7
Graphs with Given Connectivity Properties Serge Lawrencenko, 1 Qiang Luo 2 1 Sturwalnaya 7-1-45, Moscow 123363, Russia 2 Wuhan University of Hydrologic and Electronic Engineering, People’s Republic of China Received 7 November 1996; accepted 7 June 1997 Abstract: A node of a graph G , thought of as representing a communication network, is said to be redundant provided that its removal does not diminish the connectivity. In constructing networks, we require reliable connectedness in addition to the usual requirement of reliability ( i.e., the higher the connectivity, the more reliable the network ) . Two nodes are called reliably connected if they are joined by a reliable path, i.e., a path whose internal nodes (if any) are all redundant. For a given n , we construct a communication network on n nodes having a given value k of connectivity ( reliability condition ) and as few edges as possible ( optimization condition ) and in which any pair of nodes, with one exceptional node, are reliably connected ( reliable connectedness condition ) . We also studied the associated-with- G ‘‘strong redundancy graph’’ and a sequence of ‘‘weak redundancy graphs,’’ quotient graphs which display decompositions of G into connected subgraphs in accordance with the contributions of the nodes to the connectivity k( G ). q 1997 John Wiley & Sons, Inc. Networks 30: 255–261, 1997 1. INTRODUCTION u -essential node cuts of G will be denoted by C u ( G ). To see that C u ( G ) x M for any u V ( G ), pick a node, u *, The symbol G will always denote a finite simple con- adjacent to u ; clearly, then, V ( G ) " { u * } C u ( G ). The nected nontrivial graph. By V ( G ), E ( G ), n( G ), and redundancy of a node u in the graph G is defined to be k( G ), we denote the node set, the edge set, the order, and the connectivity of graph G , respectively. A node cut r G ( u ) Å min{É AÉ 0 k( G ): A C u ( G )}. (1) of G is a subset A of V ( G ) such that the induced subgraph G[V 0 A ], also denoted by G 0 A , is disconnected or A node u is called ( connectivity -) essential if r G ( u ) Å 0, trivial. A node cut is strong if G 0 A is disconnected; it and m-redundant if r G ( u ) ¢ m ¢ 1. By a redundant is weak if G 0 A is trivial. Thus, A is a node cut of G if node, we shall mean a 1-redundant node. and only if k( G 0 A ) Å 0. An ,-node cut is a node cut Redundancy of a node is a measure ‘‘inverse’’ to the of size , ; in particular, a strong 1-node cut is called a contribution of that node to the connectivity of G . This cut node. Our general graph-theoretic terminology and measure is more subtle than is the previously studied [1] notation are essentially those of Bondy and Murty [2]. cohesiveness of a node u in graph G , defined by c G ( u ) For u V ( G ), let A be a node cut of a graph G Å k( G ) 0 k( G 0 u ). Clearly, u is redundant ( respec- containing u . We will say that A is a u-essential node cut tively, essential) if and only if c G ( u ) ° 0 ( respectively, if A 0 u is no longer a node cut of G . The family of all ú0 ) . A number of papers ( see, e.g., [1, 4 ] ) were devoted to the study of the number of redundant nodes in a given graph. Correspondence to: S. Lawrencenko q 1997 John Wiley & Sons, Inc. CCC 0028-3045/97/040255-07 255 787 / 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Graphs with given connectivity properties

  • Upload
    qiang

  • View
    215

  • Download
    2

Embed Size (px)

Citation preview

Page 1: Graphs with given connectivity properties

Graphs with Given Connectivity Properties

Serge Lawrencenko,1 Qiang Luo2

1 Sturwalnaya 7-1-45, Moscow 123363, Russia

2 Wuhan University of Hydrologic and Electronic Engineering, People’s Republic of China

Received 7 November 1996; accepted 7 June 1997

Abstract: A node of a graph G , thought of as representing a communication network, is said to beredundant provided that its removal does not diminish the connectivity. In constructing networks, werequire reliable connectedness in addition to the usual requirement of reliability ( i.e., the higher theconnectivity, the more reliable the network) . Two nodes are called reliably connected if they are joinedby a reliable path, i.e., a path whose internal nodes ( if any) are all redundant. For a given n , we constructa communication network on n nodes having a given value k of connectivity ( reliability condition) and asfew edges as possible (optimization condition) and in which any pair of nodes, with one exceptionalnode, are reliably connected (reliable connectedness condition) . We also studied the associated-with-G ‘‘strong redundancy graph’’ and a sequence of ‘‘weak redundancy graphs,’’ quotient graphs whichdisplay decompositions of G into connected subgraphs in accordance with the contributions of the nodesto the connectivity k(G ) . q 1997 John Wiley & Sons, Inc. Networks 30: 255–261, 1997

