Rochester Institute of Technology
Stochastic Inverse Problems in Stochastic Elliptic and Parabolic PDEs
Abstract
dc:description.abstractThis dissertation presents a computational framework for solving stochastic inverse problems in partial differential equations (PDEs) with random data. The objective is to recover unknown random coefficients in parabolic and elliptic diffusion equations, as well as nearly incompressible linear elasticity models. To achieve this, optimization-based approaches are used to minimize regularized Energy Least Squares (ELS) and Output Least Squares (OLS) objective functionals. Both direct and adjoint methods are used to estimate the gradient and Hessian of the objective functionals with respect to the coefficients. To avoid the expensive computation of the full Hessian matrix, the Newton-CG-like method is also adopted, which relies only on the Hessian-vector products. Finally, we develop Stochastic Galerkin discretizations for both the forward and inverse problems, providing discrete expressions for the objective functionals, their gradients, and their Hessian (and Hessian-vector product), and we finally provide numerical examples. We also present a stochastic auxiliary problem principle for solving variational inequalities. We establish its convergence under suitable assumptions and then apply it to estimate deterministic coefficients in PDEs with random data. The dissertation concludes by proposing future research in uncertainty quantification and the Calderón problem.
Degree
thesis:*- Name thesis:degree_name
- Mathematical Modeling (Ph.D)
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Mathematical Sciences, School of
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gong, Zi-Jia
- Contributors dc:contributor
-
- Akhtar A. Khan
Subjects
dc:subject × 4Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.rit.edu/theses/12724