Forbidden subposet problems for traces of set families

In this paper we introduce a problem that bridges forbidden subposet and forbidden subconfiguration problems. The sets $F_1,F_2, \dots,F_{|P|}$ form a copy of a poset $P$, if there exists a bijection $i:P\rightarrow \{F_1,F_2, \dots,F_{|P|}\}$ such that for any $p,p'\in P$ the relation $p<_P p'$ implies $i(p)\subsetneq i(p')$. A family $\mathcal{F}$ of sets is \textit{$P$-free} if it does not contain any copy of $P$. The trace of a family $\mathcal{F}$ on a set $X$ is $\mathcal{F}|_X:=\{F\cap X: F\in \mathcal{F}\}$. We introduce the following notions: $\mathcal{F}\subseteq 2^{[n]}$ is $l$-trace $P$-free if for any $l$-subset $L\subseteq [n]$, the family $\mathcal{F}|_L$ is $P$-free and $\mathcal{F}$ is trace $P$-free if it is $l$-trace $P$-free for all $l\le n$. As the first instances of these problems we determine the maximum size of trace $B$-free families, where $B$ is the butterfly poset on four elements $a,b,c,d$ with $a,b<c,d$ and determine the asymptotics of the maximum size of $(n-i)$-trace $K_{r,s}$-free families for $i=1,2$. We also propose a generalization of the main conjecture of the area of forbidden subposet problems.


Introduction
In this paper we introduce a problem that bridges two areas of extremal finite set theory, namely forbidden subposet problems and traces of set families. We denote by [n] the set of the first n positive integers and for a set X we use the notation 2 X , X k , X ≤k , X ≥k to denote the family of all subsets of X, all subsets of X of size k (that we also call k-subsets of X), all subsets of X of size at most k, and all subsets of X of size at least k, respectively. The family X k is often called the kth level of X. Throughout the paper we use standard order notions.
Using this, we will assume several times throughout the paper that all members of a family F have cardinalities in the interval [n/2 − n 2/3 , n/2 + n 2/3 ] (as this way we lose only o( 1 n n ⌊n/2⌋ ) sets). Note that for our purposes it is always going to be enough to use the interval [n/3, 2n/3] instead of [n/2 − n 2/3 , n/2 + n 2/3 ].
Forbidden subposet problems. The very first result in extremal finite set theory is due to Sperner [18], who proved that if a family F ⊆ 2 [n] does not contain two sets in inclusion, then the size of F is at most n ⌊n/2⌋ and the only families achieving this size are [n] ⌊n/2⌋ and [n] ⌈n/2⌉ . This was later generalized by Erdős [6], who showed that if F ⊆ 2 [n] does not contain a chain of length k + 1 (i.e. nested sets F 1 F 2 · · · F k+1 ), then the size of F is at most k i=1 n ⌊ n−k 2 ⌋+i , the sum of the k largest binomial coefficients of order n. There is a vast literature of Sperner type problems (see the not very recent monograph of Engel [5]), we focus on forbidden subposet problems introduced by Katona and Tarján [11]. We say that the sets F 1 , F 2 , . . . , F |P | form a copy of a poset P , if there exists a bijection i : P → {F 1 , F 2 , . . . , F |P | } such that for any p, p ′ ∈ P the relation p < P p ′ implies i(p) i(p ′ ). A family F of sets is P -free if it does not contain any copy of P . Katona and Tarján initiated the study of the parameter La(n, P ), the maximum size of a P -free family F ⊆ 2 [n] . Note that with this notation the above-mentioned result of Erdős can be formulated as where P k+1 denotes the chain of size k + 1. As a copy of a chain of length |P | in a family F is always a copy of P , the result of Erdős implies La(n, P ) ≤ (|P | − 1) n ⌊n/2⌋ .
