This paper proves that under certain degree conditions, a directed graph contains two disjoint cycles covering a specified subset, and conjectures a similar property for multiple cycles with a partition of the subset.
Contribution
It establishes a degree condition ensuring two disjoint cycles cover a subset with a given partition, and proposes a conjecture for multiple cycles with sharper conditions.
Findings
01
Degree condition guarantees two disjoint cycles covering a subset
02
Conjecture extends the result to multiple cycles with larger partitions
03
Degree condition is shown to be sharp in general
Abstract
Let D=(V,A) be a directed graph of order n≥6. Let W be a subset of V with ∣W∣≥6. Suppose that every vertex of W has degree at least (3n−3)/2 in D. Then for any integer partition ∣W∣=n1+n2 with n1≥3 and n2≥3, D contains two disjoint directed cycles C1 and C2 such that ∣V(C1)∩W∣=n1 and ∣V(C2)∩W∣=n2. We conjecture that for any integer partition ∣W∣=n1+n2+⋯+nk with k≥3 and ni≥3(1≤i≤k), D contains k disjoint directed cycles C1,C2,…,Ck such that ∣V(Ci)∩W∣=ni for all 1≤i≤k. The degree condition is sharp in general.
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
TopicsLimits and Structures in Graph Theory · graph theory and CDMA systems · Graph Labeling and Dimension Problems
Full text
Partition of a Subset into Two Directed Cycles
with Partial Degrees
Hong Wang
Department of Mathematics
The University of Idaho
Moscow, Idaho, USA 83844
Let D=(V,A) be a directed graph of order n≥6. Let W be a subset of V with ∣W∣≥6. Suppose that every vertex of W has degree at least (3n−3)/2 in D. Then for any integer partition ∣W∣=n1+n2 with n1≥3 and n2≥3, D contains two disjoint directed cycles C1 and C2 such that ∣V(C1)∩W∣=n1 and ∣V(C2)∩W∣=n2. We conjecture that for any integer partition ∣W∣=n1+n2+⋯+nk with k≥3 and ni≥3(1≤i≤k), D contains k disjoint directed cycles C1,C2,…,Ck such that ∣V(Ci)∩W∣=ni for all 1≤i≤k. The degree condition is sharp in general.
1 Introduction
We discuss only finite simple graphs and strict directed graphs. The terminology and notation concerning graphs is that of [2], except as indicated. A set of graphs or directed graphs is said to be disjoint if no two of them have any common vertices. El-Zahar [5] conjectured that if G is a graph of order n=n1+⋯+nk with ni≥3(1≤i≤k) and the minimum degree of G is at least ⌈n1/2⌉+⋯+⌈nk/2⌉ then G contains k disjoint cycles C1,…,Ck of orders n1,…,nk, respectively. This conjecture has been verified for a number of cases (for example, see, [3], [5] and [10]). Sauer and Spencer in their work [8] conjectured that if the
minimum degree of G is at least 2n/3 then G contains every
graph of order n with maximum degree of at most 2. This
conjecture was proved by Aigner and Brandt [1]. Shi [9] showed that if G is 2-connected, W⊆V(G) and
d(x)≥n/2 for each x∈W then G contains a cycle passing
through all the vertices of W.
This generalizes the classic result of Dirac [4].
In [13], we proposed a conjecture on disjoint cycles of a graph G if we only know a degree condition on vertices in a subset of V(G):
Conjecture A [13] Let G be a graph of order n≥3. Let W be a subset of V(G) with ∣W∣≥3k where k is a
positive integer. Suppose that d(x)≥2n/3 for each x∈W.
Then for any integer partition ∣W∣=n1+⋯+nk with ni≥3(1≤i≤k), G contains k disjoint cycles
C1,…,Ck such that ∣V(Ci)∩W∣=ni for all 1≤i≤k.
In [13], we showed that G contains ⌊∣W∣/3⌋ disjoint cycles such that each of them contains at least three vertices of W. This conjecture was verified in [14] for the case k=2, i.e., ∣W∣=n1+n2. This result further supports Conjecture A. In [7] and [11], we proved the following theorem.
**Theorem B ** ([7],[11])Let D be a directed graph of order n≥4. Suppose that the minimum degree of D is at least (3n−3)/2. Then for any two integers s and t with s≥2, t≥2 and s+t≤n, D contains two disjoint
directed cycles of orders s and t, respectively.
Conjecture C [12] Let D be a directed graph of order n≥4. Suppose that the minimum degree of D is at least (3n−3)/2. Then for any directed graph H with Δ+(H)≤1 and Δ−(H)≤1, D contains a directed subgraph isomorphic to H.
To support this conjecture, we proved the following theorem:
**Theorem D ** [12] If D is a directed graph of order n≥4 with minimum degree at least (3n−3)/2, then D contains ⌊n/3⌋ disjoint directed triangles.
Continuing the work along this line, we propose the following conjecture:
Conjecture E *Let D=(V,A) be a directed graph of order n≥6. Let W be a subset of V. Suppose that every vertex of W has degree at least (3n−3)/2 in D. Then for any integer partition ∣W∣=n1+n2+⋯+nk with k≥2 and ni≥3(1≤i≤k), D contains k disjoint directed cycles C1,C2,…,Ck such that ∣V(Ci)∩W∣=ni for all 1≤i≤k. *
To demonstrate that the degree condition in Conjecture E is sharp in general for each n≥6, we construct the following directed graph Bn of order n. For each positive integer k, we use Kk∗
to denote the complete directed graph of order k, i.e., Kk∗ contains both (x,y) and (y,x) for any two distinct vertices x and y of Kk∗. The directed graph Bn consists of two disjoint complete directed subgraphs D′ and D′′ of order ⌊n/2⌋ and ⌈n/2⌉, respectively, and all the arcs (x,y) with x∈V(D′) and y∈V(D′′). Then the minimum degree of Bn is ⌊(3n−4)/2⌋. Clearly, Bn does not contain a directed cycle of order greater than ⌈n/2⌉.
