This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

On the clique behavior and Hellyness of the complements of regular graphs

Rafael Villarroel-Flores
Universidad Autónoma del Estado de Hidalgo
Carretera Pachuca-Tulancingo km. 4.5
Pachuca 42184 Hgo. México
email: [email protected]
MSC: 05C76
keywords: helly property, clique graphs
Abstract

A collection of sets is intersecting, if any pair of sets in the collection has nonempty intersection. A collection of sets 𝒞\mathcal{C} has the Helly property if any intersecting subcollection has nonempty intersection. A graph is Helly if the collection of maximal complete subgraphs of GG has the Helly property. We prove that if GG is a kk-regular graph with nn vertices such that n>3k+2k2kn>3k+\sqrt{2k^{2}-k}, then the complement G¯\overline{G} is not Helly. We also consider the problem of whether the properties of Hellyness and convergence under the clique graph operator are equivalent for the complement of kk-regular graphs, for small values of kk.

1 Introduction

In this paper we consider only finite simple graphs. As a reference for Graph Theory, we follow [8]. We usually identify subsets of vertices of a graph GG with the subgraph they induce. A complete in a graph GG is a set of vertices which are adjacent by pairs, and a clique is a maximal complete. The clique graph K(G)K(G) of a graph GG is the intersection graph of the cliques of GG. We can then define a sequence of iterated clique graphs as: K0(G)=GK^{0}(G)=G, Kn(G)=K(Kn1(G))K^{n}(G)=K(K^{n-1}(G)) for n1n\geq 1. If this sequence of graphs has a finite number of different graphs up to isomorphism, we say that GG is convergent, otherwise, GG is said to be divergent. There is a number of criteria in the literature in order to determine which of these behaviors correspond to a given graph GG, however it is conjectured that the problem in general is algorithmically unsolvable for finite graphs ([1]).

A collection of sets is intersecting if any two members of the collection has nonempty intersection. The collection has the Helly property if any intersecting subcollection has nonempty intersection. We say that a graph GG is Helly if the collection of the cliques of GG has the Helly property (in the literature, this is usually called the clique-Helly property). The Helly property on cliques can be tested in polynomial time ([6], [16]), and has been studied in several papers (for example, [5], [13]) and generalized ([4], [3]).

A Helly graph is convergent ([7]) but there are convergent graphs which are not Helly. However, there are several graph classes for which it is has been proven that a graph in such a class is convergent if and only if it is Helly. Such classes include that of cographs ([9]), complements and powers of cycles ([15], [10]), chessboard graphs ([12]) and circulants with three small jumps ([11]).

In this note we consider a class of regular graphs. We want to study regular graphs with high degree, hence it will be convenient for us to consider the complement. We prove that for k=1,2k=1,2, the complements of kk-regular graphs are convergent if and only if they are Helly, and we show that this happens exactly when the order of the graph is sufficiently small. Then for each k3k\geq 3, we determine a number N(k)N(k) such that all kk-regular graphs with at least N(k)N(k) vertices are such that their complements are not Helly. In contrast to the cases k=1,2k=1,2, we can show examples of 33-regular graphs GG such that G¯\overline{G} is not Helly, and still G¯\overline{G} is convergent. In fact, we conjecture that there are arbitrarily large 33-regular graphs GG where the complement G¯\overline{G} is convergent but not Helly.

2 Preliminaries

2.1 Definitions

The disjoint union GHG\cup H of two graphs G,HG,H is the graph that has as vertex set the disjoint union of the vertex sets of GG and HH, and as edge set the union of the edge sets of GG and HH. We can extend this definition to the disjoint union of a finite collection of graphs. The disjoint union of mm copies of the graph GG is denoted by mGmG. We denote by G+HG+H the graph obtained from the disjoint union of GG and HH, adding all edges between a vertex in GG and a vertex in HH. It is immediate that GH¯=G¯+H¯\overline{G\cup H}=\overline{G}+\overline{H}.

The mm-th octahedron OmO_{m} is the complement of the graph mK2mK_{2}.

A triangle in a graph GG is a complete with three vertices. If TT is a triangle, then its extended triangle T^\hat{T} in GG is the induced subgraph of GG on the vertices that are adjacent to at least two elements of TT. A graph is a cone if it has a vertex adjacent to all other vertices in the graph.

