This paper explores a bipartite graph variation of the Erd ext{"o}s-S ext{"o}s Conjecture, determining maximum sizes of bipartite graphs that exclude certain bipartite trees and characterizing extremal cases.
Contribution
It extends the Erd ext{"o}s-S ext{"o}s Conjecture to bipartite graphs, identifying maximum sizes and characterizing extremal bipartite graphs avoiding specific tree subgraphs.
Findings
01
Maximum size of bipartite graphs avoiding all (n,m)-bipartite trees determined
02
Maximum size of bipartite graphs avoiding all (k,2)-bipartite trees determined
03
Extremal graphs characterized for these cases
Abstract
The Erd\H{o}s-S\'{o}s Conjecture states that every graph with average degree more than k−2 contains all trees of order k as subgraphs. In this paper, we consider a variation of the above conjecture: studying the maximum size of an (n,m)-bipartite graph which does not contain all (k,l)-bipartite trees for given integers n≥m and k≥l. In particular, we determine that the maximum size of an (n,m)-bipartite graph which does not contain all (n,m)-bipartite trees as subgraphs (or all (k,2)-bipartite trees as subgraphs, respectively). Furthermore, all these extremal graphs are characterized.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsGraph theory and applications · Limits and Structures in Graph Theory · Advanced Graph Theory Research
Full text
A Variation of the Erdős-Sós Conjecture in Bipartite Graphs
††thanks: This work is supported by the Joint NSFC-ISF Research Program (jointly funded by the National Natural Science Foundation of China and the Israel Science Foundation (No. 11561141001)), the National Natural Science Foundation of China (Nos.11531001 and 11271256), Innovation Program of Shanghai Municipal Education Commission (No. 14ZZ016) and Specialized Research Fund for the Doctoral Program of Higher Education (No.20130073110075).
The Erdős-Sós Conjecture states that every graph with average degree more than k−2 contains all trees of order k as subgraphs. In this paper, we consider a variation of the above conjecture: studying the maximum size of an (n,m)-bipartite graph which does not contain all (k,l)-bipartite trees for given integers n≥m and k≥l. In particular, we determine that the maximum size of an (n,m)-bipartite graph which does not contain all (n,m)-bipartite trees as subgraphs (or all (k,2)-bipartite trees as subgraphs, respectively). Furthermore, all these extremal graphs are characterized.
The graphs considered in this paper are finite, undirected, and simple (no loops or multiple edges). Let G=G[V;E] be a graph with vertex set V and edge set E. The number of vertices in V is called order of G and the number of edges in E is called size of G, denoted by e(G).
The degree of v∈V, the number of edges incident to v, is denoted by dG(v) and the set of neighbors of v is denoted by NG(v). Moreover, a vertex of degree one is called a pendent vertex. If u and v in V are adjacent, we say that uhitsv and vhitsu. If u and v are not adjacent, we say that umissesv and vmissesu. The path with n vertices is denoted by Pn, the star with n vertices is denoted by K1,n−1 (K1,0 is an isolated vertex, and K1,1 is an edge), the cycle with n vertices is denoted by Cn, and the double star with k1+k2 vertices which is obtained from two stars K1,k1−1 and K1,k2−1 by joining an edge between two central vertices with degree k1−1,k2−1 is denoted by Sk1,k2.
Let G and H be two vertex disjoint graphs. Denote by G∪H the vertex disjoint union of G and H and by k⋅G the vertex disjoint union of k copies of a graph G.
In addition, δ(G), Δ(G) and avedeg(G)=∣V(H)∣2e(H) are denoted by the minimum, maximum and average degree in V(G), respectively. If S⊆V(G), the induced subgraph of G by S is denoted by G[S]. Let T be a tree of order k. If there exists an injection f:V(T)→V(G) such that f(u)f(v)∈E(G) if uv∈E(T) for u,v∈V(T), we call f an embedding of T into G and G contains a copy of T as a subgraph, denoted by T⊆G.
An (n,m)-bipartite graph (or bigraph)G[U,V;E], or Bn,m, is a bipartite graph of order m+n whose vertices can be divided into two disjoint sets U and V with ∣U∣=n and ∣V∣=m such that every edge joins one vertex in U to another vertex in V. Moreover, denote by Kn,m the complete (n,m)-bipartite graph.
Furthermore, denote by Bn,m the set of all bipartite graphs Bn,m, and Tn,m the set of all the (n,m)-bipartite trees Tn,m with two partitions ∣U∣=n and ∣V∣=m, respectively.
We call a problem a Turaˊn type extremal problem if a family L of graphs are given from universe, such as Gn is a graph of order n, we try to maximize e(Gn) of Gn under the condition that Gn does not contain L∈L. The maximum value is denoted by ex(n,L). Similarly, for a given family L of bipartite graphs, the maximum value e(Bn,m) of Bn,m under the condition that Bn,m does not contain L∈L is denoted by ex(n,m;L). Furthermore, if a bipartite graph Bn,m with ex(n,m;L) edges does not contain L∈L, then this bipartite graph is called an extremal bipartite graph for L.
In 1959, Erdős and Gallai [10] proved the following theorem.
Theorem 1.1
Let G be a graph with avedeg(G)>k−2. Then G contains a path of order k as a subgraph.
Based on the above theorem and related results, Erdős and Sós proposed the following well known conjecture (for example, see [11]).
Conjecture 1.2
Let G be a graph with avedeg(G)>k−2.
Then G contains all trees of order k.
Furthermore,
[TABLE]
where Tk is the set of all trees of order k.
In [1, 2, 3], Ajtai, Komlós, Simonovits and Szemerédi proved that the Erdős-Sós Conjecture is true for sufficiently large k. Fan [12] proved that the Erdős-Sós Conjecture holds for the spiders of large size. More results on this conjecture can be referred to [4, 6, 8, 9, 18, 20, 22, 23, 24, 25, 26]. On the extremal problems on complete bipartite graph, Kővári, Sós and Turán [17] proved the following result:
Theorem 1.3
[17]**
The maximum size of a graph containing no complete bipartite graph Ka,b is at most
21ab−1n2−1/a+21(a−1)n.
Füredi and Simonovits [14] written a survey on extremal graph theory focusing on the cases when one of the excluded graphs is bipartite.
For example, Győri [16] proved that ex(n,m;C6)<2n+2m2.
Gyárfás, Rousseau and Schelp [15] proved that the following theorem.
*Furthermore, (1). If m≤l−1, then all extremal graphs are Kn,m.
(2). If l−1<m<2(l−1), then all extremal graphs are Kl−1,n∪(m−l+1)⋅K1.
