\tkzSetUpPoint
[fill=black, size = 3pt]\tkzSetUpLine[color=black, line width=0.6pt]\tkzSetUpCircle[color=black, line width=0.6pt]
James Davies 111Department of Pure Mathematics and Mathematical Statistics, University of Cambridge, Cambridge, UK (jgd37@cam.ac.uk). Robert Hickingbotham 222School of Mathematics, Monash University, Melbourne, Australia (robert.hickingbotham@monash.edu). Research supported by an Australia Mathematical Society Lift-Off Fellowship and the Monash Postgraduate Publication Award.
Freddie Illingworth 333Department of Mathematics, University College London, London, UK (f.illingworth@ucl.ac.uk). Research supported by EPSRC grant EP/V521917/1 and the Heilbronn Institute for Mathematical Research. Rose McCarty 444School of Mathematics and School of Computer Science, Georgia Institute of Technology, Atlanta, USA (rmccarty3@gatech.edu). Supported by the National Science Foundation under Grant No.DMS-2202961.
Abstract
We disprove the conjecture of Georgakopoulos and Papasoglu that a length space (or graph) with no -fat minor is quasi-isometric to a graph with no minor. Our counterexample is furthermore not quasi-isometric to a graph with no 2-fat minor or a length space with no minor.On the other hand, we show that the following weakening holds: any graph with no -fat minor is quasi-isometric to a graph with no -fat minor.
1 Introduction
Inspired by connections to metric geometry, fat minors were independently introduced by Chepoi, Dragan, Newman, Rabinovich, and Vaxes[CDN+12] and Bonamy, Bousquet, Esperet, Groenland, Liu, Pirot, and Scott[BBE+23] as a coarse variant of graph minors.They have been a key tool in resolving open problems at the intersection of structural graph theory and coarse geometry [BBE+23].Georgakopoulos and Papasoglu[GP23] recently gave a systematic overview of this blossoming area which can be described as “coarse graph theory”.At the heart of their paper, they proposed a natural conjecture that would effectively lift graph minor theory to the coarse setting: for every finite graph , graphs with no fat minor are quasi-isometric to graphs with no minor. Quasi-isometry is a fundamental notion from geometric group theory which preserves the large-scale geometry of a metric space. So, if true, this conjecture would imply coarse analogues to various results from graph minor theory, such as the Graph Minor Structure Theorem of Robertson and Seymour[RS03].
Georgakopoulos and Papasoglu’s conjecture has been settled for particular graphs . Manning[Man05] characterised quasi-trees which implies the case (see[GP23]). The case was proved by Chepoi, Dragan, Newman, Rabinovich, and Vaxes[CDN+12], and more generally, Fujiwara and Papasoglu[FP23] recently characterised quasi-cacti. The case was recently settled by Georgakopoulos and Papasoglu[GP23].
Our main contribution is a counterexample to the conjecture of Georgakopoulos and Papasoglu.
Theorem 1.
There exists a finite graph such that, for every , there is a graph that does not contain as a -fat minor and is not -quasi-isometric to any graph with no (-fat) minor.
In fact, Theorem1 gives a strong counterexample to Georgakopoulos and Papasoglu’s conjecture, since we can find as a 2-fat minor, rather than just as a minor.One might hope that an analogue of the conjecture might still hold for edge weighted graphs or, more generally, length spaces, but this is not the case.
Theorem 2.
There exists a finite graph such that, for every , there is a graph that does not contain as a -fat minor and is not -quasi-isometric to any length space with no -fat minor.
It turns out that Theorem1 is essentially tight;we prove that graphs with no -fat minor are -quasi-isometric to graphs with no -fat minor.By scaling,Theorem2 is also similarly tight up to some positive factor depending on (see 10).
Theorem 3.
Let be a positive integer, a non-empty collection of graphs, and a graph that has no -fat minor for every . Then is -quasi-isometric to a graph that does not contain a -fat minor for each .
Our proof of Theorem1 is based on a recent construction of Nguyen, Scott, and Seymour[NSS24] which was used to disprove a conjectured coarse version of Menger’s theorem. The connection between Menger’s theorem and Theorem1 intuitively comes from the fact that Menger’s theorem can be interpreted as a statement about rooted minors.
2 Preliminaries
A length space is a metric space such that, for every and , there is an -arc of length at most whenever is finite.In particular, every graph (where the edges are arcs of length one) is a length space (in which one can even take ). Graphs with arbitrary non-negative real-valued edge-lengths also induce a length space. See [BH99] for further background on length spaces.
For point sets in a length space , an -path is a path in with one end-point in and the other in . The distance between and , denoted is the infimum of the lengths of the -paths in . Note that this corresponds to the usual notion of distance for graphs.
We say that a set of points in a length space is path-connected if, for every , there is an -path contained in .Note that this corresponds to the usual notion of connectivity for graphs.For a length space , a non-negative real , and a set of points , we let denote the set of points such that there is an -path in of length at most . In graphs this is just the set of points at distance at most from .Note that for any point , is always path-connected.
For a positive integer , a -fat minor model of a graph in a length space is a collection of path-connected sets in such that
- •
whenever is an end of in ; and
- •
for any pair of distinct not covered by the above condition, we have .
If contains an -fat minor model of , then we say that is a -fat minor of .
