Machine-learning interatomic potential for BaTiO₃: phase transitions, domain walls, and grain boundaries.

Area of Science:

• Materials Science
• Computational Physics
• Machine Learning

Background:

• Barium Titanate (BaTiO3) is a crucial ferroelectric material with complex phase transitions.
• Accurate atomistic simulations are needed to understand its behavior under various conditions.
• Existing models may lack the accuracy or transferability for diverse applications.

Purpose of the Study:

• To develop a machine learning interatomic potential for BaTiO3.
• To enable accurate atomistic simulations of phase transitions, defects, and domain walls.
• To validate the potential against experimental and density-functional theory data.

Main Methods:

• Utilized the atomic cluster expansion formalism for potential development.
• Trained the potential on a large dataset of density-functional theory calculations.
• Performed atomistic simulations to study phase transitions, pressure effects, and domain structures.

Main Results:

• The potential accurately reproduces temperature-driven phase transitions (rhombohedral, orthorhombic, tetragonal, cubic).
• It correctly describes the influence of pressure on transition temperatures, matching experimental observations.
• Accurate prediction of 180° domain-wall structures and grain boundary energetics in the rhombohedral phase.

Conclusions:

• The developed machine learning potential provides a reliable tool for simulating BaTiO3.
• It enables in-depth studies of ferroelectric phase transitions and defect properties.
• This work advances computational materials science for ferroelectric materials.