1. INTRODUCTION u-essential node cuts of G will be denoted by Cu(G) . Tosee that Cu(G) x M for any u √ V (G) , pick a node, u *,

The symbol G will always denote a finite simple con- adjacent to u ; clearly, then, V (G)" {u *} √ Cu(G) . Thenected nontrivial graph. By V (G) , E(G) , n(G) , and redundancy of a node u in the graph G is defined to bek(G) , we denote the node set, the edge set, the order,and the connectivity of graph G , respectively. A node cut rG(u) Å min{ÉAÉ 0 k(G) : A √ Cu(G)}. (1)of G is a subset A of V (G) such that the induced subgraphG[V 0 A] , also denoted by G 0 A , is disconnected or

A node u is called (connectivity-) essential if rG(u) Å 0,trivial. A node cut is strong if G 0 A is disconnected; it

and m-redundant if rG(u) ¢ m ¢ 1. By a redundantis weak if G 0 A is trivial. Thus, A is a node cut of G if

node, we shall mean a 1-redundant node.and only if k(G 0 A) Å 0. An ,-node cut is a node cut

Redundancy of a node is a measure ‘‘inverse’’ to theof size , ; in particular, a strong 1-node cut is called a

contribution of that node to the connectivity of G . Thiscut node. Our general graph-theoretic terminology and

measure is more subtle than is the previously studied [1]notation are essentially those of Bondy and Murty [2] .

cohesiveness of a node u in graph G , defined by cG(u)For u √ V (G) , let A be a node cut of a graph G Å k(G) 0 k(G 0 u) . Clearly, u is redundant (respec-

containing u . We will say that A is a u-essential node cuttively, essential) if and only if cG(u) ° 0 (respectively,

if A 0 u is no longer a node cut of G . The family of all ú0). A number of papers (see, e.g., [1, 4]) were devotedto the study of the number of redundant nodes in a givengraph.Correspondence to: S. Lawrencenko

q 1997 John Wiley & Sons, Inc. CCC 0028-3045/97/040255-07

255

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 2: Graphs with given connectivity properties

256 LAWRENCENKO AND LUO

We shall think of G as representing a communication and also infinitely many graphs I satisfying Ià Rm(I) . Tocharacterize all such graphs I remains an open problem.network; its nodes correspond to the communication sta-

tions, and the edges to the communication links betweenthose stations. Usually, a network is regarded as reliableas the value of its connectivity (see [2]) . In addition to 2. PROPERTIES OF m-REDUNDANTthis characteristic, here we introduce the concept of reli- NODESable connectedness in a communication network. A (u ,£)-path is called a reliable (u , £)-path, for communica- Let a graph G represent a communication network. Bytion between nodes u and £, if it internally avoids essential choosing m-reliable paths for communication in the sys-nodes. Nodes u and £ are called reliably connected if tem (such choices are usually possible in an m-reliablythere is a reliable (u , £)-path. (Note that a breakdown of connected communication network), we gain an im-a communication station represented by an essential node portant advantage: namely, a blockage of a number (inwould decrease the value of connectivity of the network; total less than m) of their internal nodes at a time doestherefore, intensively using essential nodes may jeopar- not create new essential nodes (which are undesirable todize communication in the system.) More generally, use for communication) provided that the blocked nodesnodes u and £ are called m-reliably connected (m ¢ 1) constitute a subset of V (G) stable in the sense that itsif there is an m-reliable (u , £)-path, i.e., a (u , £)-path removal does not alter the value of connectivity (bywhose internal nodes (if any) are all m-redundant. Adja- Lemma 2.3, this condition is normally satisfied).cent nodes are m-reliably connected for any m . We now state the result formally: Let m ¢ 2. A