For , a -quasi-isometry of a length space into a length space is a map such that, for every ,
and, for every , there is a with . If is a graph, then we restrict to quasi-isometries with .
Let be a graph and . We write for the subgraph of induced on . We refer to and interchangeably, whenever there is no chance of confusion. A graph is a subdivision of a if can be obtained from by replacing each edge of by a path with end-vertices and . If each path has exactly internal vertices, then we say that is the -subdivision of .
For a graph , a tree-decomposition of is a collection of bags indexed by a tree such that for each , is a non-empty subtree of ; and for each , there is a node such that . The width of is . The treewidth of a graph is the minimum width of a tree-decomposition of . Treewidth is a fundamental parameter in algorithmic and structural graph theory and is the standard measure of how similar a graph is to a tree.
3 Removing coarseness counterexample
3.1 The construction
We start with Nguyen, Scott, and Seymour’s counterexample to a conjectured coarse version of Menger’s theorem[NSS24]. For a positive integer , they defined the graph shown in Figure1 where the complete binary treehas depth with each solid line being an edge and each dotted line represents a path of length .
![Fat minors cannot be thinned (by quasi-isometries) (1) Fat minors cannot be thinned (by quasi-isometries) (1)](https://i0.wp.com/arxiv.org/html/2405.09383v1/x1.png)
They proved that has the following property.
Lemma 4 ([NSS24]).
Let and be the sets of the three vertices in as shown in the figure,and let , , be three -paths in . Then there exist distinct such that .
We prove an additional property of .
Lemma 5.
has treewidth at most seven.
Proof.
Subdividing an edge does not change the treewidth of a graph and so has the same treewidth as the graph where each dotted segment has been replaced by an edge. Contract the following edges of the path at the bottom of : the edges joining the 1st and 2nd vertices, 3rd and 4th vertices, the 5th and 6th vertices, …. The resulting graph is a minor of the graph depicted in Figure2.
![Fat minors cannot be thinned (by quasi-isometries) (2) Fat minors cannot be thinned (by quasi-isometries) (2)](https://i0.wp.com/arxiv.org/html/2405.09383v1/x2.png)
The graph in Figure2 is a subgraph of a Halin graph111A Halin graph is a planar graph that can be constructed from a planar embedding of a tree by connecting its leaves with a cycle which passes around the tree’s boundary. and so has treewidth at most three[Bod88]. Thus . When contracting the edges to get from to , the bags drop in size by a factor of at most two and so , as required.∎
Now we turn to the construction of and in Theorem1.
Let and be vertex-disjoint copies of with vertex-sets and , respectively. Let be the graph obtained from the union of and by adding the edges , , as shown in Figure3.We will use the following properties of :
- (i)
and are both -connected graphs with treewidth ; and
- (ii)
and are isomorphic.
The graph is obtained as follows and shown in Figure4.Let be a -subdivision of and let be a -subdivision of . We let and denote the high degree vertices of and , respectively.Take disjoint copies of each of , , and .Label the vertices of as , , and the vertices of as , , .For each add a path of length from to and add a path of length from to .
3.2 Avoiding fat minors
We now show that does not contain as a 3-fat minor.
Lemma 6.
For every , does not contain as a -fat minor.
Proof.
Suppose that contains as a -fat minor. Let and be a -fat minor model of in . We begin by considering the -fat sub-model of in .
Claim 1:There exists such that or .
Proof.
Let . Since is obtained from by adding paths, each with one end-vertex on , it follows that . Now suppose that for every . Let and . Then is a minor of .Since every path in from to contains a vertex from , it follows that . Therefore , contradicting Lemma5. As such, there exists such that . Since is connected, it follows that either or .∎
By symmetry, we may assume without loss of generality that .
Claim 2: For every , .
Proof.
Suppose there is some vertex such that . Then every -path in contains a vertex from . So contains at most three vertex-disjoint -paths, which contradicts and being a -fat model of a -connected graph.∎
Claim 3: Every vertex of degree at least in is contained in for some .
Proof.
Let . Since , contains a vertex in of degree at least . If , then contains a vertex of degree at least in . If , then Claim 2 implies that contains , , or . So each contain a vertex of degree at least in . Since , the claim then follows.∎
We now consider the -fat sub-model of in .
Claim 4: For every , .
Proof.
By symmetry, Claims 1 & 2 imply that either for every , or for every . Suppose that for every . Then there is a vertex such that . Since , it follows that contains a vertex of degree at least from . But by Claim 3, this implies that for some , a contradiction. Therefore, for every .∎
To conclude the proof, consider the edges .For each , let be the path in our model that corresponds to . Since is connected, Claims 2 & 4 imply that there is an -path in . By Lemma4, there are distinct such that . Therefore, there is and such that , contradicting and being -fat.∎
3.3 Finding fat minors
In this subsection, we shall find as a fat minor in length spaces (or graphs) that are quasi-isometric to .
Lemma 7.
Let be a length space that is -quasi-isometric to .If is a graph, then let , otherwise let .Then contains as a -fat minor.
Lemmas6 and7 immediately implies Theorems1 and2. Before proving Lemma7, we need the following two lemmas to help us build our fat model in . The first lemma allows us to find path-connected sets in which spans a set of elements, while the second lemma allows us to find path-connected sets in that are sufficiently far away from each other.