On the anti-Ramsey number of forests
Chunqiu Fang
[email protected], supported in part by CSC(No. 201806210164), and NNSFC (No. 11771247).
Department of Mathematical Sciences, Tsinghua University, Beijing 100084, China
Alfréd Rényi Institute of Mathematics, Hungarian Academy of Sciences, Reáltanoda u.13-15, 1053 Budapest, Hungary
Yau Mathematical Sciences Center, Tsinghua University, Beijing 10084, China
Ervin Győri
[email protected], supported in part by the National Research, Development and Innovation Office NKFIH, grants K116769, K117879 and K126853.
Alfréd Rényi Institute of Mathematics, Hungarian Academy of Sciences, Reáltanoda u.13-15, 1053 Budapest, Hungary
Mei Lu
[email protected], supported in part by NNSFC (No. 11771247).
Department of Mathematical Sciences, Tsinghua University, Beijing 100084, China
Jimeng Xiao
[email protected], supported in part by CSC(No. 201706290171).
Alfréd Rényi Institute of Mathematics, Hungarian Academy of Sciences, Reáltanoda u.13-15, 1053 Budapest, Hungary
Department of Applied Mathematics, Northwestern Polytechnical University, Xi’an, China
Xi’an-Budapest Joint Research Center for Combinatorics, Northwestern Polytechnical University, Xi’an, China
Abstract
We call a subgraph of an edge-colored graph rainbow subgraph, if all of its edges have different colors. The anti-Ramsey number of a graph G in a complete graph Kn, denoted by ar(Kn,G), is the maximum number of colors in an edge-coloring of Kn with no rainbow subgraph copy of G. In this paper, we determine the exact value of the anti-Ramsey number for star forests and the approximate value of the anti-Ramsey number for linear forests. Furthermore, we compute the exact value of
ar(Kn,2P4) for n≥8 and ar(Kn,Sp,q) for large n, where Sp,q is the double star with p+q leaves.
Keywords: Anti-Ramsey number, star forest, linear forest, double star.
1. Introduction
Let G be a simple undirected graph. For x∈V(G), we denote the neighborhood (the set of neighbors of x) and the degree of x in G by NG(x) and dG(x), respectively. The maximum degree and the minimum degree of G are denoted by Δ(G) and δ(G), respectively. For ∅=X⊂V(G), G[X] is the subgraph of G induced by X and G−X is the subgraph of G induced by V(G)∖X. If X={x}, then G−X will be denoted by G−x for short. Given a graph G=(V,E), for any (not necessarily disjoint) vertex sets A,B⊂V, let EG(A,B):={uv∈E(G)∣u=v,u∈A,v∈B}. A star forest is a forest whose components are stars and a linear forest is a forest whose components are paths. We use tG and G to denote t vertex-disjoint copies and the complement of G, respectively. Given two vertex disjoint graphs G1 and G2, we denote by G1+G2 the join of graphs G1 and G2, that is the graph obtained from G1∪G2 by joining each vertex of G1 with each vertex of G2.
We call a subgraph of an edge-colored graph rainbow, if all of its edges have different colors. Let G be a graph. The anti-Ramsey number ar(Kn,G) is the maximum number of colors in an edge-coloring of Kn which has no rainbow copy of G. The Turán number ex(n,G) is the maximum number of edges of a simple graph on n vertices without a copy of G.
The anti-Ramsey number was first studied by Erdős, Simonovits and Sós [6]. They showed that the anti-Ramsey number is closely related to Turán number. Since then, there are plentiful results in this field, including cycles [1, 17], cliques [16, 19], trees [12, 13] and so on. See Fujita, Magnant and Ozeki [7, 8] for an abundant survey. Among these results, almost the considered graphs are connected graphs, and a few unconnected graphs are considered including matchings [4, 11], vertex-disjoint cliques [21]. In this paper, we will consider the cases that G is a star forest or a linear forest.
We mention some of the results, which are relevant to our work.
Jiang [12] and Montellano-Ballesteros [18] independently found the anti-Ramsey number for stars.
Theorem 1. ([12],[18]) For n≥p+2≥3,
[TABLE]
By Theorem 1, ar(Kn,K1,p)=⌊2(p−2)n⌋+1 for n≥3p+4 and p≥2.
Simonovits and Sós [20] considered the anti-Ramsey number for paths and obtained the following result.
Theorem 2. ([20]) Let Pk+1 be a path of length k≥2 and k≡r (mod2),0≤r≤1. For large enough n (n≥45k+C for some universal constant C),
[TABLE]
Let Ωk denote the family of graphs that contain k vertex-disjoint cycles. Jin and Li [14] computed the anti-Ramsey number for Ω2.
Theorem 3. ([14]) For any n≥6, ar(Kn,Ω2)=max{2n−2,11}.
The Turán number of tK2 was determined by Erdős and Gallai [5] as ex(n,tK2)=max{(22t−1),(2t−1)+(t−1)(n−t+1)} for n≥2t≥2.
