Machine-learning-assisted catalytic performance predictions of binary alloy catalysts for glucose hydrogenation

Abstract

Sorbitol production involves glucose hydrogenation, which requires co-adsorption of glucose and hydrogen on the catalyst surface for high catalytic performance. Binary alloy catalysts can modulate substrate adsorption to achieve better catalytic performance than Raney Ni. Herein, we established a pioneering DFT/ML approach to investigate the adsorption energies of glucose (ΔE_GCHO) and H atoms (ΔE_H) on 1155 binary alloy catalysts. The Light Gradient Boosting Machine (LGBM) algorithm proved the most effective ML model, predicting ΔE_GCHO and ΔE_H with R² values of 0.785 and 0.636, respectively. Microkinetic simulation demonstrated a correlation between catalytic activity and adsorption energy, revealing high-performance catalyst screening criteria as ΔE_GCHO = −1.45 to −0.65 eV and ΔE_H = −0.55–0.00 eV. Nine possible binary alloy catalysts with high predicted activity were identified, with Pd₃Mg performing best. The present study highlights the potential of the DFT/ML-assisted approach in the development of efficient glucose hydrogenation binary alloy catalysts.