This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Blow-ups and extensions of trees in tournaments

Pierre Aboulker111DIENS, École normale supérieure, CNRS, PSL University, Paris, France. Research supported by the ANR project DAGDigDec (JCJC) ANR-21-CE48-0012 and by the group Casino/ENS Chair on Algorithmics and Machine Learning.    Frédéric Havet222Université Côte d’Azur, CNRS, Inria, I3S, Sophia-Antipolis, France. Research supported by research grant DIGRAPHS ANR-19-CE48-0013.    William Lochet333LIRMM, Université de Montpellier, CNRS, Montpellier, France. Research supported by research grant ELiT ANR-20-CE48-0008-01.    Raul Lopes11footnotemark: 1 ,33footnotemark: 3    Lucas Picasarri-Arrieta444National Institute of Informatics, Japan. Research supported by Japan Science and Technology Agency (JST) as part of Adopting Sustainable Partnerships for Innovative Research Ecosystem (ASPIRE), Grant Number JPMJAP2302.    Clément Rambaud22footnotemark: 2
Abstract

A class of acyclic digraphs 𝒞\mathcal{C} is linearly unavoidable if there exists a constant cc such that every digraph D𝒞D\in\mathcal{C} is contained in all tournaments of order c|V(D)|c\cdot|V(D)|. The class of all acyclic digraphs is not linearly avoidable, and Fox, He, and Widgerson recently showed that this is not even the case for acyclic digraphs with bounded maximum degree. On the positive side, Thomason and Häggkvist proved that the class of oriented trees is linearly unavoidable. In this work, we generalize this result to acyclic digraphs obtained from an oriented tree by adding at most kk vertices, and kk-blow-ups of oriented trees, for every fixed integer kk.

More precisely, we show that if DD is obtained from an oriented tree FF of order nn by adding kk universal vertices, then DD is contained in every tournament of order 23(k+1)(2k+1)n2\cdot 3^{(k+1)(2k+1)}\cdot n; and if DD is obtained from FF by replacing each vertex uu by an independent set XuX_{u} of size kk and every arc uvuv by all possible arcs from XuX_{u} to XvX_{v}, then DD is contained in every tournament of order 210+18kkn2^{10+18k}k\cdot n.

1 Introduction

A digraph is pp-unavoidable, for an integer pp, if it is contained in every tournament of order pp. For a digraph DD, the unavoidability of DD, denoted by unvd(D)\operatorname{unvd}(D), is the smallest integer pp such that DD is pp-unavoidable, if such a pp exists. It is easy and well-known that a digraph is pp-unavoidable for some integer pp if and only if it is acyclic. Indeed, every digraph containing a directed cycle is not contained in any transitive tournament, and so is not unavoidable. Moreover, as observed by Erdős and Moser [8], an immediate induction and Proposition 1 shows that the transitive tournament of order nn is 2n12^{n-1}-unavoidable and thus every acyclic digraph of order nn is 2n12^{n-1}-unavoidable.

Proposition 1.

Let DD be an acyclic digraph. For every source or sink xx of DD,

unvd(D)2unvd(Dx).\operatorname{unvd}(D)\leqslant 2\operatorname{unvd}(D-x).

This is somehow tight in the sense that any upper bound on the unavoidability of acyclic digraphs as a function of their order must be exponential. Indeed, considering a random tournament and using the First Moment Method, Erdős and Moser [8] established unvd(TTn)2(n1)/2\operatorname{unvd}(TT_{n})\geqslant 2^{(n-1)/2}. Since this seminal paper, only little progress has been made on the unavoidability of transitive tournaments. Using the Local Lemma, one can slightly improve on the lower bound and show that unvd(TTn)2(n+1)/2\operatorname{unvd}(TT_{n})\geqslant 2^{(n+1)/2} when nn is large enough. The unavoidability of small transitive tournaments have been established. Clearly unvd(TT1)=1\operatorname{unvd}(TT_{1})=1, unvd(TT2)=2\operatorname{unvd}(TT_{2})=2, unvd(TT3)=4\operatorname{unvd}(TT_{3})=4 (because of the directed 33-cycle), and unvd(TT4)=8\operatorname{unvd}(TT_{4})=8 (because of the Paley tournament on 77 vertices). Reid and Parker [22] showed unvd(TT5)=14\operatorname{unvd}(TT_{5})=14, unvd(TT6)=28\operatorname{unvd}(TT_{6})=28, and Sanchez-Flores [25] proved unvd(TT7)=54\operatorname{unvd}(TT_{7})=54. This result and Proposition 1 imply unvd(TTn)54×2n7\operatorname{unvd}(TT_{n})\leqslant 54\times 2^{n-7} for all n7n\geqslant 7.

The average degree of a nonempty digraph DD is the average degree of its underlying graph, that is 2|A(D)||V(D)|2\frac{|A(D)|}{|V(D)|}. The maximum average degree of a digraph is the maximum average degree over all its nonempty subdigraphs. Similarly to Erdős and Moser [8], one can show that the unavoidability of an acyclic digraph is exponential in its maximum average degree.

Proposition 2.

Let DD be an acyclic digraph with maximum average degree α\alpha. Then unvd(D)>2α/2\displaystyle\operatorname{unvd}(D)>2^{\alpha/2}.

Proof.

Consider a random tournament TT on p2α/2p\leqslant 2^{\alpha/2} vertices. Let v1,,vnv_{1},\dots,v_{n} be a labelling of the vertices of a subdigraph HH of DD with average degree α\alpha. For each ordered nn-uple (w1,,wn)(w_{1},\dots,w_{n}) of distinct vertices in TT, the probability that T{w1,,wn}T\langle\{w_{1},\dots,w_{n}\}\rangle contains a labelled copy of HH (i.e. such that wiwjw_{i}w_{j} is an arc in TT whenever vivjv_{i}v_{j} is an arc in HH) is (12)|A(H)|=(12)αn/2\left(\frac{1}{2}\right)^{|A(H)|}=\left(\frac{1}{2}\right)^{\alpha n/2}. Thus the expected number of labelled copies of HH is p!(pn)!(12)αn/2<pn2αn/2=(p2α/2)n1\frac{p!}{(p-n)!}\left(\frac{1}{2}\right)^{\alpha n/2}<\frac{p^{n}}{2^{\alpha n/2}}=\left(\frac{p}{2^{\alpha/2}}\right)^{n}\leqslant 1. ∎

Proposition 2 gives a non-trivial lower bound on the unavoidability of dense acyclic digraphs, that is with maximum average degree larger than logarithmic in their order. However, smaller bounds on the unavoidability are expected for sparse digraphs. In particular, one can ask which families {\cal F} of digraphs are linearly unavoidable, that is such that there exists a constant CC such that unvd(D)C|V(D)|\operatorname{unvd}(D)\leqslant C\cdot|V(D)| for every digraph DD in {\cal F}, and which are polynomially unavoidable, that is such that there exists a polynomial PP such that unvd(D)P(|V(D)|)\operatorname{unvd}(D)\leqslant P(|V(D)|) for every digraph DD in {\cal F}.

Note that by Proposition 2, the digraphs DD of a linearly unavoidable family must have maximum average degree at most 2log(|V(D)|)+𝒪(1)2\log(|V(D)|)+\mathcal{O}(1). For every integer nn with n2n\geqslant 2, Linial, Saks and Sós [19] constructed an nn-unavoidable digraph with nn vertices and nlognCnloglognn\log n-C\cdot n\log\log n arcs for some fixed constant CC. Those digraphs form a linearly unavoidable family with logarithmic maximum average degree.

Several other classes of acyclic digraphs have been proved to be linearly unavoidable. Much attention has been devoted to oriented paths, oriented trees, and oriented forests which are orientations of paths, trees, and forests respectively. It started with Rédei’s theorem [21] which states that the unavoidability of Pn\vec{P}_{n}, the directed path on nn vertices, is exactly nn.

Theorem 3 (Rédei [21]).

Every tournament has a Hamiltonian directed path.

In 1971, Rosenfeld [24] conjectured that there is an integer N>7N>7 such that unvd(P)=|P|\operatorname{unvd}(P)=|P| for every oriented path of order at least NN. Several papers gave partial answers to this conjecture [11, 1, 9, 26] until Rosenfeld’s conjecture was verified by Thomason, who proved in [27] Rosenfeld’s conjecture for N=2128N=2^{128}. Finally, Havet and Thomassé [15] showed that unvd(P)=|V(P)|\operatorname{unvd}(P)=|V(P)| for every oriented path PP except the antidirected paths of order 33, 55, and 77.

Regarding oriented trees, Sumner (see [23]) made the following celebrated conjecture.

Conjecture 4 (Sumner).

Every oriented tree of order n>1n>1 is (2n2)(2n-2)-unavoidable.

Since every forest is the subdigraph of a tree, this conjecture is equivalent to the analogue for oriented forests. If true, Sumner’s conjecture would be tight. Indeed, the out-star Sn+S^{+}_{n}, which is the digraph on nn vertices consisting of a vertex dominating the n1n-1 others, is not contained in the regular tournaments of order 2n32n-3. The first linear upper bound was given by Häggkvist and Thomason [12]. Following improvements of Havet [13], Havet and Thomassé [14], and El Sahili [7], Dross and Havet [6] used the notion of median order to prove that every oriented tree of order n2n\geqslant 2 is 218n4716\left\lceil\frac{21}{8}n-\frac{47}{16}\right\rceil-unavoidable. Kühn, Mycroft and Osthus [16] proved that Sumner’s conjecture is true for all sufficiently large nn.

Theorem 5 (Kühn, Mycroft, and Osthus [16]).

There is an integer n0n_{0} such that for every nn0n\geqslant n_{0}, every oriented tree of order nn is (2n2)(2n-2)-unavoidable.

Motivated by those results and a work of Lee [18], Bucić, Letzter and Sudakov [3] asked whether the digraphs with bounded maximum average degree are linearly unavoidable. Draganić et al. [5] proved that this is the case for kk-th powers of directed paths. The kk-th power of a digraph DD is the digraph, denoted by DkD^{k}, with vertex set V(D)V(D) in which uvuv in an arc if and only if distD(u,v)k\operatorname{dist}_{D}(u,v)\leqslant k. Fox, He, and Widgerson [10] showed that this is not the case in general: they proved that for any Δ2\Delta\geqslant 2 and any sufficiently large nn, there exists an acyclic digraph DD with nn vertices and maximum degree Δ\Delta such that

unvd(D)nΩ(Δ2/3/log5/3Δ).\operatorname{unvd}(D)\geqslant n^{\Omega(\Delta^{2/3}/\log^{5/3}\Delta)}.

This result raises the two following questions. Firstly, are the digraphs with bounded maximum average degree polynomially unavoidable?

Problem 6.

Let α\alpha be a positive real number. Does there exist a polynomial PαP_{\alpha} such that unvd(D)Pα(|V(D)|)\operatorname{unvd}(D)\leqslant P_{\alpha}(|V(D)|) for every digraph with maximum average degree at most α\alpha?

Secondly, which families of acyclic digraphs (with bounded maximum average degree) are linearly unavoidable? In this direction, Fox, He, and Widgerson [10] showed that for every acyclic digraph DD of maximum degree Δ\Delta, if there is a partition S1,,ShS_{1},\dots,S_{h} of V(D)V(D) such that for every arc uvuv of DD, there exists i[h1]i\in[h-1] such that uSiu\in S_{i} and vSi+1v\in S_{i+1}, then unvd(D)h10ΔlogΔ|V(D)|\operatorname{unvd}(D)\leqslant h^{10\Delta\log\Delta}|V(D)|.

Our results

In this paper, we give some more partial answers to the above second question by proving some new families of acyclic digraphs to be linearly unavoidable. We are in particular interested in operations to form linearly unavoidable families from others.

