Speaker
Description
An overview of the GRILLIX [1] code is presented, a three-dimensional turbulence simulation code for the self-consistent modeling of the plasma edge and scrape-off layer. The code is based on a global drift-reduced fluid model coupled to a three-moment fluid neutral model. It employs the Flux-Coordinate Independent (FCI) approach, enabling efficient treatment of complex magnetic geometries, and has recently been extended to non-axisymmetric stellarator configurations [2]. Key scientific results obtained with GRILLIX are summarized, including simulations of advanced confinement regimes such as the X-point radiator [3] and Quasi-Continuous Exhaust regime [4], along with their experimental validation, studies of confinement transitions (L–H) [5], and first simulations of island divertor configurations in Wendelstein 7-X.
We present the modeling, numerical, and technical developments required to obtain these results. On the modeling side, in particular the self-consistent treatment of electromagnetic effects and extensions to low-collisionality regimes proved essential. While the Flux-Coordinate Independent (FCI) approach provides an efficient framework for accurately capturing strongly anisotropic dynamics, it also introduces several challenges, including the treatment of boundary conditions, the coupling to neutral models, and the achievement of good parallel scalability. Furthermore, the HPC aspects of GRILLIX are discussed. Next-generation supercomputing systems increasingly require efficient execution on GPU-accelerated hardware. We therefore present the porting of the elliptic field solver to GPUs [6], which represents both the most algorithmically complex and the most computationally demanding component of GRILLIX, while also featuring a well-defined interface that enables modular porting. Significant performance improvements are achieved with the GPU-accelerated implementation across different architectures.
The presentation concludes with an overview of ongoing work on open problems. In particular, the breakdown of the fluid assumptions may necessitate a kinetic description, for which the GENE-X code is being developed in close collaboration with GRILLIX. To address the challenges associated with boundary conditions, we present an approach that departs from the currently employed immersed boundary method and offers the potential for a more accurate and flexible implementation of boundary conditions within the FCI framework. Finally, lessons learned from the porting of the elliptic field solver have important implications for the GPU porting of the remaining GRILLIX components, which will be discussed.
[1] A. Stegmeir, D. Coster, A. Ross et al., PPCF 60:035005 (2018).
[2] A. Stegmeir, M. Finkbeiner, C.Pitzal et al., CPC 318:109874 (2026).
[3] K. Eder, W. Zholobenko, A. Stegmeir et al., NF 65:096029 (2025).
[4] K. Zhang, W. Zholobenko, A. Stegmeir et al., PRL submitted (2026).
[5] W. Zholobenko, F. Jenko, K. Zhang et al., PRL 136:075101 (2026).
[6] A. Stegmeir, C. Lalescu, M. Lin et al., PASC26 (accepted) (2026)