixed positive integer and \(n = \sum_{i=1}^k n_i\). The entries in and above the main diagonal of the \(n \times n\) random symmetric matrix \({P}\) are independent uniformly bounded random variables (common bound is \(K\)) with zero expectation and with the same variance \(\sigma^2\). Rank estimates for the eigenvalues of \({W}\) are given when \(K\), \(k\), \(\sigma^2\), \(\mu_i\) and \(\nu_i\) are kept fixed while \(n_1, \ldots, n_k\) tend to infinity in the same order. It follows from this rough characterization that there is a spectral gap between the \(k\) largest and the remaining eigenvalues of \({W}\). The distance between the corresponding eigenspaces is investigated. On the basis of these results a finer estimation of protruding eigenvalues is given. It is proved that they asymptotically follow a \(k\)-variate normal distribution. The matrix data can also be viewed as a random graph with application at communication and other networks.