Improving detailed kinetic models through automated transition state theory calculations

Abstract

For decades, researchers have been developing detailed kinetic models to understand chemical reactions and intermediate species in complex systems such as combustion and catalysis. These models routinely contain hundreds of species and thousands of reactions for which thermodynamic and kinetic parameters must be specified. Researchers have traditionally constructed these mechanisms by hand, but these methods are tedious and prone to errors. In addition, models are often reported in the Chemkin format that was originally designed to fit on an 80-column punch card. The Chemkin format has led to creative naming conventions that make it difficult to translate which names correspond to which species, further increasing the complexity of these models. Also, thermodynamic and kinetic parameters in these models originate from a variety of sources with differing fidelities. Ideally, these parameters come from experimentation or ab initio calculations, but given the number of parameters that need to be specified, are often estimated or approximated. This body of work aims to improve these mechanisms through the use of a tool that allows for side-by-side comparison of kinetic models, the Importer, and a code base that performs automated transition state theory calculations, AutoTST. The Importer was used to convert mechanisms from the hard-to-interpret Chemkin format into an unambiguous format. To do this, the Importer relies on Reaction Mechanism Generator (RMG) to represent species as chemical graphs where atoms are nodes connected by bonds which are edges. It also utilizes reaction templates to propose all possible reactions between known species and a database of thermochemical parameters. These methods in RMG enable the Importer to propose species-name matches for manual (human) verification. The process is repeated as the human-computer team works in tandem to identify all species present in a model. In this work, we utilize the Importer to identify species and reactions in 98 detailed kinetic models. Through this process, we have identified significant disagreements with thermochemistries and reaction rates. We have observed many enthalpies disagree by over 100 kcal/mol and kinetics disagree by over 100 orders of magnitude. Through this work, we have also developed upon AutoTST, an automated rate calculator. AutoTST matches reactions to one of three supported reaction templates and predicts transition state geometries using a set of training data. The geometry undergoes a series of partial optimizations to arrive at a validated transition state which is used to calculate a rate expression. In this thesis, we have included a systematic conformer search, parallelized \textit{ab initio} calculations, 1D hindered rotor treatment, and vibrational analysis methods to improve the success rate and accuracy of rates calculated by AutoTST. Finally, we utilized the Importer to fully import a detailed mechanism for butanol and attempted to calculate rates of all possible reactions using AutoTST. We then swapped our calculated rates into this butanol model one at a time and assessed the impact of our calculations by comparing these alternate models to a set of experimental data using PyTeCK (Python tool for Testing Chemical Kinetics). We also performed a literature search to identify the sources of kinetics for high-impact parameters. This work provides new insight into how detailed kinetic models are constructed and methods for improvement. More importantly, it also provides avenues of future exploration so future researchers are able to generate increasingly accurate kinetic models.