Machine Learning-Assisted Design of Transition-Metal-Doped Cobalt Phosphide Electrocatalysts for Hydrogen Evolution

Electrochemical water splitting has emerged as a promising approach for sustainable hydrogen production, yet it faces several challenges such as low efficiency, high overpotential, and instability. Efficient electrocatalysts play a crucial role in enhancing hydrogen evolution reaction (HER) activity by lowering activation energy and optimizing Gibbs free energy for hydrogen in promising HER electrocatalysts. Cobalt phosphide (CoP) stands out among various nanomaterials due to its wide range of active sites and pH applicability; however, the rational screening and prediction of highly efficient transition metal doping CoP electrocatalysts is still a challenge. Herein, we highlight the rational design of transition-metal-doped CoP HER electrocatalysts by integrating machine learning (ML) with density functional theory. A model was developed for transition-metal-doped CoP electrocatalysts with 29 different transition metals, and the Gibbs free energy of hydrogen adsorption (ΔG_H*) was calculated using the generalized gradient approximation method, which served as the data set for ML. Extreme gradient boosting model shows a mean absolute error of 0.079 eV and a higher coefficient of determination of 0.931 for predicting ΔG_H*. The significant features impacting ΔG_H* are doping sites, electronic transfer, and first ionization energy. This work provides effective insights into the reverse design and development of transition-metal-doped CoP electrocatalysts.