shiruosong/MagPT

MagPT is an open dataset and base model for non-collinear magnetic materials, covering 48 metal elements. The full dataset contains 912,492 entries, generated by active learning from spin-lattice coupled MD trajectories (0–1000 K, 0–10 GPa) with DFT calculations using non-collinear magnetism, SOC, and DeltaSpin constraints. This release provides a subset of 399,168 entries from the full dataset.

• Format: extxyz
• Read with ASE:

from ase.io import read

for atoms in read("./examples/example.xyz", index=":"):
    print(atoms.get_chemical_formula(), atoms.get_potential_energy())

• Units: position in $\mathrm{\AA}$, spin in $\mu_{\mathrm{B}}$, energy in eV, force in eV/$\mathrm{\AA}$, magnetic force in eV/$\mu_{\mathrm{B}}$, virial in eV

• Citation:

@misc{MagPT2026,
      title={Atomistic prediction of magnetic behaviour across chemistry, structure and temperature}, 
      author={Ruosong Shi and Chengbing Chen and Bicheng Liu and Daye Zheng and Duo Zhang and Zhengtao Huang and Teng Yang and Zefeng Cai and Xingliang Peng and Xinzijian Liu and Shenshen Yan and Xueqing Zhao and Jianbo Wang and Bin Liu and Han Wang and Ke Wang and Liuliu Han and Ben Xu},
      year={2026},
      eprint={2610.24407},
      archivePrefix={arXiv},
      primaryClass={cond-mat.mtrl-sci},
      url={https://arxiv.org/abs/2610.24407}, 
}