---
title: Joint Degree-Leafdegree Distribution
url: https://www.emergentmind.com/topics/joint-degree-leafdegree-distribution
type: topic
---

# Joint Degree-Leafdegree Distribution

The joint degree-leafdegree distribution provides a two-parameter statistical characterization of vertices in discrete trees and certain random network models, combining information about a vertex's total degree and the number of its nearest neighbors that are leaves. Formally, for a given graph or growing network, $n_{k,\ell}$ denotes the limiting fraction of vertices with degree $k$ and leafdegree $\ell$ (i.e., those with exactly $k$ neighbors, among which $\ell$ are of degree $1$). This joint distribution refines and generalizes the classical degree distribution and captures fine-structure local correlations relevant for both structural and dynamical network properties.

## 1. Definitions and Fundamental Properties

Let $G_N$ denote a finite graph (typically a tree or a sparse network) with $N$ vertices. For each vertex $j$:
- The degree $k_j$ is the number of neighbors of $j$.
- The leafdegree $\ell_j$ is the number of neighbors of $j$ that are leaves (i.e., have degree $1$).

The empirical count $N_{k,\ell}(N)$ gives the number of vertices with $(k_j, \ell_j) = (k, \ell)$. For broad classes of growing random trees (preferential attachment, random recursive trees, etc.), it can be shown via martingale concentration arguments that for each fixed $(k, \ell)$,
$$
\frac{N_{k,\ell}(N)}{N} \xrightarrow[N\to\infty]{\mathrm{prob.}} n_{k,\ell},
$$
where $n_{k,\ell}$ is the limiting (nonrandom) joint degree-leafdegree distribution. The normalization
$$
\sum_{0\le \ell < k < \infty} n_{k,\ell} = 1
$$
holds since every vertex is accounted for. Marginal distributions are given by
$$
n_k = \sum_{\ell=0}^{k-1} n_{k,\ell} \qquad\text{(degree marginal)},
$$
$$
m_{\ell} = \sum_{k=\ell+1}^\infty n_{k,\ell} \qquad\text{(leafdegree marginal)},
$$
with $\sum_k n_k = \sum_\ell m_\ell = 1$ [2602.01021].

## 2. Generating Functions and Recursion Relations

The joint degree-leafdegree distribution admits a bivariate generating function
$$
g(y, z) = \sum_{0 \leq \ell < k < \infty} n_{k,\ell} y^k z^\ell.
$$
For the classical preferential attachment (PA) tree, the limiting $n_{k,\ell}$ satisfy a two-index recursion:
$$
(k-1)n_{k-1,\ell-1} - (2 + k + \ell) n_{k,\ell} + (\ell + 1) n_{k,\ell+1} + 2\delta_{k,1}\delta_{\ell,0} = 0,
$$
valid for $0 \leq \ell < k < \infty$. Transforming by the generating function gives a first-order linear PDE:
$$
y (1 - yz) \frac{\partial g}{\partial y}
- (1-z) \frac{\partial g}{\partial z}
+ 2 g = 2 y.
$$
The unique analytic solution regular at $(y, z) = (0, 0)$ with $g(1,1) = 1$ is
$$
g(y, z) = 4y \int_0^1 \frac{u^2}{2 - y[(1-u)^2 + z(1-u^2)]} \, du.
$$
This generating-function framework encodes not only $n_{k,\ell}$ but also fluctuations, marginals, and structural ratios. The method generalizes to other growing tree models, such as the random recursive tree (RRT) and redirection models, each yielding its own recursion and integral expression for $g(y, z)$ [2602.01021].

## 3. Explicit Formulas and Asymptotics

From the integral form of $g$, an explicit expression for $n_{k,\ell}$ is obtained:
$$
n_{k,\ell} =
\binom{k-1}{\ell}\,2^{-(k-2)}
\int_0^1 u^2 (1-u)^{2(k-1-\ell)} (1+u)^\ell \, du, \qquad 0 \leq \ell < k.
$$
Extremal values include:
- $n_{k,0} = \frac{1}{2^k[k(k^2-1/4)]}$ (protected nonleaves with no leaf neighbors)
- $n_{k,k-1} = \frac{(k-1)! \sqrt{\pi}}{2^k \Gamma(k+\tfrac{3}{2})}$

For large $k$, the conditional mean and variance of the leafdegree given degree $k$ are
$$
\bar{\ell}(k) = 1 + \frac{k}{2} - \frac{6}{k+3}, \qquad
\mathrm{Var}[\ell|k] = \frac{k}{4} + O(1).
$$
Thus for high-degree vertices, the leafdegree concentrates sharply around $k/2$, i.e., half of the neighbors of large-$k$ vertices are leaves with vanishingly small relative fluctuations [2602.01021].

