The Graph Ramsey Number R(Fℓ,K6)
Shin-ya Kadota
,
Tomokazu Onozuka
and
Yuta Suzuki
Abstract.
For a given pair of two graphs (F,H),
let R(F,H) be the smallest positive integer r
such that for any graph G of order r,
either G contains F as a subgraph
or
the complement of G contains H as a subgraph.
Baskoro, Broersma and Surahmat (2005) conjectured that
[TABLE]
for ℓ≥n≥3, where Fℓ is the join of K1 and ℓK2.
In this paper, we prove that this conjecture is true for the case n=6.
Key words and phrases:
Ramsey number; Fan graph; Complete graph
2010 Mathematics Subject Classification:
Primary 05C55, Secondary 05D10
1. Introduction
Throughout this paper, all graphs are finite and simple.
For a given pair of two graphs (F,H),
let R(F,H) be the smallest positive integer r
such that for any graph G of order r,
G contains F as a subgraph
or
the complement of G contains H as a subgraph.
In general, it is quite difficult to calculate the exact values of R(F,H).
However, for sparse graphs F and H,
there are many results on the exact values of R(F,H).
In this paper, we consider fan graph Fℓ,
which is the join of K1 and ℓK2.
This fan graph is one example of such sparse graphs.
Recently,
although Kn itself is the densest graph,
some authors succeeded in determining the value R(Fℓ,Kn)
for large ℓ and small n.
As for this problem,
Baskoro, Broersma and Surahmat [1] conjectured:
Conjecture** (The Baskoro-Broersma-Surahmat conjecture [1]).**
**
For any ℓ≥n≥3, we have R(Fℓ,Kn)=2ℓ(n−1)+1.
In the last 20 years,
this conjecture was proved to be true for the case n=3,4,5:
Theorem A** **(Gupta, Gupta and Sudan [5],
Li and Rousseau [6, Proposition 1]).
**
For any ℓ≥3, we have R(Fℓ,K3)=4ℓ+1.
Theorem B** (Baskoro, Broersma and Surahmat [1]).**
**
For any ℓ≥4, we have R(Fℓ,K4)=6ℓ+1.
Theorem C** (Chen and Zhang [2]).**
**
For any ℓ≥5, we have R(Fℓ,K5)=8ℓ+1.
In this paper,
we prove the Baskoro-Broersma-Surahmat conjecture for n=6.
Theorem 1**.**
**
For any ℓ≥6, we have R(Fℓ,K6)=10ℓ+1.
Our main strategy is
based on the approach of Chen and Zhang [2].
In particular, we shall heavily use their structural setting.
However, we need some additional insight on the Chen-Zhang structure
than their paper [2].
We briefly summarize our notation and terminology.
Let G=(V,E) be a graph.
For any subset S⊂V, we use G[S] to denote the subgraph induced by S.
For any vertex v∈V,
we denote the neighborhood of v by N(v),
i.e.
[TABLE]
and we let N[v]=N(v)∪{v}.
We let
[TABLE]
Moreover, we denote by ω(G) the clique number of G,
i.e. the order of the largest clique in G,
and denote by α(G) the independence number of G,
i.e. the order of the largest independent set in G.
We refer to the book [4]
for other graph theoretical notation and terminology
not described in this paper.
2. The lower bound
The lower bound of R(Fℓ,K6) is given by the following theorem,
which is just a special case of the Chvátal-Harary lemma [3, Lemma 4].
Theorem 2**.**
For any ℓ≥6,
we have
R(Fℓ,K6)≥10ℓ+1.
Proof.
It is sufficient to give a graph G of order 10ℓ
such that G does not contain Fℓ
and G does not contain K6.
For example, the graph
G=5K2ℓ
satisfies this condition.
Hence the theorem follows.
∎
Hence we prove the upper bound R(Fℓ,K6)≤10ℓ+1 in the following sections.
Theorem 3**.**
For any ℓ≥6,
we have
R(Fℓ,K6)≤10ℓ+1.
We prove this upper bound by contradiction.
In the remaining part of this paper,
we assume ℓ≥6 and that there exists a graph G of order 10ℓ+1
such that Fℓ⊂G and K6⊂G.
By Theorem C,
we may assume that α(G)=5.
3. Preliminary lemmas
In this section,
we prove several lemmas on basic properties of the graph G.
