**A sufficient condition for pre-Hamiltonian cycles in bipartite digraphs
**
**Samvel Kh. Darbinyan and Iskandar A. Karapetyan
**
Institute for Informatics and Automation Problems, Armenian National Academy of Sciences
E-mails: [email protected], [email protected]
Abstract
Let D be a strongly connected balanced bipartite directed graph of order 2a≥10 other than a directed cycle. Let x,y be distinct vertices in D. {x,y} dominates a vertex z if x→z and y→z; in this case, we call the pair {x,y} dominating. In this paper we prove:
*If max{d(x),d(y)}≥2a−2 for every dominating pair of vertices {x,y}, then D contains cycles of all lengths 2,4,…,2a−2 or D is isomorphic to a certain digraph of order ten which we specify.
Keywords: Digraphs, pre-Hamiltonian cycles, bipartite balanced digraphs, even pancyclic.
1 Introduction
It is sequel to the paper [13] by the first author. We consider digraphs (directed graphs) in the sense of [4], and use standard graph theoretical terminology and notation (see Section 2 for details). A cycle passing through all the vertices of a digraph is called Hamiltonian. A digraph containing a Hamiltonian cycle is called a Hamiltonian digraph. A digraph D of order n is called pancyclic if it contains cycles of every lengths 3,4,…,n. Various sufficient conditions for a digraph to be Hamiltonian have been given in terms of the vertex degree of the digraph. Here we recall some of them which are due to Ghouila-Houri [16], Nash-Williams [23], Woodall [27], Meyniel [22], Thomassen [25] and Darbinyan [11]. The Meyniel theorem is a generalization Nash-Williams’, Ghouila-Houri’s and Woodall’s theorems.
Bondy suggested (see [9] by Chvátal) the following metaconjecture:
Metaconjecture. *Almost any non-trivial condition of a graph (digraph) which implies that the graph (digraph) is Hamiltonian also implies that the graph (digraph) is pancyclic. (There may be a ”simple” family of exceptional graphs (digraphs)).
In fact various sufficient conditions for a digraph to be Hamiltonian are also sufficient for the digraph to be pancyclic. Namely,
in [20, 24, 10, 12], it was shown that if a digraph D satisfies one of the above mentioned conditions for hamiltonicity digraphs, then the digraph D also is pancyclic (unless some extremal cases which are characterized). For additional information on Hamiltonian and pancyclic digraphs, see, e.g, the book by Bang-Jensen and Gutin [4] and the surveys
[7] by Bermond and Thomassen, [21] by Kühn and Ostus and [17] by Gutin.
Each of aforementioned theorems imposes a degree condition on all vertices (or, on all pairs of nonadjacent vertices). In [5] and [3], it was described a type of sufficient conditions for a digraph to be Hamiltonian, in which a degree condition requires only for some pairs of nonadjacent vertices. Let us recall only the following theorem of them.
Theorem 1.1 (Bang-Jensen, Gutin, H.Li [5]). *Let D be a strongly connected digraph of order n≥2. Suppose that min{d(x),d(y)}≥n−1 and d(x)+d(y)≥2n−1 for any pair of non-adjacent vertices x,y with a common in-neighbour. Then D is Hamiltonian.
A digraph D is called a bipartite digraph if there exists a partition X, Y of its vertex set into two partite sets such that every arc of D has its end-vertices in different partite sets.
It is called balanced if ∣X∣=∣Y∣.
There are analogies results to the Nash-Williams, Ghouila-Houri, Woodall, Meyniel and Thomassen theorems
for balanced bipartite digraphs (see e.g., [2] and the papers cited there).
An analogue of Theorem 1.1 for bipartite digraphs was given by R. Wang [26] and recently a different result was given by Adamus [1].
Theorem 1.2 (R. Wang [26]). Let D be a strongly connected balanced bipartite digraph of order 2a, where a≥1. Suppose that, for every dominating pair of vertices {x,y}, either d(x)≥2a−1 and d(y)≥a+1 or d(y)≥2a−1 and d(x)≥a+1. Then D is Hamiltonian.
Theorem 1.3 (Adamus [1]). Let D be a strongly connected balanced bipartite digraph of order 2a, where a≥3. If d(x)+d(y)≥3a for every pair of vertices {x,y} with a common in-neighbour or a common out-neighbour, then D is Hamiltonian.
Let D be a balanced bipartite digraph of order 2a, where a≥2. For integer k≥0, we say that
D satisfies condition Bk when max{d(x),d(y)}≥2a−2+k for every pair of dominating vertices x and y.
Before stating the next theorems we need to define three digraphs.
Example 1. Let D(10) be a bipartite digraph with partite sets X={x0,x1,x2,x3,x4} and
Y={y0,y1,y2,y3,y4} satisfying the following conditions: The induced subdigraph ⟨{x1,x2,x3,y0,y1}⟩ is a complete bipartite digraph with partite sets {x1,x2,x3} and {y0,y1}; {x1,x2,x3}→{y2,y3,y4}; x4↔y1; x0↔y0 and xi↔yi+1 for all i∈[1,3]. D(10) contains no other arcs.
It is easy to check that the digraph D(10) is strongly connected and satisfies condition B0, but the underlying undirected graph of D(10) is not 2-connected and D(10) has no cycle of length 8. (It follows from the facts that d(x0)=d(x4)=2 and x0 (x4) is on 2-cycle). It is not difficult to check that any digraph obtained from D(10) by adding a new arc whose one end-vertex is x0 or x4 contains no cycle of length eight. Moreover, if to A(D) we add some new arcs of the type yixj, where i∈[2,4] and j∈[1,3], then always we obtain a digraph which does not satisfy condition B0.
Example 2. Let K2,3∗ be a complete bipartite digraph with partite sets {x1,x2} and {y1,y2,y3}. Let H(8) be the bipartite digraph obtained from the digraph K2,3∗ by adding three new vertices x0,y0,x3 and the following new arcs x0y0, y0x0, x0y1, y1x0, x3y3 and y3x3.
It is not difficult to check that the digraph H(8) is strongly connected and satisfies condition B0, but the underlying undirected graph of H(8) is not 2-connected and H(8) has no cycle of length 6. Hence, the bound on order of D in Theorem 3.4 is sharp.
