DFT datasets for training machine learning atomistic potentials, the final version of all DNPs, LAMMPS/VASP validation calculation scripts, and structure files.
We trained deep neural network potentials (DNPs) using the deepMD-kit for 23 elements across the periodic table.
Three randomly seeded DNPs were generated for each potential. Up to three iterations of adaptive learning refinement were applied to reduce standard deviations among these randomly seeded DNPs. The final iterations of these DNPs can be located in the "DNPs" directory
The "Training_Data" directory contained all of the training data used to fit the DNPs for each element. The VASP trajectories and the *.npy files are in these compressed files.
All of the structures used for validation of these DNPs can be found in the "Valdiation" directories for each element. The relaxed.in is the LAMMPS script for generating relaxed structures and calculating the energies of these structures as well and reporting the cell dimensions. The heat.in LAMMPS script was used to heat 10x10x10 supercell from 0 to the melting temperature for each element to demonstrate the DNP's stability for the solid phase. The INCAR file is the VASP input script we used to generate the corresponding VASP reference data and compare with the LAMMPs-DNP results. Lastly, the LAMMPS and VASP scripts for elastic calculations are supplied.
We utilized the LAVA code for calculating the DNP surface energies for these elements using the DNPs and LAMMPs due to the code's ease of use for this application (there are many other interesting tools in this package as well for LAMMPS and VASP validations of atomistic potentials).
The data used to produce figures are located in the "Figures" directories, along with the Python scripts to generate these plots.