Our research is organised into three themes, which occasionally overlap. The list is constantly changing, but these are representative of recent DPhil and Part II research projects.
Research
Development and testing of methods for crystallographic analysis. These projects may end up incorporated in publically released software tools. Examples of past projects include:
- Coding and testing uses of crystallographic displacement parameter restraints. Atomic displacement parameters tend to be less well determined than positional parameters and as a result they may take non-physical values. Restraints can help condition the crystal structure refinement to give physically realistic parameters, but traditional software offers relatively few options to do this.
- Implementing atomic distributions for realistic modelling of atomic motion and disorder. Traditional models treat atoms as point scatterers convoluted with an atomic density and some simple (anisotropic) atomic displacements. We have previously implemented line, ring and shell scatterers which allow modelling of some disordered species. Recent work reimplemented and tested the 'hindered rotor' model which describes atoms oscillating harmonically around an axis, for example, a CF3 group.
- Comparison of atomic scattering factor models. Most crystal structure analysis uses spherically averaged models of the scattering from atomic electron density. The results are good enough for accurate structure determination, but there is room for improvement when the atomic density deviates significantly from a spherical distribution. Notably, electron density in chemical bonds or lone pairs breaks the spherical symmetry. The effect is especially pronounced in hydrogen atoms where a significant portion of the atom's electron density is found in the bond.
Applications of machine learning and statistics to crystallographic data. These projects continue in a long tradition of identifying rules and patterns in datasets.
- Prediction of crystallization or co-crystallization propensity. The primary application of this type of analysis is in crystal engineering of pharmaceutical materials, but there is also interest in finding links between material properties and molecular structure.
- Representation of crystal structures for AI/ML applications. Finding an efficient and meaningful representation of molecular crystal structures for particular applications (e.g. structure validation, structure interpolation, similarity measures and structure prediction) can be a surprisingly powerful means of making statistical models more robust and generalizable to unseen examples.
Crystallographic projects. Determination and comparison of structures of series, or missing members of a congeneric series of molecules.
- Crystal structure validation. Re-synthesis and determination of published structures which contain apparent errors, or which exemplify certain crystallographic problems.
- Molecular solid solutions. Crystalline solid solutions (mixtures) of molecules can significantly alter physical properties of the material. Molecular similarity measures can be used to target potential molecules to form these isomorphous structures.
Image credits
Computer photo by Joshua Reddekopp on Unsplash
Colourful confetti photo by Sebastian Schuster on Unsplash
Avocado photo by Anne Nygård on Unsplash