On metastability and Markov state models for non-stationary molecular dynamics

Unlike for systems in equilibrium, a straightforward definition of a metastable set in the non-stationary, non-equilibrium case may only be given case-by-case—and therefore it is not directly useful any more, in particular in cases where the slowest relaxation time scales are comparable to the time scales at which the external field driving the system varies. We generalize the concept of metastability by relying on the theory of coherent sets. A pair of sets A and B is called coherent with respect to the time interval [t1, t2] if (a) most of the trajectories starting in A at t1 end up in B at t2 and (b) most of the trajectories arriving in B at t2 actually started from A at t1. Based on this definition, we can show how to compute coherent sets and then derive finite-time non-stationary Markov state models. We illustrate this concept and its main differences to equilibrium Markov state modeling on simple, one-dimensional examples.