Research on Machine Learning Interatomic Potentials for Titanium Oxide Ceramic Materials

By adjusting the dataset size and hyperparameter settings, we have developed a Gaussian Approximation Potential (GAP) for titanium oxides based on machine learning methods. Through large-scale molecular dynamics simulations, we have calculated properties of rutile-type titanium oxide, such as lattice constants, thermal expansion, and thermal conductivity, which show excellent consistency with first-principles calculation results. In our work, we have discussed in detail the impact of hyperparameter selection in GAP on the accuracy of the potential function, laying the foundation for subsequent related research. The developed potential file is TiO-2.xml, and its calling format in LAMMPS is: pair_style quip pair_coeff * * TiO-2.xml "Potential xml_label=GAP_2023_3_3_480_2_14_8_334" 22 8. The potential file has been uploaded to this dataset.