An improved discrete least-squares/reduced-basis method for parameterized elliptic PDEs (Q2333686)

Computational efficiency of approximation of solutions of stochastic elliptic PDEs with random input data that has an affine dependence on the random variables are improved by incorporating a reduced-basis method into the discrete least-squares (DLS) framework. DLS approximation is determined by evaluating the reduced-basis approximation instead of the full finite element approximation, resulting in huge savings in both the storage of the DLS coefficients and the online cost of evaluating the DLS approximation. Numerical experiments demonstrate significant cost savings in both the offline and online phases of the DLS reduced basis method compared to that for the original DLS algorithm. As mentioned in the paper a the drawback of the approach is the slow decay of the Kolmogorov width of the PDE solution, requiring a large number of reduced basis functions in order to maintain the accuracy of reduced basis approximation. This could potentially make the DLS-RB method more expensive than the original DLS method. Furthermore, the quasi-optimal polynomial basis used in this paper only applies to the parametrized diffusion equation, for more complicated PDEs a different polynomial basis would have to be used.