Luigi Bonati
(Italian Institute of Technology)

Machine learning (ML) and enhanced sampling methods can be combined in different ways to successfully study rare events. On the one hand, advanced sampling methodologies allow us to construct machine learning potentials for reactive processes by collecting all relevant configurations for training datasets.¹ On the other, we can use ML techniques to learn collective variables in a data-driven manner, ² as well as to enhance the sampling of quantum-mechanical observables. These methodological advances enable ab initio-quality MD simulations of rare events for systems of realistic size and on long timescales. A revealing example from heterogeneous catalysis is the dynamics of iron surfaces at high temperatures and its influence on nitrogen decomposition.³

Zoom registration link:
https://zoom.us/meeting/register/tJEtd-uorTguGNEwLv26w1temkW743JcMnSk