Verification, Validation, and Uncertainty Quantification
VVUQ answers three separate questions: is the code solved right, does the model match reality, and how confident are the results.
Three distinct questions
These three words are often run together but mean different things. Verification asks whether the code correctly solves the equations it claims to. Validation asks whether those equations describe the real world. Uncertainty quantification asks how much confidence the answer deserves. A result is only trustworthy when all three are addressed.
Verification
- Compare code output to analytic solutions where they exist.
- Check that refining the mesh converges at the expected rate.
- Confirm conservation laws and other invariants hold.
- Cross-check independent methods against each other.
Validation
Validation compares the model to experimental data, ideally data not used to build the model. Because Kronos machines are pre-construction, validation currently uses data from other devices and from irradiation tests, with the design's own data to come after first tritium around 2030. Stating that gap honestly is part of validation.
Uncertainty quantification
UQ propagates input, model, and numerical uncertainty to the output, producing distributions rather than point values, as described in uncertainty-driven decisions. Without UQ, verification and validation are incomplete.
For Kronos claims
The frozen numbers behind the Hyperion breeder and the burner are subjected to VVUQ, and the open reconciliations are exactly the places where validation or UQ is not yet complete, flagged rather than hidden.
Automation
Verification is enforced continuously by automated checks, so VVUQ is a standing discipline, not a one-time report.