dc.description.abstract | In this study, a novel approach combining quantum mechanical solvation free energy calculations and Peng-Robinson equations of state (PR EOS) is proposed for the prediction of phase equilibria of both pure and mixture fluids. For pure substances the critical properties and acentric factor must be used to determine the energy a(T) and volume b parameters. An appropriate mixing rule and often the binary interaction parameters are necessary in order to have a better description of the phase behavior for mixtures. The application of PR EOS is therefore limited by the need of input of experimental data. In this study, we found that the temperature and composition dependence of the energy parameter a(T,x) in the EOS can be derived from the attractive contribution of the solvation free energy. The volume parameter b(x) is estimated to be the mole-fraction weighted average of the molecular solvation cavity. Combined with first-principle solvation calculations, both parameters a(T,x) and b(x) can be obtained without the use of any experimental data (e.g., critical properties or acentric factor) and binary interaction parameters. The Peng-Robinson EOS combined with a solvation model based on COSMO-SAC calculation, denoted as PR+COSMOSAC, contains neither species dependent parameter nor binary interaction parameters, and can be used to predict vapor pressure, liquid density, and critical properties of pure substances, and vapor-liquid equilibrium (VLE), liquid-liquid equilibrium (LLE), and vapor-liquid-liquid equilibrium (VLLE) of mixtures. It is found that the overall relative average error from PR+COSMOSAC is 49% in vapor pressure, 21% in liquid density at normal boiling point, 10% in critical pressure, 4% in critical temperature, and 5% in critical volume for 1296 pure substances; and 28% in total pressure, and 5% in vapor phase composition for 230 binary mixtures in vapor-liquid equilibrium. The errors in binary VLE predictions can be reduced significantly down to 6% and 2% if experimental data (vapor pressures or critical properties and acentric factor) are used to correct for any error in the calculated charging free energy of pure species. The overall root-mean-square errors in the mutual solubility of 68 binary and 39 ternary mixtures predicted from PR+COSMOSAC are 0.0689 (80%) and 0.0775 (72%), respectively. This method provides the prediction of LLE and VLLE with accuracy similar to that from the widely used group contribution method, the modified UNIFAC model. The overall RMS errors of PR+COSMOSAC in drug solubility prediction of 52 drugs in 37 pure solvents and 156 mixture solvents are 1.78 (495%) and 1.40 (304%), respectively. This accuracy is better than that from COSMO-SAC model. The overall RMS of drug solubility prediction in mixture solvents can be greatly reduced to 0.65 (91%) when the experimental solubility of the drug in the relevant pure solvent is available. Since there is no issue of missing parameters or group definitions, this model is particular useful for the design of new processes involving chemicals whose interaction parameters are not available due to the lack of experimental data. | en |