Abstract 1 Introduction 2 Proof of Theorem 2 3 Concluding remarks References

From Local Pair-Crossing Number to Local Crossing Number

Jacob Fox ORCID Department of Mathematics, Stanford University, CA, USA János Pach ORCID HUN-REN Alfréd Rényi Institute of Mathematics, Budapest, Hungary Andrew Suk ORCID Department of Mathematics, University of California San Diego, La Jolla, CA, USA
Abstract

We prove that if a graph can be drawn in the plane such that each edge crosses at most k other edges, then it can be redrawn so that each edge participates in at most k3+O(k2) crossings. This improves the previous exponential bound that follows from a result of Schaefer and Štefankovič and answers a question of Ackerman and Schaefer.

Keywords and phrases:
Crossing numbers, pair crossing numbers
Funding:
Jacob Fox: supported by NSF awards DMS-2452737 and DMS-2154129.
János Pach: Rényi Institute, Budapest. Supported by NKFIH grant K-176529, Austrian Science Fund Z 342-N31 and ERC Advanced Grant “GeoScape.”
Andrew Suk: Supported by NSF grants DMS-1952786 and DMS-2246847.
Copyright and License:
[Uncaptioned image] © Jacob Fox, János Pach, and Andrew Suk; licensed under Creative Commons License CC-BY 4.0
2012 ACM Subject Classification:
Mathematics of computing Combinatorics
Funding:
This manuscript was completed while the authors were in residence at SLMath Berkeley, during the Spring semester of 2025, supported by NSF Grant DMS-1928930.
Acknowledgements:
The authors want to express their gratitude to Géza Tóth for directing their attention to the problem discussed in this note, and for his many valuable remarks and suggestions.
Editors:
Vida Dujmović and Fabrizio Montecchiani

1 Introduction

In much of the early literature about Turán’s brick factory problem and other questions related to graph drawing, the precise problem setting was ambiguous. The basic notions were thought to be so self-evident, that it was often felt that there was no need for definitions. This anomaly was pointed out by Pach and G. Tóth [11, 12], who made a distinction between several different crossing numbers. The two most important among them were the crossing number, cr(G), and the pairwise crossing number, pair-cr(G), of a graph G. The first one is the minimum possible number of crossing points between the edges of G, in a drawing of G, where

  1. 1.

    two edges, including edges that have a common endpoint, are allowed to cross any finite number of times,

  2. 2.

    no three edges meet at the same point,

  3. 3.

    the edges are drawn as simple non-self-intersecting Jordan curves.

On the other hand, pair-cr(G), is the minimum number of crossing pairs of edges, that is, if a pair of edges cross multiple times, it still contributes only 1 to this quantity. Obviously, we have pair-cr(G)cr(G) for every graph G. It was often assumed with no explanation whatsoever that these two parameters coincide. It is one of the most annoying open problems in the area to decide if this is indeed the case.

Problem 1 ([11, 12]).

Is it true that pair-cr(G)=cr(G) for every graph G?

It was proved in [12] that cr(G)(pair-cr(G)2) holds for every graph G. This result was subsequently improved several times [6, 7, 18, 20, 21, 22]. The best presently known bound, cr(G)=O((pair-cr(G))3/2) is due to O. Solé Pi [13].

The separation of the corresponding two parameters can also be observed at a local level. It was probably Ringel [14, 15], who first studied loc-cr(G), the local crossing number, of a graph G. It is defined as the minimum number k such that G can be drawn in such a way that the number of crossings along every edge is at most k. As above, we can also define loc-pair-cr(G), the local pair-crossing number of G, by counting the maximum number of different edges that cross a given edge, without multiplicities. In [2], Ackerman and Schaefer showed that if loc-pair-cr(G)2, then loc-pair-cr(G)=loc-cr(G). Slightly modifying the example depicted on Figure 1 of [9] (see also Exercise (7.1) in [16]), we can construct a graph G with 4=loc-pair-cr(G)<loc-cr(G)=5, showing that these two parameters are not always equal.

Graphs with local crossing number at most k are often called k-planar. For getting the best constant in the crossing lemma of Ajtai, Chvátal, Newborn, Szemerédi [3] and Leighton [8], it was useful to estimate the maximum number of edges that a k-planar graph with n vertices can have, for k4; see [1, 10, 9]. The crossing lemma, in turn, has many applications in discrete geometry, in additive combinatorics and elsewhere [19, 4].

A drawing of a graph is called simple (or a simple topological graph) if any two edges cross at most once. Hoffmann, Liu, Reddy, and C. D. Tóth [5] proved that there exists a function f(k)(3+o(1))k such that, if loc-cr(G)k, then G also has a simple drawing in which every edge has at most f(k) crossings.

