Generative Scenario Design
Proposing candidate operating scenarios with learned models, then screening them with physics.
What a scenario is
A plasma scenario is a plan for a discharge: the time histories of current, heating, fueling, and shape that take the plasma from start-up through a target operating point and back down. Designing scenarios that reach performance targets while staying stable is a hard, high-dimensional search.
Where generation helps
Generative and optimization-based methods can propose many candidate scenarios quickly, sampling the space of plausible actuator plans rather than hand-crafting each. This widens the search beyond what a human designer would try, surfacing non-obvious candidates.
- Sample actuator waveforms from a learned distribution
- Optimize toward targets with surrogate-in-the-loop search
- Interpolate between known good scenarios in a latent space
Physics screening is mandatory
A proposed scenario is only a hypothesis. Each candidate must pass transport, stability, and engineering-limit checks before it means anything. The generative step is a proposer; the physics simulations are the judge. Skipping the screening turns a useful tool into a source of plausible-looking nonsense.
Constraints
Real scenarios must respect actuator limits, avoid known instability boundaries, and stay within safe operating regions. These constraints are built into the search so that proposals are feasible, not merely optimal on paper.
Design context
For the Kronos breeder design, scenario studies live in simulation: generated candidates are screened against the frozen physics basis before any is considered. Scenarios are design explorations, not operational plans for built hardware, and no performance is claimed beyond what the simulations support.