The sampling problem in molecular simulations and the role of machine learning

Abstract
Molecular simulations play an increasingly important role in scientific research, providing valuable insights to accelerate the discovery of new drugs, materials and catalysts. One of the main challenges of molecular simulations is to sample exhaustively the equilibrium distribution of the given system. Most phenomena of interest, such as ligand binding, phase transitions or chemical reactions, are inherently multimodal and exhibit multiple long-lived metastable states, making the equilibrium distribution difficult to sample with molecular dynamics or Markov chain Monte Carlo methods. Since the early days of molecular simulations, several methods have been proposed to mitigate the sampling problem, but they have often been tailored to specific systems. In recent years, machine learning approaches are taking over many aspects of molecular simulations, bringing new life to old ideas and enabling more general and accurate simulation methods. I will briefly present the main ways in which machine learning is being combined with molecular simulations, focusing on my own contributions.

Speaker 
Michele Invernizzi is a PostDoc researcher at Freie Universität Berlin. He received his PhD in Physics from ETH in 2020 with a thesis on rare events sampling methods for molecular simulations, developed under the supervision of Prof. M. Parrinello. 

Best wishes,