ScholarWorks@성균관대학교

초록

Alloy nanoparticles are a promising class of materials for catalyzing the hydrogen evolution reaction (HER) but predicting their behavior at large scale and balancing catalytic performance with Pt utilization through composition and morphology control remain major challenges. To accelerate catalyst discovery and evaluate HER activity, we present a machine-learning-based framework to predict the performance of alloy spherical nanoparticles (SNPs), by integrating moment tensor potentials (MTPs) with active learning. This approach enables efficient evaluation of binding free energies and turnover frequencies (TOF) across a wide range of morphologies, compositions, and particle sizes, including PtAu (core–shell), PtCo (size/composition-dependent), and PtNi (solid-solution/intermetallic). Catalytic activity evaluations reveal optimal compositions for the highest HER performance. Notably, Pt@Au exhibits the best activity at Pt0.92Au0.08, while Co@Pt performs well even at the lowest Pt content (Pt0.08Co0.92), suggesting a tunable strategy for balancing catalytic performance and Pt utilization. In the PtNi system, intermetallic Pt0.75Ni0.25 exhibits the highest activity across all particle sizes, highlighting the influence of order–disorder transitions in solid-solution alloys. While Pt remains the most active HER site, our results show that catalytic performance varies with the local environment, which is strongly affected by nanoparticle composition and morphology. To better capture this relationship, we introduce a local environment descriptor that outperforms the conventional generalized coordination number (GCN) in predicting catalytic activity. These findings highlight the importance of atomic-level local environments and demonstrate how machine learning interatomic potentials can guide the efficient design of cost-effective HER catalysts.

키워드