We start with the following simple observation,
which is related to the number of independent edges in G.
Lemma 1**.**
There is no vertex v∈G for which G[N(v)] contains ℓK2.
Proof.
If ℓK2⊂G[N(v)],
then Fℓ⊂G[N[v]]. This is a contradiction.
∎
The next lemma is a special case of Stahl’s lemma [7].
Lemma 2**.**
For any ℓ≥1, we have R(ℓK2,K6)=2ℓ+4.
Proof.
See [7, pp. 586–587].
∎
As a consequence of Lemma 2 and α(G)=5,
any subgraph H⊂G of order n≥6
contains ⌊n/2⌋−2 or more independent edges.
The following is an immediate consequence
of Theorem C and Lemma 2.
Lemma 3**.**
We have 2ℓ≤δ(G)≤Δ(G)≤2ℓ+3.
Proof.
First we show that δ(G)≥2ℓ.
Assume, to the contrary, that there exists a vertex v with d(v)≤2ℓ−1.
Let S=V(G)\N[v].
Then ∣S∣≥8ℓ+1.
Also, since α(G)=5, it follows that α(G[S])≤4.
Then G[S] contains Fℓ by Theorem C, a contradiction.
To show that Δ(G)≤2ℓ+3, suppose that d(v)≥2ℓ+4 for some vertex v.
Since α(G)=5, Lemma 2 guarantees the existence of ℓK2 in G[N(v)], producing Fℓ in G.
This cannot occur and so Δ(G)≤2ℓ+3, as claimed.
∎
The next lemma estimates the clique number ω(G).
Lemma 4**.**
We have ω(G)≤2ℓ−2.
Proof.
Assume, to the contrary, that G contains a clique H of order 2ℓ−1.
Select a vertex v0∈V(G)\V(H) such that
[TABLE]
The graph G−H−v0 is of order 8ℓ+1
so that α(G−H−v0)=5 by Theorem C.
Let U be a 5-set of independent vertices in the graph G0=G−H−v0.
Since U∪{v} cannot be independent for each v∈V(G)\U, there are at least 2ℓ−1 edges between H and U.
Hence, there exists a vertex u0∈U with ∣N(u0)∩V(H)∣≥(2ℓ−1)/5>2.
Consequently, ∣N(v0)∩V(H)∣≥∣N(u0)∩V(H)∣≥3.
Next we show that, for each vertex v∈V(G0), if w∈N(v0)∩N(v)∩V(H), then N(v)∩V(H)={w}.
If this is not the case, say there exists a vertex w′∈N(v)∩V(H) with w=w′, then (ℓ−1)K2 in G[(V(H)\{w,w′})∪{v0}] and the edge vw′ form ℓK2 in G[N(w)], which is impossible.
Now, let S1=N(v0)∩V(H)={w1,w2,…,wt}, where t=∣N(v0)∩V(H)∣.
Then by the above observations,
we can find a t-set U1={u1,u2,…,ut}⊂U
such that N(ui)∩V(H)={wi} for 1≤i≤t.
Also, let S2=V(H)∖S1 and U2=U\U1.
Recall that ∣N(u0)∩V(H)∣≥3, so u0∈U2.
Hence, t=3,4.
Note that there is no edge between U1 and S2
and also note that U∪{v} cannot be independent for each v∈S2.
Thus there are at least ∣S2∣=2ℓ−1−t edges between U2 and S2.
However then,
[TABLE]
which cannot occur.
This completes the proof.
∎
4. Structural observation
In this section,
we give some observation on the structure of the graph G
following the argument of Chen and Zhang [2].
Let
[TABLE]
be a 5-set of independent vertices and let
[TABLE]
for 1≤i≤5.
Since α(G)≤5,
we have a partition
[TABLE]
4.1. On the sets Xi
Obviously, we have
[TABLE]
and, since d(ui)≤Δ(G)≤2ℓ+3 by Lemma 3, we have
[TABLE]
We will use (1) and (2)
throughout this paper.
For example,
by (1) and (2)
we have
[TABLE]
which implies ∣X1∣≥10ℓ−23.
Also, we have
[TABLE]
Let I={1,2,3,4,5}.
For each pair i,j∈I,
we define the set
Xij=Xi∩N(uj).
Then we have a partition
[TABLE]
We next find that each X1i induces a clique in G.
Lemma 5**.**
For each i∈I, the graph G[X1i] is complete.
