• Tidak ada hasil yang ditemukan

Embedding Trees into Dense Graphs and the Tree-Star Ramsey Number

N/A
N/A
Intan Sherlin

Academic year: 2024

Membagikan "Embedding Trees into Dense Graphs and the Tree-Star Ramsey Number"

Copied!
12
0
0

Teks penuh

(1)

Tree Embeddings and Tree-Star Ramsey Numbers

Zilong Yan

Yuejian Peng

January 3, 2022

Abstract

We say that a graphFcan be embedded into a graphGifGcontains an isomorphic copy ofF as a subgraph. Guo and Volkmann [6] conjectured that ifGis a connected graph with at leastnvertices and minimum degree at leastn−3, then any tree withnvertices and maximum degree at mostn−4 can be embedded intoG. In this paper, we give a result slightly stronger than this conjecture and obtain a sufficient and necessary condition that a tree withnvertices and maximum degree at most n−3 can be embedded into a connected graph G with at leastn vertices and minimum degree at leastn−3. Our result implies that the conjecture of Guo and Volkmann is true with one exception.

We also give an application to the Ramsey number of a tree versus a star.

Keywords: Tree, Embedding, Ramsey number, Tree-Star Ramsey number.

1 Introduction

All graphs considered throughout the paper are simple graphs, i.e. without loops and multiple edges. Let V(G) denote the vertex set ofGand letE(G) denote the edge set ofG. Forv∈V(G), letN(v) ={u∈ V(G)|uv∈E(G)},N[v] =N(v)∪{v}, andd(v) =|N(v)|. ForS⊆V(G), denoteN(S) =∪v∈SN(v) and N[S] =N(S)∪S. Let δ(G) =min{d(v)|v∈V(G)}and ∆(G) =max{d(v)|v∈V(G)}. Let Kn denote the complete graph on n vertices andK1,m denote the star withm+ 1 vertices. For graphs GandH, theRamsey NumberR(G, H) is the smallest integerN such that any red-blue-coloring ofE(KN) yields a redGor a blueH. For graphsF andG, we say that an injectionφ:V(F)→V(G) is anembedding ofF intoGif for any edgexyinF,φ(x)φ(y) is an edge inG. We say thatF can be embedded intoGif there is an embedding ofF intoG.

Degree conditions for tree embedding have been studied actively. A well-known conjecture of Erd˝os- S´os states that any tree with n vertices can be embedded into a simple graph with average degree exceedingn−2. In [7], Havet, Reed, Stein and Wood studied the conditions on the maximum degree and the minimum degree of a graph to embed any tree withn vertices, they proposed an interesting conjecture that any tree with n vertices can be embedded into a simple graph with maximum degree more thann+ 1 and minimum degree at leastb2(n+1)3 c. The Loebl-Koml´os-S´os conjecture [5] states that any tree with nvertices can be embedded into a simple graph with median degree at least n+ 1. We

School of Mathematics, Hunan University, Changsha 410082, P.R. China. Email: [email protected].

Corresponding author. School of Mathematics, Hunan University, Changsha, 410082, P.R. China. Email:

[email protected]. Supported in part by National Natural Science Foundation of China (No. 11931002).

arXiv:2112.14955v1 [math.CO] 30 Dec 2021

(2)

study minimum degree conditions to embed trees, and it has natural applications on tree-star Ramsey numbers.

In [8], Parson determined the path-star Ramsey numbers. A key ingredient in the proof is thatPn, a path withnvertices, can be embedded into a graph with minimum degreen−1. This is generalized to the following well-known result (see [3]).

Lemma 1.1 If Gis a graph with minimum degreen−1, then any tree withnvertices can be embedded intoG.

This lemma implies the Ramsey result of Burr [2]: R(Tn, K1,m)≤m+n−1, whereTn is a tree with nvertices and K1,m is the star withm+ 1 vertices. Applying Lemma 1.1, Cockayne [3] improved the upper bound ofR(Tn, K1,m) tom+n−2 for a special class of trees withnvertices and some values ofm andn. Further, Guo and Volkmann [6] generalized the result of Cockayne to any tree withnvertices. A key making the generalization possible is that Guo and Volkmann showed that any tree withnvertices other thanK1,n−1can be embedded into a connected graph with at leastnvertices and minimum degree n−2. They remarked that there are connected graphsGwith at leastnvertices and minimum degree n−3, and trees with n vertices and maximum degreen−3 which cannot be embedded into G. And they proposed the following conjecture.

Conjecture 1 ([6])IfGis a connected graph with at leastnvertices and minimum degree at leastn−3, then any tree withnvertices and maximum degree at mostn−4 can be embedded into G.

We show a result slightly stronger than this conjecture and obtain a sufficient and necessary condition that trees withn vertices and maximum degree at most n−3 can be embedded into a graphG with minimum degree at leastn−3.

Letpandqbe positive integers. LetT(p, q) be a tree with a longest pathv1v2v3v4v5satisfying that d(v3) = 2, andv2andv4havepandqleaves respectively (see Figure 1). The following is the main result

Figure 1: T(p, q) in this paper.

