Structural Properties of Bipartite Subgraphs Arxiv:2106.12099V2 [Math.CO] 29 Jun 2021

Structural Properties of Bipartite Subgraphs

Robert Hickingbotham,∗ David R. Wood†

June 30, 2021

Abstract

This paper establishes suﬃcient conditions that force a graph to contain a bipartite subgraph with a given structural property. In particular, let β be any of the following graph parameters: Hadwiger number, Hajós number, treewidth, pathwidth, and treedepth. In each case, we show that there exists a function f such that every graph G with β(G) > f(k) contains a bipartite ˆ ˆ subgraph G ⊆ G with β(G) > k.

1 Introduction

A classical result by Erdős [18] states that every graph contains a bipartite subgraph with at least half of the edges and that this is best possible for complete graphs. Many papers study the maximum number of edges in a bipartite subgraph if the graph satisfies some structural property such as being triangle-free (see [1–3, 17, 19, 56]). Instead of maximising the number of edges, we consider other structural properties of bipartite subgraphs. In general, we are interested in sufficient conditions that force a graph to have a bipartite subgraph with a given property. Typically, we arXiv:2106.12099v2 [math.CO] 29 Jun 2021 study this question for a graph parameter β.1 In such cases, the first question is whether there exists a function f such that every graph G with β(G) > f(k) ∗School of Mathematics, Monash University, Melbourne, Australia ([email protected]). Research supported by an Australian Government Research Training Program Scholarship. †School of Mathematics, Monash University, Melbourne, Australia ([email protected]). Research supported by the Australian Research Council. 1 A graph parameter is a function β such that β(G) ∈ R for every graph G and β(G1) = β(G2) for all isomorphic graphs G1 and G2.

1 contains a bipartite subgraph Gˆ ⊆ G with β(Gˆ) > k. We show that such a function exists for each of the following parameters: Hadwiger number (Section3), Hajós number (Section3), treewidth (Section4), pathwidth (Section 4.1), and treedepth (Section 4.1). These parameters are important in structural graph theory, especially in Robertson-Seymour graph minor theory. The second question for such parameters is what is the least possible function f. We discuss this question for each of these parameters in the relevant section. One motivation for exploring the properties of bipartite subgraphs is their application in the study of direct products.2 For example, a graph G is not necessarily a minor

of G × K2 [8], but it is easy to see that every bipartite subgraph of G is a subgraph

of G × K2. So in studying the structural properties of G × K2, it is very helpful to know about the behaviour of bipartite subgraphs in G. We explore this line of research in our companion papers [27, 28].

1.1 Relevant Literature

For integers a, b > 1 where a 6 b, deﬁne [b] := {1, . . . , b} and [a, b] := {a, a + 1, . . . , b − 1, b}. We now mention several results from the literature concerning structural properties of bipartite subgraphs. For minimum degree, it is folklore that every graph with minimum degree at least 2d contains a bipartite subgraph with minimum degree at least d (see [54] for a proof). For connectivity, Mader [45] showed that every graph with suﬃciently large average degree contains a highly connected subgraph. In particular, for every integer k > 1, every graph with average degree at least 4k contains a (k + 1)-connected subgraph. Now if a graph is 8k-connected, then its average degree is at least 8k. By Erdős’ result [18], the graph contains a bipartite subgraph with average degree at least 4k. Hence, Mader’s and Erdős’ results together imply the following.

Theorem 1. For every integer k > 1, every graph with average degree at least 8k (in particular, every 8k-connected graph) contains a (k + 1)-connected bipartite subgraph.

See [4, 62] for improved bounds to Mader’s result.

2 Constructing Bipartite Subgraphs

Suppose we have a proper black-white colouring of a graph. To switch the colouring means to recolour the black vertices white and the white vertices black. For a graph

2 For graphs G1 and G2, the direct product, G1 × G2, is deﬁned with vertex set V (G1) × V (G2) where (u1, u2)(v1, v2) ∈ E(G1 × G2) if u1v1 ∈ E(G1) and u2v2 ∈ E(G2).

2 G with vertex-disjoint subgraphs H1 and H2, a path P = (v1, . . . , vn) in G joins H1 and H2 if V (P ) ∩ V (H1) = {v1} and V (P ) ∩ V (H2) = {vn}. The next lemma allows us to construct bigger bipartite subgraphs from smaller ones.

Lemma 2. Let G be a graph and let H1,...,Ht be vertex-disjoint bipartite subgraphs in G. Suppose there exists a set of internally disjoint paths P = {P1,...,Pk} in G such that for all i ∈ [k], Pi joins Hj and H` for some distinct j, ` ∈ [t] and that Pi and Hb are disjoint for all b ∈ [t] \{j, `}. Then there exists a subset Q ⊆ P where ˆ S S | Q | > k/2 such that the graph G := ( i∈[t] Hi) ∪ ( Q∈Q Q) is a bipartite subgraph of G.

