Two bounds for generalized 3-connectivity of Cartesian product graphs
Hui Gao Benjian Lv Kaishun Wang
Sch. Math. Sci. & Lab. Math. Com. Sys.,
Beijing Normal University, Beijing, 100875, China
Abstract
The generalized k-connectivity κk(G) of a graph G, which was introduced by Chartrand et al.(1984) is a generalization of the concept of vertex connectivity. Let G and H be nontrivial connected graphs. Recently, Li et al. gave a lower bound for the generalized 3-connectivity of the Cartesian product graph G□H and proposed a conjecture for the case that H is 3-connected. In this paper, we give two different forms of lower bounds for the generalized 3-connectivity of Cartesian product graphs. The first lower bound is stronger than theirs, and the second confirms their conjecture.
Keywords: Connectivity, Generalized connectivity, Cartesian product
MSC 2010: 05C76 , 05C40.
††E-mail addresses: [email protected] (H. Gao), [email protected] (B. Lv),
[email protected] (K. Wang).
1 Introduction
All graphs in this paper are undirected, finite and simple. We refer to the book [1] for graph theoretic notations and terminology not described here. The generalized connectivity of a graph G, which was introduced by Chartrand et al. [2], is a natural generalization of the concept of vertex connectivity.
A tree T is called an S-tree if S⊆V(T). A family of S-trees T1, T2,…,Tr are internally disjoint if E(Ti)∩E(Tj)=ϕ and V(Ti)∩V(Tj)=S for any pair of integers i and j, where 1≤i<j≤r. We denote by κ(S) the greatest number of internally disjoint S-trees. For an integer k with 2≤k≤v(G), the generalized k-connectivity κk(G) are defined to be the least value of κ(S) when S runs over all k-subsets of V(G). Clearly, when k=2, κ2(G)=κ(G).
In addition to being a natural combinatorial notation, the generalized connectivity can be motivated by its interesting interpretation in practice. For example, suppose that G represents a network. If one considers to connect a pair of vertices of G, then a path is used to connect them. However, if one wants to connect a set S of vertices of G with ∣S∣≥3, then a tree has to be used to connect them. This kind of tree with minimum order for connecting a set of vertices is usually called Steiner tree, and popularly used in the physical design of VLSI, see [16]. Usually, one wants to consider how tough a network can be, for the connection of a set of vertices. Then, the number of totally independent ways to connect them is a measure for this purpose. The generalized k-connectivity can serve for measuring the capability of a network G to connect any k vertices in G.
Determining κk(G) for most graphs is a difficult problem. In [4], Li and Li derived that for any fixed integer k≥2, given a graph G and a subset S of V(G), deciding whether there are k internally disjoint trees connecting S, namely deciding whether κ(S)≥k is NP-complete. The exact value of κk(G) is known for only a small class of graphs. Examples are complete graphs [3], complete bipartite graphs [5], complete equipartition 3-partite graphs [6], star graphs [15], bubble-sort graphs [15], and connected Cayley graphs on Abelian groups with small degrees [17]. Upper bounds and lower bounds of generalized connectivity of a graph have been investigated by Li et al. [9, 10, 14] and Li and Mao [12]. And Li et al. investigated Extremal problems in [8, 7]. We refer the readers to [13] for more results.
In [9], Li et al studied the generalized 3-connectivity of Cartesian product graphs and showed the following result.
Theorem 1.1** ([9]).**
Let G and H be connected graphs such that κ3(G)≥κ3(H). The following assertions hold:
(i) if κ(G)=κ3(G), then κ3(G□H)≥κ3(G)+κ3(H)−1. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G), then κ3(G□H)≥κ3(G)+κ3(H). Moreover, the bound is sharp.
Later in [11], Li et al gave a better result when H becomes a 2-connected graph.
Theorem 1.2** ([11]).**
Let G be a nontrivial connected graph, and let H be a 2-connected graph. The following assertions hold:
(i) if κ(G)=κ3(G), then κ3(G□H)≥κ3(G)+1. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G), then κ3(G□H)≥κ3(G)+2. Moreover, the bound is sharp.
