Particle swarm optimization: Understanding order-2 stability guarantees

Abstract

This paper’s primary aim is to provide clarity on which guarantees about particle stability can actually be made. The particle swarm optimization algorithm has undergone a considerable amount of theoretical analysis. However, with this abundance of theory has come some terminological inconstancies, and as a result it is easy for a practitioner to be misguided by overloaded terminology. Specifically, the criteria for both order-1 and order-2 stability are well studied, but the exact definition of order-2 stability is not consistent amongst researchers. A consequence of this inconsistency in terminology is that the existing theory may in fact misguide practitioners instead of assisting them. In this paper it is theoretically and empirically demonstrated which practical guarantees can in fact be made about particle stability. Specifically, it is shown that the definition of order-2 stability which accurately reflects PSO behavior is that of convergence in second order moment to a constant, and not to zero.