13
Edge Disjoint Hamilton Cycles in Graphs Guojun Li DEPARTMENT OF MATHEMATICS SHANDONG UNIVERSITY JINAN 250100, P. R. CHINA Received November 4, 1998 Abstract: Let G be a graph of order n and k 0 an integer. It is con- jectured in [8] that if for any two vertices u and v of a 2(k 1)-connected graph G,d G (u,v 2 implies that max{d(u;G),d(v ;G)} (n/2) 2k, then G has k 1 edge disjoint Hamilton cycles. This conjecture is true for k 0,1 (see cf. [3] and [8]). It will be proved in this paper that the conjecture is true for every integer k 0. ß 2000 John Wiley & Sons, Inc. J Graph Theory 35: 8–20, 2000 Keywords: Graph; Fan 2k-type graph; edge disjoint Hamilton cycles 1. INTRODUCTION All graphs under consideration are undirected, finite, and simple. Let G be a graph. We use the symbols V G and EG, respectively, to denote the vertex set and edge set of G, and dx; G to denote the degree of x in G, and G to denote the minimum degree of G. The complement of G, denoted G, is a graph with vertex set V G of which two vertices are adjacent iff they are not adjacent in G. —————————————————— Contract grant sponsor: ARO Contract grant number: DAAH04-96-1-0233 Contract grant sponsors: National Science Foundation of China; Doctoral Discipline Foundation of China. ß 2000 John Wiley & Sons, Inc.

Edge disjoint Hamilton cycles in graphs

Embed Size (px)

Citation preview

Edge Disjoint HamiltonCycles in Graphs

Guojun LiDEPARTMENT OF MATHEMATICS

SHANDONG UNIVERSITY

JINAN 250100,

P. R. CHINA

Received November 4, 1998

Abstract: Let G be a graph of order n and k� 0 an integer. It is con-jectured in [8] that if for any two vertices u and v of a 2(k� 1)-connectedgraph G,dG(u,v�� 2 implies that max{d(u;G),d(v;G)}� (n/2)� 2k, then G hask� 1 edge disjoint Hamilton cycles. This conjecture is true for k� 0,1 (seecf. [3] and [8]). It will be proved in this paper that the conjecture is true forevery integer k� 0. ß 2000 John Wiley & Sons, Inc. J Graph Theory 35: 8±20, 2000

Keywords: Graph; Fan 2k-type graph; edge disjoint Hamilton cycles

1. INTRODUCTION

All graphs under consideration are undirected, ®nite, and simple. Let G be agraph. We use the symbols V�G� and E�G�, respectively, to denote the vertex setand edge set of G, and d�x; G� to denote the degree of x in G, and ��G� to denotethe minimum degree of G. The complement of G, denoted �G, is a graph withvertex set V�G� of which two vertices are adjacent iff they are not adjacent in G.

ÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐ

Contract grant sponsor: AROContract grant number: DAAH04-96-1-0233Contract grant sponsors: National Science Foundation of China; DoctoralDiscipline Foundation of China.

ß 2000 John Wiley & Sons, Inc.

For U � V�G�, we use G�U� to denote the subgraph of G induced by U. Fora 2 V�G�;A � V�G�;B � V�G� ÿ A, we set

NA�a; G� � fx 2 A : ax 2 E�G�g; dA�a; G� � jNA�a; G�j;NA�B; G� �

[y2B

NA�y; G�; dA�B; G� � jNA�B; G�j;

EG�A;B� � fxy 2 E�G� : x 2 A and y 2 Bg; eG�A;B� � jEG�A;B�j:

Other terminology and notation unde®ned here can be found in [2].The following condition is suf®cient for a graph to have a Hamilton cycle is

due to G. Fan [3].

Theorem 1 [3]. Let G be a 2-connected graph of order n. If for any two verticesu and v of G, dG�u; v� � 2 implies that maxfd�u; G�; d�v; G�g � �n=2�, then G

has a Hamilton cycle.

In 1994, S. Zhou got a suf®cient condition for a graph to have two edgedisjoint Hamilton cycles.

Theorem 2 [8]. Let G be a 4-connected graph of order n. If for any two verticesu and v of G, dG�u; v� � 2 implies that maxfd�u; G�; d�v; G�g � �n=2� � 2, then

G has two edge disjoint Hamilton cycles.

In general, for nonnegative integer k, S. Zhou conjectured in his paper [8] thatif for any two vertices u and v of a 2�k � 1�-connected graph G; dG�u; v� � 2implies that maxfd�u; G�; d�v; G�g � �n=2� � 2k, then G has k � 1 edge disjointHamilton cycles. Simultaneously, the graphs satisfying the above degreecondition were refered to as Fan 2k-type graphs. So the above conjecture canbe written as follows.

