Molecular simulation and modeling of the phase equilibria of polar compounds.

Abstract

The initial phase of the project involved an investigation into the modeling of binary carboxylic acid vapour-liquid equilibrium (VLE) data. This stemmed from the Masters research that led into the current study, in which the conventional gamma-phi formulation of VLE was found to inadequately describe the complicated acid chemistry. In an effort to correctly describe the dimerization occurring in both the liquid and vapour phases, the chemical theory of vapour-phase imperfections was applied. The chemical theory technique allowed the experimental liquid-phase activity coefficients to be accurately calculated by taking the vapour phase dimerization into account. Once these activity coefficients had been determined, standard Gibbs excess energy models were fitted to permit analysis of the VLE data's thermodynamic consistency. In addition, the typical bubble-point iteration scheme used for VLE data regression was adapted to include the chemical theory expressions necessary for satisfactory modeling of the carboxylic acids. The primary focus of this study was to determine the ability of currently available computer simulation techniques and technology to correctly predict the phase equilibria of polar molecules. Thus, Monte Carlo simulations in the NVT- and NPT- Gibbs ensembles were used to predict pure component and binary phase equilibrium data (respectively), for a variety of polar compounds. The average standard deviations for these simulation results lay between 1 and 2 % for the saturated liquid densities, and varied between 5 and 10 % for the saturated vapour pressures and densities. Pure component data were simulated for alcohols, carboxylic acids, hydrogen sulfide (ELS), sulfur dioxide (SO2) and nitrogen dioxide (NO2). For H2S, S02 and NO2, a potential model parameterized as part of this project was used to describe the molecular interactions. All the other compounds were simulated using the TraPPE-UA force field. The simulation results for the alcohols and acids showed a consistent saturated vapour pressure over-prediction of 5 - 20 % depending on the species and the system temperature. The liquid density predictions were, in general, good and on average differed from experiment by 1 - 2 %. The critical temperatures and densities were estimated from the pure component data by fitting to the scaling law and the law of rectilinear diameters. They were found to lie within 1 and 2 % of the experimental values for the carboxylic acids and alcohols, respectively. Clausius-Clapeyron plots of the saturated vapour pressures allowed the critical pressure and normal boiling points to be determined. The critical pressures were, as expected, over-predicted for both compound classes and the normal boiling points were under-estimated somewhat for the acids, but deviated from experiment by less than 0.5 % for the alcohols. A Lennard-Jones 12-6 plus Coulombic potential energy surface was parameterized for H2S, SO2 and NO2. For FbS, the proposed force field offers improved saturated vapour pressure and vapour density predictions when compared to the existing NERD force field, and comparable accuracy with the recent models of Kamath and co workers. SO2 and NO2 had not previously been parameterized for a Lennard-Jones 12-6 based force field. For SO2, there was excellent agreement with experimental data. In the case of NO2, the saturated liquid density predictions were very good, but the vapour pressures and densities were over-predicted. Binary VLE simulations were carried out for systems consisting purely of carboxylic acids, and also for H2S and SO2 with a selection of alkanes and alcohols. The liquid and vapour composition predictions were good for the acid systems, but the anticipated pressure and temperature deviations were observed in the isothermal and isobaric simulations, respectively. The H2S + alkane systems were generally good, as were the SO2 + alkane systems. For both H2S and SO2, the systems involving an alcohol displayed a characteristic pressure over-estimation. The azeotropes were, in most cases, predicted fairly well; the exception was the SO2 + methane binary. A sensitivity analysis of the Lennard-Jones unlike interaction parameters was also conducted. It was demonstrated that even minor changes to these parameters can have a significant effect on the final simulation results. The considerable affect that these parameters have on the simulation outputs was emphasized by studying the influence of different combining rules on the H2S + methane and H2S + ethane binary systems. Analysis of the radial distribution functions indicated that hydrogen bonding and dimerization were occurring in the alcohol and carboxylic acid systems, respectively. The H2S, SO2 and NO2 distribution functions showed little sign of any association, except for a small plateau in that of SO2. A radial distribution function from one of the carboxylic acid binary simulations was also analysed, and supported the assumption made in the chemical theory modeling work of using a geometric mean (instead of twice the geometric mean, which is favoured by some researchers) to determine the heterodimerization constant, KAB-