Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations

Active learning (AL) requires massive time for comprehensive sampling of complex potential energy surfaces to achieve desirable accuracy and stability of machine learning (ML) potentials. Here, we develop an active delta-learning (ADL) protocol for speeding up AL and building delta-learning models y...

Full description

Saved in:
Bibliographic Details
Main Authors: Yaohuang Huang, Yi-Fan Hou, Pavlo O Dral
Format: Article
Language:English
Published: IOP Publishing 2025-01-01
Series:Machine Learning: Science and Technology
Subjects:
Online Access:https://doi.org/10.1088/2632-2153/adeb46
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850112766229086208
author Yaohuang Huang
Yi-Fan Hou
Pavlo O Dral
author_facet Yaohuang Huang
Yi-Fan Hou
Pavlo O Dral
author_sort Yaohuang Huang
collection DOAJ
description Active learning (AL) requires massive time for comprehensive sampling of complex potential energy surfaces to achieve desirable accuracy and stability of machine learning (ML) potentials. Here, we develop an active delta-learning (ADL) protocol for speeding up AL and building delta-learning models yielding stable simulations. ADL converges after a few iterations and needs tenfold fewer sampled points than without delta-learning while leading to models of similar accuracy, as we show on the test simulations of Diels–Alder reactions. The test reactions include one small (ethene + 1,3-butadiene) and one relatively big (C _60 + 2,3-dimethyl-1,3-butadiene) system, treated with a target density functional theory level (U)B3LYP(-D4)/6-31G* and a baseline semi-empirical quantum mechanical method, GFN2-xTB. The crucial advantage of the models built with the delta-learning protocol is their remarkable simulation stability: even models from the initial ADL iterations yield reasonable results. In contrast, the pure ML potentials built without delta-learning often lead to the collapse in simulations, i.e. to unphysical structures.
format Article
id doaj-art-7ca137dee62746ab98402fe1c83ccbd8
institution OA Journals
issn 2632-2153
language English
publishDate 2025-01-01
publisher IOP Publishing
record_format Article
series Machine Learning: Science and Technology
spelling doaj-art-7ca137dee62746ab98402fe1c83ccbd82025-08-20T02:37:18ZengIOP PublishingMachine Learning: Science and Technology2632-21532025-01-016303500410.1088/2632-2153/adeb46Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulationsYaohuang Huang0https://orcid.org/0000-0001-6942-3499Yi-Fan Hou1https://orcid.org/0000-0001-9188-5323Pavlo O Dral2https://orcid.org/0000-0002-2975-9876State Key Laboratory of Physical Chemistry of Solid Surfaces, Department of Chemistry, College of Chemistry and Chemical Engineering, and Fujian Provincial Key Laboratory of Theoretical and Computational Chemistry, Xiamen University , Xiamen 361005, People’s Republic of ChinaState Key Laboratory of Physical Chemistry of Solid Surfaces, Department of Chemistry, College of Chemistry and Chemical Engineering, and Fujian Provincial Key Laboratory of Theoretical and Computational Chemistry, Xiamen University , Xiamen 361005, People’s Republic of ChinaState Key Laboratory of Physical Chemistry of Solid Surfaces, Department of Chemistry, College of Chemistry and Chemical Engineering, and Fujian Provincial Key Laboratory of Theoretical and Computational Chemistry, Xiamen University , Xiamen 361005, People’s Republic of China; Institute of Physics, Faculty of Physics, Astronomy, and Informatics, Nicolaus Copernicus University in Toruń , ul. Grudziądzka 5, 87-100 Toruń, Poland; Aitomistic , Shenzhen 518000, People’s Republic of ChinaActive learning (AL) requires massive time for comprehensive sampling of complex potential energy surfaces to achieve desirable accuracy and stability of machine learning (ML) potentials. Here, we develop an active delta-learning (ADL) protocol for speeding up AL and building delta-learning models yielding stable simulations. ADL converges after a few iterations and needs tenfold fewer sampled points than without delta-learning while leading to models of similar accuracy, as we show on the test simulations of Diels–Alder reactions. The test reactions include one small (ethene + 1,3-butadiene) and one relatively big (C _60 + 2,3-dimethyl-1,3-butadiene) system, treated with a target density functional theory level (U)B3LYP(-D4)/6-31G* and a baseline semi-empirical quantum mechanical method, GFN2-xTB. The crucial advantage of the models built with the delta-learning protocol is their remarkable simulation stability: even models from the initial ADL iterations yield reasonable results. In contrast, the pure ML potentials built without delta-learning often lead to the collapse in simulations, i.e. to unphysical structures.https://doi.org/10.1088/2632-2153/adeb46machine learning interatomic potentialsdelta learningactive learningchemical reactionsmolecular dynamicstransition state
spellingShingle Yaohuang Huang
Yi-Fan Hou
Pavlo O Dral
Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
Machine Learning: Science and Technology
machine learning interatomic potentials
delta learning
active learning
chemical reactions
molecular dynamics
transition state
title Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
title_full Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
title_fullStr Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
title_full_unstemmed Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
title_short Active delta-learning for fast construction of interatomic potentials and stable molecular dynamics simulations
title_sort active delta learning for fast construction of interatomic potentials and stable molecular dynamics simulations
topic machine learning interatomic potentials
delta learning
active learning
chemical reactions
molecular dynamics
transition state
url https://doi.org/10.1088/2632-2153/adeb46
work_keys_str_mv AT yaohuanghuang activedeltalearningforfastconstructionofinteratomicpotentialsandstablemoleculardynamicssimulations
AT yifanhou activedeltalearningforfastconstructionofinteratomicpotentialsandstablemoleculardynamicssimulations
AT pavloodral activedeltalearningforfastconstructionofinteratomicpotentialsandstablemoleculardynamicssimulations