In [15], we showed that if ∣W∣≥3k then D contains k disjoint directed cycles covering W such that each of them contains at least three vertices of W. In this paper, we prove the following result:
Theorem F *Let D=(V,A) be a directed graph of order n≥6. Let W be a subset of V. Suppose that every vertex of W has degree at least (3n−3)/2 in D. Then for any integer partition ∣W∣=n1+n2 with n1≥3 and n2≥3, D contains two disjoint directed cycles C1 and C2 such that ∣V(C1)∩W∣=n1 and ∣V(C2)∩W∣=n2. *
For convenience, we need some special terminology and notation. Let D be a digraph and G a
graph with the same vertex set, i.e., V=V(D)=V(G). We use V(D) and A(D) to denote the vertex set and the arc set of
D, respectively. Use E(G) to denote the edge set of G and let e(G)=∣E(G)∣. Let x∈V. The out-degree, in-degree and degree of x in D are denoted by a+(x), a−(x) and a(x), respectively. Let each of X and Y be a subset of V or a sequence of some vertices of V or a subgraph of G or a subdigraph of D. We use ND+(x,Y) denote the set of all the vertices y that are contained in Y with (x,y)∈A(D). We similarly define ND−(x,Y) and let
ND(x,Y)=ND+(x,Y)∪ND−(x,Y). We also define a+(x,Y)=∣N+(x,Y)∣,
a−(x,Y)=∣N−(x,Y)∣ and a(x,Y)=a+(x,Y)+a−(x,Y). Thus a+(x,D)=a+(x), a−(x,D)=a−(x) and a(x,D)=a(x). Define a(X,Y)=∑ua(x,Y), where u runs over all the vertices of X.
We define NG(x,Y) to be the set of all the vertices y in Y with xy∈E and let d(x,Y)=∣NG(x,Y)∣. Hence d(x,G)=d(x,D)=d(x,V)=d(u) which is the degree
of u in G. For a subset U⊆V, G[U] is the subgraph of G
induced by U and D[U] is the subdigraph
of D induced by U. Define d(X,Y)=∑ud(x,Y), where u runs over all the vertices of X.
We define I(xy,X)=NG(x,X)∩NG(y,X) and let i(xy,X)=∣I(xy,X)∣. We define I(xy,X)=ND+(x,X)∩ND−(y,X) and let i(xy,X)=∣I(xy,X)∣.
A graph or a digraph is said to be traceable if it contains a hamiltonian path or directed hamiltonian path, respectively.
A graph or a digraph is called hamiltonian if it
contains a hamiltonian cycle or directed hamiltonian cycle, respectively.
If C is a dicycle of D or a cycle of G and x is a vertex of C, we use x+ and x− to denote the successor and the predeccssor of x on C, respectively. If a and b are two vertices of C, we use C[a,b] to represent the path of C from a to b along the direction of C. We adopt the notation C(a,b]=C[a,b]−a, C[a,b)=C[a,b]−b and C(a,b)=C[a,b]−a−b.
For two vertices x and y of D, we say that x is adjacent to y in D if either (x,y) or (y,x) is an arc of D.
For any two vertices x and y of G, we define μ(xy)=1 if xy is an edge of G and μ(xy)=0 otherwise. For any integer m, let λm=1 if m is odd and otherwise λm=0.
2 Lemmas
Let D=(V,A) be a digraph. Let G=(V,E) be the underlining graph of D, i.e., V(G)=V and E(G)={xy∣(x,y)∈Aand(y,x)∈A}. We will use the following lemmas which are generally well known.
Lemma 2.1
Let P=x1x2…xk be a path of G and {y,z}⊆V−V(P). The following two statements hold:
(a)* If d(y,P)>k/2 then {xi,xi+1}⊆NG(y,P) for some i∈{1,2,…,k−1} unless k is odd, d(y,P)=(k+1)/2 and NG(y,P)={x1,x3,x5,…,xk};*
(b)* If d(y,P)≥k/2, then P+y is traceable;*
(c)* If d(x1y,P)≥k, then P+y is traceable;*
(d)* If d(yz,P)≥k+2 then P+y+z is traceable.*
Proof. To show (a), it is easy to see that if d(y,xixi+1)≤1 for all i∈{1,…,k−1}, then k must be odd and {x1,xk}⊆NG(y,P) as d(y,P)>k/2 and it follows that NG(y,P)={x1,x3,x5,…,xk}.
To show (b), it is easy to see that if {x1,xk}∩NG(y,P)=∅, then there exists i∈{2,…,k−2}
such that {xi,xi+1}⊆NG(y,P). Thus the path P′=x1…xiyxi+1…xk satisfies the requirement.
To show (c), it is easy to see that if d(xk,x1y)≥1 then (c) holds. So we may assume that d(xk,x1y)=0. As d(x1y,P)≥k, it follows that {xiy,xi+1x1}⊆E for some i∈{1,…,k−2}. Therefore yxixi−1…x1xi+1xi+1…xk is a hamiltonian path of P+y.
