Verification and Validation in Computer Simulation

Verification and validation (V&V) are complementary processes used to determine whether a computer simulation is implemented correctly and whether it represents the real-world system or phenomenon of interest. Verification addresses whether the code and numerical methods solve the governing equations and meet specified requirements, while validation assesses whether model outputs agree with reference data within stated tolerances. Together, V&V support credible use of simulations in engineering, scientific research, and high-stakes decision-making.

Overview

In many simulation workflows, developers and analysts build models that approximate physical systems such as fluid flow, structural response, or climate dynamics. The term verification is often associated with checking that a model’s implementation corresponds to its specifications. In contrast, validation focuses on whether the simulation is an adequate representation of the intended use case, frequently using comparisons to experiments, observational data, or trusted high-fidelity references such as benchmarks.

A common distinction is between solving the equations you think you specified and solving the real equations of the system you want to represent. This distinction is closely related to the broader ideas of scientific modeling and the credibility of computational results emphasized in communities such as computational science. In practice, V&V is supported by uncertainty quantification, because both numerical error and parameter uncertainty can influence whether validation claims are justified.

Verification

Verification evaluates whether the simulation is correct relative to the model description and requirements. A widely used framing divides verification into activities such as requirements verification, code verification, and numerical verification. Code verification often includes review and testing of software components, including unit testing, to ensure that algorithms are implemented as intended.

Numerical verification is concerned with discretization and solver behavior. For example, analysts may study grid or mesh refinement to demonstrate convergence, and they may check conservation properties and asymptotic behavior expected from the governing equations. Many organizations use systematic documentation and traceability to requirements to support auditability, especially when simulations are used for design decisions under regulatory scrutiny, as in safety engineering practices discussed in systems engineering.

Validation

Validation establishes whether the simulation is an adequate model for its intended purpose. This can involve comparing predicted outputs to experimental results, in which case the reference data may include measurements from laboratory tests, field experiments, or high-quality datasets. When the simulation includes uncertain or unknown parameters, validation may be accompanied by calibration, where parameters are adjusted within plausible bounds to improve agreement with observations.

Because validation is not a single pass/fail test, analyses frequently report metrics such as bias, error distributions, and confidence intervals. The strength of a validation claim depends on factors including the representativeness of test cases, the quality and uncertainty of measurement data, and the alignment of the model’s assumptions with the validation scenario. In some contexts, validation is discussed alongside the concept of model credibility and the appropriate use of computational models in decision processes that involve risk assessment.

Verification and validation process in practice

In a typical engineering or research setting, V&V is integrated with model development and study design. Analysts define the intended use and the model scope, identify requirements, and create test plans that cover both normal and edge cases. Verification evidence often includes code checks, solver verification studies, and traceability from requirements to implementation. Validation evidence often includes a documented comparison to relevant data, along with explicit statements about tolerances and limitations.

A practical approach is to use a structured workflow aligned with established guidance for simulation credibility. For example, standards and best-practice frameworks may specify documentation expectations, assessment of uncertainty, and the use of independent verification steps. Organizations in aerospace and defense have emphasized simulation credibility and V&V practices in publications related to simulation, and many teams incorporate these ideas into model governance and lifecycle management, similar to practices in software verification and validation.

Challenges and extensions

V&V can be difficult when simulations involve multiscale physics, complex material behavior, or models with many tunable parameters. In such cases, equifinality (multiple parameter sets producing similar outputs) can weaken simple validation conclusions. Moreover, model-form uncertainty—uncertainty about whether the governing model structure is appropriate—often cannot be resolved through code-level verification alone.

To address these challenges, practitioners increasingly combine V&V with model calibration and formal treatment of uncertainties via Bayesian inference or other statistical methods. Some workflows also emphasize verification of numerical approximations and quantification of numerical error using tools like method-of-manufactured-solutions approaches, and they may adopt more comprehensive approaches to credibility when simulations are used for safety-critical decisions. These extensions aim to ensure that validation claims remain consistent with both the numerical limitations and the uncertainty in observational references.

See also

• Software verification and validation
• Uncertainty quantification
• Model calibration
• Computational science
• Model validation

Categories: Verification, Validation, Computer simulation, Software testing, Numerical analysis