New bounds for a hypergraph Bipartite Tur\'an problem
Beka Ergemlidze, Tao Jiang, Abhishek Methuku

TL;DR
This paper establishes that the maximum size of triple systems avoiding a specific hypergraph grows asymptotically as t^{1+o(1)} when t becomes large, refining previous bounds and answering an open question.
Contribution
The authors determine the asymptotic growth rate of the extremal function for avoiding K_{2,t}^{(3)} hypergraphs, improving bounds and resolving an open problem.
Findings
g(t) = heta(t^{1+o(1)}) as t o
Established asymptotic growth rate of extremal function
Improved understanding of hypergraph Turán problems
Abstract
Let be an integer such that . Let denote the triple system consisting of the triples , for , where the elements are all distinct. Let denote the maximum size of a triple system on elements that does not contain . This function was studied by Mubayi and Verstra\"ete, where the special case was a problem of Erd\H{o}s that was studied by various authors. Mubayi and Verstra\"ete proved that and that for infinitely many , . These bounds together with a standard argument show that exists and that \[\frac{2t-1}{3}\leq g(t)\leq t^4.\] Addressing the question of…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsLimits and Structures in Graph Theory · Mathematical Approximation and Integration
\setstackEOL
\xpatchcmd
New bounds for a hypergraph Bipartite Turán problem
Beka Ergemlidze Tao Jiang Abhishek Methuku Alfréd Rényi Institute of Mathematics, Budapest. E-mail: [email protected] of Mathematics, Miami University, Oxford, OH 45056, USA. E-mail: [email protected] of Mathematics, École Polytechnique Fédérale de Lausanne, Switzerland. E-mail: [email protected]
Abstract
Let be an integer such that . Let denote the triple system consisting of the triples , for , where the elements are all distinct. Let denote the maximum size of a triple system on elements that does not contain . This function was studied by Mubayi and Verstraëte [8], where the special case was a problem of Erdős [1] that was studied by various authors [3, 8, 9].
Mubayi and Verstraëte proved that and that for infinitely many , . These bounds together with a standard argument show that exists and that
[TABLE]
Addressing the question of Mubayi and Verstraëte on the growth rate of , we prove that as ,
[TABLE]
1 Introduction
An -graph is an -uniform hypergraph. Let be a family of -graphs and let denote the maximum number of edges in an -graph on vertices containing no member of . We call the Turán number of . Determining the asymptotic order of is generally very difficult. For an excellent survey on the study of hypergraph Turán numbers, see [7]. In this paper, we study a hypergraph Turán problem that is motivated by the study of Turán numbers of complete bipartite graphs as well as by a question of Erdős.
Definition 1**.**
Let be an integer. Let be a bipartite graph with an ordered bipartition . Suppose that . Let be disjoint sets of size that are disjoint from . Let denote the -graph with vertex set and edge set .
Let be positive integers. If is the complete bipartite graph with an ordered bipartition where , then let be denoted by .
Definition 2**.**
For all , let denote the maximum number of edges in an -vertex -graph containing no four edges with and .
Note that , and in general . Erdős [1] asked whether when . Füredi [3] answered Erdős’ question affirmatively. More precisely, he showed that for integers with and ,
[TABLE]
The lower bound is obtained by taking the family of all -element subsets of containing a fixed element, say , and adding to the family any collection of pairwise disjoint -element subsets not containing . For , Füredi also gave an alternative lower bound construction using Steiner systems. An -Steiner system is an -uniform hypergraph on in which every -element subset of is contained in exactly one hyperedge. Füredi observed that if we replace every hyperedge in by all its -element subsets then the resulting triple system has triples and contains no copy of . This slightly improves the lower bound in (1) for to , for those for which exists. The upper bound in (1) was improved by Mubayi and Verstraëte [8] to . They obtain this bound by first showing , and then combining it with a simple reduction lemma. This was later improved to by Pikhurko and Verstraëte [9].