To show (d), we see that {yxi,yxj,zxi+1,zxj+1}⊆E for some {i,j}⊆{1,2,…,k} with i<j, where xk+1=x1. If j=k, zPy is a path of G and if j<k then
[TABLE]
is a hamiltonian path of P+y+z.
Lemma 2.2
Let C be a dicycle of D or a cycle of G. Let y∈V−V(C). Suppose that d(y,C)≥l(C)/2. Then either D[V(C)∪{y}] is hamiltonian if C is a cycle of D and G[V(C)∪{y}] is hamiltonian if C is a cycle of G, or l(C) is even and there is a list x1,…,xl(C) of vertices of C along the direction of C such that NG(y,C)={x2i−1∣1≤i≤l(C)/2}.
Proof. If the last statement does not hold, then there exists x∈V(C) such that {x,x+}⊆NG(y) and the lemma follows.
Lemma 2.3
Let P=x1x2…xk be a path of G with k≥3. If either a+(x1,P)+a+(xk,P)≥k or a−(x1,P)+a−(xk,P)≥k, then D[V(P)] is hamiltonian. Therefore if a(x1,P)+a(xk,P)≥2k−1 then D[V(P)] is hamiltonian.
Proof. If a+(x1,P)+a+(xk,P)≥k, then {(x1,xi+1),(xk,xi)}⊆A for some i∈{1,…,k−1} and so (x1,xi+1,xi+2,…,xk,xi,xi−1,…,x1) is a dicycle of D[V(P)]. Similarly, we can easily see that the lemma holds if a−(x1,P)+a−(xk,P)≥k. Since a(x1,P)+a(xk,P)≥2k−1 implies either a+(x1,P)+a+(xk,P)≥k or a−(x1,P)+a−(xk,P)≥k, the lemma follows.
Lemma 2.4
[6]* Let P=xtxt−1…x1 be a longest path starting at xt in G. Let c>t/2.
Suppose that for each v∈V(G), if there
exists a longest path starting at xt in G such that the path
ends at v then dG(v)≥c. Then there exists r≥c+1 such that N(xi)⊆{x1,x2,…,xr}, G[P] has an xt-xi
hamiltonian path containing xtxt−1…xr and dG(xi)≥c for all i∈{1,2,…,r−1}. Moreover, if t>r then xr is a cut-vertex of G.*
3 Proof of Theorem F
Let D=(V,A) be a directed graph of order n. Let W⊆V with ∣W∣≥6. Suppose that a(x)≥(3n−2−λn)/2 for all x∈W. Suppose, for a contradiction, that for some integer partition ∣W∣=n1+n2 with n1≥3 and n2≥3, D does not contain two disjoint dicycles C1 and C2 such that ∣V(C1)∩W∣=n1 and ∣V(C2)∩W∣=n2. For the proof, we may assume that D−W has no arcs. Let G=(V,E) be the underlining graph of D.
For any subgraph G′ of G, we use W(G′) to denote W∩V(G′). For convenince, let A′={xy∣{x,y}⊆Vandeither(x,y)∈Aor(y,x)∈A}, i.e., x and y are adjacent in D if and only if xy∈A′. For a dicycle or a cycle C, we use l(C) to denote the length of C and use lw(C) to denote ∣V(C)∩W∣ which is called the W-length of C. Similarly, we define l(P) and lw(P) for a dipath or a path P.
For each x∈W, we have
[TABLE]
For each partition (X,Y) of V and u∈X, if a(u,X)≤∣X∣−1, then we have
[TABLE]
Lemma 3.1
There is a path P in G such that V(P)⊇W.
Proof. We choose a path P in G with lw(P) maximal. Subject to this, we
choose P with l(P) minimal. Assume there exists x0∈W−V(P). Say P=x1x2…xt. Then x0x1∈E, I(x1x0,V−V(P))=∅ and so d(x1x0,V−V(P))≤n−t−1. By (2), d(x1x0,P)≥(n−λn)−(n−t−1)≥t. By Lemma 2.1(c), P+x0 is traceable, a contradiction.
By Lemma 3.1, G has two disjoint paths P1 and P2 with lw(P1)=n1 and lw(P2)=n2. Let G1=G[V(P1)] and G2=G[V(P2)]. We choose P1 and P2 such that
[TABLE]
Subject (7), we choose P1 and P2 such that
[TABLE]
Say P1=x1x2…xs, P2=y1y2…yt, R=V−V(G1∪G2) and r=∣R∣=n−s−t. Let H=G[V(G1∪G2)], D1=D[V(G1)] and D2=D[V(G2)]. By (8), we obtain the following lemma.
Lemma 3.2
Let x be an endvertex of a hamiltonian
path of G1 and y an endvertex of a hamiltonian path of G2. If
both G1−x+y and G2−y+x are traceable, then
[TABLE]
In particular, (9) holds if d(x,G2)≥(t+1)/2 and
d(y,G1)≥(s+1)/2 hold.
Proof. We have
[TABLE]
If both G1−x+y and G2−y+x are traceable, then e(G1)+e(G2)≥e(G1−x+y)+e(G2−y+x) by (8) and it follows that (9) holds. Note that by Lemma 2.1(b), if d(x,G2)≥(t+1)/2 and d(y,G1)≥(s+1)/2, both G1−x+y and G2−y+x are traceable.
Similarly, by (8), one can easily see that the following Lemma 3.3 holds.
