Project Details
Projekt Print View

Localized Reduced Basis Methods for PDE-constrained Parameter Optimization

Subject Area Mathematics
Term from 2018 to 2023
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 415818537
 
Final Report Year 2024

Final Report Abstract

Within this very successful project we developed and fully analyzed a novel Reduced Basis Trust Region (RB-TR) model reduction framework for large scale and multiscale PDE-constrained optimization and inverse parameter identification problems. The framework comes with a full a-posteriori error estimation for the approximation errors in the underlying primal and dual problems as well as for the corresponding sensitivities, the approximated objective functional and its gradient with respect to the parameters. Moreover convergence of the overall iterative methods to accumulation points could be shown, including the situation of box constraints for the parameters. Concerning localization for large scale or multiscale forward problems, we investigated two approaches. The first approach is particularly suited for multiscale problems and is based on a novel reduced basis approximation of the local orthogonal decomposition (LOD) method, while the second approach, based on the localized reduced basis method in a discontinuous Galerkin setting is more general and therefore also applicable to wider classes of large scale problems. Moreover, we extended and applied the novel approaches also to parabolic forward operators, as well as elliptic-parabolic systems. Finally, we also addressed elliptic inverse parameter identification problems, where in addition to the primal and dual states, also the parameter space is infinite or very high dimensional and needs to be reduced. To this end we suggested a new strategy for combined parameter and space reduction where the adaptive enrichment of the reduced parameter space is naturally inherited from the Tikhonov regularization within an iteratively regularized Gauß-Newton method. The scientific output of this project is considerably large which is also documented by the number and quality of the resulting publications. We emphasize that the results of this project evidently show, how the research achievements profited from the complementary expertise that was brought in from the participating PIs.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung