1.
Introduction
Graph theory can be used to analyze the connections between things, arguably in the most general sense, because it is based on set theory. Social networks, traffic networks, power networks, distributed control systems, molecular structures, interconnection networks and so on, any system you can think of that involves connections between things can be modeled using graphs. Faults are inevitable in these application. Many researchers have done much works to improve the reliability of these systems, see [1,2,3,4]. The stability of systems is only under very strict conditions. Maintain systems stability in case of failure brings great challenges to theoretical research. And because of the complexity of many systems in engineering applications. Many systems are modeling of interconnected systems for security and reliability. Therefore, effective fault-tolerant control techniques are urgently needed. For this purpose, two methods are commonly used. One method is the algorithm in the proof, various algorithms are used in a lot of works, such as [5,6,7,8,9,10,11]. The other method is the discussion of whether the connectivity condition is satisfied after removing some vertices or edges in the topology structure of the interconnection network, this method is used by a large number of researchers, such as [12,13,14,15]. In this paper, we use the second method to discuss the application of graph theory to the interconnection network. An interconnection network can be modeled as an undirected graph G. The edge connectivity is one of the basic parameters to estimate the fault tolerance of the interconnection network. The edge-disjoint paths relate closely with the edge connectivity. In this paper, we discuss the edge-fault-tolerant strong Menger edge connectivity of a special graph.
Let G=(V(G),E(G)) be an undirected simple connected graph, where V(G) is the vertex set and E(G) is the edge set. For any vertex u∈V(G), NG(u) denotes the set of neighbors of u and EG(u) denotes the set of edges which are associated with u. dG(u)=|EG(u)| denotes the degree of u and δ(G)=min{dG(u):u∈V(G)} denotes the minimum degree of vertices in G. G is called k-regular when dG(u)=k for each u∈V(G). Let F⊆V(G) be a vertex set, Fe⊆E(G) be an edge set, and EG(F) be the set of edges with only one end in F. Let G−F be the subgraph of G whose vertex set is V(G)−F and edge set is E(G)−{uv∈E(G):{u,v}∩F≠∅}. Let G−Fe be the subgraph of G whose vertex set is V(G) and edge set is E(G)−Fe. When G−F is disconnected or has only 1 vertex, F is called a vertex cut of G. When G−Fe is disconnected, Fe is called an edge cut of G. The (vertex) connectivity κ(G) of G is the minimal cardinality of F. The edge connectivity λ(G) of G is the minimal cardinality of Fe. Pk=ux1x2⋯xk−2v denotes a (u,v)-path on k distinct vertices u,x1,x2,⋯,xk−2,v. Let P1=ux1x2⋯xkv and P2=uy1y2⋯ymv be two distinct (u,v)-paths of G. P1 and P2 are called disjoint (u,v)-paths when V(P1)∩V(P2)={u,v}. P1 and P2 are called edge-disjoint (u,v)-paths when E(P1)∩E(P2)=∅. If G−F has no (u,v)-path, then F⊆V(G)−{u,v} is a (u,v)-cut. If G−Fe has no (u,v)-path, then Fe⊆E(G) is a (u,v)-edge-cut. Much attention has been paid to the vertex connectivity and edge connectivity in recent years. Menger [16] proposed the local connectivity.
Theorem 1.1. [16] Let u,v∈V(G) be two distinct vertices. Then the minimal size of a (u,v)-edge-cut is equal to the maximum number of edge-disjoint (u,v)-paths.
On the base of Menger's Theorem, Oh et al. [10] proposed the strong Menger connectivity (the maximal local-connectivity) and Qiao et al. [12] proposed the strong Menger edge connectivity.
Definition 1.2. [12] Let u,v∈V(G) be two distinct vertices. If there are min{dG(u),dG(v)} edge-disjoint (u,v)-paths, then G is strongly Menger edge connected.
Since the vertices and edges may fault, the discussion of fault-tolerance of interconnection networks is crucial. Oh et al. [10,11] proposed the m-fault-tolerant strong Menger connectivity. Shih et al. [15] proposed the m-conditional fault-tolerant strong Menger connectivity.
Definition 1.3. Let m≥1 be an integer, Fe⊆E(G) be an edge set with |Fe|≤m.
(1) [17] If G−Fe is strongly Menger edge connected, then G is m-edge-fault-tolerant strongly Menger edge connected.
(2) [12] If G−Fe is strongly Menger edge connected for any edge set Fe with δ(G−Fe)≥2, then G is m-conditional edge-fault-tolerant strongly Menger edge connected.
The fault-tolerant strong Menger edge connectivity is used in many topologies of interconnection networks, such as the n-dimensional star graph Sn, the hypercube-like network HLn, the augmented cubes AQn, the bubble-sort star graph BSn, the folded hypercube FQn, the hypercube Qn and so on. Table 1 is a summary of some recent researches.
