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

Abstract

Rigid nonpolarizable water models with fixed point charges have been widely employed in molecular dynamics simulations due to their efficiency and reasonable accuracy for the potential energy surface. However, the dipole moment surface of water is not necessarily well-described by the same fixed charges, leading to failure in reproducing dipole-related properties. Here, we developed a machine-learning model trained against electronic structure data to assign point charges for water, and the resulting dipole moment surface significantly improved the predictions of the dielectric constant and the low-frequency IR spectrum of liquid water. Our analysis reveals that within our atom-centered point-charge description of the dipole moment surface, the intermolecular charge transfer is the major source of the peak intensity at 200 cm −1 , whereas the intramolecular polarization controls the enhancement of the dielectric constant. The effects of exact Hartree−Fock exchange in the hybrid density functional on these properties are also discussed.

Accurate Free Energies of Aqueous Electrolyte Solutions from Molecular Simulations with Non-polarizable Force Fields

“…20− 30% compared to experiments. 45,46 Recently, Han et al 30 have successfully simulated the dielectric constant of water using non-polarizable force fields.…”

“…For this purpose, many classical force fields for water have been developed. − TIP4P/2005 is a computationally efficient and popular water force field, which accurately predicts many properties of water, such as the shear viscosity, diffusivity, density, temperature of maximum density, and surface tension, despite being rigid and non-polarizable. ,,, Clearly, the effective interactions of TIP4P/2005 (dictated by the relative energy differences in the PES) in the liquid phase are well-described. , Despite this, the TIP4P/2005 force field does not yield accurate predictions of the VLE of water, because predictions for vaporization enthalpies and saturated vapor pressures are poor. ,, Describing the VLE of water requires accurate modeling of (1) effective interactions between water molecules and (2) the excess chemical potential (with respect to the ideal gas reference state) of the liquid phase (μ w ex ) (dictated by the absolute value of the PES), because the coexistent pressures have an exponential dependency upon μ w ex . TIP4P/2005 consistently underestimates μ w ex compared to experiments (e.g., by ca.…”

“…Recently, Han et al have successfully simulated the dielectric constant of water using non-polarizable force fields. This study shows that the charges used in TIP4P/2005 water should only be used to model the PES, from which effective interactions between molecules are computed, and a different set of charges (derived from quantum mechanical simulations) should be used to model the dipole moment of the aqueous system, from which the dielectric constant is computed. , Similarly, Blazquez et al reproduced the experimental electrical conductivities of aqueous NaCl and KCl solutions up to the solubility limit (1) using non-polarizable scaled charge force fields to describe the PES of ions and (2) using integer charges to compute the dipole moment of the aqueous solution from which electrical conductivities are calculated.…”

“… 23 − 29 TIP4P/2005 23 is a computationally efficient and popular water force field, which accurately predicts many properties of water, such as the shear viscosity, diffusivity, density, temperature of maximum density, and surface tension, despite being rigid and non-polarizable. 15 , 21 , 23 , 30 Clearly, the effective interactions of TIP4P/2005 (dictated by the relative energy differences in the PES 18 ) in the liquid phase are well-described. 18 , 30 Despite this, the TIP4P/2005 force field does not yield accurate predictions of the VLE of water, because predictions for vaporization enthalpies and saturated vapor pressures are poor.…”

Fully First-Principles Surface Spectroscopy with Machine Learning

“…Several other ML approaches with varying architectures have been reported in the last five years to predict molecular dipole moments − and linear vibrational spectroscopy. − However, attempts to simulate VSFG spectroscopy have been lagging behind, most likely due to the destructive interference of the signal, which made those approaches inappropriate for this target. While the strategy presented here could be easily applied to other kernel regression models that use local representations, it remains to be seen if similar ideas could be implemented on neural network architectures.…”