Académique Documents
Professionnel Documents
Culture Documents
K23798 2015/2/2
i i
Application of
Fuzzy Logic to
Social Choice
Theory
John N. Mordeson
Creighton University
Omaha, Nebraska, USA
Davender S. Malik
Creighton University
Omaha, Nebraska, USA
Terry D. Clark
Creighton University
Omaha, Nebraska, USA
i i
i i
i i
K23798 2015/2/2
i i
Series Editors
John A. Burns
Thomas J. Tucker
Miklos Bona
Michael Ruzhansky
Chi-Kwong Li
Published Titles
Application of Fuzzy Logic to Social Choice Theory, John N. Mordeson, Davender S. Malik
and Terry D. Clark
Blow-up Patterns for Higher-Order: Nonlinear Parabolic, Hyperbolic Dispersion and
Schrdinger Equations, Victor A. Galaktionov, Enzo L. Mitidieri, and Stanislav Pohozaev
Iterative Optimization in Inverse Problems, Charles L. Byrne
Modeling and Inverse Problems in the Presence of Uncertainty, H. T. Banks, Shuhua Hu,
and W. Clayton Thompson
Set Theoretical Aspects of Real Analysis, Alexander B. Kharazishvili
Signal Processing: A Mathematical Approach, Second Edition, Charles L. Byrne
Sinusoids: Theory and Technological Applications, Prem K. Kythe
Special Integrals of Gradshetyn and Ryzhik: the Proofs Volume l, Victor H. Moll
Forthcoming Titles
Actions and Invariants of Algebraic Groups, Second Edition, Walter Ferrer Santos
and Alvaro Rittatore
Analytical Methods for Kolmogorov Equations, Second Edition, Luca Lorenzi
Complex Analysis: Conformal Inequalities and the Bierbach Conjecture, Prem K. Kythe
Cremona Groups and Icosahedron, Ivan Cheltsov and Constantin Shramov
Dictionary of Inequalities, Second Edition, Peter Bullen
Difference Equations: Theory, Applications and Advanced Topics, Third Edition,
Ronald E. Mickens
Geometric Modeling and Mesh Generation from Scanned Images, Yongjie Zhang
Groups, Designs, and Linear Algebra, Donald L. Kreher
Handbook of the Tutte Polynomial, Joanna Anthony Ellis-Monaghan and Iain Moffat
Lineability: The Search for Linearity in Mathematics, Juan B. Seoane Sepulveda,
Richard W. Aron, Luis Bernal-Gonzalez, and Daniel M. Pellegrinao
Line Integral Methods and Their Applications, Luigi Brugnano and Felice Iaverno
Microlocal Analysis on Rn and on NonCompact Manifolds, Sandro Coriasco
Monomial Algebra, Second Edition, Rafael Villarreal
Partial Differential Equations with Variable Exponents: Variational Methods and
Quantitative Analysis, Vicentiu Radulescu
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
2015 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copy-
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro-
vides licenses and registration for a variety of users. For organizations that have been granted a photo-
copy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
i i
K23798 2015/2/2
i i
Contents
Preface ix
i i
i i
i i
K23798 2015/2/2
i i
vi CONTENTS
7 Rationality 209
7.1 Fuzzy Preference and Preference-Based Choice Functions . . . 209
7.2 Fuzzy Choice Functions, Revealed Preference and Rationality . 222
7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
7.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
i i
i i
i i
K23798 2015/2/2
i i
CONTENTS vii
Index 333
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Preface
This book provides a comprehensive study of fuzzy social choice theory. Hu-
man thinking is marked by imprecision and vagueness. These are the very
qualities that fuzzy logic seeks to capture. Thus, social choice theory suggests
itself as a means for modeling the uncertainty and imprecision endemic in
social life. Nonetheless, fuzzy logic has seen little application in the social sci-
ences to include social choice theory. We attempt to partially fill this lacuna.
The main focus of Chapter 1 is the concept of a fuzzy maximal subset of
a set of alternatives X. A fuzzy maximal subset gives the degree to which the
elements of X are maximal with respect to a fuzzy relation on X and a fuzzy
subset of X. This allows for alternative notions of maximality not allowed in
the crisp case. The main result states that a fuzzy maximal subset is not the
zero function if and only if the fuzzy preference relation involved is partially
acyclic.
Chapter 2 deals with fuzzy choice functions. Classical revealed preference
theory postulates a connection between choices and revealed preferences. If it
is assumed that preferences of a set of individuals are not cyclic, then their col-
lective choices are rationalizable. We examine this question in some detail for
the fuzzy case. We determine conditions under which a fuzzy choice function
is rationalizable. We also present results inspired by Georgescu [24]. Desai [16]
characterized the rationality of fuzzy choice functions with reflexive, strongly
connected, and quasi-transitive rationalization in terms of path independence
and the fuzzy Condorcet property. We consider these results as well as those
of Chaudhari and Desai [11] on various types of rationality.
Chapter 3 is concerned with the factorization of a fuzzy preference rela-
tion into the union of a strict fuzzy relation and an indifference operator,
where union here means conorm. Such a factorization was motivated by Fono
and Andjiga [12]. The factorization is useful in the examination of Arrowian
type results. We show that there is an inclusion reversing correspondence be-
tween conorms and fuzzy strict preference relations in factorizations of fuzzy
preference relations into their strict preference and indifference components.
In Chapter 4, we consider fuzzy non-Arrowian results. The first author
felt that presenting this material before Arrowian type results would help
emphasize the difference between the fuzzy case and the crisp case. We pro-
vide two types of fuzzy aggregation rules that satisfy certain fuzzy versions of
ix
i i
i i
i i
K23798 2015/2/2
i i
x PREFACE
Arrowian conditions, but which are not dictatorial. We conclude the chapter
by presenting the work of Duddy, Perote-Pena and Piggins [17] that demon-
strates the central role max- transitivity plays in fuzzy preference theory,
and in particular Arrows Theorem.
Chapter 5 is one of the main chapters of the book. Arrows Theorem is one
of the most important discoveries in social theory. It states that if preferences
of individuals are aggregated under certain reasonable conditions: universal
admissibility, transitivity, unanimity, and independence of irrelevant alterna-
tives, then the aggregation method must be dictatorial. We consider versions
of Arrows conditions and show that they lead to fuzzy versions of Arrows
Theorem, and versions involving representation rules, oligarchies, and veto
players. In the fuzzy situation, there are many types of transitivity, indepen-
dence of irrelevant alternatives, and strict preference relations. We examine
how various combinations of these concepts yield impossibility theorems. In
particular, we consider the work of Dutta [11], where he demonstrates that
a fuzzy aggregation rule is oligarchic rather than dictatorial when a rather
strong definition of transitivity is used. However, dictatorship holds if positive
responsiveness is assumed. We also present the ideas introduced by Banerjee
[4]. These include the notion that for exact preference relations and fuzzy so-
cial preference relations, it is possible to distinguish between different degrees
of power of a dictator.
One solution to the Arrowian Impossibility Theorem is to place restric-
tions on preferences. Among the more promising approaches is that of Blacks
Median Voter Theorem, which restricts preferences to single-peaked profiles.
In Chapter 6, we show that the Median Voter Theorem holds for fuzzy pref-
erences. We also demonstrate that Blacks Median Voter Theorem holds for
fuzzy single-dimensional spatial models even though there are significant dif-
ferences between conventional models and their fuzzy counterparts. In crisp
spatial models, preferences in space are most often represented by a single
ideal point whereas in the fuzzy case, a political actors ideal policy position
encompasses a range of alternatives.
In Chapter 7, we consider the questions of how unambiguous or exact
choices are generated by fuzzy preferences and whether exact choices induced
by fuzzy preferences satisfy certain plausible rationality relations. The results
are based on those of Barrett, Pattanaik, and Salles [9]. Several alternative
rules for generating exact choice sets from fuzzy weak preference relations are
presented. To what extent exact choice sets generated by these alternative
rules satisfy certain fairly weak rationality conditions is also examined. We
also consider Banerjees results [5] concerning the situation where the domain
of the choice function consists of all crisp finite subsets of the universal set
of alternatives, but where the choice is fuzzy in the sense that the decision
maker can state the degree of his choice of an alternative.
We extend some known Arrowian results involving fuzzy set theory to
results involving intuitionistic fuzzy sets in Chapter 8. We show how the use of
an involutive fuzzy complement can be used to obtain results. Nana and Fono
i i
i i
i i
K23798 2015/2/2
i i
xi
Acknowledgments
We are indebted to the political science students in the mathematics collo-
quium, especially Crysta Price. We thank Dr. George Haddix and his late
wife Sally for their generous endowments to the Department of Mathematics
at Creighton University. We thank the journal New Mathematics and Natural
Computation for allowing us to reuse some of our work.
John N. Mordeson
Davender S. Malik
Terry D. Clark
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 1
i i
i i
i i
K23798 2015/2/2
i i
Further exacerbating the problem is that fuzzy logic has not been able to
overcome what has become the paradigmatic hold of probability theory on
empirical studies in the discipline. Claudio Cioffi-Rivilla [3] was the first to
argue that fuzzy mathematics could have important applications in political
science. Charles Ragin [11] subsequently proposed a fuzzy logic method for
testing necessary and sufficient conditions as well as identifying multiple paths
of causality between a set of conditions as well as identifying multiple paths
of causality between a set of conditions and an outcome. Pennings [10] used
Ragins method in a cross-national analysis of executive accountability in
parliamentary systems; Arfi [1], Koenig-Archibugi [7], and Sanjian [12-17] used
fuzzy logic to test hypotheses of decision-making; and Tabor [21] and Seitz [19]
employed fuzzy expert systems in computer simulations of regional rivalries.
Despite these promising efforts, the property ranking issue has proven an
insurmountable obstacle to the wider application of fuzzy logic in empirical
studies in the social sciences. See [20, 22]. Fuzzy logic approaches for estab-
lishing correlations between variables require that the data be measured on
comparable scales, otherwise there is no way to overcome the charge that the
conclusions reached are little more than consequences of how the data are
measured. In short, fuzzy logic methods are substantially more sensitive to
data measurement issues than are the conventional statistical methods that
dominate empirical research in the social sciences.
Fuzzy logic applications in formal-theoretical approaches do not suffer from
these problems. This is particularly the case for theoretical work in social
choice, a research agenda that has arguably been the most dynamic formal-
theoretical agenda in the social sciences. This book is a purposeful effort to lay
out how fuzzy logic can contribute to this agenda. In the rest of this chapter,
we consider some basic concepts in fuzzy social choice theory. In subsequent
chapters, we explore more nuanced issues and applications.
i i
i i
i i
K23798 2015/2/2
i i
Definition 1.1.1 Let be a fuzzy binary operation on [0, 1]. Then is called
a t-norm if the following conditions hold a, b, c [0, 1] :
(1) (a, 1) = a (boundary condition);
(2) b c implies (a, b) (a, c) (monotonicity);
(3) (a, b) = (b, a) (commutativity);
(4) (a, (b, c)) = ((a, b), c) (associativity).
Example 1.1.2 Some basic t-norms are defined as follows. Let a, b [0, 1].
Standard intersection: (a, b) = a b;
Algebraic product: (a, b) = ab;
Bounded difference: (a, b) = 0 (a + b 1);
a if b = 1,
Drastic intersection: (a, b) = b if a = 1, .
0 otherwise.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Definition 1.1.6 Let F R(X). Define the fuzzy binary relations (i) , i =
x, y X,
0, 1, 2, 3 as follows:
(x, y) if (x, y) > (y, x) = 0
(1) (0) (x, y) =
0 otherwise.
(x, y) if (x, y) > (y, x)
(2) (1) (x, y) =
0 otherwise.
(3) (2) (x, y) = 1 (y, x).
(4) (3) (x, y) = 0 ((x, y) (y, x)).
The strict preference (1) requires only that the degree of strict preference
of x to y is greater than its degree of preference of y to x. Then the degree
of strict preference of x to y equals its preference. The size of the degree of
preference of y to x is immaterial. This is in contrast to (3) where the size of
the degree of preference of y to x is taken into account.
i i
i i
i i
K23798 2015/2/2
i i
Example 1.2.2 Let X = {x, y, z}. Let FP (X) be such that (x) =
0, (y) = 1/4,and (z) = 1/2. Let F R (X) be such that (w, w) = 1
w X, (x, y) = 1/4, (y, z) = 1/8, (x, z) = 3/4, ( , ) = 0 otherwise.
Then since (x) = 0, it is immediate that MG (, )(x) = 0. Now
MG (, )(y) = (y) [({t [0, 1] | (y) (y, y) t (y, y)})
~({t [0, 1] | (z) (z, y) t (y, z)})]
= (1/4) (1 ~ 1)
= 1/4
and
MG (, )(z) = (z) [({t [0, 1] | (y) (y, z) t (z, y)})
~({t [0, 1] | (z) (z, z) t (z, z)})]
= (1/2) (0 ~ 1)
= 0.
i i
i i
i i
K23798 2015/2/2
i i
Example 1.2.4 Let X = {x, y, z} and R = X {(y, z), (z, x)}. Let S =
{x, y}. Then since 1R (y, x) = 0 and 1S (z) = 0,
MG (1R , 1S )(x) = 1S (x) [({t [0, 1] | 1S (x) 1R (x, x) t 1R (x, x)}) ~
({t [0, 1] | 1S (y) 1R (y, x) t 1R (x, y)})]
= 1 (1 ~ 1) = 1.
Note that R is not complete. However, x S, but (x, v) / R v S since
(x, y)
/ R. That is, Proposition 1.2.5 below doesnt hold. Thus this definition
of MG (, ) doesnt contain the crisp case.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Let = (0) . Then not (xn , x1 ) > 0 not ((xn , x1 ) > 0 and (x1 , xn ) =
0) (xn , x1 ) = 0 or (x1 , xn ) > 0 (x1 , xn ) > 0, where (xn , x1 ) = 0
(x1 , xn ) > 0 assuming is complete. That is, for complete the definitions
of partial acyclicity and strict acyclicity are equivalent for = (0) .
Now let be a fuzzy preference relation on X and define by x, y
X, (x, y) = 1 if (x, y) = 1, (y, x) = 0; and (x, y) = 0 otherwise. Then
is asymmetric. Suppose (x1 , xn ) = 0 and (xn , x1 ) = 1/2. Then (x1 , xn ) =
0 = (xn , x1 ), Thus (xn , x1 ) = 0 ; (x1 , xn ) > 0. Note also if is a
fuzzy preference relation such that (x1 , xn ) = 1/2 and (xn , x1 ) = 3/4, then
(xn , x1 ) > 0 for many types of strict preference relations associated with .
This discussion shows that strict acyclicity does not imply partial acyclicity
and partial acyclicity does not imply strict acyclicity.
The preceding result and the following corollary give corresponding results
to Theorem 1.2.10 for regular fuzzy preference relations.
i i
i i
i i
K23798 2015/2/2
i i
Suppose is regular. Then (x, y) > 0, (y, z) > 0, (z, x) > 0 (since 6=
(0) ). Thus is not strictly acyclic. Let FP (X). Then M (, ) 6= since
(u, v) > 0 u, v X. Note Md (, ) = .
i i
i i
i i
K23798 2015/2/2
i i
Then Theorem 1.2.18 holds for this definition with the same proof ex-
cept for the conclusion: Since is strictly acyclic and is reflexive, not
(xi , xn ) > (xn , xi ) for i = 1, . . . , n Thus (xn , xi ) > 0 for i = 1, . . . , n.
Hence M (, )(xn ) > 0 and so M (, ) 6= .
The following result gives a necessary and sufficient condition for an ele-
ment to belong to a level set of a fuzzy maximal subset.
i i
i i
i i
K23798 2015/2/2
i i
x M (, )t M (, )(x) t
(x) (wX ({r [0, 1] | C(x, w) r (x, w)})) t
x t and wX ({r [0, 1] | C(x, w) r (x, w)})) t
x t and {r [0, 1] | C(x, w) r (x, w)}) t w X
x t and C(x, w) t (x, w) w X.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Corollary 1.2.31 Let F R (X). Suppose that there exists t (0, 1] such
that t is reflexive, complete, and acyclic. Then FP (X) such that
t 6= , MM (, ) 6= .
i i
i i
i i
K23798 2015/2/2
i i
1.3. Exercises 17
1.3 Exercises
1. Let X be a finite set and a R relation on X. Prove that a necessary
condition for the maximal set M (R, S) to be nonempty for all subsets S
of X is that R be reflexive and complete. Prove also that transitivity of
R together with completeness and reflexivity are sufficient for M (R, S)
to be nonempty for all subsets S of X.
2. Show by example that the transitivity of R is not necessary for M (R, S)
to be nonempty for all S.
3. Let R be reflexive and complete. Prove that M (R, S) 6= for all subsets
S of X if and only if R is acyclic.
1.4 References
1. B. Arfi, Fuzzy decision making in politics: A linguistic fuzzy set approach
(LFSA), Political Analysis, 13 (2005) 2356.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
1.4. References 19
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 2
21
i i
i i
i i
K23798 2015/2/2
i i
a fuzzy Arrow axiom, a weak fuzzy congruence axiom, and a strong fuzzy
congruence axiom.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
It follows that C ({x, y}) = C ({x, z}) = C ({x, y, z}) = {x} and C ({y,
z}) = {y, z}. Define : X [0, 1] as follows: (x) = 1, (y) = (z) = 1/4.
Then C()(x) = 1, C()(y) = C()(z) = 0. Hence C (1 ) = C ({x}) =
{x} = C()1 and C (1/4 ) = C ({x, y, z}) = {x} = C()1/4 . Note that
C(1{x,z} )(x) = 1/2 and so C(1{x,z} ) 6= 1T for some T {x, z}. Thus it is
not the case that S P (X), C(1S ) = 1T for some T S.
Define : X [0, 1] by (x) = (y) = (z) = 1/8. Since t = for t
(1/8, 1], 1/8 = X, and (y, x), (z, x)
/ t t (0, 1], it follows that C()(x) =
1/8, C()(y) = 0, and C()(z) = 0.
Example 2.1.5 shows that it is not always the case C restricted to P (X)
can be considered a choice function on X. In fact, Proposition 1.2.35 shows
that this can only be the case if is crisp. Nevertheless, if this condition is
assumed, a theory involving fuzzy subsets of X is still possible.
We now proceed with a formal definition of the degree to which a fuzzy
choice function will select an outcome in the fuzzy maximal set, i.e., whether
a fuzzy choice function is rationalizable.
Let t = ({(x, y) | (x, y) XX, (x, y) > 0}. Then t > 0 since X is finite.
Since is complete, we have that x, y X either (x, y) t or (y, x) t .
Let x1 , . . . , xn X and t (0, t ]. Let be the strict preference relation asso-
ciated with . Since is acyclic, it follows that (x1 , x2 ), (x2 , x3 ), . . . , (xn1 , xn )
t = Supp() implies (x1 , xn ) t = Supp(). Since Supp() is the strict
i i
i i
i i
K23798 2015/2/2
i i
Example 2.1.7 Let X = {x, y, z}. Suppose has no zero divisors. Define the
fuzzy relation on X as follows:
For the Lukasiewicaz t-norm, MG (, 1{x,y} )(y) = 1/2 since 1/2 t = 0 for
t = 1/2. Recall that the Lukasiewicaz t-norm has zero divisors.
i i
i i
i i
K23798 2015/2/2
i i
x M (R, S)
x S and xRy y S 1S (x) = 1
and 1R (x, y) = 1 y such that 1S (y) = 1
1S (x) (~{{t [0, 1] | C(y, x) t 1R (x, y)} | y X} = 1
M (1R , 1S )(x) = 1
C(1S )(x) = 1 1T (x) = 1 x T
x Supp(C(1S )) x C (S).
i i
i i
i i
K23798 2015/2/2
i i
demand functions. Since then Uzawa [49] and Arrow [1] have developed re-
vealed preference theory in an abstract framework. The work of Uzawa and
Arrow was continued mainly by Richter [40], Sen [42, 44] and Suzumura [47,
48]. The work of Uzawa, Arrow and Sen was based on the assumption that the
domain of choice functions under consideration contains all non-empty finite
subsets of the set of alternatives of the universal set, whereas Richter [40] and
Suzumura [47] defined choice functions on an arbitrary non-empty class of
non-empty subsets of the universal set and studied rationality of choice func-
tions by means of properties as the revealed preference axioms, the congruence
axioms and the consistency conditions.
Example 2.2.1 Consider the crisp example, X = {x, y, z}, S = {y, z}, and
R = X {(x, y), (y, z), (x, z)}. Then M (R, S) = {w S | (w, v) R for all
v S} = {y}.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Proof. Suppose C(1{x1 ,...,xn } )(xi ) > 0 for some i = 2, . . . , n. Then since
C satisfies condition ,
0 < C(1{x1 ,...,xn } )(xi ) 1{xi1 ,xi } (xi ) = (C(1{x1 ,...,xn } ) 1{xi1 ,xi } )(xi )
C(1{xi1 ,xi } )(xi ) = C (xi , xi1 ),
0 < C(1{x1 ,...,xn } )(x1 ) 1{x1 ,xn } (x1 ) = C(1{x1 ,...,xn } ) 1{x1 ,xn } )(x1 )
C(1{x1 ,xn } )(x1 ) = C (x1 , xn ).
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Example 2.2.12 Let X = {x, y}. Let C be the choice function on X such
that
Then
C (x, x) = C (y, y) = 1, C (x, y) = 1/2, C (y, x) = 0.
It follows easily that C(1S ) = MG (C , 1S ) S = {x} or {y}, but C(1X )(x) =
1/2 < 1 = MG (C , 1X )(x). Thus C(1X ) MG (C , 1X ). Hence C does not
rationalize C. Clearly, C is reflexive,complete, and acyclic.
Path independence in the crisp case allows for dividing the set of alterna-
tives into smaller subsets, say S and T, then choosing a subset from S T
and making the final decision from this subset.
i i
i i
i i
K23798 2015/2/2
i i
(2) Define \ by x X,
(x) if (x) < (x),
(\)(x) =
0 if (x) (x).
(3) Define \ by x X,
(x) (x)if (x) < (x),
(\)(x) =
0 if (x) (x).
Let \ be a set difference of FP(X). Let S(\) = {(, ) | , FP(X),
(\) = and (\) = 1 }. We note that if \ is defined as in Definition
2.2.14(1), then (\) = 1 and (\) = for those and such
that Supp()Supp() 6= and = on Supp()Supp().
i i
i i
i i
K23798 2015/2/2
i i
Then
C satisfies condition .
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Lemma 2.3.1 For any a, b, c [0, 1], the following properties hold:
(1) a b c a b c;
(2) a (a b) = a b.
Proof. (1) follows immediately from the definition of b c.
(2) Since a (a b) b, a b a b. Also, a (a b) b since is
continuous and clearly, a (a b) a.
(x) {(x, y) | (y) > (x, y), y Supp()}
G(, )(x) = if y Supp() such that (y) > (x, y),
(x) if (y) (x, y) y Supp().
i i
i i
i i
K23798 2015/2/2
i i
and
{(x, y) | y S} if x S,
G(1S , )(x) =
0 if x
/ S.
Recall that our definition of a fuzzy maximal set is M (, )(x) = (x)
(~{{t [0, 1] | C(y, x)t (x, y)} | y X}). We obtain M above by letting
C(y, x) = (y) (y, x) and we obtain G above by letting C(y, x) = (y). Note
that M above is MG . We obtain MM by letting C(y, x) = 1.
The negation operator associated with a continuous t-norm is defined by
a = a 0 = {t [0, 1] | a t = 0}.
i i
i i
i i
K23798 2015/2/2
i i
e(x, y) is the degree of truth of the statement there exists a criterion
with respect to which x is chosen and y is rejected.
Let = . Since C()(y) = 1 if C()(y) = 0 and C()(y) = 0 if
C()(y) > 0, we have that
{C()(x) (y) | B} if C()(y) = 0 for some B,
e(x, y) =
0 if C()(y) > 0 for all B.
Thus e(x, y) = 1 if C()(y) > 0 for all B and e(x, y) = 0 if there exists
B such that C()(y) = 0, (y) > 0, and C()(x) > 0.
Hence it follows that (x1 , x01 ) = 0 and (x01 ) > 0. Since is reflexive, x1 6= x01 .
Let y1 = x1 and y2 = x01 . By applying the same argument to y2 , there exists
i i
i i
i i
K23798 2015/2/2
i i
Definition 2.3.13 [24] Let C be a fuzzy choice function on (X, B) and let
be the associated revealed preference relation (Definition 2.3.7). Define the
functions G : B FP(X) and M : B FP(X) by B, x X,
and
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Theorem 2.3.15 (Georgescu [24]) Let C be a fuzzy choice function then the
following properties hold.
(1) Conditions (i), (ii) are equivalent.
(2) The implication (i) (iii) holds; if = , then the implication (iii)
(i) holds;
(3) Conditions (iii) and (iv) are equivalent;
(4) If is the Lukasiewicz t-norm, then conditions (iii), (v), (vi), (vii) are
equivalent;
(5) The implication (viii) (iii) holds.
Let C be a fuzzy choice function. Let C b denote the fuzzy choice func-
tion G = G( , ), [22, p. 109]. Recall is the associated revealed preference
relation of C.
i i
i i
i i
K23798 2015/2/2
i i
Thus
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
By condition F ,
Since these inequalities hold for all B, (x) C()(y) (x, y) C()(x).
Thus W F CA holds for C.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 2.4.1 [21, 24, 31, 32] If is the fuzzy implication associated with
the Godel t-norm on [0, 1], then
(1) a b c a b c;
(2) a b a b = 1;
(3) 1 a = a.
i i
i i
i i
K23798 2015/2/2
i i
Clearly, .
Lemma 2.4.2 Let C be a fuzzy choice function on (X, B). Then and
both are reflexive and strongly connected.
Proof. Since B is the base domain and the fuzzy choice function C is
normal, we have for every x, y X, 1{x} , 1{y} , 1{x,y} B. Also, C(1{x} )(x) = 1
and C(1{x,y} )(x) = 1 or C(1{x,y} )(y) = 1. Thus and are reflexive and
strongly connected.
Let be a fuzzy preference relation on X and B a nonempty family of
nonzero fuzzy subsets of X. For all B, define the fuzzy set G(, ) of X
is as follows: x X,
i i
i i
i i
K23798 2015/2/2
i i
Definition 2.4.3 [10, 22] Let C be a fuzzy choice function. Then C is said
to be G-rational, if there exists a fuzzy preference relation on X such that
C() = G(, ) for all B. The fuzzy preference relation for which
the fuzzy choice function C is G-rational is called a rationalization for C.
The fuzzy choice function C is called full rational if C is G-rational with
reflexive, strongly connected, and max-min transitive rationalization.
Theorem 2.4.5 [10]A fuzzy choice function C defined on base domain is full
rational if and only if C satisfies the fuzzy T-congruence axiom.
Fuzzy Chernoff axiom: [24] For all 1 , 2 B and x X,
The following theorem was presented by Georgescu in [22] for general domains.
As shown in [17], the result also holds for fuzzy choice functions defined on a
base domain.
i i
i i
i i
K23798 2015/2/2
i i
Chaudhari [17]. We show that the weak form of fuzzy T-congruence axiom
does not imply the full rationality of the fuzzy choice function.
Weak fuzzy T-congruence axiom: For any x, y, z X and B,
The weak fuzzy T-congruence axiom and the fuzzy T-congruence axiom
coincide if the domain of fuzzy choice function contains only characteristic
functions of single and two-element subsets of the universal set.
Example 2.4.8 Let X = {x, y, z, w} and B = {1{x} , 1{y} , 1{z} , 1{w} , 1{x,y} ,
1{x,z} , 1{x,w} , 1{y,z} , 1{y,w} , 1{z,w} , 1X }. Let r, s be such that 0 < r < s < 1.
