This online talk is part of Xiamen Theoretical Chemistry Lectures (XTCL)

Title: What is the target of a machine-learned potential energy surface?  How the answer informs the approach taken.

Speaker: Prof. Joel M. Bowman

Emory University

Data & Time: 9:00 October 22nd (Fri.), 2021

Venue: B312 Zengchengkui Building

Koushare: https://www.koushare.com/lives/room/802307

Abstract:

There has been dramatic progress in developing so-called “machine-learned potential energy surfaces”.  After going the basics of what this means and also a quick survey of the ML methods currently employed my talk will emphasize that potential energy surfaces are of course a means to an end.  Namely, to high-quality computational chemistry ranging from reaction dynamics, spectroscopy, properties of clusters, hydrate clathrates and the condensed phase.  The “target” of the potential is the science of interest and more specifically whether the approach is quantum or classical or semi-quantum/semi-classical.  I will illustrate this with a number of specific case studies ranging from the tunneling splitting in malonaldehyde and the formic acid dimer to the conformations of glycine which include consideration of rigorous zero-point motion.