Motivated by Füredi’s work, Mubayi and Verstraëte [8] initiated the study of the general problem of determining for any . They showed that for any and
[TABLE]
and that for infinitely many , , where the lower bound is obtained by replacing each hyperedge in with all its -element subsets.
Mubayi and Verstraëte noted that exists and raised the question of determining the growth rate of . It follows from their results that
[TABLE]
In this paper, we prove that as ,
[TABLE]
showing that their lower bound is close to the truth. More precisely, we prove the following.
Theorem 1**.**
For any , we have
[TABLE]
Notation. Given a hypergraph (or a graph) , throughout the paper, we also denote the set of its edges by . For example denotes the number of edges of . Given two vertices in a graph , let denote the common neighborhood of and in . We drop the subscript when the context is clear.
2 Proof of Theorem 1: -free hypergraphs
We will use the a special case of a well-known result of Erdős and Kleitman [2].
Lemma 1**.**
Let be a -graph on vertices. Then contains a -partite -graph, with all parts of size , and with at least hyperedges.
Let us define the sets , and . Throughout the proof we define various -partite -graphs whose parts are and .
Suppose is a -free -partite -graph on vertices with parts and . First let us show that it suffices to prove the following inequality.
[TABLE]
It is easy to see that inequuity (4) and Lemma 1 together imply that any -free -graph on vertices contains at most hyperedges, from which Theorem 1 would follow after replacing by .
In the remainder of the section, we will prove (4). Let us introduce the following notion of sparsity.
Definition 3** (-sparse and -dense pairs).**
Let be a positive integer. Let be a bipartite graph with parts . Let be two different vertices such that or . Then we call a -dense pair of if . We call a -sparse pair of if but are still contained in a copy of in . Note that it is possible that is neither -sparse nor -dense.
The following Procedure about making a bipartite graph -free lies at the heart of the proof. (We think of as the parameter of the Procedure , that is changed throughout the proof.)
In the procedure , initially for all the pairs (with and ) the sets are set to be empty. Then as the edges are being deleted during the procedure, possibly, new -sparse pairs are being created. When this happens, Step 1 redefines the sets and gives them some non-empty values. (They get non-empty values due to the fact that is -sparse, which implies that is contained in a copy of , so there is at least one -dense pair in the common neighborhood of .) Therefore, these values stay unchanged throughout the rest of the procedure.
Notice that at the point was redefined, the pair was -sparse, so number of common neighbors is less than . Therefore, as is a subset of the common neighborhood of and , we also have . Moreover, since is defined as a spanning forest with the vertex set , we have . Also, it easily follows from the definition of that . Finally, notice that does not contain any isolated vertices, because its vertex set spans all of its edges, by definition. Therefore, . At the end of the procedure, the sets are renamed as . Note also that if a pair never becomes -sparse in the process then .
Observation 1**.**
For every and , we have
- (1)
. 2. (2)
. 3. (3)
. 4. (4)
.
For convenience, throughout the paper we (informally) say that the sets , , are defined by applying Procedure to a graph to obtain the graph , instead of saying that the input to Procedure is and the output is the graph and the sets , , .
Claim 1**.**
Let the sets (for and ) be defined by applying Procedure to a bipartite graph to obtain . Let denote the number of common neighbors of vertices in the graph . Then
[TABLE]
Moreover .
Proof.
Combining the parts (3) and (4) of Observation 1, we have . Combining the parts (1) and (2) of Observation 1, we obtain , proving the first part of the claim.
To prove the second part, notice that is a common neighborhood of in some subgraph of , we have . Combining this with part (3) of Observation 1, we obtain , as required. ∎
Finally, let us note the following properties of the graph obtained after applying the procedure.
Observation 2**.**
Let the sets (for and ) be defined by applying Procedure to a bipartite graph to obtain . Then
Every edge in is contained in at most members of and in at most members of . 2. 2.
For any set of edges in , removing the edges of from decreases the number of -dense pairs by less than .
