Quantifying robustness of DFT predicted pathways and activity determining elementary steps for electrochemical reactions.

Density functional theory calculations are being routinely used to screen for new catalysts. Typically, this involves invoking scaling relations leading to the Sabatier-type volcano relationship for the catalytic activity, where each leg represents a unique potential determining an elementary step. The success of such screening efforts relies heavily not only on the prediction robustness of the activity determining step, but also on the choice of the descriptor. This becomes even more important as these methods are being applied to determine selectivity between a variety of possible reaction products. In this work, we develop a framework to quantify the confidence in the classification problem of identifying the potential determining step for material candidates and subsequently the pathway selectivity toward different reaction products. We define a quantity termed as the classification efficiency, which is a quantitative metric to rank descriptors on the basis of robustness of predictions for identifying selectivity toward different reaction products and the limiting step for the corresponding pathway. We demonstrate this approach for the reactions of oxygen reduction and oxygen evolution, and identify that ΔGOOH* is the optimal descriptor to classify between 2e- and 4e- oxygen reduction. We further show that ΔGOH* and ΔGOOH* have comparable performance in identifying the limiting step for 4e- oxygen reduction reaction. In the case of oxygen evolution, we study all possible 2 descriptor models and identify that {ΔGOOH*,ΔGO* } and {ΔGOH* ,ΔGO* } both are highly efficient at classifying between 2e- and 4e- water oxidation. The presented methodology can directly be applied to other multi-electron electrochemical reactions such as CO2 and N2 reduction for improved mechanistic insights.