Efficient bifunctional catalysts for the OER and ORR are vital for clean energy, but the best-known catalysts (Pt and Ir) are scarce and expensive. Nonprecious materials such as heteroatom-doped graphene (sp2-carbon) show promise due to high conductivity and stability, but exhaustively screening their many configurations via density functional theory (DFT) is prohibitive. To address this, we developed Padarth Khoj, a machine learning-powered GUI for catalyst screening. The GUI ingests DFT outputs (e.g., DOSCAR/CONTCAR), computes electronic descriptors (such as π-orbital occupancy and density of states of π-orbital at Fermi level), and uses SVR/MLR models to predict adsorption energies ΔGOH and ΔGO – ΔGOH. The trained SVR model achieved high accuracy (R2 ≈ 0.85–0.89 for the OER/ORR). The interface supports batch analysis and visualization, streamlining workflow. Using only 160 initial DFT calculations, Padarth Khoj predicted overpotentials for ∼8000 active sites (vs ∼24,080 by brute-force DFT). The tool identified 31 catalysts with predicted bifunctional activity, and DFT validation on 10 of these confirmed ∼85% of the predictions. This corresponds to a reduction in computational effort and time by several orders of magnitude, which allows one to explore very large catalyst spaces. Overall, this ML-driven pipeline drastically reduces the screening effort and supplies physical insights to aid the rational design of low-cost OER/ORR electrocatalysts for sustainable energy technologies.
Pathak et al. (Sun,) studied this question.