Likelihood-based non-Markovian models from molecular dynamics

TL;DR

This paper presents a data-driven method to derive non-Markovian models for molecular dynamics, accurately capturing memory effects and enabling efficient long-time simulations of collective variables.

Contribution

It introduces a likelihood-based approach to estimate generalized Langevin equations from trajectory data, improving modeling of complex molecular processes.

Findings

Successfully recovers memory kernels and noise from data

Enables accurate long-time sampling of collective variables

Applicable to non-equilibrium and metastable state analysis

Abstract

We introduce a new method to accurately and efficiently estimate the effective dynamics of collective variables in molecular simulations. Such reduced dynamics play an essential role in the study of a broad class of processes, ranging from chemical reactions in solution to conformational changes in biomolecules or phase transitions in condensed matter systems. The standard Markovian approximation often breaks down due to the lack of a proper separation of time scales and memory effects must be taken into account. Using a parametrization based on hidden auxiliary variables, we obtain a generalized Langevin equation by maximizing the statistical likelihood of the observed trajectories. Both the memory kernel and random noise are correctly recovered by this procedure. This data-driven approach provides a reduced dynamical model for multidimensional collective variables, enabling the accurate sampling of their long-time dynamical properties at a computational cost drastically reduced with respect to all-atom numerical simulations. The present strategy, based on the reproduction of the dynamics of trajectories rather than the memory kernel or the velocity-autocorrelation function, conveniently provides other observables beyond these two, including e.g. stationary currents in non-equilibrium situations, or the distribution of first passage times between metastable states.