Consequently, we have
[TABLE]
Proof.
Let v,v′∈X1i.
Since the 6-set (U\{ui})∪{v,v′} cannot be independent,
it follows that v and v′ are adjacent.
Therefore, G[X1i∪{ui}] is a complete graph of order
∣X1i∣+1≤ω(G)≤2ℓ−2.
In addition, ∣X1∣=∑i=15∣X1i∣≤10ℓ−15.
∎
Assume, without loss of generality, that
[TABLE]
in the rest of our discussion.
Thus (3) implies that
[TABLE]
In particular, each X1i is nonempty.
The next lemma on the sets X1i is immediate yet useful.
We will use this lemma mainly with the choice t=2.
Lemma 6**.**
Let i,j,k∈I be distinct indices and v∈X1i.
Suppose that ∣X1i∣≥2ℓ−2t for some integer with 2≤t≤6.
Then we have
∣N(v)∩(X1j∪X1k)∣≤2t and
∣N(v)∩X1j∣≤2t−1.
In addition, if ∣N(v)∩X1j∣≥2 also holds,
then we have ∣N(v)∩X1k∣≤2t−3.
Proof.
Note that (ℓ−t)K2⊂G[(X1i\{v})∪{ui}].
Thus, we must avoid tK2 in G[N(v)∩(X1j∪X1k)]
and so the result is immediate as each of X1j and X1k induces a clique.
∎
4.2. On the sets N(ui)\X1i
Besides the same kind of information as Chen and Zhang obtained,
we need some new information on their structural setting.
We start with some observations
on the set N(ui)\X1i.
By Lemma 3 and Lemma 5
[TABLE]
In general,
suppose that A and B are disjoint nonempty sets of vertices in a graph.
Then it is well-known that,
if ∣N(v)∩B∣≥∣A∣ for every v∈A,
then there are at least ∣A∣ independent edges between A and B.
The following is then immediate.
Lemma 7**.**
Let i∈I and t=2ℓ−∣X1i∣.
Every t-set S⊂N(ui)\X1i contains a vertex v for which ∣N(v)∩X1i∣<t.
Proof.
If S={v1,v2,…,vt}⊂N(ui)\X1i
and ∣N(vj)∩X1i∣≥t for 1≤j≤t, then X1i contains t distinct vertices w1,w2,…,wt such that vjwj∈E(G) for 1≤j≤t.
However then, (ℓ−t)K2 in G[X1i\{w1,w2,…,wt}] with the t edges vjwj (1≤j≤t) forms ℓK2 in G[N(ui)], a contradiction.
∎
Let i∈I.
Under certain conditions, there exists a 4-set Qi⊂V(G) satisfying
[TABLE]
For example, such Qi exists if the degree of ui equals 2ℓ+3,
which we verify next.
Lemma 8**.**
*Let i∈I.
If d(ui)=2ℓ+3, then there exists a 4-set Qi
satisfying *(6).
Proof.
Note that ∣N(ui)\X1i∣=d(ui)−∣X1i∣≥6
since d(ui)=2ℓ+3.
Suppose first that ∣X1i∣ is odd, say ∣X1i∣=2t−1 for some positive integer t.
Then G[N(ui)\X1i] contains (ℓ−t)K2
by Lemma 2 with four remaining vertices v1,v2,v3,v4.
Let Qi={v1,v2,v3,v4}.
Then for any w∈X1i,
G[X1i\{w}] contains (t−1)K2 as G[X1i] is complete.
Thus Qi∪{w} must be independent
in order to avoid ℓK2 in G[N(ui)].
Suppose next that ∣X1i∣=2t for some positive integer t.
Then G[N(ui)\X1i] contains (ℓ−t−1)K2
again by Lemma 2 with five remaining vertices v1,v2,…,v5.
Since G[X1i] obviously contains tK2,
the set Q={v1,v2,…,v5} must be independent
to avoid ℓK2 in G[N(ui)].
For any w∈X1i,
∣N(w)∩Q∣≥1 since Q∪{w} cannot be independent.
Also, there cannot be two or more independent edges
between X1i and Q to avoid ℓK2 in G[N(ui)].
Thus we can find a vertex in Q, say v5,
such that every w∈X1i is adjacent to v5.
Let Qi={v1,v2,v3,v4}.
Recall that ∣X1i∣=2t≥2.
