Almost disjoint spanning trees: relaxing the conditions for completely independent spanning trees
Benoit Darties (Le2i), Nicolas Gastineau (LAMSADE), Olivier Togni, (Le2i)

TL;DR
This paper introduces (i, j)-disjoint spanning trees, a generalized concept that relaxes conditions for independent spanning trees, and explores their existence and computational complexity across various graph classes.
Contribution
It defines (i, j)-disjoint spanning trees, proves NP-completeness of their existence, and identifies cases where such trees exist in specific graph classes.
Findings
NP-completeness for all i, j in general graphs
Existence of (i, j)-disjoint spanning trees in certain graph classes
Nuanced relationships between disjoint spanning trees and dominating sets
Abstract
The search of spanning trees with interesting disjunction properties has led to the introduction of edge-disjoint spanning trees, independent spanning trees and more recently completely independent spanning trees. We group together these notions by defining (i, j)-disjoint spanning trees, where i (j, respectively) is the number of vertices (edges, respectively) that are shared by more than one tree. We illustrate how (i, j)-disjoint spanning trees provide some nuances between the existence of disjoint connected dominating sets and completely independent spanning trees. We prove that determining if there exist two (i, j)-disjoint spanning trees in a graph G is NP-complete, for every two positive integers i and j. Moreover we prove that for square of graphs, k-connected interval graphs, complete graphs and several grids, there exist (i, j)-disjoint spanning trees for interesting values ofβ¦
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 Β· Graph theory and applications Β· Limits and Structures in Graph Theory
Almost disjoint spanning trees: relaxing the conditions for completely independent spanning trees
Benoit Darties
UniversitΓ© de Bourgogne, 21078 Dijon cedex, France, Le2i, UMR CNRS 6303
Nicolas Gastineau
PSL, UniversitΓ© Paris-Dauphine, LAMSADE UMR CNRS 7243, France
Olivier Togni
UniversitΓ© de Bourgogne, 21078 Dijon cedex, France, Le2i, UMR CNRS 6303
Abstract
The search of spanning trees with interesting disjunction properties has led to the introduction of edge-disjoint spanning trees, independent spanning trees and more recently completely independent spanning trees. We group together these notions by defining -disjoint spanning trees, where (, respectively) is the number of vertices (edges, respectively) that are shared by more than one tree. We illustrate how -disjoint spanning trees provide some nuances between the existence of disjoint connected dominating sets and completely independent spanning trees. We prove that determining if there exist two -disjoint spanning trees in a graph is NP-complete, for every two positive integers and . Moreover we prove that for square of graphs, -connected interval graphs, complete graphs and several grids, there exist -disjoint spanning trees for interesting values of and .
1 Introduction
The graphs considered are assumed to be connected, since spanning trees are only interesting for connected graphs. Let be an integer and be spanning trees in a graph . The spanning trees are edge-disjoint if . A vertex is said to be an inner vertex in a tree if it has degree at least 2 in and a leaf if it has degree 1. We denote by the set of inner vertices of tree . The spanning trees are internally vertex-disjoint if are pairwise disjoint. Finally, the spanning trees are completely independent spanning trees if they are both pairwise edge-disjoint and internally vertex-disjoint.
In this paper, we introduce -disjoint spanning trees:
Definition 1.1**.**
Let be an integer and be spanning trees in a graph . We let be the set of vertices which are inner vertices in at least two spanning trees among , and we let be the set of edges which belong to at least two spanning trees among . The spanning trees are -disjoint for two positive integers and , if the two following conditions are satisfied:
- i)
; 2. ii)
.
By we denote a large enough integer, i.e. an integer larger than , for a graph . Remark that -disjoint spanning trees are completely independent spanning trees and that -disjoint spanning trees are edge-disjoint spanning trees. Notice also that there are infinitely many -disjoint trees in , for and , being the minimum size of a connected dominating set in (one can repeat infinitely the same tree with inner vertices).
1.1 Related work
Completely independent spanning trees were introduced by Hasunuma [11] and then have been studied on different classes of graphs, such as underlying graphs of line graphs [11], maximal planar graphs [13], Cartesian product of two cycles [14], complete graphs, complete bipartite and tripartite graphs [24], variant of hypercubes [5, 26] and chodal rings [25]. Moreover, determining if there exist two completely independent spanning trees in a graph is a NP-hard problem [13]. Recently, sufficient conditions inspired by the sufficient conditions for hamiltonicity have been determined in order to guarantee the existence of two completely independent spanning trees: Diracβs condition [1] and Oreβs condition [6]. Moreover, Diracβs condition has been generalized to more than two trees [4, 15, 17] and has been independently improved [15, 17] for two trees. Also, a recent paper has studied the problem on the class of -trees, for which the authors have proven that there exist at least completely independent spanning trees [22].
