Speaker: Jonathan Smith (Temple University, USA)
Title: Theoretical Anharmonic Vibrational Spectroscopy on Machine Learning Optimized Potential Energy Surfaces
Abstract: Molecular vibrational spectra encode structure, conformation, energetics, and, in condensed phases, local solvent environment. Computations have become increasingly capable of accurately predicting a range of key properties and observables for isolated molecules as well as clusters and in some casesthe condensed phase. Challenging cases occur where species have substantial conformational flexibility due to a combination of energy content and/or molecular floppiness near or above barriers along conformational or reaction coordinates and in time domain spectroscopies. Modelling solvent also presents challenges. In many casessampling a wide range of an anharmonic potential energy surface is necessary to reproduce spectra and to better elucidate the underlying encoded chemistry. Computing spectra from ab initio molecular dynamics (MD) affords some advantages in the above cases due to sampling the potential energy surface without resorting to solely the potential’s gradient near minima in static calculations. MD also implicitly includes mechanical and electrical anharmonicity. Ab initio molecular dynamics on coupled cluster level surfaces or directly afford a best case in accurately capturing these effects but direct dynamics at this level of theory are presently only viable for small molecular systems. Here we present an approach which leverages machine learning to optimize accurate potential energy surfaces with up to the coupled cluster level accuracy and fast evaluation of energetics and gradients at approximately 105 faster per gradient. This permits sufficiently long and statistically averaged trajectories and even permits the treatment of nuclear dynamics through the path integral molecular dynamics approximation which can be particularly important for hydroxyl groups and solvation dynamics. Early results on acetylene, formic acid, and a formic acid-water complex highlight the promise of this approach.
The Hylleraas seminars alternate between Oslo (Room V205) and Tromsø (Room C329), and are broadcasted by video conference to the other place.