Large homogeneous subgraphs in bipartite graphs with forbidden induced subgraphs
Maria Axenovich, Casey Tompkins, Lea Weber

TL;DR
This paper investigates the structure of bipartite graphs avoiding certain subgraphs, showing that for strongly acyclic graphs, the largest homogeneous subgraphs grow linearly with the size of the graph, except for four special cases.
Contribution
It proves that the maximum size of homogeneous subgraphs in bipartite graphs avoiding strongly acyclic subgraphs is linear in the number of vertices, with four exceptions.
Findings
h(Forb(n, H)) is linear in n for all strongly acyclic H except four.
For non-strongly acyclic H, h(Forb(n, H)) is O(n^{1-s}) for some s>0.
The paper characterizes the growth of homogeneous subgraphs in bipartite graphs with forbidden induced subgraphs.
Abstract
For a bipartite graph G, let h(G) be the largest t such that either G or the bipartite complement of G contain K_{t,t}. For a class F of graphs, let h(F)= min {h(G): G\in F}. We say that a bipartite graph H is strongly acyclic if neither H nor its bipartite complement contain a cycle. By Forb(n, H) we denote a set of bipartite graphs with parts of sizes n each, that do not contain H as an induced bipartite subgraph respecting the sides. One can easily show that h(Forb(n,H))= O(n^{1-s}) for a positive s if H is not strongly acyclic. Here, we prove that h(Forb(n, H)) is linear in n for all strongly acyclic graphs except for four graphs.
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.
Large homogeneous subgraphs in bipartite graphs with forbidden induced subgraphs
Maria Axenovich Karlsruhe Institute of Technology, Karlsruhe, Germany, [email protected]. ββ
Casey Tompkins Karlsruhe Institute of Technology, Karlsruhe, Germany, [email protected]. ββ
Lea Weber Karlsruhe Institute of Technology, Karlsruhe, Germany, [email protected].
Abstract
For a bipartite graph , let be the largest such that either contains , a complete bipartite subgraph with sides of size each or a bipartite complement of contains . For a class , let . We say that a bipartite graph is strongly acyclic if neither nor its bipartite complement contain a cycle. By we denote a set of bipartite graphs with parts of sizes each, that do not contain as an induced bipartite subgraph respecting the sides. One can easily show that for a positive if is not strongly acyclic. Here, we prove that is linear in for all strongly acyclic graphs except for four graphs.
Introduction
A conjecture of ErdΕs and Hajnal [3] states that for any graph there is a constant such that any -vertex graph that does not contain as an induced subgraph has either a clique or a coclique on at least vertices. While this conjecture remains open, see for example a survey by Chudnovsky [2], we address the bipartite setting of the problem.
Let be a bipartite graph with parts of size each, we write , and further write . We shall often depict the sets and as sets of points on two horizontal lines in the plane and call the top part and the bottom part. We say that a graph is the *bipartite complement * of if it has the same vertex set as and its edge set is . We denote the bipartite complement of a graph by . By we denote the largest integer such that there are with and for all , i.e., and form a biclique. By we denote the largest integer such that there are with and for all , i.e., and form a co-biclique. Let .
For a bipartite graph and a bipartite graph we say that is an *induced bipartite subgraph * of respecting sides if , , and for any , if and only if . An induced subgraph of is a *copy * of in if is isomorphic to such that the isomorphic image of is contained in and the isomorphic image of is contained in . Let denote the set of all bipartite graphs with parts of size each that do not contain a copy of as an induced bipartite subgraph respecting sides. We call a bipartite graph -free if it does not contain a copy of . Let
[TABLE]
It is implicit from a result shown by ErdΕs, Hajnal, and Pach [4], that for any bipartite with the smallest part of size , that . A standard probabilistic argument shows that if or its bipartite complement contains a cycle, then for a positive . Here, we address the question of when is linear in . We say that a bipartite graph is strongly acyclic if neither nor its bipartite complement contain a cycle. It is not difficult to show that could be linear only if is strongly acyclic. We show that for all but at most four strongly acyclic graphs , is linear in . Moreover, for several graphs we determine exactly.
Theorem 1**.**
There is a set of at most four graphs such that for any strongly acyclic bipartite graph , such that neither nor is in , there is a positive constant such that .