Theorem 1.2 Let G be a connected graph with at least nvertices and minimum degree at least n−3.

Let Tn be a tree withn vertices and∆(Tn)≤n−3. ThenTn can be embedded intoGif and only if the following situations do not happen.

(i)G=Kn−3,n−3 andTn=T(p, n−3−p), wherep≥1.

(3)

(ii) G is a balanced complete (k+ 1)-partite graph with a vertices in each part and ka = n−3, and Tn=T(1, n−4), wherek≥2 is a positive integer.

This result implies that Conjecture 1 holds except in situation (i) in Theorem 1.2. A non-negative integer Nis called a linear combination of two positive integerspandqif there exist non-negative integerskand l such that N =kp+lq. Applying Theorem 1.2, we are going to prove the following tree-star Ramsey numbers in Section 4.

Theorem 1.3 Let Tn be a tree withn vertices,∆(Tn)≤n−3, andTn6=T(p, n−3−p). Ifm+n−3 is not a linear combination ofn−1andn−2, thenR(Tn, K1,m)≤m+n−3.

Theorem 1.4 LetTnbe a tree withnvertices and∆(Tn)≤n−3. Letm=k(n−1)+3and0≤k≤n−5.

(i) If Tn 6=T(p, n−3−p), thenR(Tn, K1,m) =m+n−3.

(ii) If m+n−3 is not a linear combination of n−1, n−2, n−3 +a1, ..., n−3 +ad, where a1, a2, ..., ad are positive divisors of n−3 no less than 3, then R(T(1, n−4), K1,m) = m+n−3.

OtherwiseR(T(1, n−4), K1,m) =m+n−2.

(iii) Letp≥2. Ifm+n−3 is not a linear combination of2n−6,n−1 andn−2, thenR(T(p, n−3− p), K1,m) =m+n−3. Otherwise R(T(p, n−3−p), K1,m) =m+n−2.

The conjecture of Erd˝os-S´os states that any tree withnvertices can be embedded into a graph with average degree exceedingn−2. It is easy to show that a connected graph with average degree greater thann−2 contains a connected subgraph with minimum degree greater than 12(n−2). So if one can characterize what kind of trees withnvertices can be embedded into a connected graphGwith at least nvertices andδ(G)≥ 12(n−1), then these trees will satisfy the conjecture. In general, it is interesting to study what kind ofTncan be embedded into a connected graphGwith at leastnvertices andδ(G)≥t.

In Section 2, we give some crucial lemmas in the proof of Theorem 1.2. The proof of Theorem 1.2 will be given in Section 3. In Section 4, we prove Theorems 1.3 to 1.4.

2 Preparations

Definition 2.1 A labelling {v1, v2, ..., vn} of vertices of a tree with n vertices is called a conventional labelling if for each j ∈[2, n],|N(vj)∩ {v1, ..., vj−1}|= 1. To simplify the notation, we always denote the unique vertex inN(vj)∩ {v1, ..., vj−1}by vj0 for a conventional labelling.

Note that a treeT always has a conventional labelling. We may take any vertex inV(T) as v1 and order the other vertices asv2, ..., vn in increasing order of the distances tov1.

For given graphs F andG, we say that a subset U ⊆V(F) has been embedded into G(or φ is an embedding ofU to G) if an embeddingφ of F[U] into Ghas been established. If U ⊆V(F) has been embedded toG via an embeddingφ, we say thatφcan be extended to a vertex v∈V(F)\U if there existsv0 ∈V(G) such that the extension function ˆφ:U∪ {v} →V(G) defined by ˆφ(u) =φ(u) foru∈U and ˆφ(v) =v0 is an embedding ofF[U∪ {v}] into G.

Lemma 2.2 Let G be a connected graph with at least n vertices. Let {v1, v2, ..., vn} be a conventional labelling ofTn. Let j∈ {2, ..., n}. Letφbe an embedding of {v1, ..., vj−1}intoG.

(i) If

|N(φ(vj0))\ {φ(v1), ..., φ(vj−1)}| ≥1, (1)

(4)

thenφcan be extended to vj.

(ii) Ifδ(G)≥n−1, then φcan be extended tovj.

(iii) Let δ(G) ≥ n−2. If 2 ≤ j ≤ n−1, then φ can be extended to vj. If j = n and there exists k∈ {1,2, ..., n−1} \ {n0} such thatφ(vk)∈/ N(φ(vn0)), thenφcan be extended to vj.

(iv) Letδ(G)≥n−3. If2≤j ≤n−2, thenφcan be extended tovj. Ifj=n−1and|{φ(v1), ..., φ(vn−2)}\

N(φ(v(n−1)0))| ≥2, then φ can be extended tovj. Ifj =n and |{φ(v1), ..., φ(vn−1)} \N(φ(vn0))| ≥3, thenφcan be extended to vj.

