In a recent preprint, entitled: "Development of a machine learning interatomic potential for exploring pressure-dependent kinetics of phase transitions in Germanium", we presented a novel Neural-Network (NN) interatomic potential for Ge. We recall that Ge phases different from the cubic-diamond one are of particular interest for applications. Hexagonal Ge, for instance, displays superior optical properties. It is therefore important to investigate how, exploiting pressure, Ge can be transformed into different allotropes. In order to build a potential tackling kinetics of pressure-induced phase transformations, several kinetic paths (mainly sampled using the solid-state Nudged Elastic Band method) were added to the database, following a suitable active-learning procedure. Energies, forces, and stressed relative to the various configurations were computed ab initio using VASP with the PBE functional. The NN potential was trained using the Deep Potential Molecular Dynamic package (DeePMDkit). The potential greatly reproduces the relative stability of several Ge phases and yields at least a semi-quantitative description of the energetics along complex phase-transformation paths.
In the present archive, we provide the full potential for use in LAMMPS and ASE, together with the full database produced using VASP.