(j) Proof. For all j ∈ [t], let P be the set of paths in P that join any Hi and Hi0 where i, i0 ∈ [j]. Note that ∅ = P(1) ⊆ P(2) ⊆ ... ⊆ P(t) = P. We prove the following by induction on j. (j) (j) (j) (j) Claim: For all j ∈ [t], there exists a subset Q ⊆ P where | Q | > | P |/2 ˆ(j) S S such that the graph G := ( i∈[j] Hi) ∪ ( Q∈Q(j) Q) is a bipartite subgraph of G. For j = 1, the claim holds trivially since P(1) = ∅. Now assume it holds for j − 1. Then there exists a set of paths Q(j−1) with size at least | P(j−1) |/2 such that Gˆ(j−1) (j−1) is bipartite. Since Gˆ and Hj are vertex-disjoint bipartite subgraphs, we may take a proper black-white colouring of their vertices. Let A(j) := P(j) \P(j−1). Then (j) each path in A joins Hj to some Hi where i ∈ [j − 1]. We say that a path (j) P = (v1, . . . , vn) ∈ A is agreeable if there exists a proper black-white colouring of the vertices of P such that the colouring of v1 and vn corresponds with the colour they were previously assigned. Otherwise we say that P is disagreeable. Observe that if we switch the black-white colouring for the vertices in Hj, then all the agreeable paths in A(j) become disagreeable and all the disagreeable paths become agreeable. So if there are more disagreeable paths than agreeable, then switch the black-white (j) (j) colouring of the vertices in Hj. In doing so, the set of agreeable paths B ⊆ A (j) (j) (j−1) S (j) has size at least |A |/2. Let Q := Q ∪( P ∈B(j) P ). Then Gˆ is bipartite and (j) (j−1) (j) (j−1) (j) (j) | Q | = | Q ∪B | > | P |/2 + |A |/2 = | P |/2 as required. The result follows when j = t.

Note that Lemma2 generalises Erdős’ result [ 18]: if H1,...,Ht are the vertices of a graph G and P is the set of the edges in G, then by Lemma2, G contains a bipartite subgraph with at least half the edges.

3 Minors

A graph H is a minor of a graph G if a graph isomorphic to H can be obtained from G by vertex deletion, edge deletion, and edge contraction. The study of graph

3 minors within the broader field of structural graph theory is a fundamental topic of research that was initiated by the seminal work of Robertson and Seymour. We now prove a sufficient condition under which a graph has a bipartite subgraph that contains a given graph as a minor. For a graph G, let ∇(G) be the maximum average degree of a minor of G. For a graph H, let f(H) be the infimum of all real numbers d such that every graph with average degree at least d contains H as a minor. Mader [44] showed that f(H) exists for every graph H. The function f(H) is called the extremal function for H minors and has been studied for many graphs (see [10, 14, 16, 23, 24, 26, 29, 34–38, 44, 46, 49, 55, 57–59, 61]).

Theorem 3. For all graphs G and H, if ∇(G) > 2f(H) then G contains a bipartite subgraph Gˆ ⊆ G where H is a minor of Gˆ.

Proof. Let H1 be a minor of G with average degree at least 2f(H) and let G1 be

a minimal subgraph of G that contains H1 as a minor. By minimality, G1 consists

of pairwise disjoint subtrees T1,...,Tt of G such that for each ij ∈ E(H1), there

exists an edge vivj ∈ E(G1) where vi ∈ V (Ti) and vj ∈ V (Tj). Let P := {vivj : ij ∈

E(H1)}. Since T1,...,Tt are bipartite, by Lemma2 there exists a subset Q ⊆ P ˆ S S where | Q | > | P |/2 such that the subgraph G := ( i∈[t] Ti) ∪ ( Q∈Q Q) is bipartite.

Let H2 be the minor of Gˆ obtained by contracting each Ti into a vertex. Then |V (H2)| = |V (H1)| and |E(H2)| = | Q | > |E(H1)|/2. As such, the average degree of H2 is at least f(H). By the deﬁnition of the extremal function, H is a minor of H2 and hence H is a minor of Gˆ.

A graph H is a topological-minor of a graph G if a subgraph of G is isomorphic to a subdivision of H. For a graph G, let ∇˜ (G) be the greatest average degree of a topological-minor of G. For a graph H, let f˜(H) be the inﬁmum of all real numbers d such that every graph with average degree at least d contains H as a topological-minor. Mader [43] proved that f˜(H) exists for every graph H. The function f˜(H) is called the extremal function for topological H minors (see [7, 20, 30, 32, 33, 39–42, 44]). Using a similar approach to our proof of Theorem3, we prove an analogous result for topological-minors.

