Discrete Applied Mathematics 157 (2009) 3268–3280

Contents lists available at ScienceDirect

Discrete Applied Mathematics

journal homepage: www.elsevier.com/locate/dam

Locating and paired-dominating sets in graphs

John McCoy, Michael A. Henning ∗
School of Mathematical Sciences, University of KwaZulu-Natal, Pietermaritzburg, 3209, South Africa

article info abstract

Article history: In this paper, we continue the study of paired-domination in graphs introduced by Haynes
Received 3 September 2008 and Slater [T.W. Haynes, P.J. Slater, Paired-domination in graphs, Networks 32 (1998),
Received in revised form 8 May 2009 199–206]. A paired-dominating set of a graph G with no isolated vertex is a dominating
Accepted 12 June 2009
set S of vertices whose induced subgraph has a perfect matching. We consider paired-
Available online 10 July 2009
dominating sets which are also locating sets, that is distinct vertices of G are dominated by
distinct subsets of the paired-dominating set. We consider three variations of sets which
Keywords:
Differentiating-paired-dominating set
are paired-dominating and locating sets and investigate their properties.
Locating-paired-dominating set © 2009 Elsevier B.V. All rights reserved.
Metric-locating-paired-dominating set

1. Introduction

The problem of placing monitoring devices in a system such that every site in the system (including the monitoring
devices themselves) is adjacent to a monitor and every monitor is paired with a backup monitor, can be modeled by paired-
domination in graphs. Applications where it is also important that if there is a problem in the system its location can be
uniquely identified by the set of monitors, can be modeled by a combination of paired-domination and locating sets. We
consider three variations of this combination.
Let S = {v1 , v2 , . . . , vk } be a set of vertices in a connected graph G = (V , E ), and let v ∈ V . The k-vector (ordered k-tuple)
cS (v) of v with respect to S is defined by
cS (v) = (d(v, v1 ), d(v, v2 ), . . . , d(v, vk )),
where d(v, vi ) is the distance between v and vi (1 ≤ i ≤ k). The set S is called a locating set if the k-vectors cS (v), v ∈ V , are
distinct. This concept is studied in [1,8,20,22].
A set S of vertices of a graph G = (V , E ) is a dominating set (DS) of G if every vertex in V \ S is adjacent to a vertex of S,
and S is a total dominating set (TDS) of G if every vertex in V is adjacent to a vertex in S. The minimum cardinality of a TDS
is the total domination number γt (G). A TDS of cardinality γt (G) we call a γt (G)-set. Domination and its variations in graphs
are now well studied. The literature on this subject has been surveyed and detailed in the two books [9,10].
A matching in a graph G is a set of independent edges in G. A perfect matching M in G is a matching in G such that every
vertex of G is incident to an edge of M. A paired-dominating set, abbreviated PDS, of a graph G is a set S of vertices of G such
that every vertex is adjacent to some vertex in S and the subgraph G[S ] induced by S contains a perfect matching M (not
necessarily induced). Two vertices joined by an edge of M are said to be paired and are also called M-partners in S. Every graph
without isolated vertices has a PDS since the end-vertices of any maximal matching form such a set. The paired-domination
number of G, denoted by γpr (G), is the minimum cardinality of a PDS. Paired-domination was introduced by Haynes and
Slater [12,13] as a model for assigning backups to guards for security purposes, and is studied, for example, in [3–6,12,13,
17,18], and elsewhere.

∗ Corresponding author. Tel.: +27 33 2605648.

E-mail address: [email protected] (M.A. Henning).

0166-218X/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.dam.2009.06.019
 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3269

In this paper we combine the concepts of a locating set and a paired-dominating set and define three new
sets: locating-paired-dominating sets, differentiating-paired-dominating sets, and metric-locating-paired-dominating sets.
Similar concepts for total domination have been studied by Haynes et al. [11].

1.1. Notation

For notation and graph theory terminology we in general follow [9]. Specifically, let G = (V , E ) be a graph with vertex
set V of order n and edge set E. For a set S ⊆ V , the subgraph induced by S is denoted by G[S ]. If v ∈ S and w ∈ V \ S, then
the vertex w is an S-external private neighbor of v if N (w) ∩ S = {v}. The open neighborhood of vertex v ∈ V is denoted by
N (v) = {u ∈ V | uv ∈ E } while its closed neighborhood is given by N [v] = N (v) ∪ {v}. For a set S ⊆ V , N (S ) = v∈S N (v)
S
and N [S ] = N (S ) ∪ S. Hence, the set S is a DS if N [S ] = V , while S is a TDS if N (S ) = V . For disjoint subsets U and W of V ,
we let [U , W ] denote the set of all edges of G joining a vertex of U and a vertex of W . We denote the degree of a vertex v in
G by dG (v), or simply by d(v) if the graph G is clear from the context. The diameter of G, denoted diam(G), is the maximum
distance between two vertices of G. A path on n vertices is denoted by Pn , while a cycle on n vertices is denoted by Cn . The
girth g (G) is the length of a shortest cycle in G.
A leaf or end-vertex of G is a vertex of degree 1, while a support vertex of G is a vertex adjacent to a leaf. A support vertex
that is adjacent to at least two leaves we call a strong support vertex. A k-support vertex is a support vertex that is adjacent to
exactly k leaves. We call an edge of a graph incident with an end-vertex a pendant edge of the graph. A star is the tree K1,n−1
of order n ≥ 2. A subdivided star is a star where each edge is subdivided exactly once. A tree is a double star if it contains
exactly two vertices that are not leaves; if one of these vertices is adjacent to r leaves and the other to s leaves, then we
denote the double star by S (r , s). For k ≥ 1, the k-corona of a graph H is the graph of order (k + 1)|V (H )| obtained from H
by attaching a path of length k to each vertex of H so that the resulting paths are vertex disjoint. In particular, the 1-corona
of H, also called the corona of H and denoted by H ◦ K1 , is obtained from H by adding a pendant edge to each vertex of H.

1.2. Paired domination and locating sets in graphs

In this section we consider three different variations of paired-domination and locating sets in graphs. We begin by
mentioning variations of dominating and locating sets which occur in the literature.
Slater [21,22] defined a locating-dominating set in a connected graph G to be a dominating set S of G such that for every
two distinct vertices u and v in V \ S, N (u) ∩ S 6= N (v) ∩ S. This concept is studied in [4,5,19,21–23] and elsewhere.
Gimbel et al. [7] defined a set S to be a differentiating dominating set, abbreviated DDS, if S is a DS and for every pair of
distinct vertices u and v in V , N [u]∩ S 6= N [v]∩ S. The differentiating domination number, denoted γD (G), of G is the minimum
cardinality of a DDS of G.
Henning and Oellermann [16] merged the concepts of a locating set and a dominating set by defining the metric-locating-
dominating set in a connected graph G to be a set of vertices of G that is both a dominating set and a locating set in G.
Haynes et al. [11] extended these definitions to use total domination and we now consider analogous extensions to
paired-dominating sets.
Let S be a PDS in a connected graph G. We call the set S a
• locating-paired-dominating set, abbreviated LPDS, if for every pair of distinct vertices u and v in V \ S, N (u)∩ S 6= N (v)∩ S;
• differentiating-paired-dominating set, abbreviated DPDS, if for every pair of distinct vertices u and v in V , N [u] ∩ S 6=
N [v] ∩ S;
• metric-locating-paired-dominating set, abbreviated MLPDS, if S is also a locating set in G.
The locating-paired-domination number of G, denoted γpr
L
(G), is the minimum cardinality of a LPDS of G. A LPDS of
cardinality γpr
L
(G) we call a γprL (G)-set.
The differentiating-paired-domination number of G, denoted γpr
D
(G), is the minimum cardinality of a DPDS of G. A DPDS of
cardinality γpr
D
(G) we call a γprD (G)-set.
The metric-location-paired-domination number of G, denoted γpr
M
(G), is the minimum cardinality of a MLPDS in G. A MLPDS
in G of cardinality γpr M
(G) we call a γprM (G)-set.
We remark that our terminology used for paired-domination and locating in graphs is similar to that used for domination
and locating in graphs, as well as for total domination and locating in graphs. For example, a locating-total-dominating set in
a graph G is abbreviated by LTDS and the minimum cardinality of a LTDS of G is denoted by γtL (G), while a LTDS of cardinality
γtL (G) is called a γtL (G)-set, etc.

2. Locating-paired-dominating sets

We have two immediate aims in this section. Our first aim is to investigate which graphs possess a locating-paired-
dominating set. Our second aim is to establish bounds for the locating-paired-domination number of such graphs in terms
of their order.
3270 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

(a) S (3, 3). (b) K2,4 . (c) K2∗,4 .

Fig. 1. Graphs that do not possess a LPDS.

2.1. Which graphs possess a LPDS?

If a graph has a LPDS, then such a set is also a PDS of the graph implying that γpr
L
(G) ≥ γpr (G) for all such graphs G. In the
special case when G is a path, every PDS of G is also a LPDS of G. Thus the locating-paired-domination number of a path Pn
on n vertices is precisely its paired-domination number which is computed, for example, in [14].

Proposition 1 ([14]). For n ≥ 2, γpr (Pn ) = γpr (Pn ) = 2

L
n
4
.

Not every graph with no isolated vertex has a LPDS. Examples of graphs that do not possess a LPDS include a double star
S (3, 3) (Fig. 1(a)), a complete bipartite graph K2,4 (Fig. 1(b)), and the graph K2∗,4 obtained from K2,4 by adding an edge joining
the two degree-4 vertices (Fig. 1(c)).
The following result characterizes trees that possess a LPDS.

Proposition 2. A tree T has a LPDS if and only if every support vertex of T is adjacent to at most two leaves.

Proof. First we consider the necessity. Suppose that T has a LPDS S. Since S is a PDS, the set S contains every support vertex
of T and at most one leaf-neighbor of every support vertex. Hence if T has a support vertex v that is adjacent to three or
more leaves, then at least two leaf-neighbors x and y of v are not in S. But then N (x) ∩ S = {v} = N (y) ∩ S, a contradiction.
Hence, every support vertex of T is adjacent to at most two leaves. This establishes the necessity.
To prove the sufficiency, suppose that every support vertex of T is adjacent to at most two leaves. Among all maximum
matchings in T , let M be one that contains as many pendant edges as possible. If there is a support vertex v of T that is
not incident with a pendant edge that belongs to M, then we can remove from M the edge incident with v and replace it
with a pendant edge incident with v to produce a new maximum matching containing more pendant edges than does M,
a contradiction. Hence, every support vertex of T is incident with a pendant edge that belongs to M. Let S be the set of M-
matched vertices of T . The set S is thus a PDS. If there exist two distinct M-unmatched vertices u and v (and so, {u, v} ⊆ V \ S)
such that N (u) ∩ S = N (v) ∩ S, then since T has no cycles, u and v have exactly one vertex w in common. Since every M-
unmatched vertex is only adjacent with M-matched vertices, it follows that both u and v are leaf-neighbors of w . As observed
earlier, there is a leaf-neighbor of w that is M-matched, implying that the support vertex w has at least three leaf-neighbors,
a contradiction. Hence for every two distinct vertices u and v in V \ S, N (u) ∩ S 6= N (v) ∩ S. Thus the set S is a LPDS. 

We establish next a sufficient condition for a graph with minimum degree at least 2 to possess a LPDS. We call a C4 a
quadrilateral and K4 − e a diamond. If G does not contain a graph F as an induced subgraph, then we say that G is F -free. In
particular, we say a graph is quadrilateral-free if it is C4 -free and diamond-free if it is (K4 − e)-free.

Proposition 3. Every quadrilateral-free and diamond-free graph with minimum degree at least two has a LPDS.

Proof. Let G be a (C4 , K4 − e)-free graph with δ(G) ≥ 2. Let M be a maximum matching in G and let S be the set of M-matched
vertices of G. The set S is thus a PDS. Assume that there exist two distinct vertices u and v in V \S such that N (u)∩S = N (v)∩S.
Since every vertex in V \ S is an M-unmatched vertex, it is adjacent only with M-matched vertices and therefore, by the
minimum degree condition, is adjacent to at least two vertices of S. Let x and y be two neighbors of u (and v ) in S. If xy is an
edge of G, then G[{u, v, x, y}] induces a diamond K4 − e, while if xy is not an edge of G, then G[{u, v, x, y}] induces a 4-cycle
C4 . Both cases produce a contradiction. Hence for every two distinct vertices u and v in V \ S, N (u) ∩ S 6= N (v) ∩ S. Thus the
set S is a LPDS. 

As a special case of Proposition 3 we remark that every quadrilateral-free bipartite graph with minimum degree at least
two has a LPDS.
Given any graph F , we can construct a graph GF by attaching to every vertex in F a pendant edge. The graph F is an
induced subgraph of GF and V (GF ) is a LPDS of GF . We thus have the following result.

Proposition 4. There is no induced subgraph characterization of graphs that have a LPDS.

 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3271

2.2. Lower bounds

In this section, we establish lower bounds on the locating-paired-domination number of a graph in terms of its order.

Proposition 5. If G is a connected graph of order n ≥ 2 which has a LPDS, then

γprL (G) ≥ blog2 nc,
and this bound is sharp.
Proof. Let S be a γpr
L
(G)-set. Then, |S | = k for some even integer k ≥ 2. For each v ∈ V \ S, let Nv = N (v) ∩ S. Then, Nv is
a non-empty subset of the k-element set S. Since there are 2k − 1 distinct non-empty subsets of a k-element set, and since
Nu 6= Nv for every pair of distinct vertices u and v in V \ S, we have that n − k = |V \ S | ≤ 2k − 1. Hence, n ≤ 2k + k − 1.
For k ≥ 2, we have that k − 1 < 2k , and so n < 2 · 2k = 2k+1 . Thus, k > (log2 n) − 1, implying that γprL
(G) = k ≥ blog2 nc.
That this bound is sharp, may be seen as follows. Consider the bipartite graph Gk formed by taking as one partite set a set
S of k elements, where k ≥ 2 is even, and as the other partite set all the distinct non-empty subsets of S, and joining each
element of S to those subsets it is a member of. Finally, let G be obtained from Gk by adding the edges of a perfect matching
to the elements of S, and so G[S ] = 2k K2 . Then, G has order n = k + 2k − 1 and γpr
L
(G) = |S | = k = blog2 nc. 
If we fix the maximum degree of the graph, then the lower bound in Proposition 5 can be improved.

Proposition 6. If G is a graph of order n ≥ 3 and maximum degree ∆ ≥ 2 which has a LPDS, then
2n
γprL (G) ≥ ,
∆+2
and this bound is sharp for a given ∆ ≥ 2 and γpr
L
(G) ≥ ∆ − 1.
Proof. Let S be a γpr L
(G)-set of cardinality k. We count the number of edges |[S , V \ S ]| between S and V \ S. Let L1 denote
the set of all vertices in V \ S that are adjacent to exactly one vertex in S, and let L≥2 denote the set of vertices in V \ S
that are adjacent to at least two vertices in S. Let |L1 | = `, and so |L≥2 | = n − |S | − |L1 | = n − k − `. Since S is a
LPDS of G, no two vertices in L1 have the same neighbor in S, and so ` ≤ k. Since every vertex in L1 is incident with
exactly one edge in [S , V \ S ], while every vertex in L≥2 is incident with at least two edges in [S , V \ S ], we have that
|[S , V \ S ]| ≥ |L1 | + 2|L≥2 | = ` + 2(n − k − `) = 2n − 2k − ` ≥ 2n − 3k. However, every vertex in S is adjacent to at least
one other vertex in S and therefore, by the maximum degree condition, is adjacent to at most ∆ − 1 vertices in V \ S. Hence,
|[S , V \ S ]| ≤ k(∆ − 1). Consequently, 2n − 3k ≤ k(∆ − 1), or, equivalently, k ≥ 2n/(∆ + 2). The desired bound follows
since γprL
(G) = |S | = k.
To observe that the bound is sharp, suppose we are given ∆ ≥ 2 and γpr L
(G) = k ≥ ∆ − 1, k an even integer. Let G0 be
the graph constructed by attaching a pendant edge to every vertex in the graph 2k K2 . Construct G by choosing k(∆ − 2)/2
 ∆ − 2 such 2-element
2-element subsets of the set of vertices of degree two in G0 such that each of these vertices is in exactly
subsets. We can always choose such 2-element subsets since k ≥ ∆ − 1 and so k(∆ − 2)/2 ≤
k
2
. For each of these 2-
element subsets, add a vertex to G and join it to the two vertices in the 2-element subset. The vertices of degree two in G0
0

form a γpr
L
(G)-set of cardinality k and G has maximum degree ∆. The graph G has order 2k + (∆ − 2)k/2 vertices and so
k = 2n/(∆ + 2). 
We show next that if we restrict our attention to trees, then the lower bound in Proposition 5 can be significantly
improved. We present two lower bounds on the locating-paired-domination number of a tree in terms of its order and
its number of leaves and support vertices. The first lower bound we present is based on a bound due to Chellai [2]. Let T1
be the family of trees that can be obtained from k disjoint copies of P3 and P4 by first adding k − 1 edges in such a manner
that they are incident only with support vertices and the resulting graph is connected, and then subdividing each new edge
exactly once.

Proposition 7. If T is a tree of order n ≥ 2 with ` leaves and s support vertices and which has a LPDS, then
2
γprL (T ) ≥ (n + ` − s + 1),
5
with equality if and only if T ∈ T1 .
Proof. Let T be a tree of order n ≥ 2 which has a LPDS. Since every LPDS of a graph is also a LTDS of the graph, we have that
γprL (G) ≥ γtL (G) for all graphs G that have a LPDS. In particular, γprL (T ) ≥ γtL (T ). By a result due to Chellai [2], we have that
γtL (T ) ≥ 25 (n + ` − s + 1), with equality if and only if T ∈ T1 . If T1 ∈ T1 , then every minimum LTDS consists of the set of
support vertices of T1 and one leaf-neighbor of every strong support vertex of T1 . However such a set is also a LPDS of T1 ,
whence γpr
L
(T1 ) ≤ γtL (T1 ) = 25 (|V (T1 )| + `(T1 ) − s(T1 ) + 1). Consequently, γprL (T1 ) = 52 (|V (T1 )| + `(T1 ) − s(T1 ) + 1). Hence,
γprL (T ) ≥ 25 (n + ` − s + 1), with equality if and only if T ∈ T1 . 
3272 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

We remark that as a consequence of Proposition 7, we have that if T is a tree of order n ≥ 2 which has a LPDS, then
γprL (T ) ≥ 25 (n + 1), with equality if and only if T ∈ T10 , where T10 is the subfamily of T1 obtained from k disjoint copies of P4 .
The second lower bound we present is based on a bound in [11]. Let T2 (respectively, T20 ) be the family of trees T that
can be obtained from any tree T 0 by attaching at least two (respectively, exactly two) leaves to each vertex of T 0 and, if T 0 is
nontrivial, subdividing each edge of T 0 exactly once. Note that the family T20 of trees is a subset of the family T2 of trees.

Proposition 8. If T is a tree of order n ≥ 2 with ` leaves and s support vertices and which has a LPDS, then
n + 2(` − s) + 1
γprL (T ) ≥ ,
3
with equality if and only if T ∈ T20 .

Proof. By Theorem 3 in [11], we have that γtL (T ) ≥ (n + 2(` − s) + 1)/3, with equality if and only if T ∈ T2 . The desired
lower bound follows readily from the observation that γpr L
(T ) ≥ γtL (T ). If T ∈ T2 \ T20 , then there is a support vertex in T
that is adjacent to at least three leaves, and so, by Proposition 2, T has no LPDS, a contradiction. Hence, T ∈ T20 . But then
every minimum LTDS of T consists of all support vertices of T and one leaf-neighbor of every support vertex of T . However
such a set is also a LPDS of T , whence γpr
L
(T ) = γtL (T ) = (n + 2(` − s) + 1)/3. Consequently, γprL ≥ (n + 2(` − s) + 1)/3,
0
with equality if and only if T ∈ T2 . 

2.3. Upper bounds

In this section, we establish upper bounds on the locating-paired-domination number of a graph in terms of its order.

Proposition 9. If G is a connected graph of order n ≥ 2 that has a LPDS, then γpr

L
(G) ≤ n, with equality if and only if G = Kn
and n is even.

Proof. Clearly, γpr

L
(G) ≤ n. Suppose γprL (G) = n. Let S be a γprL (G)-set. Then, |S | = n and G has a perfect matching, and so
n is even. Let n = 2k and let S = ∪ki=1 {ui , vi } where ui and vi are paired in S. For i = 1, 2, . . . , k, let Si = ∪ij=1 {uj , vj } and
let Gi = G[Si ]. In particular, G = Gk . Since G is connected, we may assume, by renaming vertices if necessary, that Gi is
connected for all i = 1, 2, . . . , k.
We show that Gi = K2i for all i = 1, 2, . . . , k. We proceed by induction on i. When i = 1, G1 = K2 . This establishes the base
case. Assume, then, that 1 < i ≤ k and that Gj = K2j for all 1 ≤ j < i. We show that Gi = K2i . By induction, Gi−1 = K2(i−1) .
Since Gi is connected, we may assume for notational convenience that v1 vi ∈ E (G). Since the set S 0 = S \ {u1 , ui } is a PDS of G
but not a LPDS of G, we have that N (u1 ) ∩ S 0 = N (ui ) ∩ S 0 . Since V (Gi−1 ) \ {u1 } ⊆ N (u1 ) ∩ S 0 and vi ∈ N (ui ) ∩ S 0 , we therefore
have that u1 is adjacent to vi , and that ui is adjacent to every vertex of Gi−1 , except possibly for u1 . In particular, u1 vi ∈ E (G)
and ui v1 ∈ E (G). Similarly, considering the set S \ {u1 , vi }, we have that u1 ui ∈ E (G) and that vi is adjacent to every vertex
of Gi−1 . Hence, Gi = K2i . Thus, by induction, Gi = K2i for all i = 1, 2, . . . , k. In particular, G = Gk = K2k = Kn . 

As a consequence of Proposition 9, we have the following result.

Corollary 10. If G 6= Kn is a connected graph of order n ≥ 2 that has a LPDS, then γpr
L
(G) ≤ n − 1, and this bound is sharp even
for arbitrarily large minimum degree.

Proof. By Proposition 9, γpr L

(G) ≤ n − 1 for G 6= Kn . To see that this bound is sharp even for arbitrarily large minimum
degree, let δ ≥ 2 be a fixed even integer. Let Gδ be obtained from k disjoint copies of Kδ by adding a new vertex v and joining
v to every other vertex. Then, Gδ is a connected graph of order n = kδ + 1 with minimum degree δ . Let S be a LPDS of Gδ . If
v 6∈ S, then S must contain every other vertex of G. If v ∈ S, then S must contain every vertex from those copies of Kδ that
do not contain the partner of v in S and S must contain all except one vertex in the copy of Kδ that contain the partner of v .
In both cases, |S | = n − 1. Hence, γpr
L
(Gδ ) = n − 1. 
Recall that a subdivided star is a star where each edge is subdivided exactly once. Let K1∗,r denote the subdivided star with
r leaves. Hence, K1∗,r is the tree obtained from a star K1,r , r ≥ 1, by subdividing every edge exactly once. For a positive integer
r, let S2,r denote the tree obtained from a star K1,r +2 by subdividing r edges of the star exactly once. We next consider the
upper bound for the locating-paired-domination number of a tree.

Proposition 11. If T is a tree of order n ≥ 3 which has a LPDS, then γpr

L
(T ) ≤ n − 1, with equality if and only if T is a subdivided
star or T = S2,r for some positive integer r.

Proof. By Proposition 9, γpr

L
(T ) ≤ n − 1. To characterize the trees achieving equality in this bound, we proceed by induction
on the order n ≥ 3 of a tree T with γpr
L
(T ) = n − 1. If n = 3, then T = P3 = K1∗,1 . This establishes the base case. For the
inductive hypothesis, let n ≥ 4 and assume that if T 0 is a tree of order n0 , where 3 ≤ n0 < n, satisfying γpr
L
(T 0 ) = n0 − 1,
 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3273

a b

Fig. 2. Two trees that have no DPDS.

Table 1
γprD (Pn ) for small n.
n 4 5 6 7 8 9
γpr (Pn ) 4 4 4 6 6 8

then T 0 = K1∗,r or T 0 = S2,r for some integer r ≥ 1. Let T be a tree of order n satisfying γpr
L
(T ) = n − 1. Since γprL (T ) is always
even, we note that n is odd.
If T has two or more strong support vertices, then since every LPDS of T contains exactly one leaf neighbor of every
strong support vertex, it follows that γpr L
(T ) ≤ n − 2, a contradiction. Hence, T contains at most one strong support vertex.
By Proposition 2, if such a strong support vertex exists, it has exactly two leaf neighbors. Hence since n ≥ 4, the tree T has
at least two support vertices.
Let P be a longest path in T . If P has length 2, then T = P = P3 , a contradiction. Hence, P has length at least 3. By the
maximality of P and since T contains at most one strong support vertex, at least one of the two support vertices on the path
P has degree exactly 2 in T . Let v be such a support vertex of P and let u be its leaf neighbor. Let w be the neighbor of v
different from u, and let T 0 = T − {u, v}.
The subgraph T 0 is a tree of odd order n0 = n − 2. By Proposition 9, γpr L
(T 0 ) ≤ n0 − 1. If γprL (T 0 ) < n0 − 1, then every
γprL (T 0 )-set can be extended to a LPDS of T 0 by adding to it the pair {u, v}, implying that γprL (T ) ≤ γprL (T 0 )+ 2 < n0 + 1 = n − 1,
a contradiction. Hence, γpr L
(T 0 ) = n0 − 1. Applying the inductive hypothesis to T 0 , we have that T 0 = K1∗,r or T 0 = S2,r for
some integer r ≥ 1.
Suppose T 0 = K1∗,r . If w is a support vertex of T 0 , then γpr
L
(T ) < n − 1, a contradiction. If w is a leaf of T 0 and r ≥ 2, then
γpr (T ) < n − 1, a contradiction. If w is a leaf of T and r = 1, then T is the subdivided star K1∗,2 . Hence we may assume
L 0

that w is the central vertex of the subdivided star T 0 and that r ≥ 2. But then T is the subdivided star K1∗,r +1 . Hence we may
assume that T 0 is not a subdivided star. Thus, T 0 = S2,r for some integer r ≥ 2. If w is a leaf of T 0 or a support vertex of
T 0 different from the central vertex of T 0 , then γpr L
(T ) < n − 1, a contradiction. Hence, w must be the central vertex of T 0 ,
whence T = S2,r +1 . 

3. Differentiating-paired-dominating sets

Not all graphs possess a DPDS. It is observed in [15] that a tree containing a vertex with three or more leaf-neighbors has
no DPDS. Even if every vertex of a tree is adjacent to at most two leaves, the tree does not necessarily have a DPDS. The tree
shown in Fig. 2(a) obtained from the disjoint union of two stars K1,3 by identifying a leaf from each star into a new common
vertex is an example of such a tree that has no DPDS. The tree shown in Fig. 2(b) obtained from the disjoint union of two
stars K1,3 by adding a new vertex and joining it to a leaf from each star is another example of such a tree that has no DPDS.
A constructive characterization of those trees which do not possess a DPDS is presented in [15]. As a consequence of
this characterization, we remark that every tree of order at least 4 that has no strong support vertex has a DPDS. We now
determine the differentiating-paired-domination number of a path on at least four vertices.

Proposition 12. For n ≥ 4,

 jnk
4
 +4 if n ≡ 5 (mod 6)
γpr (Pn ) =
D j6
n k 
n mod 6

4
 +2 otherwise.
6 2

Proof. We proceed by induction on n ≥ 4. For 4 ≤ n ≤ 9, it is a simple exercise to determine the value of γpr D
(Pn ) (see
Table 1) and thereby establish the base cases.
Let n ≥ 10 and assume that the result holds for all paths of order less than n. Let P be the path v1 v2 . . . vn of order n and
let P 0 = P − {v1 , v2 , . . . , v6 }. We show that γpr
D
(P ) = γprD (P 0 ) + 4. Every γprD (P 0 )-set can be extended to a DPDS of P by adding
to it the vertices in the set {v2 , v3 , v4 , v5 }, and so γpr
D
(P ) ≤ γprD (P 0 ) + 4. To prove the reverse inequality, let S be a γprD (P )-set
and let S = S ∩ V (P ). If S = V (P ), then n ≥ 10 is even and S \ {v1 , vn } is a DPDS of P, contradicting the minimality of S.
0 0

Hence, S 6= V (P ). If v1 ∈ S, then we can replace v1 in S by the vertex vj of smallest subscript that is not in S to produce a new
γprD (T )-set. Hence we may choose S so that v1 6∈ S. With this assumption, {v2 , v3 , v4 , v5 } ⊆ S (with v2 paired with v3 , and
3274 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

with v4 paired with v5 ). Suppose v6 ∈ S. Then, v7 ∈ S with v6 and v7 paired. If v9 6∈ S, then v8 6∈ S, and we can replace the
pair {v6 , v7 } in S with the pair {v8 , v9 } to produce a new γpr
D
(T )-set. Hence we may assume that v9 ∈ S. Note that if v8 ∈ S,
then v9 is paired with v8 , while if v8 6∈ S, then v9 is paired with v10 . In any event, we can replace v6 in S by the vertex vi of
smallest subscript such that i ≥ 8 that is not in S to produce a new γpr D
(T )-set. Hence, we may assume that v6 6∈ S. But then
S 0 is a DPDS of P 0 , and so γpr
D
(P 0 ) ≤ |S 0 | = |S | − 4 = γprD (P ) − 4. Consequently, γprD (P ) = γprD (P 0 ) + 4. Hence, since P ∼
= Pn
0 ∼
and P = Pn−6 , the desired result now follows by applying the inductive hypothesis to the path P 0 . 
We establish next a sufficient condition for a graph with minimum degree at least 2 to possess a DPDS.

Proposition 13. Every graph with minimum degree at least 2 and girth at least 5 has a DPDS.
Proof. Let G be a graph with δ(G) ≥ 2 and girth g (G) ≥ 5. Let M be a maximum matching in G and let S be the set of
M-matched vertices of G. We claim that S is a DPDS of G. The set S is a PDS of G. Assume that there exist distinct vertices u
and v in G such that N [u] ∩ S = N [v] ∩ S. We first consider the case where u and v are both in S and are thus adjacent. If both
u and v are adjacent to vertices not in S, then they are only adjacent to common vertices not in S, since M is a maximum
matching. Hence, G has a triangle, contradicting g (G) ≥ 5. Thus at least one of u and v , say v , is only adjacent to vertices in
S. By the minimum degree condition, v must be adjacent to a vertex w 6= u in S. Since N [u] ∩ S = N [v] ∩ S, the vertex u is
also adjacent to w , and so the vertices u, v , and w , form a triangle in G, contradicting g (G) ≥ 5.
We next consider the case where u is in S and v in V \ S. Since N [u] ∩ S = N [v] ∩ S the vertex v is adjacent to both u and
the partner u0 of u in S. But then u, u0 , and v form a triangle, contradicting g (G) ≥ 5. Lastly, we consider the case where both
u and v are in V \ S. Since M is a maximum matching, V \ S is an independent set and so, by the minimum degree condition,
u is adjacent to distinct vertices x and y in S. But N [u] ∩ S = N [v] ∩ S and so v is also adjacent to x and y. Hence, u, x, v , and
y form a 4-cycle in G, contradicting g (G) ≥ 5. Hence for every two distinct vertices u and v in V (G), N [u] ∩ S 6= N [v] ∩ S.
Thus the set S is a DPDS of G. 
Given an arbitrary graph F with no isolated vertices, we can construct a graph GF by attaching a pendant edge to every
vertex in F . The graph F is an induced subgraph of GF and V (GF ) is a DPDS of GF . We thus have the following result.

Proposition 14. There is no induced subgraph characterization of graphs that have a DPDS.

3.1. Lower bounds

In this section we establish lower bounds on the differentiating-paired-domination number of a graph in terms of its
order.
If a graph has a DPDS, then such a set is also a LPDS of the graph, implying that γpr
L
(G) ≤ γprD (G) for all such graphs G. We
thus have the following result.

Proposition 15. If G is a connected graph of order n ≥ 4 which has a DPDS, then

γprD (G) ≥ blog2 nc,
and this bound is sharp.
Proof. By Proposition 5, we have that γpr
L
(G) ≥ blog2 nc for all graphs G of order n ≥ 3 which have a LPDS. The desired lower
bound follows readily from the observation that γpr D
(G) ≥ γprL (G) for all graphs G having a DPDS. That this bound is sharp
may be seen as follows. Consider the bipartite graph Gk formed by taking as one partite set, the set S of k elements, where
k ≥ 2 is even, and as the other partite set all the distinct non-empty subsets of S, and joining each element of S to those
subsets of which it is a member. Finally, let G be obtained from Gk by adding the edges of a perfect matching to the elements
of S and adding the minimum number of edges to ensure that every component of G[S ] contains at least four vertices. Then,
G has order n = k + 2k − 1 and γpr D
(G) = |S | = k = blog2 nc. 
If we fix the maximum degree of the graph, then the lower bound in Proposition 15 can be improved.

Proposition 16. If G is a graph of order n ≥ 4 and maximum degree ∆ ≥ 2 which has a DPDS, then
4n
γprD (G) ≥ ,
2∆ + 3
and this bound is sharp for a given ∆ ≥ 2 and γpr
D
(G) ≥ ∆ − 1.
Proof. Let S be a γprD
(G)-set, with |S | = k. We count the number of edges |[S , V \ S ]| between S and V \ S. Let L1 denote
the set of all vertices in V \ S that are adjacent to exactly one vertex in S, and let L≥2 denote the set of vertices in V \ S
that are adjacent to at least two vertices in S. Let |L1 | = `, and so |L≥2 | = n − |S | − |L1 | = n − k − `. Since S is a
DPDS of G, no two vertices in L1 have the same neighbor in S, and so ` ≤ k. Since every vertex in L1 is incident with
exactly one edge in [S , V \ S ], while every vertex in L≥2 is incident with at least two edges in [S , V \ S ], we have that
 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3275

|[S , V \ S ]| ≥ |L1 | + 2|L≥2 | = ` + 2(n − k − `) = 2n − 2k − ` ≥ 2n − 3k. Let S1 be the vertices in S of degree 1 in G[S ],
and let S2 = S \ S1 . Since G[S ] contains a perfect matching, all the vertices in S are adjacent to at least one other vertex in
S. Since G[S ] contains no K2 -components, at least half of the vertices in S are adjacent to at least two vertices in S and thus
|S1 | ≤ k/2. Hence,
X X
2|E (G[S ])| = dG[S ] (v) + dG[S ] (v)
v∈S1 v∈S2
≥ |S1 | + 2|S2 |
= 2|S | − |S1 |
≥ 2k − k/2
= 3k/2.

Hence, by the maximum degree condition, |[S , V \S ]| ≤ k∆−2|E (G[S ])| ≤ k∆−3k/2. As shown earlier, 2n−3k ≤ |[S , V \S ]|.
Consequently, 2n − 3k ≤ k∆ − 3k/2, or equivalently, k ≥ 4n/(2∆ + 3). Hence, γprD
(G) ≥ 4n/(2∆ + 3).
We observe that the bound is sharp as follows. Suppose that we are given ∆ ≥ 2 and γpr D
(G) = k ≥ ∆ − 1,
k an even integer. Let G0 consist of k disjoint copies of a corona P4 ◦ K1 . Construct the graph G as follows. Choose
((∆ − 2)(k/2) + (∆ − 3)(k/2))/2 = (2k∆ − 5k)/4 2-element subsets of vertices with degree 2 or 3 in G0 such that
every vertex of degree 2 in G0 is in ∆ − 2 such subsets and every vertex of degree 3 in G0 is in ∆ − 3 
such subsets. It is always
possible to choose such subsets since k ≥ ∆ − 1 with k an even integer, and so (2k∆ − 5k)/4 ≤
k
2
. For each 2-element
subset add a vertex to G and join it to the two elements in the subset. The maximum degree of G is ∆ and the order of G is
0

2k + (2k∆ − 5k)/4, and so k = 4n/(2∆ + 3). 

We next show that if we restrict our attention to trees, then the lower bound in Proposition 15 can be significantly
improved. The proof of the following lower bound is based on a proof of Theorem 2 (a lower bound on the locating-total-
domination number of a tree in terms of its order) in [11]. Let T3 be the family of trees that can be obtained from k disjoint
copies of a corona P4 ◦ K1 by first adding k − 1 edges in such a manner that they are incident only with support vertices and
the resulting graph is connected, and then subdividing each new edge with a single vertex. Note that if T ∈ T3 , then T has a
DPDS.

Proposition 17. If T is a tree of order n ≥ 4 which has a DPDS, then

4
γprD (T ) ≥ (n + 1),
9
with equality if and only if T ∈ T3 .

Proof. Let T be a tree of order n ≥ 4 which has a DPDS. Let S be a γpr D

(T )-set and let T1 , T2 , . . . , Tk be the components of T [S ].
Since every component of T [S ] has at least four vertices, |S | ≥ 4k, and so k ≤ |S |/4. Let P be the set of all S-external private
neighbors of vertices in S. Since every vertex of S has at most one S-external private neighbor, |P | ≤ |S |. Let R = V \ (S ∪ P ),
and let |R| = r. Let K be a set of k vertices corresponding to the components of T [S ]. Let F be a forest of order k + r with
V (F ) = K ∪ R where two vertices in R are adjacent in F if and only if they are adjacent in T , and where u ∈ K is joined to
v ∈ R if and only if vertex v is adjacent in T to a vertex in the component of T [S ] corresponding to u. Since each vertex of
R is adjacent to at least two vertices in S, we have that |E (F )| ≥ 2|R| = 2r, and so k + r = |V (F )| ≥ |E (F )| + 1 ≥ 2r + 1.
Thus, r ≤ k − 1. Hence, n − |S | = |V \ S | = |P | + |R| ≤ |S | + (k − 1) ≤ 5|S |/4 − 1, and so n ≤ 9|S |/4 − 1. Consequently,
γprD (T ) = |S | ≥ 4(n + 1)/9.
This bound is sharp if and only if equality is achieved in each of the above equalities. In particular, k = |S |/4 implying
that each component of T [S ] is a P4 . Also, r = k − 1. It follows that R is an independent set and that T [R ∪ S ] is a tree in
which each vertex in R has degree 2. Moreover, |P | = |S |, and so, since T [R ∪ S ] is a tree, T [P ∪ S ] is the union of k disjoint
copies of a corona P4 ◦ K1 where each vertex of P is a leaf of T . Hence, T ∈ T3 . 

Next we present a lower bound on the differentiating-paired domination number of a tree in terms of its order and the
number of support vertices. Let T4 be the family of trees T that can be obtained as follows. Let F be the forest consisting
of disjoint copies of P4 . Let T1 be obtained from F by adding edges between vertices in distinct components of F such that
the resulting graph is a tree and then subdividing each added edge twice. Let T be obtained from T1 by adding exactly one
pendant edge to any number of vertices in T1 such that each vertex in V (F ) has degree at least 2 in T .

Proposition 18. If T is a tree of order n ≥ 4 with s support vertices and which has a DPDS, then
2(n − s + 2)
γprD (T ) ≥ ,
3
with equality if and only if T ∈ T4 .
3276 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

Proof. Let S be a γpr D

(T )-set with |S | = k. Let L be the set of leaves in T with |L| = ` and let s be the number of support
vertices. Since T has a DPDS every support vertex of T is adjacent to at most two leaf neighbors. There are thus ` − s support
vertices adjacent to exactly two leaves. Every support vertex of T is in S and every 2-support vertex of T has exactly one leaf
neighbor in S. Consequently, there are at least ` − s leaves in S and so at most ` − (` − s) = s leaves in V \ S.
Let Q = V \(S ∪L) with |Q | = q. For each component Ti , 1 ≤ i ≤ r, of T [S ], we associate a vertex ci and let R = {c1 , . . . , cr }.
Let T 0 be the tree with vertex set R ∪ Q where vertices in Q are adjacent in T 0 if and only if they are adjacent in T and where a
vertex v ∈ Q is adjacent to ci if and only if v is adjacent in T to a vertex in Ti . Let Q1 be the set of vertices of Q that are adjacent
to exactly one vertex of R in T 0 , and let Q2 = Q \ Q1 . Let |Q1 | = q1 and |Q2 | = q2 . Since T 0 is a tree, |E (T 0 )| = r + q − 1. Since
every vertex in Q has degree at least 2, |E (T 0 [Q ])| ≥ q1 /2. Thus, r + q − 1 ≥ q1 + 2q2 + |E (T 0 [Q ])| ≥ 3q1 /2 + 2q2 ≥ 3q/2,
or equivalently, q ≤ 2(r − 1). Hence there are at most 2(r − 1) non-leaf vertices in V \ S.
Every component of T [S ] contains at least four vertices since S is a DPDS, and so r ≤ k/4. Consequently, n − k = |V \ S | ≤
s + 2(k/4 − 1) = s + k/2 − 2, or, equivalently, k ≥ 2(n − s + 2)/3. Hence, γpr D
(T ) ≥ 2(n − s + 2)/3. The trees that achieve
equality in the bound achieve equality in each of the above inequalities. In particular, T [S ] consists of k/4 disjoint copies of
P4 , and there are exactly 2(r − 1) non-leaf vertices in V \ S and exactly s leaves in V \ S. Thus, γpr D
(T ) = 2(n − s + 2)/3 if
and only if T ∈ T4 . 

3.2. Upper bounds

In this section we establish upper bounds on the differentiating-paired-domination number of a tree in terms of its order.

Proposition 19. If T 6= P4 is a tree of order n ≥ 4 which has a DPDS, then γpr

D
(T ) ≤ n − 1.
Proof. Assume, to the contrary, that T 6= P4 is a tree of order n ≥ 4 with γpr D
(T ) = n. Then, S = V (T ) is the unique γprD (T )-set.
Hence, every vertex in T is paired in S. In particular, every leaf is paired with the support vertex adjacent to it, implying that
T has no strong support vertex. Let u–v be a longest path in T . Necessarily, u and v are leaves of T . Let u0 be the neighbor of u,
and let v 0 be the neighbor of v . Let x be the vertex on the path u–v such that x is a support vertex on the u–v path, different
from u0 , at minimum distance from u (possibly, x = v 0 ). Let x0 be the leaf neighbor of x. As observed earlier, x and x0 are paired
in S. But then the set S 0 = S \ {u, x0 } is a DPDS of T satisfying |S 0 | < |S |, contradicting the fact that S is a γprD
(T )-set. 
We remark that, in general, the corona Kk ◦ K1 of a complete graph Kk , where k ≥ 2, is a graph of order n = 2k that
satisfies γpr
D
(G) = n. It remains an open problem to characterize the graphs G of order n ≥ 4 which have a DPDS and satisfy
γprD (G) = n.
We next characterize the trees T of order n ≥ 5 that achieve equality in the upper bound of Proposition 19. For this
purpose, we introduce three families of trees.
For integers p, q ≥ 0 with p + q ≥ 1, let Tp,q denote the tree obtained from a star K1,p+q by subdividing p of its edges
exactly once and the remaining q edges three times. Note that T1,0 = P3 while T2,0 = T0,1 = P5 . If p + q = 1, then we select
one of the two leaves of Tp,q and call it the pivot vertex of Tp,q , while if p + q ≥ 2, then we call the central vertex of the star
K1,p+q used to construct Tp,q the pivot vertex of Tp,q .
For p + q ≥ 1, let Tp,q,1 denote the tree obtained from Tp,q by adding two pendant edges to its pivot vertex. We define
the pivot vertex of Tp,q,1 to be the pivot vertex of Tp,q used to construct Tp,q,1 .
For p + q ≥ 0, let Tp,q,2 denote the tree obtained from Tp+1,q by adding two pendant edges to a leaf of Tp+1,q at distance 2
from its pivot vertex. We define the pivot vertex of Tp,q,2 to be the pivot vertex of Tp+1,q used to construct Tp,q,2 . Note that
T1,0,1 = T0,0,2 is the double star S (2, 1) which is obtained from a star K1,3 by subdividing one edge once.
Let T5 be the family of trees T such that T = Tp,q for some integers p, q ≥ 0 where p + q ≥ 1 and p ≥ 2 if q = 0, or
T = Tp,q,1 for some integers p, q ≥ 0 and p + q ≥ 1, or T = Tp,q,2 for some integers p, q ≥ 0.

