Which Connected Spatial Networks on Random Points have Linear Route-Lengths?

In a model of a connected network on random points in the plane, one expects that the mean length of the shortest route between vertices at distance $r$ apart should grow only as $O(r)$ as $r \to \infty$, but this is not always easy to verify. We give a general sufficient condition for such linearity, in the setting of a Poisson point process. In a $L \times L$ square, define a subnetwork $\GG_L$ to have the edges which are present regardless of the configuration outside the square; the condition is that the largest component of $\GG_L$ should contain a proportion $1 - o(1)$ of the vertices, as $L \to \infty$. The proof is by comparison with oriented percolation. We show that the general result applies to the relative neighborhood graph, and establishing the linearity property for this network immediately implies it for a large family of proximity graphs.

💡 Research Summary

The paper investigates a fundamental scaling property of spatial networks built on random points in the plane: whether the expected length of the shortest route between two vertices separated by Euclidean distance r grows only linearly, i.e., O(r), as r → ∞. While this linearity is intuitively expected for many deterministic geometric graphs, proving it for stochastic networks is non‑trivial. The authors work in the setting of a homogeneous Poisson point process (PPP) of unit intensity and consider a broad class of “proximity graphs” defined by deterministic rules that decide whether an edge is present between any pair of points.

The central contribution is a general sufficient condition for linear route‑lengths. For each side length L, they define a subnetwork 𝔊_L consisting of those edges whose existence is determined solely by the configuration of points inside the L × L square, irrespective of points outside. The condition states that if the largest connected component of 𝔊_L contains a proportion 1 – o(1) of all vertices in the square as L → ∞, then the infinite‑volume network satisfies ℓ(r) = O(r). In other words, almost all points must belong to a “giant component” inside arbitrarily large finite windows.

The proof proceeds by coupling the finite‑window structure to oriented percolation. The square is partitioned into a grid of smaller cells; a cell is declared “open” if its interior points are sufficiently well‑connected to allow passage to neighboring cells, and “closed” otherwise. Using geometric arguments specific to the underlying proximity rule, the authors show that the probability a cell is open exceeds the critical threshold p_c for oriented percolation when L is large. Standard results then guarantee the existence of an infinite directed open path, which translates into a network path whose length is bounded by a constant multiple of the Euclidean distance. This coupling yields the desired O(r) bound for the whole network.

To demonstrate the applicability of the condition, the authors focus on the Relative Neighborhood Graph (RNG). In the RNG, an edge between points x and y exists only if no third point z lies simultaneously closer to both x and y than they are to each other. The paper shows that, for a PPP, the RNG’s subnetwork 𝔊_L satisfies the giant‑component condition. The argument involves estimating the “forbidden region” around each potential edge and proving that, with high probability, a positive fraction of cells are open. Consequently, the RNG possesses linear route‑lengths.

Because many other proximity graphs (Gabriel graph, Delaunay triangulation, β‑skeletons, etc.) either contain the RNG as a subgraph or can be obtained from it by adding or deleting edges in a controlled way, the linearity result automatically extends to this whole family. Thus the paper provides a unified framework that replaces case‑by‑case analyses with a single percolation‑based criterion.

The work has several notable implications. First, it links geometric network scaling to well‑studied percolation theory, offering a powerful tool for future investigations of random spatial networks, including those in higher dimensions or with non‑Poisson point processes. Second, it supplies a rigorous justification for using RNG‑type structures in practical systems (wireless ad‑hoc networks, sensor deployments, transportation routing) where linear travel cost is essential. Finally, the authors acknowledge that the condition is sufficient but not necessary; identifying necessary conditions or extending the approach to weighted edges and heterogeneous point patterns remains an open research direction.

In summary, the paper establishes that if a spatial network’s finite‑window subgraph almost always contains a giant component, then the network’s shortest‑path distances scale linearly with Euclidean distance. This theorem is proved via an oriented‑percolation coupling and is verified for the Relative Neighborhood Graph, thereby guaranteeing linear route‑lengths for a broad class of proximity graphs built on Poisson point processes.