Definition 4**.**
Let be a -partite -graph with parts and .
For each , let be the bipartite graph with parts and , whose edge set is . The graphs and are defined similarly.
Definition 5** (Applying Procedure to a hypergraph).**
Let be a -partite -graph with parts and . We define the hypergraph as follows:
For each , let , , be the graphs obtained by applying the procedure to the graphs , , respectively.
For each edge which was removed from by the procedure (i.e. ) we remove the hyperedge from (it may have been removed already). Similarly for each edge (resp. ) which was removed from (resp. ) by the procedure we remove the hyperedge (resp. ) from . Let the resulting hypergraph be . More precisely,
[TABLE]
We say is obtained from by applying the Procedure .
Remark 1**.**
Let be obtained by applying the Procedure to the hypergraph . Then,
[TABLE]
Indeed, if then it is easy to see that or or .
Lemma 2**.**
Let be an even integer and be a bipartite graph with parts and . Suppose is the graph obtained by applying Procedure to . Then is -free.
Proof.
Let us define a -broom of size to be a set of -sparse pairs (with ), and a -dense pair such that is contained in the common neighborhood of for every . Note that either and or and .
Claim 2**.**
There is no -broom of size in .
Proof.
Suppose by contradiction that there is a set of -sparse pairs (with ), and a -dense pair such that is contained in the common neighborhood of and for every . Then the edge is contained in the sets for every , which contradicts Observation 2. ∎
Let us suppose for a contradiction (to Lemma 2) that contains a copy of . Then contains at least one -dense pair. Without loss of generality we may assume there is a -dense pair in . Suppose (for ) are all the -dense pairs of containing the vertex . For each , let be the common neighborhood of and in . By definition, for .
Claim 3**.**
For any , we have .
Proof.
Let us assume for contradiction that there exists a such that . Let be obtained from by deleting all the edges from to . For each , the pair has no common neighbor in since we have removed all the edges from to . Thus the pair is not -dense in . So in forming from the number of -dense pairs decreases by at least , while the number of edges decreases by edges, contradicting Observation 2. ∎
Let . For each vertex and let
[TABLE]
[TABLE]
In the next two claims, we will prove two useful inequalities concerning and .
Claim 4**.**
For each , .
Proof.
Suppose for contradiction that there is a vertex such that . Let us delete all the edges of the form , , from and let the resulting graph be . Since we deleted edges, by Observation 2, the number of -dense pairs decreases by less than . So there exists such that is (still) -dense in . That is, , where denotes the common neighborhood of and in . Clearly each pair of vertices in is contained in a copy of in (and hence in ).
For each pair of vertices in , since it is contained in a copy of in , it is either -sparse or -dense in . Note that . If all the pairs with are -sparse in then the set of these pairs together with is a -broom of size at least in , which contradicts Claim 2. So there exists a vertex such that is -dense in . Since is adjacent to both and , by the definition of , for some . However, by definition, in forming we have removed from . This contradicts and completes the proof. ∎
Claim 5**.**
[TABLE]
Proof.
Fix any with . Since is -dense in , every pair is contained in some copy of and hence is either -dense or -sparse in . Let be any vertex in and let S(v)=\{y\in B_{j}\mid\{v,y\}\text{ is qG^{\prime}}\}. By definition, the set together with is a -broom of size . By Claim 2, . Since , we have
[TABLE]
Note that (5) holds for every and every .
Let us define an auxiliary bipartite graph with a bipartition in which a vertex is joined to a vertex if and only if . Let be an arbitrary subset of . The neighborhood of in is precisely . By Claim 3, . Since this holds for every , by Hall’s theorem [5] there exist distinct vertices , for . By (5), for every . Hence
[TABLE]
∎
If we view as a hypergraph on the vertex set , then the degree of a vertex in it is precisely and the degree sum formula yields
[TABLE]
Using Claim 4 and Claim 5 we have
[TABLE]
which contradicts (6). This completes proof of Lemma 2. ∎
In the next subsection we will prove a general lemma about making an arbitrary hypergraph -free (for any given value of ). This lemma is used several times in the following subsections.
