Elsevier

Discrete Optimization

Volume 34, November 2019, 100548
Discrete Optimization

Small 1-defective Ramsey numbers in perfect graphs

https://doi.org/10.1016/j.disopt.2019.06.001Get rights and content

Abstract

In this paper, we initiate the study of defective Ramsey numbers for the class of perfect graphs. Let PG be the class of all perfect graphs and R1PG(i,j) denote the smallest n such that all perfect graphs on n vertices have either a 1-dense i-set or a 1-sparse j-set. We show that R1PG(3,j)=j for any j2, R1PG(4,4)=6, R1PG(4,5)=8, R1PG(4,6)=10, R1PG(4,7)=13, R1PG(4,8)=15 and R1PG(5,5)=13. We exhibit all extremal graphs for R1PG(4,7)=13 (there are exactly three). We also obtain the 1-defective Ramsey number of order (4,7) in triangle-free perfect graphs, namely R1ΔPG(4,7)=12.

Introduction

For any two positive integers i and j, the Ramsey number R(i,j) is the smallest positive integer such that every graph on at least R(i,j) vertices has a clique of size i or an independent set of size j. Ramsey numbers along with several variations are extensively studied in the literature. Among various generalizations of the classical Ramsey numbers, we note that defective Ramsey numbers have been the focus of several research papers [1], [2], [3], [4]. This variation is based on the following relaxation of the notions of cliques and independent sets. A k-sparsei-set is a set S of i vertices of a graph G such that each vertex in S has degree at most k in the subgraph of G induced by S. A k-densej-set is a set D of j vertices of a graph G that is k-sparse in the complement of G; in other words, each vertex in D misses at most k other vertices in D. The term k-defective (or k-uniform) set is used to denote a k-sparse or k-dense set. The defective Ramsey number RkG(i,j) is the smallest n such that all graphs on n vertices in the class G have either a k-dense i-set or a k-sparse j-set. In [3] some 1-defective Ramsey numbers are reported under the name of 1-dependent Ramsey numbers. In [2], [4] more 1-defective and 2-defective Ramsey numbers are found using direct proof techniques and several bounds are derived on general defective Ramsey numbers. However, direct proof techniques seem to have reached their limits in finding new values of defective Ramsey numbers. Indeed, this is rather not surprising given the great difficulty of computing specific Ramsey numbers (of all kinds). Having noticed this fact, some computer based generation methods are used in [1] and [2] to improve the known bounds on defective Ramsey numbers (and some other defective parameters).

Noting that computing all Ramsey numbers is notoriously unlikely, various approaches have been adopted in the literature to cope with this difficulty. One way to attack Ramsey numbers is to consider restricted graph families. In [5], all Ramsey numbers in planar graphs are computed. In [6], [7], [8], the authors achieved with the computation of some Ramsey numbers for graphs with bounded degree. Exact values as well as upper and lower bounds on some Ramsey numbers for claw-free graphs are proved in [9]. This approach has gained some importance more recently and become popular again. In [10], the authors conduct a systematic study of Ramsey numbers in various graph classes. They determine some classes of graphs such as claw-free graphs for which there are infinitely many non-trivial Ramsey numbers as for arbitrary graphs. Besides, they exhibit all (classical) Ramsey numbers for perfect graphs and some well-known subclasses of claw-free graphs. Similarly, a very recent work [11] focuses on the complexity of the coloring problem where every color class is a k-sparse set (called the defective coloring problem) when restricted to subclasses of perfect graphs.

In this paper, we adopt this natural approach of considering hard-to-compute parameters in special cases. We initiate the study of defective Ramsey numbers for perfect graphs and compute first values for small numbers. Namely, we show that R1PG(3,j)=j for any j2, R1PG(4,4)=6, R1PG(4,5)=8, R1PG(4,6)=10, R1PG(4,7)=13, R1PG(4,8)=15 and R1PG(5,5)=13 where PG denotes the class of perfect graphs. We also exhibit all (there are exactly three) extremal graphs for R1PG(4,7)=13. To find these extremal graphs, we also obtain (as a byproduct) R1ΔPG(4,7)=12 where ΔPG denotes the class of triangle-free perfect graphs.

Our main proofs do not use any previously known results. They are based on techniques from classical Ramsey theory such as considering the neighborhood of specific vertices and on the structural properties of perfect graphs and 1-defective sets. We also invoke a computer assisted proof for one of these values. We conclude with several research directions arising from our work.

Let G=(V,E) be a graph. A (partial) subgraph HG is a graph on VV and EE with both end-vertices of each edge of E in V. If all edges with both end-vertices in V are in E, then H is said to be an induced subgraph of G. In our context, whenever we say that a graph contains a subgraph, we always mean as a partial subgraph, unless stated otherwise. For a vertex xV, we denote by N(x) the set of neighbors of x, that is, vertices adjacent to x. The degree of a vertex x is d(x)=|N(x)|. We also have N[x]=N(x){x}. For a vertex xV and a subgraph HG, we denote by NH(x) the set of neighbors of x in H, that is N(x)V(H). Similarly, the degree of x in H is dH(x)=|NH(x)|. For a subset of vertices XV, the neighborhood of X, denoted by N(X), is defined as N(X)=(xXN(x))X. For a subset of vertices XV and a subgraph HG, we define NH(X) similarly to NH(x).

