MagPES-MACE: DFT datasets and model checkpoints for FeN4/O2, Mn–Au–O2, and Au/CeO2

Source data for the MagPES-MACE study, covering FeN4/O2, Mn–Au–O2 and Au/CeO2.The archive contains DFT-labelled training structures, VASP inputs and AIMDtrajectories, trained MACE checkpoints, and example run scripts for the threesystems. Contents FeN4/O2 — datasets/AIMD.extxyz (9996 frames: two 4998-frame MD segments withdifferent magnetic inputs), datasets/OPT.extxyz (1105 PES pretraining OPTframes), VASP INCAR/KPOINTS/POSCAR/XDATCAR for the AFM and FM AIMD runs (5000configurations each), AFM/FM starting templates, atomic reference energies,three checkpoints (md_mhc.pt, md_mtc.pt, pes.pt), and MD/PES example scripts. Mn–Au–O2 — datasets/AIMD.extxyz (5000 frames: 2500 AFM and 2500 FM),datasets/OPT.extxyz (778 constrained-scan frames: 400 AFM and 378 FM), VASPAIMD inputs and trajectories for both spin states, one checkpoint (model.pt),and train/validation/test split counts. Au/CeO2 — datasets/AIMD.extxyz (10953 frames with mixed oxygen-defectcompositions), a 2O reference AIMD trajectory with its VASP inputs, onecheckpoint (model.pt), and split counts. Formats and usage Structures are provided as extended XYZ (extxyz) files; DFT trajectories areVASP XDATCAR files with the corresponding INCAR, KPOINTS and POSCAR; models arePyTorch .pt checkpoints retaining their training metadata; auxiliary files areJSON and SLURM scripts. In the training files, state_id 0/1 denotes the AFM/FM(S0/S2) branch, and scan distances are encoded in the frameid field. README.mdgives the stage frame ranges, branch frame counts and the file layout. The archive contains 40 files (about 461 MiB uncompressed), distributed as asingle tar.gz file.