• Aucun résultat trouvé

4. Minimum status and number of pendant vertices

N/A
N/A
Protected

Academic year: 2022

Partager "4. Minimum status and number of pendant vertices"

Copied!
21
0
0

Texte intégral

(1)

https://doi.org/10.1051/ro/2020015 www.rairo-ro.org

MINIMUM STATUS OF TREES WITH GIVEN PARAMETERS

Zhene Peng and Bo Zhou

Abstract.The status of a vertex vin a connected graph Gis defined as the sum of the distances from vto all other vertices in G. The minimum status of Gis the minimum of status of all vertices ofG. We give the smallest and largest values for the minimum status of a tree with fixed parameters such as the diameter, the number of pendant vertices, the number of odd vertices, and the number of vertices of degree two, and characterize the unique extremal trees.

Mathematics Subject Classification. 05C12, 05C35, 05C90.

Received September 29, 2019. Accepted February 12, 2020.

1. Introduction

We consider simple and undirected graphs. LetGbe a connected graph of ordernwith vertex setV(G). The distance between verticesuandv inG, denoted bydG(u, v), is the length of a shortest path connectinguand v inG. The status of a vertexuin Gis defined as [9,15]

sG(u) = X

v∈V(G)

dG(u, v).

The minimum status ofG, denoted bys(G), is defined as [9,24,27]

s(G) = min{sG(u) : u∈V(G)}.

In the literature, the status of a vertex is also known as the distance [12,26], the transmission [21,25,29,30], or the total distance [6] of a vertex, and if n≥2, the normalized status n−11 sG(u) of vertex uinG is called the average distance from vertexu to all other vertices ofG, see [2]. The status or normalized status of a vertex may be used to measure its closeness centrality in the network [13,14]. The minimum status of a graphGis also studied through the proximity ofG, defined as n−11 s(G), see,e.g., [1–5]. Recall that the mean vertex derivation ofG, defined as 1ns(G), was also studied in [31].

Aouchiche and Hansen [2] gave sharp lower and upper bounds for the minimum status of a graph as a function of its order, and characterized the extremal graphs, and they also gave a sharp lower bound for the minimum status of a graph with fixed order and diameter. A trivial lower bound for the minimum status of a connected graph of ordernisn−1, which is achieved if and only if there is a vertex of degreen−1. LetGbe a connected

Keywords.Minimum status, closeness centrality, diameter, pendant vertices, odd vertices, series-reduced trees.

School of Mathematical Sciences, South China Normal University, Guangzhou 510631, P.R. China.

Corresponding author:zhoubo@scnu.edu.cn

Article published by EDP Sciences c EDP Sciences, ROADEF, SMAI 2021

(2)

graph of order n. They showed that s(G)≤ bn42cwith equality if and only if Gis either the cycle Cn or the pathPn. If Gis a tree, then this result has also been given in [31]. Aouchiche and Hansen [1] gave Nordhaus- Gaddum-type inequalities for the minimum status of a graph,i.e., lower and upper bounds fors(G) +s(G) and s(G)s(G) in terms of the order of G with both G and Gbeing connected, where G is the complement of G.

Lin et al. [24] established sharp lower and upper bounds for the minimum status of a graph with maximum degree, and they characterized extremal graphs for the lower bound and gave a necessary condition for graphs attaining the upper bound. Rissner and Burkard [27] gave alternate and simple proofs for the results in [24], and showed that the minimum status and the radius achieve their minimum (maximum, respectively) values at the same type of trees when order and maximum degree are given. Liang et al. [22] gave sharp lower and upper bounds for the minimum status of a tree with fixed matching number (domination, respectively), and characterize the extremal trees. Related work may be found in [2,3,10,28].

A vertex in a graph is a pendant vertex if its degree is one, and a vertex in a graph is an odd (even, respectively) vertex if its degree is odd (even, respectively). By Handshaking Theorem, a graph possesses an even number of odd vertices. A vertex is an internal vertex if its degree is at least two. A tree whose internal vertices all have degree at least three is called a series-reduced tree or a homeomorphically irreducible tree [7,16,17]. That is, a series-reduced tree has no vertices of degree two.

Following the above work, in this paper, we give sharp lower and upper bounds for the minimum status of trees using parameters, including the diameter, the number of pendant vertices, the number of odd vertices, and the number of vertices of degree two, and we also characterize all extremal cases. In some cases, the bounds may be extended to connected graphs.

2. Preliminaries

For a proper subset U of vertices of a graph G, G−U denotes the subgraph of G obtained by deleting the vertices from U (and the incident edges), and in particular, we write G−v forG− {v} if U ={v}. For a subset E1 of edges ofG,G−E1 denotes the subgraph obtained from Gby deleting all the edges inE1, and in particular, we writeG−uvforG− {uv}ifE1={uv}. For a subsetE2 of unordered vertex pairs of distinct vertices of G, if each element of E2 is not an edge of G, then G+E2 denotes the graph obtained from Gby adding all elements ofE2 as edges, and in particular, we writeG+uvforG+{uv}ifE2={uv}.

For a vertexv in a graphG, NG(v) denotes the set of vertices adjacent tov (i.e., neighbors ofv) in G, and the degree of v, denoted by δG(v), is equal to |NG(v)|. For k ≥1, we say a path P =v0. . . vk in a graph G is a pendant path (of length k) at v0 ifδG(v0)>2, δG(vk) = 1, and ifk ≥2,δG(v1) =· · · =δG(vk−1) = 2.

A pendant path of length one is also known as a pendant edge.

IfPis a pendant path of length`at vertexvin a graphG, thenGis said to be obtained fromG−(V(P)\{v}) by attaching a pendant path of length ` at v. Particularly, if `= 1, say P =vw, then we also say thatG is obtained fromG−wby attaching a pendant edge atv(or a pendant vertex to v).

A caterpillar is a tree with a path such that each vertex of the tree either lies on this path or is adjacent to a vertex in this path. Note thatSn andPn are caterpillars.

A quasi-pendant vertex is a vertex that is adjacent to a pendant vertex.

The median of a connected graphGis the set of vertices of Gwith minimum status. The median of a tree consists of either one vertex or two adjacent vertices [12,18,31].

Lemma 2.1([19,20,31]). Let T be a nontrivial tree of order n. Then a vertexxis in the median of T if and only if|V(C)| ≤n2 for every componentC of T−x.

It follows from Lemma2.1that any pendant vertex in a tree with at least 3 vertices can not be in the median of this tree.

For a graphGwith an cut edgeuv, letGuv=G− {vw:w∈NG(v)\ {u}}+{uw:w∈NG(v)\ {u}}.

