Subdivisions of vertex-disjoint cycles in bipartite
graphs ††thanks: Supported by NSFC (No. 11001214).
Shengning Qiao
School of Mathematics and Statistics,
Xidian University,
Xi’an 710071, Shaanxi, China
Bing
Chen
Department of Applied Mathematics, School of Science,
Xi’an University of Technology,
Xi’an 710048, Shaanxi, China
Corresponding author. E-mail address: [email protected] (S. Qiao).
Abstract
Let n≥6,k≥0 be two integers. Let H be a graph of order
n with k components, each of which is an even cycle of length at
least 6 and G be a bipartite graph with bipartition (X,Y)
such that ∣X∣=∣Y∣≥n/2. In this paper, we show that if the
minimum degree of G is at least n/2−k+1, then G contains a
subdivision of H. This generalized an older result of Wang.
Keywords: subdivision; vertex-disjoint cycles;
bipartite graph
1 Introduction
We use Bondy and Murty [2] for terminology and notation not
defined here and consider finite simple graphs only.
Let G be a graph. A set of subgraphs of G is said to be vertex-disjoint if no two of them have a common vertex in G. Wang
[3] considered the minimum degree condition for existence
of vertex-disjoint large cycles in a bipartite graph, as follows.
Theorem 1** (Wang [3]).**
Let G be a bipartite graph with bipartition (X,Y) such that
∣X∣=∣Y∣≥sk, where s≥3 and k≥1 are two integers.
If the minimum degree of G is at least (s−1)k+1, then G
contains k vertex-disjoint cycles of length at least 2s.
A *subdivision * of a graph H is a graph that can be obtained
from H by a sequence of edge subdivisions. Let H be a graph on
n vertices and k non-trivial components such that every
component contains at most one cycle. A *cyclic subdivision * of
H is one such that only cyclic edges of H are subdivided.
Recently, Babu and Diwan [1] gave the following result.
Theorem 2** (Babu and Diwan [1]).**
Let H be a graph with n vertices and k non-trivial components
such that every component contains at most one cycle. Let G be a
graph with at least n vertices. If the minimum degree of G is at
least n−k, then G contains a cyclic subdivision of H.
Motivated by Theorem 2, this paper focuses on the analogous problem
for existence of subdivisions of vertex-disjoint cycles in bipartite
graphs. Our result is as follows.
Theorem 3**.**
Let H be a graph of order n with k components, each of which
is an even cycle of length at least 6. Suppose that G is a
bipartite graph with bipartition (X,Y) such that ∣X∣=∣Y∣≥n/2. If the minimum degree of G is at least n/2−k+1, then G
contains a subdivision of H.
It is easy to see that Theorem 3 generalized Theorem 1. In fact, we
conjecture the following.
Conjecture 1**.**
Let H be a graph of order n with k components, each of which
is an even cycle. Suppose that G is a bipartite graph with
bipartition (X,Y) such that ∣X∣=∣Y∣≥n/2. If the minimum
degree of G is at least n/2−k+1, then G contains a subdivision
of H.
If the Conjecture is true, then the minimum degree condition is
sharp. Let k be an even. Let H be an union of k cycles such
that k−1 cycles of length 4 and a cycle of length 6. Let G be a
bipartite graph with bipartition (X1∪X2∪{u},Y1∪Y2∪{v}) such that the two induced subgraphs by X1∪Y1
and X2∪Y2 are isomorphic to Kk,k. Further, u is
adjacent to every vertex in Y1∪{v} and v is adjacent to
every vertex in X2∪{u}. The vertices of X1 are matched
with vertices of Y2 by k independent edges of G. Clearly, G
contains exactly 4k+2 vertices and the minimum degree of G is
k+1<k+2. It is not difficult to see that if G contains H then
the edge uv will be contained by the cycle of length exactly 6.
Let C be the cycle of length exactly 6 with contained the edge
uv. But, G−V(C) does not contain a spanning subgraph with a
union of k−1 cycles of length 4.
We postpone the proof of Theorem 3 to the next section.
2 Proof of Theorem 3
We first introduce some further terminology and notations that will
be used later.
Let G be a graph. The order of G is denoted by ∣G∣. For a
vertex v and a subgraph H of G, we use N(v,H) and d(v,H)
to denote the set of vertices and the number of vertices in H
which are adjacent to v, respectively. Thus d(v,G) is the degree
of v in G. By δ(G) we denote the minimum degree of G.
For a subset S of the vertices or the edges in G, we use G[S]
to denote the subgraph of G induced by S. If v∈V(H),
then we denote G[V(H)∪{v}] by H+v. If v∈V(H), then we
denote G[V(H)∖{v}] by H−v. By G−H we denote
G[V(G)∖V(H)]. Let x,y be two nonadjacent vertices in
G. We use G+xy to denote the graph by adding the edge xy in
G. Let C be a cycle with a given orientation. We use v+
(resp. v−) to denote the successor (resp. the predecessor) of v
along C according to this orientation.
