主催: The Japan Society of Vacuum and Surface Science
会議名: 2023年日本表面真空学会学術講演会
開催地: 名古屋
開催日: 2023/10/31 - 2023/11/02
Single-atom alloy (SAA), which consists of a small amount of active metal atomically dispersed in a more inert host metal, has garnered interest as a catalyst for oxygen reduction reaction (ORR) [1]. In the past years, researchers aimed to characterize SAAs with good reactivity and stability. The stability can be analyzed using the segregation energy, which measures the preference of the dopant atom to segregate to the topmost layer of the surface. While segregation happens on pristine surfaces, it is important to note that adsorbates can also induce dopant segregation. Due to the complexity of the SAA system, identifying the significant factors influencing dopant segregation remains a challenge [2].
Hence, we investigated dopant segregation and identified the significant factors influencing it by performing density functional theory (DFT)-based calculations and machine learning (ML) methods. We generated SAA surfaces of Ag, Au, Co, Cu, Ir, Ni, Pd, Pt, and Rh and used O and OH as adsorbates, key ORR reactants. We calculated the adsorption energies and the segregation energies with and without the presence of the adsorbates. We considered a set of 44 features encompassing the elemental, energetics, and electronic properties of the SAAs. We performed a two-stage feature selection method for both O- and OH-SAA systems which reduced the features to the top five most influential – formation energies, metallic radius difference, d-band centers of the dopant at the surface, and subsurface layer, the difference in surface energy between the host and dopant atom, and difference in the total number of d-electrons between the host and dopant atom. Using these identified features, we implemented various ML models – linear regression (LR), support vector machine regression (SVR), Gaussian process regression (GPR), and extra trees regression (ETR) – to predict adsorbate-induced dopant segregation energies. We found that the SVR model, both for O-SAA and OH-SAA, exhibited the best performance among the models. For O-SAA, SVR performance metrics are R2=0.92, RMSE=0.11, and MAE=0.09 for the train set; and R2=0.94, RMSE=0.10, and MAE=0.07 for the test set while the performance metrics for OH-SAA are R2=0.81, RMSE=0.016, and MAE=0.13 for the train set; and R2=0.91, RMSE=0.13, and MAE=0.10 for the test set. Also, we identified Rh-Au(111) as a potential ORR catalyst based on the criteria – good reactivity for ORR catalysis and good stability with and without adsorbates.
References:
[1] Hannagan, R. T., Giannakakis, G., Flytzani-Stephanopoulos, M., & Sykes, E. C. H. (2020). Single-Atom Alloy Catalysis. Chemical Reviews. doi:10.1021/acs.chemrev.0c00078
[2] Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018). Machine learning for molecular and materials science. Nature, 559(7715), 547–555. doi:10.1038/s41586-018-0337-2