Abstract: Advances in theory and methods of quantum mechanics (QM) are making it practical for first principles (de novo) predictions of electrocatalysis reactions including the full effects of solvent and as a function of potential (Grand Canonical QM). We will report results using these methods to extract Turn-over-Frequency and Tafel slope for CO2 reduction, oxygen evolution, hydrogen evolution, and oxygen reduction on well characterized surfaces.
However, the best electrocatalysts are often nanoparticles or nanowires of sizes 10nm and larger that may have 200K atoms and 10K surface sites, far too big for QM. We solve this problem by using the ReaxFF reactive force field to grow the catalyst, while retaining QM accuracy for the reaction barriers. But with 10K surface sites, examining each one is impractical. To solve this problem, we show how to use machine learning to predict all 10K surface sites with QM accuracy to identify the optimum sites for electrocatalysis.