Proof of Theorem 3. By contradiction. Suppose that G is a
graph satisfying the conditions of the theorem but containing no
subdivisions of H such that the number of edges of G is as large
as possible. This implies that that G is not a complete bipartite
graph. Let a∈X and b∈Y be two nonadjacent vertices of G.
Denote the components of H by C1,C2,…,Ck. It follows
from the choice of G that G+ab contains a subdivision H∗ of
H. Let Ci∗ be the subdivision of Ci in H∗ for 1≤i≤k. Without loss of generality, we assume that ab∈E(Ck∗). Then G contains a subdivision of H−Ck with
[TABLE]
Let G1 be the subgraph of G induced by
⋃i=1k−1V(Ci∗) and
[TABLE]
Set G2=G−G1 and let P be a longest path in G2. We give an
orientation to each Ci∗ for 1≤i≤k−1.
We assume that
C1∗,C2∗,…,Ck−1∗ are chosen in G such that
(i) ∣G1∣ is as small as possible;
(ii) P is as long as possible, subject to (i);
(iii) β(G1) is as large as possible, subject to (i) and (ii).
First, we prepare some claims.
Claim 1**.**
Let u be a vertex of G2 and v be a vertex of Ci∗ with
1≤i≤k−1. If ∣Ci∗∣>∣Ci∣, then d(u,Ci∗)<∣Ci∣/2 and
d(v,Ci∗)<∣Ci∣/2. If d(u,Ci∗)=∣Ci∣/2 or
d(v,Ci∗)=∣Ci∣/2, then ∣Ci∗∣=∣Ci∣.
Proof.
If d(u,Ci∗)≥∣Ci∣/2, then it is not difficult to see that
Ci∗+u contains a cycle of length less than ∣Ci∗∣ but at
least ∣Ci∣, which can be seen as a subdivision of Ci. This
contradicts the choice of C1∗,C2∗,…,Ck−1∗ in (i).
If d(v,Ci∗)≥∣Ci∣/2, then we can get a similar
contradiction. So we have d(u,Ci∗)<∣Ci∣/2 and d(v,Ci∗)<∣Ci∣/2. The result ∣Ci∗∣=∣Ci∣ if d(u,Ci∗)=∣Ci∣/2 or
d(v,Ci∗)=∣Ci∣/2 follows immediately.
∎
By Claim 1, we can immediately obtain that for any vertices u∈V(G2),v∈V(Ci∗), we have d(u,Ci∗)≤∣Ci∣/2 and
d(v,Ci∗)≤∣Ci∣/2 for each i with 1≤i≤k−1.
Claim 2**.**
Let u′ be a vertex in V(G2)∩X and u′′ be a vertex in
V(G2)∩Y. If d(u′,G2)+d(u′′,G2)<∣Ck∣, then there exists
an Ci∗ with 1≤i≤k−1 such that
d(u′,Ci∗)+d(u′′,Ci∗)≥∣Ci∣−1. Moreover, we have
∣Ci∗∣=∣Ci∣, and there exists a vertex v∈N(u′,Ci∗) such
that Ci∗+u′′−v contains Ci.
Proof.
If d(u′,Cj∗)+d(u′′,Cj∗)≤∣Cj∣−2 for all j with 1≤j≤k−1, then
[TABLE]
contradicting the condition δ(G)≥n/2−k+1. Thus, there
exists an Ci∗ with 1≤i≤k−1 such that
d(u′,Ci∗)+d(u′′,Ci∗)≥∣Ci∣−1. This implies that
d(u′,Ci∗)=∣Ci∣/2 or d(u′′,Ci∗)=∣Ci∣/2. By Claim 1 we have
∣Ci∗∣=∣Ci∣. If d(u′′,Ci∗)=∣Ci∣/2, then any v∈N(u′,Ci∗) can be chosen as the required vertex. If
d(u′′,Ci∗)=∣Ci∣/2−1, then d(u′,Ci∗)=∣Ci∣/2. Since
∣Ci∣≥6, we can also find a vertex v∈N(u′,Ci∗) such
that Ci∗+u′′−v contains Ci.
∎
Now let P=u1u2⋯us and assume that u1∈X.
Claim 3**.**
P is a Hamilton path of G2.
Proof.
Suppose that P is not a Hamilton path of G2. Then from the
choice of P and the fact that G2 does not contain a subdivision
of Ck, we have
[TABLE]
and
[TABLE]
Let G2′=G2−V(P) and M={a1b1,a2b2,…,ambm} be a
maximum matching in G2′ with {a1,a2,…,am}⊆X. We distinguish two cases as follows.
Case 1. M is not a perfect matching of G2′.
First note that G2 has an even number of vertices. If us∈Y,
then P has also an even number of vertices. Thus, ∣G2′∣ is even
and ∣V(G2′)∩X∣=∣V(G2′)∩Y∣. If us∈X, then
∣V(G2′)∩Y∣=∣V(G2′)∩X∣+1. Since M is not a perfect
matching of G2′, we have (V(G2′)∩Y)∖{b1,b2,…,bm}=∅.
