Improvements on Spectral Bisection

We investigate combinatorial properties of certain configurations of a graph partition which are related to the minimality of a cut. We show that such configurations are related to the third eigenvector of the Laplacian matrix. It is well known that the second eigenvector encodes structural information, and that can be used to approximate a minimum bisection. In this paper, we show that the third eigenvector carries structural information as well. We then provide a new spectral bisection algorithm using both eigenvectors. The new algorithm is guaranteed to return a cut that is smaller or equal to the one returned by the classic spectral bisection. Also, we provide a spectral algorithm that can refine a given partition and produce a smaller cut.