Proposing explainable descriptors towards enhanced N ₂ reduction performance on the two-dimensional bismuthine nanosheets modified by p-block element-based electrocatalysts

p-Block element-based electrocatalysts featuring a tunable electronic structure to achieve exceptional N₂ activation and proton suppression have garnered extensive interest for the electrochemical N₂ reduction reaction (NRR). Albeit various reaction mechanisms were proposed to understand and optimize the NRR performance, methods to effectively design and rapidly screen potential candidates are still elusive. Herein, a couple of explicit and interpretable descriptors on the entire p-block element-based electrocatalysts are put forward to predict NRR activity and selectivity <i>via</i> high-throughput theoretical simulations and a symbolic regression algorithm, taking two-dimensional (2D) bismuthine with p-block elements doped in or adsorbed as an example. The descriptors are merely composed of inherent atomic properties (p orbital electron number, electron affinity, electronegativity, atomic radius, <i>etc.</i>) combined with algebraic operators, independent of the intricate DFT calculations. Multi-task regression results demonstrate that the doped and adsorbed bismuthine systems possess the same descriptors, namely, the descriptors of doped-Bi can accurately forecast the NRR performance of adsorbed-Bi, and <i>vice versa</i>. Five potential candidates (5/40) with outstanding NRR activity, selectivity and stability are screened. C-doped and Si-doped bismuthine possess less negative limiting potentials of the NRR [<i>U</i>_L(NRR)] of -0.46 and -0.68 V and positive [<i>U</i>_L(NRR) - <i>U</i>_L(HER)] values of 1.15 and 0.13 V, respectively, superior to those of the majority of reported p-block element-based electrocatalysts, which are expected to be verified by the experimental research. This work offers a feasible solution for developing promising electrocatalysts for the NRR and potentially other electrochemical reactions on the basis of explainable descriptors using geometric information and intrinsic atomic quantities.