Spectral analysis for stochastic models of large-scale complex dynamical networks

Abstract

Research on large-scale complex networks has important applications in diverse systems of current interest, including the Internet, the World-Wide Web, social, biological, and chemical networks. The growing availability of massive databases, computing facilities, and reliable data analysis tools has provided a powerful framework to explore structural properties of such real-world networks. However, one cannot efficiently retrieve and store the exact or full topology for many large-scale networks. As an alternative, several stochastic network models have been proposed that attempt to capture essential characteristics of such complex topologies. Network researchers then use these stochastic models to generate topologies similar to the complex network of interest and use these topologies to test, for example, the behavior of dynamical processes in the network. In general, the topological properties of a network are not directly evident in the behavior of dynamical processes running on it. On the other hand, the eigenvalue spectra of certain matricial representations of the network topology do relate quite directly to the behavior of many dynamical processes of interest, such as random walks, Markov processes, virus/rumor spreading, or synchronization of oscillators in a network. This thesis studies spectral properties of popular stochastic network models proposed in recent years. In particular, we develop several methods to determine or estimate the spectral moments of these models. We also present a variety of techniques to extract relevant spectral information from a finite sequence of spectral moments. A range of numerical examples throughout the thesis confirms the efficacy of our approach. Our ultimate objective is to use such results to understand and predict the behavior of dynamical processes taking place in large-scale networks.