Isolation of k-cliques
Peter Borg
Department of Mathematics
Faculty of Science
University of Malta
Malta
[email protected]
Kurt Fenech
Department of Mathematics
Faculty of Science
University of Malta
Malta
[email protected]
Pawaton Kaemawichanurat
Department of Mathematics
King Mongkut’s University of Technology Thonburi
Thailand
[email protected]
Abstract
For any positive integer k and any n-vertex graph G, let ι(G,k) denote the size of a smallest set D of vertices of G such that the graph obtained from G by deleting the closed neighbourhood of D contains no k-clique. Thus, ι(G,1) is the domination number of G. We prove that if G is connected, then ι(G,k)≤k+1n unless G is a k-clique or k=2 and G is a 5-cycle. The bound is sharp. The case k=1 is a classical result of Ore, and the case k=2 is a recent result of Caro and Hansberg. Our result solves a problem of Caro and Hansberg.
1 Introduction
Unless stated otherwise, we use small letters such as x to denote non-negative integers or elements of a set, and capital letters such as X to denote sets or graphs. The set of positive integers is denoted by N. For n∈{0}∪N, the set {i∈N:i≤n} is denoted by [n]. Note that [0] is the empty set ∅. Arbitrary sets are assumed to be finite. For a set X, the set of 2-element subsets of X is denoted by (2X) (that is, (2X)={{x,y}:x,y∈X,x=y}).
If Y is a subset of (2X) and G is the pair (X,Y), then G is called a graph, X is called the vertex set of G and is denoted by V(G), and Y is called the edge set of G and is denoted by E(G). A vertex of G is an element of V(G), and an edge of G is an element of E(G). We call G an n-vertex graph if ∣V(G)∣=n. We may represent an edge {v,w} by vw. If vw∈E(G), then we say that w is a neighbour of v in G (and vice-versa).
For v∈V(G), NG(v) denotes the set of neighbours of v in G, NG[v] denotes NG(v)∪{v}, and dG(v) denotes ∣NG(v)∣ and is called the degree of v in G. For S⊆V(G), NG[S] denotes ⋃v∈SNG[v] (the closed neighbourhood of S), G[S] denotes the graph (S,E(G)∩(2S)) (the subgraph of G induced by S), and G−S denotes G[V(G)\S] (the graph obtained from G by deleting S). We may abbreviate G−{v} to G−v. Where no confusion arises, the subscript G is omitted from any of the notation above that uses it; for example, NG(v) is abbreviated to N(v).
If G and H are graphs, f:V(H)→V(G) is a bijection, and E(G)={f(v)f(w):vw∈E(H)}, then we say that G is a copy of H, and we write G≃H. Thus, a copy of H is a graph obtained by relabeling the vertices of H.
For n≥1, the graphs ([n],(2[n])) and ([n],{{i,i+1}:i∈[n−1]}) are denoted by Kn and Pn, respectively. A copy of Kn is called a complete graph or an n-clique. A copy of Pn is called an n-path or simply a path.
If G and H are graphs such that V(H)⊆V(G) and E(H)⊆E(G), then H is called a subgraph of G, and we say that G contains H.
If F is a set of graphs and F is a copy of a graph in F, then we call F an F-graph. If G is a graph and D⊆V(G) such that G−N[D] contains no F-graph, then D is called an F-isolating set of G. Let ι(G,F) denote the size of a smallest F-isolating set of G. The study of isolating sets was introduced recently by Caro and Hansberg [1]. It is an appealing and natural generalization of the classical domination problem [2, 3, 4, 5, 6, 7]. Indeed, D is a {K1}-isolating set of G if and only if D is a dominating set of G (that is, N[D]=V(G)), so ι(G,{K1}) is the domination number of G (the size of a smallest dominating set of G). In this paper, we obtain a sharp upper bound for ι(G,{Kk}), and consequently we solve a problem of Caro and Hansberg [1].
We call a subset D of V(G) a k-clique isolating set of G if G−N[D] contains no k-clique. We denote the size of a smallest k-clique isolating set of G by ι(G,k). Thus, ι(G,k)=ι(G,{Kk}).
