Fuzzy Graphs and Fuzzy Hypergraphs - Part1'

Fuzzy Graphs and
Fuzzy Hypergraphs

Physica- Verlag

A Springer-Verlag Company

John N. Mordeson Premchand s. Nair

Fuzzy Graphs and Fuzzy Hypergraphs

With 29 Figures and 10 Tables

Physica- Verlag

A Springer-Verlag Company

John N. Mordeson Director

Center for Research in Fuzzy Mathematics and Computer Science

Premchand S. Nair

Associate Professor

Department of Mathematics

and Computer Science

Creighton University

Omaha, Nebraska 68178

USA

E-mail: mordes@creighton.edu psnair@creighton.edu

ISSN 1434-9922

ISBN 3-7908-1286-2 Physica-Verlag Heidelberg New York

Cataloging-in-Publication Data applied for

Die Deutsche Bibliothek - CIP-Einheitsaufnahme

Mordeson, John N.: Fuzzy graphs and fuzzy hypergraphs: with 10 tables I John N. Mordeson; Premchand S. Nair. - Heidelberg; New York: Physica-Verl., 2000

(Studies in fuzziness and soft computing; Vol. 46) ISBN 3-7908-1286-2

This work is subject to copyright. All rights are reserved. whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Physica- Verlag. Violations are liable for prosecution under the German Copyright Law.

Physica- Verlag is a company in the specialist publishing group BertelsmannSpringer. © Physica- Verlag Heidelberg 2000

Printed in Germany

The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

Hardcover Design: Erich Kirchner, Heidelberg

SPIN 1076 I 577 88/2202-5 4 3 2 I 0 - Printed on acid-free paper

FOREWORD

In the course of fuzzy technological development, fuzzy graph theory was identified quite early on for its importance in making things work. Two very important and useful concepts are those of granularity and of nonlinear approximations. The concept of granularity has evolved as a cornerstone of Lotfi A.Zadeh's theory of perception, while the concept of nonlinear approximation is the driving force behind the success of the consumer electronics products manufacturing.

It is fair to say fuzzy graph theory paved the way for engineers to build many rule-based expert systems. In the open literature, there are many papers written on the subject of fuzzy graph theory. However, there are relatively books available on the very same topic. Professors' Mordeson and Nair have made a real contribution in putting together a very comprehensive book on fuzzy graphs and fuzzy hypergraphs. In particular, the discussion on hypergraphs certainly is an innovative idea.

For an experienced engineer who has spent a great deal of time in the laboratory, it is usually a good idea to revisit the theory. Professors Mordeson and Nair have created such a volume which enables engineers and designers to benefit from referencing in one place. In addition, this volume is a testament to the numerous contributions Professor John N. Mordeson and his associates have made to the mathematical studies in so many different topics of fuzzy mathematics.

The Center for Research in Fuzzy Mathematics and Computer Science, under the direction of Dr. John N. Mordeson, is one of the earliest of these establishments in the world. The scholarly and academic products that

VI FOREWORD

have grown out from the center certainly are very impressive indeed, both in terms of quality as well as quantity.

In a sense, fuzzy mathematics is a generalization of traditional mathematics. In this regard, I have no doubt that Professor John N. Mordeson and his associates will be recognized as important leading researchers and the Center for Research in Fuzzy Mathematics and Computer Science will have its place in the annals of fuzzy theory as an important innovation and institution.

Paul P. Wang Duke University

PREFACE

In 1965, L. A. Zadeh introduced the concept of a fuzzy subset of a set as a way for representing uncertainty. Zadeh's ideas stirred the interest of researchers worldwide. His ideas have been applied to a wide range of scientific areas. Theoretical mathematics has also been touched by the notion of a fuzzy subset. \Ve consider two areas of mathematics here.

The book deals with fuzzy graph theory and fuzzy hypergraph theory.

The book is based on papers that have appeared in journals and conference proceedings. The purpose of this book is to present an up to date account of results from these two areas and to give applications of the results. The book should be of interest to research mathematicians and to engineers and computer scientists interested in applications. For the purpose of a comprehensive presentation of fuzzy graph theory, we include not only much of what appears in volume 20 of this series, but also a greatly expanded version.

In Chapter 1, basic concepts of fuzzy subset theory are given. The notion of a fuzzy relation and its basic properties are presented. The concept of a fuzzy relation is fundamental to many applications given, e. g., cluster analysis and pattern classification. Chapter 1 is based primarily on the work of Rosenfeld and Yeh and Bang.

Chapter 2 presents many concepts and theoretical results of fuzzy graphs.

The material from this chapter is the result of the work of many authors including that of the authors of this book. However much of the work is an outgrowth of the ideas of Rosenfeld. We acknowledge the authors at the beginning of each section. This chapter deals with the fuzzification of such concepts as paths, connectedness, bridges, cut vertices, trees, forests,

viii PREFACE

cut sets, chords, cotrees, twigs, f-chains, cocycles, line graphs, intersection graphs, and interval graphs.

In Chapter 3, applications of fuzzy graph theory are presented. Here again many of the results of this chapter are based on the work of Rosenfeld and Yeh and Bang. Applications of fuzzy graphs to cluster analysis and database theory are presented. Applications of fuzzy graphs to the problem concerning group structure are also given.

In Chapter 4, we present theoretical aspects of fuzzy hypergraph theory with applications to portfolio management, managerial decision making with an example to waste management, and to neural cell-assemblies. The results of this chapter are taken mainly from the work of Goetschel and his coauthors. We have reorganized Goetschel's work and added some examples. This chapter deals with the concepts of fuzzy transversals of fuzzy hypergraphs, colorings of fuzzy hypergraphs, and intersecting fuzzy hypergraphs. In 1982, Z. Pawlak introduced the idea of a rough set in order to provide a systematic approach for the study of indiscernibility of objects. We show how (fuzzy) hypergraphs and rough sets are related in such a way that ideas may be carried back and forth between the two areas.

John N. Mordesoti Premchand S. Nair

ACKNOWLEDG11ENTS

The authors are grateful to the editorial and production staffs of PhysicaVerlag, especially Janusz Kacpryzk, Martina Bihn, and Gabriele Keidel. We are indebted to Paul Wang and Hu Cheng-ming for their support of fuzzy mathematics. We are also appreciative of the support of Frs. Michael Proterra and Albert Agresti, Deans, Creighton College of Arts and Sciences and Dr. and Mrs. George Haddix. The first author dedicates the book to his children-in-law, James, Brigid, Todd, and Christy. The second author is very grateful to his wife, Suseela, and his parents, Mr. Sukumaran Nair and Ms. Sarada Devi, for supporting his dreams.

CONTENTS

FOREWORD

v

PREFACE

vii

ACKNOWLEDGMENTS

ix

1 FUZZY SUBSETS

1.1 FUzzy Relations . . . . . . . . 1.2 FUzzy Equivalence Relations . 1.3 Pattern Classification

1.4 Similarity Relations

1.5 References ....

1 4 6 9

11 16

2 FUZZY GRAPHS 19

2.1 Paths and Connectedness 20

Bridges and Cut Vertices 21

Forests and Trees . . . . . 22

Trees and Cycles . . . . . 25

A Characterization of Fuzzy Trees 26

(Fuzzy) Cut Sets . . . . . . . . . . 31

(Fuzzy) Chords, (Fuzzy) Cotrees, and (Fuzzy) Twigs . 33

(Fuzzy) I-Chain with Boundary 0, Coboundary, and Cocyles 35 (Fuzzy) Cycle Set and (Fuzzy) Cocycle Set 36

2.2 Fuzzy Line Graphs . . . . . . . . . . . . . . . . . . . . . .. 40

XII

CONTENTS

2.3 Fuzzy Interval Graphs Fuzzy Intersection Graphs Fuzzy Interval Graphs ..

The Fulkerson and Gross Characterization . The Gilmore and Hoffman Characterization

2.4 Operations on Fuzzy Graphs ....

Cartesian Product and Composition

Union and Join .

2.5 On Fuzzy Tree Definition 2.6 References.........

45 46 47 49 51 62 62 66 70 78

3 APPLICATIONS OF FUZZY GRAPHS 83

3.1 Clusters..... 86

3.2 Cluster Analysis 87

Cohesiveness 92

Slicing in Fuzzy Graphs 93

3.3 Application to Cluster Analysis 97

3.4 Fuzzy Intersection Equations 102

Existence of Solutions . . . . . 102

3.5 Fuzzy Graphs in Database Theory 108

Representation of Dependency Structure T(X, Y) 111

3.6 A Description of Strengthening and Weakening Members of

a Group. . . . . . . . . . . . . . . . 112

Connectedness Criteria. . . . . . . . 114

Inclusive Connectedness Categories . 117

Exclusive Connectedness Categories 118

3.7 An Application to the Problem Concerning Group Structure 120

Connectedness of a Fuzzy Graph . . . . . . . . . . . . . 121

Weakening and Strengthening Points a Directed Graph. 123

3.8 References.......................... 129

4 FUZZY HYPERGRAPHS 135

4.1 Fuzzy Hypergraphs . . . . . . . . . . . . . 135

4.2 Fuzzy Transversals of Fuzzy Hypergraphs 141

Properties of Tr(H) . . . . . . 153

Construction of 'H.8 • • • • • • • 158

4.3 Coloring of Fuzzy Hypergraphs 170

~-degree Coloring Procedures . 179

Chromatic Values of Fuzzy Colorings. 194

4.4 Intersecting Fuzzy Hypergraphs . . . . 199

Characterization of Strongly Intersecting Hypergraphs 205

Simply Ordered Intersecting Hypergrapbs 207

'H.-dominant Transversals 212

4.5 Hebbian Structures. . . 217

4.6 Additional Applications 221

CONTENTS xiii

4.7 References............................ 229

LIST OF FIGURES LIST OF TABLES LIST OF SYMBOLS INDEX

233 235 237 243

1

FUZZY SUBSETS

In 1965, Lofti Zadeh published his seminal paper "Fuzzy Sets" [11] which described fuzzy set theory and consequently fuzzy logic. The purpose of Zadeh's paper was to develop a theory which could deal with ambiguity and imprecision of certain classes or sets in human thinking, particularly in the domains of pattern recognition, communication of information, and abstraction. This theory proposed making the grade of membership of an element in a subset of a universal set a value in the closed interval [0,1] of real numbers.

Zadeh's ideas have found applications in computer science, artificial intelligence, decision analysis, information science, system science, control engineering, expert systems, pattern recognition, management science, operations research, and robotics. Theoretical mathematics has also been touched by fuzzy set theory. The ideas of fuzzy set theory have been introduced into topology, abstract algebra, geometry, graph theory, and analysis.

Before introducing the concept of a fuzzy subset, we review briefly some basic properties of sets, relations, and functions. We assume the reader is familiar with the basic ideas from set theory. Let S be a set and let A and B be subsets of S. We use the notation A U B and A n B to denote the union and intersection of A and B, respectively. We also let B \ A denote the relative complement of A in B. The (relative) complement of A in S, S \ A, is sometimes denoted by AC when S is understood. Then it is easily verified that (AUB)C = AcnBc and (AnB)C = AcuBc. These equations are known as DeMorgan's Laws.

Let x be an element of S. We write x E A if x is an element of A. otherwise we write x ¢ A. We use the notation A ~ B or B ;2 A to denote

2 1. FUZZY SUBSETS

that A is a subset of B. IT A ~ B, but there exists x E B such that x rt A, then we write A c B or B ::> A and say that A is a proper subset of B. The cardinality of A is denoted by IAI or card{A). The power set of A, written p{A), is defined to be the set of all subsets of A, i. e., p{A) = {U I U ~ A}. A partition of S is a set 'P of nonempty subsets of S such that 'r/U, V E P, either (I) U = V or Un V = 0, the empty set, and (2) S = u U.

UEP

We let N denote the set of positive integers, Z the set of integers, and IR

the set of real numbers.

Let X and Y be sets. IT x E X and y E Y, then (x, y) denotes the ordered pair of x with y. The Cartesian cross product of X with Y is defined to be the set {(x, y) I x E X, Y E Y} and is denoted by X x Y. We occasionally write X2 for X x X. In fact, for n EN, n ~ 2, we let X" denote the set of all ordered n-tuples of elements from X. A relation R of X into Y is a subset of X x Y. Let R be such a relation. Then the domain of R, written Dom(R), is {x E X I 3y E Y such that (x,y) E R} and the image of R, written Im(R), is {y E Y I 3x E X such that (x, y) E R}. IT (x, y) E R, we sometimes write xRy or R(x) = y. If R is a relation from X into X, we say that R is a relation on X. A relation R on X is called

(1) reflexive if'r/x E X, (x, x) E R;

(2) symmetric if'r/x,y E X, (x,y) E R implies (y,x) E R;

(3) transitive if 'r/x, y, Z E X, (x, y) and (y, z) E R implies (x, z) E R.

A relation R on X which is reflexive, symmetric, and transitive is called an equivalence relation. If R is an equivalence relation on X, we let [xl denote the equivalence class of x with respect to R. Hence [xl = {a E X I aRx}. If R is an equivalence relation on X, then {[xli x E X} is a partition of X. Also if P is a partition of X and R is the relation on X defined by 'r/x, y E X, (x, y) E R if 3U E 'P such that x, y E U, then R is an equivalence relation on X whose equivalence classes are exactly those members of P.

Let R be a relation on X. Then R is called antisymmetric if 'r/x, y E X, (x, y) E R and (y, x) E R implies x = y. IT R is a reflexive, antisymmetric, and transitive relation on X, then R is called a partial order on X and X is said to be partially ordered by R.

Let R be a relation of X into Y and T a relation of Y into a set Z. Then the composition of R with T, written ToR, is defined to be the relation {(x, z) E X x Z I 3y E Y, such that (x, y) E Rand (y, z) E T}.

Suppose that f is a relation of X into Y such that DomU) = X and 'r/x, x' E X, x = x' implies f(x) = f(x'). Then f is called a function of X into Y and we write f ; X --+ Y. Let f be a function of X into Y. Then f is sometimes called a mapping and f is said to map X into Y. If 'r/y E Y, 3x E X such that f(x) = y, then f is said to be onto Y or to map X onto Y. If 'r/x, x' E X, f(x) = f(x') implies that x = x', then f is said to be one-to-one and f is called an injection. If f is a one-to-one function of X

1. FUZZY SUBSETS 3

onto Y, then f is called a bijection. If 9 is a function of Y into a set Z, then the composition of f with g, go t. is a function of X into Z which is one-to-one if f and 9 are and which is onto Z if f is onto Y and 9 is onto Z. If Im(f) is finite, the we say that f is finite-valued. We say that an infinite set X is countable if there exists a one-to-one function of X onto N; otherwise we call X uncountable.

We now introduce the notion of a fuzzy subset of a set S. A fuzzy subset of S is a mapping /.L : S -+ [0,1]' where [0,1] denotes the set {t E IR I o ::; t ::; I}. We think of /.L as assigning to each element xES a degree of membership, 0 ::; /.L( x) ::; 1. Let /.L be a fuzzy subset of S. We let /.Lt = {x E S I /.L(x) 2: t} for all t E [0,1]. The sets /.Lt are called level sets or t-cuts of /.L. We let supp(/.L) = {x E S I /.L(x) > O]. We call supp(/.L) the support of /.L. A fuzzy set /.L is nontrivial is supp(u) i= 0. The set of all fuzzy subsets of S is denoted by and is called the fuzzy power set of S. Throughout we use the notation V for supremum and 1\ for infimum. Let h be the function of J~(S) into [0,1] defined by h(/.L) = V {/.L(x) I x E S}V/.L E J~(S). Then h(/.L) is called the height of /.L.

