How graph neural network interatomic potentials extrapolate: Role of the message-passing algorithm DOI
Sungwoo Kang

The Journal of Chemical Physics, Год журнала: 2024, Номер 161(24)

Опубликована: Дек. 23, 2024

Graph neural network interatomic potentials (GNN-IPs) are gaining significant attention due to their capability of learning from large datasets. Specifically, universal based on GNN, usually trained with crystalline geometries, often exhibit remarkable extrapolative behavior toward untrained domains, such as surfaces and amorphous configurations. However, the origin this extrapolation is not well understood. This work provides a theoretical explanation how GNN-IPs extrapolate geometries. First, we demonstrate that can capture non-local electrostatic interactions through message-passing algorithm, evidenced by tests toy models density-functional theory data. We find GNN-IP models, SevenNet MACE, accurately predict forces in indicating they have learned exact functional form Coulomb interaction. Based these results, suggest ability learn interactions, coupled embedding nature GNN-IPs, explains ability. GNN-IP, SevenNet-0, effectively infers domains but fails arising kinetic term, which supports suggested theory. Finally, address impact hyperparameters performance potentials, SevenNet-0 MACE-MP-0, discuss limitations capabilities.

Язык: Английский

New trends in nanoparticle exsolution DOI Creative Commons
Alfonso J. Carrillo, Andrés López‐García, Blanca Delgado-Galicia

и другие.

Chemical Communications, Год журнала: 2024, Номер 60(62), С. 7987 - 8007

Опубликована: Янв. 1, 2024

Many relevant high-temperature chemical processes require the use of oxide-supported metallic nanocatalysts. The harsh conditions under which these operate can trigger catalyst degradation

Язык: Английский

Процитировано

8
Applications of machine learning in surfaces and interfaces DOI Open Access
Shaofeng Xu, Jing‐Yuan Wu, Ying Guo

и другие.

Chemical Physics Reviews, Год журнала: 2025, Номер 6(1)

Опубликована: Март 1, 2025

Surfaces and interfaces play key roles in chemical material science. Understanding physical processes at complex surfaces is a challenging task. Machine learning provides powerful tool to help analyze accelerate simulations. This comprehensive review affords an overview of the applications machine study systems materials. We categorize into following broad categories: solid–solid interface, solid–liquid liquid–liquid surface solid, liquid, three-phase interfaces. High-throughput screening, combined first-principles calculations, force field accelerated molecular dynamics simulations are used rational design such as all-solid-state batteries, solar cells, heterogeneous catalysis. detailed information on for

Язык: Английский

Процитировано

1
Application of Machine Learning Interatomic Potentials in Heterogeneous Catalysis DOI

Gbolagade Olajide,

Khagendra Baral, Sophia Ezendu

и другие.

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Application of machine learning interatomic potentials in heterogeneous catalysis DOI

Gbolagade Olajide,

Khagendra Baral, Sophia Ezendu

и другие.

Journal of Catalysis, Год журнала: 2025, Номер unknown, С. 116202 - 116202

Опубликована: Май 1, 2025

Язык: Английский

Процитировано

0
Machine learning perspective: Revealing deep mechanisms and new advances in adsorption and catalysis of gaseous molecules DOI
Yue Zhu, Fengyu Gao,

Lei Yi

и другие.

Applied Energy, Год журнала: 2025, Номер 396, С. 126241 - 126241

Опубликована: Июнь 6, 2025

Язык: Английский

Процитировано

0
How graph neural network interatomic potentials extrapolate: Role of the message-passing algorithm DOI
Sungwoo Kang

The Journal of Chemical Physics, Год журнала: 2024, Номер 161(24)

Опубликована: Дек. 23, 2024

Graph neural network interatomic potentials (GNN-IPs) are gaining significant attention due to their capability of learning from large datasets. Specifically, universal based on GNN, usually trained with crystalline geometries, often exhibit remarkable extrapolative behavior toward untrained domains, such as surfaces and amorphous configurations. However, the origin this extrapolation is not well understood. This work provides a theoretical explanation how GNN-IPs extrapolate geometries. First, we demonstrate that can capture non-local electrostatic interactions through message-passing algorithm, evidenced by tests toy models density-functional theory data. We find GNN-IP models, SevenNet MACE, accurately predict forces in indicating they have learned exact functional form Coulomb interaction. Based these results, suggest ability learn interactions, coupled embedding nature GNN-IPs, explains ability. GNN-IP, SevenNet-0, effectively infers domains but fails arising kinetic term, which supports suggested theory. Finally, address impact hyperparameters performance potentials, SevenNet-0 MACE-MP-0, discuss limitations capabilities.

Язык: Английский

Процитировано

0