Searcharxiv⌕ Search

arXiv subjects

Paul Seymour

Publications and source records attributed to Paul Seymour.

At least 73 records · Page 4Linked to original sources
arXiv2021

Induced subgraphs of graphs with large chromatic number. V. Chandeliers and strings

It is known that every graph of sufficiently large chromatic number and bounded clique number contains, as an induced subgraph, a subdivision of any fixed forest, and a subdivision of any fixed cycle. Equivalently, forests and triangles are pervasive, where H is pervasive (in some class of graphs) if for all s>0, every graph in the class with bounded clique number and sufficiently large chromatic number contains an induced subdivision of H, with every edge subdivided at least s times. Which other graphs are pervasive? Chalopin, Esperet, Li and Ossona de Mendez proved that every such graph is a forest of lanterns: roughly, the blocks are lanterns (graphs obtained from a tree by adding one extra vertex), and there are rules about how blocks fit together. It is not known whether every forest of lanterns is pervasive; but in another paper two of us prove that banana trees (multigraphs obtained from a forest by adding parallel edges) are pervasive, thus generalizing the two results above. This paper contains the first half of the proof, which works for any forest of lanterns, not just for banana trees. A class of graphs is r-controlled if for every graph in the class, its chromatic number is at most some function (determined by the class) of the largest chromatic number of an r-ball in the graph. In this paper we prove that for all r>1, every forest of lanterns is pervasive in every r-controlled class These results turn out particularly nicely when applied to string graphs (intersection graphs of sets of curves in the plane). A chandelier is a graph obtained from a tree by adding a vertex adjacent to its leaves. We prove that the class of string graphs is 2-controlled, and thus forests of lanterns are pervasive in this class. Furthermore, string graphs of sufficiently large chromatic number and bounded clique number contain any fixed chandelier as an induced subgraph.

math.CO↗
arXiv2021

Pure pairs. VI. Excluding an ordered tree

A pure pair in a graph $G$ is a pair $(Z_1,Z_2)$ of disjoint sets of vertices such that either every vertex in $Z_1$ is adjacent to every vertex in $Z_2$, or there are no edges between $Z_1$ and $Z_2$. With Maria Chudnovsky, we recently proved that, for every forest $F$, every graph $G$ with at least two vertices that does not contain $F$ or its complement as an induced subgraph has a pure pair $(Z_1,Z_2)$ with $|Z_1|,|Z_2|$ linear in $|G|$. Here we investigate what we can say about pure pairs in an {\em ordered} graph $G$, when we exclude an ordered forest $F$ and its complement as induced subgraphs. Fox showed that there need not be a linear pure pair; but Pach and Tomon showed that if $F$ is a monotone path then there is a pure pair of size $c|G|/\log |G|$. We generalise this to all ordered forests, at the cost of a slightly worse bound: we prove that, for every ordered forest $F$, every ordered graph $G$ with at least two vertices that does not contain $F$ or its complement as an induced subgraph has a pure pair of size $|G|^{1-o(1)}$.

math.CO↗
arXiv2021

Erdos-Hajnal for graphs with no 5-hole

The Erdos-Hajnal conjecture says that for every graph H there exists c>0 such that every graph G not containing H as an induced subgraph has a clique or stable set of cardinality at least |G|^c. We prove that this is true when H is a cycle of length five. We also prove several further results: for instance, that if C is a cycle and H is the complement of a forest, there exists c>0 such that every graph G containing neither of C,H as an induced subgraph has a clique or stable set of cardinality at least |G|^c.

math.CO↗
arXiv2021

Pure pairs. VII. Homogeneous submatrices in 0/1-matrices with a forbidden submatrix

For integer $n>0$, let $f(n)$ be the number of rows of the largest all-0 or all-1 square submatrix of $M$, minimized over all $n\times n$ $0/1$-matrices $M$. Thus $f(n)= O(\log n)$. But let us fix a matrix $H$, and define $f_H(n)$ to be the same, minimized over over all $n\times n$ $0/1$-matrices $M$ such that neither $M$ nor its complement (that is, change all $0$'s to $1$'s and vice versa) contains $H$ as a submatrix. It is known that $f_H(n)\ge εn^c$, where $c, ε>0$ are constants depending on $H$. When can we take $c=1$? If so, then one of $H$ and its complement must be an acyclic matrix (that is, the corresponding bipartite graph is a forest). Korandi, Pach, and Tomon conjectured the converse, that $f_H(n)$ is linear in $n$ for every acyclic matrix $H$; and they proved it for certain matrices $H$ with only two rows. Their conjecture remains open, but we show $f_H(n)=n^{1-o(1)}$ for every acyclic matrix $H$; and indeed there is a $0/1$-submatrix that is either $Ω(n)\times n^{1-o(1)}$ or $n^{1-o(1)}\times Ω(n)$.

