Contents lists available atScienceDirect
Discrete Applied Mathematics
journal homepage:www.elsevier.com/locate/dam
The Turán number of the square of a path
Chuanqi Xiao
a, Gyula O.H. Katona
b,∗, Jimeng Xiao
c,b, Oscar Zamora
a,daCentral European University, Budapest, Hungary
bMTA Rényi Institute, Budapest, Hungary
cNorthwestern Polytechnical University, Xi’an, China
dUniversidad de Costa Rica, San José, Costa Rica
a r t i c l e i n f o
Article history:
Received 2 March 2021
Received in revised form 3 October 2021 Accepted 5 October 2021
Available online 21 October 2021 Keywords:
Turán number Extremal graphs Square of a path
a b s t r a c t
The Turán number of a graphH, ex(n,H), is the maximum number of edges in a graph onnvertices which does not haveHas a subgraph. LetPkbe the path withkvertices, the squarePk2ofPkis obtained by joining the pairs of vertices with distance one or two inPk. The powerful theorem of Erdős, Stone and Simonovits determines the asymptotic behavior of ex(n,Pk2). In the present paper, we determine the exact value of ex(n,P52) and ex(n,P62) and pose a conjecture for the exact value of ex(n,Pk2).
©2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
In this paper, all graphs considered are undirected, finite and contain neither loops nor multiple edges. LetGbe such a graph, the vertex set ofGis denoted by V(G), the edge set ofGbyE(G), and the number of edges inGbye(G). We denote the degree of a vertex
v
byd(v
), the minimum degree in graphGbyδ
(G), the neighborhood ofv
byN(v
) and the chromatic number of graphGbyχ
(G). Denote bymHthe graph of the vertex-disjoint union ofmcopies of the graphH.Two disjoint vertex setsUand W are completely joined inGifu
w ∈
E(G) for allu∈
U,w ∈
W. Given graphsG1and G2, whereG1andG2with disjoint vertex setsV(G1) andV(G2) and edge setsE(G1) andE(G2). The unionG=
G1∪
G2 is the graph withV(G)=
V(G1)∪
V(G2) andE(G)=
E(G1)∪
E(G2). Denote byG1⨂G2the graph obtained fromG1
∪
G2by adding all edges betweenV(G1) andV(G2).The Turán number of a graphH, ex(n
,
H), is the maximum number of edges in a graph onnvertices which does not haveHas a subgraph. The Erdős–Stone–Simonovits Theorem [4,5] asymptotically determines ex(n,
H) for all non-bipartite graphsH:ex(n
,
H)=
(1−
1χ
(H)−
1)(n 2 )
+
o(n2)There are currently very few known exact results known. For example, Füredi and Gunderson [7] described all the graphs having ex(n
,
C2k+1) edges (k≥
2) containing noC2k+1. Erdős, Füredi, Gould and Gunderson [2] determined the exact value of ex(n,
Fk) (k≥
1,n≥
50k2) fork-fan (a graph on 2k+
1 vertices consisting ofktriangles which intersect in exact one common vertex is called ak-fan and denoted byFk).The following graphs will be studied in the present paper. LetPkbe the path withkvertices, the squarePk2 ofPkis obtained by joining the pairs of vertices with distance one or two inPk, seeFig. 1. Our goal in this paper is to study
∗ Corresponding author.
E-mail addresses: chuanqixm@gmail.com(C. Xiao),katona.gyula.oh@renyi.hu(G.O.H. Katona).
https://doi.org/10.1016/j.dam.2021.10.003
0166-218X/©2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.
org/licenses/by-nc-nd/4.0/).
Fig. 1. GraphPk2.
Fig. 2. GraphEni.
ex(n
,
Pk2) and the extremal graphs for Pk2. Sinceχ
(Pk2)=
3, k≥
3, by Erdős–Stone–Simonovits Theorem, we have ex(n,
Pk2)=
n24
+
o(n2). Yet, it still remains interesting to determine the exact value of ex(n,
Pk2).The very first result of extremal graph theory gave the value of ex(n
,
P32).Theorem 1(Mantel [10]).The maximum number of edges in an n-vertex triangle-free graph is
⌊
n42⌋
, that isex(n,
P32)= ⌊
n42⌋
. Furthermore, the only triangle-free graph with⌊
n42⌋
edges is the complete bipartite graph K⌊n2⌋,⌈n 2⌉. The casek
=
4 was solved by Dirac in a more general context.Theorem 2(Dirac [1]).The maximum number of edges in an n-vertex P42-free graph is
⌊
n24
⌋
, that isex(n,
P42)= ⌊
n24
⌋
,(n≥
4).Furthermore, when n
≥
5, the only extremal graph is the complete bipartite graph K⌊n 2⌋,⌈n2⌉.
