Distributed Optimization with Coupling Constraints in Multi-Cluster Networks Based on Dual Proximal Gradient Method

Abstract

In this work, we consider solving a distributed optimization problem in a multi-agent network with multiple clusters. In each cluster, the involved agents cooperatively optimize a separable composite function with a common decision variable. Meanwhile, a global cost function of the whole network is considered associated with an affine coupling constraint across the clusters. To solve this problem, we propose a cluster-based dual proximal gradient algorithm by resorting to the dual problem, where the global cost function is optimized when the agents in each cluster achieve an agreement on the optimal strategy and the global coupling constraint is satisfied. In addition, the proposed algorithm allows the agents to only communicate with their immediate neighbors. The computational complexity of the proposed algorithm with simple-structured cost functions is discussed and an ergodic convergence with rate O(1/T) is guaranteed (T is the index of iterations).