1. Introduction
The issue of reducing the number of crossings on edges of simple graphs is interesting in a lot of areas. Probably one of the most popular areas is the implementation of the VLSI layout because it caused a significant revolution in circuit design and thus had a strong effect on parallel calculations. Crossing numbers have also been studied to improve the readability of hierarchical structures and automated graphs. The visualized graph should be easy to read and understand. For the sake of clarity of graphic drawings, some reduction of an edge crossing is probably the most important. Note that examining number of crossings of simple graphs is an NP-complete problem by Garey and Johnson [
1].
The crossing number cr
of a simple graph
G with the vertex set
and the edge set
is the minimum possible number of edge crossings in a drawing of
G in the plane (for the definition of a drawing see Klešč [
2]). One can easily verify that a drawing with the minimum number of crossings (an optimal drawing) is always a good drawing, meaning that no two edges cross more than once, no edge crosses itself, and also no two edges incident with the same vertex cross. Let
D be a good drawing of the graph
G. We denote the number of crossings in
D by cr
. Let
and
be edge-disjoint subgraphs of
G. We denote the number of crossings between edges of
and edges of
by cr
, and the number of crossings among edges of
in
D by cr
. For any three mutually edge-disjoint subgraphs
,
, and
of
G by [
2], the following equations hold:
Throughout this paper, some parts of proofs will be based on Kleitman’s result [
3] on crossing numbers for some complete bipartite graphs
on
vertices with a partition
and
containing an edge between every pair of vertices from
and
of sizes
m and
n, respectively. He showed that
For an overview of several exact values of crossing numbers for specific graphs or some families of graphs, see Clancy [
4]. The main goal of this survey is to summarize all such published results for crossing numbers along with references also in an effort to give priority to the author who published the first result. Chapter 4 is devoted to the issue of crossing numbers of join product with all simple graphs of order at most six mainly due to unknown values of
for both
more than six in (
1). The join product of two graphs
and
, denoted
, is obtained from vertex-disjoint copies of
and
by adding all edges between
and
. For
and
, the edge set of
is the union of the disjoint edge sets of the graphs
,
, and the complete bipartite graph
. Let
denote the discrete graph (sometimes called empty graph) on
n vertices, and let
be the complete graph on
n vertices. The exact values for crossing numbers of
for all graphs
G of order at most four are given by Klešč and Schrötter [
5], and also for a lot of connected graphs
G of order five and six [
2,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24]. Note that
are known only for some disconnected graphs
G, and so the purpose of this paper is to extend known results concerning this topic to new disconnected graphs [
25,
26,
27,
28].
Let
be the disconnected graph of order five consisting of two components isomorphic to the complete graphs
and
, respectively, and let also
. We cannot determine the crossing number of the join product
by a similar technique like in [
2,
18] because
. From the topological point of view, number of crossings of any drawing
D of
placed on surface of the sphere does not matter which of regions is unbounded, but on how many times edges of the graph
could be crossed by a subgraph
in
D. This representation of
best describes idea of a configuration utilizing some cyclic permutation on pre-numbered vertices of
.
Theorem 1. and for , i.e., for n even and for n odd at least 3.
All subcases of the proof of Theorem 2 will be clearer if a graph of configurations
is used as a graphical representation of minimum numbers of crossings between two different subgraphs. Moreover, in the case of our symmetric graph
, the graph
can be linked to parity properties of configurations. Our proof of the main Theorem 2 is therefore an inevitable combination of topological analysis of existing configurations with their parity properties. The color resolution of weighted edges in
will also serve us for a simpler description of existence of its possible subgraphs in the examined drawing
D of
. Software COGA [
29] should be also very helpful in certain parts of presented proofs mainly due to possibility of generating all cyclic permutations of five elements and counting of their subsequent interchanges of adjacent elements.
The obtained crossing number of the join product is in very special form which is caused by a completely different behavior for n even and odd number. The paper concludes by giving crossing numbers of and with same values in Corollaries 3 and 4, respectively, that is something unique in the crossing number theory.
