Optimal proper connection of graphs
Shinya Fujita

TL;DR
This paper explores efficient methods to transform connected graphs into properly connected ones by recoloring edges, focusing on minimizing total recoloring and applying to specific graph classes like trees and bipartite graphs.
Contribution
It introduces algorithms for converting monochromatic graphs into properly connected graphs with minimal recoloring, especially for trees, bipartite graphs, and graphs with independence number 2.
Findings
Efficient recoloring algorithms for trees and bipartite graphs.
Optimal recoloring strategies for graphs with independence number 2.
Minimization of total recoloring and color changes in graph transformation.
Abstract
An edge-colored graph is called properly colored if no two adjacent edges share a color in . An edge-colored connected graph is called properly connected if between every pair of distinct vertices, there exists a path that is properly colored. In this paper, we discuss how to make a connected graph properly connected efficiently. More precisely, we consider the problem to convert a given monochromatic graph into properly connected by recoloring edges with colors so that is as small as possible. We discuss how this can be done efficiently for some restricted graphs, such as trees, complete bipartite graphs and graphs with independence number .
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Graph Theory Research · Graph Labeling and Dimension Problems · Limits and Structures in Graph Theory
Optimal proper connection of graphs
Shinya Fujita
School of Data Science, Yokohama City University,
22-2 Seto, Kanazawa-ku, Yokohama 236-0027, Japan,
Email: [email protected]
Abstract
An edge-colored graph is properly colored if no two adjacent edges share a color in . An edge-colored connected graph is properly connected if between every pair of distinct vertices, there exists a path that is properly colored. In this paper, we discuss how to make a connected graph properly connected efficiently. More precisely, we consider the problem to convert a given monochromatic graph into properly connected by recoloring edges with colors so that is as small as possible. We discuss how this can be done efficiently for some restricted graphs, such as trees, complete bipartite graphs and graphs with independence number .
1 Introduction
All graphs considered in this paper are finite and simple. Our notation in this paper is standard. For a graph , let be the independence number of . Also, let be the size of a maximum matching of . Let be the vertex-connectivity of . Let be the diameter of . For a vertex , let . For a vertex subset of , stands for an induced subgraph of induced by . For other terminology and notation not defined here, we refer the reader to [12]
An edge-colored graph is properly colored if no two adjacent edges share a color in . Properly colored paths and cycles appear in a variety of fields such as genetics [4, 5] and social sciences [3]. An edge-colored connected graph is properly connected if between every pair of distinct vertices, there exists a path that is properly colored. In [1], Borozan et al. defined a new notion called the proper connection number of a connected graph , where is the minimum number of colors needed to color the edges of to make it properly connected. As described in [8, 10], this concept has a real application to build an efficient communication network with no radio-frequency interference between each pair of wireless signal towers. There, roughly speaking, to avoid interference, it is important not to share the same frequency when a wireless transmission passes through a signal tower. In fact, the proposed network model can be regarded as an edge-colored graph that is properly connected. For a more precise description, see [8, 10].
Recently, the notion of proper connection number attracts much attention from both theoretical and practical aspects, and thus a lot of work has been done extensively (see e.g., [2, 6, 7, 9, 11]). For details in this recent topic, we refer the reader to the nice survey of Li and Magnant [10].
In this paper, we are concerned with making an edge-colored graph properly connected efficiently. Let be a connected graph with a given edge-coloring . Now we consider how to make properly connected by recoloring some edges with some colors. To minimize our effort to make properly connected, it would be natural to focus on the minimum value on the sum of numbers of edges and colors among such recolorings. Note that, such a value should be zero when is already properly connected.
Perhaps the most fundamental and laborious case to this problem would be the case where assigns a common color on every edge of , that is, the case where is a monochromatic colored graph. Therefore, in this paper, we shall initiate this study by assuming that all edges of have already been colored by a common color, say color [math]. For an integer , color is called a new color.
Keeping this assumption in mind, we define the following cost function of edge-colored graphs called the optimal proper connection number for a monochromatic connected graph .
we can make properly connected
by recoloring edges of with new colors
For a monochromatic connected graph , suppose that becomes properly connected by recoloring edges of with new colors such that . Then we call such an edge-coloring of an optimal recoloring of .
By definition, note that holds because any monochromatic complete graph is properly connected. Indeed, we see that a graph satisfies if and only if is isomorphic to a complete graph. We can easily determine for small graphs and some basic family of graphs. For example, we can check that , , .