Proof of Lemma 2.2. The condition in (i) guarantees that there is a free vertex inN(φ(vj0)) to embed vj, therefore vj can be embedded to a vertex in N(φ(vj0)), and this proves (i). (ii), (iii) and (iv) are

direct implications of (i). 2

Lemma 2.3 ([4]) Let G be a connected graph with minimum degreeδ(G). Then G contains a path of lengthmin{n,2δ(G) + 1}.

Lemma 2.4 Let Gbe a connected graph with minimum degreeδ. LetS0⊆V(G)and let S1be a proper subset of S0 such thatN(S1)⊆S0. If there existsα∈S0 such that N(α)6⊂S0, then there exists a path of length at least|S1|+δ− |S0|+ 1starting from a neighbor αoutsideS0.

Proof of Lemma 2.4. Let α1 ∈ N(α)\S0. Let α1α2...αg be a maximum path outside S0. Since N(S1) ⊆ S0, N(αg) ⊆ S0∪ {α1, α2, ..., αg−1} \S1. Since |N(ag)| ≥ δ, |S0|+g−1− |S1| ≥ δ. So

g≥ |S1|+δ− |S0|+ 1. 2

Lemma 2.5 Let G be a non-complete and connected graph, then there exist α, β, γ ∈V(G) such that αβ, αγ∈E(G)andβγ /∈E(G).

Proof of Lemma 2.5. Take a shortest path between two non-adjacent vertices, then three consecutive

vertices in this path satisfy the condition. 2

3 Embedding trees

We give the proof of Theorem 1.2 below.

Proof of Theorem 1.2. Let Tn be a tree with n vertices. Let P be a longest path of Tn. Let |P| be the number of vertices of path P and i = |P| −2. Since |P| = 3 implies that Tn = K1,n−1, it contradicts ∆(Tn) ≤ n−3. So|P| ≥ 4. For |P| ≥ 5, we label the two end vertices as vn−1 and vn, all other vertices inP as v1, v2, ..., vi such thatP =vn−1v1v2...vivn, i.e. vn−1 is connected tov1,v1 is connected tov2, · · ·, vi−1 is connected to vi, andvi is connected to vn. We order the vertices outside P as {vi+1, ..., vn−2} in non-decreasing order of the distances to P. For example, vi+1 is connected to a vertex inP. Note that{v1, v2, ..., vn−1, vn}is a conventional labelling. We will show that there is no embeddingφ:V(Tn)→V(G) if and only if situation (i) or (ii) in Theorem 1.2 happens.

By Lemma 1.1, we can easily embed{v1, ..., vn−2} intoG. So what we need to work on is to embed v1 andviproperly so that there is a free vertex for us to embedvn−1 in the neighborhood of the vertex v1 embedded to, and there is a free vertex for us to embed vn in the neighborhood of the vertex vi

embedded to.

(5)

By Lemma 2.5, there existu1, u2 and u3 in Gsatisfying thatu2∈N(u1) and u3 ∈N(u2)\N(u1).

Letφ(vj) =uj forj ∈[3]. By Lemma 2.2 (iv),φ can be extended to{v4, ..., vi−1}. Denote φ(vj) =uj

forj∈[i−1]. DenoteKi={φ(v1), ..., φ(vi−1)}.

Case 1. |P| ≥6.

Case 1.1. There exists a vertex, denoted byui, inN(ui−1)\Kisuch that |Ki\N(ui)| ≥2.

In this case, let φ(vi) = ui. Since δ(G) ≥ n−3, u1, u3 ∈/ N(u1), by Lemma 2.2 (iv), φ can be extended to{v1, ..., vn−1}. Further |Ki\N(ui)| ≥2 implies that |{φ(v1), ..., φ(vn−1)} \N(ui)| ≥3, by Lemma 2.2 (iv), there exists an embedding fromTn intoG.

Case 1.2. There exists a vertex, denoted by ui, in N(ui−1)\Ki such that |Ki\N(ui)| = 1, and

|Ki\N(u)| ≤1 for anyu∈N(ui−1)\Ki.

Note that|Ki\N(ui)|= 1 implies that there exists a vertexuxin Ki such that uxui∈/E(G).

If d(ui) ≥ n−2, let φ(vi) = ui. Since u3 ∈/ N(u1), by Lemma 2.2 (iv), φ can be extended to {vi+1, ..., vn−1}. Since ux, ui ∈/ N(ui), |N(ui)\ {φ(v1), ..., φ(vn−1)}| ≥ 1 and there is at least one free vertex in N(ui) to embedvn. So it is sufficient to consider d(ui) =n−3. Let K =Ki∪N[ui], then

|K|=n−1.

Claim 3.1 N(u1)⊆K if Tn cannot be embedded into G.

Proof of Claim 3.1. Suppose that there exists a vertex denoted byun−1inN(u1)\K. Sinceu3∈/N(u1), by Lemma 2.2 (iv),φcan be extended to{v1, ..., vn−1}. Furthermore, we can guarantee that some vertex va∈Tn\ {vn}is embedded toun−1, wherea∈[i+ 1, n−1] (This is always possible since we can always embedvn−1 toun−1ifun−1is not previously used). Nowux, ui, un−1∈/ N(ui), then there exists a free vertex inN(ui)\ {φ(v1), ..., φ(vn−2), φ(vn−1)} to embedvn andTn can be embedded intoG. 2 Claim 3.2 N(u2)⊆K if Tn cannot be embedded into G.