Define a fuzzy choice function C on B as follows: C(1{v} )(v) = 1 for all
v X, C(1{x,y} )(x) = 1, C(1{x,y} )(y) = r, C(1{x,z} )(x) = 1, C(1{x,z} )(z) =
r, C(1{y,z} )(y) = 1, C(1{y,z} )(z) = r, C(1{y,w} )(y) = r, C(1{y,w} )(w) = 1,
C(1{z,w} )(z) = r, C(1{z,w} )(w) = 1, C(1{x,w} )(x) = 1, C(1{x,w} )(w) = s,
C(1X )(x) = 1, C(1X )(y) = s, C(1X )(z) = r and C(1X )(w) = s. Then the
fuzzy revealed preference relation and are as follows:
x y z w x y z w
x 1 1 1 1 x 1 1 1 1
= y
s 1 1 s and =
y
r 1 1 r
z r r 1 r z r r 1 r
w s 1 1 1 w s 1 1 1
Clearly, C satisfies the weak fuzzy T-congruence axiom, but for 1{y,w} , we
have (y, x) (x, w) C(1y,w )(w) 1{y,w} (y) = s and C(1{y,w} )(y) = r.
Hence C does not satisfy the fuzzy T-congruence axiom.
i i
i i
i i
K23798 2015/2/2
i i
Example 2.4.8 also shows the weak fuzzy T-congruence axiom does not
imply the full rationality of the fuzzy choice function by Theorem 2.4.5.
In the following theorem, we characterize the full rationality of a fuzzy
choice function by combining the weak fuzzy T-congruence axiom and the
fuzzy Chernoff axiom. First we show that the weak fuzzy T-congruence axiom
and the fuzzy Chernoff axioms are independent.
In Example 2.4.8, the fuzzy choice function C satisfies the weak fuzzy T-
congruence axiom, but for 1{x,y} and 1X we have, I(1{x,y} , 1X ) 1{x,y} (y)
C(1X )(y) = s and I(1{x,y} C(1X ), C(1{x,y} )) = r. Hence C does not satisfy
fuzzy the Chernoff axiom.
Example 2.4.9 Let X = {x, y, z} and B = {1{x} , 1{y} , 1{z} , 1{x,y} , 1{x,z} ,
1{y,z} }. Let r, s be such that 0 < r < s < 1. Define a fuzzy choice function C on
B as follows: C(1{v} )(v) = 1 for all v X, C(1{x,y} )(x) = 1, C(1{x,y} )(y) =
r, C(1{x,z} )(x) = r, C(1{x,z} )(z) = 1, C(1{y,z} )(y) = 1, and C(1{y,z} )(z) = s.
Then is given by
x y z
x 1 1 r
=
y r 1 1
z 1 s 1
Clearly, C satisfies the fuzzy Chernoff axiom. However, for x, y X, we have
(x, y) (y, z) C(1{x,z} )(z) 1{x,z} (x) = 1 and C(1{x,z} )(x) = r. Thus C
does not satisfy the weak fuzzy T-congruence axiom.
Theorem 2.4.10 (Desai and Chaudhari [17]) Let be a fuzzy preference re-
lation on X. Let C be a fuzzy choice function defined on a base domain such
that C() = G(, ) for all B. Then is max-min transitive if and only
if C satisfies the weak fuzzy T-congruence axiom.
Proof. Let be a fuzzy preference relation on X such that C() = G(, )
for all B. Then is reflexive and for all x, y X, it follows that
C(1{x,y} )(x) = (x, y). For all x, y, z X and B, we have
i i
i i
i i
K23798 2015/2/2
i i
Theorem 2.4.11 (Desai and Chaudhari [17]) A fuzzy choice function C de-
fined on the base domain B is full rational if and only if it satisfies the weak
fuzzy T-congruence axiom and the fuzzy Chernoff axiom.
Proof. Suppose C satisfies the weak fuzzy T-congruence axiom and the
fuzzy Chernoff axiom. Then by Theorems 2.4.5 and 2.4.6, C is full rational.
Conversely, suppose C is full rational with rationalization . Then by Theo-
rem 2.4.5 and Lemma 2.4.2(1), C satisfies the weak fuzzy T-congruence axiom.
Let , B and x X. By Lemma 2.4.1(2), we have for all y, z X that
I(, ) (x) C()(x) (y) C()(y) (z)
(z) I(, ) C()(y)
(z) [(z) (z)] C()(y)
= (z) (z) C()(y)
(z) C()(y)
= (z) G(, )(y)
(z) [(z) (y, z)]
= (z) (y, z)
(y, z).
i i
i i
i i
K23798 2015/2/2
i i
Theorem 2.4.13 Let C be any fuzzy choice function defined on the base do-
main. Then the following statements are equivalent
(1) C is full rational.
(2) C satisfies fuzzy T-congruence axiom.
(3) C satisfies the weak fuzzy T-congruence axiom and fuzzy Chernoff ax-
iom.
(4) C satisfies the fuzzy Chernoff axiom and C is G-rational with transitive
rationalization.
i i
i i
i i
K23798 2015/2/2
i i
Chaudhari and Desai [10, 11] have studied full and acyclic rationality,
G-rationality, G-normality by introducing various congruence axioms and re-
vealed preference axioms.
i i
i i
i i
K23798 2015/2/2
i i
a b = (a b) (b a),
for all a, b [0, 1]. The corresponding negation operation has the following
form (
1 if a = 0
a = a 0 =
0 if a > 0
for all a, b [0, 1].
Lemma 2.5.1 [8, 30, 31] For any a, b, c [0, 1] the following properties hold:
(1) (a b) (b c) a c;
(2) a b = c a c b and b c a c;
(3) a a = 0;
(4) a b a b = 0
i i
i i
i i
K23798 2015/2/2
i i
Definition 2.5.2 [21, 23, 25] Let C : B F(X) be a fuzzy choice function.
Then C is said to be
(1) G-rational if there exists a fuzzy preference relation on X such that
C() = G(, ) for all B; M -rational if there exists a fuzzy preference
relation on X such that C() = M (, )) for all B;
(2) G-normal if C() = G(, ) for all B; M -normal if C() =
M (, ) for all B;
(3) full rational if there exists a reflexive, strongly connected and max-
min transitive fuzzy preference relation on X such that C() = G(, ) for
all B.
If a fuzzy choice function C is G-rational with rationalization , then
.
The fuzzy path independence property and the fuzzy Condorcet prop-
erty are needed to characterize quasi-transitive rationality of fuzzy choice
functions.
We first discuss the Condorcet property briefly. Approval voting allows a
voter to vote for as many candidates as he wishes in a multicandidate race.
Condorcet argued that the winner should be the candidate who is preferred
by a simple majority of voters to each of the other candidates in pairwise
contests, assuming such a candidate exists. In the following definition, assume
is crisp. If z belongs to and x is chosen over z, then x is chosen with
respect to . Recall that a fuzzy choice function C : B F(X) is said to
satisfy fuzzy path independence if C( ) = C(C() C()) for all , B.
Lemma 2.5.5 Let C be a fuzzy choice function. If C satisfies the fuzzy path
independence property, then is quasi-transitive.
Proof. We shall prove that for all x, y, z X,
i i
i i
i i
K23798 2015/2/2
i i
Hence
(x, y) (y, z) (x, z) (2.3)
Now
By the idempotent property of the Godel t-norm and inequalities (2.3) and
(2.4), inequality (2.2) holds.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let C satisfy the fuzzy Condorcet property. Then G(, ) C()
for all B. Since , by Lemma 2.5.1(3), it follows that
Hence G(, ) G(, ) for all B. However, C() G(, ) for all
B. Thus C() G(, ). Hence C() = G(, ) for all B. Therefore,
C is G-rational.
Conversely, suppose C() = G(, ) for some fuzzy preference relation
on X. Then for all B and x X, we have by Lemmas 2.3.1(2) and 2.5.5
that
Example 2.5.7 Let X = {x, y, z} and B = {1{x} , 1{y} , 1{z} , 1{x,y} , 1{x,z} ,
1{y,z} , 1X }. Define a fuzzy choice function C on B as follows: C(1{v} )(v) = 1
for all v X; C(1{x,y} )(x) = 1; C(1{x,y} )(y) = 1; C(1{x,z} )(x) = 1;
C(1{x,z} )(z) = 1; C(1{y,z} )(y) = 1; C(1{y,z} )(z) = 1; C(1X )(x) = 0.5;
C(1X )(y) = 1 and C(1X )(z) = 1. Then (u, v) = 1 for all u, v X. Hence C
satisfies the fuzzy path independence property. However, for 1X B, we have
1X (x) ({1X (z) (x, z) | z X}) = 1 and C(1X )(x) = 0.5. Thus C
does not satisfy the fuzzy Condorcet property.
i i
i i
i i
K23798 2015/2/2
i i
Example 2.5.8 Let X = {x, y, z} and B = {1{x} , 1{y} , 1{z} , 1{x,y} , 1{x,z} ,
1{y,z} , 1X }. Define a fuzzy choice function C on B as follows: C(1{v} )(v) = 1
for all v X, C(1{x,y} )(x) = 1, C(1{x,y} )(y) = 0, C(1{x,z} )(x) = 1,
C(1{x,z} )(z) = 1, C(1{y,z} )(y) = 1, C(1{y,z} )(z) = 0, C(1X )(x) = 1,
C(1X )(y) = 0.5 and C(1X )(z) = 0. Then the fuzzy revealed preference re-
lation is given as follows: (y, x) = 0 = (z, y) and (u, v) = 1
otherwise. Hence C satisfies the fuzzy Condorcet property. However, for
1{x,y} , 1{x,z} B, we have C(1{x,y} 1{y,z} )(y) = C(1X )(y) = 0.5 and
C(C(1{x,y} ) C(1{y,z} ))(y) = C(1{x,y} )(y) = 0. Thus C does not satisfy
the fuzzy path independence property.
The following is a characterization theorem for quasi-transitive rationality
of fuzzy choice functions.
Hence
i i
i i
i i
K23798 2015/2/2
i i
Thus
I(, ) I( C(), C()). (2.6)
Since and , we have that I(, ) = 1 and I(, ) = 1.
Hence by the above inequality, it follows that
C( )(x) I(, )
I( C( ), C())
(x) C( )(x) C()(x).
Therefore,
C( ) C(C() C()). (2.7)
Now for any , B and x X, we have
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Theorem 2.6.7 (Chaudhari and Desai [11]) Let C be a fuzzy choice func-
tion. If C is G-rational with max-transitive rationalization, then C satisfies
WFCA.
Proof. Let B and x, y X. Then
We are now able to establish connections between the above axioms and
connections between G-normal and FDRA, FTCCA,WFCA, and FCA.
Theorem 2.6.9 (Chaudhari and Desai [11]) Let C be a fuzzy choice function.
If C satisfies WFCA, then C satisfies FDRA.
Proof. Let B and x X. Since C is normal, there exists y X such
that C()(y) = 1. Hence {(z) (x, z) | z X} (y) (x, y) by
Lemma 2.4.1(3), we have that {(z) (x, z) | z X} (x, y). Thus
i i
i i
i i
K23798 2015/2/2
i i
Proof. Since ( )tc , (y) (x, y) (y) ( )tc (x, y) for all
B and y X. Thus for all x X,
(x) ({(y) (x, y) | y X})
(x) ({(y) ( )tc (x, y) | y X})
C()(x).
Hence G(, )(x) C()(x) for all B and x X. Thus G(, ) C().
Hence C is G-normal.
Theorem 2.6.12 (Chaudhari and Desai [11]) Let C be a fuzzy choice func-
tion. Then C is G-rational with max- transitive rationalization if and only if
C satisfies FTCCA.
Proof. Suppose that C is G-rational with max- transitive rationalization.
Then by Lemma 2.6.11, ( )tc , where is a fuzzy preference relation on X.
By Lemma 2.5.1(6), we have for all y X that (y) ( )tc (x, y) (y)
(x, y). Thus {(y) ( )tc (x, y) | y X} {(y) (x, y) | y X}.
Hence
(x) ({(y) ( )tc (x, y) | y X})
(x) ({(y) (x, y) | y X})
= G(, )(x)
= C()(x)
since C is full rational.
Conversely, suppose C satisfies FTCCA. Let = ( )tc . Then is max-
transitive. Let B and x X. By the definition of G(, ), it follows that
G(, )(x) = (x) ({(y) (x, y) | y X})
= (x) ({(y) ( )tc (x, y) | y X})
C()(x).
i i
i i
i i
K23798 2015/2/2
i i
Theorem 2.6.14 (Chaudhari and Desai [11]) Let C be a fuzzy choice func-
tion. If C satisfies FDRA and is max- transitive, then C satisfies WFCA.
Theorem 2.6.17 (Chaudhari and Desai [11]) Let C be a fuzzy choice func-
tion. Then C satisfies FTCCA if and only if C satisfies FCA.
i i
i i
i i
K23798 2015/2/2
i i
2.7. Exercises 65
2.7 Exercises
Let be a fuzzy preference relation on X. Let tr() denote the transitive
closure of . Define the consistent closure of , written b, by for all x, y
X, b(x, y) = (x, y)(b e, and be defined as in Chapter
(x, y)(y, x). Let ,
2. Define the fuzzy relation on X by (x, y) = (x, y)(y, x) for all x, y X.
Let WFCA, SFCA, FAA,WAFRP and FDRA, FTCCA, FCCCA, FICA be
defined as in Chapter 2.
1. [13] Let C be a fuzzy choice function. Prove that FTCCA FICA
FCCCA FDRA .
2. [13] Let C be a fuzzy choice function. Prove that the following statements
hold:
(i) FTCCA SFCA;
(ii) FICA WFCA;
(iii) WFCA FDRA;
(iv) FICA WAFRP;
(v) SFCA FICA.
i i
i i
i i
K23798 2015/2/2
i i
3. [13] Let C be a fuzzy choice function such that B is closed under intersec-
tion. Prove that the fuzzy Arrow axiom and the weak fuzzy congruence
axiom are equivalent for = .
4. Let be a continuous t-norm without zero divisors. Let be a complete
and -transitive fuzzy preference relation on a finite set X and (X, B)
a fuzzy choice space. Prove that G( , ) is a fuzzy choice function on
(X, B).
Let be a t-norm. Let complete and strongly complete be defined as in
this book.
5. [35] Let be fuzzy preference relation on X. If is acyclic, prove that
it is -acyclic for any t-norm that is left continuous or that has no zero
divisors.
2.8 References
1. K. J. Arrow, Rational choice functions and ordering, Economica,
26(1959) 121127.
2. D. Austen-Smith and J. S. Banks, Positive Political Theory I: Collective
Preference, The University of Michigan Press 2000.
3. D. Austen-Smith and J. S. Banks, Positive Political Theory II: Strategy
and Structure, The University of Michigan Press 2005.
4. A. Banerjee, Fuzzy choice functions, revealed preference and rationality,
Fuzzy Sets and Systems, 70 (1995) 3143.
i i
i i
i i
K23798 2015/2/2
i i
2.8. References 67
8. R. Belohl
avek, Fuzzy relational systems. Foundations and principles,
(Kluwer, 2002).
9. W. Bossert, Y. Srumont, and K. Suzumura, Rationality of choice func-
tions on general domains without full rationality, Social Choice and Wel-
fare, 27 (2006) 435458.
10. S. R. Chaudhari and S. S. Desai, Transitive and acyclic rationality of
fuzzy choice functions, International Journal of Mathematical Sciences
and Engineering Applications, 1 (2010) 209224.
11. S. R. Chaudhari and S. S. Desai, On full rationality and congruence ax-
ioms of fuzzy choice functions. International Journal of Computational
and Applied Mathematics, 5 (2010) 313324.
12. S. R. Chaudhari and S. S. Desai, Congruence axioms and rationality of
fuzzy choice functions, International Journal of Applied Mathematics, 5
(2010) 843856.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
2.8. References 69
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 3
Factorization of Fuzzy
Preference Relations
71
i i
i i
i i
K23798 2015/2/2
i i
(1) x, y X, (x, y) > 0 and (y, x) = 0 if and only if (x, y) > 0 and
(x, y) = (y, x) = 0.
(2) x, y X, (x, y) > 0 and (y, x) > 0 if and only if (x, y) = (y, x) >
0.
Proof. (1) Suppose (x, y) > 0 and (y, x) = 0. Then (x, y) Supp()
Supp() and (y, x) / Supp() Supp(). Thus (x, y) = (y, x) = 0 and so
(x, y) > 0. Conversely, suppose (x, y) > 0 and (x, y) = (y, x) = 0. Then
(x, y) Supp() by (i) and since (y, x) = 0, (y, x) = 0 by (i).
(2) Suppose (x, y) > 0 and (y, x) > 0. Then (x, y), (y, x) Supp()
Supp(). Since not both (x, y) > 0 and (y, x) > 0, either (x, y) or (y, x) is
in Supp() and so both (x, y), (y, x) Supp(). Conversely, suppose (x, y) =
(y, x) > 0. Then by (i), (x, y), (y, x) Supp().
Since Supp( ) = Supp()Supp(), it follows that = implies
Supp() = Supp()Supp(). Hence Propositions 3.1.1 and 3.1.2 hold for
factorization results to follow.
i i
i i
i i
K23798 2015/2/2
i i
Definition 3.1.3 Let : [0, 1] [0, 1] [0, 1]. Then is called a t-conorm
or simply a conorm if a, b, c [0, 1],
(1) (a, 0) = a (boundary condition);
(2) b c implies (a, b) (a, c) (monotonicity);
(3) (a, b) = (b, a) (commutativity);
(4) (a, (b, c)) = ((a, b), c) (associativity).
Let , , FR(X). Suppose is a conorm. When we write, = ,
we mean x, y X, (x, y) = (x, y) (x, y).
i i
i i
i i
K23798 2015/2/2
i i
(x, y) + (x, y)
= 0 ((x, y) (y, x)) + (x, y) (y, x)
(x, y) (y, x) + (y, x) = (x, y) if (x, y) > (y, x)
=
0 + (x, y) = (x, y) if (x, y) (y, x)
1.
The next factorization was given by Ovchinnikov [23] and was charac-
terized by Dutta [11]. We replaced the assumption of connectedness with
= max .
i i
i i
i i
K23798 2015/2/2
i i
Example 3.1.10 Let X = {x, y, z}. Let be the fuzzy relation on X defined
as follows: (x, x) = (y, y) = (z, z) = 1, (x, y) = 1, (y, x) = 0, (x, z) =
(z, x) = (y, z) = (y, z) = 1/2. Let and be as in Propositions 3.1.4 and
3.1.9, respectively. Then (x, y) = 1 and (u, v) = 0 (u, v) X X\{(x, y)}
and (x, y) = (y, x) = 0, (u, v) = 1/2 (u, v) X X\{(x, y), (y, x)}. It is
easily seen that is strongly complete and that Supp() Supp() = .
i i
i i
i i
K23798 2015/2/2
i i
1 = (x, y) = (x, y) + (x, y) = (x, y) + (y, x). Thus (x, y) = 1 (y, x).
3.2 Quasi-Subtraction
Due to the nature of the results to follow, we often use the notation for a
t-norm rather than . Let FR(X). Recall that is called a fuzzy weak
preference relation (FWPR) if is reflexive and complete.
We call = i, where is an asymmetric fuzzy relation and is
a symmetric fuzzy relation, a factorization of . We will see that under
very mild assumptions, a factorization of into i is such that (x, y) =
(x, y)(y, x) and that will be of a known type depending on the definition
of . Consequently, a better understanding of is determined.
Let be a continuous t-conorm. The quasi-subtraction of , denoted
, is the function from [0, 1]2 into [0, 1] defined by a, b [0, 1],
a b = {t [0, 1] | a t b}.
has a solution for t. The real number (y, x) (x, y) is the smallest solution.
(2) If = max or is a strict t-conorm, then (y, x) (x, y) is the
unique solution.
Proof. (1) Let g : [0, 1] [0, 1] be such that g is continuous, g(0) =
(y, x) 0 = (y, x) and g(1) = (y, x) 1 = 1. Then there exists
t0 [(y, x), 1] such that g(t0 ) = (x, y) by the Intermediate Value The-
orem. Hence Eq. (3.1) has a solution, namely t0 . Let t = {t [0, 1] |
(y, x) t (x, y)} = (y, x) (x, y). Then (y, x) t (x, y) and
(y, x) t0 = (x, y), where t0 is any solution to Eq. (3.1). Thus t t0 . That
is, (y, x) (x, y) is the smallest solution.
(2) Suppose = max. Then (y, x) (x, y) = {t [0, 1] | (y, x) t
(x, y)} = (x, y). Eq. (3.1) becomes (x, y) = (y, x) t from which the
uniqueness clearly follows.
Suppose is strict. Then there exists unique t0 such that (y, x) t0 =
(x, y).
The following result is essentially due to Fono and Andjiga [12].
i i
i i
i i
K23798 2015/2/2
i i
3.2. Quasi-Subtraction 77
Proof. Suppose (1) holds. Then statement (i) follows from Proposition
3.1.4 and statement (ii) follows from Proposition 3.1.5. For (iii), suppose
(x, y) > (y, x). Then (x, y) = (y, x). By hypothesis, (x, y) = (x, y)
(y, x). Thus (x, y) is a solution to Eq. (3.11).
Suppose (2) holds. By (i), is symmetric. By (ii), is asymmetric.
((x, y) > 0 (x, y) > (y, x) (y, x) = 0.) Suppose (x, y) = (y, x).
Then (x, y) = 0 = (y, x) by (ii). Thus is simple.
Suppose (y, x) (x, y). Then (x, y)(x, y) = 0(x, y) = 0(x, y) =
(x, y) by (i) and (ii).
Suppose (y, x) < (x, y). Then (x, y) is a solution to Eq. (3.1). Hence
(x, y) = (y, x) (x, y) = (x, y) (x, y).
Corollary 3.2.3 Let be FWPR on X and let and be fuzzy binary rela-
tions on X. Let be a strict t-conorm or = max. Then statements (1) and
(2) are equivalent.
(1) x, y X, (x, y) = (x, y) (x, y), is asymmetric, is symmetric,
and is simple.
(2) x, y X, (i) (x, y) = (x, y) (y, x) and (ii) (x, y) = (y, x)
(x, y).
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
3.2. Quasi-Subtraction 79
0 and (y, x) > 0. Since is complete either (x, y) > 0 or (y, x) > 0, say
(x, y) > 0. Thus (y, x) = 0 and so (0) (x, y) = (x, y) and (x, y) = 0. Hence
(x, y) = (0) (x, y) d (x, y) and (y, x) = 0 = 0 d 0 = (0) (y, x) d (y, x).
We present the next result in this particular form because the notation is
needed in later chapters.
Proof. (1) Let x, y X. Suppose (y, x) < (x, y). Suppose that (y, x) >
0. Hence {t [0, 1] | (y, x) d t (x, y)} does not exist since t > 0,
(y, x) d t = 1 and for t = 0, (y, x) d 0 = (y, x) < (x, y). Thus {t
[0, 1] | (y, x)d t (x, y)} exists and equals (x, y) if and only if (y, x) = 0.
Hence = (0) .
(2) For all x, y X, (x, y) = {t [0, 1] | (y, x) t (x, y)}
(x, y) if (x, y) > (y, x),
=
0 if (y, x) (x, y).
Hence = (1) .
(3) For all x, y X,
i i
i i
i i
K23798 2015/2/2
i i
(s1a 1)(s1b 1)
s (a, b) = 1 logs 1 + , s > 0, s 6= 1.
s1
i i
i i
i i
K23798 2015/2/2
i i
3.2. Quasi-Subtraction 81
(s 1)(s1(x,y) 1)
(x, y) = 1 logs +1 .
s1(y,x) 1
Proof. (1) Let x, y X. Define as in (1). Suppose (x, y) > (y, x).
Then
(x, y) H2 (x, y)
(x, y) (y, x) (x, y) (y, x)
= + (y, x) / 1 + (y, x)
1 + (x, y)(y, x) 1 + (x, y)(y, x)
(x, y) (y, x) + (y, x) (y, x)(x, y)(y, x)
=
1 (y, x)(x, y)
1 (y, x)(x, y)
1 (y, x)(x, y) (x, y)(y, x) (y, x)(y, x)
(x, y)(1 (y, x)(y, x))
= = (x, y).
1 (y, x)(y, x)
Suppose (x, y) (y, x). Then (x, y) H2 (x, y) = 0 H2 (x, y) = (x, y).
(2) Let x, y X. Define as in (2). Suppose (x, y) > (y, x). Then
(x, y) p (x, y)
1
= 1 [(1 (x, y))p + (1 (y, x))p 1] p
= 1
1 1
{[1 [1 (1 (x, y))p (1 (y, x))p + 1] p ]p + (1 (y, x)p 1} p .
We show this latter expression is (x, y). We have [(1 (x, y))p (1
1 1
(y, x))p + 1] p = 1 (1 [(1 (x, y))p (1 (y, x))p + 1] p and so
i i
i i
i i
K23798 2015/2/2
i i
1
[(1(x, y))p (1(y, x))p +1] = {1(1 [(1(x, y))p (1(y, x))p +1] p }p .
1
Hence (1 (x, y))p = {1 (1 [(1 (x, y))p (1 (y, x))p + 1] p }p + (1
(y, x))p 1. Solving this latter equation for (x, y) yields the desired result.
Suppose (x, y) (y, x). Then (x, y) p (x, y) = 0 p (x, y) = (x, y).
(3) Let x, y X. Define as in (3). Suppose (x, y) > (y, x). Then
(x, y) (x, y)
1 1
= {[((x, y) (y, x) ) ] + (y, x) }
= (x, y).
Suppose (x, y) (y, x). Then (x, y) (x, y) = 0 (x, y) = (x, y).
(4) Let x, y X. Define as in (4). Suppose (x, y) > (y, x). Then
(x, y) s (x, y)
(s1(x,y) 1)(s1(y,x) 1)
= 1 logs 1 + .
s1
1(x,y)
1)(s1(y,x) 1)
Let u = (x, y) s (x, y). Then 1 u = logs 1 + (s s1
(s1(x,y) 1)(s1(y,x) 1)
and so s1u = 1 + s1 . Hence s1u 1 =
(s1(x,y) 1)(s1(y,x) 1) 1u 1(y,x)
s1 Thus (s 1)(s
. 1)/(s 1) = s1(x,y) 1.
1u 1(y,x) 1(x,y)
1 + (s 1)(s
Hence
1u
1)/(s 1) = s and so (x, y) = 1
1)
logs 1 + (s1)(s
s1(y,x) 1
. Thus u = (x, y) by the definition of .
Henceforth all conorms are assumed to be continuous.
i i
i i
i i
K23798 2015/2/2
i i
3.2. Quasi-Subtraction 83
2 2 have the property that 1 2 for all such factorizations if and only
if 1 2 .
1 (x, y)1 (x, y) 1 (x, y)2 (x, y) and 2 (x, y)1 (x, y) 2 (x, y)2 (x, y).
Suppose 1 is strict. Then 1 (x, y) 1 (x, y) > 2 (x, y) 1 (x, y). Hence
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
3.3. Factorizations 85
Now (x, y) = (x, y) (x, y) (x, y) 0 = (x, y). Since is regular, the
desired result is now immediate.
3.3 Factorizations
The following list of t-conorms is taken from [16]. Many of the results are
based on [15].
Conorms
Standard union: a b = a b
Bounded sum: a 2 b = 1 (a + b)
a if b = 0
Drastic union: a d b = b if a = 0
1 otherwise
Algebraic sum: a A b = a + b ab
Yager: a Y b = 1 (aw + bw )1/w , w > 0
a b if a + b < 1
Nilpotent maximum: a n b =
1 otherwise
Einstein sum: a H2 b = (a + b)/(1 + ab)
Schweizer and Sklar:
Frank:
(s1a 1)(s1b 1)
a s b = 1 logs 1 + , s > 0, s 6= 1
s1
i i
i i
i i
K23798 2015/2/2
i i
Weber:
ab=1 a+b ab , > 1
1
(x, y) if (x, y) > (y, x) = 0
(x, y) =
0 otherwise.
Proof. Proposition 3.2.7(1).
i i
i i
i i
K23798 2015/2/2
i i
3.3. Factorizations 87
(
(x,y)(y,x)
1(y,x) if (x, y) > (y, x)
(x, y) =
0 otherwise.
Proof. Proposition 3.2.7(5) and Proposition 3.2.4.
[(x, y)w (y, x)w ]1/w if (x, y) > (y, x)
(x, y) =
0 otherwise.
Proof. Proposition 3.2.5 and 3.2.7(6).
(s 1)(s1(x,y) 1)
(x, y) = 1 logs +1 .
s1(y,x) 1
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.3.10 Let denote the Dubois and Prade conorm. Let
FR(X). Let x, y [0, 1]. Suppose (x, y) > (y, x) and (x, y) = (x, y)
(y, x). Then
(x, y) if 1 (y, x)
(x, y) =
1 [(1 (x, y)]/[1 (y, x)] if 1 (y, x).