The anti-Ramsey number of matchings was first considered by Schiermeyer [19].
Theorem 4.([19]) For n≥3t+3≥8, we have
[TABLE]
Later, Chen, Li and Tu [4] and independently Fujita, Kaneko, Schiermeyer and Suzuki [7] showed that ar(Kn,tK2)=ex(n,(t−1)K2)+1 for n≥2t+1≥5. The values ar(K2t,tK2)=ex(2t,(t−1)K2)+1 for 2≤t≤6 and ar(K2t,tK2)=ex(2t,(t−1)K2)+2 for t≥7 were determined in [4] and by Haas and Young [11], independently.
Gilboa and Roditty [9] considered the graphs with small connected components and proved inductive results of the form “if ar(Kn,G∪t0Ps)≤f(n,t0,G) for sufficiently large n, then ar(Kn,G∪tPs)≤f(n,t,G) for sufficiently large n and t≥t0, where s=2 or 3.” These results imply the following theorem.
Theorem 5. ([9]) For sufficiently large n,
(1) ar(Kn,P3∪tP2)=(t−1)(n−2t)+1 for t≥2;
(2) ar(Kn,P4∪tP2)=t(n−2t+1)+1 for t≥1;
(3) ar(Kn,C3∪tP2)=t(n−2t+1)+1 for t≥1;
(4) ar(Kn,tP3)=(t−1)(n−2t)+1 for t≥1;
(5) ar(Kn,Pk+1∪tP3)=(t+⌊2k⌋−1)(n−2t+⌊k/2⌋)+1+(kmod2) for k≥3 and t≥0;
(6) ar(Kn,P2∪tP3)=(t−1)(n−2t)+2 for t≥1;
*(7) ar(Kn,kP2∪tP3)=(t+k−2)(n−2t+k−1)+1 for k≥2 and t≥2.
*
The Turán number of star forests and linear forests are considered by Lidický, Liu and Palmer [15].
Theorem 6.([15]) Let F=⋃i=1tK1,pi be a star forest and p1≥p2≥⋯≥pt≥1. For n sufficiently large,
[TABLE]
Theorem 7.([15]) Let F=⋃i=1kPpi be a linear forest, where k≥2 and pi≥2 for 1≤i≤k. If at least one pi is not 3, then for n sufficiently large,
[TABLE]
where c=1 if all pi are odd and c=0 otherwise.
In Sections 2 and 3, we generalize Theorem 5 by considering the anti-Ramsey number of star forests and linear forests, respectively.
Theorem 8. Let F=⋃i=1tK1,pi be a star forest, where p1≥3, p1≥p2≥⋯≥pt≥1. Let s=max{i:pi≥2,1≤i≤t}. For n≥3t2(p1+1)2, we have
[TABLE]
where r=1 if pt−1=1 and r=2 otherwise.
Theorem 9. Let F=⋃i=1kPpi be a linear forest, where k≥2 and pi≥2 for 1≤i≤k. We have
[TABLE]
where ϵ=1 if all pi are odd and ϵ=2 otherwise.
We get the approximate value of the anti-Ramsey number for linear forests by Theorem 9 and it would be interesting to determine the exact value. Bialostocki, Gilboa and Roditty [2] and independently Gorgol and Görlich [10] showed that ar(Kn,2P3)=max{n,7} for n≥6. Gorgol and Görlich showed that ar(Kn,3P3)=2n−2 for n≥13. In Section 4, we will use Theorem 3 to compute the exact value of ar(Kn,2P4) for n≥8.
Theorem 10. For any n≥8, ar(Kn,2P4)=max{2n−2,16}.
Another motivation of this paper is the following conjecture of Gorgol and Görlich [10]:
Let G be a connected graph on n0≥3 vertices and t≥1, then for large n,
[TABLE]
if and only if G is a tree.
Statement (4) in Theorem 5(4) and Theorem 8 show that this conjecture is true for P3 and K1,p (p≥3), respectively. However, from Theorem 9, some simple calculation shows that this conjecture fails for Pl,l≥4.
Actually, for an arbitrary tree Tk with k edges, it is difficult to determine the (approximate) value of ar(Kn,Tk). Jiang and West [13] showed that for n≥2k,
[TABLE]
The upper bound comes from the well-known bound of ex(n,Tk)≤(k−1)n.
Erdős and Sós gave the following conjecture.
Conjecture 1.
[TABLE]
If Conjecture 1 is true (Ajtai, Komlós, Simonovits, Szemerédi announced it for large k), then the upper bound of ar(Kn,Tk) can also be reduced to 2k−1n. Also, Jiang and West [13] conjectured that:
Conjecture 2.
[TABLE]
Notice that if Tk is a star or a path of even length, then ar(n,Tk)=2k−2n+O(1).
The double star Sp,q, where p≥q≥1, is the graph consisting of the union of two stars K1,p and K1,q together with an edge joining their centers. In Section 5, we compute the anti-Ramsey number of double stars.
