Free energy, friction, and mass profiles from short molecular dynamics trajectories

Abstract: We address the problem of constructing accurate mathematical models of the dynamics of molecular systems projected on a collective variable. To this aim we introduce an algorithm optimizing the parameters of a standard or generalized Langevin equation until the latter reproduces in a faithful way a set of molecular dynamics trajectories. In particular, using solvated proline dipeptide as a test case, we report evidence that ~100 short trajectories initiated at the top of a high barrier encode all the information needed to reconstruct free energy, friction, and mass profiles, including non-Markovian effects. The approach allows accessing the thermodynamics and kinetics of activated processes in a conceptually direct way, it employs only standard unbiased molecular dynamics trajectories, and is competitive in computational cost with respect to existing enhanced sampling methods. Furthermore, the systematic construction of Langevin models for different choices of collective variables starting from the same initial data could help in reaction coordinate optimization.