For a given tree and a given pair of vertices of , let be the set of vertices in the unique path between and in . Remark that are internally vertex-disjoint in a graph if and only if for any pair of vertices of , . Other works on disjoint spanning trees include independent spanning trees, i.e. focus on finding spanning trees rooted at the same vertex . In independent spanning trees, for any vertex the paths between and in are pairwise internally vertex-disjoint, i.e. for each integers and , , . In contrast with the notion of completely independent spanning trees, in independent spanning trees only the paths to are considered. Thus, may share common vertices or edges, which is not admissible with completely independent spanning trees. Independent spanning trees have been studied for several classes of graphs which include product graphs [23], de Bruijn and Kautz digraphs [8, 12], and chordal rings [19]. Related works also include edge-disjoint spanning trees, i.e. spanning trees which are pairwise edge-disjoint only. Edge-disjoint spanning trees have been studied on many classes of graphs, including hypercubes [2], Cartesian product of cycles [3] and Cartesian product of two graphs [18].
Some subsets of vertices of a graph are disjoint connected dominating sets if are pairwise disjoint and each subset is a connected dominating set in . There are some works about disjoint connected dominating sets that can be transcribed in terms of internally vertex-disjoint spanning trees (the disjoint connected dominating sets can be used to provide the inner vertices of internally vertex-disjoint spanning trees). The maximum number of disjoint connected dominating sets in a graph is the connected domatic number. This parameter is denoted by and has been introduced by Hedetniemi and Laskar [16] in 1984. An interesting result about connected domatic number concerns planar graphs, for which Hartnell and Rall have proven that, except (which has connected domatic number ), their connected domatic number is bounded by 3 [10]. The problem of constructing a connected dominating set is often motivated by wireless ad-hoc networks [9, 28] for which connected dominating sets are used to create a virtual backbone in the network.
1.2 Motivation and basic facts about disjoint dominating sets
Remark that -disjoint spanning trees are internally vertex-disjoint, and consequently, are related to connected dominating sets. Hence, we call -disjoint spanning trees, trees induced by disjoint connected dominating sets and we give the properties about -disjoint spanning trees using, when possible, the concept of disjoint connected dominating sets. Figure 1 illustrates how disjoint connected dominating sets are used to construct -disjoint spanning trees. As we observe in the next proposition, trees induced by disjoint connected dominating sets satisfy interesting properties. First, an edge can only belong to at most two trees (Proposition 1.2.i)). Second, the paths between two non-adjacent vertices in trees induced by disjoint connected dominating sets are edge-disjoint (Proposition 1.2.ii)). Moreover, the fact that the paths between two adjacent vertices share a common edge implies that these vertices are inner vertices in different trees (Proposition 1.2.iii)). These properties illustrate the utility of disjoint connected dominating sets to broadcast a message following multiples routes in a network. For a spanning tree, an inner edge is an edge between two inner vertices and a leaf edge is an edge which is not an inner edge.
Proposition 1.2**.**
Let and be two integers, . Let be a graph of order at least , let be spanning trees induced by disjoint connected dominating sets and let .
- i)
every edge belongs to at most two trees among ; 2. ii)
if and are not adjacent, then ; 3. iii)
if then and .
Proof.
We prove that each of the three properties holds.
i) Suppose that is an inner edge in a spanning tree. Since the vertices and are leaves in any other tree, cannot belong to more than one spanning tree. Suppose is a leaf edge in at least two trees. The edge can belong to at most two trees, the trees for which and are inner vertices.
ii) Since the paths between and in the different trees have length at least and contain no common inner vertices, they share no common edges.
iii) By Property ii), and are adjacent. Moreover, if , then only contains inner edges of , and, as for Property i), each inner edge can not belong to another tree. Since only contains inner edges of , . The same goes if . β
Note that there is a relation between the minimum size of a connected dominating set in a graph , denoted by and (the maximum number of disjoint connected dominating sets) since . We also have to mention that Fan, Hong and Liu [6] have studied the line graph of cubic graphs of order at least and have proven that there are no two completely independent spanning trees in these cubic graphs. It could be possible, however, that it is not the case for two disjoint dominating sets.
If a graph satisfies and does not contain completely independent spanning trees, then there exist an integer such that contains -disjoint spanning trees. Hence, the notion of -disjoint spanning trees provides some nuances between the existence of disjoint connected dominating sets and completely independent spanning trees.