2.1 Applying Procedure to an arbitrary hypergraph
Let be an even integer and let . Let be an arbitrary -free -partite -graph with parts and . In this subsection we will prove the following lemma that estimates the number of edges removed from the graphs for , when the Procedure is applied to them. This lemma together with Remark 1 will allow us to estimate the number of edges removed from when the Procedure is applied to it.
Throughout this subsection, denotes the set of common neighbors of the vertices in the graph .
Lemma 3**.**
Let be an even integer. Let be an arbitrary -free -partite -graph with parts and . Let for . For each and any or , let be defined by applying the procedure to and let the resulting graph be . Then,
[TABLE]
Proof of Lemma 3.
First let us prove the following claim.
Claim 6**.**
Let or . Then is -dense in less than of the graphs , .
Proof.
Without loss of generality, suppose that . Suppose for contradiction that is -dense in of the graphs , . Without loss of generality suppose is -dense in . Then for . Therefore, we can greedily choose distinct vertices such that for each . For each , since we have . However, the set of hyperedges forms a copy of in , a contradiction. ∎
Note that when procedure is applied to (to obtain ), Step 1 and Step 2 may be applied several times (and each time one of these steps is applied it may delete an edge of ).
For each , let denote the number of -dense pairs of . By Claim 6, we know that each pair with or , is -dense in less than different graphs (for ). Therefore,
[TABLE]
For each , let denote the total number of edges that were removed by Step 1 when procedure is applied to and be the number of edges removed by Step 2 when procedure is applied to . Then , so .
First, we bound . Let . Observe that whenever a set of edges were removed by Step 2 of Procedure applied to , the number of -dense pairs decreased by at least . Hence . So summing up over all , and using (7), we get
[TABLE]
Next, we bound . Let . If an edge was removed from by Step 1 of the procedure then there are vertices such that for every or for every . So
[TABLE]
Therefore,
[TABLE]
This is equivalent to the following.
[TABLE]
Combining this inequality with (8) completes the proof of Lemma 3. ∎
2.2 The overall plan
Let us define the sequence as follows. Let where is an integer such that . For each , let and . Clearly , moreover
[TABLE]
So we have
[TABLE]
Now we apply the procedure to the hypergraph (recall Definition 5) to obtain a -free hypergraph . For each we obtain -free hypergraph by applying the procedure to the hypergraph .
This way, in the end we will get a -free hypergraph . In the following section, we will upper bound . Then in the next section, using the information that is -free, we will upper bound for each . Then we sum up these bounds to upper bound the total number of deleted edges (i.e., ) from to obtain . Finally, we bound the size of , which will provide us the desired bound on the size of .
2.3 Making -free
First, we are going to prove an auxiliary lemma that is similar to Lemma A.4 of [8]. In an edge-colored multigraph , an -frame is a collection of edges all of different colors such that it is possible to pick one endpoint from each edge with all the selected endpoints being distinct.
Lemma 4**.**
Let be an edge-colored multigraph with edges such that each edge has multiplicity at most and each color class has size at most . If contains no -frame then .
Proof.
Consider a maximum frame , say with edges such that for every , has color and that there exist with being distinct. By our assumption, . Let be any edge with a color not in . Then both vertices of must be in , otherwise give a larger frame, a contradiction. On the other hand, each edge with both of its vertices in has multiplicity at most . Hence there are at most edges with colors not in . The number of edges with color in is at most by our assumption. So . ∎
Let us recall that is partite -free hypergraph with . For convenience we denote where . For each and any or , let , and be defined by applying the procedure on and let the obtained graph be .
First, observe that according to our definition.
Claim 7**.**
Let or . Then .
Proof.
Let be an edge-colored multigraph in which a pair of vertices is an edge of color whenever is an edge of . The number of edges of color in is . By Claim 1 we have . Hence the number of edges in each color class of is less than .
