On Hamiltonian cycles in balanced k-partite graphs
Louis DeBiasio and Nicholas Spanier11footnotemark: 1
Department of Mathematics; Miami University; Oxford, OH. [email protected], [email protected] supported in part by Simons Foundation Collaboration Grant # 283194.
Abstract
For all integers k with kβ₯2, if G is a balanced k-partite graph on nβ₯3 vertices with minimum
degree at least
[TABLE]
then G has a Hamiltonian cycle unless k=2 and 4 divides n, or k=2nβ and 4 divides n. In the case where k=2 and 4 divides n, or k=2nβ and 4 divides n, we can characterize the graphs which do not have a Hamiltonian cycle and see that β2nββ+β2β2k+1ββn+2βββknβ+1 suffices. This result is tight for all kβ₯2 and nβ₯3 divisible by k.
1 Introduction
The study of Hamiltonian cycles in balanced k-partite graphs begins with the following classic results of Dirac, and Moon and Moser. Dirac [5] proved that for all graphs G on nβ₯3 vertices, if Ξ΄(G)β₯β2nββ, then G has a Hamiltonian cycle. Moon and Moser [7] proved that for all balanced bipartite graphs G on nβ₯4 vertices, if Ξ΄(G)β₯4n+2β, then G has a Hamiltonian cycle.
Over 30 years later Chen, Faudree, Gould, Jacobson, and Lesniak [2] beautifully tied these results together by proving that for all kβ₯2, if G is a balanced k-partite graph on n vertices with
[TABLE]
then G has a Hamiltonian cycle. It turns out that while their result is nearly optimal, in most cases the degree condition can be improved by 1. The purpose of this note is simply to provide the precise minimum degree condition in all cases thereby filling the lacuna in the above result (and as we point out in the Appendix, it is unfortunately not as simple as replacing the strict inequality in (1) with a weak inequality).
Theorem 1.1**.**
Let k be an integer with kβ₯2. For all balanced k-partite graphs G on n vertices, if
[TABLE]
then G has a Hamiltonian cycle unless k=2 and 4 divides n, or k=2nβ and 4 divides n.
Since a graph on n vertices can be viewed as a k-partite graph with k=n, note that when k=n, we have β2nββ+β2β2k+1ββn+2βββknβ=β2nββ and thus Theorem 1.1 reduces to Diracβs theorem. When k=2, we have β2nββ+β2β2k+1ββn+2βββknβ=β4n+2ββ and thus when 4 does not divide n, Theorem 1.1 reduces to Moon and Moserβs theorem; and when 4 does divide n, Ferrara, Jacobson, and Powell [6] characterized all balanced bipartite graphs G on nβ₯4 vertices such that Ξ΄(G)β₯4nβ, yet G does not have a Hamiltonian cycle. So our proof will only handle the cases when 3β€kβ€2nβ.
We will also prove the following which will handle the case when k=2nβ and 4 divides n. Together with the results in [6], this gives a complete characterization of balanced k-partite graphs G on n vertices which satisfy Ξ΄(G)β₯β2nββ+β2β2k+1ββn+2βββknβ, but do not have a Hamiltonian cycle.
Proposition 1.2**.**
Let nβ₯8 be divisible by 4, let k=2nβ, and let G be a balanced k-partite graph on n vertices. If Ξ΄(G)β₯2nβ+βk+2n+2βββknβ=2nββ1 and G does not have a Hamiltonian cycle, then G belongs to one of the families of examples described in Example 2.2.
1.1 Overview
We give the lower bound examples in Section 2, we collect the main lemmas in Section 2 (while it is a combination of existing results, Lemma 3.3 may be of independent interest), we deal with the first two exceptions in Section 4 before starting the main proof in Section 5. Finally, in the Appendix, we collect some numerical lemmas which are needed because of the floors and ceilings appearing in (2).
This project grew out of an earlier work of the first author together with Krueger, Pritikin, and Thompson [4], where we considered Hamiltonian cycles in unbalanced k-partite graphs for kβ₯3. The upcoming Example 2.1 first appeared in a more general form in [4]. In fact, it was this example which indicated to us that (1) is not always tight. By using Theorem 3.2 and Lemma 3.3, we were able to streamline the original proof of Chen et al.Β with the correct degree condition; however, because of the unexpected (to us) exceptional cases which arose when k=2nβ, our overall proof didnβt end up being any shorter than the original. Again, we emphasize that Chen et al.Β have a beautiful result which places Diracβs theorem and Moon and Moserβs theorem on a common spectrum. It is only because of the fundamental nature of these results that we have expended the effort necessary to provide the tight degree condition in all cases.