We say that connected dominating sets , , are -rooted connected dominating sets if the set satisfies . Remark that we can construct -disjoint spanning trees in a graph that contains -rooted connected dominating sets , , by considering that , for every integer , . Note also that trees induced by -rooted connected sets, i.e. -disjoint spanning trees, are also independent spanning trees rooted at a vertex . However, if are independent spanning trees rooted at in , then are not always -disjoint spanning trees in . This difference is illustrated by the fact that if for two vertices and two spanning trees and , , we have and , then it does not imply that .
1.3 Notation and Organization
We denote by the minimum degree of , i.e., . We denote by the usual distance between two vertices and in a graph . The graph is the graph obtained from by removing an edge from and , for , is the graph obtained from by removing the vertices from and their incident edges. For , we denote by , the graph . We say that a graph is -connected if and if for any set of vertices , with , is connected. By , and , we denote the complete graph, path and cycle, respectively, of order . Let and be positive integers. By we denote the square grid with rows and columns. The graph can be also defined as the Cartesian product of two paths and . The cylinder, denoted by , is the Cartesian product of one cycle and one path .
This article is organized as follows. Section 2 presents alternative characterizations of -disjoint spanning trees. Section 3 is about the computational complexity of the following decision problem: is it true that a graph contains two -disjoint spanning trees (with input the graph ). Section 4 deals with -connectivity and the conditions of Dirac and Ore for -disjoint spanning trees. Section 5 is about the required number of edges and distribution of inner vertices in -disjoint spanning trees. Section 6 presents some -disjoint spanning trees in square of graphs, -connected interval graphs, complete graphs, and square grids and cylinders.
2 Characterizations in terms of partitions and dominating sets
We begin this section by proving the following proposition.
Proposition 2.1**.**
Let be a connected graph of order at least and let be -disjoint spanning trees in . For every integer , , every vertex satisfies the two following properties:
- i)
if , then has a neighbor in ; 2. ii)
if has diameter at least , then has a neighbor in .
Proof.
Suppose there exist an integer , , and a vertex which has no neighbor in .
i) If , then which contradicts the hypothesis that has order at least .
ii) Since property i) holds, we suppose that . Remark that since has diameter at least , a spanning tree of has also diameter at least . Moreover, if is only adjacent to leaf vertices then it implies that is a star which contradicts the fact that has diameter at least . β
Let and be two subsets of vertices of a graph . By we denote the bipartite graph with vertex set and edge set . In the two following subsections we give alternative characterizations of -disjoint spanning trees and -rooted connected dominating sets. These characterizations are expressed in terms of partition in sets of vertices fulfilling some properties.
2.1 -disjoint spanning trees
In this subsection, we introduce a definition which is inspired by the definition of CIST-partition introduced by Araki [1].
Definition 2.2**.**
An -CIST-partition of a graph into sets is a partition of into sets of vertices such that:
- i)
* is connected, for each integer , ;* 2. ii)
* contains no isolated vertex, for every two integers , , ;* 3. iii)
, where is the number of connected component which are trees in , .
Figure 2 illustrates an -CIST partition on a specific graph. In a similar way than Araki [1], we prove that the notions of -CIST partition and -disjoint spanning trees are equivalent.
Theorem 1**.**
Let be a graph. There exist -disjoint spanning trees in if and only if has an -CIST-partition into sets.
Proof.
Suppose has an -CIST-partition into sets ,β¦,. We are going to construct -disjoint spanning trees . We begin by setting for each integer , . For each integer , , we suppose that is empty and we progressively add edges to in order to obtain spanning trees of at the end of the proof. Since is connected for each integer , , it is possible to add edges to in order to have a spanning tree with inner vertices from , for each integer .
Let and be two integers, , and let be a connected component of . We add edges in order to build a spanning tree restricted to and another spanning tree restricted to by considering two cases. Let be a vertex of . First, if is a tree, then we add an edge of incident with to both and . Thus, the edge will be common to and . Let . We add to the edges of the set and to the edges of the set . Second, if is not a tree, then we suppose that is in a cycle of . Let be an edge of this cycle incident with and let be a spanning tree of . We define as follows: . We add to the edges of the set and to the edges of the set . We repeat this process for every connected component of and every two integers and , . Since there is only one common edge between and for each connected component that is a tree and since , the set contains at most edges. Therefore, we obtain, by Property ii), -disjoint spanning trees.
