This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

The maximum crossing number of C3×C3C_{3}\times C_{3}

Michael Haythorpe Alex Newcombe
Abstract

We determine that the maximum crossing number of C3×C3C_{3}\times C_{3} is 78, which closes the previously best known range of between 68 and 80. The proof uses several techniques which may be useful in determining the maximum crossing number of other graphs.

keywords:
Maximum Crossing Number , Cartesian Product
MSC:
05C10 , 68R10
\usetikzlibrary

patterns \usetikzlibrarycalc \usetikzlibraryintersections \usetikzlibrarydecorations.text \usetikzlibrarydecorations.pathreplacing \usetikzlibraryshapes.geometric \usetikzlibrarypositioning

1 Introduction

It was shown by Piazza et al [6] that the maximum crossing number of C3×C3C_{3}\times C_{3} lies somewhere between 68 and 80, inclusive. We resolve this by showing that the maximum crossing number is exactly 78. The proof uses a mixture of theoretical arguments as well as exhaustive computer searches. Although the proofs are ad hoc, we believe that the general techniques could be used to determine the maximum crossing number for other, similarly sized graphs.

2 Preliminaries

We assume that the reader is familiar with the usual definitions from the theory of crossing numbers and graph drawings. For more detailed descriptions, we recommend [7]. A graph GG has the vertex set V(G)V(G) and edge set E(G)E(G). Let CnC_{n} denote the cycle graph on nn vertices, and let PmP_{m} denote the path graph on mm edges. Let G×HG\times H denote the graph Cartesian product between graphs GG and HH. For example, C3×C3C_{3}\times C_{3} is displayed in Figure 1 (a)(a). A drawing, D(G)D(G), provides a mapping of E(G)E(G) and V(G)V(G) into the plane. Vertices are mapped to distinct points and each edge e={u,v}e=\{u,v\} is mapped to a conitnuous arc between the points associated with uu and vv in such a way that the interior of the arc does not contain any points associated with vertices. In addition, the interiors of the arcs are only allowed to intersect at singleton points, these are the crossings and the number of crossings of a drawing is denoted crD(G)cr_{D}(G). We shall refer to the points and arcs given by a drawing as the ‘vertices’ and ‘edges’ of the drawing. A drawing is a good drawing if an edge never crosses itself, incident edges do not cross each other and pairs of edges cross at most once. All drawings considered here are good and when we use the word ‘drawing’, we will mean ‘good drawing’. The maximum crossing number, denoted as maxcr(G)\operatorname{\text{maxcr}}(G), is the maximum number of edge crossings in any good drawing of GG. As first described by Conway and Woodall [8], the thrackle number of GG, denoted Th(G)Th(G), is the number

Th(G):={u,v}E(G)12(|E(G)|d(v)d(u)+1),Th(G):=\sum_{\{u,v\}\in E(G)}\frac{1}{2}(|E(G)|-d(v)-d(u)+1), (1)

where d(v)d(v) is the degree of vertex vv. Obviously, for a given graph, the maximum crossing number may not be equal to the thrackle number, but if it is, then GG is thrackleable. Given a drawing DD of a graph GG, if a pair of non-incident edges do not cross in DD, then we say that they are a missed pair. Thus, for any good drawing DD, crD(G)cr_{D}(G) is equal to the thrackle number minus the number of missed pairs. Note that the graph C4C_{4} is not thrackleable, and similarly, it was shown in [4] that the bowtie graph, displayed in Figure 1 (b)(b), is also not thrackleable. Thus, any drawing of C4C_{4} or a bowtie has at least one missed pair. The sub-thrackle number, denoted STh(G)STh(G), is Th(G)Th(G) minus the number of subgraphs of GG that are isomorphic to C4C_{4} plus the number subgraphs isomorphic to the complete graph on 4 vertices. It follows that, for any graph GG, maxcr(G)STh(G)\operatorname{\text{maxcr}}(G)\leq STh(G). For a graph GG, we define a prescription, denoted PP, to be a mapping P:E(G)2E(G)P:E(G)\rightarrow 2^{E(G)}, where 2E(G)2^{E(G)} is the power set of E(G)E(G). For each edge ee, P(e)P(e) gives an unordered set of edges, which we call the virtual crossings of ee. If a drawing DD of GG is such that each edge ee crosses exactly those edges in P(e)P(e), then DD satisfies the prescription PP. Clearly, any drawing of GG satisfies some prescription PP, however, it may be the case that there is no drawing that satisfies a given PP. We shall be concerned with identifying when it is impossible to satisfy PP.

(a)(a)(b)(b)
Figure 1: The graph C3×C3C_{3}\times C_{3} is shown in (a)(a) and a bowtie graph is shown in (b)(b).

After discussing some technical lemmas in Section 3, we shall show the following in Section 4

Theorem 2.1

The maximum crossing number of C3×C3C_{3}\times C_{3} is 78.

