Amorphous Zirconia-doped Tantala modeling and simulations using explicit multi-element spectral neighbor analysis machine learning potentials (EME-SNAP)

We model amorphous Zirconia-doped Tantala with machine learning interactomc potentials based on explicit multielement spectral neighbor analysis (EME-SNAP). These atomic structure models can reproduce partial radial distribution functions obtained from first-principles calculations and elastic moduli found from experimental measurements. The two-body pair forces calculated from EME-SNAP further affirm that the potentials capture the atomic interactions well. Molecular dynamics simulations of simulated annealing with EME-SNAP show that the final density of the amorphous models depends on the thermal history even when the annealing rate is kept constant, which captures experimental observations of history-dependent densities. Mechanical spectroscopy is also simulated using both Morse-Beest-Kramer-Santen pair potentials and EME-SNAP. The success in applying the EME-SNAP to amorphous Zirconia-doped Tantala pushes the boundaries of simulation accuracy and system size and enables better and more realistic atomistic modeling for amorphous systems. There are still some limitations in applying the potentials generated in this paper. They are only optimized for trained amorphous phases; high-temperature stability and transferability need to be further investigated.