Lemma 2.2([22]). Let Gbe a connected graph and uv be a cut edge of G. If uv is not a pendant edge of G, thens(G)> s(Guv).

(3)

Lemma 2.3([22]). Let T be a tree with u∈V(T)andNT(u) ={u1, . . . , uk}, wherek≥3. For1≤i≤k, let Ti be the component ofT−ucontainingui. Forw∈V(T2), letT0 =T− {uui: 3≤i≤t}+{wui: 3≤i≤t}, where3≤t≤k. If |V(T1)| ≥ |V(T2)|, thens(T0)> s(T).

For integers n, rand t with r ≥t≥0 and r+t+ 2≤n, we denote by Dn(r, t) the caterpillar formed by attaching rpendant edges to one terminal vertex and t pendant edges to the other terminal vertex of a path Pn−r−t. Ifr+t=n−1, thenDn(r, t) is just the starSn. Note thatDn(1,1)∼=Pn ∼=Dn(1,0).

Lemma 2.4([22]). If r+t+ 2≤n andr≥t≥2, then s(Dn(r, t))> s(Dn(r+ 1, t−1)).

Lemma 2.5. If 2≤a≤n−1, thens(Dn(da2e,ba2c)) =bn2−a42+2ac.

Proof. It is evident fora= 2, n−1. Suppose that 3≤a≤n−2.

Note that the diameter of Dn(da2e,ba2c) is n−a+ 1. By Lemma 2.1, the vertex of distance bn−a+12 c from pendant vertex adjacent to a vertex of degreeda2e+ 1 is in the median ofDn(da2e,ba2c). Thus, ifn−ais odd, then

s Dn

la 2 m

,ja 2

k

= 2

n−a+1 2 −1

X

j=1

j+la 2

m·n−a+ 1 2 +ja

2

k·n−a+ 1 2

= n2−a2+ 2a−1

4 ,

and ifn−ais even, then s

Dnla 2 m

,ja 2

k

=

n−a 2 −1

X

j=1

j+

n−a 2

X

j=1

j+la 2

m·n−a 2 +ja

2 k·

n−a 2 + 1

=

n−a 2

2

+a(n−a) 2 +la

2 m·

Thus

s Dnla

2 m,ja

2 k=





n2−a2+2a−1

4 ifn−ais odd

n2−a2+2a

4 ifn−ais even andais even

n2−a2+2a−2

4 ifn−ais even andais odd

=

n2−a2+ 2a 4

,

as desired.

For a nontrivial treeT withu∈V(T) and integersp, q≥1, letTu;p,q be the tree obtained by attaching two pendant paths, one of lengthpand one of lengthq, atu.

Lemma 2.6. Let T be a nontrivial tree withu∈V(T). Ifp≥q+ 2 andq≥1, thens(Tu;p,q)> s(Tu;p−1,q+1).

Proof. LetH =Tu;p,q. Let P =v0. . . vp and Q=w0. . . wq be the two pendant paths at u(=v0=w0) inH. LetH0 =H−vp−1vp+wqvp. Letxbe a vertex in the median ofH. Asp > q, it follows from Lemma2.1that x6∈V(Q)\ {u}.

Case 1.x∈V(T).

By Lemma 2.1, x is also in the median of H0. As we go from H to H0, the distance between xand vp is decreased byp−(q+1) (which is at least 1) and the distance betweenxand any other vertex remains unchanged.

Therefore

s(H0)−s(H) =sH0(x)−sH(x)≤ −1<0, implying thats(H)> s(H0).

(4)

Figure 1.The treeT(r, t).

Case 2.x∈V(P)\ {u}.

Assume that x=vi, and if two vertices in V(P)\ {u} are in the median of H, we choose vi with smaller indexi. AsδH(x) = 2, there are exactly two components inH−x. Letnbe the order ofH. By Lemma2.1, we have dH(x, vp) =bn2c,i.e.,p−i=bn2c. By Lemma2.1,vi−1 is in the median ofH0. Note thatdH0(vi−1, vp)− dH(vi, vp) =i+q−n

2

,dH0(vi−1, z)−dH(vi, z) = 1 forz∈ {vi+1, . . . , vp−1}, anddH0(vi−1, z)−dH(vi, z) =−1 forz∈V(H)\ {vi−1, . . . , vp}. Thus, we have

s(H0)−s(H) =sH0(vi−1)−sH(vi)

= X

z∈V(H0)

dH0(vi−1, z)− X

z∈V(H)

dH(vi, z)

=

i+q−jn 2

k

+ 1·(p−1−i)−1·(n−p−2 +i)

=p+q−n+ 1<0,

implying thats(H)> s(H0).

Let r and t be positive integers. For a nontrivial tree T and a caterpillar C of degrees one or three with a diametral path u1. . . ur+t+1, where vi is the unique neighbor of ui outside path u1. . . ur+t+1 for each 2 ≤ i ≤ r+t, let T(r, t) be the tree of order 2(r+t)−1 +|V(T)| consisting of T and C such that they share a unique common vertexvr+1, see Figure1. Note thatT(r, t)−(V(T)\ {vr+1}) is a caterpillar of degrees one or three with diametral pathu1. . . ur+t+1, wherevi is the unique neighbor ofui outside pathu1. . . ur+t+1

for each 2≤i≤r+t.

Lemma 2.7. Let T be a nontrivial tree. Ifr≥t≥2, then

s(T(r+ 1, t−1))> s(T(r, t)).

Proof. LetH =T(r, t). LetT1 andT2 be the components ofG−ur+1 containingu1 andur+t+1, respectively.

Let

H0=H− {ur+1vr+1, ur+2vr+2}+{ur+1vr+2, ur+2vr+1}.

ThenH0∼=T(r+ 1, t−1). As|V(T2)|= 2t−1≤2r−1 =|V(T1)|, we havex /∈V(T2) by Lemma2.1.

Case 1.x∈V(T).

By Lemma2.1, x is also in the median of H0. Note that dH0(x, z)−dH(x, z) = 1 for z ∈ V(T1)∪ {ur+1}, dH0(x, z)−dH(x, z) = −1 for z∈V(T2)\ {vr+2}, and the distance betweenxand a vertex of V(T)∪ {vr+2} remains unchanged. Thus

s(H0)−s(H) =sH0(x)−sH(x)

= 1·(|V(T1)|+ 1)−1·(|V(T2)| −1)

=|V(T1)| − |V(T2)|+ 2>0, implying thats(H0)> s(H).

(5)

Case 2.x∈V(T1), orx=ur+1 and|V(T1)|=bn2c,bn2c −1.

By Lemma2.1,xis also in the median ofH0. As we go fromH toH0, the distance betweenxand a vertex ofV(T) is increased by 1, the distance betweenxandvr+2 is decreased by 1, and the distance betweenxand any other vertex remains unchanged. Thuss(H0)−s(H) =|V(T)| −1>0,i.e.,s(H0)> s(H).