2. Cyclic Permutations and Corresponding Configurations
The join product
(sometimes used notation
) consists of one copy of the graph
and
n vertices
, and any vertex
is adjacent to every vertex of the graph
. We denote the subgraph induced by five edges incident with the fixed vertex
by
, which yields that
We consider a good drawing
D of
. By the rotation
of a vertex
in
D we understand the cyclic permutation that records the (cyclic) counterclockwise order in which edges leave
, as defined by Hernández-Vélez et al. [
30] or Woodall [
31]. We use the notation
if the counter-clockwise order of edges incident with the fixed vertex
is
,
,
,
, and
. We recall that rotation is a cyclic permutation. By
, we understand the inverse permutation of
. In the given drawing
D, it is highly desirable to separate
n subgraphs
into three mutually disjoint subsets depending on how many times edges of
could be crossed by
in
D. Let us denote by
and
the set of subgraphs for which
and
, respectively. Edges of
are crossed by each remaining subgraph
at least twice in
D.
First, note that if
D is a drawing of the join product
with the empty set
, then
enforces at least
crossings in
D provided by
Based on this argument, we will only consider drawings of the graph for which there is a possibility to obtain a subgraph . Moreover, let denote the subgraph for any , which yields that each such subgraph is represented by its .
Let us discuss all possible subdrawings of
induced by
D. As edges of its subgraph isomorphic to
do not cross each other, it is obvious there are only two such possible drawings of
presented in
Figure 1.
Assume there is a good drawing
D of
with planar subdrawing of the graph
induced by
D and also the vertex notation of
in such a way as shown in
Figure 1a. Our aim is to list all possible rotations
which can appear in
D if edges of
are not crossed by
. Since there is only one subdrawing of
represented by the rotation
, there are three possibilities to obtain the subdrawing of
without the edge
depending on in which region both edges
and
are placed. Of course, there are two next ways how to place the corresponding two edges together with the edge
for each mentioned case. These
possibilities under our consideration can be denoted by
, for
. We will call them by the
configurations of corresponding subdrawings of the subgraph
in
D and suppose their drawings as shown in
Figure 2.
In the rest of the paper, we present a cyclic permutation by the permutation with 1 in the first position. Thus, the configurations , , , , , and are represented by the cyclic permutations , , , , , and , respectively. Clearly, in a fixed drawing of the graph , some configurations from need not appear. We denote by the set of all configurations that exist in the drawing D belonging to the set .
Let
,
be two configurations from
(not necessary distinct). We denote the number of edge crossings between two different subgraphs
and
with
and
in
D by
. Finally, let
among all good drawings of
with the planar subdrawing of
induced by
D given in
Figure 1a and with
. Our aim shall be to establish
for all pairs
. In particular, the configurations
and
are represented by the cyclic permutations
and
, respectively. Each subgraph
with
crosses edges of each
with
at least once provided by the minimum number of interchanges of adjacent elements of
required to produce
is one, i.e.,
. For more details see also Woodall [
31]. The same reason gives
,
,
,
,
,
,
,
,
,
,
,
,
, and
. Clearly, also
for any
. The lower bounds obtained for number of crossings between two configurations from
are summarized in the symmetric
Table 1 (here,
and
with
). Note that these values cannot be increased, i.e., the lower bounds can be achieved in some subdrawings of
for
with desired configurations.
Further, due to symmetry of mentioned configurations, let us define two functions
Let
be the functions obtained by applying
and
on corresponding cyclic permutations of configurations in
, respectively. Thus, we have
Therefore it is not difficult to show that values in rows of
Table 1 can be obtained by successive application of the mentioned transformations
and
.
3. The Graph of Configurations and Parity Properties
Low possible number of crossings between two different subgraphs from the nonempty set
is one of main problems in determining
, and graph of configurations as a graphical representation of
Table 1 is going by useful tool in our research. This idea of representation was first introduced in [
26].