Theorem 4. For all graphs G and H, if ∇˜ (G) > 2f˜(H) then G contains a bipartite subgraph Gˆ ⊆ G where H is a topological-minor of Gˆ.

Proof. Let H1 be an n-vertex graph that is a topological-minor of G with average

degree at least 2f˜(H). Let G1 be a subgraph of G that is isomorphic to a subdivision

of H1. Then G1 consists of a set of n vertices X = {x1, . . . , xn} such that for each

ij ∈ E(H1), there exists an (xi, xj)-path Pi,j in G1 where V (Pi,j) ∩ X = {xi, xj} and

these paths are internally disjoint. Let P := {Pi,j : ij ∈ E(H1)}. By Lemma2, there

4 ˆ S exists Q ⊆ P where | Q | > | P |/2 such that the subgraph G := X ∪ ( Q∈Q Q) is bipartite.

Let H2 be the topological-minor of Gˆ that is obtained by contracting each Pi,j ∈ Q into an edge. Then |V (H2)| = |V (H1)| and |E(H2)| = | Q | > |E(H1)|/2. As such, the average degree of H2 is at least f˜(H). By the deﬁnition of f˜(H), H2 contains H as a topological-minor and hence H is a topological-minor of Gˆ.

We now derive results concerning complete-minors and complete-topological-minors in bipartite subgraphs. For a graph G, the Hadwiger number, η(G), of G is the

maximum integer t such that Kt is a minor of G. Similarly, the Hajós number, η˜(G),

of G is the maximum integer t such that Kt is a topological-minor of G. Kostochka q [34, 35] and Thomason [58] independently proved that f(Kt) ∈ Θ(t log(t)). q Theorem 5 ([34, 35, 58]). For every integer t > 1, we have f(Kt) 6 (c+o(1))t log(t) for some positive constant c.

Note that Thomason [59] subsequently determined the best possible c. Theorems3 and5 immediately imply the following. q Theorem 6. For every integer t > 1 and graph G, if η(G) > (2c + o(1))t log(t) then G contains a bipartite subgraph Gˆ ⊆ G such that η(Gˆ) > t where c is the constant from Theorem5.

For the extremal function for complete-topological-minors, Bollobás and Thomason 2 [7] and Komlós and Szemerédi [33] independently proved that f˜(Kt) = Θ(t ).

˜ 2 Theorem 7 ([7, 33]). For every integer t > 1, we have f(Kt) 6 (c + o(1))t for some positive constant c.

See [41] for the best known bounds on c. Theorems4 and7 imply the following.

2 Theorem 8. For every integer t > 1 and graph G, if η˜(G) > (2c + o(1))t , then G contains a bipartite subgraph Gˆ ⊆ G with η˜(Gˆ) > t where c is the constant from Theorem7.

It is open whether the bounds from Theorems6 and8 are asymptotically best possible.

We now consider shallow minors. For an integer r > 0 and graph G, let ∇r(G) be the greatest average degree of a minor of G that can be obtained by contracting pairwise disjoint subgraphs of radius at most r in G into vertices and then taking a

subgraph. The function ∇r is of particular importance in graph sparsity theory (see [47]).

5 Theorem 9. For every integer r > 0, every graph G contains a bipartite subgraph ˆ ˆ 1 Gr ⊆ G such that ∇r(Gr) > 2 ∇r(G).

Proof. Let H be a graph with average degree ∇r(G) that can be obtained from G by contracting pairwise disjoint subgraphs of radius at most r in G into vertices and then taking a subgraph. Let G1 be a minimal subgraph of G from which H can be obtained by contracting subgraphs in G1 of radius at most r into vertices. By minimality, G1 consists of pairwise disjoint trees T1,...,Tt each with radius at most r such that for each ij ∈ E(H), there exists an edge vivj ∈ E(G1) where vi ∈ V (Ti) and vj ∈ V (Tj).

Let P := {vivj : ij ∈ E(H)}. Since T1,...,Tt are bipartite, by Lemma2 there exists ˆ S S Q ⊆ P where | Q | > | P |/2 such that the subgraph Gr := ( i∈[t] Ti) ∪ ( Q∈Q Q) is bipartite.

For each i ∈ V (H), contract the Ti subgraph in Gˆr into a vertex and let Hˆ be the graph obtained. Now |V (Hˆ )| = |V (H)| and |E(Hˆ )| = | Q | > |E(H)|/2. As ˆ 1 ˆ ˆ such, the average degree of H is at least 2 ∇r(G) and since H is obtained from Gr by contracting pairwise disjoint subgraphs with radius at most r, it follows that ˆ 1 ∇r(Gr) > 2 ∇r(G).