Also in [11], Li et al proposed a conjecture as follows:
Conjecture 1.3** ([11]).**
Let G be a nontrivial connected graph, and let H be a 3-connected graph. The following assertions hold:
(i) if κ(G)=κ3(G), then κ3(G□H)≥κ3(G)+2. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G), then κ3(G□H)≥κ3(G)+3. Moreover, the bound is sharp.
In this paper, we give two different forms of lower bounds for generalized 3-connectivity of Cartesian product graphs.
Theorem 1.4**.**
Let G and H be nontrivial connected graphs. Then κ3(G□H)≥min{κ3(G)+δ(H), κ3(H)+δ(G), κ(G)+κ(H)−1}.
Theorem 1.5**.**
Let G be a nontrivial connected graph, and let H be an l-connected graph. The following assertions hold:
(i) if κ(G)=κ3(G) and 1≤l≤7, then κ3(G□H)≥κ3(G)+l−1. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G) and 1≤l≤9, then κ3(G□H)≥κ3(G)+l. Moreover, the bound is sharp.
The paper is organized as follows. In Section 2, we introduce some definitions and notations. In Section 3, we give a proof of theorem 1.4, which induces theorem 1.1 and theorem 1.2, and confirms conjecture 1.3. In section 4, we discuss the problem which number the connectivity of H can be such that conjecture 1.3 still holds. And theorem 1.5 is our answer and there are counterexamples when l≥8 for κ(G)=κ3(G) and l≥10 for κ(G)>κ3(G).
2 Preliminaries
Let G and H be two graphs with V(G)={u1,u2,...,un} and V(H)={v1,v2,...,vm}, respectively. Let κ(G)=k, κ(H)=l, δ(G)=δ1, and δ(H)=δ2. And the discussion below is always based on the hypotheses.
Recall that the Cartesian product (also called the square product) of two graphs G and H, written as G□H, is the graph with vertex set V(G)×V(H), in which two vertices (u,v) and (u′,v′) are adjacent if and only if u=u′ and vv′∈E(H), or v=v′ and uu′∈E(G). By starting with a disjoint union of two graphs G and H and adding edges joining every vertex of G to every vertex of H, one obtains the join of G and H, denoted by G∨H.
For any subgraph G1⊆G, we use G1vj to denote the subgraph of G□H with vertex set {(ui,vj)∣ui∈V(G1)} and edge set {(ui1,vj)(ui2,vj)∣ui1ui2∈E(G1)}. Similarly, for any subgraph H1⊆H, we use H1ui to denote the subgraph of G□H with vertex set {(ui,vj)∣vj∈V(H1)} and edge set {(ui,vj1)(ui,vj2)∣vj1vj2∈E(H1)}. Clearly, G1vj≅G1, H1ui≅H1.
Let x∈V(G) and Y⊆V(G). An (x, Y)-path is a path which starts at x, ends at a vertex of Y, and whose internal vertices do not belong to Y. A family of k internally disjoint (x, Y)-paths whose terminal vertices are distinct is referred to as a k-fan from x to Y.
For some 1≤t≤⌊2k⌋ and s≥t+1, in G, a family {P1,P2,...,Ps} of s u1u2−paths is called an (s,t)-original-path-bundle with respect to (u1,u2,u3), if u3 are on t paths P1, …, Pt, and the s paths have no internal vertices in common except u3, as shown in figure 1.a. If there is not only an (s,t)-original-path-bundle {P1′,P2′,...,Ps′} with respect (u1,u2,u3), but also a family {M1,M2,...,Mk−2t} of k−2t internally disjoint (u3,X)-paths avoiding the vertices in V(P1′∪...∪Pt′)−{u1,u2,u3}, where X=V(Pt+1′∪...∪Ps′), then we call the family of paths {P1′,P2′,...,Ps′}∪{M1,M2,...,Mk−2t} an (s,t)-reduced-path-bundle with respect to (u1,u2,u3), as shown in figure 1.b.
In order to show our main results, we need the following theorems and lemmas.
Lemma 2.1**.**
([1, Fan Lemma])*
Let G be a k-connected graph, let x be a vertex of G, and let Y⊆V(G)∖{x}. Then there exists a k-fan in G from x to Y.*
Theorem 2.2**.**
([1, p.219])*
Let S be a set of three pairwise-nonadjacent edges in a simple 3-connected graph G. Then there is a cycle in G containing all three edges of S unless S is an edge cut of G.*
Theorem 2.3** ([10]).**
Let G be a connected graph with at least three vertices. If G has two adjacent vertices with minimum degree δ, then κ3(G)≤δ−1.