(3). If m≥2(l−1), then all extremal graphs are Kl−1,m−l+1∪Kl−1,n−l+1; or Kl−1,i∪Kl−1,n−i for i=0,1,…,⌊2n⌋, when m=2(l−1).*
Moreover, they [15] also determined ex(n,m;P2l+1) and the extremal graphs. The related results about the extremal graphs with focusing on the case when one of the excluded graphs is bipartite can be referred to [5, 13, 21]. Motivated by Erdős-Sós Conjecture and the above results, in this paper, we propose the following problem.
Question 1.5
Determine ex(n,m;Tk,l) and characterize all extremal graphs, where Tk,l is set of all (k,l)-bipartite trees of order k+l.
The main results in this paper are stated as follows.
Theorem 1.6
Let n≥m. Then
[TABLE]
Furthermore, (1). If n=m, then all extremal graphs for Tn,n are (n,n)-bipartite graphs G[U,V;E] such that the degree of each vertex in U (or V) is n−1.
(2). If n=m+1, then all extremal graphs for Tn,m are (n,m)-bipartite graphs G[U,V;E] such that the degree of each vertex in V is n−1, or d(v1)=1 and d(v2)=…=d(vm)=n.
(3). If n>m+1, then all extremal graphs for Tn,m are (n,m)-bipartite graphs G[U,V;E] such that the degree of each vertex in V is n−1.
Moreover, for small k and l we have the following results.
Theorem 1.7
Let n≥m≥2 and n≥k.
(1). If k=2, then ex(n,m;Tk,2)=n+m−2.
(2). If k≥3 and m=2, then
[TABLE]
(3). If 3≤m≤k, then
[TABLE]
(4). If m≥k+1≥4, then
[TABLE]
Theorem 1.8
Let n≥m≥3.
Then
[TABLE]
The rest of this paper is organized as follows. In Sections 2, 3 and 4, the proofs of Theorems 1.6, 1.7 and 1.8 are presented, respectively. Furthermore, all extremal graphs in Theorem 1.6, 1.7 and 1.8 are characterized.
First we prove a simple result: ex(Bn,n,C2n)=n(n−1)+1, and characterize all the extremal graphs.
The proof of this result depends on the following result [7] by Chvátal which strengthens a result [19] of Moon and Moser on Hamiltonian cycles in bipartite graphs.
Lemma 2.1
[19]**
Let Bn,n=G[U,V;E] be a bipartite graph with U={u1,…,un} and V={v1,…,vn}, where d(u1)≤…≤d(un) and d(v1)≤…≤d(vn). If
[TABLE]
then Bn,n is Hamiltonian. If (1) does not hold, then there exists a non-Hamiltonian bipartite graph H=G′[X,Y;E] with X={x1,…,xn} and Y={y1,…,yn}, where d(x1)≤…≤d(xn) and d(y1)≤…≤d(yn), such that d(uk)≤d(xk) and d(vk)≤d(yk), for k=1,…,n.
Lemma 2.2
Let Bn,n=G[U,V;E] be a bipartite graph with ∣U∣=∣V∣=n≥2.
Then
[TABLE]
Furthermore, if a bipartite graph Bn,n with n2−n+1 edges does not contain C2n as a subgraph, then Bn,n=Kn,n−1+e. In other words, if
[TABLE]
holds, then G is Hamiltonian unless Bn,n is the graph obtained from Kn,n−1 by adding a pendent edge, i.e., Bn,n=Kn,n−1+e.
Proof. Let U={u1,…,un} and V={v1,…,vn} with d(u1)≤…≤d(un) and d(v1)≤…≤d(vn). Suppose that Bn,n is not Hamiltonian. Then by Lemma 2.1, there is a vertex uk such that d(uk)≤k<n and d(vn−k)≤n−k.
Hence n2−n+1≤e(Bn,n)≤k2+(n−k)n, which implies k=1 or k=n−1.
If k=n−1, then d(v1)≤1, so d(v1)=1,d(v2)=…=d(un)=n,d(u1)=…=d(un−1)=n−1, and d(un)=n, and hence Bn,n=Kn,n−1+e. If k=1, then d(u1)=1,d(u2)=…=d(un)=n,d(v1)=…=d(vn−1)=n−1,d(vn)=n, and hence Bn,n=Kn,n−1+e.
In order to prove Theorem 1.6, we need some lemmas.
Lemma 2.3
Let Tn,m=T[U,V;E] be a tree with ∣U∣=n and ∣V∣=m. If all the pendent vertices of T are in U, then n>m.
Proof. We prove the statement by induction on n+m. Since all the pendent vertices of T are in U, n+m≥3. It is easy to see the assertion holds for n+m=3.
Assume the assertion holds for n+m−1. Now let u in U be a pendent vertex of Tn,m and v be its neighbour in V. If v is not a pendent vertex of Tn−1,m=Tn,m−{u}, then all pendent vertices in Tn−1,m are in U∖{u} and by the induction hypothesis, n−1>m. So the assertion holds. If v is a pendent vertex of Tn−1,m=Tn,m−{u}, then all pendent vertices in Tn−1,m−1=Tn,m−{u,v} are in U∖{u}. Hence by the induction hypothesis, n−1>m−1, which implies ∣U∣>∣V∣.
Lemma 2.4
Let Tn,m=T[U,V;E] be a tree with ∣U∣=n≥∣V∣=m. Then there are at least n−m+1 pendent vertices in U.
Proof. Suppose that there are at most n−m pendent vertices in U. Then
let Tn′,m be the tree obtained from Tn,m by deleting all pendent vertices in U. Then n′≥m and all pendent vertices of Tn′,m are in V. Hence by Lemma 2.3, we have m>n′, which is a contradiction. So the assertion holds.
Moreover, we need the following notation.
Definition 2.5
Let Bn,n=G[U,V;E] be a bipartite graph with ∣U∣=∣V∣=n and Tn,n=T[U′,V′;E′] be a tree with ∣U′∣=∣V′∣=n. If there exist two embeddings, says, f:Tn,n→Bn,n and g:Tn,n→Bn,n, such that f(U′)=U,f(V′)=V; and g(V′)=U,g(U′)=V, then we say that Bn,n strongly contains Tn,n as a subgraph.
Lemma 2.6
Let Bn,n=G[U,V;E] be a bipartite graph with U={u1,…,un} and V={v1,…,vn} such that d(u1)≤…≤d(un) and d(v1)≤…≤d(vn), where n≥3. If e(Bn,n)≥n(n−1), then
Bn,n strongly contains all trees in Tn,n as subgraphs, unless the degree of each vertex in U (or V) is n−1.
Proof. It is sufficient for us to prove that Bn,n with e(Bn,n)=n(n−1) and d(un)=d(vn)=n strongly contains all trees in Tn,n as subgraphs for n≥3.
