Second-Order Adjoint Sensitivity Analysis Methodology (2~(nd)I-ASAM) Applied to Quantification of Non-Gaussian Uncertainties in a Nonlinear Heat Conduction Benchmark

摘要

It is known (see, e.g., Ref. 1) that the second-order sensitivities have the following major impact on model response distributions: (i) they cause the mean (i.e., expectation) value of a response to differ from its computed value; (ii) they contribute second-order terms to the response variance and covariances; (iii) most importantly, they provide the largest contributions to causing asymmetries in the response distributions. In particular, they provide the leading contributions to the third-order response correlations ("skewness"), which are essential for quantifying any non-Gaussian features of the response distribution.