Lemma 3.3
Let x and y be the two endvertices of a hamiltonian path of G1 and u and v be the two endvertices of a hamiltonian path of G2. Suppose that
both G1−x−y+u+v and G2−u−v+x+y are taceable. Then
[TABLE]
Lemma 3.4
For some i∈{1,2}, Di does not have a dicycle of W-length ni and if x and y are the two endvertices of a hamiltonian path of Gi, then I(xy,R)=∅, I(yx,R)=∅ and so a(xy,R)≤2r.
Proof. If D1 contains a dicycle of W-length n1, then D[V(D2)∪R] does not contain a dicycle of W-length n2, and this implies that the lemma holds with i=2. Therefore we may assume that D1 does not contain a dicycle of W-length n1. Similarly, we may assume that D2 does not contain a dicycle of W-length n2. In particular, x1xs∈A′ and y1yt∈A′. On the contrary, say the lemma fails. Say w.l.o.g. that I(x1xs,R)=∅ and I(y1yt,R)=∅. Since D does not contain the two required dicycles, we see that for some u∈R, I(x1xs,R)=I(y1yt,R)={u}. Moreover, I(xsx1,R)⊆{u} and I(yty1,R)⊆{u}. This implies that a(x1xs,R)≤2r+2 and a(y1yt,R)≤2r+2. Then for each vw∈{x1xs,y1yt}, a(vw,H)≥3n−2−λn−2r−2=3s+3t+r−4−λn. Therefore for each vw∈{x1xs,y1yt}, we have
[TABLE]
Since both G1 and G2 are not hamiltonian, d(x1xs,G1)≤s−1 and d(y1yt,G2)≤t−1. Therefore d(x1xs,G2)≥s+t+r−λn−(s−1)≥t+1+r−λn and d(y1yt,G1)≥s+t+r−λn−(t−1)≥s+1+r−λn. Let σ=d(x1xs,G2)+d(y1yt,G1)−d(x1xs,G1)−d(y1yt,G2). Then σ≥4+2(r−λn).
First, assume that there exist two independent edges between {x1,xs} and {y1,yt} in G. Say w.l.o.g. {x1yt,xsy1}⊆E. By Lemma 3.2, d(x1,G1)+d(y1,G2)≥d(x1,G2)+d(y1,G1)−2μ(x1y1) and d(xs,G1)+d(yt,G2)≥d(xs,G2)+d(yt,G1)−2μ(xsyt). Thus 2μ(x1y1)+2μ(xsyt)≥σ. Since 4≥2μ(x1y1)+2μ(xsyt) and σ≥4+2(r−λn), it follows that equality hold in (11) and (12) with r=λn=1. In particular, ∣ND(z,H)∣=s+t−2 for all z∈{x1,xs,y1,yt}. Hence x1y2∈A′ and y1x2∈A′. Consequently, D1−x1+y1 and D2−y1+x1 are hamiltonian, a contradiction.
Therefore G does not contain two independent edges between {x1,xs} and {y1,yt} and so d(x1xs,y1yt)≤2. Assume for the moment that d(x1xs,G2)≥(t+2) and d(y1yt,G1)≥s+2. Then d(x1xs,G2−y1−yt)≥(t−2)+2 and d(y1yt,G1−x1−xs)≥(s−2)+2. By Lemma 2.1(d) and Lemma 3.3, we shall have that d(x1xs,G1)+d(y1yt,G2)≥d(x1xs,G2)+d(y1yt,G1)−2d(x1xs,y1yt). Thus s−1+t−1≥s+2+t+2−4, a contradiction. Therefore we may assume w.l.o.g.that d(x1xs,G2)≤t+1. By (12), d(x1xs,G2)≥s+t+r−λn−(s−1)=t+1+r−λn. It follows that d(x1xs,G1)=s−1, d(x1xs,G2)=t+1, r=1, λn=1 (i.e., n is odd), and equality hold in (11) and (12) with {v,w}={x1,xs} and ND(x1)=ND(xs)=V−{x1,xs}. Moreover, a(x1xs,R)=a(x1xs,u)=4, i.e., d(x1xs,R)=d(x1xs,u)=2. As x1xs∈A′ and by (1), we see that d(x1,G)≥(n+1)/2 and d(xs,G)≥(n+1)/2. As x1xs−1∈A′, D1−xs is hamiltonian and so d(xs,G1)≤(s−1)/2 by Lemma 2.2 because D1 is not hamiltonian. Similarly, d(x1,G1)≤(s−1)/2. It follows that s and t are odd and d(xi,G1)=(s−1)/2 and d(xi,G2)=(t+1)/2 for i∈{1,s}. Suppose that d(y1yt,G1)≤s+1. Then this argument allows us to see that d(yj,G2)=(t−1)/2 and d(yj,G1)=(s+1)/2 for j∈{1,t}. For each i∈{1,s} and j∈{1,t}, applying Lemma 3.2 with xi and yj in place of x and y, we see that μ(xiyj)=1. Thus d(x1xs,y1yt)=4, a contradiction.
Therefore d(y1yt,G1)≥s+2. Say w.l.o.g. d(y1,G1)≥⌈(s+2)/2⌉=(s+3)/2. Applying Lemma 3.2 with x1 and y1 in place of x and y, we see that d(y1,G2)≥(t+1)/2. Thus d(yt,G2)≤t−1−(t+1)/2=(t−3)/2. By (1), d(yt,G)≥(n+1)/2. Consequently, d(yt,G1)≥(n+1)/2−(t−3)/2−μ(ytu)≥(s+3)/2. Applying Lemma 3.2 with x1 and yt in place of x and y, we obtain a contradiction to (9).
