Gaussian Process-Supported Optimization of the Transferable Anisotropic Mie Potential Force Field for Primary Alkylamines

Abstract

This study extends the transferable anisotropic Mie potential (TAMie) to primary alkylamines. The force field parameters are fitted by minimizing the squared deviations of vapor pressures and liquid densities from experimental data of n-propylamine and n-butylamine. The physically based PC-SAFT equation of state was used as a surrogate model to approximate simulation results and additionally utilized in a linear multifidelity Gaussian process model extrapolating to temperatures that are difficult to simulate directly with Monte Carlo simulations. In the parameter optimization procedure, the deviation between simulation and experiment was approximated with Gaussian processes to determine those force field parameters, which cannot be mapped to the PC-SAFT representation. For propylamine and butylamine, the optimized force field leads to mean absolute relative deviations of 0.55% and 0.59% for liquid densities and 0.95% and 0.34% for vapor pressures in the range of reduced temperature 0.55 ≤ T/T_c ≤ 0.85 and 0.55 ≤ T/T_c ≤ 0.9. The transferability to higher alkylamines was tested for n-hexylamine to n-octylamine. Individual parameter sets are provided for methylamine and ethylamine. Dynamic properties such as the shear viscosity of pure substances are predicted in fair agreement with experimental data, even though no dynamic property was included in the parametrization. The phase behavior of binary mixtures of primary amines with alkanes is investigated, and the predictions of the TAMie model are found to be in very good agreement with experimental data.