In their seminal paper [17], Schaefer and Štefankovič proved the following (see also Lemma 9.2 in [16]). Suppose that the local pair-crossing number of G satisfies loc-pair-cr(G)k. Then one can redraw G in such a way that the total number of crossings along any edge is at most 2k. This implies that we have loc-cr(G)2k; see Remark 1 in the last section. In [2], Ackerman and Schaefer asked whether or not this bound can be improved. The aim of this note is to replace this by a polynomial upper bound.

Let G=(V,E) be a graph drawn in the plane. For each vertex vV, consider the cyclic order of the edges around v. The system of these cyclic permutations is called the rotation system of the drawing.

Theorem 2.

Let G be a graph that can be drawn in the plane such that every edge crosses at most k other edges, that is, loc-pair-cr(G)k. Then we have

loc-cr(G)k(k+1)(k+2).

That is, G can be redrawn in such a way that along every edge, the number of crossings is at most k(k+1)(k+2). We can also ensure that the rotation systems of the initial and final drawings are the same.

Combining this result with the theorem of Hoffmann et al. [5] mentioned above, we obtain the following.

Corollary 3.

Suppose that the local pair-crossing number of a graph G is at most k. Then G has a simple drawing such that along every edge there are at most 3k3+O(k2) crossings.

2 Proof of Theorem 2

We use standard terminology; see, e.g., [16]. Let G=(V,E) be a graph, which may have loops and multiple edges. A drawing of G, denoted by 𝒟(G), is a representation of G in which the vertices are distinct points in the plane, and the edges are Jordan arcs connecting the corresponding pairs of points. Whenever there is no danger of confusion, in notation and terminology, we make no distinction between the vertices and the points representing them, and the edges and the corresponding Jordan arcs. No edge is allowed to pass through a vertex, no three edges pass through the same point, and any two edges have only finitely many points in common. We may assume that no two edges “touch,” that is, if they have an interior point in common, then they must properly cross at this point. Otherwise we can locally pull apart the two edges near their touching point to get rid of the touching. We may also assume that there is no edge that crosses itself, otherwise we can shorten it by deleting the resulting loops.

Proof of Theorem 2.

Let G be a graph on n vertices and m edges, e1,,emE(G), and let 𝒟(G) be a drawing of G in the plane such that each edge in G crosses at most k other edges. Let be the rotation system of the drawing 𝒟(G). If G is not connected, we can pull apart the drawing 𝒟(G) so that the connected components do not interact with one another. So, we may assume without loss of generality that G is a connected graph, and let T be a spanning tree of G. Furthermore, let us assume that T={e1,,en1}, and for each i[n1], the edges e1,,ei form an edge-induced connected subgraph of G. We will need the following lemma.

Lemma 4.

Let TG and 𝒟(G) be as described above. Then for each eiT, there is a subcurve αiei in the plane such that

  1. 1.

    no two subcurves αi,αj cross each other properly (but one may have an endpoint on the other),

  2. 2.

    every vertex in G is an endpoint of some curve αi, and

  3. 3.

    the union 𝒯=i=1n1αi is a simply connected set in the plane.

Proof.

For each i[n1], we will greedily construct the subcurves α1,,αi with the properties described above, such that for 𝒯i=j=1iαj, 𝒯i is simply connected and contains all of the endpoints of the edges e1,,ei.

We start with α1=e1. Having defined α1,,αi1 with the desired properties, we obtain αi as follows. Consider edge eiT. Since the edges e1,,ei1 form an edge-induced connected subgraph of G, and the drawing 𝒯i1=j=1i1αj contains the endpoints of e1,,ei1, exactly one of the endpoints of ei lies in 𝒯i1. Let u be the endpoint of ei that does lie in 𝒯i1 and v be the other endpoint that does not lie in 𝒯i1. Then we define αi by starting at vertex v and tracing along the curve ei until we reach a point on 𝒯i1, which is either a crossing point or a vertex. Clearly, 𝒯i=𝒯i1αi is simply connected and contains the endpoints of the edges e1,,ei.

Let αiei, 1in1, be the subcurves from Lemma 4, and let 𝒯=i=1n1αi. We then define γ to be a simple closed curve drawn very close to 𝒯, such that 𝒯 lies inside of γ, that is, 𝒯 lies in the closed connected component of 2γ. Moreover, γ will not contain any crossing points in our drawing 𝒟(G), and apart from the crossing points on 𝒯, all other crossing points will lie outside of γ. For each subcurve αi𝒯, let βi be a simple closed curve drawn very close to αi, such that βi lies inside of γ, the endpoints of αi lie on βi, and all crossing points along αi lie inside of βi. Moreover, for ji, the crossing points along the curve αj lie outside of βi. Hence, either βi and βj are disjoint, or they are tangent if αi and αj have a common endpoint, or βi and βj have two points in common if an endpoint of αi lies in the interior of αj (or vice versa).