1.2 Notation
For SβV(G), we let N(S)=βvβN(v) and S=V(G)βS.
Given disjoint sets A,BβV(G), we let Ξ΄(A,B)=min{β£N(v)β©Bβ£:vβA}β£.
Given a cycle v1βv2ββ¦vkβv1β, iβ[k], and an integer t, we assume that the addition in the indices, such as vi+tβ, is taken modulo k.
2 Tightness examples
Example 2.1**.**
For all kβ₯2 and all n divisible by k, there exists a family F of balanced k-partite graphs on n vertices such that for all FβF,
[TABLE]
but F does not have a Hamiltonian cycle.
Proof.
First note that
[TABLE]
Since if k is even, then both sides of the equation equal βk+2n+2ββ; if k is odd and n is even, then both sides of the equation equal βk+1n+2ββ; and if k is odd and n is odd then we get that βk+1n+1ββ=βk+1n+2ββ, which is true since k+1n+2β is not an integer.
Let F be the family of graphs which can be obtained from a complete k-partite graph with parts V1β,β¦,Vkβ such that β£Viββ£=knβ for all iβ[k], by selecting some XiββViβ for all iβ[β2k+1ββ] such that β£X1ββ£β₯β―β₯β£Xβ2k+1ββββ£=ββ2k+1βββ2n+1ββββ and β£X1ββͺβ―βͺXβ2k+1ββββ£=β2n+1ββ. Add all edges between parts except for those between a vertex in Xiβ and Xjβ for all i,jβ[β2k+1ββ]. Note that every FβF has an independent set of size β2n+1ββ and thus does not contain a Hamiltonian cycle.
Finally to see that the degree condition is satisfied, let iβ[β2k+1ββ] and let vβXiβ. We have by (3)
[TABLE]
Example 2.2**.**
Let nβ₯8 be divisible by 4 and let k=2nβ.
- (i)
There exists a family F1β of balanced k-partite graphs on n vertices such that for all F1ββF1β, Ξ΄(F1β)β₯2nβ+βk+2n+2βββknβ=2nββ1, but ΞΊ(F1β)β€1 and thus F1β does not have a Hamiltonian cycle.
2. (ii)
There exists a 2-connected balanced 4-partite graph F2β on 8 vertices with Ξ±(F2β)=3 such that F2β does not have a Hamiltonian cycle.
3. (iii)
There exists a family F3β of balanced k-partite graphs on n vertices such that for all F3ββF3β, Ξ΄(F3β)β₯2nβ+βk+2n+2βββknβ=2nββ1, ΞΊ(F2β)β₯2, and Ξ±(F3β)=2nβ, but F3β does not have a Hamiltonian cycle.
Proof.
- (i)
Let Viβ={xiβ,yiβ} for all iβ[k]. Add all edges inside {x1β,β¦,xkβ}, add all edges from ykβ to {y1β,β¦,ykβ1β}, and for all iβ[kβ1] add at least kβ1 edges from yiβ to {y1β,β¦,yiβ1β,yi+1β,β¦,ykβ1β}βͺ{xkβ}. Let F1β be the family of graphs thus obtained. Note that every graph F1ββF1β has Ξ΄(F1β)=kβ1=2nββ1 and ΞΊ(F1β)β€1 and thus F1β does not have a Hamiltonian cycle.
2. (ii)
We let F2β be the graph in Figure 2 which can be seen to be a balanced 4-partite graph (with vertices of the same shape being in the same part of the partition) which is 2-connected and has Ξ±(F2β)=3. Note that F2β has no Hamiltonian cycle since Gβx1ββx4β has three components.
