Abstract (deu)
As the title predicts, in this masterthesis we are looking for a method which provides enclosures for solutions of overdetermined linear systems of equations, allowing for inaccuracies in the input data, i.e., for errors in the parameters. Assuming the system to have a solution, the (useable) system
with perturbated parameters is generally not solveable. Therefore we have to consider the least squares problem with those. Knowing bounds for the perturbations, they can be translated into so-called hybrid norms. Using those and assuming exact arithmetic, we show that theoretic bounds can be computed for the solution by a reduced QR-decomposition.
Since we have to take into account roundoff errors in floating point arithmetic, we need stronger tools for enclosure. Computing a QR-decomposition, basing on the Householder method, in a specific way, we can control these errors during the factorization, and combine them with the initial errors to hybrid norms, so that it will also be possible to obtain enclosures for the existing solutions.
In addition, Matlab code, which perform the upcoming concept, will be presented. Entering inaccurate parameters and bounds for the size of the perturbations provides an interval vector containg the solution of the overdetermined system.
Finally, we will analyze the algorithm and compare it to the evaluation of a (floating point) solution, using the Householder method.