Verification, Validation, and Uncertainty Quantification
V&V and UQ establish whether a code solves its equations correctly, whether those equations describe reality, and how confident a prediction can be.
Three distinct questions
Verification asks whether the code correctly solves the equations it claims to solve. Validation asks whether those equations describe the physical world. Uncertainty quantification asks how much confidence to place in a prediction. These are separate questions, and a credible result addresses all three rather than conflating them.
Verification
- Compare against analytic solutions where they exist
- Check that numerical error shrinks at the expected rate as the mesh refines
- Use the method of manufactured solutions to test the full code path
- Cross-compare independent codes on identical problems
Validation
Validation compares code predictions against experimental measurements, ideally ones not used to build the model. It quantifies how well the model reproduces reality and, importantly, within what range of conditions. A model validated at one regime is not automatically valid when extrapolated far beyond it, a caution that matters for novel designs.
Uncertainty quantification
Predictions carry uncertainty from inputs, model form, and numerics. UQ propagates these through the calculation, for example by sampling uncertain inputs and observing the spread in outputs, to produce a result with error bars rather than a bare number. Reporting the uncertainty is part of reporting the result.
Honest reporting
The purpose of V&V and UQ is to state clearly what a code can and cannot be trusted to predict. A design claim backed by a verified, validated code with quantified uncertainty is far stronger than a single number with no stated confidence, and it is the standard responsible fusion analysis holds itself to.
Kronos frames its physics results with explicit gates and uncertainty rather than presenting single numbers as guaranteed.