Leveraging Machine learning and active motifs-based catalyst design for discovery of oxygen reduction electrocatalysts for hydrogen peroxide production

As the demand for hydrogen peroxide (H₂O₂) increases across various industries, there is a growing need for eco-friendly production process to replace the energy-intensive and environmentally polluting anthraquinone process. In particular, the electrochemical production of H₂O₂ via the two-electron oxygen reduction reaction (2e-ORR) is being highlighted as a promising alternative. However, achieving high selectivity for 2e-ORR over the four-electron reduction reaction (4e-ORR), remains challenging. We introduce an integrative strategy that combines active motifs-based design with a machine learning to discover promising catalysts for electrochemical H₂O₂ production. Inspired by single-site alloys that destabilize the binding strength of O* adsorbate, thereby improving the 2e-ORR selectivity, we expanded the chemical space through elemental substitution and efficiently explored this expanded chemical space using machine learning methods. By employing these approaches, we discovered active, selective and stable 2e-ORR catalysts that are not present in the existing database and demonstrated better stability compared to the materials within the database. This work highlights the potential of integrating active motifs-based catalyst design with machine learning to efficiently explore the vast chemical space, accelerating the discovery of novel catalysts.