Generative models for crystalline materials design

Crystalline materials underpin many clean energy processes from light-to-electricity conversion in solar cells to electrochemical energy storage in batteries. I aim to accelerate the development of new materials that meet the needs of next-generation technologies that are higher performance and more sustainable.

The project will focus on the development and application of artificial intelligence (AI) techniques for materials design. This will involve improved representations to enable the encoding of materials into an informative latent space. The encoding process will preserve the physicochemical properties and symmetry considerations intrinsic to crystalline materials,

ensuring that generated structures are both chemically plausible and potentially superior in physical performance. Navigating the latent space is a critical aspect of this research, as it involves decoding to unveil novel materials with desired properties. Generative models will be trained to interpret the

latent space effectively, allowing for the prediction and evaluation of materials before their experimental synthesis. This may include elements of variational autoencoders, transformers, and reinforcement learning. The resulting iterative design loop will enable the exploration of an extensive combinatorial space of material configurations.

The practical outcomes of this research promise to impact the development of energy technologies and will involve physical properties and performance metrics based on electronic, ionic and thermal processes in crystals. By combining computational chemistry, machine learning, and materials science, I hope that my PhD research will pave the way for

smart materials discovery and the realisation of more sustainable energy systems.