Exponential Spectral Scaling: Robust and Efficient Stellarator Boundary Optimization via Mode-Dependent Scaling
Byoungchan Jang \aff 1 \corresp, Rory Conlin \aff 1, Matt Landreman \aff 1 · 2025
Read Exponential Spectral Scaling: Robust and Efficient Stellarator Boundary Optimization via Mode-Dependent Scaling by Byoungchan Jang \aff 1 \corresp, Rory Conlin \aff 1, Matt Landreman \aff 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-16320.txt.
Introduction
Shaping the wall of a stellarator, a twisted fusion device, means tuning waves of many sizes at once, and the small waves drown in the large ones. Byoungchan Jang, Rory Conlin, and Matt Landreman scale each wave mode by an exponential factor tied to its size. That shrinks the spread of values by orders of magnitude, so one direct run can use the full spectrum.
Older practice staged the work mode by mode, which was slow and full of judgment calls. The new scaling cut run times by a factor of two to five in tests. It also made results less sensitive to the starting guess, across two symmetry types and two toolkits.
Catalog record
| CREATOR | Byoungchan Jang \aff 1 \corresp, Rory Conlin \aff 1, Matt Landreman \aff 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-16320
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Exponential Spectral Scaling: Robust and Efficient Stellarator Boundary Optimization via Mode-Dependent Scaling Byoungchan Jang \aff 1 \corresp, Rory Conlin \aff 1, Matt Landreman \aff 1 Abstract Stellarator boundary optimization faces a fundamental numerical challenge: the extreme disparity between low- and high-mode amplitudes creates an optimization landscape in which direct full-spectrum approaches typically converge to poor local minima. Traditionally, this challenge has been addressed through a computationally expensive, multi-step Fourier continuation, in which low Fourier modes are optimized first, followed by the gradual incorporation of higher modes. We present Exponential Spectral Scaling (ESS), a technique that applies a mode-dependent exponential scaling factor to each Fourier mode. Our primary implementation uses the L ∞ L_{\infty} norm to determine the scaling pattern, creating a square spectral decay profile that effectively reduces the dynamic range of optimization variables from 6–7 orders of magnitude to 2–3. This scaling aligns with the natural spectral decay of physically meaningful configurations and enables direct single-step optimization using the full-spectrum of boundary Fourier modes. ESS eliminates arbitrary staging decisions and reduces computation time by a factor of ∼ 2 – 5 \sim 2\text{\textendash}5 in benchmark cases. In addition to accelerating optimization, ESS improves robustness, reducing sensitivity to initial conditions, and increasing confidence in avoiding local optima. We demonstrate the effectiveness of ESS across both quasi-axisymmetric (QA) and quasi-helically symmetric (QH) configurations, using two distinct optimization toolkits: simsopt and desc . 1 Introduction Stellarators present a promising pathway to steady-state magnetic confinement fusion, offering inherent stability and low recirculating power requirements due to their reliance on externally generated magnetic fields. Unlike the axisymmetric design of tokamaks, stellarators are not constrained by continuous geometric symmetry. This freedom enables the engineering of complex, fully three-dimensional magnetic fields, broadening the range of possible configurations and allowing precise tailoring of magnetic geometry to enhance performance. While this geometric flexibility is a central advantage, it also renders confinement highly sensitive to shape; without deliberate and careful design, stellarators are susceptible to increased orbit losses and degraded transport. As a result, boundary-shape optimization plays a critical role in realizing the potential benefits of stellarator-based fusion devices. Stellarator optimization traditionally separates the problem into (i) optimizing the plasma boundary shape ( Nührenberg & Zille 1988 ; Grieger et al. 1992 ; Mynick 2006 ; Bader et al. 2019 ; Landreman & Paul 2022 ) and (ii) designing coils that reproduce that boundary shape ( Merkel 1987 ; Drevlak 1998 ; Boozer 2000 ; Pomphrey et al. 2001 ; Zhu et al. 2018 ) . We focus exclusively on the first stage. For the first stage, the conventional approach often includes Fourier continuation ( Landreman et al. 2021b ) , where low-order Fourier modes are optimized first before gradually introducing higher ones. While this staged approach aids convergence, it is computationally inefficient, often over-optimizes low modes, and depends on arbitrary hyperparameters that can create artificial convergence plateaus as shown by the green loss history from Figure 1 . In contrast, the full-spectrum approach refers to optimizing all Fourier modes simultaneously from the outset, without any staged mode introduction. Despite its conceptual simplicity, the full-spectrum approach is rarely adopted in practice because the