Iterative machine learning method for screening high-performance catalysts for H2O2 production

We have proposed an iterative machine learning method with spatial coordinate information to accelerate the screen of high-performance single-atom catalysts for H2O2 production. Furthermore, a simple descriptor is proposed to effectively evaluate the performance of H2O2 production on single-atom catalysts. [Display omitted] •120 SACs (6 configurations) are designed and evaluated upon calculations.•The RhN2O2A exhibits an optimal ΔG∗OOHΔG∗OOH(4.233 eV) and ultra-low ηη(0.013 V).•∑i=1nNmfi-Nmm+Nd•Iterative machine learning method is proposed to rapidly screen idea catalyst. Electrochemical H2O2 production through a selective two-electron oxygen reduction reaction represents a promising alternative to the traditional anthraquinone oxidation process (AOP). However, it remains an unavoidable challenge to screen high-performance catalysts for the H2O2 production such as ambiguous activation mechanisms. Herein, we propose an iterative machine learning (iML) method that drastically reduces the required training set size. By introducing the feature of spatial coordinate information,we can rapidly screen out the optimal catalytic activity from hundreds of single-atom catalysts. It can be found that RhO2N2(A) is an ideal catalyst for H2O2production with an ultra-low overpotential of 0.013 V. The work sheds light on the path to accelerate the data-driven design and discovery of high-performance catalysts.