Therefore it is natural to ask for the existence and value of lim n→∞ La(n, P ) n ⌊n/2⌋ , denoted by π(P ). It is not known whether π(P ) exists for every poset, but the precise or asymptotic value of La(n, P ) has been determined for many posets and in all known cases the (asymptotically) optimal construction consists of some of the middle levels of [n]. This motivated the following conjecture that first appeared in [9]. Conjecture 1.1. For any poset P let e(P ) denote the largest integer k such that for any j and n the family j+i is P -free. Then π(P ) exists and is equal to e(P ). Conjecture 1.1 was proved for many classes of posets. Let us state one of the nicest results of the area. To do so we need the following definition. For a poset P the Hasse diagram, denoted by H(P ), is a graph whose vertices are elements of P , and xy is an edge if x < y and there is no other element z of P with x < z < y. A poset T is a tree poset if its Hasse diagram is a tree. Let h(T ) denote the length of a longest chain in T . Bukh proved the following. Traces of set families. The trace of a set family is its restriction to a subset of its underlying set. Formally, the trace of a set F on another set X is F | X := F ∩ X, and the trace of a family F on X is As different sets can have the same trace, we obtain |F | X | ≤ |F |. The fundamental result about traces of set families is the so-called Sauer-lemma proved independently by Sauer [16], Shelah [17], and Vapnik and Chervonenkis [19].
The bound in Theorem 1.3 is tight as shown by [n] ≤k−1 and ≥n−k+1 , but there are many other extremal families and a complete characterization is not yet known. This theorem leads naturally in several directions. One of them is the area of forbidden subconfigurations. If H ⊆ 2 [k] is a fixed family, then one can ask for the maximum size of a 'big' family F ⊆ 2 [n] such that for any k-subset X of [n], the trace F | X does not contain a subfamily isomorphic to H. For more details, the interested reader is referred to the survey of Anstee [2] and the references within. Naturally, one can consider several forbidden configurations at once. To mix the areas of forbidden subposet problems and forbidden subconfigurations, we will forbid all configurations that can be described by a poset structure.
We say that F ⊆ 2 [n] is l-trace P -free if for any l-subset L ⊆ [n], the family F | L is P -free. A family F is trace P -free if it is l-trace P -free for all l ≤ n. Let T r(n, P ) be the maximum size of a trace P -free family F ⊆ 2 [n] and T r l (n, P ) be the maximum size of an l-trace P -free family F ⊆ 2 [n] .
If the traces of two sets on some set X are in inclusion, then they remain in inclusion for any subset Y of X, however the traces might coincide on Y . So it is not straightforward from definition that forbidding a subposet in the trace on a smaller subset is a stronger property than doing the same on a larger subset. However, we will prove the following rather easy monotonicity result. Proposition 1.4. For a poset P let E(P ) denote the number of edges in the Hasse diagram H(P ). If E(P ) ≤ k ≤ l, then we have T r k (n, P ) ≤ T r l (n, P ). Proposition 1.4 implies that for any integer k we have T r(n, P k+1 ) = T r k (n, P k+1 ). The value T r k (n, P k+1 ) = k−1 i=0 n i follows from Theorem 1.3. The second author proved in [13] that the only k-trace P k+1 -free families are [n] ≤k−1 and ≥n−k+1 (moreover, he showed that for any fixed k ≤ l we have T r l (n, P k+1 ) = k−1 i=0 n i if n is large enough and the only extremal families are [n] ≤k−1 and ≥n−k+1 ). Note that for any poset P if y(P ) denotes the largest integer m with 2 [m] not containing a copy of P , then by Theorem 1.3 we have T r(n, P ) ≤ y(P ) i=0 n i , i.e. T r(n, P ) grows polynomially in n.
The simplest non-chain posets are and , both being a poset on 3 elements a, b, c with a < b, c and a, b < c. As they are both subposets of P 3 , we have T r(n, ), T r(n, ) ≤ n+1 and taking complements yields T r(n, ) = T r(n, ). Moreover, we know that there exist trace -free and trace -free families of size n + 1, namely [n] ≤1 and [n] ≥n−1 . The first contains and does not contain , while it is the opposite for the second family. On the other hand if we forbid both and as traces, then the family cannot have more than 2 sets.
As a first non-trivial and non-chain instance of the problem of finding T r(n, P ) we will consider the butterfly poset B on 4 elements a, b, c, d with a, b < B c, d.
As remarked above, Theorem 1.3 implies that T r(n, P ) grows polynomially in n and the same argument shows that for any fixed l > y(P ) we have T r l (n, P ) = O(n l−1 ). The situation completely changes when l is close to n. By definition, we have T r n (n, P ) = La(n, P ). Observe that if n is large enough and F consists of consecutive levels of [n], say , then F | X is P -free. This shows the inequality T r n−k (n, P ) ≥ (e(P ) − k + o(1)) n ⌊n/2⌋ .