A cotriangle in a graph GG is an independent set of three of its vertices.

If GG is a graph, and σ:GG\sigma\colon G\to G is an automorphism, we say that σ\sigma is a coaffination if σ(x)x\sigma(x)\neq x and σ(x)\sigma(x) is not a neighbor of xx for every xGx\in G. Note that the complement of every cyclic graph has a coaffination.

2.2 Theorems on clique behavior

The following theorem appeared in [6] but it was proved independently in [16].

Theorem 2.1.

([6], [16]) A graph GG is Helly if and only if for any triangle TT in GG, the extended triangle T^\hat{T} is a cone.

We can now pose a theorem that gives a sufficient condition for convergence.

Theorem 2.2.

([7]) If a graph GG is Helly, then GG is convergent.

We now mention some theorems that imply divergence of a graph.

Theorem 2.3.

([15]) The graph OmO_{m} is divergent if and only if m3m\geq 3.

Theorem 2.4.

([10], Theorem 5.4) The complement of a cycle CnC_{n} is divergent if and only if n8n\geq 8.

Theorem 2.5.

([10], Theorem 4.6) If G,HG,H are graphs that have a coaffination, and HH is connected, then G+HG+H is divergent.

Theorem 2.6.

([10], Theorem 3.6) If AA, BB, CC are graphs that have a coaffination, then A+B+CA+B+C is divergent.

3 The complements of regular graphs of low degree

3.1 The case k=1k=1

Theorem 3.1.

The complement of a 11-regular graph with nn vertices is divergent if and only if n6n\geq 6.

Proof.

A 11-regular graph is the disjoint union of mm copies of K2K_{2}. It follows that the complements of the 11-regular graphs are precisely the octahedrons OmO_{m}. From Theorem 2.3, we obtain that if the order of the graph is greater than six, then the graph is divergent. Now, the graph O1O_{1} consists of two isolated vertices and O2O_{2} is a 44-cycle, and both such graphs are Helly and convergent. ∎

3.2 The case k=2k=2

Theorem 3.2.

The complement of every 22-regular graph with nn vertices is divergent if n9n\geq 9.

Proof.

A 22-regular graph is the disjoint union of several cycles. If there is only one cycle in the union (that is, if the graph is connected), then the result follows from Theorem 2.4. If there are exactly two cycles, then one of them has at least five vertices, and so its complement is connected. We can then apply Theorem 2.5 in this case. Now, if there are three or more cycles, we can apply Theorem 2.6. ∎

Contrary to the case k=1k=1, Theorem 3.2 is not an equivalence, since there are 22-regular graphs with not more than 8 vertices and with divergent complement. Theorem 2.4 deals with the case of cycles, and in that case, we get that G=C8G=C_{8} is the only 22-regular connected graph with divergent complement and |G|8|G|\leq 8. The only other 22-regular graphs with at most 8 vertices are C3CiC_{3}\cup C_{i} for i=3,4,5i=3,4,5 and C4C4C_{4}\cup C_{4}. The complement of C3C3C_{3}\cup C_{3} is K3,3K_{3,3} which has no triangles, and thus is Helly by Theorem 2.1, therefore convergent by Theorem 2.2. The complements of both graphs C3C4C_{3}\cup C_{4} and C4C4C_{4}\cup C_{4} can also be shown to be Helly by applying Theorem 2.1. Now the complement of C3C5C_{3}\cup C_{5} is divergent, because it is isomorphic to C3¯+C5¯\overline{C_{3}}+\overline{C_{5}} and then Theorem 2.5 applies. We have thus proven that the complement G¯\overline{G} of a 22-regular graph GG is convergent if and only if G¯\overline{G} is Helly.

4 The case k3k\geq 3

Given a graph GG, denote by t(G)t(G) the number of triangles in GG.

Lemma 4.1.

([14], Lemma 1) For a kk-regular graph GG with nn vertices, we have:

t(G)+t(G¯)=(n3)12nk(nk1).t(G)+t(\overline{G})=\binom{n}{3}-\frac{1}{2}nk(n-k-1).\qed (1)

If TT is a cotriangle in a graph GG and xGx\in G, we will say that xx is adjacent to TT if xx is a neighbor of at least two vertices of TT.

