Broadcast Domination in Line Graphs of Trees

Jishnu Sen1, Srinivasa Rao Kola1
1Department of Mathematical and Computational Sciences National Institute of Technology Karnataka, Surathkal Mangalore – 575025, India.

Abstract

A dominating broadcast of a graph G is a function f:V(G){0,1,2,,diam(G)} such that f(v)e(v) for all vV(G), where e(v) is the eccentricity of v, and for every vertex uV(G), there exists a vertex v with f(v)>0 and d(u,v)f(v). The cost of f is vV(G)f(v). The minimum of costs over all the dominating broadcasts of G is called the broadcast domination number γb(G) of G. A graph $G$ is said to be radial if γb(G)=rad(G). In this article, we give tight upper and lower bounds for the broadcast domination number of the line graph L(G) of G, in terms of γb(G), and improve the upper bound of the same for the line graphs of trees. We present a necessary and sufficient condition for radial line graphs of central trees, and exhibit constructions of infinitely many central trees T for which L(T) is radial. We give a characterization for radial line graphs of trees, and show that the line graphs of the i-subdivision graph of K1,n and a subclass of caterpillars are radial. Also, we show that γb(L(C))=γ(L(C)) for any caterpillar C.

Keywords: Dominating broadcast, Broadcast domination number, Domination, Line graph

1. Introduction

Broadcast domination is a variety of distance related domination parameter. For a radio station, the range and the cost of a broadcasting tower depends on the capacity of the tower. Tactical installation of broadcasting towers of varying capacities at different locations of a region so that the whole region can hear the broadcast from the radio station is one of the real-life instances of the application of the broadcast domination.

Formally, a broadcast of a graph G is a function f:V(G){0,1,2,,diam(G)} such that f(v) is at most the eccentricity of v for all vV(G). The cost of f is vV(G)f(v) and is denoted by σ(f). A vertex v is a broadcasting vertex if f(v)1, and the set of all broadcasting vertices is denoted by Vf+(G) or simply Vf+. A vertex uV(G) is said to be f-dominated by a broadcasting vertex v if d(u,v)f(v). It is clear that a broadcasting vertex is f-dominated by itself. A vertex uV(G) is said to be over-f-dominated or overdominated (when the broadcast is clear from the context) by a broadcasting vertex v if d(u,v)<f(v). A vertex u is said to be a boundary vertex of a broadcasting vertex v if d(u,v)=f(v), and the set of all boundary vertices of a broadcasting vertex v is denoted by Bf(v). For each vertex vVf+, the closed f-neighborhood Nf[v] of v is the set {uV(G):d(u,v)f(v)}. A broadcast f is a dominating broadcast if vVf+Nf[v]=V(G), and the minimum of costs over all the dominating broadcasts of G is called the broadcast domination number γb(G) of G. A dominating broadcast f of G is said to be an optimal dominating broadcast if σ(f)=γb(G). A dominating broadcast f of G is said to be an efficient dominating broadcast if every vertex of G is f-dominated by exactly one broadcasting vertex. From now onwards, we use DB and ODB in place of dominating broadcast and optimal dominating broadcast, respectively. In a graph G, a set SV(G) is said to be a dominating set if every vertex in G is either belongs to S or adjacent to some vertex of S. The domination number γ(G) is the minimum size of a dominating set in G.

The concept of the line graph of a graph was first introduced by [1] but the formal studies on the line graph itself was by [1]. Two edges in a simple graph are said to be adjacent if they are incident on a common vertex. For a given simple graph G with at least one edge, the line graph L(G) is defined as the graph such that V(L(G))=E(G) and two vertices of L(G) are adjacent if the corresponding edges are adjacent in G.

In this paper, every graph, unless mentioned, is simple and connected. For any two vertices u and v in a graph, uv is written for “u and v are adjacent”. An edge with end vertices u and v is denoted as uv. If H is a subgraph of a connected graph G, then for any vertex uV(G), dG(u,H) or simply d(u,H) is min{dG(u,x):xV(H)}. A caterpillar C is a tree of order at least 3 such that deletion of all the leaves yield a path called the spine. A vertex of spine is called stem if it is adjacent to a pendant vertex and a vertex of spine is called trunk if it is not adjacent to any pendant vertex. It is clear that the first and last vertices of the spine are necessarily stem vertices. In a graph G, the ith-subdivision of an edge uv is an operation of deleting the edge uv, introducing an x,y-path of order i, and making u and v adjacent to x and y, respectively. For any graph G, the i-subdivision graph of G, denoted by Si(G), is the graph obtained from G by considering ith-subdivision for every edge of G.

