Files

Abstract

In order to get a better understanding of the interaction of plasma microinstabilities and associated turbulence with specific modes, an antenna is implemented in the global gyrokinetic Particle-In-Cell (PIC) code ORB5. It consists in applying an external perturbation to the plasma to excite various types of modes and study their coupling with the rest of the system. This antenna can for example be used to apply shear flows on individual microinstabilities or on fully developed turbulence. In other studies, the antenna can also be used to directly excite specific linear eigenmodes. The contributions of the antenna and plasma perturbed fields are considered separately and, optionally, the plasma response can be linearized by neglecting the perturbed plasma field contribution on particle orbits. As a first proof of principle, stationary ExB shear flows are applied to Ion Temperature Gradient (ITG) and Trapped Electron Mode (TEM) -driven instabilities. Their well-known linear stabilizing effect is successfully recovered. In cases where linear eigenmodes have a finite ballooning angle, the possibility of shear flows with a destabilizing effect is shown. Time-dependent shear flows are then applied to quantify their loss of effectiveness with frequency. When going to the nonlinear regime, applied shear flows prove to be unable to mitigate the heat flux level due to opposite zonal flows self-generated by the plasma. A reverse interaction, which is the generation of zonal structures by microinstabilities, is also investigated by exciting non-zonal modes and studying their decay into zonal modes. Future applications include the excitation of Alfvén eigenmodes with an electromagnetic antenna, for which the formalism is derived in this thesis. With a view to conduct the aforementioned studies, the ORB5 code had to be completely refactored. Indeed, the implementation choices of this legacy code were not adapted anymore to make the most of cutting-edge supercomputer performance. Data structures have been re-designed to ensure efficient memory access, enhancing data locality. The MPI parallelization scheme has been complemented by OpenMP multithreading to benefit from the shared memory of many- and multi- core devices. As more and more High Performance Computing (HPC) facilities provide GPU-equipped systems, ORB5 has also been ported to GPUs using OpenACC directives. As a result, the same source code can be run efficiently on different HPC architectures. Performance studies are performed on the Summit machine at ORNL (U.S.A.), Piz Daint at CSCS (Switzerland), and Marconi at CINECA (Italy). ORB5 performance is shown to scale up to the full machine size, which represents more than 24000 GPUs on Summit. The usage of GPUs on Piz Daint brings about a factor 4 speed-up with respect to best CPU-only performance, which is itself about 2 times faster than the pre-refactored version of the code. An alternative to the standard PIC approach is also proposed: the Particle-In-Fourier (PIF). It is shown to be particularly attractive to reduce the cost of simulations involving a low number of Fourier modes, such as linear studies or basic nonlinear mode coupling processes. However, PIC method remains more efficient when the full nonlinear spectrum is involved.

Details

Actions

Preview