Coarse-graining dynamics to maximize irreversibility

In many far-from-equilibrium biological systems, energy injected by irreversible processes at microscopic scales propagates to larger scales to fulfill important biological functions. But given dissipative dynamics at the microscale, how much irreversibility can persist at the macroscale? Here, we propose a model-free coarse-graining procedure that merges microscopic states to minimize the amount of lost irreversibility. Beginning with dynamical measurements, this procedure produces coarse-grained dynamics that retain as much information as possible about the underlying irreversibility. In synthetic and experimental data spanning molecular motors, biochemical oscillators, and recordings of neural activity, we derive simplified descriptions that capture the essential nonequilibrium processes. These results provide the tools to study the fundamental limits on the emergence of macroscopic irreversibility.