Hydrogen peroxide is widely used in everyday life, from disinfectants and medical sterilization to environmental cleanup and manufacturing. Despite its importance, most hydrogen peroxide is still produced using large-scale industrial processes that require significant energy. Researchers are thus seeking cleaner alternatives.

A team of researchers has made a breakthrough in this regard, developing a new computational framework that helps identify effective catalysts for producing hydrogen peroxide directly from water and electricity. The work focuses on the two-electron water oxidation reaction, an electrochemical process that can generate hydrogen peroxide in a more localized and potentially sustainable way.

This was no walk in the park, as Hao Li, the lead-author of the research, outlines. “Designing catalysts for this reaction has been difficult because catalysts come in many forms, such as metal alloys, metal oxides, and single-atom materials. Each type has different atomic structures, making it challenging to compare or predict their performance using a single method.

To address this problem, Li and his team developed a new way to describe catalytic active sites at the atomic level. This approach, called a weighted atom-centered symmetry function, captures both the geometric arrangement of atoms and their chemical identities in a unified format. These descriptors were combined with machine learning models and reaction modeling to predict how well different materials would perform.

Using this framework, the team successfully predicted key reaction properties across a wide range of catalyst types. The predictions closely matched results from detailed quantum-mechanical calculations and previously reported experimental data, showing that the approach can work across diverse materials.

The researchers then used the model to rapidly screen potential catalysts and identified lithium scandium oxide (LiScO₂) as a promising candidate. Experiments confirmed that this material can produce hydrogen peroxide with about 90% efficiency and remain stable for nearly one week of continuous operation.

“This framework allows us to connect atomic-scale information directly to measurable performance,” adds Li. “It helps reduce trial-and-error in catalyst development and makes the search process more systematic.”

The framework has been implemented in the Digital Catalysis Platform (the largest experimental + computational catalysis database to date with digital platform for users, developed by the Hao Li Lab), where it can be used to predict reaction properties efficiently. Because the method treats different material classes in a consistent way, it can be extended beyond hydrogen peroxide production.

The researchers expect that this approach will support the design of catalysts for other important electrochemical reactions, contributing to cleaner chemical production and energy technologies in the future.