It can be easily seen that if two nodes are m-reliably nonempty subset B of V (G ) is called an m-redundantconnected, then they cannot be separated by any subset subset if rG (u ) ¢ m ¢ 2 for all u √ B ; B is calledof V (G) of size less than m / k(G) . Here, we include stable if k(G ) Å k(G 0 B ) . Let Rm (G ) denote thean infinite series of examples showing that the converse is family of all m-redundant stable subsets of V (G ) hav-not necessarily true. Take an even cycle C Å £1£2rrr£2p£1 ing size less than m .(with p ¢ 2) and p disjoint copies K2( i) of K2 ( i Å 1,. . . , p) . Identify one node of K2( i) with the node £2i of Theorem 2.1. Assume that m ¢ 2 and B √ Rm(G) .C for each i and denote the resulting graph by Xp . Now, Then, the removal of B from G does not create new essen-let u and £ be some two nodes of Xp both having degree tial nodes, i.e., rG0B(£) Å 0 implies that rG(£) Å 0 .two. We have that Xp is a graph with k(Xp) Å 1 in whichno node can separate u and £, but u and £ are not 1- We shall prove this result at the end of this section,reliably connected. but first we develop three lemmas.

A communication network G is said to be reliably (m-reliably) connected if any pair of its nodes, possibly with Lemma 2.2. Let u be a node of G such that k(G) Å k(Gone exceptional node, are joined by a reliable (m-reliable) 0 u) . Then, rG(£)° rG0u(£)/ 1 for any £√ V (G)" {u} .path. When G x Kn , G has to have a strong node cut and

Proof. Let £√ V (G0 u) . By the connectivity hypoth-has to have an exceptional node, u0 ; furthermore, G hasesis, G 0 u is not trivial, and, therefore, C

£(G 0 u) xMa unique minimum strong node cut which is the neighbor

(see the Introduction). Pick A in C£(G 0 u) so thatset of u0 . A good communication network is to be de-

signed so that the exceptional node represents a ‘‘smallnonterminal station.’’ ÉAÉ 0 k(G 0 u) Å rG0u(£) (2)

In Section 2, we continue to discuss the advantages ofm-reliable connectedness in a network. In Section 3, we [see Eq. (1)] . Then, A < {u} √ C

£(G) . Therefore, by

solve the problem of constructing reliably (i.e., 1-reli- Eqs. (1) and (2),ably) connected communication networks; the generalproblem (for an arbitrary m) is still open. In Section 4, rG(£) ° ÉA < {u}É 0 k(G)we consider decompositions of graphs into parts so that

Å rG0u(£) / k(G 0 u) / 1 0 k(G) (3)each part is a maximum connected subgraph having onlynodes of the same redundancy or ‘‘close’’ redundancies. Å rG0u(£) / 1.To display such decompositions, we introduce special

jquotient graphs associated with a given graph G , namely,the strong redundancy graph R(G) and a sequence of

The following is a result of [1, 4] . Recall that weweak redundancy graphs Rm(G) , and characterize thoseassume G to be a connected nontrivial graph.graphs under certain connectivity assumptions. In Section

5, we construct infinitely many graphs I with the propertythat I and R(I) are isomorphic graphs, written I à R(I) , Lemma 2.3. A node u of G is essential if k(G) ú k(G

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 3: Graphs with given connectivity properties

GRAPHS WITH GIVEN CONNECTIVITY PROPERTIES 257

0 u) and redundant otherwise. Furthermore, G may have operations creates new essential nodes. This completesthe proof. jat most one node u with k(G) õ k(G 0 u) , which is the

case if and only if G can be obtained from a k *-connectedgraph G*, for some k* ¢ 2 , by adding a node, u, andjoining it to a set of k* 0 1 nodes in G*. j 3. RELIABLY CONNECTED

COMMUNICATION NETWORKSLet u be a 2-redundant node of G . In case k(G)õ k(G

The following is an optimization problem that emerges0 u) , one can easily construct examples to show that Gfrom constructing reliably connected communication net-0 u may have more essential nodes than has G . However,works.if k(G) Å k(G 0 u) , no such an example is possible.

