Introducing 3-Path Domination in Graphs

Rayan Ibrahim1, Rebecca Jackson1, Erika L.C. King1
1Hobart and William Smith Colleges, Geneva, New York, 14456.

Abstract

The dominating set of a graph G is a set of vertices D such that for every vV(G) either vD or v is adjacent to a vertex in D. The domination number, denoted γ(G), is the minimum number of vertices in a dominating set. In 1998, Haynes and Slater [1] introduced paired-domination. Building on paired-domination, we introduce 3-path domination. We define a 3-path dominating set of G to be D={Q1,Q2,,Qk|Qi is a 3-path} such that the vertex set V(D)=V(Q1)V(Q2)V(Qk) is a dominating set. We define the 3-path domination number, denoted by γP3(G), to be the minimum number of 3-paths needed to dominate G. We show that the 3-path domination problem is NP-complete. We also prove bounds on γP3(G) and improve those bounds for particular families of graphs such as Harary graphs, Hamiltonian graphs, and subclasses of trees. In general, we prove γP3(G)n3.

Keywords: Dominating set, 3-path, NP-complete

1. Introduction

In this paper, assume any graph G=(V,E) is finite and simple with vertex set V and edge set E. A dominating set DV of G is a set such that every vertex vV is either in D or adjacent to a vertex in D. The domination number of G, denoted γ(G), is defined as the minimum cardinality of a dominating set of G. An application of domination theory considers the task of assigning guards to rooms in a museum. In this application vertices represent the exhibit rooms and edges represent hallways joining them. Thus a dominating set of the representative graph is a set of vertices whose corresponding rooms could hold guards with the result that the entire museum would be under surveillance with guards monitoring adjacent rooms via the hallways.

Many variations on domination have been studied. See Haynes and Slater [2] for an introduction to many of these areas. In 1998, Haynes and Slater introduced paired-domination [1] in which the induced subgraph on a dominating set of vertices contains a perfect matching. In this version of the application, we say that every guard has another guard watching their back. The paired-domination number, denoted γpr(G), is the minimum cardinality of a paired-dominating set of G. We introduce a natural extension of paired-domination, namely 3-path domination. We say Qi is a 3-path if it is isomorphic to P3. We define a 3-path dominating set of G to be a set D={Q1,Q2,,Qk} of 3-paths such that the vertex set V(D)=V(Q1)V(Q2)V(Qk) is a dominating set of G. Here we continue our application analogy by allowing stationed guards to walk between 3 rooms and look down hallways extending from those 3 rooms. Note that while in paired domination every vertex of the perfect matching represents the position of a guard, in 3-path domination, not every vertex in V(D) necessarily will represent a position of a guard; only one guard is placed per 3-path. The 3-path domination number, denoted γP3(G), is the minimum cardinality of a 3-path dominating set. We will use the abbreviation γP3-set for a minimum 3-path dominating set. The study of the guard analogy and its variations was introduce in 1997 as the Watchman’s Walk Problem by Hartnell, Rall and Whitehead, see [3]. In 2003, Davies et al. [4] studied the Watchman’s Walk Problem on trees. The Watchman’s Walk Problem asks how many watchman are required such that each vertex is visited or observed by a watchman on a walk in a given time interval. The walks need not be uniform in length.

In this paper we prove that determining γP3 is NP-complete and then establish bounds on γP3 for all graphs, specifically γP3(G)n3. Tighter bounds and formulas for γP3 for specific families of graphs such as caterpillars, Harary graphs, banana trees, paths, and cycles are also shown

2. The 3-path Domination Problem is NP-complete

In 1998, Haynes and Slater proved that the paired domination problem was NP-complete [1]. We generalize this result to show the 3-path domination problem is also NP-complete.

Theorem 1. Deciding for a given graph H and positive integer k such that 3k||V(H), ”Is γP3(H) k?” is NP-complete.

