The Prescribed-Vertex Semidegree Threshold for Directed -Cycles in Oriented Graphs
Abstract.
For every , we prove that every oriented graph on vertices whose minimum semidegree satisfies
contains a directed cycle of length through every vertex. The semidegree bound is sharp. This closes the one-unit gap left by the prescribed-vertex theorem of Kelly, Kühn and Osthus when . We also prove that if an oriented graph has order , minimum semidegree , and , then every ordered pair of distinct vertices is joined by a path of length three, four, or five. The constant is best possible. As a consequence, the order hypothesis in the general prescribed-vertex theorem of Kelly, Kühn and Osthus can be replaced by for .
Key words and phrases:
oriented graph, prescribed vertex, directed cycle, minimum semidegree, stability2020 Mathematics Subject Classification
Primary 05C20; Secondary 05C35, 05C381. Introduction
An oriented graph is an orientation of a simple graph. For a vertex of an oriented graph , let and denote its outdegree and indegree. Denote
All paths and cycles in this paper are directed unless stated otherwise. We write for the consistently oriented cycle of length .
Degree conditions for directed cycles in oriented graphs, and in particular for Hamilton cycles, are surveyed by Bermond and Thomassen [1] and by Kühn and Osthus [11]. We consider the minimum semidegree condition that forces a directed cycle of a given length through each specified vertex.
For , let be the least integer such that every -vertex oriented graph with contains a copy of through every vertex. The superscript emphasizes that this is a prescribed-vertex, or vertex-rooted, threshold: it is stronger than merely requiring one copy of somewhere in the graph.
Kelly, Kühn and Osthus [10] proved that, for every and every , the condition
forces a copy of through any prescribed vertex. When , their modified cyclic blow-up [10] gives a prescribed vertex contained in no while
Consequently, for and ,
| (1) |
The two bounds coincide when . For , however, they leave precisely the one-unit window
Denote
| (2) |
We present the main results as follows.
Theorem 1.1.
Let and . If is an oriented graph on vertices and
then every vertex of lies on a copy of .
Corollary 1.2.
For every and ,
For and , this replaces the upper bound in (1) by .
The order bound in Theorem 1.1 follows from a three-length linking lemma. With , the hypothesis is for , in place of . If an oriented graph has order and minimum semidegree , then
guarantees, between every ordered pair of distinct vertices, a path of length three, four, or five. We give examples with , so the additive constant is sharp. In the defect parameterization , where is a positive integer, the lemma already applies when . The coefficient is asymptotically best possible.
Jackson [6] gave classical semidegree conditions for long directed paths and cycles. Darbinyan and Karapetyan [3] studied short paths, including versions with forbidden vertices. Their result bounds directed distance but does not guarantee a path whose length belongs to . Zhou and Yan [14] proved an -linkage theorem at the scale for fixed , sufficiently large order, and prescribed subdivision-path lengths at least four. Its semidegree condition is above the one-third scale considered here.
The same short-linking lemma improves the large-order hypothesis in the full prescribed-vertex theorem of Kelly, Kühn and Osthus, without the restriction . We prove below that, for every , the condition
forces a copy of through every prescribed vertex. Their direct arguments for already require only . Thus the same theorem has an explicit linear order bound for every length. To the best of our knowledge, no previous result improves the explicit condition while retaining both the prescribed vertex and the assumption . For and , this corollary still assumes . The equality analysis lowers this to ; the deletion budgets for , , and give the three values in (2). If and , Corollary 2.8 gives the stronger order bound .
Without prescribing a vertex, Czygrinow, Molla, Nagle and Oursler [2] proved that, for each fixed and all sufficiently large , the one-sided condition
forces a copy of . Their theorem does not give a rooted conclusion or an explicit order bound.
The Hamiltonian threshold was developed in [5, 9] and determined exactly for large order in [8]. At this denser scale, Kelly, Kühn and Osthus [10] proved that every prescribed vertex lies on a directed cycle of every length , while Wang, Wang and Zhang [13] treat arbitrary orientations without a prescribed vertex. These results do not apply at semidegree .
The restriction is essential. A directed triangle through a prescribed vertex has a different threshold: Kelly, Kühn and Osthus [10] showed that its asymptotic scale is , rather than . Thus does not belong to the phenomenon studied here.