This fact is the content of the following lemma whichstrengthens Lemma 2.2 in the particular case rG0u(£) Å 0

The General Problem[under the additional restriction rG(u) ¢ 2].GIVEN:

Lemma 2.4. Let u be a 2-redundant node of a graph G• A nontrivial edge-weighted graph L of order n whosesuch that k(G) Å k(G 0 u) . Then, the removal of u from

nodes are denoted by u0 , u1 , . . . , un01 , andG does not create new essential nodes, i.e., rG0u(£) Å 0implies that rG(£) Å 0 . • two positive integers k and m satisfying

Proof. Assume that £ is an essential node of G 0 u .k / m õ n 0 1. (5)Pick A in C

£(G 0 u) satisfying ÉAÉ 0 k(G 0 u)

Å rG0u(£) Å 0. Then, A < {u} is a node cut of G . ButA < {u} √/ Cu(G) , for otherwise, rG(u) ° ÉA < {u}É

FIND: A spanning subgraph S Å Sk ,n ,m of L satisfying the0 k(G) Å k(G 0 u) / 1 0 k(G) Å 1, which contradictsfollowing three conditions:the 2-redundancy of u . Hence, (A < {u})" {u} Å A is

also a node cut of G . This node cut is necessarily mini-(C1) k(S) Å k (reliability condition),mum because ÉAÉ Å k(G 0 u) Å k(G) . Hence, A(C2) S be an m-reliably connected communication net-√ C

£(G) , and therefore rG(£) ° ÉAÉ 0 k(G) Å 0. The

work with u0 as the exceptional node (reliable con-proof is complete. jnectedness condition), and

(C3) S be of minimum possible weight (optimizationUnder the hypotheses of Lemma 2.4, assume now that condition).

u is m-redundant (in G) with m ¢ 3. Then, it may be thatrG0u(£) ° m 0 2, while rG(£) ú m 0 2. To show this,

Observe that Eq. (1) gives m° n0 10 k with equalityhere we include an example for m Å 4. Consider K4 with

possible only if S Å Kn . On the other hand, Kn cannotone edge deleted. Denote by £ and x its two nodes of degree

satisfy this inequality for any positive m . Thus, Inequalitythree. Add two new nodes, u and y , and join them to x .

(5) is not really restrictive. So, S is not Kn and has atTake the resulting graph as G . We have k(G) Å k(G 0 u)

least one strong node cut (whose nodes are all essential,Å 1, rG(u) Å 4, rG0u(£) Å 2, but rG(£) Å 3.of course) . So, we have to admit one exceptional node;we assume that u0 is such a node.Proof of Theorem 2.1. By induction on m . When m

The general problem with (C2) omitted is known asÅ 2, the theorem is equivalent to Lemma 2.4. Let M ¢ 3.the problem of construction of reliable communicationSuppose that the theorem is true for all m õ M , and letnetworks. It appears to be very difficult and presently isB √ RM(G) with ÉBÉ ¢ 2. Then, by Lemma 2.3, thereunsolved in such a general setting (see [2]) . So, it isexists a node u in B such that k(G) Å k(G 0 u) . Let £reasonable to impose additional restrictions. Here, we√ B" {u}. By Lemma 2.2, rG0u(£) ¢ M 0 1. Further-shall solve the general problem in one such a particularmore,setting, namely, we shall refer to it as the ‘‘restrictedproblem’’ if the following restrictions hold:k((G 0 u) 0 (B" {u})) Å k(G 0 B) Å k(G)

Å k(G 0 u) .(4)

Restrictions

(R1) L Å Kn , a complete graph in which every edge isTherefore, B" {u} √ RM01(G 0 u) . The removal of Bfrom G is equivalent to the removal of u from G followed assigned unit weight,

(R2) k ¢ 2,by the removal of B" {u} from G 0 u . By Lemma 2.4and the induction hypothesis, neither of the two latter (R3) m Å 1,

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 4: Graphs with given connectivity properties

258 LAWRENCENKO AND LUO

(R4) n ¢ 6. Theorem 3.1. The graph Q Å Qk,n with u0 (Åz0) as theexceptional node is a solution of the restricted problem.