Theorem 11. For p≥2,1≤q≤p and n≥6(p2+2p), we have
[TABLE]
Notice that if we take Tk=Sp,p−1, then we have ar(Kn,Tk)=⌊2k−2⌋2n+O(1).
This paper is organized as follows. In Section 2, we give the proof of Theorem 8. The proof of Theorem 9 will be given in Section 3. The proof of Theorem 10 will be given in Section 4. The proof of Theorem 11 will be given in Section 5. Finally we will give a conjecture in Section 6.
Notation: Given an edge-coloring c of G, we denote the color of an edge uv by c(uv). We denote the number of colors by ∣c∣. For any v∈V(G), let C(v):={c(vw)∣w∈NG(v)} and dc(v):=∣C(v)∣. Let H be a subgraph of G. Denote C(H)={c(uv)∣uv∈E(H)}.
A color a is stared (at x) if all the edges with color a induce a star K1,r (centered at the vertex x). We let dc(v)=∣{a∈C(v)∣a is stared at v}∣.
A representing subgraph in an edge-coloring of Kn is a spanning subgraph
containing exactly one edge of each color. In the rest of this paper, we will use V to denote the vertex set of Kn for short.
2. Star forests
In this Section, we use the idea of [9] to prove Theorem 8.
Theorem 8. Let F=⋃i=1tK1,pi be a star forest, where p1≥3,p1≥p2≥⋯≥pt≥1. Let s=max{i:pi≥2,1≤i≤t}. For n≥3t2(p1+1)2,
[TABLE]
where r=1 if pt−1=1 and r=2 otherwise.
Proof.
For 1≤i≤s, we color Kn as follows. We color Ki−1+Kn−i+1 rainbow and color Kn−i+1 with new ar(Kn−i+1,K1,pi) colors without producing a rainbow copy of K1,pi. In such way, we use exactly (i-1)(n-i+1)+\binom{i-1}{2}+ar(K_{n-i+1},K_{1,p_{i}})=(i-1)n-\binom{i}{2}+\big{\lfloor}\frac{p_{i}-2}{2}(n-i+1)\big{\rfloor}+1 colors. Since any rainbow K1,pj (1≤j≤i) contains at least one vertex of V(Ki−1), we do not obtain any rainbow ⋃j=1iK1,pj. Also, we do not obtain any rainbow F.
For another lower bound, it is enough to consider the case t≥3. If pt−1=1, then
F⊃tK2. By Theorem 4, we have
[TABLE]
If pt−1≥2, we have F⊃(t−1)K1,2∪P2. By Theorem 5(6), we have
[TABLE]
Now we consider the upper bound. If t=1, then the result holds obviously by Theorem 1. Suppose t≥2.
Let
[TABLE]
By Theorems 4 and 5 (statements (1), (4), (6), (7)), we have
[TABLE]
Let c be any edge-coloring of Kn using
f(n,F)+1 colors. We will find a rainbow F by considering the following two cases.
Case 1. There is some vertex v0∈V such that dc(v0)≥∑i=1tpi+t.
Choose some color c0. Consider an edge-coloring c′ of Kn−v0: c′(e)=c(e) if c(e)∈/C(v0); else c′(e)=c0, where e∈E(Kn−v0).
Claim 1. There is a rainbow copy of F−K1,p1 in Kn−v0 with respect to c′.
Proof of Claim 1. We consider the following two subcases.
Subcase 1.1 p2≤2.
In this subcase, we have F−K1,p1=(s−1)K1,2∪(t−s)K1,1 and
[TABLE]
If F−K1,p1=K1,1, then the result holds obviously. Suppose F−K1,p1=K1,1. Then
[TABLE]
by (*). Thus there is a rainbow F−K1,p1 in Kn−v0 with respect to c′.
Subcase 1.2 p2≥3.
In this subcase, we have
[TABLE]
By induction hypothesis, there is a rainbow F−K1,p1 in Kn−v0 with respect to c′.
By Claim 1, there is a rainbow F−K1,p1 in Kn−v0 with respect to c′ (also respect to c).
Since dc(v0)≥∑i=1tpi+t, we are surely left with at least p1 edges, say v0w1,v0w2,…,v0wp1, such that w1,w2,…,wp1∈/V(F−K1,p1) and c(v0w1),c(v0w2),…,c(v0wp1)∈/{c(e)∣e∈E(F−K1,p1)}. By adding such p1 edges to F−K1,p1, we get a rainbow F.
Case 2. dc(v)≤∑i=1tpi+t−1 for all v∈V.
By induction hypothesis, Kn clearly contains a rainbow F−K1,pt.
Assume, by contradiction, that Kn does not contain a rainbow F.
Let G be a representing subgraph of Kn such that E(F−K1,pt)⊆E(G). Then we have ∣E(G)∣=f(n,F)+1 and dG(v)≤∑i=1tpi+t−1 for all v∈V.