The set is given in Figure 1.
Note that the notion of large bicliques and co-bicliques in ordered bipartite graphs with forbidden induced subgraphs corresponds to the notion of submatrices of all [math]βs or of all βs in binary matrices with forbidden submatrices. A paper of KorΓ‘ndi, Pach, and Tomon [6] addresses a similar question for matrices. In addition, one could interpret bipartite graphs as set systems consisting of all the neighborhoods of vertices from one part. Structural properties of these graphs in terms of VC-dimension of the respective system and connection to ErdΕs-Hajnal conjecture are addressed for example by Fox, Pach, and Suk [5].
The paper is structured as follows. In Section 4 we show the general bounds for completeness. In Section 1 we characterize all strongly acyclic bipartite graphs. In Section 2 we find linear lower bounds on for each of the strongly acyclic graphs with few exceptions, thus proving Theorem 1. Finally, in Section 3 we determine the constant in Theorem 1 exactly for forbidden bipartite graphs with two vertices in each part.
1 Characterization of all strongly acyclic graphs
In this section we list all strongly acyclic bipartite graphs up to a bipartite complement.
Let and be the set of graphs shown in Figures 2 and 1. We denote a cycle of length by , a path on vertices by , and a complete bipartite graph with parts of sizes and , .
Theorem 2**.**
Let be a strongly acyclic bipartite graph. Then either one of its parts has size or one of its parts has size at most and at least one of or its bipartite complement is in . If is not a strongly acyclic bipartite graph, then or its bipartite complement contain , , or .
Proof.
Let be a bipartite graph with the top part and the bottom part , where . Denote by the bipartite complement of and a degree of a vertex in .
Assume that . Consider , and their neighborhoods. We see that these neighborhoods share at most one vertex, otherwise we have a cycle of length four. The same holds for the bipartite complement of . Thus the only possibilities for or are exactly graphs from as shown in Figure 2.
Now let .
Since and are acyclic, the number of edges in and is at most , i.e., the total number of edges in these two graphs is at most . On the other hand, this number is . We see however, that if , then . Similarly, if and , we have that . Thus, and .
Let and . If there is a vertex from of degree [math], say , then , otherwise there is a in . Moreover we must have . Thus . By considering , we can assume that no vertex in has degree . Thus all vertices of have degrees or and the number of edges is , , , or . Since is strongly acyclic, there could be at most and at least edges. So, up to bipartite complementation we can assume that there are edges in with respective degrees , , and in both parts. This is only possible when is a disjoint union of and , whose bipartite complement is .
Let and . Assume there is a vertex with . Then any two other vertices in have non-neighbors in each, and thus, at least two common non-neighbors, resulting in in . By considering , we see that there are no vertices of of degree [math]. I.e., the degrees of vertices from could be , or . Since is strongly acyclic, the number of edges is at most and at least . So, has edges, and degrees of vertices in are or . In case of degrees we see that the neighborhoods of degree and vertices intersect in exactly one vertex. The vertex of degree must be adjacent to a neighbor of degree vertex that is not adjacent to degree vertex, otherwise there is a in the bipartite complement of . Thus we have that . If the degrees of vertices in are , then the only option is .
We only need to show that any bipartite graph that is not strongly acyclic, contains , or in it or its bipartite complement. If has one part of size at most , we are done, since any cycle in or has length at most . If has both parts of sizes at least , one can easily verify that either or contains a . β
2 Proof of Theorem 1
Let , and . In the following results we observe that if is an induced bipartite subgraph of respecting sides, then . We omit ceilings and floors where it is not essential.
Lemma 1**.**
Let be a bipartite graph with parts and of sizes each and degrees of vertices from less that . Then .
Proof.
Let be a subset of vertices of . Then . Let , we have and form a co-biclique. β
2.1 Forbidden
Lemma 2**.**
Let . Then
Proof of Lemma 2.