Let kk be a positive integer. The kk-blow-up of a digraph DD, denoted by D[k]D[k], is obtained by blowing-up each vertex into an independent set of size kk. Formally, D[k]D[k] is defined by

V(D[k])\displaystyle V(D[k]) =\displaystyle= {(v,i)vV(D),i[k]}, and\displaystyle\{(v,i)\mid v\in V(D),i\in[k]\},\mbox{ and }
A(D[k])\displaystyle A(D[k]) =\displaystyle= {(u,i)(v,j)uvA(D),i[k],j[k],ij}.\displaystyle\{(u,i)(v,j)\mid uv\in A(D),i\in[k],j\in[k],i\neq j\}.

We believe that if a family \cal F is linearly unavoidable, then the family [k]={D[k]D}{\cal F}[k]=\{D[k]\mid D\in{\cal F}\} of its kk-blow-ups is also linearly unavoidable.

Conjecture 7.

If \cal F is linearly unavoidable and has bounded maximum average degree, then [k]{\cal F}[k] is also linearly unavoidable.

Observe first that the bounded maximum average degree condition in the above conjecture is necessary. Indeed, for every nn, let DnD_{n} be the digraph of order nn which is the disjoint union of a transitive tournament of order logn+1\lfloor\log n\rfloor+1 and nlogn1n-\lfloor\log n\rfloor-1 isolated vertices. Each DnD_{n} is clearly nn-unavoidable, but Dn[k]D_{n}[k] has maximum average degree klognk\lfloor\log n\rfloor and so unvd(Dn)2klogn/212nk/2\operatorname{unvd}(D_{n})\geqslant 2^{k\lfloor\log n\rfloor/2}\geqslant\frac{1}{2}n^{k/2} by Proposition 2.

Observe that the kk-blow-up of the directed path of order nn is contained in the 2k2k-th power of the directed path of order knkn. Thus, the result of Draganić et al. [5] implies Conjecture 7 for directed paths. As a more general evidence to this conjecture, we show in Section 3 that it holds for oriented trees (and thus more generally oriented forests). Precisely, we prove in Theorem 18 that the kk-blow-up of an oriented tree of order nn is (210+18kkn)(2^{10+18k}\cdot kn)-unavoidable. This upper bound is certainly not tight. However, the maximum average degree of the kk-blow-up of a tree of order nn is 2k2k/n2k-2k/n. Therefore, by Proposition 2, the optimal bound is of the form 2Θ(k)kn2^{\Theta(k)}\cdot kn. Hence we are left with the following question.

Problem 8.

What is the infimum of all the constants CC such that for every large enough integer nn, the kk-blow-up of every oriented tree of order nn is (Ckkn)(C^{k}\cdot kn)-unavoidable?

Generalizing Proposition 1, we believe that adding a vertex cannot affect too much the unavoidability, in a sense that it can at most multiply it by a constant factor.

Conjecture 9.

There exists a constant CC such that unvd(D)Cunvd(Dv)\operatorname{unvd}(D)\leqslant C\cdot\operatorname{unvd}(D-v) for every acyclic digraph DD and for every vertex vv of DD.

Let AA be an acyclic digraph and let kk be a positive integer. An acyclic digraph DD is a kk-extension of AA if there exists a set SS of kk vertices such that DS=AD-S=A. The set SS is the added set and its vertices are the added vertices. A vertex is universal in a digraph DD if N(v)N+(v){v}=V(D)N^{-}(v)\cup N^{+}(v)\cup\{v\}=V(D). A kk-extension DD of AA is full, if its added vertices are universal in DD. A kk-extension DD of AA is a kk-twin extension, if all added vertices have the same in- and out-neighbourhood in V(DS)V(D-S). Conjecture 9, together with Conjecture 4, implies the following.

Conjecture 10.

There is an absolute constant CC such that for every integer kk, every kk-extension of an oriented tree of order nn is Ck(2n2)C^{k}(2n-2)-unavoidable.

More generally, Conjecture 9 would imply the following.

Conjecture 11.

If \cal F is linearly unavoidable, then the family of kk-extensions of digraphs in \cal F is also linearly unavoidable.

In Section 4, we prove this last conjecture for the family of oriented trees (and thus also of oriented forest). More precisely; we show in Corollary 25 that every kk-extension of a sufficiently large tree FF is (23(2k+22)|V(F)|)\left(2\cdot 3^{\binom{2k+2}{2}}\cdot|V(F)|\right)-unavoidable.

This upper bound on the unavoidability of extension of forests is certainly not tight. In Proposition 28, we show that the unavoidability of some kk-extensions of the arcless digraph on nn vertices is 2kno(n)2^{k}n-o(n) as n+n\to+\infty. Hence we are left with the following analogue to Problem 8.

Problem 12.

What is the minimum function ff such that every kk-extension of every forest of order nn is (2f(k)n)(2^{f(k)}\cdot n)-unavoidable?

The above-mentioned results show that the function ff is at least linear and at most quadratic. The same question applies to any subfamily, starting with that of empty digraphs. Generally, determining the precise value or good approximations of the unavoidability of kk-extensions of oriented trees would be interesting. In Section 5, we show an almost tight upper bound on the unavoidability of full kk-twin extensions and the exact unavoidability of the full 11-extensions of directed paths.

2 Preliminaries

2.1 Basic properties of tournaments

We will make use of the following folklore properties of tournaments.

Lemma 13 ([2, Proposition 2.2.2]).

Let kk be a positive integer. Every tournament has at most 2k12k-1 vertices of out-degree less than kk.

The out-section of a vertex vv in a tournament TT, denoted by ST+(v)S^{+}_{T}(v) or simply S+(v)S^{+}(v), is the set of vertices that can be reached from vv by a directed path. Similarly, the in-section of a vertex vv in a tournament TT, denoted by ST(v)S^{-}_{T}(v) or simply S(v)S^{-}(v) is the set of vertices from which vv can be reached by a directed path. We set sT+(v)=|ST+(v)|s^{+}_{T}(v)=|S^{+}_{T}(v)| and sT(v)=|ST(v)|s^{-}_{T}(v)=|S^{-}_{T}(v)|. An out-generator (resp. in-generator) of TT is a vertex vv such that sT+(v)=|V(T)|s^{+}_{T}(v)=|V(T)| (resp. sT(v)=|V(T)|s^{-}_{T}(v)=|V(T)|).

Lemma 14 ([2, Proposition 2.7.6]).

Let TT be a tournament and vv a vertex of TT. In TT, vv is the initial (resp. terminal) vertex of a directed path of order s+(v)s^{+}(v) (resp. s(v)s^{-}(v)). In particular, if vv is an out-generator (resp. in-generator) of TT, then vv is the initial vertex (resp. terminal vertex) of a directed Hamiltonian path.

Let TT be a tournament, and AA and BB be two disjoint sets of vertices of TT. If every vertex of AA dominates every vertex of BB, then we write ABA\Rightarrow B.

Lemma 15.

Let TT be a tournament and let OO be the set of vertices that are initial vertices of a directed path of order kk in TT. Then OV(T)OO\Rightarrow V(T)\setminus O.

Proof.

Suppose for contradiction that there exists uV(T)Ou\in V(T)\setminus O and vOv\in O such that uvA(T)uv\in A(T). Since vOv\in O, there is a path PP of order kk in TT with initial vertex vv. Hence s+(u)|V(P)|ks^{+}(u)\geqslant|V(P)|\geqslant k and so uu is the initial vertex of a directed path of order kk by Lemma 14. ∎

2.2 Median order

The notion of median order has been introduced in [14], see also [2, Chapter 2] and [5]. A median order of a digraph DD is a linear order (v1,v2,,vn)(v_{1},v_{2},\ldots,v_{n}) of its vertex set such that |{(vi,vj)i<j}||\{(v_{i},v_{j})\mid i<j\}|, the number of forward arcs (i.e. directed from left to right), is as large as possible. Let us first note some basic properties of median orders of tournaments.

Lemma 16.

Let TT be a tournament and (v1,v2,,vn)(v_{1},v_{2},\ldots,v_{n}) a median order of TT. Then, for any two indices i,ji,j with 1i<jn1\leqslant i<j\leqslant n:

  1. (M1)

    the suborder (vi,vi+1,,vj)(v_{i},v_{i+1},\ldots,v_{j}) is a median order of the induced subtournament T{vi,vi+1,,vj}T\langle\{v_{i},v_{i+1},\ldots,v_{j}\}\rangle;

  2. (M2)

    viv_{i} dominates at least half of the vertices vi+1,vi+2,,vjv_{i+1},v_{i+2},\ldots,v_{j}, and vjv_{j} is dominated by at least half of the vertices vi,vi+1,,vj1v_{i},v_{i+1},\ldots,v_{j-1}, in particular viv_{i} dominates vi+1v_{i+1};

  3. (M3)

    if vi+2viv_{i+2}\rightarrow v_{i}, then vi1{vi,vi+1,vi+2}v_{i-1}\Rightarrow\{v_{i},v_{i+1},v_{i+2}\} when i2i\geqslant 2, and {vi,vi+1,vi+2}vi+3\{v_{i},v_{i+1},v_{i+2}\}\Rightarrow v_{i+3} when in3i\leqslant n-3;

  4. (M4)

    if vnv1v_{n}\rightarrow v_{1}, then there exists \ell such that v1vvnv_{1}\rightarrow v_{\ell}\rightarrow v_{n}; moreover, if |{v1vvn}|=1|\{\ell\mid v_{1}\rightarrow v_{\ell}\rightarrow v_{n}\}|=1, then both (v2,,vn,v1)(v_{2},\ldots,v_{n},v_{1}) and (vn,v1,,vn1)(v_{n},v_{1},\ldots,v_{n-1}) are median orders of TT;

  5. (M5)

    if (vi,,vj)(v_{i}^{\prime},\dots,v_{j}^{\prime}) is a median order of T{vi,vi+1,,vj}T\langle\{v_{i},v_{i+1},\ldots,v_{j}\}\rangle, then (v1,,vi1,vi,,vj,vj+1,,vn)(v_{1},\ldots,v_{i-1},v_{i}^{\prime},\dots,v_{j}^{\prime},v_{j+1},\dots,v_{n}) is also a median order of TT.

Proof.

Items (M1)(M2), and (M5) follow easily from the definition.

(M3) Assume for a contradiction that vi+2viv_{i+2}\to v_{i} but there is an arc from vi,vi+1,vi+2v_{i},v_{i+1},v_{i+2} to vi1v_{i-1}. By (M2) there is exactly one such arc and its tail uu belongs to {vi+1,vi+2}\{v_{i+1},v_{i+2}\}. If u=vi+2u=v_{i+2}, then vi+2v_{i+2} has two out-neighbours in {vi1,vi,vi+1}\{v_{i-1},v_{i},v_{i+1}\}, contradicting (M2). If u=vi+1u=v_{i+1}, then (vi+1,vi1,vi+2,vi)(v_{i+1},v_{i-1},v_{i+2},v_{i}) has less forward arcs than (vi1,vi,vi+1,vi+2)(v_{i-1},v_{i},v_{i+1},v_{i+2}), a contradiction to (M1). By directional duality, we also have {vi,vi+1,vi+2}vi+3\{v_{i},v_{i+1},v_{i+2}\}\Rightarrow v_{i+3}.