The rooted problem is different from ordinary, unrooted containment. Kelly, Kühn and Osthus [10, Conjecture 5] conjectured that, if and is the smallest integer that does not divide , then, for all sufficiently large , the condition
forces a copy of in every oriented graph on vertices. The conjectured threshold is suggested by cyclic blow-ups. For and , Kühn, Osthus and Piguet [12] proved the corresponding asymptotic result: for every , the condition suffices for all sufficiently large . Grzesik and Volec [4, Theorem 1.4] later determined the corrected exact thresholds for this unrooted problem. The conjectured bound is exact when or ; the other congruence classes require the rounding corrections stated in their theorem. In particular, when and , one has , so the unrooted threshold is at most . Requiring the cycle to pass through an arbitrary vertex restores the one-third barrier, as the lower construction behind (1) shows. Related prescribed-vertex results under the additional assumption that the oriented graph contains no directed triangle were obtained by Ji, Wu and Song [7, Theorem 1.4].
Only the case requires an additional argument. Fix the prescribed vertex . If is independent, Lemma 2.9 gives the root-dominating balanced cut. Otherwise choose an arc in . If is independent, the same lemma gives the transitive-entry balanced cut; if not, an arc in produces an -butterfly. In the two balanced-cut cases, the row-family classification and matching arguments are used in Propositions 4.2 and 5.1. The butterfly case is completed by Lemmas 6.1 and 6.2. When , every outneighbourhood is non-independent at the threshold , so only the butterfly case is needed. Sections 2–6 follow this order.
2. Short linking and preliminary reductions
For an oriented graph , write for its vertex set and for its order. For , let and be its out- and inneighbourhoods, and put . We omit the subscript when the ambient graph is clear. For , define
Thus the external neighbourhood of is . For , set
We write for the subgraph induced by , for its number of arcs, and, for disjoint sets , for the number of arcs directed from to . A vertex string denotes the directed path
We use the following two elementary facts repeatedly.
Lemma 2.1.
Let be an oriented graph.
- (i)
Every nonempty satisfies . Consequently,
and the analogous reverse inequality holds for .
- (ii)
Every independent set has size at most .
Proof.
The first assertion follows because an oriented graph has at most one arc on each unordered pair. Choose a vertex of with outdegree at most to obtain the displayed inequality; reverse all arcs for its counterpart. If is independent, then for any the sets and are disjoint subsets of , giving . ∎
Kelly, Kühn and Osthus [10] proved that, for a positive integer , under the hypotheses
every ordered pair of distinct vertices is joined by a path whose length belongs to . The next lemma replaces their large-order hypothesis by a sharp linear inequality.
Lemma 2.2.
Let be an oriented graph of order , and put . If and
then every ordered pair of distinct vertices is joined by an – path of length , , or .
Proof.
The proof idea has three steps. We first choose equal-sized sets in the outneighbourhood of the initial vertex and the inneighbourhood of the terminal vertex; the absence of paths of lengths , , and then forces a rigid system of forbidden arcs between the resulting layers. Minimum semidegree makes two intermediate layers large, while the forbidden arcs give an incompatible upper bound on the total outdegree of one of them. The contradiction reduces to a quadratic inequality. Its two endpoint estimates use precisely the relation .
Step 1: the forbidden-arc structure. Put . The sets and both have size at least . Choose an -set . Since , there is an -set with . Set
Suppose, for a contradiction, that there is no – path of length , , or . Then there is no arc from to . Define
and put and . The sets , , , and are pairwise disjoint. Indeed, or would give an – path of length three, while would give one of length four. We also have
otherwise there is an – path of length four, five, or three, respectively.
Step 2: degree counting. Let
and define
Since , we have . No arc goes from to , and because . Hence every out-arc of goes inside , to , or possibly to , and therefore
The reverse argument, applied to the indegrees of , gives . Consequently,
All out-arcs of go to , inside , to , or possibly to . Orientedness between and gives . Summing the outdegrees of the vertices of , we obtain
Thus
Substituting the value of and using gives
The hypothesis is equivalent to
| (3) |
Step 3: the quadratic contradiction. Write . If , then is increasing in for , and
Indeed, , so the last factor is at least by (3).
It remains that . For fixed , the convex quadratic is minimized at , so
Since is an integer, is equivalent to . Together with , this implies . The function is concave in , so it suffices to check the two endpoints of
Write . Direct expansion gives
Thus throughout the interval, again contradicting . ∎
Substituting in Lemma 2.2 gives the following form.
