Advanced Computational Methods for NMR Crystallography

Knowledge of the atomic-level structure is key to understanding and predicting properties of materials. X-ray diffraction (XRD) is the methods of choice for structures containing well-defined long-range order. However, many materials contain various degrees of disorder and are thus not characterizable by diffraction methods. In contrast, NMR directly probes local atomic environments and thus allows for structural characterization. In solid-state NMR several types of observables (such as quadrupolar coupling constants, dipole coupling constants, 1H/1H spin diffusion and chemical shifts) can be used to extract structural information.

In chemical shift driven NMR Crystallography (NMRX) comparisons between experimental and calculated chemical shifts are used to identify the experimental structure from an ensemble of trial structures. The candidate structures are generated either by a comprehensive crystal structure prediction (CSP) search or through searches using different degrees of chemical intuition in combination with constraints extracted from experimental data.

In the present thesis we use chemical shift driven NMRX to investigate materials containing different types of structural disorder, ranging from microcrystalline solids over doped structures up to amorphous materials.

A perfect application for NMRX is the structural determination of drug polymorphs, where the samples are often only available as microcrystalline powders. Here, we investigate a combined CSP-NMRX approach for structure determination of microcrystalline molecular solids. To this end, we first evaluate the positional accuracy of the combined approach. Then, we develop empirical-based methods as well as machine learning algorithms to extend the scope of the CSP-NMRX approach. Finally, we combine the developed methods to determine the crystal structure of powdered ampicillin, for which the traditional approach to CSP-NMRX would have failed.

Another interesting class of structures to investigate with NMRX are amorphous compounds, which are an important component in many industrial devices and materials. Amorphous structures cannot be described by a single crystalline unit-cell, and therefore, the CSP-NMRX approach is no longer applicable. Here, we determine the atomic-level structure of amorphous calcium silicate hydrate by generating a constrained ensemble of local structural motifs using chemical intuition and experimental data. We then evaluate the local structural motifs by comparing calculated and experimental chemical shifts. Finally, we combine the selected local motifs to generate an extended amorphous structural model.

The last applications for NMRX which we investigate are doped structures. Doping is a key technology to design new functional materials with desired properties and has been successfully used in various industrial materials. However, the presence of dopants inevitably leads to disorder within the material. In general, the same approach we investigated for amorphous materials should be applicable. However, the systems analyzed here contain heavy atoms and thus a higher level of theory is required in order to accurately calculate chemical shifts. We investigate different hypothesis for doping mechanisms in a set of photovoltaic lead halide perovskite materials. For these materials, we show that chemical shift based NMRX is able to differentiate between interstitial dopants, surface passivation layers and the formation of segregated phase