Theorem 2.4** ([10]).**
Let G be k-connected, and u1,u2,u3∈V(G). Then for some 0≤t≤⌊2k⌋, there exists a (k,t)-reduced-path-bundle {P1,P2,...,Pk}∪{M1,M2,...,Mk−2t} such that for any 1≤i≤k−2t, the terminal vertex of Mi is on Pt+i.
Theorem 2.5** ([10]).**
let G be a connected graph with n vertices. For every two integers k and r with k≥0 and r∈{0, 1, 2, 3}, if κ(G)=4k+r, then 3k+⌈2r⌉≤κ3(G)≤κ(G). Moreover, the bound is sharp.
Lemma 2.6** ([5]).**
Let a, b be integers such that a+b≥3 and a≤b. Then,
[TABLE]
Lemma 2.7** ([11]).**
Let C1, C2, …, Ck be cycles. then κ3(C1□C2□...□Ck)=2k−1.
3 One lower bound
Lemma 3.1**.**
If S={(u1, v1), (u2, v2), (u3, v3)}, then κ(S)≥k+l−1.
Proof.
Set S′={(ui, vj)∣i, j=1, 2, 3}.
Case 1 (G□H)(S′) is not isomorphic to C3□C3.
Since (G□H)(S′) is not isomorphic to C3□C3, either G({u1, u2, u3}) or H({v1, v2, v3}) is not isomorphic C3. Without loss of generality, suppose v1 is not adjacent to v2 in H. Since H is l-connected, there exist l internally disjoint paths in H from v1 to v2, say Pj, in which vij is adjacent v2, j=1, 2, …, l. Suppose v3 is not in Pj, j=1, 2, …, l−1. Also, according to Lemma 2.1, there exists an l-fan in H from v3 to {vi1, vi2, …, vil−1, v1}, say Qj, which is a v3vij-path, j=1, 2, …, l−1; and Ql, which is a v3v1-path. Set Tj=(Pj−v2)u1∪Gvij∪(v2vij)u2∪Qju3, j=1, 2, …, l−1. It is easy notice that, for j=1, 2, …, l−1, (Pj−v2)u1 connects (u1,v1) and (u1,vij); Gvij∪(v2vij)u2 connects (u1,vij), (u2,v2) and (u3,vij); and Qju3 connects (u3,vij) and (u3,v3). Thus, Tj connects (u1,v1), (u2,v2) and (u3,v3). Since G is k-connected, there exist k internally disjoint paths in G from u1 to u2, say Rj, in which uij′ is adjacent to u2, j=1, 2, …, k. Suppose ui1′, ui2′, …, uik−2′=u1 or u3;uik−1′=u3; and uik′=u1.
Case 1.1 uik′=u3
Since G is k-connected, there exists a k-fan in G from u3 to {ui1′, ui2′, …, uik−2′, u2, uik′}, say Sj, which is a u3uij′-path, j=1, 2, …, k−2, k; and Sk−1, which is a u3u2-path. Since H is l-connected, H−{vi1, vi2, …, vil−1} is connected. Set Tj′=(Rj−u2)v1∪(H−{vi1,vi2,...,vil−1})uij′∪(uij′u2)v2∪Sjv3, j=1, 2, …, k−2, k; and Tk−1′=Rk−1v1∪(H−{vi1,vi2,...,vil−1})u2∪(Sk−1)v3. It is easy to notice that, for j=1, 2, …, k−2, k, (Rj−u2)v1 connects (u1, v1) and (uij′, v1); (H−{vi1,vi2,...,vil−1})uij′∪(uij′u2)v2 connects (uij′, v1), (u2, v2) and (uij′, v3); and Sjv3 connects ((uij′, v3)) and (u3, v3). Thus Tj′ connects (u1, v1), (u2, v2) and (u3, v3). Similarly, so does Tk−1′. It is clear that T1, …, Tl−1, T1′, …, Tk′ are pairwise disjoint except the vertex set S. See figure 2.