Proposition 20. If T is a tree of order n ≥ 5 which has a DPDS, then γpr

D
(T ) = n − 1, with equality if and only if T ∈ T5 .
Proof. By Proposition 19, γpr
D
(T ) ≤ n − 1 for trees of order n ≥ 5 which have a DPDS. To characterize the trees achieving
equality in this bound we proceed by induction on the order n ≥ 5 of a tree T which has a DPDS with γpr
D
(T ) = n − 1.
If n = 5 and γpr D
(T ) = 4, then T = P5 or T = S (2, 1). In both cases, T ∈ T5 . This establishes the base case. For the
inductive hypothesis, let n ≥ 6 and assume that if T 0 is a tree of order n0 , where 5 ≤ n0 < n, which has a DPDS and satisfies
γprD (T 0 ) = n0 − 1, then T 0 ∈ T5 . Let T be a tree of order n ≥ 6 which has a DPDS and satisfies γprD (T ) = n − 1. Since γprD (T ) is
always even, we note that n is odd, and so n ≥ 7.
If T has two or more strong support vertices, then since every DPDS of T contains exactly one leaf neighbor of every
strong support vertex, it follows that γpr D
(T ) ≤ n − 2, a contradiction. Hence, T contains at most one strong support vertex.
Since T has a DPDS, such a strong support vertex of T has exactly two leaf neighbors. Since n ≥ 6, the tree T has at least two
support vertices. Further, since T has at most one strong support vertex, diam(T ) ≥ 4.
Let P be a longest path in T . Then, |P | ≥ 5 and at least one of the two support vertices on the path P has degree exactly 2
in T . Let v be such a support vertex on P and let u be its leaf neighbor. Let w be the neighbor of v different from u and let x
be the neighbor of w on P different from v . We now consider two cases, depending on the degree of the vertex w .
 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3277

Case 1. dT (w) > 2. Let v 0 be a neighbor of w which is not on P. Since P is a longest path in T , the vertex v 0 is either a
leaf or a support vertex with w as its only non-leaf neighbor. Let T 0 = T − {u, v}. The subgraph T 0 is a tree of odd order
n0 = n − 2 ≥ 5, and so, by Proposition 19, γprD
(T 0 ) ≤ n0 − 1. Every DPDS of T 0 necessarily contains the vertex w. Hence if
γprD (T 0 ) < n0 − 1, then every γprD (T 0 )-set can be extended to a DPDS of T 0 by adding to it the vertices u and v , implying that
γprD (T ) ≤ γprD (T 0 ) + 2 < n0 + 1 = n − 1, a contradiction. Thus, γprD (T 0 ) = n0 − 1. Applying the inductive hypothesis to T 0 , we
have that T 0 ∈ T5 . Note that dT 0 (w) ≥ 2.
Suppose T 0 = Tp,q . If w is not the pivot vertex of T 0 , then γpr D
(T ) ≤ n − 2, a contradiction. Hence, w is the pivot vertex of
0
T , and so T = Tp+1,q ∈ T5 .
Suppose T 0 = Tp,q,1 . If w is the pivot vertex of T 0 , then T = Tp+1,q,1 . Suppose that w is not the pivot vertex of T 0 . If
T 0 = T0,1,1 and w is the vertex at distance 2 from the pivot vertex, then T = T2,0,2 . If T 0 = T0,1,1 and w is not the vertex
at distance 2 from the pivot vertex, or if T 0 6= T0,1,1 , then γpr D
(T ) ≤ n − 2, a contradiction. Hence if T 0 = Tp,q,1 , then either
T = Tp+1,q,1 ∈ T5 or T = T2,0,2 ∈ T5 .
Suppose T 0 = Tp,q,2 . If w is the pivot vertex of T 0 , then T = Tp+1,q,2 . Suppose that w is not the pivot vertex of T 0 . If
T = T1,0,2 and w is the strong support vertex of T 0 (which is at distance 2 from the pivot vertex), then T = T1,1,1 . If
0

T 0 = T1,0,2 and w is not the strong support vertex of T 0 , or if T 0 6= T1,0,2 , then γpr D
(T ) ≤ n − 2, a contradiction. Hence if
0
T = Tp,q,2 , then either T = Tp+1,q,2 ∈ T5 or T = T1,1,1 ∈ T5 .
Hence if dT (w) > 2, then T ∈ T5 .
Case 2. dT (w) = 2. Assume that dT (x) > 2. Let T 0 = T − {u, v, w}. Then T 0 is a tree of order n0 = n − 3 ≥ 3. If n0 ∈ {3, 4},
then γpr D
(T ) = 4 ≤ n − 2, a contradiction. Hence, n0 ≥ 5, and so, by Proposition 19, γprD (T 0 ) ≤ n0 − 1. Let w 0 be a neighbor
of x which is not on P. Since P is a longest path in T , we note that every child of w 0 , if any, is a leaf or a support vertex.
Further if w 0 has a child that is a support vertex of degree 2, then we may assume that dT (w 0 ) = 2, for otherwise we are
in Case 1, implying that the vertex x must belong to every DPDS of T 0 . If w 0 does not have a child that is a support vertex
of degree 2, then since T contains at most one strong support vertex, there is a γpr D
(T 0 )-set that contains x. This implies that
every γpr (T )-set can be extended to a DPDS of T by adding to it the pair {v, w}. Hence, γpr
D 0 D
(T ) ≤ γprD (T 0 )+ 2 ≤ n0 + 1 = n − 2,
a contradiction. Hence, dT (x) = 2. Let y be the neighbor of x different from w .
If n = 5, then T = P5 ∈ T5 . Hence, since n is odd, we may assume that n ≥ 7. Let T 0 = T − {u, v, w, x}. Then T 0 is a
tree of odd order n0 = n − 4 ≥ 3. If n0 = 3, then T = P7 = T1,1 or T = T0,1,1 (note that T0,1,1 is obtained from a star K1,3
by subdividing one edge three times). In both cases, T ∈ T5 . Hence we may assume that n0 ≥ 5, and so, by Proposition 19,
γprD (T 0 ) ≤ n0 − 1. If γprD (T 0 ) < n0 − 1, then every γprD (T )-set can be extended to a DPDS of T by adding to it the vertices in the
set {u, v, w, x}, implying that γpr D
(T ) ≤ γprD (T 0 ) + 4 < n0 + 3 = n − 1, a contradiction. Thus, γprD (T 0 ) = n0 − 1. Applying the
inductive hypothesis to T , we have that T 0 ∈ T5 .
0

Suppose T 0 = Tp,q . If y is the pivot vertex of T 0 , then T = Tp,q+1 . Suppose that y is not the pivot vertex of T 0 . If T 0 = T2,0 = P5
and y is a leaf of T 0 , then T = P9 = T0,2 . If T 0 = T2,0 and y is not a leaf of T 0 , then γpr D
(T ) ≤ n − 2, a contradiction. If
0 0
T = T1,1 = P7 and y is the vertex at distance 2 from a leaf in T that is different from the pivot vertex, then T = T1,2 . If
T 0 = T1,1 and y is not a vertex at distance 2 from a leaf in T 0 , then γpr D
(T ) ≤ n − 2, a contradiction. If T 0 6= T2,0 and if T 0 6= T1,1
(and y is not the pivot vertex of T 0 ), then γpr D
(T ) ≤ n − 2, a contradiction. Hence if T 0 = Tp,q , then either T = Tp,q+1 ∈ T5 or
T = T1,2 ∈ T5 .
Suppose T 0 = Tp,q,1 . If y is the pivot vertex of T 0 , then T = Tp,q+1,1 . Suppose that y is not the pivot vertex of T 0 . If
T 0 = T1,0,1 and y is a leaf of T 0 but not a leaf of a strong support vertex, then T = T0,1,2 . If T 0 = T1,0,1 and y is not a leaf
of T 0 , then γpr D
(T ) ≤ n − 2, a contradiction. If T 0 = T0,1,1 and y is the vertex at distance 2 from the pivot vertex in T 0 , then
T = T1,1,2 . If T 0 = T0,1,1 and y is not the vertex at distance 2 from the pivot vertex in T 0 , then γpr D
(T ) ≤ n − 2, a contradiction.
If T 6= T1,0,1 and if T 6= T0,1,1 (and y is not the pivot vertex of T ), then γpr (T ) ≤ n − 2, a contradiction. Hence if T 0 = Tp,q,1 ,
0 0 0 D

then either T = Tp,q+1,1 ∈ T5 or T ∈ {T0,1,2 , T1,1,2 } ⊂ T5 .

Suppose T 0 = Tp,q,2 . If y is the pivot vertex of T 0 , then T = Tp,q+1,2 . Suppose that y is not the pivot vertex of T 0 . If
T = T1,0,2 and y is the strong support vertex of T 0 , then T = T0,2,1 . If T 0 = T1,0,2 and y is not the strong support vertex of T 0 ,
0

then γpr D
(T ) ≤ n − 2, a contradiction. If T 0 6= T1,0,2 (and y is not the pivot vertex of T 0 ), then γprD (T ) ≤ n − 2, a contradiction.
Hence if T 0 = Tp,q,2 , then either T = Tp,q+1,2 ∈ T5 or T = T0,2,1 ∈ T5 .
Hence if dT (w) > 2, then T ∈ T5 . This completes the proof of Case 2 and Proposition 20. 

4. Metric-locating-paired-dominating sets

We begin by investigating which graphs possess a MLPD.

Proposition 21. Every LPDS of a graph is a MLPDS of the graph.

Proof. Suppose G has a LPDS and let S be such a set. Assume S is not a MLPDS of G. That is, assume that there exist vertices
u and v in G such that cS (u) = cS (v). We must have that u, v ∈ V \ S. Every vertex in N (u) ∩ S is at distance one from u and
so, since cS (u) = cS (v), is at a distance one from v . Hence every vertex in N (u) ∩ S is in N (v) ∩ S and similarly every vertex
in N (v) ∩ S is in N (u) ∩ S. Thus N (u) ∩ S = N (v) ∩ S, a contradiction since S is a LPDS. Hence, S is a MLPDS of G. 
3278 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

Proposition 22. If G has a MLPDS, then G has a LPDS.