math.CO↗
arXiv2020

Sparse graphs with no polynomial-sized anticomplete pairs

A graph is "$H$-free" if it has no induced subgraph isomorphic to $H$. A conjecture of Conlon, Fox and Sudakov states that for every graph $H$, there exists $s>0$ such that in every $H$-free graph with $n>1$ vertices, either some vertex has degree at least $sn$, or there are two disjoint sets of vertices, of sizes at least $sn^s$ and $sn$, anticomplete to each other. We prove this holds for a large class of graphs $H$, and we prove that something like it holds for all graphs $H$. Say $H$ is "almost-bipartite" if $H$ is triangle-free and $V(H)$ can be partitioned into a stable set and a set inducing a graph of maximum degree at most one. We prove that the conjecture above holds for when $H$ is almost-bipartite. We also prove a stronger version where instead of excluding $H$ we restrict the number of copies of $H$. We prove some variations on the conjecture, such as: for every graph $H$, there exists $s >0$ such that in every $H$-free graph with $n>1$ vertices, either some vertex has degree at least $sn$, or there are two disjoint sets $A, B$ of vertices with $|A||B| > s n^{1 + s}$, anticomplete to each other.

math.CO↗
arXiv2020

Concatenating bipartite graphs

Let $x,y\in(0,1]$ and let $A,B,C$ be disjoint nonempty subsets of a graph $G$, where every vertex in $A$ has at least $x|B|$ neighbours in $B$, and every vertex in $B$ has at least $y|C|$ neighbours in $C$. We denote by $ϕ(x,y)$ the maximum $z$ such that, in all such graphs $G$, there is a vertex $v$ in $C$ that is joined to at least $z|A|$ vertices in $A$ by two-edge paths. The function $ϕ$ is interesting, and we investigate some of its properties. For instance, we show that it is symmetric in $x$ and $y$, and that it has a discontinuity at $x=y=1/k$ for all integers $k>1$. We raise a number of questions and conjectures.

math.CO↗
arXiv2020

Detecting a long even hole

For each integer $\ell \geq 4$, we give a polynomial-time algorithm to test whether a graph contains an induced cycle with length at least $\ell$ and even

math.CO↗
arXiv2020

Pure pairs. I. Trees and linear anticomplete pairs

The Erdos-Hajnal Conjecture asserts that for every graph H there is a constant c > 0 such that every graph G that does not contain H as an induced subgraph has a clique or stable set of cardinality at least |G|^c. In this paper, we prove a conjecture of Liebenau and Pilipczuk, that for every forest H there exists c > 0, such that every graph G contains either an induced copy of H, or a vertex of degree at least c|G|, or two disjoint sets of at least c|G| vertices with no edges between them. It follows that for every forest H there is c > 0 so that if G contains neither H nor its complement as an induced subgraph then there is a clique or stable set of cardinality at least |G|^c.

math.CO↗
arXiv2020

Detecting a long odd hole

For each integer $t\ge 5$, we give a polynomial-time algorithm to test whether a graph contains an induced cycle with length at least $t$ and odd.

math.CO↗
arXiv2020

Holes with hats and Erdős-Hajnal

A "hole-with-hat" in a graph $G$ is an induced subgraph of $G$ that consists of a cycle of length at least four, together with one further vertex that has exactly two neighbours in the cycle, adjacent to each other, and the "house" is the smallest, on five vertices. It is not known whether there exists $ε>0$ such that every graph $G$ containing no house has a clique or stable set of cardinality at least $|G|^ε$; this is one of the three smallest open cases of the Erdős-Hajnal conjecture and has been the subject of much study. We prove that there exists $ε>0$ such that every graph $G$ with no hole-with-hat has a clique or stable set of cardinality at least $|G|^ε$

math.CO↗
arXiv2020

Pure pairs. II. Excluding all subdivisions of a graph

We prove for every graph H there exists a>0 such that, for every graph G with at least two vertices, if no induced subgraph of G is a subdivision of H, then either some vertex of G has at least a|G| neighbours, or there are two disjoint sets A,B of at least a|G| vertices such that no edge joins A and B. It follows that for every graph H, there exists c>0 such that for every graph G, if no induced subgraph of G or its complement is a subdivision of H, then G has a clique or stable set of cardinality at least |G|^c. This is related to the Erdos-Hajnal conjecture.