1.1. Literature survey

The concept of broadcast domination was first presented by [3]. For a graph G, [3] gave bounds for γb(G) in terms of the diameter, the radius and the domination number of G.

Theorem 1. [3] For a non-trivial connected graph G, diam(G)+13γb(G)min{γ(G),rad(G)}.

In the light of the upper bound given in Theorem 1, [4] proposed the following classification of graphs and related studies can be found in [3,6,78]. A graph G is

  • Type I or 1-cap graph if γb(G)=γ(G).

  • Type II or radial graph if γb(G)=rad(G).

  • Type III if γb(G)<min{γ(G),rad(G)}.

[4] proved that every graph has an efficient ODB. [6] characterized radial trees. If P is a diametrical path of a tree T, then a set M of edges of P is said to be a split-P-set if end vertices of every edge of M have degree two and for every component T of TM, the path PT is a diametrical path of T with even positive length. A tree is said to have a split-set if it has a split-P-set for some diametrical path P. In Figure 1, the set {x1y1,x2y2} is a split-set of T corresponding to the diametrical path induced by the black colored vertices.

Theorem 2. [6] A tree is radial if and only if it has no non-empty split-set.

Let G be any graph of order n and H1,H2,H3,,Hn be any graphs. Then, the generalized corona G(H1,H2,H3,,Hn) is the graph obtained by making adjacent the ith vertex of G with all the vertices of Hi.

Theorem 3. [6] For any connected graph G of order n and for any graphs H1,H2,H3,,Hn, the graph G(H1,H2,H3,,Hn) is radial.

[9] observed that for any spanning subgraph H of G, γb(G)γb(H). [1] showed that the broadcast domination number of a graph is the minimum of the broadcast domination numbers of its spanning trees. [6] proved that the broadcast domination number of a graph of order n is at most n3. For an efficient ODB f of G, [11] defined the domination graph Gf, where V(Gf)=Vf+ and uvE(Gf) if N(Nf[u])Nf[v]. Further, they proved the result below.

Theorem 4. [11] For any graph G, there exists an efficient optimal dominating broadcast f such that Gf is either a path or a cycle.

For trees T, the above theorem can be restated as, there exists an efficient ODB f such that Tf is a path.

[10] was the first author to characterize line graphs. For a brief survey on line graphs, one may look into [2].

Theorem 5. [10] A graph G is a line graph of some connected graph if and only if G has a clique decomposition such that every vertex of G lies in at most two cliques.

For a graph G, a collection K of subgraphs is said to be a Krausz decomposition if it has the following three properties.

  1. Each element of K is a complete graph.

  2. Every edge of G belongs to exactly one element of K.

  3. Every vertex of G belongs to exactly two elements of K.

Considering K1 also as a clique, Theorem 5 can be restated as, a graph G is a line graph of a connected graph if and only if G has a Krausz decomposition. For any graph G, the edges incident on each vertex form a clique in L(G) and the collection of all those cliques is a Krausz decomposition of L(G). In a Karusz decomposition, we say a clique as a trivial clique if it is K1. The theorem below implies that for any tree T except K1,3, L(T) has a unique Krausz decomposition.

Theorem 6. [2] Every connected line graph G, other than the line graphs of K1,3,K1,3+e,K4e and K4, has a unique Krausz decomposition.

Our contributions start from Section 2. In Subsection 2.1, we prove that γb(G)2γb(L(G))2γb(G) for any connected graph G and the upper bound is further improved to γb(L(T)) for trees T. As γb(L(T))γb(T) for any bicentral and radial tree T, for central trees T, we show that L(T) is radial if and only if T is radial and γb(L(T))=γb(T). We give a characterization for radial line graphs of trees in Subsection 2.2, by defining a concept of separating-set for trees. Further, we give the value of γb(L(T)) involving the maximum size of a separating-set in T. We denote the class of central trees T with L(T) is radial as TR and endow a couple of constructions of infinitely many trees of TR. We study the broadcast domination in line graphs of caterpillars and i-subdivision graph of K1,n in Subsection 2.3. We prove that the line graphs of caterpillars are Type I, and show that a subclasss of caterpillars and i-subdivision graph of K1,n belong to TR.

