This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Degree sequence condition for Hamiltonicity in tough graphs

Songling Shan111Auburn University, Department of Mathematics and Statistics, Auburn, AL 36849. Email: [email protected]. Supported in part by NSF grant DMS-2345869.   Arthur Tanyel222Auburn University, Department of Mathematics and Statistics, Auburn, AL 36849. Email: [email protected].
Abstract

Generalizing both Dirac’s condition and Ore’s condition for Hamilton cycles, Chvátal in 1972 established a degree sequence condition for the existence of a Hamilton cycle in a graph. Hoàng in 1995 generalized Chvátal’s degree sequence condition for 1-tough graphs and conjectured a tt-tough analogue for any positive integer t1t\geq 1. Hoàng in the same paper verified his conjecture for t3t\leq 3 and recently Hoàng and Robin verified the conjecture for t=4t=4. In this paper, we confirm the conjecture for all t4t\geq 4.

Keywords. Degree sequence; Hamiltonian cycle; Toughness

1 Introduction

Graphs considered in this paper are simple, undirected, and finite. Let GG be a graph. Denote by V(G)V(G) and E(G)E(G) the vertex set and edge set of GG, respectively. The degree of a vertex vv in GG is denoted by deg(v)\deg(v). If uu and vv are non-adjacent in GG, then G+uvG+uv is obtained from GG by adding the edge uvuv. We write uvu\sim v if two vertices uu and vv are adjacent in GG; and write u≁vu\not\sim v otherwise. For SV(G)S\subseteq V(G), denote by G[S]G[S] and GSG-S the subgraph of GG induced on SS and V(G)SV(G)\setminus S, respectively. For vV(G)v\in V(G), we write GvG-v for G{v}G-\{v\}. For two integer p,qp,q, we let [p,q]={i:piq}[p,q]=\{i\in\mathbb{Z}:p\leq i\leq q\}.

Let n1n\geq 1 be an integer. The non-decreasing sequence d1,d2,,dnd_{1},d_{2},\dots,d_{n} is a degree sequence of graph GG if the vertices of GG can be labeled as v1,v2,,vnv_{1},v_{2},\dots,v_{n} such that deg(vi)=di\deg(v_{i})=d_{i} for all i[1,n]i\in[1,n]. In 1972, Chvátal [3] proved the following well known result.

Theorem 1.

Let GG be a graph with degree sequence d1,d2,,dnd_{1},d_{2},\dots,d_{n}, where n3n\geq 3 is an integer. If for all i<n2i<\frac{n}{2}, diid_{i}\leq i implies dninid_{n-i}\geq n-i, then GG is Hamiltonian.

Hoàng [5, Conjecture 1] in 1995 conjectured a toughness analogue for the theorem above. We let c(G)c(G) denote the number of components of GG. For a real number t0t\geq 0, we say GG is t-tough if |S|tc(GS)|S|\geq t\cdot c(G-S) for all SV(G)S\subseteq V(G) such that c(GS)2c(G-S)\geq 2. The largest tt for which GG is tt-tough is called the toughness of GG and is denoted τ(G)\tau(G). If GG is complete, τ(G)\tau(G) is defined to be \infty. Chvátal [4] defined this concept in 1973 as a measure of a graph’s “resilience” under the removal of vertices. Hoàng’s conjecture can now be stated as follows.

Conjecture 2.

Let n3n\geq 3 and t1t\geq 1 be integers, and GG be a graph with degree sequence d1,d2,,dnd_{1},d_{2},\dots,d_{n}. Suppose that GG satisfies the following predicate P(t)P(t):

P(t):i,ti<n2,diidni+tni.P(t):\forall i,t\leq i<\frac{n}{2},d_{i}\leq i\Rightarrow d_{n-i+t}\geq n-i.

Then, if GG is tt-tough, GG is Hamiltonian.

Hoàng in the same paper [5, Theorem 3] proved that the Conjecture holds for t3t\leq 3. Since every hamiltonian graph must necessarily be 1-tough, the statement for t=1t=1 generalizes Theorem 1. Recently, Hoàng and Robin [6] proved that the Conjecture is true for t=4t=4. In this paper, we confirm Conjecture 2 for all t4t\geq 4.

Theorem 3.

Let t4t\geq 4 be an integer and GG be a tt-tough graph on n3n\geq 3 vertices with degree sequence d1,d2,,dnd_{1},d_{2},\dots,d_{n}. If for all i<n2i<\frac{n}{2} it holds that diid_{i}\leq i implies dni+tnid_{n-i+t}\geq n-i, then GG is Hamiltonian.

A graph GG is pancyclic if GG contains cycles of any length from 33 to |V(G)||V(G)|. As a consequence of Theorem 3 and a result of Hoàng [5, Theorem 7] that if a tt-tough graph GG satisfies P(t)P(t) and is Hamiltonian, then GG is pancyclic or bipartite, we also obtain the following result.

Corollary 4.

Let t4t\geq 4 be an integer and GG be a tt-tough graph on n3n\geq 3 vertices with degree sequence d1,d2,,dnd_{1},d_{2},\dots,d_{n}. If for all i<n2i<\frac{n}{2} it holds that diid_{i}\leq i implies dni+tnid_{n-i+t}\geq n-i, then GG is pancyclic or bipartite.

The proof of Theorem 3 relies on our closure lemma for tt-tough graphs GG: if xx and yy are non-adjacent in GG and deg(x)+deg(y)nt\deg(x)+\deg(y)\geq n-t, then G+xyG+xy is Hamiltonian implies that GG is Hamiltonian. We will prove Theorem 3 in the next section by applying this closure lemma and then prove the closure lemma in the subsequent section.

2 Proof of Theorem 3

We will need the following result by Bauer et al. [1] and our closure lemma for tt-tough graphs with t4t\geq 4.

Theorem 5.

Let t0t\geq 0 be any real number and GG be a tt-tough graph on n3n\geq 3 vertices . If δ(G)>nt+11\delta(G)>\frac{n}{t+1}-1, then GG is Hamiltonian.

Theorem 6 (Toughness Closure Lemma).

Let t4t\geq 4 be an integer, GG be a tt-tough graph on n3n\geq 3 vertices, and let distinct x,yV(G)x,y\in V(G) be non-adjacent with deg(x)+deg(y)nt\deg(x)+\deg(y)\geq n-t. Then GG is Hamiltonian if and only if G+xyG+xy is Hamiltonian.

The following toughness closure concept was given by Hoàng and Robin [6]. Let t1t\geq 1 be an integer, and GG be a tt-tough graph on n3n\geq 3 vertices. Then the tt-closure of GG is formed by repeatedly adding edges joining vertices xx and yy such that xx and yy are non-adjacent in the current graph and their degree sum is at least ntn-t in the current graph, until no such pair remains. By the same argument as showing that the Hamiltonian closure of a graph is well defined (e.g., see [2, Lemma 4.4.2]), the tt-closure of GG is well defined. Thus by Theorem 6, we will consider the tt-closure of GG instead when we prove Theorem 3. We mostly adopted the ideas used by Hoàng and Robin in [6].

Proof of Theorem 3.

As GG satisfies the property P(t)P(t) implies that any supergraph of GG obtained from GG by adding missing edges also satisfies the property P(t)P(t), by Theorem 6, it suffices to work with the tt-closure of GG. For the sake of notation, we just assume that GG itself is its tt-closure. We may assume that GG is not Hamiltonian. Thus GG is not complete and so δ(G)8\delta(G)\geq 8 by GG being 4-tough.

Let v1,v2,,vnv_{1},v_{2},\dots,v_{n} be all the vertices of GG such that deg(vi)=di\deg(v_{i})=d_{i} for all i[1,n]i\in[1,n]. Thus, we have that deg(vi)+deg(vj)nt\deg(v_{i})+\deg(v_{j})\geq n-t implies vivjE(G)v_{i}v_{j}\in E(G). By Theorem 1, if di>id_{i}>i for all i<n2i<\frac{n}{2}, then GG is Hamiltonian. So, we assume that there exists some positive integer k<n2k<\frac{n}{2} such that dkkd_{k}\leq k. Then as δ(G)8\delta(G)\geq 8, we have k8k\geq 8. Choose kk to be minimum with the property that dkkd_{k}\leq k. Then di>id_{i}>i for all i[1,k1]i\in[1,k-1]. Since dk1dkkd_{k-1}\leq d_{k}\leq k, we must have dk1=dk=kd_{k-1}=d_{k}=k.

