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

Wojciech G. Stark,Cas van der Oord,Ilyes Batatia,Yaolong Zhang,Bin Jiang,Gábor Csányi,Reinhard J. Maurer

2024-03-23

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, 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.

Chemical Physics

What problem does this paper attempt to address?

This paper evaluates the performance of machine-learning interatomic potentials (MLIPs) for simulating hydrogen dynamics on metal surfaces. Four different models are compared in the study: polarizable atomic interaction neural network (PaiNN), recursive embedding atomic neural network (REANN), modified attention-based graph neural network (MACE), and atomic cluster expansion potential (ACE). These models are applied on a dataset of hydrogen molecule reaction scattering on low-indexed copper surfaces to assess their accuracy, computational speed, and ability to simulate reaction adhesion probability.

The study finds that the REANN and MACE models achieve the best balance between accuracy and computational efficiency and are considered the state-of-the-art for gas-surface dynamics. Although the PaiNN model performs well in accuracy, it requires a large number of features, leading to decreased computational efficiency. On the other hand, the ACE model has the fastest computational speed but may not achieve sufficient predictive accuracy in certain cases.

The paper emphasizes the need to handle a large number of events in simulating chemical reaction probabilities, which makes first-principles electronic structure methods like density functional theory impractical. Efficient MLIPs are therefore required. While MLIPs have significantly accelerated molecular dynamics simulations, simulating surface chemical dynamics remains challenging, especially for predicting statistically reliable reaction probabilities that require a large number of MD trajectories.

Overall, the paper aims to address the problem of selecting an efficient and accurate MLIP model suitable for high-throughput simulation of surface reaction dynamics and provides guidance through a performance comparison of different models.