Deep-learning interatomic potentials of the ɛ-ZrX_{2} series (X=H, D, and T).

Abstract

The ɛ-ZrH_{2} species act as an important component in zirconium-based composite hydrides, which have various applications in nuclear energy, hydrogen storage, and catalysis. In this work, deep-learning interatomic potentials for ɛ-ZrH_{2} have been developed by training density functional theory (DFT) data. The results indicate that the developed deep-learning interatomic potentials (DP) can accurately predict the structural, mechanical, and thermodynamic properties of ɛ-ZrH_{2} with DFT level accuracy. These deep-learning interatomic potentials are shown to be superior to the conventional modified embedded atom method potential. The H-isotope effect was also taken into account in constructing the deep-learning interatomic potentials, which facilitates molecular dynamic (MD) simulations under irradiation conditions. The development of these deep-learning interatomic potentials offers improved options for MD simulations of ɛ-ZrX_{2} (X=H, D, and T).