Transferable Water Potentials Using Equivariant Neural Networks

TL;DR

This paper demonstrates that equivariant neural network-based machine learning interatomic potentials can accurately and transferably model water across different phases, including liquid, vapor, and ice, with high fidelity and stability.

Contribution

It introduces a transferable MLIP for water trained on bulk liquid data that accurately predicts properties across multiple phases and conditions.

Findings

Reproduces liquid water density within 0.003 g/cm3 from 230 to 365 K

Accurately models vapor-liquid equilibrium up to 550 K

Captures many-body interactions and vibrational states of ice phases

Abstract

Machine learning interatomic potentials (MLIPs) are an emerging modeling technique that promises to provide electronic structure theory accuracy for a fraction of its cost, however, the transferability of MLIPs is a largely unknown factor. Recently, it has been proposed (J. Chem. Phys., 2023, 158, 084111) that MLIPs trained on solely liquid water data cannot describe vapor-liquid equilibrium while recovering the many-body decomposition analysis of gas-phase water clusters, as MLIPs do not directly learn the physically correct interactions of water molecules, limiting transferability. In this work, we show that MLIPs based on an equivariant neural network architecture trained on only 3,200 bulk liquid water structures reproduces liquid-phase water properties (e.g., density within 0.003 g/cm3 between 230 and 365 K), vapor-liquid equilibrium properties up to 550 K, the many-body decomposition analysis of gas-phase water cluster up to six-body interactions, and the relative energy and the vibrational density of states of ice phases. This study highlights that state-of-the-art MLIPs have the potential to develop transferable models for arbitrary phases of water that remain stable in nanosecond-long simulations.