Dyna-Mat: End-to-end benchmarking of foundation machine learning interatomic potentials in finite-temperature ensembles
16 affiliations
- 1Department of Physics and Astronomy , University College London , 7-19 Gordon St , London WC1H 0AH , UK
- 2Thomas Young Centre and London Centre for Nanotechnology , 9 Gordon St , London WC1H 0AH , UK
- 3Debye Institute for Nanomaterials Science , Utrecht University , 3508 TA Utrecht , the Netherlands
- 4NOMATEN Centre of Excellence , National Center for Nuclear Research , ul. A. Sołtana 7 , 05-400 Swierk / Otwock , Poland
- 5Periodic Labs , Menlo Park , CA , USA
- 6Fritz-Haber-Institut der
- 7Center for Interdisciplinary Research and Innovation , Aristotle University of Thessaloniki , GR-57001 Thessaloniki , Greece
- 8Physics Department , Aristotle University of Thessaloniki , GR-54124 Thessaloniki , Greece
- 9Ruđer Boškovi'c Institute , Bijenička cesta 54 , 10000 Zagreb , Croatia
- 10The Faculty of Industrial Engineering , Mechanical Engineering, and Computer Science , University of Iceland , Reykjavík , Iceland
- 11MPI for the Structure and Dynamics of Matter , Luruper Chaussee 148 , 22761 Hamburg , Germany
- 12Yusuf Hamied Department of Chemistry , Lensfield Road , CB2 1EW Cambridge , UK
- 13Luxembourg Institute of Science and Technology (LIST) , 5 Avenue des Hauts-Fourneaux , L-4362 Esch-sur-Alzette , Luxembourg
- 14Laboratory of Computational Science and Modeling , Institut des Mat'eriaux , 'Ecole Polytechnique F'ed'erale de Lausanne , 1015 Lausanne , Switzerland
- 15Initiative for Computational Catalysis , Flatiron Institute , New York , NY 10010 , USA
- 16Faculty of Materials Science and Engineering , Warsaw University of Technology , Wołoska 141 , 02-507 Warsaw , Poland
Abstract
Foundation machine learning interatomic potentials (MLIPs) are increasingly being used as drop-in replacements for first-principles calculations, enabling simulations of materials at length and time scales that were previously inaccessible. However, due to lack of ground truth data, their accuracy on structural and dynamical observables in finite thermodynamic ensembles is yet to be established. Here, we introduce Dyna-Mat-v1.0, a benchmark dataset of condensed-phase first-principles molecular dynamics trajectories designed to test foundation MLIPs at realistic finite-temperature conditions. Using this dataset, we evaluate 15 foundation MLIPs across four model tiers by comparing both single-point energy and force errors on first-principles configurations and observables generated from MLIP-driven trajectories. We find that "on average" models with lower single-point force errors also yield lower errors for structural and dynamical observables. However, there are individual systems for which low force errors lead to qualitative failures in the predicted structure. Pressure remains poorly described across most models, pointing to limitations in the density functional theory stress labels available in current large-scale training datasets. Finally, we construct an accuracy-cost Pareto frontier to identify the best trade-offs for molecular dynamics with foundation MLIPs, finding that the latest generation of cross-trained models is close to Pareto-optimal according to the accuracy metrics considered here. Overall, Dyna-Mat-v1.0 shows that end-to-end finite-temperature validation is essential for quantifying the predictive behaviour of foundation MLIPs, and provides a simple, scalable route for assessing them beyond static and harmonic benchmarks relevant to materials design.
Related
- Finite-temperature dynamic stability separates foundation machine-learning interatomic potentials that harmonic benchmarks rank equally
- Foundation interatomic potentials learn the practical chemistry they were never taught
- Bayesian neural networks versus deep ensembles for uncertainty quantification in machine learning interatomic potentials
- Leveraging neural network interatomic potentials for a foundation model of chemistry
- Benchmarking Machine-Learning Interatomic Potentials for Dynamical Stability in Inorganic Semiconductor Nanocrystals: A CdSe Case Study
BibTeX
@misc{gawkowski2026dynamat,
title = {{Dyna-Mat: End-to-end benchmarking of foundation machine learning interatomic potentials in finite-temperature ensembles}},
author = {Mikołaj Jan Gawkowski and Nongnuch Artrith and Silvia Bonfanti and Abhijeet S. Gangan and Hendrik H. Heenen and Joseph Kioseoglou and Ivor Lončarić and Hemanadhan Myneni and Janosh Riebesell and M. Rossi and Matthias Rupp and Jonathan Schmidt and Sharma, S, K and Benjamin X. Shi and Antoni Wadowski and Lukas Hörmann and Venkat Kapil},
howpublished = {arXiv (Cornell University)},
year = {2026},
month = jul,
doi = {10.48550/arxiv.2607.03433},
url = {https://arxiv.org/abs/2607.03433},
note = {CC BY-SA},
}
RIS
TY - UNPB
TI - Dyna-Mat: End-to-end benchmarking of foundation machine learning interatomic potentials in finite-temperature ensembles
AU - Mikołaj Jan Gawkowski
AU - Nongnuch Artrith
AU - Silvia Bonfanti
AU - Abhijeet S. Gangan
AU - Hendrik H. Heenen
AU - Joseph Kioseoglou
AU - Ivor Lončarić
AU - Hemanadhan Myneni
AU - Janosh Riebesell
AU - M. Rossi
AU - Matthias Rupp
AU - Jonathan Schmidt
AU - Sharma, S, K
AU - Benjamin X. Shi
AU - Antoni Wadowski
AU - Lukas Hörmann
AU - Venkat Kapil
PY - 2026
DA - 2026/07/03
PB - arXiv (Cornell University)
DO - 10.48550/arxiv.2607.03433
UR - https://arxiv.org/abs/2607.03433
AB - Foundation machine learning interatomic potentials (MLIPs) are increasingly being used as drop-in replacements for first-principles calculations, enabling simulations of materials at length and time scales that were previously inaccessible. However, due to lack of ground truth data, their accuracy on structural and dynamical observables in finite thermodynamic ensembles is yet to be established. Here, we introduce Dyna-Mat-v1.0, a benchmark dataset of condensed-phase first-principles molecular dynamics trajectories designed to test foundation MLIPs at realistic finite-temperature conditions. Using this dataset, we evaluate 15 foundation MLIPs across four model tiers by comparing both single-point energy and force errors on first-principles configurations and observables generated from MLIP-driven trajectories. We find that "on average" models with lower single-point force errors also yield lower errors for structural and dynamical observables. However, there are individual systems for which low force errors lead to qualitative failures in the predicted structure. Pressure remains poorly described across most models, pointing to limitations in the density functional theory stress labels available in current large-scale training datasets. Finally, we construct an accuracy-cost Pareto frontier to identify the best trade-offs for molecular dynamics with foundation MLIPs, finding that the latest generation of cross-trained models is close to Pareto-optimal according to the accuracy metrics considered here. Overall, Dyna-Mat-v1.0 shows that end-to-end finite-temperature validation is essential for quantifying the predictive behaviour of foundation MLIPs, and provides a simple, scalable route for assessing them beyond static and harmonic benchmarks relevant to materials design.
SN - 2331-8422
ER -