(M4) Suppose that vnv1v_{n}\to v_{1}. By (M2), v1v_{1} dominates at least half of v2,,vnv_{2},\dots,v_{n}, and so more than half of v2,,vn1v_{2},\dots,v_{n-1}. Symmetrically, vnv_{n} is dominated by more than half of v2,,vn1v_{2},\dots,v_{n-1}. Hence there exists {2,,n1}\ell\in\{2,\dots,n-1\} such that v1vvnv_{1}\to v_{\ell}\to v_{n}. Now suppose that this \ell is unique. Hence v1v_{1} dominates exactly n12\frac{n-1}{2} vertices among v2,,vnv_{2},\dots,v_{n}. It follows that the number of forward arcs in (v2,,vn,v1)(v_{2},\dots,v_{n},v_{1}) is |{(vi,vj)1i<jn}|d+(v1)+d(v1)=|{(vi,vj)1i<jn}||\{(v_{i},v_{j})\mid 1\leqslant i<j\leqslant n\}|-d^{+}(v_{1})+d^{-}(v_{1})=|\{(v_{i},v_{j})\mid 1\leqslant i<j\leqslant n\}|, which is maximum since (v1,,vn)(v_{1},\dots,v_{n}) is a median order. This proves that (v1,,vn,v1)(v_{1},\dots,v_{n},v_{1}) is a median order. Symmetrically, (vn,v1,,vn1)(v_{n},v_{1},\dots,v_{n-1}) is a median order too. ∎

2.3 A Kővári-Sós-Turán-like lemma

The following lemma is inspired by the classical result by Kővári, Sós, and Turán [17], and a similar one was recently used in [5].

Lemma 17.

Let α,ε\alpha,\varepsilon be two positive reals and kk a positive integer. Let (A,B,E)(A,B,E) be a bipartite graph with |A|k12ε|A|\geqslant\frac{k}{\frac{1}{2}-\varepsilon} and α((12ε)|A|k)(|A|k)\displaystyle\alpha\leqslant\frac{\binom{(\frac{1}{2}-\varepsilon)|A|}{k}}{\binom{|A|}{k}}, such that d(v)(12ε)|B|d(v)\geqslant(\frac{1}{2}-\varepsilon)|B| for every vAv\in A. Then there exist AAA^{\star}\subseteq A and BBB^{\star}\subseteq B such that

  1. 1.

    |A|k|A^{\star}|\geqslant k,

  2. 2.

    |B|α|B||B^{\star}|\geqslant\alpha|B|, and

  3. 3.

    AA^{\star} is complete to BB^{\star}.

Proof.

Suppose for contradiction that the result does not hold. We will count by two different methods the number MM of pairs (X,b)(X,b) with XAX\subseteq A of size kk and bBb\in B such that XN(b)X\subseteq N(b). On the first hand, for every XAX\subseteq A of size kk, there are less than α|B|\alpha|B| vertices bb in BB such that XN(b)X\subseteq N(b). We thus have

M<(|A|k)α|B|.M<\binom{|A|}{k}\alpha|B|.

On the other hand, by Jensen’s inequality, we have

M=bB(d(b)k)\displaystyle M=\sum_{b\in B}\binom{d(b)}{k} \displaystyle\geqslant |B|(1|B|bBd(b)k)=|B|(1|B|aAd(a)k)\displaystyle|B|\binom{\frac{1}{|B|}\sum_{b\in B}d(b)}{k}=|B|\binom{\frac{1}{|B|}\sum_{a\in A}d(a)}{k}
\displaystyle\geqslant |B|(|A||B|(12ε)|B|k)(|A|k)α|B|.\displaystyle|B|\binom{\frac{|A|}{|B|}(\frac{1}{2}-\varepsilon)|B|}{k}\geqslant\binom{|A|}{k}\alpha|B|.

This contradiction proves the lemma. ∎

Note that α=2k12ε\alpha=2^{-\frac{k}{\frac{1}{2}-\varepsilon}} satisfies the hypothesis of Lemma 17. Indeed, simple computations show that ((12ε)|A|k)(|A|k)\frac{\binom{(\frac{1}{2}-\varepsilon)|A|}{k}}{\binom{|A|}{k}} is a non-increasing function of |A||A|, and as |A|k12ε|A|\geqslant\frac{k}{\frac{1}{2}-\varepsilon} we have

((12ε)|A|k)(|A|k)(kk)(k12εk)2k12ε.\frac{\binom{(\frac{1}{2}-\varepsilon)|A|}{k}}{\binom{|A|}{k}}\geqslant\frac{\binom{k}{k}}{\displaystyle\binom{\frac{k}{\frac{1}{2}-\varepsilon}}{\scriptstyle k}}\geqslant 2^{-\frac{k}{\frac{1}{2}-\varepsilon}}.

3 Blow-ups of oriented trees

In this section, we show that for every fixed integer kk, the class of kk-blow-ups of oriented trees is linearly avoidable.

Theorem 18.

Let FF be an oriented tree of order nn and kk a positive integer. Every tournament on at least 210+18kkn2^{10+18k}kn vertices contains F[k]F[k].

Proof.

We root FF on an arbitrary vertex ss. Let q=21+9kq=2^{1+9k}, p=25+32kkp=\lceil 2^{5+3^{2}k}k\rceil and ε=12q\varepsilon=\frac{1}{2q}. Observe that we have ε16\varepsilon\leqslant\frac{1}{6}, q21+k(12ε)2q\geqslant 2^{1+\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}, and pqk(12ε)2p\geqslant q\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}. Let N=12(4qn+1)p12210+18kknN=\frac{1}{2}(4qn+1)p\leqslant\frac{1}{2}2^{10+18k}k\cdot n (using 4qn+12×4qn4qn+1\leqslant 2\times 4qn and p2×25+32kkp\leqslant 2\times 2^{5+3^{2}k}k) and let TT be a tournament of order 2N2N. We will show that TT has a subdigraph isomorphic to F[k]F[k]. Let vN,,vN1v_{-N},\dots,v_{N-1} be a median order of TT. We partition vN,,vN1v_{-N},\dots,v_{N-1} into 4qn+14qn+1 intervals Vi={vpi,,vp(i+1)1}V_{i}=\{v_{pi},\dots,v_{p(i+1)-1}\} for i{2qn,,2qn}i\in\{-2qn,\dots,2qn\}. We will find a copy of F[k]F[k] in DD iteratively, by adding at each step the out-children (i.e. its children which are out-neighbours) or the in-children (i.e. its children which are in-neighbours) of a node in FF. Let F1,,FrF_{1},\dots,F_{r} be a sequence of subtrees of FF such that

  1. (i)(i)

    F1F_{1} is the singleton {s}\{s\},

  2. (ii)(ii)

    for every ir12i\leqslant\frac{r-1}{2}, F2i+1F_{2i+1} is obtained from F2iF_{2i} by adding all the out-children in FF of a vertex xV(F2i)x\in V(F_{2i}) without any out-children in F2iF_{2i},

  3. (iii)(iii)

    for every ir22i\leqslant\frac{r-2}{2}, F2i+2F_{2i+2} is obtained from F2i+1F_{2i+1} by adding all the in-children in FF of a vertex xV(F2i+1)x\in V(F_{2i+1}) without any in-children in F2i+1F_{2i+1}, and

  4. (iv)(iv)

    Fr=FF_{r}=F.

Observe that such a sequence exists with r2n1r\leqslant 2n-1. Moreover, we say that a vertex xx in FiF_{i} is fully processed if all its children in FF are also in FiF_{i}, half processed if all its out-children in FF are in FiF_{i}. Otherwise we say that xx is not processed. We prove the following property by induction:

For every j{1,,r}j\in\{1,\dots,r\}, there is an injection ι:V(Fj){2q|V(Fj)|,,2q|V(Fj)|}\iota:V(F_{j})\to\{-2q|V(F_{j})|,\dots,2q|V(F_{j})|\} and for every xV(Fj)x\in V(F_{j}) there exists a set UxVι(x)U_{x}\subseteq V_{\iota(x)} such that

  1. (i)(i)

    |Ux|k(12ε)2|U_{x}|\geqslant\frac{k}{\left(\frac{1}{2}-\varepsilon\right)^{2}} if xx is not processed,

  2. (ii)(ii)

    |Ux|k12ε|U_{x}|\geqslant\frac{k}{\frac{1}{2}-\varepsilon} if xx is half processed,

  3. (iii)(iii)

    |Ux|k|U_{x}|\geqslant k if xx is fully processed,

  4. (iv)(iv)

    if xyxy is an arc in FjF_{j}, then every vertex in UxU_{x} dominates UyU_{y}, and

  5. (v)(v)

    for every i{2q|V(Fj)|,,2q|V(Fj)|}i\in\{-2q|V(F_{j})|,\dots,2q|V(F_{j})|\} a fraction of at most 1q\frac{1}{q} of the elements of the interval {i,,2q|V(Fj)|}\{i,\dots,2q|V(F_{j})|\} (respectively {2q|V(Fj)|,,i}\{-2q|V(F_{j})|,\dots,i\}) is in {ι(x)xV(Fj)}\{\iota(x)\mid x\in V(F_{j})\}. More precisely,

    |{i,,2q|V(Fj)|}{ι(x)xV(Fj)}|\displaystyle|\{i,\dots,2q|V(F_{j})|\}\cap\{\iota(x)\mid x\in V(F_{j})\}| 1q(2q|V(Fj)|i+1),\displaystyle\leqslant\frac{1}{q}\left(2q|V(F_{j})|-i+1\right),
    and |{2q|V(Fj)|,,i}{ι(x)xV(Fj)}|\displaystyle\text{and~{}~{}~{}~{}}|\{-2q|V(F_{j})|,\dots,i\}\cap\{\iota(x)\mid x\in V(F_{j})\}| 1q(i+2q|V(Fj)|+1).\displaystyle\leqslant\frac{1}{q}\left(i+2q|V(F_{j})|+1\right).

Observe that, for j=1j=1, setting ι(s)=0\iota(s)=0 and taking for UsU_{s} any set in V0V_{0} of size pk(12ε)2p\geqslant\frac{k}{\left(\frac{1}{2}-\varepsilon\right)^{2}} satisfies the invariant. Note also that, for j=rj=r, the invariant implies the theorem. Now suppose that the property holds for some j<rj<r. Let ι\iota and (Ux)xFj(U_{x})_{x\in F_{j}} witnessing that the property holds at step jj. Let xx be the vertex in FjF_{j} such that Fj+1F_{j+1} is obtained from FjF_{j} by adding the out-children of xx in FF, or the in-children of xx in FF. Our goal is to find a new value UxU^{\prime}_{x} for UxU_{x}, and values for ι(y),Uy\iota(y),U_{y} for every out-children (respectively in-children) yy of xx.

Case 1:

Fj+1F_{j+1} is obtained from FjF_{j} by adding the \ell out-children of xx in FF.

Let A=UxA=U_{x} and B=ι(x)<i2q|V(Fj)|+qViB=\bigcup_{\iota(x)<i\leqslant 2q|V(F_{j})|+q\ell}V_{i}. Note that |B|p(2q|V(Fj)|+qι(x))qp|B|\geqslant p(2q|V(F_{j})|+\ell q-\iota(x))\geqslant\ell qp. We first justify that every vertex in Vι(x)V_{\iota(x)} dominates at least (12ε)|B|(\frac{1}{2}-\varepsilon)|B| vertices in BB. To this purpose, let viv_{i} be any vertex in Vι(x)V_{\iota(x)}, so i{pι(x),,p(ι(x)+1)1}i\in\{p\iota(x),\dots,p(\iota(x)+1)-1\}. Let XX be the set of vertices {vi+1,,vp(ι(x)+1)1}\{v_{i+1},\dots,v_{p(\iota(x)+1)-1}\}, which is disjoint from BB (XX is empty if i=p(ι(x)+1)1i=p(\iota(x)+1)-1). Observe that

N+(vi)B=(N+(vi)(XB))(N+(vi)X).N^{+}(v_{i})\cap B=\Big{(}N^{+}(v_{i})\cap(X\cup B)\Big{)}\setminus\Big{(}N^{+}(v_{i})\cap X\Big{)}.

