Benchmarking Different QM Levels for Usage with COSMO-RS

The COSMO-RS method is an established method for the prediction of fluid phase properties such as activity coefficients, liquid-liquid equilibria, and free energy of solvation. It is also frequently used in quantum chemistry-based chemical reaction modeling to predict the solvation contribution to the reactions. The COSMOtherm software, which features the currently most advanced implementation of COSMO-RS, is based on quantum chemical COSMO calculations using the BP functional with the def2-TZVPD basis set. As the accuracy of COSMO-RS depends on the accuracy of the underlying quantum chemical (QC) calculation, it is important to validate the currently used level against other common, presumably superior, approaches such as the more recently developed M06-2x hybrid density functional or wave function methods such as MP2. As compared to other applications where the electronic energy is the most important result of the QC calculation, the COSMO-RS method has a much higher dependence on the molecular polarity and thus the electron density distribution. We find that MP2, PBE0, and M06-2x perform slightly worse in fully reparametrized COSMO-RS with respect to the prediction of experimentally measured properties like pK_a or logP. Although MP2 was reported to yield better polarities than most DFT functionals for spin unpolarized molecules, this theoretical advantage does not manifest in a practical benefit for the prediction of thermodynamic properties with a refitted COSMO-RS parameter set. Other pure DFT functionals such as PBE or TPSS can be used instead of BP, but again, no practical advantage is expected as they yield extremely similar polarities to the original BP calculations.