Case 1.2 uik′=u3
Since G is k-connected, there exists a (k−1)-fan in G−u1 from u3 to {ui1′, ui2′, …, uik−2′, u2}, say Sj, which is a u3uij′-path, j=1, 2, …, k−2; and Sk−1, which is a u3u2-path. Set Tj′=(Rj−u2)v1∪(H−{vi1,vi2,...,vil−1})uij′∪(uij′u2)v2∪Sjv3, j=1, 2, …, k−2; Tk−1′=Rk−1v1∪(H−{vi1, vi2, …, vil−1})u2∪Sk−1v3; and Tk′=Qlu3∪(Rk−u2)v1∪Plu1∪Rk−1v2. It is easy to notice that Qlu3 connects (u3, v3) and (u3, v1); (Rk−u2)v1 connects (u3, v1) and (u1, v1); Plu1 connects (u1, v1) and (u1, v2); and Rk−1v2 connects (u1, v2) and (u2, v2). Thus, Tk′ connects (u1, v1), (u2, v2) and (u3, v3). It is clear that T1, …, Tl−1, T1′, …, Tk′ are pairwise disjoint except the vertex set S. See figure 3.
Case 2 (G□H)(S′)≅C3□C3.
Because of lemma 2.7, there exist 3 internally disjoint S-trees in (G□H)(S′), say Tj′′, j=1, 2, 3. Since H is l-connected, κ ((H−v3)−v1v2)≥l−2. Thus there exist l−2 internally disjoint v1v2-paths in (H−v3)−v1v2, say Pj, in which vij is adjacent of v2, j=1, 2, …, l−2. By Lemma 2.1, there exists an (l−2)-fan in H−v1−v2 from v3 to {vi1, vi2, …, vil−2}. Similarly to Case 1, we can construct l−2 internally disjoint S-trees T1, T2, …, Tl−2. Also, because κ(G)=k, another k−2 S-trees T1′, T2′, …, Tk−2′ can be constructed. It is clear that T1, …, Tl−2, T1′, …, Tk−2′, T1′′, T2′′, T3′′ are internally disjoint S-trees.
∎
Lemma 3.2**.**
If S={(u1, v1), (u1, v2), (u2, v1)}, then κ(S)≥k+l−1.
Proof.
Since H is l-connected, there exist l internally disjoint v1v2-paths in H, say Pj, in which vij is adjacent to v1, j=1, 2, …, l. Suppose vij=v2, j=1, 2, …, l−1. Set Tj=Pju1∪Gvij∪(v1vij)u2, j=1, 2, …, l−1. Since G is k-connected, there exist k internally disjoint u1u2-paths in G, say Qj, in which uij′ is adjacent to u1, j=1, 2, …, k. Suppose uij′=u2, j=1, 2, …, k−1. Set Tj′=Qjv1∪(H−{vi1,vi2,...,vil−1})uij′∪(u1uij′)v2, j=1, 2, …, k−1; and T′′=Plu1∪Qkv1. It is clear that T1, …, Tl−1, T1′, …, Tk−1′, T′′ are connected graphs containing S and are pairwise disjoint except S. See figure 4. ∎
Lemma 3.3**.**
If S={(u1, v1), (u2, v1), (u3, v2)}, then κ(S)≥k+l−1.
Proof.
Since H is l-connected, there exist l internally disjoint v1v2-paths in H, say Pj, in which vij is adjacent to v1, j=1, 2, …, l. Suppose vij=v2, j=1, 2, …, l−1. Set Tj=(v1vij)u1∪(v1vij)u2∪Gvij∪(Pj−v1)u3, j=1, 2, …, l−1. Since G is k-connected, there exist k internally disjoint u1u2-paths in G, say Qj, in which uij′ is adjacent u1, j=1, 2, …, k. Suppose uij′=u2, u3, j=1, 2, …, k−2. Due to Lemma 2.1, there exists a k-fan in G from u3 to {ui1′, ui2′, …, uik−2′, u1, u2}, say Rj, which is a u3uij′-path, j=1, 2, …, k−2; Rk−1, which is a u3u1-path; and Rk, which is a u3u2-path. Set Tj′=Qjv1∪(H−{vi1, vi2, …, vil−1})uij′∪Rjv2, j=1, 2, …, k−2; Tk−1′=Qk−1v1∪(H−{vi1, vi2, …, vil−1})u1∪Rk−1v2; and Tk′=Qkv1∪(H−{vi1, vi2, …, vil−1})u2∪Rkv2. It is clear that T1, …, Tl−1, T1′, …, Tk′ are connected graphs containing S and are pairwise disjoint except S. See figure 5. ∎
Lemma 3.4**.**
If S={(u1, v1), (u2, v1), (u3, v1)}, then κ(S)≥κ3(G)+δ(H).
