First Principles Datasets for Machine Learning Force Fields of High Entropy Alloys

Machine learning force fields (MLFFs) have gained popularity for simulating high-entropy alloys (HEAs). However, high-quality training data for HEAs remains limited. Here, we present first-principles data based on density functional theory (DFT) for CoCrFeMnNi. Training samples are generated through random equiatomic binary substitutions and atomic perturbations, resulting in 10,000 face-centered cubic (FCC) structures for all binary combinations of Co, Cr, Fe, Mn, and Ni. Additionally, random equiatomic quinary substitutions and perturbations are performed for 5,000 FCC CoCrFeMnNi supercells. The resulting force fields trained on the DFT data and deep neural networks are validated via molecular dynamics simulations, achieving overall good theory-experiment agreements. This study provides a foundational database for training MLFFs containing Co, Cr, Fe, Mn, and Ni, and the framework can be applied to other HEAs with different transition metals.