Cellular automata for distributed computing: models of agent interaction and their implications

We propose cellular automata (CA) based models as a useful mathematical idealization for modeling large-scale distributed computing in general, and large-scale multi-agent systems (MAS), in particular. The classical CA need to be modified in several important respects, in order to become an appropriate abstraction for a broad class of large-scale MAS made of autonomous reactive agents. We argue that thus generalized CA can capture many important MAS properties at the level of agent ensembles and their long-term global behavior patterns. In this paper, we focus on the issue of inter-agent communication in CA. We propose sequential CA as the first step towards the genuinely asynchronous CA as the ultimate CA-based abstraction for MAS insofar as the model of inter-agent communication is concerned. We then compare and contrast certain configuration space properties of simple threshold sequential CA with the corresponding properties of parallel, perfectly synchronous simple threshold CA. Specifically, we show that the possibility of "looping" in parallel threshold CA is solely due to the unrealistic assumption of perfect inter-agent communication synchrony.