For n=3, it follows from Figure 1 that the assertion holds.
G_{0}$$G_{1}$$G_{2}$$G_{3}Figure 1: G0 is the only (3,3)-bipartite graph with six edges and d(u3)=d(v3)=3.G1,G2,G3are all trees in T3,3
Assume the assertion holds for n−1. Let Tn,n=G[U′,V′;E′] be any tree in Tn,n with ∣U′∣=∣V′∣=n.
We may assume that e(Bn,n)=n(n−1), d(un)=n and d(v1)≤n−2. Let Bn−1,n−1′=Bn,n−{un,v1}. Note that
[TABLE]
Let y0 be a pendent vertex of Tn,n and x be the unique neighbor of y0 in Tn,n, say, y0∈V′ and x∈U′. Let {y0,y1,…,yt} be the set of all neighbors of x in Tn,n. So, 1≤t≤n−1. Note that Tn,n−{x,y0} consists of t components each containing exactly one vertex in {y1,…,yt}. Let x′ be a vertex in Tn,n that belongs to the partite set same as x. We may assume that x′ and y1 are contained in the same component in Tn,n−{x,y0}. Then let Tn−1,n−1′ be the graph obtained from Tn,n−{x,y0} by adding the edges x′yi for 2≤i≤t. Note that Tn−1,n−1′ is a tree in Tn−1,n−1. Then by the induction hypothesis, there exists an embedding f′:Tn−1,n−1′→Bn−1,n−1′ (recall that e(Bn−1,n−1′)>(n−1)(n−2), and hence the exceptional case cannot occur). Then let f be the mapping from Tn,n to Bn,n defined as follows:
•
For any vertex z with z=x,y0, we have f(z)=f′(z).
•
Define f(x)=un, and f(y0)=v1.
•
For any edge zw in Tn,n, if {z,w}∩{x,y0}=∅, then f(zw)=f′(zw).
•
For 1≤i≤t, define f(xyi) as the edge in Bn,n connecting un and f(yi) (since d(un)=n, such an edge must exist.).
•
Define f(xy0) as the edge in Tn,n connecting un and v1.
We see that this map is indeed an embedding of Tn,n into Bn,n with f(U′)=U and f(V′)=V. Since we may also assume that d(vn)=n and d(u1)≤n−2, similarly, there exists an embedding g of Tn,n into Bn,n with g(V′)=U and g(U′)=V. So we finish our proof.
Lemma 2.7
Let Bn,m=G[U,V;E] be a bipartite graph with U={u1,…,un} and V={v1,…,vm} such that d(u1)≤…≤d(un) and d(v1)≤…≤d(vm), where n>m≥1.
If e(Bn,m)≥m(n−1) holds,
then Bn,m contains all trees in Tn,m as subgraphs, unless d(v1)=…=d(vm)=n−1, or n−m=1, d(v2)=…=d(vm)=n and d(v1)=1.
Proof. Assume that Bn,m with e(Bn,m)≥m(n−1) does not satisfy that d(v1)=…=d(vm)=n−1, or n−m=1, d(v2)=…=d(vm)=n and d(v1)=1. We will prove this lemma by induction on m. For m=1,2, it is trivial and assume that m≥3. Let d(v1)≤…≤d(vm) and d(u1)≤…≤d(un). Since e(Bn,m)≥m(n−1), we can assume that d(vm)=n, and d(v1)≥2 when n=m+1. Let Tn,m=G[U′,V′;E′] with ∣U′∣=n,∣V′∣=m be any tree in Tn,m. Since n>m, by Lemma 2.4, Tn,m has a pendent vertex in U′. Let y0∈U′ be a pendent vertex of Tn,n and x∈V′ be the unique neighbor of y0 in Tn,m. Let Tn−1,m−1′ be obtained by the same way as Lemma 2.6. We consider the following four cases.
Case 1.d(u1)≤m−2. Let Bn−1,m−1′=Bn,m−{u1,vm}. Then
[TABLE]
Case 2.n>m+1,d(vm−1)=n. Let Bn−1,m−1′=Bn,m−{u1,vm}. Then
[TABLE]
and dBn−1,m−1′(vm−1)=n−1.
Case 3.n>m+1,d(u1)=m−1 and d(vm−1)<n. Since e(Bn,m)≥m(n−1), we have m−1=d(u1)≤…≤d(un). Moreover, we have d(vm−1)=n−1 and vm−1 misses ui for some i∈{1,2,…,m}. Otherwise e(Bn,m)≤n+(n−2)(m−1)<m(n−1), a contradiction. Let Bn−1,m−1′=Bn,m−{ui,vm}. Then
[TABLE]
and dBn−1,m−1′(vm−1)=n−1.
Case 4.n=m+1,d(u1)=m−1 and d(v1)≥2. We consider the following two subcases. Subcase 4.1.e(Bn,m)>m(n−1). Let Bn−1,m−1′=Bn,m−{u1,vm}. Then
[TABLE]
Subcase 4.2.e(Bn,m)=m(n−1). Then d(u1)=…=d(un)=m−1 and d(v1)≤n−2. Moreover, there must be a vertex, say vj2=v1 missing ui1∈{u1,…,un}. Otherwise, all of {u1…,un} miss v1, hence d(v1)=0, a contradiction. By d(v1)≤n−2, there exists a vertex ui2=ui1 such that ui2 misses v1. Let Bn−1,m−1′=Bn,m−{ui2,vm}. Then
[TABLE]
Let dBn−1,m−1′(v1′)≤…≤dBn−1,m−1′(vm−1′). Since dBn−1,m−1′(v1)<n−1, e(Bn−1,m−1′)=(m−1)(n−2) and ui1 misses vj2 in Bn−1,m−1′, we get dBn−1,m−1′(v1′)≥2.
It is easy to see that in all of the above cases Bn−1,m−1′ satisfies the condition of Lemma 2.7 on m−1. Since d(vm)=n and by the induction hypothesis, Bn−1,m−1′ contains all trees in Tn−1,m−1 as subgraphs. Similarly, by the method of Lemma 2.6, we can find an embedding f:Tn,m→Bn,m, and we have done.
Theorem 1.7 can be proven by the following several Lemmas.
Lemma 3.1
Let n≥m≥2. Then ex(n,m;T2,2)=n+m−2. Moreover, if m=2, then all extremal graphs are K1,p∪K1,n−p for p=0,…,⌊2n⌋; if m≥3, then all extremal graphs are K1,n−1∪K1,m−1.
Proof. Clearly there is only one tree P4 in T2,2.