Proof. We will use the known NP-complete domination problem,“For a given graph G and a positive integer k, is γ(G)k?” [5]. Let V(G)={v1,v2,,vn}. Construct graph H by letting V(Gi)={v1i,v2i,,vni} for 1i6 and letting vhivjiE(Gi) if and only if vhvjE(G). Let H be the graph created by these six disjoint copies of G and by adding the following edges. Let vh1vj2, vh3vj4, and vh5vj6 be in E(H) if and only if either h=j or vhvjE(G). Add the edges vh1vh3 and vh3vh5 for 1hn to H. Thus the graph H has 6n vertices and can be constructed from G in polynomial time.

We claim that γ(G)k if and only if γP3(H)k.

Assume DV(G) is a dominating set of G with ||Dk. Let R be a set of 3-paths with R={{vh1,vh3,vh5}|vhD}. Thus R is a 3-path dominating set of H with ||Rk, so γP3(H)k.

Now, assume R is a 3-path dominating set of G with ||Rk. Let T=QiRV(Qi). Since 3-paths are not necessarily disjoint and have 3 vertices each, ||T3k. Thus since G1G2G3G4G5G6, we can assume that ||T1,2=||T(V(G1)V(G2))k. Let T={vh2|vh1T1,2}(TV(G2)). Then ||T||T1,2k and T dominates V(G2). Hence, γ(G)k◻

Having shown that the 3-path domination problem is NP-complete, we find general bounds on γP3 and formulas and improved bounds for the 3-path domination number of specific families of graphs.

3. Bounds On the 3-path Domination Number

In this section, we prove bounds based on parameters of a graph G, namely γ(G), γpr(G), and Δ(G). Note that 3-path domination requires any component of a graph to have at least three vertices.

Theorem 2. For a graph G on n3 vertices, γP3(G)γ(G)3.

Proof. Let D be a γP3-set of G and V(D) be the set of vertices in D. Then |V(D)|3|D|=3γP3(G) since some 3-paths may share vertices. Furthermore, |V(D)|γ(G) as V(D) forms a dominating set of G. So we have 3γP3(G)|V(D)|γ(G), and solving yields γP3(G)γ(G)3◻

For the next lower bound, we generalize an argument from Haynes and Slater involving Δ(G) [1].

Theorem 3. For a connected graph G on n3 vertices, γP3(G)n3Δ(G).

Proof. Let D be a γP3-set of a graph G on n vertices, and let T be the set of edges in G having one vertex in V(D) and the other in V(G)V(D). Since Δ(G)deg(v) for all vV(D) and each vertex in V(D) has at least one neighbor in V(D), |T|=t(Δ(G)1)|V(D)|(Δ(G)1)3γP3(G). In addition, t|V(G)V(D)| since there is at least one edge of T incident to every vertex in G that is not in V(D). So, t|V(G)V(D)|=n|V(D)|n3γP3(G).

So we have n3γP3(G)t(Δ(G)1)3γP3(G), and solving yields γP3(G)n3Δ(G)◻

Having established some lower bounds on γP3, we explore some upper bounds.

Theorem 4. For a connected graph G on n3 vertices, γP3(G)γpr(G)2.

Proof. Let G be a graph with |V(G)|3 and D be a γpr-set of G. Since there are γpr(G)2 pairs of vertices in D and n3, for each pair we can create a 3-path using the pair and a neighbor. This forms a 3-path dominating set with cardinality γpr(G)2◻

Notice that since the induced subgraph on a γpr-set is a perfect matching, the edges in the induced subgraph are vertex disjoint. This is not always the case for 3-path domination, as shown in Figure 1.

For the following theorem, we note that a private neighbor of a 3-path Q is a vertex that is dominated by Q and no other 3-path of the 3-path dominating set. It is possible for the private neighbor to be a vertex of Q.

Theorem 5. For a graph G with n3, there exists a minimal 3-path dominating set that is edge-disjoint.