Then Qi∪{w} must be independent for every w∈X1i
to avoid two independent edges between X1i and Q.
∎
4.3. On the sets X2 and X3
We next study the sets X2 and X3 in more detail.
We start with the following lemma,
which can be seen as an analogue of Lemma 5.
Lemma 9**.**
Let i,j,k∈I be distinct indices.
- (a)
If a∈X1i, b∈X1j, c∈X2i∩X2j, then the set {a,b,c} is not independent.
2. (b)
If a∈X1i, b∈X1j, c∈X1k, d∈X3i∩X3j∩X3k, then the set {a,b,c,d} is not independent.
Proof.
First, (a) is immediate since the 6-set (U\{ui,uj})∪{a,b,c} cannot be independent.
Similarly, (b) holds by considering the 6-set (U\{ui,uj,uk})∪{a,b,c,d}.
∎
The next lemma is an analogue of Claim 1 of Chen and Zhang [2].
Lemma 10**.**
Let i,j,k∈I be distinct indices.
- (a)
If ∣X1i∣=∣X1j∣=2ℓ−3, then X2i∩X2j=∅.
2. (b)
If ∣X1i∣=∣X1j∣=∣X1k∣=2ℓ−3, then X3i∩X3j∩X3k=∅.
Proof.
Let us begin with (a) by assuming, to the contrary, that c∈X2i∩X2j.
Since G[N(c)] does not contain ℓK2
while G[X1i∪{ui}]≅G[X1j∪{uj}]≅K2ℓ−2,
it follows that ∣N(c)∩(X1i∪X1j)∣≤2ℓ−2.
Hence, ∣(X1i∪X1j)\N(c)∣≥2ℓ−4.
Without loss of generality, assume that ∣X1i\N(c)∣≥(2ℓ−4)/2=ℓ−2≥4.
Note that X1j\N(c)=∅ as ω(G)≤2ℓ−2.
Let b∈X1j\N(c).
By Lemma 9 (a) then,
b is adjacent to every vertex in X1i\N(c),
i.e. ∣N(b)∩X1i∣≥4, contradicting Lemma 6.
This proves the assertion (a).
For (b), suppose that d∈X3i∩X3j∩X3k.
Then by a similar argument done in (a),
one sees that ∣(X1i∪X1j∪X1k)\N(d)∣≥4ℓ−7
in order to avoid ℓK2 in G[N(d)].
Without loss of generality, suppose that
∣X1i\N(d)∣≥∣X1j\N(d)∣≥∣X1k\N(d)∣.
Then
[TABLE]
and
[TABLE]
Also, X1k\N(d)=∅ in order to avoid K2ℓ−1 in G.
Let
[TABLE]
By Lemma 6,
none of ∣N(c)∩X1i∣,∣N(c)∩X1j∣,∣N(b1)∩X1i∣ exceeds 3.
By this fact with Lemma 9 (b),
we may assume that b1c∈/E(G), and N(c)∩X1i={a1,a2,a3}.
Then again by Lemma 6,
at most one of b2,b3,b4 is adjacent to c
and so suppose that b2c∈/E(G).
Thus, by Lemma 9,
N(b1)∩X1i=N(b2)∩X1i={a4,a5,a6}.
This gives us (ℓ−2)K2 in G[(X1j\{b1,b2})∪{uj}]
with two edges a4a5 and a6b2 in G[N(b1)],
which is impossible.
∎
We let
[TABLE]
The next lemma describes how the sets Qi intersect each other.
Lemma 11**.**
**
- (a)
Let i,j∈J be indices with i<j.
If ∣X1i∣+∣X1j∣≥2ℓ−1, then we have Qi∩Qj∩X2=∅.
Consequently, if ∣X14∣+∣X15∣≥2ℓ−1, then
[TABLE]
2. (b)
Let i,j,k∈J be indices with i<j<k.
If ∣X1i∣+∣X1j∣+∣X1k∣≥6ℓ−14,
then Qi∩Qj∩Qk∩X3=∅.
Consequently,
if ∣X13∣+∣X14∣+∣X15∣≥6ℓ−14,
then
[TABLE]
Proof.
For (a), suppose that c∈Qi∩Qj∩X2.
Then no vertex in X1i∪X1j is adjacent to c by (6),
so G[X1i∪X1j] is a clique by Lemma 9 (a).
