Emsley, David LyndonBalodis, Martins2022-01-202022-01-202022-01-20202210.5075/epfl-thesis-9004https://infoscience.epfl.ch/handle/20.500.14299/184658NMR crystallography has been around for half a century, but with the advent of NMR crystal structure determination protocols in the last decade it has shown perspectives that were not seen before. Amalgamation of NMR and crystal structure determination has been successful in predicting crystal structures de novo. Still, there are many challenges to make these methods universal and applicable to any molecular crystal at any stage of a structural investigation. Larger molecules still take too much time to be solved by the current methods up to the point of being impossible. Amorphous structures of molecular solids had not been solved in general for any method. Reaction mechanisms and structures involved in the formation of solids are still challenging to investigate due to their fast nature and the low concentration of the reaction intermediates in solution. The current NMR crystal structure determination protocols involve NMR at the final step of the candidate crystal structure selection, but one of the biggest bottlenecks is actually at the first steps of structure determination that involve selecting gas phase conformers that will later be used in the trial crystal structures. Here, we develop a series of new NMR methods to address these bottlenecks. We show how unambiguous constraints extracted from solid state NMR experiments can help to significantly reduce the initial conformer space and help to select conformers that correspond more closely to what is found in the crystal structure. We then show how machine learned chemical shifts can be included in the crystal structure determination process from the start, and we determine the crystal structure of two compounds, one of whom is polymorphic. Then, we show how machine learned shifts in combination with molecular dynamics can be used to solve the atomic level structure of an amorphous compound yielding insights into the hydrogen bonding and stabilization of the amorphous form. To further aid crystal structure determination, we present a new method to assign spectra based on machine learned chemical shifts and propose a new database for small organic molecular crystals that would help in the future crystallographic investigations. Finally, we investigate carbonate speciation by using dissolution DNP.enNMR crystallography (NMRX)crystal structure prediction (CSP)ShiftMLsolid-state NMRNMR databasespectral assignmentdissolution DNPamorphous materialscrystalline materialscarbonates.thesis::doctoral thesis