Distilling universal machine-learning potentials for moiré lattices across one million atoms

Abstract

Atomic reconstruction reshapes moiré materials across multiple scales, from local structure and polarization textures to global electronic topology, yet direct \textit{ab initio} modeling becomes prohibitive for large superstructures such as marginal-twist-angle moirés and moiré-of-moirés. We develop MoiréMLIP by fine-tuning a universal atomistic model on the density functional theory labeled Moiré Kaleidoscope dataset, which spans transition metal dichalcogenide compositions, symmetries, stackings, and twist angles. MoiréMLIP reproduces \textit{ab initio} reconstruction with force errors of 6--8\,meV/Å and transfers to smaller twist angles and unseen structures. Knowledge distillation yields MoiréMLIP-mini, which retains this accuracy while extending single-GPU inference to one million atoms. Applied to an alternate-twist MoTe₂ trilayer, it reveals a hierarchical polarization network spanning tens of nanometers arising from large moiré-cell distortions sharply localized along the moiré-of-moiré domain walls. These results overcome key accuracy, transferability, and scaling bottlenecks of existing atomistic models and enable predictive simulations across emergent moiré length scales.

Distilling Large-Scale Machine Learning Interatomic Potentials to Predict Long-Timescale Electrolyte Transport Properties
Ishan Amin, Yue Jian, Nitesh Kumar +10
arXiv29 Oct 2026PreprintCC BY 4.0Full textPhysical chemistry Computational physicsarXiv
Can Universal Machine-Learning Interatomic Potentials Capture Small Polarons? A Geometry-Based Benchmark in Transition-Metal Oxides
Vladislav Korostelev, Pjotrs Žguns, Konstantin Klyukin
arXiv12 Oct 2026PreprintarXiv licenceFull textPhysical chemistry Computational physicsarXiv
From sparse quantum-computing data to atomistic simulation with universal machine-learning interatomic potentials
Tuan Minh Do, Yuichiro Yoshida, Kenji Ishihara +1
arXiv18 Sep 2026PreprintarXiv licenceFull textPhysical chemistry Quantum physics & quantum informationarXiv
Pressure-transferable universal machine-learning potentials: the WBM-MatPES benchmark and Tuga-FF
Pierre-Paul De Breuck, Thalis H. B. da Silva, Antoine Loew +2
arXiv10 Oct 2026PreprintCC BY 4.0Full textCondensed matterarXiv
Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials
Hongwei Du, Dingyang Lv, Boyan Zhang +11
arXiv17 Oct 2026PreprintCC BY 4.0Full textCondensed matterarXiv

Cite

BibTeX RIS

BibTeX

@misc{huang2026distilling,
  title = {{Distilling universal machine-learning potentials for moiré lattices across one million atoms}},
  author = {Thomas Huang and Yueyao Fan and Kaichen Xie and Bolun Li and Kaijie Yang and Jenna A. Bilbrey and Eric Bylaska and Peter V. Sushko and Di Xiao and Ting Cao},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.04115},
  eprint = {2610.04115},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04115},
}

RIS

TY  - UNPB
TI  - Distilling universal machine-learning potentials for moiré lattices across one million atoms
AU  - Thomas Huang
AU  - Yueyao Fan
AU  - Kaichen Xie
AU  - Bolun Li
AU  - Kaijie Yang
AU  - Jenna A. Bilbrey
AU  - Eric Bylaska
AU  - Peter V. Sushko
AU  - Di Xiao
AU  - Ting Cao
PY  - 2026
DA  - 2026/10/02
PB  - arXiv
DO  - 10.48550/arxiv.2610.04115
UR  - https://arxiv.org/abs/2610.04115
AB  - Atomic reconstruction reshapes moiré materials across multiple scales, from local structure and polarization textures to global electronic topology, yet direct \textit{ab initio} modeling becomes prohibitive for large superstructures such as marginal-twist-angle moirés and moiré-of-moirés. We develop MoiréMLIP by fine-tuning a universal atomistic model on the density functional theory labeled Moiré Kaleidoscope dataset, which spans transition metal dichalcogenide compositions, symmetries, stackings, and twist angles. MoiréMLIP reproduces \textit{ab initio} reconstruction with force errors of 6--8\,meV/Å and transfers to smaller twist angles and unseen structures. Knowledge distillation yields MoiréMLIP-mini, which retains this accuracy while extending single-GPU inference to one million atoms. Applied to an alternate-twist MoTe_2 trilayer, it reveals a hierarchical polarization network spanning tens of nanometers arising from large moiré-cell distortions sharply localized along the moiré-of-moiré domain walls. These results overcome key accuracy, transferability, and scaling bottlenecks of existing atomistic models and enable predictive simulations across emergent moiré length scales.
ER  -

Abstract

Related

Cite