2. Results

2.1. Upper and lower bounds for γb(L(G))

In this subsection, we present a relation between γb(G) and γb(L(G)). In the line graph L(G) of G, we denote the vertex corresponding to the edge e of G as ve. Conversely, the edge of G corresponding to the vertex u in L(G) is denoted as eu.

Theorem 7. For any connected graph G, γb(G)2γb(L(G))2γb(G).

Proof. Let f be an efficient ODB of L(G) and {u1,u2,u3,,uk}=Vf+(L(G)). Then, we define a broadcast g of G as below. g(x)={f(ui)+1when x=xi, where eui=xiyi1ik,0otherwise. It is easy to see that g is a DB of G with cost σ(f)+|Vf+(L(G))|. Therefore, γb(G)2γb(L(G)).

Let f be an efficient ODB of G and {v1,v2,v3,,vl}=Vf+(G). Let ej be an edge incident on vj, 1jl. Now, we define a DB g of L(G) as follows. g(x)={f(vj)+1when x=vej1jl,0otherwise. Therefore, γb(L(G))σ(f)+|Vf+|. Hence, γb(L(G))2γb(G)

A bistar graph B(m,n) is a graph obtained by joining centers of two star graphs K1,m and K1,n by an edge. A wheel graph Wn is a graph on n+1 vertices which contains a cycle Cn and all the vertices of Cn are adjacent to a single vertex. It is easy to verify that γb(B(m,n))2=γb(L(B(m,n))) and γb(L(Wn))=2γb(Wn).

From the proof of Theorem 7, it is evident that, working on an efficient ODBs with minimum number of broadcasting vertices give better bounds. Since, for any tree, there exists no edge between two boundary vertices of a broadcasting vertex and there exists a unique path between two broadcasting vertices, the upper bound given in Theorem 7 is improved for trees in Theorem 9.

For any shortest path u1u2u3uk of order k in a graph G, there exists a shortest ve1,vek1-path ve1ve2ve3vek1 of order k1 in L(G), where ei=uiui+1, i=1,2,3,,k1. Conversely, for every shortest w1,wl-path w1w2w3wl of order l in L(G), there exists a shortest u1,ul+1-path u1ew1u2ew2u3ew3u4ulewlul+1 of order l+1 in G. Hence, we have the following observation. We say a tree is central (bicentral) if its center is K1 (K2).

Observation 8. If T is a central tree, then rad(L(T))=rad(T), and if T is a bicentral tree, then rad(L(T))=rad(T)1.

Theorem 9. For any tree T, γb(L(T))γb(T).

Proof. For radial trees, it is easy to observe from Theorem 1 and Observation 8 that γb(L(T))γb(T). For any non-radial tree T, every ODB of T has at least two broadcasting vertices. Let f be an efficient ODB of T such that Tf is a path. Let {v1,v2,v3,,vk}=Vf+(T) and Tf:v1v2v3vk. Then, there is a v1,vk-path in T containing all the broadcasting vertices. Let ei be the edge incident on vi and lie on vi,vi+1-subpath of v1,vk-path in T, 1ik1. Let ek be the edge incident on vk and lie on vk1,vk-subpath of v1,vk-path in T. Let g be a DB of L(T) defined as follows. g(x)={f(vi)when x=vei, for 1ik,0otherwise. Therefore, γb(L(T))γb(T)

It is straightforward from Theorem 9 and Observation 8 that if a tree T is radial and bicentral, then γb(L(T))<γb(T). Hence, if a tree T is radial and bicentral, then L(T) can be radial or non-radial. Therefore, if a radial tree T has the property γb(T)=γb(L(T)), then T must be central. The following result is a direct implication of Theorem 9 and Observation 8.

Corollary 1. For a central tree T, L(T) is radial if and only if T is radial and γb(T)=γb(L(T)).

2.2. Radial line graphs of trees

In this subsection, we characterize those line graphs of trees that are radial. Further, we construct infinitely many radial line graphs of central trees. Before advancing, we give some results that we use in the subsequent part of the paper.

Observation 10. In the line graph L(T) of any tree T, there exists a unique shortest path between every pair of vertices.

Proof. Since every shortest path in L(T) implies a path in T, we have the proof. 

Lemma 1. For any tree T, the line graph L(T) has an efficient optimal dominating broadcast f such that (L(T))f is a path.