However then, ω(G)≥∣X1i∣+∣X1j∣≥2ℓ−1.
As a result, Qi∩Qj∩X2=∅.
For (b), suppose that d∈Qi∩Qj∩Qk∩X3.
Then no vertex in X1i∪X1j∪X1k is adjacent to d
by (6).
By the assumption ∣X1i∣+∣X1j∣+∣X1k∣≥6ℓ−14,
we have
[TABLE]
We consider two cases separately according to whether ∣X1k∣≤2ℓ−7 or not.
We first consider the case ∣X1k∣≤2ℓ−7.
In this case, we have ∣X1j∣≥2ℓ−4 by the assumption.
Therefore we have ∣X1i∣,∣X1j∣≥8.
Take a vertex c∈X1k.
By Lemma 6 with t=4,
we find that ∣N(c)∩X1i∣,∣N(c)∩X1j∣≤7.
Thus we can take vertices
a∈X1i∖N(c) and b∈X1j∖N(c).
By Lemma 6 with t=2,
we find that ∣N(a)∩X1j∣,∣N(b)∩X1i∣≤3,
i.e. ∣X1j∖N(a)∣,∣X1i∖N(b)∣≥5.
By Lemma 9 (b),
this implies ∣N(c)∩(X1i∪X1j)∣≥10,
contradicting Lemma 6 with t=4.
We next consider the case ∣X1k∣≥2ℓ−6.
In this case, we have ∣X1i∣≥8, ∣X1j∣≥7 and ∣X1k∣≥6.
Thus we can take vertices
[TABLE]
By Lemma 6 with t=3,
we find that ∣N(c)∩X1i∣,∣N(c)∩X1j∣,∣N(b1)∩X1i∣≤5.
Thus we may assume that a1c,b1c,a2b1,a3b1∈E(G).
By Lemma 9 (b),
we find that a1∈N(b1)∩X1i and a2,a3∈N(c)∩X1i.
Also, by Lemma 6 with t=2,
we see that ∣N(a1)∩X1j∣≤3.
Thus we may assume that a1b2,a1b3,a1b4,a1b5∈E(G).
However, again by Lemma 9 (b),
we find that b2,b3,b4,b5∈N(c)∩X1j,
contradicting Lemma 6 with t=3.
∎
4.4. On the set X1
Based on the above observations,
we next give a closer look at X1.
Recall that 10ℓ−23≤∣X1∣≤10ℓ−15.
The next lemma gives an improved upper bound for ∣X1∣.
Lemma 12**.**
We have ∣X14∣≤2ℓ−4.
Consequently, ∣X1∣≤10ℓ−17.
Proof.
If the statement is false, then ∣X1i∣=2ℓ−3 for 1≤i≤4.
Thus, X2,X3⊂N(u5) by Lemma 10.
Then X15∪X2∪X3∪X5⊂N(u5)
and so ∣X15∣+∣X2∣+∣X3∣+∣X5∣≤d(u5)≤2ℓ+3.
This then implies that
[TABLE]
so that ∣X4∣≥5.
On the other hand, using (1) and (2),
[TABLE]
so that ∣X4∣≤4.
Since this is impossible, it follows that ∣X14∣≤2ℓ−4 and so
∣X1∣≤3(2ℓ−3)+2(2ℓ−4)=10ℓ−17.
This completes the proof.
∎
Lemma 13**.**
The graph G contains a 5-set of independent vertices
[TABLE]
for which either
(A) ∣X1∣∈{10ℓ−18,10ℓ−17} or
(B) X5=∅,
i.e. each of the five vertices in U is adjacent to a common vertex.
Proof.
Let U={u1,u2,…,u5} be a 5-set of independent vertices in G
and suppose that U does not satisfy (A).
By Lemma 12,
we have ∣X1∣≤10ℓ−19.
Then by (1) and (2), we have
[TABLE]
so that d(ui)=2ℓ+3 for some i∈I.
By Lemma 8, therefore, we may assume the existence of a 4-set Qi that satisfies (6).
Select a vertex v∈X1i and consider the 5-set U′=Qi∪{v}, which must be independent.
Note also that each of the five vertices in U′ is adjacent to ui.
Thus, U′ satisfies (B).
∎
For the rest of this paper,
U denotes a fixed 5-set of independent vertices in G
satisfying either (A) or (B) in Lemma 13.
