**Steiner diameter, maximum degree and size
of a graph**
***Supported by the National Science Foundation of China
(Nos. 11601254, 11551001, 11161037, and 11461054) and the Science Found of Qinghai
Province (Nos. 2016-ZJ-948Q, and 2014-ZJ-907).
Yaping Mao1,2, Zhao Wang3
1Department of Mathematics, Qinghai Normal
University, Xining, Qinghai 810008, China
2Center for Mathematics and Interdisciplinary Sciences
of Qinghai Province, Xining, Qinghai 810008, China
3School of Mathematical Sciences, Beijing Normal
University, Beijing 100875, China
Abstract
The Steiner diameter sdiamk(G) of a graph G, introduced by
Chartrand, Oellermann, Tian and Zou in 1989, is a natural
generalization of the concept of classical diameter. When k=2,
sdiam2(G)=diam(G) is the classical diameter. The problem of
determining the minimum size of a graph of order n whose diameter
is at most d and whose maximum is ℓ was first introduced by
Erdös and Rényi. Recently, Mao considered the problem of
determining the minimum size of a graph of order n whose Steiner
k-diameter is at most d and whose maximum is at most ℓ,
where 3≤k≤n, and studied this new problem when k=3. In
this paper, we
investigate the problem when n−3≤k≤n.
Keywords: diameter; maximum degree; order; Steiner diameter.
AMS subject classification 2010: 05C05; 05C12; 05C35.
1 Introduction
All graphs in this paper are undirected, finite and simple. We refer
to [5] for graph theoretical notation and terminology not
described here. For a graph G, let V(G), E(G), e(G),
δ(G), and G denote the set of vertices, the set
of edges, the size, minimum degree, and the complement of G,
respectively. The connectivity κ(G) is defined as the
order of a minimum vertex subset S of V(G) such that G−S is
disconnected or has only one vertex. In this paper, we let Kn,
Pn, K1,n−1 and Cn be the complete graph of order n, the
path of order n, the star of order n, and the cycle of order
n, respectively. For any subset X of V(G), let G[X] denote
the subgraph induced by X; similarly, for any subset F of
E(G), let G[F] denote the subgraph induced by F. We use
G∖X to denote the subgraph of G obtained by removing
all the vertices of X together with the edges incident with them
from G; similarly, we use G∖F to denote the subgraph of
G obtained by removing all the edges of F from G. For two
subsets X and Y of V(G) we denote by EG[X,Y] the set of
edges of G with one end in X and the other end in Y. The union G∪H of two graphs G and H is the graph
with vertex set V(G)∪V(H) and edge set E(G)∪E(H). If G
is the disjoint union of k copies of a graph H, we simply write
G=kH. The join G∨H of two disjoint graphs
G and H is the graph with vertex set V(G)∪V(H) and edge
set E(G)∪E(H)∪{uv∣u∈V(G),v∈V(H)}. We divide
our introduction into the following four subsections to state the
motivations and our results of this paper.
1.1 Distance and its generalizations
Distance is one of the most basic concepts of graph-theoretic
subjects. If G is a connected graph and u,v∈V(G), then the
distance dG(u,v) between u and v is the length of a
shortest path connecting u and v. If v is a vertex of a
connected graph G, then the eccentricity e(v) of v is
defined by e(v)=max{dG(u,v)∣u∈V(G)}. Furthermore, the
radius rad(G) and diameter diam(G) of G are
defined by rad(G)=min{e(v)∣v∈V(G)} and diam(G)=max{e(v)∣v∈V(G)}. These last two concepts are related by the
inequalities rad(G)≤diam(G)≤2rad(G). The center
C(G) of a connected graph G is the subgraph induced by the
vertices u of G with e(u)=rad(G). Goddard and Oellermann gave
a survey on this subject, see [24].
The distance between two vertices u and v in a connected graph
G also equals the minimum size of a connected subgraph of G
containing both u and v. This observation suggests a
generalization of distance. The Steiner distance of a graph,
introduced by Chartrand, Oellermann, Tian and Zou in 1989, is a
natural generalization of the concept of classical graph distance.
For a graph G(V,E) and a set S⊆V(G) of at least two
vertices, an S-Steiner tree or a Steiner tree
connecting S (or simply, an S-tree) is a subgraph
T(V′,E′) of G that is a tree with S⊆V′. Let G be a
connected graph of order at least 2 and let S be a nonempty set
of vertices of G. Then the Steiner distance dG(S) among
the vertices of S (or simply the distance of S) is the minimum
size among all connected subgraphs whose vertex sets contain S.
Note that if H is a connected subgraph of G such that
S⊆V(H) and ∣E(H)∣=dG(S), then H is a tree. Observe
that dG(S)=min{e(T)∣S⊆V(T)}, where T is subtree
of G. Furthermore, if S={u,v}, then dG(S)=d(u,v) is the
classical distance between u and v. Set dG(S)=∞ when
there is no S-Steiner tree in G.
Observation 1.1
Let G be a graph of order n and k be an integer with 2≤k≤n. If S⊆V(G) and ∣S∣=k, then dG(S)≥k−1.
Let n and k be two integers with 2≤k≤n. The
Steiner k-eccentricity ek(v) of a vertex v of G is
defined by ek(v)=max{d(S)∣S⊆V(G),∣S∣=k, and v∈S}. The Steiner k-radius of G is sradk(G)=min{ek(v)∣v∈V(G)}, while the Steiner k-diameter of
G is sdiamk(G)=max{ek(v)∣v∈V(G)}. Note for every
connected graph G that e2(v)=e(v) for all vertices v of G
and that srad2(G)=rad(G) and sdiam2(G)=diam(G).
The following Table 1 shows how the generalization proceeds.
[TABLE]
Table 1. Classical distance parameters and Steiner distance
parameters
Observation 1.2
Let k,n be two integers with 2≤k≤n.
(1)* If H is a spanning subgraph of G, then sdiamk(G)≤sdiamk(H).*
(2)* For a connected graph G, sdiamk(G)≤sdiamk+1(G).*
In [9], Chartrand, Okamoto, Zhang obtained the
following upper and lower bounds of sdiamk(G).
Theorem 1.1
[9]*
Let k,n be two integers with 2≤k≤n, and let G be a
connected graph of order n. Then k−1≤sdiamk(G)≤n−1.
Moreover, the upper and lower bounds are sharp.*
In [13], Dankelmann, Swart and Oellermann obtained a
bound on sdiamk(G) for a graph G in terms of the order of G
and the minimum degree δ of G, that is, sdiamk(G)≤δ+13n+3k. Later, Ali, Dankelmann, Mukwembi
[2] improved the bound of sdiamk(G) and showed that
sdiamk(G)≤δ+13n+2k−5 for all connected graphs
G. Moreover, they constructed graphs to show that the bounds are
asymptotically best possible. In [36], Mao obtained the
Nordhaus-Gaddum-type results for the parameter sdiamk(G).
As a generalization of the center of a graph, the Steiner
k-center Ck(G) (k≥2) of a connected graph G is the
subgraph induced by the vertices v of G with ek(v)=sradk(G).
Oellermann and Tian [46] showed that every graph is
the k-center of some graph. In particular, they showed that the
k-center of a tree is a tree and those trees that are k-centers
of trees are characterized. The Steiner k-median of G is
the subgraph of G induced by the vertices of G of minimum
Steiner k-distance. For Steiner centers and Steiner medians, we
refer to [44, 45, 46].
The average Steiner distance μk(G) of a graph G,
introduced by Dankelmann, Oellermann and Swart in
[11], is defined as the average of the Steiner
distances of all k-subsets of V(G), i.e.
[TABLE]
For more details on average Steiner distance, we refer to
[11, 12].
Let G be a k-connected graph and u, v be any pair of
vertices of G. Let Pk(u,v) be a family of k inner
vertex-disjoint paths between u and v, i.e.,
Pk(u,v)={P1,P2,⋯,Pk}, where p1≤p2≤⋯≤pk and pi denotes the number of edges of path Pi. The
k-distance dk(u,v) between vertices u and v is the
minimum pk among all Pk(u,v) and the k-diameter
dk(G) of G is defined as the maximum k-distance dk(u,v)
over all pairs u,v of vertices of G. The concept of k-diameter
emerges rather naturally when one looks at the performance of
routing algorithms. Its applications to network routing in
distributed and parallel processing are studied and discussed by
various authors including Chung [10], Du, Lyuu and Hsu
[15], Hsu [31, 32], Meyer and Pradhan [43].
1.2 Application background of Steiner distance parameters
The Steiner tree problem in networks, and particularly in graphs,
was formulated in 1971-by Hakimi (see [27]) and Levi (see
[33]). In the case of an unweighted, undirected graph, this
problem consists of finding, for a subset of vertices S, a
minimal-size connected subgraph that contains the vertices in S.
The computational side of this problem has been widely studied, and
it is known that it is an NP-hard problem for general graphs (see
[30]). The determination of a Steiner tree in a graph is a
discrete analogue of the well-known geometric Steiner problem: In a
Euclidean space (usually a Euclidean plane) find the shortest
possible network of line segments interconnecting a set of given
points. Steiner trees have application to multiprocessor computer
networks. For example, it may be desired to connect a certain set of
processors with a subnetwork that uses the least number of
communication links. A Steiner tree for the vertices, corresponding
to the processors that need to be connected, corresponds to such a
desired subnetwork.
The Wiener index W(G) of the graph G is defined as
W(G)=∑{u,v}⊆V(G)dG(u,v). Details on this
oldest distance–based topological index can be found in numerous
surveys, e.g., in [16, 48, 49, 52]. Li et al.
[34] put forward a Steiner–distance–based generalization of
the Wiener index concept. According to [34], the k-center Steiner Wiener index SWk(G) of the graph G is
defined by
[TABLE]
For k=2, the above defined Steiner Wiener index coincides with the
ordinary Wiener index. It is usual to consider SWk for 2≤k≤n−1, but the above definition would be applicable also in the
cases k=1 and k=n, implying SW1(G)=0 and SWn(G)=n−1. A
chemical application of SWk was recently reported in [26].
Gutman [25] offered an analogous generalization of the
concept of degree distance. Later, Furtula, Gutman, and Katanić
[20] introduced the concept of Steiner Harary index and
gave its chemical applications. Recently, Mao and Das [38]
introduced the concept of Steiner Gutman index and obtained some
bounds for it. For more details on Steiner distance indices, we
refer to [20, 26, 25, 34, 35, 38, 40, 41, 42].
1.3 Classical extremal problem and our generalization
What is the minimal size of a graph of order n and diameter d ?
What is the maximal size of a graph of order n and diameter d ?
It is not surprising that these questions can be answered without
the slightest effort (see [3]) just as the similar
questions concerning the connectivity or the chromatic number of a
graph. The class of maximal graphs of order n and diameter d is
easy to describe and reduce every question concerning maximal graphs
to a not necessarily easy question about binomial coefficient, as in
[28, 29, 47, 51]. Therefore, the authors study the minimal
size of a graph of order n and under various additional
conditions.
Erdös and Rényi [18] introduced the following problem.
Let d,ℓ and n be natural numbers, d<n and ℓ<n. Denote
by H(n,ℓ,d) the set of all graphs of order n with
maximum degree ℓ and diameter at most d. Put
[TABLE]
If H(n,ℓ,d) is empty, then, following the usual
convention, we shall write e(n,ℓ,d)=∞. For more details on
this problem, we refer to [3, 4, 18, 19].
Mao [37] considered the generalization of the above problem.
Let d,ℓ and n be natural numbers, d<n and ℓ<n. Denote
by Hk(n,ℓ,d) the set of all graphs of order n with
maximum degree ℓ and sdiamk(G)≤d. Put
[TABLE]
If Hk(n,ℓ,d) is empty, then, following the usual
convention, we shall write ek(n,ℓ,d)=∞. From Theorem
1.1, we have k−1≤d≤n−1.
In [37], Mao focused their attention on the case k=3, and
studied the exact value of e3(n,ℓ,d) for d=n−1,n−2,n−3,2,3.
In this paper, we investigate another extreme case when n−3≤k≤n−1, and give the exact values or upper and lower bounds of
ek(n,ℓ,d) for n−3≤k≤n−1. For general k (3≤k≤n−1), d (k−1≤d≤n−1) and ℓ (2≤ℓ≤n−1), we give upper and lower bounds of ek(n,ℓ,d).
2 The case k=n,n−1
In the sequel, let Ks,t, Kn, Cn and Pn denote the
complete bipartite graph of order s+t with part sizes s and t,
complete graph of order n, cycle of order n and path of order
n, respectively.
The following observation is immediate.
Observation 2.1
[36]*
(1) For a cycle Cn, sdiamk(Cn)=⌊kn(k−1)⌋;*
(2)* For a complete graph Kn, sdiamk(Kn)=k−1.*
The following result is easily proved in [37].
Lemma 2.1
[37]*
For 2≤ℓ≤n−1 and 3≤k≤n,*
[TABLE]
For k=n, we know that sdiamn(G)=n−1 for a connected graph G,
and hence d=n−1. From Lemma 2.1, the following result is
immediate.
Proposition 2.1
For 2≤ℓ≤n−1, en(n,ℓ,n−1)=n−1.
From now on, we assume that k≤n−1.
Proposition 2.2
For 3≤k≤n, ek(n,n−1,k)=n−1.
Proof.
Let G=K1,n−1 be a star of order n. Clearly,
Δ(G)=ℓ=n−1. Since sdiamk(G)=k and e(G)=n−1, it
follows that ek(n,n−1,k)≤n−1. On the other hand, since we
only consider connected graphs, it follows that e(G)≥n−1 for a
connected graph G is of order n. So ek(n,n−1,k)=n−1.
For k=n−1, we have n−2≤sdiamn−1(G)≤n−1 by Theorem
1.1. So we only need to consider the case d=n−1 or
d=n−2. Note that 2≤ℓ≤n−1.
Theorem 2.1
(1)* For 2≤ℓ≤n−1, en−1(n,ℓ,n−1)=n−1.*
(2)* For 2≤ℓ≤n−1, en−1(n,ℓ,n−2)=n+ℓ−2.*
Proof.
(1) The result follows from Lemma 2.1.
(2) For ℓ=2, we let Cn be the cycle of order n. From
Observation 2.1, we have sdiamn−1(G)=n−2. Since
e(G)=n, it follows that en−1(n,2,n−2)≤n. Let G be graph
of order n with Δ(G)=2 and sdiamn−1(G)=n−2. Since
Δ(G)=2, it follows that G=Pn or G=Cn. If G=Pn, then
sdiamn−1(G)=n−1, a contradiction. Therefore, G=Cn and hence
e(G)≥n. So en−1(n,2,n−2)=n.
Suppose 3≤ℓ≤n−1. Let G be a graph obtained from a
cycle Cn=v1v2⋯vn by adding the edges v1vj (3≤j≤ℓ). From Observation 1.1, we have
sdiamn−1(G)≤sdiamn−1(Cn)≤n−2. Since
Δ(G)=ℓ and e(G)=n+ℓ−2, it follows that
en−1(n,ℓ,n−2)≤n+ℓ−2. Conversely, let G be a graph
such that sdiamn−1(G)=n−2 and Δ(G)=ℓ (3≤ℓ≤n−1). Then there exists a vertex u in G such that
dG(u)=ℓ. Choose S=V(G)−u. Since sdiamn−1(G)=n−2, it
follows that there exists an S-Steiner tree in G−u, say T.
Then e(G−u)≥e(T)=n−2 and hence e(G)=e(G−u)+ℓ≥n−2+ℓ, which implies that en−1(n,ℓ,n−2)≥n+ℓ−2.
From the above arguments, we conclude that en−1(n,ℓ,n−2)=n+ℓ−2 for 2≤ℓ≤n−1.
3 The case k=n−2
From Theorem 1.1, we have n−3≤sdiamn−2(G)≤n−1.
Mao et al. [39] characterized the graphs with
sdiamn−2(G)=d (n−3≤d≤n−1).
Lemma 3.1
[39]*
Let G be a connected graph of order n (n≥5). Then*
(1)* sdiamn−2(G)=n−3 if and only if κ(G)≥3.*
(2)* sdiamn−2(G)=n−2 if and only if κ(G)=2 or G
contains only one cut vertex.*
(3)* sdiamn−2(G)=n−1 if and only if there are at least two cut
vertices in G.*
For d=n−2, we have the following.
Proposition 3.1
For 2≤ℓ≤n−1 and n≥5,
[TABLE]
Proof.
For ℓ=n−1, from Proposition 2.2, we have
en−2(n,ℓ,n−2)=n−1. From now on, we suppose 2≤ℓ≤n−2. Let G be a graph obtained by a cycle Cn−ℓ+2 and a
star K1,ℓ−2 by identifying a vertex of Cn−ℓ+2 and
the center of K1,ℓ−2. Clearly, Δ(G)=ℓ and there is
exactly one cut vertex in G. From (2) of Lemma 3.1, we
have sdiamn−2(G)=n−2, and hence en−2(n,ℓ,n−2)≤n. It
suffices to show that en−2(n,ℓ,n−2)≥n. Let G be a
graph of order n with sdiamn−2(G)≤n−2 and
Δ(G)=ℓ. If G is a tree, then G contains at least two
cut vertices, since Δ(G)=ℓ≤n−2. From (3) of Lemma
3.1, sdiamn−2(G)=n−1, a contradiction. So G contains
at least one cycle, and hence e(G)≥n. Therefore, we have
en−2(n,ℓ,n−2)=n, as desired.
Let Pji be a path of order j, where 1≤i≤r+2. We call
the graph K1∨(K1∪Pji) as a (ui,vi,Pji)-Fan; see Figure 1 (a). For 1≤i≤r, we
choose (ui,vi,P2i)-Fan, and choose
(ur+1,vr+1,Pℓ−1r+1)-Fan and
(ur+2,vr+2,Psr+2)-Fan. Let Hn be a graph obtained
from the above (r+2) Fans by adding the edges in
[TABLE]
see Figure 1 (b), where 4r+ℓ+s+3=n, 2≤s≤5 and 1≤r≤4n−ℓ−4.
Proposition 3.2
For 6≤ℓ≤n−9,
[TABLE]
where 2≤s≤5. Furthermore, if s=2, then en−2(n,ℓ,n−3)=21(3n+ℓ−3).
Proof.
Let Hn be the graph constructed in Figure 1 (b). Clearly, Hn
is 3-connected. From Lemma 3.1, sdiamn−2(Hn)=n−3.
Since 6≤ℓ≤n−9 and 2≤s≤5, it follows that
Δ(Hn)=ℓ. Then
[TABLE]
and hence en−2(n,ℓ,n−3)≤21(3n+ℓ+s−5).
Conversely, we suppose that G is a graph with ∣V(G)∣=n, Δ(G)=ℓ, and sdiamn−2(G)=n−3.
Then there is a vertex in G, say u, such that dG(u)=ℓ. Since sdiamn−2(G)=n−3, it follows
from Lemma 3.1 that κ(G)≥3, and hence dG(v)≥3 for any v∈V(G)∖u.
Therefore, e(G)≥21(3n+ℓ−3), and hence en−2(n,ℓ,n−3)≥21(3n+ℓ−3).
For n−8≤ℓ≤n−1, we have the following result.
Lemma 3.2
Let k,ℓ be two integers with n−8≤ℓ≤n−1. Then
(i)* en−2(n,n−1−i,n−3)=2n−2 for n≥5+i and i=0,1;*
(ii)* en−2(n,n−3−i,n−3)=2n−3 for n≥7+2i and i=0,1;*
(iii)* en−2(n,n−5−i,n−3)=2n−4 for n≥11+2i and i=0,1;*
(iv)* en−2(n,n−7−i,n−3)=2n−5 for n≥15+2i and i=0,1.*
Proof.
For (i), we first consider the case i=0. For ℓ=n−1, let
Gn1 be a wheel of order n. From Lemma 3.1,
sdiamn−2(Gn1)=n−3 and Δ(Gn1)=n−1, and hence
en−2(n,n−1,n−3)≤2n−2. Conversely, we suppose that G is a
graph of order n such that sdiamn−2(G)=n−3 and
Δ(G)=n−1. Then there exists a vertex u such that
dG(u)=n−1. Since κ(G)≥3, it follows that
κ(G−u)≥2, and hence e(G−u)≥n−1. Then e(G)≥2n−2, and hence en−2(n,n−1,n−3)≥2n−2. So
en−2(n,n−1,n−3)=2n−2.
Next, we consider the case i=1. For ℓ=n−2, let Gn2 be a
graph of order n obtained by Gn−11 by deleting the edge
w2w3, and then adding a new vertex w and three edges
ww1,ww2,ww3; see Figure 2 (b). Since κ(Gn2)=3, it
follows from Lemma 3.1 that sdiamn−2(Gn2)=n−3. From
this together with Δ(Gn2)=n−2, we have
en−2(n,n−2,n−3)≤2n−2. Conversely, we suppose that G is a
graph of order n such that sdiamn−2(G)=n−3 and
Δ(G)=n−2. Then there exists a vertex u such that
dG(u)=n−2. Since sdiamn−2(G)=n−3, it follows that
κ(G)≥3, and hence κ(G−u)≥2. Clearly,
e(G−u)≥n−1. If e(G−u)=n−1, then G−u is a cycle of order
n−1, say G−u=v1v2…vn−1v1. Since dG(u)=n−2, it
follows that there exists some vertex vi in G−u such that
uvi∈/E(G), and hence dG(vi)=2, which contradicts to the
fact κ(G)≥3. Then e(G−u)≥n, and hence e(G)≥2n−2, and hence en−2(n,n−2,n−3)≥2n−2. So, we have
en−2(n,n−2,n−3)=2n−2.