Let y0 be a vertex in (V(G2′)∩Y)∖{b1,b2,…,bm}. Choose P1 as a longest M-alternating
path in G2′ starting from y0. Then P1 must end at a vertex
y1 with y1∈{y,b1,…,bm}. Let M′=M−E(P1) and
G2′′=G2′−V(P1)+y. Choose P2 as a longest M′-alternating
path in G2′′ starting from y0. Then P2 must end at a vertex
y2∈{y,b1,…,bm}. Therefore, Q=P1∪P2 is a
path from y1 to y2 in G2′ such that d(y1,G2′)=d(y1,Q)
and d(y2,G2′)=d(y2,Q). Since G2 does not contain a
subdivision of Ck, we have
[TABLE]
and
[TABLE]
If d(y1,G2)≤∣Ck∣/2, then by d(u1,G2)+d(y1,G2)<∣Ck∣
and Claim 2 we know that d(u1,Ci∗)+d(y1,Ci∗)≥∣Ci∣−1 for
some i with 1≤i≤k−1, and there exists a vertex v∈N(u1,Ci∗) such that Ci∗+y1−v contains Ci∗. This
contradicts the choice of C1∗,C2∗,…,Ck−1∗ in (ii).
Hence, we have d(y1,G2)≥∣Ck∣/2+1. Similarly we have
d(y2,G2)≥∣Ck∣/2+1. This implies that
[TABLE]
and
[TABLE]
It is not difficult to find a segment of P with at least
[TABLE]
vertices such that the two end-vertices of this segment are adjacent
to y1 and y2, respectively. Since Q has at least
[TABLE]
vertices, we can obtain a cycle with at least ∣Ck∣ vertices in
G[V(P∪Q)], which can be seen as a subdivision of Ck. This
contradicts our assumption that G contains no subdivisions of H.
Case 2. M is a perfect matching of G2′.
Choose P3 as a longest M-alternating path in G2′ such that
the first edge of P3 is in M. Denote the initial vertex and the
final vertex of P3 by x and y, respectively. Then we have
x∈{a1,a2,…,am} and y∈{b1,b2,…,bm}.
Since G2 does not contain a subdivision of Ck, we know that
[TABLE]
and
[TABLE]
Clearly, us∈Y. If d(x,G2)≤∣Ck∣/2, then it follows from
d(us,G2)+d(x,G2)<∣Ck∣ and Claim 2 that
d(us,Ci∗)+d(x,Ci∗)≥∣Ci∣−1 for some i with 1≤i≤k−1, and there exists a vertex v∈N(us,Ci∗) such that
Ci∗+x−v contains Ci∗. This contradicts the choice of C1∗,C2∗,…,Ck−1∗ in (ii). Hence, we have d(x,G2)≥∣Ck∣/2+1. Similarly, we can show that d(y,G2)≥∣Ck∣/2+1.
This means that
[TABLE]
and
[TABLE]
The rest of the proof of this case is just as same as that of Case
1.
∎
Since G2 does not contain a subdivision of Ck and P is a
Hamilton path of G2, we can immediately get
[TABLE]
[TABLE]
Note us∈Y. Then, by (1) and Claim 2, we know that
there exists a component Cp∗ for some p with 1≤p≤k−1
such that
[TABLE]
and
[TABLE]
Without loss of generality, we may assume d(u1,Cp∗)=∣Cp∣/2
and d(us,Cp∗)≥∣Cp∣/2−1. Let x∗ be a vertex of
V(Cp∗)∩X such that us is adjacent to all vertices in
(V(Cp∗)∩X)∖{x∗}. Let y∗ be a vertex of
V(Cp∗)∩Y such that x∗ and y∗ are not adjacent on
Cp∗.
We have the following Claim 4 and Claim 5.
Claim 4**.**
For all z∈{u1,us−1,us,x∗,y∗} we have
[TABLE]
and for u2 we have
[TABLE]
Proof.
The case z=u1 or z=us follows from (1) and the fact
∣Cp∗∣=∣Cp∣ immediately.
Suppose z=x∗. Then we have u1x∗∈E(G). If x∗ is
adjacent to at least ∣Ck∣/2 vertices in P, then P+x∗−u1
contains a subdivision of Ck and Cp∗+u1−x∗ contains Cp.
This contradicts our assumption that G contains no subdivisions of
H. Thus, we have
[TABLE]
Similarly, we can prove
[TABLE]
Suppose z=us−1. If us−1 is adjacent to a vertex v in
(V(Cp∗)∩Y)∖{x∗+,x∗−}, then from us−1∈X and ∣Cp∗∣≥6 we know that G2+v−us contains a
subdivision of Ck and Cp∗+us−v contains Cp. This
contradicts our assumption that G contains no subdivisions of H.
Together with (2), we can obtain
[TABLE]
Note that u2∈Y. If u2 is adjacent to a vertex v in
(V(Cp∗)∩X)∖{x∗}, then G2+v−u1 contains a
subdivision of Ck and Cp∗+u1−v contains Cp, a
contradiction. Thus, together with (2) and ∣Cp∣≥6, we have
[TABLE]
∎
Claim 5**.**
Let S={u1,u2,us−1,us,x∗,y∗}. Then there exists a Cq∗
with q=p and 1≤q≤k−1 such that
[TABLE]
and ∣Cq∗∣=∣Cq∣.
