To perform predictive reactive simulations of water at the atomistic level it is necessary to accurately describe the quantum nature of both the electrons and the nuclei. Ab initio molecular dynamics (AIMD) simulations based on density functional theory that explicitly include nuclear quantum effects have been shown to be able to capture many features of the structure, dynamics and vibrational spectroscopy of liquid water. However, their high computational cost restricts the systematic investigation of properties that require long simulation times and large system sizes.<br/><br/>Here, I present our recent advances in developing a machine learning potential trained to ab initio calculations that captures the properties of liquid water across the entire temperature range. Unlike conventional interaction potentials, our machine learning potential is able to describe the breaking and forming of chemical bonds, which permits the study of proton transfer reactions in the condensed phase with significantly reduced computational effort compared to direct AIMD simulations. I will demonstrate how our potential can accurately describe the transport properties and proton transfer barriers of aqueous hydronium and hydroxide defects as well as the vibrational spectrum of liquid water across the whole temperature range. Finally, by combining our machine learning potential with path-integral simulations, I illustrate how we are able to identify the structural origin of Raman bands and provide a detailed understanding of the mechanism of defect transport in aqueous solutions.

Machine learning water: Insights into vibrational spectroscopy and aqueous proton defects

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.