Proof. Let g be an efficient ODB of L(T) such that (L(T))g is a cycle. Let (L(T))g be v1v2v3vnv1, where {v1,v2,v3,,vn}=Vg+(L(T)). Now, the shortest vi,vi+1-paths for 1in1, and the shortest vn,v1-path together form a cycle C in L(T). Then, the end vertices of the edges in T corresponding to the vertices of C form a cycle in T, which is a contradiction. Therefore, L(T) has no efficient ODB g such that (L(T))g is a cycle. By Theorem 4, L(T) has an efficient ODB f such that (L(T))f is a path. 

Theorem 11. The line graph L(T) of a tree T has an efficient optimal dominating broadcast f such that the broadcasting vertices lie on a diametrical path of L(T). Moreover, the end vertices of the diametrical path are not over-f-dominated if one of the following conditions hold.

  1. T is bicentral.

  2. T is central and L(T) is non-radial.

Proof. We divide our proof into two cases. For any two vertices u and v, we denote a shortest u,v-path as uv. A broadcast f of a graph G is a radial broadcast if f(v)=rad(G) for some central vertex v, and f(u)=0 if uv.
Case 1: L(T) is radial
As L(T) is radial, the radial broadcast f of L(T) is an efficient ODB of L(T) with the broadcasting vertex lying on a diametrical path in L(T). One can note that for any diametrical path in T, there is a diametrical path of length diam(T)1 in L(T), and vice versa. If T is bicentral, then there is a unique central vertex in L(T) and diam(L(T)) is even, and hence, the end vertices of the diametrical path are not over-f-dominated.
Case 2: L(T) is non-radial
Every ODB of L(T) has at least two broadcasting vertices. By Lemma 1, let g be an efficient ODB of L(T) such that (L(T))g is a path. Let (L(T))g be v1v2v3vn, where {v1,v2,v3,,vn}=Vg+(L(T)). Let x1 and x2 be boundary vertices of v1 and vn, respectively, and xj and xj be two distinct boundary vertices of vj for j=2,3,4,,n1 such that x1x2 and xjxj+1, j=2,3,4,,n1. Let Q be the shortest v1,vn-path in L(T) described as below (see Figure 2). The dashed circle around vj in Figure 2 is Ng[vj]. v1x1x2v2x2x3v3x3x4xn2xn1vn1xn1xnvn

Let X1 (X2) be the set of farthest vertices from v1 (vn) in Ng[v1] (Ng[vn]). If there exists a vertex u1X1 (u2X2) such that d(u1,vn)=d(u1,v1)+d(v1,vn) (d(u2,v1)=d(u2,vn)+d(vn,v1)), then the path u1Qu2 is the diametrical path of L(T) which contains all the broadcasting vertices of g. If for every vertex xX1, d(x,vn)<d(x,v1)+d(v1,vn) but the vertex u2 exists in X2, then let u1 be a vertex in X1 such that u1v1 does not intersect Q. Then, the broadcast g defined as g(w1)=g(v1), g(v1)=0 and g(x)=g(x) for all xV(L(T)){w1,v1}, where w1 is the neighbor of v1 in v1u1, is an efficient ODB of L(T). If w1 is the neighbor of v1 in Q, then the path u1w1w1vnu2 is the diametrical path of L(T) which contains all the broadcasting vertices of g. If for every vertex yX2, d(y,v1)<d(y,vn)+d(vn,v1) but the vertex u1 exists in X1, then, for a vertex u2X2 with u2vn does not intersect Q, we can define an efficient ODB, similar to g, whose broadcasting vertices lie on the diametrical path u1v1w2w2u2, where w2 and w2 are the neighbors of vn in vnu2 and Q, respectively. If for every vertex xX1 and yX2, d(x,vn)<d(x,v1)+d(v1,vn) and d(y,v1)<d(y,vn)+d(vn,v1), then we can choose two vertices u1X1 and u2X2 such that u1v1 and u2vn do not intersect Q. Then, similar to the above, we can get an efficient ODB whose broadcasting vertices lie on the diametrical path u1w1w1w2w2u2, where w1 (w2) is the neighbor of v1 (vn) in v1u1 (vnu2). As g is an ODB of L(T), it is not possible that for every vertex xX1 (yX2), d(x,vn)<d(x,v1)+d(v1,vn) (d(y,v1)<d(y,vn)+d(vn,v1)) and xv1 (yvn) intersects Q; else we can define a DB g of L(T) as g(w1)=g(v1)1, g(w2)=g(vn)1, g(v1)=0, g(vn)=0 and g(x)=g(x) for all xV(L(T)){v1,vn,w1,w2} such that σ(g)<σ(g)=γb(L(T)), which is not possible.

