Verification and Validation
Two distinct checks: that a model is solved correctly, and that it represents reality.
Definition
Verification asks whether a model is implemented and solved correctly, that the equations are right and the numerics converge. Validation asks whether the model represents the real world, comparing its predictions against experiment or trusted reference data.
Uncertainty quantification completes the picture by attaching credible error bars to a simulation's predictions, distinguishing confident results from speculative ones. Together, verification, validation, and uncertainty quantification form the accepted framework for trustworthy computational science.
The two questions are distinct and both essential: verification asks whether the equations are solved correctly, validation whether the right equations were chosen, and passing one says nothing about the other. Uncertainty quantification completes the picture by attaching credible error bars to predictions. Together, this framework is what elevates a simulation from a suggestive picture to evidence, and it is the standard by which computational results earn the trust to inform real decisions.
The distinction
- Verification: are we solving the equations right?
- Validation: are we solving the right equations?
- Both are needed; passing one does not imply the other.
- Uncertainty quantification accompanies both.
Why it matters
A simulation can be perfectly verified yet invalid if its underlying model is wrong, or physically reasonable yet unverified if its numerics are flawed. Rigorous verification and validation is what turns a simulation from a plausible picture into credible evidence.
Fusion connection
Kronos treats verification and validation as central: simulation results are checked for numerical correctness and, where possible, against established physics, consistent with the program's insistence on claims that withstand scrutiny and its honest gating of what has and has not been demonstrated.