719
Views
2
CrossRef citations to date
0
Altmetric
Articles

The upper domatic number of a graph

, , , &

Abstract

Let G=(V,E) be a graph. For two disjoint sets of vertices R and S, set R dominates set S if every vertex in S is adjacent to at least one vertex in R. In this paper we introduce the upper domatic number D(G), which equals the maximum order k of a vertex partition π={V1,V2,,Vk} such that for every i,j, 1i<jk, either Vi dominates Vj or Vj dominates Vi, or both. We study properties of the upper domatic number of a graph, determine bounds on D(G), and compare D(G) to a related parameter, the transitivity Tr(G) of G.

1 Introduction

Partitioning the vertex set of a graph into subsets with a specified property is a popular subject in graph theory. Arguably the most famous problem of graph partitioning is graph coloring, where the vertex set is partitioned into independent sets. The minimum order of such a partition is the well-studied chromatic number of a graph. Considering subsets that are dominating sets, Cockayne and Hedetniemi [Citation1] in 1977 defined the domatic number to be the maximum order of a partition of the vertices of a graph into dominating sets.

More formally, a set SV(G) is a dominating set of a graph G=(V,E) if every vertex in V(G)S is adjacent to a vertex in S. The domination number of G, denoted γ(G), is the minimum cardinality of a dominating set of G. A dominating set with cardinality γ(G) is a called a γ-set of G. Given two disjoint sets of vertices R,SV, we say that R dominates S, denoted RS, if every vertex in S is adjacent to, or dominated by, at least one vertex in R. As previously mentioned, the domatic number of a graph G, denoted d(G), is the maximum order k of a vertex partition π={V1,V2,,Vk} such that for every i, 1ik, Vi is a dominating set of G. We call a partition π into dominating sets a domatic partition and a domatic partition of order d(G) is called a d-partition of G. Well over 100 papers have been published on this concept and its variations. The bibliography contains a sampling of these papers. For examples, cf. [Citation2–9].

An equivalent definition of the domatic number is that it equals the maximum order k of a vertex partition π={V1,V2,,Vk} such that for all i,j, 1i<jk, ViVj and VjVi. This definition of the domatic number suggests the following straightforward generalization. The upper domatic number, denoted D(G), equals the maximum order k of a vertex partition π={V1,V2,,Vk} such that for all i,j, 1i<jk, either ViVj or VjVi, or both. A vertex partition π meeting this condition is called an upper domatic partition and an upper domatic partition of order D(G) is called a D-partition of G.

In 2017, J. T. Hedetniemi and S. T. Hedetniemi [Citation10] gave a different generalization of the domatic number as follows. The transitivity of a graph Tr(G) is the maximum order k of a vertex partition π={V1,V2,,Vk} such that for all i,j, 1i<jk, ViVj. A vertex partition of V meeting this condition is called a transitive partition and a transitive partition of G having order Tr(G) is called a Tr-partition of G.

By definition, therefore, every domatic partition is both a transitive partition and an upper domatic partition, while every transitive partition is an upper domatic partition. Thus, we have the following basic inequalities for any graph G.

Proposition 1

For any graph G of order n, 1d(G)Tr(G)D(G)n.

Our aim in this paper is to introduce and study the upper domatic number of a graph. We present properties and bounds on D(G), and compare the upper domatic number and transitivity of graphs G.

1.1 Notation and terminology

In general, we follow the notation and terminology in [Citation11]. Let G=(V,E) be a graph. The open neighborhood of a vertex vV is the set N(v)={u|uvE}. Each vertex uN(v) is called a neighbor of v, and |N(v)| is called the degree of v, denoted deg(v). The minimum and maximum degrees of vertices in a graph G are denoted δ(G) and Δ(G), respectively. The closed neighborhood of a vertex vV is the set N[v]=N(v){v}. The open neighborhood of a set SV of vertices is N(S)=vSN(v), while the closed neighborhood of a set S is the set N[S]=vSN[v]. Let G[S] denote the subgraph induced by S in G.