For the ease of analysis in the latter part of the proof, we rename the efficient ODB whose broadcasting vertices lie on a diametrical path, as f; the broadcasting vertices as v1,v2,v3,,vn with (L(T))f:v1v2v3vn; the shortest v1,vn-path in L(T), as described earlier, as P. Let P:u1Pu2 be the diametrical path of L(T). Let v1~ and vn~ be the neighbors of v1 and v2 in P, respectively.

If both u1 and u2 are not over-f-dominated, then f is our desired efficient ODB and P is our desired diametrical path in L(T). Suppose that at least one of u1 and u2 is over-f-dominated. If u1 is over-f-dominated, then d(u1,v1)=f(v1)1, else we have another DB f defined by f(v1~)=f(v1)1, f(v1)=0 and f(x)=f(x) otherwise, having σ(f)<γb(L(T)) which is not possible. Similarly, if u2 is over-f-dominated, then d(u2,vn)=f(vn)1.

Suppose u1 is overdominated by v1, and u2 is not overdominated by v2. Then, we perform a couple of modifications on f to have a new efficient ODB whose broadcasting vertices lie on P and u1 is not overdominated. Let f be the ODB defined as f(v1~)=f(v1), f(v1)=0 and f(x)=f(x) for all xV(L(T)){v1,v1~}. Then, u1 is not over-f-dominated and all the broadcasting vertices lie on P. Now, Nf[v1~]Nf[v2] with f(v1~)+f(v2)=d(v1~,v2). Then, there exists a vertex w on v1~,v2-subpath of P such that d(v1~,w)=f(v2) and d(w,v2)=f(v1~). Now, we modify f to f as f(w)=f(v1~)+f(v2),f(x)=0 for x=v1~,v2, and f(y)=f(y) for all yV(L(T)){w,v1~,v2}. It is easy to verify that f is an efficient ODB. If u2 is over-f-dominated but not u1 or both u1 and u2 are over-f-dominated, then, in a similar manner, we can get an efficient ODB h whose broadcasting vertices lie on a diametrical path, and u1 and u2 are not over-h-dominated. 

Motivated by the concept of a split-set in a tree, we define separating-set, a collection of vertices in a tree, and prove that the radialness of L(T) depends on the existence of a separating-set in a tree T. Later, we show a relation between a split-set and a separating-set of a tree. For any diametrical path of T, let vi and vj be two vertices on that path such that deg(vi) and deg(vj) are less than or equal to 2, and P denotes the vi,vj-path in T. We define Ti,j is the subtree of T induced by S=vV(P){wV(T):dT(w,P)=dT(w,v)}.

Definition 1. Let T be a tree and P:v1v2v3vn be a diametrical path of it. We say a collection F={vi1,vi2,vi3,,vit}, i1<i2<i3<<it, of vertices of P is a separating-P-set if

  1. deg(vij)=2, for j=1,2,3,,t,

  2. d(vij1,vij), 1jt+1, is odd and greater than one, where vi0=v1 and vit+1=vn, and

  3. vij1,vij-path is a diametrical path of the subtree Tij1,ij, for j=1,2,3,,t+1.

We say a tree has a separating-set if T has a separating-P-set for some diametrical path P.

Example 1. In Figure 3, the collection {v4,v7} is a separating-set of the tree corresponding to the diametrical path v1v2v3v4v5v6v7v8v9v10.

Theorem 12. For any tree T, L(T) is radial if and only T has no separating-set.

Proof. We prove the contrapositive of the statement. Let T be a tree with a diametrical path P:u1e1u2e2u3e3u4un1en1un. Then, the path P:ve1ve2ve3ven1 is a diametrical path of L(T). Suppose T has a separating-P-set {ui1,ui2,ui3,,uit}, i1<i2<i3<<it, and let ui0=u1 and uit+1=un. Let f be a broadcast defined as follows. f(x)={ijij112when x=ve(ij+ij11)/2 for 1jt+1,0otherwise. As for each j{1,2,3,,t+1}, ui(j1),uij-path is a diametrical path of the subtree Ti(j1),ij of odd length, the corresponding vei(j1),ve(ij1)-subpath of P is the diametrical path of the subgraph L(Ti(j1),ij) of even length. Moreover, since the broadcast assigns label rad(L(Ti(j1),ij)) to the middle vertex of the vei(j1),ve(ij1)-subpath for each j{1,2,3,,t+1}, f is a DB of L(T) with σ(f)=nt22<n22n22=rad(L(T)). Therefore, L(T) is not radial.