In this paper, we shall investigate graphs with a small optimal proper connection number. Along this line, we give an upper bound on when is a graph with (see Theorem 6). We also give a formula in terms of the optimal proper connection number for trees and complete bipartite graphs (see Theorems 7 and 9).
This paper organizes as follows. In Section 2, we give some basic observation on optimal proper connection number in edge-colored graphs. In Section 3, we prove our main results (Theorems 6, 7 and 9). In Section 4, we discuss some extension and open problems in this topic.
2 Preliminaries
In order to give a good upper bound on for a monochromatic connected graph , we start with the following basic observation.
Proposition 1**.**
If a monochromatic graph contains as a spanning connected subgraph, then .
Proof.
The assertion obviously holds because an optimal recoloring of in assures us that is properly connected. ∎
For a monochromatic graph , an edge is good if can be partitioned into two parts and such that and for (possibly, for some ).
We can characterize monochromatic graphs having as follows.
Proposition 2**.**
A monochromatic connected graph of order at least has if and only if contains a good edge. Moreover, recoloring any good edge of with a new color can be an optimal recoloring of .
Proof.
The assertion obviously holds if is a complete graph. So we may assume that is not a complete graph. Suppose that contains a good edge . Then, recoloring with color , we can easily check that is properly connected by the definition of a good edge. Thus and the second assertion holds.
Next suppose that and consider an optimal recoloring of . We may assume that is now properly connected and has exactly one edge with color and all other edges have color [math]. We claim that is a good edge. If there exists a vertex , then there is no properly colored path joining and in because and is a unique edge having color . This contradicts the assumption that is properly connected. Thus we may assume that can be partitioned into two parts and such that for (possibly, for some ). Suppose that contains two vertices such that . Since , obviously there is no properly connected path joining and in . Again, this contradicts the assumption that is properly connected. Thus, by the symmetry of the roles of and , we see that for (possibly, for some ). Hence is a good edge, as claimed. This completes the proof of Proposition 2.
∎
We next consider monochromatic connected graphs with . Unlike the case where , it seems complicated to characterize those graphs. As an initial step, in this paper, we investigate what kind of graphs satisfy .
Proposition 3**.**
For a monochromatic graph , if contains a complete bipartite graph such that is a spanning subgraph of and each partite set of contains an edge in , then .
Proof.
Let and be the partite sets of . By assumption, let be an edge of for . Note that is a good edge of for . By Proposition 2, and recoloring with color , is properly connected for . Since for any pair , this implies that recoloring both and with color makes properly connected. This shows that .
∎
By Proposition 3, we see that a monochromatic complete multipartite graph has a small optimal proper connection number because it contains the graph described in Proposition 3 as a spanning subgraph.
Corollary 4**.**
Let be a monochromatic complete multipartite graph such that , where and . Then .
We finally give some observation on graphs with a forbidden subgraph condition. A graph is -free if it contains no as an induced subgraph. Although the following proposition has nothing to do with edge-coloring of graphs, it is useful when we prove our main results.
Proposition 5**.**
If is a connected -free graph, then is a complete graph or contains a spanning complete bipartite subgraph with such that one partite set of the forms a minimum cutset of .
Proof.
We may assume that is not a complete graph. Let be a minimum cutset of , and let be the components of , where . It suffices to show that every vertex of sends edges to all the vertices of . Suppose not, and take and such that . We may assume that . Since is a minimum cutset of , there exists and such that . Since is a component, we may assume that and are chosen so that (to see this, take a shortest path in ; if then reset as ; if then reset as ). Now, the path is an induced in . This is a contradiction. Thus the assertion holds.
∎
3 Main results
We firstly give a sharp upper bound on for graphs with small independence number.
Theorem 6**.**
If is a monochromatic connected graph of order such that then .
Proof.
The theorem obviously holds for small . So we may assume that . If then since is a complete graph. Thus we may assume that and hence is not a complete graph.
We firstly consider the case where is a -free graph. In view of Proposition 5, let be a minimum cutset of such that contains a spanning complete bipartite graph whose partite sets are and . Since , consists of two components such that each forms a complete graph. Suppose for the moment that , say . Take . Note that, by the structure of , is a good edge of . Consequently, by Proposition 2, .
Thus we may assume that . In view of Proposition 3, if both and contain an edge, respectively, then . Thus we may assume that either or has no edge. Since and is a minimum cutset of , it suffices to consider the case where has no edge and . Let be two independent edges joining a vertex of and a vertex of , respectively. Recoloring both and with color , we can check that is properly connected because and are complete graphs. Thus .
Hence we may assume that contains a path such that . Since , we can partition into three parts such that , and . Then, by recoloring and with color , respectively, we can easily check that is properly connected, thereby proving that .
