(521er) Machine Learning Investigation of Mixed Oxide Supports for Ammonia Synthesis Catalysts

Since liquid ammonia is a promising hydrogen carrier, interest is growing in novel heterogeneous catalysts for ammonia synthesis under reaction conditions milder than the Haber-Bosch process, which uses an iron catalyst. Supported ruthenium (Ru) nanoparticles have emerged as an alternative catalyst but suffer from hydrogen poisoning, preventing higher ammonia synthesis rates at higher pressures. Supports that can facilitate hydrogen spillover and strong metal support interactions (SMSI) can alleviate hydrogen poisoning. However, state-of-the-art supports such as nitrides, hydrides, and electrides prevent them from large-scale use due to air and moisture sensitivity and complex synthesis processes. Mixed metal oxide supports containing lanthanides such as BaCeO₃, La₂Ce₂O_7, and La₂Pr₂O₇ are promising due to their higher activity, stability, and ease of synthesis. Large amounts of lanthanide-containing mixed metal oxide supports remain unexplored in the literature.

In this work, we use machine learning (ML) combined with experimental validation to investigate novel oxide supports for Ru nanoparticles. The ML model is developed with catalyst activity data for unpromoted Ru on metal oxide supports mined from literature. Descriptors are engineered and selected for the model which represents catalytic properties of the support (basicity, reducibility, etc.) and support surface processes (SMSI, hydrogen spillover, etc.). Model interpretations led to insights about support properties that lead to higher ammonia synthesis activity. The model is then used to predict the activity of catalysts with all lanthanide metal-containing oxide supports extracted from the materials project database and filtered based on thermodynamic stability and synthesizability. The predicted oxide supports with higher activities are experimentally synthesized and tested for model validation which led to the discovery of novel oxide supports for ammonia synthesis. Selected novel supports are characterized for basicity and reducibility to verify the model interpretations.