Thus
or
For case (1), r (y, x)/(x, y). However, (y, x)/(x, y) can be arbitrarily
close to 0. Thus case (1) is impossible since r > 0 and r is fixed.
For case (2), r (y, x)/(x, y). Now (y, x)/(x, y) can be made arbitrar-
ily close to 1. Hence we cant have r < 1. Thus r 1. Let (x, y) = 1. Then
from (2), (y, x) r (r 1)(y, x) 1 (r 2)(y, x). Hence r 1.
Thus r 1. Hence r = 1.
i i
i i
i i
K23798 2015/2/2
i i
3.3. Factorizations 89
Proposition 3.3.12 Let denote the Schweizer and Sklar 2 conorm. Let
FR(X). Let x, y X. Suppose (x, y) > (y, x) and (x, y) = (x, y)(y, x).
Then 1/p
(x, y)p (y, x)p
(x, y) = .
1 (y, x)p
Proof. Suppose there exists t [0, 1] such that (x, y) = t (y, x). Then
p
(y,x)p
(x, y) = (tp + (y, x)p tp (y, x)p )1/p . Thus tp = (x,y)
1(y,x)p and so
1/p
(x, y)p (y, x)p
t= .
1 (y, x)p
Proposition 3.3.13 Let denote the Schweizer and Sklar 4 conorm. Let
FR(X). Let x, y X. Suppose (x, y) > (y, x) and (x, y) = (x, y)(y, x).
Then
1/p
(1 (x, y))p (1 (y, x))p
(x, y) = 1 .
(1 (y, x))p + (1 (x, y))p (1 (y, x))p (1 (x, y))p
Proof. Suppose there exists t [0, 1] such that (x, y) = t (y, x). Then
and so
(1 t)p (1 (y, x))p
(1 (x, y))p = ,
(1 t)p + (1 (y, x))p (1 t)p (1 (y, x))p
i.e.,
Thus
(1 (x, y))p (1 (y, x))p
(1 t)p = .
(1 (y, x))p + (1 (x, y))p (1 (y, x))p (1 (x, y))p
i i
i i
i i
K23798 2015/2/2
i i
Hence
1/p
(1 (x, y))p (1 (y, x))p
t=1 .
(1 (y, x)) + (1 (x, y))p (1 (y, x))p (1 (x, y))p
p
Proposition 3.3.14 Let be the Dombi conorm. Then it is not the case that
x, y X, FR(X) such that (x, y) > (y, x), (x, y) = t (y, x) has
a solution for t.
Proof. Suppose (y, x) > 0. Suppose there exists t (0, 1] such that
1/ 1
" #
1
1
(x, y) = t(y, x). Then (x, y) = 1 + t 1 + (y,x) 1 .
Thus
" #1/
1 1 1
1 = 1 + 1
(x, y) t (y, x)
1 1 1
1 = 1 + 1 .
(x, y) t (y, x)
Hence
1 1 1
1 = 1 1
t (x, y) (y, x)
" #1/
1 1 1
1 = 1 1 .
t (x, y) (y, x)
Thus 1
" #1/
1 1
t = 1 + 1 1 .
(x, y) (y, x)
Now
" #1
1/
1 1
1+ (x,y) 1 (y,x) 1 <0
1/
1 1
1+ (x,y) 1 (y,x) 1 < 0.
This inequality can occur by taking (y, x) sufficiently close to 0. In this case,
t < 0.
i i
i i
i i
K23798 2015/2/2
i i
3.3. Factorizations 91
Lemma 3.3.16 Let denote the Weber conorm. Let FR(X). Let x, y
X. Suppose (x, y) > (y, x) and (x, y) = (x, y) (y, x). Then x, y X,
(x, y) (y, x)
(x, y) =
if 1 < 0
1 1 (y, x)
1 (x, y) + (y, x)
or (0 < < 1 and ) or ( > 1).
1 + 2(y, x) (x, y)
Proof. Suppose there exists t [0,1] such that (x, y) = t (y, x). Then
(x, y) = t + (y, x) 1 t(y, x) = t 1 1 (y, x) + (y, x). Thus
(x, y) (y, x)
t=
.
1 1 (y, x)
Suppose 1 < < 0. Then clearly 1 < 0. Thus t > 0. Also 1 1 (y, x)
(x, y) (y, x) > 0. Hence t < 1.
Now suppose 0 < < 1. Then since 1 > 0, 1 1 (y, x) > 0
1
1 (y, x) > 0 < 1+(y,x) which holds since < 1. Hence t > 0.
Now
t 11 (y, x) (x, y) (y, x)
1
1 (y, x) (1 )((x, y) (y, x))
1 (x, y) + (y, x) + (y, x) (x, y) + (y, x)
1 (x, y) + (y, x) (1 + 2(y, x) (x, y))
1 (x, y) + (y, x)
.
1 + 2(y, x) (x, y)
i i
i i
i i
K23798 2015/2/2
i i
1(x,y)+(y,x)
Now suppose > 1. Then as above t 1 1+2(y,x)(x,y) since
1(x,y)+(y,x)
1 < 0. However, 1+2(y,x)(x,y) holds since > 1. Thus t 1. Now
t 01 (y, x) 0 1 (y, x) 0
1
1
.
1 + (y, x)
1
However, 1+(y,x) since > 1.
(x, y) (y, x)
(x, y) =
if 1 < 0 or > 1.
1 1 (y, x)
1(x,y)+(y,x)
Proof. Suppose 0 < < 1. By Lemma 3.3.16, 1+2(y,x)(x,y) . How-
1(x,y)+(y,x)
ever, 1+2(y,x)(x,y) can be made arbitrarily close to 1 by taking (y, x) suffi-
ciently close to 0. This is impossible since is fixed. The desired result follows
from Lemma 3.3.16.
Proposition 3.3.18 Let denote the Schweizer and Sklar 3 conorm. Let
FR(X). Let x, y X. Suppose (x, y) > (y, x) and (x, y) = (x, y)(y, x).
Then
(
1 if (x, y) = 1
(x, y) = p p 1/p
1 e[(ln( 1(x,y) )) | ln(1(y,x)| ]
1
if (x, y) < 1.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Suppose there exists t [0, 1) such that (x, y) = t (y, x). Then
p
+|(1(y,x)|p )1/p
(x, y) = 1 e(| ln(1t)|
p
+|(1(y,x)|p )1/p
1 (x, y) = e(| ln(1t)|
ln(1 (x, y)) = (| ln(1 t)|p + | ln(1 (y, x)|p )1/p
1
ln = (| ln(1 t)|p + | ln(1 (y, x)|p )1/p
1 (x, y)
p
1
ln = | ln(1 t)|p + | ln(1 (y, x)|p
1 (x, y)
p
p 1
| ln(1 t)| = ln | ln(1 (y, x)|p
1 (x, y)
p 1/p
1
| ln(1 t)| = ln | ln(1 (y, x)|p
1 (x, y)
p 1/p
1
ln(1 t) = ln | ln(1 (y, x)|p
1 (x, y)
p p 1/p
e[(ln( 1(x,y) )) | ln(1(y,x)| ]
1
1t =
p p 1/p
1 e[(ln( 1(x,y) )) | ln(1(y,x)| ] .
1
t =
i i
i i
i i
K23798 2015/2/2
i i
Definition 3.4.1 Let c : [0, 1] [0, 1]. Then c is a called a fuzzy comple-
ment if the following two conditions hold:
(1) c(0) = 1 and c(1) = 0 (boundary conditions).
i i
i i
i i
K23798 2015/2/2
i i
1 2 a, b [0, 1], a 1 b a 2 b
a, b [0, 1], c(a) 1 c(b) c(a) 2 c(b)
a, b [0, 1], c(c(a) 1 c(b)) c(c(a) 2 c(b))
a, b [0, 1], a 1 b a 2 b.
i i
i i
i i
K23798 2015/2/2
i i
Definition 3.5.1 Let T denote the set of all continuous t-norms. The quasi-
division of a t-norm , denoted , is the function : [0, 1]2 [0, 1] defined
by a, b [0, 1], a b = {t [0, 1] | a t b}.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 3.5.4 (Fono, Nana, Salles, and Gwet [13]) Let T and
FR(X). Let x, y X. Suppose (x, y) < (y, x). Then the following statement
holds:
(1) The equation
has a solution for t. The real number (y, x) (x, y) is the largest solution
to (3.2).
(2) If = min or if is a strict t-norm, then (y, x) (x, y) is the
unique solution.
Proof. (1) Let g : [0, 1] [0, 1] be defined by g(t) = t (y, x) t [0, 1].
Then g(0) = 0 and g(1) = (y, x). Since g is continuous, there exists t0 [0, 1]
such that (x, y) = t0 (y, x) by the Intermediate Value Theorem. That is,
Eq. (3.2) has a solution. Let
Then t (y, x) (x, y) and t0 (y, x) = (x, y), where t0 is any solution
to Eq. (3.2). Thus t t0 . Hence (y, x) (x, y) is the largest solution.
(2) Suppose = . Then
i i
i i
i i
K23798 2015/2/2
i i
(2) x, y X,
(i) (x, y) = (x, y) (y, x);
(ii) (x, y) (y, x) (y, x) = 1;
(iii) (x, y) < (y, x) (x, y) is a solution to (3.2).
Proof. Suppose (1) holds.
(i) Since is c-asymmetric, either (x, y) = 1 or (y, x) = 1. Hence either
(x, y) = (x, y) or (y, x) = (y, x). Say (x, y) = 1. Then (x, y) = (x, y) =
(y, x) (y, x), where the inequality holds since (y, x) = (y, x)(y, x) and
(y, x) < 1. Similarly, if (y, x) = 1, then (y, x) = (y, x) = (x, y) (x, y).
In either case, (x, y) = (x, y) (y, x).
(ii) Suppose (x, y) (y, x). If (x, y) = (y, x), then (x, y) = (y, x) =
1 since is simple. Suppose (x, y) < (y, x). Then (x, y)(x, y) = (x, y) <
(y, x) = (y, x) (y, x). Thus (x, y) (y, x). Since is c-asymmetric,
(y, x) = 1. Conversely, suppose (y, x) = 1. Then (y, x) = 1 (y, x) =
(y, x) (x, y) by (i).
(iii) Suppose (x, y) < (y, x). Were given (x, y) = (x, y) (x, y) =
(x, y) (y, x). Hence (x, y) is a solution to Eq. (3.2).
Suppose (2) holds. By (i), is clearly symmetric. Suppose (x, y) = (y, x).
Then by (ii), (x, y) = 1 = (y, x). Thus is simple. Suppose (x, y) < 1.
Then (x, y) 6= (y, x) since is simple. Suppose (x, y) < (y, x). Then
(x, y) is a solution to (3.2) by (iii) and so (x, y) = (x, y) (y, x) =
(x, y)(x, y). By (ii), (y, x) = 1 since is c-asymmetric. The case (x, y)
(y, x) is not possible under the assumption (x, y) < 1 by (ii). The case
(y, x) < 1 is similar. Suppose (x, y) = 1 = (y, x). Then clearly is c-
asymmetric and (x, y) = (y, x) by (ii) and (x, y) = 1(x, y) and (y, x) =
1 (y, x).
The dual result to Proposition 3.5.5 is [15, Proposition 3.2, p. 25] or [12,
Proposition 3, p. 378] and the dual result to Proposition 3.5.6 is [15, Propo-
sition 2.6, p. 20] or [24, Proposition 1.2, p. 364].
i i
i i
i i
K23798 2015/2/2
i i
c (x, y) = c((x, y) (y, x)) = c((x, y)) c((y, x)) = c (x, y) c (y, x).
i i
i i
i i
K23798 2015/2/2
i i
Proof. 1. The proof follows from Proposition 3.5.10 and [15, Theorem
3.13, p. 33].
Proof. 2. Suppose 1 2 for all such factorizations. Suppose there
exists a, b [0, 1] such that a 1 b < a 2 b. Let (x, y) = a 1 b and
(y, x) = b for some x, y X. Then (x, y) = b. Since 1 is the maxi-
mal solution, 1 (x, y) a and thus 2 (x, y) 1 (x, y) a. Therefore,
(x, y) = 2 (x, y) 2 b a 2 b > a 1 b = (x, y), a contradiction. Hence no
such a, b exist. Thus 1 2 .
i i
i i
i i
K23798 2015/2/2
i i
[i (x, y) > 0 (x, y) > 0] [ic (x, y) < 1 c (x, y) < 1].
Proof. Suppose [i (x, y) > 0 (x, y) > 0]. Suppose ic (x, y) < 1.
Then i (x, y) = icc (x, y) > 0. Hence (x, y) > 0. Thus c (x, y) < 1. The
remainder of the proof follows in a similar manner.
i i
i i
i i
K23798 2015/2/2
i i
Example 3.5.18 Let x = {x, y}. Let and be fuzzy binary relations on
X defined as follows: w X, (w, w) = 1 and (w, w) = 0,
1 3 3 1
(x, y) = , (y, x) = , (x, y) = , (y, x) = .
2 8 8 4
Then (u, v) + (u, v) < 1 u, v X. Let = and = . Then =
and = , where (u, v) = (u, v) (v, u) and (u, v) =
(u, v) (v, u) for all u, v X and
1 1
(x, y) = , (y, x) = 0, (x, y) = 1, (y, x) = .
2 4
We have that (x, y) + (x, y) > 1.
i i
i i
i i
K23798 2015/2/2
i i
Norms
Standard intersection: a b = a b
Bounded difference: a 2 b = 0 (a + b 1)
a if b = 1
Drastic intersection: a d b = b if a = 1
0 otherwise
Algebraic product: a A b = ab
w w 1/w
1 ((1 a) + (1 b) )
Yager: a w b = if (1 a) + (1 b)w [0, 1]
w
0 otherwise
(
a b if a + b > 1
Nilpotent: a n b = 0 otherwise
i i
i i
i i
K23798 2015/2/2
i i
Proof. Suppose (x, y) (y, x). Then (x, y) = 0 ((tp + (x, y)p 1)1/p
since (x, y) = (x, y). Thus t = 1. Suppose (x, y) < (y, x). Then (x, y) =
t(y, x) = 0(tp +(y, x)p 1)1/p = (x, y) for t = (1+(x, y)p (y, x)p )1/p .
(Note 2 tp (y, x)p = 2 (1 + (x, y)p (y, x)p ) (y, x)p [0, 1].)
i i
i i
i i
K23798 2015/2/2
i i
1 if (x, y) (y, x)
(x, y) =
1 [(1 (x, y))w (1 (y, x))w ]1/w if 0 < (x, y) < (y, x).
Proof. Let x, y X. Suppose 0 < (x, y) < (y, x). Then (x, y) =
t (y, x) = 1 ((1 t)w + (1 (y, x))w )1/w and so
7
(Note that if (x, y) = 0, (y, x) = 8 and w = 1, then any t [0, 81 ] is a
solution.)
Proof. Suppose (x, y) < (y, x). Then (x, y) = t (y, x) = logs [1 +
(st 1)(s(y,x) 1)/(s1)]. Thus s(x,y) 1 = (st 1)(s(y,x) 1)/(s1). Hence
st 1 = (s 1)(s(x,y) 1)/(s(y,x) 1). Thus t = logs [1 + (s 1)(s(x,y)
1)/(s(y,x) 1)].
Example 3.5.25 Let = n . Then there exists FR(X) such that does
not factor: Let X = {x, y} and let be such that (x, y) = 41 and (y, x) = 12 .
Then 41 = t 12 6= t 12 if t + 21 > 1 and 14 = t 12 6= 0 otherwise.
i i
i i
i i
K23798 2015/2/2
i i
(x, y) if (x, y) = 1,
(x, y) =
0 if (x, y) < 1.
Proof. If (x, y) < 1, then (x, y) < 1 and since (x, y) = (x, y)(y, x),
we have (x, y) = 0 and (x, y) = 0. If (y, x) = 1, then the desired result is
immediate.
In the remainder of the chapter, factorization results for are obtained
from known factorization results for and the notion of duality involving an
involutive fuzzy complement.
[(x, y)p (y, x)p + 1]1/p if (x, y) < (y, x),
(x, y) =
1 otherwise.
Thus
1 [1 [(1 (x, y))p (1 (y, x))p + 1]1/p ] if (x, y) > (y, x)
c (x, y) =
1 otherwise.
i i
i i
i i
K23798 2015/2/2
i i
Hence
[(1 (x, y))p (1 (y, x))p + 1]1/p if (x, y) > (y, x)
c (x, y) =
1 otherwise.
Thus
c [c (x, y)p c (y, x)p + 1]1/p if c (x, y) < c (y, x)
(x, y) =
1 otherwise.
Proposition 3.5.31 Let denote the Dubois and Prade norm. Let
FR(X). Suppose (x, y) > (y, x) and (x, y) = (x, y) (y, x). Then
x, y X,
(x, y) if (y, x)
(x, y) =
((x, y)/(y, x)) if (y, x).
Proof. For the Dubois and Prade conorm with (x, y) < (y, x) and
(x, y) = (x, y) (y, x), we have
(x, y) if 1 (y, x)
(x, y) =
1 [(1 (x, y)]/[1 (y, x)] if 1 (y, x).
Thus
c 1 (x, y) if 1 (y, x)
(x, y) =
1 [1 [(1 (x, y))]/[1 (y, x)] if 1 (y, x).
Hence
c (x, y) if c (y, x)
c (x, y) =
(c (x, y))/c (y, x) if c (y, x).
Proposition 3.5.32 Let denote the Schweizer and Sklar 2 norm. Let
FR(X). Suppose (x, y) < (y, x) and (x, y) = (x, y)(y, x). Then x, y
X,
1/p
(1 (x, y))p (1 (y, x))p
(x, y) = 1 .
1 (1 (y, x))p
Proof. For the Schweizer and Sklar 2 conorm with F R(X),
(x, y) > (y, x), and (x, y) = (x, y) (y, x) we have x, y X,
1/p
(x, y)p (y, x)p
(x, y) = .
1 (y, x)p
Thus 1/p
(x, y)p (y, x)p
c (x, y) = 1 .
1 (y, x)p
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.5.33 Let denote the Schweizer and Sklar 4 conorm. Let
FR(X). Suppose (x, y) > (y, x) and (x, y) = (x, y) (y, x). Then
x, y X,
1/p
(x, y)p (y, x)p
(x, y) = .
(y, x)p + (x, y)p (y, x)p (x, y)p
Thus 1/p
c (x, y)p c (y, x)p
c
(x, y) = c
(y, x)p + c (x, y)p c (y, x)p c (x, y)p
for c (x, y) < c (y, x).
Proposition 3.5.34 Let denote the Schweizer and Sklar 3 conorm. Let
FR(X). Suppose (x, y) > (y, x) and (x, y) = (x, y) (y, x). Then
x, y X,
(
0 if (x, y) = 0
(x, y) = p p 1/p
1 e[(ln( (x,y) )) | ln((y,x)| ] if (x, y) > 0.
1
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.5.36 Let be the Dombi norm. Then it is not the case that
x, y X, FR(X) such that (x, y) < (y, x), (x, y) = t (y, x) has a
solution for t.
Lemma 3.5.38 Let denote the Weber norm. Let FR(X). Let x, y X.
Suppose (x, y) < (y, x) and (x, y) = (x, y) (y, x). Then
(y, x) (x, y)
(x, y) = 1
if 1 < 0
1 1 (1 (y, x))
1 + (x, y) (y, x)
or (0 < < 1and ) or ( > 1).
1 + 2(1 (y, x)) (1 (x, y))
Proof. For the Weber conorm , we have with FR(X), (x, y) >
(y, x), and (x, y) = (x, y) (y, x) and x, y X,
(x, y) (y, x)
(x, y) =
if 1 < 0
1 1 (y, x)
1 (x, y) + (y, x)
or (0 < < 1 and ) or ( > 1).
1 + 2(y, x) (x, y)
i i
i i
i i
K23798 2015/2/2
i i
Hence
(x, y) (y, x)
c (x, y) = 1
if 1 < 0
1 1 (y, x)
1 (x, y) + (y, x)
or (0 < < 1 and ) or ( > 1).
1 + 2(y, x) (x, y)
Thus
(1 c (x, y)) (1 c (y, x))
c (x, y) = 1
if 1 < 0
1 1 (1 c (y, x))
1 (1 c (x, y)) + (1 c (y, x))
or (0 < < 1 and ) or ( > 1).
1 + 2(1 c (y, x)) (1 c (x, y))
Proposition 3.5.39 Let denote the Weber norm. Let FR(X). Let
x, y X. Suppose (x, y) < (y, x) and (x, y) = (x, y) (y, x). Then
(y, x) (x, y)
(x, y) = 1
if 1 < 0 or > 1.
1 1 (1 (y, x))
and (
1 if (x, y) (y, x)
(x, y) = (x,y)
(y,x) if (x, y) < (y, x).
i i
i i
i i
K23798 2015/2/2
i i
Now let (x, y) = 1/4, (y, x) = 1/8, (x, y) = 1/4, and (y, x) = 3/4.
Then (x, y) + (x, y) 1 and (y, x) + (y, x) 1 and
3.6 Exercises
1. [15] Let F R(X). Let h : X 2 [0, 1]2 be defined by x, y X,
h (x, y) = ((x, y), (y, x)) and let (x, y) = (x, y) (y, x). Let U =
{ | is a continuous strict t-conorm. Let P = { | : [0, 1]2 [0, 1]}.
Prove that there exists an injective function f : P U defined by
f () = such that (x, y) = (h (x, y) (x, y)) x, y X. Conclude
that = h is a factorization of .
2. [17] Let , , and be fuzzy binary relations on X. Suppose is reflexive
and strongly complete. Then (, ) is called an axiomatic factorization of
if (1) is irreflexive and asymmetric, (2) is reflexive and symmetric,
(3) = , and (4) = . Prove that if (, ) is an axiomatic
factorization of and is a continuous t-norm without zero divisors,
then is exact.
Let T be a t-norm and S the dual t-conorm of T, i.e. S(x, y) = 1 T (1
x), 1 y) for all x [0, 1]. A function : [0, 1] [0, 1] is called an order
isomorphism if it is bijective and increasing. An order isomorphism is
called reciprocal if (1 x) = 1 (x). The -transform of t is a t-norm
T defined by T (x, y) = 1 (T ((x), (y)). For example W (x, y) =
1 (0 ((x) + (y) 1)) and its dual t-conorm is W 0 (x, y) = 1
1 (0 ((1 x) + (1 y) 1)), where W is the Lukasiewicz t-norm.
3. [17] Let T be a continuous non-Archimedian t-norm with zero divisors
and S its dual conorm. Prove that there exists s (0, 1) and an order
isomorphism such that x, y [0, 1],
x = 1 or
S(x, y) = 1 y = 1 or
1y
(1 x, 1 y) (0, s)2 and ( 1x
s ) + ( s ) 1.
i i
i i
i i
K23798 2015/2/2
i i
and
((x, y)) + ((y, x)) 1)
1 ( ) (x, y) (x, y) (y, x).
2
(x, y) = 1 (y, x)
(x, y) = 1 (((x, y)) + ((y, x)) 1)
for all x, y X.
3.7 References
1. K. T. Atanassov, Intuitionistic fuzzy sets, Fuzzy Sets and Systems, 20
(1986) 8796.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 4
115
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
0 (x, y) 0 (y, x)
n
(wi 0i (x, y) wi 0i (y, x))
P
= Pi=1n
= (w (x, y) wi i (y, x)) + wj 0j (x, y) wj 0j (y, x)
Pni=1,i6=j i i
= i=1 (wi i (x, y) wi i (y, x)) wj (j (x, y) j (y, x))
+wj (0j (x, y) 0j (y, x))
= wj (j (x, y) j (y, x)) + wj (0j (x, y) 0j (y, x))
> 0,
where the inequality holds if either (1) or (2) hold. Hence fe satisfies PR with
respect to (3) . The desired result for (1) follows from Corollary 4.1.4.
(x, y) = 1/31 (x, y) + 2/32 (x, y) = 1/3 1/2 + 2/3 3/4 = 2/3
and
i i
i i
i i
K23798 2015/2/2
i i
and
20 (y, x) = 1 02 (x, y) = 1 1 = 0.
Thus [2 (y, x) > 0 and 20 (y, x) = 0] ; (2)
0
(x, y) > 0.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Theorem 4.1.15 Let strict preference be defined by (3) . Then there exists a
nondictatorial fe : H2n H2 satisfying IIA1, PR, and PC.
If the fuzzy binary relations and 0 in Definition 4.1.16 are reflexive, then
sk = 1 = tm .
Proof. Suppose (x, y) > (y, x). Let (x, y) = si . Then si > (y, x).
Hence (y, x) / 0ti . Since (x, y) si , we have (x, y) 0ti .
/ si . Thus (y, x)
Hence 0 (x, y) ti > 0 (y, x).
For x, y X, let x,y = {(x, x), (y, y), (x, y), (y, x)}.
In the following, we let tPdenote a function from T n into (0, 1]. Some
n
examples
Pn of t are t() = { i=1 wi i (x, y) P
| x, y X, x 6= y} and t() =
n
{ i=1 wi i (x, y) > 0 | x, y X}, where i=1 wi = 1 and wi > 0, i =
1, . . . , n.
Pn
where i=1 wi = 1 and wi > 0, i = 1, . . . , n.
i i
i i
i i
K23798 2015/2/2
i i
and
(x, z) < (x, y) (y, z). (4.2)
Then
0 < (x, y) = (x, y) > (y, x) (4.3)
and
0 < (y, z) = (y, z) > (z, y) (4.4)
By (4.3) and (4.4), (x, y) > 0 and (y, z) > 0 and so by (4.1) (x, z) > 0.
Suppose 0 < (x, z). Then (x, z) > (z, x). Thus (x, z) = (x, z) and so
(x, z) = (x, z) (x, y)(y, z) = (x, y)(y, z), where the latter equality
holds by (4.3) and (4.4). Since this contradicts (4.2), (x, z) = 0. Hence
Case 1: Suppose
(x, y) (y, z) = (x, y). (4.6)
i i
i i
i i
K23798 2015/2/2
i i
Then
(x, z) (x, y). (4.7)
By max-min transitivity, (y, x) (y, z)(z, x) (x, y)(x, z) = (x, y).
However, this contradicts (4.3).
Case 2: Suppose
(x, y) (y, z) = (y, z). (4.8)
Then by (4.1 ),
(x, z) (y, z). (4.9)
By max-min transitivity, either (z, y) (z, x) or (z, y) (x, y), If
(z, y) (z, x), then (z, y) (z, x) (x, z) (y, z) by (4.5) and (4.9).
However, this contradicts (4.4). If (z, y) (x, y), then (z, y) (x, y)
(y, z) by (4.8 ). However, this also contradicts (4.4). Thus is max-min
quasi-transitive.
Example 4.1.24 Let X = {x, y, z}. Define the fuzzy binary relation on X
as follows:
i i
i i
i i
K23798 2015/2/2
i i
Then
Also, a, b X,
i i
i i
i i
K23798 2015/2/2
i i
Theorem 4.2.1 Let fe be a fuzzy aggregation rule. Then the following condi-
tions are equivalent:
(1) fe is neutral;
(2) there exists a unique function fn : [0, 1]n [0, 1] such that x, y X,
= (1 , . . . , n ) FRn , fn ((1 (x, y), . . . , n (x, y)) = fe()(x, y).
Proof. Suppose condition (1) holds. Let x, y X. Define fn : [0, 1]n
[0, 1] as follows: Let (a1 , . . . , an ) [0, 1]n . Then there exists = (1 , . . . , n )
FRn such that i (x, y) = ai , i = 1, . . . , n. Define fn by fn ((a1 , . . . , an )) =
fe()(x, y). It remains to be shown that fn is single-valued. Let w, z X.
Then there exists = (1 , . . . , n ) F Rn such that i (w, z) = ai , i =
1, . . . , n. Thus i (x, y) = i (w, z), i = 1, . . . , n. Since fe is neutral, fn ()(x, y) =
fn ()(w, z). Hence fn is single-valued. The uniqueness of fn follows by its
construction.
Suppose condition (2) holds. Let , FRn and x, y, w, z X. Suppose
i (x, y) = i (w, z), i = 1, . . . , n. Then fe()(x, y) = fn ((1 (x, y), . . . , n (x, y)))
= fn ((1 (w, z)) , . . . , n (w, z))) = fn ()(w, z). Thus fe is neutral.