Let W=V−V(F−K1,pt). Since dG(v)≤∑i=1tpi+t−1 for all v∈V(F−K1,pt) and G[W] does not contain a copy of K1,pt, we have
[TABLE]
We will finish the proof by considering the following two subcases.
Subcase 2.1 pt=1.
Since p1≥3 and pt=1, we have
[TABLE]
a contradiction with n≥3t2(p1+1)2.
Subcase 2.2 pt≥2.
In this subcase, we have s=t and
[TABLE]
Thus we have n≤2t−32[(∑i=1tpi+t−1)2+(t−1)(pt+t−2)],
a contradiction with n≥3t2(p1+1)2.
∎
3. Linear forests
First, we have the lower bound of the anti-Ramsey number for linear forests.
Proposition 1. Let F be a linear forest with components of order p1,p2,…,pk, where k≥2 and pi≥2 for 1≤i≤k. Let n≥∑i=1kpi. Then we have
[TABLE]
where s=∑i=1k⌊2pi⌋−ϵ; ϵ=1 if all pi are odd and ϵ=2 otherwise; r=2 if exactly one pi is even and r=1 otherwise.
Proof.
For the first lower bound, we choose a subgraph K∑i=1kpi−2 and color it rainbow. Then we use one extra color to color the remaining edges. In this way, we use exactly (2∑i=1kpi−2)+1 colors and do not obtain a rainbow F.
For the second lower bound, we color Ks+Kn−s rainbow and color the edges of Kn−s with r new colors. Every copy of F in Kn have at least (r+1) edges in Kn−s. In this way we do not obtain a rainbow F and use exactly s(n−s)+(2s)+r=sn−(2s+1)+r colors.
∎
If all the components of the linear forest are even paths or odd paths, we can get the following corollary from Theorems 2 and 7.
Corollary 1. Let F be a linear forest with components of order p1,p2,…,pk, where k≥2 and pi≥2 for 1≤i≤k and n sufficiently large. Let s=∑i=1k⌊2pi⌋−2. If all pi are even, we have
[TABLE]
If all pi are odd, we have
[TABLE]
Proof.
The lower bound is due to Proposition 1.
For the upper bound, when all pi are even, by Theorem 2,
[TABLE]
When all pi are odd, if p1=p2=⋯=pk=3, by Theorem 5(4), ar(Kn,F)=ar(Kn,kP3)=(k−1)n−(2k)+1. If at least one pi is not 3, by Theorem 7,
[TABLE]
∎
It is enough to consider the linear forests with at least one even path.
Theorem 12. Let F be a linear forest with components of order p1,p2,…,pk, where k≥1, pi≥2 for 1≤i≤k and at least one pi is even. Then
[TABLE]
Proof.
By Proposition 1, we just need to show the upper bound.
We will use the idea of [20] to prove it. The following results of Erdős and Gallai [5] will be used in our proof.
[TABLE]
[TABLE]
Since we can regard the union of even paths as the subgraph of one long even path (see Corollary 1), we just need to prove the upper bound is correct for linear forest with exact one even path and some odd paths. So we assume that F=P2s∪P2t1+1∪P2t2+1∪⋯∪P2tk+1, where s≥1,k≥1 and tk≥tk−1≥⋯≥t1≥1. Let s+t1+t2+⋯+tk=m. In this proof, we just consider the case s≥6. The case s<6 can be proved by the similar arguments but need to distinguish more cases as in [20].
Consider an edge-coloring of Kn with ar(Kn,F) colors such that there is no rainbow F. First we take a rainbow path Pl=u1u2…ul with maximum length. If l≥2m+k, we can get a rainbow F, a contradiction. By Proposition 1 and Theorem 2, ar(Kn,F)≥ar(Kn,P2m−1)+1 for large n, which implies there is a rainbow P2m−1. Hence we assume that 2m−1≤l≤2m+k−1. Take a representing subgraph G of Kn such that Pl⊂G. Then ∣E(G)∣=ar(Kn,F). We would partition V∖V(Pl) into three sets U1,U2 and U3 as follows:
U_{1}$$U_{2}$$U_{3}$$P_{l}$$u_{1}$$u_{2}$$u_{l-1}$$u_{l}
U1 is the subset of vertices of V∖V(Pl) which are not jointed to Pl at all: neither by edges nor by paths;
U2 is the set of isolated vertices of V∖V(Pl) which are jointed to Pl by edges;
U3=V∖(V(Pl)∪U1∪U2).
Claim 1. ∣E(G[U1])∣≤(m−2)∣U1∣.
Proof of Claim 1
We first prove that there is an P2s∪P2t1+1∪…∪P2tk+1 in P2s+2t1+…+2tk−1∪P2s+2t1+…+2tk−2 by induction on k. The base case k=1 is correct since P2s∪P2t1+1⊂P2s+2t−1∪P2s+2t1−2. Suppose the statement holds for k−1. We divide P2s+2t1+…+2tk−1 and P2s+2t1+…+2tk−2 respectively into two parts P2s+2t1+…+2tk−1−1,P2tk and P2s+2t1+…+2tk−1−2,P2tk.