It is open whether there exists a function g such that every graph G contains a ˆ bipartite subgraph G ⊆ G where for every integer r > 0, if ∇r(G) > g(k) then ˆ ∇r(G) > k.

4 Treewidth

For a graph G, a tree-decomposition of G, (T, {Bt ⊆ V (G)}t∈V (T )), is a pair that satisﬁes the following properties: • T is a tree;

• For all v ∈ V (G), T [{t ∈ V (T ): v ∈ Bt}] is a non-empty subtree of T ; and

• For all uv ∈ E(G), there is a node t ∈ V (T ) such that u, v ∈ Bt.

The width of a tree-decomposition is max{|Bt| − 1 : t ∈ V (T )}. The treewidth of G, tw(G), is the minimum width of a tree-decomposition of G.A path-decomposition of G is a tree-decomposition of G where the tree is a path. The pathwidth of G, pw(G), is the minimum width of a path-decomposition of G. Treewidth and pathwidth respectively measure how similar a graph is to a tree and to a path, and are important parameters in algorithmic and structural graph theory (see [6, 25, 48]). Tree-decompositions and path-decompositions were introduced by Robertson and Seymour [50, 51]. The following theorem is our key contribution in this section.

6 Theorem 10. There exists a function f such that every graph G with tw(G) > f(k) contains a bipartite subgraph Gˆ ⊆ G such that tw(Gˆ) > k.

Before proving Theorem 10, we need to introduce two families of graphs that are relevant to the proof. The ﬁrst are grids. For integers j, k > 1, the (j × k)-grid, Gj×k, is the graph with vertex set V (Gj×k) = {(v1, v2): v1 ∈ [j], v2 ∈ [k]} where

(v1, v2)(u1, u2) ∈ E(Gj×k) if v1 = u1 and |v2 − u2| = 1, or v2 = u2 and |v1 − u1| = 1.

The edge is vertical when v1 = u1, and is horizontal when v2 = u2. For i ∈ [k], the ith-horizontal path in the grid is the unique path from (1, i) to (j, i) where each edge in the path is horizontal. ¯(i) (i) (i) The second family are walls. For integers j, k > 1 and i ∈ [k], let P = (v1 , . . . , v2j ) (1) (k) be a path on 2j vertices. The (2j ×k)-wall, W2j×k, consists of the paths P¯ ,..., P¯ (i−1) (i) where for each ` ∈ [j] and i ∈ [2, k], we add the edge v2` v2` whenever i is even and (i−1) (i) the edge v2`−1 v2`−1 whenever i is odd (see Figure1). For every i ∈ [k] and ` ∈ [2j], ¯(i) (i−1) (i) we call P a horizontal path in the wall and v` v` a vertical edge (if it exists). Observe that the (2j × k)-wall is a spanning subgraph of the (2j × k)-grid where for (i) th each i ∈ [k], P¯ is the i -horizontal path in G2j×k and every vertical edge in W2j×k

corresponds to a vertical edge in G2j×k.

Figure 1: The (8 × 6)-wall, W8×6.

Robertson and Seymour [52] showed that grids are a canonical example of graphs with large treewidth.

Theorem 11 ([52]). There exists a function f such that every graph with treewidth

at least f(k) contains Gk×k as a minor. Moreover, tw(Gk×k) = k.

In a recent breakthrough, Chekuri and Chuzhoy [9] showed that the function f in Theorem 11 is polynomial (see [13] for the current best upper bound as well as [11, 12]).

Now Gk×k is a minor of W2k×2k and hence tw(W2k×2k) > k. Furthermore, since walls have maximum degree 3, W2k×2k is a minor of a graph G if and only if W2k×2k is a topological-minor of G. Hence, Theorem 11 implies the following.

7 Corollary 12. There exists a function f such that every graph with treewidth at least f(k) contains W2k×2k as a topological-minor. Moreover, tw(W2k×2k) > k.

We now show that every subdivision of a suﬃciently large wall contains a bipartite subgraph that is a subdivision of a large wall. This lemma is the heart of the proof for Theorem 10.

k+1 Lemma 13. Every graph G that is a subdivision of the (k2 × k)-wall, Wk2k+1×k, contains a bipartite subgraph Gˆ ⊆ G which is a subdivision of the (2k × k)-wall.