Let Bn,m be an (n,m)-bipartite graph with e(Bn,m)≥n+m−1. If Bn,m is connected, then Bn,m contains P4 as a subgraph; if Bn,m is disconnected, then there is a component which contains a cycle, hence Bn,m contains P4 as a subgraph. Moreover, K1,n−1∪K1,m−1 does not contain P4 as a subgraph. Hence ex(n,m;T2,2)=n+m−2.
Furthermore, let Bn,m be any extremal graph with n+m−2 edges, then Bn,m has exactly two components which are stars. Hence Bn,m=K1,p∪K1,n−p for p=0,…,⌊2n⌋ when m=2 and Bn,m=K1,m−1∪K1,n−1 when m≥3.
The following simple proposition is useful for the proof of the next four lemmas.
Proposition 3.2
Let Bn,m be any graph which does not contain all trees in Tk,2 as subgraphs. Then
[TABLE]
for any two vertices u,v with d(u)>⌈2k⌉−1 and d(v)>k−1 in the same partition set.
Lemma 3.3
Let n≥m=2. If k≥3, then
[TABLE]
Furthermore, (1). If n>⌊23k⌋−1, then all extremal graphs for Tk,2 are (n,2)-bipartite graphs Bn,2=G[U,V;E] with d(v1)=n, and d(v2)=⌈2k⌉−1, where V={v1,v2}.
(2). If n<⌊23k⌋−1, then all extremal graphs for Tk,2 are (n,2)-bipartite graphs Bn,2=G[U,V;E] with d(v1)=d(v2)=k−1, where V={v1,v2}.
(3). If n=⌊23k⌋−1, then all extremal graphs for Tk,2 are (n,2)-bipartite graphs Bn,2=G[U,V;E] with d(v1)=n and d(v2)=⌈2k⌉−1, or d(v1)=d(v2)=k−1, where V={v1,v2}.
Proof.
Let Tk,2 be any (k,2)-bipartite tree in Tk,2. Without loss of generality, we may assume that Tk,2=G′[U′,V′;E′] with U′={u1′,…,uk′} and V′={v1′,v2′} such that N(v1′)={u1′,…,ur′} and
N(v2′)={ur′,…,uk′}, where 1≤r≤⌈2k⌉. We consider the following three cases.
Case 1.n>⌊23k⌋−1. Let Bn,2=G[U,V;E] be any (n,2)-bipartite graph with e(Bn,2)≥n+⌈2k⌉, where V={v1,v2}. Then l=∣N(v1)∩N(v2)∣≥⌈2k⌉.
Hence assume that N(v1)={u1,…,up,up+1,…,up+l} and
N(v2)={up+1,…,up+l,up+l+1,…,up+l+q} with 0≤p≤q.
Since
[TABLE]
Bn,2 contains all trees in Tk,2 as subgraphs. If Bn,2=G[U,V;E] is any (n,2)-bipartite graph with d(v1)=n and d(v2)=⌈2k⌉−1, then Bn,2 does not contain
all trees in Tk,2 as subgraphs, in particular, a tree Tk,2 with d(v1′)=d(v2′)=⌈2k⌉ when k is odd or d(v1′)=⌈2k⌉+1,d(v2′)=⌈2k⌉ when k is even. Hence ex(n,2;Tk,2)=n+⌈2k⌉−1. Now assume that an (n,2)-bipartite graph Bn,2=G[U,V;E] with n+⌈2k⌉−1 edges does not contain all trees in Tk,2 as subgraphs. It is easy to see that V={v1,v2} with d(v1)=n and d(v2)=⌈2k⌉−1.
Case 2.n<⌊23k⌋−1. Let Bn,2=G[U,V;E] be any (n,2)-bipartite graph such that e(Bn,2)≥2(k−1)+1 with V={v1,v2} and d(v1)≥d(v2). Then d(v1)≥k and d(v2)≥2(k−1)+1−d(v2)≥2(k−1)+1−n≥⌈2k⌉ by n<⌊23k⌋−1. Clearly N(v1)∩N(v2)=∅. Hence Bn,2 contains all trees in Tk,2 as subgraphs. It is easy to see that all extremal graphs are (n,2)-bipartite graphs Bn,2=G[U,V;E] with d(v1)=d(v2)=k−1, where V={v1,v2}.
Case 3.n=⌊23k⌋−1. Then n+⌈2k⌉−1=2(k−1). Let Bn,2=G[U,V;E] be any (n,2)-bipartite graph with e(Bn,2)≥2(k−1)+1. Similarly as Cases 1 and 2, it is easy to see that
ex(n,2;Tk,2)=n+⌈2k⌉−1 and all extremal graphs are
(n,2)-bipartite graphs with d(v1)=n, and d(v2)=⌈2k⌉−1, or d(v1)=d(v2)=k−1, where V={v1,v2}.
Lemma 3.4
Let n≥m and k≥m≥3. Then
[TABLE]
*Moreover,
(1). If n≥2k−1, then all extremal graphs for Tk,2 are Kk−1,m−1∪Kn−k+1,1;
or Bn,3=G[U,V;E] with d(v1)=d(v2)=⌈2k⌉−1 and d(v3)=n, where k≥5 is odd and V={v1,v2,v3}; or k=3 and Bn,3=Sn,3.
(2). If n<2k−1,
then the all extremal graphs for Tk,2 are all (n,m)-bipartite graphs Bn,m=G[U,V;E] with d(ui)≤k−1 for i=1,2,…,n and d(vj)=k−1 for j=1,2,…,m, where U={u1,…,un} and V={v1,…,vm}; or Bn,m=G[U,V;E] with n=2k−2,m=3, d(v1)=d(v2)=⌈2k⌉−1, d(v3)=2k−2 and k being odd, where V={v1,v2,v3}.*
Proof.
(1). Let n≥2k−1. Let Bn,m=G[U,V;E] be an (n,m)-bipartite graph with
[TABLE]
that does not contain all trees in Tk,2 as subgraphs.
Let U={u1,…,un} and V={v1,…,vm} with d(u1)≤…≤d(un) and d(v1)≤…≤d(vm). Since
e(Bn,m)=(m−2)(k−1)+n and n≥2k−1, we have d(vm)>k−1. Hence, let vm−l+1,…,vm be the vertices in V with degree more than k−1.
Then N(vi)∩N(vj)=∅ for m−l+1≤i=j≤m. Otherwise Bn,m contains all trees in Tk,2 as subgraphs. So
[TABLE]
Let
[TABLE]
Then
⌈2k⌉−1≤s≤k−1, otherwise,
[TABLE]
which is a contradiction.