Example 3. Let D(8) be a bipartite digraph with partite sets X={x0,x1,x2,x3} and
Y={y0,y1,y2,y3}, and the arc set A(D(8)) contains exactly the following arcs y0x1, y1x0, x2y3, x3y2 and all the arcs of the following 2-cycles:
xi↔yi, i∈[0,3], y0↔x2, y0↔x3, y1↔x2 and y1↔x3.
It is easy to see that
[TABLE]
and the dominating pairs in D(8) are: {y0,y1}, {y0,y2},{y0,y3},{y1,y2}, {y1,y3}, {x0,x2},
{x0,x3},
{x1,x2}, {x1,x3} and {x2,x3}. Note that every dominating pair satisfies condition B1. Since x0y0x3y2x2 y1x0 is a cycle of length 6 in D(8),
it is not difficult to check that D(8) is strong.
Observe that D(8) is not Hamiltonian. Indeed, if D(8) contains a Hamiltonian cycle, say C, then C would contain the arcs x1y1 and x0y0. Therefore, C must contain the path x1y1x0y0 or the path x0y0x1y1, which is impossible since N−(x0)=N−(x1)={y0,y1}.
For a≥5 Theorem 1.2 is an immediate consequence of the following theorem by the first author [14].
Theorem 1.4 (Darbinyan [14]). Let D be a strongly connected balanced bipartite digraph of order 2a, where a≥4. Suppose that, for every dominating pair of vertices {x,y}, either d(x)≥2a−1 or d(y)≥2a−1. Then either D is Hamiltonian or isomorphic to the digraph D(8).
A balanced bipartite digraph of order 2m is even pancyclic if it contains a cycle of length 2k
for any 2≤k≤m.
A cycle of a balanced bipartite digraph D is called pre-Hamiltonian if it contains all the vertices of D except two.
Characterizations of even pancyclic bipartite tournaments was given in [6] and [28]. A characterization of pancyclic ordinary k-partite (k≥3) tournaments (respectively, pancyclic ordinary complete multipartite digraphs) was established in [18] (respectively, in [19]).
Motivated by the Bondy’s metaconjecture, it is natural to set the following problem:
Problem. *Characterize those digraphs which satisfy the conditions of Theorem 1.2 (or, 1.3 or 1.4) but are not even pancyclic.
In [15], the first author have proved the following Theorems 1.5 and 1.6.
Theorem 1.5 ([15]. *Let D be a strongly connected balanced bipartite digraph of order 2a≥8 other than a directed cycle. If max{d(x),d(y)}≥2a−1 for every dominating pair of vertices {x,y}, then D contains a cycles of all even lengths less than equal 2a or D is isomorphic to the digraph D(8).
Theorem 1.6. ([15]. *Let D be a strongly connected balanced bipartite digraph of order 2a≥8 which contains a cycle of length 2a−2. If max{d(x),d(y)}≥2a−2 for every dominating pair of vertices {x,y}, then for any k, 1≤k≤a−1, D contains a cycle of length 2k.
In view of Theorem 1.6 it seems quite natural to ask whether a balanced bipartite digraph of order 2a, which satisfies condition B0
contains a pre-Hamiltonian
cycle (i.e., a cycle of length 2a−2).
In this paper we prove the following theorems.
Theorem 1.7. *Let D be a strongly connected balanced bipartite digraph of order 2a≥10 other than the directed cycle of length 2a. Suppose that D satisfies condition B0, i.e., max{d(x),d(y)}≥2a−2 for every dominating pair of vertices {x,y}. Then D contains a cycle of lengths 2a−2 unless D is isomorphic to the digraph D(10).
From Theorem 1.6 and 1.7 it follows the following theorem.
Theorem 1.8. *Let D be a balanced bipartite digraph of order 2a≥10 other than the directed cycle of length 2a. Suppose that D satisfies condition B0, i.e., max{d(x),d(y)}≥2a−2 for every dominating pair of vertices {x,y}. Then D contains cycles of all lengths 2,4,…,2a−2 unless D is isomorphic to the digraph D(10).
2 Terminology and Notation
In this paper we consider finite digraphs without loops and multiple arcs. The vertex set and the arc set of a digraph D are denoted
by V(D) and by A(D), respectively. The order of D is the number
of its vertices. For any x,y∈V(D), we also write x→y, if xy∈A(D). If xy∈A(D), then we say that x dominates y or y is an out-neighbour of x, and x is an in-neighbour of y.
The notation
x↔y denotes that x→y and
y→x (x↔y is called a 2-cycle). We denote by a(x,y) the number of arcs with end-vertices x and y.
For disjoint subsets A and B of V(D) we define A(A→B)
as the set {xy∈A(D)∣x∈A,y∈B} and A(A,B)=A(A→B)∪A(B→A). If x∈V(D)
and A={x} we write x instead of {x}. If A and B are two disjoint subsets of V(D) such that every
vertex of A dominates every vertex of B, then we say that A dominates B, denoted by A→B.
A↦B means that A→B and there is no arc from B to A.
We let N+(x), N−(x) denote the set of out-neighbours, respectively the set of in-neighbours of a vertex x in a digraph D. If A⊆V(D), then N+(x,A)=A∩N+(x) and N−(x,A)=A∩N−(x).
The out-degree of x is d+(x)=∣N+(x)∣ and d−(x)=∣N−(x)∣ is the in-degree of x. Similarly, d+(x,A)=∣N+(x,A)∣ and d−(x,A)=∣N−(x,A)∣. The degree of a vertex x in D is defined as d(x)=d+(x)+d−(x) (similarly, d(x,A)=d+(x,A)+d−(x,A)).
For integers a and b, a≤b, let [a,b] denote the set of
all integers which are not less than a and are not greater than
b.
The subdigraph of D induced by a subset A of V(D) is denoted by D⟨A⟩ or ⟨A⟩ for brevity.
The path (respectively, the cycle) consisting of the distinct vertices x1,x2,…,xm ( m≥2) and the arcs xixi+1, i∈[1,m−1] (respectively, xixi+1, i∈[1,m−1], and xmx1), is denoted by x1x2⋯xm (respectively, x1x2⋯xmx1).
