Unraveling active ensembles consisting of clusters and single atoms for oxygen reduction: a synergy of machine learning and DFT calculations

Abstract

Catalytic ensembles combining nanoparticles/clusters and atomically dispersed metal sites have demonstrated promising performance for various reactions. However, the optimal combinations between nanoparticles/clusters and metal single atoms remain unexplored. Herein, we integrate machine learning (ML) with density functional theory (DFT) calculations to explore the active ensembles consisting of platinum-based metallic clusters (Pt₃M) and nitrogen-coordinated metal single atoms on the N-doped graphene (NC) matrix (Pt₃M-M′NC). A total of 1521 candidates were screened using readily available metal properties to estimate the oxygen reduction reaction overpotential (η^ORR), resulting in the identification of 24 active Pt₃M-M′NC catalysts. Furthermore, the durability based on the ab initio molecular dynamics (AIMD) simulations, dissolution potential (U_diss), and cluster energies (E_cluster) was screened to identify four active and durable Pt₃M-M′NC ensembles. The quantitative relationship between η^ORR and metal features is deduced, enabling rapid and cost-effective screening of the optimal Pt₃M-M′NC ensemble for the ORR. This work provides a comprehensive framework for the rational design of efficient and durable catalysts for oxygen reduction, leveraging the synergistic power of machine learning and DFT calculations to optimize catalytic ensembles with high performance.