Lemma 4.2.

Let GG be a kk-regular graph such that G¯\overline{G} is Helly, and let TT be a cotriangle in GG. Then there are at least kk vertices in GG that are adjacent to TT.

Proof.

We have that TT is a triangle in G¯\overline{G} and so T^\hat{T}, its extended triangle in G¯\overline{G}, is a cone, by Theorem 2.1. Since G¯\overline{G} is (nk1)(n-k-1)-regular, considering the apex of the cone we obtain that |T^|nk|\hat{T}|\leq n-k. This means that there must be at least kk vertices that are neighbors to one or none of the vertices of TT in G¯\overline{G}. Such vertices are adjacent to TT. ∎

We will use 2-switches, as defined in ([2], page 20). This means the process of substituting the edges {a,b},{u,v}\{a,b\},\{u,v\} in GG with the edges {a,u}\{a,u\}, {b,v}\{b,v\}. Of course, this requires that a,ua,u are not already adjacent in GG, and also that b,vb,v are not adjacent in GG. Applying a 2-switch to a kk-regular graph produces another kk-regular graph.

Lemma 4.3.

Let GG be a kk-regular with nn vertices, where n4kn\geq 4k, and xGx\in G. Then xx is adjacent to at most (k2)(n2k)+(k3)\binom{k}{2}(n-2k)+\binom{k}{3} cotriangles. If this bound is met, then the connected component of xx is Kk,kK_{k,k}.

Proof.

Let xGx\in G, as in our hypothesis. We prove first that if there is an edge e={a,b}e=\{a,b\} in N=NG(x)N=N_{G}(x), we can perform a 2-switch on GG, obtaining a graph GG^{\prime}, where aa and bb are no longer adjacent, no new edges appear between vertices of NN, and xx is adjacent to at least the same amount of cotriangles in GG^{\prime} as it was in GG.

The edges in GG that cannot be switched with {a,b}\{a,b\} are of the form {r,s}\{r,s\}, and where both r,sr,s are neighbors of either aa or bb; or one of {r,s}\{r,s\} is a neighbor of both a,ba,b.

Consider then the edge e={a,b}e=\{a,b\} in NG(x)N_{G}(x). Let CC be the set of common neighbors of the vertices a,ba,b different from xx. Suppose that |C|=t|C|=t. Let ZZ be the set of vertices that are at distance at least 2 from xx, and let Z1ZZ_{1}\subseteq Z be the vertices in ZZ that are not neighbors of either a,ba,b. Then in Z1Z_{1} there are at least 4k[t+2(kt2)+k+1]=k+t+34k-[t+2(k-t-2)+k+1]=k+t+3 vertices. We claim that there is an edge from a vertex in Z1Z_{1} to a vertex that is not in CN{x}C\cup N\cup\{x\}. If not, then all edges incident with Z1Z_{1} have one end in Z1Z_{1} and the other in CNC\cup N. In this case, there are at least (k+t+3)k(k+t+3)k edges incident with Z1Z_{1} and at most t(k2)+(k2)(k1)t(k-2)+(k-2)(k-1) possible edges from CNC\cup N to Z1Z_{1}. If the former number was actually less than the latter, we would have that 6k+2t26k+2t\leq 2, which is impossible. Therefore, there is an edge joining a vertex rZ1r\in Z_{1} to a vertex sNs\not\in N that is not a neighbor of at least one among a,ba,b. Without loss, assume that ss is not a neighbor of bb.

We perform a 2-switch to GG removing the edges {a,b}\{a,b\}, {r,s}\{r,s\} and adding the edges {a,r},{b,s}\{a,r\},\{b,s\}, obtaining the graph GG^{\prime}. Since neither of r,sr,s is in NN, the graph GG^{\prime} has one less edge in NN than GG. We claim that xx is adjacent to no less cotriangles in GG^{\prime} than it was in GG. Suppose that T={u,v,w}T=\{u,v,w\} was a cotriangle in GG but is not a cotriangle in GG^{\prime}. The only way that can happen is if exactly one of the new edges involves two vertices from TT. Suppose u=a,v=ru=a,v=r. Then ww could have been any of the other elements in NN different from aa and bb and not a neighbor of rr. That is, in the process of going from GG to GG^{\prime}, the vertex xx loses at most k2k-2 adjacent cotriangles. On the other hand, any set of the form {a,b,z}\{a,b,z\} with zZ1{r,s}z\in Z_{1}-\{r,s\} was not a cotriangle in GG but is a cotriangle in GG^{\prime}. This means that the vertex xx gains at least k+t+1k+t+1 new adjacent cotriangles in the process of going from GG to GG^{\prime}.