A cycle and path on n vertices are denoted by Cn and Pn, respectively. The complete graph on n vertices is denoted by Kn, and its complement, the empty graph on n vertices, is denoted K¯n. A clique is a maximal complete subgraph of G, and the clique number ω(G) is the maximum cardinality of a clique of G.

Let π={V1,V2,,Vk} be a vertex partition of a graph G=(V,E). A subset Vi with |Vi|=1 is called a singleton, and a subset |Vi|2 is called a non-singleton.

1.2 Domination digraphs

An equivalent definition of domatic partitions can be given using domination digraphs. Let π={V1,V2,,Vk} be a vertex partition of a graph G=(V,E). From this partition one can construct a directed graph, or digraph, DG(π)=(π,E(π)), the vertices of which correspond one-to-one with the sets V1,V2,,Vk of π, with a directed arc ViVj in DG(π) if Vi dominates Vj in G, that is, VjN(Vi). We call DG(π) the domination digraph of π. Domination digraphs were introduced in 2012 by Goddard et al. [Citation12]. A digraph DG=(V,E) is called complete if for any two vertices vi,vjV, either vivj or vjvi, or both. A complete digraph is called bi-directed if for any two vertices vi and vj, both vivj and vjvi. A complete digraph DG=(V,E) is called a tournament if for any two vertices vi and vj, either vivj or vjvi, but not both. A tournament DG=(V,E) is called transitive if the vertices can be ordered v1,v2,,vn in such a way that for every 1i<jn, vivj if and only if i<j. A (complete) digraph DG of order n is said to be transitive if it contains a transitive tournament of order n as a sub-digraph, that is, DG contains a spanning transitive tournament.

A set Vi in a partition π={V1,V2,,Vk} is called a source set if the out-degree of vertex Vi in its domination digraph DG(π) is k1. And the set Vi is a sink set if the in-degree of vertex Vi in DG(π) is k1. Note that if π={V1,V2,,Vk} is a transitive partition, then V1 is a source set and Vk is a sink set.

Using domination digraphs, we can give equivalent definitions of transitive, domatic, and upper domatic partitions. A vertex partition π is transitive if its domination digraph is transitive, and is an upper domatic partition if its domination digraph is complete. The domatic number of a graph is the maximum integer k such that G has a vertex partition π={V1,V2,,Vk} whose domination digraph is a bi-directed complete graph.

2 The upper domatic number

2.1 Basic properties

We begin with some basic properties that will prove useful in the results that follow.

Proposition 2

If π={V1,V2,,Vr} is an upper domatic partition of a graph G, then the partition π={V1,V2,,Vi1,ViVj,Vi+1,,Vj1,Vj+1,,Vr} obtained from π by deleting Vi and Vj and adding to π the union ViVj, for any 1i<jr, is also an upper domatic partition.

Proof

In the vertex partition π, we only need to show that for any Vk, VkViVj, either Vk(ViVj) or (ViVj)Vk. There are essentially two cases: (i) VkVi and VkVj, or (ii) either ViVk or VjVk or both. For (i), Vk(ViVj), and for any of the three possibilities of (ii), (ViVj)Vk.

Corollary 3

If a graph G has an upper domatic partition of order k, then it has an upper domatic partition of order j, for any 1jk.

The following observations are straightforward.

Observation 4

Let π={V1,V2,,Vk}be an upper domatic partition of a graph G.

1.

For each Vi,Vj,ij, if ViVj, there are at least |Vj| edges in G between vertices in Vi and Vj.

2.

For each Vi,Vj,ij, there are at least min{|Vi|,|Vj|) edges in G between vertices in Vi and Vj.

3.

If vertices v1,v2,,vk are each in singleton sets in π, then the vertices v1,v2,,vk form a complete subgraph in G.

4.