Let S,TV(G)S,T\subseteq V(G). We say that SS is complete to TT if for all uSu\in S and vTv\in T such that uvu\neq v, we have uvu\sim v. If uvu\sim v for all uSu\in S and vV(G)v\in V(G) such that uvu\neq v, we call SS a universal clique of GG. Clearly, vertices in a universal clique have degree n1n-1 in GG. We will show that GG has a universal clique of size larger than nt+11\frac{n}{t+1}-1. In particular, this gives that δ(G)>nt+11\delta(G)>\frac{n}{t+1}-1. By Theorem 5, this proves that GG is Hamiltonian, a contradiction to the assumption that GG is not Hamiltonian. Let

Uα={vi:dinα,i[1,n]}for any integer α with 1α<n2.U^{\alpha}=\{v_{i}:d_{i}\geq n-\alpha,i\in[1,n]\}\quad\text{for any integer $\alpha$ with $1\leq\alpha<\frac{n}{2}$.}
Claim 2.1.

For all positive integer α<n2\alpha<\frac{n}{2}, UαU^{\alpha} is a clique complete to {vi:diαt,i[1,n]}\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\}.

Proof of Claim 2.1.

If vjUαv_{j}\in U^{\alpha} for some j[1,n]j\in[1,n] and v{vi:diαt,i[1,n]}v_{\ell}\in\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\} for some [1,n]\ell\in[1,n], then dj+dnα+αt=ntd_{j}+d_{\ell}\geq n-\alpha+\alpha-t=n-t. Thus, vjvv_{j}\sim v_{\ell}. This in turn implies that UαU^{\alpha} is a clique in GG, since Uα{vi:diαt,i[1,n]}U^{\alpha}\subseteq\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\}. ∎

Claim 2.2.

Let α<n2\alpha<\frac{n}{2} be any positive integer. If for every i[1,n]i\in[1,n], it holds that di<αtd_{i}<\alpha-t implies diit+1d_{i}\geq i-t+1, then UαU^{\alpha} is a universal clique in GG.

Proof of Claim 2.2.

Assume there exists a positive integer α<n2\alpha<\frac{n}{2} that satisfies the hypothesis, but UαU^{\alpha} is not a universal clique. Choose p[1,n]p\in[1,n] to be maximum such that there exists vqUαv_{q}\in U^{\alpha} for some q[1,n]q\in[1,n] such that vp≁vqv_{p}\not\sim v_{q}. By Claim 2.1, vp{vi:diαt,i[1,n]}v_{p}\notin\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\}. Thus dppt+1d_{p}\geq p-t+1 by the assumption of this claim. By the maximality of pp, we have vqvv_{q}\sim v_{\ell} for all [p+1,n]\ell\in[p+1,n]. So, dqnp1d_{q}\geq n-p-1, which gives dp+dqpt+1+np1=ntd_{p}+d_{q}\geq p-t+1+n-p-1=n-t. But, this implies vpvqv_{p}\sim v_{q}, a contradiction. ∎

Let ΩV(G)\Omega\subseteq V(G) be a universal clique in GG of maximum size.

Claim 2.3.

We have |Ω|k2|\Omega|\leq k-2.

Proof of Claim 2.3.

Suppose that |Ω|k1|\Omega|\geq k-1. As Ω\Omega is a universal clique in GG, we have di|Ω|k1d_{i}\geq|\Omega|\geq k-1 for all i[1,n]i\in[1,n]. If |Ω|>k|\Omega|>k, then d1>kd_{1}>k, which contradicts d1dk=kd_{1}\leq d_{k}=k. Thus |Ω|k|\Omega|\leq k. Note that viΩv_{i}\not\in\Omega for any i[1,k]i\in[1,k] as every vertex of Ω\Omega has degree n1>n2>kn-1>\frac{n}{2}>k. Let S=(i[1,k]N(vi)){v1,,vk}S=\left(\bigcup_{i\in[1,k]}N(v_{i})\right)\setminus\{v_{1},\ldots,v_{k}\}. As dikd_{i}\leq k for all i[1,k]i\in[1,k], each viv_{i} has at most k|Ω|k-|\Omega| neighbor from {vk+1,,vn}Ω\{v_{k+1},\ldots,v_{n}\}\setminus\Omega in GG, and so we have

|S|\displaystyle|S|\leq |Ω|=k\displaystyle|\Omega|=k if |Ω|=k|\Omega|=k,
|S|\displaystyle|S|\leq |Ω|+k2k1\displaystyle|\Omega|+k\leq 2k-1 if |Ω|=k1|\Omega|=k-1.

Since Δ(G[{v1,,vk}])1\Delta(G[\{v_{1},\ldots,v_{k}\}])\leq 1, we have c(GS)c(G[{v1,,vk}])k24c(G-S)\geq c(G[\{v_{1},\ldots,v_{k}\}])\geq\frac{k}{2}\geq 4. However, we get |S|c(GS)<4\frac{|S|}{c(G-S)}<4, contradicting the toughness of GG. Thus, Claim 2.3 must hold. ∎

Claim 2.4.

For all positive integer α<n2\alpha<\frac{n}{2} such that dααd_{\alpha}\leq\alpha, we have |Uα|αt|U^{\alpha}|\geq\alpha-t.

Proof of Claim 2.4.

Suppose vαV(G)v_{\alpha}\in V(G) such that dαα<n2d_{\alpha}\leq\alpha<\frac{n}{2}. By the hypothesis, dnα+tnαd_{n-\alpha+t}\geq n-\alpha. That is, there are at least n(nα+t)+1=αt+1n-(n-\alpha+t)+1=\alpha-t+1 vertices of degree at least nαn-\alpha, indicating |Uα|αt|U^{\alpha}|\geq\alpha-t. ∎

Claim 2.5.

We have dα>αd_{\alpha}>\alpha for all integer α\alpha with k+t1α<n2k+t-1\leq\alpha<\frac{n}{2}.

Proof of Claim 2.5.

Assume there exists α\alpha such that k+t1α<n2k+t-1\leq\alpha<\frac{n}{2} and dααd_{\alpha}\leq\alpha. Choose such an α\alpha to be minimum. It suffices to show that UαU^{\alpha} is a universal clique: by Claims 2.3 and  2.4, we have k2|Ω||Uα|αtk-2\geq|\Omega|\geq|U^{\alpha}|\geq\alpha-t. Rearranging gives k+t2αk+t1k+t-2\geq\alpha\geq k+t-1, a contradiction. Thus we show that UαU^{\alpha} is a universal clique in the following. By Claim 2.1, UαU^{\alpha} is a clique complete to {vi:diαt,i[1,n]}\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\}. Therefore, to apply Claim 2.2, we show that every vertex vjv_{j} for j[1,n]j\in[1,n] belongs to the set {vi:diαt,i[1,n]}\{v_{i}:d_{i}\geq\alpha-t,i\in[1,n]\} or satisfies dj<αtd_{j}<\alpha-t but djjt+1d_{j}\geq j-t+1.

We first show that djαtd_{j}\geq\alpha-t for all j[α,n]j\in[\alpha,n]. Consider for now that j=αj=\alpha. If α>k+t1\alpha>k+t-1, then α1k+t1\alpha-1\geq k+t-1. By the minimality of α\alpha, we get α1<dα1dαα\alpha-1<d_{\alpha-1}\leq d_{\alpha}\leq\alpha. Thus dα=α>αtd_{\alpha}=\alpha>\alpha-t. If α=k+t1\alpha=k+t-1, then dαdk=k>αtd_{\alpha}\geq d_{k}=k>\alpha-t. In either case, we have shown dααtd_{\alpha}\geq\alpha-t. For any j[α+1,n]j\in[\alpha+1,n], we have djdααtd_{j}\geq d_{\alpha}\geq\alpha-t. Now for j[1,α1]j\in[1,\alpha-1], suppose dj<αtd_{j}<\alpha-t. By the minimality of kk, we have djjjt+1d_{j}\geq j\geq j-t+1 if j[1,k]j\in[1,k]. We have djdk=k>k1jt+1d_{j}\geq d_{k}=k>k-1\geq j-t+1 if j[k+1,k+t2]j\in[k+1,k+t-2]. By the minimality of α\alpha, we have dj>j>jt+1d_{j}>j>j-t+1 for all j[k+t1,α1]j\in[k+t-1,\alpha-1]. This completes the proof. ∎

Claim 2.6.

We have kn2tk\geq\frac{n}{2}-t.

Proof of Claim 2.6.

