Decentralized non-convex optimization via bi-level SQP and ADMM

Abstract

Non-convex optimization problems arise in many problems of practical relevance-for example in distributed nonlinear MPC or distributed optimal power flow. Only few existing decentralized optimization methods have local convergence guarantees for general nonconvex problems. We present novel convergence results for non-convex problems for a bi-level SQP method that solves the inner quadratic problems via ADMM. A decentralized stopping criterion borrowed from inexact Newton methods allows the early termination of ADMM as an inner algorithm to improve computational efficiency. The method shows competitive numerical performance to existing methods for an optimal power flow problem.