If a vertex uVi is not adjacent to any vertex in Vj, then ViVj.

5.

If uVi with deg(u)=x<k, then outdeg(Vi)kx1 in DG(π). Further, if ViVj and u is adjacent to a vertex in Vj for some j, then outdeg(Vi)kx in DG(π).

It was shown in [Citation10] that the transitivity of any graph G is at least as large as the transitivity of any subgraph of G.

Proposition 5

[Citation10] If H is any subgraph of a graph G, then Tr(H)Tr(G).

This is not necessarily true for the upper domatic number. To see this, consider the bipartite graph G in . Let H be the induced subgraph Gv11, and note that π={V1={v1,v2,v3},V2={v4,v5,v6},V3={v7,v8,v9},V4={v10}} is an upper domatic partition of H in which V1V2V3V1 and V4 is a sink set. Thus, D(H)4, and it is straightforward to show that D(H)=4.

Fig. 1 D(H)=4>D(G)=3, where H=Gv11.

Fig. 1 D(H)=4>D(G)=3, where H=G−v11.

The graph G, on the other hand, satisfies D(G)<4. This follows by observing that since G has an isolated vertex v11, any upper domatic partition of G of order 4, π={V1,V2,V3,V4}, must have a source set V1 containing vertex v11. This means that the set V1{v11} is a dominating set of H. However, it can be seen that the domination number of H is 3, and that the vertices of any dominating set are incident to at least 8 of the 12 edges. Thus, one can check that there is no upper domatic partition of G of order 4 containing such a source set.

The following relaxed form of Proposition 5, however, does hold for the upper domatic number.

Proposition 6

If H is any subgraph of a graph G and H has a D-partition with a source set, then D(H)D(G).

Proof

To get an upper domatic partition for G, use the D-partition for H and add the vertices in V(G)V(H) to the source set in the D-partition for H.

Corollary 7

For any graph G, D(G)ω(G).

We note that the bound of Corollary 7 is sharp for complete graphs Kn, for which ω(Kn)=n=D(Kn). On the other hand, the difference D(G)ω(G) can be arbitrarily large; for complete bipartite graphs Kk,k, ω(Kk,k)=2<D(Kk,k)=k+1. Note that an upper domatic partition of order k+1 can be formed by placing one vertex from each partite set into singleton sets and adding k1 dominating sets containing one vertex from each partite set.

2.2 Small values of the upper domatic number

In this section we characterize the families of graphs G having upper domatic number equal to 1, 2, or at least 3. A star is a tree having at most one vertex of degree greater than one. A galaxy is a disjoint union of stars.

Theorem 8

For any graph G,

(1) D(G)3 if and only if G contains either a K3 or a P4,

(2) D(G)=2 if and only if G is a galaxy with at least one edge, and

(3) D(G)=1 if and only if G=K¯n.

Proof

(1) Assume that G contains a triangle or a P4 subgraph. Each of these subgraphs has upper domatic number 3 and a D-partition with a source set. Thus, Proposition 6 implies that D(G)3.

Conversely, assume that D(G)=k3. Let π={V1,V2,,Vk} be an upper domatic partition of G. Consider only V1, V2, and V3. Since π is an upper domatic partition, it follows that either ViVj or VjVi, or both, for every 1i<j3. From this we can assume either (i) the existence of a cyclic triple, where, without loss of generality, V1V2, V2V3 and V3V1, or (ii) the existence of a transitive triple, where V1V2, V1V3, and V2V3.

If (i) holds, let v1 be an arbitrary vertex in V1. Since V3V1, there must be a vertex v3V3, which is adjacent to v1. Similarly, since V2V3, there must be a vertex v2V2 which is adjacent to vertex v3. Finally, since V1V2, there must be a vertex v1V1 which is adjacent to v2. If v1=v1, then G has a triangle {v1,v2,v3}, but if v1v1, then G has a P4 subgraph (v1,v3,v2,v1).