high-dimensional parameter space can lead to slow convergence, poor optimization performance, and an increased likelihood of getting trapped in local minima as shown by the red loss history in Figure 1 . In this paper, we introduce Exponential Spectral Scaling (ESS) as a direct method to address the mode amplitude disparity (MAD). ESS applies a mode-specific exponential scaling that significantly reduces MAD and improves the overall optimization landscape. We evaluate ESS with different norm choices ( L_{1} , L_{2} , and L_{\infty} ), demonstrating robust convergence across various stellarator configurations. While this work focuses exclusively on local optimization, we expect ESS to be broadly compatible with global optimization frameworks and may improve their efficiency as well. In particular, because any global optimization algorithm requires box constraints, ESS also informs how to choose meaningful bounds in spectral space. As shown by the gold loss history line in Figure 1 (which will be described in more detail in Section 4.2 ), ESS consistently delivers faster and more reliable convergence within a general trust-region framework, while avoiding nonphysical outcomes such as self-intersecting surfaces. Figure 1 : QA (left) and QH (right) optimization loss history comparison and final 3D stellarator configuration. The red line shows the loss history with full-spectrum using Moore’s Jacobian scaling as described in Appendix B , the green line shows the loss history with Fourier continuation using Moore’s Jacobian scaling, and the gold line shows the loss history with full-spectrum using ESS. In various optimization settings, imbalances between low- and high-spectral content—formalized here as MAD—have been previously noted. In stellarator research, Fourier continuation ( Landreman et al. 2021b ) is typically used to mitigate such imbalance; penalization of high-mode-number amplitudes has also been used ( Singh et al. 2020 ) . In aerospace inverse design, gradient-limiting Sobolev filters or staged Fourier optimization is utilized to damp high-order shape perturbations and stabilize convergence ( Masters et al. 2017 ; Kedward et al. 2020 ) . Geophysical full-waveform inversion confronts analogous “cycle-skipping” local minima by introducing multiscale frequency sweeps and convex misfit functions ( Bunks et al. 1995 ; Sirgue & Pratt 2004 ; Pladys et al. 2021 ) . Within machine learning, spectral bias and F-principle reveal that networks learn low-frequency components first, inspiring frequency-aware loss functions and rescaling schemes ( Rahaman et al. 2019 ; Xu et al. 2019 ; Wang & Lai 2024 ) . These convergent insights across disciplines motivate our ESS as a principled physics-agnostic remedy for MAD. 2 Towards Reducing the Mode Amplitude Disparity 2.1 Spectral Representation of the Plasma Boundary Stellarator boundary shapes are typically represented using truncated double Fourier series in cylindrical coordinates. Most stellarator optimization to date has used the VMEC representation, in which R and Z are expanded as cosine and sine series in the form \cos(m\theta-nN_{\mathrm{FP}}\phi) . In this paper, we instead adopt the related but distinct DESC representation defined in Equations ( 1 )–( 2 ). \displaystyle R(\theta,\zeta) \displaystyle=\sum_{m=-M}^{M}\sum_{n=-N}^{N}R_{n}^{m}\,\mathcal{G}_{n}^{m}(\theta,\zeta), (1) \displaystyle Z(\theta,\zeta) \displaystyle=\sum_{m=-M}^{M}\sum_{n=-N}^{N}Z_{n}^{m}\,\mathcal{G}_{n}^{m}(\theta,\zeta), (2) where \theta\in[0,2\pi) , the normalized toroidal angle is defined as \zeta=\phi/N_{\mathrm{FP}}\in[0,2\pi) , and the basis functions \mathcal{G}_{n}^{m}(\theta,\zeta) are defined piecewise according to the signs of m and n : \mathcal{G}_{n}^{m}(\theta,\zeta)=\begin{cases}\cos(|m|\theta)\cos(|n|\zeta),&m,n\geq 0\\ \cos(|m|\theta)\sin(|n|\zeta),&m\geq 0,n<0\\ \sin(|m|\theta)\cos(|n|\zeta),&m<0,n\geq 0\\ \sin(|m|\theta)\sin(|n|\zeta),&m,n<0.\end{cases} (3) However, this representation leads to a wide variation in the magnitude of Fourier coefficients. This effect, which we refer to as mode amplitude disparity (MAD) , has significant consequences for optimization. 2.2 Numerical Challenges from Mode Amplitude Disparity MAD refers to the dynamic range among the Fourier coefficients used to represent stellarator boundaries. To quantify this imbalance, we define the MAD as \displaystyle\mathrm{MAD}=\frac{\max_{(m,n)}|A_{n}^{m}|}{\min_{(m,n)\,:\,A_{n}^{m}\neq 0}|A_{n}^{m}|}, (4) where A_{n}^{m} is either R^{m}_{n} or Z^{m}_{n} from ( 1 )-( 2 ). This metric captures the effective dynamic range across all modes, excluding identically vanishing coefficients, and highlights the challenges posed by extreme scale separation in spectral representations. This imbalance poses serious numerical challenges because optimization algorithms are sensitive to the scale of variables. When parameters differ significantly in magnitude, the resulting subproblems are poorly conditioned, leading to instability, slow convergence, or convergence to suboptimal minima (