We call fn the auxiliary function associated with fe.
Definition 4.2.2 Let fe be a fuzzy aggregation rule and let fn be the auxiliary
function associated with fe. Then fe is said to be linearly decomposable if
(a1 , . . . , an ) [0, 1]n ,
Note:
fn ((a1 , . . . , an )) = a1 fn ((1, 0, . . . , 0)) + . . . + an fn ((0, . . . , 0, 1))
fn ((a1 , . . . , an )) = fn ((a1 , 0, . . . , 0) + . . . + fn ((0, . . . , 0, an ))
and
fi ((0, . . . , 0, ai , 0, . . . , 0)) = ai fi (0, . . . , 0, 1, 0, . . . , 0) for i = 1, . . . , n.
Proposition 4.2.3 Let fe be a fuzzy aggregation rule and let fn be the auxil-
iary function associated with fe. If fe is linearly decomposable, then fe is addi-
tive.
Proof. Let (a1 , . . . , an ), (b1 , . . . , bn ) [0, 1]n be such that ai + bi [0, 1],
i = 1, . . . , n. Then
fn ((a1 , . . . , an ) + (b1 , . . . , bn ))
= fn ((a1 + b1 , . . . , an + bn ))
= (a1 + b1 )fn ((1, 0, . . . , 0)) + . . . + (an + bn )fn ((0, . . . , 0, 1))
= a1 fn ((1, 0, . . . , 0)) + . . . + an fn ((0, . . . , 0, 1))+
b1 fn ((1, 0, . . . , 0)) + . . . + bn fn ((0, . . . , 0, 1))
= fn ((a1 , . . . , an )) + fn ((b1 , . . . , bn )).
i i
i i
i i
K23798 2015/2/2
i i
The next two results show that a neutral and linearly decomposable ag-
gregation rule must be a weighted mean.
Theorem 4.2.4 Let fe be a fuzzy aggregation rule and let fn be the auxiliary
function associated with fe. If fe is linearly decomposable, then there exists a
unique linear transformation of Rn into R such that fc n |[0,1]n = fn .
Corollary 4.2.5 Let fe be a fuzzy aggregation rule and let fn be the aux-
iliary function associated with fe. If fe is linearly decomposable,
Pn then =
(1 , . . . , n ) FRn (X) and x, y X, fe()(x, y) = i=1 wi i (x, y), where
wi = fn (1i ) for i = 1, . . . , n.
The previous result applies to committees or voting bodies where individ-
uals do not contribute equally to social preference choices. This may happen,
for example, when seniority or rank of individuals is involved.
Corollary 4.2.7 Let fe be a fuzzy aggregation rule and let fn be the auxiliary
function associated with fe. If fe is linearly decomposable, then wi = n1 for
i = 1, . . . , n if and only if fe is anonymous.
Example 4.2.8 Let N = {1, 2} and X = {x, y}. Let 1 and 2 be WFPR
on X defined by 1 (x, x) = 1 (y, y) = 1 = 2 (x, x) = 2 (y, y) and 1 (x, y) =
2/3, 1 (y, x) = 1/3, 2 (x, y) = 1/3, 2 (y, x) = 2/3. Let w1 = 1/4 and w2 =
3/4. Suppose fe is a fuzzy aggregation rule such that u, v X,
i i
i i
i i
K23798 2015/2/2
i i
Also,
and
i i
i i
i i
K23798 2015/2/2
i i
25, 26, 27]. Differences between impossibility and possibility theorems can be
explained by the fact that each model uses different assumptions, particularly
with regard to how transitivity is modelled. This was seen in the previous
two sections. As seen in Chapter 3, the appropriate way of factoring a fuzzy
weak preference relation into a fuzzy strict preference relation and a fuzzy
indifference relation can also play a role.
The main result in this section states that an aggregation rule satisfying
certain criteria is dictatorial if and only if the triangular norm used in the
formulation of the transitivity condition has no zero divisor. Consequently,
max-min transitivity leads to dictatorship, whereas Lukasiewicz transitivity
does not.
Fuzzy aggregation rules that satisfy counterparts of unanimity and in-
dependence of irrelevant alternatives are assumed in the characterization of
the set of triangular norms that permit preference aggregation to be non-
dictatorial.
Recall that X is a set of alternatives with |X| 3 and N = {1, 2, . . . , n}
with n 2 is a finite set of individuals. For FR(X), we say that is
connected if for all x, y X, (x, y) = 0 implies (y, x) = 1. Let S be the set
of all FR(X) such that is reflexive, connected, and max- transitive.
A fuzzy aggregation rule (FAR) is a function fe : S n S. We often
write = fe(1 , . . . , n ), where fe is an FAR.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
(c, d).
(5) (a, b), (c, d) X X with a = d or b = c. Since |X| 3, there exists
e X such that a 6= e 6= c. Let (1 , . . . , n ) S n be such that j (a, b) =
j (a, e) = j (c, e) = j (c, d) = 1 for all j N. By cases (2) and (3), we
have that (a, b) = (a, e) = (c, e) = (c, d). Let W n denote the set of such
profiles. Let (1 , . . . , n ) S n be such that j (a, b) = j (c, d) = 1 for all j N.
Then there exists (01 , . . . , 0n ) W n such that j (a, b) = j (c, d) = 0j (a, b) =
0j (c, d) for all j N. By I, we have that (a, b) = (c, d) = 0 (a, b) =
0 (c, d). Consider distinct profiles (001 , . . . , 00n ), (1 , . . . , n ) S n such that
00j (a, b) = j (c, d) for all j N. Then there exists ( n
1 , . . . , n ) W such
00
that j (a, b) = j (c, d) = j (a, b) = j (c, d) for all j N. Then I implies
that 00 (a, b) = (c, d) = (a, b) = (c, d).
Lemma 4.3.4 ([17]) If has no zero divisor, then every FAR fe satisfying I
and U is dictatorial.
i i
i i
i i
K23798 2015/2/2
i i
(0) = (1 , . . . , n ),
(1) = (01 , 2 , . . . , n ),
(2) = (01 , 02 , 3 , . . . , n ),
..
.
(n) = (01 , . . . , 0n ).
At some point in this sequence, the social value of (a, b) rises from 0 to some
positive number. Assume that this happens at (2) when individual 2 changes
his or her preferences from (a, b) to 0 (a, b). We prove that this individual is
a dictator.
Consider a profile (01 , 2 , 03 , . . . , 0n ) S n . We now show that at this
profile the social value of (a, b) is zero. Let (1 , . . . , n ) S n be as fol-
lows: At this profile, every individuals (a, c) preference is the same as their
(a, b) preference at (1) . Everyones (a, b) preference is the same as their
(a, b) preference at (01 , 2 , 03 , . . . , 0n ). Finally everyones (b, c) preference is
the same as their (a, b) preference at (2) . Max- transitivity implies that
(a, c) (a, b) (b, c). Since fe is neutral, 0 ( (a, b), ), where > 0.
If = 1, then (a, b) = 0. If < 1, then since contains no zero divisor,
(a, b) = 0. By I, we have that at (01 , 2 , 03 , . . . , 0n ) S n the social value of
(a, b) is zero.
At this profile, connectedness implies that 2 (b, a) = 1 and also that the
social value of (b, a) must be equal to 1. This is true irrespective of everyone
elses (b, a) values. Neutrality therefore implies that for all (1 , . . . , n ) S n
and for all (a, b) X X, 2 (a, b) = 1 implies (a, b) = 1.
[fe((1) )(a, b) = 0, f ((2) )(a, b) > 0, 1 (a, c) = 01 (a, b), i (a, c) = i (a, b), i
= 2, . . . , n,
2 (a, c) = 2 (a, b) = 0,
1 (a, b) = 01 (a, b), 2 (a, b) = 2 (a, b) = 0, i (a, b) = 0i (a, b), i = 3, . . . , n,
1 (b, c) = 01 (a, b), 2 (b, c) = 02 (a, b), i (b, c) = i (a, b), i = 3, . . . , n.
(Thus (b, c) = fe((2) )(a, b) > 0 by neutrality.)
Neutrality yields (a, c) = fe((1) (a, b)) = 0. Hence
0 = (a, c) (a, b) (b, c), (b, c) = f ((2) )(a, b) > 0 by neutrality.
Hence
(a, b) = 0 since has no zero divisors. (a, b) = 0 so (b, a) = 1.]
Let (1 , . . . , n ) S n be such that 2 (c, b) = 2 (b, c) = 1 and i (a, c) =
2 (a, b) for all i N. From the above argument, it follows that (c, b) =
(b, c) = 1 and that U implies (a, c) = 2 (a, b). Since is max- transitive,
we have (a, c) (c, b) (a, b) and (a, b) (b, c) (a, c). In other words,
2 (a, b) 1 (a, b) and (a, b) 1 2 (a, b). Thus (a, b) = 2 (a, b). By
neutrality, we have that for all (b 1 , . . . , bn ) S n and for all (a, b) X X,
b(a, b) = b2 (a, b).
i i
i i
i i
K23798 2015/2/2
i i
Lemma 4.3.5 ([17]) If has a zero divisor, then there exists a non-dictatorial
FAR satisfying I and U.
Proof. Define the function fe : S n S as follows. For all a, b X and
all (1 , . . . , n ) S n , let (a, b) be equal to the median value of the three
numbers M (a, b), x, and m(a, b), where x is a zero divisor with the property
x x = 0. Then fe satisfies I and U and is non-dictatorial. It only remains
to prove is that fe takes values in S. Clearly, fe is reflexive and connected. It
remains to prove that fe satisfies max- transitivity.
Assume, by way of contradiction that fe does not satisfy max- transitivity.
Then there exists (1 , . . . , n ) S n and a, b, c X such that (a, b) (b, c) >
(a, c). We first show that (a, b) x and (b, c) x is impossible. Since
x x = 0, we have that if (a, b) x and (b, c) x, then (a, b) (b, c) = 0.
However, this contradicts the assumption that (a, b) (b, c) > (a, c).
Second, we show that (a, b) > x and (b, c) > x is impossible. Suppose it
is the case that (a, b) > x and (b, c) > x. By the definition of fe, (a, b) > x
and (b, c) > x imply that (a, b) = m(a, b) and (b, c) = m(b, c). Let j N
be such that j (a, c) = m(a, c). Since individual j 0 s preferences are max-
transitive, we have that j (a, b)j (b, c) j (a, c). From the definition of m, it
follows that j (a, b) m(a, b) and j (b, c) m(b, c), and so j (a, b) (a, b)
and j (b, c) (b, c). Consequently, (a, b) (b, c) is less than or equal to
j (a, b) j (b, c). Hence (a, b) (b, c) j (a, c). We have that j (a, b) =
m(a, c) and that m(a, c) (a, c) M (a, c). Thus (a, b) (b, c) (a, c).
This contradicts the assumption that (a, b) (b, c) > (a, c).
It remains to consider two other possibilities. Either (i) (a, b) > x and
(b, c) x, or (ii) (a, b) x and (b, c) > x. Assume that (i) is true.
Then (a, b) x is greater than or equal to (a, b) (b, c). Therefore, given
our earlier assumption that (a, b) (b, c) > (a, c), it must be the case
that (a, b) x > (a, c). Now 1 x = x. Since (a, b) 1, we have
(a, b) x x. Since (a, b) x > (a, c) and (a, b) = m(a, b), it follows
that x > (a, c). By the definition of fe, x > (a, c) implies (a, c) = M (a, c).
It also holds that (a, b) > x implies (a, b) = m(a, b). Thus for all i N,
i (a, b) (a, b) and i (a, c) (a, c). Hence there exists k N such
that k (a, b) (a, b), k (a, c) (a, c) and k (b, c) = M (b, c). Since k
is max- transitive, we have that k (a, b) k (b, c) k (b, c). We also have
that m(b, c) (b, c) M (b, c). Thus (b, c) M (b, c) = k (b, c). Since
i i
i i
i i
K23798 2015/2/2
i i
(a, b) k (a, b) and (b, c) k (b, c), it follows that (a, b) (b, c) is less
than or equal to k (a, b) k (b, c). Hence (a, b) (b, c) k (a, c). Thus since
k (a, c) (a, c), we have (a, b) (b, c) (a, c). However, this contradicts
the assumption that (a, b) (b, c) > (a, c).
We consider the intuitive meaning that can be given to the requirement
that a triangular norm contains, or does not contain, a zero divisor. The
max- transitivity for a fuzzy binary relation states that for all x, y, x X,
(x, z) (x, y) (y, z). If this triangular norm contains a zero divisor, then
(under one interpretation of fuzzy preferences) an individual or society can
believe that the proposition x is at least as good as y is not false i.e.,
(x, y) > 0 and that the proposition y is at least as good as z is not false,
i.e., (y, z) > 0 and yet still believe that the proposition x is at least as good
as z is false, i.e., (x, z) = 0. These beliefs are not inconsistent. However, if
the triangular norm does not contain a zero divisor, then the proposition x
is at least as good as z cannot be false, i.e., (x, z) > 0. This is a difference
between the two conditions.
4.4 Exercises
Let be a fuzzy preference relation. Then is called minimally transitive
(Tm ) if x, y, z X, (x, y) = 1 and (y, z) = 1 imply (x, z) = 1. Let be
the strict fuzzy preference relation associated with . is called partially
quasitransitive if (x, y) > 0 and (y, z) > 0 imply (x, z) > 0. is called
negatively transitive if x, y, z X, (x, y) > 0 implies (x, z) > 0 or
(z, y) > 0.
4.5 References
1. K. J. Arrow, Social Choice and Individual Values, Wiley, New York 1951.
i i
i i
i i
K23798 2015/2/2
i i
11. M. Dasgupta and R. Deb, Transitivity and fuzzy preferences, Soc. Choice
Welf., 13 (1996) 305318.
12. M. Dasgupta and R. Deb, An impossibility theorem with fuzzy prefer-
ences. In: de Swart H (ed) Logic, game theory and social choice: Pro-
ceedings of the International Conference, LGS 99, May 1316, 1999,
Tilburg University Press 1999.
13. M. Dasgupta and R. Deb, Factoring fuzzy transitivity, Fuzzy Sets and
Systems, 118 (2001) 489502.
14. F. Dietrich and C. List, The aggregation of propositional attitudes: to-
wards a general theory, forthcoming in Oxford Studies in Epistemology
2009.
15. C. Duddy and A. Piggins, Many-valued judgement aggregation: Charac-
terizing the possibility/impossibility boundary for an important class of
agendas, Working paper, Department of Economics, National University
of Ireland, Galway 2009.
i i
i i
i i
K23798 2015/2/2
i i
18. B. Dutta, Fuzzy preferences and social choice. Math Soc. Sci., 13 (1987)
215229.
23. N. Jain, Transitivity of fuzzy relations and rational choice, Ann. Oper.
Res., 23 (1990) 265278.
24. E. P. Klement, R. Mesiar, and E. Pap, Triangular norms, Kluwer Aca-
demic Publishers, Dordrecht, 2000.
25. B. Leclerc, Efficient and binary consensus functions on transitively val-
ued relations, Math. Soc. Sci., 8 (1984) 4561.
26. B. Leclerc, Aggregation of fuzzy preferences: A theoretic Arrow-like ap-
proach, Fuzzy Sets and Systems, 43 (1991) 291309.
27. B. Leclerc and B. Monjardeet, Lattical theory of consensus. In: Barnett
W., Moulin H., Salles M., and Schofeld N. (eds.) Social Choice, Welfare
and Ethics, Cambridge University Press, Cambridge, 1995.
28. J. N. Mordeson, M. B. Gibilisco, and T. D. Clark, Independence of
irrelevant alternatives and fuzzy Arrows theorem, New Mathematics
and Natural Computation, 8 (2012) 219237.
31. S. V. Ovchinnikov, Social choice and Lukasiewicz logic, Fuzzy Sets and
Systems, 43 (1991) 275289.
32. S. V. Ovchinnikov and M. Roubens, On strict preference relations, Fuzzy
Sets and Systems, 43 (1991) 319326.
33. S. V. Ovchinnikov and M. Roubens, On fuzzy strict preference, indiffer-
ence and incomparability relations, Fuzzy Sets and Systems, 49 (1992)
1520.
i i
i i
i i
K23798 2015/2/2
i i
34. J. Perote-Pe
na and A. Piggins, Strategy-proof fuzzy aggregation rules,
J. Math. Econ., 43 (2007) 564580.
35. J. Perote-Pe
na and A. Piggins, Non-manipulative social welfare func-
tions when preferences are fuzzy, J. Log. Comput., 19 (2009) 503515.
36. J. Perote-Pena and A. Piggins, Social choice, fuzzy preferences and ma-
nipulation. In. Boylan T. Gekker R. (eds), Economics, rational choice,
and normative philosophy. Routledge, London, 2009.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 5
137
i i
i i
i i
K23798 2015/2/2
i i
Let FB denote the set of all reflexive and complete fuzzy binary relations on
X.
i i
i i
i i
K23798 2015/2/2
i i
For the Pareto extension rule, there are two underlying rules: (1) if every-
one in the society is indifferent between two alternatives x and y, then the
society should be indifferent also and (2) if at least one individual strictly
prefers x to y and every individual regards x to be at least as good as y, then
the society should prefer x to y. Consider rule (1). Suppose everyone in the
society is indifferent between two alternatives x and y. Then R(x, y; ) = N
and P (x, y; ) = . Hence by the contrapositive of the Pareto extension rule,
we conclude that it is not the case that fe()(x, y) > 0, fe()(y, x) = 0 for some
x and y. For fe() complete, we thus have fe()(x, y) > 0, fe()(y, x) > 0. In the
crisp case, this yields that the society is indifferent between x and y while in
the fuzzy case, there is a positive degree with which the society is indifferent.
We sometimes repeat a definition involving fuzzy preference aggregation
rules when the context is changed, i.e., its domain is changed.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
z X\{x, y}, xD
e y xD z. (5.1)
i i
i i
i i
K23798 2015/2/2
i i
+
i (y, z) = 00i (y, z) and + 00
i (z, y) = i (z, y) i N
i+ (y, x) > 0 i N,
i+ (x, z) > 0 i Supp() and j+ (z.x) > 0 j N \Supp().
and + (z, y) = 0. Hence 00 (y, z) > 0 and 00 (z, y) = 0.) (For = (2) , we
write: Since + 00 +
i e{y,z} = i e{y,z} i N and f is IIA4, Cosupp( e{y,z} ) =
e
00 + + 00
Cosupp( e{y,z} ). Since (y, z) > 0, (z, y) < 1. Hence (z, y) < 1.)
Thus 00 (y, z) > 0 and so yD z. Hence since z is arbitrary in X\{x, y}, the
preceding two steps yield
z
/ {x, y}, xD
e y yD z. (5.2)
e y xD v (by (5.1)) xD
xD e v vD w (by (5.2)) with v replacing y.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Then is weakly transitive, but not max-min transitive since (z, x) (z, y)
(y, x).
(c) Define the fuzzy relation on X as follows:
Then is weakly transitive, but not transitive in the crisp sense. Note that
is not complete.
i i
i i
i i
K23798 2015/2/2
i i
such that is semidecisive for x against y.) This follows by Lemma 5.1.15
because then is decisive for x against y for all x, y X, where Supp() = {i}.
Hence FRn (X), x, y X, i (x, y) > 0 implies (x, y) > 0.
Since fe is weakly Paretian, a decisive for any pair of alternatives,
namely = 1N . For all (u, v) X X, let m(u, v) denote the size of the
smallest |Supp()| for semidecisive for u against v. Let m = {m(u, v) |
(u, v) X X}. Without loss of generality suppose is semidecisive for
x against y with |Supp()| = m. If m = 1, the proof is complete. Suppose
m > 1. Let i Supp(). Consider any fuzzy profile FRn such that
Definition 5.2.1 Let i N and t (0, 1]. Define the fuzzy subset it of N
by j N, it (j) = t if j = i and it (j) = 0 otherwise. (Then it is called a
fuzzy singleton.) If g : FRn {it | i N, t (0, 1]}, then g is called a
representation rule.
Let g be a representation rule. Then Supp(g) = {i0 } for some i0 N. By
the notation, Supp(g()) we mean i0 .
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Lemma 5.2.9 Let strict preferences be of type (0) . Let fe be a fuzzy aggrega-
tion rule which is partially quasi-transitive, weakly Paretian, and independent
of irrelevant alternatives IIA2. If i N has a semiveto for x against y for
some x, y X, then i has a veto over all ordered pairs (v, w) X X.
Proof. Let FRn be such that
Since i has a semiveto for x against y, not (y, x) > 0. Since fe is complete
(x, y) > 0. Since fe is weakly Paretian, (y, z) > 0. Since fe is partially quasi-
transitive, (x, z) > 0 since = (0) . Hence not (z, x) > 0. By independence
of irrelevant alternatives IIA2, i has a veto for x against z. (We have not
i i
i i
i i
K23798 2015/2/2
i i
(z, x) > 0 for this , but we need independence to get it for all . Let 0 be
such that i0 (x, z) > 0. Then not 0 (z, x) > 0 by independence.) Thus we have
z
/ {x, y}, xVei y xVi z. (5.3)
Now since i has a veto for x against z, i has a semiveto for x against z. Thus
switching y and z in the above argument implies i has a veto for x against
y. Let 00 be any fuzzy preference profile such that i00 (y, z) > 0. Let + be a
fuzzy preference profile such that
Since i has a semiveto for x against z, not + (z, x) > 0. Thus + (x, z) > 0.
Since fe is weakly Paretian, + (y, x) > 0. Since fe is partially quasi-transitive,
+ (y, z) > 0. Hence not + (z, y) > 0 since = (0) . Since only the prefer-
ences for i are specified and Supp(+ 00
i e{y,z} ) = Supp(i e{y,z} ), independence
00
of irrelevant alternatives IIA2 implies not (z, y) > 0 and so i has a veto for
y against z. Thus
z
/ {x, y}, xVei y yVi z. (5.4)
Now i has a veto for y against z, i has a semiveto for y against z. Thus by
(5.3), i has veto for y against x. We have (v, w) X X,
xVei y xVi v (by (5.3)) xVei v vVi w (by (5.4)) with v replacing y.
i i
i i
i i
K23798 2015/2/2
i i
the support of some fuzzy subset of N semidecisive for z over y and thus
decisive, contradicting the minimality of m. (Now for , 0 FRn satis-
fying (5.5), Supp(i e{y,z} ) = Supp(0i e{y,z} ), i = 1, . . . , n.) Hence for any
such , fe()(y, z) > 0 by completeness. Since fe is partially quasi-transitive,
fe()(x, z) > 0 and hence not (z, x) > 0. Hence i has a semiveto for x against
z. Thus i has a veto over all ordered pairs (u, v) X 2 by Lemma 5.2.9. Hence
fe is oligarchic since i was arbitrary.
In Theorem 5.2.10, the condition of partial transitivity for a fuzzy aggre-
gation rule of Theorem 5.1.16 is relaxed to that of partial quasi-transitivity.
The result of this change is a dispersion of the decision making from a dictator
to an oligarchy.
Since i has a semiveto for x against y, not (y, x) > 0. Thus (x, y) (y, x).
Since fe is complete (x, y) > 0. Since fe is weakly Paretian, (y, z) > 0.
Since fe is partially quasi-transitive by Lemma 5.2.8, (x, z) > 0. In fact, since
(y, z) > (z, y) and (x, y) (y, x), we have (x, z) > (z, x) by Proposition
5.1.20. Hence not (z, x) > 0. By independence of irrelevant alternatives, i
has a veto for x against z. (We have not (z, x) > 0 for this , but we need
independence to get it for all . Let 0 be such that i0 (x, z) > 0. Then not
0 (z, x) > 0 by independence.) Thus we have
z
/ {x, y}, xVei y xVi z. (5.6)
Now since i has a veto for x against z, i has a semiveto for x against z. Thus
switching y and z in the above argument implies i has a veto for x against
y. Let 00 be any fuzzy preference profile such that i00 (y, z) > 0. Let + be a
fuzzy preference profile such that
Since i has a semiveto for x against z, not + (z, x) > 0. Thus + (x, z) >
0. Since fe is weakly Paretian, + (y, x) > 0. Since fe is weakly transitive,
+ (y, z) > + (z, y). Thus + (y, z) > 0. Hence not + (z, y) > 0. Since only
the preferences for i are specified and + 00
i e{y,z} i e{y,z} , independence of
i i
i i
i i
K23798 2015/2/2
i i
irrelevant alternatives IIA3 implies not 00 (z, y) > 0 and so i has a veto for y
against z. Thus
z
/ {x, y}, xVei y yVi z. (5.7)
Now i has a veto for y against z, i has a semiveto for y against z. Thus by
(5.6), i has veto for y against x. We have (v, w) X X,
xVei y xVi v (by (5.6)) xVei v vVi w (by (5.7)) with v replacing y.
Let fe be a fuzzy aggregation rule and let L(fe) denote the set of decisive
coalitions associated with fe.
i i
i i
i i
K23798 2015/2/2
i i
i (x, y) > 0 for all i Supp(0 ), but not (x, y) > 0. Thus i (x, y) > 0 for all
i Supp(), but not (x, y) > 0. Hence is not decisive for fe.
Let L(fe). Let x, y X and FRn be such that i (x, y) > 0 for
all i Supp() and i (y, x) > 0 i N \Supp(). Then (x, y) > 0 and so
(y, x) > 0 is impossible. Hence 0 is not decisive for fe for all 0 F P(N )
such that Supp()Supp(0 ) = .
i i
i i
i i
K23798 2015/2/2
i i
e{xi ,xi+1 } (i) e{xi ,xi+1 } , i = 1, . . . , n and e{x1 ,xn+1 } (0) e{x1 ,xn+1 } .
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
not yD1N \L x z
/ {x, y}, xD z and zD y not zD1N \L x
w
/ {x, z}, xD w and wD z not zD1N \L w
v
/ {w, z}, wD v and vD z not vD1N \L w
u
/ {v, w}, wD u and uD v.
That is, if 1N \L is not decisive for some ordered pair of alternatives (y, x)
X X, then must be decisive for all ordered pairs of alternatives. Therefore,
since fe is weakly Paretian and independent of irrelevant alternatives IIA2,
fe(FRn ) FR implies that (P5) holds.
In the crisp case, the previous theorem shows the trade-off between the
concentration of power in a society and the extent to which social preferences
are rational. A fuller discussion can be found in Austen-Smith and Banks [2,
p. 48].
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Suppose (x, y) > (x, y). Then (x, y) = (x, y). Thus (x, y) = 0 by (4).
Since is symmetric, (y, x) = 0. Also, (x, y) = 0 from the asymmetry of .
Hence (y, x) = 0. But by the strong completeness of , (x, y) + (y, x) 1.
Hence (x, y) = 1, (y, x) = 0. Thus is an EWPR.
Suppose (x, y) > (x, y). By (4), (x, y) = 0. Also, (y, x) = (x, y) > 0.
Hence by (4), (y, x) = 0. Thus (x, y) = (y, x) = (x, y) = (y, x). Finally,
if (x, y) = (x, y), then (x, y) = (x, y) = (x, y) = 0 by (1) and (4). From
the strong completeness of , we have (y, x) = (y, x) = 1. Thus is exact.
Due to the previous proposition we will not require the condition that
= .
Recall Proposition 3.1.9. The derivation of and there was suggested by
Ovchinnikov [24].
There have been other suggestions in the literature concerning the deriva-
tion of the and relations from an FWPR. One possibility is the following
for all x, y X,
(x, y) = 0 ((x, y) (y, x)) and (x, y) = (x, y) (y, x). (5.9)
Definition 5.4.2 Let be an FWPR, with and its asymmetric and sym-
metric components. Let be a t-norm. Then satisfies
(1) PP-transitivity if for all x, y, z X, (x, z) (x, y) (y, z);
(2) IP-transitivity if for all x, y, z X, (x, z) (x, y) (y, z);
(3) PI-transitivity if for all x, y, z X, (x, z) (x, y) (y, z);
(4) II-transitivity if for all x, y, z X, (x, z) (x, y) (y, z).
Example 5.4.3 Let be a FWPR, with and its asymmetric and symmet-
ric components, respectively.
(1) max- transitivity of does not imply P I or IP transitivity.
(2) T2 -transitivity of does not imply PP-transitivity.
For (1), Ovchinnikow [24] shows that PI as well as IP implies = .
