Analysis And Control Of Networked Systems Using Structural And Measure-Theoretic Approaches

Network control theory provides a plethora of tools to analyze the behavior of dynamical processes taking place in complex networked systems. The pattern of interconnections among components affects the global behavior of the overall system. However, the analysis of the global behavior of large scale complex networked systems offers several major challenges. First of all, analyzing or characterizing the features of large-scale networked systems generally requires full knowledge of the parameters describing the system's dynamics. However, in many applications, an exact quantitative description of the parameters of the system may not be available due to measurement errors and/or modeling uncertainties. Secondly, retrieving the whole structure of many real networks is very challenging due to both computation and security constraints. Therefore, an exact analysis of the global behavior of many real-world networks is practically unfeasible. Finally, the dynamics describing the interactions between components are often stochastic, which leads to difficulty in analyzing individual behaviors in the network. In this thesis, we provide solutions to tackle all the aforementioned challenges. In the first part of the thesis, we adopt graph-theoretic approaches to address the problem caused by inexact modeling and imprecise measurements. More specifically, we leverage the connection between algebra and graph theory to analyze various properties in linear structural systems. Using these results, we then design efficient graph-theoretic algorithms to tackle topology design problems in structural systems. In the second part of the thesis, we utilize measure-theoretic techniques to characterize global properties of a network using local structural information in the form of closed walks or subgraph counts. These methods are based on recent results in real algebraic geometry that relates semidefinite programming to the multidimensional moment problem. We leverage this connection to analyze stochastic networked spreading processes and characterize safety in nonlinear dynamical systems.