Proof. Suppose G has a MLPDS. Amongst all MLPDSs of G, let S be one of maximum cardinality. Assume S is not a LPDS.
That is, assume that there exist distinct vertices u and v in V \ S such that N (u) ∩ S = N (v) ∩ S. Since cS (u) 6= cS (v), we
have that N (u) 6= N (v). Suppose, without loss of generality, that w ∈ N (u) but w 6∈ N (v). The vertex w is not in S since
N (u)∩ S = N (v)∩ S. But then S ∪{u, w} is a MLPDS of G of larger cardinality than S, a contradiction. Thus, S is a LPDS of G. 
We have the following corollaries as an immediate consequence of Propositions 2, 3, 21 and 22.

Corollary 23. A graph G has a MLPDS if and only if G has a LPDS.

Corollary 24. A tree T has a MLPDS if and only if every support vertex has at most two leaf neighbors.

Corollary 25. Every quadrilateral-free and diamond-free graph with minimum degree at least two has a MLPDS.
Since every LPDS of a graph is a MLPDS of the graph, we have that γpr
M
(G) ≤ γprL (G) for every graph G. As an immediate
consequence of Propositions 9 and 11, and Corollaries 10 and 25, we have the following results on the upper bounds of the
metric-locating-paired-domination number of a graph.

Corollary 26. If G is a connected graph of order n ≥ 2 which has a MLPDS, then γpr
M
(G) ≤ n, with equality if and only if G = Kn
and n is even.

Corollary 27. If G 6= Kn is a connected graph of order n ≥ 2 which has a MLPDS, then γpr
M
(G) ≤ n − 1, and this bound is sharp
even for arbitrarily large minimum degree.

Corollary 28. If T is a tree of order n ≥ 3 which has a MLPDS, then γpr

M
(T ) ≤ n − 1, with equality if and only if T is a subdivided
star or T is obtained from a subdivided star by adding two pendant edges to the central vertex.
Next, we give a lower bound on the metric-locating-paired-domination number of a tree in terms of its order. Let T6 be
the family of trees that can be obtained from q disjoint copies of P4 by adding q − 1 edges between non-leaf vertices such
that the resulting graph is a tree, and then subdividing each new edge twice.

Proposition 29. If T is a tree of order n ≥ 2 which has a MLPDS, then

n+2
γprM (T ) ≥ ,
3
with equality if and only if T ∈ T6 .
Proof. Let S be a γpr M
(T )-set with |S | = k. Since T has a MLPDS, every support vertex of T is adjacent to at most two leaf
neighbors. Every support vertex of T is in S and every 2-support vertex of T has exactly one leaf neighbor in S. Thus every
vertex in S has at most one leaf neighbor in V \ S. Consequently, there are at most k leaves in V \ S.
Let Q be the set of vertices in V \ S that have degree at least 2 in T , and let |Q | = q. For each component Ti , 1 ≤ i ≤ r,
of T [S ], we associate a vertex ci and let R = {c1 , . . . , cr }. Let T 0 be the tree with vertex set R ∪ Q where vertices in Q are
adjacent in T 0 if and only if they are adjacent in T and where a vertex v ∈ Q is adjacent to ci if and only if v is adjacent in
T to a vertex in Ti . Let Q1 be the set of vertices of Q that are adjacent to exactly one vertex of R in T 0 , and let Q2 = Q \ Q1 .
Let |Q1 | = q1 and |Q2 | = q2 . Since T 0 is a tree, |E (T 0 )| = r + q − 1. Since every vertex in Q has degree at least 2 in T ,
|E (T 0 [Q ])| ≥ q1 /2. Thus, r + q − 1 ≥ q1 + 2q2 + |E (T 0 [Q ])| ≥ 3q1 /2 + 2q2 ≥ 3q/2, or equivalently, q ≤ 2(r − 1). Hence
there are at most 2(r − 1) non-leaf vertices of T in V \ S. Since S is a MLPDS, every component of T [S ] contains at least two
vertices, and so r ≤ k/2. Consequently, n − k = |V \ S | ≤ k + 2(k/2 − 1) = 2k − 2, or, equivalently, k ≥ (n + 2)/3. Hence,
γprM (T ) ≥ (n + 2)/3.
The trees that achieve equality in this upper bound achieve equality in each of the above inequalities. In particular, T [S ]
consists of k/2 disjoint copies of K2 and there are exactly k leaves of T in V \ S. Further, Q = Q1 and |Q1 | = k − 2, and
T [Q ] = (k − 2)K2 . Thus, γpr D
(T ) = (n + 2)/3 if and only T ∈ T6 . 
We next present a lower bound on the metric-locating-paired-domination number of a tree in terms of its order and
number of support vertices. Let T7 be the family of trees T that can be obtained as follows. Let F be the forest consisting
of disjoint copies of K2 . Let T1 be obtained from F by adding edges between vertices in distinct components of F such that
the resulting graph is a tree and then subdividing each added edge twice. Let T be obtained from T1 by adding exactly one
pendant edge to any number of vertices in T1 such that no vertex of F is the only leaf neighbor of a support vertex in T .

Proposition 30. If T is a tree of order n ≥ 3 which has a MLPDS, then

n−s+2
γprM (T ) ≥ ,
2
with equality if and only if T ∈ T7 .
 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280 3279

Proof. The proof is identical to that of Proposition 18, except that there are at least two vertices in each component of the
subgraph of T induced by a γprM
(T )-set. 
Every MLPDS of a graph is also a PDS of the graph, and so γpr (G) ≤ γpr
M
(G) for every graph G. The following result
characterizes the trees T for which γpr
M
(T ) = γpr (T ).

Proposition 31. Let T = (V , E ) be a nontrivial tree which possesses a MLPDS and let S be the set of 2-support vertices of T . Let
K be the set of non-leaf neighbors of vertices in S. Then γpr
M
(T ) = γpr (T ) if and only if T satisfies the following two conditions:
(i) The set S is an independent set in T .
(ii) N [S ] = V or there exists A ⊆ K such that F = T − (N [S ] − A) is a forest with no isolates and with γpr (T ) = γpr (F ) + 2|S |.

Proof. First we consider the necessity. Let T be a tree which possesses a MLPDS and suppose that γpr (T ) = γpr
M
(T ). Since T
possesses a MLPDS it contains no k-support vertex for k ≥ 3, by Corollary 24. Let D be a γpr M
(T )-set. By assumption, the set
D is also a γpr (T )-set.
Let v ∈ S and let v 0 be a leaf neighbor of v in D. Suppose v is adjacent to a vertex u ∈ S. Then there exists a leaf neighbor
u0 of u in D since D is a MLPDS. But, D \ {u0 , v 0 } is a PDS of T , a contradiction. Hence, v is adjacent to no other vertex in S, and
so S is an independent set in T . Thus, (i) holds.
Suppose that N [S ] 6= V . Let A = D ∩ K and let F = T − (N [S ] \ A). Then, DF = D ∩ V (F ) is a PDS of F and |DF | = |D| − 2|S |.
It follows that the forest F has no isolated vertex and γpr (F ) ≤ |DF |. Any PDS of F can be extended to a PDS of T by adding to
it the set S and one leaf neighbor adjacent to every vertex in S, and so |DF |+ 2|S | = |D| = γpr M
(T ) = γpr (T ) ≤ γpr (F )+ 2|S | ≤
|DF | + 2|S |. Consequently, we must have equality throughout this inequality chain. In particular, γpr (T ) = γpr (F ) + 2|S | and
(ii) holds.
Next we consider the sufficiency. Suppose that T is a tree which possesses a MLPDS and that satisfies conditions (i) and
(ii). Since T possesses a MLPDS it contains no k-support vertex for k ≥ 3, by Corollary 24.
We show first that if S = ∅, then γpr (T ) = γpr M
(T ). Assume that S = ∅. Thus the tree T has no strong support vertex. Let
D be a PDS of T and let u, v ∈ V \ D. If N (u) ∩ D 6= N (v) ∩ D, then cD (u) 6= cD (v). Hence, suppose that N (u) ∩ D = N (v) ∩ D.
Since T is a tree, we may suppose that N (u) ∩ D = N (v) ∩ D = {w}. Since w is not a strong support vertex, at least one of u
and v cannot be a leaf. We may assume that dT (v) ≥ 2. Let x ∈ N (v) \ {w}. Since N (v) ∩ D = {w}, we note that x 6∈ D. Thus, x
is adjacent to a vertex y ∈ D. Since T is a tree, w 6= y. Therefore, d(v, y) = 2, while d(u, y) = 4, implying that cD (u) 6= cD (v).
Hence, D is a MLPDS of T , and so γpr M
(T ) ≤ γpr (T ). Consequently, γpr (T ) = γprM (T ).
We show next that if S 6= ∅ and N [S ] = V , then γpr (T ) = γpr
M
(T ). Assume that S 6= ∅ and N [S ] = V . Then adding
one leaf neighbor adjacent to every vertex in S to the set S produces a MLPDS of T , and so γpr
M
(T ) ≤ 2|S |. By (i), the set S is
independent in T , and so γpr (T ) ≥ 2|S |. Thus, γpr
M
(T ) ≤ γpr (T ). Consequently, γpr (T ) = γprM (T ).
We show finally that if S 6= ∅ and N [S ] 6= V , then γpr (T ) = γprM
(T ). Assume that S 6= ∅ and N [S ] 6= V . By (ii), there exists
A ⊆ K such that F = T − (N [S ] \ A) is a forest with no isolates and with γpr (T ) = γpr (F ) + 2|S |. Let DF be a γpr (F )-set. Let
D be the set obtained from DF by adding to it the set S and one leaf neighbor adjacent to every vertex in S. Then, D is a PDS
of T of cardinality |DF | + 2|S | = γpr (F ) + 2|S | = γpr (T ). Thus, D is a γpr (T )-set. Suppose D is not a MLPDS of T . Then, there
exist u, v ∈ V \ D such that cD (u) = cD (v). Consequently, N (u) ∩ D = N (v) ∩ D, and since T is a tree we may assume that
N (u) ∩ D = N (v) ∩ D = {w} for some w ∈ D. If w is not a strong support vertex, then at least one of u and v is not a leaf. If
w is a strong support vertex, then one of its leaf neighbors is in D, and so at least one of u and v is not a leaf. We may thus
assume that dT (v) ≥ 2. Let x ∈ N (v) \ {w}. Since N (v) ∩ D = {w}, we note that x 6∈ D. Thus, x is adjacent to a vertex y ∈ D.
Since T is a tree, w 6= y. Therefore, d(v, y) = 2, while d(u, y) = 4, implying that cD (u) 6= cD (v), a contradiction. Hence, D is
a MLPDS of T , and so γpr M
(T ) ≤ γpr (T ). Consequently, γpr (T ) = γprM (T ). 
As an immediate consequence of Proposition 31, and its proof, we have the following result.