Case 3.x=ur+1 and|V(T1)| ≤ bn2c −2.

By Lemma 2.1, ur+2 is a vertex in the median of H0. Note that dH0(ur+2, z)−dH(ur+1, z) = 1 for z ∈ V(T1), dH0(ur+2, z)−dH(ur+1, z) = −1 for z ∈ V(T2)\ {ur+2, vr+2}, and dH0(ur+2, z) = dH(ur+1, z) for z∈V(T)∪ {vr+2}. Thus

s(H0)−s(H) =sH0(ur+2)−sH(ur+1) =|V(T1)| −(|V(T2)| −2)>0,

implying thats(H0)> s(H).

Lemma 2.8. Let T be a caterpillar with the largest minimum status in the class of caterpillars of ordernwith t ≥ 1 vertices of degree two and maximum degree three, or in the class of caterpillars of order n with t ≥1 vertices of degree two and one or two vertices of maximum degree four, where each vertex of degree four has three pendant neighbors. LetU be the set of vertices of degree two inT. ThenU induces a path inT.

Proof. Suppose thatU does not induce a path inT. LetP =u0. . . ud be a diametral path ofT. Then there are three vertices ui, uj, u` with 1 ≤i < j < ` ≤d−1 withδT(uj) = 3 and ui, u` ∈ U. Let T1 and T2 be the nontrivial components ofT −uj containingui andu`, respectively. Letvj be the pendant neighbor ofuj. Assume that|V(T1)| ≥ |V(T2)|. LetT0 =T−ujvj+u`vj. By Lemma2.3,s(T0)> s(T), a contradiction.

Fora, b≥0 with 2(a+b+ 1)≤n, letCn(a, b) be the tree obtained from the pathPn−a−b with consecutive vertices u1, . . . , un−a−b by attaching a pendant vertex vi to ui for each i with 2≤i ≤a+ 1 and each i with n−a−2b≤i≤n−a−b−1. In particular Cn(0,0) is just the path Pn.

Lemma 2.9. If b≥0,a≥b+ 2, and2(a+b) + 2< n, then

s(Cn(a−1, b+ 1))> s(Cn(a, b)).

Proof. LetT =Cn(a, b). LetT0 =T−ua+1va+1+un−a−2b−1va+1. Then T0 ∼=Cn(a−1, b+ 1).

Case 1.2a≥n

2

.

Note thatT−udn−14 e+1contains exactly two nontrivial components of orders 2dn−14 e −1 andn−1−2dn−14 e, respectively. By Lemma2.1,udn−14 e+1is in the median ofT andT0. As we go fromT toT0, the distance between udn−14 e+1 and va+1 is increased by n−a−2b−1− n−1

4

+ 1

− a+ 1−(n−1

4

+ 1)

(which is equal to n−2(a+b)−2, larger than 0), and the distance betweenudn+14 e+1 and any other vertex remains unchanged.

Thuss(T0)> s(T), as desired.

Case 2.2a <n

2

.

Letx=udn2e−a and x0 =udn2e−a+1. By Lemma2.1,xis in the median ofT andx0 is in the median ofT0. LetT1be the component ofT−xcontainingu1, andT2the component ofT−x0 containingun−a−b. Note that

dT0(x0, z)−dT(x, z) =

(1 ifz∈V(T1)\ {va+1}

−1 ifz∈V(T2) and

dT0(x0, va+1)−dT(x, va+1) =n−a−2b−1−ln 2

m−a+ 1

−ln 2

m−a−(a+ 1)

=n−2ln 2 m

+ 2(a−b)−1≥2.

(6)

If 2a+ 1 =n

2

, thenx=ua+1,va+16∈V(T1), and thus s(T0)−s(T) =sT0(x0)−sT(x)

=|V(T1)| − |V(T2)|+dT0(x0, va+1)−dT(x, va+1)

≥ln 2

m−2

−jn 2

k−1

+ 2>0, and otherwise, we have

s(T0)−s(T) =sT0(x0)−sT(x)

=|V(T1)| −1− |V(T2)|+dT0(x0, va+1)−dT(x, va+1)

≥ln 2

m−1

−1−jn 2

k−1

+ 2>0.

It follows thats(T0)> s(T), as desired.

3. Minimum status and diameter

In this section, we give sharp lower and upper bounds for the minimum status of a tree with fixed diameter, and characterize the unique trees achieving these bounds.

LetFn,dbe the caterpillar obtained by attachingn−d−1 pendant vertices tovdd

2eof the pathPd+1=v0. . . vd, where 2≤d≤n−1. DefineFn,n−1=Pn. Particularly,Fn,2=Sn andFn,3=Dn(n−3,1). The lower bound in the following theorem has been obtained in [2] for a connected graph of ordern and diameterd. However, we include a proof here for completeness.

Theorem 3.1. Suppose that T is a tree of order n≥4 with diameterd, where2≤d≤n−1. Then s(T)≥n−1−d+

(d+ 1)2 4

with equality if and only if T ∼=Fn,d.

Proof. Letxbe a vertex in the median ofT andP a diametral path ofT. Then s(T) = X

v∈V(P)

dT(x, v) + X

v∈V(T)\V(P)

dT(x, v)

≥s(P) + X

v∈V(T)\V(P)

1

=n−1−d+

(d+ 1)2 4

with equality if and only if x is in the median of P and all vertices outside P are adjacent to x, i.e.,

T ∼=Fn,d.

Corollary 3.2. Suppose thatT is a tree of order n, and T Sn. Then s(T)≥n with equality if and only if T∼=Dn(n−3,1). Moreover, ifTDn(n−3,1), thens(T)≥n+1with equality if and only ifT∼=Fn,4, Dn(n−4,2).

Proof. Letd be the diameter of T. Letf(d) =n−1−d+j(d+1)2

4

k

for 2 ≤d≤n−1. It is easy to see that f(d+ 1) > f(d) for 2 ≤ d ≤ n−2. As T Sn, we have d ≥ 3. If d ≥ 4, then by Theorem 3.1, we have s(T)≥f(d)≥f(4) =n+ 1 with equalities if and only ifT ∼=Fn,4. Ifd= 3, thenT ∼=Dn(r, n−2−r) for some rwith n−22 ≤r≤n−3, and in this case, s(T) = 2n−3−r, which isnifr=n−3,n+ 1 ifr=n−4 and at

leastn+ 2 ifr≤n−5.

(7)

Theorem 3.3. Suppose that T is a tree of order n≥4 with diameterd, where2≤d≤n−1. Then s(T)≤

d(2n−d) + 1 4

with equality if and only if T ∼=Dn(dn−d+12 e,bn−d+12 c).

