3. How Much Verification Should I Do?

Engineering V&V practice draws a sharp line between code verification and solution (calculation) verification: the former proves that the program solves the stated equations correctly, while the latter measures the numerical error that still exists in a specific run of the model.

Standards such as ASME V&V 20 and NAFEMS guidance advise keeping a targeted suite of code-verification cases (typically analytic benchmarks or Method-of-Manufactured-Solutions (MMS) tests) and rerunning them after every major solver release, not after every minor operating-system patch.

MMS achieves this by adding an artificial source term to the governing Partial Differential Equations so that an exact closed-form answer is known, giving both vendors and users a rigorous yard-stick even for complex multiphysics codes.

Solution verification instead relies on mesh- and time-step-refinement studies or tools like the Grid Convergence Index (GCI) and a-posteriori error estimators to show that key quantities of interest have converged within an acceptable tolerance.

Because brute-force refinement is often prohibitive for large 3-D or transient models, commercial solvers supply global-norm and local error indicators, but analysts still carry the burden of interpreting these metrics and documenting any residual uncertainty.

Auditors and regulators increasingly demand this evidence, so aligning internal procedures and terminology with ASME, AIAA or NAFEMS recommendations helps forestall disputes over what level of verification is “good enough” during customer or certification reviews.