Since XB={vi+1,,vp(ι(x)+1)1+|B|}X\cup B=\{v_{i+1},\dots,v_{p(\iota(x)+1)-1+|B|}\}, and because vN,,vN1v_{-N},\dots,v_{N-1} is a median order, by (M2) we have |N+(vi)(XB)|12|XB||N^{+}(v_{i})\cap(X\cup B)|\geqslant\frac{1}{2}|X\cup B|. Therefore, we obtain

|N+(vi)B|12|XB||X|=12|B|12|X|12(|B|p)(1212q)|B|=(12ε)|B|.|N^{+}(v_{i})\cap B|\geqslant\frac{1}{2}|X\cup B|-|X|=\frac{1}{2}|B|-\frac{1}{2}|X|\geqslant\frac{1}{2}(|B|-p)\geqslant\left(\frac{1}{2}-\frac{1}{2q}\right)|B|=\left(\frac{1}{2}-\varepsilon\right)|B|.

This shows that every vertex in Vι(x)V_{\iota(x)} dominates at least (12ε)|B|(\frac{1}{2}-\varepsilon)|B| vertices in BB, and so does every vertex in AA. By Lemma 17 applied for α=2k(12ε)2\alpha=2^{-\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}, there exist AUxA^{\star}\subseteq U_{x} and BBB^{\star}\subseteq B such that every vertex in AA^{\star} dominates BB^{\star}, |B|2k(12ε)2|B||B^{\star}|\geqslant 2^{-\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}|B|, and |A|k12ε|A^{\star}|\geqslant\frac{k}{\frac{1}{2}-\varepsilon}. We set Ux=AU^{\prime}_{x}=A^{\star}.

Let II be the set of indices i{ι(x)+1,,2q|V(Fj)|+q}i\in\{\iota(x)+1,\dots,2q|V(F_{j})|+q\ell\} such that |BVi|k(12ε)2|B^{\star}\cap V_{i}|\geqslant\frac{k}{(\frac{1}{2}-\varepsilon)^{2}} and i{ι(z)zV(Fj)}i\notin\{\iota(z)\mid z\in V(F_{j})\}. In what follows we show that II is large enough, which will enable us to find UyU_{y} for every out-child yy of xx in FF.

Claim 1.

|I||I|\geqslant\ell

Proof of claim.  Suppose that this does not hold, so |I|1|I|\leqslant\ell-1. We define J,I,J,I^{\star}, and IιI^{\iota} as follows:

J\displaystyle J ={ι(x)+1,,2q|V(Fj)|+ql},\displaystyle=\{\iota(x)+1,\dots,2q|V(F_{j})|+ql\},
I\displaystyle I^{\star} ={iJ|ViB|k(12ε)2}, and\displaystyle=\{i\in J\mid|V_{i}\cap B^{\star}|\geqslant\tfrac{k}{(\frac{1}{2}-\varepsilon)^{2}}\},\text{ and}
Iι\displaystyle I^{\iota} ={ι(z)zV(j)}J.\displaystyle=\{\iota(z)\mid z\in V(j)\}\cap J.

Observe that I=IIιI=I^{\star}\setminus I^{\iota} by definition. By the induction hypothesis, |Iι|1q(|J|q)|I^{\iota}|\leqslant\frac{1}{q}(|J|-q\ell). Since |I|1|I|\leqslant\ell-1 by assumption, we obtain |I|1+1q(|J|q)=|J|q1|I^{\star}|\leqslant\ell-1+\frac{1}{q}(|J|-q\ell)=\frac{|J|}{q}-1. This leads to the following inequalities, in which we also use |B|=p|J||B|=p|J|

|B|=iJI|ViB|+iJI|ViB|(|J|q1)p+(q1q|J|+1)k(12ε)2=(|B|q)+(q1q|B|pk(12ε)2)+(k(12ε)2p)<(1q+kp(12ε)2)|B|2q|B|2k(12ε)2|B|.\begin{split}|B^{\star}|&=\sum_{i\in J\cap I^{\star}}|V_{i}\cap B^{\star}|+\sum_{i\in J\setminus I^{\star}}|V_{i}\cap B^{\star}|\\ &\leqslant\left(\frac{|J|}{q}-1\right)p+\left(\frac{q-1}{q}|J|+1\right)\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}\\ &=\left(\frac{|B|}{q}\right)+\left(\frac{q-1}{q}\frac{|B|}{p}\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}\right)+\left(\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}-p\right)\\ &<\left(\frac{1}{q}+\frac{k}{p(\frac{1}{2}-\varepsilon)^{2}}\right)|B|\\ &\leqslant\frac{2}{q}|B|\\ &\leqslant 2^{-\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}|B|.\end{split}

This is a contradiction since |B|2k(12ε)2|B||B^{\star}|\geqslant 2^{-\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}|B|. \lozenge

We can now extend ι\iota to the out-children of xx by taking values in II, and then we can arbitrarily take UzBVι(z)U_{z}\subseteq B^{\star}\cap V_{\iota(z)} for every out-child zz of xx. Recall that |V(Fj+1)|=|V(Fj)|+|V(F_{j+1})|=|V(F_{j})|+\ell. The invariant still holds because we add 2qp2\ell qp vertices at the end and at the beginning of the median order:

  1. (i)(i)

    For every i{2q|V(Fj)|2q,,2q|V(Fj)|}i\in\{-2q|V(F_{j})|-2q\ell,\dots,2q|V(F_{j})|\}, |{i,,2q|V(Fj)|+2q}{ι(z)zV(Fj+1)}|1q(2q|V(Fj)|i+1)+1q(2q|V(Fj+1)|i+1)|\{i,\dots,2q|V(F_{j})|+2q\ell\}\cap\{\iota(z)\mid z\in V(F_{j+1})\}|\leqslant\frac{1}{q}(2q|V(F_{j})|-i+1)+\ell\leqslant\frac{1}{q}(2q|V(F_{j+1})|-i+1).

  2. (ii)(ii)

    For every i{2q|V(Fj)|+1,,2q|V(Fj)|+q}i\in\{2q|V(F_{j})|+1,\dots,2q|V(F_{j})|+q\ell\}, |{i,,2q|V(Fj)|+2q}{ι(z)zV(Fj+1)}|1q(q)1q(2q|V(Fj+1)|i+1)|\{i,\dots,2q|V(F_{j})|+2q\ell\}\cap\{\iota(z)\mid z\in V(F_{j+1})\}|\leqslant\ell\leqslant\frac{1}{q}(q\ell)\leqslant\frac{1}{q}(2q|V(F_{j+1})|-i+1).

  3. (iii)(iii)

    For every i{2q|V(Fj)|+q+1,,2q|V(Fj)|+2q}i\in\{2q|V(F_{j})|+q\ell+1,\dots,2q|V(F_{j})|+2q\ell\}, |{i,,2q|V(Fj)|+2q}{ι(z)zV(Fj+1)}|=01q(2q|V(Fj+1)|i+1)|\{i,\dots,2q|V(F_{j})|+2q\ell\}\cap\{\iota(z)\mid z\in V(F_{j+1})\}|=0\leqslant\frac{1}{q}(2q|V(F_{j+1})|-i+1).

  4. (iv)(iv)

    For every i{2q|V(Fj)|,,2q|V(Fj)|+2q}i\in\{-2q|V(F_{j})|,\dots,2q|V(F_{j})|+2q\ell\}, |{2q|V(Fj)|2q,,i}{ι(z)zV(Fj+1)}|1q(i+2q|V(Fj)|+1)+1q(i+2q|V(Fj+1)|+1)|\{-2q|V(F_{j})|-2q\ell,\dots,i\}\cap\{\iota(z)\mid z\in V(F_{j+1})\}|\leqslant\frac{1}{q}(i+2q|V(F_{j})|+1)+\ell\leqslant\frac{1}{q}(i+2q|V(F_{j+1})|+1).

  5. (v)(v)

    For every i{2q2q|V(Fj)|,,2q|V(Fj)|1}i\in\{-2q\ell-2q|V(F_{j})|,\dots,-2q|V(F_{j})|-1\}, |{2q|V(Fj)|2q,,i}{ι(z)zV(Fj+1)}|=01q(i+2q|V(Fj+1)|+1)|\{-2q|V(F_{j})|-2q\ell,\dots,i\}\cap\{\iota(z)\mid z\in V(F_{j+1})\}|=0\leqslant\frac{1}{q}(i+2q|V(F_{j+1})|+1).

Case 2:

Fj+1F_{j+1} is obtained from FjF_{j} by adding the \ell in-children of xx in FF.

Let A=Uι(x)A=U_{\iota(x)} and B=2q|V(Fj)|qi<ι(x)ViB=\bigcup_{-2q|V(F_{j})|-q\ell\leqslant i<\iota(x)}V_{i}. Since xx is half processed, we have |A|k12ε|A|\geqslant\frac{k}{\frac{1}{2}-\varepsilon}. Moreover |B|qp|B|\geqslant q\ell p and every vertex in AA is dominated by at least 12|B||A|(12ε)|B|\frac{1}{2}|B|-|A|\geqslant(\frac{1}{2}-\varepsilon)|B| vertices in BB. Using Lemma 17, we deduce that there are sets AA,BBA^{\star}\subseteq A,B^{\star}\subseteq B with BAB^{\star}\Rightarrow A^{\star}, |A|k|A^{\star}|\geqslant k and |B|2k12ε|B|2k(12ε)2|B||B^{\star}|\geqslant 2^{-\frac{k}{\frac{1}{2}-\varepsilon}}|B|\geqslant 2^{-\frac{k}{(\frac{1}{2}-\varepsilon)^{2}}}|B|. We set Ux=AU^{\prime}_{x}=A^{\star}. We now find UyU_{y} for every yy in-children of xx in FF. With a proof almost identical to the one of Claim 1 we deduce the following claim.

Claim 2.

There is a set I{2q|V(Fj)|q,,ι(x)1}I\subseteq\{-2q|V(F_{j})|-q\ell,\dots,\iota(x)-1\} of size at least \ell such that ι(z)I\iota(z)\not\in I for every zV(Fj)z\in V(F_{j}) and |BVi|k(12ε)2|B^{\star}\cap V_{i}|\geqslant\frac{k}{(\frac{1}{2}-\varepsilon)^{2}} for every iIi\in I.

Then we build Uy,ι(y)U_{y},\iota(y) for every in-children yy of xx as in the first case. This concludes the proof of the theorem.∎

4 Extensions of oriented trees

In this section, we show that for every fixed integer kk, the class of kk-extensions of oriented trees is linearly unavoidable. Our method is based on the fact that the number of copies of TT2k+1TT_{2k+1} in any NN-vertex tournaments is in Ωk(N2k+1)\Omega_{k}(N^{2k+1}). We start with the case k=1k=1, which is a good warm-up for the general case.

Lemma 19 (Folklore).

Every tournament of order NN contains at least 18N(N1)(N3)\frac{1}{8}N(N-1)(N-3) copies of TT3TT_{3}.

Proof.

Let TT be a tournament of order NN. Let t\vec{t} be the number of directed triangles in TT and let tt3tt_{3} be the number of TT3TT_{3} in TT. The number of copies of P3\vec{P}_{3} (the directed path of order 33) in TT is vV(T)d(v)d+(v)N(N12)2=14N(N1)2\sum_{v\in V(T)}d^{-}(v)d^{+}(v)\leqslant N\left(\frac{N-1}{2}\right)^{2}=\frac{1}{4}N(N-1)^{2}. But the number of P3\vec{P}_{3} is also tt3+3t=tt3+3((N3)tt3)=3(N3)2tt3tt_{3}+3\vec{t}=tt_{3}+3(\binom{N}{3}-tt_{3})=3\binom{N}{3}-2tt_{3}. We deduce that 3(N3)2tt314N(N1)23\binom{N}{3}-2tt_{3}\leqslant\frac{1}{4}N(N-1)^{2} and so tt312(N(N1)(N2)2N(N1)24)=18N(N1)(N3)tt_{3}\geqslant\frac{1}{2}\left(\frac{N(N-1)(N-2)}{2}-\frac{N(N-1)^{2}}{4}\right)=\frac{1}{8}N(N-1)(N-3). ∎

We will also need the following well-known lemma. Recall that a hypergraph \mathcal{H} is a pair (V,E)(V,E) where VV is a finite set and EE is a subset of 2V2^{V}. We call the elements of VV the vertices of \mathcal{H}, and the elements of EE the hyperedges of \mathcal{H}. An hypergraph =(V,E)\mathcal{H}^{\prime}=(V^{\prime},E^{\prime}) is a subhypergraph of \mathcal{H}, and we write \mathcal{H}^{\prime}\subseteq\mathcal{H}, if VVV^{\prime}\subseteq V and EEE^{\prime}\subseteq E.

