Realistic transport models in non-linear magneto-hydrodynamic simulations of instabilities i H/F (Thermohydraulique et mécanique des fluides)

Job description

Domaine : Thermohydraulique et mécanique des fluides

Contrat : Stage

Description du poste :

The plasma in tokamak fusion reactors, such as ITER, will be subject to large scale instabilities. These instabilities are well described by modelling the plasma as an electrically conducting fluid embedded in a magnetic field. The model is called magneto-hydro-dynamics (MHD).  Computer simulations allow the study of these MHD instabilities to better understand the physics and the methods for controlling them.

The MHD simulation code JOREK has been developed within the European fusion program for this purpose. The code is applied to so-called edge localised modes (ELMs) and disruptions, MHD instabilities that will need to be controlled in ITER. The simulations require high performance computers, using several thousand cpus. At present, the JOREK code accurately describes the MHD aspect of the instabilities. However, the transport due to underlying turbulence (not included in the MHD model) is modelled by a very simple, ad hoc, diffusion coefficient for the energy and density. To improve the fidelity of the MHD simulations, a better model for the turbulent transport is needed.

Transport due to turbulence is calculated by very large scale so-called gyro-kinetic simulation codes. It is not feasible to couple a turbulence code and MHD code directly to obtain the relevant transport. This would be computationally out of reach. However, recent progress has allowed capturing the results of a large number of gyro-kinetic simulations in a compact form using neural network representations. It has been shown, in other applications, that these neural nets can accurately reproduce the transport coefficients on a very fast timescale.

The subject of the stage is to implement the transport model from the neural network into the JOREK code. A first step will be introduction to the JOREK code and the neural net transport routines. It has to decided how the output (i.e. energy fluxes) of the neural net can be included in the MHD equations. After its implementation, the improved JOREK code will need to be benchmarked against an existing transport code, not including the MHD instabilities to verify the implementation. Time permitting; the final step will be an application of a 3D MHD instability including realistic transport from the neural net model.


Physicist with some experience in compuational physics and numerics.

Ville : SAINT PAUL LEZ DURANCE

Langue / Niveau :

Anglais : Courant