Techniques for validating uncertainty analyses in large computer codes

The uncertainty in the output of large computer codes provides an important measure of the credibility of the code. In applications where safety is a concern, an accurate assessment of output uncertainty will allow a rational determination of margin. Recent uncertainty analyses of large computer codes have begun to use the response surface method and second order error propagation methods to determine output uncertainty. However, the accuracy of these methods cannot be guaranteed. Certain basic assumptions are implied, and, when these assumptions fail to hold, erroneous results can occur. The paper discusses techniques for validating the results of uncertainty analyses based on these methods