Distributed Averaging Dynamics and Optimization over Random Networks
Open Access Publications from the University of California

## Distributed Averaging Dynamics and Optimization over Random Networks

Abstract

In this thesis, we study Distributed Averaging Dynamics and its main application, i.e. Distributed Optimization. More specifically, the results of this thesis can be divided into two main parts: 1) Ergodicity of distributed averaging dynamics, and 2) Distributed optimization over dependent random networks.

First, we study both discrete-time and continuous-time time-varying distributed averaging dynamics. We show a necessary and a sufficient condition for ergodicity of such dynamics. We extend a well-known result in ergodicity of time-homogeneous (time-invariant) averaging dynamics and we show that ergodicity of a dynamics necessitates that its (directed) infinite flow graph has a spanning rooted tree. Then, we show that if groups of agents are connected using a rooted tree and the averaging dynamics restricted to each group is $\Pst$ and ergodic, then the dynamics over the whole networks is ergodic. In particular, this provides a general condition for convergence of consensus dynamics where \textit{groups} of agents capable of reaching consensus follow each other on a time-varying network.

Then, we study the averaging-based distributed optimization solvers over random networks for both convex and strongly convex functions. We show a general result on the convergence of such schemes for a broad class of dependent weight-matrix sequences. In addition to implying many of the previously known results on this domain, our work shows the robustness of distributed optimization results to link-failure. Also, it provides a new tool for synthesizing distributed optimization algorithms. To prove our main theorems, we establish new results on the rate of convergence analysis of averaging dynamics and non-averaging dynamics over (dependent) random networks. These secondary results, along with the required martingale-type results to establish them, might be of interest to broader research endeavors in distributed computation over random networks.