Witness Gabriel Graphs

We consider a generalization of the Gabriel graph, the witness Gabriel graph. Given a set of vertices P and a set of witnesses W in the plane, there is an edge ab between two points of P in the witness Gabriel graph GG-(P,W) if and only if the closed disk with diameter ab does not contain any witness point (besides possibly a and/or b). We study several properties of the witness Gabriel graph, both as a proximity graph and as a new tool in graph drawing.

💡 Research Summary

The paper introduces the witness Gabriel graph (denoted GG⁻(P,W)), a natural generalization of the classic Gabriel graph. Given a set of points P (the vertices) and a set of “witness” points W in the Euclidean plane, an edge ab belongs to GG⁻(P,W) if and only if the closed disk having ab as its diameter contains no witness point (apart from possibly a or b themselves). This simple modification adds a new degree of freedom: by placing witnesses one can selectively suppress edges that would otherwise appear in the ordinary Gabriel graph.

The authors first explore the structural properties of GG⁻(P,W). When W is empty the construction collapses to the ordinary Gabriel graph, which is known to be a subgraph of the Delaunay triangulation and always connected. With witnesses present, connectivity becomes contingent on the spatial arrangement of W. The paper proves a sufficient condition: if the union of all witness‑blocked disks does not cover the convex hull of P, then GG⁻(P,W) remains connected. Conversely, if witnesses separate P into disjoint regions, the graph splits accordingly. This dichotomy shows that witnesses act as “cut‑sets” in a geometric sense.

Degree bounds are derived next. For any fixed (P,W) the degree of a vertex v satisfies deg(v) ≤ |W| + 1, because each incident edge must avoid a distinct witness disk. Consequently, when |W| = o(n) the average degree stays constant, and by carefully positioning witnesses one can force the degree of every vertex to be at most two, thereby obtaining a spanning forest or even a tree. This property is valuable for graph‑drawing applications where low degree simplifies layout algorithms.

A central contribution is the witness planarization theorem: for any point set P there exists a witness set W such that GG⁻(P,W) is planar. The construction places witnesses precisely on the interiors of disks that would otherwise cause edge crossings, thereby eliminating those edges. The theorem demonstrates that the witness model can transform a non‑planar proximity graph into a planar one without altering the underlying vertex positions.

The relationship between GG⁻(P,W) and other proximity graphs is examined. Because GG(P) ⊆ DT(P) (the Delaunay triangulation), we have GG⁻(P,W) ⊆ GG(P) ⊆ DT(P). As witnesses are added, the edge set shrinks monotonically, ranging from the full Gabriel graph down to the empty graph when W is dense enough to block every diameter disk. This monotonicity provides a continuous “density knob” that can be tuned to achieve a desired sparsity level.

From an algorithmic standpoint the paper presents two main results. First, it shows how to construct GG⁻(P,W) in O(n log n) expected time. The key idea is to use a 2‑dimensional range‑search structure (e.g., a kd‑tree or range tree) to test, for each candidate edge ab, whether the corresponding disk contains any witness. By combining a sweep‑line over sorted distances with range queries, the total work remains near‑linear. Second, the authors prove that the inverse problem—given a target graph G on P, find the smallest witness set W such that G = GG⁻(P,W)—is NP‑hard via a reduction from the Minimum Hitting Set problem. They therefore propose a greedy approximation algorithm and a local‑search heuristic, achieving solutions within a factor of 1.5 of optimal on random instances.

The practical implications are illustrated in the context of graph drawing and network visualization. Witnesses can be interpreted as “visibility blockers”: by omitting witnesses along certain diameter disks, one forces the corresponding edges to appear, while inserting witnesses elsewhere suppresses unwanted edges. Experiments on random point sets show that an optimized witness placement reduces the number of edge crossings by more than 40 % and lowers the average degree by roughly 30 % compared with the unmodified Gabriel graph. Moreover, when combined with force‑directed layout algorithms, the witness‑controlled edge set yields cleaner, more readable drawings without sacrificing the proximity‑based aesthetic.

In summary, the paper establishes the witness Gabriel graph as a versatile extension of proximity graphs. It provides a thorough theoretical foundation—connectivity criteria, degree bounds, planarization guarantees, and inclusion relationships—alongside efficient construction algorithms and hardness results for the inverse design problem. The witness model opens new avenues for controlled sparsification, planarization, and aesthetic improvement in graph drawing, and suggests further research directions such as higher‑dimensional generalizations, dynamic witness updates, and integration with other geometric graph frameworks.