Abstract

Markov state models (MSMs) provide a useful framework for understanding the kinetics of highly complex systems. Accordingly, MSM -based methods are increasingly popular to analyse and drive molecular dynamics simulations, including biasing and unbiasing simulations to enhance sampling. However, the dimensionality of these models is still very large when we consider a space of selected discretised collective variables. We developed kinetics-based coarse graining methods to identify metastable and transition states using variational objective functions: the second eigenvalue corresponding to the slowest relaxation time, and the Kemeny constant, corresponding to the sum of all relaxation times of the system. We show that these objective functions are closely related to mean first passage times that may be easier available in various applications of complex networks beyond molecular simulations.