About this role
Full time: 35 hours per week
Fixed-term: 36 months
Vacancies available: 2
The opportunity:
This project seeks to realise a new approach to turbulence computation via a combination of modern AI technology (graph neural networks and transformers) with a dynamical systems framework. The dynamical systems viewpoint imagines the turbulence as a space-time tapestry of connected states, which are related to exact solutions of the governing equations in smaller domains. This viewpoint is compelling: it connects dynamical events to statistical properties of the flow, it generates a set of rules by which different recurrent patterns can coexist and interact and it is a robust platform on which to build low-order models for deployment in industrial applications. However, to date this approach has been restricted to small-scale flows because of (1) the computational challenge of identifying the coherent patterns and (2) the combinatorial challenge of identifying the rules by which the patterns join to form the space-time tapestry. These problems are a natural fit for frontier AI models. The two postdocs on this project will, together with the PI, his collaborators and industrial partners, design, train and release these models.
These positions are funded by UKRI to work with Dr Jacob Page on the design and construction of large neural networks for the embedding and prediction of high Reynolds number turbulence. The approach is motivated by Hopf’s “dynamical systems” view of turbulent flows. The project includes collaboration with a leading researcher at a major tech company, in addition to ongoing collaborations with Prof Steve Brunton’s group at the University of Washington and Dr Georgios Rigas’ group at Imperial College London. There will also be scope for involvement in other ongoing industrial collaborations.