Reinforced dynamics of large atomic and molecular systems

  • 2017-12-12 15:18:41
  • Linfeng Zhang, Han Wang, Weinan E
  • 0

Abstract

A new approach for efficiently exploring the configuration space andcomputing the free energy of large atomic and molecular systems is proposed,motivated by an analogy with reinforcement learning. There are two majorcomponents in this new approach. Like metadynamics, it allows for an efficientexploration of the configuration space by adding an adaptively computed biasingpotential to the original dynamics. Like deep reinforcement learning, thisbiasing potential is trained on the fly using deep neural networks, with datacollected judiciously from the exploration. Applications to the full-atom,explicit solvent models of alanine dipeptide and tripeptide show some ofpromise for this new approach.

 

Quick Read (beta)

loading the full paper ...