If (ii) holds, let v3V3. Since V2V3, there must exist a vertex v2V2 which is adjacent to v3. Since V1V2, there is a vertex v1V1 which is adjacent to v2. Finally, since V1V3, there is a vertex v1V1, which is adjacent to v3. If v1=v1, then {v1,v2,v3} forms a triangle in G, but if v1v1, then (v1,v2,v3,v1) is a P4 subgraph of G.

(2) Let G=(V,E) be a galaxy with at least one edge, say uv. Then the partition defined by {V{u},{u}} is an upper domatic partition, and so D(G)2. But since G is a galaxy, it does not contain a K3 or a P4. Hence, from (1) we know that D(G)<3. Thus, D(G)=2.

Conversely, assume that D(G)=2. Then by definition, G has at least one edge. By (1) we know that G has no triangle and no P4 subgraph. Any trivial component of G is the star K1. Since G has at least one edge, G has at least one nontrivial component. Consider any nontrivial component H of G (if G is connected H=G), and let u be a vertex of H having maximum degree Δ(H). Since G has no triangles, N(u) is an independent set. If u has degree 1, then since u has maximum degree, H is the star K2. Assume that u has degree 2 or more. If a neighbor of u, say v, has a neighbor zN[u], then (w,u,v,z), where wN(u) and wv is a P4 subgraph of G, a contradiction. Hence, every neighbor of u has degree 1, so H is a star, implying that G is a galaxy.

(3) Trivially, D(K¯n)=1. Assume that D(G)=1. If GK¯n, then G has at least one edge, say uv. Thus, the partition defined by {{u},V{u}} is an upper domatic partition, and so, D(G)2, a contradiction. Thus, G=K¯n.

3 Bounds on the upper domatic number

For a set of vertices S, let GS be the graph formed from G by removing the set S from G.

Proposition 9

If G is a graph with order n and S is a γ-set of G, then D(G)D(GS)+1.

Proof

Since S along with the sets of a D-partition of GS is an upper domatic partition of G, D(G)D(GS)+1.

By Proposition 1, we have that d(G)D(G). Proposition 9 suggests a way of improving this bound for graphs with a d-partition containing a set that is not independent.

Proposition 10

If π={V1,V2,,Vk} is a d-partition of G, with Gi=G[Vi] for 1ik, then D(G)d(G)1+max{D(Gi):1ik}.

Proof

Let π={V1,V2,,Vk} be a d-partition of G, and let Gi=G[Vi]. Let D(Gj)=max{D(Gi):1ik} and let π be a D-partition of Gj. We note that (π{Vj})π is an upper domatic partition of G. Thus, D(G)d(G)1+D(G[Vj]).

Corollary 11

If G is a graph with d(G)=D(G), then every d-partition of G is a partition of the vertices of G into independent dominating sets.

Proof

Let π be a d-partition of G, and let Vj be a set in π that is not independent. Using the notation of Proposition 10, D(G)d(G)1+D(Gj). Since Vj is not an independent set, Theorem 8 implies that D(Gj)2. Thus, D(G)d(G)1+D(Gj)d(G)1+2=d(G)+1.

The following upper bound on the domatic number is well known.

Proposition 12

[Citation1] For any graph G, d(G)δ(G)+1.

This bound does not hold for D(G). It is immediately obvious that for any complete graph Kn, the domatic number d(Kn)=D(Kn)=n=δ(G)+1. Consider, however, the graph G formed by adding a new vertex u adjacent to exactly one vertex, say v, of a complete graph Kn. It is easy to see that d(G)=2=δ(G)+1, while D(G)=n, since a set containing {u,v} along with n1 singleton sets, each containing a vertex of Kn{u,v}, is a D-partition of G. Thus, for n3, D(G)>2=δ(G)+1, and we see that a significant difference can exist between the domatic number and the upper domatic number. With this example, we have proved the following.