Corollary 2.3.
Let be a positive integer. If is an oriented graph on vertices with
then every ordered pair of distinct vertices is joined by a path of length , , or .
Proof.
The following construction shows that cannot be reduced and that the coefficient is asymptotically best possible.
Proposition 2.4.
Proof.
Take disjoint sets with
together with two vertices . Let and induce regular tournaments, and let and be independent. Add all arcs indicated by
and
with no other arcs between the parts. Regular tournaments exist on vertices.
The vertices in , , , , , and have respective -pairs
Hence and , so
Every – path is either the arc , has the form with , or begins . In the last case, any simple path ending at must pass successively through
arcs inside the regular tournaments can only increase its length. Thus every remaining – path has length at least six.
Set
Then , while . Consequently, there are no constants and for which the condition is a universal sufficient order hypothesis in Corollary 2.3. ∎
We next record the extension and deletion statements used below.
Lemma 2.5.
Let be an oriented graph.
- (i)
Let and be nonnegative integers. If , , , and
then contains a simple path of length starting at and otherwise avoiding .
- (ii)
Let have order and minimum semidegree at least , and let be an integer. If
and is obtained by deleting at most vertices, then every ordered pair of distinct vertices of is joined in by a path of length , , or .
- (iii)
Let have order and minimum semidegree at least . Let and be integers with , let be distinct vertices, and let . Suppose that, for each , there is an – path of length whose internal vertices lie in . Put
If
then lies on a copy of .
Proof.
For (i), extend greedily. Before the st step, at most vertices are forbidden, which is smaller than for .
For (ii), let vertices be deleted, put , and write with . Set
Then . A direct calculation shows that
| (4) |
The hypothesis gives
Since , the three expressions in (4) are respectively at least , , and . Moreover, . Lemma 2.2 now applies to .
For (iii), apply (i) with forbidden set to obtain a path of length from to a vertex . Delete and . At most vertices are deleted, so (ii) gives a – path of some length . The paths , , and the – path are internally disjoint and together form a . ∎
Remark 2.6.
The three cases in (2) come from three different deletion budgets. For , the only critical use of short linking deletes vertices in the transitive-entry branch, giving . For , a terminal-safe three-chain is already a ; the largest remaining budget is in the distinct-row subcase of the transitive-entry branch, giving . For , the terminal-safe-chain branch of Proposition 4.2 may delete vertices, giving . These are the smallest uniform cutoffs delivered by the present argument. The stable balanced-cut branches use only the matching inequality .
Following Kelly, Kühn and Osthus [10, discussion preceding Fact 18], an -butterfly consists of five distinct vertices and the six arcs
It contains – paths of lengths , , and .
The sharp linking lemma also gives a linear order bound in the full prescribed-vertex theorem of Kelly, Kühn and Osthus.
Corollary 2.7.
Let and . If is an oriented graph on vertices with
then every vertex of lies on a copy of .
Proof.
Fix and put . By Lemma 2.1(ii), every independent set has size at most . Hence contains an arc , and contains an arc . The five vertices are distinct, and
form an -butterfly. Indeed, because and , whereas and . This is the argument of [10, Fact 18]; we include it to keep track of the order bound.
Since and , we have
Lemma 2.5(i) gives a path of length from to a vertex , otherwise avoiding . Delete
and call the remaining graph . Exactly vertices are deleted.
Write with . Then
Moreover, gives since , and hence
Also . Lemma 2.2 therefore gives a – path in of some length . Choose the – path of length in the butterfly. Together with and the – path, it forms a simple cycle of length
∎
Together with Lemmas 16, 17 and 19 of Kelly, Kühn and Osthus [10], Corollary 2.7 replaces the order hypothesis in their Theorem 4 by for , and by for .
Taking in Corollary 2.7 gives the following bound when .
Corollary 2.8.
Let , let , and suppose . If is an oriented graph on vertices with
then every vertex of lies on a copy of .
Proof.
We next isolate the equality structure produced by an independent outneighbourhood.
Lemma 2.9.
Let be an oriented graph on vertices with . If is independent, then, with
we have , , and
In particular, each pair in is joined by exactly one arc.
Proof.