Proof. Let G be a graph and D be a minimal 3-path dominating set on G. Suppose there are two 3-paths in D that share an edge, Q1={v1,v2,v3} and Q2={v2,v3,v4}. Consider the following two possibilities:

Referring to Figure 2, Q1 and Q2 each have at least one private neighbor. If Q1 or Q2 do not have any private neighbors, then D is not minimal as G would still be dominated if we remove one path without private neighbors. Let p1 be a private neighbor of Q1 and p2 be a private neighbor of Q2. Note, p1 must be adjacent to v1, and p2 must be adjacent to v4, otherwise they would be dominated by both Q1 and Q2. Without loss of generality, we set Q1={v2,v1,p1}, making Q1 and Q2 edge-disjoint. Let C=DQ1+Q1. Now C is not guaranteed to be minimal. If C is not minimal, it must be the case that Q1 dominates what were the private neighbors of some other 3-path in D. Remove 3-paths from C that do not have any private neighbors in G with respect to C to obtain a new minimal 3-path dominating set D. Repeat this process until D is edge-disjoint and minimal. ◻

Corollary 1. For a connected graph G on n3 vertices, γP3(G)|E(G)|2.

Proof. Let G be a graph with |V(G)|3. By Theorem 5, there exists a minimal, edge-disjoint 3-path dominating set of G, call it D. Since no two 3-paths share an edge, and each 3-path contains two edges, |D||E(G)|2◻

Having an upper bound in terms of the number of edges of G is useful when dealing with classes of graphs whose number of edges directly relate to the number of vertices. For example, if we consider a tree T on n vertices, |E(T)|=n1, and so γP3(T)n12. We can improve this upper bound for trees. Recall a leaf of a tree is a vertex of degree 1, and the diameter of a graph G is the length of a longest path between a pair of vertices in G, denoted diam(G). We will call a vertex adjacent to a leaf a support vertex. Note that if diam(T)=2, then T is a star with n1 leaves, and n(n1)1=0 while clearly γP3(T)1 for all trees on at least three vertices. Thus for the following corollary we must assume that the diameter of the tree is at least three, so that nL11, where L is the number of leaves of T.

Corollary 2. For a tree T on n vertices with diam(T)3, γP3(T)nL12 where L is the number leaves of T.

Proof. Let T be a tree with diam(T)3. Obtain a new graph T by deleting all the leaves of T. Now T has nL vertices. Using Corollary 1, we have γP3(T)nL12. Let D be a 3-path dominating set of size nL12. Note that if T has an even number of edges, this would be as if all the edges of T were paired into disjoint 3-paths in D. Hence, if we added the L vertices of T back, all the support vertices of T would be in one of the 3-paths of D so D would form a 3-path dominating set in T as well. However, if T has an odd number of edges, |E(D)|=nL2, that is, there would be an edge of T, call it e, not included in the edge set of D. In T, it is possible that e is incident to a support vertex. Hence we may need one more 3-path in T than we needed in T. Therefore γP3(T)nL12◻

Increasing adjacencies in a graph, can only decrease the 3-path domination number. Thus we have the following observation.

Observation 6. If an edge is added to a graph G to form a new graph G, then γP3(G)γP3(G).

Thus the 3-path domination number of any spanning subgraph of G is an upper bound on the 3-path domination number of G. Refer to Figure 3 for an example. A spanning tree T of a graph G is a tree such that V(T)=V(G) and E(T)E(G). We now use Observation 6 and spanning trees to establish a bound for general graphs.

Figure 3: The First Graph, H, is a Spanning Subgraph of the Second, G. Note that γP3(H)=2, and we can choose paths {v1v2v3} and {v4v5v6} to form a γP3(G)-set. However, We only Need One 3-Path to Dominate G, namely {v2v3v4}

Theorem 7. For any connected graph G on n3 vertices, γP3(G)n32.

Proof. Let G be a graph on n vertices and let TG be a spanning tree of G. Notice, since V(TG)=V(G) and E(TG)E(G), we can construct G from TG by adding the edges in E(G)E(TG). By Observation 6, γP3(G)γP3(TG). Since TG is a tree on n vertices, γP3(TG)nL12 by Corollary 2. Since L2 for all trees with n3, γP3(G)n32◻

The next observation will be useful in this section and the following.

Observation 8. If a vertex u is a support vertex in V(G), then u must be part of a dominating 3-path.

Haynes and Slater [1] proved the following upper bound for paired-domination number.