Lemma 20 (Folklore).

Let \mathcal{H} be a hypergraph on nn vertices and with mm hyperedges. There is a hypergraph \mathcal{H}^{\prime}\subseteq\mathcal{H} such that every vertex of \mathcal{H}^{\prime} belongs to at least mn\frac{m}{n} hyperedges of \mathcal{H}^{\prime}.

Proof.

We proceed by induction on nn. If every vertex in \mathcal{H} is in at least mn\frac{m}{n} hyperedges we are done. Otherwise there is a vertex uu in less than mn\frac{m}{n} hyperedges, then u\mathcal{H}-u has mmmn=mn(n1)m^{\prime}\geqslant m-\frac{m}{n}=\frac{m}{n}(n-1) hyperedges and n=n1n^{\prime}=n-1 vertices. Hence, by the induction hypothesis, u\mathcal{H}-u has a subhypergraph with minimum degree at least mnmn\frac{m^{\prime}}{n^{\prime}}\geqslant\frac{m}{n} as wanted. ∎

Given an acyclic digraph DD, we denote by unvd(D)\operatorname{unvd}^{*}(D) the maximum of unvd(H)+|V(D)V(H)|\operatorname{unvd}(H)+|V(D)\setminus V(H)| over all the subdigraphs HH of DD.

Theorem 21.

Let DD be a 11-extension of an oriented tree FF on nn vertices, then unvd(D)8unvd(F)+3\operatorname{unvd}(D)\leqslant 8\operatorname{unvd}^{*}(F)+3.

By applying Theorem 5 to bound unvd(F)\operatorname{unvd}^{*}(F), we obtain the following.

Corollary 22.

There exists a constant n0n_{0} such that the following holds. Let DD be a 11-extension of an oriented tree FF on at least n0n_{0} vertices, then unvd(D)16|V(F)|13\operatorname{unvd}(D)\leqslant 16|V(F)|-13.

An embedding φ:HD\varphi\colon H\to D of a digraph HH in a digraph DD is an injective function φ:V(H)V(D)\varphi\colon V(H)\to V(D) such that φ(u)φ(v)A(D)\varphi(u)\varphi(v)\in A(D) for every uvA(H)uv\in A(H). Observe that GG contains a copy of HH if and only if there is an embedding of HH in GG.

Proof of Theorem 21.

Let DD be a 11-extension of FF with added vertex ss. We assume that DD is a full 11-extension of FF, for otherwise we just add missing arcs. Let TT be a tournament on N=8unvd(F)+3N=8\operatorname{unvd}^{*}(F)+3 vertices. By Lemma 19, the number of TT3TT_{3} in TT is at least 18N(N1)(N3)\frac{1}{8}N(N-1)(N-3). We deduce that there is a vertex tt which is the middle vertex (i.e. the vertex with in- and out-degree 11) of at least 18(N1)(N3)\frac{1}{8}(N-1)(N-3) copies of TT3TT_{3} in TT. In other words there are at least 18(N1)(N3)\frac{1}{8}(N-1)(N-3) arcs from N(t)N^{-}(t) to N+(t)N^{+}(t). Consider the bipartite graph H0=(NT(t),NT+(t),E)H_{0}=(N^{-}_{T}(t),N^{+}_{T}(t),E) where E={uwA(T)uNT(t),wNT+(t)}E=\{uw\in A(T)\mid u\in N^{-}_{T}(t),w\in N^{+}_{T}(t)\}. By Lemma 20, H0H_{0} has a nonempty subgraph HH with minimum degree at least (N1)(N3)8(N1)=18(N3)unvd(F)\frac{(N-1)(N-3)}{8(N-1)}=\frac{1}{8}(N-3)\geqslant\operatorname{unvd}^{*}(F). Let A=V(H)NT(v),B=V(H)NT+(t)A=V(H)\cap N^{-}_{T}(v),B=V(H)\cap N^{+}_{T}(t). Since HH is bipartite, note that in particular |A|,|B|unvd(F)|A|,|B|\geqslant\operatorname{unvd}^{*}(F).

We will now inductively build and embedding φ:FTV(H)\varphi\colon F\to T\langle V(H)\rangle such that

  • for every yND(s)y\in N^{-}_{D}(s), φ(y)A\varphi(y)\in A, and

  • for every yND+(s)y\in N^{+}_{D}(s), φ(y)B\varphi(y)\in B.

Extending such an embedding of FF by setting φ(s)=t\varphi(s)=t yields the desired embedding of DD in TT.

If FF is empty, then the result is clear. Suppose now that FF is not empty and consider the tree FF^{\prime} obtained from FF by contracting every connected component CC of DND(s)D\langle N^{-}_{D}(s)\rangle and DND+(s)D\langle N^{+}_{D}(s)\rangle into a single vertex uV(C)u_{V(C)}. Consider a leaf uZu_{Z} of FF^{\prime}, and without loss of generality, assume that ZND(s)Z\subseteq N^{-}_{D}(s). If Z=V(F)Z=V(F), then V(F)ND(s)V(F)\subseteq N^{-}_{D}(s), and we can embed FF in TAT\langle A\rangle since |A|unvd(F)unvd(F)|A|\geqslant\operatorname{unvd}^{*}(F)\geqslant\operatorname{unvd}(F). Otherwise, assume by induction that we have an embedding φ:FZTV(H)\varphi\colon F-Z\to T\langle V(H)\rangle with φ(ND(s)Z)A\varphi(N^{-}_{D}(s)\setminus Z)\subseteq A and φ(ND+(s))B\varphi(N^{+}_{D}(s))\subseteq B. Let uvuv be the arc joining ZZ to FZF-Z in FF^{\prime}. Observe that we have uZu\in Z and vND+(s)v\in N^{+}_{D}(s), for otherwise uvsuvs is a directed triangle in DD, a contradiction to DD being acyclic.

As HH has minimum degree at least unvd(F)\operatorname{unvd}^{*}(F), and because φ(v)B\varphi(v)\in B, the size of ANT(φ(v))A\cap N^{-}_{T}(\varphi(v)) is at least unvd(F)\operatorname{unvd}^{*}(F), hence there are at least unvd(F)|V(F)Z|unvd(FZ)\operatorname{unvd}^{*}(F)-|V(F)\setminus Z|\geqslant\operatorname{unvd}(F\langle Z\rangle) vertices in ANT(φ(v))φ(V(FZ))A\cap N^{-}_{T}(\varphi(v))\setminus\varphi(V(F-Z)). Thus we can extend φ\varphi to the vertices in FZF\langle Z\rangle in ANT(φ(v))φ(V(FZ))A\cap N^{-}_{T}(\varphi(v))\setminus\varphi(V(F-Z)), and this gives the desired embedding of DD. See Figure 1 for an illustration. ∎

ttAABBφ(v)\varphi(v)\Rightarrow\Rightarrow
Figure 1: Illustration of the proof of Theorem 21. The red vertices are the image of V(FZ)V(F-Z) under φ\varphi. The green vertices represent an embedding of FZF\langle Z\rangle in the set ANT(φ(v))φ(V(FZ))A\cap N^{-}_{T}(\varphi(v))\setminus\varphi(V(F-Z)), which is dashed in the figure.

We now move to the general case. We first prove an analogue of Lemma 19 for larger values of kk.

Lemma 23.

Let kk be a positive integer and let TT be a tournament on N3(k+12)N\geqslant 3^{\binom{k+1}{2}} vertices. Then the number of distinct copies of TTkTT_{k} in TT is at least 3(k+12)Nk3^{-\binom{k+1}{2}}N^{k}.

Proof.

We proceed by induction on kk. For k=1k=1 the result is clear. Now suppose k>1k>1 and that the result holds for smaller values of kk. By Proposition 13, in TT there are at least N3\frac{N}{3} vertices of out-degree at least N3\frac{N}{3}. For every such vertex uu, there are by induction at least 3(k2)(N3)k13^{-\binom{k}{2}}\left(\frac{N}{3}\right)^{k-1} copies of TTk1TT_{k-1} in the subtournament induced by N+(u)N^{+}(u). Together with uu, this gives as many copies of TTkTT_{k} in TT. We conclude that the number of copies of TTkTT_{k} in TT is at least N3×3(k2)(N3)k1=3(k+12)Nk\frac{N}{3}\times 3^{-\binom{k}{2}}\left(\frac{N}{3}\right)^{k-1}=3^{-\binom{k+1}{2}}N^{k}. ∎

This bound could be improved to (2(k2)+o(1))Nk(2^{-\binom{k}{2}}+o(1))N^{k}, see [4].

Theorem 24.

Let k,nk,n be integers with k1,n2k\geqslant 1,n\geqslant 2 and let FF be an oriented tree of order nn. Let DD be a kk-extension of FF, then unvd(D)3(2k+22)unvd(F)\operatorname{unvd}(D)\leqslant 3^{\binom{2k+2}{2}}\cdot\operatorname{unvd}^{*}(F).

Again, applying Theorem 5 to bound unvd(F)\operatorname{unvd}^{*}(F), we deduce the following corollary.

Corollary 25.

There is a constant n0n_{0} such that the following holds. Let DD be a kk-extension of an oriented tree FF on at least n0n_{0} vertices, then unvd(D)23(2k+22)|V(F)|\operatorname{unvd}(D)\leqslant 2\cdot 3^{\binom{2k+2}{2}}\cdot|V(F)| .

Proof of Theorem 24.

Let TT be a tournament on N=3(2k+22)unvd(F)N=3^{\binom{2k+2}{2}}\operatorname{unvd}^{*}(F) vertices.

Let HH be a copy of TT2k+1TT_{2k+1} with acyclic ordering (u0,u1,u1,,uk,uk)(u^{\prime}_{0},u_{1},u^{\prime}_{1},\dots,u_{k},u^{\prime}_{k}). By Lemma 23, TT contains at least 32k(2k+22)unvd(F)2k+13^{2k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{2k+1} copies of HH. By the Pigeon-hole Principle, there are kk vertices v1,,vkv_{1},\dots,v_{k} in TT such that there are at least 1Nk32k(2k+22)unvd(F)2k+1=3k(2k+22)unvd(F)k+1\frac{1}{N^{k}}3^{2k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{2k+1}=3^{k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{k+1} embeddings of HH in TT with uiu_{i} mapped to viv_{i} for every i{1,,k}i\in\{1,\dots,k\}. From now on, let us say that an embedding φ:HT\varphi\colon H\to T is valid if φ(ui)=vi\varphi(u_{i})=v_{i} for every i{1,,k}i\in\{1,\dots,k\}.