The n-dimensional bubble-sort graph Bn is an attractive topology structure of interconnection networks. It is regular and bipartite [23]. And it has been studied by many researchers, see [13,14,24,25,26]. From Table 2, we can see that Sn, Bn and BSn have the same number of vertices, but Bn has the largest diameter. The fault-tolerant strong Menger edge connectivity of Bn has not been studied on Bn. In this paper, we discuss the number of faulty edges that Bn can tolerant without failure. The main contributions are listed in the following:
(1) Discussing the minimum number of vertices in the biggest connected component of Bn when at most (2n−5) edges are removed from Bn.
(2) Proving that when Bn has (n−3) faulty edges, Bn is edge-fault-tolerant strongly Menger edge connected for n≥3.
(3) Using the inductive hypothesis to discuss the minimum number of vertices in the biggest connected component of Bn when at most (3n−8) edges are removed from Bn.
(4) Proving under the condition that each vertex in Bn has at least two neighbors is imposed, when Bn has (2n−6) faulty edges, Bn is (2n−6)-conditional edge-fault-tolerant strongly Menger edge connected for n≥5.
2.
The n-dimensional bubble-sort graph Bn
We will give the definition and propositions of the n-dimensional bubble-sort graph.
Let [1,n]={1,2,⋯,n}. Let (ij) be a transposition with i,j∈[1,n]. We use k1k2⋯kn to represent the permutation (12⋯nk1k2⋯kn). Then (12⋯i⋯j⋯n12⋯i⋯j⋯n)(ij)=(12⋯i⋯j⋯n12⋯j⋯i⋯n) (∗). In order to the convenience of discussion, the equation of (∗) is written as (12⋯i⋯ j⋯n)(ij)=(12⋯j⋯i⋯n).
Definition 2.1. [23] The n-dimensional bubble-sort graph Bn has the vertex set of the entire n! permutations of [1,n] for n≥2. Any two distinct vertices u,v∈V(Bn) are neighboring if and only if u=v(i,i+1) for 1≤i≤n−1. Each vertex u∈V(Bn) has the form of u=u1u2⋯un.
By Definition 2.1, Bn has n! vertices and it is (n−1)-regular. When n=3 and n=4, the graphs are given in Figure 1. Obviously, when the last position of each vertex u=u1u2⋯un∈V(Bn) has a fixed integer i∈[1,n], Bn can be decomposed into n sub-bubble-sort graphs B1n,B2n,⋯,Bnn along the last position. Then Bin≅Bn−1 for i∈[1,n]. For distinct i,j∈[1,n], when u∈V(Bin), u′=u(n−1,n)∈V(Bjn) is the outside neighbor of u, Eij(Bn)={uu′∈E(Bn):u∈V(Bin),u′∈V(Bjn)} is the cross-edge set of Bn. Let F1 and F2 be two distinct vertex sets of Bn, EBn(F1,F2) be the set of edges between F1 and F2. Let Fe⊆E(Bn), Fie=Fe∩E(Bin), F0e=Fe−∪ni=1Fie. Let F[t,s]e=Fte∪Ft+1e⋯∪Fse be an edge set, and B[t,s]n=Bn[V(Btn)∪V(Bt+1n)∪⋯∪V(Bsn)] be the subgraph of Bn induced by Btn,Bt+1n,⋯,Bsn, B[t,s]n−F[t,s]e=Bn[V(Btn−Fte)∪V(Bt+1n−Ft+1e)∪⋯∪V(Bsn−Fse)] be the subgraph of Bn induced by Btn−Fte,Bt+1n−Ft+1e,⋯,Bsn−Fse for t,s∈[1,n] and t<s.
Some useful propositions are given in the following.
Proposition 2.2. [25] Any vertex u∈V(Bin) has only one outside neighbor for all i∈[1,n].
Proposition 2.3. [25] |Eij(Bn)|=(n−2)! for distinct i,j∈[1,n].
Proposition 2.4. [25] Any two distinct vertices u,v∈V(Bin) have different outside neighbors for all i∈[1,n].
Proposition 2.5. [27] κ(Bn)=λ(Bn)=n−1 for n≥2.
3.
The edge-fault-tolerant strong Menger edge connectivity of Bn
In this part, we will discuss the edge-fault-tolerant strong Menger edge connectivity of Bn, firstly, one necessary conclusion is proved.
Lemma 3.1. Let Fe⊆E(Bn) with |Fe|≤2n−5 for n≥3. There is a connected component C of Bn−Fe with |V(C)|≥n!−1.
Proof. We prove this conclusion by induction on n. Let C be the largest connected component of Bn−Fe.
When n=3, |Fe|≤2×3−5=1, see from Figure 1, B3−Fe is connected. Then there is a connected component C of B3−Fe with |V(C)|=|V(B3)|=3!>3!−1=5. When n=4, |Fe|≤2×4−5=3, see from Figure 1, there is at most 1 vertex can be isolated. Then there is a connected component C of B4−Fe with |V(C)|≥4!−1=23. Assume that the conclusion holds for n−1, n≥5, we will prove that the conclusion holds for n. Without loss of generality, let |F1e|=max{|Fie|:1≤i≤n}. Let Ci be the largest connected component of Bin−Fie for all i∈[1,n].