We say that x1x2⋯xm is a path from x1 to xm or is an (x1,xm)-path.
If P is a path containing a subpath from x to y we let P[x,y] denote that subpath.
Similarly, if C is a cycle containing vertices x and y, C[x,y] denotes the subpath of C from x to y.
A digraph D is strongly connected (or, just, strong) if there exists an (x,y)-path in D for every ordered pair of distinct vertices x,y of D.
Given a vertex x of a path P or a cycle C, we denote by x+ (respectively, by x−) the successor (respectively, the predecessor) of x (on P or C), and in case of ambiguity, we precise P or C as a subscript (that is xP+ …).
Two distinct vertices x and y are adjacent if xy∈A(D) or yx∈A(D) (or both).
Let C be a non-Hamiltonian cycle in a digraph D. An (x,y)-path P is a C-bypass if ∣V(P)∣≥3, x=y and V(P)∩V(C)={x,y}. The length of the path C[x,y] is the gap of P with respect to C.
The underlying undirected graph of a digraph D is the unique graph that contains an edge xy if x→y or y→x (or both).
3 Preliminaries
Lemma 3.1 (Bypass Lemma 3.17, Bondy [8]). *Let D be a strongly connected digraph, and let H be a non-trivial proper subdigraph of D. If the underlying undirected graph of D is 2-connected, then D contains a H-bypass.
Remark. One can prove Bypass Lemma using the proof of Theorem 5.4.2 [4].
Lemma 3.2 ([14]). *Let D be a strongly connected balanced bipartite digraph of order 2a≥8 other than the directed cycle of length 2a. If D satisfies condition B0, then D contains a non-Hamiltonian cycle of length at least 4.
Lemma 3.3 ([15]). *Let D be a strongly connected balanced bipartite digraph of order 2a≥8 with partite sets X and Y.
Assume that D satisfies condition B0. Let C=x1y1x2y2…xkykx1 be a longest non-Hamiltonian cycle in D, where k≥2, xi∈X and yi∈Y, and P be a C-bypass. If the gap of P with respect to C is equal to one, then k=a−1, i.e., the longest non-Hamiltonian cycle in D has length 2a−2.
Theorem 3.4 ([13]). *Let D be a strongly connected balanced bipartite digraph of order 2a≥10. Assume that D satisfies condition B0. Then either the underlying undirected graph of D is 2-connected or D contains a cycle of length 2a−2 unless D is isomorphic to the digraph D(10).
4 Proof of the main result
**Proof of Theorem 1.7. ** Suppose, on the contrary, that a digraph D is not a directed cycle and
satisfies the conditions of the theorem, but contains no cycle of length 2a−2. Let C=x1y1x2y2…xmymx1 be a longest non-Hamiltonian cycle in D, where xi∈X and yi∈Y (all subscripts of the vertices xi and yi are taken modulo m, i.e., xm+i=xi and ym+i=yi).
By Lemma 3.2, D contains a non-Hamiltonian cycle of length at least 4, i.e., 2≤m≤a−2. By Theorem 3.4, the underlying undirected graph of D
is 2-connected. Therefore,
by Bypass Lemma, D contains a C-bypass. We choose a cycle C and a C-bypass P:=xu1…usy such that
(i) C is a longest non-Hamiltonian cycle in D;
(ii) the gap of C-bypass P with respect to C is minimum subject to (i);
(iii) the length of P is minimum subject to (i) and (ii).
Without loss of generality, we assume that x:=x1. Let R:=V(D)∖V(C), P1:=P[u1,us] and C′:=V(C[y1,yC−]). Note that ∣R∣≥4. From Lemma 3.3 it follows that ∣V(C[x,y])∣≥3. Then ∣C′∣≥s since C is a longest non-Hamiltonian cycle in D.
We first prove that ∣V(P1)∣=1, i.e., s=1.
Proof of s=1. Suppose, on the contrary, that s≥2. By the minimality of the gap ∣C′∣+1, we have
[TABLE]
It is not difficult to see that s≤3. Indeed, if s≥4, then from (1) it follows that d(yC−)≤2a−4 and d(us)≤2a−4, since each of P1 and C′ contains at least two vertices from each partite set. So, we have a contradiction, since {us,yC−}→y. Therefore, s=3 or s=2.
First consider the case s=3, i.e., P1=u1u2u3. From our assumption that x=x1 it follows that u1,u3∈Y and y∈X. Therefore, u2∈X and yC−∈Y. Note that x2∈C′ since ∣C′∣≥3. By the minimality of the gap ∣C′∣+1, we have
[TABLE]
Therefore, d(x2)≤2a−4. From the minimality of the gap ∣C′∣+1 and P1 it follows that x1u3∈/A(D). Therefore, by (2), d(u3)≤2a−3. This together with condition B0 and {yC−,u3}→y imply that d(yC−)≥2a−2. Therefore, by (2), the vertex yC− and every vertex of X∖{u2} form a 2-cycle, i.e., yC−↔X∖{u2}. In particular, yC−↔{x2,z}, where z is an arbitrary vertex of X∩R∖{u2}. By the minimality of the gap, we have a(z,u3)=d−(z,{u1})=0. Therefore, d(z)≤2a−3. This and d(x2)≤2a−4 contradict condition B0 since {z,x2}→yC−.
Now consider the case s=2, i.e., P1=u1u2.
Notice that u1,y∈Y and yC−∈X since x1∈X. It is easy to see that ∣C′∣ is even.
Assume first that ∣C′∣≥4. By the minimality of the gap ∣C′∣+1,
[TABLE]
where yC−− denotes the predecessor of yC− on C.
Therefore, d(u2)≤2a−4 and d(yC−)≤2a−2. Since {u2,yC−}→y, i.e., {u2,yC−} is a dominating pair, by condition B0 we have, d(yC−)=2a−2. This together with the second equality of (3) imply that yC− and every vertex of Y∖{u1} form a 2-cycle.
In particular, y→yC−→y1 and
yC−↔z, where z is an arbitrary vertex of Y∩R∖{u1}.
From the minimality of the gap ∣C′∣+1 it follows that a(z,u2)=d−(z,{x2})=0. Hence, d(z)≤2a−3. Now we
consider the vertex yC−−. It is easy to see that yC−−yC+∈/A(D) (for otherwise, the cycle x1u1u2yyC−y1…yC−−yC+…x1 is longer than C). From this and the first equality of (3) it follows that d(yC−−)≤2a−3.