Fork
=
5, our results are given in the next two theorems, where we separate the result for the Turán number and the extremal graphs forP52.Theorem 3. The maximum number of edges in an n-vertex P52-free graph is
⌊
n2+n4
⌋
, that isex(n,
P52)= ⌊
n2+n4
⌋
,(n≥
5).Definition 1. LetEindenote a graph obtained from a complete bipartite graphKi,n−i plus a maximum matching in the class which hasivertices, seeFig. 2.
Theorem 4. Let n be a natural number, when n
=
5, the extremal graphs for P52 are E52, E53and G0, where G0is obtained from a K4plus a pendent edge. When n≥
6, if n≡
1,
2 (mod 4), the extremal graphs for P52are E⌈n 2⌉ n and E⌊
n 2⌋
n , otherwise, the extremal graph for P52is E⌈
n 2⌉ n .
Definition 2. LetT denote the flattened tetrahedron, seeT inFig. 3.
Although the determination of ex(n
,
T) is not within the main lines of our paper, we need the exact value of ex(n,
T) in order to determine ex(n,
P62).Theorem 5. The maximum number of edges in an n-vertex T -free graph (n
̸=
5) is,ex(n
,
T)=
⎧
⎪⎪
⎨
⎪⎪
⎩
⌊n2 4
⌋
+
⌊n 2
⌋
,
n̸≡
2 (mod 4),
n24
+
n2
−
1,
n≡
2 (mod 4).
Definition 3. LetTni denote a graph obtained from a complete bipartite graphKi,n−i plus a maximum matching in the classXwhich hasivertices and a maximum matching in the classY which hasn
−
ivertices, seeTniinFig. 3. LetSni denote a graph obtained fromKi,n−i plus ani-vertex star in the classX, seeSininFig. 3.2
Fig. 3. GraphsT,TniandSni.
Fig. 4. GraphsFni,jandHni..
Theorem 6. Let n(n
̸=
5,
6)be a natural number, when n≡
0 (mod 4), the extremal graph for T is Tn n2, when n
≡
1 (mod 4), the extremal graphs for T are T⌈n 2⌉ n and S⌈
n 2⌉ n , when n
≡
2 (mod 4), the extremal graphs for T are Tn n2, T
n 2+1 n and S
n n2, when n
≡
3 (mod 4), the extremal graphs for T are T⌈n 2⌉ n and S⌈
n 2⌉ n .
These two results are known for sufficiently largen′s[9], here we are able to determine the value for smalln′s.
UsingTheorems 5and6, we are able to prove the next two results forP62. Theorem 7. The maximum number of edges in an n-vertex P62-free graph(n
̸=
5)is:ex(n
,
P62)=
⎧
⎪⎪
⎨
⎪⎪
⎩
⌊n2 4
⌋
+
⌊n
−
1 2⌋
,
n≡
1,
2,
3 (mod 6),
⌊n2 4
⌋
+
⌈n 2
⌉
,
otherwise.
Definition 4. Suppose 3
̸ |
n, and 1≤
j≤
i. LetFni,jbe the graph obtained by adding vertex disjoint triangles (possibly 0) and one star withjvertices in the classXof sizeiofKi,n−i, seeFig. 4(of course 3|
(i−
j) is supposed). On the other hand if 3|
ithen add 3i vertex disjoint triangles in the classXof sizei. The so obtained graph is denoted byHni, seeFig. 4.Theorem 8. Let n
≥
6be a natural number. The extremal graphs for P62are the following ones.When n
≡
1 (mod 6)then F⌈n 2⌉,j n and H⌊
n 2⌋ n , when n
≡
2 (mod 6)then Fn 2,j n and F
n 2+1,j
n ,
when n
≡
3 (mod 6)then F⌈n 2⌉,j n and H⌈
n 2⌉+1
n ,
when n
≡
0,
4,
5 (mod 6)then Hnn2, Hn 2+1 n and H⌈
n 2⌉
n , respectively.(j can have all the values satisfying the conditions j
≤
i and3|
(i−
j)).On the basis of these results let us pose a conjecture for the general case.