Proof.
Let T1, T2, …, Tκ3(G) be the internally disjoint {u1, u2, u3}-trees in G. Let vi1, vi2, …, viδ(H) be neighbors of v1 in H. Set Tj′=(v1vij)u1∪(v1vij)u2∪(v1vij)u3∪Gvij, j=1, 2, …, δ(H). It is clear that T1v1, T2v1, …, Tκ3(G)v1, T1′, T2′, …, Tδ(H)′ are connected graphs containing S and are pairwise disjoint except S. ∎
From Lemma 3.1 to Lemma 3.4, without loss of generality, we discuss all positions of three vertices of G□H. Hence, theorem 1.4 is obvious.
Theorem 1.4. Let G and H be nontrivial connected graphs. Then κ3(G□H)≥min{κ3(G)+δ(H), κ3(H)+δ(G), κ(G)+κ(H)−1}.
Example 3.5**.**
Let a and b be integers such that a≥1 and b≥2. Then κ3(Ka+1□Kb)=κ3((Ka∨K2)□Kb)=a+b−2.
Proof.
According to theorem 2.3, κ3(Ka+1□Kb), κ3((Ka∨K2)□Kb)≤a+b−2. It is easy to see that κ(Kb)=b−1; κ3(Kb)=b−2, if b≥3; and κ(Ka∨K2)=κ3(Ka∨K2)=a. From lemma 3.1-3.4, we can see that κ3(Ka+1□Kb), κ3((Ka∨K2)□Kb)≥a+b−2.
∎
Due to theorem 2.4, δ(G)≥κ(G)≥κ3(G), δ(H)≥κ(H)≥κ3(H). Hence, theorem 1.4 induces theorem 1.1.
Corollary 3.6**.**
Let G be a nontrivial connected graph, and H an l-connected graph, where 1≤l≤5. The following assertions hold;
(i) if κ(G)=κ3(G), then κ3(G□H)≥κ3(G)+l−1. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G), then κ3(G□H)≥κ3(G)+l. Moreover, the bound is sharp.
Proof.
Due to theorem 2.4, κ3(H)≥l−1. Hence, κ3(H)+δ(G)≥κ(G)+l−1. And theorem 1.4 induces this corollary. Example 3.5 guarantees the sharpness.
∎
Obviously, corollary 3.6 induces theorem 1.2 and confirms conjecture 1.3. However, corollary 3.6 does not answer the question whether conjecture 1.3 still holds if H is l-connected for l≥6. And this is what will discuss in the next section.
4 Another lower bound
Lemma 4.1**.**
Let 0≤t≤⌊2l⌋, and S={(u1,v1),(u1,v2),(u1,v3)}. If there exists an (l,t)-reduced-path-bundle {P1,P2,...,Pl}∪{M1,M2,...,Ml−2t} in H such that for any 1≤i≤l−2t, the terminal vertex of Mi is on Pt+i, then
[TABLE]
Proof.
For any j∈{1,2,...,t}, Pj is divided into two paths by v3, denoted by Pj1 and Pj2, where Pj1 is a v1v3-path and Pj2 is a v3v2-path. Without loss of generality, suppose u1u2, u1u3, …, u1uδ1+1∈E(G); l(Pj1), l(Pj2), l(Pl−j+1)≥2, j=1, 2, …, t−1. Set Tj=(Mj∪Pt+j)u1, j=1, 2, …, l−2t.
Case 1 t≤2.
Set Tj′=(u1uj+1)v1∪(u1uj+1)v2∪(u1uj+1)v3∪Huj+1, j=1, 2, …, δ1. If t=0, then κ(S)≥l−2t+δ1=l+δ1. If t=1, then set T′′=P1u1. So κ(S)≥l−2t+δ1+1=l+δ1−1. If t=2, then set T1′′=(Pl∪P11)u1, T2′′=(P21∪P12)u1, and T3′′=(P22∪Pl−1)u1. So κ(S)≥l−2t+δ1+3=l+δ1−1.
