Dartmouth Events

Full events calendar

Physics & Astronomy Virtual Quantum/Nano Seminar - Jon Vandermause '16, Harvard

Title: "Accelerating quantum-accurate materials simulation with on-the-fly machine learning"

Thursday, May 20, 2021
4:00pm – 5:00pm
Zoom: Email for link and password
Intended Audience(s): Public
Categories: Lectures & Seminars
Abstract: Machine-learned force fields have emerged in the last decade as a powerful tool for performing large-scale molecular dynamics simulations at quantum mechanical (QM) accuracy, but training these models is often a laborious process requiring thousands of expensive QM calculations and detailed prior knowledge of the material of interest. In this talk, I will present a closed-loop Bayesian inference method for automating the training of many-body force fields using structures drawn “on the fly” from molecular dynamics simulations. Within an online active learning algorithm, the internal uncertainty of a Gaussian process (GP) regression model is used to decide whether to accept the model prediction or to perform a QM calculation to augment the training set of the model and reoptimize its hyperparameters. I will show how mean predictions of the GP can be mapped onto equivalent and much faster polynomial models suitable for large-scale molecular dynamics simulations approaching millions of atoms, and discuss applications to a range of materials, including fast-ion diffusion in silver iodide, martensitic phase transitions in nickel titanium, and surface restructuring in palladium-silver alloys.
For more information, contact:
Tressena Manning
tressena.a.manning@dartmouth.edu
603-646-2854
Permanent URL to this event: https://physics.dartmouth.edu/events/event?event=63527

Events are free and open to the public unless otherwise noted.