Thus, we have {z,yC−−}→yC− and max{d(z),d(yC−−)}≤2a−3, which contradict condition
B0.
Assume then that ∣C′∣=2, i.e., C′={y1,x2}. Then y=y2. By the minimality of the gap ∣C′∣+1,
[TABLE]
i.e., the vertices u1 and x2 (respectively, u2 and y1) are not adjacent.
Therefore,
[TABLE]
Without loss of generality, we may assume that d(x2)=2a−2, since {u2,x2}→y2 (for otherwise, d(u2)=2a−2 and we will consider the cycle x1u1u2y2…ymx1).
Since u1 and x2 are non-adjacent and d(x2)=2a−2, it follows that x2 and every vertex of Y∖{u1} form a 2-cycle, i.e., x2↔Y∖{u1}. In particular, y2→x2→y1.
Let z be an arbitrary vertex in Y∩R∖{u1}. By the minimality of the gap ∣C′∣+1, a(z,u2)=0 and x1z∈/A(D). Hence, d(z)≤2a−3.
If y1→x3 (possibly, x3=x1), then, since y2→x2→y1, we see that x1u1u2y2x2y1x3…x1 is a cycle of length ∣C∣+2, a contradiction. We may therefore assume that y1x3∈/A(D).
This together with a(y1,u2)=0 (by (4)) gives d(y1)≤2a−3. Thus we have that {y1,z}→x2 and
max{d(y1),d(z)}≤2a−3, which contradict condition B0. This contradiction completes the proof of s=1.
From s=1 it follows that u1∈Y and y∈X since x1∈X. Without loss of generality, we may assume that y:=xr. From now on, let y:=u1.
Now we divide the proof of the theorem into two parts: ∣C′∣=1 and ∣C′∣≥2.
Part I. ∣C′∣=1, i.e., r=2 and x1→y→x2.
By condition B0,
max{d(y),d(y1)}≥2a−2 since {y,y1}→x2. Without loss of generality, assume that d(y)≥2a−2. For this part we first prove Claims 1-5 below.
Claim 1. If x∈R∩X and x↔y, then d(x)≤2a−3 and d(x1)≥2a−2.
Proof of Claim 1. Assume that x∈R∩X and x↔y, but d(x)≥2a−2.
It is easy to see that a(x,y1)=0 since C is a longest non-Hamiltonian cycle in D. This and d(x)≥2a−2
imply that the vertex x and every vertex of Y∖{y1} form a 2-cycle.
In particular, x↔{y0,y2,ym} (possibly, y2=ym), where y0 is an arbitrary vertex of Y∩R∖{y}.
Using this, it is easy to check that
[TABLE]
since C is a longest non-Hamiltonian cycle in D. From the last equalities we have d(x1)≤2a−3, and y↔x0, where x0 is an arbitrary vertex of X∩R∖{x}, since d(y)≥2a−2. Since C is a longest non-Hamiltonian cycle in D and since x↔{y0,y2,ym}, it follows
that d(x0,{y1,y0})=0. Therefore, d(x0)≤2a−4, which contradicts condition B0 since
max{d(x1),d(x0)}≤2a−3 and {x0,x1}→y. This contradiction proves that d(x)≤2a−3. From this and condition
B0 it follows that d(x1)≥2a−2 since {x,x1}→y. Claim 1 is proved.
Claim 2 follows immediately from Claim 1 and condition B0.
Claim 2. There are no two distinct vertices x,x0∈R∩X such that x↔y and
x0↔y.
Claim 3. If y↔x for some x∈R∩X, then a(y,z)=1
for all
z∈R∩X∖{x}.
Proof of Claim 3. Let x↔y for some x∈R∩X. Then Claim 2 implies that a(y,z)≤1 for all
z∈R∩X∖{x}.
We want to show that a(y,z)=1 for all
z∈R∩X∖{x}. Assume that this is not the case. Then a(y,x0)=0 for some
x0∈R∩X∖{x}. This together with Claims 1 and 2 thus imply that ∣R∣=4 since d(y)≥2a−2 by our assumption.
Then R∩X={x,x0}.
Let R∩Y={y,y0}.
Since a(y,x0)=0 and d(y)≥2a−2, it follows that
[TABLE]
Since C is a longest non-Hamiltonian cycle in D, (5) implies that
d(x,{y1,y2,…,ym})=0,
which in turn implies that the vertices x and y0 are adjacent and d(x)≤2a−4 since UG(D) is 2-connected
and m≥2. By the assumption of Claim 3 and (5), {x,xi}→y for all i∈[1,m]. Now using
condition B0, we obtain
[TABLE]
The remainder of the proof of Claim 3 is divided into two subcases depending on the value of a(x,y0).
Case 1. a(x,y0)=2, i.e., x↔y0.
Since C is a longest non-Hamiltonian cycle in D, from (5) and y↔x, x↔y0 it follows that
[TABLE]
Therefore, since UG(D) is 2-connected, the vertices y0 and x0 are adjacent. Using (5), it is not difficult to see that:
If x0→y0, then d−(x0,{y1,y2,…,ym})=0, and
if y0→x0, then d+(x0,{y1,y2,…,ym})=0.
In both cases we have a(yi,x0)≤1 for all i∈[1,m]. Together with a(yi,x)=0 this implies that d(yi)≤2a−3. On the other hand,
from (6) and (7) it follows that every vertex xi, i∈[1,m], and every vertex of Y∖{y0} form a 2-cycle.
In particular, {y1,y2}→x2, which contradicts condition B0, since max{d(y1),d(y2)}≤2a−3.
Case 2. a(x,y0)=1, i.e., y0↦x or x↦y0.
Let y0↦x. Since C is a longest non-Hamiltonian cycle in D and (5), we have
[TABLE]
Hence x0→y0 since D is strong and y0↦x. Now using (5), we obtain
[TABLE]
The last two equalities together with a(y,x0)=0 and y0↦x imply that
[TABLE]
which contradicts that D is strong.
