An accurate and computationally efficient molecular level description of mesoscopic behavior of ice-water systems remains a grand challenge. There are a number of different water models that have been developed in an attempt to adequately capture the various thermodynamic anomalies exhibited by water in its various phases. While the solid and liquid phases have received much attention, studies on the vapor phase remains scarce. As an alternative,machine-learning techniques can be used to interpolate atomistic simulations or first principles calculations. Such machine-learning potentials (MLPs) enable linear-scaling atomistic simulations with an accuracy that is close to the reference method at a fraction of the computational cost. Here, we train an artificial neural network interfaced directly with Monte-Carlo based sampling to develop an ML model for water vapor. We evaluate its predictive power and make systematic comparisons with other existing water models including the best available polarizable and non-polarizable models. The limitations and strengths of our ML model in the context of water vapor simulations will be discussed.

Machine-learned model for molecular simulations of liquid and water vapor

2019 Spring National Meeting

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