Definition 1.1 Let /.L, v be two fuzzy subsets of S. Then

(1) /.L ~ v if /.L(x) ::; vex) for all XES,

(2) /.L c v if /.L(x) ::; vex) for all xES and there exists at least one xES such that /.L(x) < v(x),

(3) /.L = v if /.L(x) = vex) for all xES.

Definition 1.2 Let /.L, v be any two fuzzy subsets of S. Then /.L U v is the fuzzy subset of S defined by

(/.L U v)(x) = /.L(x) V vex) for all xES

and /.L n v is the fuzzy subset of S defined by

(/.L n v)(x) = /.L(x) 1\ vex) for all xES.

Definition 1.3 Let /.L be any fuzzy subset of S. Then /.LC is the fuzzy subset of S defined by

/.LC(x) = 1 - /.L(x) for all xES.

If S is a collection of fuzzy subsets of S, we define the fuzzy subset n ~ l;ES (intersection) of S by Vx E S, (n ~)(x) = I\{~(x) I ~ E S} and the fuzzy l;ES

subset U ~ (union) of S by Vx E S, (U ~)(x) = V{~(x) I ~ E S}.

l;ES l;ES

4 1. FUZZY SUBSETS

Let A be a subset of S. Define XA : S -+ [0,11 by XA(x) = 1 if x E A and XA(X) = 0 if xES \ A. Then XA is called the characteristic function of A in S. Now Xs(x) = 1 for all xES and X0(x) = 0 for all xES. Let J.L, u, and € be fuzzy subsets of S. Then we have the following properties.

(1)J.LUv=vUJ.L (2) J.L n v = VnJ.L (3) J.LU X0 = J.L (4) J.Ln X0 = X0 (5) J.LU xs = Xs (6) J.Ln xs = J.L (7) J.LU J.L = J.L

(8) J.L n J.L = J.L

(9) J.LU(vU€) = (J.LUv)U€ (10) J.Ln(vn€) = (J.Lnv)n€

(11) J.Ln(vU€) = (J.Lnv)u(J.Ln{) (12) J.L U (v n €) = (J.L U v) n (J.L U {) (13) (J.LUvY=J.LcnVc

(14) (J.L n v)C = J.Lc U VC

(15) (J.LC)C = J.L

It is important to note that the properties J.LnJ.Lc = X0 and J.LUJ.Lc = Xs do not hold in general. In logic, the former property is known as the law of contradiction while the latter is known as the law of the excluded middle. Additional properties involving fuzzy subsets can be found in [1,2,4,5,6,7,141.

1.1 Fuzzy Relations

Much of the material in the first three sections is based on the work of Rosenfeld, [8]. Let Sand T be two sets and let J.L and v be fuzzy subsets of S and T, respectively. Then a fuzzy relation p from the fuzzy subset J.L into the fuzzy subset v is a fuzzy subset p of S x T such that p(x, y) ~ J.L(x) I\. v(y), '<Ix E S and YET. That is, for p to be a fuzzy relation, we require that the degree of membership of a pair of elements never exceed the degree of membership of either of the elements themselves. Also, the restriction p(x,y) ~ J.L(x) I\. v(y),'<Ix E Sand yET allows pt to be a relation from J.Lt into vt for all t E [0,1] and for supp(p) to be a relation from supp(J.L) into supp(v).

There are three special cases of fuzzy relations which are extensively found in the literature:

(1) S = T and J.L = u, In this case, p is said to be a fuzzy relation on J.L.

Note that p is a fuzzy subset of S x S such that p(x, y) ~ J.L(x) I\.J.L(Y).

(2) J.L(x) = 1.0 for all xES and v(y) = 1.0 for all yET. In this case, pis said to be a fuzzy relation from S into T.

(3) S = T,J.L(x) = 1.0 for all xES and v(y) = 1.0 for all YET. In this case, p is said to be a fuzzy relation on S.

There are many applications in which a fuzzy relation on a fuzzy subset is quite useful. Also, any result we obtain is clearly true for fuzzy relations

1.1 FUzzy Relations 5

on a set. We devote this and next two sections of this chapter to fuzzy relations on a fuzzy subset. The last section of this chapter is devoted to the study of cases 2 and 3.

Let p be a fuzzy relation on J.L. Then p is called the strongest fuzzy relation on J.L if and only if for all fuzzy relations t<l on J.L, Vx, YES, w(x, y) ::; p(x, y). The converse problem may also arise in practice. That is, we know the strength of the pairs and we want to compute the minimum strength required for the elements themselves. For a given fuzzy subset p of S x S, the weakest fuzzy subset J.L of S on which p is a fuzzy relation is defined by J.L(x) = V{p(x,y) V p(y,x)ly E S} for all xES. That is, if v is a fuzzy subset of Sand p is a fuzzy relation on u, then J.L ~ t/,

\Ve now introduce some important operations on fuzzy relations.

Definition 1.4 Let p : S x T -+ [0,1] be a fuzzy relation from a fuzzy subset J.L of S into a fuzzy subset v of T and w : T x U -+ [0,1] be a fuzzy relation from a fuzzy subset v of T into a fuzzy subset ~ of U. Define po c: : S x U -+ [0,1] by

po w(x, z) = V{p(x, y) 1\ t<l(y, z)ly E T}

for all XES, z E U. Then po t<l is called the composition of p with t<l.

Proposition 1.1 Let p, J.L, ro and vas defined in Definition 1.4. Then pet» is a fuzzy relation from J.L into ~.

Proof. Let XES, yET, and z E U. Then p(x,y) ::; J.L(x) 1\ v(y) and w(y, z) ::; v(y) 1\ ~(z). Hence p(x, y) 1\ r;;}(y, z) ::; J.L(x) 1\ v(y) 1\ ~(z). Thus (p 0 w)(x, z) = V {p(x, y) 1\ ro(y, z) lyE T} ::; J.L(x) 1\ ~(z) .•

We see that the composition of p with ro is a fuzzy relation from a fuzzy subset J.L of S into a fuzzy subset ~ of U. A closer look at the definition of the composition operation reveals that po t» can be computed similar to matrix multiplication, where the addition is replaced by V and the multiplication is replaced by 1\. Since composition is associative,we use the notation p2 to denote the composition pop, pk to denote pk-l op, k > 1. Define pOO(x, y) = V{pk(x,y)lk = 1,2, ... } for all X,y E S. Finally, it is convenient to define pO(x, y) = 0 if x 1= y and pO(x, y) = J.L(x) otherwise, for all x, y E S.We have introduced three binary operations. \Ve now introduce a unary operation on a fuzzy relation. Given a fuzzy relation p on a fuzzy subset J.L of S, define the fuzzy relation pc on J.L by pC(x,y) = 1- p(x,y) for all x,y E S.

Definition 1.5 Let p : S x T -+ [0,1] be a fuzzy relation from a fuzzy subset J.L of S into a fuzzy subset v of T. Define the fuzzy relation p-l T x S -+ [0,1] of v into J.L by p-l(y, x) = p(x, y) for all (y, x) E T x S.

6 1. FUZZY SUBSETS

Theorem 1.2 Let T, n , P and r;;:; be fuzzy relations on a fuzzy subset f.l of a set S. Then the following properties hold.

(1) pUr;;:;=tvUp (2) p n tv = r;;:; n p (3) p = (pC)C

(4) 7rU(pUr;;:;) = (7rUp)Utv (5) 7rn(pntv) = (7rnp)ntv (6) 7rO(potv) = (7rOp) o zn

(7) 7rn(pUtv) = (7rnp)u(7rntv) (8) 7ru(pnr;;:;) = (7rUp)n(7rur;;:;) (9) (pUr;;:;)C = r;;:;cnpc

(10) (p n tv)C = r;;:;c U pC

(11) For all t E [0,1], (p U r;;:;)t = pt U r;;:;t (12) For all t E [0,1]' (p n tv)t = pt n r;;:;t

(13) For all t E [0,1], (pOtv)t ;2 pt Otvt and if S is finite, (pOtv)t = pt Otvt. (14) If T ~ P and 7r ~ tv then T U 7r ~ P U tv

(15) If T ~ P and 7r ~ r;;:; then Tn 7r ~ P n tv

(16) If T ~ P and 7r ~ tv then TO 7r ~ p o tv

Proof. We provide the proofs for (13) and (16). Let x, z E S.

(13) (x, z) E (p 0 tv)t <=> (p 0 tv)(x, z) 2: t -¢::: 3y E S such that p(x, y) 1\ tv(y, z) 2: t <=> 3y E S such that p(x, y) 2: t and r;;:;(y, z) 2: t <=> 3y E S such that (x,y) E pt and (y,z) E ::;vt <=> (x,Z) E pt 0 tvt. The implication becomes an equivalence if S is finite.

(16) (T07r)(X,Z) = V{T(X,y) 1\7r(y,z) lyE S}:::; V{p(x,y) 1\::;V(Y,z) I yES} = (pOtv)(x,z) for all x,z E S .•

1.2 Fuzzy Equivalence Relations

In this section p and tv are fuzzy relations on a fuzzy subset f.l of S. It is quite natural to represent a fuzzy relation in the form of a matrix. We now use the matrix representation of a fuzzy relation to explain the properties of

1.2 Fuzzy Equivalence Relations 7

a fuzzy relation. In particular, we shall use the term "diagonal" to represent the principal diagonal of the matrix.

We call p reflexive (on J..L) if p(x, x) = J..L(x) for all xES. If p is reflexive, on Xs, we call p reflexive. If p is reflexive on J..L, then p(x, y) ~ J..L(x) /\J..L(y) ~ J..L(x) = p(x, x) and it follows that "any diagonal element of p is larger than or equal to any element in its row". Similarly, "any diagonal element is larger than or equal to any element in its column". Conversely, given a fuzzy relation p on J..L such that "any diagonal element is larger than or equal to any element in its row and column", define a fuzzy subset v of S as v(x) = p(x, x), Vx E S. Then v is the weakest fuzzy subset of S such that p is a fuzzy relation on u, Further, p is reflexive on u .

Fuzzy reflexive relations have some interesting algebraic properties.

Theorem 1.3 Let p and w be fuzzy relations on a fuzzy subset J..L of S. Then the following properties hold.

(1) If p is reflexive, w ~ wop and w ~ p o ta. (2) If p is reflexive, p ~ p2.

(3) If p is reflexive, pO ~ P <; p2 <; p3 ~ ... ~ poo.

(4) If p is reflexive, pO(x,x) = p(x,x) = p2(x,x) = p3(x,x) pOO(x, x) = J..L(x) , Vx E S.

(5) If p and w are reflexive, so is p o ta and tir o p,

(6) If p is reflexive, then pt is a reflexive relation on J..Lt for all t E [0, 1J .

Proof. Let x, z E S.

(1) (p 0 !V)(x, z) = V{p(x, y) /\ w(y, z) lyE S} ~ p(x, x) /\ w(x, z) = J..L(x) /\ w(x, z). Since w(x, z) ~ J..L(x) /\ J..L(z) , J..L(x) /\ w(x, z) = w(x, z). Thus -;;;; <; p o 'W. Similarly, 'W <; 'W 0 p.

(2) Choose wasp in (1).

(3) Choose wasp, p2,p3 and so on in (1).

(4) Note that p(x, x) = J..L(x) , Vx E S. Assume that pn(x, x) = J..L(x) , Vx E

S. Now for all XES, pn+l(x, x) = v{p(x, y) /\ pn(y, x) lyE S} ~ V{J..L(x) /\ J..L(x) lyE S} = J..L(x) and pn+l(X, x) = V{p(x,y) /\ pn(y,x) lyE S} ~ p(x, x) /\ pn(x, x). Hence pn+l(x, x) = J..L(x) , Vx E S.

(5) (pOw)(x, x) = V{p(x, y)/\w(y, x) lyE S} ~ V{J..L(x)/\J..L(x) lyE S} = J..L(x) and (pow)(x,x) = V{p(x,y) /\w(y,x) lyE S} ~ p(x,x) /\w(x,x) = J..L(x) /\ J..L(x) = J..L(x). The proof that tii c p is reflexive is similar.

(6) If x E J..Lt, then p(x,x) = J..L(x) ~ t and thus (x,x) E pt.

We call p symmetric if p( x, y) = p(y, x), for all x, yES. In other words, p is symmetric if the matrix representation of p is symmetric (with respect to the diagonal).

8 1. FUZZY SUBSETS

Theorern 1.4 Let p and rAJ be fuzzy relations on a fuzzy subset p, of S. Then the following properties hold.

(1) If p and 'I:V are symmetric, then p 0 'I:V is symmetric if and only if po ti: = r:;; 0 p.

(2) If p is symmetric, then so is every power of p.

(3) If p is symmetric, then l is a symmetric relation on Jit for all t E [0,1].

Proof. (1) (p 0 -:v) (x, z) = (p 0 'I:V)(z, x) ¢:} V{p(x, y) /\ r:;;(y, z) lyE S} = V{p(z, y) /\ 'I:V(y, x) lyE S} ¢:}

V{p(x,y)/\r:;;(y,z) lyE S} = V{rAJ(y, x)/\p(z, y) lyE S} ¢:} P0'I:V = rAJOp. (2) Assume that p" is symmetric for n E N. Then pn+l(x, z) = V{p(x, y)/\ pn(y,z) lyE S} = V{p(y,x)Apn(z,y) lyE S} = V{pn(z,y)Ap(y,x) lyE S} =pn+l(z,x).

(3) Let 0::; t ::; 1. Suppose (x,z) E pt. Then p(x,z) 2: t. Since p is symmetric, p(z, x) 2: t. Thus (z, x) E l .•

We call p transitive if p2 ~ p. It follows that poo is transitive for any fuzzy relation p.

Theorem 1.5 Let 7r, P and rAJ be fuzzy relations on a fuzzy subset p, of S. Then the following properties hold.

(1) If p is transitive and 7r ~ p, rAJ ~ p, then 7r 0 rAJ ~ p. (2) If p is transitive, then so is every power of p.

(3) If p is transitive, 'I:V is reflexive and 'I:V ~ p, then p o 'I:V = 'I:V 0 P = p. (4) If p is reflexive and transitive, then p2 = p.

(5) If p is reflexive and transitive, then po ~ p = p2 = p3 = ... = poo.

(6) If p and rAJ are transitive and p c t» = rAJ 0 p, then p o 'I:V is transitive.

(7) If p is symmetric and transitive, then p(x,y) ::; p(x,x) and p(y,x)::; p(x, x), for all x, yES.

(8) If p is transitive, then for any 0 ::; t ::; 1, l is a transitive relation on p,t.

Proof. (1) (7rOw)(x,z) = V{7r(x,y)Ar:v(y,z) lyE S}::; V{p(x, y)/\p(y, z) I yES} = p2(x,z)::; p(x,z). Hence 7rOr:v ~ p.

(2) Assume that p" 0 p" ~ p", Then pn+l 0 pn+l = p2n+2 = p2n 0 p2 ~ p" 0 p = pn+l.

1.3 Pattern Classification 9

(3) By (1), taking tt to be p, poco ~ p. (poro)(x, z) = V{p(x, y)l\ro(y, z) I yES} ~ Vp(x, z) 1\ ro(z, z) = p(x, z) 1\ J.L(z) = p(x, z). Hence po ro = p. Similarly, ro 0 p = p.

(4) Choose ro as p in (3).

(5) Note that p = p2 by (4). Assume that p" = pn+l, for n > 1. Hence p" 0 p = pn+l 0 p. That is, pn+l = pn+2.

