Molecular models for O2 and N2 from the second virial coefficient

Abstract

• O 2 and N 2 models were generated in gas phase and by rescaling the energy the interfacial properties were obtained. • The convenience to reproduce the N 2 and O 2 properties using the 2cLJQ and 2cANCQ fluids were analyzed. • We implement algorithms with HyperOpt and MPI4Py libraries to optimize the search in the parameter domain. In this work we take up the molecular models generation for O 2 and N 2 . This aim is achieved with the two center Lennard-Jones and two center ANC potentials, plus quadrupole interactions. We analyze the convenience to perform the fitting of potential functions to the experimental data of the second virial coefficient and then rescaling the energy well depth to simulate properties in the liquid phase: surface tension, liquid vapor equilibrium and vapour pressure. The main objective is to study the following: 1) To analyze if due to the fact that the ANC interaction sites have a greater number of parameters their results enhance these one obtained from the LJ interaction sites in the gas phase, and 2) to establish whether the effect of the interaction of three bodies, included through rescaling the energy well depth, is enough to get that the results of the ANC interaction sites enhance the results obtained with the LJ interaction sites in the liquid phase. Although the first objective was fulfilled satisfactorily, that is, the results with the ANC interaction sites reproduce better the experimental second virial coefficient, for the second objective, both types of pairwise interaction lead to similar results of properties in the liquid phase. As a consequence of this similarity, it is deduced that the objective function used to optimize the two center ANC leads to optimal parameters in accordance with the N 2 and O 2 models of the two center Lennard-Jones. Therefore, rescaling the depth of the energy well is not enough to overcome the results found with the two center Lennard-Jones in the liquid phase. It is worth mentioning that for reduce the computational cost, to obtain the best fitting curves in the gas phase, bayessian optimization were performed using the Python Hyperopt library. In addition, the O 2 and N 2 molecular models were also evaluated by transferring the parameters to the cross second virial coefficient for O 2 + N 2 and N 2 + NO binary gas mixtures, the second virial coefficient for NO pure gas, and the third virial coefficient of O 2 and N 2 pure gases.