Machine Learning–Accelerated Discovery of Earth-Abundant Bimetallic Electrocatalysts for the Hydrogen Evolution Reaction

Despite platinum's well-established catalytic activity for the hydrogen evolution reaction (HER), its limited supply and steep cost hinder large-scale adoption. Earth-abundant bimetallic alloys thus emerge as attractive substitutes, though their vast compositional and structural diversity makes exhaustive density functional theory (DFT) screening unfeasible. Here, we introduce a machine learning (ML)-DFT workflow for the discovery and prioritization of bimetallic HER catalysts. By integrating EquiformerV2 into the AdsorbML surrogate-DFT pipeline, we efficiently predict hydrogen adsorption energies on thousands of alloy surfaces. Sabatier-volcano filtering combined with targeted DFT validation yields a mean absolute error of 0.12 eV across the screened space. Two surface motifs stand out: (i) transition-metal dimers or isolated top sites embedded in Sn- or Sb-rich layers, and (ii) Cu-rich surfaces (Cu-Sn, Cu-Sb) featuring Cu-Cu bridge or hollow sites without direct Sn or Sb interaction. A multiobjective assessment of activity, stability, and cost highlights four synthesis-ready candidatesFe₂Sb₄, Cu₆Sb₂, Cu₆Sn₂, and Ni₂Sb₂which combine platinum-like performance with significantly lower material costs. This integrated ML-DFT strategy transforms an otherwise intractable chemical landscape into a concise, experimental roadmap for earth-abundant HER catalyst development.