To tune the mathematical models in such a way that they best describe real-world/measured date we formulate optimization problems with PDE-constraints. The constraints describing the underlying phycics are mostly time-dependent non-linear PDEs or systems of PDEs. Applying state-of-the-art techniques from numerical optimization requires the solution of linear systems in structured form. As these are too large to be treated directly, we apply iterative procedures requiring a minimal amount of work. To guarantee the robustness of this approach one needs to find a suitable accelerator the so-called preconditioner which allows us to limit the workload and make the simulation of three dimensional models possible.
- Andy Wathen (University of Oxford)
- John Pearon (University of Edinburgh)
- Tyrone Rees (Rutherford Appleton Laboratory)
- Sue Thorne (Rutherford Appleton Laboratory)