Let
D be a good drawing of
with the planar subdrawing of
induced by
D given in
Figure 1a, and let
be nonempty set of all configurations that exist in
D belonging to
. A graph of configurations
is an ordered triple
, where
is the set of vertices,
is the set of edges formed by all unordered pairs of two vertices (not necessary distinct), and a weight function
that associates with each edge of
an unordered pair of two vertices of
. The vertex
if the corresponding configuration
for some
. The edge
if
and
are two vertices of
. Finally,
for the edge
, if
m is associated lower bound between two configurations
and
in
Table 1. Based on that
is an undirected edge-weighted graph without multiple edges uniquely determined by
D and is also subgraph of
induced by
if we define
in the same way over
. The graph
corresponds to the edge-weighted complete graph
in
Figure 3, and thus will follow all subcases in the proof of the main Theorem 2 more clearly. In the rest of
Figure 3, let any loop of the mentioned graph
be presented by circle around vertex with respect to weight 4.
Let denote the number of all subgraphs with the configuration of for each . So, if we denote by and , then . Moreover, for a better understanding, we get for all : if and only if there is a subgraph with the configuration of if and only if in the graph .
Now, let us assume the configurations
of
,
of
, and
of
. The reader can easily find a subdrawing of
in which
,
, and
, i.e.,
. Further, there is a possibility to add another subgraph
that crosses edges of the graph
four times. We have to emphasize that the vertex
must be placed in the triangular region with three vertices of
on its boundary (in the subdrawing of
), i.e.,
and the subgraph
is represented by
. Clearly, the number of adding crossings cannot be smaller than 4 according to the well-known fact that
. This situation suggests one natural problem which requires the following definition of a new number
. If
,
, and
, then let us denote by
the number of subgraphs
with
. It is obvious that any subgraph
satisfies the condition
with the configurations
of
,
of
, and
of
, and the number of
that cross the graph
exactly six times is at most
. Due to symmetry of some configurations, it is appropriate to use the transform functions
defined above and by the similar way, we can also define the numbers
for any
. Thus, if
,
, and
or
,
, and
or
,
, and
or
,
, and
or
,
, and
, then let us denote by
or
or
or
or
the number of subgraphs
represented by the rotation
or
or
or
or
, respectively. The importance of the values
will be presented in the proof of the main Theorem 2 as parity properties (
6) and (
7).
4. The Crossing Number of
A drawing D of is said to be antipode-free if for any two different vertices and . In the proof of Theorem 2, the following statements related to some restricted subdrawings of the graph are required.
Lemma 1. Let D be a good and antipode-free drawing of , , with the vertex notation of the graph in such a way as shown in Figure 1a. For , let be a configuration of the corresponding subgraph for some . If there is a such that , then all possible are given in Table 2. Proof. Assume the configuration
of the subgraph
for some
, i.e.,
. The subdrawing of
induced by
D contains just five regions with
on their boundaries, see
Figure 2. If there is a
such that
, then the vertex
must be placed in the region with the four vertices
,
,
, and
of
on its boundary. Besides that only the edge
of
can be crossed by
, and
is fulfilling for
with
if
crosses
. The same idea also force that the rotations of the vertex
are
,
,
,
, and
for the remaining configurations
,
,
,
, and
of
, respectively. □
Corollary 1. Let D be a good and antipode-free drawing of , for , with the vertex notation of the graph in such a way as shown in Figure 1a. If , and are three different subgraphs with , and such that , , and have three mutually different configurations from any of the sets , , , , , and , theni.e., Proof. Let us assume the configurations of , of , and of with respect to the restrictions and recall that they are represented by the cyclic permutations , , and . If there is a subgraph with , then the subgraph can be represented only by , where the edge crosses of and either or crosses corresponding edge of . Any such subgraph must cross edges of both subgraphs and at least twice because the minimum number of interchanges of adjacent elements of required to produce and is two. Clearly, if or , we obtain the desired result . Further, if and , then the edge is crossed by in and also by in , respectively. However, then , which contradicts the fact that in .