Claim 1. For distinct i,j∈[1,n], if |Fie|≤n−3 and |Fje|≤n−3, then Bn[V(Bin−Fie)∪V(Bjn−Fje)] is connected.
By Proposition 2.5, λ(Bin)=n−2. Since |Fie|≤n−3 and |Fje|≤n−3, both Bin−Fie and Bjn−Fje are connected. Note that |F0e|≤|Fe|≤2n−5<(n−2)!=|Eij(Bn)| for n≥5, Bn[V(Bin−Fie)∪V(Bjn−Fje)] is connected by Proposition 2.3. The proof of Claim 1 is complete.
Case 1. |F1e|≤n−3.
Since |F1e|=max{|Fie|:1≤i≤n}, |Fie|≤n−3 and |F0e|≤|Fe|≤2n−5 for all i∈[1,n]. By Claim 1, Bn−Fe is connected. So |V(C)|=|V(Bn)|=n!>n!−1.
Case 2. n−2≤|F1e|≤2n−7.
Since |Fe|≤2n−5, |F[2,n]e|≤|Fe|−|F1e|≤(2n−5)−(n−2)=n−3 and |F0e|≤|Fe|−|F1e|≤n−3. So |Fje|≤n−3<n−2=λ(Bn−1) for all j∈[2,n]. By Claim 1, B[2,n]n−F[2,n]e is connected. By the assumption, there is an isolated vertex and a connected component C1 in B1n−F1e with |V(C1)|≥(n−1)!−1. Note that |EBn(V(C1),V(B2n))-Fe|≥|E12(Bn)|−|V(B1n)-V(C1)|−|F0e|≥(n−2)!-[(n−1)!−((n−1)!−1)]−(n−3)≥(n−2)!−n+2>0 for n≥5. By Proposition 2.2, Bn[V(C1)∪V(B[2,n]n−F[2,n]e)] is connected and it is C. So |V(C)|≥|V(C1)|+|V(B[2,n]n−F[2,n]e)|≥n!−1.
Case 3. 2n−6≤|F1e|≤2n−5.
In this case, |F[2,n]e|≤|Fe|−|F1e|≤(2n−5)−(2n−6)=1. |F0e|≤|Fe|−|F1e|≤1. So |Fje|≤1<n−2=λ(Bn−1) for all j∈[2,n] and n≥5. By Claim 1, B[2,n]n−F[2,n]e is connected. By Proposition 2.2, each vertex u∈V(B1n) has 1 outside neighbor in B[2,n]n, combining this with |F0e|≤1, there is at most one vertex in B1n can be isolated from the component C. So |V(C)|≥n!−1.
By Case 1–3, When Bn−Fe is disconnected, there is a largest connected component C of Bn−Fe with |V(C)|≥n!−1 when |Fe|≤2n−5.
Remark 3.2. The conclusion of Lemma 3.1 is optimal in that there is an edge set Fe⊆E(Bn) with |Fe|=2n−4 such that Bn−Fe has a connected component C with |V(C)|≤n!−2 for n≥4.
Let F={u,v} be a vertex set of Bn with uv∈E(Bn). Then |EBn(F)|=2n−4 by Proposition 2.5. Let Fe=EBn(F)⊆E(Bn). Then uv is isolated in Bn−F. So |V(C)|≤n!−2.
Theorem 3.3. Bn is (n−3)-edge-fault-tolerant strongly Menger edge connected for n≥3.
Proof. Let Se⊆E(Bn) be a faulty edge set with |Se|≤n−3. By Proposition 2.5, we have λ(Bn)=n−1 for n≥2. Since n−3<n−1, Bn−Se is connected. Let two distinct vertices u,v∈V(Bn), a=min{dBn−Se(u),dBn−Se(v)} and Ef⊆E(Bn)−Se with |Ef|≤a−1. By Definition 1.2, if Bn is (n−3)-edge-fault-tolerant strongly Menger edge connected, then there should be a edge-disjoint fault-free (u,v)-paths connect u and v. We prove this by contradiction, suppose that u and v are disconnected in Bn−Se by removing an edge set Ef. That means u and v are disconnected in Bn−Se−Ef. Since a=min{dBn−Se(u),dBn−Se(v)}≤κ(Bn)=n−1 and |Ef|≤a−1, we have |Ef|≤n−2. Note that dBn−Se(u)≤n−1, dBn−Se(v)≤n−1 and |Ef|≤n−2. Then |Se∪Ef|≤(n−3)+(n−2)=2n−5. By Lemma 3.1, Bn−Se−Ef has a connected component C with |V(C)|≥n!−1. Since |V(Bn)|=n!, combining this with u and v are disconnected in Bn−Se−Ef, we have |V(C)|=n!−1 and only one vertex can be isolated. Without loss of generality, we suppose that v is the isolated vertex. Then u∈V(C) and v∉V(C). Moreover, the edges which are attached to v belonging to Ef, that is EBn({v},NBn−Se(v))⊆Ef. This means |Ef|≥dBn−Se(v), which contradicts with |Ef|≤a−1≤dBn−Se(v)−1. Hence Bn is (n−3)-edge-fault-tolerant strongly Menger edge connected.
