Matroid invariants and counting graph homomorphisms

TL;DR

This paper characterizes which finite graphs G produce graph parameters from homomorphism counts that depend only on the cycle matroid of F, extending known results and using multilinear algebra tools.

Contribution

It provides a complete characterization of graphs G for which homomorphism counting yields a matroid invariant, including extensions to weighted graphs.

Findings

Characterization of graphs G with matroid-invariant homomorphism counts

Extension of results to weighted graphs and related problems

Use of multilinear algebra in the proof

Abstract

The number of homomorphisms from a finite graph $F$ to the complete graph $K_n$ is the evaluation of the chromatic polynomial of $F$ at $n$. Suitably scaled, this is the Tutte polynomial evaluation $T(F;1-n,0)$ and an invariant of the cycle matroid of $F$. De la Harpe and Jaeger \cite{dlHJ95} asked more generally when is it the case that a graph parameter obtained from counting homomorphisms from $F$ to a fixed graph $G$ depends only on the cycle matroid of $F$. They showed that this is true when $G$ has a generously transitive automorphism group (examples include Cayley graphs on an abelian group, and Kneser graphs). Using tools from multilinear algebra, we prove the converse statement, thus characterizing finite graphs $G$ for which counting homomorphisms to $G$ yields a matroid invariant. We also extend this result to finite weighted graphs $G$ (where to count homomorphisms from $F$ to $G$ includes such problems as counting nowhere-zero flows of $F$ and evaluating the partition function of an interaction model on $F$).