A Critical Evaluation on the Performance of COSMO-SAC Models for Vapor-Liquid and Liquid-Liquid Equilibrium Predictions Based on Different Quantum Chemical Calculations

The performance of two versions of the COSMO-SAC activity coefficient model is carefully examined based on eight sets of quantum chemical computations [VWN-BP/DNP, b3lyp/6-31G(d,p), b3lyp/6-31G(2d,p), b3lyp/6-31+G(d,p), b3lyp/6-311G(d,p), wb97xd/6-31G(d,p), wb97xd/6-31G(2d,p), and wb97xd/6-31+G(d,p)] and one semiempirical calculation (PM6). Furthermore, the effect of the molecular geometry is examined based on equilibrium structures determined both in a vacuum, representing a nonpolar environment, and in a conductor, representing a highly polar environment. The model parameters are reoptimized for each quantum chemical calculation method, and the performance is evaluated using a large set of databases covering the vapor-liquid equilibrium, liquid-liquid equilibrium, infinite-dilution activity coefficient of binary mixtures, and octanol-water partition coefficient (K_ow; containing over 22000 data points). It is found that the original COSMO-SAC model is sensitive to the quantum chemical method used, whereas the revised COSMO-SAC model is not. For the original COSMO-SAC, a method that gives higher molecular polarity often results in a better prediction accuracy. The modifications introduced in the revised COSMO-SAC model not only improve the accuracy but also allow for the use of a lower-quality quantum computational theory without much loss of accuracy.