(342q) Optimization of High-Throughput Photocatalyzed Bimetallic Nanoparticles Using Online Active Learning

Development of cost-effective industrial solutions for renewable hydrogen generation would would have other significant uses: hydrogen has applications in ammonia synthesis, iron and steel production, and the production of many specialty chemicals, including polymers and synthetic fuels. The key barrier is in catalysis technology: although the platinum-catalyzed hydrogen evolution reaction is one of the most active catalytic processes ever discovered, platinum is extremely rare, and no cheap, stable catalyst has been developed with similar activity. To address this need, we developed a high-throughput experimental workflow capable of assessing the activity of bimetallic nanoparticles for photocatalytic hydrogen evolution. Metal ions are injected into an aqueous solution containing DMSO, triethanolamine, and an Iridium-based photocatalyst, and a blue LED induces a reducing environment. This precipitates nanoparticles that start to evolve hydrogen gas, which is tracked in real time by a chemosensory tape. Approximately 1 wellplate can be run per day, with each well containing metal ions at varying concentrations, and hardware restrictions limit each day to 4 unique metals. A online active learning framework was then created to optimize this experimental space for bimetallic wells not containing platinum or palladium, aided by a database of hydrogen adsorption energies. Several promising combinations were discovered, and although further characterization of the nanoparticles is forthcoming, active site adsorption energies give some explanation of trends in activity through the Sabatier principle. This framework demonstrates the ability of active learning techniques combined with high-throughput experimentation to quickly explore complex spaces. Future work is expected to extend this analysis to similar chemical systems, including the oxidation of simple alcohols.