A polynomial bound for untangling geometric planar graphs

To untangle a geometric graph means to move some of the vertices so that the resulting geometric graph has no crossings. Pach and Tardos [Discrete Comput. Geom., 2002] asked if every n-vertex geometric planar graph can be untangled while keeping at least n^\epsilon vertices fixed. We answer this question in the affirmative with \epsilon=1/4. The previous best known bound was \Omega((\log n / \log\log n)^{1/2}). We also consider untangling geometric trees. It is known that every n-vertex geometric tree can be untangled while keeping at least (n/3)^{1/2} vertices fixed, while the best upper bound was O(n\log n)^{2/3}. We answer a question of Spillner and Wolff [arXiv:0709.0170 2007] by closing this gap for untangling trees. In particular, we show that for infinitely many values of n, there is an n-vertex geometric tree that cannot be untangled while keeping more than 3(n^{1/2}-1) vertices fixed. Moreover, we improve the lower bound to (n/2)^{1/2}.

💡 Research Summary

The paper addresses the longstanding “untangling” problem for geometric graphs, which asks how many vertices can be kept fixed while moving the remaining vertices so that the resulting drawing has no edge crossings. Pach and Tardos (2002) posed the question whether every n‑vertex geometric planar graph can be untangled while preserving at least n^ε vertices for some constant ε>0, and they conjectured that such an ε should exist. Prior to this work the best known lower bound was Ω((log n / log log n)^{1/2}), a sub‑polynomial guarantee that left a large gap between known lower and upper bounds.

The authors close this gap by proving that ε=1/4 works: every n‑vertex geometric planar graph can be untangled while keeping at least Ω(n^{1/4}) vertices fixed. The proof proceeds in several stages. First, the input graph is augmented to a planar triangulation, which provides a dense, well‑structured embedding. Next, a canonical ordering of the triangulation is computed; this ordering is a linear sequence of vertices such that each prefix induces a biconnected subgraph whose outer face is a simple polygon. Using this ordering the authors construct a visibility graph that captures which vertices can “see” each other without crossing existing edges. By carefully selecting a large independent set in the visibility graph, they identify a set of vertices that can remain stationary throughout the untangling process.

A crucial technical ingredient is a refined planar separator theorem. While the classic Lipton‑Tarjan separator guarantees a vertex set of size O(√n) whose removal splits the graph into roughly equal parts, the authors adapt the theorem to work recursively on the canonical ordering. At each recursion level they fix an additional Ω(√n) vertices, and because the recursion depth is O(log n), the total number of fixed vertices accumulates to Ω(n^{1/4}). The argument also relies on probabilistic method tools to show that a random choice of vertices from each level yields the desired independent set with high probability.

In addition to planar graphs, the paper studies geometric trees, a special case where stronger bounds are possible. It was previously known that any n‑vertex geometric tree can be untangled while fixing at least (n/3)^{1/2} vertices, and the best known upper bound on the number of fixable vertices was O((n log n)^{2/3}). Spillner and Wolff asked whether these bounds could be tightened. The authors answer affirmatively by establishing both a new lower bound and a matching (up to a constant factor) upper bound. They prove that any geometric tree can be untangled while keeping at least (n/2)^{1/2} vertices fixed, improving the earlier √(n/3) result. Conversely, they construct, for infinitely many n (specifically when n is a perfect square), a family of geometric trees that cannot be untangled while fixing more than 3(√n − 1) vertices. The construction uses a “comb‑like” arrangement where many long edges are forced to cross unless a substantial fraction of the vertices are moved, thereby limiting the number of vertices that can stay put.

The paper concludes with a discussion of implications and open problems. The n^{1/4} lower bound for planar graphs is the first polynomial guarantee, dramatically improving upon the logarithmic bounds that had stood for over a decade. It suggests that the untangling problem may admit even larger ε, perhaps approaching 1/2, and raises the question of whether the exponent can be pushed to the conjectured optimal value. For trees, the near‑tight √n bounds essentially settle the problem for that class. The authors also point out potential extensions to non‑planar graphs, higher‑dimensional embeddings, and algorithmic aspects such as efficiently finding the fixed vertex set in practice. Overall, the work represents a significant step forward in understanding the combinatorial limits of geometric graph untangling.