Proof. Let P (1),...,P (k) be the horizontal paths in G where each P (i) is a subdivision ¯(i) (i) (i) k of the path P = (v1 , . . . , vk2k+1 ) in Wk2k+1×k. For even i ∈ [2, k] and ` ∈ [k2 ], let (i) (i−1) (i) S` be the path in G that corresponds to the subdivided vertical edge v2` v2` in (i) Wk2k+1×k and for odd i ∈ [2, k], let S` be the path in G that corresponds to the (i−1) (i) (1) subdivided edge v2`−1 v2`−1 in Wk2k+1×k. To simplify our notation, let S` be an empty subgraph of G for all ` ∈ [k2k].

k k−i Initialise A1 := [k2 ]. We construct A1 ⊇ A2 ⊇ · · · ⊇ Ak where |Ai| > k2 for (j) S (i) S (i) all i ∈ [k] such that H := i∈[j](P ∪ ( `∈Aj S` )) is a bipartite subgraph of G for all j ∈ [k]. We proceed by induction on j. For j = 1, H(1) = P (1) and hence it is a bipartite subgraph of G. Now suppose that j > 1. By induction, H(j−1) is (j) (j) 0 (j) bipartite. Let S := {S` : ` ∈ Aj−1}. For distinct S,S ∈ S , the path S joins H(j−1) to P (j), and the paths S and S0 are internally disjoint. Since P (j) and H(j−1) are vertex-disjoint bipartite subgraphs of G, by Lemma2 there exists Q(j) ⊆ S(j) (j) (j) ˆ (j) (j−1) (j) S where | Q | > |S |/2 such that the graph H := H ∪ P ∪ ( S∈Q(j) S) is (j) (j) bipartite (see Figure2). By setting Aj := {` ∈ Aj−1 : S` ∈ Q }, it follows that (j) k−j (j) |Aj| = | Q | > |Aj−1|/2 > k2 and the subgraph H is bipartite. (k) Now consider H . We have |Ak| > k. Let `1, . . . , `k ∈ Ak such that ì < ì0 0 ˆ(i) (i) (i) whenever i < i . For each i ∈ [k], let P be the subpath of P from v2`1−1 to v(i) Gˆ S Pˆ(i) ∪ S S(i) Gˆ 2`k . Let := i∈[k]( ( j∈[k] `j )). Then is bipartite as it is a subgraph of H(k). We now explain how Gˆ is a subdivision of the (2k × k)-wall. For each of the following pairs of vertices in Gˆ, there is a path between them which consists v(i) , v(i) i, j ∈ k v(i) , v(j) solely of horizontal edges: ( 2`j −1 2`j ) where [ ]; and ( 2`j 2`j+1−1) where i ∈ k , j ∈ k − S(i) [ ] [ 1]. The union of these horizontal paths with the `j paths where i, j ∈ [k] defines a subdivision of the (2k × k)-wall, as required.

Theorem 10 directly follows from Corollary 12 and Lemma 13. Note that due to Lemma 13, the function f in Theorem 10 is exponential. It is open whether this can be improved to a polynomial function. The key challenge is whether every subdivision of a wall contains a bipartite subgraph with polynomially large treewidth.

8 Figure 2: The inductive proof of Lemma 13: a) the set S(j) of disjoint paths that join H(j−1) to P (j); b) the bipartite subgraph Hˆ (j); c) the bipartite subgraph H(j).

4.1 Pathwidth and Treedepth

We now move on to consider pathwidth and treedepth. The closure of a rooted tree T is the graph obtained from T by adding the edges uv whenever u is an ancestor of v. The closure of a rooted forest is the closure of each of the rooted trees in the forest. The height of a rooted tree is the number of vertices in the longest root to leaf path. The height of a rooted forest is the maximum height of its component subtrees. For a graph G, the treedepth, td(G), of G is the minimum height of a rooted forest whose closure contains G as a subgraph. Like treewidth and pathwidth, treedepth is an important parameter in structural graph theory (see [47]). We now explain how results in the literature immediately imply that a graph with sufficiently large pathwidth contains a bipartite subgraph with large pathwidth and a graph with sufficiently large treedepth contains a bipartite subgraph with large treedepth. To a certain degree, these parameters are simpler to deal with since trees have arbitrarily large pathwidth and treedepth. So it suffices to show that a graph with large pathwidth contains a subtree with large pathwidth and that a graph with large treedepth contains a subtree with large treedepth. For pathwidth, the excluded forest theorem demonstrates this.

Theorem 14 ([5, 50]). For every forest F , every graph of pathwidth at least |V (F )|−1 contains F as a minor.

9 k The complete binary tree Tk with height k has 2 − 1 vertices and pathwidth dk/2e

[53]. Moreover, since Tk has maximum degree 3, every graph that contains Tk as a minor also contains Tk as a topological-minor. Hence Theorem 14 implies the following.

(2k+1) Corollary 15. For every integer k > 1, every graph G with pw(G) > 2 − 2 contains a bipartite subgraph Gˆ ⊆ G such that pw(Gˆ) > k.

For treedepth, Nešetřil and Ossona de Mendez [47] proved that every graph with treedepth at least n contains an n-vertex path, Pn, as a subgraph. They also showed that the treedepth of Pn is at least dlog2(n + 1)e. Hence their results imply the following.

n Corollary 16. For every integer n > 1, every graph G with td(G) > 2 − 1 contains a bipartite subgraph Gˆ ⊆ G such that td(Gˆ) > n.

