Designing and Implementing Computational Methods to Study Catalysis

Abstract

In recent years, new findings in different branches of chemistry have been highly dependent on computational and theoretical tools. Advancements in computing power and developments in computational chemistry methods has provided a unique opportunity for the field of computational method development to thrive. Investigations using new computational methods are not heavily guided by experiments anymore and need minimal structural and mechanistic insight from experiments for model development. Complex chemical transformations can be studied using these methods with a reasonable computation time. A variety of tools are built by computational chemists to assess the physical properties controlling chemical processes. This dissertation evaluates the viability of two classes of computational methods for examining chemical reactions, reaction discovery tools and multivariate analysis methods. Reaction path and transition state finding methods fall under the first category and provide means to gaining insight into the reaction mechanism and transition state structure at the atomistic level while quantifying kinetic and thermodynamics of the reaction. The ultimate goal of this type of studies is to either explain the experimental observations that do not follow traditional chemical principles or modify reaction conditions or reagents to improve the desired outcome of the reaction. However, in situations where there is not enough information to develop a model for reaction path finding studies, alternative methods such as multivariate analysis could be used to uncover mechanistic details and engineer catalysts at even a cheaper computation cost. Motivations and limitation of designing and implementing new methods for surface chemistry reactions are briefly discussed in Chapter 1. Linear free energy relationships and their application in mechanistic studies and their relation to quantum mechanical methods were also briefly described. Chapter 2 introduces a new reaction path and transition state finding method called surface growing string method, for exploring surface reactions that is at least two times faster than the conventional nudged elastic band method and finds the structures along the reaction path, the transition state, and the product in one single run. Chapter 3 expands upon the growing string method and the automation process for orientation sampling, geometry optimization, driving coordinate generation, and finally performing reaction path finding. The new algorithm described in Chapter 3 resulted in the surface-ZStruct program. Using surface-ZStruct, the complete reaction cycle for atomic layer deposition of titanium nitride, a diffusion barrier in microelectronics, is uncovered computationally for the first time. In Chapter 4, the biocatalysis of coumarin cross-coupling reactions is investigated using data analysis tools in an attempt to identify controlling factors of reactivity and selectivity. The crystal structure of the enzyme's active site is not available, and the only information at hand are the structure of the native substrate, structure of other substrates used in the screening process, and the yield of the reactions. Based on the results, we believe the reaction proceeds via a di-radical pathway through double H atom abstractions. The yield is also dependent on the size of the substituent on the coumarin scaffold, however this association varies in different substrate pairs. The results suggest that the enzyme pre-organizes the substrate pair into a complex that facilitates the initial O-H homolytic bond cleavage. Using the size and bond dissociation energies of phenolic OH bonds, the yield of the reaction is predicted within a reasonable accuracy.