We can find an P2s∪P2t1+1∪…∪P2tk−1+1 in P2s+2t1+…+2tk−1−1∪P2s+2t1+…+2tk−1−2 by induction hypothesis. Since (2s+2t1+…+2tk−1−1)+(2s+2t1+…+2tk−1−2)>2s+(2t1+1)+…+(2tk−1+1), there is at least one vertex of either P2s+2t1+…+2tk−1−1 or P2s+2t1+…+2tk−1−2 which is not used in P2s∪P2t1+1∪…∪P2tk−1+1. Hence, we can find an P2s∪P2t1+1∪…∪P2tk+1 in the original two long paths.
Since G contains no F, G[U1] contains no P2m−2 by the statement above.
Thus ∣E(G[U1])∣≤(m−2)∣U1∣ by (a).
Claim 2. ∣{v∈U2:dG(v)≥m−1}∣≤(m+k)(m−1l−2).
Proof of Claim 2
It is obvious that NG(v)⊂V(Pl)∖{u1,ul} for all v∈U2.
Suppose ∣{v∈U2:dG(v)≥m−1}∣≥(m+k)(m−1l−2)+1. Note that there are (m−1l−2) subsets of order m−1 in V(Pl)∖{u1,ul}. By Pigeonhole Principle, there are A⊆V(Pl)∖{u1,ul} with ∣A∣=m−1 and B⊆U2 with ∣B∣=m+k+1 such that A⊆NG(v) for any v∈B. So there is a rainbow Km−1,m+k+1. By adding one edge whose endpoints are in the large part of Km−1,m+k+1 to Km−1,m+k+1, we can find a rainbow F, a contradiction.
The following claim 3 is Lemma 1 in [20]. We include the proof for the sake of completeness.
Claim 3. ∣E(G[U3])∣+∣EG(Pl,U3)∣≤(m−2)∣U3∣.
Proof of Claim 3
Take a component H of G[U3] and denote by r the length of the longest cycle of H. If H is a tree, let r=2. By (b), we have ∣E(H)∣≤2r(∣V(H)∣−1). For any v∈V(H), we can find an Pr in H starting from it. Hence, u1,…,ur and ul−r+1,…,ul cannot be joined to v. Otherwise, there is a rainbow Pl+1, a contradiction. For any three consecutive vertices {ui,ui+1,ui+2}, there is no two independent edges in EG({ui,ui+1,ui+2},V(H)) by the maximality of Pl. Hence, we have
[TABLE]
Adding all the components of G[U3] up, we get ∣E(G[U3])∣+∣EG(Pl,U3)∣≤3l+1∣U3∣. Note that l≤2m+k−1. Since s≥6, m≥s+k≥6+k which implies ∣E(G[U3])∣+∣EG(Pl,U3)∣≤(m−2)∣U3∣
By Claims 1, 2 and 3, we have
[TABLE]
∎
By Corollary 1 and Theorem 12, we can get Theorem 9.
4. The exact value of ar(Kn,2P4)
In this Section, we will prove Theorem 10. We denote the complete graph on n vertices minus one edge by Kn−.
The following fact is trivial.
Fact 1. Let n≥8. If there is an edge coloring of Kn using 17 colors such that there is a rainbow K6 or K6−, then there is a rainbow 2P4.
We first prove the following lemma.
Lemma 1. ar(K8,2P4)=16.
Proof.
By Proposition 1 in Section 3, we just need to show the upper bound. Consider an 17-edge-coloring c of K8. Suppose there is no rainbow 2P4 in K8. By Theorem 3, there must be a rainbow Ck∪Cl. Assume that k≤l. Then k=3. Let T1=C3=x1x2x3x1 and T2=Cl=y1…yly1. We choose a representing subgraph G such that G⊃T1∪T2. We just need to consider the following three cases.
Case 1. l=5.
We claim that c(xy)∈C(T1∪T2) for all x∈V(T1) and y∈V(T2). Otherwise, say c(x1y1)∈/C(T1∪T2), then x2x3x1y1∪y2y3y4y5 is a rainbow 2P4, a contradiction. Then the total number of colors is at most 3+(25)=13<17, a contradiction.
Case 2. l=4.
Let the remaining vertex be z. We have c(x1z)=c(x2x3); otherwise there must be a rainbow 2P4 in T1∪T2∪{x1z}. Similarly, we have c(x2z)=c(x1x3) and c(x3z)=c(x1x2). Now we have four rainbow C3’s: C31=x1x2x3x1, C32=zx2x3z, C33=zx1x2z and C34=zx1x3x1z. If there are u∈{z,x1,x2,x3}, say u=z, and 1≤j≤4, say j=1, such that c(zy1)=c(zy2) and c(zy1),c(zy2)∈/C(C31∪T2) then we have a rainbow C31 and a rainbow C5=y1zy2y3y4y1 and the situation is the same as Case 1. Hence we have ∣EG(u,T2)∣≤2 for any u∈{z,x1,x2,x3}. Then 17=∣E(G)∣≤3+4×2+(24)=17, which means G[V(T2)]=K4 and ∣EG(u,T2)∣=2 for any u∈{z,x1,x2,x3}. Thus we can find a rainbow 2P4, a contradiction.
