Research on machine learning interatomic potentials for titanium oxide ceramic materials

JL Ren and GH Zhang and XM Wang and Y Han, PHYSICA B-CONDENSED MATTER, 711, 417281 (2025).

DOI: 10.1016/j.physb.2025.417281

A Gaussian approximation potential (GAP) for TiO2 was constructed by adjusting the size and composition of the dataset, as well as the hyperparameters in the smooth overlap of atomic positions (SOAP) descriptors. Using these trained machine learning interatomic potentials, we conducted molecular dynamics simulations to determine the lattice constants, thermal conductivity, and thermal expansion coefficient of titanium oxide ceramics. The calculated root mean square errors for these properties were 0.01 angstrom, 1.2 W K-1 m-1, and 0.005 K-1, respectively. Our molecular dynamics simulation results exhibit strong consistency with first-principles calculations. This indicates that the Gaussian approximation potential (GAP) developed for titanium oxide achieves both firstprinciples accuracy and computational efficiency.

Return to Publications page