Let now x↦y0. Again using (5), it is easy to see that
d+(y0,{x,x1,x2,…,xm})=0. Therefore, y0→x0 since D is strong. Together with (5) this implies that
d+(x0,{y1,y2,…,ym,y})=0.
Therefore x0→y0 since D is strong, and d(x0)≤2a−4 since m≥2. Thus, d(x)≤2a−3 (Claim 1) and d(x0)≤2a−4, which contradicts condition B0 since {x,x0}→y0. This contradiction completes the proof of Claim 3.
Claim 4. If y↔x for some x∈R∩X, then d−(y,R∩X∖{x})=0.
Proof of Claim 4. Assume that the claim is not true, i.e., there exist vertices x∈R∩X and u∈R∩X∖{x} such that y↔x and
u→y. Then, by Claim 2, yu∈/A(D). Notice that {x,u}→y. Since d(x)≤2a−3 (Claim 1), condition B0 implies that d(u)≥2a−2. It is clear that y1u∈/A(D) (if y1→u, then x1y1uyx2y3…ymx1 is a cycle of length 2m+2, a contradiction). Using this, yu∈/A(D) and d(u)≥2a−2 we conclude that uy1∈A(D), d(u)=2a−2 and the vertex u together with every vertex of Y∖{y,y1} forms a 2-cycle.
In particular, u↔{y2,ym,v}, where v∈Y∩R∖{y} (possibly, y2=ym).
Using this,
it is not difficult to show that a(x1,v)=0 and yx1∈/A(D).
Indeed, if x1→v, then x1vuy1x2…ymx1 is a cycle of length 2m+2;
if v→x1, then
ymuvx1y1… xmym is a cycle of length 2m+2;
if y→x1, then ymuyx1y1…xmym is a cycle of length 2m+2. In each case we obtain a contradiction. Hence, a(x1,v)=0 and yx1∈/A(D). From this it follows that d(x1)≤2a−3, which contradicts that d(x1)≥2a−2 (Claim 1). Claim 4 is proved.
Claim 5. There is no x∈X∩R such that y↔x, i.e., in subdigraph D⟨R⟩ through the vertex y there is no cycle of length two.
Proof of Claim 5. Assume that the claim is not true, i.e., there exists a vertex x∈X∩R such that
y↔x.
By Claims 1, 3 and 4 we have
[TABLE]
Case 1. x2y∈/D.
Then from the last expression of (8) and d(y)≥2a−2 we conclude that ∣R∣=4, i.e., m=a−2≥3, and the vertex y and every vertex of {x1,x2,x3,x4,…,xm}∖{x2} form a 2-cycle, i.e.,
[TABLE]
Put R={y,x,y0,x0}, where x0∈X and y0∈Y. From (8) it follows that
y↦x0. Now using (9), it is not difficult to see that
[TABLE]
Indeed, if yi→x and i∈[1,m] (respectively, x→yj and j∈[1,m]∖{2}), then
xiyixyxi+1…xi (respectively, xjyxyj…yj−1xj)
is a cycle of length 2a−2, a contradiction.
Similarly,
[TABLE]
In particular, (10) implies that
[TABLE]
i.e., the vertices x and yi, i∈/{0,2}, are not adjacent. From this, (11) and y↦x0 it follows that
[TABLE]
Using (12) and condition B0, we obtain that for all u∈X and v∈Y the following holds
[TABLE]
Now we divide this case into four subcases.
Subcase 1.1. x↔y0.
Since x↔y and x↔y0, using (9) and the fact that m≥3, it is not difficult to check that the vertices x1 and y0 are not adjacent (for otherwise, D would contain a cycle of length 2a−2, a contradiction). Together with the first inequality of (13) (when u=x1) this implies that d(x1)≤2a−3, which contradicts Claim 1.
Subcase 1.2. x↦y0.
Then, by the second inequality of (13) when v=y0, x0y0∈/A(D). This together with x0y∈/A(D) and (11) imply that x0→y2 since D is strong. It is easy to see that y0x2∈/A(D) (for otherwise, x1yxy0x2y2…ymx1 is a cycle of length
2a−2, a contradiction). Combining this with x2y∈/A(D) and
d−(x2,{y1,ym})≤1 (by (13)) we obtain d(x2)≤2a−3. Thus, d(x2)≤2a−3 and d(x0)≤2a−3 (by (12))
and {x0,x2}→y2, which contradict condition B0.
Subcase 1.3. y0↦x.
If xi→y0 and i∈[1,m], then using (9) we obtain that the cycle xiy0xyxi+1…xi has length 2a−2, which is a contradiction. We may therefore assume that d−(y0,{x,x1,x2,…,xm})=0. Hence, x0→y0
since D is strong and y0↦x.
Since xm→y→x2 and C is a longest non-Hamiltonian cycle in D, it follows that the vertices x1 and y0 are not adjacent. Combing this with d−(x1,{y1,ym})≤1 (by (13)) we obtain that d(x1)≤2a−3, a contradiction to Claim 1.
Subcase 1.4. The vertices x and y0 are not adjacent.
Since the underlying undirected graph of D is 2-connected, from a(x,y0)=0 and (10) it follows that x→y2. Together with (13) this imply that x0y2∈/A(D). Therefore, by (11), we have that
d+(x0,{y,y1,y2,…,ym})=0. Hence, x0→y0 since D is strong. If y0→x2, then x1yx0y0x2y2…ymx1 is a cycle of length 2a−2, which is a contradiction. We may therefore assume that y0x2∈/A(D). Combining this with x2y∈/A(D) and (13) we obtain that d(x2)≤2a−3, which contradicts condition B0 since {x2,x}→y2 and d(x)≤2a−3 (Claim 1).
The discussion of Case 1 is completed.
Case 2. x2→y.
Since y↦X∩R∖{x} (by (8)) and d(y)≥2a−2, it follows that the vertices y and xi, where
i∈[1,m], are adjacent and y→{x1,x2,…,xm} or {x1,x2,…,xm}→y.
Therefore, without loss of generality, we may assume that
[TABLE]
(for otherwise we will have the considered Case 1). Let x0 be an arbitrary vertex in X∩R∖{x}. Then by (8) we have y↦x0. It is clear that
[TABLE]
This implies that d(x0)≤2a−3.