Conjecture 1.
ex(n
,
Pk2)≤
max {i(⌊2k3
⌋
−
2)2
+
i(n−
i) }.
If⌊2k3
⌋
−
1divides i then the following graph gives equality here. Take a complete bipartite graph with parts of size i and n−
i, add vertex disjoint complete graphs on⌊2k3
⌋
−
1vertices to the part with i elements.Observe thatTheorems 1,2,3and7justify our conjecture for the cases whenk
=
3,
4,
5,
6. We will give some hints in Section3how we arrived to this conjecture. A weaker form of this conjecture is the following one.Conjecture 2.
ex(n
,
Pk2)=
n2
4
+
(⌊k3
⌋
−
1) n 2+
Ok(1) where Ok(1)depends only on k.2. Proofs of the main results
2.1. The Turán number and the extremal graphs for P52
Proof ofTheorem 3. The fact that ex(n
,
P52)≥
⌊n2+n 4
⌋
follows from the constructionE
⌈
n2⌉
n .
We prove the inequality ex(n
,
P52)≤
⌊n2
+
n 4⌋
(n
≥
5) (1)by induction onn.
We check the base cases first. Since our induction step will go fromn
−
4 ton, we have to find a base case in each residue class mod 4.LetGbe ann-vertexP52-free graph. Whenn
≤
3,Knis the graph with the most number of edges and does not contain P52,e(Kn)≤
⌊n2+n 4
⌋
. This settles the casesn
=
1,
2,
3. However, whenn=
4,e(K4)=
6> ⌊
42+44
⌋
, the statement is not true.Then we show that the statement is true forn
=
8. IfP42⊈G,e(G)≤ ⌊
824
⌋
. IfP42⊆
GandK4⊈G, each vertexv ∈
V(G−
P42) can be adjacent to at most 2 vertices of the copy ofP42, sincee(G−
P42)≤
5, we havee(G)≤
5+
8+
5≤
18= ⌊
82+84
⌋
. If K4⊆
G, then each vertexv ∈
V(G−
K4) can be adjacent to at most one vertex of theK4, sincee(G−
P42)≤
6, we have e(G)≤
16.Suppose(1)holds for allk
≤
n−
1, the proof is divided into 3 parts, Case 1. IfP42⊈G, then byTheorem 2,e(G)≤ ⌊
n24
⌋
.Case 2. IfP42
⊆
GandK4⊈G, then each vertexv ∈
V(G−
P42) can be adjacent to at most 2 vertices of the copy ofP42, otherwise,P52⊆
G. SinceG−
P42is an (n−
4)-vertexP52-free graph, we havee(G)
≤
5+
2(n−
4)+
e(G−
P42)≤
2n−
3+
ex(n−
4,
P52).
By the induction hypothesis, ex(n−
4,
P52)≤
⌊(n−4)2+n−4 4
⌋ then
e(G)
≤
2n−
3+
ex(n−
4,
P52)≤
2n−
3+
⌊(n
−
4)2+
n−
4 4⌋
=
⌊n2
+
n 4⌋
(n
≥
5).
(2)Case 3. IfK4
⊆
G, then each vertexv ∈
V(G−
K4) can be adjacent to at most one vertex of theK4, otherwise,P52⊆
G.SinceG
−
K4is an (n−
4)-vertexP52-free graph, we have e(G)≤
6+
(n−
4)+
e(G−
K4)≤
n+
2+
ex(n−
4,
P52).
By the induction hypothesis, ex(n−
4,
P52)≤
⌊(n−4)2+n−4 4
⌋ , thus
e(G)
≤
n+
2+
⌊(n
−
4)2+
n−
4 4⌋
=
5+
⌊n2
−
3n 4⌋
≤
⌊n2
+
n 4⌋
(n
≥
5).