Proof.
By Claim 4, we can get that
[TABLE]
If ∑z∈Sd(z,Ci∗)≤3∣Ci∣−6 for all i with 1≤i≤k−1 and i=p, then we have
[TABLE]
This contradicts the condition δ(G)≥n/2−k+1. Hence, there
exists a Cq∗ for some q with 1≤q≤k−1 and q=p
such that
[TABLE]
Then, at least one of the following holds.
[TABLE]
[TABLE]
and
[TABLE]
If d(u1,Cq∗)+d(us,Cq∗)≥∣Cq∣−1 or d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1, then by Claim 2 we have
∣Cq∗∣=∣Cq∣. If d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−1, then
d(x∗,Cq∗)=∣Cq∣/2 or d(y∗,Cq∗)=∣Cq∣/2. Since
Cp∗+u1−x∗ or Cp∗+us−y∗ contains Cp, we can prove
∣Cq∗∣=∣Cq∣ as in Claim 1.
∎
In order to complete the proof of Theorem 3, we will show that
G[V(G2)∪V(Cp∗)∪V(Cq∗) contains a subdivision of
Ck∪Cp∪Cq, which contradicts our assumption that G
contains no a subdivision of H. Note that d(u1,Cq∗)+d(us,Cq∗)≤∣Cq∣, d(u2,Cq∗)+d(us−1,Cq∗)≤∣Cq∣ and d(x∗,Cq∗)+d(y∗,Cq∗)≤∣Cq∣.
We consider the two cases as follows.
Case 1. d(u1,Cq∗)+d(us,Cq∗)≥∣Cq∣−1.
Suppose that d(u1,Cq∗)+d(us,Cq∗)=∣Cq∣. Then by Claim 5 we
have d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−5≥1. This
implies that either u1 and us−1 have at least one common
neighbor in Cq∗ or u2 and us have at least one common
neighbor in Cq∗. By the symmetry, we assume that v∈V(Cq∗)
with u1v,us−1v∈E(G). It is easy to see that G2+v−us
contains a subdivision of Ck and Cq∗+us−v contains a
subdivision of Cq.
Suppose that d(u1,Cq∗)+d(us,Cq∗)=∣Cq∣−1. By the symmetry,
we assume that d(u1,Cq∗)=∣Cq∣/2 and d(us,Cq∗)=∣Cq∣/2−1. Let v be the vertex in V(Cq∗)∩X with
usv∈E(G). If ∣Cq∣≥8, then by Claim 5 we have
d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−4≥4. So, we can
deduce that either u2 has at least one neighbor in
(V(Cq∗)∖{v})∩X or us−1 has at least one
neighbor in (V(Cq∗)∖{v+,v−})∩Y. If u2 has at
least one neighbor in (V(Cq∗)∖{v})∩X, say vx,
then G2+vx−u1 contains a subdivision of Ck and
Cq∗+u1−vx contains a subdivision of Cq. If us−1 has at
least one neighbor in (V(Cq∗)∖{v+,v−})∩Y, say
vy, then G2+vy−us contains a subdivision of Ck and
Cq∗+us−vy contains a subdivision of Cq. Thus, we have
∣Cq∣=6. If u2 or us−1 has a neighbor in
V(Cq∗)∖{v−,v,v+}, then the proof is as same as that
of the case ∣Cq∣≥8. We may assume that N(u2,Cq∗)∪N(us−1,Cq∗)⊆{v−,v,v+}. Since d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−4=2, we have us−1v−∈E(G) or us−1v+∈E(G). Without loss of generality, we assume
that us−1v+∈E(G). If u2v∈E(G), then d(u2,Cq∗)=0 and d(us−1,Cq∗)=∣Cq∣−4. By Claim 5 we have
d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣. It follows from the choice of
y∗ in Cp∗ that G2+v+−us contains a subdivision of Ck,
Cp∗+us−y∗ contains a subdivision of Cp and Cq∗+y∗−v+
contains a subdivision of Cq. If u2v∈E(G), then we can
deduce that v is adjacent to every vertex in V(Cq∗)∩Y. If
not, then P+v−u1 is a path of order ∣P∣ and Cq∗+u1−v
contains a subdivision of Cq such that ∣Cq∗+u1−v∣=∣Cq∗∣
and ∣E(G[V(Cq∗+u1−v)])∣>∣E(G[V(Cq∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). Then it is not
difficult to see that G2+v+−us contains a subdivision of Ck
and Cq∗+us−v+ contains a subdivision of Cq.
Case 2. d(u1,Cq∗)+d(us,Cq∗)≤∣Cq∣−2.
We will consider three subcases as follows.
Case 2.1. d(u1,Cq∗)+d(us,Cq∗)≤∣Cq∣−4.
