Concepts and Practice of Verification, Validation, and Uncertainty Quantification

Abstract:

Verification and validation (V&V) are the primary means to assess the numerical and physics modeling accuracy, respectively, in computational simulation. Code verification assesses the reliability of the software coding and the numerical algorithms used in obtaining a solution, while solution verification addresses numerical error estimation of the computational solution of a mathematical model for a specified set of initial and boundary conditions. Validation assesses the accuracy of the mathematical model as compared to experimentally measured response quantities of the system being modeled. As these experimental data are typically available only for simplified subsystems or components of the system, model validation commonly provides limited ability to assess model accuracy directly. Uncertainty quantification (UQ), specifically in regard to predictive capability of a mathematical model, attempts to characterize and estimate the total uncertainty for conditions where no experimental data are available. Specific sources of uncertainty that can impact the total predictive uncertainty are: the assumptions and approximations in the formulation of the mathematical model, the error incurred in the numerical solution of the discretized model, the information available for stochastic input data for the system, and the extrapolation of the mathematical model to conditions where no experimental data are available. This presentation will briefly discuss the principles and practices of VVUQ from both the perspective of computational modeling and simulation, as well as the difficult issue of estimating predictive capability. Contrasts will be drawn between weak and strong code verification testing, and model validation as opposed to model calibration. Closing remarks will address what needs to be done to improve the value of information generated by computational simulation for improved decision-making.