If there is a with , then the subgraph is represented only by the cyclic permutation . Using same properties as in the previous subcase, we have and . This in turn implies that . Of course, we can apply the same idea for the case of .
To finish the proof, let us consider a subgraph with , , and . This enforces that the minimum number of interchanges of adjacent elements of required to produce , , and must be exactly two. However, it is not difficult to show that such cyclic permutation does not exist. Similar arguments can be applied for remaining five cases (or using the transformations and ), and the proof is complete. □
Corollary 2. Let D be a good and antipode-free drawing of , for , with the vertex notation of the graph in such a way as shown in Figure 1a. If , and are three different subgraphs such that , , and have three mutually different configurations from any of the sets and , theni.e., Proof. Let us assume the configurations
of
,
of
, and
of
. If there is a subgraph
with
, then the subgraph
can be represented only by the cyclic permutations
. Uniqueness of all rotations in
Table 2 confirms that
and
. Hence,
, and the similar way can be applied for the case if
or
with
. It remains to consider the case where
,
, and
, which yields that
clearly holds for any such
, as claimed. The proof proceeds in the similar way for the second triple of configurations
, and this completes the proof. □
Lemma 2. .
Proof. If we consider the configurations
of
and
of
, then one can easily find a subdrawing of
in which
, i.e.,
. The graph
contains a subgraph that is a subdivision of the complete graph
and it is well-known by Guy [
32] that
. As
, the proof of Lemma 2 is complete. □
Theorem 2. and for , i.e., for n even and for n odd at least 3.
Proof. The graph
is planar, hence
. For
, both special drawings in
Figure 4 and
Figure 5 produce
crossings, and so
. The opposite inequality can be proved by induction on
n, and the result holds for
by Lemma 2. For some
, suppose a drawing
D of
with
and that
Let us first show that
D must be antipode-free. Suppose that, without loss of generality,
. If at least one of
and
, say
, does not cross
, it is not difficult to verify in
Figure 1 that
, i.e.,
. By (
1), we already know that
, which yields that edges of the subgraph
must be crossed at least four times by each other
. So, by fixing the subgraph
in
D, we have
The obtained crossing number contradicts the assumption (
3) and confirms that the considered drawing
D is antipode-free. For easier reading, if
and
, then again (
3) together with
using (
1) imply the following inequality with respect to possible edge crossings of
in
D:
The inequality (
5) forces more than
subgraphs
by which edges of
are not crossed, that is,
and
. Of course, if
n is odd then previous inequalities could be strengthened, but this is not necessary in the following process of obtaining a contradiction with number of crossings in
D. Moreover, if
then
, and
with the assumption (
3) enforce
n at least four.
Case 1:
and choose the vertex notation of the graph
in such a way as shown in
Figure 1a. In this case, we deal with configurations from the nonempty set
. As the set
is nonempty, recall that
Let us first suppose that either
or
. For the rest of the proof we may therefore assume that
, that is,
. Since
is the subgraph of
induced by
with respect to weights 2 of all its edges (without possible loops), three possible subcases presented in
Figure 6 may occur:
- (a)
for each
. Let us assume three subgraphs
such that
,
and
have three mutually different configurations from the set
. Then,
holds for any other
by summing values in corresponding three rows of
Table 1, and
is true by Corollary 2 for any
. Then, by fixing the graph
- (b)
Assuming that
for exactly two
, without lost of generality, let us consider two different subgraphs
such that
and
have configurations
and
, respectively. As
, we have
for any
,
. Therewith, the antipode-free property of
D forces that,
trivially holds for any subgraph
with
. Hence, by fixing the graph
- (c)
for only one
. As
, in the rest of the paper, we may consider
with the configuration
of
. Then edges of each other subgraph
cross at least four times edges of
provided by
. Thus, by fixing the graph
All three subcases contradict the assumption (
3). In addition, let us suppose that
and
. Remark that the subgraph
can be either connected (consisting of a single component) or also disconnected with several components. Now, we are able to discuss over remaining possible components of
in the following subcases:
There are no two adjacent edges with weights 1 in the subgraph
, that is, there are four possibilities presented in
Figure 7.