Let , for . Let be an -free bipartite graph with top part and bottom part , both of size . We will show that . Let be an ordering of the vertices of , s.t. if . Since is isomorphic to its bipartite complement, we can assume that . Assume first that there is an with . Then we have a set of vertices, , and since , we have also a set of vertices , . But then and induce . Let . We have and by the above argument, we have , for all . Applying Lemma 1 to a subgraph of induced by and , we get . β
Remark 1**.**
In the case where , we can even show that
2.2 Forbidden or
In this section we need an auxiliary lemma about rooted trees. We call two vertex disjoint subforests of a rooted tree independent if no vertex in one forest is an ancestor of a vertex in the other forest. We say that a maximal path rooted at the root of a rooted tree with inner vertices of degree in is a handle of , denote its vertex set .
Lemma 3**.**
Any rooted tree on vertices has either height at least , or it contains two independent subforests on at least vertices each.
Proof.
We shall define vertex sets as follows. Let . Let be the vertex set of a largest component of , let be the set of vertices in all other components of . Assume that be defined, as well as and . Let be the tree induced by . Let be the set of vertices of a largest component of , let be the set of vertices in all other components of . We stop with being a star. Let be the set of leaves in . We have that is spanned by and a path built out of handles. See Figure 3 for an illustration. We see that span a forest with pairwise independent components in . Let , be the sizes of components in this forest. We would like to group these components into two balanced parts. Recall that for each . If , we are done. Assume that . Thus . Consider a partition such that is as close to as possible. Let . If , we are done. If not, then , . Then in particular, for each , otherwise we would move this from to and create a more balanced partition. In addition, we have that consists of one element, say , otherwise we can again move an some from to . This, however, contradicts the fact that each . β
Lemma 4**.**
Let . Then and .
Proof.
Let , . Let have partite sets and and such that has no induced copy of with the smaller partite set in . Assume that .
Proof outline: We shall first delete a few vertices of small degrees (at most ) and some other sets of vertices so that the remaining ones belong to blobs such that any two vertices in the same blob have degrees different by at most and any two vertices from different blobs have degrees different by at least . We shall call the resulting graph and its parts and . We introduce an auxiliary graph on whose edges correspond to two vertices with intersecting neighborhoods in and show that this auxiliary graph has a very special structure, i.e., formed of vertex-disjoint cliques that are pairwise either completely adjacent or completely disjoint. This gives rise to the second auxiliary graph for which we show that it is a tree closure of some tree.
Let be the set of vertices of from of degree at most . Assume , let , . Then if , . Thus form a co-biclique with parts of sizes at least . This contradicts our assumption that . Thus .
Let , where is the set of vertices of of degree at most . Let . Let , where . Since degree of vertices in are at least , and are empty. Let , . Assume without loss of generality that . Consider . Let . Then
Introduce an *auxiliary graph * with vertex set and two vertices adjacent iff their neighborhoods in intersect. We shall show that each component of is a closure of a rooted tree or just a tree closure, i.e., a graph obtained from a rooted tree by adding, for each vertex , all edges between and each of its ancestors. Here an ancestor is a vertex on a path from to the root.
Claim 0. Let , . Then .
Assume otherwise, then w.l.o.g. . Then , implying , that in turn implies that . Since , . Thus , a contradiction.
Claim 1. For any , is a pairwise vertex disjoint union of cliques. Call the family of these cliques , refer to this family as an blob.
Assume that is not a disjoint union of cliques. Then there are three vertices such that . We have that the degrees of , , and differ by at most . Thus . On the other hand . From Claim 0 we have . This implies that , i.e., , a contradiction.
The following two claims also follow from Claim 0 similarly. We use the fact that by definition, the degrees of vertices in from two distinct blobs differ by at least .
Claim 2. For any and , , the bipartite graph with parts is either complete or empty.
If not, there are , and such that , where either ( and ) or ( and ). Then we see that , and . Thus , a contradiction to Claim 0.
Let be a graph with vertex set with two cliques adjacent iff there is an edge between them in .
Claim 3. If , and , and , then and .
Assume , , . Then by Claim 0 . Thus and . Thus and .
Claim 4. Each induced connected subgraph of contains a vertex adjacent to all other vertices in .
We shall prove this by induction on the order of with a trivial basis of a one-vertex graph. Assume that the statement is true for all connected subgraphs with order less than . Let be a vertex of that belongs to the highest indexed blob. Let be a component of . Let be a vertex of that belongs to the highest indexed blob. By induction is adjacent to all other vertices of . Since is connected, is adjacent to some vertex of . If is not adjacent to , then either is disconnected (in case ) or the vertices contradict Claim 3. So, . If is not adjacent to some vertex of , then again contradict Claim 3. Thus is adjacent to all vertices of . Since this holds for each component of , we see that is adjacent to all other vertices of . Claim 4. implies the following.