4.
The m-conditional edge-fault-tolerant strong Menger edge connectivity of Bn
In this part, we will discuss the m-conditional edge-fault-tolerant strong Menger edge connectivity of Bn, firstly, two necessary conclusions are proved.
Lemma 4.1. Let Fe⊆E(B5) with |Fe|≤7. There is a connected component C of B5−Fe with |V(C)|≥5!−2.
Proof. Let |F1e|=max{|Fie|:1≤i≤5}. Let Ci be the largest connected component of Bi5−Fie for all i∈[1,5].
Claim 1. If |Fie|≤2, |Fje|≤2 and |Eij(B5)|>|F0e|, then B5[V(Bi5−Fie)∪V(Bj5−Fje)] is connected for distinct i,j∈[1,5].
By Proposition 2.5, λ(B5)=3, then both Bi5−Fie and Bj5−Fje are connected for distinct i,j∈[1,5]. Since |Eij(B5)|=(5−2)!=6>|F0e| for distinct i,j∈[1,5], B5[V(Bi5−Fie)∪V(Bj5−Fje)] is connected by Proposition 2.3. The proof of claim 1 is complete.
Case 1. |F1e|≤2.
In this case, |F0e|≤|Fe|−|F1e|≤7 and |Fie|≤7 for all i∈[1,5]. By Proposition 2.5, Bi5−Fie is connected for i∈[1,5]. Since |Eij(B5)|=(5−2)!=6 by Proposition 2.3 for distinct i,j∈[1,5] and |F0e|≤7, at most one Eij(B5)−Fe=∅ for distinct i,j∈[1,5]. Without loss of generality, we suppose that E12(B5)−Fe=∅. Then only B15−F1e is not connected to B25−F2e, but both B15−F1e and B25−F2e are connected to Bs5−Fse for all s∈[3,5]. So B5−Fe is connected and |V(C)|=|V(B5)|=5!>5!−2.
Case 2. 3≤|F1e|≤7.
In this case, |F0e|≤|Fe|−|F1e|≤7−3=4 and |F[2,5]e|≤|Fe|−|F1e|≤4.
Subcase 2.1. |Fje|≤2, for all j∈[2,5].
In this subcase, Bj5−Fje is connected by Proposition 2.5 for all j∈[2,5]. Since |Est(B5)|=(5−2)!=6>4≥|F0e| for distinct s,t∈[2,5], B[2,5]5−F[2,5]e is connected by Claim 1. If B15−F1e is connected, this is similar to Case 1, we have B5−Fe is connected and |V(C)|=|V(B5)|=5!>5!−2. If B15−F1e is disconnected, we assume on the contrary that |V(C)|≤5!−3, then there are at least three distinct vertices u1,u2,u3∈V(B15−F1e) but u1,u2,u3∉V(C). By Propositions 2.2 and 2.4, there are three distinct vertices u′1,u′2,u′3∈V(B[2,5]5)⊆V(C) and {u1u′1,u2u′2,u3u′3}⊆E(B5). Furthermore, {u1u′1,u2u′2,u3u′3}⊆F0e. At this moment, |F0e|≥3, |F1e|≤|Fe|−|F0e|≤7−3=4. Combining this with |F0e|≤4, we have 3≤|F0e|≤4 and 3≤|F1e|≤4. When |F1e|=3, there is only one vertex can be isolated in B15−F1e (Figure 1), which contradicts with the assumption. When |F1e|=4, we have |F0e|=3, a K2 or a 4-cycle can be isolated in B15−F1e (Figure 1.), but only the 4-cycle can satisfy the assumption. By Propositions 2.2 and 2.4, each vertex of the 4-cycle has an outside neighbor in B[2,5]5 and the outside neighbors are distinct, then |F0e|≥4, which contradicts with |F0e|=3. So the assumption does not hold. Therefore, |V(C)|≥5!−2.
Subcase 2.2. |Fje|≥3, for some j∈[2,5].
Note that |Fe|≤7 and |F1e|≥3. There is only one Fje can satisfy |Fje|≥3 for all j∈[2,5]. Otherwise, if there are two Fje's satisfy |Fje|≥3, then |Fe|≥3×3=9>7, a contradiction. Without loss of generality, we suppose that |F2e|≥3. Since |F1e|=max{|Fie|:1≤i≤5}, we have |F2e|=3. Otherwise, if |F2e|=4, then |Fe|≥2×4=8>7, a contradiction. Then 3≤|F1e|≤4, |F[3,5]e|≤1 and |F0e|≤1. Note that |Fse|≤1 and |Est(B5)|=(5−2)!=6>1≥|F0e| for distinct s,t∈[3,5], B[3,5]5−F[3,5]e is connected by Claim 1.