3. (iii)
Let the parts be labeled X1β,β¦,Xk/2β, Y1β,β¦,Yk/2β and let X=βͺi=1k/2βXiβ and Y=βͺi=1k/2βYiβ. Let yβ²βYk/2β, let yβ²β²βYβ{yβ²}, and let xβ²βX. Add all edges between X and Yβ{yβ²,yβ²β²}, all edges from yβ²β² to Xβ{xβ²}, and all edges from yβ² to {xβ²}βͺ(YβYk/2β). Furthermore, we may add any number of other edges between the parts Y1β,β¦,Yk/2β and we may add the edge xβ²yβ²β². Let F3β be the family of graphs thus obtained. Let F3ββF3β and let H be the bipartite graph induced by [X,Y]. It is easily seen that Ξ΄(F3β)β₯2nββ1, ΞΊ(F2β)β₯2, and Ξ±(F3β)=2nβ. Since X is an independent set, if F3β has a Hamiltonian cycle, it must be in H; however, since yβ² has degree 1 in H, there is no Hamiltonian cycle in H.
β
3 General lemmas
In this section we state three general results which are useful for finding Hamiltonian cycles, beginning with two classics.
Theorem 3.1** (Dirac [5]).**
Let nβ₯dβ₯3. If G is 2-connected and Ξ΄(G)β₯d/2, then G has a cycle of length at least d.
Theorem 3.2** (ChvΓ‘tal [3]).**
Let G=(U,V,E) be a bipartite graph on nβ₯4 vertices with vertex sets U={u1β,β¦,un/2β} and V={v1β,β¦,vn/2β}. If for all 1β€k<n/2,
[TABLE]
then G has a Hamiltonian cycle.
The main lemma which we use to begin the proof of Theorem 1.1 and Proposition 1.2 is the following combination of the well known result of Nash-Williams [8] and a (slight weakening of a) result of Bauer, Veldman, Morgana, Schmeichel [1]. We provide a proof for completeness.
We say that a cycle C in a graph G is strongly dominating if V(G)βV(C) is an independent set and no two vertices of βuβV(G)βV(C)βN(u) appear consecutively on C.
Lemma 3.3** (see [8, Lemmas 1,2,3,4] and [1, Lemma 8]).**
Let G be a graph on n vertices. If G is 2-connected and Ξ΄(G)β₯3n+2β, then every longest cycle of G is strongly dominating.
Proof.
Let C=v1βv2ββ¦vkβv1β be a longest cycle in G and let P=u1βu2ββ¦urβ be a longest path in GβC. If rβ€1, then we are done; so suppose rβ₯2. We have k+rβ€n and note that by Theorem 3.1, we have kβ₯2Ξ΄(G)β₯32n+4β and thus
[TABLE]
Let X=N(u1β)β©V(C) and Y=N(urβ)β©V(C) and note that
[TABLE]
The key observation is that by the maximality of C, no two vertices in XβͺY are consecutive along C, and furthermore if viββXβ©Y, then none of viβrβ,β¦,viβ1β,vi+1β,β¦,vi+rβ are in XβͺY.
First suppose that XβY or YβX; without loss of generality XβY. In this case we have by (5),
[TABLE]
First suppose r=2, in which case (6) becomes nβ2β₯3(3n+2ββ1)=nβ1, a contradiction. Now suppose rβ₯3 in which case (6) becomes
[TABLE]
contradicting (4).
Now suppose that XβYξ =β
and YβXξ =β
. There are vertices viβ,vjββV(C) with the following properties: viββXβY and the next vertex viβ²ββXβͺY which appears after viβ satisfies viβ²ββY (meaning that iβ²β₯i+r+1), and vjββYβX and the next vertex vjβ²ββXβͺY from XβͺY which appears after vjβ satisfies vjβ²ββX (meaning that jβ²β₯j+r+1). Each vertex of ((XβY)βͺ(YβX))β{viβ,vjβ} is followed by at least one vertex from V(C)β(XβͺY) and each vertex of (Xβ©Y)βͺ{viβ,vjβ} is followed by at least r vertices from V(C)β(XβͺY). So we have
[TABLE]
where the second to last inequality is seen by using (5) and splitting into cases whether β£Xβ©Yβ£=0 or not. However, 4(n+2)/3β2(rβ1)β€nβr implies nβ€3rβ14, contradicting (4).
To see that the second part of the definition of strongly dominating is satisfied, suppose that C=v1ββ¦vkβv1β is a longest cycle and suppose V(G)βV(C)={u1β,β¦,urβ} is an independent set. If β£V(G)βV(C)β£β€1, we are done, so suppose rβ₯2. Let X=N(u1β)β©V(C) and suppose (without loss of generality) for contradiction that v1ββX and v2ββN(u2β). By the maximality of C, this implies that v3βξ βN(u2β) and for all iβ₯3, if viββX, then vi+1β,vi+2βξ βN(u2β). Since kβ€nβ2, this implies that
[TABLE]
a contradiction.