∎
The upper bound on in Theorem 6 is sharp. To see this, let be disjoint cliques, and add all edges between and , where the indices are taken modulo . Let be the resulting graph. In order to make properly connected, obviously we need to recolor at least two edges with a new color.
So far, we observed graphs with small upper bound on . Now we consider a question to ask what kind of a family of graphs we can describe an equality for each , where is a certain value depending on some parameters of . In fact, complete bipartite graphs and trees belong to such family of graphs.
Theorem 7**.**
Let be a monochromatic complete bipartite graph such that and . Then for , and for .
Proof.
Let be the partite sets of with . We first consider the case . Take .
The upper bound can be obtained by recoloring with color and with color , where . (Since , it is easy to check that the resulting edge-colored is properly connected.) Toward a contradiction, suppose that . This implies that we have exactly one new color (say, color ) to recolor at most two edges in to make properly connected. We can easily check that cannot be properly connected if we recolor exactly one edge. Thus we may assume that exactly two edges, say are recolored with color so that is properly connected. If and share a vertex, then we can check that there is no properly colored path joining other two vertices of and . Thus and must be a matching in . Since and , there exist two vertices with such that neither nor is on for . Since is a complete bipartite graph, we see that there is no properly colored path joining and , a contradiction. Thus we have .
We next consider the case . Suppose that . To give an optimal recoloring of , let us firstly consider the case we will use exactly one new color, say color . In that case, we can recolor at most three edges. Since , wherever we rcolor at most three edges in , we can find two vertices with such that neither nor is on an edge with color , meaning that there is no properly colored path joining and . This is a contradiction. Thus we need at least two distinct new colors to make properly connected.
Now suppose that we gave an optimal recoloring on . By the above observation, we may assume that contains two edges with color , , respectively, and other edges have color [math]. Assume for the moment that and are independent edges. We then consider another different edge-coloring of by modifying the color of from color to color . We see that this modified edge-colored is still properly connected. However, this contradicts the assumption that we gave an optimal recoloring on before the modification. Therefore, we may assume that and share a vertex . Since , there exist two vertices with and such that neither nor is on an edge with new color. We can check that there is no properly colored path joining and . Since is now properly connected, this is a contradiction.
Thus we have . Take and . Recolor and with color , respectively, and recolor with color . It is easy to check that the resulting edge-colored is properly connected, meaning that . Consequently, .
∎
Theorem 7 together with Propositions 1 and 5 yields the following corollary.
Corollary 8**.**
If is a monochromatic -connected -free graph of order at least then .
We can determine the optimal proper connection number for trees as follows.
Theorem 9**.**
If is a monochromatic tree of order then .
Proof.
The theorem obviously holds for . Thus we may assume that and hence . Note that, if is a tree, then the statement that is properly connected is equivalent to the statement that is properly colored.
We first show that . Suppose that we gave an edge-coloring on so that is properly connected. Let be a maximal monochromatic subgraph of with color [math]. Since is now properly connected, note that forms a matching and hence . This implies that we recolored at least edges of . Since is properly connected, note that contains no monochromatic , meaning that we need at least new colors to make properly connected. Thus we have .
We next show that . It suffices to show that there is some appropriate edge-coloring on some edges in with new colors to make properly connected. To do this, choose a maximum matching in so that is as minimum as possible. Now we will give a new color on each edge of . Note that . Put (possibly, an empty set). Since is a maximum matching, note that contains no edge.
We now claim that the set is indeed empty. Suppose not, and take a vertex . Consider a maximal path in such that and for every . By definition, note that . Since and contains no edge, note that . By the maximality of , if then . Assume for the moment that . This implies that is an even number. Consequently, forms a matching of size greater than , a contradiction. Hence we may assume that , meaning that is an odd integer and . Then by replacing by , we get a contradiction to the choice of because and . Thus the claim holds. Let be the forest obtained from by deleting all edges of . The above claim implies that . Hence we can give a proper edge-coloring on by using at most new colors. Combining the properly colored with , we can make properly connected, thereby proving that , and hence , as desired.
∎
Since any connected graph contains a spanning tree, we obtain the following corollary.
Corollary 10**.**
If is a monochromatic connected graph of order such that , then is a spanning tree of .
The lower bound on Corollary 10 follows from the fact that we need to recolor at least edges on the path joining a pair of vertices with distance to make properly connected. The upper bound on Corollary 10 can be attained when is a star.
4 Some remarks, extension and open problems
There are many problems together with some extension on the optimal proper connection number of graphs.
