MP-ALOE: An r2SCAN dataset for universal machine learning interatomic potentials

MP-ALOE is an r2SCAN dataset for Universal Machine Learning Interatomic Potentials (UMLIPs).The `MP_ALOE_DATA.jsonl.gz` file is the main distribution, containing pymatgen structure objects, associated properties (e.g. energies, forces, stresses, magmoms), and metadata about the structures' origin and a few sanity checks.The `MPALOE.model`, `MatPES.model`, and `MPALOE_MatPES_combined.model` files are all MACE potentials trained on the respective dataset(s).The `off_equilibrium_forces_data.json` contains the information necessary to recreate the off-equilibrium forces benchmark from the MP-ALOE paper.The files labeled `iter_[1,2,3].zip` contain the parsed Emmet TaskDocs, which contain information such as the input parameters used, energy at each electronic step, and a bunch of other more in-depth information about each calculation. It is expected that these files will not be needed for most users.