We now may assume that Lemma 3.4 holds for i=1. As D1 is not hamiltonian, a(x1xs,P1)≤2(s−1) by Lemma 2.3 and so a(x1xs,G1+R)≤2(s−1)+2r. Thus
[TABLE]
We assume w.l.o.g. that a(x1,G1+R)≤a(xs,G1+R). Then a(x1,G1+R)≤s+r−1 and a(x1,G2)≥(3n−2−λn)/2−(s+r−1)=(3t+s+r−λn)/2. Thus
[TABLE]
It follows that t≥(n−λn)/2. Clearly, (n−λn)/2−1>(t−1)/2. By Lemma 2.1(b) and Lemma 2.2, we obtain the following corollary:
Corollary 3.5
*For each v∈V(G2), if G2−v is traceable then G2−v+x1 is traceable and if D[V(G2−v)] is hamiltonian then D[V(G2−v+x1)] is hamiltonian. *
Lemma 3.6
G2* is not hamiltonian and G2+z is not hamiltonian for each z∈R.*
Proof. On the contrary, say the lemma fails. If G2 is a hamiltonian, let C be a hamiltonian cycle and otherwise let C be a hamiltonian cycle of G2+z0 for some z0∈R. Let b1,b2,…,bn2 be a list of vertices in W(C) along the direction of C. Let hi=C[bi,bi+1) for each i∈{1,…,n2}, where the operation in the subscripts is taken in {1,2,…,n2} by modulo n2. Note that as D−W has no arcs, each hi has at most two vertices. Let R′=R−{z0} if l(C)=t+1 and otherwise let R′=R. Say r′=∣R′∣.
Let x∗=x2 if x2∈W and otherwise let x∗=x3 with x3∈W. Suppose that a(x∗xs,hi)≤2∣V(hi)∣ for all i∈{1,2,…,n2}. Then a(x∗xs,C)≤2l(C). Thus a(x∗xs,D−V(C))≥3n−2−λn−2l(C)=3s+3r′+l(C)−2−λn. As x1xs∈A′, we obtain that 2(s+r′−1)+2(s+r′−2)≥a(x∗xs,D−V(C))≥3s+3r′+l(C)−2−λn and so s+r′≥l(C)+4−λn. This implies that n≥2l(C)+4−λn, i.e., l(C)≤(n−4+λn)/2. But l(C)≥t≥(n−λn)/2, a contradiction.
Therefore a(x∗xs,hi)≥2∣V(hi)∣+1 for some i∈{1,…,n2}. Then either a+(x∗,hi)+a−(xs,hi)≥∣V(hi)∣+1 or a−(x∗,hi)+a+(xs,hi)≥∣V(hi)∣+1. As ∣V(hi)∣≤2, it follows that D[V(P1[x∗,xs])∪V(hi)] has a hamiltonian dicycle C1 of W-length n1. Therefore D−V(C1) does not contain a dicycle of W-length n2. In particular, d(x1,bi−1bi+1)≤1 and a(x1,bi−1bi+1)≤2. Recall that a(x1,G1+R)≤s+r−1. By (3) to (6) with Y=V(G2)−{bi−1,bi+1} and x1 in place of u, we obtain
[TABLE]
By (16), we now have
[TABLE]
Note that if C is a hamiltonian cycle of G2, then by (7), V(G2)⊆W, i.e., n2=t. We claim
[TABLE]
To see (19), say, for a contradiction, that a(bi−1bi+1,R′∪V(C)−{bi})≤2r′+2(l(C)−1)−1. As a(x1,bi−1bi+1)≤2, a(bi−1bi+1,R′∪V(C)∪{x1}−{bi})≤2r′+2l(C)−1=2(t+r−1)+1. This implies that a(u,R′∪V(C)∪{x1}−{bi})≤t+r−1 for some u∈{bi−1,bi+1}. By (3) to (6) with Y=V(G1)∪{bi}−{x1}, we obtain s≥(n−λn)/2. By (17), s+t≥(n−λn)/2+(n+4−λn)/2≥n+1, a contradiction. Hence (19) holds.
We break into the following four cases. Cases 2 to 4 are similar to Case 1 with more subtle details.
Case 1. bi+=bi+1 and bi−=bi−1.
Then hi=bi. Let P=C−bi. If the latter of (19) holds, a(z′,bi−1bi+1)≥3 for some z′∈R′ and we let C′ be a hamiltonian dicycle of D[V(P)∪{z′}]. Otherwise the former holds and we let C′ be a hamiltonian dicycle of D[V(P)] by Lemma 2.3. By (18), d(x1,C′)≥(t+s+r−λn)/2−1=(t+s−2+r−λn)/2>l(C′)/2. Thus D[V(C′)∪{x1}] is hamiltonian by Lemma 2.2, a contradiction.
Case 2. bi+=bi+1 and bi−−=bi−1.
