The formation mechanism of furfural in xylan pyrolysis: A machine learning study based on neural network potential

Abstract

Biomass fast pyrolysis has attracted extensive attention to sustainably producing energy, fuels, and chemicals. Xylan as a typical hemicellulose is one of the main components of biomass, and it can produce the value-added platform product furfural (FF) by fast pyrolysis. However, the formation mechanism of FF is still unclear. Herein, the stochastic surface walking reaction sampling (SSW-RS) method based on neural network (NN) potential, together with density functional theory (DFT) calculation, were employed to explore the FF formation mechanism in xylan pyrolysis. Xylobiose and xylopyranose were used as the model compounds for investigation, with the aim to find out new FF formation pathways and to reveal the effects of the glycosidic bond. It is found that the existence of glycosidic bonds promotes dehydration and inhibits ring-opening reactions, and depolymerization becomes easier after dehydration. d-Xylose is an important precursor for FF formation, and a new pathway from d-xylose decomposition to generate FF has been found, involving 2-ene-3-deoxyxylose and 1,2-dicarbonyl-3-ene-5-pentanol as key intermediates. Due to the glycosidic bonds, the anhydro xylopyranoses (AXPs) as the dehydration and depolymerization products play an important role in FF formation. Another new pathway involving 4-ene-4-deoxyxylose (a cyclic AXP with C4C5 bond) is confirmed favorable for FF generation.