Furthermore, we have d(vm−l)≤s, otherwise
[TABLE]
which implies that there exists a vertex vp with m−l+1≤p≤m such that ∣N(vm−l)∩N(vp)∣≥1. So Bn,m contains all trees in Tk,2 as subgraphs. On the other hand, we have
[TABLE]
hence l=1. Otherwise, we have l=2 with s=k−1 and ∑i=m−l+1md(vi)=n, which implies that Bn,m contains all trees in Tk,2 as subgraphs, a contradiction. In addition, if d(vm)<n−(k−1), then e(Bn,m)≤d(vm)+(m−1)s<n+(m−2)(k−1), which is a contradiction. Furthermore, we claim d(vm)=n−k+1 for m≥4; and d(vm)=n−k+1 or d(vm)=n for m=3. In fact, if
d(vm)>n−(k−1), then d(vm−1)≤k−2. Otherwise vm−1 and vm has a common neighbor which implies that Bn,m contains all trees in Tk,2 as subgraphs.
Moreover, for m≥4, we have d(vm−1)≥⌈2k⌉. Otherwise
[TABLE]
which is a contradiction. Then vm−1 and vm do not have a common neighbor, otherwise Bn,m contains all trees in Tk,2 as subgraphs. So
[TABLE]
which is a contradiction. Hence d(vm)=n−k+1.
On the other hand,
[TABLE]
implies d(v1)=…=d(vm−1)=k−1. Moreover, vi and vm do not have a common neighbor for i=1,…,m−1. So
[TABLE]
If m=3 and d(v2)≥⌈2k⌉,
then v2 and v3 have no common neighbour. By
[TABLE]
we have d(v1)=d(v2)=k−1,d(v3)=n−k+1 (If d(v2)≥k, then N(v1)∩N(v2)=∅ or N(v1)∩N(v3)=∅, contradicting Proposition 3.2), which implies
[TABLE]
If m=3 and d(v2)≤⌈2k⌉−1, then by
[TABLE]
we have d(v1)=d(v2)=⌈2k⌉−1, d(v3)=n, and k is odd.
The assertion holds by a simple observation (recall the definition of “strongly containing”, there are two kinds of possible embeddings considering the partition sets of Tk,l and Bn,m when k≤m).
(2). Let n<2k−1. Let Bn,m=G[U,V;E] be an (n,m)-bipartite graph Bn,m=G[U,V;E] with e(Bn,m)=(k−1)m which does not contain all trees in Tk,2 as subgraphs, where U={u1,…,un} and V={v1,…,vm} with d(v1)≤…≤d(vm). We consider the following two cases.
Case 1.d(vm)≥k. Let
[TABLE]
Then d(vm−1)≤s′. Otherwise d(vm−1)≥s′+1, which implies that d(vm−1)+d(vm)≥n+1. So vm−1 and vm have at least one common neighbor and Bn,m contains all trees in Tk,2 as subgraphs, which is a contradiction. Furthermore, we claim
s′=⌈2k⌉−1. In fact, if s′=n−d(vm)≥⌈2k⌉, we have
[TABLE]
by n≤2k−2 and k≥m≥3, which is a contradiction.
Hence
[TABLE]
which implies that
[TABLE]
Therefore m=3, otherwise m≥4, i.e., 4(k−⌈2k⌉)≤2k−⌈2k⌉−1 which implies 2k+1≤3⌈2k⌉, a contradiction. Furthermore, by m=3, we have k+1≤2⌈2k⌉ which implies that k is odd.
In addition, by 3(k−1)≤n+2(⌈2k⌉−1), we have n≥2k−2. So n=2k−2. Moreover, it is to see that
d(v1)=d(v2)=⌈2k⌉−1 and d(v3)=n.
If k≥5, then we are done. If k=3, the assertion hold by a simple observation (recall the definition of “strongly containing”, there are two kinds of possible embeddings considering the partition sets of Tk,l and Bn,m when k≤m).
Case 2.d(vm)≤k−1. Then (k−1)m=e(Bn,m)≤(k−1)m, which implies d(v1)=…=d(vm)=k−1. By n≤2k−2 and m≥3, Bn,m contains all trees in Tk,2 as subgraphs except the tree with one vertex of degree k. Hence d(ui)≤k−1 for i=1,…,n.
Lemma 3.5
*(1) Let n>m≥3. If 3≤k≤m+1, then any (n,m)-bipartite graph Bn,m=G[U,V;E] with e(Bn,m)≥(k−1)n contains all trees in Tk,2 as subgraphs.
(2) Let n=m≥k−1, m≥3 and k≥3. Then ex(n,n;Tk,2)=(k−1)n and all extremal graphs for Tk,2 are (k−1)-regular bipartite graphs.*
Proof. (1).
We prove the assertion holds by induction on m. If m=k−1, then the degree of each vertex in V is n. Hence Bn,m contains all trees in Tk,2 as subgraphs. Assume that the assertion holds for less than m.
Let U={u1,…,un} with d(u1)≤…≤d(up)<k−1, d(up+1)=…=d(up+q)=k−1 and k−1<d(up+q+1)≤…≤d(up+q+r), where p,q,r≥0 and p+q+r=n.
If r=0, then p=0 and q=n by e(Bn,m)≥(k−1)n. Moreover, there is a vertex in V with degree at least k by n>m, otherwise
[TABLE]
Hence Bn,m contains all trees in Tk,2 as subgraphs. So we may assume that r>0 and d(ui)=k−1+xi for i=p+q+1,…,n where xi≥1. Furthermore, if there are two vertices of {up+q+1,…,un} which share the same neighbor or one vertex of {up+1,…,up+q} and one vertex of {up+q+1,…,un} which share the same neighbor, then Bn,m contains all trees in Tk,2 and the assertion holds. So assume that
[TABLE]
for i,j=p+q+1,…,n,i=j, or i=p+1,…,p+q and j=p+q+1,…,n. On the other hand,
[TABLE]
where s=∑i=p+q+1nxi. So p≤s.
Let
[TABLE]
with
[TABLE]
where {vi,1,…,vi,xi}⊆N(ui) for i=p+q+1,…,n.
Hence e(Bn−p,m−s)=(k−1)(n−p) and all the vertices of Un−p have degree k−1 in Bn−p,m−s. Note that
[TABLE]
which implies that m−s≥k−1. Therefore by the induction hypothesis, Bn−p,m−s contains all trees in Tk,2 as subgraphs.
(2). It is sufficient to prove that any non-regular bipartite graph Bn,n with e(Bn,n)≥(k−1)n contains all trees in Tk,2 as subgraphs. If n=m=k−1, it is trivial. If n=m≥k, then there exists a vertex with degree at least k. Suppose that Bn,n does not contain all trees in Tk,2 as subgraphs. Let U={u1,…,un} with d(u1)≤…≤d(up)<k−1, d(up+1)=…=d(up+q)=k−1 and k−1<d(up+q+1)≤…≤d(up+q+r), where p≥0,q≥0,r≥1 and p+q+r=n. Recall that the vertices with degree more than ⌈2k⌉−1 can not share a common neighbor with the vertices with degree more than k−1, and hence we can consider the following three cases which are based on the number of neighbors of {up+q+1,…,up+q+r}.