Lemma 2.1(ii) gives , while gives . Thus . Since is independent, every in- and outneighbour of lies in . The two neighbourhoods are disjoint, have size at least each, and lie in the -set , so equality holds throughout. ∎
We call a partition of a balanced cut if, for some integer , the set is independent, , , and
If , we say that the balanced cut is rooted at .
In a balanced cut, put
Thus every row has size . For , define
The semidegree condition gives the fundamental inequalities
| (5) |
3. Nonextendable endpoints and row-family stability
Throughout the first part of this section, is a balanced cut and satisfies . Thus for every . A -transition into is a simple path
Lemma 3.1.
If , then at most three vertices of admit no -transition.
Proof.
Call such a vertex nonextendable and put
and
If , then every inneighbour of in lies in ; otherwise some has an inneighbour , giving .
Nonextendable vertices with . For , the only possible inneighbour of in is , so . By (5), , while gives . Hence and is the unique outneighbour of in . It follows that the sets belonging to distinct nonextendable vertices with are pairwise disjoint. Since three such sets have total size , there are at most two nonextendable vertices of this kind.
Nonextendable vertices with . We show that there is at most one. Suppose that distinct nonextendable vertices have nonempty . First, these sets are disjoint. Indeed, if is nonempty, then every has and . Moreover, , and all its inneighbours in lie in . Therefore every has at least inneighbours in , and
Thus . On the other hand, and , so , giving , a contradiction.
Write and . For , we have and hence . Since , the definition of gives , so . Nonextendability of implies that at least inneighbours of lie in . Consequently,
Symmetrically, . The two sets are disjoint and the graph is oriented, so at most one of the two possible arcs occurs on each pair in . Therefore
Furthermore,
so . Consequently,
which is impossible for . Hence at most one nonextendable vertex has nonempty , and the total number of nonextendable vertices is at most three. ∎
We now turn to the set-system statement. Let be a finite set, let , and let be a family of equal-sized subsets of , each meeting . A terminal-safe three-chain consists of four distinct indices and four pairwise distinct elements such that
Lemma 3.2.
Let be four distinct -subsets of a set . There is a cyclic ordering of these sets and pairwise distinct elements
Proof.
For a cyclic order , put
We have : membership in is required by and forbidden by . Consequently,
Thus the four sets have a system of distinct representatives if and only if each opposite pair and has two distinct representatives; representatives chosen for different opposite pairs are automatically distinct. Each is nonempty, because consecutive sets have the same size and are distinct. Hence an opposite pair fails Hall’s condition precisely when its two members are the same singleton.
Write
and call a pair coarse if . If two coarse pairs are adjacent, extend them to a Hamilton cycle on the four labels. The two coarse directed differences belong to different opposite pairs, so neither opposite pair can be the same-singleton obstruction.
It remains to consider the case where the graph of coarse pairs is a matching. Suppose first that it is nonempty, and relabel so that is coarse. Then are thin, meaning that their -value is one. Consider the cycles
In each order, the coarse edge supplies a difference set of size at least two to one opposite pair, so only the other opposite pair can obstruct a rainbow choice. If all four cycles fail, the four obstructions are as follows:
Here and . Also and , since either equality would impose contradictory membership in or , respectively. Thus are pairwise distinct, and
and
For the cycle , the four directed differences contain, in order, , a contradiction.
Finally, suppose there is no coarse pair. Write
Every -set satisfying has exactly one of the forms
A set of the first form and one of the second form have -value two. Hence have the same form, and all four sets are either or , where . In the first case any cyclic order has the distinct witnesses ; in the second it has the distinct witnesses . ∎
Corollary 3.3.
If four distinct row types occur in , then has a terminal-safe three-chain.
Proof.
Take a rainbow cycle with witnesses . If some , delete the outgoing edge , order the remaining path as
and use as the terminal representative. If no lies in , delete any edge and choose an arbitrary element of the final row in as the terminal representative. In either case the four representatives are distinct. ∎
The preceding lemma yields the following classification.
Proposition 3.4.
Let , and let every row be a -subset of meeting . The family has no terminal-safe three-chain if and only if one of the following holds.
- (i)
There is one row type.
- (ii)
There are exactly two row types , and either one has multiplicity one, or both have multiplicity at least two and .
- (iii)
There are exactly three row types of one of the following forms:
or
where are distinct, and the common intersection of the three rows is disjoint from .
