Stabilized and Adaptive Time-Stepping Schemes in GENE-X (Ibrahim Almuslimani, EPFL)

22.01.2026 11:00

We present recent developments in advanced time-stepping for GENE-X, a global, nonlinear, electromagnetic, and collisional full-f gyrokinetic code. Standard approaches such as RK4 or operator splitting can be inefficient, requiring overly small timesteps, spending more time per step, and/or introducing additional errors. To address this, we have implemented two new schemes: RK43, an adaptive Runge–Kutta scheme that dynamically adjusts the timestep based on local error estimates, and PIROCK (Partitioned Runge–Kutta–Chebyshev), which treats stiff diffusion terms appearing in the collision and neutrals operators with an explicit stabilized method while handling advection terms with standard Runge–Kutta steps. Importantly, PIROCK achieves this without relying on operator splitting, treating the full system within a unified framework and thereby requiring fewer function evaluations per timestep than splitting approaches. Preliminary tests indicate that both approaches significantly improve efficiency compared to existing schemes. We will also discuss ongoing optimization efforts and outline possible future directions for enhancing large-scale gyrokinetic simulations.

Lieu

Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique

entrée libre