Deep-learning potential molecular dynamics simulations of the structural and physical properties of rare-earth metal scandium

Abstract

The deep potential (DP) is a promising deep-learning-based approach to developing the high-accurate potential function of various materials from the data of ab-initio calculations based on density functional theory (DFT). To better understand the structural and physical properties of the rare-earth metal scandium (Sc), performing classical molecular dynamics (MD) simulations should be highly beneficial but has been straitened for lacking of available Sc potential. Therefore, the necessary interatomic potential function of Sc for MD simulations was developed first in this work by using the DP method. By systematically comparing the DP-predicted lattice constants, stable phase, vacancy and self-interstitial formation energies, surface energies, elastic constants and generalised stacking fault energy curves with the corresponding DFT results, we validated that the developed DP model of Sc enables these property-predictions with a reasonable DFT accuracy. Moreover, our DP-based MD simulations shown that the rare-earth Sc can transform from the α-HCP to the β-BCC structure at 1622 K and melt at 1710 K, quite close to the experimental values for the α-β phase transition temperature (1609 K) and the melting-point (1814 K) of Sc. Rising temperature can improve the diffusivity of Sc atoms and the self-diffusion coefficient at the melting-point is 5.7 × 10−12 m2/s, which is on the same order of magnitude as other HCP metals. The results could be used for understanding the fundamental properties of rare-earth metal Sc and as a basis for further developing the Sc-containing binary or multinary DP models.