Proof. By Lemma2.5, we haves(Dn(dn−d+12 e,bn−d+12 c)) =bd(2n−d)+14 c.

LetT be a tree of ordernwith diameterdsuch that its minimum status is as large as possible. By the value ofs(Dn(dn−d+12 e,bn−d+12 c)), we only need to show thatT ∼=Dn(dn−d+12 e,bn−d+12 c).

It is trivial ifd= 2, n−1.

Suppose that 3≤d≤n−2. Letαbe the number of quasi-pendant vertices inT. Then there are at least two pendant vertices with different neighbors,i.e., α≥2.

We claim thatα= 2. Otherwise,α≥3. LetP =v0. . . vdbe a diametral path ofT. Asv0 andvdare pendant vertices,v1andvd−1are quasi-pendant vertices inT. Sinceα≥3,δT(vi)≥3 for someiwith 2≤i≤d−2. For z∈NT(vi), letTzbe the component ofT−vicontainingz. Letwbe a neighbor ofvioutsideP such that the order ofTwis maximum among the components ofT−viexceptTvi−1 andTvi+1. Letρ= max{dT(vi, x) :x∈V(Tw)}.

Asd=dT(v0, vd) =dT(vi, v0) +dT(vi, vd), we haveρ≤dT(vi, v0) =iandρ≤dT(vi, vd) =d−i. Assume that

|V(Tvi−1)| ≥ |V(Tvi+1)|. Suppose thatρ < d−i. LetT0 =T−viw+vi+1w. ThenP is still a diametral path of T0,i.e.,T0 is a tree of ordernwith diameter d. By Lemma2.3,s(T0)> s(T), a contradiction. Thusρ=d−i.

Letxbe a vertex in the median ofT.

Suppose that|V(Tvi−1)| ≥ n2. LetT0be the tree obtained fromT−viw by deleting all edges inTwand adding all edges in {vd−1z : z ∈V(Tw)}. By Lemma 2.1, xmay be chosen in V(Tvi−1), and xis also in the median ofT0. As we go fromT toT0, the distance betweenxand any vertex different from wis increased or remains unchanged. Thuss(T0)−s(T) =sT0(x)−sT(x)≥dT0(x, w)−dT(x, w) =d−i−1>0, implying thats(T0)> s(T), a contradiction. It follows that |V(Tvi−1)|< n2. Similarly, if |V(Tw)| ≥ n2, then, as above, we may form a tree T0 from T−vivi+1 by deleting all edges inTvi+1 and adding all edges in {v1z:z∈V(Tvi+1)}, andx∈V(Tw) is in the median ofT and T0, such thats(T0)−s(T)≥dT0(x, vi+1)−dT(x, vi+1) =i−1>0, a contradiction.

So|V(Tw)|< n2. By Lemma2.1,x=vi.

Leta=|V(Tvi−1)|andc=|V(Tw)|. LetV0=V(Tw) ifa+c≤ bn2cand letV0be a subset ofV(Tw) consisting of bn2c −avertices otherwise. Let T00 be the tree obtained from T −viw by deleting all the edges in Tw and adding all edges in{v1z:z∈V0} ∪ {vd−1z:z∈V(Tw)\V0}. Thenviis also in the median ofT00by Lemma2.1.

This is true fora+c >bn2cbecause froma+c+|V(Tvi+1)| ≤n−1 we have|V(Tvi+1)|+a+c− bn2c ≤ n2. As we go from T to T00, the distance between vi and any vertex different from w is increased or remains unchanged. Thuss(T00)−s(T) =sT00(vi)−sT(vi)≥dT00(vi, w)−dT(vi, w) = min{i, d−i} −1 =d−i−1>0, implying thats(T00)> s(T), a contradiction. Now we have proved thatα= 2. That is,T =Dn(p, q) for some p,qwithp+q=n−d+ 1. By Lemma 2.4, we haveT ∼=Dn(dn−d+12 e,bn−d+12 c).

For 3 ≤ d ≤ n−1, let F = Dn(dn−d+12 e,bn−d+12 c) and h(n, d) = s(F). Let xy be the edge of F with δF(x) ≥ 2 and δF(y) = 1 +bn−d+12 c. Then Fxy ∼= Dn(dn−(d−1)+12 e,bn−(d−1)+12 c). By Lemma 2.2, we have h(n, d)> h(n, d−1).

Let Pn,i be the tree obtained from the path Pn−1 =v1. . . vn−1 by attaching a pendant edge at vertex vi, where 2≤i≤ bn2c. Note thatPn,2∼=Dn(2,1).

Corollary 3.4. Suppose thatT is a tree of ordern, andTPn. Thens(T)≤ bn24−3cwith equality if and only ifT ∼=Dn(2,1). Moreover, ifT Dn(2,1), thens(T)≤ bn24−8cwith equality if and only if T ∼=Dn(2,2),Pn,3. Proof. Let d be the diameter of T. Thend ≤ n−2. If d ≤ n−3, then by Theorem 3.3, s(T) ≤ h(n, d) ≤ h(n, n−3) =bn24−8cwith equalities if and only ifT ∼=Dn(2,2). If d=n−2, thenT ∼=Pn,i with 2≤i≤ bn2c, ands(T) is equal tobn24−3cifi= 2,bn24−8cifi= 3, and is at most bn24−8c −1 ifi≥4.

(8)

Note that, ifuandv are two nonadjacent vertices in a connected graphG, thens(G)≥s(G+uv).

Suppose thatGis a unicyclic graph of ordern≥4 with diameterd, where 2≤d≤n−2. By Theorem3.3, s(G)≤

d(2n−d) + 1 4

with equality if and only if Gis isomorphic to a graph with diameterd obtained fromDn(dn−d+12 e,bn−d+12 c) by adding an edge between two pendant vertices.

The eccentricityeG(u) of vertexuin a connected graphGis its distance to a farthest vertex. Ifuis a vertex such thatdG(u, v) =eG(v), then uis called an eccentric vertex ofv. Recall that the radius ofGis defined to be the minimum eccentricities of all vertices of G. Let diam(G) and r(G) be the diameter and radius of G, respectively. A vertex v is central if eG(v) = r(G). Buckley and Lewinter [8] characterized graphs that have diameter-preserving spanning trees. They showed that a connected graphGhas a diameter-preserving spanning tree if either

(1) diam(G) = 2r(G), or

(2) diam(G) = 2r(G) andGcontains a pair of adjacent central verticesxandythat have no common eccentric vertex.

