Universal\nPredictive Power: Introducing the Electronic\nStructure Decomposition Approach for CO Adsorption and Activation\non Al<sub>2</sub>O<sub>3</sub>‑Supported Ru Nanoparticles

Accurate\nprediction of catalyst performance is crucial for designing\nmaterials with specific catalytic functions. While the density functional\ntheory (DFT) method is widely used for its accuracy, modeling heterogeneous\nsystems, especially supported transition metals, poses significant\ncomputational challenges. To address these challenges, we introduce\nthe Electronic Structure Decomposition Approach (ESDA), a novel method\nthat identifies specific density of states (DOS) areas responsible\nfor adsorbate interaction and activation on the catalyst. As a case\nstudy, we investigate the influence of α-Al<sub>2</sub>O<sub>3</sub>(0001) as a support material on CO adsorption energy and the\nstretching frequency of the C–O bond on Ru nanoparticles (NPs).\nUsing multiple linear regression analysis, ESDA models were trained\nwith data from isolated Ru NPs and adjusted using supported NP sample\ndata. The ESDA models accurately predict the CO adsorption energies\nand C–O vibrational frequencies, demonstrating strong linear\ncorrelations between predicted and DFT-calculated values with low\nerrors across various adsorption sites for both isolated and supported\nRu NPs. Beyond pinpointing the DOS areas responsible for CO adsorption\nand C–O bond activation, this study provides insights into\nmanipulating these DOS areas to control CO activation, hence facilitating\nCO dissociation. Additionally, ESDA significantly accelerates the\ncharacterization and prediction of CO adsorption and activation on\nboth isolated and supported Ru NPs compared to DFT calculations, expediting\nthe design of new catalytic materials and advancing catalysis research.\nFurthermore, ESDA’s reliance on the electronic structure as\na descriptor suggests its potential for predicting various properties\nbeyond catalysis, broadening its applicability across diverse scientific\ndomains.