Theorem 9. [1] If a connected graph G has n6 and δ(G)2, then γpr(G)2n3.

The following result follows readily from Theorem 1 and Theorem 1.

Corollary 3. If a connected graph G has n6 and δ(G)2, then γP3(G)n3.

We show that this bound actually holds for all graphs. While this may seem obvious since each 3-path uses three vertices, the paths in 3-path dominating sets are not necessarily vertex disjoint. Note that this bound is sharp. Consider P6 (graph H in Figure 3). This requires two 3-paths to dominate. We can generalize P6 with an infinite family of graphs, F, where each graph Gk in F is formed from k disjoint copies of P6 with vertices of each P6i labeled in order: vi,1,vi,2,,vi,6, and the additional edges v1,3v2,3,v2,3v3,3,,vk1,3vk,3. See Figure 4 for an example of Gk. Each Gk has 6k vertices, and by Observation 8, we must have 2k 3-paths in the dominating set. Therefore γP3(Gk)=n3.

Theorem 10. For any tree T on n3 vertices, γP3(T)n3.

Proof. Let T be a tree of order n3. We will proceed by induction on n. For n=3 we have T=P3, and γP3(P3)=1=33.

Let n>3. Either there exists a longest path P in T such that one of the vertices on P adjacent to its endvertices has degree greater than two, or the vertices adjacent the endvertices of every longest path have degree two. Note the endvertices of a longest path are leaves, and therefore the vertices adjacent to the endvertices on the path are support vertices. Refer to Figure 5 for an illustration of an example of a support vertex, v, of degree greater than two for Case 1, and of support vertices, labeled mi, all of degree two for Case 2.

In Case 1, let P be a longest path with a support vertex, v, adjacent to k2 leaves in T. Let e be the edge of P incident to v, but not to an endvertex of P. Then Te is a forest with components T on n(k+1) vertices and K1,k. If n(k+1)2, then T is a star or a double star and we only need one P3 to dominate T, which is less than n3. If n(k+1)3, the induction hypothesis gives γP3(T)n(k+1)3. Let D be a 3-path dominating set of T with |D|n(k+1)3. Note that we only require one 3-path, say Q, to dominate K1,k. Thus D=DQ is a 3-path dominating set of T with |D|γP3(T)+1=n(k+1)+33n3, since k+13.

In Case 2, all the support vertices of the endvertices of the longest paths have degree two. Let P be a longest path and let v be a vertex on P that is not a leaf and is adjacent to the support vertex of an endvertex of P, that is, a vertex that is distance 2 away from an endvertex of P. Consider the neighbors of v including neighbors not on P and the support vertex of the endvertex of P. Label these i neighbors mi. Note that each mi is either degree two or degree one otherwise we contradict the assumption of Case 2. For any mi that has degree two, label its leaf neighbor i. Let e be the edge of P incident to v, but not to an endvertex of P. Then Te is a forest with components H (containing v and all the mi and i) and T.

If |V(T)|2, then T is either P4, P5, or a spider with center vertex v and legs of length one or two. Note that if T is one of the two paths, we need only one P3 to dominate it, which is less than n3. Suppose T is a spider with k legs of length two, and j legs of length one. If k is even, then we create 3-paths between pairs of support vertices of the legs of length two to obtain a γP3-set of size k2. If k is odd, we create pairs of support vertices using k1 of the legs of length two, and add one 3-path from v to the leaf of the unpaired leg of length two to obtain a γP3-set of size k+12. (We use this method again when we look at banana trees in Theorem 4.5.) Note that in both cases, n=2k+j+1, and k2 and k+12 are both less than n3.

Suppose |V(T)|3. Then H is a spider. Let k be the number of paths of the form {v,mi,i}, i.e. legs of length two. If k is even, then we dominate H using k2 many 3-paths and by the induction hypothesis γP3(T)n(2k+1)3=n32k+13 (H contains v, 2k vertices for each leg of length two, and possibly more vertices if v has leaves as neighbors.) So γP3(T)γP3(T)+γP3(H)n32k+13+k2=n3k+26n3. If k is odd, then k1 and we need k+12 many 3-paths to dominate H, so similarly γP3(T)γP3(T)+γP3(H)n32k+13+k+12=n3k16n3.