If G1,…,Gt are graphs such that V(Gi)∩V(Gj)=∅ for every i,j∈[t] with i=j, then G1,…,Gt are vertex-disjoint. A graph G is connected if, for every v,w∈V(G), G contains a path P with v,w∈V(P). A connected subgraph H of G is a component of G if, for each connected subgraph K of G with K=H, H is not a subgraph of K. Clearly, two distinct components of G are vertex-disjoint.
For n,k∈N, let an,k=⌊k+1n⌋ and bn,k=n−kan,k. We have an,k≤bn,k≤an,k+k. If n≤k, then let Bn,k=Pn. If n≥k+1, then let F1,…,Fan,k be copies of Kk such that Pbn,k,F1,…,Fan,k are vertex-disjoint, and let Bn,k be the connected n-vertex graph given by
[TABLE]
Thus, Bn,k is the graph obtained by taking Pbn,k,F1,…,Fan,k and joining i (a vertex of Pbn,k) to each vertex of Fi for each i∈[an,k].
For n,k∈N with k=2, let
[TABLE]
For n∈N, let
[TABLE]
In Section 2, we prove the following result.
Theorem 1.1
If G is a connected n-vertex graph, then, unless G is a k-clique or k=2 and G is a 5-cycle,
[TABLE]
Consequently, for any k≥1 and n≥3,
[TABLE]
A classical result of Ore [8] is that the domination number of a graph G with min{d(v):v∈V(G)}≥1 is at most 2n (see [4]). Since the domination number is ι(G,1), it follows by Lemma 2.2 in Section 2 that Ore’s result is equivalent to the bound in Theorem 1.1 for k=1. The case k=2 is also particularly interesting; while deleting the closed neighbourhood of a {K1}-isolating set yields the graph with no vertices, deleting the closed neighbourhood of a {K2}-isolating set yields a graph with no edges. In [1], Caro and Hansberg proved Theorem 1.1 for k=2, using a different argument. Consequently, they established that k+11≤limsupn→∞nι(n,k)≤31. In the same paper, they asked for the value of limsupn→∞nι(n,k). The answer is given by Theorem 1.1.
Corollary 1.2
For any k≥1,
[TABLE]
Proof. By Theorem 1.1, for any n≥3, k+11−(k+1)nk=n1(k+1n−k)≤nι(n,k)≤k+11, and, if n is a multiple of k+1, then nι(n,k)=k+11. Thus, limn→∞sup{pι(p,k):p≥n}=limn→∞k+11=k+11. □
2 Proof of Theorem 1.1
In this section, we prove Theorem 1.1. We start with two lemmas that will be used repeatedly.
If a graph G contains a k-clique H, then we call H a k-clique of G. We denote the set \{V(H)\colon H\mbox{ is a k-clique of }G\} by Ck(G).
Lemma 2.1
If v is a vertex of a graph G, then ι(G,k)≤1+ι(G−NG[v],k).
Proof. Let D be a k-clique isolating set of G−NG[v] of size ι(G−NG[v],k). Clearly, C∩NG[v]=∅ for each C∈Ck(G)\Ck(G−NG[v]). Thus, D∪{v} is a k-clique isolating set of G. The result follows. □
Lemma 2.2
If G1,…,Gr are the distinct components of a graph G, then ι(G,k)=∑i=1rι(Gi,k).