□ (3)4
Proof ofTheorem 4. We determine the extremal graphs forP52 by induction onn. LetGbe ann-vertexP52-free graph satisfying(1)with equality. It is easy to check, whenn
=
5, the extremal graphs forP52areG0,E25andE53. Whenn=
6,
7,
8, the extremal graphs forP52areE63andE64,E74,E84, respectively.SupposeTheorem 4is true fork
≤
n−
1, whenn≥
9, the proof is divided into 3 parts.Case 1. IfP42⊈G, the equality in(1)cannot hold, then we cannot find any extremal graph forP52in this case.
Case 2. IfP42
⊆
GandK4⊈G, the equality holds in inequality(2)if and only if each vertexv ∈
V(G−
P42) is adjacent to 2 vertices of theP42andG−
P42is an extremal graph onn−
4 vertices forP52. Leta,
b,
canddbe four vertices of a copy ofP42,dP24(b)
=
dP24(c)
=
3. By the induction hypothesis,G−
P42is obtained from a complete bipartite graphKi,n−4−i plus a maximum matching inX′, whereX′is the class ofG−
P42with sizei. It is easy to check that every vertexv ∈
V(G−
P42) can be adjacent to eitheraanddorbandc.Since
|
V(G−
P42)| ≥
5, we have|
V(X′)| ≥
2. The endpoints of an edge inG−
P42cannot be both adjacent tobandc, otherwise, they form aK4. Also, the endpoints of an edge inG−
P42which have one end vertex as a matched vertex in X′and one end vertex inY′can be both adjacent to none of{
a,
b,
c}
andd, otherwise, these would create aP52. If there exists a matched vertexv ∈
X′which is adjacent tobandc, then all verticesw ∈
N(v
) should be adjacent toaandd, these form aP52. Hence, it is only possible that all matched vertices inX′are adjacent to bothaandd, all vertices inY′ are adjacent tobandc. When there exists an unmatched vertexv
0∈
X′, sinceN(v
0)=
Y′, ifv
0 is adjacent tobandc, we haveP52⊆
G. ThusGis obtained from a complete bipartite graphKi+2,n−i−2plus a maximum matching inX, where X=
X′∪ {
b,
c}
andY=
Y′∪
a∪
d. Therefore, ifG−
P42isE⌈n−4 2 ⌉
n−4 thenGisE⌈
n 2⌉ n , ifE⌊
n−4 2 ⌋
n−4 thenGisE⌊
n 2⌋ n .
Case 3. IfK4
⊆
G, the inequality in(3)can be equality only whenn=
5 and the vertexv ∈
V(G−
K4) is adjacent to one vertex of theK4, that isG0. □2.2. The Turán number and the extremal graphs for T
To proveTheorem 5, we need the following lemmas.
Lemma 9. Let G be an n-vertex T -free nonempty graph such that for each edge
{
x,
y} ∈
E(G), d(x)+
d(y)≥
n+
2holds, then we have K4⊆
G.Proof. From the condition we know that each edge belongs to at least two triangles. Letabcandbcdbe two triangles, ifa is adjacent todthena
,
b,
canddinduce aK4, if not, since edge{
b,
d}
is contained in at least two triangles, there exists at least one vertexesuch thatbdeis a triangle. Similarly, edge{
c,
d}
is also contained in at least two triangles, then, either there exists a vertexf which is adjacent tocandd, this implies that verticesa,
b,
c,
d,
eandf induce aT, orcis adjacent toe, this implies that verticesb,
c,
dandeinduce aK4. □Lemma 10. Let G be an n-vertex(n
≥
7)T -free graph and K4⊆
G, then e(G)≤
2n−
2+
ex(n−
4,
T). For n≥
8, the equality might hold only if each vertexv ∈
V(G−
K4)is adjacent to 2 vertices of the K4.Proof. If there exists vertex
v ∈
V(G−
K4), such thatv
is adjacent to at least 3 vertices of the K4, it is simple to check that every other vertexu∈
V(G−
K4) can be adjacent to at most one vertex of the K4, otherwiseT⊆
G, then e(G)≤
6+
4+
(n−
5)+
e(G−
K4)≤
n+
5+
ex(n−
4,
T). If not, each vertex inG−
K4is adjacent to at most 2 vertices of theK4, thene(G)≤
6+
2(n−
4)+
e(G−
K4)≤
2n−
2+
ex(n−
4,
T). Whenn≥
8,e(G)≤
2n−
2+
ex(n−
4,
T), the equality holds only if each vertexv ∈
V(G−
K4) is adjacent to 2 vertices of theK4. □Proof ofTheorem 5. Let
fT(n)
=
⎧
⎪⎪
⎨
⎪⎪
⎩
⌊n2 4
⌋
+
⌊n 2
⌋
,
n̸≡
2 (mod 4),
n24
+
n2
−
1,
n≡
2 (mod 4).