4.5. Additional lemmas when ∣X13∣=2ℓ−3
We prepare some more lemmas on X2 under the condition ∣X13∣=2ℓ−3.
Lemma 14**.**
Suppose that ∣X13∣=2ℓ−3.
Then ∣X2∣=∣X24∣+∣X25∣−∣X24∩X25∣.
Furthermore, ∣X24∩X25∣≤1.
Proof.
The first assertion is immediate by Lemma 10 (a).
By Lemma 10 (a) (b),
we see that X2,X3⊂N(u4)∪N(u5).
Also, obviously X4,X5⊂N(u4)∪N(u5).
Thus
[TABLE]
so that ∣X24∩X25∣≤1.
This proves the lemma.
∎
Lemma 15**.**
Suppose that ∣X13∣=2ℓ−3 and v∈X2i for some i∈{4,5}.
- (a)
If v∈/X24∩X25, then ∣N(v)∩X1i∣≥∣X1i∣−3.
2. (b)
If v∈/X24∩X25 and ∣X1i∣=2ℓ−4,
then G[X1i∪{v}]≅K2ℓ−3.
Proof.
Suppose that v∈/X24∩X25,
that is, v∈X2j for some j∈{1,2,3}.
Since ω(G)≤2ℓ−2, there is w∈X1j\N(v).
By Lemma 9 (a),
observe that w is adjacent to each vertex in X1i\N(v).
Thus, ∣X1i\N(v)∣≤∣N(w)∩X1i∣≤3
by Lemma 6.
This proves (a).
We next prove the assertion (b).
Assume, to the contrary, that v∈/X24∩X25, ∣X1i∣=2ℓ−4
and G[X1i∪{v}] is not complete.
Suppose v∈X2j with j∈{1,2,3}.
By the same argument above, we see that
∣X1i\N(v)∣,∣X1j\N(v)∣≤3
since we can take some vertex w′∈X1i∖N(v) by the assumption.
However then, ∣N(v)∩(X1i∪X1j)∣≥∣X1i∣+∣X1j∣−6=4ℓ−13≥2ℓ−1,
producing ℓK2 in G[N(v)∩(X1i∪X1j∪{ui,uj})].
Since this does not occur, (b) also holds.
∎
Lemma 16**.**
Suppose that ∣X13∣=2ℓ−3.
If ∣X1i∣≥2ℓ−5 for some i∈{4,5},
then ∣X1i∣+∣X2i∣≤2ℓ.
Proof.
Note that ∣X1i∣∈{2ℓ−5,2ℓ−4} by Lemma 12.
If ∣X1i∣=2ℓ−4 and ∣X2i∣≥5, then X2i contains four vertices v1,v2,v3,v4 such that ∣N(vj)∩X1i∣≥∣X1i∣−3≥5 for 1≤j≤4 by Lemma 15 (a) since ∣X24∩X25∣≤1.
This however contradicts Lemma 7.
Thus, ∣X2i∣≤4, as desired.
Similarly, if ∣X1i∣=2ℓ−5 and ∣X2i∣≥6, say {v1,v2,…,v6}⊂X2i, then we may assume that v1v2∈E(G) as α(G)=5 and ∣N(vj)∩X1i∣≥∣X1i∣−3≥4 for 3≤j≤5.
Thus, there are three distinct vertices w3,w4,w5∈X1i such that vjwj∈E(G) for 3≤j≤5.
These three edges with the edge v1v2 and (ℓ−4)K2 in G[X1i\{w3,w4,w5}] result in ℓK2 in G[N(ui)], again a contradiction.
It follows that ∣X2i∣≤5.
∎
5. Completion of the proof
We are now prepared to prove Theorem 3.
Recall that we are now considering a fixed 5-set U of independent vertices in G
satisfying one of the conditions
(A) ∣X1∣∈{10ℓ−18,10ℓ−17} or (B) X5=∅.
In the following subsections,
we consider these two cases separately.
5.1. Case A: ∣X1∣∈{10ℓ−18,10ℓ−17}
In this subsection, we consider Case A.
In this case, observe that ∣X11∣=∣X12∣=2ℓ−3 and ∣X14∣=2ℓ−4
since otherwise we have ∣X1∣≤10ℓ−19 by Lemma 12.
Thus we also have ∣X13∣+∣X15∣∈{4ℓ−8,4ℓ−7}.