By dG(u) and NG(u) , we shall denote the degree of Proof. First, the graph Q indeed satisfies conditionnode u and its neighbor set in graph G , respectively. To (C1) because it contains as a subgraph the graph H Å Hk ,nsolve the restricted problem is actually to find a graph S known [2, 3] to be of connectivity k . Furthermore, byof connectivity k on n nodes, having as few edges as construction,possible to satisfy the condition that any two nodes, otherthan the exceptional node u0 , be 1-reliably connected. ÉE(Q)É Å ÉE(H)É / n /2 Å kn /2 / n /2 . (8)Thus, any solution S of the restricted problem has a uniquestrong k-node cut ( i.e., a unique minimum strong node Thus, Q also satisfies condition (C3) thanks to Inequalitycut) , namely, NS(u0) . Thus, we can estimate the number (7) . It remains to prove that Q satisfies condition (C2).of edges in S as follows: Observe that dQ(u0) Å k , while the other nodes have

higher degrees. Thus, it is enough to prove that NQ(u0)is a unique strong k-node cut of Q ( i.e., a unique minimum

2ÉE(S)É Å ∑n01

iÅ0

dS(ui ) ¢ k / (n 0 1)(k / 1), (6) strong node cut) .We shall first examine the structure of minimum strong

node cuts of H (ÅHk ,n) . By a continuous p-set, we willand, hence, mean a set of p nodes of H which occur consecutively

around the circle {z : ÉzÉ Å 1} in the complex plane. Itcan be easily verified that after the removal of any set ofÉE(S)É ¢ (kn / n 0 1)/2 Å kn /2 / n /2 . (7)k nodes which does not contain a continuous k /2-set wecan certainly identify a Hamilton cycle using appropriateWe shall first exhibit a graph, Qk ,n , having exactly kn /edges of H in Case 1 and of Hk01,n in Cases 2 and 3. It2 / n /2 edges. Then, we shall verify that Qk ,n is afollows that in Case 1 any strong k-node cut of H can besolution of the restricted problem.expressed as W1 < W2 , a disjoint union of two continuousWe start our construction with a special graph Hk ,n , kk /2-sets such that the union itself does not form a¢ 2, of order n and connectivity k , having exactly kn /continuous k-set. In Cases 2 and 3, ÉW1 < W2É Å k 0 1,2 edges. This graph was originally constructed by Hararyso that any strong k-node cut of H can be expressed as[3] . The construction can be also found in [2] , in which

it is actually shown that Hk ,n is a solution of the problemW Å W1 < W2 < {u} (9)of constructing reliable communication networks [i.e., the

restricted problem with (C2) omitted] . Here, for the sakefor some u √ V ( H) . Thanks to the edges that we addedof completeness, we include a reproduction of that con-to Hk01,n in Cases 2 and 3 to obtain H , the set W is astruction. Our reproduction differs from the presentationstrong k-node cut if and only if the two conditions areof [2]: We shall exploit the complex plane. The nodessatisfied:ur of Hk ,n are associated with the n points zr Å exp(2pri /

n) (r Å 0, . . . , n 0 1) in the complex plane. The proce-( i) W1 and W2 be separated by a single node—more

dure of adding edges splits into three cases:precisely, there should be a node £ of H such thatW1 < W2 < {£} is a continuous k-set, andCASE 1. k even. To obtain Hk ,n , join each node zr to all

nodes zs that can be expressed as zrexp(2pti /n) , where (ii ) dH(£) Å k .t √ {{1, {2, . . . , {k /2}.

Necessarily, then, u in Eq. (9) is the sole member ofCASE 2. k odd, n even. We obtain Hk ,n from Hk01,n by NH(£)"(W1 < W2) .

joining zr to zs Å 0zr for r Å 0, 1, . . . , (n 0 2)/2. Now we turn to the graph Q . Except the case k Å 3,n Å 6 (which is settled straightforwardly) , the graph H

CASE 3. k odd, n odd. We obtain Hk ,n from Hk01,n by 0 W has precisely two components, one of which is ajoining zr to zs Å 0zrexp(pi /n) for r Å 1, 2, . . . , (n single node £. Unless £ Å u0 , we have dQ(£) ú dH(£) ,/ 1)/2. so an edge in E(Q)"E(H) incident to £ joins the two

components of H 0 W , and so Q 0 W is connected.Therefore, in Cases 2 and 3, the set W is the only strongNow, to construct Qk ,n , we add n /2 more edges to