Conjecture 1 [8]. Every 2�k � 1�-connected Fan 2k-type graph has k � 1 edgedisjoint Hamilton cycles.

Theorem 1 and Theorem 2 imply that the conjecture is true for k � 0; 1. Beforejustifying Conjecture 1 for general k, we would like to recall two other interestingresults (cf [4] and [5]) concerning edge disjoint Hamilton cycels. In particular, thetechnique used in [5] is useful in solving Conjecture 1.

Theorem 3 [4]. Let G be a graph of order n and k be a positive integer.

If d�u; G� � d�v; G� � n� 2k ÿ 2 holds for each pair of nonadjacent verticesu and v, then G has k edge disjoint Hamilton cycles whenever n is suf®ciently

large.

Theorem 4 [5]. Let G be a graph of order n and k be a positive integer. If2k � 1 � ��G� � 2k � 2; n � 8k2 ÿ 5 and d�u; G� � d�v; G� � n for each pair of

nonadjacent vertices u and v, then G has k edge disjoint Hamilton cycles.

The following theorem will be used frequently in the proof of main results.

HAMILTON CYCLES IN GRAPHS 9

Theorem 5 [1]. Kn�n � 2k � 1� has k edge disjoint Hamilton cycles;Kn�n � 2k� has k edge disjoint Hamilton paths having any given k pairs of

vertices, which are mutually disjoint, as their endvertices respectively.

2. PRELIMINARIES

In this section, we suppose that G is a 2�k � 1�-connected graph, andV�G� � X [ Y is a partition of V�G� with jXj < 2�k � 1� and jY j � 2�k � 1�.Further, we suppose that both G�x� and G�Y � are complete. Let us nowdemonstrate a number of properties of such a graph which are useful for us tojustify Conjecture 1.

Proposition 1. Let G be a 2�k � 1�-connected graph, and let V�G� � X [ Y be

a partition of V�G� with jXj < 2�k � 1�; jY j � 2�k � 1�, and both of G�X� andG�Y � be complete. If, for any vertex x 2 X; d�x; G� � 2k � 3, then there is a vertex

y 2 NY�x; G� such that Gÿ fxyg is still 2�k � 1�-connected.

Proof. Suppose that there exists x 2 X with d�x; G� � 2k � 3 such thatGÿ fxyg is not 2�k � 1�-connected for any y 2 NY�x; G�. Since jXj < 2�k � 1�and d�x; G� � 2k � 3, we can take y; z 2 NY�x; G� with y 6� z. Let S � V�G� be acutset of Gÿ fxyg with jSj � 2k � 1. Then we can write V�G� ÿ S � A [ B sothat A 6� ;;B 6� ;;A \ B � ;, and EGÿfxyg�A;B� � ;. Since G is 2�k � 1�-connected, we may assume x 2 A and y 2 B.

Similarly, let T be a cutset of Gÿ fxzg with jTj � 2k � 1, and writeV�G� ÿ T � C [ D so that C 6� ;;D 6� ;; x 2 C; z 2 D;C \ D � ;, andEGÿfxzg�C;D� � ;. Since G�X� and G�Y � are complete, B;D � Y and A;C � X,and hence B \ C � A \ D � ;. Since x 2 C; xy 2 E�Gÿ fxzg�, and y 2 B, weget y 2 B \ T from B \ C � ;, and we similarly get z 2 S \ D. LetP � �S \ T� [ �S \ C� [ �A \ T� and P � �S \ T� [ �S \ D� [ �B \ T�. Then inGÿ fxy; xzg;P separates A \ C and B [ D. If A \ C � fxg, then this impliesjPj � d�x; G� ÿ 2 � 2k � 1; If A \ C 6� fxg, then in G;P [ fxg separates

10 JOURNAL OF GRAPH THEORY

A \ C ÿ fxg and B [ D, and hence jPj � 2�k � 1� ÿ 1 by the assumptionthat G is 2�k � 1�-connected. Thus in either case, jPj � 2k � 1. SincejPj � jQj � jSj � jT j � 4k � 2, this implies jQj � 2k � 1. On the other hand, Qseparates B \ D and A [ C in Gÿ fxy; xzg. Since y; z 2 Q, this means that Q

separates B \ D and A [ C in G. Since jQj � 2k � 1, this forces B \ D � ;. SinceY � �S [ B� \ �T [ D� by the assumption that G�Y � is complete, we now obtainjYj � j�S [ B� \ �T [ D�j � jQj � 2k � 1, which contradicts the assumption thatjYj � 2�k � 1�. Proposition 1 is proved. &