for some
,
, i.e., there are three cases mentioned in
Figure 7a–c. Let us consider two subgraphs
such that
have different configurations from
, where
are associated indexes. Using weights of edges in the considered component of
, one can easily verify that edges of the graph
are crossed at least five times by edges of any another subgraph
. Moreover, since the minimum number of interchanges of adjacent elements of
required to produce
is three, any subgraph
with
crosses edges of
at least thrice. Thus, by fixing the graph
for all
,
, i.e., there is only one case mentioned in
Figure 7d. Let us again consider two subgraphs
such that
have different configurations from
, where
are associated indexes. Then,
holds by summing edge-weights 4 and 3 for any other
. Hence, by fixing the graph
Both discussed cases again confirm a contradiction with (
3) in
D, and so, suppose that there are two adjacent edges with weights 1 in the subgraph
. Further, only in the case if the number
is defined, we claim that the following two properties (
6) and (
7) must be also fulfilled in
D:
For a contradiction, suppose, without loss of generality, that
, that is,
. In this case, from the definition of
, we have
,
, and
. Thus, in the rest of the paper, let us consider three subgraphs
such that
,
, and
have configurations
,
, and
, respectively. Using values in
Table 1, one can easily verify that edges of the graph
are crossed at least six times and seven times by edges of any another subgraph
with the configuration
and
of
(of course, if
for some
in
D), respectively. However, from Corollary 1 we get that
holds for any
provided by we can also assume that
and
due to the congruence property (If
and
are two cyclic permutations of odd length, and
denotes the minimum number of interchanges of adjacent elements of
required to produce the inverse cyclic permutation of
, then
for some nonnegative integer
z, for more see Woodall [
31]). Hence, by fixing the graph
The obtained crossing number also contradicts the assumption (
3) of
D and confirms that both parity properties (
6) and (
7) must be fulfilled in
D.
There are two adjacent edges with weights 1 in the subgraph
, that is, there are five possibilities presented in
Figure 8.
- (a)
Let the graph
consist of one component in such a way as shown in
Figure 8a. Without lost of generality, let us assume that
are vertices of the considered path on three vertices with weight 1 of both edges. In this case, it is obvious that
. Since the number
can be defined, the property (
6) forces
. Further, let us also assume that
with the configuration
of
. Then, by fixing the graph
- (b)
Let the graph
consist of one component in such a way as shown in
Figure 8b. Without lost of generality, let us assume that
are vertices of the considered path on three vertices with weight 1 of both edges and let
be vertices of the 3-cycle with respect to weight 2 of all its edges. In this case, it is obvious that
. The property (
6) enforces again
because the number
can be defined. Further, if
is assumed with the configuration
of
, then by fixing the graph
- (c)
Let the graph
consist of one component in such a way as shown in
Figure 8c–e. Let us take a maximal path
on
k vertices as the subgraph of
with weights 1 on all its edges. If
and
are two inner vertices of
with
for which the numbers
and
satisfy the parity properties (
6) and (
7), then addition of both inequalities thus obtained enforces a contradiction
The obtained contradictions in all three cases complete the proof for the planar subdrawing of
induced by
D given in
Figure 1a.
Case 2.
and choose the vertex notation of the graph
presented as in
Figure 1b. Since the set
is nonempty and there is only one subdrawing of a subgraph
for all
represented by the rotation
, the subgraph
is crossed at least four times by edges of each subgraph
with
. Hence, by fixing the graph
For all these mentioned cases, it turned out that there is no drawing of the graph with fewer than crossings, and the proof of Theorem 2 is complete. □