LetGbe a connected graph with diameter-preserving spanning trees, and letdbe the diameter ofG, where 2 ≤d≤n−1. Then s(G) ≤s(T) for a spanning tree T of diameter d ofG. Thus, by Theorem 3.3, we have s(G)≤s(T)≤ bd(2n−d)+14 cwith equality ifG∼=Dn(dn−d+12 e,bn−d+12 c).

4. Minimum status and number of pendant vertices

In this section, we give sharp lower and upper bounds for the minimum status of a tree with fixed number of pendant vertices, and characterize the unique trees achieving these bounds.

For 2≤p≤n−1, letP(n, p) be the set of trees of ordernwithppendant vertices. Particularly,P(n,2) ={Pn} andP(n, n−1) ={Sn}.

For 3≤p≤n−1, let Sn,p be the tree withppendant paths of almost equal lengths (i.e.,n−1−pbn−1p c pendant paths of length bn−1p c+ 1 and p+pbn−1p c −(n−1) pendant paths of length bn−1p c) at a common vertex. Particularly,Sn,n−1=Sn. LetSn,2=Pn.

Theorem 4.1. Suppose that T ∈P(n, p), where2≤p≤n−1. Then

s(T)≥n−1−pbn−1p c2−(2n−p−2)bn−1p c 2

with equality if and only if T ∼=Sn,p.

Proof. It is trivial forp= 2, n−1. Suppose that 3≤p≤n−2.

Let xbe the vertex in Sn,p of degree p. By Lemma 2.1, x is in the median of Sn,p. By direct calculation, we have

s(Sn,p) =p

bn−1p c

X

j=1

j+

n−1−p n−1

p

n−1 p

+ 1

=p·(1 +bn−1p c)bn−1p c

2 + (n−1−p) n−1

p

−p n−1

p 2

+n−1

=n−1−pbn−1p c2−(2n−p−2)bn−1p c

2 ·

(9)

Let T be a tree inP(n, p) such that its minimum status is as small as possible. By the value of s(Sn,p), it suffices to show thatT ∼=Sn,p.

Suppose that there are at least two vertices with degree at least three inT. Then we may choose two vertices, sayw1 and w2, inT with degree at least three, such that dT(w1, w2) is as small as possible. Let P =v0. . . v`

be the unique path connectingw1andw2, wherev0=w1 andv`=w2. If the length ofP is at least two, then any internal vertex of P has degree two inT. LetTw1 be the component of T−v1 containingw1 andTw2 the component ofT −v`−1 containingw2.

LetNT(w1)\ {v1}={u1, . . . , us}. Fori= 1, . . . , s, letTui be the component ofT−w1containingui. Assume that|V(Tu1)| ≤ · · · ≤ |V(Tus)|. Denote a=Ps

i=2|V(Tui)|.

Assume that |V(Tw1)| ≤ |V(Tw2)|. Let T0 = T − {w1ui : i = 2, . . . , s}+{w2ui : i = 2, . . . , s}. Note that the pendant vertices ofT0 are just pendant vertices of T, i.e., T0 ∈P(n, p). Letxbe a vertex in the median of T. Then x∈V(Tw2)∪n

vb`

2c, . . . , v`−1o

, which follows from Lemma2.1and the fact that |V(Tw2)|+d`2e ≥ n2 if`≥2, and follows from Lemma2.1if`= 1.

Suppose thatx∈V(Tw2). By Lemma 2.1,xis also in the median ofT0. As we go fromT toT0, the distance betweenxand a vertex of ∪si=2V(Tui) is decreased bydT(w1, w2), and the distance betweenxand any other vertex remains unchanged. Thus

s(T0)−s(T) =sT0(x)−sT(x) =−dT(w1, w2)·a <0,

implying thats(T0)< s(T), a contradiction. It follows that x=vi for some iwithb2`c ≤i≤`−1.

Note thats≥2 and|V(Tw1)|= 1 +Ps

j=1|V(Tuj)| ≥ |V(Tu1)|+s. Thus, we have |V(Tu1)| ≤ |V(Tw2)| −2, implying that|V(Tu1)| − |V(Tw2)|+ 1<0.

Suppose that a ≤ `−i−1, i.e., i+a ≤`−1. Then ` ≥ 2. By Lemma 2.1, vi+a is in the median of T0. Note that

dT0(vi+a, z)−dT(vi, z) =





a ifz∈ {v0, . . . , vi−1} ∪V(Tu1)

−a ifz∈V(Tw2)∪ {vi+a+1, . . . , v`−1}

`−2i−a ifz∈V(Tu2)∪. . .∪V(Tus) and

i+a−1

X

j=i+1

dT0(vi+a, vj) =

i+a−1

X

j=i+1

dT(vi, vj).

Thus

s(T0)−s(T) =sT0(vi+a)−sT(vi)

=a(i+|V(Tu1)|)−a(|V(Tw2)|+`−i−a−1) + (`−2i−a)a

=a(i+|V(Tu1)| − |V(Tw2)| −`+i+a+ 1 +`−2i−a)

=a(|V(Tu1)| − |V(Tw2)|+ 1)<0,

implying thats(T0)< s(T), a contradiction. It follows thata≥`−i, where`≥1. Thenw2is in the median of T0 by Lemma2.1. Note that

dT0(w2, z)−dT(vi, z) =





`−i ifz∈ {v0, . . . , vi−1} ∪V(Tu1)

−(`−i) ifz∈V(Tw2)\ {w2}

−i ifz∈V(Tu2)∪. . .∪V(Tus) and

`−1

X

j=i+1

dT0(w2, vj) =

`−1

X

j=i+1

dT(vi, vj).

(10)

Thus

s(T0)−s(T) =sT0(w2)−sT(vi)

= (`−i)(i+|V(Tu1)|)−(`−i)(|V(Tw2)| −1)−ai

≤(`−i)(i+|V(Tu1)| − |V(Tw2)|+ 1−i)

= (`−i)(|V(Tu1)| − |V(Tw2)|+ 1)<0,

implying that s(T0) < s(T), also a contradiction. Therefore, there is exactly one vertex with degree at least three in T. AsT ∈P(n, p), T consists ofp pendant paths at a common vertex. Now by Lemma 2.6, we have

T ∼=Sn,p.

Corollary 4.2. If2≤p≤n−2, thens(Sn,p)> s(Sn,p+1).

Proof. Letxbe the vertex of degree pin Sn,p and xy be an edge of a longest pendant path in Sn,p, where if p= 2,xis any vertex of degree two andxyis in the longer sub-path with a terminal vertexx. By Lemma2.2, s(Sn,p)> s

(Sn,p)xy

. Note that (Sn,p)xy ∈P(n, p+ 1). By Theorem4.1,s

(Sn,p)xy

≥s(Sn,p+1). It follows

thats(Sn,p)> s(Sn,p+1).

