Efficient Parametrization of Transferable Atomic Cluster Expansion for Water

Area of Science:

• Computational Chemistry
• Materials Science
• Statistical Mechanics

Background:

• Accurate interatomic potentials are crucial for simulating water properties.
• Traditional methods for developing these potentials are computationally expensive.

Purpose of the Study:

• To develop a highly accurate and transferable parametrization of water using the atomic cluster expansion (ACE).
• To introduce an efficient sampling technique for machine learning potentials in water simulations.

Main Methods:

• Utilized the atomic cluster expansion (ACE) for water parametrization.
• Employed active learning (AL) with D-optimality to select relevant liquid water configurations.
• Sampled static calculations of ice phases and few liquid configurations for training.

Main Results:

• The ACE potential, trained initially on ice and refined with AL-selected liquid configurations, accurately describes liquid water.
• The developed potential shows remarkable agreement with first-principles calculations.
• Accurately captured structural, dynamic, and thermodynamic properties of liquid water and the melting point of ice Ih.

Conclusions:

• Presents a novel and efficient sampling technique for machine learning potentials.
• Offers a transferable interatomic potential for water, demonstrating the accuracy of first-principles reference.
• Enhances understanding of ice-liquid water relationships and enables new studies of aqueous systems.