Accelerating high-throughput screening of hydrogen peroxide production via DFT and machine learning

Abstract

On-site production of hydrogen peroxide (H₂O₂) using electrochemical methods received considerable attention but was limited by the target product selectivity. In this work, an efficient approach was proposed to accelerate high-throughput screening of H₂O₂ by utilizing ΔG_*O and ΔG_*OOH as descriptors in a contour map. Using this method, 21 systems with high selectivity and activity for H₂O₂ production were screened from 150 different (nitrogen-doped) graphene-supported metal single atom catalysts. The constant-potential and kinetic analysis demonstrated that PdN4, PtN4, NiN4, and CuN4 all exhibited selectivity for the 2e⁻ ORR under both acidic and basic conditions, consistent with current experimental research findings. A compressed sensing data analytics approach further unveils the inherent correlation between the structure and catalytic performance, accurately predicting the descriptors, thereby significantly reducing the time and cost. This study enriches the understanding of ORR activity origins and advances rational SAC design for various catalytic reactions, leveraging a combination of DFT and machine learning.