Since D is strong and (15), there is a vertex y0∈Y∩R∖{y} such that x0→y0.
Now, since y→x0→y0 and xi→y for all i∈[1,m] (by (14)), we conclude that
d+(y0,{x1,x2,…,xm})=0 (for otherwise, for some i∈[1,m], y0→xi and
xi−1yx0y0xi…yi−2xi−1 is a cycle of length 2m+2, a contradiction).
The last equality together with
d+(x0,{y,y1,y2,…,ym})=0 (by (15))
imply that y0→x since D is strong (for otherwise,
A({x0,y0}→V(D)∖{x0,y0})=∅, which contradicts that D is strong).
Then using the facts that
y→x0→y0 and y0→x→y,
it is not difficult to show that x1 and y0 are not adjacent.
Indeed, by (14) we have that: If x1→y0, then
x1y0xyx2y2…ymx1 is a cycle of length 2m+2; and if y0→x1, then xmyx0y0x1y1… ym−1xm is a cycle of length 2m+2, in both cases we have a contradiction. So, x1 and y0 are not adjacent.
Together with d(x1)≥2a−2 (Claim 1) this implies that y1→x1. On the other hand, since d(x,{y1,ym})=d+(x0,{y1,ym})=0, we have that
max{d(y1),d(ym)}≤2a−3, which contradicts condition B0, because of {ym,y1}→x1. Claim 5 is proved.
Now we can finish the discussion of Part I.
From Claim 5 it follows that in D⟨R⟩ there is no cycle of length two through the vertex y. Then, since d(y)≥2a−2 and ∣R∣≥4, it follows that ∣R∣=4.
Put X∩R={x,x0} and Y∩R={y,y0}. Then a(y,x)=a(y,x0)=1 and the vertex y and every vertex of X∩C form a 2-cycle, i.e.,
[TABLE]
First consider the case d+(y,{x,x0})≥1. Assume, without loss of generality, that y↦x. Then, by (16),
d+(x,{y1,y2,…,ym})=0.
Together with xy∈/A(D) this implies that x→y0 since D is strong.
Therefore
d(x)≤2a−3 since ∣Y∩C∣≥2.
By (16), it is clear that
d+(y0,{x1,x2,…,xm})=0.
If y→x0, then analogously we obtain that
d+(x0,{y,y1,y2,…,ym})=0 and x0→y0, d(x0)≤2a−3, which contradicts condition B0 since
max{d(x),d(x0)}≤2a−3 and {x,x0}→y0.
We may assume therefore that yx0∈/A(D).
Then x0→y (by a(y,x0)=1),
d−(x0,{y,y1,y2,…,ym})=0 (by (16)) and hence, y0→x0 since D is strong. Now it is not difficult to show that
[TABLE]
Therefore d(x0)≤2a−3 and d(yi)≤2a−3. Since for all i∈[1,m], {xi,x0}→y and d(x0)≤2a−3,
from condition B0 it follows that d(xi)≥2a−2 for all i∈[1,m]. This together with a(xi,y0)=0 (by (16) and the last equalities) imply that yi↔xi.
Thus, {yi−1,yi}→xi and
max{d(yi−1),d(yi)}≤2a−3, which is a contradiction.
Now consider the case d+(y,{x,x0})=0. Then {x,x0}→y, because of a(y,x)=a(y,x0)=1, i.e., {x,x0} is a dominating pair. It is clear that
d−(x,{y1,y2})=d−(x0,{y1,y2})=0. This together with d+(y,{x0,x})=0 imply that
max{d(x),d(x0)}≤2a−3, which is a contradiction because of {x,x0}→y. This completes the discussion of the part ∣C′∣=1.
Part 2. ∣C′∣≥2.
Then ∣C′∣≥3 since ∣C′∣ is odd.
For this part we first will prove Claims 6-8.
Claim 6. If ∣C′∣≥3, then the following holds:
(a). d(y)≤2a−3 and d(yr−1)≥2a−2;
(b). There is no x∈X∩R such that x↔yr−1,
i.e., a(yr−1,x)≤1 for all x∈X∩R;
(c). a(yr−1,x)=1 for all x∈X∩R, ∣R∣=4, d(yr−1)=2a−2 and the vertex yr−1 together with every vertex of X∩V(C) forms a 2-cycle. In particular, xr↔yr−1 and yr−1↔x2;
(d). d−(yr−1,{x,x0})=0 and yr−1↦{x,x0};
(e). max{d(x0),d(x)}≤2a−3 and d−(v,{x0,x})≤1 for all v∈Y;
(f). ∣C′∣=3, i.e., r=3.
Proof of Claim 6.
(a). Suppose on the contrary, that d(y)≥2a−2. Then, since d(y,C′)=0, we have that r=3 and the vertex y together with every vertex of X∖{x2} forms a 2-cycle. In particular, y↔x1 and y↔x3.
It follows that for some i∈[3,m], xi→y→xi+1, which contradicts that C-bypass P has the minimum gap among the gaps of all C-bypasses. Therefore, d(y)≤2a−3. Now, since {y,yr−1}→xr, from condition B0 it follows that
d(yr−1)≥2a−2.
(b).
Suppose that Claim 6(b) is falls, i.e., there is a vertex x∈X∩R such that x↔yr−1. From the minimality of the gap ∣C′∣+1 it follows that
[TABLE]
From this we have d(x)≤2a−4. This together with condition B0 and {x,xr−1}→yr−1 imply that d(xr−1)≥2a−2.
Therefore, since a(xr−1,y)=0, it follows that
(i) *the vertex xr−1 and every vertex of Y∖{y} form a 2-cycle.
Let y′ be an arbitrary vertex in Y∩R∖{y}. By (i), xr−1↔y′.
By the minimality of the gap ∣C′∣+1, d+(y′,{x,xr})=d−(y′,{x1})=0.
Therefore, d(y′)≤2a−3. Since {y′,yr−2}→xr−1, condition B0 implies that
[TABLE]
First consider the case r≥4. Then xr−2∈C′ and it is not difficult to see that
xr−2y′∈/A(D) and yr−2x∈/A(D). Now, since a(xr−2,y)=0, we have
d(xr−2)≤2a−3. Using condition B0 and the fact that d(x)≤2a−4, we conclude that xyr−2∈/A(D).
