data-efficient and fast machine learning molecular dynamics through integrated active learning and knowledge distillation

This entry provides the training and test datasets, final datasets, trained models, and simulation results associated with four machine learning interatomic potentials (MLIPs) of liquid water: DeePMD and MACE models trained from scratch via active learning, a fine-tuned MACE foundation model, and a DeePMD potential obtained by knowledge distillation. All input scripts for the active learning and AIMD simulations have been included. Source data for reproducing the figures in the manuscript are provided. Detailed explanations for each subdirectory are provided inside the folder.