Aside from the case that a connected graph belongs to some basic family of graphs such as trees or complete bipartite graphs, it might be difficult to find an explicit formula on for some other family of graphs. Perhaps this could be a challenging problem.
Let be a monochromatic connected graph of order . If contains many edges, then it tends to contain a Hamiltonian path, meaning that holds by Corollary 10 and Proposition 1. It would be an interesting problem to consider what kind of graphs have a constant upper bound on . Also, considering upper bounds on for sparse graphs would be interesting. For example, what about connected cubic graphs?
For applications, constructing faster algorithms for giving optimal recolorings in graphs would be important. Note that the proofs of our results are constructive. So we can extract a polynomial time algorithm to make properly connected from there.
We can also think about some extension in this notion. In this paper, we consider the sum of the number of edges and colors when recoloring. However, one may simply consider the number of edges for the recoloring. Thus we can define the following function of edge-colored graphs for a monochromatic connected graph .
we can make properly connected
by recoloring edges of
Modifying the proofs of our results slightly, we can easily obtain the following counterparts. (Indeed, we have only to skip the argument for counting the number of new colors in the proofs of our previous theorems. So the proofs are omitted.)
Theorem 11**.**
If is a monochromatic tree of order then .
Theorem 12**.**
Let be a monochromatic complete bipartite graph such that and . Then for , and for .
Theorem 13**.**
If is a monochromatic connected graph such that then .
When we consider , we never care about the number of new colors for the recoloring to make properly connected. Conversely, note that, if we consider the number of colors but never care about the number of edges for the recoloring of , then the proper connection number can be the counterpart of .
We can think about this topic in a more strict manner: For a monochromatic connected graph , one may ask which ordered pair with gives us the optimal recoloring of . Note that, not all such pairs provide us the optimal recoloring of . For this requirement, trivially, we must have , but it is not sufficient in many cases. To describe this new direction more precisely, we define the following. For a monochromatic connected graph , is -feasible if we can make properly connected by recoloring edges with new colors; in particular, when , we say that is -optimal feasible.
In fact we already had some observation on the -optimal feasibility for small . To see this, note that, Proposition 2 implies that for any non-complete monochromatic connected graph , is -optimal feasible if and only if contains a good edge. Moreover, we can extract the following theorem from the proof of Theorem 7.
Theorem 14**.**
Let be a monochromatic complete bipartite graph such that and . If , then is -optimal feasible, and if , then is -optimal feasible.
As we can see from the above argument, our work on the optimal proper connection number could contribute to some problems on -optimal feasibility in monochromatic connected graphs. The author believe that there will be many interesting problems around this area of study.
On the other hand, as discussed in [1], we can consider the “properly -connected version” in this topic. An edge-colored -connected graph is properly -connected if between every pair of distinct vertices, there exist internally-disjoint paths that are properly connected. For a monochromatic -connected graph , we can similarly define the following function.
we can make properly -connected
by recoloring some edges of with new colors
Details on this function together with the above observation will be discussed elsewhere.
**Acknowledgments
**
This work was supported by JSPS KAKENHI (No. 15K04979)
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] Borozan, V., Fujita, S., Gerek, A., Magnant, C., Manoussakis, Y., Montero, L., and Tuza, Z.. Proper connection of graphs, Discrete Math., 312 (2012), 2550–2560.
- 2[2] Brause, C., Doan, T. D., Schiermeyer, I., Minimum degree conditions for the proper connection number of graphs, Gaphs and Combin., 33 (2017), 833–843.
- 3[3] Chou, W. S., Manoussakis, Y., Megalakaki, O., Spyratos, M., and Tuza, Z., Paths through fixed vertices in edge-colored graphs, Math. Inform. Sci. Humaines, 127 (1994), 49–58.
- 4[4] Dorninger, D., Hamiltonian circuits determining the order of chromosomes, Discrete Appl. Math., 50 (1994) 159–168.
- 5[5] Dorninger, D., and Timischl, W., Geometrical constraints on Bennet’s predictions of chromosome order, Heredity, 58 (1987), 321–325.
- 6[6] Fujita, S., Gerek, A., and Magnant, C., Proper connection with many colors, J. Comb., 3 (2012), 683–693.
- 7[7] Gu, R., Li, X., and Qin, Z., Proper connection number of random graphs Theoretical Computer Science, 609 (2016), 336–343.
- 8[8] Huang, F., Li, X., Qin, Z., Magnant, C., and Ozeki, K., On two conjectures about the proper connection number of graphs, Discrete Math., 340 (2017), 2217–2222.