Case 3. l=3.
Let the remaining vertices be z1,z2. By Case 2 and G containing no rainbow 2P4, we have ∣EG({z1,z2},T1∪T2)∣≤2 and then
∣E(G[V(T1∪T2)])∣≥17−2−1=14. Thus
G[V(T1∪T2)]≅K6 or K6− and we can get a rainbow 2P4 by Fact 1, a contradiction.∎
Now we will complete the proof of Theorem 10.
Theorem 10. For any n≥8, ar(Kn,2P4)=max{2n−2,16}.
Proof.
By Proposition 1, we just need to show the upper bound. We will prove it by induction on n.
Consider an (2n−1)-edge-coloring of Kn for n≥9. Suppose there is no rainbow 2P4. By Theorem 3, there is a rainbow Ck∪Cl. Assume that k≤l. Then k=3. Let T1=C3=x1x2x3x1 and T2=Cl=y1…yly1.
We finish the proof by considering the following four cases.
Case 1. 4≤l≤n−5.
In this case, there are at least two vertices z1,z2∈/V(T1∪T2).
Since there is no rainbow 2P4, we have c(x1zi)=c(x2x3), c(x2zi)=c(x1x3) and c(x3zi)=c(x1x2) for i=1,2. But we can have a rainbow 2P4 whatever the color of z1z2 is, a contradiction.
Case 2. l=n−4.
Let V(Kn)∖V(T1∪T2)={z}.
Since there is no rainbow 2P4, we have that c(x1z)=c(x2x3), c(x2z)=c(x1x3), c(x3z)=c(x1x2), c(xy)∈C(T1∪T2) for all x∈V(T1)∪{z} and y∈V(T2). By Case 1, we can assume that c(yiyj)∈C(T1∪T2) for all yi,yj∈V(T2). Hence the total number of colors is at most 3+n−4=n−1<2n−1, a contradiction.
Case 3. l=n−3.
Since there is no rainbow 2P4, we have that c(xy)∈C(T1∪T2) for all x∈V(T1) and y∈V(T2). By Case 1 and Case 2, we can assume that c(yiyj)∈C(T1∪T2) for all yi,yj∈V(T2). Hence the total number of colors is at most 3+n−3=n<2n−1, a contradiction.
Case 4. l=3.
In this case we will consider the following two subcases. We choose a representing subgraph G such that G⊃T1∪T2.
Subcase 4.1 n=9.
If there is a vertex v0 such that dc(v0)=0, then K9−v0 uses exactly 17 colors and there is a rainbow 2P4 in K9−v0 by Lemma 1. Hence we have dc(v)≥1, for all v∈V(K9). Thus δ(G)≥1.
Let Z=V(K9)∖V(T1∪T2)={z1,z2,z3}. We claim that ∣EG(zj,Ts)∣≤1 for any 1≤j≤3 and 1≤s≤2. Otherwise G contains an C3∪C4 and the situation is the same as Case 1. If there are zi∈Z and s∈{1,2} such that ∣EG(zi,Ts)∣=1, then ∣EG(z,Tt)∣=0 for any z∈Z∖{zi} and t∈{1,2}∖{s} by G having no 2P4. If EG(Z,T1∪T2)=∅, then ∣E(G[V(T1∪T2)])∣=17−∣E(G[Z])∣≥17−3=14 which implies G[V(T1∪T2)]≅K6 or K6− and there is a rainbow 2P4 by Fact 1, a contradiction. Hence we have EG(Z,T1∪T2)=∅. Assume z1x1∈E(G) and we will consider the following two subcases.
Subcase 4.1.1 EG(Z,T2)=∅.
Recall z1x1∈E(G) and ∣EG(zj,T1)∣≤1 for any 1≤j≤3. If there is z∈{z2,z3} such that z1z∈E(G), then EG({x2,x3},T2) =∅ which implies ∣E(G)∣≤15, a contradiction. Hence we have ∣E(G[Z])∣≤1 and then ∣E(G[V(T1∪T2)])∣≥17−3=14. Thus G[V(T1∪T2)]≅K6 or K6− and there is a rainbow 2P4 by Fact 1, a contradiction.
Subcase 4.1.2 EG(Z,T2)=∅.
In this case, we have ∣EG(Z,T1∪T2)∣=∣EG(z1,T1∪T2)∣=2 and assume z1y1∈E(G). If EG(z1,{z2,z3})=∅, say z1z2∈E(G), then EG(T1,T2)⊆{x1y1}. Hence ∣E(G)∣≤12, a contradiction. So we have EG(z1,{z2,z3})=∅, which implies z2z3∈E(G) by δ(G)≥1. Thus
∣E(G[V(T1∪T2)])∣=17−3=14
and G[V(T1∪T2)]≅K6−. We can get a rainbow 2P4 by Fact 1, a contradiction.