Again, it is open whether the bounds for pathwidth and treedepth in Corollaries 15 and 16 can be improved to polynomial functions. We now explain how a polynomial bound for treewidth would imply polynomial bounds for both pathwidth and treedepth. We make use of the following two theorems from the literature. The ﬁrst is by Groenland, Joret, Nadara, and Walczak [22].

Theorem 17 ([22]). For all integers k, h > 1, every graph G with tw(G) > k − 1 has pw(G) 6 kh + 1 or contains a subdivision of a complete binary tree of height h.

For treedepth, Kawarabayashi and Rossman [31] ﬁrst demonstrated a polynomial excluded-minor approximation for graphs with large treedepth which was later improved by Czerwiński, Nadara, and Pilipczuk [15].

Theorem 18 ([15, 31]). There exists a positive constant c such that for every integer 3 k > 1 and graph G with td(G) > ck , either tw(G) > k, or G contains a subdivision of a complete binary tree of depth k as a subgraph, or G contains a path of length 2k.

Lemma 19. Suppose there exists a polynomial function f such that every graph G with tw(G) > f(k) contains a bipartite subgraph Gˆ ⊆ G with tw(Gˆ) > k. Then there exists polynomial functions g and h such that:

1. Every graph G with pw(G) > g(k) contains a bipartite subgraph Gˆ ⊆ G such that pw(Gˆ) > k; and 2. Every graph G with td(G) > h(k) contains a bipartite subgraph Gˆ ⊆ G such that td(Gˆ) > k.

Proof. Let g(k) := 2f(k)k + 1 and h(k) := cf(k)3 where c is from Theorem 18. Let G be a graph. If tw(G) > f(k) then there exists a bipartite subgraph Gˆ ⊆ G such

10 that tw(Gˆ) > k. Since treedepth is bounded from below by pathwidth and pathwidth is bounded from below by treewidth, we are done. Therefore, we may assume that tw(G) 6 f(k). Now if pw(G) > g(k), then by Theorem 17, G contains a subdivision of the complete binary tree T2k as a subgraph. As pw(T2k) = h, by setting Gˆ to ˆ ˆ be the subdivision of T2k in G, it follows that G is bipartite and pw(G) > k. Now if td(G) > h(k), then G either contains a subdivision of a complete binary tree of depth 2k or a path of length 2k. In either case, by setting Gˆ to be the subdivision of the complete binary tree of depth 2k or the path of length 2k, Gˆ is bipartite and we have td(Gˆ) > k.

5 Conclusion

We conclude with the following question. Say a graph parameter β is unbounded on bipartite graphs if for every integer k > 0, there exists a bipartite graph G such that β(G) > k. For every such parameter β, does there exists a function f such that every graph G with β(G) > f(k) contains a bipartite subgraph Gˆ ⊆ G with β(Gˆ) > k?

Note: After the ﬁrst release of this paper, Gwenaël Joret and Piotr Micek informed us that our main results were already known. In particular, Thomassen [60] proved a generalisation of Lemma 13 and Geelen, Gerards, Reed, Seymour, and Vetta [21] proved Theorem6. Furthermore, Alex Scott answered our concluding question in the negative with the following construction. For an odd integer n > 1, let Cn be a cycle on n vertices. For a non-empty graph G, let a(G) be the maximum odd integer n > 1 such that Cn is a subgraph of G and let b(G) be the maximum integer t > 0 such that G contains Kt,t as a subgraph. Deﬁne c(G) := max{a(G), b(G)}. Then c(G) is unbounded on bipartite graphs. However, c(Cn) = n for every odd integer ˆ ˆ n > 1 and c(G) 6 1 whenever G is a bipartite subgraph of Cn, as required. Many thanks to Gwenaël Joret, Piotr Micek and Alex Scott for this feedback.

References

[1] Noga Alon. Bipartite subgraphs. Combinatorica, 16(3):301–311, 1996. [2] Noga Alon, Béla Bollobás, Michael Krivelevich, and Benny Sudakov. Maximum cuts and judicious partitions in graphs without short cycles. J. Combin. Theory Ser. B, 88(2):329–346, 2003. [3] Noga Alon, Michael Krivelevich, and Benny Sudakov. MaxCut in H-free graphs. Combin. Probab. Comput., 14(5-6):629–647, 2005. [4] Anton Bernshteyn and Alexandr Kostochka. On the number of edges in a graph with no (k + 1)-connected subgraphs. Discrete Math., 339(2):682–688, 2016. [5] Dan Bienstock, Neil Robertson, Paul Seymour, and Robin Thomas. Quickly excluding a forest. J. Combin. Theory Ser. B, 52(2):274–283, 1991.

