(607d) Accelerating Atomistic Simulation with Generative AI

The advent of generative artificial intelligence has provided new tools to model complex correlations in high-dimensional distributions. The recently proposed score dynamics (SD) is a general framework for learning accelerated evolution operators with large timesteps from molecular dynamics (MD) simulations.

SD enables us to take timesteps that are orders of magnitude larger than a typical MD time step. (1) In the position-only version of SD, we constructed graph neural network-based SD models of realistic molecular systems that are evolved with 10 ps timesteps. Both equilibrium predictions derived from the stationary distributions of the conditional probability and kinetic predictions for the transition rates and transition paths are in good agreement with MD tests on small molecules. (2) In the position-velocity version of SD, we have developed universally applicable acceleration techniques that allow one to take 10-100 fs timesteps in MD with little computational overhead, greatly extending the typical reach of molecular simulations.