(6) (p 0 ro) 0 (p 0 ro) = po (ro 0 p) 0 ro = po (p 0 ro) 0 ro = p2 0 ro2 ~ p o ro.

Hence p o ro is transitive.

(7) Since p is transitive, pop ~ p. Hence (p 0 p)( x, x) :5 p( x, x). That is, v{p(x,y) I\p(y,x) lyE S}:5 p(x,x). Since p is symmetric, V{p(x,y) 1\ p(x,y) lyE S} :5 p(x,x). Thus p(x,y) :5 p(x,x). Since p is symmetric, p(y, x) :5 p(x, x).

(8) Let 0:5 t :5 1. Let (x, y), (y, z) E pt. Hence p(x, y) ~ t and p(y, z) ~ t.

Therefore, p(x, z) = V{p(x, w) 1\ p(w, z) I w E S} ~ p(x, y) 1\ p(y, z) ~ t. Thus (x, z) E pt .

•

A fuzzy relation p on S which is reflexive, symmetric, and transitive is

called a fuzzy equivalence relation on S.

1.3 Pattern Classification

Let S be a set whose elements we think of as patterns. A fuzzy classification p on S is a symmetric fuzzy relation on S such that p( x, x) = 1 for all xES.

Since p is reflexive, p ~ ~ ~ p3 ~ ... ~ poo. Note that poo is a fuzzy equivalence relation. So for any 0 :5 t :5 1, (pOO)t is an equivalence relation on S. Let pt the partition of S induced by the equivalence relation (pOO)t.

Theorem 1.6 Let p be a fuzzy relation on S. Define 8 from S x S into lit by '<ix, YES, 8(x, y) = 1 - pOO(x, y). Then '<ix, y, z E S:

(1) 8(x,y) = 0 if and only if x = y.

(2) 8(x,y) = 8(y,x).

(3) 8(x, z) :5 8(x, y) + 8(y, z).

That is, 8 is a metric on S. Proof. Let x, y, z E S.

(1) pOO(x, y) = 1 if and only if x = y <=> 1 - pOO(x, y) = 0 if and only if x = y <=> 8(x,y) = 0 if and only if x = y.

(2) 8(x, y) = 1 - pOO(x, y) = 1 - pOO(y, x) = 8(y, x).

(3) poo is transitive =? pOO(x, z) ~ pOO(x, y) 1\ pOO(y, z) ~ pOO(x, y) + pOO(y, z) -1 =? pOO(x, z) -1 ~ pOO(x, y) -1 + pOO(y: z) -1 =? 1- pOO(x, z) ~ 1 - pOO(x, y) + 1 - pOO(y, z) =? 8(x, z) ~ 8(x, y) + 8(y, z) .•

10 1. FUZZY SUBSETS

Example 1.1 Let S = {XI,X2,X3,X4,XS} and define the symmetric fuzzy relation p on S as follows:

XI X2 X3 X4 Xs
Xl 1.0
X2 0.8 1.0
X3 0.0 0·4 1.0
X4 0.1 0.0 0.0 1.0
Xs 0.2 0.9 0.0 0.5 1.0
Now poo = p3 is given by
Xl X2 X3 X4 Xs
XI 1.0
X2 0.8 1.0
X3 0·4 0·4 1.0
X4 0.5 0.5 0·4 1.0
Xs 0.8 0.9 0·4 0.5 1.0 and we have the partitions

if 0 s t s 0.4 if 0.4 < t ~ 0.5 if 0.5 < t ~ 0.8 if 0.8 < t ~ 0.9 if 0.9 < t s 1.0

Thus there are many partitions possible and depending upon the level of detail, one could classify the patterns based on equivalence relations. Note that if s 2': t, then P" is a refinement of pt.

We now present an experiment done by Tamura, Higuchi and Tanaka [9]. Portraits obtained from 60 families were used in their experiment, each family of which was composed of between four and seven members. They chose portraits because even though parents may not resemble each other, they may be connected through their children, and consequently they could classify the portraits into families. They first divided the 60 families into 20 groups, each of which was composed of 3 families. Each group was, on the average, composed of 15 members. The portraits of each group were presented to a different student to assign the values of the subjective similarity p(x, y) between all pairs on a scale of 1 to 5. They used the 5 rank representation instead of a continuous value representation because it has been proved that human beings cannot make distinctions into more than 5 ranks. Twenty students were involved in the experiment. Since the levels of the subjective values are different according to individuals, the threshold was determined in each group as follows. As they lowered the threshold,

1.4 Similarity Relations 11

the number of classes decreased. Hence, under the assumption that the number of classes c to be classified was mown to be 3, while lowering the threshold they stopped at the value which divided the patterns into 3 classes (collection of the patterns composed of more than 2 patterns that have a stronger relation than A with each other) and some nonconnected patterns. However, when some p(x, y) are equal, sometimes there is no threshold by which the patterns can be divided into exactly c given classes. In such a case, they divided them into exactly c classes by stopping the threshold at the value where the patterns are divided into less than c classes and separating some connections randomly that have a minimum p(x, y) greater than the threshold. The correctly classified rates, the misclassified rates, and the rejected rates of 20 groups were within the range of 50-94 percent, 0-33 percent, and 0-33 percent, respectively, and they obtained the correctly classified rate 75 percent of the time, the misclassified rate 13 percent, and the rejected rate 12 percent as the averages of the 20 groups. Here, since the classes made in this experiment have no label, they calculated these rates by making a one-to-one correspondence between 3 families and 3 classes, so as to have the largest number of correctly classified patterns.

We see that Tamura, Higuchi and Tanaka [9] have studied pattern classification using subjective information and performed experiments involving classification of portraits. The method of classification proposed here is based on the procedure of finding a path connecting 2 patterns. Therefore, this method may be combined with nonsupervised learning and may also be applicable to information retrieval and path detection.

1.4 Similarity Relations

In this section we will show that the concept of a similarity relation introduced by Zadeh [13] is derivable in much the same way as a fuzzy equivalence relation. Throughout this section we shall be dealing with a fuzzy relation on a set. The results in this section are from [9J.

Definition 1.6 Let p be a fuzzy relation on a set S. We define the following notions:

(1) P is e-reflexive if V XES, p(x, x) 2: e, where E E [0,1]. (2) pis irreflexive ifVx E S,p(x,x) = O.

(3) p is weakly reflexive if for all x, y in S and for all f E [0, IJ ,p(x,y) = f =? p(x.x) 2: E.

Note that the definition of a reflexive relation as a I-reflexive relation coincides with the definition of a reflexive relation in Section 1.2.

12 1. FUZZY SUBSETS

Lemma 1.7 If p is a fuzzy relation from 8 into T, then the fuzzy relation p o p-l is weakly reflexive and symmetric.

Proof. (pop-l)(x,x') = V{p(x,y) Ap-l(y,x') 1 yET}:::; V{p(x,y)A p(x,y)IYET}=

V{p(x,y) Ap-l(y,X) 1 yET} = (pop-l)(X,X). Hence pop-l is weakly

reflexive.

(pop-l)(X,x') = v{p(x,y)Ap-l(y,X') 1 yET} = V{p-l(y,x)Ap(x',y) I

yET}

= V{p(x',y) A p-l(y,x) 1 YET} = (p 0 p-l)(x',X). Thus po p-l is

symmetric .•

Let p be a weakly reflexive and symmetric fuzzy relation on 8. Define a family of non-fuzzy subsets FP as follows:

FP = {K ~ 81(30 < f :::; l)(Vx E S)[x E K ¢:} ('Ix' E K)[p(x, x') ~ fl]}·

Hence if we let

Ff = {K ~ SI(Vx E S)[x E K ¢:} ('Ix' E K)[p(x,x') ~ EI]},

then we see that fl :::; E2 :::} Ff, ~ Ff. where "~" denotes a covering relation, i.e., every element in Ff, is a subset of an element in Fi,..

A subset J of S is called e-complete with respect to p if V x, x' E J,p(x, x') ~ E. A maximal e-compleie set is one which is not properly contained in any other e-complete set.

Lemma 1.8 FP is the family of all maximal e-cotnplete sets with respect to p for 0 :::; E :::; 1. •

Proof. Let K E FP and x, z" E K. Then there exists 0 < E :::; 1 such that 'Ix' E K, p(x, x') ~ f. Thus p(x, x") ~ E. Hence K is e-complete. Let J be subset of X such that K ~ J and J is e-complete. Let x E J. Since J is e-complete, 'Ix' E K, p(x,x') ~ E. Since K E FP, x E K. Thus J ~ K. Hence K is maximal. Now let K be a maximal e-complete set. Let x E X. Then clearly x E K ¢:} V x' E K, p(x,x') ~ E. Thus K E FP.

Lemma 1.9 Whenever p(x, x') > 0, there is some e-compleie set K E FP such that {x, x'} ~ K. •

Proof. If x = x', then {x} is certainly e-complete for e = J-LR(x, x). Suppose that x i- x'. Then since p( x, x') = p( x' , x) by symmetry, and p( x, x) ~ p(x,x') and p(x',x') ~ p(x,x') by weak reflexivity, we see that {x,x'} is e-complete, where E = p(x, x'). Denote by C. the family of all e-complete sets C which contain {x,x'}. Then C. is not empty since {x,x'} E C e- It follows easily by Zorn's lemma that C. has a maximal element K. This


element is also maximal in the family of all €-complete sets since any set including K must also include {x,x'}. Hence K E FP by Lemma 1.8 .•

We note that sometimes a subclass of FP, satisfying the condition of Lemma 1.9, will cover the set S. For example, let p be the fuzzy relation on S = {a, b, e, d, e,!} given by the following matrix.

a b e d e I
a 1.0 0.3 0.4 0.0 0.4 0.3
b 0.3 1.0 0.2 0.3 0.0 0.4
e 0.4 0.2 1.0 0.3 0.5 0.0
d 0.0 0.3 0.3 1.0 0.0 0.0
e 0.4 0.0 0.5 0.0 1.0 0.0
I 0.3 0.4 0.0 0.0 0.0 1.0 We see that the family consisting of the three maximal complete sets {a,b,!}, {b,e,d}, and {a,e,e} satisfy the condition of Lemma 1.9, but it does not contain the maximal complete set {a, b, e}. Note for example that {a, b,!} is maximal €' -complete 'if 0 < €' ::; e where € = 0.3 since p( a, d) = p(b,e) = pU,e) = O. We have

! {{a}, {b}, {e}, {d}, {e}, {J}}

{{e,e}, {a},{b},{d},{/}}}

Ft = {{a, e, e}, {b,!}, {d}} {{a,b,/},{a,e,e},{b,d},{e,d}} {{a,b,e},{b,e,d},{a,b,/},{a,e,e}}

if 0.5 < E s 1 if 0.4 < E ::; 0.5 if 0.3 < € ::; 0.4 if 0.2 < € ::; 0.3 if 0 < € s 0.2

We also note that p is not transitive: p(b, c) = .2 i:. .3 = (.3/\ .4) V (1/\ .2) V (.2/\ 1) V (.3/\ .3) (0/\ .5) V (.4/\ 0) = V{p(b, y) /\ p(y, c) lyE S} = po p(b, c).

Recall that X0 is the characteristic function of 0 in S x S.

Lemma 1.10 IIp:f: X0 is a weakly reflexive and symmetric fuzzy relation on S, then there exists a set T and a fuzzy relation 7r from S into T such that p = 7r 0 7r-1 .

Proof. Denote by T the set {K* IKE FP}. We define a fuzzy relation tt from S to T as follows:

7r(x, K*) = {t if x E ~ and t is the largest number such that K EFt o otherwise,

If p(x, x') = t > 0, then by Lemma 1.9, there is an t-complete set K E FP such that {x, x'} ~ K. Since (7r07r-1 )(x, x') = i/o [7r(x, K*) /\ 7r(x', K*)I ~ t

= p(x, x'), we conclude that p ~ tt 0 7r-1.

14 1. FUZZY SUBSETS

Suppose now that (1T01T-1 )(x, x') = s. Then there exists K" E Fs such that 1T(x,K*) = 1T(x',K*). This means that {x,x'} ~ K and hence p(x, x') ~ s.

(s = (1T01T-1)(X,X') = V [1T(x,K*)I\1T(K*, x')] = V [1T(x,K*)I\1T(X',K*)].

K" K'

There exists K* such that either 1T(X, K*) = s and 1T(X', K*) ~ s or 1T(X, K*)

~ s and 1T(X', K*) = s. Now s is largest such that K EFt Hence 1T(X, K*) = s and 1T(X', K*) = s.) Therefore, 1T 0 1T-1 ~ p .•

Combining Lemmas 1.7 and 1.10, we have the following theorem.

Theorem 1.11 A fuzzy relation P"l X0 on a set S is weakly reflexive and symmetric if and only if there is a set T and a fuzzy relation 1T from S into T such that p = 1T 0 1T-1 .•

In the remainder of this section, we shall use the notation ¢> p to denote the fuzzy relation 1T defined above.

Definition 1.7 A cover C on a set S is a family of subsets Si,i E I, of S such that UiEI Si = S, where I is a nonempty index set.

Definition 1.8 Let p be a fuzzy relation from S into T. FaT E E [0,1] , we say that:

(1) P is e-determinate if [or each XES, there exists at most one yET such that p(x,y) ~ E.

(2) p is e-productive if for each XES, there exists at least one yET such that p(x, y) ~ E.

(3) p is an e-function if it is both e-deierminaie and e-productiue.

Lemma 1.12 If p is an e-refiexioe fuzzy relation on S, then ¢>p is e-productive and for each E' ::; E, F: is a cover of S.

Proof. Let 0< 15'::; E. Since for each x E X,p(x,x) ~ 15, and because {x} is 15 -complete, there is some K in F: such that x E K. Hence, F: is a cover of X. Also, by definition of ¢>p, x E K implies that ¢p(x, K*) ~ E which implies that ¢>p is E -productive .•

In the sequel, we use the term productive (determinate, reflexive, function) for I-productive (I-determinate, l-reflexive, I-function).

Corollary 1.13 If p is reflexive, then ¢>p is productive and each Ff (0 <: 15 ::; 1) is a cover of S .•


The following result is a consequence of Theorem 1.11 and Corollary 1.13.

Corollary 1.14 p is reflexive and symmetric relation on 5 if and only if there is a set T and a productive fuzry relation 7l" from 5 into T such that p = 7l"O7l"-l .•

Lemma 1.15 Let p be a weakly reflexive, symmetric and transitive fuzry relation on 5, and let ¢~ denote the relation ¢p whose range is restricted to Ft. That is, ¢~ equals ¢p on 5 x {K"'IK EFn . Then for each 0 < € :s 1, ¢~ is €-determinate and the elements of Ft are pairwise disjoint.

Proof. Let K and K' be two not necessarily distinct elements of Ff and assume that K n K' i= 0. For any ql E K n K', we have p(q,qd ?: €, for all q in K and p( ql, q') ?: €, for all q' in K'. Since p is transitive, we see that p(q,q')?: €, for all q E K, and q' E K'. Since p is weakly reflexive and symmetric, we conclude that K U K' is €-complete. However, since K and K' are maximal €-complete, we must conclude that K = K'. Hence, K =P K' => KnK' = 0. Suppose x E K where K EFf. Then ¢p(x, K*) ?: €, and since x cannot belong to any other sets in FR ,¢~ is €-determinate .•.

Definition 1.9 A similarity relation p on 5 is a fuzry relation on 5 which is reflexive, symmetric and transitive. p is called an €-similarity relation if it is f.-reflexive for some 0 < e :s 1, symmetric, and transitive.

