In this thesis, we analyze this functional on a mathematical basis, including an introduction into the necessary framework, and discuss the position of our method in the family of cluster algorithms as well as the relation of the inherent optimization to other problems in learning theory.