Proof of Claim 3.2. If there exists a vertexu01∈N(u2)\K, then reassignφ(v1) =u01 and letφ(vi) =ui. By the choice ofu01, we know thatui∈/ N(φ(v1)). By Lemma 2.2 (iv),φcan be extended to{v1, ..., vn−1}.

Recall that there exists a vertex ux ∈ Ki such that uxui ∈/ E(G). We claim that ux 6= u1. Since d(u1)≥n−3,u1, u3∈/ N(u1),N(u1)⊆K and|K|=n−1,u1ui ∈E(G). Sinceux, u01∈/ N(ui), there is at least one free vertex inN(ui)\ {φ(v1), ..., φ(vn−1)}to embedvn. A contradiction. 2 Claim 3.3 N(ut)⊆K for1≤t≤i−1 ifTn cannot be embedded into G.

Proof of Claim 3.3. Use induction on t. Fort= 1 or 2, it is guaranteed by Claim 3.2. Let 3≤t≤i−1.

We assume thatN(us)⊆Kfor 1≤s≤t−1. IfN(ut)6⊂K, then there exists a vertexu0t−1∈N(ut)\K.

By Lemma 2.4, there exists a pathu0t−1u0t−2...u01outside K. Reassign φ(vα) =u0α for 1≤α≤t−1 and letφ(vi) =ui. Sinceu0α∈/N(ui) for 1≤α≤t−1 andt≥3, by Lemma 2.2 (iv),φcan be extended to

{vi+1, ..., vn}. 2

Claim 3.4 N(K)⊆K ifTn cannot be embedded into G.

Proof of Claim 3.4. If there exists a vertexu0i−1 ∈N(ui)\Ki such that there exists a vertexu0i−2 ∈ N(u0i−1)\ K, by Lemma 2.4, there exists a path u0i−2u0i−3...u01 outside K. Reassign φ(vα) = u0α for

(6)

1≤α≤i−1 and letφ(vi) =ui. Sincei≥4 guarantees thatu01, u02∈/N(ui), by Lemma 2.2 (iv),φcan

be extended to{vi+1, ..., vn}. 2

By Claim 3.4, if Tn cannot be embedded into G, then K is a component with n−1 vertices, contradicting thatG is a connected graph with at leastn vertices. SoTn can be embedded intoGin this case.

Case 1.3. For each vertexuin N(ui−1)\Ki,|Ki\N(u)|= 0.

Take a vertex, denote byui, inN(ui−1)\Ki, and letφ(vi) =ui.

If d(ui) ≥ n−1, since u1, u3 ∈/ N(u1), by Lemma 2.2 (iv), φ can be extended to an embedding on {v1, ..., vn−1}. Since d(ui) ≥ n−1 and φ(vi) = ui ∈/ N(ui), there is at least one free vertex in N(ui)\ {φ(v1), ..., φ(vn−1)}to embedvn.

Ifd(ui) =n−2, then|N[ui]|=n−1. We claim that∪u∈N[ui]N[u] =N[ui] ifTn cannot be embedded into G. If there exists a vertex, denote byun−1, inN(u1)\N[ui], letφ(vi) = ui and extend φ to an embedding on {v1, ..., vn−1} such that we embed some va ∈ {vi+1, ..., vn−1} to un−1 (This is always possible since we can always embedvn−1toun−1ifun−1is not previously used). Nowui, un−1∈/ N(ui), therefore there exists one free vertex in N(ui) to embedded vn and Tn can be embedded into G. So we have shown that N[u1] ⊆ N[ui]. Assume that ∪ti=1N[ui] ⊆ N[ui] for 1 ≤ t ≤ i−1, we show that N[ut+1] ⊆ N[ui]. If there exists a vertex u0t ∈ N[ut+1]\N[ui], by Lemma 2.4, there exists a pathu0tu0t−1...u01 outsideN[ui]. Reassign φ(vα) =u0α for 1≤α≤t. Since ui ∈/ N(φ(v1)), by Lemma 2.2 (iv), φ can be extended to an embedding on {v1, ..., vn−1}. Since d(ui) = n−2 and φ(vi) = ui, φ(va)∈/N(ui) for 1≤a≤t, there is at least one free vertex inN(ui)\ {φ(v1), ..., φ(vn−1)}to embedvn. Therefore we have shown thatN[Ki]⊆N[ui]. If there exist verticesu0, u00 such that u0 ∈N(ui)\Ki

and u00 ∈N[u0]\N[ui], since d(u3)≥n−3, |N[ui]| =n−1 and u1, u3 ∈/ N(u3) andN(u3)⊆N[ui], u3u0 ∈ E(G). Reassign φ(v2) = u0 and φ(v1) = u00. Since ui ∈/ N(φ(v1)), by Lemma 2.2 (iv), φ can be extended to {v1, v2, ..., vn−1}. Since u00ui ∈/ E(G) and d(ui) = n−2, there exists a free vertex in N(ui)\ {φ(v1), ..., φ(vn−1)} to embed vn and Tn can be embedded into G. So we have shown that

u∈N[ui]N[u] = N[ui] if Tn cannot be embedded into G. Therefore N[ui] is a component of n−1 vertices, contradicting thatGis a connected graph with at leastnvertices.