Proof. For each i∈[r], let Di be a smallest k-clique isolating set of Gi. Then, ⋃i=1rDi is a k-clique isolating set of G. Thus, ι(G,k)≤∑i=1r∣Di∣=∑i=1rι(Gi,k). Let D be a smallest k-clique isolating set of G. For each i∈[r], D∩V(Gi) is a k-clique isolating set of Gi, so ∣Di∣≤∣D∩V(Gi)∣. We have ∑i=1rι(Gi,k)=∑i=1r∣Di∣≤∑i=1r∣D∩V(Gi)∣=∣D∣=ι(G,k). The result follows. □
Proof of Theorem 1.1. We use induction on n. If G is a k-clique, then ι(G)=1=k+1n+1. If k=2 and G is a 5-cycle, then ι(G)=2=k+1n+1. Suppose that G is not a k-clique and that, if k=2, then G is not a 5-cycle. Suppose n≤2. If k≥3, then ι(G)=0. If k=2, then G≃K1, so ι(G)=0. If k=1, then G≃K2, so ι(G)=1=k+1n. Now suppose n≥3. If Ck(G)=∅, then ιk(G)=0. Suppose Ck(G)=∅. Let C∈Ck(G). Since G is connected and G is not a k-clique, there exists some v∈C such that N[v]\C=∅. Thus, ∣N[v]∣≥k+1 as C⊂N[v]. If V(G)=N[v], then {v} is a k-clique isolating set of G, so ι(G)=1≤k+1n. Suppose V(G)=N[v]. Let G′=G−N[v] and n′=∣V(G′)∣. Then,
[TABLE]
and V(G′)=∅. Let H be the set of components of G′. If k=2, then let H′={H∈H:H≃Kk}. If k=2, then let H′={H∈H:H≃Kk\mboxorH≃C5}. By the induction hypothesis, ι(H,k)≤k+1∣V(H)∣ for each H∈H\H′. If H′=∅, then, by Lemmas 2.1 and 2.2,
[TABLE]
Suppose H′=∅. For any H∈H and any x∈N(v), we say that H is linked to x if xy∈E(G) for some y∈V(H). Since G is connected, each member of H is linked to at least one member of N(v). One of Case 1 and Case 2 below holds.
Case 1: For each H∈H′, H is linked to at least two members of N(v). Let H′∈H′ and x∈N(v) such that H′ is linked to x. Let Hx be the set of members of H that are linked to x only. Then,
[TABLE]
and hence, by the induction hypothesis, each member H of Hx has a k-clique isolating set DH with ∣DH∣≤k+1∣V(H)∣.
Let X={x}∪V(H′) and G∗=G−X. Then, G∗ has a component Gv∗ with N[v]\{x}⊆V(Gv∗), and the other components of G∗ are the members of Hx. Let Dv∗ be a k-clique isolating set of Gv∗ of size ι(Gv∗,k). Since H′ is linked to x, xy∈E(G) for some y∈V(H′). If H′ is a k-clique, then let D′={y}. If k=2 and H′ is a 5-cycle, then let y′ be one of the two vertices in V(H′)\NH′[y], and let D′={y,y′}. We have X⊆N[D′] and ∣D′∣=k+1∣X∣. Let D=D′∪Dv∗∪⋃H∈HxDH. Since the components of G∗ are Gv∗ and the members of Hx, we have V(G)=X∪V(Gv∗)∪⋃H∈HxV(H), and, since X⊆N[D′], D is a k-clique isolating set of G. Thus,
[TABLE]
Subcase 1.1: Gv∗ is neither a k-clique nor a 5-cycle. Then, ∣Dv∗∣≤k+1∣V(Gv∗)∣ by the induction hypothesis. By (1), ι(G,k)≤k+11(∣V(Gv∗)∣+∣X∣+∑H∈Hx∣V(H)∣)=k+1n.
Subcase 1.2: Gv∗ is a k-clique. Since ∣N[v]∣≥k+1 and N[v]\{x}⊆V(Gv∗), we have V(Gv∗)=N[v]\{x}. If H′ is a k-clique, then let X′={y} and D′′={x}. If k=2 and H′ is a 5-cycle, then let X′ be the set whose members are y, y′, and the two neighbours of y′ in H′, and let D′′={x,y′}. Let Y=(X∪V(Gv∗))\({v,x}∪X′). Let GY=G−({v,x}∪X′). Then, the components of GY are the components of G[Y] and the members of Hx.
If G[Y] has no k-clique, then, since {v,x}∪X′⊆N[D′′], D′′∪⋃H∈HxDH is a k-clique isolating set of G, and hence
[TABLE]
This is the case if k=1 as we then have Y=∅.