Let us prove the converse of the previous implication. Suppose there exist -disjoint spanning trees in . The set , , induces a connected subgraph in . We begin by setting , for each integer , . If some vertices are inner vertices in no trees, we can add them to any set among . Thus, Property i) follows. Let and be two integers, . Suppose there exists one isolated vertex in . Without loss of generality, suppose . By Proposition 2.1.i), we obtain a contradiction since and has no neighbor in . Thus, Property ii) follows. Now suppose . Let be a connected component which is a tree in for some integers and and suppose that contains no edge from . Since has edges, it is impossible that every vertex of is adjacent to a vertex of in and that every vertex of is adjacent to a vertex of in , since it would require edges. Thus, for every two integers and and every connected component of , if is a tree then and we obtain a contradiction since implies . Consequently, Property iii) follows. β
2.2 -disjoint spanning trees
For a graph and a subset of vertices , let . In a similar way than Zelinka [29], we prove that the notion of -rooted connected dominating sets is equivalent to a notion of partition.
Definition 2.3**.**
An -rooted partition of into sets is a partition of into sets of vertices such that:
- i)
; 2. ii)
* is connected, for each integer , ;* 3. iii)
* contains no isolated vertex, for every two integers and , .*
Figure 2 illustrates an -rooted partition on a specific graph.
Theorem 2**.**
Let be a graph. There exist -rooted connected dominating sets in if and only if has an -rooted partition into sets.
Proof.
Suppose has an -rooted partition into sets . We begin by setting for each integer , . Since is a partition, we have . Moreover, by Property ii), the subgraphs induced by the sets ,β¦, are all connected. It remains to prove that is a dominating set, for each integer , . Since the vertices of are already dominated by a vertex of , for each integer , Property iii) implies that every vertex of has a neighbor in , for each integer .
Suppose there exist -rooted connected dominating sets in . We begin by setting . Afterward, we set , for each integer , . By definition, Property i) and Property ii) are satisfied by . It remains to prove Property iii). By contradiction, suppose that a vertex has no neighbor in , for some integers and . This fact implies that is not a dominating set and Property iii) follows. β
In the following definition we introduce the construction of a graph denoted by .
Definition 2.4**.**
Let be a graph, be an integer and be a subset of vertices. We denote by the graph obtained by replacing one by one each vertex , for , by a complete graph of order , and by adding edges between each vertex of this clique and every vertex of .
We finish by proving that determining if a graph contains -rooted connected dominating sets is equivalent to determine if the graph has disjoint connected dominating sets, for some subset of vertices . In contrast with the two previous propositions, this alternative characterization is expressed in terms of disjoint dominating sets.
Proposition 2.5**.**
There exist k -rooted connected dominating sets in a graph if and only if there exist a subset of vertices such that and .
Proof.
Let be a graph. Suppose there exist -rooted connected dominating sets in . We begin by setting . Let denote the clique from which replaces the vertex in , for . We can construct disjoint connected dominating sets in as follows: for each integer , , contains the vertices from and one different vertex by clique , for each integer , .
Suppose there exist a subset of vertices such that and . Let be disjoint connected dominating sets in . We can construct -rooted connected dominating sets in as follows: for each integer , , contains the vertices from .
β
3 An NP-complete problem for every integers and
We define the following decision problem:
--DSP
Instance : A graph .
Question: Does there exist -disjoint spanning trees in ?
Theorem 3**.**
Let and be non negative integers. The problem --DSP is an NP-complete problem for every pair of integer .
Proof.
Hasunuma [13] has proved that the following problem is NP-complete:
--CIST
Instance : A graph and two vertices and of .
Question: Does there exist two completely independent spanning trees and in with and ?
Initially, the NP-complete problem considered by Hasunuma [13] consists in determining if there exist two completely independent spanning trees in a graph . However, by analyzing Hasunumaβs reduction we can also obtain that the problem --CIST is NP-complete by using the same reduction (it suffices to consider that is and that is in Hasunumaβs reduction). Also, it is trivial to prove that the problem --DST is in NP since the description of two spanning trees in a graph (when ) ensures the existence of these two -disjoint spanning trees. We use a reduction from --CIST.
We introduce the three following operations that will be useful to describe our reduction:
- i)
-add is an operation on a graph with two prescribed vertices and that consists in adding the graph from Figure 3 and identifying with and with ; 2. ii)
-add is an operation on a graph with two prescribed vertices and that consists in adding the graph from Figure 3 and identifying with and with ; 3. ii)
-add is an operation on a graph with two prescribed vertices and that consists in adding the graph from Figure 3 and identifying with and with ;
Let be a graph and let and be two vertices of . We construct a graph from as follows. Let be a positive integer. We begin by constructing two graphs and by induction. The graph is the graph and the graph is the graph . The graph is obtained from by doing an -add on the two vertices of degree in (denoted by and in the left part of Figure 3, for ). The graph is obtained from by doing an -add on the two vertices of degree (also denoted by and in the middle part of Figure 3, for ). In and , we denote by and the two remaining vertices of degree .