Now we discussd(ui) =n−3. Since u1u3∈/E(G) andd(u3)≥n−3, there exists a vertex, denoted by u02, in N(u3)\N[ui]. If there exists a vertex u01 ∈ N(u02)\N[ui], then reassign φ(v2) = u02 and φ(v1) =u01. Sinceφ(v1), φ(v2)∈/N(φ(vi)), by Lemma 2.2 (iv),φcan be extended to{v1, ..., vn}andTn

can be embedded intoG. So N(u02)⊆N[ui]. Sinceu02ui ∈/ E(G) andd(u02)≥n−3,N(u02) =N(ui).

Since u02u1, u02u3 ∈ E(G), we can reassign φ(v2) = u02. Now ui has a non-neighbor in Ki. Note that u3∈/N(φ(v1)) still holds. This situation is exactlyCase 1.2we have proved.

Case 2. |P|= 5.

In this case, P =vn−1v1v2v3vn. IfdTn(v2)≥3, then there exists a vertex v4 connected tov2 (see Figure 2). If dTn(v2) = 2, then NTn(v2) = {v1, v3}. If Tn = P, since |V(G)| ≥ 5 andδ(G) ≥ 2, by Lemma 2.3, P = Tn can be embedded in G. So if dTn(v2) = 2, we may assume that dTn(v1)≥ 3 or dTn(v3)≥3 (see Figure 3).

Letu03be a vertex with maximum degree inV(G). Ifd(u03)≥n−2, if there exists a vertexu02∈N(u03) such that there exists a vertex u01 ∈N(u02)\N[u03], then let φ(v1) = u01, φ(v2) = u02 and φ(v3) =u03. Since u01, u03 ∈/ N(u01) and δ(G) ≥ n−3 and d(u03) ≥ n−2, by Lemma 2.2 (iv), φ can be extended to {v4, ..., vn}. If there is no such vertex u02 in N(u03), then N[u03] is a component, so d(u03) ≥ n−1.

(7)

Figure 2: CasedTn(v2)≥3 Figure 3: CasedTn(v2) = 2

If there exists a vertex u01 ∈ N(u03) such that d(u01) ≥ n−2, let φ(v1) = u01 and φ(v3) = u03 and φ(v2) ∈ N(u01), then by Lemma 2.2 (iv), φ can be extended to{v4, ..., vn}. So we may assume that d(ux) =n−3 for all ux ∈N(u03). Since|N[u03]| ≥n, for any vertexu003 ∈N(u03), we can find vertices u001, u004 ∈ N(u03)\N[u003]. Let φ(v1) = u001 and φ(v2) = u03 and φ(v3) = u003 and φ(v4) = u004. Since φ(v1), φ(v3)∈/ N(φ(v1)),d(φ(v1))≥n−3 andφ(v1), φ(v3), φ(v4)∈/N(φ(v3)), by Lemma 2.2 (iv), φcan be extended to{v4, ..., vn}and Tn can be embedded intoG. Therefore, it is sufficient to consider that Gis an (n−3)-regular connected graph.

Case 2.1. G is an (n−3)-regular connected graph with at least n vertices and dTn(v2) ≥ 3 (see Figure 2).

Claim 3.5 LetGbe (n−3)-regular anddTn(v2)≥3(see Figure 2). If there exist verticesux, uy, ua, ub∈ V(G) such thatuxuy ∈E(G) andua, ub∈N(uy)\N[ux](see Figure 4), then Tn can be embedded into G.

Proof. Letφ(v1) =ua,φ(v2) =uy,φ(v3) =uxandφ(v4) =ub. SinceGis (n−3)-regular,φ(v1), φ(v2)∈/ N(φ(v1)), andφ(v1), φ(v4), φ(v3)∈/ N(φ(v3)), by Lemma 2.2 (iv),φcan be extended to{v5, ..., vn}. 2

Figure 4: Figure for Claim 3.5 Figure 5: Figure for Case 2.1

Take a vertexu03∈V(G). Since|N[u03]|=n−2,N[u03] is not a component. So there exists vertices u01, u02such thatu02∈N(u03) andu01∈N(u02)\N[u03] (see Figure 5). By Claim 3.5,N[u02]⊆N[u03]∪ {u01}.

Ifu01 has two neighbors outside N[u03], then these two neighbors are outside N[u02]. By Claim 3.5, Tn

(8)

can be embedded intoG. If N(u01)⊆N[u03], then N(u01) = N(u03). IfTn cannot be embedded intoG, Claim 3.5 implies that∪u∈N[u0

3]N(u)∪N(u01) =N[u03]∪ {u01}. Otherwise there exists a vertexu∈N(u03) such thatuhas a neighbor woutsideN[u03]∪ {u01}, thenuhas two neighborsw andu01 outsideN[u03], contradicting Claim 3.5. ThusN[u03]∪ {u01} is a component onn−1 vertices, contradicting thatGis a connected graph with at leastnvertices. Therefore, there exists exactly one vertexua ∈N(u01)\N[u03].