For (ii), we first consider the case i=0. For ℓ=n−3, let
Gn3 be a graph of order n obtained by Gn−12 by deleting
the edge v1w1,w1wn−3, and then adding a new vertex w and
three edges wv1,ww1,wwn−3; see Figure 3 (b). Since
κ(Gn3)=3, it follows from Lemma 3.1 that
sdiamn−2(Gn3)=n−3. Note that Δ(Gn3)=n−3. Therefore,
we have en−2(n,n−3,n−3)≤2n−3. Conversely, we suppose that
G is a graph of order n such that sdiamn−2(G)=n−3 and
Δ(G)=n−3. Then there exists a vertex u such that
dG(u)=n−3. Since sdiamn−2(G)=n−3, it follows that
κ(G)≥3, and hence κ(G−u)≥2. Clearly,
e(G−u)≥n−1. If e(G−u)=n−1, then G−u is a cycle of order
n−1, say G−u=v1v2…vn−1v1. Since dG(u)=n−3, it
follows that there exists some vertex vi in G−u such that
uvi∈/E(G), and hence dG(vi)=2, which contradicts to the
fact κ(G)≥3. Then e(G−u)≥n, and hence e(G)≥2n−3, and hence en−2(n,n−3,n−3)≥2n−3. So, we have
en−2(n,n−3,n−3)=2n−3.
Next, we consider the case i=1. For ℓ=n−4, let Gn4 be a
graph of order n obtained by Gn−13 by deleting the edge
v2wn−4, and then adding a new vertex w and three edges
wv2,wwn−4,wwn−5; see Figure 3 (d). Since
κ(Gn4)=3, it follows from Lemma 3.1 that
sdiamn−2(Gn4)=n−3. Note that Δ(Gn3)=n−4. Therefore,
we have en−2(n,n−2,n−3)≤2n−3. Conversely, we suppose that
G is a graph of order n such that sdiamn−2(G)=n−3 and
Δ(G)=n−4. Then there exists a vertex u such that
dG(u)=n−4. Since sdiamn−2(G)=n−3, it follows that
κ(G)≥3, and hence κ(G−u)≥2. Clearly,
e(G−u)≥n−1. If e(G−u)=n−1, then G−u is a cycle of order
n−1, say G−u=v1v2…vn−1v1. Since dG(u)=n−3, it
follows that there exists some vertex vi in G−u such that
uvi∈/E(G), and hence dG(vi)=2, which contradicts to the
fact κ(G)≥3. If e(G−u)=n, then G−u is a graph
obtained from a cycle C=v1v2…vn−1v1 by adding some
edge vpvq (1≤p=q≤n−1). Since dG(u)=n−4, it
follows that there exists some vertex vi (1≤i≤n−1, i=p, i=q) in G−u such that uvi∈/E(G), and hence
dG(vi)=2, which contradicts to the fact κ(G)≥3. Then
e(G−u)≥n+1, and hence e(G)≥2n−3, and hence
en−2(n,n−4,n−3)≥2n−3. So, we have
en−2(n,n−4,n−3)=2n−3.
For (iii) and (iv), we only give the graph construction
operation (see Figure 3), and omit the proof of them.
From Propositions 3.1 and 3.2, and Lemmas
3.2 and 2.1, we have the following theorem.
Theorem 3.1
(1)* For 2≤ℓ≤n−1,*
[TABLE]
(2)* For 2≤ℓ≤n−1 and n≥5,*
[TABLE]
(3)* For n−8≤ℓ≤n−1, en−2(n,n−1−i,n−3)=2n−2 for
n≥5+i and i=0,1; en−2(n,n−3−i,n−3)=2n−3 for n≥7+2i and i=0,1; en−2(n,n−5−i,n−3)=2n−4 for n≥11+2i and
i=0,1; en−2(n,n−7−i,n−3)=2n−5 for n≥15+2i and i=0,1.
For 6≤ℓ≤n−9,*
[TABLE]
where 2≤s≤5. Furthermore, if s=2, then
en−2(n,ℓ,n−3)=21(3n+ℓ−3).
4 The case k=n−3
Mao et al. [36] derived the following results for Steiner
(n−3)-diameter.
Lemma 4.1
[36]*
Let G be a connected graph of order n. Then sdiamn−3(G)=n−4
if and only if κ(G)≥4, and sdiamn−3(G)=n−1 if and
only if G contains at least 3 cut vertices.*
Wang et al. [50] obtained the structural properties of
graphs with sdiamk(G)=n−1.
Lemma 4.2
[50]*
Let k,n be two integers with 3≤k≤n−1. Let G be a
connected graph of order n. Then sdiamk(G)=n−1 if and only if
the number of non-cut vertices in G is at most k.*
The following corollary is immediate from Lemma 2.1.
Corollary 4.1
For 2≤ℓ≤n−1, en−3(n,ℓ,n−1)=n−1.
Let uv be an edge in G. A double-star on uv is a
maximal tree in G which is the union of stars centered at u or
v such that each star contains the edge uv.
Proposition 4.1
For 2≤ℓ≤n−1 and n≥4,
[TABLE]
Proof.
For 2≤ℓ≤⌊2n⌋, we let G be a
graph of order n obtained from a cycle Cn−ℓ+2 and a star
K1,ℓ−2 by identifying the center and one vertex of
Cn−ℓ+2. Clearly, Δ(G)=ℓ, and G contains only one
cut vertex. From Lemma 4.2, we have sdiamn−3(G)≤n−2, and hence en−3(n,ℓ,n−2)≤n. It suffices to show
en−3(n,ℓ,n−2)=n if 2≤ℓ≤⌊2n⌋−1, or ℓ=⌊2n⌋
and n is odd. Let G be a graph of order n such that
sdiamn−3(G)≤n−2 and Δ(G)=ℓ, where 2≤ℓ≤⌊2n⌋−1, or
ℓ=⌊2n⌋ and n is odd. Then we have the
following claim.
