Identification of ORR activity of random graphene-based systems using the general descriptor and predictive model equation

Abstract

Carbon based electrocatalysts are well known promising candidates for the oxygen reduction reaction (ORR), but the random approach to find the best catalyst using experimental method delayed the screening process and it leads to a huge cost with less proficiency. Using Quantum Mechanics followed by Machine Learning (QM/ML) approach, we can predict the best catalyst faster way and the origin of the cause can be identified for further development of carbon-based catalyst. Using the π electronic descriptor unveiled using density functional theory, we applied the analytical simple fit method and six different machine learning algorithms to develop a highly effective predictive model to estimate ΔGOH. Furthermore, structural relations of ZGNR and AGNR are demonstrated to estimate the Dπ(EF), R-Oπ, and ΔGOH of different widths of ribbon that reduces additional DFT calculations. By applying both SVR predictive model and structural relations, we predicted the ORR performance of 2500 sites of GNRs and listed a few most ideal active carbon sites with lower overpotential (η < 0.5V). To validate our study, we predicted the ORR performance of different sites in 0D, 1D, 2D doped graphene systems using SVR model and confirmed the values with the DFT computed results.