Exploring the Intricacies of Glycerol Hydrodeoxygenation to Propanediol on Cu Catalysts: A Comprehensive Investigation with the Aid of Machine Learning Forcefield

Utilization of biomass to feedstock chemical relies on transforming hydroxyl containing molecules, as the hydroxyl group is found on the backbone of bio-molecules. For example, glycerol can undergo a hydro-deoxygenation reaction to produce propanediol, a valuable chemical precursor. This reaction captures the complexity and challenges of modelling surface-reactivity of flexible organic molecules in heterogenous catalysis, where surface intermediates can have many configurations. High computational costs of Density Functional Theory(DFT) restrict exhaustive exploration of the factorial reaction space, leading to having limited insights of the hydrodeoxygenation mechanism and hindering rational catalyst design. We deploy a Machine-Learned-Force-Field (MLFF) driven approach to elucidate the complex reaction network involved in the hydro-deoxygenation of glycerol on Cu(111). We present the bond cleavage activity ranking order for glycerol, and other intermediates and identify reaction pathways resulting in 1,2-PDO formation while highlighting indications of its higher selectivity over 1,3-PDO. This investigation delivers a comprehensive exploration of the transformation process from glycerol to propanediol, addressing the existing knowledge deficit through an advanced active-learning based MLFF approach. Notably, following a mere four iterations, our trained MLFF model accurately discerns 26 transition-states reconfirmed with DFT with root-mean-squared-error of 0.056 eV (0.74 meV/atom total-energy) embedded within network of seven competitive pathways.