Therefore we propose the following generalization of Conjecture 1.1.
Conjecture 1.6. For any poset P and integer k < e(P ) we have Moreover, if k ≥ e(P ), then T r n−k (n, P ) = o( n ⌊n/2⌋ ) holds. Note that to see the moreover part of Conjecture 1.6, by Proposition 1.4, it is enough to prove its statement for k = e(P ). Conjecture 1.6 was verified for chains by the second author in [14] and he obtained the exact value of T r n−1 (n, P k+1 ) for any positive integer k in [15]. We prove Conjecture 1.6 for the posets K r,s on r + s elements a 1 , a 2 , . . . , a r , b 1 , b 2 , . . . , b s with a i < b j for any 1 ≤ i ≤ r and 1 ≤ j ≤ s. We will use the notation r for K r,1 and s for K 1,s . Note that e(K r,s ) = 2 if r and s are both at least two and e( s ) = e( r ) = 1. Conjecture 1.1 was verified for K r,s by De Bonis and Katona [4]. Therefore the following theorem implies Conjecture 1.6 in the case of the posets K r,s . Theorem 1.7. For any positive integer s ≥ 1, we have The smallest poset for which Conjecture 1. 1 has not yet been proved is the diamond poset D on four elements a, b, c, d with a < b, c < d. The best known upper bound on La(n, D) is due to Grósz, Methuku, and Tompkins [10]. We will prove that the moreover part of Conjecture 1.6 holds for D.
The remainder of this paper is organized as follows: Section 2 deals with trace P -free families, Theorem 1.5 along with some further remarks are shown there. A general result on (n − 1)-traces of families that implies Theorem 1.7 is proved in Section 3 along with Theorem 1.8 and other results about l-trace P -free families. Finally, Section 4 contains some concluding remarks.
2 Trace P -free families Theorem 1.3 has many proofs in the literature. One of them (obtained independently by Alon [1] and Frankl [8]) uses down-compression. For a set F and an element i, the downcompression operator is defined as and for a family of sets F we define It was proved in [1,8] that if we are given a family F ⊆ 2 [n] such that there does not exist a k-set X with F | X = 2 X , then the same holds for D i (F ) for any i ∈ [n]. As any family F can be turned into a downward closed family (a family D for which C ⊂ D ∈ D implies C ∈ D) by applying a finite number of down-compressions, to prove Theorem 1.3 it is enough to show its statement for downward closed families, which is rather straightforward.
Observe that the trace P -free property is not preserved by down-compression, however there is a way how to obtain bounds on T r(n, P ) by considering only downward closed families. Frankl in [8] introduced the arrow relation (n, m) → (k, l) which, by definition, holds if for any family F ⊆ 2 [n] of size m, there exists a k-set X such that |F | X | ≥ l. With this notation Theorem 1.3 can be formulated as for any pair n ≥ k. Frankl used down-compression to prove the following.
. The following statements are equivalent.
We want to make use of Theorem 2.1 to determine T r(n, P ). In order to do that we make two simple observations. First note that if for some k-set X the trace F | X contains more than La(k, P ) sets, then F cannot be trace P -free. Therefore we obtain the following.
One can go one step further and improve Proposition 2.2. Suppose one determined the value of T r(k, P ) for some small integer k. Then obviously, if for some k-set X the trace F | X contains more than T r(k, P ) sets, then F cannot be trace P -free, so we obtain the following.  Proof. We start with the following simple claim.
Proof of Claim. By Theorem 2.1, it is enough to prove the statement for downward closed families D ⊆ 2 [5] of size 9. If D contains a set D of size 3, then |2 D | = 8 ≥ 6. Otherwise D contains at least 3 sets of size 2. As they are all subsets of [5], for two of them Similarly we have either three 2-sets on three vertices or two 2-sets on four vertices, both cases give 7 sets on three or four vertices.