Subcase 2.2.1. Bi5−Fie is connected for all i∈[1,2].
In this subcase, both B15−F1e and B25−F2e are connected. Since |Ei3(B5)|=6>1≥|F0e| for all i∈[1,2], B5−Fe is connected. So |V(C)|=|V(B5)|=5!>5!−2.
Subcase 2.2.2. Only B15−F1e is disconnected.
In this subcase, B25−F2e is connected. Since |E23(B5)|=6>1≥|F0e|, B[2,5]5−F[2,5]e is connected by Proposition 2.3. This is similar to Case 2.1, we have |V(C)|≥5!−2.
Subcase 2.2.3. Only B25−F2e is disconnected.
In this subcase, B15−F1e is connected. Note that |F2e|=3=2×4−5. There is a connected component C2 of B25−F2e with |V(C2)|≥4!−1 by Lemma 3.1. Since |E13(B5)|=6>1≥|F0e|, B5[V(B15−F1e)∪V(B[3,5]5−F[3,5]e)] is connected by Proposition 2.3. Since |EB5(V(C2),V(B35))−Fe|≥|E23(B5)| -|V(B25)−V(C2)|−|F0e|≥(5−2)!-[4!−(4!−1)]−1=4>0, B5[V(C2)∪V(B15-F1e)∪V(B[3,5]5−F[3,5]e)] is connected and it is C. So |V(C)|≥5!−1>5!−2.
Subcase 2.2.4. Bi5−Fie is disconnected for all i∈[1,2].
In this subcase, B15−F1e is disconnected and B25−F2e is disconnected. Since |V(C2)|≥4!−1 by Lemma 3.1, there is at most one isolated vertex u2∈V(B25−F2e−V(C2)). Note that 3≤|F1e|≤4 and B15−F1e is disconnected. Then B15−F1e−V(C1) can only be one isolated vertex u1 with 3≤|F1e|≤4, a K2=u1v1 with |F1e|=4 or a 4-cycle with |F1e|=4. When B15−F1e−V(C1) is a K2 or a 4-cycle, |F1e|=4, this moment, |F0e|= |Fe|−|F1e|−|F2e|= 7−4−3=0. |EB5(V(Ci),V(B35))−Fe|≥|E13(B5)| - |V(Bi5)−V(Ci)|−|F0e|≥(5−2)!−[4!−(4!−4)]−1 = 1>0 for all i∈[1,2], B5[V(Ci)∪V(B[3,5]5−F[3,5]e)] is connected by Proposition 2.3. Then B5[V(C1)∪V(C2)∪V(B[3,5]5−F[3,5])e)] is connected and it is C. By Proposition 2.2 and Proposition 2.4, each vertex in B15−F1e−V(C1) has an outside neighbor and the outside neighbors are distinct, combining this with B25−F2e−V(C2) has only 1 isolated vertex u2, B5−Fe is connected. So |V(C)|=|V(B5)|=5!>5!−2. When B15−F1e−V(C1) is an isolated vertex u1, |V(C1)|≥4!−1. Since |EB5(V(Ci),V(B35))-Fe|≥|E13(B5)|−|V(Bi5)−V(Ci)|−|F0e|≥(5−2)!- [4!−(4!−1)]−1=4>0 for all i∈[1,2], B5[V(Ci)∪V(B[3,5]5−F[3,5]e)] is connected by Proposition 2.3. Then B5[V(C1)∪V(C2)∪V(B[3,5]5−F[3,5])e)] is connected and it is C. There is an isolated vertex in Bi5−Fie−V(Ci) for all i∈[1,2], so |V(C)|≥5!−2.
Lemma 4.2. Let Fe⊆E(Bn) with |Fe|≤3n−8 for n≥5. There is a connected component C of Bn−Fe with |V(C)|≥n!−2.
Proof. We prove this conclusion by induction on n. When n=5, the conclusion holds by Lemma 4.1. Assume that the conclusion holds for n−1 with n≥6, we will prove that the conclusion holds for n. Let |F1e|=max{|Fie|:1≤i≤n}. Let Ci be the largest connected component of Bin−Fie for all i∈[1,n].
Claim 1. If |Fie|≤n−3, |Fje|≤n−3 and |Eij(Bn)|>|F0e|, then Bn[V(Bin−Fie)∪V(Bjn−Fje)] is connected for distinct i,j∈[1,n] and n≥6.
By Proposition 2.5, λ(Bin)=n−2. Since |Fie|≤n−3 and |Fje|≤n−3, both Bin−Fie and Bjn−Fje are connected for distinct i,j∈[1,n]. Since |Eij(Bn)|=(n−2)!>|F0e| for distinct i,j∈[1,n] and n≥6, Bn[V(Bin−Fie)∪V(Bjn−Fje)] is connected by Proposition 2.3. The proof of claim 1 is complete.