Claim 1. G is not a tree.
Proof of Claim 1. Assume, to the contrary, that G
is a tree. Since sdiamn−3(G)≤n−2, it follows from Lemma
4.2 that G contains at most two cut vertices. Then
G=K1,n−1 or G is a double star of order n. If
G=K1,n−1, then Δ(G)=n−1>⌊2n⌋, a
contradiction. Suppose that G is a double star of order n. Let
u,v be the two centers of G. Then dG(u)≥⌈2n⌉ or dG(v)≥⌈2n⌉, and
hence ℓ≥⌈2n⌉, a contradiction.
From Claim 1, G is not a tree. Then e(G)≥n, and hence
en−3(n,ℓ,n−2)=n if 2≤ℓ≤⌊2n⌋−1, or ℓ=⌊2n⌋
and n is odd.
We now show that if ⌊2n⌋+1≤ℓ≤n−1,
or ℓ=⌊2n⌋ and n is even, then
en−3(n,ℓ,n−2)=n−1. Let G be a double star of order n
such that dG(u)=ℓ and dG(v)=n−ℓ, where u,v are the two
centers of G. From Lemma 4.2, we have sdiamn−3(G)≤n−2. Since ⌊2n⌋+1≤ℓ≤n−1, or
ℓ=⌊2n⌋ and n is even, it follows that
Δ(G)=ℓ, and hence en−3(n,ℓ,n−2)≤n−1. So, we
have en−3(n,ℓ,n−2)=n−1.
A graph is said to be minimally k-connected if it is
k-connected but omitting any of the edges the resulting graph is
no longer k-connected.
Let A32 be a minimally 4-connected graph shown in Figure 4
(a) (see [3], Page 18). We now give a graph Hn of
order n (n≥96) such that Δ(Hn)=ℓ and
sdiamn−3(Hn)=n−4 constructed by the following steps.
- Step 1: For each i (1≤i≤x), we let A32i be the copy of
A32, where n=32x+y, x=⌊n/32⌋, and 0≤y≤31. Let V(A32i)={uji∣1≤j≤12}∪{vji∣1≤j≤20} such that dG(uji)=5 for
1≤j≤12, and dG(vji)=4 for 1≤j≤20; see
Figure 4 (a). Let B32x be a graph obtained from A32i (1≤i≤x) by adding the edges in
{v5iv1i+1∣1≤i≤x−1}∪{v6iv2i+1∣1≤i≤x−1}∪{v7iv3i+1∣1≤i≤x−1}∪{v8iv4i+1∣1≤i≤x−1}; see Figure 4 (b).
- Step 2: Let y=4z+a, where z=⌊y/4⌋, 0≤a≤3. For each j (1≤j≤z), we let
K4j be the complete graph of order 4. Furthermore, let
K4j,∗ be the graph obtained from K4j by adding four
pendant vertices w1j,w2j,w3j,w4j with four pendant edges
such that another end vertex of each pendant edge is attached on
only one vertex in K4j; see Figure 4 (c). For each j (1≤j≤a), we let S5j be the star of order 5 with its leaves
p1j,p2j,p3j,p4j. Since n≥96, it follows that
A321,A322,A323 all exist. Set S1={vj1∣1≤j≤4}∪{vj1∣9≤j≤20}⊆V(A321),
and S2={vj2∣9≤j≤20}⊆V(A322), and
S3={vj3∣9≤j≤20}⊆V(A323). Then
∣S1∪S2∪S3∣=40. If n≡0 (mod 32), then
Dn=B32x. If n=0 (mod 32) and n−32x≡0 (mod 4),
then Dn is a graph obtained from B32x and
K41,∗,K42,∗,…,K4z,∗ by identifying each vertex in
S′={wji∣1≤i≤4, 1≤j≤z} and only one
vertex in S1∪S2∪S3. Since ∣S′∣=4z<40=∣S1∪S2∪S3∣, for any vertex in S′, we can find a vertex in S1∪S2∪S3 and then identify the two vertices. If n=0 (mod 32) and n−32x=0 (mod 4), then Dn is a graph
obtained from B32x, K41,∗,K42,∗,…,K4z,∗ and
S51,S52,…,S5a by identifying each vertex in
S′={wji∣1≤i≤4, 1≤j≤z}∪{pji∣1≤i≤4, 1≤j≤a} and only one vertex
in S1∪S2∪S3. Since ∣S′∣=4z+4a≤28+12=40=∣S1∪S2∪S3∣, for any vertex in S′, we can find a vertex in
S1∪S2∪S3 and then identify the two vertices.
- Step 3: Let Hn be the graph Dn by adding
ℓ−5 edges between u121 and V(G)−u121.
We now in a position to give the upper and lower bounds of
en−3(n,ℓ,n−4).
Proposition 4.2
Let ℓ,n be two integers with 2≤ℓ≤n−1 and n≥96. Then
[TABLE]
where n≡ i (mod 32), 1≤i≤31.
Proof.