Because of the argument of the previous paragraph, we may now assume that the kk-regular graph GG is such that there are no edges among the vertices in N=NG(x)N=N_{G}(x). Then xx is adjacent to exactly (k3)\binom{k}{3} cotriangles where the three vertices of the cotriangle are contained in NN. We will estimate the amount of cotriangles adjacent to xx where exactly two vertices of the cotriangle are elements of NN. For each pair abab of vertices of NN, denote by nabn_{ab} the number of common neighbors of a,ba,b different from xx. Then there are 2k2nab2k-2-n_{ab} vertices different from xx that are neighbors of either aa or bb. Hence there are n(2k2nab)k1=n3k+1+nabn-(2k-2-n_{ab})-k-1=n-3k+1+n_{ab} vertices that can be used to extend the set {a,b}\{a,b\} to a cotriangle adjacent to xx of the required form. Summing over all pairs of vertices of NN we get S=(k2)(n3k+1)+nabS=\binom{k}{2}(n-3k+1)+\sum n_{ab}. We have that (k2)(n2k)S=(k2)(k1)nab\binom{k}{2}(n-2k)-S=\binom{k}{2}(k-1)-\sum n_{ab}. Since 0nabk10\leq n_{ab}\leq k-1 for each pair a,ba,b, this expression is non-negative, proving our first claim. It is equal to zero if and only if nab=k1n_{ab}=k-1 for each pair a,ba,b of vertices of NN. If this were the case, then the open neighborhood of each of the vertices in NN consists of the same vertices, therefore the connected component that contains xx is isomorphic to Kk,kK_{k,k}. ∎

5 Proof of the main theorem

Theorem 5.1.

If n>3k+2k2kn>3k+\sqrt{2k^{2}-k}, then the complement of any kk-regular graph with nn vertices is not Helly.

Proof.

Let k>1k>1 and nn as in the hypothesis of the Theorem. Then n>4kn>4k. Let GG be a kk-regular graph with nn vertices such that G¯\overline{G} is Helly. Let BB be the bipartite graph where one part is the set of vertices of GG and the other part is the set of cotriangles of GG. In BB, a vertex corresponding to a vertex in GG is a neighbor to the vertex corresponding to a cotriangle whenever the vertex is adjacent to the cotriangle. Denote with e(B)e(B) the amount of edges of BB. We will estimate e(B)e(B) from below and from above.

In the graph GG, each vertex is in at most (k2)\binom{k}{2} triangles. Therefore, there are at most n(k2)3=16nk(k1)\frac{n\binom{k}{2}}{3}=\frac{1}{6}nk(k-1) triangles. Since we have that t(G)+t(G¯)=(n3)12nk(n1k)t(G)+t(\overline{G})=\binom{n}{3}-\frac{1}{2}nk(n-1-k) by Lemma 4.1 , there are at least

Cn,k=(n3)12nk(n1k)16nk(k1)C_{n,k}=\binom{n}{3}-\frac{1}{2}nk(n-1-k)-\frac{1}{6}nk(k-1) (2)

cotriangles in GG. By Lemma 4.2, since each cotriangle is adjacent to at least kk vertices, we have that kCn,ke(B)kC_{n,k}\leq e(B).

On the other hand, by Lemma 4.3, the maximum amount of cotriangles adjacent to any fixed vertex in GG is Tn,k=(k2)(n2k)+(k3)T_{n,k}=\binom{k}{2}(n-2k)+\binom{k}{3}. Therefore, e(B)nTn,ke(B)\leq nT_{n,k}.