Recalling (4), we have
[TABLE]
We now consider two subcases, according to the value of ∣X13∣.
5.1.1.
Subcase A1: ∣X13∣=2ℓ−3.
In this case, ∣X15∣∈{2ℓ−5,2ℓ−4}.
Again by (4) with Lemmas 14 and 16,
[TABLE]
Thus,
[TABLE]
First, if X24∩X25=∅,
then it is immediate by (7) and (8)
that ∣X1i∣=2ℓ−∣X2i∣=2ℓ−4≥8 for i=4,5.
Let X24∩X25={v}.
By Lemma 15 (b),
each of the three vertices in X2i\{v}
is adjacent to every vertex in X1i for i=4,5.
Since neither G[N(u4)] nor G[N(u5)] contains ℓK2,
we find that N(v)∩(X14∪X15)=∅.
However then, G[X14∪X15]≅K4ℓ−8
by Lemma 9 (a),
which cannot occur since 4ℓ−8>2ℓ−2≥ω(G).
Thus, we may assume that X24∩X25=∅.
By (7) and (8) again,
∣X1i∣=2ℓ−∣X2i∣=2ℓ−4 for some i∈{4,5}.
Then Lemma 15 (b) implies
that each of the four vertices in X2i is adjacent to every vertex in X1i.
Thus, ∣N(v)∩X1i∣=∣X1i∣=2ℓ−4≥8 for every v∈X2i.
However, this contradicts Lemma 7.
5.1.2.
Subcase A2: ∣X13∣=2ℓ−4.
In this case, ∣X1i∣=2ℓ−4 for 3≤i≤5 and ∣X1∣=10ℓ−18.
We first verify that,
if v∈X2, then v∈X2j and ∣N(v)∩X1j∣≥5
for some j∈{3,4,5}.
To see this, suppose that v∈X2i∩X2j with 1≤i<j≤5.
If i≤2, then 3≤j≤5 as X21∩X22=∅
by Lemma 10.
Since ω(G)≤2ℓ−2 and ∣X1i∣=2ℓ−3,
there is a vertex w∈X1i such that vw∈/E(G).
Then w is adjacent to every vertex in X1j\N(v)
by Lemma 9 (a),
implying that ∣X1j\N(v)∣≤3 by Lemma 6.
If 3≤i<j≤5, on the other hand,
then a similar reasoning shows
that ∣X1i\N(v)∣≤3 or ∣X1j\N(v)∣≤3.
Consequently, ∣X1j\N(v)∣≤3 for some j∈{3,4,5} in each case,
i.e. ∣N(v)∩X1j∣≥∣X1j∣−3≥5.
We next verify that ∣X2∣=9.
If this is not the case, then ∣X2∣≥10.
By the pigeonhole principle,
there exists an integer j∈{3,4,5} and a 4-set S⊂X2j
such that ∣N(v)∩X1j∣≥5 for each v∈S,
contradicting Lemma 7.
Recalling (4), we have
[TABLE]
and so ∣X4∣=∣X5∣=0, which further tells us that ∣X3∣=5.
Also, ∑i=1i∣Xi∣=10ℓ+15, that is, d(ui)=2ℓ+3 for 1≤i≤5.
By Lemma 8,
we can find five 4-sets Q1,Q2,…,Q5 satisfying (6).
By Lemma 11 (a) (b) then,
[TABLE]
which is clearly impossible.
As a result, Subcases A1 and A2 are both impossible, i.e. Case A never occurs.
5.2. Case B: X5=∅
In this subsection, we next consider Case B.
In this case,
we see that ∣X1∣≥10ℓ−23+3∣X5∣≥10ℓ−20 by (3).
Since Case A never occurs, we may assume that ∣X1∣∈{10ℓ−20,10ℓ−19}.
We again consider two subcases.
5.2.1.
Subcase B1: ∣X1∣=10ℓ−20.
By (3),
we find that ∣X3∣=∣X4∣=0 and ∣X5∣=1, which in turn implies that ∣X2∣=15.
Also, ∑i=15i∣Xi∣=10ℓ+15
and so d(ui)=2ℓ+3 for 1≤i≤5,
which guarantees the existence of five 4-sets Q1,Q2,…,Q5
satisfying (6).
As done in (9), we can write
[TABLE]
and so ∑i=15∣Qi∩X2∣=15 and ∑i=15∣Qi∩X5∣=5.