Hk ,n ; namely, in Cases 1 and 2, we join zr for r Å 1, 2, k-node cut of Q . This node cut, in fact, coincides withNQ(u0) . Similarly, one can obtain the same conclusion. . . , n /2 to zs Å 0zrexp(02pi /n) if n is even and to

zs Å 0zrexp(0pi /n) if n is odd. In Case 3, we join zr to in Case 1; for this, take into account the edges that weadded to H in order to obtain Q .zs Å 0zrexp(0pi /n) for r Å 2, 3, . . . , (n 0 1)/2, and

also join z (n/1) /2 to zn01 . The proof is complete. j

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 5: Graphs with given connectivity properties

GRAPHS WITH GIVEN CONNECTIVITY PROPERTIES 259

4. THE REDUNDANCY QUOTIENT GRAPHS i.e., the graph obtained from H[K2 1 Kp] by deleting theedges satisfying the conditions specified above in Eq. (12).

Observe first that for any connected H and any p ¢ 2In this section, we develop another point of view on thethe graph GÅ Sp(H) has connectivity p and entirely consistsconcepts of the preceding section.of essential nodes. Hence, R(G) à Rm(G) à K1 for eachLet < Pi be a partition of V (G) such that for each im . Thus, the case of H trivial is settled by G Å Sk(K1), andPi is a nonempty subset of V (G) and the induced sub-in what follows, we shall assume that H à/ K1 .graph G[Pi ] is connected. The quotient graph associated

The following is another obvious but useful observation:with a given pair (G , < Pi ) is denoted by G /< Pi anddefined to be the graph whose nodes correspond to the

Lemma 4.1. If u is a cut node of a graph G of connectiv-parts Pi and two nodes Pi and Pj are adjacent wheneverity one, then in each component of G 0 u there is a nodei x j and a member of Pi is adjacent to a member of Pj£ with rG(£) ¢ 1 . j(as nodes of G) .

Consider the partition < Ri of V (G) , where each Ri In the following constructions, we shall process a suit-is the node set of a largest connected subgraph of G with

able superstructure on H by judiciously contracting itsthe property that any two of its nodes u and £ satisfy

edges. A contraction of an edge u£ is the operation ob-rG(u) Å rG(£) . The quotient graph G /< Ri is denoted

tained by the removal of u and £ and the addition of aby R(G) and called the strong redundancy graph of G .

new node adjacent to the nodes formerly adjacent to u orFor a given integer m satisfying 0 õ m õ n(G) 0 k(G) ,

£. Especially, when u Å (a , b , c1) and £ Å (a , b , c2) ,the (weak) m-redundancy graph is defined to be

the new node will be denoted by (a , b , min(c1 , c2)) .(Note that under the operation of edge contraction as

Rm(G) Å G / (<i Rmi ) < (<jE

mj ) , (10) defined in [2] there may arise pairs of multiple edges;

however, omitting a duplicated edge in each such pair,where each Rm

i (respectively, Emj ) is the node set of a we obtain the result of the contraction in our sense.)

largest connected subgraph of G having only nodes £ withrG(£) ¢ m (respectively, õm) . Clearly, each Rm(G) is Theorem 4.2. Let k be a given positive integer. A non-a connected bipartite graph (unless it is trivial) having trivial graph H is the strong redundancy graph of some

graph G of connectivity k if and only if H is connected<i Rmi and <j Em

j as two parts of the bipartition.and, for the case k Å 1 , has a cut node.Roughly speaking, the smaller the size of the part <i

Rmi , the more pairs of nodes are m-reliably connected. In Proof. The necessity follows from Lemma 4.1. To

particular, any solution S of the restricted problem of the prove sufficiency, let H be a connected graph with V ( H)preceding section satisfies Å {0, 1, . . . , n 0 1}. Especially in the case k Å 1, we

may assume that 0 is a cut node (as matter of notation).R 1(S) à Kc

2 Û Kck , (11) We shall obtain graph G as desired by processing the

superstructure Sn/k(H) . First, only in the case k Å 1,where the right-hand part denotes the join of the comple- contract all the edges of the form (0, b1 , c1)(0, b2 , c2)ments of (disjoint) K2 and Kk , with É<i R 1

i É Å ÉKc2É into a single node and designate that node by (0, 0, 0) .

