Abstract
Cellulose is a very important renewable resource because of its abundance in nature. However, it is known to have poor solubility and processability owing to its high crystallinity. In the present study, molecular dynamics (MD) simulations and machine learning (ML) were used to design solvents with high cellulose dissolving power by gaining deep mechanistic insights into the dissolution mechanism. High-throughput MD simulations were performed to evaluate the number of hydrogen bonds in cellulose crystals after dissolution, as the cellulose solubility, in various imidazolium-based ionic liquids (3,200 molecules) generated by ML models. Furthermore, the cation–anion distribution in the ionic liquid and the interaction energy between the cellulose molecular chains in the crystal after dissolution were used as metrics to characterize the cellulose solubility to improve the prediction performance of ML models. Several functional groups (methyl, ethyl, allyl, chloro, fluoro, and methoxymethyl groups) on the side chain of the imidazolium cation were identified to be effective for increasing the cellulose solubility, and a new series of powerful cellulose solvents is proposed.