Case 1. |F1e|≤n−3.
Since |F1e|≤n−3, |Fie|≤n−3 for all i∈[1,n] and |F0e|≤|Fe|−|F1e|≤3n−8. Note that |Eij(Bn)|=(n−2)!>3n−8≥|F0e| for n≥6. We have Bn−Fe is connected by Claim 1. Thus, |V(C)|=|V(Bn)|=n!>n!−2.
Case 2. n−2≤|F1e|≤2n−7.
In this case, |F0e|≤|Fe|−|F1e|≤(3n−8)−(n−2)=2n−6 and |F[2,n]e|≤|Fe|−|F1e|≤2n−6.
Subcase 2.1. |Fje|≤n−3, for all j∈[2,n].
Since |Est(Bn)|=(n−2)!>2n−6≥|F0e| for distinct s,t∈[2,n] and n≥6, Bn[V(Bsn−Fse)∪V(Btn−Fte)] is connected by Claim 1. Then B[2,n]n−F[2,n]e is connected and it is a subgraph of C. Since n−2≤|F1e|≤2n−7, we have |V(C1)|≥(n−1)!−1 by Lemma 3.1. Since |EBn(V(C1),V(B[2,3]n))−Fe|≥|E12(Bn)| +|E13(Bn)|−2|V(B1n)−V(C1)|−|F0e|≥2(n−2)!−2[(n−1)!−((n−1)!−1)]−(2n−6) =2(n−2)!−2n+4>0 for n≥6, we have Bn[V(C1)∪V(B[2,n]n−F[2,n]e)] is connected and it is C. So |V(C)|≥n!−1>n!−2.
Subcase 2.2. n−2≤|Fje|≤2n−7, for some j∈[2,n].
Since |Fe|≤3n−8, there is at most one Fje can satisfy n−2≤|Fje|≤2n−7 for some j∈[2,n]. Otherwise, if there are two Fje's satisfy n−2≤|Fje|≤2n−7 for some j∈[2,n], then |Fe|≥3(n−2)=3n−6>3n−8, a contradiction. Without loss of generality, we suppose that n−2≤|F2e|≤2n−7, then |F0e|≤|Fe|−2|F1e|≤(3n−8)−2(n−2)=n−4, |Fse|≤|Fe|−2|F1e|≤n−4<n−3 and |Fte|≤n−3 for distinct s,t∈[3,n] and n≥6. Since |Est(Bn)|=(n−2)!>n−4≥|F0e|, we have Bn[V(Bsn−Fse)∪V(Btn−Fte)] is connected by Claim 1 for distinct s,t∈[3,n] and n≥6. Then B[3,n]n−F[3,n]e is connected and it is a subgraph of C. There is only 1 isolated vertex ui∈V(Bin−Fie−V(Ci)) and |V(Ci)|≥(n−1)!−1 when Bin−Fie is disconnected by Lemma 3.1 for all i∈[1,2]. Since |EBn(V(Ci), V(B3n))−Fe|≥|Ei3(Bn)|−|V(Bin)−V(Ci)| −|F0e|≥(n−2)!−[(n−1)!−((n−1)!−1)]−(n−4)=(n−2)!−n+3>0 for n≥6, Bn[V(Ci)∪V(B[3,n]n−F[3,n]e)] is connected by Proposition 2.3 for all i∈[1,2]. Then Bn[V(C1)∪V(C2)∪V(B[3,n]n−F[3,n]e)] is connected.
Subcase 2.2.1. Bin−Fie is connected for all i∈[1,2].
In this subcase, both B1n−F1e and B2n−F2e are connected. since |F0e|≤n−4<(n−2)!=|Ei3(Bn)| for all i∈[1,2], Bn−Fe is connected by Proposition 2.3, |V(C)|=|V(Bn)|=n!>n!−2.
Subcase 2.2.2. Only one of B1n−F1e and B2n−F2e is disconnected.
Note that n−2≤|F1e|≤2n−7 and n−2≤|F2e|≤2n−7. Without loss of generality, we suppose that B1n−F1e is disconnected and B2n−F2e is connected. Then there is an isolated vertex in B1n−F1e−V(C1) and Bn[V(C1)∪V(B[2,n]n−F[2,n]e)] is connected by the above. So |V(C)|≥|V(Bn)|−1≥n!−1>n!−2.
Subcase 2.2.3. Bin−Fie is disconnected for all i∈[1,2].
In this subcase, B1n−F1e is disconnected and B2n−F2e is disconnected, Bn[V(C1)∪V(C2)∪V(B[3,n]n−F[3,n]e)] is connected by the above, there are at most two vertices can not be contained in C. So |V(C)|≥|V(Bn)|−2≥n!−2.
Case 3. 2n−6≤|F1e|≤3n−11.
