Planar Coil Optimization for the Eos Stellarator using Sparse Regression
A challenge in the design of stellarators for confining plasma at conditions relevant to fusion energy generation is designing a feasible set of magnetic field coils which can create the necessary confining field. One active direction of investigation involves the creation of a set of simplified planar coils to approximate the desired magnetic field, split between large plasma encircling coils which generate the majority of the field and small shaping coils near the plasma surface which correct remaining errors in the field geometry. The problem of optimizing currents in these coils is inherently ill-posed due to the Biot-Savart Law. In this work, the problem of optimizing the current distribution of the shaping coils is posed as a sparse regression to minimize coil count while targeting $B_n = \langle\mathbf{B} \cdot \mathbf{n}\rangle$, a metric of confinement quality, as a least-squares objective. The goal is to improve manufacturability and conductor efficiency by using fewer coils at higher currents. Using an ensemble of sparse optimization algorithms such as LASSO, relax-and-split, and heuristic methods, Pareto fronts between coil sparsity and $B_n$ properties are identified. Differences between the optimization algorithms are evaluated, demonstrating up to 20% reductions in mean $B_n$ at the same sparsity compared to a previously used heuristic method. The perturbation sensitivity of the sparse solutions to manufacturing misalignment is also evaluated. The results from the study demonstrate the potential and diversity of sparse optimization strategies in stellarator coil design problems.