By Claim 5 we have
[TABLE]
This implies that either d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣ and
d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−1 or d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣. If d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣ and
d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−1, then there exists an edge
vxvy∈E(Cq∗) such that x∗ is adjacent to every vertex in
(V(Cq∗)∖{vy})∩Y and y∗ is adjacent to every
vertex in (V(Cq∗)∖{vx})∩X. Thus we can get that
G2+vx+vy−u1−us contains a subdivision of Ck,
Cp∗+u1+us−x∗−y∗ contains a subdivision of Cp and
Cq∗+x∗+y∗−vx−vy contains a subdivision of Cq. The case
d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣ can be proved similarly.
Case 2.2. d(u1,Cq∗)+d(us,Cq∗)=∣Cq∣−3.
By Claim 5, we have d(x∗,Cq∗)+d(y∗,Cq∗)+d(u2,Cq∗)+d(us−1,Cq∗)≥2∣Cq∣−2. This implies that d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣, d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣, or d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣−1 and d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−1. We consider three subcases as
follows.
Case 2.2.1. d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣.
Then by Claim 5 we have d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−2>∣Cq∣/2. It is not difficult to see that there exists an
edge vxvy∈E(Cq∗) with u2vx,us−1vy∈E(G) such that
G2+vx+vy−u1−us contains a subdivision of Ck,
Cp∗+u1+us−x∗−y∗ contains a subdivision of Cp and
Cq∗+x∗+y∗−vx−vy contains a subdivision of Cq.
Case 2.2.2. d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣.
Then by Claim 5 we have d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−2.
First, we assume that ∣Cq∣≥8. Then we can see that d(u1,Cq∗)≥1. If d(y∗,Cq∗)≥∣Cq∣/2−1, then we can deduce
that there exists an edge vxvy∈E(Cq∗) with
u2vx,us−1vy∈E(G) such that G2+vx+vy−u1−us contains
a subdivision of Ck, Cp∗+us−y∗ contains a subdivision of
Cp and Cq∗+u1+y∗−vx−vy contains a subdivision of Cq.
If d(y∗,Cq∗)=∣Cq∣/2−2, then d(x∗,Cq∗)=∣Cq∣/2 and we
can also deduce that there exists an edge vxvy∈E(Cq∗) with
u2vx,us−1vy∈E(G) such that G2+vx+vy−u1−us contains
a subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a
subdivision of Cp and Cq∗+x∗+y∗−vx−vy contains a
subdivision of Cq.
Next, we assume that ∣Cq∣=6. If d(u1,Cq∗)=3, then by
d(y∗,Cq∗)≥1 and u1y∗∈E(G) we can deduce that there
exists an edge vxvy∈E(Cq∗) such that u2vx,us−1vy∈E(G) and G2+vx+vy−u1−u2 contains a subdivision of Ck,
Cp∗+us−y∗ contains a subdivision of Cp and
Cq∗+u1+y∗−vx−vy contains a subdivision of Cq. So we
assume that d(u1,Cq∗)≤2. Then d(us,Cq∗)≥1. Let
vx′ be a vertex in Cq∗ with usvx′∈E(G). We claim that
if d(x∗,Cq∗)=3, then x∗y∗∈E(G). If x∗y∗∈E(G),
then we can get that P+vx′−u1 is a path of order ∣P∣,
Cq∗+x∗−vx′ contains a subdivision of Cq with
∣Cq∗+x∗−vx′∣=∣Cq∣ and ∣E(G[V(Cq∗+x∗−vx′)])∣≥∣E(G[V(Cq∗)])∣, and Cp∗+u1−x∗ contains a subdivision of
Cp with ∣Cp∗+u1−x∗∣=∣Cp∣ and
∣E(G[V(Cp∗+u1−x∗)])∣>∣E(G[V(Cp∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). Thus, if
d(x∗,Cq∗)=3, then by d(y∗,Cq∗)≥1, it is not difficult
to see that there exists an edge vxvy∈E(Cq∗) such that
u2vx,us−1vy∈E(G) and G2+vx+vy−u1−u2 contains a
subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a subdivision
of Cp and Cq∗+x∗+y∗−vx−vy contains a subdivision of
Cq. If d(y∗,Cq∗)=3 and d(u1,Cq∗)≥1, then we can
also obtain that there exists an edge vxvy∈E(Cq∗) such that
u2vx,us−1vy∈E(G) and G2+vx+vy−u1−u2 contains a
subdivision of Ck, Cp∗+us−y∗ contains a subdivision of
Cp and Cq∗+u1+y∗−vx−vy contains a subdivision of Cq.
