Molecular dynamics simulations in the age of machine learning

Eric Vanden-Eijnden is professor of mathematics at the Courant Institute, New York University. His research focuses on mathematical tools and numerical methods for stochastic, multiscale dynamical systems with applications in molecular dynamics, chemical and biological networks, materials science, atmosphere-ocean science, and fluid dynamics. He has developed techniques for sampling rare events, multiscale analysis, quantification of the effects of random perturbations of dynamical systems, and contributed to modeling turbulence via stochastic partial differential equations. Prof. Vanden-Eijnden’s objectives include understating pathways and rate of occurrence of rare events in complex systems and multiscale simulations of random dynamical systems.

The video below is a recording of Prof. Vanden-Eijnden’s plenary lecture, delivered on September 12^th 2019, at the conference “Molecular and materials simulation at the turn of the decade: Celebrating 50 years of CECAM”. Enjoy!

Molecular dynamics simulations in the age of machine learning

The rapid developments in machine learning (ML), leading to its success in the context of complex classification tasks, offer intriguing promises for molecular dynamics simulations. Indeed, ML has already been successfully used for force field parametrization, protein structure prediction, drug design, etc. These tasks rely on processing data obtained by independent means. Here I will discuss how ML could also help in producing long time series data, specifically focusing on free energy calculations and reactive event analysis, two problems that require designing accelerated sampling strategies and can possibly lead to learning low dimensional models.