Fully Distributed Nash Equilibrium Seeking in N-Cluster Games

Distributed optimization and Nash equilibrium (NE) seeking problems have drawn much attention in the control community recently. This paper studies a class of non-cooperative games, known as $N$-cluster game, which subsumes both cooperative and non-cooperative nature among multiple agents in the two problems: solving distributed optimization problem within the cluster, while playing a non-cooperative game across the clusters. Moreover, we consider a partial-decision information game setup, i.e., the agents do not have direct access to other agents' decisions, and hence need to communicate with each other through a directed graph whose associated adjacency matrix is assumed to be non-doubly stochastic. To solve the $N$-cluster game problem, we propose a fully distributed NE seeking algorithm by a synthesis of leader-following consensus and gradient tracking, where the leader-following consensus protocol is adopted to estimate the other agents' decisions and the gradient tracking method is employed to trace some weighted average of the gradient. Furthermore, the algorithm is equipped with uncoordinated constant step-sizes, which allows the agents to choose their own preferred step-sizes, instead of a uniform coordinated step-size. We prove that all agents' decisions converge linearly to their corresponding NE so long as the largest step-size and the heterogeneity of the step-size are small. We verify the derived results through a numerical example in a Cournot competition game