Abstract of ELMU3 Project

Elmfire [1] is an electrostatic gyrokinetic particle code developed by TKK and VTT for simulating tokamak plasma microturbulence. Extensive work has been done to improve scaling of the code in collaboration with UNED, Åbo Akademi and CSC [2]. As a unique feature of Elmfire, the polarization operator is dynamically updated from the particle distribution for every time step. The computations associated with the polarization operator such as operator stacking are one of the most computationally intensive parts of the simulation, and as such, performance of interpolations and grid assignments are of crucial importance for performance both for the magnetic and electric fields. Currently a simplified grid structure (based on purely toroidal-parallel parametrization) is employed for the discretization of the electric potential, and the model needs to be made more flexible to be able to handle more complicated geometries frequently encountered in tokamak experiments. This requirement reduces computational effort by allowing a more refined grid to be used near the outer boundary of the simulation while increasing the grid spacing of the inner regions (where currently the coordinates converge). Also related to this are the improvements to be made with respect to numerical accuracy and generality of interpolations, especially due to conservation requirements of total energy and momentum. The Elmfire code participates in the ITM IMP#4 benchmarking and transport model development project, and as part of this project, improved diagnostics and CPO data structures are required.

[1] Heikkinen et al., J. Comp. Phys. 227 5582 (2008)
[2] Artur Signell et al., Comp. Phys. Comm. 179 5, 330-338 (2008)