Proof.
Corollary 3.3 excludes four row types. If there is one row type, then every difference is empty, so no terminal-safe three-chain exists.
Suppose there are two types . A four-index chain exists only if both types occur at least twice, and then the type sequence must alternate. Put and . If , choose two distinct representatives from the two copies of , choose , and choose a representative from distinct from ; this is possible because . Conversely, if , the two -positions cannot receive distinct representatives. This proves (ii).
Now suppose there are three types. Some type, say , occurs at least twice; call the other types . For the type sequence , the four representative sets are
All four sets are nonempty: the first three because the row types are distinct and equicardinal, and the last because every row meets . The first and third lie outside , whereas the second and fourth lie inside . Thus the union of the outside pair is disjoint from the union of the inside pair. As in Lemma 3.2, an SDR exists if and only if each pair has two distinct representatives, and a pair fails precisely when its two members are the same singleton. Hence Hall’s condition fails exactly when
| (6) |
or
| (7) |
For the sequence , failure is equivalent to (6) or
| (8) |
In the first case, equal row sizes give
for distinct . In the second case,
for distinct . Thus the types have one of the two forms in part (iii).
In the first form, if the common intersection contained , then the sequence would have the three distinct hub representatives together with , giving a terminal-safe three-chain. In the second form, already implies . Conversely, for either family with -free common intersection, every difference representative and every terminal representative lies in the same three-point hub. Four distinct representatives are impossible. ∎
Corollary 3.5.
If a row family indexed by , , has no terminal-safe three-chain, then either
- (i)
one row type has multiplicity at least , or
- (ii)
Proof.
In Proposition 3.4(ii), a singleton type leaves the other type with multiplicity ; otherwise the two types have total variation two. In case (iii), the total variation is exactly the three-point hub. ∎
4. The root-dominating balanced cut
In this section, has order and minimum semidegree at least , is a balanced cut, and satisfies . Put
Every row is an -subset of the -set , while . Hence
| (9) |
We first handle directly.
Proposition 4.1.
If , then lies on a copy of .
Proof.
Let be the set of nonextendable endpoints from Lemma 3.1, so . Suppose first that some satisfies . Choose a -transition
Since , we have . Choose
Then
is a .
We may therefore assume that for every . Since there is at least one such row, the inequality
implies . On the other hand, , so has no inneighbour in and
Thus . Put , so . Every extendable row has the form
Thus for every , and (5) gives
| (10) |
For ,
We claim that there are , , and such that and . Otherwise, for every arc from to and every extendable row label , we would have . The lower bound (10) is positive, and is nonempty, so all targets of arcs from to and all extendable row labels would equal a single vertex . This would give
a contradiction. Since , choose another extendable vertex . Then
is a . ∎
We now close all longer multiples of three under the linear order hypothesis of Theorem 1.1.
Proposition 4.2.
Let , put , and suppose . Then lies on a copy of .
Proof.
Since ,
In particular,
| (11) |
Let be the nonextendable-endpoint set and put . Thus and every admits a -transition. Apply Corollary 3.5 to the rows indexed by , with terminal set .
The proof follows the three outcomes in Corollary 3.5. A terminal-safe three-chain gives three possible initial lengths and is closed by short linking. In each of the two remaining outcomes, (5) gives a dense bipartite graph, and König’s theorem supplies the required matching.
Case 1: a terminal-safe three-chain. Suppose first that there is a terminal-safe three-chain. It yields distinct vertices with
| (12) |
If , this is already a . Assume . The suffixes of (12) give – paths
of lengths three and five. Since , take a transition
Then
is a simple path of length four. Different candidate paths may intersect; only one will eventually be used. Let be the union of their internal vertices. Then .
Case 2: a dominant row. Suppose that a row type occurs at least
times. For every , we have , so
| (13) |
Choose , which exists by (9), and form the bipartite graph with parts
joining to precisely when in . Deleting the source and the target removes at most arcs from (13), so this bipartite graph has at least edges. If its maximum matching had size at most , König’s theorem would give a vertex cover of size at most . Every vertex of the bipartite graph has degree at most , so the cover would meet at most edges. By (11),
a contradiction. Take a matching
and distinct -type vertices . These choices are available because by (11). Then
is a .