Suppose F ⊆ 2 [5] is a B-trace free family of size 9. Then by Claim 2.5 there exists a 3-set X with |F | X | ≥ 6. We may suppose that X = [3] and as F is B-trace free, we must have Claim 2.6. Suppose there is a set F ∈ F with 4 ∈ F . Then we have either Proof of Claim. The set F intersects [3] in a 1 or 2-element set. We separate cases according to this. We introduce the notation A := F | [4] , A i := F | [4]\{i} (i ∈ [4]). In particular, we have seen so far that F | [3] Also only one of {2} or {2, 4} can be in A as otherwise they would form a copy of B in A 3 with {1, 2, 4} and ∅ or {4}. We are done with the proof of Claim 2.6.
Let us consider the possible outcomes of Claim 2.6. Let Z = Y \ {5}, then we have either F | Z = Z 1 ∪ Z 2 or F | Z is a copy of the diamond poset. In the first case we can apply Claim 2.6 this time [3] replaced by Y ∩ [4] and 4 by 5 to obtain |F | Y | ≤ 6. In the second case notice that F | Y ⊆ F | Z ∪ {F ∪ {5} : F ∈ F | Z }. As this latter family is a copy of 2 [3] , to ensure the B-free property, we must have |F | Y | ≤ 6. 2 is either a triangle or a star. In the former case we have |D| ≤ 3 + n + 1 < ⌊3n/2⌋ + 2. In the latter case, if the star consists of at most 3 sets, then again |D| ≤ 3 + n + 1 < ⌊3n/2⌋ + 2, while if the star consists of at least 4 sets D 1 , D 2 , D 3 , D 4 , then |D| D 1 ∪D 2 ∪D 3 ∪D 4 | = 10. Now the upper bound in Theorem 1.5 follows from Proposition 2.3, Lemma 2.4 and Lemma 2.7. For the lower bound we consider a family that consists of the empty set, all the 1-element sets, and ⌊n/2⌋ pairwise disjoint 2-element sets. It is easy to see that this family does not contain the butterfly poset, and as it is downward closed, it does not contain it as a trace either. This finishes the proof of Theorem 1.5.

We state the last observation as a general lower bound. Let
La D (n, P ) := max{|F | ⊆ 2 [n] : F is P -free and downward closed}, and La U (n, P ) := max{|F | ⊆ 2 [n] : F is P -free and upward closed}. Proposition 2.8. We have T r(n, P ) ≥ max{La D (n, P ), La U (n, P )}. Let x(n, P ) be the largest integer such that [n] ≤x(n,P ) does not contain P . It is easy to see that x(n, P ) is monotone decreasing in n, so we can define its limit x(P ) and x(n, P ) = x(P ) for n large enough. It is easy to see that Remember that y(P ) is the largest integer such that 2 [y(P )] does not contain P . If the size of a family F ⊆ 2 [n] is larger than y(P ) i=0 n i , then by Theorem 1.3 it contains a subset X of size y(P ) + 1 such that F | X = 2 X holds. Obviously 2 X contains a copy of P by the definition of y(P ), thus we have T r(n, P ) ≤ y(P ) i=0 n i . By the observations above we have If for a poset P we have x(P ) = y(P ), then T r(n, P ) = x(P ) i=0 n i . In particular, Proposition 2.9. If a poset P has a unique maximum element, then T r(n, P ) = x(P ) i=0 n i .

l-traces
We start this section by showing the monotonicity of T r l (n, P ) in l.
Proof of Proposition 1.4. Let E(P ) ≤ k ≤ l and suppose F ⊆ 2 [n] is k-trace P -free. We claim that F is l-trace P -free. Assume otherwise. Then there exists an l-subset X of [n] such that F | X contains a copy F 1 | X , F 2 | X , . . . , F |P | | X of P . For every edge e of the Hasse diagram H(P ) let x e be an element from F i | X \ F j | X if F i | X and F j | X are the sets corresponding to the end vertices of e. As E(P ) ≤ k, we obtain that |{x e : e ∈ E(H(P ))}| ≤ k. Therefore, for any k-set Y containing {x e : e ∈ E(H(P ))} we have that F 1 | Y , F 2 | Y , . . . , F |P | | Y form a copy of P in F | Y contradicting the k-trace P -free property of F .
We continue with the proof of Theorem 1.7.