Proposition 2. Let G be a graph, and let V�G� � X [W [ Z be a partition ofV�G� with jX [W j � 2h and jZj � 2. Suppose that G�W [ Z� is complete, and

dZ�x; G� � 1 �1�

for all x 2 X. Suppose further that

dX�y; G� � h �2�

for all y 2 Z. Then X [W can be partitioned into h pairs fu1; v1g; . . . ; fuh; vhgsuch that dZ�fui; vig; G� � 2 for all 1 � i � h:

Proof. We proceed by induction of h. If h � 0, there is nothing to be proved.Thus let h � 1, and assume the proposition holds for hÿ 1. LetZ0 � fy 2 ZjdX�y; G� > 0g. If Z0 � ;, then X � ; by (1), and hence theproposition immediately follows from the assumption that G�W [ Z� is complete.Thus we may assume Z0 6� ;. Assume ®rst that jZ0j � 1. Write Z0 � fy0g. Thenby (1) and (2), jXj � dX�y0; G� � h, and hence W 6� ;. Take u1 2 X and v1 2 W .Then dZ�fu1; v1g; G� � jZj � 2. Further, applying the induction hypothesis toGÿ fu1; v1g, we see that �X [W� ÿ fu1; v1g can be partitioned into hÿ 1 pairshaving the desired property. Assume now that jZ0j � 2. Choose y0; z0 2 Z0 withy0 6� z0 so that dX�y0; G� � dX�z0; G� and so that dX�y; G� � dX�z0; G� for ally 2 Z ÿ fy0; z0g. Then by (1),

dX�y; G� � jXj3

� �� 2h

3

� �� hÿ 1 �3�

for all y 2 Z ÿ fy0; z0g. Take u1 2 NX�y0; G� and v1 2 NX�z0; G�. ThendZ�fu1; v1g; G� � 2. Further by (3), we can apply the induction hypothesis toGÿ fu1; v1g, to see that �X [W� ÿ fu1; v1g can be partitioned into hÿ 1 pairshaving the desired property. Proposition 2 is proved. &

Proposition 3. Let G be a 2�k � 1�-connected graph, and let V�G� � Y [ X be

a partition of V�G� with X 6� ; and jY j � 2�k � 1�. If G�X� and G�Y� are bothcomplete, then G has k � 1 Hamilton cycles C1;C2; . . . ;Ck�1 such that

E�Ci� \ E�Cj� � E�G�Y �� for any i, j with i 6� j.

HAMILTON CYCLES IN GRAPHS 11

Proof. The proof is by contradiction. Let G be a counterexample withminimal number of edges. Then jXj < 2k � 2 since otherwise the 2�k � 1�-connectedness of G guarrantees that there exists a matching with at least 2k � 2edges between X and Y, and then Theorem 5 implies that G has k � 1 edgedisjoint Hamilton cycles. Similarly, if jXj � 1, then it follows from Theorem 5that G has k � 1 edge disjoint Hamilton cycles, which contradicts the assumptionthat G is a counterexample. Thus we also have jXj � 2.

Now that jXj < 2k � 2 and jYj � 2k � 2, we can choose a proper subset W of Ysuch that jX [Wj � 2k � 2, and that eG�X;W� is as large as possible. For thesake of convenience, we write Z � Y ÿW . From jY j � 2k � 2 � jX [W j, we getjZj � jXj � 2.

Since G is a counterexample with minimal number of edges, by Proposition 1,we see that d�x; G� � 2k � 2 for all x 2 X.

Now we label the vertices in Z and W, respectively, as follows.Z : y1; y2; . . . ; ys

W : w1;w2; . . . ;wt

such that

dX�y1; G� � dX�y2; G� � � � � � dX�ys; G�:

Subsequently, we construct a sequence of graphs as follows:

X11 � fx 2 X : xy1 2 E�G� and xw1 62 E�G�g;G11 � �Gÿ EG�X11; y1�� [ E�G�X11;w1�;X12 � fx 2 X : xy1 2 E�G11� and xw2 62 E�G11�g;G12 � �G11 ÿ EG11

�X12; y1�� [ E�G11�X12;w2�;

..

.

X1t � fx 2 X : xy1 2 E�G1;tÿ1� and xwt 62 E�G1;tÿ1�g;G1t � �G1;tÿ1 ÿ EG1;tÿ1

�X1t; y1�� [ E�G1;tÿ1�X1t;wt�;

X21 � fx 2 X : xy2 2 E�G1t� and xw1 62 E�G1t�g;G21 � �G1t ÿ EG1t

�X21; y2�� [ E�G1t�X21;w1�;

X22 � fx 2 X : xy2 2 E�G21� and xw2 62 E�G21�g;G22 � �G21 ÿ EG21

�X22; y2�� [ E�G21�X22;w2�;

..