Case 1.n−∑i=p+q+1nd(ui)≥k−1. There are at most n−∑i=p+q+1nd(ui) vertices in U with degree k−1. Otherwise, the induced subgraph of Bn,n with vertex sets {up+1,…,up+q} and V∖∪i=p+q+1nN(ui) satisfies (1). Hence Bn,n will contain all trees in Tk,2 as subgraphs. Therefore
[TABLE]
Case 2.⌈2k⌉≤n−∑i=p+q+1nd(ui)≤k−2. Since N(ui)∩N(uj)=∅ for p+q+1≤i<j≤p+q+r, we have
[TABLE]
Case 3.n−∑i=p+q+1nd(ui)≤⌈2k⌉−1. We have
[TABLE]
All of the above three cases contradict e(Bn,n)≥(k−1)n.
Lemma 3.6
Let n≥m≥k+1≥4. Then
[TABLE]
*Furthermore, (1). If n−m≥k−1, all extremal graphs for Tk,2 are Bn,m=Bm−1,m−1k−1∪K1,n−m+1, where Bm−1,m−1k−1 is a (k−1)-regular bipartite graph or Bn,m=Sn,m when k=3.
(2). If n−m<k−1, all extremal graphs for Tk,2 are Bn,m with d(ui)≤k−1 and d(vj)=k−1 for i=1,2,…,n;j=1,2,…,m; or Bm+1,m=Sm+1,m when k=3.*
Proof.(1). Suppose that n−m≥k−1.
Let Bn,m=G[U,V;E] be an (n,m)-bipartite graph with (k−2)(m−1)+n edges which does not contain all trees in Tk,2 as subgraphs. Then there exists at least one vertex in V with degree at least k. Otherwise
[TABLE]
which is a contradiction.
Let U={u1,…,un} and V={v1,…,vm}
with d(u1)≤…≤d(un) and d(v1)≤…≤d(vm−l)<k≤d(vm−l+1)≤…≤d(vm). Moreover,
let
[TABLE]
[TABLE]
Furthermore, we have the following claim.
Claim:m−l≥k.
In fact, suppose that m−l≤k−1. Then l≥m−k+1≥2 by m≥k+1.
If s≥k−1, then
[TABLE]
which is a contradiction. If s≤k−2, then d(vm−l)≤s, otherwise vm−l and one vertex in {vm−l+1,…,vm} have at least one common neighbor, which implies that Bn,m contains all trees in Tk,2 as subgraphs. Recall that N(vi)∩N(vj)=∅ for any m−l+1≤i<j≤m. Hence
[TABLE]
which is a contradiction. Hence the claim holds.
Now we consider the following four cases.
Case 1.1.s>m−l. Then s>m−l≥k≥⌈2k⌉−1 which implies s=n−∑i=m−l+1md(vi). Hence
[TABLE]
which is a contradiction.
Case 1.2.s=m−l. Then s=m−l≥k≥⌈2k⌉−1 which implies s=n−∑i=m−l+1md(vi). Hence
[TABLE]
So l=1 and s=m−l=m−1. Then d(vm)=n−(m−1) by s=max{n−d(vm),⌈2k⌉−1}. By
[TABLE]
we have d(v1)=…=d(vm−1)=k−1.
Hence Bm−1,m−1 with Um−1=U′ and Vm−1=V′={v1,…,vm−1} has (k−1)(m−1) edges by Proposition 3.2. Therefore by Lemma 3.5, Bm−1,m−1 is (k−1)-regular bipartite graph. So
[TABLE]
where Bm−1,m−1k−1 is a (k−1)-regular bipartite graph.
Case 1.3.k−1≤s<m−l. Let d(v1)≤…≤d(vp)<k−1=d(vp+1)=…=d(vm−l)<k≤d(vm−l+1)≤…≤d(vm), where p=0 means that the degree of all vertices in V is at least k−1. By
[TABLE]
we have m−l−p−s≥(k−2)(l−1)≥0. Then s≤m−l−p.
On the other hand, since vi and one vertex of {vm−l+1,…,vm} have no common neighbor for i=p+1,…,m−l,
N(vi)⊆U∖∪j=m−l+1nN(vj). Hence
the (s,m−l−p)-bipartite graph Bs,m−l−p=G[Us,Vm−l−p;Es,m−l−p] with Us=U∖∪j=m−l+1nN(vj) and Vm−l−p={vp+1,…,vm−l} has (k−1)(m−l−p) edges and s≥k−1. By Lemma 3.5, we have p=m−l−s>0 and Bs,m−l−p is a (k−1)-regular bipartite graph, which contains all trees in Tk,2 as subgraphs except the tree with one vertex with degree k. Moreover, we have d(v1)=…=d(vp)=k−2, otherwise
[TABLE]
Since k≥3 and p>0, then either the neighbors of v1 lie in ∪j=m−l+1nN(vj) or U∖∪j=m−l+1nN(vj), hence Bn,m must contain the tree with one vertex with degree k as a subgraph (recall that Bs,m−l−p is a (k−1)-regular bipartite graph). Hence, Bn,m contains all trees in Tk,2 as subgraphs, which is a contradiction.
Case 1.4.s≤k−2. We claim that any vertex in V′ with degree at most s. Otherwise, there must be a vertex in V′ with degree more than ⌈2k⌉−1 sharing at least one common neighbour of a vertex with degree more than k−1, which contradicts Proposition 3.2. Hence
[TABLE]
with equality holds if and only if l=1, s=k−2 and d(vm)=n, d(v1)=…=d(vm−1)=k−2. If k≥4, then k−2≥⌈2k⌉. Hence Bn,m contains all trees in Tk,2 as subgraphs. If k=3, it is easy to see that d(vm)=n,d(v1)=…=d(vm−1)=1 (recall the definition of “strongly containing”, there are two kinds of possible embeddings considering the partition sets of Tk,l and Bn,m when k≤m, and there are only two graphs in T3,2). Hence Bn,m=Sn,m.
(2). Suppose that n−m≤k−2.
Let Bn,m=G[U,V;E] be an (n,m)-bipartite graph with (k−1)m edges which does not contain all trees in Tk,2 as subgraphs, where U={u1,…,un} and V={v1,…,vm}. We consider the following two cases.
Case 2.1. The degree of every vertex in V is at most k−1.
