Rational Design of Pt3M (M = Transition Metals) Alloy Catalyst for Hydrogen Production via Ethylene Glycol Decomposition: Combined Machine Learning, Density Functional Theory, and Experimental Approach

Abstract

The conversion of ethylene glycol (EG) to hydrogen is a critical process for advancing sustainable energy solutions. This study combines machine learning, density functional theory (DFT) simulations, and experimental methods to design and validate a highly efficient Pt3Sc alloy catalyst for hydrogen production. First, the supervised deep neural network model with the 38-dimensional feature vector of catalyst properties and the label of activity descriptor (binding energy of C2H5O2), trained on data sets generated using DFT, was employed to search the alloy catalysts, leading to the identification of the Pt3Zr, Pt3Hf, Pt3Sc, Pt3Ta, Pt3Ti, and Pt3Nb candidates. We further calculated the DFT-based free energy diagram of ML-searched candidates for EG decomposition, which was filtered to the Pt3Sc alloy. Next, DFT-based microkinetic modeling (analyzing 92 elementary reactions) was performed to confirm the enhanced activity of the Pt3Sc(111) surface, which revealed that the Pt3Sc(111) catalyst exhibited nearly triple the activity (turn-of-frequency) of Pt(111) in hydrogen production at a temperature of 500 K. Apparent activation energies were predicted from Arrhenius plots (obtained from microkinetic modeling), yielding a lower energy barrier by 0.21 eV (high-temperature region) and by 0.05 eV (low-temperature region) compared to the Pt(111) case. Finally, experimental validation further demonstrated the superior performance of the Pt3Sc/SiO2 catalyst, which achieved 100% hydrogen selectivity and a higher H2 production rate than Pt/SiO2 during the aqueous phase reforming (APR) of EG. This research marks a significant step forward in hydrogen production technology by integrating data-driven, theoretical, and experimental approaches to identify Pt3Sc as a promising catalyst for hydrogen energy solutions.