Note that a similarity relation on 5 is merely a fuzzy equivalence relation on 5.

Since clearly reflexivity implies weak reflexivity, we have the following consequence of Lemmas 1.12 and 1.15.

Corollary 1.16 If p is a similarity relation on S, then for each 0 < € :s 1, Ff is a partition of 5 .•

We see that Corollary 1.16 says that every similarity relation p can be represented as U tl, where pt is the equivalence relation induced by the

t

partition Ff. It was noted in [14] that if the pt, 0 < t :s 1, are a nested sequence of distinct equivalence relations on 5 with tl > tz if and only if /1 ~ pt2, ptl is nonempty and the domain of pit is equal to the domain of pt2, then p = U tl is a similarity relation on 5, where

t

t t(x ) _ {t if (x, y) E pt

P ,y - 0 otherwise.

16 1. FUZZY SUBSETS

The following result, which is a straightforward consequence of Theorem 1.11 and Corollary 1.16, yields another characterization of a similarity relation.

Theorem 1.17 A relation p is an €-similarity (0 < € ::; 1) relation on a set S if and only if there is another set T and an €-function 7r from S into T such that p = 7r 0 7r-1 .•

Example 1.2 Let p be the fuz7:!J relation on S = {a, b, c, d, e, J} given by the following matrix, Mp.

a b c d e f
a 1.0 0.5 0.5 0.2 0.2 0.2
b 0.5 1.0 0.5 0.2 0.2 0.2
c 0.5 0.5 1.0 0.2 0.2 0.2
d 0.2 0.2 0.2 1.0 0·4 0·4
e 0.2 0.2 0.2 0·4 1.0 0·4
f 0.2 0.2 0.2 0·4 0·4 1.0 Now M/ = Mp. Thus p is transitive. Clearly, p is reftexioe and symmetric. We have

{{{a}, {b}, {c}, {d}, {e}, {In if 0.5 < €::; 1

FP = {{a,b,c},{d},{e},{ln if0.4<€::;0.5

e {{a,b,c}, {d,e,J}} if 0.2 < €::; 0.4

{X}. ifO<€::;0.2

Let € = 0.4. Then the € - function 7r : X x {K*IK El;6.4} --> [0,1]' is defined as follows: 7r{a,{a,b,c}*) = 7r{b,{a,b,c}*) = 7r{c,{a,b,c}*) = 0.5,7r{d, {d,e, J}*) = 7r{e,{d,e,J}*) = 7r(J,{d,e,J}*) = 0.4, and 7r{x, K*) = 0 otherwise.

1. 5 References

1. Dubois, D. and Prade, H., Fuzzy Sets and Systems: Theory and Applications, Mathematics in Science and Engineering, Vol. 144, Academic Press, Inc., Orlando, Florida, 1980.

2. Kandel, A., PuZ7:!J Mathematical Techniques with Applications, Addison-Wesley Pub. Co. 1986.

3. Kaufmann, A., Introduction to the Theory of PuZ7:!J Sets, Vol. 1, Academic Press, Inc., Orlando, Florida, 1973.

1.5 References 17

4. Kill, G.J., U. St. Clair, U.H., and Yuan, B., Fuzzy Set Theory, Foundations and Applications, Prentice Hall, Upper Saddle River, N.J., 1997.

5. Klir, G.J. and Folger, T.A., Fuzzy Sets, Uncertainty and Information, Prentice Hall, Englewood Cliffs, N.J., 1988.

6. Klir, G.J. and Yuan, B., Fuzzy Sets and Fuzzy Logic: Theory and Applications, Prentice Hall, Upper Saddle River, N.J., 1995.

7. Mordeson, J. N. and Nair, P.S., Fuzzy Mathematics: An Introduction for Engineers and Scientists, Studies in Fuzziness and Soft Computing, Physica- Verlag, Heidelberg, Germany, 1998.

8. Rosenfeld, A., Fuzzy graphs. In: L. A. Zadeh, K. S. Fu and M.

Shimura, Eds., Fuzzy Sets and Their Applications, Academic Press, New York, 77-95, 1975.

9. Tamura, S., Higuchi, S., and Tanaka, K., Pattern Classification Based on Fuzzy Relations, IEEE 1tans. SMC-1, 61-66, 1971.

10. Yeh, R. T. and Bang, S.Y., Fuzzy graphs, fuzzy relations, and their applications to cluster analysis. In: Zadeh, L. A., K. S. Fu and M. Shimura, Eds., Fuzzy Sets and Their Applications, Academic Press, New York, 125-149, 1975.

11. Zadeh, L. A., Fuzzy sets, Inform. and Control, 8:338-353, 1965.

12. Zadeh, L. A., Fuzzy sets as a basis for a theory of possibility, Fuzzy Sets and Systems, 1:3-28, 1978.

13. Zadeh, L.A., Similarity relations and fuzzy orderings, Inform. Sci., 3:177-200, 1971.

14. Zimmermann, H.J., Fuzzy Set Theory and Its Applications, Second Edition, Kluwer Academic Publishers, Boston, Dordrecht/London, 1991.

2

FUZZY GRAPHS

A gmph is a pair (V, R), where V is a set and R is a relation on V. The elements of V are thought of as vertices of the graph and the elements of R are thought of as the edges. Similarly, any fuzzy relation p on a fuzzy subset /.L of a set V can be regarded as defining a weighted graph, or fuzzy graph, where the edge (x, Y) E V x V has weight or strength p( x, Y) E [0, 11. In this chapter, we shall use graph terminology and introduce fuzzy analogs of several basic graph-theoretical concepts. For simplicity, we will consider only undirected graphs through out this chapter unless otherwise specified. Therefore, all fuzzy relations are symmetric and all edges are regarded as unordered pairs of vertices. We abuse notation by writing (x, y) for an edge in an undirected graph (V, R), where z, y E V. (We need not consider loops, that is, edges of the form (x, x); we can assume, if we wish, that our fuzzy relation is reflexive.) Formally, a fuzzy gmph G = (V, J.L, p) is a nonempty set V together with a pair of functions /.L : V --T [0, IJ and p : V x V --T [0,11 such that for all z , y in V, p( x, y) s. J.L( x) /\ /.L(y). We call J.L the fuzzy vertex set of G and p the fuzzy edge set of G, respectively. Note that p is a fuzzy relation on /.L. \Ve will assume that, unless otherwise specified, the underlying set is V and that it is finite. Therefore, for the sake of notational convenience, we omit V in the sequel and use the notation G = (/.L, p). Thus in the most general case, both vertices and edges have membership value. However, in the special case where J.L(x) = 1, for all x E V, edges alone have fuzzy membership. So, in this case, we use the abbreviated notation G = (V, p). The fuzzy graph H = (v, r) is called a partial fuzzy subgmph of G = (J.L,p) if v ~ J.L and T ~ p. Similarly, the fuzzy graph H = (P, u, 7) is called a fuzzy subgmph of G = (V, /.L, p) induced by P if P ~ V, v(x) = /.L(x) for all

20 2. FUZZY GRAPHS

x E P and T(X, y) = p(x, y) for all x, YEP. For the sake of simplicity, we sometimes call H a fuzzy subgraph of G. It is worth noticing that a fuzzy subgraph (P, v, T) of a fuzzy graph (V, IL, p) is in fact a special case of a partial fuzzy subgraph obtained as follows.

v(X)

= { {

IL(X) if x E P

o jf x E V \ P

p( X, y) if (x, y) E P x P

o if (x, y) E V x V \ P x P

T(X,y)

Hence we see that a fuzzy graph can have only one fuzzy subgraph corresponding to a given subset P of V. Thus we shall use the notation {P} to denote the fuzzy subgraph of G induced by P. For any threshold t, 0 ~ t ~ 1, ILt = {x E V IIL(x) 2: t} and pt = {(x,y) E V x V I p(x,y) ;::: t}. Since p(x,y) ~ IL(X) /\ IL(Y) for all x,y E V, we have pt ~ ILt X ILt, so that (ILt,pt) is a graph with the vertex set ILt and edge set pt for all t E [0,1].

Proposition 2.1 Let G = (IL, p) be a fuzzy gmph. If 0 ~ u ~ t ~ 1, then (ILt,pt) is a subgraph of (ILU,pU) .•

Proposition 2.2 Let H = (v, T) be a partial fuzzy subgraph of G = (IL, p). For any threshold t,O ~ t ~ 1, (vt,Tt) is a subgraph of (ILt, pt) .•

We say that the partial fuzzy subgraph (v, T) spans the fuzzy graph (IL,P) if IL = u, In this case, we call (V,T) a spanning fuzzy subgraph of (IL, p). For any fuzzy subset v of V such that v ~ u, the partial fuzzy subgraph of (IL, p) induced by v is the maximal partial fuzzy subgraph of (IL,P) that has fuzzy vertex set u, This is the partial fuzzy graph (V,T), where T(X, y) = v(x) /\ v(y) /\ p(x, y) for all x, y E V.

2.1 Paths and Connectedness

Let G = (V, X) be a graph. A path of G is an alternating sequence of vertices and edges VO,et,VI, ... ,Vn-t,en,vn, where VO,Vi E V, ei EX, ei = (Vi-I, vd, i = 1, ... , n and all the vertices are distinct except Vo may be the same as Vn. A path is sometimes denoted by VOVI".Vn, where the edges are evident by context. Let VOVI",Vn be a path. If n 2: 3 and Vo = Vn, then the path is called a cycle. G = (V, X) is said to be complete if (u, v) E X 'rIu, v E V, u i= v. A clique of a graph is a maximal complete subgraph.

A path P in a fuzzy graph (IL, p) is a sequence of distinct vertices Xo, Xl , ... , Xn (except possibly Xo and z.,) such that p(Xi-I,Xi) > 0, 1 ~ i ~ n. Here n 2: 1 is called the length of the path P. The consecutive pairs (Xi-I, Xi) are


called the edges of the path. The diameter of x, y E V, written diam(x, y), is the length of the longest path joining x to y. It is shown in [4] that if diam(x,y) = 1, then poo(x,y) = p(x,y). It is also shown in [4] that if diam(x, y) = k, then poo(x, y) = V{/(x, y) I i = 1,2, ... , k}. In fact an algorithm to compute poo is given in [4]. The strength of P is defined as /\i'=lP(Xi-ll Xi). In other words, the strength of a path is defined to be the weight of the weakest edge of the path. A single vertex x may also be considered as a path. In this case, the path is of length O. If the path has length 0, it is convenient to define its strength to be fL(xo), It may be noted that any path of length n > 0 can as well be defined as a sequence of edges (Xi-I, Xi), 1 ::; i ::; n, satisfying the condition P(Xi-ll Xi) > 0 for 1 ::; i ::; n. A partial fuzzy subgraph (fL, p) is said to be connected if 'Vx, y E SUPP(fL), poo(x, y) > O.

We call P a cycle if Xo = Xn and n ~ 3. Two vertices that are joined by a path are said to be connected. It is evident that "connected" is an equivalence relation. In fact, X and y are connected if and only if poo (x, y) > O. The equivalence classes of vertices under this relation are called connected components of the given fuzzy graph. They are just its maximal connected partial fuzzy subgraphs. A strongest path joining any two vertices x, y has strength poo(x, y). We shall sometimes refer to this as the strength of connectedness between the vertices.

Proposition 2.3 If (1/, T) is a partial fuzzy subgraph of (fL, p), then TOO ~ poo .•

Bridges and Cut Vertices

Let G = (fL, p) be a fuzzy graph, let x, y be two distinct vertices, and let G' be the partial fuzzy subgraph of G obtained by deleting the edge (x, y). That is, G' = (fL, pi), where pi (x, y) = 0 and p' = p for all other pairs. We say that (x, y) is a bridge in G if p/oo(u, v) < poo(u, v) for some u, v. In other words, if deleting the edge (x, y) reduces the strength of connectedness between some pair of vertices. Thus, (x, y) is a bridge if and only if there exist vertices u, v such that (z, y) is an edge of every strongest path from u to v.

Theorem 2.4 The following statements are equivalent:

(1) (x, y) is a bridge;

(2) p/oo(x, y) < p(x, y);

(3) (x, y) is not the weakest edge of any cycle.

22 2. FUZZY GRAPHS

Proof. (2) ::} (1) If (x, y) is not a bridge, then p'OO(x, y) = pOO(x, y) > p(x, y).

(1) ::} (3) If (x, y) is a weakest edge of a cycle, then any path involving edge (x,y) can be converted into a path not involving (x,y) but at least as strong, by using the rest of the cycle as a path from x to y. Thus (x, y) cannot be a bridge.

(3) ::} (2) If p'OO(x, y) 2: p(x, y), there is a path from x to y, not involving (x, y), that has strength 2: p(x, y), and this path together with (x, y) forms a cycle of which (x, y) is a weakest edge .•

Let w be any vertex and let G' be the partial fuzzy subgraph of G obtained by deleting the vertex w. That is, G' = (IL', p') is the partial fuzzy subgraph of G such that IL' (w) = 0, IL' = IL for all other vertices, p'(w, z) = 0 for all z, and p' = p for all other edges.

We say that w is a cutvertex in G if (p,)OO (u, v) < pOO(u, v) for some u, v such that u #- w #- v. In other words, if deleting the vertex w reduces the strength of connectedness between some other pair of vertices. Hence, w is a cutvertex if and only if there exist u, v, distinct from w such that w is on every strongest path from u to v. G' is called nonseparable (or sometimes: a block) if it has no cut vertices. It should be pointed out that a block may have bridges. However this cannot happen for non-fuzzy graphs. For example, consider the fuzzy graph G = (V,p), where V = {x,y,z} and p(x,y) = 1, p(x,z) = p(y,z) = 0.5. Note that the edge (x,y) is a bridge since its deletion reduces the strength of connectedness between x and y from 1 to 0.5. However, it is easily verified that no vertex of this fuzzy graph is a cutvertex.

If between every two vertices z, y of G there exist two strongest paths that are disjoint (except for x, y themselves), G is a block. This is analogous to the "if" of the non-fuzzy graph theorem that G is a block (with at least three vertices) if and only if every two vertices of G lie on a common cycle. The "only if", on the other hand, does not hold in the fuzzy case, as the example shows.

Forests and Trees

A (crisp) graph that has no cycles is called acyclic or a forest. A connected forest is called a tree. We call a fuzzy graph a forest if the graph consisting of its nonzero edges is a forest, and a tree if this graph is also connected. More generally, we call the fuzzy graph G = (j.L, p) a fuzzy forest if it has a partial fuzzy spanning subgraph F = (j.L, r) which is a forest, where for all edges (x,y) not in F (i.e., such that r(x,y) = 0), we have p(x,y) < rOO(x,y). In other words, if (x, y) is in G but (x, y) is not in F, there is a path in F between x and y whose strength is greater than p(x, y). It is clear that a forest is a fuzzy forest.


The fuzzy graphs in Figure 2.1 are fuzzy forests and the fuzzy graphs in Figure 2.2 are not fuzzy forests.

FIGURE 2.1 Fuzzy forests.

x 1.0 Y

V

',~'

:~25

y

o([;:?

x 1]

OJ 1]

z

FIGURE 2.2 Fuzzy graphs; but not fuzzy forests.

x 1.0

a.! D.! w 1.0

1.0 Y

,.aD·!

W 0.5 Z

Z

If G is connected, then so is F since any edge of a path in G is either in F, or can be diverted through F. In this case, we call G a fuzzy tree. The examples of fuzzy forests given above are all fuzzy trees. Note that if we replaced < by ~ in the definition, then even the fuzzy graph (V, /l, p), where V = {x, y, z}, /lex) = /ley) = p,(z) = 1, p(x, y) = p(x, z) = p(y, z) = 1, would be a fuzzy forest since it has partial fuzzy spanning subgraphs such as (V, p" p'), where p'(X, y) = p'(X, z) = 1 and p(y, z) = O.

