Liquid viscosity oriented parameterization of the Mie potential for reliable predictions of normal alkanes and alkylbenzenes
DJ Carlson and NF Giles and WV Wilding and TA IV Knotts, FLUID PHASE EQUILIBRIA, 561, 113522 (2022).
Despite the significance of liquid viscosity in various industries, it is one of the most difficult properties to accurately predict. Molecular simulation has shown potential for accurate and consistent prediction of liquid viscosity, but this approach involves a staggering number of variables in the model (the mathematical description of atomic interactions), and the method (the procedures by which the system is initialized and propagated and the data analysis algorithms used to determine the liquid viscosity of the system from the simulation results). Each of these variables can greatly affect predictive capability, and the current literature has many recommendations that are often conflicting. In other words, it is difficult to properly select the approach needed to obtain reliable and reproducible values for compounds lacking experimental data or for conditions outside those used to develop the procedures. Additionally, force fields are typically parameterized using equilibrium thermodynamic properties rather than transport properties which compromises the ability of the model to accurately predict liquid viscosity. This work explores the hypothesis that standardizing the simulation method and data analysis procedures, and parameterizing a force field for simultaneous density and viscosity prediction results in a robust prediction method for viscosity for compounds outside of the training set. The compounds explored are those which are composed of primary and secondary alkyl carbon groups (-CH3 and -CH2-) as well as the aromatic carbon with and without hydrogen (-- CH- and -- C<). The approach outlines how new parameters for the Mie 16-6 potential were developed, and the results show that properly representing the repulsive part of the intermolecular interactions is crucial for accurate viscosity predictions. Care is taken to select both training and test sets of data to demonstrate transferability. The end results demonstrate that the technique can predict liquid viscosity to better than 10% error across a wide temperature range-a step change improvement over current prediction approaches.
Return to Publications page