Proposition 13

For any positive integer k, there is a graph G such that D(G)d(G)k, and such that D(G)(δ(G)+1)k.

Although the bound involving minimum degree in Proposition 12 does not hold for the upper domatic number, we show that replacing minimum degree with maximum degree is an upper bound for D(G).

Theorem 14

For any graph G, D(G)Δ(G)+1.

Proof

Let π={V1,V2,,Vk+1} be a D-partition of G ordered in such a way, that |Vi||Vj| for i<j. We claim there is at least one vertex vVk+1 with deg(v)k. By Observation 4, it follows there must be at least |Vk+1| edges between each of the sets Vi and Vk+1, for 1ik. Hence, the sum of the degrees of the vertices in Vk+1 is at least k|Vk+1|, and therefore, the average degree of the vertices in this set is at least k. Hence, D(G)=k+1Δ(G)+1.

It is shown in [Citation1] that d(G)+d(G¯)n+1. Our Nordhaus–Gaddum type bounds for the upper domatic number follow from Theorem 14.

Corollary 15

Let G be a graph of order n. Then D(G)+D(G¯)n+Δ(G)δ(G)+1.

Proof

Theorem 14 implies that D(G)+D(G¯)Δ(G)+1+Δ(G¯)+1. The result follows since Δ(G¯)=nδ(G)1.

We note that the bound of Corollary 15 is obtained by the self-complementary graph G=P4 for which D(G)+D(G¯)=3+3=4+21+1=n+Δ(G)δ(G)+1. It is also obtained the complete graph.

Corollary 16

If G is a regular graph, then D(G)+D(G¯)n+1.

An upper bound on the sum of the upper domatic number and the domination number of a graph also follows from Theorem 14. It is a well-known result (see Citation11]) that for graphs with order n, γ(G)nΔ(G).

Corollary 17

If G is a graph with order n, then D(G)+γ(G)n+1.

The following proposition was given in [Citation10].

Proposition 18

[Citation10] For any connected graph G with δ(G)=3, Tr(G)4.

Since Tr(G)D(G), we have the following corollary.

Corollary 19

For any connected graph G with δ(G)=3, D(G)4.

Based on Corollary 19, one might think that for any graph G, D(G)δ(G)+1. However, this is not the case. The following examples show that D(G)<δ(G) is possible. Indeed, the difference δ(G)D(G) can be arbitrarily large.

For integers t2 and k2, let Gt,k be the complete multipartite graph with t partite sets each having cardinality k. That is, Gt,k=K{k,k,,k} with t partite sets.

Proposition 20

For the graph Gt,k with t even, D(Gt,k)=kt+t2.

Proof

Let the partite sets be U1,U2,,Ut and let Ui={ui,1,ui,2,,ui,k} for 1it. Note that for any 1i<jt, {ui}{uj} and conversely, {uj}{ui}. Moreover, every two-element subset containing elements from different partite sets is a dominating set of Gt,k. That is, the partition π={{u1,1},{u2,1},,{ut,1},{u1,2,u2,2},{u3,2,u4,2},, {ut1,2,ut,2},,{u1,k,u2,k},{u3,k,u4,k},,{ut1,k,vt,k}} is an upper domatic partition. Thus, it follows that for this graph, D(G)t+(k1)t2=kt+t2.

To see that D(G)kt+t2, observe that any upper domatic partition of order larger than kt+t2 has at least t+1 singleton sets. But then at least two of these singleton sets are vertices in the same partite set and are not adjacent, contradicting Observation 4.

It follows from Proposition 20 that there exists a graph G for which D(G)=δ(G), and there is a graph G for which D(G)<δ(G). For examples, the graph G4,2 has D(G4,2)=δ(G4,2)=6, and the graph G6,2 has D(G6,2)=9<δ(G6,2)=10.

4 Upper domatic number versus transitivity

In this section, we first consider graphs G having equal transitivity and upper domatic number. We then turn our attention to the differences in these two parameters.

