Benchmarking of machine learning interatomic potentials for reactive hydrogen dynamics at metal surfaces - Benchmarking

Aug 5, 2026

Upload by Maeve

Simulations of chemical reaction probabilities in gas surface dynamics require ensemble averages over tens of thousands of reactive events and accurate mapping of energy landscapes, since small barrier errors strongly affect predicted reaction rates. The work benchmarks leading machine learning interatomic potentials for inference speed on CPUs and suitability for high-throughput reactive chemistry at metal surfaces. Models tested include PaiNN, REANN, MACE, and ACE on reactive hydrogen scattering at copper facets, evaluating accuracy, time-to-solution, and sticking probabilities.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Abstract

Introduction

Benchmarking approach

Models and dataset

Evaluation metrics

Results

Accuracy and inference performance

Reactive sticking probabilities

Conclusion

BENCHMARK • OPEN ACCESS Benchmarking of machine learning interatomic potentials for reactive hydrogen dynamics at metal surfaces To cite this article: Wojciech G Stark et al 2024 Mach. Learn. : Sci. Technol. 5 030501 View the article online for updates and enhancements. You may also like -Modeling and experimental investigation of multilayer DE transducers considering the influence of the electrode layers Jana Mertens, Abd Elkarim Masoud, Andreas Lars Peter Hubracht et al. -Footprint tools tiptoeing towards nitrogen sustainability James N Galloway, Rachel E Michaels, Elizabeth A Castner et al. -Can the court bridge the gap? Public perception of economic vs. generational inequalities in climate change mitigation policies Nanna Lauritz Schönhage, Theresa Wieland, Luna Bellani et al. This content was downloaded from IP address [131.111.184.102](131.111.184.102) on 27/08/2024 at 14:15 Mach. Learn.: Sci. Technol. 5 (2024) 030501 [https://doi.org/10.1088/2632-2153/ad5f11](https://doi.org/10.1088/2632-2153/ad5f11) OPEN ACCESS RECEIVED 3 March 2024 REVISED 17 May 2024 ACCEPTED FOR PUBLICATION 3 July 2024 PUBLISHED 15 July 2024 Original Content from this work may be used under the terms of the Creative Commons Attribution 4 .0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. BENCHMARK Benchmarking of machine learning interatomic potentials for reactive hydrogen dynamics at metal surfaces Wojciech G Stark1􀁂, Cas van der Oord2􀁂, Ilyes Batatia2􀁂, Yaolong Zhang3􀁂, Bin Jiang4,5􀁂, Gábor Csányi2􀁂 and Reinhard J Maurer1,6, ∗􀁂 1 Department of Chemistry, University of Warwick, Gibbet Hill Road, Coventry CV4 7AL, United Kingdom 2 Department of Engineering, Cambridge CB2 1PZ, United Kingdom 3 Department of Chemistry and Chemical Biology, Center for Computational Chemistry, University of New Mexico, Albuquerque, NM 87131, United States of America 4 Key Laboratory of Precision and Intelligent Chemistry, Department of Chemical Physics, University of Science and Technology of China, Hefei, Anhui, People’s Republic of China 5 Hefei National Laboratory, University of Science and Technology of China, Hefei 230088, People’s Republic of China 6 Department of Physics, University of Warwick, Gibbet Hill Road, Coventry CV4 7AL, United Kingdom ∗ Author to whom any correspondence should be addressed. [E-mail:](E-mail: r.maurer@warwick.ac.uk)[ r.maurer@warwick.ac.uk](E-mail: r.maurer@warwick.ac.uk) Keywords: molecular dynamics simulations, electronic structure theory, gas surface dynamics, machine learning model inference performance, reactive scattering, hydrogen surface chemistry Supplementary material for this article is available online Abstract Simulations of chemical reaction probabilities in gas surface dynamics require the calculation of ensemble averages over many tens of thousands of reaction events to predict dynamical observables that can be compared to experiments. At the same time, the energy landscapes need to be accurately mapped, as small errors in barriers can lead to large deviations in reaction probabilities. This brings a particularly interesting challenge for machine learning interatomic potentials, which are becoming well-established tools to accelerate molecular dynamics simulations. We compare state-of-the-art machine learning interatomic potentials with a particular focus on their inference performance on CPUs and suitability for high throughput simulation of reactive chemistry at surfaces. The considered models include polarizable atom interaction neural networks (PaiNN), recursively embedded atom neural networks (REANN), the MACE equivariant graph neural network, and atomic cluster expansion potentials (ACE) . The models are applied to a dataset on reactive molecular hydrogen scattering on low-index surface facets of copper. All models are assessed for their accuracy

FAQs

Why is benchmarking ML interatomic potentials important for reactive gas–surface dynamics?

Gas–surface reaction predictions need very large ensemble averages and highly accurate energy landscapes; even small barrier errors can cause large deviations. Benchmarking ensures models are both accurate and fast enough for high-throughput simulations.

Which machine learning interatomic potential models are compared in the study?

The study compares PaiNN, REANN, MACE, and ACE potentials, focusing on CPU inference performance and performance for reactive chemistry at metal surfaces.

What dataset and physical scenario are used for the benchmark?

All models are applied to a dataset of reactive molecular hydrogen scattering on low-index copper surface facets, and are assessed using reactive sticking probabilities across rovibrational and kinetic incidence conditions.

Aug 5, 2026

Upload by Maeve