Block-structured grids in full velocity space for Eulerian gyrokinetic simulations

Abstract

Global, i.e., full-torus, gyrokinetic simulations play an important role in exploring plasma microturbulence in magnetic fusion devices with strong radial variations. In the presence of steep temperature profiles, grid-based Eulerian approaches can become quite challenging as the correspondingly varying velocity space structures need to be accommodated and sufficiently resolved. A rigid velocity space grid then requires a very high number of discretization nodes resulting in enormous computational costs. To tackle this issue and reduce the computational demands, we introduce block-structured grids in the all velocity space dimensions. The construction of these grids is based on a general approach, making them suitable for various Eulerian implementations. In the current study, we explain the rationale behind the presented approach, detail the implementation, and provide simulation results obtained with the block-structured grids. The achieved reduction in the number of computational nodes depends on the temperature profile and simulation scenario provided. In the test cases at hand, about ten times fewer grid points are required for nonlinear simulations performed with block-structured grids in the plasma turbulence code GENE (http://genecode.org). With the speed-up found to scale almost exactly reciprocal to the number of grid points, the new implementation greatly reduces the computational costs and therefore opens new possibilities for applications of this software package.