edVirtus Reference

What Are Model Verification and Validation?

Model Verification and Validation: Building Confidence Without Claiming Perfection

Model verification and validation are complementary processes used to build justified confidence in a model. Verification asks whether the model has been implemented as intended. Validation asks whether its structure and behavior are sufficiently credible and useful for its stated purpose.

The distinction applies to numerical engineering models as well as behavioral and policy models. In RF exposure assessment, verification asks whether Maxwell's equations, source definitions, boundaries, meshes, solvers, and post-processing have been implemented and solved correctly. Validation asks whether the model represents the physical source, environment, human coupling, and dosimetric quantities adequately for the intended assessment.

Verification includes checking equations, units, initial conditions, numerical methods, data handling, software implementation, and correspondence between diagrams and code. Tests should include limiting and extreme cases whose correct behavior can be reasoned about independently.

RF numerical verification may include mesh or cell refinement, time-step and solver-convergence studies, energy-balance checks, boundary-condition tests, code-to-code comparison, and benchmark problems with analytical or well-established reference solutions. A stable calculation or plausible field plot is not sufficient if discretization, numerical dispersion, or an incorrect source model can materially change the result.

Validation examines whether model boundaries, causal relationships, decision rules, parameters, and generated behavior are consistent with relevant evidence. Sources may include measured data, established theory, documents, expert knowledge, stakeholder experience, and observations from comparable systems.

RF validation can compare modeled fields or absorption with calibrated measurements, standardized phantoms, physical source data, published interlaboratory results, or recognized validation procedures. The comparison must account for measurement uncertainty, model-input uncertainty, positioning tolerance, and the fact that agreement in one configuration does not automatically validate every frequency, geometry, or output quantity.

In system dynamics, validation is not a single statistical test. Structure tests, behavior-reproduction tests, sensitivity analysis, extreme-condition tests, and policy tests address different aspects of credibility. A model may reproduce historical data for the wrong structural reasons, so curve fitting alone is insufficient.

In both RF modeling and system dynamics, sensitivity analysis helps reveal which assumptions and inputs control the conclusion. For an RF model this may include source power and geometry, tissue properties, body position, mesh resolution, boundaries, and spatial averaging. Verification, validation, and uncertainty assessment should be reported together because each addresses a different reason why a result might be misleading.

The required evidence depends on purpose. A teaching model, an exploratory policy model, and a safety-critical engineering model demand different levels and forms of assurance. Limitations and unresolved uncertainties should be documented rather than hidden behind a claim that the model is simply valid.

Verification and validation continue throughout development. Failed tests reveal where assumptions, formulation, data, or implementation need improvement. The outcome is not proof of truth, but a transparent, purpose-specific case that the model is fit for its assigned decisions and learning tasks.