Conversely, suppose L(T) is not a radial graph. By Theorem 11, there exists an efficient ODB g of L(T) such that all the broadcasting vertices lie on a diametrical path, say Q, and none of the end vertices of Q are over-g-dominated. If Q is w1w2w3wn, then the corresponding path Q:v1ew1v2ew2v3ew3v4vn1ewn1vnewnvn+1 is a diametrical path in T. Let the set of broadcasting vertices of g be {wi1,wi2,wi3,,wim+1}, i1<i2<i3<<im+1, and w1 and wn be the boundary vertices of wi1 and wim+1, respectively. For every consecutive pair of broadcasting vertices wij and wij+1, there exists a pair of adjacent vertices xij2 and xij+11 such that xij2Bg(wij) and xij+11Bg(wij+1), where 1jm. If vkj is the common end vertex of exij2 and exi(j+1)1 in T for 1jm, then we consider the set F={vk1,vk2,vk3,,vkm}. Moreover, k1<k2<k3<<km.

Now, we prove that the set F is indeed a separating-Q-set of T. It is clear that the degree of each vertex in F is 2. Let v1=vk0 and vn+1=vkm+1. As each broadcasting vertex g-dominates exactly odd number of vertices of Q in L(T), d(vkj1,vkj), 1jm+1, is odd and greater than one. Moreover, since g is an efficient ODB of L(T) such that none of the end vertices of Q are over-g-dominated, vkj1,vkj-subpath is a diametrical path of the subtree Tkj1,kj, for j=1,2,3,,m+1. Therefore, the set F is a separating-Q-set of T

Remark 1.

  1. It is easy to observe that if diam(T)5, then the tree T cannot have a separating-set and hence L(T) is radial.

  2. If P is a diametrical path of T and T has a split-P-set {x1y1,x2y2,x3y3,,xtyt} (t2) of size at least two, then T has a separating-P-set {x2,x3,x4,,xt}. [6] have proved that if a central tree is not radial, then the maximum size of a split-set is at least two. So, it is straightforward that if a central tree T is not radial, then L(T) is not radial. If the maximum size of a split-set in T is one, then T may or may not have a separating-sets.

For any non-radial L(T) of a tree T with diam(T)=n1, let g be an efficient ODB of L(T) such that all the broadcasting vertices lie on a diametrical path of L(T) and none of the end vertices are over-g-dominated, and Fg be the corresponding separating-set of T obtained in the proof of Theorem 12. Then, γb(L(T))=σ(g)=n|Fg|22. If Fg is not of maximum size, then there exists a separating-set of T of size more than |Fg| and the corresponding DB defined in Equation (???) of the proof of Theorem 12, in L(T), has cost less than g, which is a contradiction. Therefore, we have the following result.

Theorem 13. Let T be a tree of diameter n1. Then, γb(L(T))={n12when L(T) is radial,nt22when L(T) is non-radial and t is the maximum sizeof a separating-set of T.

Remark 1. In order to determine γb(L(T)) when L(T) is non-radial, finding the maximum of the sizes of all the separating-sets of T is the key task. One of the naive approaches is to find a separating-set of maximum size for every diametrical path and then find the separating-set of maximum cardinality among all such separating-sets. It is to be observed that for a diametrical path P:v1v2v3vn, if the set F={vi1,vi2,vi3,,vit}, i1<i2<i3<<it, is a separating-P-set, then i2k1 is always even and i2k is always odd, for k=1,2,3,. One can exploit this property to obtain a separating-set of maximum size for a tree.

Theorem 14. For a central tree T, the following statements are equivalent.

  1. T is radial and γb(L(T))=γb(T).

  2. L(T) is radial.

  3. T has no separating-set.

  4. L(T) has no efficient optimal dominating broadcast f such that all the broadcasting vertices lie on a diametrical path and the end vertices of the diametrical path are not over-f-dominated.

