Deep Learning Guided Exploration of Transition Metal Oxide Catalysts in Acetylene Selective Hydrogenation

The development of empirical materials, hindered by the complex interplay between material properties and reaction mechanisms, typically necessitates an extensive trial-and-error process to identify optimal catalysts. In our study, we integrate density functional theory (DFT) predictions to generate mappings of electronic and molecular adsorption properties, which are then employed to identify novel materials using deep learning algorithms. These newly identified materials were synthesized and assessed for their catalytic performance in the selective hydrogenation of acetylene. Notably, CuTiO₃ catalysts demonstrated exceptional performance, achieving acetylene conversion exceeding 99% and ethylene selectivity greater than 99% at a relatively low temperature of 75 °C. Additionally, CuO-doped TiO₂ was observed to form strong acetylene adsorption sites and weaker ethylene adsorption sites (Cu-O-Ti). The p-π hybridized coupling between the oxygen p-orbitals and the π-electrons of acetylene in the Cu-O-Ti structure was found to play a critical role in facilitating the conversion of acetylene to ethylene.