Finally, the graph is obtained by taking and , identifying in with a vertex of degree one in and in with the other vertex of degree one in and doing a -add on the two vertices of degree which are labeled by and in . The graph is obtained by taking a copy of , adding , identifying the vertex with a vertex of degree one in and identifying the vertex with the other vertex of degree one in .
Suppose there exist two completely independent spanning trees and in with and . We can construct two -disjoint spanning trees in by reproducing the trees and in the graph restricted to and by using the patterns described in Figure 4 in order to extend the spanning trees to .
Suppose there exist two -disjoint spanning trees and in . Note that there are articulation vertices in the graph restricted to . These articulation vertices should be inner vertices in both and . Thus, the trees and restricted to are internally vertex-disjoint. By Proposition 2.1, the vertex (obtained by -add in and illustrated in Figure 3) should be adjacent to an inner vertex of and to an inner vertex of . Thus in order that and be connected, there must be a path from to in and a path from to in (we can exchange and if necessary). Note that the vertices and from a copy of (illustrated in Figure 3) cannot be both inner vertices of the same tree since it would be impossible to have a path from to in and another path from to in . Thus, in order that and belong to both and , the edge should belong to both and (since we already have articulation vertices). Moreover, since there are copies of in the graph , the trees and restricted to are both internally vertex-disjoint and edge-disjoint.
β
4 Sufficient conditions to have -disjoint spanning trees
4.1 -connectivity
We begin this section by proving classical properties about cut sets.
Proposition 4.1**.**
Let be a graph and let be -disjoint spanning trees in . For every subset of vertices such that and is not connected, (at least) one vertex of is in . For every subset of edges such that such that and is not connected, (at least) one edge of is in .
Proof.
Let be a subset of vertices such that and is not connected. Remark that should not be empty, for every integer , , since it would imply that is not connected. Since , a vertex of should be in . The same property holds for . β
Proposition 4.2**.**
Let be a graph and let be -disjoint spanning trees in . Let be the number of articulation vertices which do not belong to bridges in and let be the number of bridges in . We have and .
Proof.
Since an articulation vertex belongs to every spanning tree of , we have . The same goes for the bridges and their extremities. β
Since the presence of a -cut in a graph implies that there do not exist disjoint connected dominating set, it is natural to ask whether a -connected graph, for sufficiently large, contains at least two disjoint connected dominating sets [16]. In the paper in which completely independent spanning trees have been introduced [11], the same question has been asked for two completely independent spanning trees.
Using the construction from Kriesell [20] or PΓ©terfalvi [27], we can obtain a family of -connected graphs that do not contain two completely independent spanning trees. We recall the construction of the family of graphs considered by Kriesell [20].
Definition 4.3** ([20]).**
Let and be two integers such that . Let be the bipartite graph with vertex set and edge set . The graph corresponds to the incidence graph of the complete -uniform hypergraph with vertices.
Note that the graph is -connected and bipartite. Using a similar proof than that of Kriesell, we obtain the following theorem which shows that there exist -connected graphs which do not contain two -disjoint spanning trees, for every three positive integers , and .
Theorem 4**.**
Let , and be integers. For any , the graph does not contain two -disjoint spanning trees.
Proof.
Suppose there exist two -disjoint spanning trees and in . Let and be the two subsets of vertices forming a bipartition of , with and . Let . Note that, by definition of -disjoint spanning trees, . We consider a set , . By Proposition 2.1, at least one inner vertex of is adjacent to . This inner vertex of is denoted by . Inductively, since , for , we can create a set with and obtain that there exists a vertex adjacent to . The set is such that and . Hence, we have a contradiction with Proposition 2.1.ii), since has no neighbor which is a inner vertex of and has diameter greater than when . β
4.2 Diracβs and Oreβs conditions
We begin this subsection by proving that there are at least two disjoint dominating sets in some particular graphs.
Proposition 4.4**.**
There exist two disjoint connected dominating sets in and three disjoint connected dominating sets in
Proof.
