V. A. Avetisov 1, A. V. Chertovich 2, S. K. Nechaev 3, 4, O. A. Vasilyev 5, 6
Journal of Statistical Mechanics (2009) P07008
In this paper the question about statistical properties of block–hierarchical random matrices is raised for the first time in connection with structural characteristics of random hierarchical networks obtained by mipmapping procedure. 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 $q_{\gamma}$ depends on 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 $\nu\ge 1$ the power--law behavior is terminated. We also find that the vertex degree distribution of such hierarchical networks has a poly--scale fractal behavior extended to a very broad range of scales.
- 1. The Semenov Institute of Chemical Physics,
Russian Academy of Sciences - 2. Physics Department,
Moscow State University - 3. Laboratoire de Physique Théorique et Modèles Statistiques (LPTMS),
CNRS : UMR8626 – Université Paris XI – Paris Sud - 4. P. N. Lebedev Physical Institute,
Russian Academy of Science - 5. Max-Planck-Institute für Metallforschung,
Max-Planck-Institut - 6. Institut für Theoretische and Angewandte Physik,
Universität Stuttgart