Case 3: variation on at most three points. Suppose the active variation has size at most three. Put
Write and . Every row is with and . If , each point of appears in some but not all rows, so . Hence
Every belongs to all rows indexed by , and therefore . Since , we obtain
| (14) |
Fix and . Since belongs to no row, . Form the bipartite graph whose left part is if and is otherwise, whose right part is , and whose edges are the arcs directed from left to right. Deleting the target and, when necessary, the source removes at most arcs from (14); hence at least edges remain. If there were no matching of size , König’s theorem would give a vertex cover of size at most . Both parts have size at most , so such a cover would meet at most edges. By (11),
and hence there is a matching of size from to , avoiding and . Since , choose distinct with . Since every row contains and avoids ,
is again a . ∎
5. The transitive-entry balanced cut
We now consider a balanced cut rooted at , together with distinct vertices satisfying
| (15) |
The prescribed vertex is , not .
Proposition 5.1.
Let , put , and suppose . Under (15), the vertex lies on a copy of .
Proof.
Since , we have
In particular,
| (16) |
We distinguish whether sends an arc to , whether the rows are distinct, and whether all rows are equal. The first two cases use three initial path lengths and short linking. The last case uses a bipartite matching.
Case 1: an arc from to . Suppose first that for some . There are – paths of lengths one, two, and three:
With and , put . The two numerical conditions in Lemma 2.5(iii) are
Thus that lemma gives a through . This includes the case .
Case 2: two distinct rows. We may assume that and that two rows are distinct. Since belongs to every row and belongs to none, choose
with . Then . If , the three – paths
have lengths two, three, and four. Apply Lemma 2.5(iii) with and . Indeed,
and, since ,
If , choose the order of the two rows so that . This is always possible: if , then every element of differs from . Now
is a .
Case 3: all rows are equal. It remains that and all rows are equal to one -set . Then and . Every has , and hence
Form the bipartite graph with source part , target part , and an edge whenever . Deleting the source removes at most arcs. After that deletion, the target is incident with at most remaining sources. Equivalently, the known arc is not counted twice in the two deletions. Hence at least
edges remain. If there were no matching of size , König’s theorem would give a vertex cover of size at most , which meets at most edges. By (16),
a contradiction. A matching of size and distinct vertices , available because by (16), yield
a cycle of length . ∎
6. The butterfly branch and the main theorem
The next lemma adapts [10, Lemma 19]. Its point is that, when , the butterfly argument lowers the semidegree hypothesis from to .
Lemma 6.1.
Let be an oriented graph on vertices, put , and suppose and . If contains an -butterfly, then lies on a copy of .
Proof.
Write the butterfly vertices as , with arcs
We first translate the absence of a through into three restrictions on return paths from to . Two neighbourhood layers then have only one possible overlap, and a cardinality estimate forces that overlap to contain a vertex different from .
The butterfly contains the – paths
of lengths four, three, and two, respectively. Any – path of length two automatically avoids : using one of these vertices as its internal vertex would contradict, respectively, , , or . Such a path therefore closes with . A length-three – path cannot contain , because and ; if it also avoids , it closes with . Finally, a length-four – path avoiding closes with . Thus, if no through exists, we may assume that
- (i)
there is no – path of length two;
- (ii)
there is no – path of length three avoiding ;
- (iii)
there is no – path of length four avoiding .
Choose
Set
The three assumptions imply
Indeed, a vertex in gives a length-two – path. If , choose with ; then . If , choose with ; then . In both length-three paths the internal vertices avoid , since by and by the definition of . They also imply that lie in none of . Indeed, the definitions exclude from and from ; if or , then there is a – path of length two, while orientedness excludes and . By Lemma 2.1,
and hence . Since every overlap among the four sets is contained in , we obtain
Since , we have . Choose . There are vertices and such that
Since by , by , and , this is a length-four path avoiding , contradicting (iii). ∎
For , the three paths in a butterfly can be combined with the short-linking lemma.
Lemma 6.2.
Let , put , and let be an oriented graph on vertices with . If contains an -butterfly, then lies on a copy of .
Proof.
Let be the butterfly vertices, and put and . The butterfly supplies – paths of lengths two, three, and four. Moreover,
and, since ,
Lemma 2.5(iii) gives a through . ∎
We can now prove the critical case.
Theorem 6.3.
Let , put , and let be an oriented graph on vertices with . Then every vertex of lies on a copy of .
Proof.
