Where Foundation Machine-Learning Potentials Fail for Iron-Sulfide Defect Kinetics: Four Failure Modes and a DFT Benchmark of Vacancy-Anchored Hydrogen Migration in Pyrite, Marcasite, Mackinawite, and Greigite
Morozov, Igor N.;
Where Foundation Machine-Learning Potentials Fail for Iron-Sulfide Defect Kinetics: Four Failure Modes and a DFT Benchmark of Vacancy-Anchored Hydrogen Migration in Pyrite, Marcasite, Mackinawite, and Greigite
Archival release accompanying the manuscript on the foundation-MLIP (MACE-MP-0, CHGNet) vs DFT benchmark of vacancy-anchored proton migration in iron sulfides (pyrite, marcasite, mackinawite, greigite; pentlandite diagnostic). Includes QE DFT/NEB + dimer drivers (PBE U=0), partial-Hessian ZPE, zero-shot MLIP probes, figure generators, harvested result JSON, a complete DFT NEB structure dataset (extended-XYZ with energies+forces), and TM-Spec v0.3 records of every result (successes and failures). See README and CITATION.cff.