The fact that ex(n
,
T)≥
fT(n) follows from the constructionT⌈
n2⌉
n . Next, we show the inequality
ex(n
,
T)≤
fT(n) (4)by induction onn.
LetGbe ann-vertexT-free graph. first, we show the induction steps, in the end we will show the base cases which are needed to complete the induction.
Suppose (4) holds for alll
≤
n−
1, in the following cases, we will assume thatk≥
2, the proof is divided into 4 cases.Case 1. Whenn
=
4k, we divide the proof of ex(4k,
T)≤
fT(4k)=
4k2+
2kinto 2 subcases. LetGbe a 4k-vertexT-free graph.(i) If
δ
(G)≤
2k+
1, after removing a vertex of minimum degree and by the induction hypothesis ex(4k−
1,
T)=
4k2−
1, we gete(G)
≤
ex(4k−
1,
T)+
2k+
1≤
4k2−
1+
2k+
1=
fT(4k).
(5)(ii) If
δ
(G)≥
2k+
2, then for each edge{
u, v } ∈
E(G), d(u)+
d(v
)≥
4k+
4. ByLemmas 9and10 and the induction hypothesis ex(4k−
4,
T)=
4(k−
1)2+
2(k−
1), we gete(G)
≤
2n−
2+
ex(4k−
4,
T)=
8k−
2+
4(k−
1)2+
2(k−
1)=
fT(4k).
(6) Therefore, ex(4k,
T)≤
fT(4k).Case 2. Whenn
=
4k+
1, we divide the proof of ex(4k+
1,
T)≤
fT(4k+
1)=
4k2+
4kinto 3 subcases. LetGbe a (4k+
1)-vertexT-free graph.(i) If
δ
(G)≤
2k, after removing a vertex of minimum degree and by the induction hypothesis ex(4k,
T)=
4k2+
2k, we havee(G)
≤
ex(4k,
T)+
2k≤
fT(4k+
1).
(7)Now, we assume that in the following two cases
δ
(G)≥
2k+
1. Then for any pair of vertices{
u, v } ∈
E(G),d(u)+
d(v
)≥
4k+
2 holds.(ii) Suppose that there exists an edge
{
u, v } ∈
E(G), such thatd(u)+
d(v
)=
4k+
2. This implies thatuandv
have at least one common neighbor. Deleting{
u, v }
we can use the induction hypothesis ex(4k−
1,
T)=
4k2−
1. Thene(G)
≤
4k+
1+
ex(4k−
1,
T)=
fT(4k+
1).
(8)(iii) For each edge
{
u, v } ∈
E(G), d(u)+
d(v
)≥
4k+
3 holds. By Lemmas 9 and 10 and the induction hypothesis ex(4k−
3,
T)=
4(k−
1)2+
4(k−
1) we gete(G)
≤
2n−
2+
ex(4k−
3,
T)=
8k+
4(k−
1)2+
4(k−
1)=
fT(4k+
1).
(9) Therefore, ex(4k+
1,
T)≤
fT(4k+
1).Case 3. Whenn
=
4k+
2, we divide the proof of ex(4k+
2,
T)≤
fT(4k+
2)=
4k2+
6k+
1 into 2 subcases. LetGbe a (4k+
2)-vertexT-free graph.(i) If
δ
(G)≤
2k+
1, after removing a vertex of minimum degree and by the induction hypothesis ex(4k+
1,
T)=
4k2+
4k, we gete(G)
≤
ex(4k+
1,
T)+
2k+
1≤
4k2+
6k+
1=
fT(4k+
2).
(10)(ii) If
δ
(G)≥
2k+
2, then for each edge{
u, v } ∈
E(G), d(u)+
d(v
)≥
4k+
4. ByLemmas 9and10 and the induction hypothesis ex(4k−
2,
T)=
4(k−
1)2+
6(k−
1)+
1, we gete(G)
≤
2n−
2+
ex(4k−
2,
T)=
8k+
2+
4(k−
1)2+
6(k−
1)+
1=
fT(4k+
2).