Fix . If is independent, Lemma 2.9 gives the root-dominating balanced cut. Proposition 4.1 applies when , and Proposition 4.2 applies when .
Proof of Theorem 1.1.
Put and fix . If , the result is Theorem 6.3. Suppose that and put . Lemma 2.1(ii) gives
for every independent set . Since every outneighbourhood has size at least , none is independent. Choose an arc inside , and then choose an arc inside . The five vertices are distinct: cannot equal or because and . Thus
form an -butterfly. Lemma 6.1 applies when , and Lemma 6.2 applies when . ∎
The following construction of Kelly, Kühn and Osthus [10, discussion following Theorem 4], with relabelled parts, proves that the semidegree bound in Corollary 1.2 is best possible. On vertices, take three independent sets whose sizes differ by at most one, and add all arcs
Add a vertex with
and no arcs between and . The resulting oriented graph satisfies the following degree table, where , , and :
Since differ by at most one, this gives
Every directed cycle through consists of the arc from to , a path from to in the cyclic three-part core, and an arc from to . The middle path has length modulo , so the whole cycle has length modulo . Hence lies on no , proving Corollary 1.2.
Remark 6.4.
Proposition 2.4 shows that the constant in the linking inequality cannot be reduced and that the coefficient in Corollary 2.3 is asymptotically best possible. This does not determine the optimal coefficients in the order hypotheses of Theorem 1.1 or Corollary 2.7.
For , an argument that uses the remaining graph only through its order and minimum semidegree cannot lower to . Indeed, put . If and the terminal-safe-chain branch deletes vertices, the remaining parameters are
with
With , the pair is the boundary pair in Proposition 2.4. Hence a smaller cutoff requires additional information about the remaining graph, such as the structure of the deleted path or the relation between the in- and outdegree losses. It remains open to determine the smallest constant for which an order hypothesis of the form suffices in Theorem 1.1.
References
- [1] J.-C. Bermond and C. Thomassen, Cycles in digraphs—a survey, J. Graph Theory 5 (1981), no. 1, 1–43.
- [2] A. Czygrinow, T. Molla, B. Nagle and R. Oursler, On even rainbow or nontriangular directed cycles, J. Comb. 12 (2021), no. 4, 589–662.
- [3] S. Kh. Darbinyan and I. A. Karapetyan, A note on short paths in oriented graphs, Math. Probl. Comput. Sci. 33 (2010), 35–40.
- [4] A. Grzesik and J. Volec, Degree conditions forcing directed cycles, Int. Math. Res. Not. IMRN 2023 (2023), no. 11, 9711–9753.
- [5] R. Häggkvist, Hamilton cycles in oriented graphs, Combin. Probab. Comput. 2 (1993), no. 1, 25–32.
- [6] B. Jackson, Long paths and cycles in oriented graphs, J. Graph Theory 5 (1981), no. 2, 145–157.
- [7] Y. Ji, S. Wu and H. Song, On short cycles in triangle-free oriented graphs, Czechoslovak Math. J. 68 (2018), no. 1, 67–75.
- [8] P. Keevash, D. Kühn and D. Osthus, An exact minimum degree condition for Hamilton cycles in oriented graphs, J. Lond. Math. Soc. (2) 79 (2009), no. 1, 144–166.
- [9] L. Kelly, D. Kühn and D. Osthus, A Dirac-type result on Hamilton cycles in oriented graphs, Combin. Probab. Comput. 17 (2008), no. 5, 689–709.
- [10] L. Kelly, D. Kühn and D. Osthus, Cycles of given length in oriented graphs, J. Combin. Theory Ser. B 100 (2010), no. 3, 251–264.
- [11] D. Kühn and D. Osthus, A survey on Hamilton cycles in directed graphs, European J. Combin. 33 (2012), no. 5, 750–766.
- [12] D. Kühn, D. Osthus and D. Piguet, Embedding cycles of given length in oriented graphs, European J. Combin. 34 (2013), no. 2, 495–501.
- [13] G. Wang, Y. Wang and Z. Zhang, Arbitrary orientations of cycles in oriented graphs, arXiv:2504.09794v2 [math.CO], 2025.
- [14] J. Zhou and J. Yan, Semi-degree condition for arbitrary -linked oriented graphs, arXiv:2407.06675v2 [math.CO], 2024, revised 2025.