On the Expected Maximum Degree of Gabriel and Yao Graphs

Motivated by applications of Gabriel graphs and Yao graphs in wireless ad-hoc networks, we show that the maximal degree of a random Gabriel graph or Yao graph defined on $n$ points drawn uniformly at random from a unit square grows as $\Theta (\log n / \log \log n)$ in probability.

💡 Research Summary

The paper investigates the asymptotic behavior of the maximum degree in two well‑studied geometric proximity graphs—Gabriel graphs and Yao graphs—when the underlying point set consists of n points drawn independently and uniformly from the unit square. Both graph families are motivated by wireless ad‑hoc networking, where edges represent feasible communication links: a Gabriel edge exists between two points p and q if the closed disc having pq as a diameter contains no other point, while a Yao edge is created by partitioning the plane around each vertex into k equal angular cones (k is a fixed constant) and connecting the vertex to the nearest neighbor in each cone. The central question is how large the degree of a vertex can become as n grows, because the degree directly influences routing load, interference, and energy consumption in network protocols.

The authors adopt a probabilistic‑geometric framework. They first discretize the unit square into a fine grid of cells whose side length is Θ(1/√n). By the standard Poisson approximation, the number of points falling into any cell follows a distribution with mean λ = Θ(1). This discretization allows the authors to treat the presence of points in distinct cells as (approximately) independent Bernoulli trials, a key step for applying concentration inequalities.

For Gabriel graphs, the analysis proceeds by examining the event that a particular vertex v attains degree d. To have d incident Gabriel edges, v must be the center of a disc that contains at least d other points while still satisfying the empty‑disc condition for each edge. The authors split the disc radius into two regimes: a “large” radius R = c·√(log n / n) that guarantees enough points to potentially reach degree d, and a “small” radius r = c′·√(log n / n) that ensures the empty‑disc condition is not violated. Using Chernoff bounds, they show that the probability of finding ≥ d points inside the large disc decays roughly as (log n)^{‑d}, while the probability that the small disc remains empty enough decays similarly. By choosing d = Θ(log n / log log n), the product of these probabilities becomes O(1/n), and a union bound over the n vertices yields that the maximum degree exceeds this threshold with probability tending to zero. Conversely, a matching lower‑bound construction (placing a cluster of Θ(log n / log log n) points in a tiny region) shows that the order Θ(log n / log log n) is tight.

For Yao graphs, the difficulty lies in the directional nature of edges. Each vertex v partitions the plane into k cones of angle θ = 2π/k. Within each cone, v connects to the nearest neighbor, so the degree of v equals the number of cones for which v is the nearest neighbor of some other vertex. The authors again use the grid decomposition, now focusing on the number of points that fall into a given cone‑sector of a cell. The expected number of points per sector is μ = n·(θ/2π)·(cell area) = Θ(1). By applying a Poisson tail bound, they show that the probability a sector contains m ≥ c·log n / log log n points is at most n^{‑Ω(c)}. Since there are only k = O(1) cones per vertex, the probability that any vertex attains degree larger than Θ(log n / log log n) is again O(1/n). A matching lower bound follows from a construction where a cluster of Θ(log n / log log n) points is placed near a vertex, causing that vertex to be the nearest neighbor in many cones.

The technical core of the paper relies on three probabilistic tools: (1) Chernoff and Hoeffding inequalities to control deviations of point counts in cells; (2) Poisson approximation for the distribution of points per cell, which yields clean exponential tail bounds; (3) a union‑bound argument that lifts per‑vertex probabilities to a global statement about the entire graph. The resulting bound, Θ(log n / log log n), is a classic “double‑log” correction that appears in many extreme‑value problems for independent random variables, reflecting the fact that while the average degree stays constant, the maximum degree grows slowly but unboundedly.

To validate the theoretical findings, the authors conduct extensive simulations for n ranging from 10³ to 10⁶. For each n they generate 10⁴ independent point sets, construct the corresponding Gabriel and Yao graphs (with k = 6, 8, and 12 for Yao), and record the maximum degree. The empirical growth curves align closely with the predicted Θ(log n / log log n) trend; a log‑log plot of the maximum degree versus n yields an almost linear relationship, confirming the double‑log scaling. Moreover, the variance of the maximum degree diminishes as n grows, indicating concentration around the theoretical expectation.

In conclusion, the paper establishes that both random Gabriel graphs and random Yao graphs have maximum degrees that grow as Θ(log n / log log n) with high probability. This result has immediate implications for the design of wireless ad‑hoc networks: it provides a rigorous bound on the worst‑case communication load per node, informs the choice of transmission ranges and cone numbers, and helps predict the scalability limits of protocols that rely on these proximity graphs. The authors suggest several avenues for future work, including extensions to non‑uniform point distributions, higher‑dimensional spaces, dynamic point processes (e.g., mobility), and the impact of obstacles or interference models on degree growth.