Dataset: Machine-Learning Interatomic Potentials for Million-Atom Simulations of Multicomponent Alloys

File Description This dataset contains multiple types of files related to interatomic potential models and DFT datasets. The files can be grouped into three main categories: 1. DFT Dataset Files (*.pkl.gz) Files such as: 02.pkl.gz, 04.pkl.gz, 06.pkl.gz, …, 16.pkl.gz These are compressed pickle files containing DFT (Density Functional Theory) datasets. The number in the filename indicates the number of elements in the alloy system: 02.pkl.gz → binary alloy (2 elements) 04.pkl.gz → quaternary alloy (4 elements) 06.pkl.gz → 6-element alloy and so on These datasets are typically used for training and validating interatomic potential models. 2. GRACE-FS Potential Files (*.yaml) (a) Standard GRACE-FS Models Files such as: GRACE-FS-L.yaml GRACE-FS-M.yaml GRACE-FS-S.yaml GRACE-FS-M-Aug.yaml These are GRACE-FS interatomic potential model files. The suffix indicates the model size (number of parameters): L → Large M → Medium S → Small Aug indicates that the model was trained with an augmented dataset (additional training data). (b) Ensemble Models for UQ Files such as: FS-1.yaml, FS-2.yaml, …, FS-10.yaml These are multiple GRACE-FS potential models used for ensemble-based uncertainty quantification (UQ). Each file represents an independently trained model. The ensemble allows estimation of prediction uncertainty by comparing outputs across models. Summary *.pkl.gz → DFT datasets (number = alloy component count) GRACE-FS-*.yaml → GRACE-FS models (L/M/S = model size, Aug = augmented dataset) FS-*.yaml → GRACE-FS ensemble models for UQ Note: NEP Potential Files and corresponding datasets are provided in https://zenodo.org/records/11533864