By Claim 3.5,N(u01) andN(ua) differ by at most one vertex. SoN(u01)⊆N(u03)∪ {ua}. If there exists a vertexu∈N(u03)∩N(u01) connected to ua, then bothua and u01 are inN(u)\N(u03), contradicting Claim 3.5. SoN(u01) andN(ua) have no common neighbor inN[u03], contradicting Claim 3.5 again. So Tn can be embedded intoG.

Case 2.2. Gis (n−3)-regular anddTn(v2) = 2 (see Figure 3), i.e. Tn=T(p, n−3−p) for somep.

Claim 3.6 Let G be an (n-3)-regular connected graph on at leastnvertices. If there exists u01, u02, u03∈ V(G) such that u01u02, u02u03 ∈ E(G) andu01u03 ∈/ E(G). Then Tn can be embedded into G or N(u01) = N(u03).

Proof. Suppose that N(u01)6=N(u03). SinceGis regular, there exists a vertexu4∈N(u01)\N[u03]. Let φ(v1) = u01, φ(v2) = u02, φ(v3) = u03 and φ(v4) = u4. Since u01u03 ∈/ E(G), and u01, u03, u4 ∈/ N(u03), by

Lemma 2.2 (iv),φcan be extended to an embedding ofTn. 2

Assume thatTn cannot be embedded intoG. Let A1 be a maximal set inGsatisfying that it is an independent set and its any two vertices have a common neighbor. By Claim 3.6, all vertices inA1have the same set of neighbors, sayB. Then|B|=n−3. We claim thatV(G) =A1∪B. Otherwise, there exists a vertexux∈V(G)\(A1∪B) such thatuxis connected to a vertex inB and none of the vertices inA1is adjacent toux. By Claim 3.6,N(ux) =B. This contradicts the maximality ofA1. Since it holds for all such maximal sets, we can divideB intoksetsB1, B2, ..., Bk such thatBj is such a maximal set for eachj ∈ [k]. SinceGis regular, |A1|=|B1|=...=|Bk|=a,n−3 = akand |V(G)|=n−3 +a (See Figure 6). Since|V(G)| ≥n,a≥3.

Figure 6: Ka,a,...,a

If k = 1, since G is (n−3)-regular, G = Kn−3,n−3, and in this case, T(p, n−3−p) cannot be embedded into Kn−3,n−3. This is because that T(p, n−3−p) is a bipartite graph with 2 and n−2 vertices in each of the two parts. This is situation (i) in Theorem 1.2.

Ifk≥2, then let φ(v1)∈B1,φ(v2)∈B2andφ(v3)∈A1.

(9)

Figure 7: T(p, n−3−p)(2≤p≤n−5) Figure 8: T(1, n−4)

Ifd(v1)≥3 andd(v3)≥3 (see Figure 7), letφ(vk), φ(vj)∈A1 andφ(vs), φ(vt)∈B1, then the other leaves can be embedded inA∪B sinceGis a balanced complete (k+ 1)-partite graph, andTn can be embedded intoG. Now considerd(v1) = 2 ord(v3) = 2, i.e. Tn=T(1, n−4). We claim thatT(1, n−4) cannot be embedded intoG. Without loss of generality, assume thatd(v1) = 2 (see Figure 8). Suppose thatT(1, n−4) can be embedded intoGandv3is embedded intoA1 (without loss of generality). Then all n−3 vertices connected to v3 should be embedded outside A1. Since |V(G)\A1| =n−3,φ(v1) must be inA1. Since all vertices ofT(1, n−4)\ {v1, v3} are connected tov1 orv3, then every vertex in T(1, n−4)\ {v1, v3} must be embedded outside A1, i.e. there aren−2 vertices should be embedded outsideA1, but|V(G)\A1|=n−3. This is impossible. So in this caseT(1, n−4) cannot be embedded intoG. This is situation (ii) in Theorem 1.2.

Case 3. |P|= 4.

In this case we label the vertices of Tn as in Figure 9. Note that {v1, v2, ..., vn} is a conventional labelling. Sinced(v1) +d(v2) =nand ∆(Tn)≤n−3,d(v1), d(v2)≥3.

Figure 9: Tn has n-2 leaves Case 3.1. ∆(G)≥n−1.

Let u, v ∈ V(G) such that d(u) = ∆(G) and uv ∈ E(G). Embed φ(v2) = u and φ(v1) = v. By Lemma 2.2 (iv),φcan be extended to{v1, ..., vn−2}. Recall thatvn−1andvn have the same parent (see Figure 9). Sinced(u)≥n−1, there are two free vertices inN(u)\ {φ(v1), ..., φ(vn−2)} to embedvn−1 andvn. SoTn can be embedded inG.