These disjoint connected dominating sets are illustrated in Figure 5. β
A graph satisfies the condition of Dirac if and satisfies the condition of Ore if . Araki [1] proved that every graph with satisfying Diracβs condition contains two completely independent spanning trees. Moreover, Fan, Hong and Liu [6] proved that every graph with satisfying Oreβs condition contains two completely independent spanning trees. The only graphs with satisfying the Dirac condition or the Ore condition which do not contain two completely independent spanning trees are , and . Thus, by Proposition 4.4, we obtain the two following theorems:
Theorem 5** ([1]).**
Let be a graph. If , then there exist two disjoint connected dominating sets.
Theorem 6** ([6]).**
Let be a graph. If , then there exist two disjoint connected dominating sets.
Moreover, there exists a graph of order satisfying and , that does not contain two disjoint connected dominating sets. Such graph can be constructed by taking two complete graphs and , for a positive integer, and by identifying a vertex of the first clique with a vertex of the second clique. This fact implies that the bounds in the previous theorems are tight. It could be possible to improve the recent results about Diracβs condition [4, 15, 17] by only considering disjoint connected dominating sets.
5 Number of inner vertices and edges in -disjoint spanning trees
5.1 Required number of edges
We begin this section by giving necessary conditions on the number of edges of a graph in order to have -disjoint spanning trees.
Proposition 5.1**.**
Let be a graph of order and let be -disjoint spanning trees in . We have .
Proof.
Suppose contains at least -disjoint spanning trees. Since every spanning tree contains edges and since an edge in can be in at most trees, we obtain that contains at least edges. β
Note that the grid satisfies the equality for and . This last proposition can be improved for [10] since, by Proposition 1.2.i), an edge in can be in at most two trees.
Corollary 5.2**.**
Let be a graph of order and let be -disjoint spanning trees in . We have .
Moreover, for an arbitrary large , the following bound is known.
Proposition 5.3**.**
[10]** A graph of order such that has at least edges. This bound is sharp since .
5.2 Distribution of the inner vertices
The following observation illustrates the existence of an -disjoint spanning tree with possibly less inner vertices than the others.
Observation 5.4**.**
Let be a graph of order and let be -disjoint spanning trees in . There exists a tree among satisfying .
Two sets of vertices and are balanced if . We begin by proving that there exists a graph satisfying but in which no two disjoint connected dominating sets are balanced. Let be the graph constructed by taking one copy of , by adding a new vertex and by adding the edges between and the vertices of . Figure 6 illustrates the graph for .
Proposition 5.5**.**
Let . For any two disjoint connected dominating sets and in , .
Proof.
Suppose without loss of generality that . Since should be connected, it should contain consecutive vertices of . Moreover, since should be dominating, it should contain every vertex of , except its extremities. Thus, and consequently . Therefore, we have . β
Note that the graph does not contain two completely independent spanning trees. Thus, it could be true that every graph containing two completely independent spanning trees contains two completely independent spanning trees and such that . However, the following proposition illustrates that it is not the case. Let be the graph obtained by taking one copy of , by adding a new vertex and by adding the edge and the edges between and the extremities of , being the vertex of maximal degree in , being the induced path of vertices in obtained by removing . Figure 6 illustrates the graph for .
Proposition 5.6**.**
Let . For any two completely independent spanning trees and in , .
Proof.
First, observe that there exist two completely independent spanning trees in since and is a [math]-CIST-partition. Now, suppose there exist two completely independent spanning trees and and suppose without loss of generality that . Since the graph induced by the vertices of should be connected, it should contain consecutive vertices of . Moreover, should be dominating set. Since either and the subsets of or and the proper subsets of do not form a [math]-CIST partition, we have and . Therefore, we have . β
Even if there exist graphs only containing two non-balanced disjoint connected dominating sets, it could be interesting to find classes of graphs for which there always exist two disjoint connected dominating sets which are balanced. For example, the class of graphs with minimum degree at least , is such a class [17].
6 -disjoint spanning trees in some simple classes of graphs
6.1 Square of graphs
The square of a graph , denoted by , is the graph obtained from by adding edges between every two vertices and of with . Araki [1] has studied the square of graphs and has proven that there exists a tree such that there are no two completely independent spanning trees in and that in the square of every -connected graph, there are two completely independent spanning trees. Moreover, the family of trees such there are no two completely independent spanning trees in has been determined. We begin this section by proving that there exist two -disjoint spanning trees in the square of every graph.
Proposition 6.1**.**
Let be graph. There exist two -disjoint spanning trees in .
Proof.
Let be a spanning tree of let and be a bipartition of . The sets and form an -CIST-partition of since both and are connected in and since is a connected graph (which can be a tree in the case is a tree). Thus, by Theorem 1, there exist two -disjoint spanning trees in . β
We finish the section by determining which square of graph contains two completely independent spanning trees (the case of trees has already been treated [1]).
