Frequently Asked Question
Verification asks whether the equations were solved correctly. It examines implementation and numerical approximation through analytical solutions, manufactured solutions, conservation checks, and mesh or time-step refinement. Validation asks whether the chosen physical model represents reality for the intended use, usually through comparison with measurements or trusted benchmark data.
| Question | Evidence |
|---|---|
| Verification | Does the discrete solution approach the intended mathematical solution? |
| Validation | Does the mathematical model reproduce the physical observation within stated uncertainty? |
A simulation can be verified but poorly validated if the model omits important physics. It can also compare well with one experiment while still containing large numerical errors that happen to cancel modeling errors.
Evidence and uncertainty
Verification asks whether the equations and computational implementation are solved correctly. Validation asks whether the model represents physical reality for the intended use. Uncertainty should distinguish numerical error, input uncertainty, model-form uncertainty, measurement uncertainty, and operating variation. A single residual or a single comparison does not establish credibility.
The root-sum-square form is appropriate only when contributors are on compatible bases and reasonably independent. Otherwise, correlation or conservative bounds are required.
Worked example
If numerical, input, and measurement uncertainties are 2%, 3%, and 4%, the independent estimate is √(22+32+42) = 5.39%. This is an uncertainty estimate, not validation by itself. The comparison must also use the same quantity, regime, boundary condition, and reference plane.
Check: report discrepancies, data quality, confidence basis, and the decision consequence of the remaining uncertainty.