Concentration of the Stationary Distribution on General Random Directed Graphs

Abstract

We consider a random model for directed graphs whereby an arc is placed from one vertex to another with a prescribed probability which may vary from arc to arc. Using perturbation bounds as well as Chernoff inequalities, we show that the stationary distribution of a Markov process on a random graph is concentrated near that of the "expected" process under mild conditions. These conditions involve the ratio between the minimum and maximum in- and out-degrees, the ratio of the minimum and maximum entry in the stationary distribution, and the smallest singu- lar value of the transition matrix. Lastly, we give examples of applications of our results to well-known models such as PageRank and G(n, p).