Understanding activity origin for the oxygen reduction reaction on bi-atom catalysts by DFT studies and machine-learning

Abstract

Bi-atom catalysts (BACs) have attracted increasing attention in important electrocatalytic reactions such as the oxygen reduction reaction (ORR). Here, by means of density functional theory simulations coupled with machine-learning technology, we explored the structure–property correlation and catalytic activity origin of BACs, where metal dimers are coordinated by N-doped graphene (NC). We first sampled 26 homonuclear (M₂/NC) BACs and constructed the activity volcano curve. Disappointingly, only one BAC, namely Co₂/NC, exhibits promising ORR activity, leaving considerable room for enhancement in ORR performance. Then, we extended our study to 55 heteronuclear BACs (M₁M₂/NC) and found that 8 BACs possess competitive or superior ORR activity compared with the Pt(111) benchmark catalyst. Specifically, CoNi/NC shows the most optimal activity with a very high limiting potential of 0.88 V. The linear scaling relationships among the adsorption free energy of *OOH, *O and *OH species are significantly weakened on BACs as compared to a transition metal surface, indicating that it is difficult to precisely describe the catalytic activity with only one descriptor. Thus, we adopted machine-learning techniques to identify the activity origin for the ORR on BACs, which is mainly governed by simple geometric parameters. Our work not only identifies promising BACs yet unexplored in the experiment, but also provides useful guidelines for the development of novel and highly efficient ORR catalysts.