Since 6= in our construction, the proposition follows.
For (2), Let (x, x) = (y, y) = (z, z) = 1; (x, y) = (y, z) = (z, x) =
2
3 ; (x, z) = (y, x) = (z, y) = 13 . Then, the corresponding values are:
i i
i i
i i
K23798 2015/2/2
i i
2
(x, x) = (y, y) = (z, z) = 0; (x, y) = (y, z) = (z, x) = 3 ; (x, z) =
(y, x) = (z, y) = 0.
Proposition 5.4.4 (Dutta [11]) Let be a FWPR, with and its asym-
metric and symmetric components. Then maxmin transitivity of implies
PP-transitivity.
Proof. Suppose satisfies max-min transitivity, but not PP-transitivity.
Then for some x, y, z X, we must have:
and
(x, z) < (x, y) (y, z). (5.12)
Thus
0 < (x, y) = (x, y) > (y, x), (5.13)
0 < (y, z) = (y, z) > (z, y). (5.14)
By (5.11) and (5.12), we have
or
(z, y) (x, y). (5.22)
i i
i i
i i
K23798 2015/2/2
i i
Now (5.21), (5.16) and (5.20) imply (z, y) (y, z). Similarly, (5.22) and
(5.19) imply (z, y) (y, z). In either case, (5.14) is violated.
Example 5.4.3 and Proposition 5.4.4 play an important role in the possi-
bility theorems that follow.
Let HE denote the set of reflexive, strongly complete, transitive exact fuzzy
preference relations over X. Nonempty subsets of N are called coalitions at
times.
We next present the fuzzy counterpart of Gibbards oligarchy result. Its
proof is analogous to the proof of the corresponding result in [6, Theorem 3.5]
and also similar to the proof of Gibbards theorem in the exact framework.
For a sketch of the proof of the latter, see for instance Sen, [28].
Proposition 5.4.5 assumes that individual and social preferences are max
min transitive. As shown in [6], the proposition is true under any transitivity
condition on FWPRs such that the corresponding strict preference relations
satisfy for all x, y, z X,
i i
i i
i i
K23798 2015/2/2
i i
Definition 5.4.8 Let be any FWPR and A P (X). The -greatest set
in A is defined to be the function G(A, ) : P (X) [0, 1] such that for all
x X\A, G(A, )(x) = 0 and for all x A, G(A, )(x) = {(x, y) | y A}.
Let B(A, ) = {x A | G(A, )(x) G(A, )(y), for all y A}.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let A = {x, y, z}, fe and C satisfy the hypotheses of the proposi-
tion. By Proposition 5.4.5, there is a unique oligarchy G. Since f is nondicta-
torial, |G| 1.
Without loss of generality, assume that {1, 2} G. Construct (1 , 2 , . . . ,
n ) as follows:
(i) for all i N, i (x, y) = 1 and i (y, x) < 1,
(ii) 1 (y, z) > 1 (z, y) and 1 (x, z) > 1 (z, x),
(iii) 2 (z, y) > 2 (y, z) and 2 (z, x) > 2 (x, z),
(iv) for all i N \{1, 2}, i (y, z) = i (z, y) = i (z, x) = i (x, z).
It follows that there exists (1 , . . . , n ) H n satisfying restrictions (i)
(iv).
Since f satisfies PC, (x, y) = 1. Hence G({x, y}, )(x) = 1. Thus
G({x, y}, )(y) = 0. Since 1 C, (ii) implies that (z, y) = (z, x) = 0. Sim-
ilarly, (iii) implies that (y, z) = (x, z) = 0. Also, (z, y) = (y, z) = 0 im-
plies (y, z) = (z, y), and (z, x) = (x, z) = 0 implies that (z, x) = (x, y).
Hence G({y, z}, )(y) = G({y, z})(z) and G({x, z}, )(x) = G({x, z}, )(z).
Since C is H-generated by , G({x, y}, )(x) > G({x, y}, )(y)
C({x, y}) = {x}. Also, C({y, z}, )(y) = G({y, z}, )(z) C({y, z}) = {y, z}
and G({x, z}, )(x) = G({x, z}, )(z) C({x, z}) = {x, z}.
Suppose y / C(A). Since C satisfies Property (+), and y C({y, z}),
we have that z
/ C(A). However, z C({x, z}). Hence, if z / C(A), then
x/ C(A).
i i
i i
i i
K23798 2015/2/2
i i
Thus if y
/ C(A), then C(A) = . Hence y C(A) and the desired result
holds.
In contrast to Proposition 5.4.13, Propositions 5.4.5 and Theorem 4.1.23
show that the class of FARs (with domain H1n and range H1 ) satisfying IIA
and PC must be oligarchic, but not necessarily dictatorial. However, if the
nondictatorial FARs in this class are further constrained to H-generated ex-
act choice functions, then in some situations, an alternative which is Pareto-
dominated will also be chosen. Thus the necessity to generate exact social
choice may under certain circumstances lead to strong impossibility results
even in a fuzzy framework as commented by Dutta.
As noted by Dutta, the concept of H-rationability in terms of max-min
transitive FWPRs requires the exact choice function to satisfy Property (+).
n
Suppose the domain of the aggregation rule is restricted to HE , i.e., individual
preferences are assumed to be exact. If IIA and the Pareto condition are
rephrased in choice-functional terms, and if the aggregation rule is defined to
n
be a mapping from HE to the set of choice functions, then it is known that
property (+) leads to the existence of a dictator. Thus this result shows that
n
a dictator on the domain HE remains a dictator on the expanded domain H1n .
A corresponding result for T2 -transitivity is not necessarily true. The
restrictions imposed on exact choice functions by the concept of H-
rationalizability in terms of T2 -transitive FWPRs are not known. These re-
strictions could, however, be weaker than known rationality conditions such
as Property (+). Hence an impossibility result, even when social choice is
assumed to be exact, is not inevitable in a fuzzy framework.
In contrast to [6], Dutta has permitted individual and social preferences to
be fuzzy weak preference relations. Dutta has argued that differences in the
structure of fuzzy binary relations from that of exact binary relations become
apparent only if the binary relations are allowed to be weak. Also, qualitative
differences in the nature of the possibility theorems in the two frameworks do
follow form these differences in the structure of binary relations.
Dutta goes on to state that with a relatively strong version of the tran-
sitivity condition on FWPRs, the counterparts of the conditions assumed by
Arrow [1] lead to an oligarchy in the fuzzy framework. The dictatorship re-
sult is restored if Positive Responsiveness is imposed in addition to the other
conditions. However, with a weaker form of transitivity, all these conditions
can be satisfied without leading to oligarchic rules.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Proof. The proof is once again similar to the crisp case. We consider
the proof of the Group contraction part. Let M be a decisive group. Let
M = M1 M2 , where M1 and M2 are not empty. Suppose that there exists
x, y, z X such that
i (x, y) = 1 and i (y, z) = 1 for all i M1 ;
i (y, z) = 1 and i (z, x) = 1 for all i M2 ;
i (z, x) = 1 and i (x, y) = 1 for all i N \M.
Suppose (x, z) = 1. Then M1 is almost perfectly decisive over (x, z). Hence
it is a decisive group. Suppose (x, z) < 1. Then (z, x) > 0. Hence we have
(z, y) 21 (z, x) + 12 (x, y) > 0 contradicting the fact that M is decisive so
that (y, z) = 1 and (z, y) = 1 (y, z) = 0. Thus (x, y) = 0. (Note that
by the definition of T -transitivity, the weak inequality does not apply in this
case.) Hence (y, x) = 1. Thus M2 is almost decisive over (y, x). Hence it is
decisive.
It has been shown here that when individual preference relations are ex-
act, the power of the dictator increases if the transitivity property of the
social preference relation is strengthened. That is, T2 -transitivity, max-min
transitivity, and T -transitivity restrictions on the social preference relation
imply respectively, the existence of a dictator, an almost perfect dictator and
a perfect dictator.
Banerjee notes the following. Similar results can be obtained when the
domain of fe is H1n and H n instead of U n . If the definitions of a dictator, an
almost perfect dictator and a perfect dictator in Definition 5.5.8 are kept un-
changed, these similar results are immediately established. However, in these
cases the definition that is accepted in the literature declares individual j as
a dictator if j (x, y) > 0 implies (x, y) > 0 for all x, y X and for all pro-
files of individual preference relations. In this case, if we wish to define, say,
a perfect dictator, a natural definition will assign that status to individual j
if j (x, y) = implies (x, y) = for all values of lying between 0 and 1,
for all x, y X, and for all profiles of individual preference relations. How-
ever, this strong a form of the result is not true, as counterexamples are easily
constructed.
i i
i i
i i
K23798 2015/2/2
i i
that it seems unreasonable that the two situations ((x, y) = 1, (y, x) = .99)
and ((x, y) = 1, (y, x) = 0) should yield exactly the same value for (x, y).
Richardson adds that the discontinuity of (1) is also an intuitive difficulty;
while in the previous pair of cases that values of are too similar, it seems
rather unreasonable for the values of to be so different in the two situations
((x, y) = 1, (y, x) = .99) and ((x, y) = 1, (y, x) = 1).
Richardson also states the following. In the absence of strong connected-
ness, the relation (2) (x, y) = 1 (y, x) is not asymmetric. Its insensitivity
to (x, y) also becomes an intuitive problem if is not strongly connected;
(2) (x, y) takes exactly the same value in the two situations ((x, y) = 1.0,
(y, x) = .99) and ((x, y) = .01, (y, x) = .99). This is not to say that strong
connectedness is an undesirable restriction in all situations. It has proven to
be useful in [9] and [22].
Recall Proposition 3.1.8. Let , , and satisfy the following conditions:
(i) = 2 ;
(ii) x, y X, (x, y) + (x, y) 1;
(iii) is symmetric;
(iv) is asymmetric.
Then x, y X,
(x, y) = (x, y) (y, x) (5.24)
and
(x, y) = (3) (x, y) = 0 ((x, y) (y, x)). (5.25)
Clearly, (3) and (2) coincide when is strongly connected. This version
of strict preference, (3) (x, y) = 0 ((x, y) (y, x)), satisfies Banerjees
objection to (1) and is sensitive to the values of both (x, y) and (y, x).
A further advantage of (3) over (1) is its continuity ((1) (x, y) = (x, y) if
(x, y) > (y, x) and 0 otherwise), as a function of (x, y) and (y, x).
The use of (3) (x, y) as the specification of strict preference implies
For a given FWPR , (1) and (3) are clearly regular and, when is
strongly connected, so is (2) .
The general requirement for extending any concept from exact sets to
fuzzy sets is that the fuzzy extension must still agree with the original concept
when fuzzy sets are restricted to be exact. Therefore any fuzzy extension of
the concept of transitivity must be at least as strong as the following.
i i
i i
i i
K23798 2015/2/2
i i
Now suppose (x, y) > 0 and (y, z) > 0, but that (x, z) = 0. Then by (5.29),
(z, x) = 1. Also, (x, y) > 0 (y, x) = 0. Thus (x, y) = 1. However, since
is also TM, (z, x) = 1 and (x, y) = 1 (z, y) = 1. Hence by (5.29),
(y, z) = 0, a contradiction.
By requiring strong connectedness and replacing T2 -transitivity with the
even weaker condition TM, Proposition 5.6.3 clarifies and extends Proposition
5.5.2 by which T2 -transitivity implies the quasitransitivity of (2) .
Definition 5.6.5 A fuzzy aggregation rule is said to satisfy the Pareto con-
dition (PC) if W n and x, y X, (x, y) {i (x, y) | i N }.
i i
i i
i i
K23798 2015/2/2
i i
Let
H1 = the set of strongly complete FWPRs satisfying max-min tran-
sitivity.
H2 = the set of strongly complete FWPRs satisfying T2 -transitivity.
S2 = the set of strongly connected FWPRs satisfying T2 -transitivity.
SM = the set of strongly connected FWPRs satisfying TM-
transitivity.
E = the set of exact, strongly connected, and transitive weak pref-
erence relations.
In Proposition 5.4.4, it was shown with strict preferences of type (1) that
only FAR fe : H1n H1 satisfying IIA1, PC and PR is a dictatorship. There
exists a nondictatorial FAR fe : H2n H2 that satisfies IIA1, PC and PR (see
the proof of Theorem 4.1.15). Weakening the transitivity requirement seems to
create some possibility for bypassing the dictatorship. However, in Proposition
5.4.4 it is proved that if (2) were substituted for (1) , the only FAR that
satisfies IIA1 and PC is a dictatorship, even when the version of transitivity
used is only T2 . As we have seen, however, the use of (2) implies strong
connectedness (and Banerjees proof of this proposition uses facts about (2)
that are only true under strong connectedness). Hence the range and domain
of the FAR in Banerjees theorem are S2 and S2n , respectively.
The following proposition demonstrates that it does not seem to be the
choice of specification of strict preference that generates the dictatorship re-
sult, but rather it is the restriction to strong connectedness. Furthermore,
Richardson noticed that when strong connectedness is assumed, the dicta-
torship result stands even if the transitivity requirement is further weakened
beyond T2 -transitivity to the weakest possible version of fuzzy transitivity,
TM.
Theorem 5.6.8 (Richardson [26]) Let strict preference be defined by any reg-
n
ular . If fe : SM SM satisfies IIA1 and PC, there exists a dictator.
Proof. As in [26], we sketch the proof since the proof follows known ar-
guments. It is customary to prove an Arrowian impossibility theorem in two
stages ([15], [27]). First, a field expansion lemma shows that a group of
voters that are almost decisive over one pair of alternatives will be decisive
over any pair of alternatives. It is known that this lemma is implied by qua-
sitransitivity. In fuzzy preference terms, Proposition 5.6.3 demonstrates that
TM-transitivity is all that is required to obtain partial quasitransitivity of any
regular strict fuzzy preference relation, including (1) , (2) and (3) , if strong
connectedness is assumed.
Secondly, a group contraction lemma is used to show that if a group of
voters containing more than one member is decisive, it must have a proper
subset that is also decisive. The dictatorship result then follows from the
finiteness of N and the fact that N is decisive.
i i
i i
i i
K23798 2015/2/2
i i
5.7 Exercises
1. Prove Lemma 5.4.12.
2. Prove that T-transitivity max-min transitivity T2 -transitivity.
5.8 References
1. K. J. Arrow, Social Choice and Individual Values, New York, Wiley
1951.
2. D. Austen-Smith and J. S. Banks, Positive Political Theory I: Collec-
tive Preference, Michigan Studies in Political Science, The University
Michigan Press 2000.
i i
i i
i i
K23798 2015/2/2
i i
10. M. Dasgupta and R. Deb, Transitivity and fuzzy preferences, Soc Choice
Welfare, 13 (1996) 305318.
11. B. Dutta, Fuzzy preferences and social choice, Math. Soc. Sci., 13 (1987)
215229.
12. B. Dutta, S.C. Panda and P. K. Pattanaik, Exact choice and fuzzy
preferences, to appear in Mathematical Social Science (1986).
13. A. Gibbard, Intransitive Social Indifference and the Arrow Dilemma,
mimeographed (1969).
14. R. A. Heiner and P. K. Pattanaik, The structure of general probabilistic
group decision rules, in: P. K Pattanaik and M. Salles, eds., Social Choice
and Welfare, North-Holland, Amsterdam, 1983.
15. J. S. Kelly, Arrow Impossibility Theorems, Academic Press, New York,
1978.
16. G. Klir and Bo Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Applica-
tions, Prentice Hall, Englewood Cliffs, 1985.
17. A. Kaufman, Introduction to the Theory of Fuzzy Sets, Academic Press,
New York, 1975.
18. A. Mas-Collel and H. Sonnenschein, General possibility theorems for
group decisions, Review of Economic Theory, 29 (1972) 185192.
i i
i i
i i
K23798 2015/2/2
i i
22. Nguyen Hung T and E. A. Walker, A First Course in Fuzzy Logic, CRC
Press, Boca Raton, FL, 1997.
23. S. A. Orlovsky, Decision making with a fuzzy binary relation, Fuzzy Sets
and Systems, 1 (1978) 155167.
28. A. K. Sen, Social Choice Theory, to appear in: K.J. Arrow and M.J.
Intriligator, eds., Handbook of Mathematical Economics, Vol. 3, North-
Holland, Amsterdam, 1985.
29. H. J. Skala, Arrows impossibility theorem: Some new aspects, Decision
Theory and Social Ethics: Issues in Social Choice, Eds.: H. W. Gottinger
and W. Leinfellner, Vol. 17, 1978, 215225.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 6
The median voter theorem states that a majority rule voting system will select
the outcome most preferred by the median voter. The theorem assumes that
voters can place election alternatives along a one-dimensional spectrum. This
is often not the case since each party will have its own policy on each of many
issues. Similarly, in the case of a referendum, the alternatives may cover more
than one issue. The theorem also assumes that voters preferences are single-
peaked, that is voters choose the alternative closest to their own view. The
theorem also assumes that voters always vote for their true preferences which
actually is not always the case. The results in this chapter were influenced by
[23, 24].
A troubling issue in spatial models is the degree to which they are suscep-
tible to majority cycling. Under majority rules, spatial models can not assure
the existence of a maximal set, that is, a set of alternatives each of which
is not strictly dominated by some other alternative. Every alternative in the
maximal set is no less preferred by a majority than any other alternative in
X. In the absence of a maximal set, spatial models tend to be badly behaved.
Any alternative in X is as likely as any other and no prediction can be made
[2].
Single dimensional models with preference profiles that are single peaked
comprise an important exception to this general rule. If there exists some
alignment of the alternatives in X along a single continuum such that the
Euclidean preferences of each political actor making a majority decision over
those alternatives descend monotonically from their respective ideal points,
then a maximal set is guaranteed to lie at the median ideal point. Known
as Blacks Median Voter Theorem, [4], the theorem has been widely used in
175
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let f be the aggregation rule associated with fe. Since fe is a Pareto
extension rule, it follows that f is a Pareto extension rule. Since is decisive
with respect to fe, Supp() is decisive with respect to f by Proposition 6.1.5.
The desired result now follows.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
feL ()(x, y) = 0.7 0.5 = 0.7, feL ()(x, z) = 0.2 0.3 = 0.3.
We also have that feL ()(y, z) = feL ()(z, y) = 1, feL ()(y, x) = feL ()(z, x) =
0, and feL ()(x, x) = feL ()(y, y) = feL ()(z, z) = 1.
We recall from the crisp situation that the knowledge of decisive sets asso-
ciated with the aggregation rule is equivalent to knowledge of the rule itself.
This leads us to the following definition.
The following result supports the fact preceding the definition and also
helps explain our definition of a partial fuzzy simple rule.
L(fe)
= { FP(N ) | is decisive forfe}
= { FP(N ) | FRn , i (x, y) > 0 i Supp() (x, y) > 0}.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
otherwise. Then individuals that 00 and 0 differ from 0 on u, v must come from
L+ . Thus 00 (u, v) > 0 since fe is monotonic. Since P (u, v; ) = P (u, v; 00 )
and fe is decisive, (u, v) > 0. Hence since and u, v are arbitrary except for
i (u, v) > 0 if i P (x, y; ) and (u, v) > 0, it follows that P (x, y; ) L(fe).
Conversely, suppose that fe is a partial fuzzy simple rule. Then fe is neutral
since L(fe) is defined without regard to alternatives. Monotonicity follows
from Proposition 6.1.10 and the definition of feL(fe) . That fe is decisive follows
directly from the definitions of L(fe) and fe e .L(f )
The intuition for the definition of feD in the next definition is similar to
that given for the definition of feL in Definition 6.2.1.
i i
i i
i i
K23798 2015/2/2
i i
We also have that feD ()(u, v) = 1 if (y, x) 6= (u, v) 6= (x, y) and that fe(y, x) =
0.
In the crisp case, voting rules are completely characterized by their decisive
structure. Partial fuzzy voting rules are characterized in a similar manner as
can be seen by the next two propositions.
(x, y) FP D(fe) ()
(, ) D(fe), i Supp(), 1Pi (x, y) = 1,
j Supp(), 1Rj (x, y) = 1
(, ) D(fe), i Supp(1Supp() ), 1Pi (x, y) = 1,
j Supp(1Supp() ), 1Rj (x, y) = 1
S, W D(f ), i S, (x, y) Pi , j W, (x, y) Rj ,
(x, y) f (R1 , . . . , Rn )
(x, y) Supp(fe(1R1 , . . . , 1Rn ))
(x, y) (X X\Symm(FP D(fe) ())) (FP D(fe) ())
(x, y) (X X\Symm(PD(f ) (R1 , . . . , Rn ))) (PD(f ) (R1 , . . . , Rn ))
(x, y) fD(f ) (R1 , . . . , Rn ).
i i
i i
i i
K23798 2015/2/2
i i
Therefore, f = fD(f ) .
We next characterize partial fuzzy voting rules.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
In Definition 6.3.6, we write F,fe() for F,fe(, Q) when the strict order Q
is understood. F,fe() is the fuzzy subset of fe-medians given with respect
to F.
Definition 6.3.7 Let F : [0, 1]2 [0, 1]. For all FT Rn , define
CF,fe()(x) = MF (fe(), 1X )(x) for all x X. Then CF,fe() is called the
fuzzy core of fe at with respect to F.
i i
i i
i i
K23798 2015/2/2
i i
Before presenting our main result, we provide some intuition for the me-
dian voter theorem. Suppose N = {1, 2, 3} and X = {x, y, z}, where the
alternatives of X are ordered, say x < y < z. Suppose player 1 favors al-
ternative x, player 2 alternative y, and player 3 alternative z. Each player
prefers the alternative closest to its ideal choice. The following table presents
the majority decision of the players.
and
e + (x)(1) = 0 since not xQx, L
L e + (x)(2) = 2 (y, x) since yQx,
+ e + (y)(1) = 0 since not xQy,
L (x)(3) = 3 (z, x) since zQx, L
e
+ +
L (y)(2) = 0 since not yQy, L (y)(3) = 3 (z, y) since zQy,
e e
e + (z)(1) = 0 since not xQz, L
L e + (z)(2) = 0 since not yQz,
+
L (z)(3) = 0 since not zQz.
e
Let fe be a simple majority rule. Let L(fe). Then FRn , i (x, y) > 0
i Supp() (x, y) > 0. Thus {Supp() | L(fe)} = {{1, 2}, {1, 3},
i i
i i
i i
K23798 2015/2/2
i i
Theorem 6.3.9 Let fe be a fuzzy aggregation rule. Assume that F : [0, 1]2
[0, 1] is such that F (0, 1) = 0. If fe is a partial fuzzy simple rule, then for any
FS, CF,fe()(x) = F,fe(,Q)(x) for all x X, where Q is the strict order
for which is single-peaked.
Proof. Let FS and z X. Then
MF (fe(), 1X )(z) = {t [0, 1] | F (fe()(z, y), 1X (x)) t y Supp(1X )}.
= {t [0, 1] | F (fe()(z, y), 1) t y X}
= {F (fe(z, y)(), 1) | y X}.
i i
i i
i i
K23798 2015/2/2
i i
Since Supp(F,fe()) Supp(MF (fe(), 1X)) we have that Supp(MF (fe(), 1X))
= Supp( e()). Hence MF (fe(), 1X ) = e(). Now assume that Supp(L e+
F,f F,f
(x)) {Supp() | L(fe)}. Then Supp(L e + (x)) = Supp() for some
L(fe). Thus u, v X, i (u, v) > 0i L e + (x) gives (u, v) > 0.
Since is single-peaked, i (at+1 , at ) > 0i Supp(L e + (x)). Hence since L(fe)
is decisive, f ()(at+1 , at ) > 0, i.e., f ()(at+1 , x) > 0. But then as before,
e e
F (fe()(x, at+1 ) = 0, a contradiction. Thus x Supp(F,fe()). Hence
We see from Theorem 6.3.9 that a partial fuzzy simple rule results in a
fuzzy core (fuzzy maximal set) in single-dimensional spatial models when-
ever the policy preferences of actors are single-peaked. Single dimensional
fuzzy spatial models of decision making under majority, unanimity, and super-
majority rules will result in stable predictions under assumptions of single-
peakedness. Hence single-dimensional fuzzy spatial models are no less stable
than the conventional single-dimensional spatial models.
i i
i i
i i
K23798 2015/2/2
i i
for all x X.
Players will select alternatives with the highest set inclusion values with
respect to M (, ). If M (, ) = , then no alternative will be selected.
i i
i i
i i
K23798 2015/2/2
i i
The fuzzy core has substantial utility in k-dimensional spatial models. Our
domain of interest is the set of real numbers R. We next develop the notion
of fuzzy a core and connect it to a nonempty maximal fuzzy subset. We begin
with the concept of single-peakedness. We then show that the maximal fuzzy
subset of players whose preferences are single-peaked lies at the median. Thus
the players will jointly prefer the median alternative to all others. We conclude
the section by showing that the median preference under single-peakedness is
not only the fuzzy maximal subset, but it is also the fuzzy core.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Example 6.4.9 Let X = [0, 1] and x2 (0, 1). Let fe be a fuzzy aggregation
rule such that FRn and x, y X,
11 if x = y,
x2 y if x = x2 and 0 < y x2 ,
1
f ()(x, y) = x2 1 (y 1) if x = x2 and 1 > y x2 ,
e
0 if x 6= x2 and y = x2 ,
.5 if x2 6= x 6= y 6= x2 or y = 0 or y = 1.
i i
i i
i i
K23798 2015/2/2
i i
The previous theorem states that any alternative at the median position
for players with single peaked preference profiles is an element in the fuzzy
maximal subset and constitutes the fuzzy core. An element in the fuzzy core
will be chosen by players by any partial fuzzy simple rule.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Let x (0, 1). Then 1 (x) = {y X | (x, y) > 0} = [0, x) which is open
in X. Also 1 (0) = which is open and 1 (1) = [0, 1] which is open in
X. Thus is lower continuous. Let Y be a finite subset of X. If 1 / Y, then
x X such that (x, y) > 0 y Y. Suppose 1 Y. Then (1, y) > 0
y Y. Thus satisfies condition F. Now Supp() = {(1, 1)} ({(x, y)
X X | x > y}. There does not exist t [0, 1] such that Supp() = t since
(x, y) approaches 0 as x approaches 0. Note however that MM (, 1X )(1) =
{t [0, 1] | (1, y) t y X} = 1 and MM (, 1X )(x) = 0 x [0, 1) since
it is not the case (x, y) t y X; only for those y < x does (x, y) t .
Also, MG (, 1X )(1) = {t [0, 1] | (y, 1) t (1, y) y X} = 1 and
MG (, 1X )(x) = {t [0, 1] | (y, x) t (x, y) y X} = 0 x [0, 1)
since (x, y) = 0 if y > x. Let x X. Let t [0, 1]. Then
if x < t,
1
t (x) = [0, x) if x [t, 1), t < 1,
[0, 1] if t = 1 and x = 1.
Since , [0, x), and [0, 1) are open in X, t is lower continuous t [0, 1). Let
t [0, 1]. Then (1, y) t t [0, 1] and y X. Hence 1 M (t , X) and so
M (t , X) 6= t [0, 1]. In fact, M (t , X) = {1} t [0, 1].
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
For the crisp case, Theorem 6.4.24 corresponds to [2, Theorem 4.3, p. 100]
for voting rules. With single-peaked preferences, the core outcomes of f and
fL(f ) are the same.
Let L FP(N ). Let LSupp = {Supp() | L}. Let LSupp (fe) =
{Supp() | L(fe)}. Define K(L) = L Supp().
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Proof. By Lemmas 6.4.30 and 6.4.31, fe() is lower continuous and semi-
convex. The desired result is now immediate from Theorem 6.4.23. (Recall
that X is assumed to be convex and compact.)
Lemma 6.4.29 shows that for a partial fuzzy simple rule fe, the Nakamura
number s(fe) is greater than or equal to 3. The Nakamura number is the num-
ber of winning coalitions in a noncollegial subfamily. Hence as the number of
these coalitions increases past three, a fuzzy core will exist in two-dimensional
spatial models. To the extent that the number of players potentially increases
the number of winning coalitions, an increase in their numbers will result in
a greater likelihood of a core.
We conclude the chapter with a short history of the median voter theorem.