We consider the hypergraph \mathcal{H} with vertex set V(T){v1,,vk}V(T)\setminus\{v_{1},\dots,v_{k}\} which contains as a hyperedge {φ(ui)0ik}\{\varphi(u_{i}^{\prime})\mid 0\leqslant i\leqslant k\} for every valid embedding φ:HT\varphi\colon H\to T. By Lemma 20 applied to \mathcal{H}, there is a set of vertices XV()X\subseteq V(\mathcal{H}) such that every vertex in XX has degree at least 1N3k(2k+22)unvd(F)k+1\frac{1}{N}3^{k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{k+1} in \mathcal{H}. Let (X0,,Xk)(X_{0},\dots,X_{k}) be the partition of XX where XiX_{i} contains all the vertices xXx\in X such that x=φ(ui)x=\varphi(u_{i}^{\prime}) for some valid embedding φ:HT\varphi\colon H\to T.

For all i,j{0,,k}i,j\in\{0,\dots,k\} with i<ji<j and for every vertex xXix\in X_{i}, we claim that xx has at least 1Nk3k(2k+22)unvd(F)k+1=unvd(F)\frac{1}{N^{k}}3^{k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{k+1}=\operatorname{unvd}^{*}(F) out-neighbours in XjX_{j}. Assume that this is not the case, then the degree of xx in X\mathcal{H}\langle X\rangle, which is exactly the number of valid embedding φ:HT\varphi\colon H\to T with the extra conditions φ(ui)=x\varphi(u_{i}^{\prime})=x and φ(u)X\varphi(u_{\ell}^{\prime})\in X for every {0,,k}\ell\in\{0,\dots,k\}, is at most

|N+(x)Xj|{0,,k}{i,j}|X|<Nk11Nk3k(2k+22)unvd(F)k+1,|N^{+}(x)\cap X_{j}|\cdot\prod_{\ell\in\{0,\dots,k\}\setminus\{i,j\}}|X_{\ell}|<N^{k-1}\frac{1}{N^{k}}3^{k\binom{2k+2}{2}}\operatorname{unvd}^{*}(F)^{k+1},

a contradiction to the choice of XX. Similarly, if j<ij<i then xx has at least unvd(F)\operatorname{unvd}^{*}(F) in-neighbours in XjX_{j}. Since k1k\geqslant 1, this implies in particular that |Xi|unvd(F)|X_{i}|\geqslant\operatorname{unvd}^{*}(F) for every i{0,,k}i\in\{0,\dots,k\}.

Let DD be a full kk-extension of FF. As DD is acyclic, let σ\sigma be a topological ordering of V(D)V(D). Let WV(D)W\subseteq V(D) be such that F=DWF=D-W and w1,,wkw_{1},\dots,w_{k} the ordering of WW with respect to σ\sigma. We let Y0,,YkY_{0},\dots,Y_{k} be the partition of V(F)V(F) where YiY_{i} contains every vertex between wiw_{i} and wi+1w_{i+1} in σ\sigma, for every i{1,,k1}i\in\{1,\dots,k-1\}, Y0Y_{0} contains every vertex before w1w_{1} in σ\sigma, and YkY_{k} contains every vertex after wkw_{k} in σ\sigma. By construction, every arc between YiY_{i} and YjY_{j} in FF is from YiY_{i} to YjY_{j} if i<ji<j. We will now construct by induction on |V(F)||V(F)| an embedding φ:FTX\varphi\colon F\to T\langle X\rangle such that φ(Yi)Xi\varphi(Y_{i})\subseteq X_{i} for every i{0,,k}i\in\{0,\dots,k\}. By setting φ(wi)=vi\varphi(w_{i})=v_{i} for every i[k]i\in[k], we will get the desired embedding of DD in TT.

If FF is empty, then the result is clear. Now assume |V(F)|>0|V(F)|>0. Consider the tree FF^{\prime} obtained from FF by contracting every connected component CC of DYiD\langle Y_{i}\rangle into a single vertex xV(C)x_{V(C)}, for every i{0,,k}i\in\{0,\dots,k\}. Let xZx_{Z} be a leaf of FF^{\prime} and i{0,,k}i\in\{0,\dots,k\} be the index such that ZYiZ\subseteq Y_{i}. If Z=V(F)Z=V(F), then, as |Xi|unvd(F)unvd(F)|X_{i}|\geqslant\operatorname{unvd}^{*}(F)\geqslant\operatorname{unvd}(F), there is an embedding φ\varphi of FF into TXiT\langle X_{i}\rangle and we are done. Otherwise let pp be the unique neighbour in FF of a vertex in ZZ which does not belong to ZZ. Assume that pp is the tail of the (unique) arc between FZF-Z and ZZ, the other case being symmetric. Let j{0,,k}j\in\{0,\dots,k\} be the index such that pYjp\in Y_{j}. Recall that, by construction, we have j<ij<i. By induction hypothesis, there is an embedding φ:FZTX0Xk\varphi\colon F-Z\to T\langle X_{0}\cup\dots\cup X_{k}\rangle such that φ(YZ)X\varphi(Y_{\ell}\setminus Z)\subseteq X_{\ell} for every {0,,k}\ell\in\{0,\dots,k\}. We know that φ(p)\varphi(p) has at least unvd(F)\operatorname{unvd}^{*}(F) in-neighbours in XiX_{i}. Moreover unvd(F)|V(F)Z|unvd(FZ)\operatorname{unvd}^{*}(F)-|V(F)\setminus Z|\geqslant\operatorname{unvd}(F\langle Z\rangle) and so |XiN(φ(p))φ(V(FZ))|unvd(FZ)|X_{i}\cap N^{-}(\varphi(p))\setminus\varphi(V(F-Z))|\geqslant\operatorname{unvd}(F\langle Z\rangle) (by definition of unvd(F)\operatorname{unvd}^{*}(F)). It follows that FZF\langle Z\rangle can be embedded in XiN(φ(p))φ(V(FZ))X_{i}\cap N^{-}(\varphi(p))\setminus\varphi(V(F-Z)). This gives the desired embedding of FF and concludes the proof of the theorem. ∎

5 Particular extensions of some particular forests

5.1 Twin extensions of arcless digraphs

For every positive integers k,n1,n2k,n_{1},n_{2}, let Dk(n1,n2)D_{k}(n_{1},n_{2}) be full kk-twin-extension of the arcless digraph DD of order n1+n2n_{1}+n_{2} in which the kk added vertices have the same in-neighbourhood V1V_{1} of size n1n_{1} and the same out-neighbourhood V2V_{2} of size n2n_{2} (with V1,V2V_{1},V_{2} a partition of V(D)V(D)). Note that D1(n1,n2)D_{1}(n_{1},n_{2}) is simply the oriented star made of a vertex dominated by n1n_{1} vertices and dominating n2n_{2} vertices.

Proposition 26.

For every positive integers n,n1,n2n,n_{1},n_{2} with n=n1+n2+1n=n_{1}+n_{2}+1, each of the following holds

  1. 1.

    unvd(D1(0,n1))=unvd(D1(n1,0))=2n2\operatorname{unvd}(D_{1}(0,n-1))=\operatorname{unvd}(D_{1}(n-1,0))=2n-2, and

  2. 2.

    unvd(D1(n1,n2))=2n3\operatorname{unvd}(D_{1}(n_{1},n_{2}))=2n-3.

Proof.

The fact that unvd(D1(0,n1))=unvd(D1(n1,0))2n2\operatorname{unvd}(D_{1}(0,n-1))=\operatorname{unvd}(D_{1}(n-1,0))\leqslant 2n-2 and unvd(D1(n1,n2))2n3\operatorname{unvd}(D_{1}(n_{1},n_{2}))\leqslant 2n-3 follows from Lemma 13. For the lower bounds, consider first any (n2)(n-2)-regular tournament TT. Then TT has 2n32n-3 vertices and does not contain D1(0,n1)D_{1}(0,n-1). Thus unvd(D1(0,n1))=unvd(D1(n1,0))2n2\operatorname{unvd}(D_{1}(0,n-1))=\operatorname{unvd}(D_{1}(n-1,0))\geqslant 2n-2. Similarly, if T1T_{1} is an (n11)(n_{1}-1)-regular tournament and T2T_{2} an (n21)(n_{2}-1)-regular tournament, then T1T2T_{1}\Rightarrow T_{2} is a tournament of order 2n42n-4 which does not contain D1(n1,n2)D_{1}(n_{1},n_{2}), implying that unvd(D1(n1,n2))2n3\operatorname{unvd}(D_{1}(n_{1},n_{2}))\geqslant 2n-3. ∎

To prove a lower bound on unvd(Dk(0,n))\operatorname{unvd}(D_{k}(0,n)), we will use a probabilistic argument relying on Chernoff’s bound.

Proposition 27 (Chernoff’s Bound).

If XX is a random variable following a binomial law with parameters p[0,1]p\in[0,1] and n0n\geqslant 0, then for every ε[0,1]\varepsilon\in[0,1]

𝐏𝐫[X(1+ε)pn]exp(ε23pn).\mathbf{Pr}[X\geqslant(1+\varepsilon)pn]\leqslant\exp\left(-\frac{\varepsilon^{2}}{3}pn\right).
Proposition 28.

For every fixed k1k\geqslant 1, unvd(Dk(0,n))=2kno(n)\operatorname{unvd}(D_{k}(0,n))=2^{k}n-o(n) as n+n\to+\infty.

Proof.

The added vertices of Dk(0,n)D_{k}(0,n) are sources, so by Proposition 1, unvd(Dk(0,n))2kn\operatorname{unvd}(D_{k}(0,n))\leqslant 2^{k}n.

To prove the lower bound, let 0<ε<10<\varepsilon<1 and let TT be a tournament of order 2k1+εn+k\left\lfloor\frac{2^{k}}{1+\varepsilon}n+k\right\rfloor taken uniformly at random. For every XV(T)X\subseteq V(T) of size kk, let dX+d^{+}_{X} be the number of common out-neighbours of the vertices in XX, that is dX+=|xXN+(x)|d^{+}_{X}=|\bigcap_{x\in X}N^{+}(x)|. Then dX+d^{+}_{X} follows a binomial law with parameters 2k2^{-k} and 2k1+εn2k1+εn\left\lfloor\frac{2^{k}}{1+\varepsilon}n\right\rfloor\leqslant\frac{2^{k}}{1+\varepsilon}n. By Chernoff’s bound (Proposition 27),

𝐏𝐫[dX+n]𝐏𝐫[dX+(1+ε)𝐄[dX+]]exp(ε23𝐄[dX+])exp(ε232k2k1+εn)exp(ε23(1+ε)n+1).\begin{split}\mathbf{Pr}[d^{+}_{X}\geqslant n]&\leqslant\mathbf{Pr}\big{[}d^{+}_{X}\geqslant(1+\varepsilon)\mathbf{E}[d^{+}_{X}]\big{]}\\ &\leqslant\exp\left(-\frac{\varepsilon^{2}}{3}\mathbf{E}[d^{+}_{X}]\right)\\ &\leqslant\exp\left(-\frac{\varepsilon^{2}}{3}2^{-k}\left\lfloor\frac{2^{k}}{1+\varepsilon}n\right\rfloor\right)\\ &\leqslant\exp\left(-\frac{\varepsilon^{2}}{3(1+\varepsilon)}n+1\right).\\ \end{split}

It follows by the Union Bound that 𝐏𝐫[T contains Dk(0,n)](2k1+εnk)eε23(1+ε)n+1<1\displaystyle\mathbf{Pr}[T\text{ contains }D_{k}(0,n)]\leqslant\binom{\lfloor\frac{2^{k}}{1+\varepsilon}n\rfloor}{k}\mathrm{e}^{-\frac{\varepsilon^{2}}{3(1+\varepsilon)}n+1}<1 for nn large enough. This proves that unvd(Dk(0,n))>2k1+εn+k\operatorname{unvd}(D_{k}(0,n))>\left\lfloor\frac{2^{k}}{1+\varepsilon}n+k\right\rfloor for nn large enough. ∎

We will now prove a similar upper bound for Dk(n1,n2)D_{k}(n_{1},n_{2}). To do so, we will use Ramsey’s theorem, that we first recall here.