Suppose that k≥2 and G[Y] has a k-clique CY. We have
[TABLE]
Thus, ∣V(CY)∩V(Gv∗)∣=∣V(CY)\(V(H′)\X′)∣≥k−(k−1)=1 and ∣V(CY)∩V(H′)∣=∣V(CY)\(V(Gv∗)\{v})∣≥k−(k−1)=1. Let z∈V(CY)∩V(Gv∗) and Z=V(Gv∗)∪V(CY). Since z is a vertex of each of the k-cliques Gv∗ and CY,
[TABLE]
We have
[TABLE]
Let GZ=G−Z. Then, V(GZ)={x}∪(V(H′)\V(CY))∪⋃H∈HxV(H). We have that the components of GZ−x are GZ[V(H′)\V(CY)] (which is a clique or a path, depending on whether H′ a k-clique or a 5-cycle) and the members of Hx, y∈V(H′)\V(CY) (by (2)), y∈NGZ[x], and, by the definition of Hx, NGZ(x)∩V(H)=∅ for each H∈Hx. Thus, GZ is connected, and, if Hx=∅, then GZ is neither a clique nor a 5-cycle.
Suppose Hx=∅. By the induction hypothesis, ι(GZ,k)≤k+1∣V(GZ)∣. Let DGZ be a k-clique isolating set of GZ of size ι(GZ,k). By (3), {z}∪DGZ is a k-clique isolating set of G. Thus, ι(G,k)≤1+ιk(GZ)≤1+k+1∣V(GZ)∣, and hence, by (4), ι(G,k)≤k+1∣Z∣+k+1∣V(GZ)∣=k+1n.
Now suppose Hx=∅. Then, G∗=Gv∗, so V(G)=V(Gv∗)∪{x}∪V(H′). Recall that either H′ is a k-clique or k=2 and H′ is a 5-cycle.
Suppose that H′ is a k-clique. Then, n=2k+1. By (3), ∣V(G−N[z])∣≤∣V(G−Z)∣=n−∣Z∣=2k+1−∣Z∣. Suppose ∣Z∣≥k+2. Then, ∣V(G−N[z])∣≤k−1, and hence {z} is a k-clique isolating set of G. Thus, ι(G,k)=1<k+1n. Now suppose ∣Z∣≤k+1. Then, by (4), ∣Z∣=k+1 and ∣V(CY)∩V(H′)∣=1.
Let z′ be the element of V(CY)∩V(H′), and let Z′=V(CY)∪V(H′). Since z′ is a vertex of each of the k-cliques CY and H′, Z′⊆N[z′]. We have ∣Z′∣=∣V(CY)∣+∣V(H′)∣−∣V(CY)∩V(H′)∣=2k−1 and ∣V(G−N[z′])∣≤∣V(G−Z′)∣=n−∣Z′∣=(2k+1)−(2k−1)=2. If k≥3, then {z′} is a k-clique isolating set of G, and hence ι(G)=1<k+1n. Suppose k=2. Then, H′, Gv∗, and CY are the 2-cliques with vertex sets {y,z′}, {v,z}, and {z,z′}, respectively. Thus, V(G)={v,z,z′,y,x}, and G contains the 5-cycle with edge set {xv,vz,zz′,z′y,yx}. Since G is not a 5-cycle, d(w)≥3 for some w∈V(G). Since ∣V(G−N[w])∣=5−∣N[w]∣≤1, {w} is a k-clique isolating set of G, and hence ι(G,k)=1<35=k+1n.
Now suppose that k=2 and H′ is a 5-cycle. Then, V(Gv∗)={v,z} and E(H′)={yy1,y1y2,y2y3,y3y4,y4y} for some y1,y2,y3,y4∈V(G). Recall that ∣V(CY)∩V(H′)∣≥1. Let z′∈V(CY)∩V(H′). Since z and z′ are vertices of CY, zz′∈E(G). We have V(G)={v,z,x,y,y1,y2,y3,y4}, N(v)={x,z}, z′∈{y1,y2,y3,y4} (as y∈/V(CY) by (2)), and {vx,vz,xy,zz′}∪E(H′)⊆E(G). If z′ is y1 or y4, then V(G−N[{y,z′}]) is {v,y3} or {v,y2}. If z′ is y2 or y3, then V(G−N[{y,z′}])={v}. Thus, {y,z′} is a k-clique reducing set of G, and hence ι(G,k)=2<38=k+1n.
