Graph cospectrality using neighborhood matrices

From MaRDI portal





Summary: In this note we address the problem of graph isomorphism by means of eigenvalue spectra of different matrix representations: the neighborhood matrix \(\hat{M}\), its corresponding signless Laplacian \(Q_{\hat{M}}\), and the set of higher order adjacency matrices \(M_{\ell}\)s. We find that, in relation to graphs with at most 10 vertices, \(Q_{\hat{M}}\) leads to better results than the signless Laplacian \(Q\); besides, when combined with \(\hat{M}\), it even surpasses the Godsil and McKay switching method.











This page was built for publication: Graph cospectrality using neighborhood matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q456353)