Theorem 4.3. Suppose that T ∈P(n, p), where2≤p≤n−1. Then s(T)≤

n2−p2+ 2p 4

with equality if and only if T ∼=Dn(dp2e,bp2c).

Proof. By Lemma2.5, we haves(Dn(dp2e,bp2c)) =bn2−p42+2pc.

LetT be a tree inP(n, p) such that its minimum status is as large as possible. From the value of the status of Dn(dp2e,bp2c), it suffices to show thatT ∼=Dn(dp2e,bp2c). It is trivial forp= 2, n−1. Suppose that 3≤p≤n−2.

Letαbe the number of quasi-pendant vertices inT. Since the diameter ofT is at least three, we haveα≥2.

Suppose thatα≥3. Then there are at least three components, sayTv, Tw andTz, inT−ufor some vertexu, and at least two of them are not nontrivial, wherev, w, z ∈NT(u). Assume that|V(Tv)| ≥ |V(Tw)| ≥ |V(Tz)|.

Then Tv andTw are nontrivial. Let u1 be a quasi-pendant vertex of Tw. ThenT0 =T−uz+u1z is a tree in P(n, p). By Lemma2.3,s(T0)> s(T), a contradiction. Therefore, we haveα= 2. That is,T ∼=Dn(`1, `2), where

`1≥`2≥1 and`1+`2=p. By Lemma2.4, we haveT ∼=Dn(dp2e,bp2c).

We note that Theorem 4.3 follows also from Theorem 3.3. Suppose that T ∈ P(n, p), where 2 ≤ p ≤ n−1. Let d be the diameter ofT. As a diametral path contains exactly two pendant vertices, we have d≤ n−p+ 1. By Theorem 3.3, s(T) ≤ h(n, d) ≤ h(n, n−p+ 1) = bn2−p42+2pc with equalities if and only if T ∼=Dn(dn−d+12 e,bn−d+12 c) withd=n−p+ 1,i.e., T ∼=Dn(dp2e,bp2c).

5. Minimum status and number of odd vertices

In this section, we give sharp lower and upper bounds for the minimum status of a tree with fixed number of odd vertices, and characterize the unique trees achieving these bounds.

For integersnandkwith 1≤k≤ bn2c, letO(n, k) be the set of trees of ordernwith 2kodd vertices [23].

Theorem 5.1. Suppose that T ∈O(n, k), where1≤k≤ bn2c. Then s(T)≥n−1−k

n−1 2k

2

+ (n−k−1) n−1

2k

with equality if and only if T ∼=Sn,2k, where ifk=n2, thenSn,n=Sn.

(11)

Proof. The casek= 1 is trivial as in this case,T ∼=Pn∼=Sn,2. The casek=bn2cis also trivial asSn∼=Sn,2bn

2c

is the unique tree of ordernwith smallest minimum statusn−1.

Suppose that 2≤k <bn2c,i.e., 4≤2k≤n−2. Letpbe the number of pendant vertices ofT. Thenp≤2k.

By Theorem4.1and Corollary4.2, we have

s(T)≥s(Sn,p)≥s(Sn,2k) =n−1−k n−1

2k 2

+ (n−k−1) n−1

2k

with equalities if and only ifT ∼=Sn,p andp= 2k,i.e.,T ∼=Sn,2k. Theorem 5.2. Suppose that T ∈O(n, k), where1≤k≤ bn2c. Then

s(T)≤





j(n+1−k)2

4

k +j

2nk−3k2+6k−2n 4

k

ifkis odd j(n+1−k)2

4

k +j

2nk−3k2+6k−2n 4

k−1 ifkis even with equality if and only if T ∼=Cn(dk−12 e,bk−12 c).

Proof. Let a = dk−12 e, b = bk−12 c and c = dn+1−k2 e. Then a + b = k −1 and a = b, b + 1. Let H =Cn(a, b), whose vertices are labelled as before. Let x=vc. Thenxis in the median of H by Lemma 2.1.

LetU ={ui:i= 1, . . . , n+ 1−k}andW ={vi:i= 2, . . . , a+ 1, n+ 1−k−b, . . . , n−k}. By direct calculation, we have

X

u∈U

d(x, u) =s(Pn+1−k) =

(n+ 1−k)2 4

and

X

u∈W

d(x, u) =

a

X

i=1

(c−1−a+i) +

b

X

i=1

(n+ 1−k−b−c+i)

=a(c−1−a) +a(a+ 1)

2 + (n+ 1−k−b−c)b+b(b+ 1) 2

=

(a(n−k−2a) +a(a+ 1) ifa=b (a−1)(n+ 1−k−2a) +c−1−a−a2 ifa=b+ 1

=

((k−1)(2n−3k+3)

4 ifkis odd,

(k−2)(2n+2−3k)

4 +dn+1−k2 e −1 ifkis even.

Thus

s(H) =X

u∈U

d(x, u) + X

u∈W

d(x, u)

=





j(n+1−k)2

4

k +j

2nk−3k2+6k−2n 4

k

ifkis odd, j(n+1−k)2

4

k +j

2nk−3k2+6k−2n 4

k−1 ifkis even.

Let T be a tree in O(n, k) such that its minimum status is as large as possible. By the value of s(Cn(dk−12 e,bk−12 c)), it suffices to show thatT ∼=Cn(dk−12 e,bk−12 c).

Ifk= 1, then it is obvious thatT ∼=Pn∼=Cn(0,0).

Suppose that k ≥ 2. We claim that the maximum odd degree of T is 3. Otherwise, δT(u) = 2t+ 1 from some u ∈ V(T) and t ≥ 2. Let NT(u) = {u1, . . . , u2t+1}. Let Ti be the component of T −u containing ui,

(12)

where 1 ≤ i ≤ 2t+ 1. Assume that |V(T1)| ≥ |V(T2)|. Let z be a pendant vertex of T in V(T2), and let T0 = T − {uui : 3 ≤ i ≤ 2t}+{zui : 3 ≤ i ≤ 2t}. Note that the degrees of u and z are still odd in T0. ThenT0∈O(n, k). By Lemma2.3,s(T0)> s(T), a contradiction. Thus, the maximum odd degree ofT is 3, as claimed.

Ifk= n2, then by Lemma2.7,T ∼=Cn dk−12 e,bk−12 c .

Suppose thatk < n2. Then there is at least one even vertex inT. We claim that all even vertices have degree two. Otherwise, δT(w) = 2t for some w ∈ V(T), where t ≥ 2. Let NT(w) = {w1, . . . , w2t}. Let Ti be the component ofT−wcontainingwi, where 1≤i≤2t. Assume that|V(T1)| ≥ |V(T2)|. Letzbe a pendant vertex ofT inV(T2), and letT00=T−ww3+zw3. Note that, inT00, the degree ofwis odd and the degree ofzis even.

