[Paper Review] On the stretch factor of the Theta-4 graph

Luis Barba, Prosenjit Bose+3 more|arXiv (Cornell University)|Mar 21, 2013

Computational Geometry and Mesh Generation6 references8 citations

TL;DR

This paper proves that the θ₄-graph is a geometric spanner by establishing an upper bound of approximately 237 on its stretch factor, resolving the last open question for θ-graphs. The authors achieve this by showing that the θ₄-graph approximates the L∞-Delaunay triangulation and constructing spanning paths between vertices using L₁-distance analysis and geometric properties of empty triangles and cones.

ABSTRACT

In this paper we show that the θ-graph with 4 cones has constant stretch factor, i.e., there is a path between any pair of vertices in this graph whose length is at most a constant times the Euclidean distance between that pair of vertices. This is the last θ-graph for which it was not known whether its stretch factor was bounded.

Motivation & Objective

• To resolve the open problem of whether the θ₄-graph has a bounded stretch factor, completing the characterization of θₘ-graphs as geometric spanners.
• To establish that the θ₄-graph is a geometric spanner by constructing spanning paths between any two vertices with length at most a constant factor of their Euclidean distance.
• To bridge the gap between θ₄-graph and known spanners like the L∞-Delaunay triangulation by proving that the θ₄-graph can emulate paths in the L∞-Delaunay triangulation.
• To provide a tight lower bound of 7 for the stretch factor, suggesting the true stretch factor is close to this value.

Proposed method

• Use of L₁-distance (Manhattan distance) to analyze path lengths in the θ₄-graph, leveraging the fact that dₗ₁(s,t) ≤ 2·dₗ₁(s,w) under certain geometric constraints.
• Construction of spanning paths via a recursive algorithm that follows edges in cones, ensuring progress toward the target by maintaining geometric containment and emptiness of regions.
• Application of geometric lemmas (e.g., Lemma 1) that guarantee path existence when specific regions (e.g., top-right quadrant of S(s)) are empty and cone constraints are satisfied.
• Use of the L∞-Delaunay triangulation as a reference spanner with known stretch factor √(4+2√2) ≈ 2.613, to which the θ₄-graph is shown to be a close approximation.
• Geometric construction of worst-case point sets by iteratively replacing edges with longer paths via vertex insertion in empty triangles, to derive a lower bound.
• Use of orthogonal projections and diagonal bisectors (e.g., ρ, ℓ⁺) to define regions where path length can be bounded via L₁-distance comparisons.

Experimental results

Research questions

• RQ1 Is the θ₄-graph a geometric spanner with bounded stretch factor, given that this was the last unresolved case among θₘ-graphs?
• RQ2 Can the θ₄-graph emulate the path structure of the L∞-Delaunay triangulation, which is known to be a spanner?
• RQ3 What is the tightest possible upper and lower bound on the stretch factor of the θ₄-graph?
• RQ4 Does the path construction in θ₄-graphs remain effective despite the failure of the distance-decreasing property for m ≤ 6?
• RQ5 Can a path be constructed in the θ₄-graph between any two points s and t such that its length is bounded by a constant multiple of |st|?

Key findings

• The stretch factor of the θ₄-graph is at most approximately 237, proving it is a geometric spanner.
• The stretch factor of the θ₄-graph is at least 7, as demonstrated by a constructed point set with a shortest path of length arbitrarily close to 7 times the Euclidean distance.
• The θ₄-graph approximates the L∞-Delaunay triangulation, which has a known stretch factor of √(4+2√2) ≈ 2.613.
• A spanning path exists between any two vertices s and t in the θ₄-graph with length at most (√2 + 36)·|st|, under the condition that the smallest axis-aligned square with s and t as corners is empty.
• The proof relies on geometric invariants such as the emptiness of specific quadrants and the position of intermediate points relative to diagonal lines (e.g., ℓ⁺, ρ) to bound path length using L₁-distance.
• The construction of the lower bound path uv₁v₂v₃v₄w yields a stretch factor arbitrarily close to 7, with edges uv₁, v₃v₄, and v₄w each of length |uw|−ε and v₁v₂, v₂v₃ each of length 2·|uw|−ε.