3 Properties used in the proof of Theorem 2.1

Given a drawing of a graph, let m(A,B)m(A,B) be the number, modulo 2, of missed pairs with one of the edges belonging to the edge set AA and the other edge belonging to edge set BB. A property that has been used several times before, e.g. see [8], is as follows

Lemma 3.1

Let AA and BB be two vertex disjoint subgraphs of a graph GG, where AA is a cycle of length nn and BB is a cycle of length mm. Then, in any drawing of GG, m(E(A),E(B))=nm(mod2)m(E(A),E(B))=nm\pmod{2}.

For a prescription PP of GG, we shall say that PP satisfies property 1 if, for the edges of any two vertex disjoint cycles of GG, the virtual crossings given by PP do not contradict the condition in Lemma 3.1.

The graph displayed in Figure 2 (a)(a) is C3×P1C_{3}\times P_{1} (sometimes called the envelope graph) and C3×C3C_{3}\times C_{3} contains 9 subgraphs isomorphic to this graph. We have the following

Lemma 3.2

The maximum crossing number of C3×P1C_{3}\times P_{1} is 15.

Proof.   Figure 2 (b)(b) displays C3×P1C_{3}\times P_{1} drawn with 15 crossings which provides a lower bound and STh(C3×P1)=15STh(C_{3}\times P_{1})=15 which provides the matching upper bound. \qed

Next, C3×P1C_{3}\times P_{1} is small enough to use an exhaustive computer search to confirm that there are no drawings of C3×P1C_{3}\times P_{1} with exactly 14 crossings. Hence, every drawing of C3×P1C_{3}\times P_{1} has either 15 crossings or fewer than 14 crossings. We note that for larger graphs, exaustive searches such as this quickly become intractable.

For a prescription PP of GG, we shall say that PP satisfies property 2 if, for any subgraph of GG which is isomorphic to C3×P1C_{3}\times P_{1}, the virtual crossings given by PP, restricted to the subgraph, provide a total of either 15 or fewer than 14 virtual crossings.

Next, consider a graph consisting of a cycle of length 6, and one additional edge which forms a triangle. In the notation of [3], this is a DB(5,3,1)DB(5,3,-1) graph. In [5], Harborth found all thrackleable graphs on six vertices, which does not include DB(5,3,1)DB(5,3,-1) and so there must exist at least one missed pair in any drawing of DB(5,3,1)DB(5,3,-1). Hence, along with the drawing of DB(5,3,1)DB(5,3,-1) in Figure 2 (c)(c), which has 10 crossings, we conclude that maxcr(DB(5,3,1))=Th(DB(5,3,1))1=10\operatorname{\text{maxcr}}(DB(5,3,-1))=Th(DB(5,3,-1))-1=10.

For a prescription PP of GG, we shall say that PP satisfies property 3 if, for any subgraph of GG which is isomorphic to DB(5,3,1)DB(5,3,-1), the virtual crossings given by PP, restricted to the subgraph, provide at most 10 virtual crossings.

(a)(a)(b)(b)(c)(c)
Figure 2: The graph C3×P1C_{3}\times P_{1} is shown in (a)(a). C3×P1C_{3}\times P_{1} is drawn with 15 crossings in (b)(b). DB(5,3,1)DB(5,3,-1) is drawn with 10 crossings in (c)(c).

4 Proof of Theorem 2.1

4.1 Lower bound

The lower bound is determined by demonstrating a drawing of C3×C3C_{3}\times C_{3} with 78 crossings. The drawing was found using a modified version of the largely successful ‘planarisation method’ for minimising crossings [1, 2]. The modifications will be discussed in and upcoming publication and they allow us to attempt to maximise crossings by utilising longest paths. Figure 3 displays a drawing of C3×C3C_{3}\times C_{3} with 78 crossings and so maxcr(G)78\operatorname{\text{maxcr}}(G)\geq 78.

Refer to caption
Figure 3: C3×C3C_{3}\times C_{3} drawn with 78 crossings.

4.2 Upper bound

First, let G=C3×C3G=C_{3}\times C_{3}, and let GvG-v be the graph obtained by deleting vertex vv from GG. Note that Th(Gv)=55Th(G-v)=55. We will now determine that maxcr(Gv)=44\operatorname{\text{maxcr}}(G-v)=44. This is done in the following way. There are 5 different 4-cycles along with 4 bowtie subgraphs in GvG-v, each of which must have a missed pair and moreover, no missed pair is double counted. Hence there must be at least 9 missed pairs in any drawing of GvG-v.

To show that there is not exactly nine missed pairs in GvG-v, we do the following exhaustive search. We consider all possible combinations of 9 missed pairs, and each combination provides a prescription PP of GvG-v. Note that a prescription PP must satisfy properties 1-3 of Section 3, otherwise, there does not exist a drawing of GG that satisfies PP. The exhaustive search determines that there are no such sets of 9 missed pairs, whose resulting prescription satisfies properties 1-3.