Å 2, and the symbol ‘‘à’’ denotes the isomorphy of the Second, for any k , repeatedly contract the edges of thegraphs. form (a , b , c1)(a , b , c2) satisfying min(c1 , c2) ¢ a / k

In this section, we shall obtain criteria for a given 0 1. When no such edges are left, we have a graph whichgraph H to be the redundancy graph of some graph G we take as G . Observe that G has the property that everywith a prescribed value k of connectivity. By n Å n(H) , subgraph induced by the set of nodes (a , b , c) with awe shall denote the order of H , and by 0, 1, . . . , n 0 1, fixed a is isomorphic to the graph K2 1 Ka/k , with theits nodes. Let V ( Kp) be {0, 1, . . . , p 0 1}. The composi- obvious exception of a Å 0, k Å 1. It remains to provetion H[K2 1 Kp] has as its node set the product V ( H) that G satisfies R(G) à H and k(G) Å k . For this, let A1 V ( K2) 1 V ( Kp) . The nodes of the graph H[K2 1 Kp] be a node cut of G attaining the minimum in Eq. (1) forwill be denoted by (a , b , c) , where a √ {0, 1, . . . , n a given node u Å (a(u) , b(u) , c(u)) √ V (G) ; in other0 1}, b √ {0, 1}, and c √ {0, 1, . . . , p 0 1}, and its words, A is a member of Cu(G) of minimum size. It canedges by (a1 , b1 , c1)(a2 , b2 , c2) . The p th superstructure be easily seen that the projection of A to V ( H) collapseson graph H is defined to be into a single node a(u) of H and that ÉAÉ Å a(u) / k ,

so that k(G) Å k . For instance, one such node cut isgiven bySp(H) Å H[K2 1 Kp]

0 {(a1 , b1 , c1)(a2 , b2 , c2) : A Å {(a(u) , b(u) , c(u))} <

{(a , b , c) : a Å a(u) , b x b(u) , c x c(u)}.(13)

a1 x a2 and b1b2 Å 0},

(12)

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 6: Graphs with given connectivity properties

260 LAWRENCENKO AND LUO

Therefore, we have rG((a , b , c)) Å a for each node (a , problem to characterize strongly irreducible (or weaklym-irreducible) graphs with a given value k of connectivityb , c) √ V (G) . The proof is complete. j

(with given values of m and k) .The following is a construction for generating infinitelyThe next theorem is a criterion for a graph to be the

many strongly irreducible graphs with a prescribed valueweak m-redundancy graph of some graph with a givenk , k ¢ 1, of connectivity. Let T be a nontrivial tree, notvalue k of connectivity. This splits into two cases:K2 . Denote its node set by V (ÅV (T )) , and by V1 , the

CASE 1. mk ¢ 2. set of its nodes of degree one. We have V1 x M andV "V1 x M. For z in V1 , its neighbor set NT(z) consistsCASE 2. mk Å 1.of a single node which will be denoted by N(z) . Assignto the nodes y √ V positive integers n(y) so thatTheorem 4.3. Let m and k be given positive integers. A

nontrivial graph H is isomorphic to Rm(G) for somegraph G of connectivity k if and only if H is a connected ( i ) min{n(y) : y √ V "V1} Å k ,bipartite graph in Case 1 , or H is a connected bipartite ( ii ) n(y) x n(x) whenever yx √ E(T ) and y , x √/ V1 ,graph with one part, X, of the bipartition entirely con- andsisting of cut nodes in Case 2 .

( iii ) (yxN (z ) n(y) x n(N(z)) for each z √ V1 .Proof. The necessity is obvious in Case 1 and follows

from Lemma 4.1 in Case 2. To establish sufficiency, let Then, replace each node y √ V by the graph Y ( y)H be a bipartite graph with parts X Å {0, 1, . . . , l 0 1} à Kc

n (y ) [ i.e., the edgeless graph on n(y) nodes] , andand Y Å {l, l / 1, . . . , n 0 1}. To construct G in Case join each node in Y ( y) to each node in Y ( x) whenever2, it is enough to contract, for each a √ {0, 1, . . . , l yx √ E(T ) . Denote the resulting graph by Y (T ) .0 1}, all the edges in S2(H) that join the nodes havingthe same a-component into a single node, £(a) . Let us