β
4 Filtering out F1β and F2β
We want to say that every balanced k-partite graph satisfying (2) is both 2-connected and every longest cycle in strongly dominating; however, there are two exceptions and we deal with those exceptions before beginning the main proof in next Section.
First, we show that every balanced k-partite graph satisfying (2) is either 2-connected or belongs to the family F1β in Example 2.2.(i).
Lemma 4.1**.**
Let kβ₯3, let n be an integer such that nβ₯2k, and let G be a balanced k-partite graph on n vertices. If
[TABLE]
then G is 2-connected unless n=2k and GβF1β (see Example 2.2.(i)).
Proof.
Let V1β,β¦,Vkβ be the parts of G. Suppose for contradiction that ΞΊ(G)β€1. Let A,B,C be a partition of V(G) such that β£Cβ£β€1 and GβC is not connected.
First suppose that there exists iβ[k] such that ViββAβͺC or ViββBβͺC. Without loss of generality suppose ViββBβͺC and let uβA and vβBβ©Viβ. We have
[TABLE]
contradicting Fact 6.4.(i) when k is even and nβ₯3k, and contradicting Fact 6.4.(ii) when k is odd and nβ₯2k.
So unless k is even and n=2k, we must have that for all iβ[k], Viββ©Aξ =β
and Viββ©Bξ =β
. Either C=β
and we let uβA and vβB, or Cξ =β
and suppose without loss of generality that V1ββ©Cξ =β
in which case we let uβV1ββ©A and vβV1ββ©B. Either way we have
[TABLE]
contradicting Fact 6.4.(i) when k is even and nβ₯3k, and contradicting Fact 6.4.(ii) when k is odd and nβ₯2k.
Finally, suppose k is even and n=2k which implies Ξ΄(G)β₯2nββ1. For all uβA we have 2nββ1β€d(u)β€β£Aβ£+β£Cβ£β1 which implies β£Aβ£β₯2nβββ£Cβ£ and for all vβB we have 2nββ1β€d(v)β€β£Bβ£+β£Cβ£β1 which implies β£Bβ£β₯2nβββ£Cβ£. If C=β
, this implies β£Aβ£=2nβ and β£Bβ£=2nβ. If Cξ =β
, we have β£Aβ£β₯2nββ1 and β£Bβ£β₯2nββ1, so without loss of generality suppose β£Aβ£+β£Cβ£=2nβ and β£Bβ£=2nβ.
If there exists iβ[k] such that ViββAβͺC or ViββB; say ViββAβͺC, then for uβViββ©A, we have
[TABLE]
a contradiction. Thus for all iβ[k], we have β£Viββ©(AβͺC)β£=1 and β£Viββ©Bβ£=1. Let Viβ={xiβ,yiβ} for all iβ[k] and let X={x1β,β¦,xkβ} and Y={y1β,β¦,ykβ} and suppose X=AβͺC and Y=B. So it must be the case that every vertex uβA is adjacent to precisely the vertices in Xβ{u} which means G[X] is a clique. Also every vertex vβY is adjacent to at least 2nββ1 of the 2nβ vertices in (YβͺC)β{v}. Thus GβF1β.
β
We now prove a lemma which shows that when G is a 2-connected balanced k-partite graph satisfying (2), we either have that every longest cycle in G is strongly dominating or G is isomorphic to the graph F2β in Example 2.2.(ii).
Lemma 4.2**.**
Let kβ₯3 and let nβ₯2k and let G be a balanced k-partite graph on n vertices. If G is 2-connected and Ξ΄(G)β₯β2nββ+β2β2k+1ββn+2βββknβ, then every longest cycle of G is strongly dominating unless n=8 and Gβ
F2β (see Example 2.2.(ii)).
Proof.
We will show that, unless n=8 and k=4, we have Ξ΄(G)β₯β2nββ+β2β2k+1ββn+2βββknββ₯3n+2β and thus we are done by Lemma 3.3.
First suppose k is odd, in which case β2nββ+β2β2k+1ββn+2βββknβ=β2nββ+βk+1n+2βββknβ. First note that when k=3 and n=6 or n=9 we have by direct inspection that β2nββ+βk+1n+2βββknββ₯3n+2β. So in the remaining cases we have
[TABLE]
Thus, using Fact 6.2, we have
[TABLE]
as desired.
