Graph polynomials from principal pivoting

Abstract

The recursive computation of the interlace polynomial introduced by Arratia, Bollobás and Sorkin is defined in terms of a new pivoting operation on undirected simple graphs. In this paper, we interpret the new pivoting operation on graphs in terms of standard pivoting (on matrices). Specifically, we show that, up to swapping vertex labels, Arratia et al.'s pivoting operation on a graph is equivalent to a principal pivot transform on the graph's adjacency matrix, provided that all computations are performed in the Galois field F 2 . Principal pivoting on adjacency matrices over F 2 has a natural counterpart on isotropic systems. Thus, our view of the interlace polynomial is closely related to the one by Aigner and van der Holst. The observations that adjacency matrices of undirected simple graphs are skew-symmetric in F 2 and that principal pivoting preserves skew-symmetry in all fields suggest to extend Arratia et al.'s pivoting operation to fields other than F 2 . Thus, the interlace polynomial extends to polynomials on gain graphs, namely bidirected edge-weighted graphs whereby reversed edges carry non-zero weights that differ only by their sign. Extending a proof by Aigner and van der Holst, we show that the extended interlace polynomial can be represented in a non-recursive form analogous to the non-recursive form of the original interlace polynomial, i.e., the Martin polynomial. For infinite fields it is shown that the extended interlace polynomial does not depend on the (non-zero) gains, as long as they obey a non-singularity condition. These gain graphs are all supported by a single undirected simple graph. Thus, a new graph polynomial is defined for undirected simple graphs. The recursive computation of the new polynomial can be done such that all ends of the recursion correspond to independent sets. Moreover, its degree equals the independence number. However, the new graph polynomial is different from the independence polynomial.