We suppose to the contrary that k<n2tk<\frac{n}{2}-t. Let p=n12p=\lfloor\frac{n-1}{2}\rfloor. Then k+t1p<n/2k+t-1\leq p<n/2. By Claim 2.5, we have dp>pd_{p}>p. If dp=n1d_{p}=n-1, then all vertices from {vp,,vn}\{v_{p},\ldots,v_{n}\} are contained in a universal clique of GG and so we have |Ω|>n2|\Omega|>\frac{n}{2}. This gives k|Ω|>n2k\geq|\Omega|>\frac{n}{2}, a contradiction of the assumption that k<n2tk<\frac{n}{2}-t. Thus there exists i[1,n]i\in[1,n] such that vp≁viv_{p}\not\sim v_{i}. We choose such an ii to be maximum. Since vivpv_{i}\nsim v_{p}, we have di<ntdp<nt(n121)=n+12t+1dpd_{i}<n-t-d_{p}<n-t-(\frac{n-1}{2}-1)=\frac{n+1}{2}-t+1\leq d_{p}, which gives i<pi<p. Then by Claim 2.5 and the argument in the second paragraph in the proof of Claim 2.5, we have diit+1d_{i}\geq i-t+1. By the maximality of ii, we have vpvjv_{p}\sim v_{j} for all j[i+1,n]j\in[i+1,n] and so dpni1d_{p}\geq n-i-1. This gives di+dpni1+it+1=ntd_{i}+d_{p}\geq n-i-1+i-t+1=n-t, which contradicts that vpviv_{p}\nsim v_{i}. ∎

Claim 2.7.

We have δ(G)>nt+11\delta(G)>\frac{n}{t+1}-1.

Proof of Claim 2.7.

Assume δ(G)nt+11\delta(G)\leq\frac{n}{t+1}-1. Then, as 2tδ(G)2t\leq\delta(G), we have (2t+1)(t+1)n(2t+1)(t+1)\leq n. By Claim 2.2 and the choice of kk, we know that UkU^{k} is a universal clique. Therefore, by Claims 2.4 and 2.6, we get δ(G)|Uk|ktn22t\delta(G)\geq|U^{k}|\geq k-t\geq\frac{n}{2}-2t. Observe that for t3t\geq 3, we have

n2nt+1\displaystyle\frac{n}{2}-\frac{n}{t+1} =\displaystyle= n(t1)2(t+1)(2t+1)(t+1)(t1)2(t+1)\displaystyle\frac{n(t-1)}{2(t+1)}\geq\frac{(2t+1)(t+1)(t-1)}{2(t+1)}
=\displaystyle= (t+0.5)(t1)>2t1.\displaystyle(t+0.5)(t-1)>2t-1.

This gives n22t>nt+11\frac{n}{2}-2t>\frac{n}{t+1}-1. Thus δ(G)kt>nt+11\delta(G)\geq k-t>\frac{n}{t+1}-1, a contradiction. ∎

As δ(G)>nt+11\delta(G)>\frac{n}{t+1}-1, Theorem 5 implies that GG is Hamiltonian, a contradiction to our assumption that GG is not Hamiltonian. This completes the proof. \blacksquare

3 Proof of Theorem 6

Denote by C\overset{\rightharpoonup}{C} an orientation of a cycle CC. We assume that the orientation is clockwise throughout the rest of this paper. For u,vV(C)u,v\in V(C), uCvu\overset{\rightharpoonup}{C}v denotes the path from uu to vv along C\overset{\rightharpoonup}{C}. Similarly, uCvu\overset{\leftharpoonup}{C}v denotes the path between uu and vv which travels opposite to the orientation. We use u+u^{+} to denote the immediate successor of uu on C\overset{\rightharpoonup}{C} and uu^{-} to denote the immediate predecessor of uu on C\overset{\rightharpoonup}{C}. If SV(C)S\subseteq V(C), then S+={u+:uS}S^{+}=\{u^{+}:u\in S\} and S={u:uS}S^{-}=\{u^{-}:u\in S\}. We use similar notation for a path PP when it is given an orientation. Theorem 7 is needed in the proof of Theorem 6, and we prove Theorem 7 in the last section.

Theorem 7.

Let t4t\geq 4 be rational and GG be a tt-tough graph on n3n\geq 3 vertices. Suppose that GG is not Hamiltonian, but there exists zV(G)z\in V(G) such that GzG-z has a Hamilton cycle CC. Then, for any distinct x,yN(z)x,y\in N(z), we have that deg(x+)+deg(y+)<nt\deg(x^{+})+\deg(y^{+})<n-t.

Theorem 6 (Toughness Closure Lemma).

Let t4t\geq 4 be an integer, GG be a tt-tough graph on n3n\geq 3 vertices, and let distinct x,yV(G)x,y\in V(G) be non-adjacent with deg(x)+deg(y)nt\deg(x)+\deg(y)\geq n-t. Then GG is Hamiltonian if and only if G+xyG+xy is Hamiltonian.

Proof.

It is clear that GG is Hamiltonian implies that G+xyG+xy is Hamiltonian. For the converse, we suppose that G+xyG+xy is Hamiltonian but GG is not. Then again, this implies that GG is not complete and so δ(G)2t\delta(G)\geq 2t.

As G+xyG+xy is Hamiltonian, GG has a Hamilton path connecting xx and yy. Let P=v1vnP=v_{1}\ldots v_{n} be such a path, where v1=xv_{1}=x and vn=yv_{n}=y. We will orient PP to be from xx to yy, and write uvu\preceq v for two vertices uu and vv such that uu is at least as close to xx along P\overset{\rightharpoonup}{P} as vv is. Our goal is to find a cutset SS of GG with size less than 2t2t and so arriving a contradiction to the toughness of GG. For this purpose, based on the assumption that GG is not Hamiltonian, we look at how the neighbors of xx and yy are arranged along this path PP, and their adjacency relations.

The first two assertions below follow directly from the assumption that GG is not Hamiltonian, and the last two are corollaries of the first two.

Claim 3.1.

Let distinct i,j[2,n1]i,j\in[2,n-1] and suppose xvix\sim v_{i} and yvjy\sim v_{j}. Then the following holds.

  1. (1)

    If i<ji<j, then vi≁vj+v_{i}^{-}\not\sim v_{j}^{+} and y≁viy\not\sim v_{i}^{-}.

  2. (2)

    If i>ji>j, then vi+≁vj+v_{i}^{+}\not\sim v_{j}^{+} and vi≁vjv_{i}^{-}\not\sim v_{j}^{-}.

  3. (3)

    If in3i\leq n-3 and additionally xvi+2x\sim v_{i+2}, then vi+1≁vk+v_{i+1}\not\sim v_{k}^{+} for any vkv_{k} with vkyv_{k}\sim y.

  4. (4)

    If jn3j\leq n-3 and additionally yvj+2y\sim v_{j+2}, then vj+1≁vkv_{j+1}\not\sim v_{k}^{-} for any vkv_{k} with vkxv_{k}\sim x.

Since deg(x)+deg(y)nt\deg(x)+\deg(y)\geq n-t and xx and yy do not have two common neighbors that are consecutive on PP by Claim 3.1(1) above, each of xx and yy is expected to have many neighbors that are consecutive on PP. Thus we define neighbor intervals for xx and yy, respectively, as set of consecutive vertices on PP that are all adjacent to xx or yy. For z{x,y}z\in\{x,y\}, and vi,vjv_{i},v_{j} with i,j[2,n1]i,j\in[2,n-1] and iji\leq j such that zvi,vjz\sim v_{i},v_{j}, we call V(viPvj)V(v_{i}Pv_{j}) a zz-interval and denote it by Iz[vi,vj]I_{z}[v_{i},v_{j}] if V(viPvj)N(z)V(v_{i}Pv_{j})\subseteq N(z) but vi,vj+≁zv_{i}^{-},v_{j}^{+}\not\sim z.

Given Ix[vi,vj]I_{x}[v_{i},v_{j}] and Iy[vk,v]I_{y}[v_{k},v_{\ell}], by Claim 3.1(1), we know that the two intervals can have at most one vertex in common. In case that they do have a common vertex, then it must be the case that vj=vkv_{j}=v_{k}. In this case, we let Ixy[vi,vj,vk]=Ix[vi,vj]Iy[vk,v]I_{xy}[v_{i},v_{j},v_{k}]=I_{x}[v_{i},v_{j}]\cup I_{y}[v_{k},v_{\ell}] and call it a joint-interval. Finally, for i,j[3,n2]i,j\in[3,n-2] with iji\leq j, we define interval-gaps to be sets of consecutive vertices on PP that are all not adjacent to either xx or yy. A parallel-gap is J[vi,vj]:=V(viPvj)J[v_{i},v_{j}]:=V(v_{i}Pv_{j}) such that V(viPvj)(N(x)N(y))=V(v_{i}Pv_{j})\cap(N(x)\cup N(y))=\emptyset and that vi,vj+N(x)v_{i}^{-},v_{j}^{+}\in N(x), or vi,vj+N(y)v_{i}^{-},v_{j}^{+}\in N(y), or viN(x)v_{i}^{-}\in N(x) but vj+N(y)v_{j}^{+}\in N(y). A crossing-gap is J[vi,vj]:=V(viPvj)J[v_{i},v_{j}]:=V(v_{i}Pv_{j}) such that V(viPvj)(N(x)N(y))=V(v_{i}Pv_{j})\cap(N(x)\cup N(y))=\emptyset and that viN(y)v_{i}^{-}\in N(y) and vj+N(x)v_{j}^{+}\in N(x). By the range of ii and jj in the above definition, we see that each of xx and yy is not contained in any interval-gaps.

