Global Bayesian Stage-1 Optimization

The first step of stellarator design involves finding a suitable plasma target for fusion-capable confinement. In the past, several optimization methods have been used to search the space of possible stellarator configurations for plasmas with desirable physical properties, including stability, engineering constraints, and confinement-enhancing symmetries. However, the current state of this Stage-1 optimization is restricted to local optimizers which miss out on large swaths of the highly nonconvex, high dimensional, and highly anisotropic stellarator landscape. In this work, we implement a global stochastic optimization routine, TuRBO, in the stellarator optimization framework DESC. We exhibit the capability of TuRBO to solve for stellarator equilibria, optimize for geometry constraints, and target the important confinement metric of quasisymmetry.