Theorem 2.5 G is a fuzzy forest if and only if in any cycle of G, there is an edge (x, y) such that p(x, y) <p''''' (x, y), where G' = (/l, p') is the partial fuzzy subgraph obtained by the deletion of the edge (x,y) from G.

Proof. Suppose (x, y) is an edge, belonging to a cycle, which has the property of the theorem and for which p(x, y) is smallest. (If there are no cycles, G is a forest and we are done.) If we delete (x, y), the resulting partial fuzzy subgraph satisfies the path property of a fuzzy forest. If there are still cycles in this graph, we can repeat the process. Now at each stage, no previously deleted edge is stronger than the edge being currently deleted. Thus the path guaranteed by the property of the theorem involves only edges that have not yet been deleted. "When no cycles remain, the resulting

24 2. FUZZY GRAPHS

partial fuzzy subgraph is a forest F. Let (x, y) not be an edge of F. Then (x, y) is one of the edges that we deleted in the process of constructing F, and there is a path from x to y that is stronger than p(x, y) and that does not involve (x, y) nor any of the edges deleted prior to it. If this path involves edges that were deleted later, it can be diverted around them using a path of still stronger edges; if any of these were deleted later, the path can be further diverted; and so on. This process eventually stabilizes with a path consisting entirely of edges of F. Thus G is a fuzzy forest.

Conversely, if G is a fuzzy forest and P is any cycle, then some edge (x, y) of P is not in F. Thus by definition of a fuzzy forest we have p(x, y) < TOO(X, y) ~p'OO(x, y) .•

We see that if G is connected, then so is F determined by the construction in the first part of the proof.

Proposition 2.6 If there is at most one strongest path between any two vertices of G, then G must be a juZ7:]J forest.

Proof. Suppose G is not a fuzzy forest. Then by Theorem 2.5, there is a cycle P in G such that p(x, y) ~ p'(x, y) for all edges (x, y) of P. Thus (z, y) is a strongest path from x to y. If we choose (x, y) to be a weakest edge of P, it follows that the rest of the P is also a strongest path from x to y, a contradiction .•

We now show that the converse of Proposition 2.6 does not hold, that is, G can be a fuzzy forest and still have multiple strongest paths between vertices. For example, the fuzzy graph in Figure 2.3 is a fuzzy forest. Here F consists of all edges except (u, y). The strongest paths between x and y have strength 1/4, due to the edge (z, u); both x, u, v, y and x, u, y are such paths, where the former lies in F but the latter does not.

Proposition 2.7 If G is a juZ7:]J forest, then the edges of F are just the bridges of G.

Proof. An edge (x, y) not in F cannot be a bridge since p( x, y) < TOO ( x, y) ~ p'OO(x, y). Suppose that (x, y) is an edge in F. If it were not a bridge, we would have a path P from x to y, not involving (x, y), of strength ~ p( x, y). This path must involve edges not in F since F is a forest and has no cycles. However, by definition, any such edge (Ui' Vi) can be replaced by a path Pi in F of strength > p(u,v). Now Pi cannot involve (x,y) since all its edges are strictly stronger than p(u, v) ~ p(x, y). Thus by replacing each (Ui' Vi) by Pi, we can construct a path in F from x to y that does not involve (x, y), giving us a cycle in F, contradiction .•


FIGURE 2.3 A fuzzy forest with no multiple strongest paths between vertices.

0.25

Trees and Cycles

Recall that supp(JL) = {x E V I JL(x) > O} and supp(p) = {(x,y) E V x V I p(x,y) > O}. Since p(x,y) ::; JL(x) 1\ JL(Y), (x,y) E supp(p) implies x,y E supp(JL). Thus (supp(JL), supp(p)) is a graph.

\Ve now recall some definitions and give some new ones.

Definition 2.1

(1) (JL, p) is a tree if and only if (supp(JL), supp(p)) is a tree.

(2) (JL, p) is a fuzzy tree if and only if(JL, p) has a fuzzy spanning subgraph (JL, v) which is a tree such that'll (u, v) E supp(p)\ supp(v), p(u, v) < VOO(u, v), i. e., there exists a path in (JL, v) between u and v whose strength is greater than p( u, v).

Definition 2.2

(1) (JL, p) is a cycle if and only if (supp(JL), supp(p)) is a cycle.

(2) (JL, p) is a fuzzy cycle if and only if (supp(JL)' supp(p)) is a cycle and ~ unique (x,y) E supp(p) such that p(x,y) = A{ p(u,v) I (u,v) E supp(p)}.

Example 2.1 Let V = {u, v, w, S, t} and X = {(u, v), (u, w), (v, w), (w,s), (w,t),(s,t)}. Let JL(x) = 1 for all x E V and let p be the fuzzy subset of X defined by p(u, v) = 1/2, p(u, w) = p(v, w) = pew, s) = pew, t) = pes, t) = 1. Then (JL, p) is neither a fuzzy cycle nor a fuzzy tree.

Example 2.2 Let V = {w, u, v} and X = {(w, u), (w, v), (u, v)}. Let J-l(x) = 1 for all x E V and p and p' be fuzzy subsets of X defined by pew, u) = pew, v) = 1, p(u, v) = 1/2 and p' (w, v) = 1, p'(W, u) = p'(U, v) = 1/2. Then (JL, p) is a fuzzy tree, but not a tree and not a fuzzy cycle while (J-l, p') is a fuzzy cycle, but not a fuzzy tree.

26 2. FUZZY GRAPHS

Example 2.2 illustrates the next result.

Theorem 2.8 Let (J.L, p) be a cycle. Then (J.L, p) is a fuzzy cycle if and only if (J.L, p) is not a fuzzy tree.

Proof. Suppose that (J.L, p) is a fuzzy cycle. Then :3 distinct edges (Xl, YI), (X2,Y2) E supp(p) such that P(XllYl) = p(X2,Y2) = A{p(u,v) I (u,v) E supp(p)}. If (J.L, v) is any spanning tree of (J.L, p), then supp(p)\ supp(v) = { (u, v)} for some u, v E V since (J.L, p) is a cycle. Hence ~ path in (J.L, v) between u and v of greater strength that p(u, v). Thus (J.L, p) is not a fuzzy tree. Conversely, suppose that (J.L, p) is not a fuzzy tree. Since (J.L, p) is a cycle, we have V (u, v) E supp(p) that (J.L, v) is a fuzzy spanning subgraph of (J.L. p) which is a tree and v<Xl(u, v) ::; p(u, v) where v(u, v) = 0 and v(x,y) = p(x,y) V (x,y) E supp(p) \ {(u,v)}. Hence p does not attain A{ p(x,y) I (x,y) E supp(p)} uniquely. Thus (J.L, p) is a fuzzy cycle .•

Theorem 2.9 If:3 q E (0, I] such that (supp(J.L), pq) is a tree, pq a q - cut, then (J.L, p) is as fuzzy tree. Conversely, if (J.L, p) is a cycle and (J.L, p) is a fuzzy tree, then:3 q E (0, I] such that (supp(J.L), pq) is a tree.

Proof. Suppose that q exists. Let u be the fuzzy subset of V x V such that v = p on pq and vex, y) = 0 if (x, y) E V x V\pq. Then (J.L, v) is a spanning fuzzy subgraph of (J.L, p) such that (J.L, v) is a fuzzy tree since (suPP(J.L), supp(v)) is a tree. Suppose that (u,v) E V x V and (u,v) ¢:. pq. Then B a path between u and v of strength ~ q > p(u, v). Thus (J.L, p) is a fuzzy tree. For the converse, we note that since (J.L, p) is a cycle and a fuzzy tree, :3 unique (x,y) E supp(p) such that p(x,y) = A{p(u,v) I (u,v) E supp(p)}. Let q be such that p(x,y) < s < A{p(u,v) I (u,v) E supp(p) \ {(x,y)}}. Then (suPP(J.L), pq) is a tree .•

We now illustrate Theorem 2.9.

Example 2.3 Let V = {s,t,u,v,w} and X = {(s,t),(s,u),(t,u),(u,v), (u, w), (w, v)}. Let J.L(x) = 1 for all x E V and let p be the fuzzy subset of X defined by pes, t) = 1/4, pes, u) = pet, u) = 3/8, p(u, v) = 1/2, and p(u, w) = pew, v) = 1. Then ~ q E (0, I] such that (supp(J.L), pq) is a tree. However (J.L, p) is a fuzzy tree.

A Characterization of Fuzzy Trees

The results here are taken from [41]. Let G = (V, J.L, p) be a fuzzy graph. Recall that p<Xl(u, v) denotes the strength of connectedness between u and v in V.


Let C = (J.L, p) be a fuzzy graph such that C* = (SUPP(fL), supp(p)) is a cycle and let t = A {pC u, v) I p( u, v) > O}. Then all edges (u, v) such that p(u,v) > t are bridges of C.

Theorem 2.10 Let C = (fL, p) be a fuzzy graph such that C· = (SUpp(fL), supp(p)) is a cycle. Then a vertex is a cuiuertez of C if and only if it is a common vertex of two bridges.

Proof. Let w be a cutvertex of C. Then there exist u and v, other than w, such that w is on every strongest u-v path. Since C* = (supptu), supp(p)) is a cycle, there exists only one strongest u-v path containing w and all its edges are bridges. Thus w, is a common vertex of two bridges. Conversely, let w be a common vertex of two bridges (u, w) and (w, v). Then both (u, w) and (w, v) are not the weakest edges of C. Also, the path from U to v not containing the edges (u, w) and (w, v) has strength less than p(u, w) A pew, v). Hence the strongest u-v path is the path u, w, v and pOO(u, v) = p(u, w) A pew, v). Thus w is a cutvertex .•

Theorem 2.11 If w is a common vertex of at least two bridges, then w is a cuiueriex.

Proof. Let (UI' w) and (w, U2) be two bridges. Then there exist some u,v such that (UI' w) is on every strongest u-v path. If w is distinct from U and v it follows that w is a cutvertex. Next, suppose one of v, u is w so that (UI, w) is on every strongest u-w path or (w, U2) is on every strongest w-v path. Suppose that w is not a cutvertex. Then between every two vertices there exists at least one strongest path not containing w. In particular, there exists at least one strongest path P, joining UI and U2, not containing w. This path together with (UI'W) and (W,U2) forms a cycle.

\Ve now consider two cases. First suppose that UI, W, U2 is not a strongest path. Then clearly one of (UI, w), (w, U2) or both become the weakest edges of the cycle which contradicts that (u}, w) and (w, U2) are bridges.

Second suppose that UI, W, U2 is also a strongest path joining UI to U2.

Then pOO(UI, U2) = P(UI'W) A p(W,U2), the strength of P. Thus edges of P are at least as strong as p(Ul,w) and P(W,U2) which implies that (U I, w), (w, U2) or both are the weakest edges of the cycle, which again is a contradiction .•

That the condition in Theorem 2.11 is not necessary can be observed from the following example. Let V = {u,v,w,x} and X = {(u,v),(u,w),(u,x), (v, w), (v, x), (w, x)}. Let J.L(s) = 1 for all S E V and let p be the fuzzy subset of X defined by p(u, w) = p(v, x) = 1, p(v, w) = pew, x) = 0.7, and p(u, v) = p(u, x) = 0.5. Clearly, w is a cutvertex; (u, w) and (v, x) are the only bridges.

28 2. FUZZY GRAPHS

Consider the fuzzy graph G = (V,/L,p) where V = {u,v,w,x}. Let X = {(u,v),(v,w),(w,x),(x,u)}. Let /L(s) = 1 for all S E V and let p be the fuzzy subset of X defined by p(u, v) = pew, x) = 0.5, and p(v, w) = p(x, u) = 0.4. Note that (u, v) and (w, x) are the bridges and no vertex is a cutvertex. This is a significant difference from the crisp graph theory.

Theorem 2.12 If (u, v) is a bridge, then pOO(u, v) = p(u, v).

Proof. Suppose that (u, v) is bridge and that poo (u, v) > p( u, v). Then there exists a strongest u-v path with strength greater than p( u, v) and all edges of this strongest path have strength greater than p( u, v) and all edges of this strongest path have strength greater than p(u, v). Now, this path together with the edge (u, v) forms a cycle in which (u, v) is the weakest edge, contradicting that (u, v) is a bridge .•

The converse of Theorem 2.12 is not true. The condition for the converse to be true is discussed in Theorem 2.20.

Recall that a connected fuzzy graph G = (/L, p) is a fuzzy tree if it has a fuzzy spanning subgraph F = (/L, T), which is a tree, where for all edges (u,v) not in F, p(u,v) < TOO(U,V).

Equivalently, there is a path in F between u and v whose strength exceeds p(u, v).

Lemma 2.13 If (v, T) is a partial fuzzy subgraph of (/L, p), then for all u, v, TOO(U, v) S pOO(u, v) .•

Theorem 2.14 If G = (/L, p) is a fuzzy tree and G* = (suPP(/L), supp(p)) is not a tree, then there exists at least one edge (u, v) in supp(p) for which p(u, v) < pOO(u, v).

Proof. IT G is a fuzzy tree, then by definition there exists a fuzzy spanning subgraph F = (/L, T), which is a tree and p(u, v) < TOO(U, v) for all edges (u, v) not in F. Also TOO(U, v) S pOO(u, v) by Lemma 2.13. Thus p(u, v) < pOO(u, v) for all (u, v) not in F and by hypothesis there exist at least on edge (u, v) not in F .•

Definition 2.3 A complete fuzzy graph is a fuzzy graph G = (/L,p) such that p(u, v) = /L(u) /\ /L(v) for all u and v.

Lemma 2.15 If G is a complete fuzzy graph, then pOO(u, v) = p(u, v) .•


Lemma 2.16 A complete fuzzy graph has no cutuertices .•

That the converse of Lemma 2.15 is not true can be observed from the following example.

Example 2.4 Let V = {u,v,w,x} and X = {(u,v),(u,w),(u,x),(v,w), (v,x),(w,x)}. Let p(u) = 0.8, and p(v) = pew) = p(x) = 1. Let p be the fuzzy subset of X defined by p(v, w) = 1, p(v, x) = pew, x) = 0.8, and p(u, v) = p(u, w) = p(u, x) = 0.6.

FUrther, a complete fuzzy graph may have a bridge as illustrated by the next example. Let V = {u,v,w,x} and X = {(u,v),(u,w),(u,x),(v,w),

(v:x): (w,x)}. Let p(u) = 0.6: p(x) = 0.8, and p(v) = pew) = 1. Let p be the fuzzy subset of X defined by p(v, w) = 1, p(v, x) = pew, x) = 0.8, and p(u,1I) = p(u, w) = p(u, x) = 0.6.

Theorem 2.17 If G = (p, p) is a fuzzy tree, then G is not complete.

Proof. If possible, let G be a complete fuzzy graph. Then p=(u, v) = p(u, v) for all u, v by Lemma 2.15. Now G being a fuzzy tree, p(u, v) < vcx.:(u, v) for all (u, v) not in F. Thus p=(u, v) < v=(u: v), contradicting Lemma 2.13 .•

Recall that if G is a fuzzy tree, then the edges of F are the bridges of G.

Theorem 2.18 If G is a fuzzy tree, then the internal vertices of F are the cutvertices of G.

