🔬 physics

Towards joint optimization of stellarator coils and support structures

This paper introduces coil-fem, an open-source tool that integrates differentiable finite element analysis into stellarator coil optimization to simultaneously minimize magnetic field errors and mechanical stresses by jointly optimizing coil geometry and support structure placement.

Original authors: Lanke Fu, Alan A. Kaptanoglu

Published 2026-07-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lanke Fu, Alan A. Kaptanoglu

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to build a giant, complex sculpture out of very brittle, super-strong glass. This sculpture needs to hold a powerful, invisible magnetic force in place to keep a star-like fire burning inside it. This is the challenge of building a stellarator, a type of nuclear fusion reactor.

The "glass" in this story is the superconducting coil, and the invisible force is the magnetic field.

The Problem: The "Hand-Placed" Support

In the past, engineers designed these glass coils to be perfect at holding the magnetic field. Once the shape was finalized, they would try to figure out where to put support structures (like metal clamps or cages) to keep the glass from shattering under its own weight and the massive magnetic forces pushing and pulling on it.

Think of it like this: You design a beautiful, twisting glass bridge. Only after the design is finished do you try to figure out where to put the steel pillars to hold it up. If the pillars are in the wrong spot, the glass cracks. So, engineers have to move the pillars by hand, run a computer simulation to see if it holds, move them again, and repeat this dozens of times. It's slow, tedious, and often leads to a design that isn't as good as it could be.

The Solution: "Coil-Fem"

The authors of this paper, Lanke Fu and Alan Kaptanoglu, created a new software tool called coil-fem.

Instead of designing the bridge and then looking for pillars, coil-fem designs the bridge and the pillars at the same time.

Here is how it works, using a simple analogy:
Imagine you are trying to balance a heavy, wobbly stack of books on your hand.

Old Way: You arrange the books perfectly. Then you try to place your fingers (the supports) underneath them to stop them from falling. You keep moving your fingers until the stack is stable.
New Way (Coil-Fem): You move the books and your fingers simultaneously. The software instantly knows that if you move a finger slightly to the left, the books need to shift slightly to the right to stay balanced. It does this math in a split second, finding the perfect combination of book shape and finger placement that keeps the stack stable and the books from breaking.

What They Actually Did

The paper describes a "proof of concept" test using a simplified version of a real machine called W7-X.

The Test: They took a standard coil design and asked the software to find the best place to put the support clamps to reduce stress (the "cracking" force).
The Result: When they let the software move the clamps to the perfect spots, the stress on the coil dropped by 2.4 times compared to the original design where the clamps were just placed at the top and bottom.
The Surprise: They found that simply changing the shape of the coil (without moving the clamps) didn't help much. But moving the clamps to the right spots made a huge difference. This proves that you can't just design the coil; you have to design the support at the same time.

Why This Matters

The paper claims that this tool allows engineers to:

Save Time: No more manually moving clamps and re-running simulations for days.
Save Money: A less stressed coil might be cheaper to build because it doesn't need to be over-engineered.
Protect the Material: The coils use a special material called High-Temperature Superconductor (HTS) which is very brittle. If the stress is too high, it breaks. This tool helps keep that stress low.

What It Is NOT (Yet)

The paper is careful to say this is a first step.

It is not a finished product for building a real power plant today.
It currently uses simplified models (like treating the support as a simple "spring" rather than a complex metal beam).
It doesn't yet account for every single detail of heat and cooling, though it does account for the big picture of temperature changes.

In short, the paper introduces a new "smart assistant" that helps engineers design fusion reactors by figuring out the shape of the coils and the location of their supports simultaneously, rather than doing it one step at a time. This leads to stronger, safer, and more efficient designs.

Technical Summary: Towards Joint Optimization of Stellarator Coils and Support Structures

Problem Statement
The design of stellarator fusion reactors faces a critical challenge: balancing the need for an accurate magnetic field for plasma confinement with the mechanical requirements of protecting brittle high-temperature superconductor (HTS) conductors from enormous Lorentz forces and thermal stresses. Currently, the design process is sequential and labor-intensive. Coil geometries are optimized for magnetic performance in isolation, and support structures are subsequently added manually through iterative finite element analysis (FEA) to meet engineering stress limits. This "post-processing" approach fails to account for how support locations influence coil stress and how coil deformation under load introduces magnetic field errors. Existing optimization tools either rely on simplified analytic proxies that ignore support structures or use closed-source commercial FEA tools that cannot interface with gradient-based coil optimization loops.

Methodology
The authors present coil-fem, an open-source, differentiable structural mechanics tool integrated into the stellarator coil optimization loop. Built upon the JAX-FEM library, coil-fem enables the simultaneous optimization of coil geometry and support structure locations using gradient-based methods.

Key methodological components include:

Differentiable FEA: The tool solves the linear elasticity equation $\nabla \cdot \sigma(u) + F_{body} = 0$ with Robin (spring foundation) boundary conditions. This approximation replaces fixed Dirichlet boundary conditions, allowing the support clamp locations to be treated as optimizable variables without altering the sparse matrix structure of the FEA system.
Physics Model: The model accounts for body forces including gravity, mutual magnetic forces (modeled as 1D filaments), and self-forces (calculated via the Landreman-Hurwitz formula for rectangular cross-sections). It also incorporates thermal strain resulting from the $\sim$ 200K temperature drop from the stress-free state to operation.
Meshing and Geometry: Coils are modeled as finite-build bodies generated by sweeping a rectangular cross-section along a centerline using a rotation-minimizing frame. The mesh topology is fixed during optimization to ensure differentiability and computational efficiency.
Optimization Loop: The authors formulate a multi-objective penalty function that includes magnetic field error, mechanical load (volume integral of squared von Mises stress), and standard geometric constraints (coil length, spacing, curvature). The L-BFGS algorithm is used to minimize this objective with respect to both coil Fourier coefficients and support clamp parametric locations.

Key Contributions

First Combined Optimization: This paper presents the first instance in the literature of a gradient-driven, combined optimization of stellarator coil geometry and support structure locations.
coil-fem Tool: The release of an open-source tool that integrates support differentiable FEA directly into the coil optimization loop, treating support clamp locations as optimizable variables.
Demonstration of Support Importance: The study provides numerical evidence that the placement of support structures is a critical factor in reducing mechanical stress, a factor often overlooked when supports are fixed or manually placed.

Results
The authors benchmarked coil-fem against the DOLFINx FEA suite and the commercial-style surrogate approach, demonstrating agreement within 1–5% for stress and displacement metrics. They then applied the tool to a simplified W7-X coil set under four scenarios:

Baseline: Original W7-X coils with fixed top/bottom clamps.
Case A (Optimized Supports): Fixed coil geometry, optimized clamp locations.
Case B (Optimized Coils): Optimized coil geometry, fixed clamp locations.
Case C (Joint Optimization): Both coil geometry and clamp locations optimized.
Case D (Force Minimization): Optimized coil geometry to minimize Lorentz force (standard proxy) with fixed clamps.

Key Findings:

Stress Reduction: Case C achieved a 2.4× reduction in RMS von Mises stress compared to the baseline.
Role of Supports: Optimizing only the coil geometry with fixed supports (Case B) yielded only marginal improvements in stress. Conversely, optimizing only the support locations (Case A) significantly reduced stress, confirming that support placement is a dominant factor in structural mechanics.
Limitations of Proxies: Case D demonstrated that minimizing Lorentz forces (a common proxy) does not necessarily correlate with minimizing von Mises stress, justifying the need for integrated structural FEA.
Thermal Effects: While Lorentz forces are the dominant stressor, thermal contraction significantly impacts the maximum stress (shifting the peak by ~17%), highlighting the necessity of including thermal loads in the design loop.

Significance and Claims
The paper claims that coil-fem significantly accelerates the iteration rate for stellarator design by eliminating the manual, trial-and-error process of placing support structures. By enabling the simultaneous reduction of magnetic field errors and mechanical stresses, the tool offers a pathway to more economical and robust fusion reactor designs. The authors position this work as a "proof of concept" that establishes the feasibility of differentiable structural mechanics in fusion coil design. They explicitly state that while the current model uses simplifying assumptions (e.g., isotropic materials, cage-type supports modeled as springs), the framework allows for future extensions to higher-fidelity thermo-elastic models and complex support structures (e.g., beam networks or shells). The work does not claim to have solved all stellarator mechanical challenges but rather provides the first computational framework to jointly optimize coils and their supports.