Rationalizing the use of mutual prediction models in non-ideal binary mixtures

In this paper, we compared seven diffusion models in terms of prediction performances. Using vapour-liquid equilibrium (VLE) data, we calculate the thermodynamic correction factor as a function of composition for eleven binary liquid mixtures using non-random two-liquid and Redlich-Kister models. These data, together with intra-diffusion coefficients, and viscosity values are used to predict mutual diffusivity. The Darken-based models, which consider a scaling power on the thermodynamic factor, give accurate predictions, with absolute average relative deviation (AARD) values between 1 and 20 %. The removal of the scaling power leads to a decrease in prediction accuracy. The viscosity-based models with (Vis-SF) and without (Vis-nSF) scaling factor have AARD of 14 and 30 %, respectively. The dimerization model is inaccurate for most mixtures except those containing water, while the Vignes-based model (V-Gex), which is based on the Gibbs free energy, gave high AARD values of 25 %, hence, not as reliable when compared to the other models.