AbstractsPhysics

The spatial resolution of an astronomical telescope is limited by either by the diffraction of light and the amount of aberrations caused in the atmosphere. To increase the diffraction limited resolution, the size of telescopes increases such as the to be build European Extremely Large Telescope (E-ELT). To fully benefit from the increased diffraction limited resolution, also the ability of adaptive optics to compensate for the wavefront aberrations should increase. This is for example done by increase the number of spatial wavefront measurements. This demands algorithms that are computationally efficient and highly parallelizable. In this thesis it is discussed whether the recently introduced Spline based ABeration REconstruction (SABRE) method can be parallelized. It is discussed that SABRE can be solved with the null-space method, resulting in a system of equations similar to the Poisson equation. The Conjugate Gradient (CG) method with different preconditioners, namely the Approximate INVerse (AINV), Algebraic Multigrid (AMG) and Incomplete Cholesky factorization (ICHOL), are used that make use of the sparsity of SABRE are discussed, their scalability and parallelism seem to be promising but the high amount of communication lowers the expectations. Experiments on the Graphics Processing Unit (GPU) give results, that matches with these expectations. Although there is plenty of parallelism, the sparse solvers are limited by the communication and therefore, for the grid sizes discussed in this thesis, slower than their counter ones on the Central Processing Unit (CPU). But there is lot of optimization possible and especially the CG with the AINV has a lot of potential. Further research is required.