A dynamical systems approach to distributed eigenvector computation

Spyridon Leonardos, Victor M. Preciado, Kostas Daniilidis

Abstract: We study the problem of distributedly estimating the k largest/smallest eigenvalues and the associated eigenvectors of a (possibly weighted) graph. In this work, we propose a dynamical systems approach that is fully decentralized and has global convergence guarantees. We demonstrate the validity of our approach through rigorous theoretical analysis and experimental evaluation.