(580h) Deep Reaction Network of Glucose Pyrolysis: Discovering Unexpected Intermediates and Reaction Mechanisms

Predicting the reaction mechanism of biomass pyrolysis plays a central role in understanding the selectivity of each product and the evolution of possible intermediates, and ultimately aids in catalyst design. However, even the relatively simple model system, glucose monomer pyrolysis, has significant complexity in terms of the depth of the reaction network and the number of possible reactions per intermediate, which forces contemporary computational studies to rely on sampling heuristics. Here, we show how automated reaction prediction methods can be applied to construct a complex reaction network with limited chemical heuristics. Using graph-based rules to explore the reaction network and a modified Dijkstra algorithm to control the search space, the entire reaction exploration involves over 31,000 reactions computed at the low-cost semi-empirical quantum chemistry level and approximately 7,000 kinetically favored reactions optimized at the DFT level. The resulting network is the largest in the biomass pyrolysis fields. The automated algorithm (re)discovers all major products observed in the experimental work, as well as many intermediates proposed by previous computational studies, while also discovering many unintuitive reaction mechanisms that lead to much lower activation barriers. This comprehensive network also provides explanatory pathways for the high yield of hydroxymethylfurfural (HMF) and the reaction pathway that contributes most to the formation of hydroxyacetaldehyde (HAA). Due to the limited domain knowledge required to generate this network, the automated reaction prediction approach developed in the study can also be transferred to other complex reaction network prediction problems.