Enhanced learning loop framework accelerates screening of bimetallic catalysts with high oxygen reduction properties in different coordination environments

Abstract

In this work, we used density functional theory and machine learning (DFT-ML) to predict and screen the first coordination environment (FCE) and second coordination environment (SCE) of A-doped (A = C, N, O, P, and S) coordination environments of FeMA4N2 (M = Fe, Co, Ni, Cu and Zn) and validated it to improve the oxygen reduction reaction (ORR) catalytic activity of FeMA4N2. To expedite the FCE screening, we constructed the Enhanced learning loop framework (EHLN) and ORR descriptors based on convolutional neural networks and transfer learning. EHLN has high accuracy (RMSE is 0.167 V), and high characterization factor R2 of 86%. Therefore, SCE modification of the optimal overpotential structure (FeFeONCN) of FCE was found to reduce the overpotential to 0.26 V after doping with N and S at sites 4 and 6′. This study brings more possibilities for ORR catalytic activity by tailoring the FCE/SCE of the FeMA4N2 structure.