Molecular Models for the Prediction of Thermophysical Properties of Pure Fluids and Mixtures

Accurate thermophysical property data are essential for process design, but experimental databases are often limited, so reliable predictive methods are needed. Conventional equations of state and activity-coefficient (GE) models are effective for correlation but have limited predictive power. This work demonstrates molecular modeling and simulation as an alternative approach for pure fluids and mixtures with strong predictive capability. Molecular models for 78 pure substances (e.g., N₂, O₂, ethane, CO, CO₂, halogens, and alternative refrigerants) are developed using two-center Lennard–Jones models with either a point quadrupole (2CLJQ) or point dipole (2CLJD). A new parametrization strategy based on global correlations of vapor–liquid equilibrium (VLE) for these model fluids enables efficient model development, with required VLE data generated by molecular simulation. The resulting models typically match experiments within about 0.5% for saturated liquid densities, ~4% for vapor pressures, and ~3% for enthalpies of vaporization, outperforming many literature models and remaining accurate in homogeneous states far from VLE. More detailed Lennard–Jones–based models are additionally developed for ethylene oxide and methanol. The compatible pure-fluid models are then applied to 45 binary mixtures, first using purely predictive Lorentz–Berthelot combining rules, which already yield more reliable VLE predictions than the Peng–Robinson equation of state. Quantitative agreement for real mixtures is achieved by introducing a single fitted binary interaction parameter per mixture, determined from one experimental equilibrium pressure. Finally, these binary interaction models are used to predict VLE for five ternary mixtures, reproducing vapor compositions and saturated densities more accurately than Peng–Robinson. Overall, the developed molecular model set provides a broadly applicable and improvable foundation for predictive thermophysical-property modeling.