Maximum entropy for dynamic processes on networks (2504.04240v1)

Abstract: Dynamic processes on networks are fundamental to understand modern day phenomena such as information diffusion and opinion polarization on the internet or epidemic spreading. However, they are notoriously difficult to study broadly as small changes in initial conditions, the process or the network can lead to very different evolution trajectories. Here we apply the information-theoretic framework of maximum caliber to study the statistics of such systems analytically in a general way. We verify the dynamics deduced from maximum caliber by using simulations of different processes on different networks, introduce an approximation of the dynamics which significantly simplifies the problem and show that the approximation can be used to recover well-established models of population dynamics that are typically not thought of as taking place on a network. This provides a theoretical tool that allows to study dynamic processes on networks broadly without loosing sight of known results.