Hence, if we let X5={v0},
then each Qi consists of three vertices in X2 and v0
so that ⋃i=15Qi=X2∪X5.
However then, no vertex in X1∪X2 is adjacent to v0 as v0∈Qi,
implying that N(v0)=U. Thus, d(v0)=5<2ℓ=δ(G), a contradiction.
5.2.2.
Subcase B2: ∣X1∣=10ℓ−19.
First, observe that
[TABLE]
by (1) and (2).
Thus, we find that ∣X4∣=0, ∣X5∣=1 and either
(i) ∣X2∣=13 and ∣X3∣=1 or (ii) ∣X2∣=14 and ∣X3∣=0.
We consider these two cases separately.
Let X5={v0}.
If (i) occurs,
then again ∑i=15i∣Xi∣=10ℓ+15
and so we have 4-sets Q1,Q2,…,Q5 satisfying (6).
Since ∣X1∣=10ℓ−19,
we can check that ∣X13∣+∣X14∣+∣X15∣≥6ℓ−13.
Thus by Lemma 11 (a) (b),
[TABLE]
and we arrive at the same contradiction as in Subcase B.1.
Finally, suppose that (ii) occurs.
Then ∑i=15i∣Xi∣=10ℓ+14.
Thus, four of the five vertices in U have degree 2ℓ+3
and the remaining one has degree 2ℓ+2.
Let i0∈I for which d(ui0)=2ℓ+2.
We know that there is a 4-set Qi
satisfying (6) for each i∈I\{i0}.
In the set N(ui0)\X1i0,
which contains 2ℓ+2−∣X1i0∣≥5 vertices,
we show that there exists a 3-set Q⊂N(ui0)\X1i0
such that ∣N(v)∩X1i0∣≤2 for each v∈Q.
Recall Lemma 2.
First, if ∣X1i0∣=2t−1 for some positive integer t,
then G[N(ui0)\X1i0] contains (ℓ−t−1)K2
and five vertices v1,v2,…,v5.
Since G[N(ui0)] cannot contain ℓK2,
there are at most two independent edges
between X1i0 and {v1,v2,…,v5}.
Thus, we may assume that ∣N(vj)∩X1i0∣≤2 for 1≤j≤3.
Similarly, if ∣X1i0∣=2t for some positive integer t,
then G[N(ui0)\X1i0] contains (ℓ−t−1)K2
and four vertices v1,v2,v3,v4.
Then there cannot be two or more independent edges
between X1i0 and {v1,v2,v3,v4}.
Hence, we may assume that ∣N(vj)∩X1i0∣≤1 for 1≤j≤3.
As a result,
we find a 3-set Q={v1,v2,v3}⊂N(ui0)\X1i0
such that ∣N(v)∩X1i0∣≤2 for each v∈Q.
Write Qi0′=Q and Qi′=Qi for each i∈I\{i0}.
We next verify that ∑i=15∣Qi′∩X2∣≤∣X2∣
by proving that Qi′∩Qj′∩X2=∅ for distinct integers i,j∈I.
First, since ∣X1∣=10ℓ−19, it follows that
∣X14∣+∣X15∣=∣X1∣−∑i=13∣X1i∣≥4ℓ−10≥2ℓ+2.
Hence, by Lemma 11 (a),
it suffices to verify the result for i∈I\i0 and j=i0.
If v∈Qi′∩Qj′∩X2, then no vertex in X1i is adjacent to v.
Also, we have just shown that v is adjacent to at most two vertices in X1j.
Then by Lemma 9 (a),
there are at least ∣X1j∣−2 vertices in X1j
that are adjacent to every vertex in X1i.
Hence,
ω(G)≥∣X1i∣+∣X1j∣−2≥∣X14∣+∣X15∣−2≥2ℓ,
which cannot occur.
Therefore,
[TABLE]
so v0 belongs to every Qi′.
Thus, N[v0]⊂U∪Qi0′∪X1i0.
In particular, ∣N(v0)∩X1i0∣≤2
and so d(v0)≤∣U∣+∣Qi0′\{v0}∣+2=9<2ℓ.
This is again impossible.
We conclude that Case B never occurs, either.
This completes
the proof of Theorem 3 and Theorem 1.
Acknowledgements
The authors would like to thank Prof. Futaba Fujie
for her comments and suggestions,
which greatly improved the original manuscript.