Former Research Interest
The most successful predictive activity coefficient models for phase equilibria are UNIFAC and modified UNIFAC, in which a molecule is described as a collection of independent functional groups, and a mixture is considered to be a mixture of these functional groups. The activity coefficient of a molecule in a mixture is then obtained from the sum of activity coefficients of the constituent groups in the mixture. Such group contribution methods greatly reduce the number of parameters needed in describing a variety of mixture systems. However, as is true for all group contribution methods, UNIFAC and modified UNIFAC suffer from the inability to distinguish between isomers and inaccuracies when compounds with several strong, non-alkyl functional groups are considered. The functional group activity coefficients, which are dependent on the interactions between functional groups, cannot be measured; only the activity coefficient for the whole molecule can be obtained from fitting to experimental phase equilibrium data. As a consequence, the quality of UNIFAC predictions depends on the similarity of the new system to the database used in its parametrization. There is no obvious way to improve these models other than empirical approaches such as adding additional functional groups, introducing second and higher order functional groups (to account for the intramolecular interactions between functional groups), and increasing the database used to fit parameters.
We have developed a new activity coefficient model using
molecular solvation based on the COSMO-RS method. In this model quantum
mechanical COSMO calculations are performed to obtain the screening charges
for molecules in a perfect conductor. In analogy to the UNIFAC model, here
we consider molecules to be a collection of equally sized surface segments,
with their interactions determined from the screening charges they acquire
on ideal solvation. This model requires only a single set of radii for
all atoms involved in the COSMO solvation calculations and two universal
parameters to determine segment interactions. This is significantly fewer
parameters than other group contribution methods such as UNIFAC (168 parameters)
and modified UNIFAC (612 parameters) models for the phase equilibrium problems
considered here. The resulting completely a priori prediction method
results in an absolute average deviation of vapor-liquid equilibrium predictions
for 243 binary mixtures is 9% in vapor phase composition and total pressure,
and root-mean-square deviations of the octanol/water partition coefficient
log10KOW, infinite dilution activity coefficients
lng¥ in water, and hexane for
64 compounds of 0.48, 1.65, and 0.50, respectively. This model is general
and applicable for the a priori prediction of the phase behavior
of most compounds.
A Multipole Correction Method to Improve the predictions from Group Contribtuon Methods
Group contribution methods have been reasonably successful for estimating many physical and thermodynamic properties of pure substances and mixtures, such as the heat capacity, Gibbs free energy of formation, excess Gibbs free energy, etc. These methods can be used not only to predict the properties of new, more complex systems but also to design new compounds with desired properties. Although group contribution methods are simple and powerful, they are limited by three common deficiencies. First, the definition of groups is empirical and arbitrary. Second, these methods do not distinguish between isomers (structure effects). Finally, group contribution methods fail when a molecule contains two or more strong functional groups in close proximity (proximity effects).
We have developed a group contribution model, called the
GCSKOW model, for predicting the octanol-water partition coefficient, KOW.
This model was found to provide accurate predictions of KOW for monofunctional
solutes. However, when applying the same set of group parameters to multifunctional
compounds, large deviations were observed due to proximity effects. In
this work, we resolve this problem by using multipole corrections to the
free energy in the GCSKOW model. In particular, we use quantum mechanics
to determine the charge (monopole) and dipole moments of a group in an
isolated molecule, and based on this make corrections for the free energy
of the same group in different molecules. This way, structure and proximity
effects are simultaneously accounted for with the additional piece of information
from inexpensive, first principle calculations. The calculated results
for a set of 322 compounds show that with multipole corrections the GCSKOW
model is more accurate then ClogP, which heretofore had been considered
the best existing predictive model for KOW. Future work is to apply the
same idea to other group contribution models, such as UNIFAC, for phase
Quantum Mechanical Computation of the Infinite Dilution Activity Coefficient
Infinite dilution activity coefficients are of both theoretical and practical importance. Such information can be used to test thermodynamic models or to fit the parameters in existing models. Infinite dilution activity coefficients are of direct use in the design of separation and extraction processes. Recently, the infinite dilution activity coefficient has also been used to predict the value of octanol-water partition coefficient, which is the key parameter in the prediction of the fate of chemical pollutants in the environment, and to correlate properties such as biological activity and toxicity.
We have developed a Group Contribution Solvation (GCS) model to accurately predict both the infinite dilution activity coefficient and the octanol-water partition coefficient based on computational quantum mechanics. In this work, modern computational chemistry is used to determine the energies of solvation of molecules in various solvents; that is the energy of taking a molecule from the pure ideal gas state and placing it in a liquid solvent. From this and information about the size and shape of the molecule, we can then predict the infinite dilution activity coefficient directly. By making such predictions for two solvents, we get accurate values for the octanol-water, hexane-water and other partition coefficients.
From the results for a collection of small molecules,
we have developed an easy-to-use group contribution solvation (GCS) model
for the prediction of the infinite dilution activity coefficients and partition
coefficients of large molecules. An important feature of this model is
that it can be used without the need to perform any quantum mechanical
calculations (which would be expensive and time consuming for large molecules).
We have found that the GCS model is more accurate than the modified UNIFAC
model for the prediction of infinite dilution activity coefficients, especially
for alkanes in water. GCS is also more accurate than other current predictive
models for the octanol-water partition coefficient, such as those based
on linear solvation energy relations (LSER) and the method of Hansch and
Leo. The GCS model provides a new perspective on predicting physiochemical
properties from molecular interactions in solution, and using computational
chemistry to compute these interactions.
Research advisor: Prof. Stanley I. Sandler