Physical Chemistry Seminar: Overcoming Computational Bottlenecks in Quantum and Classical Molecular Dynamics Simulations

Zoom: 

Abstract:

Molecular dynamics (MD) simulations are a powerful tool, widely used in chemistry, biology, and physics, yet their application is often limited by severe computational bottlenecks. For example, simulations of quantum materials scale exponentially with the number of particles, and sampling conformational transitions of biomolecules is only possible on exceedingly short timescales. We develop algorithms to overcome these challenges, reducing the scaling of quantum simulations and sampling longer timescales in classical dynamics. I will focus on two recent efforts: 1) A method for bosonic path integral molecular dynamics (PIMD) simulations that reduces the scaling from cubic to quadratic, and its extension to periodic systems, allowing the first simulations of thousands of bosons using PIMD. 2) Sampling conformational transitions in biomolecules through stochastic resetting, leading to orders of magnitude in speedup and improved kinetic inference. We will also show how to use machine learning to automatically d

Seminar Organizer: Prof. Ilia Kaminker