Case 3.2. ∆(G) =n−2.

(10)

Let φ(v2) = u02 such that d(u02) = n−2. We claim that there exists a vertex u01 ∈ N(u02) such that there exists a vertexu03∈N(u01)\N[u02]. Otherwise ∪u∈N[u0

2] =N[u02]. So N[u02] is a component with n−1 vertices, contradicting that Gis a connected graph on at least nvertices. Let φ(v1) = u01 and φ(v3) = u03. By Lemma 2.2 (iv), φ can be extended to{v1, ..., vn−2}. Since d(u02) = n−2 and u03∈/N(u02), there are two free vertices inN(u02)\ {φ(v1), ..., φ(vn−2)}to embedvn−1andvn. SoTncan be embedded intoG.

Case 3.3. Gis (n−3)-regular.

Claim 3.7 If there exists a vertex u02 ∈ V(G) such that there exists a vertex u01 ∈ N(u02) with two neighborsux, uy∈N(u01)\N[u02], thenTn can be embedded intoG.

Proof. Let φ(v1) = u01, φ(v2) = u02, φ(v3) = ux and φ(v4) = uy. By Lemma 2.2 (iv), φ can be extended to{v1, ..., vn−2}. Sinceu02, ux, uy ∈/ N(u02) and d(u02) = n−3, there are two free vertices in N(u02)\ {φ(v1), ..., φ(vn−2)} to embedvn−1 andvn. SoTn can be embedded intoG. 2 Claim 3.8 If Tn cannot be embedded into G, then for any vertex ux ∈ V(G), there exists a vertex uy ∈N(ux)such that uy has exactly one neighbor outside N[ux].

Proof. By Claim 3.7, any vertex inN(ux) has at most one neighbor outsideN[ux]. If∪u∈N[ux]N[u]⊆ N[ux], thenN[ux] is a component onn−2 vertices, a contradiction. 2 If Tn cannot be embedded into G, let ux ∈V(G) and uy ∈ N(ux) such that uz ∈N(uy)\N[ux].

SinceG is (n-3)-regular and uz ∈N(uy)\N[ux], there is exactly one vertex ua ∈N(ux)\N(uy) and N(uy) =N[ux]∪{uz}\{uy, ua}. By Claim 3.7,N[uy]\{ux} ⊆N[uz]. Sinceuyua∈/E(G),|d(ua)|=n−3 and|N[ux]|=n−2, there is exactly one vertexub∈N(ua)\N(ux). Ifub=uz, thenN(uz) =N(ux).

For any vertex u ∈ N(ux), uz ∈ N(u)\N(ux). Then by Claim 3.7, N[ux]∪ {uz} is a component on n−1 vertices, a contradiction. So ub 6= uz. Since ua has exactly one neighbor outside N(ux), N[ua] =N[ux]∪ {ub} \ {uy}. Since all vertices inN(ux) have at most one neighbor outsideN(ux) and uzis a neighbor of all vertices inN(ux)\ {ua}=N(ua)\ {ub, ux} ∪ {uy}, by Claim 3.7,ubis not adjacent to any of these vertices, i.e, ub has no neighbor inN(ua)\ {ub}, contradicting Claim 3.7 again. SoTn

can be embedded intoGin this case. 2

4 Tree-star Ramsey number

In this section, we apply Theorem 1.2 to obtain the Ramsey number of a tree versus a star for some cases.

Proof of Theorem 1.3. For any red-blue-coloring of E(Km+n−3), let GR and GB be the red graph and the blue graph respectively. If ∆(GB) ≥m, then there is a blue K1,m. So we may assume that

∆(GB) ≤ m−1. Hence δ(GR) ≥ (m+n−4)−(m−1) = n−3. Since m+n−3 is not a linear combination ofn−1 andn−2, thenGR contains a component with order at leastn. By Theorem 1.2,

Tn can be embedded intoGR. 2

Fact 4.1 For positive integersmandn,m+n−4is a linear combination ofn−1andn−2andm+n−3 is not a linear combination ofn−1andn−2if and only if m=k(n−1) + 3 and0≤k≤n−5.

(11)

Proof of Fact 4.1. Necessity: Ifm+n−4 =k(n−1) +l(n−2),k≥0 andl≥1, then m+n−3 =k(n−1) +l(n−2) + (n−1)−(n−2) = (k+ 1)(n−1) + (l−1)(n−2),

contradicting that m+n−3 is not a linear combination of n−1 and n−2. Therefore l = 0 and m= (k−1)(n−1) + 3, wherek≥1. Let us rewritem=k(n−1) + 3, where k≥0. Ifk≥n−4, then

m+n−3 =k(n−1) +n= (k−n+ 4)(n−1) + (n−2)(n−2), andm+n−3 is a linear combination of n-1 and n-2. So 0≤k≤n−5.