Theorem 29 (Ramsey’s theorem [20]).

For every positive integers kk, aa and bb, there exists Rk(a,b)R_{k}(a,b) such that for every 22-colouring of the kk-subsets of [Rk(a,b)][R_{k}(a,b)], either there is a set X[Rk(a,b)]X\subseteq[R_{k}(a,b)] of size aa such that every kk-subset of XX is coloured 11, or there is a set Y[Rk(a,b)]Y\subseteq[R_{k}(a,b)] of size bb such that every kk-subset of YY is coloured 22.

Theorem 30.

Let ε>0\varepsilon>0 and let kk be a positive integer. There is a constant CC depending only on kk and ε\varepsilon such that unvd(Dk(n1,n2))(2k+ε)(n1+n2)+C\operatorname{unvd}(D_{k}(n_{1},n_{2}))\leqslant(2^{k}+\varepsilon)(n_{1}+n_{2})+C.

Proof.

For fixed constants k,εk,\varepsilon, simple computations show the existence of constants C0,η>0C_{0}\in\mathbb{N},\eta>0 such that (Ck)(C(12η)k)2k+ε\frac{\binom{C}{k}}{\binom{C(\frac{1}{2}-\eta)}{k}}\leqslant 2^{k}+\varepsilon for every integer CC0C\geqslant C_{0}. We let CC be the maximum between C0C_{0} and Rk(k12η,k12η)R_{k}\left(\left\lceil\frac{k}{\frac{1}{2}-\eta}\right\rceil,\left\lceil\frac{k}{\frac{1}{2}-\eta}\right\rceil\right).

We first assume that n1,n2Cη(2k+ε)n_{1},n_{2}\geqslant\frac{C}{\eta(2^{k}+\varepsilon)}. Let TT be a tournament on (2k+ε)n1+(2k+ε)n2+C\lceil(2^{k}+\varepsilon)n_{1}\rceil+\lceil(2^{k}+\varepsilon)n_{2}\rceil+C vertices. Consider a median order of TT and let B1B_{1} be the first (2k+ε)n1\lceil(2^{k}+\varepsilon)n_{1}\rceil vertices in this order, B2B_{2} be the last (2k+ε)n2\lceil(2^{k}+\varepsilon)n_{2}\rceil vertices, and AA be set of the remaining CC vertices in the middle.

By (M2), every vertex vAv\in A is dominated by at least 12|B1||A|(12η)|B1|\frac{1}{2}|B_{1}|-|A|\geqslant(\frac{1}{2}-\eta)|B_{1}| vertices in B1B_{1} (using |B1|=(2k+ε)n1Cη=|A|η|B_{1}|=\lceil(2^{k}+\varepsilon)n_{1}\rceil\geqslant\frac{C}{\eta}=\frac{|A|}{\eta}). Let AAA^{\prime}\subseteq A be any set of size k12η\left\lceil\frac{k}{\frac{1}{2}-\eta}\right\rceil. By Lemma 17, there is a subset AA^{\star} of AA^{\prime} of size kk such that AA^{\star} has at least n1n_{1} common in-neighbours in B1B_{1}. Since this holds for every subset AA^{\prime} of size k12η\left\lceil\frac{k}{\frac{1}{2}-\eta}\right\rceil, and by choice of CC, by Ramsey’s Theorem there is a subset A~\tilde{A} of AA of size k12η\left\lceil\frac{k}{\frac{1}{2}-\eta}\right\rceil such that every subset of kk vertices in A~\tilde{A} have at least kk common in-neighbours in B1B_{1}. By repeating the same reasoning, because every vertex in AA dominates at least 12|B2||A|(12η)|B2|\frac{1}{2}|B_{2}|-|A|\geqslant(\frac{1}{2}-\eta)|B_{2}| vertices in B2B_{2}, we deduce that A~\tilde{A} contains a set XX of kk vertices with at least n2n_{2} common out-neighbours in B2B_{2}. Since every kk-subset of A~\tilde{A} has at least n1n_{1} common in-neighbours in B1B_{1}, this gives the desired copy of Dk(n1,n2)D_{k}(n_{1},n_{2}).

We now consider the case n1<Cη(2k+ε)n_{1}<\frac{C}{\eta(2^{k}+\varepsilon)} or n2<Cη(2k+ε)n_{2}<\frac{C}{\eta(2^{k}+\varepsilon)}. We apply the previous case for n1,n2n_{1},n_{2} replaced respectively by n1+Cη(2k+ε)n_{1}+\left\lceil\frac{C}{\eta(2^{k}+\varepsilon)}\right\rceil and n2+Cη(2k+ε)n_{2}+\left\lceil\frac{C}{\eta(2^{k}+\varepsilon)}\right\rceil. This shows that unvd(Dk(n1,n2))(2k+ε)(n1+n2+2(Cη(2k+ε)+1))+C(2k+ε)(n1+n2)+C\operatorname{unvd}(D_{k}(n_{1},n_{2}))\leqslant(2^{k}+\varepsilon)\left(n_{1}+n_{2}+2\left(\frac{C}{\eta(2^{k}+\varepsilon)}+1\right)\right)+C\leqslant(2^{k}+\varepsilon)(n_{1}+n_{2})+C^{\prime} for C=C+2(2k+ε)(Cη(2k+ε)+1)C^{\prime}=C+2(2^{k}+\varepsilon)\left(\frac{C}{\eta(2^{k}+\varepsilon)}+1\right). ∎

While Proposition 28 shows that this upper is tight when n1n_{1} or n2n_{2} is in o(n)o(n), we do not know the precise asymptotic of unvd(Dk(n1,n2))\operatorname{unvd}(D_{k}(n_{1},n_{2})) when n1n2n_{1}\approx n_{2}.

5.2 11-extensions of directed paths

In this section, we give the exact value of unvd(D)\operatorname{unvd}(D) when DD is a full 11-extension of a directed path.

Theorem 31.

Let DD be a full 11-extension of a directed path with added vertex ss. If ss is a source or a sink, or if DD is TT3TT_{3}, then unvd(D)=2|V(D)|2\operatorname{unvd}(D)=2|V(D)|-2. Otherwise, unvd(D)=2|V(D)|3\operatorname{unvd}(D)=2|V(D)|-3.

Proof.

Observe that every orientation of K4K_{4} contains TT3TT_{3}, and that the directed triangle does not contain TT3TT_{3}, so unvd(TT3)=4\operatorname{unvd}(TT_{3})=4. We now assume that DD is not TT3TT_{3}.

Set n=|V(D)|n=|V(D)|, and let ss be a vertex of DD such that DsD-s is a directed path PP. Let V1=ND(s)V_{1}=N^{-}_{D}(s) and V2=ND+(s)V_{2}=N^{+}_{D}(s) and set n1=|V1|n_{1}=|V_{1}| and n2=|V2|n_{2}=|V_{2}|. Observe that D=D1(n1,n2)D=D_{1}(n_{1},n_{2}). By Proposition 26 we obtain unvd(D)2n3\operatorname{unvd}(D)\geqslant 2n-3, and unvd(D)2n2\operatorname{unvd}(D)\geqslant 2n-2 if ss is a sink or a source.

If n1=0n_{1}=0 or n2=0n_{2}=0, then ss is a source or a sink. Thus, by Proposition 1, we have unvd(D)2unvd(P)=2n2\operatorname{unvd}(D)\leqslant 2\operatorname{unvd}(P)=2n-2.

It remains to show that unvd(D)2n3\operatorname{unvd}(D)\leqslant 2n-3, assuming n11n_{1}\geqslant 1 and n21n_{2}\geqslant 1, and max{n1,n2}2\max\{n_{1},n_{2}\}\geqslant 2. By directional duality, assume without loss of generality that n12n_{1}\geqslant 2. Let TT be a tournament of order 2n32n-3, and let σ0=(v2n1+1,,v0,,v2n21)\sigma_{0}=(v_{-2n_{1}+1},\dots,v_{0},\dots,v_{2n_{2}-1}) be a median order of TT. Set A1={v2n1+1,,v1}A_{1}=\{v_{-2n_{1}+1},\dots,v_{-1}\} and A2={v0,,v2n21}A_{2}=\{v_{0},\dots,v_{2n_{2}-1}\}. Let B1=N(v1)A1B_{1}=N^{-}(v_{-1})\cap A_{1} and B2=N+(v1)A2B_{2}=N^{+}(v_{-1})\cap A_{2}. Observe that, by (M2), |B2|n2|B_{2}|\geqslant n_{2} and |B1|n11|B_{1}|\geqslant n_{1}-1. We first consider the case |B1|n1|B_{1}|\geqslant n_{1}.

Claim 3.

If |B1|n1|B_{1}|\geqslant n_{1}, then TT contains a copy of DD with s=v1s=v_{-1}.

Proof of claim.  Let II^{-} be the set of in-generators of TB1T\langle B_{1}\rangle and O+O^{+} be the set of vertices which are initial vertices of a directed path of order n2n_{2} in TB2T\langle B_{2}\rangle. Note that II^{-}\neq\emptyset and O+O^{+}\neq\emptyset. Moreover, by Lemma 15, (B1I)I(B_{1}\setminus I^{-})\Rightarrow I^{-} and O+(B2O+)O^{+}\Rightarrow(B_{2}\setminus O^{+}). Let viv_{i^{-}} be the vertex in II^{-} with largest index and vi+v_{i^{+}} be the vertex in O+O^{+} with smallest index. By Lemma 14, viv_{i^{-}} is the terminus of a directed path PP^{-} of order n1n_{1} in TB1T\langle B_{1}\rangle and vi+v_{i^{+}} is the initial vertex of a directed path P+P^{+} of order n2n_{2} in TB2T\langle B_{2}\rangle. Hence we may assume vi+viv_{i^{+}}\rightarrow v_{i^{-}}, for otherwise the union of v1v_{-1}, PP^{-}, P+P^{+}, the arcs between those paths and v1v_{-1}, and vivi+v_{i^{-}}v_{i^{+}} is a copy of DD in TT.

Assume first that |B2|=n2|B_{2}|=n_{2}. Then, since |A2|=2n2|A_{2}|=2n_{2}, v1v_{-1} has exactly n2n_{2} out-neighbour and n2n_{2} in-neighbours in A2A_{2}. Consequently σ=(v2n1+1,v2,v0,,v2n21,v1)\sigma^{\prime}=(v_{-2n_{1}+1},\dots v_{-2},v_{0},\dots,v_{2n_{2}-1},v_{-1}) is also a median order of TT. By (M4) for σ\sigma^{\prime}, since vi+viv_{i^{+}}\rightarrow v_{i^{-}}, there is a vertex vv_{\ell} in {vi,,vi+}{v1}\{v_{i^{-}},\dots,v_{i^{+}}\}\setminus\{v_{-1}\} such that vivvi+v_{i^{-}}\rightarrow v_{\ell}\rightarrow v_{i^{+}}. Observe that vIv_{\ell}\notin I^{-} by maximality of ii^{-} and v(B1I)v_{\ell}\notin(B_{1}\setminus I^{-}) because (B1I){vi}(B_{1}\setminus I^{-})\Rightarrow\{v_{i^{-}}\}. Thus vB1v_{\ell}\notin B_{1}, and similarly vB2v_{\ell}\notin B_{2}. In particular, v(V(P)V(P+))v_{\ell}\notin(V(P^{-})\cup V(P^{+})). If v1vv_{-1}\rightarrow v_{\ell}, remove the end-vertex from P+P^{+}, otherwise remove the initial vertex of PP^{-}. In both cases, the resulting union of v1v_{-1}, PP^{-}, P+P^{+} and the arcs between those paths and v1v_{-1} is a copy of DD in TT (with s=v1s=v_{-1}) as desired.

Henceforth, we may assume that |B2|>n2|B_{2}|>n_{2}. In particular it implies |O+|2|O^{+}|\geqslant 2: to see this, take any Hamiltonian directed path of TB2T\langle B_{2}\rangle (which exists by Theorem 3), then the two first vertices of this directed path are initial vertices of directed paths of order n2n_{2} in TB2T\langle B_{2}\rangle. If there is a vertex vv_{\ell} in {vi,,vi+}{v1}\{v_{i^{-}},\dots,v_{i^{+}}\}\setminus\{v_{-1}\} such that vivvi+v_{i^{-}}\rightarrow v_{\ell}\rightarrow v_{i^{+}}, then as above there is a copy of DD in TT (with s=v1s=v_{-1}), so assume that there is no such vv_{\ell}. Hence, by (M4) and (M5), σ~=(v2n1+1,,vi1,vi+1,,vi+,vi,vi++1,,v2n21)\tilde{\sigma}=(v_{-2n_{1}+1},\dots,v_{i^{-}-1},v_{i^{-}+1},\dots,v_{i^{+}},v_{i^{-}},v_{i^{+}+1},\ldots,v_{2n_{2}-1}) is a median order of TT. Let vj+v_{j^{+}} be the vertex of smallest index in O+{vi+}O^{+}\setminus\{v_{i^{+}}\}. Let Q+Q^{+} be a directed path of order n2n_{2} in TB2T\langle B_{2}\rangle with initial vertex vj+v_{j^{+}}. If vivj+v_{i^{-}}\rightarrow v_{j^{+}}, then the union of v1v_{-1}, PP^{-}, Q+Q^{+}, the arcs between those paths and v1v_{-1}, and vivj+v_{i^{-}}v_{j^{+}} is a copy of DD in TT. Henceforth we may assume vj+viv_{j^{+}}\rightarrow v_{i^{-}}. By (M4) for σ~\tilde{\sigma}, there is a vertex vpv_{p} with i++1pj+1i^{+}+1\leqslant p\leqslant j^{+}-1 such that vivpvj+v_{i^{-}}\rightarrow v_{p}\rightarrow v_{j^{+}}. Note that vpv1v_{p}\neq v_{-1} because i+0i^{+}\geqslant 0, and as above vpB1B2v_{p}\notin B_{1}\cup B_{2}. As above, by removing either the end-vertex of Q+Q^{+} or the initial vertex of PP^{-}, we get that the union of v1v_{-1}, PP^{-}, Q+Q^{+} and the arcs between those paths and v1v_{-1} is a copy of DD in TT. \lozenge

Since Claim 3 settles the case |B1|n1|B_{1}|\geqslant n_{1}, and because |B1|n11|B_{1}|\geqslant n_{1}-1, we now assume that |B1|=n11|B_{1}|=n_{1}-1. Then by (M4) and (M5) applied on σ0\sigma_{0}, we have that σ1=(v1,v2n1+1,,v2,v0,,v2n21)\sigma_{1}=(v_{-1},v_{-2n_{1}+1},\dots,v_{-2},v_{0},\dots,v_{2n_{2}-1}) is a median order of TT. In particular this implies v2v0v_{-2}\rightarrow v_{0} and v1v2n1+1v_{-1}\rightarrow v_{-2n_{1}+1}. Observe that v1,v2n1+1,,v1v_{-1},v_{-2n_{1}+1},\dots,v_{-1} is then a Hamiltonian directed cycle of TA1T\langle A_{1}\rangle. Not also that if |N(v2)A1|n1|N^{-}(v_{2})\cap A_{1}|\geqslant n_{1} we can apply Claim 3 with σ1,v2\sigma_{1},v_{-2} playing the roles of σ0,v1\sigma_{0},v_{-1} respectively. By repeating this process, either we eventually find a copy of DD in TT with added vertex sA1s\in A_{1} or we ensure that A1{v0}A_{1}\Rightarrow\{v_{0}\}. Henceforth we assume that A1{v0}A_{1}\Rightarrow\{v_{0}\}.

The end of the proof is similar to the proof of Claim 3, with v0v_{0} playing the role of v1v_{-1}. Let B1=A1B_{1}^{\prime}=A_{1} and B2=N+(v0)A2B_{2}^{\prime}=N^{+}(v_{0})\cap A_{2}. We have |B1|=2n11|B_{1}^{\prime}|=2n_{1}-1 and, by (M2) on σ0\sigma_{0}, |B2|n2|B_{2}^{\prime}|\geqslant n_{2}. Since v1,v2n1+1,,v1v_{-1},v_{-2n_{1}+1},\dots,v_{-1} is a directed cycle, in particular both v1v_{-1} and v2v_{-2} are the terminal vertices of a directed path of order n1n_{1} in B1B_{1}^{\prime}. Let O+{O^{+}}^{\prime} be the set of vertices which are initial vertex of a directed path of order n2n_{2} in TB2T\langle B_{2}^{\prime}\rangle. Note that O+{O^{+}}^{\prime}\neq\emptyset because |B2|n2|B_{2}^{\prime}|\geqslant n_{2} and, by Lemma 15, O+(B2O+){O^{+}}^{\prime}\Rightarrow(B_{2}\setminus{O^{+}}^{\prime}). Let vk+v_{k^{+}} be the vertex in O+{O^{+}}^{\prime} with smallest index. Let RR^{-} be the directed path vn1,,v1v_{-n_{1}},\dots,v_{-1} and R+R^{+} be a directed path of order n2n_{2} in TB2T\langle B_{2}\rangle starting on vk+v_{k^{+}}.

We may assume that vk+v1v_{k^{+}}\rightarrow v_{-1}, for otherwise v0v_{0}, RR^{-}, R+R^{+}, the arcs between those paths and v0v_{0}, and v1vk+v_{-1}v_{k^{+}} is a copy of DD in TT. If there is a vertex vv_{\ell} in {v1,,vk+1}\{v_{1},\dots,v_{k^{+}-1}\} such that v1vvk+v_{-1}\rightarrow v_{\ell}\rightarrow v_{k^{+}}, then by removing either the end-vertex of R+R^{+} or the initial vertex of RR^{-}, we get that the union of v0v_{0}, RR^{-}, R+R^{+} and the arcs between those paths and v0v_{0} is a copy of DD in TT. This holds because vB2v_{\ell}\notin B_{2}^{\prime} by choice of vk+v_{k^{+}} and R+R^{+} is a directed path in B2B_{2}^{\prime}. Henceforth, we assume that there is no such vv_{\ell}. Therefore, by (M4) and (M5) applied on σ0\sigma_{0}, σ^=(v2n1+1,,v2,vk+,v1,,vk+1,vk++1,,v2n21)\hat{\sigma}=(v_{-2n_{1}+1},\dots,v_{-2},v_{k^{+}},v_{-1},\dots,v_{k^{+}-1},v_{k^{+}+1},\ldots,v_{2n_{2}-1}) is a median order of TT. In particular, v2vk+v_{-2}\rightarrow v_{k^{+}}. Let R{R^{-}}^{\prime} be vn11,,v2v_{-n_{1}-1},\dots,v_{-2}. We thus obtain that the union of v0v_{0}, R{R^{-}}^{\prime}, R+R^{+} and the arcs between those paths and v0v_{0} is a copy of DD in TT. ∎

References

  • [1] B. Alspach and M. Rosenfeld. Realization of certain generalized paths in tournaments. Discrete Mathematics, 34(2):199–202, 1981.
  • [2] J. Bang-Jensen and G. Gutin, editors. Classes of directed graphs. Springer monographs in mathematics. Springer International Publishing, Cham, Switzerland, 1st edition, July 2018.
  • [3] M. Bucić, S. Letzter, and B. Sudakov. Directed ramsey number for trees. Journal of Combinatorial Theory, Series B, 137:145–177, 2019. arXiv:1708.04504.
  • [4] L. N. Coregliano and A. A. Razborov. On the density of transitive tournaments. Journal of Graph Theory, 85(1):12--21, 2017. arXiv:1501.04074.
  • [5] N. Draganić, F. Dross, J. Fox, A. Girão, F. Havet, D. Korándi, W. Lochet, D. M. Correia, A. Scott, and B. Sudakov. Powers of paths in tournaments. Combinatorics, Probability and Computing, 30(6):894--898, 2021. arXiv:2010.05735.
  • [6] F. Dross and F. Havet. On the unavoidability of oriented trees. Journal of Combinatorial Theory, Series B, 151:83--110, 2021. arXiv:1812.05167.
  • [7] A. El Sahili. Trees in tournaments. Journal of Combinatorial Theory, Series B, 92(1):183--187, 2004.
  • [8] P. Erdős and L. Moser. On the representation of directed graphs as unions of orderings. Magyar Tud. Akad. Mat. Kutató Int. Közl., 9:125--132, 1964.
  • [9] R. Forcade. Parity of paths and circuits in tournaments. Discrete Mathematics, 6(2):115--118, 1973.
  • [10] J. Fox, X. He, and Y. Wigderson. Ramsey numbers of sparse digraphs. Israel Journal of Mathematics, 2024. arXiv:2105.02383.
  • [11] B. Grünbaum. Antidirected hamiltonian paths in tournaments. Journal of Combinatorial Theory, Series B, 11(3):249--257, 1971.
  • [12] R. Häggkvist and A. Thomason. Trees in tournaments. Combinatorica, 11(2):123–130, June 1991.
  • [13] F. Havet. Trees in tournaments. Discrete Mathematics, 243(1):121--134, 2002.
  • [14] F. Havet and S. Thomassé. Median orders of tournaments: a tool for the second neighborhood problem and Sumner’s conjecture. Journal of Graph Theory, 35(4):244--256, 2000.
  • [15] F. Havet and S. Thomassé. Oriented Hamiltonian paths in tournaments: a proof of Rosenfeld’s conjecture. Journal of Combinatorial Theory, Series B, 78(2):243--273, 2000.
  • [16] D. Kühn, R. Mycroft, and D. Osthus. An approximate version of Sumner’s universal tournament conjecture. Journal of Combinatorial Theory, Series B, 101(6):415--447, 2011. arXiv:1010.4429.
  • [17] T. Kóvari, V. T. Sós, and P. Turán. On a problem of K. Zarankiewicz. Colloquium Mathematicum, 3(1):50–57, 1954.
  • [18] C. Lee. Ramsey numbers of degenerate graphs. Annals of Mathematics, 185(3), 2017. arXiv:1505.04773.
  • [19] N. Linial, M. Saks, and V. T. Sós. Largest digraphs contained in alln-tournaments. Combinatorica, 3(1):101--104, 1983.
  • [20] F. P. Ramsey. On a problem of formal logic. In Classic Papers in Combinatorics, pages 1--24. Springer, 1987.
  • [21] L. Rédei. Ein kombinatorischer Satz. Acta Litteraria Szeged, 7:39--43, 1934.
  • [22] K. Reid and E. Parker. Disproof of a conjecture of Erdös and Moser on tournaments. Journal of Combinatorial Theory, 9(3):225 -- 238, 1970.
  • [23] K. B. Reid and N. C. Wormald. Embedding oriented nn-trees in tournaments. Studia Scientiarum Mathematicarum Hungarica, 18(2-4):377--387, 1983.
  • [24] M. Rosenfeld. Antidirected Hamiltonian paths in tournaments. Journal of Combinatorial Theory, Series B, 12(1):93–99, Feb. 1972.
  • [25] A. Sanchez-Flores. On tournaments free of large transitive subtournaments. Graphs and Combinatorics, 14(2):181--200, Jun 1998.
  • [26] H. J. Straight. The existence of certain type of semi-walks in tournaments. Congressus Numerantium, 29:901--908, 1980.
  • [27] A. Thomason. Paths and cycles in tournaments. Transactions of the American Mathematical Society, 296(1):167--180, 1986.