Atoms that Matter: Largest Materials Datasets Are 99% Redundant, and Local Chemistry Can Guide the Required 1%
Abstract
General-purpose machine-learning force fields (MLFFs) for materials are now trained on the order of $100$ million structures, assuming transferability requires such scale. We test this premise on OMat24, the largest available materials dataset, by training MACE models on subsets of $6.5$ million atoms, less than 1% of OMat24, selected in three complementary ways: at random, by latent-space clustering of a pretrained MLFF (DIRECT), and by a new model-free method maximizing the diversity of discrete, interpretable local environments (EnvTok). All three models approach the full-data reference MACE-OMAT-0 model to within a few tens of meV/mathringA in forces, and remain close to or exceed it on formation energies, elastic moduli, heat capacities, Li+ migration barriers, and molecular-dynamics stability up to 3000K and 500GPa. This near-independence from the selection strategy shows that chemical environment diversity, not size, defines OMat24's information content: the remaining 99% of structures are largely redundant. Broad sampling of chemical space thus remains essential, but reference labels are needed only for structurally and chemically distinct configurations. Using atomic positions and species, optionally augmented by magnetic moments, EnvTok identifies these before any labelling or training, yielding compact, architecture-independent datasets. This hundredfold reduction makes hybrid-DFT or higher-level relabelling of general-purpose training sets feasible, and frees computational budget for extending environmental coverage and multi-fidelity learning.
Related
- Diversity-Aware Active Learning Builds Materials Datasets for Current and Future Objectives
- Neutral atom quantum computing for materials science and quantum chemistry
- Culinary Chemistry as a Platform for Interfacial Materials Design
- Chemistry-resolved disagreement between Materials Project-compatible PBE(+U) and r2SCAN stability verdicts in the GNoME materials database
- Measurement-efficient energy estimator for quantum chemistry of photodynamic materials
BibTeX
@misc{moerman2026atoms,
title = {{Atoms that Matter: Largest Materials Datasets Are 99% Redundant, and Local Chemistry Can Guide the Required 1%}},
author = {Evgeny Moerman and Igor Poltavsky and Anton Charkin-Gorbulin and Alexandre Tkatchenko},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.14392},
eprint = {2610.14392},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.14392},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Atoms that Matter: Largest Materials Datasets Are 99% Redundant, and Local Chemistry Can Guide the Required 1%
AU - Evgeny Moerman
AU - Igor Poltavsky
AU - Anton Charkin-Gorbulin
AU - Alexandre Tkatchenko
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.14392
UR - https://arxiv.org/abs/2610.14392
AB - General-purpose machine-learning force fields (MLFFs) for materials are now trained on the order of $100$ million structures, assuming transferability requires such scale. We test this premise on OMat24, the largest available materials dataset, by training MACE models on subsets of $6.5$ million atoms, less than 1% of OMat24, selected in three complementary ways: at random, by latent-space clustering of a pretrained MLFF (DIRECT), and by a new model-free method maximizing the diversity of discrete, interpretable local environments (EnvTok). All three models approach the full-data reference MACE-OMAT-0 model to within a few tens of meV/mathringA in forces, and remain close to or exceed it on formation energies, elastic moduli, heat capacities, Li^+ migration barriers, and molecular-dynamics stability up to 3000K and 500GPa. This near-independence from the selection strategy shows that chemical environment diversity, not size, defines OMat24's information content: the remaining 99% of structures are largely redundant. Broad sampling of chemical space thus remains essential, but reference labels are needed only for structurally and chemically distinct configurations. Using atomic positions and species, optionally augmented by magnetic moments, EnvTok identifies these before any labelling or training, yielding compact, architecture-independent datasets. This hundredfold reduction makes hybrid-DFT or higher-level relabelling of general-purpose training sets feasible, and frees computational budget for extending environmental coverage and multi-fidelity learning.
ER -