Emma Kerr

This research utilizes active learning techniques to accelerate the discovery of high- performance electrocatalysts for green hydrogen production. By optimizing platinum- group-metal doping in iridium oxides, this work aims to significantly reduce iridium loading while enhancing catalytic activity, lowering resource barriers for proton exchange membrane (PEM) electrolyzers deployment.

Globally, we produce 100 Mt of hydrogen annually. Over 99% of this is produced from fossil fuels, resulting in 1 Gt of annual CO2 emissions. Green hydrogen produced via proton exchange membrane (PEM) water electrolysis is a cornerstone of the global transition toward a carbon-neutral economy. However, the oxygen evolution reaction (OER) at the anode remains a primary technical bottleneck, traditionally requiring expensive and scarce iridium-based catalysts. This work addresses the urgent need for more sustainable and cost-effective materials by employing a data-driven approach to catalyst design. Utilizing an active learning loop centered on gaussian process regression (GPR), this research iteratively predicts and experimentally validates the performance of various platinum-group-metal dopants within a rutile iridium oxide framework. Our methodology allows for the rapid exploration of an expansive compositional space, successfully identifying mixed-metal oxide catalysts that surpass traditional activity benchmarks while simultaneously reducing total iridium loading. By leveraging uncertainty quantification to prioritize experimental synthesis, this effort demonstrates how machine learning can drastically accelerate the development of next-generation energy materials, ultimately making clean hydrogen production more economically viable at scale.