Claim 5. Each component of and thus of is a tree closure for some tree.
Claim 6. If a component of has vertex set and an underlying tree with two independent subforests on vertex sets of sizes at least each, then in there is a co-biclique with parts in of sizes at least respectively.
Since and induce independent forests in an underlying tree of a component of , we have that . Thus, without loss of generality, . Then and form a desired co-biclique.
Claim 7. Let be a subset of vertices from that induces pairwise disjoint union of cliques in . If for a constant , then .
Assume first that there is a set that induces a clique in of size at least . We can assume that if then , otherwise induce a co-biclique with parts of sizes and , that is at least for . Thus is a bipartite graph with parts of sizes at least , and this graph has no induced , otherwise these stars and a common neighbor of their centers induce . Thus by Lemma 2, .
Now, we are left with the case that all cliques in have size at most . Split the cliques of in two groups of total size at least , let the vertex sets of these groups be and . Let w.l.o.g. in . Then induces a co-biclique with parts at least .
*Final argument: * Consider a component of on a vertex set . By Lemma 3, it either has a clique of size (when the underlying tree has respective height) or gives a co-biclique in with parts of sizes and .
Case 1. At least a -fraction of the vertices in is spanned by the components of with large cliques, i.e., giving a subset of vertices that is a union of cliques in . Thus by Claim 7.
[TABLE]
Case 2. At least a -fraction of the vertices of is spanned by the components of that give large co-bicliques. Let be the family of vertex sets of respective components. So, for each , we have a subset of , that forms a co-clique with , . Then
[TABLE]
form partite sets of a co-biclique. Note that , . Thus
[TABLE]
We have that the lower bounds in Case 1 and Case 2 are larger than for . For , these bounds are at least , however we can give an explicit argument for a bound of for . Note that one can definitely improve on the bound . This concludes the proof of the Lemma for .
Now, consider an -free bipartite graph with top part and bottom part , both of size . If there is a vertex of degree , then the graph is -free. Thus, by the previous result on , we have . If does not contain a vertex of degree at most , consider the bipartite complement of . Since is (bipartite) self-complementary, does not contain either, but we have a vertex with and thus, we can apply the same argument. β
Proof of Theorem 1.
Consider a strongly acyclic graph such that neither nor is in . Then by Theorem 2 has one part of size or . If one part is of size , i.e., this part consists of a vertex that has neighbors and non-neighbors for some and . Thus if is -free, each vertex in the top part has either degree at most or degree at most in a bipartite complement. Assume without loss of generality that at least half of the vertices in the top part have degree at most . Applying Lemma 1 to these vertices in the top part and the bottom part, we see that has a bi-coclique with parts of sizes at least . If has one part of size and another part of size at least , then by Theorem 2, is , or , with , or their bipartite complement. Then the result follows from Lemmas 2 and 4. β
3 Tight bounds for all strongly acyclic graphs with two vertices in each part
We consider strongly acyclic bipartite graphs with each part of size . These are exactly , , , and , where is such a graph with exactly two adjacent edges, has two disjoint edges. We shall give bounds for for each of these graphs. Recall that , where is a bipartite complement of .
Lemma 5**.**
Let be a bipartite -free graph with vertices in each part. Then . This bound is tight for .
Proof.
Let have partite sets and . It is easy to see that is a pairwise vertex disjoint union of bicliques. Let, for some index set , these bicliques have partite sets and of sizes , , respectively, , . Observe first that , for each , otherwise biclique gives us . Moreover and , for each since otherwise a cobiclique with parts or has parts of sizes greater than .
Let be the set of indices, so that , . Let . Let , , , , , , , . Consider a cobiclique with parts . We can assume that either or is less than , say . Then since for each , . Thus .
Consider a minimal subset such that has size . Then otherwise for any , . In particular, we could have taken instead of , contradicting its minimality. Thus has size less than . This implies that and form a co-biclique with each part of size at least .
