Foresight-dependent entropy production in stochastic two-agent dynamics

Abstract

Nonreciprocal interactions can break detailed balance by sustaining stationary probability currents. Here we ask whether such currents can also emerge when the agents' immediate coordination preferences are reciprocal, but their state-adjustment rules are determined by forward-looking optimization (foresight). We study a minimal stochastic linear-quadratic game in which two agents control noisy scalar states and minimize individual discounted costs in a full-information feedback Nash equilibrium. The optimized controls yield an Ornstein-Uhlenbeck process whose stationary probability current and entropy-production rate, measuring statistical time-irreversibility, can be calculated exactly. For reciprocal coordination with unequal self-anchoring strengths, the effective drift asymmetry first appears at order $τ^5$, where $τ$ is the planning timescale, and the normalized entropy production consequently scales as $τ^8$ in the limit $τ\to0$. We derive this cancellation from the coupled Nash Riccati equations. For conformist-contrarian coupling, normalized entropy production approaches a nonzero short-foresight limit and initially decreases from that limit by an amount proportional to $τ^2$. We also determine the long-foresight limit for a conformist-contrarian pair with equal self-anchoring strengths. Numerical solutions of the full Nash equations verify the asymptotic laws and reveal the effect of foresight. Thus, forward-looking optimization can generate or reduce normalized stationary irreversibility, depending on the underlying interaction.