By construction, for each subcurve αi, there are tk edges ej1,,ejtE(G) that cross αi. Inside of βi, we can redraw each edge ej1,,ejt so that it intersects the curve αi at most once as follows. By starting at one of the endpoints of ejr, we trace along ejr until it intersects βi for the first time, which is right before it would intersect αi. We denote this subcurve as fjr (the first part of ejr). Likewise, by starting at the other endpoint of ejr, we again trace along ejr until it intersects βi for the first time, and denote this subcurve as jr (the last part of ejr). We then delete the part of ejr between fjr and jr (the middle part of ejr), and then connect fjr and jr by drawing a curve inside of βi from the endpoint of fjr on βi to the endpoint of jr on βi. Clearly, this can be done so that this new drawing of ejr crosses αi at most once. We can perform this redrawing procedure to each ej1,,ejt such that afterwards, each pair of edges ejr, ejw{ej1,,ejt} will have at most one crossing point between them inside of βi, and each such edge ejr crosses αi at most once. See Figure 1. After applying this redrawing procedure for each αi, let 𝒟(G) denote the new drawing of G in the plane. Hence, no additional crossings were created outside of γ in 𝒟(G), and the rotation system in the new drawing has not changed.

Refer to caption
(a) Before redrawing.
Refer to caption
(b) After redrawing.
Figure 1: For edge ei=uv, αi drawn in red, and βi drawn dashed, we redraw edges ej1 and ej2 inside of βi.

Given 𝒟(G), we now redraw the edges of G outside of γ as follows. First notice that for each edge eiE(G), ei will cross γ at most 2k+2 times in 𝒟(G). Indeed, starting at one of the endpoints of ei, we move along ei and first cross γ without crossing 𝒯. Continuing along ei, since ei will cross at most k subcurves αi, we have another possible 2k crossing points on γ. Finally, as we approach the other endpoint of ei, we cross γ one last time for a total of at most 2k+2 crossing points along γ. Hence, outside of γ, each edge ei will give rise to at most k+1 pairwise disjoint outer arcs, with end points on γ. See Figure 2. Hence, all of the edges in G will give rise to at most (k+1)|E(G)| outer arcs outside of γ. Given the circular structure of the endpoints of these outer arcs by γ, we can redraw all of the outer arcs outside of γ such that any two outer arcs cross at most once, and the endpoints of each outer arc remain the same. Moreover, if edges ei and ej did not cross outside of γ, then they still do not cross outside of γ in the new drawing. After we redraw the outer arcs as described above, let 𝒟′′(G) be the new and final drawing of G in the plane, and notice that the rotation system of 𝒟′′(G) is the same as the rotation system of 𝒟(G). Also notice that since each edge consists of at most k+1 outer arcs, two edges ei and ej will create at most (k+1)2 crossing points outside of γ in 𝒟′′(G).

Refer to caption
Figure 2: Example of an edge e with endpoints u and v, consisting of 5 outer arcs in bold.

Finally, we show that each edge ei in the final drawing 𝒟′′(G) has at most k(k+1)(k+2) crossing points. By the above, the number of crossing points on ei outside of γ is at most k(k+1)2, since ei crosses at most k other edges, and any two will create at most (k+1)2 crossing points outside of γ. In order to bound the number of crossing points on ei inside of γ, let us consider the following two cases.

Case 1.

Suppose that eiT. Recall that there are at most k subcurves αj that cross ei, and for each subcurve αj, there are at most k1 other edges that cross αj. Hence, ei will have at most k crossing points inside of βj, giving us a total of at most k2 crossing points on ei inside of γ.

Case 2.

Suppose eiT. Again, there are at most k subcurves αj that cross ei, and for each subcurve αj with ji, there are at most k1 other edges that cross αj. Moreover, ei could have an additional k crossing points along αi inside of βi. Hence, ei will have at most k2+k crossing points inside of γ.

Putting everything together, each edge in 𝒟′′(G) has at most k(k+1)2+k2+k=k(k+1)(k+2) crossings.

3 Concluding remarks

1.

Suppose G has local pair-crossing number at most k. If there is an edge with more than 2k crossings on it, then by applying Lemma 9.2 in [17], we can simplify the drawing without creating any new pair of crossing edges (that is, the local pair-crossing number remains at most k), and without increasing the number of crossings along any edge. Repeating this step for each edge of G, we end up with a drawing of G having local crossing number at most 2k.

2.