Therefore, by (17), x and yr−2 are not adjacent.
This together with d(yr−2)≥2a−2 (by (18)) imply that yr−2↔x0, where x0 is an arbitrary vertex in R∩X∖{x}. By the minimality of the gap ∣C′∣+1, we have x0yr−1∈/A(D), yr−3x0∈/A(D) and x0y′∈/A(D).
Therefore, d(x0)≤2a−3, which contradicts condition B0, since {x0,xr−2}→yr−2 and
d(xr−2)≤2a−3.
Now consider the case r=3. By (i), x2↔y3 and x2↔y2. If
y2→x4 (possibly, x4=x1) (respectively, x3→y2), then
the cycle
Q:=x1yx3y3x2y2x4…ymx1 (respectively, Q:=x1yx3y2x2y3x4…ymx1) has length 2m, the vertex y1 is not on this cycle and x1→y1→x2 is a Q-bypass whose gap with respect to Q is equal to 4 and d(y1)≥2a−2, (by (18) since r−2=1), which contradicts Claim 6(a).
We may therefore assume that y2x4∈/A(D) and x3y2∈/A(D).
Combining this with d(y2)≥2a−2 (by Claim 6(a) and r=3) we obtain that d(y2)=2a−2 and y2 together with every vertex of X∖{x3,x4} forms a 2-cycle. In particular,
y2↔x0.
On the other hand, from the minimality of the gap ∣C′∣+1 it follows that x0,y are not adjacent and y1x0∈/A(D). Therefore,
d(x0)≤2a−3, which is a contradiction, since d(x)≤2a−3 and {x,x0}→y2.
(c). Claim 6(c) is an immediate corollary of Claims 6(a) and 6(b).
From now on, we assume that X∩R={x,x0} and Y∩R={y,y0}.
(d). Suppose, on the contrary, that there exists a vertex in {x,x0}, say x, such that x→yr−1.
From the minimality of the gap ∣C′∣+1 it follows that d−(x,{y,y1})=0. This together with yr−1x∈/A(D) (Claim 6(b)) imply that d(x)≤2a−3.
Therefore, using Claim 6(c) and condition B0, we obtain that d(x2)≥2a−2 and d(xr−1)≥2a−2 (possibly, x2=xr−1) since {x,x2,xr−1}→yr−1. Together with d(y,{x2,xr−1})=0 this implies that xr−1→yr and
x2↔y0.
Because of gap minimality,
d−(y0,{x1})=d+(y0,{x,xr})=0. Therefore, d(y0)≤2a−3, and hence, by condition B0, d(y1)≥2a−2 since {y1,y0}→x2.
Since xr→yr−1 and yr−1→x2 (Claim 6(c)) and xr−1→yr, the cycle Q:=x1yxryr−1x2…xr−1yr…ymx1 has length equal to 2a−4.
Observe that y1∈/V(Q) and x1→y1→x2 is a Q-bypass, whose gap with respect to Q is equal to 4, but d(y1)≥2a−2, which contradicts Claim 6(a).
(e). By Claim 6(d), yr−1↦{x,x0}. Therefore, because of gap minimality, we have
[TABLE]
Therefore, max{d(x0),d(x)}≤2a−3 and, by condition B0,
the vertices x, x0 does not form a dominating pair, i.e., d−(v,{x0,x})≤1 for all v∈Y.
(f). Suppose, on the contrary, that is ∣C′∣>3. Then ∣C′∣≥5, i.e., r≥4, since ∣C′∣ is odd.
By Claim 6(c), xi↔yr−1 for all i∈[1,m]. This together with condition B0 imply that
d(xi)≥2a−2 for all i∈[1,m] maybe except one. In particular,
d(xr−1)≥2a−2 or d(xr−2)≥2a−2.
Fist consider the case d(xr−1)≥2a−2. From this and a(y,xr−1)=0 it follows that xr−1↔yr.
Using this and the fact that
xr→yr−1→x2 (Claim 6(c)), we see that the cycle
Q:=x1yxryr−1x2y2…xr−1yr xr+1…x1 has length equal to 2a−4. Observe that x1→y1→x2 is a Q-bypass whose gap with respect to Q is equal to 4, which contradicts the choice of the cycle C and C-bypass P.
Now consider the case d(xr−1)≤2a−3. Then d(xr−2)≥2a−2. Observe that the vertex xr−2 and every vertex of Y other than y form a 2-cycle (since xr−2 and y are not adjacent), in particular, y0↔xr−2. It is easy to see that xr−2=x2, i.e., r=4. Indeed, if r−2≥3, then xr−3∈C′ and xr−3y0∈/A(D) because of the minimality of the gap ∣C′∣+1 and y0↔xr−2.
This together with a(y,xr−3)=0 gives d(xr−3)≤2a−3. Thus we have that the vertices xr−3 and xr−1 both have degree less than 2a−2, which contradicts the fact that at most one vertex xi, i∈[1,m] maybe has degree less that 2a−2.
Thus, r=4. By the above observations, y4→x2 and y3→x5.
Therefore, Q:=x1yx4y4x2y2x3y3x5…x1 is a cycle of length 2a−4. Notice that y1∈/V(Q) and the Q-bypass x1→y1→x2 has gap with respect to Q equal to 4. This contradicts the choice of C and C-bypass P.
From Claims 6(c), 6(d) and 6(f)1 it follows that
(ii) *If ∣C′∣≥3, then ∣C′∣=3, ∣R∣=4, y2↦{x,x0}, d(y)≤2a−3,
d(y2)=2a−2 and the vertex y2 together with every vertex of X∩V(C) forms a 2-cycle.
Claim 7. If ∣C′∣=3, then d(y0)≤2a−3, (recall that {y0}=Y∩R∖{y}).
Proof of Claim 7. By Claim 6((e), d−(y0,{x,x0})≤1. Therefore, if x2 and y0 are not adjacent, then
d(y0)≤2a−3. We may therefore assume that x2 and y0 are adjacent. Then y0→x2 or x2→y0.
It is easy to see that if y0→x2, then x1y0∈/A(D); and if x2→y0, then y0x3∈/A(D). Therefore, d(y0,{x1,x2,x3})≤4. This and d−(y0,{x0,x})≤1 imply that
d(y0)≤2a−3.