Note, that the bound shown is best possible, by the following -free construction for every natural with . Take to be the disjoint union of complete bipartite graphs . Clearly this is -free, and . β
Lemma 6**.**
Let be a -free bipartite graph with vertices in each part. Then . This bound is tight.
Proof.
Let be bipartite -free with parts of size each. Then we have for any vertices that or . Thus there is a total ordering of the vertices in where if and only if . Consider the two subgraphs , , with
By our vertex ordering, we have that , , and thus, is a biclique. On the other hand, , , and thus, is a co-biclique. We know that . Since , one of them has to have size at least , which gives us .
Note, that is best possible. Consider a bipartite graph that is a union of a complete bipartite graph and add isolated vertices added to the smaller part. Both parts have size , we have and is -free. β
Lemma 7**.**
Let be an -free bipartite graph with vertices in each part. Then . This bound is tight for .
Proof.
Let have top part and bottom part of size each, assume is divisible by . Denote by the bipartite complement of . First observe, that , for any .
Claim: There is a set such that for a graph that is either or its bipartite complement, for all we have for some . I.e., the neighborhoods of the vertices from form a sunflower set system with petals of sizes at most one.
To prove the Claim, consider a vertex of largest degree. If for each , let . Then satisfies the conditions of the Claim with . So, we assume that there is a vertex for some vertex . In particular . Let , . We see that for each , either or .
If for all vertices , , then satisfies the conditions of the Claim with . If for all , , then the Claim is satisfied with and . If there are such that , and , we see that form a copy of , where is the symmetric difference of and . This proves the Claim.
Now that we proved the Claim, it remains to find a large biclique or (co-)biclique. Assume without loss of generality that in the Claim. If , then induces a biclique with parts of sizes at least . Assume that . Let . We see that induces a pairwise disjoint union of stars in with centers in . Let , note that . Let such that , . Let . If , then induces a co-biclique with parts of sizes at least . If , then for all . Thus , a contradiction since .
To show that the bound is tight, construct the following graph with parts and of sizes , . Let where and . Let , where and . Let have all edges between and , form a perfect matching between and , and form a perfect matching between and . Note that if has a copy of , this copy has a vertex of degree in . Thus, this copy must have a neighbor of in , that in turn is adjacent to all of and thus could not have degree in a copy of . Thus is -free. In addition, we see that . β
4 General bounds
In this section we work out known arguments for completeness.
Theorem 3**.**
Let be a bipartite graph that is not strongly acyclic. Then there is an such that for each sufficiently large , . Moreover, if or its bipartite complement contains , , or , then could be taken any positive real strictly less than or , respectively.
Proof.
First, recall from Theorem 2, that if is not strongly acyclic, then or its bipartite complement contains , or . In case of , we know by a result of Caro and Rousseau [1] that there is a bipartite graph with parts of size each that does not contain and such that . This result was shown using LovΓ‘sz Local Lemma that we abbreviate as LLL. The LLL tells us that if there are bad events and positive numbers associated with these events such that , where iff is adjacent to in the dependency graph, then with positive probability none of the bad events happen.
We use the same approach to randomly create respective graphs with no and with no and not having large co-cliques. Consider and color each edge red with probability and blue with probability . For a specific , we say that there is a red bad event if this is red. Similarly, for a specific , we say that there is a blue bad event if this is blue. We shall use LLL to prove that with positive probability there are no bad events. We say A βdependsβ on B when an event A is adjacent to B in the dependency graph.
The probability of a red bad event is . The probability of a blue bad event is .
- -
Each red bad event depends on at most other red bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each red bad event depends on at most blue bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each blue bad event depends on red bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each blue bad event depends on blue bad events (this corresponds to the number of βs sharing an edge with a given ).
Since here we have bad events of two types, let for red bad events and for blue bad events. We shall assign the values to and such that
[TABLE]
Let be small positive constants, say , Let
[TABLE]
We shall be using the fact that for small . Then, for large we have
[TABLE]
Thus Therefore, by the LLL there is an edge-coloring of with no red βs and no blue .
To see the result for , we closely follow the above argument and choose the parameters similarly and define red bad event corresponding to a red and blue bad event as before.
The probability of a red bad event is . The probability of a blue bad event is .