ThenT00∈O(n, k). By Lemma2.3,s(T00)> s(T), a contradiction. Thus, all even vertices ofT have degree two.

Now, we claim T is a caterpillar. Otherwise, as the maximum degree of T is three, there is a vertex u of degree three inT such thatustill has degree three in the tree obtained fromT by deleting all pendant vertices.

Let NT(u) = {u1, u2, u3}. Then δT(ui) ≥2 for i = 1,2,3. Let Ti be the component of T −ucontaining ui, where i= 1,2,3. LetU be the set of vertices of degree two inT. Suppose that U 6⊆V(Ti) for anyi= 1,2,3, say v1 ∈ V(T1) and v2 ∈ V(T2) for v1, v2 ∈ U. Assume that|V(T1)| ≥ |V(T2)|. Let T = T −uu3+v2u3. AsδT(u) = 2 andδT(v2) = 3, we haveT∈O(n, k). By Lemma2.3,s(T)> s(T), a contradiction. Therefore, U ⊆ V(Ti) for some i = 1,2,3, say U ⊆ V(T1). Suppose that V(T2) (V(T3), respectively) contains r (t, respectively) vertices of degree three of T. Then T ∼=T1(r+ 1, t+ 1)∈ O(n, k), where r, t ≥1. Assume that r ≥ t. Note that T1(r+ 2, t) ∈ O(n, k). By Lemma 2.7, s(T1(r+ 2, t)) > s(T), a contradiction. Thus, T is a caterpillar, as claimed. By Lemma2.8, the set of all n−2k vertices of degree two induces a path, and thus T ∼=Cn(a, b) for someaandb witha+b=k−1. By Lemma2.9, we haveT ∼=Cn(dk−12 e,bk−12 c).

A vertex in a graph is called a branching vertex if its degree is at least three. LetT be a tree of ordernwith kbranching vertices. Letpbe the number of pendant vertices inT. Thenp+ 2(n−p−k) + 3k≤2(n−1),i.e., p≥k+ 2. This implies that 2k+ 2≤k+p≤n, and thus,k≤n2 −1.

Corollary 5.3. Suppose that T is a tree of order n with k branching vertices, and 0 ≤ k ≤ n2 −1. Then s(T)≤s(Cn(dk2e,bk2c))with equality if and only if T ∼=Cn(dk2e,bk2c).

Proof. Ifk= 0, then the result follows from the known fact thatPnis the unique tree of ordernwhose minimum status is maximum [2].

Suppose thatk≥1. LetT be a tree of ordernwithkbranching vertices such that its minimum status is as large as possible.

Let ∆ be the maximum degree ofT. Suppose that ∆≥4. Letu∈V(T) andNT(u) ={u1, . . . , u}. LetTi

be the component ofT−ucontainingui, where 1≤i≤∆. Assume that|V(T1)| ≥ |V(T2)|. Letwbe a pendant vertex ofT in V(T2), and letT0 =T−uu3+wu3. Then T0 is a tree of ordernwith kbranching vertices. By Lemma 2.3, s(T0)> s(T), a contradiction. Hence ∆ = 3. Let pbe the number of pendant vertices in T. As p+ 2(n−k−p) + 3k= 2(n−1), we havep=k+ 2, and thusT is a tree of ordernwith 2k+ 2 odd vertices.

By Theorem5.2, we haveT ∼=Cn(dk2e,bk2c).

6. Minimum status and number of vertices of degree two

For integers nand t with 0≤t ≤n−2, let H(n, t) be the set of trees of ordern with t vertices of degree two. Note thatH(n, n−2) ={Pn},H(n, n−3) =∅, andH(n,0) is the class of series-reduced trees of ordern.

So we only consider trees inH(n, t) with 0≤t≤n−4.

Theorem 6.1. Suppose that T ∈H(n, t), where0≤t≤n−4. Then

s(T)≥n−1−(n−t−1)bn−t−1n−1 c2−(n+t−1)bn−t−1n−1 c 2

with equality if and only if T ∼=Sn,n−t−1.

(13)

Proof. The caset= 0 is trivial as in this caseSn(∼=Sn,n−1) is the unique tree of ordernwith smallest minimum statusn−1.

Suppose thatt≥1. Letpbe the number of pendant vertices ofT. As there is a vertex with degree at least 3, we havep≤n−t−1. By Theorem4.1and Corollary4.2, we have

s(T)≥s(Sn,p)≥s(Sn,n−t−1) =n−1−(n−t−1)bn−t−1n−1 c2−(n+t−1)bn−t−1n−1 c 2

with equalities if and only ifT ∼=Sn,p andp=n−t−1,i.e.,T ∼=Sn,n−t−1. Lemma 6.2. Let T be a tree in H(n, t) such thats(T)is maximum, where0≤t≤n−4. Then the maximum degree ofT is at most 4,T is caterpillar, and there are at most two vertices of degree four in T.

Proof. Let ∆ be the maximum degree of T. Suppose that ∆ ≥5. Let u∈ V(T) withδT(u) = ∆(T) and let NT(u) ={u1, . . . , u}. For 1≤i≤∆, letTi be the component of T−ucontainingui. Suppose without loss of generality|V(T1)| ≥ |V(T2)|. LetT0 =T − {uui : 4≤i≤∆}+{wui : 4≤i≤∆}, wherew is a pendant vertex ofT in V(T2). AsδT0(u) = 3, δT0(w) = ∆−2≥3 and δT0(v) =δT(v) for v∈ V(T)\ {u, w}, we have T0∈H(n, t). By Lemma2.3,s(T0)> s(T), a contradiction. Therefore ∆≤4.

Suppose that T is not a caterpillar. Then for some vertex uof T, T −u has three nontrivial components.

Note thatδT(u) = 3,4 as ∆≤4.

Case 1.δT(u) = 3.

Let NT(u) = {u1, u2, u3}. Let Ti be the component of T −u containing ui for i = 1,2,3. Assume that

|V(T1)| ≥ |V(T2)| ≥ |V(T3)|. Suppose first thatT has a vertexwof degree two inV(T2). LetT0=T−uu3+wu3. As δT0(u) = 2,δT0(w) = 3 and δT0(z) = δT(z) forz ∈V(T)\ {u, w}, we have T0 ∈ H(n, t). By Lemma 2.3, s(T0)> s(T), a contradiction. Therefore,T has no vertex of degree two inV(T2). Letw1be a pendant vertex of T inV(T2) such thatdT(u, w1) = max{dT(u, s) :s∈V(T2)}. LetT0 =T− {uu3, w1w2}+{uw1, w2u3}, where w2is the neighbor of w1. Note thatT0∈H(n, t).

