Design-Space Exploration
Sampling many candidate designs to map how performance changes with parameters, before committing to any single point.
What it is
A fusion device has dozens of coupled design knobs: aspect ratio, field strength, plasma current, triangularity, wall standoff, blanket thickness. Each choice moves several outputs at once. Design-space exploration means running many parameter combinations through models to build a map of how objectives (gain, neutron load, tritium breeding) respond, rather than guessing one configuration and hoping.
Why sampling beats intuition
Coupled nonlinear systems have non-obvious optima. A field increase that helps confinement can raise stress on the magnet and shift the divertor heat load. Only by sweeping combinations do you see the trade surfaces. The output is not a single answer but a Pareto front: the set of designs where you cannot improve one objective without hurting another.
Sampling strategies
- Full-factorial grids: exhaustive but explode combinatorially with dimension
- Latin hypercube: space-filling, good coverage with few samples
- Sobol / quasi-random sequences: low-discrepancy, better than pseudo-random for integration
- Adaptive sampling: place new points where the model is uncertain or where the front is forming
For expensive simulations you cannot sample densely, so a surrogate (see Surrogate Models for Design) is fitted to cheap-to-evaluate approximate outputs, and the surrogate is searched instead of the full code.
Kronos use
The Hyperion breeder design point (Q 3.424, plasma current 9.86 MA, 16.84 T peak field, negative triangularity of -0.30) was selected from a swept space, not chosen in isolation. Neighboring points are retained so that if one constraint tightens during engineering, the team already knows which direction on the front recovers margin. All of this is simulation and design work; no hardware net-gain is claimed before first-of-a-kind first tritium around 2030.
Design-space maps are living artifacts: as models improve, the sweep is re-run and the front updates. That discipline keeps the chosen point traceable to evidence.