11 [6] Hans L. Bodlaender. A tourist guide through treewidth. Acta Cybernet., 11(1-2):1–21, 1993. [7] Béla Bollobás and Andrew Thomason. Proof of a conjecture of Mader, Erdős and Hajnal on topological complete subgraphs. European J. Combin., 19(8):883–887, 1998. [8] Anne Bottreau and Yves Métivier. Some remarks on the Kronecker product of graphs. Inform. Process. Lett., 68(2):55–61, 1998. [9] Chandra Chekuri and Julia Chuzhoy. Polynomial bounds for the grid-minor theorem. J. ACM, 63(5):Art. 40, 65, 2016.

[10] Maria Chudnovsky, Bruce Reed, and Paul Seymour. The edge-density for K2,t minors. J. Combin. Theory Ser. B, 101(1):18–46, 2011. [11] Julia Chuzhoy. Excluded grid theorem: improved and simpliﬁed. In STOC’15—Proceedings of the 2015 ACM Symposium on Theory of Computing, pp. 645–654. ACM, New York, 2015. [12] Julia Chuzhoy. Improved bounds for the excluded grid theorem. arXiv preprint arXiv:1602.02629, 2016. [13] Julia Chuzhoy and Zihan Tan. Towards tight(er) bounds for the excluded grid theorem. J. Combin. Theory Ser. B, 146:219–265, 2021. [14] Endre Csóka, Irene Lo, Sergey Norin, Hehui Wu, and Liana Yepremyan. The extremal function for disconnected minors. J. Combin. Theory Ser. B, 126:162–174, 2017. [15] Wojciech Czerwiński, Wojciech Nadara, and Marcin Pilipczuk. Improved bounds for the excluded-minor approximation of treedepth. In 27th Annual European Symposium on Algorithms, vol. 144 of LIPIcs. Leibniz Int. Proc. Inform., pp. Art. No. 34, 13. Schloss Dagstuhl. Leibniz-Zent. Inform., Wadern, 2019. [16] Gabriel A. Dirac. Homomorphism theorems for graphs. Math. Ann., 153:69–80, 1964. [17] Christopher S. Edwards. Some extremal properties of bipartite subgraphs. Canadian J. Math., 25:475–485, 1973. [18] Paul Erdős. On some extremal problems in graph theory. Israel J. Math., 3:113–116, 1965. [19] Paul Erdős, Ralph Faudree, János Pach, and Joel Spencer. How to make a graph bipartite. J. Combin. Theory Ser. B, 45(1):86–98, 1988. [20] Paul. Erdős and András Hajnal. On complete topological subgraphs of certain graphs. Ann. Univ. Sci. Budapest. Eötvös Sect. Math., 7:143–149, 1964. [21] Jim Geelen, Bert Gerards, Bruce Reed, Paul Seymour, and Adrian Vetta. On the odd-minor variant of Hadwiger’s conjecture. J. Combin. Theory Ser. B, 99(1):20–29, 2009. [22] Carla Groenland, Gwenaël Joret, Wojciech Nadara, and Bartosz Walczak. Approximating pathwidth for graphs of small treewidth. In Proceedings of the 2021 ACM-SIAM Symposium on Discrete Algorithms (SODA), pp. 1965–1976. SIAM, 2021. [23] Daniel J. Harvey and David R. Wood. Cycles of given size in a dense graph. SIAM J. Discrete Math., 29(4):2336–2349, 2015. [24] Daniel J. Harvey and David R. Wood. Average degree conditions forcing a minor. Electron. J. Combin., 23(1):Paper 1.42, 8, 2016. [25] Daniel J. Harvey and David R. Wood. Parameters tied to treewidth. J. Graph Theory, 84(4):364–385, 2017. [26] Kevin Hendrey and David R. Wood. The extremal function for Petersen minors. J. Combin. Theory Ser. B, 131:220–253, 2018. [27] Robert Hickingbotham and David R. Wood. Hadwiger number for direct products. 2021, in preparation. [28] Robert Hickingbotham and David R. Wood. Structural properties of graph products. 2021, in preparation.

[29] Leif K. Jørgensen. Contractions to K8. J. Graph Theory, 18(5):431–448, 1994.

12 [30] Heinz A. Jung. Eine Verallgemeinerung des n-fachen Zusammenhangs für Graphen. Math. Ann., 187:95–103, 1970. [31] Ken-ichi Kawarabayashi and Benjamin Rossman. A polynomial excluded-minor approximation of treedepth. In Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 234–246. SIAM, 2018. [32] János Komlós and Endre Szemerédi. Topological cliques in graphs. Combin. Probab. Comput., 3(2):247–256, 1994. [33] János Komlós and Endre Szemerédi. Topological cliques in graphs. II. Combin. Probab. Comput., 5(1):79–90, 1996. [34] Alexandr V. Kostochka. The minimum Hadwiger number for graphs with a given mean degree of vertices. Metody Diskret. Analiz., (38):37–58, 1982. [35] Alexandr V. Kostochka. Lower bound of the Hadwiger number of graphs by their average degree. Combinatorica, 4(4):307–316, 1984.

