Convergence Analysis of Dual Decomposition Algorithm in Distributed Optimization: Asynchrony and Inexactness

Abstract

Dual decomposition is widely utilized in the distributed optimization of multiagent systems. In practice, the dual decomposition algorithm is desired to admit an asynchronous implementation due to imperfect communication, such as time delay and packet drop. In addition, computational errors also exist when the individual agents solve their own subproblems. In this article, we analyze the convergence of the dual decomposition algorithm in the distributed optimization when both the communication asynchrony and the subproblem solution inexactness exist. We find that the interaction between asynchrony and inexactness slows down the convergence rate from O(1/k) to O(1/ √ k). Specifically, with a constant step size, the value of the objective function converges to a neighborhood of the optimal value, and the solution converges to a neighborhood of the optimal solution. Moreover, the violation of the constraints diminishes in O(1/ √ k). Our result generalizes and unifies the existing ones that only consider either asynchrony or inexactness. Finally, numerical simulations validate the theoretical results.

On the Primal Feasibility in Dual Decomposition Methods Under Additive and Bounded Errors

“…The existing literature that designs adaptive stepsizes primarily considers subgradient methods within the primal domain. These results can be readily applied to distributed algorithms that are based on duality [36]- [39]. More specifically, in the case of dual decomposition algorithms for distributed optimization, the master problem that coordinates the subproblems at subsystems is often solved by using subgradient methods [40] where the earlier results for stepsizes can be adopted.…”