Counting the number of labelled unicyclic graphs via Lagrange inversion

combinatoricsgenerating-functionstrees

I am struggling with the following exercise:

Let $\mathcal{U}$ be the class of labelled unicyclic graphs, i.e. connected graphs with precisely one cylce. Express $\mathcal{U}$ in terms of basic constructions involving the class of $\mathcal{T}$ of Cayley trees and find a sum formula for the number $u_n$ of labelled unicyclic graphs on $n$ vertices using the Lagrange inversion theorem.

I understand that $\mathcal{U}$ can be seen as an undirected cylce of size $\ge 3$ (due to the implicit assumption of the graph being simple, so no loops or double edges), where the "nodes" are Cayley trees. Thus

$$ \mathcal{U} = UCYC_{\ge 3}(T),$$

which implies the EGF

$$U(z) = \frac{1}{2} \log \bigg( \frac{1}{1-T(z)} \biggr) – \frac{T(z)}{2} – \frac{T(z)^2}{4}.$$

However, I do not see where to use the Lagrange inversion formula here, as nowhere appears something of the form $y(z) = z \phi (y(z))$.

Could you please give me a hint?

Best Answer

Here is an answer using Egorychev method by way of enrichment. Note that the group acting on the cycle is the dihedral group $D_q$ and not just the cyclic group $C_q.$ We assume that two graphs where we get the second by turning over the first are considered the same. We thus have the combinatorial class equation

$$\def\textsc#1{\dosc#1\csod} \def\dosc#1#2\csod{{\rm #1{\small #2}}} \mathcal{U} = \textsc{DHD}_{\ge 3}(\mathcal{T}).$$

This gives the EGF

$$U(z) = \sum_{q\ge 3} \frac{T(z)^q}{2q} = \frac{1}{2} \log\frac{1}{1-T(z)} - \frac{T(z)}{2} - \frac{T(z)^2}{4}.$$

The quantity we seek is $$n! [z^n] U(z) = (n-1)! [z^{n-1}] U'(z).$$ This is given by

$$(n-1)! \;\underset{z}{\mathrm{res}}\; \frac{1}{z^n} U'(z) = (n-1)! \;\underset{z}{\mathrm{res}}\; \frac{1}{z^n} \left[\frac{1}{2}\frac{1}{1-T(z)} - \frac{1}{2} - \frac{1}{2} T(z) \right] T'(z).$$

Now put $T(z) = w.$ We have from the combinatorial class specification that

$$\mathcal{T} = \mathcal{Z} \times \textsc{SET}(\mathcal{T})$$

so that

$$T(z) = z \exp T(z) \quad\text{and}\quad z = T(z) \exp(-T(z)).$$

We thus obtain

$$(n-1)! \;\underset{w}{\mathrm{res}}\; \frac{\exp(nw)}{w^n} \left[\frac{1}{2}\frac{1}{1-w} - \frac{1}{2} - \frac{1}{2} w \right] \\ = (n-1)! [w^{n-1}] \exp(nw) \left[\frac{1}{2}\frac{1}{1-w} - \frac{1}{2} - \frac{1}{2} w \right].$$

Extract the coefficient to obtain

$$\frac{1}{2} (n-1)! \times \left[ \sum_{q=0}^{n-1} \frac{n^q}{q!} - \frac{n^{n-1}}{(n-1)!} - \frac{n^{n-2}}{(n-2)!} \right].$$

We thus have the sum formula

$$\bbox[5px,border:2px solid #00A000]{ \frac{1}{2} (n-1)! \sum_{q=0}^{n-3} \frac{n^q}{q!}.}$$

This gives the sequence OEIS A057500

$$0, 0, 1, 15, 222, 3660, 68295, 1436568, 33779340, 880107840, \ldots$$

where we see we have the right data. The traditional Lagrange inversion probably refers to the functional equation of the tree function $T(z).$

Note that the OEIS says we are counting connected labeled graphs with $n$ edges and $n$ nodes. This is the same statistic, however. Supposing such a graph had more than one cycle then there is a cycle from which we may remove one edge to obtain a graph with at least one other cycle remaining. The graph is still connected as the rest of the cycle can replace the removed edge. But if it is connected and has $n-1$ edges it must be a tree, a contradiction as there were cycles left over that we didn't touch by removing the edge. So these graphs contain at most one cycle. And this is indeed what happens since otherwise we would have a tree, again a contradiction by the number of nodes.

Related Question