By considering Theorem 8 together with the following two results proven in [Citation10], we see that graphs G satisfying D(G)3 have equal upper domatic and transitivity numbers.

Theorem 21

[Citation10] Let G be a graph.

1.

Tr(G)=1 if and only if G=K¯n.

2.

Tr(G)=2 if and only if G is a galaxy with at least one edge.

Proposition 22

[Citation10] For any graph G, Tr(G)3 if and only if G contains either a triangle or a P4 subgraph (not necessarily induced).

We now consider other examples of graph families having this property. Recall that Pn and Cn denote the path and cycle, of order n, respectively. The Cartesian product GH of two graphs G and H is the graph with vertex set V(G)×V(H) and edges such that two vertices (u,v) and (w,x) are adjacent in E(GH) if and only if either u=w and v is adjacent to x in H, or u is adjacent to w in G and v=x. The class of graphs called n-cubes is defined recursively as follows. The 0-cube, denoted Q0, is the graph K1. The n-cube Qn is the graph Qn=Qn1P2.

Using the fact that Tr(G)D(G)Δ(G)+1 (Theorem 14) and proofs similar to those in [13,10], we can easily prove the following.

Theorem 23

1.

For the path Pn, n4, Tr(Pn)=D(Pn)=Δ(Pn)+1=3.

2.

For the cycle Cn, n3, Tr(Cn)=D(Cn)=Δ(Cn)+1=3.

3.

For any cubic graph G, Tr(G)=D(G)=Δ(G)+1=4.

4.

For m,n4, Tr(PmPn)=D(PmPn)=Δ(PmPn)+1=5.

5.

For m,n3, Tr(CmCn)=D(CmCn)=Δ(CmCn)+1=5.

6.

For any positive integer n, Tr(Qn)=D(Qn)=Δ(Qn)+1=n+1.

Theorem 23 gives several families of graphs having equal transitivity and upper domatic numbers. Acyclic graphs are another family with this property.

Theorem 24

If G is acyclic, then D(G)=Tr(G).

Proof

Let G be any acyclic graph, and let π={V1,V2,,Vk} be an upper domatic partition of cardinality k for G. We show that the sets in π can be reordered V1,V2,,Vk, so that for i<j, Vi dominates Vj. We first show that π contains a source set. Let Tr be a tree in G rooted at r. We consider two cases:

Case (i). Let rVi. For Case (i), if c is a child of r, then either cVi or Vi dominates the set of π containing c. Thus, Vi dominates all the sets of π containing vertices in N[r]. Suppose there is some set Vjπ that is not dominated by Vi. Then no child of r is in Vj, implying that Vj does not dominate Vi, contradicting that π is an upper domatic partition of G. Hence, Vi is a source set in π, and we can let V1=Vi.

Case (ii). If Case (i) does not hold, let u be a vertex of maximum depth in Tr, such that parent(u)’s set, say Vi, is different from the set Vj containing u and Vi does not dominate Vj. Since Case (i) does not hold, at least one child of r has this property, so the vertex u exists. Moreover, since π is an upper domatic partition, VjVi. Also note by the choice of u, if u has a child, it is either in Vj or Vj dominates the child’s set. Again, if there is a set say Va such that Vj does not dominate Va, then Va has no vertex in N(u). But then Vj does not dominate Va and Va does not dominate Vj, contradicting that π is an upper domatic partition. Thus, Vj is a source set in π.

Therefore, there exists a source set in π. We let this set be the first set V1 in a reordered π. We can create a new acyclic graph G=GV1. The partition π=πV1 will be an upper domatic partition for G. By repeating the above process, we can find the next set in the ordering of π, and after k1 iterations the entire ordering.