Then hi=bi. In this case, as bi−∈W, C is not a hamiltonian cycle of G2 by (7). Thus l(C)=t+1 with V(C)∩R={z0} and so r≥1 . As in Case 1, we see that we may assume that D[V(C)−{bi}] is not hamiltonian. Thus bi−bi+1∈A′. If the former of (19) holds, then a(bi−1bi+1,C−bi−bi−)≥2(l(C)−2) and we let C′′ be a hamiltonian dicycle of D[V(C−bi−bi−)] by Lemma 2.3. Otherwise D[V(C−bi−bi−)∪{z′′}] is hamiltonian for some z′′∈R′ with a(bi−1bi+1,z′′)≥3 and we let C′′ be one of its hamiltonian dicycles. If x1bi−∈E, then d(x1,C′′)≥(t+s+r−λn)/2−1≥(l(C′′)+1)/2 and so D[V(C′′)∪{x1}] is hamiltonian, a contradiction. Therefore x1bi−∈E. Then x1bi+1∈A′ for otherwise D[V(C)∪{x1}−{bi}] is hamiltonian. By (14), we see that d(x1,C′′)≥d(x1,G2)−2≥(3t+s+r−λn)/2−(t−1)−2=(t+s−2+r−λn)/2≥(l(C′′)+1)/2. Again, D[V(C′′)∪{x1}] is hamiltonian, a contradiction.
Case 3. bi++=bi+1 and bi−=bi−1.
Then hi=bibi+. In this case, l(C)=t+1 with V(C)∩R={z0}. Let P′=C−bi−bi+. First, assume that the latter of (19) holds. Then D[V(P′)∪{u}] has a hamiltonian dicycle L for some u∈R′ with a(bi−1bi+1,u)≥3. Thus r′≥1 and so r≥2. Clearly, l(L)≤t and d(x1,L−u−z0)=d(x1,G2−bi−bi+)≥(t+s+r−λn)/2−2=(t+s+r−4−λn)/2≥l(L)/2. As D[V(L)∪{x1}] is not hamiltonian and by Lemma 2.2, it follows that l(L)=t, d(x1,G2−bi−bi+)=d(x1,L)=t/2, d(x1,NL(z0))=2 and so D[V(L)∪{x1}−{z0}] is hamiltonian, a contradiction. Hence the latter of (19) does not hold. Note that this argument allows us to see that D[V(P′)] does not have a hamiltonian dicycle in both Case 3 and Case 4.
Therefore a(bi−1bi+1,R′)≤2r′, and by Lemma 2.3, a(bi−1bi+1,P′)≤2(∣V(P′)∣−1). As a(x1,bi−1bi+1)≤2, it follows that a(bi−1bi+1,R′∪V(C)∪{x1}−{bi,bi+})≤2(t+r−2). Thus a(v,R′∪V(C)∪{x1}−{bi,bi+})≤t+r−2 for some v∈{bi−1,bi+1}. By (3) to (6) with Y=V(G1)∪{bi,bi+}−{x1}, we obtain s+1≥(n−λn)/2. By (17), n≥s+t+1≥(n−λn)/2+(n+4−λn)/2≥n+1, a contradiction.
Case 4. bi++=bi+1 and bi−−=bi−1.
In this case, hi=bibi+, l(C)=t+1 and r≥1. As noted in Case 3, we may assume that D[V(P′)] is not hamiltonian for otherwise D[V(P′)∪{x1}] is hamiltonian, where P′=C−bi−bi+. Similarly, D[V(C)−{bi,bi−}] is not hamitonian. Thus bi+1bi−∈A and bi−1bi+∈A′. Set P′′=C−bi−−bi−bi+.
If the former of (19) holds, then a(bi−1bi+1,P′′)≥2(l(C)−1)−4=2∣V(P′′)∣ and by Lemma 2.3, we let Q be a hamiltonian dicycle of D[V(P′′)]. Otherwise D[V(P′′)∪{c}] is hamiltonian for some c∈R′ with a(c,bi−1bi+1)≥3, and we let Q be a hamiltonian dicycle of D[V(P′′)∪{c}]. If x1bi−∈E, d(x1,Q)≥(t+s+r−λn)/2−2=(t+s−4+r−λn)/2>l(Q)/2 and so D[V(Q+x1)] is hamiltonian by Lemma 2.2, a contradiction. Therefore x1bi−∈E. In this situation, x1bi+1∈A′. By (14), d(x1,G2)≥(t+s+r+2−λn)/2. Thus d(x1,Q)≥(t+s+r+2−λn)/2−3=(t+s−4+r−λn)/2>l(Q)/2 and so D[V(Q+x1)] is hamiltonian, a contradiction.
By Lemma 3.6, we obtain
[TABLE]
Since we have d(y1yt,G)≥3n−2−λn−∣ND(y1)∣−∣ND(yt)∣, it follows that
[TABLE]
From (21) and (22), it follows that
[TABLE]
We claim
[TABLE]
Proof of (27). On the contray, say (27) fails. Since G1 is not hamiltonian, d(x1xs,G1)≤s−1. First, assume that d(x1xs,y1yt)≥3. Then there exist two independent edges of G between {x1,xs} and {y1,yt}. Say w.l.o.g. {x1yt,xsy1}⊆E. If yty2∈A′ then D[V(G2)∪{x1}−{y1}] is hamiltonian by Corollary 3.5 and so y1x2∈A′. That is, either yty2∈A′ or y1x2∈A′. By (22), we obtain d(y1yt,G1)≥s+2−λn. With (13) and (20), we obtain
[TABLE]
Thus d(x,G2)+d(y,G1)−d(x,G1)−d(y,G2)≥3 for some (x,y)∈{(x1,y1),(xs,yt)}. This contradicts Lemma 3.2. Hence d(x1xs,y1yt)≤2. Thus by (13), d(x1xs,G2−y1−yt)≥t+s+r−λn−2≥(t−2)+2. Next, assume d(y1yt,G1)≥s+2. Then d(y1yt,G1−x1−xs)≥s+2−2=(s−2)+2. By Lemma 2.1(d), both G1−x1−xs+y1+yt and G2−y1−yt+x1+xs are traceable. By Lemma 3.3, we would have that (s−1)+(t−1)≥(t+s+r−λn)+(s+2)−4, which is impossible. Hence d(y1yt,G1)≤s+1. Together with (24), (25) and (26), we obtain y1yt∈A′. Hence (27) holds. □
We now choose P2=y1y2…yt in G2 such that
[TABLE]
Lemma 3.7
For each v∈{y1,yt}, if d(v,G2)≤(t−1)/2 then d(v,G1)≤(s−1)/2.
