Uncertainty quantification for misspecified machine learned interatomic potentials

Abstract The use of high-dimensional regression techniques from machine learning has significantly improved the quantitative accuracy of interatomic potentials. Atomic simulations can now plausibly target quantitative predictions in a variety of settings, which has brought renewed interest in robust...

Full description

Saved in:
Bibliographic Details
Main Authors: Danny Perez, Aparna P. A. Subramanyam, Ivan Maliyov, Thomas D. Swinburne
Format: Article
Language:English
Published: Nature Portfolio 2025-08-01
Series:npj Computational Materials
Online Access:https://doi.org/10.1038/s41524-025-01758-4
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849735761518133248
author Danny Perez
Aparna P. A. Subramanyam
Ivan Maliyov
Thomas D. Swinburne
author_facet Danny Perez
Aparna P. A. Subramanyam
Ivan Maliyov
Thomas D. Swinburne
author_sort Danny Perez
collection DOAJ
description Abstract The use of high-dimensional regression techniques from machine learning has significantly improved the quantitative accuracy of interatomic potentials. Atomic simulations can now plausibly target quantitative predictions in a variety of settings, which has brought renewed interest in robust means to quantify uncertainties. In many practical settings where model complexity is constrained (e.g., due to performance considerations), misspecification — the inability of any one choice of model parameters to exactly match all training data — is a key contributor to errors that is often disregarded. Here, we employ a recent misspecification-aware regression technique to quantify parameter uncertainties, which is then propagated to a broad range of phase and defect properties in tungsten. The propagation is performed through both brute-force resampling and implicit Taylor expansion. The propagated misspecification uncertainties robustly quantify and bound errors on a broad range of material properties. We demonstrate application to recent foundational machine learning interatomic potentials, accurately predicting and bounding errors in MACE-MPA-0 energy predictions across the diverse materials project database.
format Article
id doaj-art-b94d3d8a6c624c7c95c29e491b81fe05
institution DOAJ
issn 2057-3960
language English
publishDate 2025-08-01
publisher Nature Portfolio
record_format Article
series npj Computational Materials
spelling doaj-art-b94d3d8a6c624c7c95c29e491b81fe052025-08-20T03:07:27ZengNature Portfolionpj Computational Materials2057-39602025-08-0111111510.1038/s41524-025-01758-4Uncertainty quantification for misspecified machine learned interatomic potentialsDanny Perez0Aparna P. A. Subramanyam1Ivan Maliyov2Thomas D. Swinburne3Theoretical Division T-1, Los Alamos National LaboratoryTheoretical Division T-1, Los Alamos National LaboratoryAix-Marseille Université, CNRS, CINaM UMR 7325, Campus de LuminyAix-Marseille Université, CNRS, CINaM UMR 7325, Campus de LuminyAbstract The use of high-dimensional regression techniques from machine learning has significantly improved the quantitative accuracy of interatomic potentials. Atomic simulations can now plausibly target quantitative predictions in a variety of settings, which has brought renewed interest in robust means to quantify uncertainties. In many practical settings where model complexity is constrained (e.g., due to performance considerations), misspecification — the inability of any one choice of model parameters to exactly match all training data — is a key contributor to errors that is often disregarded. Here, we employ a recent misspecification-aware regression technique to quantify parameter uncertainties, which is then propagated to a broad range of phase and defect properties in tungsten. The propagation is performed through both brute-force resampling and implicit Taylor expansion. The propagated misspecification uncertainties robustly quantify and bound errors on a broad range of material properties. We demonstrate application to recent foundational machine learning interatomic potentials, accurately predicting and bounding errors in MACE-MPA-0 energy predictions across the diverse materials project database.https://doi.org/10.1038/s41524-025-01758-4
spellingShingle Danny Perez
Aparna P. A. Subramanyam
Ivan Maliyov
Thomas D. Swinburne
Uncertainty quantification for misspecified machine learned interatomic potentials
npj Computational Materials
title Uncertainty quantification for misspecified machine learned interatomic potentials
title_full Uncertainty quantification for misspecified machine learned interatomic potentials
title_fullStr Uncertainty quantification for misspecified machine learned interatomic potentials
title_full_unstemmed Uncertainty quantification for misspecified machine learned interatomic potentials
title_short Uncertainty quantification for misspecified machine learned interatomic potentials
title_sort uncertainty quantification for misspecified machine learned interatomic potentials
url https://doi.org/10.1038/s41524-025-01758-4
work_keys_str_mv AT dannyperez uncertaintyquantificationformisspecifiedmachinelearnedinteratomicpotentials
AT aparnapasubramanyam uncertaintyquantificationformisspecifiedmachinelearnedinteratomicpotentials
AT ivanmaliyov uncertaintyquantificationformisspecifiedmachinelearnedinteratomicpotentials
AT thomasdswinburne uncertaintyquantificationformisspecifiedmachinelearnedinteratomicpotentials