We have seen several examples where the inequality Tr(G)D(G) is tight. On the other hand, strict inequality is possible. For example, the graph G1 in has 4=Tr(G1)<D(G1)=5. To see that D(G1)=5, we note that {{u1,u2},{u3,u4},{u5,u6},{u7},{u8}} is an upper domatic partition, so D(G1)5, and Theorem 14 implies that D(G1)5. It can be verified by a straightforward, but detailed proof or by a computer search that G1 is a graph with the fewest vertices having transitivity strictly less than its upper domatic number. Similarly, it can be proved that the bipartite graph G2 in , for which 3=Tr(G2)<D(G2)=4, is a graph with the fewest edges having this property. We note that Propositions 22 and 25 imply that Tr(G2)=3. From Theorem 14, D(G2)4. The partition {{v1,v2,v3},{v4,v5,v6},{v7,v8,v9},{v10}} is an upper domatic partition, and so, D(G2)=4.

Fig. 2 Tr(Gi)<D(Gi) for i=1, 2.

Fig. 2 Tr(Gi)<D(Gi) for i=1, 2.

We next show that the difference between D(G) and Tr(G) can be made arbitrarily large. We will need the following result from [Citation10].

Proposition 25

[Citation10] If G is a graph with Tr(G)=k, then there exist two adjacent vertices in G, each having degree k1 or more.

Theorem 26

There exists a graph G for which D(G)Tr(G) differ by at least p, for any positive integer p.

Proof

Let n=2k+1 for some positive integer k>1. We construct a bipartite graph G with n(k+1) vertices and nk(k+1) edges as follows.

Let V(G)={s0,s1,s2,,sn1}{xi,j|0in1,1jk}. For 0in1, let Xi={xi,j|1jk} and Vi={si}Xi.

For 0in1, add edges such that si is adjacent to every vertex in Xp where p is the sum i+j modulo n for 1jk. Finally, add the edges sixa,b such that a+bi(modn). Now G is a bipartite graph with partite sets S={si|0in1} and X=Xi for 1in1, where each vertex in X has degree k+1 and each vertex in S has degree k(k+1).

We note that the partition π={V0,V1,,Vn1} is an upper domatic partition, so D(G)n=2k+1. On the other hand, since every edge of G is incident to a vertex in X with degree k+1, Proposition 25 implies that Tr(G)k+2, and so, D(G)Tr(G)k1.

5 Concluding remarks

We conclude by noting another difference in the transitivity number and the upper domatic number of a graph that leads us to making a conjecture. As we saw in Section 1.2, every transitive partition π has at least one source set and at least one sink set. In fact, it was observed in [Citation10] that a graph G will always have a Tr-partition with at least two sink sets. Upper domatic partitions, on the other hand, do not necessarily behave similarly.

For example, consider the graph G2 from . This graph has no D-partition with a source set. If G2 had a D-partition with a source set V1, then as in the proof of Proposition 29 , in the graph GV1, {V2,V3,V4} would be an upper domatic partition. But if D(GV1)=3, then Tr(GV1)=3 as well. This would imply that a transitive partition of G of order four could be constructed, a contradiction. Additionally, the graph G2 in has no D-partition with two sink sets. To see this, note that the graph G2 has Tr(G)<D(G)=4. Thus, if G2 had a D-partition with two sink sets, then the partition would also be transitive, again a contradiction.

The question arises whether there are graphs with no D-partitions with a sink set. Unfortunately, we have neither found such a graph nor proven that one does not exist.

Conjecture 1

There exists a graph G such that no D-partition of G contains a sink set.

Our next results show that if Conjecture 1 is true for a graph G, then D(G)5.

Proposition 27

If G has a D-partition π of order k that contains a set that is dominated by k2 sets, then G has a D-partition π containing a sink set.

Proof

Let π={V1,V2,,Vk} be an upper domatic partition that contains a set Vj that is dominated by k2 sets. If π does not have a sink set, consider the set Vi that is dominated by Vj. Let u be a vertex in Vj that dominates a vertex in Vi. Let X=ViVj{u}. Then π=π{Vi,Vj}{X,{u}} is a D-partition of G with the sink set {u} (cf. Proposition 2).