Now suppose k is even, in which case β2nββ+β2β2k+1ββn+2βββknβ=2nβ+βk+2n+2βββknβ. Note that aside from the case n=8 and k=4 we have
[TABLE]
Thus
[TABLE]
as desired.
Finally suppose n=8 and k=4 and let C be a longest cycle of G. Since G is 2-connected and Ξ΄(G)β₯28β+β610βββ48β=3, Theorem 3.1 implies that C has length at least 6. If C had length at least 7, it would be a strongly dominating cycle, so suppose C has length 6. Let C=x1βx2βx3βx4βx5βx6β and let xβ² and xβ²β² be the two vertices in V(G)βV(C). If xβ²xβ²β²ξ βE(G), then by the maximality of C it is easily seen that, without loss of generality, N(xβ²)=N(xβ²β²)={x1β,x3β,x5β} and thus C is strongly dominating; so suppose that xβ²xβ²β²βE(G).
Without loss of generality suppose xβ²x1ββE(G). If either xβ²β²x2ββE(G) or xβ²β²x6ββE(G), then G has a Hamiltonian cycle; and if either xβ²β²x3ββE(G) or xβ²β²x5ββE(G), then G has a cycle longer than C, a contradiction. Since Ξ΄(G)β₯3, this forces xβ²β²x1β,xβ²β²x4ββE(G). By the same argument we get xβ²x4ββE(G). If x6βx3ββE(G), then x6βx3βx2βx1βxβ²xβ²β²x4βx5βx6β is a Hamiltonian cycle, so x6βx3βξ βE(G), and by symmetry x2βx5βξ βE(G). If x6βx2ββE(G), then x6βx2βx3βx4βxβ²xβ²β²x1βx6β is a cycle longer than C, a contradiction. So x6βx2βξ βE(G) and by symmetry x3βx5βξ βE(G). Since Ξ΄(G)β₯3, this forces x6βx4β,x5βx1β,x2βx4β,x3βx1ββE(G). Therefore Gβ
F2β.
β
5 Proof of Theorem 1.1 and Proposition 1.2
Let kβ₯3 and let G be a balanced k-partite graph on n vertices. Let V1β,V2β,β¦,Vkβ denote the parts and note that β£Viββ£=knβ=:m for all iβ[k]. Since the case k=n is Diracβs theorem, we suppose kβ€2nβ and since the case k=2 is handled in [7] and [6], we suppose kβ₯3. Furthermore, if k=2nβ, we suppose that Gξ βF1β and Gξ β
F2β (see Example 2.2). Now let C be a maximum length cycle and suppose for contradiction that C is not Hamiltonian. By Lemma 4.1 and Lemma 4.2 we may assume that C is strongly dominating.
Without loss of generality, let
zβV1ββV(C).
Let S=(V(G)βV(C))βͺ{vi+1β:viββN(z)} and R=(V(G)βV(C))βͺ{viβ1β:viββN(z)} and note that
[TABLE]
Since C is strongly dominating, both S and R are independent sets.
For each iβ[k], set
[TABLE]
Define β=β£{iβ[k]:Siβξ =β
}β£ and ββ²=β£{iβ[k]:Riβξ =β
}β£ and without loss of generality suppose
[TABLE]
Furthermore, without loss of generality, we may suppose that
[TABLE]
Claim 5.1**.**
β,ββ²β₯β2kββ**
Proof.
We claim that β£Rβ£,β£Sβ£β₯Ξ΄(G)+1>(β2kβββ1)knβ, which implies the result. Indeed, we have
[TABLE]
When k is even, (10) reduces to βk+2n+2ββ+1>0, and when k is odd, by Fact 6.2, (10) reduces to 2nβ+k+1nββk+1kβ3ββknβ(2k+1β)+1=k+1nββ2knβ+1βk+1kβ3β=2k(k+1)(kβ1)nβ+k+14β>0.
β
Claim 5.2**.**
- (i)
For all yβS,
β£N(y)ββSβ£β€nβ2Ξ΄(G)β1. For all yβR,
β£N(y)ββRβ£β€nβ2Ξ΄(G)β1.