.

X2t � fx 2 X : xy2 2 E�G2;tÿ1� and xwt 62 E�G2;tÿ1�g;G2t � �G2;tÿ1 ÿ EG2;tÿ1

�X2t; y2�� [ E�G1;tÿ1�X2t;wt�;

..

. ... ..

.

Xs1 � fx 2 X : xys 2 E�Gsÿ1;t� and xw1 62 E�Gsÿ1;t�g;Gs1 � �Gsÿ1;t ÿ EGsÿ1;t

�Xs1; ys�� [ E�Gsÿ1;t�Xs1;w1�;

Xs2 � fx 2 X : xys 2 E�Gs1� and xw2 62 E�Gs1�g;Gs2 � �Gs1 ÿ EGs1

�Xs2; ys�� [ E�Gs1�Xs2;w2�;

..

.

Xst � fx 2 X : xys 2 E�Gs;tÿ1� and xwt 62 E�Gs;tÿ1�g;Gst � �Gs;tÿ1 ÿ EGs;tÿ1

�Xst; ys�� [ E�Gs;tÿ1�Xst;wt�:

12 JOURNAL OF GRAPH THEORY

It follows from the construction of Gst that Gst�X [W � is complete, and thatdZ�x; Gst� � 1 for all x 2 X since d�x; Gst� � 2k � 2 and jX [W j � 2k � 2, andthat Gst�Y � is still complete.

The following fact follows instantly from the construction of Gst.

Claim 1. Let x and yi be two vertices respectively in X and Z, and xyi be an edge

of Gst. If yj 2 Z; j 6� i, is adjacent to x in G then j < i.

The following claim will play an important role in the rest of the proof ofProposition 3.

Claim 2. Gst has k � 1 Hamilton cycles C1;C2; . . . ;Ck�1 such that

E�Ci� \ E�Cj� � E�Gst�Z��for any i,j with i 6� j:

Proof. The proof is by contradiction. Suppose that the claim is not true.Noting that Gst satis®es all conditions of Proposition 2 apart from it's condition(2), we see that there is a vertex yi0 2 Z such that dX�yi0 ;G� � k � 2. Sinceotherwise, Proposition 2 guarantees that X [W can be partitioned into k � 1 pairsfu1; v1g; . . . ; fuk�1; vk�1g such that dZ�fui; vig; G� � 2 for all 1 � i � k � 1, andTheorem 5 ensures that G contains k � 1 Hamilton cycles satisfying (4), acontradiction. Let

M � NX�yi0 ; Gst�; m � jMj;L � X ÿM; l � jLj:

By the arrangement of vertices of Z, we get dX�yi;G� � dX�yi0 ; G� �dX�yi0 ; Gst� � m for all i < i0. Set

U � fyi 2 Z ÿ fyi0g : NM�yi;G� 6� ;g:By Claim 1, U � fy1; y2; . . . ; yi0ÿ1

g, and hence dX�y; G� � m for all y 2 U.If jUj � 2k ÿ t ÿ l, since jU [W [ L [ fyi0gj � 2k ÿ t ÿ l� t � l� 1 �

2k � 1 and jY j � 2k � 2, we know that U [W [ L [ fyi0g is a cut set of G,and so that ��G� � jU [W [ L [ fyi0gj � 2k � 1, contradicting the fact that G is2�k � 1�-connected. So we may assume that jUj � 2k ÿ t ÿ l� 1.

Keeping t � l� m � 2k � 2 and m � k � 2 in mind, we know that t � l � k.By the fact that we have chosen W so that eG�W ;X� is as large as possible, wehave that dX�w; G� � dX�yi0 ; G� � dX�yi0 ; Gst� � m for all w 2 W . It follows that

tjXj � eGst�X;W� � eG�W ;X� � jUjm� dX�yi0 ; G� ÿ jXj

� tm� �2k ÿ t ÿ l� 1�m� mÿ jXj� tm� �k � 1�m� mÿ �2k � 2ÿ t�� �m� 1�t � km� 2mÿ 2k ÿ 2 � �m� 1�t � km� 2�k � 2� ÿ 2k ÿ 2

> 2mt � t�2k � 4� > tjXj:

HAMILTON CYCLES IN GRAPHS 13

This ®nal contradiction means that Gst has k � 1 Hamilton cycles satisfying (4).Claim 2 is proved. &

By Claim 2, we can choose p and q such that(a) Gpq has k � 1 Hamilton cycles C1;C2; . . . ;Ck�1 satisfying

E�Ci� \ E�Cj� � E�Gpq�Z��

for any i, j with i 6� j.(b) subject to (a), p is as small as possible;(c) subject to (a) and (b), q is as small as possible.

