Computational Chemistry and Machine Learning-Assisted Screening of Supported Amorphous Metal Oxide Nanoclusters for Methane Activation

Abstract

Activating the C−H bond in methane represents a cornerstone challenge in catalytic research. While several supported metal oxide nanoclusters (MeO-NCs) have shown promise for this reaction, their optimal composition remains underexplored primarily due to the large number of possible compositions and their amorphous nature. This study addresses these challenges using computational approaches. Leveraging density functional theory (DFT) calculations, we began with a previously studied supported tetra-copper oxide nanocluster and systematically substituted its Cu sites with first-row transition metals (Mn, Fe, Co, Ni, and Zn). This process allowed us to examine the catalytic activity of 162 MeO-NCs with a variety of geometric and electronic structures, leading to 12 new compositions that outperformed the base nanocluster. Exploring the structure−activity relationships with machine learning, our analysis uncovered correlations between the intrinsic electronic and structural properties of the nanoclusters and the free energy barriers for methane activation despite the challenges posed by the structural flexibility of these amorphous nanoclusters. The results offer insights into the optimization of MeO-NCs for methane activation. Additionally, we developed a clustering model capable of distinguishing high-performing nanoclusters from less effective ones with strong tolerance to the interference from the structural flexibility of these amorphous nanoclusters. These findings help narrow down the material design space for more timeconsuming high-level quantum chemical calculations, offering a promising pathway toward advancing eco-friendly methane conversion.