Let x\mathcal{I}_{x} be the set of xx-intervals that are not joint-intervals, y\mathcal{I}_{y} be the set of yy-intervals that are not joint-intervals, and xy\mathcal{I}_{xy} be the set of joint-intervals. Let

p=|xy|,andq=|xy|.p=|\mathcal{I}_{x}\cup\mathcal{I}_{y}|,\quad\text{and}\quad q=|\mathcal{I}_{xy}|.
Claim 3.2.

Each crossing-gap contains at least two vertices and there are at least q1q-1 distinct crossing-gaps when q1q\geq 1.

Proof of Claim 3.2.

For the first part, suppose {vi}\{v_{i}\} for some i[2,n1]i\in[2,n-1] is a crossing-gap with a single vertex. Then C=vi+1xPvi1yPvi+1C=v_{i+1}x\overset{\rightharpoonup}{P}v_{i-1}y\overset{\leftharpoonup}{P}v_{i+1} gives a Hamilton cycle of GviG-v_{i}. We have vivi1,vi+1v_{i}\sim v_{i-1},v_{i+1}, and with respect to the cycle C\overset{\rightharpoonup}{C}, we have x=vi+1+x=v_{i+1}^{+} and y=vi1+y=v_{i-1}^{+}. However, deg(x)+deg(y)nt\deg(x)+\deg(y)\geq n-t, contradicting Theorem 7. For the second part, assume that q2q\geq 2. Let the qq common neighbors of xx and yy be u1,uqu_{1},\ldots u_{q} with u1u2uqu_{1}\preceq u_{2}\ldots\preceq u_{q}. Thus V(uiPui+1)V(u_{i}Pu_{i+1}) for each i[1,q1]i\in[1,q-1] is a set of vertices such that uiyu_{i}\sim y and ui+1xu_{i+1}\sim x. By the first part of this claim and Claim 3.1(1), we know that each of V(ui+Pui+1)V(u_{i}^{+}Pu^{-}_{i+1}) for i[1,q1]i\in[1,q-1] contains at least two vertices that are adjacent to neither xx nor yy. By finding a minimal sub-path of ui+Pui+1u_{i}^{+}\overset{\rightharpoonup}{P}u^{-}_{i+1} such that the predecessor of its left end is a neighbor of yy, the successor of its right end is a neighbor of xx, we can find two distinct vertices w1,w2V(ui+Pui+1)w_{1},w_{2}\in V(u_{i}^{+}Pu^{-}_{i+1}) with the following properties: w1w2w_{1}\preceq w_{2}, w1yw_{1}^{-}\sim y, w2+xw_{2}^{+}\sim x, and V(w1Pw2)(N(x)N(y))=V(w_{1}Pw_{2})\cap(N(x)\cap N(y))=\emptyset. Then J[w1,w2]J[w_{1},w_{2}] is a crossing-gap. Since V(ui+Pui+1)V(u_{i}^{+}Pu^{-}_{i+1}) and V(uj+Puj+1)V(u_{j}^{+}Pu^{-}_{j+1}) are disjoint for distinct i,j[1,q1]i,j\in[1,q-1], we can find q1q-1 distinct crossing-gaps. ∎

Let pp^{*} be the total number of distinct parallel-gaps and qq^{*} be the total number of distinct crossing-gaps. We let the set of pp^{*} parallel-gaps be {J[ui,wi]:i[1,p],u1w1u2w2upwp}\{J[u_{i},w_{i}]:i\in[1,p^{*}],u_{1}\preceq w_{1}\preceq u_{2}\preceq w_{2}\preceq\ldots\preceq u_{p^{*}}\preceq w_{p^{*}}\}, and let |J[ui,wi]|=pi|J[u_{i},w_{i}]|=p_{i}. We also let the set of qq^{*}crossing-gaps be {J[ri,si]:i[1,q],r1s1r2s2rqsq}\{J[r_{i},s_{i}]:i\in[1,q^{*}],r_{1}\preceq s_{1}\preceq r_{2}\preceq s_{2}\ldots\preceq r_{q^{*}}\preceq s_{q^{*}}\}, and let |J[ri,si]|=qi|J[r_{i},s_{i}]|=q_{i}.

Claim 3.3.

We have |xyxy|=p+qti=1p(pi1)i=1q(qi2)|\mathcal{I}_{x}\cup\mathcal{I}_{y}\cup\mathcal{I}_{xy}|=p+q\leq t-\sum\limits_{i=1}^{p^{*}}(p_{i}-1)-\sum\limits_{i=1}^{q^{*}}(q_{i}-2).

Proof of Claim 3.3.

By the definition, the three sets x,y,xy\mathcal{I}_{x},\mathcal{I}_{y},\mathcal{I}_{xy} are pairwise disjoint. Thus |xyxy|=p+q|\mathcal{I}_{x}\cup\mathcal{I}_{y}\cup\mathcal{I}_{xy}|=p+q. Also, by our definition, we have |N(x)N(y)|=|xy|=q|N(x)\cap N(y)|=|\mathcal{I}_{xy}|=q and so |N(x)N(y)|ntq|N(x)\cup N(y)|\geq n-t-q. Since |xyxy|=p+q|\mathcal{I}_{x}\cup\mathcal{I}_{y}\cup\mathcal{I}_{xy}|=p+q, and v2v_{2} and vn1v_{n-1} are contained in an xx-interval, yy-interval, or joint-interval, it follows that there are exactly p+q1=p+qp+q-1=p^{*}+q^{*} interval-gaps. By Claim 3.2, qq1q^{*}\geq q-1. As xx and yy are not contained in any interval-gaps, we get

t+q\displaystyle t+q \displaystyle\geq |V(G)(N(x)N(y))|2+i=1ppi+i=1qqi\displaystyle|V(G)\setminus(N(x)\cup N(y))|\geq 2+\sum_{i=1}^{p^{*}}p_{i}+\sum_{i=1}^{q^{*}}q_{i}
\displaystyle\geq 2+p+i=1p(pi1)+2q+i=1q(qi2).\displaystyle 2+p^{*}+\sum\limits_{i=1}^{p^{*}}(p_{i}-1)+2q^{*}+\sum\limits_{i=1}^{q^{*}}(q_{i}-2).

As p+q1=p+qp+q-1=p^{*}+q^{*} and qq1q^{*}\geq q-1, we get p+qti=1p(pi1)i=1q(qi2)p+q\leq t-\sum\limits_{i=1}^{p^{*}}(p_{i}-1)-\sum\limits_{i=1}^{q^{*}}(q_{i}-2). Therefore,

|xyxy|=p+qti=1p(pi1)i=1q1(qi2),|\mathcal{I}_{x}\cup\mathcal{I}_{y}\cup\mathcal{I}_{xy}|=p+q\leq t-\sum_{i=1}^{p}(p_{i}-1)-\sum_{i=1}^{q-1}(q_{i}-2),

as desired. ∎

Claim 3.4.

For any i[2,n2]i\in[2,n-2], if {vi,vi+1}\{v_{i},v_{i+1}\} is a crossing-gap of size 2, then vi≁vjv_{i}\not\sim v_{j} for any j[3,n2]j\in[3,n-2] such that yvj1,vj+1y\sim v_{j-1},v_{j+1}.

Proof of Claim 3.4.

We will show that vi+1v_{i+1} has less than 2t2t neighbors in GG, to arrive a contradiction to GG being tt-tough.

