Computational design of water splitting electrocatalysts: From d-band center to machine learning descriptors

Electrochemical water splitting is critical for sustainable hydrogen production, yet its commercial viability is limited by the sluggish kinetics of the hydrogen and oxygen evolution reactions (HER and OER). Computational chemistry has accelerated catalyst discovery through predictive activity descriptors. However, early descriptors are predominantly single-valued and static, built upon idealized models that omit dynamic surface reconstruction, electrolyte effects, and the coupling between multiple reaction pathways. These oversimplifications impair their capacity to break intrinsic scaling relationships and capture operando conditions, which has motivated a paradigm shift toward multidimensional, time-resolved frameworks. This review tracks the evolution of descriptors, from conventional volcano plots to machine learning-driven dynamic descriptors, and elaborates on how modern approaches capture structural evolution under operating potentials, bypass scaling constraints, and align theoretical model systems with realistic solid-liquid interfaces. Finally, we discuss prospective research directions, highlighting the integration of high-precision dynamic modeling, multi-objective optimization (which balances activity, stability, and mass transport), and closed-loop automated platforms to facilitate the on-demand design of robust electrocatalysts for industrial applications.