Proof. Let w be any vertex in G which is not an end vertex of F. Then w is the common vertex of at least two edges in F which are bridges of G and by Theorem 2.11, w is a cutvertex. Also, if w is an end vertex of F, then w is not a cutvertex; else there would exist u, v distinct from w such that w is on every strongest u-v path and one such path certainly lies in F. But since w is an end vertex of F, this is not possible .•

Corollary 2.19 A cutueriez of a fuzzy tree is the common vertex of at least two bridges .•

Theorem 2.20 G = (p, p) is a fuzzy tree if and only if the following are equivalent for all u, v :

(1) (u, v) is a bridqe.

30 2. FUZZY GRAPHS

(2) pOO{u, v) = p{u, v).

Proof. Let G = (Il, p) be a fuzzy tree and suppose that (u, v) is a bridge. Then pOO{u, v) = p{u, v) by Theorem 2.12. Now, let (u, v) be an edge in G such that pOO{u, v) = p{u, v). If G* is a tree, then clearly (u, v) is a bridge; otherwise, it follows from Theorem 2.14 that (u, v) is in F and (u, v) is a bridge.

Conversely, assume that (I) and (2) are equivalent. Construct a maximum spanning tree T : (Il, r) for G [4]. If (u, v) is in T, by an algorithm in [4], pOO{u, v) = p{u, v) and hence (u, v) is a bridge. Now, these are the only bridges of G; for, if possible let (u', v') be a bridge of G which is not in T. Consider a cycle C consisting of (u', v') and the unique u' -v' path in T. Now edges of this u'-v' path are bridges and so they are not weakest edges of C and thus (u', v') must be the weakest edge of C and thus cannot be a bridge.

Moreover, for all edges (u', v') not in T, we have p{u', v') < rOO{u', v'); for, if possible let p(u', v') ~ rOO{u', v'). But p{u', v') < pOO{u', v'), where strict inequality holds since (u', v') is not a bridge. Hence, rOO (u', v') < pOO{u', v') which gives a contradiction since rOO{u', v') is the strength of the unique u'-v' path in T and by an algorithm in [4], pOO{u',v') = rOO{u',v'). Thus T is the required spanning subgraph F, which is a tree and hence G is a fuzzy tree .•

For a fuzzy tree G, the spanning fuzzy subgraph F is unique. It follows from the proof of Theorem 2.20 that F is nothing but the maximum fuzzy spanning tree T of G.

Theorem 2.21 A fuzzy graph is a fuzzy tree if and only if it has a unique maximum fuzzy spanning tree .•

For a fuzzy graph which is not a fuzzy tree there is at least one edge in T which is not a bridge and edges not in T are not bridges of G. This observation leads to the following theorem.

Theorem 2.22 If G = (1l,P) is a fuzzy graph with SUPP{Il) IVI = p then G has at most p - 1 bridges .•

II and

Theorem 2.23 Let G = (Il, p) be a fuzzy graph and let T be a maximum fuzzy spanning tree of G. Then end vertices of T are not cutvertices of G .•

Corollary 2.24 Every fuzzy graph has at least two vertices which are not cutvertices .•


FIGURE 2.4 Graph of example 2.5.

(Fuzzy) Cut Sets

The results in the remainder of the section are from [301. Before proceeding further, we review some concepts from graph theory. \Ve associate with a graph G, two vector spaces over the field of scalars Z2 = {O, I}, where addition and multiplication is modulo 2. Then for I E Z2, I + I = O. Let V = {VI, ... , vn} denote the set of vertices of G and X = {ell ... , em} , the set of edges. A O-chain of G is a formal linear combination E fiVi of vertices and a l-chain a formal linear combination of edges E fiei, where the e, E Z2. The boundary operator a is a linear function which maps 1- chains to O-chains such that if e = (x, y), then aCe) = x+y. The coboundary operator b is a linear function which maps O-chains to l-chains such that 6(v) = E fiei whenever e, E Z2 is incident with v.

Example 2.5 LetG= (V,X), where V = {VI, ... ,V6} and X = {el, ... ,e9} and where el = (vr, V2), e2 = (vJ, V3), e3 = (V2, V3), e4 = (V2, V4), es = (V2,VS), e6 = (V3,VS), e7 = (V3,V6), es = (V4,VS), and e9 = (VS,V6)' The l-chaui II = el +e2 +e4 + e9 has boundary a(,l) = (VI + V2) + (VI + V3) + (V2+V4)+(VS+V6) = V3+V4+V5+V6. The O-chain,o = V3+X4+vS+V6 has coboundary b( 10) = (e2+e3 +es +e7) + (e4 +es) +( es + e6 +es +e9) +( e7+e9) = e2 + e3 + e4 + es·

A l-chain with boundary 0 is called a cycle vector of G which we can think of as a set of line disjoint cycles. The collection of all cycle vectors, called the cycle space, form a vector space over Z2. A cut set of a connected graph is a collection of edges whose removal results in a disconnected graph. A cocycle is a minimal cutset. A co boundary of G is the coboundary of some O-chain in G. The coboundary of a subset of V is the set of all edges joining a point in this subset to a point not in the subset. Hence every coboundary

32 2. FUZZY GRAPHS

is a cutset. Since any minimal cutset is a coboundary, a co cycle is just a minimal nonzero coboundary. The collection of all coboundaries of G is a vector space over ~ and is called the cocycle space of G. A basis of this spaces which consists entirely of co cycles is called a cocycle basis for G.

Let G be a connected graph. Then a chord of a spanning tree T is an edge of G which is not in T. The subgraph of G consisting of T and any chord of T has only one cycle. The set C(T) of cycles obtained in this way is independent. Every cycle C depends on the set C(T) since C is the symmetric difference of the cycles determined by the chords of T which lie in C. \Ve define meG), the cycle rank, to be the number of cycles in a basis for the cycle space of G. This discussion yields the following result.

Theorem 2.25 The cycle rank of a connected graph G is equal to the number of chords of any spanning tree in G .•

Similar results can be derived for the co cycle space. Again assume that G is a connected graph. The cotree T' of a spanning tree T of G is the spanning subgraph of G containing exactly those edges which are not in T. A cotree of G is the cotree of some spanning tree T. The edges of G which are not in T' are called its twigs. The subgraph of G consisting of T' and anyone of its twigs contains exactly one co cycle. The collection of cocycles obtained by adding twigs to T', on at a time, is a basis for the cocycle space of G. The cocycle rank m'(G) is the number of co cycles in a basis for the cocycle space of G. A more detailed account can be found in

[20].

Theorem 2.26 The cocycle rank of a connected graph G is the number of twigs in any spanning tree of T .•

Let x E V and let t E [0,1]. Define the fuzzy subset Xt of V byVy E V,Xt(Y) = 0 if y =I x and Xt(Y) = t if y = x. Then Xt is called a fuzzy singleton in V. If (x, y) E V x V, then (x, y)p(x,y) denotes a fuzzy singleton inVxV.

Definition 2.4 Let E be a subset of supp(p).

(1) {(x.Y)p(x,y) I (x,y) E E} is a cut set of (11, p) if E is a cut set of (supp(Il), supp(p».

(2) {(x,Y)p(x,y) I (x,y) E E} is a fuzzy cut set of(ll, p) ifj u, v E supp(ll) such that p'OO(u, v) < pOO(u, v) where p' is the fuzzy subset of V x V defined by p' = p on supp(p) \ E and p'(x, y) = 0 Vex, y) E E, i. e., the removal of E from supp(p) reduces the strength of connectedness between some pair of vertices of (J.L, p).


When E is a singleton set, a cut set is called a bridge and a fuzzy cut set is called a fuzzy bridge.

The following is an example of a fuzzy graph (J.L,p) which has no fuzzy bridges and where p is not a constant function.

Example 2.6 Let V = {t,u,v,w} and X = {(t,u),(u,v),(v,w),(w,t), (t, v)}. Let J.L(x) = 1 for all x E V and p(t, u) = p(u, v) = p(v, w) = p(w, t) = 1 and p(t, v) = 1/2. Then p is not constant, but (J.L, p) does not have a fuzzy bridge since the strength of connectedness between any pair of vertices of (J.L, p) remains 1 even after the removal of an edge.

Theorem 2.27 Let V = {v}, ... ,vn} and C = {(V},V2),(V2,V3), ... , (vn-l,Vn),(vn,vd}, n 2: 3.

(1) Suppose that supp(p) 2 C and that V (Vj, Vk) E supp(p) \ C, p(Vj, Vk) < V{p(Vi,Vi+d Ii = l, ... ,n} where Vn+I = VI' Then either p is a constant on C or (J.L, p) has a fuzzy bridge.

(2) Suppose that 0 :f:.supp(p) C C. Then (J.L, p) has a bridge.

Proof. (1) Suppose that p is not constant on C. Let (Vh, vh+d E C be such that P(Vh, Vh+l) = V{P(Vi' vi+d Ii = 1, ... , n}. Since p is not constant on C, the strength of the path C \ {(vh,vh+d} between Vh and Vh+1 is strictly less than p(Vh, Vh+1)' The strength of any other path P between Vh and Vh+l is also strictly less than p(Vh,Vh+1) since P must contain an edge from supp(p) \ C. Thus (Vh, Vh+1)P(Uh,Vh+tl is a fuzzy bridge.

(2) Immediate .•

Theorem 2.28 Suppose that the dimension of the cycle space of (supp(J.L), supp(p)) is 1. Then (J.L, p) does not have a fuzzy bridge if and only if (J.L, p) is a cycle and p is a constant function.

Proof. Suppose it is not the case that (J.L, p) is a cycle and p is a constant function. If (J.L, p) is not a cycle, then :3 edge (x, y) E supp(p) which is not part of the cycle. Then (x,Y)p(x,y) is a bridge and hence a fuzzy bridge. Suppose that (J.L, p) is a cycle, but p is not a constant function. Let (x,y) E supp(p) be such that p(x, y) is maximal. Then (x, y)p(x,y) is a fuzzy bridge. Conversely, suppose that (J.L, p) is a cycle and p is a constant function. Then the deletion of an edge (Vi, Vi+d yields a unique path between Vi and Vi+l of strength equal to p(Vi, Vi+I)' Thus (Vi, Vi+l)p(v.,v.+tl is not a fuzzy bridge .•

(Fuzzy) Chords, (Fuzzy) Co trees, and (Fuzzy) Twigs

We assume throughout this section that (supp(J.L), supp(p)) is connected.

34 2. FUZZY GRAPHS

Definition 2.5 Let (f..L, v) be a spanning fuzzy subgraph of (f..L, p) which is a tree. If (f..L, v f) is a fuzzy tree such that v ~ u I ~ p, v f = p on supp(v I), and ~ a fuzzy tree (f..L, v') such that vI C i/ ~ p and v' = p on supp(v'), then (f..L, v f) is called a fuzzy spanning tree of (f..L, p) with respect to u.

Clearly given (f..L, v) and (f..L, p) of Definition 2.5, (f..L, v I) exists. We note that (f..L, v) is a spanning fuzzy subgraph of (f..L, vf).

In Example 2.6, let u, vI, and u]' be the fuzzy subsets of X defined as follows: v = p on {(t,u), (t,v), (t,w)} and v(w, v) = v(u, v) = O,vI = P on {(t, u), (t, v), (t, w), (u, v)} and vJ(w, v) = 0, vJ = p on {(t, u), (t, v), (t, w), (w, v)} and vI (u, v) = O. Then both (f..L, v I) and (f..L, vI) are fuzzy spanning trees of (f..L,p) with respect to t/, That is, given t/, vI in Definition 2.5 is not necessarily unique.

Consider the fuzzy graph (f..L, p) of Example 2.2 . Define the fuzzy subset v of X by v( w, u) = v( u, v) = 1. Since (f..L, p) is not a fuzzy cycle, the addition of (u, v h /2 to (f..L, v) does not create a fuzzy cycle. This fact motivates the definition of a fuzzy chord below.

Definition 2.6 Let (f..L, v) be a fuzzy spanning subgraph of (f..L, p) which is a tree. Let (x, y) E supp(p).

(1) (x,Y)p(x,y) is a chord of (f..L, v) if and only if (x,y) ~ supp(v), i. e., (x, y) is a chord of (supp(f..L), supp(p)).

(2) (x, Y)p(x,y) is a fuzzy chord of tu, vf) if and only if(x, y) ~ (x, Y)p(x,y).

Example 2.7 Let V = {s,t,u,v,w} and X = {(w,s),(w,t),(w,u),(w,v), (s, t), (u, v)}. Define the fuzzy subsets f..L of V and p, u of X by f..L(x) = 1 for all x E V and p(x, y) = 1 for all (x, y) in X \ {(w, u)} and p(w, u) = 1/2, v(x,y) = 1 for all (x,y) in Y, where Y = X \ {(w,s), (w,u)}. Then vJ = p onX\{(w,s)} and vf(w,S) = O. Also (w,sh and (W,Uh/2 are chords of (f..L, v) and (w, sh is a fuzzy chord of (f..L, vI). If we define the fuzzy subset u' of X by u' = VI on sUPP(vf) and V'(W, s) = t where 0 < t < 1, then (f..L, v') is a fuzzy tree such that vJ C v', However, v' =I p on supp(v').

Definition 2.7 Let (f..L, v) be a spanning fuzzy subgraph of (f..L, p) which is a tree.

(1) Let v' be a fuzzy subset of V x V. Then (f..L, v') is the cotree of (f..L, v) if and only if\;/ (x,y) E supp(p), v'(x,y) = 0 if v(x,y) > 0 and v'(x,y) = p(x,y) ifv(x,y) = o.

(2) Let v I' be a fuzzy subset of V x V. Then (f..L, v 1') is the fuzzy cotree of(f..L, VI) if and only if\;/ (x,y) E supp(p), v/(x,y) = 0 ifvf(x,y) > 0 and v/(x,y) = p(x,y) ifvJ(x,y) = O.


Let (Il, v') be a cotree of (Il, v), where (Il, v) is a spanning fuzzy subgraph of (Il, p) which is a tree. Then (suppfjr), supp(v')) is a cotree of (suppfu), suppu-) since supp(v') n supp(v) = 0, supp(v') U supp(v) = supp(p), and (suppfrz), supp(v)) is a tree.

Definition 2.8 Let (Il, v) be a fuzzy spanning subgraph of (Il, p) which is a tree and let (x, y) E supp(p).

(1) Let (Il, v') be a coiree of(ll, v). Then (x,Y)p(x,y) is a twig of(ll, v') if and only if v'(x, y) = o.

(2) Let (Il, vI') be a fuzzy cotree of(ll, vI)' Then (x,Y)p(x,y) is a fuzzy twig of (Il, v/) if and only if v/(x,y) = o.

Example 2.8 Let (p, p), (p, v), and (Il, vI) be the fuzzy subgraphs of Example 2.7. Let (p, v') be the cotree of (Il, v) and (p, vI') be the fuzzy cotree of(ll, vI)' Then the twigs of(ll, v') are (s,th, (w,th, (w,vh, and (u,vh. The fuzzy twigs of(p, v/) are (s,th, (w,th, (w,vh, (u,vh, and (W,Uh/2.

(Fuzzy) i-Chain with Boundary 0, (Fuzzy) Co boundary, and (Fuzzy) Cocyles

We recall that a pair (M, *) is a semigroup if M is a nonempty set and * is an associative binary operation on M. We let G = (p, p). If X is a set of fuzzy singletons, we let foot(X) denote {x I Xt EX}.