Case 2 t≥3.
Since t≤⌊2l⌋, l≥6. Suppose v1vi1, v1vi2, v3vi3, v3vi4, v2vi5, v2vi6∈E(G), as shown in figure 6. It is easy to see that κ((H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6})≥l−3≥3. If {vi1vi2,vi3vi4,vi5vi6} is not an edge cut of (H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6}, then according to lemma 2.2, there is a cycle containing {vi1vi2,vi3vi4,vi5vi6} in (H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6}, which without loss of generality, we suppose to be vi1vi2C1vi3vi4C2vi5vi6C3vi1, where C1, C2, C3 are paths. Set T1′=P12u1∪(u1u2)v1∪(u1u2)vi3∪(v1vi2)u2∪C1u2, T2′=Plu1∪(u1u2)v3∪(u1u2)vi5∪(v3vi4)u2∪C2u2, and T3′=P11u1∪(u1u2)v2∪(u1u2)vi1∪(v2vi6)u2∪C3u2. Hence, we can find 3 internally disjoint S-trees in (v∈V(P1∪Pl)∪(u1u2)v)∪(P1∪Pl)u1∪Hu2.
Case 2.1 l≥7.
Since κ((H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6})≥l−3≥4, {vi1vi2,vi3vi4,vi5vi6} cannot be an edge cut of (H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6}.
Similar to the discussion above, if δ1≥t−2, without loss of generality, suppose we can find 3 internally disjoint S-trees in (v∈V(Pj∪Pl−j+1)∪(u1u1+j)v)∪(Pj∪Pl−j+1)u1∪Hu1+j respectively, j=1, 2, …, t−2. Set Tj′′=(u1uj)v1∪(u1uj)v2∪(u1uj)v3∪Huj, j=t, t+1, …, δ1+1; T1′′′=(Pl−t+2∪Pt−1,1)u1; T2′′′=(Pt1∪Pt−1,2)u1; and T3′′′=(Pt2∪Pl−t+1)u1. Hence, κ(S)≥l−2t+3(t−2)+δ1−t+2+3=l+δ1−1.
If δ1≤t−3, without loss of generality, suppose we can find 3 internally disjoint S-trees in (v∈V(Pj∪Pl−j+1)∪(u1u1+j)v)∪(Pj∪Pl−j+1)u1∪Hu1+j respectively, j=1, 2, …, δ1. Clearly, there exist at least ⌊23(t−δ1)⌋ internally disjoint S-trees in (j=δ1+1∪t(Pj1∪Pj2∪Pl−j+1))u1. Hence, κ(S)≥l−2t+3δ1+⌊23(t−δ1)⌋=l+δ1−⌈2t−δ1⌉.
Case 2.2 l=6.
Since 3≤t≤⌊2l⌋, t=3. If {vi1vi2,vi3vi4,vi5vi6} is not an edge cut of (H−{v1,v2,v3})∪{vi1vi2,vi3vi4,vi5vi6}, then similar to the case that l≥7 and δ1≥t−2, κ(S)≥5+δ1. From now on, suppose the contrary, that is κ(S)≤4+δ1. Then, ui1ui2, ui3ui4, ui5ui6∈E(H); and {vi1vi2,vi3vi4,vi5vi6} separates H−{v1,v2,v3} into two components C1 and C2, where we suppose vi1,vi3,vi5∈V(C1), and vi2,vi4,vi6∈V(C2). Suppose l(P31)≥2, and vi7 is adjacent to v1 in P31. Since κ(S)≤4+δ1, so vi1vi7, vi2vi7∈E(G), which induces that H−{v1,v2,v3}−{vi1vi2,vi3vi4,vi5vi6} is connected, a contradiction. Hence, l(P31)=1. Similarly, l(P32)=l(P4)=1. Suppose l(P11)≥3, and vi7 and vi8 are adjacent to v3 in P11 and P21 respectively. Since κ(S)≤4+δ1, vi7vi8∈E(H), which induces that H−{v1,v2,v3}−{vi1vi2,vi3vi4,vi5vi6} is connected, a contradiction. Hence, l(P11)=2. Similarly, l(P12)=l(P21)=l(P22)=l(P5)=l(P6)=2. Suppose vi1v2∈E(G). Then in H, κ({v1,v2,v3})≥5, and according to lemma 3.4, κ(S)≥5+δ1, a contradiction. Hence, vi1v2∈/E(G). Since d(vi1)≥6, there exists a vertex vi9∈V(C1)−{vi1,vi3,vi5} such that vi9vi1∈E(H). Since κ(H)=6, so there exists a 6-fan in H from vi9 to {v1,v2,v3,vi1,vi3,vi5}, disjoint from C2. Then in H, κ({v1,v2,v3})≥5, also a contradiction. ∎
Proposition 4.2**.**
Let S={(u1,v1),(u1,v2),(u1,v3)}. Then
[TABLE]
Proof.
