The transport of interstitial protons, oxygen vacancies and strontium vacancies in the cubic perovskite SrTiO \(_3\) are investigated with molecular statics (MS) and molecular dynamics (MD) simulations employing the pre-trained machine-learning (ML) potential SevenNet. In the first two cases ( \(\mathrm{H_i}\) and \(\mathrm{v_O}\) ), defect diffusion coefficients are obtained, and these are higher than experimental values reported in the literature. For all three cases ( \(\mathrm{H_i}\) , \(\mathrm{v_O}\) and \(\mathrm{v_{Sr}}\) ), the activation enthalpies of defect migration are underestimated relative to experimental data, thus yielding a lower threshold for activation enthalpies. Without the need for subsequent modification, the SevenNet potential provides a lower limit for activation barriers of ion migration and an upper limit to diffusion coefficients. Our results thus indicate that, without fine-tuning, the SevenNet potential is unlikely to accurately reproduce experimental data, but they suggest that it can be used to compare proton, anion and cation transport rates across various perovskite oxides prior to fine-tuning and eventual experimental study.