Proof of Theorem 1.7 (i).
To see the lower bound consider the following well-known construction. Let us partition Let F be the union of the s largest F i 's, and therefore we have For any x ∈ [n], the trace F [n]\{x} contains sets of size ⌊n/2⌋ and ⌊n/2⌋ − 1, so a copy of s would be possible if there existed an (⌊n/2⌋ − 1)-set G that is contained in at least s + 1 sets of F . By construction, there is no such G, therefore F is (n − 1)-trace s -free.
To prove the upper bound let F ⊆ 2 [n] be an (n − 1)-trace s -free family. Let ≥n/2−n 2/3 . Note that F 1 cannot contain a chain of length s + 1 as omitting an element of its smallest set would result in a (s + 1)-chain in the trace contradicting the (n − 1)-trace s -free property. Therefore F 1 contains an antichain F 2 with |F 2 | ≥ |F 1 |/s. We will bound the size of F 2 using the Lubell-function To this end we will count the number of pairs (F, C) with C being a maximal chain in [n] and F ∈ F 2 ∩ C. We will denote by C the set of all maximal chains in [n].
Let us consider G, the shadow of F 2 , and for a set G ∈ G let C G := {C ∈ C : G ∈ C}.
Claim 3.1. For every chain C ∈ C there exist at most s sets G ∈ G with C ∈ C G .
Proof of claim. Suppose to the contrary that C ∈ C contains G 1 for i = 2, 3, . . . , s + 1 and they are all different, as F 2 is an antichain, thus these form a copy of s . Claim 3.2. For every G ∈ G there exist at most s sets F ∈ F 2 with G ⊆ F .

Proof of claim.
As G ∈ G there exists an x / ∈ G with F = G ∪ {x} ∈ F 2 . As F 2 is an antichain, any other F ′ ∈ F 2 with G ⊆ F ′ must not contain x. So if there were F 1 , F 2 , . . . , F s ∈ F 2 other than F all containing G, then the traces of F, F 1 , F 2 , . . . , F s ∈ F 2 on [n] \ {x} would form a copy of s .
Let us now count the number of pairs (F, C) with C being a maximal chain in [n] and F ∈ F 2 ∩ C. On the one hand it is On the other hand it is at most G∈G C∈C G F ∈C∩F 2

1.
As F 2 is an antichain, no chain C ∈ C G can contain a set F ∈ F 2 with F ⊆ G. Therefore, by Claim 3.2 and the condition that F 2 consists only of sets of size from [n/2 − n 2/3 , n/2 + n 2/3 ], we have ≤n/2+n 2/3 | = o( 1 n n ⌊n/2⌋ ), the proof of Theorem 1.7 (i) is finished.
Proof of Theorem 1.7 (iii). The statement follows from part (i) and the following claim. We denote by K r,1,s the poset on r + 1 + s elements a 1 , a 2 , . . . , a r , c, b 1 , b 2 , . . . , b s with a i < c < b j for any 1 ≤ i ≤ r and 1 ≤ j ≤ s. Proof of claim. Let F ⊆ 2 [n] be a family of size T r n−1 (n 2s+1 ) + T r n−1 (n, 2r+1 ) + 1. We can find pairs (F i , x i ) for 1 ≤ i ≤ T r n−1 (n, 2r+1 ) + 1 and F i ∈ F , x i ∈ [n] such that all F i 's are distinct and F i | [n]\x i is the bottom element of a copy of 2s+1 in F | [n]\{x i } . Therefore there exists a y ∈ [n] such that contains a copy of 2r+1 , say with F 1 | [n]\{y} being the top element. We claim that F | [n]\{x 1 ,y} contains a copy of K r,1,s . Indeed, let F 1 , G 1 , G 2 , . . . , G 2s+1 ∈ F be sets the traces of which form a copy of 2s+1 on [n]\{x 1 } (these sets exist by the definition of F 1 and x 1 ). As removing one element may cause at most 2 sets to have the same trace, F 1 and at least s of the G i 's will have distinct traces on [n] \ {x 1 , y} and thus will form a copy of r with F 1 | [n]\{x 1 ,y} being the bottom element. The same reasoning shows that we can pick r of F 2 , F 3 , . . . , F 2r+2 such that their traces on [n] \ {x 1 , y} together with F 1 | [n]\{x 1 ,y} form a copy of r with F 1 | [n]\{x 1 ,y} being the top element. Putting these copies of s and r together, we obtain a copy of K r,1,s . Note that K r,s is a subposet of K r,1,s , hence Theorem 1.7 (iii) is proved.