Majoritarian voting is an ancient method of group decision making. There is
no clear statement of the median voter theorem until around 1950. Aristotles
analysis of political decision making written in 330 B.C. made no mention of
a pivotal or decisive voter. Condorcet (1785) discovered the idea of a pivotal
voter and noted how the accuracy of decisions can be improved by majority
decisions, but made no clear statement of the median voter theorem. The
median voter theorem was first stated by Black (1948) and extended by Downs
[9] to representative democracy (1957).
i i
i i
i i
K23798 2015/2/2
i i
6.5 Exercises
1. [14] Let be an FWPR. Then , the strict preference relation strict
preference relation with respect to , is said to be partial if, for all
x, y X, (x, y) > 0 (y, x) = 0. An FWPR on X is regularly
acyclic if for all {x1 , x2 , x3 , . . . , xn1 , xn } X, (x1 , x2 ) (x2 , x3 )
. . . (xn1 , xn ) > 0 implies (xn , x1 ) = 0. Let be an FWPR. If is
regular, then prove that the following properties hold:
(1) is max-min transitive implies is partially quasi-transitive.
(2) is weakly transitive implies is partially quasi-transitive.
(3) is partially quasi-transitive implies is regularly acyclic.
If (x, y) P(), then prove that gL ()(x, y) > gL ()(y, x). If strict pref-
erences are partial (regular), then L is partial (regular).
4. [14] Let D F(N ) F(N ) be such that Supp() Supp() for all
(, ) D. Let FRn and set R() = {(x, y) X X | (, )
i i
i i
i i
K23798 2015/2/2
i i
and f is called a partial fuzzy voting rule if for all FRn and all
x, y X,
f()(x, y) > 0 gD(f) ()(x, y) > 0.
(2) f is called a regular fuzzy simple rule if for all FRn and all
x, y X,
6. [14] Let D F(N ) F(N ) be such that Supp() Supp() for all
(, ) D. Let FRn and set R() = {(x, y) X X | (, )
i i
i i
i i
K23798 2015/2/2
i i
1 if (x, y)
/ Symm(R())
{{j (x, y) | j Supp()} | (, ) D, i (x, y) > 0
i Supp(), and j (y, x) = 0 j Supp()}
gD ()(x, y) =
if (x, y) R()
{{j (x, y) | j Supp()} | (, ) D, i (y, x) > 0
i Supp(), and j (x, y) = 0 j Supp()}
if (x, y) Symm(R())\
R().
and f is called a partial fuzzy voting rule if for all F Rn and all
x, y X,
f()(x, y) > 0 gD(f) ()(x, y) > 0.
8. [14] Show that the following fuzzy preference aggregation rule is mono-
tonic and decisive, but not neutral. Let X = {w, x, y, z} and N = {1, 2}.
Suppose that is regular. Let be the lexicographical (alphabetical)
order of X. Then is irreflexive, complete, and asymmetric. Define the
fuzzy aggregation rule fe : FRn FR by for all a, b X and for all
FRn ,
1 if a = b or i (a, b) > 0 i N,
if a b, not [i (a, b) > 0 i N ]
and not [i (b, a) > 0 i N ],
fe()(a, b)) =
if a b, not [i (a, b) > 0 i N ]
and not [i (b, a) > 0 i N ],
0 if i (b, a) > 0 i N,
i i
i i
i i
K23798 2015/2/2
i i
9. [14] Show that the following fuzzy preference aggregation rule is neutral
and monotonic, but not decisive. Let X = {x, y, z} and N = {1, 2}. Sup-
pose that is regular. Define the fuzzy aggregation rule fe : FRn FR
by for all a, b X and for all FRn ,
1 if a = b or i (a, b) > 0 i N,
if P (b, a; ) = , not [i (a, b) > 0 i N ]
and not [i (b, a) > 0 i N ],
fe()(a, b)) =
if P (b, a; ) 6= , not [i (a, b) > 0 i N ]
and not [i (b, a) > 0 i N ],
0 if i (b, a) > 0 i N,
10. [14] Show that the following fuzzy preference aggregation rule is de-
cisive and neutral, but not monotonic. Let X = {x, y, z} and N =
{1, 2, 3, 4, 5}. Suppose that is regular. Define the fuzzy aggregation
rule fe : FRn FR by for all a, b X and for all FRn ,
1 if a = b or i (a, b) > 0 i N,
if |P (b, a; )| ( n2 , n 1), not [i (a, b) > 0 i N ]
and not [i (b, a) > 0 i N ],
fe()(a, b)) =
if |P / ( n2 , n 1), not [i (a, b) > 0 i N ]
(b, a; )|
and not [i (b, a) > 0 i N ],
0 if i (b, a) > 0 i N.
6.6 References
1. K. J. Arrow, Social Choice and Individual Values, John Wiley and Sons,
New York, 1951.
2. D. Austen-Smith and J. S. Banks, Positive Political Theory I: Collec-
tive Preference, Michigan Studies in Political Science, The University of
Michigan Press, 2000.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 7
Rationality
209
i i
i i
i i
K23798 2015/2/2
i i
210 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
Let C (i) denote the set of all PCFs satisfying condition (i) of Definition
7.1.3, i = 1, . . . , 9.
An equivalent way of defining some of the PCFs introduced in Definition
7.1.3 is by using the following approach. Let be a fuzzy relation. Recall that
(2) (x, y) = 1 (y, x) for all x, y X. Given B P (X), C(B, ) under the
i i
i i
i i
K23798 2015/2/2
i i
212 7. Rationality
max-MF rule is the same as C(B, (2) ) under the min-mA rule. Also, C(B, )
under the max-mF rule is the same as C(B, (2) ) under the min-MA rule.
The class of max-mF PCFs was introduced by Dutta, Panda, and Pat-
tanaik [15], and Switaiski [30]. The min-MA PCFs are the same as Nurmis
[20],
If is strongly complete, then the max-mF rule here is the same as a choice
rule discussed by Switalski [29],
Proposition 7.1.5 below shows that max-mD PCFs are the same as certain
PCFs introduced by Orlovsky [21]. An interesting discussion of the intuition
underlying max-mF PCFs and Orlovskys PCFs can be found in [29].
Let B P (X) and FC(R). Define the fuzzy subset OV (B, ) of B
by for all x B,
i i
i i
i i
K23798 2015/2/2
i i
We also have for all x T (B, ), mD(B, )(x) 0 and for all x B\T (B, ),
mD(B, )(x) < 0. If |T (B, )| 2, then for all x T (B, ), mD(B, )(x) = 0.
Hence
By (7.1) and (7.2), it follows that when T (B, ) 6= , a PCF is max-mD if and
only if it is max-OV.
Suppose that T (B, ) = . Then it can be shown that for all x B,
Hence
i i
i i
i i
K23798 2015/2/2
i i
214 7. Rationality
and
0 (x, x) = 0 (y, y) = 0 (z, z) = 1,
0 (x, y) = s, 0 (x, z) = t,
0 (y, x) = r, 0 (y, z) = r,
0 (z, x) = q, 0 (z, y) = s.
where 1 > t > q > s > r > 0. Then (x, y) > (y, x), (x, z) > (z, x),
0 (x, y) > 0 (y, x), and 0 (x, z) > 0 (z, x). However, if C is max-MF, or max-
SF, or max-MD, or max-SD, then x / C(B, ). Thus RPWD and RPSD do
not hold. If C is min-MA or min-SA, then x / C(B, 0 ). Thus neither RPWD
nor RPSD hold.
Consider (2). Let FC(R)0 = FC(R). Suppose C is max-MF, or max-SF, or
min-SA, or max-MD, or max-SD. Since FT (R) FC(R), it follows from (1)
that both RPWD and RPSD do not hold. Therefore, we have only to consider
the case where C is either max-mF or min-mA. Let B = {x, y, z}. Let
FC(R) be such that
where 1 > t1 > t2 > t3 > t4 > t5 > 0. Then (x, y) > (y, x) and (x, z) >
(z, x). However, if C is either max-mF or min-mA, then x / C(B, ). Thus
RPWD and RPSD do not hold.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 7.1.8 (Barrett, Pattanaik, and Salles [9]) (1) If C C (8) , then C
satisfies both RPWD and RPSD.
(2) If FC(R) F T (R) and C C (2) C (5) C (8) , then C satisfies both
RPWD and RPSD.
Proof. (1) The proof here is clear.
(2) Let FC(R)0 FT (R). We show that if C C (2) C (5) , then C satisfies
RPWD; the result will then follow from the fact that RPWD implies RPSD
and part (i).
Let B P (X), FC(R)0 , and x B be such that
i i
i i
i i
K23798 2015/2/2
i i
216 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
where 1 > t > s > 0. If C is max-mF, then x C(B, ) and so WREJ and
hence SREJ do not hold.
(b) Let 0 FT (R) FC(R)0 be such that
where 1 > t > s > r > 0. If C is min-MA or min-SA, then x C(B, 0 ) and
so WREJ and hence SREJ do not hold.
(c) Let 00 FT (R) FC(R)0 be such that
where 1 > s > r > 0. If C is min-mA, then x C(B, ) and so SREJ fails
to hold. This completes the proof of (i).
(2) Let FC(R)0 = FC(R). Since FT (R) FC(R), from part (1), it follows
that if C is max-mF, or min-mF, or min-MA, or min-SA, then WREJ does
not hold for C. We now show that if C is max-SF or max-mD, then WREJ
does not hold. Let B = {x, y, z, w}. Let FC(R)0 be such that its restriction
to B B is given by
x y z w
x 1 s s s
y t 1 r s
z t s 1 r
w t r s 1
where 1 > t > s > r > 0. If C is max-SF or max-mD, then x C(B, R) and
so WREJ fails to hold
Given part (i) and given that FC(R)0 = FC(R) and C is max-mD, then C
violates WREJ, it follows that if FC(R)0 = FC(R) and C K(max-MF, max-
mF, max-SF, min-MA, min-mA, min-SA, max-mD), then C violates SREJ.
i i
i i
i i
K23798 2015/2/2
i i
218 7. Rationality
Theorem 7.1.12 (Barrett, Pattanaik, and Salles [9]) (1) If C C (7) C (9) ,
then C satisfies SREJ.
If C C (1) C (5) C (7) C (9) , then C satisfies WREJ.
(2) If FC(R)0 FT (R) and C C (7) C (8) C (9) , then C satisfies SREJ.
If FC(R)0 FT (R) and C C (1) C (3) C (5) C (7) C (8) C (9) , then C
satisfies WREJ.
Proof. (1) If C is max-MD, then clearly C satisfies SREJ. We now show
that if C is max-SD, then C satisfies SREJ.
Suppose that C is max-SD. Let FC(R)0 , B P (X), and x B be
such that for all y B\{x}, (x, y) (y, x). Suppose there exists y B\{x}
such that (x, y) < (y, x). Suppose that x C(B, ). Since x C(B, )
and SD(B, )(x) is negative, it follows that SD(B, )(y) < 0 for all y B.
Let B = {z1 , z2 , . . . , zn } and for all distinct i, j {1, 2, . . . , n}, let dij =
(xi , xj ) (xj , xi ). Then for all distinct i, j {1, 2, . . . , n}, dij + dji = 0.
Hence
(d12 + d13 + + d1n ) + (d21 + d23 + + d2n )+
+ (dn1 + dn2 + + dn,n1 )
= 0.
i i
i i
i i
K23798 2015/2/2
i i
000 (y, z) = t, 000 (z, y) = s, 000 (u, v) = 1 (u, v) X X\{(y, z), (z, y)}
i i
i i
i i
K23798 2015/2/2
i i
220 7. Rationality
Theorem 7.1.15 (Barrett, Pattanaik, and Salles [9]) (1) If C C (1) C (3) ,
then C satisfies UF.
(2) If C C (1) C (2) C (3) C (4) C (7) , then C satisfies LF.
(3) If FC(R)0 FT (R) and C C (2) C (3) C (6) C (8) C (9) , then C
satisfies UF.
(4) If FC(R)0 FT (R) and C 9i=1 C (i) , then C satisfies LF.
Proof. That (1) and (2) hold is clear.
Consider (3) and (4). From part (1), it follows that if FC(R)0 F T (R)
and C is max-mF or max-SF, then C satisfies UF. We next show that if
FC(R)0 FT (R) and C is min-SA, or max-mD, or max-SD, then C satisfies
UF.
Let B P (X), F C(R)0 F T (R) and x D(B, ), and suppose
z C(B, )\{x}. Suppose C is min-SA. By the max-min transitivity of , we
have since (x, z) = 1 that
(w, z) (w, x) for all w B\{x, z}. (7.7)
For z C(B, ), we have
X X
(w, z) (w, x). (7.8)
wB\{z} wB\{x}
From (7.7) and (7.8), and since (x, z) = 1, it follows that (z, x) = 1. Hence
by the max-min transitivity of and the fact that (x, w) = 1 for all w B,
it follows that (z, x) = 1 for all w B. Thus z D(B, ).
Let C be max-mD. For x D(B, ), it is clear that mD(B, )(x) 0. Since
z C(B, ), it follows that mD(B, )(z) 0 and hence (z, x) (x, z) = 1.
Therefore, by the max-min transitivity of ,we have (z, w) = 1 for all w B
since (x, w) = 1 for all w B. Thus z D(B, ).
Let C be max-SD. By the max-min transitivity of and the fact that
(x, z) = 1, we have
(w, z) (w, x) for all w B. (7.9)
Also,
(z, w) (x, w) = 1 for all w B. (7.10)
From (7.9) and (7.10), it is clear that if (z, w) < 1 for some w B,
then SD(B, )(z) < SD(B, )(x), which contradicts z C(B, ). Therefore,
(z, w) = 1 for all w B, i.e., z D(B, ).
Now consider LF. Given FC(R)0 FT (R), by part (i), if C is max-MF, or
max-mF, or max-SF, or min-MA, or max-mD, then C satisfies LF. We have
only to consider the cases where C C (5) C (6) C (7) C (9) .
Let B P (X) and FC(R)0 FT (R), and let x D(B, ) and
z C(B, ). Then by the max-min transitivity of ,we have since (x, z) = 1
that
(w, z) (w, x) for all w B. (7.11)
i i
i i
i i
K23798 2015/2/2
i i
Also,
(z, w) (x, w) = 1 for all w B. (7.12)
From (7.11) and (7.12), it follows that
RP W D RP SD W REJ SREJ UF LF
max-M F
max-mF ~ ~
max-SF ~
min-M A
min-mA ~ ~ ~
min-SA ~ ~
max-M D ~
max-mD ~ ~ ~
max-SD ~ ~
i i
i i
i i
K23798 2015/2/2
i i
222 7. Rationality
For all S E, C(S)(x) represents the degree to which x belongs to the set
of chosen alternatives when the available set of alternatives is S.
In the previous definition of a fuzzy choice function, every nonempty avail-
able set S has an element which is chosen to a positive extent. A natural
question arises as to whether it should be assumed that every such available
i i
i i
i i
K23798 2015/2/2
i i
The notion of what has been called weak transitivity appeared in Ponsard
[25].
and
(x, y) = (x, y) (y, x).
The definition of a fuzzy strict preference relation given here follows the
ideas presented in Orlovsky [21]. It can be shown that the major results of
this section are not sensitive to the choice between alternative derivation rules,
[14]. Recall that this choice of is (3) of Chapter 1.
i i
i i
i i
K23798 2015/2/2
i i
224 7. Rationality
Definition 7.2.5 Let C be a fuzzy choice function. Define the fuzzy relation
e on X by for all x, y X,
In the crisp case, Definition 7.2.5 reduces to Arrows [2] definition of re-
vealed preference. The results of this section can be established without ex-
plicitly obtaining the weak preference relation corresponding to e.
Definition 7.2.6 Let be the fuzzy preference relation revealed by the choice
function C. The image of C is defined to be the function C b : E FP (X)
f
such that S E and x S,
C(S)(x)
b = {(x, y) | y S}.
C(S)(x) = C(S)(x).
b
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
226 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
Example 7.2.14 We show that SAFRP1 does not imply normality: Let X =
{x, y, z} and r, s, t real numbers such that 0 < r < s < t < 1. Define the fuzzy
choice function C as follows:
C({x})(x) = C({y})(y) = C({z})(z) = 1,
C({x, y})(x) = 1, C({y, z})(y) = 1, C({x, z})(z) = 1,
C({x, y})(y) = t, C({y, z})(z) = s, C({x, z})(z) = 0,
C({x, y, z})(x) = 1, C({x, y, z})(y) = r, C({x, y, z})(z) = 0.
i i
i i
i i
K23798 2015/2/2
i i
228 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
Also,
(x, z) < (y, z). (7.15)
Let S = {x, y, z}. Then
C(S)(x)
b = (x, y) (x, z) = (x, z), by (7.14).
i i
i i
i i
K23798 2015/2/2
i i
230 7. Rationality
C(S)(y) t, (7.16)
and
(x, y) t. (7.17)
By (7.16), (y, z) t z S. Hence by the max-min transitivity of , (7.17)
implies (x, z) t z S. Thus C(S)(x)
b t. Hence by normality C(S)(x)
t. Consequently, FC3 by rationality.
Theorem 7.2.22 shows that fuzzy congruence constitutes a full characteri-
zation of rationality of a fuzzy choice function.
We next examine the relationship between fuzzy congruence and other
fuzzy revealed preference conditions.
If (y, x) (y, z), C(S)(y) < (y, z) so that FC1 is violated and consequently
so is rationality. Thus (x, y) > (y, z). If (y, x) (x, z), then fails to be
intensely transitive. Hence, (y, x) > (x, z). If (y, z) < (x, z), then max-
min transitivity is violated. Hence (y, z) (x, z). Thus
Since z is dominant in S, the strict inequality violates FC1 and thus rationality.
i i
i i
i i
K23798 2015/2/2
i i
Example 7.2.26 [5] We show that WAFRP2 does not imply SAFRP2. Let
X = {x, y, z} and s be a real number such that 12 < s < 1. Define the fuzzy
choice function C as follows:
C({x})(x) = C({y})(y) = C({z})(z) = 1,
C({x, y})(x) = 1, C({y, z})(y) = 1, C({x, z})(x) = 1,
C({x, y})(y) = s, C({y, z})(z) = s, C({x, z})(z) = 1,
C({x, y, z})(x) = s, C({x, y, z})(y) = s, C({x, y, z})(z) = s.
The fuzzy revealed preference relation is as follows: (x, x) = (y, y) =
(z, z) = 1. (x, y) = (y, z) = (x, z) = (z, x) = 1, (y, x) = (z, y) = s.
Thus e(x, y) = e(y, z) = 1 s,
e(x, z) =
e(y, x) = e(z, y) =
e(z, x) = 0.
For t > 1 s, there does not exist (u, v) such that e(u, v) t. If t is such
that 0 < t 1 s, e(x, y) t, (y, x) 1 t; e(y, z) t, (z, y) 1 t.
There does not exist any other (u, v) such that e(u, v) t. Thus WAFRP2
holds. However, (x, z) t and (z, x) = 1 > 1 t. Hence SAFRP2 does not
hold.
i i
i i
i i
K23798 2015/2/2
i i
232 7. Rationality
Example 7.2.27 [5] We show that SAFRP2 does not imply normality of the
choice function. We modify Example 7.2.26 by letting C({y, z})(y) = s. The
revealed preference relation is now as follows: (x, x) = (y, y) = (z, z) = 1;
(x, y) = (x, z) = (z, x) = 1 s; (y, x) = (y, z) = (z, y) = s.
The e relation is now such that e(x, y) = 1 s, e(x, z) = e(z, x) =
e(y, z) = e(y, x) =
e(z, y) = 0.
It follows that WAFRP2 holds. Since there is no ordered triple (x, y, z)
and no t, 0 < t 1, such that e(x, z) < t, but e(x, y) t and e(y, z)
t, SAFRP2 also holds. The choice function C is not normal since for S =
{x, y, z}, C(S)(x)
b = (x, y) (x, z) = 1 1 = 1, but C(S)(x) = s.
We have demonstrated that FC Rationality SAFRP2 WAFRP2
and neither of the last two arrows can be reversed.
We now consider the case where there is at least one unambiguous choice,
i.e., S E, x S such that C(S)(x) = 1. Note that the choice function
continues to be fuzzy.
We note that FC3 implies FC1. This holds since if x is dominant in S E,
but C(S)(y) < (y, x) = t, then C(S)(x) = 1 t, (y, x) = t. However,
C(S)(y) < t so that FC3 is violated. Similarly, it can be shown that FC3
implies FC2. Thus FC and FC3 are now equivalent.
Conditions A and B are now trivially satisfied by every well-defined fuzzy
choice function since if x is dominant in S E, C(S)(x) = 1 so that by
definition, (x, y) = 1 y S.
A fuzzy preference relation is called a fuzzy ordering if it is reflexive,
strongly complete, and intensely transitive.
i i
i i
i i
K23798 2015/2/2
i i
Thus e(x, y) > 1 (y, x). Let e(x, y) = t. If WAFRP2 holds, then (y, x)
1 t < (y, x), a contradiction.
For the weak transitivity of , it suffices in view of Lemma 7.2.16 to show
in addition that WAFRP2 implies FC2. If FC2 does not hold, S E and
x, y S such that y is dominant in S, (x, y) (y, x), but x is not dominant
in S. Therefore, C(S)(y) = 1 > C(S)(x). Hence C(S)(y) C(S)(x) = 1
C(S)(x) = t > 0. Thus e(y, x) t. If WAFRP2 holds, then (x, y) 1t < 1.
However, since C(S)(y) = 1, (y, x) = 1. Hence (x, y) (y, x) = 1, a
contradiction. Thus WAFRP2 does not hold.
By Proposition 7.2.30, the Theorem 7.2.31 holds for SAFRP2. Since a well-
defined choice function guarantees that is reflexive and complete, Theorem
7.2.31 shows that WAFRP2 and SAFRP2 come close to fulfilling the require-
ments of rationality. However, Example 7.2.32 shows that a gap remains.
Example 7.2.32 We show that SAFTRP2 (hence, WAFRP2) does not imply
max-min transitivity of . We modify Example 7.2.14 by letting C({x, y})(y) =
0.4, leaving other specifications unchanged. Then (y, x) = 0.4, but for all
other ordered pairs (u, v) X X, (u, v) remains the same as in Example
e(x, y) = 1 (y, x) x, y X, which is a suf-
7.2.14. It follows easily that
ficient condition for WAFRP2 and hence SAFRP2. However, 0 = (z, x) <
(z, y) (y, x) = 0.5 0.4 = 0.4. Hence is not max-min transitive.
When at least one unambiguous choice is assumed, we have FC Ratio-
nality SAFRP2 WAFRP2.
7.3 Exercises
1. Show that all the PCFs introduced in Definition 7.1.3 satisfy the prop-
erty that for all F(R)0 and for all x, y X, (x, y) (y, x) (resp.
(y, x) (x, y)) implies x C({x, y}).
Let C : P (X)F(R)0 P (X) be a PCF. Define J 1 and J 2 as follows:
(i) C J 1 if and only if there exists 0.5 such that for all B P (X)
and all F(R)0 ,
(ii) C J 2 if and only if for all B P (X) and all F(R)0 , C(B, ) =
T (B, ).
i i
i i
i i
K23798 2015/2/2
i i
234 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
7.4 References
1. M. M. Aizerman and A. V. Malishevski, General theory of best-variant
choices. Some aspects, IEEE Trans. Automat. Control, 26 (1981), 1030
1040.
2. K. J. Arrow, Rational choice functions and ordering, Economica, NS 26
(1959) 121127.
3. A. K. Banerjee, On deriving strict preference and indifference from a
fuzzy weak preference relation, Mimeo, Dept. of Economics, Calcutta
Univ. (1991).
4. A. K. Banerjee, Rational choice under fuzzy preferences: The Orlovsky
choice function, Fuzzy Sets and Systems, 54 (1993) 295299.
5. A. K. Banerjee, Fuzzy choice functions, revealed preference and ratio-
nality, Fuzzy Sets and Systems, 70 (1995) 3143.
i i
i i
i i
K23798 2015/2/2
i i
236 7. Rationality
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 8
239
i i
i i
i i
K23798 2015/2/2
i i
We point out how the use of an involutive fuzzy complement can be used to
obtain the results.
It was shown in Chapter 5 that Arrows theorem, [2, 30, 5, 32], could be
avoided by using fuzzy preferences and by using a weak version of transitivity.
In [6], it is shown that the choice of definitions for indifference and strict
preference associated with a given fuzzy preference relation also have Arrowian
implications. In [26], it is shown that the dictatorship theorem in [6] can
be strengthened to cover any version of transitivity for fuzzy preferences no
matter how weak, and that this dictatorship result holds for any regular strict
preference including the one used in [6].
In [16], the notion of intuitionistic fuzzy relations was introduced to study
Arrowian like conditions. In Chapter 5, other types of independence of irrel-
evant alternative conditions were introduced and it was shown that they can
be profitably used in the examination of Arrows theorem. In this chapter, we
extend some known Arrowian results involving fuzzy set theory to results in-
volving intuitionistic fuzzy sets. The interested reader can find results dealing
with vague preference relations to study Arrows Impossibility Theorem and
invariants in [1, 17, 12, 13, 14, 19, 21, 26, 31].
In this section, we lay the groundwork for our study of fuzzy Arrow results.
An intuitionistic fuzzy binary relation on X is an ordered pair h , i
of fuzzy binary relations on X such that x, y X, (x, y) + (x, y) 1.
Let c be a fuzzy complement on [0, 1]. Recall that c is called involutive if
c(c(a)) = a for all a [0, 1]. It follows that an involutive fuzzy complement
is a one-to-one continuous function of [0, 1] onto [0, 1]. Let c be an involutive
fuzzy complement of [0, 1]. Let be a fuzzy binary relation on X. Define
the fuzzy relation c on X by (x, y) X X, c (x, y) = c((x, y)). If is
a fuzzy relation on X, then h, c i is an intuitionistic fuzzy relation on X.
Let FR denote the set of all fuzzy relations on X and define the function
f : FR FR by FR, f () = c . Then f maps FR one-to-one
onto FR. We showed in Chapter 3 how a result concerning c can thus be
determined from a result concerning as long as the definitions involving c
are dual to those of with respect to c. This follows since a result concerning
c is a general result for fuzzy preference relation due to the fact that f is
one-to-one onto. We do not use this approach in this chapter. The proofs of
results in Sections 8.1-8.4, can be compared to those in Chapter 3 to see their
dual nature. We explain the situation further in Section 8.5.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
(x, y) {i (x, y) | i N },
(x, y) {i (x, y) | i N }.
i i
i i
i i
K23798 2015/2/2
i i
It is clear that if i and i are reflexive and complete for all i N, then
fe() in Definition 8.1.9 is reflexive and complete.
i i
i i
i i
K23798 2015/2/2
i i
0 0
j (x, y) > 0) or (2) (j (y, x) > 0 and j (y, x) = 0), where strict preference
is of type (3) . Then (x, y) = 0 ( (x, y) 0 (y, x)) and
0 0
0 (x, y) 0 (y, x)
Xn
= (wi 0i (x, y) wi 0i (y, x))
i=1
n
X
= (wi i (x, y) wi i (y, x)) + wj 0j (x, y) wj 0j (y, x)
i=1,i6=j
Xn
= (wi i (x, y) wi i (y, x)) wj (j (x, y) j (y, x))
i=1
+wj (0j (x, y) 0j (y, x))
= wj (j (x, y) j (y, x)) + wj (0j (x, y) 0j (y, x))
> 0,
where the inequality holds if either (1) or (2) hold. Hence fe satisfies PR with
respect to (3) . The desired result for (1) follows from Corollary 8.1.8.
0
We now consider the case for (3)
. Suppose that either (1) (j (x, y) = 1
0 0
and j (x, y) < 1) or (2) (j (y, x) < 1 and j (y, x) = 1), where strict
preference is of type (3) . Then (3) (x, y) = 1 (1 0 (x, y) + 0 (y, x)) and
0
as above
1 0 (x, y) + 0 (y, x)
= 1 [wj (j (x, y) j (y, x)) + wj (0j (x, y) 0j (y, x))].
1 0 (x, y) + 0 (y, x)
= 1 [wj [1 + j (x, y) j (y, x)] + wj (1 + 0j (x, y) 0j (y, x))]
0
= 1 [wj j (x, y) wj j (x, y)] < 1.
1 0 (x, y) + 0 (y, x)
= 1 [wj [1 + j (x, y) j (y, x)] + wj (1 + 0j (x, y) 0j (y, x))]
0
= 1 [wj j (y, x) + wj j (y, x)] < 1.