Proof. On the contrary, say d(y1,G2)≤(t−1)/2 and d(y1,G1)≥s/2. By Corollary 3.5, G2−y1+x1 is traceable. First, assume that G1−x1+y1 is tracable. Then by Lemma 3.2, (s−2)+(t−1)/2≥(t+s+r−λn)/2+⌈s/2⌉−2μ(x1y1). This implies that r=0, λn=1, d(x1,G1)=s−2, d(y1,G2)=(t−1)/2, d(y1,G1)=s/2 and x1y1∈E. Thus s is even, s≥4 and NG(x1,G1)={x2,…,xs−1}. It follows that d(yt,G2)≤(t−1)/2 by (20), and by (24), d(yt,G1)≥s/2. Clearly, d(xs,G1)=1 as d(x1xs,G1)≤s−1, and G1−xs+yt is traceable. As x1xs∈A′ and by (1), d(xs,G)≥(t+s+1)/2 and so d(xs,G2)≥(t+s−1)/2. Thus G2−y1+xs is traceable. Applying Lemma 3.2 with xs and y1 in place of x and y, we obtain 1+(t−1)/2≥(t+s−1)/2+s/2−2μ(xsy1), which is impossible. Hence G1−x1+y1 is not traceable. Therefore d(y1,x2xs)=0 and d(y1,P1−{x1,x2,xs})≤⌊(s−2)/2⌋ by Lemma 2.1(a). As d(y1,G1)≥s/2, it follows that d(y1,G1)=s/2 with NG(y1,G1)={x1,x3,x5,…,xs−1}. Then NG(x2,G1−x1)⊆{x3,x5,…,xs−1} and NG(xs,G1)⊆{x3,x5,…,xs−1} for otherwise G1−x1+y1 is traceable. Thus d(x2,G1)≤s/2 and d(xs,G2)≤(s−2)/2. We claim
[TABLE]
To see (29), we have that d(x2,G1)≤s/2, d(y1,x2xs)=0 and G1−xi+y1 is traceable for each i∈{2,s}. If d(xs,G2)≥(t−1)/2, then G2−y1+xs is traceable by Lemma 2.1(b) and so (s−2)/2+(t−1)/2≥(t−1)/2+s/2 by Lemma 3.2, which is impossible. Hence d(xs,G2)≤(t−2)/2. If x2∈W and d(x2,G2)≥t/2, then G2−y1+x2 is traceable by Lemma 2.1(b) and so s/2+(t−1)/2≥t/2+s/2 by Lemma 3.2, which is impossible. Therefore (29) holds.
Then d(xs,R)≥(n−λn)/2−(s−2)/2−(t−2)/2=(r+4−λn)/2. We also have that d(y1,R)≥(n−λn)/2−s/2−(t−1)/2=(r+1−λn)/2. We obtain that d(xsy1,R)≥r+5/2−λn and so i(xsy1,R)≥2. Note that r≥2. If x2∈W and a(x2,NG(y1,R))=0 then d(x2,G)≥(n+4−λn)/2 by (1) and so i(x2y1,R)≥2, a contradiction. Therefore if x2∈W then a(x2,NG(y1,R))>0. It follows that D1−x1+y1+R contains a dicycle of W-length n1. Thus D[V(G2−y1+x1)] does not contain a hamiltonian dicycle. By Corollary 3.5, D[V(G2−y1)] is not hamiltonian. In particular, y2yt∈A′. By (24), d(yt,G1)≥s/2 as d(y1,G1)=s/2. If d(yt,G2)≤(t−1)/2, then with yt in place of y1 in the above argument, we see that d(yt,G1)=s/2 and so yty2∈A′ by (25), a contradiction. Therefore d(yt,G2)≥t/2. With (28), we apply Lemma 2.4 to y2y3…yt in G2−y1, we see that there exists a∈{2,…,t−2} such that for each i∈{a+1,a+2,…,t}, there exists a hamiltonian path of G2 from y1 to yi containing y1y2…ya, d(yi,G2)≥t/2 and N(yi,G2)⊆{ya,ya+1,…,yt}. Clearly G[{ya,ya+1,…,yt}] is hamiltonian. Thus a=2 and so a≥3. By Lemma 3.6, y1yi∈E for all i∈{a+1,a+2,…,t}. As yty2∈A′ and by (20), (21), (22) and (27), we obtain that d(y1yt,G1)=s+1 and d(y1yt,G2)=t−1. It follows that NG(y1,G2)={y2,y3,…,ya}. Thus G2 has a hamiltonian path from y2 to yt. By (27) with y2 in place of y1, we obtain that y2yt∈A′, a contradiction.