Let T be a tree poset with a unique maximum element m. We define two new posets obtained from T . Let T k denote the poset obtained from T by replacing m with an antichain of size k. Equivalently, such that the m i 's form an antichain, for any t ∈ T \ {m} and 1 ≤ i ≤ k we have t < T k m i and for any t, t ′ ∈ T \ {m} we have t < T k t ′ if and only if t < T t ′ . Note that T k is not a tree poset unless there is a unique element of T that precedes m. Also, if T k is not a tree poset, then e(T k ) = e(T ) + 1 = h(T ). Let T ⊗r be the tree poset defined recursively (with respect to its height) in the following way: if T = P 1 is the poset with one element, then T ⊗r = P 1 for any r. Otherwise, if the maximum element m of T has c children in its Hasse-diagram and the posets below its children are T 1 , T 2 , . . . , T c , then the maximum element of T ⊗r has c·r children m 1 , m 2 , . . . , m cr such that m (j−1)r+i is the maximum element of a poset isomorphic to T ⊗r j for every 1 ≤ j ≤ c and 1 ≤ i ≤ r.  Proof of lemma. Let F ⊆ 2 [n] be a family of size La(n, T ⊗2 ) + T r n−1 (n, s−1 ) + 1. Then F contains a copy of T ⊗2 . Let F 1 be the set of this copy corresponding to the top element of T ⊗2 . As F \ {F 1 } is still larger than La(n, T ⊗2 ), we can pick a set F 2 ∈ F \ {F 1 } that corresponds to the top element of T ⊗2 in a copy in F \ {F 1 }. Repeating this, we can obtain sets F 1 , F 2 , . . . , F T r n−1 (n, s−1 )+1 with the property that for every F j there exists a copy of T ⊗2 in F in which they correspond to the top element. Let us write We know that there exists a copy of T ⊗2 in F with F s playing the role of the top of T ⊗2 . We claim that we can take some of the sets (including F s ) of this copy of T ⊗2 such that their traces on [n] \ {x} form a copy of T and thus together with F 1 \ {x}, F 2 \ {x}, . . . , F s−1 \ {x} they form a copy of T s in F [n]\{x} . To see this, we only need to observe that if G 1 , G 2 G and G 1 = G 2 , then for any y at least one of G 1 \ {y}, G 2 \ {y} is a strict subset of G \ {y}. So we can pick the sets of the copy of T recursively starting with F s .
Thus we indeed obtained a copy of T s in F | [n]\{x} .
Now the upper bound in Theorem 3.4 follows from Lemma 3.5, Theorem 1.7 (i) and Theorem 1.2 using the simple observation that the height of T and T ⊗r are the same and therefore we have e(T ) = e(T ⊗r ) for any integer r.
The lower bound is due to the general observation made before Conjecture 1.6 that T r n−1 (n, P ) ≥ (e(P ) − 1 − o(1)) n ⌊n/2⌋ holds for any poset P . Note that Theorem 1.7 (ii) follows by applying Theorem 3.4 to T = r .
As Aα is the smallest labeled edge in the cycles AαBβ and AαBγ, using that the smallest and the largest labeled edges must be adjacent, we obtain l(Bβ) < l(Aβ) and l(Bγ) < l(Aγ). But then in the cycle AβBγ we have l(Bβ) < l(Aβ), l(Bγ) < l(Aγ), so the smallest labeled edge is Bβ and the largest labeled edge is Aγ, contradicting that these should be adjacent.
Let us consider a (symmetric) chain partition C of 2 [n] , i.e. C consists of n ⌊n/2⌋ chains C such that ∪ C∈C C = 2 [n] and for any pair C, C ′ ∈ C we have C ∩ C ′ = ∅. For any C ∈ C let us define the graph G C with vertex set [n] and edge set {e ∈ [n] 2 : ∃C ∈ C C ∪ e ∈ F }.