Since G contains no k � 1 Hamilton cycles satisfying (1), we know that G 6� Gpq.Set

G0 � Gp;qÿ1; q > 1;Gpÿ1;t; q � 1:

�Then it follows from the generation of Gpq that Gpq � �G0 ÿ EG0 �Xpq; yp��[E�G0 �Xpq;wq�. By the choice of p and q, we know that Xpq 6� ;. For the sake ofconvenience, we use X1; y1 and w1, respectively, to denote Xpq; yp and wq. LetX1 � fx1; x2; . . . ; xdg. Then xiy1 2 E�G0� and xiw1 62 E�G0� for 1 � i � d. WriteX2 � fz 2 Xjzy1; zw1 2 E�G0�g. Now for the cycle

Ci � a1a2 � � � ana1;

where al � w1; as � y1; of Gpq, we use Qi and Qi�k�1 respectively to denote thepaths alal�1 � � � as and asas�1 � � � al. Then Q1;Q2; . . . ;Qk�1;Qk�2; . . . ;Q2�k�1� are2�k � 1� paths in Gpq connecting y1 and w1, and they satisfy

E�Qi� \ E�Qj� � E�Gpq�Z��:

Note that for 1 � i � 2�k � 1�, the vertices adjacent to y1 and w1 in Qi

respectively are determined uniquely. Set

G0 � G0 ÿ[2�k�1�

i�1

�E�Qi� \ �E�G0� ÿ E�G0�Z����:

We are going to construct k � 1 Hamilton cycles, O1;O2; . . . ;Ok�1, in G0 suchthat E�Oi� \ E�Oj� � E�G0�Z�� by replacing in C1;C2; . . . ;Ck�1 the edges in[2�k�1�

i�1 E�Qi� ÿ E�G0�, which will contradict the choice of p and q.By the generation of Gpq;[2�k�1�

i�1 E�Qi� ÿ E�G0� � EGpq�w1;X1�. Suppose there

is some xj 2 X1 such that xjw1 2 Qj1 . We let zj;1 � xj, and put

Hj1 � �Qj1 ÿ fzj;1w1; zj;2y1g� [ fzj;2w1; zj;1y1g;

14 JOURNAL OF GRAPH THEORY

where zj;2 is the vertex joined to y1 on Qj1 . Since zj;2y1 2 E�Gpq�, we havezj;2 2 X2 [ Y and hence zj;2w1 2 E�G0� \ E�Gpq�. Thus Hj1 is a path in G0

connecting y1 and w1. Therefore, Hj1 [ Qk�1�j1 is a Hamilton cycle of�Gpq ÿ fzj;1w1g� [ fzj;1y1g.

Now suppose zj;2w1 is an edge of some Qj2 and zj;2 2 X. Note that zj;2 2 X2. Weput

Hj2 � �Qj2 ÿ fzj;2w1; zj;3y1g� [ fzj;3w1; zj;2y1g;

where zj;3 is the vertex adjacent to y1 on Qj2 ; and so on.In general, we have found zj;1; zj;2; . . . ; zj;s� j� and Qj1 ;Qj2 ; . . . ;Qjs� j�ÿ1

; s� j� � 2,such that

(i) zj;1 � xj 2 X1; zj;l 2 X2, for 2 � l � s� j� ÿ 1;(ii) zj;lw1 and zj;l�1y1 are on Qjl ; 1 � l;� s� j� ÿ 1;

(iii) zj;l 6� zj;d and Qjl 6� Qjd , for any 1 � l 6� d � s� j� ÿ 1; and(iv) subject to (i), (ii) and (iii), s� j� is as large as possible.

Now we claim that

zj;s� j� 62 fzj;l : 1 � l � s� j� ÿ 1g: �5�

In fact, if zj;s� j� � zj;l for some 1 � l � s� j� ÿ 1, then from zj;s� j�y1 2 E�Gpq�and zj;1y1 62 E�Gpq�, we get l 6� 1, and hence it is clear from the property (ii) thatQjs� j�ÿ1

� Qjlÿ1and hence zj;s� j�ÿ1 � zj;lÿ1, contradicting the property (iii).

Moreover, if zj;s�j� 2 X2 and there exists some Qjs� j� such that zj;s� j�w1 2 Qjs� j� ,then there exists some zj;s� j��1 on Qjs�j� such that zj;s� j��1y1 2 Qjs� j� . IfQjs� j� 6� Qjl ; 1 � l � s� j� ÿ 1, we have a contradiction to the property (iv). Andif Qjs� j� � Qjl for some l; 1 � l � s� j� ÿ 1, then by the uniqueness of the verticesjoined to w1 and y1, respectively, on each Qi, we have zj;s� j� � zj;l, contradictingthe formula (5).