To show that there is not exactly 10 missed pairs, we consider all possible combinations of 10 missed pairs in GvG-v and look for those whose resulting prescription satisfies properties 1-3. There are 74 such sets, and for each, we have a prescription PP for GvG-v. For each edge eE(Gv)e\in E(G-v), P(e)P(e) gives a set of edges (the virtual crossings) that ee is to cross, however, we do not know the order in which they might be crossed. We would like to consider all possible orderings and show that it is impossible to produce such a drawing, but this is beyond the realm of tractability. So instead we do this heuristically in the following way. Consider a subgraph of GvG-v and the virtual crossings given by PP, restricted to the edges within the subgraph. For some subgraphs, it is tractable to consider all possible orderings of the virtual crossings given by PP. So, we search for a subgraph for which we are able to verify that it is impossible to draw this subgraph in a way which satisfies the virtual crossings given by PP, and hence, the supergraph GvG-v also cannot be drawn to satisfy PP. For each of the 74 prescriptions, we were able to find such a subgraph. At the completion of this process, we have shown that there must be at least 11 missed pairs in any drawing of GvG-v. Hence, we have that maxcr(Gv)44\operatorname{\text{maxcr}}(G-v)\leq 44 and Figure 4 displays GvG-v drawn with exactly 44 crossings which, together, show that maxcr(Gv)=44\operatorname{\text{maxcr}}(G-v)=44.

Refer to caption
Figure 4: C3×C3C_{3}\times C_{3}, minus one vertex, drawn with 44 crossings.

Next, there are 9 possibilites to delete a vertex from GG and in any given drawing of GG, each crossing appears in exactly 5 of the 9 subdrawings corresponding to the subgraph of GvG-v, hence, for any drawing DD of GG

crD(G)=15vV(G)crD(Gv)15(9×44)=79.2cr_{D}(G)=\frac{1}{5}\sum_{v\in V(G)}cr_{D}(G-v)\leq\frac{1}{5}(9\times 44)=79.2 (2)

and so maxcr(G)79\operatorname{\text{maxcr}}(G)\leq 79. To complete the argument, note that (2) shows that there is at least one more missed pair than observed by Piazza et al in [6]. There are two possible cases to consider. Firstly, if this additional missed pair is formed by edges from two vertex disjoint triangles, then by the arguments in [6], there is also a second additional missed pair and we would obtain maxcr(G)78\operatorname{\text{maxcr}}(G)\leq 78.

The only other possible case is that the additional missed pair comes from a bowtie subgraph in GG. That is, there is a bowtie in GG with at least two missed pairs. Observe that each bowtie of GG is contained in 4 of the 9 subgraphs isomorphic to GvG-v. We now show that if DD is a drawing of GvG-v with two missed pairs on any bowtie, then crD(Gv)<maxcr(Gv)cr_{D}(G-v)<\operatorname{\text{maxcr}}(G-v). This again is done by an exhaustive search. We consider all possible combinations of 11 missed pairs of GvG-v, such that there is a bowtie subgraph with at least 2 of the missed pairs and such that properties 1-3 are still satisfied. There are 13 such combinations, and for each of these, we find a subgraph that cannot be drawn to satisfy the corresponding prescription. Hence we have shown that there is no drawing of GvG-v with maxcr(Gv)\operatorname{\text{maxcr}}(G-v) crossings where one of the bowtie subgraphs has at least two missed pairs. That is, in this case,

crD(G)=15vV(G)crD(Gv)15(5×44+4×43)=78.4.cr_{D}(G)=\frac{1}{5}\sum_{v\in V(G)}cr_{D}(G-v)\leq\frac{1}{5}(5\times 44+4\times 43)=78.4. (3)

We can now conclude that, in either case, maxcr(G)78\operatorname{\text{maxcr}}(G)\leq 78, which completes the proof of Theorem 2.1.

References

  • [1] C. Buchheim, M. Chimani, C. Gutwenger, M. Jünger, P. Mutzel. Crossings and Planarization. Handbook of Graph Drawing and Visualization, Roberto Tamassia (ed.), pp. 43–80, 2013.
  • [2] K. Clancy, M. Haythorpe, A. Newcombe. An effective crossing minimisation heuristic based on star insertion. J. Graph Alg. App., 23(2):135–166, 2019.
  • [3] R. Fulek, J. Pach. A computational approach to Conway’s thrackle conjecture. Computational Geometry, 44:345–355, 2011.
  • [4] H. Gronau, H. Harborth. Numbers of nonisomorphic drawings for small graphs. Congr. Numer., 71:105–114, 1990.
  • [5] H. Harborth, S. Zahn. Maximum Number of Crossings in Drawings of Small Graphs. In: Graph Theory, Combinatorics and Algorithms: Proceedings of the Seventh Quadrennial International Conference on the Theory and Applications of Graphs, pp. 485–495, 1995.
  • [6] B.L. Piazza, R.D. Rigeisen, S.K. Stueckle. Subthrackleable graphs and four cycles. Disc. Math., 127:265–276, 1994.
  • [7] M. Schaefer. Crossing Numbers of Graphs. Series: Discrete Mathematics and Its Applications, CRC Press, 2017.
  • [8] D.R. Woodall. Thrackles and deadlock. In: eds. Welsh, D.A.J., Combinatorial Mathematics and its Applications, Proceedings of a Conference held at the Mathematical Institute, Oxford, pp. 335-348, 1969.