A discrete-time averaging theorem and its application to zeroth-order Nash equilibrium seeking

Abstract

In this paper we present an averaging technique applicable to the design of zeroth-order Nash equilibrium seeking algorithms. First, we propose a multi-timescale discrete-time averaging theorem that requires only that the equilibrium is semi-globally practically stabilized by the averaged system, while also allowing the averaged system to depend on "fast" states. Furthermore, sequential application of the theorem is possible, which enables its use for multi-layer algorithm design. Second, we apply the aforementioned averaging theorem to prove semi-global practical convergence of the zeroth-order information variant of the discrete-time projected pseudogradient descent algorithm, in the context of strongly monotone, constrained Nash equilibrium problems. Third, we use the averaging theory to prove the semi-global practical convergence of the asynchronous pseudogradient descent algorithm to solve strongly monotone unconstrained Nash equilibrium problems. Lastly, we apply the proposed asynchronous algorithm to the connectivity control problem in multi-agent systems.

Aggregative feedback optimization for distributed cooperative robotics

“…Early references of this approach in the context of distributed optimization are [14], [15]. More recent works exploiting singular perturbations are [16]- [20] in the context of distributed optimization and [21]- [23] in the one of games. In order to deal with agreement on global information in our schemes, we resort to the formalism of ς-consensus [24], [25], modeling problems in which a set of agents, each endowed with a local quantity s t i and a common generic function ς, aims to reach consensus at ς(s t 1 , .…”