Solving the Poisson partial differential equation using vector space projection methods.

摘要

This research presents a new approach at solving the Poisson partial differential equation using Vector Space Projection (VSP) methods. The work attacks the Poisson equation as encountered in two-dimensional phase unwrapping problems, and in two-dimensional electrostatic problems. Algorithms are developed by first considering simple one-dimensional cases, and then extending them to two-dimensional problems.In the context of phase unwrapping of two-dimensional phase functions, we explore an approach to the unwrapping using a robust extrapolation-projection algorithm. The unwrapping is done iteratively by modification of the Gerchberg-Papoulis (GP) extrapolation algorithm, and the solution is refined by projecting onto the available global data. An important contribution to the extrapolation algorithm is the formulation of the algorithm with the relaxed bandwidth constraint, and the proof that such modified GP extrapolation algorithm still converges. It is also shown that the unwrapping problem is ill-posed in the VSP setting, and that the modified GP algorithm is the missing link to pushing the iterative algorithm out of the trap solution under certain conditions. Robustness of the algorithm is demonstrated through its performance in a noisy environment. Performance is demonstrated by applying it to phantom phase functions, as well as to the real phase functions. Results are compared to well known algorithms in literature. Unlike many existing unwrapping methods which perform unwrapping locally, this work approaches the unwrapping problem from a globally, and eliminates the need for guiding instruments, like quality maps. VSP algorithm also very effectively battles problems of shadowing and holes, where data is not available or is heavily corrupted.In solving the classical Poisson problems in electrostatics, we demonstrate the effectiveness and ease of implementation of the VSP methodology to solving the equation, as well as imposing of the boundary conditions. Through one- and two-dimensional problems we introduce the necessary mathematical machinery, and analyze the computational complexity, as well as the memory requirements. We also compare the VSP solutions to the solutions of the well established finite element methods and finite difference methods. It is shown that VSP produces equivalent results, but uses less memory and computational resources in some cases.