- -
Each red bad event depends on at most other red bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each red bad event depends on at most blue bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each blue bad event depends on red bad events (this corresponds to the number of βs sharing an edge with a given ).
- -
Each blue bad event depends on blue bad events (this corresponds to the number of βs sharing an edge with a given ).
Let be a small positive constants, say , . Let
[TABLE]
Then, for large we have
[TABLE]
Thus Therefore, by the LLL there is an edge-coloring of with no red βs and no blue .
Now, let be a bipartite graph containing , . Consider an edge-coloring of with no red and no blue . Let be a graph formed by the red edges. Then does not contain and thus does not contain , that implies that it does not contain an induced copy of . In particular does not have . On the other hand the bipartite complement of does not contain . Thus, for sufficiently large , . β
Theorem 4**.**
[4*]** Let be a bipartite graph with parts of sizes and , . Let be a bipartite graph with parts of sizes , . Then either is -free or , where . *
We need the following Lemma.
Lemma 8**.**
Let be an -partite graph with vertex classes , , , for some integers . Let and denote and of the bipartite subgraph of induced by . Let for all . Then for any map , there exists a vertex , such that
[TABLE]
Proof.
Let be an -partite graph as in the statement and fix a function . We prove the statement by contradiction.
Assume there is no such vertex . Then for every , there must be at least one index , such that is bad for , meaning that
[TABLE]
Since there are only sets, we have a set , that is bad for vertices in . Choose such that and .
We now distinguish two cases:
Case 1: . Consider the subset of vertices, that are adjacent to all vertices in , so . Since every vertex in is non-adjacent to at most vertices, we obtain . Thus, the pair contains a copy of , which contradicts our assumption of .
Case 2: . Consider the subset of vertices, that have no neighbour in , so . Since every vertex in is adjacent to at most vertices, we obtain . Thus, the pair contains an empty bipartite graph of size , which contradicts our assumption of .
Hence, for every we find a vertex , which is good for all sets . β
Proof of Theorem 4.
Let be a bipartite graph with parts , with , let . Assume that and . We show how to find an induced copy of . Note that from the choice of and , we have that and .
Partition into subsets , each of size at least . Partition into subsets each of size at least . Since , we have that , for all . We shall apply Lemma 8 times to obtain subsets such that we can embed and for and .
In step 1, we apply Lemma 8 to the sets with and
[TABLE]
to find a vertex and subsets
[TABLE]
such that for .
Assume that after step we have subsets , .
If , we apply the Lemma again, to the sets with and
[TABLE]
to find a vertex and subsets
[TABLE]
such that for .
We finish after steps, and by our choice of , we still obtain , . Thus, we have found vertices where we can embed and nonempty sets of candidates , in which we can embed . This concludes the proof. β
Acknowledgments The authors thank Jacob Fox for bringing the connections between the considered problem and the VC-dimensions of sets systems to their attention. The first author thanks the Department of Mathematics at Stanford University for the hospitality.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] Yair Caro and Cecil Rousseau. Asymptotic bounds for bipartite Ramsey numbers . The Electronic Journal of Combinatorics, 8(1):17, 2001.
- 2[2] Maria Chudnovsky. The ErdΕs-Hajnal conjecture β a survey . Journal of Graph Theory 75 (2014) 178β190.
- 3[3] Paul ErdΕs and AndrΓ‘s Hajnal. Ramsey-type theorems . Discrete Applied Mathematics 25(1989) 37β52.
- 4[4] Paul ErdΕs, AndrΓ‘s Hajnal, and JΓ‘nos Pach. A Ramsey-type theorem for bipartite graphs .Geombinatorics, 10(DCG-ARTICLE-2000-001) (2000) 64β68.
- 5[5] Jacob Fox, JΓ‘nos Pach, and Andrew Suk. ErdΕs-Hajnal conjecture for graphs with bounded VC-dimension. 33rd International Symposium on Computational Geometry, Art. No. 43, 15 pp., LIP Ics. Leibniz Int. Proc. Inform., 77, Schloss Dagstuhl. Leibniz-Zent. Inform., Wadern, 2017.
- 6[6] Daniel KorΓ‘ndi, JΓ‘nos Pach, and IstvΓ‘n Tomon. Large homogeneous submatrices . ar Xiv:1903.06608
