How to compute the expected degree of the root of Cayley and Catalan trees?

168 Views Asked by At

Find the expected degree of the root of

  1. Cayley trees
  2. Catalan trees

I want to do this with probability generating functions, aka PGFs. Here is what I got so far:

  1. Let $\mathcal{C}$ be the class of rooted acyclic connected graphs, a.k.a. Cayley trees (non-plane rooted labelled trees), and let $X$ denote the degree of the root. Then we have

$$\mathcal{C} = \mathcal{A} \times \text{SET}\left( u\mathcal{C} \right)$$ and $$C(z,u) = z \exp(uC(z,1)).$$ Then the expected degree of the root in a random Cayley tree on $n$ vertices is given by

$$\mathbb{E}[X] = \frac{1}{[z^n]C(z,1)} [z^n] \frac{\partial}{\partial u} C(z,u) \bigg\vert_{u=1} = \ldots$$

However, I do not see how to go on from here. I know that $$[z^n]C(z) = n^{n-1},$$ so $C_n$ is the $n$-th Cayley number. However, I do not see what to do with the term $[z^n] \frac{\partial}{\partial u} C(z,u) \bigg\vert_{u=1}$.

  1. Let $\mathcal{P}$ be the class of rooted acyclic connected plane graphs, a.k.a. Catalan trees (a.k.a. plane rooted trees), and let $X$ denote the degree of the root. Then we have

$$\mathcal{P} = \mathcal{A} \times \text{SET}\left( u\mathcal{P} \right)$$ and $$P(z,u) = z \exp(uP(z,1)).$$ Then the expected degree of the root in a random Catalan tree on $n$ vertices is given by

$$\mathbb{E}[X] = \frac{1}{[z^n]P(z,1)} [z^n] \frac{\partial}{\partial u} P(z,u) \bigg\vert_{u=1} = \ldots$$

Again, I do not see how to go on from here. I know that $$[z^n]P(z) = \frac{1}{n} \cdot {{2n-2}\choose{n-1}},$$ but I do not see what to do with the term $[z^n] \frac{\partial}{\partial u} P(z,u) \bigg\vert_{u=1}$. However, I know that $P(z) = \frac{1-\sqrt{1-4z}}{2}$.

Could you please help me?

2

There are 2 best solutions below

1
On BEST ANSWER

For Cayley trees let us recall the combinatorial class equation

$$\def\textsc#1{\dosc#1\csod} \def\dosc#1#2\csod{{\rm #1{\small #2}}} \mathcal{T} = \mathcal{Z} \times \textsc{SET}(\mathcal{T}).$$

which gives the functional equation

$$T(z) = z \exp T(z).$$

Marking the root degree we get

$$\mathcal{Z} \times \mathcal{U}^0 \textsc{SET}_{=0}(\mathcal{T}) + \mathcal{Z} \times \mathcal{U}^1 \textsc{SET}_{=1}(\mathcal{T}) + \mathcal{Z} \times \mathcal{U}^2 \textsc{SET}_{=2}(\mathcal{T}) + \cdots$$

This gives the generating function

$$R(z, u) = z \sum_{k\ge 0} u^k \frac{T(z)^k}{k!} = z \exp(u T(z)).$$

As a sanity check put $u=1$ to get $z \exp T(z)$ which is correct. Differentiate and put $u=1$ to obtain

$$\left.\frac{\partial}{\partial u} R(z, u)\right|_{u=1} = \left. z \exp(u T(z)) T(z) \right|_{u=1} = z \exp(T(z)) T(z) = T(z)^2.$$

Now using Lagrange inversion by residues (Egorychev method) we seek

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

Next put $T(z)=w$ to get with $z= w \exp(-w)$

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

It follows that the average degree of the root is

$$\bbox[5px,border:2px solid #00A000]{ 2 \frac{n-1}{n}.}$$

This also works for $n=1.$

The above result can be verified using the combstruct package, see below.

with(combinat);
with(combstruct);

trees :=
proc(n)
    option remember;
    local spec, k, res;

    res := 0;

    for k to n-1 do
        spec := { T1= Prod(Z, Set(T1)),
                   Z=Atom,
                   T2 = Prod(Z, Set(T1, card=k))};
        res := res +
        k*count([T2, spec, labeled], size=n); 
    od;

    res/n^(n-1);
end;
0
On

For Catalan trees let us recall the combinatorial class equation

$$\def\textsc#1{\dosc#1\csod} \def\dosc#1#2\csod{{\rm #1{\small #2}}} \mathcal{C} = \mathcal{Z} \times \textsc{SEQ}(\mathcal{C}).$$

which gives the functional equation

$$C(z) = z \frac{1}{1-C(z)}.$$

Marking the root degree we get

$$\mathcal{Z} \times \mathcal{U}^0 \textsc{SEQ}_{=0}(\mathcal{C}) + \mathcal{Z} \times \mathcal{U}^1 \textsc{SEQ}_{=1}(\mathcal{C}) + \mathcal{Z} \times \mathcal{U}^2 \textsc{SEQ}_{=2}(\mathcal{C}) + \cdots$$

This gives the generating function

$$R(z, u) = z \sum_{k\ge 0} u^k C(z)^k = z \frac{1}{1-u C(z)}.$$

As a sanity check put $u=1$ to get $z/(1-C(z))$ which is correct. Differentiate and put $u=1$ to obtain

$$\left. \frac{\partial}{\partial u} R(z, u) \right|_{u=1} = \left. \frac{z}{(1-uC(z))^2} C(z) \right|_{u=1} = \frac{1}{z} C(z)^2 C(z) = \frac{1}{z} C(z)^3.$$

Now using Lagrange inversion by residues (Egorychev method) we seek

$$[z^n] \frac{1}{z} C(z)^3 = [z^{n+1}] C(z)^3 = \frac{1}{n+1} [z^n] 3 C(z)^2 C'(z) \\ = \frac{1}{n+1} \;\underset{z}{\mathrm{res}}\; \frac{1}{z^{n+1}} 3 C(z)^2 C'(z).$$

Next put $C(z)=w$ to get with $z= w (1-w)$

$$\frac{1}{n+1} \;\underset{w}{\mathrm{res}}\; \frac{1}{w^{n+1} (1-w)^{n+1}} 3w^2 = \frac{3}{n+1} [w^{n-2}] \frac{1}{(1-w)^{n+1}} = \frac{3}{n+1} {n+n-2\choose n} \\ = \frac{3}{n+1} {2n-2\choose n}.$$

It follows that the average degree of the root is

$$\frac{3}{n+1} {2n-2\choose n} n {2n-2\choose n-1}^{-1} = \frac{3}{n+1} \frac{(n-1)!^2}{(n-1)! (n-2)!}$$

or

$$\bbox[5px,border:2px solid #00A000]{ 3\frac{n-1}{n+1}.}$$

This result was verified with the combstruct package, see below.

with(combstruct);
with(combinat);

trees :=
proc(n)
    option remember;
    local spec, k, res;

    res := 0;
    for k to n-1 do
        spec := { T1= Prod(Z, Sequence(T1)),
                  Z=Atom,
                  T2 = Prod(Z, Sequence(T1, card=k))};
        
        res := res +
        k*count([T2, spec, unlabeled], size=n);
    od;

    res/(1/n*binomial(2*n-2,n-1));
end;