2. (ii)
For all iβ[k], if Siβξ =β
, then β£Siββ£β₯2Ξ΄(G)+1β(1βk1β)nβ₯21β(knββ(β2nβ1βββΞ΄(G))). For all iβ[k], if Riβξ =β
, then β£Riββ£β₯2Ξ΄(G)+1β(1βk1β)nβ₯21β(knββ(β2nβ1βββΞ΄(G))).
Proof.
- (i)
Since C is a longest cycle of G, the vertex subsets
N(y), S, and N(y)ββS are pairwise disjoint for all yβS.
Thus
n=β£N(y)β£+β£Sβ£+β£N(y)ββSβ£β₯2Ξ΄(G)+1+β£N(y)ββSβ£, where the inequality holds by (9). Thus β£N(y)ββSβ£β€nβ2Ξ΄(G)β1. Similarly, N(y), R, and N(y)ββR are pairwise disjoint for all yβR, so β£N(y)ββRβ£β€nβ2Ξ΄(G)β1
2. (ii)
Let yβSiβ. We have that ViββSiββN(y)ββS so by (i) we have that β£Siββ£β₯knβββ£N(y)ββSβ£β₯2Ξ΄(G)+1β(1βk1β)n as desired. Similarly, if yβRiβ, then by (i) we have that β£Riββ£β₯knβββ£N(y)ββRβ£β₯2Ξ΄(G)+1β(1βk1β)n.
Finally, we have 2Ξ΄(G)+1β(1βk1β)nβ₯21β(knββ(β2nβ1βββΞ΄(G))) by Fact 6.4.(iii).β
Claim 5.3**.**
For all iβ[k], if Siββ©Riβξ =β
, then β£N(z)β©Viββ£β€β2nβ1βββΞ΄(G).
Proof.
Let 2β€iβ€k such that Siββ©Riβξ =β
and let yβSiββ©Riβ. So y is a successor along C of some vertex
in N(z), and a predecessor along C of some vertex in N(z) as well.
Since C is a longest cycle of G, neither N(z) nor N(y) contains
two consecutive vertices of C, so N(y)β©(SβͺR)=β
. Thus,
[TABLE]
Rearranging gives the result.
β
Claim 5.4**.**
2β+ββ²β<β2k+1ββ**
Proof.
Let i1ββ€i2ββ€β―β€iββ²β be the indices such that Rijββξ =β
for all jβ[ββ²]. By Claim 5.2.(ii) and Claim 5.3 and the fact that zβS1ββ©R1β we see that each of the sets S2β,β¦,Sββ, Ri2ββ,β¦,Riββ²ββ contributes at least 21β(knββ(β2nβ1βββΞ΄(G))) to β£N(z)ββV1ββ£. So we have
[TABLE]
Solving the above inequality for 2β+ββ²β, we have
[TABLE]
as desired.
β
Since we are supposing without loss of generality that ββ€ββ², we have by Claim 5.1 and Claim 5.4 that
[TABLE]
Thus if k is odd, we have a contradiction. So for the rest of the proof we will suppose that k is even and consequently by Claim 5.1 and 5.4, we have β=2kβ.
5.1 k is even and β=2kβ
Let
[TABLE]
and let H be the bipartite graph induced by [A,B]. Label the vertices of A as u1β,β¦,un/2β such that dHβ(u1β)β€β―β€dHβ(un/2β) and label the vertices of B as v1β,β¦,vn/2β such that dHβ(v1β)β€β―β€dHβ(vn/2β). Recall that SβA.
Since we are in the case where k is even, (2) reduces to
[TABLE]
Claim 5.5**.**
- (i)
Ξ΄(S,B)β₯2nβ+2βk+2n+2βββk2nβ+1, with equality only if Siβ=Viβ for some iβ[β]
2. (ii)
Ξ΄(AβS,B)β₯βk+2n+2βββ₯β£AβSβ£+1,
3. (iii)
Ξ΄(B,A)β₯βk+2n+2βββ₯β£AβSβ£+1.
Proof.
- (i)
This follows from Claim 5.2.(i) since for all yβS,
[TABLE]
If we have equality above, this implies that N(y)ββSβB, which in particular implies that if yβViβ, then ViββSiβ=β
.
2. (ii)
We have
[TABLE]
where the third inequality holds by Fact 6.3.(i). β
3. (iii)
Since Ξ΄(B,A)β₯Ξ΄(G)β(2nββknβ), the rest of the calculation is the same as in (ii).
Before proceeding with the rest of the proof, we finally filter out F3β.