Then by (k−1)m=e(Bn,m), the degree of every vertex in V is k−1. Hence by (k−1)m>m+(k−2)≥n, there exist two vertices in V such that they have a common neighbor. So Bn,m contains all trees in Tk,2 as subgraphs
except the tree with one vertex with degree k. Hence d(ui)≤k−1 for i=1,…,n, otherwise Bn,m contains all trees in Tk,2 as subgraphs.
Case 2.2. There exists at least one vertex in V with degree at least k. Let
d(v1)≤…≤d(vm−l)<k≤d(vm−l+1)≤…≤d(vm) with l≥1, and
[TABLE]
Then
[TABLE]
[TABLE]
Moreover, let xi=d(vi)−(k−1) for i=m−l+1,…,m and d(v1)≤…≤d(vp)<k−1=d(vp+1)=…=d(vm−l), where p≥0.
Subcase 2.2.1.n−∑i=m−l+1md(vi)≥k−1. By
[TABLE]
we have p≤∑i=m−l+1mxi.
Let
[TABLE]
be a bipartite graph with Un−∑i=m−l+1md(vi)=U′ and Vm−l−p={vp+1,…,vm−l}.
Then by p≤∑i=m−l+1mxi and n−m≤k−2, we have
[TABLE]
Clearly the degree of every vertex in Vm−l−p is k−1 and n−∑i=m−l+1md(vi)≥k−1. Since Bn−∑i=m−l+1md(vi),m−l−p does not contain all trees in Tk,2 as subgraphs, and
and Bn−∑i=m−l+1md(vi),m−l−p is a (k−1)-regular bipartite graph, which contains all trees in Tk,2 as subgraphs except the tree with one vertex with degree k. Hence by n−m≤k−2, we have
[TABLE]
which implies l=1 and d(v1)=…d(vp)=k−2. We claim that p=0, otherwise, similarly as case 1.3, v1 and one vertex in {vp+1,…,vm} have a common neighbor, which implies Bn,m contains the tree with one vertex with degree k as a subgraph, and hence contains all trees in Tk,2 as subgraphs. Moreover, since p=0,l=1 and Bn−∑i=m−l+1md(vi),m−l−p=Bm−1,m−1 is a (k−1)-regular bipartite graph, we get that d(vm)=n−m+1≤k−2+1, which is a contradiction.
Subcase 2.2.2.n−∑i=m−l+1md(vi)<k−1. We claim that d(vi)≤s for i=1,2,…,m−l. Otherwise, there are one vertex in V′ with degree more than ⌈2k⌉−1 and one vertex in {vm−l+1,…,vm} have a common neighbor. Hence Bn,m contains all trees in Tk,2 as subgraphs.
Furthermore, we have s=⌈2k⌉−1. Otherwise,
[TABLE]
which implies
[TABLE]
a contradiction. By s=⌈2k⌉−1,
we have
[TABLE]
Then l=1. Furthermore, by (k−1)m≤(m−1)(⌈2k⌉−1)+m+k−2, we have (k−⌈2k⌉−1)(m−1)≤0. Hence k=3. Moreover, by 2m=(k−1)m≤(m−1)(⌈23⌉−1)+n≤(k−1)m=2m, we have n=m+1. Therefore d(vm)=n by 2m≤(m−1)+d(vm) and d(vm)≤n=m+1. Then it is easy to see that d(vm−1)=…=d(v1)=1. Since there are only two graphs in T3,2, by an easy observation, we have d(um+1)=m,d(um)=…=d(u1)=1. The assertion holds.
Now we are ready to present the proof of Theorem 1.7.
Proof.
Theorem 1.7 follows from Lemmas 3.1, 3.3, 3.4 and 3.6.
Since there are exactly three trees G1,G2 and G3 (see Figure 1) in T3,3, we have the following result for small m and n.
Lemma 4.1
Let n≥m and 1≤m≤4. Then
[TABLE]
Furthermore, (1). If m≤2, all the extremal graphs for T3,3 are Kn,m.
(2). If m=3 and n=3, then all the extremal graphs for T3,3 are B3,3=G[U,V;E] such that any vertex in U or V has degree two.
(3). If m=3 and n≥4, then all the extremal graphs for T3,3 are Bn,3=G[U,V;E] such that any vertex in U has degree two.
(4). If m=4 and n=4, then the extremal graph for T3,3 is G1′ (see Figure 2).
(5). If m=4 and n=5, then all the extremal graphs for T3,3 are B5,4=G[U,V;E] such that any vertex in U has degree two and G2′. (see Figure 2).
(6). If m=4 and n≥6, then all the extremal graphs for T3,3 are
Bn,4=G[U,V;E] such that any vertex in U has degree two.
G^{\prime}_{1}$$G^{\prime}_{3}$$G^{\prime}_{2}Figure 2: Some extremal graphs
Proof. It is trivial for m≤2. For m=3 and n=3, it follows from Theorem 1.6. For m=3 and n≥4, let Bn,3=G[U,V;E] be any bipartite graph with e(Bn,3)≥2n which does not contain all trees in T3,3 as subgraphs. If there exists a vertex in U with degree three, then it is easy to check that Bn,3 contains G1,G2 in T3,3 as subgraphs. For G3=P6, by theorem 1.4, Bn,3 contains P6 as a subgraph. Hence, Bn,3 contains all trees in T3,3 as subgraphs, a contradiction. Since Bn,3=G[U,V;E] in which degree of each vertex in U is two does not contain G1 as a subgraph, we conclude that the assertion holds for m=3 and n≥4. For m=4, and n=4,5, let B4,4=G[U,V;E] be any bipartite graph with e(B4,4)≥9 and B5,4 be any bipartite graph with e(B5,4)≥10 which does not contain all trees in T3,3 as subgraphs. If there are one vertex ui in U and one vertex vj in V with d(ui)≥3,d(vj)≥3 and ui hits vj, then it is easy to see that B4,4 and B5,4 both contain G1 and G2 as subgraphs. Furthermore, by Theorem 1.4, B4,4 and B5,4 contain P6 as a subgraph. Hence B4,4 and B5,4 contains all trees in T3,3 as subgraphs. It is easy to see that the extremal graphs in (4) and (5) do not contain G1 as a subgraph and they are the only possible extremal graphs (G1′, G2′ and B5,4=G[U,V;E] such that any vertex in U has degree two.), hence the assertion holds. For m=4 and n≥6, let Bn,4=G[U,V;E] be any bipartite graph with e(Bn,4)≥2n which does not contain all trees in T3,3 as subgraphs. If there exists a vertex in U with degree at least three, then it is easy to check that Bn,4 contains G1,G2 in T3,3 as subgraphs. Moreover, by Theorem 1.4, Bn,4 contains P6 as a subgraph. So the assertion holds.
