Benchmarking machine-learned interatomic potentials for reactive surface dynamics at metal surfaces: accuracy vs speed Wojciech Stark 1 , Julia Westermayr 1 , Cas van der Oord 2 , Lingjun Zhu 3 , Bin Jiang 3 , Gábor Csány 2 , Reinhard J. Maurer 1 1 University of Warwick, UK, 2 University of Cambridge,UK, 3 University of Science and Technology, China Machine-learned interatomic potentials (MLIP) have become widely used tools to accelerate ab initio molecular dynamics simulations in materials science. Many promising MLIPs emerged recently, ranging from simple linear models to deep neural networks (DNN), differing in stability, accuracy, and inference time. Reactive scattering dynamics are highly sensitive to potential corrugation and low reaction probabilities require extensive ensemble averaging. Therefore, MLIPs need to combine smooth and accurate landscapes with extremely efficient inference. In this study, we compare different families of MLIPs, from atomic cluster expansion (ACE) [1], message-passing based Neural Network SchNet [2] and embedded atom neural networks (EANN) [3], to equivariant neural networks such as PaiNN [4] and MACE [5] on the example of reactive molecular hydrogen scattering on copper. We compare these diverse methods by measuring accuracy and inference performance directly on dynamical observables. This provides a detailed picture of MLIP accuracy, speed, and learning rate that goes beyond simple train/test error analysis. References 1. R. Drautz, Phys. Rev. B 99, 014104 (2022) 2. K. T. Schütt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, K.-R. Müller, J. Chem. Phys. 148, 241722 (2018) 3. Y. Zhang, C. Hu, B. Jiang,J. Phys. Chem. Lett.10, 4962−4967 (2019) 4. K. T. Schütt, O. T. Unke, M. Gastegger, PMLR 139 (2021) 5. I. Batatia, D. P. Kovács, G. N. C. Simm, C. Ortner, G. Csányi, arXiv:2206.07697
DC07
© The Author(s), 2023
Made with FlippingBook Learn more on our blog