By induction, we have shown that γP3(T)n3◻

The following result follows readily from Theorem 10 by considering a spanning tree of G.

Corollary 4. For any connected graph G on n3 vertices, γP3(G)n3.

4. Classes of Graphs and Their 3-path Domination Number

While bounds for general graphs are desirable, it is useful to restrict ourselves to families of graphs to obtain formulas or tighter upper bounds for those families. We begin by looking at path and cycle graphs.

Theorem 11. For n3, γP3(Pn)=γP3(Cn)=n5.

Proof. Consider Pn for n3. Suppose n=5q+k for nonnegative integers q and k, where k<5. We can partition Pn into q vertex-disjoint segments Si, where |V(Si)|=5, and one segment Sq+1 with k vertices. (Refer to Figure 6.) We can dominate at most five vertices with a 3-path in Pn, specifically the three in the 3-path and potentially two others adjacent to the ends. Thus we need one 3-path for each Si. Hence if k=0 we need q 3-paths and if k>0 we need q+1 3-paths. Hence γP3(Pn)n5.

The same argument holds for Cn, from which we can obtain Pn by deleting one edge. ◻

Corollary 5. If G is a connected graph on n3 vertices that has a Hamiltonian path, then γP3(G)n5.

Proof. Suppose P is a Hamiltonian path of G. Then PPn and since γP3(Pn)=n5 and G is obtained from P by adding edges, by Observation 6, we find that γP3(G)n5◻

Introducing even the slightest complexity to a graph can hinder our ability to find a formula for γP3(G). One such example is the caterpillar tree. A caterpillar is a tree in which every leaf is adjacent to a central path, or stalk. See Figure 7 for an example.

We will use the following observation in the proof of the next theorem.

Observation 12. If a new vertex is connected by a single edge to a graph G to form a new graph G, then γP3(G)γP3(G).

Theorem 13. Let A be a caterpillar with stalk S and m=||V(S). Then m+25γP3(A)m3.

Proof. Let A be a caterpillar with stalk S and m=|V(S)|. Then there exists a longest path, P, in A of length m+2, such that V(S)V(P). Any vertex in V(AP) must be a leaf that is adjacent to a vertex of P that is not a leaf. Thus γP3(A)γP3(P) by Observation 12. By Theorem 11, γP3(P)m+25.

By Observation 8, a caterpillar in which every vertex of the stalk is a support vertex has a 3-path domination number at least as large as any caterpillar with in which at least one stalk vertex is not a support vertex. Thus we consider the case where every viV(S) is a support vertex Then, each of the vi must be part of a 3-path by Observation 8. Furthermore, V(S) is a dominating set of A. Thus a minimum 3-path dominating set would require that the 3-paths be as vertex disjoint as possible. At most two of the 3-paths would need to share vertices. Thus at most m3 3-paths are needed. ◻

A banana tree, denoted Bn,k for n,k1, is a tree composed of n copies of a K1,k1 graph in which one leaf from each copy is joined by an edge to a vertex called the root vertex (See Figures 9 and 10 for examples).

Theorem 14. For n+k3, the following formulas hold for γP3(Bn,k):

  1. γP3(Bn,1)=1.

  2. γP3(Bn,2)=n2.

  3. For k3, γP3(Bn,k)=n.

Proof.

  1. Note that Bn,1 is a star, K1,n. Any 3-path in a star will contain the center vertex and thus dominate all vertices. Hence, γP3(Bn,1)=1.

  2. For k=2, we pick our 3-paths with the following process. In Bn,2, the support vertices are the vertices adjacent to the root vertex. We make unique pairs of support vertices, and choose the unique 3-path between each pair to be in our dominating set. If n is odd, then one support vertex is not in a pair and the last 3-path will include the leaf adjacent to the lone support vertex. See Figure [fig:banana1]. By Observation 1, every support vertex must be in a 3-path, so this is best possible.