We are now in a position to complete our proof. By (24) , we may assume w.l.o.g. that d(y1,G1)≥s/2. By Lemma 3.6 and Lemma 3.7, d(y1,G2−yt)≥t/2. With (28), applying Lemma 2.4 to yt−1yt−2…y1, there exists p≥⌈t/2⌉ such that for each i∈{1,…,p}, NG(yi,G2)⊆{y1,…,yp+1}, G2 has a hamiltonian path from yt to yi containing ytyt−1…yp+1 and d(yi,G2)≥d(y1,G1). By Lemma 3.6, NG(yt,G2)⊆{yp+1,yp+1,…,yt−1} and so d(yt,G2)≤t−1−p≤⌊(t−2)/2⌋. By Lemma 3.7, d(yt,G1)≤⌊(s−1)/2⌋. Thus d(y1,G1)≥d(y1yt,G1)−d(yt,G1)≥⌈(s+1)/2⌉ by (24). By (27) with yi in place of y1, ytyi∈A′ for all i∈{1,2,…,p}. Thus for all j∈{1,2,…,p}, D[V(G2)−yj] is hamiltonian and so D[V(G2)−yj+x1] is hamiltonian by Corollary 3.5. We conclude that D[V(G1−x1+yi+R)] does not contain a dicycle of W-length n1 for all i∈{1,2,…,p}. This implies that a(yi,x2xs)≤2 for all i∈{1,2,…,p} and so a(x2xs,y1y2…yp})≤2p.
Assume that x2∈W. By Corollary 3.5, G2−y1+x1 is traceable. By (7), y1x3∈E and y1xs∈E. By (7) and Lemma 2.1(a), d(y1,P1−{x1,x2,x3,xs})≤(s−3)/2. As d(y1,G1)≥(s+1)/2, it follows that d(y1,x1x2)=2 and d(y1,P1−{x1,x2,x3,xs})=(s−3)/2. Thus y1xs∈A′ for otherwise D[V(G1−x1+y1)] is hamiltonian. By (20), (21), (22) and (27), we obtain that d(y1yt,G1)=s+1 and so d(yt,G1)=(s+1)/2, a contradiction since we have d(yt,G1)≤⌊(s−1)/2⌋ in the above paragraph.
Therefore x2∈W. We claim that a(x2xs,R)≤2r and x2xs∈A′. If this is not true, then either e(z,x2xs)≥3 for some z∈R or x2xs∈A′. Since D[V(G1−x1+R+y1)] does not contain a dicycle of W-length n2, we see that d(y1,x2xs)≤1 and d(y1,xixi+1)≤1 for all i∈{2,…,s−1}. By Lemma 2.1(a), this yields that d(y1,P1−x1)≤(s−1)/2. It follows that d(y1,G1)=(s+1)/2, y1x1∈E and d(y1,x2xs)=1. Thus either y1x2∈A′ or y1xs∈A′. By (20), (21), (22) and (27), it follows that d(y1yt,G1)=s+1 and so d(yt,G1)=(s+1)/2, a contradiction. Therefore a(x2xs,R)≤2r and x2xs∈A′. Thus s≥4.
Next, we claim that a(x2xs,P1−x1)≤2(s−2). If this is not true, then by Lemma 2.3, D[V(G1−x1)] has a hamiltonian dicycle C. By Lemma 2.2, d(x1,C)≤(s−1)/2 and d(y1,C)≤(s−1)/2. It follows that d(y1,C)=(s−1)/2 and y1x1∈E. Since d(y1,G1)=(s+1)/2, G1−x1+y1 is traceable by Lemma 2.1(b). By Corollary 3.5, G2−y1+x1 is traceable. By Lemma 3.2, (s−1)/2+d(y1,G2)≥d(x1,G2)+(s+1)/2−2. Thus p≥d(y1,G2)≥d(x1,G2)−1≥(t+s+r−λn)/2−1. Then d(yt,G2)≤t−1−p≤(t−s−r+λn)/2. By Lemma 3.7, d(yt,G1)≤(s−1)/2. Thus
[TABLE]
This contradicts (2).
Therefore a(x2xs,P1−x1)≤2(s−2). Clearly, a(x2xs,x1)=2 as x1xs∈A′. Let X=R∪V(G1)∪{y1,…,yp}. We obtain that a(x2xt,X)≤2r+2(s−2)+2+2p=2(∣X∣−1). By (6), n−∣X∣=t−p≥(t+s+r−λn)/2. Thus t≥2p+s+r−λn≥t+s+r−λn>t, a contradiction. This proves the theorem.
Bibliography15
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] M. Aigner and S. Brandt, Embedding arbitrary graphs of maximum degree two, J. London Math. Soc. , (2) 48 (1993), 39–51.
2[2] J.A. Bondy and U.S.R. Murty, Graph Theory with Applications , The Macmillan Press, London, 1976.
3[3] K. Corrádi and A. Hajnal, On the maximal number of independent circuits in a graph, Acta Math. Acad. Sci. Hungar . 14(1963), 423–439.
4[4] G. Dirac, Some theorems on abstract graphs, Proc. London. Math. 21(1972), 111–113.
5[5] M.H. El-Zahar, On circuits in graphs, Discrete Math. 50(1984), 227–230.
6[6] P. Erdős and T. Gallai, On maximal paths and circuits of graphs, Acta Math. Acad. Sci. Hungar. 10(1959), 337–356.
7[7] C. Little and H. Wang, Vertex-disjoint cycles in a directed graph, The Australasian Journal of Combinatorics , 12(1995), 113-119.
8[8] N. Sauer and J. Spencer, Edge disjoint placement of graphs, J. Combin. Theory B , 25(1978), 295–302.