jhaens/solidacid-0_model_dataset

SolidAcid-0 Dataset

DFT reference data for solid acid proton conductors from ab initio MD with CP2K (NVT, 0.5 fs, Nose-Hoover).

• Preprint: Enabling Domain-Specific Atomistic Models: A Machine Learning Potential for the Solid Acid Family. Hänseroth, von Stackelberg, Dreßler (2026) - arXiv:2609.38531
• High-throughput solid-acid study: High-throughput screening and mechanistic insights into solid acid proton conductors. Hänseroth, Großmann, Grunert, Runge, Dreßler in Commun. Mater. 7, 222 (2026) - doi:10.1038/s43246-026-01338-z.
• Models trained on this data: jhaens/solidacid-0_model

Contents

Path	Level of theory	Content
train_small.zip, val_small.zip	PBE-D3	training/validation frames (smaller subset)
train_medium.zip, val_medium.zip	PBE-D3	training/validation frames (2x larger subset)
test_set.zip	PBE-D3	held-out test frames
hse06_transfer_learning/<system>/{train,val,test}.xyz	HSE06-ADMM-D3	per-compound sets for transfer learning
aimd_pbe.inp, hse06_transfer_learning/aimd_hse06.inp	-	CP2K input files used to generate the data

All frames are extended XYZ with per-frame REF_TotEnergy (eV) and per-atom REF_Force (eV/A); <system> is one of csh2po4, csh2aso4, cshso4, cshseo4.

Please cite the preprint and publication above.

@unpublished{solidacid0,
      title={Enabling Domain-Specific Atomistic Models: A Machine Learning Potential for the Solid Acid Family}, 
      author={Jonas H{\"a}nseroth and Rose Asuka Baroness von Stackelberg and Christian Dre{\ss}ler},
      year={2026},
      eprint={2609.38531},
      archivePrefix={arXiv},
      primaryClass={cond-mat.mtrl-sci},
      url={https://arxiv.org/abs/2609.38531}, 
}

@article{haenseroth2026htscreening,
      title = {High-throughput screening and mechanistic insights into solid acid proton conductors},
      volume = {7},
      doi = {10.1038/s43246-026-01338-z},
      number = {222},
      journal = {Commun. Mater.},
      publisher = {Springer Science and Business Media LLC},
      author = {Jonas H{\"a}nseroth and Max Gro{\ss}mann and Malte Grunert and Erich Runge and Christian Dre{\ss}ler},
      year = {2026},
}

Released under the MIT license.