Sum of squares of chromatic roots of a bipartite graph

66 Views Asked by At

Given a graph $G = (V, E)$, we can calculate its chromatic polynomial $P(G, k)$, and it has $n$ (complex) roots, also known as chromatic roots. It is a well-known fact that the sum of chromatic roots of the graph $G$ is equal to the number of edges $e(G)$.

I am working on a problem and it somehow is related to the following pattern, which I also experiment with SageMath:

Want to prove: Given a bipartite graph $G$ of minimum degree $2$ and $G$ is not a forest, with its chromatic polynomial $P(G, k)$ and chromatic roots $r_{1}, r_{2}, \ldots, r_{n}$, then we have

$$\sum_{i = 1}^{n} r_{i}^{2} = e(G)$$

However, I could not seem to prove it. I tried some Cauchy-Schwarz inequality calculations. I also couldn't find relevant literatures containing this result, or relevant research on (complex) chromatic roots of bipartite graphs.

Thank you!

1

There are 1 best solutions below

4
On BEST ANSWER

The property seems to be true for any bipartite graph unless I missed something in my proof.

Let $P = X^n+c_1X_{n-1}+c_2X_{n-2}+...$ be the chromatic polynomial of $G$.

Properties of roots allow to write: $$c_1 = -\sum r_i$$ $$c_2 = \sum_{i\neq j} r_ir_j$$

So $\sum r_i^2 = c_1^2-2c_2$

We already know by deletion contraction that $c_1 = -e$ Lets show by induction using the deletion contraction that $c_2=\frac{e^2-e}{2}$.

This is true for a graph with no edges, since $c_2=0$.

Suppose this is true for bipartite graphs with $e$ edges. Let $G$ be a bipartite graph with $e+1$ edges.

Let $uv$ be an edge. We know that $G/uv$ (the graph obtained by contracting $uv$) has $e$ edges because the contraction deletes exactly one edge (the contracted one). A contraction can delete more edges by replacing the multi-edges formed by the contraction with a single edge. However, in this case, it does not lose any other edge, otherwise, that would mean that $G$ had a triangle, which can't be possible.

Using the contraction-deletion formula, $c_2(G) = c_2(G-uv) - c_1(G/uv) = \frac{e^2-e}{2} + e = \frac{(e+1)^2-(e+1)}{2}$, which prove the property is hereditary and finish the induction.

It follows that $\sum r_i^2 = c_1^2-2c_2 = e^2 - 2\frac{e^2-e}{2} = e$

Edit: As suggested by @MishaLavrov, the result holds more generally for triangle-free graphs. For a general graph, we have $c_2 = \frac{e^2-e}{2}-t$ where $t$ is the number of triangles because we lose one edge for each triangle in the induction. So $\sum r_i^2 = e+2t$