Typical questions that V&V can answer …

• Why should we believe that the predictions of our numerical simulations are any better than crystal-ball reading?
• What is the prediction accuracy of the model, especially away from those settings that can be measured experimentally?
• What is the validation domain for a given application?
• Is the computer code free of programming mistakes?
• Does the computational mesh (or grid) provide converged solutions?
• Which feature of the response best provides physical insight about the phenomenon?
• Where is an observed variability coming from?
• Which parameters of the numerical simulation control the “spread” of output results?
• What is the effect of modeling uncertainty on the predictions?
• Can the physics-based simulation be replaced by a fast-running surrogate?
• How to meaningfully compare physical measurements to numerical predictions?
• Where is the modeling error coming from, and how can it be reduced?
• Which numerical modeling technique is better for a particular application?
• How robust are predictions to the modeling error?
• How to study the trade-offs between prediction accuracy and modeling lack-of-knowledge?