Machine Learning for Impurity Transport
Surrogates predict how impurities move and accumulate in the plasma, informing radiation control and avoidance of performance-limiting buildup.
Impurities matter both ways
Impurity ions enter the plasma from the wall or by deliberate seeding. In the edge they radiate power usefully to protect the divertor, but in the core they dilute the fuel and radiate away energy, degrading performance. Predicting their transport is therefore central to both protection and performance.
The transport problem
Impurity transport combines classical, neoclassical, and turbulent processes, with a diffusive part and a convective pinch that can drive impurities inward or outward. Computing these from first principles is expensive, so learned surrogates trained on transport-code outputs provide fast estimates of impurity diffusivity and pinch.
- Inputs: plasma gradients, collisionality, impurity charge and mass
- Outputs: impurity diffusion coefficient and convective velocity
- Use: predict core accumulation and edge radiation distribution
Control relevance
Avoiding central impurity accumulation, which can trigger radiative collapse, requires knowing when conditions favor an inward pinch. A fast surrogate lets a controller anticipate accumulation and adjust heating or fueling, while edge seeding for divertor protection is tuned to radiate where wanted without contaminating the core.
Limits
Impurity transport is sensitive to the turbulence regime and to the specific impurity species, so surrogates must span the relevant conditions and be validated per species. High-charge impurities are especially consequential and warrant careful modeling and cross-checks.
Impurity control protects performance and components in any breeder concept, including Hyperion. In Kronos design work these surrogates are developed against simulated conditions ahead of construction; they estimate transport in modeled plasmas because the machine is not yet built, and results are computational.