Analytical property prediction models for complex fluids

 

The figure bellow depicts a schematic cycle of modeling chemical systems in a multi-scale manner. The ultimate goal is to obtain models that could predict properties at each scale by having information either from experiments or from models at other scales.

By designing a substance one means the acquirement of a chemical system where a certain set of its properties would satisfy a desirable target value or value range. As it can be seen from the figure, we do the modeling at different length and time scales. This is necessary partly for it allows for understanding many macroscopic phenomena by investigating their underlying microscopic processes and partly for the reason that experimental measurements of some properties are either impossible or very hard, so that for obtaining predictions for them one needs to rely on models that cover these properties but are validated using another set of properties at different scales.

 

Property prediction (forward modeling)

It is attempted to maximize the diversity of properties of a chemical system that can be predicted having a minimum of available information on the system. By increasing the quality of these predictions, a gradually more realistic image of the target system emerges which in turn would be of great assistance throughout various phases of the overall process design.

An example of a phenomenon that poses a challenge on the current models is the dissolution of cellulose, lignin and other compounds obtained from the processing of wood in various solvent candidates. These have not only no well-defined structure, but their size results in a conformational flexibility that is hard to capture by simple conformation search methods. Nevertheless, we have performed a screening of ionic liquids for high cellulose solubility that yielded interesting candidates for new solvents.

Other examples include solution systems where the molecules show associative behaviour in the solvent, that is they appear to build clusters rather than being independently solvated. Correspondingly, the models that cannot describe this associative behavior, fail to predict the properties of these systems. Hydrogen bonding and ionic interactions are two examples of the underlying causes that lead to the associative behavior.

 

Structural analysis of mixtures

One of the most important properties of every mixture is its molecules’ conformational space. For a certain molecule this could be given as a sub-set of the 3N-dimensional cartesian space, where N is the number of atoms in that molecule. Every point in this sub-set, a matrix of Nx3 dimensions would represent a single conformation. This information is either impossible or very hard to obtain experimentally. It is, on the other hand, crucial if one wants to build models that relate the structure of the molecule to its macroscopic properties.

We use molecular dynamics simulations of mixtures to scan their conformational space. The efficiency with which the simulations are run and the reliability of their results are important parameters that should be optimized if the obtained information on the conformational space is to be later used in other models for successful predictions.

 

Thermodynamic properties of complex fluid systems

Thermodynamic modeling of chemical systems is the basis for predictions of a vast number of properties. These include for example partition coefficient or vapor pressure which play a role in the further extractive or distillatory separation process analyses, respectively. Other examples would be surface tension or viscosity where a sensible underlying thermodynamic theory can provide us with insightful predictions.

One of the models that we apply for the calculation of thermodynamic properties of the mixtures is the COSMO-RS. Here each molecule is treated as being in a cavity with a well-defined surface area. Using quantum mechanical calculations the charge distribution is then calculated all around the molecular surface areas. Besides the electrostatic factors, other interactions such as the entropic factors, hydrogen boding, etc. are taken into account by simple empirical relationships with terms to fitted experimental data. For more information please refer to the Cosmologic website.

 

Statistical analysis of structure-property data sets

When a phenomenon under study is so complicated that an analytic modeling of the system is not possible, Quantitative Structure-Property Relationships (QSPR) can be used to establish a correlation between descriptors obtained from the structure of the molecule and the corresponding property of interest. An example of such a complex property is the lubricity of a fluid that makes an analytical physical modeling impossible.

Using COSMO-RS for the calculation of some molecular descriptors one we have then related these to experimental data by statistical probing of the importance of the respective descriptor concerning the measured lubricity in cooperation with IFAS. This lead to a new lubricity model for bio fuels.