Proposition 6.2**.**
Let be a connected graph which is not a tree. There exist two completely independent spanning trees in .
Proof.
Since is not a tree, there exists an induced cycle in . Let be a vertex of which has a neighbor not belonging to . If such vertex does not exist, then is cycle and contains two completely independent spanning trees [1]. Let be an edge of , let be a spanning tree of and let and be a bipartition of . Remark that both and are connected and that every edge of belongs to . Our goal is to prove that there is one more edge in , i.e, that is connected and is not a tree. First, if is of even length, then and (or and , by symmetry) and . Second if is of odd length, then , and (or , and , by symmetry) and . Thus, by Theorem 1, there exist two completely independent spanning trees in . β
Note that the square of a star (a tree of diameter at most 2) is a clique and can contain an arbitrary large number of completely independent spanning trees (this number depends on the degree of the central vertex). Thus, it could be interesting to determine which square of graph contains completely independent spanning trees for .
6.2 -connected interval graph
We begin by recalling the definition of a path-decomposition of a graph .
Definition 6.3**.**
Let be a graph. A sequence of subsets of vertices of is a path-decomposition of if the two following properties are satisfied:
- i)
for each edge of , there exists an integer such that both extremities of belong to the subset ; 2. ii)
for every three integers , .
An interval graph is the intersection graph of a family of intervals of the real line. We recall that an interval graph has a path-decomposition for which each , , forms a maximal clique in . We also recall that for a -connected interval graph with path-decomposition , we have , for every integer , (otherwise would be a cut set of order less than ). The following property is true for -connected interval graphs.
Theorem 7**.**
Let be an integer. Every -connected interval graph satisfies .
Proof.
Let be a -connected interval graph. Let be a path-decomposition of , for which every forms a maximal clique. If , then is a -connected complete graph, i.e., for . Thus satisfies . Hence, suppose . Our goal is to construct disjoint connected dominating sets by setting , for .
By hypothesis, , and there exist different vertices forming disjoint connected dominating sets on the graph . We set , for every integer , . Suppose and that, by induction, that we have already determined , for every integer , and that are disjoint connected dominating sets on the graph , for . Now our goal is to construct disjoint connected dominating sets on the graph .
Let be an integer, . If , then we set , otherwise we set to a vertex not in . Such a vertex exists since otherwise it would imply that . Finally, the sets are disjoint connected dominating sets on the graph . Consequently, by induction, we can construct disjoint connected dominating sets on the graph . β
The previous theorem can not be generalized to chordal graphs since there exist -connected chordal graphs, for , which do not contain two disjoint connected dominating sets [27].
6.3 Complete graphs
By we denote the maximum number of -disjoint spanning trees in . Remark that there are disjoint connected dominating sets in and that there are completely independent spanning trees in [22].
We give the following intermediate result about -disjoint spanning trees.
Proposition 6.4**.**
Let be an integer. We have , where if is odd and [math] otherwise.
Proof.
First, suppose is even. Let . We begin by proving that for . Suppose that there are -disjoint spanning trees. By Corollary 5.2, we have . Observe that . Since , we have , contradicting the definition of . We are going to prove that we can construct -disjoint spanning trees in , for . We construct two kinds of spanning trees. First, we construct spanning trees which are spanning stars. Second, we construct spanning trees in each tree with two inner vertices, as in [22] (with disjoint inner vertices). The left part of Figure 7 illustrates this construction for . There are common edges between the spanning stars and common edges between the inner vertices of and the remaining vertices. Thus, there are common edges and by definition .
Second, suppose is odd. Let . We begin by proving that for . Suppose that there are -disjoint spanning trees. By Corollary 5.2, we have . Observe that . Since , we have , contradicting the definition of . We begin by constructing -disjoint spanning trees in , for . We construct two kinds of spanning trees. First, we construct spanning trees which are spanning stars. Second, we construct spanning trees in , each tree having two inner vertices, following the construction described in [22] (with disjoint inner vertices). There are common edges between the spanning stars and common edges between the inner vertices of and the remaining vertices. Thus, there are common edges and by definition . β
The middle part of Figure 7 depicts three -disjoint spanning trees in .
Proposition 6.5**.**
Let be a positive integer. For , we have , where if is even, and [math] otherwise. Moreover, if , then is not finite.
Proof.
Observe that a connected dominating set of can contain only one vertex. Thus, if , then is not finite.
