Göttingen 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

P: Fachverband Plasmaphysik

P 17: Poster Session II

P 17.5: Poster

Mittwoch, 2. April 2025, 16:15–18:15, ZHG Foyer 1. OG

A novel optimization approach for stellarator design — •Issra Ali, Alan Goodman, and Ryan Wu — Max-Planck-Institut für Plasmaphysik, Greifswald, Germany

Stellarators are a class of fusion reactor candidates which confine a hot plasma in a toroidal shape using powerful magnetic fields. One of stellarators' great strengths is their vast design space, which offers incredible flexibility in their physical properties. This flexibility is exploited via numerical optimization to carefully mold magnetic field geometries and corresponding coil designs to maximize confinement quality.

Gradient-based optimization methods are the industry standard for stellarator and coil optimization problems, and have recently proven effective in finding reactor-relevant stellarator designs. However, these problems are often multimodal and highly sensitive to initial conditions, meaning that gradient-based methods are at best unlikely to find global minima, and at worst unable to converge altogether without careful selection of initial conditions and optimizer hyperparameters. Global optimization methods, such as Bayesian Optimization, may offer a better way to explore the entire parameter space, but tend to scale poorly to high-dimensional problems such as stellarator optimization. Moreover, they do not exploit gradient information, which is a powerful tool in high-dimensional optimization spaces. In this work, we present a novel global optimization method that seeks global optima while still being guided by gradients.

Keywords: Optimization; Stellarators