The above argument and the fact that zj;s� j�y1 2 E�Qjs� j�ÿ1� show that either

zj;s� j�2Y , or zj;s� j� 2 X2 and zj;s� j�w1 2 E�G0�. Thus we de®ne in general

Hjl � �Qjl ÿ fzj;lw1; zj;l�1y1g� [ fzj;l�1w1; zj;ly1g; for 1 � l � s� j� ÿ 1: �6�

We conclude from this de®nition that these Hjl ; 1 � l � s� j� ÿ 1, are pathsconnecting y1 and w1 and pairwise having no common edge in E�G0� ÿ E�G�Z��.Moreover, since fzj;ly1; zj;l�1w1 : 1 � l � s� j� ÿ 1g � [s� j�ÿ1

l�1 E�Qjl� [ E�G0�,each of Hjl ; 1 � l � s� j� ÿ 1, has in E�G0� ÿ E�G�Z�� no common edge withany of the remaining Qi; i 2 f1; 2; . . . ; 2�k � 1�g ÿ f j1; j2; . . . ; js� j�ÿ1g.

If xi and xj are two different elements of X1 such that xiw1 is on some Qi1

and xjw1 is on some Qj1 , then we may get xi � zi;1; zi;2; . . . ; zi;s�i�;Qi;1;Qi;2; . . . ;Qi;s�i�ÿ1 and xj � zj;1; zj;2; . . . ; zj;s� j�;Qj1 ;Qj2 ; . . . ;Qjs� j�ÿ1

respectively, bythe above procedure.

HAMILTON CYCLES IN GRAPHS 15

We claim that zi;l 6� zj;d for any 1 � l � s�i� and any 1 � d � s� j� (except thatit is possible that zi;s�i� � zj;s� j� if zi;s�i�; zj;s� j� 2 Y�, and Qil 6� Qjd for any1 � l � s�i� ÿ 1 and any 1 � d � s� j� ÿ 1. Otherwise, suppose that l is theminimal number such that there exist some d; 1 � d � s� j�, with zi;l � zj;d. Ifd � 1, then since zi;ry1 2 E�Gpq� for each 2 � r � s�i� and zj;1y1 62 E�Gpq�, weget l � 1, which contradicts the assumption that xi 6� xj. Thus d 6� 1. Similarlyl 6� 1. Now, since the vertices adjacent to w1 and y1, respectively, on each Qi arerespectively unique, it is easy to see from the property (ii) that Qilÿ1

� Qjdÿ1

(unless l � s�i�; d � s� j� and zi;s�i�; zj;s� j� 2 Y�, and hence zi;lÿ1 � zj;dÿ1, contra-dicting the choice of l.

Based on the above discussion, for any xj in X1 such that xjw1 is on some Qj1,we de®ne Hjl ; 1 � l � s� j� ÿ 1, as in (6). Put Hj � Qj for the remaining Qj thatdo not appear in the above procedure. Then we obtain H1;H2; . . . ;H2�k�1� in G0,which are pairwise edge disjoint in E�G0� ÿ E�G�Z��. By this de®nition and theabove argument, we put

Oi � Hi [ Hi�k�1; for 1 � i � k � 1;

which are k � 1 Hamilton cycles of G0 such that E�Oi� \ E�Oj� � E�G0�Z��,contradicting the choice of p and q. Proposition 3 is proved. &

3. PROOF OF CONJECTURE 1

In this section we turn to justify the following theorem which was everconjectured by Zhou [8].

Theorem 6. Every 2�k � 1�-connected Fan 2k-type graph has k � 1 edge

disjoint Hamilton cycles.

Proof. The proof is by contradiction. Suppose that G is a 2�k � 1�-connectedFan 2k-type graph, but G contains no k � 1 edge disjoint Hamilton cycles.Furthermore, let G be a counterexample with the maximum number of edges, i.e.,the graph G itself contains no k � 1 edge disjoint Hamilton cycles, but for eachpair of nonadjacent vertices u and v, either G� fuvg contains k � 1 edge disjointHamilton cycles or G� fuvg is no longer a Fan 2k-type graph. Put

Y � v 2 V�G� : d�v; G� � n

2� 2k

n o:

Since G is 2�k � 1�-connected, n � 2k � 3. By Theorem 5 and the fact that G is acounterexample, we know that G is not complete, and that Y 6� ;.

