Asynchronous Modeling and Simulation with Orthogonal Agents

Abstract

This paper considers a class of systems of autonomous self-governed agents with purpose-specific behavior. Agents of this class contribute most to the overall performance if they have an unobstructed transparent access to the environment. Many examples of such systems can be found in swarm technologies and asynchronous simulation of discrete and continuous systems. An efficiency metric for a multi-agent system operating within a given environment is proposed as a dot product of the system's characteristic time-vectors: one of an agent's demands for resources and the other of the resources' availability. It is shown that the smaller the dot product the higher the efficiency of the agents. In some cases, the better efficiency of individual agents translates into improvement of the overall performance of the system. This observation is postulated as the principle of orthogonality: under some conditions, the asynchronous, ungoverned systems outperform the systems with synchronized actions. It is shown that the asynchronous "chaotic" multi-agent models, properly devised to achieve a higher level of transparency, can produce a better throughput beyond the level achieved by simply improving the latency of the system. Examples of orthogonal systems include many discrete-continuous physical, financial, control and some machine learning multi-agent models. Conditions of convergence of asynchronous models are presented. Some experimental results are shown, as well. More general observations are made in the context of natural decomposition.