Improving the Description of Adsorbed Hydroxyl to Make Predictive Catalytic Activity Models

Artificial intelligence may take catalytic models to a level that matches reality only when provided with high-quality data. Currently, activity plots based on the Sabatier principle are among the most extensively used tools to design materials in thermocatalysis and electrocatalysis. While salient examples attest to their usefulness, activity plots based on density functional theory calculations often fail for various reasons, one of them being the inaccurate approximation of the exchange-correlation energy for free molecules, catalysts, and adsorbed species. Here, we combine experiments and computational modeling to assess the errors in adsorbed hydroxyl (*OH), which are otherwise overly difficult to assess. Upon correcting a systematic adsorbed-phase error in *OH, our predictions match the experimental results for ethylene oxide electroreduction on transition-metal electrodes. The transferability of *OH corrections is illustrated by enhancing activity estimations of CO electrooxidation, where *OH is a key intermediate. Finally, we provide an alternative to cases in which the assessment of the adsorbed-phase errors is unfeasible. In summary, we show here how to improve electrocatalytic estimations by depurating their input data.