Unconstrained learning of equivariant atomistic properties

Abstract

Machine-learned interatomic potentials based on graph neural networks (GNNs) have made first-principles accuracy routine for the potential energy surface (PES). Their backbones, often O(3)-equivariant by design, are increasingly used to learn other atomistic properties. Many of these are of high angular order or pseudotensorial character, defined per atom, or expanded on large atom-centered basis sets. Such targets strain equivariant GNNs, whose tensor products grow steeply in cost with angular order and can fail to represent the target if truncated, forcing difficult trade-offs between hyperparameters. Unconstrained models like the Point-Edge Transformer (PET) instead replace explicit O(3) algebra with edge-to-edge attention and learn equivariance from data, so that the same backbone can in principle learn any equivariant property. We demonstrate how to use PET to learn molecular dipoles and polarizabilities, NMR chemical shielding tensors, and auxiliary-basis expansions of the electron density. Increasing the backbone size systematically improves both accuracy and equivariance error, and targeted architectural and training choices reduce the symmetry breaking further. PET matches or improves on the released reference models at lower inference cost and, compared with equivariant models, a smaller memory footprint. The predictions give accurate IR and Raman spectra and isotropic shieldings; the predicted densities accelerate self-consistent field calculations and give one-shot ground-state properties, and pretrained density models transfer to higher levels of theory by fine-tuning on little data. The impact of approximate equivariance is small for most observables, further validating this approach to train fast, accurate, and useful surrogate models for properties beyond the PES.

NEMORA: Neural Equivariant Multipole Operators for Long-Range Atomistic Learning
Jay L. Kaplan, Samuel Varner, Rebecca Willett +1
arXiv7 Oct 2026PreprintCC BY 4.0Full textPhysical chemistry Computational physicsarXiv
Fast and Accurate Equivariant Foundation Models for Atomistic Simulation
Seán R. Kavanagh, Chuin Wei Tan, Menghang Wang +12
arXiv30 Jul 2026PreprintCC BY 4.0Full text12 citedPhysical chemistry Applied physics & instrumentationarXivOpenAlex
On the Theoretical Limits of Learning Unconstrained Interatomic Potentials
Prajwal Ananth, Shuwen Yue
arXiv10 Oct 2026PreprintarXiv licenceFull textPhysical chemistryarXiv
A Lightweight Universal Machine-Learning Interatomic Potential via Knowledge Distillation for Scalable Atomistic Simulations
Sangmin Oh, J.H. You, Jaesun Kim +4
Journal of Chemical Information and Modeling20 Jul 2026Peer-reviewedLicence not statedFull text9 citedMaterials & polymersOpenAlex
Overcoming the limitations of body-ordered potentials for atomistic machine learning
Filippo Bigi, Michelangelo Domina, Maximilian Lloyd Ach +5
arXiv7 Oct 2026PreprintarXiv licenceFull textPhysical chemistryarXiv

Cite

BibTeX RIS

BibTeX

@misc{abbott2026unconstrained,
  title = {{Unconstrained learning of equivariant atomistic properties}},
  author = {Joseph W. Abbott and Paolo Pegolo and Pol Febrer and Michelangelo Domina and Sandra Saade and Matthias Kellner and Filippo Bigi and Philip Loche and Michele Ceriotti},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.31183},
  eprint = {2610.31183},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.31183},
}

RIS

TY  - UNPB
TI  - Unconstrained learning of equivariant atomistic properties
AU  - Joseph W. Abbott
AU  - Paolo Pegolo
AU  - Pol Febrer
AU  - Michelangelo Domina
AU  - Sandra Saade
AU  - Matthias Kellner
AU  - Filippo Bigi
AU  - Philip Loche
AU  - Michele Ceriotti
PY  - 2026
DA  - 2026/10/22
PB  - arXiv
DO  - 10.48550/arxiv.2610.31183
UR  - https://arxiv.org/abs/2610.31183
AB  - Machine-learned interatomic potentials based on graph neural networks (GNNs) have made first-principles accuracy routine for the potential energy surface (PES). Their backbones, often O(3)-equivariant by design, are increasingly used to learn other atomistic properties. Many of these are of high angular order or pseudotensorial character, defined per atom, or expanded on large atom-centered basis sets. Such targets strain equivariant GNNs, whose tensor products grow steeply in cost with angular order and can fail to represent the target if truncated, forcing difficult trade-offs between hyperparameters. Unconstrained models like the Point-Edge Transformer (PET) instead replace explicit O(3) algebra with edge-to-edge attention and learn equivariance from data, so that the same backbone can in principle learn any equivariant property. We demonstrate how to use PET to learn molecular dipoles and polarizabilities, NMR chemical shielding tensors, and auxiliary-basis expansions of the electron density. Increasing the backbone size systematically improves both accuracy and equivariance error, and targeted architectural and training choices reduce the symmetry breaking further. PET matches or improves on the released reference models at lower inference cost and, compared with equivariant models, a smaller memory footprint. The predictions give accurate IR and Raman spectra and isotropic shieldings; the predicted densities accelerate self-consistent field calculations and give one-shot ground-state properties, and pretrained density models transfer to higher levels of theory by fine-tuning on little data. The impact of approximate equivariance is small for most observables, further validating this approach to train fast, accurate, and useful surrogate models for properties beyond the PES.
ER  -

Abstract

Related

Cite