(11) Therefore, ex(4k+
2,
T)≤
fT(4k+
2).Case 4. Whenn
=
4k+
3, we divide the proof of ex(4k+
3,
T)≤
fT(4k+
3)=
4k2+
8k+
3 into 2 subcases. LetGbe a (4k+
3)-vertexT-free graph.(i) If
δ
(G)≤
2k+
2, after removing a vertex of minimum degree and by the induction hypothesis ex(4k+
2,
T)=
4k2+
6k+
1, we gete(G)
≤
ex(4k+
2,
T)+
2k+
2≤
4k2+
8k+
3=
fT(4k+
3).
(12)(ii). If
δ
(G)≥
2k+
3, then for each edge{
u, v } ∈
E(G),d(u)+
d(v
)≥
4k+
6. By Lemmas 9and 10and the induction hypothesis ex(4k−
1,
T)=
4(k−
1)2+
8(k−
1)+
3, we gete(G)
≤
2n−
2+
ex(4k−
1,
T)=
8k+
4+
4(k−
1)2+
8(k−
1)+
3=
fT(4k+
3).
(13) Therefore, ex(4k+
3,
T)≤
fT(4k+
3).Now we show the base cases which are needed to complete the induction steps. Since our induction steps will go from n
−
1 ton,n−
2 tonandn−
4 ton, we will require to show the statement is true for cases whenn=
3,
4,
6 and 9.Whenn
≤
4,Knis the graph with the most number of edges, ande(Kn)=
fT(n).Whenn
=
5,e(K5)=
10>
fT(5), the statement is not true, but we will see that the statement is true forn=
9.Whenn
=
6, letv
be a vertex with minimum degree. Ifδ
(G)=
1, sincee(G− v
)≤
10, we gete(G)≤
11. Ifδ
(G)=
2 ande(G)=
12, then the only possibility is that G− v
isK5, but thenT⊆
G, and we havee(G)≤
11. Suppose nowδ
(G)≥
3. IfK4⊆
Gand there exists a vertexu∈
V(G−
K4) which is adjacent to at least 3 vertices of the copy ofK4, thenw ∈
V(G−
K4−
u) can be adjacent to at most one vertex of theK4, otherwise,T⊆
G. This contradictsδ
(G)≥
3. Then in this case it is only possible that{
u, w } ∈
E(G) and bothuandw
are adjacent to 2 vertices of theK4which implies that e(G)≤
11. IfK4⊈G, then by Turán’s Theorem, we havee(G)≤
12 and the Turán graphT(6,
3) is the uniqueK4-free graph which has 12 edges, however,T⊆
T(6,
3), thene(G)≤
11=
fT(6). Summarizing:e(G)≤
11≤
fT(6).6
Fig. 5.
Fig. 6.
Whenn
=
9, suppose first that there exists a pair of vertices{
u, v } ∈
E(G), such thatd(u)+
d(v
)≤
10. Deleting{
u, v }
and using ex(7,
T)=
15, we gete(G)≤
9+
15=
24=
fT(9). If for each pair of vertices{
u, v } ∈
E(G),d(u)+
d(v
)≥
11 holds, byLemma 9, we obtainK4⊆
G. LetG′denote the graphG−
K4. Ife(G′)≤
8, since the number of edges betweenK4 andG′is at most 10, we havee(G)≤
6+
10+
8=
24. Ife(G′)≥
9, thenK4⊆
G′and the vertexw ∈
G′−
K4is adjacent to at least 3 vertices of the copy ofK4inG′. This implies that each vertex fromG−
G′can be adjacent to at most 1 vertex ofG′− w
, then the number of edges betweenG−
G′andG′is at most 8, we can conclude that,e(G)≤
6+
8+
10=
24, e(G)≤
24=
fT(9).It is easy to see that the casen
=
7 can be proved usingn=
3 andn=
6 (Case 4). Similarly, the casen=
8 follows byn=
7 andn=
4 (Case 1). Hence the casesn=
6,
7,
8,
9 are settled forming a good bases for the induction. □Now, we determine the extremal graphs forT.
Proof ofTheorem 6. Similarly to the proof ofTheorem 5, first, we show the induction steps, in the end we will show the base cases which are needed to complete the induction.
