Machine Learning Study of Methane Activation by O-Centered Radicals over Metal Oxide Clusters.

Abstract

Methane activation, a "holy grail" in chemistry, is crucial for producing value-added chemicals. Metal oxide clusters (MOCs) that activate methane through oxygen-centered radicals (O^•-) have been extensively studied. However, a systematic and quantitative understanding of the electronic factors that govern the reactivity of the O^•- radicals toward methane is still missing. Herein, a machine learning model has been developed to quantitatively describe the reactivity of MOCs toward CH₄ by incorporating 17 newly obtained experimental reaction rate constants alongside data accumulated from the literature, a total of 107 in number, as well as descriptors derived from density functional theory calculations. Utilizing the back propagation artificial neural network algorithm, the model described with only two key features-unpaired spin density (UPSD) and local charge (Q_L)-is capable of predicting CH₄ activation reactivity of O^•- containing MOCs across a wide range of metal elements and cluster compositions. Further investigations indicate that a feature related to the detachment or attachment of electrons can replace Q_L while UPSD is irreplaceable. By using artificial intelligence, this study has made a big step forward in understanding methane activation by reactive oxygen species.