## 4. Marginal Distributions and Protected Vertices

The marginal leafdegree distribution,
$$
m_\ell = 4 \int_0^1 \frac{u^2}{1+2u-u^2} \Bigl(1 - \frac{2u}{1+2u-u^2}\Bigr)^\ell du, \quad \ell \geq 0,
$$
exhibits an algebraic tail, $m_\ell \sim \ell^{-3}$ for large $\ell$. The protected vertex fraction (vertices that are not leaves and have leafdegree zero) is encoded in $n_{k,0}$. Summing, the overall fraction is
$$
p = \sum_{k \geq 2} n_{k,0} = m_0 - n_1 \approx 0.039447,
$$
with mean degree among protected vertices $\bar k_p \approx 2.20249$ [2602.01021].

## 5. Fluctuations and Limit Theorems

For fixed $(k, \ell)$ and large $N$,
$$
\mathbb{E}[N_{k,\ell}] \sim n_{k,\ell} N, \qquad
\mathrm{Var}(N_{k,\ell}) \sim \chi_{k,\ell} N,
$$
and the normalized counts are asymptotically (multivariate) normal. For example:
- Number of leaves $N_1 \sim \mathcal{N}(2N/3, N/9)$,
- Number of degree-2 vertices $N_2 \sim \mathcal{N}(N/6, 23N/180)$,
- Number of degree-2, leafdegree-1 vertices $N_{2,1} \sim \mathcal{N}(2N/15, 49N/600)$.

Joint distributions of these counts converge to bivariate normal with explicitly computable covariances (e.g., $\sigma_{12} = -4/45$, $\sigma_{1,(2,1)} = -1/18$). All higher cumulants are $O(N)$ [2602.01021, 2102.07570].

## 6. Extensions and Related Models

The bivariate recursion/generating-function method is applicable to a range of tree growth mechanisms:
- **Random Recursive Tree (RRT):** The recursion simplifies, and the unique solution is $g(y,z) = y\int_0^1 v^{1-y} e^{-y(1-z)(1-v)} dv$. Degree and leafdegree distributions, as well as protected fractions, can be recovered explicitly; for the protected fraction, $p = 1/2 - 1/e \approx 0.13212$.
- **Redirection Model:** The recursion alters, leading to a more complicated, but tractable, integral representation and varying marginals and fluctuation laws.
- **Pólya Trees:** The joint profile of (degree-$d$, leaf) vertices at a given level $k$ in random Pólya trees of fixed size $n$ has a generating function-based (cycle-index) construction. For $k=\lfloor \kappa\sqrt{n} \rfloor$, both the expectation and covariance of node counts admit asymptotic expansions, with joint normalized processes converging to linearly dependent multiples of Brownian excursion local time [1104.4271].

## 7. Connection to Degree-Degree and Assortativity Statistics

The joint degree-leafdegree framework is related, but distinct from, the classical joint degree distribution $p(k, \ell)$ and neighbor-degree conditional distributions $p(\ell|k)$ as studied in degree-correlation and assortativity literature [1308.5169, 2008.10506]. Unlike those, the joint degree-leafdegree distribution directly counts the configuration of local neighborhoods in terms of leaf status, offering refined resolution for studying robustness, percolation, and motif frequency in networks.

In edge-based degree-degree frameworks, the probability that a random edge joins nodes of degrees $j$ and $k$ is $e_{jk}$, with marginals and conditionals given by:
$$
p_k = \frac{\langle k \rangle}{k} \sum_j e_{jk}, \qquad
P(j|k) = \frac{e_{jk}}{k p_k/\langle k \rangle}.
$$
The probability to reach a degree-$k$ neighbor from a leaf (degree $1$ node) is
$$
P(k|1) = \frac{e_{k,1}}{\sum_{\ell} e_{\ell,1}},
$$
a statistic resembling, but not equivalent to, the $\ell$-marginals of $n_{k,\ell}$ [2008.10506].

## 8. Significance and Applications

The joint degree-leafdegree distribution $n_{k,\ell}$ enables a comprehensive local (and nearly local) statistical characterization of the topology of random trees and related networks. Applications include:
- Quantitative analysis of motif abundance and protected vertices,
- Rigorous fluctuation (central limit) theorems for empirical substructure counts,
- Parameterization and comparison of generative tree models,
- Input to network robustness, percolation, and dynamics studies, especially in the analysis of epidemic thresholds and resilience under failures,
- Refinement beyond traditional degree and edge-based correlation statistics [2602.01021, 1308.5169, 2008.10506].

The methodology provides both explicit formulas and asymptotics, with generating-function machinery that is tractable for both analytic and algorithmic purposes. The approach is general and adaptable to new models, supporting ongoing research into network microstructure and its macroscopic implications.

Source: https://www.emergentmind.com/topics/joint-degree-leafdegree-distribution