Suppose that the extremal graphs forTare as shown inTheorem 5forl
≤
n−
1. In the following cases, we will assume thatk≥
2.LetGbe ann-vertexT-free graph withe(G)
=
fT(n). The proof is divided into 4 cases following the steps of the proof ofTheorem 5.Case 1. Whenn
=
4k,fT(n)=
4k2+
2k.(i) If
δ
(G)≤
2k+
1, the equality in(5)holds only when there exists av ∈
V(G), such thatd(v
)= δ
(G)=
2k+
1 and G− v
is an extremal graph forT on 4k−
1 vertices. By the induction hypothesis,G− v
can be eitherT4k2k−1orS4k2k−1. Let X′andY′be the classes inG− v
with size 2kand 2k−
1, respectively.WhenG
− v
isT4k2k−1, it can be easily checked thatv
cannot be adjacent to the two endpoints of an edge which have two matched vertices located in different classes, otherwise,T⊆
G, seeFig. 5. Letw
be the unmatched vertex inY′. Sinced(v
)=
2k+
1,N(v
) must contain the unmatched vertexw ∈
Y′, then the only way to avoidT⊆
Gis choosing N(v
)= w ∪
X′. Consequently,G=
T4k2kholds.WhenG
− v
isS4k2k−1, letx1denote the center of the star inX′. Ifv
is adjacent to the two endpoints of the edge{
xi,
yj}
(xi∈
X′,
yi∈
Y′,
2≤
i≤
2k,
1≤
j≤
2k−
1), thenT⊆
G(seeFig. 6). We obtained a contradiction. Butd(v
)=
2k+
1 implies that this is always the case.(ii) If
δ
(G)≥
2k+
2, this implies thate(G)≥
2k(2k+
2)=
4k2+
4k, which contradicts the fact that ex(4k,
T)=
4k2+
2k.That is,Gcan only beT
n 2 n .
Case 2. Whenn
=
4k+
1,fT(n)=
4k2+
4k.(i) If
δ
(G)≤
2k, the equality in(7)holds only if there existsv ∈
V(G), such thatd(v
)= δ
(G)=
2kandG− v
is an extremal graph forT on 4kvertices. By the induction hypothesis,G− v
isT4k2k. All neighbors ofv
should be located in the same class, otherwise,T⊆
G, we get thatGisT4k2k++11, that isT⌈n 2⌉ n .
Fig. 7.
If
δ
(G)≥
2k+
1, then for any pair of vertices{
u, v } ∈
V(G),d(u)+
d(v
)≥
4k+
2. Here we distinguish two subcases.(ii)Suppose that there exists an edge
{
u, v } ∈
E(G) such thatd(u)+
d(v
)=
4k+
2. The equality in(8)holds only if whend(u)=
d(v
)=
2k+
1 andG−
u− v
is an extremal graph forT on 4k−
1 vertices. By the induction hypothesis, G−
u− v
can be eitherT4k2k−1orS4k2k−1. LetX′andY′be the classes inG−
u− v
with size 2kand 2k−
1, respectively.WhenG
−
u− v
isT4k2k−1, as in the previous case, neitherunorv
can be adjacent to the two endpoints of an edge which have two matched vertices located in different classes, seeFig. 5. If N(u)− v ̸=
X′, thenu is adjacent to the unmatched vertexw
inY′and the other 2k−
1 neighbors ofuare all located inX′, say,N(u)− v − w = {
x1, . . . ,
x2k−1}
and{
x2k−1,
x2k} ∈
E(X′), otherwise,T⊆
G. Since|
X′| ≥
4, in this case,v
cannot be adjacent toxi (1≤
i≤
2k−
2), otherwise,T⊆
G, seeFig. 7. Nowv
should choose 2kneighbors among the rest 2k+
1 vertices inV(G−
u− v −
⋃2k−2i=1 xi), which implies that
v
is adjacent to the two endpoints of an edge which have two matched vertices located in different classes as endpoints, thenT⊆
G. Hence,N(u)− v =
X′, similarly,N(v
)−
u=
X′. Thus,GisT4k2k++11=
T4k2k+1, that isT⌈n 2⌉ n . Let us now consider the case whenG
−
u− v
isS4k2k−1. Letx1denote the center of the star inX′. Ifuis adjacent to the two endpoints of the edge{
xi,
yj}
(2≤
i≤
2k,
1≤
j≤
2k−
1), thenT⊆
G. Thus, there are only two possibilities for T ⊈G:N(u)− v =
X′orN(u)− v =
Y′∪
x1. The same holds forv
and it is easy to check that ifN(u)− v =
N(v
)−
u, thenT⊆
G. From the above, the only possibility forT ⊈Gis that whenN(u)− v =
X′andN(v
)−
u=
Y′∪
x1or in the another way around, which implies thatGisS4k2k++11, that isS⌈n 2⌉ n .