The proof of Theorem 6 can be divided into a series of the following lemmas.

Lemma 1. G[Y] is complete.

16 JOURNAL OF GRAPH THEORY

Proof. Suppose, to the contrary, that there are two vertices x; y 2 Y such thatxy 62 E�G�. It is easy to verify that G� fxyg is still a Fan 2k-type graph. By theassumption that G is a maximum counterexample, G� fxyg must have k � 1edge disjoint Hamilton cycles C1;C2; . . . ;Ck�1, one of them, say Ck�1, passesthrough the edge fxyg in G� fxyg. Let

G0 � Gÿ[ki�1

E�Ci�:

Then d�x; G0� � d�y; G0� � d�x; G� � d�y; G� ÿ 4k � n. By the proof of Ore'stheorem [6], there are two consecutive vertices u and v along the pathCk�1 ÿ fxyg from the starting vertex x to the terminating vertex y such thatboth xv and yu are edges of G0, and therefore Ck�1 ÿ fxy; uvg � fxv; yug is aHamilton cycle in G0 and so in G which is certainly disjoint from the cyclesC1;C2; . . . ;Ck, contradicting the assumption that G is a counterexample. Nowwe write &

X � V�G� ÿ Y :

By Lemma 1, Theorem 5, and the fact that G is a counterexample, we knowthat X 6� ;. We use Gi � �Xi;Ei�; 1 � i � !, to denote the components ofG�X�, and let Yi � NY�Xi; G�; 1 � i � !. By the hypothesis, there is not apair of vertices in X at distance two. Consequently, the following lemma istrivial.

Lemma 2. (1) Gi�1 � i � !� is complete; (2) Yi \ Yj � ;; i 6� j:

To justify Theorem 6, the following lemma is useful.

Lemma 3. jYij � 2�k � 1�; i � 1; 2; . . . ; !:

Proof. Suppose, to the contrary, that jYij < 2�k � 1�. Then since G is2�k � 1�-connected and Xi 6� ;, we know that ! � 1, i.e., Xi � X and Yi � Y . G

being incomplete guarrantees that there is at least one vertex y 2 Y which is notadjacent to some vertex in X. Therefore, we get the inequality

jY j � d�y; G� � 1ÿ �jXj ÿ 1� � n

2� 2k ÿ jXj � 2:

It follows from jXj � jY j � n that n � 4k � 4, and from jY j � 2k � 1 thatjXj � nÿ jY j � 4k � 4ÿ �2k � 1� � 2k � 3 � jY j � 2. So we obtain theinequality

jY jjXj ÿ 1 � ÿ 2

jXj.Using the equality

Px2X dY�x; G� �Py2Y dX�y; G�, we can derive the

following inequality jXj maxx2X dY�x; G� � jY j miny2YdX�y; G�:

HAMILTON CYCLES IN GRAPHS 17

If follows from the above inequalityjY jjXj ÿ 1 � ÿ 2

jXj < 0 that

maxx2X

dY�x; G� � miny2Y

dX�y; G� � jYjjXj ÿ 1

� �miny2Y

dX�y; G�

� miny2Y

dX�y; G� � jYjjXj ÿ 1

� �jXj

� miny2Y

dX�y; G� � jYj ÿ jXj:

Then a contradiction will be derived from the series of following inequalities.

n

2� 2k > max

x2Xd�x; G� � jXj ÿ 1�max

x2XdY�x; G�

� jXj ÿ 1�miny2Y

dX�y; G� � jY j ÿ jXj� jYj ÿ 1�min

y2YdX�y; G�

� miny2Y

d�y; G� � n

2� 2k: &

The following lemma, which will be the key for justifying Theorem 6, followsinstantly from Proposition 3.

Lemma 4. For i � 1; 2; . . . ; !;G�Yi [ Xi� has k � 1 Hamilton cycles

Ci1 ;Ci2 ; . . . ;Cik�1such that

E�Cis� \ E�Cit� � E�G�Yi��

for any s,t with s 6� t.

Lemma 5.

E�Ci; j� \ E�G�Y �� 6� ;; i � 1; 2; . . . ; !; j � 1; 2; . . . ; k � 1:

Proof. If jXij � 2k � 2, by the proof of Proposition 3, each Hamilton cycleCi;j; j � 1; 2; . . . ; k � 1 in G�Xi [ Yi� is constituted by two independent edgesbetween Xi and Yi, and a Hamilton path in G�Xi�, and a Hamilton path in G�Yi�. SoE�Ci;j� \ E�G�Y �� 6� ;; j � 1; 2; . . . ; k � 1:

If jXij < 2k � 2, since jYij � 2k � 2 there must be two consecutive vertices oneach cycle Ci; j; j � 1; 2; . . . ; k � 1 being in Yi. Hence E�Ci;j� \ E�G�Y �� 6� ;;j � 1; 2; . . . ; k � 1. &

Lemma 6. If M is a matching of Kn, then Kn has a Hamilton cycle passing

through all edges of M.