[36] Alexandr V. Kostochka and Noah Prince. On Ks,t-minors in graphs with given average degree. Discrete Math., 308(19):4435–4445, 2008.

[37] Alexandr V. Kostochka and Noah Prince. Dense graphs have K3,t minors. Discrete Math., 310(20):2637–2654, 2010.

[38] Alexandr V. Kostochka and Noah Prince. On Ks,t-minors in graphs with given average degree, II. Discrete Math., 312(24):3517–3522, 2012. [39] Daniela Kühn and Deryk Osthus. Topological minors in graphs of large girth. J. Combin. Theory Ser. B, 86(2):364–380, 2002. [40] Daniela Kühn and Deryk Osthus. Large topological cliques in graphs without a 4-cycle. Combin. Probab. Comput., 13(1):93–102, 2004. [41] Daniela Kühn and Deryk Osthus. Extremal connectivity for topological cliques in bipartite graphs. J. Combin. Theory Ser. B, 96(1):73–99, 2006. [42] Daniela Kühn and Deryk Osthus. Improved bounds for topological cliques in graphs of large girth. SIAM J. Discrete Math., 20(1):62–78, 2006. [43] Wolfgang Mader. Homomorphieeigenschaften und mittlere Kantendichte von Graphen. Math. Ann., 174:265–268, 1967. [44] Wolfgang Mader. Homomorphiesätze für Graphen. Math. Ann., 178:154–168, 1968. [45] Wolfgang Mader. Existenzn-fach zusammenhängender teilgraphen in graphen genügend großer kantendichte. In Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg, vol. 37, pp. 86–97. Springer, 1972. [46] Joseph Samuel Myers and Andrew Thomason. The extremal function for noncomplete minors. Combinatorica, 25(6):725–753, 2005. [47] Jaroslav Nešetřil and Patrice Ossona de Mendez. Sparsity: graphs, structures, and algorithms, vol. 28. Springer, 2012. [48] Bruce Reed. Tree width and tangles: a new connectivity measure and some applications. In Surveys in combinatorics, 1997 (London), vol. 241 of London Math. Soc. Lecture Note Ser., pp. 87–162. Cambridge Univ. Press, Cambridge, 1997. [49] Bruce Reed and David R. Wood. Forcing a sparse minor. Combin. Probab. Comput., 25(2):300–322, 2016. [50] Neil Robertson and Paul Seymour. Graph minors. I. Excluding a forest. J. Combin. Theory Ser. B, 35(1):39–61, 1983. [51] Neil Robertson and Paul Seymour. Graph minors. II. Algorithmic aspects of tree-width. J. Algorithms, 7(3):309–322, 1986. [52] Neil Robertson and Paul Seymour. Graph minors. V. Excluding a planar graph. J. Combin. Theory Ser. B, 41(1):92–114, 1986. [53] Petra Scheffler. Die Baumweite von Graphen als ein Maßfür die Kompliziertheit algo-

13 rithmischer Probleme, vol. 89 of Report MATH. Akademie der Wissenschaften der DDR, Karl-Weierstrass-Institut für Mathematik, Berlin, 1989. [54] Alex Scott and David R. Wood. Separation dimension and degree. Math. Proc. Cambridge Philos. Soc., 170(3):549–558, 2021.

[55] Zi-Xia Song and Robin Thomas. The extremal function for K9 minors. J. Combin. Theory Ser. B, 96(2):240–252, 2006.

[56] Benny Sudakov. Making a K4-free graph bipartite. Combinatorica, 27(4):509–518, 2007. [57] Robin Thomas and Youngho Yoo. The extremal functions for triangle-free graphs with excluded minors. European J. Combin., 75:1–10, 2019. [58] Andrew Thomason. An extremal function for contractions of graphs. Math. Proc. Cambridge Philos. Soc., 95(2):261–265, 1984. [59] Andrew Thomason. The extremal function for complete minors. J. Combin. Theory Ser. B, 81(2):318–338, 2001. [60] Carsten Thomassen. On the presence of disjoint subgraphs of a speciﬁed type. J. Graph Theory, 12(1):101–111, 1988. [61] Klaus Wagner. Über eine Eigenschaft der ebenen Komplexe. Math. Ann., 114(1):570–590, 1937. [62] Raphael Yuster. A note on graphs with k-connected subgraphs. Ars Combin., 67:231–235, 2003.