(iii) Suppose that for each edge
{
u, v } ∈
E(G),d(u)+
d(v
)≥
4k+
3 holds. Letd(v
)= δ
(G), then eitherd(v
)=
2k+
1 ord(v
)≥
2k+
2, but in both cases, each neighbor ofv
has degree at least 2k+
2. Then all 4k+
1 vertices have degree at least 2k+
1, but 2k+
1 of them, which are the neighbors ofv
, have degree at least one larger. This implies that e(G)≥
(4k+1)(2k+1)+2k+12
=
4k2+
4k+
1, which contradicts the fact that ex(4k+
1,
T)=
4k2+
4k.That is,Gcan be eitherT⌈
n 2⌉ n orS⌈
n 2⌉ n .
Case 3. Whenn
=
4k+
2 we havefT(n)=
4k2+
6k+
1.(i)If
δ
(G)≤
2k+
1, the equality holds in(10)only if there existsv ∈
V(G), such thatd(v
)= δ
(G)=
2k+
1 andG− v
is an extremal graph forT on 4k+
1 vertices. By the induction hypothesis,G− v
can be eitherT4k2k++11orS4k2k++11.Suppose first thatG
− v
isT4k2k++11. LetX′anyY′be the classes inG− v
with size 2k+
1 and 2k,w
be the unmatched vertex inX′. The vertexv
cannot be adjacent to the two endpoints of an edge which have two matched vertices located in different classes. Sinced(v
)=
2k+
1, there are two possibilities to avoidT:N(v
)=
X′orN(v
)=
Y′∪ w
, which implies thatGis eitherT4k2k++21orT4k2k++22, that isTn 2 n orT
n 2+1
n .
WhenG
− v
isS4k2k++11. LetX′be the class inG− v
which contains a star andY′be the other class of theG− v
. Also, letx1 denote the center of the star inX′. Since,d(v
)=
2k+
1 andv
cannot be adjacent to the two endpoints of an edge which is not incident withx1, we get eitherN(v
)=
Y′∪
x1orN(v
)=
X′. IfN(v
)=
X′,GisS4k2k++21, that isSn 2
n. IfN(
v
)=
Y′∪
x1,G isS4k2k++22, that isSn 2+1
n . It is easy to see thatS
n 2+1
n is isomorphic toS
n n2.
(ii)If
δ
(G)≥
2k+
2, thene(G)≥
(k+
1)(4k+
2)=
4k2+
6k+
2, which contradicts the fact that ex(4k+
2,
T)=
4k2+
6k+
1.Therefore,Gcan beT
n 2 n,T
n 2+1
n orS
n 2 n.
Case 4. Whenn
=
4k+
3 we havefT(n)=
4k2+
8k+
3.(i)If
δ
(G)≤
2k+
2, the equality holds in(12)only if there existsv ∈
V(G), such thatd(v
)= δ
(G)=
2k+
2 andG− v
is an extremal graph forT on 4k+
2 vertices. By the induction hypothesis,G− v
can beT4k2k++21,T4k2k++22orS4k2k++21.WhenG
− v
isT4k2k++21orT4k2k++22, similarly to Case 1 (i),Gcan only beT4k2k++32, that isT⌈
n2⌉
n .
WhenG
− v
isS2k4k++21, similarly to Case 2 (ii),Gcan only beS4k2k++23, that isS⌈
n2⌉
n .
(ii) If
δ
(G)≥
2k+
3, then e(G)≥
(2k+3)(4k+3)2
>
4k2+
9k+
4>
4k2+
8k+
3, which contradicts the fact that ex(4k+
3,
T)=
4k2+
8k+
3.Therefore, in this case,Gis eitherT
⌈
n2⌉
n orS
⌈
n2⌉
n .
Now we check the base cases which are needed to complete the induction.
8