Subcase 1.3: Gv∗ is a 5-cycle. If k=2, then the result follows as in Subcase 1.1. Suppose k=2. We have E(Gv∗)={vv1,v1v2,v2v3,v3v4,v4v} for some v1,v2,v3,v4∈V(G). Let Y={v2,v3,v4}. Recall that the components of G∗ are Gv∗ and the members of Hx. Thus, G−Y is connected and V(G−Y)={v,v1,x}∪V(H′)∪⋃H∈HxV(H).
Suppose that G−Y is not a 5-cycle. By the induction hypothesis, G−Y has a k-clique isolating set D with ∣D∣≤k+1∣V(G−Y)∣=3n−3=3n−1. Since Y⊆N[v3], {v3}∪D is a k-clique isolating set of G, so ι(G,k)≤3n=k+1n.
Now suppose that G−Y is a 5-cycle. Then, H′ is a 2-clique and V(G−Y)={v,v1,x,y,z}, where {z}=V(H′)\{y}. Since v1v,vx,xy,yz∈E(G−Y) and G−Y is a 5-cycle, E(G−Y)={v1v,vx,xy,yz,zv1}. We have V(G−N[{v,v1}])⊆{v3,y}. If v3y∈/E(G), then {v,v1} is a k-clique isolating set of G. If v3y∈E(G), then V(G−N[v,v3])⊆{z}, so {v,v3} is a k-clique isolating set of G. Therefore, ι(G,k)=2<38=k+1n.
Case 2: For some x∈N(v) and some H′∈H′, H′ is linked to x only. Let H1={H∈H′:H is linked to x only} and H2={H∈H\H′:H is linked to x only}. Let h1=∣H1∣ and h2=∣H2∣. Since H′∈H1, h1≥1. For each H∈H1, yH∈N(x) for some yH∈V(H). Let X={x}∪⋃H∈H1V(H).
For each k-clique H∈H1, let DH={x}. If k=2, then, for each 5-cycle H∈H1, let yH′ be one of the two vertices in V(H)\NH[yH], and let DH={x,yH′}. Let DX=⋃H∈H1DH. Then, DX is a k-clique isolating set of G[X]. If k=2, then DX={x}, so ∣DX∣=1≤k+11+k∣H1∣=k+1∣X∣. If k=2 and we let h1′=∣{H∈H1:H≃C5}∣, then ∣DX∣=1+h1′≤31+5h1′+2(h1−h1′)=k+1∣X∣.
Let G∗=G−X. Then, G∗ has a component Gv∗ with N[v]\{x}⊆V(Gv∗), and the other components of G∗ are the members of H2. By the induction hypothesis, ι(H,k)≤k+1∣V(H)∣ for each H∈H2. For each H∈H2, let DH be a k-clique isolating set of H of size ι(H,k).
If Gv∗ is a k-clique, then let Dv∗={x}. If k=2 and Gv∗ is a 5-cycle, then let v′ be one of the two vertices in V(Gv∗)\NGv∗[v], and let Dv∗={x,v′}. If neither Gv∗ is a k-clique nor k=2 and Gv∗ is a 5-cycle, then, by the induction hypothesis, Gv∗ has a k-clique isolating set Dv∗ with ∣Dv∗∣≤k+1∣V(Gv∗)∣.
Let D=Dv∗∪DX∪⋃H∈H2DH. By the definition of H1 and H2, the components of G−x are Gv∗ and the members of H1∪H2. Thus, D is a k-clique isolating set of G since x∈D, v∈V(Gv∗)∩N[x], and DX is a k-clique isolating set of G[X]. Let D′=DX∪⋃H∈H2DH and n∗=∣V(Gv∗)∣. We have
[TABLE]
If Gv∗ is a k-clique, then ∣D∣=∣D′∣<k+1n. If k=2 and Gv∗ is a 5-cycle, then
[TABLE]
If neither Gv∗ is a k-clique nor k=2 and Gv∗ is a 5-cycle, then ∣D∣=∣Dv∗∣+∣D′∣≤k+1n∗+k+1n−n∗=k+1n. □