BFaster linear iterations for distributed averaging

Abstract: Distributed averaging problems are a subclass of distributed consensus problems, which have received substantial attention from several research communities. Although many of the proposed algorithms are linear iterations, they vary both in structure and state dimension. In this paper, we investigate the performance benefits of adding extra states to distributed averaging iterations. We establish conditions for convergence and discuss possible ways of optimizing the convergence rates. By numerical examples, it is shown that the performance can be significantly increased by adding extra states. Finally, we provide necessary and sufficient conditions for convergence of a more general version of distributed averaging iterations. 1