Lemma 4.2
Let n≥m≥5. Then
[TABLE]
Furthermore, if n=m=5, all extremal graphs for G1 are K2,3∪K2,3 and G3′.
If m≥5,n≥6, then all extremal graphs for G1 are K2,n−2∪K2,m−2.
Proof.
If n=m=5, it is easy to see that the assertion holds. Now assume that m≥5 and n≥6. Let Bn,m=G[U,V;E] be an (n,m)-bipartite graph with e(Bn,m)=2n+2m−8 which does not contain G1 as a subgraph. Let U1 and V1 be set of the vertices in U and V with degree more than two, respectively. Denote by U2=U∖U1 and V2=V∖V1. Since Bn,m does not contain G1 as a subgraph, any vertex in U1 does not hit any vertex in V1. So
[TABLE]
Hence
[TABLE]
On the other hand, we have ∣U1∣≥1, otherwise e(Bn,m)≤2n which contradicts e(Bn,m)=2n+2m−8 and m≥5. Furthermore, we have ∣V1∣≥2, otherwise there is at most one vertex in V with degree at most n−1 and e(Bn,m)≤n−1+2(m−1) which contradicts e(Bn,m)=2n+2m−8. Therefore ∣U1∣=∣V1∣=2, otherwise by ∣U1∣+∣V1∣≤4, we have ∣U1∣=1 and 2≤∣V1∣≤3 which implies
[TABLE]
a contradiction.
Moreover, e(U2,V2)=0. So each vertex in U1 (V1) hits each vertex in V2 (U2), respectively. Hence Bn,m=K2,n−2∪K2,m−2.
Lemma 4.3
Let n≥m≥5. If an (n,m)-bipartite graph Bn,m=G[U,V;E] with e(Bn,m)≥2n+2m−8 is not an extremal graph in Lemma 4.2, then Bn,m contains G2 as a subgraph.
Proof.
Suppose that Bn,m=G[U,V;E] with e(Bn,m)≥2n+2m−8 is not an extremal graph in Lemma 4.2 and does not contain G2 as a subgraph.
By Lemma 4.2, Bn,m contains G1 as a subgraph. Hence there exists a vertex u in U with degree at least three and a vertex v in V with degree at least three such that u hits v. Since Bn,m does not contain G2 as a subgraph, all vertices in N(u)∪N(v)∖{u,v} are pendent vertices. Hence the induced subgraph by the vertex set N(u)∪N(v) is a component of Bn,m with d(u)+d(v) vertices and d(u)+d(v)−1 edges. Let C1,…,Cp be the components of Bn,m such that there are ui∈U∩Ci and vi∈V∩Ci with ui hits vi and d(ui)≥3, d(vi)≥3. Let Bn1,m1=G[Un1,Vm1;En1,m1] be the union of all these p components with ∣Un1∣=n1 and ∣Vm1∣=m1. Then
e(Bn1,m1)=n1+m1−p. Furthermore any two vertices with degree at least three in Bn,m−Un1∪Vm1=Bn2,m2 are not adjacent, where n2=n−n1 and m2=m−m1. Hence Bn,m−Un1∪Vm1 does not contain G1 as a subgraph. Without loss of generality, let n2≥m2. If m2≥5, then
[TABLE]
If m2=4,n2≥5, then
[TABLE]
If m2=4,n2=4, then
[TABLE]
If m2=3, then
[TABLE]
Then by Lemmas 4.1 and 4.2, Bn2,m2 contains G1 as a subgraph, which is a contradiction. If m2≤2, then
Proof.
If 3≤m≤4, then the assertion holds from Lemma 4.1. If m≥5, by Lemmas 4.2, 4.3 and Theorem 1.4, any bipartite graph Bn,m with e(Bn,m)≥2(n+m−4) edges contains G1, G2 and G3 as subgraphs except the extremal graph in Lemma 4.2. So we finish the proof. Moreover, the extremal graphs are described in Lemmas 4.1 and 4.2.
Remark Let Bm−l+1,m−l+1k−1 be a (k−1)-regular bipartite graph with order 2m−2l+2 and Bn,m;k−1=G[U,V;E] be a bipartite graph with d(ui)≤k−1, d(vj)=k−1 for i=1,2,…,n;j=1,2,…,m, where U={u1,…,un}, V={v1,…,vj}. It is easy to see that
Kn−l+1,l−1∪Kl−1,m−l+1, Kn−m+l−1,l−1∪Bm−l+1,m−l+1k−1
and Bn,m;k−1 does not contain all trees in Tk,l as subgraphs. Base on the proof of Theorems 1.7 and 1.8, we propose the following conjecture:
Conjecture 4.4
Let n≥m,k≥l,n≥k and m≥l.
(1).
If k<2l−2,m≥2l, then
[TABLE]
(2). If k≥2l−1,m−l+1≥k−1,n−m+l−1≥k, then
[TABLE]
(3). If k≥2l−1,m−l+1≥k−1,n−m+l−1≤k−1, then
[TABLE]
Acknowledgements:
The authors would like to thank the anonymous referee for many helpful and constructive suggestions to an earlier version of this paper, in particular for giving new short proofs of Lemmas 2.6 and 2.7.
Bibliography26
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] Ajtai, M., Komlós, J., Simonovits, M., Szemerédi, E.: On the approximative solution of the Erdős-Sós conjecture on trees. (manuscript)
2[2] Ajtai, M., Komlós, J., Simonovits, M., Szemerédi, E.: Some elementary lemmas on the Erdős-Sós conjecture for trees. (manuscript)
3[3] Ajtai, M., Komlós, J., Simonovits, M., Szemerédi, E.: The solution of the Erdős-Sós conjecture for large trees. (manuscript, in preparation)
4[4] Balasubramanian, S., Dobson, E.: On the Erdős-Sós conjecture for graphs with no K 2 , s . subscript 𝐾 2 𝑠 K_{2,s}. J. Graph Theory 56 , 301-310 (2007)
5[5] Balbuena, C., Carcía-Vázquez, P., Marcote, X., Valenzuela, J. C.: Counterexample to a conjecture of Győri on C 2 l subscript 𝐶 2 𝑙 C_{2l} -free bipartite graphs. Discrete Math. 307 , 748-749 (2007)
6[6] Brandt, S., Dobson, E.: The Erdős-Sós conjecture for graphs of girth 5. Discrete Math. 150 (1), 411-414 (1996)
7[7] Chvátal, V.: On Hamilton’s ideals. J. Combin. Theory Ser. B 12 , 163-168 (1972)
8[8] Dobson, E.: Constructing trees in graphs whose complement has no K 2 , s . subscript 𝐾 2 𝑠 K_{2,s}. Combin. Probab. Comput. 11 (04), 343-347 (2002)