First suppose is even. Let . We prove that we can construct -disjoint spanning trees in , for . Let be an induced of . We begin by creating completely independent spanning trees in , as in [22]. Let be a vertex of . We are going to extend these trees in order they span the whole graph . To each tree , , we add the vertex and an edge incident to and to a vertex of . We finally add to the edges incident to and to every vertex of . We now construct the remaining trees as follows. Each tree , , has two inner vertices: and a vertex of different for each tree. Each tree also contains the edges incident to and to every vertex of and the edges incident to and to every vertex of . It is easy to verify that the trees have only one common vertex (the vertex ) and common edges (the edges incident to in ).
Second, suppose is odd. Let . We prove that we can construct -disjoint spanning trees in , for . Let be an induced of . We begin by creating completely spanning trees in , as in [22] and extend them to the whole graph as for the case even. We construct the trees similarly as in the case even. It is easy to verify that the trees have only one common vertex (the vertex ) and common edges (the edges incident to in ). β
The right part of Figure 7 depicts four -disjoint spanning trees in . Note that we can obtain a lower bound on the number of -disjoint trees in by using Proposition 5.1. However, in this case, we do not obtain a tight bound. Moreover, Proposition 5.1 implies that for every positive integer .
6.4 Cylinders
Let and be positive integers with and . Let and .
Theorem 8**.**
There exist two -disjoint spanning trees in the cylinder .
Proof.
We describe these two trees by giving their edge sets:
.
. β
Observe that contains edges. Hence, by Corollary 5.2, we can conclude that there does not exist two -disjoint spanning trees in , for .
6.5 square grids
Let and be positive integers with and . Let and .
In two papers [7, 21], the trees with a maximum number of leaves in have been determined. In particular, Fujie [7] has shown that a spanning tree of has at least inner vertices. Hartnell and Rall [10] have proven that there do not exist two disjoint connected dominating sets in , except if or . However, this is not the case for -rooted connected dominating set. We finish this paper by giving a construction of two -rooted connected dominating sets in for and . In Figure 9, we exhibit two trees induced by two -rooted connected dominating sets in . In this example, we have minimized the number of common edges.
Theorem 9**.**
There exist two -rooted connected dominating sets in the grid , for every and .
Proof.
Suppose without loss of generality that . If , then one can easily construct two -rooted connected dominating sets by setting and by setting .
Now suppose that . We construct as follows:
.
The set is . Note that .
Figure 9 illustrates this construction, with circle vertices corresponding to and triangles to (the square vertex being both in and ). β
Open questions
In this introducing paper about -disjoint spanning trees, we tried to cover a large number of issues. However, there still remains a lot of interesting properties to be found about this notion. We finish this paper by giving some open questions:
Does there exist -disjoint spanning trees in every -connected graph of order ? 2. 2.
Determine conditions in order to guarantee the existence of completely independent spanning trees, , in the square of graphs. 3. 3.
Determine conditions on chordal graphs in order to guarantee the existence of two disjoint connected dominating set. 4. 4.
Determine for the remaining cases. 5. 5.
Determine for the complete -partite graphs. 6. 6.
Determine the minimum number of common edges in order to have two -disjoint spanning trees in the square grid.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] T. Araki, Diracβs condition for completely independent spanning trees, Journal of Graph Theory 77 (2014), 171β179.
- 2[2] B. Barden, J. Davis, R. Libeskind-Hadas, W. Williams, On edge-disjoint spanning trees in hypercubes, Information Processing Letters 70 (1999), 13β16.
- 3[3] D. M. Blough, H. Wang, Multicast in wormhole-switched torus networks using edge-disjoint spanning trees, Journal of Parallel and Distributed Computing 61 (2001), 1278β1306.
- 4[4] H-Y. Chang, H-L. Wang, J-S. Yang, J-M. Chang A note on the degree condition of completely independent spanning trees, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences E 98.A (2015), 2191β2196.
- 5[5] B. Cheng, D. Wang, J. Fan Constructing completely independant spanning trees in crossed cubes, Discrete Applied Mathematics 219 (2017), 100β109.
- 6[6] G. Fan, Y. Hong, Q. Liu, Oreβs condition for completely independent spanning trees, Discrete Applied Mathematics 177 (2014), 95β100.
- 7[7] T. Fujie, An exact algorithm for the maximum leaf spanning tree problem, Computers and Operations Research 30 (2003), 1931β1944.
- 8[8] Z. Ge, S. L. Hakimi, Disjoint rooted spanning trees with small depths in de Bruijn and Kautz graphs, SIAM J. Comput 26 (1997), 79β92.
