Distributed strategies for average consensus in directed graphs

We address the average consensus problem for a distributed system whose components (nodes) can exchange information via interconnections (links) that form an arbitrary, strongly connected but possibly directed, topology (graph). Specifically, we discuss how the nodes can asymptotically reach average consensus (i.e., obtain the average of their initial values) with linear-iterative algorithms in which each node updates its value using a weighted linear combination of its own value and the values of neighboring nodes. In the process, the strategies we develop allow the nodes to adapt their weights in a distributed fashion, so that asymptotically they obtain a doubly stochastic weight matrix, which is useful for many algorithms that utilize linear- or nonlinear-iterative schemes to perform various estimation and optimization tasks.