THIS FOLDERS CONTAINS DATA RELATED TO THE WORK "Simulating diffusion properties of solid-state electrolytes via a neural network potential: performance and training scheme" by A. Marcolongo, T. Binninger, F.Zipoli and T.Laino, https://arxiv.org/abs/1910.10090https://arxiv.org/abs/1910.10090 [1] , which tests the DeepMD code : https://github.com/deepmodeling/deepmd-kit [2] A description of folders follows: ------------ STARTING_CONFIGURATIONS ---------------------- We report the initial configurations of the training workflows (see Appendix A of [1] ) in a "pfe" format which is an extension of an .xyz format containing also energy and stress and which should be self-explaining. FPMD and PFF refer to training worflows started either from short first principles molecular dynamics trajectories or from long polarizable force field trajectories. To generate the "raw" files needed to launch a training of the DeepMD code [2] go to the directory of the chosen material, where the corresponding pfe file is, and use the perl script, like in the following : ../../../SCRIPTS/deepmd_from_pfe_to_deepmd_from_mapping.pl pfe_orig_1.dat If the virial is used in the training one may need also: ../../../SCRIPTS/scale_virial.py mv virial_resc.raw virial.raw This operation is needed to be consistent with units in [2], which scales the virial with the volume. ----------- CONVERGED_MODELS ------------------------ For each material we report a set of configurations used for generating final models, after the corresponding training workflow has converged. Only models resurting the training workflows started from PFF trajectories are reported, which are found to converge faster. The same steps described for extracting data from the "STARTING_CONFIGURATIONS" folder apply in this case to generate the raw files and can be used to regenerate the model using the DeepMD code. ------------- DATA_PLOTS ---------------------- The diffusion data needed to reproduce the figures in [1]. -------------- SCRIPTS ------------------------ Scripts used for extracting the data, as discussed in the descriptions of the relevant folders. --------------- EXAMPLES INPUT ----------------- Examples of inputs for a DeepMD training and a Lammps NVT calculation, reported just for reference.