Hence fe satisfies PR with respect to (3) . The desired result for (1)
follows from Corollary 8.1.8.
i i
i i
i i
K23798 2015/2/2
i i
and
n
X n
X
(2) (x, y) = (y, x) = wi i (y, x) = wi i (y, x).
i=1 i=1
i i
i i
i i
K23798 2015/2/2
i i
and
n
X n
X n
X
(x, z) = wi i (x, z) wi i (x, y) + wi i (y, z)
i=1 i=1 i=1
= (x, y) + (y, z).
i i
i i
i i
K23798 2015/2/2
i i
Theorem 8.2.2 Let strict preference be defined by (3) . Then there exists a
nondictatorial fuzzy aggregation rule fe : H2n H2 satisfying IIA1, PR, and
PC.
Proof. Let fe be the fuzzy aggregation rule defined in Definition 8.1.9.
Clearly, fe is not dictatorial. By Propositions 8.1.10, 8.1.13, and 8.1.16, it only
remains to show that fe satisfies IIA1. However, this is immediate.
Proof. Suppose (x, y) > (y, x). Let (x, y) = si . Then si > (y, x).
Hence (y, x) / (0 )ti . Now (x, y) ( )si so (x, y)
/ ( )si . Thus (y, x)
0 0 0
( )ti . Hence (x, y) ti > (y, x). A similar argument holds for and
0 .
0
i N fe ()e{x,y} fe ( )e{x,y} ].
In the following, we let t and tP denote functions from FRn into (0, 1].
n
Some examples Pn of t are t () = { i=1 wi i (x, y)
Pn| x, y X, x 6= y} and
t () = { i=1 wi i (x, y) > 0 | x, y X}, where i=1 wi = 1 and wi > 0,
i = 1, . . . , n.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 8.2.7 Let strict preference be defined by (1) or (3) . Let (ff , f )
e
be the fuzzy aggregation rule defined in Definition 8.2.6. Suppose Im(t ),
Im(t ) (0, 1). Then (ff
, f ) is independent of irrelevant alternatives IIA3.
e
n
Definition 8.2.8 Let (ff , f ) : T
e T be a fuzzy aggregation rule. Then
(f , f ) is said to be weakly Paretian with respect to a given strict preference
f e
if x, y X, (h1 , 1 i, . . . , hn , n i) T n , i (x, y) > 0 i N implies
(x, y) > 0 and i (x, y) < 1 i N implies (x, y) < 1.
i i
i i
i i
K23798 2015/2/2
i i
Thus
1 > (x, y) = (x, y) < (y, x), (8.2)
1 > (y, z) = (y, z) < (z, y) (8.3)
and so
1 = (x, z) (x, z) (z, x) (8.4)
Suppose
(x, y) (y, z) = (x, y) (8.5)
Then
(x, z) (x, y). (8.6)
Now
(y, x) (y, z) (z, x). (8.7)
But since (x, y) (y, z) = (x, y) and (y, x) (y, z),we have (y, x)
(x, y), contradicting (8.2). Similarly, (8.4), (8.6) and (y, x) (z, x)
imply (y, x) (x, y). Hence (y, z) (z, x) < (y, x) contradicting
(8.7).
Suppose
(x, y) (y, z) = (y, z). (8.8)
Then
(x, z) (y, z). (8.9)
By the max-min transitivity of ,
i i
i i
i i
K23798 2015/2/2
i i
or
(z, y) (x, y). (8.11)
Now (8.10), (8.4), and (8.9) imply (z, y) (y, z). Similarly, (8.11) and
imply (z, y) (y, z). However, this contradicts (8.3).
Let FR denote the set of all intuitionistic fuzzy binary relations on X
which are reflexive and complete. Let FRT r = {h , i F R | h , i is
transitive}. We say that a fuzzy aggregation rule (ff , f ) is max-min transi-
e
tive if ff
( ) and f ( ) are max-min transitive for every intuitionistic fuzzy
e
preference profile (h1 , 1 i, . . . , hn , n i) T n , where = (1 , . . . , n )
and = (1 , . . . , n ).
i i
i i
i i
K23798 2015/2/2
i i
Theorem 8.2.14 There exists a weakly Paretian (with respect to (2) ) fuzzy
n
aggregation rule fe = (ff , f ) : T T that is reflexive, complete, max-min
e
transitive, satisfies IIA3 and is such that every member of N is a dictator.
Proof. Define fe : T n T by (h1 , 1 i, . . . , hn , n i) T n , x, y
X,
1 if x = y,
ff
( )(x, y) =
t ( ) if x 6= y,
where 0 < t ( ) < 1 and
0 if x = y,
fe ( )(x, y) =
t ( ) if x 6= y,
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
z X\{x, y}, xD
e y xD z. (8.12)
+
i (y, z) = 00vi (y, z) and + 00
vi (z, y) = vi (z, y) i N,
+
vi (y, x) < 1 i N,
+ +
vi (x, z) < 1 i Supp() and j (z, x) < 1 j N \Supp().
+
Then vi (y, z) < 1 i Supp(). Since xD z, + (x, z) < 1. Since fe is weakly
Paretian, + (y, x) < 1. Since fe is partially quasi-transitive, + (y, z) < 1. For
i i
i i
i i
K23798 2015/2/2
i i
00
i N and fe is IIA2, Cosupp(+ +
e{y,z} ) = Cosupp( e{y,z} ). Since (y, z) <
+ + 00 00
1, (y, z) < 1 and (z, y) = 1. Hence (y, z) < 1 and (z, y) = 1.)
(For = (2) , we write: Since + 00
vi e{y,z} = vi e{y,z} i N and f is IIA4,
e
00
Supp(+ e
{y,z} ) = Supp( e
{y,z} ). Since +
(y, z) < 1, +
(z, y) > 0. Hence
00 (z, y) > 0.) Thus 00 (y, z) < 1 and so yD z. Hence since z is arbitrary in
X\{x, y}, the preceding two steps yield
z
/ {x, y}, xD
e y yD z. (8.13)
e y xD v (by (4.1)) xD
xD e v vD w (by (8.13)) with v replacing y.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
(x, z) and (x, y) (y, x), we have (z, y) < (y, z), a contradiction.
Hence (x, z) (z, x).
Let FCR denote the set of all reflexive and complete intuitionistic fuzzy
preference relations that are consistent.
Proof. The proof for can be found in Theorem 5.1.22. Since fe is con-
sistent, fe is partially quasi-transitive by Corollary 8.3.17. By Lemma 8.3.11,
it suffices to show that i N, x, y X such that {i} is semidecisive for
x against y. (That is, with Supp() = {i} such that is semidecisive for
x against y.) This follows by Lemma 8.3.11 because then is decisive for x
against y for all x, y X, where Supp() = {i}. Hence h , i FRn ,
x, y X, i (x, y) > 0 implies (x, y) > 0 and i (x, y) < 1 implies
(x, y) < 1.
Since fe is weakly Paretian, a decisive for any pair of alternatives,
namely = 1N . For all (u, v) X X, let m(u, v) denote the size of the
smallest |Supp()| for semidecisive for u against v. Let m = {m(u, v) |
(u, v) X X}. Without loss of generality suppose is semidecisive for x
against y with |Supp()| = m. If m = 1, the proof is complete. Suppose m > 1.
Let i Supp(). Consider any fuzzy profile (h1 , 1 i, . . . , hn , n i) T n
such that
i (x, y) > 0, i (y, z) > 0, i (x, z) > 0,
j Supp()\{i}, j (z, x) > 0, j (x, y) > 0, j (z, y) > 0,
k / Supp(), k (y, z) > 0, k (z, x) > 0, k (y, x) > 0.
i i
i i
i i
K23798 2015/2/2
i i
Definition 8.4.1 Let i N and t (0, 1]. Define the fuzzy subset it of N by
j N, it (j) = t if j = i and it (j) = 0 otherwise. If g : T n {it | i N,
t (0, 1]}, then g is called a representation rule for T.
i i
i i
i i
K23798 2015/2/2
i i
and
Cosupp(i e{x,y} ) = Cosupp(0i e{x,y} )
i N implies Cosupp(Supp(g()) e{x,y} ) = Cosupp(Supp(g(0 ) e{x,y} ).
(3) g is called independent of irrelevant alternatives IIA3 if , 0
T , x, y X, i e{x,y} 0i e{x,y} i N implies Supp(g()) e{x,y}
n
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Since i has a semiveto for x against y, not (y, x) < 1. Since fe is complete,
(x, y) < 1 and in fact (x, y) < 1. Since fe is weakly Paretian, (y, z) < 1.
Since fe is partially quasi-transitive, (x, z) < 1. By IIA2, i has a veto for
x against z. (We have (x, z) < 1 for this , but we need independence
to get it for all . Let 0 be such that i
0
(x, z) < 1. Then 0 (x, z) < 1 by
independence.) Thus we have
z
/ {x, y}, xVei y xVi z. (8.14)
Now since i has a veto for x against z, i has a semiveto for x against z. Thus
switching y and z in the above argument implies i has a veto for x against
y. Let 00 be any fuzzy preference profile such that i00 (y, z) < 1. Let + be a
fuzzy preference profile such that
+ + +
i (y, x) < 1, i (y, z) < 1, i (x, z) < 1,
+ +
j 6= i, j (z, x) < 1, j (y, x) < 1.
z
/ {x, y}, xVei y yVi z. (8.15)
Now i has a veto for y against z, i has a semiveto for y against z. Thus by
(8.14), i has a veto for y against x. We have (v, w) X X,
with v replacing y.
i i
i i
i i
K23798 2015/2/2
i i
Now since i has a veto for x against z, i has a semiveto for x against z. Thus
switching y and z in the above argument implies i has a veto for x against y.
Let h00 , 00 i F QRn be any fuzzy preference profile such that i
00
(y, z) > 0
i i
i i
i i
K23798 2015/2/2
i i
00 n
and i (y, z) < 1. Let h+ +
, i FQR be another fuzzy preference profile
such that
+ + +
i (y, x) > 0, i (y, z) > 0, i (x, z) > 0
+ +
j 6= i, j (z, x) > 0, j (y, x) > 0.
+ + +
i (y, x) < 1, i (y, z) < 1, i (x, z) < 1
+ +
j 6= i, j (z, x) < 1, j (y, x) < 1.
Since i has a semiveto for x against z, + (z, x) = 0 and + (z, x) = 1. Thus
+ + + +
(x, z) (z, x) and (x, z) (z, x). Since f is weakly Paretian,
e
(y, x) > 0 and (y, x) < 1. Now suppose (z, y) > 0 and + (z, y) < 1.
+ + +
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Cornelis et al. [9, Theorem 2, pp. 60-61] showed that, given a t-norm T
and a t-conorm S, (T, S) is a t-representable intuitionistic fuzzy t-norm if
a1 , a2 [0, 1], T (a1 , a2 ) 1 S(1 a1 , 1 a2 ).
It follows that T = (min, max) and J = (max, min) are respectively a
t-representable intuitionistic fuzzy t-norm and a t-representable intuitionistic
fuzzy t-conorm. Condition G defined below and introduced by Fono et al. [16]
is an important condition that holds for a t-representable intuitionistic fuzzy
t-conorm J = (S, T ).
(x) and (x) denote, respectively, the degree of membership and the
degree of non-membership of the element x in the intuitionistic fuzzy set. The
number (x) = 1 (x) (x) is called the index of the element x in X.
Clearly, if x X, (x) = 1 (x), then the IFS is a fuzzy subset of X.
An intuitionistic fuzzy binary relation represents a (weak) preference
which for x, y X can be interpreted as saying that (x, y) is the degree to
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Proposition 8.5.8 [16] Let be an IFR and its regular strict component.
Then
(1) is pos-transitive if x, y, z X,
(i)( (x, y) > (y, x) and (y, z) > (z, y)) (x, z) > (z, x)
(ii)( (x, y) < (y, x) and (y, z) < (z, y)) (x, z) < (z, x)
i i
i i
i i
K23798 2015/2/2
i i
and (i) or (ii) or (iii) or (iv) implies (x, z) (z, x), where
(y, x) > (x, y) (y, x) = (x, y)
(i) , (ii) ,
(z, y) < (y, z) (z, y) > (y, z)
(y, x) > (x, y) (x, y) = (y, x)
(iii) , (iv) .
(z, y) = (y, z) (y, z) = (z, y)
In [16, Example 5.3, pp. 21-22], it is shown that there exists an IFR such
that pos-transitivity and negative transitivity do not hold for a regular strict
component. Necessary and sufficient conditions on an IFR such that satisfy
each of these properties is also established. We now recall these results.
The following four conditions of an IFR are from [16].
For all x, y, z X,
i i
i i
i i
K23798 2015/2/2
i i
(3) satisfies condition C2 if x, y, z X,
(y, x) = (x, y) (y, x) < (x, y)
or
(z, y) < (y, z) (z, y) = (y, z)
(x, z) [2 (x, y, z), 3 (x, y, z)]
(z, x) < (x, z) .
(z, x) [2 (x, y, z), 3 (x, y, z)]
i i
i i
i i
K23798 2015/2/2
i i
Lemma 8.5.12 (Nana and Fono [24]) Let T = (T, S) and be an IFR.
(1) If T = (min, max), then satisfies C1 and C1 .
(2) If T = (min, max) and is (min, max)-transitive, then
(i) conditions C2 and T are equivalent,
(ii) conditions C2 and T are equivalent.
Proof. Suppose that T = (min, max).
(1i) We show that satisfies C1 . Let x, y, z X be such that (y, x) <
(x, y) and (z, y) < (y, z). We show that (x, z) [2 (x, y, z), 3 (x, y,
z)] and (z, x) [2 (x, y, z), 3 (x, y, z)] imply (z, x) < (x, z). It suffices
to show that conjunction (x, z) [2 (x, y, z), 3 (x, y, z)] and (z, x)
[2 (x, y, z), 3 (x, y, z) is false. We distinguish the following cases: (z, y)
(y, x) < (x, y) (y, z) or (z, y) (y, x) < (y, z) < (x, y) or
(z, y) < (y, z) (y, x) < (x, y) or (y, x) < (x, y) (z, y) <
(y, z) or (y, x) < (z, y) < (x, y) (y, z) or (y, x) < (z, y) <
(y, z) < (x, y). Since T = (min, max), each of these cases yield 3 (x, y, z)
< 2 (x, y, z). Thus the interval [2 (x, y, z), 3 (x, y, z)] is empty and then the
above conjunction is false.
(1ii) By -symmetry, we show that satisfies C1 .
(2ia) Assume T . We show that C2 holds. Let x, y, z X be such that
(y, x) = (x, y) (y, x) < (x, y)
or (8.20)
(z, y) < (y, z) (z, y) = (y, z)
It suffices to show that (x, z) [2 (x, y, z), 3 (x, y, z)] and (z, x)
[2 (x, y, z), 3 (x, y, z)] imply (z, x) < (x, z). (8.20) is equivalent to the
disjunction
i i
i i
i i
K23798 2015/2/2
i i
and 2 (x, z, y) = min( (x, z), (z, y)) = (z, y). We obtain 2 (x, z, y) =
3 (x, z, y) = (z, y). Because of hypothesis (z, y) = (x, y) = (y, x),
we have (y, x), (x, y) [2 (x, z, y), 3 (x, z, y)]. Since satisfies (z, x) <
(x, z) and (y, z) = (z, y), C2R implies (y, x) < (x, y). That con-
tradicts the equality (y, x) = (x, y). Hence, (x, z) > (z, x) is not
possible.
If (x, z) < (z, x), then the transitivity of implies
Thus
We obtain
3 (z, x, y) = min(Imin ( (x, y), (x, z)), 1 Jmax ( (x, y), (x, z)),
Imin ( (z, x), (y, x)), 1 Jmax ( (z, x), (y, x)))
= min(1, (y, x)) = (y, x)
i i
i i
i i
K23798 2015/2/2
i i
and 2 (z, x, y) = min( (z, x), (x, y)) = (x, y). Then 2 (x, z, y) = 3 (x,
z, y) = (x, y). Since (x, y) = (z, y) = (y, z), we have (y, z), (z, y)
[2 (z, x, y), 3 (z, x, y)]. Since satisfies ( (x, z) < (z, x) and (y, x) =
(x, y)), C2 gives (y, z) < (z, y). That contradicts the equality (y, z)
= (z, y). Hence, (z, x) > (x, z) is impossible.
(2ii) Since is -symmetric, the proof of the equivalence of conditions C2
and T follows from the previous result.
In the case where T = (min, max), conditions C1 and C1 hold and
have become simplified C2 and C2 . Therefore, it is important to rewrite
characterizations of the pos-transitivity and the negative transitivity of a strict
component of an IFR in this case.
The following result shows that, when T = (min, max), each regular strict
component of a T -transitive IFR is pos-transitive. It also establishes that
when T = (min, max), condition T or T is necessary and sufficient on a
T -transitive IFR such that each regular strict component of is negative
transitive.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
implies
0 (x, y) > 0
(x, y) > 0
0
(x, y) < 1 0 (x, y) < 1
(y, x) > 0 (y, x) > 0
(y, x) < 1 0 (y, x) < 1
(3) C is decisive if Gn , x, y X,
C (x, y) > 0 (x, y) > 0
.
C (x, y) < 1 (x, y) < 1
(6) i is a dictator if Gn , x, y X,
i (x, y) > 0 (x, y) > 0
.
i (x, y) < 1 (x, y) < 1
i i
i i
i i
K23798 2015/2/2
i i
Lemma 8.6.3 (Nana and Fono[24]) (Field Expansion Lemma for an IFAR)
Let fe : (W1T )n W1T be an IFAR satisfying IIA and IFPC, C P (N ) and
x, y X. If C is almost decisive on (x, y), then C is decisive.
Proof. Suppose that C is an almost decisive coalition on (x, y). We show
that C is decisive. Let 0 (W1T )n and u, v X be such that
0
C (u, v) > 0
0
C (u, v) < 1
We show that
0 (u, v) > 0
(8.21)
0 (u, v) < 1.
Consider the profile 00 (W1T )n satisfying the following conditions on
{u, v, x, y} :
(1) i C,
00 00
i (u, x) > 0 i (y, v) > 0
00 , 00 , (8.22)
i (u, x) < 1 i (y, v) < 1
00 00 0
i (x, y) > 0 i (u, v) = i (u, v)
00 and 00 0
i (x, y) < 1 i (u, v) = i (u, v)
(2) j N \C,
00 00 00
j (u, x) > 0 j (y, v) > 0 j (y, x) > 0
00 , 00 , 00 (8.23)
j (u, x) < 1 j (y, v) < 1 j (y, x) < 1
i i
i i
i i
K23798 2015/2/2
i i
00 (y, v) > 0
(8.26)
00 (y, v) < 1
Since C is almost decisive on (x, y), we have by the third system of (8.22))
and the third system of (8.23) that
00
(x, y) > 0
(8.27)
00 (x, y) < 1
Since 00 = fe(00 ) W1T , the first result of Theorem 8.5.10 implies that 00
is pos-transitive. The pos-transitivity of 00 , (8.25) and (8.27) imply
00
(u, y) > 0
(8.28)
00 (u, y) < 1
i i
i i
i i
K23798 2015/2/2
i i
0 00
The last system of (8.22) and the fact that N \C and N \C coincide on {u, v}
imply 00 0
N (u, v) = N (u, v)
00 0 (8.30)
N (u, v) = N (u, v)
By (8.29) and (8.30),
00 0
i (u, v) > 0 i (u, v) > 0
00 0
(u, v) < 1 0i (u, v) < 1
i N, i
00 (8.31)
i (v, u) > 0 i (v, u) > 0
00 0
i (v, u) < 1 i (v, u) < 1
Since fe satisfies IIA, (8.31) and (8.24) imply 0 (u, v) > 0 and 0 (u, v) < 1.
Thus the desired result holds.
The following lemma is an intuitionistic fuzzy version of the Group con-
traction lemma for crisp aggregation rules.
Lemma 8.6.4 (Nana and Fono [24]) (Group Contraction Lemma for IFARs)
Let fe : (W1T )n W2T be an IFAR satisfying IIA and IFPC and, D P (N )
with |D| 2.
If D is decisive, then there exists a coalition K P (N ) such that K D
and K is decisive.
Proof. Let D P (N ) be such that |D| 2 and D is decisive. We find a
coalition K D which is decisive. Let {D1 , D2 } be a partition of D and let
the profile (W1T )n satisfy the following conditions on {x, y, z} :
(1) i D1 ,
i (x, y) > 0 i (y, z) > 0
and , (8.32)
i (x, y) < 1 i (y, z) < 1
(2) j D2 ,
j (x, y) > 0 j (z, x) > 0
and , (8.33)
j (x, y) < 1 j (z, x) < 1
(3) k N \D,
k (y, z) > 0 k (z, x) > 0
and . (8.34)
k (y, z) < 1 k (z, x) < 1
Otherwise, by the first system of (8.32) and the first system of (8.33) we have
that D (x, y) > 0 and D (x, y) < 1, by the second system of (8.32) and the
i i
i i
i i
K23798 2015/2/2
i i
first system of (8.34) we have that N \D2 (y, z) > 0 and N \D2 (y, z) < 1,
and by the second system of (8.33) and the second system of (8.34), we have
that
N \D1 (z, x) > 0
. (8.36)
N \D1 (z, x) < 1
Since D (x, y) > 0 and D (x, y) < 1 and D is decisive, we have (x, y) > 0
and (x, y) < 1. Moreover, fe() = W2T and so by the second result of
Theorem 8.5.10, we have that is negative transitive. Hence by the previous
system and the negative transitivity of , we have that
(x, z) > 0 (z, y) > 0
or .
(x, z) < 1 (z, y) < 1
i i
i i
i i
K23798 2015/2/2
i i
Hence (x, y)+ (y, x) = 1+ > 1 and (x, y)+ (y, x) = 0+(1) <
1.
i (y, x) > i (x, y)
(2) If i N, , then
i (y, x) < i (x, y)
(y, x) = 1 (x, y) =
and .
(y, x) = 0 (x, y) = 1
Thus (y, x)+ (x, y) = 1+ > 1 and (y, x)+ (x, y) = 0+(1) < 1.
If i N such that
i (x, y) i (y, x)
i (x, y) i (y, x)
or j N such that
j (y, x) j (x, y)
,
j (y, x) j (y, x)
then
(x, y) = (y, x) =
and .
(x, y) = 1 (y, x) = 1
Hence (x, y)+ (y, x) = 2 > 1 and (x, y)+ (y, x) = (1)+(1) =
2 2 < 1. Thus is strongly complete.
i i
i i
i i
K23798 2015/2/2
i i
or
(x, z) > max( (x, y), (y, z)).
We consider two cases.
(1) If (x, z) < min( (x, y), (y, z)), then (x, z) = and (x, y) =
(y, z) = 1. By the definition of , we have that
i i
i i
i i
K23798 2015/2/2
i i
The result follows upon rewriting this system by the means of and 0 .
We show that fe is non-dictatorial. Assume to the contrary that fe is dicta-
torial. Then there exists i0 N such that W n , x, y X, i0 (x, y) > 0
and i0 (x, y) < 1 implies (x, y) > 0 and (x, y) < 1. Consider the profile
W n defined on {x, y} by k N \{i0 },
k (x, y) < k (y, x) N \{k} (x, y) > N \{k} (y, x)
and
k (x, y) > k (y, x) N \{k} (x, y) < N \{k} (y, x)
i i
i i
i i
K23798 2015/2/2
i i
(x, y) = 0 and (x, y) = 1 which contradicts (x, y) > 0 and (x, y) < 1.
Hence fe is non-dictatorial.
We now show that if we do not assume conditions (C1 and C2 ) or (C1
and C2 ), we obtain an oligarchy IFAR which is an intuitionistic fuzzy version
of Gibbards theorem.
Theorem 8.6.7 (Nana and Fono [24]) Let fe : (W1T )n W1T be an IFAR.
If fe satisfies conditions IIA and IFPC, then there exists a unique oligarchy.
To establish this theorem, we prove the following two Lemmas. We first
establish an intuitionistic fuzzy version of the Veto-field expansion lemma.
Lemma 8.6.8 (Nana and Fono [24]) (Veto-field Expansion Lemma for
IFARs) Let fe : (W1T )n W1T be an IFAR satisfying IIA and IFPC, i N
and u, v X. If i is almost semi-decisive over (u, v), then i has a veto.
Proof. Let i N and u, v X such that i is almost semi-decisive over
(u, v). Let x, y X and 0 (W1T )n be such that 0 (x, y) > 0, 0 (x, y) < 1.
We show that 0 (y, x) = 0 and 0 (y, x) = 1.
Since i is almost semi-decisive over (u, v), (W1T )n such that i satis-
fies (4) of Definition 8.6.1. Let the profile 00 (W1T )n satisfy the following
conditions: u, v, x, y X,
00 00
i (u, v) = i (u, v) > 0
00 00
(u, v) = (u, v) < 1
00 i i
00
i (x, y) = i (x, y) > 0
00
00 00 N (x, u) > 0
( i (x, y) = i (x, y) < 1
00
N (x, u) < 1
00 00 and . (8.40)
(v, u) = (v, u) > 0 00
N (v, y) > 0
N \i N \i
00 00
(v, u) = (v, u) < 1
00
N (v, y) < 1
( N \i N \i
00 00
N \i (y, x) = N \i (y, x) > 0
00 00
N \i (y, x) = N \i (y, x) < 1
By the two last systems of (8.40) and condition IFPC, we have that
00 00
(x, u) > 0 (v, y) > 0
and . (8.41)
00 (x, u) < 1 00 (v, y) < 1
Since i is perfect antisymmetric for all i, we have the first and the third
systems of (8.40) that
00 and coincide on (u, v). . (8.42)
By (4) of Definition 8.6.1, the first and the third systems of (8.40) yield
(v, u) = 0 and (v, u) = 1. By this system, condition IIA and (8.41),
it follows that 00
(v, u) = 0
(8.43)
00 (v, u) = 1.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 8.6.9 (Nana and Fono [24]) (Veto-field Expansion Lemma for
IFARs) Let fe : (W1T )n W2T be an IFAR satisfying IIA and IFPC. Then
there exists a unique decisive coalition S of N such that i S, i has a veto.
Since S is decisive, we have by the first system of (i) of (8.45) and the first
system of (ii) of (8.45) that
(x, y) > 0
. (8.46)
(x, y) < 1
i i
i i
i i
K23798 2015/2/2
i i
8.7 Exercises
1. Prove Proposition 8.1.7.
2. [22] Let fe be the fuzzy aggregation rule defined in Definition 8.1.9. Show
by example that fe does not satisfy PR with respect to (2) .
3. [22] Construct an example of (2) that is T2 -transitive and partially
transitive, but not quasi-transitive.
i i
i i
i i
K23798 2015/2/2
i i
8.8 References
1. F. B. Abdelaziz, J. R. Figueira, and O. Meddeb, On the manipulability
of the fuzzy social choice functions, Fuzzy Sets and Systems, 159 (2008)
177184.
2. K. J. Arrow, Social Choice and Individual Values, New York, Wiley
1951.
3. K. T. Atanassov, Intuitionistic fuzzy sets, Fuzzy Sets and Systems, 20
(1986) 8796.
i i
i i
i i
K23798 2015/2/2
i i
15. L.A. Fono, H. Gwet and S. Fotso. On strict lower and upper sections
of weakly complete fuzzy pre-orders based on co-implication. Fuzzy Sets
and Systems, 159 (2008) 22402255.
16. L. A. Fono, G. N. Nana, M. Salles, and H. Gwet, A binary intuitionistic
fuzzy relation: some new results, a general factorization, and two prop-
erties of strict components, International Journal of Mathematics and
Mathematical Sciences, Hindaw Publishing Corporation Volume 2009,
Article ID 580918, 38 pages dos: 10.1155/2009/580918.
17. L.A. Fono and M. Salles. Continuity of utility functions representing
fuzzy preferences. Social Choice and Welfare, 37 (2011) 669682, doi:
10.1007/s00355-011-0571-0.
18. S. Fotso and L.A. Fono. On the consistency of some crisp choice func-
tions based on a strongly complete fuzzy preorder. New Mathematics
and Natural Computation, 8 (2012) 257272.