Proof. The equivalencies of 1, 2 and 3 are straightforward from Corollary 1 and Theorem 12. From Theorem 11, if L(T) is non-radial, then L(T) has an efficient ODB f such that all the broadcasting vertices lie on a diametrical path and the end vertices of the diametrical path are not over-f-dominated. Therefore, 4 implies 2. Let g be an efficient ODB of L(T) such that all the broadcasting vertices lie on a diametrical path and the end vertices of the diametrical path are not over-g-dominated. As T is central, L(T) has two adjacent minimum eccentricity vertices on any diametrical path. So, |Vg+(L(T))|2. Then, by the second part of the proof of Theorem 12, T has a separating-set and hence, by Theorem 12, L(T) is non-radial. Therefore, 2 implies 4

We denote the class of central trees T for which L(T) is radial as TR. Now, we give some types of constructions of infinitely many trees of TR.

Theorem 15. For any central tree H of order t, if T=H(K¯n1,K¯n2,K¯n3,,K¯nt) (ni1 for i=1,2,3,,t), then TTR.

Proof. As H is central, T is also central, and by Observation 8, rad(L(T))=rad(T). As degree of each non-pendant vertex of H is at least 3 in T, T has no separating-set. So, by Theorem 12, L(T) is radial. Hence, we have the proof. 

Next, we present a construction of a bigger tree of TR from a tree of TR. Let T be a central tree and v be its central vertex. A new tree T is obtained from T by merging a leaf of K1,n (n2) to a leaf u of T such that d(u,v)rad(T)1 (Method 1), or by merging the center of K1,n (n2) to a leaf u of T such that d(u,v)rad(T) (Method 2). In Figure 4, examples of Method 1 and Method 2 are given. The tree T is obtained using Method 1 from T by merging vertices marked with in T and K1,4, and the tree T is obtained using Method 2 from T by merging vertices marked with in T and K1,4.

Theorem 16. If T is a tree obtained from a tree TTR by Method 1 or Method 2 of construction, then TTR.

Proof. The constructed tree T is always a central tree, and its radius, with respect to T, depends on the position of merging of a vertex of K1,n. Further, as the tree T is a subtree of the tree T, L(T) is a subgraph of L(T). Let v be the central vertex of T and u be the leaf of T where the merging of a vertex of K1,n has happened. We use the equivalencies of Theorem 14 to prove our claim.
Case 1: rad(T)=rad(T)
Let f be an efficient ODB of L(T). Then, the broadcast g of L(T) defined as
g(x)={f(v)if for some vVf+(L(T))dL(T)(v,x)=dL(T)(v,L(T)),0otherwise. is a DB of cost γb(L(T)). So, γb(L(T))γb(L(T)). As rad(L(T))γb(L(T))γb(L(T))=rad(L(T))=rad(T)=rad(T)=rad(L(T)), L(T) is radial. Therefore, TTR.
Case 2: rad(T)=rad(T)+1
Let f be an efficient ODB of L(T) such that every broadcasting vertex lie on a diametrical path of L(T). Then, there exists a broadcasting vertex v which f-dominates the vertices of L(K1,n). Let w be the neighbor of v on the diametrical path, such that dL(T)(w,L(K1,n))=dL(T)(v,L(K1,n))+1. Then, we define a DB g of L(T) such that g(w)=f(v)1, g(v)=0 (if vV(L(T))) and g(x)=f(x) for other vertices of L(T). Hence, γb(L(T))+1γb(L(T)). As, rad(L(T))γb(L(T))γb(L(T))+1=rad(L(T))+1=rad(T)+1=rad(T)=rad(L(T)), L(T) is radial. Therefore, TTR

2.3. Broadcast domination number of line graphs of some graphs

In this subsection, we explore the line graphs of two classes of graphs, namely, caterpillar graph and i-subdivision graph of star graph. First, we show that the line graph of caterpillar is Type I.

Theorem 17. For any caterpillar C of order at least 3, γb(L(C))=γ(L(C)).

Proof. Let f be an efficient ODB of L(C). A broadcasting vertex v f-dominates all the vertices of at most 2f(v) cliques. Then, it can be observed from Figure 5 that those vertices can be dominated by at most f(v) vertices of Sv={u1,u2,u(f(v)1),uf(v)}. The dashed circles in Figure 5 represent cliques of L(C) in its Krausz decomposition. Then, the set vVf+(L(C))Sv forms a dominating set of L(C) of size at most γb(L(C)). Hence, γ(L(C))γb(L(C)). Therefore, by Theorem 1, we have the proof.