Corollary 32. If a nontrivial tree T contains no strong support vertex, then every PDS of T is a MLPDS of T . In particular,
γprM (T ) = γpr (T ).
We next establish a relationship between the paired-domination number and the metric-location-paired-domination
number of a tree in terms of the number of leaves and the number of support vertices of the tree.

Proposition 33. If T is a tree with ` leaves and s support vertices and which possesses a MLPDS, then
γpr (T ) ≤ γprM (T ) ≤ γpr (T ) + 2(` − s)
and this bound is sharp.
Proof. The lower bound is immediate since every MLPDS of a graph is a PDS of the graph. To prove the upper bound, let T
be a tree with ` leaves and s support vertices which possesses a MLPDS. Since T has a MLPDS, every support vertex of T is
adjacent to at most two leaves. If T contains no 2-support vertex, then by Corollary 32, γpr
M
(T ) = γpr (T ) and since ` = s
we have γpr
M
(T ) = γpr (T ) + 2(` − s). Hence we may assume that T has at least one 2-support vertex, since otherwise the
3280 J. McCoy, M.A. Henning / Discrete Applied Mathematics 157 (2009) 3268–3280

result follows. Let T 0 be the tree obtained from T by deleting exactly one leaf-neighbor of every 2-support vertex of T . Then,
γpr (T 0 ) = γpr (T ). Let S 0 be a γpr (T 0 )-set. Since T 0 is a tree with no strong support vertex, by Corollary 32, the set S 0 is a MLPDS
of T 0 . Let S be a MLPDS of T constructed from S 0 as follows. Consider each of the (` − s) 2-support vertices of T in turn. Let
v be a 2-support of T . If v has a leaf in T 0 which is in S 0 , then we consider the next 2-support of T . If v is paired in S 0 with a
non-leaf neighbor u and epn(u, S 0 ) = ∅, then we add to S 0 a leaf of v and remove u from S 0 . If v is paired in S 0 with a non-leaf
neighbor u and epn(u, S 0 ) 6= ∅, then we let u0 ∈ epn(u, S 0 ) and add both u0 and a leaf-neighbor of v to S 0 . The resulting set S
so constructed is a MLPDS of T . Thus, γpr M
(T ) ≤ |S | ≤ |S 0 | + 2(` − s) = γpr (T ) + 2(` − s). This establishes the upper bound.
Equality in the lower bound is characterized in Proposition 31. Equality in the upper bound is achieved if T has no strong
support vertices. There are also infinite families of trees that contain strong support vertices and achieve equality in the
upper bound. For example, such a family of trees T can be constructed as follows. Begin with any path of order 3k + 1, k ≥ 0,
with vertices labeled from v1 through v3k+1 . For each j = 0, . . . , k, add two pendant edges and a path of length 2 to the
vertex v3j+1 . Then, γpr (T ) = 2(k + 1), γpr M
(T ) = 4(k + 1), ` = 3(k + 1), and s = 2(k + 1). Thus, γpr (T ) + 2(` − s) = γprM (T ),
and so the tree T achieves equality in the upper bound. 
We next establish upper and lower bounds on γpr
M
(T ), for a tree T possessing a LPDS, in terms of γprL (T ).

Proposition 34. For any tree T having a LPDS,

γprM (T ) ≤ γprL (T ) ≤ 2γprM (T ) − 2,
and these bounds are sharp.
Proof. The proof is identical to the proof of Theorem 13 presented in [11] (which gives a similar bound for total domination),
except for one difference. The proof in [11] uses dominating sets and the fact that the domination number of a graph with no
isolated vertices is at most one-half the order of the graph. The proof in the paired-domination case uses paired-dominating
sets and the fact that the paired-domination number of a graph with no isolated vertices is at most the order of the graph.
Equality is achieved in the lower bound by, for example, any path. That the upper bound is sharp may be seen as follows.
Let T be the tree obtained from the disjoint union of k copies of P4 by joining a support vertex from one of these paths to a
support vertex from each of the other k − 1 paths and then subdividing each added edge twice. Then, γpr M
(T ) = 2k (the set
of 2k support vertices of T form a minimum MLPDS of T ) while γpr
L
(T ) = 2k + 2(k − 1) (the set consisting of the 2(k − 1)
subdivided vertices and the 2k support vertices forms a minimum LPDS of T ). Hence, γpr
L
(T ) = 2γprM (T ) − 2. 

Acknowledgements

Research of both authors supported in part by the South African National Research Foundation and the University of
KwaZulu-Natal.

References

[1] G. Chartrand, L. Eroh, M. Johnson, O.R. Oellermann, Resolvability in graphs and the metric dimension of a graph, Discrete Appl. Math. 105 (2000)
99–113.
[2] M. Chellai, On locating and differentiating-total domination in trees, Discuss. Math. Graph Theory 28 (2008) 383–392.
[3] M. Chellali, T.W. Haynes, Trees with unique minimum paired-dominating sets, Ars Combin. 73 (2004) 3–12.
[4] C.J. Colbourn, P.J. Slater, L.K. Stewart, Locating-dominating sets in series–parallel networks, Congr. Numer. 56 (1987) 135–162.
[5] A. Finbow, B.L. Hartnell, On locating dominating sets and well-covered graphs, Congr. Numer. 65 (1988) 191–200.
[6] S. Fitzpatrick, B. Hartnell, Paired-domination, Discuss. Math. Graph Theory 18 (1998) 63–72.
[7] J. Gimbel, B. van Gorden, M. Nicolescu, C. Umstead, N. Vaiana, Location with dominating sets, Congr. Numer. 151 (2001) 129–144.
[8] F. Harary, R.A. Melter, On the metric dimension of a graph, Ars Combin. 2 (1976) 191–195.
[9] T.W. Haynes, S.T. Hedetniemi, P.J. Slater, Fundamentals of Domination in Graphs, Marcel Dekker, New York, 1998.
[10] T.W. Haynes, S.T. Hedetniemi, P.J. Slater (Eds.), Domination in Graphs: Advanced Topics, Marcel Dekker, New York, 1998.
[11] T.W. Haynes, M.A. Henning, J. Howard, Locating and total dominating sets in trees, Discrete Appl. Math. 154 (8) (2006) 1293–1300.
[12] T.W. Haynes, P.J. Slater, Paired-domination and the paired-domatic number, Congr. Numer. 109 (1995) 65–72.
[13] T.W. Haynes, P.J. Slater, Paired-domination in graphs, Networks 32 (1998) 199–206.
[14] M.A. Henning, Graphs with large paired-domination number, J. Combin. Optim. 13 (2007) 61–78.
[15] M.A. Henning, J. McCoy, Which trees have a differentiating-paired-dominating set? (submitted for publication).
[16] M.A. Henning, O.R. Oellermann, Metric-locating-dominating sets in graphs, Ars Combin. 73 (2004) 129–141.
[17] M.A. Henning, M.D. Plummer, Vertices contained in all or in no minimum paired-dominating set of a tree, J. Combin. Optim. 10 (2005) 283–294.
[18] H. Qiao, L. Kang, M. Cardei,Ding-Zhu, Paired-domination of trees, J. Global Optim. 25 (2003) 43–54.
[19] D.F. Rall, P.J. Slater, On location-domination numbers for certain classes of graphs, Congr. Numer. 45 (1984) 97–106.
[20] P.J. Slater, Leaves of trees, Ars Combin. 2 (1976) 549–559.
[21] P.J. Slater, Dominating and location in acyclic graphs, Networks 17 (1987) 55–64.
[22] P.J. Slater, Dominating and reference sets in graphs, J. Math. Phys. Sci. 22 (1988) 445–455.
[23] P.J. Slater, Locating dominating sets and locating-dominating sets, in: Y. Alavi, A. Schwenk (Eds.), Graph Theory, Combinatorics, and Applications,
in: Proc. Seventh Quad. Internat. Conf. on the Theory and Applications of Graphs, John Wiley & Sons, Inc., 1995, pp. 1073–1079.