Solubility of selected polymers in cyclohexane: comparison between Flory-Huggins interaction parameters calculated...

Accurate prediction of solubility of polymers in solvents a priori is highly desirable in practice. To this end, the Flory-Huggins interaction parameter χ is commonly used and molecular dynamics simulation, a powerful computational tool, has been used for such a purpose. To calculate χ, there exist three possible strategies using molecular dynamics simulation. One is through the calculation of Hildebrand solubility parameters of the pure components while the other two are to calculate the enthalpy of solvation and Gibbs free energy of solvation for the solution, respectively. This study evaluated these three strategies using binary solutions containing a hydrophobic or hydrophilic polymer (polyisobutylene, polystyrene, cis and trans polybutadiene, cis and trans polyisoprene, poly(ethylene oxide), and polyacrylamide) and an aliphatic solvent - cyclohexane. We found that χ determined via solubility parameters predicted the solubility trend but deviated significantly from experimental values. On the other hand, the enthalpy of solvation approach provided the most accurate χ values, compared to experiment, at a reasonable computational demand, especially for hydrocarbon polymers, while the Gibbs free energy of solvation approach, though more computationally intensive, did not significantly improve χ from the enthalpy of solvation approach. In particular, the Gibbs free energy of solvation approach overestimated χ for non-polar polymers. A conformational analysis of the solvated polymers revealed that all polymers collapsed in cyclohexane with polyethylene oxide and polyacrylamide collapsed the most as expected. For the two polar polymers used, the collapse was evidenced by abrupt changes in radius of gyration (R_g) and solvent accessible surface area (SASA) in the early stage of molecular dynamics simulation trajectories, and plateauing at much lower final values. Conversely, the hydrocarbon polymers exhibited minimal deviation from the expected R_g and barely any change in SASA with time. Our findings demonstrated that there exist differences in the accuracy and computational resources used when different molecular dynamics simulation strategies are used in the determination of χ.