Since H is l-connected, so according to theorem 2.4, for some 0≤t≤⌊2l⌋, there exists an (l,t)-reduced-path-bundle {P1,P2,...,Pl}∪{M1,M2,...,Ml−2t} such that for any 1≤i≤l−2t, the terminal vertex of Mi is on Pt+i. When δ1≥⌊2l⌋−2, Since t≤⌊2l⌋, δ1≥t−2. Hence, due to lemma 4.1, κ(S)≥l+δ1−1. When δ1≤⌊2l⌋−3, if δ1≥t−2, then κ(S)≥l+δ1−1≥l+δ1−⌈2⌊2l⌋−δ1⌉; if δ1≤t−3, according to lemma 4.1, κ(S)≥l+δ1−⌈2t−δ1⌉≥l+δ1−⌈2⌊2l⌋−δ1⌉. ∎
Theorem 1.5 Let G be a nontrivial connected graph, and let H be an l-connected graph. The following assertions hold:
(i) if κ(G)=κ3(G) and 1≤l≤7, then κ3(G□H)≥κ3(G)+l−1. Moreover, the bound is sharp;
(ii) if κ(G)>κ3(G) and 1≤l≤9, then κ3(G□H)≥κ3(G)+l. Moreover, the bound is sharp.
Proof.
Let S be a 3-vertex-set of V(G□H). If there does not exist a vertex u∈V(G) such that S⊆V(Hu), then according to lemma 3.1-lemma 3.4, κ(S)≥κ3(G)+δ(H)≥κ3(G)+l or κ(S)≥κ(G)+l−1. Next, without loss of generality, suppose S={(u1,v1),(u1,v2),(u1,v3)}. If l≤7, then δ1≥⌊2l⌋−2, which according to proposition 4.2, induces that κ(S)≥l+δ1−1. If l=8 or 9 and κ(G)>κ3(G), then δ1≥2=⌊2l⌋−2, which according to proposition 4.2, induces that κ(S)≥l+δ1−1≥κ3(G)+l. Example 3.5 guarantees the sharpness. ∎
An S-tree T in G is said to be minimal if for any S-tree T′, V(T′)⊆V(T) and ∂(S)∩E(T′)⊆∂(S)∩E(T) induce V(T′)=V(T) and ∂(S)∩E(T′)=∂(S)∩E(T), where ∂(S) denotes the set of edges with one vertex in S.
Example 4.3**.**
Let a, b, c be integers such that 1≤a≤b≤c. Then,
[TABLE]
Proof.