By Claim 3.1(1)-(2), we know that for any vkyv_{k}\sim y with vkviv_{k}\preceq v_{i} on PP, we have vi+1≁vk1v_{i+1}\not\sim v_{k-1}; and for vkyv_{k}\sim y with vivkv_{i}\preceq v_{k} on PP, we have vi+1≁vk+1v_{i+1}\not\sim v_{k+1}. Thus vertices from (N(y)V(v2Pvi))(N(y)\cap V(v_{2}Pv_{i}))^{-} and (N(y)V(vi+2Pvn1))+(N(y)\cap V(v_{i+2}Pv_{n-1}))^{+} are non-neighbors of vi+1v_{i+1}. Let

C=\displaystyle C= vjviPxvi+2Pvj1yPvj\displaystyle v_{j}v_{i}\overset{\leftharpoonup}{P}xv_{i+2}\overset{\rightharpoonup}{P}v_{j-1}y\overset{\leftharpoonup}{P}v_{j} if i<ji<j,
C=\displaystyle C= vjviPvj+1yPvi+2xPvj\displaystyle v_{j}v_{i}\overset{\leftharpoonup}{P}v_{j+1}y\overset{\leftharpoonup}{P}v_{i+2}x\overset{\rightharpoonup}{P}v_{j} if i>ji>j.

Then CC is a Hamilton cycle of Gvi+1G-v_{i+1}. The predecessors and successors of vertices below are all taken with respect to C\overset{\rightharpoonup}{C}. As GG is not Hamiltonian, both N(vi+1)N(v_{i+1})^{-} and N(vi+1)+N(v_{i+1})^{+} are independent sets in GG. When i<ji<j, since vi+1vi+2v_{i+1}\sim v_{i+2} and x=vi+2x=v_{i+2}^{-}, it then follows that vi+1≁z+v_{i+1}\not\sim z^{+} for any zN(x)z\in N(x). As a consequence, we get N(x)+N(vi+1)=N(x)^{+}\cap N(v_{i+1})=\emptyset. When i>ji>j, since vi+1vi+2v_{i+1}\sim v_{i+2} and x=vi+2+x=v_{i+2}^{+}, it then follows that vi+1≁zv_{i+1}\not\sim z^{-} for any zN(x)z\in N(x). As a consequence, we get N(x)N(vi+1)=N(x)^{-}\cap N(v_{i+1})=\emptyset.

The arguments above indicate that for every distinct vertex zN(x)N(y)z\in N(x)\cup N(y), there is a unique non-neighbor of vi+1v_{i+1} that is corresponding to zz. Thus vi+1v_{i+1} has at least |N(x)N(y)||N(x)\cup N(y)| non-neighbors on CC. Then by Claim 3.3 that qtq\leq t, we get

deg(vi+1)\displaystyle\deg(v_{i+1}) \displaystyle\leq n1|N(x)N(y)|\displaystyle n-1-|N(x)\cup N(y)|
\displaystyle\leq n1(ntq)\displaystyle n-1-(n-t-q)
\displaystyle\leq 2t1,\displaystyle 2t-1,

a contradiction. ∎

We now construct a cutset SS of GG such that |S|<2t|S|<2t. To do so, we define the following sets:

Sx\displaystyle S_{x} =\displaystyle= {vj,vj+1:vj is the right endvertex of an x-interval that is not a joint-interval},\displaystyle\{v_{j},v_{j+1}:\text{$v_{j}$ is the right endvertex of an $x$-interval that is not a joint-interval}\},
Sy\displaystyle S_{y} =\displaystyle= {vi,vj:Iy[vi,vj] is a y-interval that is not a joint-interval},\displaystyle\{v_{i},v_{j}:\text{$I_{y}[v_{i},v_{j}]$ is a $y$-interval that is not a joint-interval}\},
Sxy\displaystyle S_{xy} =\displaystyle= {vj,vk:Ixy[vi,vj,vk] is a joint-interval},\displaystyle\{v_{j},v_{k}:\text{$I_{xy}[v_{i},v_{j},v_{k}]$ is a joint-interval}\},
T1\displaystyle T_{1} =\displaystyle=  J[vi,vj] is a parallel-gap of size at least 2J[vi,vj],\displaystyle\bigcup_{\text{ $J[v_{i},v_{j}]$ is a parallel-gap of size at least 2}}J[v_{i},v_{j}],
T2\displaystyle T_{2} =\displaystyle=  J[vi,vj] is a crossing-gap of size 3(J[vi,vj]{vj}),\displaystyle\bigcup_{\text{ $J[v_{i},v_{j}]$ is a crossing-gap of size 3}}\left(J[v_{i},v_{j}]\setminus\{v_{j}\}\right),
T3\displaystyle T_{3} =\displaystyle=  J[vi,vj] is a crossing-gap of size at least 4J[vi,vj].\displaystyle\bigcup_{\text{ $J[v_{i},v_{j}]$ is a crossing-gap of size at least 4}}J[v_{i},v_{j}].

Let

S=\displaystyle S= SxSySxyT1T2T3\displaystyle S_{x}\cup S_{y}\cup S_{xy}\cup T_{1}\cup T_{2}\cup T_{3} if {vn1}\{v_{n-1}\} is a yy-interval,
S=\displaystyle S= (SxSySxyT1T2T3){vn1}\displaystyle\left(S_{x}\cup S_{y}\cup S_{xy}\cup T_{1}\cup T_{2}\cup T_{3}\right)\setminus\{v_{n-1}\} otherwise.

We prove the following claims regarding what vertices are in V(G)SV(G)\setminus S and the size of SS.

Claim 3.5.

Let viV(G)Sv_{i}\in V(G)\setminus S for some i[2,n2]i\in[2,n-2]. Then xvi,vi+1x\sim v_{i},v_{i+1}, or yvi1,vi+1y\sim v_{i-1},v_{i+1}, or viv_{i} is contained in a parallel-gap of size one such that yvi1,vi+1y\sim v_{i-1},v_{i+1}, or viv_{i} is contained in a crossing-gap of size two, or viv_{i} is the right endvertex of a crossing-gap of size three.

Proof of Claim 3.5.

By the definition of SS, we know that either viv_{i} is a neighbor of xx or yy, or viv_{i} is contained in a parallel-gap of size one, or a crossing-gap of size two or three. If xvix\sim v_{i}, then by the definition of SxS_{x}, we have xvi+1x\sim v_{i+1}. If yviy\sim v_{i}, then by the definition of SyS_{y}, we have yvi1,vi+1y\sim v_{i-1},v_{i+1}. If viv_{i} is contained in a parallel-gap of size one, then by the definition of SxS_{x}, we know that yvi1y\sim v_{i-1}. As {vi}\{v_{i}\} is a parallel-gap, yvi1y\sim v_{i-1} implies yvi+1y\sim v_{i+1}. If viv_{i} is contained in crossing-gap of size three, then viv_{i} is the right endvertex of a crossing-gap of size three by the definition of T3T_{3}. ∎

Claim 3.6.

We have |S|2t1|S|\leq 2t-1.

Proof of Claim 3.6.

For each crossing-gap J[ri,si]J[r_{i},s_{i}] of size qiq_{i}, we let qi=qiq_{i}^{*}=q_{i} if qi4q_{i}\geq 4, qi=qi1q_{i}^{*}=q_{i}-1 if qi=3q_{i}=3, and qi=0q_{i}^{*}=0 if qi=2q_{i}=2. Note that by the definition of SS, only one vertex was deleted from the yy-interval containing vn1v_{n-1}. Now by the definition of SS and Claim 3.3, we have

|S|\displaystyle|S| \displaystyle\leq 2(p+q)1+i=1,pi2ppi+i=1qqi\displaystyle 2(p+q)-1+\sum_{i=1,p_{i}\geq 2}^{p^{*}}p_{i}+\sum_{i=1}^{q^{*}}q_{i}^{*}
\displaystyle\leq 2(ti=1p(pi1)i=1q(qi2))1+i=1,pi2ppi+i=1qqi\displaystyle 2\left(t-\sum\limits_{i=1}^{p^{*}}(p_{i}-1)-\sum\limits_{i=1}^{q^{*}}(q_{i}-2)\right)-1+\sum_{i=1,p_{i}\geq 2}^{p^{*}}p_{i}+\sum_{i=1}^{q^{*}}q_{i}^{*}
=\displaystyle= 2t1+i=1,pi2p(pi2(pi1))+i=1q(qi2(qi2))\displaystyle 2t-1+\sum_{i=1,p_{i}\geq 2}^{p^{*}}\left(p_{i}-2(p_{i}-1)\right)+\sum\limits_{i=1}^{q^{*}}\left(q_{i}^{*}-2(q_{i}-2)\right)
\displaystyle\leq 2t1,\displaystyle 2t-1,

where the last inequality follows as pi2(pi1)0p_{i}-2(p_{i}-1)\leq 0 when pi2p_{i}\geq 2, and qi2(qi2)0q_{i}^{*}-2(q_{i}-2)\leq 0 by the definition of qiq_{i}^{*} and the fact that qi2q_{i}\geq 2 for all i[1,q]i\in[1,q^{*}] from Claim 3.2. ∎

Claim 3.7.