In this case, |F0e|≤|Fe|−|F1e|≤(3n−8)−(2n−6)≤n−2 and |F[2,n]e|≤|Fe|−|F1e|≤n−2.
Subcase 3.1. |Fje|≤n−3, for all j∈[2,n].
Since |Fse|≤n−3, |Fte|≤n−3, |Est(Bn)|=(n−2)!>n−2≥|F0e| for distinct s,t∈[2,n] and n≥6, we have B[2,n]n−F[2,n]e is connected by Claim 1. Then B[2,n]n−F[2,n]e is a subgraph of C. Note that 2n−6≤|F1e|≤3n−11 and Bn≅Bn−1 for n≥6. By the assumption, there is a connected component of B1n−F1e with |V(C1)|≥(n−1)!−2. Since |EBn(V(C1),V(B2n))−Fe|≥|E12(Bn)|-|V(B1n)−V(C1)|−|F0e|≥(n−2)!−[(n−1)!−((n−1)!−2)]-(n−2)=(n−2)!−n>0 for n≥6, Bn[V(C1)∪V(B[2,n]n−F[2,n]e)] is connected and it is C. So |V(C)|≥n!−2.
Subcase 3.2. n−2≤|Fje|≤2n−7, for some j∈[2,n].
Note that |Fe|≤3n−8 and |F[2,n]e|≤n−2. There is only one Fje can satisfy |Fje|=n−2 for all j∈[2,n]. Without loss of generality, we suppose that |F2e|=n−2. This moment, |F1e|=|Fe|−|F2e|=(3n−8)−(n−2)=2n−6. Then |Fse|=0 for all s∈[3,n] and |F0e|=0. Bn[V(Bsn−Fse)∪V(Btn−Fte)] is connected by Claim 1. Then B[3,n]n−F[3,n]e is connected and it is a subgraph of C.
Subcase 3.2.1. Bin−Fie is connected for all i∈[1,2].
In this subcase, both B1n−F1e and B2n−F2e are connected. Since |F0e|=0, we have Bn−Fe is connected by Proposition 2.3, |V(C)|=|V(Bn)|=n!>n!−2.
Subcase 3.2.2. Only B1n−F1e is disconnected.
In this subcase, B2n−F2e is connected. Note that |F1e|=2n−6<2n−5. There is a connected component C1 of B1n−F1e with |V(C1)|≥n!−1 by Lemma 3.1. |E23(Bn)|=(n−2)!>0=|F0e|, we have B[2,n]n−F[2,n]e is connected. Since |EBn(V(C1),V(B3n))−Fe|≥|E13(Bn)|-|V(B1n)−V(C1)|−|F0e|≥(n−2)!-[(n−1)!−((n−1)!−1)]−0=(n−2)!−1>0 for n≥6, we have Bn[V(C1)∪V(B[2,n]n−F[2,n]e)] is connected and it is C. So |V(C)|≥n!−1>n!−2.
Subcase 3.2.3. Only B2n−F2e is disconnected.
In this subcase, B1n−F1e is connected. Since |F2e|=n−2<2n−5 for n≥6, there is a connected component C2 of B2n−F2e with |V(C2)|≥n!−1 by Lemma 3.1. By Proposition 2.3 and |F0e|=0, Bn[V(B1n−F1e)∪V(B[3,n]n−F[3,n]e)] is connected. Since |EBn(V(C2),V(B3n))−Fe|≥|E23(Bn)|−|V(B2n)−V(C2)|-|F0e|≥(n−2)!−[(n−1)!−((n−1)!−1)]-0=(n−2)!−1>0 for n≥6, we have Bn[V(B1n−F1e)∪V(C2)∪V(B[3,n]n−F[3,n]e)] is connected and it is C. So |V(C)|≥n!−1>n!−2.
Subcase 3.2.4. Bin−Fie is disconnected for all i∈[1,2].
In this subcase, B1n−F1e is disconnected and B2n−F2e is disconnected. By Subcase 3.2.2 and Subcase 3.2.3, we have |V(C1)|≥n!−1, |V(C2)|≥n!−1. Moreover, we have Bn[V(C1)∪V(C2)∪V(B[3,n]n−F[3,n]e)] is connected and it is C. So |V(C)|≥n!−2.
Case 4. 3n−10≤|F1e|≤3n−8.
In this case, |F0e|≤|Fe|−|F1e|≤(3n−8)−(3n−10)=2 and |F[2,n]e|≤|Fe|−|F1e|=2. Since |Eij(Bn)|=(n−2)!>2≥|F0e| for distinct i,j∈[2,n] and n≥6, B[2,n]n−F[2,n]e is connected by Claim 1. If B1n−F1e is connected, then Bn−Fe is connected by Proposition 2.3 and |V(C)|=|V(Bn)|=n!>n!−2. If B1n−F1e is disconnected, note that |F0e|≤2, then at most two vertices in B1n cannot be contained in C, thus, |V(C)|≥n!−2.
