Solid-state NMR is a highly sensitive probe of atomic scale structure and dynamics. Experimental NMR often relies on electronic structure calculations based on DFT to provide the assignment and interpretation of observed spectra. Over the past two decades, Yates and co-workers have developed computational tools to predict NMR parameters from materials modelling simulations. These calculation are now widely used by experimentalists, and have helped drive the field known as NMR Crystallography.
Despite these advances many materials problems have remained out of reach due to the computational cost of DFT simulations - for example the effects of long-time scales (for thermal vibrations or ionic motion), or large disordered materials. The recent advances in machine learning have provided a potential route to address these challenges. We have recently publish a method to machine learn NMR tensors using graph-neural-networks (https://arxiv.org/abs/2412.15063).
This project will build on these developments. There are two different strands of work: (a) application of these ML methods to study challenging problems in Materials Science - for example ionic motion in ionic conductors. This will be in close collaboration with experimental NMR studies (b) fundamental improvements in the accuracy and applicability of these methods.
For more information contact Prof. Jonathan Yates.