We have c(GS)2c(G-S)\geq 2.

Proof of Claim 3.7.

For the sake of contradiction, suppose G=GSG^{\prime}=G-S is connected. Let X=NG(x){x}X^{\prime}=N_{G^{\prime}}(x)\cup\{x\} and Y=NG(y){y}Y^{\prime}=N_{G^{\prime}}(y)\cup\{y\}. Then, there must exists a path PP^{\prime} in GG^{\prime} connecting a vertex of XX^{\prime} and a vertex YY^{\prime} and is internally-disjoint with XYX^{\prime}\cup Y^{\prime}. Suppose that P=uu1uhvP^{\prime}=uu_{1}\ldots u_{h}v for some uXu\in X^{\prime} and vYv\in Y^{\prime}. By Claim 3.5, we know that v=yv=y, or v,v+yv^{-},v^{+}\sim y, or y=yn1y=y_{n-1} when the yy-interval containing yn1y_{n-1} has size at least two, and that u+xu^{+}\sim x. By Claim 3.1(1) and (4), we know that PuvP^{\prime}\neq uv. Thus PP^{\prime} contains at least three vertices. As PP^{\prime} is internally-disjoint with XYX^{\prime}\cup Y^{\prime}, u1,,uhu_{1},\ldots,u_{h} are from interval-gaps of PP.

As again, v=yv=y, or v,v+yv^{-},v^{+}\sim y, or y=yn1y=y_{n-1} when the yy-interval containing yn1y_{n-1} has size at least two. Since uhvu_{h}\sim v, Claim 3.1(4) implies that uh+≁xu_{h}^{+}\not\sim x. Thus uhu_{h} is not the right endvertex of any crossing-gap. By Claim 3.4, uhu_{h} is not the left endvertex of any crossing-gap of size two. Thus by Claim 3.5, {uh}\{u_{h}\} is a parallel-gap of size one such that yuh,uh+y\sim u_{h}^{-},u_{h}^{+}. Now with uhu_{h} in the place of vv, the same arguments as above imply that {uh1}\{u_{h-1}\}, if exists, is a parallel-gap of size one such that yuh1,uh1+y\sim u_{h-1}^{-},u_{h-1}^{+}. Similarly, for any i[1,h2]i\in[1,h-2], if exists, we deduce that {ui}\{u_{i}\} is a parallel-gap of size one such that yui,ui+y\sim u_{i}^{-},u_{i}^{+}. As u1uu_{1}\sim u and u+xu^{+}\sim x, we get a contradiction to Claim 3.1(4). ∎

Now Claims 3.6 and 3.7 together give a controduction to the toughness of GG, completing the proof of Theorem 6. \blacksquare

4 Proof of Theorem 7

Theorem 7.

Let t4t\geq 4 be rational and GG be a tt-tough graph on n3n\geq 3 vertices. Suppose that GG is not Hamiltonian, but there exists zV(G)z\in V(G) such that GzG-z has a Hamilton cycle CC. Then, for any distinct x,yN(z)x,y\in N(z), we have that deg(x+)+deg(y+)<nt\deg(x^{+})+\deg(y^{+})<n-t.

Proof.

Suppose to the contrary that there are distinct x,yN(z)x,y\in N(z) for which deg(x+)+deg(y+)nt\deg(x^{+})+\deg(y^{+})\geq n-t. As GG is not hamiltonian, GG is not a complete graph. Thus deg(z)=deg(z,C)2t\deg(z)=\deg(z,C)\geq 2t.

For SV(G)S\subseteq V(G) and xV(G)x\in V(G), let N(S)=vSN(v)N(S)=\bigcup_{v\in S}N(v) and N(x,S)=N(x)SN(x,S)=N(x)\cap S. For u,vV(C)u,v\in V(C), we let Vuv+=V(uCv)V_{uv}^{+}=V(u\overset{\rightharpoonup}{C}v) and Vuv=V(uCv)V_{uv}^{-}=V(u\overset{\leftharpoonup}{C}v). We will construct a cutset SS of GG such that |S|c(GS)<t\frac{|S|}{c(G-S)}<t. For this purpose, we define the following sets:

Y1=N(y+,Vy+x+),Y2=N(y+,Vy+x)+,Y=Y1Y2,X=N(x+),Z=N(z)+,R=V(G)(XYZ).\begin{array}[]{lll}Y_{1}=N(y^{+},V^{+}_{y^{+}x})^{-},&Y_{2}=N(y^{+},V^{-}_{y^{+}x})^{+},&Y=Y_{1}\cup Y_{2},\\ X=N(x^{+}),&Z=N(z)^{+},&R=V(G)\setminus(X\cup Y\cup Z).\end{array}

In the following, we prove some properties of these sets.

Claim 4.1.

We have XY=X\cap Y=\emptyset.

Proof of Claim 4.1.

Suppose to the contrary that there exists aXYa\in X\cap Y. If aY1a\in Y_{1}, then y+Cax+CyzxCa+y+y^{+}\overset{\rightharpoonup}{C}ax^{+}\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}a^{+}y^{+} is a Hamilton cycle of GG. If aY2a\in Y_{2}, then y+Cax+CyzxCa+y+y^{+}\overset{\rightharpoonup}{C}ax^{+}\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}a^{+}y^{+} is a Hamilton cycle of GG. ∎

If there are u,vZu,v\in Z with uN(v)u\in N(v), then uvCuzvCuuv\overset{\rightharpoonup}{C}u^{-}zv^{-}\overset{\leftharpoonup}{C}u is a Hamilton cycle in GG. Thus we have the following claim.

Claim 4.2.

The set ZZ is an independent set in GG.

Claim 4.3.

We have |R(ZY)|t|R\cup(Z\setminus Y)|\leq t and |YZ||R|+t|Y\cap Z|\geq|R|+t.

Proof of Claim 4.3.

Clearly |XYZ|n|R||X\cup Y\cup Z|\leq n-|R|. Observe that |X|=deg(x+)|X|=\deg(x^{+}) and |Y|=deg(y+)|Y|=\deg(y^{+}). By Claim 4.1, we have |XY|=|X|+|Y|nt|X\cup Y|=|X|+|Y|\geq n-t; and by Claim 4.2, we have XZ=X\cap Z=\emptyset . Thus,

n|R||XYZ|\displaystyle n-|R|\geq|X\cup Y\cup Z| \displaystyle\geq |X|+|Y|+|Z||XZ||YZ|\displaystyle|X|+|Y|+|Z|-|X\cap Z|-|Y\cap Z| (1)
\displaystyle\geq nt+|Z||YZ|=nt+|ZY|,\displaystyle n-t+|Z|-|Y\cap Z|=n-t+|Z\setminus Y|,

which gives |R(ZY)|t|R\cup(Z\setminus Y)|\leq t. For the second part, it follows from (1) by noting that |Z|2t|Z|\geq 2t. ∎

We will take a subset UU of (YX+)(YX)(Y\cap X^{+})\cup(Y\cap X^{-}) with size at least tt and show that deleting less than 4t4t vertices from GG produces at least tt components, and thus contradicts the assumption that GG is 4-tough. We let

U1=YX+Vyx+,U2=YXVyx,U=U1U2.U_{1}=Y\cap X^{+}\cap V^{+}_{yx},\quad U_{2}=Y\cap X^{-}\cap V^{-}_{yx},\quad U=U_{1}\cup U_{2}.
Claim 4.4.

We have |U|t|U|\geq t.

Proof of Claim 4.4.