Let G be a graph of order n such that sdiamn−3(G)=n−4 and
Δ(G)=ℓ, where 2≤ℓ≤n−1. Since
Δ(G)=ℓ, it follows that there exists a vertex v in G
such that dG(v)=ℓ. Since sdiamn−3(G)=n−4, it follows from
Lemma 4.2 that δ(G)≥κ(G)≥4. For any
vertices in V(G)−v, its degree is at least 4. Then 2e(G)≥ℓ+4(n−1), and hence en−3(n,ℓ,n−3)≥2n−2−⌈ℓ/2⌉.
It suffices to show en−3(n,ℓ,n−3)≤74⌊32n⌋+2i+ℓ−9, where n≡ i (mod 32),
and 1≤i≤31. Let G=Hn. Clearly, Δ(G)=ℓ. Since
κ(G)≥4, it follows from Lemma 4.2 that
sdiamn−3(G)=n−4, and hence en−3(n,ℓ,n−3)≤e(G)=74⌊32n⌋+2i+ℓ−9, where
n≡ i (mod 32), 1≤i≤31.
For the remain case d=n−3, we have the following.
Proposition 4.3
Let ℓ,n be two integers with 2≤ℓ≤n−1 and n≥5.
(i)* If ⌈n/2⌉+1≤ℓ≤n−1, then
en−3(n,ℓ,n−3)≤2n−ℓ+1.*
(ii)* If 5≤ℓ≤⌈n/2⌉, then*
[TABLE]
where n=(ℓ+1)x+i and 0≤i≤ℓ.
(iii)* If 2≤ℓ≤n−1, then en−3(n,ℓ,n−3)≥max{n−1+⌈ℓ/2⌉,⌈23n−ℓ−3⌉}.*
Proof.
(i) Let G be a graph obtained from a wheel Wn−ℓ+3 with
center w and a star K1,ℓ−3 by identifying the center of
the star and one vertex of Wn−ℓ+3−w. Note that
Wn−ℓ+3−w is a cycle of order n−ℓ+2, say C=u1u2…un−ℓ+2u1. Let u1 be the identifying vertex. Clearly,
dG(w)=n−ℓ+2, dG(u1)=ℓ and dG(ui)=3 for each i (2≤i≤n−ℓ+2). Since ⌈n/2⌉+1≤ℓ≤n−1, it follows that ℓ≥n−ℓ+2, and hence
Δ(G)=ℓ. Since G contains only one cut vertex, one can
easily check that sdiamn−3(G)≤n−3. So, we have
en−3(n,ℓ,n−3)≤e(G)=2n−ℓ+1.
(ii) Let n=(ℓ+1)x+y, where x=⌊n/(ℓ+1)⌋,
0≤y≤ℓ. For each i (1≤i≤x), we let
Wℓ+1i be the wheel of order ℓ+1, with its center
wi. Note that Wℓ+1i−wi is a cycle of order ℓ, say
v1iv2i…vℓiv1i. Let F(ℓ+1)x be a graph
obtained from Wℓ+1i (1≤i≤x) by adding the edges
in {vjivji+1∣1≤i≤x−1, 1≤j≤3}. We
now give a graph Dn of order n (n≥ℓ+1) constructed in
the following way.
If n≡ 0 (mod ℓ+1), then Dn=F(ℓ+1)x.
If n≡ 1 (mod ℓ+1), then Dn is the graph obtained from F(ℓ+1)x
by adding a new vertex u and three edges uv11,uv21,uv31.
If n≡ 2 (mod ℓ+1), then Dn is the graph obtained from F(ℓ+1)x
by adding two new vertices u1,u2 and six edges
{u1u2,u1v11,u1v21,u2v31,u2v41,u2v51}.
If n≡ 3 (mod ℓ+1), then Dn is the graph obtained from F(ℓ+1)x
by adding two new vertices u1,u2,u3 and six edges
{u1u2,u1u3,u2u3,u1v11,u2v21,u3v31}.
If n≡ y (mod ℓ+1) (4≤y≤ℓ), then Dn is the graph obtained from
F(ℓ+1)x and a new wheel Wy∗ with center w∗ and y
vertices by adding three edges
{v2∗v11,v2∗v21,v3∗v31}, where
V(Wy∗)−w∗={v2∗,v2∗,…,vy−1∗}.
Let Hn be the graph Dn by adding ℓ−5 edges between
v11 and V(G)−v11. Clearly, Δ(Hn)=ℓ. Since
Hn is 3-connected, one can easily check that
sdiamn−3(G)≤n−3. Since
[TABLE]
it follows that
[TABLE]
where n=(ℓ+1)x+i and 0≤i≤ℓ.
(iii) Let G be a graph of order n such that
sdiamn−3(G)≤n−3 and Δ(G)=ℓ, where 2≤ℓ≤n−1. Then G is 2-connected, or G contains only one cut
vertex. If G is 2-connected, then there exists a vertex v in
G such that dG(v)=ℓ, since Δ(G)=ℓ. For any
vertices in V(G)−v, its degree is at least 2. Then 2e(G)≥ℓ+2(n−1), and hence en−3(n,ℓ,n−3)≥n−1+⌈ℓ/2⌉. Suppose that G contains only one cut vertex, say
v. Then each connected component of G∖v is a connected
subgraph of order at least 3, or an edge of G, or an isolated
vertex. Let w1,w2,…,wr be the isolated vertices,
e1,e2,…,es be the edges, and C1,C2,…,Ct be the
connected components of order at least 3 in G∖v. Then
we have the following claim.
Claim 1. For any w∈⋃i=1tV(Ci), if dG(w)=2,
then wv∈E(G).