If d(y∗,Cq∗)=3 and d(us,Cq∗)=3, then we claim that
x∗y∗∈E(G). If x∗y∗∈E(G), then we can get that
P+vy′+x∗−u1−us is a path of order ∣P∣, where vy′∈N(x∗,Cq∗), Cq∗+us−vy′ contains a subdivision of Cq
with ∣Cq∗+us−vy′∣=∣Cq∣ and ∣E(G[V(Cq∗+us−vy′)])∣≥∣E(G[V(Cq∗)])∣, and Cp∗+u1−x∗ contains a subdivision of
Cp with ∣Cp∗+u1−x∗∣=∣Cp∣ and
∣E(G[V(Cp∗+u1−x∗)])∣>∣E(G[V(Cp∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). Note that
d(x∗,Cq∗)≥1. So, there exists an edge vxvy∈E(Cq∗)
such that u2vx,us−1vy∈E(G) and G2+vx+vy−u1−u2
contains a subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a
subdivision of Cp and Cq∗+x∗+y∗−vx−vy contains a
subdivision of Cq. We may assume that
d(x∗,Cq∗)=d(y∗,Cq∗)=2. Let Cq∗=v1v2v3v4v5v6v1
with v1∈X. Without loss of generality, we suppose that
v2x∗,v4x∗∈E(G). If v1y∗,v3y∗∈E(G) or
v3y∗,v5y∗∈E(G), by the symmetry, say v1y∗,v3y∗∈E(G), then G2+v5+v6−u1−u2 contains a subdivision of Ck,
Cp∗+u1+us−x∗−y∗ contains a subdivision of Cp and
Cq∗+x∗+y∗−v5−v6 contains a subdivision of Cq. Thus let
v1y∗,v5y∗∈E(G). If u1v2 or u1v4∈E(G), by the
symmetry, say u1u2∈E(G), then G2+v3+v4−u1−u2 contains
a subdivision of Ck, Cp∗+us−y∗ contains a subdivision of
Cp and Cq∗+u1+y∗−v3−v4 contains a subdivision of Cq.
This implies that d(u1,Cq∗)≤1 and d(us,Cq∗)≥2. If
u1v6,usv1,usv5∈E(G), then we can see that G2+v6−us
contains a subdivision of Ck and Cq∗+us−v6 contains a
subdivision of Cq. Thus we assume that usv1,usv3∈E(G) or
usv3,usv5∈E(G). By the symmetry, we suppose that
usv1,usv3∈E(G). Then it is not difficult to see that
G2+v5+v6−u1−u2 contains a subdivision of Ck,
Cp∗+u1−x∗ contains a subdivision of Cp and
Cq∗+us+x∗−v5−v6 contains a subdivision of Cq.
Case 2.2.3. d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣−1 and
d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−1.
If ∣Cq∣≥8, or ∣Cq∣=6 and N(x∗,Cq∗)∪N(y∗,Cq∗)=N(u2,Cq∗)∪N(us−1,Cq∗), then it is easy to
deduce that there exists an edge vxvy∈E(Cq∗) with
u2vx,us−1vy∈E(G) such that G2+vx+vy−u1−us contains
a subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a
subdivision of Cp and Cq∗+x∗+y∗−vx−vy contains a
subdivision of Cq. Now suppose that ∣Cq∣=6 and N(x∗,Cq∗)∪N(y∗,Cq∗)=N(u2,Cq∗)∪N(us−1,Cq∗). Let
Cq∗=v1v2v3v4v5v6v1 with v1∈X. Without loss of
generality, we assume that u2v1,v1y∗∈E(G). Then we have
u2v3,u2v5,v3y∗,v5y∗∈E(G). If d(u1,Cq∗)=3, then we
can obtain that G2+v3+v4−u1−us contains a subdivision of
Ck, Cp∗+us−y∗ contains a subdivision of Cp and
Cq∗+u1+y∗−v3−v4 contains a subdivision of Cq. If
d(u1,Cq∗)≤2, then d(us,Cq∗)≥1, say vx∈V(Cq∗) with usvx∈E(G). We claim that x∗y∗∈E(G). If
not, then by d(x∗,Cq∗)=3 we can deduce that P+vx−u1 is a
path of order ∣P∣, Cq∗+x∗−vx contains a subdivision of Cq
with ∣Cq∗+x∗−vx∣=∣Cq∣ and ∣E(G[V(Cq∗+x∗−vx)])∣≥∣E(G[V(Cq∗)])∣, and Cp∗+u1−x∗ contains a subdivision of
Cp with ∣Cp∗+u1−x∗∣=∣Cp∣ and
∣E(G[V(Cp∗+u1−x∗)])∣>∣E(G[V(Cp∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). Thus we can
see that G2+v3+v4−u1−us contains a subdivision of Ck,
Cp∗+u1+us−x∗−y∗ contains a subdivision of Cp and
Cq∗+x∗+y∗−v3−v4 contains a subdivision of Cq.
Case 2.3. d(u1,Cq∗)+d(us,Cq∗)=∣Cq∣−2.
By the symmetry, we let d(u1,Cq∗)≥∣Cq∗∣/2−1. Note that
d(x∗,Cq∗)+d(y∗,Cq∗)≤∣Cq∣. It follows from Claim 5
that ∣Cq∣−3≤d(u2,Cq∗)+d(us−1,Cq∗)≤∣Cq∣.
