Model Validation and Benchmarking
Validation checks a model against reality; benchmarking checks it against other trusted codes and known reference cases.
Is the model right?
Verification asks whether a code solves its equations correctly. Validation asks the harder question: do those equations describe the real world well enough for the intended use? Validation is settled only by comparison with experiment or with reality-grounded reference data.
Benchmarking against peers
Before or alongside experimental validation, a code is compared against other established codes on shared problems, and against analytic solutions where they exist. Agreement across independent implementations raises confidence; disagreement locates a bug or a modeling difference to investigate.
What a good benchmark set contains
- Analytic cases with exact known answers.
- Standard community reference problems.
- Cross-comparison with independently developed codes.
- Where available, experimental data with quantified uncertainty.
Validation is scoped
A model is never validated in the abstract, only for a specific range of conditions and a specific quantity. A code validated for one regime may be invalid in another. Stating the domain of validation is as important as reporting the agreement, because it defines where the model may be trusted.
Honesty about the gap
For machines not yet built, full experimental validation of end-to-end performance is not yet possible, which is why fusion programs are careful to distinguish validated component physics from projected integrated performance, and make no hardware net-gain claim before first-of-a-kind operation.