As |ZY||R|+t|Z\cap Y|\geq|R|+t, it suffices to show |(ZY)U||R||(Z\cap Y)\setminus U|\leq|R|. Let u(YZVyx+)U1u\in(Y\cap Z\cap V_{yx}^{+})\setminus U_{1}. Then we have uXu^{-}\not\in X by the definition of U1U_{1}. Also, we have uZu^{-}\not\in Z because uZu\in Z and ZZ is an independent set by Claim 4.2. Furthermore, uYu^{-}\not\in Y, as otherwise y+uy^{+}\sim u that contradicts ZZ being independent in GG. Thus uRV(y+Cx)u^{-}\in R\cap V(y^{+}\overset{\rightharpoonup}{C}x^{-}). Consider next that u(ZYVyx)U2u\in(Z\cap Y\cap V_{yx}^{-})\setminus U_{2}. Then we have u+Xu^{+}\not\in X by the definition of U2U_{2}. Also, we have u+Zu^{+}\not\in Z and u+Yu^{+}\not\in Y by the same argument as above. Thus u+RV(x+Cy)u^{+}\in R\cap V(x^{+}\overset{\rightharpoonup}{C}y). Therefore we have

|(ZY)U|\displaystyle|(Z\cap Y)\setminus U| =\displaystyle= |(YZVyx+)U1|+|(ZYVyx)U2|\displaystyle|(Y\cap Z\cap V_{yx}^{+})\setminus U_{1}|+|(Z\cap Y\cap V_{yx}^{-})\setminus U_{2}|
=\displaystyle= |((ZYVyx+)U1)|+|((ZYVyx)U2)+|\displaystyle|\left((Z\cap Y\cap V_{yx}^{+})\setminus U_{1}\right)^{-}|+|\left((Z\cap Y\cap V_{yx}^{-}\right)\setminus U_{2})^{+}|
\displaystyle\leq |RV(y+Cx)|+|RV(x+Cy)||R|,\displaystyle|R\cap V(y^{+}\overset{\rightharpoonup}{C}x^{-})|+|R\cap V(x^{+}\overset{\rightharpoonup}{C}y)|\leq|R|,

as desired. ∎

Claim 4.5.

The set U{z}U\cup\{z\} is an independent set in GG.

Proof of Claim 4.5.

Since ZZ is an independent set by Claim 4.2, for any uU1u\in U_{1}, since y+u+y^{+}\sim u^{+} and y+Zy^{+}\in Z, it follows that z≁uz\not\sim u; and for any uU2u\in U_{2}, since x+u+x^{+}\sim u^{+} and x+Zx^{+}\in Z, it follows that z≁uz\not\sim u. Thus zz it not adjacent to any vertex from UU. Next, let distinct u,vUu,v\in U such that uvu\sim v. Consider first that u,vU1u,v\in U_{1}. By symmetry, we assume that uu is in between yy and vv along C\overset{\rightharpoonup}{C}. Then xCvuCy+u+Cvx+Cyzxx\overset{\leftharpoonup}{C}vu\overset{\leftharpoonup}{C}y^{+}u^{+}\overset{\rightharpoonup}{C}v^{-}x^{+}\overset{\rightharpoonup}{C}yzx is a Hamilton cycle of GG. Next consider u,vU2u,v\in U_{2}. By symmetry, we assume that uu is in between xx and vv along C\overset{\rightharpoonup}{C}. Then xCy+vCu+x+CuvCyzxx\overset{\leftharpoonup}{C}y^{+}v^{-}\overset{\leftharpoonup}{C}u^{+}x^{+}\overset{\rightharpoonup}{C}uv\overset{\rightharpoonup}{C}yzx is a Hamilton cycle of GG. Finally, consider uU1u\in U_{1} and vU2v\in U_{2}. Then xCu+y+CuvCx+v+Cyzxx\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}uv\overset{\leftharpoonup}{C}x^{+}v^{+}\overset{\rightharpoonup}{C}yzx is a Hamilton cycle in GG. Therefore, U{z}U\cup\{z\} is an independent set in GG. ∎

We show that all except at most 2t2t vertices of N(U)N(U) correspond to a vertex from UU. For this purpose, we introduce three new sets as follows.

N(U1)\displaystyle N^{*}(U_{1}) =\displaystyle= uU1(N(u,Vux+)N(u,Vux)+),\displaystyle\bigcup_{u\in U_{1}}(N(u,V^{+}_{ux})^{-}\cup N(u,V^{-}_{ux})^{+}),
N(U2)\displaystyle N^{*}(U_{2}) =\displaystyle= uU2(N(u,Vuy+)N(u,Vuy)+),\displaystyle\bigcup_{u\in U_{2}}(N(u,V^{+}_{uy})^{-}\cup N(u,V^{-}_{uy})^{+}),
N(U)\displaystyle N^{*}(U) =\displaystyle= N(U1)N(U2)\displaystyle N^{*}(U_{1})\cup N^{*}(U_{2})

We can think of the definition of N(U)N^{*}(U) above as a mapping from N(U)N(U) to vertices in N(U)+N(U)N(U)^{+}\cup N(U)^{-}. For vN(U)v\in N^{*}(U), we say that a vertex uUu\in U generates vv if vN(u,Vux+)N(u,Vux)+v\in N(u,V^{+}_{ux})^{-}\cup N(u,V^{-}_{ux})^{+} when uU1u\in U_{1}, and if vN(u,Vuy+)N(u,Vuy)+v\in N(u,V^{+}_{uy})^{-}\cup N(u,V^{-}_{uy})^{+} when uU2u\in U_{2}.

A chord of CC is an edge uvuv with u,vV(C)u,v\in V(C) and uvE(C)uv\not\in E(C). Two chords uaua and vbvb of CC that do not share any endvertices form a crossing if the four vertices u,a,v,bu,a,v,b appear along C\overset{\rightharpoonup}{C} in the order u,v,a,bu,v,a,b or u,b,a,vu,b,a,v. We say that uN(U)u\in N^{*}(U) form a crossing with v{x+,y+}v\in\{x^{+},y^{+}\} if there exist distinct vertices aN(u)a\in N(u) and bN(v)b\in N(v) such that such that uaua and vbvb are crossing chords of CC.

Claim 4.6.

For uUu\in U and v{x+,y+}v\in\{x^{+},y^{+}\}, there exist no a,bV(C)a,b\in V(C) such that abE(C)ab\in E(C), aN(U)a\in N^{*}(U), and uaua and vbvb form a crossing.

Proof of Claim 4.6.

We proceed by contradiction. Assume that u,v,au,v,a, and bb are as described in the claim. The definitions of U1U_{1} and U2U_{2} are symmetric up to reversing the direction of C\overset{\rightharpoonup}{C} and exchanging the roles of xx and yy. Thus we assume that uU1u\in U_{1} and consider two cases regarding v=x+v=x^{+} or v=y+v=y^{+} below. In each case, we construct a Hamilton cycle of GG, thereby achieving a contradiction to the assumption that GG is not Hamiltonian.

Consider first that v=x+v=x^{+}. We let a Hamilton cycle CC^{*} of GG be defined as follows according to the location of the vertex aa on C\overset{\rightharpoonup}{C}:

C=\displaystyle C^{*}= uaCy+u+CxzyCx+bCu\displaystyle ua\overset{\leftharpoonup}{C}y^{+}u^{+}\overset{\rightharpoonup}{C}xzy\overset{\leftharpoonup}{C}x^{+}b\overset{\rightharpoonup}{C}u if aVy+u+a\in V_{y^{+}u}^{+} (in this case b=a+b=a^{+}). See Figure 1(a).
C=\displaystyle C^{*}= uaCxzyCx+bCu+y+Cu\displaystyle ua\overset{\leftharpoonup}{C}xzy\overset{\rightharpoonup}{C}x^{+}b\overset{\rightharpoonup}{C}u^{+}y^{+}\overset{\leftharpoonup}{C}u if aVu+x+a\in V_{u^{+}x}^{+} (in this case b=ab=a^{-}).
C=\displaystyle C^{*}= uaCx+bCyzxCu+y+Cu\displaystyle ua\overset{\leftharpoonup}{C}x^{+}b\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}u if aVx+y+a\in V_{x^{+}y}^{+} (in this case b=a+b=a^{+}).

Consider then that v=y+v=y^{+}. We let a Hamilton cycle CC^{*} of GG be defined as follows according to the location of the vertex aa on C\overset{\rightharpoonup}{C}:

C=\displaystyle C^{*}= uaCy+bCux+CyzxCu\displaystyle ua\overset{\leftharpoonup}{C}y^{+}b\overset{\rightharpoonup}{C}u^{-}x^{+}\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}u if aVy+u+a\in V_{y^{+}u}^{+} (in this case b=a+b=a^{+}). See Figure 1(b).
C=\displaystyle C^{*}= uaCxzyCx+uCy+bCu\displaystyle ua\overset{\leftharpoonup}{C}xzy\overset{\rightharpoonup}{C}x^{+}u^{-}\overset{\rightharpoonup}{C}y^{+}b\overset{\rightharpoonup}{C}u if aVu+x+a\in V_{u^{+}x}^{+} (in this case b=ab=a^{-}).
Refer to caption
Refer to caption
Figure 1: Illustration of the cycle CC^{*}, drawn in red.

