GPU-Accelerated Structure-Preserving Finite Element Simulation of plasma physics
Dnr:
NAISS 2026/4-1662
Type:
NAISS Small
Principal Investigator:
Junjie Wen
Affiliation:
Uppsala universitet
Start Date:
2026-09-18
End Date:
2027-02-01
Primary Classification:
10105: Computational Mathematics
Webpage:
Allocation
- Arrhenius Disk at NAISS: 250 GiB
- Arrhenius GPU at NAISS: 200 GPU-h/month
- Arrhenius CPU at NAISS: 10 x 1000 core-h/month
Abstract
We simulate kinetic plasma dynamics via the Vlasov–Maxwell system using a novel second-order, positivity- and structure-preserving finite element method, which combines low-order graph viscosity with high-order residual-based convex limiting on a tensor-product phase-space discretization. The solver is implemented in C++/DOLFINx with MPI parallelization and GPU acceleration (CUDA-enabled PETSc). We request compute resources for production runs of plasma-physics benchmarks (two-stream, Weibel, Landau damping) in up to 2D2V/2D3V phase space, and for parallel scaling and GPU-performance studies.