First, we assume that ∣Cq∣≥8. If d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and d(u1,Cq∗)=∣Cq∗∣/2, then d(us,Cq∗)=∣Cq∗∣/2−2≥2 and there exists at least one vertex vx
in V(Cq∗)∩X such that u2vx,usvx∈E(G). So we can
obtain that G2+vx−u1 contains a subdivision of Ck and
Cq∗+u1−vx contains a subdivision of Cq. If d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and d(u1,Cq∗)=∣Cq∗∣/2−1≥3, then d(us,Cq∗)=∣Cq∗∣/2−1≥3. By
the symmetry we let d(u2,Cq∗)=∣Cq∣/2. Then there exists at
least one vertex vx in V(Cq∗)∩X such that u2vx,usvx,u1vx−,u1vx+∈E(G). It is easy to see that G2+vx−u1
contains a subdivision of Ck and Cq∗+u1−vx contains a
subdivision of Cq. If d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−3,
then by ∣Cq∣−3>∣Cq∗∣/2, there exists at least an edge
vxvy∈E(Cq∗) such that u2vx,us−1vy∈E(G). By Claim
5, we can know that d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣. It follows
from the choices of x∗,y∗ in Cp∗ that G2+vx+vy−u1−u2
contains a subdivision of Ck, Cp∗+u1+u2−x∗−y∗ contains a
subdivision of Cp and Cq∗+x∗+y∗−vx−vy contains a
subdivision of Cq. If d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−2,
then by Claim 5 we have d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−1.
Since ∣Cq∣−2−∣Cq∗∣/2≥2, we can get that there exists at
least an edge vxvx+∈E(Cq∗) such that u2vx,us−1vx+,vx−x∗,vx++y∗∈E(G). Further, we have d(x∗,Cq∗−vx−vx+)+d(y∗,Cq∗−vx−vx+)≥∣Cq∗−vx−vx+∣−1.
Thus we can obtain that G2+vx+vx+−u1−u2 contains a
subdivision of Ck, Cp∗+u1+u2−x∗−y∗ contains a subdivision
of Cp and Cq∗+x∗+y∗−vx−vx+ contains a subdivision of
Cq.
Next, we assume that ∣Cq∣=6. Let Cq∗=v1v2v3v4v5v6v1
with v1∈X.
Suppose that d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and
d(u1,Cq∗)=∣Cq∗∣/2=3. If there exists a vertex vx in
V(Cq∗)∩X such that u2vx,usvx∈E(G), then we can
obtain that G2+vx−u1 contains a subdivision of Ck and
Cq∗+u1−vx contains a subdivision of Cq. Note that d(us,Cq∗)=1. So we may assume that d(u2,Cq∗)=∣Cq∣/2−1 and
N(u2,Cq∗)∩N(us,Cq∗)=∅. Without loss of
generality, we let usv1,u2v3,u2v5∈E(G). We claim that
v1v4∈E(G). If not, then P+v1−u1 is a path of order ∣P∣
and Cq∗+u1−v1 contains a subdivision of Cq with
∣Cq∗+u1−v1∣=∣Cq∣ and
∣E(G[V(Cq∗+u1−v1)])∣>∣E(G[V(Cq∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). By Claim 5 we
have d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−2=4. If y∗ has at
least two neighbors in Cq∗, then by d(u1,Cq∗)=∣Cq∗∣/2,d(us−1,Cq∗)=∣Cq∗∣/2 and the choice of y∗ in Cp∗,
there exists a vertex vy in Cq∗ such that G2+vy−us
contains a subdivision of Ck, Cp∗+us−y∗ contains a
subdivision of Cp and Cq∗+y∗−vy contains a subdivision of
Cq. So we have d(x∗,Cq∗)=3 and d(y∗,Cq∗)=1. We claim
that x∗y∗∈E(G). If not, then P+v1−u1 is a path of order
∣P∣, Cq∗+x∗−v1 contains a subdivision of Cq with
∣Cq∗+x∗−v1∣=∣Cq∣ and
∣E(G[V(Cq∗+x∗−v1)])∣=∣E(G[V(Cq∗)])∣, and Cp∗+u1−x∗
contains a subdivision of Cp with ∣Cp∗+u1−x∗∣=∣Cp∣ and
∣E(G[V(Cp∗+u1−x∗)])∣>∣E(G[V(Cp∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). If v1y∗∈E(G), then we can get that G2+v2+v3−u1−us contains a
subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a subdivision
of Cp and Cq∗+x∗+y∗−v2−v3 contains a subdivision of
Cq. The cases v3y∗∈E(G) and v5y∗∈E(G) can be proved
similarly.