math.CO↗
arXiv2020

A survey of $χ$-boundedness

If a graph has bounded clique number, and sufficiently large chromatic number, what can we say about its induced subgraphs? András Gyárfás made a number of challenging conjectures about this in the early 1980's, which have remained open until recently; but in the last few years there has been substantial progress. This is a survey of where we are now.

math.CO↗
arXiv2020

Even-hole-free graphs still have bisimplicial vertices

A {\em hole} in a graph is an induced subgraph which is a cycle of length at least four. A hole is called {\em even} if it has an even number of vertices. An {\em even-hole-free} graph is a graph with no even holes. A vertex of a graph is {\em bisimplicial} if the set of its neighbours is the union of two cliques. In an earlier paper \cite{bisimplicial}, Addario-Berry, Havet and Reed, with the authors, claimed to prove a conjecture of Reed, that every even-hole-free graph has a bisimplicial vertex, but we have recently been shown that the "proof" has a serious error. Here we give a proof using a different method.

math.CO↗
arXiv2020

Finding a shortest odd hole

An odd hole in a graph is a induced cycle with odd length greater than 3. In an earlier paper (with Sophie Spirkl), solving a longstanding open problem, we gave a polynomial-time algorithm to test if a graph has an odd hole. We subsequently showed that, for every t, there is a polynomial time algorithm to test whether a graph contains an odd hole of length at least t. In this paper, we give an algorithm that finds a shortest odd hole, if one exists.

math.CO↗
arXiv2020

Induced subgraphs of bounded treewidth and the container method

A hole in a graph is an induced cycle of length at least 4. A hole is long if its length is at least 5. By $P_t$ we denote a path on $t$ vertices. In this paper we give polynomial-time algorithms for the following problems: the Maximum Weight Independent Set problem in long-hole-free graphs, and the Feedback Vertex Set problem in $P_5$-free graphs. Each of the above results resolves a corresponding long-standing open problem. An extended $C_5$ is a five-vertex hole with an additional vertex adjacent to one or two consecutive vertices of the hole. Let $\mathcal{C}$ be the class of graphs excluding an extended $C_5$ and holes of length at least $6$ as induced subgraphs; $\mathcal{C}$ contains all long-hole-free graphs and all $P_5$-free graphs. We show that, given an $n$-vertex graph $G \in \mathcal{C}$ with vertex weights and an integer $k$, one can in time $n^{\Oh(k)}$ find a maximum-weight induced subgraph of $G$ of treewidth less than $k$. This implies both aforementioned results.

cs.DS↗
arXiv2020

New Examples of Minimal Non-Strongly-Perfect Graphs

A graph is strongly perfect if every induced subgraph H has a stable set that meets every nonempty maximal clique of H. The characterization of strongly perfect graphs by a set of forbidden induced subgraphs is not known. Here we provide several new minimal non-strongly-perfect graphs.

math.CO↗
arXiv2019

Subdivided Claws and the Clique-Stable Set Separation Property

Let $\mathcal{C}$ be a class of graphs closed under taking induced subgraphs. We say that $\mathcal{C}$ has the {\em clique-stable set separation property} if there exists $c \in \mathbb{N}$ such that for every graph $G \in \mathcal{C}$ there is a collection $\mathcal{P}$ of partitions $(X,Y)$ of the vertex set of $G$ with $|\mathcal{P}| \leq |V(G)|^c$ and with the following property: if $K$ is a clique of $G$, and $S$ is a stable set of $G$, and $K \cap S =\emptyset$, then there is $(X,Y) \in \mathcal{P}$ with $K \subseteq X$ and $S \subseteq Y$. In 1991 M. Yannakakis conjectured that the class of all graphs has the clique-stable set separation property, but this conjecture was disproved by Göös in 2014. Therefore it is now of interest to understand for which classes of graphs such a constant $c$ exists. In this paper we define two infinite families $\mathcal{S}, \mathcal{K}$ of graphs and show that for every $S \in \mathcal{S}$ and $K \in \mathcal{K}$, the class of graphs with no induced subgraph isomorphic to $S$ or $K$ has the clique-stable set separation property.

math.CO↗