Claim 5.6**.**
If Ξ΄(G)β₯2nββ1, then either G has a Hamiltonian cycle or k=2nβ and GβF3β.
Proof.
By (9) and the fact that β=2kβ, we have β£Sβ£=2nβ and thus A=S which means A is an independent set. So by Claim 5.5.(i), we have Ξ΄(A,B)β₯2nββ1. Furthermore we have by Claim 5.5.(iii) that Ξ΄(B,A)β₯βk+2n+2ββ. If Ξ΄(B,A)β₯2, then by Theorem 3.2, G has a Hamiltonian cycle.
So suppose Ξ΄(B,A)=1 which implies n=2k. In this case there is a vertex yβ²βB such that yβ² only has one neighbor in A, say xβ². We have Ξ΄(A,B)β₯2nββ1=β£Bβ£β1 so every vertex in Aβ{xβ²} is adjacent to every vertex in Bβ{yβ²}. Since d(yβ²,A)=1 and d(yβ²)=2nββ1, it must be the case that yβ² is adjacent to everything in B except the other vertex in its own part. Now we have all the edges between Aβ{xβ²} and Bβ{yβ²}, the edge xβ²yβ², all the edges from yβ² to B excluding the vertex in its own part, and we have all but possibly one edge from xβ² to Bβ{yβ²}, so GβF3β.
β
Now for the rest of the proof we may suppose that nβ₯3k (i.e. mβ₯3). We now use Theorem 3.2 to show that H, and therefore G, has a Hamiltonian cycle.
Suppose there exists iβ[2nβ] such that dHβ(viβ)β€i. By Claim 5.5.(iii), we must have
[TABLE]
Case 1 (2nββiβ€β£AβSβ£) By Claim 5.5.(ii) we have dHβ(u2nββiβ)β₯Ξ΄(AβS,B)β₯β£AβSβ£+1β₯2nββi+1.
Case 2 (2nββiβ₯β£AβSβ£+1)
Case 2.1 (kβ₯6) Note that since kβ₯6, when 3kβ€nβ€5k, we have Ξ΄(G)β₯2nββ1 and thus we are done by Claim 5.6. So for remainder of this case, suppose nβ₯6k (i.e. mβ₯6).
By Claim 5.5.(i)
[TABLE]
where the third inequality holds by Fact 6.3.(ii) since mβ₯6 and kβ₯6.
Thus the conditions of Theorem 3.2 are satisfied and therefore H has a Hamiltonian cycle.
Case 2.2 (k=4)
By Claim 5.5.(i), we either donβt have equality and thus
[TABLE]
where the third inequality holds by Fact 6.3.(ii) since mβ₯3 and kβ₯4, and thus we are done as in the previous case; or Ξ΄(S,B)=2nβ+2βk+2n+2βββk2nβ+1 and without loss of generality, S1β=V1β.
When 3kβ€nβ€4k, we have Ξ΄(G)β₯2nββ1 and thus we are done by Claim 5.6. So for remainder of this case, suppose nβ₯5k (i.e. mβ₯5). Also note that
since k=4, (2) reduces to
[TABLE]
Claim 5.7**.**
- (i)
β£S2ββ£β₯β6n+2ββ+1.
2. (ii)
Ξ΄(S1β,B)β₯2β6n+2ββ+1.
3. (iii)
Ξ΄(S2β,B)β₯Ξ΄(G)β₯4nβ+β6n+2ββ**
Proof.
- (i)
Since β£Sβ£β₯Ξ΄(G)+1, we have
[TABLE]
2. (ii)
Each vertex in S1β has at most β£V2ββS2ββ£ neighbors in V2β and thus by (i), at least 4nβ+β6n+2βββ(4nβββ£S2ββ£)β₯β6n+2ββ+β6n+2ββ+1=2β6n+2ββ+1
neighbors in B.
3. (iii)
Since S1β=V1β, the vertices in S2β have no neighbors in A and thus all of their neighbors are in B. β
We are in the case where 2nββiβ₯β£AβSβ£+1, so if
[TABLE]
then by Claim 5.7.(ii) we have
[TABLE]
where the third inequality holds by Fact 6.3.(ii) (in particular 3β6n+2ββ+1β₯3(6n+2β4β)+1=2nβ).
