From Edisonian trial and error to the inverse design of materials and molecules


The goals of this chair are to develop methods to dramatically accelerate the rate at which new materials can be discovered. We will do this using a two-pronged strategy. On one hand, we will use physics-informed equivariant neural networks to accelerate quantum a mechanical calculations of material properties in the search for new spin-crossover materials. On the other, we will use generative modelling to teach machine learning models how to suggest entirely new atomic arrangements that ate statistically likely to possess the materials properties we are seeking. Together these represent a significant step forward for the field and will be part of a transformation in the way we discover new materials.


Two PhDs have been hired, one to work on "development of physics-based machine learning models for accelerated discovery of materials" and the other on "development of generative ML methods for the inverse design of materials and molecules."


Upcoming invited talk at "Machine Learning Interatomic Potentials: Theory and Practice" conference, November 6-10, Helsinki, Finland.

Published on  January 25, 2024
Updated on January 25, 2024