Proof of Claim 1. Assume, to the contrary, that
wv∈/E(G) for any w∈⋃i=1tV(Ci) with
dG(w)=2. Without loss of generality, let w∈V(C1). Then
there exist two vertices u1,u2 in C1 such that u1w∈E(C1) and u2w∈E(C1). Choose S⊆V(G) with
∣S∣=n−3 such that w∈S but u1,u2,v∈/S. Then any
S-Steiner tree must occupy v and one of u1,u2, and hence
dG(S)≥n−2, a contradiction.
From Claim 1, we suppose that there are x vertices
u1,u2,…,ux in ⋃i=1tV(Ci) such that its degree
is 2. Then for any vertex in (⋃i=1tV(Ci))∖{u1,u2,…,ux}, its degree is at least 3.
If dG(v)=ℓ, then x≤ℓ−r−2s and r≤ℓ.
Furthermore, we have
[TABLE]
and hence e(G)≥⌈23n−ℓ−3⌉.
If dG(v)=ℓ, then there exists a vertex u∈V(G)−v such
that dG(u)=ℓ. Clearly, r+2s+t≤dG(v)≤ℓ. Since
x+2s+r≤ℓ−1, it follows that
[TABLE]
and hence e(G)≥⌈23n−5⌉.
From the above argument, we conclude that en−3(n,ℓ,n−3)≥max{n−1+⌈ℓ/2⌉,⌈23n−ℓ−3⌉}.
From Propositions 4.1, 4.2 and 4.3, we
have the following theorem.
Theorem 4.1
(i)* For 2≤ℓ≤n−1, en−1(n,ℓ,n−1)=n−1.*
(ii)* For 2≤ℓ≤n−1 and n≥4,*
[TABLE]
(iii)* For 2≤ℓ≤n−1 and n≥96,*
[TABLE]
where n≡ i (mod 32), 1≤i≤31.
(iv)* If 2≤ℓ≤n−1, then en−3(n,ℓ,n−3)≥max{n−1+⌈ℓ/2⌉,23n−ℓ−5}. If ⌈n/2⌉+1≤ℓ≤n−1, then en−3(n,ℓ,n−3)≤2n−ℓ+1. If 5≤ℓ≤⌈n/2⌉, then*
[TABLE]
where n=(ℓ+1)x+i and 0≤i≤ℓ.
5 For general k
In [39], Mao et al. obtained the following
result.
Lemma 5.1
Let ℓ,n be two integers with 1≤ℓ≤n−2, and let G
be a graph of order n. Then κ(G)≥ℓ if and only if
sdiamn−ℓ+1(G)=n−ℓ.
In this section, we construct a graph and give an upper bound of ek(n,ℓ,d)
for general k, ℓ, and d.
Theorem 5.1
Let k,ℓ,d be three integers with 2≤k≤n, 2≤ℓ≤n−1, and k−1≤d≤n−1.
(i)* If d=k−1, ⌈2n+1⌉≤k≤n, and
max{n−k+1,⌈2n⌉}<ℓ≤n−1, then*
[TABLE]
(ii)* If 2≤k≤d, k≤d≤n−1, and 2+⌈d−k+1n−d+k−3⌉≤ℓ≤n−1, then*
[TABLE]
Proof.
(i) We first consider the lower bound. From Lemma 5.1,
for a connected graph G of order n, sdiamk(G)=k−1 if and
only if κ(G)≥n−k+1. Let G be a graph of order n such
that sdiamk(G)=k−1 and Δ(G)=ℓ, where 2≤ℓ≤n−1. Since sdiamk(G)=k−1, it follows that δ(G)≥κ(G)≥n−k+1. Since Δ(G)=ℓ, it follows that there
exists a vertex v in G such that dG(v)=ℓ. For any vertex
in V(G)−v, its degree is at least n−k+1. Then 2e(G)≥ℓ+(n−1)(n−k+1), and hence ek(n,ℓ,k−1)≥⌈2ℓ+(n−1)(n−k+1)⌉.
Next, we consider the upper bound. For
max{n−k+1,⌈2n⌉}≤ℓ≤n−1, we let
Ka,b be a complete bipartite graph of order n=a+b with a≥b. Let U,V={v1,v2,…,vb} be the parts of order a,b in
Ka,b. Let G be a graph obtained from Ka,b by adding
edges v1vi (2≤i≤ℓ−a), where ℓ≤a+b−1=n−1.
Then Δ(G)=a+(ℓ−a)=ℓ. Since
max{n−k+1,⌈2n⌉}<ℓ≤n−1 and a≥b,
it follows that for any S⊆V(G) and ∣S∣=k, we have
S∩U=∅ and S∩U=∅, and hence
sdiamk(G)=k−1. So ek(n,ℓ,d)≤ab+ℓ−a=a(n−a)−a+ℓ≤4(n−1)2+ℓ.
(ii) Let Pd−k+3=v1v2…vd−k+3 be a path of order
d−k+3. Set x=⌈d−k+1n−d+k−3⌉,
Ui={ui,1,ui,2,…,ui,x∣2≤i≤d−k+1}, and
Ud−k+2={ud−k+2,1,ud−k+2,2,…,ud−k+2,p}, where
p=n−(d−k)(x+1)+3. Let T′ be a tree of order n obtained from
Pd−k+3 by adding all the vertices in
⋃i=2d−k+2Ui, and then adding the edges in
⋃i=2d−k+2ET′[vi,Ui]. Let T be a tree obtained
from T′ by gradually deleting ℓ−x−2 vertices in
V(T′)∖({v1,v2}⋃U2) and then adding ℓ−x−2
pendant edges at v2. Clearly, dT(v2)=ℓ=Δ(T). Since
sdiamk(T)≤d, it follows that ek(n,ℓ,d)=n−1.