Letxbe a vertex in the median ofT. As|V(T1)| ≥ |V(T2)| ≥ |V(T3)|, we havex∈V(T1)∪{u}by Lemma2.1.

Case 1.1.x∈V(T1).

By Lemma 2.1,xis also in the median of T0. As we go from T to T0, the distance betweenxand a vertex ofV(T3) is increased bydT(u, w2), the distance betweenxandw1is decreased by dT(u, w2), and the distance betweenxand any other vertex remains unchanged. Thus

s(T0)−s(T) =sT0(x)−sT(x)

=dT(u, w2)|V(T3)| −dT(u, w2)

=dT(u, w2)(|V(T3)| −1)>0, implying thats(T0)> s(T), a contradiction.

Case 1.2.x=u

Letx0 be a vertex in the median ofT0. By Lemma2.1, x0 lies on the path connectinguandw2 inT0. Note that dT0(x0, s)−dT(x, s) = dT(u, x0) fors ∈ V(T1), dT0(x0, s)−dT(x, s)≥ −dT(u, x0) for s ∈ V(T2)\ {w1}, dT0(x0, s)−dT(x, s) =dT(w2, x0) fors∈V(T3), anddT0(x0, w1)−dT(x, w1) =−dT(w2, x0). Thus

s(T0)−s(T) =sT0(x0)−sT(x)

≥dT(u, x0)(|V(T1)| − |V(T2)|+ 1) +dT(x0, w2)(|V(T3)| −1)

≥dT(u, x0) +dT(x0, w2)>0, a contradiction.

(14)

Case 2.δT(u) = 4.

LetNT(u) ={u1, u2, u3, u4}. Let Ti be the component ofT −ucontainingui fori= 1, . . . ,4. Assume that T1,T2andT3are nontrivial, and|V(T1)| ≥ |V(T2)|. Suppose thatT has a vertex, sayw, of degree two inV(T2).

LetT0=T− {uu3, uu4}+{wu3, wu4}. AsδT0(u) = 2,δT0(w) = 4 andδT0(v) =δT(v) for anyv∈V(T)\ {u, w}, we have T0 ∈H(n, t). By Lemma 2.3, s(T0)> s(T), a contradiction. Thus, T has no vertex of degree two in V(T2). Then there is a vertexz∈V(T2) such thatδT(z) = 3,4. LetT0=T−uu3+zu3. ThenT0∈H(n, t). By Lemma2.3, s(T0)> s(T), a contradiction.

By combining Cases 1 and 2, we conclude thatT is a caterpillar.

Suppose that there are three vertices, sayv1, v2, andv3 inT with degree four. AsT is a caterpillar,v1, v2

and v3 lie on a diametral path of T, and so one of them, say v3, lies on the path P connecting the other two verticesv1andv2. LetTi be the component ofT−v3containingvi fori= 1,2. Assume that|V(T1)| ≥ |V(T2)|.

Letzbe a neighbor ofv3 outside the pathP inT. LetT00=T−v3z+v2z. ThenT00∈H(n, t). By Lemma2.3, s(T00)> s(T), a contradiction. Therefore, there are at most two vertices of degree four inT.

Lemma 6.3. LetT ∈H(n, t)such thats(T)is maximum, where0≤t≤n−4. Letki be the number of vertices of degreei inT, where1≤i≤4. Then





k1=n−t+32 k3=n−t−52 k4= 1

ifn−t is odd,

and





k1=n−t+22 k3=n−t−22 k4= 0

or





k1=n−t+42 k3=n−t−82 k4= 2

ifn−tis even.

Proof. Trivially, k2 = t. By Lemma 6.1, the maximum degree of T is at most 4 and k4 ≤ 2. Now the result follows from the facts thatk1+k2+k3+k4=nandk1+ 2k2+ 3k3+ 4k4= 2(n−1).

Lemma 6.4. Let T ∈H(n, t) such thats(T)is maximum, where 0 ≤t≤n−4. Ifu∈V(T) withδT(u) = 4, thenT−uhas at most one nontrivial component.

Proof. From Lemma6.1,T is a caterpillar with at most two vertices of degree four. LetNT(u) ={u1, u2, u3, u4}.

Suppose that T−uhas two nontrivial components, sayT1 andT2, withui ∈V(Ti) for i= 1,2. Assume that

|V(T1)| ≥ |V(T2)|. IfT has a vertexwof degree two inV(T2), then by settingT0=T−{uu3, uu4}+{wu3, wu4}, we haveT0∈H(n, t), and by Lemma2.3,s(T0)> s(T), which is a contradiction. ThusT has no vertex of degree two inV(T2). ThenδT(w) = 3,4 for somew∈V(T2). LetT0=T−uu3+wu3. ThenT0∈H(n, t). By Lemma2.3,

s(T0)> s(T), a contradiction.

For nonnegative integersa, band positive integernwith 2(a+b) + 5≤n, letRn(a, b) be the tree of ordern obtained from a path u1. . . un−a−b by attaching pendant vertex v1 to u2, and then attaching pendant vertex vi to ui for each 2≤i≤a+ 2 and eachi withn−a−2b−2≤i≤n−a−b−3. The structure ofRn(a, b) is shown in Figure2.

Lemma 6.5. Fora≥max{b,1} and2(a+b) + 5< n, we have

s(Rn(a−1, b+ 1))> s(Rn(a, b)).

Proof. LetRn=Rn(a, b). LetR0n =Rn−ua+2va+2+un−a−2b−3va+2. ThenRn0 ∼=Rn(a−1, b+ 1).

Références

Documents relatifs

Since there are no binary trees with an even number of vertices, it would perhaps be useful to consider inead binary trees with k leaves (a leaf being a vertex without a

VODEV, Sharp polynomial bounds on the number of scattering poles for metric perturbations of the Laplacian in Rn, Math. VODEV, Sharp bounds on the number of

A numerical approximation K R,∗ num of the Rayleigh conductivity obtained by a direct numerical solution of problem (1.7) using the boundary element code CESC of CERFACS is

tation of P and Q and prove that P and Q are equivalent polyhedra. Obviously, this correspondence between the faces of P and Q preserves the inclusion and

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

We bring bounds of a new kind for kernel PCA: while our PAC bounds (in Section 3) were assessing that kernel PCA was efficient with a certain amount of confidence, the two

Given a subset of k points from S and a family of quadrics having d degrees of freedom, there exists a family of quadrics with d−k degrees of freedom that are passing through

Figure 3: The set P 19 (black and gray), the set of input vertices (gray circles) and the set of output vertices (gray squares)... Note that not every labeling is possible since