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

被引:0
|
作者
Stark, Wojciech G. [1 ]
van der Oord, Cas [2 ]
Batatia, Ilyes [2 ]
Zhang, Yaolong [3 ]
Jiang, Bin [4 ,5 ]
Csanyi, Gabor [2 ]
Maurer, Reinhard J. [1 ,6 ]
机构
[1] Univ Warwick, Dept Chem, Gibbet Hill Rd, Coventry CV4 7AL, England
[2] Dept Engn, Inst Mfg, Dept Engn, Cambridge CB2 1PZ, England
[3] Univ New Mexico, Ctr Computat Chem, Dept Chem & Chem Biol, Albuquerque, NM 87131 USA
[4] Univ Sci & Technol China, Dept Chem Phys, Key Lab Precis & Intelligent Chem, Hefei, Anhui, Peoples R China
[5] Univ Sci & Technol China, Hefei Natl Lab, Hefei 230088, Peoples R China
[6] Univ Warwick, Dept Phys, Gibbet Hill Rd, Coventry CV4 7AL, England
来源
基金
英国工程与自然科学研究理事会;
关键词
molecular dynamics simulations; electronic structure theory; gas surface dynamics; machine learning model inference performance; reactive scattering; hydrogen surface chemistry; MODIFIED SHEPARD INTERPOLATION; ENERGY SURFACES; DISSOCIATIVE CHEMISORPTION; H-2; ADSORPTION; MOLECULE; CU(111); TEMPERATURE; SCATTERING;
D O I
10.1088/2632-2153/ad5f11
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
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, time-to-solution, and ability to simulate reactive sticking probabilities as a function of the rovibrational initial state and kinetic incidence energy of the molecule. REANN and MACE models provide the best balance between accuracy and time-to-solution and can be considered the current state-of-the-art in gas-surface dynamics. PaiNN models require many features for the best accuracy, which causes significant losses in computational efficiency. ACE models provide the fastest time-to-solution, however, models trained on the existing dataset were not able to achieve sufficiently accurate predictions in all cases.
引用
收藏
页数:17
相关论文
共 50 条
  • [1] Machine Learning Interatomic Potentials for Reactive Hydrogen Dynamics at Metal Surfaces Based on Iterative Refinement of Reaction Probabilities
    Stark, Wojciech G.
    Westermayr, Julia
    Douglas-Gallardo, Oscar A.
    Gardner, James
    Habershon, Scott
    Maurer, Reinhard J.
    [J]. JOURNAL OF PHYSICAL CHEMISTRY C, 2023, 127 (50): : 24168 - 24182
  • [2] Benchmarking machine learning interatomic potentials via phonon anharmonicity
    Bandi, Sasaank
    Jiang, Chao
    Marianetti, Chris A.
    [J]. MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 2024, 5 (03):
  • [3] Accelerating the prediction of inorganic surfaces with machine learning interatomic potentials
    Noordhoek, Kyle
    Bartel, Christopher J.
    [J]. NANOSCALE, 2024, 16 (13) : 6365 - 6382
  • [4] Lattice dynamics simulation using machine learning interatomic potentials
    Ladygin, V. V.
    Korotaev, P. Yu
    Yanilkin, A., V
    Shapeev, A., V
    [J]. COMPUTATIONAL MATERIALS SCIENCE, 2020, 172
  • [5] Benchmarking structural evolution methods for training of machine learned interatomic potentials
    Waters, Michael J.
    Rondinelli, James M.
    [J]. JOURNAL OF PHYSICS-CONDENSED MATTER, 2022, 34 (38)
  • [6] Machine Learning Interatomic Potentials for Heterogeneous Catalysis
    Tang, Deqi
    Ketkaew, Rangsiman
    Luber, Sandra
    [J]. CHEMISTRY-A EUROPEAN JOURNAL, 2024,
  • [7] Performance Assessment of Universal Machine Learning Interatomic Potentials: Challenges and Directions for Materials' Surfaces
    Focassio, Bruno
    Freitas, Luis Paulo M.
    Schleder, Gabriel R.
    [J]. ACS APPLIED MATERIALS & INTERFACES, 2024,
  • [8] Machine Learning of Reactive Potentials
    Yang, Yinuo
    Zhang, Shuhao
    Ranasinghe, Kavindri D.
    Isayev, Olexandr
    Roitberg, Adrian E.
    [J]. ANNUAL REVIEW OF PHYSICAL CHEMISTRY, 2024, 75 : 371 - 395
  • [10] Adaptive loss weighting for machine learning interatomic potentials
    J. Mike Walker ’66 Department of Mechanical Engineering, Texas A&M University, College Station
    TX
    77843, United States
    不详
    TX
    78249, United States
    不详
    TX
    77843, United States
    [J]. Comput Mater Sci, 2024,