Repeatability in simulation of large-scale agent-based social behavior

Abstract

In large-scale agent-based social simulations situations are tested and compared based on measurements. These measurements measure the behavior of the agents in the simulation. Due to the concurrency within the simulation model of large-scale agent-based social simulations the ordering of some actions cannot be ensured which can lead to noise within the simulation model. Noise within the simulation model means that the simulation is less accurate due to the fact that it no longer models the real world system it tries to represent. Given the same input one would expect a similar output, but this is not always the case in large-scale agent-based social simulations. Ensuring this means ensuring repeatability. We propose synchronization techniques in multi-agent systems for ensuring repeatability in large-scale agent-based social simulations and show the correctness of these techniques in our attempt to ensure repeatability in large-scale agent-based social simulations.