Stochastic approximation for consensus with general time-varying weight matrices

This paper considers consensus problems with delayed noisy measurements, and stochastic approximation is used to achieve mean square consensus. For stochastic approximation based consensus algorithms with switching topologies, the existing convergence analysis heavily relies on quadratic Lyapunov functions, whose existence may be difficult to guarantee for switching digraphs. The main contribution of this paper is to introduce a new approach for proving convergence. This is achieved by obtaining ergodicity results for backward products of degenerating stochastic matrices via a discrete time dynamical system approach. Our approach does not require the double stochasticity condition typically assumed for the existence of a quadratic Lyapunov function.