Lastly, let aVx+y+a\in V_{x^{+}y}^{+}. In this case, we have b=ab=a^{-}. Let cUc\in U be the vertex that generates aa. Then CC^{*} is constructed according to the location of cc on C\overset{\rightharpoonup}{C}:

C=\displaystyle C^{*}= uaCyzxCu+y+Ccx+CbcCu\displaystyle ua\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}c^{-}x^{+}\overset{\rightharpoonup}{C}bc\overset{\rightharpoonup}{C}u if cVy+u+c\in V_{y^{+}u}^{+}. See Figure 2.
C=\displaystyle C^{*}= uaCyzxCcbCx+cCu+y+Cu\displaystyle ua\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}cb\overset{\leftharpoonup}{C}x^{+}c^{-}\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}u if cVu+x+c\in V_{u^{+}x}^{+}.
C=\displaystyle C^{*}= uaCac+x+Cca+CyzxCu+y+Cu\displaystyle ua\overset{\leftharpoonup}{C}ac^{+}x^{+}\overset{\rightharpoonup}{C}ca^{+}\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}u if cVx+a+c\in V_{x^{+}a}^{+}.
C=\displaystyle C^{*}= uaCcbCx+c+CyzxCu+y+Cu\displaystyle ua\overset{\rightharpoonup}{C}cb\overset{\leftharpoonup}{C}x^{+}c^{+}\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}u^{+}y^{+}\overset{\rightharpoonup}{C}u if cVa+y+c\in V_{a^{+}y}^{+}.
Refer to caption
Figure 2: Illustration of the cycle CC^{*} when aVx+y+a\in V_{x^{+}y}^{+} and cVy+u+c\in V_{y^{+}u}^{+}, drawn in red.

We will show next that |N(U)|4|U||N(U)|\leq 4|U|. In this way, after deleting N(U)N(U) of at most 4|U|4|U| vertices, we will get at least |U|+1|U|+1 components and so get a contradiction to the toughness of GG.

Claim 4.7.

We have |N(U)|2t+2|U||N(U)|\leq 2t+2|U|.

Proof of Claim 4.7.

By the definition of N(U)N^{*}(U), we know that N(U)(N(U))+(N(U))N(U)\subseteq(N^{*}(U))^{+}\cup(N^{*}(U))^{-}. Now to prove Claim 4.7, it suffices to show that |N(U)U|t|N^{*}(U)\setminus U|\leq t because if this is true then we get

|N(U)|\displaystyle|N(U)| \displaystyle\leq |(N(U)U)+|+||(N(U)U)|+|U+|+|U|2t+2|U|.\displaystyle|(N^{*}(U)\setminus U)^{+}|+||(N^{*}(U)\setminus U)^{-}|+|U^{+}|+|U^{-}|\leq 2t+2|U|.

We show that |N(U)U||R(ZY)||N^{*}(U)\setminus U|\leq|R\cup(Z\setminus Y)|, which would get us the desired upper bound by the first part of Claim 4.3.

The proof requires several cases. In most cases, we show that for each distinct element of N(U)UN^{*}(U)\setminus U in the given case, there is a distinct element of R(ZY)R\cup(Z\setminus Y). Let uN(U)Uu\in N^{*}(U)\setminus U and vUv\in U such that vv generates uu. Recall that U1=YX+Vyx+U_{1}=Y\cap X^{+}\cap V^{+}_{yx} and U2=YXVyxU_{2}=Y\cap X^{-}\cap V^{-}_{yx}. Since the definitions of U1U_{1} and U2U_{2} are symmetric up to reversing the direction of C\overset{\rightharpoonup}{C} and exchanging the roles of xx and yy, we prove the case vU1v\in U_{1} only.

Consider first that uYu\not\in Y. We may assume uZu\not\in Z as otherwise uZYu\in Z\setminus Y. Now we must have uXu\not\in X since otherwise x+ux^{+}u and vuvu^{-} form a crossing if uN(v)u^{-}\in N(v) and x+ux^{+}u and vu+vu^{+} form a crossing if u+N(v)u^{+}\in N(v), contradicting Claim 4.6. Therefore uXYZu\not\in X\cup Y\cup Z and so uRu\in R. Thus in the following cases, we assume uYu\in Y. Recall that we assumed vU1v\in U_{1}.

Suppose first that uVvx+u\in V_{vx}^{+}. Then uN(U)Uu\in N^{*}(U)\setminus U implies uYX+u\not\in Y\cap X^{+}. Since uYu\in Y, we must have uX+u\not\in X^{+}. This implies that uXu^{-}\not\in X. We next claim that uYu^{-}\not\in Y, as otherwise y+uy^{+}u^{-} and vu+vu^{+} form a crossing. Thus u(ZY)Ru^{-}\in(Z\setminus Y)\cup R.

Suppose then that uVx+y+u\in V_{x^{+}y}^{+}. Then uN(U)Uu\in N^{*}(U)\setminus U implies uYXu\not\in Y\cap X^{-}. As uYu\in Y, we get u+Xu^{+}\notin X. Also, u+Yu^{+}\notin Y. Otherwise, y+uCyzxCvuCx+vCy+y^{+}u\overset{\rightharpoonup}{C}yzx\overset{\leftharpoonup}{C}vu^{-}\overset{\leftharpoonup}{C}x^{+}v^{-}\overset{\leftharpoonup}{C}y^{+} is a Hamilton cycle in GG. Thus u+R(ZY)u^{+}\in R\cup(Z\setminus Y). In particular, in this case, uyu\neq y. For otherwise, suppose u=yu=y, then vyCx+vCyzxCvvy^{-}\overset{\leftharpoonup}{C}x^{+}v^{-}\overset{\leftharpoonup}{C}yzx\overset{\leftharpoonup}{C}v is a Hamilton cycle in GG. Thus u+y+u^{+}\neq y^{+}.

Lastly, consider uVy+v+u\in V_{y^{+}v}^{+}. Then uN(U)Uu\in N^{*}(U)\setminus U implies uYX+u\not\in Y\cap X^{+}. As uYu\in Y, we must have uX+u\not\in X^{+}, which gives uXu^{-}\not\in X. By Claim 4.6, uYu^{-}\notin Y. Lastly, uZu^{-}\notin Z, as otherwise zuCx+vCuvCxzzu^{--}\overset{\leftharpoonup}{C}x^{+}v^{-}\overset{\leftharpoonup}{C}u^{-}v\overset{\rightharpoonup}{C}xz is a Hamilton cycle in GG. Thus u(ZY)Ru^{-}\in(Z\setminus Y)\cup R. Since uy+u\neq y^{+}, it follows that uyu^{-}\neq y.

The three sets Vvx+V_{vx}^{+}, Vx+y+V_{x^{+}y}^{+}, and Vy+v+V_{y^{+}v}^{+} are disjoint, we have u+y+u^{+}\neq y^{+} when uVx+y+u\in V_{x^{+}y}^{+}, and we have uyu^{-}\neq y when uVy+v+u\in V_{y^{+}v}^{+}. Thus the argument above implies that distinct vertices from N(U)UN^{*}(U)\setminus U correspond to distinct vertices from (ZY)R(Z\setminus Y)\cup R. Therefore |N(U)U||R(ZY)||N^{*}(U)\setminus U|\leq|R\cup(Z\setminus Y)|, as desired. ∎

Now, set S=N(U)S=N(U). Then |S|2t+2|U|4|U||S|\leq 2t+2|U|\leq 4|U| by Claims 4.7 and 4.4. By Claim 4.5, c(GS)|U|+1c(G-S)\geq|U|+1, where |U||U| of the components are isolated vertices from UU, and one component contains the vertex zz. This gives |S|c(GS)4|U||U|+1<4\frac{|S|}{c(G-S)}\leq\frac{4|U|}{|U|+1}<4, which contradicts that GG is 4-tough. This completes the proof of Theorem 7. \blacksquare

References

  • [1] D. Bauer, H. J. Broersma, J. van den Heuvel, and H. J. Veldman. Long cycles in graphs with prescribed toughness and minimum degree. Discrete mathematics, 141(1-3):1–10, 1995.
  • [2] J. A. Bondy and U. S. R. Murty. Graph theory with applications. American Elsevier Publishing Co., Inc., New York, 1976.
  • [3] V. Chvátal. On hamilton’s ideals. Journal of Combinatorial Theory, Series B, 12(2):163–168, 1972.
  • [4] V. Chvátal. Tough graphs and hamiltonian circuits. Discrete Mathematics, 5(3):215–228, 1973.
  • [5] C. T. Hoàng. Hamiltonian degree conditions for tough graphs. Discrete mathematics, 142(1-3):121–139, 1995.
  • [6] C. T. Hoàng and C. Robin. A closure lemma for tough graphs and hamiltonian degree conditions. Discrete Mathematics, 347(4):113872, 2024.