For a graph G and a subgraph H, we use the notation GH to mean the subgraph of G induced by all vertices in V(G)V(H). We also use the same notation when we remove a set of vertices from a graph. For a graph H and G, we say that G is H-free if it does not contain H as an induced subgraph. A path on n vertices is denoted by Pn, and a cycle on n vertices, also called an n-cycle, is denoted by Cn. The distance between two vertices is the length of a shortest path between them. The girth of a graph G, denoted by g(G) is the length of a shortest induced cycle in it. A clique of size i, denoted by Ki, is a set of i vertices which are pairwise adjacent. Given a graph G, the maximum size of a clique is denoted by ω(G). A set of vertices is called independent if all vertices in it are pairwise non-adjacent.

The chromatic number of G, denoted by χ(G) is the minimum number of colors to color the vertices of it in such a way that no two adjacent vertices receive the same color. A graph G is called perfect if for every induced subgraph H of G, we have χ(H)=ω(H). According to the Strong Perfect Graph Theorem, a graph is perfect if and only if it contains no induced odd cycles of length at least 5 and their complements [12]. We will be using this result heavily in our proofs. An extremal graph for RkG(i,j) is a graph in G on RkG(i,j)1 vertices containing neither a k-dense i-set nor a k-sparse j-set. We omit the superscript G when we consider a Ramsey number in arbitrary graphs.

Section snippets

Preliminary remarks

In [10], the authors find Ramsey numbers for perfect graphs, denoted by RPG(i,j), as follows:

Theorem 2.1

Theorem 6 of [10]

Let PG be the class of perfect graphs or a subclass of it containing all disjoint unions of complete graphs. ThenRPG(i,j)=(i1)(j1)+1 for alli,j1.

Indeed, the disjoint union of (j1) copies of Ki1 shows that RPG(i,j)>(i1)(j1). Now, let GPG be of size (i1)(j1)+1. If χ(G)=ω(G)i then we already have clique of size i and we are done. Consider the case where χ(G)=ω(G)i1. Since we have (i1)(j1)+1

The Ramsey number R1PG(4,7)

The first value of a 1-defective Ramsey number which does not follow from previous works is R1PG(4,7). It is shown in [3] that R1(4,7)=16 with the only extremal graph being H in Fig. 1. Since H is not perfect, this implies that R1PG(4,7)15. However, it turns out that in this situation, the 1-defective Ramsey numbers in general and in perfect graphs are further apart for the same parameters. In what follows, we will show that R1PG(4,7)=13.

The following fact will be repetitively used in our

Triangle-free perfect graphs

In this section, we show that if we have a perfect graph with 12 vertices, one can still guarantee the existence of a 1-dense 4-set or a 1-sparse 7-set under the additional condition that the graph is triangle-free, that is of girth at least 4. Moreover, this result is best possible in the sense that if we have fewer vertices, then we cannot guarantee it. We now present this result which will be then used to obtain all extremal graphs for R1PG(4,7); we will note that there are exactly three

Extremal graphs for R1PG(4,7)

Now, we are ready to find all extremal graphs for R1PG(4,7)=13. Following Corollary 4.2, we will consider extremal graphs containing an induced C10, C8 and C6 separately in Propositions 5.1, Proposition 5.2, Proposition 5.3 respectively. We then combine them in Theorem 5.4 to state that G1,G2,G3 in Fig. 5 are the only perfect graphs on 12 vertices with neither 1-dense 4-set nor 1-sparse 7-set.

Recall that by Corollary 4.2, an extremal graph for R1PG(4,7)=13 has no induced cycles of length 7, 9

The Ramsey number R1PG(4,8)

We continue with the computation of R1PG(4,8). In [3], it is shown that R1(4,8)=17 (however not all extremal graphs are provided). Here, we show that when restricted to perfect graphs, we obtain R1PG(4,8)=15.

Theorem 6.1

With the preceding notation, R1PG(4,8)=15.

Proof

Observe that the Heawood graph is perfect, has 14 vertices and has no 4-cycle or 1-sparse 8-set. It follows that R1PG(4,8)15.

In what follows, let G be a perfect graph on 15 vertices, we will show that G has either a C4 or a 1-sparse 8-set. To this

Conclusion

In this paper, we initiated the study of defective Ramsey numbers in perfect graphs and computed some values. In addition to those, we also showed the following with a computer assisted proof.

Theorem 7.1

With the preceding notation, R1PG(5,5)=13.

Proof

Using a computer program, we considered all graphs on 9 vertices taken from the online available source [15]. We filtered out all non-perfect graphs on 9 vertices by searching for induced odd cycles of length at least 5 or their complements. In this way, we

References (16)

There are more references available in the full text version of this article.

Cited by (5)

1

Tınaz Ekim conducted part of this research when she was visiting University of Oregon, Department of Computer and Information Science under Fulbright Visiting Scholar Grant and TUBITAK 2219 Programme, all of whose support is greatly appreciated. She also acknowledges the support of the Turkish Academy of Science TUBA GEBIP award.

2

This work was initiated when John Gimbel visited Istanbul Center for Mathematical Sciences (IMBM) whose support is greatly acknowledged.

View full text