It is easy to see that κ3(K1,1,1)=1 and κ3(K1,1,c)=2 for c≥2. Next, suppose a+b≥3. Let V1={v11,v21,...,va1}, V2={v12,v22,...,vb2} and V3={v13,v23,...,vc3} be the maximal independent sets of Ka,b,c. Let S be a 3-vertex-set of V(Ka,b,c). If S⊆V1∪V2, then κ(S)≥c+κ3(Ka,b); If S⊆V1∪V3, then κ(S)≥b+κ3(Ka,c); If S⊆V2∪V3, then κ(S)≥a+κ3(Kb,c). Moreover, equalities can hold in the three inequalities above. Next, without loss of generality, suppose S={v11,v12,v13}. Then figure 7 shows three types of minimal S-trees, namely A, B and C. Let T be the set of κ(S) internally disjoint minimal S-trees, containing u trees of type A (or A trees), v trees of type B (or B trees) and w trees of type C (or C trees). We notice that v+2w≤3 and 2u+v≤a+b+c−3. Hence, κ(S)=u+v+w≤⌊2a+b+c⌋. Besides, κ(S)≤d(v13)=a+b. Set T1=(v11v12)∪(v12v13) and T2=(v11v13)∪(v11v23)∪(v12v23). If a+b≥c+1, then it is easy to see that there exist ⌊2a−1+b−1+c−2⌋=⌊2a+b+c⌋−2 internally disjoint A trees in Ka,b,c−v23−{v11v12,v11v13,v12v13}. So κ(S)=⌊2a+b+c⌋. If a+b≤c, then it is easy to see that there exist a−1+b−1=a+b−2 internally disjoint A trees in Ka,b,c−v23−{v11v12,v11v13,v12v13}. So κ(S)=a+b. Finally, because of lemma 2.6, the problem is solved by comparing sizes. ∎
Example 4.4**.**
Let a and b be integers such that a≥1 and b≥2. Then,
[TABLE]
and
[TABLE]
Proof.
Since the proofs are quite similar for Kb□Ka,a,a and Kb□Ka,a+1,a+1, we give our proof only for the latter. Let V(Kb)={u1,u2,...,ub}; Let V1={v11,v21,...,va1}, V2={v12,v22,...,va+1,2} and V3={v13,v23,...,va+1,3} be the maximal independent sets of Ka,a+1,a+1; and Let S be a 3-vertex-set of V(Kb□Ka,a+1,a+1). Because of theorem 2.3, κ3(Kb□Ka,a+1,a+1)≤2a+b−1. If for any u∈V(G), S⊈V(Hu), then due to lemma 3.1-3.4, κ(S)≥2a+b−1; If there exists a vertex u∈V(G) such that for some i∈{1,2,3}, S⊆(V(H)−Vi)u, then according to the proof of lemma 3.4, κ(S)≥2a+b−1. Next, without loss of generality, let S={(u1,v11),(u1,v12),(u1,v13)}. Then according to proposition 4.2, we have
[TABLE]
Hence, when b≥a−1, κ3(Kb□Ka,a+1,a+1)=2a+b−1. Next, we claim that when b≤a−2, κ(S)≤⌊23a+3b−1⌋.
Let F be the graph obtained from Kb□Ka,a+1,a+1 by joining every pair of nonadjacent vertices in {(u,v)∣u∈V(Kb)−u1,v∈V(Ka,a+1,a+1)}. Obviously, the maximum number of internally disjoint S-trees in F is not less than the maximum number of internally disjoint S-trees in Kb□Ka,a+1,a+1. In F, all types of minimal S-trees are shown in figure 7 and figure 8. Let T be the set with maximum number of internally disjoint minimal S-trees in F and as many F trees as possible. Let the number of A, B, C, D, E and F trees T contains be u, v, w, x, y and z, denoted by T=uA+vB+wC+xD+yE+zF. It is easy to see that one A tree and one D tree can be replaced by two F trees, denoted by A+D→2F. So A trees and D trees can not be both in T. Similarly, we have A+E→B+F. If x>0 or y>0, then u=0. We notice that 3x+2y+z≤3(b−1) and v+2w+y≤3, so ∣T∣=v+w+x+y+z≤3b<⌊23a+3b−1⌋. If x=y=0, then z=3(b−1) and ∣T∣=3(b−1)+κ3(Ka−b+1,a−b+2,a−b+2)=⌊23a+3b−1⌋. ∎
Also, we can similarly prove that when a≥4, κ3(P3□Ka,a,a)=κ3(K2□Ka,a,a)=⌊23a+3⌋, and κ3(P3□Ka,a+1,a+1)=κ3(K2□Ka,a+1,a+1)=⌊23a+5⌋. In theorem 1.5, let G=P3, and H=Ka,a,a or Ka,a+1,a+1. Then l≥8, and κ3(G□H)<κ3(G)+l−1. When a≥5, κ3(K3□Ka,a,a)=⌊23a+6⌋, and κ3(K3□Ka,a+1,a+1)=⌊23a+8⌋. In theorem 1.5, let G=K3, and H=Ka,a,a or Ka,a+1,a+1. Then l≥10, and κ3(G□H)<κ3(G)+l.