Let be an arbitrary edge of and let . For each , the pair is -dense in by the definition of . Therefore, by Claim 6, we have . So has multiplicity less than in . Since is arbitrary, the multiplicity of each edge of is less than .
Next, observe that contains no -frame. Indeed, otherwise without loss of generality we may assume that contains edges , where has color for each and are distinct. For each since , in particular (where denotes the common neighborhood of and in ), which means that . But now, forms a copy of , contradicting being -free.
Therefore, applying Lemma 4, we have . By Claim 1, we have
[TABLE]
So
[TABLE]
which proves the claim. ∎
By Lemma 3 we have
[TABLE]
Combining it with Claim 7 we get
[TABLE]
Therefore, as , we have
[TABLE]
So,
[TABLE]
By symmetry, using the same arguments, we have
[TABLE]
and
[TABLE]
Therefore, by Remark 1, we have
[TABLE]
2.4 Making a -free hypergraph -free
In this subsection, we fix a with . Recall that is -free, and is obtained by applying the to . Our goal in this subsection is to estimate . The key difference between arguments in this subsection and in the previous subsection is that now in addition to being -free we can also utilize the fact that is -free. In particular, this extra condition leads to Claim 8, which improves upon Claim 7.
For convenience of notation, in this subsection, let for each . For every and every or let the sets and be defined by applying the procedure to the graph , to obtain the graph .
Claim 8**.**
Let or . Then .
Proof.
For each we denote the set of common neighbors of in as . For each , since is -free, is -free and so .
Without loss of generality let us assume . For each vertex of , let . We claim that . Indeed, for each , we have . So the set of hyperedges form a copy of in . Thus if , then contains a copy of , a contradiction. Therefore, , as desired.
Consider an auxiliary bipartite graph with parts and where the vertex is adjacent to in if and only if . Then by the discussion in the previous paragraph, each vertex has degree , and each vertex has degree . In other words, the maximum degree in is less than .
We claim that does not contain a matching of size . Indeed, suppose for a contradiction that the edges (i.e., for ) form a matching of size in . Then the set of hyperedges , , form a copy of in , a contradiction, as desired.
Since does not contain a matching of size , by the König-Egerváry theorem it has a vertex cover of size less than . This fact combined with the fact that the maximum degree of is less than , implies that the number of edges of is less than . On the other hand, the number of edges in is . Therefore, . This, combined with the fact that for each , (see Claim 1), completes the proof of the lemma. ∎
By Lemma 3, we have
[TABLE]
Now using Claim 8, we have
[TABLE]
Since we have
[TABLE]
So,
[TABLE]
By symmetry, using the same arguments, we have
[TABLE]
and
[TABLE]
Therefore, by Remark 1, we have
[TABLE]
2.5 Putting it all together
[TABLE]
By (10) we have , so we obtain,
[TABLE]
Notice that is -free and . Therefore is -free. Moreover, we know that the hypergraph is -partite and -free with parts (as it is a subhypergraph of ). Now we bound the size of .
Claim 9**.**
We have .
Proof.
Suppose for a contradiction that . For any pair of vertices with and , let denote the number of hyperedges of containing the pair . Then the number of copies of in of the form where , , is
[TABLE]
As the average codegree (over all the pairs ) is more than , by convexity, this expression is more than
[TABLE]
This means there exist a pair and a set of pairs such that whenever . Let be a bipartite graph whose edges are elements of . Since has edges, it either contains a matching with edges or a vertex of degree (see Lemma A.3 in [8] or the last paragraph of our proof of Claim 8 for a proof). In the former case, the set of all hyperedges of the form with , form a copy of in , a contradiction. In the latter case, let be the neighbors of in . Then the set of hyperedges form a copy of in , a contradiction again. This completes the proof of the claim. ∎
Combining (13) with Claim 9, we have thus proving (4), which implies Theorem 1, as desired.
