Multilevel Based All-At-Once Methods in PDE Constrained Optimization with Applications to Shape Optimization of Active Microfluidic Biochips

This project within the area of PDE constrained optimization focuses on the development, analysis and implementation of optimization algorithms that combine efficient solution techniques from the numerics of PDEs, namely multilevel iterative solvers, and state-of-the-art optimization approaches, the so-called `all-at-once' optimization methods. It is well-known that multilevel techniques provide efficient PDE solvers of optimal algorithmic complexity. On the other hand, optimization methods within the all-at-once approach, such as sequential quadratic programming (SQP) methods and primal-dual Newton interior-point methods, have the appealing feature that in contrast to more traditional approaches, the numerical solution of the state equations is an integral part of the optimization routine. This is realized by incorporating the PDEs as constraints into the optimization routine. These strategies allow to save a considerable amount of computational work compared to methods that treat the PDE solution as an implicit function of the control/design variables. Moreover, the proper combination of multilevel techniques and optimization algorithms makes it possible to extract essential structural information from the originally infinite dimensional optimization problem. This can not be done with respect to a single grid. We aim to develop and analyze multilevel preconditioners for optimization subproblems arising in SQP and primal-dual Newton interior-point methods including strategies to control the level of inexactness allowable in optimization subproblems, when using iterative subproblem solvers. Moreover, we will investigate strategies to use multilevel methods for detection of negative curvature and in path following methods.