    Figure 9: The Banana Trees B5,2 and B6,2. We Pair the Support Vertices si and Join Each Pair by a 3-Path With Middle Vertex r. In the n Odd Case, We Use the Leaf Adjacent to the Unpaired Vertex to Create the Last 3-Path
  3. For k3, by Observation 8, the center vertex of every copy of K1,k1 must be part of a 3-path. It is impossible to include any two centers in the same 3-path. So we must have a 3-path for each copy of K1,k1. We can insure that at least one of these 3-paths contains or is adjacent to the root vertex. Then one 3-path for each K1,k1 is best possible. Since we have n copies, γP3(Bn,k)=n. See Figure 10.

 ◻

A Harary graph, denoted Hk,n, is a k-regular graph of order n with kn1, V(G)={v1,v2,,vn} and adjacencies as follows. If k is even, then k=2j for some integer j and we join vi to {vij,vij+1,,vi1,vi+1,,vi1+j,vi+j}. If k is odd, then k=2j+1 for some j and n=2 for some , and we join vi to {vij,vij+1,vi1,vi+1,,vi1+j,vi+j} and vi+. See Figure 11 for an example.

Theorem 15. For k even, we have γP3(Hk,n)=n2k+1.

Proof. Suppose we have G=Hk,n where k=2j for some integer j. Note that a Hamiltonian cycle exists in each Harary graph and we can assume the vertices around this cycle are labeled v1,,vn. To construct a 3-path dominating set of G, we create 3-paths as follows. Define a 3-path, Q, by choosing any vertex vm to be the middle vertex and then choose neighbors vmj and vm+j to be the end vertices such that there is a path of length k2 from vm to vmj and vm+j. Notice, vm dominates k+1 vertices, and each of vmj and vm+j has an additional k2 private neighbors with respect to Q so long as n is sufficiently large. So, Q dominates k+1+k2+k2=2k+1 vertices. Note, Q dominates the largest number of vertices possible by a 3-path in G. In addition, all the dominated vertices lie on a single path (Figure 11). Suppose n=(2k+1)q+r where 0r2k. Then we can partition the labeled Hamiltonian cycle of G into q segments of 2k+1 vertices and one segment with r vertices. We choose a 3-path Q for each of those segments, and thus we can dominate G with n2k+1 3-paths and this is best possible. ◻

The case when k is odd is more complicated because the neighbors of a vertex are not confined to a sequential path along a Hamiltonian cycle. Ignoring the neighbors not along a sequential path, we can use the proof of Theorem 15 to obtain the following upper bound on the 3-path domination number of Harary graphs of odd degree.

Corollary 6. For k odd, we have γP3(Hk,n)n2k1.

Acknowledgments

The authors wish to thank Bert Hartnell of Saint Mary’s University in Halifax for his suggestions and support. The authors also wish to thank Kevin O’Bryant of CUNY – Staten Island and Daniel Cranston of Virginia Commonwealth University for their feedback and suggestions.

This article is based on the research completed at the NSF HWS-REU (grant no. DMS 1757616) in the summer of 2018. At the time, Rayan Ibrahim was a student at the College of Staten Island, and Rebecca Jackson was a student at Charleston Southern University. As of January 2020, Ibrahim is a graduate student at Virginia Commonwealth University and Jackson is a graduate student at the University of Arkansas.

Conflict of Interest

The authors declare no conflict of interest.

References:

  1. Haynes, T. W. and Slater, P. J., 1998. Paired-domination in graphs. Networks, 32(3), pp.199-206.

  2. Haynes, T.W., Hedetniemi, S. and Slater, P., 2013. Fundamentals of Domination in Graphs. CRC press.
  3. Gary, M.R. and Johnson, D.S., 1979. Computers and Intractability: A Guide to the Theory of NP-completeness.
  4. Davies, C., Finbow, S., Hartnell, B., Li, Q. and Schmeisser, K., 2003. The watchman problem on trees. Bulletin of the Institute of Combinatorics and its Applications, 37, pp.35-43.

  5. Hartnell, B.L., Rall, D.F. and Whitehead, C.A., 1998. The Watchman’s Walk Problem. Congressus Numerantium, pp.149-156.