| name | computational-physics-validation |
| description | Verify computational physics software, quantify numerical uncertainty, and validate models against experiments. Use for convergence studies, manufactured solutions, conservation audits, code-to-code comparisons, calibration, uncertainty quantification, and simulation credibility. |
| license | MIT |
Computational Physics Validation
Separate the questions
- Code verification: is the discretized algorithm implemented correctly?
- Solution verification: how large is numerical error for this run?
- Validation: how well does the mathematical model represent the physical experiment for its intended use?
- Uncertainty quantification: how do input, numerical, and model uncertainties affect outputs?
Never use agreement with one experiment to replace code verification, or grid convergence to claim physical validity.
Build the evidence ladder
- Define quantities of interest, intended-use domain, acceptance criteria, and failure consequences.
- Test pure components, invariants, exact solutions, and method of manufactured solutions where possible.
- Vary mesh, time step, solver tolerance, precision, and stochastic sample count independently.
- Estimate observed order and numerical uncertainty; investigate non-asymptotic refinement behavior.
- Calibrate only designated parameters, then validate against separate measurements with uncertainty.
- Record code revision, inputs, environment, seeds, raw outputs, and post-processing.
Verify
- Audit conservation, symmetries, positivity, boundary conditions, and dimensional consistency.
- Compare with a simpler analytic limit and an independent implementation or formulation.
- Check sensitivity to uncertain inputs and numerical controls; identify dominant contributors.
- Compare residual patterns and quantities of interest with experimental uncertainty, not point values alone.
- Define the validated domain; treat extrapolation outside it as a new claim.
Deliver
Provide a credibility matrix separating implementation, numerical, input, experimental, and model-form evidence,
with pass/fail criteria, unresolved discrepancies, provenance, and intended-use limits.
Source basis
Original synthesis informed by NASA verification, validation, and uncertainty-quantification guidance and open
numerical-analysis sources recorded in ../../docs/TEXTBOOK_SOURCES.md.