Proposition 5.1. The graph Y (T ) is a strongly irreduc-now construct G in Case 1 to satisfy k(G)Å k and Rm(G)ible graph of connectivity k.à H . For this, repeatedly contract in Sm/k(H) the edges

of the form (a , b , c1)(a , b , c2) satisfying both conditions: Proof. Consider the projection p : V (Y (T )) r V de-fined by p(u) Å y , where y is such that u √ Y ( y) . For

(i) a √ {0, 1, . . . , l 0 1}, and u √ V (Y (T )) , let A be a member of Cu(Y (T )) of mini-( ii ) min(c1 , c2) ¢ k 0 1; mum size. Since NY (T ) (u) Å NY (T ) (£) whenever p(u)

Å p(£) , it follows that Y (p(u)) , A whenever u √ A ,in the case k Å 1, condition (ii ) is replaced by and, therefore, p(A) is a member of Cp(u ) (T ) of minimum(ii *) min(c1 , c2) ¢ 1. size. If p(u) √/ V1 , then {p(u)} is the only member of

Cp(u ) (T ) of minimum size. If p(u) √ V1 , the only suchFinally, only in the case k Å 1 (necessarily, then, m member is V " {N(p(u))}. Therefore, by condition (i) ,

¢ 2 as we are now considering Case 1), to make the k(Y (T )) Å k . Furthermore,resulting graph (which is to be taken as G) be of connec-tivity one, we adjoin a cycle of length five to the node

rY (T ) (u)(0, 0, 0) . Then, except the node (0, 0, 0) having rG((0,0, 0)) Å 0, the nodes of that cycle have redundancy oneas well as the nodes of the form (0, b , c) . Hence, all the Å

n(p(u)) 0 k if p(u) √ V "V1

∑yxN (p(u ) )

n(y) 0 k if p(u) √ V1(14)

named nodes are in the same set Emj of Eq. (10). Proceed-

ing as in the proof of the preceding theorem, one canverify that the so-constructed graph G is as desired. The

Now, by conditions (ii ) and (iii ) , the proof is complete.proof is complete. jj

5. IRREDUCIBLE GRAPHS A similar construction applies to obtain infinitely manyweakly m-irreducible graphs with a given positive valuek of connectivity. Let m be a positive integer. Let T be aGiven a connected nontrivial graph, by repeatedly con-

tracting the edges having both ends of the same redun- nontrivial tree, not K2 , in which the distance between anytwo nodes of degree one is even. Then, V splits into twodancy, we finally obtain a graph I with the property R(I)

à I . We will say that such a graph I is a strongly irreduc- classes: (1) nodes for which the distance to any node ofdegree one is even, and (2) nodes for which that distanceible graph. Similarly, a weakly m-irreducible graph is

defined to be a graph I satisfying Rm(I) à I . It is an open is odd. For y in class 1, assign to y any integer n(y)

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks

Page 7: Graphs with given connectivity properties

GRAPHS WITH GIVEN CONNECTIVITY PROPERTIES 261

satisfying n(y) ¢ m / k . For y in class 2, y is assigned The authors are grateful to the referees and personally toProfessor Frank Boesch for their helpful comments in improv-n(y) Å k . Denote the resulting graph by Ym(T ) .ing the presentation.

Proposition 5.2. Ym(T ) is a weakly m-irreducible graphof connectivity k. REFERENCES

Proof. This proceeds similarly to the proof of the pre-[1] J. Akiyama, F. Boesch, H. Era, F. Harary, and R. Tindell,ceding proposition. j

The cohesiveness of a point of a graph. Networks 11(1981) 65–68.

However, in this way, we do not obtain all irreducible [2] J. A. Bondy and U. S. R. Murty, Graph Theory with Appli-graphs. For example, the graph Xp of connectivity one cations. MacMillan, London (1976).constructed in the Introduction is strongly irreducible and [3] F. Harary, The maximum connectivity of a graph. Proc.weakly 1-irreducible for each p . But, unless p Å 2, no Nat. Acad. Sci. U.S.A. 48 (1962) 1142–1146.pair of nodes of Xp have the same neighbor set, and, [4] S. Lawrencenko and J. Mao, A sharp lower bound for thetherefore, Xp cannot be obtained from a tree as described number of connectivity-redundant nodes. Ars Combin., to

appear.above.

787/ 8u19$$0787 10-13-97 23:51:43 netwa W: Networks