The chemical industry is facing several important challenges in the coming decade, mainly as a result of the tremendous impact of society on the environment, but also to maintain current living standards taking in to account the economic viability of chemical processes. These challenges
entail the use of alternative feedstocks compared to the conventional fossil resources, and the design and optimization of chemical processes with a focus on a minimal use of energy and materials next to avoiding pollutant formation. An important methodology to solve these challenges are experimental measurements to search for optimal process parameters such as reactor configurations, reactor conditions and inlet concentrations. These experiments, however,
are time consuming, costly and can only cover a limited range of the aforementioned parameters.
Modeling of chemical processes can be a great aid to accelerate the understanding of these parameters on the process outcome and minimize the number of necessary experiments. These modeling methodologies rely on a very detailed knowledge of the underlying chemistry to the
overall process. This requires the inclusion of all important elementary chemical reactions and their reactants and products. Even more importantly, the reaction rate coefficient of each reaction and the properties of each species are needed, constructing the so called kinetic model. When
dealing with common chemical processes, the number of chemical reactions often exceeds a thousand, with hundreds of molecules and intermediates. The amount of data to construct these kinetic models is thus very large. This thesis focusses on the latter, i.e. on how a large number of
accurate data can be retrieved automatically without the need for human interaction, which is part of a larger framework of automatic kinetic model generation. For the latter, many parameters, e.g.
reaction rate coefficients or thermodynamic properties of molecules, are calculated using approximation methods such as group additivity or other quantitative structure-property
relationships. Several of these methods have been proven to yield results of sufficient accuracy, given that they are not used outside their application range. The latter depends on the data used by the approximation model. These data originate from either experimental measurements or from high-level quantum chemical calculations. Today data from these high-level quantum chemical calculations are scarce due to the necessity of user involvement and expertise.
Therefore, in this work, methods are developed to allow automatic calculations of a large number of reaction rate coefficients and thermodynamic parameters using ab initio calculations. The results can then be used to develop new approximation schemes, to improve existing one or to
extend their applicability range.
State of the art
Automatically performing ab initio calculations has already been implemented in several computer codes, with a different application in mind than the current work. These methods prove
to be of great value and were used to investigate the state of the art. One of these computer codes is “KinBot”, developed at Sandia National Laboratories, which automatically searches for
reactions on a potential energy surface. KinBot has been used to identify reactions during npentanol pyrolysis and to calculate accurate rate coefficients for these reactions. They were implemented in a kinetic model which shows good agreement to experimental data.
To automatically perform ab initio calculations, the framework of the kinetic model generator tool “Genesys” is used, providing a computational representation of molecules, intermediates and reactions. The representation entails information on the connectivity of atoms within a species or
transition state, in the form of a mathematical graph or connectivity matrix. This information is insufficient to start ab initio calculations, for which three dimensional coordinates of all the atoms are needed. Using “distance geometry” and force field calculations, the connectivity of a species is translated to three dimensional coordinates. Although these methods have already been applied to molecules and radicals, in this work the application range has been extended to transition states. In several steps, the coordinates are optimized and the lowest energy conformer can be selected. For the latter, high-level ab initio calculations are done, e.g. using the CBS-QB3
level of theory.
From the CBS-QB3 calculations, species properties are calculated using statistical thermodynamics, for which the partition functions are calculated using the molecular frequencies.
Internal modes resembling rotations around a chemical bond are treated as one dimensional hindered rotors, and the frequencies corresponding to these rotations need to be removed from the partition function calculations. This is automatically done by verifying if the normal mode of a
frequency resembles an internal rotation. From the thermodynamic properties calculated for reactants, products and the transition state of a reaction, kinetic parameters can be easily calculated. These parameters are further improved by calculating tunneling corrections.
Thermodynamics and kinetics
The newly developed methodology has been tested by calculating the thermodynamic properties and rate coefficients of several species and reactions respectively, exhibiting a wide variety of functional groups and molecular structures. These values have been compared to literature data, both experimental as well as theoretic, and an excellent agreement is found, similar to the accuracy of the CBS-QB3 calculations themselves. Automatically performing the calculations
has thus no impact on accuracy compared to the “manual” calculations, and can be done much faster and on a larger scale.
Using the newly developed methods to automatically perform ab initio calculations, reaction rate coefficients have been calculated for a large set of intramolecular hydrogen abstraction reactions in hydrocarbons, including abstractions by a carbon radical from a neighboring carbon atom up to
abstractions by a carbon radical to another carbon atom with a carbon chain of five atoms in between. The influence of the ligands of the attacking and attacked carbon atom has been studied, as well as the influence of substituents on the carbon chain between both reactive atoms. This has led to a new group additivity model for intramolecular hydrogen abstractions, which is able to reproduce the ab initio calculations with sufficient accuracy.
With the new group additivity model for intramolecular hydrogen abstractions, a kinetic model for the pyrolysis of n-heptane has been automatically constructed using “Genesys”. By accounting for the chemistry of small molecules and for the formation of aromatics through literature data, the new model was able to reproduce two datasets in a very wide range of
pressures (400 – 1 105 Pa) and temperatures (800 – 1800 K) without any adjustments to the kinetic model using the experiment data itself. A rate of production analysis has unraveled the major reaction pathways, and has shown the necessity of including intramolecular hydrogen
abstraction reactions for the correct prediction of several pyrolysis products.