Corollary 28

If D(G)4, then there exists a D-partition that contains a sink set.

Proof

It follows from previous results that if D(G)3, then Tr(G)=D(G), and there is a Tr-partition π that is a D-partition. Thus, π has a sink set. Hence, assume that D(G)=4. Let π be a D-partition for G, and consider the domination digraph for π. There are four vertices and at least six arcs in this digraph, so some vertex must have in-degree at least 2=D(G)2. By Proposition 27, the result holds.

We also note that if Conjecture 1 holds, the following is true.

Proposition 29

If there exist graphs that have no D-partitions containing a sink set, then there exist graphs that have no D-partitions containing either a sink set or a source set.

Proof

Let G be a graph for which D(G)=k, and G has no D-partition containing a sink set. Assume that k is the minimum integer for which a graph G exists for which D(G)=k and G has no D-partitions containing a sink set. Let π={V1,V2,,Vk} be a D-partition of G, where, by assumption, Vk is not a sink set, and assume that V1 is a source set. This means that V1 is a dominating set of G. Then π={V2,V3,,Vk} is an upper domatic partition of order k1 of the graph G=GV1.

First, we claim that D(G)=k1. From the upper domatic partition π, we know that D(G)k1. Suppose D(G)>k1. Let {W1,W2,,Wr} be a D-partition of G, where rk. Then it follows that {V1,W1,W2,,Wr} is an upper domatic partition of G of order greater than k; a contradiction. Thus, D(G)=k1.

Now suppose that G has a D-partition {U1,U2,,Uk1} in which Uk1 is a sink. Then {V1,U1,U2,,Uk1} is a D-partition of G containing a sink; a contradiction.

Thus, G has no D-partition containing a sink, but D(G)=k1<k, which contradicts our assumption that k is the minimum integer for which a graph G exists having D(G)=k and no D-partitions containing a sink set.

Thus, it follows that G has no D-partitions containing either a sink set or a source set.

References

  • CockayneE.J., HedetniemiS.T., Towards a theory of domination in graphs, Networks, 7 1977 247–261
  • ChangG.J., The domatic number problem, Discrete Math., 125 1994 115–122
  • LuT.L., HoP.H., ChangG.J., The domatic number problem in interval graphs, SIAM J. Discrete Math., 3 1990 531–536
  • ManacherG.K., MankusT.A., Finding a domatic partition of an interval graph in time O(n), SIAM J. Discrete Math., 9 1996 167–172
  • PengS.L., ChangM.S., A simple linear time algorithm for the domatic partition problem on strongly chordal graphs, Inform. Process. Lett., 43 1992 297–300
  • RaoA.S., Pandu RanganC., Linear algorithm for domatic number problem on interval graphs, Inform. Process. Lett., 33 1989-1990 29–33
  • ZelinkaB., On k-domatic numbers of a graph, Czechoslovak Math. J., 33 1983 309–313
  • ZelinkaB., Domatically critical graphs, Czechoslovak Math. J., 30 1980 486–489
  • ZelinkaB., Domatic number and linear arboricity of cacti, Math. Slovaca, 36 1986 41–54
  • HedetniemiJ.T., HedetniemiS.T., The transitivity of a graph, J. Combin. Math. Combin. Comput., 104 2018 75–91
  • HaynesT.W., HedetniemiS.T., SlaterP.J., Fundamentals of Domination in Graphs 1998Marcel Dekker New York
  • GoddardW., HedetniemiS.M., HedetniemiS.T., McRaeA.A., The algorithmic complexity of domination digraphs, J. Combin. Math. Combin. Comput., 80 2012 367–384
  • HaynesT.W., HedetniemiJ.T., HedetniemiS.T., McRaeA.A., PhillipsN., The transitivity of special graph classes, J. Combin. Math. Combin. Comput.2018 (in press)