It follows that kCn,ke(B)nTn,kkC_{n,k}\leq e(B)\leq nT_{n,k}. Now, the difference nTn,kkCn,knT_{n,k}-kC_{n,k} is equal to kn6a(n,k)-\frac{kn}{6}\,a(n,k), where a(n,k)=n26kn+(7k2+k)a(n,k)=n^{2}-6kn+(7k^{2}+k). Hence, we must have a(n,k)0a(n,k)\leq 0. Fixing kk, and considering a(n,k)a(n,k) as a quadratic polynomial on nn, its roots are 3k±2k2k3k\pm\sqrt{2k^{2}-k}. From this, the theorem follows. ∎

6 The cubic case

Theorem 5.1 implies that the complement of a cubic graph is not Helly if the graph has at least 14 vertices. This bound is tight, since the graph K3,3K3,3K_{3,3}\cup K_{3,3} has 12 vertices and its complement is Helly. However, unlike the cases k=1k=1 and k=2k=2, in this case there are graphs that are not Helly but are convergent under the clique graph operator. As examples of that behavior we can give the graphs in Figure 1, which were found by a computer search. We conjecture that there are cubic graphs GG with an arbitrarily large number of vertices, such that G¯\overline{G} is Helly.

Refer to caption
(a) 14 vertices
Refer to caption
(b) 16 vertices
Refer to caption
(c) 18 vertices
Figure 1: Cubic graphs with convergent (non Helly) complements

7 Acknowledgments

Partially supported by CONACYT, grant A1-S-45528.

References

  • [1] C. Cedillo and M. A. Pizaña. Clique-convergence is undecidable for automatic graphs. Journal of Graph Theory, 96(3):414–437, 2021.
  • [2] G. Chartrand, L. Lesniak, and P. Zhang. Graphs & digraphs. Textbooks in Mathematics. CRC Press, Boca Raton, FL, sixth edition, 2016.
  • [3] M. C. Dourado, L. N. Grippo, and M. D. Safe. On the generalized Helly property of hypergraphs, cliques, and bicliques. arXiv, 2022. https://arxiv.org/abs/2201.12610.
  • [4] M. C. Dourado, F. Protti, and J. L. Szwarcfiter. Characterization and recognition of generalized clique-Helly graphs. Discrete Appl. Math., 155(18):2435–2443, 2007.
  • [5] M. C. Dourado, F. Protti, and J. L. Szwarcfiter. Complexity aspects of the Helly property: graphs and hypergraphs. Electron. J. Comb., DS17:53, 2009.
  • [6] F. F. Dragan. Centers of graphs and the Helly property. PhD thesis, Moldova State University, 1989.
  • [7] F. Escalante. Über iterierte Clique-Graphen. Abh. Math. Sem. Univ. Hamburg, 39:59–68, 1973.
  • [8] F. Harary. Graph theory. Addison-Wesley Publishing Co., Reading, Mass.-Menlo Park, Calif.-London, 1969.
  • [9] F. Larrión, C. P. de Mello, A. Morgana, V. Neumann-Lara, and M. A. Pizaña. The clique operator on cographs and serial graphs. Discrete Math., 282(1-3):183–191, 2004.
  • [10] F. Larrión, V. Neumann-Lara, and M. A. Pizaña. On expansive graphs. European J. Combin., 30(2):372–379, 2009.
  • [11] F. Larrión, M. A. Pizaña, and R. Villarroel-Flores. The clique behavior of circulants with three small jumps. Ars Combin., 113A:147–160, 2014.
  • [12] F. Larrión, M. A. Pizaña, and R. Villarroel-Flores. The clique operator on matching and chessboard graphs. Discrete Math., 309(1):85–93, 2009.
  • [13] M. Lin and J. L Szwarcfiter. Faster recognition of clique-Helly and hereditary clique-Helly graphs. Information Processing Letters, 103(1):40–43, 2007.
  • [14] G. Lorden. Blue-empty chromatic graphs. Amer. Math. Monthly, 69:114–120, 1962.
  • [15] V. Neumann-Lara. On clique-divergent graphs. Problèmes combinatoires et théorie des graphes, Orsay 1976, Colloq. int. CNRS No. 260, 313-315 (1978)., 1978.
  • [16] J. L. Szwarcfiter. Recognizing clique-Helly graphs. Ars Combin., 45:29–32, 1997.