Exactly solvable Gaussian and non-Gaussian mean-field games and collective swarms dynamics

Abstract

The collective behaviour of stochastic multi-agents swarms driven by Gaussian and non-Gaussian environments is analytically discussed in a mean-field approach. We first exogenously implement long range mutual interactions rules with strengths that are modulated by the real-time distance separating each agent with the swarm barycentre. Depending on the form of this barycentric modulation, a transition between drastically collective behaviours can be unveiled. A behavioural bifurcation threshold due to the tradeoff between the desynchronisation effects of the stochastic environment and the synchronising interactions is analytically calculated. For strong enough interactions, the emergence of a swarm soliton propagating wave is observable. Alternatively, weaker interactions cannot overcome the environmental noise and evanescent diffusive waves result. In a second and complementary approach, we show the the emergent solitons can alternatively be interpreted as being the optimal equilibrium of mean-field games (MFG) models with ad-hoc running cost functions which are here exactly determined. The MFG's equilibria resulting from the optimisation of individual utility functions are solitons that are therefore endogenously generated. Hence for the classes of models here proposed, an explicit correspondence between exogenous and endogenous interaction rules ultimately producing similar collective effects can be explicitly constructed. For both Gaussian and non-Gaussian environments our exact results unveil new classes of exactly solvable mean-field games dynamics.