Let the spine of a caterpillar C be v1v2v3vn. Ignoring all the leaves, except one leaf each attached to v1 and vn, we get a diametrical path of C. Then, if exists, finding a separating-set of maximum size on this diametrical path is enough to find γb(L(C)). Hence, by Theorem 17, we can determine the domination number of L(C). Now, we give a subclass of caterpillars C which satisfy the equality γb(L(C))=γb(C).

Proposition 1. Let C be a caterpillar with the spine v1v2v3vn, n is odd. If each v2i+1, i=1,2,3,,n32, is a stem, then γb(L(C))=γb(C).

Proof. As C has odd number of spine vertices, C is central. Moreover, it is not hard, due to the way C is defined, to observe that it has no separating-set. Hence, by Theorem 12, L(C) is radial. Therefore, by Theorem 14, γb(L(C))=γb(C)

Finally, we show that L(Si(K1,n)) is radial.

Theorem 18. For n3, γb(L(Si(K1,n)))=γb(Si(K1,n)).

Proof. It is easy to observe that Si(K1,n) is central, and let v be the central vertex. Let P be a diametrical path of Si(K1,n). As Si(K1,n) has at least three diametrical paths and v lies on every diametrical path, due to 1 of Definition [def], Si(K1,n) cannot have a separating-set. Hence, by Theorem 12, L(Si(K1,n)) is radial, and by Theorem 14, we have the proof. 

3. Conclusion

We conclude the article with some remarks which require further investigation.

  1. In this article, we gave a characterization of line graphs of trees to be radial, by introducing the concept of separating-set on trees. One may generalize the concept to graphs to deal with the radialness of line graphs.

  2. The next interesting problem is to characterize Type I line graphs of trees.

  3. In this article, for radial trees T, we completely settled the case when γb(T)=γb(L(T)). The same problem is open for non-radial trees.

  4. We show that the line graphs of caterpillars are 1-cap graphs, and in Proposition 1, we present a subclass of caterpillars whose line graphs are radial. So, we propose to determine the complete subclass of caterpillars whose line graphs are both radial and 1-cap.

Conflict of Interest

The authors declare no conflict of interests.

References:

  1. Whitney, H., 1992. Congruent graphs and the connectivity of graphs. Hassler Whitney Collected Papers, pp.61-79.[Google Scholor]
  2. Krausz, J., 1943. Démonstration nouvelle d’une théoreme de Whitney sur les réseaux. Matematikai és fizikai lapok, 50, pp.75-85.[Google Scholor]
  3. Erwin, D. J., 2001. Cost domination in graphs. Ph.D. Thesis, Department of Mathematics and Statistics, Western Michigan University.[Google Scholor]
  4. Dunbar, J. E., Erwin, D. J., Haynes, T. W., Hedetniemi, S. M., and Hedetniemi, S. T., 2006. Broadcasts in graphs. Discrete Applied Mathematics, 154(1), pp.59-75. [Google Scholor]
  5. Herke, S., 2009. Dominating broadcasts in graphs. Master’s Thesis, Department of Mathematics and Statistics, University of Victoria.
  6. Herke, S., and Mynhardt, C. M., 2009. Radial trees. Discrete mathematics, 309(20), pp.5950-5962. [Google Scholor]
  7. Cockayne, E. J., Herke, S., and Mynhardt, C. M., 2011. Broadcasts and domination in trees. Discrete mathematics, 311(13), pp.1235-1246. https://doi.org/10.1016/j.disc.2009.12.012[Google Scholor]
  8. Mynhardt, C. M., and Wodlinger, J., 2013. A class of trees with equal broadcast and domination numbers. Australasian Journal of Combinatorics, 56, pp.3-22.[Google Scholor]
  9. Brešar, B., and Špacapan, S. (2009). Broadcast domination of products of graphs. Ars Combinatoria, 92, pp.303-320.[Google Scholor]
  10. Hemminger, R. L., and Beineke, L. W. (1978). Line Graphs and Line Digraphs. In L. W. Beineke & R. J. Wilson (Eds.), Selected topics in graph theory (Vol. 1, pp. 271-305). Academic Press London.[Google Scholor]
  11. Heggernes, P., and Lokshtanov, D. (2006). Optimal broadcast domination in polynomial time. Discrete Mathematics, 306(24), pp.3267-3280. [Google Scholor]