19. Garcia-Lapresesta, J. Luis, and B. Llamazares, Aggregation of fuzzy
preferences: Some rules of the mean, Social Choice and Welfare, 17
(2000) 673690.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 9
Manipulability of Fuzzy
Social Choice Functions
Sections 9.1-9.3 are based on the work of Abdelaziz, Figueria, and Med-
deb [1]. In social decision-making contexts, a manipulator often has incentive
to change the social choice in his/her favor by strategically misrepresenting
his/her preference. The Gibbard-Satterthwaite theorem [19, 26] states that
a social choice function over three or more alternatives that does not pro-
vide incentive to individuals to misrepresent their sincere preferences must
be dictatorial. This chapter extends their result to the case of fuzzy weak
preference relations. Here we use the approach given in [1] by defining the
best alternative set in three ways and thus provide three generalizations of
the Gibbard-Satterthwaite theorem in the fuzzy context.
A voting choice procedure is subject to strategic manipulation when an
individual reveals a insincere preference relation with the purpose of changing
the social choice in his favor. Attempts to avoid the negative result of Gibbard
and Satterthwaite have been made by relaxing its assumptions [3, 29]. This
occurs when individuals have difficulty expressing their preference over the
set of alternatives but they can, however, specify a preference degree between
0 and 1 for each ordered pair of alternatives [5, 16, 17, 30]. Thus the choice
of a single alternative can involve the use of a fuzzy social choice function.
How alternatives are to be compared and how the social choice is to be
interpreted must be defined in order to introduce fuzzy manipulation. The
concept of best alternative set, as proposed by Dutta, Panda, and Pattanaik
[15], is used to provide three possible alternative sets. Hence three new defini-
tions of fuzzy manipulability were presented in [1]. In each case, the strategic
manipulation of a fuzzy social choice function by an individual is possible
when two conditions are satisfied. The first condition is that the sincere social
choice does not belong to the best alternative set of the manipulator. The
second condition is that there exists a fuzzy relation assuring the manipulator
289
i i
i i
i i
K23798 2015/2/2
i i
an outcome at least as good as the sincere social choice. A dictator for a fuzzy
social choice function is defined to the individual who secures the social choice
in his/her best alternative set in all situations. Under these new definitions,
it is shown that the Gibbard-Satterthwaite theorem continues to hold.
9.1 Preliminaries
In this section, we present three fuzzy preference relations and their best
alternative sets.
The social choice problem consists of finding the best alternative in ac-
cordance with the preferences of all individuals. The best alternative is also
called the social choice. Fuzzy rather than crisp preferences are considered.
Fuzzy binary relations on the set of alternatives X are used to model the fuzzy
preferences.
i i
i i
i i
K23798 2015/2/2
i i
(3) Dominance degree rule (DDR) : d(i , X)(x) d(i , X)(y), where
d(i , X)(x) = {i (x, z) | z X}, is a dominance degree of alternative x with
respect to all the remaining alternatives in X.
The best alternative set is defined as follows:
Example 9.1.7 Let X = {x, y, z} and N = {1, 2, 3}. Let the fuzzy binary
relations i , i = 1, 2, 3, be given by the following table:
1 x y z 2 x y z 3 x y z
x 1 .6 .64 x 1 .7 .9 x 1 .7 .6
, ,
y .7 1 .65 y .4 1 .7 y .65 1 .6
z .65 .6 1 z .4 .4 1 z 1 .9 1
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
01 x y z
x 1 0 0
y .95 1 1
z .9 0 1
x y z
x 1 .47 .5
y .67 1 .77
z .77 .43 1
(2) Let t = .45 be a confidence threshold for all individuals. Then alterna-
tive x belongs to every best alternative set P2 (X, i ) for all i {1, 2, 3}. Hence
no individual can 2-manipulate the FSCF in this situation.
i i
i i
i i
K23798 2015/2/2
i i
(3) Individual 3 can 3-manipulate the FSCF. Alternative x does not belong
to individual 3s best alternative set P3 (X, 3 ) = {z}. Individual 3 can thus
reveal the non-sincere fuzzy relation 03 to obtain z as the social choice.
03 x y z
x 1 0 0
y .9 1 0
z .95 1 1
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
For i 6= 1,
0
i is a CLO on P1 (X, 1 ),
0 (y, x) = 1 and 0i (x, y) = 0 if x P1 (X, 1 ), y X\P1 (X, 1 ).
i0
i is a CLO on X\P1 (X, 1 ).
xj = v(|01 , 02 , . . . , 0j ) P1 (X, 1 ),
xj1 = v1 (|01 , 02 , . . . , 0j1 )
/ P1 (X, 1 ).
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
(2) if (x) = N \0 (y), then one of v(x) and v 0 (y) is greater than 1/2 if and
only if the other is less than 1/2.
v (z) = 0 for all z X\{x, y}. Thus if either v (x) or v (y) is less than 1/2,
then the other is greater than 1/2. This holds for v(x) and v(y) since f satisfies
MI by Lemma 9.4.2 and v(x) = v (x) and v(y) = v (y) by MI.
By replacing with N and 0 with 1, a similar argument shows that if
either v(x) and v 0 (y) is greater than 1/2, then the other must be less than
1/2.
(2) Suppose x 6= y and (x) = 0 (y). Let z X\{x, y}. Let 00 be a profile
such that 00 (z) = N \(x). Also, 00 (x) = N \0 (y). By case (1), if either one
of v 00 (z) or v(x) is less than 1/2, then the other is greater than 1/2. This is
also the case for v 00 (z) and v 0 (y). Thus it follows that v(x) and v 0 (y) are both
greater than or both less than or both equal to 1/2.
(3) Suppose x = y and (x) = 0 (y). Then v(x) = v 0 (y) by MI.
(4) Suppose x = y and (x) = N \0 (y). Let z X\{x} and let be a
profile such that (z) = (x). Then by case (2), v (z) and v(x) are both
greater than, both less than or both equal to 1/2. Since (z) = N \0 (y), we
have by (1) that if either of v (z) and v 0 (y) is less than 1/2, then the other is
greater than 1/2. Hence it follows that if either of v(x) and v 0 (y) is less than
1/2, then the other is greater than 1/2. Thus neutrality holds.
i i
i i
i i
K23798 2015/2/2
i i
is less than 1/2. Since f satisfies neutrality, this must also be true of every
other alternative.
9.5 Application
We consider a group of individuals N = {1, . . . , n} that wishes to classify a
finite set of alternatives X as belonging to one of two sets. The set X may
be countries and the classification may classify the countries as those who
are politically stable and those who are not. Individuals may disagree on the
classification and so an aggregation rule is applied. The aggregation rule is a
function from the set V n into V, where
V = { | X [0, 1] | x X, (x) = 1 y X, (y) < 1/2
and x X, (x) = 0 y X, (y) > 1/2}.
We justify our definition of aggregation rule as follows. Suppose the number 1
corresponds to the classification definitely politically stable. Since politically
stable and not politically stable are relative concepts, a country cannot be
classified definitely stable unless another country is classified as more politi-
cally unstable than politically stable. Similar reasoning can be applied to the
classification of 0.
We do not want the social classification to be imposed. The nonimposi-
tion axiom guarantees that every country is definitely stable for some profile
of individual partitions and definitely unstable for some other profile of in-
dividual partitions. The axiom ensures that the social classification of any
country takes at least two values. This is a defensible requirement ensuring
that the social classification of any country is at least minimally responsive
to individual evaluations.
It is also undesirable for the social classification to be potentially manip-
ulable. If the social classification is manipulable, then we cannot guarantee
that individual evaluators will submit partitions that are sincere. It is also
desirable for the social classification to be non-dictatorial.
Impossibility theorems have been discovered by Mirkin [24], Leclerc [22],
and Fishburn and Rubuinstein [16], Barthelemy et al. [4], Dimitrov et al. [12],
and Chambers and Miller [6]. The formal model presented here is different
to the ones presented by these authors. The model here is applicable to the
possible world problem posed by List [23] and to the aggregation of so-called
dichotomous preferences Inada [20].
i i
i i
i i
K23798 2015/2/2
i i
P3
For C X, define sC as follows: sC = i=1 vi (C). Let
( 3 )
X
M = max vi (C) | C X
i=1
and ( )
3
X
m = min vi (C) | C X .
i=1
9.6 Exercises
1. A function C : F(X)n F(X) is called a fuzzy choice function. Let C
be a fuzzy choice function.
(1) Then C is called manipulable if there exists x X, F(X)n ,
i N, and i0 F(X) such that (a) C()(x) < i (x) and C()(x) <
C(M \i , 0 )(x) or (b) C()(x) > i (x) and C()(x) > C(M \i , 0 )(x).
(2) C is called strategy-proof if its not manipulable.
(3) C is called weakly Paretian if for all F(X)n and all x X,
max{i (x) | i N } C()(x) min{i (x) | i N }.
(4) C is said to satisfy the -only condition if for all , 0 F(X)n and
all x X, i (x) i0 (x) i N, implies C()(x) = C 0 )(x).
(5) C is said to be monotonic if for all x X and all , 0
F (X)n , i (x) i0 (x) implies C()(x) ( 0 )(x).
If C is a strategy-proof fuzzy choice function, prove that it satisfies the
-only condition.
i i
i i
i i
K23798 2015/2/2
i i
9.7 References
1. F. B. Abdelaziz, J. R. Figueira, and O. Meddeb, On the manipulability
of the fuzzy social choice functions, Fuzzy Sets and Systems, 159 (2008)
177184.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Chapter 10
In this chapter, we focus on the work of Georgescu [19, 22]. Zadeh [47] in-
troduced the notion of a similarity relation in order to generalize the notion
of an equivalence relation. Trillas and Valverde [40] extended the notion of a
similarity relation for an arbitrary t-norm. See also [27, p. 254].
Samuelson [34] introduced revealed preference theory in order to define
the rational behavior of consumers in terms of a preference relation. Others
including Uzawa [42], Arrow [1], Sen [35], Suzumura [38] have developed a
revealed preference theory in a context of choice functions. Preferences of the
individuals can be imprecise, due possibly to human subjectivity or incom-
plete information. Vague preferences can be mathematically modelled by fuzzy
relations, [9-12, 14, 28]. However, the choice can be exact or vague even if the
preference is ambiguous. When the choice is exact, it can be mathematically
described by a crisp choice function. The case of vague preferences and exact
choices considered in [3] has been presented in Chapter 7. In situations such
as negotiations on electronic markets, the choices are fuzzy.
Various types of fuzzy choice functions have been presented in Chapters 2
and 7. See also [2, 33, 45]. Georgescu [16 , 17] presented a fuzzy generalization
of classic theory on revealed preference which generalized the results of [1, 34,
35].
In this chapter, Georgescus degree of similarity of two fuzzy choice func-
tions is presented. It is analogous to the degree of similarity of two fuzzy sets
and induces a similarity relation on the set of all choice functions defined on
a collection of fuzzy choice functions. The degree of similarity of two choice
functions is defined to be the pair (, t), where is a t-norm and t is an element
of the interval [0, 1].
There is an interesting notion involving similarity relations and that is
Poincares paradox. Poincare stated that in the physical world equal re-
305
i i
i i
i i
K23798 2015/2/2
i i
Lemma 10.1.1 (Belohl avek [6], H ajek [25]) The following properties hold for
all a, b, c [0, 1]:
(1) (a b) (b c) a c.
(2) a (a b) a b; if is continuous, then a (a b) = a b.
Lemma 10.1.2 (Belohl ajek [25]) For any index set I and a, ai , bj
avek [6], H
[0, 1], the following properties hold:
(1) a ({ai | i I}) = {a ai | i I}
(2) ({ai | i I}) a = {ai a | i I};
(3) {ai a | i I} ({ai | i I}) a;
(4) ({ai | i I}) a = {ai a | i I};
(5) ({ai | i I}) ({bj | j I}) {ai bj | i, j I}.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 10.1.3 (Turunen [41]) The following properties hold for all
a, b, c, d [0, 1].
(1) (a, 1) = a;
(2) a = b
(a, b) = 1;
(3) (a, b) =
(b, a);
(4) (a, b) (a, b);
(5) (a, b)
(b, c)
(a, c);
(6) (a, b) (c, d)
(a c, b d);
(7) (a, b) (c, d) (a c, b d);
(8) (a, b)
(c, d) (a c, b d).
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Definition 10.2.1 Let C1 , C2 be two fuzzy choice functions on (X, B). The
degree of similarity E(C1 , C2 ) of C1 and C2 is defined to be
E(C , C ) = {
1 2
(C () (x) , C () (x)) | B, x X}.
1 2
For t [0, 1], we say that C1 and C2 are (, t)-equal, written C1 =(,t) C2 , if
E(C1 , C2 ) t.
i i
i i
i i
K23798 2015/2/2
i i
Let FCF(X, B) denote the set of fuzzy choice functions on (X, B) and by
FR(X) the set of fuzzy preference relations on X. By Proposition 10.2.2, E
is a similarity relation on FCF(X, B).
Let A, B be two non-empty sets. The notation A B denotes a partial
function from A to B, i.e., a function f whose domain Dom(f ) is a subset of
A and whose image Im(f ) is a subset of B.
Consider the partial functions i : FR(X) FCF(X, B), i = 1, 2, where
Dom (1 ) = { FR(X) | M ( , ) FCF(X, B)},
Dom (2 ) = { FR(X) | G( , ) FCF(X, B)},
and are defined by 1 () = M ( , ) for all Dom(1 ) and 2 () = G( , )
for all Dom(2 ).
Consider the partial functions, i : FCF(X, B) FR(X), i = 1, 2, 3,
defined by for all C F CF(X, B),
1 (C) = C , 2 (C) = C , and 3 (C) = C .
The functions i and i provide correspondences between the fuzzy
choice functions on (X, B) and certain fuzzy preference relations on X. They
provide the means by which the results for fuzzy choice functions are proved.
We are interested in the way these functions preserve similarity.
The next result shows to what extent the similarity of the fuzzy pref-
erence relations is transformed by the functions 1 and 2 into the similarity
of the corresponding choice functions.
i i
i i
i i
K23798 2015/2/2
i i
E(2 (1 ), 2 (2 ))
= {
(1 (1 )()(x), 1 (2 )()(x)) | B, x X}
= {
(G(, 1 )(x), G(, 2 )(x)) | B, x X}
= {
((x) {(y) 1 (x, y) | y X},
(x) {(y) 2 (x, y) | y X}) | B, x X}
{
({(y) 1 (x, y) | y X},
{(y) 2 (x, y)) | y X}) | B, x X}
{
((y) 1 (x, y), ((y) 2 (x, y)) | B, x X}
{
(1 (x, y), 2 (x, y)) | x, y X} = E(1 , 2 ).
Theorem 10.2.5 (Georgescu [19]) Let C1 , C2 FCF(X, B). Then the fol-
lowing statements hold.
(1) E(1 (C1 ), 1 (C2 )) E(C1 , C2 );
(2) E(2 (C1 ), 2 (C2 )) E(C1 , C2 );
(3) E(3 (C1 ), 3 (C2 )) E(C1 , C2 ) E(C1 , C2 ).
Proof. (1) By Lemma 10.1.3(2), (7) and Lemma 10.1.4(2), we have that
i i
i i
i i
K23798 2015/2/2
i i
{
(C1 (1{x,y} )(x, y), C2 (1{x,y} )(x)) | x, y X}
{
(C1 (1{x,y} )(x), C2 (1{x,y} )(x))
= E(C1 , C2 ) | x, y X}.
(3) By Lemma 10.1.3(2), (4), (7) and Lemma 10.1.4(2), it follows that
i i
i i
i i
K23798 2015/2/2
i i
E(C, C)
b = {(C()(x), C()(x))
b | x X, B}
= {(C()(x) C()(x)) | x X, B}
b
since
(C()(x), C()(x))
b = C()(x)
b C()(x).
By Lemma 10.1.1(5) and Lemma 10.1.2(5), we have that
I(, ) (x) C(
b )(x)
= {((u) (u)) | u X} (x) (x)
({ (u) C (x, u) | u X})
(x) ({((u) (u)) ( (u) C (x, u)) | u X})
(x) ({(u) C (x, u) | u X} C()(x).
b
b we have since C C
Proof. By the definition of C, b that
E(C, C)
b = {C()(x)
b C()(x) | x X, B}
= {C()(x) C()(x) | x X, B}.
b
i i
i i
i i
K23798 2015/2/2
i i
Thus
b {x,y} )(x) C (x, y)
C (x, y) C (x, y) = C(1
b {x,y} )(x) C(1{x,y} )(x).
= C(1
Hence
E(C, C)
b = {C()(x)
b C()(x) | x X, B}
{C(1
b {x,y} )(x)) C(1{x,y} )(x) | x, y X}
= {C (x, y) C (x, y) | x, y X}
= E(C , C ).
C()(x)
b = (x) [(x) C (x, x)] [(y) C (x, y)] = s (t q) ,
C()(y)
b = (y) [(x) C (y, x)] [(y) C (y, y)] = s (s r) = 1.
C()(x)
b = 1 and C()(y)
b = t (s r) .
i i
i i
i i
K23798 2015/2/2
i i
E(C, C)
b = {(C()(z)
b C() (z)) | {, }, z {x, y}}
= [C()(x)
b C()(x)] [C()(y)
b C()(y)]
[C()(x) C()(x)] [C()(y) C()(y)]
b b
= [(s (t q)) q] [(t (s r)) r]
= {[(s t) (s q)] q} {[(t s) (t r) r]}
t) q] [(s t) r] = (s t) (q r)
= [(s
1 if s t = q r
=
q r if s t > q r.
In the case when the results of several independent experts are used, we
can consider that a fuzzy choice problem models a multicriterial decision-
making problem by interpreting the available fuzzy sets as vague attributes
or criteria. Then it would be beneficial to rank the choice functions in order
to be able to select the best ones.
The normality of the fuzzy choice function C expresses the rationality of
C by the fuzzy revealed preference relation C . Thus the similarity degree
E(C, C)
b can be considered a measure of the rationality of C. For two choice
c1 ) E(C2 , C
functions C1 and C2 , if E(C1 , C c2 ), then we can consider C1 to
be more rational than C2 .
In the next example, we illustrate how a ranking of a collection of fuzzy
choice functions can be obtained by using the similarity degree E(C, C).
b
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
properties AA and full rationality were later integrated by Sen [35] in a result
regarding the characterization of the rationality of a crisp choice function
by various conditions such as the axiom of revealed preferences (WARP),
(SARP) and the congruence axioms (WCA), (SCA). The result is known as
the Arrow-Sen Theorem.
In [20], Georgescu extended the classical result of [1] by showing that
(FAA) is equivalent with the full rationality of fuzzy choice functions. A new
concept defined in [20] is the Arrow index A(C) of a fuzzy choice function C.
The A(C) expresses the degree to which C satisfies the Fuzzy Arrow Axiom.
As previously discussed, revealed preference is a concept introduced by
Samuelson in 1938 [34] to measure the rationality of a consumers behavior
in terms of a preference relation associated with a demand function. This is
a major concern of classical economic theory. The work of Uzawa [40], Arrow
[1], Richter [32], Sen [35, 36, 37], Suzumura [38, 39] and others extended the
area.
In Chapter 7, we presented the work by Banerjee where the choice function
has the domain consisting of crisp sets and the range consisting of fuzzy sets.
In Chapter 7, we also presented the work of Barrett, Pattanaik and Salles
where the choice is unambiguous even though the preferences are ambiguous.
See also Dasgupta and Deb [8].
In [16, 17, 18, 19, 20, 21] Georgescu worked with a general definition of the
fuzzy choice function in order to develop a theory of revealed preference. Pre-
viously, we presented the fuzzy Arrow axiom (FAA) introduced by Georgescu
[19] who showed that a fuzzy choice function satisfies FAA if and only if it is
full rational. This result extends a classical result of [1] which established the
equivalence between the Arrow Axiom and the full rationality of crisp choice
functions (see also [39]).
We show that A(C) characterizes the degree to which the fuzzy choice
function C is full rational. This result refines the above mentioned result from
[19] which is obtained as a particular case.
In [20], the equivalence between FAA and the full rationality of fuzzy choice
functions was established. Using the Arrow index to measure the degree of full
rationality of a fuzzy choice function, the results of this section allow for the
comparison of them from the point of view of their rationality. Then by using
the Arrow index one can obtain a ranking of fuzzy choice functions with
respect to their full rationality.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
Lemma 10.3.3 ([16]) Suppose C is a fuzzy choice function on (X, B). Then
C C and C , C are reflexive and strongly total.
Lemma 10.3.4 ([20]) Suppose C is a fuzzy choice function on (X, B). Then
C() G(, C ) for all B.
i i
i i
i i
K23798 2015/2/2
i i
FAA can be interpreted as follows: For all x, 1 , and 2 , the maximum degree
that x is chosen from 2 , x belongs to 1 and 1 is included in 2 , is less than
or equal to the degree that the set of alternatives chosen from 2 , and also in
1 is equal to the set of alternatives chosen from 1 , [21].
In [20], Georgescu extended the classical result of [1] involving AA by
showing that FAA is equivalent with the full rationality of fuzzy choice func-
tions. A new concept defined in [20] is the Arrow index A(C) of a fuzzy choice
function C. The Arrow index provides the degree to which C verifies the fuzzy
Arrow axiom.
It is shown in this section that A(C) characterizes the degree to which C
is full rational. As stated by Georgescu in [22]: This type of result changes the
perspective of studying the rationality of fuzzy choice functions. The attention
is no longer focused on rational fuzzy choice functions, but the entire class
of fuzzy choice functions has been taken into account with respect to their
rationality. By the Arrow index, one can appreciate how rational any fuzzy
choice function is. This fact has direct consequences in concrete problems.
More often, the choice of an agent does not fulfill conditions of rationality or
full rationality. By using Arrow index, from a set of fuzzy choice functions we
can select the ones with the maximum degree of rationality.
i i
i i
i i
K23798 2015/2/2
i i
Definition 10.3.7 Let C be a choice function on (X, B). The Arrow index
A(C) of C is defined by
A(C) is the degree to which the fuzzy choice function C satisfies the fuzzy
Arrow axiom. It follows that A(C) = 1 if and only if C verifies FAA.
By (10.3),
A(C) C(1{x,y} )(x) C(1{y,z} )(y) A(C) C(1{x,y,z} )(y) C(1{y,z} )(x)
C(1{x,y,z} )(x)
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
C(1 )(z) = I(1 , 2 ) 1 (x) C(2 )(x) C(1 )(z) 2 (v) (z, v)
Also,
Therefore,
i i
i i
i i
K23798 2015/2/2
i i
Thus we have that 1 (z) C(2 )(z) 1 (y) (z, y) for all y X.
Hence
{(1 (z) C(2 )(z)) C(1 )(z) | x X} = E(1 C(2 ), C(1 )).
Thus (10.8) was proved, hence (10.7) is verified. Since (10.7) holds for all
1 , 2 B and x X, we obtain (10.6).
i i
i i
i i
K23798 2015/2/2
i i
Proof. From (10.6), i.e., Lemma 10.3.11, we can derive the inequality
Thus A(C) C()(x) G(, C )(x) and so (10.13) holds. In order to prove
(10.14), we consider an element y X such that C()(y) = 1. Since C() is
a normal fuzzy subset of X, (y) = 1. Hence
We note that
Therefore,
A(C) G(, C )
(I(1{x,y} , ) 1{x,y} (y) C()(y))
[(I(1{x,y} , ) 1{x,y} (y) C()(y)) E(1{x,y} C(), C(1{x,y} ))]
E(1{x,y} C(), C(1{x,y} )).
i i
i i
i i
K23798 2015/2/2
i i
A(C) G(, C )
C(1{x,y} (x)) E(1{x,y} C(), C(1{x,y} ))
C(1{x,y} (x) [(1{x,y} (x) C()(x) C(1{x,y} (x))]
= C (x, y) [C()(x) C (x, y)]
C (x, y) [C (x, y) C()(x)]
= C (x, y) C()(x)
C()(x).
Thus inequality (10.12) holds. Since C is reflexive and strongly total, we have
Ref(C ) =ST(C ) = 1. Thus by Theorem 10.3.9 and (10.12), we get
Corollary 10.3.13 Assume that is the minimum t-norm . Then the fol-
lowing assertions are equivalent:
(1) C satisfies FAA;
(2) E(C, G( , )) = Trans() = Ref() = ST() = 1 for some FR(X)
(3) C is full rational.
By Corollary 10.3.13, we have that the main result of [20] holds, namely,
FAA is equivalent with the full rationality.
Let tc
C denote the transitive closure of the fuzzy revealed preference re-
lation C . Recall the following congruence axioms for fuzzy choice functions
defined in Chapter 2.
Weak fuzzy congruence axiom (WFCA)
For all B and x, y X,
tc
C (x, y) C()(y) (x) C()(x).
i i
i i
i i
K23798 2015/2/2
i i
WFCA and SFCA are fuzzy versions of the classical congruence axioms
(WCA), (SCA) [30, 33, 36, 37].
We next define indicators of the axioms WFCA and SFCA.
WFCA(C) = SFCA(C)
= E(C, G( , C )) Trans(C ) = E(C, G( , C )) Trans(C ).
i i
i i
i i
K23798 2015/2/2
i i
In Georgescu [20], the equivalence between FAA and the full rationality of
the fuzzy choice function was shown. The notion of full rationality combines
two types of properties, namely, properties of fuzzy choice functions with
rationality and properties of fuzzy preference relations with reflexivity, max-
transitivity, and strong connectedness.
In this section, the relation between FAA and the full rationality has been
analyzed from a different point of view. The Arrow index A(C) of a fuzzy
choice function, an indicator which expresses the degree to which C satis-
fies FAA, was used rather than the Arrow Axiom. Also a numerical measure
of the full rationality, expressed by the degree of similarity of fuzzy choice
functions and by the indicators of reflexivity, max- transitivity, and strong
completeness of the fuzzy preference relation is used.
The equivalence between FAA and the full rationality is transformed here
into the equality between A(C) and the degree of full rationality. The result
of this section strengthens the one in [20] since, as noted by Georgescu, it
consists of a numerical relation valid for any fuzzy choice function. It also
provides a method of ranking the fuzzy choice functions with respect to their
rationality.
10.4 Exercises
The following exercises are based on [23]. If is a fuzzy relation on X, then
the associated fuzzy strict of is defined by (x, y) = (x, y) (y, x)
for all x, y X. Let C be a fuzzy choice function. Define the fuzzy subsets
of B : RatG (C) = {E(C, G( , ) | F R}, NormG (C) = E(C, G( , C )
and RATM (C) = {E(C, M ( , ) | FR}, Norm(C) = E(C, M ( , C )).
RatG (C) is called the indicator of G-rationality of C and RatM (C) is called
the indicator of M -rationality of C. NormG (C) is called the indicator of
G-normality of C and NormM (C) is called the indicator of M -normality
of C.
1. Let 1 , 2 be fuzzy relations on X. Prove that for all x, y X, E(1 , 2 )
1 (x, y) 2 (x, y).
2. Let C be a fuzzy choice function on (X, B). Prove the following state-
ments.
(i) NormM (C) RatM (C) RatG (C) = NormG (C);
(ii) If C satisfies H1 , H2 , then RatG (C) E(C, G( , C ).
The function Ac () = {{((x1 , x2 ) . . . (xn1 xn ) (xn , x1 ))
(x1 , xn ) | x1 , . . . , xn X} | n N, n 2} is called degree of acyclic-
ity of .
3. Prove that Ac () = {I( n1 1 , ) | n N, n 2}, where n1
denotes the max-min composition of with itself n 1 times.
i i
i i
i i
K23798 2015/2/2
i i
10.5 References
1. K. J. Arrow, Rational choice functions and orderings, Economica, 26
(1959) 121127.
2. A. Banerjee, Fuzzy choice functions, revealed preference and rationality,
Fuzzy Sets and Systems, 70 (1995) 3143.
7. K.Y. Cai, -Equalities of fuzzy sets, Fuzzy Sets and Systems, 76 (1995)
97112.
8. M. Dasgupta and R. Deb, Fuzzy Choice functions, Soc. Choice Welfare,
8 (1991) 171182.
9. B. De Baets and J. Fodor, Twenty years of fuzzy preference relations
(1978-1997), Belg. J. Oper. Res. Statist. Comput. Sci., 37 (1997) 6182.
10. B. De Baets and J. Fodor (Eds.), Principles of Fuzzy Preference Mod-
elling and Decision Making, Academia Press, Gent, 2003.
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i
i i
K23798 2015/2/2
i i
i i
i i