Understanding the role of metal oxide support in ruthenium-based catalysts for ammonia synthesis via interpretable machine learning

The role of metal oxide supports is complex in heterogeneous catalysis due to acidity, basicity, and surface defects. Interpretable machine learning models trained on experimental data could lead to new insights about these complexities, which are rarely verified through detailed catalyst characterization. This study explores the role of metal oxide supports for the Ru-based ammonia synthesis catalysts using Shapley additive explanations (SHAP). The support metal nitride formation energy and the support metal hydride formation energy were identified as critical descriptors that could describe the ammonia synthesis activity. These descriptors and the related catalyst characterization by Ammonia-Temperature Programmed Desorption and Hydrogen-Temperature Programmed Desorption suggest new processes that could govern the ammonia synthesis reaction. It is suggested that in addition to basicity, the metal oxide support should possess a certain acidity to alleviate ammonia inhibition and form metal hydrides to alleviate hydrogen inhibition.