Otherwise together with (11), we have
2nββi=2nβββ6n+2ββ,
so by Claim 5.7.(i),(iii) we have
[TABLE]
where the third inequality holds by Fact 6.3.(i) since mβ₯5 and k=4.
This completes the proof of Theorem 1.1 and Proposition 1.2. β
6 Appendix: Numerical lemmas
The level of precision in (2) necessitates a careful handling of floors and ceilings throughout the paper. We collect a number of such required facts here.
First, to see how (2) compares to (1) we note the following fact which in particular implies that (2) is not obtained by simply replacing the strict inequality in (1) with a weak inequality (or the combination of a weak inequality and a floor).
Fact 6.1**.**
Let nβ₯kβ₯2 such that n is divisible by k. Then
[TABLE]
if and only if
- (i)
k* is even and nβ‘kmod(k+2)*
2. (ii)
k* is odd, n is even, and nβ‘kβ1mod(k+1)*
3. (iii)
k* is odd, n is odd, and nβ‘jmod(k+1) for some jβ{kβ1,1,3,β¦,2k+1β}*
and
[TABLE]
otherwise.
Proof.
- (i)
Since k divides n and k is even, this implies n is even. So (13) holds when βk+2n+2ββ=βk+2nββ and (14) holds when βk+2n+2ββ=βk+2nββ.
2. (ii)
Suppose k is odd and n is even. So (13) holds when βk+1n+2ββ=βk+1nββ and (14) holds when βk+1n+2ββ=βk+1nββ.
3. (iii)
Suppose k is odd and n is odd. So (13) holds when 2n+1β+βk+1n+2ββ=β2nβ+k+1nββ which is equivalent to βk+1n+2ββ=βk+1nβ2k+1βββ and (14) holds when 2n+1β+βk+1n+2ββ=β2nβ+k+1nββ which is equivalent to βk+1n+2ββ=βk+1nβ2k+1βββ. β
In the case when k is odd, the following fact shows that in order to estimate β2nββ+β2β2k+1ββn+2βββknβ from below, it suffices to consider the case when n is even.
Fact 6.2**.**
Let k be an odd integer with kβ₯3 and let n be divisible by k. Then
[TABLE]
Proof.
When n is odd we have
[TABLE]
and when n is even we have
[TABLE]
The following two technical facts will be used throughout the proof.
Fact 6.3**.**
Let m and k be positive integers and let n=mk (note that if k is even, then n is even).
- (i)
If k is even, then 2βk+2n+2βββknββ₯2(k+2n+2βkβ)βknβ=k+2(mβ2)(kβ2)β.
2. (ii)
If k is even, then 3βk+2n+2βββk2nββ₯3(k+2n+2βkβ)βk2nβ=k+2(mβ3)(kβ4)β6β.
3. (iii)
If k is odd and n is even, then 2βk+1n+2βββknββ₯2(k+1n+2β(kβ1)β)βknβ=k+1(mβ2)(kβ1)+4β.
4. (iv)
If k is odd and n is even, then 3βk+1n+2βββk2nββ₯3(k+1n+2β(kβ1)β)βk2nβ=k+1(mβ3)(kβ2)+3β.
Fact 6.4**.**
Let m and k be positive integers with kβ₯3 and let n=mk.
- (i)
If k is even and nβ₯3k, then 2(β2nββ+β2β2k+1ββn+2βββknβ)>(1βk1β)n.
2. (ii)
If k is odd and nβ₯2k, then 2(β2nββ+β2β2k+1ββn+2βββknβ)>(1βk1β)n.
3. (iii)
If nβ₯2k, then 3(β2nββ+β2β2k+1ββn+2βββknβ)β₯(2βk1β)nββ2nβ1βββ2.
Proof.
- (i)
Since n is divisible by k and k is even, n is even. Since m=knββ₯3 and kβ₯3, we have by Fact 6.3.(i)
[TABLE]
2. (ii)
Since k is odd, we may assume by Fact 6.2 that n is even. Since m=knββ₯2 and kβ₯3, we have by Fact 6.3.(iii)
[TABLE]
3. (iii)
As before, if k is even, then n is even. Also by Fact 6.2, if k is odd, we may assume that n is even. So we have
[TABLE]
Now if k is even, then we have kβ₯4 and thus by Fact 6.3.(ii) and m=knββ₯2 we have
[TABLE]
If k is odd, then since m=knββ₯2 and kβ₯3, we have by Fact 6.3.(iv) that
[TABLE]