In the forthcoming journal version of this note, we show that it is possible to improve the bound in Theorem 2 to loc-cr(G)loc-pair-cr(G)2.5+o(1) by a more involved argument.

3.

Using the redrawing technique used in the proof of Theorem 2, one can show that for any multigraph G, cr(G)=O(pair-cr(G)5/3). Although this is weaker than the result stated in the introduction due to Solé Pi, we plan to include the proof in the forthcoming journal version for the interested reader.

References

  • [1] E. Ackerman. On topological graphs with at most four crossings per edge. Comput. Geom., 85:101574, 2019. doi:10.1016/J.COMGEO.2019.101574.
  • [2] E. Ackerman and M. Schaefer. A crossing lemma for the pair-crossing number. In Graph Drawing and Network Visualization: 22nd International Symposium, GD 2014, Würzburg, Revised Selected Papers, pages 222–233. Springer, 2014. doi:10.1007/978-3-662-45803-7_19.
  • [3] M. Ajtai, V. Chvátal, M. Newborn, and E. Szemerédi. Crossing-free subgraphs. In Theory and Practice of Combinatorics, North-Holland Math. Stud., pages 9–12. North-Holland, Amsterdam, 1982.
  • [4] G. Elekes. On the number of sums and products. Acta Arithmetica, 81(4):365–367, 1997.
  • [5] M. Hoffmann, C.-H. Liu, M. M. Reddy, and C. D. Tóth. Simple topological drawings of k-planar graphs. In Graph Drawing and Network Visualization: 28nd International Symposium, GD 2020, Vancouver, Revised Selected Papers, pages 390–402. Springer, 2020. doi:10.1007/978-3-030-68766-3_31.
  • [6] J. Karl and G. Tóth. Crossing lemma for the odd-crossing number. Comput. Geom., 108:101901, 2023. doi:10.1016/J.COMGEO.2022.101901.
  • [7] J. R. Lee. Separators in region intersection graphs. arxiv:1608.01612, 2016.
  • [8] F.T. Leighton. Complexity issues. In VLSI: Optimal Layouts for the Shuffle-Exchange Graph and Other Networks. MIT Press, Cambridge, MA, 1983.
  • [9] J. Pach, R. Radoičić, G. Tardos, and G. Tóth. Improving the crossing lemma by finding more crossings in sparse graphs. Discrete Comput. Geom., 36:527–552, 2006. doi:10.1007/S00454-006-1264-9.
  • [10] J. Pach and G. Tóth. Graphs drawn with few crossings per edge. Combinatorica, 17:427–439, 1997. doi:10.1007/BF01215922.
  • [11] J. Pach and G. Tóth. Thirteen problems on crossing numbers. Geombinatorics, 9:194–207, 2000.
  • [12] J. Pach and G. Tóth. Which crossing number is it anyway? J. Combin. Theory Ser. B, 80:225–246, 2000. doi:10.1006/JCTB.2000.1978.
  • [13] O. S. Pi. Pair crossing number, cutwidth, and good drawings on arbitrary point sets. Discrete Comput. Geom., 73:310–326, 2025. doi:10.1007/S00454-024-00708-Z.
  • [14] G. Ringel. Ein Sechsfarbenproblem auf der Kugel. Abh. Math. Sem. Univ. Hamburg, 29:107–117, 1965.
  • [15] G. Ringel. A nine color theorem for the torus and the Klein bottle. In The theory and applications of graphs (Kalamazoo, Mich., 1980), pages 507–515. Wiley, New York, 1981.
  • [16] M. Schaefer. Crossing Numbers of Graphs. CRC Press, 2017.
  • [17] M. Schaefer and D. Štefankovič. Decidability of string graphs. J. Comput. System Sci., 68:319–334, 2004. doi:10.1016/J.JCSS.2003.07.002.
  • [18] J. Matoušek. Near-optimal separators in string graphs. Combin. Probab. Comput., 23:135–139, 2014. doi:10.1017/S0963548313000400.
  • [19] J. Spencer, E. Szemerédi, and W. Trotter. Unit distances in the Euclidean plane. In Graph theory and combinatorics (Cambridge, 1983), pages 293–303. Academic Press, London, 1984.
  • [20] G. Tóth. Note on the pair-crossing number and the odd-crossing number. Discrete Comput. Geom., 39:791–799, 2008. doi:10.1007/S00454-007-9024-Z.
  • [21] G. Tóth. A better bound for the pair-crossing number. In Thirty Essays on Geometric Graph Theory, pages 563–567. Springer, New York, NY, 2013.
  • [22] P. Valtr. On the pair-crossing number. In Combinatorial and Computational Geometry, Math. Sci. Res. Inst. Publ., pages 569–575. Cambridge Univ. Press, Cambridge, 2005.