Title:On spectra of random hierarchical networks

Abstract: We propose a new model of hierarchical random graphs with power--law tail in the spectral density of a graph adjacency matrix. In our construction the adjacency matrices are randomized Parisi matrices. The peculiarity of our approach consists in the fact that we build clusters of edges with hierarchically organized probabilities, while the typical procedure consists in hierarchical grouping of vertices. Our procedure does not impose any additional metric structure on the graph and leaves the graph (network) purely topological. In particular, we compute numerically the spectral density of large random adjacency matrices defined by a hierarchy of the Bernoulli distributions $\{q_1,q_2,...\}$ on matrix elements, where each $q_{\gamma}$ depends on a hierarchy level $\gamma$ as $q_{\gamma}=p^{-\mu \gamma}$ ($\mu>0$). For the spectral density we clearly see the free--scale behavior. We show also that for the Gaussian distributions on matrix elements with zero mean and variances $\sigma_{\gamma}=p^{-\nu \gamma}$, the tail of the spectral density, $\rho_G(\lambda)$, behaves as $\rho_G(\lambda) \sim |\lambda|^{-(2-\nu)/(1-\nu)}$ for $|\lambda|\to\infty$ and $0<\nu<1$, while for $\nu\ge 1$ one sees termination of the power--law behavior.