Subcase 4.2 n≥10.
If there is a vertex v0 such that dc(v0)≤2, then Kn−v0 uses 2n−1−dc(v0)≥2n−1−2=2(n−1)−1 colors and we get that Kn−v0 contains a rainbow 2P4 by induction hypothesis, a contradiction. So we assume that dc(v)≥3 for all v∈V(Kn). Thus δ(G)≥3.
Let Z=V(Kn)∖V(T1∪T2)={z1,z2,…,zn−6}. Since G contains no rainbow 2P4, we have that if there are z∈Z and s∈{1,2} such that EG(z,Ts)=∅, then EG(z′,Tt)=∅ for any z′∈Z∖{z} and t∈{1,2}∖{s}.
Subcase 4.2.1 EG(Z,T1)=∅ and EG(Z,T2)=∅.
In this case, there is exactly one vertex in Z, say z1, such that EG(z1,T1),EG(z1,T2)=∅ and EG(Z∖{z1},T1∪T2)=∅. Since dG(v)≥3, G[T1∪T2∪{z1}] contains a cycle with length at least 4 and G[Z∖{z1}] contains a cycle of length at least 3. Then we can have a contradiction by Cases 1 to 3.
Subcase 4.2.2 EG(Z,T1)=∅ or EG(Z,T2)=∅.
Assume that EG(Z,T1)=∅. By Case 1, we can assume that ∣EG(z,T2)∣≤1 for any z∈Z. Then δ(G[Z])≥2 and there is a cycle in G[Z]. Since dG(xi)≥3, we have ∣EG(xi,T2)∣≥1 for any 1≤i≤3 and there is a cycle with length at least 4 in G[T1∪T2]. Then we can have a contradiction by Cases 1 to 3.
∎
5. Double stars
The following Lemma 2 is an extension of Theorem 1. The idea of the proof is the same as the idea used in [12]. We include the proof for the sake of completeness.
Lemma 2. Let V(Kn)={v1,v2,…,vn}. For 1≤p1≤p2≤⋯≤pn≤3n, the maximum number of colors of an edge-coloring of Kn such that dc(vi)≤pi for any 1≤i≤n is at most 2∑i=1npi−n+1.
Proof.
Let c be an edge-coloring of Kn such that dc(vi)≤pi for any 1≤i≤n. Let G be a representing subgraph of Kn. Then dG(vi)≤dc(vi)≤pi for 1≤i≤n and ∣E(G)∣=∣c∣.
Write V=V(G)=V(Kn). Let S={vi∈V:dG(vi)=pi}. If S=∅, then dG(vi)≤pi−1 for 1≤i≤n and hence
[TABLE]
and we are done.
Hence we may assume that S=∅. For v∈V, let CG(v) be the set of colors used on the edges incident to v in G. Clearly, we have CG(v)⊂C(v) for all v∈V and CG(v)=C(v) for v∈S. Particularly, {c(uv)}=C(u)∩C(v)=CG(u)∩CG(v) for u,v∈S.
Claim 1. G[S] is a clique.
Proof of Claim 1. Let u,v∈S. Since c(uv)∈C(u)∩C(v)=CG(u)∩CG(v), uv∈E(G). Hence G[S] is a clique.
Claim 2. Let u,v∈S and w∈V−S. If c(uw)=c(vw), then c(uw)=c(vw)=c(uv).
Proof of Claim 2. Since c(uw)=c(vw)∈C(u)∩C(v)=CG(u)∩CG(v)={c(uv)}, we have c(uw)=c(vw)=c(uv).
Claim 3. Let u∈S,v∈/S and uv∈/E(G). Then c(uv)∈/CG(v).
Proof of Claim 3. Suppose c(uv)∈CG(v). Since u∈S, we have c(uv)∈CG(u). Hence c(uv)∈CG(u)∩CG(v) which implies uv∈E(G), a contradiction.
Claim 4. For all vi∈/S, dG(vi)≤pi−2∣S∖NG(vi)∣.
Proof of Claim 4. Let S∖NG(vi)={u1,u2,…,uk}. By Claim 3, c(ujvi)∈/CG(vi) for all 1≤j≤k. Furthermore, by Claim 2, if c(ujvi)=c(ulvi), then c(ujvi)=c(ulvi)=c(ujul). This implies that no three edges in the set {u1vi,u2vi,…,ukvi} have the same color; otherwise, by Claim 1, G[S] would contain a monochromatic triangle, a contradiction. Hence at least 2k distinct colors are used on the edges viu1,viu2,⋯,viuk, and those colors are not in CG(vi). So ∣C(vi)∣≥∣CG(vi)∣+2k=dG(vi)+2k. Since ∣C(vi)∣=dc(vi)≤pi, we have dG(vi)+2k≤pi, which yields dG(vi)≤pi−2k=pi−2∣S∖NG(vi)∣.
