Abstract Details
Abstracts
Author: Stefan Buller
Requested Type: Poster
Submitted: 2023-03-24 23:05:02
Co-authors: R. Gaur, P. Kim, R. Jorge, M. Landreman,
Contact Info:
University of Maryland
8223 Paint Branch Drive
College Park, Maryland 20740
USA
Abstract Text:
Stellarator optimization has been largely successful in producing magnetic geometries with reduced energy and particle losses due to neoclassical transport. In these optimized geometries, turbulence is expected to be the dominant transport mechanism. Indeed, in Wendelstein 7-X, the world's largest optimized stellerator, the achievable ion temperatures appear to be limited by turbulence due to ion-temperature-gradient (ITG) modes. Thus, we would like to account for turbulence when designing future stellarators.
Linear gyrokinetics flux-tube simulations can be run cheaply, and are thus suitable for inclusion in optimization loops. However, it is not certain that growth rates from linear simulations accurately reflect the saturated amplitudes in nonlinear simulations. Generally, lower growth rates are expected to correlate with lower mode amplitudes and lower transport for a given geometry, but since this correlation depends on geometry, this is not true when varying the geometry. This makes the linear growth rates an unreliably turbulence proxy for geometry optimizations.
To try to remedy this problem, we take inspiration from multi-fidelity optimization. We fit parameters of a low-fidelity model (here, a linear ITG gyrokinetics simulations) to match a high-fidelity model (a nonlinear simulation) in a neighborhood of a point (a magnetic geometry) in the optimization space. We investigate the size of the region of validity of low-fidelity models around various geometries. Optimization proceeds by iteratively finding new low-fidelity models as the optimization point strays from the point at which the model was fitted.
Another downside of stellarator flux-tube simulations is that they may not reflect the transport from the full flux-surface. We compare optimization results that were optimized on varying number of flux-tubes.
Comments: