Stellarator island divertor shape optimization for reduced peak heat fluxes

This paper presents an automated algorithm that leverages Bayesian optimization to efficiently design robust stellarator island divertors with significantly reduced peak heat fluxes, achieving a 95% reduction in computational cost compared to traditional parameter scans.

The Gist

Imagine you are trying to build a house that generates its own power by mimicking the sun (nuclear fusion). The biggest challenge isn't just making the fire; it's managing the heat. If the fire gets too hot in one spot, it will melt the floor.

In a fusion reactor called a stellarator, the "floor" that catches the excess heat is called a divertor. Think of the divertor as a specialized gutter system designed to catch the "rain" of super-hot particles flowing out of the reactor core.

Here is the problem: In current designs, this "rain" tends to pour down in a single, concentrated stream, like a firehose hitting a single spot on the gutter. This creates a "hotspot" that is so intense it could melt the metal (tungsten) used to build the gutter.

The Solution: A Smart, Shaped Gutter

The authors of this paper created a new, automated way to design these gutters (divertors) so that the heat is spread out evenly, like a gentle drizzle instead of a firehose.

Here is how they did it, using some simple analogies:

1. The "Two-Finger" Trick
Usually, designing a complex 3D shape for a reactor is like trying to sculpt a masterpiece while blindfolded, with thousands of variables to adjust.
The authors' algorithm is much simpler. It only needs two starting points (like placing two fingers on a map) on the edge of the magnetic field. Once you give it these two points, the computer automatically draws the rest of the gutter shape, curving it perfectly to catch the heat.

2. The "Ski Slope" Analogy
Imagine you are skiing down a mountain. If you hit a steep cliff straight on, you crash (high heat flux). But if the slope is gentle and angled just right, you glide smoothly, spreading your energy over a long distance.
The computer shapes the divertor plates like a perfectly angled ski slope. It ensures the magnetic "wind" hits the plate at a very shallow angle (about 3 degrees). This forces the heat to stretch out over a large area rather than concentrating in one spot.

3. The "Smart Search" (Bayesian Optimization)
Finding the perfect two starting points is like trying to find the lowest point in a foggy, mountainous valley.

• The Old Way (Grid Scan): You could walk every single step of the valley, checking the ground at every inch. This is accurate but takes forever and costs a lot of money (computing power).
• The New Way (Bayesian Optimization): The authors used a "smart search" algorithm. It's like having a GPS that learns from every step you take. If you go up a hill, it knows not to go that way next time. If you find a dip, it investigates that area more closely.
• The Result: This smart search found the best design 95% faster than the old "walk every step" method. It saved massive amounts of time and money.

4. The "Safety Net"
The computer also checks for "bad spots." Imagine if the gutter was shaped wrong and the rain splashed over the edge onto the house foundation (the reactor wall). The algorithm includes a "safety net" that penalizes designs where heat hits the wrong places, ensuring the solution is safe for engineers to build.

Why This Matters

This paper is a "proof of concept." It shows that we can use smart computer algorithms to design fusion reactors that don't melt under their own heat.

• Before: We guessed the shape of the heat catcher, hoping it wouldn't melt.
• Now: We can mathematically calculate the perfect shape to spread the heat out, making fusion power plants much more feasible and safe.

In short, the authors built a digital architect that designs a heat-dissipating gutter for a star-in-a-box, ensuring the heat is spread out like a warm blanket rather than a burning laser, all while doing the math 20 times faster than before.

Technical Summary

1. Problem Statement

In the design of future Fusion Power Plants (FPPs), particularly for stellarators, managing the intense heat fluxes transported to the plasma edge is a critical engineering challenge. Current divertor materials (e.g., tungsten) have a melting limit of approximately 10 MW/m². Exceeding this threshold causes material degradation, cracking, and recrystallization.

While stellarator island divertors offer advantages due to long magnetic connection lengths that promote perpendicular heat diffusion, the specific geometry of the divertor plates significantly impacts the peak heat load. Existing optimization methods for tokamaks exist, but there is a lack of automated, efficient tools to optimize the shape of stellarator island divertors to minimize peak heat fluxes while adhering to engineering constraints (such as strike angle limits). The goal is to find an optimal divertor geometry that spreads heat loads effectively without requiring computationally prohibitive simulations.

2. Methodology

The authors developed a fully automated workflow combining a geometric construction algorithm with a Bayesian optimization loop, utilizing the FLARE code for heat flux approximation.

A. Divertor-Making Algorithm

The core of the approach is a low-dimensional parametrization algorithm that constructs divertor plates based on a fixed magnetic equilibrium.

• Inputs: The algorithm requires only two initial coordinates ($p_{L0}, p_{R0}$) on the island separatrix (or a control surface near it) at a specific toroidal angle ($\phi_{init}$).
• Geometry Construction:
  • The algorithm generates a V-shaped divertor where the tip is closest to the core plasma, and the plates slope outward toward the vessel wall.
  • It iteratively steps through toroidal angles ($\phi$). At each step, it searches for new points on the separatrix such that the magnetic field vector makes a specific strike angle ($\alpha$) with the vector connecting the previous and current points.
  • This enforces a maximum strike angle constraint (set to 3° in this study) to maximize the surface area exposed to the plasma, thereby spreading the heat flux.
  • The output is a set of divertor plate files in Kisslinger format.
• Cost Function: To evaluate performance, a cost function $J$ is defined:
  $$J(\theta_L, \theta_R) = W_1 q_{peak} + W_2 (\text{end-plate \%})$$
  • $q_{peak}$: Maximum heat flux on the divertor.
  • End-plate %: The percentage of field lines hitting "undesirable" regions (e.g., the leading edge or behind the plate), which serves as a proxy for engineering viability and self-shadowing.

B. Heat Flux Simulation (FLARE)

• The FLARE code is used to approximate heat loads using a field line diffusion model.
• Instead of full kinetic simulations, it models heat transport via parallel streaming along field lines and perpendicular diffusive kicks (Monte-Carlo method).
• This approach is significantly faster (100–1000x) than full particle trajectory integration, making it suitable for an optimization loop.

C. Optimization Strategy

• Grid Scan: A brute-force 2D scan of the initial angle parameters ($\theta_L, \theta_R$) was performed to map the cost function landscape.
• Bayesian Optimization (BO): To overcome the high computational cost of the grid scan (approx. 5–10 mins/divertor), the authors employed Bayesian Optimization.
  • BO uses a Gaussian Process (with a Matérn kernel) to model the cost function.
  • It utilizes an Expected Improvement acquisition function to balance exploration and exploitation.
  • A soft penalty was applied to steer the optimizer away from parameter sets that caused the divertor algorithm to crash (non-convergence).

3. Key Results

A. Optimization Efficiency

• Cost Reduction: The Bayesian optimization achieved a result nearly identical to the best solution found in the exhaustive grid scan but with a 95% reduction in computational cost (simulating far fewer divertor configurations).
• Optimal Parameters: The optimal divertor was found at initial angles $(\theta_L, \theta_R) \approx (4.05, 1.21)$.
• Performance Metrics:
  • Peak Heat Flux ($q_{peak}$): Reduced to ~3 MW/m² (well below the 10 MW/m² material limit).
  • End-plate Hits: Reduced to ~3.5%, indicating minimal self-shadowing or unwanted hits.

B. Pareto Frontier Analysis

By varying the weight ratio ($W_2/W_1$) in the cost function, the authors generated a Pareto frontier. This demonstrates the trade-off between minimizing peak heat flux and minimizing end-plate hits. The results show that achieving end-plate hit percentages below 5% requires a peak heat flux of at least 3 MW/m².

C. Robustness and Scaling

• Strike Angle: The optimized divertor maintained a fairly uniform strike angle with a mean of 1.56°, successfully enforcing the design constraint.
• Diffusivity Scaling: The authors tested the divertor's robustness by varying the cross-field diffusion coefficient ($D$).
  • The heat flux width ($\lambda_{q,t}$) scaled as $\sqrt{D}$ for low diffusivities, consistent with theoretical predictions.
  • At high diffusivities, the scaling broke down because a significant fraction of particles struck the vessel wall rather than the divertor, indicating a limit to the divertor's capture efficiency under extreme transport conditions.

4. Significance and Contributions

1. First Automated Optimization: This work represents the first step toward a comprehensive, automated optimization process for stellarator island divertors. It moves beyond manual tuning or semi-automated tilting to a fully automated search for optimal geometries.
2. Computational Efficiency: By combining a low-dimensional parametrization with Bayesian Optimization, the authors demonstrated a viable path to optimizing complex plasma-facing components without the prohibitive cost of full physics simulations.
3. Engineering-Physics Integration: The study successfully bridges the gap between magnetic equilibrium physics and engineering constraints (heat flux limits, strike angles, and self-shadowing).
4. Scalability: The methodology is designed to be extensible. Future iterations could include more degrees of freedom (e.g., optimizing the strike angle $\alpha$ or toroidal extent) and integrate with full-device optimization suites.

5. Limitations and Future Work

• Physics Simplification: The current model uses a field line diffusion proxy and does not include neutral transport, impurity dynamics, or detailed plasma detachment physics.
• Engineering Constraints: The cost function currently focuses on heat flux and geometry; it does not yet account for cooling channel placement, structural support, or manufacturing tolerances.
• Validation: While FLARE is efficient, the authors note that final designs should be validated with higher-fidelity codes like EMC3-EIRENE before construction.

In conclusion, this paper provides a robust framework for designing stellarator divertors that minimize peak heat loads, offering a critical tool for the development of viable stellarator fusion power plants.