Suppose that d(u2,Cq∗)+d(us−1,Cq∗)≥∣Cq∣−1 and
d(u1,Cq∗)=d(us,Cq∗)=∣Cq∗∣/2−1=2. By the symmetry, we may
assume that d(u2,Cq∗)=∣Cq∣/2=3 and d(us−1,Cq∗)≥∣Cq∗∣/2−1=2. Without loss of generality, we let
u1v2,u1v6,us−1v2∈E(G). If usv1∈E(G), then it is
easy to see that G2+v1−u1 contains a subdivision of Ck and
Cq∗+u1−v1 contains a subdivision of Cq. Thus, we consider
usv1∈E(G) and by d(us,Cq∗)=∣Cq∗∣/2−1=2, we have
usv3,usv5∈E(G). We claim that v1v4∈E(G). If not,
then P+v3+v4−u1−u2 is a path of order ∣P∣ and
Cq∗+u1+u2−v3−v4 contains a subdivision of Cq with
∣Cq∗+u1+u2−v3−v4∣=∣Cq∣ and
∣E(G[V(Cq∗+u1+u2−v3−v4)])∣>∣E(G[V(Cq∗)])∣, which
contradicts the choice of C1∗,C2∗,…,Ck−1∗ in
(iii). It follows from Claim 5 that d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−3=3. If v1y∗ or v5y∗∈E(G), then
G2+v2+v3−u1−us contains a subdivision of Ck,
Cp∗+us−y∗ contains a subdivision of Cp and
Cq∗+y∗+u1−v2−v3 contains a subdivision of Cq. If
v3y∗∈E(G), then G2+v1+v2−u1−us contains a subdivision
of Ck, Cp∗+us−y∗ contains a subdivision of Cp and
Cq∗+y∗+u1−v1−v2 contains a subdivision of Cq. So we let
d(y∗,Cq∗)=0 and d(x∗,Cq∗)=3. Then we can see that
G2+v1+v2−u1−us contains a subdivision of Ck,
Cp∗+u1−x∗ contains a subdivision of Cp and
Cq∗+x∗+us−v1−v2 contains a subdivision of Cq.
Suppose that ∣Cq∣−3≤d(u2,Cq∗)+d(us−1,Cq∗)≤∣Cq∣−2. From Claim 5, we can obtain that d(x∗,Cq∗)+d(y∗,Cq∗)≥∣Cq∣−1. First, we assume that d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣. If there exists an edge vxvy in
Cq∗ with u2vx,us−1vy∈E(G), then G2+vx+vy−u1−us
contains a subdivision of Ck, Cp∗+u1+us−x∗−y∗ contains a
subdivision of Cp and Cq∗+x∗+y∗−vx−vy contains a
subdivision of Cq. So, we consider d(u2,Cq∗)=3 or
d(us−1,Cq∗)=3. By the symmetry, we let d(u2,Cq∗)=3.
Note that d(us,Cq∗)>0. This implies that there exists a vertex
vx∈V(Cq∗) such that u2vx,usvx∈E(G). Then
G2+vx−u1 contains a subdivision of Ck, Cp∗+u1−x∗
contains a subdivision of Cp and Cq∗+x∗−vx contains a
subdivision of Cq. Next, we assume that d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣−1=5. It follows from Claim 5 that d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−2=4. Since d(x∗,Cq∗)+d(y∗,Cq∗)=∣Cq∣−1=5, we have d(x∗,Cq∗)=3 or d(y∗,Cq∗)=3.
If d(x∗,Cq∗)=3, then d(y∗,Cq∗)=2. We claim that
x∗y∗∈E(G). If not, then by d(us,Cq∗)≥1, say
usv1∈E(G), we can get that P+v1−u1 is a path of order
∣P∣, Cq∗+x∗−v1 contains a subdivision of Cq with
∣Cq∗+x∗−v1∣=∣Cq∣ and ∣E(G[V(Cq∗+x∗−v1)])∣≥∣E(G[V(Cq∗)])∣, and Cp∗+u1−x∗ contains a subdivision of
Cp with ∣Cp∗+u1−x∗∣=∣Cp∣ and
∣E(G[V(Cp∗+u1−x∗)])∣>∣E(G[V(Cp∗)])∣, which contradicts the
choice of C1∗,C2∗,…,Ck−1∗ in (iii). From d(u2,Cq∗)+d(us−1,Cq∗)=∣Cq∣−2=4, we can deduce that there exists
an edge vxvy∈E(Cq∗) such that u2vx,us−1vy∈E(G)
and G2+vx+vy−u1−us contains a subdivision of Ck,
Cp∗+u1+us−x∗−y∗ contains a subdivision of Cp and
Cq∗+x∗+y∗−vx−vy contains a subdivision of Cq. Now we let
d(y∗,Cq∗)=3. If d(u1,Cq∗)≥2 or
N(us−1,Cq∗)∖N(u1,Cq∗)=∅, then we can
obtain that there exists an edge vxvy∈E(Cq∗) such that
u2vx,us−1vy∈E(G) and G2+vx+vy−u1−us contains a
subdivision of Ck, Cp∗+us−y∗ contains a subdivision of
Cp and Cq∗+u1+y∗−vx−vy contains a subdivision of Cq.
If d(u1,Cq∗)=d(us−1,Cq∗)=1 and there exists the vertex
vy∈V(Cq∗) with u1vy,us−1vy∈E(G), then by
d(us,Cq∗)=3 we can get that G2+vy−us contains a
subdivision of Ck and Cq∗+us−vy contains a subdivision of
Cq.
The proof of Theorem 3 is complete. □