Stability of Evolving Agent Populations

Stability is perhaps the most desired feature in the systems that we design. It is important for us to be able to predict the response of a Multi-Agent System (MAS) to various environmental conditions prior to its actual deployment. The Chli-DeWilde agent stability measure views a MAS as a discrete time Markov chain with a potentially unknown transition probabilities. A MAS is considered to be stable when its state, a stochastic process, has converged to an equilibrium distribution. We investigate an extension of their agent stability definition to include MASs with evolutionary dynamics, focusing on evolving agent populations. Additionally, using our extended agent stability measure, we construct an entropy-based definition for the degree of instability. An example system, the Digital Ecosystem, is considered in detail to investigate the stability of an evolving agent population through simulations. The results are consistent with the original Chli-DeWilde measure.