Machine-Learned Surrogate for Tritium Breeding
A focused surrogate predicts the tritium breeding ratio from blanket design parameters, accelerating the search for a self-sufficient fuel cycle.
Tritium self-sufficiency
A deuterium-tritium device consumes tritium that is scarce in nature, so it must breed its own by capturing fusion neutrons in a lithium-bearing blanket. The tritium breeding ratio (TBR) is tritium bred per tritium burned; a value above one, with margin, is required for a closed fuel cycle.
What drives TBR
- Lithium-6 enrichment in the breeder material
- Neutron multiplier fraction, such as beryllium or lead
- Blanket thickness and coverage fraction
- Structural material and coolant, which parasitically absorb neutrons
Each of these interacts nonlinearly, and evaluating a configuration requires a neutron-transport simulation. A surrogate trained on many such simulations learns the TBR response surface, letting designers see instantly how enrichment or multiplier fraction trades against coverage.
Optimization use
With a fast TBR surrogate, gradient-free optimizers or Bayesian optimization can search the blanket design space for configurations that reach a target with adequate margin. The Hyperion breeder concept uses a design target TBR of 1.8, and a surrogate makes exploring the many-parameter path to such a value practical.
Reconciliation and honesty
A single design-point TBR figure can hide distinctions between local and net, three-dimensional coverage effects, and penetrations for heating and diagnostics that lower the achievable value. Surrogate predictions must be reconciled with full three-dimensional transport that includes these penetrations before the number is trusted.
The 1.8 target is a simulation result for a concept under design, not a measured breeding rate. Final TBR is always confirmed with reference neutronics, and no fuel-cycle performance is claimed for hardware that has not been built.