Machine-Learning-Assisted Prediction of Catalytic Activity of Alloy Spherical Nanoparticles for the Hydrogen Evolution Reaction

Abstract

Alloy nanoparticles are essential catalysts for the hydrogen evolution reaction (HER), but predicting their catalytic behavior at experimental scales remains a challenge. To address this, we developed an approach integrating machine learning-based moment tensor potentials (MTPs) with active learning to predict the binding free energies of active sites on alloy spherical nanoparticles (SNPs). We evaluated HER activities across various morphologies and compositions, including PtAu (core-shell), PtNi (solid solution), and PtCo (size/composition-dependent morphology). Analyzing the correlation between the generalized coordination number (GCN) and HER overpotential (ηHER), we identified the critical role of local environments in optimizing HER performance. We propose the "crown-jewel" structure as the optimal morphology for Pt-based alloy SNPs, leveraging synergistic effects between Pt and heteroatoms. Additionally, we predicted turnover frequency (TOF) at different electrode potentials, showing a similar trend to the GCN-ηHER correlation. Our results reveal that Pt@Au core-shell SNPs exhibit the highest HER activity at Pt0.92Au0.08, while Co@Pt core-shell SNPs perform best at Pt0.08Co0.92​, reducing Pt dependency. For PtNi SNPs, the solid solution structure at Pt0.75Ni0.25 shows the highest HER activity across all sizes. This study provides key insights for catalyst optimization and experimental design.

Keywords: Alloy nanoparticle, Moment tensor potential, Hydrogen evolution reaction, Energetic span model, Turnover frequency