Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials

Abstract

Universal machine-learning interatomic potentials are increasingly used across diverse materials, yet their reliability under directional deformation remains unclear. We compare compression and extension along paired deformation paths from the same parent structures for eleven pretrained potentials. The benchmark comprises 1,963 Materials Project trajectory (MPtrj) parents and 1,286 experimental-source MinX-Eq parents at six strain magnitudes from 1% to 10%. Density-functional theory provides references for isotropic, uniaxial and biaxial loading in MPtrj and isotropic loading in MinX-Eq. At ± 10%, every model has a larger median force error under compression, with pooled isotropic compression-to-extension ratios of 2.12 and 3.30 for MPtrj and MinX-Eq. Decomposing error growth into DFT force-scale growth and relative-error evolution shows that relative errors decrease on both branches, faster under compression in MPtrj but slower under compression in MinX-Eq. Force-scale matching moves the error ratios toward unity, although MinX-Eq retains a compression-side excess across models. Radial and SOAP analyses place compressed structures at shorter separations and farther from public reference pools, while readout-associated model features show larger compression-side displacements. These results distinguish stronger physical responses from changes in relative predictive accuracy and provide a basis for assessing universal potentials in deformation simulations.

Benchmarking machine-learning interatomic potentials using classical force fields
Johannes Sandberg, Geneviève Dusson, Julien Lam
arXiv9 Oct 2026PreprintCC BY 4.0Full textCondensed matterarXiv
Why Multi-Layer Message Passing Works: Completeness Theory for Graph Neural Network Interatomic Potentials
Pingbing Ming, Han Wang
arXiv2 Sep 2026PreprintCC BY-NC-SA 4.0Full textPhysical chemistry Mathematical physicsarXiv
MLIP Detective: Active Failure Mode Discovery Beyond Benchmark Scores for Machine-Learning Interatomic Potentials
Ryuhei Okuno, Nontawat Charoenphakdee, Kaoru Hisama +1
arXiv8 Sep 2026PreprintCC BY 4.0Full textPhysical chemistry Condensed matterarXiv
Cluster-based Structural Similarity for Dataset Visualization and Data Selection for Machine Learning Interatomic Potentials
Yuto Iwasaki, Miguel A. Caro
arXiv30 Sep 2026PreprintarXiv licenceFull textCondensed matterarXiv
BranchIP: Learning Adaptive Equivariant Computation for Interatomic Potentials
Laura Zichi, Gil Harari, Chuin Wei Tan +6
arXiv1 Oct 2026PreprintarXiv licenceFull textPhysical chemistry Applied physics & instrumentationarXiv

Cite

BibTeX RIS

BibTeX

@misc{du2026understanding,
  title = {{Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials}},
  author = {Hongwei Du and Dingyang Lv and Boyan Zhang and Yu Ren and Feng Yu and Xin He and Bonan Zhu and Yongda Huang and Jiahui Liu and Yongheng Li and Siqi Shi and Jianjun Liu and Ziheng Lu and Hong Wang},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.24514},
  eprint = {2610.24514},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.24514},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials
AU  - Hongwei Du
AU  - Dingyang Lv
AU  - Boyan Zhang
AU  - Yu Ren
AU  - Feng Yu
AU  - Xin He
AU  - Bonan Zhu
AU  - Yongda Huang
AU  - Jiahui Liu
AU  - Yongheng Li
AU  - Siqi Shi
AU  - Jianjun Liu
AU  - Ziheng Lu
AU  - Hong Wang
PY  - 2026
DA  - 2026/10/17
PB  - arXiv
DO  - 10.48550/arxiv.2610.24514
UR  - https://arxiv.org/abs/2610.24514
AB  - Universal machine-learning interatomic potentials are increasingly used across diverse materials, yet their reliability under directional deformation remains unclear. We compare compression and extension along paired deformation paths from the same parent structures for eleven pretrained potentials. The benchmark comprises 1,963 Materials Project trajectory (MPtrj) parents and 1,286 experimental-source MinX-Eq parents at six strain magnitudes from 1% to 10%. Density-functional theory provides references for isotropic, uniaxial and biaxial loading in MPtrj and isotropic loading in MinX-Eq. At ± 10%, every model has a larger median force error under compression, with pooled isotropic compression-to-extension ratios of 2.12 and 3.30 for MPtrj and MinX-Eq. Decomposing error growth into DFT force-scale growth and relative-error evolution shows that relative errors decrease on both branches, faster under compression in MPtrj but slower under compression in MinX-Eq. Force-scale matching moves the error ratios toward unity, although MinX-Eq retains a compression-side excess across models. Radial and SOAP analyses place compressed structures at shorter separations and farther from public reference pools, while readout-associated model features show larger compression-side displacements. These results distinguish stronger physical responses from changes in relative predictive accuracy and provide a basis for assessing universal potentials in deformation simulations.
ER  -

Abstract

Related

Cite