A Kokkos-Accelerated Moment Tensor Potential Implementation for LAMMPS

This repository provides benchmark data for: A Kokkos-Accelerated Moment Tensor Potential Implementation for LAMMPS. The benchmarking was conducted on a block of unstrained BCC potassium under NVE conditions with a 1-femtosecond timestep for 100 timesteps. The training data used is relatively simplistic and shared across all potentials, which may limit their accuracy outside of this benchmark. Default hyperparameters and cutoff distances were applied, resulting in a comparatively low number of neighbors per neighborhood. The results presented here are therefore intended as general guidance or scaling indicators and are expected to vary significantly with changes in the number of neighbors per neighborhood. List of directories and files in dataset (zenodo.zip): Potentials: Contains potentials used for inference (levels 6–28). Active Learning Potentials: Contains active learning potentials (levels 6–28) for one to three species, including configuration and neighborhood mode. Inference: Contains inference speed data for 1, 16, 32, and 64 AMD 7532 cores, as well as 1, 2, 3, and 4 NVIDIA A100 GPUs (40 GB). This directory also includes plotting scripts, some of which were partially developed using generative AI. Active Learning: Contains active learning speed data for 1, 16, 32, and 64 AMD 7532 cores, and 1, 2, and 4 NVIDIA A100 GPUs (40 GB). Active Learning Plotting: Contains plotting scripts for active learning speed data, some of which were partially developed using generative AI. inference.lmp: Script used for benchmarking and example invocations. active_learning.lmp: Script used for benchmarking and example invocations. Example scripts for the 3 illustrative examples are also available (example_scripts.zip).