By Claim 4, we have
[TABLE]
Notice that ∑vi∈/S∣S∖NG(vi)∣ counts exactly the number of non-edges in G between S and V∖S. We have ∑vi∈/S∣S∖NG(vi)∣=∑vi∈S(n−1−pi). Hence,
[TABLE]
On the other hand, for vi∈/S,dG(vi)≤pi−1. Hence, we have
[TABLE]
We have
[TABLE]
Since n≥3pn, we have min{∣S∣,n−2∑vi∈S(n−1−pi)}≤2. Therefore,
[TABLE]
∎
Now we will prove Theorem 11.
Theorem 11. For p≥2,1≤q≤p and n≥6(p2+2p),
[TABLE]
Proof.
If 1≤q≤p−1, we have
[TABLE]
When q=p, we color the edges of Kn as follows. We color K1,n−1 rainbow and color Kn−1 with new ar(Kn−1,K1,p) colors without producing a rainbow K1,p. In such way, we use exactly
[TABLE]
colors and do not obtain a rainbow Sp,p.
Now we consider the upper bound. Let
[TABLE]
Let c be any edge-coloring of Kn using g(n,p,q)+1 colors. We will find a rainbow Sp,q by considering the following two cases.
Case 1. There is some vertex v0 such that dc(v0)≥p+2q+1.
Let U={v∈V∖{v0}: the color c(v0v) is stared at v0}.
Consider the induced edge-coloring c′ of c on Kn−v0. We have
[TABLE]
Claim 1. Kn−v0 contains a rainbow K1,q with center in U or a rainbow K1,q+1 with center in V(Kn−v0)∖U with respect to c′ (also c).
Proof of Claim 1. Suppose dc′(u)≤q−1 for all u∈U and dc′(v)≤q for all v∈V(Kn−v0)∖U, by Lemma 2, we have
[TABLE]
While ∣c′∣≥g(n,p,q)+1−∣U∣,
we can get ∣U∣≥n, a contradiction.
By Claim 1, we get a rainbow K1,q in Kn−v0 with respect to c whose center is u0 such that the color c(v0u0) does not present in c(K1,q). The vertex v0 is the endpoint of at least p+2q+1 edges with distinct colors (with respect to c). Since dc(v0)≥p+2q+1, there are at least p edges, say v0v1,v0v2,…,v0vp, such that vi∈/V(K1,q) and c(v0vi)∈/C(K1,q)∪{c(v0u0)} for all i=1,…,p. So we can get a rainbow Sp,q.
Case 2. dc(v)≤p+2q for all v∈V.
Since g(n,p,q)+1≥ar(Kn,K1,p+1)+1, we can find a rainbow K1,p+1. Choose some color c0 which is not in C(K1,p+1). Consider an edge-coloring c′′ of Kn−V(K1,p+1): for e∈E(Kn−V(K1,p+1)), if c(e)∈/C(K1,p+1), then c′′(e)=c(e); else c′′(e)=c0. Then
[TABLE]
So there is a rainbow K1,q+1 in Kn−K1,p+1 with respect to c′′ (also c). K1,q+1 has at most one edge e with color c0. By joining the centers K1,p+1 and K1,q+1 and deleting one edge of K1,p+1 and K1,q+1 respectively, we can find a rainbow Sp,q.
∎
6. Open problems
A spider is a tree with at most one vertex of degree more than 2, called the center of the spider (if no vertex of degree more than two, then any vertex can be the center). A leg of a spider is a path from the center to a vertex of degree 1. Thus, a star with p edges is a spider of p legs, each of length 1, and a path is a spider of 1 or 2 legs.
The number of edges in a maximum matching of a graph G is called the matching number of G and denoted by ν(G).
Observation 1. Let p≥2 and T be a spider of p legs, each of length at least 2. We have
[TABLE]
the equality holds if and only if T has exactly one leg with length even.
Let p≥2 and T be a spider of p legs, each of length at least 2. Let β(T)=min{ν(T−e):e∈E(T)}.
Proposition 2. Let p≥2 and T be a spider of p legs, each of length at least 2. We have
[TABLE]
where r=2 if there is exactly one leg of T with length even and r=1 otherwise.
Proof.
Let β=β(T). We take an Kβ−1+Kn−β+1 and color it rainbow, and use r extra colors for all the remaining edges. Suppose there is a rainbow T in this coloring. Then
T−e contains a matching of size β for any e∈E(T) (or T−e1−e2 contains a matching of size β for any e1,e2∈E(T) if T has exactly one leg with length even). But Kβ−1+Kn−β+1 does not contain a matching of size β, a contradiction.
∎
If we regard Pk+1 as a spider of 2 legs, the lower bound of Proposition 2 is sharp for p=2 and large n by Theorem 2.
We conjecture that the lower bound is sharp for p≥3 and large n.
Conjecture 3.
Let p≥3 and T be a spider of p legs, each of length at least 2. For large n,
[TABLE]
where r=2 if there is exactly one leg of T with length even and r=1 otherwise.