Incorporating Polarization and Charge Transfer into a Point-Charge Model for Water Using Machine Learning

Area of Science:

• Computational Chemistry
• Materials Science
• Statistical Mechanics

Background:

• Rigid nonpolarizable water models with fixed charges are efficient but struggle with dipole-related properties.
• Accurate dipole moment surfaces are essential for predicting dielectric and spectroscopic behavior of water.

Purpose of the Study:

• To develop a machine-learning model for assigning accurate point charges to water molecules.
• To improve the prediction of dipole-related properties in molecular dynamics simulations of water.

Main Methods:

• Trained a machine-learning model using electronic structure data to determine atom-centered point charges for water.
• Performed molecular dynamics simulations using the new charge model.
• Analyzed the contributions of intermolecular charge transfer and intramolecular polarization to key properties.

Main Results:

• The machine-learning model significantly improved predictions of the dielectric constant and low-frequency infrared spectrum of liquid water.
• Identified intermolecular charge transfer as the primary source of spectral intensity at 200 cm^-1.
• Determined that intramolecular polarization drives the enhancement of the dielectric constant.

Conclusions:

• Machine-learned point charges offer a superior description of the water dipole moment surface compared to fixed charges.
• This approach enhances the accuracy of molecular simulations for crucial water properties.
• The study provides insights into the molecular origins of dielectric and spectroscopic behavior in water.