Optimizing Magnetically Confined Plasmas with GENE

The Second Grand Challenge

To create fusion energy, hydrogen plasmas of more than 100 million degrees are to be confined in strong and suitably shaped toroidal magnetic fields. The so-called energy confinement time is controlled by turbulent heat transport from the centre to the edge. GENE is a gyrokinetic turbulence code that can quantitatively describe the underlying physical processes, and it has been successfully compared to experimental data from existing mid-scale fusion devices numerous times. The big challenge for the worldwide fusion community is to extend such capabilities to next step devices like ITER and DEMO, and GENE is well positioned to pioneer this effort.

Exascale platforms will enable GENE to predict the temperature and density profiles and therefore the plasma performance of the international flagship experiment ITER and help guide the design of the pilot power plant DEMO. This amounts to a breakthrough in fusion research and will only be possible by adopting cutting edge software like GENE to some of the world’s largest supercomputers.

How can exascale simulations solve this grand challenge? Turbulence is known to be a multi-scale problem, and the range of spatial/temporal scales one needs to resolve to predict the plasma profiles and performance of ITER and DEMO calls for extreme computing on emerging exascale platforms. (We note in passing that we intend to employ the code versions GENE and GENE-X in tandem here, covering, respectively, the core and edge region of the plasma, and being loosely coupled.) Taking simulations of existing mid-size devices like ASDEX Upgrade as a reference point, tackling ITER will require at least a 10-fold increase of resolution in 5D position-velocity space and an extension of the simulation times by a factor of about 100. Recently developed multi-scale techniques can reduce this effort by one order of magnitude, leaving us with a 100-fold increase of the overall computational effort. Single such simulations may take up to several 100 million core-hours. They will enable a new kind of fusion plasma science, providing one of the few pathways to significantly accelerating fusion research.