Machine-Learned Surrogate for RF Heating
Surrogates approximate radio-frequency wave heating and current drive, replacing slow wave-propagation solvers in scenario and control loops.
Radio-frequency heating
Radio-frequency (RF) systems launch electromagnetic waves that deposit energy at resonances in the plasma, heating ions or electrons and driving current. Computing where a wave deposits requires solving wave propagation and absorption through an inhomogeneous plasma, which is expensive for repeated evaluation.
Surrogate scope
- Deposition location and profile for the launched frequency and geometry
- Absorbed power split between species
- Driven current from the wave
- Sensitivity to density and temperature profiles
A surrogate trained on wave-solver outputs maps launcher and plasma settings to these quantities in milliseconds. This enables scenario scans over frequency, launch angle, and power, and supports real-time control where the full solver would be too slow.
Resonance sensitivity
RF deposition is sharply sensitive to resonance locations, which shift with the magnetic field and plasma profiles. A surrogate must resolve this sensitivity, sampling densely near resonances, or it will misplace the deposition. This is the main modeling challenge and the focus of validation.
Integration
Combined with beam and transport surrogates, an RF surrogate completes a fast actuator model for integrated scenario design. It lets designers balance RF and beam heating to shape profiles, and lets controllers predict the effect of RF commands before applying them.
RF heating is part of the heating mix considered for breeder concepts like Hyperion. In Kronos design work these surrogates are developed against wave solvers ahead of construction; they model deposition in a plasma that does not yet exist, and every result is a computational estimate to be validated on hardware.