Definition 2.9 Let (x,y) E V x V. Then (x,y) is called exceptional in G if and only if3 a cycle C ~ V x V such that (x, y) E C and (x, y) is unique with respect to p(x,y) = A{p(u,v) I (u,v) E C}. Let E = {(x,y) E V x V I (x, y) is exceptional}. Let PE be the fuzzy subset of V x V defined by PE = P on V x V \ E and P d x, y) = 0 V( x, y) E E.

Let Sp = {(x,Y)t I (x,y) E supp(p), t E (0, I]} U {O, It E (O,l]}. Let addition of elements of supp(p) be a formal addition modulo 2, That is, Vex, y), (u, v) E supp(p), we write (x, y)+(u, v) if (x, y) =f. (u, v) and (x, y) + (u,v) = 0 if (x,y) = (u,v). Then V(x,Y)t, (u,v)s ESp, (x,Y)t + (u,v)s = «x, y)+(u, V»r where r = tr.s. Also Vex, y}t ESp, (x, y}t +08 = (x, Y)r and Ot + Os = Or, where r = t A s. Clearly (Sp, +) is a commutative semigroup with identity, 01, If S is a set of fuzzy singletons of a set W, we let foot(S) ={wEWlwtES}.

Since Z2 = {O, I} is the field of integers modulo 2, 1 + 1 = O. We have that L Ei(Xi, Yi)p(x"y;J + L E/(Xi, Yi)p(x"y.) = :L(€i + d)(Xi, Yi)p(Xi,y;), Ei(Xi, Yi)p(x"y,) = (Xi, ydp(Xi,y;) if €i = 1 and Ei(Xi, Yi)p(X"Yi) = 0p(Xi'Y') if Ei

36 2. FUZZY GRAPHS

= 0, Ei, E/ E Z:!. We have that LEi(Xi,Yi)p(Xi,y;J = (LEi(Xi,Yi»m where m = ;:. {p(Xi,Yi)} .

•

Definition 2.10

(1) L Ei(Xi, Yi)P(Xi,Yi) is a l-chain with boundary 0 in (/-L, p) where (Xi, Yi) E supp(p) if and only if L Ei (Xi, Yi) is a 1-chain with boundary 0 in ( supp(/-L ), supp(p».

(2) L Ei(Xi, Yi)P(Xi,Yi) is a fuzzy l-chain with boundary 0 in (/-L, p) where (Xi, Yi) E SUpp(PE) if and only ifL Ei(Xi, Yi) is a 1-chain with boundary 0 in (supp(/-L), SUpp(pE»'

A (fuzzy) l-chain with boundary 0 in (/-L, p) is called a (fuzzy) cycle vector.

Definition 2.11

(1) L Ei(Xi, Yi)p(x"y.) is a coboundary of (/-L, p) where (Xi, Yi) E supp(p) if and only if L Ei(Xi, Yi) is a coboundary of (SUpp(/-L), supp(p».

(2) Lf'i(Xi,Yi)p(Xi,y,) is a fuzzy coboundary of (/-L, p) where (Xi,Yi) E SUPP(PE) if and only if LEi(Xi,Yi) is a coboundary of (supp(/-L), SUpp(PE»'

S' ~ Sp is called a (fuzzy) cocycle of (/-L, p) if and only if foot(S') is a co cycle of ((SUpp(/-L), SUpp(PE») (supp(/-L), supp(p».

(Fuzzy) Cycle Set and (Fuzzy) Cocycle Set Definition 2.12

(1) The set of all (fuzzy) cycle vectors of (/-L, p) is called the (fuzzy) cycle set of (/-L, p).

(2) The set of all (fuzzy) coboundaries of (/-L, p) is called the (fuzzy) co cycle set of (/-L, p).

The following examples show that the fuzzy cycle, cycle, fuzzy cocycle, and cocycle sets are not and do not necessarily generate vector spaces over Z2.

Example 2.9 Let V = {t,u,v,w} and X = {(t,u),(u,v),(v,w),(w,t), (t, v)}. Define the fuzzy subsets /-L of V and p of X as follows: /-L(x) = 1 for all X E V, p(t, u) = p(u, v) = 1, p(v, w) = p(w, t) = 1/2, and p(t, v) = 1/4. Then the cycle set is {(t, uh + (u, vh + (t, Vh/4' (v, Wh/2 + (w, th/2 +


(t, V)1/4, (t, U)l +(U, vh +(v, Wh/2+(W, th/2, 01/4, 01/2}. The fuzzy cycle set is {(t, uh + (u, vh + (v, Wh/2 + (w, th/2, 01/2}. The cocycle set is {(t, uh + (t, Vh/4 + (w, th/2, (u, vh + (t, Vh/4 + (v, Wh/2' (t, uh + (u, vh, (v, Wh/2 + (w, th/2, (w, th/2+(t, Vh/4+(U, vh, (t, uh +(t,Vh/4+(V, Wh/2' (w, th/2+ (t,uh + (u,vh + (V,Wh/2,01/4,01/2}. The fuzzy cocycle set is {(t,U)l + (w, th/2, (u, vh + (v, Wh/2' (t, uh + (u, vh, (v, Wh/2 + (w, th/2, (w, th/2 + (u, vh, (t, uh + (v, Wh/2, (w, th/2 + (t, uh + (u, v). + (v, Wh/2' 01/2}. The cycle set and cocycle set are not and do not generate vector spaces over ~ because of the presence of 01/2 and 01/4. Note also that in the cycle set, «t,uh + (u,vh + (t,Vh/4) + «V,Wh/2 + (W,th/2 + (t,Vh/4) = (t,uh + (u, vh + (v, Wh/2 + (w, th/2 + 01/4 =P (t, uh + (u, vh + (v, Wh/2 + (w, th/2. The fuzzy cycle set is a vector space over;£2 in this example. The fuzzy cocycle set is not a vector space over;£2 since «v, Wh/2 + (w, th/2) + «W,th/2 + (t,uh + (u,vh + (V,Wh/2) = (t,uh + (u,vh + 01/2 =P (t, uh + (u, vh.

Example 2.10 Let V, J1, and X be as in Example 2.8. Let X' = X u {(u,w)}. Define the fuzzy subset p' of X' by p' = p on X and p'(U,W) = 1/8. Then the fuzzy cycle set and the fuzzy cocycle set of (J1, p') coincides with the cycle set and the cocycle set of (J1, p) of Example 2.8, respectively.

Examples 2.8 and 2.9 illustrate the results which follow.

Let CS(J1,p), FCS(J1,p), CoS(J1,p), and FCoS(J1,p) denote the cycle set, the fuzzy cycle set, the co cycle set, and the fuzzy cocycle set of (J1, p), respectively. When the fuzzy graph (J1, p) is understood, we sometimes write CS, FCS, COS, and FCoS for these sets, respectively.

\Ve now show that even though CS, FCS, CoS, and FCOS are not necessarily vector spaces over ;£2, they are nearly so. In fact, it will be clear by the following results that the concepts of (fuzzy) twigs and (fuzzy) chords introduced here will have consequences similar to what their counterparts have in the crisp case.

For ease of notation, we sometimes use the notation e, f,or g for members of supp(p).

Clearly, CS, FCS, CoS, and FCoS are subsets of Sp = ret leE supp(p), t E (0, In U {Ot I t E (O,ll}. Let S be a subset of Sp. We let < S > denote the intersection of all subsemigroups of Sp which contain S. Then < S > is the smallest subsemigroup of Sp which contains S. Let S+ = {(edt, + ... + (enhn I (eik E S,i = I, ... ,n,n E N} where N denotes the set of positive integers. Then S+ is a subsemigroup of Sp.

Theorem 2.29 < CS > = (CS)+ = CSU{ea +Ob I ea E CS, Ob E (CS)+}. < CS > has Om as its identity where m = V{b I Ob E (CS)+}.

38 2. FUZZY GRAPHS

Proof. Since (CS)+ is a subsemigroup of Sp which contains CS, < CS > ~ (CS)+. However, it is clear that < CS > ;2 (CS)+ since < CS > is closed under +. Thus < CS > = (CS)+. Clearly < CS > ;2 CS U {ea + Ob I ea E CS, o, E (CS)+}. Now ioot(CS) is a vector space over Z2. Let gt, is E CS. Then 9 + f E foot(CS). Also gt + fs = (g + f)r where r = t /I. s. Now gt = (u},vdt, + ... + (u",vnk, and t, = (PI,ql)Sl + ... + (Pm,qm)sm' where (Ui,Vi),(Pj,qj) EX, i = 1, .. ,n and j = 1, ... ,m. Let I = {(Ui,Vi) I i = 1, .. ,n} n {(Pj,qj) I j = 1, ... ,m}. Suppose that I i= 0. By rearranging the summands in the representations of gt and fs if necessary, we have gt

+ is = (U1, vdt, + ... + (Ui-I, vi-dt,_l + (PI, qt}Sl + + (pj-I, qj-dSj-l +

Ob where Ob = (Ui,Vik + ... + (u",vn)tn+ (pj,qj)sJ + + (P771,qm)s"", b =

/I.{ t«, ... ,tn, Sj, ... ,sm}, and I = {(Ui,Vi), ... ,(Un,vn)} = {(Pj,qj), ... , (Pm,qm)). Now (u},vd + ... + (Ui-bVi-d + (P1,ql) + ... + (Pj-l,qj-l) E foot(CS) since foote CS) is a vector space over Z2' Also (g + f)a = (UI,Vt}t1 + ... + (Ui-l,Vi-J)ti_l+ (PI,q1)SI + ... + (pj-l,qj-I)Sj_l' where a = /I.{t), ... , ti--l, Sl, ... , Sj-d· Now (g + f)r + Ob = (g + f)a + Ob. Hence gt + is E CS U {ea + Oe I ea E CS, Ob E (CS)+}. That is, the sum of any two elements from CS is in CS U {ea + Ob I ea E CS, Ob E (CS)+}. From this it follows easily that (CS)+ ~ CSU {ea+ o, lea E CS, Ob E (CS)+}. The case where I = 0 is similar .•

Corollary 2.30 < FCS > = (FCS)+ = FCS U {ea + o, I ea E FCS, o, E (FCS)+}. < FCS > has Om as its identity where m = V{b I Ob E (FCS)+}.

Proof. FCS is the cycle set of (JL, P E)' •

In a similar manner, we obtain the next two results.

Theorem 2.31 < CoS> = (Co S)+ = CoS U {ea + Ob I ea E CoS, Ob E (CoS)+}. < CoS> has am as its identity where m = V{b I Ob E (CoS)+} .•

Corollary 2.32 < FCoS > = ( FCoS)+ = FCoS U {ea + Ob I ea E FCoS, O, E (FCoS)+}. < FCoS > has Om as its identity where m = V{b I Ob E (FCoS)+} .•

Since CS, FCS, Co S, and FCoS are nearly vector spaces over Z2, we can define the (fuzzy) cycle rank and (fuzzy) cocycle rank of a fuzzy graph in a meaningful way.

Definition 2.13 The cycle rank of (JL, p), written m(JL, p), is defined to be m(J1., p) = v{ 2:7=1 ti I (eik E CS, i = 1, ... , n, {e), ... , en} is a basis for foot(CS)}. The fuzzy cycle rank of (JL, p), written fm(J1., p), is defined to be the cycle rank of (JL, PE)' If {e), ... , en} is a basis for foot(CS) such that m(fl, p) = 2:7=1 tc. where (eik E CS, i = 1, ... , n, then {E), ... , en} is called a cycle basis of < CS > . If {e}, ... ,en} is a basis for ioot(FCS(JL,p))


FIGURE 2.5 Non-isomorphic fuzzy graphs with the same cycle rank.

a b

LSJ

d c

a b

Z

d c

such that fm(fl,p) = L~lti' where (ei}t. E FCS(fl,P), i = 1, ... ,n, then {el, ... , en} is called a fuzzy cycle basis of < CS(fl, p) > .

Theorem 2.33 Let {el, ... , en} be a cycle basis of < CS > . Then V et E

CS, there is a reordering of el, , en such that et = (el)tl + ... + (emkn,

m ~ n, where ti = peed, i = 1, , m.

Proof. Since {el, ... , en} is a basis for foot(CS), there is a reordering of

el, ... , en such that e = el + + em, m ~ n. Suppose that t > tl /\ ... /\ tTn·

Then there is a i; i = 1, ,m, such that t > ti· Nowe <J.< {ei + ... +

en}\{ei} > . Hence ({e) + +en} \ fed) U {e} is a basis for foot(CS).

However this contradicts the hypothesis that {e 1, ... , en} is a cycle basis of < CS > since t > ti. Thus t ~ tl /\ ... /\tTn. Now e = (Ul,Vl) + ... + (Urvr) where (Ui, Vi) E V x V, i = 1, ... , r. Hence et = (UI, Vl)a, + ... + (ur, vr)ar and (edt, + ... + (emkn = (UI, VI)a, + ... + (u,, vr)ar + Oa for some

a E (0,1] where P(Ui,Vi) = ai, i = 1, ... ,1', and tl /\ ... /\tm = al /\ /\ar/\a.

1\0\". al /\ ... /\ ar /\ a ~ al /\ ... /\ a; = t ~ tl /\ ... /\ tm = a) /\ /\ ar /\ a.

Hence t = tl /\ ... /\ tm and so et = (el)tl + ... + (eTnk,,· •

Corollary 2.34 Let {ej , ... , en} be a fuzzy cycle basis of < CS(fl, p) > .

Then V et E FCS(fl, p), there is a reorderinq oj el, , en such that et =

(c] )t, + ... + (em)tm, m ~ n, uihere t, = p(ei)' i = 1, , m.

Proof. FCS is the cycle set of (fl, PE) and the fuzzy cycle rank of (fl, p) is the cycle rank of (fl, P E)' •

The following graphs have the same cycle rank, but they are of course not isomorphic since one has four vertices and the other has five vertices.

Definition 2.14 The cocycle rank oj (fl, p), written mc(fl, p), is defined to be

mc(fl,p) = V{ L:7=lti I (eik ECoS, i = 1, ... ,n,{el,···,€n}

is a basis fOT Joot(CoS)}. The fuzzy cocycle rank of (11, p), written !mc(l1, p), is defined to be the cocycle rank of (fl,PE)' If {el,""C,,} is a basis fOT

40 2. FUZZY GRAPHS

foot(CoS) such that mcU.L, p) = L:~=l ti, where (ei)t; E CoS, i = 1, ... , n, then {ell ... , en} is called a co cycle basis of < CoS > . If {el, ... , en} is a basis for foot(FCoS({1, p)) such that fmc({1, p) = L:~=1 ti, where (eik E FCoS({1,p), i = 1, ... ,n, then {el, ... ,en} is called a fuzzy cocycle basis of < CS({1, p) > .

In a similar manner, we obtain the next two results.

Theorem 2.35 Let {e}, ... , en} be a cocycle basis of < CoS > . Then V et E CoS, there is a reordering ofel,···,en such that e, = (edtl+ ... + (emk", m :s n, where ti = p(ei), i = 1, ... , m .•

Corollary 2.36 Let {ell ... , en} be a fuzzy cocycle basis of < CS({1, p) > .

Then V et E FCS({1, p), there is a reordering of el, , en such that et =

(edt! + ... + (em}t,,, , m :s n, where ti = p(ei), i = 1, , m .•

2.2 Fuzzy Line Graphs

The results of this section are taken from [29]. The line graph, L(G), of a graph G is the intersection graph of the set of edges of G. Hence the vertices of L( G) are the edges of G with two vertices of L( G) adjacent whenever the corresponding edges of G are. In this section, the concept of a fuzzy line graph is introduced and its basic properties are developed. We give a necessary and sufficient condition for a fuzzy graph to be isomorphic to its corresponding fuzzy line graph. We also examine when the isomorphism between two fuzzy graphs follows from an isomorphism of their corresponding fuzzy line graphs. A necessary and sufficient condition for a fuzzy line graph to be the fuzzy line graph of some fuzzy graph is also presented in this section. Let G = (V. X) and G' = (V', X') be graphs. If {1 is a fuzzy subset of V and p is a fuzzy subset of V x V such that ({1, p) is a fuzzy graph and supp(p) t;;;; X, we call ({1, p) a partial fuzzy subgraph of G.

