Floquet Driving of Enzymatic Reactions: Counting Statistics and Long-Time Currents

What the math gives to ML

The paper provides a constructive Floquet formalism for periodically driven continuous-time Markov processes: long-time behavior is controlled by the one-period propagator, its effective Floquet generator, and derivatives of the propagator with respect to a counting field. This transfers naturally to periodically switched neural-network optimization, where different optimizers, learning rates, losses, or parameter subsets are applied in alternating phases. The key asset is a stability criterion based on the spectral radius of the product of phase-wise Jacobians, rather than on the stability of either phase in isolation. High-frequency driving additionally yields an effective averaged update with commutator corrections, predicting when the order of optimizer phases matters.