18 JOURNAL OF GRAPH THEORY

Lemma 7. G contains k � 1 Hamilton cycles C1;C2; . . . ;Ck�1 such that

E�Ci� \ E�Cj� � E�G�Y �� �7�

Proof. In Lemma 5, let j � 1. Then Lemma 5 guarantees that there existsy0i;1y00i;1 2 E�G�Yi�� \ E�Ci;1� for 1 � i � !. Write Y0 � Y ÿ [!i�1Yi.

Then by Lemma 6, G�Y0 [ �S!

i�1fy0i;1; y00i;1g�� has a Hamilton cycle passingthrough the edges fy0i;1y00i;1 : 1 � i � !g. Now for each i; 1 � i � !, we substitutethe path Ci;1 ÿ fy0i;1y00i;1g for the edge y0i;1y00i;1 to get a Hamilton cycle C1 of G. Forj � 2; . . . ; k � 1, using the same method, we can then get the cycles C2; . . . ;Ck�1.By Lemma 3, we can see that the k � 1 cycles are just our desired. &

Lemma 8. G has k � 1 edge disjoint Hamilton cycles if and only if G has k � 1Hamilton cycles satisfying (7).

Proof. It suf®ces to show the suf®ciency. Let C1;C2; . . . ;Ck�1 be k � 1Hamilton cycles satisfying (7) such that p �Pi6�j jE�Ci� \ E�Cj� \ E�G�Y ��j isas small as possible. We assume, without loss of generality, that p 6� 0 andxy 2 E�C1� \ E�C2� \ E�G�Y��. Set

G0 � Gÿ[k�1

i�2

Ci;

Then dG0 �x� � dG0 �y� � n. By the proof of Ore's theorem [6], we know that thereare two consecutive vertices u and v on C1 such that both xu and yv are the edgesof G0 and C01 � �C1 ÿ fxy; uvg� � fxu; yvg is a Hamilton cycle. LetC0i � Ci; 2 � i � k � 1. Then C01;C

02; . . . ;C0k�1 are k � 1 Hamilton cycles of G

satisfying (7) andXi6�j

jE�C0i� \ E�C0j� \ E�G�Y ��j � pÿ 1 < p;

contradicting the choice of C1;C2; . . . ;Ck�1. &

Lemma 7, together with Lemma 8, ensures that G has k � 1 edge disjointHamilton cycles, violating our assumption at the beginning. Theorem 6 is proved.

&

We close this paper by pointing out that the 2�k � 1�-connectedness inTheorem 6 is essential. For example, consider the graph G consisting of twovertex disjoint complete graphs K2k�1 and Kp�p � 6k � 3� and a matchingwith 2k � 1 edges between the vertex sets of K2k�1 and Kp. Evidently G is a2k � 1-connected Fan 2k-type graph, but contains no k � 1 edge disjointHamilton cycles.

HAMILTON CYCLES IN GRAPHS 19

ACKNOWLEDGMENT

I am sincerely grateful to an anonymous referee who presented novel short proofsfor Proposition 1 and Proposition 2, and many useful suggestions for the polish ofthe manuscript.

REFERENCES

[1] L. W. Beineke and R. L. Wilson, Selected Topics in Graph Theory, AcademicPress, New York, 1978, 161 p.

[2] J. A. Bondy, U. S. R. Murty, Graph Theory with Applications, MacMillan,London, 1976.

[3] G. H. Fan, New suf®cient condition for cycles in graphs, J Combin TheoryB37 (1984), 221±227.

[4] R. J. Faudree, C. C. Rousseau, and R. H. Schelp, Edge disjoint Hamiltoniancycles. Graph Theory with Applications to Algorithms and Computer Science(Kalamazoo, Mich; 1984), Wiley-Interscience New York (1985) 231±249,MR 87c:05104.

[5] H. Li, Edge disjoint cycles in graphs, J Graph Theory 3 (1989), 313±322.

[6] O. Ore, Note on Hamilton circuits. Am Math Monthly 67 (1960), 55.

[7] F. Tian, A short proof of a theorem about the circumference of a graph,J Combin Theory B45 (1988), 373±375.

[8] S. Zhou, Disjoint Hamiltonian cycles in Fan 2k-type graphs, J Graph Theory 3(1989), 313±322.

20 JOURNAL OF GRAPH THEORY