Combining Claims 6(a), 6(e) and 7 we obtain that if ∣C′∣=3, then
[TABLE]
in particular, by condition B0, we have
[TABLE]
Claim 8. If ∣C′∣=3, then the following holds:
(a). d(x2)≤2a−3 and d(x3)≥2a−2;
(b). d(y1)≥2a−2;
(c). If x→yk or x0→yk, where k∈[3,m],
then xky1∈/A(D).
Proof of Claim 8.
(a).
Suppose, on the contrary, that ∣C′∣=3 and d(x2)≥2a−2. This and a(x2,y)=0 imply that d(x2)=2a−2 and x2 together with every vertex of Y∖{y} form a 2-cycle. In particular, x2↔y0 and
x2↔y3. Since d(y0)≤2a−3 (by (19)) and {y1,y0}→x2, from condition B0 it follows that d(y1)≥2a−2.
On the other hand, using the facts that x3→y2→x2 (Claim 6(c)) and x2→y3 we see that Q:=x1yx3y2x2y3…x1 is a cycle of length 2a−4.
Notice that x1→y1→x2 is a Q-bypass, whose the gap with respect to Q is equal to 4. This contradicts the minimality of the gap ∣C′∣+1 or Claim 6(a) since d(y1)≥2a−2. Therefore, d(x2)≤2a−3. Together with condition B0 this implies that d(x3)≥2a−2 since {x2,x3}→y2.
(b). Suppose, on the contrary, that ∣C′∣=3 and d(y1)≤2a−3. By condition B0, y1x3∈/A(D) since
d(y)≤2a−3 (Claim 6(a)) and y→x3.
From (20) we have that y0x3∈/A(D) since yx3∈A(D).
Thus, the arcs y1x3 and y0x3 are not in A(D).
This together with d(x3)≥2a−2 (Claim 8(a)) imply that x3→y1 and y3→x3.
By (ii), y2→x4 and hence,
Q:=x1yx3y1x2y2x4…x1 is a cycle of length 2a−4, which does not contain the vertices x0,y0,x and y3. The path x3y3x4 is a Q-bypass whose the gap with respect to Q is equal to 4.
Therefore, by the minimality of the gap ∣C′∣+1 and Claim 6(a), d(y3)≤2a−3 but this is a contradiction since
{y,y3}→x3 and
d(y)≤2a−3 (Claim 6(a)).
(c). Suppose that the claim is not true. Without loss of generality, assume that for some k∈[3,m], x→yk and
xk→y1. Then, by (ii) we have that y2→{x,x0} and hence, the cycle
x1yx3…xky1x2y2xyk …ymx1 is a cycle of length 2a−2, a contradiction.
Now we are ready to complete the proof of Theorem 1.7.
Combining (19) and Claim 8(a), we obtain
[TABLE]
This and condition B0 imply
[TABLE]
Therefore, since d(y1)≥2a−2 (Claim 8(b)), the vertex y1 and every vertex of X∖{x,x0,x2} form a 2-cycle, i.e.,
[TABLE]
Now using Claim 8(c), we obtain
A({x,x0}→{y2,y3,…,ym})=∅.
From y2↦{x,x0} (Claim 6(d)) and the minimality of the gap ∣C′∣+1 it follows that d−(y,{x,x0})=0. The last two equalities imply that
[TABLE]
By (20), in particular, we have
[TABLE]
Since D is strong, from (23) and (24) it follows that x→y0 or x0→y0. Again using (24), we obtain that if
x→y0, then x0y0∈/A(D) and x0→y1; if x0→y0, then xy0∈/A(D) and x→y1.
Because of the symmetry between the vertices x and x0, we can assume that x0→y0, x→y1 and x0y1∈/A(D), xy0∈/A(D).
It is not difficult to show that x2 and every vertex yi with i∈[3,m] are not adjacent. Indeed, if
yi→x2, then, by (22), y1→xi+1 and hence, x1yx3y3…yix2y2xy1xi+1…x1 is a cycle of length 2a−2;
if
x2→yi, then, by (ii), xi→y2 and hence, x1yx3y3…xiy2xy1x2yi…ymx1 is a cycle of length 2a−2. Thus, in both cases we have a contradiction.
Therefore, a(yi,x2)=0 for all i∈[3,m]. This and (20) imply that d(yi)≤2a−3 for all i∈[3,m].
From y→x3, d(y)≤2a−3 (Claim 6(a)) and condition B0 it follows that
d−(x3,{y0,y3,y4,…ym})=0.
Now frome d(x3)≥2a−2 (Claim 8(a)), we have that m=3, i.e., a=5.
Since x→y1, x0→y0 and (21), it follows that d+(x2,{y0,y1})=0. Because of d(y1)≥2a−2 (Claim 8(b)) and d−(y1,{x0,x2})=0 we have y1→x0, x3↔y1 and y1→x1.
From this it follows that y0x2∈/A(D) (for otherwise, y1x0y0x2y2…x1y1 is a cycle of length 2a−2, a contradiction). Therefore, a(x2,y0)=0. If y0→x1, then the cycle x1yx3y1x2y2x0y0x1 is a cycle of length 2a−2=8, a contradiction. Therefore, y0x1∈/A(D). So, we have d+(y0,{x1,x2,x3})=0. Then d+(y0,{x,x0})≥1 since D is strong. It is easy to see that y0x∈/A(D) (for otherwise, x1yx3y2x0y0xy1x1 is a cycle of length 8, a contradiction. Therefore, d+(y0,{x1,x2,x3,x})=0.
On the other hand, from (23) and x0y1∈/A(D) we have N+(x0)={y0}. Now it is not difficult to see that there is no path from x0 to any vertex of V(C) since N+(y0)={x0}, which contradicts that D is strong. So, the discussion of the case ∣C′∣≥3 is completed. Theorem 1.7 is proved.
5 Concluding remarks
In view of Theorem 1.3 it is natural to set the following problem.
Problem. Characterize those strongly connected balanced bipartite digraphs of order 2a≥6 in which
d(x)+d(y)≥3a for every pair of vertices x, y with a common in-neighbour or a common out-neighbour but are not even pancyclic.