Preview

PDF, English - main document Download (36MB) | Terms of use

Abstract

Lewis acids are defined as electron pair acceptors, and the introduction of structural strain was shown a powerful lever to increase the Lewis acidity of p-block atom-based molecules. This dissertation presents experimental, quantum chemical, and data scientific contributions to the research area of p-block element-based Lewis acidity. The reactivity of calix[4]pyrrolato aluminates (Cx[4]Al) was investigated experimentally. Cx[4]Al was found to heterolytically split the O-H bond of protic substrates (alcohols, carboxylic acids) in a reversible manner. With triplet dioxygen as substrate, Cx[4]Al forms an alkyl peroxido aluminate. A subsequent rearrangement reaction of the peroxido aluminates which yields α,β-unsaturated carboxylato aluminates was discovered and its reaction mechanism investigated. Furthermore, the reactivity of Cx[4]Al with nitrogen monoxide, nitrosobenzene, and phenyl isocyanate was studied. Characteristic for all reactions is an aluminum-ligand cooperative mechanism. The chemistry of the calix[4]pyrrolato ligands was expanded to gallium, and calix[4]pyrrolato gallates(III) (Cx[4]Ga) were synthesized and fully characterized. They are the first examples for ideally square planar-coordinated Ga(III) atoms. Cx[4]Ga’s reactivity with small molecules (e.g., CO2) as well as its redox chemistry were researched. More than 250 small p-block atom-based molecules in square planar structural configuration were investigated with density functional theory and wavefunction-based quantum chemistry. Substituent effects on the planarization energies were discussed and related to the second-order Jahn-Teller theorem. Indeed, the square planar configurations can serve as transition states for the stereochemical inversion of the respective molecules. In addition to that, a new stepwise mechanism for stereoinversion was discovered. Lastly, the FIA49k dataset, encompassing 48,986 fluoride ion affinities of molecules based on 13 different p-block elements, is presented herein. The data was generated with double-hybrid DFT calculations. It was used to train FIA-GNN, a graph neural network machine learning model that predicts FIA values. FIA-GNN’s usefulness was demonstrated in four different case studies.