Definition 2.15 Let ({1,p) and ({1',p') be partial fuzzy subgraphs of G and G', respectively. Let f be a one-to-one function of V onto V'. Then

(1) f is called a (weak) vertex-isomorphism of ({1, p) onto ({1', p') if and onlyif Vv E V, ({1(v):S {1'(f(v)) andsupp({1') = (f(supp({1))) {1(v) = {1'(f(v)).

(2) f is called a (weak) line-isomorphism of ({1,p) onto ({1',p') if and only if V(u, v) EX, (p(u, v) :s p'(f(u), f(v)) and supp(p') = {(f(u), f(v)) I (u, v) E supp(p)}) p(u, v) = p'(f(u), f(v)).

2.2 Fuzzy Line Graphs 41

If ! is a (weak) vertex-isomorphism and a (weak) line isomorphism of (f..L,p) onto (f..L',p'), then! is called a (weak) isomorphism of (f..L,p) onto (f..L',p'). If (f..L,p) is isomorphic to (f..L',p'), then we write (f..L,p) ~ (f..L',p').

Our definition of isomorphism is equivalent to the one in [5, Definition 3.2, p. 160]. For homomorphisms [5, Definition 3.1, p. 159], our definitions of vertex-isomorphism and line-isomorphism are equivalent to the definitions of weak isomorphism and co-weak isomorphism, respectively, which appear in [5, Definition 3.3, 3.5, p.160]. We use the term 'weak' in a different manner than in [5].

Let G = (V, X) be a graph where V = {VI" ... , vn}. Let S, = {Vi, Xil, ... , Xiqi} where Xii E X and Xii has Vi as a vertex, j = 1, ... , qi; i = 1, ... , n. Let S = {Sl"",Sn}' Let T = {(Si,Si) I s.;s, E S, s, ivs, ::j:. 0,i::j:. j}. Then I(S) = (S, T) is an intersection graph and G ~ I(S). Any partial fuzzy subgraph (~, ')') of I(S) with suppl-y) = T is called a fuzzy intersection graph.

Let (f..L, p) be a partial fuzzy subgraph of G. Let I(S) be the intersection graph described above. Define the fuzzy subsets c, ')' of Sand T, respectively, as follows:

VSi E S, ~(Si) = f..L(Vi);

VSi E T, ')'(Si, Sj) = p(Vi, Vj).

Proposition 2.37 Let (f..L, p) be a partial fuzzy subgraph of G . Then

(1) (L, ')') is a partial fuzzy subgraph of I(S); (2) (f..L, p) ~ (t, ')').

Proof. (1) ')'(Si, Sj) = p(Vi, Vj) 'S f..L(Vi) 1\ f..L(Vj) = l(Si) 1\ [(Sj)

(2) Define f : V ~ S by !(Vi) = Si, i = 1, ... , n. Clearly, f is a one-toone function of V onto S. Now (Vi,Vj) E X if and only if (Si,Sj) E T and so T = {(f(vd,J(vj)) I (Vi,Vj) E X}. Also L(f(Vi)) = L(Si) = f..L(Vi) and ')'(f(vi),f(vj)) = ')'(Si,Sj) = P(Vi,Vj). Thus f is an isomorphism of (f..L,p) onto (~,')') .•

Let I(S) be the intersection graph of (V, X). Let (L, ')') be the fuzzy subgraph of I(S) as defined above. We call (~, ')') the fuzzy intersection graph of (f..L, p). Proposition 2.37 shows that any fuzzy graph is isomorphic to a fuzzy intersection graph.

Now L(G), the line graph of G, is by definition the intersection graph I(X). That is, L(G) = (Z, W) where Z = {{x}U{ux,vx} I x E X, Ux,Vx E V, x = (ux,vx)} and W = {(Sx,Sy) I s; n Sy ::j:. 0, x E X, x ::j:. y} and where Sx = {x} U {ux, vx}, x E X. Let (f..L, p) be a partial fuzzy subgraph of G. Define the fuzzy subsets >., w of Z, W, respectively, as follows:

VSx E Z, >'(Sx) = p(x);

V(Sx, Sy) E W, w(Sx, Sy) = p(x) 1\ p(y).

42 2. FUZZY GRAPHS

Proposition 2.38 (>., w) is a fuzzy subgraph of L( G), called the fuzzy line graph corresponding to (/-L, p).

Proof. w(Sx, Sy) = p(x) 1\ p(y) = >'(Sx) 1\ >'(Sy) .•

\Ve recall that a cutpoint of a graph is a vertex whose deletion increases the number of components, and a bridge is such an edge. Every cutpoint of L(G) is a bridge of G which is not an endline, and conversely, [20, p.71].

Recall that (u, v) E X is defined to be a bridge of (/-L, p) jf and only if deleting (u, v) reduces the strength of connectedness between some pair of vertices. Also, recall that v E V is defined to be a cutpoint in (/-L, p) if and only if deleting v reduces the strength of connectedness between some pair of vertices not including v.

The following example shows that the relationship between cutpoints in L( G) and bridges in G does not carryover to the fuzzy case for the above definitions of cutpoints and bridges.

Example 2.11 Let GI be the graph defined in [20, p.72J, that is, G1 = (V,X) where V = {VI,V2,V3,V4} and X = {Xl = (Vt.V2),X2 = (Vt.V3), X3 = (V2,V3),X4 = (V3,V4)). Let /-L(VI) = /L(V2) = /-L(va) = /-L(V4) = 1, p(xd = p(X3) = I, and p(X2) = p(X4) = 1/2. Then >'(SX1) = >'(SX3) = 1,>'(Sx2) = >'(SxJ = 1/2, andw(Sx"Sx2) = 1,w(Sx2,Sx3) = W(SX3'SX4) = W(SX2' Sx.) = 1/2. If we delete Xl from GI, then the strength of connectedness between VI and V2 is 1/2 = p(X2) 1\ p(X3) while the strength of connectedness between VI and V2 before the deletion of Xl is 1 = p(xd. Thus Xl is a bridge of(/-L, p) (and not an endline of Gd. However the strength of connectedness between any pair of vertices SX2' SX3' SX4 is 1/2 before and after the deletion of SXI' Thus SXI is not a cutvertex of (>., w).

Proposition 2.39 Let (/-L, p) and (/-L', p') be partial fuzzy subgraphs of G and G', respectively. If f is a weak isomorphism of (/-L, p) onto (/-L', p'), then f is an isomorphism of (supp(/-L), supp(p)) onto (supp(/-L'), supp(p')).

Proof. v E supp[rz] ¢:} f(v) E supptu') and (u, v) E supp(p) ¢:} (f(u), f(v)) E supp(p') .•

Proposition 2.40 If (>., w) is the fuzzy line graph of (/-L, p), then (supp(>.), supp(w)) is the line graph of (supp(/-L), supp(p)).

Proof. (/-L,p) is a partial fuzzy subgraph of Gand (>.,w) is a partial fuzzy subgraph of L(G). Now >'(Sx) = p(x) 'Ix E X and so Sx E suppfA) ¢:} X E supp(p). Also w(Sx, Sy) = p(x) 1\ p(y) V(Sx, Sy) E W and so supp(w) = {(Sx,Sy) I Sx n s, 1= 0, x,y E supp(p), X 1= y} .•

2.2 Fuzzy Line Graphs 43

We also see in Proposition 2.40 that (A ISUpp(A),W ISl1pp(W») is the fuzzy line graph corresponding to (/-L ISl1PP(P.)'P Isupp(p»).We now give a necessary and sufficient condition for fuzzy graph (/-L, p) to be isomorphic to its fuzzy line graph (A, w) .

Theorem 2.41 Let (A, w) be the fuzzy line graph corresponding to (/-L, p). Suppose that (8Upp(/-L), supp(p)) is connected. Then

(1) :3 a weak isomorphism of (/-L,p) onto (A,W) if and only if (supptu}, supp(p)) is a cycle and "tv E supplu}, /-L(v) = p(x), that is, /-L and p are constant functions of supp(/-L) and supp(p), respectively, taking on the same value.

(2) If f is a weak isomorphism of (/-L, p) onto (A, w), then f is an isomorphism.

Proof. Suppose that f is a weak isomorphism of (/-L,p) onto (A,W). By Proposition 2.39, f is an isomorphism of (SUpp(/-L), supp(p)) onto (SUpp(A), supp(w)). By Proposition 2.40, (supptu}, supp(p)) is a cycle, [20, Theorem

8.2, p.72]. Let suppfjr) = {vt, , vn} and supp(p)

= { (VI, V2), (V2, V3), ... , (Vn, vd} where VI V2V3 VnVI is a cycle. Let 11( Vi)

= 81 and p(VI, VI+I) = Ti,i = 1, ... , n where Vn+l = VI· Then for 8n+l = 81,

(2.1)

Now SUpp(A) = {S(v;,v;+d I i = 1, ... ,n} and supp(w) {(S(v;,v;+d' S(11i+l ,V;+2») I i = 1, ... ,n-l}. Also for Tn+l = Tl, A(S(Vi,Vi-ttl) = P(Vi' vi+J) = Ti and w(S(v;,v;+d' S(v;+I,V;+2») = p(Vi' Vi+l)/\P(Vi+l, Vi+2) = Ti/\Ti+l, i = 1, ... , n, where Vn+2 = V2. Since f is an isomorphism of (suppfjz}, supp(p)) onto (SUpp(A), supp(w)), f maps supptu) one to one onto SUpp(A) = {S(l.",V2)' ... ,S(vn,vtl}· Also f preserves adjacency. Hence f induces a permut.ation 11 of {I, ... ,n} such that f(Vi) = S(V"(i),V"'(i)+tl and (vi,vi+d-+ (f(Vi), f( Vi+l)) = (S(v".(;) ,V"(;j+,),S(V,,.(i_l) ,V"C;+I)+!l)' i = 1, ... , n - 1. Now s; = /-L(Vi) :::; A(J(Vi)) = A(S(v"..:;),v,,:;)_Il) = T1r(i) and Ti = P(Vi, Vi+l) :::; w(J(vd, f(Vi+I)) = W(S(V"'(i) ,V7r(;)+tl,S(V"'(Hl) ,v7rci+l)+Il) = P(V1r(i), V1r(i)+1) /\ p( V1r(i+l), V1r(i+l)+l) = T 1r(i) /\ T 1r(i+1) , i = 1, ... ,n. That is,

8i :::; T1r(i) and Ti :::; T1r(i) /\ T1r(i+l), i = 1, ... , n. (2.2)

By the second part of (2.2), we have that Ti :::; T1r(i) , i = 1, ... , n, and so T1r(i) :::; T1r(1r(i»' i = 1, ... , n. Continuing, we have that Ti :::; T1r(i) :::; ... :::; T1ri(i) :::; Ti and so r; = T1r(i) , i = 1, ... , n, where 1Ij+l is the identity map. By (2.2) again, we have r, :::; T1r(i+l) = Ti+I, i = 1, ... , n where Tn+1 = TI. Hence by (2.1) and (2.2), Tl = ... = Tn = 51 = .,. = 8n. Thus we have not only proved the conclusion about /-L and p being constant functions, but we have also shown that (2) holds. Conversely, suppose that (suPP(/-L),

44 2. FUZZY GRAPHS

supp(p)) is a cycle and'r/v E suppiu), 'r/x E supp(p), p.(v) = p(x). By Proposition 2.40, (supp(.A), supp(w)) is the line graph of (supp(p.), supp(p)). Since (supp(p.), supp(p)) is a cycle, (supp(p.), supp(p)) ~ (suppf X), supp(w)) by [20, Theorem 8.2, p.72]. This isomorphism induces an isomorphism of (p.,p) onto (.A, w) since p.(v) = p(x) 'r/v E V, 'r/v E X and so p. = p = .A = w on their respective domains .•

Theorem 2.42 Let (J.l-, p) and (p.', p') be the partial fuzzy subgraphs of G and a, respectively, such that (supp(p.), supp(p)) and (supp(p.'), supp(p')) are connected. Let (.A, w) and (.A', w') be the line graphs corresponding to (J.l-, p) and (J.l-', p'), respectively. Suppose that it is not the case that one of (supp(p.), supp(p)) and (supp(p.'), supp(p')) is K3 and the other is K1,3. If (>.,w) ~ (.A',w'), then (J.l-,p) and (p.',p') are line-isomorphic.

Proof. Since (.A,w) ~ (.A',w'),(suPP(.A), supp(w)) ~ (supp(>"), supp(w')) by Proposition 2.39. Since (supp(.A), supp(w)) and (suppfx'), supp(w')) are line graphs of (supptu), supp(p)) and (supp(p.'), supp(p')) respectively, by Proposition 2.40, we have that (supp(J.l-), supp(p)) c:= (supp(u'], supp(p')) by [20, Theorem 8.3, p.72j. Let g denote the isomorphism of (>.,w) onto (>.',w') and f the isomorphism of (suppfu), supp(p)) onto (suppfjr'}, supp(p')). Then >'(S(u,v») = .A' (g(S(u,v»)) = .A' (S(!(u),f(v») where the latter equality holds the proof of [20, Theorem 8.3, p.72] and so p(u, v) = p' (f (u), f (v)). Hence (p., p) and (p.', p') are line isomorphic .•

Proposition 2.43 Let (r, v) be a partial fuzzy subgraph of L( G). Then (r, v) is a fuzzy line graph of some partial fuzzy subgraph of G if and only if'r/(S", Sy) E W, v(S", Sy) = r(S,,) /\ r(Sy).

Proof. Suppose that v(S", Sy) = r(S,,) /\ r(Sy) 'r/(S", Sy) E W. 'r/x E X, define p(x) = r(S,,). Then v(S",Sy) = r(S,,) /\ r(Sy) = p(x) /\ p(y). Any p. that yields the property p(u, v) $ J.l-(u) /\ J.l-(v) will suffice, e.g., p.(v) = 1 'r/v E V. The converse is immediate .•

Not every graph is a line graph of some graph. The following result tells us when a fuzzy graph is a fuzzy line graph of some fuzzy graph.

Theorem 2.44 (J.l-, p) is a fuzzy line graph if and only if (supp(J.l-), supp(p)) is a line graph and 'r/(u, v) E supp(p),p(u, v) = p.(u) /\ p.(v).

Proof. Suppose that (J.l-, p) is a fuzzy line graph. Then the conclusion holds by the Proposition 2.40 and 2.43. Conversely, suppose that (supp(p.), supp(p)) is a line graph and 'r/(u, v) E supp(p), p(u, v) = J.l-(u) /\ p.(v). Then the conclusion holds from Proposition 2.43 .•

Fuzzy Graphs and Fuzzy Hypergraphs - Part1'

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fuzzy Graphs and Fuzzy Hypergraphs - Part1'

Uploaded by

Copyright:

Available Formats

Fuzzy Graphs and

You might also like

Fuzzy Graphs and Fuzzy Hypergraphs - Part1'

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fuzzy Graphs and Fuzzy Hypergraphs - Part1'

Uploaded by

Copyright:

Available Formats

﻿Fuzzy Graphs and

You might also like

Fuzzy Graphs and