Let e C denote the number of edges in G C and let us bound C∈C e C . Every F ∈ F contains |F | 2 pairs and each of them belongs to different chains. Moreover, for every C and every edge e ∈ E(G C ) there can be at most 3 sets F ∈ F containing e and F \ e ∈ C (as otherwise these sets would form a 4-chain, i.e. a special copy of the diamond), so we obtain 1 54 On the other hand for any C ∈ C let us define the labeling ℓ : E(G C ) → {0, 1, . . . , n} by letting ℓ(e) := |C| with C ∈ C, C ∪ e ∈ F (if there are more such sets C, then take the size of an arbitrary one). Note that G C and the labeling ℓ satisfy the conditions of Lemma 3.6. Indeed, if e 1 , e 2 , e 3 , e 4 are consecutive edges of a 4-cycle in G C with C 1 , C 2 , C 3 , C 4 ∈ C and e i ∪ C i = F i ∈ F such that |C 1 | ≤ |C 2 |, |C 4 | ≤ |C 3 |, then the traces of the F i 's on [n] \ e 1 form a copy of the diamond poset. Lemma 3.6 implies G C does not contain a complete bipartite graph with parts of size 3 and 17. Therefore the celebrated Kővári -T. Sós -Turán theorem [12] implies e C = O(n 2−1/3 ) for all C ∈ C. Summing over C we obtain 1 54 Rearranging yields the theorem.

Concluding remarks
We finish this article by posing some remarks and problems concerning our results.
• We introduced the functions La D (n, P ) and La U (n, P ) as lower bounds on T r(n, P ). They seem to be interesting on their own right, and we are not aware of any earlier study on them. Natural questions arise about the order of magnitude of La D (n, P ).
It is natural to ask if we can find an upper bound on La D (n, P ) using x(P ). However, we show a poset P m for every m such that x(P m ) = 1 and La D (n, P ) = Ω(n m ).
Let (P m , <) consist of a minimal element a, 2 m + 1 elements b 1 , . . . , b 2 m +1 with a < b i for 1 ≤ i ≤ 2 m + 1 and m ′ := 2 m +1 2 elements c 1 , . . . , c m ′ such that for every two different b k , b l there is exactly one c j with b k , b l < c j . Observe that we have x(P m ) = 1 as a family consisting of sets of size at most 2 is P m -free if and only if its 2-element sets do not contain a copy of the complete graph K m . On the other hand consider a partition of [n] into m sets A 1 , . . . , A m of almost equal size. Consider the family F of sets that intersect every A j in at most one element. It is obvious that F is downward closed and has cardinality Ω(n m ). We will show it is P m -free.
Suppose by contradiction that F contains a copy of P m . Let F 1 be the subfamily consisting of the sets that correspond to b 1 , . . . , b 2 m +1 . If two distinct element of A j are both contained in members of F 1 , then they are both contained in a set corresponding to c k for some k, which is impossible. Thus ∪F 1 intersects every A j in at most one vertex, which implies | ∪ F 1 | ≤ m. Therefore we have |F 1 | ≤ 2 m , a contradiction.
• Concerning the connection of La D (n, P ), La U (n, P ) and T r(n, P ), the obvious question is the following: is Proposition 2.8 sharp for n large enough? We know that 6 = T r(3, B) > max{La D (3, B), La U (3, B)} = 5, but also that T r(n, B) = max{La D (n, B), La U (n, B)} if n > 4. The sharpness of Proposition 2.8 would mean that we could use down-compression in forbidden subposet problems for traces, similarly to Theorem 2.1.
Another possible improvement that would essentially be equivalent to using down-compressions is at Proposition 2.3. Can we replace T r by max{La D , La U } in Proposition 2.3? For the butterfly poset and n = 3 these are different but (n, m) → (5, 9) would give the same bound. On the other hand, note that Proposition 2.3 is sharp for any poset P . Indeed, T r(n, P ) ≥ min{m : ∃k (n, m) → (k, T r(k, P ) + 1)} − 1, as shown by k = n. The question is if we can chose a small k. More precisely, is there a constant c(P ) for every poset P such that determining T r(c(P ), P ) and using Proposition 2.3 is enough to find T r(n, P ) for every n (like c(P ) = 5 for the butterfly poset)?