By Case 1-4, When Bn−Fe is disconnected, there is a largest connected component C of Bn−Fe with |V(C)|≥n!−2 when |Fe|≤3n−8.
Remark 4.3. Lemma 4.2 does not hold for n=4. When n=4, |Fe|=3×4−8=4. See from Figure 1, B4−Fe has a 4-cycle as its component except the largest connected component C. Then |V(C)|=4!−4.
Remark 4.4. The conclusion of Lemma 4.2 is optimal in that there is an edge set Fe⊆E(Bn) with |Fe|=3n−7 such that Bn−Fe has a connected component C with |V(C)|≤n!−3 for n≥6.
Let a 2-path be uvw with the vertices in this order and F={u,v,w} be the vertex set of the 2-path. Then |EBn(F)|=3n−7 by Proposition 2.5. Let Fe=EBn(F)⊆E(Bn). Then uvw is a component of Bn−Fe except C. So |V(C)|≤n!−3.
Theorem 4.5. Bn is (2n−6)-conditional edge-fault-tolerant strongly Menger edge connected for n≥5.
Proof. Let Se⊆E(Bn) be an faulty edge set with |Se|≤2n−6 and δ(Bn−Se)≥2. Let two distinct vertices u,v∈V(Bn), a=min{dBn−Se(u),dBn−Se(v)} and Ef⊆E(Bn)−Se with |Ef|≤a−1. By Theorem 1.1, Definition 1.1 and Definition 1.2, it is sufficient to prove that u is connected to v in Bn−Se−Ef. We prove this by contradiction, assume that u and v are disconnected in Bn−Se−Ef for Ef⊆E(Bn)−Se with |Ef|≤a−1. Note that dBn−Se(u)≤n−1, dBn−Se(v)≤n−1 and |Ef|≤n−2 by Proposition 2.5. Then |Fe|=|Se∪Ef|≤(2n−6)+(n−2)=3n−8. By Lemma 4.2, Bn−Se−Ef has a connected component C with |V(C)|≥n!−2. Note that u and v are disconnected in Bn−Se−Ef, |{u,v}∩V(C)|≤1. Without loss of generality, we suppose that u∈V(C) and v∉V(C).
Case 1. |V(C)|=n!−1.
In this case, v is the isolated vertex of Bn−Se−Ef. Then EBn({v},NBn−Se(v))⊆Ef. This means |Ef|≥dBn−Se(v), which contradicts with |Ef|≤a−1≤dBn−Se(v)−1. The assumption does not hold.
Case 2. |V(C)|=n!−2.
In this case, there is another vertex w∈V(Bn−Se−Ef) and w∉V(C). If v and w are connected, then K2=vw and EBn−Se({vw})⊆Ef. Since δ(Bn−Se)≥2, v has at least one degree in Ef. Then |Ef|≥dBn−Se(v)−1+1=dBn−Se(v), which contradicts with |Ef|≤a−1≤dBn−Se(v)−1. If v and w are disconnected, then EBn({v},NBn−Se(v))⊆Ef. This means |Ef|≥dBn−Se(v), which contradicts with |Ef|≤a−1≤dBn−Se(v)−1. The assumption does not hold.
Hence, Bn is (2n−6)-conditional edge-fault-tolerant strongly Menger edge connected for n≥5.
5.
Results discussion
The edge connectivity is one of the basic parameters to estimate the fault tolerance of the interconnection network. We show that Bn is (n−3)-edge-fault-tolerant strongly Menger edge connected for n≥3 and (2n−6)-conditional edge-fault-tolerant strongly Menger edge connected for n≥5. There are some problems need further discussion.
(1) The fault tolerance of an interconnected network is the maximum number of vertices or edges allowed to fail without affecting the communication of other fault-free vertices or edges. Let Fe be the set of faulty edges in Bn. We show that when Bn has (n−3) (|Fe|=n−3) edges fault, there are min{dBn−Fe(u),dBn−Fe(v)} edge-disjoint paths that can connect any two distinct vertices in Bn. This conclusion can ensure that Bn can work normally even with (n−3) faulty edge and improve the reliability of the system. When the condition of any vertex in Bn has at least two neighbors is imposed, the conclusion can ensure that Bn can work normally even with (2n−6) faulty edges and improve the reliability of the system.
(2) In fact, we don't have a counter example to show that Bn is not edge-fault-tolerant strongly Menger edge connected for larger |Fe|, the researchers can explore the more general |Fe|.
(3) In this paper, we discuss under the condition that each vertex or edge in Bn has a uniform and independent failure probability. But actually, vertex or edge may have different failure probabilities. In addition, vertices or edges may be related and fail at the same time, so that vertices or edges failures may not be independent. These need to be discussed furthermore.
Acknowledgments
This work is supported by the National Science Foundation of China (61772010) and the Graduate Quality Curriculum Construction Project of Henan Normal University(5101019500604).
Conflict of interest
All authors declare no conflict of interest in this paper.