Skip to content
Technology How it works Breeder — Hyperion Burner — Aegis Burner — MetroVolt AI-Native Architecture Magnets Fuel cycle Safety Roadmap
Solutions AI & Data Centers Defense & Government Grid & Baseload Neutron Detection Quantum
Learn Technical Library
Proof Publications Whitepapers Technical Library Open Science & Reproducibility The Honest Gates
Company About / Mission Leadership Environment Health & Safety Investors Careers Press Contact
3D Model
AI Architecture › L0 · Foundation
L0 · Foundation

Surrogate Model Generation

Building fast emulators of expensive solvers so the twin can evaluate neutronics, transport, and equilibrium in real time.

THE STACK · click to jumpL7Ecosystem & StrategyL6Experience & VisualizationL5Applications & CopilotsL4OrchestrationL3Twin Modeling & AIL2Data FabricL1Control PlaneL0Foundation▲tlmctl▼L0 · FOUNDATIONThe offline compute substrate — multi-physics & batch training.1Cloud HPCelastic burst2Bare-Metal ClusterGPU / CPU3Supercomputingmulti-physics runs4Batch Trainingmodel builds5Simulation FarmGrad-Shafranov · MHD6Object StorecheckpointsMACHINE TIETrains the models that ship UP to L3 — no real-time path to the machine.KRONOS FUSION ENERGYAI-NATIVE S.M.A.R.T. GENERATORFOUNDATIONSHEET 02REV. 2026-08L0 · AI-NATIVE STACK
L0 · Foundation — its place in the stack (left, click any layer) and its internal components (right). Telemetry rises; control descends.

Emulating the expensive solvers

Many L0 solvers, gyrokinetic turbulence, Monte Carlo neutronics, coupled transport, are far too slow for the real-time twin. Surrogate model generation is the L0 process that replaces each with a fast emulator: a model trained to reproduce the solver's input-output mapping in a single cheap evaluation. Surrogates are what let the twin carry physics it could never solve live.

Design of experiments

A surrogate is only as good as the samples it learns from. Kronos uses design-of-experiments sampling, space-filling designs over the input parameters, to run the expensive solver at well-chosen points, then fits the surrogate to that data. The sampling strategy matters more than raw sample count, because coverage of the input space determines where the surrogate can be trusted.

Knowing the surrogate's limits

Every surrogate carries an error estimate and a domain of validity. Outside the sampled region, its output is extrapolation and must not drive control. Kronos records these bounds with the model so the twin knows when a prediction is inside its trained envelope and when it should defer or trigger a fresh solve. A surrogate that hides its uncertainty is dangerous.

Surrogates serve both machines and many modules. Breeder neutronics response tables, burner ambipolar-potential emulators, transport-coefficient surrogates from gyrokinetics, and equilibrium surrogates all follow the same generate-validate-bound pattern. The physics differs; the discipline of building a bounded, validated emulator does not.

Surrogate generation is coupled to retraining: as new solver runs accumulate, surrogates are refreshed and their validity domains extended. This is how the twin's fast physics stays anchored to the slow ground truth, and it is a central reason the offline foundation exists.

Content reviewed August 2026 · design-and-simulation stage