Gradient-Free Distributed Optimization with Exact Convergence

TL;DR

This paper introduces a gradient-free distributed optimization algorithm that guarantees exact convergence over directed networks, applicable when gradient information is unavailable, with proven convergence rates and numerical validation.

Contribution

It presents a novel gradient-free distributed method with surplus-based updates that ensures exact convergence without requiring doubly stochastic matrices.

Findings

Guarantees exact convergence to the optimal value with non-summable step-sizes.

Achieves exact solution convergence when step-size is square-summable.

Validated effectiveness through numerical simulations.

Abstract

In this paper, a gradient-free distributed algorithm is introduced to solve a set constrained optimization problem under a directed communication network. Specifically, at each time-step, the agents locally compute a so-called pseudo-gradient to guide the updates of the decision variables, which can be applied in the fields where the gradient information is unknown, not available or non-existent. A surplus-based method is adopted to remove the doubly stochastic requirement on the weighting matrix, which enables the implementation of the algorithm in graphs having no associated doubly stochastic weighting matrix. For the convergence results, the proposed algorithm is able to obtain the exact convergence to the optimal value with any positive, non-summable and non-increasing step-sizes. Furthermore, when the step-size is also square-summable, the proposed algorithm is guaranteed to achieve the exact convergence to an optimal solution. In addition to the standard convergence analysis, the convergence rate of the proposed algorithm is also investigated. Finally, the effectiveness of the proposed algorithm is verified through numerical simulations.