High precision machine learning force field development for BaTiO3 phase transitions, amorphous, and liquid structures

The phase transition behavior of BaTiO3 (BTO) plays a crucial role in determining its material properties. However, studying these transitions typically requires large supercells to minimize constraints on atomic movements, significantly increasing computational time and cost. In this study, we developed a machine learning potential by fine-tuning the MACE-MP-0 model with a customized BTO database. The fine-tuned MACE model demonstrated high accuracy in reproducing the structural and dynamic characteristics of BTO systems. Notably, the model effectively represented changes in lattice constants and polarization during phase transitions (rhombohedral → orthorhombic → tetragonal → cubic). Furthermore, we employed the fine-tuned MACE model to investigate the melting point of BTO and subsequently generated liquid-phase and amorphous structures via thermal annealing, providing valuable insights into the high-temperature behavior of BTO.