PDE learning for verification

• Sponsored by: Max Planck Institute for plasma physics
• Scientific Lead: Prof. Dr. Eric Sonnendrücker and Dr. Ahmed Ratnani
• Project Lead: Dr. Ricardo Acevedo Cabra
• Term: Summer semester 2018

The Max Planck Institute for plasma physics is investigating the principles underlying a power plant, which like the sun will produce energy from the fusion of light atomic nuclei. To this aim very hot charged particles are confined by a strong magnetic field away from the reactor wall in toroidal devices like Tokamaks or Stellarators, which are operated by the institute in Garching and Greifswald respectively. In order to understand the complex physics of magnetic fusion numerical simulations play an essential role and become more and more complex. Designing efficient, robust and reliable simulation software is an essential task of the institute. To this aim a careful verification and validation process needs to be undertaken. Recently developed machine learning techniques provide new tools for the verification of the model used in a simulation directly from the simulation data. Based on recent research papers, this project shall implement and compare some of these techniques using python and the scikit-learn package.

Results: The results of this project were summarised in the presentation.