Machine learning interatomic potential to study radiation-induced damage in 3C-SiC

This dataset contains atomic structures used to train a machine learning interatomic potential (MLIP) with the Gaussian Approximation Potential (GAP) framework. Designed to study radiation damage at the atomistic level, it includes the following configurations: dimers; elastically distorted bulk 3C-SiC, bulk Si, and bulk C; thermalized supercells at different temperatures and lattice constants; vacancies, di-vacancies, and tri-vacancies; antisites; tetrahedral, hexagonal, and split interstitials; liquid; mid-quench; and amorphous phases. The structures are stored in extended XYZ format. Each configuration is tagged with the total energy, atomic forces, and virial stresses calculated with DFT at the PBE level using the VASP code. Each structure is a member of a configurational category identified by the "config_type" keyword. Additional information about each structure is stored under the "sub_config" keyword. Details regarding the dataset's creation and DFT calculations are presented in the paper's supplementary material.