Sufficiency: If m = k(n−1) + 3 and 0 ≤ k ≤ n−5, then m+n−4 = (k+ 1)(n−1) is a linear combination of n−1 and n−2. If m+n−3 is a linear combination of n−1 and n−2, then m+n−3 =k(n−1) +n=k0(n−1) +l0(n−2) (k0, l0≥0). So (k+ 2−k0)(n−1) = (l0+ 1)(n−2). Since n−1 is relatively prime ton−2,k+ 2−k0=c(n−2) forc≥1. Thereforek≥(n−2)−2 +k0 ≥n−4,

a contradiction. 2

For a positive integer a dividing n−3, let B(n−3 +a) be the balanced complete n−3+aa -partite graph onn−3 +avertices.

Proof of Theorem 1.4. Since m+n−4 = (k+ 1)(n−1), we can partition V(Km+n−4) intok+ 1 disjoint equal parts withn−1 vertices in each part. Color edges in each part red and other edges blue.

Then ∆(GB)≤m+n−4−(n−1) =m−3. So this coloring yields neither a redTn nor a blueK1,m. ThereforeR(Tn, K1,m)≥m+n−3.

(i). By Fact 4.1 and Theorem 1.3,R(Tn, K1,m)≤m+n−3. ThereforeR(Tn, K1,m) =m+n−3.

(ii). For any red-blue-coloring ofE(Km+n−3), letGR andGB be the red graph and the blue graph respectively. If ∆(GB) ≥ m, then there is a blue K1,m. So we may assume that ∆(GB) ≤ m−1.

Hence δ(GR)≥(m+n−4)−(m−1) =n−3. By Fact 4.1, GR contains a component with at least n vertices. By Theorem 1.2, ifm+n−3 is not a linear combination of n−1, n−2,n−3 +a1, ..., n−3 +ad, thenT(1, n−4) can be embedded intoGR. SoR(T(1, n−4), K1,m)≤m+n−3. Therefore R(T(1, n−4), K1,m) =m+n−3.

If m+n−3 =k1(n−1) +l1(n−2) +s1(n−3 +a1) +...+sd(n−3 +ad), for red-blue-coloring ofE(Km+n−3), letGR andGB be the red graph and the blue graph respectively. Let GR=k1Kn−1∪ l1Kn−2∪s1B(n−3 +a1)∪...∪sdB(n−3 +ad) andGB=GcR. By Theorem 1.2,T(1, n−4) cannot be embedded into GR. Sinceδ(GR)≥n−3, then ∆(GB)≤(m+n−4)−(n−3) = m−1. So there is neither a redTn nor a blueK1,m. ThereforeR(T(1, n−4), K1,m)≥m+n−2. The same lower bound was given in [6]. SoR(T(1, n−4), K1,m) =m+n−2.

(iii). For any red-blue-coloring of E(Km+n−3), letGR andGB be the red graph and the blue graph respectively. If ∆(GB)≥m, then there is a blueK1,m. So we may assume that ∆(GB)≤m−1. Hence δ(GR)≥n−3. By Fact 4.1 andm+n−3 is not a linear combination of 2n−6,n−1 and n−2, there exists a componentG0such thatn≤ |V(G0)| 6= 2n−5. By Theorem 1.2,T(p, n−3−p) can be embedded intoG0. ThereforeR(T(p, n−3−p), K1,m) =m+n−3. Ifm+n−3 =r(2n−6) +s(n−1) +t(n−2), then we can partition Km+n−3 into r+s+t disjoint parts such that r parts have order 2n−6 and s parts have order n−1 and t parts have order n−2. Let GR = rKn−3,n−3∪sKn−1∪tKn−2 and GB = GcR. Then ∆(GB) ≤m+n−4−(n−3) = m−1 and T(p, n−3−p) cannot be embedded into G. Therefore R(T(p, n−3−p), K1,m)≥m+n−2. The same lower bound was given in [6]. So

(12)

R(T(1, n−4), K1,m) =m+n−2. 2 Acknowledgement The research is supported in part by National Natural Science Foundation of China (No. 11931002).

References

[1] J. A. Bondy and U. S. R. Murty, Graph theory with applications, The Macmillan Press Ltd., 1976.

[2] S. A. Burr: Generalized Ramsey theory for graphs - a survey, Graphs and Combinatorics, Lecture Notes in Mathematics, 406(1974), 52-75.

[3] E. J. Cockayne: Some tree-star ramsey numbers, J. Combinatorial Theory, Ser. B, 17(1974), 183-187.

[4] G. A. Dirac: Some theorems on abstract graphs, Proc. London Math. Soc., 2(1952), 69-81.

[5] P. Erd˝os, Z. F¨uredi, M. Loebl, and V. T. S´os: Discrepency of trees, Studia Sci. Math. Hungra, 30(1995), 47-57.

[6] Y. Guo and L. Volkmann: Tree-Ramsey numbers, Australasian J. Combin., 11(1995), 169-175.

[7] F. Havet, B. Reed, M. Stein, and D. Wood: A variant of the Erd˝os and S´os conjecture, Journal of Graph Theory, 94(2019), 131-158.

[8] T. D. Parsons: Path-star ramsey numbers, J. Combinatorial Theory, Ser. B, 17(1974), 51-58.

Referensi

Dokumen terkait