Numerical approach for quantification of epistemic uncertainty (Q975137): Difference between revisions

The authors consider solutions \(u\) of an initial-boundary value problem for a partial differential equation which depends on a set \(Z\) of some random variables (parameters). A methodology is proposed for the analysis of uncertainty in the values of \(u\) at a given space-time point. The case is considered when the distribution of \(Z\) is unknown. At first a confidence domain \(I_X\) for \(Z\) must be identified to which \(Z\) belongs wit high probability. Then an approximate solution \(u_n\) is calculated on a discrete grid on \(I_X\) and interpolated over all \(I_X\), and the obtained solution is analyzed. In fact the authors concentrate on the accuracy of the approximation and interpolation discussing different approaches (such as sparse grid interpolation and Galerkin approximation). A damped linear oscillator equation and a homogeneous diffusion equation are considered as examples.