Adaptive finite element methods for parameter estimation problems in partial differential equations

Abstract: Physical and chemical phenomena are often described by a system of partial di®erential equations. These equations usually involve unknown parameters, which cannot be measured directly but which can be adjusted to make the model predictions match the observed data. The process of ¯tting these para- meters to laboratory or plant data is called parameter estimation. In order to recover these parameters, the well-known output least squares formulation is of- ten utilized. To solve the optimization problem governed by partial di®erential equations, the in¯nite-dimensional problem must be approximated by introduc- ing discretizations such as a ¯nite elements or di®erences. It is clear that the e±ciency of the numerical methods dealt with here will be in°uenced by the discretization scheme. The goal of this thesis is to develop e±cient numerical methods for the parameter estimation problems governed by partial di®erential equations, based on adaptive ¯nite element methods. This work was initiated by an investigation into an a posteriori error esti- mator of residual type for parameter estimation problems with a ¯nite number of unknown parameters. It appears that an adaptive ¯nite element algorithm guided by the derived a posteriori error estimator produces a sequence of eco- nomical, locally re¯ned meshes. The methods are then applied to the identi¯ca- tion of elastic constants in paper from measured displacements. Further, some a posteriori error estimators of gradient recovery type are derived for the error in parameters due to the discretization. The main advantages of using error estimators of this type are the simplicity of their implementation and their cost e®ectiveness. Often, the unknown parameters are functions, which also need to be dis- cretized. Adaptive ¯nite element method is developed for the estimation of distributed parameters in elliptic equations with multi-mesh techniques. Finally, a goal-oriented adaptive method, dual weighted residual methods (DWR methods) are employed determining the elastic constants in paper from measured displacements.