3 Concluding remarks
Recall that given a bipartite graph with an ordered bipartition , where , is the -graph with vertex set and edge set , where are disjoint -sets that are disjoint from . A standard reduction argument such as the one used in the proof of Theorem 1.4 in [8] can be used to show the following.
Proposition 1**.**
Let be integers and a bipartite graph with an ordered bipartition . There exists a constant depending only on such that
[TABLE]
Thus, by Theorem 1 and Proposition 1, for all , we have for some constant , depending only on . On the other hand, taking the family of all -element subsets of containing a fixed element shows that . Recall that in the case, a better lower bound of was shown by Mubayi and Verstraëte [8]. For , we are able to improve the lower bound to as follows.
Proposition 2**.**
We have
[TABLE]
Proof.
(Sketch.) Consider a -free graph with edges where each vertex has degree . (Such a graph exists by a construction of Füredi [3].) Let us a define a -graph and . In other words, let the edges of be the vertex sets of induced -matchings in . Via standard counting, it is easy to show that . It remains to show is -free.
Claim 10**.**
If , then there is a vertex such that .
Proof.
By our assumption, and both induce a -matching in . Without loss of generality, suppose . If then we are done. Otherwise, we have or , both contradicting being an induced matching in . ∎
Suppose for contradiction that has a copy of with edge set . By Claim 10, for each , there exists a vertex such that . This yields a copy of in , a contradiction. ∎
For , we do not yet have a lower bound that is asymptotically larger than . It would be interesting to narrow the gap between the lower and upper bounds on .
It will be interesting to have a systematic study of the function . Mubayi and Verstraëte [8] showed that and that if then and speculated that is the correct order of magnitude. The case when is a tree is studied in [4], where the problem considered there is slightly more general. The case when is an even cycle has also been studied. Let denote where is the even cycle of length . It was shown by Jiang and Liu [6] that , for some positive constants depending on . Using results in this paper and new ideas, we are able to narrow the gap to , for some positive constants depending on . We would like to postpone this and other results on the topic for a future paper.
Finally, motivated by results on and , we pose the following question.
Question 1**.**
Let . Let be the family of bipartite graphs with an ordered bipartition in which every vertex in has degree at most in . Is it true that there is a constant depending on such that ?
Acknowledgments
The research of the first and third authors was supported by the Doctoral Research Support Grant of CEU, and by the Hungarian National Research, Development and Innovation Office NKFIH, grant K116769. The first and third authors are especially grateful for the generous hospitality of Miami University.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] P. Erdős, Problems and results in combinatorial analysis, Proceedings of the Eighth Southeastern Conference on Combinatorics, Graph Theory and Computing, Louisana State University, Baton Rouge, LA, 1977, Congressus Numerantium , Vol XIX, Utilitas Mathematics, Winnipeg, Manchester, 1977, 3–12.
- 2[2] P. Erdős, D. Kleitman, On coloring graphs to maximize the proportion of multicolored k 𝑘 k -edges. J. Combinatorial Theory 5 (1968) 164–169.
- 3[3] Z. Füredi, Hypergraphs in which all disjoint pairs have distinct unions. Combinatorica 4(2-3) (1984) 161–168.
- 4[4] Z. Füredi, T. Jiang, D. Mubayi, A. Kostochka, J. Verstraëte, The extremal number for ( a , b ) 𝑎 𝑏 (a,b) -paths and other hypergraph trees, manuscript, 19 pp., April 12, 2018.
- 5[5] P. Hall, On Representatives of Subsets. J. London Math. Soc 10 , (1935) 26–30.
- 6[6] T. Jiang, X. Liu, Turán numbers of enlarged cycles, manuscript.
- 7[7] P. Keevash, Hypergraph Turán problems, Surveys in Combinatorics, Cambridge University Press, 2011, 83–140.
- 8[8] D. Mubayi, J. Verstraëte, A hypergraph extension of the bipartite Turán problem. J. Combinatorial Theory, Series A 106.2 (2004): 237–253.
